**** BEGIN LOGGING AT Tue Feb 16 03:00:33 2021 Feb 16 03:50:44 m4t: not clear what's going on with builds lately... seems a lot is broken of late... Feb 16 03:59:48 mangix: how does PR #14760 look? Feb 16 04:08:26 Is there a way to disable CI processing on a PR? Feb 16 04:08:45 It's a waste of resources because it keeps running out of space. Feb 16 04:14:35 5-5.5 hours of build time to error with: 2021-02-15T19:50:06.3800728Z ##[warning]You are running out of disk space. The runner will stop working when the machine runs out of disk space. Free space left: 89 MB Feb 16 04:14:45 and it really doesn't need to run ;/ Feb 16 04:15:17 is this for a package? Feb 16 04:15:27 Yes Feb 16 04:15:31 if so you should ping aparcar[m] Feb 16 04:16:40 Thanks lipnitsk :) ping apacar[m] Feb 16 04:17:07 err aparcar[m] Feb 16 04:21:35 Grommish: pong Feb 16 04:21:48 Is there a way to disable CI for a package PR? Feb 16 04:22:09 Grommish: i hope not Feb 16 04:22:16 Its a draft and chew for 5-5.5 hours before failing beacuse it runs out of space.. Doesn't need to actually run for a draft Feb 16 04:22:22 2021-02-15T19:50:06.3800728Z ##[warning]You are running out of disk space. The runner will stop working when the machine runs out of disk space. Free space left: 89 MB Feb 16 04:22:33 and.. it runs 9 times Feb 16 04:22:38 Wow what are you building? Feb 16 04:22:44 Its the rust-lang package Feb 16 04:22:52 Oh sweet Feb 16 04:23:11 It's been working for a while, but its why I had the PKG_HASH as skip, but people complained hehe Feb 16 04:23:26 CI won't touch the hash mismatch Feb 16 04:23:42 but I'd rather not do that Feb 16 04:24:08 if not, I'm not stressing, but I find its a huge waste of resources Feb 16 04:25:24 https://github.com/openwrt/packages/runs/1904349643 for an example Feb 16 04:26:12 philipp64: i guess you didn't see my comment linking the PR https://github.com/openwrt/packages/pull/14761 Feb 16 04:27:15 aparcar[m] Interestingly enough, the ones that "pass' are ones rust won't build for Feb 16 04:27:55 It totally fails for those packages :D but gets the check Feb 16 04:28:29 m4t: no, just saw it. what the hell... why is it still failing? Feb 16 04:29:04 lol Feb 16 04:29:10 it wasn't me :) Feb 16 04:29:23 i've seen the perl failures on x86 before Feb 16 04:29:47 * pkg_run_script: Internal error: perlbase-i18n has a NULL tmp_unpack_dir. Feb 16 04:31:41 i dunno if this is relevant https://patchwork.openembedded.org/patch/81415/ Feb 16 04:33:17 thanks mangix Feb 16 04:34:31 i wonder if there is a middle ground between relative & absolute that makes both local out-of-tree and CI builds happy Feb 16 04:37:56 aparcar[m]: What do you think of the above patchwork? Feb 16 04:38:35 Grommish, aparcar[m]: It's probably the fact that the container/VM runs out of space that causes it to take so long Feb 16 04:39:31 Its a 2h20m build for the rust toolchain on my i7-7500u w/28Gb ram.. It's just big for the first compile Feb 16 04:39:49 and it would die at 6 hours anyway if Openwrt's CI is like the personal ones Feb 16 04:40:10 CI not beefy enough probably, too little RAM/CPU? Feb 16 04:40:47 Dunno.. Feb 16 04:43:12 Grommish: CI jobs can be manually canceled by package maintainers Feb 16 04:45:12 m4t: the perl packages are defined...weird. I initially added --force-removal-of-dependent-packages to fix CI problems. It's more of a workaround though. Feb 16 04:45:35 ah Feb 16 04:45:50 i wonder if that patch should be picked up, then Feb 16 04:45:59 the migration to apk should solve this Feb 16 04:46:00 * m4t has no idea where/when/if opkg diverged from upstream Feb 16 04:48:36 aparcar[m]: what's the story there actually? Anything missing? Feb 16 04:50:52 mangix: im wondering what's up with libubox in snapshots Feb 16 04:50:56 Uhttpd doesn't start Feb 16 04:51:57 i was asking about apk. This ABI stuff is voodoo Feb 16 04:52:42 sorry trying to fix my homes network Feb 16 04:52:47 I'll read the log once done Feb 16 04:54:07 sounds like btrfs snapshot magic is needed :) Feb 16 04:56:15 aparcar[m]: nbd was working on the ubox stuff earlier Feb 16 04:56:22 but I'm not sure if he got it fixed or not Feb 16 04:57:50 I know I updated from main branch earlier and it wasn't an issue anymore for the uci/ubox error I was having.. then it was spreading to dmesg and something else I saw before I stepped away Feb 16 05:53:41 mangix: want to help with the APK project? Feb 16 06:00:38 what do you mean? Feb 16 06:46:47 Hauke: what packages are you still getting the errors with? Feb 16 06:47:04 i tried building images with fdisk=y and it worked just fine for me Feb 16 06:59:23 nbd: I'm trying to profile my SW offload changes on mt7621 and not getting much difference in measurement. in fact CPU0 seems swamped with IRQs from the NIC. Do you expect any benefit while using sw offload or should I just try to work out the HW offload piece? Testing with VLAN+PPPoE on eth0 Feb 16 07:02:24 as a corollary, on 19.07 with SW offload I'm also seeing ksoftirqd taking up CPU, but it seems distributed across the 4 cores: Feb 16 07:02:25 24 2 root RW 0 0% 25% [ksoftirqd/3] Feb 16 07:02:25 14 2 root SW 0 0% 21% [ksoftirqd/1] Feb 16 07:02:25 19 2 root SW 0 0% 20% [ksoftirqd/2] Feb 16 07:02:25 7 2 root RW 0 0% 10% [ksoftirqd/0] Feb 16 07:02:56 vs on master I see: Feb 16 07:02:57 21 2 root RW 0 0% 27% [ksoftirqd/2] Feb 16 07:03:22 so maybe some SMP issue too (btw not quite master, I am testing 5.10) Feb 16 07:03:45 lipnitsk: SMP issue is correct Feb 16 07:05:09 is it an issue on 5.4 too (snapshot)? Feb 16 07:05:35 probably. i have no mt7621 hardware Feb 16 07:05:56 lipnitsk: https://github.com/gnubee-git/GnuBee_Docs/issues/75#issuecomment-390516728 and https://github.com/gnubee-git/GnuBee_Docs/issues/75#issuecomment-390520301 Feb 16 07:07:27 that's a crusty old long thread :) I'll take a look tomorrow Feb 16 07:07:30 thanks Feb 16 07:33:58 build #797 of ramips/rt3883 is complete: Failure [failed tools] Build details are at http://buildbot.openwrt.org/master/images/builders/ramips%2Frt3883/builds/797 blamelist: ?lvaro Fern?ndez Rojas , Paul Spooren , Rafa? Mi?ecki , Felix Fietkau , R. Diez , Feb 16 07:33:58 Hauke Mehrtens Feb 16 07:35:13 build #639 of x86/64 is complete: Failure [failed pkgbuild] Build details are at http://buildbot.openwrt.org/master/images/builders/x86%2F64/builds/639 blamelist: ?lvaro Fern?ndez Rojas , Paul Spooren , Rafa? Mi?ecki , Felix Fietkau , R. Diez , Hauke Feb 16 07:35:13 Mehrtens Feb 16 07:47:29 i suppose it's about Feb 16 07:47:30 sh: 1: /builder/shared-workdir/build/staging_dir/host/bin/autom4te: not found Feb 16 07:47:39 (buildbot fail) Feb 16 07:49:49 build #789 of lantiq/xway_legacy is complete: Failure [failed tools] Build details are at http://buildbot.openwrt.org/master/images/builders/lantiq%2Fxway_legacy/builds/789 blamelist: Rafa? Mi?ecki , ?lvaro Fern?ndez Rojas , Felix Fietkau , R. Diez , Paul Spooren Feb 16 07:49:49 Feb 16 08:07:47 yep, getting ready http://buildbot.openwrt.org/openwrt-21.02/images/ Feb 16 08:08:26 so I needed to shuffle some buildworkers around Feb 16 08:13:29 ynezz: when is 19.07.7 getting released? Feb 16 08:29:03 * rsalvaterra wakes up and notices the openwrt-21.02 branch… Feb 16 08:29:57 Time to break out the champagne? :) Feb 16 08:53:44 build #0 of bcm53xx/generic is complete: Exception [exception interrupted] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/bcm53xx%2Fgeneric/builds/0 Feb 16 08:53:45 build #0 of mediatek/mt7622 is complete: Exception [exception interrupted] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/mediatek%2Fmt7622/builds/0 Feb 16 08:53:45 build #0 of mediatek/mt7623 is complete: Exception [exception gitcheckout interrupted] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/mediatek%2Fmt7623/builds/0 Feb 16 08:53:45 build #0 of layerscape/armv7 is complete: Exception [exception] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/layerscape%2Farmv7/builds/0 Feb 16 08:53:46 build #0 of octeontx/generic is complete: Exception [exception interrupted] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/octeontx%2Fgeneric/builds/0 Feb 16 09:04:59 Updating feed 'freifunk' from 'https://github.com/freifunk/openwrt-packages.git;openwrt-21.02' ... Feb 16 09:05:07 warning: Could not find remote branch openwrt-21.02 to clone. Feb 16 09:06:53 I'm wondering who got this idea to include feed which we don't have under full control Feb 16 09:06:56 :) Feb 16 09:06:58 c/elar Feb 16 09:13:16 I think, that we should handle it as packages, git.openwrt.org as source and github should mirror to that Feb 16 09:22:01 hmmm * pkg_hash_fetch_best_installation_candidate: Packages for fdisk found, but incompatible with the architectures configured Feb 16 09:22:01 * opkg_install_cmd: Cannot install package fdisk. Feb 16 09:32:48 ldir: try reverting c92165038217e49075098479704da58a2a3a89bd Feb 16 09:33:42 nbd: your ABI stuff rework broke ipk metadata preparation Feb 16 09:33:59 dependencies appaear to not contain the abi version suffixes anymore Feb 16 09:36:57 jow: thanks, will try when the current build fails ;-) Feb 16 09:37:22 heh https://github.com/freifunk/openwrt-packages/issues/37 Feb 16 09:38:10 build #648 of mediatek/mt7629 is complete: Failure [failed toolchain] Build details are at http://buildbot.openwrt.org/master/images/builders/mediatek%2Fmt7629/builds/648 blamelist: Rafa? Mi?ecki , ?lvaro Fern?ndez Rojas , Felix Fietkau , R. Diez , Paul Spooren Feb 16 09:38:10 Feb 16 09:38:37 ^ /builder/shared-workdir/build/tmp/ccVlOak8.s:729266: Warning: end of file not at end of a line; newline inserted Feb 16 09:38:45 smells like out of disk space Feb 16 09:39:25 /dev/mapper/lvm-builder 394G 32G 363G 9% /builder Feb 16 09:39:31 would be surprising :) Feb 16 09:40:04 that's fsf-dock-15 Feb 16 09:40:15 aaarggh c92165038217e49075098479704da58a2a3a89bd doesn't revert cleanly. Feb 16 09:40:48 ldir: since nbd drover over that entire abi version handling with a tank and reassembled the scrap afterwards, I have no idea how to fix or debug Feb 16 09:41:00 doing the stuff took me about a week when I implemented it Feb 16 09:41:20 I've just removed all the ABI commits Feb 16 09:41:23 lol Feb 16 09:42:13 * jow is slightly annoyed Feb 16 09:42:33 whenever I don't pay attention for two or three weeks and return to it, everything is f'ed up Feb 16 09:43:03 maybe we should just drop binary package repos Feb 16 09:43:20 everyone is just building his own perfectly customized images, right? Feb 16 09:43:23 so why bother Feb 16 09:43:29 :3 Feb 16 09:44:07 jow: there are people not building their own images ? Feb 16 09:44:11 * ldir hides the RPG launcher from jow Feb 16 09:44:28 blogic: apparently not Feb 16 09:46:05 Unfortunately that's a necessity for some people Feb 16 09:46:28 why do we even have wolfssl ?! Feb 16 09:47:18 blogic: wolfssl people added patch Feb 16 09:47:33 but why ?! Feb 16 09:47:42 its worse than openssl and no one uses it Feb 16 09:47:57 let's nuke it. less is more? Feb 16 09:47:58 * stintel hides Feb 16 09:48:04 with openssl you at least have 1bn users effected by bugs that will then get fixed fast Feb 16 09:48:14 stintel: ... from orbit Feb 16 09:48:19 stintel: from orbit? Feb 16 09:48:38 yeah openssl is also more t Feb 16 09:48:38 han half the size Feb 16 09:49:05 openssl is ~1MB Feb 16 09:49:11 you are right, not noticing bugs is a fair tradeoff for safing some flash space Feb 16 09:49:20 *sving Feb 16 09:49:50 hahaha. The question you meant to ask is why doesn't hostapd support mbedtls. Feb 16 09:50:06 no Feb 16 09:50:22 not at all, first thing i do on any build is enable hapd full with openssl Feb 16 09:50:39 If it did I think wolfssl can be nuked Feb 16 09:50:44 we've had those ABI issue with other packages as well Feb 16 09:51:13 hostapd doesn't support mbedtls because nobody cared enough to add support for it Feb 16 09:51:32 AFAIK it took more then year to include wolfssl into hostapd Feb 16 09:52:04 Code was contributed by wolfssl team, no? Feb 16 09:52:15 I think so Feb 16 09:52:24 does it matter? Feb 16 09:52:32 problem with wolfssl is it's horrible ifdef forest Feb 16 09:52:38 Now that there is an openwrt-21.02 branch, can we start to expect builds going into https://downloads.openwrt.org/releases/21.02-SNAPSHOT/ just like with 19.07? Feb 16 09:52:51 and upstream's apparent inability to maintain a proper verisoned abi Feb 16 09:53:50 Person X needs to write mbedtls for hostapd. I don't know who person X is. Feb 16 09:54:43 this is rather naive, mbedtls can have similar set of issues Feb 16 09:55:03 you don't find out unless you start using it Feb 16 09:55:53 ynezz: which commits did you revert exactly? do you have a list of hashes? Feb 16 09:56:06 jow: I've rebased and removed all of them Feb 16 09:56:13 One other advantage of mbedTLS. LTS Feb 16 09:56:18 reverting was tedious Feb 16 09:56:25 ynezz: just the recent ones by nbd or the original ones as well? Feb 16 09:56:31 recent ones Feb 16 09:56:37 k Feb 16 09:57:02 maybe we should revert them in the tree for now until nbd can spin up a fixed series? Feb 16 09:57:10 or at least revert them in the branch Feb 16 09:58:08 depends on nbd, if he has time to look into it and fix it Feb 16 09:58:21 pmelange: yes! we're still on it. Feb 16 09:58:59 we can't start builds anyway, as that freifunk package repository doesn't have openwrt-21.02 branch yet Feb 16 09:59:17 * lynxis gives jow a cookie and a tschunk Feb 16 09:59:28 Cool. Is the "guessed" directory correct? Feb 16 10:00:09 yes. you can find also the url in the first commit of the branch (changing defaults..) Feb 16 10:00:24 There is an issue at freifunk to have the feed removed https://github.com/freifunk/openwrt-packages/issues/37 Feb 16 10:00:36 i will look into it now Feb 16 10:00:48 i just need some more information on what dependencies are still wrong Feb 16 10:00:51 because i can't reproduce it myself Feb 16 10:00:55 blogic: BTW I've nothing against openssl, actually would prefer that, but it's huge, so we would make openwrt probably unusable on bunch of devices Feb 16 10:01:35 so we either have ABI issues which might be fixed OR flash space issues Feb 16 10:01:50 is installable, but broken/non-functional better? Feb 16 10:01:50 how is wolfssl broken? Feb 16 10:03:57 pmelange: it sounds to me it's not yet a final decision. but as I wrote in the comment we can remove it if freifunk want to withdraw the repo. Feb 16 10:03:58 nbd: you have me & my very slow macbook for 30 mins before I have to go to $dayjob. Feb 16 10:05:19 I think it's difficult to know what freifunk wants as it seems like there is very little to no participation from anyone. Feb 16 10:06:13 But since creating of the branch is stopping openwrt-21.02, I can just go and make a branch. The feed can be removed at a later time. Feb 16 10:06:29 pmelange: andi already created it Feb 16 10:06:58 pmelange: just propose the patch :p Feb 16 10:09:21 ldir: which platform are you building for? Feb 16 10:09:45 x86-64 for a pcengines APU2 Feb 16 10:09:51 @ynezz first I thought that I would get some opinions before submitting a patch. Feb 16 10:11:05 ldir: please show me the output of cat build_dir/target-x86*/util-linux-2.36.1/ipkg-*/*/CONTROL/control Feb 16 10:11:26 ynezz: openssl isn't a requirement to run openwrt though. I don't have openssl installed on any of my owrt devices. Feb 16 10:11:34 (or any ssl lib) Feb 16 10:12:34 nbd: https://pastebin.com/X6ytHjZj Feb 16 10:12:45 it wouldn't strike me as particularly odd that to run ssl-enabled you'd need something a bit beefier than baseline 8/64 Feb 16 10:13:29 wpa3 Feb 16 10:13:38 ldir: hm, that looks normal Feb 16 10:13:59 ynezz: same remark though. wpa3 requires more beef Feb 16 10:14:23 but we don't want to differentiate Feb 16 10:14:29 ldir: and the full output of make package/install V=s Feb 16 10:14:36 that's the point: maybe we should? Feb 16 10:15:11 Heresy!!! Feb 16 10:15:15 f00b4r0: wpa3 capable system with openssl fits in 5120 here, so 8 MiB flash should be enough Feb 16 10:15:28 PaulFertser: without luci Feb 16 10:15:41 ynezz: well, yes Feb 16 10:16:17 ynezz: how big LuCI is these days? Feb 16 10:16:20 nbd: will do - just waiting for build to complete just in case something had magic'd itself better. Feb 16 10:17:22 PaulFertser: opkg install it find out Feb 16 10:17:29 s/it/it to/ Feb 16 10:18:45 jow: since you’re here, any plans on updating the buildbots to Debian 10? Feb 16 10:19:36 mangix: it's prepared https://git.openwrt.org/?p=buildbot.git;a=shortlog;h=refs/heads/staging/ynezz/buildbot-2.10.0 Feb 16 10:19:43 but need to find time to deploy that Feb 16 10:20:31 OK. At least 1 package is held back because of C++17 Feb 16 10:20:56 one out of 5000? :p Feb 16 10:22:06 Well, I assume mpd is widely used :) Feb 16 10:22:50 speaking of C++, I found out today that GCC6 is broken with lambdas using this. Feb 16 10:23:00 ynezz, jow: I don't know if this has been considered but if buildbot is upgraded, one might want to re-enable network locks which were implemented in 0edb4934 and disabled in 1e1a326 due to lack of support in 0.9 Feb 16 10:26:48 nbd: sorry felix, my machine is busy melting itself on ' make[3] -C target/linux install' - frustrating 'cos I only reverted & then restored one commit and it needs to build the whole darn thing again. Feb 16 10:28:26 ldir: well, if it's doing target/linux install it means that it already did package/install Feb 16 10:28:41 so it's working Feb 16 10:30:06 it is working - I've changed nothing. WTH!? Feb 16 10:30:33 maybe on the last build you still had some stale stuff in there Feb 16 10:30:38 anyway, sorry have to get on the road to work. If I get a moment I'll make clean and try again. Feb 16 10:30:58 thanks Feb 16 10:30:59 I've done make dirclean and rebuiltd entire toolchain. Feb 16 10:31:48 jow: sorry for the mess i've caused Feb 16 10:32:37 the removal of abiversion from metadata was necessary in order to deal with the wolfssl config dependent ABI version Feb 16 10:40:25 nbd: as long as it ends up in the reverse dependencies it is fine for me. From a cursory look it seems you pushed the info from a generated makefile to per-package files instead Feb 16 10:43:27 for intra-package dependencies it uses the ABIV_$(1) variable now, for inter-package dependencies the .version files Feb 16 10:43:51 had to change the order of a few BuildPackage calls in package makefiles to make it work Feb 16 10:49:56 luci's way smallerthan openssl... Feb 16 10:50:03 sorry, lost in the scrollback again Feb 16 10:50:26 mangix: why were you using gcc6? Feb 16 10:50:53 its used by buildbots Feb 16 10:51:14 User had issues with some changes I did on Debian 9. Non OpenWrt related. Feb 16 10:51:40 Apparently GCC6 has a compiler bug with lambdas Feb 16 10:51:52 I told him to just use clang Feb 16 10:51:55 debian, "stable" again I suppose.... Feb 16 10:52:56 It’s an LTS release Feb 16 10:53:18 Debian is on a 5 year support cycle Feb 16 10:53:38 Not infinite like CentOS :) Feb 16 10:57:31 speaking of build systems, on the 9th I asked in irc if openwrt is still dependent on python2. The wiki page for imagebuilder still has python as a requirement on "Arch / Manjaro" or "Debian / Ubuntu". If anyone can confirm that they can build on either of these systems without python2 being installed, then I would be happy to update the wiki Feb 16 10:57:31 page. Feb 16 11:02:35 19.07 needs Python2, snapshot and 21.02 shouldn't Feb 16 11:02:54 They don’t . Feb 16 11:04:43 nbd: would appreciate feedback on some OSX/BSD "tools/libelf" build questions: https://github.com/openwrt/packages/issues/9927#issuecomment-766299179 Feb 16 11:05:14 nbd: related to a BPF PR: https://github.com/openwrt/openwrt/pull/3855 Feb 16 11:06:56 I will leave the wiki page as it is for now, and once 21.02 is release, then that would probably be the right time to update it. Feb 16 11:07:29 build #1 of ath25/generic is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/ath25%2Fgeneric/builds/1 Feb 16 11:10:29 pfew Feb 16 11:10:49 guidosarducci_: there's a reference to libelf for gcc Feb 16 11:10:55 not sure if it's still needed Feb 16 11:11:06 but that might be one reason for libelf Feb 16 11:11:33 i'd be fine with replacing libelf with elfutils if we can get elfutils to compile on macOS Feb 16 11:11:45 ynezz: nice :) Feb 16 11:12:28 well, I assume there is still long road towards 21.02.1 :] Feb 16 11:15:54 nbd: on Linux it can be completely removed and builds still work, thought maybe same on OSX but I have no way to test Feb 16 11:16:29 linux systems probably typically have some variant of libelf on the host Feb 16 11:17:09 the same can't be said for macOS, since it doesn't use ELF natively Feb 16 11:18:04 Looks like I should try with my macOS VM. Feb 16 11:18:11 nbd: I did some experiments removing libelf-dev too. *But* I didn't rebuild toolchains so there may be a gcc dep. Feb 16 11:18:56 nbd: how supported is macOS? Feb 16 11:19:38 nbd: problem I stumbled across (during BPF work) was we have 3 sources of libbpf: host build from elfutils pkg, system prereq, and the 9-year old BSD tools/libelf Feb 16 11:19:44 mangix: not everything works, but the core stuff does. i use it for almost all of my dev work Feb 16 11:20:10 nbd: which can cause build conflicts. Feb 16 11:20:24 OK. Feb 16 11:20:32 guidosarducci_: right, it's a mess Feb 16 11:20:53 i would like to switch to elfutils, but it might be quite a bit of work to get it building on macOS Feb 16 11:21:38 if i remember correctly, last time i looked at it, it was having lots of portability issues Feb 16 11:21:44 but that was years ago Feb 16 11:22:10 nbd: in my BPF/BTR PR, for now I just isolate tools/libelf to non-Linux, drop the host-build from target pkg elfutils, and instead add a host pkg elfutils. Feb 16 11:22:36 but I wanted to clean thing up a bit more than that... Feb 16 11:23:16 I kind of wish I had a proper VM. That was I can build stuff comfortably. Feb 16 11:23:43 mangix: it's possible to run macOS in KVM Feb 16 11:24:39 Yeah I have one running with QEMU. It’s quite subpar. Small resolution and significant lag. Feb 16 11:25:20 nbd: what I've read show elfutils working on FreeBSD, but request for OSX ports marked "wontfix", but I can't be certain of any of it. Feb 16 11:25:43 mangix: i got it working reasonably well Feb 16 11:25:52 i can give you my libvirt config if you're interested Feb 16 11:30:08 +1 Feb 16 11:30:23 nbd: on Linux, is the OWRT preference for adding host packages or system prerequisites? I needed to also add host libdw-dev, and thought an elfutils host pkg best Feb 16 11:30:47 guidosarducci_: depends on how common the dependency is Feb 16 11:32:45 nbd: atm building dwarves tools needs it, and in past some target pkgs needed host libelf. But I saw there are no make prereq checks for libelf, so we can't always assume they're there... Feb 16 11:33:57 nbd: I'd be OK with adding a system prerequisite, but don't know the change process for it, and it means possibly updated a lot of build bot too no? Feb 16 11:35:00 nbd: No build bot updates needed with a host package... Feb 16 11:38:01 build #2 of imx6/generic is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/imx6%2Fgeneric/builds/2 Feb 16 11:49:06 build #1 of at91/sam9x is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/at91%2Fsam9x/builds/1 Feb 16 11:54:43 build #1 of mpc85xx/p2020 is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/mpc85xx%2Fp2020/builds/1 Feb 16 11:55:11 guidosarducci_: looks like elfutils on macOS is not feasible Feb 16 11:55:18 it uses a lot of non-portable gcc specific nonsense Feb 16 11:55:26 like nested functions, variable length arrays, etc. Feb 16 11:56:10 build #1 of sunxi/cortexa53 is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/sunxi%2Fcortexa53/builds/1 Feb 16 12:02:00 nbd: thanks, what I was afraid of. I'll leave things as in my PR, with tools/libelf restricted to non-Linux build hosts Feb 16 12:02:33 nbd: libvirt config would be great Feb 16 12:05:18 mangix: http://nbd.name/macos.xml - contains a lot of host specific stuff with absolute paths, so maybe just copy relevant settings from it Feb 16 12:05:49 it emulates a vmware graphics card, exposed via vnc Feb 16 12:05:51 so no local display Feb 16 12:12:11 nbd: any other feedback on https://github.com/openwrt/openwrt/pull/3855 would be great if you have a chance Feb 16 12:12:53 nbd: BTW would it be normal for some kernel build options to be unavailable on OSX host because of this? Feb 16 12:18:14 yes Feb 16 12:23:41 nbd: how do we include a Kconfig dependency on host arch then? Can't remember seeing that tbh Feb 16 12:26:00 at the moment we don't have that in kconfig, just as command line override Feb 16 12:26:19 i have an idea on how to do it though Feb 16 12:26:52 write a file with the non-linux config overrides and use kconfig.pl to merge it in the place that generates the kernel config Feb 16 12:28:08 f00b4r0: makes sense, lets first do the upgrade, then finetune it :p Feb 16 12:29:18 ynezz: sure :) Feb 16 12:32:06 nbd: I suppose documentation works too :-) "Only supported on Linux". Otherwise the complexity is spiraling out of control. Feb 16 12:34:04 nbd: thanks for the help, going away for a while Feb 16 12:38:17 build #1 of bcm47xx/generic is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/bcm47xx%2Fgeneric/builds/1 Feb 16 12:48:04 build #1 of bcm63xx/smp is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/bcm63xx%2Fsmp/builds/1 Feb 16 12:50:12 nbd: i finally installed iperf2 Feb 16 12:50:17 with iperf3 i was getting 860 - 870 Mb/s, with iperf2 single stream I get them same (870 Mb/s) Feb 16 12:50:20 i tried iperf -P 4 and I get 870 - 880 Mb/s Feb 16 12:50:36 nbd: is CPU not a real bottleneck? Feb 16 12:51:18 my testing decices are i3-8130U (WAN) and AMD Ryzen Mobile 2500U (LAN) Feb 16 12:56:02 aparcar: https://github.com/openwrt/openwrt/commit/a015d9170851509a58e063dd82a054e98fb48700 has an abbreviated name Feb 16 12:57:45 not sure whether we care, but consider revert/reapply with proper name Feb 16 13:01:52 with -P 4 in top I still see 25% [ksoftirqd/0] - does it mean only 1 CPU gets user even with multiple TCP streams? Feb 16 13:02:01 build #1 of ipq806x/generic is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/ipq806x%2Fgeneric/builds/1 Feb 16 13:10:52 rmilecki: I've now seen Adrian removing ipq807x target from 21.02, perhaps bcm4908 should be removed as well? Feb 16 13:11:35 blogic: see above, what about realtek target, is anyone going to bother with it in 21.02? I doubt that, perhaps it should be removed as well Feb 16 13:12:06 at least it was discused on some past meeting and it was decided, that realtek shouldn't be included in 21.02 Feb 16 13:12:17 ynezz: blogic at least said that he planned to remove realtek Feb 16 13:12:30 (in general, not recently) Feb 16 13:12:55 it's better to double check, then to be sorry :) Feb 16 13:13:22 of course, I also thought whether I should ping him again or not ;-) Feb 16 13:13:32 anyone against gcc6+ for master https://patchwork.ozlabs.org/project/openwrt/patch/20191112200129.19396-1-ynezz@true.cz/ ? Feb 16 13:15:43 ok, nobody :) so I'm going to rebase that patch and send it out for final review Feb 16 13:19:53 ynezz: i'm againt removing bcm4908, it's source-only, doesn't bother anyone, i'm planning to use it Feb 16 13:23:10 ynezz: but muh centos 7 VM Feb 16 13:23:43 ynezz: speaking of which, could we now bump master to gcc 10? I've been using if for months on x86-64, mvebu, mips (24/74kc) and mipsel. :) Feb 16 13:25:04 build #1 of arc770/generic is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/arc770%2Fgeneric/builds/1 Feb 16 13:28:34 build #1 of tegra/generic is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/tegra%2Fgeneric/builds/1 Feb 16 13:34:49 rsalvaterra: openwrt.org/submitting-patches :p Feb 16 13:36:32 rmilecki: like you're going to keep it updated as source-only in 21.02? Feb 16 13:37:06 any additional target/kernel patch is headache during kernel rebasing/bumps Feb 16 13:37:22 ynezz: hey, I know that. :P Feb 16 13:38:09 if it's in tree, it means, that it's maintained and it screams "we accept patches for that" Feb 16 13:38:39 build #1 of ath79/nand is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/ath79%2Fnand/builds/1 Feb 16 13:44:15 well, maybe we simply should skip source-only targets during version bumps Feb 16 13:44:45 hm, but then what's the point to have it in the tree if it won't likely compile Feb 16 13:44:49 or run Feb 16 13:52:47 mangix: I plan to do some minor tests and tag it today Feb 16 13:53:10 mangix: zorun: I plan to tag 19.07.7 today in the evening, ayn objections? Feb 16 13:56:17 mangix: you wanted something merged in package feed, should I merge it or are you around in ~5 hours? Feb 16 14:10:14 Hauke: yes please. https://github.com/openwrt/packages/pull/14647 Feb 16 14:10:37 I take it you have push access. Feb 16 14:19:18 ynezz: i'm going to personally build & daily use 21.02 and bcm4908 Feb 16 14:19:51 ynezz: there are two missging things for daily usage now: sysupgrade & PCIe Feb 16 14:20:00 ynezz: i don't use PCIe (no wifi) Feb 16 14:20:35 ynezz: don't worry about kernel bumps, my targets are the cleanest in the whole tree Feb 16 14:20:39 build #1 of rockchip/armv8 is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/rockchip%2Farmv8/builds/1 Feb 16 14:20:58 ynezz: i may consider dropping "source-only" if I backport sysupgrade support Feb 16 14:23:16 how do one measure that cleanest feature, asking for a friend Feb 16 14:23:58 ynezz: ls target/linux/bcm4908/patches-5.4/ Feb 16 14:24:04 ynezz: 95% of patches are backports Feb 16 14:24:09 I see plenty patches Feb 16 14:24:36 + every patch is cleanly described (source kernel version) Feb 16 14:25:15 it doesn't matter during kernel bumps Feb 16 14:25:29 you need to rebase onto the next 5.4 stable release Feb 16 14:25:49 the more you need to rebase the more issues you might hit and need to solve Feb 16 14:26:06 build #1 of pistachio/generic is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/pistachio%2Fgeneric/builds/1 Feb 16 14:26:35 let's hope you can make it out of source-only state before 21.02.1 Feb 16 14:26:53 thanks, i hope for that, i work on it daily now Feb 16 14:27:02 good luck :) Feb 16 14:27:29 ynezz: you'd always have the option to nuke it later, though Feb 16 14:27:31 * f00b4r0 hides Feb 16 14:28:20 does anyone remember how we used to set CPUs mask to handle interface RX irqs? Feb 16 14:28:24 there was sysfs entry for that Feb 16 14:28:28 *is Feb 16 14:29:05 build #1 of malta/be is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/malta%2Fbe/builds/1 Feb 16 14:30:35 build #1 of bcm27xx/bcm2711 is complete: Success [build successful] Build details are at http://buildbot.openwrt.org/openwrt-21.02/images/builders/bcm27xx%2Fbcm2711/builds/1 Feb 16 14:31:03 adrianschmutzler: BTW what is so special about ipq807x? Why simply not remove all of the source-only targets (minus the bcm4908) ? Feb 16 14:32:35 ynezz: I just removed that one because blogic specifically stated it and it's obviously incomplete Feb 16 14:32:54 so one thing less to care with an easy decision Feb 16 14:33:12 adrianschmutzler: do you plan to address the rest as well? Feb 16 14:33:33 nbd: with echo f > /sys/class/net/eth0/queues/rx-0/rps_cpus I get 880 Mb/s and 20% [ksoftirqd/1] Feb 16 14:33:41 what the hell is limiting that NAT speed :( Feb 16 14:33:44 I considered removing realtek as well, but since that one looks so ready, I would prefer to leave that to blogic, and not have my name on the removal Feb 16 14:34:15 rmilecki can surely decide himself on bcm4908 ... Feb 16 14:34:24 do we have other SOURCEONLY targets left? Feb 16 14:34:37 yeah, but it would be quite strange to make it out of source-only in 21.02.3 :p Feb 16 14:35:14 I don't really have a strong opinion on that Feb 16 14:35:45