**** BEGIN LOGGING AT Wed Aug 24 02:59:58 2016 Aug 24 08:54:07 hello all. I've an udp server running on bbb. UDP payloads sent from clients are small (8-32 bytes). When running server, I can see ksoftirqd eating a lot of CPU for handling interrupts. Then my first idea was to try changing some NIC parameters using ethtool. However i always get "operation not supported", even for queries. (e.g. ethtool -g eth0) Aug 24 08:54:08 can you help me? Aug 24 08:58:19 alfatau: have you checked the driver source? maybe it does not support such options Aug 24 09:05:38 lindi: no I haven't. I took a look at driver user's guide http://processors.wiki.ti.com/index.php/AM335x_CPSW_%28Ethernet%29_Driver%27s_Guide and I didn't find any know limitations... Aug 24 09:09:43 think you'll find ksoftirqd is probably hung up on emmc/uSD accesses .. not your network interface. Aug 24 09:10:15 or at least that was the last thing I remember bein associated with it. Aug 24 09:17:27 veremit: this http://pastebin.com/CLwFZ8fD is what pointed me to network interface Aug 24 09:32:42 alfatau: ah interesting Aug 24 09:32:54 that just tells you the number though .. Aug 24 09:33:14 you would expect that if there is data flowing, you -would- have interrupts unless its DMA'ing Aug 24 09:33:53 have you tried a 'top' or 'htop' ? Aug 24 09:34:01 at least one has a tree view too Aug 24 09:44:19 veremit: i tried top. i've not understood what you mean with "if there is data flowing, you -would- have interrupts unless its DMA'ing"... I also tried to enable "interrupt pacing" (ethtool -C eth0 rx-usecs 500) as suggested here: http://processors.wiki.ti.com/index.php/AM335x_CPSW_%28Ethernet%29_Driver%27s_Guide#Interrupt_Pacing. But there's no di Aug 24 09:44:19 fference. Given the previous link, do you know what are Timer 5 & 6? Aug 24 09:44:47 newp, sorry .. that's about all I can shed light on :/ Aug 24 13:36:30 has anyone used the seeed fusion service? Aug 24 15:02:08 Ciao Aug 24 15:27:56 could someone advise on a board i could use to compile software and deploy my bbb system images? i need a compatible board (except for the capes, of course) but with more performance (ram, cpu cores, etc) Aug 24 15:30:17 is PWM output subject to distortion when the CPU becomes heavily loaded? Aug 24 15:31:48 samael: what level of compatability do you need? would a wandboard (imx6 arm) work? It wouldn't be able to run the same kernel Aug 24 15:36:50 MrCurious: Afaik PWM is a completel independent hardware subsystem that doesn't require the CPU except to configure its registers. The datasheet would say more, but that's pretty standard for SoC. Aug 24 15:46:03 ddrown: the best fit would be kernel compatible, so that i'd use the dev board to compile software, setup the system, and test all the software parts prior to clone it to BBBs Aug 24 15:47:34 samael: ah, that's harder Aug 24 15:53:49 ddrown: it's not entirely clear to me from where the incompatibility comes from. different cpu arch? Aug 24 15:54:16 samael: arm is a mess of vendor-specific kernel trees Aug 24 15:55:47 wonderful. how could i proceed to indentify the right board, then? Aug 24 15:56:23 your best bet if you want the same kernel is to use a beaglebone itself Aug 24 15:56:49 it's easy(ish) to cross compile the kernel on a different machine Aug 24 15:57:13 True that. I do it all day every day. Aug 24 15:57:19 Ragnorok: ty! Aug 24 15:57:37 One is happy to be of service. (wink) Aug 24 15:59:01 samael: I have a debian VM I build on. I do mixed BBB / Win development so I run X on win and connect to CCS on the VM. Drag and drop in Dolphin to put builds on the device, run them. It's not one-click automagic, but it's repeatable and works well enough. Aug 24 16:00:55 I could probably configure a post-build step to drop them on the device automagically but I've never gotten around to that. Aug 24 16:04:32 i never expored similar solutions (until now) because at the moment we're happy with the stock rcn kernels. but instead i usually need to do a lot of work at system level, and together with software tests they take considerable time to be done on the BBB itself Aug 24 18:01:07 Hi, I was looking for modern replacements to http://www.febo.com/time-freq/ntp/soekris/ for an accurate time server. This led me to https://www.febo.com/pipermail/time-nuts/2014-November/088385.html which indicate the BBB can indeed be a nanosecond accurate time server which is great news Aug 24 18:01:31 Reading http://blog.dan.drown.org/tcxo-beaglebone-black/ the author states "Another problem is that since this is a 1.364MHz clock, the clock's precision is limited to 733ns. This is much less precise than what the hardware could do (41ns)." Aug 24 18:02:10 Strykar: yeah, that's because I was using an external 1.364MHz clock Aug 24 18:02:19 you can use an external 24MHz clock Aug 24 18:02:38 Oh hi! nice writeup Aug 24 18:03:26 the driver just works on the 3.8 kernel now (the debian 7.x image). I've been working on porting it to 4.4 (debian 8.x image) Aug 24 18:04:24 ddrown, and a 24Mhz clock get us close to the max precision the system will allow? In part 2 of the blog, you state "After a day, it looks like it's around 100ppb/C. An improvement, but I think it can do better than this." Aug 24 18:04:32 what would you do to better this!? Aug 24 18:05:05 24MHz is the upper limit of the input frequency, so that would give you the highest precision Aug 24 18:05:13 I suppose a higher quality TCXO? Aug 24 18:05:21 yeah, or an OCXO Aug 24 18:05:30 or something even more fancy Aug 24 18:05:53 you might also be interested in my holdover tests - https://blog.dan.drown.org/gps-pps-drift-when-it-has-no-signal/ Aug 24 18:08:27 damn Aug 24 18:08:41 yeah, much better results than I would have expected Aug 24 18:09:11 what's fancier than an ocxo? I'm not an electronics guy Aug 24 18:09:41 rubidium, cesium - atomic clocks Aug 24 18:09:52 tbh, you've appeared to hit the lower measurement floor of ntpd Aug 24 18:10:31 yeah, chrony seems to have better results at this scale than ntpd/ntpsec Aug 24 18:10:58 ah, yea, way over budget. My idea was to use a good tc/ocxo and test with an RTC and a GPS PPS for clock discipline Aug 24 18:11:48 heh, there are used rubidium standards on ebay but they are more expensive Aug 24 18:12:04 shipping kills me on those, I'd love to Aug 24 18:14:12 I wanted to do this with a newer ublox chip that does GPS, Galileo, GLONASS and Beidou figuring averaging all 4 systems should give me even better accuracy and holdover Aug 24 18:14:50 cool. My beaglebone right now has a GPS+GLONASS module Aug 24 18:15:19 but I haven't had problems with signal to my GPS-only modules with an external antenna on the window Aug 24 18:15:38 so not the navspark Aug 24 18:17:06 I'm using four modules right now, two adafruit (MTK3339), a Navspark mini (venus 8x), and a Navspark-GL (venus 8x) Aug 24 18:17:45 mini does GPS+Beidou, -GL does GPS+GLONASS Aug 24 18:18:01 quick question, someone on the time-nuts list said "Also, simon the plls can be entirely bypassed. The circuits aren't clearly exposed but I believe you could get the peripheral clock covered and definitely the Ethernet clock - interesting from a 1588 perspective." Any ideas on how to provide the ethernet clock with the new clock source? PTP would be interesting on the BBB Aug 24 18:18:53 do you query and sample all 4 concurrently and then use an average dropping outliers? Aug 24 18:18:58 so the new model of 1588 in linux is to sync the ethernet clock off the system clock Aug 24 18:19:05 but the beaglebone's drivers don't seem to be using this new model Aug 24 18:19:17 I have multiple systems, each with one GPS module Aug 24 18:19:43 I didn't see the point in connecting all the modules to one system - that seems overkill Aug 24 18:19:50 oh so it's purely a software fix to make ethernet on the BBB use the new clock source? Aug 24 18:20:05 ddrown, of course not, I was thinking out aloud :) Aug 24 18:20:35 I should restate that - the ethernet driver doesn't use a different clock source, you use software to sync the two clock sources Aug 24 18:21:03 https://www.mankier.com/8/phc2sys Aug 24 18:21:34 this might be a better page for that - https://www.suse.com/documentation/sles-12/book_sle_tuning/data/tuning_ptp_phc2sys.html Aug 24 18:21:58 they're talking about slaving the system clock off the ethernet clock Aug 24 18:22:09 but you can go the other way if you're a PTP source instead of a client Aug 24 18:25:31 gotcha, thanks. what's the chipset on the ethernet, I'm lookin at the wiki and it just says 10/00 Aug 24 18:26:35 built into the ti chip - uses the cpsw driver Aug 24 18:27:02 ddrown, assuming you've seen these - http://www.leobodnar.com/shop/index.php?main_page=product_info&cPath=107&products_id=234 http://www.force12inc.com/products/gps-locked-precision-frequency-reference-low-jitter-gps-clock-450-hz-to-800-mhz-output-custom-low-phase-noise-xtal-version.html Aug 24 18:28:16 not those specifically, but I have seen GPSDO's Aug 24 18:47:36 ddrown, silly question, will a 24.576Mhz or a 25Mhz xo do? seems 24Mhz aren't easily available, I'm still looking Aug 24 18:47:53 I think both of those are out of spec Aug 24 18:48:09 I believe the datasheet has the limits in it Aug 24 18:48:14 for the ti cpu Aug 24 19:53:11 Strykar: you mean for the main osc of an am335x? 19.6 MHz, 24 MHz, 25 MHz, or 26 MHz .. although the last option comes with some caveats iirc Aug 24 19:53:19 sorry, 19.2 Aug 24 19:57:16 (they're not "limits", only those four specific values are permitted and you need to strap sysboot14-15 appropriately to indicate the choice made. you could deviate from this but this means ROM cannot correctly setup the PLLs hence timing-sensitive boot options will not work) Aug 24 20:00:01 zmatt: yeah, changing the main osc is an option too but harder than using tclkin Aug 24 20:00:46 I didn't read the whole context, I saw noted the "will a 24.576Mhz or a 25Mhz xo do? seems 24Mhz aren't easily available" and assumed it was about the main osc Aug 24 20:03:51 unfortunately neither tclkin nor an externally provided ethernet phy clock can be used as ethernet tsync reference clock Aug 24 20:06:26 tsync is the clock running the 1588 timer? Aug 24 20:06:37 yes, used for packet timestamping Aug 24 20:06:43 ok, that's fine Aug 24 20:06:58 that's how most other 1588 systems are setup too Aug 24 20:07:11 you just have to sync between the system clock and the NIC's 1588 clock Aug 24 20:07:50 the 1588 clock is synchronous to the system clock Aug 24 20:08:50 (via divider m5 of the core PLL) Aug 24 20:09:02 well, as long as your system clock is from the internal clocksource Aug 24 20:09:27 with "system clock" I mean the clock which the am335x considers to be the system clock, i.e. its main osc Aug 24 20:09:34 that's not true if your system clock (I'm using this to mean the kernel tracking of time) comes from tclkin Aug 24 20:09:48 yeah, I'm using a different meaning of system clock :) Aug 24 20:10:21 I think kernel tracking time by default derives from the system clock also, but I'm not 100% sure Aug 24 20:10:22 maybe kernel timebase would be a better term for me Aug 24 20:10:22 zmatt, thanks for clearing that up. Aug 24 20:10:49 yeah, it defaults to using a timer on the (processor) system clock Aug 24 20:11:41 from a power management point of view (and possibly long-time stability) it would be better if it used the 32768 Hz rtc clock instead probably Aug 24 20:12:18 (I recently saw some e2e post mentioning time tends to drift relative to the rtc) Aug 24 20:12:28 that has a downside of losing precision, but some people might value the lower power use Aug 24 20:12:55 well you can combine timebases for precision probably Aug 24 20:13:20 the absolute ppm error of the timebase isn't as big of an issue as the changes due to temperature (and other factors) Aug 24 20:13:44 although, that also depends on your application Aug 24 20:14:00 if you can calibrate it once. if you can't then absolute ppm error is just as big a problem Aug 24 20:14:07 unless you don't care at all, in which case neither is a problem :P Aug 24 20:14:11 exactly Aug 24 20:14:22 yep, for the casual ntpd, temps cause the biggest swings Aug 24 21:05:10 how hard is it to replace the main oscillator? can't seem to find articles that have done it Aug 24 21:08:30 the oscillator? Aug 24 21:09:51 you'll need to adjust the am335x pins too .. the ones that config the chip Aug 24 21:10:19 the oscillator itself is on the AM335x die Aug 24 21:11:13 is there a P/N for the new clock you would like? Aug 24 21:12:48 Strykar: https://github.com/beagleboard/BeagleBone-Black/blob/master/BBB_SCH.pdf - Y2 Aug 24 21:13:22 part Y2 Aug 24 21:13:32 on "Sheet 3 of 11" Aug 24 21:15:47 SYSBOOT[15:14] values tell the processor what speed to expect (Sheet 6 of 11) Aug 24 21:24:58 thank you Aug 24 21:26:04 I'm assuming you're using the Beaglebone Black Aug 24 21:48:46 the bbb is what I was looking at Aug 24 21:50:05 x15 availability? Aug 24 21:50:21 Mouser no longer has a delivery date listed Aug 25 01:02:29 I have plans to build a network of about 250 Beaglebone Greens for a data acquisition and control setup. Does anybody know if another Sitara-based board is in the pipeline? Aug 25 01:03:46 I just came across Octavo's OSD335x, but that's a bit too low-level for me Aug 25 01:04:02 and given the things I'd want to attach, I don't think it'd actually be any cheaper than a Green Aug 25 01:06:17 As for the BeagleBoard x15, that's way too much muscle; I think the price per analog acquisition channel is higher than the Green **** ENDING LOGGING AT Thu Aug 25 02:59:58 2016