03 Cortex A Programmer Guide
User Manual:
Open the PDF directly: View PDF .
Page Count: 421
Download | |
Open PDF In Browser | View PDF |
ARM Cortex -A Series ® ™ Version: 4.0 Programmer’s Guide Copyright © 2011 – 2013 ARM. All rights reserved. ARM DEN0013D (ID012214) ARM Cortex-A Series Programmer’s Guide Copyright © 2011 – 2013 ARM. All rights reserved. Release Information The following changes have been made to this book. Change history Date Issue Confidentiality Change 25 March 2011 A Non-Confidential First release 10 August 2011 B Non-Confidential Second release. Updated to include Virtualization, Cortex-A15 processor, and LPAE. Corrected and revised throughout 25 June 2012 C Non-Confidential Updated to include Cortex-A7 processor, and big.LITTLE. Index added. Corrected and revised throughout. 22 January 2014 D Non-Confidential Updated to include Cortex-A12 processor, Cache Coherent Interconnect, expanded GIC coverage, Multi-core processors, Corrected and revised throughout. Proprietary Notice This Cortex-A Series Programmer’s Guide is protected by copyright and the practice or implementation of the information herein may be protected by one or more patents or pending applications. No part of this Cortex-A Series Programmer’s Guide may be reproduced in any form by any means without the express prior written permission of ARM. No license, express or implied, by estoppel or otherwise to any intellectual property rights is granted by this Cortex-A Series Programmer’s Guide. Your access to the information in this Cortex-A Series Programmer’s Guide is conditional upon your acceptance that you will not use or permit others to use the information for the purposes of determining whether implementations of the information herein infringe any third party patents. This Cortex-A Series Programmer’s Guide is provided “as is”. ARM makes no representations or warranties, either express or implied, included but not limited to, warranties of merchantability, fitness for a particular purpose, or non-infringement, that the content of this Cortex-A Series Programmer’s Guide is suitable for any particular purpose or that any practice or implementation of the contents of the Cortex-A Series Programmer’s Guide will not infringe any third party patents, copyrights, trade secrets, or other rights. This Cortex-A Series Programmer’s Guide may include technical inaccuracies or typographical errors. To the extent not prohibited by law, in no event will ARM be liable for any damages, including without limitation any direct loss, lost revenue, lost profits or data, special, indirect, consequential, incidental or punitive damages, however caused and regardless of the theory of liability, arising out of or related to any furnishing, practicing, modifying or any use of this Programmer’s Guide, even if ARM has been advised of the possibility of such damages. The information provided herein is subject to U.S. export control laws, including the U.S. Export Administration Act and its associated regulations, and may be subject to export or import regulations in other countries. You agree to comply fully with all laws and regulations of the United States and other countries (“Export Laws”) to assure that neither the information herein, nor any direct products thereof are; (i) exported, directly or indirectly, in violation of Export Laws, either to any countries that are subject to U.S. export restrictions or to any end user who has been prohibited from participating in the U.S. export transactions by any federal agency of the U.S. government; or (ii) intended to be used for any purpose prohibited by Export Laws, including, without limitation, nuclear, chemical, or biological weapons proliferation. Words and logos marked with ® or ™ are registered trademarks or trademarks of ARM Limited, except as otherwise stated below in this proprietary notice. Other brands and names mentioned herein may be the trademarks of their respective owners. Copyright © 2011 – 2013 ARM Limited, 110 Fulbourn Road Cambridge, CB1 9NJ, England Figure 1-1 on page 1-2 is supplied courtesy of The Centre for Computing History , www.computinghistory.org.uk This document is Non-Confidential but any disclosure by you is subject to you providing notice to and the acceptance by the recipient of, the conditions set out above. In this document, where the term ARM is used to refer to the company it means “ARM or any of its subsidiaries as appropriate”. Web Address http://www.arm.com ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential ii Contents ARM Cortex-A Series Programmer’s Guide Preface Preface to the 4th Edition ................................................................................................ ix Glossary ............................................................................................................................ x Typographical conventions ............................................................................................ xiv Feedback on this book .................................................................................................... xv References .................................................................................................................... xvi Chapter 1 Introduction 1.1 1.2 1.3 Chapter 2 ARM Architecture and Processors 2.1 2.2 2.3 2.4 2.5 Chapter 3 Registers ....................................................................................................................... 3-6 Introduction to Assembly Language 4.1 4.2 4.3 4.4 4.5 ARM DEN0013D ID012214 Architectural profiles ..................................................................................................... 2-2 Architecture history and extensions .............................................................................. 2-3 Processor properties ..................................................................................................... 2-8 Cortex-A series processors ........................................................................................ 2-10 Key architectural points of ARM Cortex-A series processors ..................................... 2-16 ARM Processor Modes and Registers 3.1 Chapter 4 History ........................................................................................................................... 1-2 System-on-Chip (SoC) .................................................................................................. 1-4 Embedded systems ...................................................................................................... 1-5 Comparison with other assembly languages ................................................................ 4-2 The ARM instruction sets .............................................................................................. 4-3 Introduction to the GNU Assembler .............................................................................. 4-5 ARM tools assembly language ..................................................................................... 4-9 Interworking ................................................................................................................ 4-11 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential iii Contents 4.6 4.7 Chapter 5 ARM/Thumb Unified Assembly Language Instructions 5.1 5.2 5.3 5.4 5.5 5.6 Chapter 6 ARM memory ordering model ..................................................................................... 10-3 Memory barriers .......................................................................................................... 10-6 Cache coherency implications .................................................................................. 10-11 Exception Handling 11.1 11.1.1 11.2 11.3 11.4 ARM DEN0013D ID012214 Virtual memory .............................................................................................................. 9-3 The Translation Lookaside Buffer ................................................................................. 9-4 Choice of page sizes .................................................................................................... 9-6 First level address translation ....................................................................................... 9-7 Level 2 translation tables ............................................................................................ 9-11 Memory attributes ....................................................................................................... 9-14 Multi-tasking and OS usage of translation tables ....................................................... 9-17 Memory Ordering 10.1 10.2 10.3 Chapter 11 Why do caches help? ................................................................................................... 8-3 Cache drawbacks ......................................................................................................... 8-4 Memory hierarchy ......................................................................................................... 8-5 Cache architecture ........................................................................................................ 8-6 Cache policies ............................................................................................................ 8-13 Write and Fetch buffers .............................................................................................. 8-15 Cache performance and hit rate ................................................................................. 8-16 Invalidating and cleaning cache memory .................................................................... 8-17 Point of coherency and unification .............................................................................. 8-19 Level 2 cache controller .............................................................................................. 8-22 Parity and ECC in caches ........................................................................................... 8-23 The Memory Management Unit 9.1 9.2 9.3 9.4 9.5 9.6 9.7 Chapter 10 SIMD ............................................................................................................................. 7-2 NEON architecture overview ........................................................................................ 7-5 NEON C Compiler and assembler .............................................................................. 7-11 Caches 8.1 8.2 8.3 8.4 8.5 8.6 8.7 8.8 8.9 8.10 8.11 Chapter 9 Floating-point basics and the IEEE-754 standard ........................................................ 6-2 VFP support in GCC ..................................................................................................... 6-8 VFP support in the ARM Compiler ................................................................................ 6-9 VFP support in Linux .................................................................................................. 6-10 Floating-point optimization .......................................................................................... 6-11 Introducing NEON 7.1 7.2 7.3 Chapter 8 Instruction set basics .................................................................................................... 5-2 Data processing operations .......................................................................................... 5-6 Memory instructions .................................................................................................... 5-13 Branches ..................................................................................................................... 5-15 Saturating arithmetic ................................................................................................... 5-16 Miscellaneous instructions .......................................................................................... 5-17 Floating-Point 6.1 6.2 6.3 6.4 6.5 Chapter 7 Identifying assembly code .......................................................................................... 4-12 Compatibility with ARMv8-A ....................................................................................... 4-13 Types of exception ...................................................................................................... 11-3 Exception priorities ..................................................................................................... 11-6 Exception handling ................................................................................................... 11-10 Other exception handlers .......................................................................................... 11-12 Linux exception program flow ................................................................................... 11-14 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential iv Contents Chapter 12 Interrupt Handling 12.1 12.2 Chapter 13 Boot Code 13.1 13.2 13.3 Chapter 14 Idle management ........................................................................................................ Hotplug ....................................................................................................................... Dynamic Voltage and Frequency Scaling ................................................................... Assembly language power instructions ...................................................................... Power State Coordination Interface ............................................................................ 20-3 20-6 20-7 20-8 20-9 Security 21.1 ARM DEN0013D ID012214 Amdahl’s law ............................................................................................................... 19-2 Decomposition methods ............................................................................................. 19-3 Threading models ....................................................................................................... 19-5 Threading libraries ...................................................................................................... 19-6 Performance issues .................................................................................................... 19-9 Synchronization mechanisms in the Linux kernel ..................................................... 19-11 Profiling in SMP systems .......................................................................................... 19-13 Power Management 20.1 20.2 20.3 20.4 20.5 Chapter 21 Multi-processing ARM systems .................................................................................. 18-3 Symmetric multi-processing ........................................................................................ 18-5 Asymmetric multi-processing ...................................................................................... 18-7 Heterogeneous multi-processing ................................................................................ 18-8 Cache coherency ........................................................................................................ 18-9 TLB and cache maintenance broadcast ................................................................... 18-13 Handling interrupts in an SMP system ...................................................................... 18-14 Exclusive accesses ................................................................................................... 18-15 Booting SMP systems ............................................................................................... 18-17 Private memory region .............................................................................................. 18-19 Parallelizing Software 19.1 19.2 19.3 19.4 19.5 19.6 19.7 Chapter 20 Compiler optimizations ............................................................................................... 17-2 ARM memory system optimization ............................................................................. 17-7 Source code modifications ........................................................................................ 17-12 Multi-core processors 18.1 18.2 18.3 18.4 18.5 18.6 18.7 18.8 18.9 18.10 Chapter 19 Profiler output ............................................................................................................. 16-3 Optimizing Code to Run on ARM Processors 17.1 17.2 17.3 Chapter 18 Procedure Call Standard ............................................................................................ 15-2 Mixing C and assembly code ...................................................................................... 15-8 Profiling 16.1 Chapter 17 Endianness ................................................................................................................. 14-2 Alignment .................................................................................................................... 14-5 Miscellaneous C porting issues .................................................................................. 14-7 Porting ARM assembly code to ARMv7-A ................................................................ 14-10 Porting ARM code to Thumb .................................................................................... 14-11 Application Binary Interfaces 15.1 15.2 Chapter 16 Booting a bare-metal system ...................................................................................... 13-2 Configuration .............................................................................................................. 13-6 Booting Linux .............................................................................................................. 13-7 Porting 14.1 14.2 14.3 14.4 14.5 Chapter 15 External interrupt requests .......................................................................................... 12-2 The Generic Interrupt Controller ................................................................................. 12-7 TrustZone hardware architecture ................................................................................ 21-2 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential v Contents Chapter 22 Virtualization 22.1 22.2 22.3 22.4 Chapter 23 big.LITTLE 23.1 23.2 23.3 23.4 Chapter 24 Linux distributions ......................................................................................................... B-2 Useful tools ................................................................................................................... B-5 Software toolchains for ARM processors ...................................................................... B-7 ARM DS-5 ................................................................................................................... B-10 Example platforms ...................................................................................................... B-12 Building Linux for ARM Systems C.1 C.2 C.3 ARM DEN0013D ID012214 Instruction Summary ..................................................................................................... A-2 Tools, Operating Systems and Boards B.1 B.2 B.3 B.4 B.5 Appendix C ARM debug hardware ................................................................................................. 24-2 ARM trace hardware ................................................................................................... 24-4 Debug monitor ............................................................................................................ 24-7 Debugging Linux applications ..................................................................................... 24-8 DS-5 debug and trace ............................................................................................... 24-10 Instruction Summary A.1 Appendix B Structure of a big.LITTLE system ............................................................................... 23-2 Software execution models in big.LITTLE .................................................................. 23-4 big.LITTLE MP ............................................................................................................ 23-7 Using big.LITTLE ...................................................................................................... 23-10 Debug 24.1 24.2 24.3 24.4 24.5 Appendix A ARMv7-A Virtualization Extensions ............................................................................ 22-3 Hypervisor software .................................................................................................... 22-6 Relationship between virtualization and ARM Security Extensions ............................ 22-9 Large Physical Address Extensions ......................................................................... 22-10 Building the Linux kernel ............................................................................................... C-2 Creating the Linux filesystem ........................................................................................ C-6 Putting it together .......................................................................................................... C-8 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential vi Preface It is estimated that the number of mobile phones in the world will exceed the human population sometime in 2014. It is also estimated that over 90% of all these mobile devices have an ARM processor inside them. This book provides an introduction to ARM technology for programmers using ARM Cortex-A series processors conforming to the ARMv7–A architecture. The book is meant to complement rather than replace other ARM documentation available for Cortex-A series processors, such as the ARM Technical Reference Manuals (TRMs) for the processors themselves, documentation for individual devices or boards or, most importantly, the ARM® Architecture Reference Manual (the ARM ARM). The purpose of this book is to provide a single guide for programmers who want to develop applications for the Cortex-A series of processors, bringing together information from a wide variety of sources that will be useful to both assembly language and C programmers. Hardware concepts such as caches and Memory Management Units are covered, but only where this is valuable to the application writer. We will also look at the way operating systems such as Linux make use of ARM features, and how to take full advantage of the capabilities of the ARM processor, in particular writing software for multi-core processors. Although much of the content of this book is also applicable to older ARM processors, it does not explicitly cover processors that implement older versions of the Architecture. The Cortex-R series and M-series processors are mentioned but not described. Our intention is to provide an approachable introduction to the ARM architecture, covering the feature set in detail and providing practical advice on writing both C and assembly language programs to run efficiently on a Cortex-A series processor. This is not an introductory level book. It assumes some knowledge of the C programming language and microprocessors, but not of any ARM-specific background. In the allotted space, we cannot hope to cover every topic in detail. In some chapters, we suggest additional reading (referring either to books or web sites) that can give a deeper level of background to the topic in hand, but in this ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential vii Preface book we focus on the ARM-specific detail. We do not assume the use of any particular tool chain. We will mention both GNU and ARM tools in the course of the book. We hope that the book is suitable for programmers who have a desktop PC or x86 background and are taking their first steps into the ARM processor based world. The first chapters of the book cover the basic features of the ARM Cortex-A series processors. An introduction to the fundamentals of the ARM architecture, covering the various registers, and modes and some background on individual processors is provided in Chapter 2 and 3. Chapters 4 and 5 provide a brisk introduction to ARM assembly language programming, and assembly language instructions. We look at floating-point and the ARM Advanced SIMD extensions (NEON™) in Chapters 6 and 7. These chapters are only an introduction to the relevant topics. We then switch our focus to the memory system and look at Caches, Memory Management and Memory Ordering in Chapters 8, 9 and 10. Dealing with exceptions and interrupts are covered in Chapters 11 and 12. The remaining chapters of the book provide more advanced programming information. Chapter 13 provides an overview of boot code. Chapter 14 looks at issues with porting C and assembly code to the ARMv7 architecture, from other architectures and from older versions of the ARM architecture. Chapter 15 covers the Application Binary Interface, knowledge of which is useful to both C and assembly language programmers. Profiling and optimizing of code is covered in Chapters 16 and 17. Many of the techniques presented are not specific to the ARM architecture, but we also provide some processor-specific hints. Chapters 18 and 19 cover the area of multi-core processors. We take a detailed look at how these are implemented by ARM and how you can write code to take advantage of them. Power management is an important part of ARM programming and is covered in Chapter 20. The final chapters of the book provide a brief coverage of the ARM Security Extensions (TrustZone®) in Chapter 21, the ARM Virtualization extensions in Chapter 22, big.LITTLE™ technology in Chapter 23, and the hardware debug features available to programmers in Chapter 24. Appendix A gives a summary of the available ARM instructions. Appendix B gives a brief introduction to some of the tools and platforms available to those getting started with ARM programming, and Appendix C gives step-by-step instructions for configuring and building Linux for ARM systems. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential viii Preface Preface to the 4th Edition The ARM architecture continues to evolve. ARM recently announced the ARMv8 architecture, which, although not covered by this book, has influenced its content. Small changes have been made to the ARMv7 instruction set architecture for compatibility reasons. Those changes are included in this version. We have also taken advantage of the opportunity to carry out a general revision of the contents of the Guide. You will find some sections moved, some extensively rewritten, and some with minor changes. Chapters that covered the same areas have been combined. The chapter on Registers has been largely rewritten, to include the effects of the Security and Virtualization Extensions and the introduction of Privilege levels. The chapters on Exception Handling have been consolidated, so that they now tell a coherent story. A similar approach has been applied to multiprocessing, and parallelization. In all cases, what was two chapters is now one, with the contents updated. LPAE is now covered in the chapter on Virtualization. The chapter on Tools, Operating Systems and Boards has been moved to the Appendices, while the chapter on big.LITTLE technology has been extensively revised to keep pace with rapidly changing technology. ARM has added another processor to the ARMv7 Cortex-A series, and the new Cortex-A12 processor is covered in this new edition. You will also notice that what we call the devices has changed. Processor now refers to the marketed device, such as the Cortex-A15 processor, elsewhere you will find references to cores, and clusters (of cores). Some content that existed in the previous editions of the Cortex™-A Series Programmer’s Guide has been removed. You will notice that the NEON/VFP Appendix and the chapter on Writing NEON Code have been removed. We originally said that the information on NEON would need a book of its own. It now has one, in the form of the ARM® NEON™ Programmer’s Guide. Additional information has also been provided in the form of programming hints and tips, for developers using the Cortex-A series processors. While these cannot cover every eventuality that the developer might encounter we hope that the examples included will prove useful. The ARM® Cortex™-A Series Programmer’s Guide has proved to be a very popular addition to the ARM documentation set, and now also forms the reference textbook for the ARM Accredited Engineer (AAE) examinations. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential ix Preface Glossary Abbreviations and terms used in this document are defined here. ARM DEN0013D ID012214 AAPCS ARM Architecture Procedure Call Standard. ABI Application Binary Interface. ACP Accelerator Coherency Port. AEABI ARM Embedded ABI. AHB Advanced High-Performance Bus. AMBA® Advanced Microcontroller Bus Architecture. AMP Asymmetric Multi-Processing. APB Advanced Peripheral Bus. ARM ARM The ARM Architecture Reference Manual. ASIC Application Specific Integrated Circuit. APSR Application Program Status Register. ASID Address Space ID. ATPCS ARM Thumb® Procedure Call Standard. AXI Advanced eXtensible Interface. BE8 Byte Invariant Big-Endian Mode. BIU Bus Interface Unit. BSP Board Support Package. BTAC Branch Target Address Cache. BTB Branch Target Buffer. CISC Complex Instruction Set Computer. CP15 Coprocessor 15. System control coprocessor. CPSR Current Program Status Register. DAP Debug Access Port. DBX Direct Bytecode Execution. DDR Double Data Rate (SDRAM). DMA Direct Memory Access. DMB Data Memory Barrier. DPU Data Processing Unit. DS-5™ The ARM Development Studio. DSB Data Synchronization Barrier. DSP Digital Signal Processing. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential x Preface DSTREAM® An ARM debug and trace unit. DVFS Dynamic Voltage/Frequency Scaling. EABI Embedded ABI. ECC Error Correcting Code. ECT Embedded Cross Trigger. EOF End Of File. ETB Embedded Trace Buffer™. ETM Embedded Trace Macrocell™. FDT Flattened Device Tree. FIQ An interrupt type (formerly fast interrupt). FPSCR Floating-Point Status and Control Register. GCC GNU Compiler Collection. GIC Generic Interrupt Controller. GIF Graphics Interchange Format. GPIO General Purpose Input/Output. Gprof GNU profiler. Harvard architecture Architecture with physically separate storage and signal pathways for instructions and data. ARM DEN0013D ID012214 HCR Hyp Configuration Register. HMP Heterogenous Multi-Processing. ICU Instruction Cache Unit. IDE Integrated development environment. I/F Interface (abbreviation used in some diagrams). IPA Intermediate Physical Address. IRQ Interrupt Request (normally external interrupts). ISA Instruction Set Architecture. ISB Instruction Synchronization Barrier. ISR Interrupt Service Routine. Jazelle™ The ARM bytecode acceleration technology. JIT Just In Time. L1/L2 Level 1/Level 2. LPAE Large Physical Address Extension. LSB Least Significant Bit. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential xi Preface ARM DEN0013D ID012214 MESI A cache coherency protocol with four states; Modified, Exclusive, Shared and Invalid. MMU Memory Management Unit. MOESI A cache coherency protocol with five states; Modified, Owned, Exclusive, Shared and Invalid. MPU Memory Protection Unit. MSB Most Significant Bit. NEON™ The ARM Advanced SIMD Extensions. NMI Non-Maskable Interrupt. Normal world The execution environment when the core is in the Non-secure state. Oprofile A Linux system profiler. QEMU A processor emulator. PCI Peripheral Component Interconnect. A computer bus standard. PCS Procedure Call Standard. PFU Prefetch Unit. PIPT Physically Indexed, Physically Tagged. PLE Preload Engine. PLI Preload Instruction. PMU Performance Monitor Unit. PoC Point of Coherency. PoU Point of Unification. PPI Private Peripheral Input. Privilege The ability to perform certain tasks that cannot be done from User (Unprivileged) mode. PSCI Power State Coordination Interface. PSR Program Status Register. RCT Runtime Compiler Target. RISC Reduced Instruction Set Computer. RVCT RealView® Compilation Tools (the “ARM Compiler”). SBZP Should Be Preserved. SCU Snoop Control Unit. Secure world The execution environment when the core is in the Secure State. SGI Software Generated Interrupt. SIMD Single Instruction, Multiple Data. SiP System in Package. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential xii Preface ARM DEN0013D ID012214 SMP Symmetric Multi-Processing. SoC System on Chip. SP Stack Pointer. SPI Shared Peripheral Interrupt. SPSR Saved Program Status Register. Streamline A graphical performance analysis tool. SVC Supervisor Call instruction. (Previously SWI) SWI Software Interrupt instruction. (Replaced with SVC) SYS System Mode. TAP Test Access Port (JTAG Interface). TDMI® Thumb, Debug, Multiplier, ICE. TEX Type Extension. Thumb® An instruction set extension to ARM. Thumb-2 A technology extending the Thumb instruction set to support both 16-bit and 32-bit instructions. TLB Translation Lookaside Buffer. TLS Thread Local Storage. TrustZone The ARM security extension. TTB Translation Table Base. TTBR Translation Table Base Register. UAL Unified Assembly Language. UART Universal Asynchronous Receiver/Transmitter. UEFI Unified Extensible Firmware Interface. U-Boot A Linux Bootloader. UNK Unknown. USR User mode, a non-privileged processor mode. VFP The ARM floating-point instruction set. Before ARMv7, the VFP extension was called the Vector Floating-Point architecture, and was used for vector operations. VIC Vectored Interrupt Controller. VIPT Virtually Indexed, Physically Tagged. VMID Virtual Machine ID. VMSA Virtual Memory Systems Architecture. XN Execute Never. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential xiii Preface Typographical conventions This book uses the following typographical conventions: italic Highlights important notes, introduces special terminology, denotes internal cross-references, and citations. bold Used for terms in descriptive lists, where appropriate. monospace Denotes text that you can enter at the keyboard, such as commands, file and program names, instruction names, parameters and source code. monospace italic Denotes arguments to monospace text where the argument is to be replaced by a specific value. < and > Enclose replaceable terms for assembler syntax where they appear in code or code fragments. For example: MRC p15, 0,, , , ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential xiv Preface Feedback on this book We have tried to ensure that the Cortex-A Series Programmer’s Guide is both easy to read and still covers the material in enough depth to provide the comprehensive introduction to using the processors that we originally intended. If you have any comments on this book, don’t understand our explanations, think something is missing, or think that it is incorrect, send an e-mail to errata@arm.com. Give: • The title, The Cortex-A Series Programmer’s Guide. • The number, ARM DEN0013D. • The page number(s) to which your comments apply. • What you think needs to be changed. ARM also welcomes general suggestions for additions and improvements. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential xv Preface References Cohen, D. “On Holy Wars and a Plea for Peace”, USC/ISI IEN April, 1980 http://www.ietf.org/rfc/ien/ien137.txt. Furber, Steve. “ARM System-on-chip Architecture”, 2nd edition, Addison-Wesley, 2000, ISBN: 9780201675191. Hohl, William. “ARM Assembly Language: Fundamentals and Techniques” CRC Press, 2009. ISBN: 9781439806104. Sloss, Andrew N.; Symes, Dominic.; Wright, Chris. “ARM System Developer's Guide: Designing and Optimizing System Software”, Morgan Kaufmann, 2004, ISBN: 9781558608740. ANSI/IEEE Std 754-1985, “IEEE Standard for Binary Floating-Point Arithmetic”. ANSI/IEEE Std 754-2008, “IEEE Standard for Binary Floating-Point Arithmetic”. ANSI/IEEE Std 1003.1-1990, “Standard for Information Technology - Portable Operating System Interface (POSIX) Base Specifications, Issue 7”. ANSI/IEEE Std 1149.1-2001, “IEEE Standard Test Access Port and Boundary-Scan Architecture”. ANSI/IEEE Std 1275-1994, “IEEE Standard for Boot Firmware (Initialization Configuration) Firmware: Core Requirements and Practices”. The ARM® Architecture Reference Manual (known as the ARM ARM) is a must-read for any serious ARM programmer. It is available (after registration) from the ARM web site. It fully describes the ARMv7 instruction set architecture, programmer’s model, system registers, debug features and memory model. It forms a detailed specification to which all implementations of ARM processors must adhere. References to the ARM® Architecture Reference Manual in this document are to: ARM® Architecture Reference Manual - ARMv7-A and ARMv7-R edition (ARM DDI 0406). Note In the event of a contradiction between this book and the ARM ARM, the ARM ARM is definitive and must take precedence. In most instances, however, the ARM ARM and the Cortex-A Programmer’s Guide cover two separate world views. The ARM ARM is for processor implementers, while this book is for processor users. The most likely scenario is that this book will describe something in a way that does not cover all architecturally permitted behaviors, or rewords an abstract concept in more practical terms. ARM® Generic Interrupt Controller Architecture Specification version 2.0 (ARM IHI 0048). ARM® Compiler Toolchain Assembler Reference (DUI 0489). ARM® C Language Extensions (IHI 0053). ARM® NEON™ Programmer’s Guide (DEN 0018). Procedure Call Standard for the ARM® Architecture (ARM IHI 0042). Power State Coordination Interface (PSCI) Platform Design Document (ARM DEN 0022) ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential xvi Preface The individual processor Technical Reference Manuals provide a detailed description of the processor behavior. They can be obtained from the ARM web site documentation area http://infocenter.arm.com. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential xvii Chapter 1 Introduction You can find ARM processors everywhere. More than 10 billion ARM processor based devices had been manufactured by the end of 2008. At the end of 2013, over 52 billion ARM processors had been shipped. It is likely that readers of this book own at least one product containing ARM processor based devices – a mobile phone, tablet device, personal computer, television or even a car. It might come as a surprise to programmers more used to the personal computer to learn that the highly successful x86 architecture occupies a much smaller position in terms of total microprocessor shipments, with over three billion devices. The ARM architecture has advanced significantly since the first ARM1 silicon in 1985. ARM produces a whole family of processors that share common instruction sets and programmer’s models and have some degree of backward compatibility. Let’s begin, however, with a brief look at the history of ARM. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 1-1 Introduction 1.1 History The first ARM processor, the ARM1, was designed at Acorn Computers Limited by a team led by Sophie Wilson and Steve Furber, with the first silicon (which worked first time!) produced in April 1985. The ARM1 was quickly replaced by the ARM2, which added multiplier hardware, and was used in real systems, including the Acorn Archimedes personal computer. Photograph courtesy of The Centre for Computing History, www.computinghistory.org.uk Figure 1-1 Acorn Archimedes ARM as a separate company was formed in Cambridge, England in November 1990, as Advanced RISC Machines Ltd. It was a joint venture between Apple Computers, Acorn Computers and VLSI Technology and has outlived two of its parents. The original 12 employees came mainly from the team within Acorn Computers. One reason for spinning ARM off as a separate company was that the processor had been selected by Apple Computers for use in its Newton product. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 1-2 Introduction Figure 1-2 The Apple Newton The new company quickly decided that the best way forward for their technology was to license their Intellectual Property (IP). Instead of designing, manufacturing and selling the chips themselves, they would sell rights to their designs to semiconductor companies. These companies would design the ARM processor into their own products, in a partnership model. This IP licensing business is how ARM continues to operate today. ARM was able to sign up licensees, with Sharp, Texas Instruments and Samsung prominent names among the first customers. In 1998, ARM floated on the London Stock Exchange and Nasdaq. At the time of writing, ARM has over 2000 employees and has expanded somewhat from its original remit of processor design. ARM also licenses Physical IP – libraries of cells (NAND gates, RAM and so forth), graphics and video accelerators and software development products such as compilers, debuggers and development boards. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 1-3 Introduction 1.2 System-on-Chip (SoC) Designers today can assemble computer chips containing a billion or more transistors. Designing and verifying such complex circuits has become an extremely difficult task. It is increasingly rare for all of the parts of such systems to be designed by a single company. In response to this, ARM and other semiconductor IP companies design and verify components (so-called IP blocks or processors). These are licensed by semiconductor companies who use these blocks in their own designs and include microprocessors, DSPs, 3D graphics and video controllers, along with many other functions. The semiconductor companies take these blocks and include many other parts to create a complete system on the chip, forming a System-on-Chip (SoC). The architects must select the appropriate core(s), memory controllers, on-chip memory, peripherals, bus interconnect and other logic (perhaps including analog or radio frequency components), in order to produce a system. Application Specific Integrated Circuit (ASIC) is another term that we will use in the book. This is an Integrated Circuit design that is specific to a particular application. An ASIC might well contain an ARM core, memory and other components. Clearly there is a large overlap between ASICs and SoCs. The term SoC usually refers to a device with a higher degree of integration, including many of the parts of the system in a single device, possibly including analog, mixed-signal or radio frequency circuits. Semiconductor companies investing tens of millions of dollars to create these devices will normally invest a great deal in software to run on their platform. It would be uncommon to produce a complex system with a powerful processor without at least having ported one or more operating systems to it and written device drivers for peripherals. Of course, operating systems like Linux require significant amounts of memory to run, more than is usually possible on a single silicon device. The term System-on-Chip is therefore not always entirely accurate, because the SoC does not always contain the whole system. Apart from the issue of silicon area, it is also often the case that many useful parts of a system require specialist silicon manufacturing processes that preclude them from being placed on the same die. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 1-4 Introduction 1.3 Embedded systems An embedded system is conventionally defined as a piece of computer hardware running software designed to perform a specific task. Examples of such systems might be TV set-top boxes, smartcards, routers, disk drives, printers, automobile engine management systems, MP3 players or photocopiers. These contrast with what is generally considered a computer system, that is, one that runs a wide range of general purpose software and possesses input and output devices such as a keyboard, and a graphical display of some kind. This distinction is becoming increasingly blurred. Consider the cellular or mobile phone. A basic model might only perform the task of making phone calls, but modern smartphones can run a complex operating system to which many thousands of applications are available for download. Embedded systems can contain very simple 8-bit microprocessors, such as an Intel 8051 or PIC micro-controllers, or some of the more complex 32 or 64-bit processors, such as the ARM family described in this book. They require RAM and some form of non-volatile storage to hold the program(s) to be executed by the system. Systems will almost always have additional peripherals, relating to the actual function of the device – typically including UARTs, interrupt controllers, timers, GPIO controllers, but also potentially quite complex blocks such as GPUs, or DMA controllers. Software running on such systems is typically grouped into two separate parts: the Operating System (OS) and applications that run on top of the OS. In this book, we will concentrate mainly on examples from Linux. The source code for Linux is readily available for inspection by the reader and is likely to be familiar to many programmers. Nevertheless, lessons learned from Linux are equally applicable to other operating systems. Memory Footprint In many systems, to minimize cost (and power), memory size can be limited. You might be forced to consider the size of the program and how to reduce memory usage while it runs. Real-time behavior A feature of certain systems is that there are deadlines to respond to external events. This might be a “hard” requirement (a car braking system must respond within a certain time) or “soft” requirement (audio processing must complete within a certain time-frame to avoid a poor user experience – but failure to do so under rare circumstances might not render the system worthless). Power In many embedded systems the power source is a battery, and programmers and hardware designers must take great care to minimize the total energy usage of the system. This can be done, for example, by slowing the clock, reducing supply voltage or switching off the core when there is no work to be done. Cost Reducing the bill of materials can be a significant constraint on system design. Time to market In competitive markets, the time to develop a working product can significantly impact the success of that product. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 1-5 Chapter 2 ARM Architecture and Processors ARM does not manufacture silicon devices. Instead, ARM creates microprocessor designs, that are licensed to semiconductor companies and OEMs, who then integrate them into System-on-Chip devices. To ensure compatibility between implementations, ARM defines architecture specifications that define how compliant products must behave. Processors implementing the ARM architecture conform to a particular version of the architecture. The ARM architecture supports implementations across a very wide range of performance points. Its simplicity leads to very small implementations, and this enables very low power consumption. The Cortex-A series processors covered in this book conform to the ARMv7-A architecture. There might also be multiple processors with different internal implementations and micro-architectures, different cycle timings and clock speeds that conform to the same version of the architecture, in that they execute the ARM instruction set defined for the architecture and pass the ARM Validation System tests. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-1 ARM Architecture and Processors 2.1 Architectural profiles Periodically, new versions of the architecture are announced by ARM. These add new features or make changes to existing behaviors. Such changes are almost always backwards compatible, meaning that user code that ran on older versions of the architecture will continue to run correctly on new versions. Of course, code written to take advantage of new features will not run on older processors that lack these features. In all versions of the architecture, some system features and behaviors are left as implementation-defined. For example, the architecture does not define cache sizes or cycle timings for individual instructions. These are determined by the individual core and SoC. Each architecture version can also define optional extensions. These might be implemented in a particular implementation of a processor. For example, in the ARMv7 architecture, the Advanced SIMD (NEON) technology is available as an optional extension, and is described in Chapter 7 Introducing NEON. The ARMv7 architecture also has the concept of profiles. These are variants of the architecture describing processors targeting different markets and usages. The profiles are as follows: A The Application profile defines an architecture aimed at high performance processors, supporting a virtual memory system using a Memory Management Unit (MMU) and therefore capable of running fully featured operating systems. Support for the ARM and Thumb instruction sets is provided. ARMv7-A, the Application profile, is implemented by all Cortex-A series processors, and by processors developed by companies who have licensed the ARM architecture. At the beginning of 2014, just under three billion Cortex-A Series chips had been shipped. The ARMv8-A architecture, which is not described in this book, supports the AArch32 state, a 32-bit implementation of the architecture that is backwards compatible with ARMv7-A. ARM DEN0013D ID012214 R The Real-time profile defines an architecture aimed at systems that require deterministic timing and low interrupt latency. There is no support for a virtual memory system, but memory regions can be protected using a simple Memory Protection Unit (MPU). M The Microcontroller profile defines an architecture aimed at low cost systems, where low-latency interrupt processing is vital. It uses a different exception handling model to the other profiles and supports only a variant of the Thumb instruction set. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-2 ARM Architecture and Processors 2.2 Architecture history and extensions The ARM architecture changed relatively little between the first test silicon in the mid-1980s through to the first ARM6 and ARM7 devices of the early 1990s. In the first version of the architecture, the majority of the load, store and arithmetic operations along with the exception model and register set was implemented by the ARM1. Version 2 added multiply and multiply-accumulate instructions and support for coprocessors, plus other innovations. These early processors only supported 26-bits of address space. Version 3 of the architecture separated the program counter and program status registers and added several new modes, enabling support for 32-bits of address space. Version 4 adds support for halfword load and store operations and an additional kernel-level privilege mode. Readers unfamiliar with the ARM architecture should not worry if parts of this description use terms they have not met, because all of these topics are covered in the following chapters. The ARMv4T architecture, which introduced the Thumb (16-bit) instruction set, was implemented by the ARM7TDMI® and ARM9TDMI® processors, products that have shipped in their billions. The ARMv5TE architecture added improvements for DSP-type operations and saturated arithmetic and to ARM/Thumb interworking. ARMv6 made a number of enhancements, including support for unaligned memory accesses, significant changes to the memory architecture and for multi-core support, plus some support for SIMD operations operating on bytes or halfwords within the 32-bit registers. It also provided a number of optional extensions, notably Thumb-2 and Security Extensions (TrustZone). Thumb-2 extends Thumb to be a mixed length (16-bit and 32-bit) instruction set. The ARMv7-A architecture makes the Thumb-2 extensions mandatory and adds the Advanced SIMD extensions (NEON), described in Chapter 7. For a number of years, ARM adopted a sequential numbering system for processors with ARM9 following ARM8, that came after ARM7. Various numbers and letters were appended to the base family to denote different variants. For example, the ARM7TDMI processor has T for Thumb, D for Debug, M for a fast multiplier and I for EmbeddedICE. For the ARMv7 architecture, ARM Limited adopted the brand name Cortex for its processors, with a supplementary letter indicating which of the three profiles (A, R, or M) the processor supports. Figure 2-1 on page 2-4 shows how different versions of the architecture correspond to different processor implementations. The figure is not comprehensive and does not include all architecture versions or processor implementations. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-3 ARM Architecture and Processors Architecture v4 / v4T Architecture v5 ARM7TDMI ARM920T StrongARM ARM926EJ-S ARM946E-S XScale Architecture v6 ARMv7-A Cortex-A5 Cortex-A7 Cortex-A8 Cortex-A9 Cortex-A12 Cortex-A15 Architecture v7 ARM1136J-S ARM1176JZ-S ARM1156T2-S ARMv7-R Cortex-R4 Cortex-R5 Cortex-R7 ARMv7-M Cortex-M3 ARMv6-M Cortex-M0+ Cortex-M0 Cortex-M1 ARMv7E-M Cortex-M4 Figure 2-1 Architecture and processors In Figure 2-2, we show the development of the architecture over time, illustrating additions to the architecture at each new version. Almost all architecture changes are backwards-compatible, meaning software written for the ARMv4T architecture can still be used on ARMv7 processors. 4T 5 Halfword and signed halfword/byte support Improved ARM/Thumb Interworking System mode CLZ Thumb instruction set Saturated arithmetic DSP multiply-accumulate Instructions Extensions: Jazelle (v5TEJ) 6 7 SIMD instructions Thumb-2 technology Multi-processing NEON TrustZone Profiles: v7-A (Applications) NEON LPAE Virtualization v6 memory architecture Unaligned data support Extensions: Thumb-2 (v6T2) TrustZone (v6Z) Multiprocessor (v6K) Thumb only (v6-M) v7-R (Real-time) Hardware divide NEON v7-M (Microcontroller) Hardware divide Thumb only Floating-point Figure 2-2 Architecture history Individual chapters of this book will cover these topics in greater detail, but here we will briefly introduce a number of architecture elements. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-4 ARM Architecture and Processors 2.2.1 DSP multiply-accumulate and saturated arithmetic instructions These instructions, added in the ARMv5TE architecture, improve the capability for digital signal processing and multimedia software and are denoted by the letter E. The new instructions provide many variations of signed multiply-accumulate, saturated add and subtract, and count leading zeros and are present in all later versions of the architecture. In many cases, this made it possible to remove a simple separate DSP from the system. 2.2.2 Jazelle Jazelle-DBX (Direct Bytecode eXecution) was introduced in ARMv5TEJ to accelerate Java performance while conserving power. A combination of increased memory availability and improvements in Just-In-Time (JIT) compilers have subsequently reduced its value in application processors. For this reason, many ARMv7-A processors do not implement this hardware acceleration. Jazelle-DBX is best suited to providing high performance Java in systems with very limited memory (for example, feature phone or low-cost embedded use). In today’s systems, it is mainly used for backwards compatibility. 2.2.3 Thumb Execution Environment (ThumbEE) Introduced in ARMv7-A, ThumbEE is sometimes referred to as Jazelle-RCT (Runtime Compilation Target). It involves small changes to the Thumb instruction set that make it a better target for code generated at runtime in controlled environments, for example, by managed languages like Java, Dalvik, C#, Python or Perl. ThumbEE is designed to be used by Just-In-Time (JIT) or Ahead-Of-Time (AOT) compilers, where it can reduce the code size of recompiled code. The use of ThumbEE is now deprecated by ARM. 2.2.4 Thumb The ARMv7 architecture contains two main instruction sets, the ARM and Thumb instruction sets. Much of the functionality available is identical in the two instruction sets. The Thumb instruction set is a subset of the most commonly used 32-bit ARM instructions. Thumb instructions are each 16 bits long, and have a corresponding 32-bit ARM instruction that has the same effect. The main reason for using Thumb code is to reduce code density. Because of its improved density, Thumb code tends to cache better than the equivalent ARM code and can reduce the amount of memory required.. You can still use the ARM instruction set and you may want to, for particular code sections which require the highest performance. See The ARM instruction sets on page 4-3. 2.2.5 Thumb-2 Despite continued rumours to the contrary, there is no such thing as a Thumb-2 instruction set. Thumb-2 technology was introduced in ARMv6T2, and is required in ARMv7. This technology extends the original 16-bit Thumb instruction set to include 32-bit instructions. The range of 32-bit Thumb instructions included in ARMv6T2 permits Thumb code to achieve performance similar to ARM code, with code density better than that of the purely 16-bit Thumb code. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-5 ARM Architecture and Processors 2.2.6 Security Extensions (TrustZone) The optional Security Extensions referred to as TrustZone introduced with ARMv6K have been implemented in all ARM Cortex-A processors. TrustZone provides a separate Secure world to isolate sensitive code and data from the normal world that contains the operating system and applications. The software in the Secure world is therefore intended to provide security services to the Normal (non-secure) world applications. TrustZone is described in Chapter 21. 2.2.7 VFP Before ARMv7, the VFP extension was called the Vector Floating-Point Architecture, and supported vector operations. VFP is an extension that implements single-precision and optionally, double-precision floating-point arithmetic, compliant with the ANSI/IEEE Standard for Floating-Point Arithmetic. 2.2.8 Advanced SIMD (NEON) The ARM NEON technology provides an implementation of the Advanced SIMD instruction set, with separate register files (shared with VFP). Some implementations have a separate NEON pipeline back-end. It supports 8, 16, 32 and 64-bit integer, and single-precision (32-bit) floating-point data, that can be operated on as vectors in 64-bit and 128-bit registers. NEON is described in Chapter 7, and in the ARM® NEON™ Programmer’s Guide. 2.2.9 Coprocessors The ARM architecture supports a way of extending the instruction set by using Coprocessors, to extend the functionality of an ARM processor. There are 16 coprocessors with numbers from 0 to 15, On earlier ARM cores, a dedicated hardware interface was provided to permit external coprocessors to be connected. On the Cortex-A series processors, only internal coprocessors are supported. These are CP15 (system control for such things as cache and MMU, CP14 for debug, and CP10 and 11 for NEON and VFP operations. Coprocessors are described in Registers on page 3-6. Coprocessor operations are described in Miscellaneous instructions on page 5-17. 2.2.10 Large Physical Address Extension (LPAE) LPAE is optional in the v7-A architecture and is presently supported by the Cortex-A7, Cortex-A12, and Cortex-A15 processors. It enables 32-bit processors that are normally limited to addressing a maximum of 4GB of address space to access up to 1TB of address space by translating 32-bit virtual memory addresses into 40-bit physical memory addresses. See Large Physical Address Extensions on page 22-10. 2.2.11 Virtualization The ARM Virtualization Extensions are optional extensions to the ARMv7-A architecture profile. The extensions support the use of a virtual machine monitor, known as the hypervisor, to switch from one operating system to another. When implemented in a single core or in a multi-core system, the Virtualization Extensions support running multiple virtual machines on a single cluster. See Chapter 22. 2.2.12 big.LITTLE big.LITTLE processing was introduced in ARMv7 to enable devices to balance the requirements for processing power and power efficiency. big.LITTLE uses a high performance cluster, such as the Cortex-A15 processor, coupled with an energy efficient cluster, such as the ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-6 ARM Architecture and Processors Cortex-A7 processor. The “big” cluster can be utilized for heavy workloads, while the “LITTLE” cluster can take over for the majority of mobile device workloads. big.LITTLE is described in Chapter 23. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-7 ARM Architecture and Processors 2.3 Processor properties In this section, we consider some ARM processors and identify which processor implements which architecture version. In Cortex-A series processors on page 2-10 we take a slightly more detailed look at some of the individual processors that implement architecture version v7-A. Some terminology will be used in this chapter that might be unfamiliar to the first-time user of ARM processors and will not be explained until later in the book. Table 2-1 lists the architecture version implemented by a number of older ARM processors. Table 2-1 Older ARM processors and architectures Architecture version Applications processor Embedded processor v4T ARM720T™ ARM920T™ ARM922T™ ARM7TDMI™ v5TE - ARM946E-S™ ARM966E-S™ ARM968E-S v5TEJ ARM926EJ-S™ - v6K ARM1136J(F)-S™ ARM11™ MPCore™ - v6T2 - ARM1156T2-S™ v6K + security extensions ARM1176JZ(F)-S™ - Table 2-2 lists the architecture version implemented by the Cortex family of processors. Table 2-2 Cortex processors and architecture versions v7-A (Applications) v7-R (Real Time) v6-M/v7-M (Microcontroller) Cortex-A5 (Single/MP) Cortex-R4 Cortex-M0+ (ARMv6-M) Cortex-A7 (MP) Cortex-R5 Cortex-M0 (ARMv6-M) Cortex-A8 (Single) Cortex-R7 Cortex-M1™ (ARMv6-M) Cortex-A9 (Single/MP) Cortex-M3™ (ARMv7-M) Cortex-A12 (MP) Cortex-M4(F) (ARMv7E-M) Cortex-A15 (MP) Table 2-3 on page 2-9 compares the properties of Cortex-A series processors. For processor cache information, see Table 8-1 on page 8-11. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-8 ARM Architecture and Processors Table 2-3 Some properties of Cortex-A series processors Processor ARM DEN0013D ID012214 Cortex-A5 Cortex-A7 Cortex-A8 Cortex-A9 Cortex-A12 Cortex-A15 Release date Dec 2009 Oct 2011 July 2006 March 2008 June 2013 April 2011 Typical clock speed ~1GHz ~1GHz on 28nm ~1GHz on 65nm ~2GHz on 40nm ~2GHz on 28nm ~2.5GHz on 28nm Execution order In-order In-order In-order Out of order Out of order Out of order Cores 1 to 4 1 to 4 1 1 to 4 1 to 4 1 to 4 Peak integer throughput 1.6DMIPS/ MHz 1.9DMIPS/ MHz 2DMIPS/MHz 2.5DMIPS/M Hz 3.0DMIPS/ MHz 3.5DMIPS/ MHz VFP architecture VFPv4 VFPv4 VFPv3 VFPv3 VFPv4 VFPv4 NEON architecture NEON NEONv2 NEON NEON NEONv2 NEONv2 Half precision extension Yes Yes No Yes Yes Yes Hardware Divide No Yes No No Yes Yes Fused Multiply Accumulate Yes Yes No No Yes Yes Pipeline stages 8 8 13 9 to 12 11 15+ Instructions decoded per cycle 1 Partial dual issue 2 (Superscalar) 2 (Superscalar) 2 (Superscalar) 3 (Superscalar) Return stack entries 4 8 8 8 8 48 LPAE No Yes No No Yes Yes Floating Point Unit Optional Yes Yes Optional Yes Optional AMBA interface 64-bit AMBA 3 128-bit AMBA 4 64 or 128-bit AMBA 3 2× 64-bit AMBA 3 128-bit AMBA 4 128-bit AMBA 4 Generic Interrupt Controller (GIC) Included Optional Not included Included External Optional Trace Optional ETM Optional ETM separate macrocell Integrated ETM Integrated PTM Integrated PTM Integrated PTM Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-9 ARM Architecture and Processors 2.4 Cortex-A series processors In this section, we take a closer look at each of the processors that implement the ARMv7-A architecture. Only a general description is given in each case, for more specific information on each processor, see Table 2-3 on page 2-9 and Table 8-1 on page 8-11. 2.4.1 The Cortex-A5 processor The Cortex-A5 processor is the smallest ARM multi-core applications processor. Devices based on this processor are typically low-cost, capable of delivering the internet to the widest possible range of devices from low-cost entry-level smartphones and smart mobile devices, to embedded, consumer and industrial devices. ARM CoreSight Multicore Debug and Trace Generic Interrupt Controller NEON Data Engine Cortex-A5 processor Floating-point unit Instruction Cache 4 2 Data Cache Core SCU 1 3 ACP Dual 64-bit AMBA3 AXI Figure 2-3 Cortex-A5 processor The Cortex-A5 processor has the following features: ARM DEN0013D ID012214 • Full application compatibility with other Cortex-A series processors. • Multiprocessing capability for scalable, energy efficient performance. • Optional Floating-point or NEON units for media and signal processing. • High-performance memory system including caches and memory management unit. • High value migration path from older ARM processors. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-10 ARM Architecture and Processors 2.4.2 The Cortex-A7 processor The ARM Cortex-A7 processor is the most energy efficient application processor developed by ARM and extends ARM’s low-power leadership in entry level smart phones, tablets and other advanced mobile devices. ARM CoreSight Multicore Debug and Trace Generic Interrupt Controller NEON Data Engine Cortex-A7 processor Floating-point unit Instruction Cache 4 Data Cache Core SCU 1 2 3 L2 Cache W/ECC 128-bit AMBA ACE Coherent Bus Interface Figure 2-4 Cortex-A7 processor The Cortex-A7 processor has the following features: 2.4.3 • Architecture and feature set identical to the Cortex-A15 processor, enabling big.LITTLE configuration. • Less than 0.5mm2, using 28nm process technology. • Full application compatibility with all Cortex-A series processors. • Tightly-coupled low latency level 2 cache (up to 4MB). • Floating-point unit. • NEON technology for multimedia and SIMD processing. The Cortex-A8 processor The ARM Cortex-A8 processor, has the ability to scale in speed from 600MHz to greater than 1GHz. The Cortex-A8 processor can meet the requirements for power-optimized mobile devices needing operation in less than 300mW; and performance-optimized consumer applications requiring 2000 Dhrystone MIPS. It is available in a number of different devices, including the S5PC100 from Samsung, the OMAP3530 from Texas Instruments and the ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-11 ARM Architecture and Processors i.MX515 from Freescale. From high-end feature phones to netbooks, DTVs, printers and automotive-infotainment, the Cortex-A8 processor offers a proven high-performance solution with millions of units shipped annually ARM CoreSight Multicore Debug and Trace NEON Data Engine Cortex-A8 processor Dynamic branch prediction L1 Instruction Cache L1 Data Cache Floating-point unit Integrated L2 Cache 64- or 128-bit AMBA3 Bus Interface Figure 2-5 Cortex-A8 processor The Cortex-A8 processor has the following features: 2.4.4 • Frequency from 600MHz to more than 1GHz. • High performance superscalar architecture. • NEON technology for multi-media and SIMD processing. • Compatibility with older ARM processors. The Cortex-A9 processor The ARM Cortex-A9 processor is a power-efficient and popular high performance choice in low power or thermally constrained cost-sensitive devices. It is currently shipping in large volumes for smartphones, digital TV, consumer and enterprise applications. The Cortex-A9 processor provides an increase in performance of greater than 50% compared to the Cortex-A8 processor. The Cortex-A9 processor can be configured with up to four cores delivering peak performance when required. Configurability and flexibility makes the Cortex-A9 processor suitable for wide variety of markets and applications. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-12 ARM Architecture and Processors ARM CoreSight Multicore Debug and Trace Generic Interrupt Controller NEON Data Engine Cortex-A9 processor Floating-point unit L1 Instruction Cache 4 L1 Data Cache Core SCU 1 2 3 ACP Dual 64-bit AMBA3 AXI Figure 2-6 Cortex-A9 processor Devices containing the Cortex-A9 processor include nVidia’s dual-core Tegra-2, the SPEAr1300 from ST and TI’s OMAP4 platform. The Cortex-A9 processor has the following features: 2.4.5 • Out-of-order speculating pipeline. • 16, 32 or 64KB four way associative L1 caches. • Floating-point unit. • NEON technology for multi-media and SIMD processing. • Available as a speed or power optimized hard macro implementation. The Cortex-A12 processor The Cortex-A12 processor is a high performance mid-range mobile processing solution designed for mobile applications, for example, use in smartphones and tablet devices. The Cortex-A12 processor is a successor to the highly successful Cortex-A9 processor and is optimized for highest performance in the mainstream mobile power envelope leading to best-in-class efficiency. The high performance and high-end feature set of the Cortex-A12 processor is suitable for many use cases. Mid-range devices can build on the success of high-end devices and continue driving the fastest growing market segment in mobile. Architecturally, the Cortex-A12 processor is based on the latest ARMv7-A architecture and features extensions that are aligned with processors such as the Cortex-A15 processor. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-13 ARM Architecture and Processors ARM CoreSight Multicore Debug and Trace NEON Data Engine Cortex-A12 processor Floating-point unit L1 Instruction Cache L1 Data Cache Core SCU ACP 4 Integrated L2 cache w/ECC 1 2 3 Peripheral Port Dual 64-bit AMBA3 AXI Figure 2-7 Cortex-A12 processor The Cortex-A12 processor has the following features: 2.4.6 • 40-bit Large Physical Address Extensions (LPAE) addressing up to 1 TB of RAM. • Full application compatibility with all Cortex-A series processors. • NEON technology for multi-media and SIMD processing. • Virtualization and TrustZone security technology The Cortex-A15 processor The ARM Cortex-A15 processor is designed to deliver unprecedented flexibility and processing capability. This processor is designed with advanced power reduction techniques, and enables products in a wide range of markets ranging from mobile computing, high-end digital home, servers and wireless infrastructure. The Cortex-A15 MPCore processor has full application compatibility with all other Cortex-A series processors. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-14 ARM Architecture and Processors ARM CoreSight Multicore Debug and Trace Generic Interrupt Controller NEON Data Engine Cortex-A15 processor Floating-point unit L1 Instruction Cache 4 L1 Data Cache Core SCU ACP 1 2 3 Integrated L2 cache w/ECC 128-bit AMBA ACE Coherent Bus Interface Figure 2-8 Cortex-A15 processor The Cortex-A15 processor has the following features: ARM DEN0013D ID012214 • Highly scalable, up to 2.5GHz performance. • Full application compatibility with all Cortex-A series processors. • Out-of-order superscalar processor. • Tightly coupled low-latency level 2 cache (up to 4MB). • Floating-point unit. • NEON technology for multi-media and SIMD processing. • Available as a quad-core hard macro implementation. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-15 ARM Architecture and Processors 2.5 Key architectural points of ARM Cortex-A series processors A number of key points are common to the Cortex-A family of devices: • 32-bit RISC core, with 16 × 32-bit visible registers with mode-based register banking. • Modified Harvard Architecture (separate, concurrent access to instructions and data). • Load/Store Architecture. • Thumb-2 technology as standard. • VFP and NEON options. • Backward compatibility with code from previous ARM processors. • 4GB of virtual address space and a minimum of 4GB of physical address space. • Hardware translation table walking for virtual to physical address translation. • Virtual page sizes of 4KB, 64KB, 1MB and 16MB. Cacheability attributes and access permissions can be set on a per-page basis. • Big-endian and little-endian data access support. • Unaligned access support for basic load/store instructions. • Symmetric Multi-processing (SMP) support on MPCore™ variants, that is, multi-core versions of the Cortex-A series processors, with full data coherency at the L1 cache level. Automatic cache and Translation Lookaside Buffer (TLB) maintenance propagation provides high efficiency SMP operation. • Physically indexed, physically tagged (PIPT) data caches. See Virtual and physical tags and indexes on page 8-11. All of these architectural points are described in the chapters which follow. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 2-16 Chapter 3 ARM Processor Modes and Registers The ARM architecture is a modal architecture. Before the introduction of Security Extensions it had seven processor modes, summarized in Table 3-1. There were six privileged modes and a non-privileged user mode. Privilege is the ability to perform certain tasks that cannot be done from User (Unprivileged) mode. In User mode, there are limitations on operations that affect overall system configuration, for example, MMU configuration and cache operations. Modes are associated with exception events, which are described in Chapter 11 Exception Handling. Table 3-1 ARM processor modes before ARMv6 ARM DEN0013D ID012214 Mode Function Privilege User (USR) Mode in which most programs and applications run Unprivileged FIQ Entered on an FIQ interrupt exception IRQ Entered on an IRQ interrupt exception Supervisor (SVC) Entered on reset or when a Supervisor Call instruction (SVC) is executed Abort (ABT) Entered on a memory access exception Undef (UND) Entered when an undefined instruction executed System (SYS) Mode in which the OS runs, sharing the register view with User mode Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential Privileged 3-1 ARM Processor Modes and Registers The introduction of the TrustZone Security Extensions (see Figure 2-2 on page 2-4) created two security states for the processor that are independent of Privilege and processor mode, with a new Monitor mode to act as a gateway between the Secure and Non-secure states and modes existing independently for each security state. Monitor Non-secure Privileged Secure Privileged Non-secure User Secure User Normal world Secure world Figure 3-1 Secure and Non-secure worlds The TrustZone Security Extensions are described in Chapter 21, but for the present, for processors that implement the TrustZone extension, system security is achieved by dividing all of the hardware and software resources for the device so that they exist in either the Secure world for the security subsystem, or the Normal (Non-secure) world for everything else. When a processor is in the Non-secure state, it cannot access the memory that is allocated for Secure state. In this situation the Secure Monitor acts as a gateway for moving between these two worlds. If Security Extensions are implemented, software executing in Monitor mode controls transition between Secure and Non-secure processor states. The ARMv7-A architecture Virtualization Extensions add a hypervisor mode (Hyp), in addition to the existing privileged modes. Virtualization enables more than one Operating System to co-exist and operate on the same system. The ARM Virtualization Extensions therefore makes it possible to operate multiple Operating Systems on the same platform. Normal world Secure world Application Application Guest OS Guest OS Trusted services Trusted OS Hypervisor Secure monitor Figure 3-2 The hypervisor ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-2 ARM Processor Modes and Registers If the Virtualization Extensions are implemented there is a privilege model different to that of previous architectures. In Non-secure state there can be three privilege levels, PL0, PL1 and PL2. PL0 The privilege level of application software, that executes in User mode. Software executed in User mode is described as unprivileged software. This software cannot access some features of the architecture. In particular, it cannot change many of the configuration settings. Software executing at PL0 can make only unprivileged memory accesses. PL1 Software execution in all modes other than User mode and Hyp mode is at PL1. Normally, operating system software executes at PL1. The PL1 modes refers to all the modes other than User mode and Hyp mode. An Operating System is expected to execute across all PL1 modes and its applications executing in PL0 (User Mode). PL2 Hyp mode is normally used by a hypervisor, that controls, and can switch between Guest Operating Systems that execute at PL1. If Virtualization Extensions are implemented, a hypervisor will execute at PL2 (Hyp mode). A hypervisor will control and enable multiple Operating Systems to co-exist and execute on the same processor system. These privilege levels are separate from the TrustZone Secure and Normal (Non-secure) settings. Note The privilege level defines the ability to access resources in the current security state, and does not imply anything about the ability to access resources in the other security state. Figure 3-3 on page 3-4 shows the available processor modes under different processor states. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-3 ARM Processor Modes and Registers Non-secure state Secure state Non-secure PL0 USER mode Secure PL0 USER mode Non-secure PL1 Secure PL1 System mode(SYS) Supervisor mode (SVC) FIQ mode IRQ mode Undef (UND) mode Abort (ABT) mode System mode(SYS) Supervisor mode (SVC) FIQ mode IRQ mode Undef (UND) mode Abort (ABT) mode Non-secure PL2 Hyp mode Secure PL1 Monitor mode (MON) Figure 3-3 Privilege levels The presence of particular processor modes and states depends on whether the processor implements the relevant architecture extension, as shown in Table 3-2. Table 3-2 ARMv7 processor modes Mode Encoding Function Security state Privilege level User (USR) 10000 Unprivileged mode in which most applications run Both PL0 FIQ 10001 Entered on an FIQ interrupt exception Both PL1 IRQ 10010 Entered on an IRQ interrupt exception Both PL1 Supervisor (SVC) 10011 Entered on reset or when a Supervisor Call instruction (SVC) is executed Both PL1 Monitor (MON) 10110 Implemented with Security Extensions. See Chapter 21 Secure only PL1 Abort (ABT) 10111 Entered on a memory access exception Both PL1 Hyp (HYP) 11010 Implemented with Virtualization Extensions. See Chapter 22 Non-secure PL2 Undef (UND) 11011 Entered when an undefined instruction executed Both PL1 System (SYS) 11111 Privileged mode, sharing the register view with User mode Both PL1 A general purpose Operating System, such as Linux, and its applications, are expected to run in Non-secure state. The Secure state is expected to be occupied by vendor-specific firmware, or security-sensitive software. In some cases, software running in the Secure state is even more privileged than that running in Non-secure. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-4 ARM Processor Modes and Registers The current processor mode and execution state is contained in the Current Program Status Register (CPSR). Changing processor state and modes can be either explicit by privileged software, or as a result of taking exceptions link to relevant section. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-5 ARM Processor Modes and Registers 3.1 Registers The ARM architecture provides sixteen 32-bit general purpose registers (R0-R15) for software use. Fifteen of them (R0-R14) can be used for general purpose data storage, while R15 is the program counter whose value is altered as the core executes instructions. An explicit write to R15 by software will alter program flow. Software can also access the CPSR, and a saved copy of the CPSR from the previously executed mode, called the Saved Program Status Register (SPSR). R0 R1 R2 R3 R4 Low Registers R5 R6 R7 R8 R9 R10 R11 R12 High Registers Stack pointer R13 (SP) Link register R14 (LR) Program Counter R15 (PC) CPSR Current Program Status Register Figure 3-4 Programmer visible registers for user code Although software can access the registers, depending on which mode the software is executing in and the register being accessed, a register might correspond to a different physical storage location. This is called banking, the shaded registers in Figure 3-5 on page 3-7 are banked. They use physically distinct storage and are usually accessible only when a process is executing in that particular mode. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-6 ARM Processor Modes and Registers R0 R0 R0 R0 R0 R0 R0 R0 R0 R1 R1 R1 R1 R1 R1 R1 R1 R1 R2 R2 R2 R2 R2 R2 R2 R2 R2 R3 R3 R3 R3 R3 R3 R3 R3 R3 R4 R4 R4 R4 R4 R4 R4 R4 R4 R5 R5 R5 R5 R5 R5 R5 R5 R5 R6 R6 R6 R6 R6 R6 R6 R6 R6 R7 R7 R7 R7 R7 R7 R7 R7 R7 R8 R8 R8 R8 R8 R8 R8 R8 R8_fiq R9 R9 R9_fiq R9 R9 R9 R9 R9 R9 R10 R10 R10_fiq R10 R10 R10 R10 R10 R10 R11 R11 R11_fiq R11 R11 R11 R11 R11 R11 R12 R12_fiq R12 R12 R12 R12 R12 R12 SP_hyp R12 R13 (sp) R13 (sp) SP_fiq SP_svc SP_abt SP_svc SP_und SP_mon R14 (lr) R14 (lr) LR_fiq LR_svc LR_abt LR_svc LR_und LR_mon R15 (pc) R15 (pc) R15 (pc) R15 (pc) (A/C)PSR CPSR CPSR CPSR R15 (pc) R15 (pc) CPSR CPSR R15 (pc) CPSR R14 (lr) R15 (pc) R15 (pc) CPSR CPSR SPSR_fiq SPSR_irq SPSR_abt SPSR_svc SPSR_und SPSR_mon SPSR_hyp ELR_hyp User Sys FIQ IRQ ABT SVC UND MON HYP Banked Figure 3-5 The ARM register set In all modes, 'Low Registers' and R15 share the same physical storage location. Figure 3-5 shows that some 'High Registers' are banked for certain modes. For example, R8-R12 are banked for FIQ mode, that is, accesses to them go to a different physical storage location. For all modes other than User and System modes, R13 and the SPSRs are banked. In the case of banked registers, software does not normally specify which instance of the register is to be accessed, this is implied by the mode from which the access is made. For example, a program executing in User mode which specifies R13 will access R13_usr. A program executing in SVC mode which specifies R13 will access R13_svc. R13 (in all modes) is the OS stack pointer, but it can be used as a general purpose register when not required for stack operations. R14 (the Link Register) holds the return address from a subroutine entered when you use the branch with link (BL) instruction. It too can be used as a general purpose register when it is not supporting returns from subroutines. R14_svc, R14_irq, R14_fiq, R14_abt and R14_und are used similarly to hold the return values of R15 when interrupts and exceptions arise, or when Branch and Link instructions are executed within interrupt or exception routines. R15 is the program counter and holds the current program address (actually, it always points eight bytes ahead of the current instruction in ARM state and four bytes ahead of the current instruction in Thumb state, a legacy of the three stage pipeline of the original ARM1 processor). When R15 is read in ARM state, bits [1:0] are zero and bits [31:2] contain the PC. In Thumb state, bit [0] always reads as zero. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-7 ARM Processor Modes and Registers The reset values of R0-R14 are unpredictable. SP, the stack pointer, must be initialized (for each mode) by boot code before making use of the stack. The ARM Architecture Procedure Call Standard (AAPCS) or ARM Embedded ABI (AEABI) (see Chapter 15 Application Binary Interfaces) specifies how software should use the general purpose registers in order to interoperate between different toolchains or programming languages. 3.1.1 Hypervisor mode Implementations that support the Virtualization Extensions have additional registers available in Hypervisor (Hyp) mode. Hypervisor mode operates at the PL2 level of privilege. It has access to its own versions of R13 (SP) and SPSR. It uses the User mode link register for storing function return addresses, and has a dedicated register, called ELR_hyp, to store the exception return address. Hyp mode is available only in the Normal world and provides facilities for virtualization, that are only accessible in this mode. See Chapter 22 Virtualization, for more information. 3.1.2 Program Status Registers At any given moment, you have access to 16 registers (R0-R15) and the Current Program Status Register (CPSR). In User mode, a restricted form of the CPSR called the Application Program Status Register (APSR) is accessed instead. The Current Program Status Register (CPSR) is used to store: • The APSR flags. • The current processor mode. • Interrupt disable flags. • The current processor state, that is, ARM, Thumb, ThumbEE, or Jazelle. • The endianness. • Execution state bits for the IT block. The Program Status Registers (PSRs) form an additional set of banked registers. Each exception mode has its own Saved Program Status Register (SPSR) where a copy of the pre-exception CPSR is stored automatically when an exception occurs. These are not accessible from User modes. Application programmers must use the APSR to access the parts of the CPSR that can be changed in unprivileged mode. The APSR must be used only to access the N, Z, C, V, Q, and GE[3:0] bits. These bits are not normally accessed directly, but instead set by condition code setting instructions and tested by instructions that are executed conditionally. For example, the CMP R0, R1 instruction compares the values of R0 and R1 and sets the zero flag (Z) if R0 and R1 are equal. Figure 3-6 shows the bit assignments in the CPSR. 31 27 26 25 24 23 N Z C V Q IT [1:0] J 20 19 Reserved 16 15 GE[3:0] 10 IT[7:2] 9 8 7 6 5 E A I F T 4 0 M[4:0] Figure 3-6 CPSR bits ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-8 ARM Processor Modes and Registers The individual bits represent the following: • N – Negative result from ALU. • Z – Zero result from ALU. • C – ALU operation Carry out. • V – ALU operation oVerflowed. • Q – cumulative saturation (also described as sticky). • J – indicates whether the core is in Jazelle state. • GE[3:0] – used by some SIMD instructions. • IT [7:2] – If-Then conditional execution of Thumb-2 instruction groups. • E bit controls load/store endianness. • A bit disables asynchronous aborts. • I bit disables IRQ. • F bit disables FIQ. • T bit – indicates whether the core is in Thumb state. • M[4:0] – specifies the processor mode (FIQ, IRQ, as described in Table 3-1 on page 3-1). The core can change between modes using instructions that directly write to the CPSR mode bits. More commonly, the processor changes mode automatically as a result of exception events. In User mode, you cannot manipulate the PSR bits [4:0] that control the processor mode or the A, I and F bits that govern the exceptions to be enabled or disabled. We will consider these bits in more detail in Chapter 5 and Chapter 11. 3.1.3 Coprocessor 15 CP15, the System Control coprocessor, provides control of many features of the core. It can contain up to sixteen 32-bit primary registers. Access to CP15 is privilege controlled and not all registers are available in User mode. The CP15 register access instructions specify the required primary register, with the other fields in the instruction used to define the access more precisely and increase the number of physical 32-bit registers in CP15. The 16 primary registers in CP15 are named c0 to c15, but are often referred to by name. For example, the CP15 System Control Register is called CP15.SCTLR. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-9 ARM Processor Modes and Registers Table 3-3 summarizes the function of some of the more relevant coprocessor registers used in Cortex-A series processors. We will consider some of these in more detail when we look at the operation of the cache and MMU. Table 3-3 CP15 Register summary Register Description Main ID Register (MIDR) Gives identification information for the processor (including part number and revision). Multiprocessor Affinity Register (MPIDR) Provides a way to uniquely identify individual cores within a cluster. CP15 c1 System Control registers System Control Register (SCTLR) The main processor control register (see System control register (SCTLR) on page 3-12). Auxiliary Control Register (ACTLR) IMPLEMENTATION DEFINED. Implementation specific additional control and configuration options. Coprocessor Access Control Register (CPACR) Controls access to all coprocessors except CP14 and CP15. Secure Configuration Register (SCR) Used by TrustZone (Chapter 21). CP15 c2 and c3, memory protection and control registers Translation Table Base Register 0 (TTBR0) Base address of level 1 translation table (see Chapter 9). Translation Table Base Register 1 (TTBR1) Base address of level 1 translation table (see Chapter 9). Translation Table Base Control Register (TTBCR) Controls use of TTB0 and TTB1 (see Chapter 9). CP15 c5 and c6, memory system fault registers Data Fault Status Register (DFSR) Gives status information about the last data fault (see Chapter 11). Instruction Fault Status Register (IFSR) Gives status information about the last instruction fault (see Chapter 11). Data Fault Address Register (DFAR) Gives the virtual address of the access that caused the most recent precise data abort (see Chapter 11). Instruction Fault Address Register (IFAR) Gives the virtual address of the access that caused the most recent precise prefetch abort (see Chapter 11). CP15 c7, cache maintenance and other functions Cache and branch predictor maintenance functions See Chapter 8. Data and instruction barrier operations See Chapter 10. CP15 c8, TLB maintenance operations ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-10 ARM Processor Modes and Registers Table 3-3 CP15 Register summary (continued) Register Description CP15 c9, performance monitors CP15 c12, Security Extensions registers Vector Base Address Register (VBAR) Provides the exception base address for exceptions that are not handled in Monitor mode. Monitor Vector Base Address Register (MVBAR) Holds the exception base address for all exceptions that are taken to Monitor mode. CP15 c13, process, context and thread ID registers Context ID Register (CONTEXTIDR) See description of ASID Chapter 9). Software thread ID registers See description of ASID Chapter 9). CP15 c15, IMPLEMENTATION DEFINED registers Configuration Base Address Register (CBAR) Provides a base address for the GIC and Local timer type peripherals. All system architecture functions are controlled by reading or writing a general purpose processor register (Rt) from or to a set of registers (CRn) located within Coprocessor 15. The Op1, Op2, and CRm fields of the instruction can also be used to select registers or operations. The format is shown in Example 3-1. Example 3-1 CP15 Instruction syntax MRC p15, Op1, Rt, CRn, CRm, Op2 ; read a CP15 register into an ARM register MCR p15, Op1, Rt, CRn, CRm, Op2 ; write a CP15 register from an ARM register We will not go through each of the various CP15 registers in detail, because this would duplicate reference information that can readily be obtained from the ARM Architecture Reference Manual or product documentation. AS an example of how we can read information from one of these registers, we will consider the read-only Main ID Register (MIDR), the format is shown in Figure 3-7. 31 24 23 Implementer 20 19 Variant 16 15 Arch 4 3 Primary part number 0 Rev Figure 3-7 Main ID Register format In a privileged mode, we can read this register, using the instruction: MRC p15, 0, R1, c0, c0, 0 The result, placed in register R1, tells software the processor it is running on. For an ARM Cortex processor the interpretation of the results is as follows: • ARM DEN0013D ID012214 Bits [31:24] – implementer, will be 0x41 for an ARM designed processor. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-11 ARM Processor Modes and Registers 3.1.4 • Bits [23:20] – variant, shows the revision number of the processor. • Bits [19:16] – architecture, will be 0xF for ARM architecture v7. • Bits [15:4] – part number, for example 0xC08 for the Cortex-A8 processor. • Bits [3:0] – revision, shows the patch revision number of the processor. System control register (SCTLR) The SCTLR is another of a number of registers that are accessed using CP15, and controls standard memory, system facilities and provides status information for functions implemented in the core. The System Control Register is only accessible from PL1 or higher. 31 30 29 28 27 26 25 24 22 21 20 19 18 U TE NMFI EE 14 13 12 11 10 9 3 2 1 0 V I Z C A M FI Figure 3-8 Simplified view of the system control register The individual bits represent the following: • TE – Thumb exception enable. This controls whether exceptions are taken in ARM or Thumb state. • NMFI – Non-maskable FIQ (NMFI) support. See External interrupt requests on page 12-2. • EE = Exception endianness. This defines the value of the CPSR.E bit on entry to an exception. See Endianness on page 14-2. • U – Indicates use of the alignment model. See Alignment on page 14-5. • FI – FIQ configuration enable. See External interrupt requests on page 12-2 • V – This bit selects the base address of the exception vector table. See The Vector table on page 11-7. • I – Instruction cache enable bit. See Example 3-2 on page 3-13 and Invalidating and cleaning cache memory on page 8-17. • Z – Branch prediction enable bit. See Example 3-2 on page 3-13 and Invalidating and cleaning cache memory on page 8-17 • C – Cache enable bit. See Example 3-2 on page 3-13 and Invalidating and cleaning cache memory on page 8-17 • A – Alignment check enable bit. See Alignment on page 14-5. • M – Enable the MMU. See Configuring and enabling the MMU on page 9-3. Part of the boot code sequence will typically be to set the Z bit in the CP15:SCTLR, System Control Register, that enables branch prediction. A code sequence to do this is shown in Example 3-2 on page 3-13. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-12 ARM Processor Modes and Registers Example 3-2 Setting bits in the SCTLR MRC ORR ORR ORR MCR p15, 0, r0, r0, r0, r0, r0, r0, p15, 0, r0, #(1 #(1 #(1 r0, c1, c0, 0 << 2) << 12) << 11) c1, c0, 0 ; ; ; ; ; Read System Control Register configuration data Set C bit Set I bit Set Z bit Write System Control Register configuration data A similar form of this code can be found in Example 13-3 on page 13-3. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 3-13 Chapter 4 Introduction to Assembly Language Assembly language is a low-level programming language. There is in general, a one-to-one relationship between assembly language instructions (mnemonics) and the actual binary opcode executed by the core. Many programmers writing at the application level will have little reason to code in assembly language. However, knowledge of assembly code can be useful in cases where highly optimized code is required, when writing compilers, or where low level use of features not directly available in C is required. It might be required for portions of boot code, device drivers or when performing OS development. Finally, it can be useful to be able to read disassembled code when debugging C programs, and particularly, to understand the mapping between assembly instructions and C statements. Programmers seeking a more detailed description of ARM assembly language can also refer to the ARM Compiler Toolchain Assembler Reference or the ARM Architecture Reference Manual. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-1 Introduction to Assembly Language 4.1 Comparison with other assembly languages An ARM processor is a Reduced Instruction Set Computer (RISC) processor. Complex Instruction Set Computer (CISC) processors, like the x86, have a rich instruction set capable of doing complex things with a single instruction. Such processors often have significant amounts of internal logic that decode machine instructions to sequences of internal operations (microcode). RISC architectures, in contrast, have a smaller number of more general purpose instructions, that might be executed with significantly fewer transistors, making the silicon cheaper and more power efficient. Like other RISC architectures, ARM cores have a large number of general-purpose registers and many instructions execute in a single cycle. It has simple addressing modes, where all load/store addresses can be determined from register contents and instruction fields. The ARMv7-Architecture has basic data processing instructions that permit cores to perform arithmetic operations (such as ADD) and logical bit manipulation (such as AND). They also transfer program execution from one part of the program to another, in order to support loops and conditional statements. The architecture also has instructions to read and write external memory. The ARM instruction set is generally considered to be simple, logical and efficient. It has features not found in some processors, while at the same time lacking operations found in others. For example, it cannot perform data processing operations directly on memory. To increment a value in a memory location, the value must be loaded to an ARM register, the register incremented and a third instruction is required to write the updated value back to memory. The Instruction Set Architecture (ISA) includes instructions that combine a shift with an arithmetic or logical operation, auto-increment and auto-decrement addressing modes for optimized program loops, Load, and Store Multiple instructions that enable efficient stack and heap operations, plus block copying capability and conditional execution of almost all instructions. Like the x86 (but unlike the 68K), ARM instructions typically have a two or three operand format, with the first operand in most cases specifying the destination for the result. Load multiple and store instructions are an exception to this rule. The 68K, by contrast, places the destination as the last operand. For ARM instructions, there are generally no restrictions on which registers can be used as operands. Example 4-1 and Example 4-2 give a flavor of the differences between the different assembly languages. Example 4-1 Instructions to add 100 to a value in a register x86: add eax, #100 68K: ADD #100, D0 ARM: add r0, r0, 100 Example 4-2 Load a register with a 32-bit value from a register pointer ARM DEN0013D ID012214 x86: mov eax, DWORD PTR [ebx] 68K: MOVE.L (A0), D0 ARM: ldr r0, [r1] Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-2 Introduction to Assembly Language 4.2 The ARM instruction sets The ARMv7 architecture is a 32-bit processor architecture. It is also a load/store architecture, meaning that data-processing instructions operate only on values in general purpose registers. Only load and store instructions access memory. General purpose registers are also 32 bits. Throughout this book, when we refer to a word, we mean 32 bits. A doubleword is therefore 64 bits and a halfword is 16 bits wide. Though the ARMv7 architecture is a 32-bit architecture, individual processor implementations do not necessarily have 32-bit width for all blocks and interconnections. For example, it is possible to have 64-bit, or wider paths for instruction fetches or data accesses. Most ARM processors support two or even three different instruction sets, while some (for example, the Cortex-M3 processor) do not in fact execute the original ARM instruction set. There are at least two instruction sets that ARM processors can use. ARM (32-bit instructions) This is the original ARM instruction set. Thumb The Thumb instruction set was first added in the ARM7TDMI processor and contained only 16-bit instructions, which gave much smaller programs (memory footprint can be a major concern in smaller embedded systems) at the cost of some performance. Recent processors, including those in the Cortex-A series, support Thumb-2 technology that extends the Thumb instruction set to provide a mix of 16-bit and 32-bit instructions. This gives the best of both worlds, performance similar to that of the ARM instruction set, with code size similar to that of Thumb instructions. Because of its size and performance advantages, it is increasingly common for all code to be compiled or assembled to take advantage of Thumb-2 technology. In older ARM processors, systems often contained code which was compiled for ARM state and code which was compiled for Thumb state. ARM code, with 32-bit instructions, was more powerful and required fewer instructions to perform a particular task and so might be preferred for performance critical parts of the system. It was also used for exception handler code, because exceptions could not be handled in Thumb state on ARM7 or ARM9 Series processors. Thumb code, using 16-bit instructions, required more instructions to carry out the same task, when compared with ARM code. Thumb code could typically encode smaller constant values within instructions and has shorter relative branches. See Branches on page 5-15. The available range for relative branches is approximately ±32MB for ARM instructions and ±16MB for the Thumb-2 extension. Thumb is further limited where only 16-bit instructions are used, with conditional branches having a range of ±256 Bytes and unconditional relative branches being limited to ±2048 bytes. However, because Thumb instructions were only half of the size, programs would be typically a third smaller than their ARM code equivalent. Thumb instructions are therefore used for reasons of code density and to reduce system memory requirements. Thumb code can also outperform ARM when the processor is directly connected to a narrow (16-bit) memory, without the benefit of cache. One Thumb instruction can be fetched on each cycle, whereas each 32-bit ARM instruction requires two clock cycles per fetch. When executing a Thumb instruction, the PC reads as the address of the current instruction plus 4. The only 16-bit Thumb instructions which can directly modify the PC are certain encodings of MOV and ADD. The value written to the PC is forced to be halfword-aligned by ignoring its least significant bit, treating that bit as being 0. In ARMCC, the option --thumb or –arm (the default) allows selection of the instruction set used for compilation. A program can branch between these two instruction sets at run-time. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-3 Introduction to Assembly Language The currently used instruction set is indicated by the CPSR T bit and the core is said to be in ARM state (T = 0) or Thumb state (T = 1). Code has to be explicitly compiled or assembled to one state or the other. An explicit instruction is used to change between instruction sets. Calling functions that are compiled for a different state is known as interworking. We’ll take a more detailed look at this in Interworking on page 4-11. For Thumb assembly code, there is often a choice of 16-bit and 32-bit instruction encodings, with the 16-bit versions being generated by default. The .W (32-bit) and .N (16-bit) width specifiers can be used to force a particular encoding (if such an encoding exists), for example: BCS.W B.N ARM DEN0013D ID012214 label label ; forces 32-bit instruction even for a short branch ; faults if label out of range for 16-bit instruction Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-4 Introduction to Assembly Language 4.3 Introduction to the GNU Assembler The GNU Assembler, part of the GNU tools, is used to convert assembly language source code into binary object files. The assembler is extensively documented in the GNU Assembler Manual, that can be found online at http://sourceware.org/binutils/docs/as/index.html or (if you have GNU tools installed on your system) in the gnutools/doc sub-directory. GNU Assembler documentation is also available in the /gcc-doc/ package on Ubuntu. What follows is a brief description, intended to highlight differences in syntax between the GNU Assembler and standard ARM assembly language, and to provide enough information to enable programmers to get started with the tools. The names of GNU tool components will have prefixes indicating the target options selected, including operating system. An example would be arm-none-eabi-gcc, that might be used for bare metal systems using the ARM EABI. 4.3.1 Invoking the GNU Assembler You can assemble the contents of an ARM assembly language source file by running the arm-none-eabi-as program. arm-none-eabi-as -g -o filename.o filename.s The option -g requests the assembler to include debug information in the output file. When all of your source files have been assembled into binary object files (with the extension .o), you use the GNU Linker to create the final executable in ELF format. This is done by executing: arm-none-eabi-ld -o filename.elf filename.o For more complex programs, where there are many separate source files, it is more common to use a utility like make to control the build process. You can use the debugger provided by either arm-none-eabi-gdb or arm-none-eabi-insight to run the executable files on your host machine, as an alternative to a real target core. 4.3.2 GNU Assembler syntax The GNU Assembler can target many different processor architectures and is not ARM-specific. This means that its syntax is somewhat different from other ARM assemblers, such as the ARM toolchain. The GNU Assembler uses the same syntax for all of the many processor architectures that it supports. Assembly language source files consist of a sequence of statements, one per line. Each statement has three optional parts, ordered as follows: label: instruction @ comment A label lets you identify the address of this instruction. This can then be used as a target for branch instructions or for load and store instructions. A label can be a letter followed (optionally) by a sequence of alphanumeric characters, followed by a colon. The instruction can be either an ARM assembly instruction, or an assembler directive. These are pseudo-instructions that tell the assembler itself to do something. These are required, amongst other things, to control sections and alignment, or create data. Everything on the line after the @ symbol is treated as a comment and ignored (unless it is inside a string). C style comment delimiters “/*” and “*/” can also be used. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-5 Introduction to Assembly Language At link time an entry point can be specified on the command line if one has not been explicitly provided in the source code. 4.3.3 Sections An executable program with code will have at least one section, by convention this is called .text. Data can be included in a .data section. Directives with the same names enable you to specify which of the two sections should hold what follows in the source file. Executable code should appear in a .text section and read or write data in the .data section. Also read-only constants can appear in a .rodata section. Zero initialized data will appear in .bss. The Block Started by Symbol (bss) segment defines the space for uninitialized static data. 4.3.4 Assembler directives This is a key area of difference between GNU tools and other assemblers. All assembler directives begin with a period “.” A full list of these is described in the GNU documentation. Here, we give a subset of commonly used directives. .align This causes the assembler to pad the binary with bytes of zero value, in data sections, or NOP instructions in code, ensuring the next location will be on a word boundary. .align n gives 2^n alignment on ARM cores. .ascii “string” Insert the string literal into the object file exactly as specified, without a NUL character to terminate. Multiple strings can be specified using commas as separators. .asciiz Does the same as .ascii, but this time additionally followed by a NUL character (a byte with the value 0 (zero)). .byte expression, .hword expression, .word expression Inserts a byte, halfword, or word value into the object file. Multiple values can be specified using commas as separators. The synonyms .2byte and .4byte can also be used. .data Causes the following statements to be placed in the data section of the final executable. .end Marks the end of this source code file. The assembler does not process anything in the file after this point. .equ symbol, expression Sets the value of symbol to expression. The “=” symbol and .set have the same effect. .extern symbol Indicates that symbol is defined in another source code file. .global symbol Tells the assembler that symbol is to be made globally visible to other source files and to the linker. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-6 Introduction to Assembly Language .include “filename” Inserts the contents of filename into the current source file and is typically used to include header files containing shared definitions. .text This switches the destination of following statements into the text section of the final output object file. Assembly instructions must always be in the text section. For reference, Table 4-1 shows common assembler directives alongside GNU and ARM tools. Not all directives are listed, and in some cases there is not a 100% correspondence between them. Table 4-1 Comparison of syntax ARM DEN0013D ID012214 GNU Assembler armasm Description @ ; Comment #& #0x An immediate hex value .if IFDEF, IF Conditional (not 100% equivalent) .else ELSE .elseif ELSEIF .endif ENDIF .ltorg LTORG | :OR: OR & :AND: AND << :SHL: Shift Left >> :SHR: Shift Right .macro MACRO Start macro definition .endm ENDM End macro definition .include INCLUDE GNU Assembler requires “filename” .word DCD A data word .short DCW .long DCD .byte DCB .req RN .global IMPORT, EXPORT .equ EQU Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-7 Introduction to Assembly Language 4.3.5 Expressions Assembly instructions and assembler directives often require an integer operand. In the assembler, this is represented as an expression to be evaluated. Typically, this will be an integer number specified in decimal, hexadecimal (with a 0x prefix) or binary (with a 0b prefix) or as an ASCII character surrounded by single quotes. In addition, standard mathematical and logical expressions can be evaluated by the assembler to generate a constant value. These can utilize labels and other pre-defined values. These expressions produce either absolute or relative values. Absolute values are position-independent and constant. Relative values are specified relative to some linker-defined address, determined when the executable image is produced – such as target addresses for branches. 4.3.6 GNU tools naming conventions Registers are named in GCC as follows: • General registers: R0 - R15. • Stack pointer register: SP (R13). • Frame pointer register: FP (R11). • Link register: LR (R14). • Program counter: PC (R15). • Program Status Register flags: xPSR, xPSR_all, xPSR_f, xPSR_x, xPSR_ctl, xPSR_fs, xPSR_fx, xPSR_f, xPSR_cs, xPSR_cf, xPSR_cx (where x = C current or S saved). See Program Status Registers on page 3-8. Note Chapter 15 Application Binary Interfaces describes how all of the registers are assigned a role within the procedure call standard and how the GNU Assembler lets you refer to the registers using their Procedure Call Standard (PCS) names. See Table 15-1 on page 15-2. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-8 Introduction to Assembly Language 4.4 ARM tools assembly language The Unified Assembly Language (UAL) format now used by ARM tools enables the same canonical syntax to be used for both ARM and Thumb instruction sets. The assembler syntax of ARM tools is not identical to that used by the GNU Assembler, particularly for preprocessing and pseudo-instructions that do not map directly to opcodes. In the next chapter, we will look at the individual assembly language instructions in a little more detail. Before doing that, we take a look at the basic syntax used to specify instructions and registers. Assembly language examples in this book use both UAL and GNU Assembly syntax. UAL gives the ability to write assembler code that can be assembled to run on all ARM cores. In the past, it was necessary to write code explicitly for ARM or Thumb state. Using UAL the same code can be assembled for different instruction sets at the time of assembly, not at the time the code is written. This can be either through the use of command line switches or inline directives. Legacy code will still assemble correctly. It is worth noting that GNU Assembler now supports UAL through use of the .syntax directive, though it might not be identical syntax to the ARM tools assembler. 4.4.1 ARM assembler syntax ARM assembler source files consist of a sequence of statements, one per line. Each statement has three optional parts, ordered as follows: label instruction ; comment A label lets you identify the address of this instruction. This can then be used as a target for branch instructions or for load and store instructions. The instruction can be either an assembly instruction, or an assembler directive. These are pseudo-instructions that tell the assembler itself to do something. These are required, amongst other things, to control sections and alignment, or create data. Everything on the line after the ; symbol is treated as a comment and ignored (unless it is inside a string). C style comment delimiters “/*” and “*/” can also be used. 4.4.2 Labels A label is required to start in the first character of a line. If the line does not have a label, a space or tab delimiter is required to start the line. If there is a label, the assembler makes the label equal to the address in the object file of the corresponding instruction. Labels can then be used as the target for branches or for loads and stores. Example 4-3 A simple example showing use of a label Loop MUL R5, R5, R1 SUBS R1, R1, #1 BNE Loop In Example 4-3, Loop is a label and the conditional branch instruction (BNE Loop) will be assembled in a way that makes the offset encoded in the branch instruction point to the address of the MUL instruction that is associated with the label Loop. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-9 Introduction to Assembly Language 4.4.3 Directives Most lines will normally have an assembly language instruction, to be converted by the tool into its binary equivalent, or a directive that tells the assembler to do something. It can also be a pseudo-instruction (one that will be converted into one or more real instructions by the assembler). We’ll look at the actual instructions available in hardware in Chapter 5 and focus mainly on the assembler directives here. These perform a wide range of tasks. They can be used to place code or data at a particular address in memory, create references to other programs and so forth. For example, the Define Constant (DCD, DCB, DCW) directives let you place data into a piece of code. This can be expressed numerically (in decimal, hex, binary) or as ASCII characters. It can be a single item or a comma separated list. DCB is for byte sized data, DCD can be used for word sized data, and DCW for half-word sized data items. For example, we might have: MESSAGE DCB “Hello World!”,0 This will produce a series of bytes corresponding to the ASCII characters in the string, with a 0 termination. MESSAGE is a label that we can use to get the address of this data. Similarly, we might have data items expressed in hex: Masks DCD 0x100, 0x80, 0x40, 0x20, 0x10 The EQU directive lets you assign names to address or data values. For example: CtrlD EQU 4 TUBE EQU 0x30000000 We can then use these names in other instructions, as parts of expressions to be evaluated. EQU does not actually cause anything to be placed in the program executable – it merely sets a name to a value, for use in other instructions, in the symbol table for the assembler. It is convenient to use such names to make code easier to read, but also so that if we change the address or value of something in a piece of code, we must modify the original definition, rather than having to change all of the references to it individually. It is usual to group EQU definitions together, often at the start of a program or function, or in separate include files. The AREA pseudo-instruction is used to tell the assembler about how to group together code or data into logical sections for later placement by the linker. For example, exception vectors might require to be placed at a fixed address. The assembler keeps track of where each instruction or piece of data is located in memory. The AREA directive can be used to modify that location. The ALIGN directive lets you align the current location to a specified boundary. It usually does this by padding (where necessary) with zeros or NOP instructions, although it is also possible to specify a pad value with the directive. The default behavior is to set the current location to the next word (four byte) boundary, but larger boundary sizes and offsets from that boundary can also be specified. This can be required to meet alignment requirements of certain instructions (for example LDRD and STRD doubleword memory transfers), or to align with cache boundaries. As with the .align directive on GNU Assembler, the ALIGN n directive gives 2^n alignment on ARM cores. END is used to denote the end of the assembly language source program. Failure to use the END directive will result in an error being returned. INCLUDE tells the assembler to include the contents of another file into the current file. Include files can be used as an easy mechanism for sharing definitions between related files. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-10 Introduction to Assembly Language 4.5 Interworking When the core executes ARM instructions, it is said to be operating in ARM state. When it is operating in Thumb state, it is executing Thumb instructions. A core in a particular state can only execute instructions from that instruction set. We must make sure that the core does not receive instructions of the wrong instruction set. Each instruction set includes instructions to change processor state. ARM and Thumb code can be mixed, if the code conforms to the requirements of the ARM and Thumb Procedure Call Standards (described in Chapter 15). Compiler generated code will always do so, but assembly language programmers must take care to follow the specified rules. Selection of processor state is controlled by the T bit in the CPSR. See Figure 3-6 on page 3-8. When T is 1, the processor is in Thumb state. When T is 0, the processor is in ARM state. However, when the T bit is modified, it is also necessary to flush the instruction pipeline (to avoid problems with instructions being decoded in one state and then executed in another). Special instructions are used to accomplish this. These are BX (Branch with eXchange) and BLX (Branch and Link with eXchange). LDR of PC and POP/LDM of PC also have this behavior. In addition to changing the processor state with these instructions, assembly programmers must also use the appropriate directive to tell the assembler to generate code for the appropriate state. The BX or BLX instruction branches to an address contained in the specified register, or an offset specified in the opcode. The value of bit [0] of the branch target address determines whether execution continues in ARM state or Thumb state. Both ARM (aligned to a word boundary) and Thumb (aligned to a halfword boundary) instructions do not use bit [0] to form an instruction address. This bit can therefore safely be used to provide the additional information about whether the BX or BLX instruction should change the state to ARM (address bit [0] = 0) or Thumb (address bit [0] = 1). The BL label will be turned into a BLX label as appropriate at link time if the instruction set of the caller is different from the instruction set of label, assuming that it is unconditional. A typical use of these instructions is when a call from one function to another is made using the BL or BLX instruction, and a return from that function is made using the BX LR instruction. Alternatively, we can have a non-leaf function that pushes the link register onto the stack on entry and pops the stored link register from the stack into the program counter, on exit. Here, instead of using the BX LR instruction to return, we have a memory load. Memory load instructions that modify the PC might also change the processor state depending on the value of bit [0] of the loaded address. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-11 Introduction to Assembly Language 4.6 Identifying assembly code When faced with a piece of assembly language source code, it can be useful to be able to determine which instruction set will be used and which kind of assembler it is targeted at. Older ARM Assembly language code can have three (or even four) operand instructions present (for example, ADD R0, R1, R2) or conditional execution of non-branch instructions (for example, ADDNE R0, R0, #1). Filename extensions will typically be .s or .S. Code targeted for the newer UAL, will contain the directive .syntax unified but will otherwise appear similar to traditional ARM Assembly language. The pound (or hash) symbol # can be omitted in front of immediate operands. Conditional instruction sequences must be preceded immediately by the IT instruction, described in Chapter 5. Such code assembles either to fixed-size 32-bit (ARM) instructions, or mixed-size (16-bit and 32-bit) Thumb instructions, depending on the presence of the directives .thumb or .arm. You can, on occasion, encounter code written in 16-bit Thumb assembly language. This can contain directives such as .code 16, .thumb or .thumb_func but will not specify .syntax unified. It uses two operands for most instructions, although ADD and SUB can sometimes have three. Only branches can be executed conditionally. All GCC inline assembler code types, such as .c, .h, .cpp, .cxx, and .c++ can be built for Thumb or ARM, depending on GCC configuration and command-line switches (-marm or –mthumb). ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-12 Introduction to Assembly Language 4.7 Compatibility with ARMv8-A With the introduction of the ARMv8-A architecture a number of changes have been made to the ARMv7 ISA to provide backward compatibility. The following instructions have been added to the ARMv7 ISA. Table 4-2 New instructions Instruction Description LDA Load-Acquire Word LDAB Load-Acquire Byte LDAEX Load-Acquire Exclusive Word LDAEXB Load-Acquire Exclusive Byte LDAEXD Load-Acquire Exclusive Double LDAEXH Load-Acquire Exclusive Halfword LDAH Load-Acquire Halfword STL Store-Release Word STLB Store-Release Byte STLEX Store-Release Exclusive Word STLEXB Store-Release Exclusive Byte STLEXD Store-Release Exclusive Double STLEXH Store-Release Exclusive Halfword STLH Store-Release Halfword These are described in Appendix A Instruction Summary. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 4-13 Chapter 5 ARM/Thumb Unified Assembly Language Instructions This chapter is a general introduction to ARM/Thumb assembly language. It does not provide detailed coverage of every instruction, descriptions of individual instructions can be found in Appendix A Instruction Summary. Instructions can broadly be placed in one of a number of classes: • Data processing operations (ALU operations such as ADD). • Memory access (load and stores to memory). • Control flow (for loops, goto, conditional code and other program flow control). • System (coprocessor, debug, mode changes and so forth). We will take a brief look at each of these in turn. Before that, let us examine capabilities that are common to different instruction classes. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 5-1 ARM/Thumb Unified Assembly Language Instructions 5.1 Instruction set basics There are a number of features common to all parts of the instruction set. 5.1.1 Constant and immediate values ARM or Thumb assembly language instructions have a length of only 16 or 32 bits. This presents something of a problem. It means that you cannot encode an arbitrary 32-bit value within the opcode. In the ARM instruction set, as opcode bits are used to specify condition codes, the instruction itself and the registers to be used, only 12 bits are available to specify an immediate value. You have to be somewhat creative in how these 12 bits are used. Rather than enabling a constant of size –2048 to +2047 to be specified, instead the 12 bits are divided into an 8-bit constant and 4-bit rotate value. The rotate value enables the 8-bit constant value to be rotated right by a number of places from 0 to 30 in steps of 2, that is, 0, 2, 4, 6, 8... So, you can have immediate values like 0x23 or 0xFF. You can produce other useful immediate values, for example, addresses of peripherals or blocks of memory. As an example, 0x23000000 can be produced by expressing it as 0x23 ROR 8 (see ROR on page A-35). But many other constants, like 0x3FF, cannot be produced within a single instruction. For these values, you must either construct them in multiple instructions, or load them from memory. Programmers do not typically concern themselves with this, except where the assembler gives an error complaining about an invalid constant. Instead, you can use assembly language pseudo-instructions to do whatever is necessary to generate the required constant. Constant values encoded in an instruction can be one of the following in Thumb: • a constant that can be produced by rotating an 8-bit value by any even number of bits within a 32-bit word • a constant of the form 0x00XY00XY • a constant of the form 0xXY00XY00 • a constant of the form 0xXYXYXYXY. where XY is a hexadecimal number in the range 0x00 to 0xFF. The MOVW instruction (move wide), will move a 16-bit constant into a register, while zeroing the top 16 bits of the target register. MOVT (move top) will move a 16-bit constant into the top half of a given register, without changing the bottom 16 bits. This permits a MOV32 pseudo-instruction that is able to construct any 32-bit constant. The assembler provides some more help here. The prefixes :upper16: and :lower16: enable you to extract the corresponding half from a 32-bit constant: MOVW R0, #:lower16:label MOVT R0, #:upper16:label Although this requires two instructions, it does not require any extra space to store the constant, and there is no requirement to read a data item from memory. You can also use pseudo-instructions LDR Rn, = or LDR Rn, =label. This was the only option for older processors that lacked MOVW and MOVT. The assembler will then use the best sequence to generate the constant in the specified register (one of MOV, MVN or an LDR from a literal pool). A literal pool is an area of constant data held within the code section, typically after the end of a function and before the start of another. If it is necessary to manually control literal pool placement, this can be done with an assembler directive – LTORG for armasm, or .ltorg when using GNU tools. The register loaded could be the program counter, that would cause a branch. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 5-2 ARM/Thumb Unified Assembly Language Instructions This can be useful for absolute addressing or for references outside the current section; obviously this will result in position-dependent code. The value of the constant can be determined either by the assembler, or by the linker. ARM tools also provides the related pseudo-instruction ADR Rn, =label. This uses a PC-relative ADD or SUB, to place the address of the label into the specified register, using a single instruction. If the address is too far away to be generated this way, the ADRL pseudo-instruction is used. This requires two instructions, that gives a better range. This can be used to generate addresses for position-independent code, but only within the same code section. 5.1.2 Conditional execution A feature of the ARM instruction set is that nearly all instructions can be conditional. On most other architectures, only branches or jumps can be executed conditionally. This can be useful in avoiding conditional branches in small if/then/else constructs or for compound comparisons. As an example of this, consider code to find the smaller of two values, in registers R0 and R1 and place the result in R2. This is shown in Example 5-1. The suffix LT indicates that the instruction must be executed only if the most recent flag-setting instruction returned less than; GE means greater than or equal. Example 5-1 Example code showing branches (GNU) @ Code using branches CMP R0, R1 BLT .Lsmaller MOV R2, R1 B .Lend .Lsmaller: MOV R2, R0 .Lend: @ if R0 > 2 is done as MOV R0, R1, LSR #2. Equally, it is common to combine shifts with ADD, SUB or other instructions. For example, to multiply R0 by 5, you might write: ADD R0, R0, R0, LSL #2 A left shift of n places is effectively a multiply by 2 to the power of n, so this effectively makes R0 = R0 + (4 × R0). A right shift provides the corresponding divide operation, although ASR rounds negative values differently than would division in C. Apart from multiply and divide, another common use for shifted operands is array index look-up. Consider the case where R1 points to the base element of an array of int (32-bit) values and R2 is the index that points to the nth element in that array. You can obtain the array value with a single load instruction that uses the calculation R1 + (R2 × 4) to get the appropriate address. Example 5-3 on page 5-8 provides examples of differing operand 2 types used in ARM instructions. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 5-7 ARM/Thumb Unified Assembly Language Instructions Example 5-3 ARM instructions showing a variety of operand 2 types add add add add 5.2.2 R0, R0, R0, R0, R1, R1, R1, R1, #1 R2 R2, LSL #4 R2, LSL R3 @ @ @ @ R0 R0 R0 R0 = = = = R2 R1 R1 R1 + + + + 1 R2 R2<<#4 R2< × 2 Let’s consider an example: The decimal value +0.5 is represented as a single precision float by the hexadecimal value 0x3F000000. This has a sign value of 0 (positive). The value of the mantissa is 1.0, though the integral part (1) is implicit and is not stored. The exponent value is specified in bits [30:23] – that hold 0b01111110, or 126 – offset by 127 to represent an exponent of -1. The value is therefore given by (-1)sign × mantissa × 2exponent = 1 × 1 × 2-1 = 0.5 (decimal) Denormal numbers are a special case. If you set the exponent bits to zero, you can represent very small numbers other than zero, by setting mantissa bits. Because normal values have an implied leading 1, the closest value to zero you can represent as a normal value is ±2-126. To get smaller numbers, the 1.m interpretation of the mantissa value is replaced with a 0.m interpretation. Now, the number's magnitude is determined only by bit positions. When using these extremely-small numbers, the available precision does not scale with the magnitude of the value. Without the implied 1 attached to the mantissa, all bits to the left of the lowest set bit are leading zeros, so the smallest representable number is 1.401298464e-45, represented by 0x00000001. For performance reasons, such denormal values are often ignored and are flushed to zero. This is strictly a violation of IEEE-754, but denormal values are used rarely enough in real programs that the performance benefit is worth more than correct handling of these extremely small numbers. Cortex processors with VFP enable code to select between flush-to-zero mode and full denormal support. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 6-3 Floating-Point Because a 32-bit floating-point number has a 23-bit mantissa there are many values of a 32-bit int that if converted to 32-bit float cannot be represented exactly. This is referred to as loss of precision. If you convert one of these values to float and back to int you will get a different, but nearby value. In the case of double-precision floating-point numbers, the exponent field has 11 bits (giving an exponent range from –1022 to +1023) and a mantissa field with 52 bits. 6.1.1 Rounding algorithms The IEEE 754-1985 standard defines four different ways in which results can be rounded, as follows: • Round to nearest (ties to even). This mode causes rounding to the nearest value. If a number is exactly midway between two possible values, it is rounded to the nearest value with a zero least significant bit. • Round toward 0. This causes numbers to always be rounded towards zero (this can be also be viewed as truncation). • Round toward +∞ .This selects rounding towards positive infinity. • Round toward -∞. This selects rounding towards negative infinity. The IEEE 754-2008 standard adds an additional rounding mode. In the case of round to nearest, it is now also possible to round numbers that are exactly halfway between two values, away from zero (in other words, upwards for positive numbers and downwards for negative numbers). This is in addition to the option to round to the nearest value with a zero least significant bit. At present VFP does not support this rounding mode. 6.1.2 ARM VFP VFP is an optional but rarely omitted extension to the instruction sets in the ARMv7-A architecture. It can be implemented with either thirty-two, or sixteen double-word registers. The terms VFPv3-D32 and VFPv3-D16 are used to distinguish between these two options. If the Advanced SIMD (NEON) extension is implemented together with VFPv3, VFPv3-D32 is always present. VFPv3 can also be optionally extended by the half-precision extensions that provide conversion functions in both directions between half-precision floating-point (16-bit) and single-precision floating-point (32-bit). These operations only enable half-precision floats to be converted to and from other formats. VFPv4 adds both the half-precision extensions and the Fused Multiply-Add instructions to the features of VFPv3. In a Fused Multiply-Add operation, only a single rounding occurs at the end. This is one of the new facets of the IEEE 754-2008 specification. Fused operations can improve the accuracy of calculations that repeatedly accumulate products, such as matrix multiplication or dot product calculation. The VFP version supported by individual Cortex-A series processors is given in Table 2-3 on page 2-9. In addition to the registers described, there are a number of other VFP registers: Floating-Point System ID Register (FPSID) This can be read by system software to determine which floating-point features are supported in hardware. Floating-Point Status and Control register (FPSCR) This holds comparison results and flags for exceptions. Control bits select rounding options and enable floating-point exception trapping. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 6-4 Floating-Point Floating-Point Exception Register (FPEXC) The FPEXC register contains bits that enable system software that handles exceptions to determine what has happened. Media and VFP Feature registers 0 and 1 (MVFR0 and MVFR1) These registers enable system software to determine which Advanced SIMD and floating-point features are provided on the processor implementation. User mode code can only access the FPCSR. One implication of this is that applications cannot read the FPSID to determine which features are supported unless the host OS provides this information. Linux provides this through /proc/cpuinfo, for example, but the information is not nearly as detailed as that provided by the VFP hardware registers. Unlike ARM integer instructions, no VFP operations will affect the flags in the APSR directly. The flags are stored in the FPSCR. Before the result of a floating-point comparison can be used by the integer processor, the flags set by a floating-point comparison must be transferred to the APSR, using the VMRS instruction. This includes use of the flags for conditional execution, even of other VFP instructions. Example 6-1 shows a simple piece of code to illustrate this. The VCMP instruction performs a comparison the values in VFP registers d0 and d1 and sets FPSCR flags as a result. These flags must then be transferred to the integer processor APSR, using the VMRS instruction. You can then conditionally execute instructions based on this. Example 6-1 Example code illustrating usage of floating-point flags VCMP d0, d1 VMRS APSR_nzcv, FPSCR BNE label Flag meanings The integer comparison flags support comparisons that are not applicable to floating-point numbers. For example, floating-point values are always signed, so there is no requirement for unsigned comparisons. On the other hand, floating-point comparisons can result in the unordered result (meaning that one or both operands was NaN, or Not a Number). IEEE-754 defines four testable relationships between two floating-point values, that map onto the ARM condition codes as follows: Table 6-2 ARM APSR flags IEEE-754 relationship ARM DEN0013D ID012214 ARM APSR flags N Z C V Equal 0 1 1 0 Less Than (LT) 1 0 0 0 Greater Than (GT) 0 0 1 0 Unordered (At least one argument was NaN) 0 0 1 1 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 6-5 Floating-Point Compare with zero Unlike the integer instructions, most VFP (and NEON) instructions can operate only on registers, and cannot accept immediate values encoded in the instruction stream. The VCMP instruction is a notable exception in that it has a special-case variant that enables quick and easy comparison with zero. Interpreting the flags When the flags are in the APSR, they can be used almost as if an integer comparison had set the flags. However, floating-point comparisons support different relationships, so the integer condition codes do not always make sense. Table 6-3 describes floating-point comparisons rather than integer comparisons: Table 6-3 Interpreting the flags Code Meaning (when set by vcmp) Meaning (when set by cmp) Flags tested EQ Equal to Equal to Z =1 NE Unordered, or not equal to Not equal to. Z=0 CS or HS Greater than, equal to, or unordered Greater than or equal to (unsigned). C=1 CC or LO Less than. Less than (unsigned). C=0 MI Less than Negative. N=1 PL Greater than, equal to, or unordered Positive or zero. N=0 VS Unordered. (At least one argument was NaN.) Signed overflow. V=1 VC Not unordered. (No argument was NaN.) No signed overflow. V=0 HI Greater than or unordered Greater than (unsigned). (C = 1) && (Z = 0) LS Less than or equal to Less than or equal to (unsigned). (C = 0) || (Z = 1) GE Greater than or equal to. Greater than or equal to (signed). N==V LT Less than or unordered. Less than (signed). N!=V GT Greater than. Greater than (signed). (Z==0) && (N==V) LE Less than, equal to or unordered. Less than or equal to (signed). (Z==1) || (N!=V) AL (or omitted) Always executed. Always executed. None tested. It is clear that the condition code is attached to the instruction reading the flags, and the source of the flags makes no difference to the flags that are tested. It is the meaning of the flags that differs when you perform a vcmp rather than a cmp. Similarly, it is clear that the opposite conditions still hold. For example, HS is still the opposite of LO. When set by CMP the flags generally have analogous meanings to the flags set by VCMP. For example, GT still means greater than. However, the unordered condition and the removal of the signed conditions can confuse matters. Often, for example, it is desirable to use LO, normally an unsigned less than check, in place of LT, because it does not match in the unordered case. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 6-6 Floating-Point 6.1.3 Enabling VFP If an ARMv7-A core includes VFP hardware, it must be explicitly enabled before applications can make use of it. Several steps are required to do this: ARM DEN0013D ID012214 • The EN bit in the FPEXC register must be set. • If access to VFP is required in the Normal world, access to CP10 and CP11 must be enabled in the Non-Secure Access Control Register (CP15.NSACR). This would normally be done inside the Secure bootloader. • Access to CP10 and CP11 must be enabled in the Coprocessor Access Control Register (CP15.CACR). This can be done on demand by the operating system. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 6-7 Floating-Point 6.2 VFP support in GCC Use of VFP is fully supported by GCC, although some builds can be configured to default to assume no VFP support, in which case floating-point calculations will use library code. The main option to use for VFP support is: • -mfpu=vfp specifies that the target has VFP hardware. As does specifying the option -mfpu=neon. Other options can be used to specify support for a specific VFP implementation on an ARM Cortex-A series processor: • -mfpu=vfpv3 or -mfpu=vfpv3-d16 (for Cortex-A8 and Cortex-A9 processors). • -mfpu=vfpv4 or -mfpu=vfpv4-d16 (for Cortex-A5 and Cortex-A15 processors). These options can be used for code that will run only on these VFP implementations, and do not require backward compatibility with older VFP implementations. • -mfloat-abi=softfp (or hard) specify which ABI to use to enable the use of VFP. softfp uses a Procedure Call Standard compatible with software floating-point, and so provides binary compatibility with legacy code. This permits running older soft float code with new libraries that support hardware floating-point, but still makes use of hardware floating-point registers between function calls. hard has floating-point values passed in floating-point registers. This is more efficient but is not backward compatible with the softfp ABI variant. Particular care is required with libraries, including the C platform library. See VFP and NEON register use on page 15-4 for more information on efficient parameter passing. C programmers must note that there can be a significant function call overhead when using -mfloat-abi=softfp, if many floating-point values are being passed. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 6-8 Floating-Point 6.3 VFP support in the ARM Compiler Use of VFP is fully supported by the ARM Compiler (although some builds can be configured by default to assume no VFP support, in which case floating-point calculations will use library code). The main option to use with the ARM Compiler for VFP support is: • --fpu=name that lets you specify the target floating-point hardware. The options used to specify support for a specific VFP implementation on an ARM Cortex-A series processor are: • --fpu=vfpv3 or --fpu=vfpv3_d16 (for the Cortex-A8 and Cortex-A9 processors). • --fpu=vfpv4 or --fpu=vfpv4_d16 (for all other Cortex-A series processors). These options can be used for code that will run only on these VFP implementations, and do not require backward compatibility with older VFP implementations. Use --fpu=list to see the full list of FPUs supported. The following options can be used for linkage support: • --apcs=/hardfp generates code for hardware floating-point linkage. • --apcs=/softfp generates code for software floating-point linkage. Hardware floating-point linkage uses the FPU registers to pass the arguments and return values. Software floating-point linkage means that the parameters and return value for a function are passed using the ARM integer registers R0 to R3 and the stack. --apcs=/hardfp and --apcs=/softfp interact with or override explicit or implicit use of --fpu. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 6-9 Floating-Point 6.4 VFP support in Linux An application that uses VFP (or calls into a library that uses VFP) places some additional requirements on the Linux kernel. For the application to run correctly, the kernel must save and restore the VFP registers during context switches. The kernel might also have to decode and emulate VFP instructions where the VFP hardware is not present. 6.4.1 Context switching In addition to saving and restoring integer registers, the kernel might also have to perform saving and restoring of VFP registers on a context switch. To avoid wasting cycles, this is done only when an application actually used VFP. Because the VFP initialization code leaves VFP disabled, the first time a thread actually tries to access the floating-point hardware, an undefined exception occurs. The kernel function that handles this, sees that VFP is disabled and that a new thread wants to use VFP. It saves the current VFP state and restores the state for the new thread. On clusters, where threads can migrate to a different core, this simple system will no longer work correctly. Instead, the kernel saves the state if the VFP was used by the previous thread. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 6-10 Floating-Point 6.5 Floating-point optimization This section contains some suggestions for developers writing FP assembly code. Some caution is required when applying these points, as recommendations can be specific to a particular piece of hardware. A code sequence that is optimal for one core can be sub-optimal on different hardware. ARM DEN0013D ID012214 • Avoid mixing of VFP and NEON instructions on the Cortex-A9 processor, as there is a significant overhead in switching between data engines. • Moves to and from VFP system control registers, such as FPSCR are not typically present in high-performance code, and might not be optimized. These must not be placed in time-critical loops, if possible. For example, accesses to control registers on the Cortex-A9 processor are serializing, and will have a significant performance impact if used in tight loops or performance-critical code. • Register transfer between the integer processor register bank and the floating-point register bank must similarly be avoided in time-critical loops. For the Cortex-A8 processor, this is particularly true of register transfers from VFP registers to integer registers. • Load/store multiple operations are preferred to the use of multiple, individual floating-point loads and stores, to make efficient use of available transfer bandwidth. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 6-11 Chapter 7 Introducing NEON NEON technology provides Single Instruction Multiple Data (SIMD) operations in ARM processors implementing the Advanced SIMD architecture extensions and can be used to accelerate the performance of multimedia applications running on ARM Cortex-A series processors. These operations can significantly accelerate repetitive operations on large data sets. This can be useful in applications such as media codecs. NEON is implemented as a separate hardware unit that is available as an option on Cortex-A series processors. Making the NEON hardware optional enables ARM SoCs to be optimized for specific markets. In most general-purpose applications processor SoCs, NEON will probably be included. However, for an embedded application such as a network router, NEON can be omitted, enabling a small saving in silicon area that translates to a small cost saving. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-1 Introducing NEON 7.1 SIMD SIMD is a computational technique for processing a number of data values (generally a power of two) using a single instruction, with the data for the operands packed into special wide registers. One instruction can therefore do the work of many separate instructions. This type of parallel processing instruction is commonly called SIMD (single instruction, multiple data), one of four classifications of computer architectures defined by Michael J. Flynn in 1966 based on the number of instruction and data streams available in the architecture. Single Instruction Multiple Data (SIMD) A technique for processing multiple data values using a single instruction, with the data for the operands packed into wide registers. One instruction can therefore do the work of many. SIMD instructions are very powerful for computations on media data. Single Instruction, Single Data (SISD) A single core executes a single instruction stream, to operate on data stored in a single memory one operation at a time. There is often a central controller that broadcasts the instruction stream to all the processing elements. Almost all ARM processors prior to the ARMv6 architecture use SISD processing. Multiple Instruction, Single Data (MISD) A type of parallel computing architecture where many functional units perform different operations on the same data. Fault-tolerant computers executing the same instructions in order to detect and mask errors might be considered to belong to this type. One example of this was the Space Shuttle flight control computers Multiple Instruction, Multiple Data (MIMD) Multiple computer instructions, that might be the same, and that might be synchronized with each other, perform actions simultaneously on two or more pieces of data. A multi-core superscalar processor is an MIMD processor. For code that can be parallelized, large performance improvements can be achieved. SIMD extensions exist on many 32-bit architectures – PowerPC has AltiVec, while x86 has several variants of MMX/SSE. SIMD is described in Integer SIMD instructions on page 5-8. Many software programs operate on large datasets. The data items can be less than 32 bits in size. 8-bit pixel data is common in video, graphics and image processing, 16-bit samples in audio codecs. In such cases, the operations to be performed are simple, repeated many times and have little requirement for control code. SIMD can offer considerable performance improvements for this type of data processing. It is particularly beneficial for digital signal processing or multimedia algorithms, such as: ARM DEN0013D ID012214 • Block-based data processing. • Audio, video, and image processing codecs. • 2D graphics based on rectangular blocks of pixels • 3D graphics • Color-space conversion. • Physics simulations. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-2 Introducing NEON On a 32-bit core such as the Cortex-A series processors, it is relatively inefficient to perform large numbers of 8-bit or 16-bit single operations one at a time. The processor ALU, registers and datapath are designed for the 32-bit calculations. SIMD enables a single instruction to treat a register value as multiple data elements (for example, as four 8-bit values in a 32-bit register) and to perform multiple identical operations on those elements. 32-bit scalar add A0 SIMD parallel add C0 B0 A1 B1 C1 A2 B2 C2 A3 B3 A0 B0 C0 A1 B1 C1 A2 B2 C2 A3 B3 C3 C3 Figure 7-1 Comparing SIMD parallel add with 32-bit scalar add To achieve four separate additions without using SIMD requires you to use four ADD instructions, as shown in Figure 7-1, and additional instructions to prevent one result from overflowing into the adjacent byte. SIMD requires only one instruction to do this. 31 16 15 24 23 0 8 7 Vector R2 31 8 7 16 15 24 23 0 R1 + + 31 8 7 16 15 24 23 + + 0 R0 Lane 3 Lane 2 Lane 1 Lane 0 Figure 7-2 4-way 8-bit add operation Figure 7-2 shows the operation of the SIMD UADD8 R0, R1, R2 instruction. This operation performs a parallel addition of four pairs of 8-bit elements (called lanes) packed into vectors stored in general purpose registers R1 and R2, and places the result into a vector in register R0. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-3 Introducing NEON It is important to note that the addition operations are truly independent. Any overflow or carry from bit [7] of lane 0 does not affect bit [0] of lane 1 (bit [8] of the whole register), which is a separate calculation. ARM NEON technology is designed to build on the concept of SIMD. NEON is a combined 64-bit and 128-bit SIMD instruction set that provides 128-bit wide vector operations, compared to the 32-bit SIMD in the ARMv6 architecture. NEON technology introduced in the ARMv7 architecture is at present only available with ARM Cortex-A and Cortex-R series processors. It is a SIMD technology targeted at advanced media and signal processing applications and embedded processors, and can accelerate multimedia and signal processing algorithms such as video encode/decode, 2D/3D graphics, gaming, audio and speech processing, image processing, telephony, and sound synthesis by at least twice the performance of ARMv6 SIMD. NEON is included by default in the Cortex-A7, Cortex-A12, and Cortex-A15 processors, but is optional in all other ARMv7 Cortex-A series processors. NEON can execute MP3 audio decoding on processors running at 10 MHz. It features a comprehensive instruction set, a separate register file and independent execution hardware. NEON supports 8-, 16-, 32- and 64-bit integer and single-precision (32-bit) floating-point data and SIMD operations for handling audio and video processing as well as graphics and gaming processing. NEON supports up to 16 operations at the same time. The NEON hardware shares the same registers as used in VFP. It is implemented as part of the ARM processor, but has its own execution pipelines and a register bank that is distinct from the ARM core register bank. NEON data is organized into very long registers (64 or 128 bits wide). These registers can hold data items that are 8, 16, 32 or 64 bits long. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-4 Introducing NEON 7.2 NEON architecture overview NEON was designed as an additional load/store architecture to provide good vectorizing compiler support from languages such as C/C++. A rich set of NEON instructions operate on wide 64-bit and 128-bit vector registers, enabling a high level of parallelism. The NEON instructions are straightforward and easy to understand, that also makes hand-coding easy for applications that require the very highest performance. A key advantage of NEON technology is that instructions form part of the normal ARM or Thumb code, making programming simpler than with an external hardware accelerator. There are NEON instructions available to read and write external memory, move data between NEON registers and other ARM registers and to perform SIMD operations. The NEON architecture uses a 32 × 64-bit register file. These are actually the same registers used by the floating-point unit (VFPv3). It does not matter that the floating-point registers are re-used as NEON registers. All compiled code and subroutines will conform to the EABI, that specifies which registers can be corrupted and which registers must be preserved. The compiler is free to use any NEON or VFPv3 registers for floating-point values or NEON data at any point in the code. The NEON architecture permits 64-bit or 128-bit parallelism. This choice was made to keep the size of the NEON unit manageable (a vector ALU can easily become quite large), while still offering good performance benefits from vectorization. The NEON architecture also does not specify instruction timings and might require different numbers of cycles to execute the same instruction on different processors. 7.2.1 Commonality with VFP The ARM architecture can support a wide range of different NEON and VFP options, but in practice you see only the combinations: • No NEON or VFP. • VFP only. • NEON and VFP. These are vendor implementation options for the architecture, and so are fixed for a particular implementation of an ARM-based design. The key differences between NEON and VFP are that NEON only works on vectors, does not support double-precision floating-point (double-precision is supported by the VFP), and does not support certain complex operations such as square root and divide. NEON has a register bank of thirty-two 64-bit registers. If both NEON and VFPv3 are implemented, this register bank is shared between them in hardware. This means that VFPv3 must be present in its VFPv3-D32 form, that has 32 double-precision floating-point registers. This makes support for context switching simpler. Code that saves and restores VFP context also saves and restores NEON context. 7.2.2 Data types NEON instructions operate on elements of the following types: ARM DEN0013D ID012214 • 32-bit single precision floating-point • 8, 16, 32 and 64-bit unsigned and signed integers • 8 and 16-bit polynomials. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-5 Introducing NEON Data type specifiers in NEON instructions comprise a letter indicating the type of data and a number indicating the width. They are separated from the instruction mnemonic by a point, for example, VMLAL.S8. So you have the following possibilities: • Unsigned integer U8 U16 U32 U64. • Signed integer S8 S16 S32 S64. • Integer of unspecified type I8 I16 I32 I64. • Floating-point number F16 F32. • Polynomial over {0,1} P8. Note F16 is not supported for data processing operations. It is only supported as a format to be converted to, or from. Polynomial arithmetic is useful when implementing certain cryptography or data integrity algorithms. Adding two polynomials over {0,1} is the same as a bitwise exclusive OR. Polynomial additional results in different values to a conventional addition. Multiplying two polynomials over {0,1} involves first determining the partial products as done in conventional multiply, then the partial products are exclusive ORed instead of being added conventionally. Polynomial multiplication results in different values to conventional multiplication because it requires polynomial addition of the partial products. NEON technology is IEEE 754-1985 compliant, but only supports round-to-nearest rounding mode. This is the rounding mode used by most high-level languages, such as C and Java. Additionally, NEON instructions always treats denormals as zero. 7.2.3 NEON registers The register bank can be viewed as either sixteen 128-bit registers (Q0-Q15) or as thirty-two 64-bit registers (D0-D31). Each of the Q0-Q15 registers maps to a pair of D registers, as shown in Figure 7-3. D0 D1 D2 D3 D30 D31 Q0 Q1 Q15 Figure 7-3 NEON register bank The view of registers in Figure 7-4 on page 7-7 is determined by form of the instruction used. so that the software does not have to explicitly change state. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-6 Introducing NEON 4 x 32 bit Data 8 x 16 bit Data 128 bit Q register 16 x 8 bit Data D1 D0 4 x 32 bit Data Q0 Figure 7-4 NEON registers Individual elements can also be accessed as scalars.The advantage of the dual view is that it accommodates mathematical operations that widen or narrow the result. For example multiplying two D registers gives a Q register result. The dual-view enables the register bank to be used more efficiently. NEON data processing instructions are typically available in Normal, Long, Wide, Narrow and Saturating variants. • Normal instructions can operate on any vector types, and produce result vectors the same size, and usually the same type, as the operand vectors. • Long instructions operate on doubleword vector operands and produce a quadword vector result. The result elements are usually twice the width of the operands, and of the same type. Long instructions are specified using an L appended to the instruction. Figure 7-5 shows this, with input operands being promoted before the operation. Dn Dm Qd Figure 7-5 NEON long instructions • ARM DEN0013D ID012214 Wide instructions operate on a doubleword vector operand and a quadword vector operand, producing a quadword vector result. The result elements and the first operand are twice the width of the second operand elements. Wide instructions have a W appended to the instruction. Figure 7-6 on page 7-8 shows this, with the input doubleword operands being promoted before the operation. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-7 Introducing NEON Qn Dn Qd Figure 7-6 NEON wide instructions • Narrow instructions operate on quadword vector operands, and produce a doubleword vector result. The result elements are usually half the width of the operand elements. Narrow instructions are specified using an N appended to the instruction. Figure 7-7 shows this, with input operands being demoted before the operation. Qn Qm Dd Figure 7-7 NEON narrow instructions Some NEON instructions act on scalars together with vectors. The scalars can be 8-, 16-, 32-, or 64-bit. Instructions that use scalars can access any element in the register bank, although there are differences for multiply instructions. The instruction uses an index into a doubleword vector to specify the scalar value. Multiply instructions only support 16-bit or 32-bit scalars, and can only access the first 32 scalars in the register bank (that is, D0-D7 for 16-bit scalars or D0-D15 for 32-bit scalars). 7.2.4 NEON instruction set All mnemonics for NEON instructions (as with VFP) begin with the letter “V”. Instructions are generally able to operate on different data types, with this being specified in the instruction encoding. The size is indicated with a suffix to the instruction. The number of elements is indicated by the specified register size. For example, looking at the instruction ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-8 Introducing NEON VADD.I8 D0, D1, D2 where: • VADD indicates a NEON ADD operation. • The I8 suffix indicates that 8-bit integers are to be added • D0, D1 and D2 specify the 64-bit registers used (D0 for the result destination, D1 and D2 for the operands). So this instruction performs eight additions in parallel. There are operations that have different size registers for input and output. VMULL.S16 Q2, D8, D9 This instruction performs four 16-bit multiplies of data packed in D8 and D9 and produces four 32-bit results packed into Q2. The VCVT instruction converts elements between single-precision floating-point and 32-bit integer, fixed-point, and (if implemented) half-precision floating-point. NEON includes load and store instructions that can load or store individual or multiple values to a register. In addition, there are instructions that can transfer blocks of data between multiple registers and memory. It is also possible to interleave or de-interleave data during such multiple transfers. The following modifiers can be used with certain Advanced SIMD instructions (some modifiers can be used only with a small subset of the available instructions): Q The instruction uses saturating arithmetic, so that the result is saturated within the range of the specified data type. The sticky QC bit in the FPSCR is set if saturation occurs in any lane. VQADD is an example of such an instruction. H The instruction will halve the result. It does this by shifting right by one place (effectively a divide by two with truncation). VHADD is an example of such an instruction – it could be used to calculate the mean of two inputs. D The instruction doubles the result and saturates. This is commonly required when multiplying numbers in Q15 format, where an additional doubling is required to get the result into the correct form. R The instruction will perform rounding on the result, equivalent to adding 0.5 to the result before truncating. VRHADD is an example of this. Instructions have the following general format: V{ } { }{ }{. }{ }, src1, src2 where: is one of the previously described Modifiers (Q, H, D, R) - operation (for example, ADD, SUB, MUL) - Shape (L, W or N, as described in NEON registers on page 7-6) - Condition, used with IT instruction <.dt> - Data type - Destination ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-9 Introducing NEON - Source operand 1 - Source operand 2. The NEON instruction set includes a range of vector addition and subtraction operations, including pairwise adding, that adds adjacent vector elements together. There are a number of multiply operations, including multiply-accumulate and multiply-subtract and doubling and saturating options. There is no SIMD division operation, but such an operation can be performed using the VRECPE (Vector Reciprocal Estimate) and VCREPS (Vector Reciprocal Step) instructions to perform Newton-Raphson iteration. Similarly, there is no vector square root instruction, but VRSQRTE, VRSQRTS, and multiplies can be used to compute square roots. Shift left, right and insert operations are also available, along with instructions that select minimum or maximum values. Common logic operations (AND, OR, EOR, AND NOT and OR NOT) can be performed. The instruction set also includes the ability to count numbers of bits set in an element or to count leading zeros or sign bits. There are a number of different instructions to move data between registers, or between elements. It is also possible for instructions to swap or duplicate registers, to perform reversal, matrix transposition and extract individual vector elements. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-10 Introducing NEON 7.3 NEON C Compiler and assembler Code targeted at NEON hardware can be written in C or in assembly language and a range of tools and libraries is available to support this. In many cases, it might be preferable to use NEON code within a larger C/C++ function, rather than in a separate file to be processed by the assembler. This can be done using NEON intrinsics. 7.3.1 Vectorization A vectorizing compiler can take your C or C++ source code and parallelize it in a way that enables efficient usage of NEON hardware. This means you can write portable C code, while still obtaining the levels of performance made possible by NEON. The C language does not specify parallelizing behavior, so it can be necessary to provide hints to the compiler about this. For example, it might be necessary to use the __restrict keyword when defining pointers. This has the effect of guaranteeing that pointers will not address overlapping regions of memory. It can also be helpful to ensure that the number of loop iterations is a multiple of four or eight. Automatic vectorization is specified with the GCC option -ftree-vectorize (along with –mfpu=neon). Using the ARM Compiler, you must specify optimization level –O2 (or –O3), -Otime and –-vectorize. 7.3.2 Detecting NEON As NEON hardware can be omitted from a processor implementation, it might be necessary to test for its presence. Build-time NEON selection This is the easiest way to select NEON. In armcc (RVCT 4.0 and later), or GCC, the predefined macro __ARM_NEON__ is defined when a suitable set of processor and FPU options is provided to the compiler. The armasm equivalent predefined macro is TARGET_FEATURE_NEON. This could be used to have a C source file that has both NEON and non-NEON optimized versions. Run-time NEON detection To detect NEON at run-time requires help from the operating system, since the ARM architecture intentionally does not expose processor capabilities to user-mode applications. Under Linux, /proc/cpuinfo contains this information in human-readable form. On Tegra2 (a dual-core Cortex-A9 processor with FPU), cat /proc/cpuinfo reports: … Features … : swp half thumb fastmult vfp edsp thumbee vfpv3 vfpv3d16 The ARM quad-core Cortex-A9 processor with NEON gives a slightly different result: … Features … : swp half thumb fastmult vfp edsp thumbee neon vfpv3 As the /proc/cpuinfo output is text based, it is often preferred to look at the auxiliary vector /proc/self/auxv. This contains the kernel hwcap in a binary format. The /proc/self/auxv file can be easily searched for the AT_HWCAP record, to check for the HWCAP_NEON bit (4096). ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-11 Introducing NEON Some Linux distributions, for example, Ubuntu 09.10 or later, take advantage of NEON transparently. The ld.so linker script is modified to read the hwcap using glibc, and add an additional search path for NEON-enabled shared libraries. In the case of Ubuntu, a new search path /lib/neon/vfp contains NEON-optimized versions of libraries from /lib. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 7-12 Chapter 8 Caches The word cache derives from the French verb cacher, “to hide”.The application of this word to a processor is obvious – a cache is where the processor stores instructions and data, hidden from the programmer and system. In many cases, it would be true to say that the cache is transparent to, or hidden from you. But very often, as we shall see, it is important to understand the operation of the cache in detail. When the ARM architecture was first developed, the clock speed of the processor and the access speeds of memory were broadly similar. Processor cores today are much more complicated and can be clocked orders of magnitude faster. However, the frequency of the external buses and of memory devices has not scaled to the same extent. It is possible to implement small blocks of on-chip SRAM that can operate at the same speeds as the core, but such RAM is very expensive in comparison to standard DRAM blocks, that can have thousands of times more capacity. In many ARM processor-based systems, access to external memory will take tens or even hundreds of core cycles. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 8-1 Caches Core Internal cache (L1 Cache) Main memory External cache (L2 cache) Bus Figure 8-1 A basic cache arrangement Essentially, a cache is a small, fast block of memory that (conceptually at least) sits between the core and main memory. It holds copies of items in main memory. Accesses to the cache memory happen significantly faster than those to main memory. Because the cache holds only a subset of the contents of main memory, it must store both the address of the item in main memory and the associated data. Whenever the core wants to read or write a particular address, it will first look for it in the cache. If it finds the address in the cache, it will use the data in the cache, rather than having to perform an access to main memory. This significantly increases the potential performance of the system, by reducing the effect of slow external memory access times. It also reduces the power consumption of the system, by avoiding the need to drive external signals. Cache sizes are small relative to the overall memory used in the system. Larger caches make for more expensive chips. In addition, making an internal core cache larger can potentially limit the maximum speed of the core. Efficient use of this limited resource is a key part of writing efficient applications to run on a core. On-chip SRAM can be used to implement caches, that hold temporary copies of instructions and data from main memory. Code and data have the properties of temporal and spatial locality. This means that programs tend to re-use the same addresses over time (temporal locality) and tend to use addresses that are near to each other (spatial locality). Code, for instance, can contain loops, meaning that the same code gets executed repeatedly or a function can be called multiple times. Data accesses (for example, to the stack) can be limited to small regions of memory. It is this fact that access to RAM by the core exhibits such locality, and is not truly random, that enables caches to be successful. The write buffer is a block that decouples writes being done by the core when executing store instructions from the external memory bus. The core places the address, control and data values associated with the store into a set of hardware buffers. Like the cache, it sits between the core and main memory. This enables the core to move on and execute the next instructions without having to stop and wait for the slow main memory to actually complete the write operation. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 8-2 Caches 8.1 Why do caches help? Caches speed things up, as we have seen, because program execution is not random. Programs tend to access the same sets of data repeatedly and execute the same sets of instructions repeatedly. By moving code or data into faster memory when it is first accessed, subsequent accesses to that code or data become much faster. The initial access that provided the data to the cache is no faster than normal. It is any subsequent accesses to the cached values that are faster, and it is from this that the performance increase derives. The core hardware will check all instruction fetches and data reads or writes in the cache, although obviously you must mark some parts of memory (those containing peripheral devices, for example) as non-cacheable. Because the cache holds only a subset of main memory, you require a way to determine (quickly) whether the address you are looking for is in the cache. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 8-3 Caches 8.2 Cache drawbacks It might appear that caches and write buffers are automatically a benefit, as they speed up program execution. However, they also add some problems that are not present in an uncached core. One such drawback is that program execution time can become non-deterministic. What this means is that, because the cache is small and holds only a subset of main memory, it fills rapidly as a program executes. When the cache is full, existing code or data must be removed to make room for new items. So, at any given time, it is not normally possible for an application to be certain whether or not a particular instruction or data item is to be found in the cache. This means that the execution time of a particular piece of code can vary significantly. This can be something of a problem in hard real-time systems where strongly deterministic behavior is required. Furthermore, you require a way to control how different parts of memory are accessed by the cache and write buffer. In some cases, you want the core to read up-to-date data from an external device, such as a peripheral. It would not be sensible to use a cached value of a timer peripheral, for example. Sometimes you want the core to stop and wait for a store to complete. So caches and write buffers give you some extra work to do. Occasionally the contents of cache and external memory might not be the same, this is because the processor can update the cache contents, which have not yet been written back to main memory. Alternatively, an agent might update main memory after a core has taken its own copy. This is a problem of coherency. This can be a particular problem when you have multiple cores or memory agents like an external DMA controller. Coherency issues are described later in the book. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 8-4 Caches 8.3 Memory hierarchy In computer science, a memory hierarchy refers to a hierarchy of memory types, with faster and smaller memories closer to the core and slower and larger memory farther away. In most systems, you can have secondary storage, such as disk drives and primary storage such as Flash, SRAM and DRAM. In embedded systems, you typically sub-divide this into on-chip and off-chip memory. Memory that is on the same chip (or at least in the same package) as the core will typically be much faster. A cache can be included at any level in the hierarchy and can improve system performance where there is an access time difference between different parts of the memory system. In ARM processor-based systems, level 1 (L1) caches are typically connected directly to the core logic that fetches instructions and handles load and store instructions. These are Harvard caches, that is, there are separate caches for instructions and for data that effectively appear as part of the core. Data cache Core Instruction cache Main memory L2 cache Bus Figure 8-2 Typical Harvard cache Over the years, the size of L1 caches has increased, because of SRAM size and speed improvements. At the time of writing, 16KB or 32KB cache sizes are most common, as these are the largest RAM sizes capable of providing single cycle access at a core speed of 1GHz or more. Many ARM systems have, in addition, a level 2 (L2) cache. This is larger than the L1 cache (typically 256KB, 512KB or 1MB), but slower and unified (holding both instructions and data). It can be inside the core itself, or be implemented as an external block, placed between the core and the rest of the memory system. The ARM L2C-310 is an example of such an external L2 cache controller block. In addition, cores can be implemented in clusters in which each core has its own level 1 cache. Such systems require mechanisms to maintain coherency between caches, so that when one core changes a memory location, that change is made visible to other cores that share that memory. This is described in more detail when we look at multi-core processors. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 8-5 Caches 8.4 Cache architecture In a von Neumann architecture, a single cache is used for instruction and data (a unified cache). A modified Harvard architecture has separate instruction and data buses and therefore there are two caches, an instruction cache (I-cache) and a data cache (D-cache). In many ARM systems, you can have distinct instruction and data level 1 caches backed by a unified level 2 cache. The cache requires to hold an address, some data and some status information. The top bits of the 32-bit address tells the cache where the information came from in main memory and is known as the tag. The total cache size is a measure of the amount of data it can hold; the RAMs used to hold tag values are not included in the calculation. The tag does, however, take up physical space in the cache. It would be inefficient to hold one word of data for each tag address, so several locations are typically grouped together under the same tag. This logical block is commonly known as a cache line. The middle bits of the address, or index, identify the line. The index is used as address for the cache RAMs and does not require storage as a part of the tag. This will be covered in more detail later in this chapter. A cache line is said to be valid when it contains cached data or instructions, and invalid when it does not. This means that the bottom few bits of the address (the offset) are not required to be stored in the tag – you require the address of a whole line, not of each byte within the line, so the five or six least significant bits will always be 0. Associated with each line of data are one or more status bits. Typically, you will have a valid bit, that marks the line as containing data that can be used. (This means that the address tag represents some real value.) In a data cache you might also have one or more dirty bits that mark whether the cache line (or part of it) holds data that is not the same as (newer than) the contents of main memory. 8.4.1 Cache terminology A brief summary of some of the terms used might be helpful: Line Offset 31 Data RAM Tag RAM 32-bit address Tag Index 0 Offset Index Set Way Tag Figure 8-3 Cache terminology ARM DEN0013D ID012214 • A line refers to the smallest loadable unit of a cache, a block of contiguous words from main memory. • The index is the part of a memory address that determines in which line(s) of the cache the address can be found. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 8-6 Caches 8.4.2 • A way is a subdivision of a cache, each way being of equal size and indexed in the same fashion. The line associated with a particular index value from each cache way grouped together forms a set. • The tag is the part of a memory address stored within the cache that identifies the main memory address associated with a line of data. Direct mapped caches There are different ways of implementing caches, the simplest of which is a direct mapped cache. In a direct mapped cache, each location in main memory maps to a single location in the cache. However, as main memory is many times larger than the cache, many addresses will map to the same cache location. Figure 8-4 shows a small cache, with four words per line and four lines. Main memory Cache 0x0000.0000 0x0000.0010 0x0000.0020 0x0000.0030 0x0000.0040 0x0000.0050 0x0000.0060 0x0000.0070 0x0000.0080 0x0000.0090 Figure 8-4 Direct mapped cache operation This means that the cache controller will use two bits of the address (bits [3:2]) as the offset to select a word within the line and two bits of the address (bits [5:4]) as the index to select one of the four available lines. The remaining bits of the address (bits [31:6]) will be stored as a tag value. Address Tag Index 6 5 31 4 Line 3 2 1 0 Figure 8-5 Cache address To look up a particular address in the cache, the hardware extracts the index bits from the address and reads the tag value associated with that line in the cache. If the two are the same and the valid bit indicates that the line contains valid data, it has a hit. It can then extract the data value from the relevant word of the cache line, using the offset and byte portion of the address. If the line contains valid data, but does not generate a hit (that is, the tag shows that the cache holds a different address in main memory) then the cache line is removed and is replaced by data from the requested address. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 8-7 Caches It should be clear that all main memory addresses with the same value of bits [5:4] will map to the same line in the cache. Only one of those lines can be in the cache at any given time. This means a problem called thrashing can easily occur. Consider a loop that repeatedly accesses address 0x00, 0x40 and 0x80, as in the following code: void add_array(int *data1, int *data2, int *result, int size) { int i; for (i=0 ; i This performs a PC-relative branch. It is suitable for calling exception handler code that is close enough in memory that the 24-bit field provided in the branch instruction is large enough to encode the offset. LDR PC, [PC, #offset] This loads the PC from a memory location whose address is defined relative to the address of the exception instruction. This lets the exception handler be placed at any arbitrary address within the full 32-bit memory space (but takes some extra cycles relative to the simple branch). When the core is operating in Hyp mode, it uses Hyp mode vector entries, that are taken from a dedicated vector table belonging to the hypervisor. Hypervisor mode is entered through a special exception process called Hyp trap entry that makes use of the previously reserved 0x14 address within the vector table. A dedicated register (the Hyp Syndrome Register) gives information to the hypervisor about the exception or other reason for entering the hypervisor (a trapped CP15 operation, for example). 11.1.4 FIQ and IRQ FIQ is reserved for a single, high-priority interrupt source that requires a guaranteed fast response time, with IRQ used for all of the other interrupts in the system. As FIQ is the last entry in the vector table, the FIQ handler can be placed directly at the vector location and run sequentially from that address. This avoids a branch instruction and any associated delay, speeding up FIQ response times. The extra banked registers available in FIQ mode relative to other modes allows state to be retained between calls to the FIQ handler, again increasing execution speed by potentially removing the need to push some registers before using them. A further key difference between IRQ and FIQ is that the FIQ handler is not expected to generate any other exceptions. FIQ is therefore reserved for special system-specific devices which have all their memory mapped and no need to make SVC calls to access kernel functions (so FIQ can be used only by code which does not need to use the kernel API). FIQ is not typically used by Linux. As the kernel is architecture-independent, it does not have a concept of multiple forms of interrupt. Some systems running Linux can still make use of FIQ, but as the Linux kernel never disables FIQs they have priority over anything else in the system and so some care is required. 11.1.5 The return instruction The Link Register (LR) is used to store the appropriate return address for the PC after the exception has been handled. Its value must be modified as shown in Table 11-3 on page 11-9, depending on the type of exception occurred.The ARM Architecture Reference Manual defines ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 11-8 Exception Handling the LR values that are appropriate (the definition derives from the values that were convenient for early hardware implementations) Table 11-3 also supplies an example return instruction from the exception that includes this adjustment. Table 11-3 Link Register Adjustments ARM DEN0013D ID012214 Exception Adjustment Return instruction Instruction returned to SVC 0 MOVS PC, R14 Next instruction Undef 0 MOVS PC, R14 Next instruction Prefetch Abort -4 SUBS PC, R14, #4 Aborting instruction Data abort -8 SUBS PC, R14, #8 Aborting instruction if precise FIQ -4 SUBS PC, R14, #4 Next instruction IRQ -4 SUBS PC, R14, #4 Next instruction Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 11-9 Exception Handling 11.2 Exception handling When an exception occurs, the ARM core automatically does the following: 1. Copies the CPSR to the SPSR_ , the banked register specific to the (non-user) mode of operation. 2. Stores a return address in the Link Register (LR) of the new mode. 3. Modifies the CPSR mode bits to a mode associated with the exception type. • The other CPSR mode bits are set to values determined by bits in the CP15 System Control Register. • The T bit is set to the value given by the CP15 TE bit. • The J bit is cleared and the E bit (Endianness) is set to the value of the EE (Exception Endianness) bit. This enables exceptions to always run in ARM or Thumb state and in little or big-endian, irrespective of the state the core was in before the exception. 4. Sets the PC to point to the relevant instruction from the exception vector table. 1 Save status Change status Program flow Figure 11-1 Taking the exception When in the new mode the core will access the register associated with that mode, as shown in Figure 3-5 on page 3-7. It will almost always be necessary for the exception handler software to save registers onto the stack immediately on exception entry. FIQ mode has more banked registers and so a simple handler might be able to be written in a way that requires no stack usage. A special assembly language instruction is provided to assist with saving the necessary registers, called SRS (Store Return State). This instruction pushes the LR and SPSR onto the stack of any mode; the stack to be used is specified by the instruction operand. 11.2.1 Exit from an exception handler To return from an exception handler, two separate operations must take place atomically: 1. ARM DEN0013D ID012214 Restore the CPSR from the saved SPSR. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 11-10 Exception Handling 1 Save status Change status Program flow Exception handler 2 Return from exception Figure 11-2 Returning from an exception 2. Set the PC to the return address offset, see Table 11-1 on page 11-6. In the ARM architecture this can be achieved either by using the RFE instruction or any flag-setting data processing operation (with the S suffix) with the PC as the destination register, such as SUBS PC, LR, #offset (note the S). The Return From Exception (RFE) instruction pops the link register and SPSR off the current mode stack. There are a number of ways to achieve this. • You can use a data processing instruction to adjust and copy the LR into the PC, for example: SUBS pc, lr, #4 Specifying the S means the SPSR is copied to the CPSR at the same time. If the exception handler entry code uses the stack to store registers that must be preserved while it handles the exception, it can return using a load multiple instruction with the ^ qualifier. For example, an exception handler can return in one instruction using: LDMFD sp! {pc}^ LDMFD sp!,{R0-R12,pc}^ The ^ qualifier in this example means the SPSR is copied to the CPSR at the same time. To do this, the exception handler must save the following onto the stack: — All the work registers in use when the handler is invoked. — The link register, modified to produce the same effect as the data processing instructions. Note You cannot use 16-bit Thumb instructions to return from exceptions because these are unable to restore the CPSR. • The RFE instruction (See RFE on page A-35) restores the PC and SPSR from the stack of the current mode. RFEFD sp! ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 11-11 Exception Handling 11.3 Other exception handlers This section briefly describes handlers for aborts, undefined instructions and SVC exceptions and considers how interrupts are handled by the Linux kernel. Reset handlers are covered in Chapter 13 Boot Code. 11.3.1 Abort handler Abort handler code can vary significantly between systems. In many embedded systems, an abort indicates an unexpected error and the handler will record any diagnostic information, report the error and have the application (or system) quit gracefully. In systems that support virtual memory using an MMU, the abort handler can load the required virtual page into physical memory. In effect, it tries to fix the cause of the original abort and then return to the instruction that aborted and re-execute it. Chapter 9 gives more information about how Linux does this. CP15 registers provide the address of the memory access that caused an abort (the Fault Address Register) and the reason for the abort (Fault Status Register). The reason might be lack of access permission, an external abort or an address translation fault. In addition, the link register (with a –8 or –4 adjustment, depending on whether the abort was caused by an instruction fetch or a data access), gives the address of the instruction that caused the abort exception. By examining these registers, the last instruction executed and possibly other things in the system, for example, translation table entries, the abort handler can determine what action to take. 11.3.2 Undefined instruction handling An undefined instruction exception is taken if the core tries to execute an instruction with an opcode, described in the ARM architecture specification as UNDEFINED, or when a coprocessor instruction is executed but no coprocessor recognizes it as an instruction that it can execute. In some systems, it is possible that code includes instructions for a coprocessor (such as a VFP coprocessor), but that no corresponding VFP hardware is present in the system. In addition, it is possible that the VFP hardware cannot handle the particular instruction and wants to call software to emulate it. Alternatively, the VFP hardware might be disabled, and you take the exception so that it can be enabled and then re-execute the instruction. Such emulators are called through the undefined instruction vector. They examine the instruction opcode that caused the exception and determine what action to take (for example, perform the appropriate floating-point operation in software). In some cases, such handlers might have to be daisy-chained together (for example, there might be multiple coprocessors to emulate). If there is no software that makes use of undefined or coprocessor instructions, the handler for the exception must record suitable debug information and kill the application that failed because of this unexpected event. An additional use for the undefined instruction exception in some cases is to implement user breakpoints, see Chapter 24 Debug for more information on breakpoints. See also the description of the Linux context switch for VFP in Context switching on page 6-10. 11.3.3 SVC exception handling A supervisor call (SVC) is typically used to enable User mode code to access OS functions. For example, if user code wants to access privileged parts of the system (for example to perform file I/O) it will typically do this using an SVC instruction. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 11-12 Exception Handling Parameters can be passed to the SVC handler either in registers or (less frequently) by using the comment field within the opcode. An exception handler might have to determine whether the core was in ARM or Thumb state when the exception occurred. SVC handlers, especially, might have to read the instruction set state. This is done by examining the SPSR T-bit. This bit is set for Thumb state and clear for ARM state. Both ARM and Thumb instruction sets have the SVC instruction. When calling SVCs from Thumb state, you must consider the following: • The instruction address is at LR-2, rather than LR-4. • The instruction itself is 16-bit, and so requires a halfword load, • The SVC number is held in 8 bits instead of the 24 bits in ARM state. • The SVC number is held in 8 bits instead of the 24 bits in ARM state. Code to illustrate SVC usage with the Linux kernel is shown in Example 11-1. Example 11-1 Linux kernel SVC usage _start: MOV R0, #1 ADR R1, msgtext MOV R2, #13 MOV R7, #4 SVC #0 .... .align 2 msgtxt: .asciz "Hello World\n" @ @ @ @ STDOUT Address Length sys_write The SVC #0 instruction makes the ARM core take the SVC exception, the mechanism to access a kernel function. Register R7 defines which system call you want (in this case, sys_write). The other parameters are passed in registers; for sys_write you have R0 telling where to write to, R1 pointing to the characters to be written and R2 giving the length of the string. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 11-13 Exception Handling 11.4 Linux exception program flow Linux utilizes a cross-platform framework for exception handling that does not distinguish between different privileged core modes when handling exceptions. Therefore, the ARM implementation uses an exception handler stub to enable the kernel to handle all exceptions in SVC mode. All exceptions other than SVC and FIQ use the stub to switch to SVC mode and invoke the correct exception handler. 11.4.1 Boot process During the boot process, the kernel will allocate a 4KB page as the vector page. It maps this to the location of the exception vectors, virtual address 0xFFFF0000 or 0x00000000. This is done by devicemaps_init() in the file arch/arm/mm/mmu.c. This is invoked very early in the ARM system boot. After this, trap_init (in arch/arm/kernel/traps.c), copies the exception vector table, exception stubs and kuser helpers into the vector page. The exception vector table obviously has to be copied to the start of the vector page, the exception stubs being copied to address 0x200 (and kuser helpers copied to the top of the page, at 0x1000 - kuser_sz), using a series of memcpy() operations, as shown in Example 11-2. Example 11-2 Copying exception vectors during Linux boot unsigned long vectors = CONFIG_VECTORS_BASE; memcpy((void *)vectors, __vectors_start, __vectors_end - __vectors_start); memcpy((void *)vectors + 0x200, __stubs_start, __stubs_end - __stubs_start); memcpy((void *)vectors + 0x1000 - kuser_sz, __kuser_helper_start, kuser_sz); When the copying is complete, the kernel exception handler is in its runtime dynamic status, ready to handle exceptions 11.4.2 Interrupt dispatch There are two different handlers, __irq_usr and __irq_svc. These save all of the core registers and use a macro get_irqnr_and_base that indicates if there is an interrupt pending. The handlers loop around this code until no interrupts remain. If there is an interrupt, the code will branch to do_IRQ that exists in arch/arm/kernel/irq.c. At this point, the code is the same in all architectures and you call an appropriate handler written in C. There is, however, an additional point to consider. When the interrupt is completed, you would normally have to check whether or not the handler has done something that requires the kernel scheduler to be called. If the scheduler decides to go to a different thread, the one that was originally interrupted stays dormant until it is selected to run again. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 11-14 Chapter 12 Interrupt Handling Older versions of the ARM architecture enabled implementers considerable freedom in the design of an external interrupt controller, with no agreement over the number or types of interrupts or the software model to be used to interface to the interrupt controller block. The Generic Interrupt Controller v2 (GIC) architecture provides a much more tightly controlled specification, with a greater degree of consistency between interrupt controllers from different manufacturers. This enables interrupt handler code to be more portable. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 12-1 Interrupt Handling 12.1 External interrupt requests Types of exception on page 11-3, described how all ARM cores have two external interrupt requests, FIQ and IRQ. Both of these are level-sensitive active-LOW inputs. Individual implementations have interrupt controllers that accept interrupt requests from a wide variety of external sources and map them onto FIQ or IRQ, causing the core to take an exception. In general, an interrupt exception can be taken only when the appropriate CPSR disable bit (the F and I bits respectively) is clear and if the corresponding input is asserted. The CPS instruction provides a simple mechanism to enable or disable the exceptions controlled by CPSR A, I and F bits (asynchronous abort, IRQ and FIQ respectively). CPS IE or CPS ID will enable or disable exceptions respectively. The exceptions to be enabled or disabled are specified using one or more of the letters A, I and F. Exceptions whose corresponding letters are omitted will not be modified. In Cortex-A series processors, it is possible to configure the core so that FIQs cannot be masked by software. This is known as Non-Maskable FIQ and is controlled by a hardware configuration input signal that is sampled when the core is reset. They will still be masked automatically on taking an FIQ exception. 12.1.1 Assigning interrupts A system will always have an interrupt controller that accepts and arbitrates interrupts from multiple sources. This typically contains a number of registers enabling software running on the core to mask individual interrupt sources, to acknowledge interrupts from external devices, to assign priorities to individual interrupt sources and to determine which interrupt sources are currently requesting attention or require servicing. This interrupt controller can be a design specific to the system, or it can be an implementation of the ARM Generic Interrupt Controller (GIC) architecture, described in The Generic Interrupt Controller on page 12-7. 12.1.2 Simplistic interrupt handling This represents the simplest kind of interrupt handler. On taking an interrupt, additional interrupts of the same kind are disabled until explicitly enabled later. We can only handle additional interrupts at the completion of the first interrupt request and there is no scope for a higher priority or more urgent interrupt to be handled during this time. This is not generally suitable for complex embedded systems, but it is useful to examine before proceeding to a more realistic example, in this case of a non re-entrant interrupt handler. The steps taken to handle an interrupt are as follows: 1. ARM DEN0013D ID012214 An IRQ exception is raised by external hardware. The core performs several steps automatically. The contents of the PC in the current execution mode are stored in LR_IRQ. The CPSR register is copied to SPSR_IRQ. The CPSR content is updated so that the mode bits reflects the IRQ mode, and the I bit is set to mask additional IRQs. The PC is set to the IRQ entry in the vector table. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 12-2 Interrupt Handling rq _i PC -> LR R PS C Program flow R -> SP rq _i R ASM IRQ Handler Figure 12-1 Save the context of the program 2. The instruction at the IRQ entry in the vector table (a branch to the interrupt handler) is executed. 3. The interrupt handler saves the context of the interrupted program, that is, it pushes onto the stack any registers that will be corrupted by the handler. These registers will be popped from stack when the handler finishes execution. Stack LR_irq rq _i PC -> LR P SR C Program flow -> rq _i R PR S ASM IRQ Handler Figure 12-2 4. ARM DEN0013D ID012214 The interrupt handler determines which interrupt source must be processed and calls the appropriate device driver. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 12-3 Interrupt Handling 5. Prepare the core to switch to previous execution state by copying the SPSR_IRQ to CPSR, and restoring the context saved earlier, and finally the PC is restored from LR_IRQ. Stack LR_irq rq _i PC -> LR R PS R SP -> C Program flow SP ASM IRQ Handler LR SR _i rq _i R rq _i -> rq -> C PS PC R Figure 12-3 The same sequence is also applicable to an FIQ interrupt. A very simple interrupt handler is shown in Example 12-1. Example 12-1 Simple interrupt handler IRQ_Handler PUSH {r0-r3, r12, lr} BL BL POP {r0-r3, r12, lr} SUBS pc, lr, #4 12.1.3 @ @ @ @ @ Store AAPCS registers and LR onto the IRQ mode stack identify_and_clear_source C-irq_handler Restore registers and return from exception using modified LR Nested interrupt handling Nested interrupt handling is where the software is prepared to accept another interrupt, even before it finishes handling the current interrupt. This enables you to prioritize interrupts and make significant improvements to the latency of high priority events at the cost of additional complexity. It is worth noting that nested interrupt handling is a choice made by the software, by virtue of interrupt priority configuration and interrupt control, rather than imposed by hardware. A reentrant interrupt handler must save the IRQ state and then switch core modes, and save the state for the new core mode, before it branches to a nested subroutine or C function with interrupts enabled. This is because a fresh interrupt could occur at any time, which would cause the core to store the return address of the new interrupt and overwrite the original interrupt. When the original interrupt attempts to return to the main program, it will cause the system to fail. The nested handler must change into an alternative kernel mode before re-enabling interrupts in order to prevent this. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 12-4 Interrupt Handling SPSR_irq is corrupted rq SR -> S Program flow SR -> R PS S P P C rq _i _i R PS C ASM IRQ Handler Figure 12-4 Nested interrupts Note A computer program is reentrant if it can be interrupted in the middle of its execution and then be called again before the previous version has completed. In Figure 12-4 the value of SPSR must be preserved before interrupts are re-enabled. If it is not, any new interrupt will overwrite the value of SPSR_irq. The solution to this is to stack the SPSR before re-enabling the interrupts by using the following: SRSFD sp!, #0x12 Additionally, using the BL instruction within the interrupt handler code will cause LR_IRQ corruption. The solution is to switch to Supervisor mode before using the BL instruction. A reentrant interrupt handler must therefore take the following steps after an IRQ exception is raised and control is transferred to the interrupt handler in the way previously described. ARM DEN0013D ID012214 1. The interrupt handler saves the context of the interrupted program (that is, it pushes onto the alternative kernel mode stack any registers that will be corrupted by the handler, including the return address and SPSR_IRQ). 2. It determines which interrupt source must be processed and clears the source in the external hardware (preventing it from immediately triggering another interrupt). 3. The interrupt handler changes to the core SVC mode, leaving the CPSR I bit set (interrupts are still disabled). 4. The interrupt handler saves the exception return address on the stack (a stack for the new mode, located in kernel memory) and re-enables interrupts. 5. It calls the appropriate handler code. 6. On completion, the interrupt handler disables IRQ and pops the exception return address from the stack. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 12-5 Interrupt Handling 7. It restores the context of the interrupted program directly from the alternative kernel mode stack. This includes restoring the PC, and the CPSR which switches back to the previous execution mode. If the SPSR does not have the I bit set then the operation also re-enables interrupts. Sample code for a nested interrupt handler (for non-vectored interrupts) is given in Example 12-2. Example 12-2 Nested interrupt handler IRQ_Handler SUB lr, lr, #4 SRSFD sp!, #0x1f @ use SRS to save LR_irq and SPSR_irq in one step onto the @ System mode stack CPS #0x1f @ Use CPS to switch to system mode PUSH AND {r0-r3, r12} r1, sp, #4 @S tore remaining AAPCS registers on the System mode stack @ Ensure stack is 8-byte aligned. Store adjustment and @ LR_sys to stack SUB PUSH sp, sp, r1 {r1, lr} BL ARM DEN0013D ID012214 @ identify_and_clear_source CPSIE i @ Enable IRQ with CPS BL C_irq_handler CPSID i @ Disable IRQ with CPS POP ADD POP RFEFD {r1, lr} sp, sp, r1 {r0-r3, r12} sp! @ @ @ @ Restore LR_sys Unadjust stack Restore AAPCS registers Return using RFE from the System mode stack. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 12-6 Interrupt Handling 12.2 The Generic Interrupt Controller The GIC architecture defines a Generic Interrupt Controller (GIC) that comprises a set of hardware resources for managing interrupts in a single or multi-core system. The GIC provides memory-mapped registers that can be used to manage interrupt sources and behavior and (in multi-core systems) for routing interrupts to individual cores. It enables software to mask, enable and disable interrupts from individual sources, to prioritize (in hardware) individual sources and to generate software interrupts. It also provides support for the TrustZone Security Extensions described in Chapter 21 Security. The GIC accepts interrupts asserted at the system level and can signal them to each core it is connected to, potentially resulting in an IRQ or FIQ exception being taken. From a software perspective, a GIC has two major functional blocks: Distributor to which all interrupt sources in the system are wired. The distributor has registers to control the properties of individual interrupts such as priority, state, security, routing information and enable status. The distributor determines which interrupt is to be forwarded to a core, through the attached CPU interface. CPU Interface through which a core receives an interrupt. The CPU interface hosts registers to mask, identify and control states of interrupts forwarded to that core. There is a separate CPU interface for each core in the system. Interrupts are identified in the software by a number, called an interrupt ID. An interrupt ID uniquely corresponds to an interrupt source. Software can use the interrupt ID to identify the source of interrupt and to invoke the corresponding handler to service the interrupt. The exact interrupt ID presented to the software is determined by the system design, Interrupts can be of a number of different types: Software Generated Interrupt (SGI) This is generated explicitly by software by writing to a dedicated distributor register, the Software Generated Interrupt Register (ICDSGIR). It is most commonly used for inter-core communication. SGIs can be targeted at all, or a selected group of cores in the system. Interrupt numbers 0-15 are reserved for this. The exact interrupt number used for communication is at the discretion of software. Private Peripheral Interrupt (PPI) This is generated by a peripheral that is private to an individual core. Interrupt numbers 16-31 are reserved for this. These identify interrupt sources private to the core, and is independent of the same source on another core, for example, per-core timer. Shared Peripheral Interrupt (SPI) This is generated by a peripheral that the Interrupt Controller can route to more than one core. Interrupt numbers 32-1020 are used for this. SPIs are used to signal interrupts from various peripherals accessible across the whole the system. Interrupts can either be edge-triggered (considered to be asserted when the Interrupt Controller detects a rising edge on the relevant input – and to remain asserted until cleared) or level-sensitive (considered to be asserted only when the relevant input to the Interrupt Controller is HIGH). ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 12-7 Interrupt Handling An interrupt can be in a number of different states: • Inactive – this means that the interrupt is not asserted yet. • Pending – this means that the interrupt source has been asserted, but is waiting to be handled by a core. Pending interrupts are candidates to be forwarded to the CPU interface and then later on to the core. • Active – this describes an interrupt that has been acknowledged by a core and is currently being serviced. • Active and pending – this describes the situation where a core is servicing the interrupt and the GIC also has a pending interrupt from the same source. The priority and list of cores to which an interrupt can be delivered to are all configured in the distributor. An interrupt asserted to the distributor by a peripheral will be marked in Pending state (or Active and Pending if was already Active). The distributor determines the highest priority pending interrupt that can be delivered to a core and forwards that to the CPU interface of the core. At the CPU interface, the interrupt is in turn signalled to the core, at which point the core takes the FIQ or IRQ exception. The core executes the exception handler in response. The handler must query the interrupt ID from a CPU interface register and begin servicing the interrupt source. When finished, the handler must write to a CPU interface register to report the end of processing. Later on the CPU interface is prepared to signal the next interrupt forwarded to it by the distributor. While servicing an interrupt, the distributor cycles through Pending, Active states, ending in Inactive state when it has finished. The state of an interrupt is therefore reflected in the distributor registers. More detailed information on GIC behavior can be found in the TRMs for the individual processor types and in the ARM Generic Interrupt Controller Architecture specification. 12.2.1 Configuration The GIC is accessed as a memory-mapped peripheral. All cores can access the common distributor block, but the CPU interface is banked, that is, each core uses the same address to access its own private CPU interface. It is not possible for a core to access the CPU interface of another core. See Handling interrupts in an SMP system on page 18-14 for more details. The distributor hosts a number of registers that you can use to configure the properties of individual interrupts. These configurable properties are: • An interrupt priority. The distributor uses this to determine which interrupt is next forwarded to the CPU interface. • An interrupt configuration. This determines if an interrupt is level- or edge-sensitive. • An interrupt target. This determines a list of cores to which an interrupt can be forwarded. • Interrupt enable or disable status. Only those interrupts that are enabled in the distributor are eligible to be forwarded when they become pending. • Interrupt security determines whether the interrupt is allocated to Secure or Normal world software. • An Interrupt state. The distributor also provides priority masking by which interrupts below a certain priority are prevented from reaching the core. The distributor uses this when determining whether a pending interrupt can be forwarded to a particular core. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 12-8 Interrupt Handling The CPU interfaces on each core helps with fine-tuning interrupt control and handling on that core: 12.2.2 Initialization Both the distributor and the CPU interfaces are disabled at reset.The GIC must be initialized after reset before it can deliver interrupts to the core. In the distributor, software must configure the priority, target, security and enable individual interrupts. The distributor block must subsequently be enabled through its control register. For each CPU interface, software must program the priority mask and preemption settings. Each CPU interface block itself must be enabled through its control register. This prepares the GIC to deliver interrupts to the core. Before interrupts are expected in the core, software prepares the core to take interrupts by setting a valid interrupt vector in the vector table, and clearing interrupt masks bits in the CPSR. The entire interrupt mechanism in the system can be disabled by disabling the distributor block. Interrupt delivery to an individual core can be disabled by disabling its CPU interface block, or by setting mask bits in CPSR of that core. Individual interrupts can also be disabled (or enabled) in the distributor. For an interrupt to reach the core, the individual interrupt, distributor and CPU interface must all be enabled, and the CPSR interrupt mask bits cleared. 12.2.3 Interrupt handling When the core takes an interrupt, it jumps to the top-level interrupt vector obtained from the vector table and begins execution. The top-level interrupt handler reads the Interrupt Acknowledge Register from the CPU Interface block to obtain the interrupt ID. As well as returning the interrupt ID, the read causes the interrupt to be marked as active in the distributor. Once the interrupt ID is known (identifying the interrupt source), the top-level handler can now dispatch a device-specific handler to service the interrupt. When the device-specific handler finishes execution, the top-level handler writes the same interrupt ID to the End of Interrupt register in the CPU Interface block, indicating the end of interrupt processing. Apart from removing the active status, which will make the final interrupt status either Inactive, or Pending (if the state was Active and Pending), this will enable the CPU Interface to forward more pending interrupts to the core. This concludes the processing of a single interrupt. It is possible for there to be more than one interrupt waiting to be serviced on the same core, but the CPU Interface can signal only one interrupt at a time. The top-level interrupt handler repeats the above sequence until it reads the special interrupt ID value 1023, indicating that there are no more interrupts pending at this core. This special interrupt ID is called the spurious interrupt ID. The spurious interrupt ID is a reserved value, and cannot be assigned to any device in the system. When the top-level handler has read the spurious interrupt ID it can complete its execution, and prepare the core to resume the task it was doing before taking the interrupt. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 12-9 Chapter 13 Boot Code This chapter considers the boot code running in an ARM processor based system, and focuses on two distinct areas: ARM DEN0013D ID012214 • Code to be run immediately after the core comes out of reset, on a so-called bare-metal system, that is, one in which code is run without the use of an operating system. This is a situation that is often encountered when first starting up a chip or system. • How a bootloader loads and runs the Linux kernel. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 13-1 Boot Code 13.1 Booting a bare-metal system When the core has been reset, it will commence execution at the location of the reset vector in the exception vector table (at either address 0x00000000 or 0xFFFF0000, see Table 11-3 on page 11-9). The reset handler code must do some, or all of the following: • In a multi-core system, enable non-primary cores to sleep See Booting SMP systems on page 18-17. • Initialize exception vectors. • Initialize the memory system, including the MMU. • Initialize core mode stacks and registers. • Initialize any critical I/O devices. • Perform any necessary initialization of NEON or VFP. • Enable interrupts. • Change core mode or state. • Handle any set-up required for the Secure world (see Chapter 21). • Call the main() application. The first consideration is placement of the exception vector table. You must make sure that it contains a valid set of instructions that branch to the appropriate handlers. The _start directive in the GNU Assembler tells the linker to locate code at a particular address and can be used to place code in the vector table. The initial vector table will be in non-volatile memory and can contain branch to self instructions (other than the reset vector) as no exceptions are expected at this point. Typically, the reset vector contains a branch to the boot code in ROM. The ROM can be aliased to the address of the exception vector. The ROM then writes to some memory-remap peripheral that maps RAM into address 0 and the real exception vector table is copied into RAM. This means the part of the boot code that handles remapping must be position-independent, as only PC-relative addressing can be used. Example 13-1 shows an example of typical code that can be placed in the exception vector table. Example 13-1 Typical exception vector table code start B B B B B NOP B Reset_Handler Undefined_Handler SWI_Handler Prefetch_Handler Data_Handler @ Reserved vector IRQ_Handler @ FIQ_Handler will follow directly after this table You might then have to initialize stack pointers for the various modes that your application can make use of. Example 13-2 on page 13-3 gives a simple example, showing code that initializes the stack pointers for FIQ and IRQ modes. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 13-2 Boot Code Example 13-2 Code to initialize the stack pointers LDR R0, stack_base @ Enter each mode in turn and set up the stack pointer MSR CPSR_c, #Mode_FIQ:OR:I_Bit:OR:F_Bit ; MOV SP, R0 SUB R0, R0, #FIQ_Stack_Size MSR CPSR_c, #Mode_IRQ:OR:I_Bit:OR:F_Bit ; MOV SP, R0 The next step is to set up the caches, MMU and branch predictors. An example of such code is shown in Example 13-3. We begin by disabling the MMU and caches and invalidating the caches and TLB. This example code is for the Cortex-A9 processor. Some of the Cortex-A processors automatically invalidate the L1 and/or L2 caches at reset, others require manual invalidation. You must check the TRM for a particular core to determine which options have been implemented. The MMU TLBs must be invalidated. The branch target predictor hardware might not have to be explicitly invalidated, but it must be enabled by boot code. Branch prediction can safely be enabled at this point; this will improve performance. Example 13-3 Setting up caches, MMU and branch predictors @ Disable MMU MRC p15, 0, r1, c1, c0, 0 BIC r1, r1, #0x1 MCR p15, 0, r1, c1, c0, 0 @ Disable L1 Caches MRC p15, 0, r1, c1, c0, 0 BIC r1, r1, #(0x1 << 12) BIC r1, r1, #(0x1 << 2) MCR p15, 0, r1, c1, c0, 0 @ Read Control Register configuration data @ Write Control Register configuration data @ @ @ @ Read Control Register configuration data Disable I Cache Disable D Cache Write Control Register configuration data @ Invalidate L1 Caches @ Invalidate Instruction cache MOV r1, #0 MCR p15, 0, r1, c7, c5, 0 @ Invalidate Data cache @ to make the code general purpose, we calculate the @ cache size first and loop through each set + way MRC p15, 1, r0, c0, c0, 0 LDR r3, #0x1ff AND r0, r3, r0, LSR #13 MOV r1, #0 way_loop: MOV r3, #0 set_loop: MOV r2, r1, LSL #30 ORR r2, r3, LSL #5 MCR p15, 0, r2, c7, c6, 2 ADD r3, r3, #1 CMP r0, r3 BGT set_loop ADD r1, r1, #1 CMP r1, #4 ARM DEN0013D ID012214 @ Read Cache Size ID @ r0 = no. of sets - 1 @ r1 = way counter way_loop @ r3 = set counter set_loop @ @ @ @ @ @ @ @ r2 = set/way cache operation format Invalidate line described by r2 Increment set counter Last set reached yet? if not, iterate set_loop else, next Last way reached yet? Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 13-3 Boot Code BNE way_loop @ if not, iterate way_loop @ Invalidate TLB MCR p15, 0, r1, c8, c7, 0 @ Branch Prediction Enable MOV r1, #0 MRC p15, 0, r1, c1, c0, 0 ORR r1, r1, #(0x1 << 11) MCR p15, 0, r1, c1, c0, 0 @ Read Control Register configuration data @ Global BP Enable bit @ Write Control Register configuration data After this, you can create some translation tables, as shown in the example code of Example 13-4. The variable ttb_address is used to denote the address to be used for the initial translation table. This must be a 16KB area of memory (whose start address is aligned to a 16KB boundary), to which an L1 translation table can be written by this code. Example 13-4 Create translation tables @ Enable D-side Prefetch MRC p15, 0, r1, c1, c0, 1 @ Read Auxiliary Control Register ORR r1, r1, #(0x1 <<2) @ Enable D-side prefetch MCR p15, 0, r1, c1, c0, 1 ; @ Write Auxiliary Control Register DSB ISB @ DSB causes completion of all cache maintenance operations appearing in program @ order before the DSB instruction @ An ISB instruction causes the effect of all branch predictor maintenance @ operations before the ISB instruction to be visible to all instructions @ after the ISB instruction. @ Initialize PageTable @ We will create a basic L1 page table in RAM, with 1MB sections containing a flat (VA=PA) mapping, all pages Full Access, Strongly Ordered @ It would be faster to create this in a read-only section in an assembly file LDR r0, =2_00000000000000000000110111100010 @ r0 is the non-address part of descriptor LDR r1, ttb_address LDR r3, = 4095 @ loop counter write_pte ORR r2, r0, r3, LSL #20 @ OR together address & default PTE bits STR r2, [r1, r3, LSL #2] @ write PTE to TTB SUBS r3, r3, #1 @ decrement loop counter BNE write_pte @ for the write-back, BIC r0, ORR r0, BIC r0, ORR r0, ORR r0, STR r0, very first entry in the table, we will make it cacheable, normal, write allocate r0, #2_1100 @ clear CB bits r0, #2_0100 @ inner write-back, write allocate r0, #2_111000000000000 @ clear TEX bits r0, #2_101000000000000 @ set TEX as write-back, write allocate r0, #2_10000000000000000 @ shareable [r1] @ Initialize MMU MOV r1,#0x0 MCR p15, 0, r1, c2, c0, 2 LDR r1, ttb_address ARM DEN0013D ID012214 @ Write Translation Table Base Control Register Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 13-4 Boot Code MCR p15, 0, r1, c2, c0, 0 @ Write Translation Table Base Register 0 @ In this simple example, we don't use TRE or Normal Memory Remap Register. @ Set all Domains to Client LDR r1, =0x55555555 MCR p15, 0, r1, c3, c0, 0 @ Write Domain Access Control Register @ Enable MMU MRC p15, 0, r1, c1, c0, 0 ORR r1, r1, #0x1 MCR p15, 0, r1, c1, c0, 0 @ Read Control Register configuration data @ Bit 0 is the MMU enable @ Write Control Register configuration data The L2 cache, if present, and if running without an operating system, might also require invalidating and enabling at this point. NEON or VFP access must also be enabled. If the system makes use of the TrustZone Security Extensions, it might have to switch to the Normal world when the Secure world is initialized. See Chapter 21 Security for details of this. The next steps will depend on the exact nature of the system. It might be necessary, for example, to zero-initialize memory that will hold uninitialized C variables, copy the initial values of other variables from a ROM image to RAM, and set up application stack and heap spaces. It might also be necessary to initialize C library functions, call top-level constructors (for C++ code) and other standard embedded C initialization. A common approach is to permit a single core within the cluster to perform system initialization, while the same code, if run on a different core, will cause it to sleep, that is, enter WFI state, as described in Chapter 20. The other cores might be woken after core 0 has created a simple set of L1 translation table entries, as these could be used by all cores in the system. Example 13-5 shows example code that determines which core it is running on and either branches to initialization code (if running on core 0), or goes to sleep otherwise. The secondary cores are typically woken up later by an SMP OS. Example 13-5 Determining which core is running @ Only CPU 0 performs initialization. Other CPUs go into WFI @ to do this, first work out which CPU this is @ this code typically is run before any other initialization step MRC AND CMP BEQ p15, 0, r1, c0, c0, 5 r1, r1, #0x3 r1, #0 initialize @ Read Multiprocessor Affinity Register @ Extract CPU ID bits @ if we’re on CPU0 goto the start wait_loop: @ Other CPUs are left powered-down ..... ..... ..... initialize: @ next section of boot code goes here ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 13-5 Boot Code 13.2 Configuration There are a number of control register bits within the core that will typically be set by boot code. In all cases, for best performance, code must run with the MMU, instruction and data caches and branch prediction enabled. Translation table entries for all regions of memory that are not peripheral I/O devices must be marked as L1 Cacheable and (by default) set to read-allocate, write-back cache policy. For multi-core systems, pages must be marked as Shareable and the broadcasting feature for CP15 maintenance operations must be enabled. In addition to the CP15 registers required by the ARM architecture, cores typically have registers that control implementation-specific features. Programmers of boot code should refer to the relevant technical reference manual for the correct usage of these. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 13-6 Boot Code 13.3 Booting Linux It is useful to understand what happens from the core coming out of reset and executing its first instruction at the exception base address 0x00000000 or 0xFFFF0000 if HIVECS (known as high vectors) is selected, until the Linux command prompt appears. (See The Vector table on page 11-7.) When the kernel is present in memory, the sequence on an ARM processor based system is similar to what might happen on a desktop computer. However, the bootloading process can be very different, as ARM processor based phones or more deeply embedded devices can lack a hard drive or PC-like BIOS. Typically, what happens when you power the system on is that hardware specific boot code runs from flash or ROM. This code initializes the system, including any necessary hardware peripheral code and then launches the bootloader (for example U-Boot). This initializes main memory and copies the compressed Linux kernel image into main memory (from a flash device, memory on a board, MMC, host PC or elsewhere). The bootloader passes certain initialization parameters to the kernel. The Linux kernel then decompresses itself and initializes its data structures and running user processes, before starting the command shell environment. Let’s take a more detailed look at each of those processes. 13.3.1 Reset handler There is typically a small amount of system-specific boot monitor code that configures memory controllers and performs other system peripheral initialization. It sets up stacks in memory and typically copies itself from ROM into RAM, before changing the hardware memory mapping so that RAM is mapped to the exception vector address, rather than ROM. In essence this code is independent of that operating system is to be run on the board and performs a function similar to a PC BIOS. When it has completed execution, it will call a Linux bootloader, such as U-Boot. 13.3.2 Bootloader Linux requires a certain amount of code to be run out of reset, to initialize the system. This performs the basic tasks required for the kernel to boot: • Initializing the memory system and peripherals. • Loading the kernel image to an appropriate location in memory (and possibly also an initial RAM disk). • Generate the boot parameters to be passed to the kernel (including machine type). • Set up a console (video or serial) for the kernel. • Enter the kernel. The exact steps taken vary between different bootloaders, so for detailed information, refer to documentation for the one that you want to use. U-Boot is a widely used example, but other bootloader possibilities include Apex, Blob, Bootldr and Redboot. When the bootloader starts, it is typically not present in main memory. It must start by allocating a stack and initializing the core (for example invalidating its caches) and installing itself to main memory. It must also allocate space for global data and for use by malloc() and copy exception vector entries into the appropriate location. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 13-7 Boot Code 13.3.3 Initialize memory system This is very much a board or system specific piece of code. The Linux kernel has no responsibility for the configuration of the RAM in the system. It is presented with the physical memory layout, but has no other knowledge of the memory system. In many systems, the available RAM and its location are fixed and the bootloader task is straightforward. In other systems, code must be written that discovers the amount of RAM available in the system. 13.3.4 Kernel images The kernel image from the build process is typically compressed in zImage format (the conventional name given to the bootable kernel image). Its head code contains a magic number, used to verify the integrity of the decompression, plus start and end addresses. The kernel code is position independent and can be located anywhere in memory. Conventionally, it is placed at a 0x8000 offset from the base of physical RAM. This gives space for the parameter block placed at a 0x100 offset (used for translation tables etc). Many systems require an initial RAM disk (initrd), as this lets you have a root filesystem available without other drivers being setup. The bootloader can place an initial ramdisk image into memory and pass the location of this to the kernel using ATAG_INITRD2 (a tag that describes the physical location of the compressed RAM disk image) and ATAG_RAMDISK. The bootloader will typically setup a serial port in the target, enabling the kernel serial driver to detect the port and use it for a console. In some systems, another output device such as a video driver can be used as a console. The kernel command line parameter console= can be used to pass the information. 13.3.5 Kernel parameters using ATAGs Historically, the parameters passed to the kernel are in the form of a tagged list, placed in physical RAM with register R2 holding the address of the list. Tag headers hold two 32-bit unsigned ints, with the first giving the size of the tag in words and the second providing the tag value (indicating the type of tag). For a full list of parameters that can be passed, consult the appropriate documentation. Examples include ATAG_MEM to describe the physical memory map and ATAG_INITRD2 to describe where the compressed ramdisk image is located. The bootloader must also provide an ARM Linux machine type number (MACH_TYPE). This can be a hard-coded value, or the boot code can inspect the available hardware and assign a value accordingly. There is a more flexible, or generic method for passing this information using Flattened Device Trees (FDTs). 13.3.6 Kernel parameters using Flattened Device Trees The Linux device tree or FDT support was introduced for the PowerPC kernel as a part of the merger of a 32-bit and 64-bit kernel to standardize the firmware interface by using an Open Firmware interface for all PowerPC platforms, servers, desktop and embedded. It has become the configuration methodology used in the Linux kernel for PowerPC, Micro Blaze and SPARC architectures. A device tree is a data structure that describes the hardware configuration. It includes information about processors, memory sizes and banks, interrupt configuration, and peripherals. The data structure is organized as a tree with a single root node named /. With the exception of the root node, each node has a single parent. Each node has a name and can have any number of child nodes. Nodes can also contain named properties values with arbitrary data, and they are expressed in key-value pairs. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 13-8 Boot Code The device tree data format follows the conventions defined in IEEE standard 1275. To simplify system description, a device tree source format (.dts) is used to express device tree data. A device tree node must comply with the following syntax: [label:] node-name[@unit-address] { [properties definitions] [child nodes] } Nodes are defined with a name and a unit-address. Braces mark the beginning and end of the node definition. You can use a Device Tree Compiler (DTC) tool to convert the device tree source file (.dts) to the device tree blob (dtb) format. The dtb, or blob, is known as the Flattened Device Tree and is a firmware independent description of the system, in a compressed format that requires no firmware calls to retrieve its properties. The Linux kernel loads the dtb before it loads the operating system. The chosen node is a placeholder for any environment information that does not belong anywhere else, such as boot arguments for the kernel and default console. Chosen node properties are usually defined by the boot loader, but the dts file can specify a default value. The following code fragment shows a root node description for an ARM Versatile Platform Board. The model and compatible properties are assigned the name of the platform in the form , . This string concatenation is the universal identifier for the machine and must be defined at the top node. / { model = "arm,versatilepb"; compatible = "arm,versatilepb"; #address-cells = <1>; #size-cells = <1>; memory { name = "memory"; device_type = "memory"; reg = <0x0 0x08000000>; }; chosen { bootargs = “console=ttyAMA0 debug”; } }; 13.3.7 Kernel entry Kernel execution must commence with the core in a fixed state. The bootloader calls the kernel image by branching directly to its first instruction, the start label in arch/arm/boot/compressed/head.S. The MMU and data cache must be disabled. The core must be in Supervisor mode, with CPSR I and F Bits set (IRQ and FIQ disabled). R0 must contain 0, R1 the MACH_TYPE value and R2 the address of the tagged list of parameters. The first step in getting the kernel working is to decompress it. This is mostly architecture independent. The parameters passed from the bootloader are saved and the caches and MMU are enabled. Checks are made to see if the decompressed image will overwrite the compressed image, before calling decompress_kernel() in arch/arm/boot/compressed/misc.c, The cache is then cleaned and invalidated before being disabled again. We then branch to the kernel startup entry point in arch/arm/kernel/head.S. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 13-9 Boot Code 13.3.8 Platform-specific actions A number of architecture specific tasks are now undertaken. The first checks core type using __lookup_processor_type() that returns a code specifying which core it is running on. The function __lookup_machine_type() is then used (unsurprisingly) to look up machine type. A basic set of translation tables is then defined which map the kernel code. The cache and MMU are initialized and other control registers set. The data segment is copied to RAM and start_kernel() is called. 13.3.9 Kernel start-up code In principle, the rest of the startup sequence is the same on any architecture, but in fact some functions are still hardware dependent. ARM DEN0013D ID012214 1. IRQ interrupts are disabled with local_irq_disable(), while lock_kernel() is used to stop FIQ interrupts from interrupting the kernel. It initializes the tick control, memory system and architecture-specific subsystems and deals with the command line options passed by the bootloader. 2. Stacks are set up and the Linux scheduler is initialized. 3. The various memory areas are set-up and pages allocated. 4. The interrupt and exception table and handlers are setup, along with the GIC. 5. The system timer is setup and at this point IRQs are enabled. Additional memory system initialization occurs and then a value called BogoMips is used to calibrate the core clock speed. 6. Internal components of the kernel are set up, including the filesystem and the initialization process, followed by the thread daemon that creates kernel threads. 7. The kernel is unlocked (FIQ enabled) and the scheduler started. 8. The function do_basic_setup() is called to initialize drivers, sysctl, work queues and network sockets. At this point, the switch to User mode is performed. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 13-10 Boot Code 0xFFFF.0000 Vectors page Dynamic mappings Kernel linear mapping 0xC000.0000 0xBF00.0000 Kernel memory Static mappings Modules User stacks Dynamic libraries 0x4000.0000 User memory Heap Data ZI data Code 0x0000.0000 Figure 13-1 Linux virtual memory view The memory map used by Linux is shown in Figure 13-1. ZI refers to zero-initialized data. There is a broad split between kernel memory, above address 0xBF000000 and user memory, below that address. Kernel memory uses global mappings, while user memory uses non-global mappings, although both code and data can be shared between processes. As already mentioned, application code starts at 0x1000, leaving the first 4KB page unused, to enable trapping of NULL pointer references. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 13-11 Chapter 14 Porting New projects will normally use an existing operating system and re-use code from existing applications. New code might be targeted at the ARMv7-A architecture, but might eventually require porting to a different board. There are a number of issues associated with porting code from a different architecture to run on an ARM processor, or from older versions of the ARM architecture to ARMv7-A. For many applications, particularly those coded with portability issues in mind, this will mean recompiling the source code. For example, a large amount of Linux application software is designed to run in many different environments and tends to make fewer assumptions about the underlying hardware. However, there are a number of areas where C code is not fully portable. In particular, low level, hardware-specific code such as device drivers might require more effort than porting applications. There is an additional consideration when porting code between processors: that of efficiency. It might be the case that optimizations applied to code running on another processor, or to older versions of the ARM architecture, do not apply to that code when running on ARMv7-A. Equally, there might be scope for making code smaller or faster on ARMv7-A class processors. Optimizations might differ between processors or systems. Code that is optimal for one processor might not be optimal for others. ARM-specific optimization is covered in Chapter 17. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-1 Porting 14.1 Endianness The use of the terms little-endian and big-endian was introduced by Danny Cohen in his 1980 paper “On Holy Wars and a Plea for Peace”. Cohen has also been responsible for many advances in the fields of networks and computer graphics. It is a reference to Gulliver’s Travels, a famous satire from the early 18th century, by Irish writer Jonathan Swift, in which a war is fought between the fictional countries of Lilliput and Blefuscu over the correct end to open a boiled egg. There are two basic ways of viewing bytes in memory – little-endian and big-endian. On big-endian machines, the most significant byte of an object in memory is stored at the least significant (closest to zero) address. On little-endian machines, the most significant byte is stored at the highest address. The term byte-ordering can also be used rather than endian. Other kinds of endianness do exist, notably middle-endian and bit-endian, but we will not discuss these. Consider the following simple piece of code (Example 14-1): Example 14-1 Endian access int i = 0x44332211; unsigned char c = *(unsigned char *)&i; On a 32-bit big-endian machine, c is given the value of the most significant byte of i: 0x44. On little-endian machines, c is the least significant byte of i: 0x11. Figure 14-1 on page 14-3 illustrates the two differing views of memory. It should be stated at this point that many people find endianness confusing and that even the act of drawing a diagram to illustrate it can reveal a personal bias. The diagram shows a 32-bit value in a register being written to address 0x1000, using a STR instruction. The core then performs a read of a byte, using a LDRB instruction. A different value will be returned by this instruction sequence depending on whether you have a little- or big-endian memory system. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-2 Porting Storing (writing) Bit index 31 0x44 0x33 Little-endian 0 0x22 0x11 r0=0x44332211 Big-endian STR r0, [r1] Address offset (bytes) 3 2 1 0 0x44 0x33 0x22 0x11 Address offset (bytes) 3 2 1 0 External device or memory 0x11 0x22 0x33 0x44 LDRB r2, [r1] 31 0x00 0x00 External device or memory LDRB r2, [r1] 0 r2=0x11 0x00 0x11 31 0x00 0x00 0 r2=0x44 0x00 0x44 Loading (reading) 3 Address offset 2 1 0 0x44 0x33 0x22 0x11 Little-endian 31 Bit index 0x44 0x33 External device or memory Big-endian LDR r0, [r1] 0 0x22 0x11 r0=0x44332211 Bit index 31 0x11 0x22 0 0x33 0x44 r0=0x11223344 Figure 14-1 Different endian behaviors ARM cores support both modes, but are most commonly used in, and typically default to little-endian mode. Most Linux distributions for ARM tend to be little-endian only. The x86 architecture is little-endian. The PowerPC or the venerable 68K, on the other hand, are generally big-endian, although the Power architecture can also handle little-endian. Several common file formats and networking protocols specify different endianness. For example, .BMP and .GIF files are little-endian, while .JPG is big-endian, and TCP/IP is big-endian, but USB and PCI are little-endian. So, there are two issues to consider – code portability and data sharing. Systems are built from multiple blocks and can include one or more cores, DSPs, peripherals, memory, and network connections. Whenever data is shared between these elements, there is a potential endianness conflict. If code is being ported from a system with one endianness to a system with different endianness, it might be necessary to modify that code, either to make it endian-neutral or to work with the opposite byte-ordering. Cortex-A series processors provide support for systems of either endian configuration, controlled by the CPSR E bit that enables software to switch dynamically between viewing data as little or big-endian. Instructions in memory are always treated as little-endian. The REV instruction (see Byte reversal on page 5-18) can be used to reverse bytes within an ARM register, providing simple conversion between big and little-endian formats. In principle, it is straightforward to support mixed endian systems. Typically this means the system is natively of one endian configuration, but there are peripherals which are of the opposite endianness. The CPSR E bit can be modified dynamically by software, and there is a SETEND assembly instruction provided to do this. The CP15:SCTLR (System Control Register, c1), contains the EE bit (see Coprocessor 15 on page 3-9) that defines the endian mode to switch to on an exception and also the endianness of translation table lookups. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-3 Porting It would be difficult if exception code had to worry about which endian state the core was in on arrival at the handler. In practice, however, it can be difficult to tell the compiler that part of the system is of a different endian configuration to the rest of memory. Modern ARM processors support a big-endian format known architecturally as BE8 that is only applied to the data memory system. Older ARM processors used a different format known as BE-32 that applied to both instructions and data. BE8 corresponds to what most other computer architectures call big-endian. Example 14-2 provides a simple piece of code that behaves differently when run on architectures with different endianness. Example 14-2 Non-portable code int i- 0x12345678; char *buf = (char*)&i; char i0, i1, i2, i3; i0 i1 i2 i3 = = = = buf[0]; buf[1]; buf[2]; buf[3]; The values of i0…i3 are not guaranteed to be the same if the system endianness changes. This kind of code is therefore inherently non-portable. When inspecting code in which you suspect endianness problems, you must look for the following potential causes of problems: Unions A union can hold objects of different types and sizes. You must keep track of what the data member represents at any particular time. Code that uses unions must be carefully checked. If the union is used to access the same data, but with different data types, there exists a possible endianness, alignment, and packing problem. Any time that halfword, word (or longer) data types are combined or viewed as an array of bytes is a potential issue. Casting of data types Anywhere that data is accessed in a way outside of its native data type is a potential problem. Similarly, if there are arrays of bytes, they must not be accessed other than as a byte data type. Casting of pointers changes how data is addressed and can be endian sensitive. Bitfields To avoid endianness problems code that defines bitfields or performs bit operations must not be used in code that is intended to be portable. Data sharing Any code that reads shared data from another block, or exports data to another block, must be checked to see whether the two blocks agree endian definitions. If the two are different, it might be necessary to implement byte swapping at one location. Network code Code that accesses networking or other I/O devices must be reviewed to see if there is any endian dependency. Again, it might be necessary to re-write code for greater efficiency, or swap bytes at the interface. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-4 Porting 14.2 Alignment The alignment of accesses is significant on ARM cores. On older ARM processors, accesses to addresses that are not aligned are possible, but with a different behavior to those using the ARMv7 architecture. On ARM7 and ARM9 processors, an unaligned LDR is performed in the memory system in the same way as an aligned access, but with the data returned being rotated so that the data at the requested address is placed in the least significant byte of the loaded register. Some older compilers and operating systems were able to use this behavior for clever optimizations. This can represent a portability problem when moving code from an ARMv4 or ARMv5 to ARMv7 architecture. ARM MMUs can be configured to automatically detect such unaligned accesses and abort them (using the CP15:SCTL A bit), see System control register (SCTLR) on page 3-12 and Coprocessor 15 on page 3-9. For the Cortex-A series of processors, unaligned accesses are supported, although you must enable this by setting the U bit in the CP15:SCTL register, indicating that unaligned accesses are permitted. This means that instructions to read or write words or halfwords can access addresses that are not aligned to word or halfword boundaries. However, load and store multiple instructions (LDM and STM) and load and store double-word (LDRD or STRD) must be aligned to at least a word boundary. Loads and stores of floating-point values must always be aligned. Additional alignment constraints might be imposed by the ABI that are stronger than those imposed by the ARM architecture. These unaligned accesses can take additional cycles in comparison with aligned accesses and therefore alignment is also a performance issue. In addition, such accesses are not guaranteed to be atomic. This means that a external agent (another core in the system) might perform a memory access that appears to occur part way through the unaligned access. For example, it might read the accessed location and see the new value of some bytes and the old value of others. The word at address 0 3 2 1 0 0x100 7 6 5 4 0x104 B A 9 8 0x108 The word at address 8 Figure 14-2 Aligned words at address 0 or 8 A word aligned address is one that is a multiple of four, for example 0x100, 0x104, 0x108, 0x10C, 0x110. Figure 14-2 shows examples of aligned accesses. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-5 Porting 3 2 1 0 7 6 5 4 B A 9 8 The word at address 1 The word at address 8 Figure 14-3 An unaligned word An unaligned word at address 1 is shown in Figure 14-3. It takes three bytes from the word at 0 and one byte from the word at 4. A simple example where alignment effects can have significant performance effects is the use of memcpy(). Copying small numbers of bytes between word aligned addresses will be compiled into LDM or STM instructions. Copying larger blocks of memory aligned to word boundaries will typically be done with an optimized library function that will also use LDM or STM. Copying blocks of memory whose start or end points do not fall on a word boundary can result in a call to a generic memcpy() function that can be significantly slower. Although, if the source and destination are similarly unaligned then only the start and end fragments are non-optimal. Whenever explicit typecasting is performed, that cast always carries alignment implications. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-6 Porting 14.3 Miscellaneous C porting issues In this section we consider some other possible causes of problems when porting C code. 14.3.1 unsigned char and signed char The piece of code in Example 14-3 illustrates a very simple example of a possible issue when porting code to ARM. Example 14-3 Use of unsigned char char c = -1; if (c > 0) printf("c is positive \n"); else printf("c is negative \n”); When this is compiled for some architectures (for example, for x86) the result is the one you might intuitively expect, which is that it reports the variable c as negative, but compiling the code on an ARM Compiler will produce code that reports c as positive, and typically a warning will be emitted by the compiler too. The ANSI C standard specifies a range for both signed (at least -127 to +127) and unsigned (at least 0 to 255) chars. Simple chars are not specifically defined and it is compiler dependent whether they are signed or unsigned. Although the ARM architecture has the LDRSB instruction, that loads a signed byte into a 32-bit register with sign extension, the earliest versions of the architecture did not. It made sense at the time for the compiler to treat simple chars as unsigned, whereas on the x86 simple chars are, by default, treated as signed. One workaround for users of GCC is to use the -fsigned-char command line switch or --signed-chars for RVCT, that forces all chars to become signed, but a better practice is to write portable code by declaring char variables appropriately. Unsigned char must be used for accessing memory as a block of bytes or for small unsigned integers. Signed char must be used for small signed integers and simple char must be used only for ASCII characters and strings. In fact, on an ARM core, it is usually better to use ints rather than chars, even for small values, for performance reasons. You can read more on this in Chapter 17 Optimizing Code to Run on ARM Processors. A second piece of code, in Example 14-4, illustrates another possible problem with chars. Here EOF is compared with an unsigned char. On an ARM core, the while loop will never complete. The value of EOF is defined as -1 and when it is converted to be compared with a char (which is unsigned and therefore in the range 0 to 255), it can never be equal and so the loop does not terminate. Example 14-4 Use of EOF char c; while ((c = getchar()) != EOF) putchar(c); You must declare the variable as int instead of char to avoid the problem, in fact, this is how the functions in stdio.h are defined. Similar cases to look out for include the use of getopt() and getc() – both are defined as returning an int. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-7 Porting 14.3.2 Compiler packing of structures Compilers are not permitted to re-order members of a structure and have to follow the alignment restrictions of the core architecture. This means that compilers might have to add unused bytes into user defined structures, for best performance and code size. Such padding is architecture specific and can therefore lead to portability problems if assumptions have been made about the location and size of this padding. Marking a structure as __packed in the ARM Compiler or using the attribute __packed__ in GCC, will remove any padding. This reduces the size of the structure and can be useful when porting code or for structures being passed from external hardware, but can reduce performance and increase code size, although generally it will be relatively efficient on Cortex-A series processors. If you have some simple struct code, as shown in Example 14-5: Example 14-5 A typical C struct struct test { unsigned char c; unsigned int i ; unsigned short s; } Then the arrangement of data within the struct will be as in Figure 14-4. 0 1 3 C 4 7 I 8 9 S. Unpacked structure layout Figure 14-4 Unpacked structure If you now mark the structure as packed, as in Example 14-6: Example 14-6 Packed structure struct test { unsigned char c; unsigned int i ; unsigned short s; } __attribute__((__packed__)); ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-8 Porting 0 1 3 C 4 I I 5 6 S. Packed structure layout Figure 14-5 Packed structure In Figure 14-5 byte 0 now holds C, bytes 1-4 hold I and bytes 5-6 will hold S. To access S or I will require the core to perform an unaligned access. 14.3.3 Use of the stack Code that makes assumptions about the stack structure can require porting effort. For example, functions with a variable number and type of arguments can receive their variables through the stack. The macros dealing with accessing these arguments will walk through the stack frame and provide compatibility between systems, but code that does not use standard libraries or macros can have a problem. 14.3.4 Other issues A function prototype is a declaration that omits the function body but gives the function name, argument and return types. It effectively gives a way to specify the interface of a function separately from its definition. Incorrectly prototyped functions can behave differently between different compilers. Compilers can allocate different numbers of bytes to enum. Care is therefore required when enumerations are used; cross-linking of code and libraries between different compilers might not be possible if enums are used. Code written for 8-bit or 16-bit cores might assume that integer variables are 16-bit. On ARMv7-A processors, they will always be 32 bits. The program might rely on 16-bit behaviors. In general, this is easily fixed by the use of the C short type. Use of short ints can be less efficient, as we shall see in the chapter on optimization and in cases where the code does not rely on 16-bit behavior, it is usually better to promote these variables to a 32-bit int. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-9 Porting 14.4 Porting ARM assembly code to ARMv7-A So far in this chapter, we have looked at porting C code from other architectures to ARM. It is sometimes necessary to port assembly code from older ARM processors to the Cortex-A series. Sometimes, it can be difficult to determine which ARM architecture variant your code was originally targeting. GCC has a series of macros, with names like __ARM_ARCH_6__ which are mutually exclusive. The ARM Compiler has a similar set of macros such as __TARGET_ARCH_7_A. In general, ARM assembler code is backward compatible and will work unmodified. There are a few special cases to look for: CP15 Operations The architecture specifies a consistent mapping of CP15 to designed system control operations. In general, you should attempt to understand the purpose of code that performs CP15 instructions and ensure that this code is appropriate for the ARMv7-A Architecture. In addition, a number of CP15 registers (for example CP15:ACTL, the Auxiliary Control Register) are implementation specific. Code that references such registers will always require attention when being ported. SWP 14.4.1 The SWP (or SWPB) instruction was used to implement atomic operations in older versions of the ARM architecture, but is deprecated and its use strongly discouraged in the ARMv7 Architecture. There is no encoding for SWP in Thumb at all, so SWP is not permitted when building for Thumb. In the ARMv7 Architecture, SWP is disabled by default, but can be re-enabled by setting CP15:SYSCTL bit [10]. See System control register (SCTLR) on page 3-12. Code that uses SWP must be rewritten to make use of LDREX or STREX (and possibly also barrier instructions – see Chapter 10). Alternatively, the GCC __sync_... intrinsics could be used. It is usually preferable to use library functions for such things as spinlocks, semaphores, and mutexes, rather than writing such primitives yourself. The mechanisms used are different from those used by SWP, so it is necessary to port all code accessing an atomic object, not just some of it. Usually an atomic object will be managed by a piece of library code shared between the threads that access it, so this is not typically a problem. Cortex-A series cores treat SWP as Undefined out of reset. Memory access ordering and memory barriers The ARMv7 architecture has a weakly-ordered memory model. Code for older processors that makes implicit assumptions about ordering of memory accesses might not operate correctly on ARMv7 devices. This might be particularly true for code that interacts with other devices, such as a DMA controller or other bus master. Such code must be inspected and modified, possibly by the insertion of appropriate barrier instructions or by the use of suitable atomic primitives. See Chapter 10 for a more detailed description of memory ordering and memory barriers. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-10 Porting 14.5 Porting ARM code to Thumb We also consider problems associated with porting code written for the ARM instruction set to the Thumb instruction set. As we have seen, use of Thumb is often preferred because of its combination of small code size with higher performance relative to the older 16-bit only Thumb instruction set. 14.5.1 Use of PC as an operand Instructions that use PC (R15) as an explicit operand can cause problems when assembling in Thumb. As we have seen, it is not possible to encode any arbitrary 32-bit address into any instruction as an operand. For this reason, it is common to address data stored in a literal pool using offsets relative to the current instruction location. In ARM code, the PC value can be used (with some adjustment) to determine the address of the currently executing instruction, for this purpose and this enables position-independent coding. However, the PC value obtained in this way can show some inconsistencies between ARM and Thumb states and can also depend on the type of instruction executed. For this reason, ARM assembly code that directly references the PC register might require modifying to work correctly in Thumb. It is better to avoid explicit PC arithmetic and instead to use an instruction such as: LDR r0, = This automatically puts somewhere in the text section and assembles an appropriate PC-relative LDR instruction. It will, however, only do this if cannot be encoded in a MOV instruction.You can still do a load from a local text section label that you declare explicitly, as shown below, but again, the assembler or linker can be permitted to perform the PC offset calculation: LDR ... data: r0, data .long Sometimes the required PC-relative address offset is too large to encode in a single LDR instruction, causing the assembler to complain that a literal pool is out of range. This can be resolved by explicitly placing a literal pool, using the .ltorg directive that tells the assembler where to insert literal data. You must ensure that the literal data is not located where it might be executed as code. This typically means the .ltorg directive is placed after an unconditional branch or function return instruction. 14.5.2 Branches and interworking When using Thumb, the system will typically have both ARM and Thumb functions. Even if you compile your application for Thumb, you might still have to think about such things as libraries and prebuilt binaries. The core must know which instruction set is to be used for the code being executed after a branch, procedure call or return. This interworking between instruction sets is described in Interworking on page 4-11. When writing C code, the linker takes care of this for us, but a little more care is required when porting assembly code. The target instruction set state is determined in different ways depending on the type of branch. We can consider a number of different instructions: Function return Legacy code might use the MOV PC, LR instruction. This is unsafe for systems that contain a mix of ARM and Thumb code and must be replaced by BX LR for code running on all later architectures. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-11 Porting Function return from the stack This is done using the LDMFD SP!, {registers, pc} instruction that will continue to work correctly in the ARMv7-A architecture, although a newer, equivalent form, POP { , pc} is also available. This is used when registers that must be preserved by the function are PUSHed at the start of the function. Branch A simple B instruction will work in the same fashion on all ARM architectures. If ARM and Thumb instructions are mixed in a single source file (this is unusual), there is no automatic instruction set switch for local symbols. The assembler might introduce a veneer depending on whether it knows that the destination is in a different instruction set and is definitely a code symbol (such as a .type , %function or .thumb_func). Because a symbol appears in a code section it is not assumed to be a code symbol unless specifically tagged in this way. If the label is in a different file, the linker will take care of any necessary instruction set change. Similar considerations apply for a function call (BL). Note Veneers are small pieces of code that are automatically inserted by the linker when it detects that a branch target is out of range or is a conditional branch to code in the other state, for example, from Thumb to ARM or ARM to Thumb. The veneer becomes an intermediate target of the original branch with the veneer itself then being a branch to the target address. Often these veneers can be inlined. The linker can reuse a veneer generated for a previous call, for other calls to the same function if it is in range from both calls. Occasionally, such veneers can be a performance factor. If you have a loop that calls multiple functions through veneers, you will get many pipeline flushes and therefore sub-optimal performance. Placing related code together in memory can avoid this. The ARM linker can be made to export information on this by specifying the –-info veneers option. PC modification Care might be required with other instructions that modify the PC and produce a branch effect. For example, MOV PC, register must be replaced with BX register in systems that contain both ARM and Thumb code. Function call to register address If code contains a sequence like MOV LR, PC followed by MOV PC, register, this will not work in a system that has both ARM and Thumb code. You must replace it with the single instruction BLX . When a destination or return address is variable or calculated at run-time, take care to appropriately set the Thumb bit (bit [0]) in the address correctly and to use the correct type of branch, to make sure that the call (and return, if applicable) switches instruction set appropriately. If an external label or function defined in another object is referenced, the linker will produce an address with the Thumb bit (bit [0]) set appropriately. However, if you reference a symbol internal to the object, things are more complicated. For C functions, or code tagged as Thumb, bit [0] will be set appropriately, but it will not be set appropriately for other symbols. In particular, GNU Assembler local labels will not have the Thumb bit set appropriately, nor will the GNU current assembly location symbol “.”. Therefore, when coding in assembler, if an address will be passed to any other function or object, for example, as a return address, method address or callback, you must handle the Thumb bit setting yourself, setting bit [0] of the address where required. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-12 Porting 14.5.3 Operand combinations Thumb and ARM Assembly code have different restrictions on instruction operands. You might find that existing ARM code can produce assembler errors when targeted for Thumb. Branch out of range errors occur when the distance between the current instruction and the branch target is too large to be encoded in a Thumb instruction. To resolve this, it might be necessary to use a different type of branch, move code sections or to use two separate branches, a so-called trampoline. Similarly, index out of range errors might be produced on load and store operations, and to resolve these it might be necessary to manually add part (or all) of the required index offset to the base register in a separate explicit instruction. Generally, use of SP is limited to stack operations, other forms of use might not be permitted in Thumb code. This means that PUSH, POP, LDMFD SP!, STMFD SP!, ADD, SUB or MOV instructions that use the SP are permitted, but other operations should be treated as possible problems. Similarly, operations that directly operate on the PC must be checked (other than the usual function or exception return operations, or literal pool loads). 14.5.4 Other ARM/Thumb differences There are a number of other differences that can require attention by the assembly language programmer. ARM DEN0013D ID012214 • The RSC instruction is not available in Thumb. Therefore, code that uses RSC must be re-coded using RSB or SBC, or be built in ARM state. • Most ARM instructions can optionally be made conditional. This is not the case in Thumb, other than for branches. Instead, small instruction sequences can be executed conditionally by preceding them with the IT instruction. For compatibility with both ARM and Thumb, the IT block construct is always understood when using unified assembler syntax. Manually modifying code to use IT instructions can be tedious. Fortunately, the assembler command-line option -mimplicit-it= , where can be one of never, arm, thumb or always. In this case you don’t have to add IT instructions, the assembler will work out the right thing to do for the given target. When assembling for Thumb, however, it is sensible to use -mimplicit-it=thumb. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 14-13 Chapter 15 Application Binary Interfaces The C compiler is able to generate code from many separately compiled modules. For programs to execute successfully these modules must be able to interoperate, with the operating system code and any code that is written in assembler or any other compiled language. For that reason, we must define a set of conventions to govern inter-operability between separate pieces of code. The Application Binary Interface (ABI) for the ARM architecture specification describes a set of rules that an ARM executable must adhere to in order to execute in a specific environment. It specifies conventions for executables, including file formats and ensures that objects from different compilers or assemblers can be linked together successfully. There are variants of the ABI for specific purposes, for example, the Linux ABI for the ARM architecture or the Embedded ABI (EABI). The ARM Architecture Procedure Call Standard (AAPCS) is part of the ABI (although the ABI actually calls it the Procedure Call Standard for the ARM Architecture). It specifies conventions for register and stack usage by the compiler and during subroutine calls. Knowledge of this is vital for inter-working C and assembly code and can be useful for writing optimal code. The AAPCS supersedes the previous ARM-Thumb Procedure Call Standard (ATPCS). The AAPCS specifies rules that must be adhered to by callers to enable a callee function to run and what callee routines must do in order to ensure that callers can continue function correctly when the callee returns. It describes the way that data is laid out in memory and how the stack is laid out, plus permitted variations for processor extensions. It defines how code that has been separately compiled or assembled works together. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 15-1 Application Binary Interfaces 15.1 Procedure Call Standard As we have seen, there are sixteen 32-bit integer registers available in the core. These are labeled R0 - R15. Table 15-1 shows the role assigned to registers within the procedure call standard. Table 15-1 APCS registers Register PCS name PCS role R0 a1 argument 1/scratch register/result R1 a2 argument 2/scratch register/result R2 a3 argument 3/scratch register/result R3 a4 argument 4/scratch register/result R4 v1 register variable R5 v2 register variable R6 v3 register variable R7 v4 register variable R8 v5 register variable R9 tr/sb/v6 static base/ register variable R10 v7 register variable R11 v8 register variable R12 IP scratch register/new -sb in inter-link-unit calls R13 SP SP always points at the top of the stacka R14 LR link register/scratch register R15 PC program counter a. The ARM instruction set provides instructions which can facilitate the implementation of different types of stacks but the ABI only uses full descending stacks. For the purposes of function calls, the registers are divided into three groups: • Argument registers R0-R3 (a1-a4). These can be used as scratch registers or as caller-saved register variables that can hold intermediate values within a routine, between calls to other functions. • Callee-saved registers, normally used as register variables. Typically, the registers R4-R8, R10 and R11 (v1-v5, v7 and v8) are used for this purpose. • Registers that have a dedicated role. The function of the program counter, link register and stack pointer ought to be clear. If not, read Registers on page 3-6. The IP (R12) register can be used by the linker, as a scratch register between a routine and any subroutine it calls, or as an additional local variable within a function. Because the BL instructions cannot address the full 32-bit address space, the linker might have to insert a ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 15-2 Application Binary Interfaces veneer between the caller and callee. Veneers can also be used for ARM-Thumb inter-working or dynamic linking. Veneers are permitted to modify the contents of IP (R12). Register R9 has a role that is specific to a particular environment. It can be used as the static base register (SB) to point to position-independent data, or as the thread register (TR) where thread-local storage is used. In code that has no requirement for such a special register, it can be used as an extra callee-saved variable register, v6. The first four word-sized parameters passed to a function will be transferred in registers R0-R3. Sub-word sized arguments, for example, char, will still use a whole register. Arguments larger than a word will be passed in multiple registers. If more arguments are passed, the fifth and subsequent words will be passed on the stack. Passing arguments on the stack always requires additional instructions and memory accesses and therefore reduces performance. For optimal code, you must always try to limit arguments to four words or fewer. If this is not possible, the most commonly used parameters must be defined in the first four positions of the function definition. If the arguments are part of a structure then it might be more efficient to pass a pointer to the structure instead. C++ uses the first argument to pass the this pointer to member functions, so only three arguments can be passed in registers. There are additional rules about 64-bit types. 64-bit types must always be 8-byte aligned in memory. Recall that in Alignment on page 14-5 we described how there are limitations on use of LDRD and STRD double-word instructions to unaligned addresses. In addition, 64-bit arguments to functions must be passed in an even + consecutive odd register pair (for example, R0 + R1 or R2 + R3). If 64-bit arguments are passed on the stack, they must be at an 8-byte aligned location. Again, this is because of restrictions on LDRD and STRD instructions. If such 64-bit arguments are listed in a sub-optimal fashion, there can be wasted space in registers or on the stack. When considering such issues, it is important to take into account the this pointer in R0 present in all non-static C++ member functions. Stack R0 R1 R2 R3 Stack Foo1 (int i0, double d, int i1) i0 Unused d d i1 Stack Foo2 (int i0, int i1, double d) i0 i1 d Unused d Figure 15-1 Efficient parameter passing Figure 15-1 shows some examples of how sub-optimal listing of arguments can cause unnecessary spilling of variables to the stack. The figure shows how two different function calls, that pass identical parameters, make use of the registers and stack. The first function passes an int, a double and an additional int. The first parameter is passed in R0. The second argument is 64-bits and must be passed in an even and consecutive odd register (or in an 8-byte aligned location on the stack). It is therefore passed in R2 and R3. This means that the final argument is passed on the stack. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 15-3 Application Binary Interfaces As the stack pointer must be 8-byte aligned (a constraint imposed by the ABI to simplify the interface at function boundaries), there will be an additional unused word pushed and popped. In the second function call, you can pass the two int values in registers R0 and R1 and the double value in R2 and R3. This means that no values are spilled to the stack, which gives more efficient code, requiring both fewer instructions and fewer memory accesses. The registers R4-R11 (v1-v8) are used to hold the values of the local variables of a subroutine. A subroutine is required to preserve (on the stack), the contents of the registers R4-R8, R10, R11 and SP (and R9 in PCS variants that designate R9 as v6), if they are used by that subroutine. A caller function will have code like this: @ may need to preserve r0-r3 @ does not need to preserve r4-r11 BL Func while the callee function will have code like this: Func: @ Must preserve r4-r11, lr (if used) @ May corrupt r0-r3, r12 PUSH {r4-r11, lr} ... ... ... POP {r4-r11, pc} @ Returns value in r0 - char, short or int @ Returns value in r0 & r1 - double The PUSH/POP instructions must maintain 8-byte stack alignment, and so use an even number of registers. Leaf functions do not have to do so. The example callee code as shown pushes or pops R4-R11 and LR and PC, which would not preserve an 8-byte aligned stack. It is shown like this to indicate which registers must be saved. In practice, the compiler will normally push an extra register, depending on whether the function is leaf and which registers are modified by the function. Actual instructions will usually be one of PUSH/POP {r4, lr/pc}, PUSH/POP {r4-r8, lr/pc}, PUSH/POP {r4-r10, lr/pc} or PUSH/POP {r4-r12, lr/pc}. In each case, you would PUSH lr and POP pc. 15.1.1 VFP and NEON register use Readers unfamiliar with ARM floating-point might want to refer to Chapter 6 Floating-Point before reading this section. VFPv3 has 32 single-precision registers s0-s31 that can also be accessed in pairs as double-precision registers d0-d15. There are an additional 16 double-precision registers, d16-d31. NEON can also view these as quadword registers q0-q15. Registers s16-s31 (d8-d15, q4-q7) must be preserved across subroutine calls; registers s0-s15 (d0-d7, q0-q3) do not have to be preserved (and can be used for passing arguments or returning results in standard procedure-call variants). Registers d16-d31 (q8-q15), do not have to be preserved. The Procedure Call Standard specifies two ways in which floating-point parameters can be passed. For software floating-point, they will be passed using ARM registers R0-R3 and on the stack, if required. An alternative, where floating-point hardware exists in the core, is to pass parameters in the VFP or NEON registers. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 15-4 Application Binary Interfaces This hardware floating-point variant behaves in the following way: • Integer arguments are treated in exactly the same way as in softfp. So, if you consider the function f below, you will see that the 32-bit value a will be passed to the function in R0, and because the value b must be passed in an even or odd register pair, it will go into R2 or R3, leaving R1 unused. void f(uint32_t a, uint64_t b) r0: a r1: unused r2: b[31:0] r3: b[63:32] • FP arguments fill d0-d7 (or s0-s15), independently of any integer arguments. This means that integer arguments can flow onto the stack and FP arguments will still be slotted into FP registers (if there are enough available). • FP arguments are able to back-fill, so it's less common to get the unused slots that we see in integer arguments. Consider the following examples: void f(float a, double b) d0: s0: a s1: unused d1: b Here, b is aligned automatically by being assigned to d1 (that occupies the same physical registers as VFP s2 or VFP s3). void f( float a, double b, float c) d0: s0: a s1: c d1: b In this example, the compiler is able to place c into s1; it does not have to be placed into s4. In practice, this is implemented (and described) by using separate counters for s, d and q arguments, and the counters always point at the next available slot for that size. In the second FP example, a is allocated first because it is first in the list, and it goes into first available s register, which is s0. Next, b is allocated into the first available d register, which is d1 because a is using part of d0. When c is allocated, the first available s register is s1. A subsequent double or single argument would go in d2 or s4, respectively. • There is an additional case when filling FP registers for arguments. When an argument must be spilled to the stack, no back-filling can occur, and stack slots are allocated in exactly the same way for additional parameters as they are for integer arguments. void f( double a, double b, double c, double d, double e, double f, float g, double h, double i, float j) d0: a d1: b d2: c d3: d d4: e d5: f d6: s12: g s13: unused d7: h *sp: i *sp+8: j *sp+12: unused (4 bytes of padding for 8-byte sp alignment) Arguments a-f are allocated to d0-d5 as expected. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 15-5 Application Binary Interfaces The single-precision g is allocated to s12, and h goes to d7. The next argument, i, cannot fit in registers, so it is stored on the stack. It would be interleaved with stacked integer arguments if there were any. However, while s13 is still unused, j must go on the stack because we cannot back-fill to registers when FP arguments have hit the stack. 15.1.2 • Double-precision and quad-precision registers can also be used to hold vector data. This would not occur in typical C code. • No VFP registers are used for variadic procedures, that is, a procedure that does not have a fixed number of arguments. They are instead treated as in softfp, in that they are passed in integer registers (or the stack). Single-precision variadic arguments are converted to doubles, as in softfp. Linkage If the platform has hardware support for NEON technology or a Floating Point Unit (FPU), the highest performance is achieved by passing NEON and FPU parameters and return values in NEON and FPU registers. This is called hardware floating-point linkage. In some situations, using the general-purpose registers for parameter passing might be preferred, to simplify software compatibility between platforms with and without hardware floating-point support. This is called software floating-point linkage. Floating-point arithmetic might use either hardware coprocessor instructions, or library functions. Floating-point linkage, however, is concerned with passing arguments between functions that use floating-point variables. Software floating-point linkage means that the parameters and return value for a function are passed using the ARM integer registers R0 to R3 and the stack. Hardware floating-point linkage uses the VFP coprocessor registers to pass the arguments and return value. The benefit of using software floating-point linkage is that the code can be run on a core with or without a VFP coprocessor and is therefore more portable. The benefit of using hardware floating-point linkage is that it is more efficient than software floating-point linkage, but only runs on systems that have a VFP coprocessor. You cannot mix objects with different floating-point linkage in a single image. Any dynamic libraries loaded while the application is running must also use the same linkage. Any system that supports both NEON and VFP instructions uses a common register bank for these instructions, therefore configuration options that affect the floating-point calling convention also affect how NEON parameters are passed and returned. 15.1.3 Stack and heap The stack implementation is full-descending, with the current top of the stack pointed to by R13 (SP). The stack must be contiguous in the virtual memory space. Detection of a stack overflow is usually handled with memory management. The stack must always be aligned to a word boundary, except at an external interface, when it must be double-word aligned. The heap is an area (or areas) of memory that are managed by the process itself, for example, with the C malloc() function. It is typically used for the creation of dynamic data objects. 15.1.4 Returning results A function that returns a char, short, int or single-precision float value will do so using R0. A function that returns a 64-bit value (a double-precision float or long long) does so in R0 and R1. As mentioned in VFP and NEON register use on page 15-4, floating-point and NEON return ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 15-6 Application Binary Interfaces values will be in s0, d0, or q0 when using hardware linkage. If the software uses hardware linkage it will return floating-point values in s0, d0, or q0. If the software uses software linkage, it will return single-precision float in r0. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 15-7 Application Binary Interfaces 15.2 Mixing C and assembly code One example of why it can be useful to understand the AAPCS is to write assembly code that is compatible with C code. One way to do this is write separate modules and assemble them with GNU Assembler. They can be defined as extern functions in C and called. Provided the AAPCS rules are followed, there should be no problem. We can also insert assembly code into our C code, through the GCC asm statement. This is very simple to use. For example, we can implement a NOP as shown in Example 15-1. Example 15-1 NOP asm("nop"); The time taken to carry out a NOP is undefined. In fact, it is likely that this NOP will have no effect, because the C compiler might optimize it away, or the core might discard the instruction. When you include assembly language code with inline assembler, the resulting code is still subject to optimization by the C compiler. This is a very important point that must be taken into account whenever inline assembly is used. Even if the compiler does not optimize-away the NOP instruction, the core itself can filter out the NOP from the instruction stream so that it never reaches the execute stage. Inline assembly code has a different syntax to regular assembly code. Registers and constants have to be specified differently, if they refer to C expressions. A slightly more complicated example takes the value of an int and uses the USAD8 assembly instruction to calculate the sum of difference of bytes and then store the result in a different int. Example 15-2 shows the relevant code. Example 15-2 Using the USAD8 instruction asm volatile (“usad8 %0, %1, %2” : “=r” (result): “r”(value1), “r”(value2)); The general format of such inline assembly code is: asm volatile (assembler instructions : output operands (optional) : input operands (optional) : clobbered registers (optional) ); The colons divide the statement up into parts. The first part “usad8 %0, %1, %2” is the actual assembly instruction. The second part is an (optional) list of output values from the sequence. If more than one output is required, commas are used to separate the entries. You might then optionally have a list of input values for the sequence, with the same format as the output. If you don’t specify an output operand for an assembly sequence, it is quite likely that the C compiler optimizer will decide that it is not serving any useful purpose and optimize it away! A way to avoid this is to use the volatile attribute that tells GCC not to optimize the sequence. In the actual assembly language statement, operands are referenced by a percent sign followed by the symbolic name in square brackets. The symbolic name references the item with the same name in either the input or output operand list. This name is completely distinct from any other symbol within your C code (although clearly it is less confusing to use symbols that do have a meaning within your code). Alternatively, the name can be omitted and the operand can be specified using a percent sign followed by a digit indicating the position of the operand in the list (that is, %0, %1 … %9), as shown in the example. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 15-8 Application Binary Interfaces There is an optional fourth part to an asm statement, called the clobber list. This enables you to specify to the compiler what will be changed by the assembly code. We can specify registers (for example, R0), the condition code flags (cc) or memory. This makes the compiler store affected values before and reload them after executing the instruction. The constraints mentioned when we talked about input and output operand lists relate to the fact that assembly language instructions have specific operand type requirements. When passing parameters to inline assembly statements, the compiler must know how they are represented. For example, the constraint “r“ specifies one of the registers R0-R15 in ARM state, while “m” is a memory address and “w” is a single precision floating-point register. These characters have an = placed before them to indicate a write-only output operand, a “+” for a read/write output operand (that is, one that is both input and output to the instruction). The “&” modifier instructs the compiler not to select any register for the output value that is used for any of the input operands. You can force the inline assembler to use a particular register to hold a local variable by using something like the code shown in Example 15-3. Example 15-3 Inline assembler local variable usage void func (void) { register unsigned int regzero asm(“r0”); and later asm volatile("rev r0, r0"); This usage can interfere with the compiler optimization and does not guarantee that the register will not be re-used, for example when the local variable is no longer referenced. Hard coding register usage is always bad practice. It is almost always better to use local variables instead. Example 15-4 gives a longer example of inline assembler, taken from the Linux kernel. It shows how a series of inline assembly language instructions can be used. The code manipulates the CPSR, to change modes. This would not be possible using C code. Example 15-4 Inline assembler void __naked get_fiq_regs(struct pt_regs *regs) { register unsigned long tmp; asm volatile ( "mov ip, sp\n\ stmfd sp!, {fp, ip, lr, pc}\n\ sub fp, ip, #4\n\ mrs %0, cpsr\n\ msr cpsr_c, %2 @ select FIQ mode\n\ mov r0, r0\n\ stmia %1, {r8 - r14}\n\ msr cpsr_c, %0 @ return to SVC mode\n\ mov r0, r0\n\ ldmfd sp, {fp, sp, pc}" : "=&r" (tmp) : "r" (®s->ARM_r8), "I" (PSR_I_BIT | PSR_F_BIT | FIQ_MODE)); } ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 15-9 Application Binary Interfaces The ARM compiler tools have a similar concept, albeit with different syntax. In addition to inline assembly, they also support embedded assembly that is assembled separately from the C code and produces a compiled object that is combined with the object from the C compilation. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 15-10 Chapter 16 Profiling Donald Knuth once famously observed that “Premature optimization is the root of all evil”. However, Knuth's comment does not argue against appropriate optimization. Code optimization is an important part of the work of software engineers– modifying software so that it runs more quickly, uses less power and makes less use of memory or other resources. To do this, you must first identify which part or parts of the code should be optimized. It is worth noting, however, that a programmer’s best guesses about where the time is spent can be notoriously inaccurate, hence the importance of profiling. Profiling is a technique that lets you identify sections of code that consume large proportions of the total execution time. It is usually more productive to focus optimization efforts on code segments that are executed very frequently, or that take a significant proportion of total execution time than to optimize rarely used functions or code that takes only a small proportion of total execution time. A profiler will tell you which parts of the code are frequently executed and which occupy the most core cycles. A profiler can help you identify bottlenecks, situations where the performance of the system is constrained by a small number of functions. This data is collected using instrumentation, an execution trace or sampling. When you have identified some slow part of your code it is important to consider whether you can change the algorithm, before attempting to improve the existing code. For example, if the time is being spent searching a linked list, it is probably much more beneficial to change to using a tree or hash table instead of spending effort to speed up the linked list search. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 16-1 Profiling Profiling can be considered as a form of dynamic code analysis. Profiling tools can gather information in a number of different ways. We can distinguish two basic approaches to gathering information. Time based sampling Here, the state of the system is sampled at a periodic, time-based interval. The size of this interval can affect the results, a smaller sampling interval can increase execution time but produce more detailed data. Event based sampling Here, sampling is driven by occurrences of an event, which means that the time between sampling intervals is usually variable. Events can often be hardware related, for example, cache misses. It is also important to understand that profilers typically operate on a statistical basis, they might not necessarily produce absolute counts of events. In complex systems, it might be necessary to control profiling information by use of annotation options to specify which events are to be recorded, which events shown, or thresholds to avoid displaying large numbers of functions with low count numbers. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 16-2 Profiling 16.1 Profiler output Profiler tools normally provide two kinds of information: Call graph The call graph tells you the number of times each function was called. This can help point out which function calls can be eliminated or replaced and shows inter-relations between different functions. Viewing a call graph can suggest code to optimize and reveal hidden bugs, for example, if code is unexpectedly calling an error function many times. Collecting call graph information can require building the code with special options. Flat profile A flat profile, as in Example 16-1 shows how much core time each function uses and the number of times it was called. This enables a simple identification of which functions consume large fractions of run-time and should therefore be considered first for possible optimizations. Example 16-1 Example flat profile Flat profile: Each sample counts as 0.01 seconds. % cumulative self self time seconds seconds calls ms/call 33.34 0.02 0.02 6275 0.00 16.67 0.03 0.01 192 0.07 16.67 0.04 0.01 15 1.20 16.67 0.05 0.01 7 1.41 total ms/call 0.00 0.21 1.20 1.41 name start func1 memcpy write It might be useful at this point to consider some example profiling tools which can be used in Cortex-A series processors. 16.1.1 Gprof GProf is a GNU tool that provides an easy way to profile your C/C++ application and find the locations that require work. Using GCC, you can generate profile information by compiling with a special flags. The source code has to be compiled with the -pg option. For line-by-line profiling, the –g option would also be required. You then execute the compiled program and the profiling data is collected. You then run gprof on the resulting statistics file to view the data, in a variety of convenient formats. When you compile with -pg the compiler adds profiling instrumentation that collects data at function entry and exits at runtime. It therefore profiles only the user application code and not anything that happens in code that has not been built for profiling (for example, libc) or the kernel. Gprof can give misleading results if the performance limitations of the code come from kernel or I/O issues (memory fragmentation or file accesses for example). It might be necessary to remove GCC optimization flags, as some compiler optimizations can cause problems while profiling. It is also the case that the use of the profiling flags will actually slow the program down. This can be an important consideration in some types of real-time system, where it might be that the interaction of real-time events has a significant effect on the performance of the profiled code. A binary file called gmon.out containing profiling information is generated. This file can then be operated on by the gprof tool. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 16-3 Profiling 16.1.2 OProfile OProfile is a whole system profiling tool that runs on Linux and includes the kernel in its metrics. Unlike gprof, it works using a statistical sampling method. OProfile can examine the system at regular intervals, determine what code is running, and update appropriate counters. If a long enough profile is taken, with a sufficient sample rate, an accurate picture of the execution is obtained. Like other profilers that make use of interrupts, code that disables interrupts can cause inaccuracies. For this reason, the Linux function spinlock_irq_restore() that re-enables interrupts after a spinlock has been relinquished can erroneously appear to be a major system bottleneck, as the time for which interrupts were disabled can be counted against it. OProfile can also be made to trigger on hardware events and will record all system activity including kernel and library code execution. OProfile does not require code to be recompiled with any special flags (provided symbol information is available). It provides useful hardware information about such things as clock cycles and cache misses. Call graphs are statistically generated, so might not be completely accurate. 16.1.3 DS-5 Streamline DS-5 Streamline is a graphical performance analysis tool that can be used to analyze the performance of an Linux or Android system. It is a component of ARM DS-5 and combines a kernel driver, target daemon and Eclipse-based user interface. (For more information about DS-5, see ARM DS-5 on page B-10.) DS-5 Streamline takes sampling data and system trace information and produces reports that present the data visually and in a statistical form. It uses hardware performance counters and kernel metrics to provide an accurate representation of system resources. DS-5 Streamline enables the application source code to be annotated to augment its graphical display with additional textual or visual information, such as labelled timing markers or screen shots from the target. For example, DS-5 Streamline has a display mode called Core map. In this mode the process trace view changes from an intensity map of time to a view that highlights core affinity. Figure 16-1 on page 16-5 shows the DS-5 Streamline view of a multi-threaded Linux application running on a multi-core cluster. Threads are being assigned to the cores by the Linux kernel: ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 16-4 Profiling Figure 16-1 Core map in ARM DS-5 Streamline 16.1.4 ARM performance monitor The performance monitor hardware is able to count several events, using multiple counters. Normally, you combine together multiple values to generate useful parameters to optimize. For example, you can choose to count the total number of clock cycles and the number of instructions executed and use this to derive a cycles per instruction figure that is a useful proxy for the efficiency with which the core is operating. We can generate information about cache hit or miss rates (separately for both L1 data and instruction caches) and examine how code changes can affect these. Cortex-A series processors contain event counting hardware that can be used to profile and benchmark code, including generation of cycle and instruction count figures and to derive figures for cache misses and so forth. The performance counter block contains a cycle counter that can count core cycles, or be configured to count every 64 cycles. There are also a number of configurable 32-bit wide event counters that can be set to count instances of events from a wide-ranging list (for example, instructions executed, or MMU TLB misses). These counters can be accessed through debug tools, or by software running on the core, through the CP15 Performance Monitoring Unit (PMU) registers. They provide a non-invasive debug feature and do not change the behavior of the core. CP15 also provides a number of controls for enabling and resetting the counters and to indicate overflows (there is an option to generate an interrupt on a counter overflow). The cycle counter can be enabled independently of the event counters. It is important to understand that information generated by such counters might not be exact. In a superscalar, out-of-order core, for example, it can be difficult to guarantee that the number of instructions executed is precise at the time any other counter is updated. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 16-5 Profiling The standard countable events, common to all ARMv7-A processors are listed in Table 16-1. The TRM for the specific processor being used provides more information on the lists of events that can be monitored, that can include a large number of additional possibilities in addition to those listed here. Table 16-1 Performance monitor events ARM DEN0013D ID012214 Number Event counted 0x00 Software increment of the Software Increment Register 0x01 Instruction fetch that causes a Level 1 instruction cache refill 0x02 Instruction fetch that causes a Level 1 instruction TLB refill 0x03 Data Read or Write operation that causes a Level 1 instruction TLB refill 0x04 Data Read or Write operation that causes a Level 1 data cache access 0x05 Data Read or Write operation that causes a Level 1 data TLB refill 0x06 Memory-reading instruction executed 0x07 Memory-writing instruction executed 0x09 Exception taken 0x0A Exception return executed 0x0B Instruction that writes to the Context ID register 0x0C Software change of program counter 0x0D Immediate branch instruction executed 0x0F Unaligned load or store 0x10 Branch mispredicted or not predicted 0x11 Cycle count; the register is incremented on every cycle 0x12 Predictable branch speculatively executed 0x13 Data memory access 0x14 Level 1 instruction cache access 0x15 Level 1 data cache write-back 0x16 Level 1 data cache write-back 0x17 Level 2 data cache refill 0x18 Level 2 data cache write-back 0x19 Bus access 0x1A Local memory error 0x1B Instruction speculatively executed 0x1C Instruction write to TTBR 0x1D Bus cycle 0x1E-0x3F Reserved Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 16-6 Profiling Multi-core clusters include a significant number of additional events relating to SMP capabilities, described in Chapter 18 Multi-core processors. These include numbers of barrier instructions executed, measures of coherent cache activity and counts of exclusive access failures. In Linux, these counters are normally accessed through the kernel using the Linux OProfile tool or the Linux perf events framework. However, User mode access to these counters can be enabled for direct access if required. 16.1.5 Linux perf events Linux contains patches (as of Linux-2.6.34) to support the Ingo Molnar Perf Events framework. It handles multiple events, including core cycles and cache-misses but also measures kernel events like context switches. It provides simple, platform independent access to the ARM performance counter registers and also to software events. Through this framework, tools can produce graphs and statistics about such things as function calls. You can also record execution traces, profile on a per-core, per-application and per-thread basis and generate summaries of events. 16.1.6 Ftrace Ftrace is an increasingly widely used trace tool on Linux. It provides visualization of the flow within the kernel by tracing each function call. It enables interactions between parts of the kernel to be viewed and enables to developers to investigate potential problem paths where interrupts or pre-emption are disabled for long periods. 16.1.7 Valgrind and Cachegrind Valgrind is a widely used tool, commonly used for detection of memory leaks and other memory handling problems; however, it can also be used for profiling memory usage and is potentially able to give much more detailed results than OProfile. Valgrind translates a program into a core-neutral intermediate representation. Other tools associated with Valgrind are able to operate on this and then Valgrind translates the code back into native machine code. This process makes the code run an order of magnitude slower, but enables checks for accesses to undefined memory, off-by-one errors and so forth to be performed by tools like Memcheck. For memory access optimization, the Cachegrind tool can be used. Because Valgrind simulates the execution of the program, you can use Cachegrind to record all uses of program memory. By simulating the operation of the core caches, you can generate statistics about cache usage. Some care is required – this is a simulation of the cache and it is possible that it does not represent the real cache hardware completely accurately. Nevertheless, it can be a very useful tool to examine cache and memory usage by an application. When writing high level applications, it can be difficult to have much appreciation for (or control over) addresses used by a program. The linker will typically generate many of the virtual addresses used within an image, while the runtime loader will control positioning of libraries and so forth. Finally, the kernel is responsible for placement of code and data in physical memory. Memory profiling tools can therefore be a useful aid. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 16-7 Chapter 17 Optimizing Code to Run on ARM Processors Optimization does not necessarily mean optimizing to make programs faster. In embedded systems, you might prefer to optimize for battery life, code density or memory footprint. Writing code that is more efficient delivers not only higher levels of performance, but can also be crucial in conserving battery life. If you can get a job done faster, in fewer cycles, you can turn off the power for longer periods. Many compilers provide options to help with this. For example, both the ARM Compiler and the GNU GCC compiler have -O flags to specify the level of optimization. Although cycle timing information can be found in the Technical Reference Manual (TRM) for the processor that you are using, it is very difficult to work out how many cycles even a trivial piece of code will take to execute. The movement of instructions through the pipeline is dependent on the progress of the surrounding instructions and can be significantly affected by memory system activity. Pending loads or instruction fetches that miss in the cache can stall code for tens of cycles. Standard data processing instructions (logical and arithmetic) will take only one or two cycles to execute, but this does not give the full picture. Instead, you must use profiling tools, or the system performance monitor built-in to the core, to extract useful information about performance. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-1 Optimizing Code to Run on ARM Processors 17.1 Compiler optimizations The ARM Compiler and GNU GCC give you a wide range of options that aim to increase the speed, or reduce the size, of the executable files they generate. For each line in the source code there are generally many possible choices of assembly instructions that could be used. The compiler must trade-off a number of resources, such as registers, stack and heap space, code size (number of instructions), compilation time, ease of debug, and number of cycles per instruction in order to produce the best image file. 17.1.1 Function inlining When a function is called, there is a certain overhead. A called function must store its own return address on the stack if it has to reuse R14. Instructions might also be required to place arguments in the appropriate registers and push registers on the stack, in accordance with the Procedure Call Standard. There is a possible overhead when returning to the original point of execution when the function ends, again requiring a branch (and corresponding instruction pipeline flush) and possibly popping registers from the stack. However, the pipeline will not be flushed if the return is correctly predicted using the return stack. This function-call overhead can become significant when there are functions that contain only a few instructions, and where these functions represent a significant amount of the total run-time. Also, executing branches uses branch predictor resources, that can affect overall program performance. Function inlining eliminates this overhead by replacing calls to a function by a copy of the actual code of the function itself (known as placing the code inline). Inlining for critical code paths is always a worthwhile optimization if there is only one place where the function is called. It is always worthwhile if calling the function requires more instructions (memory) than inlining the function body. An additional consideration is that inlining can help permit other optimizations. Clearly, increasing the number of times that a function is called will increase the number of inlined copies of the function that are made and this will increase the cost in code size. GCC performs inlining only within each compilation unit. The inline keyword can be used to request that a specific function must be inlined wherever possible, even in other files. The GCC documentation gives more details of this and how its use can be combined with static and extern. We will look at inlining in a little more detail when we consider cache optimizations. 17.1.2 Eliminating common sub-expressions Another simple source-level optimization is re-using already computed results in a later expression. This common sub-expression elimination is performed automatically when optimization command line switches are used and can make code both smaller and faster. However, the compiler might not necessarily catch all cases, and it can sometimes be more useful to do this by hand. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-2 Optimizing Code to Run on ARM Processors Example 17-1 illustrates how this works: Example 17-1 Common sub-expression i = a * b + c; j = a * b * d; The compiler can treat this code as if it had been written as in Example 17-2. It must be noted though, that it can only do this if neither a nor b is volatile. Example 17-2 Common sub-expression elimination tmp = a * b; i = tmp + c; j = tmp * d; This reduces both the instruction count and cycle count. 17.1.3 Loop unrolling Every iteration of a loop has a certain penalty associated with it. Every conditional loop must include a test for the end of loop on each iteration. Furthermore, there is a branch instruction to iterate over the loop, that can take a number of cycles to execute. We can avoid this penalty by unrolling loops, partially or fully. Consider the simple code shown in Example 17-4, to initialize an array. Example 17-3 Initializing an array for (i = 0; i < 10; i++) { x[i] = i; } Each iteration of the loop contains an assembler sequence of the form in Example 17-4. Example 17-4 Loop termination assembly code CMP i,#10 BLT for_loop A large proportion of the total run time will have been spent checking if the loop has terminated and in executing a branch to re-execute the loop. The same code can be written by unrolling the loop, as shown in Example 17-5. Example 17-5 Unrolled loop x[0] = 0; x[1] = 1; ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-3 Optimizing Code to Run on ARM Processors x[2] x[3] x[4] x[5] x[6] x[7] x[8] x[9] = = = = = = = = 2; 3; 4; 5; 6; 7; 8; 9; When the code is written in this way, we remove the compare and branch instruction and have a sequence of stores and adds. This is clearly larger than the original code but can execute considerably faster. Conventionally, loop unrolling is often considered to increase the speed of the program but at the expense of an increase in code size (except for very short loops). However, in practice this might not always be the case on many hardware platforms. In many systems, an access to external memory takes significant numbers of cycles and an instruction cache is provided. Code that loops will often fit into the cache very well. The code is fetched into the cache during the first loop iteration and is executed directly from cache after that. Unrolling the loop can mean that the code is executed only once and, because it is larger, does not cache so well. This is more likely to be the case for functions that are executed only once. Loops that are executed frequently might be cached whether they are unrolled or not. An additional consideration is that modern ARM processors typically include branch prediction logic that can hide the effect of pipeline flushes from you by speculatively predicting whether a branch will or will not be taken ahead of the actual evaluation of a condition. In some cases, the branch instruction can be folded, so that it does not require an actual processor cycle to execute. Cortex-A series processors can have long, complex instruction pipelines, with interdependencies between instructions, particularly loads and instructions that set condition code flags. The compiler understands the rules associated with a particular processor and can often re-arrange instructions so that pipeline interlocks are avoided. This is called scheduling and typically involves re-arranging the order of instructions in ways that do not alter the logical correctness of the program or its size, but that reduce its execution time. This can significantly increase the compiler effort, increasing both the time and memory required for the compilation. It can also restrict the ability to perform source level debug. There might no longer be a strict one-to-one link between a line of C source and a sequence of assembly instructions. We can instead have a couple of instructions from a C statement followed by instructions for the next statement and then some more instructions for the first statement. 17.1.4 GCC optimization options GCC has a range of optimization levels, plus individual options to enable or disable particular optimizations. The overall compiler optimization level is controlled by the command line option -On, where n is the required optimization level, as follows: • -O0. (default). No optimization is performed. Each source code command relates directly to the corresponding instructions in the executable file. This gives the clearest view for source level debugging. • -O1. This enables most common forms of optimization that requires no size versus speed decisions, including function inlining. It can often actually produce a faster compile than –O0, because the resulting files are smaller. • -O2.This enables additional optimizations, such as instruction scheduling. Again, optimizations that can have speed versus size implications will not be used. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-4 Optimizing Code to Run on ARM Processors • -O3. This enables additional optimizations, such as aggressive function inlining and can therefore increase the speed at the expense of image size. Furthermore, this option enables -ftree-vectorize - causing the compiler to attempt to automatically generate NEON code from standard C or C++. See Chapter 20 Writing NEON Code. • -funroll-loops. This option is independent of the -On option, and enables loop unrolling. Loop unrolling can increase code size and might not have a beneficial effect in all cases. • -Os. This selects optimizations that attempt to minimize the size of the image, even at the expense of speed. Higher levels of optimization can restrict debug visibility and increase compile times. It is usual to use -O0 for debugging, and -O2 for finished code. When using the above optimization options with the –g (debug) switch, it can be difficult to see what is happening. The optimizations can change the order of statements or remove (or add) temporary variables among other things. But an understanding of the kinds of things the compiler will do means that satisfactory debug is normally still possible with –O2 -g. For optimal code, it is important to specify to the compiler as much detailed information about the target platform as practically possible. Many useful options are documented on http://gcc.gnu.org/onlinedocs/gcc/ARM-Options.html. The main platform-specifying parameters are: -march= where is the architecture version to compile for. This defines the instruction set supported, It can make a significant difference to performance to specify –march=armv7-a if this is supported by your platform but is not used by default by your compiler. -mcpu= More specific than –march, -mcpu specifies which processor to optimize for, including scheduling instructions in the way most efficient for that processor’s pipeline. -mtune= This option provides processor specific tuning options for code, even when only an architecture version is specified on the command line. For instance, the command line might contain -march=armv5te -mtune=cortex-a8. This selects instructions for the architecture ARMv5TE but tunes the selected instructions for execution on a Cortex-A8 processor. -mfpu= If your target platform supports hardware floating-point or NEON, specify this to ensure that the compiler can make use of these instructions. For a Cortex-A5 target, you would specify –mfpu=neon-vfpv4. -mfloat-abi= This option specifies the floating-point ABI to use. Values for are: ARM DEN0013D ID012214 soft causes GCC to generate code containing calls to the software floating-point library for floating-point operations. softfp enables GCC to generate code containing hardware floating-point instructions, but still uses the software floating-point linkage. hard enables GCC to generate code containing hardware floating-point instructions and uses FPU-specific hardware floating-point linkage. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-5 Optimizing Code to Run on ARM Processors The default depends on the target configuration. You must compile your entire program with the same ABI, and link with a compatible set of libraries. Table 17-1 shows a few examples of code generation for floating-point operations. Table 17-1 Floating-point code generation 17.1.5 -mfpu -mfloat-abi Resultant code Any value soft Floating-point emulation using software floating-point library vfpv3 softfp VFPv3 floating-point code vfpv3-d16 softfp VFPv3 floating-point code neon hard VFPv3 and Advanced SIMD code, where the floating-point and SIMD types use the hardware FP registers armcc optimization options The armcc compiler enables you to compile your C and C++ code. It is an optimizing compiler with a range of command-line options to enable you to control the level of optimization. The command line option gives a choice of optimization levels, as follows: • -Ospace. This option instructs the compiler to perform optimizations to reduce image size at the expense of a possible increase in execution time. • -Otime. This option instructs the compiler to perform optimizations to reduce execution time at the expense of a possible increase in image size. • -O0. Turns off most optimizations. It gives the best possible debug view and the lowest level of optimization. • -O1. Removes unused inline functions and unused static functions. Turns off optimizations that seriously degrade the debug view. If used with --debug, this option gives a satisfactory debug view with good code density. • -O2 (default). High optimization. If used with --debug, the debug view might be less satisfactory because the mapping of object code to source code is not always clear. • -O3. performs the same optimizations as -O2 however the balance between space and time optimizations in the generated code is more heavily weighted towards space or time compared with -O2. That is: — -O3 -Otime aims to produce faster code than -O2 -Otime, at the risk of increasing your image size — -O3 -Ospace aims to produce smaller code than -O2 -Ospace, but performance might be degraded. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-6 Optimizing Code to Run on ARM Processors 17.2 ARM memory system optimization Writing code that is optimal for the system it will run on is a key part of the art of programming. It requires you to understand how the compiler and underlying hardware will carry out the tasks described in the lines of code. If you can do the job with less access to external memory, you can save power by keeping everything on-chip. Furthermore, by accessing the external memory less frequently, you improve the performance of the system, enabling software to run faster, or the processor to be clocked more slowly or for shorter periods, to save power. 17.2.1 Data cache optimization In most Cortex-A series processors, there is a significant gap in performance between memory accesses that hit in the cache and those that do not. Cache misses can take tens of cycles to resolve. Cache hits return data within a few cycles and the compiler can often schedule instructions in a way that hides latency. For most algorithms, therefore, ensuring that cache misses are minimized is the most important possible optimization. The most important improvements are those that affect the level 1 cache. Consider the problem of data cache misses. Optimization is particularly significant for pieces of code that use a dataset larger than the available cache size. It is important for you to understand the arrangement of data in memory and how that corresponds to data cache accesses. Code must be structured in a way that ensures maximum re-use of data already loaded into the cache. It is this principle of data locality, the degree to which accesses to the same cache line are concentrated during program execution, in both space and time, which gives best performance. Several techniques to improve this locality can be considered. 17.2.2 Loop tiling Loop tiling divides loop iterations into smaller pieces, in a way which promotes data cache re-use. Large arrays are divided into smaller blocks (tiles) that match the accessed array elements to the cache size. The classic example to illustrate this approach is a large matrix vector product. Consider two square matrices a and b, each of size 1024 × 1024. Example 17-6 shows code to compute a matrix vector product. This requires you to multiply each element in each array with each element in the other array. Example 17-6 Matrix vector product code for (i = 0; i < 1024; i++) for (j = 0; j < 1024; j++) for (k = 0; k < 1024; k++) result[i][j] = result[i][j] + a[i][k] * b[k][j]; In this case, the contents of matrix a are accessed sequentially, but matrix b advances in the inner loop, by row. It is therefore, highly probable that you will encounter a cache miss for each multiply operation. It is obvious that the order in which the additions for each element of the result matrix are calculated does not change the result, ignoring the effect of such things as overflows. Code can be rewritten in a way that improves the cache hit rate. In the example, the elements of matrix b are accessed in the following way (0,0), (1,0), (2,0)… (1023, 0), (0,1), (1,1)… (1023,1). The elements are stored in memory in the order (0,0), (0,1) etc. For word sized elements, it means that the elements (0,0), (0,1)…(0,7) will be stored in the same cache line. For simplicity, we will assume that the start address of the matrix is aligned to a cache line. Alignment will be ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-7 Optimizing Code to Run on ARM Processors mentioned again in Structure alignment on page 17-9. Therefore, elements (0,0), (0,1), (0,2) etc. will be in the same cache line; when you load (0,0) into the cache, you get (0,1...7) too. By the time the inner loop completes, it is likely that this cache line will be evicted. If you modify the code so that two (or indeed four, or eight) iterations of the middle loop are performed immediately while executing the inner loop, as in Example 17-7 you can make a big improvement. Similarly, you can unroll the outer loop two (or four, or eight) times as well. Example 17-7 Code using tiles for (io = 0; io < 1024; io += 8) for (jo = 0; jo < 1024; jo += 8) for (ko = 0; ko < 1024; ko += 8) for (ii = 0, rresult = &result[io][jo], ra = &a[io][ko]; ii < 8; ii++, rresult += 1024, ra += 1024) for (ki = 0, rb = &b[ko][jo]; ki < 8; ki++, rb += 1024) for (ji = 0; ji < 8; ji++) rresult[ji] += ra[ki] * rb[ji]; There are now six nested loops. The outer loops iterate with steps of 8, representing the fact that eight int sized elements are stored in each line of the level 1 cache. Some additional optimizations have also been introduced. The order of ji and ki has been reversed as only one expression uses ki, but two use ji. In addition, you can optimize by removing common expressions from the inner loops. All pointer accesses are potential sources of aliasing in C, so by using result, ra and rb to access array elements, the array indexing is speeded up. This is covered in more detail in Source code modifications on page 17-12. Figure 17-1 illustrates the changing cache access pattern that results from changes to the C code. Figure 17-1 Effect of tiling on cache usage 17.2.3 Loop interchange In many programs, there will be nested loops, a very simple example would be code that stepped through the items in a 2-dimensional array. For reasonably complex code, you can sometimes get better performance by re-arrangement of the loops. It is better to have the loop with the smaller number of iterations as the outer loop and the one with the highest iteration count as the innermost loop. This gives two potential advantages. One is that the compiler can potentially unroll the inner loop. Perhaps more importantly for complex loops where the size of the nested loop is sufficiently large that it might not all be held in the level 1 cache at the same time, the overall ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-8 Optimizing Code to Run on ARM Processors cache hit rate will be improved by this change. Some compilers can make this change automatically at higher levels of optimization. For example, GCC 4.4 adds the switch -floop-interchange to do this. 17.2.4 Structure alignment Efficient placement of structure elements and alignment are not the only aspects of data structures that influence cache efficiency. Where code has a large working set, it is important to make efficient use of the available cache space. To achieve this, it might be necessary to rearrange data structures. It is common to have data structures that span multiple cache lines, but where the program uses only a few parts of the structure at any particular time. If there are many objects of this type, it can make sense to try to split the structure so that it fits within a cache line. For example, you can split an array of structures into two or more arrays of smaller structures. This only makes sense if the object itself is aligned to a cache boundary. For example, consider the case where you have a very large array of instances of a 64-byte structure (much larger than the cache size). Within that structure, you have a byte-sized quantity and you have a commonly used function that iterates through the array looking only at that byte-sized quantity. This function would make inefficient use of the cache, as you would have to load an entire cache line to read the 8-bit value. If instead those 8-bit values were stored in their own array (rather than as part of a larger structure), you would get 32 or 64 values per cache linefill. As we saw in Chapter 14 Porting, unaligned accesses are supported, but can take extra cycles in comparison to aligned accesses. For performance reasons, therefore, it can be sensible to remove or reduce unaligned accesses. 17.2.5 Associativity effects As we have seen, ARM L1 caches are normally 4-way set-associative, but L2 caches typically have 8- or 16-way associativity. There can be performance problems if more than four of the locations in the data fall into the same cache set, as there can be repeated cache misses, even though other parts of the cache can be unused. The ARM L1 Cache uses physical rather than virtual addresses, so it can be difficult for programmers operating in User mode to take care of this. A particularly common cause of this problem is arranging data so that it is on boundaries of powers of two. If the cache size is 16KB, each way is 4KB in size. If you have multiple blocks of data arranged on boundaries that are multiples of 4KB, the first access to each block will go into line 0 of a way. If code accesses the first line in several such blocks then you can get cache misses even if only five cache lines in total are being used. Unaligned accesses can increase the likelihood of this, as each access might require two cache lines rather than one. 17.2.6 Optimizing instruction cache usage The C programmer does not directly have control over how the instruction cache is used by code. Code is linear between branch instructions and this pattern of sequential accesses uses the cache efficiently. The branch prediction logic of the core will try to minimize the stalls because of branches, so there is little you can do to assist. The main goal for you is to reduce the code footprint. Many of the compiler optimizations enabled at -O2 and -O3 for the ARM Compiler and GCC deal with loop optimizations and function inlining. These optimizations will improve performance if the code accounts for a significant part of the total program execution. In particular, function inlining has multiple potential benefits. Obviously, it can reduce branch penalties by removing branches on both function call and exit, and potentially also stack usage. Equally importantly, it enables the compiler to optimize over a larger block of code that can lead to better optimizations for value range propagation and elimination of unused code. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-9 Optimizing Code to Run on ARM Processors However, modifications intended for speed optimizations that increase code size can actually reduce performance because of cache issues. Larger code is less likely to fit in the L1 cache (or indeed the L2 cache) and the performance lost by the additional cache linefills can well outweigh any benefits of the optimization. It is often better to use the armcc -Ospace or gcc –Os option to optimize for code density rather than speed. Clearly, using Thumb code will also improve code density and cache efficiency. There are some interesting decisions to be made around function inlining and in some cases human judgment can improve on that of the compiler. A function that is only ever called from one place will always give a benefit if inlined. One might think that inlining very small functions always gives a benefit, but this is not the case. An instance of a tiny function that is called from many places is likely to be re-used many times within the instruction cache. If the same function is repeatedly inlined, it is much more likely that it will cause a cache miss and also evict other potentially useful code from the cache. The branch prediction logic within Cortex-A series processors is efficient and an unconditional function call and return consumes few cycles, much less than would be used for a cache linefill. You might want to use the GCC function attributes noinline or always_inline to control such cases. This is a general problem and not specific to inlining functions. Whenever conditional execution is used and it is lopsided, that is, the expression far more often leads to one result than the other, there is the potential for false static branch prediction and bubbles (a delay in execution of an instruction) in the pipeline. It is usually better to order conditional blocks so that the often-executed code is linear, while the less commonly executed code has to be branched to and does not get pre-fetched unless it is actually used. The GCC attribute __builtin_expect used with the –freorder-blocks optimization option can help with this. The performance monitor block of the processor (and OProfile) can be used to measure branch prediction rates in code. There are two effects at play here. Correct branch prediction saves clock cycles by avoiding pipeline flushes, but taking fewer conditional branches that skip forward over code can help performance by making more of the program fit within the L1 cache. 17.2.7 Optimizing L2 and outer cache usage Everything said about optimizations for using the L1 cache also applies to the L2 cache accesses. Best performance results from having a working dataset that is smaller than the L2 cache size, and where the data is used more than once; there is little benefit caching data that is used only once, other than possibly producing more optimal bus accesses. If the dataset is larger than the cache size, you can consider similar techniques to those already described for the L1 cache. There is, however, an additional point to consider with outer caches, which is that they might well be shared with other cores and therefore the effective size for an individual processor can be less than the actual size. In addition, when writing generic code to run on a number of ARM families, it can be difficult to make optimal use of the L2 cache. The presence of such a cache is not guaranteed and its size can vary significantly between systems. 17.2.8 Optimizing TLB usage In general, the scope for optimizing usage of the Translation Lookaside Buffer (see Chapter 9) is much less than for optimizing cache accesses. The key points are to minimize the number of pages in use (this obviously gives fewer TLB misses) and to use large MMU mappings (supersections or sections in preference to 4KB pages) as this reduces the cost of individual translation table walks (one external memory access rather than two) and also means that a larger amount of memory is represented within an individual TLB entry (also giving fewer TLB misses). In practice, however, an operating system like Linux uses 4KB pages everywhere, so the main optimization technique available is to separate the frequently accessed code and data from the infrequently accessed code and data (for example exception handling code can be moved to a different page) and trying to limit the number of frequently accessed pages to below ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-10 Optimizing Code to Run on ARM Processors the maximum number supported by the processor hardware. The main optimization would be to try to process multiple cache lines’ worth of data per page, so that the L1 cache is the limiting factor rather than TLB entries. 17.2.9 Data abort optimization Chapter 9 The Memory Management Unit, in the context of Linux, describes how data aborts will be generated by page faults on the first time that a memory page is accessed and again when the page is first written to. This means that the kernel abort handler is called to take appropriate action and there is a certain performance overhead to this. Simplistically, you can reduce this overhead by using fewer pages. Again, code optimizations that make code smaller will help, as will reducing the size of the data space. 17.2.10 Prefetching a memory block access ARM Cortex-A series processors contain sophisticated cache systems, and support for speculation and out of order execution that can hide latencies associated with memory accesses. However, accesses to the external memory system are usually sufficiently slow that there will still be some penalty. If you can prefetch instructions or data into the cache before you require them, you can hide this latency. ARM processors provide support for preloading of data, using the PLD instruction. The PLD instruction is a hint that enables you to request that data is loaded to the data cache in advance of it actually being read or written by the application. The PLD operation might generate a cache linefill or a data cache miss, independent of load and store instruction execution, while the core continues to execute other instructions. If supported and used correctly, PLD can significantly improve performance by hiding memory access latencies. There is also a PLI instruction that enables you to hint to the processor that an instruction load from a particular address is likely in the near future. This can cause the processor to preload the instructions to its cache. In addition to this programmer-initiated prefetch, the core might also support automatic data prefetching. Essentially, the core can detect a series of sequential accesses to memory. When it does, it automatically requests the following cache lines speculatively, in advance of the program actually using them. In many systems, significant numbers of cycles are consumed initializing or moving blocks of memory, using the memset() or memcpy() functions. Optimized ARM libraries will typically implement such functions by using Store Multiple instructions, with each store aligned to a cache line boundary. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-11 Optimizing Code to Run on ARM Processors 17.3 Source code modifications Profiling tools enable you to identify code segments or functions that can benefit from optimization and how different compiler options can enable compiler optimizations to our code. We will now consider a variety of source code modifications that can yield faster or smaller code on the ARM. 17.3.1 Loop termination For loops that have been identified by the profiler, it might be appropriate to have integer loop counters that end at 0 (zero), rather than start from 0 (zero). This is because a compare with zero comes for free with the ADD or SUB instruction used to update the loop counter, whereas a compare with a non-zero value will typically require an explicit CMP instruction. Replace a loop that counts up to a terminating value: for (i = 1; i<= total; i++) with one that counts down to zero: for (i = total; i != 0; i--) This will remove a CMP instruction from each iteration of the loop. It is also good practice to use int (32-bit) variables for loop counters. This is because the ARM is natively a 32-bit machine. Its ADD assembly language instruction operates on two 32-bit registers. If it carries out an ADD (or other data processing operation) with a smaller quantity, the compiler might insert additional instructions to handle overflow (see also Variable selection on page 17-13). 17.3.2 Loop fusion This is one of a variety of other possible loop techniques that can be employed either by you, or by an optimizing compiler. It essentially means merging loops that have the same iteration count and no interdependencies (Example 17-8 and Example 17-9). Example 17-8 Loop fusion for (i { x[i] } for (j { y[j] } = 0; i < 10; i++) = 1; = 0; j < 10; j++) = j; It is immediately apparent that this can be optimized to: Example 17-9 Fused loops for (i = 0; i < 10; i++) { x[i] = 1; y[i] = i; ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-12 Optimizing Code to Run on ARM Processors } It is worth mentioning that this approach can sometimes lead to a reduction in performance because of cache effects such as thrashing, depending on the cache associativity and the addresses of the data being accessed. 17.3.3 Reducing stack and heap usage In general, it is a good idea to try to minimize memory usage by code. The ARM processor has a register set that provides a relatively limited set of resources for the compiler to keep variables in. When all registers are allocated with currently live variables, additional variables will be spilled to the stack, causing memory operations and extra cycles for the code to execute. There are a number of ways available to you, to try to help. A key rule is to try to limit the number of live variables at any one time. Chapter 15 stated that up to four parameters can be passed in registers to a function. Additional parameters are passed on the stack. It is therefore significantly more efficient to pass four or fewer parameters than to pass five or more. Of course, the ARM registers in question are 32-bits in size and therefore if you pass a 64-bit variable, it will take two of our four register slots. For similar reasons, recursive functions do not typically yield efficient processor register usage. Remember also that non-static C++ functions also consume one argument slot with the this pointer. 17.3.4 Variable selection ARM integer registers are 32-bit sized and optimal code is therefore produced most readily when using 32-bit sized variables, as this avoids the requirement to provide extra code to deal with the case where a 32-bit result overflows an 8-bit or 16-bit sized variable. Consider the following code: unsigned int i, j, k; i = j+k; The compiler would typically emit assembly code similar to: ADD R0, R1, R2 If these variables were instead short (16-bit) or char (8-bit), the compiler must ensure the result does not overflow the halfword or byte. The same code might be as shown in Example 17-10, for signed halfwords (shorts). Example 17-10 Addition of 2 signed shorts (assembly code) ADD SXTH R0, R1, R2 R0, R0 Or for unsigned halfwords as in Example 17-11. Example 17-11 Addition of 2 unsigned shorts (assembly code) ADD BIC ARM DEN0013D ID012214 R0, R1, R2 R0, R0, #0x10000 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-13 Optimizing Code to Run on ARM Processors This has the effect of clipping the result to the defined size. Although the compiler can sometimes cope with such things as an incorrect type specification for a loop counter variable, it is generally best to use the correct type in the first place. 17.3.5 Pointer aliasing If a function has two pointers pa and pb, with the same value, we say the pointers alias each other. This introduces constraints on the order of instruction execution. If two write accesses that alias occur in program order, they must happen in the same order on the processor and cannot be re-ordered. This is also the case for a write followed by a read, or a read followed by a write. Two read accesses to aliases are safe to re-order. Because any pointer could alias any other pointer in C, the compiler must assume that memory regions accessed through these pointers can overlap, which prevents many possible optimizations. C++ enables more optimizations, as pointer arguments will not be treated as possible aliases if they point to different types. C99 introduces the restrict keyword that specifies that a particular pointer argument does not alias any other. If you know that pointers do not overlap, using this keyword to give the compiler this information can yield significant improvements. However, misusing it can lead to incorrect program function. The restrict keyword qualifies the pointer and not the object being pointed to. This consideration is not specific to the ARM architecture. When using GCC, you can enable the C99 standard by adding -std=c99 to your compilation flags. In code that cannot be compiled with C99, use either __restrict or __restrict__ to enable the keyword as a GCC extension. Consider the following simple code sequence: void foo(unsigned int *ptr1, unsigned int *ptr2, unsigned int *i) { *ptr1 += *i; *ptr2 += *i; } The pointers could possibly refer to the same memory location and this causes the compiler to generate code that is less efficient. In this example, it must read the value *i from memory twice, once for each add, as it cannot be certain that changing the value of *ptr1 does not also change the value of *i. If the function is instead declared as: void foo(unsigned int *restrict ptr1, unsigned int *restrict ptr2, unsigned int *restrict i) This means that the compiler can assume that the three pointers might not refer to the same location and optimize accordingly. You must ensure that the pointers never overlap. 17.3.6 Division and modulo Not all ARM processors have hardware support for division (See Table 2-3 on page 2-9). For these processors, C division typically calls a library routine that takes tens of cycles to run for divides of 32-bit integers, Note Division is slower than multiplication even in hardware. In performance-critical code it is almost always worth replacing it if possible. This must be done as a trade-off against code maintainability. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-14 Optimizing Code to Run on ARM Processors Where possible, divides must be avoided, or removed from loops. Division with a fixed divisor, that is, one that is known at compile time, is faster than dividing two variable quantities. The compiler can replace a divide by a shift-multiply pair in this case. A 32 × 32 multiply by fixed constant, then shift right to adjust the most significant word. Modulo arithmetic is another case to be aware of, as this will also use division library routines. The code minutes = (minutes + 1) % 60; will run significantly faster on machines with no hardware divide, if coded as if (++minutes == 60) minutes=0; that substitutes a two cycle add and compare in place of a call to a library function. 17.3.7 Extern data Accessing external variables requires the processor to execute a series of load instructions to acquire the address of the variable through a base pointer and then read the actual variable value. If multiple variables are defined as members of a structure, they can share a base pointer, saving cycles and instructions. It is therefore good practice to define the variables inside the same struct. 17.3.8 Inline or embedded assembler In some cases, it can be a worthwhile optimization to use assembly code, in addition to C. The general principle here is for you to code in a high level language, use a profiler to determine which sections will produce the most benefit if optimized and then inspect the compiler-produced assembly code to look for possible improvements. If a code section is identified as being a performance bottleneck, don’t reach immediately for the assembly language manual. Improvements to the algorithm should first be sought and then compiler optimizations tried before considering use of assembly code. Even then, it is often the case that poor performance is because of cache misses and memory access delays rather than the actual assembly code. The ARM Compiler, GCC, and most other C compilers use the –s flag to tell the compiler to produce assembly code output. The –fverbose-asm command line option can also be useful in gcc. Interleaved source and assembler can be produced by the ARM Compiler with the --interleave option. Chapter 14 Porting provides more information about the use of an inline assembler. 17.3.9 Complex addressing modes It is often better to avoid complex addressing modes. In cases where the address to be used for a load or store requires a complex calculation, dual-issue of instructions is not possible. Only the addressing mode that uses a base register plus an offset, specified either by a register or an immediate value, with an optional shift left by an immediate value of two is fast. Other, less commonly used, addressing modes can be executed more quickly by splitting into two instructions that might be dual-issued. For example: MOV R2, R1 LSL#3; LDR R2,[R0, R2] can be faster than LDR R2, [R0, R1 LSL #3] ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-15 Optimizing Code to Run on ARM Processors LDRH and LDRB have no extra penalty, but LDRSH and LDRSB have a single cycle load-use penalty, but no early forwarding path and can incur additional latency if a subsequent instruction uses the loaded value. 17.3.10 Unaligned access Unaligned LDRs have an extra cycle penalty compared with aligned loads, but unaligned LDRs that cross cache-lines have many cycles of additional penalty. In general, stores are less likely to stall the system compared to loads. STRB and STRH have similar performance to STR, because of the merging write buffer. Because there are four slots in the load/store unit, more than four consecutive pending loads will always cause a pipeline stall. 17.3.11 Linker optimizations Some code optimizations can be performed at the link, rather than the compile stage of the build, for example, unused section elimination and linker feedback. Multi-file optimization can be carried out across multiple C files, and unused sections can be removed. Similarly, multi-file compilation enables the compiler to perform optimization across multiple files instead of on individual files. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 17-16 Chapter 18 Multi-core processors Up to this point, we have considered the ARM processor core as a single entity. Most Cortex-A series processors, however, can include up to four processing cores. Multi-core systems can potentially deliver higher performance, because more processing units (cores) are available. This enables multiple tasks to be executed in parallel, potentially reducing the amount of time required to perform the allocated task. Multi-processing can be defined as running two or more sequences of instructions simultaneously within a single device containing two or more cores. The concept of multi-processing has been a subject of research for a number of decades, and has seen widespread commercial use over the past 15 years. Multi-processing is now a widely adopted technique in both systems intended for general-purpose application processors and in areas more traditionally defined as embedded systems. The overall energy consumption of a multi-core system can be significantly lower than that of a system based on a single processor core. Multiple cores enable execution to be completed faster and so some elements of the system might be completely powered down for longer periods. Alternatively, a system with multiple cores might be able to operate at a lower frequency than that required by a single processor to achieve the same throughput. A lower power silicon process or a lower supply voltage can result in lower power consumption and reduced energy usage. Most current systems do not permit the frequency of cores to be changed independently. However, each core can be dynamically clock gated, giving additional power and energy savings. Multi-core systems also add flexibility and scalability to system designs. A system that contains one or two cores could be scaled up for more performance by adding additional cores, without requiring redesign of the whole system or significant changes to software. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-1 Multi-core processors Having multiple cores at our disposal also enables more options for system configuration. For example, you might have a system that uses separate cores, one to handle a hard real-time requirement and another for an application requiring high, uninterrupted performance. These could be consolidated into a single multi-processor system. A multi-core device is also likely to be more responsive than one with a single core. When interrupts are distributed between cores there will be more than one core available to respond to an interrupt and fewer interrupts per core to be serviced. Multiple cores will also enable an important background process to progress simultaneously with an important but unrelated foreground process. Multi-core systems can also extract more performance from high latency memory systems (for example, DDR memory) by enabling a memory controller to queue up and optimize requests to the memory. Processors working on coherent data can benefit from reductions in linefills and evictions. When the data is not shared, it is likely that performance will be adversely affected. An L2 cache can mean improved utilization for shared memory regions (including file caches), shared libraries and kernel code. Additionally, if the number of cores is increased, then without a corresponding increase in memory bandwidth performance will also deteriorate. In the past, much software was written to operate within the context of a single core. Some operating systems provide support for time-slicing. this gives the illusion of multiple processes or tasks running simultaneously. It is important to clearly understand the difference between multi-threading, for example POSIX threads, or Java and multi-processing. A multi-threaded application can be run on a single core, but only with multi-processing can the threads truly execute in parallel. Migrating multi-threaded software from a single core system to a multi-core one can trigger problems with incorrect programs that could not be exposed by running the same program time-sliced on a single core. It can also cause very infrequent bugs to become very frequently triggered. What it cannot do is to cause correctly written multi-threaded programs to misbehave, only expose previously unnoticed errors. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-2 Multi-core processors 18.1 Multi-processing ARM systems From early in the history of the architecture, ARM processors were likely to be implemented in systems that contained other processors. This commonly meant a heterogeneous system, perhaps containing an ARM processor plus a separate DSP processor. Such systems have different software executing on different cores and the individual processors can have differing privileges and views of memory. Many widely used ARM systems, such as the TI OMAP series, or the Freescale i.MX, are examples of this. GIC-400 Interrupts Interrupts A15 A15 A15 core A7 A15 A15 core L2 Cache L2 Cache CCI-400 Memory controller ports System port Figure 18-1 Example of a multi-cluster system We can distinguish between systems that contain: • A single processor containing a single core, such as the Cortex-A8 processor. • A multi-core processor, that contains several cores capable of independent instruction execution, that can be externally viewed as a single unit or cluster, either by the system designer or by an operating system that can abstract the underlying resources from the application layer. • Multiple clusters (such as that shown in Figure 18-1), in which each cluster contains multiple cores. ARM was among the first companies to introduce multi-core processors to the SoC market, when it introduced the ARM11 MPCore processor in 2004. All the processors described in this book, with the exception of the Cortex-A8 processor, are examples of such multi-core systems. An ARM multi-core processor can contain between one and four cores. Each core can be individually configured to take part (or not) in a data cache coherency management scheme. A Snoop Control Unit (SCU) device inside the processor has the task of automatically maintaining level 1 data cache coherency, between cores within the cluster without software intervention. ARM multi-core processors include an integrated interrupt controller. Multiple external interrupt sources can be independently configured to target one or more of the individual processor cores. Furthermore, each core is able to signal (or broadcast) any interrupt to any other core or set of cores in the system, from software (software triggered interrupts). These mechanisms enable the OS to share and distribute interrupts across all cores and to coordinate activities using the low-overhead signaling mechanisms provided. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-3 Multi-core processors Cortex-A MPCore processors also provide hardware mechanisms to accelerate OS kernel operations such as system-wide cache and TLB maintenance operations. (This feature is not found in the ARM11 MPCore.) Each of the Cortex-A series multi-core processors have the following features: • Configurable between one and four cores (at design time). • Level 1 data cache coherency. • Integrated interrupt controller. • Local timers and watchdogs. • An optional Accelerator Coherency Port (ACP). Figure 18-2 illustrates the structure of the Cortex-A9 MPCore processor, though a generalized form of this description also applies to other multi-core processors. These features are described in more detail in the course of this chapter and those that follow. CoreSight MultiCore Debug and Trace Architecture FPU/ NEON PTM I/F Cortex-A9 processor I-Cache D-Cache FPU/ NEON PTM I/F Cortex-A9 processor I-Cache D-Cache FPU/ NEON PTM I/F Cortex-A9 processor I-Cache D-Cache FPU/ NEON PTM I/F Cortex-A9 processor I-Cache D-Cache Snoop Control Unit (SCU) Generic Interrupt Control and Distribution Cache-to-Cache Transfers Snoop Filtering Timers Accelerator Coherency Port Advanced Bus Interface Unit Primary AMBA3 64-Bit Interface Optional 2nd I/F with Address Filtering Figure 18-2 Cortex-A9 MPCore processor ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-4 Multi-core processors 18.2 Symmetric multi-processing Symmetric multi-processing (SMP) is a software architecture that dynamically determines the roles of individual processors. Each core in the cluster has the same view of memory and of shared hardware. Any application, process or task can run on any core and the operating system scheduler can dynamically migrate tasks between cores to achieve optimal system load. We expect that readers will be familiar with the fundamental operating principles of an OS, but OS terminology will be briefly reviewed here. An application that executes under an operating system is known as a process. It performs many operations through calls to the system library that provides certain functions from library code, but also acts a wrapper for system calls to kernel operations. Individual processes have associated resources, including stack, heap and constant data areas, and properties such as scheduling priority settings. The kernel view of a process is called a task. For the purposes of describing SMP operation, we will use the term kernel to represent that portion of the operating system that contains exception handlers, device drivers and other resource and process management code. We will also assume the presence of a task scheduler that is typically called using a timer interrupt. The scheduler is responsible for time-slicing the available cycles on cores between multiple tasks, dynamically determining the priority of individual tasks and deciding which task to run next. Threads are separate tasks executing within the same process space that enable separate parts of the application to execute in parallel on different cores. They also permit one part of an application to keep executing while another part is waiting for a resource. In general, all threads within a process share a number of global resources (including the same memory map and access to any open file and resource handles). Threads also have their own local resources, including their own stacks and register usage that will be saved and restored by the kernel on a context switch. The fact that these resources are local does not, however, mean that the local resources of any thread are guaranteed to be protected from incorrect accesses by other threads. Threads are scheduled individually and can have different priority levels even within a single process. An SMP-capable operating system provides an abstracted view of the available core resources to the application. Multiple applications can run concurrently in an SMP system without re-compilation or source code changes. A conventional multitasking OS enables the system to perform several task or activities at the same time, in both single-core and multi-core processors. There is, however, a key distinction. In the single (uniprocessor) case, multitasking is performed through time-slicing of a single core, giving the illusion of many tasks being performed at the same time. In a multi-core system, we can have true concurrency; multiple tasks are actually run at the same time, in parallel, on separate cores. The role of managing the distribution of such tasks across the available cores is performed by the operating system. Typically, the OS task scheduler can distribute tasks across available cores in the system. This feature, known as load balancing, is aimed at obtaining better performance, or energy savings or even both. For example, with certain types of workloads, energy savings can be achieved if the tasks making up the workload are scheduled on fewer cores. This would allow more resources to be left idling for longer periods, thereby saving energy. In other cases, the performance of the workload could be increased if the tasks were spread across more cores. These tasks could make faster forward progress, without getting perturbed by each other than if they ran on fewer cores. In another case, it might be worth running tasks on more cores at reduced frequencies as compared to fewer cores at higher frequencies. Doing this could provide a better trade-off between energy savings and performance. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-5 Multi-core processors The scheduler in an SMP system can dynamically re-prioritize tasks. This dynamic task prioritization enables other tasks to run while the current task sleeps. In Linux, for example, tasks whose performance is bound by I/O activity can have their priority decreased in favor of tasks whose performance is limited by processor activity. The I/O-bound task will typically schedule I/O activity and then sleep pending such activity. Interrupt handling can also be load balanced across cores. This can help improve performance or save energy. Balancing interrupts across cores or reserving cores for particular types of interrupts can result in reduced interrupt latency. This might also result in reduced cache use which will help improve performance. Using fewer cores for handling interrupts could result in more resources idling for longer periods, resulting in an energy saving at the cost of reduced performance. The Linux kernel does not support automatic interrupt load balancing. However, the kernel provides mechanisms to change the binding of interrupts to particular cores. There are open source projects such as irqbalance https://github.com/Irqbalance/irqbalance which use these mechanisms to arrange a spread of interrupts across the available cores. irqbalance is made aware of system attributes such as the shared cache hierarchy (which cores have a common cache) and power domain layout (which cores can be powered off independently). It can then determine the best interrupt to core binding. An SMP system will by definition have shared memory between cores in the cluster. To maintain the required level of abstraction to application software, the hardware must take care of providing a consistent and coherent view of memory for you. Changes to shared regions of memory must be visible to all cores without any explicit software coherency management. Likewise, any updates to the memory map (for example because of demand paging, allocation of new memory or mapping a device into the current virtual address space) of either the kernel or applications must be consistently presented to all cores. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-6 Multi-core processors 18.3 Asymmetric multi-processing In an Asymmetric Multi-processing (AMP) system, you can statically assign individual roles to a core within a cluster, so that in effect, you have separate cores, each performing separate jobs, within each cluster. This can be referred to as a function-distribution software architecture and typically means that you have separate operating systems running on the individual cores. The system can appear to you as a single-core system with dedicated accelerators for certain critical system services. In general AMP does not refer to systems in which tasks or interrupts are associated with a particular core. In an AMP system, each task can have a different view of memory and there is no scope for a core that is highly loaded to pass work to one that is lightly loaded. There is no requirement for hardware cache coherency in such systems, although there will typically be mechanisms for communication between the cores through shared resources, possibly requiring dedicated hardware. The system described in Cache coherency on page 18-9 can help reduce the overheads associated with sharing data between the systems. Reasons for implementing an AMP system using a multi-core processor might include security, a requirement for guaranteeing meeting of real-time deadlines, or because individual cores are dedicated to perform specific tasks. There are classes of systems that have both SMP and AMP features. This means that we have two or more cores running an SMP operating system, but the system has additional elements that do not operate as part of the SMP system. The SMP sub-system can be regarded as one element within the AMP system. Cache coherency is implemented between the SMP cores, but not necessarily between SMP cores and AMP elements within the system. In this way, independent subsystems can be implemented within the same cluster. It is entirely possible (and normal) to build AMP systems in which individual cores are running different operating systems (these are called Multi-OS systems). The selection of software MP model is determined by the characteristics of the applications running in the system. In networking systems, for example, it can be convenient to provide a separation between control-plane (AMP) and data-plane (SMP) sections of the system. In other systems, it might be desirable to isolate those parts of the system that require hard real-time response from applications that require raw performance and to implement these on separate cores. Note Where synchronization is required between these separate cores, it can be provided through message passing communication protocols, for example, the Multicore Communications Association API (MCAPI). These can be implemented by using shared memory to pass data packets and by the use of software triggered interrupts to implement a so-called door-bell mechanism. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-7 Multi-core processors 18.4 Heterogeneous multi-processing The term Heterogeneous multi-processing (HMP) finds application in many different contexts. It is often conflated with AMP to describe systems that are composed of different types of processors, such as a multi-core ARM applications processor and an application specific processor (such as a baseband controller chip or an audio codec chip). ARM uses HMP to mean a system composed of clusters of application processors that are 100% identical in their instruction set architecture but very different in their microarchitecture. All the processors are fully cache coherent and a part of the same coherency domain. This is best explained using the ARM implementation of HMP technology known as big.LITTLE. In a big.LITTLE system energy efficient LITTLE cores are coherently coupled with high performance big cores to form a system that can accomplish both high intensity and low intensity tasks in the most energy efficient manner. GIC-400 Generic Interrupt Controller Cortex-A7 core Cortex-A7 core Cortex-A15 core Cortex-A15 core L2 Cache L2 Cache Cortex-A7 Cluster Cortex-A15 Cluster IO coherent master CCI-400 Cache Coherent Interconnect Memory controller ports System ports Figure 18-3 A typical big.LITTLE system The central principle of big.LITTLE is that application software can run unmodified on either type of processor. For a detailed overview of big.LITTLE technology, including the software execution models see Chapter 23. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-8 Multi-core processors 18.5 Cache coherency Coherency is about ensuring all processors, or bus masters in the system see the same view of memory. It means that changes to data held in the cache of one core are visible to the other cores, making it impossible for cores to see stale copies of data (the old data from before it was changed by the first core). For example, if you have a processor that is creating a data structure then passing it to a DMA engine to move, both the processor and DMA must see the same data. If that data were cached in the core and the DMA reads from external memory, the DMA will read old, stale data. There are three mechanisms to maintain coherency: Disable caching This is the simplest mechanism but might cost significant core performance. To get the highest performance processors are pipelined to run fast, and to run from caches that offer a very low latency. Caching of data that is accessed multiple times increases performance significantly and reduces DRAM accesses and power. Marking data as “non-cached” could impact performance and power. Software managed coherency Software managed coherency is the traditional solution to the data sharing problem. Here the software, usually device drivers, must clean or flush dirty data from caches, and invalidate old data to enable sharing with other processors or masters in the system. This takes processor cycles, bus bandwidth, and power. Where there are high rates of sharing between requesters the cost of software cache maintenance can be significant, and can limit performance. Hardware managed coherency Hardware Coherency is the most efficient solution. Any data marked ‘shared’ in a hardware coherent system will always be up to date. All cores and bus masters in that sharing domain see the exact same value. While hardware coherency might add some complexity to the interconnect and clusters, it greatly simplifies the software and enables applications that would not be possible with software coherency. 18.5.1 MESI and MOESI protocols There are a number of standard ways by which cache coherency schemes can operate. Most ARM processors use the MOESI protocol, while the Cortex-A9 uses the MESI protocol. Depending on which protocol is in use, the SCU marks each line in the cache with one of the following attributes: M (Modified), O (Owned), E (Exclusive), S (Shared) or I (Invalid). These are described below: ARM DEN0013D ID012214 Modified The most up-to-date version of the cache line is within this cache. No other copies of the memory location exist within other caches. The contents of the cache line are no longer coherent with main memory. Owned This describes a line that is dirty and in possibly more than one cache. A cache line in the owned state holds the most recent, correct copy of the data. Only one core can hold the data in the owned state. The other cores can hold the data in the shared state. Exclusive The cache line is present in this cache and coherent with main memory. No other copies of the memory location exist within other caches. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-9 Multi-core processors Shared The cache line is present in this cache and coherent with main memory. Copies of it can also exist in other caches in the coherency scheme. Invalid The cache line is invalid. The rules for the standard implementation of the protocol are as follows: • A write can only be done if the cache line is in the Modified or Exclusive state. If it is in the Shared state, all other cached copies must be invalidated first. A write moves the line into the Modified State. • A cache can discard a Shared line at any time, changing to the Invalid state. A Modified line is written back first. • If a cache holds a line in the Modified state, reads from other caches in the system will get the updated data from the cache. Conventionally, this is done by first writing the data to main memory and then changing the cache line to the Shared state, before performing a read. • A cache that has a line in the Exclusive state must move the line to the Shared state when another cache reads that line. • The Shared state might not be precise. If one cache discards a Shared line, another cache might not be aware that it could now move the line to Exclusive status. ARM multi-core processors also implement optimizations that can copy clean data and move dirty data directly between participating L1 caches, without having to access (and wait for) external memory. This activity is handled in multi-core systems by the Snoop Control Unit (SCU). 18.5.2 Snoop Control Unit The SCU maintains coherency between the L1 data cache of each core. Since executable code changes much less frequently, this functionality is not extended to the L1 instruction caches. The coherency management is implemented using a MOESI-based protocol, optimized to decrease the number of external memory accesses. In order for the coherency management to be active for a memory access, all of the following must be true: • The SCU is enabled, through its control register located in the private memory region. See Private memory region on page 18-19. The SCU has configurable access control, restricting which processors can configure it. • The core performing the access is configured to participate in the Inner Shareable domain, configured using the operating system at boot time, by setting the somewhat misleadingly named SMP bit in the CP15:ACTLR, Auxiliary Control Register. The following code example sets the SMP bit in either a Cortex-A7 or Cortex-A15 ACTLR: MRC ORR MCR DSB ARM DEN0013D ID012214 p15, 0, r0, c1, c0, 1 r0, r0, #0x040 p15, 0, r0, c1, c0, 1 ; Read ACTLR ; Set bit[6] SMP (coherency) ; Write ACTLR • The MMU is enabled. • The page being accessed is marked as Normal Shareable, with a cache policy of write-back, write-allocate. Device and Strongly-ordered memory, however, are not cacheable, and write-through caches behave like uncached memory from the point of view of the core. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-10 Multi-core processors The SCU can only maintain coherency within a single cluster. If there are additional processors or other bus masters in the system, explicit software synchronization is required when these share memory with the MP block. Alternatively, the Accelerator Coherency Port (ACP) can be used. 18.5.3 Accelerator Coherency Port (ACP) The Accelerator Coherency Port (ACP) is a feature of the Cortex-A5, Cortex-A9, Cortex-A12 and Cortex-A15 processors. It provides an AXI slave interface into the Snoop Control Unit of the processor. The AXI bus interface is defined in the ARM AMBA specification. This slave interface can be connected to an external uncached AXI master, such as a DMA engine, for example, which is able to initiate both read and write memory transfers to the ACP. It enables such a device to snoop the L1 caches of all cores, avoiding the requirement for synchronization through external memory. A cached device can also be connected, but this requires manual coherency management through software. The behavior of accesses performed on the ACP is as follows: • Addresses used by the ACP are physical addresses that can be snooped by the SCU to be fully coherent. • ACP reads can hit in the L1 D-cache of any core. • Writes on the ACP invalidate any stale data in L1 and write-through to L2. The ACP enables an external device to see core-coherent data without knowledge of where the data is in the memory hierarchy. Memory transfers are automatically coherent in the same way as happens between the L1 D-caches of the cores in the cluster. Use of the ACP can both increase performance and save power, as there will be reduced traffic to external memory and faster execution. Programmers writing device drivers that use the ACP do not have to be concerned with coherency issues, because no cache cleaning or invalidation is required to ensure coherency. However, the use of memory barriers (DMB) or external cache synchronization operations can still be necessary, if a particular ordering must be enforced. 18.5.4 The Cache Coherent Interface (CCI) Extending hardware coherency to a multi-cluster system requires a coherent bus protocol. In 2011 ARM released the AMBA 4 ACE specification that introduces the AXI Coherency Extensions (ACE) on top of the popular AXI protocol. The full ACE interface enables hardware coherency between clusters and enables an SMP operating system to extend to more cores. In the ACE protocol, three coherency channels are added in addition to the normal five channels of AXI. If you have two clusters, any shared access to memory can snoop into the cache of the other cluster to see if the data is already on chip, if not, it is fetched from external memory. The AMBA 4 ACE-Lite interface is designed for I/O (or one-way) coherent system masters like DMA engines, network interfaces and GPUs. These devices might not have any caches of their own, but they can read shared data from the ACE processors. The CoreLink™ CCI-400 is one of the first implementations of AMBA 4 ACE and supports up to two ACE processor clusters enabling up to eight cores to see the same view of memory and run an SMP OS. If we return to our example of a multi-cluster system: ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-11 Multi-core processors GIC-400 Interrupts ACE Interface Interrupts A15 A15 A15 core A7 A15 A15 core L2 Cache L2 Cache 1. Coherent read request 3. Read data 2 Data available CC1-400 Cache coherent interconnect Figure 18-4 Cache coherency in a multi-cluster system Figure 18-4 shows the steps in a coherent data read from the Cortex-A7 cluster to the Cortex-A15 cluster. This starts with the Cortex-A7 cluster issuing a Coherent Read Request. The CCI-400 hands over the request to the Cortex-A15 processor to snoop into Cortex-A15 cluster cache. When the request from the CCI-400 is received, the Cortex-A15 cluster checks the data availability and reports this information back. If the requested data is in the cache, the CCI-400 moves the data from the Cortex-A15 cluster to the Cortex-A7 cluster, resulting in a cache linefill in the Cortex-A7 cluster. The CCI-400 and the ACE protocol enable full coherency between the Cortex-A15 and Cortex-A7 clusters, enabling data sharing to take place without external memory transactions. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-12 Multi-core processors 18.6 TLB and cache maintenance broadcast An individual core can broadcast Translation Lookaside Buffer and cache maintenance operations to other cores in the Inner Shareable coherency domain. This can be required whenever shared translation tables are modified, for example. This behavior might have to be enabled by you. For example, in the Cortex-A9 processor, this is controlled by the FW bit in the Auxiliary Control Register (ACTLR). Maintenance operations can only be broadcast and received when the processor is configured to participate in the Inner Shareable domain, using the SMP bit in ACTLR. Only Inner Shareable operations are broadcast, for example: • To invalidate TLB entry by virtual address. • To clean or invalidate data cache line by virtual address. • To invalidate instruction cache line by virtual address. Some care is required with cache maintenance activity in multi-core systems that include a L2C-310 L2 cache (or similar). Cleaning or invalidating the L1 cache and L2 cache will not be a single atomic operation. A core might therefore perform cache maintenance on a particular address in both L1 and L2 caches only as two discrete steps. If another core were to access the affected address between those two actions, a coherency problem can occur. Such problems can be avoided by following two simple rules. ARM DEN0013D ID012214 • When cleaning, always clean the innermost (L1) cache first and then clean the outer cache(s). • When invalidating, always invalidate the outermost cache first and the L1 cache last. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-13 Multi-core processors 18.7 Handling interrupts in an SMP system Multi-core processors include an integrated interrupt controller that implements the GIC architecture (see The Generic Interrupt Controller on page 12-7 for additional details). This controller provides 32 private interrupts per core, of which the lower 16 are Software Generated Interrupts (SGI) that can be generated only through software operations, and the rest are Private Peripheral Interrupts (PPI). It also provides a configurable number of Shared Peripheral Interrupts (SPI), up to 224 in current multi-core implementations). It supports interrupt prioritization, pre-emption and routing to different cores. In a multi-core processor, the GIC control registers are memory-mapped and located within the Private memory region, see Private memory region on page 18-19. The Interrupt Processor Targets registers configure that cores individual interrupts are routed to. They are ignored for private interrupts. The registers controlling the private interrupts (0-31) are banked, so that each core can have its own configuration for these. This includes priority configuration and the enabling or disabling of individual interrupts. The Software Generated Interrupt (SGI) Register can assert a private SGI on any core, or a groups of cores. The priority of a software interrupt is determined by the priority configuration of the receiving core, not the one that sends the interrupt. The interrupt acknowledge register bits [12:10] will provide the ID of the core that made the request. The target list filter field within this register provides shorthand for an SGI to be sent to all processors, all but self or to a target list. For a software generated interrupt in a multi-core processor, the Interrupt Acknowledge Register also contains a bitfield holding the ID of the core that generated it. When the interrupt service routine has completed, it must write-back the value previously read from the Interrupt Acknowledge Register into the End Of Interrupt Register. For an SGI, this must also match the ID of the signalling core. In both AMP and SMP systems, it is likely that cores will trigger interrupts on other cores (or themselves), a so called softirq. These can be used for kernel synchronization operations, or for communicating between AMP cores. For operations requiring more information passed than a raised interrupt, you can use a shared buffer to store messages. Before the core can receive an interrupt, some initialization steps are required both in the distributor and in the core interface. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-14 Multi-core processors 18.8 Exclusive accesses In an SMP system, data accesses frequently have to be restricted to one modifier at any particular time. This can be true of peripheral devices, but also for global variables and data structures accessed by more than one thread or process. Furthermore, library code that is used by multiple threads must be designed to ensure that concurrent access or execution is possible – it must be reentrant. Note Code is reentrant if it can be interrupted in the middle of its execution and then be called again before the previous invocation has completed. Protection of such shared resources is often through a method known as mutual exclusion. The section of code that is being executed by a core while accessing such a shared resource is known as the critical section. In a single core system, mutual exclusion can be achieved by disabling interrupts when inside critical sections. This is not sufficient in a multi-core system, as disabling interrupts on one core will not prevent others from entering the critical section. It is also not ideal since interrupts cannot be disabled from within User mode. Of course, there are other problems with this technique, including reduced responsiveness to real-time events, particularly if the critical section is long. In a multi-core system, we can use a spinlock – effectively a shared flag with an atomic (indivisible) mechanism to test and set its value. We perform this operation in a tight loop to wait until another thread (or core) clears the flag. We require hardware assistance in the form of special machine instructions to implement this. Application developers should not worry about the low-level implementation detail, but should instead become familiar with the lock and unlock calls available in their OS or threading library API. The ARM architecture provides three instructions relating to exclusive access. Variants of these instructions, that operate on byte, halfword, word or doubleword sized data, are also provided. The instructions rely on the ability of the core or memory system to tag particular addresses to be monitored for exclusive access by that core, using an exclusive access monitor. • LDREX (Load Exclusive) performs a load of memory, but also tags the physical address to be monitored for exclusive access by that core. • STREX (Store Exclusive) performs a conditional store to memory, succeeding only if the target location is tagged as being monitored for exclusive access by that core. This instruction returns the value of 1 in a general purpose register if the store does not take place, and a value of 0 if the store is successful. • CLREX (Clear Exclusive) clears any exclusive access tag for that core. Load Exclusive and Store Exclusive operations must be performed only on Normal memory (see Normal memory on page 10-4) and have slightly different effect depending on whether the memory is marked as Shareable or not. If the core reads from Shareable memory with an LDREX, the load happens and that physical address is tagged to be monitored for exclusive access by that core. If any other core writes to that address and the memory is marked as Shareable, the tag is cleared. If the memory is not Shareable then any attempt to write to the tagged address by the one that tagged it results in the tag being cleared. If the core does an additional LDREX to a different address, the tag for the previous LDREX address is cleared. Each core can only have one address tagged. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-15 Multi-core processors STREX can be considered as a conditional store. The store is performed only if the physical address is still marked as exclusive access (this means it was previously tagged by this core and no other core has since written to it). STREX returns a status value showing if the store succeeded. STREX always clears the exclusive access tag. The use of these instructions is not limited to multi-core systems. In fact, they are frequently employed in single core systems, to implement synchronization operations between threads running on the same core. In hardware, the core includes a device named the local monitor. This monitor observes the core. When the core performs an exclusive load access, it records that fact in the local monitor. When it performs an exclusive store, it checks that a previous exclusive load was performed and fails the exclusive store if this was not the case. The architecture enables individual implementations to determine the level of checking performed by the monitor. The core can only tag one physical address at a time. An LDREX from a particular address can be followed shortly after by an STREX to the same location, before an LDREX from a different address is performed. This is because the local monitor does not have to store the address of the exclusive tag (although it can do, if the processor implementer decides to do this). The architecture enables the local monitor to treat any exclusive store as matching a previous LDREX address. For this reason, use of the CLREX instruction to clear an existing tag is required on context switches. Where exclusive accesses are used to synchronize with external masters outside the core, or to regions marked as Sharable even between cores in the same cluster, it is necessary to implement a global monitor within the hardware system. This acts as a wrapper to one or more memory slave devices and is independent of the individual cores. This is specific to a particular SoC and might not exist in any particular system. An LDREX/STREX sequence performed to a memory location that has no suitable exclusive access monitor will fail, with the STREX instruction always returning 1. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-16 Multi-core processors 18.9 Booting SMP systems Initialization of the external system might have to be synchronized between cores. Typically, only one of the cores in the system has to run code that initializes the memory system and peripherals. Similarly, the SMP operating system initialization typically runs on only one core – the primary core. When the system is fully booted, the remaining cores are brought online and this distinction between the primary core and the others (secondary cores) is lost. If all of the cores come out of reset at the same time, they will normally all start executing from the same reset vector. The boot code will then read the cluster ID to determine which core is the primary. The primary core will perform the initialization and then signal to the secondary ones that everything is ready. An alternative method is to hold the secondary cores in reset while the primary core does the initialization. This requires hardware support to co-ordinate the reset. In an AMP system, the bootloader code will determine the suitable start address for the individual cores, based on their cluster ID (as each core will be running different code). Care might be required to ensure correct boot order in the case where there are dependencies between the various applications running on different cores. 18.9.1 Processor ID Booting provides a simple example of a situation where particular operations must be performed only on a specific core. Other operations perform different actions dependent on the core on which they are executing. The CP15:MPIDR Multiprocessor Affinity Register provides an identification mechanism in a multi-core system. This register was introduced in version 7 of the ARM architecture, but was in fact already used in the same format in the ARM11 MPCore. In its basic form, it provides up to three levels of affinity identification, with 8 bits identifying individual blocks at each level (Affinity Level 0, 1 and 2). This information can also be of value to an operating system scheduler, as an indication of the order of magnitude of the cost of migrating a process to a different core, processor or cluster. The format of the register was slightly extended with the ARMv7-A multiprocessing extensions. This extends the previous format by adding an identification bit to reflect that this is the new register format, and adds the U bit that indicates whether the current core is the only core in a single-core implementation or not. 18.9.2 SMP boot in Linux The boot process for the primary core is as described in Boot process on page 11-14. The method for booting the secondary cores can differ somewhat depending on the SoC being used. The method that the primary core invokes in order to get a secondary core booted into the operating system is called boot_secondary() and must be implemented for each mach type that supports SMP. Most of the other SMP boot functionality is extracted out into generic functions in linux/arch/arm/kernel. The method below describes the process on an ARM Versatile Express development board (mach-vexpress). While the primary core is booting, the secondary cores will be held in a standby state, using the WFI instruction. It will provide a startup address to the secondary cores and wake them using an Inter-Processor Interrupt (IPI), meaning an SGI signalled through the GIC (see Handling interrupts in an SMP system on page 18-14). Booting of the secondary cores is serialized, using the global variable pen_release. Conceptually, you can think of the secondary cores being in a ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-17 Multi-core processors holding pen and being released one at a time, under control of the primary core. The variable pen_release is set by the kernel code to the ID value of the processor to boot and then reset by that core when it has booted. When an inter-processor interrupt occurs, the secondary core will check the value of pen_release against their own ID value using the MPIDR register. Booting of the secondary core will proceed in a similar way to the primary. It enables the MMU (setting the TTB register to the new translation tables already created by the primary). It enables the interrupt controller interface to itself and calibrates the local timers. It sets a bit in cpu_online_map and calls cpu_idle(). The primary processor will see the setting of the appropriate bit in cpu_online_map and set pen_release to the next secondary core. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-18 Multi-core processors 18.10 Private memory region In the Cortex-A5, and Cortex-A9 MPCore processors, all of the internal peripherals are mapped to the private address space. This is an 8KB region location within the memory map at an address determined by the hardware implementation of the specific device used (this can be read using the CP15 Configuration Base Address Register). The registers in this region are fixed in little-endian byte order, so some care is required if the CPSR E bit is set when accessing it. Some locations within the region exist as banked versions, dependent on the processor ID. The Private memory region is not accessible through the ACP. Table 18-1 shows the layout of this Private memory region. Table 18-1 Private memory region layout Base Address offset Function 0x0000 Snoop Control Unit (SCU) 0x0100 Interrupt controller CPU Interface 0x0200 Global Timer 0x0600 Local Timer/Watchdog 0x1000 Interrupt Controller Distributor 18.10.1 Timers and watchdogs Each core in a multi-core processor implements a standard timer and a watchdog, both private to that core. These can be configured to trigger after a number of core cycles, using a 32-bit start value and an 8-bit pre-scaler. They can be operated using interrupts, or by periodic polling, supported by the Timer and Watchdog Interrupt Status Registers. They stop counting while the core is in debug state. The timer can be configured in single-shot or auto-reload mode. The watchdog can be operated in classic watchdog fashion, where it asserts the core reset signal, for that specific core on time-out. Alternatively, it can be used as a second timer. The Cortex-A9, and Cortex-A5 processors also include a global timer, shared between all cores, but with banked comparator and auto-increment registers for each core. It is a single, incrementing 64-bit counter, accessible only through 32-bit accesses. It can be configured to trigger an interrupt when the comparator value is reached. The auto-increment feature causes the processor comparator register to be incremented after each match. This is typically used by the OS scheduler, to trigger the scheduler on each core, at different times. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 18-19 Chapter 19 Parallelizing Software Chapter 18 Multi-core processors, described how an SMP system can enable you to run multiple threads efficiently and concurrently across multiple cores. In this case, the parallelization is, in effect, handled on our behalf by the OS scheduler. In many cases, however, this is insufficient and you must take steps to rewrite code to take advantage of speed-ups available through parallelization. An obvious example is where a single application requires more performance than can be delivered by a single core. More commonly, we can have the situation where an application requires much more performance than all of the others within a system, when it is said to be dominant. This prevents efficient energy usage, as we cannot perform optimal load-balancing. An unbalanced load distribution does not permit efficient dynamic voltage and frequency scaling. The operating system cannot automatically parallelize an application. It is limited to treating that application as a single scheduling unit. In such cases, the application itself has to be split into multiple smaller tasks by you. Of course, this means each of these tasks must be able to be independently scheduled by the OS, as separate threads. A thread is a part of a program that can be run independently and concurrently with other parts of a program. If you decompose an application into smaller execution entities that can be separately scheduled, the OS can spread the threads of the application across multiple cores. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-1 Parallelizing Software 19.1 Amdahl’s law Amdahl’s Law defines the theoretical maximum speedup achievable by parallelizing an application. The maximum speedup is given by the formula: Max speedup = 1/ ((1-P) + (P/N)) where: P = parallelizable proportion of program, N = Number of cores. This is, of course, an abstract, academic view. In practice, this provides a theoretical maximum speedup, as there are a number of overheads associated with concurrency. Synchronization overheads occur when a thread must wait for another task or tasks before it can continue execution. If a single task is slow, the whole program must wait. In addition, you might have critical sections of code, where only a single task is able to run at a time. There might also be occasions when all tasks are contending for the same resource or where no other tasks can be scheduled to run by the OS. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-2 Parallelizing Software 19.2 Decomposition methods The best approach to decomposition of an application into smaller tasks capable of parallel execution depends on the characteristics of the original application. Large data-processing algorithms can be broken down into smaller pieces by sub-division into a number of similar threads that execute in parallel on smaller portions of a dataset. This is known as data decomposition. Consider the example of color-space conversion, from RGB to YUV. Start with an array of pixel data. The output is a similar array giving chrominance and luminance data for each pixel. Each output value is calculated by performing a small number of multiplies and adds. Crucially, the output Y, U and V values for each pixel depend only on the input R, G and B values for that pixel. There is no dependency on the data values of other pixels. Therefore, the image can be divided into smaller blocks and you can perform the calculation using any number of instances of your code. This does not require any change to your original algorithm, only changes to the amount of data supplied to each thread. Split the image into stripes (1/N arrays, where we have N threads) so that each thread works on a stripe. The level of detail of the stripes can be an important consideration, given that it is clearly better for cacheability if each thread works on a contiguous block of pixels in array order. The code does not have to be modified to take care of scheduling, the operating system takes care of it. Color space conversion would be a task where the NEON unit could significantly improve performance. Splitting the task across several cores can provide additional parallelization gains than using NEON instructions alone. A different approach is that of task decomposition. Here, areas of code that are independent of each other and capable of being executed concurrently can be identified. This is a little more difficult, as you must consider the discrete operations being carried out and the interactions between them. A simple example might be the start-up sequence of a program. One task might be to check that you have a valid license for the software. Another task might be to display a start-up banner with a copyright message. These are independent tasks with no dependency on each other and can be performed in separate threads. Again, no change is required to the source code that carries out these isolated tasks. These must be supplied to the OS kernel Scheduler as separate execution threads. Of course, not all algorithms are able to be handled through data or task decomposition. Instead, you must analyze the program with the aim of identifying functional blocks. These are independent pieces of code with defined inputs and outputs that have some scope to be parallelized. Such functional blocks often depend on input from other blocks (they have a serial dependency), but do not have a corresponding dependency on time (a temporal dependency). This is (in some respects) analogous to the hardware pipelining employed in the core itself. The software for an MPEG video encoder provides a good example of this. Input data, in the form of an analog video signal is sampled and processed through a pipeline of discrete functional blocks. First, both inter-frame and intra-frame redundancies are removed. Then, quantization takes place to reduce the number of bits required to represent the video. After this, motion vector compensation takes place, run length compression and finally the encoded sub-stream is stored. At the same time that data from one frame is being run-length compressed and stored, you can also start to process the next frame. Within a frame, the motion vector compensation process can be parallelized. You can use multiple parallel threads to operate on a frame (an example of data decomposition). ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-3 Parallelizing Software When decomposing an application using these techniques, you must consider the overheads associated with task creation and management. An appropriate level of granularity is required for best performance. If you make your datasets too small, too big, or have too many datasets, it can reduce performance. In the color-space conversion example, it would not be sensible to have a separate thread for each pixel, even though this is logically possible. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-4 Parallelizing Software 19.3 Threading models When an algorithm has been analyzed to determine the potential changes that can be made for parallelization, you must modify code to map the algorithm to smaller, threaded execution units. There are two widely-used threading models, the workers’ pool model and the fork-join model, not to be confused with the UNIX fork system call. The latter creates (spawns) a new thread whenever one is required (that is, threads are created on-demand.) The operating system then schedules the various threads across the available cores. Each of the newly spawned threads is typically considered to be either a detached thread, or a joinable thread. A detached thread executes in the background and terminates when it has completed, without any message to the parent process. Of course, communication to or from such processes can be implemented manually by you, through the available signaling mechanisms, or using global variables. A joinable thread, in contrast, will communicate back to the main thread, at a point set by you. The parent process might have to wait for all joinable threads to return before proceeding with the next execution step. In the fork-join model, individual threads have explicit start and end conditions. There is an overhead associated with managing their creation and destruction and latencies associated with the synchronization point. This means that threads must be sufficiently long-lived to justify these costs. If you know that some execution threads will be repeatedly required to consume input data, you can instead use the workers’ pool threading model. Here, you create a pool of worker threads at the start of the application. The pool can consist of multiple instances of the same algorithm, where the distributor (also called producer or boss) will dispatch the task to the first available worker (consumer) thread. Alternatively, the workers’ pool can contain several different data processing operators and data-items will be tagged to show which worker can consume the data. The number of worker threads can be changed dynamically to handle peaks in the workload. Each worker thread performs a task until it is finished, then interrupts the boss to be assigned another task. Alternatively, the boss can periodically poll workers to see whether one is ready to receive another task. The work queue model is similar. The boss places tasks in a queue, and workers check the queue and take tasks to perform. An additional variant is to have multiple bosses, sharing the pool of workers. The boss threads place tasks onto a queue, from where they are taken by the worker threads. In each of these models, it must be understood that the amount of work to be performed by a thread can be variable and unpredictable. Even for threads that operate on a fixed quantity of data, it can be the case that data dependencies cause different execution times for similar threads. There is always likely to be some synchronization overhead associated with the requirement for a parent thread to wait for all spawned threads to return (in the fork-join model) or for a pool of workers to complete data consumption before execution can be resumed. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-5 Parallelizing Software 19.4 Threading libraries We have looked at how to make our target application capable of concurrent execution. We must now consider actual source code modifications. This is normally done using a threading library, normally utilizing multi-threading support available in the OS. When modifying existing code, you must take care to ensure that all shared resources are protected by proper synchronization. This includes any libraries used by the code, as all libraries are not reentrant. In some cases, there can be separate reentrant libraries for use in multi-threaded applications. A library that is designed to be used in multi-threaded applications is called thread-safe. If a library is not known to be thread-safe, only one thread can be permitted to make calls to the library functions. The most commonly used standard in this area is POSIX threads (Pthreads), a subset of the wider POSIX standard. POSIX (IEEE std. 1003) is the Portable Operating System Interface, a collection of OS interface standards. Its goal is to assure interoperability and portability of code between systems. Pthreads defines a set of API calls for creating and managing threads. Pthreads libraries are available for Linux, Solaris, and Windows. There are several other multi-threading frameworks, such as OpenMP that can simplify multi-threaded development by providing high-level primitives, or even automatic multi-threading. OpenMP is a multi-platform, multi-language API that supports shared memory multi-processing through a set of libraries and compiler directives plus environment variables that affect run-time behavior. Pthreads provides a set of C primitives that enable you to create, manage, and terminate threads and to control thread synchronization and scheduling attributes. Let you examine, in general terms, how you can use Pthreads to build multi-threaded software to run on our SMP system. We’ll deal with the following types: • pthread_t – thread identifier • pthread_mutex_t – mutex • sem_t - semaphore. You must modify your code to include the appropriate header files. #include #include You must also link your code using the pthread library with the switch -lpthread. To create a thread, you must call pthread_create(), a library function that requires four arguments. The first of these is a pointer to a pthread_t, which is where you will store the thread identifier. The second argument is the attribute that can point to a structure that modifies the thread's attributes (for example scheduling priority), or be set to NULL if no special attributes are required. The third argument is the function the new thread will start by executing. The thread will be terminated if this function returns. The fourth argument is a void * pointer supplied to the thread. This can receive a pointer to a variable or data structure containing relevant information to the thread function. A thread can complete either by returning, or calling pthread_exit(). Both will terminate the thread. A thread can be detached, using pthread_detach(). A detached thread will automatically have its associated data structures (but not explicitly allocated data) released on exit. For a thread that has not been detached, this resource cleanup will happen as part of a pthread_join() call from another thread. The library function pthread_join() enables you to make a thread stall and wait for completion of another thread. Take care, as so-called zombie threads can be created by joining a thread that has already completed. It is not possible to join a detached thread (one that has called pthread_detach()). ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-6 Parallelizing Software Mutexes are created with the pthread_mutex_init() function. The functions pthread_mutex_lock() and pthread_mutex_unlock() are used to lock or unlock a mutex. pthread_mutex_lock() blocks the thread until the mutex can be locked. pthread_mutex_trylock() checks whether the mutex can be claimed and returns an error if it cannot, rather than blocking. A mutex can be deleted when no longer required with the pthread_mutex_destroy() function. Semaphores are created in a similar way, using sem_init() – one key difference being that you must specify the initial value of the semaphore. sem_post() and sem_wait() are used to increment and decrement the semaphore. The GNU tools for ARM cores support full thread-local storage using the Native POSIX Thread library (NPTL) that enables efficient use of POSIX threads with the Linux kernel. There is a one-to-one correspondence between threads created with pthread_create() and kernel tasks Example 19-1 provides a simple example of using the Pthreads library. Example 19-1 Pthreads code void *thread(void *vargp); int main(void) { pthread_t tid; pthread_create(&tid, NULL, /* Parallel execution area pthread_join(tid, NULL); return 0; } /* thread routine */ void *thread(void *vargp) { /* Parallel execution area printf(“Hello World from a return NULL; } 19.4.1 thread, NULL); */ */ POSIX thread!\n”); Inter-thread communications Semaphores can be used to signal to another thread. A simple example would be where one thread produces a buffer containing shared data. It could use a semaphore to indicate to another thread that the data can now be processed. For more complex signaling, a message passing protocol might be required. Threads within a process use the same memory space, so an easy way to implement message passing is by posting in a previously agreed-upon mailbox and then incrementing a semaphore. 19.4.2 Threaded performance There are a few general points to consider when writing a multi-threaded application: ARM DEN0013D ID012214 • Each thread has its own stack space and care might be required with the size of this if large numbers of threads are in use. • Multiple threads contending for the same mutex or semaphore creates contention and wasted core cycles. There is a large body of research on programming techniques to reduce this performance loss. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-7 Parallelizing Software • 19.4.3 There is an overhead associated with thread creation. Some applications avoid this by creating a thread pool at startup. These threads are used on demand and then returned to the thread pool for later re-use, rather than being closed completely. Thread affinity Thread affinity refers to the practice of assigning a thread to a particular core or cores. When the scheduler wants to run a particular thread, it will use only the selected core(s) even if others are idle. This can be a problem if too many threads have an affinity set to a specific core. By default, threads are able to run on any core in an SMP system. ARM DS-5 Streamline is able to reveal the affinity of a thread by using a display mode called Core map. This mode can be used to visualize how tasks are divided up by the kernel and shared amongst several cores. See DS-5 Streamline on page 16-4. 19.4.4 Thread safety and reentrancy Functions that can be used concurrently by more than one thread concurrently must be both thread-safe and reentrant. This is particularly important for device drivers and for library functions. For a function to be reentrant, it must fulfill the following conditions: • All data must be supplied by the caller. • The function must not hold static or global data over successive calls. • The function cannot return a pointer to static data. • The function cannot itself call functions that are not reentrant. For a function to be thread-safe, it must protect shared data with locks. This means that the implementation must be changed by adding synchronization blocks to protect concurrent accesses to shared resources, from different threads. Reentrancy is a stronger property, this means that not every thread-safe function is reentrant. There are number of common library functions that are not reentrant. For example, the function ctime() returns a pointer to static data that is over-written on each call. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-8 Parallelizing Software 19.5 Performance issues There are several multi-core specific issues relating to performance of threads: Bandwidth The connection to external memory is shared between all cores within a cluster. The individual cores run at speeds far higher than the external memory and so are potentially limited (in I/O intensive code) by the available bandwidth. Thread dependencies and priority inversion The execution of a higher priority thread can be stalled by a lower priority thread holding a lock to some shared data. Alternatively, an incorrect split in thread functionality can lead to a situation where no benefit is seen because the threads have fully serialized dependencies. Cache contention and false sharing If multiple threads are using data that reside within the same coherent cache lines, there can be cache line migration overhead even if the actual variables are not shared. 19.5.1 Bandwidth concerns Bandwidth issues can be optimized in a number of ways. Clearly, the code itself must be optimized using the techniques described in Chapter 17, to minimize cache misses and therefore reduce the bandwidth utilization. Another option is to pay attention to thread allocation. The kernel scheduler does not pay any attention to data usage by threads; instead it makes use of priority to decide which threads to run. You can provide hints that enable more efficient scheduling through the use of thread affinity. 19.5.2 Thread dependencies In real systems you can have threads with higher or lower priority that both access a shared resource. This gives scope for some potential difficulties. The term starvation is used to describe the situation where a thread is unable to get access to a resource after repeated attempts to claim it. Priority inversion is said to occur when a lower priority task has a lock on a resource that a higher priority requires in order to be able to execute. In other words, a lower priority task prevents the higher priority task from executing. Priority inheritance resolves this by temporarily raising the priority of the task that has the lock to the highest level. This causes that task to execute as quickly as possible and relinquish the shared resource as soon as it can. Operating systems (particularly real time operating systems) have ways to avoid such problems automatically. One method is not to permit lower-priority threads from directly accessing resources required by higher-priority threads, they might have to use a higher-priority proxy thread to perform the operation. A similar approach is to temporarily increase the priority of the low-priority thread while it is holding the critical resource, ensuring that the scheduler will not pre-empt execution of that thread while in the critical selection. A program that relies on threads executing in a particular sequence to work correctly might have a race condition. Single-core real-time systems often implicitly rely on tasks being executed in a priority based order. Tasks will then execute to completion, without pre-emption. Later tasks can rely on earlier tasks having completed. This can cause problems if such software is moved to a multi-core system without careful checking for such assumptions. A lower-priority task can run at the same time as a higher-priority task and the expected execution order of the original single-core system is no longer guaranteed. There are number of ways to resolve this. A simple ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-9 Parallelizing Software approach is to set task affinity to make those tasks run on the same core. This requires little change to the legacy code, but does break the symmetry of the system and remove scope for load balancing. A better approach is to enforce serial execution through the use of the kernel synchronization mechanisms that give you explicit control over the execution flow and better SMP performance, but does require the legacy code to be modified. 19.5.3 Cache thrashing Cortex-A series processors use physically tagged caches that remove the requirement for flushing caches on context switch. In an SMP system, it is possible for tasks to migrate between the different cores in the system. The scheduler starts a task on a core. It runs for a certain period and is then replaced by a different task. When that task is restarted at a later time by the scheduler, this could be on a different core. This means that the task does not get the potential benefit of cache data already being present in the core cache. Memory intensive tasks that quickly fill the data cache might thrash each others’ cached data. This has an impact on both performance (slower execution because of the higher number of cache misses) and system energy usage (because of additional interaction with external memory). Multi-core optimizations for cache line migration mitigate the effects of this. In addition, the OS scheduler can try to reduce the problem by aiming to keep tasks on the same core. As we have seen, you can also do this by setting core affinity to threads and processes. 19.5.4 False sharing This is a problem of systems with shared coherent caches and is effectively a form of involuntary memory contention. It can happen when a core regularly accesses data that is never changed, and shares a cache line with data that will be altered by another core. The MESI protocol can end up migrating data that is not truly shared between different parts of the memory system, costing clock cycles and power. Even though there is no actual coherency to be maintained, the MESI protocol invalidates the cache line, forcing it to be re-loaded on each write. However, the cache-to-cache migration capability of multi-core clusters reduces the overhead. Therefore, you must avoid having cores operating on independent data that is stored within the same cache line and increasing the level of detail for inner loop parallelization. 19.5.5 Deadlock and livelock When writing code that includes critical sections, it is important to be aware of common problems that can break correct execution of the program: • Deadlock is the situation where two (or more) threads are each waiting for another thread to release a resource. Such threads are effectively blocked, waiting for a lock that can never be released. • Livelock occurs when multiple threads are able to execute, without blocking indefinitely (the deadlock case), but the system as a whole is unable to proceed, because of a repeated pattern of resource contention. Both deadlocks and livelocks can be avoided either by correct software design, or by use of lock-free software techniques. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-10 Parallelizing Software 19.6 Synchronization mechanisms in the Linux kernel When porting software from a single core environment to run on multi-core cluster, there can be situations where you must modify code to enforce a particular order of execution or to control parallel access to shared peripherals or global data. The Linux kernel (like other operating systems) provides a number of different synchronization primitives for this purpose. Most such primitives are implemented using the same architectural features as application-level threading libraries like Pthreads. Understanding which of these is best suited for a particular case will give software performance benefits. Serialization and multiple threads contending for a resource can cause suboptimal use of the increased processing throughput provided by the multiple cores. In all cases, minimizing the size of the critical section provides best performance. 19.6.1 Completions Completions are a feature provided by the Linux kernel that can be used to serialize task execution. They provide a lightweight mechanism with limited overhead that essentially provides a flag to signal completion of an event between two tasks. The task that is waiting can sleep until it receives the signal, using wait_for_completion (struct completion *comp) and the task that is sending the signal typically uses either complete (struct completion *comp), that will wake up one waiting process, or complete_all (struct completion *comp) that wakes all processes that are waiting for the event. Kernel version 2.6.11 added support for completions that can time out and for interruptible completions. 19.6.2 Spinlocks A spinlock provides a simple binary locking mechanism, designed for protection of critical sections. It implements a busy-wait loop. A spinlock is a generic synchronization primitive that can be accessed by any number of threads. More than one thread might be spinning for obtaining the lock. However, only one thread can obtain the lock. The waiting task executes spin_lock (spinlock_t *lock) and the signaling task uses spin_unlock(spinlock_t *lock). Spinlocks do not sleep and disable pre-emption. 19.6.3 Semaphores Semaphores are a widely used method to control access to shared resources, and can also be used to achieve serialization of execution. They provide a counting locking mechanism that can cope with multiple threads attempting to lock. They are designed for protection of critical sections and are useful when there is no fixed latency requirement. However, where there is a significant amount of contention for a semaphore, performance will be reduced. The Linux kernel provides a straightforward API with functions down(struct semaphore *sem) and up(struct semaphore *sem); to lower and raise the semaphore. Unlike spinlocks, which spin in a busy wait loop, semaphores have a queue of pending tasks. When a semaphore is locked, the task yields, so that some other task can run. Semaphores can be binary (in which case they are also mutexes) or counting. 19.6.4 Lock-free synchronization The use of lock-free data structures, such as circular buffers, is widespread and can avoid the overheads associated with spinlocks or semaphores. The Linux kernel also provides two synchronization mechanisms that are lock-free, the Read-Copy-Update (RCU) and seqlocks. Neither of these mechanisms is normally used in device drivers. If you have multiple readers and writers to a shared resource, using a mutex might not be very efficient. A mutex would prevent concurrent read access to the shared resource because only a single thread is permitted inside the critical section. Large numbers of readers might delay a ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-11 Parallelizing Software writer from being able to update the shared resource. RCUs can help in the case where the shared resource is mainly accessed by readers. Reader threads execute with little synchronization overhead. A thread that writes the shared resource has a much higher overhead, but is executed relatively infrequently. The writer thread must make a copy of the shared resource (access to shared resources must be done though pointers). When the update is complete, it publishes the new data structure, so that it is visible to all readers. The original copy is preserved until the next context switch on all cores. This guarantees that all current read operations can complete. RCUs are more complex to use than standard mutexes and are typically used only when traditional solutions are not suitable. Examples include shared file buffers or networking routing tables and garbage collection. Seqlocks are also intended to provide quick access to shared resources, without use of a lock. They are optimized for short critical sections. Readers are able to access the shared resource with no overhead, but must explicitly check and re-try if there is a conflict with a write. Writes, of course, still require exclusive access to the shared resource. They were originally developed to handle things like system time, a global variable that can be read by many processes and is written only by a timer-based interrupt (on a frequent basis, obviously) The timer write has a high priority and a hard deadline, in order to be accurate. Using a seqlock instead of a mutex enables many readers to share access, without locking out the writer from accessing the critical section. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-12 Parallelizing Software 19.7 Profiling in SMP systems ARM multi-core processors contain additional performance counter functions, that enable counting of the following SMP cache events: • Coherent linefill missed in all cores. • Coherent linefill hit in other core caches. ARM DS-5 Streamline configures a default set of hardware performance counters that are a best-fit for optimizing applications. See DS-5 Streamline on page 16-4 for more information. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 19-13 Chapter 20 Power Management Many ARM systems are mobile devices, powered by batteries. In such systems, optimization of power usage (in fact, it would be more accurate to look at total energy usage) is a key design constraint. Programmers often spend significant amounts of time trying to save battery life in such systems. Power-saving can also be of concern even in systems that do not use batteries. For example, you might want to minimize energy usage for reduction of electricity costs to the consumer or for environmental reasons. Built into ARM cores are many hardware design methods aimed at reducing power usage. Energy usage can be divided into two components – dynamic and static. Both are important. Static power consumption occurs whenever the core logic or RAM blocks have power applied to them. In general terms, the leakage currents (any current that flows when the ideal current is zero) are proportional to the total silicon area – the bigger the chip, the more the leakage. The proportion of power consumption due to leakage gets significantly higher as you move to more advanced manufacturing processes – they are much worse on fabrication geometries of 130nm and below. Dynamic power consumption occurs because of transistors switching and is a function of the core clock speed and the numbers of transistors that change state per cycle. Clearly, higher clock speeds and more complex cores will consume more power. Power management aware operating systems dynamically change the power states of cores, balancing the available compute capacity to the current workload, while attempting to use the minimum amount of power. Some of these techniques dynamically switch cores on and off, or place them into quiescent states, where they no longer perform computation. This means they consume very little power. The main examples of these techniques are: ARM DEN0013D ID012214 • Idle management on page 20-3. • Hotplug on page 20-6. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 20-1 Power Management • Dynamic Voltage and Frequency Scaling on page 20-7. Performance management for something like a big.LITTLE system intersects directly with power management. Operating System Power Management (OSPM) might have to turn cores and clusters on and off as it transfers computation from a big core to a LITTLE one, or from a LITTLE core to a big one. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 20-2 Power Management 20.1 Idle management When a core is idle the OSPM transitions it into a low power state. Typically, a choice of states is available, with different entry and exit latencies, and different levels of power consumption, associated with each state. The state that is used typically depends on how quickly the core will be required again. The power states that can be used at any one time might also depend on the activity of other components in an SoC, beside the cores. Each state is defined by the set of components that will be clock-gated or power-gated when the state is entered. States are sometimes described as being shallow or deep. The time required to move from a low power state to a running state, known as the wakeup latency, is longer in deeper states. Although idle power management is driven by thread behavior on a core, the OSPM can place the platform into states that affect many other components beyond the core itself. If the last core in a cluster becomes idle, the OSPM can target power states that affect the whole cluster. Equally, if the last core in a SoC becomes idle the OSPM can target power states that affect the whole SoC. The choice is also driven by the usage of other components in the system. A typical example is placing memory in self-refresh when all cores, and any other bus masters, are idle. The OSPM has to provide the necessary power management software infrastructure to determine the correct choice of state. In idle management, when a core or cluster has been placed into a low power state, it can be reactivated at any time by a core wakeup event. That is, an event that can wake up a core from a low power state, such as an interrupt. No explicit command is required by the OSPM to bring the core or cluster back into operation. The OSPM considers the affected core or cores to be available at all times even if they are currently in a low power state. 20.1.1 Power and clocking One way in which you can reduce energy usage is to remove power, which removes both dynamic and static currents (sometimes called power gating) or to stop the clock of the core which removes dynamic power consumption only and can be referred to as clock gating. ARM cores typically support a number of levels of power management, as follows: • Standby. • Power down on page 20-4. For certain operations, there is a requirement to save and restore state before and after removing power and both the time taken to do this and power consumed by this extra work can be an important factor in software selection of the appropriate power management activity. The SoC device that includes the core can have additional low power states, with names such as “STOP” and “Deep sleep.” These refer to the ability for the hardware Phase Locked Loop (PLL) and voltage regulators to be controlled by power management software. 20.1.2 Standby In the standby mode of operation, the core is left powered-up, but most of its clocks are stopped, or clock-gated. This means that almost all parts of the core are in a static state and the only power drawn is because of leakage currents and the clocking of the small amount of logic that looks out for the wake-up condition. This mode is entered using either the WFI (Wait For Interrupt) or WFE (Wait For Event) instructions. ARM recommends the use of a Data Synchronization Barrier (DSB) instruction before WFI or WFE, to ensure that pending memory transactions complete before changing state. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 20-3 Power Management If a debug channel is active, it will remain active. The core stops execution until a wakeup event is detected. The wakeup condition is dependent on the entry instruction. For WFI an interrupt or external debug request will wake the core. For WFE, a number of specified events exist, including another core in the cluster executing the SEV instruction. A request from the Snoop Control Unit (SCU) can also wake up the clock for a cache coherency operation in an multi-core system. This means that the cache of a core that is in standby state will continue to be coherent with caches of other cores. A core reset will always force the core to exit from the standby condition. Various forms of dynamic clock gating can also be implemented in hardware. For example the SCU, GIC, timers, CP15, instruction pipeline or NEON blocks can be automatically clock gated when an idle condition is detected, to save power. Standby mode can be entered and exited quickly (typically in two-clock-cycles). It therefore has an almost negligible affect on the latency and responsiveness of the core. To an operating system managing power, a standby state is mostly indistinguishable from a retention state. The difference is evident to an external debugger, and in hardware implementation, but not evident to the idle management subsystem of an operating system. 20.1.3 Retention The core state, including the debug settings, is preserved in low-power structures, enabling the core to be at least partially turned off. Changing from low- power retention to running operation does not require a reset of the core. The saved core state is restored on changing from low-power retention state to running operation. From an operating system point of view there is no difference between a retention state and standby state, other than method of entry, latency and usage- related constraints. However, from an external debugger point of view the states differ as External Debug Request debug events stay pending and debug registers in the core power domain cannot be accessed. 20.1.4 Power down In this state the core is powered off. Software on the device must save all core state, so that it can be preserved over the power-down. Changing from power-down to running operation must include: • A reset of the core, after the power level has been restored. • Restoring the saved core state. The defining characteristic of power down states is that they are destructive of context. This affects all the components that are switched off in a given state, including the core, and in deeper states other components of the system such as the GIC or platform-specific IP. Depending on how debug and trace power domains are organized, in some power-down states one or both of debug and trace context might be lost. Mechanisms must be provided to enable the operating system to perform the relevant context saving and restoring for each given state. Resumption of execution starts at the reset vector, after which each OS must restore its context. For power down states, the interface requires a return address. This is the address at which the calling OS expects resumption of execution on wakeup at its level of privilege. From a powered down state, the core will restart at the reset vector, typically in secure mode. After initializing, the Secure world must resume the OS that called the power down interface, at the required return address. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 20-4 Power Management 20.1.5 Dormant mode In dormant mode, the core logic is powered down, but the cache RAMs are left powered up. Often the RAMs will be held in a low-power retention state where they hold their contents but are not otherwise functional. This provides a far faster restart than complete shutdown, as live data and code persists in the caches. Again, in a multi-core system, individual cores can be placed in dormant mode. In a multi-core system that permits individual cores within the cluster to go into dormant mode, there is no scope for maintaining coherency while the core has its power removed. Such cores must therefore first isolate themselves from the coherence domain. They will clean all dirty data before doing this and will typically be woken up using another core signaling the external logic to re-apply power. The woken core must then restore the original core state before rejoining the coherency domain. Because the memory state might have changed while the core was in dormant mode, it might have to invalidate the caches anyway. Dormant mode is therefore much more likely to be useful in a single core environment rather than in a cluster. This is because of the additional expense of leaving and rejoining the coherency domain. In a cluster, dormant mode is typically likely to be used only by the last core when the other cores have already been shutdown. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 20-5 Power Management 20.2 Hotplug CPU hotplug is a technique that can dynamically switch cores on or off. Hotplug can be used by the OSPM to change available compute capacity based on current compute requirements. Hotplug is also sometimes used for reliability reasons. There are a number of differences between hotplug and use of a power-down state for idle: 1. When a core is hot unplugged, the supervisory software stops all use of that core in interrupt and thread processing. The core is no longer considered to be available by the calling OS. 2. The OSPM has to issue an explicit command to bring a core back online, that is, hotplug a core. The appropriate supervisory software will only start scheduling on or enabling interrupts to that core after this command. Operating systems typically perform much of the kernel boot process on one primary core, bringing secondary cores online at a later stage. Secondary boot behaves very similarly to hotplugging a core into the system.The operations in both cases are almost identical. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 20-6 Power Management 20.3 Dynamic Voltage and Frequency Scaling Many systems operate under conditions where their workload is very variable. It would be useful to have the ability to reduce or increase the core performance to match the expected core workload. If you could clock the core more slowly when it is less busy, you could save dynamic power consumption. Dynamic Voltage and Frequency Scaling (DVFS) is an energy saving technique that exploits: • The linear relationship between power consumption and operational frequency. • The quadratic relationship between power consumption and operational voltage. This relationship is given as: P = C × V2 × f Where: P Is the dynamic power. C Is the switching capacitance of the logic circuit in question. V Is the operational voltage. f Is the operational frequency. Power savings are achieved by modulating the frequency of a core in proportion to its current load. Some frequencies can be served at lower voltages and a net quadratic power saving is achieved. If the core is running more slowly, it is also the case that its supply voltage can be reduced somewhat. The advantage of reducing supply voltage is that it reduces both dynamic and static power. Compared to the alternative of running fast, then entering standby, then running fast and so forth, the approach of running slowly at a lower supply can save energy. To do this successfully requires two difficult engineering problems to be solved. The SoC requires a way in which software running on the ARM core can reliably modify the clock speed and supply voltage of the core, without causing problems in the system. This requires such things as voltage level-shifters and split power supplies on chip to cope with the variable supply, plus synchronizers between voltage domains to cope with timing changes. Of equal importance is the ability of software running in the system to make accurate predictions about future workloads to set the voltage and clock speed accordingly. There is an implementation-specific relationship between the operational voltage for a given circuit and the range of frequencies that circuit can safely operate at. A given frequency of operation together with its corresponding operational voltage is expressed as a tuple and is known as an Operating Performance Point (OPP). For a given system, the range of attainable OPPs are collectively termed as the system DVFS curve. Operating systems use DVFS to save energy and, where necessary, keep within thermal limits. The load on a core modulates its frequency of operation. The OS provides DVFS policies to manage the power consumed and the required performance. A policy aimed at high-performance selects higher frequencies and uses more energy. A policy aimed at saving energy selects lower frequencies and therefore results in lower performance. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 20-7 Power Management 20.4 Assembly language power instructions ARM assembly language includes instructions that can be used to place the core in a low power state. The architecture defines these instructions as hints – the core is not required to take any specific action when it executes them. In the Cortex-A processor family, however, these instructions are implemented in a way that shuts down the clock to almost all parts of the core. This means that the power consumption of the core is significantly reduced – only static leakage currents will be drawn, and there will be no dynamic power consumption. The WFI instruction has the effect of suspending execution until the core is woken up by one of the following conditions: • An IRQ interrupt, even if the CPSR I-bit is set. • An FIQ interrupt, even if the CPSR F-bit is set. • An asynchronous abort. • A Debug Entry request, even if JTAG Debug is disabled. In the event of the core being woken by an interrupt when the relevant CPSR interrupt flag is disabled, the core will implement the next instruction after WFI. On older versions of the ARM architecture, the wait for interrupt function (also called standby mode) was accessed using a CP15 operation, rather than a dedicated instruction. The WFI instruction is widely used in systems that are battery powered. For example, mobile telephones can place the core in standby mode many times a second, while waiting for you to press a button. WFE is similar to WFI. It suspends execution until an event occurs. This can be one the events listed, or an additional possibility – an event signaled by another core in a cluster. Other cores can signal events by executing the SEV instruction. SEV signals an event to all cores in a cluster. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 20-8 Power Management 20.5 Power State Coordination Interface The ARM architecture provides a convenient method of partitioning a software stack through its architectural definition of privilege levels, and through the Security and Virtualization Extensions. This introduces different modes of execution in the architecture, which in turn provide a way to partition the systems that compose the software stack used on a device. As we have seen, the ARMv7 architecture has the following privilege levels in which software from different software vendors can operate: • PL0 - For User Application vendors, such as apps downloaded from an App Store. • PL1 - Rich OS vendors, such as the Linux Kernel used by Android. • PL2 - Hypervisor Vendors. • Secure PL0 - Trusted OS applications, from Trusted OS vendors. • Secure PL1 - Trusted OS. • Secure PL1 - OEMs providing secure firmware. As operating systems from different vendors can be simultaneously executing in an ARM system, it is necessary to have a method of collaboration when performing power control. This means that, if the operating system that is managing power, be it at supervisor (PL1) or hypervisor level (PL2), wants to enter an idle state, power up/down a core, or perform a big.LITTLE migration, operating systems at other levels of privilege must react to this request. Equally, if a core is roused from a power state by a wake up event, it might be necessary for operating systems running at different levels of privilege to perform actions such as restoring state. There are no freely available interfaces that allow interoperation, and integration amongst the various operating systems. This presents difficulties for the OS vendors. ARM provides a software interface, the Power State Coordination Interface (PSCI) so that the OSPM can place a core into a low power state when it has no work for it. Using this interface operating systems and firmware can implement power management techniques such as idle, hotplug and state migration on big.LITTLE systems. The messages sent using this interface are received by all relevant levels of execution. That is, if Virtualization and Security Extensions are implemented, a message sent by a Rich OS must be received by a hypervisor. If the latter sends it, the message must be received by the secure firmware that then coordinates with a Trusted OS. This enables each operating system to determine whether context saving is required. PSCI specifies the following functions: CPU_SUSPEND Suspend the execution on a core. This call is intended for use in idle subsystems where the core is expected to return to execution through a wake up event. CPU_OFF Power down a core. This call is intended for use in hotplug. A core that is powered down by CPU_OFF can only be powered up again by a CPU_ON. CPU_ON Power up a core. This call is used to power up cores that either: • Have not yet been booted into the calling OS. • Have been previously powered down with a CPU_OFF call. MIGRATE This is used to request the Trusted OS of a single core to migrate its context to a specific core. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 20-9 Chapter 21 Security The term security is used in the context of computer systems to cover a wide variety of features. In this chapter, we will use a narrower definition. A secure system is one that protects assets (resources that require protecting, for example passwords, or credit card details) from a range of plausible attacks to prevent them from being copied or damaged or made unavailable (denial of service). Confidentiality is a key security concern for assets such as passwords and cryptographic keys. Defense against modification and proof of authenticity is vital for security software and on-chip secrets used for security. Examples of secure systems might include entry of Personal Identification Numbers (PINs) for such things as mobile payments, digital rights management, and e-Ticketing. Security is harder to achieve in the world of open systems where a wide range of software can be downloaded onto a platform. This gives the potential for malevolent or untrusted code to tamper with the system. ARM processors include specific hardware extensions to enable construction of secure systems. Creating secure systems is outside the scope of this book. In the remainder of this chapter, we present the basic concepts behind the ARM Security Extensions (TrustZone). If your system is one that makes use of these extensions, be aware that this imposes some restrictions on the operating system and on unprivileged code (in other words, code that is not part of the secure system). TrustZone is of little or no use without memory system support. It must, of course, be emphasized, that no security is absolute! ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 21-1 Security 21.1 TrustZone hardware architecture The TrustZone hardware architecture aims to provide resources that enable a system designer to build secure systems. It does this through a range of components and infrastructure additions. Low-level programmers must have some awareness of the restrictions placed on the system by the TrustZone architecture, even if they are not intending to make use of the security features. In essence, system security is achieved by dividing all of the device’s hardware and software resources, so that they exist in either the Secure world for the security subsystem, or the Normal world for everything else. System hardware ensures that no Secure world resources can be accessed from the Normal world. A secure design places all sensitive resources in the Secure world, and has robust software running that can protect assets against a wide range of possible attacks. The use of the term Non-Secure is used in the ARM Architecture Reference Manual as a contrast to Secure state, but this does not imply that there is a security vulnerability associated with this state. We will refer to this as Normal operation here. The use of the word world is to emphasize the relationship between the Secure world and other states the device is capable of. The additions to the architecture enable a single physical core to execute code from both the Normal world and the Secure world in a time-sliced fashion. The memory system is similarly divided. An additional bit, indicating whether the access is Secure or Non-Secure (the NS bit) is added to all memory system transactions, including cache tags and access to system memory and peripherals. This can be considered as an additional address bit, giving a 32-bit physical address space for the Secure world and a completely separate 32-bit physical address space for the Normal world. Normal World Secure World SMC Secure Monitor Mode Platform OS RFE SVC Privileged Mode User Mode SMC RFE Secure Kernel Privileged Mode SVC User Mode Application Code Secure Service Figure 21-1 Switching between Normal and Secure worlds As the cores execute code from the two worlds in a time-sliced fashion, context switching between them is done using an additional core mode (like the existing modes for IRQ, FIQ etc.) called Monitor mode. A limited set of mechanisms by which the core can enter Monitor mode from the Normal world is provided. Entry to monitor can be through a dedicated instruction, the Secure Monitor Call (SMC) instruction, or by hardware exception mechanisms. IRQ, FIQ and ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 21-2 Security external aborts can all be configured to cause the core to switch into Monitor mode. In each case, this will appear as an exception to be dealt with by the Monitor mode exception handler. Figure 21-1 on page 21-2 provides a conceptual summary of this switching. Figure 21-2 shows how, in many systems, FIQ is reserved for use by the secure world (it becomes, in effect, a non-maskable secure interrupt). An IRQ that occurs when in the Normal world is handled in the normal way, described in the chapters on exception handling. An FIQ that occurs while executing in the Normal world is vectored directly to Monitor mode. Monitor mode handles the transition to Secure world and transfers directly to the Secure world FIQ handler. If the FIQ occurs when in the Secure world, it is handled through the Secure vector table and routed directly to the Secure world handler. IRQs are typically disabled during execution in the Secure world. Monitor FIQ IRQ SVC SVC FIQ FIQ IRQ IRQ Undef Undef Abort Abort System System Privileged Modes Privileged Modes FIQ (Secure Interrupt) IRQ (Normal Interrupt) FIQ User User Normal Secure Figure 21-2 Normal and Secure worlds The software handler for Monitor mode is implementation specific, but will typically save the state of the current world and restore the state of the world being switched to, much like a normal context switch. The NS bit in the Secure Configuration Register (SCR) in CP15 indicates which world the core is currently in. In Monitor mode, the core is always executing in the Secure world, regardless of the value of the SCR NS-bit that is used to signal which world you were previously in. The NS-bit also enables code running in Monitor mode to snoop security banked registers, to see what is in either world. TrustZone hardware also effectively provides two virtual MMUs, one for each virtual core. This enables each world to have a local set of translation tables, with the Secure world mappings hidden and protected from the Normal world. The translation table descriptions include a NS bit that is used to determine whether accesses are made to the secure or non-secure physical address space. Although the translation table entry bit is still present, the Normal virtual core hardware ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 21-3 Security does not use this field, and memory accesses are always made with NS = 1. The Secure virtual core can therefore access either Secure or Normal memory. Cache and TLB hardware permits Normal and Secure entries to co-exist. It is good practice for code that modifies translation table entries and that does not care about TrustZone based security, to always set the translation table NS-bit to zero. This means that it will be equally applicable when the code is executing in the Secure or Normal worlds. The ability to direct aborts, IRQ and FIQ directly to the monitor, enables trusted software to route the interrupt request accordingly, enabling a design to provide secure interrupt sources immune from manipulation by the Normal world software. Similarly, the Monitor mode routing means that from the point of view of Normal world code, an interrupt that occurs during Secure world execution appears to occur in the last Normal world instruction that occurred before the Secure world was entered. A typical implementation is to use FIQ for the Secure world and IRQ for the Normal world. Exceptions are configured to be taken by the current world (whether Secure or Normal), or to cause an entry to the monitor. The monitor has its own vector table. Because of this, the core has three sets of exception vector tables. It has a table for the Normal world, one for the Secure world, and one for Monitor mode. The hardware must also provide the illusion of two separate cores within CP15. Sensitive configuration CP15 registers can only be written by Secure world software. Other settings are normally banked in the hardware, or by the Monitor mode software, so that each world sees its own version. Implementations that use TrustZone will typically have a light-weight kernel (Trusted Execution Environment) that hosts services (for example, encryption) in the Secure world. A full OS runs in the Normal world and is able to access the secure services using the SMC instruction. In this way, the Normal world gets access to functions of the service, without any ability to see keys or other protected data. 21.1.1 Multi-core systems with security extensions Each core in a multi-core system has the programmer’s model features described for single cores earlier in this book. Any number of the cores in the cluster can be in the Secure world at any point in time, and cores are able to transition between the worlds independently of each other. The Snoop Control Unit is aware of security settings. Additional registers are provided to control whether Normal world code can modify SCU settings. Similarly, the generic interrupt controller that distributes prioritized interrupts across the Multi-core cluster must also be modified to be aware of security concerns. Theoretically, the Secure world OS on an SMP system could be as complicated as the Normal world OS. However, this is undesirable when aiming for security. In general, it is expected that a Secure world OS will actually only execute on one core of an SMP system (with security requests from the other cores being routed to this chosen core). This does provide some bottleneck issues. To some extent these will be balanced by the Normal world OS performing load balancing against the core that it will see as busy for unknown reasons. Beyond that this limitation has to be seen as one of the compromises that can be reached to hit a particular target level of security. 21.1.2 Interaction of Normal and Secure worlds If you are writing code in a system that contains some secure services, it can be useful to understand how these are used. As we have seen, a typical system will have a light-weight kernel, Trusted Execution Environment (TEE) hosting services (for example, encryption) in the ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 21-4 Security Secure world. This interacts with a full OS in the Normal world that can access the secure services using the SMC call. In this way, the Normal world is able to have access to functions of the service, without getting to see keys (for example). Generally applications developers won’t directly interact with TrustZone (or TEEs or Trusted Services). Instead, one makes use of a high level API (for example, it might be called reqPayment()) provided by a Normal world library. The library would be provided by the same vendor as the Trusted Service (for example, a credit card company), and would handle the low level interactions. Figure 21-3 shows this interaction and illustrates the flow from user application calling the API that makes an appropriate OS call, which then passes to the TrustZone driver code, and then passes execution into the TEE through the Secure monitor. Normal Secure Application Vendor Specific Library Operating System Trusted Services Trustzone Driver Trusted Execution Environment Secure Monitor Figure 21-3 Interaction with TrustZone It is common to share data between the Secure and Normal worlds. For example, in the Secure world you might have a signature checker. The Normal world can request that the Secure world verifies the signature of a downloaded update, using the SMC call. The Secure world requires access to the memory used by the Normal world to store the package. The Secure world can use the NS-bit in its translation table descriptors to ensure that it used non-secure accesses to read the data. This is important because data relating to the package might already be in the caches, because of the accesses done by the Normal world. These accesses with addresses marked as non-secure. As mentioned previously the security attribute can be thought of as an additional address bit. If the core used secure access to try to read the package, it would not hit on data already in the cache. If you are a Normal world programmer, in general, you can ignore something happening in the Secure world, as its operation is hidden from you. One side effect is that interrupt latency can increase slightly, if an interrupt goes off in the Secure world, but this increase is small compared to the overall latency on a typical OS. If you do have to access a secure application, you will require a driver-like function to talk to the Secure world OS and Secure applications, but the details of creating that Secure world OS and applications are beyond the scope of this book. Those writing code for the Normal world only require the particular protocol for the secure application being called. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 21-5 Security Finally, the TrustZone System also controls availability of debug provision. Separate hardware over full JTAG debug and trace control is separately configurable for Normal and Secure software worlds, so that no information about the Secure system leaks. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 21-6 Chapter 22 Virtualization Modern compute subsystems are powerful but often under utilized. The domains that these systems address increasingly require multiple software environments working simultaneously on the same physical processor systems. Software applications and their environments might have to be completely separated for reasons of robustness, differing requirements for real-time behavior, or to have them isolated from each other. This is done by providing virtual cores for the software to execute on. Implementing such virtual cores in an efficient fashion requires both dedicated hardware extensions (to accelerate switching between virtual machines) and hypervisor software. The volume of legacy software to be ported to new hardware might make it more expedient to provide a virtual machine matching the legacy OS requirements than to port the OS and drivers to new hardware and OS versions. The Virtualization Extensions to the ARMv7 architecture provide a standard hardware accelerated implementation enabling the creation of high performance hypervisors. Writing a hypervisor is outside the scope of this book, as such software is typically produced by specialized teams. In this chapter, we will provide an overview of how the Virtualization Extensions work and the effects of this on OS and application code, in a fashion similar to our previous discussion of the TrustZone Security Extensions. The term hypervisor actually dates back to the 1960s and was first used by IBM to describe software running on mainframes. Today’s hypervisors are conceptually very similar and might be thought of as operating at a level higher than the supervisor or operating system. They typically permit multiple instances of one or more different operating systems, called Guest Operating Systems (Guest OS), to run on the system. The hypervisor provides a virtual system to each of the Guest operating systems and monitors their execution. The Guest OS does not typically need to be aware that it is running under a hypervisor or that other operating systems might be using the system. The term Virtual Machine Monitor (VMM) is sometimes used to describe a hypervisor. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-1 Virtualization There are two types of hypervisors being deployed on ARM. Type1 is as described with each Virtual Machine (VM) containing Guest OS. In Type 2, the hypervisor is an extension of the Host OS with each subsequent Guest OS contained in a separate VM. One possible use of virtual machine hypervisors is in so-called cloud computing solutions, where software can be partitioned into client or server devices or where large amounts of data or content exist. This scenario is likely to increase the amount of addressable physical memory required in the system. For this reason, the Virtualization Extensions require that the core also implement the Large Physical Address Extension (LPAE) described on page 22-10. This enables each of the multiple software environments, as well as different processes within, to access separate windows of physical addresses. LPAE provides an additional level of MMU translation table, so that each 32-bit virtual memory address can be mapped within a 40-bit physical memory range. In this scenario, this permits software to allocate enough physical memory to each virtual machine, even when total demands on memory exceed the range of 32-bit addressing. It would also be possible for a single operating system kernel to handle up to 40 bits of physical address space, with up to 4GB available at any given time. In theory, this can mean up to ~3GB per process. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-2 Virtualization 22.1 ARMv7-A Virtualization Extensions For the most part, this book assumes that a system is owned and managed by a single privileged Operating System that deploys many unprivileged applications. Most main stream Operating Systems are also built on this assumption. Virtualization is the concept where more than one Operating System is enabled to co-exist and operate on a same system. The ARM Virtualization Extensions make it possible to operate multiple Operating Systems on the same system, while offering each such Operating System an illusion of sole ownership of the system by virtue of introduction of new architectural features. These are: • A hypervisor mode, in addition to the existing privileged modes. This PL2 mode is even more privileged than PL1 modes. Hyp mode is expected to be occupied by Hypervisor software managing multiple guest operating systems occupying PL1 and PL0 modes. Hyp mode only exists in the Normal (Non-secure) world. • An additional memory translation, called Stage 2 is introduced. Previously, the Virtual Address (VA) is translated to Physical Address (PA) by the PL1 and PL0 MMU described in Chapter 9. This translation is now known as Stage 1 and the old Physical Address is now called Intermediate Physical Address (IPA). The IPA is subjected to another level of translation in Stage 2 to get the final PA corresponding to the VA. • Interrupts can be configured to be taken (routed) to the Hypervisor. The hypervisor will then take care of delivering interrupts to the appropriate guest. • A Hypervisor Call instruction (HVC) for guests to request Hypervisor services. ARM Virtualization Extensions aim to run conventional Operating Systems as guests on a virtualized system with no or little modification. 22.1.1 Types of virtualization Virtualization solutions can be broadly classified as bare metal or hosted according to their design. Regardless of the classification, the functional role of a Hypervisor remains the same, arbitration of platform resources, and seamless operation of individual guests with minimal porting effort and run time sacrifices. In bare metal virtualization, the Hypervisor is the most privileged non-secure software to boot on the core. It initializes various aspects of PL2 for housekeeping, for example its own translation tables for PL2 translation regime. The Hypervisor will then initialize settings for each guest operating system that it will launch. This will involve setting up stage 2 translation tables for individual guests, interrupt delivery mechanisms etc., and goes on to launching them. The Hypervisor will schedule guest operating systems on available cores, handle exceptions on behalf of the guests, and deliver interrupts to appropriate guests at run time. Figure 22-1 on page 22-4 contrasts conventional systems with systems deploying bare-metal virtualization. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-3 Virtualization System without virtualization System with virtualization Virtual Machine 1 App 1 App 2 Operating system App 1 Virtual Machine 1 App 2 App 1 Guest OS 1 App 2 Guest OS 1 Hypervisor Hardware Hardware Figure 22-1 Bare metal virtualization 22.1.2 Memory translation A number of memory translation regimes are possible in a virtualized system in the Normal world. A translation regime is a broad term encompassing the privileges and execution mode of the core and the set of translation tables used. These translations are carried out using the MMUs and the Translation Table structures created by the software that controls the translation. A stage of translation comprises of a set of translation tables that translates an input address to an output address. The input and output addresses take different names depending on the stage of translation. PL1&0 Stage 1 This is the translation regime that an Operation System will setup and control, exactly as described in Chapter 9 The Memory Management Unit. This translation regime is applied when the core is in one of execution modes that falls under PL1 (Kernel) and PL0 (application) privilege levels. On a conventional system, this regime is used to translate virtual addresses to physical addresses. In a virtualized system however, this physical address is treated as an Intermediate Physical Address (IPA) because it is subjected to another stage (Stage 2) of translation. In the presence of an oncoming Stage 2, this translation regime qualifies as Stage 1. IPAs cannot be used to address system memory. Although called an intermediate physical address, from the perspective of a guest, the output of this translation regime is what it sees and uses as physical address. PL1&0 Stage 2 This comprises of a set of translation tables that the Hypervisor sets up for each of the guests it manages. This regime will translate the IPA that was output by Stage 1, and translates it to a physical address that can finally be used to address system memory. The Virtualization Extensions add a set of core registers to control the Stage 2 translation tables. The hypervisor saves and restores these registers whenever it schedules a different guest on the core. Individual guests that are managed by the Hypervisor have no control over, nor are aware of the presence of a Stage 2 translation. When virtualization is in effect, all PL1&0 Stage 1 translations are implicitly subjected to this Stage 2 translation. It is also important to note that a Stage 2 translation is applied even if Stage 1 translation is turned off by the guest, As a result, Stage 2 translation is always applied when virtualization is in effect. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-4 Virtualization Because it is managed by the Hypervisor, exceptions arising from Stage 2 translation are taken in Hyp mode so that Hypervisor can respond to them. Exceptions from Stage 1, as before, are taken to PL1 Abort mode for the respective guest to handle. What Stage 2 translation achieves is virtualization of the guest's view of physical memory. Every virtual address used in PL1&0 stage is first translated to an IPA by Stage 1. This IPA is translated to the actual physical address by Stage 2. The guest is unaware of, and cannot control, translation by Stage 2. By appropriately setting up the Stage 2 translation tables the hypervisor can manage and allocate physical memory to guests. This can be thought of a natural extension of what an operating system does, managing and allocating physical memory to its applications. In that respect, the guests could be considered as applications of the Hypervisor. PL2 ARM DEN0013D ID012214 This comprises of a set of translation tables that the Hypervisor sets up for itself to manage its own virtual memory. This regime will translate virtual addresses used in Hyp mode to Physical Addresses. No additional stages are applied to this translation, therefore it is not qualified by stages. The Virtualization Extension includes a set of registers for the Hypervisor to manage its own translation tables, much like an operating system kernel manages its own translation tables Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-5 Virtualization 22.2 Hypervisor software This section lists some of the functions that a Hypervisor performs. As we have seen, these functions do not depend on the type of virtualization solution deployed. 22.2.1 Memory management The hypervisor is responsible for memory management for itself and for the guest operating systems it manages. The entire physical memory is at the direct disposal of the Hypervisor. The PL2 MMU is used in Hyp mode to translate the virtual addresses that the Hypervisor uses to address physical memory. Apart from setting up and managing its own translation tables, a Hypervisor has to create and manage Stage 2 translation tables for each of its guests. In addition to both Secure and Non Secure versions of the TTBR, there are also a Virtualization Translation Table Base Register (VTTBR) and Hyp Translation Table Base Register (HTTBR) where the VTTBR is used to point to second stage translation tables (See Large Physical Address Extensions on page 22-10) and the HTTBR is used to point to tables used for Hypervisor memory mapping. Secure functions are the same as the Non-secure TTBR and can use the short-descriptor format, but VTTBR and HTTBR are long descriptor only. Stage 2 translation tables setup by the Hypervisor translate intermediate physical memory addresses to physical memory addresses. Any aborts resulting from attempting to translate addresses in Stage 2 are taken in Hyp mode. The Hypervisor is responsible for receiving the aborts, and handling them appropriately. For intended and legitimate faults, the Hypervisor might take remedial measures such as emulating a device or allocating more memory to a guest operating systems. For unexpected faults the Hypervisor can choose to terminate the guest operating system or report aborts to the guest in turn. 22.2.2 Device emulation Platform devices are memory-mapped, and guest accesses to devices are subject to at least Stage 2 translation when virtualization is in effect. When Virtualization is in effect, there are use cases for a Hypervisor to emulate a device in software, or to hide it from the guests. Device emulation is necessary where more than one guest is aware of a platform device (using the physical address), and attempt to access it. Because of the shared nature, guests cannot be permitted direct accesses to it without arbitration. In such cases the Hypervisor can prohibit access to the said device region to all such guests, by means of their respective stage 2 translation table descriptors. Alternatively, the hypervisor might choose to hide a device from a chosen set of guests either because the device is not actually present or has already been assigned to a different guest. (Guests would typically detect a platform device by reading its ID register.) By employing the same trapping mechanism as before, the Hypervisor can return dummy values for guest reads and ignore writes, effectively giving the guest the impression that the device does not exist on the platform. One of the important jobs of the Hypervisor is to schedule guests on available cores in the system. The task is accomplished in much the same way that an operating system schedules different tasks. In a virtualized system the Hypervisor schedules a guest operating system on a core to execute. A scheduled guest assumes sole ownership of the core, and schedules its tasks on the core, as it would do on a conventional system. Guests will not be aware of the fact that a Hypervisor is functioning and that they are being constantly scheduled in and out of the core. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-6 Virtualization 22.2.3 Device assignment Device emulation is necessary but turns out to be expensive as all accesses to the device by the guest have to be trapped and emulated in software. The Hypervisor has the option of assigning individual devices to individual guests so that the guest can own and operate the device without requiring Hypervisor arbitration. The challenge is to hide from the guest the fact that the device is actually located at a different physical address, and generates a different interrupt ID than that which the guest is expecting. Transparent Stage 2 mappings, and interrupt virtualization can circumvent these challenges. 22.2.4 Exception handling Asynchronous exceptions (IRQ, FIQ and asynchronous aborts) can be routed to Hypervisor mode. This behavior can be selected through dedicated control bits in the Hyp Configuration Register (HCR). An additional bit in this register also enables synchronous exceptions (Undef, SVC or precise aborts) to be routed to the Hypervisor Trap entry. The hypervisor can also be called through the HVC instruction. If this is performed while in privilege level PL1, it causes a hypervisor trap exception and entry into Hypervisor mode. If performed when already in Hypervisor mode, it causes an HVC exception. The Virtualization Extensions introduce the concept of virtual exceptions. The exceptions we have met before are real events occurring in the system. Virtual exceptions are events of similar nature, that is, interrupts and aborts, but are manufactured by the Hypervisor, either in response to an actual physical exception, or manufactured (as a result of Device emulation) with no relation to a physical exception. These are taken in their conventional exception mode, for example, a virtual abort is taken in Abort mode. Physical exceptions can be configured to be consumed by the Hypervisor, only virtual exceptions are delivered to guests. This concept is useful only when the corresponding physical exception is routed to Hyp mode. This means that the hypervisor has control of masking and generating a virtual exception. When a real physical exception occurs, and is routed to the hypervisor software, the Hypervisor then signals a virtual exception to the current Guest OS. The Guest OS handles the exception as it would do for an equivalent physical exception and need not be aware that the hypervisor has been involved. The virtual exception is signalled through the virtual GIC, or by using dedicated HCR bits. When exceptions are routed to the hypervisor in this fashion, the CPSR A, I and F bits no longer mask physical exceptions, instead they mask the handling of virtual exceptions within the Guest OS. 22.2.5 Interrupt handling When virtualization is in effect, interrupts reaching a core can be targeted to a guest that is running on a different core, or even to a guest that does not currently occupy a core at all (suspended because of inactivity). In a virtualized system the hypervisor must be the primary arbitrator, and configures the core so that interrupts delivered to a core are handled in Hyp mode. This enables the Hypervisor to be in charge of receiving and delivering interrupts to guests. The Hypervisor Configuration Register (HCR) in the core, that is accessible to guests, has a set of three bits corresponding to IRQ, FIQ and abort virtual exceptions. When any of these bits are set while the core is in PL1&0 modes, This has the same effect as the core taking the corresponding exception in a conventional system, that is, they will not be taken in Hypervisor mode again. Physical exceptions take the core to Hyp mode. If, after handling the exception, the hypervisor chooses to deliver the exception to the guest, it sets the required bits and resumes the guest, the guest, being in one of PL1&0 modes, will take the exception as if it were running standalone. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-7 Virtualization 22.2.6 Scheduling One of the important jobs of the hypervisor is to schedule guests on available cores in the system. This task is accomplished in much the same way that an operating system schedules different tasks. In a virtualized system the hypervisor schedules a guest on a core to execute. A scheduled guest assumes sole ownership of the core, and schedules its tasks on the core, as it would do on a conventional system. Guests are not aware of the fact that a Hypervisor is functioning beneath and is being constantly scheduled in and out of the core. 22.2.7 Context switch When the Hypervisor schedules another guest on a core, it must perform a context switch, that is, save the context of the currently running guest to memory, and then restore the context of the new guest from memory. The goal is to recreate the environment for the new guest on the current core before it resumes, creating the illusion of uninterrupted execution. By performing a context switch, the Hypervisor ensures that the execution environment follows the guest, and offers the illusion of a virtual core that the guest always occupies. The following elements of a guest context must be saved and restored: • The general purpose registers of the core including the banked registers of all modes, • System register contents for such things as memory management and access control. • The pending and active states of private interrupts on the core. • In case of guests using core private timers, the timer registers must be saved and restored so that they generate interrupts at the expected intervals. The physical memory assigned to a guest, that it sees as RAM, stays in place and does not have to be saved or restored. By using two stages of memory translation, the physical memory that the guest uses stays private and distinct from any others, even for identical guests. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-8 Virtualization 22.3 Relationship between virtualization and ARM Security Extensions We discussed the concept of privilege levels, PL0, PL1 and PL2, in Chapter 3 ARM Processor Modes and Registers. Figure 22-2 illustrates how in the Normal world, you have PL0 (User mode), PL1 (for exception modes) and PL2 for the hypervisor, while in the Secure world, you have only PL0 and PL1, with secure monitor mode at PL1. Normal World PL0 PL1 Secure World User User SVC SVC Abort Abort IRQ IRQ FIQ FIQ Undef Undef System System PL0 PL1 Monitor PL2 HYP Figure 22-2 Privilege levels and security In general, the hypervisor has no impact on the Secure world. It is unable to access secure parts of memory or otherwise interfere in the operations of the secure system. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-9 Virtualization 22.4 Large Physical Address Extensions Processors that implement the ARMv7-A Large Physical Address Extension (LPAE), expand the range of accessible physical addresses from 4GB (232 bytes) to 1024GB (240 bytes) a terabyte, by translating 32-bit virtual memory addresses into 40-bit physical memory addresses. To do this they use the Long-descriptor format. The existing short-descriptor format translation tables are still supported, as are the security extensions described in Chapter 21. The Virtualization Extensions provide an additional second stage of address translation when running virtual machines. The first stage of this translation produces an Intermediate Physical Address (IPA) and the second stage then produces the physical address. The second stage of this conversion process is controlled by the Hypervisor, TLB entries can also have an associated Virtual Machine ID (VMID), in addition to an ASID. Again, it is possible to disable the stage 2 MMU and have a flat mapping from IPA to PA. Virtual memory map Under control of guest OS Physical memory map as seen by guest, controlled by Hypervisor Real physical memory map Peripherals Peripherals Peripherals OS FLASH RAM Peripherals Translation tables Translation tables RAM RAM FLASH Application TTBR{n} VTTBR Figure 22-3 Stage 2 translation Long-descriptor format memory management includes the following features: • 64-bit page descriptors. • Up to three levels of translation tables. • Supports specifying up to 40-bit physical addresses. • 1GB, 2MB and 4KB block or page sizes are supported. • Optional second stage memory translation used in virtualization • An additional access permission setting – Privileged eXecute Never (PXN). This marks a page as containing code that can be executed only in a non-privileged (user) mode. This setting is also added to the legacy descriptor format. There is also a privileged execute setting (PX) that means that code can be executed only in privileged mode. There are certain similarities between the short-descriptor format and the long-descriptor format: ARM DEN0013D ID012214 • Both have an input 32-bit virtual address from the processor. • Both use TTBR0/TTBR1 to point to level 1 translation tables (with TTBCR selecting which part of the address space is covered by each). Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-10 Virtualization • Long descriptors also provides a way to indicate “Outer-shareable” or “Inner-shareable” as opposed to “Shareable”. Table descriptor 63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32 Attributes SBZ Ignored 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 Ignored Next level table address out 1 1 Block or page descriptor 63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32 Ignored SBZ Attr First level PA 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 2nd level PA 3rd level PA Lower Attributes 1 1 Figure 22-4 Format of long-descriptor table entries 22.4.1 Long descriptor translation tables Long descriptor translation tables output a 40-bit intermediate physical address. • The first level translation table is 4 entries – one entry for each 1 GB of virtual memory — • • First level page descriptors contain either: — The upper 10 bits of physical address for that 1 GB of virtual memory — A pointer to the second level translation table Second level translation table is 512 entries – one entry for each 2MB of virtual memory (in the 1GB address range of the first level table entry). — • • Indexed by bits [29:21] of the VA. Second level page descriptors contain: — The upper 19 bits of physical address for that 2MB of virtual memory — A pointer to the third level translation table Third level translation table is 512 entries – one entry for each 4KB of virtual memory (in the 2MB address range of the second level table entry) — • Indexed by [31;30] of the VA. Indexed by bits [20:12] of the VA. Third level page descriptors contain: — Upper 28 bits of physical address for that 4KB of virtual memory. This process is shown in Figure 22-5 on page 22-12. The translation table walking steps are similar to those previously described for processors that do not implement LPAE. Individual translation table entries are now 64-bit in size. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-11 Virtualization 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 Bits [29:21] of VA Bits [20:12] of VA 2nd level table index 3rd level table index 2-bit table index 9-bit table index Bits [11:0] of VA Page offset address 9-bit table index 12-bit address offset 1st level table 4 entries 4KB 2nd level table 512 entries 4KB 3rd level table IPA 512 entries Figure 22-5 Long -descriptor table walk Recall that long-descriptor translation tables output a 40-bit physical address. Therefore, the physical base address given by this translation will be of the form [39:0], with bit [39] of the descriptor producing bit [39] of the physical address. It is not necessary to map all of the physical address space, but all of the 4GB virtual address space has to be taken care of. There must be a descriptor for each virtual address – although it might be a “fault” descriptor. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 22-12 Chapter 23 big.LITTLE Modern software stacks place conflicting requirements on mobile systems. On the one hand is a demand for very high performance for tasks such as games, while on the other is a continuing requirement to be frugal with energy reserves for low intensity applications like audio playback. Traditionally, it has not been possible to have a single processor design that can be capable of both high peak performance as well as high energy efficiency. This meant that a lot of energy was wasted because the high performance core would be used for low intensity tasks leading to reduced battery life. Performance would itself be affected by the thermal limits at which the cores could run for sustained periods. big.LITTLE technology from ARM solves this problem by coupling together an energy efficient “LITTLE” core with a high performance “big” core. big.LITTLE is an example of a heterogeneous processing system. Such systems typically include several different processor types with different microarchitectures, like general purpose processors and specialized ASICs. big.LITTLE takes the heterogeneity one step further in that it includes general purpose processors that are different in their micro-architecture but compatible in their instruction set architecture. A term that is often used with such systems is Heterogeneous Multi-processing (HMP). What makes HMP different from AMP is that all the processors in an HMP system are fully coherent and run the same operating system image. The basic premise is that software can run on big or LITTLE processors depending on performance requirements. When peak performance is required software can run on big processors. At most other times, software can run on LITTLE processors. Through this combination, big.LITTLE provides a solution that is capable of delivering the high peak performance demanded by the latest mobile devices, within the thermal bounds of the system, with maximum energy efficiency. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 23-1 big.LITTLE 23.1 Structure of a big.LITTLE system Both types of core in a big.LITTLE system are fully cache coherent and share the same instruction set architecture (ISA). The same application binary will run unmodified on either. Differences in the internal microarchitecture of the processors enable them to provide the different power and performance characteristics that are fundamental to the big.LITTLE concept. These are typically managed by the operating system. big.LITTLE software models require transparent and efficient transfer of data between big and LITTLE clusters. Hardware coherency enables this, transparently to the software, such as the ARM CoreLink CCI-400 described in The Cache Coherent Interface (CCI) on page 18-11. Without hardware coherency, the transfer of data between big and LITTLE cores would always occur through main memory - this would be slow and not power efficient. In addition, it would require complex cache management software, to enable data coherency between big and LITTLE clusters. In addition, such a system also requires a shared interrupt controller, such as the GIC-400, enabling interrupts to be migrated between any cores in the clusters. All cores can signal each other using distributed interrupt controllers such as the CoreLink GIC-400. Task switching is handled entirely within the OS scheduler, and is invisible to the application software. A typical system is shown in Figure 23-1. GIC-400 Generic Interrupt Controller Cortex-A7 core Cortex-A7 core Cortex-A15 core Cortex-A15 core L2 Cache L2 Cache Cortex-A7 Cluster Cortex-A15 Cluster IO coherent master CCI-400 Cache Coherent Interconnect Memory controller ports System ports Figure 23-1 Typical big.LITTLE system 23.1.1 big.LITTLE configurations A number of big.LITTLE configurations are possible, Figure 23-1 uses a Cortex-A7 processor as the LITTLE cluster and a Cortex-A15 processor as the big cluster, though other configurations are possible. The LITTLE cluster is capable of handling most low intensity tasks such as audio playback, web-page scrolling, operating system events, and other always on, always connected tasks. As such, it is likely that the LITTLE cluster is where the software stack will remain until intensive tasks such as gaming or video processing are run. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 23-2 big.LITTLE The big cluster can be utilized for heavy workloads such as high performance graphics. A coupling of these two cluster types provides opportunities to save energy as well as satisfy the increasing performance demands of applications stacks in mobile devices. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 23-3 big.LITTLE 23.2 Software execution models in big.LITTLE There are two primary execution models for big.LITTLE: • Migration. • Global Task Scheduling. Migration models are a natural extension to power-performance management techniques such as DVFS, (see Dynamic Voltage and Frequency Scaling on page 20-7). The Migration model has two types: • Cluster migration. • CPU migration. A migration action is similar to a DVFS operating point transition. Operating points on the DVFS curve of a core will be traversed in response to load variations. When the current core (or cluster) has attained the highest operating point, if the software stack requires more performance, a core (or cluster) migration action is effected (see Figure 23-2). Execution then continues on the other core (or cluster) with the operating points on this core (or cluster) being traversed. When performance is not required, execution can switch back. Figure 23-2 Typical Cortex-A15 and Cortex-A7 DVFS curves The following sections describe the execution models: ARM DEN0013D ID012214 • Cluster migration on page 23-5. • CPU migration on page 23-5. • Global Task Scheduling on page 23-5. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 23-4 big.LITTLE 23.2.1 Cluster migration Only one cluster, either big or LITTLE, is active at any one time, except very briefly during a cluster context switch to the other cluster. To achieve the best power and performance efficiency, the software stack runs mostly on the energy-efficient LITTLE cluster and only runs for short time periods on the big cluster. This model requires the same number of cores in both clusters. This model does not cope well with unbalanced software workloads, that is, workloads that place significantly different loads on cores within a cluster. In such situations, cluster migration will result in a complete switch to the big cluster even though not all the cores need that level of performance. For this reason Cluster migration is less popular than other methods. 23.2.2 CPU migration In this model, each big core is paired with a LITTLE core. Only one core in each pair is active at any one time, with the inactive core being powered down. The active core in the pair is chosen according to current load conditions. Using the example in Figure 23-3, the operating system sees four logical cores. Each logical core can physically be a big or LITTLE core. This choice is driven by Dynamic Voltage and Frequency Scaling (DVFS). This model requires the same number of cores in both the clusters. Low load big cluster CPU0 CPU1 CPU2 CPU3 High load big cluster LITTLE cluster CPU0 CPU1 CPU2 CPU3 CPU0 CPU2 Inactive core CPU1 LITTLE cluster CPU0 CPU1 CPU2 CPU3 CPU3 Active core Figure 23-3 CPU migration The system actively monitors the load on each core. High load causes the execution context to be moved to the big core, and conversely, when the load is low, the execution is moved to the LITTLE core. Only one core in the pairing can be active at any time. When the load is moved from an outbound core to an inbound core, the former is switched off. This model allows a mix of big and LITTLE cores to be active at any one time, and support the use of asymmetric topologies, that is, an unequal number of big and LITTLE cores. The In Kernel Switcher (IKS) solution from Linaro is an example of this model. 23.2.3 Global Task Scheduling Through the development of big.LITTLE technology, ARM has evolved the software models starting with various migration models through to Global Task Scheduling (GTS) that forms the basis for all future development in big.LITTLE technology. The ARM implementation of GTS is called big.LITTLE Multi-processing (MP). ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 23-5 big.LITTLE In this model the operating system task scheduler is aware of the differences in compute capacity between big and LITTLE cores. Using statistical data, the scheduler tracks the performance requirement for each individual software thread, and uses that information to decide which type of core to use for each. Unused cores can be powered off. If all cores in a cluster are off, the cluster itself can be powered off. This model can work on a big.LITTLE system with any number of cores in any cluster. This is shown in Figure 23-4. This approach has a number of advantages over the migration models. The system can have different numbers of big and LITTLE cores. • Any number of cores can be active at any one time. When peak performance is required the system can deploy all cores. • It is possible to isolate the big cluster for the exclusive use of intensive threads, while light threads run on the LITTLE cluster. This enables heavy compute tasks to complete faster, as there are no additional background threads. • It is possible to target interrupts individually to big or LITTLE cores. Core 0 Core 0 Core 1 Core 2 Core 3 Linux kernel Scheduler Core 1 Core 2 Core 3 Figure 23-4 Global Task Scheduling ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 23-6 big.LITTLE 23.3 big.LITTLE MP For big.LITTLE MP the fundamental requirement is for the scheduler to decide when a software thread can run on a LITTLE core or a big core. The scheduler does this by comparing the tracked load of software threads against tunable load thresholds, an up migration threshold and a down migration threshold as shown in Figure 23-5. Task eligible for up migration Up migration threshold Down migration threshold Task eligible for down migration Task tracked load Task state Figure 23-5 Migration thresholds When the tracked load average of a thread currently allocated to a LITTLE core exceeds the up migration threshold, the thread is considered eligible for migration to a big core. Conversely, when the load average of a thread that is currently allocated to a big core drops below the down migration threshold, it is considered eligible for migration to a LITTLE core. In big.LITTLE MP these basic rules govern task migration between big and LITTLE cores. Within the clusters, standard Linux scheduler load balancing applies. This tries to keep the load balanced across all the cores in one cluster. The model is refined by adjusting the tracked load metric based on the current frequency of a core. A task that is running when the core is running at half speed, will accrue tracked load at half the rate that it would if the core was running at full speed. This enables big.LITTLE MP and DVFS management to work together in harmony. big.LITTLE MP uses a number of mechanisms to determine when to migrate a task between big and LITTLE cores: 23.3.1 Fork migration This operates when the fork system call is used to create a new software thread. At this point, clearly no historical load information is available. The system defaults to a big core for new threads on the assumption that a ‘light’ thread will quickly migrate down to a LITTLE core as a result of Wake migration. Fork migration benefits demanding tasks without being expensive. Threads that are low intensity and persistent, such as Android system services, will only be moved to big cores at creation time, quickly moving to more suitable LITTLE cores thereafter. Threads that are clearly demanding throughout, will not be penalized by being made to launch on LITTLE cores first. Threads that are episodic but tend to require performance on the whole will benefit from being launched on the big cluster and will continue to run there as required. 23.3.2 Wake migration When a task that was previously idle becomes ready to run, the scheduler must decide which cluster will execute the task. To choose between big and LITTLE, big.LITTLE MP uses the tracked load history of a task. Generally, the assumption is that the task will resume on the same ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 23-7 big.LITTLE cluster as before. The load metric is not updated for a task that is sleeping. Therefore, when scheduler checks the load metric of a task at wake up, before choosing a cluster to execute it on, the metric will have the value it had when the task last ran. This property means that tasks that a periodically busy will always tend to wake up on a big core. A task has to actually modify its behavior to change cluster. Up migration uses previous value of tracked load Up migration threshold Wake migration (up) Down migration threshold Task state Core residency B L Wakeup on big core Figure 23-6 Wake migration on a big core Up migration threshold Down migration uses previo value of tracked load Down migration threshold Wake m (dow Task state Core residency B L Wake up on LITTLE core Figure 23-7 Wake migration on a LITTLE core If a task modifies its behavior, and the load metric has crossed either of the up or down migration thresholds, the task can be allocated to a different cluster. Figure 23-6 and Figure 23-7 illustrate this process. Rules are defined that ensure that big cores generally only run a single intensive thread and run it to completion, so upward migration only occurs to big cores which are idle. When migrating downwards, this rule does not apply and multiple software threads can be allocated to a little core. 23.3.3 Forced migration Forced migration deals with the problem of long running software threads that do not sleep, or do not sleep very often. The scheduler periodically checks the current thread running on each LITTLE core. If the tracked load exceeds the up migration threshold the task is transferred to a big core, as in Figure 23-8 on page 23-9. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 23-8 big.LITTLE Forced migration Up migration threshold Down migration threshold Task state Core residency B L Task migrated to big multiprocessor Figure 23-8 Forced migration 23.3.4 Idle pull migration Idle pull migration is designed to make best use of active big cores. When a big core has no task to run, a check is made on all LITTLE cores to see if a currently running task on a LITTLE core has a higher load metric that the up migration threshold. Such a task can then be immediately migrated to the idle big core. If no suitable task is found, then the big core can be powered down. This technique ensures that big cores, when they are running, always take the most intensive tasks in a system and run them to completion. 23.3.5 Offload migration Offload migration requires that normal scheduler load balancing be disabled. The downside of this is that long-running threads can concentrate on the big cores, leaving the LITTLE cores idle and under-utilized. Overall system performance, in this situation, can clearly be improved by utilizing all the cores. Offload migration works to periodically migrate threads downwards to LITTLE cores to make use of unused compute capacity. Threads that are migrated downwards in this way remain candidates for up migration if they exceed the threshold at the next scheduling opportunity. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 23-9 big.LITTLE 23.4 Using big.LITTLE For a representative sample of common use cases, a big.LITTLE system running the ARM big.LITTLE MP solution shows significant power savings when compared to a system composed of only Cortex-A15 processors. Figure 23-9 Typical big.LITTLE MP power savings compared to the Cortex-A15 shows how big.LITTLE MP benefits benchmarks. The comparison is between a big.LITTLE system composed of four LITTLE cores and four big cores and a system composed only of four big cores. Figure 23-10 big.LITTLE benchmark improvements The software thread affinity management techniques discussed earlier result in substantial performance gains for threaded benchmarks where the number of threads is greater than four. In this situation, on the system under test big.LITTLE MP enables the use of more cores to aid ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 23-10 big.LITTLE the benchmark. Offload migration assists with spreading the number of compute intensive benchmark threads to the LITTLE cores when the big cores are busy or overloaded. Idle-pull migration results in the best utilisation of the big processors which effectively work as accelerators. For benchmarks with fewer threads, using big.LITTLE MP either provides no degradation or a small improvement. Compared to the test system with only four big cores, dynamic software thread affinity management will promote better utilisation of the big cores which will not be encumbered with low intensity and frequent running threads (such as system services) or interrupts. ARM big.LITTLE MP technology has been tested with Android on multiple silicon implementations. The code is self-contained and available as a drop-in into the software stack without significant modification or tuning. The only requirement is that the platform board-support package is well tuned in terms of DVFS and idle power management, allowing the scheduler extensions to focus on getting the job done. The big.LITTLE MP scheduler extensions are available in two forms: • As a part of monthly Linaro Stable Kernel (LSK) releases. These releases contain a complete Android software stack based on a very recent Linux Stable Kernel. The stack is available in source form and also as a pre-built binary set complete with boot firmware, boot loaders, ramdisk images and an Android root filesystem image. See https://releases.linaro.org Select the latest release and android/vexpress-lsk for details on the LSK. • As an isolated patch set against the LSK kernel. See https://wiki.linaro.org/ARM/VersatileExpress Select Attachments and then chose get for the latest lsk.tar.bz2 version of attachment:big-LITTLE-MP-scheduler-patchset. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 23-11 Chapter 24 Debug Debugging is a key part of software development and is often considered to be the most time consuming (and therefore expensive) part of the process. Bugs can be difficult to detect, reproduce and fix and it can be difficult to predict how long it will take to resolve a defect. The cost of resolving problems grows significantly when the product is delivered to a customer. In many cases, when a product has a small time window for sales, if the product is late, it can miss the market opportunity. Therefore, the debug facilities provided by a system are a vital consideration for any developer. Many embedded systems using ARM processors have limited input/output facilities. This means that traditional desktop debug methods (such as use of printf()) might not be appropriate. In such systems in the past, developers might have used expensive hardware tools like logic analyzers or oscilloscopes to observe the behavior of programs. The processors described in this book have caches and are part of a complex system-on-chip containing memory and many other blocks. There might be no processor signals that are visible off-chip and therefore no ability to monitor behavior by connecting up a logic analyzer (or similar). For this reason, ARM systems typically include dedicated hardware to provide wide-ranging control and observation facilities for debug. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-1 Debug 24.1 ARM debug hardware Cortex-A series processors provide hardware features that enable debug tools to provide significant levels of control over core activity and to non-invasively collect large amounts of data about program execution. We can sub-divide the hardware features into two broad classes, invasive and non-invasive. Invasive debug provides facilities that enable you to stop programs and step through them line by line (either at the C source level, or stepping through assembly language instructions). This can be by means of an external device that connects to the core using the chip JTAG pins, or (less commonly) by means of debug monitor code in system ROM. JTAG stands for Joint Test Action Group and refers to the IEEE-1149.1 specification, originally designed to standardize testing of electronic devices on boards, but now widely re-used for core debug connection. A JTAG connection typically has five pins – two inputs, plus a clock, a reset and an output. The debugger gives the ability to control execution of the program, enabling you to run code to a certain point, halt the core, step through code and resume execution. We can set breakpoints on specific instructions (causing the debugger to take control when the core reaches that instruction). These work using one of two different methods. Software breakpoints work by replacing the instruction with the opcode of the BKPT instruction. Obviously, these can only be used on code that is stored in RAM, but have the advantage that they can be used in large numbers. The debug software must keep track of where it has placed software breakpoints and what opcodes were originally located at those addresses, so that it can put the correct code back when you want to execute the breakpointed instruction. Hardware breakpoints use comparators built into the core and stop execution when execution reaches the specified address. These can be used anywhere in memory, as they do not require changes to code, but the hardware provides limited numbers of hardware breakpoint units (typically four in the Cortex-A family). Debug tools can support more complex breakpoints (for example stopping on any instruction in a range of addresses, or only when a specific sequence of events occurs or hardware is in a specific state). Data watchpoints give debugger control when a particular data address or address range is read or written. These can also be called data breakpoints. On hitting a breakpoint, or when single-stepping, you can inspect and change the contents of ARM registers and memory. A special case of changing memory is code download. Debug tools typically enable you to change your code, recompile and then download the new image to the system. 24.1.1 Single stepping Single step refers to the ability of the debugger to move through a piece of code, one instruction at a time. The difference between Step-In and Step-Over can be explained with reference to a function call. If you Step-Over the function call, the entire function is executed as one step, enabling you to continue after a function that you do not want to step through. Step-in would mean that you single step through the function instead. 24.1.2 Debug events A debug event is some part of the process being debugged that causes the system to notify the debugger. Debug events can be synchronous or asynchronous. Breakpoints, the BKPT instruction, and Watchpoints are all synchronous debug events. When any of these events occur, the core can respond in one of a number of ways: ARM DEN0013D ID012214 • It can ignore the debug event. • It can takes a debug exception. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-2 Debug • It will enter one of two debug modes, depending on the setup of the Debug Status and Control Register (DSCR): — Monitor debug mode. — Halt Debug mode. Both of these are examples of invasive debug. Halt debug mode In Halt Debug mode, a debug event causes the core to enter Debug state. The core is halted and isolated from the rest of the system. This means that the debugger displays memory as seen by the core, and the effects of memory management and cache operations will become visible. In Debug state, the core stops executing instructions from the location indicated by the program counter, and is instead controlled through the external debug interface, in particular using the Debug Instruction Transfer Register (DBGITR). This enables an external agent, such as a debugger, to interrogate core context and control all subsequent instruction execution. Both the core and system state can be modified. Because the core is stopped, no interrupts will be handled until execution is restarted by the debugger. Monitor debug-mode In Monitor debug-mode, a debug event causes a debug exception to occur, either related to the instruction execution that generates a Prefetch Abort exception, or a data access that generates a Data Abort exception. Both of these must be handled by the software debug monitor. Since the core is still operating, interrupts can still be serviced. 24.1.3 Semihosting debug Semihosting is a mechanism that enables code running on an ARM target to use the facilities provided on a host computer running a debugger. Examples of this might include keyboard input, screen output, and disk I/O. For example, you might use this mechanism to enable C library functions, such as printf() and scanf(), to use the screen and keyboard of the host. Development hardware often does not have a full range of input and output facilities, but semihosting enables the host computer to provide these facilities. Semihosting is implemented by a set of defined software instructions that generate an exception. The application invokes the appropriate semihosting call and the debug agent then handles the exception. The debug agent provides the required communication with the host. The semihosting interface is common across all debug agents provided by ARM. Tools from ARM use SVC 0x123456 (ARM state) or SVC 0xAB (Thumb) to represent semi-hosting debug functions. Of course, outside of the development environment, a debugger running on a host is not normally connected to the system. It is therefore necessary for the developer to re-target any C library functions that use semi-hosting, for example, by using fputc(). This would involve replacing the library code that used an SVC call with code that could output a character. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-3 Debug 24.2 ARM trace hardware Non-invasive debug, often called trace in ARM documentation, enables observation of the core behavior while it is executing. It is possible to record memory accesses performed (including address and data values) and generate a real-time trace of the program, seeing peripheral accesses, stack and heap accesses and changes to variables. For many real-time systems, it is not possible to use invasive debug methods. Consider, for example, an engine management system – while you can stop the core at a particular point, the engine will keep moving and you will not be able to do useful debug. Even in systems with less onerous real-time requirements, trace can be very useful. Trace is typically provided by an external hardware block connected to the core. This is known as an Embedded Trace Macrocell (ETM) or Program Trace Macrocell (PTM) and is an optional part of an ARM processor based system. System-on-chip designers can omit this block from their silicon to reduce costs. These blocks observe, but do not affect core behavior and are able to monitor instruction execution and data accesses. There are two main problems with capturing trace. The first is that with current very high core clock speeds, even a few seconds of operation can mean trillions of cycles of execution. Clearly, to make sense of this volume of information would be extremely difficult. The second, related problem is that current cores can potentially perform one or more 64-bit cache accesses per cycle, and to record both the data address and data values can require a large bandwidth. This presents a problem in that typically, only a few pins might be provided on the chip and these outputs can be switched at significantly lower rates than the core can be clocked. If the core generates 100 bits of information every cycle at a speed of 1GHz, but the chip can only output four bits of trace at a speed of 200MHz, then there is a problem. To solve this latter problem, the trace macrocell will try to compress information to reduce the bandwidth required. However, the main method to deal with these issues is to control the trace block so that only selected trace information is gathered. For example, you might trace only execution, without recording data values, or you might trace only data accesses to a particular peripheral or during execution of a particular function. In addition, it is common to store trace information in an on-chip memory buffer (the Embedded Trace Buffer (ETB)). This alleviates the problem of getting information off-chip at speed, but has an additional cost in terms of silicon area (and therefore price of the chip) and also provides a fixed limit on the amount of trace that can be captured. The ETB stores the compressed trace information in a circular fashion, continuously capturing trace information until stopped. The size of the ETB varies between chip implementations, but a buffer of 8 or 16KB is typically enough to hold a few thousand lines of program trace. When a program fails, if the trace buffer is enabled, you can see a portion of program history. With this program history, it is easier to walk back through your program to see what happened before the point of failure. This is particularly useful for investigating intermittent and real-time failures that can be difficult to identify through traditional debug methods that require stopping and starting the core. The use of hardware tracing can significantly reduce the amount of time required to find these failures, as the trace shows exactly what was executed, what the timing was and what data accesses occurred. 24.2.1 CoreSight The ARM CoreSight™ technology expands on the capabilities provided by the ETM. Again its presence and capabilities in a particular system are defined by the system designer. CoreSight provides a number of extremely powerful debug facilities. It enables debug of multi-core systems (both asymmetric and SMP) that can share debug access and trace pins, with full control ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-4 Debug of which cores are being traced at which times. The embedded cross trigger mechanism enables tools to control multiple cores in a synchronized fashion, so that, for example when one core hits a breakpoint, all of the other cores will also be stopped. Commercial debug tools can use trace data to provide features such as real-time views of registers, memory and peripherals, enabling you to step forward and backward through the program execution. Profiling tools can use the data to show where the program is spending its time and what performance bottlenecks exist. Code coverage tools can use trace data to provide call graph exploration. Operating system aware debuggers can make use of trace, and in some cases additional code instrumentation. to provide high level system context information. A brief description of some of the available CoreSight components follows: Debug Access Port (DAP) The DAP is an optional part of an ARM CoreSight system. Not every device will contain a DAP. It enables an external debugger to directly access the memory space of the system without having to put the core into debug state. To read or write memory without a DAP might require the debugger to stop the core and have it execute Load or Store instructions. The DAP gives an external debug tool access to all of the JTAG scan chains in a system and therefore to all debug and trace configuration registers of the available cores. Embedded Cross Trigger (ECT) The ECT block is a CoreSight component that can be included within in a CoreSight system. Its purpose is to link together the debug capabilities of multiple devices in the system. For example, you can have two cores that run independently of each other. When you set a breakpoint on a program running on one core, it would be useful to be able to specify that when that core stops at the breakpoint, the other one must also be stopped (regardless of the instruction it is currently executing). The Cross Trigger Matrix and Interface within the ECT enable debug status and control information to be propagated between cores and trace macrocells. AHB Trace Macrocell The AMBA AHB Trace Macrocell enables the debugger to have visibility of what is happening on the system memory bus. This information is not directly obtainable from the core ETM, as the integer core is unable to determine whether data comes from a cache or external memory. CoreSight Serial Wire CoreSight Serial Wire Debug gives a 2-pin connection using a Debug Access Port (DAP) that is equivalent in function to a 5-pin JTAG interface. System Trace Macrocell (STM) This provides a way for multiple cores (and processes) to perform printf() style debugging. Software running on any master in the system is able to access STM channels without having to be aware of usage by others, using very simple fragments of code. This enables timestamped software instrumentation of both kernel and user space code. The timestamp information gives a delta with respect to previous events and can be extremely useful. Trace Memory Controller (TMC) As already described, adding additional pins to a packaged IC can significantly increase its cost. In situations where you have multiple cores (or other blocks capable of generating trace information) on a single ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-5 Debug device, it is likely that economics preclude the possibility of providing multiple trace ports. The CoreSight Trace Memory Controller can be used to combine multiple trace sources into a single bus. Controls are provided to enable prioritize and select between these multiple input sources. The trace information can be exported off-chip using a dedicated trace port, through the JTAG or serial wire interface or by re-using I/O ports of the SoC. Trace information can be stored in an ETB or in system memory. You must consult documentation specific to the device you are using to determine what trace capabilities are present and the tools available to make use of them. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-6 Debug 24.3 Debug monitor We have seen how the ARM architecture provides a wide range of features accessible to an external debugger. Many of these facilities can also be used by code running on the core – a so called debug monitor that is resident on the target system. Monitor systems can be inexpensive, as they might not require any additional hardware. However, they take up memory space in the system and can only be used if the target system itself is actually running. They are of little value on a system that does not at least boot correctly. The breakpoint and watchpoint hardware facilities of the core are available to a debug monitor. When Monitor mode debug is selected, breakpoint units can be programmed by code running on the ARM processor. If a BKPT instruction is executed, or a hardware breakpoint unit matches, the system behaves differently in Monitor mode. Instead of stopping the core under control of an external hardware debugger, the core instead takes an abort exception and this can recognize that the abort was generated by a debug event and call the Monitor code. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-7 Debug 24.4 Debugging Linux applications Linux is a multi-tasking operating system in which each process has its own process address space, complete with private translation table mappings. This can make debug of some kinds of problems quite tricky. We can broadly define two different debug approaches used in Linux systems. • Linux applications are typically debugged using a GDB debug server running on the target, communicating with a host computer, usually through Ethernet. The kernel continues to operate normally while the debug session takes place. This method of debug does not provide access to the built-in hardware debug facilities. The target system is permanently in a running state. The server receives a connection request from the host debugger and then receives commands and provides data back to the host. The host debugger sends a load request to the GDB server, which responds by starting a new process to run the application being debugged. Before execution begins, it uses the system call ptrace() to control the application process. All signals from this process are forwarded to the GDB server. Any signals sent to the application will go instead to the GDB server that can deal with the signal or forward it to the application being debugged. To set a breakpoint, the GDB server inserts code that generates the SIGTRAP signal at the required location in the code. When this is executed, the GDB server is called and can then perform classic debugger tasks such as examining call stack information, variables or register contents. • For kernel debug, a JTAG-based debugger is used. The system is halted when a breakpoint is executed. This is the easiest way to examine problems such as device driver loading or incorrect operation or the kernel boot failure. Another common method is through printk() function calls. The strace tool shows information about user system calls. Kgdb is a source-level debugger for the Linux kernel that works with GDB on a separate machine and enables inspection of stack traces and view of kernel state (such as PC value, timer contents, and memory). The device /dev/kmem enables run-time access to the kernel memory. Of course, a Linux-aware JTAG debugger can be used to debug threads. It is usually possible only to halt all processes; one cannot halt an individual thread or process and leave others running. A breakpoint can be set either for all threads, or it can be set only on a specific thread. As the memory map depends on which process is active, software breakpoints can usually only be set when a particular process is mapped in. The ARM DS-5 Debugger is able to debug Linux applications using gdbserver and Linux kernel and Linux kernel modules using JTAG. The debug and trace features of DS-5 are described in the next section. 24.4.1 The call stack Application code uses the call stack to pass parameters, store local data and store return addresses. The data each function pushes on the stack is organized into a stack frame. When a debugger stops a core, it might be able to analyze the data on the stack to provide you with a call stack, that is, a list of function calls leading up to the current situation. This can be extremely useful when debugging, as it enables you to determine why the application has reached a particular state. In order to reconstruct the call stack, the debugger must be able to determine which entries on the stack contain return address information. This information might be contained in debugger information (DWARF debug tables) if the code was built with these included, or by following ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-8 Debug a chain of frame pointers pushed on the stack by the application. To do this, the code must be built to use frame pointers. If neither of these types of information are present, the call stack can not be constructed. In multi-threaded applications, each thread has its own stack. The call stack information will therefore only relate to the particular thread being examined. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-9 Debug 24.5 DS-5 debug and trace DS-5 Debugger provides a powerful tool for debugging applications on both hardware targets and models using ARM architecture-based processors. You can have complete control over the flow of the execution so that you can quickly isolate and correct errors. DS-5 Debugger provides a wide range of debug features such as: • Loading images and symbols. • Running images. • Breakpoints and watchpoints. • Source and instruction level stepping. • Controlling variables and register values. • Viewing the call stack. • Support for handling exceptions and Linux signals. • Debug of multi-threaded Linux and Android applications. • Debug of Linux kernel and Android modules, boot code and kernel porting. • Application rewind, that allows you to debug backwards as well as forwards through Linux and Android applications. The debugger supports a comprehensive set of DS-5 Debugger commands that can be executed in the Eclipse IDE, script files, or a command-line console. In addition, there is a small subset of CMM-style commands sufficient for running target initialization scripts. DS-5 Debugger supports bare-metal debug using JTAG, Linux application debug using gdbserver, Linux kernel debug using JTAG, and Linux kernel module debug using JTAG. Debug and trace support for bare-metal SMP systems, including cross-triggering and core-dependent views and breakpoints, PTM trace, and up to 4 GB trace with DSTREAM. This support is described in the following sections. In addition, DS-5 Debugger supports ARM CoreSight ETM, PTM and STM, to provide non-intrusive program trace that enables you to review instructions (and the associated source code) as they have occurred. It also provides the ability to debug time-sensitive issues that would otherwise not be picked up with conventional intrusive stepping techniques. The DS-5 Debugger currently uses DSTREAM to capture trace on the ETB. 24.5.1 Debugging Linux or Android applications using DS-5 DS-5 Debugger takes care of downloading and connecting to the debug server. Developers must specify the platform and the IP address. This reduces a complex task using several applications and a terminal to a couple of steps in the IDE. To debug a Linux or Android application you can use a TCP or serial connection: • To gdbserver running on a model that is pre-configured to boot Linux or Android. • To gdbserver running on a hardware target. This type of development requires gdbserver to be installed and running on the target. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-10 Debug 24.5.2 Debugging Linux kernel modules Linux kernel modules provide a way to extend the functionality of the kernel, and are typically used for things such as device and filesystem drivers. Modules can either be built into the kernel or can be compiled as a loadable module and then dynamically inserted and removed from a running kernel during development without having to frequently recompile the kernel. However, some modules must be built into the kernel and are not suitable for loading dynamically. An example of a built-in module is one that is required during kernel boot and must be available prior to the root filesystem being mounted. You can use DS-5 Debugger to set source-level breakpoints in a module provided that the debug information is loaded into the debugger. Attempts to set a breakpoint in a module before it is inserted into the kernel results in the breakpoint being pended. When debugging a module, you must ensure that the module on your target is the same as that on your host. The code layout must be identical, but the module on your target does not have to contain debug information. Built in module To debug a module that has been built into the kernel using DS-5 Debugger, the procedure is the same as for debugging the kernel itself: 1. Compile the kernel together with the module. 2. Load the kernel image on to the target. 3. Load the related kernel image with debug information into the debugger. 4. Debug the module as you would for any other kernel code. Loadable module The procedure for debugging a loadable kernel module is more complex. From a Linux terminal shell you can use the insmod and rmmod commands to insert and remove a module. Debug information for both the kernel and the loadable module must be loaded into the debugger. When you insert and remove a module, DS-5 Debugger automatically resolves memory locations for debug information and existing breakpoints. To do this, DS-5 Debugger intercepts calls within the kernel to insert and remove modules. This introduces a small delay for each action while the debugger stops the kernel to interrogate various data structures. 24.5.3 Debugging Linux kernels using DS-5 To debug a Linux kernel module you can use a debug hardware agent connected to the host workstation and the running target. 24.5.4 Debugging a multi-threaded applications using DS-5 DS-5 Debugger tracks the current thread using the debugger variable, $thread. You can use this variable in print commands or in expressions. Threads are displayed in the Debug Control view with a unique ID that is used by the debugger and a unique ID from the Operating System (OS). For example: Thread 1 (OS ID 1036) where Thread 1 is the ID used by the debugger and OS ID 1036 is the ID from the OS. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-11 Debug A separate call stack is maintained for each thread and the selected stack frame is shown in bold text. All the views in the DS-5 Debug perspective are associated with the selected stack frame and are updated when you select another frame. Figure 24-1 Threading call stacks in the DS-5 Debug Control view 24.5.5 Debugging shared libraries Shared libraries enable parts of your application to be dynamically loaded at runtime. You must ensure that the shared libraries on your target are the same as those on your host. The code layout must be identical, but the shared libraries on your target do not have to contain debug information. You can set standard execution breakpoints in a shared library but not until it is loaded by the application and the debug information is loaded into the debugger. Pending breakpoints however, enable you to set execution breakpoints in a shared library before it is loaded by the application. When a new shared library is loaded DS-5 Debugger re-evaluates all pending breakpoints, those with addresses that it can resolve, are set as standard execution breakpoints. Unresolved addresses remain as pending breakpoints. The debugger automatically changes any breakpoints in a shared library to a pending breakpoint when the library is unloaded by your application. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-12 Debug Figure 24-2 Adding shared libraries for debug using DS-5 24.5.6 Trace support in DS-5 DS-5 enables you to perform trace on your application or system. You can capture in real-time a historical, non-intrusive trace of instructions. Tracing is a powerful tool that enables you to investigate problems while the system runs at full speed. These problems can be intermittent, and are difficult to identify through traditional debugging methods that require starting and stopping the core. Tracing is also useful when trying to identify potential bottlenecks or to improve performance-critical areas of your application. Before the debugger can trace function executions in your application you must ensure that: • You have a debug hardware agent, for example, an ARM DSTREAM unit with a connection to a trace stream. • The debugger is connected to the debug hardware agent. Trace view When the trace has been captured the debugger extracts the information from the trace stream and decompresses it to provide a full disassembly, with symbols, of the executed code. This view shows a graphical navigation chart that displays function executions with a navigational timeline. In addition, the disassembly trace shows function calls with associated addresses and if selected, instructions. Clicking on a specific time in the chart synchronizes the disassembly view. In the left-hand column of the chart, percentages are shown for each function of the total trace. For example, if a total of 1000 instructions are executed and 300 of these instructions are associated with myFunction() then this function is displayed with 30%. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-13 Debug In the navigational timeline, the color coding is a “heat” map showing the executed instructions and the amount of instructions each function executes in each timeline. The darker red color showing more instructions and the lighter yellow color showing less instructions. At a scale of 1:1 however, the color scheme changes to display memory access instructions as a darker red color, branch instructions as a medium orange color, and all the other instructions as a lighter green color. Figure 24-3 DS-5 Debugger Trace view Trace-based profiling Based on trace data received from a trace buffer such as the ETB, DS-5 Debugger can generate timeline charts with information to help developers to quickly understand how their software executes on the target and which functions are using the core the most. The timeline offers various zoom levels, and can display a heat-map based on the number of instructions per time unit or, at its highest resolution, provide per-instruction visualization color-coded by the typical latency of each group of instructions. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential 24-14 Appendix A Instruction Summary A summary of the instructions available in ARM/Thumb assembly language is given in this Appendix. For most instructions, additional explanation can be found in Chapter 5. The optional condition code field (denoted by cond) is described in Conditional execution on page 5-3. The format of the flexible operand2 used by data processing operations is described in Operand 2 and the barrel shifter on page 5-7, while the various addressing mode options for loads and stores is given in Addressing modes on page 5-13. This Appendix is intended for quick reference. If more detail about the precise operation of an instruction is required, refer to the ARM Architecture Reference Manual, or to the official ARM documentation (for example the ARM Compiler Toolchain Assembler Reference) that can be found at http://infocenter.arm.com/help/index.jsp. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-1 Instruction Summary A.1 Instruction Summary Instructions are listed in alphabetic order, with a description of the syntax, operands and behavior of the instruction. Not all usage restrictions are documented here, nor do we show the associated binary encoding or any detail of changes associated with older architecture versions. A.1.1 ADC ADC (Add with Carry) adds together the values in Rn and Operand2, with the carry flag. Syntax ADC{S}{cond} {Rd}, Rn, where: S (if specified) means that the condition code flags will be updated depending on the result of the instruction. cond is an optional condition code. See Conditional execution on page 5-3. Rd is the destination register. Rn is the register holding the first operand. Operand2 is a flexible second operand. See Operand 2 and the barrel shifter on page 5-7. A.1.2 ADD ADD adds together the values in Rn and Operand2 (or Rn and imm12). Syntax ADD{S}{cond} {Rd,} Rn, ADD{cond} {Rd,} Rn, #imm12 (Only available in Thumb) where: S (if specified) means that the condition code flags will be updated depending on the result of the instruction. cond is an optional condition code. See Conditional execution on page 5-3. Rd is the destination register. Rn is the register holding the first operand. Operand2 is a flexible second operand. See Operand 2 and the barrel shifter on page 5-7. imm12 is in the range 0-4095. A.1.3 ADR ADR (Address) is an instruction that loads a program or register-relative address (short range). It generates an instruction that adds or subtracts a value to the PC (in the PC-relative case). Alternatively, it can be some other register for a label defined as an offset from a base register defined with the MAP directive (see the ARM tools documentation for more detail). ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-2 Instruction Summary Syntax ADR{cond} Rd, label where: cond is an optional condition code. See Conditional execution on page 5-3. Rd is the destination register. label is a PC or register-relative expression. A.1.4 ADRL ADRL (Address) is a pseudo-instruction that loads a program or register-relative address (long range). It always generates two instructions. Syntax ADRL{cond} Rd, label where: cond is an optional condition code. See Conditional execution on page 5-3. Rd is the destination register. label is a PC-relative expression. The offset between label and the current location has some restrictions. The ADRL pseudo-instruction can generate a wider range of addresses than ADR. A.1.5 AND AND does a bitwise AND on the values in Rn and Operand2. Syntax AND{S}{cond} {Rd,} Rn, where: S (if specified) means that the condition code flags will be updated depending on the result of the instruction. cond is an optional condition code. See Conditional execution on page 5-3. Rd is the destination register. Rn is the register holding the first operand. Operand2 is a flexible second operand. See Operand 2 and the barrel shifter. A.1.6 ASR ASR (Arithmetic Shift Right) shifts the value in Rm right, by the number of bit positions specified and copies the sign bit into vacated bit positions on the left. Permitted shift values are in the range 1-32. It can be considered as giving the signed value of a register divided by a power of two. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-3 Instruction Summary Syntax ASR{S}{cond} {Rd,} Rm, Rs ASR{S}{cond} {Rd,} Rm, imm where: S (if specified) means that the condition code flags will be updated depending on the result of the instruction. cond is an optional condition code. See Conditional execution on page 5-3. Rd is the destination register. Rm is the register holding the operand to be shifted. Rs is the register that holds a shift value to apply to the value in Rm. Only the least significant byte of the register is used. imm is a shift amount, in the range 1-32. A.1.7 B B (Branch) transfers program execution to the address specified by label. Syntax B{cond}{.W} label where: cond is an optional condition code. See Conditional execution on page 5-3. label is a PC-relative expression. .W is an optional instruction width specifier to force the use of a 32-bit instruction in Thumb. A.1.8 BFC BFC (Bit Field Clear) clears bits in a register. A number of bits specified by width are cleared in Rd, starting at lsb. Other bits in Rd are unchanged. Syntax BFC{cond} Rd, #lsb, #width where: cond is an optional condition code. See Conditional execution on page 5-3 Rd is the destination register. lsb specifies the least significant bit to be cleared. width is the number of bits to be cleared. A.1.9 BFI BFI (Bit Field Insert) copies bits into a register. A number of bits in Rd specified by width, starting at lsb, are replaced by bits from Rn, starting at bit[0]. Other bits in Rd are unchanged. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-4 Instruction Summary Syntax BFI{cond} Rd, Rn, #lsb, #width where: cond is an optional condition code. See Conditional execution on page 5-3. Rd is the destination register. Rn is the register that contains the bits to be copied. lsb specifies the least significant bit in Rd to be written to. width is the number of bits to be copied. A.1.10 BIC BIC (bit clear) does an AND operation on the bits in Rn, with the complements of the corresponding bits in the value of Operand2. Syntax BIC{S}{cond} {Rd}, Rn, where: S (if specified) means that the condition code flags will be updated depending on the result of the instruction. cond is an optional condition code. See Conditional execution on page 5-3. Rd is the destination register. Rn is the register holding the first operand. Operand2 is a flexible second operand. See Operand 2 and the barrel shifter on page 5-7. A.1.11 BKPT BKPT (Breakpoint) causes the core to enter Debug state. Syntax BKPT #imm where: imm is an integer in the range 0 – 65535 (ARM) or 0 – 255 (Thumb). This integer is not used by the core itself, but can be used by Debug tools. A.1.12 BL BL (Branch with Link) transfers program execution to the address specified by label and stores the address of the next instruction in the LR (R14) register. Syntax BL{cond} label where: ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-5 Instruction Summary cond is an optional condition code. See Conditional execution on page 5-3. label is a PC-relative expression. A.1.13 BLX BLX (Branch with Link and eXchange) transfers program execution to the address specified by label and stores the address of the next instruction in the LR (R14) register. BLX can change the core state from ARM to Thumb, or from Thumb to ARM. BLX label always changes the core state from Thumb to ARM, or ARM to Thumb. BLX Rm will set the state based on bit[0] of Rm: • Rm bit[0]=0 ARM state. • Rm bit[0]=1 Thumb state. Syntax BLX{cond} label BLX{cond} Rm where: cond is an optional condition code. See Conditional execution on page 5-3. label is a PC-relative expression. Rm is a register that holds the address to branch to. A.1.14 BX BX (Branch and eXchange) transfers program execution to the address specified in a register. BX can change the core state from ARM to Thumb, or from Thumb to ARM. BX Rm will set the state based on bit[0] of Rm: • Rm bit[0] = 0 ARM state. • Rm bit[0] = 1 Thumb state. Syntax BX{cond} Rm where: cond is an optional condition code. See Conditional execution on page 5-3. Rm is a register that holds the address to branch to. A.1.15 BXJ BXJ (Branch and eXchange Jazelle) enter Jazelle State or perform a BX branch and exchange to the address contained in Rm. Syntax BXJ{cond} Rm where: cond is an optional condition code. See Conditional execution on page 5-3. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-6 Instruction Summary Rm is a register that holds the address to branch to if entry to Jazelle fails. A.1.16 CBNZ CBNZ (Compare and Branch if Nonzero) causes a branch if the value in Rn is not zero. It does not change the PSR flags. There is no ARM or 32-bit Thumb versions of this instruction. Syntax CBNZ Rn, label where: label is a pc-relative expression. Rn is a register that holds the operand. A.1.17 CBZ CBZ (Compare and Branch if Zero) causes a branch if the value in Rn is zero. It does not change the PSR flags. There is no ARM or 32-bit Thumb versions of this instruction. Syntax CBZ Rn, label where: label is a PC-relative expression. Rn is a register that holds the operand. A.1.18 CDP CDP (Coprocessor Data Processing operation) performs a coprocessor operation. The purpose of this instruction is defined by the coprocessor implementer. Syntax CDP{cond} coproc, #opcode1, CRd, CRn, CRm{, #opcode2} where: cond is an optional condition code. See Conditional execution on page 5-3. coproc is the name of the coprocessor the instruction is for. This is usually of the form pn, where n is an integer in the range 0 to 15. opcode1 is a 4-bit coprocessor-specific opcode. opcode2 is an optional 3-bit coprocessor-specific opcode. CRd, CRn, CRm are coprocessor registers. A.1.19 CDP2 CDP2 (Coprocessor Data Processing operation) performs a coprocessor operation. The purpose of this instruction is defined by the coprocessor implementer. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-7 Instruction Summary Syntax CDP2{cond} coproc, #opcode1, CRd, CRn, CRm{, #opcode2} where: cond is an optional condition code See Conditional execution on page 5-3. coproc is the name of the coprocessor the instruction is for. This is usually of the form pn, where n is an integer in the range 0 to 15. opcode1 is a 4-bit coprocessor-specific opcode. opcode2 is an optional 3-bit coprocessor-specific opcode. CRd, CRn, CRm are coprocessor registers. A.1.20 CHKA CHKA (Check array) is a ThumbEE instruction. If the value in the first register is less than or equal to, the second, the IndexCheck handler is called. This instruction is only available in 16-bit ThumbEE and only when Thumb-2EE support is present. Syntax CHKA Rn, Rm where: Rn holds the size of the array. Rm contains the array index. A.1.21 CLREX CLREX (Clear Exclusive) moves a local exclusive access monitor to its open-access state. Syntax CLREX{cond} where: cond is an optional condition code. See Conditional execution on page 5-3. A.1.22 CLZ CLZ (Count Leading Zeros) counts the number of leading zeros in the value in Rm and returns the result in Rd. The result returned is 32 if no bits are set in Rm, or 0 if bit [31] is set. Syntax CLZ{cond} Rd, Rm where: cond is an optional condition code. See Conditional execution on page 5-3. Rd is the destination register. Rm is the register holding the operand. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-8 Instruction Summary A.1.23 CMN CMN (Compare Negative) performs a comparison by adding the value of Operand2 to the value in Rn. The condition flags are changed, based on the result, but the result itself is discarded. Syntax CMN{cond} Rn, where: cond is an optional condition code. See Conditional execution on page 5-3. Rn is the register holding the first operand. Operand2 is a flexible second operand. See Operand 2 and the barrel shifter on page 5-7. A.1.24 CMP CMP (Compare) performs a comparison by subtracting the value of Operand2 from the value in Rn. The condition flags are changed, based on the result, but the result itself is discarded. Syntax CMP{cond} Rn, where: cond is an optional condition code. See Conditional execution on page 5-3. Rn is the register holding the first operand. Operand2 is a flexible second operand. See Operand 2 and the barrel shifter on page 5-7. A.1.25 CPS CPS (Change Processor State) can be used to change the processor mode or to enable or disable individual exception types. Syntax CPS #mode CPSIE iflags{, #mode} CPSID iflags{, #mode} where: mode is the number of a mode for the processor to enter. IE Interrupt or Abort Enable. ID Interrupt or Abort Disable. iflags specifies one or more of: • • • ARM DEN0013D ID012214 a = asynchronous abort. i = IRQ. f = FIQ. Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-9 Instruction Summary A.1.26 DBG DBG (Debug) is a hint operation, treated as a NOP by the processor, but can provide information to debug systems. Syntax DBG{cond} {option} where: cond is an optional condition code. See Conditional execution on page 5-3. option is in the range 0-15. A.1.27 DMB DMB (Data Memory Barrier) requires that all explicit memory accesses in program order before the DMB instruction are observed before any explicit memory accesses in program order after the DMB instruction. See Chapter 10 for a detailed description. Syntax DMB{cond} {option} where: cond is an optional condition code. See Conditional execution on page 5-3. option is covered in depth in Chapter 10. A.1.28 DSB DSB (Data Synchronization Barrier) requires that no further instruction executes until all explicit memory accesses, cache maintenance operations, branch prediction and TLB maintenance operations before this instruction complete. See Chapter 10 for a detailed description. Syntax DSB{cond} {option} where: cond is an optional condition code. See Conditional execution on page 5-3. option is covered in depth in Chapter 10. A.1.29 ENTERX ENTERX causes a change from Thumb state to ThumbEE state, or has no effect in ThumbEE state. It is not available in the ARM instruction set. Syntax ENTERX A.1.30 EOR EOR performs an Exclusive OR operation on the values in Rn and Operand2. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-10 Instruction Summary Syntax EOR{S}{cond} {Rd,} Rn, where: S (if specified) means that the condition code flags will be updated depending on the result of the instruction. cond is an optional condition code. See Conditional execution on page 5-3. Rd is the destination register. Rn is the register holding the first operand. Operand2 is a flexible second operand. See Operand 2 and the barrel shifter on page 5-7. A.1.31 ERET ERET (Exception Return) loads the PC from the ELR-hyp and loads the CPSR from SPSR-hyp when executed in Hyp mode. When executed in a Secure or Non-Secure PLI mode, ERET behaves as: • MOVS PC, LR in the ARM instruction set. • SUBS PC, LR, #0 in the Thumb instruction set. Syntax ERET{cond} {q} where: cond is the optional condition code. See Conditional execution on page 5-3. q specifies optional assembler qualifiers on the instruction. The following qualifiers are defined: • .N(narrow), specifies that the assembler must select a 16-bit encoding for the instruction. • .W(wide), specifies that the assembler must select a 32-bit encoding for the intruction. If neither .W nor .N are specified, the assembler can select either 16-bit or 32-bit encodings. if both are available, it must select a 16-bit encoding. A.1.32 HB HB (Handler Branch) branches to a specified handler (available in ThumbEE only). Syntax HB{L} #HandlerID HB{L}P #imm, #HandlerID where: L indicates that the instruction saves a return address in the LR. P means that the instruction passes the value of imm to the handler in R8. imm is an immediate value in the range 0-31 (if L is present), otherwise in the range 0-7. ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-11 Instruction Summary HandlerID is the index number of the handler to be called, in the range 0-31 (if P is specified), otherwise in the range 0-255. A.1.33 ISB ISB (Instruction Synchronization Barrier) flushes the processor pipeline and ensures that context altering operations (such as ASID or other CP15 changes, branch prediction or TLB maintenance activity) before the ISB, are visible to the instructions fetched after the ISB. See Chapter 10 for a detailed description of barriers. Syntax ISB{cond} {option} where: cond is an optional condition code. See Conditional execution on page 5-3. option can be SY (full system), which is the default and so can be omitted. A.1.34 IT IT (If-then) makes up to four following instructions conditional (known as the IT block). The conditions can all be the same, or some can be the logical inverse of others. IT is a pseudo-instruction in ARM state. Syntax IT{x{y{z}}} {cond} where: cond is a condition code. See Conditional execution on page 5-3 that specifies the condition for the first instruction in the IT block. x, y and z specify the condition switch for the second, third and fourth instructions in the IT block, for example, ITTET. The condition switch can be either: A.1.35 • T (Then) applies the condition cond to the instruction. • E (Else) applies the inverse condition of cond to the instruction. LDA LDA (Load-Acquire Word) loads a word from memory and writes it to a register. This instruction was introduced to provide backward compatibility for the ARMv8 architecture AArch32 state. Note LDA imposes ordering restrictions on memory accesses to the same shareability domain. For further information, see Section E2.7.3 in the ARMv8 Architecture Reference Manual (ARM DDI 0487) ARM DEN0013D ID012214 Copyright © 2011 – 2013 ARM. All rights reserved. Non-Confidential A-12 Instruction Summary Syntax LDA{cond}{q}
Source Exif Data:
File Type : PDF File Type Extension : pdf MIME Type : application/pdf PDF Version : 1.3 Linearized : No XMP Toolkit : XMP Core 5.4.0 Title : ARM Cortex-A Series Programmer’s Guide Creator : ARM Limited Format : application/pdf Description : Cortex-A Series Programmer’s Guide, Understanding the ARMv7-A Architecture Metadata Date : 2014:01:24 09:41:05Z Create Date : 2014:01:22 10:49:19Z Modify Date : 2014:01:24 09:41:05Z Creator Tool : FrameMaker 8.0 Instance ID : uuid:a804084a-efda-4494-a247-fed07477a1e9 Document ID : uuid:34e0933d-fe95-4013-9966-c73ed0fa96a4 Copyright : Copyright © 2011 – 2013 ARM. All rights reserved. Producer : Acrobat Distiller 8.3.1 (Windows) Page Count : 421 Author : ARM Limited Subject : Cortex-A Series Programmer’s Guide, Understanding the ARMv7-A Architecture Keywords : Cortex-A Apple Keywords : Cortex-AEXIF Metadata provided by EXIF.tools