I have a fix which works for me: https://github.com/mhaberler/linuxcnc/commit/10924b5509a2d2e38533a9f9e876d139a2cf5d4b (branch unified-build-candidate-3)
I'd appreciate reports - this should repair the "latency-test: jitter going down" error NB: this error did show only on i386 kernels, not on x86_64 the clocksource was not involved - Michael Am 26.01.2014 um 18:02 schrieb Peter C. Wallace <[email protected]>: > On Sun, 26 Jan 2014, Michael Haberler wrote: > >> Date: Sun, 26 Jan 2014 17:40:19 +0100 >> From: Michael Haberler <[email protected]> >> To: Peter C. Wallace <[email protected]>, >> EMC developers <[email protected]> >> Subject: Re: [Emc-developers] reports on rt-preempt / latency-test: max >> jitter >> going down >> >> Am 26.01.2014 um 17:28 schrieb Peter C. Wallace <[email protected]>: >> >>> On Sun, 26 Jan 2014, Michael Haberler wrote: >>> >>>> Date: Sun, 26 Jan 2014 09:59:52 +0100 >>>> From: Michael Haberler <[email protected]> >>>> To: EMC developers <[email protected]> >>>> Subject: Re: [Emc-developers] reports on rt-preempt / latency-test: max >>>> jitter >>>> going down >>>> could the folks observing this behaviour please run the following commands >>>> on their systems: >>>> >>>> cat /sys/devices/system/clocksource/clocksource0/available_clocksource >>>> >>>> cat /sys/devices/system/clocksource/clocksource0/current_clocksource >>>> >>>> background: Answer 2 of: >>>> http://stackoverflow.com/questions/5930621/problem-of-understanding-clock-gettime >>>> >>>> thanks! >>> >>> >>> cat /sys/devices/system/clocksource/clocksource0/available_clocksource >>> tsc acpi_pm >>> >>> cat /sys/devices/system/clocksource/clocksource0/current_clocksource >>> tsc >> >> hm, I would have expected to find 'hpet' there (see also link below) >> maybe for some reason HPET isnt enabled in the kernel so not settable >> >> (is that by chance my howegrown 3.1x-mah rt-preempt kernel..?) > > Yes > >> >> >>> >>> >>> I changed the clocksource to acpi_pm but the error remains >> >> acpi_pm is probably the worst of all in terms of resolution, 3.58Mhz: >> http://stackoverflow.com/questions/7987671/what-is-the-acpi-pm-linux-clocksource-used-for-what-hardware-implements-it >> >> it's probably a good idea to figure out the best clock source, and make sure >> in the startup phase it's actually selected (assuming this is related at all) >> >>> One interesting thing is that the error is periodic with a ~4 second period >>> (like an uncaught 32 bit ns count wrap) >> >> hm, thats an interesting observation. Need to read the timedelta and >> rtapi_time code once more to see if there's something smelly. >> >> does that 4 second period show up with either clock source just alike? if >> so, this would suggest the clock source not be the cause > > > > Yes, same ~4 second period with either clock source, I HAL 'scoped > the baddelta staircase and its 4.3 seconds as you would expect with a > uncaught 32 bit wrap > >> >> I should be next to real hardware soon >> >> thanks! >> >> - Michael >> >>> >>> >>> >>> >>>> >>>> - Michael >>>> >>>> >>>> Am 25.01.2014 um 17:03 schrieb Michael Haberler <[email protected]>: >>>> >>>>> >>>>> Peter Wallace kindly help to locate the error, it is rtapi_get_time() >>>>> going backwards occasionally on rt-preempt, suggesting the cause is this >>>>> function: >>>>> >>>>> https://github.com/mhaberler/linuxcnc/blob/e3de5bb74857bbd9935c1a1d1f031a0558ca0f85/src/rtapi/rtapi_time.c#L166 >>>>> >>>>> this cause the 'del' variable in >>>>> https://github.com/mhaberler/linuxcnc/blob/e3de5bb74857bbd9935c1a1d1f031a0558ca0f85/src/hal/components/timedelta.comp >>>>> to become negative at times, causing the decreasing jitter >>>>> >>>>> no fix yet, but we know where to look >>>>> >>>>> -m >>>>> >>>>> >>>>> Am 25.01.2014 um 08:43 schrieb Michael Haberler <[email protected]>: >>>>> >>>>>> Reading back on IRC I found hints at rt-preempt ubc3 builds showing the >>>>>> max latency going down occasionally. >>>>>> >>>>>> Please: >>>>>> >>>>>> 1. report the exact kernel version, in particular if this is stock >>>>>> debian rt-preempt kernel. >>>>>> 2. check if ntp is running or not, i.e. the output of 'ps ax|grep ntp' >>>>>> 3. If yes, please stop the ntp demon with 'sudo /etc/init.d/ntp stop' or >>>>>> equivalent >>>>>> 4. Re-run latency-test after ntp is stopped. Does the problem reoccur? >>>>>> 5. Please report your results using the ubc issue tracker as mentioned >>>>>> here >>>>>> http://static.mah.priv.at/public/html/common/UnifiedBuild.html#_issue_tracker >>>>>> , for this issue: https://github.com/zultron/linuxcnc/issues/72 >>>>>> >>>>>> background: there is a chance that the issue is related to the behavior >>>>>> described here, in particular the subtle difference between >>>>>> CLOCK_MONOTONIC and CLOCK_MONOTONIC_RAW: >>>>>> >>>>>> http://stackoverflow.com/questions/3657289/linux-clock-gettimeclock-monotonic-strange-non-monotonic-behavior >>>>>> http://markmail.org/thread/54bb663vi47kjxnu >>>>>> >>>>>> - Michael >>>>>> >>>>>> >>>>>> >>>>>> ------------------------------------------------------------------------------ >>>>>> CenturyLink Cloud: The Leader in Enterprise Cloud Services. >>>>>> Learn Why More Businesses Are Choosing CenturyLink Cloud For >>>>>> Critical Workloads, Development Environments & Everything In Between. >>>>>> Get a Quote or Start a Free Trial Today. >>>>>> http://pubads.g.doubleclick.net/gampad/clk?id=119420431&iu=/4140/ostg.clktrk >>>>>> _______________________________________________ >>>>>> Emc-developers mailing list >>>>>> [email protected] >>>>>> https://lists.sourceforge.net/lists/listinfo/emc-developers >>>>> >>>>> >>>>> ------------------------------------------------------------------------------ >>>>> CenturyLink Cloud: The Leader in Enterprise Cloud Services. >>>>> Learn Why More Businesses Are Choosing CenturyLink Cloud For >>>>> Critical Workloads, Development Environments & Everything In Between. >>>>> Get a Quote or Start a Free Trial Today. >>>>> http://pubads.g.doubleclick.net/gampad/clk?id=119420431&iu=/4140/ostg.clktrk >>>>> _______________________________________________ >>>>> Emc-developers mailing list >>>>> [email protected] >>>>> https://lists.sourceforge.net/lists/listinfo/emc-developers >>>> >>> >>> Peter Wallace >>> Mesa Electronics >>> >>> (\__/) >>> (='.'=) This is Bunny. Copy and paste bunny into your >>> (")_(") signature to help him gain world domination. >> > > Peter Wallace > Mesa Electronics > > (\__/) > (='.'=) This is Bunny. Copy and paste bunny into your > (")_(") signature to help him gain world domination. ------------------------------------------------------------------------------ CenturyLink Cloud: The Leader in Enterprise Cloud Services. Learn Why More Businesses Are Choosing CenturyLink Cloud For Critical Workloads, Development Environments & Everything In Between. Get a Quote or Start a Free Trial Today. http://pubads.g.doubleclick.net/gampad/clk?id=119420431&iu=/4140/ostg.clktrk _______________________________________________ Emc-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/emc-developers
