**** BEGIN LOGGING AT Tue Jan 05 01:02:38 2010 Jan 05 01:03:24 oh, kernel finally started boot, but I think it will ne never finished due to ubi warnings Jan 05 01:22:10 JaMa, cmair, PaulFertser and all who can be interested, I confirm that I have booting shr-u default kernel from ubifs rootfs Jan 05 01:22:37 configure process on go, of course very slow because of debug info Jan 05 01:23:07 does Qi support ubifs? Jan 05 01:23:46 plz disable all ubi debug except CONFIG_MTD_UBI_DEBUG=y and CONFIG_MTD_UBI_DEBUG_DISABLE_BGT=y it will keep kernel ubifs working, and no so slow Jan 05 01:24:05 tmzt, I replaced jffs2 mtd6 with ubifs, not whole nand Jan 05 01:24:11 ok Jan 05 01:24:27 but would it work so that kernel could be loaded from Qi? Jan 05 01:24:29 so if Qi will be compiled with correct params for kernel, kernel can be booted from nand Jan 05 01:24:36 yes Jan 05 01:24:53 parameters not filesystem? Jan 05 01:25:20 ubifs as rootfs on mtd6 doesn't touch kernel boot at all Jan 05 01:25:48 just, rootfstype=ubifs ubi.mtd=6,2048 root=ubi0:rootfs Jan 05 01:26:16 kernel will start ubi, mount rootfs, and so on Jan 05 01:31:24 max_posedon: how did you install it? ubiupdatevol from a system on sd card? Jan 05 01:33:26 boot from sdcard, flash_eraseall /dev/mtd6, ubiformat /dev/mtd6 -s 2048 -O 2048, ubimkvol /dev/ubi0 -N rootfs -m, mount -t ubifs ubi0:rootfs /mnt Jan 05 01:33:49 I think ubitools for some reason doesn't fit kernel implementation Jan 05 01:34:02 ok, that might be possible. Jan 05 01:34:03 and as much as possible should be done by kernel Jan 05 01:34:18 but plz, don't try it with default shr) Jan 05 01:34:27 phone boots alreay 30m Jan 05 01:34:44 and I think its need 2h more) Jan 05 01:34:45 :D Jan 05 01:34:59 2k blocks? Jan 05 01:35:06 cool, this might work on msm Jan 05 01:35:36 ubi0 is a block device? Jan 05 01:35:41 nope Jan 05 01:35:48 its MTD device Jan 05 01:35:58 not block, not char Jan 05 01:36:00 hmm, but it's virtual? Jan 05 01:36:30 yep, virtual mtd "block" device Jan 05 01:37:27 ubi is something like lvm. You get a volume on top of your mtd device. Jan 05 01:37:48 ubifs is like lvm, on top on ubi device Jan 05 01:37:49 yes, I was going to say why not use lvm/dm for this? Jan 05 01:38:13 because lvm works on top of block devices. Jan 05 01:38:19 ubi is mtd layer implementation, which helps as use nand very effitiently Jan 05 01:38:33 I don't think that is like lvm Jan 05 01:38:46 its more complicated, and have different idea Jan 05 01:38:57 max_posedon: any idea why shr doesn't boot properly? Jan 05 01:39:04 PaulFertser, its boots! Jan 05 01:39:06 It represents logical volumes on top of a (managed) mtd device. Jan 05 01:39:21 04:34 < max_posedon> phone boots alreay 30m Jan 05 01:39:24 PaulFertser, just very-very-very slow, because ubi full debug maked it really slow Jan 05 01:39:29 max_posedon: ah, i see Jan 05 01:39:51 Just cut power and disable debug. Ubi should survive it :) Jan 05 01:39:52 I even can ping it now, and nmap found rpcbind service Jan 05 01:40:13 cmair, I want give it a chanse to have a phone tomorrow) Jan 05 01:41:00 PaulFertser, plz keep only CONFIG_MTD_UBI_DEBUG=y and CONFIG_MTD_UBI_DEBUG_DISABLE_BGT=y for debuggin, (and 2nd one is really important!) Jan 05 01:41:13 if you can edit def config or smth like this Jan 05 01:41:32 I hope we will have nice shr-aware-ubifs kernel tomorrow Jan 05 01:41:34 i must say that i'm quite happy that someone is trying ubi on the freerunner. i couldn't think of a better way to illustrate that nand is just too insanely complex to use ;-))) Jan 05 01:43:18 I just don't understand the issue about the bg thread. My FR is still working (but with .32). Jan 05 01:43:48 cmair, I don't too Jan 05 01:44:03 but I have same problem with vanilla kernel on sheevaplug I think (not sure) Jan 05 01:44:07 (.320 Jan 05 01:44:10 (.32) Jan 05 01:44:20 and without it I see segfault during mount sometimes Jan 05 01:44:59 wpwrak: otoh i'm not sure built-in controller in SDs use NAND as effectively as this new ubi implementation Jan 05 01:45:36 PaulFertser: oh, i think ubi gets a lot of things right. perhaps more than those built-in controllers. Jan 05 01:46:14 ubi have a lot of ram, SDs have only about ~200k ram, and its not enough for firmware Jan 05 01:46:19 PaulFertser: but then, your sd card will never present bare nand, so love them or hate them, those controller are there when you're using significant amounts of flash. Jan 05 01:46:48 (I was near becoming a SD firmware developer) Jan 05 01:46:55 max_posedon: are you sure SHR kernel has any UBI support? I can't see anything in defconfig. Jan 05 01:47:02 SD? Jan 05 01:47:21 PaulFertser, yes, SHR have a patch, JaMa confirmed it Jan 05 01:47:25 wish there were some open projects to understand how they work Jan 05 01:47:30 max_posedon: At my workplace we have a device using kernel 2.6.32-rc8 running ubifs without errors since a month or so. They don't have heavy I/O going on, but they survived lots of reboots with the bg thread running. Jan 05 01:47:33 tmzt, SD, SSD, (Sandisk) Jan 05 01:47:35 other than EyeFi Jan 05 01:48:04 cmair: would you swap over ubi? Jan 05 01:48:11 with or wothout ubifs Jan 05 01:48:17 max_posedon: hm, why JaMa didn't add it to the OM kernel directly? Jan 05 01:48:31 max_posedon: so you ran in time :) interesting, i wouldn't even have expected as much as 200 kB Jan 05 01:48:44 PaulFertser, because nobody before me can't confirm it working Jan 05 01:48:56 people use .32, JaMa had problems Jan 05 01:49:11 but I defenetly see booting sustem, just very slow) Jan 05 01:49:45 tmzt: if you really need swap on nand, ubi should help you there, but I would not recommend it. I also did not try it yet. Jan 05 01:49:45 wpwrak, I'm not sure that its correct for all such devices, but at least people says it) Jan 05 01:49:51 * wpwrak wonders by what time our GTA01/GTA02 user base will have shrunk to the point where a mandatory migration to SD-only would be feasible Jan 05 01:50:22 people = developers who doing firmware Jan 05 01:51:20 max_posedon: firmware for sd cards must be fun. i'm glad others are doing it for us :) what do they say - do they trust it ? Jan 05 01:52:40 they trust and have big test center and love their work, and say thats a lot of place for improvments) Jan 05 01:53:17 but you should understand they mainly work to improve performance with ntfs/fat on top of it Jan 05 01:54:40 ntfs? Jan 05 01:54:47 max_posedon: trust is good :) my main worry about sd is that it may have catastrophic failure modes when power fails and such Jan 05 01:54:56 hey, I just want to be able to secure erase a card Jan 05 01:55:00 (all ones) Jan 05 01:55:05 without xdestroying ftl Jan 05 01:55:27 tmzt, I suggest writing random data, not zeroes!) Jan 05 01:55:32 max_posedon: there are rumors of such failures, but then other people are running massive long-term tests that don't end in disaster. so i'm not sure what the truth is. Jan 05 01:55:34 freesmartphone.org: 03Frederik.Sdun 07cornucopia * r4427f29128b4 10/fsomusicd/src/playlist.vala: fsomusicd: implement load_from_uri Jan 05 01:55:50 max_posedon: for a phase change type medium? Jan 05 01:55:56 max_posedon: ... which of course puts me into a bit of a dilemma on my crusade against bare nand :) Jan 05 01:56:01 only ones get written I think Jan 05 01:56:20 wpwrak, whey work for sandisk, you know, all firmware companies don't share info Jan 05 01:56:31 if its true for one, it can be false for others Jan 05 01:56:33 also, there's at least one device worked on in #mer where power failure leads to nand failure Jan 05 01:56:38 it's not hare nand Jan 05 01:56:47 but embeddedsd or similar Jan 05 01:57:44 max_posedon: well, sandisk are a good enough data point for me. you can get them everywhere and they generally have a good reputation. Jan 05 01:58:41 tmzt: well, bare nand has no such catastrophic failure mode. its failure are all "local" (a block of so) Jan 05 01:58:42 tmzt: eMMC Jan 05 01:58:47 I think if I'll sucked with current work, I'll become developer, and than (and if) I'll be able answer more your questions. Or just after convesations with them) Jan 05 01:58:56 tmzt: the problem is that nand can have lots of failures. so you need to deploy that gazillion of defense mechanisms. Jan 05 01:59:30 max_posedon: i do not see a reason to not expose some interface to bare NAND since a real OS can manage all those task better. Jan 05 01:59:57 max_posedon: if you could perhaps throw "catastrophic failure" into the discussion after a few beers ... ? ;-) Jan 05 01:59:59 but you break compatibility in non sd mode Jan 05 02:00:09 PaulFertser: exposing interface requires ressources, ie someone to garantee the API/ABI, and also maybe more space in firmware Jan 05 02:00:25 wpwrak: that's what I'm saying Jan 05 02:00:34 PaulFertser: oh right, was that smartq? Jan 05 02:00:47 PaulFertser: at the cost of handing the responsibility over to you ... Jan 05 02:00:59 PaulFertser, I'm not sure about nand in our devices, but regular ssd/mmc have upto 8-16-32 "write proccesses" Jan 05 02:01:26 hmm? Jan 05 02:01:32 like running threads? Jan 05 02:01:41 are these 9051s or mips? Jan 05 02:01:45 8051s Jan 05 02:03:06 tmzt: yes Jan 05 02:03:54 PaulFertser: that means your kernel, your boot loader, any maintenance tools that need to know about it, and so on Jan 05 02:03:54 max_posedon: what do you call a "write process" ? Jan 05 02:03:54 (write/read) Jan 05 02:03:55 its not so easy for cpu Jan 05 02:03:56 and can become a bottleneck Jan 05 02:03:57 but, may be I'm wrong and our cpu already do it Jan 05 02:03:58 have we some open ssd/flash with ability to read/write firmware sources? Jan 05 02:03:58 I afraid of we not Jan 05 02:04:05 max_posedon: so do you think http://cgit.openembedded.org/cgit.cgi/openembedded/tree/recipes/linux/linux-openmoko-shr-devel/0007-Enable-UBI-UBIFS.patch should be applied to andy-tracking branch except for the debugging options? Jan 05 02:04:16 wpwrak: why not expose both interfaces? Jan 05 02:04:46 max_posedon: what cpu? 8051/mips? Jan 05 02:05:20 tmzt, I don't know; wpwrak just mini-controller-task Jan 05 02:05:49 PaulFertser: they would yield divergent results. in theory, why not. in practice, probably more trouble and worth for the card manufacturers Jan 05 02:05:55 internal clock or bus clocked? Jan 05 02:05:59 PaulFertser, yes Jan 05 02:06:13 but plz keep CONFIG_MTD_UBI_DEBUG=y and CONFIG_MTD_UBI_DEBUG_DISABLE_BGT=y Jan 05 02:06:22 all other debug can be disabled Jan 05 02:06:33 I think CONFIG_UBIFS_FS_DEBUG=y CONFIG_UBIFS_FS_DEBUG_MSG_LVL=0 CONFIG_UBIFS_FS_DEBUG_CHKS=y Jan 05 02:06:36 is ok too Jan 05 02:06:42 max_posedon: the recipe will fail to build then. Why do you need those debugging options at all? Jan 05 02:07:07 I'm not sure that ubifs will work without it Jan 05 02:07:15 mainly without CONFIG_MTD_UBI_DEBUG_DISABLE_BGT=y Jan 05 02:07:27 I had problems without tham on sheevaplug Jan 05 02:07:34 background tasks? Jan 05 02:07:41 tmzt, yep Jan 05 02:07:58 anybody futz/fuzz mmc/sd? Jan 05 02:08:06 like the old sc's? Jan 05 02:08:09 PaulFertser: also, they would have to document what exactly is beneath the interface. might be innovative stuff that patent trolls could use, etc. Jan 05 02:08:09 max_posedon: quite a lot of processes Jan 05 02:08:33 yeah, sandisk has some power management patents Jan 05 02:08:42 without them sd would be impractical Jan 05 02:08:49 wpwrak, this is how ssd hardware devs improving speed now I think Jan 05 02:09:47 hey, any reason an sd couldn't start out doing lpc decode? Jan 05 02:10:00 then switch to nand with a command? Jan 05 02:11:01 max_posedon: you need processes/threads of there's something that you're waiting for and there's something else you can do while waiting. how is the nand in sd structured that you have this sort of situation ? e.g., with a single nand block, there wouldn't be much point in having multiple threads. maybe two or three, one for foreground activity, the rest for housekeeping. Jan 05 02:12:04 hm, yep, maybe Jan 05 02:12:29 I would guess block groups, given how the power domains must work Jan 05 02:12:39 also, write is a lot harder than read Jan 05 02:12:57 max_posedon: lpc? Jan 05 02:13:38 tmzt, I just don't know, don't know much Jan 05 02:14:03 it would be cool, you could boot a pc from nand without nor/pflash part Jan 05 02:14:15 but nobody does it Jan 05 02:15:22 max_posedon: http://paste.debian.net/55630/ Jan 05 02:15:36 tmzt, don't linux bios projest just do it? Jan 05 02:16:17 max_posedon: it would be nice if you provide exactly working and checked configuration and i'll commit it. Jan 05 02:16:20 no, they don't make motherboards Jan 05 02:16:28 and it's coreboot now Jan 05 02:16:56 I just watch as bom's fall evrywhere but pc Jan 05 02:17:05 but pc prices (netbooks) fall Jan 05 02:17:15 PaulFertser, from my sheevaplug http://pastebin.ca/1738257 Jan 05 02:18:00 we're still using nor, we still boot from ssd etc. Jan 05 02:18:08 what's an ssd but nand+controller Jan 05 02:18:19 but it needs a sata phy? Jan 05 02:18:28 it can't be on the bus? Jan 05 02:18:30 tmzt, sheevaplug boots from nand directly I think Jan 05 02:18:37 interesting Jan 05 02:18:43 what cpu? Jan 05 02:18:55 e.g. it boots uboot, which doing all other work Jan 05 02:19:10 Feroceon 88FR131 rev 1 (v5l) Jan 05 02:19:23 but how does it boot uboot? Jan 05 02:19:27 which is? Jan 05 02:19:34 I thought it was arm Jan 05 02:20:01 yes its arm, I don't know so low-level stuff Jan 05 02:20:15 but I think cpu open enough to be able find docs Jan 05 02:20:50 PaulFertser, so, I can't provide excactly working andy-trucking configuration now Jan 05 02:20:56 tmzt: s3c6410 boots directly from SD Jan 05 02:21:08 using rom bootloader? Jan 05 02:21:22 max_posedon: are you ok with waiting till morning when you provide the configuration? I'll commit it then. Jan 05 02:21:28 tmzt: From what I know, the CPU contains a very small bootloader which initializes nand and loads the next bootloader from there. Jan 05 02:21:36 tmzt: yes, i think that's mask-rom in SoC. Jan 05 02:21:42 you will not get them at morning, I'm going to sleep) Jan 05 02:21:47 lpc would let a modern x86 fetch the rom bootblock over lpc bus Jan 05 02:21:53 at an instruction level Jan 05 02:21:57 just rom decode Jan 05 02:22:23 PaulFertser, I suggest just wait for andy-trycking, I'll touch shr people Jan 05 02:22:23 no stage0 in rom Jan 05 02:22:35 [we] will test before Jan 05 02:23:08 max_posedon: after testing the configuration can go directly to andy-tracking, i see no reason to have additional patches in OE. Jan 05 02:23:19 ok Jan 05 02:23:37 but, I can't do it today, may be tomorrow at evening I'll play with kernel compilation Jan 05 02:23:51 max_posedon: ok Jan 05 02:24:02 max_posedon: i've just woken up btw ;) Jan 05 02:32:28 PaulFertser: also the 2450 has that irom. seems that all modern samsung s3c do. Jan 05 02:33:27 wpwrak: now they make p5c or smth Jan 05 02:35:29 wpwrak: if SD vendors really optimize for fat or ntfs that might result in suboptimal results for those who doesn't want to use crappy FS's. Why not just expose two interfaces concurrently: raw NAND interface and traditional SD? Without providing any info, it should be like if one uses raw interface traditional just doesn't work. Jan 05 02:35:56 because the sd would be useless to anyone else Jan 05 02:36:00 max_posedon: (sandisk) any chance you have inside contacts in sandisk? Jan 05 02:36:30 tmzt: to anyone who? You just compile e.g. ubifs support in your kernel and use it... Jan 05 02:36:35 PaulFertser, I have contacts inside people who developing firmware (outsourcing) not in sandisk directly Jan 05 02:36:50 PaulFertser: the external sd would appear corrupted Jan 05 02:36:56 it wouldn't even respond Jan 05 02:37:02 max_posedon: i mean it would be nice to help people who "bricked" their internal eMMC in SmartQ's. Jan 05 02:37:22 he, I'd just be hAppy to know how many/what cpu cores there are Jan 05 02:37:39 maybe what commands aren't documented Jan 05 02:37:46 tmzt: the controller can respond something meaningful over the traditional interface even after switched to "raw" mode. Jan 05 02:37:47 as sdmi is deader than dead Jan 05 02:38:04 the original reasons for restricting this makes no sense Jan 05 02:38:34 PaulFertser: what? if you use ubi or other ftl the orginal fw won't read the ftl and present a block interface Jan 05 02:38:41 you assume it can format itself Jan 05 02:38:51 given the rom/ram that's not likely Jan 05 02:38:52 I just saw, "Init runlevel 3", looks like its forewer task) Jan 05 02:39:05 PaulFertser, to you think they can help? Jan 05 02:39:24 who makes the chip in smartq? Jan 05 02:39:35 e.g. sandisk outsource algorithmics parts (as I know) Jan 05 02:39:36 tmzt: i mean SD vendors should alter their firmware to present a choice to the user. Jan 05 02:39:46 max_posedon: if they're involved in eMMC, definetely. Jan 05 02:40:13 ok, I'll try to ask Jan 05 02:40:17 but making the card no longer an SD seems ineffective Jan 05 02:40:21 max_posedon: looks like eMMC got disabled when it was low on power and can't enable itself again the usual SD way. Jan 05 02:40:48 also, this raw nand interface would likely have to involve programming power/clock distribution and other lowlevel things Jan 05 02:41:02 things the companies don't expose Jan 05 02:41:06 does anybody have recent shr-U kernel's configs right now? Jan 05 02:41:18 looks like I want try compile kernel Jan 05 02:41:28 tmzt: aren't that usual documented NAND chips inside the SDs? Jan 05 02:41:49 max_posedon: it's andy-tracking packaging defconfig with UBI patch. Jan 05 02:41:54 what documented nand chips? Jan 05 02:42:23 tmzt: i haven't disassembled any SD but i assume they use the same chips one can buy for embedding. Jan 05 02:42:48 with a separate controller chip? Jan 05 02:42:57 also, those are block level? Jan 05 02:43:09 how do you know they don't have firmware of sometype? Jan 05 02:43:43 I'm trying to think how poweron, decode, erase, write, poweroff cycle might work Jan 05 02:43:48 tmzt: at least i think i've read about usb mass storage that they're using regular NAND with common interface, just like the one used in freerunner. Jan 05 02:44:07 usb maybe, since the if isbmore complex Jan 05 02:44:08 guys! X started!) Jan 05 02:44:16 on 32? Jan 05 02:44:23 which driver? Jan 05 02:44:48 tmzt, I still first time boot phone with ubifs) Jan 05 02:45:06 which X driver Jan 05 02:45:09 max_posedon: Are you planning some speed measurement? Jan 05 02:45:10 drm one or fbdev Jan 05 02:45:29 tmzt, shr-unstable-recent-default, I don't know) Jan 05 02:45:30 max_posedon: on ubifs i mean Jan 05 02:45:47 gena2x, ubifs is faster than jffs2, really faster Jan 05 02:45:57 I just shoul disable some debug info Jan 05 02:46:02 maybe faster than yaffs? Jan 05 02:46:12 defenetly Jan 05 02:46:34 http://www.plugcomputer.org/plugwiki/index.php/Enabling_UBIFS Jan 05 02:46:49 this is tests on sheevaplug, but... you may find them interesting too Jan 05 02:50:55 max_posedon: btw, i noticed strange option in kconfig: CONFIG_MTD_NAND_VERIFY_WRITE Jan 05 02:51:58 max_posedon: it makes kernel to rereall alldata it wrote as far as i understanding Jan 05 02:52:18 I'm happy http://pastebin.ca/1738295 !!) Jan 05 02:52:30 gena2x, I don't want touch so low-level now Jan 05 02:52:39 I just happy to have running system) Jan 05 02:52:51 max_posedon: congrats Jan 05 02:53:10 29-3? Jan 05 02:53:20 .32 should work now? Jan 05 02:54:43 PaulFertser, I'm a bit lost, last commit to andy-tracking branch was done on 26Nov, is it correct branch to use? Jan 05 02:54:49 gena2x, I use .32 on sheevaplug Jan 05 02:55:03 max_posedon: yes, it's in maintenance-only mode. Jan 05 02:57:59 and I hope latest(sorry, my mind becomeing crazy), as I understand shr only patches config for ubi, thay haven't patches some bugs and so on, yes? Jan 05 02:58:42 max_posedon: yes, only ubifs **** ENDING LOGGING AT Tue Jan 05 02:59:56 2010