**** BEGIN LOGGING AT Fri Oct 07 02:59:58 2016 Oct 07 09:46:48 I had some problems with uio_pruss and the RT-kernel, switching back to the 4.4-ti kernel (non RT) I have problems with terrible scheduling performance.. CONFIG_HAVE_PREEMPT_LAZY=y, CONFIG_PREEMPT_VOLUNTARY=y, CONFIG_PREEMPT__LL is not set, CONFIG_PREEMPT_RTB is not set, CONFIG_PREEMPT_RT_FULL is not set. Does anyone know how I can get a bit more performance in the non RT kernel? Oct 07 10:14:33 _LL should be set Oct 07 10:17:16 voluntary preemption isn't preemption :P userspace still gets preempted, but the kernel itself effective does cooperative multitasking by only switching at predefined points Oct 07 10:17:27 *effectively Oct 07 10:19:12 I'd also suggest trying a -bone kernel. I don't know how much impact using an SMP kernel on a non-SMP cpu has, but presumably the impact is not zero Oct 07 10:53:56 how to connect eeprom(93c56)to beaglebone black through spi communucation Oct 07 10:59:28 by waiting more then a few minutes after asking your question :P Oct 07 10:59:59 jkridner: ping Oct 07 11:01:23 jkridner: https://beagleboard.org/chat has two dead links. the guide to asking smart questions should use http instead of https to fix it, the irc survival guide is archived here: https://web.archive.org/web/20150531122326/http://www.mahdiyusuf.com/post/6176918750/survival-guide-irc-1 Oct 07 11:02:30 jkridner: it might be useful to include the most important points inline on the https://beagleboard.org/chat page: 1 and 6, and maybe 2 Oct 07 11:02:57 I see soo many people leaving before they could have gotten any useful answer Oct 07 11:03:29 it probably hurts the reputation of the beaglebone community. e.g. I've seen some youtube video describe it as a desolate wasteland or something like that Oct 07 11:05:34 remind newbies that people may be in different timezones, or just paying occasional attention to irc, so they should try to ask their actual question as clearly as they can, and then be patient and leave their irc client connected Oct 07 11:06:13 (and never ask to ask, or for "help", since the probability of getting any reply at all is negligible) Oct 07 11:14:15 Привет Oct 07 11:23:05 zmatt: Do you think the bone kernel (non TI) would be better for multithreading? The RT kernel did the trick when it came to simultaneous spi reads, but then the uio_pruss driver does not work on that kernel.. Oct 07 11:24:45 I think recent bone kernels also use preempt-voluntary for some reason. just reconfigure and recompile the kernel to fix it Oct 07 11:26:03 thats something I have never done before.. Oct 07 11:26:37 it's not hard, you just need some linux host system (preferably a bit beefy to speed up the compilation) Oct 07 11:27:29 zmatt: there is no helping with people who dont understand that irc is not a paid-for help system Oct 07 11:28:36 tommer: do i understand you correctly, that you are doing two spi transactions from user space and have performance problems? Oct 07 11:28:59 KotH: I have no reason to believe those people think that Oct 07 11:29:43 zmatt: they expect help within minutes. they dont get that people on irc are just people like them and might not have the time to answer questions Oct 07 11:30:23 zmatt: over the years i've met a lot of people who think that on an "official" channel, people are being paid for helping others, especially if they paid for something like hardware Oct 07 11:30:27 KotH: I've had some performance issues when doing SPI reads in two different threads (on SPI0 and SPI1). I'll try to reconfigure and recompile Oct 07 11:30:46 KotH: the preferred behaviour here would be considered very bizarre in the Real World™: proclaim your question to a room full of people who don't seem to be paying attention to you and then just sit in a chair for hours Oct 07 11:31:02 zmatt: hehe Oct 07 11:31:03 zmatt: yes Oct 07 11:31:08 KotH: it is not an intuitive thing, even if it makes sense once you get it Oct 07 11:31:21 but it is something newbies unfamiliar with this need to know Oct 07 11:31:33 zmatt: it's intuitive on explicitly asynchronous media, like web forums Oct 07 11:31:57 zmatt: but people dont get that an user help channel is basically the same like a forum, minus the logging of questions and answers Oct 07 11:32:09 they are logged Oct 07 11:32:23 yes, but it's not the main form of communication Oct 07 11:32:29 you have to go looking for the log Oct 07 11:32:46 but yes, that's why a reminder of this should be on http://bbb.io/chat Oct 07 11:34:03 tommer: https://github.com/RobertCNelson/bb-kernel includes instructions (first select the branch corresponding to the kernel series you want) Oct 07 11:34:15 zmatt: ack Oct 07 11:35:23 I personally also prefer to add a line to version.sh, BUILD+="-mycustomsuffix" (replace by your preferred suffix) to distinguish the resulting build from the standard one Oct 07 11:36:23 if AUTO_BUILD is not set in system.sh then you'll be presented with an opportunity to reconfigure the kernel when you run ./build_deb.sh Oct 07 11:36:33 So, is there any reason to use the TI kernel at all? Oct 07 11:37:15 no idea, I use bone kernels anyway Oct 07 11:38:11 or rather, customized versions thereof ( https://github.com/dutchanddutch/bb-kernel ) Oct 07 11:48:50 zmatt: should I build debian package or kernel image? Oct 07 11:49:20 debian package is greatly more convenient Oct 07 11:49:50 like I said however I'd suggest making sure the version has a custom prefix so it can't be confused with the standard package Oct 07 11:49:55 erh Oct 07 11:49:59 custom suffix Oct 07 11:50:08 will do ! Oct 07 11:50:50 when it's done you'll find your linux-image-*.deb in deploy/ and you can copy it to your bbb and use dpkg -i to install it Oct 07 12:05:20 zmatt: where can I change the CONFIG_PREEMPT parameters? Oct 07 12:14:22 kernel features Oct 07 12:19:47 hmm, maybe we should send a search&rescue team into the kernel config menu to retrieve tommer Oct 07 12:22:44 haha :D No I tried to do this on the beefy server with 32 cores, but the lzop package was missing and I dont have sudo.. Just waiting for the IT guys now.. in the meantime Im going to try this on my laptop, cloning the linux-stable repo takes forever over wifi... Oct 07 12:23:37 yeah, the first time will take a bit Oct 07 12:24:02 but then I have to run the "build_kernel.sh" first Oct 07 12:24:11 no Oct 07 12:24:31 you generally use either build_kernel.sh or build_deb.sh Oct 07 12:25:01 and where do you find the kernel features then? Oct 07 12:25:33 both of them will present a kernel config menu after they've set up the source tree (unless AUTO_BUILD:=1 is set in system.sh) Oct 07 12:26:24 btw on the server I'd also suggest making sure ccache is installed, it helps recompilation time a lot Oct 07 12:27:14 (the scripts automatically take advantage of it) Oct 07 12:28:14 I must have missed something.. I cloned the correct branch in the bb-kernel, but I cant find any menu config.. Oct 07 12:29:41 it will take a bit after you've started ./build_deb.sh before it will present you with it since it first needs to prepare the source tree Oct 07 12:30:56 ok, then I will cancel build_kernel :P Oct 07 12:31:38 hmm, it should likewise present the menu though (after it has applied all the patches, but before compiling) Oct 07 12:32:14 adding BUILD+="-custom" in version.sh resulted in an error.. Oct 07 12:32:21 oh? Oct 07 12:32:45 BUILD+=-test not found.. weird Oct 07 12:33:04 maybe += is a bashism and your /bin/sh is something like dash Oct 07 12:33:09 yeah Oct 07 12:33:11 it didnt like that one Oct 07 12:33:17 works fine now Oct 07 12:33:23 in that case BUILD=${BUILD}-test Oct 07 12:57:00 zmatt: im still alive.. :D just extremely slow git cloning... I'll be back when something changes Oct 07 12:57:25 I'm probably afk for a bit, need to drag my ass to work Oct 07 12:57:55 later then :) **** BEGIN LOGGING AT Fri Oct 07 20:22:30 2016 Oct 08 01:22:18 zmatt, laurittr well I got some dev_dbg() in mcp251x, http://pastebin.com/MjLqwuA4 Oct 08 01:22:52 I'm still trying to figure out how to use dev_dbg() correctly and what structures, variables, etc I should be reading Oct 08 01:23:15 but looks like the clock is not getting set? Oct 08 01:24:52 or maybe something with pdat Oct 08 01:24:55 pdata Oct 08 01:26:35 pdata Oct 08 01:31:16 517 is EPROBE_DEFER Oct 08 01:31:39 ok funky, I thought we made that a fixed oscillator so it makes no sense it's getting deferred Oct 08 01:33:45 ohh wait Oct 08 01:34:20 hm no I didn't screw that up afaict Oct 08 01:38:16 the pdata struct btw is part the pre-DeviceTree way of doing board configuration, you can ignore it Oct 08 01:39:34 are you using an overlay or a custom device tree? Oct 08 01:42:32 does the clock show up in /sys/kernel/debug/clk ? Oct 08 02:14:41 I'm using that DT you created Oct 08 02:14:52 overlay you mean Oct 08 02:14:57 the clk folder is there Oct 08 02:15:02 yeah overlay, sorry Oct 08 02:15:22 that is... odd Oct 08 02:15:33 lot more folders in clk Oct 08 02:15:38 obviously Oct 08 02:15:53 there are quite a few clock signals on the am335x Oct 08 02:16:15 I don't see any with spi or mcp251x Oct 08 02:16:44 tt3201_clk Oct 08 02:16:53 is what I had named it in my overlay Oct 08 02:17:03 nope, not there Oct 08 02:17:16 * zmatt scratches head Oct 08 02:18:31 well that explains why the mcp driver couldn't find it Oct 08 02:18:44 doesn't explain why it isn't there Oct 08 02:19:45 ohh, wait a minute to create that clock I'm overlaying a node by path instead of by reference... I don't think I've ever tested that before Oct 08 02:20:26 https://github.com/mvduin/overlay-utils/blob/master/TT3201-001-01.dtsi#L31 Oct 08 02:20:55 try sticking it into &ocp instead of into / Oct 08 02:21:18 I'm not sure what you mean Oct 08 02:21:28 change l31 to &ocp ? Oct 08 02:21:33 &ocp { Oct 08 02:24:04 nope no clk Oct 08 02:27:35 weird, it looks like the overlay should be correct even with the / Oct 08 02:28:00 maybe a clock only shows up in debugfs once it's been requested.. not sure about that Oct 08 02:29:26 how did you find out what -517 was Oct 08 02:30:04 well the fact that it was repeatedly trying again already strongly suggested to me it was EPROBE_DEFER Oct 08 02:30:48 http://lxr.free-electrons.com/source/include/linux/errno.h you can find these special error codes here (always returned negated in the linux kernel) Oct 08 02:33:32 I notice in the old overlay there is no clk per say Oct 08 02:33:35 weird, maybe there's a problem with adding clock nodes using an overlay, since the custom clock node I added in my main device tree shows up just fine in /sys/kernel/debug/clk Oct 08 02:33:51 the current driver requires it though Oct 08 02:35:38 ok using / as target in an overlay works just fine, there's definitely something going wrong with introducing the clock node Oct 08 02:37:10 it does show up in /sys/bus/platform/devices though Oct 08 02:38:13 but clearly clocks are treated specially somehow, none of the clocks that show up in /sys/bus/platform/devices have a driver Oct 08 02:38:17 hmz Oct 08 02:39:16 https://groups.google.com/forum/#!searchin/beagleboard/SPI$20Robert%7Csort:date/beagleboard/vCFU_NWRdgo/7QGgua-QCgAJ Oct 08 02:39:46 something to late Oct 08 02:41:01 that's about pinmux though, and probably the issue that has been solved in the latest kernel releases Oct 08 02:43:20 I haven't gotten it to load the dtbo at bootup anyway Oct 08 02:43:30 I put the file in /lib/firmware Oct 08 02:43:40 but always fails Oct 08 02:44:02 easiest is probably changing the main device tree instead of using an overlay, or alternatively patching the driver to avoid the need for an actual clock node Oct 08 02:46:00 a dirty hack would be replacing the clock error case (lines 1069-1070 in http://pastebin.com/MjLqwuA4 ) by: freq = 16000000; Oct 08 02:47:34 but, well, apparently this is yet another example of stuff you can't do in an overlay (yet)... add new clock nodes Oct 08 02:48:05 probably the kernel scans for clock nodes once during early boot Oct 08 02:51:15 so is probably laurittr issue too then Oct 08 02:51:36 no idea, I thought he was using a custom device tree rather than an overlay Oct 08 02:58:41 if he is, he should also try an mcp driver similarly endowed with debug messages... without diagnostics it's impossible to say where it goes wrong Oct 08 02:59:39 if we had just an overlay with a clock in it, does that work? Oct 08 02:59:45 or that fails too? **** ENDING LOGGING AT Sat Oct 08 02:59:58 2016