**** BEGIN LOGGING AT Thu May 09 02:59:56 2019 May 09 07:01:13 Pavel40: it may be too complex to discuss/resolve it here May 09 07:01:37 Pavel40: i suggest describing that properly & sending an e-mail of creating a bug report May 09 08:44:32 Can a clever person remind me of the feature/reason that we have our own dhcpv6 server over using dnsmasq's dhcpv6/ra service? May 09 08:47:43 :) May 09 08:52:05 ldir:as not all features in odhcpd are supported by dnsmasq May 09 08:53:36 dedeckeh: Greetings (again) :-) Can you remind me the specifics ? May 09 08:54:18 and thanks for the netifd fix - it's still fixed :-) May 09 08:54:28 one of the features is force reconfigure May 09 08:55:19 immediate invalidation of a prefix when it is revoked on the wan May 09 08:56:31 ah ok, dnsmasq I think marks it as deprecated and assumes the clients timeout and use new prefixes for new connections May 09 08:56:40 ldir: I think broadly speaking, there needs to be a way to directly control the prefix pool inside dnsmasq May 09 08:56:50 through some kind of control socket or wahtever May 09 08:57:16 right now dnsmasq is a black box in this regard May 09 08:57:58 Does anyone know the magic blogic dns/dnsmasq replacement service is coming along? May 09 08:57:58 the other common complaint was code smell, but given that dnsmasq is such a vital piece in various cloud deployments nowadays, I'd tend to trust its stability & security enough May 09 08:59:30 indeed code readability/structure is err 'interesting' - but in terms of security it has survived fairly well....and better thanks to a round of google fuzzing it to death May 09 09:00:26 there's rfc6644/reconfigure support, though I'm not sure if there are clients apart from odhcp6c that support it May 09 09:03:03 if you want to support the IPv6 ready logo force reconfigure is required May 09 09:04:41 jow: what is the purpose of iconv-stub? May 09 09:07:09 * ldir decides to look at going back to odhcpv6d6cc6d6vd or whatever it's called for lan dhcpv6/ra stuff instead of using dnsmasq-full May 09 09:11:57 mangix: the purpose of the stub was to provide an external iconv() implementaiton for programs that need it May 09 09:12:29 the stub is only a few KB in size, has no dynamic library depends and supports conversion from/to utf-8 and various 8 bit charsets (cp1252, iso8859-1, ...) May 09 09:13:00 the "normal" solution for iconv() would be GNU's libiconv which is very heavy May 09 09:14:35 or rather used to be, seems its in the 10-11KB ballpark nowadays May 09 09:24:06 * ldir looks into the abyss of /etc/init.d/dnsmasq May 09 09:38:12 * ldir so far, so painless May 09 09:44:52 just because you got used to the pain doesn't mean it's not there anymore ldir ;) May 09 09:45:54 Borromini: you mean I have to stop taking cocaine *AS WELL*???? May 09 09:49:47 hehe May 09 09:53:56 Today I tried configuring an OpenWRT router via LUCI... from an 1996 Macintosh Performa 5260 with Netscape Navigator Gold 3.0.1 May 09 09:54:11 it crashed. May 09 09:56:11 have you tried contacting Apple support? ;-) May 09 09:56:47 I'd rather not :^) May 09 09:57:09 fun fact: router's RAM size > computer's RAM size May 09 09:58:44 router crashed? May 09 09:59:05 no, macintosh crashed :P May 09 09:59:24 apparently the issue is "too much javascript" (disabling it makes stuff not crash) **** BEGIN LOGGING AT Thu May 09 12:02:30 2019 May 09 12:56:18 hmm, does odhcpd not maintain state across process restarts? May 09 12:59:17 ldir:no it does not atm May 09 13:03:59 hmm, so I restart odhcpd and lose all my hostname to ipv6 address mappings May 09 13:14:46 is there a design specific reason/requirement why this is the case? May 09 13:16:04 because I'd personally describe the current behaviour as unfortunate/undesirable. May 09 13:38:28 ldir: probably "nobody bothered to implement it yet" May 09 13:39:51 KanjiMonster: I suspect so - and that leads to 'if the itch annoys you, you scratch it" :-) May 09 13:42:26 ok, ideally odhcpd should die/need restart that often, so maybe I'm being fussy. May 09 13:43:10 and probably there's a nice way to remove a lease? May 09 13:48:51 ldir: And that leads to 'why the fuck is all the skin coming off my arm'.. :) May 09 13:49:10 lol May 09 13:52:21 and I wonder about CERT VU#598349 mitigations as well. May 09 13:53:35 * ldir glares at his samsung tv trying to claim 'localhost' May 09 14:01:39 * ldir goes back to dnsmasq May 09 14:57:19 tmn505: BTW where do you buy espressobins in EU? May 09 15:02:09 it's on amazon.com, they ship to EU. Globalscale also but their shipping prices are horrendous. May 09 15:03:21 search for SBUD102 May 09 15:58:57 jow: i ask since iconv is not compiled in uClibc-ng and fails even with programs that include the stub as a DEPEND May 09 16:01:59 you still doing synopsis work for them I see... May 09 16:02:06 hope you're getting paid May 09 16:53:42 Yesterday I did a build of OpenWRT for my UAP-AC-PRO with GCC 9.1. Surprisingly, it mostly worked fine. There is one warning (treated as error) in ubox and busybox ntpd miscompiles with link-time optimization though. May 09 17:51:54 mamarley: do you have some patches for gcc 9.1? May 09 17:54:20 Hauke: Not at the moment, but I should be able to put something together and submit it tomorrow or Saturday. May 09 17:54:48 mangix: thanks May 09 17:56:00 Hauke: Is there any automated way of refreshing patches, or do I have to do that manually? (Right now, I made all the applicable ones apply, but there are still offsets and fuzz.) May 09 17:56:55 mamarley: make package/... refresh May 09 17:57:00 should do that afaik. May 09 17:57:07 Thanks! May 09 17:57:14 package being the path to what you are updating :) May 09 17:57:15 yw May 09 19:02:29 i compiled the latest stable openwrt for the "Mediatek MT7628AN evaluation board" to run it on the a Hi-Link HLK-7688A module. But i get a kernel panic because it cant mount root because spi doens't work: May 09 19:02:31 [ 0.693994] spi-mt7621 10000b00.spi: could not find pctldev for node /pinctrl/spi, deferring probe May 09 19:02:36 has someone an idea? May 09 19:03:00 the same error for uartlite and uart1 May 09 19:36:26 Mirici: well, Hi-Link HLK-7688A != Mediatek MT7628AN evaluation board - you'll have to bite the bullet and port OpenWrt to that particular device properly, there is nothing generic in embedded development May 09 20:02:18 @pkgadd mediatek states that the MT7688 and MT7628 are the same SoC except wifi May 09 20:05:52 Mirici: that has nothign to do with the supporting peripehrals like the spi flash however... May 09 20:06:52 i know karlp, but in my case it doesn't even try to talk to the flash May 09 20:07:03 ... because spi doenst't work May 09 20:11:30 spi and uart are SoC internal peripehrals, so it depents only on the SoC right? May 09 20:18:59 why would it try on a completely different devcie? May 09 20:33:40 pkgadd, sure im going to modify the device-tree file to make sure i have the right peripehrals at the right ports May 09 20:35:44 but the soc should run with any MT76x8 image as long flash is compatible and spi uses the same pins May 09 20:44:32 no, it shouldn't May 09 20:44:47 there is no magic detection going on May 09 20:45:05 the device tree file is either correct, or the kernel fails May 09 20:51:42 . May 09 20:52:26 .. May 09 20:52:47 ... May 09 20:53:36 .{4} May 09 20:58:25 my dsl line has been flaky for 3 days, i was testing if i am online :-) May 09 20:58:37 can't wait for the technician to come tomorrow May 09 20:58:52 i am getting disconnect every 10-20 minutes May 09 20:58:59 I try to run perf on my mt7621 router but I get irq 7: nobody cared see log: https://paste.debian.net/1082416/ May 09 20:59:19 you're not online - this is a figment of your .... brrrrrrrrrrrrrrrrrrrrrrrrr May 09 21:00:25 blogic: I feel your pain May 09 21:00:53 yikees i just had ldir's ghsost appear in my dreams May 09 21:00:53 scary shit May 09 21:01:11 are you on ADSL (as in, direct (ancient) cable all the way to the post office or VDSL2, as in just a couple hundreds of metres to the outdoor DSLAM)? I had that kind of fun with ADSL for years (it only got better when they replaced parts of the cable that was getting flooded when it rained); now with VDSL2 the situation is much better May 09 21:01:52 Rene__: well 5.1, shouldn't you complain upstream? :) May 09 21:02:14 pkgadd: adsl May 09 21:02:26 i live in germany, digital middle age May 09 21:02:43 blogic: I still only have VDSL2, it's not much better. May 09 21:02:44 i get 9mbit downlink on a good day May 09 21:02:57 Family friend on a farm barely knows how to use a phone, has fibre outside his house.. May 09 21:03:02 Monkeh: i signed up for fiber, but there is a 6 month lead time May 09 21:03:33 ouch, yep, been there until summer 2017 - when DTAG finally decided to deploy outdoor DSLAMs/ VDSL on a large(r) scale May 09 21:05:00 pk there is fiber here now so dtag abandond us May 09 21:05:24 ynezz: Maybe but I am not sure, I have seen this also with a gpio interrupt from the i2c gpio expander but with a different IRQ number. May 09 21:07:10 blogic: Remember to thank the guy nicely for diving into pits and/or climbing up poles to fix the last guy's shitty work. :) May 09 21:07:11 pkgadd: in this case flash is the only externel part on the board that is mandatory to boot up, therefore every device tree for a MT76x8 (with a maching flash configuration, most of the boards uses exacly the same, exept flash size) should work May 09 21:08:20 understandable to some extent, but yes, bad times. I had to 'voluntarily' revert to 1 MBit/s ADSL, as 6 MBit/s simply weren't stable enough, for many years, before I finally got up to first 6 MBit/s and later ~11 MBit/s (and now 100/40 MBit/s) May 09 21:08:58 Mirici: for an initramfs image booted via tftp, yes - when flashed, no May 09 21:09:59 ? May 09 21:11:59 thare are not so much (common) possibilities of flash configuration, in witch point is should fail? May 09 21:15:59 Rene__: this is happening on the clean upstream kernel or you've some mods on top of it? what is in `cat /proc/interrupts` ? May 09 21:17:03 Mirici: but why do it wrong if you can do it right ? May 09 21:18:21 blogic thats what i'm trying to do May 09 21:20:28 unfortunately i could figure out where the error [ 0.693994] spi-mt7621 10000b00.spi: could not find pctldev for node /pinctrl/spi comes from May 09 21:21:41 ynezz: it is clean on the mips code. /proc/interrupt output: https://paste.debian.net/1082418/ May 09 21:22:37 ynezz: https://github.com/vDorst/linux-1/tree/net-next-phylink This is the current code, working on the phylink, mtk_eth_soc and mt7530 driver. May 09 21:39:15 Rene__: well, that's quite huge patchset :) I would simply start with clean net-next and incrementaly add more on top of it to isolate the issue May 09 21:42:28 hmm, annoyed. We could be smarter with odhcpd's poking of dnsmasq. by using --hostsdir option instead of --addn-hosts dnsmasq will automagically read any new or updated hosts file in pointed to directory. Then we only need to sighup dnsmasq on deletion of a lease/host. May 09 21:42:42 It looks like most .so files in OpenWrt have the symbols stripped out. Is that correct? May 09 21:42:49 ynezz: I do, I also going to test current openwrt image see if it is the same. May 09 21:45:45 odhcpd's dhcpv6_ia_write_statefile() could be called with a 'I delete stuff' flag and then if set only call vfork/execve the poke dnsmasq script May 09 21:46:36 * ldir goes to bed as is tired and typing semi-gibberish May 09 21:48:42 Rene__: have you tried to disable that i2c expander? May 09 21:53:07 ynezz: I removed the DTS part that it can be a interrupt-controller see dts https://github.com/vDorst/linux-1/blob/net-next-phylink/drivers/staging/mt7621-dts/UBNT-ERX-SFP.dts#L87 May 09 21:57:17 ynezz: SNAPSHOT, r9851+1-d1739c6c9a with 4.14.111 doesn't seems to have the perf issue. But it is missing IRQ 7 in the list /proc/interrupts May 09 21:58:43 so you've almost fixed it :) May 09 22:00:51 Almost..., thanks for helping, I am gone, bedtime bye May 09 22:12:56 Trying again... I have a package where I need to build the .so files without the symbols being stripped. I tried to set "CONFIG_NO_STRIP" but that did not do anything. Is there another magic make variable for this? May 09 22:14:43 brucethompson: uhm, the upstream package itself could be supplying gcc -s or similar, possibly May 09 22:15:17 I was under the impression that the current build system strips symbols by default May 09 22:15:22 Not true? May 09 22:15:28 it does May 09 22:15:40 Right. May 09 22:15:53 when i say upstream i mean if you're building libfoo, libfoo might do its own strip May 09 22:16:24 You mean the Makefile for libfoo? May 09 22:16:34 yes, not the openwrt makefile May 09 22:16:57 I'm searching through rules.mk it looks like it does the stripping May 09 22:17:03 I think May 09 22:17:14 brucethompson: what's your use case? May 09 22:17:33 for most people, having the unstripped file locally still allows debugging, what do you need unstripped on the target for? May 09 22:17:49 I am building a package where the .so files in the package need to have the symbols not stripped May 09 22:18:01 there's even a menuconfig option for stripping, you shouldn't have to be looking into makefiles for this. May 09 22:18:06 It's a python package. May 09 22:18:07 yes, but what's the usecase of that. May 09 22:18:41 you should be filing bugs upstream if they are relying on symbols that aren't public. May 09 22:19:05 For some reason, the package is actually checking for specific symbols in the .so file May 09 22:19:11 Not sure why May 09 22:19:40 It checks for the symbols when the package initializes May 09 22:20:22 On other Linux targets the .so file does not have the symbols stripped May 09 22:20:40 i'm not sure if stirpping is your problem.... May 09 22:20:44 On OpenWrt, the symbols are stripped so the package fails initialization May 09 22:21:00 I think it is. May 09 22:22:46 which package is complaining about missing symbols? and which library is it looking for those symbols in? May 09 22:23:09 stripping unneccesary symbols is default May 09 22:23:53 could be an api version mismatch, or the library is being compiled without a feature your python package requires May 09 22:24:08 It's a python package called "magic workhole" May 09 22:24:12 wormhole May 09 22:24:18 I ported it to OpenWrt May 09 22:24:29 I just checked the error again. May 09 22:24:56 It is looking for a specific symbol in libsodium when it starts May 09 22:25:29 Its a copy of libsodium specific to the wormhole package May 09 22:25:47 So all I need is to not strip the symbols from that specific .so file May 09 22:26:28 I just don't know what makefile variable I need to set to keep from stripping symbols in a .so file May 09 22:27:06 did you see where I said there's menuconfig options on stipping? May 09 22:27:23 Oh. I did not. Sorry. May 09 22:27:41 Now I see it May 09 22:28:22 But, I always need the symbols present whenever this package is built. May 09 22:28:32 Only this particular package, May 09 22:28:53 So shouldn't I make a change to the Makefile for this? May 09 22:31:37 I see this option in menuconfig: May 09 22:32:11 "Compile packages with debugging info" May 09 22:32:17 Is that what you mean? May 09 22:36:25 i see SKIP_STRIP=y in busybox/Makefile May 09 22:37:21 Great!!. Let me check that May 09 22:37:30 false alarm May 09 22:37:39 ?? May 09 22:37:40 looks like it's used for busybox's build system May 09 22:37:48 OK May 09 22:38:41 brucethompson: try something like https://github.com/openwrt/openwrt/blob/master/package/firmware/layerscape/ppfe-firmware/Makefile#L21 May 09 22:39:36 OK. Great!!! I will try this. May 09 22:39:44 Gotta eat dinner right now. May 09 22:39:59 Thank you VERY MUCH. May 09 23:05:40 Ugh. That did not work. May 09 23:22:57 Oops. That DID work!! May 10 00:09:47 (IMO, yuou should still absolutely be filing tickets upstream if things don't work with stripped libs) May 10 01:02:36 i've seen sstrip make some things unhappy May 10 01:02:41 ltrace iirc **** ENDING LOGGING AT Fri May 10 03:01:14 2019