request_module: runaway loop modprobe net-pf-1

request_module: runaway loop modprobe net-pf-1

[Ashlesha Shintre]

Hi,

During boot up on the Encore M3 board (AU1500 MIPS) of the 2.6.14.6
kernel, the process stops after the NFS filesystem has been mounted,
memory freed  and spits out the following message:


> request_module: runaway loop modprobe net-pf-1

I can ping to the board and see that the control periodically goes back
to the flush_local_tlb_all function in the arch/mips/mm/tlb-r4k.c file
and prints a line that I have put in..

What does the net-pf-1 mean?

Thanks
Ashlesha.

Re: request_module: runaway loop modprobe net-pf-1

[sjhill]

> During boot up on the Encore M3 board (AU1500 MIPS) of the 2.6.14.6
> kernel, the process stops after the NFS filesystem has been mounted,
> memory freed  and spits out the following message:
> 
> 
> > request_module: runaway loop modprobe net-pf-1
> 
If I had to guess, you have a big endian kernel and you used a little
endian filesystem or the converse.

-Steve

Re: request_module: runaway loop modprobe net-pf-1

[Ralf Baechle]

On Wed, Nov 22, 2006 at 11:42:39AM -0800, Ashlesha Shintre wrote:

> During boot up on the Encore M3 board (AU1500 MIPS) of the 2.6.14.6
> kernel, the process stops after the NFS filesystem has been mounted,
> memory freed  and spits out the following message:
> 
> 
> > request_module: runaway loop modprobe net-pf-1

The kernel tried to open UNIX domain socket but because support is not
compiled it will load the module instead.  Now, glibc-based programs
happen to try to connect to nscd via a UNIX domain socket on startup
and the whole show starts all over.  After a few iterations the kernel
gets tired of the whole game and prints this friendly message.

> What does the net-pf-1 mean?

net-pf-1 is PF_UNIX, see the definitions in include/linux/socket.h.  So
you should set CONFIG_UNIX to y.  Building it as a module won't work
as you just found :).

  Ralf

Re: request_module: runaway loop modprobe net-pf-1

[Ashlesha Shintre]

On Wed, 2006-11-22 at 22:17 +0000, Ralf Baechle wrote:
> On Wed, Nov 22, 2006 at 11:42:39AM -0800, Ashlesha Shintre wrote:
> 
> > During boot up on the Encore M3 board (AU1500 MIPS) of the 2.6.14.6
> > kernel, the process stops after the NFS filesystem has been mounted,
> > memory freed  and spits out the following message:
> > 
> > 
> > > request_module: runaway loop modprobe net-pf-1
> 
> The kernel tried to open UNIX domain socket but because support is not
> compiled it will load the module instead.  Now, glibc-based programs
> happen to try to connect to nscd via a UNIX domain socket on startup
> and the whole show starts all over.  After a few iterations the kernel
> gets tired of the whole game and prints this friendly message.
> 
> > What does the net-pf-1 mean?
> 
> net-pf-1 is PF_UNIX, see the definitions in include/linux/socket.h.  So
> you should set CONFIG_UNIX to y.  

Thanks for your reply Ralf, but I dont see he CONFIG_UNIX option in my
Makefile, so I created one, but still get the same error!
is there anything else that i should try?

Thanks again,
Ashlesha.
> Building it as a module won't work
> as you just found :).
> 
>   Ralf

Re: request_module: runaway loop modprobe net-pf-1

[mlachwani]

CONFIG_UNIX is defined in net/unix/Kconfig or simply do a "make 
menuconfig" and search for UNIX.

It depends on CONFIG_NET. Check your .config ...

thanks,
Manish Lachwani


Ashlesha Shintre wrote:
> On Wed, 2006-11-22 at 22:17 +0000, Ralf Baechle wrote:
>   
>> On Wed, Nov 22, 2006 at 11:42:39AM -0800, Ashlesha Shintre wrote:
>>
>>     
>>> During boot up on the Encore M3 board (AU1500 MIPS) of the 2.6.14.6
>>> kernel, the process stops after the NFS filesystem has been mounted,
>>> memory freed  and spits out the following message:
>>>
>>>
>>>       
>>>> request_module: runaway loop modprobe net-pf-1
>>>>         
>> The kernel tried to open UNIX domain socket but because support is not
>> compiled it will load the module instead.  Now, glibc-based programs
>> happen to try to connect to nscd via a UNIX domain socket on startup
>> and the whole show starts all over.  After a few iterations the kernel
>> gets tired of the whole game and prints this friendly message.
>>
>>     
>>> What does the net-pf-1 mean?
>>>       
>> net-pf-1 is PF_UNIX, see the definitions in include/linux/socket.h.  So
>> you should set CONFIG_UNIX to y.  
>>     
>
> Thanks for your reply Ralf, but I dont see he CONFIG_UNIX option in my
> Makefile, so I created one, but still get the same error!
> is there anything else that i should try?
>
> Thanks again,
> Ashlesha.
>   
>> Building it as a module won't work
>> as you just found :).
>>
>>   Ralf
>>     
>
>
>   

init cannot spawn shell

[Ashlesha Shintre]

Hi,

So the ">>>> request_module: runaway loop modprobe net-pf-1" problem is
now solved, I dont get the same error again --
Thank you Ralf, Manish and Steve for your prompt responses..

However, the kernel gets stuck now after mounting the filesystem and
freeing memory.  It does not execute the shell.  When I give a boot
argument such as init=/bin/sh, it attempts to kill init.

I m sure that the endianness is not an issue - everything is big endian.

I statically compiled a hello world program and instead of the rcS
script in the init.d directory, put in this entry in the inittab

sys::sysinit:/etc/init.d/hello

however, I dont see "Hello World" on the console! -- meaning the problem
is in userspace?

Any ideas on how to debug this situation?

Thanks,
Ashlesha.


On Wed, 2006-11-22 at 16:10 -0800, mlachwani wrote:
> CONFIG_UNIX is defined in net/unix/Kconfig or simply do a "make 
> menuconfig" and search for UNIX.
> 
> It depends on CONFIG_NET. Check your .config ...
> 
> thanks,
> Manish Lachwani
> 
> 
> Ashlesha Shintre wrote:
> > On Wed, 2006-11-22 at 22:17 +0000, Ralf Baechle wrote:
> >   
> >> On Wed, Nov 22, 2006 at 11:42:39AM -0800, Ashlesha Shintre wrote:
> >>
> >>     
> >>> During boot up on the Encore M3 board (AU1500 MIPS) of the 2.6.14.6
> >>> kernel, the process stops after the NFS filesystem has been mounted,
> >>> memory freed  and spits out the following message:
> >>>
> >>>
> >>>       
> >>>> request_module: runaway loop modprobe net-pf-1
> >>>>         
> >> The kernel tried to open UNIX domain socket but because support is not
> >> compiled it will load the module instead.  Now, glibc-based programs
> >> happen to try to connect to nscd via a UNIX domain socket on startup
> >> and the whole show starts all over.  After a few iterations the kernel
> >> gets tired of the whole game and prints this friendly message.
> >>
> >>     
> >>> What does the net-pf-1 mean?
> >>>       
> >> net-pf-1 is PF_UNIX, see the definitions in include/linux/socket.h.  So
> >> you should set CONFIG_UNIX to y.  
> >>     
> >
> > Thanks for your reply Ralf, but I dont see he CONFIG_UNIX option in my
> > Makefile, so I created one, but still get the same error!
> > is there anything else that i should try?
> >
> > Thanks again,
> > Ashlesha.
> >   
> >> Building it as a module won't work
> >> as you just found :).
> >>
> >>   Ralf
> >>     
> >
> >
> >   
> 

RE: init cannot spawn shell

[Andre.Messerschmidt]

Hi,
 

> I m sure that the endianness is not an issue - everything is big 
> endian.

We had some problem in the past, where the reverse endianess bit was
set, that would result in a crash when the system starts init. Not sure
what bootloader you are using, but maybe it is worth checking the RE bit
in CP0.STATUS, since otherwise your userspace applications will run in
little endian.

Regards
Andre

RE: init cannot spawn shell

[Andre.Messerschmidt]

Hi,
 

> I m sure that the endianness is not an issue - everything is big 
> endian.

We had some problem in the past, where the reverse endianess bit was
set, that would result in a crash when the system starts init. Not sure
what bootloader you are using, but maybe it is worth checking the RE bit
in CP0.STATUS, since otherwise your userspace applications will run in
little endian.

Regards
Andre

Re: init cannot spawn shell

[Ralf Baechle]

On Thu, Nov 23, 2006 at 07:45:25AM +0100, Andre.Messerschmidt@infineon.com wrote:

> > I m sure that the endianness is not an issue - everything is big 
> > endian.
> 
> We had some problem in the past, where the reverse endianess bit was
> set, that would result in a crash when the system starts init. Not sure
> what bootloader you are using, but maybe it is worth checking the RE bit
> in CP0.STATUS, since otherwise your userspace applications will run in
> little endian.

This was fixed in February for 2.6.16 already.  Also Ashleha is using
a different CPU ...

  Ralf