Li, Aubrey wrote:
> [snip]
>>>
>>> the sum of these functions is ~100. That's the tick frequency.
>>> But take cbe_hres_tick as an example,
>>> it looks like cbe_hres_tick is called only one time per second.
>>> Actually it's called once per tick, every nsec_per_tick ns.
>>> So this is not correct, please remove this option, :-)
>> I agree it looks a bit strange, but it is an artifact of the cyclic
>> subsystem batch processing cyclics. It just happens that
>> unix`cbe_hres_tick is usually fired after `clock and `cyclic_timer. So
>> the DTrace script is only reporting the first event per expire
>> timestamp.
>
> Yeah, I see the point, it always count the first cyclic being batch
> processed.
> And it proves the cyclic is really being batch processed.(sum= tick
> freq).
>> I think this is a good feature for developers interested in the cyclic
>> subsystem, and the behavior could be explained in the manual. What do
>> you think ?
>
> I'm sorry I still can't agree with you, :-)
> because the report is wrong. the frequency is wrong, and from the
> report,
> We can't see cyclic is being batch processed, that means cyclic behavior
> is wrong.
>
> Both if you persist it's a good feature, you have to persuade me why,
> :-)
Well, I just think it's a cool feature since cyclic batch processing is
not a well documented behavior. I'll look for a better way of displaying it.
>>> ["-c" option] keeps the current script, but I think help message is
>>> not okay.
>>>
>>> "-c --cyclic report cyclic subsystem activety\n"
>> Hmm.. did you import my patch? That's not what I'm seeing here, could
>> you double check?
>>
> ================================================
> [aubrey-laptop at amd64]./powertop -h
> Usage: powertop [OPTION]
> -d, --dump Read wakeups once and print list of top
> offenders
> -t, --time=DOUBLE Default time to gather data in seconds, ranging
> between 1 and 100s
> -h, --help Show this help message
> -c, --cyclic Report cyclic subsystem activity
> -C, --collapse Collapse events that are batch processed by the
> cyclic subsyste
> ================================================
>
> It would be better something like this, instead of just cyclic
> ----snip----
> -v, --verbose show more kernel events
> ----snip----
Hmm.. I still prefer -c, as verbose is pretty general and doesn't tell
much to the user. But as I said earlier, I just joined the project. I'll
go for what you guys think it's best :)
>>> It would be better to use this option to place powertop into the
>>> verbose mode, reporting more events in the kernel.
>>>
>>> [default option], I'd like to do the following change, because clock
>>> event is missing.
>>>
>>> static const char *g_prog_A =
>>> ----snip----
>>> +++ "sdt:::cyclic-start"
>>> +++ "/(caddr_t)((cyclic_t *)arg0)->cy_handler == (caddr_t)&`clock/"
>>> +++ "{" +++" @events_k[(caddr_t)((cyclic_t
> *)arg0)->cy_handler] =
>>> count(); "
>>> +++"}";
>> That goes a little bit against your first option of not reporting
>> cyclics at all :) But I agree with you.
>
> er..., I don't want to see the batch processed cyclic, but I still
> want to see the clock frequency, so clock is the best one, :-)
>
> Thanks,
> -Aubrey
> _______________________________________________
> tesla-dev mailing list
> tesla-dev at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/tesla-dev
