HP-UX Reference (11i v2 07/12) - 1M System Administration Commands A-M (vol 3)

i
intctl(1M) intctl(1M)
NAME
intctl - manage the interrupt configuration of the system
SYNOPSIS
/usr/contrib/bin/intctl
[-h
| -F | -p | -c cpu_id]
/usr/contrib/bin/intctl
[-C
class][-H hw_path]
/usr/contrib/bin/intctl
[-M -H
hw_path -I intr_id -c cpu_id]
/usr/contrib/bin/intctl
[
-r file | -s file]
/usr/contrib/bin/intctl
[-b
[-w][-a algorithm]
[
-i io:hw_path:intr_id]...
[
-i cpu:hw_path]...
[-i drv:driver_name]...
[-o drv:driver_name:weight]...
[
-o trig:balance_on_cpu:distribute_to_cpu]]
DESCRIPTION
A processor receives an interrupt when either the processor’s interrupt pin is asserted (for line based inter-
rupts) or if a processor detects an interrupt message bus transaction on the system bus (for transaction
based interrupts).
Interrupts from the interface cards can be line or transaction based. Interrupts are routed to different pro-
cessors during boot time.
The
intctl command is a tool that allows a performance expert to display and modify these interrupt
assignments. The tool only supports migration of external device interrupts. The performance analyst can
also save and restore the interrupt configuration. If interrupt migration process completes successfully, a
message is logged to the console and/or to the /var/adm/syslog/syslog.log
file.
intctl resides in /usr/contrib/bin
, and the command can be executed only by the superuser. The
intctl command is not a general system administration command. It should be used only by perfor-
mance tuning experts with a high level of system knowledge. The performance specialist can use the
intctl command to view the interrupt configuration of the system and modify the interrupt assignments
of the CPUs to re-distribute the system load across the CPUs.
intctl is synchronized with other High Availability (HA) events happening simultaneously on the sys-
tem. An HA event can be a PCI OLA/R or Processor allocation/de-allocation. If any of these events are hap-
pening when intctl is trying to display interrupt information or is trying to migrate an interrupt to a
CPU, intctl will exit with the error message below, and the user should retry the intctl command:
"Another HA event is in progress, try again!"
Interrupt migration is not enabled on workstations. Also non-MP safe drivers do not support interrupt
migration. The tool will display an error message if the user tries to move the interrupts of a non-MP safe
driver to a different CPU.
On a system with virtual partitions (vPars), intctl will only display CPUs in the current partition.
Using the
-b option, the command can be used to balance interrupt distribution on a system. Interrupt
assignments in a given system may not be distributed in a balanced manner. Most of the time, imbalance
in distribution is caused by CPU migrations.
These migrations may cause the interrupts to get assigned to CPUs available in the system in a non-
optimal fashion and they will not get distributed when more CPUs become available.
By default, HP-UX will distribute interrupts at boot time based on the round robin method. Each interrupt
owned gets assigned to the available CPUs in a round robin fashion. However, CPU migrations, which may
happen because of Work Load Manager (WLM) configuration, vPar administration activity, and Instant
Capacity (iCAP/TiCAP) administration activity, can cause the interrupts to be assigned to a smaller set of
CPUs causing an imbalance and thus a non-optimally configured system.
Using the -b option allows the user to manually balance the interrupt distribution of the system. Users
can choose one of these two balancing algorithm of their choice to balance interrupts:
• driver_weight
The default balancing algorithm used by intctl is driver_weight, which associates weights to
each driver based on its interrupt frequency and balances the system such that each CPU is loaded with
HP-UX 11i Version 2: December 2007 Update − 1 − Hewlett-Packard Company 401