Re: [s390] networking related oops during boot on Hercules (was: build failure)

Re: [s390] networking related oops during boot on Hercules (was: build failure)

[Eric W. Biederman]

Christian Borntraeger <borntraeger@de.ibm.com> writes:

> Am Mittwoch, 28. November 2007 schrieb Frans Pop:
> [...]
>> During boot I get the following oops in the Hercules emulator.
>> 2.6.22 runs fine on Hercules; I've not tried .23 on it.
> [...]
>> ------------  cut here !------------
>> kernel BUG at net/core/dev.c:852 +
>> illegal operation: 0001  #1!
>> Modules linked in: ctc fsm tape_34xx cu3088 tape ccwgroup tape_class dm_mirror
> d
>> m_snapshot dm_mod dasd_eckd_mod dasd_mod
>> CPU:    1    Not tainted
>> Process hwup (pid: 990, task: 0f034c00, ksp: 0f305be0)
>> Krnl PSW : 070c0000 8019d0da (dev_alloc_name+0x1e/0x58)
>>            R:0 T:1 IO:1 EX:1 Key:0 M:1 W:0 P:0 AS:0 CC:0 PM:0
>> Krnl GPRS: 00132700 00000000 0f398800 0f398800
>>            00000000 80132796 0f305d7a 00000002
>>            0f1ada00 0f3c1f08 0f3c1f00 0f398800
>>            0f398800 8019d0c2 0f305ca0 0f305c30
>> Krnl Code: 8019d0cc: f000bf1f2460       srp     3871(1,%r11),1120(%r2),0
>>            8019d0d2: a7740004           brc     7,8019d0da
>>            8019d0d6: a7f40001           brc     15,8019d0d8
>>           >8019d0da: 5810d04e           l       %r1,78(%r13)
>>            8019d0de: 41a0f060           la      %r10,96(%r15)
>>            8019d0e2: 5820c460           l       %r2,1120(%r12)
>>            8019d0e6: 184a               lr      %r4,%r10
>>            8019d0e8: 0de1               basr    %r14,%r1
>> Call Trace:
>> ( <0000000000000000>! _ehead+0xfffee000/0x80)
>>   <000000000019d710>! register_netdev+0x34/0x6c
>>   <00000000108a561e>! ctc_new_device+0x3ee/0x590  ctc!
>>   <0000000010861398>! ccwgroup_online_store+0xb0/0x13c  ccwgroup!
>>   <00000000000c962a>! sysfs_write_file+0xca/0x130
>> <00000000000840e6>! vfs_write+0x92/0x128 +
>>   <00000000000847e8>! sys_write+0x40/0x70
>>   <000000000002098a>! sysc_do_restart+0x12/0x16
>>   <0000000077f0abaa>! 0x77f0abaa
>
>
> This seems to be related to the new network namespace code by Eric
> Biederman (CCed). Can you try the following (untested) patch? I also 
> CCed Ursula and Peter as they know the ctc code better than me.

Can you instead change the allocation to be alloc_netdev instead of
just a plain old kzalloc.    That will fix the problem in a much
more forward compatible way.  The current code is also going to
have problems with dynamically sized array at the end of struct
net_device.

You ought to also be able to make the allocation unconditional
as that function only has one caller and it does not pass in
a struct net_device.

Eric


>
> CC: Eric W. Biederman <ebiederm@xmission.com>
> CC: Ursula Braun <braunu@de.ibm.com>
> CC: Peter Tiedemann <ptiedem@de.ibm.com>
> Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
>
> ---
>  drivers/s390/net/ctcmain.c |    2 ++
>  1 file changed, 2 insertions(+)
>
> Index: linux-2.6/drivers/s390/net/ctcmain.c
> ===================================================================
> --- linux-2.6.orig/drivers/s390/net/ctcmain.c
> +++ linux-2.6/drivers/s390/net/ctcmain.c
> @@ -56,6 +56,7 @@
>  #include <linux/skbuff.h>
>  #include <linux/ctype.h>
>  #include <net/dst.h>
> +#include <net/net_namespace.h>
>  
>  #include <asm/io.h>
>  #include <asm/ccwdev.h>
> @@ -2823,6 +2824,7 @@ ctc_init_netdevice(struct net_device * d
>  	dev->type = ARPHRD_SLIP;
>  	dev->tx_queue_len = 100;
>  	dev->flags = IFF_POINTOPOINT | IFF_NOARP;
> +	dev->nd_net = &init_net;
>  	return dev;
>  }
>  

Re: [s390] networking related oops during boot on Hercules (was: build failure)

[Stephen Hemminger]

On Wed, 28 Nov 2007 10:26:36 -0700
ebiederm@xmission.com (Eric W. Biederman) wrote:

> Christian Borntraeger <borntraeger@de.ibm.com> writes:
> 
> > Am Mittwoch, 28. November 2007 schrieb Frans Pop:
> > [...]
> >> During boot I get the following oops in the Hercules emulator.
> >> 2.6.22 runs fine on Hercules; I've not tried .23 on it.
> > [...]
> >> ------------  cut here !------------
> >> kernel BUG at net/core/dev.c:852 +
> >> illegal operation: 0001  #1!
> >> Modules linked in: ctc fsm tape_34xx cu3088 tape ccwgroup tape_class dm_mirror
> > d
> >> m_snapshot dm_mod dasd_eckd_mod dasd_mod
> >> CPU:    1    Not tainted
> >> Process hwup (pid: 990, task: 0f034c00, ksp: 0f305be0)
> >> Krnl PSW : 070c0000 8019d0da (dev_alloc_name+0x1e/0x58)
> >>            R:0 T:1 IO:1 EX:1 Key:0 M:1 W:0 P:0 AS:0 CC:0 PM:0
> >> Krnl GPRS: 00132700 00000000 0f398800 0f398800
> >>            00000000 80132796 0f305d7a 00000002
> >>            0f1ada00 0f3c1f08 0f3c1f00 0f398800
> >>            0f398800 8019d0c2 0f305ca0 0f305c30
> >> Krnl Code: 8019d0cc: f000bf1f2460       srp     3871(1,%r11),1120(%r2),0
> >>            8019d0d2: a7740004           brc     7,8019d0da
> >>            8019d0d6: a7f40001           brc     15,8019d0d8
> >>           >8019d0da: 5810d04e           l       %r1,78(%r13)
> >>            8019d0de: 41a0f060           la      %r10,96(%r15)
> >>            8019d0e2: 5820c460           l       %r2,1120(%r12)
> >>            8019d0e6: 184a               lr      %r4,%r10
> >>            8019d0e8: 0de1               basr    %r14,%r1
> >> Call Trace:
> >> ( <0000000000000000>! _ehead+0xfffee000/0x80)
> >>   <000000000019d710>! register_netdev+0x34/0x6c
> >>   <00000000108a561e>! ctc_new_device+0x3ee/0x590  ctc!
> >>   <0000000010861398>! ccwgroup_online_store+0xb0/0x13c  ccwgroup!
> >>   <00000000000c962a>! sysfs_write_file+0xca/0x130
> >> <00000000000840e6>! vfs_write+0x92/0x128 +
> >>   <00000000000847e8>! sys_write+0x40/0x70
> >>   <000000000002098a>! sysc_do_restart+0x12/0x16
> >>   <0000000077f0abaa>! 0x77f0abaa
> >
> >
> > This seems to be related to the new network namespace code by Eric
> > Biederman (CCed). Can you try the following (untested) patch? I also 
> > CCed Ursula and Peter as they know the ctc code better than me.
> 
> Can you instead change the allocation to be alloc_netdev instead of
> just a plain old kzalloc.    That will fix the problem in a much
> more forward compatible way.  The current code is also going to
> have problems with dynamically sized array at the end of struct
> net_device.
> 
> You ought to also be able to make the allocation unconditional
> as that function only has one caller and it does not pass in
> a struct net_device.
> 
> Eric
> 

Any code that calls register_netdev with a kzalloc'd pointer is
fundamentally flawed.  Read Documentation/networking/netdevices