LDP
/
LDP
mirror of https://github.com/tLDP/LDP
0
1
Fork 0
Code Releases Activity
LDP/LDP/howto/linuxdoc/KernelAnalysis-HOWTO.sgml

4670 lines
117 KiB
Plaintext
Raw Permalink Blame History

<!doctype linuxdoc system>
<article>
<!-- LyX 1.3 created this file. For more info see http://www.lyx.org/ -->
<title>
KernelAnalysis-HOWTO
</title>
<author>
Roberto Arcomano berto@bertolinux.com
</author>
<date>
v0.7, March 26, 2003
</date>
<abstract>
This document tries to explain some things about the Linux Kernel,
such as the most important components, how they work, and so on.
This HOWTO should help prevent the reader from needing to browse
all the kernel source files searching for the"right function," declaration,
and definition, and then linking each to the other. You can find
the latest version of this document at <url url="http://www.bertolinux.com" name="http://www.bertolinux.com"> If you have suggestions to
help make this document better, please submit your ideas to me at
the following address: <url url="mailto:berto@bertolinux.com" name="berto@bertolinux.com">
</abstract>
<sect>
Introduction
<sect1>
Introduction
<p>
This HOWTO tries to define how parts of the<bf> </bf>Linux Kernel work,
what are the main functions and data structures used, and how the
"wheel spins". You can find the latest version of this document at
<url url="http://www.bertolinux.com" name="http://www.bertolinux.com"> If you have suggestions to help make this document better, please
submit your ideas to me at the following address: <url url="mailto:berto@bertolinux.com" name="berto@bertolinux.com">Code used within
this document refers to the Linux Kernel version 2.4.x, which is
the last stable kernel version at time of writing this HOWTO.
</p>
<sect1>
Copyright
<p>
Copyright (C) 2000,2001,2002 Roberto Arcomano. This document
is free; you can redistribute it and/or modify it under the terms
of the GNU General Public License as published by the Free Software
Foundation; either version 2 of the License, or (at your option)
any later version. This document is distributed in the hope that
it will be useful, but WITHOUT ANY WARRANTY; without even the implied
warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
See the GNU General Public License for more details. You can get
a copy of the GNU GPL <url url="http://www.gnu.org/copyleft/gpl.html" name="here">
</p>
<sect1>
Translations
<p>
If you want to translate this document you are free to do so.
However, you will need to do the following:
</p>
<p>
<enum>
<item>
Check that another version of the document doesn't already exist
at your local LDP
<item>
Maintain all 'Introduction' sections (including 'Introduction',
'Copyright', 'Translations' , 'Credits').
</enum>
</p><p>
Warning! You don't have to translate TXT or HTML file, you have
to modify LYX file, so that it is possible to convert it all other
formats (TXT, HTML, RIFF, etc.): to do that you can use "LyX" application
you download from <url url="http://www.lyx.org" name="http://www.lyx.org">.
</p>
<p>
No need to ask me to translate! You just have to let me know
(if you want) about your translation.
</p>
<p>
Thank you for your translation!
</p>
<sect1>
Credits
<p>
Thanks to <url url="http://www.tldp.org" name="Linux Documentation Project"> for publishing and uploading my document quickly.
</p>
<p>
Thanks to Klaas de Waal for his suggestions.
</p>
<sect>
Syntax used
<sect1>
Function Syntax
<p>
When speaking about a function, we write:
</p>
<p>
<verb>
&quot;function_name &lsqb; file location . extension &rsqb;&quot;
</verb>
</p><p>
For example:
</p>
<p>
<verb>
&quot;schedule &lsqb;kernel/sched.c&rsqb;&quot;
</verb>
</p><p>
tells us that we talk about
</p>
<p>
&quot;schedule&quot;
</p>
<p>
function retrievable from file
</p>
<p>
&lsqb; kernel/sched.c &rsqb;
</p>
<p>
Note: We also assume /usr/src/linux as the starting directory.
</p>
<sect1>
Indentation
<p>
Indentation in source code is 3 blank characters.
</p>
<sect1>
InterCallings Analysis
<sect2>
Overview
<p>
We use the"InterCallings Analysis "(ICA) to see (in an indented
fashion) how kernel functions call each other.
</p>
<p>
For example, the sleep_on command is described in ICA below:
</p>
<p>
<verb>
|sleep_on
|init_waitqueue_entry --
|__add_wait_queue | enqueuing request
|list_add |
|__list_add --
|schedule --- waiting for request to be executed
|__remove_wait_queue --
|list_del | dequeuing request
|__list_del --
sleep_on ICA
</verb>
</p><p>
The indented ICA is followed by functions' locations:
</p>
<p>
<itemize>
<item>
sleep_on &lsqb;kernel/sched.c&rsqb;
<item>
init_waitqueue_entry &lsqb;include/linux/wait.h&rsqb;
<item>
__add_wait_queue
<item>
list_add &lsqb;include/linux/list.h&rsqb;
<item>
__list_add
<item>
schedule &lsqb;kernel/sched.c&rsqb;
<item>
__remove_wait_queue &lsqb;include/linux/wait.h&rsqb;
<item>
list_del &lsqb;include/linux/list.h&rsqb;
<item>
__list_del
</itemize>
</p><p>
Note: We don't specify anymore file location, if specified just
before.
</p>
<sect2>
Details
<p>
In an ICA a line like looks like the following
</p>
<p>
<verb>
function1 -&gt; function2
</verb>
</p><p>
means that &lt; function1 &gt; is a generic pointer to another
function. In this case &lt; function1 &gt; points to &lt; function2
&gt;.
</p>
<p>
When we write:
</p>
<p>
<verb>
function:
</verb>
</p><p>
it means that &lt; function &gt; is not a real function. It is
a label (typically assembler label).
</p>
<p>
In many sections we may report a ''C'' code or a ''pseudo-code''.
In real source files, you could use ''assembler'' or ''not structured''
code. This difference is for learning purposes.
</p>
<sect2>
PROs of using ICA
<p>
The advantages of using ICA (InterCallings Analysis) are many:
</p>
<p>
<itemize>
<item>
You get an overview of what happens when you call a kernel function
<item>
Function locations are indicated after the function, so ICA could
also be considered as a little ''function reference''
<item>
InterCallings Analysis (ICA) is useful in sleep/awake mechanisms,
where we can view what we do before sleeping, the proper sleeping
action, and what we'll do after waking up (after schedule).
</itemize>
</p><sect2>
CONTROs of using ICA
<p>
<itemize>
<item>
Some of the disadvantages of using ICA are listed below:
</itemize>
</p><p>
As all theoretical models, we simplify reality avoiding many
details, such as real source code and special conditions.
</p>
<p>
<itemize>
<item>
Additional diagrams should be added to better represent stack
conditions, data values, and so on.
</itemize>
</p><sect>
Fundamentals
<sect1>
What is the kernel?
<p>
The kernel is the "core" of any computer system: it is the "software"
which allows users to share computer resources.
</p>
<p>
The kernel can be thought as the main software of the OS (Operating
System), which may also include graphics management.
</p>
<p>
For example, under Linux (like other Unix-like OSs), the XWindow
environment doesn't belong to the Linux Kernel, because it manages
only graphical operations (it uses user mode I/O to access video
card devices).
</p>
<p>
By contrast, Windows environments (Win9x, WinME, WinNT, Win2K,
WinXP, and so on) are a mix between a graphical environment and kernel.
</p>
<sect1>
What is the difference between User Mode and Kernel Mode?
<sect2>
Overview
<p>
Many years ago, when computers were as big as a room, users ran
their applications with much difficulty and, sometimes, their applications
crashed the computer.
</p>
<sect2>
Operative modes
<p>
To avoid having applications that constantly crashed, newer OSs
were designed with 2 different operative modes:
</p>
<p>
<enum>
<item>
Kernel Mode: the machine operates with critical data structure,
direct hardware (IN/OUT or memory mapped), direct memory, IRQ, DMA,
and so on.
<item>
User Mode: users can run applications.
</enum>
<p>
<verb>
| Applications /|&bsol;
| ______________ |
| | User Mode | |
| ______________ |
| | |
Implementation | _______ _______ | Abstraction
Detail | | Kernel Mode | |
| _______________ |
| | |
| | |
| | |
&bsol;|/ Hardware |
</verb>
</p><p>
Kernel Mode "prevents" User Mode applications from damaging the
system or its features.
</p>
<p>
Modern microprocessors implement in hardware at least 2 different
states. For example under Intel, 4 states determine the PL (Privilege
Level). It is possible to use 0,1,2,3 states, with 0 used in Kernel
Mode.
</p>
<p>
Unix OS requires only 2 privilege levels, and we will use such
a paradigm as point of reference.
</p>
<sect1>
Switching from User Mode to Kernel Mode
<sect2>
When do we switch?
<p>
Once we understand that there are 2 different modes, we have
to know when we switch from one to the other.
</p>
<p>
Typically, there are 2 points of switching:
</p>
<p>
<enum>
<item>
When calling a System Call: after calling a System Call, the
task voluntary calls pieces of code living in Kernel Mode
<item>
When an IRQ (or exception) comes: after the IRQ an IRQ handler
(or exception handler) is called, then control returns back to the
task that was interrupted like nothing was happened.
</enum>
</p><sect2>
System Calls
<p>
System calls are like special functions that manage OS routines
which live in Kernel Mode.
</p>
<p>
A system call can be called when we:
</p>
<p>
<itemize>
<item>
access an I/O device or a file (like read or write)
<item>
need to access privileged information (like pid, changing scheduling
policy or other information)
<item>
need to change execution context (like forking or executing some
other application)
<item>
need to execute a particular command (like ''chdir'', ''kill",
''brk'', or ''signal'')
</itemize>
<p>
<verb>
| |
-------&gt;| System Call i | (Accessing Devices)
| | | | &lsqb;sys_read()&rsqb; |
| ... | | | |
| system_call(i) |-------- | |
| &lsqb;read()&rsqb; | | |
| ... | | |
| system_call(j) |-------- | |
| &lsqb;get_pid()&rsqb; | | | |
| ... | -------&gt;| System Call j | (Accessing kernel data structures)
| | | &lsqb;sys_getpid()&rsqb;|
| |
USER MODE KERNEL MODE
Unix System Calls Working
</verb>
</p><p>
System calls are almost the only interface used by User Mode
to talk with low level resources (hardware). The only exception to
this statement is when a process uses ''ioperm'' system call. In
this case a device can be accessed directly by User Mode process
(IRQs cannot be used).
</p>
<p>
NOTE: Not every ''C'' function is a system call, only some of
them.
</p>
<p>
Below is a list of System Calls under Linux Kernel 2.4.17, from
&lsqb; arch/i386/kernel/entry.S &rsqb;
</p>
<p>
<verb>
.long SYMBOL_NAME(sys_ni_syscall) /* 0 - old &quot;setup()&quot; system call*/
.long SYMBOL_NAME(sys_exit)
.long SYMBOL_NAME(sys_fork)
.long SYMBOL_NAME(sys_read)
.long SYMBOL_NAME(sys_write)
.long SYMBOL_NAME(sys_open) /* 5 */
.long SYMBOL_NAME(sys_close)
.long SYMBOL_NAME(sys_waitpid)
.long SYMBOL_NAME(sys_creat)
.long SYMBOL_NAME(sys_link)
.long SYMBOL_NAME(sys_unlink) /* 10 */
.long SYMBOL_NAME(sys_execve)
.long SYMBOL_NAME(sys_chdir)
.long SYMBOL_NAME(sys_time)
.long SYMBOL_NAME(sys_mknod)
.long SYMBOL_NAME(sys_chmod) /* 15 */
.long SYMBOL_NAME(sys_lchown16)
.long SYMBOL_NAME(sys_ni_syscall) /* old break syscall holder */
.long SYMBOL_NAME(sys_stat)
.long SYMBOL_NAME(sys_lseek)
.long SYMBOL_NAME(sys_getpid) /* 20 */
.long SYMBOL_NAME(sys_mount)
.long SYMBOL_NAME(sys_oldumount)
.long SYMBOL_NAME(sys_setuid16)
.long SYMBOL_NAME(sys_getuid16)
.long SYMBOL_NAME(sys_stime) /* 25 */
.long SYMBOL_NAME(sys_ptrace)
.long SYMBOL_NAME(sys_alarm)
.long SYMBOL_NAME(sys_fstat)
.long SYMBOL_NAME(sys_pause)
.long SYMBOL_NAME(sys_utime) /* 30 */
.long SYMBOL_NAME(sys_ni_syscall) /* old stty syscall holder */
.long SYMBOL_NAME(sys_ni_syscall) /* old gtty syscall holder */
.long SYMBOL_NAME(sys_access)
.long SYMBOL_NAME(sys_nice)
.long SYMBOL_NAME(sys_ni_syscall) /* 35 */ /* old ftime syscall holder */
.long SYMBOL_NAME(sys_sync)
.long SYMBOL_NAME(sys_kill)
.long SYMBOL_NAME(sys_rename)
.long SYMBOL_NAME(sys_mkdir)
.long SYMBOL_NAME(sys_rmdir) /* 40 */
.long SYMBOL_NAME(sys_dup)
.long SYMBOL_NAME(sys_pipe)
.long SYMBOL_NAME(sys_times)
.long SYMBOL_NAME(sys_ni_syscall) /* old prof syscall holder */
.long SYMBOL_NAME(sys_brk) /* 45 */
.long SYMBOL_NAME(sys_setgid16)
.long SYMBOL_NAME(sys_getgid16)
.long SYMBOL_NAME(sys_signal)
.long SYMBOL_NAME(sys_geteuid16)
.long SYMBOL_NAME(sys_getegid16) /* 50 */
.long SYMBOL_NAME(sys_acct)
.long SYMBOL_NAME(sys_umount) /* recycled never used phys() */
.long SYMBOL_NAME(sys_ni_syscall) /* old lock syscall holder */
.long SYMBOL_NAME(sys_ioctl)
.long SYMBOL_NAME(sys_fcntl) /* 55 */
.long SYMBOL_NAME(sys_ni_syscall) /* old mpx syscall holder */
.long SYMBOL_NAME(sys_setpgid)
.long SYMBOL_NAME(sys_ni_syscall) /* old ulimit syscall holder */
.long SYMBOL_NAME(sys_olduname)
.long SYMBOL_NAME(sys_umask) /* 60 */
.long SYMBOL_NAME(sys_chroot)
.long SYMBOL_NAME(sys_ustat)
.long SYMBOL_NAME(sys_dup2)
.long SYMBOL_NAME(sys_getppid)
.long SYMBOL_NAME(sys_getpgrp) /* 65 */
.long SYMBOL_NAME(sys_setsid)
.long SYMBOL_NAME(sys_sigaction)
.long SYMBOL_NAME(sys_sgetmask)
.long SYMBOL_NAME(sys_ssetmask)
.long SYMBOL_NAME(sys_setreuid16) /* 70 */
.long SYMBOL_NAME(sys_setregid16)
.long SYMBOL_NAME(sys_sigsuspend)
.long SYMBOL_NAME(sys_sigpending)
.long SYMBOL_NAME(sys_sethostname)
.long SYMBOL_NAME(sys_setrlimit) /* 75 */
.long SYMBOL_NAME(sys_old_getrlimit)
.long SYMBOL_NAME(sys_getrusage)
.long SYMBOL_NAME(sys_gettimeofday)
.long SYMBOL_NAME(sys_settimeofday)
.long SYMBOL_NAME(sys_getgroups16) /* 80 */
.long SYMBOL_NAME(sys_setgroups16)
.long SYMBOL_NAME(old_select)
.long SYMBOL_NAME(sys_symlink)
.long SYMBOL_NAME(sys_lstat)
.long SYMBOL_NAME(sys_readlink) /* 85 */
.long SYMBOL_NAME(sys_uselib)
.long SYMBOL_NAME(sys_swapon)
.long SYMBOL_NAME(sys_reboot)
.long SYMBOL_NAME(old_readdir)
.long SYMBOL_NAME(old_mmap) /* 90 */
.long SYMBOL_NAME(sys_munmap)
.long SYMBOL_NAME(sys_truncate)
.long SYMBOL_NAME(sys_ftruncate)
.long SYMBOL_NAME(sys_fchmod)
.long SYMBOL_NAME(sys_fchown16) /* 95 */
.long SYMBOL_NAME(sys_getpriority)
.long SYMBOL_NAME(sys_setpriority)
.long SYMBOL_NAME(sys_ni_syscall) /* old profil syscall holder */
.long SYMBOL_NAME(sys_statfs)
.long SYMBOL_NAME(sys_fstatfs) /* 100 */
.long SYMBOL_NAME(sys_ioperm)
.long SYMBOL_NAME(sys_socketcall)
.long SYMBOL_NAME(sys_syslog)
.long SYMBOL_NAME(sys_setitimer)
.long SYMBOL_NAME(sys_getitimer) /* 105 */
.long SYMBOL_NAME(sys_newstat)
.long SYMBOL_NAME(sys_newlstat)
.long SYMBOL_NAME(sys_newfstat)
.long SYMBOL_NAME(sys_uname)
.long SYMBOL_NAME(sys_iopl) /* 110 */
.long SYMBOL_NAME(sys_vhangup)
.long SYMBOL_NAME(sys_ni_syscall) /* old &quot;idle&quot; system call */
.long SYMBOL_NAME(sys_vm86old)
.long SYMBOL_NAME(sys_wait4)
.long SYMBOL_NAME(sys_swapoff) /* 115 */
.long SYMBOL_NAME(sys_sysinfo)
.long SYMBOL_NAME(sys_ipc)
.long SYMBOL_NAME(sys_fsync)
.long SYMBOL_NAME(sys_sigreturn)
.long SYMBOL_NAME(sys_clone) /* 120 */
.long SYMBOL_NAME(sys_setdomainname)
.long SYMBOL_NAME(sys_newuname)
.long SYMBOL_NAME(sys_modify_ldt)
.long SYMBOL_NAME(sys_adjtimex)
.long SYMBOL_NAME(sys_mprotect) /* 125 */
.long SYMBOL_NAME(sys_sigprocmask)
.long SYMBOL_NAME(sys_create_module)
.long SYMBOL_NAME(sys_init_module)
.long SYMBOL_NAME(sys_delete_module)
.long SYMBOL_NAME(sys_get_kernel_syms) /* 130 */
.long SYMBOL_NAME(sys_quotactl)
.long SYMBOL_NAME(sys_getpgid)
.long SYMBOL_NAME(sys_fchdir)
.long SYMBOL_NAME(sys_bdflush)
.long SYMBOL_NAME(sys_sysfs) /* 135 */
.long SYMBOL_NAME(sys_personality)
.long SYMBOL_NAME(sys_ni_syscall) /* for afs_syscall */
.long SYMBOL_NAME(sys_setfsuid16)
.long SYMBOL_NAME(sys_setfsgid16)
.long SYMBOL_NAME(sys_llseek) /* 140 */
.long SYMBOL_NAME(sys_getdents)
.long SYMBOL_NAME(sys_select)
.long SYMBOL_NAME(sys_flock)
.long SYMBOL_NAME(sys_msync)
.long SYMBOL_NAME(sys_readv) /* 145 */
.long SYMBOL_NAME(sys_writev)
.long SYMBOL_NAME(sys_getsid)
.long SYMBOL_NAME(sys_fdatasync)
.long SYMBOL_NAME(sys_sysctl)
.long SYMBOL_NAME(sys_mlock) /* 150 */
.long SYMBOL_NAME(sys_munlock)
.long SYMBOL_NAME(sys_mlockall)
.long SYMBOL_NAME(sys_munlockall)
.long SYMBOL_NAME(sys_sched_setparam)
.long SYMBOL_NAME(sys_sched_getparam) /* 155 */
.long SYMBOL_NAME(sys_sched_setscheduler)
.long SYMBOL_NAME(sys_sched_getscheduler)
.long SYMBOL_NAME(sys_sched_yield)
.long SYMBOL_NAME(sys_sched_get_priority_max)
.long SYMBOL_NAME(sys_sched_get_priority_min) /* 160 */
.long SYMBOL_NAME(sys_sched_rr_get_interval)
.long SYMBOL_NAME(sys_nanosleep)
.long SYMBOL_NAME(sys_mremap)
.long SYMBOL_NAME(sys_setresuid16)
.long SYMBOL_NAME(sys_getresuid16) /* 165 */
.long SYMBOL_NAME(sys_vm86)
.long SYMBOL_NAME(sys_query_module)
.long SYMBOL_NAME(sys_poll)
.long SYMBOL_NAME(sys_nfsservctl)
.long SYMBOL_NAME(sys_setresgid16) /* 170 */
.long SYMBOL_NAME(sys_getresgid16)
.long SYMBOL_NAME(sys_prctl)
.long SYMBOL_NAME(sys_rt_sigreturn)
.long SYMBOL_NAME(sys_rt_sigaction)
.long SYMBOL_NAME(sys_rt_sigprocmask) /* 175 */
.long SYMBOL_NAME(sys_rt_sigpending)
.long SYMBOL_NAME(sys_rt_sigtimedwait)
.long SYMBOL_NAME(sys_rt_sigqueueinfo)
.long SYMBOL_NAME(sys_rt_sigsuspend)
.long SYMBOL_NAME(sys_pread) /* 180 */
.long SYMBOL_NAME(sys_pwrite)
.long SYMBOL_NAME(sys_chown16)
.long SYMBOL_NAME(sys_getcwd)
.long SYMBOL_NAME(sys_capget)
.long SYMBOL_NAME(sys_capset) /* 185 */
.long SYMBOL_NAME(sys_sigaltstack)
.long SYMBOL_NAME(sys_sendfile)
.long SYMBOL_NAME(sys_ni_syscall) /* streams1 */
.long SYMBOL_NAME(sys_ni_syscall) /* streams2 */
.long SYMBOL_NAME(sys_vfork) /* 190 */
.long SYMBOL_NAME(sys_getrlimit)
.long SYMBOL_NAME(sys_mmap2)
.long SYMBOL_NAME(sys_truncate64)
.long SYMBOL_NAME(sys_ftruncate64)
.long SYMBOL_NAME(sys_stat64) /* 195 */
.long SYMBOL_NAME(sys_lstat64)
.long SYMBOL_NAME(sys_fstat64)
.long SYMBOL_NAME(sys_lchown)
.long SYMBOL_NAME(sys_getuid)
.long SYMBOL_NAME(sys_getgid) /* 200 */
.long SYMBOL_NAME(sys_geteuid)
.long SYMBOL_NAME(sys_getegid)
.long SYMBOL_NAME(sys_setreuid)
.long SYMBOL_NAME(sys_setregid)
.long SYMBOL_NAME(sys_getgroups) /* 205 */
.long SYMBOL_NAME(sys_setgroups)
.long SYMBOL_NAME(sys_fchown)
.long SYMBOL_NAME(sys_setresuid)
.long SYMBOL_NAME(sys_getresuid)
.long SYMBOL_NAME(sys_setresgid) /* 210 */
.long SYMBOL_NAME(sys_getresgid)
.long SYMBOL_NAME(sys_chown)
.long SYMBOL_NAME(sys_setuid)
.long SYMBOL_NAME(sys_setgid)
.long SYMBOL_NAME(sys_setfsuid) /* 215 */
.long SYMBOL_NAME(sys_setfsgid)
.long SYMBOL_NAME(sys_pivot_root)
.long SYMBOL_NAME(sys_mincore)
.long SYMBOL_NAME(sys_madvise)
.long SYMBOL_NAME(sys_getdents64) /* 220 */
.long SYMBOL_NAME(sys_fcntl64)
.long SYMBOL_NAME(sys_ni_syscall) /* reserved for TUX */
.long SYMBOL_NAME(sys_ni_syscall) /* Reserved for Security */
.long SYMBOL_NAME(sys_gettid)
.long SYMBOL_NAME(sys_readahead) /* 225 */
</verb>
</p><sect2>
IRQ Event
<p>
When an IRQ comes, the task that is running is interrupted in
order to service the IRQ Handler.
</p>
<p>
After the IRQ is handled, control returns backs exactly to point
of interrupt, like nothing happened.
</p>
<p>
<verb>
Running Task
|-----------| (3)
NORMAL | | | &lsqb;break execution&rsqb; IRQ Handler
EXECUTION (1)| | | -------------&gt;|---------|
| &bsol;|/ | | | does |
IRQ (2)----&gt;| .. |-----&gt; | some |
| | |&lt;----- | work |
BACK TO | | | | | ..(4). |
NORMAL (6)| &bsol;|/ | &lt;-------------|_________|
EXECUTION |___________| &lsqb;return to code&rsqb;
(5)
USER MODE KERNEL MODE
User-&gt;Kernel Mode Transition caused by IRQ event
</verb>
</p><p>
The numbered steps below refer to the sequence of events in the
diagram above:
</p>
<p>
<enum>
<item>
Process is executing
<item>
IRQ comes while the task is running.
<item>
Task is interrupted to call an "Interrupt handler".
<item>
The "Interrupt handler" code is executed.
<item>
Control returns back to task user mode (as if nothing happened)
<item>
Process returns back to normal execution
</enum>
</p><p>
Special interest has the Timer IRQ, coming every TIMER ms to
manage:
</p>
<p>
<enum>
<item>
Alarms
<item>
System and task counters (used by schedule to decide when stop
a process or for accounting)
<item>
Multitasking based on wake up mechanism after TIMESLICE time.
</enum>
</p><sect1>
Multitasking
<sect2>
Mechanism
<p>
The key point of modern OSs is the "Task". The Task is an application
running in memory sharing all resources (included CPU and Memory)
with other Tasks.
</p>
<p>
This "resource sharing" is managed by the "Multitasking Mechanism".
The Multitasking Mechanism switches from one task to another after
a "timeslice" time. Users have the "illusion" that they own all resources.
We can also imagine a single user scenario, where a user can have
the "illusion" of running many tasks at the same time.
</p>
<p>
To implement this multitasking, the task uses "the state" variable,
which can be:
</p>
<p>
<enum>
<item>
READY, ready for execution
<item>
BLOCKED, waiting for a resource
</enum>
</p><p>
The task state is managed by its presence in a relative list:
READY list and BLOCKED list.
</p>
<sect2>
Task Switching
<p>
The movement from one task to another is called ''Task Switching''.
many computers have a hardware instruction which automatically performs
this operation. Task Switching occurs in the following cases:
</p>
<p>
<enum>
<item>
After Timeslice ends: we need to schedule a "Ready for execution"
task and give it access.
<item>
When a Task has to wait for a device: we need to schedule a new
task and switch to it *
</enum>
</p><p>
* We schedule another task to prevent "Busy Form Waiting", which
occurs when we are waiting for a device instead performing other
work.
</p>
<p>
Task Switching is managed by the "Schedule" entity.
</p>
<p>
<verb>
Timer | |
IRQ | | Schedule
| | | ________________________
|-----&gt;| Task 1 |&lt;------------------&gt;|(1)Chooses a Ready Task |
| | | |(2)Task Switching |
| |___________| |________________________|
| | | /|&bsol;
| | | |
| | | |
| | | |
| | | |
|-----&gt;| Task 2 |&lt;-------------------------------|
| | | |
| |___________| |
. . . . .
. . . . .
. . . . .
| | | |
| | | |
------&gt;| Task N |&lt;--------------------------------
| |
|___________|
Task Switching based on TimeSlice
</verb>
</p><p>
A typical Timeslice for Linux is about 10 ms.
</p>
<p>
<verb>
| |
| | Resource _____________________________
| Task 1 |-----------&gt;|(1) Enqueue Resource request |
| | Access |(2) Mark Task as blocked |
| | |(3) Choose a Ready Task |
|___________| |(4) Task Switching |
|_____________________________|
|
|
| | |
| | |
| Task 2 |&lt;-------------------------
| |
| |
|___________|
Task Switching based on Waiting for a Resource
</verb>
</p><sect1>
Microkernel vs Monolithic OS
<sect2>
Overview
<p>
Until now we viewed so called Monolithic OS, but there is also
another kind of OS: ''Microkernel''.
</p>
<p>
A Microkernel OS uses Tasks, not only for user mode processes,
but also as a real kernel manager, like Floppy-Task, HDD-Task, Net-Task
and so on. Some examples are Amoeba, and Mach.
</p>
<sect2>
PROs and CONTROs of Microkernel OS
<p>
PROS:
</p>
<p>
<itemize>
<item>
OS is simpler to maintain because each Task manages a single
kind of operation. So if you want to modify networking, you modify
Net-Task (ideally, if it is not needed a structural update).
</itemize>
</p><p>
CONS:
</p>
<p>
<itemize>
<item>
Performances are worse than Monolithic OS, because you have to
add 2*TASK_SWITCH times (the first to enter the specific Task, the
second to go out from it).
</itemize>
</p><p>
My personal opinion is that, Microkernels are a good didactic
example (like Minix) but they are not ''optimal'', so not really
suitable. Linux uses a few Tasks, called "Kernel Threads" to implement
a little microkernel structure (like kswapd, which is used to retrieve
memory pages from mass storage). In this case there are no problems
with perfomance because swapping is a very slow job.
</p>
<sect1>
Networking
<sect2>
ISO OSI levels
<p>
Standard ISO-OSI describes a network architecture with the following
levels:
</p>
<p>
<enum>
<item>
Physical level (examples: PPP and Ethernet)
<item>
Data-link level (examples: PPP and Ethernet)
<item>
Network level (examples: IP, and X.25)
<item>
Transport level (examples: TCP, UDP)
<item>
Session level (SSL)
<item>
Presentation level (FTP binary-ascii coding)
<item>
Application level (applications like Netscape)
</enum>
</p><p>
The first 2 levels listed above are often implemented in hardware.
Next levels are in software (or firmware for routers).
</p>
<p>
Many protocols are used by an OS: one of these is TCP/IP (the
most important living on 3-4 levels).
</p>
<sect2>
What does the kernel?
<p>
The kernel doesn't know anything (only addresses) about first
2 levels of ISO-OSI.
</p>
<p>
In RX it:
</p>
<p>
<enum>
<item>
Manages handshake with low levels devices (like ethernet card
or modem) receiving "frames" from them.
<item>
Builds TCP/IP "packets" from "frames" (like Ethernet or PPP ones),
<item>
Convers ''packets'' in ''sockets'' passing them to the right
application (using port number) or
<item>
Forwards packets to the right queue
</enum>
<p>
<verb>
frames packets sockets
NIC ---------&gt; Kernel ----------&gt; Application
| packets
--------------&gt; Forward
- RX -
</verb>
</p><p>
In TX stage it:
</p>
<p>
<enum>
<item>
Converts sockets or
<item>
Queues datas into TCP/IP ''packets''
<item>
Splits ''packets" into "frames" (like Ethernet or PPP ones)
<item>
Sends ''frames'' using HW drivers
</enum>
<p>
<verb>
sockets packets frames
Application ---------&gt; Kernel ----------&gt; NIC
packets /|&bsol;
Forward -------------------
- TX -
</verb>
</p><sect1>
Virtual Memory
<sect2>
Segmentation
<p>
Segmentation is the first method to solve memory allocation problems:
it allows you to compile source code without caring where the application
will be placed in memory. As a matter of fact, this feature helps
applications developers to develop in a independent fashion from
the OS e also from the hardware.
</p>
<p>
<verb>
| Stack |
| | |
| &bsol;|/ |
| Free |
| /|&bsol; | Segment &lt;---&gt; Process
| | |
| Heap |
| Data uninitialized |
| Data initialized |
| Code |
|____________________|
Segment
</verb>
</p><p>
We can say that a segment is the logical entity of an application,
or the image of the application in memory.
</p>
<p>
When programming, we don't care where our data is put in memory,
we only care about the offset inside our segment (our application).
</p>
<p>
We use to assign a Segment to each Process and vice versa. In
Linux this is not true. Linux uses only 4 segments for either Kernel
and all Processes.
</p>
<sect3>
Problems of Segmentation
<p>
<verb>
____________________
-----&gt;| |-----&gt;
| IN | Segment A | OUT
____________________ | |____________________|
| |____| | |
| Segment B | | Segment B |
| |____ | |
|____________________| | |____________________|
| | Segment C |
| |____________________|
-----&gt;| Segment D |-----&gt;
IN |____________________| OUT
Segmentation problem
</verb>
</p><p>
In the diagram above, we want to get exit processes A, and D
and enter process B. As we can see there is enough space for B, but
we cannot split it in 2 pieces, so we CANNOT load it (memory out).
</p>
<p>
The reason this problem occurs is because pure segments are continuous
areas (because they are logical areas) and cannot be split.
</p>
<sect2>
Pagination
<p>
<verb>
____________________
| Page 1 |
|____________________|
| Page 2 |
|____________________|
| .. | Segment &lt;---&gt; Process
|____________________|
| Page n |
|____________________|
| |
|____________________|
| |
|____________________|
Segment
</verb>
</p><p>
Pagination splits memory in &quot;n&quot; pieces, each one with
a fixed length.
</p>
<p>
A process may be loaded in one or more Pages. When memory is
freed, all pages are freed (see Segmentation Problem, before).
</p>
<p>
Pagination is also used for another important purpose, "Swapping".
If a page is not present in physical memory then it generates an
EXCEPTION, that will make the Kernel search for a new page in storage
memory. This mechanism allow OS to load more applications than the
ones allowed by physical memory only.
</p>
<sect3>
Pagination Problem
<p>
<verb>
____________________
Page X | Process Y |
|____________________|
| |
| WASTE |
| SPACE |
|____________________|
Pagination Problem
</verb>
</p><p>
In the diagram above, we can see what is wrong with the pagination
policy: when a Process Y loads into Page X, ALL memory space of the
Page is allocated, so the remaining space at the end of Page is wasted.
</p>
<sect2>
Segmentation and Pagination
<p>
How can we solve segmentation and pagination problems? Using
either 2 policies.
</p>
<p>
<verb>
| .. |
|____________________|
-----&gt;| Page 1 |
| |____________________|
| | .. |
____________________ | |____________________|
| | |----&gt;| Page 2 |
| Segment X | ----| |____________________|
| | | | .. |
|____________________| | |____________________|
| | .. |
| |____________________|
|----&gt;| Page 3 |
|____________________|
| .. |
</verb>
</p><p>
Process X, identified by Segment X, is split in 3 pieces and
each of one is loaded in a page.
</p>
<p>
We do not have:
</p>
<p>
<enum>
<item>
Segmentation problem: we allocate per Pages, so we also free
Pages and we manage free space in an optimized way.
<item>
Pagination problem: only last page wastes space, but we can decide
to use very small pages, for example 4096 bytes length (losing at
maximum 4096*N_Tasks bytes) and manage hierarchical paging (using
2 or 3 levels of paging)
</enum>
<p>
<verb>
| | | |
| | Offset2 | Value |
| | /|&bsol;| |
Offset1 | |----- | | |
/|&bsol; | | | | | |
| | | | &bsol;|/| |
| | | ------&gt;| |
&bsol;|/ | | | |
Base Paging Address ----&gt;| | | |
| ....... | | ....... |
| | | |
Hierarchical Paging
</verb>
</p><sect>
Linux Startup
<p>
We start the Linux kernel first from C code executed from ''startup_32:''
asm label:
</p>
<p>
<verb>
|startup_32:
|start_kernel
|lock_kernel
|trap_init
|init_IRQ
|sched_init
|softirq_init
|time_init
|console_init
|&num;ifdef CONFIG_MODULES
|init_modules
|&num;endif
|kmem_cache_init
|sti
|calibrate_delay
|mem_init
|kmem_cache_sizes_init
|pgtable_cache_init
|fork_init
|proc_caches_init
|vfs_caches_init
|buffer_init
|page_cache_init
|signals_init
|&num;ifdef CONFIG_PROC_FS
|proc_root_init
|&num;endif
|&num;if defined(CONFIG_SYSVIPC)
|ipc_init
|&num;endif
|check_bugs
|smp_init
|rest_init
|kernel_thread
|unlock_kernel
|cpu_idle
</verb>
<p>
<itemize>
<item>
startup_32 &lsqb;arch/i386/kernel/head.S&rsqb;
<item>
start_kernel &lsqb;init/main.c&rsqb;
<item>
lock_kernel &lsqb;include/asm/smplock.h&rsqb;
<item>
trap_init &lsqb;arch/i386/kernel/traps.c&rsqb;
<item>
init_IRQ &lsqb;arch/i386/kernel/i8259.c&rsqb;
<item>
sched_init &lsqb;kernel/sched.c&rsqb;
<item>
softirq_init &lsqb;kernel/softirq.c&rsqb;
<item>
time_init &lsqb;arch/i386/kernel/time.c&rsqb;
<item>
console_init &lsqb;drivers/char/tty_io.c&rsqb;
<item>
init_modules &lsqb;kernel/module.c&rsqb;
<item>
kmem_cache_init &lsqb;mm/slab.c&rsqb;
<item>
sti &lsqb;include/asm/system.h&rsqb;
<item>
calibrate_delay &lsqb;init/main.c&rsqb;
<item>
mem_init &lsqb;arch/i386/mm/init.c&rsqb;
<item>
kmem_cache_sizes_init &lsqb;mm/slab.c&rsqb;
<item>
pgtable_cache_init &lsqb;arch/i386/mm/init.c&rsqb;
<item>
fork_init &lsqb;kernel/fork.c&rsqb;
<item>
proc_caches_init
<item>
vfs_caches_init &lsqb;fs/dcache.c&rsqb;
<item>
buffer_init &lsqb;fs/buffer.c&rsqb;
<item>
page_cache_init &lsqb;mm/filemap.c&rsqb;
<item>
signals_init &lsqb;kernel/signal.c&rsqb;
<item>
proc_root_init &lsqb;fs/proc/root.c&rsqb;
<item>
ipc_init &lsqb;ipc/util.c&rsqb;
<item>
check_bugs &lsqb;include/asm/bugs.h&rsqb;
<item>
smp_init &lsqb;init/main.c&rsqb;
<item>
rest_init
<item>
kernel_thread &lsqb;arch/i386/kernel/process.c&rsqb;
<item>
unlock_kernel &lsqb;include/asm/smplock.h&rsqb;
<item>
cpu_idle &lsqb;arch/i386/kernel/process.c&rsqb;
</itemize>
</p><p>
The last function ''rest_init'' does the following:
</p>
<p>
<enum>
<item>
launches the kernel thread ''init''
<item>
calls unlock_kernel
<item>
makes the kernel run cpu_idle routine, that will be the idle
loop executing when nothing is scheduled
</enum>
</p><p>
In fact the start_kernel procedure never ends. It will execute
cpu_idle routine endlessly.
</p>
<p>
Follows ''init'' description, which is the first Kernel Thread:
</p>
<p>
<verb>
|init
|lock_kernel
|do_basic_setup
|mtrr_init
|sysctl_init
|pci_init
|sock_init
|start_context_thread
|do_init_calls
|(*call())-&gt; kswapd_init
|prepare_namespace
|free_initmem
|unlock_kernel
|execve
</verb>
</p><sect>
Linux Peculiarities
<sect1>
Overview
<p>
Linux has some peculiarities that distinguish it from other OSs.
These peculiarities include:
</p>
<p>
<enum>
<item>
Pagination only
<item>
Softirq
<item>
Kernel threads
<item>
Kernel modules
<item>
''Proc'' directory
</enum>
</p><sect2>
Flexibility Elements
<p>
Points 4 and 5 give system administrators an enormous flexibility
on system configuration from user mode allowing them to solve also
critical kernel bugs or specific problems without have to reboot
the machine. For example, if you needed to change something on a
big server and you didn't want to make a reboot, you could prepare
the kernel to talk with a module, that you'll write.
</p>
<sect1>
Pagination only
<p>
Linux doesn't use segmentation to distinguish Tasks from each
other; it uses pagination. (Only 2 segments are used for all Tasks,
CODE and DATA/STACK)
</p>
<p>
We can also say that an interTask page fault never occurs, because
each Task uses a set of Page Tables that are different for each Task.
There are some cases where different Tasks point to same Page Tables,
like shared libraries: this is needed to reduce memory usage; remember
that shared libraries are CODE only cause all datas are stored into
actual Task stack.
</p>
<sect2>
Linux segments
<p>
Under the Linux kernel only 4 segments exist:
</p>
<p>
<enum>
<item>
Kernel Code &lsqb;0x10&rsqb;
<item>
Kernel Data / Stack &lsqb;0x18&rsqb;
<item>
User Code &lsqb;0x23&rsqb;
<item>
User Data / Stack &lsqb;0x2b&rsqb;
</enum>
</p><p>
&lsqb;syntax is ''Purpose &lsqb;Segment&rsqb;''&rsqb;
</p>
<p>
Under Intel architecture, the segment registers used are:
</p>
<p>
<itemize>
<item>
CS for Code Segment
<item>
DS for Data Segment
<item>
SS for Stack Segment
<item>
ES for Alternative Segment (for example used to make a memory
copy between 2 different segments)
</itemize>
</p><p>
So, every Task uses 0x23 for code and 0x2b for data/stack.
</p>
<sect2>
Linux pagination
<p>
Under Linux 3 levels of pages are used, depending on the architecture.
Under Intel only 2 levels are supported. Linux also supports Copy
on Write mechanisms (please see Cap.10 for more information).
</p>
<sect2>
Why don't interTasks address conflicts exist?
<p>
The answer is very very simple: interTask address conflicts
cannot exist because they are impossible. Linear -&gt; physical
mapping is done by "Pagination", so it just needs to assign physical
pages in an univocal fashion.
</p>
<sect2>
Do we need to defragment memory?
<p>
No. Page assigning is a dynamic process. We need a page only
when a Task asks for it, so we choose it from free memory paging
in an ordered fashion. When we want to release the page, we only
have to add it to the free pages list.
</p>
<sect2>
What about Kernel Pages?
<p>
Kernel pages have a problem: they can be allocated in a dynamic
fashion but we cannot have a guarantee that they are in contiguous
area allocation, because linear kernel space is equivalent to physical
kernel space.
</p>
<p>
For Code Segment there is no problem. Boot code is allocated
at boot time (so we have a fixed amount of memory to allocate), and
on modules we only have to allocate a memory area which could contain
module code.
</p>
<p>
The real problem is the stack segment because each Task uses
some kernel stack pages. Stack segments must be contiguous (according
to stack definition), so we have to establish a maximum limit for
each Task's stack dimension. If we exceed this limit bad things happen.
We overwrite kernel mode process data structures.
</p>
<p>
The structure of the Kernel helps us, because kernel functions
are never:
</p>
<p>
<itemize>
<item>
recursive
<item>
intercalling more than N times.
</itemize>
</p><p>
Once we know N, and we know the average of static variables for
all kernel functions, we can estimate a stack limit.
</p>
<p>
If you want to try the problem out, you can create a module with
a function inside calling itself many times. After a fixed number
of times, the kernel module will hang because of a page fault exception
handler (typically write to a read-only page).
</p>
<sect1>
Softirq
<p>
When an IRQ comes, task switching is deferred until later to
get better performance. Some Task jobs (that could have to be done
just after the IRQ and that could take much CPU in interrupt time,
like building up a TCP/IP packet) are queued and will be done at
scheduling time (once a time-slice will end).
</p>
<p>
In recent kernels (2.4.x) the softirq mechanisms are given to
a kernel_thread: ''ksoftirqd_CPUn''. n stands for the number of CPU
executing kernel_thread (in a monoprocessor system ''ksoftirqd_CPU0''
uses PID 3).
</p>
<sect2>
Preparing Softirq
<sect2>
Enabling Softirq
<p>
''cpu_raise_softirq'' is a routine that will wake_up ''ksoftirqd_CPU0''
kernel thread, to let it manage the enqueued job.
</p>
<p>
<verb>
|cpu_raise_softirq
|__cpu_raise_softirq
|wakeup_softirqd
|wake_up_process
</verb>
<p>
<itemize>
<item>
cpu_raise_softirq &lsqb;kernel/softirq.c&rsqb;
<item>
__cpu_raise_softirq &lsqb;include/linux/interrupt.h&rsqb;
<item>
wakeup_softirq &lsqb;kernel/softirq.c&rsqb;
<item>
wake_up_process &lsqb;kernel/sched.c&rsqb;
</itemize>
</p><p>
''__cpu_raise_softirq'' routine will set right bit in the vector
describing softirq pending.
</p>
<p>
''wakeup_softirq'' uses ''wakeup_process'' to wake up ''ksoftirqd_CPU0''
kernel thread.
</p>
<sect2>
Executing Softirq
<p>
TODO: describing data structures involved in softirq mechanism.
</p>
<p>
When kernel thread ''ksoftirqd_CPU0'' has been woken up, it will
execute queued jobs
</p>
<p>
The code of ''ksoftirqd_CPU0'' is (main endless loop):
</p>
<p>
<verb>
for (;;) &lcub;
if (!softirq_pending(cpu))
schedule();
__set_current_state(TASK_RUNNING);
while (softirq_pending(cpu)) &lcub;
do_softirq();
if (current-&gt;need_resched)
schedule
&rcub;
__set_current_state(TASK_INTERRUPTIBLE)
&rcub;
</verb>
<p>
<itemize>
<item>
ksoftirqd &lsqb;kernel/softirq.c&rsqb;
</itemize>
</p><sect1>
Kernel Threads
<p>
Even though Linux is a monolithic OS, a few ''kernel threads''
exist to do housekeeping work.
</p>
<p>
These Tasks don't utilize USER memory; they share KERNEL memory.
They also operate at the highest privilege (RING 0 on a i386 architecture)
like any other kernel mode piece of code.
</p>
<p>
Kernel threads are created by ''kernel_thread &lsqb;arch/i386/kernel/process&rsqb;''
function, which calls ''clone'' &lsqb;arch/i386/kernel/process.c&rsqb;
system call from assembler (which is a ''fork'' like system call):
</p>
<p>
<verb>
int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
&lcub;
long retval, d0;
__asm__ __volatile__(
&quot;movl &percnt;&percnt;esp,&percnt;&percnt;esi&bsol;n&bsol;t&quot;
&quot;int &dollar;0x80&bsol;n&bsol;t&quot; /* Linux/i386 system call */
&quot;cmpl &percnt;&percnt;esp,&percnt;&percnt;esi&bsol;n&bsol;t&quot; /* child or parent? */
&quot;je 1f&bsol;n&bsol;t&quot; /* parent - jump */
/* Load the argument into eax, and push it. That way, it does
* not matter whether the called function is compiled with
* -mregparm or not. */
&quot;movl &percnt;4,&percnt;&percnt;eax&bsol;n&bsol;t&quot;
&quot;pushl &percnt;&percnt;eax&bsol;n&bsol;t&quot;
&quot;call *&percnt;5&bsol;n&bsol;t&quot; /* call fn */
&quot;movl &percnt;3,&percnt;0&bsol;n&bsol;t&quot; /* exit */
&quot;int &dollar;0x80&bsol;n&quot;
&quot;1:&bsol;t&quot;
:&quot;=&amp;a&quot; (retval), &quot;=&amp;S&quot; (d0)
:&quot;0&quot; (__NR_clone), &quot;i&quot; (__NR_exit),
&quot;r&quot; (arg), &quot;r&quot; (fn),
&quot;b&quot; (flags | CLONE_VM)
: &quot;memory&quot;);
return retval;
&rcub;
</verb>
</p><p>
Once called, we have a new Task (usually with very low PID number,
like 2,3, etc.) waiting for a very slow resource, like swap or usb
event. A very slow resource is used because we would have a task
switching overhead otherwise.
</p>
<p>
Below is a list of most common kernel threads (from ''ps x''
command):
</p>
<p>
<verb>
PID COMMAND
1 init
2 keventd
3 kswapd
4 kreclaimd
5 bdflush
6 kupdated
7 kacpid
67 khubd
</verb>
</p><p>
'init' kernel thread is the first process created, at boot time.
It will call all other User Mode Tasks (from file /etc/inittab) like
console daemons, tty daemons and network daemons (''rc'' scripts).
</p>
<sect2>
Example of Kernel Threads: kswapd &lsqb;mm/vmscan.c&rsqb;.
<p>
''kswapd'' is created by ''clone() &lsqb;arch/i386/kernel/process.c&rsqb;''
</p>
<p>
Initialisation routines:
</p>
<p>
<verb>
|do_initcalls
|kswapd_init
|kernel_thread
|syscall fork (in assembler)
</verb>
</p><p>
do_initcalls &lsqb;init/main.c&rsqb;
</p>
<p>
kswapd_init &lsqb;mm/vmscan.c&rsqb;
</p>
<p>
kernel_thread &lsqb;arch/i386/kernel/process.c&rsqb;
</p>
<sect1>
Kernel Modules
<sect2>
Overview
<p>
Linux Kernel modules are pieces of code (examples: fs, net, and
hw driver) running in kernel mode that you can add at runtime.
</p>
<p>
The Linux core cannot be modularized: scheduling and interrupt
management or core network, and so on.
</p>
<p>
Under "/lib/modules/KERNEL_VERSION/" you can find all the modules
installed on your system.
</p>
<sect2>
Module loading and unloading
<p>
To load a module, type the following:
</p>
<p>
<verb>
insmod MODULE_NAME parameters
example: insmod ne io=0x300 irq=9
</verb>
</p><p>
NOTE: You can use modprobe in place of insmod if you want the
kernel automatically search some parameter (for example when using
PCI driver, or if you have specified parameter under /etc/conf.modules
file).
</p>
<p>
To unload a module, type the following:
</p>
<p>
<verb>
rmmod MODULE_NAME
</verb>
</p><sect2>
Module definition
<p>
A module always contains:
</p>
<p>
<enum>
<item>
"init_module" function, executed at insmod (or modprobe) command
<item>
"cleanup_module" function, executed at rmmod command
</enum>
</p><p>
If these functions are not in the module, you need to add 2 macros
to specify what functions will act as init and exit module:
</p>
<p>
<enum>
<item>
module_init(FUNCTION_NAME)
<item>
module_exit(FUNCTION_NAME)
</enum>
</p><p>
NOTE: a module can "see" a kernel variable only if it has been
exported (with macro EXPORT_SYMBOL).
</p>
<sect2>
A useful trick for adding flexibility to your kernel
<p>
<verb>
// kernel sources side
void (*foo_function_pointer)(void *);
if (foo_function_pointer)
(foo_function_pointer)(parameter);
// module side
extern void (*foo_function_pointer)(void *);
void my_function(void *parameter) &lcub;
//My code
&rcub;
int init_module() &lcub;
foo_function_pointer = &amp;my_function;
&rcub;
int cleanup_module() &lcub;
foo_function_pointer = NULL;
&rcub;
</verb>
</p><p>
This simple trick allows you to have very high flexibility in
your Kernel, because only when you load the module you'll make "my_function"
routine execute. This routine will do everything you want to do:
for example ''rshaper'' module, which controls bandwidth input traffic
from the network, works in this kind of matter.
</p>
<p>
Notice that the whole module mechanism is possible thanks to
some global variables exported to modules, such as head list (allowing
you to extend the list as much as you want). Typical examples are
fs, generic devices (char, block, net, telephony). You have to prepare
the kernel to accept your new module; in some cases you have to create
an infrastructure (like telephony one, that was recently created)
to be as standard as possible.
</p>
<sect1>
Proc directory
<p>
Proc fs is located in the /proc directory, which is a special
directory allowing you to talk directly with kernel.
</p>
<p>
Linux uses ''proc'' directory to support direct kernel communications:
this is necessary in many cases, for example when you want see main
processes data structures or enable ''proxy-arp'' feature on one
interface and not in others, you want to change max number of threads,
or if you want to debug some bus state, like ISA or PCI, to know
what cards are installed and what I/O addresses and IRQs are assigned
to them.
</p>
<p>
<verb>
|-- bus
| |-- pci
| | |-- 00
| | | |-- 00.0
| | | |-- 01.0
| | | |-- 07.0
| | | |-- 07.1
| | | |-- 07.2
| | | |-- 07.3
| | | |-- 07.4
| | | |-- 07.5
| | | |-- 09.0
| | | |-- 0a.0
| | | `-- 0f.0
| | |-- 01
| | | `-- 00.0
| | `-- devices
| `-- usb
|-- cmdline
|-- cpuinfo
|-- devices
|-- dma
|-- dri
| `-- 0
| |-- bufs
| |-- clients
| |-- mem
| |-- name
| |-- queues
| |-- vm
| `-- vma
|-- driver
|-- execdomains
|-- filesystems
|-- fs
|-- ide
| |-- drivers
| |-- hda -&gt; ide0/hda
| |-- hdc -&gt; ide1/hdc
| |-- ide0
| | |-- channel
| | |-- config
| | |-- hda
| | | |-- cache
| | | |-- capacity
| | | |-- driver
| | | |-- geometry
| | | |-- identify
| | | |-- media
| | | |-- model
| | | |-- settings
| | | |-- smart_thresholds
| | | `-- smart_values
| | |-- mate
| | `-- model
| |-- ide1
| | |-- channel
| | |-- config
| | |-- hdc
| | | |-- capacity
| | | |-- driver
| | | |-- identify
| | | |-- media
| | | |-- model
| | | `-- settings
| | |-- mate
| | `-- model
| `-- via
|-- interrupts
|-- iomem
|-- ioports
|-- irq
| |-- 0
| |-- 1
| |-- 10
| |-- 11
| |-- 12
| |-- 13
| |-- 14
| |-- 15
| |-- 2
| |-- 3
| |-- 4
| |-- 5
| |-- 6
| |-- 7
| |-- 8
| |-- 9
| `-- prof_cpu_mask
|-- kcore
|-- kmsg
|-- ksyms
|-- loadavg
|-- locks
|-- meminfo
|-- misc
|-- modules
|-- mounts
|-- mtrr
|-- net
| |-- arp
| |-- dev
| |-- dev_mcast
| |-- ip_fwchains
| |-- ip_fwnames
| |-- ip_masquerade
| |-- netlink
| |-- netstat
| |-- packet
| |-- psched
| |-- raw
| |-- route
| |-- rt_acct
| |-- rt_cache
| |-- rt_cache_stat
| |-- snmp
| |-- sockstat
| |-- softnet_stat
| |-- tcp
| |-- udp
| |-- unix
| `-- wireless
|-- partitions
|-- pci
|-- scsi
| |-- ide-scsi
| | `-- 0
| `-- scsi
|-- self -&gt; 2069
|-- slabinfo
|-- stat
|-- swaps
|-- sys
| |-- abi
| | |-- defhandler_coff
| | |-- defhandler_elf
| | |-- defhandler_lcall7
| | |-- defhandler_libcso
| | |-- fake_utsname
| | `-- trace
| |-- debug
| |-- dev
| | |-- cdrom
| | | |-- autoclose
| | | |-- autoeject
| | | |-- check_media
| | | |-- debug
| | | |-- info
| | | `-- lock
| | `-- parport
| | |-- default
| | | |-- spintime
| | | `-- timeslice
| | `-- parport0
| | |-- autoprobe
| | |-- autoprobe0
| | |-- autoprobe1
| | |-- autoprobe2
| | |-- autoprobe3
| | |-- base-addr
| | |-- devices
| | | |-- active
| | | `-- lp
| | | `-- timeslice
| | |-- dma
| | |-- irq
| | |-- modes
| | `-- spintime
| |-- fs
| | |-- binfmt_misc
| | |-- dentry-state
| | |-- dir-notify-enable
| | |-- dquot-nr
| | |-- file-max
| | |-- file-nr
| | |-- inode-nr
| | |-- inode-state
| | |-- jbd-debug
| | |-- lease-break-time
| | |-- leases-enable
| | |-- overflowgid
| | `-- overflowuid
| |-- kernel
| | |-- acct
| | |-- cad_pid
| | |-- cap-bound
| | |-- core_uses_pid
| | |-- ctrl-alt-del
| | |-- domainname
| | |-- hostname
| | |-- modprobe
| | |-- msgmax
| | |-- msgmnb
| | |-- msgmni
| | |-- osrelease
| | |-- ostype
| | |-- overflowgid
| | |-- overflowuid
| | |-- panic
| | |-- printk
| | |-- random
| | | |-- boot_id
| | | |-- entropy_avail
| | | |-- poolsize
| | | |-- read_wakeup_threshold
| | | |-- uuid
| | | `-- write_wakeup_threshold
| | |-- rtsig-max
| | |-- rtsig-nr
| | |-- sem
| | |-- shmall
| | |-- shmmax
| | |-- shmmni
| | |-- sysrq
| | |-- tainted
| | |-- threads-max
| | `-- version
| |-- net
| | |-- 802
| | |-- core
| | | |-- hot_list_length
| | | |-- lo_cong
| | | |-- message_burst
| | | |-- message_cost
| | | |-- mod_cong
| | | |-- netdev_max_backlog
| | | |-- no_cong
| | | |-- no_cong_thresh
| | | |-- optmem_max
| | | |-- rmem_default
| | | |-- rmem_max
| | | |-- wmem_default
| | | `-- wmem_max
| | |-- ethernet
| | |-- ipv4
| | | |-- conf
| | | | |-- all
| | | | | |-- accept_redirects
| | | | | |-- accept_source_route
| | | | | |-- arp_filter
| | | | | |-- bootp_relay
| | | | | |-- forwarding
| | | | | |-- log_martians
| | | | | |-- mc_forwarding
| | | | | |-- proxy_arp
| | | | | |-- rp_filter
| | | | | |-- secure_redirects
| | | | | |-- send_redirects
| | | | | |-- shared_media
| | | | | `-- tag
| | | | |-- default
| | | | | |-- accept_redirects
| | | | | |-- accept_source_route
| | | | | |-- arp_filter
| | | | | |-- bootp_relay
| | | | | |-- forwarding
| | | | | |-- log_martians
| | | | | |-- mc_forwarding
| | | | | |-- proxy_arp
| | | | | |-- rp_filter
| | | | | |-- secure_redirects
| | | | | |-- send_redirects
| | | | | |-- shared_media
| | | | | `-- tag
| | | | |-- eth0
| | | | | |-- accept_redirects
| | | | | |-- accept_source_route
| | | | | |-- arp_filter
| | | | | |-- bootp_relay
| | | | | |-- forwarding
| | | | | |-- log_martians
| | | | | |-- mc_forwarding
| | | | | |-- proxy_arp
| | | | | |-- rp_filter
| | | | | |-- secure_redirects
| | | | | |-- send_redirects
| | | | | |-- shared_media
| | | | | `-- tag
| | | | |-- eth1
| | | | | |-- accept_redirects
| | | | | |-- accept_source_route
| | | | | |-- arp_filter
| | | | | |-- bootp_relay
| | | | | |-- forwarding
| | | | | |-- log_martians
| | | | | |-- mc_forwarding
| | | | | |-- proxy_arp
| | | | | |-- rp_filter
| | | | | |-- secure_redirects
| | | | | |-- send_redirects
| | | | | |-- shared_media
| | | | | `-- tag
| | | | `-- lo
| | | | |-- accept_redirects
| | | | |-- accept_source_route
| | | | |-- arp_filter
| | | | |-- bootp_relay
| | | | |-- forwarding
| | | | |-- log_martians
| | | | |-- mc_forwarding
| | | | |-- proxy_arp
| | | | |-- rp_filter
| | | | |-- secure_redirects
| | | | |-- send_redirects
| | | | |-- shared_media
| | | | `-- tag
| | | |-- icmp_echo_ignore_all
| | | |-- icmp_echo_ignore_broadcasts
| | | |-- icmp_ignore_bogus_error_responses
| | | |-- icmp_ratelimit
| | | |-- icmp_ratemask
| | | |-- inet_peer_gc_maxtime
| | | |-- inet_peer_gc_mintime
| | | |-- inet_peer_maxttl
| | | |-- inet_peer_minttl
| | | |-- inet_peer_threshold
| | | |-- ip_autoconfig
| | | |-- ip_conntrack_max
| | | |-- ip_default_ttl
| | | |-- ip_dynaddr
| | | |-- ip_forward
| | | |-- ip_local_port_range
| | | |-- ip_no_pmtu_disc
| | | |-- ip_nonlocal_bind
| | | |-- ipfrag_high_thresh
| | | |-- ipfrag_low_thresh
| | | |-- ipfrag_time
| | | |-- neigh
| | | | |-- default
| | | | | |-- anycast_delay
| | | | | |-- app_solicit
| | | | | |-- base_reachable_time
| | | | | |-- delay_first_probe_time
| | | | | |-- gc_interval
| | | | | |-- gc_stale_time
| | | | | |-- gc_thresh1
| | | | | |-- gc_thresh2
| | | | | |-- gc_thresh3
| | | | | |-- locktime
| | | | | |-- mcast_solicit
| | | | | |-- proxy_delay
| | | | | |-- proxy_qlen
| | | | | |-- retrans_time
| | | | | |-- ucast_solicit
| | | | | `-- unres_qlen
| | | | |-- eth0
| | | | | |-- anycast_delay
| | | | | |-- app_solicit
| | | | | |-- base_reachable_time
| | | | | |-- delay_first_probe_time
| | | | | |-- gc_stale_time
| | | | | |-- locktime
| | | | | |-- mcast_solicit
| | | | | |-- proxy_delay
| | | | | |-- proxy_qlen
| | | | | |-- retrans_time
| | | | | |-- ucast_solicit
| | | | | `-- unres_qlen
| | | | |-- eth1
| | | | | |-- anycast_delay
| | | | | |-- app_solicit
| | | | | |-- base_reachable_time
| | | | | |-- delay_first_probe_time
| | | | | |-- gc_stale_time
| | | | | |-- locktime
| | | | | |-- mcast_solicit
| | | | | |-- proxy_delay
| | | | | |-- proxy_qlen
| | | | | |-- retrans_time
| | | | | |-- ucast_solicit
| | | | | `-- unres_qlen
| | | | `-- lo
| | | | |-- anycast_delay
| | | | |-- app_solicit
| | | | |-- base_reachable_time
| | | | |-- delay_first_probe_time
| | | | |-- gc_stale_time
| | | | |-- locktime
| | | | |-- mcast_solicit
| | | | |-- proxy_delay
| | | | |-- proxy_qlen
| | | | |-- retrans_time
| | | | |-- ucast_solicit
| | | | `-- unres_qlen
| | | |-- route
| | | | |-- error_burst
| | | | |-- error_cost
| | | | |-- flush
| | | | |-- gc_elasticity
| | | | |-- gc_interval
| | | | |-- gc_min_interval
| | | | |-- gc_thresh
| | | | |-- gc_timeout
| | | | |-- max_delay
| | | | |-- max_size
| | | | |-- min_adv_mss
| | | | |-- min_delay
| | | | |-- min_pmtu
| | | | |-- mtu_expires
| | | | |-- redirect_load
| | | | |-- redirect_number
| | | | `-- redirect_silence
| | | |-- tcp_abort_on_overflow
| | | |-- tcp_adv_win_scale
| | | |-- tcp_app_win
| | | |-- tcp_dsack
| | | |-- tcp_ecn
| | | |-- tcp_fack
| | | |-- tcp_fin_timeout
| | | |-- tcp_keepalive_intvl
| | | |-- tcp_keepalive_probes
| | | |-- tcp_keepalive_time
| | | |-- tcp_max_orphans
| | | |-- tcp_max_syn_backlog
| | | |-- tcp_max_tw_buckets
| | | |-- tcp_mem
| | | |-- tcp_orphan_retries
| | | |-- tcp_reordering
| | | |-- tcp_retrans_collapse
| | | |-- tcp_retries1
| | | |-- tcp_retries2
| | | |-- tcp_rfc1337
| | | |-- tcp_rmem
| | | |-- tcp_sack
| | | |-- tcp_stdurg
| | | |-- tcp_syn_retries
| | | |-- tcp_synack_retries
| | | |-- tcp_syncookies
| | | |-- tcp_timestamps
| | | |-- tcp_tw_recycle
| | | |-- tcp_window_scaling
| | | `-- tcp_wmem
| | `-- unix
| | `-- max_dgram_qlen
| |-- proc
| `-- vm
| |-- bdflush
| |-- kswapd
| |-- max-readahead
| |-- min-readahead
| |-- overcommit_memory
| |-- page-cluster
| `-- pagetable_cache
|-- sysvipc
| |-- msg
| |-- sem
| `-- shm
|-- tty
| |-- driver
| | `-- serial
| |-- drivers
| |-- ldisc
| `-- ldiscs
|-- uptime
`-- version
</verb>
</p><p>
In the directory there are also all the tasks using PID as file
names (you have access to all Task information, like path of binary
file, memory used, and so on).
</p>
<p>
The interesting point is that you cannot only see kernel values
(for example, see info about any task or about network options enabled
of your TCP/IP stack) but you are also able to modify some of it,
typically that ones under /proc/sys directory:
</p>
<p>
<verb>
/proc/sys/
acpi
dev
debug
fs
proc
net
vm
kernel
</verb>
</p><sect2>
/proc/sys/kernel
<p>
Below are very important and well-know kernel values, ready to
be modified:
</p>
<p>
<verb>
overflowgid
overflowuid
random
threads-max // Max number of threads, typically 16384
sysrq // kernel hack: you can view istant register values and more
sem
msgmnb
msgmni
msgmax
shmmni
shmall
shmmax
rtsig-max
rtsig-nr
modprobe // modprobe file location
printk
ctrl-alt-del
cap-bound
panic
domainname // domain name of your Linux box
hostname // host name of your Linux box
version // date info about kernel compilation
osrelease // kernel version (i.e. 2.4.5)
ostype // Linux!
</verb>
</p><sect2>
/proc/sys/net
<p>
This can be considered the most useful proc subdirectory. It
allows you to change very important settings for your network kernel
configuration.
</p>
<p>
<verb>
core
ipv4
ipv6
unix
ethernet
802
</verb>
</p><sect3>
/proc/sys/net/core
<p>
Listed below are general net settings, like "netdev_max_backlog"
(typically 300), the length of all your network packets. This value
can limit your network bandwidth when receiving packets, Linux has
to wait up to scheduling time to flush buffers (due to bottom half
mechanism), about 1000/HZ ms
</p>
<p>
<verb>
300 * 100 = 30 000
packets HZ(Timeslice freq) packets/s
30 000 * 1000 = 30 M
packets average (Bytes/packet) throughput Bytes/s
</verb>
</p><p>
If you want to get higher throughput, you need to increase netdev_max_backlog,
by typing:
</p>
<p>
<verb>
echo 4000 &gt; /proc/sys/net/core/netdev_max_backlog
</verb>
</p><p>
Note: Warning for some HZ values: under some architecture (like
alpha or arm-tbox) it is 1000, so you can have 300 MBytes/s of average
throughput.
</p>
<sect3>
/proc/sys/net/ipv4
<p>
"ip_forward", enables or disables ip forwarding in your Linux box.
This is a generic setting for all devices, you can specify each
device you choose.
</p>
<sect4>
/proc/sys/net/ipv4/conf/interface
<p>
I think this is the most useful /proc entry, because it allows
you to change some net settings to support wireless networks (see
<url url="http://www.bertolinux.com" name="Wireless-HOWTO"> for more information).
</p>
<p>
Here are some examples of when you could use this setting:
</p>
<p>
<itemize>
<item>
"forwarding", to enable ip forwarding for your interface
<item>
"proxy_arp", to enable proxy arp feature. For more see Proxy arp
HOWTO under <url url="http://www.tldp.org" name="Linux Documentation Project"> and <url url="http://www.bertolinux.com" name="Wireless-HOWTO"> for proxy arp use in Wireless networks.
<item>
"send_redirects" to avoid interface to send ICMP_REDIRECT (as before,
see <url url="http://www.bertolinux.com" name="Wireless-HOWTO"> for more).
</itemize>
</p><sect>
Linux Multitasking
<sect1>
Overview
<p>
This section will analyze data structures--the mechanism used
to manage multitasking environment under Linux.
</p>
<sect2>
Task States
<p>
A Linux Task can be one of the following states (according to
&lsqb;include/linux.h&rsqb;):
</p>
<p>
<enum>
<item>
TASK_RUNNING, it means that it is in the "Ready List"
<item>
TASK_INTERRUPTIBLE, task waiting for a signal or a resource (sleeping)
<item>
TASK_UNINTERRUPTIBLE, task waiting for a resource (sleeping),
it is in same "Wait Queue"
<item>
TASK_ZOMBIE, task child without father
<item>
TASK_STOPPED, task being debugged
</enum>
</p><sect2>
Graphical Interaction
<p>
<verb>
______________ CPU Available ______________
| | ----------------&gt; | |
| TASK_RUNNING | | Real Running |
|______________| &lt;---------------- |______________|
CPU Busy
| /|&bsol;
Waiting for | | Resource
Resource | | Available
&bsol;|/ |
______________________
| |
| TASK_INTERRUPTIBLE / |
| TASK-UNINTERRUPTIBLE |
|______________________|
Main Multitasking Flow
</verb>
</p><sect1>
Timeslice
<sect2>
PIT 8253 Programming
<p>
Each 10 ms (depending on HZ value) an IRQ0 comes, which helps
us in a multitasking environment. This signal comes from PIC 8259
(in arch 386+) which is connected to PIT 8253 with a clock of 1.19318
MHz.
</p>
<p>
<verb>
_____ ______ ______
| CPU |&lt;------| 8259 |------| 8253 |
|_____| IRQ0 |______| |___/|&bsol;|
|_____ CLK 1.193.180 MHz
// From include/asm/param.h
&num;ifndef HZ
&num;define HZ 100
&num;endif
// From include/asm/timex.h
&num;define CLOCK_TICK_RATE 1193180 /* Underlying HZ */
// From include/linux/timex.h
&num;define LATCH ((CLOCK_TICK_RATE + HZ/2) / HZ) /* For divider */
// From arch/i386/kernel/i8259.c
outb_p(0x34,0x43); /* binary, mode 2, LSB/MSB, ch 0 */
outb_p(LATCH &amp; 0xff , 0x40); /* LSB */
outb(LATCH &gt;&gt; 8 , 0x40); /* MSB */
</verb>
</p><p>
So we program 8253 (PIT, Programmable Interval Timer) with LATCH
= (1193180/HZ) = 11931.8 when HZ=100 (default). LATCH indicates the
frequency divisor factor.
</p>
<p>
LATCH = 11931.8 gives to 8253 (in output) a frequency of 1193180
/ 11931.8 = 100 Hz, so period = 10ms
</p>
<p>
So Timeslice = 1/HZ.
</p>
<p>
With each Timeslice we temporarily interrupt current process
execution (without task switching), and we do some housekeeping work,
after which we'll return back to our previous process.
</p>
<sect2>
Linux Timer IRQ ICA
<p>
<verb>
Linux Timer IRQ
IRQ 0 &lsqb;Timer&rsqb;
|
&bsol;|/
|IRQ0x00_interrupt // wrapper IRQ handler
|SAVE_ALL ---
|do_IRQ | wrapper routines
|handle_IRQ_event ---
|handler() -&gt; timer_interrupt // registered IRQ 0 handler
|do_timer_interrupt
|do_timer
|jiffies++;
|update_process_times
|if (--counter &lt;= 0) &lcub; // if time slice ended then
|counter = 0; // reset counter
|need_resched = 1; // prepare to reschedule
|&rcub;
|do_softirq
|while (need_resched) &lcub; // if necessary
|schedule // reschedule
|handle_softirq
|&rcub;
|RESTORE_ALL
</verb>
</p><p>
Functions can be found under:
</p>
<p>
<itemize>
<item>
IRQ0x00_interrupt, SAVE_ALL &lsqb;include/asm/hw_irq.h&rsqb;
<item>
do_IRQ, handle_IRQ_event &lsqb;arch/i386/kernel/irq.c&rsqb;
<item>
timer_interrupt, do_timer_interrupt &lsqb;arch/i386/kernel/time.c&rsqb;
<item>
do_timer, update_process_times &lsqb;kernel/timer.c&rsqb;
<item>
do_softirq &lsqb;kernel/soft_irq.c&rsqb;
<item>
RESTORE_ALL, while loop &lsqb;arch/i386/kernel/entry.S&rsqb;
</itemize>
</p><p>
Notes:
</p>
<p>
<enum>
<item>
Function "IRQ0x00_interrupt" (like others IRQ0xXY_interrupt) is
directly pointed by IDT (Interrupt Descriptor Table, similar to Real
Mode Interrupt Vector Table, see Cap 11 for more), so EVERY interrupt
coming to the processor is managed by "IRQ0x&num;NR_interrupt" routine,
where &num;NR is the interrupt number. We refer to it as "wrapper
irq handler".
<item>
wrapper routines are executed, like "do_IRQ","handle_IRQ_event" &lsqb;arch/i386/kernel/irq.c&rsqb;.
<item>
After this, control is passed to official IRQ routine (pointed
by "handler()"), previously registered with "request_irq" &lsqb;arch/i386/kernel/irq.c&rsqb;,
in this case "timer_interrupt" &lsqb;arch/i386/kernel/time.c&rsqb;.
<item>
"timer_interrupt" &lsqb;arch/i386/kernel/time.c&rsqb; routine is
executed and, when it ends,
<item>
control backs to some assembler routines &lsqb;arch/i386/kernel/entry.S&rsqb;.
</enum>
</p><p>
Description:
</p>
<p>
To manage Multitasking, Linux (like every other Unix) uses a
''counter'' variable to keep track of how much CPU was used by the
task. So, on each IRQ 0, the counter is decremented (point 4) and,
when it reaches 0, we need to switch task to manage timesharing (point
4 "need_resched" variable is set to 1, then, in point 5 assembler routines
control "need_resched" and call, if needed, "schedule" &lsqb;kernel/sched.c&rsqb;).
</p>
<sect1>
Scheduler
<p>
The scheduler is the piece of code that chooses what Task has
to be executed at a given time.
</p>
<p>
Any time you need to change running task, select a candidate.
Below is the ''schedule &lsqb;kernel/sched.c&rsqb;'' function.
</p>
<p>
<verb>
|schedule
|do_softirq // manages post-IRQ work
|for each task
|calculate counter
|prepare_to__switch // does anything
|switch_mm // change Memory context (change CR3 value)
|switch_to (assembler)
|SAVE ESP
|RESTORE future_ESP
|SAVE EIP
|push future_EIP *** push parameter as we did a call
|jmp __switch_to (it does some TSS work)
|__switch_to()
..
|ret *** ret from call using future_EIP in place of call address
new_task
</verb>
</p><sect1>
Bottom Half, Task Queues. and Tasklets
<sect2>
Overview
<p>
In classic Unix, when an IRQ comes (from a device), Unix makes
"task switching" to interrogate the task that requested the device.
</p>
<p>
To improve performance, Linux can postpone the non-urgent work
until later, to better manage high speed event.
</p>
<p>
This feature is managed since kernel 1.x by the "bottom half" (BH).
The irq handler "marks" a bottom half, to be executed later, in scheduling
time.
</p>
<p>
In the latest kernels there is a "task queue"that is more dynamic
than BH and there is also a "tasklet" to manage multiprocessor environments.
</p>
<p>
BH schema is:
</p>
<p>
<enum>
<item>
Declaration
<item>
Mark
<item>
Execution
</enum>
</p><sect2>
Declaration
<p>
<verb>
&num;define DECLARE_TASK_QUEUE(q) LIST_HEAD(q)
&num;define LIST_HEAD(name) &bsol;
struct list_head name = LIST_HEAD_INIT(name)
struct list_head &lcub;
struct list_head *next, *prev;
&rcub;;
&num;define LIST_HEAD_INIT(name) &lcub; &amp;(name), &amp;(name) &rcub;
''DECLARE_TASK_QUEUE'' &lsqb;include/linux/tqueue.h, include/linux/list.h&rsqb;
</verb>
</p><p>
"DECLARE_TASK_QUEUE(q)" macro is used to declare a structure named
"q" managing task queue.
</p>
<sect2>
Mark
<p>
Here is the ICA schema for "mark_bh" &lsqb;include/linux/interrupt.h&rsqb;
function:
</p>
<p>
<verb>
|mark_bh(NUMBER)
|tasklet_hi_schedule(bh_task_vec + NUMBER)
|insert into tasklet_hi_vec
|__cpu_raise_softirq(HI_SOFTIRQ)
|soft_active |= (1 &lt;&lt; HI_SOFTIRQ)
''mark_bh''&lsqb;include/linux/interrupt.h&rsqb;
</verb>
</p><p>
For example, when an IRQ handler wants to "postpone" some work,
it would "mark_bh(NUMBER)", where NUMBER is a BH declarated (see section
before).
</p>
<sect2>
Execution
<p>
We can see this calling from "do_IRQ" &lsqb;arch/i386/kernel/irq.c&rsqb;
function:
</p>
<p>
<verb>
|do_softirq
|h-&gt;action(h)-&gt; softirq_vec&lsqb;TASKLET_SOFTIRQ&rsqb;-&gt;action -&gt; tasklet_action
|tasklet_vec&lsqb;0&rsqb;.list-&gt;func
</verb>
</p><p>
"h-&gt;action(h);" is the function has been previously queued.
</p>
<sect1>
Very low level routines
<p>
set_intr_gate
</p>
<p>
set_trap_gate
</p>
<p>
set_task_gate (not used).
</p>
<p>
(*interrupt)&lsqb;NR_IRQS&rsqb;(void) = &lcub; IRQ0x00_interrupt,
IRQ0x01_interrupt, ..&rcub;
</p>
<p>
NR_IRQS = 224 &lsqb;kernel 2.4.2&rsqb;
</p>
<sect1>
Task Switching
<sect2>
When does Task switching occur?
<p>
Now we'll see how the Linux Kernel switchs from one task to another.
</p>
<p>
Task Switching is needed in many cases, such as the following:
</p>
<p>
<itemize>
<item>
when TimeSlice ends, we need to give access to some other task
<item>
when a task decide to access a resource, it sleeps for it, so
we have to choose another task
<item>
when a task waits for a pipe, we have to give access to other
task, which would write to pipe
</itemize>
</p><sect2>
Task Switching
<p>
<verb>
TASK SWITCHING TRICK
&num;define switch_to(prev,next,last) do &lcub; &bsol;
asm volatile(&quot;pushl &percnt;&percnt;esi&bsol;n&bsol;t&quot; &bsol;
&quot;pushl &percnt;&percnt;edi&bsol;n&bsol;t&quot; &bsol;
&quot;pushl &percnt;&percnt;ebp&bsol;n&bsol;t&quot; &bsol;
&quot;movl &percnt;&percnt;esp,&percnt;0&bsol;n&bsol;t&quot; /* save ESP */ &bsol;
&quot;movl &percnt;3,&percnt;&percnt;esp&bsol;n&bsol;t&quot; /* restore ESP */ &bsol;
&quot;movl &dollar;1f,&percnt;1&bsol;n&bsol;t&quot; /* save EIP */ &bsol;
&quot;pushl &percnt;4&bsol;n&bsol;t&quot; /* restore EIP */ &bsol;
&quot;jmp __switch_to&bsol;n&quot; &bsol;
&quot;1:&bsol;t&quot; &bsol;
&quot;popl &percnt;&percnt;ebp&bsol;n&bsol;t&quot; &bsol;
&quot;popl &percnt;&percnt;edi&bsol;n&bsol;t&quot; &bsol;
&quot;popl &percnt;&percnt;esi&bsol;n&bsol;t&quot; &bsol;
:&quot;=m&quot; (prev-&gt;thread.esp),&quot;=m&quot; (prev-&gt;thread.eip), &bsol;
&quot;=b&quot; (last) &bsol;
:&quot;m&quot; (next-&gt;thread.esp),&quot;m&quot; (next-&gt;thread.eip), &bsol;
&quot;a&quot; (prev), &quot;d&quot; (next), &bsol;
&quot;b&quot; (prev)); &bsol;
&rcub; while (0)
</verb>
</p><p>
Trick is here:
</p>
<p>
<enum>
<item>
''pushl &percnt;4'' which puts future_EIP into the stack
<item>
''jmp __switch_to'' which execute ''__switch_to'' function, but
in opposite of ''call'' we will return to valued pushed in point
1 (so new Task!)
</enum>
<p>
<verb>
U S E R M O D E K E R N E L M O D E
| | | | | | | |
| | | | Timer | | | |
| | | Normal | IRQ | | | |
| | | Exec |------&gt;|Timer_Int.| | |
| | | | | | .. | | |
| | | &bsol;|/ | |schedule()| | Task1 Ret|
| | | | |_switch_to|&lt;-- | Address |
|__________| |__________| | | | | |
| | |S | |
Task1 Data/Stack Task1 Code | | |w | |
| | T|i | |
| | a|t | |
| | | | | | s|c | |
| | | | Timer | | k|h | |
| | | Normal | IRQ | | |i | |
| | | Exec |------&gt;|Timer_Int.| |n | |
| | | | | | .. | |g | |
| | | &bsol;|/ | |schedule()| | | Task2 Ret|
| | | | |_switch_to|&lt;-- | Address |
|__________| |__________| |__________| |__________|
Task2 Data/Stack Task2 Code Kernel Code Kernel Data/Stack
</verb>
</p><sect1>
Fork
<sect2>
Overview
<p>
Fork is used to create another task. We start from a Task Parent,
and we copy many data structures to Task Child.
</p>
<p>
<verb>
| |
| .. |
Task Parent | |
| | | |
| fork |----------&gt;| CREATE |
| | /| NEW |
|_________| / | TASK |
/ | |
--- / | |
--- / | .. |
/ | |
Task Child /
| | /
| fork |&lt;-/
| |
|_________|
Fork SysCall
</verb>
</p><sect2>
What is not copied
<p>
New Task just created (''Task Child'') is almost equal to Parent
(''Task Parent''), there are only few differences:
</p>
<p>
<enum>
<item>
obviously PID
<item>
child ''fork()'' will return 0, while parent ''fork()'' will
return PID of Task Child, to distinguish them each other in User
Mode
<item>
All child data pages are marked ''READ + EXECUTE'', no "WRITE''
(while parent has WRITE right for its own pages) so, when a write
request comes, a ''Page Fault'' exception is generated which will
create a new independent page: this mechanism is called ''Copy on
Write'' (see Cap.10 for more).
</enum>
</p><sect2>
Fork ICA
<p>
<verb>
|sys_fork
|do_fork
|alloc_task_struct
|__get_free_pages
|p-&gt;state = TASK_UNINTERRUPTIBLE
|copy_flags
|p-&gt;pid = get_pid
|copy_files
|copy_fs
|copy_sighand
|copy_mm // should manage CopyOnWrite (I part)
|allocate_mm
|mm_init
|pgd_alloc -&gt; get_pgd_fast
|get_pgd_slow
|dup_mmap
|copy_page_range
|ptep_set_wrprotect
|clear_bit // set page to read-only
|copy_segments // For LDT
|copy_thread
|childregs-&gt;eax = 0
|p-&gt;thread.esp = childregs // child fork returns 0
|p-&gt;thread.eip = ret_from_fork // child starts from fork exit
|retval = p-&gt;pid // parent fork returns child pid
|SET_LINKS // insertion of task into the list pointers
|nr_threads++ // Global variable
|wake_up_process(p) // Now we can wake up just created child
|return retval
fork ICA
</verb>
<p>
<itemize>
<item>
sys_fork &lsqb;arch/i386/kernel/process.c&rsqb;
<item>
do_fork &lsqb;kernel/fork.c&rsqb;
<item>
alloc_task_struct &lsqb;include/asm/processor.c&rsqb;
<item>
__get_free_pages &lsqb;mm/page_alloc.c&rsqb;
<item>
get_pid &lsqb;kernel/fork.c&rsqb;
<item>
copy_files
<item>
copy_fs
<item>
copy_sighand
<item>
copy_mm
<item>
allocate_mm
<item>
mm_init
<item>
pgd_alloc -&gt; get_pgd_fast &lsqb;include/asm/pgalloc.h&rsqb;
<item>
get_pgd_slow
<item>
dup_mmap &lsqb;kernel/fork.c&rsqb;
<item>
copy_page_range &lsqb;mm/memory.c&rsqb;
<item>
ptep_set_wrprotect &lsqb;include/asm/pgtable.h&rsqb;
<item>
clear_bit &lsqb;include/asm/bitops.h&rsqb;
<item>
copy_segments &lsqb;arch/i386/kernel/process.c&rsqb;
<item>
copy_thread
<item>
SET_LINKS &lsqb;include/linux/sched.h&rsqb;
<item>
wake_up_process &lsqb;kernel/sched.c&rsqb;
</itemize>
</p><sect2>
Copy on Write
<p>
To implement Copy on Write for Linux:
</p>
<p>
<enum>
<item>
Mark all copied pages as read-only, causing a Page Fault when
a Task tries to write to them.
<item>
Page Fault handler creates a new page.
</enum>
<p>
<verb>
| Page
| Fault
| Exception
|
|
-----------&gt; |do_page_fault
|handle_mm_fault
|handle_pte_fault
|do_wp_page
|alloc_page // Allocate a new page
|break_cow
|copy_cow_page // Copy old page to new one
|establish_pte // reconfig Page Table pointers
|set_pte
Page Fault ICA
</verb>
<p>
<itemize>
<item>
do_page_fault &lsqb;arch/i386/mm/fault.c&rsqb;
<item>
handle_mm_fault &lsqb;mm/memory.c&rsqb;
<item>
handle_pte_fault
<item>
do_wp_page
<item>
alloc_page &lsqb;include/linux/mm.h&rsqb;
<item>
break_cow &lsqb;mm/memory.c&rsqb;
<item>
copy_cow_page
<item>
establish_pte
<item>
set_pte &lsqb;include/asm/pgtable-3level.h&rsqb;
</itemize>
</p><sect>
Linux Memory Management
<sect1>
Overview
<p>
Linux uses segmentation + pagination, which simplifies notation.
</p>
<sect2>
Segments
<p>
Linux uses only 4 segments:
</p>
<p>
<itemize>
<item>
2 segments (code and data/stack) for KERNEL SPACE from &lsqb;0xC000
0000&rsqb; (3 GB) to &lsqb;0xFFFF FFFF&rsqb; (4 GB)
<item>
2 segments (code and data/stack) for USER SPACE from &lsqb;0&rsqb;
(0 GB) to &lsqb;0xBFFF FFFF&rsqb; (3 GB)
</itemize>
<p>
<verb>
__
4 GB---&gt;| | |
| Kernel | | Kernel Space (Code + Data/Stack)
| | __|
3 GB---&gt;|----------------| __
| | |
| | |
2 GB---&gt;| | |
| Tasks | | User Space (Code + Data/Stack)
| | |
1 GB---&gt;| | |
| | |
|________________| __|
0x00000000
Kernel/User Linear addresses
</verb>
</p><sect1>
Specific i386 implementation
<p>
Again, Linux implements Pagination using 3 Levels of Paging,
but in i386 architecture only 2 of them are really used:
</p>
<p>
<verb>
------------------------------------------------------------------
L I N E A R A D D R E S S
------------------------------------------------------------------
&bsol;___/ &bsol;___/ &bsol;_____/
PD offset PF offset Frame offset
&lsqb;10 bits&rsqb; &lsqb;10 bits&rsqb; &lsqb;12 bits&rsqb;
| | |
| | ----------- |
| | | Value |----------|---------
| | | | |---------| /|&bsol; | |
| | | | | | | | |
| | | | | | | Frame offset |
| | | | | | &bsol;|/ |
| | | | |---------|&lt;------ |
| | | | | | | |
| | | | | | | x 4096 |
| | | PF offset|_________|------- |
| | | /|&bsol; | | |
PD offset |_________|----- | | | _________|
/|&bsol; | | | | | | |
| | | | &bsol;|/ | | &bsol;|/
_____ | | | ------&gt;|_________| PHYSICAL ADDRESS
| | &bsol;|/ | | x 4096 | |
| CR3 |--------&gt;| | | |
|_____| | ....... | | ....... |
| | | |
Page Directory Page File
Linux i386 Paging
</verb>
</p><sect1>
Memory Mapping
<p>
Linux manages Access Control with Pagination only, so different
Tasks will have the same segment addresses, but different CR3 (register
used to store Directory Page Address), pointing to different Page
Entries.
</p>
<p>
In User mode a task cannot overcome 3 GB limit (0 x C0 00 00
00), so only the first 768 page directory entries are meaningful
(768*4MB = 3GB).
</p>
<p>
When a Task goes in Kernel Mode (by System call or by IRQ) the
other 256 pages directory entries become important, and they point
to the same page files as all other Tasks (which are the same as
the Kernel).
</p>
<p>
Note that Kernel (and only kernel) Linear Space is equal to Kernel
Physical Space, so:
</p>
<p>
<verb>
________________ _____
|Other KernelData|___ | | |
|----------------| | |__| |
| Kernel |&bsol; |____| Real Other |
3 GB ---&gt;|----------------| &bsol; | Kernel Data |
| |&bsol; &bsol; | |
| __|_&bsol;_&bsol;____|__ Real |
| Tasks | &bsol; &bsol; | Tasks |
| __|___&bsol;_&bsol;__|__ Space |
| | &bsol; &bsol; | |
| | &bsol; &bsol;|----------------|
| | &bsol; |Real KernelSpace|
|________________| &bsol;|________________|
Logical Addresses Physical Addresses
</verb>
</p><p>
Linear Kernel Space corresponds to Physical Kernel Space translated
3 GB down (in fact page tables are something like &lcub; "00000000",
"00000001" &rcub;, so they operate no virtualization, they only report
physical addresses they take from linear ones).
</p>
<p>
Notice that you'll not have an "addresses conflict" between Kernel
and User spaces because we can manage physical addresses with Page
Tables.
</p>
<sect1>
Low level memory allocation
<sect2>
Boot Initialization
<p>
We start from kmem_cache_init (launched by start_kernel &lsqb;init/main.c&rsqb;
at boot up).
</p>
<p>
<verb>
|kmem_cache_init
|kmem_cache_estimate
</verb>
</p><p>
kmem_cache_init &lsqb;mm/slab.c&rsqb;
</p>
<p>
kmem_cache_estimate
</p>
<p>
Now we continue with mem_init (also launched by start_kernel&lsqb;init/main.c&rsqb;)
</p>
<p>
<verb>
|mem_init
|free_all_bootmem
|free_all_bootmem_core
</verb>
</p><p>
mem_init &lsqb;arch/i386/mm/init.c&rsqb;
</p>
<p>
free_all_bootmem &lsqb;mm/bootmem.c&rsqb;
</p>
<p>
free_all_bootmem_core
</p>
<sect2>
Run-time allocation
<p>
Under Linux, when we want to allocate memory, for example during
"copy_on_write" mechanism (see Cap.10), we call:
</p>
<p>
<verb>
|copy_mm
|allocate_mm = kmem_cache_alloc
|__kmem_cache_alloc
|kmem_cache_alloc_one
|alloc_new_slab
|kmem_cache_grow
|kmem_getpages
|__get_free_pages
|alloc_pages
|alloc_pages_pgdat
|__alloc_pages
|rmqueue
|reclaim_pages
</verb>
</p><p>
Functions can be found under:
</p>
<p>
<itemize>
<item>
copy_mm &lsqb;kernel/fork.c&rsqb;
<item>
allocate_mm &lsqb;kernel/fork.c&rsqb;
<item>
kmem_cache_alloc &lsqb;mm/slab.c&rsqb;
<item>
__kmem_cache_alloc
<item>
kmem_cache_alloc_one
<item>
alloc_new_slab
<item>
kmem_cache_grow
<item>
kmem_getpages
<item>
__get_free_pages &lsqb;mm/page_alloc.c&rsqb;
<item>
alloc_pages &lsqb;mm/numa.c&rsqb;
<item>
alloc_pages_pgdat
<item>
__alloc_pages &lsqb;mm/page_alloc.c&rsqb;
<item>
rm_queue
<item>
reclaim_pages &lsqb;mm/vmscan.c&rsqb;
</itemize>
</p><p>
TODO: Understand Zones
</p>
<sect1>
Swap
<sect2>
Overview
<p>
Swap is managed by the kswapd daemon (kernel thread).
</p>
<sect2>
kswapd
<p>
As other kernel threads, kswapd has a main loop that wait to
wake up.
</p>
<p>
<verb>
|kswapd
|// initialization routines
|for (;;) &lcub; // Main loop
|do_try_to_free_pages
|recalculate_vm_stats
|refill_inactive_scan
|run_task_queue
|interruptible_sleep_on_timeout // we sleep for a new swap request
|&rcub;
</verb>
<p>
<itemize>
<item>
kswapd &lsqb;mm/vmscan.c&rsqb;
<item>
do_try_to_free_pages
<item>
recalculate_vm_stats &lsqb;mm/swap.c&rsqb;
<item>
refill_inactive_scan &lsqb;mm/vmswap.c&rsqb;
<item>
run_task_queue &lsqb;kernel/softirq.c&rsqb;
<item>
interruptible_sleep_on_timeout &lsqb;kernel/sched.c&rsqb;
</itemize>
</p><sect2>
When do we need swapping?
<p>
Swapping is needed when we have to access a page that is not
in physical memory.
</p>
<p>
Linux uses ''kswapd'' kernel thread to carry out this purpose.
When the Task receives a page fault exception we do the following:
</p>
<p>
<verb>
| Page Fault Exception
| cause by all these conditions:
| a-) User page
| b-) Read or write access
| c-) Page not present
|
|
-----------&gt; |do_page_fault
|handle_mm_fault
|pte_alloc
|pte_alloc_one
|__get_free_page = __get_free_pages
|alloc_pages
|alloc_pages_pgdat
|__alloc_pages
|wakeup_kswapd // We wake up kernel thread kswapd
Page Fault ICA
</verb>
<p>
<itemize>
<item>
do_page_fault &lsqb;arch/i386/mm/fault.c&rsqb;
<item>
handle_mm_fault &lsqb;mm/memory.c&rsqb;
<item>
pte_alloc
<item>
pte_alloc_one &lsqb;include/asm/pgalloc.h&rsqb;
<item>
__get_free_page &lsqb;include/linux/mm.h&rsqb;
<item>
__get_free_pages &lsqb;mm/page_alloc.c&rsqb;
<item>
alloc_pages &lsqb;mm/numa.c&rsqb;
<item>
alloc_pages_pgdat
<item>
__alloc_pages
<item>
wakeup_kswapd &lsqb;mm/vmscan.c&rsqb;
</itemize>
</p><sect>
Linux Networking
<sect1>
How Linux networking is managed?
<p>
There exists a device driver for each kind of NIC. Inside it,
Linux will ALWAYS call a standard high level routing: "netif_rx &lsqb;net/core/dev.c&rsqb;",
which will controls what 3 level protocol the frame belong to, and
it will call the right 3 level function (so we'll use a pointer to
the function to determine which is right).
</p>
<sect1>
TCP example
<p>
We'll see now an example of what happens when we send a TCP packet
to Linux, starting from ''netif_rx &lsqb;net/core/dev.c&rsqb;'' call.
</p>
<sect2>
Interrupt management: "netif_rx"
<p>
<verb>
|netif_rx
|__skb_queue_tail
|qlen++
|* simple pointer insertion *
|cpu_raise_softirq
|softirq_active(cpu) |= (1 &lt;&lt; NET_RX_SOFTIRQ) // set bit NET_RX_SOFTIRQ in the BH vector
</verb>
</p><p>
Functions:
</p>
<p>
<itemize>
<item>
__skb_queue_tail &lsqb;include/linux/skbuff.h&rsqb;
<item>
cpu_raise_softirq &lsqb;kernel/softirq.c&rsqb;
</itemize>
</p><sect2>
Post Interrupt management: "net_rx_action"
<p>
Once IRQ interaction is ended, we need to follow the next part
of the frame life and examine what NET_RX_SOFTIRQ does.
</p>
<p>
We will next call ''net_rx_action &lsqb;net/core/dev.c&rsqb;''
according to "net_dev_init &lsqb;net/core/dev.c&rsqb;".
</p>
<p>
<verb>
|net_rx_action
|skb = __skb_dequeue (the exact opposite of __skb_queue_tail)
|for (ptype = first_protocol; ptype &lt; max_protocol; ptype++) // Determine
|if (skb-&gt;protocol == ptype) // what is the network protocol
|ptype-&gt;func -&gt; ip_rcv // according to ''struct ip_packet_type &lsqb;net/ipv4/ip_output.c&rsqb;''
**** NOW WE KNOW THAT PACKET IS IP ****
|ip_rcv
|NF_HOOK (ip_rcv_finish)
|ip_route_input // search from routing table to determine function to call
|skb-&gt;dst-&gt;input -&gt; ip_local_deliver // according to previous routing table check, destination is local machine
|ip_defrag // reassembles IP fragments
|NF_HOOK (ip_local_deliver_finish)
|ipprot-&gt;handler -&gt; tcp_v4_rcv // according to ''tcp_protocol &lsqb;include/net/protocol.c&rsqb;''
**** NOW WE KNOW THAT PACKET IS TCP ****
|tcp_v4_rcv
|sk = __tcp_v4_lookup
|tcp_v4_do_rcv
|switch(sk-&gt;state)
*** Packet can be sent to the task which uses relative socket ***
|case TCP_ESTABLISHED:
|tcp_rcv_established
|__skb_queue_tail // enqueue packet to socket
|sk-&gt;data_ready -&gt; sock_def_readable
|wake_up_interruptible
*** Packet has still to be handshaked by 3-way TCP handshake ***
|case TCP_LISTEN:
|tcp_v4_hnd_req
|tcp_v4_search_req
|tcp_check_req
|syn_recv_sock -&gt; tcp_v4_syn_recv_sock
|__tcp_v4_lookup_established
|tcp_rcv_state_process
*** 3-Way TCP Handshake ***
|switch(sk-&gt;state)
|case TCP_LISTEN: // We received SYN
|conn_request -&gt; tcp_v4_conn_request
|tcp_v4_send_synack // Send SYN + ACK
|tcp_v4_synq_add // set SYN state
|case TCP_SYN_SENT: // we received SYN + ACK
|tcp_rcv_synsent_state_process
tcp_set_state(TCP_ESTABLISHED)
|tcp_send_ack
|tcp_transmit_skb
|queue_xmit -&gt; ip_queue_xmit
|ip_queue_xmit2
|skb-&gt;dst-&gt;output
|case TCP_SYN_RECV: // We received ACK
|if (ACK)
|tcp_set_state(TCP_ESTABLISHED)
</verb>
</p><p>
Functions can be found under:
</p>
<p>
<itemize>
<item>
net_rx_action &lsqb;net/core/dev.c&rsqb;
<item>
__skb_dequeue &lsqb;include/linux/skbuff.h&rsqb;
<item>
ip_rcv &lsqb;net/ipv4/ip_input.c&rsqb;
<item>
NF_HOOK -&gt; nf_hook_slow &lsqb;net/core/netfilter.c&rsqb;
<item>
ip_rcv_finish &lsqb;net/ipv4/ip_input.c&rsqb;
<item>
ip_route_input &lsqb;net/ipv4/route.c&rsqb;
<item>
ip_local_deliver &lsqb;net/ipv4/ip_input.c&rsqb;
<item>
ip_defrag &lsqb;net/ipv4/ip_fragment.c&rsqb;
<item>
ip_local_deliver_finish &lsqb;net/ipv4/ip_input.c&rsqb;
<item>
tcp_v4_rcv &lsqb;net/ipv4/tcp_ipv4.c&rsqb;
<item>
__tcp_v4_lookup
<item>
tcp_v4_do_rcv
<item>
tcp_rcv_established &lsqb;net/ipv4/tcp_input.c&rsqb;
<item>
__skb_queue_tail &lsqb;include/linux/skbuff.h&rsqb;
<item>
sock_def_readable &lsqb;net/core/sock.c&rsqb;
<item>
wake_up_interruptible &lsqb;include/linux/sched.h&rsqb;
<item>
tcp_v4_hnd_req &lsqb;net/ipv4/tcp_ipv4.c&rsqb;
<item>
tcp_v4_search_req
<item>
tcp_check_req
<item>
tcp_v4_syn_recv_sock
<item>
__tcp_v4_lookup_established
<item>
tcp_rcv_state_process &lsqb;net/ipv4/tcp_input.c&rsqb;
<item>
tcp_v4_conn_request &lsqb;net/ipv4/tcp_ipv4.c&rsqb;
<item>
tcp_v4_send_synack
<item>
tcp_v4_synq_add
<item>
tcp_rcv_synsent_state_process &lsqb;net/ipv4/tcp_input.c&rsqb;
<item>
tcp_set_state &lsqb;include/net/tcp.h&rsqb;
<item>
tcp_send_ack &lsqb;net/ipv4/tcp_output.c&rsqb;
</itemize>
</p><p>
Description:
</p>
<p>
<itemize>
<item>
First we determine protocol type (IP, then TCP)
<item>
NF_HOOK (function) is a wrapper routine that first manages the
network filter (for example firewall), then it calls ''function''.
<item>
After we manage 3-way TCP Handshake which consists of:
</itemize>
<p>
<verb>
SERVER (LISTENING) CLIENT (CONNECTING)
SYN
&lt;-------------------
SYN + ACK
-------------------&gt;
ACK
&lt;-------------------
3-Way TCP handshake
</verb>
<p>
<itemize>
<item>
In the end we only have to launch "tcp_rcv_established &lsqb;net/ipv4/tcp_input.c&rsqb;"
which gives the packet to the user socket and wakes it up.
</itemize>
</p><sect>
Linux File System
<p>
TODO
</p>
<sect>
Useful Tips
<sect1>
Stack and Heap
<sect2>
Overview
<p>
Here we view how "stack" and "heap" are allocated in memory
</p>
<sect2>
Memory allocation
<p>
<verb>
FF.. | | &lt;-- bottom of the stack
/|&bsol; | | |
higher | | | | stack
values | | | &bsol;|/ growing
| |
XX.. | | &lt;-- top of the stack &lsqb;Stack Pointer&rsqb;
| |
| |
| |
00.. |_________________| &lt;-- end of stack &lsqb;Stack Segment&rsqb;
Stack
</verb>
</p><p>
Memory address values start from 00.. (which is also where Stack
Segment begins) and they grow going toward FF.. value.
</p>
<p>
XX.. is the actual value of the Stack Pointer.
</p>
<p>
Stack is used by functions for:
</p>
<p>
<enum>
<item>
global variables
<item>
local variables
<item>
return address
</enum>
</p><p>
For example, for a classical function:
</p>
<p>
<verb>
|int foo_function (parameter_1, parameter_2, ..., parameter_n) &lcub;
|variable_1 declaration;
|variable_2 declaration;
..
|variable_n declaration;
|// Body function
|dynamic variable_1 declaration;
|dynamic variable_2 declaration;
..
|dynamic variable_n declaration;
|// Code is inside Code Segment, not Data/Stack segment!
|return (ret-type) value; // often it is inside some register, for i386 eax register is used.
|&rcub;
we have
| |
| 1. parameter_1 pushed | &bsol;
S | 2. parameter_2 pushed | | Before
T | ................... | | the calling
A | n. parameter_n pushed | /
C | ** Return address ** | -- Calling
K | 1. local variable_1 | &bsol;
| 2. local variable_2 | | After
| ................. | | the calling
| n. local variable_n | /
| |
... ... Free
... ... stack
| |
H | n. dynamic variable_n | &bsol;
E | ................... | | Allocated by
A | 2. dynamic variable_2 | | malloc &amp; kmalloc
P | 1. dynamic variable_1 | /
|_______________________|
Typical stack usage
Note: variables order can be different depending on hardware architecture.
</verb>
</p><sect1>
Application vs Process
<sect2>
Base definition
<p>
We have to distinguish 2 concepts:
</p>
<p>
<itemize>
<item>
Application: that is the useful code we want to execute
<item>
Process: that is the IMAGE on memory of the application (it depends
on memory strategy used, segmentation and/or Pagination).
</itemize>
</p><p>
Often Process is also called Task or Thread.
</p>
<sect1>
Locks
<sect2>
Overview
<p>
2 kind of locks:
</p>
<p>
<enum>
<item>
intraCPU
<item>
interCPU
</enum>
</p><sect1>
Copy_on_write
<p>
Copy_on_write is a mechanism used to reduce memory usage. It
postpones memory allocation until the memory is really needed.
</p>
<p>
For example, when a task executes the "fork()" system call (to
create another task), we still use the same memory pages as the
parent, in read only mode. When a task WRITES into the page, it causes
an exception and the page is copied and marked "rw" (read, write).
</p>
<p>
<verb>
1-) Page X is shared between Task Parent and Task Child
Task Parent
| | RO Access ______
| |----------&gt;|Page X|
|_________| |______|
/|&bsol;
|
Task Child |
| | RO Access |
| |----------------
|_________|
2-) Write request
Task Parent
| | RO Access ______
| |----------&gt;|Page X| Trying to write
|_________| |______|
/|&bsol;
|
Task Child |
| | RO Access |
| |----------------
|_________|
3-) Final Configuration: Either Task Parent and Task Child have an independent copy of the Page, X and Y
Task Parent
| | RW Access ______
| |----------&gt;|Page X|
|_________| |______|
Task Child
| | RW Access ______
| |----------&gt;|Page Y|
|_________| |______|
</verb>
</p><sect>
80386 specific details
<sect1>
Boot procedure
<p>
<verb>
bbootsect.s &lsqb;arch/i386/boot&rsqb;
setup.S (+video.S)
head.S (+misc.c) &lsqb;arch/i386/boot/compressed&rsqb;
start_kernel &lsqb;init/main.c&rsqb;
</verb>
</p><sect1>
80386 (and more) Descriptors
<sect2>
Overview
<p>
Descriptors are data structure used by Intel microprocessor i386+
to virtualize memory.
</p>
<sect2>
Kind of descriptors
<p>
<itemize>
<item>
GDT (Global Descriptor Table)
<item>
LDT (Local Descriptor Table)
<item>
IDT (Interrupt Descriptor Table)
</itemize>
</p><sect>
IRQ
<sect1>
Overview
<p>
IRQ is an asyncronous signal sent to microprocessor to advertise
a requested work is completed
</p>
<sect1>
Interaction schema
<p>
<verb>
|&lt;--&gt; IRQ(0) &lsqb;Timer&rsqb;
|&lt;--&gt; IRQ(1) &lsqb;Device 1&rsqb;
| ..
|&lt;--&gt; IRQ(n) &lsqb;Device n&rsqb;
_____________________________|
/|&bsol; /|&bsol; /|&bsol;
| | |
&bsol;|/ &bsol;|/ &bsol;|/
Task(1) Task(2) .. Task(N)
IRQ - Tasks Interaction Schema
</verb>
</p><sect2>
What happens?
<p>
A typical O.S. uses many IRQ signals to interrupt normal process
execution and does some housekeeping work. So:
</p>
<p>
<enum>
<item>
IRQ (i) occurs and Task(j) is interrupted
<item>
IRQ(i)_handler is executed
<item>
control backs to Task(j) interrupted
</enum>
</p><p>
Under Linux, when an IRQ comes, first the IRQ wrapper routine
(named "interrupt0x??") is called, then the "official" IRQ(i)_handler
will be executed. This allows some duties like timeslice preemption.
</p>
<sect>
Utility functions
<sect1>
list_entry &lsqb;include/linux/list.h&rsqb;
<p>
Definition:
</p>
<p>
<verb>
&num;define list_entry(ptr, type, member) &bsol;
((type *)((char *)(ptr)-(unsigned long)(&amp;((type *)0)-&gt;member)))
</verb>
</p><p>
Meaning:
</p>
<p>
"list_entry" macro is used to retrieve a parent struct pointer,
by using only one of internal struct pointer.
</p>
<p>
Example:
</p>
<p>
<verb>
struct __wait_queue &lcub;
unsigned int flags;
struct task_struct * task;
struct list_head task_list;
&rcub;;
struct list_head &lcub;
struct list_head *next, *prev;
&rcub;;
// and with type definition:
typedef struct __wait_queue wait_queue_t;
// we'll have
wait_queue_t *out list_entry(tmp, wait_queue_t, task_list);
// where tmp point to list_head
</verb>
</p><p>
So, in this case, by means of *tmp pointer &lsqb;list_head&rsqb;
we retrieve an *out pointer &lsqb;wait_queue_t&rsqb;.
</p>
<p>
<verb>
____________ &lt;---- *out &lsqb;we calculate that&rsqb;
|flags | /|&bsol;
|task *--&gt; | |
|task_list |&lt;---- list_entry
| prev * --&gt;| | |
| next * --&gt;| | |
|____________| ----- *tmp &lsqb;we have this&rsqb;
</verb>
</p><sect1>
Sleep
<sect2>
Sleep code
<p>
Files:
</p>
<p>
<itemize>
<item>
kernel/sched.c
<item>
include/linux/sched.h
<item>
include/linux/wait.h
<item>
include/linux/list.h
</itemize>
</p><p>
Functions:
</p>
<p>
<itemize>
<item>
interruptible_sleep_on
<item>
interruptible_sleep_on_timeout
<item>
sleep_on
<item>
sleep_on_timeout
</itemize>
</p><p>
Called functions:
</p>
<p>
<itemize>
<item>
init_waitqueue_entry
<item>
__add_wait_queue
<item>
list_add
<item>
__list_add
<item>
__remove_wait_queue
</itemize>
</p><p>
InterCallings Analysis:
</p>
<p>
<verb>
|sleep_on
|init_waitqueue_entry --
|__add_wait_queue | enqueuing request to resource list
|list_add |
|__list_add --
|schedule --- waiting for request to be executed
|__remove_wait_queue --
|list_del | dequeuing request from resource list
|__list_del --
</verb>
</p><p>
Description:
</p>
<p>
Under Linux each resource (ideally an object shared between many
users and many processes), , has a queue to manage ALL tasks requesting
it.
</p>
<p>
This queue is called "wait queue" and it consists of many items
we'll call the"wait queue element":
</p>
<p>
<verb>
*** wait queue structure &lsqb;include/linux/wait.h&rsqb; ***
struct __wait_queue &lcub;
unsigned int flags;
struct task_struct * task;
struct list_head task_list;
&rcub;
struct list_head &lcub;
struct list_head *next, *prev;
&rcub;;
</verb>
</p><p>
Graphic working:
</p>
<p>
<verb>
*** wait queue element ***
/|&bsol;
|
&lt;--&lsqb;prev *, flags, task *, next *&rsqb;--&gt;
*** wait queue list ***
/|&bsol; /|&bsol; /|&bsol; /|&bsol;
| | | |
--&gt; &lt;--&lsqb;task1&rsqb;--&gt; &lt;--&lsqb;task2&rsqb;--&gt; &lt;--&lsqb;task3&rsqb;--&gt; .... &lt;--&lsqb;taskN&rsqb;--&gt; &lt;--
| |
|__________________________________________________________________|
*** wait queue head ***
task1 &lt;--&lsqb;prev *, lock, next *&rsqb;--&gt; taskN
</verb>
</p><p>
"wait queue head" point to first (with next *) and last (with prev
*) elements of the "wait queue list".
</p>
<p>
When a new element has to be added, "__add_wait_queue" &lsqb;include/linux/wait.h&rsqb;
is called, after which the generic routine "list_add" &lsqb;include/linux/wait.h&rsqb;,
will be executed:
</p>
<p>
<verb>
*** function list_add &lsqb;include/linux/list.h&rsqb; ***
// classic double link list insert
static __inline__ void __list_add (struct list_head * new, &bsol;
struct list_head * prev, &bsol;
struct list_head * next) &lcub;
next-&gt;prev = new;
new-&gt;next = next;
new-&gt;prev = prev;
prev-&gt;next = new;
&rcub;
</verb>
</p><p>
To complete the description, we see also "__list_del" &lsqb;include/linux/list.h&rsqb;
function called by "list_del" &lsqb;include/linux/list.h&rsqb; inside
"remove_wait_queue" &lsqb;include/linux/wait.h&rsqb;:
</p>
<p>
<verb>
*** function list_del &lsqb;include/linux/list.h&rsqb; ***
// classic double link list delete
static __inline__ void __list_del (struct list_head * prev, struct list_head * next) &lcub;
next-&gt;prev = prev;
prev-&gt;next = next;
&rcub;
</verb>
</p><sect2>
Stack consideration
<p>
A typical list (or queue) is usually managed allocating it into
the Heap (see Cap.10 for Heap and Stack definition and about where
variables are allocated). Otherwise here, we statically allocate
Wait Queue data in a local variable (Stack), then function is interrupted
by scheduling, in the end, (returning from scheduling) we'll erase
local variable.
</p>
<p>
<verb>
new task &lt;----| task1 &lt;------| task2 &lt;------|
| | |
| | |
|..........| | |..........| | |..........| |
|wait.flags| | |wait.flags| | |wait.flags| |
|wait.task_|____| |wait.task_|____| |wait.task_|____|
|wait.prev |--&gt; |wait.prev |--&gt; |wait.prev |--&gt;
|wait.next |--&gt; |wait.next |--&gt; |wait.next |--&gt;
|.. | |.. | |.. |
|schedule()| |schedule()| |schedule()|
|..........| |..........| |..........|
|__________| |__________| |__________|
Stack Stack Stack
</verb>
</p><sect>
Static variables
<sect1>
Overview
<p>
Linux is written in ''C'' language, and as every application
has:
</p>
<p>
<enum>
<item>
Local variables
<item>
Module variables (inside the source file and relative only to
that module)
<item>
Global/Static variables present in only 1 copy (the same for
all modules)
</enum>
</p><p>
When a Static variable is modified by a module, all other modules
will see the new value.
</p>
<p>
Static variables under Linux are very important, cause they are
the only kind to add new support to kernel: they typically are pointers
to the head of a list of registered elements, which can be:
</p>
<p>
<itemize>
<item>
added
<item>
deleted
<item>
maybe modified
</itemize>
<p>
<verb>
_______ _______ _______
Global variable -------&gt; |Item(1)| -&gt; |Item(2)| -&gt; |Item(3)| ..
|_______| |_______| |_______|
</verb>
</p><sect1>
Main variables
<sect2>
Current
<p>
<verb>
________________
Current ----------------&gt; | Actual process |
|________________|
</verb>
</p><p>
Current points to ''task_struct'' structure, which contains all
data about a process like:
</p>
<p>
<itemize>
<item>
pid, name, state, counter, policy of scheduling
<item>
pointers to many data structures like: files, vfs, other processes,
signals...
</itemize>
</p><p>
Current is not a real variable, it is
</p>
<p>
<verb>
static inline struct task_struct * get_current(void) &lcub;
struct task_struct *current;
__asm__(&quot;andl &percnt;&percnt;esp,&percnt;0; &quot;:&quot;=r&quot; (current) : &quot;0&quot; (&tilde;8191UL));
return current;
&rcub;
&num;define current get_current()
</verb>
</p><p>
Above lines just takes value of ''esp'' register (stack pointer)
and get it available like a variable, from which we can point to
our task_struct structure.
</p>
<p>
From ''current'' element we can access directly to any other
process (ready, stopped or in any other state) kernel data structure,
for example changing STATE (like a I/O driver does), PID, presence
in ready list or blocked list, etc.
</p>
<sect2>
Registered filesystems
<p>
<verb>
______ _______ ______
file_systems ------&gt; | ext2 | -&gt; | msdos | -&gt; | ntfs |
&lsqb;fs/super.c&rsqb; |______| |_______| |______|
</verb>
</p><p>
When you use command like ''modprobe some_fs'' you will add a
new entry to file systems list, while removing it (by using ''rmmod'')
will delete it.
</p>
<sect2>
Mounted filesystems
<p>
<verb>
______ _______ ______
mount_hash_table ----&gt;| / | -&gt; | /usr | -&gt; | /var |
&lsqb;fs/namespace.c&rsqb; |______| |_______| |______|
</verb>
</p><p>
When you use ''mount'' command to add a fs, the new entry will
be inserted in the list, while an ''umount'' command will delete
the entry.
</p>
<sect2>
Registered Network Packet Type
<p>
<verb>
______ _______ ______
ptype_all ------&gt;| ip | -&gt; | x25 | -&gt; | ipv6 |
&lsqb;net/core/dev.c&rsqb; |______| |_______| |______|
</verb>
</p><p>
For example, if you add support for IPv6 (loading relative module)
a new entry will be added in the list.
</p>
<sect2>
Registered Network Internet Protocol
<p>
<verb>
______ _______ _______
inet_protocol_base -----&gt;| icmp | -&gt; | tcp | -&gt; | udp |
&lsqb;net/ipv4/protocol.c&rsqb; |______| |_______| |_______|
</verb>
</p><p>
Also others packet type have many internal protocols in each
list (like IPv6).
</p>
<p>
<verb>
______ _______ _______
inet6_protos -----------&gt;|icmpv6| -&gt; | tcpv6 | -&gt; | udpv6 |
&lsqb;net/ipv6/protocol.c&rsqb; |______| |_______| |_______|
</verb>
</p><sect2>
Registered Network Device
<p>
<verb>
______ _______ _______
dev_base ---------------&gt;| lo | -&gt; | eth0 | -&gt; | ppp0 |
&lsqb;drivers/core/Space.c&rsqb; |______| |_______| |_______|
</verb>
</p><sect2>
Registered Char Device
<p>
<verb>
______ _______ ________
chrdevs ----------------&gt;| lp | -&gt; | keyb | -&gt; | serial |
&lsqb;fs/devices.c&rsqb; |______| |_______| |________|
</verb>
</p><p>
''chrdevs'' is not a pointer to a real list, but it is a standard
vector.
</p>
<sect2>
Registered Block Device
<p>
<verb>
______ ______ ________
bdev_hashtable ---------&gt;| fd | -&gt; | hd | -&gt; | scsi |
&lsqb;fs/block_dev.c&rsqb; |______| |______| |________|
</verb>
</p><p>
''bdev_hashtable'' is an hash vector.
</p>
<sect>
Glossary
<sect>
Links
<p>
<url url="http://www.kernel.org" name="Official Linux kernels and patches download site">
</p>
<p>
<url url="http://jungla.dit.upm.es/~jmseyas/linux/kernel/hackers-docs.html" name="Great documentation about Linux Kernel">
</p>
<p>
<url url="http://www.uwsg.indiana.edu/hypermail/linux/kernel/index.html" name="Official Kernel Mailing list">
</p>
<p>
<url url="http://www.tldp.org/guides.html" name="Linux Documentation Project Guides">
</p>
</article>
View Git Blame Copy Permalink