**** BEGIN LOGGING AT Tue May 21 09:32:48 2019 May 21 09:33:32 what do you mean by updating? May 21 09:34:12 sorry, after sysupgrade May 21 09:34:42 well, squashfs is read only, so it doesn't make sense May 21 09:34:59 I guess correct would be squashfs + jffs2 May 21 09:35:24 it seems to me, that either your SPI bus or flash is slow, I don't think, that other filesystem is going to make it better May 21 09:35:30 the issue isn't the squashfs, it's the jffs2 used for the overlay (which requires erasing/initializing all sectors used, which can take quite a while on slower NOR flash chips) May 21 09:35:47 KanjiMonster: Thanks, that is what I meant May 21 09:36:10 so I guess it is jffs2 that I am looking for alternatives to (if there are any) May 21 09:38:11 ubifs on ubi is the only alternative I guess. Not sure why you have power-cut issues; ubi(fs) is supposed to not have them, but then again ubi(fs) on NOR is probably less tested May 21 09:38:25 kristrev: "takes so long time", what amount of time are we talking about? May 21 09:38:43 SwedeMike: 5-6 minutes on the 32MB device I am working on now May 21 09:39:27 relative to how long the device will be switched on, it is not a lot of time, but it would anyway be nice to reduce it May 21 09:40:19 also I'm not sure if powercuts during the initialization time won't cause issues later on May 21 09:40:49 I will give ubifs another go. I tried the rootflags=sync work-around for ubi, but still had the issue with empty files. Maybe I did something wrong or there are some other options I can set May 21 09:41:26 kristrev: hm, ok, interesting. I have devices here that take 1-2 minutes extra on first boot after sysupgrade. We had to crank up the timeouts to 120 seconds to make sure all components would reliably work on first start and not timeout. May 21 09:41:59 software components May 21 09:42:45 SwedeMike: It sounds like it could be that for example a slow flash is the reason. I will see if increasing the frequency will improve performance (and not destroy anything) May 21 09:43:35 With 32MB or larger flash? May 21 09:43:51 kristrev: yes, that's what I deduced as well. The device I am testing on has smaller flash than yours so if it's the same speed then I am not totally surprised about your 5-6 minutes. May 21 09:44:13 With 16MB I also usually see times of around 1-2 minutes May 21 09:45:13 kristrev: WNDR3800 , 16MB flash May 21 09:50:43 SwedeMike: I see that the frequence of the SPI in the WNDR3800 is set to th esame as in most of the 16MB devices I work with, so I guess that explains our similar performance numbers :) May 21 09:52:20 kristrev: yes, you're not alone! :) May 21 11:15:50 kristrev: ususally flash chip datasheets should tell you the max frequency they support (though probably write and erase speed is capped at much lower frequencies, and only reads will profit from higher frequencies) May 21 11:17:17 I've never really understood how much of openwrt user space can just start up while the rootfs is still being prepared. May 21 11:17:28 where do init scripts that write things to disk go until that's happened? May 21 11:20:25 tmpfs May 21 11:23:26 which then get replaced when the rootfs finalyl finishes and are "gone" though right? May 21 11:23:33 the overlay lands on top of it? May 21 11:28:22 yeah, mount_root: jffs2 not ready yet, using temporary tmpfs overlay May 21 11:30:59 then it's switched over here https://lxr.openwrt.org/source/fstools/libfstools/overlay.c#L276 May 21 11:31:06 mount_root: syncronizing overlay May 21 11:31:10 so, ifyou write to /blah before the jffs2 is ready, it goes to tmpfs, but does that ever get written to jffs2 when it comes up? May 21 11:31:23 is that what the "synchronizing overlay" is doing? May 21 11:31:50 system("cp -a /tmp/root/upper/* / 2>/dev/null") May 21 11:35:04 kristrev: On ath79/ar71xx, currently supported max SPI frequency is around 1MHz May 21 11:36:13 wat May 21 11:36:57 ath79-spi is a bitbang driver, which means that it's performance is terrible. May 21 11:40:04 With this patch: https://github.com/openwrt/openwrt/blob/master/target/linux/ar71xx/patches-4.14/461-spi-ath79-add-fast-flash-read.patch , reading operation on flash smaller than 16M reads directly from memory-mapped area and this reading operation uses a fixed 25MHz clock. Other operations is done by bit-banging, I checked with a logic analyzer an May 21 11:40:04 d it shows that the clock speed is around 1MHz. May 21 11:43:23 gch981213: Thanks for the info. My device with a bigger flash is based on mt7620, but maybe there are similar limitations in the driver there May 21 11:43:52 kristrev: MT7620 doesn't have this problem :) May 21 11:44:17 I was under the impression the bitbanging ws only for the second spi periphs, not the first one, that has the flash attached to normally. May 21 11:44:36 thanks, I will make some more tests and see if I can find some optimizations to the driver May 21 11:44:46 or maybe there are other limiting factors May 21 12:13:19 karlp: thanks for that Lua testing May 21 12:17:10 you're welcome May 21 12:29:04 ynezz: the iwinfo bump is pushed to master May 21 13:13:13 stintel: I disabled your buildslave for now. Since it is a PPC host, it produces IB and SDK binaries unusable by x86 users May 21 13:13:22 xback: thanks May 21 13:55:48 Does anybody know how the packages' CPE_IDs are being used? May 21 14:02:09 jow: ok. is that expected behavior? May 21 14:02:23 maybe I can enable it for phase2 instead May 21 15:44:22 Is there anything special to be aware on when adding a dependency on a selection in a choice? I am trying to get libcurl to use openssl by having my package depend on @LIBCURL_OPENSSL, but I still see that mbedtls is selected May 21 15:57:46 menuconfig, go to mbedtls, press ? and look what's sleecting it. May 21 16:00:21 mbedtls is selected by several packages, but my problem is not mbedtls. My problem is making curl us openssl and not mbedtls (openssl is also enabled) using dependencies May 21 16:01:09 mbedtls is the default choice for curl SSL library, and I guess that is why depending on @LIBCURL_OPENSSL doesn't work May 21 16:09:35 don't suppose anyone hwere knows how to figure out the url to do a fossil clone from for a given fossil repo? (lua.sqlite.org in this case) May 21 16:11:23 nvm, pebcakc, have been using lua.sqlite3.org, which is some domain squatting. May 21 16:13:27 and now the damn thing downloaded everything into the current directoy. good job sqlite duders. May 21 16:22:08 Is there a way to keep `make V=s` from firing off `* Restart config...` and asking for selection of all the kernel params? May 21 16:22:20 (Or V=sc, for that matter) May 21 16:22:41 Meh, NM, only a couple needed May 21 16:25:11 make defconfig May 21 16:25:22 you must have been mangling something else for that to have been popping up May 21 16:26:54 I'll chase it down -- runs "cleanly" with my normal "make -j12 clean download world" so might be missing something in the (new) target's config files May 21 16:31:02 cotequeiroz: if you're still wondering, they're used by uscan May 21 17:05:30 Thanks mangix. That's what I thought until I grepped devscripts for cpe and found nothing. The reason I'm asking can be seen here: https://github.com/openwrt/packages/pull/9047 May 21 17:07:28 hmmmm I don't know that it will be scanned like this. I remember seeing PKG_CPE_ID in the uscan scripts May 21 17:07:43 sdwalker is not here looks like May 21 17:09:43 The perl Makefile builds 3 packages, so it will have 3 CPE_IDs. At least for the metadata used in openwrt build tools, PKG_* variables are used as defaults for the per-package * variables ... May 21 17:11:05 PKG_MAINTAINER is the default for every package's MAINTAINER in the same Makefile. PKG_CPE_ID is not handled by build system. Will a scanning tool know about this. I imagine the answer to be NO, May 21 17:11:30 ping jow May 21 17:11:31 mangix: you got a d too much, it's just swalker_ May 21 17:11:45 ah ty May 21 17:12:25 but I don't know whether it just scans for a CPE, not caring about variable name, or if it would even matter to add the CPE ids for pip and setuptools. May 21 17:13:46 I remember seeing a part of the script. It tries to find the CPE ID if PKG_CPE_ID is missing. let's see if I can find it in my logs. May 21 17:17:09 uscan only checks PKG_CPE_ID May 21 17:18:01 OK. Thanks. **** ENDING LOGGING AT Wed May 22 03:02:00 2019