**** BEGIN LOGGING AT Wed Aug 29 03:00:03 2018 Aug 29 06:46:13 jow: https://github.com/openwrt/openwrt/pull/1298 seen this ? Aug 29 07:37:28 huaracheguarache: i remember telling you my nexus 6p could connect w/80211w enabled, and chalked it up to lineageos. but uhm, no, it doesn't ;( i had wpad-mini installed so 80211w was a nop Aug 29 07:46:19 blogic: good idea, poor implementation Aug 29 07:46:20 m4t: I would be surprised if it doesn't support it Aug 29 07:47:04 blogic: everything about that script misses details or has subtle bugs Aug 29 07:47:11 yeah, it doesn't. i just tested it with wpad-full which is what made me think of that. Aug 29 07:47:14 it is a complete nogo Aug 29 07:56:12 blogic: did a review, but still almost every single line has a bug Aug 29 07:58:24 gninrom Aug 29 08:02:55 m4t: ah, ok Aug 29 08:03:18 nothing on my 5ghz seems to auth with it on :( Aug 29 08:03:27 =2 anyways Aug 29 08:04:32 apple stuff seems to support it (iphones, ipads, etc) Aug 29 08:05:04 and my laptop with a intel 7260 ac card supports it Aug 29 08:05:28 i just have non apple phone/tablet and some misc embedded devices Aug 29 08:05:52 ok Aug 29 08:06:40 i wonder if there's a way to see which devices are using it, if any? something with wpa_cli or..? Aug 29 08:08:37 I found out by running some speedtests on different devices since ath10k-ct firmware has a bug on the qca9984 that leads to low throughput on 802.11w clients Aug 29 08:09:36 hah Aug 29 08:15:17 jow: ok :-) i never looked that deeply Aug 29 08:20:35 hrm, list_sta/all_sta/etc return nothing with hostapd_cli Aug 29 08:36:36 mkresin: yes, xrx200 with vdsl2+vectoring and FXS, so only one core is usable and this probably makes the situation worse and the watchdog easier to trigger Aug 29 08:38:30 mkresin: and unfortunately I can't see a way to have the watchdog trigger the crashlog mechanism before rebooting, so my logs are empty Aug 29 08:40:54 mkresin: I had always assumed that it was related to the sudden ethernet link failures I was experiencing, so the log buffer would probably grow without bounds and cause OOM, triggering a reboot. Aug 29 09:48:21 carldani: if you are using the xdsl it is another issue Aug 29 09:49:24 carldani: a crashlog is only created if a kernel crash happens. if the watchdog kicks in due to high system load, there will be no crashlog Aug 29 10:03:58 m4t: you can do # iw dev station dump Aug 29 10:03:58 and check the value of MFP. If it says yes, the station is using 11w Aug 29 10:08:59 luaraneda: thanks Aug 29 10:13:54 I did some tests, and had to make 11w optional, because chromecasts can't connect when it's mandatory. Also, my laptop with an intel 7260 connects with 11w on Windows, but not in arch, which I attribute to Networkmanager/WPA_supplicant Aug 29 10:34:25 luaraneda: you need to patch wpa_supplicant afaik Aug 29 10:34:41 (I'm doing NM with 11w enabled) Aug 29 10:44:17 stintel: how are you doing NW with 11w? Mine doesn't like to do out of the box Aug 29 10:45:48 mkresin: NM supports it since a few releases back, but need to patch wpa_supplicant Aug 29 10:47:00 I actually submitted patches for that to wpa_supplicant, but apparently they were wrong Aug 29 10:47:58 wait a minute Aug 29 10:49:53 mkresin: http://w1.fi/cgit/hostap/commit/?id=3cdb4ac074f76accf24a51d143db545afad2c90b Aug 29 10:49:59 I believe this should actually be enough Aug 29 10:53:06 stintel: thanks. yet another distribution package I need to patch... Aug 29 10:53:28 mkresin: why do you think I am running Gentoo Aug 29 10:55:43 it's far easier to mkdir -p /etc/portages/patches/net-wireless/wpa_supplicant ; cd /etc/portages/patches/net-wireless/wpa_supplicant; curl -O http://some.patch; emerge -1 wpa_supplicant Aug 29 10:56:05 than to e.g. apt-get source, dpkg -S, blah blah, debuild, dpkg -i ../*.deb Aug 29 10:57:07 stintel: I've a script for that :-) Aug 29 10:57:47 let me know if just that patch is actually enough to enable it for you Aug 29 10:58:01 I submitted more patches but apparently they were wrong, and someone recently sent reverts for them Aug 29 11:06:32 stintel: sure, will report if I'm back on the laptop Aug 29 11:10:20 thanks Aug 29 11:11:59 me goes hunting for a kernel bug Aug 29 11:12:04 [ 18.891385] NOHZ: local_softirq_pending 08 Aug 29 11:12:18 introduced between 4.14.52 and 4.14.67 Aug 29 11:12:19 mkresin: you might also want to import some of the recent changes in http://w1.fi/cgit/hostap/log/wpa_supplicant/scan.c Aug 29 11:12:20 oh joy Aug 29 11:12:56 so that wpa_supplicant makes better decisions for selecting 2.4 vs 5 GHz Aug 29 11:13:02 (if you have both on same SSID that is) Aug 29 11:19:27 I previously only had http://w1.fi/cgit/hostap/commit/wpa_supplicant/scan.c?id=14204148781cf8e64827eae12dffacc869586073 but seems there might be a few more that could help Aug 29 15:28:48 anyone know why ldir hasn't been on irc for such a long time? Aug 29 15:30:24 he mentioned real life obligations Aug 29 15:31:38 oh, ok Aug 29 16:27:39 blogic: I saw the same "NOHZ: local_softirq_pending" on zynq a few days ago, but I assumed it was a zynq specific kernel option that I would have to send a patch to disable. Now I know zynq is not the only target with that message. ipq40xx is working fine Aug 29 17:11:28 stintel: your capabilties patch did the trick. still no option in the NM gui but MFP: yes Aug 29 17:49:18 Will someone give me a quote to gen a version of 18.06.1 for the Archer C7 v1.1 that does not try to use the 5GHz radio? There is a 15.05.1 build that does boot successfully. 8e58789494341d57a7f04ca0aaff8b11c33f740f97739ed3f34ebac467013f8b https://archive.openwrt.org/chaos_calmer/15.05.1/ar71xx/generic/ openwrt-15.05.1-ar71xx-generic-archer-c7-v1-squashfs-factory.bin Aug 29 18:07:35 profmac: can you not just disable that radio? Aug 29 18:09:24 iirc both radios are ath10k Aug 29 18:10:01 I tried to set the selector switch to "wireless off" and it still bootloops. I opened the case and removed the pci card and it does boot. lol. I should check if it is up at 2.5. Aug 29 18:11:01 I want to either come up with a solution for the entire community, whether it be a software gen or a replacement pci card, or else re-flash it to TP-Link stock and sell it without putting more time in. Aug 29 18:11:14 i wonder whether greearb_ can make ath10k-ct work on that early variant of the 9880 chip Aug 29 18:12:44 (Or at least not KP?) Aug 29 18:18:47 I was thinking, since 15.05.1 works and failsafe (18.06.1) works, there would be something pretty "simple" for someone who already has the toolchain set up. I'd rather work on my "real" project instead of doing this first. Aug 29 18:32:01 I see the ssid for the 2.4G radio running 15.05.1 Aug 29 18:32:50 I think a lot of devices with ath10k for the 5gHz still have ath9k for the 2.4gHz. I know my UAP-AC-LR does that. Aug 29 18:37:40 mamarley: i have an archer c7 v2, and it's ath10k(-ct) for both Aug 29 18:37:52 Interesting. Aug 29 18:38:53 I have both a V2 and a V1.1. The radios are different. Aug 29 18:39:06 DonkeyHotei: I have about a dozen archer c7's and they all are ath9k for 2.4GHz Aug 29 18:39:25 DonkeyHotei: including some version 1, some 2, some 3 Aug 29 18:40:08 DonkeyHotei: note: not disagreeing with your accessment, just pointing out that only god know what these hardware makers do Aug 29 18:40:17 Does anyone have a part number or whatever to order a 5GHz radio that I can swap in the v1.1? Aug 29 18:41:21 Almost 30 years ago, I managed software development at a small hardware mfr. The owner said in a meeting "we should introduce model numbers" and the hardware designer said "we have them now, we just call them serial numbers" Aug 29 18:43:19 jow: now i get the same problem in 'staging_dir/host/bin' all the '.bins' files are missing as well. Aug 29 18:43:27 Zero_Chaos: you are mistaken. the v2 does not use ath9k in any way, shape, or form, and i doubt the v3 does, either Aug 29 18:44:12 v1.x i was not sure about Aug 29 18:44:14 'staging_dir/toolchain-*' works fine now Aug 29 18:44:53 I have both routers online, and am ssh'd into them, fwiw. Aug 29 18:46:10 Does some system log somewhere have whatever ID is read from the actual pci device at sysinit? Aug 29 18:46:57 I'm seeing verious cards for about the same price as I paid on eBay for the router. Aug 29 18:50:18 (18.06.1) dmesg | grep ath --> [ 12.264516] ath10k_pci 0000:01:00.0: qca988x hw2.0 target 0x4100016c chip_id 0x043202ff sub 0000:0000 Aug 29 18:50:42 Off topic: Someone likes James BOND? Aug 29 18:51:05 profmac: i'm assuming that's on the v2 Aug 29 18:52:50 DonkeyHotei: interesting, you are correct. it loads ath9k but it isn't actually in use Aug 29 18:53:24 Zero_Chaos: In what manner are you determining that it isn't in use? Aug 29 18:54:06 mamarley: dmesg | grep ath Aug 29 18:54:24 (I actually thought this was true of my UAP until I rmmodded it and my 2.4gHz interface went down.) Aug 29 18:55:17 mamarley: "lspci -k" is also a good indicator Aug 29 18:55:29 (Also, I said "-LR" previously but that was an error, it is actually a "-PRO".) Aug 29 19:00:45 DonkeyHotei, yes the v2. I am reading the v1.1 dmesg dump, I find switches, SoC, not anything I recognize as the radio. Aug 29 20:19:59 Here is the PCI card on the Archer C7 v2, this router works with 18.06.1. Aug 29 20:32:41 profmac: that's the difference between QCA9880-AR1A and QCA9880-BR4A, the later is supported - the former isn't Aug 29 20:49:39 pkgadd, I agree. If I buy an BR4A aftermarket card and drop it into a V1.1 running 18.05.1, should it work, and has anyone confirmed that? Aug 29 20:50:50 Here is the AR1A in the v1.1, just to be complete. https://photos.google.com/photo/AF1QipON4oPh7naSHbs4CjAWKXBQ8hoTs5OecRbHFo9Z Aug 29 20:52:10 it should work and afaik was quite common practice early on during the archer c7's support cycle under openwrt, but I don't own any archer c7 Aug 29 21:09:45 mkresin: Do you remember the bug number for the xrx200 xdsl high load problem? I'd like to investigate further once I have some spare time. Aug 29 21:10:45 mkresin: Getting a backtrace when the watchdog fires would be helpful to determine where the time is spent... Aug 29 21:13:00 hold up Aug 29 21:13:21 the c7v2 does not use ath10k for 2.4ghz. Aug 29 21:13:35 ath10k requires firmware. ath9k does not Aug 29 21:16:06 [Wed 2018-08-29 02:13:20 PM PDT] the c7v2 does not use ath10k for 2.4ghz. <----- yes it does and it requires ath10k(-ct) firmware to operate that 2.4ghz Aug 29 21:19:27 What's the reason that ARM builds with uClibc and not MUSL? Aug 29 21:20:35 philipp64|laptop: this is news to me Aug 29 21:21:25 mangix: never mind, i had it confused with a different device. the c7v2 has AR9550 Aug 29 21:21:28 Let me ask a different question then… why are we still supporting uClibc? Aug 29 21:21:50 What does uClibc offer at this point that MUSL doesn't have? Aug 29 21:22:05 My c7 v2 has a QCA9558-AT4A chip on the motherboard, in addition to the pci card. See https://photos.google.com/photo/AF1QipPzmeF9O7YGq_h7KCdV8zDTMZwB5pQ8B-f1-2Oz Aug 29 21:22:08 we technically support glibc and eglibc too Aug 29 21:22:40 DonkeyHotei: yeah i just grepped the kernel. QCA955X is located in ath9k Aug 29 21:23:08 DonkeyHotei: Sure, but I thought the default across all platforms was MUSL since 2015. seems it's not. Aug 29 21:23:45 philipp64|laptop: i'm pretty sure uClibc has been non-default across the board ever since Aug 29 21:23:49 what uses uclibc? Aug 29 21:24:00 mvebu and ipq806x default to musl Aug 29 21:24:02 mkresin: I think it's also a bit weird that there's no kernel.softlockup_panic sysctl on xrx200, at least in OpenWrt 18.06.1 Aug 29 21:25:39 after I removed "option htmode" and added "option legacy_rates 1", iwinfo still shows "Type: nl80211 HW Mode(s): 802.11bgn", why is it not 802.11bg instead? tried htmode NONE with no help either Aug 29 21:26:00 Well, what I want to do is to run 18.06.1 on the v1.1 without pulling the pci card. I thought someone who has the toolchain set up could gen-up a version faster than I could get another router and/or pci card or learn the tool-chain myself. Aug 29 21:26:19 mkresin: otherwise, a softlockup would trigger a panic before triggering the generic watchdog, and that might give us a better understanding of why xrx200 has this high load issue. Aug 29 21:26:55 profmac: if the 2.4ghz indeed uses ath9k, then all you gotta do is build an image with no ath10k drivers Aug 29 21:27:12 imagebuilder will do this Aug 29 21:28:30 ausjke: erm. what are you trying to achieve? Aug 29 21:29:42 mangix: still the zbt1326, someone said put it to legacy mode can make 2.4G work Aug 29 21:30:11 i doubt that Aug 29 21:30:42 are you using dangole's staging tree? Aug 29 21:30:56 (my zbt we1326 is still happily serving 2.4gig, fwiw. on r7742-6031ab345d) Aug 29 21:31:03 turns out no iw, iwinfo, whatever, to tell you which mode the wifi is in, be it 802.11abgn-ac, very ambiguous it seems Aug 29 21:31:31 karlp: are you using it 24x7, it sometimes disconnect for a while, then resume Aug 29 21:31:43 probaly, but that's transparent to me. Aug 29 21:31:53 you wewre normally complaing that it didn't work at all. Aug 29 21:32:00 i run a script to monitor it all the time, and yes it disconnects, another router with qualcommn never disconnect, 2.4G only for both Aug 29 21:32:07 carldani: there is no open bug to my knowledge. and the issue shows only if the dsl daemon runs but no xdsl is connected Aug 29 21:32:27 karlp: you're subjective, this is serious for me and i have to make sure it works 100% of the time Aug 29 21:32:43 otherwise i can just throw this router away and buy a new one Aug 29 21:32:56 I thought you already had, I'm kinda surprised you're still insisting on it. Aug 29 21:33:52 already invested quite some money, once a while i ping back and check see if mt76 saves the day, and i saw someone reporting legacy mode can do it, though at slower speed, better than dead connections Aug 29 21:35:11 now i don't know how to check to make sure openwrt is in bg or n mode, neither iw nor iwinfo does it explicitly Aug 29 21:35:37 for me they are both bgn no matter what htmode is in(no AC here) Aug 29 21:35:51 Type: nl80211 HW Mode(s): 802.11bgn Aug 29 21:36:10 mkresin: hm... in that case, my bug seems to be unrelated. Aug 29 21:38:41 anyway fwitw 'iw dev wlan0 station dump' can tell, 11n will have MCS info in 'tx bitrate' Aug 29 21:39:19 legacy does not have MCS info, that should confim it,sadly, 2.4G/legacy still disconnects, check back later then Aug 29 21:39:21 carldani: teh softlock thingy is only available if CONFIG_SOFTLOCKUP_DETECTOR = y Aug 29 21:39:31 carldani: which most likely isn't by default Aug 29 21:40:01 mkresin: so it boils down to recompiling the kernel and testing... Aug 29 21:40:52 carldani: might be a good idea to enable the hardlock detector as well Aug 29 21:41:32 mkresin: and activate panic for all lockups and problems so I have a crashlog after the automatically triggered reboot on panic Aug 29 21:42:04 ausjke: AFAIK that's an EEPROM issue Aug 29 21:42:13 carldani: which might even allow to investigate the watchdog reboot I'm talking about Aug 29 21:42:19 mkresin: Sounds like a plan. Aug 29 21:42:57 ausjke: my advice is, ask karlp for his and flash it. if that doesn't work, trash can. Aug 29 21:43:14 mkresin: I have one spare o2 box 6431 and one in production, so I'm somewhat free to experiment. The only downside is that sometimes it takes more than a week to spuriously reboot. Aug 29 21:43:34 mkresin: (wifi is off, I handle that on a different box) Aug 29 21:43:35 mangix: worse, someone from taiwan suggesting it is a chip issue, sensitive to heat or something, only happens to 7603e Aug 29 21:44:06 karlp: mind sharing your eeprom...? Aug 29 21:45:10 it can at least be used for a 5G/AC router, so not totally trash Aug 29 21:45:27 mkresin: what's the fastest way to rebuild an image with a new kernel? sdk, then imagebuilder? full git checkout? I'd rather not have to recompile the whole toolchain. Aug 29 21:45:51 carldani: new kernel means git checkout Aug 29 21:46:20 if you give me instructions, sure. Aug 29 21:46:46 mkresin: Meh. Can I at least reuse the toolchain from the sdk? Aug 29 21:47:25 carldani: no idea. never used the sdk Aug 29 21:48:12 karlp: dd copy /dev/mtd2 or 3 Aug 29 21:48:20 carldani: hardlock detector will not work. mips/xrx200 obviously isn't a HARDLOCKUP_DETECTOR_ARCH Aug 29 21:49:27 carldani: if you have the o2 box selected in menuconfig, make kernel_menuconfig is your friend Aug 29 21:49:35 mkresin: so it's a pure hardware hard-reboot watchdog which can't be redirected to function as sort of NMI? Aug 29 21:50:33 ausjke: note you will have to modify it since the MAC address of your router is located there. or just flash if you don't care Aug 29 21:50:52 carldani: yes, hard reboot. it is embedded into the xrx200 SoC. if the watchdog is armed but not pinged, the SoC is doing a reset Aug 29 21:52:00 mkresin: do you know if this condition can be detected in the subsequent boot, e.g. via some bits in a config register or somesuch? Aug 29 21:52:29 carldani: yes, possible. let me search my patches dir Aug 29 21:53:30 mkresin: That way we could print a message "Watchdog was triggered before this boot" Aug 29 21:53:48 carldani: https://www.dropbox.com/s/q776xbqbt14sbwp/lantiq-print-reset-status.patch?dl=0 Aug 29 21:54:10 carldani: meaning of the bits will follow in a sec Aug 29 21:55:24 mkresin: Any objections if I take that patch and extend it with verbose printing, then submit? Aug 29 21:56:12 mkresin: I have extended the crashlog patch as well to print a message if a crashlog is available, but need to check if it works as intended. Aug 29 21:57:18 karlp: cat /proc/mtd is mtd2 is "factory" then most like it will be the eeprom Aug 29 21:57:37 dd if=/dev/mtd2 of=/tmp/factory.bin bs=1k count=64 Aug 29 21:57:47 carldani: unlikely that it will be merged. it is really just for debugging. using /dev/mem would have the same effect Aug 29 21:58:21 md5sum /tmp/factory.bin -- my md5sum is: 79929ef44150f0f5a2cc0f9d92ca3105 Aug 29 21:58:28 mkresin: AFAICS xrx200 is tens of thousands of lines of unmerged patches anyway... Aug 29 21:58:48 carldani: I'm to lazy to check which bits the relvant are. but found the following in my sent folder: 0x70390300 - normal reboot, 0x58390300 - boot after power on, 0xD0390300 - reboot forced by Aug 29 21:59:02 mkresin: thanks. Aug 29 21:59:38 carldani: you might want to check what is already upstreamed. custom/owrt code is only a fraction of it Aug 29 21:59:59 mkresin: IMHO it makes sense to have some debug functionality in production as long as the device is somewhat unstable. Aug 29 22:00:42 ausjke: no need for bs or count Aug 29 22:00:43 mkresin: I thought the PCI stuff and a few other patches were still unmerged, but yes, I need to check against current upstream. Aug 29 22:00:53 dd reads untill EOF Aug 29 22:01:48 yes but paranoid Aug 29 22:01:56 mtd read also works Aug 29 22:02:03 mtd write is needed for writing Aug 29 22:02:11 dd of=/dev/mtd2 does not work Aug 29 22:03:02 carldani: sure it isn't everything upstream. but a lot is. pci driver would need a full rewrite to get accepted upstream. same for everything xdsl related Aug 29 22:03:33 carldani: hauke wrote a new xrx200 switch driver and is working on getting it accepted upstream Aug 29 22:04:35 carldani: I don't see the gain of adding the debug stuff. most user don't care/have no clue what to do with and the ones who have are enabling the debug stuff if required Aug 29 22:06:08 mkresin: oh, I didn't know that Hauke is working on a new xrx200 switch driver. My apologies. Aug 29 22:06:38 mkresin: Indeed, pushing old stuff is counterproductive if new stuff is in the works and actually mergeable. Aug 29 22:08:10 carldani: everything that is keep out of tree was either rejected before is known to be in a state not being accepted upstream Aug 29 22:08:53 carldani: pushing what we have in the state we have would only wastes peoples time Aug 29 22:10:40 mkresin: I didn't know that, I assumed xrx200 was in maintenance mode because I didn't see stuff getting upstreamed recently. Aug 29 22:11:16 auske: https://nc.beeroclock.net/s/f35bdc2syLeL7wk Aug 29 22:13:26 carldani: in the end it is a matter of resources and time. quite a lot of work to do for free ;-) Aug 29 22:20:51 mkresin: ah yes, I know that from working on flashrom Aug 29 22:21:13 mkresin: there also is offline life, and it sometimes takes precedence Aug 29 22:22:48 karlp: is that domain yours? Aug 29 22:25:20 carldani: soem of the drivers are written as contracted work or were sponsored Aug 29 22:25:56 carldani: writing a switch driver isn't done in a few hours Aug 29 22:26:27 mangix: enough to be able to share links from it? Aug 29 22:26:50 file sharing still sucks in the 21st century Aug 29 22:29:05 mkresin: I'm happy that xrx200 works as well as it does, and for the Vodafone Easybox 904 xDSL as well as the Telekom Speedport W921V there are even people who try to get unsupported models supported. Aug 29 22:31:40 mkresin: Yes, I can imagine that writing a switch driver is not easy. When I reverse engineered the closed nvidia nforce network driver to create the free forcedeth driver in Linux, I saw firsthand that network drivers in general are not the easiest target. Aug 29 22:34:07 mangix: why do you ask? Aug 29 23:11:19 karlp: just wondering what people use to host nextcloud. Aug 29 23:11:26 ARM or x86 Aug 29 23:15:10 I'm using an odroid c2 Aug 29 23:16:57 not running openwrt i'm guessing Aug 29 23:17:06 no :) **** ENDING LOGGING AT Thu Aug 30 02:25:41 2018