Buildroot a linux 2.6.26-lab126 kernal. (3.3)

[geekmaster]

The amazon GPL source code contains patch files to be used if you replace any source files with unpatched versions. Perhaps some of those patches may be needed when using newer source code too (with suitable modification).

Before downgrading to an older version, it may be worth seeing if one of amazon's patches makes the newer version work.

[NiLuJe]

@knc1: And to expand on that, GCC's auto-vectorization isn't enabled at anything lower than -O3, IIRC .

[twobob]

Works with the 2006q3 CS TC as knc1 said. I feel this is probably the best route. I'm having trouble with the patches applying as the build directory in BR is "./built/linux-custom"

I have also amended it to do the:

Spoiler:

@@ -52,9 +52,7 @@
 # CONFIG_RELAY is not set
 # CONFIG_NAMESPACES is not set
 CONFIG_BLK_DEV_INITRD=y
-CONFIG_INITRAMFS_SOURCE="../../initramfs/skeleton-luigi.list ../../initramfs/image/"
-CONFIG_INITRAMFS_ROOT_UID=0
-CONFIG_INITRAMFS_ROOT_GID=0
+CONFIG_INITRAMFS_SOURCE=""
 CONFIG_CC_OPTIMIZE_FOR_SIZE=y
 CONFIG_SYSCTL=y
 # CONFIG_RADIX_TREE_CONCURRENT is not set
@@ -328,7 +326,7 @@

in arch/arm/configs/imx35_luigi_defconfig but I'm still fumbling my way through using patch and BR currently requires a typed in 'fix' to reference the correct file. I'll RTFM - again.

Anyways: the modules look much happier now:

[root@kindle root]# insmod /mnt/us/modules/smbfs.ko

[root@kindle root]# lsmod
Module Size Used by
smbfs 64800 0

Narry a moan

[root@kindle root]# rmmod smbfs
[root@kindle root]# strace -o /mnt/us/opt/insmod.trace insmod /mnt/us/modules/smbfs.ko

Spoiler:

execve("/sbin/insmod", ["insmod", "/mnt/us/modules/smbfs.ko"], [/* 21 vars */]) = 0
brk(0)                                  = 0x12000
uname({sys="Linux", node="kindle", ...}) = 0
mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x4001b000
access("/etc/ld.so.preload", R_OK)      = -1 ENOENT (No such file or directory)
open("/var/run/ld.so.cache", O_RDONLY)  = 3
fstat64(3, {st_mode=S_IFREG|0644, st_size=13788, ...}) = 0
mmap2(NULL, 13788, PROT_READ, MAP_PRIVATE, 3, 0) = 0x4001c000
close(3)                                = 0
open("/lib/libc.so.6", O_RDONLY)        = 3
read(3, "\177ELF\1\1\1\0\0\0\0\0\0\0\0\0\3\0(\0\1\0\0\0\250J\1\0004\0\0\0"..., 512) = 512
fstat64(3, {st_mode=S_IFREG|0755, st_size=1348230, ...}) = 0
mmap2(NULL, 1135004, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_DENYWRITE, 3, 0) = 0x40025000
mprotect(0x4012e000, 28672, PROT_NONE)  = 0
mmap2(0x40135000, 12288, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0x108) = 0x40135000
mmap2(0x40138000, 8604, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0x40138000
close(3)                                = 0
mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x40020000
set_tls(0x400204a0, 0x400204a0, 0x40024058, 0x40020b78, 0x40) = 0
mprotect(0x40135000, 8192, PROT_READ)   = 0
munmap(0x4001c000, 13788)               = 0
brk(0)                                  = 0x12000
brk(0x33000)                            = 0x33000
open("/mnt/us/modules/smbfs.ko", O_RDONLY) = 3
read(3, "\177ELF\1\1\1\0\0\0\0\0\0\0\0\0\1\0(\0\1\0\0\0\0\0\0\0\0\0\0\0"..., 16384) = 16384
read(3, "\2600\33\345\0 \303\345xA\33\345\0 \304\345\264\340\33\345\0 \316\345\220pK\345\234\21\33\345"..., 16384) = 16384
read(3, "\5\0\240\341\376\377\377\353\360\250\235\350\r\300\240\341\0\330-\351\4\260L\342\2726\320\341#3\1\340"..., 32768) = 32768
read(3, "\270\0\0\0\2\0\1\0\27\0\0\0dc\0\0\0\0\0\0\0\0\1\0P\0\0\0\24d\0\0"..., 65536) = 11088
read(3, "", 54448)                      = 0
close(3)                                = 0
init_module(0x12018, 76624, "")         = 0
exit_group(0)                           = ?
+++ exited with 0 +++

[smbfs.ko for the 3 from the lab126 2.6.26 kernel, via the 2006q3 CS TC]

So I will be using that TC for my testing on the arm of 'USING THE RIGHT HAMMER FOR THE JOB'

The link for the 2006q3 TC is available on the KEK project tools repository.
That's where I got mine.

I'll update the BR - "doing it like an Amazon" - guide when I have it slicker and solid. but in essence it still works. I will update the Best Practice choice of tools now.

Hopefully these Best Practices will help to form an accessible reference for the future digital hitch-hiker.

Insert requests for custom modules here. Thanks.

I'll remove the binfmt_misc.ko in the OP just in case...

[twobob]

[root@kindle root]# modinfo /mnt/us/modules/cifs.ko

filename:       /mnt/us/modules/cifs.ko
author:         Steve French <sfrench@us.ibm.com>
license:        GPL
description:    VFS to access servers complying with the SNIA CIFS Specification e.g. Samba and Windows
version:        1.53
vermagic:       2.6.26-rt-lab126 mod_unload modversions ARMv6 
depends:        
srcversion:     55B503E37F4DFB20AA9BE56
parm:           cifs_max_pending:Simultaneous requests to server. Default: 50 Range: 2 to 256 (int)
parm:           cifs_min_small:Small network buffers in pool. Default: 30 Range: 2 to 256 (int)
parm:           cifs_min_rcv:Network buffers in pool. Default: 4 Range: 1 to 64 (int)
parm:           CIFSMaxBufSize:Network buffer size (not including header). Default: 16384 Range: 8192 to 130048 (int)

[root@kindle root]# insmod /mnt/us/modules/cifs.ko

[root@kindle root]# lsmod
Module                  Size  Used by
cifs                  238340  0 
binfmt_misc             8552  0 
smbfs                  64800  0 
g_ether                21096  0

[twobob]

A quick eyeball comparison of what modinfo thinks about an internal module and a recreated one using the BR - lab126 method

[root@kindle root]# modinfo fuse

filename:       /lib/modules/2.6.26-rt-lab126/kernel/fs/fuse/fuse.ko
alias:          char-major-10-229
license:        GPL
description:    Filesystem in Userspace
author:         Miklos Szeredi <miklos@szeredi.hu>
depends:        
vermagic:       2.6.26-rt-lab126 mod_unload modversions ARMv6

[root@kindle root]# modinfo /mnt/us/modules/fuse.ko

filename:       /mnt/us/modules/fuse.ko
alias:          char-major-10-229
author:         Miklos Szeredi <miklos@szeredi.hu>
description:    Filesystem in Userspace
license:        GPL
vermagic:       2.6.26-rt-lab126 mod_unload modversions ARMv6 
depends:

Looks like it will swallow that.

[twobob]

The new trimmed down /usr/bin built with BR (The ln -s type binaries are elided)

Spoiler:

[root@kindle bin]# ls
aclocal                   df_input                  fbterm                    links                     ntfscluster
addr2line                 df_video                  fbv                       lite_bench                ntfscmp
animate                   df_video_particle         fc-cache                  lite_checktest            ntfsfix
autoconf                  df_window                 fc-cat                    lite_dfbspy               ntfsinfo
autoheader                dfbtest_blit              fc-list                   lite_listtest             ntfsls
autom4te                  dfbtest_blit2             fc-match                  lite_msgbox               opkg-cl
automake                  dfbtest_fillrect          fftw-wisdom               lite_progressbar          opkg-key
autoreconf                dfbtest_font              fftw-wisdom-to-conf       lite_run                  pgpewrap
autoscan                  dfbtest_mirror            file                      lite_scrollbartest        pgpring
autoupdate                dfbtest_prealloc          flea                      lite_simple               readelf
bison                     dfbtest_reinit            fusion_call               lite_slider               rsync
bsdiff                    dfbtest_resize            fusion_fork               lite_textbuttontest       screen
bspatch                   dfbtest_scale             fusion_reactor            lite_textlinetest         sip_reg
bzdiff                    dfbtest_sync              fusion_skirmish           lite_textlisttest         size
bzgrep                    dfbtest_window            fusion_stream             logsurfer                 smime_keys
bzip2                     dfbtest_window_cursor     gdb                       lowntfs-3g                speexdec
bzip2recover              dfbtest_window_flip_once  gdbserver                 m4                        speexenc
bzmore                    dfbtest_window_surface    gperf                     make                      spooky
c++filt                   dialog                    gprof                     microperl                 stream
ccache                    direct_stream             identify                  mogrify                   strings
compare                   direct_test               ifnames                   montage                   ts_calibrate
composite                 directfb-config           import                    mutt                      ts_finddev
conjure                   display                   inputattach               muttbug                   ts_harvest
convert                   distcc                    jpegtran                  nano                      ts_print
coretest_blit2            distccd                   jsmin                     ntfs-3g                   ts_print_raw
ctorrent                  divine-config             libhid-detach-device      ntfs-3g.probe             ts_test
cvs                       elfedit                   libtool                   ntfs-3g.secaudit          update-alternatives
df_drivertest             fbdump                    libtoolize                ntfs-3g.usermap           yacc
df_fonts                  fbgrab                    libusb-config             ntfscat
[root@kindle bin]#

[twobob]

[root@kindle sbin]# ./usb_modeswitch --help

 * usb-modeswitch: handle USB devices with multiple modes
 * Version 1.1.2alpha (C) Josua Dietze 2010
 * Based on libusb0 (0.1.12 and above)

 ! PLEASE REPORT NEW CONFIGURATIONS !

Usage: usb-modeswitch [-hvpVPmMrdHn] [-c filename]

 -h, --help                    this help
 -e, --version                 print version number and exit
 -v, --default-vendor NUM      vendor ID to look for (mandatory)
 -p, --default-product NUM     product ID to look for (mandatory)
 -V, --target-vendor NUM       target vendor (optional, for success check)
 -P, --target-product NUM      target model (optional, for success check)
 -C, --target-class NUM        target device class
 -m, --message-endpoint NUM    where to direct the message (optional)
 -M, --message-content <msg>   command to send (hex number as string)
 -2 <msg>, -3 <msg>            additional commands to send
 -n, --need-response           read a response to the message transfer
 -r, --response-endpoint NUM   where from read the response (optional)
 -d, --detach-only             just detach the storage driver
 -H, --huawei-mode             apply a special procedure
 -S, --sierra-mode             apply a special procedure
 -O, --sony-mode               apply a special procedure
 -G, --gct-mode                apply a special procedure
 -R, --reset-usb               reset the device in the end
 -c, --config <filename>       load different config file
 -Q, --quiet                   don't show progress or error messages
 -W, --verbose                 print all settings before running
 -D, --sysmode                 specific result and syslog message
 -s, --success NUM             check switching result after NUM secs
 -I, --no-inquire              do not get device details (default on)

 -i, --interface NUM           select initial USB interface (default 0)
 -u, --configuration NUM       select USB configuration
 -a, --altsetting NUM          select alternative USB interface setting

Looks helpful.

[knc1]

Here is one, well, one pair, of programs to add to the collection:
http://www.daemonology.net/bsdiff/

Should be easy to add-in (and push upstream) to Buildroot.
The general use would be to build bsdiff for the host and bspatch for the target but the makefile isn't organized that way.
So just build the pair for both host and target with two calls to GENTARGETS ( ... )

If you haven't added a package file to Buildroot before, for this one just copy one of the simplest ones you can find - text edit the various names.
The GENTARGETS macro will write the actual makefile glue between Buildroot and the package's makefile.

This package depends on one external package, libbz2 but I suspect that package is already defined in Buildroot.

Upstream **might** want a patch included to change that CFLAGS line (which would over-ride the optimization level selected in Buildroot and also contains a LDFLAGS item.

- - - -

Oh, as I am sure you have learned by now - you can build any individual package in Buildroot with the correct "make ... ... ..." Voodo command.

[knc1]

Now that we have them (Samba and CIFS), I guess it is time to google up the correct config file, its contents, decide where to put it, decide what to export to M$-Windows (Linux, MacOSx), figure out a nice little patch to the firewall rules.

Maybe keep it safe and simple, with a read-only export of some part of the filesystem.

[knc1]

Another little chore . . . .
FAT-32 does not support symbolic links (required to get the symantics of *nix library sonames correct).

There are (at least) two ways to deal with that:
FAT{,16,32} does partially support hard-links (at least until you run dosfsck on it) ;
The target library of the links could be copied to each of the less specific names (makes a large increase in disk usage, but we have 3Gbytes) ;
or
some combination of the above.

It would probably be appropriate to do that in a custom, post-build, script.


Decisions, decisions, decisions. . . .

[twobob]

Originally Posted by knc1

Another little chore . . . .
FAT-32 does not support symbolic links (required to get the symantics of *nix library sonames correct).

There are (at least) two ways to deal with that:
FAT{,16,32} does partially support hard-links (at least until you run dosfsck on it) ;
The target library of the links could be copied to each of the less specific names (makes a large increase in disk usage, but we have 3Gbytes) ;
or
some combination of the above.

It would probably be appropriate to do that in a custom, post-build, script.


Decisions, decisions, decisions. . . .

if you use fileroller to export to fat it does that for you. 'ln -s' become the actual files.

this tells me that doubtless some underlying process is being invoked that supports this. possibly tar -

tar -h, --dereference follow symlinks; archive and dump the files they point to
--hard-dereference follow hard links; archive and dump the files they refer to

EDIT: I looked around this issue some more and have experience created junction points / hard links in MS server application but AIUI FAT would see this as a file error (IIRC).

It's Not impossible by any means:
I found example code to create symbolic link support for FAT drives but it WOULD REQUIRE at the very minimum a windows related process on the filesytem (invoking wine might do it), there are also various MS utilities throughout the ages that could do this.

BUT I don't think this is a good solution given the PITA and probably fschk complaints. There would need to be a good reason and IMHO I don't see one yet.

In this case the 'rolled out' version of the file would probably be far less painful to manage.

I am using these 'Rolled Out' copies successfully now.

It would probably be appropriate to do that in a custom, post-build, script.

me@dev ~/somewhere_else CD YOUR_BUILDROOT_OUTPUT_FOLDER/images ~(let's say ~/BLDS/kindle/images for ease)

cp the rootfs.tar to a temp.

cp rootfs.tar rootfs.temp.tar

me@dev ~/BLDS/kindle $ tar --delete -vf ./rootfs.temp.tar ./dev/ ./home/ ./media/ ./mnt/ ./opt/ ./proc/ ./root/ ./run/ ./sys/ ./tmp/ ./var/ ./linuxrc

copy that onto your device in the root of mnt/us (or somewhere on the fat partition anyways) and run tar with --extract --dereference on it.

I'll go check the exact local tar command as I don't use it (I extract via sftp)

But it will be something like
[root@kindle]# tar -xvf ./rootfs.temp.tar --dereference



err no. busybox tar doesn't support that. I'll have another think.
That process will get us the good stuff easily splattered into a nice tar.

[knc1]

The way I had in mind, once discovered by dos-fsck, would result in identical copies under different names being created to "fix" what isn't broken.

So might as well just start out with multiple copies.
Then one only must figure out a way to enforce: ldconfig -X ...
Otherwise it would try to replace the copies with the symbolic links.

That sort of thing is usually done by replacing /sbin/ldconfig with a shell script that calls the re-named ldconfig with option -X plus an edited version of the original arguments.

Hmm...
Now we are talking about a change to the filesystem that would break any OTA update when (if) /sbin/ldconfig was ever tested.

Hmm...
Unless we learn to always patch the md5sum manifest file when we change a system file.
Yup, I forsee the need for that bsdiff/bspatch pair of routines mentioned above.

[knc1]

I don't think that busybox tar understands: --dereference
Check for that on your K3 - (mine is still torn apart).

[twobob]

Originally Posted by knc1

I don't think that busybox tar understands: --dereference
Check for that on your K3 - (mine is still torn apart).

your right it doesn't. however creating an interim tar with the -h will probably sort it.

[knc1]

There was a patch posted on the BR M.L. yesterday to fix the QT build.

In the top level Makefile, in the list of unexported symbols, add:
unexport QMAKESPEC

The list should be somewhere near line 209.