root/lm-sensors/tags/V2-10-3/BACKGROUND

This package consists of many parts. This document offers a small tour
around the most important pieces. Not all aspects are discussed 
exhaustively, but it should be enough to give you the feeling you
understand what you are seeing happen.

When you want to communicate with a piece of hardware, you need a kernel
driver (well, that is not quite true, but it is in most cases the only
way to do it safely). In the past, this meant you had to patch the kernel
and recompile it. These days, you can use kernel modules. We support both
approaches.

We have chooses for a very modular (no pun intended) setup. There are a
few general-purpose base kernel modules, which you always need. In 
addition, there is one kernel module for each piece of hardware - whether
this is an I2C bus adapter, an SMBus adapter or a sensors chip.

The kernel modules communicate their information through both the /proc
and sysctl interfaces (Linux 2.4) or the /sys interface (Linux 2.6). To
keep things uncomplicated, the sensor chips always advert their
measurements 'as is'. This means that the values they report are not
immediately relevant to you - they must first be scaled and translated
to correspond to the real world.

It is also possible to communicate directly with chips on an I2C bus
or SMBus. This is done through /dev files. This is useful if you 
quickly want to test how a certain chip behaves, without having to
write a kernel driver.

Note that all other parts of this package function in so-called user-space.
This is important, because bugs in kernel-space might crash your
computer or do other bad things. And kernel memory can not be swapped out!

Applications could (and can) directly read the sensor values through the 
/proc, sysctl or /sys interfaces. This is harder than it sounds; because
no two chips are the same, the information they communicate may also
be very unlike. This would mean that every application would have to know
about every type of chip. But there is a better solution.

libsensors is a (shared or static) library of access functions. It
knows about every type of chip supported by the kernel modules (or it
should, if it is up-to-date). It offers a simple-to-use interface for
applications to access the sensor chip readings, to set new limits,
and all other commonly needed things. From the application's point of
view, there is no need to know very much about a specific sensors chip. 
Having some inside information can still be useful, but it is possible to
write a generic fall-back function that takes care of newer, unknown
chips, and to display all really important information.

libsensors takes care of one other thing. The kernel modules report 
so-called 'as is' values. They have to be scaled or translated to be 
relevant in the real world. libsensors reads a configuration file 
(usually /etc/sensors.conf) which specifies how this translation should
be done (with some other things). Again, the application does not have
to know about it. And because the configuration file is reread each time
a new application is started, you can change configuration values without
having to recompile anything.

This package does not contain a nice graphical monitor. Look at the file
doc/useful_addresses.html for pointers to such programs. It does contain
an example console program that reports all current sensors values. This
program is called 'sensors'. You can use it as a reference implementation
for more intricate programs.

There are many, many kernel modules in this package, and there are lots
of different sensor chips supported. Sometimes, it can be hard to 
determine what chips and adapters you have, and which modules correspond
to them. Fortunately, there is a user-space application 'sensors-detect'
that should tell you exactly what is available, and what you need to
do. This perl script uses the /dev interface, and you may use it as an
example how you can do this.