bri@cs.uchicago.edu
Should you read this document? Well, see if you've got any of the following symptoms:
Some of the examples in this document assume that you have GNU tar,
find, and xargs. These are quite standard; this should
not cause problems. It is also assumed that
you know your system's filesystem structure; if you don't, it is critical
that you keep a written copy of the mount command's output during
normal system operation (or a listing of /etc/fstab, if you can
read it). This information is important, and does not change unless you
repartition your disk, add a new one, reinstall your system, or something
similar.
The latest ``production'' kernel version at the time of this writing was 2.2.9, meaning that the references and examples correspond to that release. Even though I try to make this document as version-independent as possible, the kernel is constantly under development, so if you get a newer release, it will inevitably have some differences. Again, this should not cause major problems, but it may create some confusion.
There are two versions of the linux kernel source, ``production'' and ``development.'' Production releases are the even-minor-numbered releases; 1.2.x was production, 2.0.x is production, as well as 2.2.x. These kernels are considered to be the most stable, bug-free versions available at the time of release. The development kernels (2.1.x, 2.3.x, etc) are meant as testing kernels, for people willing to test out new and possibly very buggy kernels. You have been warned.
Text that looks like this is either something that appears on
your screen, a filename, or something that can be directly typed in, such as a
command, or options to a command (if you're looking
at a plain-text file, it doesn't look any different). Commands and other
input are frequently quoted (with ` '), which causes the following
classic punctuation problem: if such an item appears at the end of a sentence
in quotes, people often type a `.' along with the command, because the
American quoting style says to put the period inside of the quotation
marks. Even though common sense (and unfortunately, this assumes that the
one with the ``common sense'' is used to the so-called American style of
quotation) should tell one to strip off the
punctuation first, many people simply do not remember, so I will place
it outside the quotation marks in such cases. In other words, when
indicating that you should type ``make config'' I would write
`make config', not `make config.'
The Unix kernel acts as a mediator for your programs and your hardware. First, it does (or arranges for) the memory management for all of the running programs (processes), and makes sure that they all get a fair (or unfair, if you please) share of the processor's cycles. In addition, it provides a nice, fairly portable interface for programs to talk to your hardware.
There is certainly more to the kernel's operation than this, but these basic functions are the most important to know.
Newer kernels generally offer the ability to talk to more types of hardware (that is, they have more device drivers), they can have better process management, they can run faster than the older versions, they could be more stable than the older versions, and they fix silly bugs in the older versions. Most people upgrade kernels because they want the device drivers and the bug fixes.
See the Hardware-HOWTO. Alternatively, you can look at the
`config.in' file in the linux source, or just find out
when you try `make config'.
This shows you all hardware supported by the
standard kernel distribution, but not everything that linux supports; many
common device drivers (such as the PCMCIA drivers and some tape drivers) are
loadable modules maintained and distributed separately.
Linus recommends a version of gcc in the README file included with
the linux source. If you don't have this version, the documentation in the
recommended version of gcc should tell you if you need to upgrade your libc.
This is not a difficult procedure, but it is important to follow the
instructions.
These are pieces of kernel code which are not linked (included) directly in the kernel. One compiles them separately, and can insert and remove them into the running kernel at almost any time. Due to its flexibility, this is now the preferred way to code certain kernel features. Many popular device drivers, such as the PCMCIA drivers and the QIC-80/40 tape driver, are loadable modules.
It depends on your particular system configuration. First, the compressed linux source is nearly 14 megabytes large at version 2.2.9. Many sites keep this even after unpacking. Uncompressed and built with a moderate configuration, it takes up another 67 MB.
With newer machines, the compilation takes dramatically less time than older ones; an AMD K6-2/300 with a fast disk can do a 2.2.x kernel in about four minutes. As for old Pentiums, 486s, and 386s, if you plan to compile one, be prepared to wait, possibly hours, days..
If this troubles you, and you happen to have a faster machine around to compile on, you can build on the fast machines (assuming you give it the right parameters, that your ulilities are up-to-date, and so on), and then transfer the kernel image to the slower machine.
You can obtain the source via anonymous ftp from ftp.kernel.org in
/pub/linux/kernel/vx.y, where x.y is the version (eg 2.2),
and as mentioned before, the ones that end with an odd number are
development releases and may be unstable.
It is typically labelled linux-x.y.z.tar.gz, where x.y.z
is the version number. The sites also typically carry ones with a suffix of
.bz2, which have been compressed with bzip2 (these files will be
smaller and take less time to transfer).
It's best to use ftp.xx.kernel.org where xx is your
country code; examples being ftp.at.kernel.org for Austria,
and ftp.us.kernel.org for the United States.
Log in as or su to `root', and cd to
/usr/src.
If you installed kernel source when you first installed linux (as most do),
there will already be a directory called `linux' there, which
contains the entire old source tree.
If you have the disk space and you want to play it safe, preserve that
directory. A good idea is to figure out
what version your system runs now and rename the directory
accordingly. The command `uname -r' prints the current
kernel version.
Therefore, if `uname -r' said `1.0.9', you would
rename (with `mv') `linux' to `linux-1.0.9'.
If you feel mildly reckless, just wipe out the entire
directory. In any case, make certain there is no `linux' directory
in /usr/src before unpacking the full source code.
Now, in /usr/src, unpack the source with
`tar zxpvf linux-x.y.z.tar.gz'
(if you've just got a .tar file with no .gz at the end,
`tar xpvf linux-x.y.z.tar' works.).
The contents of the source will fly by. When finished, there will be
a new `linux' directory in /usr/src. cd to
linux and look over the README file.
There will be a section with the label `INSTALLING the kernel'.
Carry out the instructions when appropriate -- symbolic links that should
be in place, removal of stale .o files, etc.
If you have a .bz2 file and the bzip2 program (read about it at
http://www.muraroa.demon.co.uk/), do this:
bz2cat linux-x.y.z.tar.bz2 | tar xvf -
Note: Some of this is reiteration/clarification of a similar
section in Linus' README file.
The command `make config' while in /usr/src/linux starts
a configure script which asks you many questions. It requires bash,
so verify that bash is /bin/bash, /bin/sh,
or $BASH.
However, there are some much more pleasant alternatives to `make
config' and you may very well find them easier and more comfortable to
use.
`make menuconfig' is probably the most widely-used. Whatever you
choose, it's best to get familiar with the interface because you may find
yourself back at it sooner than you think.
For those ``running X,'' you can try `make xconfig' if you have Tk
installed (`click-o-rama' - Nat). `make menuconfig' is for those
who have
(n)curses and would prefer a text-based menu. These interfaces have a rather
clear advantage: If you goof up and make a
wrong choice during configuration, it is simple to go back and fix it.
The configuration options will appear in hierarchies with `make
menuconfig' and `make xconfig'.
You are ready to answer the questions, usually with `y' (yes) or
`n' (no). Device drivers typically have an `m' option.
This means ``module,'' meaning that the system will compile it, but not
directly into the kernel, but as a loadable module. A more comical way to
describe it is as ``maybe.'' Some of the
more obvious and non-critical options are not described here; see the section
``Other configuration options'' for short descriptions of a few others.
With `make menuconfig', the space bar toggles the selection.
In 2.0.x and later, there is a `?' option, which provides a brief description of the configuration parameter. That information is likely to be the most up-to-date. Here are a listing of some of the important features, which hierarchy they are in, and brief description.
If you don't have a math coprocessor (you have a bare 386 or
486SX), you must say `y' to this. If you do have a coprocessor and
you still say `y', don't worry too much -- the coprocessor is
still used and the emulation ignored. For any halfway modern machine, the
answer will be no, but don't worry if you say yes accidentally; if not
needed, it is not used.
You probably need to support this; it means that the kernel will support standard PC hard disks, which most people have. This driver does not include SCSI drives; they come later in the configuration.
You will then be asked about the ``old disk-only'' and ``new IDE'' drivers. You want to choose one of them; the main difference is that the old driver only supports two disks on a single interface, and the new one supports a secondary interface and IDE/ATAPI cdrom drives. The new driver is 4k larger than the old one and is also supposedly ``improved,'' meaning that aside from containing a different number of bugs, it might improve your disk performance, especially if you have newer (EIDE-type) hardware.
In principle, you would only say `y' if your machine is on a network
such as the internet, or you want to use SLIP, PPP, term, etc to
dial up for internet access. However, as many packages (such as the X
window system)
require networking support even if your machine does not live on a real
network, you should say `y'. Later on, you will be asked if you
want to support TCP/IP networking; again, say `y' here if you
are not absolutely sure.
One of the best definitions of IPC (Interprocess Communication) is in the
Perl book's glossary. Not surprisingly, some Perl programmers employ it to
let processes talk to each other, as well as many other packages (DOOM,
most notably), so it is not a good idea to say n unless you know
exactly what you are doing.
(in older kernels: Use -m486 flag for 486-specific optimizations)
Traditionally, this compiled in certain optimizations for a particular processor; the kernels ran fine on other chips, but the kernel was perhaps a bit larger. In newer kernels, however, this is no longer true, so you should enter the processor for which you are compiling the kernel. A ``386'' kernel will work on all machines.
If you have SCSI devices, say `y'. You will be prompted for
further information, such as support for CD-ROM, disks, and what kind
of SCSI adapter you have. See the SCSI-HOWTO for greater detail.
If you have a network card, or you would like to use SLIP, PPP, or a
parallel port adapter for connecting to the Internet,
say `y'. The config script will prompt
for which kind of card you have, and which protocol to use.
The configure script then asks if you wish to support the following filesystems:
Standard (minix) - Newer distributions don't create minix filesystems, and many people don't use it, but it may still be a good idea to configure this one. Some ``rescue disk'' programs use it, and still more floppies may have a minix filesystem, since the minix filesystem is less painful to use on a floppy.
Second extended - This is the standard Linux filesystem. You
almost definitely have one of these, and need to say `y'.
msdos - If you want to use your MS-DOS hard disk
partitions, or mount MS-DOS formatted floppy disks, say `y'.
There are various other foreign operating system filesystem types available.
/proc - (idea from Bell Labs, I guess). One doesn't make a proc
filesystem on a disk; this is a filesystem interface to the kernel and
processes. Many process listers (such as `ps') use it. Try
`cat /proc/meminfo' or `cat /proc/devices' sometime.
Some shells (rc, in particular) use /proc/self/fd (known as
/dev/fd
on other systems) for I/O. You should almost certainly say `y' to
this; many important linux tools depend on it.
NFS - If your machine lives on a network and you want to use filesystems which
reside on other systems with NFS, say `y'.
ISO9660 - Found on most CD-ROMs. If you have a CD-ROM drive and you wish to
use it under Linux, say `y'.
Ok, type `mount'. The output will look something like this:
blah# mount
/dev/hda1 on / type ext2 (defaults)
/dev/hda3 on /usr type ext2 (defaults)
none on /proc type proc (defaults)
/dev/fd0 on /mnt type msdos (defaults)
Look at each line; the word next to `type' is the filesystem
type. In this example, my / and /usr filesystems are
second extended, I'm using /proc, and there's a floppy
disk mounted using the msdos (bleah) filesystem.
You can try `cat /proc/filesystems' if you have /proc
currently enabled; it will list your current kernel's filesystems.
The configuration of rarely-used, non-critical filesystems can cause kernel bloat; see the section on modules for a way to avoid this and the ``Pitfalls'' section on why a bloated kernel is undesirable.
Here, you enable the drivers for your printer (parallel printer, that is),
busmouse, PS/2 mouse (many notebooks use the PS/2 mouse protocol for their
built-in trackballs), some tape drives, and other such ``character''
devices. Say `y'
when appropriate.
Note: gpm is a program which
allows the use of the mouse outside of the X window system for cut and paste
between virtual consoles. It's fairly nice if you have a serial mouse,
because it coexists well with X, but you need to do special tricks
for others.
If you feel a great desire to hear biff bark, say
`y',
and you can tell the configuration program all about your
sound board. (A note on sound card configuration: when it asks you if you
want to install the full version of the driver, you can say `n'
and save some kernel memory by picking only the features which you deem
necessary.)
If you are serious about sound card support, have a look at both the free
drivers at http://www.linux.org.uk/OSS/ and the commercial
Open Sound System at http://www.opensound.com/.
Not all of the configuration options are listed here because they change
too often or fairly self-evident (for instance, 3Com 3C509 support to
compile the device drive for this particular ethernet card).
There exists a fairly comprehensive list of all the options (plus a way to
place them into the Configure script) in an effort started and
maintained by Axel Boldt (boldt@math.ucsb.edu) and it's the online
help. It's also available as one big file at the
Documentation/Configure.help in your Linux kernel source tree as
of version 2.0.
>From Linus' README:
the ``kernel hacking'' configuration details usually result in a bigger or slower kernel (or both), and can even make the kernel less stable by configuring some routines to actively try to break bad code to find kernel problems (kmalloc()). Thus you should probably answer `n' to the questions for a ``production'' kernel.
After you finish configuration, a message tells you that your kernel has
been configured, and to ``check the top-level Makefile for
additional configuration,'' etc.
So, look at the Makefile. You probably will not need to change it,
but it never hurts to look. You can also change its options
with the `rdev' command once the new kernel is in place. If you're
feel lost when you look at the file, then don't worry about it.
When the configure script ends, it also tells you to `make dep'
and (possibly) `clean'.
So, do the `make dep'. This insures that all of the
dependencies, such the include files, are in place. It does not take long,
unless your computer is fairly slow to begin with.
For older versions of the kernel, when finished, you should do a
`make clean'.
This removes all of the object files and some other things
that an old version leaves behind. In any case,
do not forget this step before attempting to recompile a kernel.
After depending and cleaning, you may now `make
bzImage' or `make bzdisk' (this is the part that takes a long
time.).
`make bzImage' will compile the kernel, and leave
a file in arch/i386/boot called `bzImage'
(among other things). This is the new compressed kernel. `make bzdisk'
does the same thing, but also places the new bzImage
on a floppy disk which you hopefully put in drive ``A:''.
`bzdisk' is fairly handy for testing new
kernels; if it bombs (or just doesn't work right), just remove the floppy
and boot with
your old kernel. It can also be a handy way to boot if you accidentally
remove your kernel (or something equally as dreadful). You can also use it
to install new systems when you just dump the contents of one disk onto the
other (``all this and more! NOW how much would you pay?'').
All even halfway reasonably recent kernels are compressed, hence the
`bz' in front of the names. A compressed kernel automatically
decompresses itself when executed.
In older kernels, you don't have the option to build a bzImage; it
was simply a zImage. That option is at the moment still available,
however, given the code size of newer kernels, it is now more or
less mandatory to build a bzImage because the older methods can't
handle a kernel that's just too large.
`make mrproper' will do a more extensive `clean'ing.
It is sometimes necessary; you may wish to do it at every patch. `make
mrproper' will also delete your configuration file, so you might want
to make a backup of it (.config) if you see it as valuable.
`make oldconfig' will attempt to configure the kernel from an old
configuration file; it will run through the `make config' process
for you. If you haven't ever compiled a kernel before or don't have an old
config file, then you probably shouldn't do this, as you will most likely
want to change the default configuration.
See the section on modules for a description of `make modules'.
After you have a new kernel that seems to work the way you want it
to, it's time to install it. Most people use LILO (Linux Loader) for
this. `make bzlilo' will install the kernel, run LILO on it,
and get you all ready to boot, BUT ONLY if lilo is configured in the
following way on your system: kernel is
/vmlinuz, lilo is in /sbin, and your lilo config
(/etc/lilo.conf) agrees with this.
Otherwise, you need to use LILO directly. It's a fairly easy package to
install and work with, but it has a tendency to
confuse people with the configuration file.
Look at the config file (either /etc/lilo/config for older
versions or /etc/lilo.conf for new versions), and see what the
current setup is. The config file looks like this:
image = /vmlinuz
label = Linux
root = /dev/hda1
...
The `image =' is set to the currently installed kernel.
Most people use /vmlinuz. `label'
is used by lilo to determine which kernel or operating system to
boot, and `root' is the / of that particular operating
system. Make a backup
copy of your old kernel and copy the bzImage
which you just
made into place (you would say `cp bzImage /vmlinuz' if you
use `/vmlinuz'). Then, rerun lilo -- on newer systems, you can
just run `lilo', but on older stuff, you might have to do an
/etc/lilo/install or even an
/etc/lilo/lilo -C /etc/lilo/config.
If you would like to know more about LILO's configuration, or you don't have LILO, get the newest version from your favorite ftp site and follow the instructions.
To boot one of your old kernels off the hard disk
(another way to save yourself in case you screw up the new kernel),
copy the lines below (and including) `image = xxx' in the LILO config
file to the bottom of the file, and change the `image = xxx' to
`image = yyy', where `yyy' is the full pathname of
the file you saved your
backup kernel to. Then, change the `label = zzz' to
`label = linux-backup' and rerun lilo. You may need to
put a line in
the config file saying `delay=x', where x is an amount in tenths of a
second, which tells LILO to wait that much time before booting, so that you
can interrupt it (with the shift key, for example), and type in the label
of the backup boot image (in case unpleasant things happen).
Incremental upgrades of the kernel are distributed as patches. For
example, if you have version 1.1.45, and you notice that there's a
`patch46.gz' out there for it, it means you can upgrade to version
1.1.46 through application of the patch. You might want to make a backup of the
source tree first (`make clean' and then
`cd /usr/src; tar zcvf old-tree.tar.gz linux'
will make a compressed tar archive for you.).
So, continuing with the example above, let's suppose that
you have `patch46.gz' in /usr/src. cd to
/usr/src and do a `zcat patch46.gz | patch -p0'
(or `patch -p0 < patch46'
if the patch isn't compressed). You'll see things whizz by
(or flutter by, if your
system is that slow) telling you that it is trying to apply hunks,
and whether it succeeds or not. Usually, this action goes by too quickly for
you to read, and you're not too sure whether it worked or not, so you might
want to use the -s flag to patch, which tells patch
to only report error messages (you don't get as much of the ``hey, my
computer is actually doing something for a change!'' feeling, but you may
prefer this..). To look for
parts which might not have gone smoothly, cd to /usr/src/linux and
look for files with a .rej extension. Some versions of patch
(older versions which may have been compiled with on an inferior
filesystem) leave the rejects with a # extension. You can use
`find' to look for you;
find . -name '*.rej' -print
prints all files who live in the current directory or any subdirectories with
a .rej extension to the standard output.
If everything went right, do a `make clean', `config',
and `dep' as described in sections 3 and 4.
There are quite a few options to the patch command. As mentioned
above, patch -s
will suppress all messages except the errors. If you keep your kernel
source in some other place than /usr/src/linux, patch -p1
(in that directory) will patch things cleanly. Other patch options are
well-documented in the manual page.
(Note: this section refers mostly to quite old kernels)
The most frequent problem that used to arise was when a patch modified
a file called `config.in' and it didn't look quite right,
because you changed the options to suit your machine. This has been
taken care of, but one still might encounter it with an older release.
To fix it, look at the config.in.rej file, and see what remains
of the original patch.
The changes will typically be marked with `+' and `-'
at the beginning of the
line. Look at the lines surrounding it, and remember if they were set to
`y' or `n'. Now, edit config.in, and change
`y' to `n' and `n' to `y'
when appropriate. Do a
patch -p0 < config.in.rej
and if it reports that it
succeeded (no fails), then you can continue on with a configuration and
compilation. The config.in.rej file will remain, but you can get
delete it.
If you encounter further problems, you might have installed a patch out
of order. If patch says `previously applied patch detected: Assume
-R?', you are probably trying to apply a patch which is below your current
version number; if you answer `y', it will attempt to degrade
your source, and will most likely fail; thus, you will need to get a whole new
source tree (which might not have been such a bad idea in the first place).
To back out (unapply) a patch, use `patch -R' on the original patch.
The best thing to do when patches really turn out wrong is to start over
again with a clean, out-of-the-box source tree (for example, from one
of the linux-x.y.z.tar.gz files), and start again.
After just a few patches, the .orig files will start to pile up. For
example, one 1.1.51 tree I had was once last cleaned out at 1.1.48.
Removing the .orig files saved over a half a meg.
find . -name '*.orig' -exec rm -f {} ';'
will take care of it for you. Versions of patch which use
# for rejects use a tilde instead of .orig.
There are better ways to get rid of the .orig files, which
depend on GNU xargs:
find . -name '*.orig' | xargs rm
or the ``quite secure but a little more verbose'' method:
find . -name '*.orig' -print0 | xargs --null rm --
There are other patches (I'll call them ``nonstandard'') than the ones Linus distributes. If you apply these, Linus' patches may not work correctly and you'll have to either back them out, fix the source or the patch, install a new source tree, or a combination of the above. This can become very frustrating, so if you do not want to modify the source (with the possibility of a very bad outcome), back out the nonstandard patches before applying Linus', or just install a new tree. Then, you can see if the nonstandard patches still work. If they don't, you are either stuck with an old kernel, playing with the patch or source to get it to work, or waiting (possibly begging) for a new version of the patch to come out.
How common are the patches not in the standard distribution? You will probably hear of them. I used to use the noblink patch for my virtual consoles because I hate blinking cursors (This patch is (or at least was) frequently updated for new kernel releases.). With most newer device drivers being developed as loadable modules, though, the frequecy of ``nonstandard'' patches is decreasing significantly.
Your linux kernel has many features which are not explained in the kernel source itself; these features are typically utilized through external packages. Some of the most common are listed here.
The linux console probably has more features than it deserves. Among these are the ability to switch fonts, remap your keyboard, switch video modes (in newer kernels), etc. The kbd package has programs which allow the user to do all of this, plus many fonts and keyboard maps for almost any keyboard, and is available from the same sites that carry the kernel source.
Rik Faith (faith@cs.unc.edu) put together a large collection of
linux utilities which are, by odd coincidence, called util-linux. These are
now maintained by Andries Brouwer (util-linux@math.uio.no).
Available via
anonymous ftp from sunsite.unc.edu in /pub/Linux/system/misc, it
contains programs such as setterm, rdev, and
ctrlaltdel, which are relevant to the kernel. As Rik says, do
not install without thinking; you do not need to install everything in
the package, and it could very well cause serious problems if you do.
As with many packages, this was once a kernel patch and support programs. The patches made it into the official kernel, and the programs to optimize and play with your hard disk are distributed separately.
gpm stands for general purpose mouse. This program allows you to cut and paste text between virtual consoles and do other things with a large variety of mouse types.
If your new kernel does really weird things after a routine kernel upgrade,
chances are you forgot to make clean before compiling the new
kernel. Symptoms can be anything from
your system outright crashing, strange I/O problems, to crummy
performance. Make sure you do a make dep, too.
If your kernel is sucking up a lot of memory, is too large, and/or just takes forever to compile even when you've got your new Quadbazillium-III/4400 working on it, you've probably got lots of unneeded stuff (device drivers, filesystems, etc) configured. If you don't use it, don't configure it, because it does take up memory. The most obvious symptom of kernel bloat is extreme swapping in and out of memory to disk; if your disk is making a lot of noise and it's not one of those old Fujitsu Eagles that sound like like a jet landing when turned off, look over your kernel configuration.
You can find out how much memory the kernel is using by taking the
total amount of memory in your machine and subtracting from it the
amount of ``total mem'' in /proc/meminfo or the output of the command
`free'.
Configuration options for PCs are: First, under the category `General Setup', select `Parallel port support' and `PC-style hardware'. Then under `Character devices', select `Parallel printer support'.
Then there are the names. Linux 2.2 names the printer devices differently
than previous releases. The upshot of this is that if you had an lp1
under your old kernel, it's probably an lp0 under your new one.
Use `dmesg' or look through the logs in /var/log to find
out.
If it does not compile, then it is likely that a patch failed, or your
source is somehow corrupt. Your version of gcc also might not
be correct, or could also be corrupt (for example, the include files
might be in error). Make sure that the symbolic links which
Linus describes in the README are set up correctly. In general, if
a standard kernel
does not compile, something is seriously wrong with the system, and
reinstallation of certain tools is probably necessary.
In some cases, gcc can crash due to hardware problems. The error message will be something like ``xxx exited with signal 15'' and it will generally look very mysterious. I probably would not mention this, except that it happened to me once - I had some bad cache memory, and the compiler would occasionally barf at random. Try reinstalling gcc first if you experience problems. You should only get suspicious if your kernel compiles fine with external cache turned off, a reduced amount of RAM, etc.
It tends to disturb people when it's suggested that their hardware has
problems. Well, I'm not making this up. There is an FAQ for it -- it's at
http://www.bitwizard.nl/sig11/.
You did not run LILO, or it is not configured correctly. One thing that
``got'' me once was a problem in the config file; it said `boot =
/dev/hda1'
instead of `boot = /dev/hda' (This can be really annoying at first,
but once you have a working config file, you shouldn't need to
change it.).
Ooops! The best thing you can do here is to boot off of a floppy disk or
CDROM and
prepare another bootable floppy (such as `make zdisk' would do).
You need to know where your root (/) filesystem is and what type
it is (e.g. second extended, minix). In the example below, you also need
to know what filesystem your /usr/src/linux source
tree is on, its type, and where it is normally mounted.
In the following example, / is /dev/hda1, and the
filesystem which holds /usr/src/linux
is /dev/hda3, normally mounted at /usr. Both are
second extended filesystems. The working kernel image in
/usr/src/linux/arch/i386/boot is called bzImage.
The idea is that if there is a functioning
bzImage, it is possible to use that
for the new floppy. Another alternative, which may or may not work better
(it depends on the particular method in which you messed up your system) is
discussed after the example.
First, boot from a boot/root disk combo or rescue disk, and mount the filesystem which contains the working kernel image:
mkdir /mnt
mount -t ext2 /dev/hda3 /mnt
If mkdir tells you that the directory already exists, just ignore
it. Now, cd to the place where the working kernel image was. Note
that
/mnt + /usr/src/linux/arch/i386/boot - /usr = /mnt/src/linux/arch/i386/bootPlace a formatted disk in drive ``A:'' (not your boot or root disk!), dump the image to the disk, and configure it for your root filesystem:
cd /mnt/src/linux/arch/i386/boot
dd if=bzImage of=/dev/fd0
rdev /dev/fd0 /dev/hda1
cd to / and unmount the normal /usr filesystem:
cd /
umount /mnt
You should now be able to reboot your system as normal from this floppy. Don't forget to run lilo (or whatever it was that you did wrong) after the reboot!
As mentioned above, there is another common alternative. If you
happened to have a working kernel image in / (/vmlinuz
for example), you can use that for a boot disk. Supposing all of the above
conditions, and that my kernel image is /vmlinuz, just make these
alterations to the example above: change
/dev/hda3 to /dev/hda1 (the / filesystem),
/mnt/src/linux to
/mnt, and if=bzImage to if=vmlinuz. The
note explaining how to derive /mnt/src/linux may be ignored.
Using LILO with big drives (more than 1024 cylinders) can cause problems. See the LILO mini-HOWTO or documentation for help on that.
This can be a severe problem. Starting with a kernel release
after 1.0 (around 20 Apr 1994), a program called `update' which
periodically flushes out the filesystem buffers, was upgraded/replaced. Get
the sources to `bdflush'
(you should find it where you got your kernel source), and install it (you
probably want to run your system under the old kernel while doing this). It
installs itself as `update' and after a reboot, the new kernel
should no longer complain.
Strangely enough, lots of people cannot get their ATAPI drives working, probably because there are a number of things that can go wrong.
If your CD-ROM drive is the only device on a particular IDE interface, it must be jumpered as ``master'' or ``single.'' Supposedly, this is the most common error.
Creative Labs (for one) has put IDE interfaces on their sound cards now. However, this leads to the interesting problem that while some people only have one interface to being with, many have two IDE interfaces built-in to their motherboards (at IRQ15, usually), so a common practice is to make the soundblaster interface a third IDE port (IRQ11, or so I'm told).
This causes problems with linux in that versions 1.2.x don't support a third IDE interface (there is support in starting somewhere in the 1.3.x series but that's development, remember, and it doesn't auto-probe). To get around this, you have a few choices.
If you have a second IDE port already, chances are that you are not using it or it doesn't already have two devices on it. Take the ATAPI drive off the sound card and put it on the second interface. You can then disable the sound card's interface, which saves an IRQ anyway.
If you don't have a second interface, jumper the sound card's interface (not the sound card's sound part) as IRQ15, the second interface. It should work.
Get new versions of the route program and any other programs
which do route manipulation.
/usr/include/linux/route.h (which is actually a file in
/usr/src/linux) has changed.
Upgrade to at least version 1.2.1.
Don't use the vmlinux file created in /usr/src/linux as
your boot image; [..]/arch/i386/boot/bzImage is the right
one.
Change the word dumb to linux in the console termcap
entry in /etc/termcap. You may also have to make a terminfo entry.
The linux kernel source includes a number of include files (the things that
end with .h) which are referenced by the standard ones in
/usr/include. They are typically referenced like this (where
xyzzy.h would be something in /usr/include/linux):
#include <linux/xyzzy.h>
Normally, there is a link called linux in /usr/include to
the include/linux directory of your kernel source
(/usr/src/linux/include/linux in the typical system). If this link
is not there, or points to the wrong place, most things will not compile at
all. If you decided that the kernel source was taking too much room on the
disk and deleted it, this will obviously be a problem. Another way it might
go wrong is with file permissions; if your root has a umask
which doesn't allow other users to see its files by default, and you
extracted the kernel source without the p (preserve filemodes)
option, those users also won't be able to use the C compiler. Although you
could use the chmod command to fix this, it is probably easier to
re-extract the include files. You can do this the same way you did the
whole source at the beginning, only with an additional argument:
blah# tar zxvpf linux.x.y.z.tar.gz linux/include
Note: ``make config'' will recreate the /usr/src/linux
link if it isn't there.
The following few example commands may be helpful to those wondering how to increase certain soft limits imposed by the kernel:
echo 4096 > /proc/sys/kernel/file-max echo 12288 > /proc/sys/kernel/inode-max echo 300 400 500 > /proc/sys/vm/freepages
Kernel versions 2.0.x and 2.2.x introduced quite a bit of changes for kernel
installation. The file Documentation/Changes in the 2.0.x source
tree contains information that you should know when upgrading to either of
these versions. You will most likely need to upgrade several key packages,
such as gcc, libc, and SysVInit, and perhaps alter some system files, so
expect this. Don't panic, though.
Loadable kernel modules can save memory and ease configuration. The scope of modules has grown to include filesystems, ethernet card drivers, tape drivers, printer drivers, and more.
The module utilities are available from wherever you got your kernel
source as modutils-x.y.z.tar.gz; choose the highest
patchlevel x.y.z that is equal to or below that of your current
kernel. Unpack it with `tar zxvf modutils-x.y.z.tar.gz',
cd to the directory it creates (modutils-x.y.z), look
over the README, and carry out its installation instructions
(which is usually something simple, such as make install). You
should now have the programs
insmod, rmmod, ksyms, lsmod,
genksyms, modprobe, and depmod in
/sbin. If you wish,
test out the utilities with the ``hw'' example driver in insmod; look
over the INSTALL file in that subdirectory for details.
insmod inserts a module into the running kernel. Modules
usually have a .o extension; the example driver mentioned above
is called drv_hello.o, so to insert this, one would say
`insmod drv_hello.o'. To see the modules that the kernel is
currently using, use lsmod. The output looks like this:
blah# lsmod
Module: #pages: Used by:
drv_hello 1
`drv_hello' is the name of the module, it uses one page (4k) of
memory, and no other kernel modules depend on it at the moment. To remove
this module, use `rmmod drv_hello'. Note that rmmod
wants a module name, not a filename; you get this from
lsmod's listing. The other module utilities' purposes are documented
in their manual pages.
As of version 2.0.30, most of everything is available as a loadable
modules. To use
them, first make sure that you don't configure them into the regular
kernel; that is, don't say y to it during `make config'.
Compile a new kernel and reboot with it. Then, cd to
/usr/src/linux again, and do a `make modules'. This
compiles all of the modules which you did not specify in the kernel
configuration, and places links to them in /usr/src/linux/modules.
You can use them straight from that directory or execute `make
modules_install', which installs them in
/lib/modules/x.y.z, where x.y.z is the kernel release.
This can be especially handy with filesystems. You may not use the minix
or msdos filesystems frequently. For example, if I encountered an msdos
(shudder) floppy, I would insmod /usr/src/linux/modules/msdos.o,
and then rmmod msdos when finished. This procedure saves about
50k of RAM in the kernel during normal operation. A small note is in order for
the minix filesystem: you should always configure it directly into the
kernel for use in ``rescue'' disks.
If you would like logs of what those `make' or `patch'
commands did, you can redirect output to a file. First,
find out what shell you're running:
`grep root /etc/passwd' and look for something like
`/bin/csh'.
If you use sh or bash,
(command) 2>&1 | tee (output file)
will place a copy of (command)'s output in the
file `(output file)'.
For csh or tcsh, use
(command) |& tee (output file)
For rc (Note: you probably do not use rc) it's
(command) >[2=1] | tee (output file)
Other than using floppy disks, there are several methods of testing out a new kernel without touching the old one. Unlike many other Unix flavors, LILO has the ability to boot a kernel from anywhere on the disk (if you have a large (500 MB or above) disk, please read over the LILO documentation on how this may cause problems). So, if you add something similar to
image = /usr/src/linux/arch/i386/boot/bzImage
label = new_kernel
to the end of your LILO configuration file, you can choose to run a newly
compiled kernel without touching your old /vmlinuz (after running
lilo, of course). The easiest way to tell LILO to boot a new
kernel is to press the shift key at bootup time (when it says
LILO on the screen, and nothing else), which gives you a prompt.
At this point, you can enter `new_kernel' to boot the new kernel.
If you wish to keep several different kernel source trees on your system at
the same time (this can take up a lot of disk space; be careful), the
most common way is to name them /usr/src/linux-x.y.z, where
x.y.z is the kernel version. You can then ``select'' a source
tree with a symbolic link; for example, `ln -sf linux-1.2.2
/usr/src/linux' would make the 1.2.2 tree current. Before creating a
symbolic link like this, make certain that the last argument to
ln is not a real directory (old symbolic links are fine); the
result will not be what you expect.
Russell Nelson (nelson@crynwr.com) summarizes the changes in new
kernel releases. These are short, and you might like to look at them
before an upgrade. They are available with anonymous ftp from
ftp.emlist.com in pub/kchanges or through the URL
http://www.crynwr.com/kchanges
The author and maintainer of the Linux Kernel-HOWTO is Brian Ward
(bri@cs.uchicago.edu). Please send me any comments, additions,
corrections (Corrections are, in particular, the most important to me.).
You can take a look at my `home page' at one of these URLs:
http://www.math.psu.edu/bri/
http://blah.math.tu-graz.ac.at/~bri/
Even though I try to be attentive as possible with mail, please remember that I get a lot of it every day, so it may take a little time to get back to you. Especially when emailing me with a question, please try extra hard to be clear and detailed in your message. If you're writing about non-working hardware (or something like that), I need to know what your hardware configureation is. If you report an error, don't just say ``I tried this but it gave an error;'' I need to know what the error was. I would also like to know what versions of the kernel, gcc, and libc you're using. If you just tell me you're using this-or-that distribution, it won't tell me much at all. I don't care if you ask simple questions; remember, if you don't ask, you may never get an answer! I'd like to thank everyone who has given me feedback.
If your question does not relate to the kernel, or is in some language that I don't understand, I may not answer.
If you mailed me and did not get an answer within a resonable amount of time (three weeks or more), then chances are that I accidentally deleted your message or something (sorry). Please try again.
I get a lot of mail about thing which are actually hardware problems or issues. That's OK, but please try to keep in mind that I'm not familiar with all of the hardware in the world. I use AMD processors, Adaptec and Sybios SCSI controllers, and IBM SCSI disks.
Version -0.1 was written on October 3, 1994. This document is available in SGML, PostScript, TeX, roff, and plain-text formats.
The ``Tips and tricks'' section is a little small. I hope to expand on it with suggestions from others.
So is ``Additional packages.''
More debugging/crash recovery info needed.
A small part of Linus' README (kernel hacking options) is inclusive. (Thanks, Linus!)
uc@brian.lunetix.de (Ulrich Callmeier): patch -s and xargs.
quinlan@yggdrasil.com (Daniel Quinlan): corrections and
additions in many sections.
nat@nat@nataa.fr.eu.org (Nat Makarevitch): mrproper, tar -p, many
other things
boldt@math.ucsb.edu (Axel Boldt): collected descriptions of
kernel configuration options on the net; then provided me with the list
lembark@wrkhors.psyber.com (Steve Lembark): multiple boot
suggestion
kbriggs@earwax.pd.uwa.edu.au (Keith Briggs): some corrections and
suggestions
rmcguire@freenet.columbus.oh.us (Ryan McGuire): makeables
additions
dumas@excalibur.ibp.fr (Eric Dumas): French translation
simazaki@ab11.yamanashi.ac.jp (Yasutada Shimazaki): Japanese translation
jjamor@lml.ls.fi.upm.es (Juan Jose Amor Iglesias): Spanish translation
mva@sbbs.se (Martin Wahlen): Swedish translation
jzp1218@stud.u-szeged.hu (Zoltan Vamosi): Hungarian translation
bart@mat.uni.torun.pl (Bartosz Maruszewski): Polish translation
donahue@tiber.nist.gov (Michael J Donahue): typos, winner of the
``sliced bread competition''
rms@gnu.ai.mit.edu (Richard Stallman):
``free'' documentation concept/distribution notice
dak@Pool.Informatik.RWTH-Aachen.DE (David Kastrup): NFS thing
esr@snark.thyrsus.com (Eric Raymond): various tidbits
The people who have sent me mail with questions and problems have also been quite helpful.
Copyright © Brian Ward, 1994-1999.
Permission is granted to make and distribute copies of this manual provided the copyright notice and this permission notice are preserved on all copies.
Permission is granted to copy and distribute modified versions of this manual under the conditions for verbatim copying, provided that the derived work is distributed under the terms of a permission notice identical to this one. Translations fall under the catagory of ``modified versions.''
Warranty: None.
Recommendations: Commercial redistribution is allowed and encouraged; however, it is strongly recommended that the redistributor contact the author before the redistribution, in the interest of keeping things up-to-date (you could send me a copy of the thing you're making while you're at it). Translators are also advised to contact the author before translating. The printed version looks nicer. Recycle.