[Xenomai-core] Clockevent/clocksource.

[Xenomai-core] Clockevent/clocksource.

[Gilles Chanteperdrix]

Hi,

I would like to implement the (I-pipe) AT91 timer management as
clocksource/clockevent, however, I have a question, how does one choose
the "shift" value in struct clocksource and struct clock_event_device ?

TIAç

-- 


					    Gilles Chanteperdrix.

Re: [Xenomai-core] Clockevent/clocksource.

[Philippe Gerum]

Gilles Chanteperdrix wrote:
> Hi,
> 
> I would like to implement the (I-pipe) AT91 timer management as
> clocksource/clockevent, however, I have a question, how does one choose
> the "shift" value in struct clocksource and struct clock_event_device ?
> 

http://ixlabs.cs.pub.ro/category/projects/linux-kernel-library/

> TIAç
> 


-- 
Philippe.

Re: [Xenomai-core] Clockevent/clocksource.

[Gilles Chanteperdrix]

Philippe Gerum wrote:
 > Gilles Chanteperdrix wrote:
 > > Hi,
 > > 
 > > I would like to implement the (I-pipe) AT91 timer management as
 > > clocksource/clockevent, however, I have a question, how does one choose
 > > the "shift" value in struct clocksource and struct clock_event_device ?
 > > 
 > 
 > http://ixlabs.cs.pub.ro/category/projects/linux-kernel-library/

Looks like the thing is working. A concern though: the hardware timer
used on AT91 wraps fast, so if we want the tsc to work correctly, the
timer needs to tick every 20ms (which makes using the clock events
framework pretty useless, but it is mostly useful for other reasons: it
allows me to test Xenomai on ARM in the clock events case). I have set
the clock event max_delta_ns member to these 20ms, however when
inserting the nucleus module, the max_delta_ns member is set to
ULONG_MAX, so my question is: will the nucleus take care of ticking
every 20ms ?

-- 


					    Gilles Chanteperdrix.

Re: [Xenomai-core] Clockevent/clocksource.

[Gilles Chanteperdrix]

Gilles Chanteperdrix wrote:
 > Philippe Gerum wrote:
 >  > Gilles Chanteperdrix wrote:
 >  > > Hi,
 >  > > 
 >  > > I would like to implement the (I-pipe) AT91 timer management as
 >  > > clocksource/clockevent, however, I have a question, how does one choose
 >  > > the "shift" value in struct clocksource and struct clock_event_device ?
 >  > > 
 >  > 
 >  > http://ixlabs.cs.pub.ro/category/projects/linux-kernel-library/
 > 
 > Looks like the thing is working. A concern though: the hardware timer
 > used on AT91 wraps fast, so if we want the tsc to work correctly, the
 > timer needs to tick every 20ms (which makes using the clock events
 > framework pretty useless, but it is mostly useful for other reasons: it
 > allows me to test Xenomai on ARM in the clock events case). I have set
 > the clock event max_delta_ns member to these 20ms, however when
 > inserting the nucleus module, the max_delta_ns member is set to
 > ULONG_MAX, so my question is: will the nucleus take care of ticking
 > every 20ms ?

Ok. This is now how max_delta_ns is supposed to be used anyway: enabling
NO_HZ breaks the tsc, even if max_delta_ns is not set to ULONG_MAX.

-- 


					    Gilles Chanteperdrix.