**** BEGIN LOGGING AT Fri May 10 03:01:14 2019 May 10 07:09:27 why is cake so slow on the wrt3200acm? May 10 07:09:30 https://forum.openwrt.org/t/cake-at-300-mbit-on-wrt3200acm-in-2019/36557/5 May 10 07:09:44 Glad to find out it's not just me! May 10 07:20:10 Tapper: I can get 235 out of a 250 meg connection May 10 07:20:20 Tapper: with cake on WRT1200AC. 40% CPU. May 10 07:20:40 Tapper: what are your numbers compared to what you get without SQM enabled? May 10 07:21:21 with SQM off I get 220 down May 10 07:22:41 Queue setup script May 10 07:22:41 layer_cake.qos May 10 07:23:34 SwedeMike I have a cable connection May 10 07:23:56 doxes 3 if that's how you spell it May 10 07:24:59 more like DOCSIS 3.0 or 3.1 May 10 07:26:30 DOCSIS 3.0 yeah May 10 07:27:27 virgin UK May 10 07:35:28 Tapper: did you change the overhead setting? May 10 07:36:01 (link layer adaptation) May 10 07:39:41 SwedeMike no mate what should it be? May 10 07:42:34 Tapper: I was just wondering if it's still "none (default)" or if you changed it to something. May 10 07:43:12 SwedeMike o no i left it blank. May 10 07:44:34 You should leave at none to use the built-in overhead calculations. May 10 07:45:10 Tapper: can you go to http://www.dslreports.com/speedtest and perform a test? mine is here: http://www.dslreports.com/speedtest/49513110 with 250/100 setting. If you can share yours, that'd be good and might give some clues. May 10 07:46:55 http://www.dslreports.com/speedtest/49513135 May 10 07:47:13 and you probably need to tell cake some overhead keywords - which means enabling both 'use advanced configurations' and 'use dangerous configurations' May 10 07:47:46 ldir what should i put in the boxes mate May 10 07:49:09 don't know off the top of my head - and have to dash out the door to go to work - maybe catch you later May 10 07:49:39 ldir np mate have a good day. May 10 07:49:43 Tapper: I don't think you'll get much better than the result you're showing there. With DOCSIS there is a scheduler you're fighting against that's looking at queue fill on the segment etc, and with SQM you're purposely not filling the buffer meaning it might give you less slots to transmit in. To give you even better SQM results you might even have to lower the rate even more. May 10 07:49:51 am on vdsl myself May 10 07:49:59 ldir kk May 10 07:50:10 Tapper: when I had this on my DOCSIS connection it never worked as well as on my FTTH connection, mostly because of the way DOCSIS works. May 10 07:50:17 SwedeMike ok mate It's not a biggy really May 10 07:50:40 180 down is still fast anyway May 10 09:35:32 can we use clang-format for the firewall3.git? It uses indents partly spaces and tabs. May 10 09:35:39 jow: -^ May 10 09:38:48 lynxis: -^ May 10 12:19:12 Well, pressed "send" on upstream patches for spi-nand support for GigaDevice. Now to see how badly I get beaten up May 10 13:00:09 gch981213: any idea how to solve this different mdio probing order on ar71xx Vs ath79 https://github.com/openwrt/openwrt/pull/1990#issuecomment-490396451 ? May 10 13:23:27 and my dsl is fixed and amazingly i suddenly get whooping 16mbit May 10 13:23:42 this feels like the millenium falcon May 10 13:24:03 blogic: Welcome to 2007 May 10 13:24:11 :) May 10 13:24:36 welcome to the 20th century! May 10 13:25:49 ynezz: the quick&dirty solution is to reorder the nodes in the dtsi file (done already for mvebu IIRC) - the proper solution might be to add support for renaming interfaces to procd/hotplug (whereever appropriate) to rename them to their expected names based on their full node paths May 10 13:26:36 KanjiMonster: thanks for the hint! May 10 13:26:49 didn't knew, that the order in the DT matters May 10 13:27:28 and that it's somehow guaranteed :) May 10 13:27:54 I don't think this is guaranteed, just the way linux works currently May 10 13:28:14 yeah, so it could change anytime soon, right? May 10 13:28:46 dts(i) rules also specify that nodes should be ordered by their addresses, so reordering them to fix a software "issue" is technically wrong ;) May 10 13:40:21 KanjiMonster: this is the rename you've had in mind https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=target/linux/mvebu/base-files/lib/preinit/06_set_iface_mac;h=534a271efbfbbe02abdb485ffad0f57588b4f393;hb=HEAD#l18 ? May 10 13:41:48 ynezz: right - forgot we dropped the reordering May 10 13:54:24 ynezz: Implement a proper probe defer for ag71xx and drop my hacky mdio/eth driver split: https://git.openwrt.org/?p=openwrt/openwrt.git;a=commit;h=83d2dbc59924c90dd865de113a55d60826ac0c04 May 10 13:58:14 ynezz: The problem here is that eth0 needs to use a PHY on builtin switch, which can only be accessed from mdio bus on eth1. I thought atheros made separated mdio/gmac reset because they physically splitted mdio and gmac so I made the above commit. May 10 14:00:35 ynezz: and now I think they just want to workaround some unknown gmac problem by resetting only gmac block. (after we found out that only asserting both gmac/mdio resets will reset register values) May 10 14:03:43 ynezz: We can delay probing of gmac0 if driver failed to get a phy for current interface, and eth0/eth1 orders will be exactly the same as ar71xx. May 10 14:11:04 * gch981213 is trying to implement that. May 10 14:11:36 gch981213: is gmac0 directly wired to a phy? else it shouldn't try to get a phy, and should be just a fixed-connection for the switch May 10 14:13:44 KanjiMonster: It is. Atheros made a mux so that either the first or the last PHY on builtin switch is wired directly to gmac0 May 10 14:25:08 gch981213: on the MT7621 SOC with MT7530 onboard you can also mux PHY0 or PHY4 to 2nd GMAC. I just added the phy in the mdio node. Like this https://github.com/vDorst/linux-1/blob/9a677382901a806b833f2e6e3d2824cceee2af95/drivers/staging/mt7621-dts/UBNT-ER-e50.dtsi#L142 May 10 14:59:04 gch981213: ah, makes sense, I've forget about this May 10 15:00:48 gch981213: any idea where to hook that ar8xxx MIB config on ar71xx? I'm probably going to implement it during weekend, send the series to the mailing list for review/comments and push it afterwards May 10 15:02:41 ok, preinit script May 10 15:35:17 It works: https://github.com/981213/openwrt/tree/ag71xx_eth_swap And here's related kernel log: https://paste.ubuntu.com/p/NrQPpBxwnG/ May 10 15:36:10 Now I need to walk through every currently supported routers and swap eth0/eth1 in /etc/board.d/02_network for affected routers :( May 10 15:36:47 * karlp 's spider sense tingles May 10 16:28:47 * ldir also wonders if this has any nasty implications for lan/wan partitioning when the device is booting up - let alone the fallout of connecting lan to wan and wan to lan once booted May 10 16:30:01 what is the advantage of this change? May 10 16:31:55 ldir: the eth0/eth1 naming matches ar71xx, so switching to ath79 won't require config fixups May 10 16:33:08 Am I right in thinking that the Archer C7 v2 is ag71xx? May 10 16:33:53 ldir: ag71xx is tge May 10 16:34:01 ethernet driver May 10 16:34:22 I ask because I've swapped between ath79 & ar71xx builds for that device and haven't been (obviously) bitten by any eth0/eth1 swaps? May 10 16:34:53 radio0/radio1 does swap... or certainly used to do so. May 10 16:36:10 sounds like a race condition May 10 16:36:44 shouldn't happen with ethernet since that's specified in dts. May 10 16:37:52 which sounds like a race condition? May 10 16:38:46 the radio swap or the eth swap? May 10 16:42:01 radio May 10 16:44:54 well all the more that this eth0/eth1 swap change is thought through May 10 17:14:38 ldir: that's only a problem on routers using built-in switch. May 10 20:31:35 Hi can any one help with is pleas? my build keeps crapping out with this in the logs May 10 20:31:41 home/tapper/openwrt/openwrt/include/target.mk:45: /home/tapper/openwrt/openwrt/target/linux/mvebu/cortexa9/target.mk: Permission denied May 10 20:34:33 Tapper: if you want to get "full" log (ie lots of spam), run "make V=s 2>&1 | tee build.log" and look towards end what has really failed, often it is not last line, but fairly near the end May 10 20:36:08 olmari running now thanks. May 10 20:36:39 Tapper: care to share your /etc/config/sqm with me ? May 10 20:37:02 ldir KK stand by. May 10 20:37:48 ldir my firefox is crashing I will have to reboot I think May 10 20:38:22 hmm... out of storage? would explain both as a symptom May 10 20:38:39 I can at least have a think about what options you have set :-) May 10 20:40:46 it's not storage I have 180 gb free May 10 20:43:06 my mobo is dodge I have 6 gig of ram but onley 4 shows up to my os and I run 64 bit windows. May 10 20:43:41 I think I have a broken stick of ram May 10 20:45:43 could be - or it just needs reseating. May 10 20:54:06 ldir https://pastebin.com/YGJZCCSy May 10 21:03:01 ok I'll have a think, but first off I can see you're not compensating for DOCSIS overheads anywhere. May 10 21:04:15 as extras I'd put in May 10 21:04:58 option iqdisc_opts 'dual-dsthost docsis nat ingress' May 10 21:05:39 option eqdisc_opts 'dual-srchost docsis nat ack-filter' May 10 21:07:57 at a 17:1 or so ingress to egress ratio there may be benefit to thinning out the ack packets on egress so they don't consume all your egress bandwidth with 'send me the next part' messages. May 10 21:10:55 the 'dual-dsthost' on ingress and 'dual-srchost' on egress prevent one 'host' monopolising all the bandwidth by starting up many flows. e.g. without it if 1 host starts 9 downloads and another starts 1, 1 host gets 90% of the bandwidth whilst the other gets 10%. With the option they'd both get 50%. May 10 21:14:22 nat option makes the 'host' bits of the dual-srchost/dsthost bits work on the internal lan addresses of a NAT network...I'm assuming your router is the NAT.masquerading device. docsis option puts in some overhead calculation values. ingress changes the way that packets are accounted in the shaper on ingress. May 10 21:17:57 the iqdisc_opts & eqdisc_opts are the equivalent of the "Advanced option string " options in the sqm gui if you don't want to edit the file directly. May 10 21:39:26 ldir I just add those lines to the botem of the fiel? May 10 21:39:32 file* May 10 21:42:12 Then restart SQM May 10 21:47:34 yarp - but I'm off to sleep now so if you want to wait so you can throw things at me I'll understand. May 10 21:51:45 Ok mate chat later. May 10 21:51:49 thanks May 10 22:35:22 I am looking at a .o file that was created by default gcc flags for OpenWrt make May 10 22:36:27 When I run nm on the resulting .o file, all the symbols as listed as undefined "U" May 10 22:37:14 This is causing problems for a package I am porting as it tries to look at the symbols in the resulting .so file May 10 23:56:35 Pepe: ping May 11 00:48:59 pong **** ENDING LOGGING AT Sat May 11 02:59:57 2019