Re: 2.6.14-rc4-rt1

[Fernando Lopez-Lezcano <nando@ccrma.Stanford.EDU>]

On Wed, 2005-10-12 at 15:08 -0700, George Anzinger wrote:
> Steven Rostedt wrote:
> > On Wed, 12 Oct 2005, Ingo Molnar wrote:
> > 
> > 
> >>i'm not sure latency traces will uncover anything useful for this bug.
> >>Your problems could be timer issues: timers going off too fast cause
> >>high keyboard repeat rates, and the same goes for the screensaver. Does
> >>'sleep 1' work as expected, or is that timing out in an "accelerated"
> >>way too?
> >>
> > 
> > 
> > I usually recommend doing a 'sleep 10'.  It really shows you if things are
> > wrong.  If a sleep 1 returns 2 seconds, or 0.5 seconds later it may not be
> > detected.  But a sleep 10 returning 20 seconds or 5 seconds later is
> > obvious.
> 
> Or maybe:
> 'time sleep 10'
> 
> Lets the machine time it.

My first thought was "this can't work" as I imagined the same timing
services would be used and you would get always 10 secs or so...

Ingo: I tried with PREEMPT_RCU=y and it made no difference. 

When the system starts to misbehave I tried 'time sleep 10' and got
really wild results:

# time sleep 10

real    0m10.007s
user    0m0.001s
sys     0m0.003s
# time sleep 10

real    0m10.006s
user    0m0.000s
sys     0m0.003s
# time sleep 10
[the return key "autorepeated" here :-]



real    0m10.006s
user    0m0.001s
sys     0m0.003s
#
#
#
# time sleep 10

real    0m0.016s
user    0m0.002s
sys     0m0.001s
[yes I really got the prompt back that fast!]
#
#
# time sleep 10

real    73m18.087s
user    0m0.000s
sys     0m0.003s
[this last one was also very fast, it did not take 73 minutes...]
# time sleep 10

real    73m18.089s
user    0m0.000s
sys     0m0.003s
#
#
# time sleep 10

real    0m10.006s
user    0m0.000s
sys     0m0.004s
# time sleep 10

real    73m18.069s
user    0m0.000s
sys     0m0.003s
# time sleep 10

And so on and so forth. At least this shows the problem clearly.
I still have to test with a up kernel and try again PREEMPT_RT.

-- Fernando