/proc/stat: fix scalability of irq sum of all cpu
authorKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Wed, 27 Oct 2010 22:34:15 +0000 (15:34 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 28 Oct 2010 01:03:13 +0000 (18:03 -0700)
commit478735e38887077ac77a9756121b6ce0cb956e2f
tree7b42110f71b1711f56cbbc0a01f00894b221337d
parentf2c66cd8eeddedb440f33bc0f5cec1ed7ae376cb
/proc/stat: fix scalability of irq sum of all cpu

In /proc/stat, the number of per-IRQ event is shown by making a sum each
irq's events on all cpus.  But we can make use of kstat_irqs().

kstat_irqs() do the same calculation, If !CONFIG_GENERIC_HARDIRQ,
it's not a big cost. (Both of the number of cpus and irqs are small.)

If a system is very big and CONFIG_GENERIC_HARDIRQ, it does

for_each_irq()
for_each_cpu()
- look up a radix tree
- read desc->irq_stat[cpu]
This seems not efficient. This patch adds kstat_irqs() for
CONFIG_GENRIC_HARDIRQ and change the calculation as

for_each_irq()
look up radix tree
for_each_cpu()
- read desc->irq_stat[cpu]

This reduces cost.

A test on (4096cpusp, 256 nodes, 4592 irqs) host (by Jack Steiner)

%time cat /proc/stat > /dev/null

Before Patch:  2.459 sec
After Patch :   .561 sec

[akpm@linux-foundation.org: unexport kstat_irqs, coding-style tweaks]
[akpm@linux-foundation.org: fix unused variable 'per_irq_sum']
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Tested-by: Jack Steiner <steiner@sgi.com>
Acked-by: Jack Steiner <steiner@sgi.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/proc/stat.c
include/linux/kernel_stat.h
kernel/irq/irqdesc.c