[PATCH 0/4] Fix periodic itimers precision -v4

[Stanislaw Gruszka <sgruszka@redhat.com>]

Hello,

I'm resending patches for fixing itimers periodic ticks precision,
for reference see: http://lkml.org/lkml/2009/4/2/230

Changelog:

v3 -> v4
- Rebase for 2.6.31

v2 -> v3 (resend 5):
- Use CPUCLOCK_PROF CPUCLOCK_VIRT defines in kernel/fork.c
- Using inline function cputime_sub_ns() with sanity check against negative
  results instead of macro.
- Add patch which optimize jiffies_to_cputime(1) for PowerPC with
  CONFIG_VIRT_CPU_ACCOUNTING 

v1 -> v2 (resend 4):
- Add comment about CPUCLOCK_PROF and CPUCLOCK_VIRT are 0 and 1 respectively and
  can be used for indexing itimers array.
- Remove not needed BUG().
- Coding style fixes.
- Better algorithm (idea form Thomas) to calculate next signal time for periodic
  itimers,  we use now precomputed nanosecond value of cputime 1 tick.
- Add patch with simplify arm_timer() function.

v0 -> v1 (resend 3):
- Fix bug: using utime instead of sum stime + utime when setup PROF itimer.
- Fix bug: using value->it_interval instead of ovalue->it_interavl.
- No new definitions IT_PROF and IT_VIRT only CPUCLOCK_PROF and CPUCLOCK_VIRT
  are used.

Cheers
Stanislaw