Re: realtime-preempt-2.6.12-final-V0.7.51-11 glitches [no more]

["Rui Nuno Capela" <rncbc@rncbc.org>]

> * Rui Nuno Capela <rncbc@rncbc.org> wrote:
>
>> OTOH, I'll take this chance to show you something that is annoying me
>> for quite some time. Just look to the attached chart where I've marked
>> the spot with an arrow and a question mark. Its just one example of a
>> strange behavior/phenomenon while running the jack_test4.2 test on
>> PREEMPT_RT kernels: the CPU usage, which stays normally around 50%,
>> suddenly jumps to 60% steady, starting at random points in time but
>> always some time after the test has been started. Note that this
>> randomness surely adds to the the slight differences found on the
>> above results.
>
> how long does this condition persist? Firstly, please upgrade to the
> -51-16 kernel, previous kernels had a condition where interrupt storms
> (or repeat interrupts) could occur. (Your irqs/sec values dont suggest
> such a condition, but it could still occur.)
>
> Then could you enable profiling (CONFIG_PROFILING=y and profile=1 boot
> parameter), and create a script like this to capture a kernel profile
> for a fixed amount of time:
>
>  #!/bin/bash
>
>  readprofile -r          # reset profile
>  sleep 10
>  readprofile -n -m /home/mingo/System.map | sort -n
>
> and start it manually when the anomaly triggers. Also start it during a
> 'normal' period of the test. The output should give us a rough idea of
> what is happening. This type of profiling is very low-overhead so it
> wont disturb the condition.
>
> Note that you can increase the frequency and the quality of profiling by
> enabling the NMI watchdog (LOCAL_APIC in the .config and nmi_watchdog=2
> boot option), in the -RT kernel it will automatically switch the
> profiling tick to occur from NMI context. Such tracing will also show
> overhead occuring in irqs-off functions.
>

After several trials, with CONFIG_PROFILING=y and profile=1 nmi_watchdog=2
as boot parameters, I'm almost convinced I'm doing something wrong :)

- `readprofile` always just outputs one line:

     0 total                                    0.0000

- `readprofile -a` gives the whole kernel symbol list, all with zero times.

Is there anything else I can check around here?
-- 
rncbc aka Rui Nuno Capela
rncbc@rncbc.org

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/