linux-apus-devel

libc reduction and powerpc/apus woody boot floppies.

[Sven L. <lu...@dp...>]

Hello, ...

As i was no more able to test the boot floppies for the apus powerpc subarch,
for lack of working hardware and time, Giorgio Terzi did volunteer to do the
work. Many thanks go to him for this.

After some time, here is what he did find :

----- Forwarded message from Giorgio Terzi <de...@ip...> -----

now i can load new root.bin image!!!!

The problem was that the reduced copy of libc2.2.2 won't work!
I have sobstituted it with the standard long sized one and all 
worked well !

Simple it isn't ? But maked me sweat! :)))))

Note that the reduced libld2.2.2 works. 

Now there are 2 alternative strategies: 
1) the build software must avoid to reduce that library.
2) To fix the bug in the library

----- End forwarded message -----

Note : The symptoms were that init (from busybox) did not launch correctly,
which makes boot-floppies rather unusable on powerpc/apus right now.

Now, i have been out of boot-floppies development for a long time, and also am
not familiar at all libc's internal.

What is the best way to solve this, should we fill a bug against busybox,
boot-floppies or libc6 ?

Is this a apus specific problem, or something related to the powerpc arch ?

Is it possible to build the boot floppies while not reducing libc ? or is
there size constraints (apus don't has 1.4 MB floppies anyway, so it is just a
file on the harddisk) ?

Any help on further work will be welcome ...

Friendly,

Sven Luther

Re: libc reduction and powerpc/apus woody boot floppies.

[Ben C. <bco...@de...>]

On Tue, May 01, 2001 at 06:01:59PM +0200, Sven LUTHER wrote:
> 
> Note : The symptoms were that init (from busybox) did not launch correctly,
> which makes boot-floppies rather unusable on powerpc/apus right now.

What exactly do you mean by "did not launch correctly"? Were there
missing symbols? Did it segv or sigbus, or something else? Note, I
testes a reduced libc with bash, and it worked fine. Perhaps something
in busybox is broken? Maybe you are building with a different libc6-pic
package on your system that the libc6 that is installed on
boot-floppies?

Ben

-- 
 -----------=======-=-======-=========-----------=====------------=-=------
/  Ben Collins  --  ...on that fantastic voyage...  --  Debian GNU/Linux   \
`  bco...@de...  --  bco...@op...  --  bco...@li...  '
 `---=========------=======-------------=-=-----=-===-======-------=--=---'

Re: libc reduction and powerpc/apus woody boot floppies.

[Sven L. <lu...@dp...>]

On Tue, May 01, 2001 at 03:19:26PM -0400, Ben Collins wrote:
> On Tue, May 01, 2001 at 06:01:59PM +0200, Sven LUTHER wrote:
> > 
> > Note : The symptoms were that init (from busybox) did not launch correctly,
> > which makes boot-floppies rather unusable on powerpc/apus right now.
> 
> What exactly do you mean by "did not launch correctly"? Were there
> missing symbols? Did it segv or sigbus, or something else? Note, I

I did not test this myself, as i have not the hardware, but the report of
Giorgio Terzi is that there is no output anywhere. Err, this is not really
understandable, isn't it ? 

Anyway, this is when using the busybox init to launch the boot-floppies
root.bin root image. What happens is that the kernel boots, then tries to
launch INIT (from busybox) as the parent process, and fails, without any
fursther output, not even the log on VC 3 or 4. And since the only way to go
out of that is rebooting, there is no significative log file to look trough.

Now, it seems the same init will work well when launched from the shell, not
sure though, at least dbootstrap will.

Did i get that right Giorgio ?

> testes a reduced libc with bash, and it worked fine. Perhaps something
> in busybox is broken? Maybe you are building with a different libc6-pic
> package on your system that the libc6 that is installed on
> boot-floppies?

mmm, yes, this seems like a plausible thing to me, busybox is built with the
installed libc6, but then is tried to run with the reduced one on the root.bin
root image. This would explain this.

How can we test that this is the case ? And if it is, how should we build
busybox to have it working right ? Maybe Girgio did not have the good version
of libc6 installed when building busybox, but then, this would be a
build-depend bug or something such in busybox ?

Thanks for caring,

Friendly,

Sven Luther

Re: libc reduction and powerpc/apus woody boot floppies.

[Giorgio T. <de...@ip...>]

Hello Sven,

i had the time to download the mail only today, 
sorry...but i was very busy working. 

Your question:

The dimension of the normal libc2.2.2 is about 1.2megs
and its reduced counterpart is about 700 kbytes.
Note that the previous reduced libc 2.1.3 (from potato root-disk) 
was about 900 kbytes, this can be not so important but has 
left me some suspects... 
Ah, forgotten to mail you that tuesday i have uploaded the floppy image
to your ftp site.

--------------------------------------------------------------
 
For the mailing list:

Please don't think i have made any type of in-deep modification: 
i have worked in this test only sobstituting files, 
with a "follow the stream" logic. :)))

Last but not least: I don't know nothing about libc incarnations,
modifications etc.

For a better explanation here are my testing steps:
 
1) In first i have built floppy image "as-is" downloading all the
   debs needed, letting the software to create all the floppy-images
   for all the platforms involved.
   The build process worked ok.
   
   Dbootstrap worked if called by a shell.
   I was not sure of Busybox's init subroutine because was impossible  
   to test it in a right manner by shell. 
   (almost something worked... what is device z6 ? It complained for it).
   I have tried some other its builtin commands as cd, ls, du 
   and they worked.  

2) I have tested the "as-is" root1440apus.bin.
   Failed.
   It locked just before or during init process because after the following 
   kernel messages:
      Ramdisk: compressed image found at block 0
      VFS Mounted root (ext2 filesystem)
      Freing unused kernel memory: etc...
   all the processes stopped.
   The cursor stopped but was blinking... 
   probably something was still running or waiting :)))
   The 2.2.10 kernel i have used was compiled with the syskeys ctrl.
   With this i have seen that the processes that was running was:
      1) init
      2) kflushd
      3) kpiod
      4) kswapd
      5) md_thread
   So init process was started but was locked before the inittab reading
   because there wasn't ash messages after, and none VC was working.

3) I have downloaded busybox source and compiled it with the general debug
   option enabled and the special debug options in the init.c file too.
   I was hoping for some debug messages.
   I have loop mounted the image file and i have sobstituted the old busybox 
   with the new that i've compiled. 
   Test Failed: same results as above.
   
4) Changed strategy. 
   I have sobstituted only new busybox ( the standard version ) and 
   the new dbootstrap in a old working root-floppy image 
   (taken from potato)
   Test Succeded :  the root.image bootstrapped, but new dbootstrap 
   complained because i had left the old libc2.1.3 in that old image.
   
5) Sobstituted old library with the new reduced libc2.2.2.  
   Test Failed: got locked as the first tries.

6) Tried to sobstitute the small library with the standard big libc2.2.2 
   Test Succeded: the image began to bootstrap another time. 

Of course i have sobstituted it also in the "as-is" image and it began to work
too. It works also with our last 2.4.3 kernel.
As you can see, what i have done was a simple file sobstitution.
Nothing more.

I don't want to give any opinion about because i don't know in-deep what
libc does. Also is the same for dbootstrap and busybox.

But i have a question:
May be now dboostrap & busybox need more resources than in the past?  

Friendly,

-- 
                         Giorgio Terzi

Re: libc reduction and powerpc/apus woody boot floppies.

[Sven L. <lu...@dp...>]

On Thu, May 03, 2001 at 06:21:36PM +0200, Giorgio Terzi wrote:
> 6) Tried to sobstitute the small library with the standard big libc2.2.2 
>    Test Succeded: the image began to bootstrap another time. 

Yes, i guess this is the real problem, ...

Ben Collins (the libc maintainer) seems to think that this behavior is because
you are using the wrong libc will linking the busybox executable.

I cannot do it here, but please mount the root.bin root image, and try an ldd
on the sbin/init executable.

BTW, i am not sure about what the meaning of the following are :

EXT2-fs warning: mounting unchecked fs, running e2fsck is recommended
VFS: Mounted root (ext2 filesystem).
Freeing unused kernel memory: 100k init 32k prep 4k pmac 8k open firmware
Warning: unable to open an initial console.
Restarting system.

It tells you about being unable to open an initial console, why is that ? Does
this happen early, or later on when you try rebooting. In any case, this could
be due to missing devices, or somethine wrong with the fbcon in the kernel,
not sure though, maybe some of the other apus guys have more idea on this ? Or
even to the libc6 linkage problem.

Anyway we have to make sure that when you build busybox, it is against the
reduced libc6, which will go onto the boot floppy, and not the full one which
you have installed on your system.

Please also send me the output of :

$ dpkg -l libc6\*

And maybe if you could ftp me a log of the build of busybox ? simply type :

$ scirpt /tmp/busybox.build.log

before you build, and once it is builded, you can :

$ exit

and then send me the /tmp/busybox.buil.log file.

Friendly,

Sven Luther

Re: libc reduction and powerpc/apus woody boot floppies.

[Giorgio T. <de...@ip...>]

Hello Sven,
This is VERY important!!!
...
> On Thu, May 03, 2001 at 06:21:36PM +0200, Giorgio Terzi wrote:
>> 6) Tried to sobstitute the small library with the standard big libc2.2.2 
>>    Test Succeded: the image began to bootstrap another time. 
> 
> Yes, i guess this is the real problem, ...
> 
> Ben Collins (the libc maintainer) seems to think that this behavior is
> because you are using the wrong libc will linking the busybox executable.

I must apologize.
Sorry, i wasn't clear in my explanation. 
When i wrote "standard version" about busybox i was meaning:  

"the version taken from busybox debian package that is automatically 
added when boot-floppyes are build"!

From the 4th step to end of my past e-mail i haven't used nothing
of build by myself!

I am very, very sorry. 

Regards
-- 
                                                 Giorgio Terzi

Re: libc reduction and powerpc/apus woody boot floppies.

[Sven L. <lu...@dp...>]

On Fri, May 04, 2001 at 05:20:33PM +0200, Giorgio Terzi wrote:
> Hello Sven,
> This is VERY important!!!
> ...
> > On Thu, May 03, 2001 at 06:21:36PM +0200, Giorgio Terzi wrote:
> >> 6) Tried to sobstitute the small library with the standard big libc2.2.2 
> >>    Test Succeded: the image began to bootstrap another time. 
> > 
> > Yes, i guess this is the real problem, ...
> > 
> > Ben Collins (the libc maintainer) seems to think that this behavior is
> > because you are using the wrong libc will linking the busybox executable.
> 
> I must apologize.
> Sorry, i wasn't clear in my explanation. 
> When i wrote "standard version" about busybox i was meaning:  
> 
> "the version taken from busybox debian package that is automatically 
> added when boot-floppyes are build"!
> 
> >From the 4th step to end of my past e-mail i haven't used nothing
> of build by myself!
> 
> I am very, very sorry. 

No problem, please run ldd on init, and tell us about it.

mmm, can do this also myself, will be doing it right now.

Friendly,

Sven Luther

Re: libc reduction and powerpc/apus woody boot floppies.

[Sven L. <lu...@dp...>]

On Fri, May 04, 2001 at 05:20:33PM +0200, Giorgio Terzi wrote:
> Hello Sven,
> This is VERY important!!!
> ...
> > On Thu, May 03, 2001 at 06:21:36PM +0200, Giorgio Terzi wrote:
> >> 6) Tried to sobstitute the small library with the standard big libc2.2.2 
> >>    Test Succeded: the image began to bootstrap another time. 
> > 
> > Yes, i guess this is the real problem, ...
> > 
> > Ben Collins (the libc maintainer) seems to think that this behavior is
> > because you are using the wrong libc will linking the busybox executable.
> 
> I must apologize.
> Sorry, i wasn't clear in my explanation. 
> When i wrote "standard version" about busybox i was meaning:  
> 
> "the version taken from busybox debian package that is automatically 
> added when boot-floppyes are build"!
> 
> >From the 4th step to end of my past e-mail i haven't used nothing
> of build by myself!
> 
> I am very, very sorry. 

Sorry, forgot,n cannot run ldd on it, am not using a ppc system :(((

Friendly,

Sven Luther

Re: libc reduction and powerpc/apus woody boot floppies.

[Giorgio T. <de...@ip...>]

Hello Sven,
these are the informations you've asked me taken from
the linux installation i used for boot-floppy building:

# dpkg -l 'libc6*'

Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Installed/Config-files/Unpacked/Failed-config/Half-installed
|/ Err?=(none)/Hold/Reinst-required/X=both-problems (Status,Err: uppercase=bad)
||/ Name           Version        Description
+++-==============-==============-============================================
ii  libc6          2.2.2-1        GNU C Library: Shared libraries and Timezone
un  libc6-bin      <none>         (no description available)
pn  libc6-dbg      <none>         (no description available)
ii  libc6-dev      2.2.2-1        GNU C Library: Development Libraries and Hea
un  libc6-doc      <none>         (no description available)
ii  libc6-pic      2.2.2-1        GNU C Library: PIC archive library
pn  libc6-prof     <none>         (no description available)
un  libc6.1        <none>         (no description available)


# dpkg -l 'busy*'

Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Installed/Config-files/Unpacked/Failed-config/Half-installed
|/ Err?=(none)/Hold/Reinst-required/X=both-problems (Status,Err: uppercase=bad)
||/ Name           Version        Description
+++-==============-==============-============================================
ii  busybox        0.51-1         Tiny utilities for small and embedded system
pn  busybox-static <none>         (no description available)


# ldd -v -d -r /sbin/init

    libc.so.6 => /lib/libc.so.6 (0x0feb3000)
    /lib/ld.so.1 => /lib/ld.so.1 (0x30000000)

    Version information:
    /sbin/init:
        libc.so.6 (GLIBC_2.1) => /lib/libc.so.6
        libc.so.6 (GLIBC_2.0) => /lib/libc.so.6
    /lib/libc.so.6:
        ld.so.1 (GLIBC_2.1.1) => /lib/ld.so.1
        ld.so.1 (GLIBC_2.1) => /lib/ld.so.1
        ld.so.1 (GLIBC_2.2) => /lib/ld.so.1
        ld.so.1 (GLIBC_2.0) => /lib/ld.so.1


# ldd -v -d -r /bin/busybox

    libc.so.6 => /lib/libc.so.6 (0x0feb3000)
    /lib/ld.so.1 => /lib/ld.so.1 (0x30000000)

    Version information:
    /bin/busybox:
        libc.so.6 (GLIBC_2.1) => /lib/libc.so.6
        libc.so.6 (GLIBC_2.0) => /lib/libc.so.6
    /lib/libc.so.6:
        ld.so.1 (GLIBC_2.1.1) => /lib/ld.so.1
        ld.so.1 (GLIBC_2.1) => /lib/ld.so.1
        ld.so.1 (GLIBC_2.2) => /lib/ld.so.1
        ld.so.1 (GLIBC_2.0) => /lib/ld.so.1
  
Regards,
-- 
                                                 Giorgio Terzi

Re: libc reduction and powerpc/apus woody boot floppies.

[Sven L. <lu...@dp...>]

On Sat, May 05, 2001 at 11:33:50AM +0200, Giorgio Terzi wrote:
> Hello Sven,
> these are the informations you've asked me taken from
> the linux installation i used for boot-floppy building:
> 
> # dpkg -l 'libc6*'
> 
> Desired=Unknown/Install/Remove/Purge/Hold
> | Status=Not/Installed/Config-files/Unpacked/Failed-config/Half-installed
> |/ Err?=(none)/Hold/Reinst-required/X=both-problems (Status,Err: uppercase=bad)
> ||/ Name           Version        Description
> +++-==============-==============-============================================
> ii  libc6          2.2.2-1        GNU C Library: Shared libraries and Timezone
> un  libc6-bin      <none>         (no description available)
> pn  libc6-dbg      <none>         (no description available)
> ii  libc6-dev      2.2.2-1        GNU C Library: Development Libraries and Hea
> un  libc6-doc      <none>         (no description available)
> ii  libc6-pic      2.2.2-1        GNU C Library: PIC archive library
> pn  libc6-prof     <none>         (no description available)
> un  libc6.1        <none>         (no description available)
> 
> 
> # dpkg -l 'busy*'
> 
> Desired=Unknown/Install/Remove/Purge/Hold
> | Status=Not/Installed/Config-files/Unpacked/Failed-config/Half-installed
> |/ Err?=(none)/Hold/Reinst-required/X=both-problems (Status,Err: uppercase=bad)
> ||/ Name           Version        Description
> +++-==============-==============-============================================
> ii  busybox        0.51-1         Tiny utilities for small and embedded system
> pn  busybox-static <none>         (no description available)
> 

Ok, fine ...

> # ldd -v -d -r /sbin/init
> 
>     libc.so.6 => /lib/libc.so.6 (0x0feb3000)
>     /lib/ld.so.1 => /lib/ld.so.1 (0x30000000)

Is this the /lib of the root disk, or the one from the normal install.

My guess is that this is the /lib of the normal install, and thus the reason
for the init program not working on the root disk.

Could you try mounting the root image, and then chroot to it (you need to
install that package).

Then you can try one of the other functions of the busybox (i don't know, but
i guess something like ls or so will do, or even mount).

Try this either in the chrooted environment and in the nomral one.

Anyway, i am not sure i can help you much here, we need help from the
boot-floppies people, ...

That said, since you are using the precompiled busybox, this clearly warrants
a bug against busybox.

Friendly,

Sven Luther

Re: libc reduction and powerpc/apus woody boot floppies.

[Erik A. <and...@co...>]

On Sun May 06, 2001 at 11:28:42AM +0200, Sven LUTHER wrote:
> 
> Then you can try one of the other functions of the busybox (i don't know, but
> i guess something like ls or so will do, or even mount).

All busybox functions are symlinks to /bin/busybox.  So running more
'ldd' checks isn't going to tell us much.  

> Try this either in the chrooted environment and in the nomral one.
> 
> Anyway, i am not sure i can help you much here, we need help from the
> boot-floppies people, ...
> 
> That said, since you are using the precompiled busybox, this clearly warrants
> a bug against busybox.

I'm dubious.  I still havn't figured out exactly what the problem is.
So far I have only seen vague guesses to a problem, but I don't see
what the original problem is that people are guessing about.
So far, I'm guessing that the problem is some version of boot floppies
of powerpc failed to boot...

Is this a serial console?  What kernel verison is being used?
Do we have a log of what does happen on boot?

 -Erik

--
Erik B. Andersen   email:  and...@de..., and...@li...
--This message was written using 73% post-consumer electrons--

Re: libc reduction and powerpc/apus woody boot floppies.

[Sven L. <lu...@dp...>]

On Sun, May 06, 2001 at 09:47:42PM -0600, Erik Andersen wrote:
> On Sun May 06, 2001 at 11:28:42AM +0200, Sven LUTHER wrote:
> > 
> > Then you can try one of the other functions of the busybox (i don't know, but
> > i guess something like ls or so will do, or even mount).
> 
> All busybox functions are symlinks to /bin/busybox.  So running more
> 'ldd' checks isn't going to tell us much.  

well, but running ls (it is a busybox function isn't it ?) will enable you to
test if the libc compiled in will work or not, without doing things we want to
do with init.

Just to see if it will work in a chrooted environment.

My guess, conforted by ben collins mail, is that the busybox is linked toward
the not reduced libc6 that is in the standard install, while there is the
reduced version in the root image.

This can be tested with chrooting into the boot image and testing there if any
busybox program will run, isn't it ?

Note : i have no ppc box myself, so i cannot test it, Giorgio volunteered to do
the building/testing, altough he is not a debian developper, and thus not
really all that much familiar with boot floppies internals.

> > Try this either in the chrooted environment and in the nomral one.
> > 
> > Anyway, i am not sure i can help you much here, we need help from the
> > boot-floppies people, ...
> > 
> > That said, since you are using the precompiled busybox, this clearly warrants
> > a bug against busybox.
> 
> I'm dubious.  I still havn't figured out exactly what the problem is.
> So far I have only seen vague guesses to a problem, but I don't see
> what the original problem is that people are guessing about.

See above, do you think this seems reasonable ?

> So far, I'm guessing that the problem is some version of boot floppies
> of powerpc failed to boot...

Well, it boots all right, but init will not work, and as thus he gets only
blinking cursor on VC1 and nothing on VC 2-3.

he noticed that chaging the reduced libc6 in the root image by the not reduced
one, then it will boot.

As for me, i have only experience with the potato boot floppies on apus, not
worked on them since then, and no access to a ppc box.

> Is this a serial console?  What kernel verison is being used?

No it is not, standard (fbdev) console. Kernel being either 2.2.10 (latest
2.2.x on apus) or 2.4.x.

> Do we have a log of what does happen on boot?

Well, since he has no serial console, and init is causing problem, there is no
meaningfull log, apart from what you will find at :

ftp://lambda.u-strasbg.fr/boot-floppies

Friendly,

Sven Luther
> 
>  -Erik
> 
> --
> Erik B. Andersen   email:  and...@de..., and...@li...
> --This message was written using 73% post-consumer electrons--