Linux Kernel and Driver Development Training

Linux Kernel and Driver

Development Training
Free Electrons

Free Electrons
Embedded Linux
Developers

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Latest update: June 10, 2015.
Document updates and sources:
http://free- electrons.com/doc/training/linux- kernel
Corrections, suggestions, contributions and translations are welcome!
Send them to feedback@free-electrons.com

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

1/477

Rights to copy

Copyright 2004-2015, Free Electrons

License: Creative Commons Attribution - Share Alike 3.0

http://creativecommons.org/licenses/by-sa/3.0/legalcode
You are free:
I to copy, distribute, display, and perform the work
I to make derivative works
I to make commercial use of the work
Under the following conditions:
I Attribution. You must give the original author credit.
I 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.
I For any reuse or distribution, you must make clear to others the license terms of
this work.
I 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.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

2/477

Hyperlinks in the document

There are many hyperlinks in the document
I

Regular hyperlinks:
http://kernel.org/

Kernel documentation links:

Documentation/kmemcheck.txt

Links to kernel source files and directories:

drivers/input
include/linux/fb.h

Links to the declarations, definitions and instances of kernel

symbols (functions, types, data, structures):
platform_get_irq()
GFP_KERNEL
struct file_operations

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

3/477

Free Electrons at a glance

I

Engineering company created in 2004

(not a training company!)

Locations: Orange, Toulouse, Lyon (France)

Serving customers all around the world

See http://free-electrons.com/company/customers/

Head count: 9
Only Free Software enthusiasts!

Focus: Embedded Linux, Linux kernel, Android Free Software

/ Open Source for embedded and real-time systems.

Activities: development, training, consulting, technical

support.

Added value: get the best of the user and development

community and the resources it offers.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

4/477

Free Electrons on-line resources

I

All our training materials:

http://free-electrons.com/docs/
Technical blog:
http://free-electrons.com/blog/
Quarterly newsletter:
http://lists.freeelectrons.com/mailman/listinfo/newsletter
News and discussions (Google +):
https://plus.google.com/+FreeElectronsDevelopers
News and discussions (LinkedIn):
http://linkedin.com/groups/Free-Electrons-4501089
Quick news (Twitter):
http://twitter.com/free_electrons
Linux Cross Reference - browse Linux kernel sources on-line:
http://lxr.free-electrons.com

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

5/477

Generic course information

Generic course
information

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

6/477

Hardware used in this training session

BeagleBone Black, from CircuitCo
I Texas Instruments AM335x (ARM Cortex-A8)
I Powerful CPU, with 3D acceleration,

additional processors (PRUs) and lots of

peripherals.
I 512 MB of RAM
I 2 GB of on-board eMMC storage

(4 GB in Rev C)
I USB host and USB device ports
I microSD slot
I HDMI port
I 2 x 46 pins headers, with access to many

expansion buses (I2C, SPI, UART and more)

I A huge number of expansion boards, called

capes. See http://beagleboardtoys.com/.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

7/477

Do not damage your BeagleBone Black!

Do not remove power abruptly:

I

I
I

Boards components have been damaged by removing the

power or USB cable in an abrupt way, not leaving the PMIC
the time to switch off the components in a clean way. See
http://bit.ly/1FWHNZi
Reboot (reboot) or shutdown (halt) the board in software
when Linux is running.
You can also press the RESET button to reset and reboot.
When there is no software way, you can also switch off the
board by pressing the POWER button for 8 seconds.

Do not leave your board powered on a metallic surface (like a

laptop with a metal finish).

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

8/477

Shopping list: hardware for this course

I BeagleBone Black - Multiple distributors:

See http://beagleboard.org/Products/
I Nintendo Nunchuck with UEXT connector:

Olimex: http://j.mp/1dTYLfs
I Breadboard jumper wires - Male ends:

Olimex: http://j.mp/IUaBsr
I USB Serial Cable - Male ends:

Olimex: http://j.mp/1eUuY2K
I USB Serial Cable - Female ends:

Olimex: http://j.mp/18Hk8yF
I Note that both USB serial cables are the same.

Only the gender of their connector changes.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

9/477

Participate!
During the lectures...
I

Dont hesitate to ask questions. Other people in the audience

may have similar questions too.

This helps the trainer to detect any explanation that wasnt

clear or detailed enough.

Dont hesitate to share your experience, for example to

compare Linux / Android with other operating systems used
in your company.

Your point of view is most valuable, because it can be similar

to your colleagues and different from the trainers.

Your participation can make our session more interactive and

make the topics easier to learn.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

10/477

Practical lab guidelines

During practical labs...
I

We cannot support more than 8 workstations at once (each

with its board and equipment). Having more would make the
whole class progress slower, compromising the coverage of the
whole training agenda (exception for public sessions: up to 10
people).

So, if you are more than 8 participants, please form up to 8

working groups.

Open the electronic copy of your lecture materials, and use it

throughout the practical labs to find the slides you need again.

Dont copy and paste from the PDF slides.

The slides contain UTF-8 characters that look the same as
ASCII ones, but wont be understood by shells or compilers.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

11/477

Cooperate!

As in the Free Software and Open Source community, cooperation

during practical labs is valuable in this training session:
I

If you complete your labs before other people, dont hesitate

to help other people and investigate the issues they face. The
faster we progress as a group, the more time we have to
explore extra topics.

Explain what you understood to other participants when

needed. It also helps to consolidate your knowledge.

Dont hesitate to report potential bugs to your instructor.

Dont hesitate to look for solutions on the Internet as well.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

12/477

Command memento sheet

I

This memento sheet gives

command examples for the most
typical needs (looking for files,
extracting a tar archive...)

It saves us 1 day of UNIX / Linux

command line training.

Our best tip: in the command line

shell, always hit the Tab key to
complete command names and file
paths. This avoids 95% of typing
mistakes.

Get an electronic copy on

http://free-electrons.com/
doc/training/embeddedlinux/command_memento.pdf

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

13/477

vi basic commands
I

The vi editor is very useful to

make quick changes to files in an
embedded target.

Though not very user friendly at

first, vi is very powerful and its
main 15 commands are easy to
learn and are sufficient for 99% of
everyones needs!

Get an electronic copy on

http://free-electrons.com/
doc/training/embeddedlinux/vi_memento.pdf

You can also take the quick tutorial

by running vimtutor. This is a
worthy investment!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

14/477

Practical lab - Training Setup

Prepare your lab environment

I

Download the lab archive

Enforce correct permissions

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

15/477

Linux Kernel Introduction

Linux Kernel
Introduction

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

16/477

Linux Kernel Introduction

Linux features

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

17/477

History

The Linux kernel is one component of a system, which also

requires libraries and applications to provide features to end
users.
The Linux kernel was created as a hobby in 1991 by a Finnish
student, Linus Torvalds.
I

Linux quickly started to be used as the kernel for free software

operating systems

Linus Torvalds has been able to create a large and dynamic

developer and user community around Linux.

Nowadays, more than one thousand people contribute to each

kernel release, individuals or companies big and small.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

18/477

Linux kernel key features

Portability and hardware

support. Runs on most
architectures.

Security. It cant hide its

flaws. Its code is reviewed
by many experts.

Scalability. Can run on

super computers as well as
on tiny devices (4 MB of
RAM is enough).

Stability and reliability.

Modularity. Can include

only what a system needs
even at run time.

Easy to program. You can

learn from existing code.
Many useful resources on
the net.

Compliance to standards
and interoperability.

Exhaustive networking
support.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

19/477

Linux kernel in the system

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

20/477

Linux kernel main roles

Manage all the hardware resources: CPU, memory, I/O.

Provide a set of portable, architecture and hardware

independent APIs to allow user space applications and
libraries to use the hardware resources.
Handle concurrent accesses and usage of hardware
resources from different applications.

Example: a single network interface is used by multiple user

space applications through various network connections. The
kernel is responsible to multiplex the hardware resource.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

21/477

System calls

The main interface between the kernel and user space is the
set of system calls
About 300 system calls that provide the main kernel services
I

File and device operations, networking operations,

inter-process communication, process management, memory
mapping, timers, threads, synchronization primitives, etc.

This interface is stable over time: only new system calls can
be added by the kernel developers

This system call interface is wrapped by the C library, and

user space applications usually never make a system call
directly but rather use the corresponding C library function

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

22/477

Pseudo filesystems

Linux makes system and kernel information available in user

space through pseudo filesystems, sometimes also called
virtual filesystems

Pseudo filesystems allow applications to see directories and

files that do not exist on any real storage: they are created
and updated on the fly by the kernel
The two most important pseudo filesystems are

proc, usually mounted on /proc:

Operating system related information (processes, memory
management parameters...)
sysfs, usually mounted on /sys:
Representation of the system as a set of devices and buses.
Information about these devices.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

23/477

Inside the Linux kernel

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

24/477

Supported hardware architectures

See the arch/ directory in the kernel sources

Minimum: 32 bit processors, with or without MMU, and gcc

support

32 bit architectures (arch/ subdirectories)

Examples: arm, avr32, blackfin, c6x, m68k, microblaze,
mips, score, sparc, um

64 bit architectures:
Examples: alpha, arm64, ia64, tile

32/64 bit architectures

Examples: powerpc, x86, sh, sparc

Find details in kernel sources: arch/<arch>/Kconfig,

arch/<arch>/README, or Documentation/<arch>/

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

25/477

Embedded Linux Kernel Usage

Embedded Linux
Kernel Usage

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

26/477

Embedded Linux Kernel Usage

Linux kernel sources

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

27/477

Location of kernel sources

I

The official versions of the Linux kernel, as released by Linus

Torvalds, are available at http://www.kernel.org
I
I

Many chip vendors supply their own kernel sources

I
I
I

These versions follow the development model of the kernel

However, they may not contain the latest development from a
specific area yet. Some features in development might not be
ready for mainline inclusion yet.
Focusing on hardware support first
Can have a very important delta with mainline Linux
Useful only when mainline hasnt caught up yet.

Many kernel sub-communities maintain their own kernel, with

usually newer but less stable features
I

Architecture communities (ARM, MIPS, PowerPC, etc.),

device drivers communities (I2C, SPI, USB, PCI, network,
etc.), other communities (real-time, etc.)
No official releases, only development trees are available.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

28/477

Getting Linux sources

The kernel sources are available from

http://kernel.org/pub/linux/kernel as full tarballs
(complete kernel sources) and patches (differences between
two kernel versions).
However, more and more people use the git version control
system. Absolutely needed for kernel development!
I

Fetch the entire kernel sources and history

git clone git://git.kernel.org/pub/scm/linux/
kernel/git/torvalds/linux.git
Create a branch that starts at a specific stable version
git checkout -b <name-of-branch> v3.11
Web interface available at http://git.kernel.org/cgit/
linux/kernel/git/torvalds/linux.git/tree/.
Read more about Git at http://git-scm.com/

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

29/477

Linux kernel size (1)

I

Linux 3.10 sources:

Raw size: 573 MB (43,000 files, approx 15,800,000 lines)
gzip compressed tar archive: 105 MB
bzip2 compressed tar archive: 83 MB (better)
xz compressed tar archive: 69 MB (best)

Minimum Linux 3.17 compiled kernel size, booting on the

ARM Versatile board (hard drive on PCI, ext2 filesystem, ELF
executable support, framebuffer console and input devices):
876 KB (compressed), 2.3 MB (raw)

Why are these sources so big?

Because they include thousands of device drivers, many
network protocols, support many architectures and
filesystems...

The Linux core (scheduler, memory management...) is pretty

small!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

30/477

Linux kernel size (2)

As of kernel version 3.10.

I

drivers/: 49.4%

arch/: 21.9%

fs/: 6.0%

include/: 4.7%

sound/: 4.4%

Documentation/: 4.0%

net/: 3.9%

firmware/: 1.0%

kernel/: 1.0%

tools/: 0.9%

scripts/: 0.5%

mm/: 0.5%

crypto/: 0.4%

security/: 0.4%

lib/: 0.4%

block/: 0.2%

...

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

31/477

Practical lab - Get Linux Kernel Source Code

Clone the mainline Linux source

tree with git

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

32/477

Kernel Source Code

Kernel Source
Code

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

33/477

Kernel Source Code

Linux Code and Device Drivers

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

34/477

Programming language

Implemented in C like all Unix systems. (C was created to

implement the first Unix systems)
A little Assembly is used too:
I
I

I
I

CPU and machine initialization, exceptions

Critical library routines.

No C++ used, see http://www.tux.org/lkml/#s15-3

All the code compiled with gcc
I

I
I

Many gcc specific extensions used in the kernel code, any

ANSI C compiler will not compile the kernel
A few alternate compilers are supported (Intel and Marvell)
See http://gcc.gnu.org/onlinedocs/gcc-4.9.0/gcc/CExtensions.html

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

35/477

No C library

The kernel has to be standalone and cant use user space

code.

User space is implemented on top of kernel services, not the

opposite.

Kernel code has to supply its own library implementations

(string utilities, cryptography, uncompression ...)

So, you cant use standard C library functions in kernel code.

(printf(), memset(), malloc(),...).

Fortunately, the kernel provides similar C functions for your

convenience, like printk(), memset(), kmalloc(), ...

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

36/477

Portability

The Linux kernel code is designed to be portable

All code outside arch/ should be portable

To this aim, the kernel provides macros and functions to
abstract the architecture specific details

Endianness
I
I
I
I

I
I
I

cpu_to_be32()
cpu_to_le32()
be32_to_cpu()
le32_to_cpu()

I/O memory access

Memory barriers to provide ordering guarantees if needed
DMA API to flush and invalidate caches if needed

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

37/477

No floating point computation

Never use floating point numbers in kernel code. Your code

may be run on a processor without a floating point unit (like
on certain ARM CPUs).
Dont be confused with floating point related configuration
options
I

They are related to the emulation of floating point operation

performed by the user space applications, triggering an
exception into the kernel.
Using soft-float, i.e. emulation in user space, is however
recommended for performance reasons.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

38/477

No stable Linux internal API

The internal kernel API to implement kernel code can undergo

changes between two releases.

In-tree drivers are updated by the developer proposing the API

change: works great for mainline code.

An out-of-tree driver compiled for a given version may no

longer compile or work on a more recent one.

See Documentation/stable_api_nonsense.txt in kernel

sources for reasons why.

Of course, the kernel to userspace API does not change

(system calls, /proc, /sys), as it would break existing
programs.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

39/477

Kernel memory constraints

No memory protection

Accessing illegal memory locations result in (often fatal)

kernel oopses.

Fixed size stack (8 or 4 KB). Unlike in user space, theres no

way to make it grow.

Kernel memory cant be swapped out (for the same reasons).

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

40/477

Linux kernel licensing constraints

I

The Linux kernel is licensed under the GNU General Public

License version 2
I

However, when the software is redistributed, either modified

or unmodified, the GPL requires that you redistribute the
software under the same license, with the source code
I

This license gives you the right to use, study, modify and share
the software freely

If modifications are made to the Linux kernel (for example to

adapt it to your hardware), it is a derivative work of the kernel,
and therefore must be released under GPLv2
The validity of the GPL on this point has already been verified
in courts

However, youre only required to do so

I
I

At the time the device starts to be distributed

To your customers, not to the entire world

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

41/477

Proprietary code and the kernel

It is illegal to distribute a binary kernel that includes statically

compiled proprietary drivers
The kernel modules are a gray area: are they derived works of
the kernel or not?
I

The general opinion of the kernel community is that

proprietary drivers are bad: http://j.mp/fbyuuH
From a legal point of view, each driver is probably a different
case
Is it really useful to keep your drivers secret?

There are some examples of proprietary drivers, like the Nvidia

graphics drivers
I
I

They use a wrapper between the driver and the kernel

Unclear whether it makes it legal or not

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

42/477

Advantages of GPL drivers

I

You dont have to write your driver from scratch. You can
reuse code from similar free software drivers.

You could get free community contributions, support, code

review and testing, though this generally only happens with
code submitted for the mainline kernel.

Your drivers can be freely and easily shipped by others (for

example by Linux distributions or embedded Linux build
systems).

Pre-compiled drivers work with only one kernel version and

one specific configuration, making life difficult for users who
want to change the kernel version.

Legal certainty, you are sure that a GPL driver is fine from a
legal point of view.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

43/477

Advantages of in-tree kernel drivers

Once your sources are accepted in the mainline tree, they are
maintained by people making changes.

Near cost-free maintenance, security fixes and improvements.

Easy access to your sources by users.

Many more people reviewing your code.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

44/477

User space device drivers 1/3

In some cases, it is possible to implement device drivers in

user space!
Can be used when
I

The kernel provides a mechanism that allows userspace

applications to directly access the hardware.
There is no need to leverage an existing kernel subsystem such
as the networking stack or filesystems.
There is no need for the kernel to act as a multiplexer for
the device: only one application accesses the device.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

45/477

User space device drivers 2/3

Possibilities for userspace device drivers:

I
I
I
I

USB with libusb, http://www.libusb.org/

SPI with spidev, Documentation/spi/spidev
I2C with i2cdev, Documentation/i2c/dev-interface
Memory-mapped devices with UIO, including interrupt
handling, Documentation/DocBook/uio-howto/

Certain classes of devices (printers, scanners, 2D/3D graphics

acceleration) are typically handled partly in kernel space,
partly in user space.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

46/477

User space device drivers 3/3

I

Advantages
I

I
I
I

I
I
I
I

No need for kernel coding skills. Easier to reuse code between

devices.
Drivers can be written in any language, even Perl!
Drivers can be kept proprietary.
Driver code can be killed and debugged. Cannot crash the
kernel.
Can be swapped out (kernel code cannot be).
Can use floating-point computation.
Less in-kernel complexity.
Potentially higher performance, especially for memory-mapped
devices, thanks to the avoidance of system calls.

Drawbacks
I
I

Less straightforward to handle interrupts.

Increased interrupt latency vs. kernel code.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

47/477

Kernel Source Code

Linux sources

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

48/477

Linux sources structure 1/5

I

arch/<ARCH>
I
I
I
I

block/
I

Linux copying conditions (GNU GPL)

CREDITS
I

Block layer core

COPYING
I

Architecture specific code

arch/<ARCH>/mach-<machine>, machine/board specific code
arch/<ARCH>/include/asm, architecture-specific headers
arch/<ARCH>/boot/dts, Device Tree source files, for some
architectures

Linux main contributors

crypto/
I

Cryptographic libraries

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

49/477

Linux sources structure 2/5

I

Documentation/

drivers/

firmware/

fs/

include/

All device drivers except sound ones (usb, pci...)

Legacy: firmware images extracted from old drivers
Filesystems (fs/ext3/, etc.)
Kernel headers

include/linux/
I

Kernel documentation. Dont miss it!

Linux kernel core headers

include/uapi/
I

User space API headers

init/

ipc/

Linux initialization (including main.c)

Code used for process communication

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

50/477

Linux sources structure 3/5

I

Kbuild
I

Maintainers of each kernel part. Very useful!

Makefile
I

Misc library routines (zlib, crc32...)

MAINTAINERS
I

Linux kernel core (very small!)

lib/
I

Top level description file for configuration parameters

kernel/
I

Part of the kernel build system

Kconfig

Top Linux Makefile (sets arch and version)

mm/
I

Memory management code (small too!)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

51/477

Linux sources structure 4/5

I

net/
I

Security model implementations (SELinux...)

sound/
I

Scripts for internal or external use

security/
I

Sample code (markers, kprobes, kobjects...)

scripts/
I

Bug report instructions

samples/
I

Overview and building instructions

REPORTING-BUGS
I

Network support code (not drivers)

README

Sound support code and drivers

tools/
I

Code for various user space tools (mostly C)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

52/477

Linux sources structure 5/5

usr/
I

Code to generate an initramfs cpio archive

virt/
I

Virtualization support (KVM)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

53/477

Kernel Source Code

Kernel source management tools

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

54/477

Cscope
I

Tool to browse source code (mainly C, but also C++ or Java)

Supports huge projects like the Linux kernel. Typically takes

less than 1 min. to index the whole Linux sources.
In Linux kernel sources, two ways of running it:

cscope -Rk
All files for all architectures at once
make cscope
cscope -d cscope.out
Only files for your current architecture

Allows searching for a symbol, a definition, functions, strings,

files, etc.

Integration with editors like vim and emacs.

Dedicated graphical front-end: KScope

http://cscope.sourceforge.net/

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

55/477

Cscope screenshot

[Tab]: move the cursor between search results and commands

[Ctrl] [D]: exit cscope
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

56/477

LXR: Linux Cross Reference

I

Generic source indexing tool and code browser

Web server based, very easy and fast to use

Very easy to find the declaration, implementation or usage of

symbols

Supports C and C++

Supports huge code projects such as the Linux kernel (431

MB of source code in version 3.0).

Takes a little time and patience to setup (configuration,

indexing, web server configuration)

You dont need to set up LXR by yourself. Use our

http://lxr.free-electrons.com server!

http://sourceforge.net/projects/lxr

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

57/477

LXR screenshot

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

58/477

Practical lab - Kernel Source Code - Exploring

Explore kernel sources manually

Use automated tools to explore the

source code

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

59/477

Kernel Source Code

Kernel configuration

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

60/477

Kernel configuration and build system

The kernel configuration and build system is based on

multiple Makefiles

One only interacts with the main Makefile, present at the

top directory of the kernel source tree
Interaction takes place

I
I

using the make tool, which parses the Makefile

through various targets, defining which action should be done
(configuration, compilation, installation, etc.). Run
make help to see all available targets.

Example
I
I

cd linux-3.6.x/
make <target>

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

61/477

Kernel configuration (1)

The kernel contains thousands of device drivers, filesystem

drivers, network protocols and other configurable items

Thousands of options are available, that are used to

selectively compile parts of the kernel source code

The kernel configuration is the process of defining the set of

options with which you want your kernel to be compiled
The set of options depends

I
I

On your hardware (for device drivers, etc.)

On the capabilities you would like to give to your kernel
(network capabilities, filesystems, real-time, etc.)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

62/477

Kernel configuration (2)

The configuration is stored in the .config file at the root of

kernel sources
I

I
I
I

Simple text file, key=value style

As options have dependencies, typically never edited by hand,

but through graphical or text interfaces:
make xconfig, make gconfig (graphical)
make menuconfig, make nconfig (text)
You can switch from one to another, they all load/save the
same .config file, and show the same set of options

To modify a kernel in a GNU/Linux distribution: the

configuration files are usually released in /boot/, together
with kernel images: /boot/config-3.2.0-31-generic

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

63/477

Kernel or module?
I

The kernel image is a single file, resulting from the linking

of all object files that correspond to features enabled in the
configuration
I
I

This is the file that gets loaded in memory by the bootloader

All included features are therefore available as soon as the
kernel starts, at a time where no filesystem exists

Some features (device drivers, filesystems, etc.) can however

be compiled as modules
I

These are plugins that can be loaded/unloaded dynamically to

add/remove features to the kernel
Each module is stored as a separate file in the filesystem,
and therefore access to a filesystem is mandatory to use
modules
This is not possible in the early boot procedure of the kernel,
because no filesystem is available

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

64/477

Kernel option types

There are different types of options

I

bool options, they are either

I
I

tristate options, they are either

I
I
I

I
I
I

true (to include the feature in the kernel) or

false (to exclude the feature from the kernel)
true (to include the feature in the kernel image) or
module (to include the feature as a kernel module) or
false (to exclude the feature)

int options, to specify integer values

hex options, to specify hexadecimal values
string options, to specify string values

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

65/477

Kernel option dependencies

There are dependencies between kernel options

For example, enabling a network driver requires the network

stack to be enabled
Two types of dependencies

depends on dependencies. In this case, option A that depends

on option B is not visible until option B is enabled
select dependencies. In this case, with option A depending
on option B, when option A is enabled, option B is
automatically enabled
make xconfig allows to see all options, even the ones that
cannot be selected because of missing dependencies. In this
case, they are displayed in gray

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

66/477

make xconfig

make xconfig
I

The most common graphical interface to configure the kernel.

Make sure you read

help -> introduction: useful options!

File browser: easier to load configuration files

Search interface to look for parameters

Required Debian / Ubuntu packages: libqt4-dev g++

(libqt3-mt-dev for older kernel releases)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

67/477

make xconfig screenshot

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

68/477

make xconfig search interface

Looks for a keyword in the parameter name. Allows to select or
unselect found parameters.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

69/477

Kernel configuration options

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

70/477

Corresponding .config file excerpt

Options are grouped by sections and are prefixed with CONFIG_.
#
# CD-ROM/DVD Filesystems
#
CONFIG_ISO9660_FS=m
CONFIG_JOLIET=y
CONFIG_ZISOFS=y
CONFIG_UDF_FS=y
CONFIG_UDF_NLS=y
#
# DOS/FAT/NT Filesystems
#
# CONFIG_MSDOS_FS is not set
# CONFIG_VFAT_FS is not set
CONFIG_NTFS_FS=m
# CONFIG_NTFS_DEBUG is not set
CONFIG_NTFS_RW=y
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

71/477

make gconfig

make gconfig
I

GTK based graphical

configuration interface.
Functionality similar to that
of make xconfig.

Just lacking a search

functionality.

Required Debian packages:

libglade2-dev

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

72/477

make menuconfig

make menuconfig
I

Useful when no graphics are

available. Pretty convenient
too!

Same interface found in

other tools: BusyBox,
Buildroot...

Required Debian packages:

libncurses-dev

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

73/477

make nconfig

make nconfig
I

A newer, similar text

interface

More user friendly (for

example, easier to access
help information).

Required Debian packages:

libncurses-dev

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

74/477

make oldconfig

make oldconfig
I

Needed very often!

Useful to upgrade a .config file from an earlier kernel release

Issues warnings for configuration parameters that no longer

exist in the new kernel.

Asks for values for new parameters (while xconfig and

menuconfig silently set default values for new parameters).

If you edit a .config file by hand, its strongly recommended to

run make oldconfig afterwards!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

75/477

Undoing configuration changes

A frequent problem:
I

After changing several kernel configuration settings, your

kernel no longer works.

If you dont remember all the changes you made, you can get
back to your previous configuration:
$ cp .config.old .config

All the configuration interfaces of the kernel (xconfig,

menuconfig, oldconfig...) keep this .config.old backup
copy.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

76/477

Configuration per architecture

The set of configuration options is architecture dependent

I
I

Some configuration options are very architecture-specific

Most of the configuration options (global kernel options,
network subsystem, filesystems, most of the device drivers) are
visible in all architectures.

By default, the kernel build system assumes that the kernel is

being built for the host architecture, i.e. native compilation

The architecture is not defined inside the configuration, but at

a higher level

We will see later how to override this behaviour, to allow the

configuration of kernels for a different architecture

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

77/477

Kernel Source Code

Compiling and installing the kernel

for the host system

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

78/477

Kernel compilation
I

make
I
I

in the main kernel source directory

Remember to run multiple jobs in parallel if you have multiple
CPU cores. Example: make -j 4
No need to run as root!

Generates
I

vmlinux, the raw uncompressed kernel image, in the ELF

format, useful for debugging purposes, but cannot be booted
arch/<arch>/boot/*Image, the final, usually compressed,
kernel image that can be booted
I

bzImage for x86, zImage for ARM, vmImage.gz for Blackfin,

etc.

arch/<arch>/boot/dts/*.dtb, compiled Device Tree files

(on some architectures)
All kernel modules, spread over the kernel source tree, as .ko
files.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

79/477

Kernel installation
I

make install
I

Installs
I

Does the installation for the host system by default, so needs

to be run as root. Generally not used when compiling for an
embedded system, as it installs files on the development
workstation.
/boot/vmlinuz-<version>
Compressed kernel image. Same as the one in
arch/<arch>/boot
/boot/System.map-<version>
Stores kernel symbol addresses
/boot/config-<version>
Kernel configuration for this version

Typically re-runs the bootloader configuration utility to take

the new kernel into account.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

80/477

Module installation

make modules_install
I

Does the installation for the host system by default, so needs

to be run as root

Installs all modules in /lib/modules/<version>/

I

kernel/
Module .ko (Kernel Object) files, in the same directory
structure as in the sources.
modules.alias
Module aliases for module loading utilities. Example line:
alias sound-service-?-0 snd_mixer_oss
modules.dep
Module dependencies
modules.symbols
Tells which module a given symbol belongs to.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

81/477

Kernel cleanup targets

Clean-up generated files (to force

re-compilation):
make clean

Remove all generated files. Needed when

switching from one architecture to another.
Caution: it also removes your .config file!
make mrproper

Also remove editor backup and patch reject files

(mainly to generate patches):
make distclean

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

82/477

Kernel Source Code

Cross-compiling the kernel

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

83/477

Cross-compiling the kernel

When you compile a Linux kernel for another CPU architecture

I

Much faster than compiling natively, when the target system

is much slower than your GNU/Linux workstation.

Much easier as development tools for your GNU/Linux

workstation are much easier to find.

To make the difference with a native compiler, cross-compiler

executables are prefixed by the name of the target system,
architecture and sometimes library. Examples:
mips-linux-gcc, the prefix is mips-linuxarm-linux-gnueabi-gcc, the prefix is arm-linux-gnueabi-

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

84/477

Specifying cross-compilation (1)

The CPU architecture and cross-compiler prefix are defined through

the ARCH and CROSS_COMPILE variables in the toplevel Makefile.
I

ARCH is the name of the architecture. It is defined by the

name of the subdirectory in arch/ in the kernel sources
I

Example: arm if you want to compile a kernel for the arm

architecture.

CROSS_COMPILE is the prefix of the cross compilation tools

I

Example: arm-linux- if your compiler is arm-linux-gcc

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

85/477

Specifying cross-compilation (2)

Two solutions to define ARCH and CROSS_COMPILE:
I

Pass ARCH and CROSS_COMPILE on the make command line:

make ARCH=arm CROSS_COMPILE=arm-linux- ...
Drawback: it is easy to forget to pass these variables when
you run any make command, causing your build and
configuration to be screwed up.

Define ARCH and CROSS_COMPILE as environment variables:

export ARCH=arm
export CROSS_COMPILE=arm-linuxDrawback: it only works inside the current shell or terminal.
You could put these settings in a file that you source every
time you start working on the project. If you only work on a
single architecture with always the same toolchain, you could
even put these settings in your ~/.bashrc file to make them
permanent and visible from any terminal.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

86/477

Predefined configuration files

Default configuration files available, per board or per-CPU

family
I

They are stored in arch/<arch>/configs/, and are just

minimal .config files
This is the most common way of configuring a kernel for
embedded platforms

Run make help to find if one is available for your platform

To load a default configuration file, just run

make acme_defconfig
I

This will overwrite your existing .config file!

To create your own default configuration file

I
I

make savedefconfig, to create a minimal configuration file

mv defconfig arch/<arch>/configs/myown_defconfig

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

87/477

Configuring the kernel

After loading a default configuration file, you can adjust the

configuration to your needs with the normal xconfig,
gconfig or menuconfig interfaces
As the architecture is different from your host architecture
I

Some options will be different from the native configuration

(processor and architecture specific options, specific drivers,
etc.)
Many options will be identical (filesystems, network protocols,
architecture-independent drivers, etc.)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

88/477

Device Tree
I

Many embedded architectures have a lot of non-discoverable

hardware.

Depending on the architecture, such hardware is either

described using C code directly within the kernel, or using a
special hardware description language in a Device Tree.

ARM, PowerPC, OpenRISC, ARC, Microblaze are examples of

architectures using the Device Tree.
A Device Tree Source, written by kernel developers, is
compiled into a binary Device Tree Blob, passed at boot time
to the kernel.

There is one different Device Tree for each board/platform

supported by the kernel, available in
arch/arm/boot/dts/<board>.dtb.

The bootloader must load both the kernel image and the
Device Tree Blob in memory before starting the kernel.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

89/477

Building and installing the kernel

I
I

Run make
Copy the final kernel image to the target storage
I

make install is rarely used in embedded development, as

the kernel image is a single file, easy to handle
I

can be uImage, zImage, vmlinux, bzImage in

arch/<arch>/boot
copying the Device Tree Blob might be necessary as well, they
are available in arch/<arch>/boot/dts

It is however possible to customize the make install behaviour

in arch/<arch>/boot/install.sh

make modules_install is used even in embedded

development, as it installs many modules and description files
I
I

make INSTALL_MOD_PATH=<dir>/ modules_install

The INSTALL_MOD_PATH variable is needed to install the
modules in the target root filesystem instead of your host root
filesystem.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

90/477

Booting with U-Boot

I
I

Recent versions of U-Boot can boot the zImage binary.

Older versions require a special kernel image format: uImage
I

uImage is generated from zImage using the mkimage tool. It

is done automatically by the kernel make uImage target.
On some ARM platforms, make uImage requires passing a
LOADADDR environment variable, which indicates at which
physical memory address the kernel will be executed.

In addition to the kernel image, U-Boot can also pass a

Device Tree Blob to the kernel.
The typical boot process is therefore:
1. Load zImage or uImage at address X in memory
2. Load <board>.dtb at address Y in memory
3. Start the kernel with bootz X - Y or bootm X - Y
The - in the middle indicates no initramfs

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

91/477

Kernel command line

I

In addition to the compile time configuration, the kernel

behaviour can be adjusted with no recompilation using the
kernel command line
The kernel command line is a string that defines various
arguments to the kernel
I
I
I
I

It is very important for system configuration

root= for the root filesystem (covered later)
console= for the destination of kernel messages
Many more exist. The most important ones are documented in
Documentation/kernel-parameters.txt in kernel sources.

This kernel command line is either

I

Passed by the bootloader. In U-Boot, the contents of the

bootargs environment variable is automatically passed to the
kernel
Built into the kernel, using the CONFIG_CMDLINE option.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

92/477

Practical lab - Kernel compiling and booting

1st lab: board and bootloader setup:
I

Prepare the board and access its

serial port

Configure its bootloader to use

TFTP

2nd lab: kernel compiling and booting:

I

Set up a cross-compiling
environment

Cross-compile a kernel for an ARM

target platform

Boot this kernel from a directory

on your workstation, accessed by
the board through NFS

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

93/477

Kernel Source Code

Using kernel modules

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

94/477

Advantages of modules

Modules make it easy to develop drivers without rebooting:

load, test, unload, rebuild, load...

Useful to keep the kernel image size to the minimum

(essential in GNU/Linux distributions for PCs).

Also useful to reduce boot time: you dont spend time

initializing devices and kernel features that you only need later.

Caution: once loaded, have full control and privileges in the

system. No particular protection. Thats why only the root
user can load and unload modules.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

95/477

Module dependencies

Some kernel modules can depend on other modules, which

need to be loaded first.

Example: the usb-storage module depends on the

scsi_mod, libusual and usbcore modules.

Dependencies are described in

/lib/modules/<kernel-version>/modules.dep
This file is generated when you run make modules_install.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

96/477

Kernel log
When a new module is loaded, related information is available in
the kernel log.
I

The kernel keeps its messages in a circular buffer (so that it

doesnt consume more memory with many messages)

Kernel log messages are available through the dmesg

command (diagnostic message)

Kernel log messages are also displayed in the system console

(console messages can be filtered by level using the loglevel
kernel parameter, or completely disabled with the quiet
parameter).

Note that you can write to the kernel log from user space too:
echo "<n>Debug info" > /dev/kmsg

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

97/477

Module utilities (1)

modinfo <module_name>
modinfo <module_path>.ko
Gets information about a module: parameters, license,
description and dependencies.
Very useful before deciding to load a module or not.

sudo insmod <module_path>.ko

Tries to load the given module. The full path to the module
object file must be given.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

98/477

Understanding module loading issues

When loading a module fails, insmod often doesnt give you

enough details!

Details are often available in the kernel log.

Example:
$ sudo insmod ./intr_monitor.ko
insmod: error inserting './intr_monitor.ko': -1 Device or resource busy
$ dmesg
[17549774.552000] Failed to register handler for irq channel 2

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

99/477

Module utilities (2)

sudo modprobe <module_name>

Most common usage of modprobe: tries to load all the
modules the given module depends on, and then this module.
Lots of other options are available. modprobe automatically
looks in /lib/modules/<version>/ for the object file
corresponding to the given module name.

lsmod
Displays the list of loaded modules
Compare its output with the contents of /proc/modules!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

100/477

Module utilities (3)

sudo rmmod <module_name>

Tries to remove the given module.
Will only be allowed if the module is no longer in use (for
example, no more processes opening a device file)

sudo modprobe -r <module_name>

Tries to remove the given module and all dependent modules
(which are no longer needed after removing the module)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

101/477

Passing parameters to modules

I

Find available parameters:

modinfo snd-intel8x0m

Through insmod:
sudo insmod ./snd-intel8x0m.ko index=-2

Through modprobe:
Set parameters in /etc/modprobe.conf or in any file in
/etc/modprobe.d/:
options snd-intel8x0m index=-2

Through the kernel command line, when the driver is built

statically into the kernel:
snd-intel8x0m.index=-2
I
I
I

snd-intel8x0m is the driver name

index is the driver parameter name
-2 is the driver parameter value

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

102/477

Check module parameter values

How to find the current values for the parameters of a loaded

module?
I

Check /sys/module/<name>/parameters.

There is one file per parameter, containing the parameter

value.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

103/477

Useful reading
Linux Kernel in a Nutshell, Dec 2006
I

By Greg Kroah-Hartman, OReilly

http://www.kroah.com/lkn/

A good reference book and guide on

configuring, compiling and managing the
Linux kernel sources.

Freely available on-line!

Great companion to the printed book for
easy electronic searches!
Available as single PDF file on
http://free-electrons.com/
community/kernel/lkn/

Our rating: 2 stars

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

104/477

Developing Kernel Modules

Developing Kernel
Modules

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

105/477

Hello Module 1/2

/* hello.c */
#include <linux/init.h>
#include <linux/module.h>
#include <linux/kernel.h>
static int __init hello_init(void)
{
pr_alert("Good morrow to this fair assembly.\n");
return 0;
}
static void __exit hello_exit(void)
{
pr_alert("Alas, poor world, what treasure hast thou lost!\n");
}
module_init(hello_init);
module_exit(hello_exit);
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("Greeting module");
MODULE_AUTHOR("William Shakespeare");

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

106/477

Hello Module 2/2

__init
I

removed after initialization (static kernel or module.)

__exit
discarded when module compiled statically into the kernel, or
when module unloading support is not enabled.

Example available on
http://git.free-electrons.com/trainingmaterials/plain/code/hello/hello.c

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

107/477

Hello Module Explanations

I

Headers specific to the Linux kernel: linux/xxx.h

I

An initialization function
I

Called when the module is loaded, returns an error code (0 on

success, negative value on failure)
Declared by the module_init() macro: the name of the
function doesnt matter, even though <modulename>_init()
is a convention.

A cleanup function
I
I

No access to the usual C library, were doing kernel

programming

Called when the module is unloaded

Declared by the module_exit() macro.

Metadata information declared using MODULE_LICENSE(),

MODULE_DESCRIPTION() and MODULE_AUTHOR()

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

108/477

Symbols Exported to Modules 1/2

From a kernel module, only a limited number of kernel

functions can be called

Functions and variables have to be explicitly exported by the

kernel to be visible to a kernel module
Two macros are used in the kernel to export functions and
variables:

EXPORT_SYMBOL(symbolname), which exports a function or

variable to all modules
EXPORT_SYMBOL_GPL(symbolname), which exports a function
or variable only to GPL modules

A normal driver should not need any non-exported function.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

109/477

Symbols exported to modules 2/2

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

110/477

Module License
I

Several usages
I

Used to restrict the kernel functions that the module can use if
it isnt a GPL licensed module
I

Difference between EXPORT_SYMBOL() and

EXPORT_SYMBOL_GPL()

Used by kernel developers to identify issues coming from

proprietary drivers, which they cant do anything about
(Tainted kernel notice in kernel crashes and oopses).
Useful for users to check that their system is 100% free (check
/proc/sys/kernel/tainted)

Values
I

GPL compatible (see include/linux/license.h: GPL,

GPL v2, GPL and additional rights, Dual MIT/GPL,
Dual BSD/GPL, Dual MPL/GPL)
Proprietary

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

111/477

Compiling a Module

Two solutions
I

Out of tree
I

When the code is outside of the kernel source tree, in a

different directory
Advantage: Might be easier to handle than modifications to
the kernel itself
Drawbacks: Not integrated to the kernel
configuration/compilation process, needs to be built
separately, the driver cannot be built statically

Inside the kernel tree

I

Well integrated into the kernel configuration/compilation

process
Driver can be built statically if needed

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

112/477

Compiling an out-of-tree Module 1/2

I

I
I

The below Makefile should be reusable for any single-file

out-of-tree Linux module
The source file is hello.c
Just run make to build the hello.ko file

ifneq ($(KERNELRELEASE),)
obj-m := hello.o
else
KDIR := /path/to/kernel/sources
all:
<tab>$(MAKE) -C $(KDIR) M=$$PWD
endif
I

For KDIR, you can either set:

I

full kernel source directory

(configured + make modules_prepare)
or just kernel headers directory (make headers_install)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

113/477

Compiling an out-of-tree Module 2/2

The module Makefile is interpreted with KERNELRELEASE

undefined, so it calls the kernel Makefile, passing the module
directory in the M variable
The kernel Makefile knows how to compile a module, and
thanks to the M variable, knows where the Makefile for our
module is. The module Makefile is interpreted with
KERNELRELEASE defined, so the kernel sees the obj-m
definition.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

114/477

Modules and Kernel Version

To be compiled, a kernel module needs access to the kernel

headers, containing the definitions of functions, types and
constants.
Two solutions
I
I

The sources or headers must be configured

I

Full kernel sources

Only kernel headers (linux-headers-* packages in
Debian/Ubuntu distributions)
Many macros or functions depend on the configuration

A kernel module compiled against version X of kernel headers

will not load in kernel version Y
I

modprobe / insmod will say Invalid module format

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

115/477

New Driver in Kernel Sources 1/2

To add a new driver to the kernel sources:

I

Add your new source file to the appropriate source directory.

Example: drivers/usb/serial/navman.c
Single file drivers in the common case, even if the file is several
thousand lines of code big. Only really big drivers are split in
several files or have their own directory.
Describe the configuration interface for your new driver by
adding the following lines to the Kconfig file in this directory:

config USB_SERIAL_NAVMAN
tristate "USB Navman GPS device"
depends on USB_SERIAL
help
To compile this driver as a module, choose M
here: the module will be called navman.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

116/477

New Driver in Kernel Sources 2/2

Add a line in the Makefile file based on the Kconfig setting:

obj-$(CONFIG_USB_SERIAL_NAVMAN) += navman.o
It tells the kernel build system to build navman.c when the
USB_SERIAL_NAVMAN option is enabled. It works both if
compiled statically or as a module.
I
I
I

Run make xconfig and see your new options!

Run make and your new files are compiled!
See Documentation/kbuild/ for details and more elaborate
examples like drivers with several source files, or drivers in their
own subdirectory, etc.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

117/477

Hello Module with Parameters 1/2

/* hello_param.c */
#include <linux/init.h>
#include <linux/module.h>
MODULE_LICENSE("GPL");
/* A couple of parameters that can be passed in: how many
times we say hello, and to whom */
static char *whom = "world";
module_param(whom, charp, 0);
static int howmany = 1;
module_param(howmany, int, 0);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

118/477

Hello Module with Parameters 2/2

static int __init hello_init(void)
{
int i;
for (i = 0; i < howmany; i++)
pr_alert("(%d) Hello, %s\n", i, whom);
return 0;
}
static void __exit hello_exit(void)
{
pr_alert("Goodbye, cruel %s\n", whom);
}
module_init(hello_init);
module_exit(hello_exit);

Thanks to Jonathan Corbet for the example!

Source code available on:
http://git.free-electrons.com/trainingmaterials/plain/code/hello-param/hello_param.c
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

119/477

Declaring a module parameter

module_param(
name, /* name of an already defined variable */
type, /* either byte, short, ushort, int, uint, long, ulong,
charp, bool or invbool. (checked at run time!) */
perm /* for /sys/module/<module_name>/parameters/<param>,
0: no such module parameter value file */
);
/* Example */
static int irq=5;
module_param(irq, int, S_IRUGO);

Modules parameter arrays are also possible with

module_param_array().

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

120/477

Practical lab - Writing Modules

Create, compile and load your first

module

Add module parameters

Access kernel internals from your

module

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

121/477

Useful general-purpose kernel APIs

Useful
general-purpose
kernel APIs

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

122/477

Memory/string utilities

In linux/string.h
I

I
I

Memory-related: memset(), memcpy(), memmove(),

memscan(), memcmp(), memchr()
String-related: strcpy(), strcat(), strcmp(), strchr(),
strrchr(), strlen() and variants
Allocate and copy a string: kstrdup(), kstrndup()
Allocate and copy a memory area: kmemdup()

In linux/kernel.h
I

String to int conversion: simple_strtoul(),

simple_strtol(), simple_strtoull(),
simple_strtoll()
Other string functions: sprintf(), sscanf()

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

123/477

Linked lists
I

Convenient linked-list facility in linux/list.h

I

Used in thousands of places in the kernel

Add a struct list_head member to the structure whose

instances will be part of the linked list. It is usually named
node when each instance needs to only be part of a single list.

Define the list with the LIST_HEAD() macro for a global list,
or define a struct list_head element and initialize it with
INIT_LIST_HEAD() for lists embedded in a structure.
Then use the list_*() API to manipulate the list

I
I

I
I

Add elements: list_add(), list_add_tail()

Remove, move or replace elements: list_del(),
list_move(), list_move_tail(), list_replace()
Test the list: list_empty()
Iterate over the list: list_for_each_*() family of macros

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

124/477

Linked Lists Examples (1)

From include/linux/atmel_tc.h
/*
* Definition of a list element, with a
* struct list_head member
*/
struct atmel_tc
{
/* some members */
struct list_head node;
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

125/477

Linked Lists Examples (2)

I

From drivers/misc/atmel_tclib.c

/* Define the global list */

static LIST_HEAD(tc_list);
static int __init tc_probe(struct platform_device *pdev) {
struct atmel_tc *tc;
tc = kzalloc(sizeof(struct atmel_tc), GFP_KERNEL);
/* Add an element to the list */
list_add_tail(&tc->node, &tc_list);
}
struct atmel_tc *atmel_tc_alloc(unsigned block, const char *name)
{
struct atmel_tc *tc;
/* Iterate over the list elements */
list_for_each_entry(tc, &tc_list, node) {
/* Do something with tc */
}
[...]
}
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

126/477

Linux device and driver model

Linux device and

driver model

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

127/477

Linux device and driver model

Introduction

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

128/477

The need for a device model?

The Linux kernel runs on a wide range of architectures and

hardware platforms, and therefore needs to maximize the
reusability of code between platforms.

For example, we want the same USB device driver to be

usable on a x86 PC, or an ARM platform, even though the
USB controllers used on these platforms are different.

This requires a clean organization of the code, with the device

drivers separated from the controller drivers, the hardware
description separated from the drivers themselves, etc.

This is what the Linux kernel Device Model allows, in

addition to other advantages covered in this section.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

129/477

Kernel and Device Drivers

In Linux, a driver is always
interfacing with:
I

a framework that allows the

driver to expose the
hardware features in a
generic way.

a bus infrastructure, part

of the device model, to
detect/communicate with
the hardware.

This section focuses on the

device model, while kernel
frameworks are covered later in
this training.
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

130/477

Device Model data structures

I

The device model is organized around three main data

structures:
I

The struct bus_type structure, which represent one type of

bus (USB, PCI, I2C, etc.)
The struct device_driver structure, which represents one
driver capable of handling certain devices on a certain bus.
The struct device structure, which represents one device
connected to a bus

The kernel uses inheritance to create more specialized versions

of struct device_driver and struct device for each bus
subsystem.
In order to explore the device model, we will
I

First look at a popular bus that offers dynamic enumeration,

the USB bus
Continue by studying how buses that do not offer dynamic
enumerations are handled.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

131/477

Bus Drivers
I

The first component of the device model is the bus driver

I

One bus driver for each type of bus: USB, PCI, SPI, MMC,
I2C, etc.

It is responsible for
I
I

I
I

Registering the bus type (struct bus_type)

Allowing the registration of adapter drivers (USB controllers,
I2C adapters, etc.), able to detect the connected devices, and
providing a communication mechanism with the devices
Allowing the registration of device drivers (USB devices, I2C
devices, PCI devices, etc.), managing the devices
Matching the device drivers against the devices detected by
the adapter drivers.
Provides an API to both adapter drivers and device drivers
Defining driver and device specific structures, mainly
struct usb_driver and struct usb_interface

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

132/477

Linux device and driver model

Example of the USB bus

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

133/477

Example: USB Bus 1/2

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

134/477

Example: USB Bus 2/2

Core infrastructure (bus driver)

I
I

Adapter drivers
I
I

drivers/usb/core
struct bus_type is defined in
drivers/usb/core/driver.c and registered in
drivers/usb/core/usb.c
drivers/usb/host
For EHCI, UHCI, OHCI, XHCI, and their implementations on
various systems (Atmel, IXP, Xilinx, OMAP, Samsung, PXA,
etc.)

Device drivers
I

Everywhere in the kernel tree, classified by their type

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

135/477

Example of Device Driver

To illustrate how drivers are implemented to work with the

device model, we will study the source code of a driver for a
USB network card
I
I
I

It is USB device, so it has to be a USB device driver

It is a network device, so it has to be a network device
Most drivers rely on a bus infrastructure (here, USB) and
register themselves in a framework (here, network)

We will only look at the device driver side, and not the
adapter driver side

The driver we will look at is drivers/net/usb/rtl8150.c

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

136/477

Device Identifiers
I

Defines the set of devices that this driver can manage, so that
the USB core knows for which devices this driver should be
used
The MODULE_DEVICE_TABLE() macro allows depmod to
extract at compile time the relation between device identifiers
and drivers, so that drivers can be loaded automatically by
udev. See
/lib/modules/$(uname -r)/modules.{alias,usbmap}

static struct usb_device_id rtl8150_table[] = {

{ USB_DEVICE(VENDOR_ID_REALTEK, PRODUCT_ID_RTL8150) },
{ USB_DEVICE(VENDOR_ID_MELCO, PRODUCT_ID_LUAKTX) },
{ USB_DEVICE(VENDOR_ID_MICRONET, PRODUCT_ID_SP128AR) },
{ USB_DEVICE(VENDOR_ID_LONGSHINE, PRODUCT_ID_LCS8138TX) },
[...]
{}
};
MODULE_DEVICE_TABLE(usb, rtl8150_table);
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

137/477

Instantiation of usb driver

I

struct usb_driver is a structure defined by the USB core.

Each USB device driver must instantiate it, and register itself
to the USB core using this structure

This structure inherits from struct device_driver, which

is defined by the device model.

static struct usb_driver rtl8150_driver = {

.name = "rtl8150",
.probe = rtl8150_probe,
.disconnect = rtl8150_disconnect,
.id_table = rtl8150_table,
.suspend = rtl8150_suspend,
.resume = rtl8150_resume
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

138/477

Driver (Un)Registration
I

When the driver is loaded or unloaded, it must register or

unregister itself from the USB core

Done using usb_register() and usb_deregister(),

provided by the USB core.

static int __init usb_rtl8150_init(void)

{
return usb_register(&rtl8150_driver);
}
static void __exit usb_rtl8150_exit(void)
{
usb_deregister(&rtl8150_driver);
}
module_init(usb_rtl8150_init);
module_exit(usb_rtl8150_exit);
I

Note: this code has now been replaced by a shorter

module_usb_driver() macro call.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

139/477

At Initialization
I

The USB adapter driver that corresponds to the USB

controller of the system registers itself to the USB core

The rtl8150 USB device driver registers itself to the USB

core

The USB core now knows the association between the

vendor/product IDs of rtl8150 and the struct usb_driver
structure of this driver

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

140/477

When a Device is Detected

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

141/477

Probe Method

The probe() method receives as argument a structure

describing the device, usually specialized by the bus
infrastructure (struct pci_dev, struct usb_interface,
etc.)
This function is responsible for
I

Initializing the device, mapping I/O memory, registering the

interrupt handlers. The bus infrastructure provides methods to
get the addresses, interrupt numbers and other device-specific
information.
Registering the device to the proper kernel framework, for
example the network infrastructure.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

142/477

Probe Method Example

static int rtl8150_probe(struct usb_interface *intf,
const struct usb_device_id *id)
{
rtl8150_t *dev;
struct net_device *netdev;
netdev = alloc_etherdev(sizeof(rtl8150_t));
[...]
dev = netdev_priv(netdev);
tasklet_init(&dev->tl, rx_fixup, (unsigned long)dev);
spin_lock_init(&dev->rx_pool_lock);
[...]
netdev->netdev_ops = &rtl8150_netdev_ops;
alloc_all_urbs(dev);
[...]
usb_set_intfdata(intf, dev);
SET_NETDEV_DEV(netdev, &intf->dev);
register_netdev(netdev);
return 0;
}
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

143/477

The Model is Recursive

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

144/477

Linux device and driver model

Platform drivers

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

145/477

Non-discoverable buses

On embedded systems, devices are often not connected

through a bus allowing enumeration, hotplugging, and
providing unique identifiers for devices.

For example, the devices on I2C buses or SPI buses, or the

devices directly part of the system-on-chip.

However, we still want all of these devices to be part of the

device model.
Such devices, instead of being dynamically detected, must be
statically described in either:

I
I

The kernel source code

The Device Tree, a hardware description file used on some
architectures.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

146/477

Platform devices

Amongst the non-discoverable devices, a huge family are the

devices that are directly part of a system-on-chip: UART
controllers, Ethernet controllers, SPI or I2C controllers,
graphic or audio devices, etc.

In the Linux kernel, a special bus, called the platform bus has
been created to handle such devices.

It supports platform drivers that handle platform devices.

It works like any other bus (USB, PCI), except that devices are
enumerated statically instead of being discovered dynamically.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

147/477

Implementation of a Platform Driver

I

The driver implements a struct platform_driver structure

(example taken from drivers/serial/imx.c)

static struct platform_driver serial_imx_driver = {

.probe = serial_imx_probe,
.remove = serial_imx_remove,
.driver = {
.name = "imx-uart",
.owner = THIS_MODULE,
},
};
I

And registers its driver to the platform driver infrastructure

static int __init imx_serial_init(void) {

ret = platform_driver_register(&serial_imx_driver);
}
static void __exit imx_serial_cleanup(void) {
platform_driver_unregister(&serial_imx_driver);
}
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

148/477

Platform Device Instantiation: old style (1/2)

I

As platform devices cannot be detected dynamically, they are

defined statically
I

By direct instantiation of struct platform_device

structures, as done on some ARM platforms. Definition done
in the board-specific or SoC specific code.
By using a device tree, as done on Power PC (and on some
ARM platforms) from which struct platform_device
structures are created

Example on ARM, where the instantiation is done in

arch/arm/mach-imx/mx1ads.c

static struct platform_device imx_uart1_device = {

.name = "imx-uart",
.id = 0,
.num_resources = ARRAY_SIZE(imx_uart1_resources),
.resource = imx_uart1_resources,
.dev = {
.platform_data = &uart_pdata,
}
};
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

149/477

Platform device instantiation: old style (2/2)

I

The device is part of a list

static struct platform_device *devices[] __initdata = {

&cs89x0_device,
&imx_uart1_device,
&imx_uart2_device,
};
I

And the list of devices is added to the system during board

initialization

static void __init mx1ads_init(void)

{
[...]
platform_add_devices(devices, ARRAY_SIZE(devices));
}
MACHINE_START(MX1ADS, "Freescale MX1ADS")
[...]
.init_machine = mx1ads_init,
MACHINE_END
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

150/477

The Resource Mechanism

Each device managed by a particular driver typically uses

different hardware resources: addresses for the I/O registers,
DMA channels, IRQ lines, etc.

Such information can be represented using

struct resource, and an array of struct resource is
associated to a struct platform_device

Allows a driver to be instantiated for multiple devices

functioning similarly, but with different addresses, IRQs, etc.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

151/477

Declaring resources

static struct resource imx_uart1_resources[] = {

[0] = {
.start = 0x00206000,
.end = 0x002060FF,
.flags = IORESOURCE_MEM,
},
[1] = {
.start = (UART1_MINT_RX),
.end = (UART1_MINT_RX),
.flags = IORESOURCE_IRQ,
},
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

152/477

Using Resources

When a struct platform_device is added to the system

using platform_add_device(), the probe() method of the
platform driver gets called

This method is responsible for initializing the hardware,

registering the device to the proper framework (in our case,
the serial driver framework)

The platform driver has access to the I/O resources:

res = platform_get_resource(pdev, IORESOURCE_MEM, 0);

base = ioremap(res->start, PAGE_SIZE);
sport->rxirq = platform_get_irq(pdev, 0);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

153/477

platform data Mechanism

In addition to the well-defined resources, many drivers require

driver-specific information for each platform device

Such information can be passed using the platform_data

field of struct device (from which
struct platform_device inherits)
As it is a void * pointer, it can be used to pass any type of
information.

Typically, each driver defines a structure to pass information

through struct platform_data

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

154/477

platform data example 1/2

I

The i.MX serial port driver defines the following structure to

be passed through struct platform_data

struct imxuart_platform_data {
int (*init)(struct platform_device *pdev);
void (*exit)(struct platform_device *pdev);
unsigned int flags;
void (*irda_enable)(int enable);
unsigned int irda_inv_rx:1;
unsigned int irda_inv_tx:1;
unsigned short transceiver_delay;
};
I

The MX1ADS board code instantiates such a structure

static struct imxuart_platform_data uart1_pdata = {

.flags = IMXUART_HAVE_RTSCTS,
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

155/477

platform data Example 2/2

I

The uart_pdata structure is associated to the

struct platform_device structure in the MX1ADS board
file (the real code is slightly more complicated)

struct platform_device mx1ads_uart1 = {

.name = "imx-uart",
.dev {
.platform_data = &uart1_pdata,
},
.resource = imx_uart1_resources,
[...]
};
I

The driver can access the platform data:

static int serial_imx_probe(struct platform_device *pdev)

{
struct imxuart_platform_data *pdata;
pdata = pdev->dev.platform_data;
if (pdata && (pdata->flags & IMXUART_HAVE_RTSCTS))
sport->have_rtscts = 1;
[...]
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

156/477

Device Tree
I

On many embedded architectures, manual instantiation of

platform devices was considered to be too verbose and not
easily maintainable.

Such architectures are moving, or have moved, to use the

Device Tree.

It is a tree of nodes that models the hierarchy of devices in

the system, from the devices inside the processor to the
devices on the board.

Each node can have a number of properties describing various

properties of the devices: addresses, interrupts, clocks, etc.

At boot time, the kernel is given a compiled version, the

Device Tree Blob, which is parsed to instantiate all the
devices described in the DT.

On ARM, they are located in arch/arm/boot/dts.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

157/477

Device Tree example

uart0: serial@44e09000 {
compatible = "ti,omap3-uart";
ti,hwmods = "uart1";
clock-frequency = <48000000>;
reg = <0x44e09000 0x2000>;
interrupts = <72>;
status = "disabled";
};
I

serial@44e09000 is the node name

uart0 is an alias, that can be referred to in other parts of the

DT as &uart0

other lines are properties. Their values are usually strings, list
of integers, or references to other nodes.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

158/477

Device Tree inheritance (1/2)

I

Each particular hardware platform has its own device tree.

However, several hardware platforms use the same processor,

and often various processors in the same family share a
number of similarities.
To allow this, a device tree file can include another one. The
trees described by the including file overlays the tree described
by the included file. This can be done:

Either by using the /include/ statement provided by the

Device Tree language.
Either by using the #include statement, which requires
calling the C preprocessor before parsing the Device Tree.

Linux currently uses either one technique or the other,

(different from one ARM subarchitecture to another, for
example).

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

159/477

Device Tree inheritance (2/2)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

160/477

Device Tree: compatible string

I

With the device tree, a device is bound with the

corresponding driver using the compatible string.

The of_match_table field of struct device_driver lists

the compatible strings supported by the driver.

#if defined(CONFIG_OF)
static const struct of_device_id omap_serial_of_match[] = {
{ .compatible = "ti,omap2-uart" },
{ .compatible = "ti,omap3-uart" },
{ .compatible = "ti,omap4-uart" },
{},
};
MODULE_DEVICE_TABLE(of, omap_serial_of_match);
#endif
static struct platform_driver serial_omap_driver = {
.probe
= serial_omap_probe,
.remove
= serial_omap_remove,
.driver
= {
.name
= DRIVER_NAME,
.pm
= &serial_omap_dev_pm_ops,
.of_match_table = of_match_ptr(omap_serial_of_match),
},
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

161/477

Device Tree Resources

The drivers will use the same mechanism that we saw

previously to retrieve basic information: interrupts numbers,
physical addresses, etc.

The available resources list will be built up by the kernel at

boot time from the device tree, so that you dont need to
make any unnecessary lookups to the DT when loading your
driver.

Any additional information will be specific to a driver or the

class it belongs to, defining the bindings

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

162/477

Device Tree bindings

I

The compatible string and the associated properties define

what is called a device tree binding.

Device tree bindings are all documented in

Documentation/devicetree/bindings .
Since the Device Tree is normally part of the kernel ABI, the
bindings must remain compatible over-time.

I
I

A new kernel must be capable of using an old Device Tree.

This requires a very careful design of the bindings. They are all
reviewed on the devicetree@vger.kernel.org mailing list.

A Device Tree binding should contain only a description of the

hardware and not configuration.
I

An interrupt number can be part of the Device Tree as it

describes the hardware.
But not whether DMA should be used for a device or not, as it
is a configuration choice.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

163/477

sysfs

The bus, device, drivers, etc. structures are internal to the

kernel

The sysfs virtual filesystem offers a mechanism to export

such information to user space

Used for example by udev to provide automatic module

loading, firmware loading, device file creation, etc.
sysfs is usually mounted in /sys

I
I
I

/sys/bus/ contains the list of buses

/sys/devices/ contains the list of devices
/sys/class enumerates devices by class (net, input,
block...), whatever the bus they are connected to. Very useful!

Take your time to explore /sys on your workstation.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

164/477

References

Device Tree for Dummies, Thomas Petazzoni (Apr. 2014):

http://j.mp/1jQU6NR
Kernel documentation
I
I
I

Documentation/driver-model/
Documentation/devicetree/
Documentation/filesystems/sysfs.txt

The kernel source code

I

Full of examples of other drivers!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

165/477

Introduction to the I2C subsystem

Introduction to
the I2C subsystem

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

166/477

What is I2C?

A very commonly used low-speed bus to connect on-board

devices to the processor.

Uses only two wires: SDA for the data, SCL for the clock.

It is a master/slave bus: only the master can initiate

transactions, and slaves can only reply to transactions
initiated by masters.

In a Linux system, the I2C controller embedded in the

processor is typically the master, controlling the bus.

Each slave device is identified by a unique I2C address. Each

transaction initiated by the master contains this address,
which allows the relevant slave to recognize that it should
reply to this particular transaction.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

167/477

An I2C bus example

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

168/477

The I2C subsystem

Like all bus subsystems, the I2C subsystem is responsible for:

I
I

Providing an API to implement I2C controller drivers

Providing an API to implement I2C device drivers, in kernel
space
Providing an API to implement I2C device drivers, in user
space

The core of the I2C subsystem is located in drivers/i2c.

The I2C controller drivers are located in

drivers/i2c/busses.

The I2C device drivers are located throughout drivers/,

depending on the type of device (ex: drivers/input for
input devices).

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

169/477

Registering an I2C device driver

I

Like all bus subsystems, the I2C subsystem defines a

struct i2c_driver that inherits from
struct device_driver, and which must be instantiated and
registered by each I2C device driver.
I

As usual, this structure points to the ->probe() and

->remove() functions.
It also contains an id_table field that must point to a list of
device IDs (which is a list of tuples containing a string and
some private driver data). It is used for non-DT based probing
of I2C devices.

The i2c_add_driver() and i2c_del_driver() functions

are used to register/unregister the driver.

If the driver doesnt do anything else in its init()/exit()

functions, it is advised to use the module_i2c_driver()
macro instead.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

170/477

Registering an I2C device driver: example

static const struct i2c_device_id <driver>_id[] = {
{ "<device-name>", 0 },
{ }
};
MODULE_DEVICE_TABLE(i2c, <driver>_id);
#ifdef CONFIG_OF
static const struct of_device_id <driver>_dt_ids[] = {
{ .compatible = "<vendor>,<device-name>", },
{ }
};
MODULE_DEVICE_TABLE(of, <driver>_dt_ids);
#endif
static struct i2c_driver <driver>_driver = {
.probe
= <driver>_probe,
.remove
= <driver>_remove,
.id_table
= <driver>_id,
.driver = {
.name
= "<driver-name>",
.owner = THIS_MODULE,
.of_match_table = of_match_ptr(<driver>_dt_ids),
},
};
module_i2c_driver(<driver>_driver);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

171/477

Registering an I2C device: non-DT

On non-DT platforms, the struct i2c_board_info

structure allows to describe how an I2C device is connected to
a board.
Such structures are normally defined with the
I2C_BOARD_INFO() helper macro.
I

Takes as argument the device name and the slave address of

the device on the bus.

An array of such structures is registed on a per-bus basis using

i2c_register_board_info(), when the platform is
initialized.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

172/477

Registering an I2C device, non-DT example

static struct i2c_board_info <board>_i2c_devices[] __initdata = {

{
I2C_BOARD_INFO("cs42l51", 0x4a),
},
};
void board_init(void)
{
/*
* Here should be the registration of all devices, including
* the I2C controller device.
*/
i2c_register_board_info(0, <board>_i2c_devices,
ARRAY_SIZE(<board>_i2c_devices));
/* More devices registered here */
}

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

173/477

Registering an I2C device, in the DT

In the Device Tree, the I2C controller device is typically

defined in the .dtsi file that describes the processor.
I

Normally defined with status = "disabled".

At the board/platform level:

I
I

the I2C controller device is enabled (status = "okay")

the I2C bus frequency is defined, using the clock-frequency
property.
the I2C devices on the bus are described as children of the I2C
controller node, where the reg property gives the I2C slave
address on the bus.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

174/477

Registering an I2C device, DT example (1/2)

Definition of the I2C controller, .dtsi file

i2c@7000c000 {
compatible = "nvidia,tegra20-i2c";
reg = <0x7000c000 0x100>;
interrupts = <GIC_SPI 38 IRQ_TYPE_LEVEL_HIGH>;
#address-cells = <1>;
#size-cells = <0>;
clocks = <&tegra_car TEGRA20_CLK_I2C1>,
<&tegra_car TEGRA20_CLK_PLL_P_OUT3>;
clock-names = "div-clk", "fast-clk";
status = "disabled";
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

175/477

Registering an I2C device, DT example (2/2)

Definition of the I2C device, .dts file

i2c@7000c000 {
status = "okay";
clock-frequency = <400000>;
alc5632: alc5632@1e {
compatible = "realtek,alc5632";
reg = <0x1e>;
gpio-controller;
#gpio-cells = <2>;
};
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

176/477

probe() and remove()

The ->probe() function is responsible for initializing the

device and registering it in the appropriate kernel framework.
It receives as argument:
I

A struct i2c_client pointer, which represents the I2C

device itself. This structure inherits from struct device.
A struct i2c_device_id pointer, which points to the I2C
device ID entry that matched the device that is being probed.

The ->remove() function is responsible for unregistering the

device from the kernel framework and shut it down. It receives
as argument:
I

The same struct i2c_client pointer that was passed as

argument to ->probe()

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

177/477

Probe/remove example
static int <driver>_probe(struct i2c_client *client,
const struct i2c_device_id *id)
{
/* initialize device */
/* register to a kernel framework */
i2c_set_clientdata(client, <private data>);
return 0;
}
static int <driver>_remove(struct i2c_client *client)
{
<private data> = i2c_get_clientdata(client);
/* unregister device from kernel framework */
/* shut down the device */
return 0;
}

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

178/477

Practical lab - Linux device model for an I2C driver

Modify the Device Tree to

instantiate an I2C device.

Implement a driver that registers as

an I2C driver.

Make sure that the probe/remove

functions are called when there is a
device/driver match.

Explore the sysfs entries related to

your driver and device.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

179/477

Communicating with the I2C device: raw API

The most basic API to communicate with the I2C device provides
functions to either send or receive data:
I

int i2c_master_send(struct i2c_client *client,

const char *buf, int count);
Sends the contents of buf to the client.

int i2c_master_recv(struct i2c_client *client,

char *buf, int count);
Receives count bytes from the client, and store them into
buf.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

180/477

Communicating with the I2C device: message transfer

The message transfer API allows to describe transfers that

consists of several messages, with each message being a
transaction in one direction:
I

int i2c_transfer(struct i2c_adapter *adap, struct

i2c_msg *msg, int num);

The struct i2c_adapter pointer can be found by using

client->adapter

The struct i2c_msg structure defines the length, location,

and direction of the message.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

181/477

I2C: message transfer example

struct i2c_msg msg[2];
int error;
u8 start_reg;
u8 buf[10];
msg[0].addr = client->addr;
msg[0].flags = 0;
msg[0].len = 1;
msg[0].buf = &start_reg;
start_reg = 0x10;
msg[1].addr = client->addr;
msg[1].flags = I2C_M_RD;
msg[1].len = sizeof(buf);
msg[1].buf = buf;
error = i2c_transfer(client->adapter, msg, 2);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

182/477

SMBus calls
I

SMBus is a subset of the I2C protocol.

It defines a standard set of transactions, for example to read

or write a register into a device.

Linux provides SMBus functions that should be used instead

of the raw API, if the I2C device supports this standard type of
transactions. The driver can then be used on both SMBus and
I2C adapters (cant use I2C commands on SMBus adapters).
Example: the i2c_smbus_read_byte_data() function allows
to read one byte of data from a device register.

It does the following operations:

S Addr Wr [A] Comm [A] S Addr Rd [A] [Data] NA P
Which means it first writes a one byte data command
(Comm), and then reads back one byte of data ([Data]).

See Documentation/i2c/smbus-protocol for details.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

183/477

List of SMBus functions

I

Read/write one byte

I s32 i2c_smbus_read_byte(const struct i2c_client *client);
I s32 i2c_smbus_write_byte(const struct i2c_client *client, u8 value);

Write a command byte, and read or write one byte

I s32 i2c_smbus_read_byte_data(const struct i2c_client *client, u8 command);
I s32 i2c_smbus_write_byte_data(const struct i2c_client *client, u8 command, u8
value);

Write a command byte, and read or write one word

I s32 i2c_smbus_read_word_data(const struct i2c_client *client, u8 command);
I s32 i2c_smbus_write_word_data(const struct i2c_client *client, u8 command, u16
value);

Write a command byte, and read or write a block of data

(max 32 bytes)
I s32 i2c_smbus_read_block_data(const struct i2c_client *client, u8 command, u8
*values);
I s32 i2c_smbus_write_block_data(const struct i2c_client *client, u8 command, u8
length, const u8 *values);

Write a command byte, and read or write a block of data (no

limit)
I s32 i2c_smbus_read_i2c_block_data(const struct i2c_client *client, u8 command, u8
length, u8 *values);
I s32 i2c_smbus_write_i2c_block_data(const struct i2c_client *client, u8 command, u8
length, const u8 *values);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

184/477

I2C functionality
I

Not all I2C controllers support all functionalities.

The I2C controller drivers therefore tell the I2C core which
functionalities they support.

An I2C device driver must check that the functionalities they

need are provided by the I2C controller in use on the system.

The i2c_check_functionality() function allows to make

such a check.

Examples of functionalities: I2C_FUNC_I2C to be able to use

the raw I2C functions, I2C_FUNC_SMBUS_BYTE_DATA to be
able to use SMBus commands to write a command and
read/write one byte of data.

See include/uapi/linux/i2c.h for the full list of existing

functionalities.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

185/477

References

http://en.wikipedia.org/wiki/I2C, general presentation

of the I2C protocol
Documentation/i2c/ , details about the Linux support for
I2C
I
I
I
I
I

writing-clients, how to write I2C device drivers

instantiating-devices, how to instantiate devices
smbus-protocol, details on the SMBus functions
functionality, how the functionality mechanism works
and many more documentation files

http://free-electrons.com/pub/video/2012/elce/
elce-2012-anders-board-bringup-i2c.webm, excellent
talk: You, me and I2C from David Anders at ELCE 2012.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

186/477

Introduction to pin muxing

Introduction to pin
muxing

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

187/477

What is pin muxing?

Modern SoCs (System on Chip) include more and more

hardware blocks, many of which need to interface with the
outside world using pins.

However, the physical size of the chips remains small, and

therefore the number of available pins is limited.

For this reason, not all of the internal hardware block features
can be exposed on the pins simultaneously.

The pins are multiplexed: they expose either the functionality

of hardware block A or the functionality of hardware block B.

This multiplexing is usually software configurable.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

188/477

Pin muxing diagram

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

189/477

Pin muxing in the Linux kernel

I
I

Since Linux 3.2, a pinctrl subsystem has been added.

This subsystem, located in drivers/pinctrl provides a
generic subsystem to handle pin muxing. It offers:
I
I

A pin muxing consumer interface, for device drivers.

A pin muxing driver interface, to implement the
system-on-chip specific drivers that configure the muxing.

Most pinctrl drivers provide a Device Tree binding, and the

pin muxing must be described in the Device Tree.
I

The exact Device Tree binding depends on each driver. Each

binding is documented in
Documentation/devicetree/bindings/pinctrl .

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

190/477

pinctrl subsystem diagram

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

191/477

Device Tree binding for consumer devices

I

The devices that require certains pins to be muxed will use the
pinctrl-<x> and pinctrl-names Device Tree properties.
The pinctrl-0, pinctrl-1, pinctrl-<x> properties link to
a pin configuration for a given state of the device.
The pinctrl-names property associates a name to each
state. The name default is special, and is automatically
selected by a device driver, without having to make an explicit
pinctrl function call.
In most cases, the following is sufficient:
i2c@11000 {
pinctrl-0 = <&pmx_twsi0>;
pinctrl-names = "default";
...
};

See Documentation/devicetree/bindings/pinctrl/
pinctrl-bindings.txt for details.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

192/477

Defining pinctrl configurations

I

The different pinctrl configurations must be defined as child

nodes of the main pinctrl device (which controls the muxing
of pins).
The configurations may be defined at:
I

the SoC level (.dtsi file), for pin configurations that are often
shared between multiple boards
at the board level (.dts file) for configurations that are board
specific.

The pinctrl-<x> property of the consumer device points to

the pin configuration it needs through a DT phandle.

The description of the configurations is specific to each pinctrl

driver. See
Documentation/devicetree/bindings/pinctrl for the
DT bindings documentation.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

193/477

Example on OMAP/AM33xx
I

On OMAP/AM33xx, the
pinctrl-single driver is
used. It is common between
multiple SoCs and simply
allows to configure pins by
writing a value to a register.
I In each pin configuration,
a pinctrl-single,pins
value gives a list of
(register, value) pairs
needed to configure the
pins.
To know the correct values,
one must use the SoC and
board datasheets.

am33xx_pinmux: pinmux@44e10800 {
i2c0_pins: pinmux_i2c0_pins {
pinctrl-single,pins = <
/* i2c0_sda.i2c0_sda */
0x188 (PIN_INPUT_PULLUP | MUX_MODE0)
/* i2c0_scl.i2c0_scl */
0x18c (PIN_INPUT_PULLUP | MUX_MODE0)
>;
};
};
i2c0: i2c@44e0b000 {
pinctrl-names = "default";
pinctrl-0 = <&i2c0_pins>;
status = "okay";
clock-frequency = <400000>;
tps: tps@2d {
reg = <0x2d>;
};
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

194/477

Example on Allwinner SoC

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

195/477

Practical lab - Communicate with the Nunchuk

Configure the pinmuxing for the

I2C bus used to communicate with
the Nunchuk

Validate that the I2C

communication works with user
space tools.

Extend the I2C driver started in the

previous lab to communicate with
the Nunchuk.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

196/477

Kernel frameworks for device drivers

Kernel frameworks
for device drivers

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

197/477

Kernel and Device Drivers

In Linux, a driver is always
interfacing with:
I

a framework that allows the

driver to expose the
hardware features to user
space applications.

a bus infrastructure, part

of the device model, to
detect/communicate with
the hardware.

This section focuses on the

kernel frameworks, while the
device model was covered earlier
in this training.
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

198/477

Kernel frameworks for device drivers

User space vision of devices

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

199/477

Types of devices
Under Linux, there are essentially three types of devices:
I

Network devices. They are represented as network

interfaces, visible in user space using ifconfig.

Block devices. They are used to provide user space

applications access to raw storage devices (hard disks, USB
keys). They are visible to the applications as device files in
/dev.

Character devices. They are used to provide user space

applications access to all other types of devices (input, sound,
graphics, serial, etc.). They are also visible to the applications
as device files in /dev.

Most devices are character devices, so we will study these in

more details.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

200/477

Major and minor numbers

Within the kernel, all block and character devices are

identified using a major and a minor number.

The major number typically indicates the family of the device.

The minor number typically indicates the number of the

device (when they are for example several serial ports)

Most major and minor numbers are statically allocated, and

identical across all Linux systems.

They are defined in Documentation/devices.txt .

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

201/477

Devices: everything is a file

A very important Unix design decision was to represent most

of the system objects as files

It allows applications to manipulate all system objects with

the normal file API (open, read, write, close, etc.)

So, devices had to be represented as files to the applications

This is done through a special artifact called a device file

It is a special type of file, that associates a file name visible to

user space applications to the triplet (type, major, minor) that
the kernel understands

All device files are by convention stored in the /dev directory

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

202/477

Device files examples

Example of device files in a Linux system
$ ls -l /dev/ttyS0 /dev/tty1 /dev/sda1 /dev/sda2
brw-rw---- 1 root disk
8, 1 2011-05-27 08:56
brw-rw---- 1 root disk
8, 2 2011-05-27 08:56
crw------- 1 root root
4, 1 2011-05-27 08:57
crw-rw---- 1 root dialout 4, 64 2011-05-27 08:56
crw-rw-rw- 1 root root
1, 5 2011-05-27 08:56

/dev/zero
/dev/sda1
/dev/sda2
/dev/tty1
/dev/ttyS0
/dev/zero

Example C code that uses the usual file API to write data to a
serial port
int fd;
fd = open("/dev/ttyS0", O_RDWR);
write(fd, "Hello", 5);
close(fd);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

203/477

Creating device files

On a basic Linux system, the device files have to be created

manually using the mknod command
I
I
I

mknod /dev/<device> [c|b] major minor

Needs root privileges
Coherency between device files and devices handled by the
kernel is left to the system developer

On more elaborate Linux systems, mechanisms can be added

to create/remove them automatically when devices appear
and disappear
I
I

devtmpfs virtual filesystem

udev daemon, solution used by desktop and server Linux
systems
mdev program, a lighter solution than udev

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

204/477

Kernel frameworks for device drivers

Character drivers

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

205/477

A character driver in the kernel

From the point of view of an application, a character device is

essentially a file.

The driver of a character device must therefore implement

operations that let applications think the device is a file:
open, close, read, write, etc.

In order to achieve this, a character driver must implement

the operations described in the struct file_operations
structure and register them.

The Linux filesystem layer will ensure that the drivers

operations are called when a user space application makes the
corresponding system call.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

206/477

From user space to the kernel: character devices

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

207/477

File operations

Here are the most important operations for a character driver.

All of them are optional.

#include <linux/fs.h>
struct file_operations {
ssize_t (*read) (struct file *, char __user *,
size_t, loff_t *);
ssize_t (*write) (struct file *, const char __user *,
size_t, loff_t *);
long (*unlocked_ioctl) (struct file *, unsigned int,
unsigned long);
int (*mmap) (struct file *, struct vm_area_struct *);
int (*open) (struct inode *, struct file *);
int (*release) (struct inode *, struct file *);
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

208/477

open() and release()

I

int foo_open(struct inode *i, struct file *f)

I
I

Called when user space opens the device file.

struct inode is a structure that uniquely represents a file in
the system (be it a regular file, a directory, a symbolic link, a
character or block device)
struct file is a structure created every time a file is opened.
Several file structures can point to the same inode structure.
I

I
I

Contains information like the current position, the opening

mode, etc.
Has a void *private_data pointer that one can freely use.
A pointer to the file structure is passed to all other
operations

int foo_release(struct inode *i, struct file *f)

I

Called when user space closes the file.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

209/477

read()

ssize_t foo_read(struct file *f, char __user *buf,

size_t sz, loff_t *off)
I

I
I

Called when user space uses the read() system call on the
device.
Must read data from the device, write at most sz bytes in the
user space buffer buf, and update the current position in the
file off. f is a pointer to the same file structure that was
passed in the open() operation
Must return the number of bytes read.
On UNIX, read() operations typically block when there isnt
enough data to read from the device

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

210/477

write()

ssize_t foo_write(struct file *f,

const char __user *buf, size_t sz, loff_t *off)
I

Called when user space uses the write() system call on the
device
The opposite of read, must read at most sz bytes from buf,
write it to the device, update off and return the number of
bytes written.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

211/477

Exchanging data with user space 1/3

Kernel code isnt allowed to directly access user space

memory, using memcpy() or direct pointer dereferencing
I
I

Doing so does not work on some architectures

If the address passed by the application was invalid, the
application would segfault.

To keep the kernel code portable and have proper error

handling, your driver must use special kernel functions to
exchange data with user space.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

212/477

Exchanging data with user space 2/3

I

A single value
I

get_user(v, p);
I

put_user(v, p);
I

The kernel variable v gets the value pointed by the user space
pointer p
The value pointed by the user space pointer p is set to the
contents of the kernel variable v.

A buffer
I

unsigned long copy_to_user(void __user *to,

const void *from, unsigned long n);

unsigned long copy_from_user(void *to,

const void __user *from, unsigned long n);

The return value must be checked. Zero on success, non-zero

on failure. If non-zero, the convention is to return -EFAULT.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

213/477

Exchanging data with user space 3/3

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

214/477

Zero copy access to user memory

Having to copy data to or from an intermediate kernel buffer

can become expensive when the amount of data to transfer is
large (video).
Zero copy options are possible:
I

mmap() system call to allow user space to directly access

memory mapped I/O space. See our mmap() chapter.
get_user_pages_fast() to get a mapping to user pages
without having to copy them. See http://j.mp/1sML7lP
(Kernel API doc). This API is more complex to use though.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

215/477

unlocked ioctl()

long unlocked_ioctl(struct file *f,

unsigned int cmd, unsigned long arg)
I
I

I
I

Associated to the ioctl() system call.

Called unlocked because it didnt hold the Big Kernel Lock
(gone now).
Allows to extend the driver capabilities beyond the limited
read/write API.
For example: changing the speed of a serial port, setting video
output format, querying a device serial number...
cmd is a number identifying the operation to perform
arg is the optional argument passed as third argument of the
ioctl() system call. Can be an integer, an address, etc.
The semantic of cmd and arg is driver-specific.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

216/477

ioctl() example: kernel side

static long phantom_ioctl(struct file *file, unsigned int cmd,

unsigned long arg)
{
struct phm_reg r;
void __user *argp = (void __user *)arg;
switch (cmd) {
case PHN_SET_REG:
if (copy_from_user(&r, argp, sizeof(r)))
return -EFAULT;
/* Do something */
break;
case PHN_GET_REG:
if (copy_to_user(argp, &r, sizeof(r)))
return -EFAULT;
/* Do something */
break;
default:
return -ENOTTY;
}
return 0; }

Selected excerpt from drivers/misc/phantom.c

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

217/477

Ioctl() Example: Application Side

int main(void)
{
int fd, ret;
struct phm_reg reg;
fd = open("/dev/phantom");
assert(fd > 0);
reg.field1 = 42;
reg.field2 = 67;
ret = ioctl(fd, PHN_SET_REG, & reg);
assert(ret == 0);
return 0;
}
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

218/477

Kernel frameworks for device drivers

The concept of kernel frameworks

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

219/477

Beyond character drivers: kernel frameworks

Many device drivers are not implemented directly as character

drivers
They are implemented under a framework, specific to a given
device type (framebuffer, V4L, serial, etc.)
I

The framework allows to factorize the common parts of drivers

for the same type of devices
From user space, they are still seen as character devices by the
applications
The framework allows to provide a coherent user space
interface (ioctl, etc.) for every type of device, regardless of
the driver

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

220/477

Kernel Frameworks

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

221/477

Kernel frameworks for device drivers

Example of the framebuffer

framework

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

222/477

Example: Framebuffer Framework

Kernel option CONFIG_FB

I

menuconfig FB
I

Implemented in drivers/video/
I

fb.c, fbmem.c, fbmon.c, fbcmap.c, fbsysfs.c, modedb.c,

fbcvt.c

Implements a single character driver and defines the

user/kernel API
I

tristate "Support for frame buffer devices"

First part of include/linux/fb.h

Defines the set of operations a framebuffer driver must

implement and helper functions for the drivers
I
I

struct fb_ops
Second part of include/linux/fb.h

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

223/477

Framebuffer Driver Skeleton

I

Skeleton driver in drivers/video/skeletonfb.c

Implements the set of framebuffer specific operations defined

by the struct fb_ops structure

xxxfb_open()

xxxfb_read()

xxxfb_write()

xxxfb_release()

xxxfb_checkvar()

xxxfb_setpar()

xxxfb_setcolreg()

xxxfb_blank()

xxxfb_pan_display()

xxxfb_fillrect()

xxxfb_copyarea()

xxxfb_imageblit()

xxxfb_cursor()

xxxfb_rotate()

xxxfb_sync()

xxxfb_ioctl()

xxxfb_mmap()

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

224/477

Framebuffer Driver Skeleton

I

After the implementation of the operations, definition of a

struct fb_ops structure
static struct fb_ops xxxfb_ops = {
.owner = THIS_MODULE,
.fb_open = xxxfb_open,
.fb_read = xxxfb_read,
.fb_write = xxxfb_write,
.fb_release = xxxfb_release,
.fb_check_var = xxxfb_check_var,
.fb_set_par = xxxfb_set_par,
.fb_setcolreg = xxxfb_setcolreg,
.fb_blank = xxxfb_blank,
.fb_pan_display = xxxfb_pan_display,
.fb_fillrect = xxxfb_fillrect,
/* Needed !!! */
.fb_copyarea = xxxfb_copyarea,
/* Needed !!! */
.fb_imageblit = xxxfb_imageblit, /* Needed !!! */
.fb_cursor = xxxfb_cursor,
/* Optional !!! */
.fb_rotate = xxxfb_rotate,
.fb_sync = xxxfb_sync,
.fb_ioctl = xxxfb_ioctl,
.fb_mmap = xxxfb_mmap,
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

225/477

Framebuffer Driver Skeleton

I

In the probe() function, registration of the framebuffer

device and operations
static int xxxfb_probe (struct pci_dev *dev,
const struct pci_device_id *ent)
{
struct fb_info *info;
[...]
info = framebuffer_alloc(sizeof(struct xxx_par), device);
[...]
info->fbops = &xxxfb_ops;
[...]
if (register_framebuffer(info) > 0)
return -EINVAL;
[...]
}

register_framebuffer() will create the character device

that can be used by user space applications with the generic
framebuffer API.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

226/477

Driver-specific Data Structure

Each framework defines a structure that a device driver must

register to be recognized as a device in this framework
I

struct uart_port for serial ports, struct netdev for

network devices, struct fb_info for framebuffers, etc.

In addition to this structure, the driver usually needs to store

additional information about its device
This is typically done
I
I
I

By subclassing the appropriate framework structure

By storing a reference to the appropriate framework structure
Or by including your information in the framework structure

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

227/477

Driver-specific Data Structure Examples 1/2

I

i.MX serial driver: struct imx_port is a subclass of

struct uart_port
struct imx_port {
struct uart_port port;
struct timer_list timer;
unsigned int old_status;
int txirq, rxirq, rtsirq;
unsigned int have_rtscts:1;
[...]
};

ds1305 RTC driver: struct ds1305 has a reference to

struct rtc_device
struct ds1305 {
struct spi_device
struct rtc_device
[...]
};

*spi;
*rtc;

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

228/477

Driver-specific Data Structure Examples 2/2

rtl8150 network driver: struct rtl8150 has a reference to

struct net_device and is allocated within that framework
structure.
struct rtl8150 {
unsigned long flags;
struct usb_device *udev;
struct tasklet_struct tl;
struct net_device *netdev;
[...]
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

229/477

Link Between Structures 1/4

The framework typically contains a struct device * pointer

that the driver must point to the corresponding
struct device
I

Its the relation between the logical device (for example a

network interface) and the physical device (for example the
USB network adapter)

The device structure also contains a void * pointer that the

driver can freely use.
I

Its often used to link back the device to the higher-level

structure from the framework.
It allows, for example, from the struct platform_device
structure, to find the structure describing the logical device

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

230/477

Link Between Structures 2/4

static int serial_imx_probe(struct platform_device *pdev)

{
struct imx_port *sport;
[...]
/* setup the link between uart_port and the struct
* device inside the platform_device */
sport->port.dev = &pdev->dev;
[...]
/* setup the link between the struct device inside
* the platform device to the imx_port structure */
platform_set_drvdata(pdev, sport);
[...]
uart_add_one_port(&imx_reg, &sport->port);
}
static int serial_imx_remove(struct platform_device *pdev)
{
/* retrieve the imx_port from the platform_device */
struct imx_port *sport = platform_get_drvdata(pdev);
[...]
uart_remove_one_port(&imx_reg, &sport->port);
[...]
}

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

231/477

Link Between Structures 3/4

static int ds1305_probe(struct spi_device *spi)
{
struct ds1305
*ds1305;
[...]
/* set up driver data */
ds1305 = devm_kzalloc(&spi->dev, sizeof(*ds1305), GFP_KERNEL);
if (!ds1305)
return -ENOMEM;
ds1305->spi = spi;
spi_set_drvdata(spi, ds1305);
[...]
/* register RTC ... from here on, ds1305->ctrl needs locking */
ds1305->rtc = devm_rtc_device_register(&spi->dev, "ds1305",
&ds1305_ops, THIS_MODULE);
[...]
}
static int ds1305_remove(struct spi_device *spi)
{
struct ds1305 *ds1305 = spi_get_drvdata(spi);
[...]
}

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

232/477

Link Between Structures 4/4

static int rtl8150_probe(struct usb_interface *intf,
const struct usb_device_id *id)
{
struct usb_device *udev = interface_to_usbdev(intf);
rtl8150_t *dev;
struct net_device *netdev;
netdev = alloc_etherdev(sizeof(rtl8150_t));
dev = netdev_priv(netdev);
[...]
dev->udev = udev;
dev->netdev = netdev;
[...]
usb_set_intfdata(intf, dev);
SET_NETDEV_DEV(netdev, &intf->dev);
[...]
}
static void rtl8150_disconnect(struct usb_interface *intf)
{
rtl8150_t *dev = usb_get_intfdata(intf);
[...]
}

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

233/477

The input subsystem

The input
subsystem

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

234/477

What is input subsystem?

I

The input subsystem takes care of all the input events coming
from the human user.

Initially written to support the USB HID (Human Interface

Device) devices, it quickly grew up to handle all kind of inputs
(using USB or not): keyboards, mice, joysticks, touchscreens,
etc.
The input subsystem is split in two parts:

Device drivers: they talk to the hardware (for example via

USB), and provide events (keystrokes, mouse movements,
touchscreen coordinates) to the input core
Event handlers: they get events from drivers and pass them
where needed via various interfaces (most of the time through
evdev)

In user space it is usually used by the graphic stack such as

X.Org, Wayland or Androids InputManager.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

235/477

Input subsystem diagram

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

236/477

Input subsystem overview

I

Kernel option CONFIG_INPUT

I

menuconfig INPUT
I

Implemented in drivers/input/
I

input.c, input-polldev.c, evbug.c

Implements a single character driver and defines the

user/kernel API
I

tristate "Generic input layer (needed for keyboard,

mouse, ...)"

include/uapi/linux/input.h

Defines the set of operations a input driver must implement

and helper functions for the drivers
I
I
I

struct input_dev for the device driver part

struct input_handler for the event handler part
include/linux/input.h

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

237/477

Input subsystem API 1/3

I

An input device is described by a very long

struct input_dev structure, an excerpt is:

struct input_dev {
const char *name;
[...]
unsigned long evbit[BITS_TO_LONGS(EV_CNT)];
unsigned long keybit[BITS_TO_LONGS(KEY_CNT)];
[...]
int (*getkeycode)(struct input_dev *dev,
struct input_keymap_entry *ke);
[...]
int (*open)(struct input_dev *dev);
[...]
int (*event)(struct input_dev *dev, unsigned int type,
unsigned int code, int value);
[...]
};

Before being used this struct must be allocated and initialized:

struct input_dev *input_allocate_device(void);

After unregistering struct input_dev, it must be freed:

void input_free_device(struct input_dev *dev);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

238/477

Input subsystem API 2/3

I

Depending on the type of event that will be generated, the

input bit fields evbit and keybit must be configured: For
example, for a button we only generate EV_KEY type events,
and from these only BTN_0 events code:
set_bit(EV_KEY, myinput_dev.evbit);
set_bit(BTN_0, myinput_dev.keybit);

set_bit() is an atomic operation allowing to set a particular

bit to 1 (explained later).

Once the input device is allocated and filled, the function to

register it is:
int input_register_device(struct input_dev *);

When the driver is unloaded, the input device will be

unregistered using:
void input_unregister_device(struct input_dev *);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

239/477

Input subsystem API 3/3

The events are sent by the driver to the event handler using
input_event(struct input_dev *dev, unsigned int
type, unsigned int code, int value);
I

The event types are documented in

Documentation/input/event-codes.txt
An event is composed by one or several input data changes
(packet of input data changes) such as the button state, the
relative or absolute position along an axis, etc..
After submitting potentially multiple events, the input core
must be notified by calling:
void input_sync(struct input_dev *dev):
The input subsystem provides other wrappers such as
input_report_key(), input_report_abs(), ...

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

240/477

Polled input subclass

I

The input subsystem provides a subclass supporting simple

input devices that do not raise interrupts but have to be
periodically scanned or polled to detect changes in their state.

A polled input device is described by a

struct input_polled_dev structure:

struct input_polled_dev {
void *private;
void (*open)(struct input_polled_dev *dev);
void (*close)(struct input_polled_dev *dev);
void (*poll)(struct input_polled_dev *dev);
unsigned int poll_interval; /* msec */
unsigned int poll_interval_max; /* msec */
unsigned int poll_interval_min; /* msec */
struct input_dev *input;
/* private: */
struct delayed_work work;
}
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

241/477

Polled input subsystem API

I

Allocating/freeing the struct input_polled_dev structure

is done using
I
I

I
I

input_allocate_polled_device()
input_free_polled_device()

Among the handlers of the struct input_polled_dev only

the poll() method is mandatory, this function polls the
device and posts input events.
The fields id, name, evkey and keybit of the input field
must be initialized too.
If none of the poll_interval fields are filled then the default
poll interval is 500ms.
The device registration/unregistration is done with:
I

input_register_polled_device(struct input_polled_
dev *dev).
input_unregister_polled_device(struct input_
polled_dev *dev)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

242/477

evdev user space interface

I

The main user space interface to input devices is the event

interface
Each input device is represented as a /dev/input/event<X>
character device
A user space application can use blocking and non-blocking
reads, but also select() (to get notified of events) after
opening this device.
Each read will return struct input_event structures of the
following format:

struct input_event {
struct timeval time;
unsigned short type;
unsigned short code;
unsigned int value;
};

A very useful application for input device testing is evtest,

from http://cgit.freedesktop.org/evtest/

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

243/477

Practical lab - Expose the Nunchuk to user space

Extend the Nunchuk driver to

expose the Nunchuk features to
user space applications, as an input
device.

Test the operation of the Nunchuk

using sample user space
applications.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

244/477

Memory Management

Memory
Management

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

245/477

Physical and Virtual Memory

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

246/477

Virtual Memory Organization

I

1GB reserved for kernel-space

I

Contains kernel code and core data

structures, identical in all address spaces
Most memory can be a direct mapping
of physical memory at a fixed offset

Complete 3GB exclusive mapping

available for each user space process
I

I
I

Process code and data (program, stack,

...)
Memory-mapped files
Not necessarily mapped to physical
memory (demand fault paging used for
dynamic mapping to physical memory
pages)
Differs from one address space to
another

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

247/477

Physical / virtual memory mapping

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

248/477

Accessing more physical memory

I
I

Only less than 1GB memory addressable directly through

kernel virtual address space
If more physical memory is present on the platform, part of
the memory will not be accessible by kernel space, but can be
used by user space
To allow the kernel to access more physical memory:
I

Change 1GB/3GB memory split (2GB/2GB)

(CONFIG_VMSPLIT_3G) reduces total memory available for
each process
Change for a 64 bit architecture ;-) See
Documentation/x86/x86_64/mm.txt for an example.
Activate highmem support if available for your architecture:
I

I
I

Allows kernel to map parts of its non-directly accessible

memory
Mapping must be requested explicitly
Limited addresses ranges reserved for this usage

See http://lwn.net/Articles/75174/ for useful

explanations

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

249/477

Accessing even more physical memory!

If your 32 bit platform hosts more than 4GB, they just cannot
be mapped

PAE (Physical Address Expansion) may be supported by your

architecture

Adds some address extension bits used to index memory areas

Allows accessing up to 64 GB of physical memory on x86

Note that each user space process is still limited to a 3 GB

memory space

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

250/477

Notes on user space memory

New user space memory is allocated either from the already

allocated process memory, or using the mmap system call
Note that memory allocated may not be physically allocated:
I

Kernel uses demand fault paging to allocate the physical page

(the physical page is allocated when access to the virtual
address generates a page fault)
... or may have been swapped out, which also induces a page
fault

User space memory allocation is allowed to over-commit

memory (more than available physical memory) can lead to
out of memory

OOM killer kicks in and selects a process to kill to retrieve

some memory. Thats better than letting the system freeze.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

251/477

Back to kernel memory

Kernel memory allocators (see following slides) allocate

physical pages, and kernel allocated memory cannot be
swapped out, so no fault handling required for kernel memory.

Most kernel memory allocation functions also return a kernel

virtual address to be used within the kernel space.

Kernel memory low-level allocator manages pages. This is the

finest granularity (usually 4 KB, architecture dependent).

However, the kernel memory management handles smaller

memory allocations through its allocator (see SLAB allocators
used by kmalloc()).

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

252/477

Allocators in the Kernel

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

253/477

Page Allocator
I

Appropriate for medium-size allocations

A page is usually 4K, but can be made greater in some

architectures (sh, mips: 4, 8, 16 or 64 KB, but not
configurable in x86 or arm).

Buddy allocator strategy, so only allocations of power of two

number of pages are possible: 1 page, 2 pages, 4 pages, 8
pages, 16 pages, etc.

Typical maximum size is 8192 KB, but it might depend on the

kernel configuration.
The allocated area is virtually contiguous (of course), but also
physically contiguous. It is allocated in the identity-mapped
part of the kernel memory space.

This means that large areas may not be available or hard to

retrieve due to physical memory fragmentation.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

254/477

Page Allocator API: Get free pages

I

unsigned long get_zeroed_page(int flags)

I
I

unsigned long __get_free_page(int flags)

I

Returns the virtual address of a free page, initialized to zero

flags: see the next pages for details.

Same, but doesnt initialize the contents

unsigned long __get_free_pages(int flags,

unsigned int order)
I

Returns the starting virtual address of an area of several

contiguous pages in physical RAM, with order being
log2(number_of_pages).Can be computed from the size
with the get_order() function.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

255/477

Page Allocator API: Free Pages

void free_page(unsigned long addr)

I

Frees one page.

void free_pages(unsigned long addr,

unsigned int order)
I

Frees multiple pages. Need to use the same order as in

allocation.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

256/477

Page Allocator Flags

The most common ones are:

I

GFP_KERNEL
I

GFP_ATOMIC
I

RAM allocated from code which is not allowed to block

(interrupt handlers or critical sections). Never blocks, allows
to access emergency pools, but can fail if no free memory is
readily available.

GFP_DMA
I

Standard kernel memory allocation. The allocation may block

in order to find enough available memory. Fine for most
needs, except in interrupt handler context.

Allocates memory in an area of the physical memory usable

for DMA transfers. See our DMA chapter.

Others are defined in include/linux/gfp.h

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

257/477

SLAB Allocator 1/2

The SLAB allocator allows to create caches, which contains a

set of objects of the same size

The object size can be smaller or greater than the page size

The SLAB allocator takes care of growing or reducing the size

of the cache as needed, depending on the number of allocated
objects. It uses the page allocator to allocate and free pages.
SLAB caches are used for data structures that are present in
many many instances in the kernel: directory entries, file
objects, network packet descriptors, process descriptors, etc.

See /proc/slabinfo

They are rarely used for individual drivers.

See include/linux/slab.h for the API

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

258/477

SLAB Allocator 2/2

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

259/477

Different SLAB Allocators

I

There are three different, but API compatible,

implementations of a SLAB allocator in the Linux kernel. A
particular implementation is chosen at configuration time.
I

SLAB: legacy, well proven allocator.

Still the default in most ARM defconfig files.
SLOB: much simpler. More space efficient but doesnt scale
well. Saves a few hundreds of KB in small systems (depends
on CONFIG_EXPERT)
Linux 3.17 on ARM: used in 5 defconfig files
SLUB: more recent and simpler than SLAB, scaling much
better (in particular for huge systems) and creating less
fragmentation.
Linux 3.17 on ARM: used in 0 defconfig files

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

260/477

kmalloc Allocator
I
I
I
I
I

I
I

The kmalloc allocator is the general purpose memory allocator

in the Linux kernel
For small sizes, it relies on generic SLAB caches, named
kmalloc-XXX in /proc/slabinfo
For larger sizes, it relies on the page allocator
The allocated area is guaranteed to be physically contiguous
The allocated area size is rounded up to the size of the
smallest SLAB cache in which it can fit (while using the SLAB
allocator directly allows to have more flexibility)
It uses the same flags as the page allocator (GFP_KERNEL,
GFP_ATOMIC, GFP_DMA, etc.) with the same semantics.
Maximum sizes, on x86 and arm (see http://j.mp/YIGq6W):
- Per allocation: 4 MB
- Total allocations: 128 MB
Should be used as the primary allocator unless there is a
strong reason to use another one.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

261/477

kmalloc API 1/2

I

#include <linux/slab.h>

void *kmalloc(size_t size, int flags);

I

I
I

void kfree(const void *objp);

I

Allocate size bytes, and return a pointer to the area (virtual

address)
size: number of bytes to allocate
flags: same flags as the page allocator

Free an allocated area

Example: (drivers/infiniband/core/cache.c)
struct ib_update_work *work;
work = kmalloc(sizeof *work, GFP_ATOMIC);
...
kfree(work);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

262/477

kmalloc API 2/2

void *kzalloc(size_t size, gfp_t flags);

I

Allocates a zero-initialized buffer

void *kcalloc(size_t n, size_t size, gfp_t flags);

I

Allocates memory for an array of n elements of size size, and

zeroes its contents.

I void *krealloc(const void *p, size_t new_size, gfp_t flags);

I

Changes the size of the buffer pointed by p to new_size, by

reallocating a new buffer and copying the data, unless
new_size fits within the alignment of the existing buffer.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

263/477

devm kmalloc functions

Automatically free the allocated buffers when the

corresponding device or module is unprobed.

Need to have a reference to a struct device.

I void *devm_kmalloc(struct device *dev, size_t size, int flags);

I void *devm_kzalloc(struct device *dev, size_t size, int flags);

I void *devm_kcalloc(struct device *dev, size_t n, size_t size, gfp_t flags)

I

void *devm_kfree(struct device *dev, void *p);

I

Useful to immediately free an allocated buffer

See Documentation/driver-model/devres.txt for details

about managed device resources.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

264/477

vmalloc Allocator
I

The vmalloc() allocator can be used to obtain virtually

contiguous memory zones, but not physically contiguous. The
requested memory size is rounded up to the next page.

The allocated area is in the kernel space part of the address

space, but outside of the identically-mapped area

Allocations of fairly large areas is possible (almost as big as

total available memory, see http://j.mp/YIGq6W again),
since physical memory fragmentation is not an issue, but areas
cannot be used for DMA, as DMA usually requires physically
contiguous buffers.
API in include/linux/vmalloc.h

void *vmalloc(unsigned long size);

I

Returns a virtual address

void vfree(void *addr);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

265/477

Kernel memory debugging

Kmemcheck
I
I

Dynamic checker for access to uninitialized memory.

Only available on x86 so far (Linux 3.17 status), but will help
to improve architecture independent code anyway.
See Documentation/kmemcheck.txt for details.

Kmemleak
I
I
I

Dynamic checker for memory leaks

This feature is available for all architectures.
See Documentation/kmemleak.txt for details.

Both have a significant overhead. Only use them in development!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

266/477

I/O Memory and Ports

I/O Memory and

Ports

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

267/477

Port I/O vs. Memory-Mapped I/O

MMIO
I
I
I

Same address bus to address memory and I/O devices

Access to the I/O devices using regular instructions
Most widely used I/O method across the different architectures
supported by Linux

PIO
I
I
I

Different address spaces for memory and I/O devices

Uses a special class of CPU instructions to access I/O devices
Example on x86: IN and OUT instructions

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

268/477

MMIO vs PIO

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

269/477

Requesting I/O ports

I

Tells the kernel which driver is using which I/O ports

Allows to prevent other drivers from using the same I/O ports,
but is purely voluntary.

struct resource *request_region(

unsigned long start,
unsigned long len,
char *name);

Tries to reserve the given region and returns NULL if

unsuccessful.

request_region(0x0170, 8, "ide1");

void release_region(
unsigned long start,
unsigned long len);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

270/477

/proc/ioports example (x86)

0000-001f
0020-0021
0040-0043
0050-0053
0070-0077
0080-008f
00a0-00a1
00c0-00df
00f0-00ff
0170-0177
01f0-01f7
0376-0376
03f6-03f6
03f8-03ff
0800-087f
...

:
:
:
:
:
:
:
:
:
:
:
:
:
:
:

dma1
pic1
timer0
timer1
rtc
dma page reg
pic2
dma2
fpu
ide1
ide0
ide1
ide0
serial
0000:00:1f.0

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

271/477

Accessing I/O ports

I

Functions to read/write bytes (b), word (w) and longs (l) to

I/O ports:
I

unsigned in[bwl](unsigned long port)

void out[bwl](value, unsigned long port)

And the strings variants: often more efficient than the

corresponding C loop, if the processor supports such
operations!
I

void ins[bwl](unsigned port, void *addr,

unsigned long count)

void outs[bwl](unsigned port, void *addr,

unsigned long count)

Examples
I

read 8 bits
I

oldlcr = inb(baseio + UART_LCR)

write 8 bits
I

outb(MOXA_MUST_ENTER_ENCHANCE, baseio + UART_LCR)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

272/477

Requesting I/O memory

Functions equivalent to request_region() and

release_region(), but for I/O memory.

struct resource *request_mem_region(

unsigned long start,
unsigned long len,
char *name);

void release_mem_region(
unsigned long start,
unsigned long len);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

273/477

/proc/iomem example
00000000-0009efff
0009f000-0009ffff
000a0000-000bffff
000c0000-000cffff
000f0000-000fffff
00100000-3ffadfff
00100000-0030afff
0030b000-003b4bff
3ffae000-3fffffff
40000000-400003ff
40001000-40001fff
40400000-407fffff
40800000-40bfffff
a0000000-a0000fff
e8000000-efffffff
...

:
:
:
:
:
:
:
:
:
:
:
:
:
:
:

System RAM
reserved
Video RAM area
Video ROM
System ROM
System RAM
Kernel code
Kernel data
reserved
0000:00:1f.1
0000:02:01.0
PCI CardBus #03
PCI CardBus #03
pcmcia_socket0
PCI Bus #01

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

274/477

Mapping I/O memory in virtual memory

Load/store instructions work with virtual addresses

To access I/O memory, drivers need to have a virtual address

that the processor can handle, because I/O memory is not
mapped by default in virtual memory.

The ioremap function satisfies this need:

#include <asm/io.h>
void __iomem *ioremap(phys_addr_t phys_addr,
unsigned long size);
void iounmap(void __iomem *addr);

Caution: check that ioremap() doesnt return a NULL

address!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

275/477

ioremap()

ioremap(0xFFEBC00, 4096) = 0xCDEFA000

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

276/477

Managed API

Using request_mem_region() and ioremap() in device drivers is

now deprecated. You should use the below managed functions
instead, which simplify driver coding and error handling:
I

devm_ioremap()

devm_iounmap()
devm_ioremap_resource()

Takes care of both the request and remapping operations!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

277/477

Accessing MMIO devices

I

Directly reading from or writing to addresses returned by

ioremap() (pointer dereferencing) may not work on some
architectures.

To do PCI-style, little-endian accesses, conversion being done

automatically
unsigned read[bwl](void *addr);
void write[bwl](unsigned val, void *addr);

To do raw access, without endianness conversion

unsigned __raw_read[bwl](void *addr);
void __raw_write[bwl](unsigned val, void *addr);
Example

32 bits write
__raw_writel(1 << KS8695_IRQ_UART_TX,
membase + KS8695_INTST);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

278/477

Avoiding I/O access issues

Caching on I/O ports or memory already disabled

Use the macros, they do the right thing for your architecture
The compiler and/or CPU can reorder memory accesses,
which might cause troubles for your devices is they expect one
register to be read/written before another one.

I
I

I
I

Memory barriers are available to prevent this reordering

rmb() is a read memory barrier, prevents reads to cross the
barrier
wmb() is a write memory barrier
mb() is a read-write memory barrier

Starts to be a problem with CPUs that reorder instructions

and SMP.

See Documentation/memory-barriers.txt for details

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

279/477

/dev/mem

Used to provide user space applications with direct access to

physical addresses.

Usage: open /dev/mem and read or write at given offset.

What you read or write is the value at the corresponding
physical address.

Used by applications such as the X server to write directly to

device memory.

On x86, arm, arm64, tile, powerpc, unicore32, s390:

CONFIG_STRICT_DEVMEM option to restrict /dev/mem
non-RAM addresses, for security reasons (Linux 3.10 status).

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

280/477

Practical lab - I/O Memory and Ports

Add UART devices to the board

device tree

Access I/O registers to control the

device and send first characters to
it.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

281/477

The misc subsystem

The misc
subsystem

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

282/477

Why a misc subsystem?

I

The kernel offers a large number of frameworks covering a

wide range of device types: input, network, video, audio, etc.
I

However, there are some devices that really do not fit in any
of the existing frameworks.
I

These frameworks allow to factorize common functionality

between drivers and offer a consistent API to user space
applications.

Highly customized devices implemented in a FPGA, or other

weird devices for which implementing a complete framework is
not useful.

The drivers for such devices could be implemented directly as

raw character drivers.
But there is a subsystem that makes this work a little bit
easier: the misc subsystem.
I

It is really only a thin layer above the character driver API.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

283/477

Misc subsystem diagram

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

284/477

Misc subsystem API (1/2)

I

The misc subsystem API mainly provides two functions, to

register and unregister a single misc device:
I
I

int misc_register(struct miscdevice * misc);

int misc_deregister(struct miscdevice *misc);

A misc device is described by a struct miscdevice

structure:
struct miscdevice {
int minor;
const char *name;
const struct file_operations *fops;
struct list_head list;
struct device *parent;
struct device *this_device;
const char *nodename;
umode_t mode;
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

285/477

Misc subsystem API (2/2)

The main fields to be filled in struct miscdevice are:

I

minor, the minor number for the device, or

MISC_DYNAMIC_MINOR to get a minor number automatically
assigned.

name, name of the device, which will be used to create the

device node if devtmpfs is used.

fops, pointer to a struct file_operations structure, that

describes which functions implement the read, write, ioctl,
etc. operations.

parent, the struct device that represents the hardware

device exposed by this driver.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

286/477

User space API for misc devices

I
I

misc devices are regular character devices

The operations they support in user space depends on the
operations the kernel driver implements:
I

The open() and close() system calls to open/close the

device.
The read() and write() system calls to read/write to/from
the device.
The ioctl() system call to call some driver-specific
operations.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

287/477

Practical lab - Output-only serial port driver

Extend the driver started in the

previous lab by registering it into
the misc subsystem.

Implement serial output

functionality through the misc
subsystem.

Test serial output using user space

applications.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

288/477

Processes, scheduling and interrupts

Processes,
scheduling and
interrupts

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

289/477

Processes, scheduling and interrupts

Processes and scheduling

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

290/477

Process, thread?

I
I

Confusion about the terms process, thread and task

In Unix, a process is created using fork() and is composed of
I

I
I

An address space, which contains the program code, data,

stack, shared libraries, etc.
One thread, that starts executing the main() function.
Upon creation, a process contains one thread

Additional threads can be created inside an existing process,

using pthread_create()
I

They run in the same address space as the initial thread of the
process
They start executing a function passed as argument to
pthread_create()

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

291/477

Process, thread: kernel point of view

I

The kernel represents each thread running in the system by a

structure of type struct task_struct

From a scheduling point of view, it makes no difference

between the initial thread of a process and all additional
threads created dynamically using pthread_create()

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

292/477

A thread life

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

293/477

Execution of system calls

The execution of system calls takes place in the context of the

thread requesting them.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

294/477

Processes, scheduling and interrupts

Sleeping

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

295/477

Sleeping

Sleeping is needed when a process (user space or kernel space) is

waiting for data.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

296/477

How to sleep 1/3

I
I

Must declare a wait queue

A wait queue will be used to store the list of threads waiting
for an event
I

Static queue declaration

I
I

useful to declare as a global variable

DECLARE_WAIT_QUEUE_HEAD(module_queue);

Or dynamic queue declaration

I

Useful to embed the wait queue inside another data structure

wait_queue_head_t queue;
init_waitqueue_head(&queue);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

297/477

How to sleep 2/3

Several ways to make a kernel process sleep

I

void wait_event(queue, condition);

I

int wait_event_killable(queue, condition);

I

Sleeps until the task is woken up and the given C expression is

true. Caution: cant be interrupted (cant kill the user space
process!)

Can be interrupted, but only by a fatal signal (SIGKILL).

Returns -ERESTARSYS if interrupted.

int wait_event_interruptible(queue, condition);

I

Can be interrupted by any signal. Returns -ERESTARTSYS if

interrupted.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

298/477

How to sleep 3/3

int wait_event_timeout(queue, condition, timeout);

I

Also stops sleeping when the task is woken up and the timeout
expired. Returns 0 if the timeout elapsed, non-zero if the
condition was met.

int wait_event_interruptible_timeout(queue,
condition, timeout);
I

Same as above, interruptible. Returns 0 if the timeout elapsed,

-ERESTARTSYS if interrupted, positive value if the condition
was met.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

299/477

How to Sleep - Example

ret = wait_event_interruptible
(sonypi_device.fifo_proc_list,
kfifo_len(sonypi_device.fifo) != 0);
if (ret)
return ret;

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

300/477

Waking up!

Typically done by interrupt handlers when data sleeping

processes are waiting for become available.
I

wake_up(&queue);

wake_up_interruptible(&queue);

Wakes up all processes in the wait queue

Wakes up all processes waiting in an interruptible sleep on the
given queue

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

301/477

Exclusive vs. non-exclusive

I

wait_event_interruptible() puts a task in a

non-exclusive wait.
I

All non-exclusive tasks are woken up by wake_up() /

wake_up_interruptible()

wait_event_interruptible_exclusive() puts a task in

an exclusive wait.
I

wake_up() / wake_up_interruptible() wakes up all

non-exclusive tasks and only one exclusive task
wake_up_all() / wake_up_interruptible_all() wakes up
all non-exclusive and all exclusive tasks

Exclusive sleeps are useful to avoid waking up multiple tasks

when only one will be able to consume the event.

Non-exclusive sleeps are useful when the event can benefit

to multiple tasks.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

302/477

Sleeping and waking up - Implementation 1/2

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

303/477

Sleeping and waking up - Implementation 2/2

The scheduler doesnt keep evaluating the sleeping condition!

I

wait_event(queue, condition);
I

The process is put in the TASK_UNINTERRUPTIBLE state.

wake_up(&queue);
I

All processes waiting in queue are woken up, so they get

scheduled later and have the opportunity to evaluate the
condition again and go back to sleep if it is not met.

See include/linux/wait.h for implementation details.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

304/477

Processes, scheduling and interrupts

Interrupt Management

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

305/477

Registering an interrupt handler 1/2

The managed API is recommended:
int devm_request_irq(struct device *dev,
unsigned int irq,
irq_handler_t handler,
unsigned long irq_flags,
const char *devname,
void *dev_id);
I
I
I
I
I
I
I

Register an interrupt handler.

device for automatic freeing at device or module release time.
irq is the requested IRQ channel. For platform devices, use
platform_get_irq() to retrieve the interrupt number.
handler is a pointer to the IRQ handler
irq_flags are option masks (see next slide)
devname is the registered name
dev_id is a pointer to some data. It cannot be NULL as it is
used as an identifier for free_irq() when using shared IRQs.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

306/477

Releasing an interrupt handler

void devm_free_irq(struct device *dev,

unsigned int irq, void *dev_id);
I

Explicitly release an interrupt handler. Done automatically in

normal situations.

Defined in include/linux/interrupt.h

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

307/477

Registering an interrupt handler 2/2

Main irq_flags bit values

(can be combined, 0 when no flags are needed):
I IRQF_SHARED
I

The interrupt channel can be shared by several devices.

Requires a hardware status register telling whether an IRQ was
raised or not.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

308/477

Interrupt handler constraints

No guarantee in which address space the system will be in

when the interrupt occurs: cant transfer data to and from
user space.

Interrupt handler execution is managed by the CPU, not by

the scheduler. Handlers cant run actions that may sleep,
because there is nothing to resume their execution. In
particular, need to allocate memory with GFP_ATOMIC.

Interrupt handlers are run with all interrupts disabled on the

local CPU (see http://lwn.net/Articles/380931).
Therefore, they have to complete their job quickly enough, to
avoiding blocking interrupts for too long.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

309/477

/proc/interrupts on a Panda board

39:
41:
42:
43:
44:
52:
...
IPI0:
IPI1:
IPI2:
IPI3:
IPI4:
LOC:
Err:

CPU0
4
0
0
0
20294
0

CPU1
0
0
0
0
0
0

GIC
GIC
GIC
GIC
GIC
GIC

0
23095
0
231
0
196407
0

0
25663
0
173
0
136995

Timer broadcast interrupts

Rescheduling interrupts
Function call interrupts
Single function call interrupts
CPU stop interrupts
Local timer interrupts

TWL6030-PIH
l3-dbg-irq
l3-app-irq
prcm
DMA
gpmc

Note: interrupt numbers shown on the left-most column are virtual numbers
when the Device Tree is used. The real physical interrupt numbers are either
shown as an additional column, or can be seen in
/sys/kernel/debug/irq_domain_mapping.
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

310/477

Interrupt handler prototype

irqreturn_t foo_interrupt(int irq, void *dev_id)

I
I

irq, the IRQ number

dev_id, the opaque pointer that was passed to
devm_request_irq()

Return value
I
I

IRQ_HANDLED: recognized and handled interrupt

IRQ_NONE: not on a device managed by the module. Useful to
share interrupt channels and/or report spurious interrupts to
the kernel.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

311/477

Typical interrupt handlers job

Acknowledge the interrupt to the device (otherwise no more

interrupts will be generated, or the interrupt will keep firing
over and over again)

Read/write data from/to the device

Wake up any waiting process waiting for the completion of an

operation, typically using wait queues
wake_up_interruptible(&module_queue);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

312/477

Threaded interrupts
In 2.6.30, support for threaded interrupts has been added to the
Linux kernel
I The interrupt handler is executed inside a thread.
I Allows to block during the interrupt handler, which is often
needed for I2C/SPI devices as the interrupt handler needs to
communicate with them.
I Allows to set a priority for the interrupt handler execution,
which is useful for real-time usage of Linux
int devm_request_threaded_irq(
struct device *dev,
unsigned int irq,
irq_handler_t handler, irq_handler_t thread_fn
unsigned long flags, const char *name, void *dev);
I
I

handler, hard IRQ handler

thread_fn, executed in a thread

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

313/477

Top half and bottom half processing

Splitting the execution of interrupt handlers in 2 parts

I

Top half
I

This is the real interrupt handler, which should complete as

quickly as possible since all interrupts are disabled. If possible,
take the data out of the device and schedule a bottom half to
handle it.

Bottom half
I

Is the general Linux name for various mechanisms which allow

to postpone the handling of interrupt-related work.
Implemented in Linux as softirqs, tasklets or workqueues.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

314/477

Top half and bottom half diagram

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

315/477

Softirqs
I

Softirqs are a form of bottom half processing

The softirqs handlers are executed with all interrupts enabled,

and a given softirq handler can run simultaneously on multiple
CPUs

They are executed once all interrupt handlers have completed,

before the kernel resumes scheduling processes, so sleeping is
not allowed.

The number of softirqs is fixed in the system, so softirqs are

not directly used by drivers, but by complete kernel
subsystems (network, etc.)

The list of softirqs is defined in

include/linux/interrupt.h: HI, TIMER, NET_TX, NET_RX,
BLOCK, BLOCK_IOPOLL, TASKLET, SCHED, HRTIMER, RCU

The HI and TASKLET softirqs are used to execute tasklets

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

316/477

Tasklets
I

Tasklets are executed within the HI and TASKLET softirqs.

They are executed with all interrupts enabled, but a given
tasklet is guaranteed to execute on a single CPU at a time.

A tasklet can be declared statically with the

DECLARE_TASKLET() macro or dynamically with the
tasklet_init() function. A tasklet is simply implemented
as a function. Tasklets can easily be used by individual device
drivers, as opposed to softirqs.
The interrupt handler can schedule the execution of a tasklet
with

tasklet_schedule() to get it executed in the TASKLET

softirq
tasklet_hi_schedule() to get it executed in the HI softirq
(higher priority)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

317/477

Tasklet Example: simplified atmel serial.c 1/2

/* The tasklet function */

static void atmel_tasklet_func(unsigned long data) {
struct uart_port *port = (struct uart_port *)data;
[...]
}
/* Registering the tasklet */
init function(...) {
[...]
tasklet_init(&atmel_port->tasklet,
atmel_tasklet_func, (unsigned long)port);
[...]
}

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

318/477

Tasklet Example: simplified atmel serial.c 2/2

/* Removing the tasklet */

cleanup function(...) {
[...]
tasklet_kill(&atmel_port->tasklet);
[...]
}
/* Triggering execution of the tasklet */
somewhere function(...) {
tasklet_schedule(&atmel_port->tasklet);
}

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

319/477

Workqueues

Workqueues are a general mechanism for deferring work. It is

not limited in usage to handling interrupts.
The function registered as workqueue is executed in a thread,
which means:
I
I

All interrupts are enabled

Sleeping is allowed

A workqueue is registered with INIT_WORK() and typically

triggered with queue_work()

The complete API, in include/linux/workqueue.h

provides many other possibilities (creating its own workqueue
threads, etc.)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

320/477

Interrupt management summary

I

Device driver
I

Interrupt handler
I
I
I

Called when an interrupt is raised.

Acknowledge the interrupt
If needed, schedule a tasklet taking care of handling data.
Otherwise, wake up processes waiting for the data.

Tasklet
I
I

When the device file is first opened, register an interrupt

handler for the devices interrupt channel.

Process the data

Wake up processes waiting for the data

Device driver
I

When the device is no longer opened by any process,

unregister the interrupt handler.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

321/477

Practical lab - Interrupts

Adding read capability to the

character driver developed earlier.

Register an interrupt handler.

Waiting for data to be available in

the read file operation.

Waking up the code when data are

available from the device.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

322/477

Concurrent Access to Resources: Locking

Concurrent Access
to Resources:
Locking

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

323/477

Sources of concurrency issues

I

In terms of concurrency, the kernel has the same constraint as

a multi-threaded program: its state is global and visible in all
executions contexts
Concurrency arises because of
I

Interrupts, which interrupts the current thread to execute an

interrupt handler. They may be using shared resources.
Kernel preemption, if enabled, causes the kernel to switch from
the execution of one system call to another. They may be
using shared resources.
Multiprocessing, in which case code is really executed in
parallel on different processors, and they may be using shared
resources as well.

The solution is to keep as much local state as possible and for

the shared resources, use locking.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

324/477

Concurrency protection with locks

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

325/477

Linux mutexes

The kernels main locking primitive

The process requesting the lock blocks when the lock is

already held. Mutexes can therefore only be used in contexts
where sleeping is allowed.
Mutex definition:

Initializing a mutex statically:

I

#include <linux/mutex.h>
DEFINE_MUTEX(name);

Or initializing a mutex dynamically:

I

void mutex_init(struct mutex *lock);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

326/477

Locking and Unlocking Mutexes 1/2

void mutex_lock(struct mutex *lock);

I
I

int mutex_lock_killable(struct mutex *lock);

I

Tries to lock the mutex, sleeps otherwise.

Caution: cant be interrupted, resulting in processes you
cannot kill!

Same, but can be interrupted by a fatal (SIGKILL) signal. If

interrupted, returns a non zero value and doesnt hold the
lock. Test the return value!!!

int mutex_lock_interruptible(struct mutex *lock);

I

Same, but can be interrupted by any signal.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

327/477

Locking and Unlocking Mutexes 2/2

int mutex_trylock(struct mutex *lock);

I

int mutex_is_locked(struct mutex *lock);

I

Never waits. Returns a non zero value if the mutex is not

available.

Just tells whether the mutex is locked or not.

void mutex_unlock(struct mutex *lock);

I

Releases the lock. Do it as soon as you leave the critical

section.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

328/477

Spinlocks
I

I
I
I
I

Locks to be used for code that is not allowed to sleep

(interrupt handlers), or that doesnt want to sleep (critical
sections). Be very careful not to call functions which can
sleep!
Originally intended for multiprocessor systems
Spinlocks never sleep and keep spinning in a loop until the
lock is available.
Spinlocks cause kernel preemption to be disabled on the CPU
executing them.
The critical section protected by a spinlock is not allowed to
sleep.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

329/477

Initializing Spinlocks

Statically
I

DEFINE_SPINLOCK(my_lock);

Dynamically
I

void spin_lock_init(spinlock_t *lock);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

330/477

Using Spinlocks 1/2

I

Several variants, depending on where the spinlock is called:

I

void spin_lock(spinlock_t *lock);

void spin_unlock(spinlock_t *lock);

I

Doesnt disable interrupts. Used for locking in process context

(critical sections in which you do not want to sleep).

void spin_lock_irqsave(spinlock_t *lock,

unsigned long flags);

void spin_unlock_irqrestore(spinlock_t *lock,

unsigned long flags);
I
I

Disables / restores IRQs on the local CPU.

Typically used when the lock can be accessed in both process
and interrupt context, to prevent preemption by interrupts.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

331/477

Using Spinlocks 2/2

void spin_lock_bh(spinlock_t *lock);

void spin_unlock_bh(spinlock_t *lock);

I
I

Disables software interrupts, but not hardware ones.

Useful to protect shared data accessed in process context and
in a soft interrupt (bottom half).
No need to disable hardware interrupts in this case.

Note that reader / writer spinlocks also exist.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

332/477

Spinlock example
I

Spinlock structure embedded into struct uart_port

struct uart_port {
spinlock_t lock;
/* Other fields */
};

Spinlock taken/released with protection against interrupts

static unsigned int ulite_tx_empty
(struct uart_port *port) {
unsigned long flags;
spin_lock_irqsave(&port->lock, flags);
/* Do something */
spin_unlock_irqrestore(&port->lock, flags);
}

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

333/477

Deadlock Situations
I

They can lock up your system. Make sure they never happen!
Dont call a function that can try to get access to the same
lock

Holding multiple locks is risky!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

334/477

Kernel lock validator

From Ingo Molnar and Arjan van de Ven

I
I

Adds instrumentation to kernel locking code

Detect violations of locking rules during system life, such as:
I

Locks acquired in different order (keeps track of locking

sequences and compares them).
Spinlocks acquired in interrupt handlers and also in process
context when interrupts are enabled.

Not suitable for production systems but acceptable overhead in

development.

See Documentation/locking/lockdep-design.txt for

details

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

335/477

Alternatives to Locking

As we have just seen, locking can have a strong negative

impact on system performance. In some situations, you could
do without it.
I
I

By using lock-free algorithms like Read Copy Update (RCU).

RCU API available in the kernel (See
http://en.wikipedia.org/wiki/RCU).
When available, use atomic operations.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

336/477

Atomic Variables 1/2

I
I
I

Useful when the shared resource is an integer value

Even an instruction like n++ is not guaranteed to be atomic
on all processors!
Atomic operations definitions
I

atomic_t
I

#include <asm/atomic.h>
Contains a signed integer (at least 24 bits)

Atomic operations (main ones)

I

Set or read the counter:

I

void atomic_set(atomic_t *v, int i);

int atomic_read(atomic_t *v);

Operations without return value:

I

void atomic_inc(atomic_t *v);

void atomic_dec(atomic_t *v);

void atomic_add(int i, atomic_t *v);

void atomic_sub(int i, atomic_t *v);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

337/477

Atomic Variables 2/2

Similar functions testing the result:

I

int atomic_inc_and_test(...);

int atomic_dec_and_test(...);

int atomic_sub_and_test(...);

Functions returning the new value:

I

int atomic_inc_return(...);

int atomic_dec_return(...);

int atomic_add_return(...);

int atomic_sub_return(...);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

338/477

Atomic Bit Operations

I

Supply very fast, atomic operations

On most platforms, apply to an unsigned long type.

Apply to a void type on a few others.

Set, clear, toggle a given bit:

void set_bit(int nr, unsigned long * addr);

void clear_bit(int nr, unsigned long * addr);

void change_bit(int nr, unsigned long * addr);

Test bit value:

I

int test_bit(int nr, unsigned long *addr);

Test and modify (return the previous value):

I

int test_and_set_bit(...);

int test_and_clear_bit(...);

int test_and_change_bit(...);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

339/477

Practical lab - Locking

Add locking to the driver to

prevent concurrent accesses to
shared resources

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

340/477

Kernel Debugging

Free Electrons

Kernel Debugging

Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

341/477

Debugging Using Messages

I

Three APIs are available

I

The old printk(), no longer recommended for new debugging

messages
The pr_*() family of functions: pr_emerg(), pr_alert(),
pr_crit(), pr_err(), pr_warning(), pr_notice(),
pr_info(), pr_cont()
and the special pr_debug() (see next page)
I
I

They take a classic format string with arguments

Defined in include/linux/printk.h

The dev_*() family of functions: dev_emerg(),

dev_alert(), dev_crit(), dev_err(), dev_warning(),
dev_notice(), dev_info()
and the special dev_dbg() (see next page)
I

I
I

They take a pointer to struct device as first argument, and

then a format string with arguments
Defined in include/linux/device.h
To be used in drivers integrated with the Linux device model

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

342/477

pr debug() and dev dbg()

When the driver is compiled with DEBUG defined, all these

messages are compiled and printed at the debug level. DEBUG
can be defined by #define DEBUG at the beginning of the
driver, or using ccflags-$(CONFIG_DRIVER) += -DDEBUG
in the Makefile
When the kernel is compiled with CONFIG_DYNAMIC_DEBUG,
then these messages can dynamically be enabled on a per-file,
per-module or per-message basis
I
I

See Documentation/dynamic-debug-howto.txt for details

Very powerful feature to only get the debug messages youre
interested in.

When DEBUG is not defined and CONFIG_DYNAMIC_DEBUG is

not enabled, these messages are not compiled in.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

343/477

Configuring The Priority

Each message is associated to a priority, ranging from 0 for

emergency to 7 for debug.
All the messages, regardless of their priority, are stored in the
kernel log ring buffer
I

Typically accessed using the dmesg command

Some of the messages may appear on the console, depending

on their priority and the configuration of
I

The loglevel kernel parameter, which defines the priority

above which messages are displayed on the console. See
Documentation/kernel-parameters.txt for details.
The value of /proc/sys/kernel/printk, which allows to
change at runtime the priority above which messages are
displayed on the console. See
Documentation/sysctl/kernel.txt for details.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

344/477

DebugFS

A virtual filesystem to export debugging information to user

space.
I

Kernel configuration: DEBUG_FS

I

I
I
I

Kernel hacking -> Debug Filesystem

The debugging interface disappears when Debugfs is

configured out.
You can mount it as follows:
sudo mount -t debugfs none /sys/kernel/debug

First described on http://lwn.net/Articles/115405/

API documented in the Linux Kernel Filesystem API:
I

Documentation/DocBook/filesystems/

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

345/477

DebugFS API
I

Create a sub-directory for your driver:

I

struct dentry *debugfs_create_dir(const char *name,

struct dentry *parent);

Expose an integer as a file in DebugFS:

I

struct dentry *debugfs_create_{u,x}{8,16,32}

(const char *name, mode_t mode, struct dentry *parent,
u8 *value);
I
I

Expose a binary blob as a file in DebugFS:

I

u for decimal representation

x for hexadecimal representation

struct dentry *debugfs_create_blob(const char *name,

mode_t mode, struct dentry *parent,
struct debugfs_blob_wrapper *blob);

Also possible to support writable DebugFS files or customize

the output using the more generic debugfs_create_file()
function.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

346/477

Deprecated Debugging Mechanisms

Some additional debugging mechanisms, whose usage is now

considered deprecated
I

Adding special ioctl() commands for debugging purposes.

DebugFS is preferred.
Adding special entries in the proc filesystem. DebugFS is
preferred.
Adding special entries in the sysfs filesystem. DebugFS is
preferred.
Using printk(). The pr_*() and dev_*() functions are
preferred.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

347/477

Using Magic SysRq

Allows to run multiple debug / rescue commands even when

the kernel seems to be in deep trouble
I
I

Example commands:
I
I
I
I
I

On PC: [Alt] + [SysRq] + <character>

On embedded: break character on the serial line +
<character>
n: makes RT processes nice-able.
w: shows the kernel stack of all sleeping processes
t: shows the kernel stack of all running processes
b: reboot the system
You can even register your own!

Detailed in Documentation/sysrq.txt

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

348/477

kgdb - A Kernel Debugger

The execution of the kernel is fully controlled by gdb from

another machine, connected through a serial line.

Can do almost everything, including inserting breakpoints in

interrupt handlers.

Feature supported for the most popular CPU architectures

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

349/477

Using kgdb 1/2

Details available in the kernel documentation:

Documentation/DocBook/kgdb/

Recommended to turn on CONFIG_FRAME_POINTER to aid in

producing more reliable stack backtraces in gdb.

You must include a kgdb I/O driver. One of them is kgdb

over serial console (kgdboc: kgdb over console, enabled by
CONFIG_KGDB_SERIAL_CONSOLE)
Configure kgdboc at boot time by passing to the kernel:

I
I

kgdboc=<tty-device>,<bauds>.
For example: kgdboc=ttyS0,115200

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

350/477

Using kgdb 2/2

Then also pass kgdbwait to the kernel: it makes kgdb wait

for a debugger connection.

Boot your kernel, and when the console is initialized, interrupt

the kernel with Alt + SysRq + g.
On your workstation, start gdb as follows:

I
I
I

gdb ./vmlinux
(gdb) set remotebaud 115200
(gdb) target remote /dev/ttyS0

Once connected, you can debug a kernel the way you would
debug an application program.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

351/477

Debugging with a JTAG Interface

I

Two types of JTAG dongles

I

The ones offering a gdb compatible interface, over a serial port

or an Ethernet connection. gdb can directly connect to them.
The ones not offering a gdb compatible interface are generally
supported by OpenOCD (Open On Chip Debugger):
http://openocd.sourceforge.net/
I

OpenOCD is the bridge between the gdb debugging language

and the JTAG interface of the target CPU.
See the very complete documentation: http://openocd.
sourceforge.net/documentation/online-docs/
For each board, youll need an OpenOCD configuration file
(ask your supplier)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

352/477

More Kernel Debugging Tips

Enable CONFIG_KALLSYMS_ALL
I

General Setup > Configure standard kernel features

To get oops messages with symbol names instead of raw
addresses
This obsoletes the ksymoops tool

On ARM, if your kernel doesnt boot or hangs without any

message, you can activate early debugging options
(CONFIG_DEBUG_LL and CONFIG_EARLYPRINTK), and add
earlyprintk to the kernel command line.

Techniques to locate the C instruction which caused an oops:

http://j.mp/18oMRHx

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

353/477

Kernel Crash Analysis with kexec/kdump

kexec system call: makes it

possible to call a new kernel,
without rebooting and going
through the BIOS / firmware.

Idea: after a kernel panic, make

the kernel automatically execute a
new, clean kernel from a reserved
location in RAM, to perform
post-mortem analysis of the
memory of the crashed kernel.

See Documentation/kdump/
kdump.txt in the kernel sources
for details.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

354/477

Practical lab - Kernel debugging

Use the dynamic printk feature.

Add debugfs entries

Load a broken driver and see it

crash

Analyze the error information

dumped by the kernel.

Disassemble the code and locate

the exact C instruction which
caused the failure.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

355/477

Porting the Linux Kernel to an ARM Board

Porting the Linux

Kernel to an ARM
Board

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

356/477

Porting the Linux kernel

I
I

The Linux kernel supports a lot of different CPU architectures

Each of them is maintained by a different group of
contributors
I

The organization of the source code and the methods to port

the Linux kernel to a new board are therefore very
architecture-dependent
I

See the MAINTAINERS file for details

For example, some architectures use the Device Tree, some do

not.

This presentation is focused on the ARM architecture only

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

357/477

Architecture, CPU and Machine

I

In the source tree, each architecture has its own directory

I

This directory contains generic ARM code

I

arch/arm for the ARM architecture

boot, common, configs, kernel, lib, mm, nwfpe, vfp,
oprofile, tools

And many directories for different SoC families

I

mach-* directories: mach-pxa for PXA CPUs, mach-imx for

Freescale iMX CPUs, etc.
I

Before the ARM cleanup, these directories contained support

for the SoC family (GPIO, clocks, pinmux, power
management, interrupt controller, etc.) and for the various
boards.
Nowadays, they contain a lot less code, essentially a small
SoC description file, power management and SMP code.

Some CPU types share some code, in directories named

plat-*
Device Tree source files in arch/arm/boot/dts.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

358/477

Before the Device Tree and ARM cleanup

I

Until 2011, the ARM architecture wasnt using the Device

Tree, and a large portion of the SoC support was located in
arch/arm/mach-<foo>.

Each board supported by the kernel was associated to an

unique machine ID.

The entire list of machine ID can be downloaded at

http://www.arm.linux.org.uk/developer/machines/
download.php and one could freely register an additional one.

The Linux kernel was defining a machine structure for each

board, which associates the machine ID with a set of
information and callbacks.

The bootloader had to pass the machine ID to the kernel in a

specific ARM register.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

359/477

The Device Tree and the ARM cleanup

As the ARM architecture gained significantly in popularity,

some major refactoring was needed.

First, the Device Tree was introduced on ARM: instead of

using C code to describe SoCs and boards, a specialized
language is used.
Second, many driver infrastructures were created to replace
custom code in arch/arm/mach-<foo>:

I
I
I
I

The
The
The
The

common clock framework in drivers/clk

pinctrl subsystem in drivers/pinctrl
irqchip subsystem in drivers/irqchip
clocksource subsystem in drivers/clocksource

The amount of code in mach-<foo> has now significantly

reduced.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

360/477

Adding the support for a new ARM board

Provided the SoC used on your board is supported by the Linux

kernel:
1. Create a Device Tree file in arch/arm/boot/dts, generally
named <soc-name>-<board-name>.dts, and make it
include the relevant SoC .dtsi file.
I

Your Device Tree will describe all the SoC peripherals that are
enabled, the pin muxing, as well as all the devices on the
board.

2. Modify arch/arm/boot/dts/Makefile to make sure your

Device Tree gets built as a DTB during the kernel build.
3. If needed, develop the missing device drivers for the devices
that are on your board outside the SoC.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

361/477

Example of the Freescale iMX28 SoCs

The hardware platform used in this training is based on the

AM335x processor from Texas Instruments.

This platform inherits from the OMAP family of TI, for which
kernel support has been around for a long time.

Due to this, and the complexity of the platform, the AM335x

and OMAP support in the kernel hasnt fully migrated yet to
all the infrastructures created during the ARM cleanup.

Therefore, to illustrate this section, we will take the example

of the Freescale iMX28 platform, on which Free Electrons has
worked specifically.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

362/477

Studying the Crystalfontz CFA-10036 platform

Crystalfontz CFA-10036

Uses the Freescale iMX28 SoC, from the

MXS family.

128MB of RAM

1 serial port, 1 LED

1 I2C bus, equipped with an OLED display

1 SD-Card slot

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

363/477

Crystalfontz CFA-10036 Device Tree, header

I

Mandatory Device Tree language definition

/dts-v1/

Include the .dtsi file describing the SoC

#include "im28.dtsi"

Start the root of the tree

A human-readable string to describe the machine

/ {

model = "Crystalfontz CFA-10036 Board";

A list of compatible strings, from the most specific one to the

most general one. Can be used by kernel code to do a SoC or
board-specific check.

compatible = "crystalfontz,cfa10036", "fsl,imx28";

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

364/477

Crystalfontz CFA-10036 Device Tree, chosen/memory

Definition of the default kernel command line. Some

additional operating-system specific entries can be added in
chosen:

chosen {
bootargs = "console=ttyS0,115200 earlyprintk";
};
I

Definition of the size and location of the RAM:

memory {
device_type = "memory";
reg = <0x40000000 0x8000000>; /* 128 MB */
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

365/477

Crystalfontz CFA-10036, bus/UART

Start of the internal SoC peripherals.

apb@80000000 {
apbh@80000000 {
apbx@80040000 {

The CFA-10036 has one debug UART, so the corresponding

controller is enabled:

duart: serial@80074000 {
pinctrl-names = "default";
pinctrl-0 = <&duart_pins_b>;
status = "okay";
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

366/477

Crystalfontz CFA-10036 Device Tree, Muxing

I

Definition of a few pins that will be muxed as GPIO, for LEDs

and reset.

pinctrl@80018000 {
ssd1306_cfa10036: ssd1306-10036@0 {
reg = <0>;
fsl,pinmux-ids = <
0x2073 /* MX28_PAD_SSP0_D7__GPIO_2_7 */
>;
fsl,drive-strength = <0>;
fsl,voltage = <1>;
fsl,pull-up = <0>;
};
led_pins_cfa10036: leds-10036@0 {
reg = <0>;
fsl,pinmux-ids = <
0x3043 /* MX28_PAD_AUART1_RX__GPIO_3_4 */
>;
fsl,drive-strength = <0>;
fsl,voltage = <1>;
fsl,pull-up = <0>;
};
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

367/477

Crystalfontz CFA-10036 Device Tree, LED

One LED is connected to this platform. Note the reference to

the led_pins_cfa10036 muxing configuration.

leds {
compatible = "gpio-leds";
pinctrl-names = "default";
pinctrl-0 = <&led_pins_cfa10036>;
power {
gpios = <&gpio3 4 1>;
default-state = "on";
};
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

368/477

Crystalfontz CFA-10036 Device Tree, SD Card/USB

The platform also has a USB port

usb0: usb@80080000 {
pinctrl-names = "default";
pinctrl-0 = <&usb0_otg_cfa10036>;
status = "okay";
};

and an SD Card slot:

ssp0: ssp@80010000 {
compatible = "fsl,imx28-mmc";
pinctrl-names = "default";
pinctrl-0 = <&mmc0_4bit_pins_a
&mmc0_cd_cfg &mmc0_sck_cfg>;
bus-width = <4>;
status = "okay";
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

369/477

Crystalfontz CFA-10036 Device Tree, I2C bus

An I2C bus, with a Solomon SSD1306 OLED display

connected on it:

i2c0: i2c@80058000 {
pinctrl-names = "default";
pinctrl-0 = <&i2c0_pins_b>;
clock-frequency = <400000>;
status = "okay";
ssd1306: oled@3c {
compatible = "solomon,ssd1306fb-i2c";
pinctrl-names = "default";
pinctrl-0 = <&ssd1306_cfa10036>;
reg = <0x3c>;
reset-gpios = <&gpio2 7 0>;
solomon,height = <32>;
solomon,width = <128>;
solomon,page-offset = <0>;
};
};

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

370/477

Crystalfontz CFA-10036 Device Tree, Breakout Boards

The CFA-10036 can be plugged in other breakout boards, and

the device tree also allows us to describe this, using includes.
For example, the CFA-10057:

#include "imx28-cfa10036.dts"

This allows to have a layered description. This can also be

done for boards that have a lot in common, like the
BeagleBone and the BeagleBone Black, or the AT91
SAMA5D3-based boards.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

371/477

Crystalfontz CFA-10036: build the DTB

To ensure that the Device Tree Blob gets built for this board
Device Tree Source, one need to ensure it is listed in
arch/arm/boot/dts/Makefile:

dtb-$(CONFIG_ARCH_MXS) += imx28-cfa10036.dtb \
imx28-cfa10037.dtb \
imx28-cfa10049.dtb \
imx28-cfa10055.dtb \
imx28-cfa10056.dtb \
imx28-cfa10057.dtb \
imx28-cfa10058.dtb \
imx28-evk.dtb

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

372/477

Understanding the SoC support

I

Lets consider another ARM platform here, the Marvell

Armada 370/XP.

For this platform, the core of the SoC support is located in

arch/arm/mach-mvebu
The armada-370-xp.c (see code on the next slide) contains
the entry point of the SoC definition, the
DT_MACHINE_START .. MACHINE_END definition:

Defines the list of platform compatible strings that will match

this platform, in this case marvell,armada-370-xp. This
allows the kernel to know which DT_MACHINE structure to use
depending on the DTB that is passed at boot time.
Defines various callbacks for the platform initialization, the
most important one being the .init_machine callback, which
calls of_platform_populate(). This function travels
through the Device Tree and instantiates all the devices.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

373/477

arch/arm/mach-mvebu/armada-370-xp.c

static void __init armada_370_xp_dt_init(void)

{
of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
}
static const char * const armada_370_xp_dt_compat[] = {
"marvell,armada-370-xp",
NULL,
};
DT_MACHINE_START(ARMADA_XP_DT, "Marvell Armada 370/XP (Device Tree)")
.smp
= smp_ops(armada_xp_smp_ops),
.init_machine
= armada_370_xp_dt_init,
.map_io
= armada_370_xp_map_io,
.init_time
= armada_370_xp_timer_and_clk_init,
.restart
= mvebu_restart,
.dt_compat
= armada_370_xp_dt_compat,
MACHINE_END

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

374/477

Components of the minimal SoC support

The minimal SoC support consists in

I An SoC entry point file, arch/arm/mach-mvebu/armada-370-xp.c
I At least one SoC .dtsi DT and one board .dts DT, in

arch/arm/boot/dts
I A interrupt controller driver, drivers/irqchip/irq-armada-370-xp.c
I A timer driver, drivers/clocksource/time-armada-370-xp.c
I An earlyprintk implementation to get early messages from the console,

arch/arm/Kconfig.debug and arch/arm/include/debug

I A serial port driver in drivers/tty/serial. For Armada 370/XP, the

8250 driver drivers/tty/serial/8250 is used.

This allows to boot a minimal system up to user space, using a

root filesystem in initramfs.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

375/477

Extending the minimal SoC support

Once the minimal SoC support is in place, the following core
components should be added:
I

Support for the clocks. Usually requires some clock drivers, as

well as DT representations of the clocks. See
drivers/clk/mvebu for Armada 370/XP clock drivers.

Support for pin muxing, through the pinctrl subsystem. See

drivers/pinctrl/mvebu for the Armada 370/XP drivers.

Support for GPIOs, through the GPIO subsystem. See

drivers/gpio/gpio-mvebu.c for the Armada 370/XP
GPIO driver.

Support for SMP, through struct smp_operations. See

arch/arm/mach-mvebu/platsmp.c.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

376/477

Adding device drivers

Once the core pieces of the SoC support have been implemented,
the remaining part is to add drivers for the different hardware
blocks:
I

Ethernet driver, in drivers/net/ethernet/mvneta.c

SATA driver, in drivers/ata/sata_mv.c

I2C driver, in drivers/i2c/busses/i2c-mv64xxx.c

SPI driver, in drivers/spi/spi-orion.c

PCIe driver, in drivers/pci/host/pci-mvebu.c

USB driver, in drivers/usb/host/ehci-orion.c

etc.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

377/477

Power Management

Power
Management

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

378/477

PM Building Blocks

Several power management building blocks

I
I
I
I
I

Suspend and resume

CPUidle
Runtime power management
Frequency and voltage scaling
Applications

Independent building blocks that can be improved gradually

during development

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

379/477

Clock Framework (1)

Generic framework to manage clocks used by devices in the

system

Allows to reference count clock users and to shutdown the

unused clocks to save power
Simple API described in
Documentation/DocBook/kernel-api/clk.html .

I
I
I
I
I

clk_get() to get a reference to a clock

clk_enable() to start the clock
clk_disable() to stop the clock
clk_put() to free the clock source
clk_get_rate() to get the current rate

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

380/477

Clock Framework (2)

The common clock framework

I

I
I

Allows to declare the available clocks and their association to

devices in the Device Tree (preferred) or statically in the
source code (old method)
Provides a debugfs representation of the clock tree
Is implemented in drivers/clk

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

381/477

Diagram overview of the common clock framework

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

382/477

Clock Framework (3)

The interface of the CCF divided into two halves:

I Common Clock Framework core
I
I

Common definition of struct clk

Common implementation of the clk.h API (defined in
drivers/clk/clk.c)
struct clk_ops: operations invoked by the clk API
implementation
Not supposed to be modified when adding a new driver

Hardware-specific
I

Callbacks registered with struct clk_ops and the

corresponding hardware-specific structures
Has to be written for each new hardware clock

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

383/477

Clock Framework (4)

Hardware clock operations: device tree

I

The device tree is the mandatory way to declare a clock and

to get its resources, as for any other driver using DT we have
to:
I

Parse the device tree to setup the clock: the resources but
also the properties are retrieved.
Declare the compatible clocks and associate it with an
initialization function using CLK_OF_DECLARE

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

384/477

Suspend and Resume

I
I

Infrastructure in the kernel to support suspend and resume

Platform hooks
I

I
I

prepare(), enter(), finish(), valid() in a

struct platform_suspend_ops structure
Registered using the suspend_set_ops() function
See arch/arm/mach-at91/pm.c

Device drivers
I

suspend() and resume() hooks in the *_driver structures

(struct platform_driver, struct usb_driver, etc.)
See drivers/net/macb.c

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

385/477

Board-specific Power Management

Typically takes care of battery and charging management.

Also defines presuspend and postsuspend handlers.

Example: arch/arm/mach-pxa/spitz_pm.c

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

386/477

arch/arm/mach-cpu/sleep.S

Assembly code implementing CPU specific suspend and

resume code.

Note: only found on arm, just 3 other occurrences in other

architectures, with other paths.

First scenario: only a suspend function. The code goes in

sleep state (after enabling DRAM self-refresh), and continues
with resume code.

Second scenario: suspend and resume functions. Resume

functions called by the bootloader.
Examples to look at:

I
I

arch/arm/mach-omap2/sleep24xx.S (1st case)

arch/arm/mach-pxa/sleep.S (2nd case)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

387/477

Triggering Suspend

Whatever the power management implementation, CPU

specific struct suspend_ops functions are called by the
enter_state() function.

enter_state() also takes care of executing the suspend and

resume functions for your devices.
The execution of this function can be triggered from user
space. To suspend to RAM:

echo mem > /sys/power/state

Can also use the s2ram program from

http://suspend.sourceforge.net/

Read kernel/power/suspend.c

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

388/477

Runtime Power Management

According to the kernel configuration interface: Enable

functionality allowing I/O devices to be put into energy-saving
(low power) states at run time (or autosuspended) after a
specified period of inactivity and woken up in response to a
hardware-generated wake-up event or a drivers request.

New hooks must be added to the drivers:

runtime_suspend(), runtime_resume(), runtime_idle()

API and details on Documentation/power/runtime_pm.txt

See also Kevin Hilmans presentation at ELC Europe 2010:

http://elinux.org/images/c/cd/ELC-2010-khilmanRuntime-PM.odp

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

389/477

Saving Power in the Idle Loop

The idle loop is what you run when theres nothing left to run
in the system.

Implemented in all architectures in

arch/<arch>/kernel/process.c

Example to read: look for cpu_idle in

arch/arm/kernel/process.c

Each ARM cpu defines its own arch_idle function.

The CPU can run power saving HLT instructions, enter NAP
mode, and even disable the timers (tickless systems).

See also http://en.wikipedia.org/wiki/Idle_loop

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

390/477

Managing Idle

Adding support for multiple idle levels

I

Modern CPUs have several sleep states offering different power

savings with associated wake up latencies
Since 2.6.21, the dynamic tick feature allows to remove the
periodic tick to save power, and to know when the next event
is scheduled, for smarter sleeps.
CPUidle infrastructure to change sleep states
I

I
I

Platform-specific driver defining sleep states and transition

operations
Platform-independent governors (ladder and menu)
Available for x86/ACPI, not supported yet by all ARM cpus.
(look for cpuidle* files under arch/arm/)
See Documentation/cpuidle/ in kernel sources.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

391/477

PowerTOP

https://01.org/powertop/
I

I
I
I

With dynamic ticks, allows to fix parts of kernel code and

applications that wake up the system too often.
PowerTOP allows to track the worst offenders
Now available on ARM cpus implementing CPUidle
Also gives you useful hints for reducing power.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

392/477

Frequency and Voltage Scaling (1)

Frequency and voltage scaling possible through the cpufreq

kernel infrastructure.
I

Generic infrastructure: drivers/cpufreq/cpufreq.c and

include/linux/cpufreq.h
Generic governors, responsible for deciding frequency and
voltage transitions
I
I
I
I

I
I

performance: maximum frequency

powersave: minimum frequency
ondemand: measures CPU consumption to adjust frequency
conservative: often better than ondemand. Only increases
frequency gradually when the CPU gets loaded.
userspace: leaves the decision to a user space daemon.

This infrastructure can be controlled from

/sys/devices/system/cpu/cpu<n>/cpufreq/

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

393/477

Frequency and Voltage Scaling (2)

CPU support code in architecture dependent files. Example to

read: arch/arm/plat-omap/cpu-omap.c
Must implement the operations of the cpufreq_driver
structure and register them using
cpufreq_register_driver()
I
I
I
I

init() for initialization

exit() for cleanup
verify() to verify the user-chosen policy
setpolicy() or target() to actually perform the frequency
change

See Documentation/cpu-freq/ for useful explanations

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

394/477

PM Quality Of Service interface

I

Kernel and user mode interface for registering performance

expectations by drivers, subsystems and user space
applications.
Two different PM QoS frameworks are available:
I

PM QoS classes for CPU DMA latency, network latency and

and network throughput.
The per-device PM QoS framework API to manage per-device
latency.

According to these requirements, PM QoS allows kernel

drivers to adjust their power management

See Documentation/power/pm_qos_interface.txt

Still in very early deployment (only used in about 15 drivers in

3.12).

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

395/477

Regulator Framework

Modern embedded hardware have hardware responsible for

voltage and current regulation
The regulator framework allows to take advantage of this
hardware to save power when parts of the system are unused
I
I
I
I

A consumer interface for device drivers (i.e users)

Regulator driver interface for regulator drivers
Machine interface for board configuration
sysfs interface for user space

Merged in Linux 2.6.27.

See Documentation/power/regulator/ in kernel sources.

See Liam Girdwoods presentation at ELC 2008

http://free-electrons.com/blog/elc-2008report#girdwood

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

396/477

BSP Work for a New Board

In case you just need to create a BSP for your board, and your
CPU already has full PM support, you should just need to:
I
I

I
I

I
I

Create clock definitions and bind your devices to them.

Implement PM handlers (suspend, resume) in the drivers for
your board specific devices.
Implement runtime PM handlers in your drivers.
Implement board specific power management if needed (mainly
battery management)
Implement regulator framework hooks for your board if needed.
All other parts of the PM infrastructure should be already
there: suspend / resume, cpuidle, cpu frequency and voltage
scaling.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

397/477

Useful Resources

Documentation/power/ in the Linux kernel sources.

I

Will give you many useful details.

http://wiki.linaro.org/WorkingGroups/PowerManagement/
I Ongoing developments on the ARM platform.

Tips and ideas for prolonging battery life

I

http://j.mp/fVdxKh

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

398/477

The kernel development and contribution process

The kernel
development and
contribution
process

Free Electrons
Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

399/477

The kernel development and contribution process

Linux versioning scheme and

development process

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

400/477

Until 2.6 (1)

One stable major branch every 2 or 3 years

I
I

One development branch to integrate new functionalities and

major changes
I
I
I

Identified by an even middle number

Examples: 1.0.x, 2.0.x, 2.2.x, 2.4.x

Identified by an odd middle number

Examples: 2.1.x, 2.3.x, 2.5.x
After some time, a development version becomes the new base
version for the stable branch

Minor releases once in while: 2.2.23, 2.5.12, etc.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

401/477

Until 2.6 (2)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

402/477

Changes since Linux 2.6

Since 2.6.0, kernel developers have been able to introduce

lots of new features one by one on a steady pace, without
having to make disruptive changes to existing subsystems.

Since then, there has been no need to create a new

development branch massively breaking compatibility with the
stable branch.

Thanks to this, more features are released to users at a

faster pace.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

403/477

3.x stable branch

From 2003 to 2011, the official kernel versions were named

2.6.x.

Linux 3.0 was released in July 2011

This is only a change to the numbering scheme

I
I

Official kernel versions are now named 3.x (3.0, 3.1, 3.2,
etc.)
Stabilized versions are named 3.x.y (3.0.2, 3.4.3, etc.)
It effectively only removes a digit compared to the previous
numbering scheme

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

404/477

New development model

Using merge and bug fixing windows

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

405/477

New development model - Details

After the release of a 3.x version (for example), a two-weeks

merge window opens, during which major additions are
merged.

The merge window is closed by the release of test version

3.(x+1)-rc1

The bug fixing period opens, for 6 to 10 weeks.

At regular intervals during the bug fixing period,

3.(x+1)-rcY test versions are released.

When considered sufficiently stable, kernel 3.(x+1) is

released, and the process starts again.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

406/477

More stability for the kernel source tree

I

Issue: bug and security fixes only released

for most recent stable kernel versions.

Some people need to have a recent kernel,

but with long term support for security
updates.

You could get long term support from a

commercial embedded Linux provider.

You could reuse sources for the kernel

used in Ubuntu Long Term Support
releases (5 years of free security updates).

The http://kernel.org front page

shows which versions will be supported for
some time (up to 2 or 3 years), and which
ones wont be supported any more
(EOL: End Of Life)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

407/477

Whats new in each Linux release?

I

The official list of changes for each Linux release is just a

huge list of individual patches!
commit aa6e52a35d388e730f4df0ec2ec48294590cc459
Author: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Date:
Wed Jul 13 11:29:17 2011 +0200
at91: at91-ohci: support overcurrent notification
Several USB power switches (AIC1526 or MIC2026) have a digital output
that is used to notify that an overcurrent situation is taking
place. This digital outputs are typically connected to GPIO inputs of
the processor and can be used to be notified of these overcurrent
situations.
Therefore, we add a new overcurrent_pin[] array in the at91_usbh_data
structure so that boards can tell the AT91 OHCI driver which pins are
used for the overcurrent notification, and an overcurrent_supported
boolean to tell the driver whether overcurrent is supported or not.
The code has been largely borrowed from ohci-da8xx.c and
ohci-s3c2410.c.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Nicolas Ferre <nicolas.ferre@atmel.com>

Very difficult to find out the key changes and to get the global
picture out of individual changes.

Fortunately, there are some useful resources available

I
I
I

http://wiki.kernelnewbies.org/LinuxChanges
http://lwn.net
http://linuxfr.org, for French readers

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

408/477

The kernel development and contribution process

Contributing to the Linux kernel

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

409/477

Solving Issues

If you face an issue, and it doesnt look specific to your work

but rather to the tools you are using, it is very likely that
someone else already faced it.

Search the Internet for similar error reports.

You have great chances of finding a solution or workaround, or

at least an explanation for your issue.

Otherwise, reporting the issue is up to you!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

410/477

Getting Help

I
I

If you have a support contract, ask your vendor.

Otherwise, dont hesitate to share your questions and issues
I

Either contact the Linux mailing list for your architecture (like
linux-arm-kernel or linuxsh-dev...).
Or contact the mailing list for the subsystem youre dealing
with (linux-usb-devel, linux-mtd...). Dont ask the maintainer
directly!
Most mailing lists come with a FAQ page. Make sure you read
it before contacting the mailing list.
Useful IRC resources are available too (for example on
http://kernelnewbies.org).
Refrain from contacting the Linux Kernel mailing list, unless
youre an experienced developer and need advice.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

411/477

Reporting Linux Bugs

First make sure youre using the latest version

Make sure you investigate the issue as much as you can: see
Documentation/BUG-HUNTING

Check for previous bugs reports. Use web search engines,

accessing public mailing list archives.

If the subsystem you report a bug on has a mailing list, use it.
Otherwise, contact the official maintainer (see the
MAINTAINERS file). Always give as many useful details as
possible.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

412/477

How to Become a Kernel Developer?

Recommended resources
I

See Documentation/SubmittingPatches for guidelines and

http://kernelnewbies.org/UpstreamMerge for very
helpful advice to have your changes merged upstream (by Rik
van Riel).
Watch the Write and Submit your first Linux kernel Patch talk
by Greg. K.H:
http://www.youtube.com/watch?v=LLBrBBImJt4
How to Participate in the Linux Community (by Jonathan
Corbet) A Guide To The Kernel Development Process
http://j.mp/tX2Ld6

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

413/477

Contribute to the Linux Kernel (1)

I

Clone Linus Torvalds tree:

I

Keep your tree up to date

I

git clone git://git.kernel.org/pub/scm/linux/

kernel/git/torvalds/linux.git
git pull

Look at the master branch and check whether your issue /

change hasnt been solved / implemented yet. Also check the
maintainers git tree and mailing list (see the MAINTAINERS
file).You may miss submissions that are not in mainline yet.
If the maintainer has its own git tree, create a remote branch
tracking this tree. This is much better than creating another
clone (doesnt duplicate common stuff):
I

git remote add linux-omap git://git.kernel.org/

pub/scm/linux/kernel/git/tmlind/linux-omap.git
git fetch linux-omap

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

414/477

Contribute to the Linux Kernel (2)

I

Either create a new branch starting from the current commit

in the master branch:
I

git checkout -b feature

Or, if more appropriate, create a new branch starting from the

maintainers master branch:
I

git checkout -b feature linux-omap/master (remote

tree / remote branch)

In your new branch, implement your changes.

Test your changes (must at least compile them).

Run git add to add any new files to the index.

Check that each file you modified is ready for submission:

scripts/checkpatch.pl --strict --file <file>

If needed, fix indenting rule violations:

I

indent -linux <file>

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

415/477

Configure git send-email

Make sure you already have configured your name and e-mail
address (should be done before the first commit).
I
I

git config --global user.name My Name

git config --global user.email me@mydomain.net

Configure your SMTP settings. Example for a Google Mail

account:
I

I
I
I

git config -global sendemail.smtpserver smtp.googlemail.com

git config --global sendemail.smtpserverport 587
git config --global sendemail.smtpencryption tls
git config -global sendemail.smtpuser jdoe@gmail.com
git config --global sendemail.smtppass xxx

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

416/477

Contribute to the Linux Kernel (3)

I

Group your changes by sets of logical changes, corresponding

to the set of patches that you wish to submit.
Commit and sign these groups of changes (signing required by
Linux developers).
I
I

The easiest way is to look at previous commit summaries on

the main file you modify
I

git commit -s
Make sure your first description line is a useful summary and
starts with the name of the modified subsystem. This first
description line will appear in your e-mails

git log --pretty=oneline <path-to-file>

Examples subject lines ([PATCH] omitted):

Documentation: prctl/seccomp_filter
PCI: release busn when removing bus
ARM: add support for xz kernel decompression

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

417/477

Contribute to the Linux Kernel (4)

I

Remove previously generated patches

I

rm 00*.patch

Have git generate patches corresponding to your branch

I

If your branch is based on mainline

I

scripts/checkpatch.pl --strict 00*.patch

Now, send your patches to yourself

I

git format-patch <remote>/<branch>..<your branch>

You can run a last check on all your patches (easy)

I

git format-patch master..<your branch>

If your branch is based on a remote branch

git send-email --compose -to me@mydomain.com 00*.patch

If you have just one patch, or a trivial patch, you can remove
the empty line after In-Reply-To:. This way, you wont add
a summary e-mail introducing your changes (recommended
otherwise).

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

418/477

Contribute to the Linux Kernel (5)

I
I

Check that you received your e-mail properly, and that it looks
good.
Now, find the maintainers for your patches
scripts/get_maintainer.pl ~/patches/00*.patch
Russell King <linux@arm.linux.org.uk> (maintainer:ARM PORT)
Nicolas Pitre <nicolas.pitre@linaro.org>
(commit_signer:1/1=100%)
linux-arm-kernel@lists.infradead.org (open list:ARM PORT)
linux-kernel@vger.kernel.org (open list)

Now, send your patches to each of these people and lists

I

git send-email --compose --to linux@arm.linux.

org.uk --to nicolas.pitre@linaro.org --to linuxarm-kernel@lists.infradead.org --to linuxkernel@vger.kernel.org 00*.patch

Wait for replies about your changes, take the comments into
account, and resubmit if needed, until your changes are
eventually accepted.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

419/477

Contribute to the Linux Kernel (6)

I

If you use git format-patch to produce your patches, you

will need to update your branch and may need to group your
changes in a different way (one patch per commit).
Heres what we recommend
I

Update your master branch

I

Back to your branch, implement the changes taking

community feedback into account. Commit these changes.
Still in your branch: reorganize your commits and commit
messages
I
I

git checkout master; git pull

git rebase --interactive origin/master

git rebase allows to rebase (replay) your changes starting
from the latest commits in master. In interactive mode, it also
allows you to merge, edit and even reorder commits, in an
interactive way.

Third, generate the new patches with git format-patch.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

420/477

Kernel Resources

Free Electrons

Kernel Resources

Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

421/477

Kernel Development News

Linux Weekly News

I

http://lwn.net/

The weekly digest off all Linux and free software information
sources

In depth technical discussions about the kernel

Subscribe to finance the editors ($7 / month)

Articles available for non subscribers after 1 week.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

422/477

Useful Reading (1)

Essential Linux Device Drivers, April 2008
I

http://elinuxdd.com/

By Sreekrishnan Venkateswaran, an
embedded IBM engineer with more than
10 years of experience

Covers a wide range of topics not covered

by LDD: serial drivers, input drivers, I2C,
PCMCIA and Compact Flash, PCI, USB,
video drivers, audio drivers, block drivers,
network drivers, Bluetooth, IrDA, MTD,
drivers in user space, kernel debugging,
etc.

Probably the most wide ranging and

complete Linux device driver book Ive
read Alan Cox

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

423/477

Useful Reading (2)

Writing Linux Device drivers, September 2009
I

http://www.coopj.com/

Self published by Jerry Cooperstein

Available like any other book (Amazon

and others)

Though not as thorough as the previous

book on specific drivers, still a good
complement on multiple aspects of kernel
and device driver development.

Based on Linux 2.6.31

Multiple exercises. Updated solutions for

2.6.36.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

424/477

Useful Reading (3)

Linux Device Drivers, 4th edition, July 2015
(estimated)
I

http://shop.oreilly.com/product/
0636920030867.do

By Jonathan Corbet, Alessandro Rubini,

Greg Kroah-Hartman, Jessica McKellar,
OReilly

Expected to be a great book, if as good

as the previous edition (Free PDF:
http://free-electrons.com/
community/kernel/ldd3/), which is
now out of date.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

425/477

Useful Reading (4)

Linux Kernel Development, 3rd Edition,

Jun 2010
I
I

Robert Love, Novell Press

http://free-electrons.com/redir/
lkd3-book.html
A very synthetic and pleasant way to
learn about kernel subsystems (beyond
the needs of device driver writers)

The Linux Programming Interface, Oct

2010
I
I
I

Michael Kerrisk, No Starch Press

http://man7.org/tlpi/
A gold mine about the kernel interface
and how to use it

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

426/477

Useful Online Resources

I

Kernel documentation (Documentation/ in kernel sources)

I

Linux kernel mailing list FAQ

I
I
I

http://www.tux.org/lkml/
Complete Linux kernel FAQ
Read this before asking a question to the mailing list

Kernel Newbies
I
I

Available on line:
http://free-electrons.com/kerneldoc/ (with HTML
documentation extracted from source code)

http://kernelnewbies.org/
Glossary, articles, presentations, HOWTOs, recommended
reading, useful tools for people getting familiar with Linux
kernel or driver development.

Kernel glossary
I

http://kernelnewbies.org/KernelGlossary

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

427/477

International Conferences
I

Embedded Linux Conference:

http://embeddedlinuxconference.com/
I
I
I
I

Linux Plumbers: http://linuxplumbersconf.org

I

Conference on the low-level plumbing of Linux: kernel, audio,

power management, device management, multimedia, etc.

linux.conf.au: http://linux.org.au/conf/
I
I

Organized by the Linux Foundation:

in California (San Francisco, in the spring)
in Europe (October-November)
Very interesting kernel and user space topics for embedded
systems developers.
Presentation slides freely available

In Australia / New Zealand

Features a few presentations by key kernel hackers.

Dont miss our free conference videos on http://freeelectrons.com/community/videos/conferences/

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

428/477

ARM resources
I

ARM Linux project: http://www.arm.linux.org.uk/

I

Linaro: http://linaro.org
I

Many optimizations and resources for recent ARM CPUs

(toolchains, kernels, debugging utilities...).

ARM Limited: http://www.linux-arm.com/

I

Developer documentation:
http://www.arm.linux.org.uk/developer/
linux-arm-kernel mailing list:
http://lists.infradead.org/mailman/listinfo/linuxarm-kernel
FAQ:
http://www.arm.linux.org.uk/armlinux/mlfaq.php

Wiki with links to useful developer resources

See our Embedded Linux course for details about toolchains:

http://free-electrons.com/training/embedded-linux/

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

429/477

Last slides

Free Electrons

Last slides

Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

430/477

Last slide

Thank you!
And may the Source be with you

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

431/477

Backup slides

Free Electrons

Backup slides

Embedded Linux
Developers

Free Electrons

Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Corrections, suggestions, contributions and translations are welcome!

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

432/477

Backup slides

DMA

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

433/477

DMA Integration

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

434/477

Peripheral DMA

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

435/477

DMA Controllers

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

436/477

DMA descriptors

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

437/477

Backup slides

DMA Usage

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

438/477

Constraints with a DMA

A DMA deals with physical addresses, so:

I

Programming a DMA requires retrieving a physical address at

some point (virtual addresses are usually used)
The memory accessed by the DMA shall be physically
contiguous

The CPU can access memory through a data cache

I

Using the cache can be more efficient (faster accesses to the

cache than the bus)
But the DMA does not access to the CPU cache, so one need
to take care of cache coherency (cache content vs memory
content)
Either flush or invalidate the cache lines corresponding to the
buffer accessed by DMA and processor at strategic times

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

439/477

DMA Memory Constraints

Need to use contiguous memory in physical space.

Can use any memory allocated by kmalloc() (up to 128 KB)

or __get_free_pages() (up to 8MB).

Can use block I/O and networking buffers, designed to

support DMA.

Can not use vmalloc() memory (would have to setup DMA

on each individual physical page).

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

440/477

Memory Synchronization Issues

Memory caching could interfere with DMA

I

Before DMA to device

I

After DMA from device

I

Need to make sure that all writes to DMA buffer are

committed.
Before drivers read from DMA buffer, need to make sure that
memory caches are flushed.

Bidirectional DMA
I

Need to flush caches before and after the DMA transfer.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

441/477

Linux DMA API

The kernel DMA utilities can take care of:

I
I
I
I

Either allocating a buffer in a cache coherent area,

Or making sure caches are flushed when required,
Managing the DMA mappings and IOMMU (if any).
See Documentation/DMA-API.txt for details about the
Linux DMA generic API.
Most subsystems (such as PCI or USB) supply their own DMA
API, derived from the generic one. May be sufficient for most
needs.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

442/477

Coherent or Streaming DMA Mappings

Coherent mappings
I

I
I
I
I

The kernel allocates a suitable buffer and sets the mapping for
the driver.
Can simultaneously be accessed by the CPU and device.
So, has to be in a cache coherent memory area.
Usually allocated for the whole time the module is loaded.
Can be expensive to setup and use on some platforms.

Streaming mappings
I

I
I

The kernel just sets the mapping for a buffer provided by the
driver.
Use a buffer already allocated by the driver.
Mapping set up for each transfer. Keeps DMA registers free on
the hardware.
The recommended solution.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

443/477

Allocating Coherent Mappings

The kernel takes care of both buffer allocation and mapping

#include <asm/dma-mapping.h>
void *
dma_alloc_coherent(
struct device *dev,
size_t size,
dma_addr_t *handle,
gfp_t gfp
);

/* Output: buffer address */

/*
/*
/*
/*

device structure */
Needed buffer size in bytes */
Output: DMA bus address */
Standard GFP flags */

void dma_free_coherent(struct device *dev,

size_t size, void *cpu_addr, dma_addr_t handle);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

444/477

Setting up streaming mappings

Works on buffers already allocated by the driver

#include <linux/dmapool.h>
dma_addr_t dma_map_single(
struct device *,
void *,
size_t,
enum dma_data_direction

/*
/*
/*
/*
*
*

device structure */
input: buffer to use */
buffer size */
Either DMA_BIDIRECTIONAL,
DMA_TO_DEVICE or
DMA_FROM_DEVICE */

);
void dma_unmap_single(struct device *dev, dma_addr_t handdle,
size_t size, enum dma_data_direction dir);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

445/477

DMA Streaming Mapping Notes

When the mapping is active: only the device should access

the buffer (potential cache issues otherwise).

The CPU can access the buffer only after unmapping!

Another reason: if required, this API can create an

intermediate bounce buffer (used if the given buffer is not
usable for DMA).

The Linux API also supports scatter / gather DMA streaming

mappings.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

446/477

Backup slides

DMA transfers

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

447/477

Starting DMA transfers

If the device youre writing a driver for is doing peripheral

DMA, no external API is involved.
If it relies on an external DMA controller, youll need to
I

Ask the hardware to use DMA, so that it will drive its request
line
Use Linux DMAEngine framework, especially its slave API

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

448/477

DMAEngine Slave API

I

In order to start a DMA transfer, you need to call the

following functions from your driver
1. Request a channel for exclusive use with
dma_request_channel, or one of its variants
2. Configure it for our use case, by filling a dma_slave_config
structures with various parameters (source and destination
adresses, accesses width, etc.) and passing it as an argument
to dmaengine_slave_config
3. Start a new transaction with dmaengine_prep_slave_single
or dmaengine_prep_slave_sg
4. Put the transaction in the driver pending queue using
dmaengine_submit
5. And finally ask the driver to process all pending transactions
using dmaengine_issue_pending

I
I

Of course, this needs to be done in addition to the DMA

mapping seen previously
Some framework abstract it away from you, such as SPI and
ASoC

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

449/477

Backup slides

mmap

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

450/477

mmap

Possibility to have parts of the virtual address space of a

program mapped to the contents of a file

Particularly useful when the file is a device file

Allows to access device I/O memory and ports without having

to go through (expensive) read, write or ioctl calls
One can access to current mapped files by two means:

I
I

/proc/<pid>/maps
pmap <pid>

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

451/477

/proc/<pid>/maps

start-end
...
7f4516d04000-7f4516d06000
7f4516d07000-7f4516d0b000
...
7f4518728000-7f451874f000
7f451874f000-7f451894f000
7f451894f000-7f4518951000
7f4518951000-7f4518952000
...
7f451da4f000-7f451dc3f000
7f451de3e000-7f451de41000
7f451de41000-7f451de4c000
...

perm offset major:minor inode

mapped file name

rw-s 1152a2000 00:05 8406

rw-s 120f9e000 00:05 8406

/dev/dri/card0
/dev/dri/card0

r-xp
---p
r--p
rw-p

/lib/x86_64-linux-gnu/libexpat.so.1.5.2
/lib/x86_64-linux-gnu/libexpat.so.1.5.2
/lib/x86_64-linux-gnu/libexpat.so.1.5.2
/lib/x86_64-linux-gnu/libexpat.so.1.5.2

00000000
00027000
00027000
00029000

08:01
08:01
08:01
08:01

268909
268909
268909
268909

r-xp 00000000 08:01 1549

r--p 001ef000 08:01 1549
rw-p 001f2000 08:01 1549

/usr/bin/Xorg
/usr/bin/Xorg
/usr/bin/Xorg

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

452/477

mmap Overview

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

453/477

How to Implement mmap - User Space

I
I

Open the device file

Call the mmap system call (see man mmap for details):
void * mmap(
void *start,
size_t length,
int prot,
int flags,
int fd,
off_t offset
);

/*
/*
/*
/*
/*
/*

Often 0, preferred starting address */

Length of the mapped area */
Permissions: read, write, execute */
Options: shared mapping, private copy... */
Open file descriptor */
Offset in the file */

You get a virtual address you can write to or read from.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

454/477

How to Implement mmap - Kernel Space

Character driver: implement an mmap file operation and add it

to the driver file operations:
int (*mmap) (
struct file *,
struct vm_area_struct *
);

/* Open file structure */

/* Kernel VMA structure */

Initialize the mapping.

I

Can be done in most cases with the remap_pfn_range()

function, which takes care of most of the job.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

455/477

remap pfn range()

pfn: page frame number

The most significant bits of the page address (without the

bits corresponding to the page size).

#include <linux/mm.h>
int remap_pfn_range(
struct vm_area_struct *, /* VMA struct */
unsigned long virt_addr, /* Starting user
* virtual address */
unsigned long pfn,
/* pfn of the starting
* physical address */
unsigned long size,
/* Mapping size */
pgprot_t prot
/* Page permissions */
);

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

456/477

Simple mmap Implementation

static int acme_mmap
(struct file * file, struct vm_area_struct *vma)
{
size = vma->vm_end - vma->vm_start;
if (size > ACME_SIZE)
return -EINVAL;
if (remap_pfn_range(vma,
vma->vm_start,
ACME_PHYS >> PAGE_SHIFT,
size,
vma->vm_page_prot))
return -EAGAIN;
return 0;
}
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

457/477

devmem2

http://free-electrons.com/pub/mirror/devmem2.c, by
Jan-Derk Bakker
Very useful tool to directly peek (read) or poke (write) I/O
addresses mapped in physical address space from a shell
command line!
I

I
I

Very useful for early interaction experiments with a device,

without having to code and compile a driver.
Uses mmap to /dev/mem.
Examples (b: byte, h: half, w: word)
I
I

devmem2 0x000c0004 h (reading)

devmem2 0x000c0008 w 0xffffffff (writing)

devmem is now available in BusyBox, making it even easier to

use.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

458/477

mmap Summary

The device driver is loaded. It defines an mmap file operation.

A user space process calls the mmap system call.

The mmap file operation is called.

It initializes the mapping using the device physical address.

The process gets a starting address to read from and write to

(depending on permissions).

The MMU automatically takes care of converting the process

virtual addresses into physical ones.

Direct access to the hardware without any expensive read or

write system calls

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

459/477

Backup slides

Introduction to Git

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

460/477

What is Git?

A version control system, like CVS, SVN, Perforce or

ClearCase

Originally developed for the Linux kernel development, now

used by a large number of projects, including U-Boot,
GNOME, Buildroot, uClibc and many more
Contrary to CVS or SVN, Git is a distributed version control
system

I
I
I
I
I

No central repository
Everybody has a local repository
Local branches are possible, and very important
Easy exchange of code between developers
Well-suited to the collaborative development model used in
open-source projects

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

461/477

Install and Setup

Git is available as a package in your distribution

I

Everything is available through the git command

I

sudo apt-get install git

git has many commands, called using git <command>, where
<command> can be clone, checkout, branch, etc.
Help can be found for a given command using
git help <command>

Set up your name and e-mail address

I
I
I

They will be referenced in each of your commits

git config --global user.name My Name
git config --global user.email me@mydomain.net

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

462/477

Clone a Repository

To start working on a project, you use Gits clone operation.

With CVS or SVN, you would have used the checkout

operation, to get a working copy of the project (latest version)

With Git, you get a full copy of the repository, including the
history, which allows to perform most of the operations offline.

Cloning Linus Torvalds Linux kernel repository

git clone git://git.kernel.org/pub/scm/linux/
kernel/git/torvalds/linux.git

git:// is a special Git protocol. Most repositories can also

be accessed using http://, but this is slower.

After cloning, in linux/, you have the repository and a

working copy of the master branch.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

463/477

Explore the History

I

git log will list all the commits. The latest commit is the
first.
commit 4371ee353c3fc41aad9458b8e8e627eb508bc9a3
Author: Florian Fainelli <florian@openwrt.org>
Date: Mon Jun 1 02:43:17 2009 -0700
MAINTAINERS: take maintainership of the cpmac Ethernet driver
This patch adds me as the maintainer of the CPMAC (AR7)
Ethernet driver.
Signed-off-by: Florian Fainelli <florian@openwrt.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

I
I

git log -p will list the commits with the corresponding diff
The history in Git is not linear like in CVS or SVN, but it is a
graph of commits
I

Makes it a little bit more complicated to understand at the

beginning
But this is what allows the powerful features of Git
(distributed, branching, merging)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

464/477

Visualize the History: gitk

I
I

gitk is a graphical tool that represents the history of the

current Git repository
Can be installed from the gitk package

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

465/477

Visualize the History: cgit

I

Another great tool is cgit, a web interface to Git. For the

kernel, it is available at http://git.kernel.org/

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

466/477

Update your Repository

The repository that has been cloned at the beginning will

change over time

Updating your local repository to reflect the changes of the

remote repository will be necessary from time to time

git pull
Internally, does two things

Fetch the new changes from the remote repository

(git fetch)
Merge them in the current branch (git merge)

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

467/477

Tags

The list of existing tags can be found using

I

To check out a working copy of the repository at a given tag

I

git log v2.6.30..master

List of changes with diff on a given file between two tags

I

git checkout <tagname>

To get the list of changes between a given tag and the latest
available version
I

git tag -l

git log -p v2.6.29..v2.6.30 MAINTAINERS

With gitk
I

gitk v2.6.30..master

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

468/477

Branches

To start working on something, the best is to make a branch

I
I
I

It is local-only, nobody except you sees the branch

It is fast
It allows to split your work on different topics, try something
and throw it away
It is cheap, so even if you think youre doing something small
and quick, do a branch

Unlike other version control systems, Git encourages the use

of branches. Dont hesitate to use them.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

469/477

Branches

Create a branch
I

Move to this branch

I

git checkout -b <branchname>

List of local branches

I

git checkout <branchname>

Both at once (create and switch to branch)

I

git branch <branchname>

git branch

List of all branches, including remote branches

I

git branch -a

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

470/477

Making Changes
I
I

Edit a file with your favorite text editor

Get the status of your working copy
I

Git has a feature called the index, which allows you to stage
your commits before committing them. It allows to commit
only part of your modifications, by file or even by chunk.
On each modified file
I

git add <filename>

Then commit. No need to be on-line or connected to commit

I
I

git status

Linux requires the -s option to sign your changes

git commit -s

If all modified files should be part of the commit

I

git commit -as

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

471/477

Sharing Changes: E-mail

The simplest way of sharing a few changes is to send patches

by e-mail
The first step is to generate the patches
I
I

git format-patch -n master..<yourbranch>

Will generate one patch for each of the commits done on
<yourbranch>
The patch files will be 0001-...., 0002-...., etc.

The second step is to send these patches by e-mail

I

git send-email --compose -to email@domain.com 00*.patch

Required Ubuntu package: git-email

In a later slide, we will see how to use git config to set the
SMTP server, port, user and password.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

472/477

Sharing Changes: Your Own Repository

I
I

If you do a lot of changes and want to ease collaboration with

others, the best is to have your own public repository
Use a git hosting service on the Internet:
I

Gitorious (https://gitorious.org/)
I

For public repositories. Have to pay for private repositories.

Publish on your own web server

I
I
I

Open Source server. Easiest. For public repositories.

GitHub (https://github.com/)

Easy to implement.
Just needs git software on the server and ssh access.
Drawback: only supports http cloning (less efficient)

Set up your own git server

I
I

Most flexible solution.

Todays best solutions are gitolite
(https://github.com/sitaramc/gitolite) for the server
and cgit for the web interface
(http://git.zx2c4.com/cgit/about/).

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

473/477

Sharing changes: HTTP Hosting

Create a bare version of your repository

I
I
I

cd /tmp
git clone --bare ~/project project.git
touch project.git/git-daemon-export-ok

Transfer the contents of project.git to a publicly-visible

place (reachable read-only by HTTP for everybody, and
read-write by you through SSH)

Tell people to clone

http://yourhost.com/path/to/project.git
Push your changes using

git push ssh://yourhost.com/path/toproject.git

srcbranch:destbranch

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

474/477

Tracking Remote Trees

In addition to the official Linus Torvalds tree, you might want

to use other development or experimental trees
I

The git remote command allows to manage remote trees

I

git remote add rt git://git.kernel.org/pub/scm/

linux/kernel/git/rt/linux-stable-rt.git

Get the contents of the tree

I

The OMAP tree at git://git.kernel.org/pub/scm/

linux/kernel/git/tmlind/linux-omap.git
The stable realtime tree at git://git.kernel.org/pub/
scm/linux/kernel/git/rt/linux-stable-rt.git

git fetch rt

Switch to one of the branches

I

git checkout rt/master

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

475/477

About Git

We have just seen the very basic features of Git.

A lot more interesting features are available (rebasing,

bisection, merging and more)
References

Git Manual

Git Book

I
I

http://git-scm.com/book

Git official website

I

http://schacon.github.com/git/user-manual.html

http://git-scm.com/

Video: James Bottomleys tutorial on using Git

I

http://free-electrons.com/pub/video/2008/ols/
ols2008-james-bottomley-git.ogg

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

476/477

Practical lab - Going further: git

Get familiar with git by

contributing to a real project: the
Linux kernel

Send your patches to the

maintainers and mailing lists.

Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com

477/477