Programming Embedded Linux

Programming Embedded Linux ®
Unable to handle kernel paging request at virtual address 4d1b65e8
Covers Linux version 2.6.24
Unable to handle kernel paging request at virtual address 4d1b65e8
pgd = c0280000
pgd = c0280000
Version 1.4
<1>[4d1b65e8] *pgd=00000000[4d1b65e8] *pgd=00000000
Internal error: Oops: f5 [#1]
Internal error: Oops: f5 [#1]
Modules linked in:Modules linked in: hx4700_udc hx4700_udc asic3_base asic3_base
CPU: 0
CPU: 0
PC is at set_pxa_fb_info+0x2c/0x44
PC is at set_pxa_fb_info+0x2c/0x44
LR is at hx4700_udc_init+0x1c/0x38 [hx4700_udc]
LR is at hx4700_udc_init+0x1c/0x38 [hx4700_udc]
pc : [<c00116c8>] lr : [<bf00901c>] Not tainted
sp : c076df78 ip : 60000093 fp : c076df84
pc : [<c00116c8>] lr : [<bf00901c>] Not tainted
© Copyright 20062004, Michael Opdenacker For full copyright information see last page.
© Copyright 20032006, Oron Peled Creative Commons AttributionShareAlike 2.0 license
© Copyright 20042006 Codefidence Ltd. 1
Rights to Copy
This kit contains work by the
following authors:
Attribution – ShareAlike 2.0
You are free © Copyright 20042006
to copy, distribute, display, and perform the work Michael Opdenacker
to make derivative works
michael@freeelectrons.com
to make commercial use of the work
Under the following conditions
http://www.freeelectrons.com
Attribution. You must give the original author credit.
© Copyright 20032006
Share Alike. If you alter, transform, or build upon this work,
Oron Peled
you may distribute the resulting work only under a license
identical to this one. oron@actcom.co.il
For any reuse or distribution, you must make clear to others the http://www.actcom.co.il/~oron
license terms of this work.
Any of these conditions can be waived if you get permission from © Copyright 2004–2008
the copyright holder.
Codefidence ltd.
Your fair use and other rights are in no way affected by the above.
info@codefidence.com
License text: http://creativecommons.org/licenses/bysa/2.0/legalcode
http://www.codefidence.com
What is Linux?
Linux is a kernel that implements the POSIX and Single Unix Specification
standards which is developed as an opensource project.
Usually when one talks of “installing Linux”, one is referring to a Linux
distribution.
A distribution is a combination of Linux and other programs and library that
form an operating system.
There exists many such distribution for various purposes, from highend
servers to embedded systems.
They all share the same interface, thanks to the LSB standard.
Linux runs on 24 main platforms and supports applications ranging from
ccNUMA super clusters to cellular phones and micro controllers.
Linux is 15 years old, but is based on the 40 years old Unix design philosophy.
Layers in a Linux System
●
Kernel
User programs
●
Kernel Modules
●
C library
●
System libraries Kernel
●
Application libraries
C library
●
User programs
Linux vs. Legacy RTOS
Process
Library
Process
Kernel Shell RTOS

Process
Process
Process
RTOS are like the Linux kernel:

Single program with single memory space that
manages memory, scheduling and interrupts.
Linux also has user tasks that run in their own

memory space. One of them is the shell.
This Course
Process
In this course we will go
Library from the shell, through
Process the system libraries and
applications and unto the
Shell
kernel.
Kernel
Process
Process
Process
Table of Content
Basic Interaction Kernel
Files and file system Source layout
The shell Coding convention
Processes Versions
Setting up networking Kernel modules
Application Programming Memory Management
Makefile Character device
Processes Scheduling
Threads Soft and hard interrupts
IPC Networking
The Unix and GNU/Linux CommandLine
The Unix Filesystem
Everything is a File
Almost everything in Unix is a file!
Regular files Devices and peripherals
Read and write from devices as
Directories
with regular files
Directories are just files
listing a set of files Pipes
Used to cascade programs
Symbolic links
cat *.log | grep error
Files referring to the name of
another file Sockets
Inter process communication
Filenames
File name features since the beginning of Unix:
Case sensitive.
No obvious length limit.
Can contain any character (including whitespace, except /).
File types stored in the file (“magic numbers”).
File name extensions not needed and not interpreted. Just used for
user convenience.
File name examples:
README .bashrc Windows Buglist
index.htm index.html index.html.old
File Paths
A path is a sequence of nested directories with a file or directory at
the end, separated by the / character.
Relative path: documents/fun/microsoft_jokes.html
Relative to the current directory
Absolute path: /home/bill/bugs/crash9402031614568
/ : root directory.
Start of absolute paths for all files on the system (even for files on
removable devices or network shared).
GNU/Linux Filesystem Structure (1)
Not imposed by the system. Can vary from one system to the other,
even between two GNU/Linux installations!
/ Root directory
/bin/ Basic, essential system commands
/boot/ Kernel images, initrd and configuration files
/dev/ Files representing devices
/dev/hda: first IDE hard disk
/etc/ System configuration files
/home/ User directories
/lib/ Basic system shared libraries
/lost+found Corrupt files the system tried to recover
/mnt/ Mounted filesystems
/mnt/usbdisk/, /mnt/windows/ ...
/opt/ Specific tools installed by the sysadmin
/usr/local/ often used instead
/proc/ Access to system information
/proc/cpuinfo, /proc/version ...
/root/ root user home directory
/sbin/ Administratoronly commands
/sys/ System and device controls
(cpu frequency, device power, etc.)
/tmp/ Temporary files
/usr/ Regular user tools (not essential to the system)
/usr/bin/, /usr/lib/, /usr/sbin...
/usr/local/ Specific software installed by the sysadmin
(often preferred to /opt/)
/var/ Data used by the system or system servers
/var/log/, /var/spool/mail (incoming
mail), /var/spool/lpd (print jobs)...
Shells and File Handling
CommandLine Interpreters
Shells: tools to execute user commands.
Called “shells” because they hide the details on the
underlying operating system under the shell's surface.
Commands are entered using a text terminal: either a window
in a graphical environment, or a textonly console.
Results are also displayed on the terminal. No graphics are
needed at all.
Shells can be scripted: provide all the resources to write
complex programs (variable, conditionals, iterations...)
ls Command
Lists the files in the current directory, in alphanumeric order,
except files starting with the “.” character.
ls a (all) ls S (size)
Lists all the files (including .* Lists the biggest files first
files)
ls r (reverse)
ls l (long) Reverses the sort order
Long listing (type, date, size,
owner, permissions) ls ltr (options can be
combined)
ls t (time) Long listing, most recent files at
Lists the most recent files first the end
Filename Pattern Substitutions
Better introduced by examples:
ls *txt
The shell first replaces *txt by all the file and directory names
ending by txt (including .txt), except those starting with .,
and then executes the ls command line.
ls d .*
Lists all the files and directories starting with .
d tells ls not to display the contents of directories.
ls ?.log
Lists all the files which names start by 1 character and end by
.log
Special Directories (1)
./
The current directory. Useful for commands taking a directory
argument. Also sometimes useful to run commands in the current
directory (see later).
So ./readme.txt and readme.txt are equivalent.
../
The parent (enclosing) directory. Always belongs to the . directory
(see ls a). Only reference to the parent directory.
Typical usage:
cd ..
The cd and pwd Commands
cd <dir>
Change the current directory to <dir>.
pwd
Displays the current directory ("working directory").
The cp Command
cp <source_file> <target_file>
Copies the source file to the target.
cp file1 file2 file3 ... dir
Copies the files to the target directory (last argument).
cp i (interactive)
Asks for user confirmation if the target file already exists
cp r <source_dir> <target_dir> (recursive)
Copies the whole directory.
The mv and rm Commands
mv <old_name> <new_name> (move)
Renames the given file or directory.
mv i (interactive)
If the new file already exits, asks for user confirm
rm file1 file2 file3 ... (remove)
Removes the given files.
rm i (interactive)
Always ask for user confirm.
rm r dir1 dir2 dir3 (recursive)
Removes the given directories with all their contents.
Creating and Removing Directories
mkdir dir1 dir2 dir3 ... (make dir)
Creates directories with the given names.
rmdir dir1 dir2 dir3 ... (remove dir)
Removes the given directories
Safe: only works when directories and empty.
Alternative: rm r (doesn't need empty directories).
Displaying File Contents
Several ways of displaying the contents of files.
cat file1 file2 file3 ... (concatenate)
Concatenates and outputs the contents of the given files.
more file1 file2 file3 ...
After each page, asks the user to hit a key to continue.
Can also jump to the first occurrence of a keyword
(/ command).
less file1 file2 file3 ...
Does more than more with less.
Doesn't read the whole file before starting.
Supports backward movement in the file (? command).
Task Control
Full Control Over Tasks
Since the beginning, Unix supports true preemptive
multitasking.
Ability to run many tasks in parallel, and abort them even if
they corrupt their own state and data.
Ability to choose which programs you run.
Ability to choose which input your programs takes, and
where their output goes.
Processes
“Everything in Unix is a file
Everything in Unix that is not a file is a process”
Processes
Instances of a running programs
Several instances of the same program can run at the same time
Data associated to processes:
Open files, allocated memory, process id, parent, priority, state...
Running Jobs in Background
Same usage throughout all the shells.
Useful:
For command line jobs which output can be examined later,
especially for time consuming ones.
To start graphical applications from the command line and
then continue with the mouse.
Starting a task: add & at the end of your line:
find_prince_charming cute clever rich &
Background Job Control
jobs
Returns the list of background jobs from the same shell
[1] Running ~/bin/find_meaning_of_life withoutgod &
[2]+ Running make mistakes &
fg
fg %<n>
Puts the last / nth background job in foreground mode
Moving the current task in background mode:
[Ctrl] Z
bg
kill %<n>
Aborts the nth job.
Job Control Example
> jobs
[1] Running ~/bin/find_meaning_of_life withoutgod &
[2]+ Running make mistakes &
> fg
make mistakes
> [Ctrl] Z
[2]+ Stopped make mistakes
> bg
[2]+ make mistakes &
> kill %1
[1]+ Terminated ~/bin/find_meaning_of_life withoutgod
Listing All Processes
... whatever shell, script or process they are started from
ps ux
Lists all the processes belonging to the current user
ps aux (Note: ps edf on System V systems)
Lists all the processes running on the system
ps aux
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
bart 3039 0.0 0.2 5916 1380 pts/2 S 14:35 0:00 /bin/bash
bart 3416 0.0 0.0 0 0 pts/2 R 15:07 0:00 [bash] ...
PID: Process id
VSZ: Virtual process size (code + data + stack)
RSS: Process resident size: number of KB currently in RAM
TTY: Terminal
STAT: Status: R (Runnable), S (Sleep), D (Uninterrupted sleep), Z (Zombie), T(Traced)
Live Process Activity
top Displays most important processes, sorted by cpu percentage
top 15:44:33 up 1:11, 5 users, load average: 0.98, 0.61, 0.59
Tasks: 81 total, 5 running, 76 sleeping, 0 stopped, 0 zombie
Cpu(s): 92.7% us, 5.3% sy, 0.0% ni, 0.0% id, 1.7% wa, 0.3% hi, 0.0% si
Mem: 515344k total, 512384k used, 2960k free, 20464k buffers
Swap: 1044184k total, 0k used, 1044184k free, 277660k cached
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
3809 jdoe 25 0 6256 3932 1312 R 93.8 0.8 0:21.49 bunzip2
2769 root 16 0 157m 80m 90m R 2.7 16.0 5:21.01 X
3006 jdoe 15 0 30928 15m 27m S 0.3 3.0 0:22.40 kdeinit
3008 jdoe 16 0 5624 892 4468 S 0.3 0.2 0:06.59 autorun
3034 jdoe 15 0 26764 12m 24m S 0.3 2.5 0:12.68 kscd
3810 jdoe 16 0 2892 916 1620 R 0.3 0.2 0:00.06 top
You can change the sorting order by typing
M: Memory usage, P: %CPU, T: Time.
You can kill a task by typing k and the process id.
Killing Processes (1)
kill <pids>
Sends a termination signal (SIGTERM) to the given processes. Lets
processes save data and exit by themselves. Should be used first.
Example:
kill 3039 3134 3190 3416
kill 9 <pids>
Sends an immediate termination signal (SIGKILL). The system
itself terminates the processes. Useful when a process is really stuck
(doesn't answer to kill 1).
kill 9 1
Kills all the processes of the current user. 1: means all processes.
Killing Processes (2)
killall [<signal>] <command>
Kills all the jobs running <command>. Example:
killall bash
Sequential Commands
Can type the next command in your terminal even when the
current one is not over.
Can separate commands with the ; symbol:
echo n “I love thee”; sleep 5; echo “ not”
Conditionals: use || (or) or && (and):
more God || echo “Sorry, God doesn't exist”
Runs echo only if the first command fails
ls mydir && cat mydir/*
Only cats the directory contents if the ls command succeeds
(means read access).
System Administration Basics
Shutting Down
shutdown h +5 (h: halt)
Shuts the system down in 5 minutes.
Users get a warning in their consoles.
shutdown r now (r: reboot)
init 0
Another way to shutdown (used by shutdown).
init 6
Another way to reboot (used by shutdown).
[Ctrl][Alt][Del]
Also works on GNU/Linux (at least on PCs!).
Network Setup (1)
ifconfig a
Prints details about all the network interfaces available
on your system.
ifconfig eth0
Lists details about the eth0 interface
ifconfig eth0 192.168.0.100
Assigns the 192.168.0.100 IP address
to eth0 (1 IP address per interface).
ifconfig eth0 down
Shuts down the eth0 interface
(frees its IP address).
Network Setup (2)
route add default gw 192.168.0.1
Sets the default route for packets outside the local network.
The gateway (here 192.168.0.1) is responsible for
sending them to the next gateway, etc., until the final
destination.
route
Lists the existing routes
route del default
route del <IP>
Deletes the given route
Useful to redefine a new route.
Network Testing
ping freshmeat.net
ping 192.168.1.1
Tries to send packets to the given machine and get acknowledgment
packets in return.
PING 192.168.1.1 (192.168.1.1) 56(84) bytes of data.
64 bytes from 192.168.1.1: icmp_seq=0 ttl=150 time=2.51 ms
When you can ping your gateway, your network interface works fine.
When you can ping an external IP address, your network settings are
correct!
Network Setup Summary
Only for simple cases with one interface, no DHCP server...
Connect to the network (cable, wireless card or device...)
Identify your network interface:
ifconfig a
Assign an IP address to your interface (assuming eth0)
ifconfig eth0 192.168.0.100 (example)
Add a route to your gateway (assuming 192.168.0.1) for
packets outside the network:
route add default gw 192.168.0.1
Name Resolution
Your programs need to know what IP address corresponds to a
given host name (such as kernel.org)
Domain Name Servers (DNS) take care of this.
You just have to specify the IP address of 1 or more DNS
servers in your /etc/resolv.conf file:
nameserver 217.19.192.132
nameserver 212.27.32.177
The changes takes effect immediately!
Mounting Devices (1)
To make filesystems on any device (internal or external storage)
visible on your system, you have to mount them.
The first time, create a mount point in your system:
mkdir /mnt/usbdisk (example)
Now, mount it:
mount t vfat /dev/sda1 /mnt/usbdisk
/dev/sda1: physical device
t: specifies the filesystem (format) type
(ext2, ext3, vfat, reiserfs, iso9660...)
Mounting Devices (2)
Lots of mount options are available, in particular to choose permissions or the
file owner and group... See the mount manual page for details.
Mount options for each device can be stored in the /etc/fstab file.
You can also mount a filesystem image stored in a regular file (loopback
devices)
Useful to access the contents of an ISO cdrom image without having to
burn it.
Useful to create a Linux partition on a hard disk with only Windows
partitions
cp /dev/sda1 usbkey.img
mount o loop t vfat usbkey.img /mnt/usbdisk
Listing Mounted Filesystems
Just use the mount command with no argument:
/dev/hda6 on / type ext3 (rw,noatime)
none on /proc type proc (rw,noatime)
none on /sys type sysfs (rw)
none on /dev/pts type devpts (rw,gid=5,mode=620)
usbfs on /proc/bus/usb type usbfs (rw)
/dev/hda4 on /data type ext3 (rw,noatime)
none on /dev/shm type tmpfs (rw)
/dev/hda1 on /win type vfat (rw,uid=501,gid=501)
none on /proc/sys/fs/binfmt_misc type binfmt_misc (rw)
Or display the /etc/mtab file
(same result, updated by mount and umount each time they are run)
Unmounting Devices
umount /mnt/usbdisk
Commits all pending writes and unmounts the given device,
which can then be removed in a safe way.
To be able to unmount a device, you have to close all the
open files in it:
Close applications opening data in the mounted partition
Make sure that none of your shells have a working directory in
this mount point.
You can run the lsof command (list open files) to view
which processes still have open files in the mounted partition.
Creating Filesystems
Examples
mkfs.ext2 /dev/sda1
Formats your USB key (/dev/sda1: 1st partition raw data) in ext2 format.
mkfs.ext2 F disk.img
Formats a disk image file in ext2 format
mkfs.vfat v F 32 /dev/sda1 (v: verbose)
Formats your USB key back to FAT32 format.
mkfs.vfat v F 32 disk.img
Formats a disk image file in FAT32 format.
Blank disk images can be created as in the below example:
dd if=/dev/zero of=disk.img bs=1024 count=65536
Help
Command Help
Some Unix commands and most GNU/Linux commands offer
at least one help argument:
h
( is mostly used to introduce 1character options)
help
( is always used to introduce the corresponding “long”
option name, which makes scripts easier to understand)
You also often get a short summary of options when you input
an invalid argument.
Manual Pages
man [section] <keyword>
Displays one or several manual pages for <keyword> from
optional [section].
man fork
Man page of the fork() system call
man fstab
Man page of the fstab configuration file
man printf
Man of printf() shell command
man 3 printf
Man of printf() library function
Coding Embedded Linux Applications
Writing Applications
A Simple Makefile
CC=/usr/local/arm/2.95.3/bin/armlinuxgcc
Makefile
variables CFLAGS=g
LDFLAGS=lpthreads
To build target
Always build the .PHONY: clean all all build target
clean and all test
targets, if asked all: test
To compile test.o
test.o: test.c test.h from test.c and
These must be a test.h, run the
TAB, not a $(CC) $(CFLAGS) C test.c following command
SPACE!
test: test.o
To link test from
test.o, run the $(CC) $(LDFLAGS) test.o o test
following To clean, run the
command
clean: following
command
@rm f *.o *~ test
A NotSoSimple Makefile
.PHONY: clean all
SRC := $(shell ls *.c) This Makefile automatically
EXE := test compiles all files ending in .c in the
OBJ := $(SRC:.c=.o)
current directory, computes
LDFLAGS := lsqlite3
CFLAGS := g dependency information for them
all: $(EXE) and links them into an executable
%.o : %.c called test.
$(COMPILE.c) MD o $@ $<
@cp $*.d $*.P; \
sed e 's/#.*//' e 's/^[^:]*: *//' e 's/ *\\$$//' \
e '/^$$/ d' e 's/$$/ :/' < $*.d >> $*.P; \
rm f $*.d
$(EXE): $(OBJ)
$(CC) $(LDFLAGS) o $@ $(OBJ)
clean:
rm f $(EXE) $(OBJ) $(SRC:.c=.P) *~
include $(SRC:.c=.P)
# Source: http://make.paulandlesley.org/autodep.html
Hello World!
Linux Application API is ISO C and POSIX!
#include <stdio.h>

int main(int argc, char **argv)
{
int i=1;
printf("Hello World! %d\n", i);
return 0;
}

Creating Processes Using fork()
pid_t pid;
if (pid = fork()) {
int status;
printf("I'm the parent!\n");
wait(&status);
if (WIFEXITED(status))
printf("Child exist with status of %d\n", WEXITSTATUS(status));
} else {
printf("I'm the child!\n");
exit(0);
}
How fork() Seems to Work
A complete copy
Parent Process is created of the
parent process.
Both parent and

child continue
execution from
the same spot.
Copy
Parent Process Child Process
How fork() Really Works
Child process gets

RW a copy of stack and
Parent file descriptors.
Memory
Process Copy On Write
reference is used
for the memory.
Parent RO RO Child
Memory
Process Process
What Happens During Write?
When write is
attempted,
a single memory
Original Memory
page is copied RO
and references
updated.
This is called:
“breaking the Child
CoW”. RW Process
RO
Parent
Process
Creating Processes (2)
This call will pid_t pid;

replace the
program memory if (pid = fork()) {
(code and data)
with the image int status;
from storage at the
specified path.
printf("I'm the parent!\n");
wait(&status);
Open file
descriptors, } else {
priority and other
properties remains printf("I'm the child!\n");
the same.
execve(“/bin/ls”, argv, envp);
}
Exiting Processes
A process exists when either of the following occurs:
Return from the main function.
Calling the exit() function.
Exception (more on those later).
A process should return an exit status to it's parent process
Upon exit:
All exit handlers are called (use atexit() to register them)
All memory, file descriptors and other resources are released
by the system.
Linux Process Stack
Linux process stack is auto expanding.
A default stack (8Mb) is allocated at process creation.
By default, use of additional stack will automatically trigger
allocation of more stack space.
This behavior can be limited by setting a resource limit on
the stack size.
See setrlimit()
Linux Priorities
Nice Real Time priority

level 99
-20 98
-19 97 Real time processes
-18 ... SCHED_FIFO
SCHED_RR
... 4
16 3
17 2
18 1
19 Non real-time processes
0 SCHED_OTHER
POSIX Priorities (2)
SCHED_OTHER: Default Linux timesharing scheduling
Priority 0 is reserved for it.
Fair, no starvation.
Use this for any non time critical tasks.
SCHED_FIFO: First InFirst Out scheduling
Priorities 1 – 99.
Preemptive.
SCHED_RR: Round Robin scheduling
Like SCHED_FIFO + time slice
Changing Real Time Priorities
int sched_setscheduler(pid_t pid, int policy, const struct sched_param *p);
struct sched_param {
int sched_priority
};
sched_setscheduler() sets both the scheduling policy and the associated
parameters for the process identified by pid. If pid equals zero, the
scheduler of the calling process will be set. The interpretation of the
parameter p depends on the selected policy. Currently, the following three
scheduling policies are supported under Linux: SCHED_FIFO,
SCHED_RR, and SCHED_OTHER;
There is also a sched_getscheduler().
Locking Memory
int mlock(const void *addr, size_t len);
mlock disables paging for the memory in the range starting at addr with length len
bytes.
int mlockall(int flags);
mlockall() disables paging for all pages mapped into the address space of the calling
process.
MCL_CURRENT locks all pages which are currently mapped into the address
space of the process.
MCL_FUTURE locks all pages which will become mapped into the address space
of the process in the future. These could be for instance new pages required by a
growing heap and stack as well as new memory mapped files or shared memory
regions.
Must lock memory to guarantee real time responses!
Real Time Responses
To get deterministic real time response from a Linux process,
make sure to:
Put the process in a real time scheduling domain.
mlockall() process memory.
Prefault stack pages
To do this call a dummy function that allocates on stack an
automatic variable big enough for your entire future stack
usage and writes to it.
Threads and Processes
Stack Stack Stack Stack Stack
State State State State State

Signal Signal Signal Signal Signal
Mask Mask Mask Mask Mask
Priority Priority Priority Priority Priority
Thread Thread Thread Thread

Thread 1
1 2 3 4
Process 123 Process 124
File Signal File Signal

Memory Memory
Descriptors Handlers Descriptors Handlers
POSIX Threads
Linux uses the POSIX Threads threading mechanism.
Linux threads are Light Weight Processes – each thread is a task
scheduled by the kernel's scheduler.
Process creation time is roughly double than that of a thread's.
But Linux process creation time is relatively low...
To use threads in your code:
#include <pthread.h>
In your Makefile:
Add -lpthread to CFLAGS.
Creating Threads
Function: pthread_create()
int pthread_create(pthread_t *thread, const pthread_attr_t *attr,
void *(*start_routine)(void *), void * arg);
The pthread_create() routine creates a new thread within a
process. The new thread starts in the start routine start_routine
which has a start argument arg. The new thread has attributes
specified with attr, or default attributes if attr is NULL.
If the pthread_create() routine succeeds it will return 0 and put the
new thread ID into thread, otherwise an error number shall be
returned indicating the error.
Creating Thread Attributes
Function: pthread_attr_init()
int pthread_attr_init(pthread_attr_t *attr);
Setting attributes for threads is achieved by filling a thread attribute
object attr of type pthread_attr_t, then passing it as a second argument
to pthread_create(3). Passing NULL is equivalent to passing a thread
attribute object with all attributes set to their default values.
pthread_attr_init() initializes the thread attribute object attr and fills it
with default values for the attributes.
Each attribute attrname can be individually set using the function
pthread_attr_setattrname() and retrieved using the function
pthread_attr_getattrname().
Destroying Thread Attributes
Function: pthread_attr_destroy()
int pthread_attr_destroy(pthread_attr_t *attr);
pthread_attr_destroy() destroys a thread attribute object, which
must not be reused until it is reinitialized. pthread_attr_destroy()
does nothing in the LinuxThreads implementation.
Attribute objects are consulted only when creating a new thread.
The same attribute object can be used for creating several threads.
Modifying an attribute object after a call to pthread_create() does
not change the attributes of the thread previously created.
Detach State
Thread Attribute: detachstate
Control whether the thread is created in the joinable state or in the detached
state. The default is joinable state.
In the joinable state, another thread can synchronize on the thread termination
and recover its termination code using pthread_join(3), but some of the thread
resources are kept allocated after the thread terminates, and reclaimed only
when another thread performs pthread_join(3) on that thread.
In the detached state, the thread resources are immediately freed when it
terminates, but pthread_join(3) cannot be used to synchronize on the thread
termination.
A thread created in the joinable state can later be put in the detached thread
using pthread_detach(3).
Sched Policy
Thread Attribute: schedpolicy
Select the scheduling policy for the thread: one of
SCHED_OTHER (regular, nonrealtime scheduling),
SCHED_RR (realtime, roundrobin) or SCHED_FIFO (realtime,
firstin firstout).
Default value: SCHED_OTHER.
The realtime scheduling policies SCHED_RR and SCHED_FIFO
are available only to processes with superuser privileges.
The scheduling policy of a thread can be changed after creation
with pthread_setschedparam(3).
Sched Param
Thread Attribute: schedparam
Contain the scheduling parameters (essentially, the scheduling
priority) for the thread.
Default value: priority is 0.
This attribute is not significant if the scheduling policy is
SCHED_OTHER; it only matters for the realtime policies
SCHED_RR and SCHED_FIFO.
The scheduling priority of a thread can be changed after creation
with pthread_setschedparam(3).
Inherit Sched
Thread Attribute: inheritsched
Indicate whether the scheduling policy and scheduling parameters
for the newly created thread are determined by the values of the
schedpolicy and schedparam attributes
(PTHREAD_EXPLICIT_SCHED) or are inherited from the parent
thread (value PTHREAD_INHERIT_SCHED).
Default value: PTHREAD_EXPLICIT_SCHED.
Destroying Threads
Function: pthread_exit()
void pthread_exit(void *status);
The pthread_exit() routine terminates the currently running thread
and makes status available to the thread that successfully joins,
pthread_join(), with the terminating thread. In addition,
pthread_exit() executes any remaining cleanup handlers in the
reverse order they were pushed, pthread_cleanup_push(), after
which all appropriate thread specific destructors are called.
An implicit call to pthread_exit() is made if any thread, other than
the thread in which main() was first called, returns from the start
routine specified in pthread_create().
Thread and Process Termination
Action Main Thread Other Thread
Thread Thread
pthread_exit()
terminates terminates
All threads All threads

exit()
terminate terminate
Thread
All threads Thread
function
terminate terminates
returns
Waiting For a Thread to Finish
Function: pthread_join()
int pthread_join(pthread_t thread, void **status);
If the target thread thread is not detached and there are no other threads
joined with the specified thread then the pthread_join() function
suspends execution of the current thread and waits for the target thread
thread to terminate. Otherwise the results are undefined.
On a successful call pthread_join() will return 0, and if status is non
NULL then status will point to the status argument of pthread_exit().
On failure pthread_join() will return an error number indicating the
error.
Also exists pthread_tryjoin_np() and pthread_timedjoin_np().
Canceling Threads
Function: pthread_cancel()
int pthread_cancel(pthread_t thread);
Cancellation is the mechanism by which a thread can terminate the
execution of another thread. More precisely, a thread can send a
cancellation request to another thread. Depending on its settings, the
target thread can then either ignore the request, honor it immediately, or
defer it till it reaches a cancellation point.
When a thread eventually honors a cancellation request, it performs as
if pthread_exit(PTHREAD_CANCELED) has been called at that point.
pthread_cancel() sends a cancellation request to the thread denoted by
the thread argument.
Cancel State
Function: pthread_setcancelstate()
int pthread_setcancelstate(int state, int *oldstate);
pthread_setcancelstate() changes the cancellation state for the calling
thread that is, whether cancellation requests are ignored or not.
The state argument is the new cancellation state: either
PTHREAD_CANCEL_ENABLE to enable cancellation, or
PTHREAD_CANCEL_DISABLE to disable cancellation
(cancellation requests are ignored).
If oldstate is not NULL, the previous cancellation state is stored in the
location pointed to by oldstate, and can thus be restored later by
another call to pthread_setcancelstate().
Cancel Type
Function: pthread_setcanceltype()
int pthread_setcanceltype(int type, int *oldtype);
pthread_setcanceltype() changes the type of responses to cancellation
requests for the calling thread: asynchronous (immediate) or deferred.
The type argument is the new cancellation type: either
PTHREAD_CANCEL_ASYNCHRONOUS to cancel the calling thread
as soon as the cancellation request is received, or
PTHREAD_CANCEL_DEFERRED to keep the cancellation request
pending until the next cancellation point. If old type is not NULL, the
previous cancellation state is stored in the location pointed to by oldtype,
and can thus be restored later by another call to
pthread_setcanceltype().
Cancellation Point Defaults
Threads are always created by pthread_create(3) with cancellation
enabled and deferred. That is, the initial cancellation state is
PTHREAD_CANCEL_ENABLE and the initial type is
PTHREAD_CANCEL_DEFERRED.
Cancellation Points
Cancellation points are those points in the program execution where a
test for pending cancellation requests is performed and cancellation is
executed if positive.
The following POSIX threads functions are cancellation points:
pthread_join(3), pthread_cond_wait(3), pthread_cond_timedwait(3),
pthread_testcancel(3), sem_wait(3), sigwait(3).
All other POSIX threads functions are guaranteed not to be cancellation
points. That is, they never perform cancellation in deferred cancellation
mode.
A number of system calls (basically, all system calls that may block,
such as read(2), write(2), wait(2), etc.) and library functions that may
call these system calls (e.g. fprintf(3)) are cancellation points.
Test Cancel
Function: pthread_testcancel()
void pthread_testcancel(void);
pthread_testcancel() does nothing except testing for pending
cancellation and executing it. Its purpose is to introduce explicit
checks for cancellation in long sequences of code that do not call
cancellation point functions otherwise.
Linux IPC
POSIX IPC
SysV IPC
Pipes
Unix Domain Sockets
Signals
TCP/IP Sockets
Creating a Mutex
Function: pthread_mutex_init()
int pthread_mutex_init(pthread_mutex_t *mutex, const
pthread_mutex_attr *attr);
The pthread_mutex_init() routine creates a new mutex, with
attributes specified with attr, or default attributes if attr is NULL.
If the pthread_mutex_init() routine succeeds it will return 0 and
put the new mutex ID into mutex, otherwise an error number shall
be returned indicating the error.
Destroying a Mutex
Function: pthread_mutex_destroy()
int pthread_mutex_destroy(pthread_mutex_t *mutex);
The pthread_mutex_destroy() routine destroys the mutex specified
by mutex.
If the pthread_mutex_destroy() routine succeeds it will return 0,
otherwise an error number shall be returned indicating the error.
Locking Mutexes
Function: pthread_mutex_lock()
int pthread_mutex_lock(pthread_mutex_t *mutex);
The pthread_mutex_lock() routine shall lock the mutex specified by mutex. If the
mutex is already locked, the calling thread blocks until the mutex becomes available.
If the pthread_mutex_lock() routine succeeds it will return 0, otherwise an error
number shall be returned indicating the error.
Function: pthread_mutex_trylock()
int pthread_mutex_trylock(pthread_mutex_t *mutex);
The pthread_mutex_trylock() routine shall lock the mutex specified by mutex and
return 0, otherwise an error number shall be returned indicating the error. In all cases
the pthread_mutex_trylock() routine will not block the current running thread.
Locking Mutexes
Function: pthread_mutex_timedlock()
int pthread_mutex_timedlock(pthread_mutex_t *mutex, struct timespec *restrict
abs_timeout);
The pthread_mutex_timedlock() function shall lock the mutex object referenced
by mutex. If the mutex is already locked, the calling thread shall block until the
mutex becomes available as in the pthread_mutex_lock() function. If the mutex
cannot be locked without waiting for another thread to unlock the mutex, this
wait shall be terminated when the specified timeout expires.
The timeout shall expire when the absolute time specified by abs_timeout
passes, as measured by the clock on which timeouts are based (that is, when the
value of that clock equals or exceeds abs_timeout), or if the absolute time
specified by abs_timeout has already been passed at the time of the call.
Unlocking Mutexes
Function: pthread_mutex_unlock()
int pthread_mutex_unlock(pthread_mutex_t *mutex);
If the current thread is the owner of the mutex specifed by mutex,
then the pthread_mutex_unlock() routine shall unlock the mutex.
If there are any threads blocked waiting for the mutex, the
scheduler will determine which thread obtains the lock on the
mutex, otherwise the mutex is available to the next thread that
calls the routine pthread_mutex_lock(), or
pthread_mutex_trylock().
If the pthread_mutex_unlock() routine succeeds it will return 0,
otherwise an error number shall be returned indicating the error.
Priority Inversion
2. High Priority task preempts low

priority task
99
3. Hi Priority task block on mutex
Task Priority
50
4. Medium Priority task preempts
low priority task and high
priority task
3
1. Low Priority task takes mutex
Time
Priority Inheritance and Ceilings
Priority inheritance and ceilings are methods to protect
against priority inversions.
Linux only got support for them in 2.6.18.
Patches to add support for older version exists.
Embedded Linux vendors usually provide patched kernels.
If the kernel version you're using is not patched, make sure to
protect against this scenario in design
One possible way: raise the priority of each tasks trying to
grab a mutex to the maximum priority of all possible
contenders.
POSIX Condition Variables
int pthread_cond_wait(pthread_cond_t *cond,
pthread_mutex_t *mutex);
int pthread_cond_timedwait(pthread_cond_t *cond,
pthread_mutex_t *mutex, const struct timespec
*abstime);
int pthread_cond_signal(pthread_cond_t *cond);
int pthread_cond_broadcast(pthread_cond_t *cond);
The first two function calls are used for waiting on the condition var.
The latter two function calls are used to wake up waiting tasks:
pthread_cond_signal() wakes up one single task that is waiting for the
condition variable.
pthread_cond_broadcast() wakes up all waiting tasks.
Condition Variable Usage Example
void my_wait_for_event(pthread_mutex_t *lock, pthread_cond_t *cond) {
pthread_mutex_lock(lock);
while (flag == 0)
pthread_cond_wait(cond, lock);
flag = 0;
pthread_mutex_unlock(lock);
}

void my_post_event(pthread_mutex_t *lock, pthread_cond_t *cond) {
pthread_mutex_lock(lock);
flag = 1;
pthread_cond_signal(cond);
pthread_mutex_unlock(lock);
}
The condition variable function calls are used within an area protected by the
mutex that belong to the condition variable. The operating system releases the
mutex every time it blocks a task on the condition variable; and it has locked
the mutex again when it unblocks the calling task from the signaling call.
POSIX Semaphores
POSIX Semaphores also available:
#include <semaphore.h>
int sem_init(sem_t *sem, int pshared, unsigned int value);
int sem_wait(sem_t *sem);
int sem_trywait(sem_t *sem);
int sem_post(sem_t *sem);
int sem_getvalue(sem_t *sem, int *sval);
int sem_destroy(sem_t *sem);
The pshared flag is meaningless in Linux and must be 0.
Shared Memory Using mmap()
void *mmap(void *start, size_t length, int prot , int flags, int fd,
off_t offset);
The mmap() function asks to map length bytes starting at offset
offset from the file (or device) specified by the file descriptor fd
into process virtual memory at a kernel chosen address, but
preferably start with protection of prot.
The actual place where the object is mapped is returned by
mmap() (a pointer).
The return value in case of an error is MAP_FAILED (1) and
NOT 0!
MMAP Flags
MAP_SHARED: Share this mapping with all other processes that map this
object. Storing to the region is equivalent to writing to the file. The file may
not actually be updated until msync(2) or munmap(2) are called.
MAP_FIXED: Do not select a different address than the one specified. If the
specified address cannot be used, mmap() will fail. If MAP_FIXED is
specified, start must be a multiple of the page size.
MAP_PRIVATE: Create a private copyonwrite mapping. Stores to the
region do not affect the original file. It is unspecified whether changes made
to the file after the mmap() call are visible in the mapped region.
MAP_ANONYMOUS: The mapping is not backed by any file; the fd and
offset arguments are ignored.
POSIX Shared Memory
void *shm_open(const char *name, int oflag, mode_t mode);
shm_open() creates and opens a new, or opens an existing, POSIX shared
memory object. A POSIX shared memory object is in effect a handle which
can be used by unrelated processes to mmap(2) the same region of shared
memory. The shm_unlink() function performs the converse operation,
removing an object previously created by shm_open().
The operation of shm_open() is analogous to that of open(2). name specifies
the shared memory object to be created or opened. For portable use, name
should have an initial slash (/) and contain no embedded slashes.
Tip: make sure to reserve the size of the shared memory object using
ftruncate()!
SysV Mailboxes
key_t ftok(const char *pathname, int proj_id);
int msgget(key_t key, int msgflg);
int msgsnd(int msqid, struct msgbuf *msgp, size_t msgsz, int msgflg);
ssize_t msgrcv(int msqid, struct msgbuf *msgp, size_t msgsz, long msgtyp,
int msgflg);
ftok() generates a key from the pathname and project ID proj_id.
msgget() creates a mailbox from the key.
msgsnd() sends a message and msgrcv() retrieves it.
The different flags can control messages types (priorities), if receiving is a
blocking or nonblocking operations and mailbox permissions.
SysV Mailbox Messages
struct msgbuf {
long mtype; /* message type, must be > 0 */
char mtext[1]; /* message data */
};
The mtext field is an array (or other structure) whose size is
specified by msgsz parameter.
The mtype field must have a positive integer value that can
be used by the receiving process for message selection.
SysV Mailbox Message Types
For msgrcv(), the msgtyp argument specifies the type of
message requested as follows:
If msgtyp is 0, then the first message in the queue is read.
If msgtyp is greater than 0, then the first message on the queue
of type msgtyp is read, unless MSG_EXCEPT was asserted in
msgflg, in which case the first message on the queue of type
not equal to msgtyp will be read.
If msgtyp is less than 0, then the first message on the queue
with the lowest type less than or equal to the absolute value of
msgtyp will be read.
SysV Mailboxes Flags
The msgflg argument asserts none, one or more (oring them)
of the following flags:
IPC_NOWAIT For immediate return if no message of the
requested type is on the queue. The system call fails with errno
set to ENOMSG.
MSG_EXCEPT Used with msgtyp greater than 0 to read the
first message on the queue with message type that differs from
msgtyp.
MSG_NOERROR To truncate the message text if longer
than msgsz bytes.
POSIX Pipes
int pipe(int filedes[2]);
pipe() creates a pair of file descriptors, pointing to a pipe inode,
and places them in the array pointed to by filedes. filedes[0] is
for reading, filedes[1] is for writing.
FILE *popen(const char *command, const char *type);
The popen() function opens a process by creating a pipe, forking,
and invoking the shell. Since a pipe is by definition unidirectional,
the type argument may specify only reading or writing, not both;
the resulting stream is correspondingly readonly or writeonly.
A named version of pipes, called fifo also exists.
UNIX Domain Sockets
Files in the file system that acts like sockets.
All normal socket operations applies.
struct sockaddr_un server;
int sock;
sock = socket(AF_UNIX, SOCK_STREAM, 0);
server.sun_family = AF_UNIX;
strcpy(server.sun_path, SOCKET_NAME);
bind(sock, (struct sockaddr *) &server, sizeof(struct sockaddr_un));
UDS can be used to pass file descriptors between processes using a
special socket option operation.
Signals
Signals are asynchronous notifications sent to a process by
the kernel or another process
Signals interrupt whatever the process was doing at the time
to handle the signal.
Each signal may have a signal handler, which is a function
that gets called when the process receives that signal.
Two default signal handlers also exist:
SIG_IGN: Causes the process to ignore the specified signal.
SIG_DFL: Causes the system to set the default signal handler
for the given signal.
Regular Signals
•
SIGHUP Hangup detected on controlling terminal •
SIGBUS Bus error (bad memory access)
•
SIGINT Interrupt from keyboard •
SIGPOLL Pollable event (Sys V). Synonym of SIGIO
•
SIGQUIT Quit from keyboard •
SIGPROF Profiling timer expired
•
SIGILL Illegal Instruction •
SIGSYS Bad argument to routine (SVID)
•
SIGABRT Abort signal from abort(3) •
SIGTRAP Trace/breakpoint trap
•
SIGFPE Floating point exception •
SIGURG Urgent condition on socket (4.2 BSD)
•
SIGKILL Kill signal •
SIGIO I/O now possible (4.2 BSD)
•
SIGSEGV Invalid memory reference
•
SIGPIPE Broken pipe: write to pipe with no readers
•
SIGALRM Timer signal from alarm(2) See signal(7) for default behaviors.
•
SIGTERM Termination signal
•
SIGUSR1 Userdefined signal 1
•
SIGUSR2 Userdefined signal 2 There are two signal handlers you
•
SIGCHLD Child stopped or terminate cannot modify or ignore – SIGKILL
•
SIGCONT Continue if stopped
SIGSTOP Stop process and SIGSTOP.
SIGTSTP Stop typed at tty
SIGTTIN tty input for background process
SIGTTOU tty output for background process
Limitations of Regular Signals
Too few signals
Only 2 signals available for user defined purposes: USR1, USR2. All the rest have
predefined meaning.
No queuing
If the same signal is sent multiple times until a task processes it, it is only delivered once.
No priority
Multiple signals delivered according to order sent.
Portability
Sys V signal handlers are “popped” before executed (sadly, also a racecondition).
Signal Masking
Installing the same signal handler for multiple signals may cause reentrancy.
No passing of a value with the signal
Real Time Signals
Additional 32 signals from SIGRTMIN to SIGRTMAX
No predefined meaning...
But LinuxThreads lib makes use of the first 3.
Multiple instances of the same signals are queued.
Value can be sent with the signal.
Priority is guaranteed:
Lowest number real time signals are delivered first. Same
signals are delivered according to order they were sent.
Regular signals have higher priority then real time signals.
Signal Action
int sigaction(int signum, const struct sigaction *act, struct sigaction
*oldact);
Register a signal handler.
signum: signal number.
act: pointer to new struct sigaction.
oldact: pointer to buffer to be filled with current sigaction (or
NULL, if not interested).
Signal Action cont.
The sigaction structure is defined as something like:
struct sigaction {
void (*sa_handler)(int);
void (*sa_sigaction)(int, siginfo_t *, void *);
sigset_t sa_mask;
int sa_flags;
...
}
sa_mask gives a mask of signals which should be blocked
during the execution of the signal handler.
The signal which triggered the handler will also be blocked,
unless the SA_NODEFER or SA_NOMASK flags are used.
Real Time Signals Flags
sa_flags can be used to pass flags to change behavior:
SA_ONESHOT: Restore the signal action to the default state
once the signal handler has been called.
SA_RESTART: Make blocking system calls restart
automatically aftyer a signal is received.
SA_NODEFER: Do not prevent the signal from being
received from within its own signal handler.
SA_SIGINFO: The signal handler takes 3 arguments, not one.
In this case, sa_sigaction should be set instead of sa_handler.
For details about siginfo_t structure, see sigaction(2).
Sending Signals
int sigqueue(pid_t pid, int sig, const union sigval value);
Queue signal to process.
pid is the process ID to send the signal to.
sig is the signal number.
sigval is:
union sigval {
int sival_int;
void *sival_ptr;
};
The sigval is available to the handler via the sig_value field of
siginfo_t.
Signal Masking
The sigprocmask() call is used to change the list of currently
blocked signals.
int sigprocmask(int how, const sigset_t *set, sigset_t *oldset);
The behavior of the call is dependent on the value of how, as
follows:
SIG_BLOCK: The set of blocked signals is the union of the
current set and the set argument.
SIG_UNBLOCK: The signals in set are removed from the
current set of blocked signals.
SIG_SETMASK: The set of blocked signals is set to the
argument set.
Signal Sets
These functions allow the manipulation of POSIX signal sets:
int sigemptyset(sigset_t *set);
Initializes the signal set given by set to empty, with all signals
excluded from the set.
int sigfillset(sigset_t *set);
Initializes set to full, including all signals.
int sigaddset(sigset_t *set, int signum);
int sigdelset(sigset_t *set, int signum);
Add and delete respectively signal signum from set.
Signals & Threads
Signal masks are per thread.
Signal handlers are per process.
Exception signals (SIGSEGV, SIGBUS...) will be caught by
thread doing the exception.
Other signals will be caught by any thread in the process
whose mask does not block the signal – use
pthread_sigmask() to modify the thread's signal mask.
Tip: Use a “signal handler” thread that does sigwait(3) to
make thread catching less random!
Processes Timers
#include <sys/time.h>
int getitimer(int which, struct itimerval
*value);
int setitimer(int which, const struct itimerval
*value, struct itimerval *ovalue);
The system provides each process with three interval timers,
each decrementing in a distinct time domain.
When any timer expires, a signal is sent to the process, and
the timer (potentially) restarts.
Timers Expiry
The timer interval is controlled by the following structures:
struct itimerval {
struct timeval it_interval; /* next value */
struct timeval it_value; /* current value */
};
struct timeval {
long tv_sec; /* seconds */
long tv_usec; /* microseconds */
};
Timers decrement from it_value to zero, generate a signal, and
reset to it_interval.
A timer which is set to zero (it_value is zero or the timer expires
and it_interval is zero) stops.
Time Domains
ITIMER_REAL
Decrements in real time, and delivers SIGALRM upon
expiration.
ITIMER_VIRTUAL
Decrements only when the process is executing, and delivers
SIGVTALRM upon expiration.
ITIMER_PROF
Decrements both when the process executes and when the
system is executing on behalf of the process.
SIGPROF is delivered upon expiration.
More on Timers
Timers will never expire before the requested time, instead
expiring some short, constant time afterwards, dependent on
the system timer resolution.
If the timer expires while the process is active (always true
for ITIMER_VIRT), the signal will be delivered immediately
when generated.
Otherwise, the delivery will be offset by a small time
dependent on the system load.
An alternate interface called timer_create() allows
specifying which signal will be sent (utilize real time signals)
and spawning of a thread on timer expiry.
Debugging
GDB: The GNU Debugger
GDB runs on the host
Either standalone or via a graphical front end like Eclipse/CDT.
GDBserver runs on the target
GDBserver can attach to an already running processes or start new
processes under the debugger.
GDB talks to GDBserver via TCP or UART
Need to have the executable with debug information on the host.
Also, system and application dynamic libraries, if used.
No need to have debug symbols in executable on target.
See http://www.codefidence.com/sourcedrop/course/gdb_with_embedded_linux_cheat_sheet.pdf
Embedded Linux Driver Development
Kernel Overview
Linux Features
Linux Kernel Development Timeline
Linux Stable Releases
Major versions
1 major version every 2 or 3 years
Examples: 1.0, 2.0, 2.4, 2.6 Even number
Stable releases
1 stable release every 1 or 2 months
Examples: 2.0.40, 2.2.26, 2.4.27, 2.6.7 ...
Stable release updates (since March 2005)
Updates to stable releases up to several times a week
Address only critical issues in the latest stable release
Examples: 2.6.11.1 to 2.6.11.7
Linux Development and Testing Releases
Testing releases
Several testing releases per month, before the next stable one.
You can contribute to making kernel releases more stable by
testing them!
Example: 2.6.12rc1
Development versions
Unstable versions used by kernel developers
before making a new stable major release
Examples: 2.3.42, 2.5.74 Odd number
Continued Development in 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 major changes in existing subsystems.
Opening a new Linux 2.7 (or 2.9) development branch will be
required only when Linux 2.6 is no longer able to accommodate
key features without undergoing traumatic changes.
Thanks to this, more features are released to users at a faster pace.
However, the internal kernel API can undergo changes between
two 2.6.x releases. A module compiled for a given version may
no longer compile or work on a more recent one.
2.4 vs. 2.6
Linux 2.4 Linux 2.6
Mature 3.5 years old stable Linux release!
But developments stopped; very Support from the Linux
few developers willing to help. development community and all
commercial vendors.
Now obsolete and lacks recent
features. Now mature and more exhaustive.
Most drivers upgraded.
Still fine if you get your
sources, tools and support from Cutting edge features and
commercial Linux vendors. increased performance.
Linux Kernel Key Features
Portability and hardware support Security
Runs on most architectures. It can't hide its flaws. Its code is
reviewed by many experts.
Scalability
Can run on super computers as Stability and reliability.
well as on tiny devices
Modularity
(4 MB of RAM is enough).
Can include only what a system
Compliance to standards and needs even at run time.
interoperability.
Easy to program
Exhaustive networking support. You can learn from existing code.
Many useful resources on the net.
Supported Hardware Architectures
See the arch/ directory in the kernel sources
Minimum: 32 bit processors, with or without MMU
32 bit architectures (arch/ subdirectories)
alpha, arm, cris, frv, h8300, i386, m32r, m68k, m68knommu,
mips, parisc, ppc, s390, sh, sparc, um, v850, xtensa
64 bit architectures:
ia64, mips64, ppc64, sh64, sparc64, x86_64
See arch/<arch>/Kconfig, arch/<arch>/README, or
Documentation/<arch>/ for details
Kernel Overview
Kernel Code
Linux Sources Structure (1)
arch/<arch> Architecture specific code
arch/<arch>/mach<mach> Machine / board specific code
COPYING Linux copying conditions (GNU GPL)
CREDITS Linux main contributors
crypto/ Cryptographic libraries
Documentation/ Kernel documentation. Don't miss it!
drivers/ All device drivers (drivers/usb/, etc.)
fs/ Filesystems (fs/ext3/, etc.)
include/ Kernel headers
include/asm<arch> Architecture and machine dependent headers
include/linux Linux kernel core headers
init/ Linux initialization (including main.c)
ipc/ Code used for process communication
Linux Sources Structure (2)
kernel/ Linux kernel core (very small!)
lib/ Misc library routines (zlib, crc32...)
MAINTAINERS Maintainers of each kernel part. Very useful!
Makefile Top Linux makefile (sets arch and version)
mm/ Memory management code (small too!)
net/ Network support code (not drivers)
README Overview and building instructions
REPORTINGBUGS Bug report instructions
scripts/ Scripts for internal or external use
security/ Security model implementations (SELinux...)
sound/ Sound support code and drivers
usr/ Early userspace code (initramfs)
LXR: Linux Cross Reference
http://sourceforge.net/projects/lxr Takes a little bit of time and patience to setup
(configuration, indexing, server
Generic source indexing tool and code
configuration).
browser
Initial indexing quite slow:
Web server based
Linux 2.6.11: 1h 40min on P4 M
Very easy and fast to use
1.6 GHz, 2 MB cache
Identifier or text search available
You don't need to set up LXR by yourself.
Very easy to find the declaration, Use our http://lxr.freeelectrons.com server!
implementation or usages of symbols Other servers available on the Internet:
http://freeelectrons.com/community/kernel/lxr/
Supports C and C++
Supports huge code projects such as the
Linux kernel (260 M in Apr. 2006)
Implemented in C
Implemented in C like all Unix systems.
(C was created to implement the first Unix systems)
A little Assembly is used too:
CPU and machine initialization, critical library routines.
See http://www.tux.org/lkml/#s153
for reasons for not using C++
(main reason: the kernel requires efficient code).
Compiled with GNU C
Need GNU C extensions to compile the kernel.
So, you cannot use any ANSI C compiler!
Some GNU C extensions used in the kernel:
Inline C functions
Inline assembly
Structure member initialization
in any order (also in ANSI C99)
Branch annotation (see next page)
Help gcc Optimize Your Code!
Use the likely and unlikely statements
(include/linux/compiler.h)
Example:
if (unlikely(err)) {
...
}
The GNU C compiler will make your code faster
for the most likely case.
Used in many places in kernel code!
Don't forget to use these statements!
No C library
The kernel has to be standalone and can't use userspace code.
Userspace 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 can't use standard C library functions in kernel code.
(printf(), memset(), malloc()...).
You can also use kernel C headers.
Fortunately, the kernel provides similar C functions for your
convenience, like printk(), memset(), kmalloc()...
Kernel Stack
Very small and fixed stack.
2 page stack (8k), per task.
Or 1 page stack, per task and one for interrupts.
2.6 Chosen in build time via menu.
Not for all architectures
For some architectures, the kernel provides debug facility to
detect stack overruns.
Managing Endianess
Linux supports both little and big endian architectures
Each architecture defines __BIG_ENDIAN or __LITTLE_ENDIAN
in <asm/byteorder.h>
Can be configured in some platforms supporting both.
To make your code portable, the kernel offers conversion macros
(that do nothing when no conversion is needed). Most useful ones:
u32 cpu_to_be32(u32); // CPU byte order to big endian
u32 cpu_to_le32(u32); // CPU byte order to little endian
u32 be32_to_cpu(u32); // Little endian to CPU byte order
u32 le32_to_cpu(u32); // Big endian to CPU byte order
Kernel Coding Guidelines
Never use floating point numbers in kernel code. Your code
may be run on a processor without a floating point unit (like on
arm). Floating point can be emulated by the kernel, but this is
very slow.
Define all symbols as static, except exported ones (avoid name
space pollution)
All system calls return negative numbers (error codes) for
errors:
#include <linux/errno.h>
See Documentation/CodingStyle for more guidelines
Kernel Log
Printing to the kernel log is done via the printk() function.
The kernel keeps the messages in a circular buffer
(so that doesn't consume more memory with many messages).
Kernel log messages can be accessed from user space through system
calls, or through /proc/kmsg
Kernel log messages are also displayed in the system console.
printk()
The printk function:
Similar to stdlib's printf(3)
No floating point format.
Log message are prefixed with a “<0>”, where the number
denotes severity, from 0 (most severe) to 7.
Macros are defined to be used for severity levels:
KERN_EMERG, KERN_ALERT, KERT_CRIT,
KERN_ERR, KERN_WARNING, KERN_NOTICE,
KERN_INFO, KERN_DEBUG.
Usage example:
printk(KERN_DEBUG “Hello World number %d\n”, num);
Accessing the Kernel Log
Many ways are available!
Watch the system console logread
syslogd/klogd Same. Often found in small
embedded systems with no
Daemon gathering kernel messages
/var/log/messages or no
in /var/log/messages
dmesg. Implemented by Busybox.
Follow changes by running:
tail f /var/log/messages cat /proc/kmsg
Caution: this file grows! Waits for kernel messages and
Use logrotate to control this displays them.
dmesg Useful when none of the above
Found in all systems user space programs are available
Displays the kernel log buffer (tiny system)
Linked Lists
Many constructs use doublylinked lists.
List definition and initialization:
struct list_head mylist = LIST_HEAD_INIT(mylist);
or
LIST_HEAD(mylist);
or
INIT_LIST_HEAD(&mylist);
List Manipulation
List definition and initialization:
void list_add(struct list_head *new, struct
list_head *head);
void list_add_tail(struct list_head *new, struct
list_head *head);
void list_del(struct list_head *entry);
void list_del_init(struct list_head *entry);
void list_move(struct list_head *list, struct
list_head *head);
List Manipulation (cont.)
List splicing and query:
void list_splice(struct list_head *list, struct
list_head *head);
void list_add_splice_init(struct list_head *list,
struct list_head *head);
void list_empty(struct list_head *head);
In 2.6, there are variants of these API's for RCU protected
lists (see section about Locks ahead).
List Iteration
Lists also have iterator macros defined:
list_for_each(pos, head);
list_for_each_prev(pos, head);
list_for_each_safe(pos, n, head);
list_for_each_entry(pos, head, member);
Example:
struct mydata *pos;
list_for_each_entry(pos, head, dev_list) {
pos>some_data = 0777;
}
Kernel Overview
Boot Sequence
Linux Boot Process
BIOS and/or bootloader initializes hardware.
Bootloader loads kernel image into memory.
Bootloader can get kernel image from flash, HD, network.
Possibly also loads a filesystem to RAM in the form of initrd
or initramfs archives.
Bootloader or kernel decompress compressed kernel.
Kernel performs internal (hash table, lists etc.) and hardware
(device driver) setup.
Kernel finds and mounts the root filesystem.
Kernel executes the “/sbin/init” application.
Boot Sequences
x86 PC Arm PowerPC

Init
Kernel
Decompression Boot loader
Setup and
bzImage [ U-boot
decompresses the
zImage
Real Mode kernel image
trampoline on it's own ]
BIOS and
GRUB
RedBoot U-Boot
BIOS
Root Filesystem Options
External FS Initrd Style Initramfs style

Kernel
2.4 and 2.6 2.4 and 2.6 2.6 only
Version
Filesystem on
Storage
storage device or Filesystem image CPIO archive
format
network
Passed to kernel Loaded and Like initrd or
Provided
as parameter location provided statically linked
by
(“root=”) by boot loader into kernel image
Dynamically
Fixed allocation
Stored in On device allocated
RAM disk
page cache
Run time Copied to Copied to Already in
Location page cache page cache page cache
init – the First Process
The first (and only) process the kernel starts is init.
By default it is searched in /sbin/init
Can be overridden by the kernel parameter “init=”.
init has three roles:
To setup the system configuration and start applications.
To shut down the system applications.
The serve as the parent process of all child processes whose
parent has exited.
The default init implementation reads its instructions from
the file /etc/inittab
inittab
This is an Busybox style inittab (for an # Startup the system
example for Sys V inittab see the ::sysinit:/bin/mount o remount,rw /
Appendix.) ::sysinit:/bin/mount t proc proc /proc
::sysinit:/bin/mount a
Format:
::sysinit:/sbin/ifconfig lo 127.0.0.1 up
id: runlevel : action : command
# Put a getty on the serial port
id: To which device should std ::respawn:/sbin/getty L ttyS1 115200 vt100
input/output go (empty means the
console)
# Start system loggers
runlevel: ignored. For compatibility null::respawn:/sbin/syslogd n m 0
with Sys V init. null::respawn:/sbin/klogd n
action: one of sysinit, respawn,

askfirst, wait, shutdown and once # Stuff to do before rebooting
null::shutdown:/usr/bin/killall klogd
command: shell command to execute null::shutdown:/usr/bin/killall syslogd
null::shutdown:/bin/umount a r
Kernel Overview
Kernel Subsystems
Kernel Architecture
App1 App2 ...
User
space
C library
System call interface
Process Memory Filesystem Device

management Networking
management support control Kernel
space
Filesystem
types
CPU support CPU / MMU Storage Character Network

code support code drivers device drivers device drivers
Hardware
CPU RAM Storage

KernelMode vs. UserMode
All modern CPUs support a dual mode of operation:
Usermode, for regular tasks.
Supervisor (or privileged) mode, for the kernel.
The mode the CPU is in determines which instructions the
CPU is willing to execute:
“Sensitive” instructions will not be executed when the CPU is
in user mode.
The CPU mode is determined by one of the CPU registers,
which stores the current “Ring Level”
0 for supervisor mode, 3 for user mode, 12 unused by Linux.
The System Call Interface
When a userspace tasks needs to use a kernel service, it will make a
“System Call”.
The C library places parameters and number of system call in registers
and then issues a special trap instruction.
The trap atomically changes the ring level to supervisor mode and the
sets the instruction pointer to the kernel.
The kernel will find the required system called via the system call table
and execute it.
Returning from the system call does not require a special instruction,
since in supervisor mode the ring level can be changed directly.
Linux System Call Path
do_name() Function call

Kernel
Trap
sys_name()
entry.S
Glibc
Task
Task
Driver Development
Loadable Kernel Modules
Loadable Kernel Modules (1)
Modules: add a given functionality to the kernel (drivers,
filesystem support, and many others).
Can be loaded and unloaded at any time, only when their
functionality is need. Once loaded, have full access to the
whole kernel. No particular protection.
Useful to keep the kernel image size to the minimum
(essential in GNU/Linux distributions for PCs).
Loadable Kernel Modules (2)
Useful to support incompatible drivers (either load one or the
other, but not both).
Useful to deliver binaryonly drivers (bad idea) without
having to rebuild the kernel.
Modules make it easy to develop drivers without rebooting:
load, test, unload, rebuild, load...
Modules can also be compiled statically into the kernel.
Hello Module
/* hello.c */
#include <linux/init.h>
#include <linux/module.h>
#include <linux/kernel.h>
static int __init hello_init(void)
{
printk(KERN_ALERT "Good morrow"); __init:
printk(KERN_ALERT "to this fair assembly.\n");
return 0; removed after initialization
}
(static kernel or module).
static void __exit hello_exit(void)
{
printk(KERN_ALERT "Alas, poor world, what treasure");
__exit: discarded when
printk(KERN_ALERT "hast thou lost!\n");
}
module compiled statically
into the kernel.
module_init(hello_init);
module_exit(hello_exit);
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("Greeting module");
MODULE_AUTHOR("William Shakespeare");
Example available on http://freeelectrons.com/doc/c/hello.c
Module License Usefulness
Used by kernel developers to identify issues coming from
proprietary drivers, which they can't do anything about.
Useful for users to check that their system is 100% free.
Useful for GNU/Linux distributors for their release policy
checks.
Possible Module License Strings
Available license strings explained in include/linux/module.h
GPL Dual BSD/GPL
GNU Public License v2 or later GNU Public License v2 or
GPL v2 BSD license choice
GNU Public License v2 Dual MPL/GPL
GPL and additional GNU Public License v2 or
rights Mozilla license choice
Proprietary
Non free products
Compiling a Module
The below Makefile should be reusable for any Linux 2.6 module.
Just run make to build the hello.ko file
Caution: make sure there is a [Tab] character at the beginning of
the $(MAKE) line (make syntax) Either
full kernel source
# Makefile for the hello module directory
(configured and
objm := hello.o compiled)
KDIR := /lib/modules/$(shell uname r)/build or just kernel
PWD := $(shell pwd)
headers directory
[Tab]! default:
$(MAKE) C $(KDIR) SUBDIRS=$(PWD) modules (minimum needed )
(no spaces)
Example available on http://freeelectrons.com/doc/c/Makefile
Using the Module
Need to be logged as root
Load the module:
insmod ./hello.ko
You will see the following in the kernel log:
Good morrow
to this fair assembly
Now remove the module:
rmmod hello
You will see:
Alas, poor world, what treasure
hast thou lost!
Module Dependencies
Module dependencies stored in
/lib/modules/<version>/modules.dep
They don't have to be described by the module writer.
They are automatically computed during kernel building from
module exported symbols. module2 depends on module1 if
module2 uses a symbol exported by module1.
You can update the modules.dep file by running (as root)
depmod a [<version>]
Module Utilities (1)
modinfo <module_name>
modinfo <module_path>.ko
Gets information about a module: parameters, license,
description. Very useful before deciding to load a module or not.
insmod <module_name>
insmod <module_path>.ko
Tries to load the given module, if needed by searching for its
.ko file throughout the default locations (can be redefined by
the MODPATH environment variable).
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.
lsmod
Displays the list of loaded modules
Compare its output with the contents of /proc/modules!
rmmod <module_name>
Tries to remove the given module
modprobe r <module_name>
Tries to remove the given module and all dependent modules
(which are no longer needed after the module removal)
Driver Development
Module Parameters
Hello Module with Parameters
/* hello_param.c */
#include <linux/init.h> Thanks to
#include <linux/module.h>
#include <linux/moduleparam.h> Jonathan Corbet
for the example!
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);
static int __init hello_init(void)
{
int i;
for (i = 0; i < howmany; i++)
printk(KERN_ALERT "(%d) Hello, %s\n", i, whom);
return 0;
}
static void __exit hello_exit(void)
{
printk(KERN_ALERT "Goodbye, cruel %s\n", whom);
}
module_init(hello_init);
module_exit(hello_exit);
Example available on http://freeelectrons.com/doc/c/hello_param.c
Passing Module Parameters
Through insmod or modprobe:
insmod ./hello_param.ko howmany=2 whom=universe
Through modprobe
after changing the /etc/modprobe.conf file:
options hello_param howmany=2 whom=universe
Through the kernel command line, when the module is built statically
into the kernel:
options hello_param.howmany=2 hello_param.whom=universe
module name
module parameter name
module parameter value
Declaring a Module Parameter
#include <linux/moduleparam.h>
module_param(
name, /* name of an already defined variable */
type, /* either byte, short, ushort, int, uint, long,
ulong, charp, bool or invbool
(checked at compile time!) */
perm /* for /sys/module/<module_name>/<param>
0: no such module parameter value file */
);
Example
int irq=5;
module_param(irq, int, S_IRUGO);
Declaring a Module Parameter Array
#include <linux/moduleparam.h>
module_param_array(
name, /* name of an already defined array */
type, /* same as in module_param */
num, /* address to put number of elements in the array, or NULL */
perm /* same as in module_param */
);
Example
static int base[MAX_DEVICES] = { 0x820, 0x840 };
module_param_array(base, int, &count, 0);
Driver Development
Using the proc Filesystem Interface
Proc Filesystem Interface
/proc is a virtual filesystem that exports kernel internal structures to user
space
/proc/cpuinfo: processor information
/proc/meminfo: memory status
/proc/version: version and build information
/proc/cmdline: kernel command line
/proc/<pid>/fd: process used file descriptors
/proc/<pid>/cmdline: process command line
/proc/sys/kernel/panic: time in second until reboot in case of
fatal error
...
Userspace Interface Documentation
Lots of details about the /proc interface are available in
Documentation/filesystems/proc.txt
(almost 2000 lines) in the kernel sources.
You can also find other details in the proc manual page:
man proc
See the New Device Model section for details about /sys
Hello Module with proc File
#include <linux/proc_fs.h> The proc file name
#define MYNAME "driver/my_proc_file"
The proc_dir _entry
static struct proc_dir_entry *my_proc_dir = NULL;
struct
int mymodule_proc_read(char *page, char **start, off_t off,
int count, int *eof, void *data) {
Callback function
int len = 0;

len += sprintf(page + len, "io=%d\n", io);
len += sprintf(page + len, "irq=%d\n", irq); page is the buffer we
if (len <= off+count) write to
*eof = 1;
*start = page + off;
len = off;
if (len > count) len = count; Setting *eof to 1 means
if (len < 0) len = 0;
return len;
end of file.
}
int __init startup_mymodule(void) { start is set to a pointer
my_proc_dir = create_proc_entry(MYNAME, 0, NULL); to where we wrote
my_proc_dir>read_proc = mymodule_proc_read;
return 0;
}
second parameters is
void __exit shutdown_mymodule(void) {
remove_proc_entry(MYNAME, NULL); file permission, last
} parameter is a handle
of directory.
© Copyright 20032006, Oron Peled Creative Commons AttributionShareAlike 2.0 license Don't forget to check
© Copyright 20042006 Codefidence Ltd. for NULL here! 178
Some More proc Details
You can also register a proc_write() callback.
proc_dir_entry has a data field. The kernel does not use it, but
whatever you set there will be returned to you as the last parameter of the
callback.
The permissions (2nd) parameter of create_proc_entry() is the same
as the mode flags of the open(2) system call.
0 means use the system wide defaults.
The directory handle (3rd) parameter of create_proc_entry() is an
address of proc_dir_entry for a proc directory.
NULL means the proc root directory.
For large and complex files use the seq_file wrapper.
Using a proc File
Once the module is loaded, you can access the registered proc file:
From the shell:
Read cat /proc/driver/my_proc_file
Write echo “123” > /proc/driver/my_proc_file
Programatically, using open(2), read(2) write(2) and related
functions.
You can't delete, move or rename a proc file.
proc files usually don't have reported size.
Driver Development
Memory Management
Physical Memory
In ccNUMA1 machines:
The memory of each node is represented at pg_data_t
These memories are linked into pgdat_list
In uniform memory access systems:
There is just one pg_data_t named contig_page_data
If you don't know which of these is your machine, you're
using a uniform memory access system :)
1
ccNUMA: Cache Coherent Non Uniform Memory Access
Memory Zones
Each pg_data_t is split to three zones
Each zone has different properties:
ZONE_DMA
DMA operations on address limited buses is possible.
ZONE_NORMAL
Maps directly to linear addressing (<~1Gb on i386)
Always mapped to kernel space.
ZONE_HIMEM
Rest of memory.
Mapped into kernelspace on demand.
Physical and Virtual Memory
Physical address space Virtual address spaces
0xFFFFFFFFF 0xFFFFFFFFF 0xFFFFFFFFF
I/O memory 3
Kernel Process1
I/O memory 2
0x00000000 0x00000000
I/O memory 1
Flash MMU CPU
0xFFFFFFFFF
Memory
RAM 1 Management All the processes have
Unit their own virtual address Process2
RAM 0
space, and run as if they
had access to the whole 0x00000000
0x00000000 address space.
3:1 Virtual Memory Map
Physical address space Random Virtual address spaces
0xFFFFFFFFF opportunistic 0xFFFFFFFFF
Kernel
I/O memory mappings Virtual Addresses
VMALLOC_START
Kernel
Logical
Addresses
Rest of RAM PAGE_OFFSET
0xC0000000
User Space
840 Mb 1:1
persistent
mapping Zero Page
0x00000000 0x00000000
Address Types
Physical address
Physical memory as seen from the CPU, without MMU1
translation.
Bus address
Physical memory as seen from device bus.
May or may not be virtualized (via IOMMU, GART, etc).
Virtual address
Memory as seen from the CPU, with MMU1 translation.
1
MMU: Memory Management Unit
Address Translation Macros
bus_to_phys(address)
*_phys_to_bus(address)
phys_to_virt(address)
virt_to_phys(address)
*_bus_to_virt(address)
virt_to_bus(address)
Where * are different bus names.
The Memory Management Unit
Task Virtual Physical Permission
12 0x8000 0x5340 RWX
12 0x8001 0x1000 RX
15 0x8000 0x3390 RX
CPU MMU Memory
Translation Lookaside Buffers
The tables that describes the virtual to physical translations
are called page tables.
They reside in system memory.
The MMU caches the content of these tables in a CPU local
cache dubbed the TLB, or Translation Lookaside Buffer.
Making the changes to the page tables often requires TLB
flushes.
For example, while context switching between two processes.
TLB flushes are expensive, at least on some architectures.
kmalloc() and kfree()
Basic allocators, kernel equivalents of glibc's malloc() and
free().
#include <linux/slab.h>
static inline void *kmalloc(size_t size, int flags);
size: number of bytes to allocate
flags: priority (see next page)
void kfree (const void *objp);
Example:
data = kmalloc(sizeof(*data), GFP_KERNEL);
...
kfree(data);
kmalloc() Features
Quick (unless it's blocked waiting for memory to be freed).
Doesn't initialize the allocated area.
You can use kcalloc or kzalloc to get zeroed memory.
The allocated area is contiguous in physical RAM.
Allocates by 2n sizes, and uses a few management bytes.
So, don't ask for 1024 when you need 1000! You'd get 2048!
Caution: drivers shouldn't try to kmalloc
more than 128 KB (upper limit in some architectures).
Memory Allocation Flags (1)
Defined in include/linux/gfp.h (GFP: get_free_pages)
GFP_KERNEL
Standard kernel memory allocation. May block. Fine for most needs.
GFP_ATOMIC
Allocated RAM from interrupt handlers or code not triggered by user
processes. Never blocks.
GFP_USER
Allocates memory for user processes. May block. Lowest priority.
Memory Allocation Flags (2)
Extra flags (can be added with |)
__GFP_DMA __GFP_NORETRY
Allocate in DMA zone If allocation fails, doesn't try to
__GFP_REPEAT get free pages.
Ask to try harder. May still Example:
block, but less likely. GFP_KERNEL | __GFP_DMA
__GFP_NOFAIL
Must not fail. Never gives up.
Caution: use only when
mandatory!
Slab Caches
Also called lookaside caches
Slab: name of the standard Linux memory allocator
Slab caches: Objects that can hold any number
of memory areas of the same size.
Optimum use of available RAM and reduced fragmentation.
Mainly used in Linux core subsystems: filesystems (open files, inode
and file caches...), networking... Live stats on /proc/slabinfo.
May be useful in device drivers too, though not used so often.
Linux 2.6: used by USB and SCSI drivers.
Slab Cache API (1)
#include <linux/slab.h>
Creating a private cache:
cache = kmem_cache_create(
name, /* Name for /proc/slabinfo */
size, /* Cache object size */
flags, /* Options: alignment, DMA... */
constructor, /* Optional, called after each allocation */
destructor); /* Optional, called before each release */
Slab Cache API (2)
Allocating from the cache:
object = kmem_cache_alloc(cache, flags);
Freing an object:
kmem_cache_free(cache, object);
Destroying the whole cache:
kmem_cache_destroy(cache);
More details and an example in the Linux Device Drivers book:
http://lwn.net/images/pdf/LDD3/ch08.pdf
Memory Pools
Useful for memory allocations that cannot fail.
Kind of lookaside cache trying to keep a minimum number
of preallocated objects ahead of time.
Use with care: otherwise can result in a lot of unused
memory that cannot be reclaimed! Use other solutions
whenever possible.
Memory Pool API (1)
#include <linux/mempool.h>
Mempool creation:
mempool = mempool_create(
min_nr,
alloc_function,
free_function,
pool_data);
Memory Pool API (2)
Allocating objects:
object = mempool_alloc(pool, flags);
Freeing objects:
mempool_free(object, pool);
Resizing the pool:
status = mempool_resize(
pool, new_min_nr, flags);
Destroying the pool (caution: free all objects first!):
mempool_destroy(pool);
Memory Pool Implementation
Call alloc
mempool_create function min_nr
times
No Take an
Call alloc
mempool_alloc Success? object from
function
the pool
Yes
Yes
pool count Add freed
mempool_free New object
< min_nr? object to pool
No
Call free
function
on object
Memory Pools Using Slab Caches
Idea: use slab cache functions to allocate and free objects.
The mempool_alloc_slab and mempool_free_slab
functions supply a link with slab cache routines.
So, you will find many code examples looking like:
cache = kmem_cache_create(...);
pool = mempool_create(
min_nr,
mempool_alloc_slab,
mempool_free_slab,
cache);
The Buddy System
Kernel memory page allocation follows the “Buddy”
System.
Free Page Frames are allocated in powers of 2:
If suitable page frame is found, allocate.
Else: seek higher order frame, allocate half, keep
“buddy”
When freeing page frames, coalescing occurs.
Buddy System 1
16 Mb
We need 8 Mb of memory, but

don't find an exact match.
We do have a block of 16 Mb
memory though.
Buddy System 2
8 Mb 8 Mb
So we'll split the 16 Mb into two 8

Mb areas.
Buddy System 3
8 Mb 8 Mb
We'll use 8 Mb and keep the rest

as a free block of 8 Mb.
Buddy System 4
8 Mb 8 Mb
When the allocated memory has

been freed...
Buddy System 5
16 Mb
We can once again combine the

two blocks in a single 16 Mb free
block.
Because of the order of 2

allocation, it's easy to spot our
“buddy”.
Allocating by Pages
More appropriate when you need big slices of RAM:
unsigned long get_zeroed_page(int flags);
Returns a pointer to a free page and fills it up with zeros
unsigned long __get_free_page(int flags);
Same, but doesn't initialize the contents
unsigned long __get_free_pages(int flags,
unsigned long order);
Returns a pointer on a memory zone of several contiguous pages in
physical RAM.
order: log2(<number_of_pages>)
maximum: 8192 KB (MAX_ORDER=11 in linux/mmzone.h)
Freeing Pages
void free_page(unsigned long addr);
void free_pages(unsigned long addr,
unsigned long order);
Need to use the same order as in allocation.
vmalloc()
vmalloc() can be used to obtain contiguous memory zones in
virtual address space (even if pages may not be contiguous in
physical memory).
void *vmalloc(unsigned long size);
void vfree(void *addr);
Memory Utilities
void *memset(void *s, int c, size_t count);
Fills a region of memory with the given value.
void *memcpy(void *dest,
const void *src,
size_t count);
Copies one area of memory to another.
Use memmove() for overlapping areas.
Lots of functions equivalent to standard C library ones defined in
include/linux/string.h
Memory Management Summary
Small allocations Bigger allocations
kmalloc, kzalloc __get_free_page[s],
(and kfree!) get_zeroed_page,
free_page[s]
slab caches
vmalloc, vfree
memory pools
Libc like memory utilities
memset, memcopy,
memmove...
Driver Development
I/O Memory and Ports
Requesting I/O Ports
/proc/ioports example struct resource *request_region(
0000001f : dma1 unsigned long start,
00200021 : pic1
00400043 : timer0
unsigned long len,
00500053 : timer1 char *name);
0060006f : keyboard
00700077 : rtc
0080008f : dma page reg
00a000a1 : pic2 Tries to reserve the given region and returns NULL if
00c000df : dma2
00f000ff : fpu unsuccessful. Example:
0100013f : pcmcia_socket0
01700177 : ide1
01f001f7 : ide0
03760376 : ide1
request_region(0x0170, 8, "ide1");
0378037a : parport0
03c003df : vga+ void release_region(
03f603f6 : ide0
03f803ff : serial unsigned long start,
0800087f : 0000:00:1f.0
08000803 : PM1a_EVT_BLK
unsigned long len);
08040805 : PM1a_CNT_BLK
0808080b : PM_TMR
08200820 : PM2_CNT_BLK
See include/linux/ioport.h and
0828082f : GPE0_BLK kernel/resource.c
...
Reading/Writing on I/O Ports
The implementation of the below functions and the exact unsigned
type can vary from architecture to architecture!
bytes
unsigned inb(unsigned port);
void outb(unsigned char byte, unsigned port);
words
unsigned inw(unsigned port);
void outw(unsigned char byte, unsigned port);
"long" integers
unsigned inl(unsigned port);
void outl(unsigned char byte, unsigned port);
Reading/Writing Strings on I/O Ports
Often more efficient than the corresponding C loop, if the processor
supports such operations!
byte strings
void insb(unsigned port, void *addr, unsigned long count);
void outsb(unsigned port, void *addr, unsigned long count);
word strings
void insw(unsigned port, void *addr, unsigned long count);
void outsw(unsigned port, void *addr, unsigned long count);
long strings
void inbsl(unsigned port, void *addr, unsigned long count);
void outsl(unsigned port, void *addr, unsigned long count);
Requesting I/O Memory
/proc/iomem example
Equivalent functions with the same interface
000000000009efff : System RAM
0009f0000009ffff : reserved struct resource *request_mem_region(
000a0000000bffff : Video RAM area unsigned long start,
000c0000000cffff : Video ROM
000f0000000fffff : System ROM unsigned long len,
001000003ffadfff : System RAM char *name);
001000000030afff : Kernel code
0030b000003b4bff : Kernel data
3ffae0003fffffff : reserved
void release_mem_region(
40000000400003ff : 0000:00:1f.1 unsigned long start,
4000100040001fff : 0000:02:01.0 unsigned long len);
4000100040001fff : yenta_socket
4000200040002fff : 0000:02:01.1
4000200040002fff : yenta_socket
40400000407fffff : PCI CardBus #03
4080000040bfffff : PCI CardBus #03
40c0000040ffffff : PCI CardBus #07
41000000413fffff : PCI CardBus #07
a0000000a0000fff : pcmcia_socket0
a0001000a0001fff : pcmcia_socket1
e0000000e7ffffff : 0000:00:00.0
e8000000efffffff : PCI Bus #01
e8000000efffffff : 0000:01:00.0
...
Mapping I/O Memory into Virtual Memory
To access I/O memory, drivers need to have a virtual address
that the processor can handle.
The ioremap() functions satisfy this need:
#include <asm/io.h>
void *ioremap(unsigned long phys_addr,
unsigned long size);
void iounmap(void *address);
Caution: check that ioremap doesn't return a NULL address!
Differences from Standard Memory
Reads and writes on memory can be cached.
The compiler may choose to write the value in a CPU
register, and may never write it in main memory.
The compiler may decide to optimize or reorder read and
write instructions.
Avoiding I/O Access Issues
Caching on I/O ports or memory already disabled, either by
the hardware or by Linux init code.
Memory barriers are supplied to avoid reordering:
Hardware independent Hardware dependent
#include <asm/kernel.h> #include <asm/system.h>
void barrier(void); void rmb(void);
void wmb(void);
Only impacts the behavior of the void mb(void);
compiler. Doesn't prevent reordering Safe on all architectures!
in the processor!
Accessing I/O Memory
Directly reading from or writing to addresses returned by ioremap()
(“pointer dereferencing”) may not work on some architectures.
Use the below functions instead. They are always portable and safe:
unsigned int ioread8(void *addr); (same for 16 and 32)
void iowrite8(u8 value, void *addr); (same for 16 and 32)
To read or write a series of values:
void ioread8_rep(void *addr, void *buf, unsigned long count);
void iowrite8_rep(void *addr, const void *buf, unsigned long count);
Other useful functions:
void memset_io(void *addr, u8 value, unsigned int count);
void memcpy_fromio(void *dest, void *source, unsigned int count);
void memcpy_toio(void *dest, void *source, unsigned int count);
Driver Development
Character Drivers
Usefulness of Character Drivers
Except for storage device drivers, most drivers for devices with
input and output flows are implemented as character drivers.
So, most drivers you will face will be character drivers
You will regret if you sleep during this part!
Creating a Character Driver
Userspace
Userspace needs
The name of a device file in /dev to interact Read
buffer
Write
string
with the device driver through regular file
operations (open, read, write, close...) read write
The kernel needs /dev/foo
To know which driver is in charge of device
Copy from user
Copy to user
major / minor
files with a given major / minor number pair
For a given driver, to have handlers (“file
Read Write
operations”) to execute when userspace opens, handler handler
reads, writes or closes the device file. Device driver
Kernel space
Declaring a Character Driver
Device number registration
Need to register one or more device numbers (major/minor pairs),
depending on the number of devices managed by the driver.
Need to find free ones!
File operations registration
Need to register handler functions called when user space programs
access the device files: open, read, write, ioctl, close...
Information on Registered Devices
Registered devices are visible in /proc/devices:
Character devices: Block devices:
1 mem 1 ramdisk
4 /dev/vc/0 3 ide0
4 tty 8 sd
4 ttyS 9 md Can be used to
5 /dev/tty
5 /dev/console
22 ide1
65 sd
find free major
5 /dev/ptmx 66 sd numbers
6 lp 67 sd
7 vcs 68 sd
10 misc 69 sd
13 input
14 sound
... Major Registered
number name
dev_t Structure
Kernel data structure to represent a major/minor pair.
Defined in <linux/kdev_t.h>
Linux 2.6: 32 bit size (major: 12 bits, minor: 20 bits)
Macro to create the structure:
MKDEV(int major, int minor);
Macros to extract the numbers:
MAJOR(dev_t dev);
MINOR(dev_t dev);
Allocating Fixed Device Numbers
#include <linux/fs.h>
int register_chrdev_region(
dev_t from, /* Starting device number */
unsigned count, /* Number of device numbers */
const char *name); /* Registered name */
Returns 0 if the allocation was successful.
Example
if (register_chrdev_region(MKDEV(202, 128),
acme_count, “acme”)) {
printk(KERN_ERR “Failed to allocate device number\n”);
...
Dynamic Allocation of Device Numbers
Safer: have the kernel allocate free numbers for you!
#include <linux/fs.h>
int alloc_chrdev_region(
dev_t *dev, /* Output: starting device number */
unsigned baseminor, /* Starting minor number, usually 0 */
unsigned count, /* Number of device numbers */
const char *name); /* Registered name */
Returns 0 if the allocation was successful.
Example
if (alloc_chrdev_region(&acme_dev, 0, acme_count, “acme”)) {
printk(KERN_ERR “Failed to allocate device number\n”);
...
Creating Device Files
Issue: you can no longer create /dev entries in advance!
You have to create them on the fly after loading the driver according to
the allocated major number.
Tip: read /proc/devices after module loads to find out which
major number your driver got.
Trick: the script loading the module can then use /proc/devices:
rm f /dev/my_dev_file
insmod my_driver.ko
cat /proc/devices
mknod /dev/my_dev_file c 254 0
File Operations (1)
Before registering character devices, you have to define
file_operations (called fops) for the device files.
Here are the main ones:
int (*open)(
struct inode *, /* Corresponds to the device file */
struct file *); /* Corresponds to the open file descriptor */
Called when userspace opens the device file.
int (*release)(
struct inode *,
struct file *);
Called when userspace closes the file.
The file Structure
Is created by the kernel during the open call. Represents open files.
Pointers to this structure are usually called "fips".
mode_t f_mode;
The file opening mode (FMODE_READ and/or FMODE_WRITE)
loff_t f_pos;
Current offset in the file.
struct file_operations *f_op;
Allows to change file operations for different open files!
struct dentry *f_dentry
Useful to get access to the inode: filp>f_dentry>d_inode.
File Operations (2)
ssize_t (*read)(
struct file *, /* Open file descriptor */
char *, /* Userspace buffer to fill up */
size_t, /* Size of the userspace buffer */
loff_t *); /* Offset in the open file */
Called when userspace reads from the device file.
ssize_t (*write)(
struct file *, /* Open file descriptor */
const char *, /* Userspace buffer to write to the device */
size_t, /* Size of the userspace buffer */
loff_t *); /* Offset in the open file */
Called when userspace writes to the device file.
Exchanging Data With UserSpace (1)
In driver code, you can't just memcpy between
an address supplied by userspace and
the address of a buffer in kernelspace!
Correspond to completely different
address spaces (thanks to virtual memory).
The userspace address may be swapped out to disk.
The userspace address may be invalid
(user space process trying to access unauthorized data).
Exchanging Data With UserSpace (2)
You must use dedicated functions such as the following ones in your
read and write file operations code:
include <asm/uaccess.h>
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);
Make sure that these functions return 0!
Another return value would mean that they failed.
File Operations (3)
int (*ioctl) (struct inode *, struct file *,
unsigned int, unsigned long);
Can be used to send specific commands to the device, which are neither
reading nor writing (e.g. formatting a disk, configuration changes).
int (*mmap) (struct file *,
struct vm_area_struct);
Asking for device memory to be mapped into the address space of a user
process
struct module *owner;
Used by the kernel to keep track of who's using this structure and count
the number of users of the module. Set to THIS_MODULE.
Read Operation Example
static ssize_t
acme_read(struct file *file, char __user *buf, size_t count, loff_t * ppos)
{
/* The hwdata address corresponds to a device I/O memory area */
/* of size hwdata_size, obtained with ioremap() */
int remaining_bytes;
/* Number of bytes left to read in the open file */
remaining_bytes = min(hwdata_size (*ppos), count);

if (remaining_bytes == 0) {
/* All read, returning 0 (End Of File) */
return 0;
}
if (copy_to_user(buf /* to */, *ppos+hwdata /* from */, remaining_bytes)) {
return EFAULT;
/* Increase the position in the open file */
*ppos += remaining_bytes;
return remaining_bytes;
}
Read method Piece of code available on
http://freeelectrons.com/doc/c/acme_read.c
Write Operation Example
static ssize_t
acme_write(struct file *file, const char __user *buf, size_t count, loff_t * ppos)
{
/* Assuming that hwdata corresponds to a physical address range */
/* of size hwdata_size, obtained with ioremap() */
/* Number of bytes not written yet in the device */
remaining_bytes = hwdata_size (*ppos);

if (count > remaining_bytes) {
/* Can't write beyond the end of the device */
return EIO;
}
if (copy_from_user(*ppos+hwdata /* to */, buf /* from */, count)) {
return EFAULT;
/* Increase the position in the open file */
*ppos += count;
return count;
}
Write method Piece of code available on
http://freeelectrons.com/doc/c/acme_write.c
File Operations Definition Example
Defining a file_operations structure
include <linux/fs.h>
static struct file_operations acme_fops =
{
.owner = THIS_MODULE,
.read = acme_read,
.write = acme_write,
};
You just need to supply the functions you implemented!
Defaults for other functions (such as open, release...)
are fine if you do not implement anything special.
Character Device Registration (1)
The kernel represents character drivers using the cdev structure.
Declare this structure globally (within your module):
#include <linux/cdev.h>
static struct cdev *acme_cdev;
In the init function, allocate the structure and set its file operations:
acme_cdev = cdev_alloc();
acme_cdev>ops = &acme_fops;
acme_cdev>owner = THIS_MODULE;
Character Device Registration (2)
Now that your structure is ready, add it to the system:
int cdev_add(
struct cdev *p, /* Character device structure */
dev_t dev, /* Starting device major / minor number */
unsigned count); /* Number of devices */
Example (continued):
if (cdev_add(acme_cdev, acme_dev, acme_count)) {
printk (KERN_ERR “Char driver registration failed\n”);
...
Character Device Unregistration
First delete your character device:
void cdev_del(struct cdev *p);
Then, and only then, free the device number:
void unregister_chrdev_region(dev_t from,
unsigned count);
Example (continued):
cdev_del(acme_cdev);
unregister_chrdev_region(acme_dev, acme_count);
Linux Error Codes
Try to report errors with error numbers as accurate as possible!
Fortunately, macro names are explicit and you can remember
them quickly.
Generic error codes:
include/asmgeneric/errnobase.h
Platform specific error codes:
include/asm/errno.h
Char Driver Example Summary (1)
static void *hwdata;
static hwdata_size=8192;
static int acme_count=1;
static dev_t acme_dev;
static struct cdev *acme_cdev;
static ssize_t acme_write(...) {...}
static ssize_t acme_read(...) {...}
static struct file_operations acme_fops =
{
.owner = THIS_MODULE,
.read = acme_read,
.write = acme_write,
};
Char Driver Example Summary (2)
static int __init acme_init(void)
{ if (cdev_add(acme_cdev, acme_dev,
int err; acme_count)) {
hwdata = ioremap(PHYS_ADDRESS, err=ENODEV;
hwdata_size); goto err_free_cdev;
}
if (!acme_buf) {
err = ENOMEM; return 0;
goto err_exit;
} err_free_cdev:
kfree(acme_cdev);
if (alloc_chrdev_region(&acme_dev, 0, err_dev_unregister:
acme_count, “acme”)) { unregister_chrdev_region(
err=ENODEV; acme_dev, acme_count);
goto err_free_buf; err_free_buf:
} iounmap(hwdata);
err_exit:
acme_cdev = cdev_alloc(); return err;
}
if (!acme_cdev) {
err=ENOMEM; static void __exit acme_exit(void)
goto err_dev_unregister; {
} cdev_del(acme_cdev);
unregister_chrdev_region(acme_dev,
acme_cdev>ops = &acme_fops; acme_count);
acme_cdev>owner = THIS_MODULE; iounmap(hwdata);
}
Show how to handle errors and deallocate resources in the right order!
Character Driver Summary
Character driver writer
Define the file operations callbacks for the device file: read, write, ioctl...
Kernel
In the module init function, get major and minor numbers with alloc_chrdev_region(),
init a cdev structure with your file operations and add it to the system with cdev_add().
In the module exit function, call cdev_del() and unregister_chrdev_region()
System administration
Load the character driver module
In /proc/devices, find the major number it uses.
Userspace
Create the device file with this major number
The device file is ready to use!
System user
Open the device file, read, write, or send ioctl's to it.
Kernel
Kernel
Executes the corresponding file operations
Driver Development
mmap()
mmap() (1)
Possibility to have parts of the virtual address space of a program
mapped to the contents of a file!
> cat /proc/1/maps (init process)
start end perm offset major:minor inode mapped file name
007710000077f000 rxp 00000000 03:05 1165839 /lib/libselinux.so.1
0077f00000781000 rwp 0000d000 03:05 1165839 /lib/libselinux.so.1
0097d00000992000 rxp 00000000 03:05 1158767 /lib/ld2.3.3.so
0099200000993000 rp 00014000 03:05 1158767 /lib/ld2.3.3.so
0099300000994000 rwp 00015000 03:05 1158767 /lib/ld2.3.3.so
0099600000aac000 rxp 00000000 03:05 1158770 /lib/tls/libc2.3.3.so
00aac00000aad000 rp 00116000 03:05 1158770 /lib/tls/libc2.3.3.so
00aad00000ab0000 rwp 00117000 03:05 1158770 /lib/tls/libc2.3.3.so
00ab000000ab2000 rwp 00ab0000 00:00 0
0804800008050000 rxp 00000000 03:05 571452 /sbin/init (text)
0805000008051000 rwp 00008000 03:05 571452 /sbin/init (data, stack)
08b4300008b64000 rwp 08b43000 00:00 0
f6fdf000f6fe0000 rwp f6fdf000 00:00 0
fefd4000ff000000 rwp fefd4000 00:00 0
ffffe000fffff000 p 00000000 00:00 0
mmap() (2)
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!
X server example (maps excerpt)
start end perm offset major:minor inode mapped file name
08047000081be000 rxp 00000000 03:05 310295 /usr/X11R6/bin/Xorg
081be000081f0000 rwp 00176000 03:05 310295 /usr/X11R6/bin/Xorg
...
f4e08000f4f09000 rws e0000000 03:05 655295 /dev/dri/card0
f4f09000f4f0b000 rws 4281a000 03:05 655295 /dev/dri/card0
f4f0b000f6f0b000 rws e8000000 03:05 652822 /dev/mem
f6f0b000f6f8b000 rws fcff0000 03:05 652822 /dev/mem
A more user friendly way to get such information: pmap <pid>
How to Implement mmap() UserSpace
Open the device file
Call the mmap system call (see man mmap for details):
void *mmap(
void *start, /* Often 0, preferred starting address */
size_t length, /* Length of the mapped area */
int prot , /* Permissions: read, write, execute */
int flags, /* Options: shared mapping, private copy... */
int fd, /* Open file descriptor */
off_t offset /* Offset in the file */
);
Read from the return virtual address or write to it.
How to Implement mmap() KernelSpace
Character driver: implement a mmap file operation
and add it to the driver file operations:
int (*mmap)(
struct file *, /* Open file structure */
struct vm_area_struct /* Kernel VMA structure */
);
Initialize the mapping.
Can be done in most cases with the remap_pfn_range()
function, which takes care of most of the job.
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 /* Page permissions */
);
PFN: Page Frame Number, the number of the page (0, 1, 2, ...).
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;
}
Driver Development
Debugging
Usefulness of a Serial Port
Most processors feature a serial port interface (usually very
well supported by Linux). Just need this interface to be
connected to the outside.
Easy way of getting the first messages of an early kernel
version, even before it boots. A minimum kernel with only
serial port support is enough.
Once the kernel is fixed and has completed booting, possible
to access a serial console and issue commands.
The serial port can also be used to transfer files to the target.
Debugging With printk()
Universal debugging technique used since the beginning of
programming (first found in cavemen drawings).
Printed or not in the console or /var/log/messages
according to the priority. This is controlled by the loglevel
kernel parameter, or through /proc/sys/kernel/printk
(see Documentation/sysctl/kernel.txt).
Available priorities (include/linux/kernel.h):
#define KERN_EMERG "<0>" /* system is unusable */
#define KERN_ALERT "<1>" /* action must be taken immediately */
#define KERN_CRIT "<2>" /* critical conditions */
#define KERN_ERR "<3>" /* error conditions */
#define KERN_WARNING "<4>" /* warning conditions */
#define KERN_NOTICE "<5>" /* normal but significant condition */
#define KERN_INFO "<6>" /* informational */
#define KERN_DEBUG "<7>" /* debuglevel messages */
Debugging With /proc or /sys
Instead of dumping messages in the kernel log, you can have your
drivers make information available to userspace:
Through a file in /proc or /sys, which contents are handled by
callbacks defined and registered by your driver.
Can be used to show any piece of information
about your device or driver.
Can also be used to send data to the driver or to control it.
Caution: anybody can use these files.
You should remove your debugging interface in production!
kgdb Kernel Patch
http://kgdb.linsyssoft.com/
The execution of the patched kernel is fully controlled by
gdb from another machine, connected through a serial line.
Can do almost everything, including inserting breakpoints in
interrupt handlers.
Supported architectures: i386, x86_64, ppc and s390.
Kernel Oops
Caused by an exception in the kernel code itself.
The kernel issues a diagnostic messaged, called an Oops!. The message
contains debug information, such as register content, stack trace, code
dump etc.
After the message is sent to the log and console
From within a task – the kernel kills the task.
From interrupt – the kernel panics and may reboot automatically if
given apriori the panic=secs parameter.
In 2.4, stack trace addresses can be translated via the ksymoops utility.
In 2.6, with the CONFIG_KALLSYMS option on, the kernel does
translation internally.
Kernel Oops Example
Unable to handle kernel NULL pointer dereference at virtual address
00000014
printing eip: d8d4d545
*pde = 00000000
Oops: 0000
CPU: 0
EIP: 0010:[<d8d4d545>] Tainted: PF
EFLAGS: 00013286
eax: c8b078c0 ebx: c8b078c0 ecx: 00000019 edx: c8b078c0
esi: 00000000 edi: 00000000 ebp: bffff75c esp: c403bf84
ds: 0018 es: 0018 ss: 0018
Process vmware (pid: 5194, stackpage=c403b000)
Stack: c8b078c0 00000000 bffff35c 00000000 c0136d64 c403bfa4 c8b078c0
c0130185
Call Trace: [sys_stat64+100/112] [filp_close+53/112]
[sys_close+67/96] [system_call+51/56]

Code: f6 46 14 01 74 1c 83 c4 f4 8b 06 50 e8 da 62 43 e7 83 c4 f8
Driver Development
Concurrent Access to Resources
Sources of Concurrency Issues
The same resources can be accessed by several kernel processes in
parallel, causing potential concurrency issues
Several userspace programs accessing the same device data or
hardware. Several kernel processes could execute the same code on
behalf of user processes running in parallel.
Multi processor (SMP): the same driver code can be running on
another processor. This can also happen with single CPUs with
hyperthreading.
Kernel preemption, interrupts: kernel code can be interrupted at any
time (just a few exceptions), and the same data may be access by another
process before the execution continues.
Avoiding Concurrency Issues
Avoid using global variables and shared data whenever possible
(cannot be done with hardware resources)
Don't make resources available to other kernel processes until
they are ready to be used.
Use techniques to manage concurrent access to resources.
See Rusty Russell's Unreliable Guide To Locking
Documentation/DocBook/kernellocking/
in the kernel sources.
Concurrency Protection With Semaphores
Process 1 Process 2
Failed
Acquire lock Wait lock release
Try again
Success Critical code section Success
Shared resource
Release lock
Kernel Semaphores
Also called “mutexes” (Mutual Exclusion)
V: “Verhoog”
1 “Increment” in Dutch
(free)
P V
(down) (up)
0
P: “Probeer” (locked)
“Try” (to decrement) in Dutch
Initializing a Semaphore
Statically
DECLARE_MUTEX(name);
DECLARE_MUTEX_LOCKED(name);
Dynamically
void init_MUTEX(struct semaphore *sem);
void init_MUTEX_LOCKED(struct semaphore *sem);
Locking and Unlocking Semaphores
void down(struct semaphore *sem);
Decrements the semaphore if set to 1, waits otherwise.
Caution: can't be interrupted, causing processes you cannot kill!
int down_interruptible(struct semaphore *sem);
Same, but can be interrupted. If interrupted, returns a non zero value
and doesn't hold the semaphore. Test the return value!!!
int down_trylock(struct semaphore *sem);
Never waits. Returns a non zero value if the semaphore is not
available.
void up(struct semaphore *sem);
Releases the semaphore. Make sure you do it as soon as possible!
Reader/Writer Semaphores
Allow shared access to unlimited readers, or to only one writer. Writers get
priority.
void init_rwsem(struct rw_semaphore *sem);
void down_read(struct rw_semaphore *sem);
int down_read_trylock(struct rw_semaphore *sem);
int up_read(struct rw_semaphore *sem);
void down_write(struct rw_semaphore *sem);
int down_write_trylock(struct rw_semaphore *sem);
int up_write(struct rw_semaphore *sem);
Well suited for rare writes, holding the semaphore briefly. Otherwise, readers get
starved, waiting too long for the semaphore to be released.
Spinlocks
Locks to be used for code that can't sleep (critical sections,
interrupt handlers... Be very careful not to call functions which
can sleep!
Still locked?
Intended for multiprocessor systems
Spinlock
Spinlocks are not interruptible,
don't sleep and keep spinning in a loop
until the lock is available.
Spinlocks cause kernel preemption to be disabled on the CPU
executing them, similar to VxWorks's task lock.
May require interrupts to be disabled too.
Initializing Spinlocks
Static
spinlock_t my_lock = SPIN_LOCK_UNLOCKED;
Dynamic
void spin_lock_init(spinlock_t *lock);
Using Spinlocks
void spin_[un]lock(spin_lock_t *lock);
void spin_[un]lock_irq[save|restore](spin_lock_t
*lock, unsigned long flags);
Disables IRQs on the local CPU
void spin_[un]lock_irq(spin_lock_t *lock);
Disables IRQs without saving flags. When you're sure that nobody
already disabled interrupts.
void spin_[un]lock_bh(spin_lock_t *lock);
Disables software interrupts, but not hardware ones
Note that reader/writer spinlocks also exist.
Deadlock Situations
They can lock up your system. Make sure they never happen!
Don't call a function that can try Holding multiple locks is risky!
to get access to the same lock
Get lock1 Get lock2
Get lock1
call
Dead
Get lock2 Lock! Get lock1
Wait for lock1
Dead
Lock!
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.
By using lockfree algorithms like Read Copy Update (RCU).
RCU API available in the kernel
(See http://en.wikipedia.org/wiki/RCU).
When available, use atomic operations.
Atomic Variables
Useful when the shared resource is an
Operations without return value:
integer value void atomic_inc(atomic_t *v);
void atomic_dec(atomic_ *v);
Even an instruction like n++ is not void atomic_add(int i, atomic_t *v);
guaranteed to be atomic on all processors! void atomic_sub(int i, atomic_t *v);
Header Similar functions testing the result:
int atomic_inc_and_test(...);
#include <asm/atomic.h> int atomic_dec_and_test(...);
int atomic_sub_and_test(...);
Type
Functions returning the new value:
atomic_t int atomic_inc_and_return(...);
int atomic_dec_and_return(...);
contains a signed integer (use 24 bits only) int atomic_add_and_return(...);
int atomic_sub_and_return(...);
Atomic operations (main ones)
Set or read the counter:
atomic_set(atomic_t *v, int i);
int atomic_read(atomic_t *v);
Atomic Bit Operations
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:
int test_bit(int nr, unsigned long *addr);
Test and modify (return the previous value):
int test_and_set_bit(...);
int test_and_clear_bit(...);
int test_and_change_bit(...);
Driver Development
Processes and Scheduling
Processes and Threads – a Reminder
A process is an instance of a running program.
Multiple instances of the same program can be running.
Program code (“text section”) memory is shared.
Each process has its own data section, address space, open files
and signal handlers.
A thread is a single task in a program.
It belongs to a process and shares the common data section,
address space, open files and pending signals.
It has its own stack, pending signals and state.
It's common to refer to single threaded programs as processes.
The Kernel and Threads
In 2.6 an explicit notion of processes and threads was
introduced to the kernel.
Scheduling is done on a thread by thread basis.
The basic object the kernel works with is a task, which is
analogous to a thread.
task_struct
Each task is represented by a task_struct.
The task is linked in the task tree via:
parent Pointer to its parent
children A linked list
sibling A linked list
task_struct contains many fields:
comm: name of task
priority, rt_priority: nice and realtime priorities
uid, euid, gid, egid: task's security credentials
Task Identifiers
Each task_struct has the following identities:
PID Globally unique ID. Different one for each thread.
TGID Thread Group ID. Returned to userspace as getpid()
Shared by all threads of a process.
For single threaded process == PID.
PGID Process Group ID (Posix.1).
SID Session ID (Posix.1).
Current Task
current points to the current process task_struct
When applicable – not valid in interrupt context.
Current is a macro that appears to the programmer as a
magical global variable which updated each context switch.
Real value is either in register or computed from start of stack
register value.
On SMP machine current will point to different structure on
each CPU.
A Process Life
TASK_ZOMBIE
Parent process Task terminated but its
Calls fork() resources are not freed yet.
and creates The process is elected
Waiting for its parent
a new process by the scheduler to acknowledge its death.
TASK_RUNNING TASK_RUNNING
Ready but
Actually running
not running
The process is preempted
by to scheduler to run
a higher priority task
Decides to sleep
The event occurs
TASK_INTERRUPTIBLE on a wait queue
or the process receives
or TASK_UNINTERRUPTIBLE for a specific event
a signal. Process becomes Waiting
runnable again
Process Context
Userspace programs and system calls are scheduled together:
Process continuing in user space...
Process executing in user space...
(or replaced by a higher priority process)
(can be preempted)
(can be preempted)
System call
or exception
Kernel code executed
Still has access to process
on behalf of user space
data (open files...)
(can be preempted too!)
Kernel Threads
The kernel does not only react to userspace programs (system calls,
exceptions) or hardware events (interrupts). It also runs its own
“processes”.
Kernel threads are standard processes scheduled and preempted in the same
manner as any other processes (you can view them with top or ps!). They
just have no special address space and usually run forever.
Kernel thread examples:
pdflush: regularly flushes “dirty” memory pages to disk (file
changes not committed to disk yet).
ksoftirqd: manages soft irqs.

Timer Frequency
Timer interrupts are raised every HZ th of second (= 1 jiffy)
HZ is now configurable (in Processor type and features):
100 (i386 default), 250 or 1000.
Supported on i386, ia64, ppc, ppc64, sparc64, x86_64
See kernel/Kconfig.hz.
Compromise between system responsiveness and global throughput.
Caution: not any value can be used. Constraints apply!
The Effect of Timer Frequency
Requested sleep time, in ms
20 HZ=100
10
10 15 20
Real sleep time, in ms
The Effect of Timer Frequency cont.
Requested sleep time, in ms
20 HZ=1000
10
10 15 20
Real sleep time, in ms
HighRes Timers and Tickless Kernel (2.6.20)
The highres timers feature enables POSIX timers and
nanosleep() to be as accurate as the hardware allows (around
1usec on typical hardware) by using non RTC interrupt timer
sources if supported by hardware.
This feature is transparent if enabled it just makes these
timers much more accurate than the current HZ resolution.
The tickless kernel feature enables 'ondemand' timer
interrupts.
On x86 test boxes the measured effective IRQ rate drops to to
12 timer interrupts per second.
Both require outside kernel patches.
O(1) Scheduler
The O(1) scheduler was the algorithm used from the
beginning of the 2.6 kernel up till (and including) 2.6.22.
The kernel maintains 2 priority arrays:
the active and the expired array.
Each array contains 140 entries (100 realtime priorities + 40
regular ones), 1 for each priority, each containing a list of
processes with the same priority.
The arrays are implemented in a way that makes it possible
to pick a process with the highest priority in constant time
(regardless of the number of running processes).
Choosing and Expiring Processes
The scheduler finds the highest process priority
It executes the first process in the priority queue for this priority.
If the process is a non real time one, once the process has
exhausted its time slice, it is moved to the expired array.
Otherwise, it's run until it yields the CPU[*].
The scheduler gets back to selecting another process with the
highest priority available, and so on...
Once the active array is empty, the 2 arrays are swapped!
Again, everything is done in constant time!
[*] SCHED_RR tasks will go to the back of the list for a specific priority when their time slice expires.
When is Scheduling Run?
Each process has a need_resched flag which is set:
After a process exhausted its time slice.
After a process with a higher priority is awakened.
This flag is checked (possibly causing the execution of the scheduler):
When returning to userspace from a system call.
When returning from an interrupt handler (including the CPU timer).
Scheduling also happens when kernel code explicitly calls
schedule() or executes an action that sleeps.
Time Slices
The scheduler also prioritizes high priority (but nonrealtime)
processes by giving them a bigger time slice.
Initial process time slice: parent's time slice split in 2
(otherwise process would cheat by forking).
Minimum priority: 5 ms or 1 jiffy (whichever is larger)
Default priority in jiffies: 100 ms
Maximum priority: 800 ms
Note: actually depends on HZ.
See kernel/sched.c for details.
Dynamic Priorities
Only applies to regular processes.
For a better user experience, the Linux scheduler boots the priority
of interactive processes (processes which spend most of their time
sleeping, and take time to exhaust their time slices). Such
processes often sleep but need to respond quickly after waking up
(example: word processor waiting for key presses).
Priority bonus: up to 5 points.
Conversely, the Linux scheduler reduces the priority of compute
intensive tasks (which quickly exhaust their time slices).
Priority penalty: up to 5 points.
Problems With the O(1) Scheduler
The O(1) scheduler uses heuristics for deciding when to
move a process to the expired queue.
It is sometimes wrong, and is also subject to a number of
attacks which can “fool” it and get more CPU cycles
(fiftyp.c, thud.c, chew.c, ringtest.c, massive_intr.c).
Also as a result of the heuristics, the code is complicated –
hard to understand and to maintain.
Therefore, the socalled Completely Fair Scheduler (CFS)
was introduced by Red Hat's Ingo Molnar in kernel 2.6.23.
For more information see the annexes.
Driver Development
Sleeping
How to Sleep (1)
Sleeping is needed when a user process is waiting for data which
are not ready yet. The process then puts itself in a waiting queue.
#include <linux/wait.h>
Static queue declaration:
DECLARE_WAIT_QUEUE_HEAD(module_queue);
Dynamic queue declaration:
wait_queue_head_t queue;
init_waitqueue_head(&queue);
How to Sleep (2)
There are several ways to make a kernel process sleep:
wait_event(queue, condition);
Sleeps until the given boolean expression is true.
Caution: can't be interrupted (i.e. by killing the client process in userspace)
wait_event_interruptible(queue, condition);
Can be interrupted
wait_event_timeout(queue, condition, timeout);
Sleeps and automatically wakes up after the given timeout.
wait_event_interruptible_timeout(queue, condition, timeout);
Same as above, interruptible.
Waking Up!
Typically done by interrupt handlers when data sleeping
processes are waiting for are available.
wake_up(&queue);
Wakes up all the waiting processes on the given queue
wake_up_interruptible(&queue);
Does the same job. Usually called when processes waited
using wait_event_interruptible().
Atomically Going to Sleep
We must have
cond to support
the event
happening before
We must have a If (!cond) going to sleep.
queue so we
can put the task
on it before sleep();
checking the
condition to
This is a naïve implementation of
avoid race.
wait_for_event() used to explain
why we need both queue and
cond.
Driver Development
Interrupt Management
Need for Interrupts
Internal processor interrupts used by the processor, for
example for multitask scheduling.
External interrupts needed because most internal and external
devices are slower than the processor. Better not keep the
processor waiting for input data to be ready or data to be
output. When the device is ready again, it sends an interrupt
to get the processor attention again.
Interrupt Handler Constraints
Not run from a user context:
Can't transfer data to and from user space
(need to be done by system call handlers)
Interrupt handler execution is managed by the CPU, not by
the scheduler. Handlers can't run actions that may sleep,
because there is nothing to resume their execution.
In particular, need to allocate memory with GFP_ATOMIC
Have to complete their job quickly enough:
they shouldn't block their interrupt line for too long.
Registering an Interrupt Handler (1)
Defined in include/linux/interrupt.h
int request_irq(
unsigned int irq, Requested irq channel
irqreturn_t (*handler) (...), Interrupt handler
unsigned long irq_flags, Option mask (see next page)
const char *devname, Registered name
void *dev_id); Pointer to some handler data
Cannot be NULL and must be unique for shared irqs!
void free_irq(unsigned int irq, void *dev_id);
Registering an Interrupt Handler (2)
irq_flags bit values (can be combined, none is fine too):
SA_INTERRUPT
"Quick" interrupt handler. Run with all interrupts disabled on the current CPU.
Shouldn't need to be used except in specific cases (such as timer interrupts)
SA_SHIRQ
Run with interrupts disabled only on the current IRQ line and on the local
CPU. The interrupt channel can be shared by several devices.
Requires a hardware status register telling whether an IRQ was raised or not.
SA_SAMPLE_RANDOM
Interrupts can be used to contribute to the system entropy pool used by
/dev/random and /dev/urandom. Useful to generate good random
numbers. Don't use this if the interrupt behavior of your device is predictable!
When to Register the Handler
Either at driver initialization time:
consumes lots of IRQ channels!
Or at device open time (first call to the open file operation):
better for saving free IRQ channels.
Need to count the number of times the device is opened, to
be able to free the IRQ channel when the device is no longer
in use.
Information on Installed Handlers
/proc/interrupts
CPU0
0: 5616905 XTPIC timer # Registered name
1: 9828 XTPIC i8042
2: 0 XTPIC cascade
3: 1014243 XTPIC orinoco_cs
7: 184 XTPIC Intel 82801DBICH4
8: 1 XTPIC rtc
9: 2 XTPIC acpi
11: 566583 XTPIC ehci_hcd, uhci_hcd
12: 5466 XTPIC i8042
14: 121043 XTPIC ide0
15: 200888 XTPIC ide1
NMI: 0 # Non Maskable Interrupts
ERR: 0
The Interrupt Handler's Job
Acknowledge the interrupt to the device
(otherwise no more interrupts will be generated).
Read/write data from/to the device.
Wake up any waiting process waiting for the completion of
this read/write operation:
wake_up_interruptible(&module_queue);
Interrupt Handler Prototype
irqreturn_t (*handler)(
int, /* irq number */
void *dev_id, /* Pointer used to keep track of the
corresponding device. Useful
when several devices are
managed by the same module */
struct pt_regs *regs /* CPU register snapshot, rarely
needed */
);
Return value:
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.
TopHalf and BottomHalf Processing (1)
Tophalf and bottomhalf is a logical paradigm to divide the tasks
that needs to be performed when an interrupt occurs:
Tophalf: the interrupt handler must complete as quickly as
possible. Once it acknowledged the interrupt, it just schedules the
lengthy rest of the job taking care of the data, for a later execution.
Bottomhalf: completing the rest of the interrupt handling job in a
different context. Handles data, and then wakes up any waiting
user process.
Can be implemented by tasklets, timers or work queues or user
space tasks.
Softirq
A fixed set (max 32) of software interrupts (prioritized):
HI_SOFTIRQ Runs low latency tasklets
TIMER_SOFTIRQ Runs timers
NET_TX_SOFTIRQ Network stack Tx
NET_RX_SOFTIRQ Network stack Rx
SCSI_SOFTIRQ SCSI sub system
TASKLET_SOFTIRQ Runs normal tasklets
Can run concurrently on SMP systems (even the same softirq).
Linux Contexts
Interrupt Handlers
Interrupt
SoftIRQs
Context
Kernel
Regular
tasklets
tasklets
Hi prio
Timers
Stack
Space
Net
...
Process
Scheduling Process Context Thread
Points
Kernel
User Space Thread
Handling Floods
Normally, pending softirqs will be run after each interrupt.
A pending softirq is marked in a special bit field.
The function that handles this is called do_softirq() and it is
called by do_IRQ() function.
If after do_softirq() called the handler for that softirq, the
softirq is still pending (the bit is on), it will not call the
softirq again.
Instead, a low priority kernel thread, called ksoftirqd, is
woken up. It will execute the softirq handler when it is next
scheduled.
Tasklets
Are run from softirqs (normal or lowlatency)
Each tasklet runs only on a single CPU (serialization)
You can initialize a tasklet via:
init tasklet_init(struct tasklet_struct *t
void (*func)(unsigned long),
unsigned long data));
Or declare the tasklet in the module source file:
DECLARE_TASKLET(module_tasklet, /* name */
module_do_tasklet, /* function */
0 /* data */);
Tasklet Scheduling and Killing
Scheduling a tasklet in the top half part (interrupt handler):
For regular tasklets:
tasklet_schedule(&module_tasklet);
Or for low latency tasklets (runs first):
tasklet_hi_schedule().
If this tasklet was already scheduled – it is run only once.
If this tasklet was already running – it is rescheduled for later.
On module exit, the tasklet should be killed:
tasklet_kill(&module_tasklet);
Tasklet Masking
Tasklets may be temporarily disabled/enabled:
tasklet_enable(&module_tasklet);
tasklet_disable(&module_tasklet);
tasklet_hi_enable(&module_tasklet);
tasklet_disable_nosync(&module_tasklet);
Timers
Run via softirq like tasklets...
But at a specific time.
A timer is represented by a timer_list structure:
struct timer_list {
/* ... */
unsigned long expires; /* In Jiffies */
void (*function )(unsigned int);
unsigned long data; /* Optional */
);
Timer Operations
Manipulated with:
void init_timer(struct timer_list *timer);
void add_timer(struct timer_list *timer);
void init_timer_on(struct timer_list *timer, int cpu);
void del_timer(struct timer_list *timer);
void del_timer_sync(struct timer_list *timer);
void mod_timer(struct timer_list *timer, unsigned long
expires);
void timer_pending(const struct timer_list *timer);
Work Queues (2.6 only)
Each work queue has a kernel thread (task) per CPU.
Since 2.6.6 also a single threaded version exists.
Code in a work queue:
Has a process context.
May sleep.
New work queues may be created/destroyed via:
struct workqueue_struct *create_workqueue(const
char *name);
struct workqueue_struct
*create_singlethread_workqueue(const char *name);
void destroy_workqueue(const char *name);
Working the Work Queue
Work is delivered to a work queue via:
DECLARE_WORK(work, func, data);
INIT_WORK(work, func, data);
int queue_work(struct workqueue_struct *wq, struct
work_struct *work);
int queue_delayed_work(struct workqueue_struct *wq,
struct work_struct *work, unsigned long delay);
int flush_workqueue(struct workqueue_struct *wq);
The Default Work Queue
One “default” work queue is run by the events kernel thread
(also known as the keventd_wq in the sources).
For the events work queue, we have the more common:
int schedule_work(struct work_struct *work);
int schedule_delayed_work(struct work_struct
*work, unsigned long delay);
int cancel_delayed_work(struct work_struct *work);
int flush_scheduled_work(void);
int current_is_keventd(void);
Disabling Interrupts
May be useful in regular driver code...
Can be useful to ensure that an interrupt handler will not preempt your
code (this includes kernel preemption).
Disabling interrupts on the local CPU:
unsigned long flags;
local_irq_save(flags); // Interrupts disabled
...
local_irq_restore(flags); // Interrupts restored to their previous state.
Note: must be run from within the same function!
Masking Out an Interrupt Line
Useful to disable interrupts on a particular device.
void disable_irq(unsigned int irq);
Disables the irq line for all processors in the system.
Waits for all currently executing handlers to complete.
void disable_irq_nosync(unsigned int irq);
Same, except it doesn't wait for handlers to complete.
void enable_irq(unsigned int irq);
Restores interrupts on the irq line.
void synchronize_irq(unsigned int irq);
Waits for irq handlers to complete (if any).
Checking Interrupt Status
Can be useful for code which can be run from both process or
interrupt context, to know whether it is allowed or not to call
code that may sleep.
irqs_disabled()
Tests whether local interrupt delivery is disabled.
in_interrupt()
Tests whether code is running in interrupt context.
in_irq()
Tests whether code is running in an interrupt handler.
Interrupt Management Summary
Device driver Tasklet/Work Queue
When the device file is first open, Process the data.
register an interrupt handler for the
Wake up processes waiting for
device's interrupt channel.
the data.
Interrupt handler
Device driver
Called when an interrupt is raised.
When the device is no longer
Acknowledge the interrupt. opened by any process,
unregister the interrupt handler.
If needed, schedule a tasklet or work
queue taking care of handling data.
Otherwise, wake up processes waiting
for the data.
The Network Subsystem
and Network Device Drivers
A View of the Linux Networking Subsystem
Stack <> App App 1 App2 App3
Socket Layer
UDP TCP ICMP
Networking Stack IP
Bridge
Networking Stack
Driver <> Stack
Driver
Driver <> Hardware
Network Device Driver Hardware Interface
xxx xxx xxx xxx xxx
Memory
Tx Access
Send Send Send SentOK SendErr Free
Driver
Rx
Free Free RcvOk RcvErr RecvCRC RcvOK
xxx xxx xxx xxx

●Driver allocates Ring Buffers.
DMA ●Driver resets descriptors to initial state.
●Driver puts packet to be sent in Tx buffers.
●Device puts received packet in Rx buffers.
●Driver/Device update descriptors to indicate state.
●Usually, device indicates Rx and end of Tx with
interrupt, unless polling or NAPI.
Socket Buffers
We need to manipulate packets through the network stack.
This manipulation involves efficiently:
Adding protocol headers/trailers down the stack.
Removing protocol headers/trailers up the stack.
Packets can be chained together.
Each protocol should have convenient access to header
fields.
To do all this the kernel provides the sk_buff structure.
struct sk_buff
An sk_buff represents a single packet.
This structure is passed through the protocol stack.
It holds pointers to a buffer with the packet data:
sk_buff
mac end
h
head data tail
nh
headroom mac ip tcp telnet tailroom
sk_buff Manipulation (1)
Manipulate an sk_buff:
unsigned char *skb_push(struct sk_buff *skb,
unsigned int len);
data = len
unsigned char *skb_pull(struct sk_buff *skb,
unsigned int len);
data += len
sk_buff Manipulation (2)
Manipulate an sk_buff:
int skb_headroom(const struct sk_buff *skb);
data head
int skb_tailroom(const struct sk_buff *skb);
end tail
int skb_reserve(const struct sk_buff *skb,
unsigned int len);
tail = (data +=len)
sk_buff Allocation
Low level allocation is done via:
struct sk_buff *alloc_skb(unsigned int
size, int gfp_mask);
But it is better to use the wrapper:
struct sk_buff *dev_alloc_skb(unsigned int
size);
Which reserves some space for optimization.
sk_buff Allocation Example
Immediately after allocation, we should reserve the needed
headroom:
struct sk_buff *skb;
skb = dev_alloc_skb(1500);
if (unlikely(!skb))
break;
/* Mark as being used by this device */
skb>dev = dev;
/* Align IP on 16 byte boundaries */
skb_reserve(skb, 2);
Softnet
Was introduced in kernel 2.4.x.
Parallelize packet handling on SMP machines.
Packet transmit/receive is handled via two softirqs:
NET_TX_SOFTIRQ feeds packets from the network stack to
the driver.
NET_RX_SOFTIRQ feeds packets from the driver to the
network stack.
Like any other soiftirq, these are called on return from
interrupt or via the low priority ksoftirqd kernel thread.
Transmit/receive queues are stored in perCPU softnet_data.
Packet Reception
The driver:
Allocates an skb.
sets up a descriptor in the ring buffers for the hardware.
The driver Rx interrupt handler calls netif_rx(skb).
netif_rx(skb)
Deposits the sk_buff in the perCPU input queue.
Marks the NET_RX_SOFTIRQ to run.
Later, net_rx_action() is called by NET_RX_SOFTIEQ, which
calls the driver's poll() method to feed the packet up.
Normally poll() is set to process_backlog() by net_dev_init().
Packet Reception Overview
Packet Transmission
Each network device defines a method:
int (*hard_start_xmit)(struct sk_buff *skb, struct
net_device *dev);
This method is indirectly called from NET_TX_SOFTIRQ
Calls to this method are serialized via dev>xmit_lock_owner
The driver can manage the transmit queue:
void netif_start_queue(struct net_device *net);
void netif_stop_queue(struct net_device *net);
void netif_wake_queue(struct net_device *net);
int netif_queue_stopped(struct net_device *net);
Packet Transmission Overview
Network Device Allocation
Each network device is represented by a struct net_device.
These structures are allocated using:
struct net_device *alloc_netdev(size, mask,
setup_func);
size – size of our priv data part
mask – a naming pattern (e.g. “eth%d”)
setup_func – A function to prepare the rest of net_device.
And deallocated with:
void free_netdev(struct *net_device);
Network Device Allocation (cont.)
For Ethernet device drivers, we have a short version:
struct net_device *alloc_etherdev(size);
which calls alloc_netdev(size, “eth%d”, ether_setup);
Network Device Registration
A network device driver provides interface to the network
stack.
It does not have or use major/minor numbers, like character
devices and has no /dev file.
A network driver is represented by a struct net_device.
The structure is registered with the kernel via:
int register_netdev(struct net_device *dev);
int unregister_netdev(struct net_device *dev);
After filling in some important bits...
Network Device Initialization
The net_device should be filled with numerous methods:
open() – request resources, register interrupts, start queues.
stop() – deallocates resources, unregister irq, stop queue.
get_stats() – report statistics.
set_multicast_list() – configure device for multicast.
do_ioctl() – device specific IOCTL function.
change_mtu() – Control device MTU setting.
hard_start_xmit() – called by the stack to initiate Tx.
Network Device Initialization (Cont.)
Also, the dev>flags should be set according to device
capabilities:
IFF_MULTICAST – device support multicast.
IFF_NOARP – device does not support the ARP protocol.
NAPI
Network “New API”.
Optional – provides interrupt mitigation under high load.
Requirements:
A DMA ring buffer.
Ability to turn off receive interrupts.
It is used by defining a new method:
int (*poll) (struct net_device *dev, int *budget);
Called by the network stack periodically when signaled by
the driver to do so.
NAPI (cont.)
When a receive interrupt occurs, driver:
Turns off receive interrupts.
Calls netif_rx_schedule(dev) to get stack to start
calling its poll method.
Poll method:
Scans receive ring buffers, feeding packets to the stack via:
netif_receive_skb(skb).
If work finished within the budget parameter, reenables
interrupts and calls netif_rx_complete(dev).
Else, the stack will call the poll() method again.
Embedded Linux driver development
Advice and Resources
Getting Help and Contributions
Information Sites (1)
Linux Weekly News
http://lwn.net/
The weekly digest off all Linux and free software
information sources.
Indepth technical discussions about the kernel.
Subscribe to finance the editors ($5 / month).
Articles available for nonsubscribers
after 1 week.
Information Sites (2)
KernelTrap
http://kerneltrap.org/
Forum website for kernel developers.
News, articles, whitepapers, discussions, polls, interviews.
Perfect if a digest is not enough!
Useful Reading (1)
Linux Device Drivers, 3rd edition, Feb 2005
By Jonathan Corbet, Alessandro Rubini, Greg KroahHartman, O'Reilly
http://www.oreilly.com/catalog/linuxdrive3/
Freely available online!
Great companion to the printed book for easy electronic searches!
http://lwn.net/Kernel/LDD3/ (1 PDF file per chapter)
http://freeelectrons.com/community/kernel/ldd3/ (single PDF file)
A musthave book for Linux device driver writers!
Useful Reading (2)
Linux Kernel Development, 2nd Edition, Jan 2005
Robert Love, Novell Press
http://rlove.org/kernel_book/
A very synthetic and pleasant way to learn about kernel
subsystems (beyond the needs of device driver writers)
Understanding the Linux Kernel, 3rd edition, Nov 2005
Daniel P. Bovet, Marco Cesati, O'Reilly
http://oreilly.com/catalog/understandlk/
An extensive review of Linux kernel internals, covering Linux 2.6 at last.
Unfortunately, only covers the PC architecture.
Useful Reading (3)
Building Embedded Linux Systems, April 2003
Karim Yaghmour, O'Reilly
http://www.oreilly.com/catalog/belinuxsys/
Not very fresh, but doesn't depend too much on kernel versions.
See http://www.linuxdevices.com/articles/AT2969812114.html
for more embedded Linux books.
Useful Online Resources
Linux kernel mailing list FAQ
http://www.tux.org/lkml/
Complete Linux kernel FAQ.
Read this before asking a question to the mailing list.
Kernel Newbies
http://kernelnewbies.org/
Glossaries, articles, presentations, HOWTOs,
recommended reading, useful tools for people
getting familiar with Linux kernel or driver
development.
CE Linux Forum Resources
CE Linux Forum's Wiki
is full of useful resources for embedded systems developers:
Kernel patches not available in mainstream yet.
Many howto documents of all kinds.
Details about ongoing projects, such as reducing kernel size,
boot time, or power consumption.
Contributions are welcome!
http://tree.celinuxforum.org/CelfPubWiki
International Conferences (1)
Useful conferences featuring Linux kernel presentations
Ottawa Linux Symposium (July): http://linuxsymposium.org/
Right after the (private) kernel summit.
Lots of kernel topics. Many core kernel hackers still present.
Fosdem: http://fosdem.org (Brussels, February)
For developers. Kernel presentations from wellknown kernel hackers.
CE Linux Forum: http://celinuxforum.org/
Organizes several international technical conferences, in particular in
California (San Jose) and in Japan. Now open to non CELF members!
Very interesting kernel topics for embedded systems developers.
International Conferences (2)
linux.conf.au: http://conf.linux.org.au/ (Australia/New Zealand)
Features a few presentations by key kernel hackers.
Linux Kongress (Germany, September/October)
http://www.linuxkongress.org/
Lots of presentations on the kernel but very expensive registration fees.
Don't miss our free conference videos on
http://freeelectrons.com/community/videos/conferences/!
Use the Source, Luke!
Many resources and tricks on the Internet find you will, but
solutions to all technical issues only in the Source lie.
Thanks to LucasArts
Annexes
Additional Materials
For Your Reference
Annexes
More Shell Tips & Tricks
The head and tail Commands
head [<n>] <file>
Displays the first <n> lines (or 10 by default) of the given file.
Doesn't have to open the whole file to do this!
tail [<n>] <file>
Displays the last <n> lines (or 10 by default) of the given file.
No need to load the whole file in RAM! Very useful for huge files.
tail f <file> (follow)
Displays the last 10 lines of the given file and continues to display new lines
when they are appended to the file.
Very useful to follow the changes in a log file, for example.
Examples
head windows_bugs.txt
tail f outlook_vulnerabilities.txt
Environment Variables
Shells let the user define variables.
They can be reused in shell commands.
Convention: lower case names
You can also define environment variables: variables
that are also visible within scripts or executables called
from the shell.
Convention: upper case names.
env
Lists all defined environment variables and their value.
Shell Variables Examples
Shell variables (bash)
projdir=/home/marshall/coolstuff
ls la $projdir; cd $projdir
Environment variables (bash)
cd $HOME
export DEBUG=1
./find_extraterrestrial_life
(displays debug information if DEBUG is set)
Measuring Elapsed Time
time find_expensive_housing near
<...command output...>
real 0m2.304s (actual elapsed time)
user 0m0.449s (CPU time running program code)
sys 0m0.106s (CPU time running system calls)
real = user + sys + waiting
waiting = I/O waiting time + idle time (running other tasks)
The grep Command
grep <pattern> <files>
Scans the given files and displays the lines which match the given pattern.
grep error *.log
Displays all the lines containing error in the *.log files.
grep i error *.log
Same, but case insensitive.
grep ri error .
Same, but recursively in all the files in . and its subdirectories.
grep v info *.log
Outputs all the lines in the files except those containing info.
The sort Command
sort <file>
Sorts the lines in the given file in character order and
outputs them.
sort r <file>
Same, but in reverse order.
sort ru <file>
u: unique. Same, but just outputs identical lines once.
More possibilities described later!
Symbolic Links
A symbolic link is a special file which is just a reference to the
name of another one (file or directory):
Useful to reduce disk usage and complexity when 2 files have the
same content.
Example:
anakin_skywalker_biography > darth_vador_biography
How to identify symbolic links:
ls l displays > and the linked file name.
GNU ls displays links with a different color.
Creating Symbolic Links
To create a symbolic link (same order as in cp):
ln s file_name link_name
To create a link with to a file in another directory, with the
same name:
ln s ../README.txt
To create multiple links at once in a given directory:
ln s file1 file2 file3 ... dir
To remove a link:
rm link_name
Of course, this doesn't remove the linked file!
Hard Links
The default behavior for ln is to create hard links.
A hard link to a file is a regular file with exactly the same
physical contents.
While they still save space, hard links can't be distinguished
from the original files.
If you remove the original file, there is no impact on the hard
link contents.
The contents are removed when there are no more files (hard
links) to them.
Files Names and inodes
Makes hard and symbolic (soft) links easier to understand!
Users
File name interface
Soft link File Hard link

rm
rm
Inode Inode
interface
Filesystem
File Ownership
chown R sco /home/linux/src (R: recursive)
Makes user sco the new owner of all the files in
/home/linux/src.
chgrp R empire /home/askywalker
Makes empire the new group of everything in
/home/askywalker.
chown R borg:aliens usss_entreprise/
chown can be used to change the owner and group at the
same time.
File Access Rights
Use ls l to check file access rights
3 types of access rights 3 types of access levels
Read access (r) User (u): for the owner of the
file
Write access (w)
Group (g): each file also has a
Execute rights (x)
“group” attribute, corresponding
to a given list of users
Others (o): for all other users
Access Right Constraints
x without r is legal but is useless...
You have to be able to read a file to execute it.
Both r and x permissions needed for directories:
x to enter, r to list its contents.
You can't rename, remove, copy files in a directory if you don't
have w access to this directory.
If you have w access to a directory, you CAN remove a file even if
you don't have write access to this file (remember that a directory
is just a file describing a list of files). This even lets you modify
(remove + recreate) a file even without w access to it.
Access Rights Examples
rwrr
Readable and writable for file owner, only readable for others
rwr
Readable and writable for file owner, only readable for users
belonging to the file group.
drwx
Directory only accessible by its owner.
rx
File executable by others but neither by your friends nor by
yourself. Nice protections for a trap...
chmod: Changing Permissions
chmod <permissions> <files>
2 formats for permissions:
Octal format (abc):
a,b,c = r*4+w*2+x (r, w, x: booleans)
Example: chmod 644 <file>
(rw for u, r for g and o)
Or symbolic format. Easy to understand by examples:
chmod go+r: add read permissions to group and others.
chmod uw: remove write permissions from user.
chmod ax: (a: all) remove execute permission from all.
More chmod (1)
chmod R a+rX linux/
Makes linux and everything in it available to
everyone!
R: apply changes recursively.
X: x, but only for directories and files already executable
Very useful to open recursive access to directories,
without adding execution rights to all files.
More chmod (2)
chmod a+t /tmp
t: (sticky). Special permission for directories, allowing
only the directory and file owner to delete a file in a
directory.
Useful for directories with write access to anyone,
like /tmp.
Displayed by ls l with a t character.
Standard Output
More about command output.
All the commands outputting text on your terminal do it by
writing to their standard output.
Standard output can be written (redirected) to a file using the
> symbol
Standard output can be appended to an existing file using the
>> symbol
Standard Output Redirection Examples
ls ~saddam/* > ~gwb/weapons_mass_destruction.txt
cat obiwan_kenobi.txt > starwars_biographies.txt
cat han_solo.txt >> starwars_biographies.txt
echo “README: No such file or directory” > README
Useful way of creating a file without a text editor.
Nice Unix joke too in this case.
Standard Input
More about command input:
Lots of commands, when not given input arguments, can take their
input from standard input.
sort
sort takes its input from
windows
linux the standard input: in this case,
[Ctrl][D] what you type in the terminal
linux (ended by [Ctrl][D])
windows
sort < participants.txt
The standard input of sort is taken from the given file.
Pipes
Unix pipes are very useful to redirect the standard output of a
command to the standard input of another one.
Examples
cat *.log | grep i error | sort
grep ri error . | grep v “ignored” | sort u \
> serious_errors.log
cat /home/*/homework.txt | grep mark | more
This one of the most powerful features in Unix shells!
The tee Command
tee [a] file
The tee command can be used to send standard output
to the screen and to a file simultaneously.
make | tee build.log
Runs the make command and stores its output to
build.log.
make install | tee a build.log
Runs the make install command and appends its
output to build.log.
Standard Error
Error messages are usually output (if the program is well written)
to standard error instead of standard output.
Standard error can be redirected through 2> or 2>>
Example:
cat f1 f2 nofile > newfile 2> errfile
Note: 1 is the descriptor for standard output, so 1> is equivalent
to >.
Can redirect both standard output and standard error to the same
file using &> :
cat f1 f2 nofile &> wholefile
Special Devices (1)
Device files with a special behavior or contents:
/dev/null
The data sink! Discards all data written to this file.
Useful to get rid of unwanted output, typically log information:
mplayer black_adder_4th.avi &> /dev/null
/dev/zero
Reads from this file always return \0 characters
Useful to create a file filled with zeros:
dd if=/dev/zero of=disk.img bs=1k count=2048
See man null or man zero for details.
Special Devices (2)
/dev/random
Returns random bytes when read. Mainly used by cryptographic
programs. Uses interrupts from some device drivers as sources of
true randomness (“entropy”).
Reads can be blocked until enough entropy is gathered.
/dev/urandom
For programs for which pseudo random numbers are fine.
Always generates random bytes, even if not enough entropy is
available (in which case it is possible, though still difficult, to
predict future byte sequences from past ones).
See man random for details.
Embedded Linux driver development
Annexes
Init Runlevels (Sys V Init)
System V Init Runlevels (1)
Introduced by System V Unix
Much more flexible than in BSD. /etc/initab excerpt:
Make it possible to start or stop id:5:initdefault:
different services for each # System initialization.
si::sysinit:/etc/rc.d/rc.sysinit
runlevel.
l0:0:wait:/etc/rc.d/rc 0
Correspond to the argument given l2:2:wait:/etc/rc.d/rc 2
to /sbin/init. l4:4:wait:/etc/rc.d/rc 4
Runlevels defined in
/etc/inittab.
System V Init Runlevels (2)
Standard levels: Customizable levels: 2, 3, 4, 5
init 0 init 3
Halt the system. Often multiuser mode, with only
init 1 commandline login.
Single user mode for maintenance. init 5
init 6 Often multiuser mode, with
Reboot the system. graphical login.
init S
Single user mode for maintenance.
Mounting only /. Often identical to 1
Init Scripts
According to /etc/inittab settings, init <n> runs:
First /etc/rc.d/rc.sysinit for all runlevels
Then scripts in /etc/rc<n>.d/
Starting services (1, 3, 5, S):
runs S* scripts with the start option.
Killing services (0, 6):
runs K* scripts with the stop option.
Scripts are run in file name lexical order
Just use ls l to find out the order!
/etc/init.d
Repository for all available init scripts.
/etc/rc<n>.d/ only contains links to the /etc/init.d/
scripts needed for runlevel n
/etc/rc1.d/ example (from Fedora Core 3):
K01yum > ../init.d/yum S00single > ../init.d/single
K02cupsconfigdaemon > ../init.d/cups S01sysstat > ../init.d/sysstat
configdaemon S06cpuspeed > ../init.d/cpuspeed
K02haldaemon > ../init.d/haldaemon
K02NetworkManager >
../init.d/NetworkManager
K03messagebus > ../init.d/messagebus
K03rhnsd > ../init.d/rhnsd
K05anacron > ../init.d/anacron
K05atd > ../init.d/atd
Handling Init Scripts by Hand
Simply call the /etc/init.d scripts!
/etc/init.d/sshd start
Starting sshd: [ OK ]
/etc/init.d/nfs stop
Shutting down NFS mountd: [FAILED]
Shutting down NFS daemon:
[FAILED]Shutting down NFS quotas:
[FAILED]
Shutting down NFS services: [ OK ]
/etc/init.d/pcmcia status
cardmgr (pid 3721) is running...
/etc/init.d/httpd restart
Stopping httpd: [ OK ]
Starting httpd: [ OK ]
Annexes
The Completely
Fair Scheduler
The Completely Fair Scheduler
The socalled Completely Fair Scheduler (CFS) was introduced by Red
Hat's Ingo Molnar in kernel 2.6.23.
As Ingo puts it: “CFS basically models an ideal, precise multitasking
CPU”.
CFS picks tasks to run according to the p>wait_runtime value, which
stands for the amount of time (in nanoseconds) that the task should
now run on the CPU for it to become completely fair and balanced.
CFS also uses the rq>fair_clock value to track the CPU time a
runnable task would have fairly gotten; For example, if there are four
tasks in the system, the fair clock will increase at ¼ speed of the actual
wall time.
The CFS Algorithm
The CFS holds all tasks in a redblack tree, sorted according
to “rq>fair_clock – p>wait_runtime”.
Therefore, the leftmost task in the tree (smallest value) is the
one which the scheduler should pick next.
This redblack tree algorithm is O(log n), which is a small
drawback, considering the previous scheduler was O(1).
Lab Information
Login:
User 'gby', password 'qwerty' or user 'root', password 'secret'
Running the Eclipse IDE:
/opt/course/scripts/eclipse &
Running the QEMU Emulator:
/opt/course/scripts/runqemu &
The emulated board files are at:
/opt/course/emulation/root
The kernel module template is at:
/opt/course/skeleton/
Copyrights and Trademarks
© Copyright 20062004, Michael Opdenacker
© Copyright 20032006, Oron Peled
© Copyright 20042008 Codefidence Ltd.
Portions of this text are reprinted and reproduced in electronic form from IEEE Std 1003.1,
2003 Edition, Standard for Information Technology Portable Operating System Interface
(POSIX), The Open Group Base Specifications Issue 6, Copyright (C) 20012003 by the
Institute of Electrical and Electronics Engineers, Inc and The Open Group. In the event of
any discrepancy between this version and the original IEEE and The Open Group Standard,
the original IEEE and The Open Group Standard is the referee document. The original
Standard can be obtained online at http://www.opengroup.org/unix/online.html .
Portions © Xavier Leroy <Xavier.Leroy@inria.fr>
Tux Image Copyright: (C) 1996 Larry Ewing
Linux is a registered trademark of Linus Torvalds.
All other trademarks are property of their respective owners.
Used and distributed under a Creative Commons AttributionShareAlike 2.0 license

Programming Embedded Linux

Uploaded by

Copyright:

Available Formats

Programming Embedded Linux

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Programming Embedded Linux

Uploaded by

Copyright:

Available Formats

Programming Embedded Linux ®

Kernel Shell RTOS

RTOS are like the Linux kernel:

Linux also has user tasks that run in their own

Linux Application API is ISO C and POSIX!

Both parent and

Child process gets

This call will pid_t pid;

Nice Real Time priority

Stack Stack Stack Stack Stack

State State State State State

Thread Thread Thread Thread

Process 123 Process 124

File Signal File Signal

Action Main Thread Other Thread

All threads All threads

2. High Priority task preempts low

x86 PC Arm PowerPC

External FS Initrd Style Initramfs style

action: one of sysinit, respawn,

Process Memory Filesystem Device

CPU support CPU / MMU Storage Character Network

CPU RAM Storage

do_name() Function call

Flash MMU CPU

Task Virtual Physical Permission

12 0x8000 0x5340 RWX

CPU MMU Memory

We need 8 Mb of memory, but

So we'll split the 16 Mb into two 8

We'll use 8 Mb and keep the rest

When the allocated memory has

We can once again combine the

Because of the order of 2

Requested sleep time, in ms

Requested sleep time, in ms

Stack <> App App 1 App2 App3

UDP TCP ICMP

Driver <> Hardware

xxx xxx xxx xxx

●Driver puts packet to be sent in Tx buffers.

●Device puts received packet in Rx buffers.

●Driver/Device update descriptors to indicate state.

●Usually, device indicates Rx and end of Tx with

interrupt, unless polling or NAPI.

headroom mac ip tcp telnet tailroom

Soft link File Hard link

You might also like