**** BEGIN LOGGING AT Wed Aug 15 02:59:59 2018 Aug 15 05:44:24 stintel: :-) Aug 15 06:18:07 hello All! Aug 15 06:18:34 Experts, Which do you recommend, compile everything from source, or install trunk image, with packages ? Aug 15 06:39:00 Experts, Which do you recommend, compile everything from source, or install trunk image, with packages ? Aug 15 06:40:09 depends on what you're aiming for Aug 15 06:54:13 Borromini: When stable home use for few month Aug 15 06:54:13 s Aug 15 06:54:25 *For Aug 15 06:55:29 is your device not supported by 18.06? Aug 15 06:56:03 or do you have requirements satisfied only by master? Aug 15 07:00:31 Borromini: I usually like the latest drivers, firmware for my Archer C7 v2 Aug 15 07:01:41 ok. master is a moving target and i think kernel modules need to match the image's kernel still Aug 15 07:02:00 the image generator allows you to pack everything into the image prior to flashing Aug 15 07:02:24 Borromini: So compilation seems to be better Aug 15 07:02:30 If I understand well Aug 15 07:03:04 xcoom: compilation won't change the scenario for kernel modules Aug 15 07:04:20 they still need to match, however, if you compile yourself and keep the binaries around, you won't run into needing modules that aren't available anymore Aug 15 07:04:32 so yes, in that case it's handier to compile yourself Aug 15 07:32:32 xcoom: Borromini: that's not quite an issue anymore, as we keep an archive of older kmods, see e.g. https://downloads.openwrt.org/snapshots/targets/mvebu/cortexa9/kmods/ Aug 15 07:34:23 KanjiMonster: These sites are available via Ipv6 too, right? Aug 15 07:35:40 xcoom: yes Aug 15 07:46:04 KanjiMonster: neat, thanks Aug 15 07:47:17 ofc you still get bleeding edge with the snapshots, so be prepared for bugs or stuff missing/not working Aug 15 07:48:00 * Borromini is only test driving ath79 atm :) Aug 15 07:56:25 KanjiMonster: Current 18.06 == Trunk around 2018. March ? Aug 15 07:56:37 KanjiMonster: Or it's not this simple? Aug 15 07:57:15 ath79 is fine for 'production' on my tl-wdr3600 (and I've finally been able to ditch ar71xx for the last two devices yesterday) Aug 15 08:00:46 xcoom: more around May IIRC, but a lot of fixes and simple board support were backported from master to 18.06 Aug 15 08:03:40 pkgadd: what other ar71xx stuff are you running? Aug 15 08:07:41 Borromini: tl-wdr3600, tl-wdr4300, tl-wr1043ndv1, tl-wr941ndv2 Aug 15 08:08:05 ok Aug 15 08:08:19 1043 v1 working well (wireless)? Aug 15 08:08:31 btw did you replace your uboot on that? Aug 15 08:10:59 wireless works as 'well' as before on ar71xx, which is 'o.k.', but not perfect (hardware quirks with that draft-n wireless). OEM u-boot (but one of the latest upgrades that retrofitted push-button tftp recovery, by far not as convenient as u-boot_mod, but it has saved my cookie more than once) Aug 15 08:12:49 :) Aug 15 08:13:02 wl-wdr3600/4300, bt hh5a or nbg6817 provide /much/ better wlan stability Aug 15 08:13:10 s/^wl/tl/ Aug 15 08:15:14 yeah the 1043 v1/2 i have are nothing more than backup devices at this point Aug 15 08:16:18 any one know anything about disabling gso. gro and tso with ethtool? I just would like to see if they have any effect on SQM. Aug 15 08:17:26 I don't think it will. but it's something I read in the forums and thought I would give it a quick test. Aug 15 08:22:16 ethtool changes dont last past reboots anyway Aug 15 08:22:37 Borromini: the v2 is much newer (720 MHz and 64 MB RAM and rather current ath9k hardware), that one shouldn't be bad at all (except for the obvious, 2.4-GHz-only) Aug 15 08:23:43 Borromini: v1/ ath79 dmesg, http://paste.debian.net/1038001/ Aug 15 08:25:46 speaking of good wifi, does anyone here have the glinet B1300 (ipq40xx), are those good? Aug 15 08:26:24 pkgadd: thanks Aug 15 08:26:35 I have some legacy devices that are only 2.4ghz and the current zbt boxes are too flaky Aug 15 08:27:16 abenz: zbt being mt760x? Aug 15 08:27:25 abenz: I don't have any personal experience with ipq40xx devices, but I've gotten feedback from someone who uses three AVM Fritz!Box 4040 as APs (only), he's very happy with it (ipq4018 vs ipq4028) Aug 15 08:27:41 Borromini: https://wikidevi.com/wiki/ZBT_WG3526 Aug 15 08:28:11 pkgadd: thanks for the feedback, but what does this part mean: (ipq4018 vs ipq4028) ? Aug 15 08:28:36 I should aim for 4028 as opposed to 4018 ? Aug 15 08:28:48 here's the B1300: https://wikidevi.com/wiki/GL.iNet_GL-B1300 Aug 15 08:28:50 abenz: the Fritz!Box 4040 uses ipq4018 SOC/ wlan, the GL.Inet B1300 is based on the ipq4028 instead Aug 15 08:29:02 I see Aug 15 08:29:50 quite the heatsink Aug 15 08:29:55 https://www.qualcomm.com/products/ipq4018 vs https://www.qualcomm.com/products/ipq4028 Aug 15 08:32:50 I'm looking for ipq40xx myself (fritz!box 4040 or zyxel nbg6617), but only if I can find one for a bargain (I don't /really/ need it, so I'm not going to pay the normal price - even if prices aren't bad at all) Aug 15 08:33:33 I'm contemplating the GLInet since it seems they are quite open source friendly Aug 15 08:34:13 reading their forums. but hesitant because I see people with the same chipset as in my box without problems. So I'm suspecting it could be the cheaper construction of the zbt box thats causing the issues Aug 15 08:34:36 power delivery or cheapening out in secondary components or something Aug 15 08:35:22 so the fritzbox could be great, but not the alibaba-sourced copy with the exact same hardware specs Aug 15 08:35:24 it's a nice device, but I really want to retain the option of using it as a full blown router <-- 4+1 ethernet ports Aug 15 08:35:54 I will keep you posted on my unit when I get it.. I think i wil order one and see Aug 15 08:36:14 dissent tells me the 40xx even outperforms the ipq806x! Aug 15 08:36:15 thanks Aug 15 08:36:33 abenz: isn't 40xx a crippled 806x? Aug 15 08:36:37 the are some nice (but expensive) triple-radio devices based on ipq40xx Aug 15 08:37:26 the 806x has the nss and all those fancy bits that are not used in offloading everything when using LEDE. There's a bottleneck in the ipq806x in hardware design Aug 15 08:37:57 ok :) Aug 15 08:39:42 Borromini: I wouldn't go that far, it's a pretty different SOC. lower cost 4*618 MHz cortex A7 (ipq40xx) and no NSS cores vs 2* 1.7 GHz KRAIT300 ~cortex a15 + 2* 800 MHz NSS cores. so yes, CPU wise it's (much) slower - but at the same time it has two integrated (cheaper) dual-band radios, rather than connecting full blown PCIe cards on the board Aug 15 08:41:17 pkgadd: ok, so different trade-offs then Aug 15 08:41:59 yes, also - you can get ipq40xx devices for ~70 EUR, while you have to pay ~180 EUR for ipq806x Aug 15 08:42:32 i found ipq806x devices to be quite pricey... i'm not forking over 250 € for a home router :-/ Aug 15 08:43:22 Borromini: keep an eye on amazon warehouse deals, I paid 124 EUR for my nbg6817 - which was more than I would have liked, but at least not 180 EUR... Aug 15 08:43:52 I had a soft limit of ~100 EUR and a pretty hard one around 130 EUR Aug 15 08:44:12 :) Aug 15 08:44:24 so far i'm still quite satisfied with my dir-860l Aug 15 08:44:42 i'd push them to friends but the 2,4 GHz still seems to be kind of a dealbreaker. Aug 15 08:45:37 pkgadd: was that new or refurb? Aug 15 08:45:39 that's most of the reason why I went with ipq8065 instead of mt7621 (2.4 GHz is more important than 5 GHz for me) Aug 15 08:46:44 Borromini: claimed to be used (probably sent back during the first 14 days), but complete package and very good condition Aug 15 08:46:58 ok Aug 15 08:47:01 neat Aug 15 08:49:57 one pretty important use case for me is a WDS link (nbg6817 <--> tl-wdr3600) diagonally through the house, so one of the worst tasks for 5 GHz - so I really need 2.4 GHz there for its indoor range (2.4 GHz is twice as fast as 5 GHz over that range) Aug 15 08:50:51 around 4-9 MByte/s over 2.4 GHz (HT20) Aug 15 08:51:00 err, 8-9 Aug 15 08:51:53 ath9k <--> ath10k? Aug 15 08:51:58 yes Aug 15 08:52:05 I tried that with a tl 4300 and c2600 but it didn't seem stable Aug 15 08:52:19 but that was long ago Aug 15 08:52:39 it's stable Aug 15 08:53:05 and performance is better than tl-wdr4300 <--> tl-wdr3600 over the same distance before Aug 15 08:53:42 I see Aug 15 08:55:25 I do notice better range with 802.11ac over 802.11n in general Aug 15 09:46:19 going to power on my Gemini PDA - let's see if ath10k-ct and ct-wmi-mgt fw survives that :) Aug 15 10:41:47 3700 v1 and 1043 v2 on ath79 now, but something's wrong with my per device rootfs, clearly. no added packages or opkg at all :P Aug 15 13:31:19 I'm building 18.06 from source. The whole interaction on x86/64 and the HID drivers with what openwrt make menuconfig and make kernel_menuconfig offers is just very confusing. Also, it doesn't make sense to make these HID drivers as packages, they should be built in from the beginning as it's impossible to add these packages if HID doesn't work. Where is this defined, I'd like to see if I can Aug 15 13:31:26 make a pull request suggesting change. Aug 15 13:34:00 so for instance kmod-usb-hid_4.14.62-1_x86_64.ipk kmod-hid_4.14.62-1_x86_64.ipk and so on (and whatever they depend on) should just be already in the image and not .ipk files that one has to install after boot Aug 15 13:36:05 SwedeMike: in target.mk under your target/subtarget Aug 15 13:39:21 Borromini: thanks, so add to "DEFAULT_PACKAGES += kmod-button-hotplug kmod-e1000e kmod-e1000 kmod-r8169 kmod-igb" that is in the x86/64 target already. Aug 15 13:39:54 is there a reason this is so stripped out? Aug 15 13:40:51 not that i know of. I thought there had been a patch before Aug 15 13:41:23 mm, should probably be enough by default to use a usb keyboard, yes Aug 15 13:41:38 shouldn't be any real space constraints on x86 boxes Aug 15 13:42:42 jwh: that's what I thought as well. Doesn't make sense to not include any usb support at all from start basically. Aug 15 13:43:07 although mine work, don't think I had to select anything Aug 15 13:43:07 hm Aug 15 13:43:09 so my proposal is going to be usb/usb2/usb3 and the hid drivers to be included by default Aug 15 13:43:15 maybe thats just usb1 Aug 15 13:43:43 jwh: perhaps your motherboard emulates this as ps2 keyboard somehow? That worked last I tried, nothing true usb Aug 15 13:43:56 because there are no usb hid drivers or usb drivers installed at all by default Aug 15 13:44:00 SwedeMike: you don't need usb2/3 Aug 15 13:44:10 i have this in my config: Aug 15 13:44:10 SwedeMike: its a bmc thinger, usb Aug 15 13:44:11 CONFIG_PACKAGE_kmod-hid=y Aug 15 13:44:12 CONFIG_PACKAGE_kmod-hid-generic=y Aug 15 13:44:32 CONFIG_PACKAGE_kmod-usb-hid=y Aug 15 13:44:39 Borromini: Right up until you have a system with no OHCI or UHCI controller. Aug 15 13:44:47 SwedeMike: https://git.openwrt.org/cdedb24ef89ffbab8206450fbfd021af6ad90b30, it's enabled by default, those kmod's that You installed are empty ones. Aug 15 13:45:07 Monkeh: no backwards compatibility in newer controllers? Aug 15 13:45:25 Borromini: Why use three drivers when you can use one? Aug 15 13:45:26 tmn505: I have tried on multiple machines, by default I get nothing when I insert usb keyboard if I don't include those other drivers. Aug 15 13:45:44 Most still pretend it's 1998, but a silly assumption to make. Aug 15 13:45:50 Monkeh: i'm betting the two others got enabled by usb-hid :P Aug 15 13:46:03 I've wasted hours on this problem in the past weeks. Aug 15 13:46:04 -ash: lsusb: not found Aug 15 13:46:05 hnng Aug 15 13:46:07 Borromini: .. no, not the HID stuff.. Aug 15 13:46:14 and I don't even have new hardware, most of it is from 2009 Aug 15 13:46:29 when I tried to do this on 2016 hardware, it was even worse. Aug 15 13:46:33 and now its hung Aug 15 13:46:38 Monkeh: ok. Aug 15 13:46:43 I think I have broken mtu somewhere :/ Aug 15 13:47:49 it's also that if I in kernel_menuconfig say something should be built in and I happen to have selected it in openwrt menuconfig, then it's built as ipk anyway Aug 15 13:48:05 it's all really confusing to get to work Aug 15 13:49:03 built in just means that its installed onto the image Aug 15 13:49:08 still packaged as an ipk Aug 15 13:49:09 SwedeMike: if the ipk is empty, then it is built-in Aug 15 13:49:16 yeah Aug 15 13:49:33 SwedeMike: usb keyboard should work (TM) on 18.06 ootb, may bee that it's requiring usb controller drivers, most of them are built-in by default https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=target/linux/x86/config-4.14;h=bde549b1e31930dea8eb42c7293219554af289cd;hb=refs/heads/openwrt-18.06#l433 Aug 15 13:49:52 I will re-do this tomorrow from scratch and figure out what's actually included by default Aug 15 13:50:02 SwedeMike: What machine were you trying this on? Aug 15 13:50:19 SwedeMike: ./scripts/diffconfig.sh will show you what's deviating from the stock config Aug 15 13:50:21 Monkeh: an i7-940 x58 motherboard machine. Aug 15 13:50:23 usually that's pretty limited Aug 15 13:52:00 since it has dual port r8169 I can unplug the 10GE cards I have so those become eth0/eth1 and do this via network, but otherwise it defaults to eth2/eth3 and then these 8169 can't be used to login to the machine with. I guess I can edit /etc/config/network before I boot but... it's still a huge pain Aug 15 13:52:15 SwedeMike: Right, so that'd be UHCI - so, which genius included OHCI but not UHCI in the kernel? Aug 15 13:52:37 Monkeh: I have no idea, that's what I'm trying to figure out. I can't be alone in this pain. Aug 15 13:52:48 or I don't care who did what, I just want to improve things. Aug 15 13:53:10 this was a pain in 17.01 when I tried that last year, and it's a pain now with 18.06 Aug 15 13:53:46 SwedeMike: default network config would just configure the first NIC on x86 i think? Aug 15 13:54:05 SwedeMike: Well, enabling UHCI is probably all you need to do Aug 15 13:54:19 Monkeh: right, I'll look into that. Aug 15 13:54:23 hm, I must have enabled ehci in mine Aug 15 13:54:27 don't remember doing that Aug 15 13:54:44 i have ehci=y, ohci=m, uhci=m Aug 15 13:55:02 basically, I'll try to figure out how I can make this work out of the box on both this X58 motherboard and the X170 motherboard machine I have, if I can do that I imagine it should work for a lot of people Aug 15 13:55:46 Borromini: Define 'first NIC' Aug 15 13:56:05 Monkeh: that which get listed as eth0 afaik. Aug 15 13:56:12 it's what bit me when i first set up my APU2. Aug 15 13:56:13 So which one is that? Aug 15 13:56:20 Monkeh: excellent question. Aug 15 13:56:36 i know on my APU2 it was the first out of the three it has :D Aug 15 13:56:41 It is heavily dependent on hardware, namely the order in which the PCI(e) devices are enumerated. Aug 15 13:56:55 DHCP server was only listening on eth0 on my APU2, by default Aug 15 13:57:35 mamarley: That is what I'm trying to get at - it will depend on device address and load order Aug 15 13:59:50 on x86/64 it makes eth0 and eth1 wan/lan respectively. Aug 15 14:00:03 but if you change what nic drivers you have, these move around Aug 15 14:00:22 so when I use kernel_menuconfig to install mlx4_core they all of a sudden become eth0/eth1 and r8169 is then eth2/eth3 Aug 15 14:00:41 without the mlx4_core driver, r8169 is eth0/eth1 Aug 15 14:01:04 need some mac address matching Aug 15 14:01:32 ubuntu has changed now to pci slot based numbering of ethernet interfaces Aug 15 14:01:43 thats systemd Aug 15 14:01:57 No, that's just udev Aug 15 14:01:58 ok, doesn't matter, things are less prone to move around now anyway. Aug 15 14:02:01 they just disabled it befpre Aug 15 14:02:03 befoe Aug 15 14:02:07 udev is part of systemd now. Aug 15 14:02:08 Monkeh: systemd-udev Aug 15 14:02:14 And it sucks because it's trivially broken and totally unpredictable without precise foreknowledge of the hardware. Aug 15 14:02:40 its pretty shit until you work it out yeah, but then it stays consistent Aug 15 14:02:48 Consistently broken, too Aug 15 14:02:54 why Aug 15 14:03:12 Because they didn't consider all cases so it breaks when reality meets their good intentions Aug 15 14:03:26 you can always change it Aug 15 14:03:31 Turn it off then, if it doesn't work for you. Aug 15 14:03:32 Like this stack of cards I have which have a single chip with a single function to drive two ports Aug 15 14:03:34 either the method it uses to name, or disable it Aug 15 14:03:46 yeah, on this ubuntu 16.04 I have other problems, sometimes things move around a little bit but it's more subtle so it's hard to find what it is and how to fix it. Aug 15 14:03:52 mamarley: I turned it off long ago Aug 15 14:03:59 Monkeh: tahts why it has an fX suffix Aug 15 14:04:06 my cards are similar, they end up being f0, f1 Aug 15 14:04:11 for port 0, 1 Aug 15 14:04:14 jwh: That's for function. Aug 15 14:04:19 It doesn't have two functions. Aug 15 14:04:30 it has 2 ports Aug 15 14:04:40 But only one function. Aug 15 14:04:44 not really Aug 15 14:04:49 each port is independant Aug 15 14:05:02 .. no, literally, only one PCI function. Aug 15 14:05:07 That's how things get named. Aug 15 14:05:14 yes Aug 15 14:05:28 same pci path, but differered by fX Aug 15 14:05:36 ..., yes, by PCI function Aug 15 14:05:41 Of which these cards have only 1. Aug 15 14:05:50 Therefore both network interfaces are function 0. Aug 15 14:05:59 hm Aug 15 14:06:36 btw, what NIC drivers are available in openwrt make menuconfig, how were those chosen? I'd like to include more? I'd rather stay out of make kernel_menuconfig, it just seems to break. Right now I'm in a situation where the kernel keeps asking me questions every time I have done make (openwrt) menuconfig , becuase there is something that openwrt menuconfig doesn't set, and it asks me the Aug 15 14:06:42 "N/y/m/whatever" question every time I trry to compile after that Aug 15 14:06:53 jwh: As I said, their good intentions met reality, they didn't think it through, and they've flatly refused to improve it Aug 15 14:07:05 SwedeMike: make oldconfig Aug 15 14:07:13 and use profiles (./scripts/env) Aug 15 14:07:14 Monkeh: how common are single function chips though? Aug 15 14:07:20 jwh: *shrug* Aug 15 14:07:30 they wm Aug 15 14:07:44 won't fix it until its commonly seen Aug 15 14:08:12 Ah, so it doesn't just have to be a bug, but a BIG bug! Aug 15 14:08:19 heh Aug 15 14:08:24 The whole scheme sucks anyway Aug 15 14:08:34 not sure how they'd fix it Aug 15 14:08:42 The 'predictable' names are 100% unpredictable without booting and looking first on commodity hardware Aug 15 14:08:45 as systemd would need to care about the way chips operate Aug 15 14:08:54 Now if they'd just gone with MAC based I'd have not a care in the world, that makes SENSE! Aug 15 14:09:13 Sure, if you swap a card it changes - and that's predictable. Aug 15 14:09:30 I'm not convinced mac based is any better Aug 15 14:09:45 as you'd need to know the macaddr in advance too Aug 15 14:09:49 I'm pretty sure you can put it in MAC mode, and it does default to this mode for USB adapters. Aug 15 14:09:52 If I want to know what the port is called, on most machines I just read it off the label rather than guess what their PCI topology is. Aug 15 14:10:11 mamarley: You can, but every system still boots with the unpredictable broken default Aug 15 14:10:17 that works if you have access to the machines and they're actually printed Aug 15 14:10:49 I have never seen the PCI topology of any board printed on the board, and it is often difficult to even find documentation on it at all. Aug 15 14:10:53 it depends on your situation eally Aug 15 14:10:58 really* Aug 15 14:11:00 jwh: And if not you need to boot and look - so, not a downside from the default. Aug 15 14:11:18 Monkeh: yeah, but also not really useful for the large majority of cases Aug 15 14:11:29 IMO MAC based is a far saner and more predictable default, if you require physical ports to have a consistent name across different machines of the same model, change to that default and know your hardware Aug 15 14:11:43 In real-world practical every-day computers, + a few semi-corner cases, that default sucks. Aug 15 14:11:57 its mostly designed for end users Aug 15 14:12:13 For end users it's stupid, because for them it was always going to be 'eth0' Aug 15 14:12:20 but in reality end users don't tend to have multiple interfaces anymoe Aug 15 14:12:22 anymore Aug 15 14:12:29 most may not even have ethernet now Aug 15 14:12:53 And now it's enp$RANDOMNUMBERs$RANDOMNUMBERf0 Aug 15 14:12:56 personally I've always disabled it Aug 15 14:13:14 not because its unpredictable at first boot, but because I cba typing the interface name in :D Aug 15 14:13:19 That too. Aug 15 14:14:02 although I can see the benefits Aug 15 14:14:21 As can I, I also see the downsides and reality breaking it Aug 15 14:14:23 my openwrt boxes as an example, some of them have >=8 interfaces Aug 15 14:14:35 But in true systemd fashion, "We know best, we're never wrong" Aug 15 14:14:36 and adding pcie cards moves the onboard Aug 15 14:14:49 (and the rest, but the onboard is the most useful since it has management on) Aug 15 14:15:06 so I can see why they did it Aug 15 14:15:34 There are ways to handle that without changing the naming scheme, too Aug 15 14:15:54 mac address matching is the easiest one Aug 15 14:16:00 but afaik theres no such thing? Aug 15 14:16:32 Writing a udev script to match and name interfaces by MAC is trivial and many distros shipped with that to stop interfaces shuffling Aug 15 14:16:34 thats how I do it in systemd, I leave them as ethX but use mac address matching in networkd Aug 15 14:17:01 well ideally it shouldn't matter what the names are Aug 15 14:17:07 As for dealing with it on OWRT that's probably another matter. Aug 15 14:17:19 (we're a corner case in ourselves) Aug 15 14:17:22 also the problem with renaming ethX to ethY (which is why systemd doesn't) is that you can end up with race conditions Aug 15 14:17:26 yes Aug 15 14:17:33 jwh: Which is why you use temporary names Aug 15 14:17:54 honestly one rename is bad enough when trying to figure it out from dmesg Aug 15 14:18:28 That's why I just identify ports by MAC Aug 15 14:18:38 yeah Aug 15 14:18:47 i just have a text file with mac addresses in now :D Aug 15 14:18:52 so I can figure out whats changed Aug 15 14:19:10 the best bit is some of my ports are reversed due to the way they designed the card Aug 15 14:19:17 :D Aug 15 14:20:38 Probably the fault of the IC designers, blame them Aug 15 14:20:50 They love to put things in illogical places for the rest of us Aug 15 14:21:54 I think in this case its that the way the mainboard is designed it inverts the expected card orientation Aug 15 14:22:03 as the physical ports are upside down :D Aug 15 14:22:24 they're 1U boxes, so Aug 15 14:23:05 but do see the same on boards with multiple built in nics Aug 15 14:25:05 mmm, netifd isn't going to be upset by something else creating and destroying interfaces is it? Aug 15 14:25:46 if they're not mentioned in the config it will just ignore them right? Aug 15 14:38:29 Rene__: which branch are you working on atm? Aug 15 14:39:09 master Aug 15 14:39:32 I mean in your repo, for the erx stuff Aug 15 14:39:55 theres two, erx sfp and one with gpio? Aug 15 14:40:13 or is that something different to the DSA support? Aug 15 14:40:51 https://github.com/vDorst/openwrt/tree/mt7621-ramips-dsa2 Aug 15 14:40:56 aah ok Aug 15 14:41:01 I got the right one then :D Aug 15 14:41:46 It is a bit messy. Aug 15 14:41:50 yeah Aug 15 14:41:59 ubnt kinda suck for hiding all that stuff in their own modules Aug 15 14:44:14 I have seen more SDK's all with similar code. Aug 15 14:44:28 yeah Aug 15 14:45:14 so does the sfp port on the erx actually share the single 1gbps uplink to the cpu? I thought they were doing some magic to use the 2nd link for the sfp Aug 15 14:46:43 mikrotik has a good picture of it. https://i.mt.lv/cdn/rb_files/RB760iGS-esw-180517144516.png Aug 15 14:47:16 aah is it the same config? Aug 15 14:47:32 that is, 2 lanes until sfp is plugged in, then one lane is used by the sfp? Aug 15 14:48:17 a lane is a ethernet port. CPU has 2 of them Aug 15 14:48:25 yeah Aug 15 14:48:34 so does the erx-sfp follow the same behaviour as that mikrotik? Aug 15 14:48:39 first is always connected to the switch port 6. Aug 15 14:48:46 just trying to work out if they did something dumb and its sharing a lane or something Aug 15 14:48:49 given its ubnt :P Aug 15 14:51:02 I checked out a completely new tree and did what I usually do. Lots of usb/hid things are now "y" in target/linux/x86/config-4.14 . Wonder if my problem stems from me doing make kernel_menuconfig and this makes a lot of these things go away. We'll see if the resulting image doesn't have the problems I typically get Aug 15 14:51:17 probably Aug 15 14:51:25 2nd gmac, switch por5 and external phy can connect to eachother. Aug 15 14:51:33 aaah Aug 15 14:51:35 cute Aug 15 14:52:11 I should probably get a 760igs, I think I saw a commit for support? Aug 15 14:52:25 It is not a bad idea but without documentation it is hard to see how to works Aug 15 14:53:00 yeah Aug 15 14:54:39 In my case I would like to connect the SFP port to 2nd gmac. So I can get 1gbit routing. Aug 15 14:55:02 yeah, thats what I would like also Aug 15 14:55:17 but I thought that was down to the device maker, not something that can be software configued Aug 15 14:55:20 configured* Aug 15 14:55:27 Rene__: i've imported the patches into my local tree and am cleaning them up atm Aug 15 14:56:52 jwh: lucky you can configure it. You can redirect P0 or P4 of the switch to P5 so that make it a WAN port Aug 15 14:57:05 So you get also 1gbit routing. Aug 15 14:57:29 blogic: great Aug 15 14:57:37 Rene__: thats pretty awesome actually Aug 15 14:58:28 Only downside is that SFP is not usable. But I think that is not an issue for most users. Aug 15 14:58:43 well its a start Aug 15 14:59:00 SFPs are what I'm most interested in though, plenty of devices that have copper ports Aug 15 15:00:13 Me too, I connect my fiber directly to my SFP. Aug 15 15:00:43 yeah Aug 15 15:01:17 hm, I was sure I saw support for the rb760igs Aug 15 15:01:20 maybe I'm imagining it Aug 15 15:03:35 oh, I'm thinking of the 750gr3 Aug 15 15:05:39 mamarley: my usb ethernet adapter show justs up as enp0s20u1u1, and moves based aroudn where the usb root hub is connected, so if it had a usb mode, it's not useing it on fedora at least. Aug 15 15:06:47 too bad it doesn't have wifi. So that I can replace all my 3 devices (fiber to rj45 convert, isp router and 2nd switch) with one device. Aug 15 15:06:56 heh Aug 15 15:08:29 like this one https://mikrotik.com/product/RB962UiGS-5HacT2HnT Aug 15 15:08:51 mmm Aug 15 15:09:11 I wanna get some of those mikrotiks but they have no console port, gonna make it tricky to see if it boots or not Aug 15 15:09:19 maybe they have uart headers though Aug 15 15:12:23 so I now verified again that on my i7-940 x58 mobo out of the box x86/64 doesn't come with the correct usb/hid drivers to make my usb keyboard work. There are also no ipk files in packages to install that I can imagine would make it work either Aug 15 15:12:31 so something is definitely missing Aug 15 15:13:15 SwedeMike: you mean x86 fails to probe usb keyboards ? Aug 15 15:14:51 blogic: it fails to include kmod-something that is needed for the usb/hid subsystem to work properly. Aug 15 15:15:16 it's either *hci or *hid* or combination Aug 15 15:15:28 from what i can tell usb hid isn't included by default Aug 15 15:16:12 I install hid and hid-generic now. But what worries me, is that before I would insert the usb keyboard and it detected the device, but hid driver was missing so still didn't work. This time it didn't even detect the device. Aug 15 15:16:28 the "usb device inserted .... " in dmesg is what I am referring to Aug 15 15:17:07 I just want to figure out what's missing without even going into make kernel_menuconfig and potentially adding yet another variable into this Aug 15 15:17:13 karlp: Interesting, on Ubuntu, my two appear as enx00051bb18260 and enx00051bb18d4b. Aug 15 15:18:07 i remapped my interfaces through udev rules in debian Aug 15 15:18:21 at least like that i have read- and typeable names again now :-/ Aug 15 15:19:52 SwedeMike: kmod-usb-uhci Aug 15 15:22:29 Monkeh: yeah, kmod-hid[-generic] wasn't enough. I'll try kmod-usb-uhci next. Aug 15 15:22:48 No point in HID drivers without a driver for the USB controller.. Aug 15 15:23:06 right. Aug 15 15:28:35 Monkeh: right, now it works. Aug 15 15:28:48 is there anything like iotop for openwrt? Aug 15 15:29:13 so is the answer then to require kmod-usb-uhci kmod-hid kmod-hid-generic then for x86-64? Aug 15 15:29:30 does kmod-usb-uhci work on "all" x86 machines, or do we need more? Aug 15 15:30:29 uhci tends to be intel and ohci tends to be amd Aug 15 15:31:38 because I really think we should include enough by default on x86 so that usb keyboards just work out of the box for most motherboards Aug 15 15:31:49 I see a xx% io in top, but not supppper useful Aug 15 15:32:05 least astonishment factor by including more than less.... Aug 15 15:37:15 not included right now that look potential are kmod-usb-ohci kmod-usb2 kmod-usb3 kmod-usb-hid whatever they do that uhci doesn't do Aug 15 15:37:57 what about kmod-usb-storage ? Aug 15 15:41:13 DonkeyHotei: i thought it was a matter of usb generations, not vendor implementations? Aug 15 15:41:32 that's for ehci and xhci Aug 15 15:41:34 it's both Aug 15 15:41:41 but usb 1.1 is ohci/uhci Aug 15 15:41:49 (like keyboards) Aug 15 15:42:01 Borromini: Yes, no, sort of, not at all, precisely. Aug 15 15:42:13 Wonderful thing about standards.. Aug 15 15:44:37 lovely Aug 15 15:44:42 I'll add the first three drivers to target.mk on a new checkout and see if it works "out of the box" without any changes Aug 15 15:44:56 so we'd need them all eh :) Aug 15 15:45:03 something must have been built in if you didn't need to add the others... Aug 15 15:51:04 karlp: yep, probably it added support for some other usb controller than the one I had, perhaps even some hid drivers. So it migh tbe that adding "kmod-usb-uhci kmod-hid kmod-hid-generic" is actually too much, but I'll see what happens Aug 15 15:51:56 for x86 I'd rather add too much when it comes to usb controllers and hid than less, so that people can interact with their device Aug 15 15:57:38 it certainly won't hurt. space is not really the issue on x86 Aug 15 15:58:14 right. Aug 15 15:59:55 HID stands for 'human interface device' afaik no? so i'd throw in kmod-usb-hid as well Aug 15 16:00:19 yeah, but if it wasn't required, it would have already been included, Aug 15 16:00:30 probably because someone rightuflly said, "you need this built in so youcan boot nicelyt" Aug 15 17:00:18 ok, so that was successful. So I suggest adding "kmod-usb-uhci kmod-hid kmod-hid-generic" to target.mk for x86/64 or for all x86. Aug 15 17:12:47 probably actually just needs CONFIG_USB_UHCI_HCD uncommented in target/linux/x86/config-4.14 ? Aug 15 17:12:59 it already has ehci, hid, and ohci there, but not uhci... Aug 15 17:14:16 karlp: what about 4.9 then? But sure, I can try this way of solving it as well Aug 15 17:14:29 sure, config-XX. Aug 15 17:14:40 I'll try it. Aug 15 17:15:45 I don't know if uhci and ochi conflict, you ðll want to use kernel_menuconfig or similar to get the "right" syntax in the config file, Aug 15 17:16:21 They don't conflict. In fact, many desktop-oriented distros have both compiled into the kernel. Aug 15 17:21:11 yes, I compiled in all of them before when I couldn't figure out exactly what was missing and that worked fine Aug 15 17:22:45 blogic: I think the ath10k-ct changes you made in your staging tree might be bugged. I get: install -m0644 /home/michael/Source/openwrt/uap-ac-pro/build_dir/target-mips_24kc_musl/ath10k-ct-firmware/QCA988X/hw2.0/board.bin /home/michael/Source/openwrt/uap-ac-pro/build_dir/target-mips_24kc_musl/ath10k-ct-firmware/ipkg-mips_24kc/ath10k-firmware-qca988x-ct-htt/lib/firmware/ath10k/QCA988X/hw2.0/ Aug 15 17:22:45 install: cannot stat '/home/michael/Source/openwrt/uap-ac-pro/build_dir/target-mips_24kc_musl/ath10k-ct-firmware/QCA988X/hw2.0/board.bin': No such file or directory Aug 15 17:26:11 jow: is it known/expected that ubus mangles big numbers in json? https://zerobin.net/?e645d982acfbc974#umLcW8/ZGqNiRBnTiR3BxXF7uWndtnwm1zK816CV/i4= Aug 15 17:26:21 I thought I'd seen a patch for that ages ago? Aug 15 18:00:09 karlp: your suggestion on CONFIG_USB_UHCI_HCD worked as well Aug 15 19:55:42 that should make sure it's built in, so works in failsafe and so on too iirc. Aug 15 20:13:02 mamarley: ok, i'll have a look. dont use -ct in my staging tree Aug 15 20:13:42 blogic: Thanks! (For the time being, I worked around it by manually copying the board.bin off the device and dropping it in the directory where it was expected.) Aug 15 20:14:13 blogic: Oh, wait, you're telling me not to use it? Why? Aug 15 20:17:23 because the tree is there to test mac80211 Aug 15 20:17:25 not CT Aug 15 20:17:29 well use it if you want Aug 15 20:17:41 but testing mac80211 might make more sense Aug 15 20:18:38 Sorry, I thought ath10k(-ct) was a mac80211 driver. I guess I was wrong. Anyway, the device also has an ath9k chip, which definitely uses mac80211. Aug 15 20:20:30 mamarley: there is a non-ct version of ath10k Aug 15 20:20:54 I know, but that one is about as stable as a house of cards in an EF-5 tornado. Aug 15 20:21:24 becaue we had a shitty random testing version from 1 year ago with a randomy shitty FW from QCA Aug 15 20:21:35 my patch pulls it up to 4.18-stable Aug 15 20:24:39 blogic: Isn't all firmware from QCA random and shitty? Aug 15 20:25:44 :D Aug 15 20:26:53 there are levels of shittiness Aug 15 20:51:18 sorry was in a call Aug 15 20:51:37 so for me using the FW inside stable witht he driver from stable actually yields a pretty good ath10k experience Aug 15 20:52:03 Last I used upstream firmware it was very explody Aug 15 20:52:15 ok Aug 15 20:53:19 blogic: I believe stintel still experiences issues with the mac80211 firmware Aug 15 20:53:30 crap Aug 15 20:54:13 blogic: We're working on that - .11w performance issues Aug 15 20:54:53 what is that about ? Aug 15 20:56:15 blogic: Radio firmware issues, not related to mac80211 afaik Aug 15 20:57:09 the performance issues are ath10k-ct related Aug 15 20:57:27 https://github.com/greearb/ath10k-ct/issues/31 Aug 15 20:57:49 Maybe. Aug 15 20:57:52 I was thinking about this: https://bugs.openwrt.org/index.php?do=details&task_id=333 Aug 15 21:01:49 blogic: yeah, stock ath10k is still shit Aug 15 21:02:06 stintel: ok Aug 15 21:02:15 blogic: testing -ct again now, but performance is much worse than stock Aug 15 21:02:18 we should consider making ath10k-ct the default Aug 15 21:02:39 if we can get the performance issues with -ct sorted, I would suggest to drop ath10k entirely Aug 15 21:02:40 Currently seeing painful performance issues with the 10.1 firmware, though Aug 15 21:02:42 it's useless Aug 15 21:02:48 stintel: Try with 10.1 upstream firmware. Aug 15 21:03:30 that's going to be for another day Aug 15 21:03:36 almost headed to bed Aug 15 21:04:03 good idea Aug 15 21:04:14 its been a long day and i did 3 hours of sports today Aug 15 21:04:28 crazy :P Aug 15 21:04:30 stintel: Have hardware on the way to do some more tests Aug 15 21:04:38 Monkeh: cool Aug 15 21:05:08 it's funny though, my Gemini PDA (MediaTek SoC running Android) was off, and didn't see FS#333 for days Aug 15 21:05:27 there are some clients that can really mess up ath10k Aug 15 21:05:34 Certain devices really trigger it Aug 15 21:05:39 My phone is good for that Aug 15 21:05:52 I have a nice combination of clients Aug 15 21:06:13 oneplus one, nexus 5x, gemini pda, intel 8265, xbox one x Aug 15 21:06:41 oh I forgot my huawei phone Aug 15 21:07:37 and I could add a oneplus 2, another huawei phone and another xbox one Aug 15 21:07:56 oh and my samsung tablet :D Aug 15 21:08:01 lol Aug 15 21:14:28 nbd, nice! ZBT WE3526 with default firewall+NAT gets 1 Gbit/s when offload is activated in 18.06.0 :) Aug 15 21:16:21 system type : MediaTek MT7621 ver:1 eco:3 Aug 15 21:16:22 machine : ZBT-WE3526 Aug 15 21:16:25 o_O Aug 15 21:16:38 ... Aug 15 21:16:38 processor : 3 Aug 15 21:16:38 cpu model : MIPS 1004Kc V2.15 Aug 15 21:17:02 hmmm openwrt wiki and hw vendor page says dual core, cpuinfo reports 4 cores Aug 15 21:17:14 2c/4t Aug 15 21:17:22 stintel, ah ok thx Aug 15 21:17:27 iirc Aug 15 21:18:18 stintel, indeed https://wikidevi.com/wiki/MediaTek_MT7621 Aug 15 21:18:26 https://wikidevi.com/wiki/MediaTek_MT7621 Aug 15 21:18:41 CPU architecture: MIPS 1004Kc @880MHz 2C/4T (Dual Core/Quad Thread) Aug 15 21:19:16 stintel, for the record pics case opened: https://pano.tetaneutral.net/data/20180815-ZBT-WE3526/ Aug 15 21:19:59 good to know that it performs so well Aug 15 21:20:49 maybe we should do a kickstarter for a device based on that, PoE-PD, console port, etc Aug 15 21:21:10 with no wifi Aug 15 21:21:13 ? Aug 15 21:21:25 mt7615 is still not supported Aug 15 21:21:29 DonkeyHotei, wifi seems to work (not tested for perf yet) Aug 15 21:21:43 let me check Aug 15 21:21:47 mt7612 is limited to two streams Aug 15 21:21:58 DonkeyHotei: well I would actually like both Aug 15 21:22:13 DonkeyHotei: Those aren't your only choices you know Aug 15 21:22:17 I'm running only APs combined with wired routers Aug 15 21:22:52 multiple APs with same SSID, so you can reboot one without clients losing connection, and my wired routers are redundantly configured Aug 15 21:23:23 besides mt7612, the choices for 11ac are mwlwifi, ath10k, brcmfmac, and that's about it iirc Aug 15 21:23:27 that reminds me I should finish part 3 of the blog post series about that redundant setup Aug 15 21:23:41 stintel: where's your blog? Aug 15 21:23:49 wigyori: https://stijn.tintel.eu/ Aug 15 21:23:56 should've guessed... Aug 15 21:23:56 :) Aug 15 21:24:05 thanks Aug 15 21:24:19 wigyori: welcome Aug 15 21:25:26 I'm going to add a reminder in my calendar to do part 3 the coming weekend Aug 15 21:26:44 and I did some fancy stuff with suricata on openwrt too Aug 15 21:26:49 I should force myself to blog more Aug 15 21:29:22 2.4 GHz (wget firewall+NAT) 2018-08-15 23:29:00 (10.8 MB/s) - ?/dev/null? saved [1073741824/1073741824] Aug 15 21:29:55 That.. is absolutely amazing. Aug 15 21:30:03 Every SD reader I've got apparently causes this BIOS to hard lock Aug 15 21:30:49 No.. it's the CARD Aug 15 21:30:51 What?! Aug 15 21:31:39 stintel I like your blog btw Aug 15 21:31:45 Tapper: thanks Aug 15 21:32:10 I will be bugging you about pt3 now tho! hahaha Aug 15 21:32:17 been thinking to migrate it to a static page generator or so Aug 15 21:32:20 coz Drupal ... Aug 15 21:32:24 you put it out there and now we all know! Aug 15 21:32:40 :P Aug 15 21:33:13 hmmm I don't see anymore the 5 GHz SSID, I was seeing it before Aug 15 21:33:40 Tapper: I really like the macvlan trick to make it all possible Aug 15 21:34:24 stintel: but PHP is so fancy. Aug 15 21:34:45 stintel: i'm very happy with pelican, but my needs are basic (volatilesystems.org ;-) ) Aug 15 21:34:57 Borromini: the problem is I cannot get MySQL to perform acceptably Aug 15 21:35:29 and I was thinking it was due to using btrfs, but after migrating to ext4 for the mysql data dir I still have horrible performance Aug 15 21:35:56 that sounds bad. i'm glad i have migrated to plain ntext being turned into html Aug 15 21:35:58 I should just order a new server with SSD and migrate everything there Aug 15 21:36:01 :P Aug 15 21:36:12 SSDs do make everythign better Aug 15 21:36:35 I have 15 VMs running off the same mdraid1 Aug 15 21:36:48 my laptop doesn't get nearly as hot since the ssd Aug 15 21:36:50 stintel: Moar RAM Aug 15 21:37:01 Monkeh: RAM is not the problem Aug 15 21:37:23 Monkeh: he already bought all the RAM on sale in Bulgaria ;) Aug 15 21:37:56 but I have 2x MySQL, zimbra (which runs its own MySQL), icinga2 exporting data to carbon, observium, 2x pdns auth nameservers also exporting data to carbon Aug 15 21:38:03 I just have too many writes Aug 15 21:38:31 stintel ssds for the win dude Aug 15 21:38:35 yep Aug 15 21:38:45 but can't have 2 SSDs added to my current server Aug 15 21:38:51 so have to migrate everything Aug 15 21:39:04 pane! Aug 15 21:39:16 Okay, something about the partition table in the x86 images really, really angers this machine. Aug 15 21:39:19 maybe later this year I will take a few weeks of holiday and get it all done Aug 15 21:39:28 Monkeh: what issues are you seeing? Aug 15 21:39:58 hmmm I see briefly wlan1 in "ip a" then it disappears Aug 15 21:40:03 Borromini: Uh, it locks the BIOS up. Aug 15 21:40:06 nothing in dmesg Aug 15 21:40:10 Why would you put APs in vlans? There is a dude on the forums asking about putting 4 APs in to 4 diffrent vlans. Aug 15 21:40:36 I just use dumb APs and don't see the need for vlans Aug 15 21:40:44 Tapper: guest network? Aug 15 21:40:58 or limit broadcast domain if you have many clients Aug 15 21:41:02 Monkeh: is that even with non-SD devices? Aug 15 21:41:15 OO for isolation? Aug 15 21:41:31 DonkeyHotei: It shouldn't really matter, but let's see Aug 15 21:41:37 * Monkeh upturns the tub Aug 15 21:42:23 Tapper: my guest network is in a VLAN too Aug 15 21:42:31 straight to the outside Aug 15 21:42:55 O rite I don't have a guest network. I just let all on my wifi Aug 15 21:43:10 Random USB stick does it too.. Aug 15 21:43:11 Bad opsec! I know. Aug 15 21:43:32 Tapper: well, I'm self employed and have client data on my network - it's protected but still Aug 15 21:43:53 My WAP is attached to a trunk interface, so all traffic to it is in one VLAN or another. Aug 15 21:43:58 stintel Yeah makes sense. Aug 15 21:44:08 Monkeh: both with the ext4 images and the squashfs? Aug 15 21:45:17 Tapper: all my friends and family continously moan because i outfit their wireless with guest networks :P Aug 15 21:45:19 DonkeyHotei: Well the partition table is identical.. Aug 15 21:45:55 Borromini lol Aug 15 21:46:01 also, usb sticks and usb sdreaders are all usb Aug 15 21:46:06 No, really? Aug 15 21:46:09 I never would've guessed Aug 15 21:46:15 Perhaps there's a reason I want to USB boot it Aug 15 21:46:38 perhaps that boot is what's broken Aug 15 21:46:53 I've USB booted this too many times to count Aug 15 21:47:07 with openwrt? Aug 15 21:47:09 Never been a problem before Aug 15 21:47:13 No.. Aug 15 21:47:19 i have booted openwrt from usb last week Aug 15 21:47:30 x86/64 squashfs Aug 15 21:49:09 It's long before it would boot and a manually created table doesn't break, so.. interesting. Aug 15 21:49:18 Borromini: haha :) Aug 15 21:49:52 the funniest thing is my guest network password, it contains "boobs" written in Dutch slang Aug 15 21:50:13 always funny when you have to give it to $gf Aug 15 21:50:18 All my passwords are XKCD-style passphrases. Aug 15 21:50:24 (WiFi passwords) Aug 15 21:50:39 openssl rand -base64 20 Aug 15 21:51:18 is my usual password generator Aug 15 21:51:44 but for wifi, especially guest network, it's something that is easier to remember Aug 15 21:57:07 Fascinating Aug 15 21:57:16 Created an identical table and it's fine Aug 15 22:02:00 stintel, DonkeyHotei hmm I can't seem to activate 5 GHz wifi on my ZBT WE3526 Aug 15 22:02:10 I can scan and it finds 5 GHz SSID Aug 15 22:02:21 but can't seem to make wlan1 appear Aug 15 22:03:55 "SSID: OpenWrt5 | Mode: Master Aug 15 22:03:55 Wireless is not associated" Aug 15 22:04:03 hmmm Aug 15 22:06:14 no BSSID displayed Aug 15 22:08:00 guerby: DFS? did you select a non-DFS channel? Aug 15 22:12:53 stintel, yes 5180 Aug 15 22:13:00 stintel, I'm flashing a snapshot Aug 15 22:14:13 * Borromini tips his hat Aug 15 22:14:15 night gents Aug 15 22:16:01 stintel, with the snapshot no pb, I see both SSID Aug 15 22:21:18 stintel, DonkeyHotei wget over 5 GHz wifi + NAT + firewall : Aug 15 22:21:27 2018-08-16 00:19:51 (38.8 MB/s) - ?/dev/null? saved [1073741824/1073741824] Aug 15 22:23:58 so there's something fixed in the snapshot vs 18.06.0 for this device Aug 15 22:24:02 gn Aug 15 22:31:01 snapshot for me on zbt we1326 is much better for me than a few weeks ago, haven't tried 18.06 Aug 15 22:43:27 mt7603 seesm to be much more stable now! Aug 15 22:43:49 though i think txpower wa lost somewhere, iw does not report it Aug 15 22:48:45 ugh Aug 15 22:48:49 I gave up on txpower :) Aug 15 23:01:44 Hmm. Aug 15 23:06:46 I was really hoping this would be easy. Aug 15 23:11:03 why is txpower so difficult, isnt it just some attribute from the driver? Aug 15 23:11:29 but really, its working now… Aug 15 23:23:16 I really need to get a SATA dock for this thing Aug 16 02:25:08 out of curiosity, any non x86 10gbit device? **** ENDING LOGGING AT Thu Aug 16 03:00:00 2018