| 01:22.19 | *** join/#arm-netbook fjp` (~fjp@ip4da6b463.direct-adsl.nl) |
| 08:05.59 | fjp | meh. wraplinux is i386 only :-( |
| 08:11.33 | fjp | Guess I'll have to experiment with mkimage. |
| 09:05.29 | fjp | Hmm. I think the bootpImage target in the kernel build system is probably the thing to use. |
| 11:14.14 | *** join/#arm-netbook lkcl (~lkcl@nat67.mia.three.co.uk) |
| 11:54.27 | *** join/#arm-netbook SQlvpapir (~teis@188.177.95.62) |
| 13:51.11 | fjp | lkcl: I have managed to uncompress the image (after removing the header of 13428), but I don't see any sign of an initrd in what's left. |
| 13:54.16 | fjp | But the uncompressed update image is still ~1.4 MB larger than the regular kernel image... |
| 14:01.17 | fjp | It looks as if any addition is in the early part of the file, not the end. |
| 14:05.28 | lkcl | hmm, intriguing. |
| 14:05.43 | lkcl | ohhh - of course. linux looks for the zcat header, right? |
| 14:06.20 | lkcl | and the ... yes, of course. so the early part is the cramfs, direct in memory, right at the "start" |
| 14:06.31 | lkcl | fjp: ^ |
| 14:07.34 | fjp | The two kernels are similar but not identical. |
| 14:08.15 | fjp | Sure that it is cramfs and not initramfs they are using? Know what the sig and format is of cramfs? |
| 14:08.22 | lkcl | yes. root= line says .... ok 1sec... |
| 14:09.26 | lkcl | root=/dev/mtdblock2 rootfstype=cramfs init=/linuxrc |
| 14:15.55 | lkcl | fjp: did you see the message "trailing garbage ignored" on zcat, when decompressing? |
| 14:16.03 | fjp | Where exactly did you find that? |
| 14:16.13 | fjp | Yes. Just a few 00 bytes. |
| 14:17.36 | fjp | thinks that's just the default cmdline from the kernel config. Which is not very interesting. |
| 14:19.06 | fjp | Both kernels have it, which IMO means it cannot be very relevant. And it should get overridden by the cmdline passed by u-boot anyway. |
| 14:19.25 | lkcl | ah HA! |
| 14:19.37 | lkcl | there's a _second_ zcat header :) |
| 14:19.47 | lkcl | at offset 13400 |
| 14:20.00 | fjp | Yay! |
| 14:20.01 | lkcl | sorry 0x13400 |
| 14:20.24 | lkcl | _after_ you've decompressed the image, let's try again.... |
| 14:20.47 | lkcl | which is 154 * 512 |
| 14:21.13 | fjp | wonders if that is the value set in the Makefile.boot... |
| 14:21.38 | fjp | when using the bootpImage target |
| 14:22.11 | lkcl | YES that's a tar file. |
| 14:22.48 | lkcl | ok it looks ... hmmm.. |
| 14:22.58 | fjp | $ file image.uncompressed |
| 14:22.58 | fjp | image.uncompressed: ASCII cpio archive (SVR4 with no CRC) |
| 14:23.03 | fjp | initramfs! |
| 14:23.16 | fjp | Progress........ |
| 14:23.35 | lkcl | ok it's.... oh yeah |
| 14:23.53 | lkcl | excellent |
| 14:24.34 | lkcl | ah HA veery interesting |
| 14:24.40 | lkcl | there's mention of hhtech |
| 14:24.48 | lkcl | commented out |
| 14:25.06 | lkcl | so this _is_ using the mer-smartq 2.6.24.7 code/patches |
| 14:25.36 | fjp | /etc |
| 14:25.36 | fjp | /etc/pointercal-240x320 |
| 14:25.36 | fjp | /etc/mdev.conf |
| 14:25.37 | fjp | /etc/modules.conf |
| 14:25.38 | fjp | /etc/protocols |
| 14:25.38 | fjp | /etc/hosts |
| 14:25.40 | fjp | /etc/shadow |
| 14:25.40 | fjp | /etc/MiniGUI.cfg |
| 14:25.43 | fjp | /etc/init.d |
| 14:25.43 | fjp | /etc/init.d/rcS~ |
| 14:25.45 | fjp | /etc/init.d/rcS_cp |
| 14:25.45 | fjp | /etc/init.d/rcS |
| 14:25.47 | fjp | ....... |
| 14:25.56 | fjp | cpio -it < file |
| 14:26.22 | fjp | /etc/init.d/rcS_update |
| 14:26.22 | fjp | /etc/init.d/rcS_dt_update |
| 14:26.22 | fjp | /etc/init.d/rcS_up |
| 14:26.38 | fjp | That's probably the scripts that do the update of the root fs. |
| 14:27.09 | fjp | Bingo. Now we just need to reproduce this. |
| 14:27.34 | fjp | But we know it's a compressed initramfs they're using. |
| 14:28.09 | fjp | I have the offset as 78848 (bytes?) |
| 14:29.03 | fjp | Yep. 0x13400 |
| 14:29.03 | lkcl | ack. |
| 14:29.10 | lkcl | 78848. |
| 14:29.51 | fjp | Let me try if I can recreate this using that bootpImage target |
| 14:31.54 | fjp | Hmm. An existing Makefile.boot says: |
| 14:31.57 | fjp | # Note: the following conditions must always be true: |
| 14:31.57 | fjp | # ZRELADDR == virt_to_phys(TEXTADDR) |
| 14:31.58 | fjp | # PARAMS_PHYS must be within 4MB of ZRELADDR |
| 14:31.58 | fjp | # INITRD_PHYS must be in RAM |
| 14:31.59 | fjp | <PROTECTED> |
| 14:31.59 | fjp | params_phys-y := 0x20000100 |
| 14:32.02 | fjp | initrd_phys-y := 0x20410000 |
| 14:32.31 | fjp | is slightly out of his depth here |
| 14:33.03 | lkcl | :) |
| 14:51.03 | fjp | <PROTECTED> |
| 14:51.03 | fjp | <PROTECTED> |
| 14:51.16 | fjp | (with random adresses) |
| 14:54.05 | fjp | fjp@aragorn:~/projects/ct-pc89e$ ./uncompress-image bootpImage |
| 14:54.05 | fjp | Offset is: 13660 |
| 14:54.05 | fjp | gzip: image.clean: decompression OK, trailing garbage ignored |
| 14:54.06 | fjp | fjp@aragorn:~/projects/ct-pc89e$ ./uncompress-image image.uncompressed |
| 14:54.06 | fjp | Offset is: 99232 |
| 14:54.07 | fjp | gzip: image.clean: decompression OK, trailing garbage ignored |
| 14:54.19 | fjp | That looks quite similar :-) |
| 14:55.34 | fjp | So now all we need is the correct values for zreladdr-y, params_phys-y and initrd_phys-y. |
| 14:55.45 | fjp | The second could possibly remain unset. |
| 14:56.25 | fjp | Those offsets are decimal BTW. |
| 14:59.33 | fjp | This will make for slightly strange building of D-I images: build the kernel first, then the D-I rootfs and then the kernel again :-P |
| 15:00.38 | fjp | Unless we make it so that no kernel modules are needed in the initrd. Then we can just include a standard initrd in every kernel. |
| 15:00.47 | fjp | Should not be a problem actually. |
| 15:03.32 | fjp | writes arm kernel list |
| 15:05.42 | lkcl | :) |
| 15:19.45 | fjp | Hmmm. Actually, the values _are_already defined. So why is the build complaining.... |
| 15:19.54 | fjp | - zreladdr-$(CONFIG_PLAT_S3C64XX) := 0x50008000 |
| 15:19.54 | fjp | -params_phys-$(CONFIG_PLAT_S3C64XX) := 0x50000100 |
| 15:22.23 | fjp | Ah, no. initrd_phys- is missing! |
| 15:23.57 | fjp | Common seems to be to use 800000, so 0x50080000 |
| 15:26.02 | fjp | fingers crossed... |
| 15:29.38 | fjp | Hmm. The flash from my update kernel hangs :-( |
| 15:29.47 | fjp | Reboot is fine. |
| 15:31.37 | fjp | Ah. 0 to few. |
| 15:32.23 | fjp | 0x50800000 |
| 15:33.15 | fjp | boots... |
| 15:33.42 | fjp | Kernel panic - not syncing: Attempt to kill init! |
| 15:33.54 | fjp | Yay! It works! |
| 15:33.59 | fjp | Almost that is. |
| 15:37.55 | lkcl | wha-heey |
| 15:39.30 | lkcl | ok, so let's see if i can replicate this.... it's |
| 15:40.29 | fjp | You need to add this in arch/arm/mach-s3c2410/Makefile.boot: |
| 15:40.40 | fjp | initrd_phys-$(CONFIG_MACH_CTPC89E) := 0x50800000 |
| 15:41.07 | fjp | Or in your case CONFIG_MACH_SMDK64XX |
| 15:41.09 | lkcl | ohh ok |
| 15:41.46 | fjp | And then make bootpImage with INITRD=<file.gz> |
| 15:41.53 | lkcl | okaaay |
| 15:43.12 | fjp | Eh, CONFIG_MACH_SMDK6410 |
| 15:43.20 | lkcl | so 0x50080000 or 0x50800000 ? |
| 15:43.29 | fjp | The latter |
| 15:43.39 | lkcl | ok |
| 15:43.40 | lkcl | ta |
| 15:44.29 | fjp | The race is on. What will we have first: a working kernel or a booting installer ;-) |
| 15:44.58 | lkcl | beh? this machine does not support INITRD ? |
| 15:45.00 | lkcl | beh? |
| 15:45.02 | lkcl | what have i missed |
| 15:45.13 | fjp | That line above |
| 15:45.35 | fjp | Note that it needs to be added in _24_10 |
| 15:46.29 | lkcl | oops got it |
| 15:46.35 | lkcl | okay! |
| 15:48.08 | fjp | What initrd are you using? Theirs? |
| 15:48.14 | lkcl | theirs, yep. |
| 15:48.17 | lkcl | i want to see what happens |
| 15:48.20 | fjp | That would be a good test if it works or not. |
| 15:48.45 | fjp | If that works then it's just a matter of tuning the contents of my initrd. |
| 15:49.07 | lkcl | seems so! |
| 15:49.25 | lkcl | ok diffficult to tell... |
| 15:51.09 | lkcl | try again.. ok, couldn't find valid ramdisk at 0 |
| 15:52.23 | fjp | Did you include the tarball on your SD? |
| 15:52.35 | lkcl | lemme get the right file, first, as the initrd :) |
| 15:52.44 | fjp | Duh! |
| 15:53.22 | lkcl | found |
| 15:53.37 | lkcl | ok compressed image found |
| 15:53.45 | lkcl | whoops what fstype do i need? |
| 15:53.59 | fjp | none. |
| 15:54.00 | lkcl | tried ext ext4dev cramfs msdos vfat romfs |
| 15:54.04 | lkcl | beh? |
| 15:54.20 | fjp | Huh? What exactly did you do? |
| 15:54.47 | fjp | initramfs is a cpio archive; it does not have a specific file system |
| 15:54.52 | lkcl | put the initrd compressed we extracted earlier |
| 15:55.06 | lkcl | hmmm... will try the uncompressed one next |
| 15:55.42 | fjp | Should be the compressed one I think. |
| 15:55.57 | lkcl | shrugs :) |
| 15:56.28 | lkcl | yep it should be compressed |
| 15:56.32 | lkcl | you're right |
| 15:56.33 | fjp | What are you seeing? Does the kernel run init and fail after, or does it fail in the boot itself. |
| 15:57.20 | lkcl | 1sec... |
| 15:57.30 | lkcl | lemme do it again... |
| 15:57.59 | fjp | should probably just try himself :-) |
| 15:58.19 | lkcl | it looks for a partition |
| 15:58.31 | lkcl | RAMDISK: Compressed image found at block 0 |
| 15:58.34 | lkcl | List of all partitions: |
| 15:58.51 | lkcl | No filesystem could mount root, tried: ext3 ext4dev cramfs msdos vfat romfs |
| 15:59.08 | lkcl | Kernel panic - not syncing: VFS unable ... unknown-block(0,0) |
| 15:59.42 | fjp | Hmm. That's at least different from what I get. |
| 16:00.02 | fjp | Which makes me think you are past init. But not sure. |
| 16:02.18 | lkcl | mrmmmhh.. |
| 16:06.00 | fjp | Maybe, just maybe, all I need is to create a symlink linuxrc->init... |
| 16:07.03 | fjp | Fun. They have a few command_*.txt files for fdisk, i.e. 'fdisk <command_240.txt' |
| 16:07.32 | lkcl | yes i saw |
| 16:07.42 | fjp | init is a direct symlink to busybox |
| 16:07.58 | lkcl | or, in kernel/printk.c or elsewhere you could just modify the boot parms hard-coded :) |
| 16:08.28 | lkcl | i'm not past init. this is rootfs failing to be recognised. |
| 16:09.14 | fjp | Weird. Is that a kernel that boots correctly without initrd? |
| 16:09.30 | fjp | tries the symlink |
| 16:10.25 | lkcl | fjp: yep. |
| 16:10.52 | lkcl | so it's showing no partitions, after decompress, hmmm what boot params you using? |
| 16:11.11 | fjp | Nothing special. |
| 16:11.34 | fjp | Oh! I've unset the default boot params option in the kernel config. |
| 16:11.42 | lkcl | ah? |
| 16:11.44 | lkcl | ahh |
| 16:12.01 | fjp | Not sure if that should make a difference though. |
| 16:12.05 | lkcl | which config option? |
| 16:12.08 | lkcl | CONFIG_xxxx |
| 16:12.25 | fjp | CMDLINE IIRC |
| 16:12.50 | lkcl | ok |
| 16:15.06 | fjp | Can you also try setting that to their values? |
| 16:15.50 | fjp | AFAIK support for initramfs should be standard and the kernel should start /init if nothing else is specified. |
| 16:15.50 | lkcl | ok |
| 16:15.56 | lkcl | root=/dev/ram |
| 16:16.19 | fjp | Yep. That's default too. |
| 16:16.37 | fjp | At least, we don't explicity set it for D-I anymore. |
| 16:17.08 | lkcl | i have root=/dev/mtdblock2 in CONFIG_CMDLINE |
| 16:17.20 | fjp | That may explain. |
| 16:18.00 | lkcl | nope |
| 16:19.17 | lkcl | ok let's try mounting the dt fs, recreating it as ext3. |
| 16:19.27 | lkcl | how do i mount that cramfs on loopback? |
| 16:19.46 | fjp | It's not a cramfs... |
| 16:20.05 | fjp | It's a cpio archive. All you can do is extract it, not mount it. |
| 16:20.21 | lkcl | ahh. |
| 16:20.30 | lkcl | cpio?? ok. hmm... |
| 16:21.02 | lkcl | then how the heck did you manage to... ? ok :) |
| 16:24.03 | lkcl | FUCK me fucking cpio. |
| 16:24.25 | lkcl | fucking overwriting stuff in / on my desktop |
| 16:27.16 | lkcl | *sigh* |
| 16:27.54 | lkcl | fjp: can you recall the dpkg command to verify files on a debian system, to check what's been overwritten? |
| 16:28.09 | fjp | cpio -i --no-absolute-filenames <archive |
| 16:28.22 | fjp | YES! cpio is seriously fucked. |
| 16:28.24 | lkcl | no, i mean: if you haven't used --no-absolute-filenames |
| 16:28.33 | fjp | Did you run as root? |
| 16:28.38 | lkcl | yyyep |
| 16:28.46 | fjp | OUCH! |
| 16:29.34 | fjp | Run md5sums using files in /var/lib/dpkg/info/*.md5sums |
| 16:29.53 | fjp | There may be a command to do that, but I don't know that... |
| 16:30.42 | lkcl | debsums |
| 16:30.52 | fjp | yep. |
| 16:34.34 | lkcl | fuckme /bin/rm has been overwritten. going to need to get coreutils .deb and dpkg-deb -X it |
| 16:39.34 | lkcl | ok getting there. coreutils, busybox, bzip, e2fsprogs... |
| 16:42.54 | lkcl | halt (!!) |
| 16:42.57 | lkcl | sysvinit. ... |
| 16:42.59 | lkcl | *sigh* |
| 16:43.10 | fjp | I made the same mistake, but was protected by not running as root :-P |
| 16:44.47 | lkcl | ha ha |
| 16:44.54 | lkcl | debsums is good |
| 16:47.07 | lkcl | mount, procps... gaah :) |
| 16:50.03 | lkcl | ok, i'm happy with that. |
| 16:50.05 | lkcl | rrrright. |
| 16:54.57 | lkcl | nearly there... |
| 16:55.41 | mnemoc | go go go |
| 16:56.46 | lkcl | :) |
| 16:57.08 | lkcl | boot... |
| 17:01.15 | lkcl | made it an ext3 |
| 17:01.19 | lkcl | err |
| 17:01.24 | lkcl | beh? |
| 17:01.35 | lkcl | looks like vaguely... |
| 17:01.56 | lkcl | yeah.... looks like it's working |
| 17:01.57 | fjp | did you remove the extra scripts from /etc/rcS ? |
| 17:02.27 | lkcl | i modified /etc/init.d/rcS to put "echo hello" and reboot |
| 17:02.39 | lkcl | but.... i think i modified the wrong file... |
| 17:03.34 | lkcl | so it's currently churning through a tar -xvzf |
| 17:05.31 | fjp | As long as it's not doing the fdisk... |
| 17:05.48 | fjp | OTOH, it will probably do that on a device that does not exist. |
| 17:07.16 | lkcl | ok let's create a /dev/tty and /dev/tty1... |
| 17:07.28 | lkcl | assume echo didn't work because i should have put /bin/echo |
| 17:08.46 | lkcl | but - yeah, reformatting as ext3 works fine. |
| 17:09.09 | fjp | Their initrd seems to be working for me. |
| 17:09.42 | fjp | At least, I get no errors after the last boot messages about mmcblk0 |
| 17:10.08 | fjp | No idea if it's actually doing anything though... |
| 17:12.17 | lkcl | yehh we need to get something on-screen, to prove it's working |
| 17:12.30 | lkcl | i'm trying mknod /dev/fb0 then echo "hello" > /dev/fb0 |
| 17:12.33 | lkcl | my favourite |
| 17:14.53 | fjp | My god. What truly horrible scripts in /etc/init/ |
| 17:15.38 | fjp | guesses it will take a while to extract the tarball. |
| 17:15.47 | fjp | And as our poweroff does not work..... |
| 17:16.27 | fjp | Should be able to tell after a reboot though as I'll no longer have my wireless configured. |
| 17:16.56 | lkcl | yes - something like that |
| 17:17.02 | lkcl | and yes they are :) |
| 17:22.17 | lkcl | ok i don't believe this is working. |
| 17:22.44 | fjp | Yay! It did flash the rootfs. |
| 17:22.53 | lkcl | yes? excellent! |
| 17:23.17 | lkcl | ok so what's different about the setup that you have, neh? |
| 17:23.22 | fjp | Eh, maybe not... |
| 17:23.42 | fjp | The system date was 2010, but wireless still works. |
| 17:25.23 | fjp | And some files in /root/ are still there :-( |
| 17:25.52 | fjp | Bah. Enough for today. Still feels like progress. |
| 17:33.19 | fjp | Right. We have: |
| 17:33.20 | fjp | 50000000-5f1fffff : System RAM |
| 17:33.20 | fjp | <PROTECTED> |
| 17:33.21 | fjp | <PROTECTED> |
| 17:33.50 | fjp | So 0x50800000 is in RAM and does not conflict with the kernel. |
| 17:34.12 | fjp | But we could easily try a higher value. |
| 17:55.58 | lkcl | excellent |
| 17:56.01 | lkcl | good stuff. |
| 18:02.26 | fjp | And the command line that is used is the one from u-boot. From dmesg with both original and custom kernel: |
| 18:02.28 | fjp | Kernel command line: root=/dev/mmcblk0p2 rootwait init=/init console=ttySAC0,115200 |
| 18:09.34 | fjp | Ahhh. |
| 18:09.58 | fjp | The hard size limit for the uncompressed kernel |
| 18:09.58 | fjp | se limitations this target is rarely used in practice. |
| 18:10.20 | fjp | Eh, I mean: |
| 18:10.33 | fjp | The hard size limit for the uncompressed kernel and initrd of 4Megabytes applies. Because of these limitations this target is rarely used in practice. |
| 18:10.59 | fjp | And kernel + my D-I initrd is quite a bit larger... |
| 18:12.20 | fjp | Their uncompressed kernel image is 4187476 |
| 18:13.20 | fjp | Mine is 4519901. |
| 18:14.05 | fjp | Which is 4.3MB |
| 18:14.31 | fjp | So I'll need to trim quite a bit... |
| 18:15.06 | fjp | The kernel itself is 2943400, so there should be enough space... |
| 18:15.52 | fjp | Good documentation: http://www.simtec.co.uk/products/SWLINUX/files/booting_article.pdf |
| 19:12.08 | fjp | Woot! |
| 19:12.30 | fjp | "udev requires a kernel >= 2.6.26, not started. |
| 19:13.03 | fjp | mount: mounting none on /proc failed: Device or resource busy |
| 19:13.05 | fjp | " |
| 19:13.24 | fjp | It's running init from my D-I image! |
| 19:14.14 | fjp | It then panics again, but that's not important :-) |
| 19:14.45 | fjp | lkcl: ^^^^^^^^^^^^^^ |
| 19:16.10 | fjp | So now I just have to create an installation method that has a very small initial initrd and then loads all additional udebs from the SD card. |
| 19:16.27 | fjp | Something like our old floppy installation methods. |
| 19:22.42 | lkcl | ack |
| 19:23.08 | lkcl | thaat |
| 19:23.11 | lkcl | yaay! |
| 19:23.19 | lkcl | well doone |
| 19:23.44 | lkcl | there's a way to increase that size |
| 19:23.56 | lkcl | i managed it before, did a 12mb initrd for the HTC Blueangel |
| 19:27.06 | fjp | OK. But their initrd is 1939536; the one I used now (extremely stripped - too much) is only 1272113. |
| 19:27.19 | fjp | So there's room to play with. |
| 19:27.38 | lkcl | excellent |
| 19:27.40 | fjp | But as kernels seem to get bigger and bigger and bigger... |
| 19:27.56 | fjp | It would be nice to have more room to play with. |
| 19:30.42 | fjp | could really use a working keyboard now :-) |
| 19:34.32 | lkcl | yehh... |
| 19:36.15 | fjp | No wonder their initrd is so large. They have a full libc in there... |
| 19:40.36 | fjp | Eh, hmm. Did you do a 12MB initrd piggybacked onto the kernel or just a 12 MB initrd that's loaded using initrd=? The latter is not a problem. |
| 19:41.04 | fjp | The problem is that the kernel size (which now includes the initrd) is limited to 4MB. |
| 19:53.23 | lkcl | a 12mb initrd iirc |
| 19:53.25 | lkcl | ohh ok |
| 19:53.51 | lkcl | yes and rather than symlinks to busybox, applications dropped into / |
| 19:54.48 | fjp | It not a huge problem. D-I is amazingly modular and flexible. Just takes a bit of creativity. |
| 22:20.39 | lkcl | yes. phil loved explaining to me how he would come up with something for the unattended installer, and joey would simplify it by _removing_ bits of d-i, _not_ putting in what phil did, but replacing it with something else even more powerful and compact :) |
| 22:21.13 | lkcl | eighteen months of that, and phil's unattended installer modifications entirely disappeared and got sucked into d-i |
| 22:21.23 | lkcl | ok. |
| 22:21.26 | lkcl | sleep for me |
| 22:21.32 | lkcl | 3G modem goes off. |
| 22:21.34 | lkcl | night folks. |