**** BEGIN LOGGING AT Tue Mar 16 03:03:12 2021 Mar 16 03:03:50 why not? Mar 16 03:23:54 i had issues with ccache when I was buildig with CONFIG_ALL Mar 16 03:24:13 I would like to know which. I already fixed a bunch. Mar 16 03:24:40 In other news, AliExpress refunded my purchase of the totolink :( Mar 16 03:24:47 no ax for me Mar 16 03:25:45 was with the luci packages if I remember right Mar 16 03:27:41 Maybe I should look into the linksys e8450 instead... Mar 16 03:28:30 :(. more expensive Mar 16 03:29:37 Why did they cancel the order? Mar 16 03:30:36 No idea Mar 16 03:30:49 Just says closed Mar 16 03:31:38 Huh. AE doesn't usually stop and order. sell worthless fakes, yes, but not cancel Mar 16 03:31:42 maybe the shop got shut down Mar 16 03:31:53 and you got saved from being scammed Mar 16 03:35:35 I've been really interested in the e8450/ rt3200 for the last couple of days - but shipping- and import/ customs fees for post-Brexit Britain have gotten quite uncomfortable Mar 16 03:36:25 the price over all is still reasonable, but ~25 EUR shipping/ customs for a device that already sells with 21% VAT included just feels wrong Mar 16 03:37:56 pkgadd: more expensive than the totolink :( Mar 16 03:38:05 oof.. yeah.. They've still got teh 25% tarrifs on damn near all silicon parts here.. it's nuts Mar 16 03:38:18 I might still bite, as I have a need for a device faster (or at least as fast as-) ipq8065 (especially considering wired performance), but so far I haven't gotten through with it Mar 16 03:40:08 I've skipped looking at the totolink because of its mt7621 SOC, independent of the price (and I'm not toooo keen on importing from outside the EU either) - mt7622b is just so much more interesting Mar 16 03:42:01 hmmm so the Linksys E8450 from what I see sells for around ~200 Mar 16 03:42:14 the belkin rt3200 sells for around ~100 Mar 16 03:42:22 seems to be the same hardware though Mar 16 03:42:39 yep, https://www.amazon.co.uk/Belkin-Wireless-Dual-Band-Streaming-Parental/dp/B08L4PJKKB ~= 92 EUR, plus shipping and customs fees Mar 16 03:43:38 oh I see. it's for the european market Mar 16 03:44:10 found the reason order was canceled Mar 16 03:44:13 Seller failed to ship your order in time. A full refund will be issued automatically to your original payment method within 3 to 20 business days. Mar 16 03:44:25 Ah, see.. probably a scam store Mar 16 03:44:44 but I've often noticed low promotional prices on amazon.co.uk not seen on e.g. amazon.de, the Linksys ea6350v3 sold for ~35 GBP there as well Mar 16 03:44:51 Grommish: No it's the official one Mar 16 03:44:59 mangix: Interesting then Mar 16 03:45:10 maybe they ran out Mar 16 03:46:17 Now I have to convince myself to get the rt3200 Mar 16 03:48:26 I'd be fine with the price (~92 EUR), but not the shipping/ import duties on top Mar 16 03:51:11 the xiaomi ax3600 could be a nice alternative (ipq8071a), but sadly there seem to be quite some roadblocks to get ethernet/ wireless going with ipq807x Mar 16 03:53:37 right. I assume there are more mediatek devices that can be added Mar 16 03:53:45 I don't see anything on the mailing list Mar 16 03:53:54 btw. there seems to be one difference between e8450 and rt3200, the number of LEDs (the rt3200 only seems to have two, the e8450 also activity LEDs for the LAN ports on the bottom front) Mar 16 03:54:28 hard to tell from the vendor pictures though Mar 16 03:55:15 meh I hate LEDs Mar 16 03:55:45 nbd seems to have a patch for the D-Link DIR-X1860 Mar 16 03:55:48 * mangix looks Mar 16 03:56:08 furthermore the e8450 advertizes 2500 sqft coverage, the rt3200 2000 sqft - which doesn't need to mean anything Mar 16 03:56:21 (or could refer to FCC vs ETSI power levels) Mar 16 03:56:51 yeah...no. DIR-X1860 costs too much here Mar 16 03:56:54 and it's mt7621 Mar 16 03:58:42 btw. has anyone seen encrypted/ signed firmwares for mt7621 so far? or is it 'safe' to assume that mt7621 based devices can be ported/ flashed 'easily'? (I know that the TP-Link mt7622b+mt7915 devices seem to make it difficult that way) Mar 16 04:05:59 mangix: https://www.aliexpress.com/item/1005002132960438.html would be a cheaper option, but according to błogıc with secure boot (locked via arm trusted firmware); https://wikidevi.wi-cat.ru/TP-LINK_TL-XDR3230 Mar 16 04:07:26 which means a bootloader exploit would be needed Mar 16 04:07:30 ugh Mar 16 04:10:23 pkgadd: btw MIPS has no such thing as TrustZone Mar 16 04:11:47 thanks, that's what I thought - but wasn't 100% sure about (no experience with mt7621 for far) Mar 16 04:28:38 anyone have RTL8367rb experience? Mar 16 04:30:58 Or, just switch/swconfig knowledge? <_< Mar 16 04:32:31 I have devices using it (ath79/ tl-wr1043nd v1 and lantiq/ Arcadyan/Astoria Easybox 904xDSL (VGV952CJW33-E-IR)), but not really any developer side experience Mar 16 04:33:08 ugh Mar 16 04:33:17 how did these openssl failures creep in Mar 16 04:33:57 I've got the switch working, but only 4 of the 5 ports ;/ port0 on the switch refuses to work Mar 16 04:34:07 Build [#2](https://buildbot.openwrt.org/master/images/#builders/23/builds/2) of `mpc85xx/p2020` completed successfully. Mar 16 04:34:17 port1-4 work normally.. eth0-eth4 on the mt7621 work as well Mar 16 04:34:45 Buggin the hell outta me as to why it refuses to work Mar 16 04:35:24 Grommish: (boring, almost default) network config of the tl-wr1043nd v1, http://paste.debian.net/hidden/6e5e1b82/ Mar 16 04:36:51 pkgadd: This is what I've got so far.. https://gist.github.com/Grommish/c9ac775fc9041e676e092a3f5750ccf9 Mar 16 04:37:19 Port 6 is the DSA CPU port for the switch Mar 16 04:38:46 but I have no experience with switch configs under Openwrt.. Not used them before Mar 16 04:39:43 swconfig configuration is quite nice with luci Mar 16 04:40:36 VLAN 1: info: VLAN 1: Ports: '06', members=0041, untag=0041, fid=0 ports: 0 6 Mar 16 04:40:47 Wonder if that's why Mar 16 04:42:25 Well.. that broke it good Mar 16 04:42:28 Grommish: this would be a slightly more complex swconfig configuration, http://paste.debian.net/1189560/ Mar 16 04:43:33 It's wierd because eth5 (port0 on the switch) is competely dead.. doesn't even recognize a cable plugged in.. but the rest of the switch is afine Mar 16 04:46:17 just to visualize the config above in luci, https://imgur.com/a/U4G6HN3 Mar 16 04:53:34 pkgadd: *nod* I was going to try luci to see what it showed, but the port being dead just annoys me to no end hehe Mar 16 04:55:12 I need to figure out the dead port, then try to figure out the leds Mar 16 05:00:01 Grommish: you can disable ports by corresponding status definition in device tree node Mar 16 05:00:32 if it is actually considered dead Mar 16 05:00:40 damex: It works under EdgeOS Mar 16 05:00:48 It flashes alive on boot Mar 16 05:00:53 edgeos this. edgeos that Mar 16 05:01:15 The physical port is fine, whatever in openwrt is not bring it up while the rest of the switch ports are fine Mar 16 05:02:13 this is that wierd ass er10x with the external phy swith Mar 16 05:02:39 So you'd think it would ALL work, or not work at all ;/ Mar 16 05:06:56 damex: ;p and I"m trying to get away from EdgeOS Mar 16 05:07:19 Grommish: the constellation is so special or ubiquiti guys dodn't know any better Mar 16 05:08:50 so 5 port mtk switch connected via rgmii link to mt7621 mac ? Mar 16 05:09:24 5 port rtl8367rb connects via rgmii2(?) to the mt7621 Mar 16 05:10:26 Grommish: does rtl8367rb even have driver with DSA support? Mar 16 05:10:36 damex: Nope.. swconfig Mar 16 05:10:51 At least, not that I can find Mar 16 05:11:20 but like I said, rtl8367rb port 1-4 are fine.. port 0 just refuses to acknowledge it's even there Mar 16 05:11:27 eww, whole mt7621 is (mostly/or fully by now) a dsa target Mar 16 05:11:46 dsa is setup on port 6 for the mt7621 Mar 16 05:11:56 but it all funnels thru eth5 it seems Mar 16 05:12:00 i don't think you want have dsa+swconfig on single system Mar 16 05:13:37 mangix: Holy sh!t.. I just ran the ld vs ld.gold Mar 16 05:15:02 mangix: https://gist.github.com/Grommish/c3e4645a9d81045d3b381321a9c37fb3 Both builds from a rm -rf bin/ build_dir/ staging_dir tmp/ .ccache/ Mar 16 05:15:22 damex: I don't think I have to choice unless I want to lose half the ports on the device Mar 16 05:15:40 Grommish: that is...impressive Mar 16 05:16:12 I run make -j5 to build Mar 16 05:16:17 Grommish: https://forum.openwrt.org/t/ubiquiti-edgerouter-10x-er-10x/42980/41 so just 5th port does not work? Mar 16 05:16:55 damex: Right.. eth5 should be port0 on switch0. port1-port4 work Mar 16 05:18:09 Grommish: do you have dmesg? Mar 16 05:18:41 lipnitsk: if not, I can opkg install it.. I'm using defconfig, but adding the rtl8367rb drivers in the 5.10 kernel Mar 16 05:19:14 Grommish: pastebin your dts and dmesg output Mar 16 05:21:08 lipnitsk: One sec.. I've got to rebuild from testing Mar 16 05:21:54 lipnitsk: dts https://gist.github.com/Grommish/e595092b4702ff581045fb85eb6d4a8e Mar 16 05:23:40 Grommish: how did you get those numbers. Was it just a difference between the second commit? Mar 16 05:24:18 mangix: Right.. I have my standard kernel-5.10 branch.. that's the without.. the only different between the two are your two patches Mar 16 05:24:37 with the removal of the directories before each build Mar 16 05:25:07 and just a time make -j5 for each *shrug* seemed like the easiest way to A/B test Mar 16 05:26:09 Those numbers make no sense to me. Can you try just by applying both patches and reverting/keeping the last one? Mar 16 05:28:45 What do you mean? I rebased from master, removed bin/ build_dir/ staging_dir/ tmp/ .ccache/, built with time make -j5 Switched branches from master, cherry-picked your 2 commits, removed the same dirs and built again Mar 16 05:28:58 So the only difference is your 2 commits *shrug* Mar 16 05:30:04 but, I'll run it again later for ya if you'd like Mar 16 05:30:33 I need to get the toolchain for the er10x rebuilt (I built for the Itus device) so it's doing that at the moment Mar 16 05:32:41 I can only report the output of the time command *shrug* Mar 16 05:33:32 Grommish: time command is good enough Mar 16 05:34:10 but yeah, I Mar 16 05:34:18 err I'll rerun it later Mar 16 05:34:28 >KGB-0< https://tests.reproducible-builds.org/openwrt/openwrt_tegra.html has been updated. (0% images and 98.1% packages reproducible in our current test framework.) Mar 16 05:34:40 if I cherry pick and revert, the toolchain will be set to build with gold Mar 16 05:34:47 so I'd have to remove it anyway Mar 16 05:35:33 hmm funny Mar 16 05:35:54 -mips64r2 -mtune=mips64r2 -mabi=64 does not support PAUSE but -mips32r2 -mtune=mips32r2 -mabi=32 does Mar 16 05:35:58 figure that one out Mar 16 05:36:15 We dealt with that before Mar 16 05:36:23 Something with MIPS16 being defaulted to on Mar 16 05:36:30 which predated PAUSE opcode Mar 16 05:36:43 might start there Mar 16 05:37:00 nope. has nothing to do with mips16 Mar 16 05:37:16 gotcha.. that' the only time I've seen an issue with PAUSE Mar 16 05:37:18 those three parameters are the only difference Mar 16 05:37:20 gotta love it Mar 16 05:37:52 If you can search the irc logs, it was.. a month ago? maybe? Mar 16 05:38:25 I remember the discussion. I assumed it was just an octeon problem. looks like no :) Mar 16 05:38:27 Grommish: rtl8367b only has port5, right? the other 5 ports (eth0-eth4) come up ok? Mar 16 05:38:59 lipnitsk: eth0-4 are fine.. eth6-9 (switch0:1-4) are fine.. eth5 which should be switch0:0 is dead Mar 16 05:39:08 eth0-4 are on th mt7621 Mar 16 05:39:18 5-9 on the rtl Mar 16 05:39:27 Grommish: okay, i see. Mar 16 05:39:40 Grommish: 6-9 come up but not 5? Mar 16 05:39:45 Right Mar 16 05:40:03 eth5 doesn't even acknowledge a cable plugged in Mar 16 05:40:27 how come eth5 is so special? And where does eth0 map if eth5 is switch0:0? Mar 16 05:40:29 * mangix needs to learn vim badly Mar 16 05:40:42 mangix: :help cmd Mar 16 05:40:43 eth5 is the external phy Mar 16 05:41:22 lipnitsk: I know it at a basic level, but that's not good enough Mar 16 05:41:26 lipnitsk: https://github.com/torvalds/linux/blob/master/Documentation/devicetree/bindings/net/dsa/mt7530.txt#L43 and https://forum.openwrt.org/t/ubiquiti-edgerouter-10x-er-10x/42980/47 if you're interested Mar 16 05:41:43 I recently figured out that dd is not delete but cut... Mar 16 05:43:29 lipnitsk: didn't know there was such a thing as 1.2gb ethernet... Mar 16 05:43:39 internally Mar 16 05:43:46 just for the CPU port Mar 16 05:46:34 mangix: https://lore.kernel.org/netdev/20190616182010.18778-1-opensource@vdorst.com/ Mar 16 05:52:07 mangix: yeah but, effectively dd is delete. unless you care about what's in your default register or whatever Mar 16 05:52:08 interesting to say the least Mar 16 05:52:25 lipnitsk: dd + p = cut + paste Mar 16 05:52:31 ya Mar 16 05:52:53 i was googling how to paste in vim Mar 16 05:53:13 p - paste after cursor, P - paste before cursor Mar 16 05:53:42 you can also paste with mouse in insert mode ;) Mar 16 05:54:33 one cool thing I found out is that this vim setup (well nvim actually) automatically wraps commit messages to 72 characters per line Mar 16 05:54:37 can't say I'm a super expert, but been using it for a number of years and what I know is enough for me.. Mar 16 05:55:04 quite convenient for OpenWrt commits Mar 16 05:55:08 yeah, there are some cool plugins that do all sorts of stuff too Mar 16 05:55:38 I saw this guy's nvim config and replicated to force myself to use nvim. didn't last long. Mar 16 05:55:52 but now I need to since nano doesn't work right on this terminal Mar 16 05:58:58 fascinating. just saw an ebay listing for a broadcome router. wonder what that is. Mar 16 06:13:34 Grommish: maybe eth5 is an internal port similar to 'dsa' and there is a conflict. have you tried using a different label for port@5? Mar 16 06:14:23 Grommish: ie both mt7530 and rtl8367 drivers are trying to bring it up? Mar 16 06:14:35 lipnitsk: Not yet.. this compile is about finished and I can change things quick and easy then. i'm flashing from TFTP to sdram so its quick Mar 16 06:15:14 Grommish: I'm going to bed now, sorry. but it seems you are close. probably need to make it work with DSA though anyway.. but it's a start. Mar 16 06:15:33 I can't find a dsa driver for it .. meh Mar 16 06:15:51 but yeah, it'll get fixed eventually :) Gnight! Mar 16 06:38:17 Hello, I want to add a procd trigger to watch some files, but not the uci files. How to? Mar 16 06:39:17 It seems that "procd_add_reload_trigger" does not take effect. Mar 16 06:40:20 https://openwrt.org/docs/guide-developer/procd-init-scripts#specifying_triggers Mar 16 07:34:17 lipnitsk: For tomorrow - https://gist.github.com/Grommish/8fef54a79f6dc6e07d8dff3bd495ec20 Mar 16 07:43:12 Build [#3](https://buildbot.openwrt.org/master/images/#builders/10/builds/3) of `bcm63xx/smp` failed. Mar 16 08:13:37 'mornin! Mar 16 08:14:01 rsalvaterra1: Hey hey :) Mar 16 08:15:42 mangix: Don't drop those. I only added patches to fix things, not to backport the latest device tree changes, although I have them in my tree. :) Mar 16 08:16:36 mangix: https://github.com/rsalvaterra/openwrt/commit/fb214850fba0c9cefbe1840ab5b8b16bb2d37b21 Mar 16 08:19:23 Grommish: Hey, mate! Looks like you're making progress, heh? :) Mar 16 08:20:28 port0 on the switch still is dead for whatever reason :) It's super odd :) Any idea how I can set kernel CONFIG flags for just the er10x without changing config-5.10? Mar 16 08:23:13 Grommish: You can't, as far as I know… :/ Mar 16 08:23:46 Hmm.. I've got 5 changes to the kernel I need to set that won't be used in any other ramips device i wouldn't think Mar 16 08:24:33 Grommish: Which additional symbols do you need? Mar 16 08:25:01 CONFIG_MDIO_DEVRES, CONFIG_RTL8366_SMI, CONFIG_SWCONFIG, CONFIG_RTL8367B_PHY, CONFIG_RTL8367_PHY Mar 16 08:25:30 Ouch… a swconfig driver. Mar 16 08:25:42 no dsa that I can find for the rtl8367rb Mar 16 08:26:45 But yeah, the entire rtl8367 is piped thru eth5, which is why its odd only one port on that switch doesn't work, but the other 4 do *shrug* Mar 16 08:27:38 Probably we'll just have to bite the bullet and enable the rtl8367rb on ramips/mt7621, for the time being… Mar 16 08:28:15 I already have to hacky out target/linux/generic/files/drivers/net/phy.. this is getting ugly hehe Mar 16 08:29:34 rsalvaterra: https://github.com/Itus-Shield/openwrt/commit/789faf443404d89d211ebcab5a2da2459d6526cc Mar 16 08:30:38 That gives me 9 of the 10 ports *shrug* Mar 16 08:32:06 wonder if I should change 0:lan to 1:lan and shift everything down Mar 16 08:32:16 on the switch define Mar 16 08:32:30 Grommish: Could be… port numbering is weird. Mar 16 08:33:16 Grommish: perhaps read a bit into the easybox 904 xDSL porting, which also uses two switches (lantiq and rtl8367b), with weird internal VLAN connections Mar 16 08:33:48 pkgadd: I'll take a look.. the switch works, but just the one port refuses to work Mar 16 08:34:08 it's like it disabled, doesn't even recognize a cable plugged into it Mar 16 08:37:56 rsalvaterra: Nope.. No change.. meh Mar 16 08:38:03 Grommish: FYI, I just added the module for CONFIG_MDIO_DEVRES, kmod-mdio-devres Mar 16 08:38:54 guidosarducci: Oo.. Thanks! :) Mar 16 08:40:26 rsalvaterra, 0/ Mar 16 08:41:10 nitroshift: Hey, mate! How's it going? :) Mar 16 08:42:09 rsalvaterra, did you forget about HAVE_ARM_ARCH_TIMER in 5.10 config? Mar 16 08:42:21 it doesn't seem to be enabled Mar 16 08:42:37 *enabled --> defined Mar 16 08:42:46 * nitroshift needs more coffee Mar 16 08:42:55 Grommish: yeah, I would normally expect the switch not to work (respectively not allowing access to the SOCs sshd without proper VLAN setups) Mar 16 08:43:03 nitroshift: Nope. The Armada 385 is Cortex-A9, doesn't implement the ARM architectural (per-CPU) timer. Mar 16 08:43:17 Only the ARM global timer. Mar 16 08:43:23 ah i see Mar 16 08:44:14 it should be defined as not set nonetheless, shouldn't it? Mar 16 08:44:38 The architectural timer is implemented starting in Cortex-A15. Mar 16 08:45:14 rsalvaterra, i'm doing a build from master right now and it stopped asking about that Mar 16 08:45:16 nitroshift: I think that's already defined in the generic config, isn't it? Mar 16 08:47:38 Oh, wait, it probably doesn't even need to be specified, if it's one of the filtered symbols. Mar 16 08:49:52 nitroshift: Right, it's a CONFIG_HAVE_*, it doesn't have to be specified. https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=target/linux/generic/config-filter;h=719487be5fc73a448a734bca0eb9f8a8fc3a0b70;hb=HEAD Mar 16 08:50:41 rsalvaterra, strange Mar 16 08:51:10 must be my local clone.... Mar 16 08:55:32 git log Mar 16 08:55:40 Ouch. Wrong terminal, sorry. Mar 16 08:56:01 rsalvaterra: you say that like not everyone does it :) Mar 16 08:57:09 :P Mar 16 10:03:48 OMG, I just noticed my jffs2 zstd implementation has two perfectly avoidable *backwards* goto. The perils of copy-paste… Mar 16 10:04:47 zx2c4: congrats on the FreeBSD port. sucks to get that kind of flak from netgate for actually helping *them* out >_> Mar 16 10:06:16 Borromini: Come on, it's not like Netgate is making money out of it… *rolls eyes* Mar 16 10:07:46 :P Mar 16 10:08:28 seems they kind of stonewalled him and now they're complaining about fallout and bad press for them Mar 16 10:08:40 https://lists.zx2c4.com/pipermail/wireguard/2021-March/006499.html Mar 16 10:14:48 nbd: I've started looking at luci/dsa again to get it in shape for the release Mar 16 10:15:00 nbd: I encountered two issues I consider blockers Mar 16 10:15:47 nbd: 1) when we have conflicting settings (config interface lan + option type bridge and config device; option type bridge; option name br-lan) - which one is preferred? Currently it seems that the behaviour is undefined Mar 16 10:17:09 nbd: 2) what is the proper way to attach wifi-iface specific wireless netdevs as "config device; option type bridge" ports? It seems ucrrently the only ways are either a) "list ifname wlanX" in "config device" which is highly discourages due to the dynamic nature of wireless netdev naming Mar 16 10:17:46 karlp: thanks for updating libwebsockets, turns out that did not fix my ttyd ssl issues but still appreciated Mar 16 10:17:49 or b) utilizing "option network" in "config wifi-iface", but that requires an intermediate "config interface" referencing the target bridge Mar 16 10:19:24 you proposed an "option network_vlan" a while back (http://lists.openwrt.org/pipermail/openwrt-devel/2020-July/030397.html) but it seems it didn't materialize in netifd yet? Mar 16 10:20:46 grift: no problems, it was well overdue, and it's much bigger than it used to be, but hey, "latest is greatest" Mar 16 10:20:56 and at least this version works on both ttyd and mosquitto. Mar 16 10:21:05 I can work around 1) in LuCI by forcing a migration from "config interface lan; option type bridge; option ifname ..." to "config interface lan; option ifname br-lan" plus "config device; option name br-lan; option type bridge; option ifname ..." Mar 16 11:59:44 and.... what a surprrise, wolfssl + libwebsockets doesn't actually work.... Mar 16 12:53:14 jow: with dsa, we'd have a switch0 bridge with two vlans. we should name the vlan devs something like sw0-lan, sw0-wan and use them as ifname in the config interface lan/wan section Mar 16 12:53:28 lan/wan interfaces will then no longer be bridges, but you can still use option network lan in wifi-iface sections Mar 16 12:53:39 there is no need to create device sections for wifi ifaces Mar 16 12:53:57 netifd will figure out that the vlan device sits on top of a vlan-filtering bridge Mar 16 12:54:06 and add the ports dynamically Mar 16 12:54:23 I know, currently writign a summary mail with the problems I see Mar 16 12:54:44 a lot can be solved by shipping a different default config that makes the switch0 bridge declaration explicit Mar 16 12:54:50 we should phase out implicit bridging entirely Mar 16 12:54:56 i agree Mar 16 12:55:20 we should keep support for it in netifd, but once configs have been migrated, we can phase out simple ui support for it Mar 16 12:56:29 one piece of the puzzle I miss is the ability to attach wifi-ifaces to config device bridges without requiring a dummy config interface Mar 16 12:57:37 right now the netdefs are bridged into (wireless.wifi-iface.network -> network.interface.ifname -> network.device.name) Mar 16 12:57:44 *wireless netdevs Mar 16 12:57:51 for normal configurations you have a interface set up on each relevant vlan, right? Mar 16 12:59:01 Not necessarily Mar 16 12:59:18 I mean sure, if with "normal" you mean lan + wan Mar 16 12:59:25 and nothing else Mar 16 12:59:34 if you don't, you could still set up a proto=none interface for the relevant vlan Mar 16 12:59:54 and use that in option network in wifi-iface Mar 16 13:00:02 I know, and I don't like it Mar 16 13:00:04 so you wouldn't refer to dummy interfaces from bridge port listings Mar 16 13:01:24 requiring dummy config interfaces w/ proto none made a little bit more sense when we did implicit bridging at the same time, so that these actually declared the target bridge Mar 16 13:01:46 now with bridge devices declared as config device, they're just redundant config churn Mar 16 13:02:37 I'd rather wish for an wireless.wifi-iface.bridge which simply refers to a netdev (e.g. br-lan) Mar 16 13:03:02 no implicit behaviour choices based on type of target interface (and type of device referenced by target interface) Mar 16 13:04:40 so you'd prefer to specify bridge and vlan settings explicitly as two options? Mar 16 13:05:34 you mean bridge membership or layer 3 settings? VLAN settings are separate in either case Mar 16 13:05:50 bridge port vlan membership Mar 16 13:05:50 either option network + option network_vlan to select vlan within target bridge Mar 16 13:05:56 brb Mar 16 13:06:07 yes, I'd prefer bridge port membership to be explicit Mar 16 13:06:36 option network is too overloaded with different configuration semantics Mar 16 13:09:04 if the target config interface is unbridged (option type bridge removed or ifname changed to something not being a bridge), I'd rather see the wireless netdevs become orphaned than suddenly inheriting layer 3 protcol settings in an undefined manner Mar 16 13:09:48 so that you end up with two interfaces running a dhcp client, or worse having the same static ip settings or just one of the former members Mar 16 13:09:58 ... having them Mar 16 13:22:23 jow: from a ui consistency point of view, i actually strongly prefer the option network variant Mar 16 13:22:36 because it means that wifi interface network settings can be assigned with one single drop down Mar 16 13:22:44 from being the one implementing the ui I can assure you that there is no consistency at all Mar 16 13:22:51 the code has grown extremely complex Mar 16 13:23:26 why not just allow selecting any interface as network for the wifi iface? Mar 16 13:23:51 because in my tests it results in undefined behaviour Mar 16 13:24:03 which tests? Mar 16 13:24:04 multiple interfaces inheriting the same layer 3 settings for example Mar 16 13:24:37 also keep in mind that option network is actually a list Mar 16 13:24:48 you can attach a wifi-iface to multiple networks Mar 16 13:25:01 got knows what happens if one is a bridge or happens to reference a bridge and the other is not Mar 16 13:25:48 what was the use case for supporting network as a list? Mar 16 13:27:07 maybe the first step in reducing complexity would be to get rid of that Mar 16 13:28:20 wireless wan Mar 16 13:28:25 having both wan and wan6 Mar 16 13:30:43 one way to deal with that would be to add a way to assign wan6 to wan without making it depend on wan being up Mar 16 13:30:58 then network could go back to being a single option Mar 16 13:31:02 and we'd get rid of some corner cases that way Mar 16 13:31:36 and we'd get rid of some duplication of ifnames in /etc/config/network while we're at it Mar 16 13:31:42 specify the wan device only in the wan section Mar 16 13:32:06 there is no duplication of ifnames Mar 16 13:32:23 typically in the non-bridge case, the target interface has no option ifname Mar 16 13:32:34 i mean for the non-wireless wan case Mar 16 13:32:47 where we typically throw the same ifname in both wan and wan6 Mar 16 13:33:17 usually we throw @wan in wan6 Mar 16 13:33:28 i think we got rid of that Mar 16 13:33:45 because that would mean with dhcp, ipv6 connectivity depends on ipv4 connectivity Mar 16 13:34:22 if wan6 would use @wan, there would be no need to specify both wan and wan6 as wifi-iface network Mar 16 13:34:37 correct Mar 16 13:34:56 so all we need is a way to make wan6 inherit the l2 device of wan Mar 16 13:35:12 I just tried just setting "option network wan" in a wifi-iface btw, didn't work Mar 16 13:35:34 with @wan in wan6 ifname? Mar 16 13:35:39 or what kind ofconfig? Mar 16 13:35:45 wan kept its ip settings, wlan0-1 ended up being unconfigured (which is fine, better than both inheriting settings or even worse, randomly either wlan or ethernet getting them) Mar 16 13:35:56 ah Mar 16 13:36:08 but it also means that it is not consistent from an ui perspective Mar 16 13:36:39 you still need to evaluate the target network, not offer interfaces which already have an ifname set (or if they have an ifname set, check if its a bridge or not, and only offer it if it is a bridge etc.) Mar 16 13:37:06 or let the user configure it, but report an error when attempting to set it up Mar 16 13:37:14 an error which clearly indicates what's wrong Mar 16 13:37:15 so in short, the same magic wrt. option network that happens in netifd, needs to be replicated in the ui, only with less state available Mar 16 13:37:49 we could make netifd report an error for a wifi-iface that couldn't be assigned to a network Mar 16 13:37:55 and then the ui could easily tell the user what's wrong Mar 16 13:38:53 yes, would be an improvement Mar 16 13:39:27 next problem is config interface xxx + option type bridge vs. config device + option type bridge + option name br-xxx clashes Mar 16 13:39:34 also something the ui needs to handle atm Mar 16 13:39:53 which becomes a non-issue once I remove support for implicit bridging Mar 16 13:40:07 but right now it is an issue, not sure how netifd handles it Mar 16 13:40:36 ideally the config device should take precedence since it offers more bridge specific settings afair Mar 16 13:45:17 i think it merges the device settings currently Mar 16 13:45:32 let me check Mar 16 13:46:11 no, wait. it overwrites the config blob Mar 16 13:46:26 since device gets parsed first, and the interface can overwrite it Mar 16 13:46:49 i guess i could change the code to simply pick the device instead Mar 16 13:46:57 nbd: If possible, when you have time, would you please take a look at why mt76 breaks with mangix's musl 1.2.2 patch series? It builds, loads and runs, but something in broken in the EEPROM reading. I find it extremely odd, since musl is userspace and these are kernel modules… :/ Mar 16 13:47:24 rsalvaterra1: what device? Mar 16 13:47:50 nbd: All of mt76. In my specific case, MT7603 and MT7615. Mar 16 13:48:19 and what do you mean by "something in broken in the EEPROM reading"? Mar 16 13:49:16 nbd: Simple: on my Redmi AC2100, when I build OpenWrt with musl 1.2.2, mt76 fails to read the correct MAC addresses and caldata. Mar 16 13:50:16 The system works, but the signal strength is extremely low and the Wi-Fi MAC addresses are random. Mar 16 13:52:17 wow, that's really odd Mar 16 13:52:47 It's particularly musl causing this… :P Mar 16 13:53:21 But mangix has more details. Something related to stddef and endianness. Mar 16 13:53:42 possibly an issue with the caldata scripts? maybe something goes wrong in printf or other base tools Mar 16 13:54:15 zorun: No scripts involved. The driver reads the data directly. Mar 16 13:54:42 zorun: It's not the same as ath9k or ath10k. Mar 16 13:54:46 ah ok, it's different then Mar 16 13:55:40 zorun: eeprom location is set in the dts for mt7621 Mar 16 13:55:51 unless i'm speaking out of turn Mar 16 13:57:02 Borromini: You're not, it's exactly like that. Mar 16 13:58:17 rsalvaterra: is this a nand device? Mar 16 13:58:40 Borromini: Yes, it is. 128 MiB. Mar 16 13:59:28 but it works when you roll back the musl bump right Mar 16 14:00:34 Borromini: Yes, it works with the current musl. Weird as hell. Mar 16 14:00:57 Borromini: Here's the relevant device tree: https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=target/linux/ramips/dts/mt7621_xiaomi_router-ac2100.dtsi;h=7e6b3afcdf008dc6506bd09b207ec64770a59bbf;hb=HEAD Mar 16 14:02:10 this might be silly, but hexdumping the factory partition returns the same contents before and after? you probably already checked that Mar 16 14:02:41 nbd: last question; for wifi-iface netdevs, the bridge vlan membership is inherited from the target interface's ifname, right? So if there's a config interface vlan200; option ifname switch0.200 and a wifi-iface with option network vlan200, It'll end up as port in the bridge vlan with vid 200? Mar 16 14:03:54 Borromini: Yes, I hexedumped to compare. It's untouched by OpenWrt (and it better be :P). Mar 16 14:04:00 *hexdumped Mar 16 14:05:16 grift: are you https://github.com/openwrt/packages/issues/14577#issuecomment-800265515 ? (doverride on github?) Mar 16 14:05:29 sounds like your ttyd ssl issue should be fixed? Mar 16 14:05:38 yes thats me Mar 16 14:06:21 took the time to dig a little deeper after it became clear that libwebsockets had nothing to do with it Mar 16 14:10:29 jow: yes Mar 16 14:10:54 karlp: and yes ttyd finally now works with acme/letsencrypt Mar 16 14:20:14 rsalvaterra: nbd: kernel modules need -most since passed with musl 1.2.0 Mar 16 14:20:28 otherwise endianness issues occur Mar 16 14:20:50 *-nostdinc Mar 16 14:21:15 mangix: Thanks for explaining it better than me. :) Mar 16 14:22:04 This big was always there but musl 1.2 exposed it Mar 16 14:22:15 *bug Mar 16 14:27:22 Build [#3](https://buildbot.openwrt.org/master/images/#builders/14/builds/3) of `bcm63xx/generic` failed. Mar 16 14:32:02 mangix: shouldn't that be up to the kernel build system to pass that flag? Mar 16 14:32:09 is there an upstream commit that fixes it? Mar 16 16:06:05 Hi all, if I build the kernel using make target/linux/install does that automatically update the firmware image as well? Or do I have to run another command to build the firmware image? Mar 16 16:09:44 Build [#4](https://buildbot.openwrt.org/openwrt-21.02/images/#builders/61/builds/4) of `ramips/mt7621` completed successfully. Mar 16 16:52:00 Something's fishy with pinmux on ubnt XM hardware… Mar 16 17:50:02 blocktrron: do you think https://github.com/openwrt/openwrt/pull/3960 aged enough to merge it into 21.02? Mar 16 17:52:19 works fine here :) Mar 16 18:45:38 Grommish: RE: https://github.com/openwrt/openwrt/commit/7379f8bd: what we really need is to teach people how to use a subset of buildbot. Mar 16 18:46:47 I spend like ... no time actually articulating builds these days. It's just commit/push, add diffconfig + files, then `make output-${remote}_${revision}` and wait Mar 16 19:29:19 Grommish: I updated https://git.laboratoryb.org/hurricos/docker-builder.git with much better documentatoin Mar 16 19:34:00 i am toying with an idea to hold recurring test day's, i shamelessly copied that idea from fedora test day. it would look something like this (rough sketch) http://sprunge.us/s1NYlF Mar 16 19:36:55 now sure i dont expect much participation in the first 3 or 4 test day's but it might pick up as people get used to the announcements posted in forums, mailing lists and other mediums Mar 16 19:43:59 ill refine the idea but i can sense the excitement for the prospect, thats a good sign and inspirational Mar 16 19:44:38 inspiring* Mar 16 20:14:34 hurricos: Half the time a build fails it's a PITA to figure out why the build-bot died :p I honestly find it's easier to build locally for anything I need, but maybe it's because I never learned to use Git actions to build it for me.. Mar 16 20:16:04 I just do a make -j5 || make -j1 V=sc and come back in 15 minutes to see what happened Mar 16 20:16:45 Grommish: isn't it usually searching the build log for "Error" or similar? Mar 16 20:17:14 lipnitsk: or "***", but "error" can occur in many places that arent fatal Mar 16 20:27:43 nbd: the Makefile overrides all NOSTDINC_FLAGS Mar 16 20:28:20 It should be adding to them Mar 16 20:37:23 Grommish: turn onbuild logs and just do "grep -r '\*\*\*' logs" instead ? Mar 16 20:57:46 karlp: Probably, but the original conversation revovled about ImageBuilder and making it too easy for people who don't want to get hands dirty making silly demands Mar 16 20:58:36 karlp: It's why I just OR-gate the -j1 V=sc if it fails, it'll tell me why pretty easily Mar 16 21:01:20 mangix: https://patchwork.ozlabs.org/project/openwrt/patch/20210308231452.1585178-1-mail@aparcar.org/ Mar 16 21:01:24 can you please test that? Mar 16 21:38:27 Grommish: fair shake. My strat was to strip out the base imgae of the buildbot and script / layer away the process of cloning down everything from scratch ... Mar 16 21:39:05 so it gives me a really really clean environment with exactly whatever crappy tree I want Mar 16 21:39:47 and you can use the debug- rather than output- target from my docker-builder to persist the docker image after it dies to see what messed up Mar 16 21:39:55 or rather it drops you to a shell. Mar 16 23:09:39 Grommish: yeah, but it takes way longer? Mar 16 23:16:11 aparcar[m]: bookmarked Mar 16 23:18:10 mangix: gracias Mar 16 23:23:23 Is it normal to see a one-armed router's switch configured with a single untagged VLAN during boot? My laptop is connected to an TL-WDR3600 and is receiving the IP address from the DHCP server upstream of the WDR3600(!!). Mar 16 23:26:29 I mean, it's not completely surprising, but still… Mar 16 23:31:15 rsalvaterra: did you try getting musl 1.2 to work with mt76? Mar 16 23:34:18 mangix: I did, but no cigar… that's why I asked for nbd's help, I was hoping it would be obvious to him, but he was as surprised as I was. Mar 16 23:37:21 rsalvaterra: so why do we need GENERIC_CLOCKEVENTS_BROADCAST if we are not using cpuidle? Mar 16 23:41:10 lipnitsk: We don't strictly need it, but it's a form of documentation. It means we have a global timer (GIC) capable of waking up the cores if they go to sleep. Of course, we have no cpuidle driver for the 1004Kc, and I don't even know if it makes sense to write one (haven't dived in the datasheet), but at least we have that information. Mar 16 23:42:45 At least this is what I understood from reading the code and the documentation I found about GENERIC_CLOCKEVENTS_BROADCAST. If I'm lying through my teeth, someone please yell. Mar 16 23:43:09 rsalvaterra: then should it be a weak dependency, 'imply'? Or better yet, an actual documentation entry? :) How much extra code are we talking? Mar 16 23:43:26 rsalvaterra: just seems weird to compile in code that is unused Mar 16 23:44:23 rsalvaterra: the patch does need to be split up regardless though, so I hope your change gets merged. we can optimize the code size as a future activity Mar 16 23:44:24 lipnitsk: It's not that much code: https://elixir.bootlin.com/linux/v5.10.23/A/ident/CONFIG_GENERIC_CLOCKEVENTS_BROADCAST Mar 16 23:46:10 Actually, I think I found something… Mar 16 23:46:58 The 1004Kc is a CPS (Coherent Processor System) arch. Mar 16 23:52:39 so can we use https://github.com/torvalds/linux/blob/fcadab740480e0e0e9fa9bd272acd409884d431a/drivers/cpuidle/cpuidle-cps.c ? Mar 16 23:54:00 in other news, mt7621 NAND driver is something that should be upstreamed... And there have been attempts, but not too successful so far. Mar 16 23:54:09 1. https://patchwork.ozlabs.org/project/linux-mtd/patch/1585678079-5999-1-git-send-email-weijie.gao@mediatek.com/ - OpenWrt uses this one Mar 16 23:54:15 2. https://lore.kernel.org/lkml/20191107073521.11413-1-gerg@kernel.org/ Mar 16 23:54:22 lipnitsk: For some reason I wasn't able to select any MIPS cpuidle driver… I need to see what's missing. Mar 16 23:54:45 3. https://github.com/torvalds/linux/blob/master/drivers/mtd/nand/raw/mtk_nand.c - similar, but not quite the same driver? Mar 16 23:55:28 rsalvaterra: add --nostdinc to NOSTDINC_FLAGS Mar 16 23:55:37 sorry -nostdinc Mar 16 23:55:38 * lipnitsk < https://matrix.org/_matrix/media/r0/download/matrix.org/YgisXhxFptKQtQMqeVgNTBok/message.txt > Mar 16 23:56:12 lipnitsk: it never will Mar 16 23:56:54 the issue is the driver is too similar to the already existing one Mar 16 23:57:10 mangix: Ok, I'll try a new build with -nostdinc tomorrow. You mean here, right? https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=package/kernel/mt76/Makefile;h=b60afb355d2ac7ed221a7603e1ca6fc2205a3bcf;hb=HEAD#l321 Mar 16 23:57:34 mangix: well right, maybe extending the existing won't be that hard? Mar 16 23:57:36 rsalvaterra: no 2 lines down Mar 16 23:57:55 lipnitsk: close to impossible without breaking the hardware already using it Mar 16 23:58:13 mangix: has someone tried? Mar 16 23:58:27 let me find some logs for you Mar 16 23:58:52 mangix: After modules? Mar 16 23:59:08 lipnitsk: Ok, I was able to enable the driver! Mar 16 23:59:37 rsalvaterra: yay. and it selects GENERIC_CLOCKEVENTS_BROADCAST when needed, too, btw Mar 17 00:00:02 rsalvaterra: so I think we either enable it (if it works) or take out GENERIC_CLOCKEVENTS_BROADCAST altogether Mar 17 00:00:23 rsalvaterra: i can test on er-x. is it supposed to save power or something? Mar 17 00:00:44 rsalvaterra: add -nostdinc as part of NOSTDINC_FLAGS Mar 17 00:01:13 lipnitsk: Sure, the ER10X is also MT7621, so you should be fine. Mar 17 00:01:54 ugh Mar 17 00:02:04 things become difficult when i uninstall vim Mar 17 00:02:18 mangix: I use nano… :P Mar 17 00:02:21 * rsalvaterra runs Mar 17 00:02:29 rsalvaterra: need word wrapping Mar 17 00:02:47 Ah, well… Mar 17 00:04:36 lipnitsk: https://gist.github.com/neheb/bfab060cf1b0b06779e290eef1866d66 Mar 17 00:05:35 lipnitsk: So tomorrow I'll do a build with MIPS cpuidle and the TEO governor. :P Mar 17 00:06:43 lipnitsk: btw the guy who wrote the mt7621 NAND driver is a mediatek employee. keep that in mind. Mar 17 00:09:27 mangix: well right... so is the guy who wrote the one in the kernel. Then Greg came in and wrote his version for whatever reason, which has references like: Mar 17 00:09:41 * lipnitsk < https://matrix.org/_matrix/media/r0/download/matrix.org/lOZXMqgBgeWxQyovNXYYGSpI/message.txt > Mar 17 00:09:58 no mention of that at all in the other ones? unless that's part of mtk_ecc Mar 17 00:10:54 lipnitsk: different employees different divisions Mar 17 00:12:11 mangix: yeah so they tried to upstream but didn't have time to do it properly Mar 17 00:12:55 the mt7621 stuff? Yeah most of the drivers were upstreamed by third parties Mar 17 00:13:53 interesting gist Mar 17 00:17:20 one benefit of logging IRC :) Mar 17 00:18:19 mangix: Yeah, but isn't logging thrice a bit… excessive? :P Mar 17 00:18:28 https://freenode.irclog.whitequark.org/openwrt-devel/2021-03-17 Mar 17 00:18:46 hahha Mar 17 00:19:08 Why do we have three whiteloggers here? :P Mar 17 00:20:02 rsalvaterra: irssi did it by default. Quassel does it too. except the latter doesn't come with a builtin search Mar 17 00:21:07 Alright, off to bed now… cheers, everyone! **** ENDING LOGGING AT Wed Mar 17 02:59:56 2021