[Buildroot] init segfaults when compiled with glibc

[Buildroot] init segfaults when compiled with glibc

[Sasha Sirotkin]

I compiled an external toolchain with glibc support using crosstools-ng 
as recommended. I managed to compile buildroot without any issues. 
However, when I try to use the resulting image /sbin/init segfaults. 
With uclibc there are no such issues. I'm using the latest versions of 
everything on ARM with EABI.

Any suggestions ?

[Buildroot] init segfaults when compiled with glibc

[Yann E. MORIN]

Sasha, All,

On Tuesday 30 March 2010 142701 Sasha Sirotkin wrote:
> I compiled an external toolchain with glibc support using crosstools-ng 
> as recommended. I managed to compile buildroot without any issues. 
> However, when I try to use the resulting image /sbin/init segfaults. 
> With uclibc there are no such issues. I'm using the latest versions of 
> everything on ARM with EABI.

By default, gcc will emit armv5t instructions. What's your target CPU?
init segfaulting can be a symptom of using a bad instruction set, so if
your CPU is armv4, and you did not tell gcc to default to armv4, then
you will hit this segfault issue.

regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +0/33 662376056 | Software  Designer | \ / CAMPAIGN     |   ^                |
| --==< O_o >==-- '------------.-------:  X  AGAINST      |  /e\  There is no  |
| http://ymorin.is-a-geek.org/ | (*_*) | / \ HTML MAIL    |  """  conspiracy.  |
'------------------------------'-------'------------------'--------------------'

[Buildroot] init segfaults when compiled with glibc

[Sasha Sirotkin]

On Tue, Mar 30, 2010 at 4:20 PM, Yann E. MORIN 
<yann.morin.1998@anciens.enib.fr> wrote:

    Sasha, All,

    On Tuesday 30 March 2010 142701 Sasha Sirotkin wrote:
     > I compiled an external toolchain with glibc support using
    crosstools-ng
     > as recommended. I managed to compile buildroot without any issues.
     > However, when I try to use the resulting image /sbin/init segfaults.
     > With uclibc there are no such issues. I'm using the latest
    versions of
     > everything on ARM with EABI.

    By default, gcc will emit armv5t instructions. What's your target CPU?
    init segfaulting can be a symptom of using a bad instruction set, so if
    your CPU is armv4, and you did not tell gcc to default to armv4, then
    you will hit this segfault issue.



As I already mentioned, it works fine with uClibc, so it is definitely 
not a CPU instructions issue. At any rate, my CPU is AT91, i.e. armvte 
instruction set.

[Buildroot] init segfaults when compiled with glibc

[Yann E. MORIN]

Sasha, All,

On Tuesday 30 March 2010 16:26:14 Sasha Sirotkin wrote:
> On Tue, Mar 30, 2010 at 4:20 PM, Yann E. MORIN 
> <yann.morin.1998@anciens.enib.fr> wrote:
>     On Tuesday 30 March 2010 142701 Sasha Sirotkin wrote:
>      > I compiled an external toolchain with glibc support using
>     crosstools-ng
>      > as recommended. I managed to compile buildroot without any issues.
>      > However, when I try to use the resulting image /sbin/init segfaults.

Standard init, or busybox' ?

>      > With uclibc there are no such issues. I'm using the latest
>     versions of
>      > everything on ARM with EABI.

In crosstool-NG, there's a sample that is quite up-to-date and that builds
an arm-unknown-linux-gnueabi:
# ct-ng show-arm-unknown-linux-gnueabi
arm-unknown-linux-gnueabi  [l  ]
    OS             : linux-2.6.31.12
    Companion libs : gmp-4.3.1 mpfr-2.4.2 libelf-0.8.12
    binutils       : binutils-2.19.1
    C compiler     : gcc-4.3.2 (C,C++,Fortran,Java)
    C library      : glibc-2.9
    Tools          : sstrip dmalloc-5.5.2 duma-2_5_15 gdb-6.8 ltrace-0.5.3 strace-4.5.19

>     By default, gcc will emit armv5t instructions. What's your target CPU?
>     init segfaulting can be a symptom of using a bad instruction set, so if
>     your CPU is armv4, and you did not tell gcc to default to armv4, then
>     you will hit this segfault issue.
> As I already mentioned, it works fine with uClibc, so it is definitely 
> not a CPU instructions issue. At any rate, my CPU is AT91, i.e. armvte 
> instruction set.

Ahem. Right, sorry.

I'll read you back on crossgcc... :-)

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'

[Buildroot] init segfaults when compiled with glibc

[Sasha Sirotkin]

On Tue, Mar 30, 2010 at 7:58 PM, Yann E. MORIN 
<yann.morin.1998@anciens.enib.fr> wrote:

    Sasha, All,

    On Tuesday 30 March 2010 16:26:14 Sasha Sirotkin wrote:
     > On Tue, Mar 30, 2010 at 4:20 PM, Yann E. MORIN
     > <yann.morin.1998 at anciens.enib.fr
    <mailto:yann.morin.1998@anciens.enib.fr>> wrote:
     >     On Tuesday 30 March 2010 142701 Sasha Sirotkin wrote:
     >      > I compiled an external toolchain with glibc support using
     >     crosstools-ng
     >      > as recommended. I managed to compile buildroot without any
    issues.
     >      > However, when I try to use the resulting image /sbin/init
    segfaults.

    Standard init, or busybox' ?

     >      > With uclibc there are no such issues. I'm using the latest
     >     versions of
     >      > everything on ARM with EABI.

    In crosstool-NG, there's a sample that is quite up-to-date and that
    builds
    an arm-unknown-linux-gnueabi:
    # ct-ng show-arm-unknown-linux-gnueabi
    arm-unknown-linux-gnueabi  [l  ]
       OS             : linux-2.6.31.12
       Companion libs : gmp-4.3.1 mpfr-2.4.2 libelf-0.8.12
       binutils       : binutils-2.19.1
       C compiler     : gcc-4.3.2 (C,C++,Fortran,Java)
       C library      : glibc-2.9
       Tools          : sstrip dmalloc-5.5.2 duma-2_5_15 gdb-6.8
    ltrace-0.5.3 strace-4.5.19

     >     By default, gcc will emit armv5t instructions. What's your
    target CPU?
     >     init segfaulting can be a symptom of using a bad instruction
    set, so if
     >     your CPU is armv4, and you did not tell gcc to default to
    armv4, then
     >     you will hit this segfault issue.
     > As I already mentioned, it works fine with uClibc, so it is
    definitely
     > not a CPU instructions issue. At any rate, my CPU is AT91, i.e.
    armvte
     > instruction set.

    Ahem. Right, sorry.

    I'll read you back on crossgcc... :-)



It happens with both busybox and standard init.

Compilation of crossgcc with EABI work OK, but than I got this 
segmentation fault issue...

It looks like glibc does not work at all with buildroot right now. 
Should I open a bug ticket or something?

[Buildroot] init segfaults when compiled with glibc

[Thomas Petazzoni]

Hi Sasha,

On Tue, 30 Mar 2010 23:03:55 +0300
Sasha Sirotkin <buildroot@browserseal.com> wrote:

> It looks like glibc does not work at all with buildroot right now. 
> Should I open a bug ticket or something?

Unfortunately, without more details, a bug report would be fairly
useless. Can you send your Buildroot .config and kernel .config so that
I can try to reproduce your problem on my AT91 systems ?

Thanks,

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

[Buildroot] init segfaults when compiled with glibc

[Eric BENARD]

Le 30/03/2010 16:26, Sasha Sirotkin a ?crit :
> As I already mentioned, it works fine with uClibc, so it is definitely
> not a CPU instructions issue. At any rate, my CPU is AT91, i.e. armvte
> instruction set.
>
AT91RM9200 is armv4

Eric

[Buildroot] init segfaults when compiled with glibc

[Sasha Sirotkin]

On Tue, Mar 30, 2010 at 11:34 PM, Thomas Petazzoni 
<thomas.petazzoni@free-electrons.com> wrote:

    Hi Sasha,

    On Tue, 30 Mar 2010 23:03:55 +0300
    Sasha Sirotkin <buildroot@browserseal.com> wrote:

    > It looks like glibc does not work at all with buildroot right now.
    > Should I open a bug ticket or something?

    Unfortunately, without more details, a bug report would be fairly
    useless. Can you send your Buildroot .config and kernel .config so that
    I can try to reproduce your problem on my AT91 systems ?

Attached please find both config files. 
Thanks.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: buildroot.config.bz2
Type: application/x-bzip
Size: 3643 bytes
Desc: not available
URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20100331/a0cc21e1/attachment-0002.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: kernel.config.bz2
Type: application/x-bzip
Size: 7096 bytes
Desc: not available
URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20100331/a0cc21e1/attachment-0003.bin>

[Buildroot] init segfaults when compiled with glibc

[Alexander Sirotkin]

On Wed, Mar 31, 2010 at 1:20 AM, Eric BENARD <ebenard@free.fr> wrote:

> Le 30/03/2010 16:26, Sasha Sirotkin a ?crit :
>
>  As I already mentioned, it works fine with uClibc, so it is definitely
>> not a CPU instructions issue. At any rate, my CPU is AT91, i.e. armvte
>> instruction set.
>>
>>  AT91RM9200 is armv4
>
You are right, nevertheless the isssue, i.e. init sudden death is not
related to the instruction set - as I already mentioned, it works just fine
with uClibc. The problem appears only when I use external glibc based
toolchain.


>
> Eric
>
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20100331/689a90b7/attachment.htm>

[Buildroot] init segfaults when compiled with glibc

[Javier Viguera]

Alexander (Sasha) Sirotkin wrote:
> 
> You are right, nevertheless the isssue, i.e. init sudden death is not 
> related to the instruction set - as I already mentioned, it works just 
> fine with uClibc. The problem appears only when I use external glibc 
> based toolchain.

I had the same problem a month ago more or less. With a glibc toolchain 
built with crosstool-NG the init process was dying and the kernel 
panic'ing as a result because init (pid=1) was not supposed to die.

Unfortunately i could not manage to find what the problem was. All i 
found is that it was happening with some versions of glibc while it was 
working with other versions: glibc-2.9 was failing, so my workaround was 
finally to select EGLIBC-2.10 to build the toolchain. With that version 
it worked for me.

Give it a try to a newer version of glibc/eglibc (starting with 2.10) 
and let us know if it works.

-- 
Regards,

Javier Viguera
Software Engineer

Digi International Spain, S.A.
http://www.digi.com

[Buildroot] init segfaults when compiled with glibc

[Alexander Sirotkin]

On Wed, Mar 31, 2010 at 5:37 PM, Javier Viguera <javier.viguera@digi.com>wrote:

> Alexander (Sasha) Sirotkin wrote:
>
>>
>> You are right, nevertheless the isssue, i.e. init sudden death is not
>> related to the instruction set - as I already mentioned, it works just fine
>> with uClibc. The problem appears only when I use external glibc based
>> toolchain.
>>
>
> I had the same problem a month ago more or less. With a glibc toolchain
> built with crosstool-NG the init process was dying and the kernel panic'ing
> as a result because init (pid=1) was not supposed to die.
>
> Unfortunately i could not manage to find what the problem was. All i found
> is that it was happening with some versions of glibc while it was working
> with other versions: glibc-2.9 was failing, so my workaround was finally to
> select EGLIBC-2.10 to build the toolchain. With that version it worked for
> me.
>

I did not manage to compile crosstoll-ng with eglibc yet. It looks like
eglibc support is broken in crosstool-ng and glibc support is broken in
buildroot.

Dead end...


>
> Give it a try to a newer version of glibc/eglibc (starting with 2.10) and
> let us know if it works.
>
> --
> Regards,
>
> Javier Viguera
> Software Engineer
>
> Digi International Spain, S.A.
> http://www.digi.com
>
>
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20100331/6c65079c/attachment-0001.htm>

[Buildroot] init segfaults when compiled with glibc

[Peter Korsgaard]

>>>>> "Alexander" == Alexander (Sasha) Sirotkin <buildroot@browserseal.com> writes:

 Alexander> I did not manage to compile crosstoll-ng with eglibc yet. It
 Alexander> looks like eglibc support is broken in crosstool-ng and
 Alexander> glibc support is broken in buildroot.

uClibc? We don't have glibc support (except for external toolchains), so
it cannot really be broken ;)

-- 
Bye, Peter Korsgaard

[Buildroot] init segfaults when compiled with glibc

[Sasha Sirotkin]

Peter Korsgaard wrote:
>>>>>> "Alexander" == Alexander (Sasha) Sirotkin <buildroot@browserseal.com> writes:
>>>>>>             
>
>  Alexander> I did not manage to compile crosstoll-ng with eglibc yet. It
>  Alexander> looks like eglibc support is broken in crosstool-ng and
>  Alexander> glibc support is broken in buildroot.
>
> uClibc? We don't have glibc support (except for external toolchains), so
> it cannot really be broken ;)
>
>   
Well... it is broken in a sense that buildroot does not work with an 
external toolchain based on glibc. Please see above.

[Buildroot] init segfaults when compiled with glibc

[Thomas Petazzoni]

On Wed, 31 Mar 2010 13:47:13 +0300
Sasha Sirotkin <buildroot@browserseal.com> wrote:

> Attached please find both config files. 

Which exact AT91 CPU are you using ?

Your Buildroot .config specifies an ARM926 core, which is ARMv5.
However, if you use AT91RM9200, as ?ric B?nard noted, you have an
ARM920 core, which is ARMv4. Therefore, if you have an AT91RM9200 CPU,
it is normal that the Buildroot configuration you sent does not work.
Select BR2_arm920t instead of BR2_arm926t.

Your .config also states that you're using a crosstool-NG external
toolchain. Can you send the .config you used in Crosstool-NG to
generate the toolchain ?

In other words:
 * Give us more details about your hardware
 * Give us your Crosstool-NG configuration file

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

[Buildroot] init segfaults when compiled with glibc

[H Hartley Sweeten]

On Wednesday, March 31, 2010 11:28 AM, Sasha Sirotkin wrote:
> Peter Korsgaard wrote:
>>>>>>> "Alexander" == Alexander (Sasha) Sirotkin <buildroot@browserseal.com> writes:
>>>>>>>             
>>
>>  Alexander> I did not manage to compile crosstoll-ng with eglibc yet. It
>>  Alexander> looks like eglibc support is broken in crosstool-ng and
>>  Alexander> glibc support is broken in buildroot.
>>
>> uClibc? We don't have glibc support (except for external toolchains), so
>> it cannot really be broken ;)
>>
>>   
> Well... it is broken in a sense that buildroot does not work with an 
> external toolchain based on glibc. Please see above.

FWIW, I'm using an external glibc toolchain (created with crosstoo-NG)
with no problems.

Regards,
Hartley

[Buildroot] init segfaults when compiled with glibc

[Sasha Sirotkin]

H Hartley Sweeten wrote:
> On Wednesday, March 31, 2010 11:28 AM, Sasha Sirotkin wrote:
>   
>> Peter Korsgaard wrote:
>>     
>>>>>>>> "Alexander" == Alexander (Sasha) Sirotkin <buildroot@browserseal.com> writes:
>>>>>>>>             
>>>>>>>>                 
>>>  Alexander> I did not manage to compile crosstoll-ng with eglibc yet. It
>>>  Alexander> looks like eglibc support is broken in crosstool-ng and
>>>  Alexander> glibc support is broken in buildroot.
>>>
>>> uClibc? We don't have glibc support (except for external toolchains), so
>>> it cannot really be broken ;)
>>>
>>>   
>>>       
>> Well... it is broken in a sense that buildroot does not work with an 
>> external toolchain based on glibc. Please see above.
>>     
>
> FWIW, I'm using an external glibc toolchain (created with crosstoo-NG)
> with no problems.
>
> Regards,
> Hartley
I'd be extremely grateful to know how you managed to do this. As you can 
see above I could not compile crosstool-ng with eglibc and when I used 
glibc version created with crosstool-ng I ran into this "init death" 
problem that at least one other person could reproduce.

[Buildroot] init segfaults when compiled with glibc

[Thomas Petazzoni]

On Wed, 31 Mar 2010 21:28:29 +0300
Sasha Sirotkin <buildroot@browserseal.com> wrote:

> Well... it is broken in a sense that buildroot does not work with an 
> external toolchain based on glibc. Please see above.

Could you please stop making false assertions, and start being nice so
that we can have some interest in helping you ?

The fact that using Buildroot, with glibc external toolchain, does not
work in *your* *particular* configuration, does not mean at all that
Buildroot does not support glibc external toolchains. Just like
Hartley, I successfully build and run systems generated by Buildroot,
and that used a glibc external toolchain.

It is quite possible that the configuration mistake is on your side, so
instead of being offensive about the status of Buildroot, could you
just be nice and help us to constructively solve the problem you're
having ?

Thanks!

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

[Buildroot] init segfaults when compiled with glibc

[Alexander Sirotkin]

On Wed, Mar 31, 2010 at 9:52 PM, Thomas Petazzoni
<thomas.petazzoni@free-electrons.com> wrote:
> On Wed, 31 Mar 2010 13:47:13 +0300
> Sasha Sirotkin <buildroot@browserseal.com> wrote:
>
>> Attached please find both config files.
>
> Which exact AT91 CPU are you using ?

AT91SAM9260

>
> Your Buildroot .config specifies an ARM926 core, which is ARMv5.
> However, if you use AT91RM9200, as ?ric B?nard noted, you have an
> ARM920 core, which is ARMv4. Therefore, if you have an AT91RM9200 CPU,
> it is normal that the Buildroot configuration you sent does not work.
> Select BR2_arm920t instead of BR2_arm926t.
>
> Your .config also states that you're using a crosstool-NG external
> toolchain. Can you send the .config you used in Crosstool-NG to
> generate the toolchain ?

Attached.

>
> In other words:
> ?* Give us more details about your hardware
> ?* Give us your Crosstool-NG configuration file
>
> 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
-------------- next part --------------
A non-text attachment was scrubbed...
Name: .config.bz2
Type: application/x-bzip2
Size: 2299 bytes
Desc: not available
URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20100331/f1db35fd/attachment.bin>

[Buildroot] init segfaults when compiled with glibc

[Alexander Sirotkin]

On Wed, Mar 31, 2010 at 10:16 PM, Thomas Petazzoni
<thomas.petazzoni@free-electrons.com> wrote:
> On Wed, 31 Mar 2010 21:28:29 +0300
> Sasha Sirotkin <buildroot@browserseal.com> wrote:
>
>> Well... it is broken in a sense that buildroot does not work with an
>> external toolchain based on glibc. Please see above.
>
> Could you please stop making false assertions, and start being nice so
> that we can have some interest in helping you ?

Sorry about that, it is simply a frustration of somebody who's been
trying to compile buildroot with glibc for a few weeks already.

>
> The fact that using Buildroot, with glibc external toolchain, does not
> work in *your* *particular* configuration, does not mean at all that
> Buildroot does not support glibc external toolchains. Just like
> Hartley, I successfully build and run systems generated by Buildroot,
> and that used a glibc external toolchain.
>
> It is quite possible that the configuration mistake is on your side, so
> instead of being offensive about the status of Buildroot, could you
> just be nice and help us to constructively solve the problem you're
> having ?
>
> Thanks!
>
> 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] init segfaults when compiled with glibc

[Alexander Sirotkin]

On Wed, Mar 31, 2010 at 5:37 PM, Javier Viguera <javier.viguera@digi.com> wrote:
> Alexander (Sasha) Sirotkin wrote:
>>
>> You are right, nevertheless the isssue, i.e. init sudden death is not
>> related to the instruction set - as I already mentioned, it works just fine
>> with uClibc. The problem appears only when I use external glibc based
>> toolchain.
>
> I had the same problem a month ago more or less. With a glibc toolchain
> built with crosstool-NG the init process was dying and the kernel panic'ing
> as a result because init (pid=1) was not supposed to die.
>
> Unfortunately i could not manage to find what the problem was. All i found
> is that it was happening with some versions of glibc while it was working
> with other versions: glibc-2.9 was failing, so my workaround was finally to
> select EGLIBC-2.10 to build the toolchain. With that version it worked for
> me.
>
> Give it a try to a newer version of glibc/eglibc (starting with 2.10) and
> let us know if it works.

Eventually I managed to compile crosstool-ng with eglibc 2.10 after
downgrading the compiler to 4.3.2 from 4.4.1 and binutils to 2.19 from
2.20. This one seems to work OK with buildroot.

>
> --
> Regards,
>
> Javier Viguera
> Software Engineer
>
> Digi International Spain, S.A.
> http://www.digi.com
>
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
>