**** BEGIN LOGGING AT Thu Oct 29 02:59:59 2015 Oct 29 08:19:52 Hello, is it possible to remove a recipe from the PNBLACKLIST, without edditing the recipe, where it was added to the list? Oct 29 08:30:26 DatGizmo: grep PNBLACKLIST to find it Oct 29 08:31:39 I think it is listed in conf/distro/*.conf Oct 29 08:45:05 5a1/wii stephan__ Oct 29 08:45:07 oops Oct 29 08:45:19 :) Oct 29 09:10:05 bluelightning: ping Oct 29 09:30:47 parrot1: pong Oct 29 09:30:50 morning all Oct 29 09:36:27 <[Sno]> RP: can I somehow help on the autotools fix approval I submitted? Oct 29 09:46:57 [Sno]: he's away on holiday atm I believe Oct 29 09:48:40 <[Sno]> bluelightning: thanks, I just hoped I catch him or rburton because of the mails both wrote this morning Oct 29 09:49:20 <[Sno]> holiday is a perfect excuse for being unresponsive :D Oct 29 11:50:45 which would be a latest x86_64 machine target ? Oct 29 12:34:26 <[Sno]> bluelightning: view the pictures at https://github.com/libusb/libusb/issues/16 Oct 29 12:35:06 <[Sno]> I rate this - maybe not critical, but - highly important for long running embedded devices ;) Oct 29 12:35:57 [Sno]: as I said in my reply, at this point we'd still apply the patch in preference to the upgrade unless applying the patch was more risky than the upgrade Oct 29 12:36:28 <[Sno]> so the patch fixing that issue would be ok, but the update not Oct 29 12:40:58 it's not up to me, but in theory yes - it's still getting pretty late for any kind of patch though Oct 29 12:56:12 * zeddii is back after 3 days travelling .. sees his name highlighted .. and can't find it :) Oct 29 12:58:01 zeddii: I think Crofton was trying to get hold of you Oct 29 12:58:36 * zeddii knows you linked me to the archives before. but I lost that link too :) Oct 29 12:58:36 heh Oct 29 12:58:47 three times in fact, must have been something important ;) Oct 29 12:58:49 maybe it's a ETOOMANYINPUTS defence mechanism Oct 29 12:58:55 https://www.yoctoproject.org/irc/ Oct 29 12:59:07 I need to remember to shut down my freenode connection when I'm stuffed onto planes Oct 29 12:59:19 otherwise, I just look like I'm ignoring everyone Oct 29 12:59:25 maybe I am .. maybe I'm not ;) Oct 29 12:59:47 probably safer to have the option of doing that ;) Oct 29 13:00:42 yep it was Crofton, each day Oct 29 13:01:05 Crofton or Crofton|work .. I'm back at my desk if you still need me! Oct 29 13:01:20 or if you had a complaint. I'm on a plane Oct 29 13:01:22 and this is a bot Oct 29 13:01:53 hehe Oct 29 13:13:55 hi folks, I'm trying to boot from a .hddimg on a usb but hangs while booting because of an infinite loop in init_live.sh when looking for rootfs in /run/media. Any help would be appreciated. Oct 29 13:45:11 <[Sno]> bluelightning: wrt libusb - I like the proposal from otavio Oct 29 13:45:24 [Sno]: that's definitely a possibility Oct 29 13:47:00 zeddii, figured out problem Oct 29 13:47:15 thought I was in the kernel for a bit , but was confused Oct 29 13:47:45 zeddii, patch works slight different on F20 and F21 :( Oct 29 13:50:23 ech. fun. Oct 29 13:50:31 this bot, is not amused! Oct 29 13:52:17 I wasn't either Oct 29 13:52:29 respects exec flag on 20, but not on 21 Oct 29 14:00:43 <_valle_> Hi, I'm having trouble with Qt Bluetooth. Would really appreciate if someone could give me a pointer Oct 29 14:00:48 <_valle_> http://pastebin.com/0hfenDnu Oct 29 14:06:05 NULL Oct 29 14:09:46 _valle_ caught signal: 11 Oct 29 14:11:23 jatac: how did you flash the usb key? Oct 29 14:12:07 bluelightning: hi Oct 29 14:12:24 bluelightning: any python build system experts here? Oct 29 14:13:07 <_valle_> fredcadete: No Oct 29 14:13:30 <_valle_> fredcadete: ;) Oct 29 14:17:18 hi - I'm working on a project that recently moved from fido to jethro - most things worked just fine but a kernel (3.19 + some vendor patches) that I was building for an mx5 target no longer works (no output after u-bot loads the kernel) Oct 29 14:18:15 the kernel tree and config are the same, the first thing that springs to mind is maybe it's a gcc 4 → 5 problem? (the dtb from the new build works fine wth the old kernel uimage) Oct 29 14:18:48 can I tell the jethro setup to use gcc4 instead to see if that's the problem? Oct 29 14:19:53 fledermaus: you can, GCCVERSION = "4.9%" (or "4.8%" if you prefer) Oct 29 14:20:10 bluelightning: cheers. Oct 29 14:20:32 env var or dropped into the kernel .bb or ... ? Oct 29 14:20:41 fledermaus: no, at the configuration level Oct 29 14:20:46 it almost certainly will fix the issue - there was at least one, possibly two gcc5 fixes required for the kernel Oct 29 14:20:56 zeddii: not sure if you have any pointers ^ Oct 29 14:20:59 bluelightning: interesting. got a link? Oct 29 14:21:04 (to the fixes) Oct 29 14:21:37 fledermaus: not directly, I've just seen the commits mentioning gcc5-related fixes going into linux-yocto during this dev cycle Oct 29 14:22:43 bluelightning: cheers. been tearing what's left of my hair out for a couple of days. Oct 29 14:23:31 * zeddii reads Oct 29 14:23:58 yah. we have a patch that the Wind River guys did to get ARM booting with gcc5. I have it in all the linux-yocto variants. they sent it to Linaro as well. IIRC. Oct 29 14:24:28 http://git.yoctoproject.org/cgit/cgit.cgi/linux-yocto-3.19/commit/?h=standard/base&id=151571a39785218a57c3ae3355cd63694890cc8d Oct 29 14:26:25 zeddii: optimisation funkiness in gcc5? Oct 29 14:26:40 errm, how is the automount from udev-extraconf supposed to be working in fido and later branches? Oct 29 14:27:08 systemd now runs udev in its own mount namespace, so the automounts are going into a namespace that is only visible to the udev process Oct 29 14:27:20 yup. but no analysis to the point of pointing exactly at what has gone wrong ..only empirical results of working kernels Oct 29 14:28:16 zeddii: sometimes I hate software. Oct 29 14:28:30 oh well, if it was easy they wouldn't pay me. Oct 29 14:31:49 abelal: dd if=/dev/zero of=/dev/sdb, then dd if=.hddimg of=/dev/sdb, then sync & ejected. moved device to target machine & booted from usb Oct 29 14:40:32 <_4urele_> hi everyone! Oct 29 14:41:34 <_4urele_> I'm using systemd and some of my installed services are masked (via "/etc/systemd/system") by default, I don't know why... if someone knows something about that... Oct 29 15:24:10 bluelightning zeddii - that two-liner makes the kernel boot again on mx5. so it's one of the optimisations that's -On dependent that's on by default now in gcc5 Oct 29 15:24:41 the flag differences being: http://paste.lisp.org/display/157700 Oct 29 15:37:25 fledermaus, bluelightning, zeddii: from a quick look I'd say this is relevant - http://lists.infradead.org/pipermail/linux-arm-kernel/2015-September/367610.html Oct 29 15:39:22 ntl: hmm... my symptom though is there's no console output _at all_. Just a brick. Until I pull the plug physically and re-power to get to u-boot. Oct 29 15:40:12 bad code generation can have different manifestations. the GCC bug reports non-booting systems Oct 29 15:40:19 aha Oct 29 15:40:23 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65932 Oct 29 15:40:24 Bug 65932: was not found. Oct 29 15:40:49 yocti: link wfm Oct 29 15:40:50 fledermaus: Error: "link" is not a valid command. Oct 29 15:41:00 oh, it's a bot? Oct 29 15:41:19 it's a bot yeah Oct 29 15:41:44 yocti: ☠̈ Oct 29 15:41:45 fledermaus: Error: "☠̈" is not a valid command. Oct 29 15:50:33 fledermaus: are you going to try with the -fno-ipa-sra kernel patch or just back off to GCC4? Oct 29 15:51:50 the kernel patch sounds promising. Oct 29 15:51:58 was it in one of the links above? Oct 29 15:52:03 If so I missed it. Oct 29 15:52:33 ah diff --git a/arch/arm/Makefile b/arch/arm/Makefile Oct 29 15:52:39 I'll try that next. Oct 29 15:54:56 https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit?id=a077224fd35b2f7fbc93f14cf67074fc792fbac2 Oct 29 16:03:59 ta. compiling ⏳ Oct 29 16:07:26 if that fix works I'll try to get it added to the migration section of the manual Oct 29 16:13:50 ok, that rebuild was entirely too quick. not convinced it isn't using a lo of cached stuff there. Oct 29 16:14:02 * fledermaus scrubs harder Oct 29 16:16:52 bluelightning: can I purge just the kernel package's objects from the sstate cache? Oct 29 16:17:08 fledermaus: bitbake -c cleansstate virtual/kernel Oct 29 16:17:12 thanks Oct 29 16:17:33 fledermaus: but if you *have* to do that then you may not have actually made the appropriate change Oct 29 16:19:40 that's the trouble with SSDs, I can't tell the difference between "failed to rebuild" and "just hella fast" any more :) Oct 29 16:34:23 bluelightning ntl - that no-fips-sra-on-arm patch lets me boot. Oct 29 16:34:56 fledermaus: great thanks Oct 29 16:36:02 additional datum: -fipa-sra is only enabled at -O2, -O3 and -Os Oct 29 16:36:24 so the previous two-liner to force vsprintf.o to -O0 has a mechanism for working too Oct 29 17:05:49 JaMa: did you CC zeddii on that linux-yocto warnings email? Oct 29 17:06:16 I guess I've kind of just done that, but still Oct 29 17:10:59 bluelightning: no I didn't, thanks for doing it Oct 29 17:12:19 bluelightning: there is also some upgrade from Saul, maybe it's related, but I'm not using meta-yocto-bsp Oct 29 17:31:38 fledermaus: FYI: http://www.yoctoproject.org/docs/2.0/ref-manual/ref-manual.html#moving-to-the-yocto-project-2.0-release Oct 29 17:31:48 (just updated, thanks for the info) Oct 29 17:35:22 anyone using qt5 with cross-compiling? that is meta-toolchain-qt5? Oct 29 17:36:09 i don't understand how to use qmake as it returns an error Oct 29 17:43:18 bluelightning: cheers. Oct 29 17:46:11 has anyone looked at building multiple kernels for a single platform within the same build? Oct 29 17:53:04 qknight: we use qt5 cross-compiling, but not with meta-toolchain-qt5 Oct 29 17:57:04 neverpanic: so how are you doing the setup then? Oct 29 17:57:06 * qknight is curious Oct 29 18:00:41 We have a completely different approach to doing an SDK, so our method wouldn't apply here Oct 29 18:01:08 neverpanic: can you just outline it for me? Oct 29 18:02:43 we ship a subdir in the sysroot that has binaries for the architecture of your host machine and use user & mount namespaces to completely replace a user's system Oct 29 18:02:55 so we basically install a nativesdk version of qmake and that's it. Oct 29 18:04:12 and the compiler in that chroot on steroids cross-compiles automatically, so from qmake's PoV it seems to be a local build, because we can run target binaries (either because we can run the target arch or because we use qemu in user mode) Oct 29 18:04:56 neverpanic: sounds like a very cool setup, especially with the qemu stuff Oct 29 18:05:11 neverpanic: so you actually kind of compile 'natively' Oct 29 18:05:18 It's easy for developers. That's a plus. Oct 29 18:05:25 yup Oct 29 18:05:32 neverpanic: did you documentat that somewhere? Oct 29 18:06:11 Nope. We're planning to contribute it back upstream at some point, but haven't gotten around to doing it yet. Oct 29 18:06:40 Plus it only works on Linux, so is not really a replacement if you have devs on OS X Oct 29 18:07:09 neverpanic: OS x ppl could use a VM tough, but true Oct 29 19:11:00 bluelightning: it would be better to put the -fno-ipa-sra workaround in your kernel(s). The -O0 workaround only prevents miscompiling those two files, and one of them is even misspelled in the patch (there is no asprintf.c but there is a kasprintf.c) Oct 29 19:25:07 zeddii: ping Oct 29 19:29:49 What is the best recommendation for forcing -march=pentium-m -mtune=pentium-m when compiling kernel, and all binaries that end up on IMAGE?? Oct 29 19:32:22 We get illegal instructions on applications since SSE3 is probably used. Oct 29 19:33:15 FRAY: This is in your wheelhouse. Are you around? Oct 29 19:33:41 sorry I'm missing hte context.. Oct 29 19:33:54 is the issue that you have a core that does not have specific instructions, but they are being used when compiling the apps? Oct 29 19:35:35 FRAY: I have some binaries on the real hardware (Pentium-M - MMX/SSE/SSE2) that are throwing illegal instructions. It must be because of the -march flag. Oct 29 19:35:40 the 'tune' you select is responsible for telling the system what -userspace- attributes should be set.. (the kernel recipe is responsible for it's own attributes..) Oct 29 19:35:56 so you will have to adjust the kernel "in some way" that it uses the right arguments for the compiler to select the right values.. Oct 29 19:35:57 I can compile the source on the actual native hardware, and it works fine. Oct 29 19:36:50 for userspace, you need to select a proper tune for the hardware. I'd start with tune-i586 (DEFAULTTUNE = "i586") Oct 29 19:36:57 zeddii - not sure if you have been following the discussion above re the upstream fix that ntl is referring to (as I've just got Scott to note in the manual - http://www.yoctoproject.org/docs/2.0/ref-manual/ref-manual.html#moving-to-the-yocto-project-2.0-release ) Oct 29 19:37:21 the i586 tune should not require anything more then basic Pentium support. I do not think it has MMX enabled, but I might be wrong there.. Oct 29 19:37:32 it selectes "-march=i586" Oct 29 19:37:37 FRAY: explain a little more for me please about. Are there multiple files that must be changed to accomplish this? Which ones?? Oct 29 19:37:48 if there is a more appropriate arch/tune value for your CPU then you'll have to write a custom tune for your BSP Oct 29 19:37:54 ok start from the beginning.. Oct 29 19:37:54 ntl: er which patch misspells the filename? Oct 29 19:38:22 The machine .conf file is responsible for including the Oct 29 19:38:30 'tune file' as well as selecteding the default tune.. Oct 29 19:38:31 i.e.: Oct 29 19:38:44 qemumips64.conf Oct 29 19:38:51 require conf/machine/include/tune-mips64.inc Oct 29 19:39:16 that will include the file "tune-mips64.inc". You won't see a 'DEFAULTTUNE' setting as it inherits the one specified in tune-mips64. Oct 29 19:39:30 looksing at qemux86-64 Oct 29 19:39:35 DEFAULTTUNE ?= "core2-64" Oct 29 19:39:35 require conf/machine/include/tune-core2.inc Oct 29 19:39:38 * zeddii hasn't been following at all. too many inputs. Oct 29 19:39:45 at this point, I'm not inclined to change what works. Oct 29 19:39:51 post release, I can do new patches. Oct 29 19:39:56 that specifies the default of 'core2-64' (?= says only if not defined).. and includes the tune-core2.inc file Oct 29 19:40:00 but we were running ahead of the curve when all those changes went in. Oct 29 19:40:10 so it is what it is for now. Oct 29 19:40:10 inside of the tune-*.inc files is where teh system defines all of the values for how the compiler is invoked -for userspace- Oct 29 19:40:17 denix, pong Oct 29 19:40:32 see the README file: meta/conf/machine/include/README for more information.. Oct 29 19:40:45 your BSP needs to select the correct tune file and DEFAULTTUNE value for the board you are building for.. Oct 29 19:40:46 zeddii: have you looked into building multiple kernels in one got? Oct 29 19:40:52 yah. Oct 29 19:40:54 within oe-core/poky the lowest for IA32 is 'i586'.. Oct 29 19:41:01 there was even a bugzilla around for it .. I thought. Oct 29 19:41:08 if that works, you can "move up" to another.. Oct 29 19:41:12 or define your own Oct 29 19:41:20 and we had some hacks at wind to do it as well. Oct 29 19:41:57 Hold, I am reading this and making sure I understand exactly the best approach. Oct 29 19:42:07 zeddii: were there any public patches available? Oct 29 19:47:06 denix. there was a derived patch from an internal approach: it was this series, which had some issues and didn't merget yet:[OE-core] [PATCH v4 1/1] kernel: Add KERNEL_IMAGETYPES to build multi types of kernel at one time Oct 29 19:47:27 it probably wouldn't meet your needs exactly, but it is a start. Oct 29 19:47:38 zeddii: thanks! Oct 29 19:53:37 zeddii: it's a good read! although AFAICS, it just packages the same built kernel multiple times, it doesn't rebuild/reinstall/redeploy more than once, right? Oct 29 19:57:06 it can do that, with a minor tweak. we wanted to use it to build a main, and a kexec kernel, which were different configs. Oct 29 19:57:50 FRAY: First question, the directory (meta-yocto-bsp/conf/machine) has a genericx86.conf file. My file (build/conf/local.conf) has MACHINE ??= genericx86. Oct 29 19:58:55 QUESTION: Should I modify the genericx86.conf variable DEFAULTTUNE ?= pentium-m and make an associated configuration file? Oct 29 19:58:55 The MACHINE = value in bitbake is parsed and causes the load of the conf/machine/.conf file Oct 29 19:59:15 since you do NOT apparently have a machine that is compatible, you should create your own machine (in your own new layer) Oct 29 19:59:25 zeddii: what if I need to build 2 branches, e.g. w/ and w/o a large feature, say RT or something like that? Any pointers? I'm now trying to hack it in, but the more I do, the more of an ugly hack it looks :) Oct 29 19:59:30 you can include the genericx86.conf if the only thing you are changing is the defaulttune or similar.. Oct 29 19:59:34 but you should add your own Oct 29 20:01:40 OK. I have not successfully added a new layer yet since we are only tweaking existing files. This is the right way forward then? Oct 29 20:02:13 in almost all cases, twaking existing files is the wrong development strategy.. if you do that it will be very difficult for you to unwind or upgrade to follow the community.. this is why the layer mechanism was created.. Oct 29 20:02:27 it allows you to put your changes in an area where you have full control. Oct 29 20:02:31 denix. that's when looping and building comes into play. I looked at what angstrom did (not sure if it is still there or not) for that, and had it roughly working. but the installs into the sysroot got crazy pretty fast. Oct 29 20:04:06 zeddii: I have that code in meta-ti as well for building multiple configs, but from the same source code. I'm actually extending it for switching source branches/commits... Oct 29 20:04:33 aha. Oct 29 20:04:39 so you are deep into the chaos :) Oct 29 20:05:03 I admit to doing this outside of bitbake in general. multiple runs within a build. Oct 29 20:18:07 bluelightning, zeddii: back, sorry for the latency, had to run an errand. this patch: Oct 29 20:18:09 http://git.yoctoproject.org/cgit/cgit.cgi/linux-yocto-3.19/commit/?h=standard/base&id=151571a39785218a57c3ae3355cd63694890cc8d Oct 29 20:18:53 ah I see.. hmm Oct 29 20:20:22 but really, I think you want the upstream workaround as soon as your process allows Oct 29 20:20:55 it's been filtering out to the -stable trees, but maybe not as far back as 3.19 Oct 29 20:21:13 * zeddii doesn't really care about 3.19 .. and tried to remove it completely for 2.0 Oct 29 20:21:24 I'm sitting on several thousand commits until after 2.0 releases. Oct 29 20:23:04 lol Oct 29 20:23:42 * Crofton hopes zeddii is joking Oct 29 20:24:13 based on arch changes, board changes, and bug fixes in new linux.. I'm betting he's not Oct 29 20:24:34 I'm talking about the -stable updates + BSP changes in genearl. Oct 29 20:24:43 so yah. they number in the thousands. Oct 29 20:24:47 zeddii, https://lists.fosdem.org/pipermail/fosdem/2015-October/002197.html Oct 29 20:25:21 Crofton: oooh! Oct 29 20:25:24 * zeddii tries to make a note Oct 29 20:30:22 * fray grabs aleve.. maybe it'll help eliminate the screaming headache that is prelink Oct 29 20:30:39 (FYI, I have IA32 and MIPS (n32/n64) working now -- testing o32 right now) Oct 29 20:39:58 fray, if you need, I can probably test on ppc32 with my sbc8641d ; last time I tried (1-2wks ago?) /sbin/init would fail Oct 29 20:40:08 (for both systemdisease and sysV) Oct 29 20:41:59 paulg w/ or w/o prelink enabled (if you don't know that means with) Oct 29 20:42:07 and I do believe PPC is on the still f'ed list Oct 29 20:42:49 (I have a coworker helping me with ARM currently.. PPC is next on my list after I complete the MIPS testing..) Oct 29 20:42:51 I tested w/ the defaults which I assume is prelink enabled. Oct 29 20:43:00 yup.. it likely broke things Oct 29 20:43:03 have not tested pl disabled. Oct 29 20:44:40 * fray is VERY glad MIPS64 emulation works.. but man it's slow Oct 29 20:58:26 Hi guys, another prob, I want to download kernel image from kernel.org (git) and also drivers (another git) and put the drivers inside the kernel for compiling, what's the way to do this Oct 29 20:58:31 (best) Oct 30 01:44:25 hello, how do I edit my recipe so the app installs to /usr/lib64 rather than /usr/lib ? Been getting Files//directories were installed but not shipped error Oct 30 01:52:21 behanw: i don't suppose you could poke someone about this? Oct 30 01:52:22 https://llvm.org/bugs/show_bug.cgi?id=8220 Oct 30 01:52:24 Bug 8220: normal, Medium, 2.1, brian.avery, NEW , We need to enable an exclusion list in toasterconf.json to drop layers from layers.openembedded.org Oct 30 01:52:48 * nerdboy slaps yocti with a large trout Oct 30 01:53:47 makes it really really hard to pass correct flags to both gcc and clang Oct 30 01:54:35 really need a way to override flags from llvm-config Oct 30 01:58:28 plus it's plainly wrong to (blindly) hard-code another compiler's flags in the config tool when they're not compatible Oct 30 02:03:27 seebs: http://www.openembedded.org/wiki/Adding_a_secondary_toolchain Oct 30 02:03:44 this was written with multiple toolchains in mind, even multiple that used different configuration commands Oct 30 02:30:23 'er.. not sure why it did "seebs" that was for nerdboy Oct 30 02:42:59 Hm... anyone dealt with ninja output before? Kind of struggling to build a good debug package from it's output. It outputs a bunch of object files, and dwarf object files in a separate directory. Haven't managed to get this setup to process core dumps properly as of yet. **** ENDING LOGGING AT Fri Oct 30 02:59:59 2015