ARM Linux specifics

ARM Linux specifics

Thomas Petazzoni / Michael Opdenacker Free Electrons http://free-electrons.com/

Created with OpenOffice.org 2.x

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

Rights to copy
Copyright 2004-2008 Free Electrons [email protected] Document sources, updates and translations: http://free-electrons.com/docs/arm-linux Corrections, suggestions, contributions and translations are welcome!

Attribution ShareAlike 3.0 You are free to copy, distribute, display, and perform the work to make derivative works to make commercial use of the work Under the following conditions Attribution. You must give the original author credit. Share Alike. If you alter, transform, or build upon this work, you may distribute the resulting work only under a license identical to this one. For any reuse or distribution, you must make clear to others the license terms of this work. Any of these conditions can be waived if you get permission from the copyright holder. Your fair use and other rights are in no way affected by the above. License text: http://creativecommons.org/licenses/by-sa/3.0/legalcode

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

Best viewed with...

This document is best viewed with a recent PDF reader or with OpenOffice.org itself! Take advantage of internal or external hyperlinks. So, dont hesitate to click on them! Find pages quickly thanks to automatic search Use thumbnails to navigate in the document in a quick way If youre reading a paper or HTML copy, you should get your copy in PDF or OpenOffice.org format on http://free-electrons.com/training/devtools!

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

Training contents
Floating point and ABIs Floating point support on ARM Linux Different ABIs Thumb mode Introduction to Thumb Thumb and ARM code together Interworking on your system Other ARM extensions

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

Floating point and ABIs

Floating point and ABIs

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

Floating point support (1)

Many ARM platforms do not have an hardware floating point unit Two solutions exists to emulate floating point Hard float: let userspace binaries use floating point instructions, and emulate them in the kernel using the illegal instruction exception Soft float: add the emulation code in userspace at compile time, using gcc -msoft-float option The solution traditionally used in Linux is hard float, with FPA instructions However, hard float is very slow due to the exception handling and context switch.

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

Floating point support (2)

In the Linux kernel, two floating point emulators are available NWFPE, NetWinder Floating Point Emulator CONFIG_FPE_NWFPE FastFPE, faster that NWFPE, but not fully complete and not recommended for scientific applications CONFIG_FPE_FASTFPE Support for VFP is also available VFP is a coprocessor extension for floating point computations available in ARM10, ARM11 and Cortex processor families

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

Mixing hard and soft float

Due to ABI calling conventions differences, it was not possible with the traditional ABI to mix hard and soft float code in userspace An application and all its libraries have to be compiled either for hard float or soft float One of the reason for which floating point emulation in the kernel was preferred over soft float Binaries could take advantage of floating point capable hardware immediately, with no recompiling. Fortunately, the new EABI solves this issue

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

EABI (1)
EABI is a new standardized ABI for the ARM platforms It has several advantages
Ability to mix hard and soft float code, so that general code can be compiled with soft float and several versions of optimized libraries can be provided using hard float Allows to link with code generated by other compilers and provided by other vendors Has integrated support for Thumb interworking

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

EABI (2)
Other changes coming from EABI Structure packing and alignment rules change : no minimum alignement in structures Stack alignment on function entry is 8 bytes instead of 4 bytes Alignment of 64 bits types is 8 bytes instead of 4 System call interface The system call number was passed as part of the swi instruction The kernel had to read and decode the swi instruction, polluting the data cache with instructions Now, the system call number is passed in r7 64 bits function arguments are aligned to an even-number register instead of using the next available pair
ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

10

EABI in gcc and Linux

Support for EABI was added in GCC 4.1.0 Buildroot allows to select the target ABI of the toolchain Support for EABI was added in Linux 2.6.16 CONFIG_AEABI Compiles EABI support in the Linux kernel, so that applications can be compiled with the new EABI CONFIG_OABI_COMPAT In an EABI-able kernel, provides compatibility with old ABI userspace binaries Works only for non-Thumb binaries Running an EABI binary on a non-EABI kernel doesn't work

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

11

Introduction to Thumb

Introduction to Thumb

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

12

Two instruction sets

Default mode on ARM : instructions on 32 bits With the ARMV4T ISA, a new execution mode is added, with 16 bits instructions : Thumb mode ARMV4T ISA is used in ARM7TDMI, ARM9TDMI, ARM7x0T, ARM9xxT In the ARMV5TE ISA, improvements to ease the switch between ARM and Thumb modes 16 bits instructions can be useful to increase code density, and decrease the overall code size

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

13

Instruction encoding
15 001 10 Thumb instruction 0 Rd 8-bit immediate

Major opcode denoting format 3 move/compare/add/sub with immediate value

Minor opcode denoting add instruction

Destination and source register

Immediate value

31 1110 001
Condition: always

0 01001 0 Rd 0 Rd 0000 8-bit immediate ARM instruction

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

14

Compiling a thumb program

Any ARM toolchain is able to produce binaries using the Thumb instruction set Using the -mthumb option of the GNU C Compiler
int bar(int c, int d) { return c + d; } int foo(int a, int b) { a += 3; b -= 2; return bar(b, a); }
arm-linux-gcc -mthumb -c arm-linux-objdump -S 0000000 0: 2: 4: 6: 8: a: c: e: 10: 12: [...] <bar>: b580 b082 af00 1d3b 6018 1c3b 6019 1d3b 1c3a 6819 push sub add adds str adds str adds adds ldr {r7, lr} sp, #8 r7, sp, #0 r3, r7, #4 r0, [r3, #0] r3, r7, #0 r1, [r3, #0] r3, r7, #4 r2, r7, #0 r1, [r3, #0]

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

15

Branches on two instructions

In Thumb mode, branch and link instructions take two instructions
0000000 <bar>: 0: b580 [...] 00000020 <foo>: [...] 4e: f7ff fffe [...] push {r7, lr}

bl

0 <bar>

A.7.1.17 BL, BLX instructions in Thumb mode These Thumb instructions must always occur in the pairs described above

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

16

Size gains
Size gains on a small, non-representative example
int bar(int c, int d) { return c + d; } int foo(int a, int b) { a += 3; b -= 2; return bar(b, a); }
arm-linux-gcc -c

test.arm.o

$ sizediff test.arm.o test.thumb.o text data bss dec 124 0 0 124 96 0 0 96 -28 0 0 -28

hex 7c 60 -1C

filename test.arm.o test.thumb.o +/-

arm-linux-gcc -c -mthumb

test.thumb.o

28 bytes reduction, 22% code size reduction

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

17

Thumb and ARM code together

Thumb and ARM code together Interworking

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

18

Thumb and ARM code together (2)

For several reasons, one might need to combine ARM and Thumb code together Performance-critical code in ARM Libraries compiled in ARM mode The ARM achitecture provides instructions to switch back and forth bx and blx instructions, the lowest bit of the address set allowing to select Thumb or ARM mode ldr and ldm instructions that load the pc register can also be used T bit (bit 5) in the CPSR controls the mode

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

19

Interworking
The GNU C Compiler provides a transparent mechanism called interworking to allow the mix of ARM and Thumb code Interworking-enabled code can be generated using
-mthumb-interwork

The toolchain must be interwork capable

--enable-interwork binutils configuration option --enable-interwork gcc configuration option

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

20

Interworking (2)
000081c4 <main>: 81c4: b580 81c6: af00 [...] 81cc: f005 fb30 [...] 00008220 <foo>: 8220: e1a0c00d 8224: e92dd800 [...] 8254: e12fff1e 0000d830 <__foo_from_thumb>: d830: 4778 d832: 46c0 push add bl {r7, lr} r7, sp, #0 d830 <__foo_from_thumb>

Function main(), compiled in Thumb, calls foo() in ARM mode.

mov push bx

ip, sp {fp, ip, lr, pc} lr

bx nop

pc (mov r8, r8)

0000d834 <__foo_change_to_arm>: d834: eaffea79 b

8220 <foo>

GCC generated wrappers around foo() to switch to ARM mode

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

21

Interworking (3)
Two thumb b instructions
lr = pc + (immediate << 12) pc = lr + (immediate) lr = addr of next instruction | 1
f005 fb30 bl d830 <__foo_from_thumb>

Switch to ARM mode

pc has the lowest bit to 0, switch to ARM
4478 bx pc

Call the correct function Return to Thumb mode at the calling site
lr has the lowest bit to 0, switch to Thumb

eaffea79

b 8220 <foo>

e12fff1e

bx lr

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

22

Interworking on your system

Interworking on your system

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

23

Several solutions
ARM and Thumb mode of the kernel and userspace are independent Can use a ARM kernel with a Thumb mode userspace, the system call ABI remains the same Full Thumb userspace, including the libc uClibc doesn't seem to support Thumb mode correctly, at least gcc 4.2 is not able to compile it Thumb userspace, excluding the libc The solution chosen for our experiments

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

24

Generating the toolchain

Binutils and gcc --enable-interwork Uclibc -mthumb-interwork USE_BX configuration option Automated using Buildroot BR2_INTERWORKING_SUPPORT
Using a Free-Electrons contributed patch, not merged in the official Buildroot version yet
ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

25

Compiling your applications

Manually Add the -mthumb option to the compilation command line CFLAGS+=-mthumb Automated using Buildroot BR2_THUMB_BINARIES Using a Free-Electrons contributed patch, not merged in the official Buildroot version yet Using Scratchbox Need to integrate the new toolchain inside Scratchbox Follow http://www.scratchbox.org/wiki/ForeignToolchains

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

26

Jazelle and Thumb 2

Jazelle, allows to execute some Java bytecode in hardware Need a Jazelle-aware Java Virtual Machine Support in ARM5vTEJ, ARMv6 and ARMv7 http://www.arm.com/products/esd/jazelle_home.html Thumb 2 extends the traditional 16 bits Thumb instruction set with 32 bits instruction Goal is to achieve similar density as Thumb code with performance similar to ARM code Support in ARMv6T2 (ARM1156T2) and ARMv7 (Cortex). Cortex-M3 has only Thumb 2 support. http://www.arm.com/products/CPUs/archi-thumb2.html Linux 2.6.26 adds support for Thumb 2 userspace.
ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

27

ThumbEE
ThumbEE stands for Thumb Execution Environment
Adds more instructions designed for runtime generated code, for example by JIT compilation (automatic null pointer checks or array boundary checks, branch to handlers, etc.)
http://www.arm.com/pdfs/JazelleRCTWhitePaper_final1-0_.pdf

ARM Linux specifics Copyright 2004-2008, Free Electrons Creative Commons Attribution-ShareAlike 3.0 license http://free-electrons.com Sep 15, 2009

28

Related documents

All our technical presentations on http://free-electrons.com/docs Linux kernel Device drivers Architecture specifics Embedded Linux system development
Free Electrons. Kernel, drivers and embedded Linux development, consulting, training and support. http//free-electrons.com

How to help
You can help us to improve and maintain this document... By sending corrections, suggestions, contributions and translations By asking your organization to order development, consulting and training services performed by the authors of these documents (see http://free-electrons.com/). By sharing this document with your friends, colleagues and with the local Free Software community. By adding links on your website to our on-line materials, to increase their visibility in search engine results.

Free Electrons. Kernel, drivers and embedded Linux development, consulting, training and support. http//free-electrons.com

Linux kernel Linux device drivers Board support code Mainstreaming kernel code Kernel debugging Embedded Linux Training All materials released with a free license! Unix and GNU/Linux basics Linux kernel and drivers development Real-time Linux, uClinux Development and profiling tools Lightweight tools for embedded systems Root filesystem creation Audio and multimedia System optimization

Free Electrons
Our services
Custom Development System integration Embedded Linux demos and prototypes System optimization Application and interface development Consulting and technical support Help in decision making System architecture System design and performance review Development tool and application support Investigating issues and fixing tool bugs