**** BEGIN LOGGING AT Tue Mar 26 02:59:56 2019 Mar 26 07:26:13 mangix: I don't know, sorry Mar 26 07:27:59 [ 25.958433] br-pub: received packet on eth0 with own address as source address (addr:68:72:51:xx:yy:zz, vlan:0) Mar 26 07:28:19 russell--: ralink ? Mar 26 07:28:26 anybody see the before? from a ubnt nano loco m2 Mar 26 07:28:46 pretty sure it's some kind of hardware fault Mar 26 07:29:04 its a unroutable package Mar 26 07:29:20 the fdb inside the switch fabric will bounce packets to the cpu port when the lookup fails Mar 26 07:29:46 interesting Mar 26 07:30:09 morning Mar 26 07:30:14 hi Mar 26 07:38:21 hi Mar 26 07:51:53 i need to find cheap HW x86 based that has 2,5 or 5gbit ports Mar 26 07:52:18 anyone have a some pointers ? Mar 26 07:58:15 10gbit networkcard? Mar 26 08:18:49 blogic: look at recent NAS, they are starting to move > 1 Gbits Mar 26 08:20:44 e.g. https://www.qnap.com/en/product/tbs-453dx Mar 26 08:30:24 they generally have aquantia nics that support 2.5G/5G/10G on copper Mar 26 08:57:23 zorun_: good idea Mar 26 08:57:31 need to see if i can install debian on them Mar 26 08:57:49 actually need it to test throughput on an AQ phy Mar 26 08:57:52 we need cheap fiber equipment maaan Mar 26 08:58:08 kafigaz: i have rtl9062c on my desk Mar 26 08:58:14 tiny gpon boards Mar 26 08:58:23 but caps at around 600mbit Mar 26 08:58:27 I have gpon from isp Mar 26 08:58:30 uwu Mar 26 08:58:49 doing 256byte udp frames 200 interleaved flows that is Mar 26 08:58:53 but yeah, cheap 10G LAN at home when Mar 26 08:59:05 i need it for my testbed for now Mar 26 08:59:16 at home i dont have the need for that kind of bandwidth Mar 26 08:59:27 serving 2 HD flows is tops what i ever need Mar 26 08:59:39 and that can be done on 100mbit with plenty spare Mar 26 08:59:45 ye Mar 26 09:00:03 same but I dont get why 10G isnt on mainstream yet Mar 26 09:00:10 i mean, mainstream home use Mar 26 09:00:22 because no one needs it Mar 26 09:00:32 you only see 10G nics on HEDT motherboards Mar 26 09:08:12 1) mostly lack of real need, which then 2) no conusmer prised devices... also 3) 10g is starting to be on limit on 4-pair copper wiring.... traditional 100meters is hard to achieve (while not often needed ofcourse), also rj45 connector is limiting Mar 26 09:10:20 I think we start seeing 1000Base-T1 earlier than 10gbit. Mar 26 09:12:25 1000Base-T1 is already used in cars. And in my industry oil&gas they also making a 10mbit standard for it. Mar 26 10:32:25 how do i debug Out of memory errors on openwrt? Mar 26 11:32:10 lazy command of the day: echo "Fixes: $(git log -1 --pretty='format:%h ("%s")' --abbrev=12 8de907c441db)" Mar 26 11:36:37 kafigaz: 10G Ethernet also needs a lot of power, 3 Watts is normal and it does not work over 100 meters CAT 5 cable, the next step will be 2.5G Mar 26 11:42:58 ok guys, I need some help with network performance Mar 26 11:43:22 the last kernel with high network routing/NAT performance for me is 4.4 Mar 26 11:43:37 what is weird that i see performance drop when switching from 4.4 to 4.4.1 Mar 26 11:43:49 i first thought of regression in upstream kernel but it's NOT the case Mar 26 11:44:07 if I revert all changes from 4.4.1 i still see lower performance Mar 26 11:44:34 i used: "git diff v4.4..v4.4.1 > ~/4.4.1.diff" and then "patch -p1 -R < ~/4.4.1.diff" Mar 26 11:45:54 also, applying 4.4.1 changes to the 4.4 kernel (without version bump in OpenWrt's build root) also gives me good routing/NAT performance Mar 26 11:46:45 rmilecki: the only thing you changed is the LINUX_VERSION? Mar 26 11:46:49 in other words I get good NAT/routing performance when using: Mar 26 11:46:53 1) OpenWrt with kernel 4.4 Mar 26 11:46:54 2) OpenWrt with kernel 4.4 + patch applying changes from 4.4.1 Mar 26 11:47:12 I get BAD routing/NAT performance when using: Mar 26 11:47:13 1) OpenWrt with kernel 4.4.1 Mar 26 11:47:43 Hauke: yes, it's about "LINUX_VERSION-4.4 =" vs. "LINUX_VERSION-4.4 = .1" Mar 26 11:47:46 does your 4.4.1 patch also changes the version number of the kernel? Mar 26 11:48:21 Who cares cat_5_ cable... problem is that while we still can physically build a cable that can make 100 meters possible in 10G/s speeds it is also the connector that is lacking.. ie for any longetivity we need basically new standard (or, to use existing more widely).. but like said earlier, first there need to be general need... cat6a can still work in "traditional lets change Mar 26 11:48:22 nothing" in physichal etc form... we might see 2.5 or not in between or then totally new system... cat7 at least tried new connector stuff, maybe too early Mar 26 11:48:27 was this done with current openwrt master? Mar 26 11:48:30 yes, my 4.4.1.diff modifies SUBLEVEL in the Makefile Mar 26 11:48:33 hmz .. maybe grep'ing for this would show a clue? : KERNEL_VERSION(4,4,0) Mar 26 11:48:46 Hauke: i'm testing old OpenWrt commit that still got 4.4 support Mar 26 11:50:10 rmilecki: is the performance regression on UDP and TCP? or only TCP? Mar 26 11:50:15 xback: i tried "grep -R "LINUX_VERSION" package/*" but didn't find anything interesting and it also seems that my 4.4.1.diff makes kernel detected as 4.4.1 anyway when using the LINUX_VERSION / KERNEL_VERSION Mar 26 11:50:20 iperf TCP Mar 26 11:50:31 none as scheduler Mar 26 11:50:35 rmilecki: we should support reproducable builds, you could check which binaries are different in both builds to get a clue where to look further Mar 26 11:50:39 sorry, pfifo_fast as scheduler Mar 26 11:50:44 rmilecki: 4.4.0 was around the time that Eric Dumazet refined TSO Mar 26 11:50:55 by default, TCP traffic is buffered for max 1ms Mar 26 11:51:04 if latency goes above, TCP will hold back Mar 26 11:51:45 I'm countering this locally by altering TCP TSO in source from 1ms to ~8ms Mar 26 11:52:11 but how can that issue be kernel code related, if applying 4.4.1 changes without bumping the LINUX_VERSION doesn't break the performance? Mar 26 11:52:47 the kernel marco should work the same way in both cases Mar 26 11:53:10 do you have the openwrt patches applied? Mar 26 11:53:16 maybe it's worth the shot to test this? Mar 26 11:53:17 was the toolchain rebuild? Mar 26 11:53:38 i'm wondering if one on OpenWrt packages could be affected by the LINUX_VERSION-4.4 value & apply some different configuration? Mar 26 11:53:41 jow? Mar 26 11:53:57 no toolchain rebuild Mar 26 11:54:13 i've dropped quite a lot of OpenWrt patches Mar 26 11:54:23 I would really suggest to check which binaries are different Mar 26 11:54:51 Hauke: dropped patches: https://pastebin.com/YEKhKcsf Mar 26 11:55:42 Hauke: good point, should I look for changing files in the bin/? Mar 26 11:56:27 (btw. I know all of it sounds a bit of nonsense, but I verified it multiple times before coming here) Mar 26 11:56:49 I would look here: build_dir/target-mips_24kc_musl/root-lantiq/bin/busybox Mar 26 11:57:11 or build_dir//root- Mar 26 11:57:21 and then look at the kernel binary Mar 26 11:57:25 ok, thanks! Mar 26 11:57:29 probably before compression Mar 26 11:57:58 rmilecki: purely fyi: the TSO patch using locally: https://pastebin.com/6upQPJSY Mar 26 11:59:19 xback: oh, that' snot a big change, I'll look at it, thanks Mar 26 12:28:22 blogic: what is rtl9062c ? Mar 26 12:28:30 google has no results Mar 26 12:41:03 Hauke: http://szdragon.com/html/FTTH/32.html Mar 26 12:45:50 https://forum.archive.openwrt.org/viewtopic.php?id=72425 Mar 26 12:50:01 funny how many private keys you find when searching for "PRIVATE KEY" in a regular vendor firmware update ;-) Mar 26 12:57:26 somebody in #openwrt asked whether their board is supported in openwrt (it currently isn't) and i'm curious what sort of performance it would have if it were Mar 26 12:57:44 octeon cn7010 Mar 26 13:00:07 And how do you suppose we answer that until there is support, if ever Mar 26 13:00:45 my response was "it would probably be supportable but nobody has done the work to port it" Mar 26 13:00:49 ..i think matrix puts replies to irc quite poorly.. Mar 26 13:00:56 yes Mar 26 13:02:51 but i haven't looked at the octeon target in almost 4 years and idk its state Mar 26 13:04:04 Sidenote: I proposed to matrix that as irc is what it is that all really is needed is namehandle and the reply medssge, maybe text (reply) if warranted.. any other is just waste in irc :) Mar 26 13:04:48 matrix _is_ a waste of irc Mar 26 13:08:07 Well... personally I like matrix and its bridges better than irc as a whole, but Mar 26 13:09:11 ...but I also realise matrix shouldn't try to implement a thing that is not there Mar 26 13:10:09 But well.. offtopic, so lets just get along :p Mar 26 13:19:47 Hauke: interesting, "uname -r" reports kernel version as it was before applying a stable release patch Mar 26 13:20:21 so e.g. after "patch -p1 < ~/4.4.1.diff" (which e.g. modifies Makefile and SUBLEVEL) i'm still getting "4.4" Mar 26 13:21:38 ah yes I think something overwrites it Mar 26 13:22:03 could you check if the kernel version test macro is also affected? Mar 26 13:22:31 will do Mar 26 13:23:45 /home/build/openwrt/staging_dir/toolchain-arm_arm1176jzf-s+vfp_gcc-7.4.0_musl_eabi/arm-openwrt-linux-muslgnueabi/include/c++/7.4.0/cstdlib:75:15: fatal error: stdlib.h: No such file or directory Mar 26 13:23:56 any clues how to solve that? Mar 26 13:24:17 stintel: did you do anything specail? Mar 26 13:24:38 there was a chnage to some gcc compile options merged yesterday Mar 26 13:25:05 Hauke: it's a Qt app. I think I had this problem before and manually copied the stdlib.h somewhere but can't remember where Mar 26 13:27:59 cp staging_dir/toolchain-arm_arm1176jzf-s+vfp_gcc-7.4.0_musl_eabi/include/stdlib.h staging_dir/target-arm_arm1176jzf-s+vfp_musl_eabi/usr/include/ Mar 26 13:28:47 that fixes it, but I'm wondering why this file isn't there in the first place Mar 26 13:29:43 Hauke: LINUX_VERSION_CODE works as expected, it seems to respect SUBLEVEL correctly. i'm getting compilation error for: Mar 26 13:29:46 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 4, 1) Mar 26 13:29:48 #error test Mar 26 13:40:31 ok Mar 26 13:40:50 are the modules getting loaded? Mar 26 13:54:00 stintel: more like why isn't your thing looking at the toolcahin? Mar 26 13:54:16 the target usr/include never has the std headers. (for me at least) Mar 26 13:56:41 Hauke: yeah, no problem with module Mar 26 14:03:58 how do you turn on buddyinfo in openwrt kernel? Mar 26 14:09:50 how does one normally turn it on? Mar 26 14:10:27 target/linux/generic/hack-4.19/902-debloat_proc.patch:- proc_create_seq("buddyinfo", 0444, NULL, &fragmentation_op); Mar 26 14:10:30 it's patched out Mar 26 14:11:23 karlp: normally its included in the kernel by default Mar 26 14:12:11 how can it be patched out? how do you get the buddyinfo without it? Mar 26 14:12:33 that's exactly what that patch does, and you don't Mar 26 14:12:58 unless you build with CONFIG_PROC_STRIPPED=n Mar 26 14:13:43 ohh, i'll try that Mar 26 14:17:15 hmm seems that can only be done in the kernel config, not in the openwrt .config Mar 26 14:17:35 I guess it would be nice to have that available in openwrt menuconfig Mar 26 14:18:48 that's where i tried looking for it at first, but wouldn't have found it anyways because i didn't know about the proc stripping Mar 26 14:18:55 or not, out there niche use case, leave it where it belongs. Mar 26 14:53:50 man, czmq got let it with no description, broken chars in the description, no url. thos cz.nic people are special Mar 26 14:59:20 :-Ø Mar 26 14:59:24 harhar Mar 26 15:38:03 karlp: I'm sorry, but yeah, you complain about it here. Next time, I'd like to give your feedback in pull request, which was created. https://github.com/openwrt/packages/pull/8014 nobody noticed it there. Mar 26 15:38:28 rip my English. :-D Mar 26 15:39:16 * I'd like to kindly ask you to provide your feedback next time in pull requests. It was reviewed by community. Mar 26 15:39:38 and I do, normally. Mar 26 15:39:41 but I was on holidays. Mar 26 15:39:50 so I filed an issue for them to improve it in the future. Mar 26 15:42:39 Thanks. I see you created an issue! We'll address it ASAP! I'm contributing to OpenWrt since January and I haven't seen you in any pull request. No need to complain that we're special. You could avoid it as there isn't a reason to be sarcastic. :) This one slip through my eyes. Mar 26 15:42:56 It could happen to anyone. Mar 26 15:43:17 it can, and it does, so I filed an issue to improve it down the road. noen of us are capable of following everything. Mar 26 15:56:08 xback: ynezz: looking into the git thing now Mar 26 16:08:02 jow: thanks! Mar 26 16:53:46 rmilecki: any luck on your investigation? Mar 26 17:01:16 * russell-- tried to buy a gigapoint from calix recently, they wanted me to sign an NDA before they'd show me a spec sheet, i replied i'd just buy one on ebay. Mar 26 17:12:49 Definitely something strange with batman-adv building on ath79 Linux 4.19 -- works on 4.14, fails to compile with re-DEFINE -- looking into header include changes Mar 26 17:17:20 my x86/64 kernel is too large for the default KERNEL_PARTSIZE (4MB). this results in an unbootable image being created. I propose this change to avoid creation of unbootable image: https://gist.github.com/3d47f0ab1aa12dc5fdd3338a41222f88 Mar 26 17:17:31 any thoughts? Mar 26 17:19:57 stintel: ack, I'd also suggest updating the default kernel size to 16MB or so Mar 26 17:20:02 in a separate commit Mar 26 17:20:44 4MB on x86 is extremely conservative, even when considering very old boards (wrap, soekris) and old mmc cards Mar 26 17:20:47 it's already 16MB by default Mar 26 17:20:51 ah ok Mar 26 17:20:53 ah, it changed ? Mar 26 17:21:02 then it's just my config that contained the previous default Mar 26 17:21:14 anyway, the fix is good Mar 26 17:21:21 ack, commit it Mar 26 17:22:05 will do, thanks! I'll add acked-by from both of you ? Mar 26 17:22:20 yep Mar 26 17:23:19 but, maybe it would make sense to fix it properly? Mar 26 17:23:48 if any of the steps fail in that file it could be a problem Mar 26 17:25:02 not sure how to handle that in "Image/Build/grub2" Mar 26 17:25:23 maybe a simple set -e in that script Mar 26 17:27:12 make_ext4fs fails with this message: error: ext4_allocate_best_fit_partial: failed to allocate 33 blocks, out of space? Mar 26 17:27:26 which should be enough info, no need for an extra message Mar 26 17:27:33 let me try set -x Mar 26 17:27:47 yes, but `set -e` might need handling of the `rm -f "$OUTPUT"` on line 16 Mar 26 17:28:11 this could probably fail if you're building from scratch (not sure) Mar 26 17:28:32 rm -f $OUTPUT || true ? Mar 26 17:28:59 possibly oneline fix becomes weekend project :) Mar 26 17:29:03 :D Mar 26 17:29:29 ynezz: rm -f returns 0 even if the path doesn't exist Mar 26 17:29:52 hm, I'm having problem with that in Makefiles so I'm not 100% sure Mar 26 17:30:48 (but maybe it was something else) Mar 26 17:31:20 yeah, you're correct Mar 26 17:33:35 set -e causes make to fail as well, now trying with a KERNEL_PARTSIZE that is large enough Mar 26 17:40:36 what should we answer the guy from OnLife Project Edge? Mar 26 17:40:49 ynezz: jow: https://git.openwrt.org/?p=openwrt/staging/stintel.git;a=commitdiff;h=751a3e3fc5ff0572c5d5235b5c2d68b2dca7c26e - tested in good an bad scenario Mar 26 17:46:41 let me know if I can add both your acked-by and I'll push it Mar 26 17:46:52 xback: i did one more check only, I verified my kernel sources for bad and good are the same Mar 26 17:46:54 they are Mar 26 17:47:00 xback: i'll continue tomorro Mar 26 17:48:27 jow: you know quite a lot about packaging, do you know what else than a kernel can be affected by the "LINUX_VERSION-4.4 =" vs. "LINUX_VERSION-4.4 =.1"? Mar 26 17:48:29 jow: I'm getting different routing performance for 4.4 + patch updating sources to 4.4.1 vs. 4.4.1 Mar 26 18:09:58 xback, you around? Mar 26 18:10:39 I tested 9880 IBSS on my 4.20 kernel, first time I associated, I got one way traffic, but after resetting a node, I see bi-dir traffic around 170Mbps one direction, 317Mbps the other (UDP) Mar 26 18:11:22 I see a few SWBA messages, but not many Mar 26 18:12:17 SWBA appears to happen when WMI communication is being slow... You might could enable HTC credit logging and/or WMI logging to better understand why that is happening (credits slow to be returned, very large number of on-going WMI messages, etc) Mar 26 18:16:54 rmilecki: uhm good question, I could only imagine random kmods getting enabled or diasabled based on version guards Mar 26 18:19:07 jow: hm, 4.4 vs 4.4.1 seems unlikely... Mar 26 18:19:10 but i'll check it Mar 26 18:21:29 rmilecki: I also think thgere is some check like foo <= 4.4 and this is true for 4.4 and not for 4.4.1 Mar 26 19:48:03 stintel: so simple, nice, thanks! Mar 26 20:59:27 hi! i need some help with quilt and kernel patches for a router i'm trying to add support for Mar 26 21:00:25 i've followed the steps on the dev page for adding a kernel patch, but when i build it's as if my patch is applied in the wrong order compared to all the others, so a subsequent patch fails Mar 26 21:02:35 in fact it's the last kernel patch that fails, are the patches applied in the order to how they're numbered? Mar 26 21:11:21 AKAIK, they are by "class", then by "collation order" within each class -- `quilt series` should show the order Mar 26 21:12:01 You *may* need to update/refresh a subsequent patch Mar 26 21:44:04 jeffsf: re https://forum.openwrt.org/t/archer-c7-v4-5ghz-slow/34100/5 just nitpicking, but PPPoE also adds considerably to the CPU load Mar 26 21:44:44 Agree -- thanks, edited Mar 26 21:45:04 An Archer C7 ain't gonna hit 600 Mbps Mar 26 21:46:40 my tl-wdr4300 (560 MHz) is close to its limit at 100/40 MBit/s with PPPoE, it can do it, but there are frequent load spikes under heavy utilization Mar 26 21:47:45 it certainly won't do anything close to 400 MBit/s (or even 300 MBit/s, accounting for the difference in clock speed) with PPPoE enabled Mar 26 21:49:16 Thankfully mine have generally only needed to route Ethernet / 802.11 packets. Mar 26 21:49:30 I could get ~450 Mbps on a good day Mar 26 21:49:30 I'm still kinda amazed that PPPoE is still a thing at certain.. well.. countries I guess... as in general means of broadband providing :) Mar 26 21:50:12 unfortunately it's still in wide use for ADSL/ VDSL Mar 26 21:50:27 exactly, and for no real reason :) Mar 26 21:51:05 in general sense I don't see what benefit does it really add.... to the ISP even taht is Mar 26 21:52:18 before vectoring, it allowed resellers to offer their services. nowadays it's probably mostly a matter of not touching a proven configuration Mar 26 21:53:15 (even including two parallel PPPoE sessions to different ISPs over the same line) Mar 26 21:55:00 I was just about to add that unless multiple ISP offer services at the very same time on same line.. but you beat me to it =) I'm just glad they got rid of that in generally Finland long time ago when ADSL was still a major thing :) Mar 26 21:55:51 Also, I did not say it would never has a usecaes, but in default general it's just meh.. Mar 26 22:06:00 git log Mar 26 22:06:05 heh Mar 26 22:08:16 Sidetrack: I wonder that with what method Accelerated (nowadays Digi) SR-6350 and such they make LTE-passthrough happening :) Mar 26 22:09:44 there is sierra wireless MC7455 on mine as LTE-modem, it can be configured so that an downstream router gets IP from ISP that LTE is connected to, all while modem itself is still capable of accessing internet too (say, update modem firmware list and ping and whatnot) Mar 26 22:09:59 Hauke: i just thought of that, hopefully that's it Mar 26 22:35:00 *yawn* Mar 26 22:35:22 note to self, working from 7am till 11:30pm does wear Mar 26 22:35:46 i'll go an check out if that thing in the bedroom is still as soft as last night Mar 26 22:38:06 jeffsf_: thanks i'll have a closer look at that Mar 26 22:38:42 Sleep well blogic Mar 26 22:45:50 Hauke: jow: i compared my image with 4.4 kernel with image with 4.4.1 Mar 26 22:46:36 Hauke: jow: opkg list-installed output contains the same list of packages (only version of kernel & kmod differs: e.g. 4.4-1-c5c793c5d547b65de656af6b9c07d246 vs. 4.4.1-1-c5c793c5d547b65de656af6b9c07d246) Mar 26 22:47:02 Hauke: jow: the output of "ls /lib/modules/4.4*/" is the same for both images Mar 26 22:47:12 it's slowly killing me... Mar 26 22:47:17 xback: FYI ^^ Mar 26 23:01:00 I FOUND SOMETHING Mar 26 23:01:07 jow: can you take a look at https://pastebin.com/VSKzF9dF please? Mar 26 23:01:23 nice cat :) Mar 26 23:03:48 rmilecki: https://i.imgur.com/YPv7oXK.png Mar 26 23:04:00 oh :( Mar 26 23:04:37 mirror: https://paste.ubuntu.com/p/8cxkKBzSJv/ Mar 26 23:05:21 the pastebin link worked for me Mar 26 23:08:43 same here Mar 26 23:09:23 now, any hints why starting with kernel 4.4.1 OpenWrt uses different firewall setup? what is it about? can it indeed affect NAT/routing performance>/ Mar 26 23:13:31 rmilecki: hm, which commit is 4.4.1 ? I can find only bump to 4.4.3 cb04b8d58201f6aa35f99b76ee8b3435beb7a01e Mar 26 23:14:30 so you should probably try to bisect `git log --pretty=oneline f875e18fd28cf744a504f0f8ddbfd9d4c211098c..cb04b8d58201f6aa35f99b76ee8b3435beb7a01e` ? Mar 26 23:16:09 for fw3 it would be probably this one 6064710b90b6138c3471129090ab4192710be42c firewall: drop invalid by default, remove chain indirection, fix invert flags (#21738) Mar 26 23:18:50 and this commits in fw3 http://paste.ubuntu.com/p/BNVpsTPcrc/ (git log --pretty=oneline 980b7859bbd1db1e5e46422fccccbce38f9809ab..8957be6c026858fe414aef69281d8aa06f7ea122) Mar 26 23:20:16 rmilecki: in my current master build __flags_v4 and __flags_v6 are not set by default Mar 27 01:59:43 wdr3600 with 4.19.x: Mar 27 01:59:49 [37479.371158] do_page_fault(): sending SIGSEGV to cat for invalid write access to 7fe5ec20 Mar 27 01:59:52 [37479.379516] epc = 00403d71 in busybox[400000+47000] Mar 27 01:59:55 [37479.384578] ra = 0040a8db in busybox[400000+47000] Mar 27 02:00:31 r9716-616ec4365c **** ENDING LOGGING AT Wed Mar 27 02:59:57 2019