[Buildroot] Annoying... nothing work at all :-(

[Buildroot] Annoying... nothing work at all :-(

[sani.broyeur]

Hi all.

I'm trying for several days to get a simple working result with
buildroot. Nothing.

I'm very surprised!

The source PC is a regular Fedora12 AMD quadcore.

The target is a MiniITX single board regular Atom PC, able to boot with
all desktop OS, on a regular disk or a CompactFlash reader.

- When invoking "make menuconfig && make" an ext2 image is created. OK
but unbootable, without bootloader nor kernel.

- When asking to add a bootloader, grub doesn't install properly on
target image, config is not done. This part is broken.

- When asking to add a CF/FAT16 bootloader, syslinux doesn't install at
all on target image, this part is also broken since it does really
nothing on the image :-)

- When asking a kernel, attempts lead often to cryptic make errors and
the whole buildroot tree has to be recreated from .tar.gz without kernel
generation. The "kernel" menu in menuconfig is broken, and no kernel
config is invoked at build or within menuconfig.

- When invoking linux26-menuconfig (I'v found that one time), either
nothing, or make errors, this one is also broken.

Etc, etc...

- Why asking for installation of a bootloader that is never installed,
nor configured on the target ?

- Why asking for generation of a kernel that is never created and lead
to corruption of buildroot tree ?

etc, etc...

All produced forms of "target" are either incomplete, or unusable, or
unbootable, when the commands are not ending with errors!

Is there somewhere an USABLE documentation?

A step-by-step tutorial ? With exemple and results ?

I'm not a newbie to linux, but this "buildroot project" seems for sure
not usable before years of improvements...

Comments would be appreciated.

Regards.

[Buildroot] Annoying... nothing work at all :-(

[Thomas Petazzoni]

Hello,

On Sun, 13 Jun 2010 05:13:14 +0200
"sani.broyeur" <sani.broyeur@free.fr> wrote:

> - When invoking "make menuconfig && make" an ext2 image is created. OK
> but unbootable, without bootloader nor kernel.

We've made the choice by default to only build a toolchain + a simple
root filesystem image with only Busybox in it. So that's expected.

> - When asking to add a bootloader, grub doesn't install properly on
> target image, config is not done. This part is broken.

What do you expect exactly ? How would you want grub to be installed ?
If you're not more specific about what you expect, it's going to be
hard to help you.

(FYI: I'm currently doing a fairly big bootloader code cleanup, for
which I've sent a 49 patches patch set yesterday on this list. At least
with this patch set applied, Grub compiles properly. However, I'm not
sure what to do with the resulting binaries.)

> - When asking to add a CF/FAT16 bootloader, syslinux doesn't install
> at all on target image, this part is also broken since it does really
> nothing on the image :-)

What do you call "target image" ? The ext2 filesystem ? Then it is
expected that syslinux isn't installed *in* it. However, the syslinux
image is supposed to be available in output/images/.

But again, I know syslinux builds properly with my bootloaders-cleanup
branch applied. The bootloader code in Buildroot hasn't been
unmaintained for quite some time, so he doesn't surprise me that you
encounter issues.

Another thing is that Buildroot is used a lot for non-x86 architectures
(ARM, PowerPC, etc.), and for these architectures, the bootloader is
never *in* the root filesystem image. The fact the bootloader is *in*
the root filesystem image is a x86-specific thing, and as we haven't
had a lot of x86-related contributions lately, it may well be that
we're not doing what should be done for this specific architecture.

> - When asking a kernel, attempts lead often to cryptic make errors and
> the whole buildroot tree has to be recreated from .tar.gz without
> kernel generation. The "kernel" menu in menuconfig is broken, and no
> kernel config is invoked at build or within menuconfig.

Please provide detailed reproduction steps, because ? attemps lead
often to cryptic make errors ? is by no means an useful description of
what's happening.

FWIW :

 * I've built succesfully a kernel yesterday with Buildroot, both ARM
   and PowerPC.

 * I'm working on a big cleanup of the linux compilation code in
   Buildroot, because the current code is way too complicated.

> - When invoking linux26-menuconfig (I'v found that one time), either
> nothing, or make errors, this one is also broken.

Be more specific ?

> All produced forms of "target" are either incomplete, or unusable, or
> unbootable, when the commands are not ending with errors!

Be more specific ?

> Is there somewhere an USABLE documentation?

The only documentation we have so far is
http://buildroot.org/downloads/buildroot.html. Patches welcome.

> I'm not a newbie to linux, but this "buildroot project" seems for sure
> not usable before years of improvements...

We certainly have rough edges, particularly in the area of
bootloader/kernel compilation, which haven't seen any
cleanup/refactoring since a very, very long time.

I'm sorry, but your message is by no means useful :

 * You're not giving an appropriate level of precision that would give
   us the opportunity to understand and solve the problems you're
   facing. For each problem: the .config file that allows to produce
   the issue, the behaviour that you see, and the behaviour that you
   would expect Buildroot to have.

 * You're insulting the work done by the developers on the Buildroot
   project, reducing their incentive to help you and to solve your
   problems.

Thomas
-- 
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com

[Buildroot] Annoying... nothing work at all :-(

[sani.broyeur]

OK thanks Thomas, let's be more accurate, step by step:

0) My experience is not an "insult", juste a "result" :-)

1) tar -xvzf buildroot-2010.05.tar.gz --> OK.

2) cd buildroot-2010.05 --> OK.

3) make menuconfig
- no changes at all made to default options
- save on exit

4) make --> OK. rootfs.ext2 created in output/images

5) make menuconfig again to ask for a kernel
- kernel type -> advanced configuration
        linux kernel version -> latest
        patch -> no
        linux kernel configuration -> Run make ARCH=$(ARCH) [xconfig|
menuconfig] before build
        kernel binary -> bZimage
        destination -> copy kernel to root file
- save on exit

6) make

make[1]: *** [insmod.static] Erreur 1
make[1]: *** Attente des t?ches non termin?es....
make[1]: quittant le r?pertoire
? /devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2 ?
make: ***
[/devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2/depmod] Erreur 2
[root at hercules buildroot-2010.05]# 

-> Message means "waiting for unfinished tasks": the "-j 2" option in
make invocation (defaut = 2) is perhaps the cause. Relaunch!

7) make again

rm -rf /devel/buildroot-2010.05/output/build/buildroot-config
mkdir -p /devel/buildroot-2010.05/output/build
cp -dpRf
package/config/buildroot-config /devel/buildroot-2010.05/output/build/buildroot-config
/usr/bin/make -j2
-C /devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2
make[1]: entrant dans le r?pertoire
? /devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2 ?
/usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o lsmod  lsmod.o  
/usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o insmod  insmod.o  
/usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o rmmod  rmmod.o  
/usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o modprobe  modprobe.o
zlibsupport.o  
/usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o depmod  depmod.o
moduleops.o tables.o zlibsupport.o  
/usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o modinfo  modinfo.o
zlibsupport.o  
/usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o insmod.static -static
insmod.o  
/usr/bin/ld: cannot find -lc
collect2: ld a retourn? 1 code d'?tat d'ex?cution
make[1]: *** [insmod.static] Erreur 1
make[1]: quittant le r?pertoire
? /devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2 ?
make: ***
[/devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2/depmod] Erreur 2

-> ld perhaps invoked with an empty variable leading to syntax mismatch,
option taken for parameter or missing one, or something alike. Relaunch!

8) make again

This time, "linux26-menuconfig" is automagically invoked as expected.
- Just changed: SMP support [disable] / CPU type [ATOM]
- No optimization nor kernel tayloring for the embedded x86 target.
Let's make it simple :-)
- save changes on exit
make goes on, generating all the regular stuff.
[...] MODPOST 1975 modules
[...]
Root device is (8, 2)
Setup is 13516 bytes (padded to 13824 bytes).
System is 3183 kB
CRC f8992a55
Kernel: arch/x86/boot/bzImage is ready  (#1)
[...]
/devel/buildroot-2010.05/output/target/boot/bzImage -> OK a kernel is
generated.
[...]
>>>   Generating root filesystem image rootfs.ext2
rm -f /devel/buildroot-2010.05/output/build/_fakeroot.fs
touch /devel/buildroot-2010.05/output/build/.fakeroot.00000
cat /devel/buildroot-2010.05/output/build/.fakeroot*
> /devel/buildroot-2010.05/output/build/_fakeroot.fs
echo "chown -R 0:0 /devel/buildroot-2010.05/output/target"
>> /devel/buildroot-2010.05/output/build/_fakeroot.fs
echo "/devel/buildroot-2010.05/output/host/usr/bin/makedevs -d
target/generic/device_table.txt /devel/buildroot-2010.05/output/target"
>> /devel/buildroot-2010.05/output/build/_fakeroot.fs
echo "
PATH="/devel/buildroot-2010.05/output/toolchain/bin:/devel/buildroot-2010.05/output/host/bin:/devel/buildroot-2010.05/output/host/usr/bin:/devel/buildroot-2010.05/output/host/usr/sbin/:/devel/buildroot-2010.05/output/staging/bin:/devel/buildroot-2010.05/output/staging/usr/bin:/usr/lib64/qt-3.3/bin:/usr/kerberos/sbin:/usr/kerberos/bin:/usr/lib64/ccache:/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin:/root/bin" fs/ext2/genext2fs.sh -d /devel/buildroot-2010.05/output/target  -U /devel/buildroot-2010.05/output/images/rootfs.ext2" >> /devel/buildroot-2010.05/output/build/_fakeroot.fs
chmod a+x /devel/buildroot-2010.05/output/build/_fakeroot.fs
/devel/buildroot-2010.05/output/host/usr/bin/fakeroot
-- /devel/buildroot-2010.05/output/build/_fakeroot.fs
rootdir=/devel/buildroot-2010.05/output/target
table='target/generic/device_table.txt'
genext2fs: couldn't allocate a block (no free space)
make: *** [/devel/buildroot-2010.05/output/images/rootfs.ext2] Erreur 1 

-> KO the ext2 fs is NOT generated.

Of course, there is plenty of Gb available on disk, this error is
surprising.

9) trying to generate something else, i.e cramfs, squashfs and
initramfs:
- make menuconfig -> target filesystem option
        ENABLED cramfs
        ENABLED squashfs
        ENABLED initramfs
        DISABLED ext2
- save on exit

10) make again
- building of cramfs
[...]
>>>   Generating root filesystem image rootfs.cramfs
warning: estimate of required size (upper bound) is 661MB, but maximum
image size is 272MB, we might die prematurely
mkcramfs: filesystem too big
make: *** [/devel/buildroot-2010.05/output/images/rootfs.cramfs] Erreur
8

11) let see what we got after attempting to generate either a root fs
and the corresponding kernel:

[root at hercules buildroot-2010.05]# ll output/images/
total 7156
-rw-r--r-- 1 root root 3272576 juin  13 15:50 bzImage
-rw-r--r-- 1 root root       0 juin  13 15:50 rootfs.cramfs
-rw-r--r-- 1 root root 3645440 juin  13 14:08 rootfs.ext2
-rw-r--r-- 1 root root  394744 juin  13 15:49 rootfs.initramfs

- kernel seems to be generated (no evidence it will load at this time)
- rootfs.cramfs isn't generated
- rootfs.ext2 seems to be generated BUT is not (error above at point 8)
- rootfs.initramfs is only a text file without any CPIO content:

[root at hercules buildroot-2010.05]# head output/images/rootfs.initramfs 

#####################
# /devel/buildroot-2010.05/output/target
# Last modified: 1276436908.0000000000

dir /bin 775 0 0
slink /bin/dmesg busybox 777 0 0
slink /bin/tar busybox 777 0 0
slink /bin/watch busybox 777 0 0
slink /bin/zcat busybox 777 0 0

[root at hercules buildroot-2010.05]# tail output/images/rootfs.initramfs
slink /usr/sbin/rdate ../../bin/busybox 777 0 0
slink /usr/sbin/readprofile ../../bin/busybox 777 0 0
slink /usr/sbin/setlogcons ../../bin/busybox 777 0 0
slink /usr/sbin/dnsd ../../bin/busybox 777 0 0
slink /usr/sbin/chroot ../../bin/busybox 777 0 0
dir /usr/share 775 0 0
dir /usr/share/udhcpc 775 0 0
file /usr/share/udhcpc/default.script /devel/buildroot-2010.05/output/target/usr/share/udhcpc/default.script 755 0 0
dir /home 775 0 0
dir /home/default 2755 1000 1000
[root at hercules buildroot-2010.05]# 

This "rootfs.initramfs" is NOT an initramfs as expected but something
else I cannot state on.

12) additional investigations:
- asking for a bootloader without kernel prevent errors in target
filesystems generation, since the size issue is avoided.
- asking a bootloader in make menuconfig lead to strange things:
        -> grub only create a /boot/grub subdir in target, but doesn't
create
menu.lst nor invoque root (hd0,0) nor setup (hd0,0). Doing it manually
afterwards is refused.
        -> sysconfig is an external bootloader, your're right, so
including it
in menuconfig could be understood as "prepare a subdir in output/image
with all the relevant stuff configured, that is: bzImage+ldlinux.sys
+rootfs.cramfs+syslinux.cfg. Nothing this way is done. What means
"syslinux" in menuconfig?

13) Pending questions...
- Is buildroot suitable at this time to create tiny x86 embedded
systems, either by regular filesystem on a disk (ext2) or ramdisk in a
flashcard (syslinux+initrd) ?
- Why is there so many unrecoverable errors as I used default presets ?
- Is there rock-solid examples of bootable x86 systems generated, or
only "I sware it works, I was aware of an alien CPU card using it long
time ago in a distant galaxy" :-)))

Please remember: this not an "insult", just a "result" ;-)

Best regards.



Le dimanche 13 juin 2010 ? 11:19 +0200, Thomas Petazzoni a ?crit :
> Hello,
> 
> On Sun, 13 Jun 2010 05:13:14 +0200
> "sani.broyeur" <sani.broyeur@free.fr> wrote:
> 
> > - When invoking "make menuconfig && make" an ext2 image is created. OK
> > but unbootable, without bootloader nor kernel.
> 
> We've made the choice by default to only build a toolchain + a simple
> root filesystem image with only Busybox in it. So that's expected.
> 
> > - When asking to add a bootloader, grub doesn't install properly on
> > target image, config is not done. This part is broken.
> 
> What do you expect exactly ? How would you want grub to be installed ?
> If you're not more specific about what you expect, it's going to be
> hard to help you.
> 
> (FYI: I'm currently doing a fairly big bootloader code cleanup, for
> which I've sent a 49 patches patch set yesterday on this list. At least
> with this patch set applied, Grub compiles properly. However, I'm not
> sure what to do with the resulting binaries.)
> 
> > - When asking to add a CF/FAT16 bootloader, syslinux doesn't install
> > at all on target image, this part is also broken since it does really
> > nothing on the image :-)
> 
> What do you call "target image" ? The ext2 filesystem ? Then it is
> expected that syslinux isn't installed *in* it. However, the syslinux
> image is supposed to be available in output/images/.
> 
> But again, I know syslinux builds properly with my bootloaders-cleanup
> branch applied. The bootloader code in Buildroot hasn't been
> unmaintained for quite some time, so he doesn't surprise me that you
> encounter issues.
> 
> Another thing is that Buildroot is used a lot for non-x86 architectures
> (ARM, PowerPC, etc.), and for these architectures, the bootloader is
> never *in* the root filesystem image. The fact the bootloader is *in*
> the root filesystem image is a x86-specific thing, and as we haven't
> had a lot of x86-related contributions lately, it may well be that
> we're not doing what should be done for this specific architecture.
> 
> > - When asking a kernel, attempts lead often to cryptic make errors and
> > the whole buildroot tree has to be recreated from .tar.gz without
> > kernel generation. The "kernel" menu in menuconfig is broken, and no
> > kernel config is invoked at build or within menuconfig.
> 
> Please provide detailed reproduction steps, because ? attemps lead
> often to cryptic make errors ? is by no means an useful description of
> what's happening.
> 
> FWIW :
> 
>  * I've built succesfully a kernel yesterday with Buildroot, both ARM
>    and PowerPC.
> 
>  * I'm working on a big cleanup of the linux compilation code in
>    Buildroot, because the current code is way too complicated.
> 
> > - When invoking linux26-menuconfig (I'v found that one time), either
> > nothing, or make errors, this one is also broken.
> 
> Be more specific ?
> 
> > All produced forms of "target" are either incomplete, or unusable, or
> > unbootable, when the commands are not ending with errors!
> 
> Be more specific ?
> 
> > Is there somewhere an USABLE documentation?
> 
> The only documentation we have so far is
> http://buildroot.org/downloads/buildroot.html. Patches welcome.
> 
> > I'm not a newbie to linux, but this "buildroot project" seems for sure
> > not usable before years of improvements...
> 
> We certainly have rough edges, particularly in the area of
> bootloader/kernel compilation, which haven't seen any
> cleanup/refactoring since a very, very long time.
> 
> I'm sorry, but your message is by no means useful :
> 
>  * You're not giving an appropriate level of precision that would give
>    us the opportunity to understand and solve the problems you're
>    facing. For each problem: the .config file that allows to produce
>    the issue, the behaviour that you see, and the behaviour that you
>    would expect Buildroot to have.
> 
>  * You're insulting the work done by the developers on the Buildroot
>    project, reducing their incentive to help you and to solve your
>    problems.
> 
> Thomas
> -- 
> Thomas Petazzoni, Free Electrons
> Kernel, drivers, real-time and embedded Linux
> development, consulting, training and support.
> http://free-electrons.com
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
> 

[Buildroot] Annoying... nothing work at all :-(

[Thomas Petazzoni]

Hello,

Thanks for sharing more precisions about the issues you're having, this
report is actually helpful, even though a little bit more details on
some of the issues are still needed.

On Sun, 13 Jun 2010 17:19:08 +0200
"sani.broyeur" <sani.broyeur@free.fr> wrote:

> 0) My experience is not an "insult", juste a "result" :-)

Sorry, but:

?
I'm not a newbie to linux, but this "buildroot project" seems for sure
not usable before years of improvements...
?

is indeed insulting the work of the developers contributing to
Buildroot.

> 6) make
> 
> make[1]: *** [insmod.static] Erreur 1
> make[1]: *** Attente des t?ches non termin?es....
> make[1]: quittant le r?pertoire
> ? /devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2 ?
> make: ***
> [/devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2/depmod]
> Erreur 2 [root at hercules buildroot-2010.05]# 
> 
> -> Message means "waiting for unfinished tasks": the "-j 2" option in
> make invocation (defaut = 2) is perhaps the cause. Relaunch!

Please provide the rest of the build log, the real error occured before
that.

> /usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o insmod.static
> -static insmod.o  
> /usr/bin/ld: cannot find -lc
> collect2: ld a retourn? 1 code d'?tat d'ex?cution
> make[1]: *** [insmod.static] Erreur 1
> make[1]: quittant le r?pertoire
> ? /devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2 ?
> make: ***
> [/devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2/depmod]
> Erreur 2
> 
> -> ld perhaps invoked with an empty variable leading to syntax
> mismatch, option taken for parameter or missing one, or something
> alike. Relaunch!

Here I'm sorry, but it looks like something is broken on your side:
we're just using the gcc/ld of your host. Could you try to disable
ccache on your system and see if it works better ?

If you try to statically compile a simple program on your host, does it
work ?

> make goes on, generating all the regular stuff.
> [...] MODPOST 1975 modules

Ouch, huge number of modules. Certainly not an use case we've been
testing too much.

> >> -- /devel/buildroot-2010.05/output/build/_fakeroot.fs
> rootdir=/devel/buildroot-2010.05/output/target
> table='target/generic/device_table.txt'
> genext2fs: couldn't allocate a block (no free space)
> make: *** [/devel/buildroot-2010.05/output/images/rootfs.ext2] Erreur
> 1 
> 
> -> KO the ext2 fs is NOT generated.

Here, it seems that our help script fs/ext2/genext2fs.sh made a wrong
estimate of the number of blocks the target filesystem should be to
contain all the files of your system.

This is definitely a bug, I'll have a look at it.

> Of course, there is plenty of Gb available on disk, this error is
> surprising.

No, it isn't :-)

> 9) trying to generate something else, i.e cramfs, squashfs and
> initramfs:
> - make menuconfig -> target filesystem option
>         ENABLED cramfs
>         ENABLED squashfs
>         ENABLED initramfs
>         DISABLED ext2
> - save on exit
> 
> 10) make again
> - building of cramfs
> [...]
> >>>   Generating root filesystem image rootfs.cramfs
> warning: estimate of required size (upper bound) is 661MB, but maximum
> image size is 272MB, we might die prematurely
> mkcramfs: filesystem too big
> make: *** [/devel/buildroot-2010.05/output/images/rootfs.cramfs]
> Erreur 8

By design, cramfs filesystems are limited to ~272 MB. See
http://en.wikipedia.org/wiki/Cramfs:

? Maximum file system size is a little under 272MB. (The last file on
the file system must begin before the 256MB block, but can extend past
it.) ?

Nothing we can really do about: cramfs filesystems are useful for some
embedded systems, and having a 661 MB root filesystem in Buildroot is
definitely not the common case.

Moreover, the message is pretty clear, it's saying: ? filesystem too
big ?

> 11) let see what we got after attempting to generate either a root fs
> and the corresponding kernel:
> 
> [root at hercules buildroot-2010.05]# ll output/images/
> total 7156
> -rw-r--r-- 1 root root 3272576 juin  13 15:50 bzImage
> -rw-r--r-- 1 root root       0 juin  13 15:50 rootfs.cramfs
> -rw-r--r-- 1 root root 3645440 juin  13 14:08 rootfs.ext2
> -rw-r--r-- 1 root root  394744 juin  13 15:49 rootfs.initramfs
> 
> - kernel seems to be generated (no evidence it will load at this time)
> - rootfs.cramfs isn't generated

Yes, see above explanation.

> - rootfs.ext2 seems to be generated BUT is not (error above at point
> 8)

Yes, see above explanation.

> - rootfs.initramfs is only a text file without any CPIO content:

This is normal, this file is *not* supposed to be a CPIO archive. It is
supposed to be used by the kernel as the CONFIG_INITRAMFS_SOURCE value.
Its purpose is to describe what the initramfs should contain.

> This "rootfs.initramfs" is NOT an initramfs as expected but something
> else I cannot state on.

rootfs.initramfs *IS* the text file that will allow the kernel to
include an initramfs.

> 12) additional investigations:
> - asking for a bootloader without kernel prevent errors in target
> filesystems generation, since the size issue is avoided.

Yes, as these errors are all due to the fact that your filesystem is
much, much bigger than the typical cases we do (< 50 MB).

> - asking a bootloader in make menuconfig lead to strange things:
>         -> grub only create a /boot/grub subdir in target, but doesn't
> create
> menu.lst nor invoque root (hd0,0) nor setup (hd0,0). Doing it manually
> afterwards is refused.

What would you want Buildroot to do exactly with Grub ? Remember that
Buildroot doesn't run as root, and this might limit the actions that
Buildroot can take with regard to the installation of a bootloader.

>         -> sysconfig is an external bootloader, your're right, so
> including it
> in menuconfig could be understood as "prepare a subdir in output/image
> with all the relevant stuff configured, that is: bzImage+ldlinux.sys
> +rootfs.cramfs+syslinux.cfg. Nothing this way is done. What means
> "syslinux" in menuconfig?

I'm not sure what the current syslinux code is doing. In my cleanup,
I've made sure it copies isolinux.bin to output/images. If you'd like
it to do more things, please describe that in detail and I can have a
look at it.

> 13) Pending questions...
> - Is buildroot suitable at this time to create tiny x86 embedded
> systems, either by regular filesystem on a disk (ext2) or ramdisk in a
> flashcard (syslinux+initrd) ?

Yes, but as you can see, there might be issues here and there, mostly
because the bootloader code isn't very good for x86 bootloaders. Some
help on *what* Buildroot should do would be very helpful.

> - Why is there so many unrecoverable errors as I used default
> presets ?

See above explanations. I have no explanation for the module-init-tools
error you had, except that something seems strange with your ccache set
up.

Thomas
-- 
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com

[Buildroot] Annoying... nothing work at all :-(

[Will Moore]

> -----Original Message-----
> From: buildroot-bounces at busybox.net [mailto:buildroot-bounces at busybox.net]
> On Behalf Of sani.broyeur
> Sent: 13 June 2010 16:19
> To: buildroot at busybox.net
> Subject: Re: [Buildroot] Annoying... nothing work at all :-(
> 
> OK thanks Thomas, let's be more accurate, step by step:
> 
> 0) My experience is not an "insult", juste a "result" :-)
> 
> 1) tar -xvzf buildroot-2010.05.tar.gz --> OK.
> 
> 2) cd buildroot-2010.05 --> OK.
> 
> 3) make menuconfig
> - no changes at all made to default options
> - save on exit
> 
> 4) make --> OK. rootfs.ext2 created in output/images
> 
> 5) make menuconfig again to ask for a kernel
> - kernel type -> advanced configuration
>         linux kernel version -> latest
>         patch -> no
>         linux kernel configuration -> Run make ARCH=$(ARCH) [xconfig|
> menuconfig] before build
>         kernel binary -> bZimage
>         destination -> copy kernel to root file
> - save on exit
> 
> 6) make
> 
> make[1]: *** [insmod.static] Erreur 1
> make[1]: *** Attente des t?ches non termin?es....
> make[1]: quittant le r?pertoire
> ? /devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2 ?
> make: ***
> [/devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2/depmod]
> Erreur 2
> [root at hercules buildroot-2010.05]#
> 
> -> Message means "waiting for unfinished tasks": the "-j 2" option in
> make invocation (defaut = 2) is perhaps the cause. Relaunch!
> 
> 7) make again
> 
> rm -rf /devel/buildroot-2010.05/output/build/buildroot-config
> mkdir -p /devel/buildroot-2010.05/output/build
> cp -dpRf
> package/config/buildroot-config /devel/buildroot-
> 2010.05/output/build/buildroot-config
> /usr/bin/make -j2
> -C /devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2
> make[1]: entrant dans le r?pertoire
> ? /devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2 ?
> /usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o lsmod  lsmod.o
> /usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o insmod  insmod.o
> /usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o rmmod  rmmod.o
> /usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o modprobe  modprobe.o
> zlibsupport.o
> /usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o depmod  depmod.o
> moduleops.o tables.o zlibsupport.o
> /usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o modinfo  modinfo.o
> zlibsupport.o
> /usr/lib64/ccache/gcc  -g -O2 -Wunused -Wall   -o insmod.static -static
> insmod.o
> /usr/bin/ld: cannot find -lc
> collect2: ld a retourn? 1 code d'?tat d'ex?cution
> make[1]: *** [insmod.static] Erreur 1
> make[1]: quittant le r?pertoire
> ? /devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2 ?
> make: ***
> [/devel/buildroot-2010.05/output/toolchain/module-init-tools-3.2.2/depmod]
> Erreur 2
> 
> -> ld perhaps invoked with an empty variable leading to syntax mismatch,
> option taken for parameter or missing one, or something alike. Relaunch!
> 
> 8) make again
> 
> This time, "linux26-menuconfig" is automagically invoked as expected.
> - Just changed: SMP support [disable] / CPU type [ATOM]
> - No optimization nor kernel tayloring for the embedded x86 target.
> Let's make it simple :-)
> - save changes on exit
> make goes on, generating all the regular stuff.
> [...] MODPOST 1975 modules
> [...]
> Root device is (8, 2)
> Setup is 13516 bytes (padded to 13824 bytes).
> System is 3183 kB
> CRC f8992a55
> Kernel: arch/x86/boot/bzImage is ready  (#1)
> [...]
> /devel/buildroot-2010.05/output/target/boot/bzImage -> OK a kernel is
> generated.
> [...]
> >>>   Generating root filesystem image rootfs.ext2
> rm -f /devel/buildroot-2010.05/output/build/_fakeroot.fs
> touch /devel/buildroot-2010.05/output/build/.fakeroot.00000
> cat /devel/buildroot-2010.05/output/build/.fakeroot*
> > /devel/buildroot-2010.05/output/build/_fakeroot.fs
> echo "chown -R 0:0 /devel/buildroot-2010.05/output/target"
> >> /devel/buildroot-2010.05/output/build/_fakeroot.fs
> echo "/devel/buildroot-2010.05/output/host/usr/bin/makedevs -d
> target/generic/device_table.txt /devel/buildroot-2010.05/output/target"
> >> /devel/buildroot-2010.05/output/build/_fakeroot.fs
> echo "
> PATH="/devel/buildroot-2010.05/output/toolchain/bin:/devel/buildroot-
> 2010.05/output/host/bin:/devel/buildroot-
> 2010.05/output/host/usr/bin:/devel/buildroot-
> 2010.05/output/host/usr/sbin/:/devel/buildroot-
> 2010.05/output/staging/bin:/devel/buildroot-
> 2010.05/output/staging/usr/bin:/usr/lib64/qt-
> 3.3/bin:/usr/kerberos/sbin:/usr/kerberos/bin:/usr/lib64/ccache:/usr/local/
> sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin:/root/bin"
> fs/ext2/genext2fs.sh -d /devel/buildroot-2010.05/output/target  -U
> /devel/buildroot-2010.05/output/images/rootfs.ext2" >> /devel/buildroot-
> 2010.05/output/build/_fakeroot.fs
> chmod a+x /devel/buildroot-2010.05/output/build/_fakeroot.fs
> /devel/buildroot-2010.05/output/host/usr/bin/fakeroot
> -- /devel/buildroot-2010.05/output/build/_fakeroot.fs
> rootdir=/devel/buildroot-2010.05/output/target
> table='target/generic/device_table.txt'
> genext2fs: couldn't allocate a block (no free space)
> make: *** [/devel/buildroot-2010.05/output/images/rootfs.ext2] Erreur 1
> 
> -> KO the ext2 fs is NOT generated.
> 
> Of course, there is plenty of Gb available on disk, this error is
> surprising.
> 
> 9) trying to generate something else, i.e cramfs, squashfs and
> initramfs:
> - make menuconfig -> target filesystem option
>         ENABLED cramfs
>         ENABLED squashfs
>         ENABLED initramfs
>         DISABLED ext2
> - save on exit
> 
> 10) make again
> - building of cramfs
> [...]
> >>>   Generating root filesystem image rootfs.cramfs
> warning: estimate of required size (upper bound) is 661MB, but maximum
> image size is 272MB, we might die prematurely
> mkcramfs: filesystem too big
> make: *** [/devel/buildroot-2010.05/output/images/rootfs.cramfs] Erreur
> 8
> 
> 11) let see what we got after attempting to generate either a root fs
> and the corresponding kernel:
> 
> [root at hercules buildroot-2010.05]# ll output/images/
> total 7156
> -rw-r--r-- 1 root root 3272576 juin  13 15:50 bzImage
> -rw-r--r-- 1 root root       0 juin  13 15:50 rootfs.cramfs
> -rw-r--r-- 1 root root 3645440 juin  13 14:08 rootfs.ext2
> -rw-r--r-- 1 root root  394744 juin  13 15:49 rootfs.initramfs
> 
> - kernel seems to be generated (no evidence it will load at this time)
> - rootfs.cramfs isn't generated
> - rootfs.ext2 seems to be generated BUT is not (error above at point 8)
> - rootfs.initramfs is only a text file without any CPIO content:
> 
> [root at hercules buildroot-2010.05]# head output/images/rootfs.initramfs
> 
> #####################
> # /devel/buildroot-2010.05/output/target
> # Last modified: 1276436908.0000000000
> 
> dir /bin 775 0 0
> slink /bin/dmesg busybox 777 0 0
> slink /bin/tar busybox 777 0 0
> slink /bin/watch busybox 777 0 0
> slink /bin/zcat busybox 777 0 0
> 
> [root at hercules buildroot-2010.05]# tail output/images/rootfs.initramfs
> slink /usr/sbin/rdate ../../bin/busybox 777 0 0
> slink /usr/sbin/readprofile ../../bin/busybox 777 0 0
> slink /usr/sbin/setlogcons ../../bin/busybox 777 0 0
> slink /usr/sbin/dnsd ../../bin/busybox 777 0 0
> slink /usr/sbin/chroot ../../bin/busybox 777 0 0
> dir /usr/share 775 0 0
> dir /usr/share/udhcpc 775 0 0
> file /usr/share/udhcpc/default.script /devel/buildroot-
> 2010.05/output/target/usr/share/udhcpc/default.script 755 0 0
> dir /home 775 0 0
> dir /home/default 2755 1000 1000
> [root at hercules buildroot-2010.05]#
> 
> This "rootfs.initramfs" is NOT an initramfs as expected but something
> else I cannot state on.
> 
> 12) additional investigations:
> - asking for a bootloader without kernel prevent errors in target
> filesystems generation, since the size issue is avoided.
> - asking a bootloader in make menuconfig lead to strange things:
>         -> grub only create a /boot/grub subdir in target, but doesn't
> create
> menu.lst nor invoque root (hd0,0) nor setup (hd0,0). Doing it manually
> afterwards is refused.
>         -> sysconfig is an external bootloader, your're right, so
> including it
> in menuconfig could be understood as "prepare a subdir in output/image
> with all the relevant stuff configured, that is: bzImage+ldlinux.sys
> +rootfs.cramfs+syslinux.cfg. Nothing this way is done. What means
> "syslinux" in menuconfig?
> 
> 13) Pending questions...
> - Is buildroot suitable at this time to create tiny x86 embedded
> systems, either by regular filesystem on a disk (ext2) or ramdisk in a
> flashcard (syslinux+initrd) ?

I believe it is, yes.

I have used various versions of buildroot since early 2009 to generate cross
tools, build Linux kernel and a tar of a rootfs for both vortex86SX (486SX)
and vortex86dx (486DX) PC104 embedded X86 boards.  I use the host's syslinux
to generate a Linux boot partition on an IDE flash disk then copy the kernel
across.  I am not sure what you would term tiny, but I am generating tar
rootfs in the order of 2.3MB MB which I then copy across to an ext2 format
partition on the flash disk.

> - Why is there so many unrecoverable errors as I used default presets ?
> - Is there rock-solid examples of bootable x86 systems generated, or
> only "I sware it works, I was aware of an alien CPU card using it long
> time ago in a distant galaxy" :-)))

Yes.  Naturally I have a number of rock-solid examples of configurations for
bootable x86 systems generated by buildroot, as above.  I am running ICOP
VSX-6154 and ICOP VSX-6354 PC104 cards.

> 
> Please remember: this not an "insult", just a "result" ;-)
> 
> Best regards.
> 
> 
> 
> Le dimanche 13 juin 2010 ? 11:19 +0200, Thomas Petazzoni a ?crit :
> > Hello,
> >
> > On Sun, 13 Jun 2010 05:13:14 +0200
> > "sani.broyeur" <sani.broyeur@free.fr> wrote:
> >
> > > - When invoking "make menuconfig && make" an ext2 image is created. OK
> > > but unbootable, without bootloader nor kernel.
> >
> > We've made the choice by default to only build a toolchain + a simple
> > root filesystem image with only Busybox in it. So that's expected.
> >
> > > - When asking to add a bootloader, grub doesn't install properly on
> > > target image, config is not done. This part is broken.
> >
> > What do you expect exactly ? How would you want grub to be installed ?
> > If you're not more specific about what you expect, it's going to be
> > hard to help you.
> >
> > (FYI: I'm currently doing a fairly big bootloader code cleanup, for
> > which I've sent a 49 patches patch set yesterday on this list. At least
> > with this patch set applied, Grub compiles properly. However, I'm not
> > sure what to do with the resulting binaries.)
> >
> > > - When asking to add a CF/FAT16 bootloader, syslinux doesn't install
> > > at all on target image, this part is also broken since it does really
> > > nothing on the image :-)
> >
> > What do you call "target image" ? The ext2 filesystem ? Then it is
> > expected that syslinux isn't installed *in* it. However, the syslinux
> > image is supposed to be available in output/images/.
> >
> > But again, I know syslinux builds properly with my bootloaders-cleanup
> > branch applied. The bootloader code in Buildroot hasn't been
> > unmaintained for quite some time, so he doesn't surprise me that you
> > encounter issues.
> >
> > Another thing is that Buildroot is used a lot for non-x86 architectures
> > (ARM, PowerPC, etc.), and for these architectures, the bootloader is
> > never *in* the root filesystem image. The fact the bootloader is *in*
> > the root filesystem image is a x86-specific thing, and as we haven't
> > had a lot of x86-related contributions lately, it may well be that
> > we're not doing what should be done for this specific architecture.
> >
> > > - When asking a kernel, attempts lead often to cryptic make errors and
> > > the whole buildroot tree has to be recreated from .tar.gz without
> > > kernel generation. The "kernel" menu in menuconfig is broken, and no
> > > kernel config is invoked at build or within menuconfig.
> >
> > Please provide detailed reproduction steps, because ? attemps lead
> > often to cryptic make errors ? is by no means an useful description of
> > what's happening.
> >
> > FWIW :
> >
> >  * I've built succesfully a kernel yesterday with Buildroot, both ARM
> >    and PowerPC.
> >
> >  * I'm working on a big cleanup of the linux compilation code in
> >    Buildroot, because the current code is way too complicated.
> >
> > > - When invoking linux26-menuconfig (I'v found that one time), either
> > > nothing, or make errors, this one is also broken.
> >
> > Be more specific ?
> >
> > > All produced forms of "target" are either incomplete, or unusable, or
> > > unbootable, when the commands are not ending with errors!
> >
> > Be more specific ?
> >
> > > Is there somewhere an USABLE documentation?
> >
> > The only documentation we have so far is
> > http://buildroot.org/downloads/buildroot.html. Patches welcome.
> >
> > > I'm not a newbie to linux, but this "buildroot project" seems for sure
> > > not usable before years of improvements...
> >
> > We certainly have rough edges, particularly in the area of
> > bootloader/kernel compilation, which haven't seen any
> > cleanup/refactoring since a very, very long time.
> >
> > I'm sorry, but your message is by no means useful :
> >
> >  * You're not giving an appropriate level of precision that would give
> >    us the opportunity to understand and solve the problems you're
> >    facing. For each problem: the .config file that allows to produce
> >    the issue, the behaviour that you see, and the behaviour that you
> >    would expect Buildroot to have.
> >
> >  * You're insulting the work done by the developers on the Buildroot
> >    project, reducing their incentive to help you and to solve your
> >    problems.
> >
> > Thomas
> > --
> > Thomas Petazzoni, Free Electrons
> > Kernel, drivers, real-time and embedded Linux
> > development, consulting, training and support.
> > http://free-electrons.com
> > _______________________________________________
> > buildroot mailing list
> > buildroot at busybox.net
> > http://lists.busybox.net/mailman/listinfo/buildroot
> >
> 
> 
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot

[Buildroot] Annoying... nothing work at all :-(

[Jonathan dumaresq]

Hi,

I think you should provide the error messages and .config file too to have
some answer...

Jonathan
buildroot-bounces at busybox.net wrote:
> Hi all.
> 
> I'm trying for several days to get a simple working result with
> buildroot. Nothing. 
> 
> I'm very surprised!
> 
> The source PC is a regular Fedora12 AMD quadcore.
> 
> The target is a MiniITX single board regular Atom PC, able to boot
> with all desktop OS, on a regular disk or a CompactFlash reader. 
> 
> - When invoking "make menuconfig && make" an ext2 image is created.
> OK but unbootable, without bootloader nor kernel. 
> 
> - When asking to add a bootloader, grub doesn't install properly on
> target image, config is not done. This part is broken. 
> 
> - When asking to add a CF/FAT16 bootloader, syslinux doesn't install
> at all on target image, this part is also broken since it does really
> nothing on the image :-)  
> 
> - When asking a kernel, attempts lead often to cryptic make errors
> and the whole buildroot tree has to be recreated from .tar.gz without
> kernel generation. The "kernel" menu in menuconfig is broken, and no
> kernel config is invoked at build or within menuconfig.   
> 
> - When invoking linux26-menuconfig (I'v found that one time), either
> nothing, or make errors, this one is also broken. 
> 
> Etc, etc...
> 
> - Why asking for installation of a bootloader that is never
> installed, nor configured on the target ? 
> 
> - Why asking for generation of a kernel that is never created and
> lead to corruption of buildroot tree ? 
> 
> etc, etc...
> 
> All produced forms of "target" are either incomplete, or unusable, or
> unbootable, when the commands are not ending with errors! 
> 
> Is there somewhere an USABLE documentation?
> 
> A step-by-step tutorial ? With exemple and results ?
> 
> I'm not a newbie to linux, but this "buildroot project" seems for
> sure not usable before years of improvements... 
> 
> Comments would be appreciated.
> 
> Regards.
> 
> 
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot

[Buildroot] Annoying... nothing work at all :-(

[sani.broyeur at free.fr]

Thanks Jonathan, I'm working on it :-)

Let's say that my bad experience is a mix (on my own) of misunderstanding,
unusual config and... one bug found by Thomas :-)

I'll try last snapshot this evening.

Regards.


Selon Jonathan dumaresq <jdumaresq@cimeq.qc.ca>:

> Hi,
>
> I think you should provide the error messages and .config file too to have
> some answer...
>
> Jonathan
> buildroot-bounces at busybox.net wrote:
> > Hi all.
> >
> > I'm trying for several days to get a simple working result with
> > buildroot. Nothing.
> >
> > I'm very surprised!
> >
> > The source PC is a regular Fedora12 AMD quadcore.
> >
> > The target is a MiniITX single board regular Atom PC, able to boot
> > with all desktop OS, on a regular disk or a CompactFlash reader.
> >
> > - When invoking "make menuconfig && make" an ext2 image is created.
> > OK but unbootable, without bootloader nor kernel.
> >
> > - When asking to add a bootloader, grub doesn't install properly on
> > target image, config is not done. This part is broken.
> >
> > - When asking to add a CF/FAT16 bootloader, syslinux doesn't install
> > at all on target image, this part is also broken since it does really
> > nothing on the image :-)
> >
> > - When asking a kernel, attempts lead often to cryptic make errors
> > and the whole buildroot tree has to be recreated from .tar.gz without
> > kernel generation. The "kernel" menu in menuconfig is broken, and no
> > kernel config is invoked at build or within menuconfig.
> >
> > - When invoking linux26-menuconfig (I'v found that one time), either
> > nothing, or make errors, this one is also broken.
> >
> > Etc, etc...
> >
> > - Why asking for installation of a bootloader that is never
> > installed, nor configured on the target ?
> >
> > - Why asking for generation of a kernel that is never created and
> > lead to corruption of buildroot tree ?
> >
> > etc, etc...
> >
> > All produced forms of "target" are either incomplete, or unusable, or
> > unbootable, when the commands are not ending with errors!
> >
> > Is there somewhere an USABLE documentation?
> >
> > A step-by-step tutorial ? With exemple and results ?
> >
> > I'm not a newbie to linux, but this "buildroot project" seems for
> > sure not usable before years of improvements...
> >
> > Comments would be appreciated.
> >
> > Regards.
> >
> >
> > _______________________________________________
> > buildroot mailing list
> > buildroot at busybox.net
> > http://lists.busybox.net/mailman/listinfo/buildroot
>
>
>