Subversion Repositories or1k

This file describes the strategy for dynamically loadable modules

in the Linux kernel. This is not a technical description on

the internals of module, but mostly a sample of how to compile

and use modules.

Note: You should ensure that the modules-X.Y.Z.tar.gz you are using

is the most up to date one for this kernel. The "X.Y.Z" will reflect

the kernel version at the time of the release of the modules package.

Some older modules packages aren't aware of some of the newer modular

features that the kernel now supports. (If you are unsure, you can

usually find out what the current release of the modules-X.Y.Z.tar.gz

package is by looking up the URL listed for "Bjorn Ekwall" in the

file ./linux/CREDITS)

In the beginning...

-------------------

Anyway, your first step is to compile the kernel, as explained in the

file linux/README.  It generally goes like:

        make config

        make dep

        make clean

        make zImage or make bzImage or make zlilo

In "make config", you select what you want to include in the "resident"

kernel and what features you want to have available as loadable modules.

You will generally select the minimal resident set that is needed to boot:

        The filesystem of your root partition

        A scsi driver, but see below for a list of SCSI modules!

        Normal hard drive support

        Net support (CONFIG_NET)

        TCP/IP support (CONFIG_INET), but no drivers!

        plus those things that you just can't live without...

The set of modules is constantly increasing, and you will be able to select

the option "m" in "make config" for those features that the current kernel

can offer as loadable modules.

You also have a possibility to create modules that are less dependent on

the kernel version.  This option can be selected during "make config", by

enabling CONFIG_MODVERSIONS, and is most useful on "stable" kernel versions,

such as the kernels from the 1.2 and 2.0 series.

If you have modules that are based on sources that are not included in

the official kernel sources, you will certainly like this option...

Here is a sample of the available modules included in the kernel sources:

        Most filesystems: minix, xiafs, msdos, umsdos, sysv, isofs, hpfs,

                          smbfs, nfs

        Mid-level SCSI support (required by top and low level scsi drivers).

        Most low-level SCSI drivers: (i.e. aha1542, in2000)

        All SCSI high-level drivers: disk, tape, cdrom, generic.

        Most ethernet drivers: (too many to list, please see the file

                                ./Documentation/networking/net-modules.txt)

        Most CDROM drivers:

                aztcd:     Aztech,Orchid,Okano,Wearnes

                cm206:     Philips/LMS CM206

                gscd:      Goldstar GCDR-420

                mcd, mcdx: Mitsumi LU005, FX001

                optcd:     Optics Storage Dolphin 8000AT

                sjcd:      Sanyo CDR-H94A

                sbpcd:     Matsushita/Panasonic CR52x, CR56x, CD200,

                           Longshine LCS-7260, TEAC CD-55A

                sonycd535: Sony CDU-531/535, CDU-510/515

        And a lot of misc modules, such as:

                lp: line printer

                binfmt_elf: elf loader

                binfmt_java: java loader

                isp16: cdrom interface

                serial: the serial (tty) interface

When you have made the kernel, you create the modules by doing:

        make modules

This will compile all modules and update the linux/modules directory.

In this directory you will then find a bunch of symbolic links,

pointing to the various object files in the kernel tree.

Now, after you have created all your modules, you should also do:

        make modules_install

This will copy all newly made modules into subdirectories under

"/lib/modules/kernel_release/", where "kernel_release" is something

like 2.0.1, or whatever the current kernel version is...

As soon as you have rebooted the newly made kernel, you can install

and remove modules at will with the utilities: "insmod" and "rmmod".

After reading the man-page for insmod, you will also know how easy

it is to configure a module when you do "insmod" (hint: symbol=value).

Nifty features:

---------------

You also have access to two utilities: "modprobe" and "depmod", where

modprobe is a "wrapper" for (or extension to) "insmod".

These utilities use (and maintain) a set of files that describe all the

modules that are available for the current kernel in the /lib/modules

hierarchy as well as their interdependencies.

Using the modprobe utility, you can load any module like this:

        /sbin/modprobe module

without paying much attention to which kernel you are running, or what

other modules this module depends on.

With the help of the modprobe configuration file: "/etc/conf.modules"

you can tune the behaviour of modprobe in many ways, including an

automatic setting of insmod options for each module.

And, yes, there _are_ man-pages for all this...

To use modprobe successfully, you generally place the following

command in your /etc/rc.d/rc.S script.  (Read more about this in the

"rc.hints" file in the module utilities package, "modules-x.y.z.tar.gz".)

        /sbin/depmod -a

This computes the dependencies between the different modules.

Then if you do, for example

        /sbin/modprobe umsdos

you will automatically load _both_ the msdos and umsdos modules,

since umsdos runs piggyback on msdos.

The "ultimate" utility:

-----------------------

OK, you have read all of the above, and feel amply impressed...

Now, we tell you to forget all about how to install and remove

loadable modules...

With the kerneld daemon, all of these chores will be taken care of

automatically.  Just answer "Y" to CONFIG_KERNELD in "make config",

and make sure that /sbin/kerneld is started as soon as possible

after boot and that "/sbin/depmod -a" has been executed for the

current kernel. (Read more about this in the module utilities package.)

Whenever a program wants the kernel to use a feature that is only

available as a loadable module, and if the kernel hasn't got the

module installed yet, the kernel will ask the kerneld daemon to take

care of the situation and make the best of it.

This is what happens:

        - The kernel notices that a feature is requested that is not

          resident in the kernel.

        - The kernel sends a message to kerneld, with a symbolic

          description of the requested feature.

        - The kerneld daemon asks e.g. modprobe to load a module that

          fits this symbolic description.

        - modprobe looks into its internal "alias" translation table

          to see if there is a match.  This table can be reconfigured

          and expanded by having "alias" lines in "/etc/conf.modules".

        - insmod is then asked to insert the module(s) that modprobe

          has decided that the kernel needs.  Every module will be

          configured according to the "options" lines in "/etc/conf.modules".

        - modprobe exits and kerneld tells the kernel that the request

          succeeded (or failed...)

        - The kernel uses the freshly installed feature just as if it

          had been configured into the kernel as a "resident" part.

The icing of the cake is that when an automatically installed module

has been unused for a period of time (usually 1 minute), the module

will be automatically removed from the kernel as well.

This makes the kernel use the minimal amount of memory at any given time,

making it available for more productive use than as just a placeholder for

unused code.

Actually, this is only a side-effect from the _real_ benefit of kerneld:

You only have to create a minimal kernel, that is more or less independent

of the actual hardware setup.  The setup of the "virtual" kernel is

instead controlled by a configuration file as well as the actual usage

pattern of the current machine and its kernel.

This should be good news for maintainers of multiple machines as well as

for maintainers of distributions.

To use kerneld with the least amount of "hassle", you need modprobe from

a release that can be considered "recent" w.r.t. your kernel, and also

a configuration file for modprobe ("/etc/conf.modules").

Since modprobe already knows about most modules, the minimal configuration

file could look something like this:

        alias scsi_hostadapter aha1542  # or whatever SCSI adapter you have

        alias eth0 3c509        # or whatever net adapter you have

        # you might need an "options" line for some net adapters:

        options 3c509 io=0x300 irq=10

        # you might also need an "options" line for some other module:

        options cdu31a cdu31a_port=0x1f88 sony_pas_init=1

You could add these lines as well, but they are only "cosmetic":

        alias net-pf-3 off      # no ax25 module available (yet)

        alias net-pf-4 off      # if you don't use the ipx module

        alias net-pf-5 off      # if you don't use the appletalk module

Finally, for the "purists":

You can name the modprobe configuration either "/etc/conf.modules" or

"/etc/modules.conf", since modprobe knows what to do in each case...

Written by:

        Jacques Gelinas 

        Bjorn Ekwall