lkml.org 
[lkml]   [2012]   [Apr]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: /proc/stat information incorrect
Thanks for your consideration Martin.
> CONFIG_IRQ_TIME_ACCOUNTING comes to mind. Is this enabled for your
> kernel ?

This configuration was not yet enabled.
I tried this configuration with and without the patch you mentioned
(git a25cac5198d4ff28).
The results are still not completely as expected:

With patch, with CONFIG_IRQ_TIME_ACCOUNTING:
user nice system idle iowait irq softirq sum
cpu 5 0 373 7610 0 55 382 8425
cpu0 0 0 1 1042 0 18 1 1062
cpu1 0 0 1 1058 0 1 0 1060
cpu2 0 0 1 1060 0 0 0 1061
cpu3 0 0 0 1061 0 1 0 1062
cpu4 2 0 75 505 0 36 380 998
cpu5 1 0 82 983 0 0 0 1066
cpu6 1 0 139 911 0 0 0 1051
cpu7 1 0 74 988 0 0 0 1063

Revoked patch and CONFIG_IRQ_TIME_ACCOUNTING:
user nice system idle iowait irq softirq sum
cpu 8 0 461 7854 0 55 448 8826
cpu0 0 0 1 1056 0 15 1 1073
cpu1 0 0 1 1073 0 0 1 1075
cpu2 1 0 1 1073 0 0 0 1075
cpu3 0 0 1 1217 0 0 1 1219
cpu4 3 0 147 511 0 39 447 1147
cpu5 1 0 106 967 0 0 0 1074
cpu6 1 0 99 972 0 0 0 1072
cpu7 1 0 106 985 0 1 1 1094

Looking at the sum column still some unexpected behaviour is seen.
The CONFIG_IRQ_TIME_ACCOUNTING seems to help but the interrupt
handling CPU (CPU4) is still "lazy" when the original code is used
(with patch).
If the patch is revoked some other CPUs seem to spend more time than
is really available.
Both scenarios should not be possible imho.


\
 
 \ /
  Last update: 2012-04-25 09:27    [W:2.149 / U:0.644 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site