boot_cpuset(4)

NAME
     boot_cpuset.conf, boot_cpuset.so - boot cpuset configuration tool

SYNOPSIS
     boot_cpuset.so is loaded and executed by init

DESCRIPTION
     The boot_cpuset.so library provides a method for putting init(1M) into a
     cpuset(4). The cpuset that init(1M) will be placed into is named boot.
     Since all processes (threads of execution associated with a process
     identifier, or PID) are descendents of init(1M), they too will be
     confined to the boot cpuset.  For a process to execute outside the boot
     cpuset, another cpuset must be created and the process placed in the new
     cpuset.  Then that process, and all processses descended from it after it
     is placed in the new cpuset, will execute outside the boot cpuset.


     It is sometimes desirable to create a cpuset for system daemons and
     kernel threads.  This cpuset protects them from starvation and is called
     the system cpuset.  If this cpuset is created, unbound kernel threads and
     NFS server processes will run within it.


     The boot_cpuset.so library is normally installed on SGI ccNUMA systems.
     If it is found in /lib32 by init, it will be loaded and executed.  The
     behaviour of the boot cpuset is controlled by /etc/config/boot_cpuset and
     /etc/config/boot_cpuset.config. The former file follows the chkconfig(1M)
     convention and may be manipulated by chkconfig. If it contains the word
     on, creation of the cpuset will proceed unless an error occurs. If off is
     found, the library will check if a system cpuset is desired.  This is
     controlled by the /etc/config/system_cpuset and
     /etc/config/system_cpuset.config files.  These files operate the same as
     the boot cpuset files.  The library will lastly exit and normal init
     processing will resume.  The /etc/config/boot_cpuset.config and
     /etc/config/system_cpuset.config files follows the syntax convention of
     cpuset(4). All restrictions specified in miser_cpuset(1) apply.

     Confinement of bound kernel system and interrupt threads within the boot
     or system cpusets may be achieved by using the XThread Control Interface
     (XTCI).  This is sometimes desirable to keep IRIX kernel level daemons
     from competing with applications for certain processors.  The XTCI is
     documented in the realtime(5) man page.


FILES
     /etc/config/boot_cpuset           Follows chkconfig(1M) convention.

     /etc/config/boot_cpuset.config    Configuration file specifying the
                                       cpuset. See cpuset(4) for more
                                       information.
     /etc/config/system_cpuset           Follows chkconfig(1M) convention.  If
                                         off, no further processing occurs and
                                         init will resume.

     /etc/config/system_cpuset.config    Configuration file specifying the
                                         cpuset.  See cpuset(4) for more
                                         information.


WARNINGS
     No attempt is made to check the configuration file for a reasonable
     configuration. Specifying an unreasonable configuration may leave system
     non-functional.

     Note: The cpuset is not activated if the system is booted into single
     user mode. This provides an escape mechanism should the configuration
     prove to be unreasonable.

     When the boot cpuset is defined, processes that are moved out of the
     other cpusets do not, by default, return to the boot cpuset. To move
     processes from the other cpusets to the boot cpuset, specify the boot
     cpuset as the destination for the move operation. (see cpuset(1) for more
     information about moving processes between cpusets).

NOTES
     Allthough this mechanism can be used in a standalone fashion with the
     cpuset(1) command, it is primarily intended for use by batch systems. Due
     to the above noted limitation of moving processes between cpusets, it may
     not practical to dynamically re-adjust resources for running processes or
     jobs.

SEE ALSO
     init(1M), miser_cpuset(1), cpuset(4), realtime(5).