Bug#657916: linux-source-2.6.32: ps time doubled then constant: missing lock for task_utime?

[paul.szabo@sydney.edu.au]

I wrote:
  I only observed this for multi-threaded
  processes compiled with  -fopenmp  .
I think I now observed the same issue with a single-threaded process:

$ ps u -p 14252
USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
duncans  14252  150  9.7 2458868 2408272 ?     RN   Jan03 71589:24 ./compact
$ grep . /proc/14252/stat /proc/14252/task/*/stat
/proc/14252/stat:14252 (compact) R 1 14222 14218 0 -1 4202496 366809489 0 0 0 429428470 108008 0 0 36 16 1 0 130475208 2517880832 602068 4294967295 1 1 0 0 0 0 0 0 132 4294967295 0 0 17 0 0 0 0 0 0
/proc/14252/task/14252/stat:14252 (compact) R 1 14222 14218 0 -1 4202496 366809489 0 0 0 429428468 95107 0 0 36 16 1 0 130475208 2517880832 602068 4294967295 1 1 0 0 0 0 0 0 132 4294967295 0 0 17 0 0 0 0 0 0

Should I investigate, should I try to reproduce, and check by how much
do TIME and %CPU jump when the "wrong results" start?

Thanks,

Paul Szabo   psz@maths.usyd.edu.au   http://www.maths.usyd.edu.au/u/psz/
School of Mathematics and Statistics   University of Sydney    Australia