**** BEGIN LOGGING AT Thu Feb 11 02:59:57 2021 Feb 11 05:59:03 can I just say how much I hate Android's process for managing code Feb 11 09:56:23 I'm a bit miffed I still see the GPIO IRQ handler fucking off after a while, on the Omnia (not mine, mind you, as I disabled GPIO IRQs as a workaround). It really needs to survive spurious interrupts. Feb 11 09:56:53 I'm going to try and fix it myself. Feb 11 10:44:12 pkgadd: ping Feb 11 11:41:48 Hm. The potential fix isn't trivial to backport. Whatever. I just diffed the v5.4 gpio-pca953x driver against the current v5.11-rc7 and made a patch. Hopefully it'll build/run. Feb 11 11:43:15 The world changed a lot between 5.4 and 5.11… Feb 11 12:00:52 rsalvaterra: mt7621? Feb 11 12:01:11 wait, the Omnia... that's mvebu right Feb 11 12:27:22 Borromini: Yeah, mvebu. Feb 11 12:30:59 … and boom. Build failed. Oh, well… Feb 11 12:31:42 … speleology time. Feb 11 13:10:07 anyone here understands NAT / routing implementation defailts in Linux kernel? Feb 11 13:10:38 is that even possible to use multiple CPU cores for doing RX, NAT, TX? Feb 11 13:10:51 afaik it's not implemented Feb 11 13:10:59 but is that doable at all? Feb 11 13:11:11 so we get multiple CPU cores handle single TCP stream traffif Feb 11 13:11:14 nbd: ? Feb 11 13:34:05 rmilecki: a single stream needs to be pinned to one CPU, otherwise you get reordering Feb 11 13:36:49 nbd: is that possible to improve it anyway? Feb 11 13:38:29 not for me ofc, but for someone with good routing, nat, napi, skb, whatever knowledge? Feb 11 13:49:49 rmilecki: well, the way to improve it is to make sure that as few cycles as possible are spent handling it Feb 11 13:50:11 i don't think it's worth trying to parallelize a single flow Feb 11 13:51:06 nbd: some devices have 2.5 Gbit Ethernet now but OpenWrt/Linux is still capable of doing only 800 - 900 Mb/s NAT Feb 11 13:51:40 what platform? Feb 11 13:51:46 nbd: even with optimized paths I'm afraid we won't hit 2000 - 2500 Mb/s Feb 11 13:51:51 BCM4908 Feb 11 13:52:03 4 x ARM A53 Feb 11 13:52:18 how much can it do with multiple flows? Feb 11 13:52:34 i didn't get iperf2 yet, let me do testing later this week Feb 11 14:00:02 make sure you don't run iperf on the router itself Feb 11 14:00:26 and if you have some time, you could make a perf flamegraph Feb 11 14:00:30 to narrow down where the bottleneck is Feb 11 14:01:07 i've merged an updated version of the netfilter flow offload code to my staging tree 5.10 kernel Feb 11 14:01:11 maybe it'll help as well Feb 11 14:01:21 and i'll pull in the latest version of napi threading support as well Feb 11 14:01:24 with >1Gbit there are some effect of jumbo frames on single flow too Feb 11 14:02:42 so the "PPS" packet per second might be relevant to compare since some firewalls have these benchmarks beside Mbit/Gbits too Feb 11 15:13:21 Borromini: The gpio-pca953x backport requires me to frob with the bitmap library… I only added API, so it should be fine. Feb 11 15:13:49 * rsalvaterra can't wait for Linux 5.10 to hit master… Feb 11 15:36:52 :) Feb 11 15:37:14 i switched to master thinking we'd get to 20.xx or 21.xx rather quickly :P Feb 11 15:39:32 Borromini You just made me chuckle! Feb 11 15:41:01 Borromini I have bin using OpenWrt since BB so.... Feb 11 15:47:04 Tapper: if we're doing e-penises, i since 10.03 :P Feb 11 15:47:43 Borromini so long! hahahah Feb 11 15:48:06 yeah getting old xD Feb 11 17:01:50 build #744 of malta/be is complete: Failure [failed tools] Build details are at http://buildbot.openwrt.org/master/images/builders/malta%2Fbe/builds/744 blamelist: INAGAKI Hiroshi , John Audia Feb 11 17:30:45 Borromini: I miss the cocktail version names… :P Feb 11 17:38:41 Hello. I just received an ubnt unifi 6-lr. Is there currently someone working to port openwrt to it? Do you need help with something (ex: mtd dump)? Feb 11 17:41:16 jlsalvador: https://git.openwrt.org/?p=openwrt/staging/blocktrron.git;a=summary Feb 11 17:41:35 if you bought it blindly, you're a *very* lucky guy. Feb 11 17:43:31 Borromini: :) Feb 11 17:48:02 Borromini: Hehe, I wasn't fast enough. :) Feb 11 17:50:24 BTW, blocktrron thank you! Feb 11 18:05:14 FunFact about this device: It has an Aquantia AQR112 2.5G PHY which Ubiquiti crippled to 1 Gbit/s in software Feb 11 18:06:18 I'm seriously interested in the stuff their PM smokes Feb 11 18:07:38 blocktrron: Why ubnt software-limit it? To prevent thermal problems? Feb 11 18:08:24 Most likely so they don't piss off their hardcore fans who bought the Early Access LR Feb 11 18:09:05 The LR which is on the market was intended to be the U6-Pro Feb 11 18:09:45 But yeah, I've ordered a 2.5G Adapter, which is set to arrive on monday to see whetehr or not 2.5G works Feb 11 18:10:11 sadly linux has no support for this PHY, so we will have to see Feb 11 18:10:57 It's a good reason. I think that Ubnt its more PR than tech-y. Feb 11 18:11:45 Oh! You lifted the restriction? :D Feb 11 19:03:43 credit where credit is due eh blocktrron ;) Feb 11 19:22:16 well, technically 2.5G should work with the hardware, I'm just not sure whether or not the PHY will advertise it. Feb 11 19:23:03 If not, I'll either have to find a datasheet, wait for someone with one to add this PHY to the linux driver or hope the U-Boot sources supporting this PHY work with 2.5G Feb 11 19:27:39 blocktrron: the 10G Aquantia Phys need some special FW and this FW can restrict the supported modes Feb 11 19:27:55 is the FW loaded at runtime or is theer an extra SPI flash? Feb 11 19:28:26 or is there no extra FW needed for this PHY Feb 11 19:29:49 Hauke there are indeed two SPI flashes Feb 11 19:32:07 i can dump the firmware of the second one, as it's a multilayer PCB i can't say for yure whether or not it is wired to the marvell PHY Feb 11 19:36:42 The host system does not interact with the PHY at all. It's configured for a reference board with a RTL switch Feb 11 19:40:30 blocktrron: why does it use a RTL switch and a Aquantia PHY? The RTL switch has integrated PHYs and only supports 1G on the LAN ports Feb 11 19:41:03 Hauke: It doesn't have a RTL switch at all. The U-Boot as well as the kernel are however configured for an RTL switch Feb 11 19:41:04 do you see the PHY on the MDIO bus? Feb 11 19:41:31 The Phy is at ID8 on the MDIo bus, yes Feb 11 19:42:50 blocktrron: and it's a production device? as in: available to consumers? Feb 11 19:43:13 you didn't get your hands on some early dev version? (just curious) Feb 11 19:45:01 Hauke: see for example: https://github.com/blocktrron/u6-lr-gpl/blob/master/uboot-uap-mtk-1.1.43/config-uap6-lr#L35 Feb 11 19:45:37 It also does not detect link state in the OS at all, the link is fixed to 1G Feb 11 19:47:14 Borromini: nope Feb 11 19:51:36 Hmpf… My gpio-pca953x backport patch is weighing at about 27 kiB. Not great, not terrible… Feb 11 19:51:53 tmn505 just a follow up from what we discussed on tuesday. The device kept the config after a flash so I believe that the filesystem marker found at 0x0 is ok to ignore. I've try to remove my dts node with the wdt declaration but I the device enter a bootloop as I mentioned.  do you have any other sugestions to ignore de node imported from the Feb 11 19:51:54 dtsi? Feb 11 19:57:23 I'm trying to port CF-E375AC from ar71xx to ath79, PR available https://github.com/openwrt/openwrt/pull/3319/ for reference the match file from ar71xx https://github.com/openwrt/openwrt/blob/openwrt-18.06/target/linux/ar71xx/files/arch/mips/ath79/mach-cf-e316n-v2.c#L356 Feb 11 20:08:03 joaohcca: then the watchdog from dtsi needs to be enabled, but it's weird that it spews error when activating and works. Btw. the mach file You pointed activates watchdog by toggling gpio to high but Your dts uses low, maybe there's the issue? Feb 11 20:09:08 tmn505 I'll change to ACtive_high and test Feb 11 20:15:18 tmn505 still got the same messages from the wdt-ath79 [    1.673024] ath79-wdt 18060008.wdt: unable to register misc device, err=-16 Feb 11 20:15:18 [    1.680269] ath79-wdt: probe of 18060008.wdt failed with error -16 Feb 11 20:18:21 my /dev has watchdog and watchdog0. Could be an issue when trying to register /dev/watchdog? Feb 11 20:29:10 joaohcca: probably, if it doesn't reboot then it's somehow toggled. I'm out of ideas about this since I don't know much about it. Consider the support done, since all of the devices on the unit work. Feb 11 20:33:12 tmn505 I've seen this messages on other ported devices from COMFAST such as https://github.com/openwrt/openwrt/pull/1957 Feb 11 20:39:33 yeah, so don't bother Yourself with it too much. Feb 11 20:41:18 would like to understand what happen so if anyone has any suggestions feel free to drop a comment at the PR Feb 11 20:42:05 tmn505 thanks for taking some time for this, much appreciate Feb 11 22:33:47 nbd: do you have a better version of the package/kernel/mac80211/patches/subsys/300-mac80211-optimize-skb-resizing.patch patch which works better with mac80211 5.10? Feb 11 22:35:00 joaohcca: I think the GPIO and the ath79 watchdig driver want to register a misc char device for the watchdog and it looks like only one is possible Feb 11 23:10:56 nbd: sorry for the drive-by. i am still trying to find the commit causing the timeout errors on the mt7613 radio, but by now i'm essentially at 2021-01-14 and it looks like even that shows timeout errors, although the AP had been up over 24h now. my backup testing device isn't showing any timeout errors yet with the same build, but it's only been up half as long for now. will keep you posted. Feb 11 23:11:17 bbl. **** ENDING LOGGING AT Fri Feb 12 02:59:56 2021