[Buildroot] busybox login problem on AT91SAM9263

[Buildroot] busybox login problem on AT91SAM9263

[Stridh, Robin ActeSE]

Hi,

After updating to a fresh buildroot (r24922) configured with at91sam9263ek_defconfig, the busybox login with "root" and no password doesn't work any more. The login prompt shows up, and the login seems to work fine, but then it quits and throws me back to the login prompt again:


Welcome to the Erik's uClibc development environment running on an AT91SAM9261EK
AT91SAM9263EK login: root
Dec 31 17:00:15 login[925]: root login on 'ttyS0'



Welcome to the Erik's uClibc development environment running on an AT91SAM9261EK
AT91SAM9263EK login: 


Changing the busybox version from 1.13.x to 1.10.x solves the problem, so I guess something must have been changed in that time affecting the login process.

Best regards,
Robin Stridh
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.busybox.net/pipermail/buildroot/attachments/20090121/e5b748f2/attachment.htm 

[Buildroot] busybox login problem on AT91SAM9263

[Peter Korsgaard]

>>>>> "Robin" == Stridh, Robin ActeSE <robin.stridh@acteembedded.se> writes:

Hi,

 Robin> Changing the busybox version from 1.13.x to 1.10.x solves the
 Robin> problem, so I guess something must have been changed in that
 Robin> time affecting the login process.

Then it's time to debug. Do you have an /etc/securetty on the target
and does it list ttyS0?

If that's ok, then I would suggest you enable strace, boot with
init=/bin/sh and send the strace output of running login here to the
list.

-- 
Bye, Peter Korsgaard

[Buildroot] busybox login problem on AT91SAM9263

[Stridh, Robin ActeSE]

Hi Peter and thanks for your quick reply!

Peter> Then it's time to debug. Do you have an /etc/securetty on the target
Peter> and does it list ttyS0?

Yes, ttyS0 is listed in /etc/securetty. The only change i made from the non 
working to the working version is changing the busybox version and rebuild buildroot. 

Peter> If that's ok, then I would suggest you enable strace, boot with
Peter> init=/bin/sh and send the strace output of running login here to the
Peter> list.

I tried passing init=/bin/sh to the kernel, but then I end up without a shell. 
The console is still responsive and echoes everything I type and produces 
newlines when I hit return but nothing more.

Regards,
Robin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.busybox.net/pipermail/buildroot/attachments/20090121/9ba5ddae/attachment.htm 

[Buildroot] busybox login problem on AT91SAM9263

[Peter Korsgaard]

>>>>> "Robin" == Stridh, Robin ActeSE <robin.stridh@acteembedded.se> writes:

Hi,

 Robin> Hi Peter and thanks for your quick reply!

You're welcome.

 Peter> Then it's time to debug. Do you have an /etc/securetty on the target
 Peter> and does it list ttyS0?

 Robin> Yes, ttyS0 is listed in /etc/securetty. The only change i made
 Robin> from the non working to the working version is changing the
 Robin> busybox version and rebuild buildroot.

Ok, are you using the default busybox config
(package/busybox/busybox-1.13.x.config) or a custom one?

 Peter> If that's ok, then I would suggest you enable strace, boot with
 Peter> init=/bin/sh and send the strace output of running login here to the
 Peter> list.

 Robin> I tried passing init=/bin/sh to the kernel, but then I end up
 Robin> without a shell.  The console is still responsive and echoes
 Robin> everything I type and produces newlines when I hit return but
 Robin> nothing more.

Strange, that doesn't sound right. What is you console= on the kernel
cmdline set to?

-- 
Bye, Peter Korsgaard

[Buildroot] busybox login problem on AT91SAM9263

[Stridh, Robin ActeSE]

 Peter> Then it's time to debug. Do you have an /etc/securetty on the target
 Peter> and does it list ttyS0?

 Robin> Yes, ttyS0 is listed in /etc/securetty. The only change i made
 Robin> from the non working to the working version is changing the
 Robin> busybox version and rebuild buildroot.

Peter> Ok, are you using the default busybox config
Peter> (package/busybox/busybox-1.13.x.config) or a custom one?

I'm using everything as default as set up by at91sam9263ek_defconfig. When checking
closer I realize the config file used is package/busybox/busybox-1.6.0. A diff
between this one and package/busybox/busybox-1.13.x.config gives a lot of differeces,
so I suspect this might be the cause of the problem. I'll try to recompile with a 
newer config file.

 Peter> If that's ok, then I would suggest you enable strace, boot with
 Peter> init=/bin/sh and send the strace output of running login here to the
 Peter> list.

 Robin> I tried passing init=/bin/sh to the kernel, but then I end up
 Robin> without a shell.  The console is still responsive and echoes
 Robin> everything I type and produces newlines when I hit return but
 Robin> nothing more.

Peter> Strange, that doesn't sound right. What is you console= on the kernel
Peter> cmdline set to?

console=ttyS0,115200

Regards,
Robin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.busybox.net/pipermail/buildroot/attachments/20090121/faeefc2d/attachment.htm 

[Buildroot] busybox login problem on AT91SAM9263

[Stridh, Robin ActeSE]

 Peter> Then it's time to debug. Do you have an /etc/securetty on the target
 Peter> and does it list ttyS0?

 Robin> Yes, ttyS0 is listed in /etc/securetty. The only change i made
 Robin> from the non working to the working version is changing the
 Robin> busybox version and rebuild buildroot.

Peter> Ok, are you using the default busybox config
Peter> (package/busybox/busybox-1.13.x.config) or a custom one?

I'm using everything as default as set up by at91sam9263ek_defconfig. When checking
closer I realize the config file used is package/busybox/busybox-1.6.0. A diff
between this one and package/busybox/busybox-1.13.x.config gives a lot of differeces,
so I suspect this might be the cause of the problem. I'll try to recompile with a 
newer config file.

 Peter> If that's ok, then I would suggest you enable strace, boot with
 Peter> init=/bin/sh and send the strace output of running login here to the
 Peter> list.

 Robin> I tried passing init=/bin/sh to the kernel, but then I end up
 Robin> without a shell.  The console is still responsive and echoes
 Robin> everything I type and produces newlines when I hit return but
 Robin> nothing more.

Peter> Strange, that doesn't sound right. What is you console= on the kernel
Peter> cmdline set to?

console=ttyS0,115200

Regards,
Robin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.busybox.net/pipermail/buildroot/attachments/20090121/0ed6d836/attachment.htm 

[Buildroot] busybox login problem on AT91SAM9263

[Ulf Samuelsson]

ons 2009-01-21 klockan 13:11 +0100 skrev Peter Korsgaard:
> >>>>> "Robin" == Stridh, Robin ActeSE <robin.stridh@acteembedded.se> writes:
> 
> Hi,
> 
>  Robin> Hi Peter and thanks for your quick reply!
> 
> You're welcome.
> 
>  Peter> Then it's time to debug. Do you have an /etc/securetty on the target
>  Peter> and does it list ttyS0?
> 
>  Robin> Yes, ttyS0 is listed in /etc/securetty. The only change i made
>  Robin> from the non working to the working version is changing the
>  Robin> busybox version and rebuild buildroot.
> 
> Ok, are you using the default busybox config
> (package/busybox/busybox-1.13.x.config) or a custom one?

Last time I loaded the SAM9263 board was 10-12 days ago,
and at that time I managed to get the board running
without any such problem, 
- I might have upgraded the uclibc-config though.

BR
Ulf Samuelsson




>  Peter> If that's ok, then I would suggest you enable strace, boot with
>  Peter> init=/bin/sh and send the strace output of running login here to the
>  Peter> list.
> 
>  Robin> I tried passing init=/bin/sh to the kernel, but then I end up
>  Robin> without a shell.  The console is still responsive and echoes
>  Robin> everything I type and produces newlines when I hit return but
>  Robin> nothing more.
> 
> Strange, that doesn't sound right. What is you console= on the kernel
> cmdline set to?
> 

[Buildroot] busybox login problem on AT91SAM9263

[Stridh, Robin ActeSE]

Ulf> Last time I loaded the SAM9263 board was 10-12 days ago,
Ulf> and at that time I managed to get the board running
Ulf> without any such problem, 
Ulf> - I might have upgraded the uclibc-config though.

The log for at91sam9263ek_defconfig shows you last modified it in revision 24822:

svn log target/device/Atmel/at91sam9263ek/at91sam9263ek_defconfig 
------------------------------------------------------------------------
r24822 | ulf | 2009-01-13 19:45:19 +0100 (Tue, 13 Jan 2009) | 1 line

Update at91sam9263ek BSP
------------------------------------------------------------------------
r24790 | ulf | 2009-01-12 11:25:31 +0100 (Mon, 12 Jan 2009) | 1 line

Update sam9263EK  BSP
------------------------------------------------------------------------
r24728 | ulf | 2009-01-09 00:01:50 +0100 (Fri, 09 Jan 2009) | 1 line

Update sam9263ek configs, build completes

I checked out and built this version, an here the login works fine:
svn co -r24822 svn://uclibc.org/trunk/buildroot


The conclusion is that the problem occured somewhere between r24822 and r24922
(jan 13 and jan 21)

Best regards,
Robin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.busybox.net/pipermail/buildroot/attachments/20090122/5ea39022/attachment.htm 

[Buildroot] busybox login problem on AT91SAM9263

[Peter Korsgaard]

>>>>> "Robin" == Stridh, Robin ActeSE <robin.stridh@acteembedded.se> writes:

Hi,

 Robin> The conclusion is that the problem occured somewhere between
 Robin> r24822 and r24922 (jan 13 and jan 21)

Strange, nothing significantly have changed to busybox since then.

Could you do little binary search between those 2 revisions to close
in on the problem commit (E.G. try 24872, if that works then 897, else
847 and so on)?

-- 
Bye, Peter Korsgaard

[Buildroot] busybox login problem on AT91SAM9263

[Stridh, Robin ActeSE]

 Robin> The conclusion is that the problem occured somewhere between
 Robin> r24822 and r24922 (jan 13 and jan 21)

Peter> Strange, nothing significantly have changed to busybox since then.
Peter> 
Peter> Could you do little binary search between those 2 revisions to close
Peter> in on the problem commit (E.G. try 24872, if that works then 897, else
Peter> 847 and so on)?

Well, I could do that, but the build time is quite long for a complete buildroot.
I could try with external toolchain, but if the problem comes from uClibc it 
won't show.

Has anyone else experienced the problem?

Ulf, could you try a recent buildroot rfs on your board?

Regards,
Robin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.busybox.net/pipermail/buildroot/attachments/20090122/0e9c9dd9/attachment.htm 

[Buildroot] busybox login problem on AT91SAM9263

[Peter Korsgaard]

>>>>> "Robin" == Stridh, Robin ActeSE <robin.stridh@acteembedded.se> writes:

Hi,

 Peter> Could you do little binary search between those 2 revisions to
 Peter> close in on the problem commit (E.G. try 24872, if that works
 Peter> then 897, else 847 and so on)?

 Robin> Well, I could do that, but the build time is quite long for a
 Robin> complete buildroot.  I could try with external toolchain, but
 Robin> if the problem comes from uClibc it won't show.

You don't have to trace it completely down to a single commit, but if
you could do it 1 or 2 times it would make it a lot easier to track
down.

 Robin> Has anyone else experienced the problem?

I for one haven't seen it (I'm normally using powerpc though).

-- 
Bye, Peter Korsgaard

[Buildroot] busybox login problem on AT91SAM9263

[Stridh, Robin ActeSE]

Hi,

I configured buildroot to use an external toolchain and tried using the 
toolchain built with the working configuration (r24822) to build the 
current version of the root file system (r24966). Here the login actually 
works!

Using an newer toolchain, but still external (r24936) still gives the
same login problem. Also building todays buildroot with todays toolchain 
gives the same problem. Seems the problem lies within the toolchain.
Ulf: you said something about that you maybe updated the uClibc 
configuration. Can you explain this?

 Peter> Could you do little binary search between those 2 revisions to
 Peter> close in on the problem commit (E.G. try 24872, if that works
 Peter> then 897, else 847 and so on)?

 Robin> Well, I could do that, but the build time is quite long for a
 Robin> complete buildroot.  I could try with external toolchain, but
 Robin> if the problem comes from uClibc it won't show.

Peter> You don't have to trace it completely down to a single commit, but if
Peter> you could do it 1 or 2 times it would make it a lot easier to track
Peter> down.

If I have to rebuild the toolchain each time it will take time, but I will
make some more experiments.

Regards,
Robin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.busybox.net/pipermail/buildroot/attachments/20090122/3e98a680/attachment-0001.htm 

[Buildroot] busybox login problem on AT91SAM9263

[Peter Korsgaard]

>>>>> "Robin" == Stridh, Robin ActeSE <robin.stridh@acteembedded.se> writes:

 Robin> Hi,

 Robin> I configured buildroot to use an external toolchain and tried
 Robin> using the toolchain built with the working configuration
 Robin> (r24822) to build the current version of the root file system
 Robin> (r24966). Here the login actually works!

 Robin> Using an newer toolchain, but still external (r24936) still
 Robin> gives the same login problem. Also building todays buildroot
 Robin> with todays toolchain gives the same problem. Seems the
 Robin> problem lies within the toolchain.  Ulf: you said something
 Robin> about that you maybe updated the uClibc configuration. Can you
 Robin> explain this?

Strange. Between those two versions we changed the default gcc version
from 4.2.4 to 4.3.2 - Are both compilations with gcc 4.2.4?

 Robin> If I have to rebuild the toolchain each time it will take
 Robin> time, but I will make some more experiments.

Thanks, please do it as a binary search instead of more-or-less random
versions.

-- 
Bye, Peter Korsgaard

[Buildroot] busybox login problem on AT91SAM9263

[Stridh, Robin ActeSE]

 Robin> I configured buildroot to use an external toolchain and tried
 Robin> using the toolchain built with the working configuration
 Robin> (r24822) to build the current version of the root file system
 Robin> (r24966). Here the login actually works!

 Robin> Using an newer toolchain, but still external (r24936) still
 Robin> gives the same login problem. Also building todays buildroot
 Robin> with todays toolchain gives the same problem. Seems the
 Robin> problem lies within the toolchain.  Ulf: you said something
 Robin> about that you maybe updated the uClibc configuration. Can you
 Robin> explain this?

Peter> Strange. Between those two versions we changed the default gcc version
Peter> from 4.2.4 to 4.3.2 - Are both compilations with gcc 4.2.4?

Both are built with gcc 4.2.4, binutils 2.18, uClibc 0.9.30

 Robin> If I have to rebuild the toolchain each time it will take
 Robin> time, but I will make some more experiments.

Peter> Thanks, please do it as a binary search instead of more-or-less random
Peter> versions.

Yes, I will try that. The reason for the versions I tried is that I already had
those toolchains built. I'm currently running a new build, but it takes time...

Regards,
Robin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.busybox.net/pipermail/buildroot/attachments/20090122/d1cd152b/attachment.htm 

[Buildroot] busybox login problem on AT91SAM9263

[Stridh, Robin ActeSE]

 Robin> I configured buildroot to use an external toolchain and tried
 Robin> using the toolchain built with the working configuration
 Robin> (r24822) to build the current version of the root file system
 Robin> (r24966). Here the login actually works!

Seems I've been searching a dead end. I compared the configurations and 
found that I had Target ABI set to EABI, but Ulf has used OABI in 
at91sam9263ek_defconfig. When selecting OABI the login works for all 
versions I have tried, including r24966 from today.

I know I have selected EABI in buildroot sucessfully in the past and my kernel 
is built to support both. Now I have no idea when this thing broke - probably 
months ago.

What ABI is recommended for Buildroot? I though that if selectable you should 
always choose EABI since it's newer?

Best regards,
Robin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.busybox.net/pipermail/buildroot/attachments/20090122/c7ea09db/attachment.htm 

[Buildroot] busybox login problem on AT91SAM9263

[Peter Korsgaard]

>>>>> "Robin" == Stridh, Robin ActeSE <robin.stridh@acteembedded.se> writes:

Hi,

 Robin> I know I have selected EABI in buildroot sucessfully in the
 Robin> past and my kernel is built to support both. Now I have no
 Robin> idea when this thing broke - probably months ago.

 Robin> What ABI is recommended for Buildroot? I though that if
 Robin> selectable you should always choose EABI since it's newer?

Well, it's newer, but potentinally less stable (as you see).
Sorry, I don't work much with ARM, so I cannot say much more than that
- Ulf?

-- 
Bye, Peter Korsgaard

[Buildroot] busybox login problem on AT91SAM9263

[Sven Neumann]

Hi,

On Thu, 2009-01-22 at 20:12 +0100, Stridh, Robin ActeSE wrote:

> Seems I've been searching a dead end. I compared the configurations
> and
> found that I had Target ABI set to EABI, but Ulf has used OABI in
> at91sam9263ek_defconfig. When selecting OABI the login works for all
> versions I have tried, including r24966 from today.
> 
> I know I have selected EABI in buildroot sucessfully in the past and
> my kernel
> is built to support both. Now I have no idea when this thing broke -
> probably
> months ago.

Most probably you have run into a similar problem as the one that Daniel
Mack described last week on this list. In the end it turned out that we
had configured gcc with the wrong target name. This resulted in gcc
producing OABI binaries even though EABI was selected in buildroot. See
my mail from 20 Jan 2009 where I suggest that buildroot should try
harder to make sure that the gcc target suffix is set correctly.


Sven

[Buildroot] busybox login problem on AT91SAM9263

[Stridh, Robin ActeSE]

  Robin> What ABI is recommended for Buildroot? I though that if
  Robin> selectable you should always choose EABI since it's newer?
 
 Peter> Well, it's newer, but potentinally less stable (as you see).
 Peter> Sorry, I don't work much with ARM, so I cannot say much more than that
 Peter> - Ulf?
 
Ulf> I OABI works and EABI doesn't then I am not going to drive EABI.
Ulf> Still way too many packages that does not build...

Ok, I guess I'll stick with OABI then as well for now. 
Could someone describe the differences? What is the reason for EABI
if it's not commonly used?

Best regards,
Robin



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.busybox.net/pipermail/buildroot/attachments/20090123/bcbf975a/attachment.htm 

[Buildroot] busybox login problem on AT91SAM9263

[Ulf Samuelsson]

fre 2009-01-23 klockan 12:06 +0100 skrev Peter Korsgaard:
> >>>>> "Robin" == Stridh, Robin ActeSE <robin.stridh@acteembedded.se> writes:
> 
> Hi,
> 
>  Robin> I know I have selected EABI in buildroot sucessfully in the
>  Robin> past and my kernel is built to support both. Now I have no
>  Robin> idea when this thing broke - probably months ago.
> 
>  Robin> What ABI is recommended for Buildroot? I though that if
>  Robin> selectable you should always choose EABI since it's newer?
> 
> Well, it's newer, but potentinally less stable (as you see).
> Sorry, I don't work much with ARM, so I cannot say much more than that
> - Ulf?
> 

I OABI works and EABI doesn't then I am not going to drive EABI.
Still way too many packages that does not build...

BR
Ulf Samuelsson

[Buildroot] busybox login problem on AT91SAM9263

[Sven Neumann]

Hi,

On Fri, 2009-01-23 at 14:02 +0100, Ulf Samuelsson wrote:

> I OABI works and EABI doesn't then I am not going to drive EABI.
> Still way too many packages that does not build...

Oh, please. EABI works nicely as long as the toolchain is correctly
configured. And there are very good reasons to prefer EABI over OABI.
The performance difference is definitely worth it.


Sven

[Buildroot] busybox login problem on AT91SAM9263

[Ulf Samuelsson]

fre 2009-01-23 klockan 15:30 +0100 skrev Sven Neumann:
> Hi,
> 
> On Fri, 2009-01-23 at 14:02 +0100, Ulf Samuelsson wrote:
> 
> > I OABI works and EABI doesn't then I am not going to drive EABI.
> > Still way too many packages that does not build...
> 
> Oh, please. EABI works nicely as long as the toolchain is correctly
> configured. And there are very good reasons to prefer EABI over OABI.
> The performance difference is definitely worth it.
> 

If it is so easy that you just need to change OABI to EABI,
then do I need to get involved?
I do not say that it is a bad idea to have EABI,
but you have to choose your battleground,
and my primary focus is ironing
out inherent issues with current packages.

Everything else has less importance to me, RIGHT NOW.

Since it is a collaborative effort, anyone
can take up the EABI ball.

BR
Ulf Samuelsson

> Sven
> 
> 
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot