Re: [RFC] Reentrant clock sources

["Magnus Damm" <magnus.damm@gmail.com>]

On Wed, Nov 26, 2008 at 6:22 AM, Thomas Gleixner <tglx@linutronix.de> wrote:
> On Tue, 25 Nov 2008, Magnus Damm wrote:
>> Hi everyone,
>>
>> Is there any special reason behind the non-reentrant clock source
>> code? I'm writing some timer help code and getting the struct
>> clocksource as argument to the callbacks would make the code much
>> cleaner and better.
>
> Why do you want that ? And what has reentrancy to do with the
> clocksource argument to read() ?

I should have picked my words more carefully, sorry about that. I
simply want to get some context so my callbacks can access
per-instance private data such as ioport address or irq number. Right
now I have to get the context from somewhere else.

>> Extending the callbacks to be able to start and stop clock sources
>> for improved power management would be good too in my opinion.
>> Any thoughts?
>
> What have you in mind there ? Starting / stopping a clocksource when
> what happens ? You can't stop them randomly except you want to screw
> timekeeping.

Is changing clock source using sysfs known to screw up the time keeping?

# echo jiffies >
/sys/devices/system/clocksource/clocksource0/current_clocksource

The line above seems to work well with dynamic ticks disabled, but the
current code doesn't seem to handle the nohz -> periodic transition
very well.

>> +     cycle_t (*vread)(struct clocksource *cs);
>
> This is crap. vread can not access the clocksource.

Yeah, sorry about that. =)

Thanks for your comments.

/ magnus