Kernel boot command-line parameter reference: Chapter 9 - Linux Kernel in a Nutshell

Options

Send kernel console data across the network using UDP packets to another machine. Options are:

To listen to this data, the remote machine can use the syslogd program, or run the netcat program as follows:

	                    netcat -u -l -p 
                     port

For more background on how to use this option, see the file Documentation/networking/netconsole.txt.

Cause the kernel log level to be set to the debug level, so that all debug messages will be printed to the console at boot time.

Set the default kernel log level to KERN_WARNING (4), which suppresses all messages during boot except extremely serious ones. (Log levels are defined under the loglevel parameter.)

Show kernel log messages that precede the initialization of the traditional console. These messages are typically never seen on the console unless you use this option. Enabling this can be very useful for tracking down hardware issues. Currently, the option can specify either the VGA device or the serial port, but not both at the same time. Also, only the ttyS0 or ttyS1 serial devices will work. Interaction with the standard serial driver is not very good, and the VGA output will eventually be overwritten by the real console.

Append ,keep in order not to disable the messages shown by this option when the real kernel console is initialized and takes over the system.

Specify the initial console log level. Any log messages with levels less than this (that is, of higher priority) will be printed to the console, whereas any messages with levels equal to or greater than this will not be displayed.

The console log level can also be changed by the klogd program, or by writing the specified level to the /proc/sys/kernel/printk file.

The kernel log levels are:

Set the size of the kernel's internal log buffer. n must be a power of 2, if not, it will be rounded up to be a power of two. This value can also be changed by the CONFIG_LOG_BUF_SHIFT kernel configuration value.

Cause the kernel to trace all functions that are called by the kernel during initialization of the system as the kernel boots. This option is useful for determining where the kernel is dying during startup.

Specify how many words from the kernel stack should be printed in the kernel oops dumps. n is an integer value.

Cause the kernel to prefix every kernel log message with a timestamp.

Control how much information the APIC subsystem generates when booting the kernel.

Prevent the kernel from using any of the IOAPICs that might be present in the system.

Cause the kernel to enable the local APIC even if the BIOS had disabled it.

Tell the kernel not to use the local APIC.

Disable all of the built-in kernel IRQ balancing logic.

When an interrupt is not handled, search all known interrupt handlers for it. This is intended to get systems with badly broken firmware running.

When an interrupt is not handled, search all known interrupt handlers for it and also check all handlers on each timer interrupt. This is intended to get systems with badly broken firmware running.

By default, the kernel attempts to detect and disable unhandled interrupt sources because they can cause problems with the responsiveness of the rest of the kernel if left unchecked. This option will disable this logic.

Force the highmem memory zone to have an exact size of n bytes. This will work even on boxes that have no highmem zones by default. It can also reduce the size of the highmem zone for machines with a lot of memory.

The hugetlb feature lets you configure Linux to use 4MB pages, one thousand times the default size. If Linux is configured this way, this options sets the maximum number of hugetlb pages to be n.

Override the default number of hash buckets for the kernel's inode cache. Recommended only for kernel experts.

Cause the kernel to ignore all physical memory greater than or equal to the physical address n.

Set the specific ammount of memory used by the kernel. When used with the memmap= option, physical address space collisions can be avoided. Without the memmap= option, this option could cause PCI devices to be placed at addresses that belong to unused RAM. n specifies the amount of memory to force and is measured in units of kilobytes (K), megabytes (M), or gigabytes (G).

Disable the use of huge (4MB) pages for kernel memory.

Use a specific memory map. The exactmap lines can be constructed based on BIOS output or other requirements.

Force the kernel to use a specific memory region. n is the size of the memory location and startis the start location in memory of the range. Units can be kilobytes (K), megabytes (M), or gigabytes (G).

Enable or disable the kernel's ability to map sections of memory as non-executable. By default, the mapping is enabled (on).

Force the kernel to ignore some of the I/O memory areas.

Force vmalloc to have the exact size specified by n. This can be used to increase the minimum size of the vmalloc area (which is 128 MB on the x86 processor). It can also be used to decrease the size and leave more room for directly mapped kernel RAM.

By default, the kernel randomizes the address space of all programs when they are started. This option disables this feature. It is equivalent to writing 0 to the file /proc/sys/kernel/randomize_va_space.

Disable (0) or enable (1) the VDSO (Virtual Dynamic Shared Object) mapping option. By default, it is enabled.

Tell the kernel which disk device contains the suspended kernel image. If the data on the image is a valid kernel image created by the software suspend subsystem, it will be loaded into memory and the kernel will run it instead of continuing on with the normal boot process. suspend_device is the kernel device name, which might be different from what userspace thinks the device name is, so be careful with this option.

Disable the resume functionality of the kernel. Any swap partitions that were being used to hold system images to which the kernel could be restored will revert back to available swap space.

Sometimes CPU hardware bugs make them report the cache size incorrectly. The kernel will attempt to work around and fix known problems with most CPUs, but for some CPUs it is not possible to determine what the correct size should be. This option provides an override for these situations. n is measured in units of bytes.

Specify the loops per jiffy that should be used by the kernel, and thus have the kernel avoid the time-consuming boot-time autodetection of this value. If n is 0, the value will be autodetected as usual.

This is a debugging feature that allows the user to override the default non-maskable interrupt (NMI) watchdog value. 0 specifies that no NMI watchdog should be used. 1 specifies that the APIC should be used if present. 2 specifies that the local APIC should be used if present. 3 means that the NMI watchdog is invalid, so do not use it.

Always use the 387 math emulation library, even if a 387 math coprocessor is present in the system.

Disable the x86 floating-point extended register save and restore. The kernel will save only legacy floating-point registers on a task switch.

This option is available because the HLT instruction does not work correctly for some x86 processors. This option tells the kernel not to use the instruction.

Some processors can check for machine errors (usually errors in the hardware). This option turns this subsystem on, if it has been built into the kernel configuration.

This option turns the subsystem off.

Disable x86 SYSENTER/SYSEXIT support in the kernel. This can cause some system calls to take longer.

Tell an SMP kernel to act as a uniprocessor kernel, even on a multiprocessor machine.

Disable the time stamp counter hardware in the system, if present.

Specify the maximum number of processors that a SMP kernel should use, even if there are more processors present in the system.

Remove the specified CPUs, as defined by the cpu_number values, from the general kernel SMP balancing and scheduler algroithms. The only way to move a process onto or off an "isolated" CPU is via the CPU affinity syscalls. cpu_number begins at 0, so the maximum value is 1 less than the number of CPUs on the system.

This option is the preferred way to isolate CPUs. The alternative, manually setting the CPU mask of all tasks in the system, can cause problems and suboptimal load balancer performance.

This is a debugging option that overrides the default scheduler migration cost matrix. The numbers specified by level-N-useconds are indexed by the "CPU domain distance" and are measured in microseconds.

An example of this option is migration_cost=1000,2000,3000 for a SMT NUMA machine. It sets up an intra-core migration cost of 1 ms, another inter-core migration cost of 2 ms, and another internode migration cost of 3 ms.

Set the migration cost debug level. If 0 is specified, no extra messages will be printed to the kernel log. This is the default value. 1 prints some information on how the matrix is determined. 2 is very verbose and is useful only if you use a serial console, as the amount of information will overflow the kernel log buffer.

Modify the default migration costs by the specified percentage. This is a debugging option that can be used to proportionally increase or decrease the auto-detected migration costs for all entries of the migration matrix. For example, migration_factor=150 increases migration costs by 50%, so the scheduler will be less eager to migrate cache-hot tasks. migration_factor=80 decreases migration costs by 20%, thus making the scheduler will be more eager to migrate tasks.

Specify where the initial ramdisk for the kernel boot is located.

If n is set to 1, a ramdisk is loaded by the kernel at boot time from the floppy drive.

Do not load any initial ramdisk, even if it is configured in other options passed to the kernel.

Prompt the user for the initial ramdisk before attempting to read it from the floppy drive.

Tell the ramdisk driver how many bytes to use per block. The default size is 1024.

Specify the size of the initial RAM disk in kilobytes. The default size is 4096 (4 MB). This option should be used instead of the older ramdisk command-line option.

The default for the kernel is to mount the root device as read-only at boot time. This option ensures that this is the mode the kernel uses. It overrides the rw command-line option, if it had been specified earlier on the boot command line.

Tell the kernel which disk device the root filesystem image is on. device can be specified in one of the following ways:

Wait n seconds before trying to mount the root filesystem. This can be useful if the root filesystem is on a USB or Firewire device, as those disk devices take a bit longer to be discovered by the kernel.

Mount options that the kernel should use in mounting the root filesystem. The options value depend on the filesystem type; see the documentation for the individual types for details on what is valid.

Try to mount the root filesystem as this type of filesystem. For instance, rootfstype=ext3.

The default for the kernel is to mount the root device as read-only at boot time. This option mounts the root device as read-write instead.

Run the specified binary as the init process instead of the default /sbin/init program.

Run the program specified by full_path_name as the init process. This file must be on the kernel ramdisk instead of on the root filesystem.

The default for the kernel is to run init in multi-user mode. This option runs init in single-user mode instead.

The kexec subsystem likes to have a portion of physical memory reserved for it. This option reserves that memory from the rest of the kernel and will switch to use it if the kernel panics. n specifies the amount of memory to reserve, and start specifies the location for this memory chunk. Both are measured in units of kilobytes (K), megabytes (M), or gigabytes (G).

The kernel, like every Linux executable, is stored in ELF format. This option specifies the physical address where the kernel core image's ELF header starts. This is used by kexec to find the kernel when booting the secondary kernel image.

Set the maximum number of finished RCU callbacks to process in one batch.

Batch limiting is disabled when the number of queued RCU callbacks rises above n.

Batch limiting is re-enabled when the number of queued RCU callbacks falls below n.

Set the number of additional RCU callbacks that should bee queued before forcing a reschedule on all CPUs.

This is the main option for the Advanced Configuration and Power Interface (ACPI). Values are:

During S3 resume (which happens after the machine has been suspended to RAM), hardware needs to be reinitialized properly. For most devices this is simple, except for video cards, which are normally initialized by the BIOS. The kernel does not have enough information to restore the video device, because that information is in the BIOS and not accessable at all. This option lets the kernel try to use the ACPI subsystem to restore the video card in two different ways.

See the file Documentation/power/video.txt for more information on this option and how to find the proper value for your type of hardware.

Set the ACPI System Control Interrupt trigger mode.

Cause ACPI to balance the active IRQs. This is the default option when operating in APIC mode.

Cause ACPI not to move the active IRQs. This is the default option when operating in PIC mode.

If the IRQ balance option is enabled, mark the listed IRQs as used by the ISA subsystem.

If the IRQ balance option is enabled, mark the listed IRQs as used by the PCI subsystem.

Tell the ACPI BIOS that the name of the running operating system is name. This can be useful to spoof the BIOS into thinking that Windows is running instead of Linux, which can help solve some ACPI issues for older BIOSes. As an example, use the string Microsoft 2001 to spoof the BIOS into thinking that Windows 2001 is running on the machine.

This is actually a binary option despite the integer value. If n is absent, ACPI will disable the _OSI method. If n is present, _OSI will not be disabled.

Force the serialization of ACPI Machine Language methods.

Allow the ACPI layer to recognize and ignore IRQ0/pin2 Interrupt Override issues for broken nForce2 BIOSes that result in the XT-PIC timer acting up.

Set the ACPI debug layers. n is an integer in which each bit indicates a different ACPI debug layer. After the system has booted, the debug layers can be set via the /proc/acpi/debug_layer file.

If present, this allows ACPI to workaround BIOS failures when it lacks an Embedded Controller Description Table.

This allows the ACPI consolidated generic hotkey driver to override the platform-specific driver if one is present.

Force the kernel to assume that the machine's pmtimer latches its value and always returns good values.

Specify the ACPI embedded controller interrupt mode. If n is 0, polling mode will be used, otherwise interrupt mode will be used. Interrupt mode is the default.

Marks a specific location and range of memory as ACPI data. n is the size of the memory location and startis the start location in memory of the range. Both are measured in units of kilobytes (K), megabytes (M), or gigabytes (G).

This marks a specific location and range of memory as reserved. n is the size of the memory location and start is the start location in memory of the range.

Disable the PnP ACPI functionality.

Limit the processor to a maximum C-state, no matter what the ACPI tables say it can support. n is a valid C-state value. A value of 9 overrides any DMI blacklist limit that might be present for this processor.

Causes the ACPI core to ignore the _CST method of determining the processor C-states and use the legacy FADT method instead.

Specify the maximum number of SCSI LUNS that the system should probe. n is an integer from 1 to 4294967295.

Specifu the maximum number of SCSI LUNs that the system can receive. n is an integer from 1 to 16384.

This option lets the user add entries to the SCSI black/white list for a specific vendor and model of device.

Each option can be one of the following:

Disable the ISA PnP subsystem, if it has been enabled in the kernel configuration.

Set the main PnP BIOS settings. on enables the PnP BIOS subsystem. off disables the PnP BIOS subsystem. curr tells the PnP BIOS subsystem to use the current static settings and no-curr tells the subsystem to probe for dynamic settings if possible.

List of the IRQs that the PnP BIOS subsystem should not use for autoconfiguration.

List of the DMAs that the PnP BIOS subsystem should not use for autoconfiguration.

I/O ports that the PnP BIOS subsystem should not use for autoconfiguration. Each port is listed by its starting location and size.

Memory regions that the PnP BIOS subsystem should not use for autoconfiguration. Each region is listed by its starting location and size.

Set the initial checkreqprot flag value. 0 means that the check protection will be applied by the kernel and will include any implied execute protection. 1 means that the check protection is requested by the application. The default value is set by a kernel configuration option.

The value can be changed at runtime via the /selinux/checkreqprot file.

Specify whether SELinux enforces its rules upon boot. 0 means that SELinux will just log policy violations but wil not deny access to anything. 1 means that the enforcement will be fully enabled with denials as well as logging. The default value is 0.

The value can be changed at runtime via the /selinux/enforce file.

This option allows SELinux to be enabled (1) or disabled (0) to boot time. The default value is set by a kernel configuration option.

If SELinux is enabled at boot time, the /selinux/disable file can be used later to disable it prior to the initial policy load.

Set the initial value for the SELinux network control model. 0 uses the new secmark-based packet controls, and 1 uses the legacy packet controls. 0 is the default and preferred value.

This value can be changed at runtime via the /selinux/compat_net file.

Specify network device parameters, which are specific to the driver used by the network device. Some drivers' source files document the applicable options. This option does not usually apply to PCI, USB, or other plug-and-play network devices. It is intented for use only on devices that can not discover their own resource assignments.

This option lets you override the default number of hash buckets for the kernel's route cache. Recommended only for kernel network experts. are doing.

This option lets you set the maximum number of network shapers that the kernel can use.

This option lets you override the default number of hash buckets for the kernel's TCP connection cache.

Set the NFS lock manager grace period. n is measured in seconds.

Set the TCP port that the NFS lock manager should use. port must be a valid TCP port value.

Override the default time value for the NFS lock manager. n is measured in seconds. If this option is not specified the default of 10 seconds will be used.

Set the UDP port that the NFS lock manager should use. port must be a valid UDP port value.

Set the NFS root filesystem for diskless boxes, to enable them to boot properly over NFS. If this parameter is not set, the value /tftpboot/ client_ip_address will be used as the root filesystem with the default NFS options.

Specify the TCP port that the NFSv4 callback channel should listen on. port must be a valid TCP port value.

Specify the maximum lifetime for idmapper cache entries. n is measured in seconds.

If this option is present, the USB subsystem will not be initialized.

Specify the parallel port to use. The lp=port1,port2... format associates a sequence of parallel ports to devices, starting with lp0. An example is lp=none,parport0, which would suppress configuration of the lp0 device and cause the lp1 device to use the first parallel port.

lp=0 disables the printer driver.

lp=reset causes the attached printers to be reset. This option can be combined with the port specifications.

lp=auto causes the kernel to examine the device ID from each port to determine whether a IEEE 1284-compatible printer is attached. If so, the kernel will manage that printer.

Specify specific settings for parallel port drivers. Parallel ports are assigned in the order they are specified on the command line, starting with parport0.

auto forces the driver to use any IRQ/DMA settings detected (the default is to ignore detected IRQ/DMA settings because of possible conflicts). You can also specify the base address, IRQ, and DMA settings in the format 0x nnnn[,irq[,dma]]. irq and DMA can be numbers, auto to use detected settings on that particular port, or nofifo to avoid using a FIFO even if it is detected.

Specifies the mode that the parallel port should be operated in. This is necessary on the Pegasos computer where the firmware has no options for setting up the parallel port mode. This option works for parallel port chips of type 686a and 8231.

Specifies the maximum number of different UARTs that can be registered in the kernel.

Enable pin 1 of the APIC timer. This option can be useful to work around chipset bugs (on some ATI chipsets in particular.) The kernel tries to set a reasonable default, but sometimes this option is necessary to override it.

Disable pin 1 of the APIC timer. Useful for the same reasons as enable_timer_pin_1.

Enable interrupt 0 timer routing over the 8254 chip in addition to routing over the IO-APIC. The kernel tries to set a reasonable default but sometimes this option is necessary to override it.

Disable interrupt 0 timer routing over the 8254 chip in addition to routing over the IO-APIC. The kernel tries to set a reasonable default but sometimes this option is necessary to override it.

Disable the HPET timer source and tell the kernel to use the PIT timer source instead.

Override the default kernel clocksource and use the clocksource with the specified name instead.

This option lets you override the default number of hash buckets for the kernel's dentry cache. Recommended only for kernel experts.

Specify the I/O scheduler. See IOSCHED_NOOP for a list of the different I/O schedulers available, and what they do.

Large hashes that are allocated during the boot process on the IA-64 platform are, by default, distributed across the different NUMA nodes. This option lets the user turn this option on or off.

Control which driver uses the IDE ports in combined mode: the legacy IDE driver, libata, or both. Note that using the ide or libata options may affect your device naming (e.g., by changing hdc to sdb).

Specify the maximum number of loopback filesystem devices that can be mounted at the same time. n is an integer from 1 to 256.

Specify the ammount of time in seconds that the kernel should wait after a panic happens before it reboots. If this is set to 0 (the default value) the kernel will not reboot after panicking; it will simply halt.

Tell the kernel to halt all CPUs after the first oops for n seconds before continuing. This is useful if oopses keep scrolling off of the screen before you can write them down or take a picture of them.

This option affects how the kernel profiler is calculated. If schedule is specified, the schedule points are affected by the value set in number. If schedule is not specified, number is the step size as a power of two for statistical time-based profiling in the kernel.

The most common use of this option is profile=2