[Xenomai-core] [BUG] timerbench tests don't work with SVN

[Xenomai-core] [BUG] timerbench tests don't work with SVN

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 550 bytes --]

Hi Philippe,

likely a regression of latest timer base patches: with periodic support
enabled (haven't tested the off case yet), latency -t1/2 no longer work.
I have no time to look at this, so I just dump my report here. :)

Jan


PS: Is there an official nucleus API to convert dates between timebases?
I'm asking because of the open question how to share RTDM drivers
between skins on different timer bases. Right now it looks to me like
RTDM is locked to one base (configured at compile or module load time)
and cannot be shared.


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]

[Xenomai-core] Re: [BUG] timerbench tests don't work with SVN

[Philippe Gerum]

On Thu, 2007-01-04 at 15:52 +0100, Jan Kiszka wrote:
> Hi Philippe,
> 
> likely a regression of latest timer base patches: with periodic support
> enabled (haven't tested the off case yet), latency -t1/2 no longer work.
> I have no time to look at this, so I just dump my report here. :)
> 
> Jan
> 
> 
> PS: Is there an official nucleus API to convert dates between timebases?
> I'm asking because of the open question how to share RTDM drivers
> between skins on different timer bases. Right now it looks to me like
> RTDM is locked to one base (configured at compile or module load time)
> and cannot be shared.
> 

This is the same situation than before, except that now, RTDM can choose
the timebase it is locked on. There is no conversion API right now, but
we could define xntbase_convert(inbase, rtdm_base), with inbase being
extracted from xnthread_timebase(xnpod_current_thread()) for primary
context services. The question remains open on how to get the inbase
pointer for seconday mode callers though.

-- 
Philippe.

[Xenomai-core] Re: [BUG] timerbench tests don't work with SVN

[Philippe Gerum]

On Thu, 2007-01-04 at 15:52 +0100, Jan Kiszka wrote:
> Hi Philippe,
> 
> likely a regression of latest timer base patches: with periodic support
> enabled (haven't tested the off case yet), latency -t1/2 no longer work.
> I have no time to look at this, so I just dump my report here. :)
> 

Btw, it works fine on my testboxes UP/SMP, so this may be a
config-dependent issue which I have overlooked. Please send me
your .config. TIA,

> Jan
> 
> 
> PS: Is there an official nucleus API to convert dates between timebases?
> I'm asking because of the open question how to share RTDM drivers
> between skins on different timer bases. Right now it looks to me like
> RTDM is locked to one base (configured at compile or module load time)
> and cannot be shared.
> 
-- 
Philippe.

Re: [Xenomai-core] Re: [BUG] timerbench tests don't work with SVN

[Philippe Gerum]

On Thu, 2007-01-04 at 16:16 +0100, Philippe Gerum wrote:
> On Thu, 2007-01-04 at 15:52 +0100, Jan Kiszka wrote:
> > Hi Philippe,
> > 
> > likely a regression of latest timer base patches: with periodic support
> > enabled (haven't tested the off case yet), latency -t1/2 no longer work.
> > I have no time to look at this, so I just dump my report here. :)
> > 
> > Jan
> > 
> > 
> > PS: Is there an official nucleus API to convert dates between timebases?
> > I'm asking because of the open question how to share RTDM drivers
> > between skins on different timer bases. Right now it looks to me like
> > RTDM is locked to one base (configured at compile or module load time)
> > and cannot be shared.
> > 
> 
> This is the same situation than before, except that now, RTDM can choose
> the timebase it is locked on. There is no conversion API right now, but
> we could define xntbase_convert(inbase, rtdm_base), with inbase being
> extracted from xnthread_timebase(xnpod_current_thread()) for primary
> context services. The question remains open on how to get the inbase
> pointer for seconday mode callers though.
> 

Mm, silly me: xnthread_timebase(xnshadow_thread(current)) if the
argument is non-null, or we would force the master time base if current
is not a real-time shadow, IOW plain Linux tasks would always call -nrt
services with nanosecond based timeout and delays. So this should be a
no-brainer.

-- 
Philippe.

[Xenomai-core] Re: [BUG] timerbench tests don't work with SVN

[Philippe Gerum]

On Thu, 2007-01-04 at 15:52 +0100, Jan Kiszka wrote:
> Hi Philippe,
> 
> likely a regression of latest timer base patches: with periodic support
> enabled (haven't tested the off case yet), latency -t1/2 no longer work.
> I have no time to look at this, so I just dump my report here. :)

Ok, fixed by #2054.

-- 
Philippe.