The Linux kernel accepts certain `command line options' or `boot time
parameters' at the moment it is started. In general this is used to supply the
kernel with information about hardware parameters that the kernel would not be
able to determine on its own, or to avoid/override the values that the kernel
would otherwise detect.
When the kernel is booted directly by the BIOS (say from a floppy to which you
copied a kernel using `cp zImage /dev/fd0'), you have no opportunity to
specify any parameters. So, in order to take advantage of this possibility you
have to use software that is able to pass parameters, like LILO or loadlin.
For a few parameters one can also modify the kernel image itself, using rdev,
see rdev(8) for further details.
The LILO program (LInux LOader) written by Werner Almesberger is the most
commonly used. It has the ability to boot various kernels, and stores the
configuration information in a plain text file. (See lilo(8) and
lilo.conf(5).) LILO can boot DOS, OS/2, Linux, FreeBSD, UnixWare, etc.,
and is quite flexible.
The other commonly used Linux loader is `LoadLin' which is a DOS program that
has the capability to launch a Linux kernel from the DOS prompt (with
boot-args) assuming that certain resources are available. This is good for
people that want to launch Linux from DOS.
It is also very useful if you have certain hardware which relies on the supplied
DOS driver to put the hardware into a known state. A common example is
`SoundBlaster Compatible' sound cards that require the DOS driver to twiddle a
few mystical registers to put the card into a SB compatible mode. Booting DOS
with the supplied driver, and then loading Linux from the DOS prompt with
loadlin avoids the reset of the card that happens if one rebooted instead.
The kernel command line is parsed into a list of strings (boot arguments)
separated by spaces. Most of the boot args take the form of:
where `name' is a unique keyword that is used to identify what part of the
kernel the associated values (if any) are to be given to. Note the limit of 10
is real, as the present code only handles 10 comma separated parameters per
keyword. (However, you can re-use the same keyword with up to an additional 10
parameters in unusually complicated situations, assuming the setup function
Most of the sorting goes on in linux/init/main.c. First, the kernel checks to
see if the argument is any of the special arguments `root=', `nfsroot=',
`nfsaddrs=', `ro', `rw', `debug' or `init'. The meaning of these special
arguments is described below.
Then it walks a list of setup functions (contained in the bootsetups array) to
see if the specified argument string (such as `foo') has been associated with
a setup function (`foo_setup()') for a particular device or part of the
kernel. If you passed the kernel the line foo=3,4,5,6 then the kernel would
search the bootsetups array to see if `foo' was registered. If it was, then it
would call the setup function associated with `foo' (foo_setup()) and hand it
the arguments 3, 4, 5 and 6 as given on the kernel command line.
Anything of the form `foo=bar' that is not accepted as a setup function as
described above is then interpreted as an environment variable to be set. A
(useless?) example would be to use `TERM=vt100' as a boot argument.
Any remaining arguments that were not picked up by the kernel and were not
interpreted as environment variables are then passed onto process one, which
is usually the init program. The most common argument that is passed to the
init process is the word `single' which instructs init to boot the computer in
single user mode, and not launch all the usual daemons. Check the manual page
for the version of init installed on your system to see what arguments it
This sets the initial command to be executed by the kernel. If this is not set,
or cannot be found, the kernel will try /etc/init, then
/bin/init, then /sbin/init, then /bin/sh and panic if all
of this fails.
(Only when CONFIG_BUGi386 is defined.) Some i387 coprocessor chips have bugs
that show up when used in 32 bit protected mode. For example, some of the
early ULSI-387 chips would cause solid lockups while performing floating point
calculations. Using the `no387' boot arg causes Linux to ignore the maths
coprocessor even if you have one. Of course you must then have your kernel
compiled with math emulation support!
(Only when CONFIG_BUGi386 is defined.) Some of the early i486DX-100 chips have a
problem with the `hlt' instruction, in that they can't reliably return to
operating mode after this instruction is used. Using the `no-hlt' instruction
tells Linux to just run an infinite loop when there is nothing else to do, and
to not halt the CPU. This allows people with these broken chips to use Linux.
This argument tells the kernel what device is to be used as the root filesystem
while booting. The default of this setting is determined at compile time, and
usually is the value of the root device of the system that the kernel was
built on. To override this value, and select the second floppy drive as the
root device, one would use `root=/dev/fd1'. (The root device can also be set
The root device can be specified symbolically or numerically. A symbolic
specification has the form /dev/XXYN, where XX designates the device type
(`hd' for ST-506 compatible hard disk, with Y in `a'-`d'; `sd' for SCSI
compatible disk, with Y in `a'-`e'; `ad' for Atari ACSI disk, with Y in
`a'-`e', `ez' for a Syquest EZ135 parallel port removable drive, with Y=`a',
`xd' for XT compatible disk, with Y either `a' or `b'; `fd' for floppy disk,
with Y the floppy drive number - fd0 would be the DOS `A:' drive, and fd1
would be `B:'), Y the driver letter or number, and N the number (in decimal)
of the partition on this device (absent in the case of floppies). Recent
kernels allow many other types, mostly for CD-ROMs: nfs, ram, scd, mcd,
cdu535, aztcd, cm206cd, gscd, sbpcd, sonycd, bpcd. (The type nfs specifies a
net boot; ram refers to a ram disk.)
Note that this has nothing to do with the designation of these devices on your
file system. The `/dev/' part is purely conventional.
The more awkward and less portable numeric specification of the above possible
root devices in major/minor format is also accepted. (E.g., /dev/sda3 is major
8, minor 3, so you could use `root=0x803' as an alternative.)
The `ro' option tells the kernel to mount the root filesystem as `readonly' so
that filesystem consistency check programs (fsck) can do their work on a
quiescent file system. No processes can write to files on the filesystem in
question until it is `remounted' as read/write capable, e.g., by `mount -w -n
-o remount /'. (See also mount(8).)
The `rw' option tells the kernel to mount the root filesystem read/write. This
is the default.
The choice between read-only and read/write can also be set using
This is used to protect I/O port regions from probes. The form of the command
In some machines it may be necessary to prevent device drivers from checking for
devices (auto-probing) in a specific region. This may be because of hardware
that reacts badly to the probing, or hardware that would be mistakenly
identified, or merely hardware you don't want the kernel to initialize.
The reserve boot-time argument specifies an I/O port region that shouldn't be
probed. A device driver will not probe a reserved region, unless another boot
argument explicitly specifies that it do so.
For example, the boot line
keeps all device drivers except the driver for `blah' from probing 0x300-0x31f.
The BIOS call defined in the PC specification that returns the amount of
installed memory was only designed to be able to report up to 64MB. Linux uses
this BIOS call at boot to determine how much memory is installed. If you have
more than 64MB of RAM installed, you can use this boot arg to tell Linux how
much memory you have. The value is in decimal or hexadecimal (prefix 0x), and
the suffixes `k' (times 1024) or `M' (times 1048576) can be used. Here is a
quote from Linus on usage of the `mem=' parameter.
``The kernel will accept any `mem=xx' parameter you give it, and if it turns out
that you lied to it, it will crash horribly sooner or later. The parameter
indicates the highest addressable RAM address, so `mem=0x1000000' means you
have 16MB of memory, for example. For a 96MB machine this would be
NOTE NOTE NOTE: some machines might use the top of memory for BIOS cacheing or
whatever, so you might not actually have up to the full 96MB addressable. The
reverse is also true: some chipsets will map the physical memory that is
covered by the BIOS area into the area just past the top of memory, so the
top-of-mem might actually be 96MB + 384kB for example. If you tell linux that
it has more memory than it actually does have, bad things will happen: maybe
not at once, but surely eventually.''
(Only when CONFIG_BUGi386 is defined.) Since 2.0.22 a reboot is by default a
cold reboot. One asks for the old default with `reboot=warm'. (A cold reboot
may be required to reset certain hardware, but might destroy not yet written
data in a disk cache. A warm reboot may be faster.) By default a reboot is
hard, by asking the keyboard controller to pulse the reset line low, but there
is at least one type of motherboard where that doesn't work. The option
`reboot=bios' will instead jump through the BIOS.
(Only when __SMP__ is defined.) A command-line option of `nosmp' or `maxcpus=0'
will disable SMP activation entirely; an option `maxcpus=N' limits the maximum
number of CPUs activated in SMP mode to N.
Kernel messages are handed off to the kernel log daemon klogd so that they may
be logged to disk. Messages with a priority above console_loglevel are
also printed on the console. (For these levels, see <linux/kernel.h>.)
By default this variable is set to log anything more important than debug
messages. This boot argument will cause the kernel to also print the messages
of DEBUG priority. The console loglevel can also be set at run time via an
option to klogd. See klogd(8).
It is possible to enable a kernel profiling function, if one wishes to find out
where the kernel is spending its CPU cycles. Profiling is enabled by setting
the variable prof_shift to a nonzero value. This is done either by
specifying CONFIG_PROFILE at compile time, or by giving the `profile=' option.
Now the value that prof_shift gets will be N, when given, or
CONFIG_PROFILE_SHIFT, when that is given, or 2, the default. The significance
of this variable is that it gives the granularity of the profiling: each clock
tick, if the system was executing kernel code, a counter is incremented:
profile[address >> prof_shift]++;
The raw profiling information can be read from /proc/profile. Probably
you'll want to use a tool such as readprofile.c to digest it. Writing to
/proc/profile will clear the counters.
Set the eight parameters max_page_age, page_advance, page_decline,
page_initial_age, age_cluster_fract, age_cluster_min, pageout_weight,
bufferout_weight that control the kernel swap algorithm. For kernel tuners
(Only if the kernel was compiled with CONFIG_BLK_DEV_RAM.) In general it is a
bad idea to use a ramdisk under Linux - the system will use available memory
more efficiently itself. But while booting (or while constructing boot
floppies) it is often useful to load the floppy contents into a ramdisk. One
might also have a system in which first some modules (for filesystem or
hardware) must be loaded before the main disk can be accessed.
In Linux 1.3.48, ramdisk handling was changed drastically. Earlier, the memory
was allocated statically, and there was a `ramdisk=N' parameter to tell its
size. (This could also be set in the kernel image at compile time, or by use
of rdev(8).) These days ram disks use the buffer cache, and grow
dynamically. For a lot of information (e.g., how to use rdev(8) in
conjunction with the new ramdisk setup), see
There are four parameters, two boolean and two integral.
(Only if the kernel was compiled with CONFIG_BLK_DEV_RAM and
CONFIG_BLK_DEV_INITRD.) These days it is possible to compile the kernel to use
initrd. When this feature is enabled, the boot process will load the kernel
and an initial ramdisk; then the kernel converts initrd into a
"normal" ramdisk, which is mounted read-write as root device; then
/linuxrc is executed; afterwards the "real" root file system is
mounted, and the initrd filesystem is moved over to /initrd; finally the usual
boot sequence (e.g. invocation of /sbin/init) is performed.
For a detailed description of the initrd feature, see
The `noinitrd' option tells the kernel that although it was compiled for
operation with initrd, it should not go through the above steps, but leave the
initrd data under /dev/initrd. (This device can be used only once - the
data is freed as soon as the last process that used it has closed
iobase -- the first I/O port that the SCSI host occupies. These are
specified in hexidecimal notation, and usually lie in the range from 0x200 to
irq -- the hardware interrupt that the card is configured to use. Valid
values will be dependent on the card in question, but will usually be 5, 7, 9,
10, 11, 12, and 15. The other values are usually used for common peripherals
like IDE hard disks, floppies, serial ports, etc.
scsi-id -- the ID that the host adapter uses to identify itself on the
SCSI bus. Only some host adapters allow you to change this value, as most have
it permanently specified internally. The usual default value is 7, but the
Seagate and Future Domain TMC-950 boards use 6.
parity -- whether the SCSI host adapter expects the attached devices to
supply a parity value with all information exchanges. Specifying a one
indicates parity checking is enabled, and a zero disables parity checking.
Again, not all adapters will support selection of parity behaviour as a boot
A SCSI device can have a number of `sub-devices' contained within itself. The
most common example is one of the new SCSI CD-ROMs that handle more than one
disk at a time. Each CD is addressed as a `Logical Unit Number' (LUN) of that
particular device. But most devices, such as hard disks, tape drives and such
are only one device, and will be assigned to LUN zero.
Some poorly designed SCSI devices cannot handle being probed for LUNs not equal
to zero. Therefore, if the compile time flag CONFIG_SCSI_MULTI_LUN is not set,
newer kernels will by default only probe LUN zero.
To specify the number of probed LUNs at boot, one enters `max_scsi_luns=n' as a
boot arg, where n is a number between one and eight. To avoid problems as
described above, one would use n=1 to avoid upsetting such broken devices.
Some boot time configuration of the SCSI tape driver can be achieved by using
The first two numbers are specified in units of kB. The default buf_size
is 32kB, and the maximum size that can be specified is a ridiculous 16384kB.
The write_threshold is the value at which the buffer is committed to
tape, with a default value of 30kB. The maximum number of buffers varies with
the number of drives detected, and has a default of two. An example usage
Full details can be found in the README.st file that is in the scsi directory of
the kernel source tree.
The aha1542 series cards have an i82077 floppy controller onboard, while the
aha1540 series cards do not. These are busmastering cards, and have parameters
to set the ``fairness'' that is used to share the bus with other devices. The
boot arg looks like the following.
Valid iobase values are usually one of: 0x130, 0x134, 0x230, 0x234, 0x330,
0x334. Clone cards may permit other values.
The buson, busoff values refer to the number of microseconds that
the card dominates the ISA bus. The defaults are 11us on, and 4us off, so that
other cards (such as an ISA LANCE Ethernet card) have a chance to get access
to the ISA bus.
The dmaspeed value refers to the rate (in MB/s) at which the DMA (Direct
Memory Access) transfers proceed. The default is 5MB/s. Newer revision cards
allow you to select this value as part of the soft-configuration, older cards
use jumpers. You can use values up to 10MB/s assuming that your motherboard is
capable of handling it. Experiment with caution if using values over 5MB/s.
The extended value, if non-zero, indicates that extended translation for
large disks is enabled. The no_reset value, if non-zero, tells the
driver not to reset the SCSI bus when setting up the host adaptor at boot.
The AdvanSys driver can accept up to four i/o addresses that will be probed for
an AdvanSys SCSI card. Note that these values (if used) do not effect EISA or
PCI probing in any way. They are only used for probing ISA and VLB cards. In
addition, if the driver has been compiled with debugging enabled, the level of
debugging output can be set by adding an 0xdeb[0-f] parameter. The 0-f allows
setting the level of the debugging messages to any of 16 levels of verbosity.
For an extensive discussion of the BusLogic command line parameters, see
/usr/src/linux/drivers/scsi/BusLogic.c (lines 3149-3270 in the kernel
version I am looking at). The text below is a very much abbreviated extract.
The parameters N1-N5 are integers. The parameters S1,... are strings. N1 is the
I/O Address at which the Host Adapter is located. N2 is the Tagged Queue Depth
to use for Target Devices that support Tagged Queuing. N3 is the Bus Settle
Time in seconds. This is the amount of time to wait between a Host Adapter
Hard Reset which initiates a SCSI Bus Reset and issuing any SCSI Commands. N4
is the Local Options (for one Host Adapter). N5 is the Global Options (for all
The string options are used to provide control over Tagged Queuing (TQ:Default,
TQ:Enable, TQ:Disable, TQ:<Per-Target-Spec>), over Error Recovery
(ER:Default, ER:HardReset, ER:BusDeviceReset, ER:None,
ER:<Per-Target-Spec>), and over Host Adapter Probing (NoProbe,
where S is a comma-separated string of items keyword[:value]. Recognized
keywords (possibly with value) are: ioport:addr, noreset, nosync:x, period:ns,
disconnect:x, debug:x, proc:x. For the function of these parameters, see
If the card doesn't use interrupts, then an IRQ value of 255 (0xff) will disable
interrupts. An IRQ value of 254 means to autoprobe. More details can be found
in the file /usr/src/linux/drivers/scsi/README.g_NCR5380.
where S is a comma-separated string of items keyword:value. Recognized keywords
are: mpar (master_parity), spar (scsi_parity), disc (disconnection), specf
(special_features), ultra (ultra_scsi), fsn (force_sync_nego), tags
(default_tags), sync (default_sync), verb (verbose), debug (debug), burst
(burst_max). For the function of the assigned values, see
The IDE driver accepts a number of parameters, which range from disk geometry
specifications, to support for broken controller chips. Drive specific options
are specified by using `hdX=' with X in `a'-`h'.
Non-drive specific options are specified with the prefix `hd='. Note that using
a drive specific prefix for a non-drive specific option will still work, and
the option will just be applied as expected.
Also note that `hd=' can be used to refer to the next unspecified drive in the
(a, ..., h) sequence. For the following discussions, the `hd=' option will be
cited for brevity. See the file README.ide in linux/drivers/block for more
These options are used to specify the physical geometry of the disk. Only the
first three values are required. The cylinder/head/sectors values will be
those used by fdisk. The write precompensation value is ignored for IDE disks.
The IRQ value specified will be the IRQ used for the interface that the drive
resides on, and is not really a drive specific parameter.
The dual IDE interface CMD-640 chip is broken as designed such that when drives
on the secondary interface are used at the same time as drives on the primary
interface, it will corrupt your data. Using this option tells the driver to
make sure that both interfaces are never used at the same time.
This tells the IDE driver that there is an ATAPI compatible CD-ROM attached in
place of a normal IDE hard disk. In most cases the CD-ROM is identified
automatically, but if it isn't then this may help.
The standard disk driver can accept geometry arguments for the disks similar to
the IDE driver. Note however that it only expects three values (C/H/S) -- any
more or any less and it will silently ignore you. Also, it only accepts `hd='
as an argument, i.e. `hda=' and so on are not valid here. The format is as
If there are two disks installed, the above is repeated with the geometry
parameters of the second disk.
If you are unfortunate enough to be using one of these old 8 bit cards that move
data at a whopping 125kB/s then here is the scoop. If the card is not
recognised, you will have to use a boot arg of the form:
The type value specifies the particular manufacturer of the card, overriding
autodetection. For the types to use, consult the drivers/block/xd.c
source file of the kernel you are using. The type is an index in the list
xd_sigs and in the course of time types have been added to or deleted
from the middle of the list, changing all type numbers. Today (Linux 2.5.0)
the types are 0=generic; 1=DTC 5150cx; 2,3=DTC 5150x; 4,5=Western Digital;
6,7,8=Seagate; 9=Omti; 10=XEBEC, and where here several types are given with
the same designation, they are equivalent.
The xd_setup() function does no checking on the values, and assumes that you
entered all four values. Don't disappoint it. Here is an example usage for a
WD1002 controller with the BIOS disabled/removed, using the `default' XT
where `port' is the base address, `pro' is the protocol number, `uni' is the
unit selector (for chained devices), `mod' is the mode (or -1 to choose the
best automatically), `slv' is 1 if it should be a slave, and `dly' is a small
integer for slowing down port accesses. The `nice' parameter controls the
driver's use of idle CPU time, at the expense of some speed.
This CD-ROM interface is found on some of the Pro Audio Spectrum sound cards,
and other Sony supplied interface cards. The syntax is as follows:
Specifying an IRQ value of zero tells the driver that hardware interrupts aren't
supported (as on some PAS cards). If your card supports interrupts, you should
use them as it cuts down on the CPU usage of the driver.
The is_pas_card should be entered as `PAS' if using a Pro Audio Spectrum
card, and otherwise it should not be specified at all.
The wait_value is used as an internal timeout value for people who are
having problems with their drive, and may or may not be implemented depending
on a compile time #define. The Mitsumi FX400 is an IDE/ATAPI CD-ROM player and
does not use the mcd driver.
The driver assumes numbers between 3 and 11 are IRQ values, and numbers between
0x300 and 0x370 are I/O ports, so you can specify one, or both numbers, in any
order. It also accepts `cm206=auto' to enable autoprobing.
Different drivers make use of different parameters, but they all at least share
having an IRQ, an I/O port base value, and a name. In its most generic form,
it looks something like this:
The first non-numeric argument is taken as the name. The param_n values (if
applicable) usually have different meanings for each different card/driver.
Typical param_n values are used to specify things like shared memory address,
interface selection, DMA channel and the like.
The most common use of this parameter is to force probing for a second
ethercard, as the default is to only probe for one. This can be accomplished
with a simple:
Note that the values of zero for the IRQ and I/O base in the above example tell
the driver(s) to autoprobe.
The Ethernet-HowTo has extensive documentation on using multiple cards and on
the card/driver specific implementation of the param_n values where used.
Interested readers should refer to the section in that document on their
Sets the bitmask of allowed drives to mask. By default, only units 0 and 1 of
each floppy controller are allowed. This is done because certain non-standard
hardware (ASUS PCI motherboards) mess up the keyboard when accessing units 2
or 3. This option is somewhat obsoleted by the cmos option.
Sets the cmos type of drive to type. Additionally, this drive is allowed in the
bitmask. This is useful if you have more than two floppy drives (only two can
be described in the physical cmos), or if your BIOS uses non-standard CMOS
types. Setting the CMOS to 0 for the first two drives (default) makes the
floppy driver read the physical cmos for those drives.
Don't print a message when an unexpected interrupt is received. This is needed
on IBM L40SX laptops in certain video modes. (There seems to be an interaction
between video and floppy. The unexpected interrupts only affect performance,
and can safely be ignored.)
The sound driver can also accept boot args to override the compiled in values.
This is not recommended, as it is rather complex. It is described in the
Readme.Linux file, in linux/drivers/sound. It accepts a boot arg of the form:
where each deviceN value is of the following format 0xTaaaId and the bytes are
used as follows:
If this option is used, it should have precisely six parameters. Syntax:
The parameters maybe given as integers, or as strings. If strings are used, then
iobase and membase should be given in hexadecimal. The integer arguments
(fewer may be given) are in order: status (Enable(1) or Disable(0) this card),
type (PC/Xi(0), PC/Xe(1), PC/Xeve(2), PC/Xem(3)), altpin (Enable(1) or
Disable(0) alternate pin arrangement), numports (number of ports on this
card), iobase (I/O Port where card is configured (in HEX)), membase (base of
memory window (in HEX)). Thus, the following two boot prompt arguments are
More details can be found in /usr/src/linux/Documentation/digiboard.txt.
There are precisely 3 parameters; for several cards, give several `baycom='
commands. The modem parameter is a string that can take one of the values
ser12, ser12*, par96, par96*. Here the * denotes that software DCD is to be
used, and ser12/par96 chooses between the supported modem types. For more
details, see /usr/src/linux/drivers/net/README.baycom.
All parameters except the last are integers; the dummy 0 is required because of
a bug in the setup code. The mode parameter is a string with syntax hw:modem,
where hw is one of sbc, wss, wssfdx and modem is one of afsk1200, fsk9600.
You can tell the printer driver what ports to use and what ports not to use. The
latter comes in handy if you don't want the printer driver to claim all
available parallel ports, so that other drivers (e.g. PLIP, PPA) can use them
The format of the argument is multiple port names. For example, lp=none,parport0
would use the first parallel port for lp1, and disable lp0. To disable the
printer driver entirely, one can use lp=0.