2004-01-21 06:39:12

by Valdis Kletnieks

[permalink] [raw]
Subject: 2.6.1-mm5 - oops during network initialization

(linux-net people, please cc: on replies, am only on lkml)

Under 2.6.1-mm4, at boot I'd get the following:

Jan 20 10:00:46 turing-police kernel: Initializing IPsec netlink socket
Jan 20 10:00:46 turing-police kernel: NET: Registered protocol family 1
Jan 20 10:00:46 turing-police kernel: NET: Registered protocol family 10
Jan 20 10:00:46 turing-police kernel: IPv6 over IPv4 tunneling driver
Jan 20 10:00:46 turing-police kernel: NET: Registered protocol family 17
Jan 20 10:00:46 turing-police kernel: NET: Registered protocol family 15
Jan 20 10:00:46 turing-police kernel: RAMDISK: Compressed image found at block 0

and the initrd would kick off and we'd be happy.

Under 2.6.1-mm5, I get this: (hand-copied..

NET: Registered protocol family 10
Unable to handle kernel NULL pointer dereference at virtual address 00000068
printing eip:
c01186f9
*pde = 00000000
Oops: 0000 [#1]
PREEMPT
CPU: 0
EIP: 0060:[<c01180f9>] Not tainted VLI
EFLAGS: 0010217
EIP is at do_page_fault+0x53/0x4b2
eax: cfe84000 ebx: cfe86000 ecx: 0000007b edx: 00000000
esi: 00000000 edi: 00000000 ebp: cfe84048 esp: cfe8405c
ds: 007b es: 007b ss: 0068
process ksoftirqd/0 (pid:2, threadinfo=cfe82000 task=cff81310
stack: 00000000 00000068 00000000 00000000 00000000 00030001 00000000 00000000
00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
call trace:

(I had to stick a 'for (;;);' into the code at this point to keep it from scrolling off the
screen - was do_page_fault and 2 other routines in a loop over and over again).

So it's choking somewhere in IPv6 init. Only change I can spot in -mm5 in that
area is ipv6-sysctl-oops-fix.patch but I'm not seeing how that one can *cause* this oops.

For the record, built *without* regparm-3, with the Fedora gcc-3.3.2-5 compiler.

IPv6-related .config:

CONFIG_IPV6=y
CONFIG_IPV6_PRIVACY=y
CONFIG_INET6_AH=y
CONFIG_INET6_ESP=y
CONFIG_INET6_IPCOMP=y
# CONFIG_IPV6_TUNNEL is not set
# CONFIG_DECNET is not set
# CONFIG_BRIDGE is not set
CONFIG_NETFILTER=y
# CONFIG_NETFILTER_DEBUG is not set

#
# IPv6: Netfilter Configuration
#
# CONFIG_IP6_NF_QUEUE is not set
CONFIG_IP6_NF_IPTABLES=m
CONFIG_IP6_NF_MATCH_LIMIT=m
CONFIG_IP6_NF_MATCH_MAC=m
CONFIG_IP6_NF_MATCH_RT=m
CONFIG_IP6_NF_MATCH_OPTS=m
CONFIG_IP6_NF_MATCH_FRAG=m
CONFIG_IP6_NF_MATCH_HL=m
CONFIG_IP6_NF_MATCH_MULTIPORT=m
CONFIG_IP6_NF_MATCH_OWNER=m
CONFIG_IP6_NF_MATCH_MARK=m
CONFIG_IP6_NF_MATCH_IPV6HEADER=m
CONFIG_IP6_NF_MATCH_AHESP=m
CONFIG_IP6_NF_MATCH_LENGTH=m
CONFIG_IP6_NF_MATCH_EUI64=m
CONFIG_IP6_NF_FILTER=m
CONFIG_IP6_NF_TARGET_LOG=m
CONFIG_IP6_NF_MANGLE=m
CONFIG_IP6_NF_TARGET_MARK=m
CONFIG_XFRM=y
CONFIG_XFRM_USER=y

#
# SCTP Configuration (EXPERIMENTAL)
#
CONFIG_IPV6_SCTP__=y

I'll play binary-search on the IPv6 config options, see if one of them is
involved, but that will have to wait for morning....


Attachments:
(No filename) (226.00 B)

Subject: Re: 2.6.1-mm5 - oops during network initialization

On Wed, 21 Jan 2004, Valdis Kletnieks wrote:

I can confirm I get this also.

> (linux-net people, please cc: on replies, am only on lkml)
>
> Under 2.6.1-mm4, at boot I'd get the following:
>
> Jan 20 10:00:46 turing-police kernel: Initializing IPsec netlink socket
> Jan 20 10:00:46 turing-police kernel: NET: Registered protocol family 1
> Jan 20 10:00:46 turing-police kernel: NET: Registered protocol family 10
> Jan 20 10:00:46 turing-police kernel: IPv6 over IPv4 tunneling driver
> Jan 20 10:00:46 turing-police kernel: NET: Registered protocol family 17
> Jan 20 10:00:46 turing-police kernel: NET: Registered protocol family 15
> Jan 20 10:00:46 turing-police kernel: RAMDISK: Compressed image found at block 0
>
> and the initrd would kick off and we'd be happy.
>
> Under 2.6.1-mm5, I get this: (hand-copied..
>
> NET: Registered protocol family 10
> Unable to handle kernel NULL pointer dereference at virtual address 00000068
> printing eip:
> c01186f9
> *pde = 00000000
> Oops: 0000 [#1]
> PREEMPT
> CPU: 0
> EIP: 0060:[<c01180f9>] Not tainted VLI
> EFLAGS: 0010217
> EIP is at do_page_fault+0x53/0x4b2
> eax: cfe84000 ebx: cfe86000 ecx: 0000007b edx: 00000000
> esi: 00000000 edi: 00000000 ebp: cfe84048 esp: cfe8405c
> ds: 007b es: 007b ss: 0068
> process ksoftirqd/0 (pid:2, threadinfo=cfe82000 task=cff81310
> stack: 00000000 00000068 00000000 00000000 00000000 00030001 00000000 00000000
> 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> call trace:
>
> (I had to stick a 'for (;;);' into the code at this point to keep it from scrolling off the
> screen - was do_page_fault and 2 other routines in a loop over and over again).
>
> So it's choking somewhere in IPv6 init. Only change I can spot in -mm5 in that
> area is ipv6-sysctl-oops-fix.patch but I'm not seeing how that one can *cause* this oops.
>
> For the record, built *without* regparm-3, with the Fedora gcc-3.3.2-5 compiler.
>
> IPv6-related .config:
>
> CONFIG_IPV6=y
> CONFIG_IPV6_PRIVACY=y
> CONFIG_INET6_AH=y
> CONFIG_INET6_ESP=y
> CONFIG_INET6_IPCOMP=y
> # CONFIG_IPV6_TUNNEL is not set
> # CONFIG_DECNET is not set
> # CONFIG_BRIDGE is not set
> CONFIG_NETFILTER=y
> # CONFIG_NETFILTER_DEBUG is not set
>
> #
> # IPv6: Netfilter Configuration
> #
> # CONFIG_IP6_NF_QUEUE is not set
> CONFIG_IP6_NF_IPTABLES=m
> CONFIG_IP6_NF_MATCH_LIMIT=m
> CONFIG_IP6_NF_MATCH_MAC=m
> CONFIG_IP6_NF_MATCH_RT=m
> CONFIG_IP6_NF_MATCH_OPTS=m
> CONFIG_IP6_NF_MATCH_FRAG=m
> CONFIG_IP6_NF_MATCH_HL=m
> CONFIG_IP6_NF_MATCH_MULTIPORT=m
> CONFIG_IP6_NF_MATCH_OWNER=m
> CONFIG_IP6_NF_MATCH_MARK=m
> CONFIG_IP6_NF_MATCH_IPV6HEADER=m
> CONFIG_IP6_NF_MATCH_AHESP=m
> CONFIG_IP6_NF_MATCH_LENGTH=m
> CONFIG_IP6_NF_MATCH_EUI64=m
> CONFIG_IP6_NF_FILTER=m
> CONFIG_IP6_NF_TARGET_LOG=m
> CONFIG_IP6_NF_MANGLE=m
> CONFIG_IP6_NF_TARGET_MARK=m
> CONFIG_XFRM=y
> CONFIG_XFRM_USER=y
>
> #
> # SCTP Configuration (EXPERIMENTAL)
> #
> CONFIG_IPV6_SCTP__=y
>
> I'll play binary-search on the IPv6 config options, see if one of them is
> involved, but that will have to wait for morning....
>

---
Catalin(ux) BOIE
[email protected]

2004-01-21 16:01:58

by Laszlo 'GCS' Boszormenyi

[permalink] [raw]
Subject: Re: 2.6.1-mm5 - oops during network initialization

On Wed, Jan 21, 2004 at 10:31:23AM +0200, Catalin BOIE <[email protected]> wrote:
> On Wed, 21 Jan 2004, Valdis Kletnieks wrote:
>
> I can confirm I get this also.
> > CONFIG_IPV6_PRIVACY=y
Can you both try it without the above? At least it's solved my problem, and
I can have 'CONFIG_IPV6=y' and ipv6 netfilter options as modules.

Hope this helps,
GCS

2004-01-21 17:00:10

by Valdis Klētnieks

[permalink] [raw]
Subject: Re: 2.6.1-mm5 - oops during network initialization

On Wed, 21 Jan 2004 16:46:27 +0100, GCS said:

> > > CONFIG_IPV6_PRIVACY=y
> Can you both try it without the above? At least it's solved my problem, and
> I can have 'CONFIG_IPV6=y' and ipv6 netfilter options as modules.

Confirm on that. Same config, turn off CONFIG_IPV6_PRIVACY, and the
kernel boots just fine. I'm willing to test patches if needed....


Attachments:
(No filename) (226.00 B)

2004-01-21 18:58:26

by Andrew Morton

[permalink] [raw]
Subject: Re: 2.6.1-mm5 - oops during network initialization

[email protected] wrote:
>
> On Wed, 21 Jan 2004 16:46:27 +0100, GCS said:
>
> > > > CONFIG_IPV6_PRIVACY=y
> > Can you both try it without the above? At least it's solved my problem, and
> > I can have 'CONFIG_IPV6=y' and ipv6 netfilter options as modules.
>
> Confirm on that. Same config, turn off CONFIG_IPV6_PRIVACY, and the
> kernel boots just fine. I'm willing to test patches if needed....
>

Which kernel fails to boot? There were ipv6 fixes applied to 2.6.2-rc1.

2004-01-21 19:35:25

by Valdis Klētnieks

[permalink] [raw]
Subject: Re: 2.6.1-mm5 - oops during network initialization

On Wed, 21 Jan 2004 10:58:36 PST, you said:
> [email protected] wrote:
> >
> > On Wed, 21 Jan 2004 16:46:27 +0100, GCS said:
> >
> > > > > CONFIG_IPV6_PRIVACY=y
> > > Can you both try it without the above? At least it's solved my problem,
and
> > > I can have 'CONFIG_IPV6=y' and ipv6 netfilter options as modules.
> >
> > Confirm on that. Same config, turn off CONFIG_IPV6_PRIVACY, and the
> > kernel boots just fine. I'm willing to test patches if needed....
> >
>
> Which kernel fails to boot? There were ipv6 fixes applied to 2.6.2-rc1.

2.6.1-mm4 worked, 2.6.1-mm5 failed, haven't tried 2.6.2-rc1 (will do so this evening).


Attachments:
(No filename) (226.00 B)

2004-01-21 23:37:28

by Laszlo 'GCS' Boszormenyi

[permalink] [raw]
Subject: Re: 2.6.1-mm5 - oops during network initialization

On Wed, Jan 21, 2004 at 02:35:07PM -0500, [email protected] <[email protected]> wrote:
> On Wed, 21 Jan 2004 10:58:36 PST, you said:
> > [email protected] wrote:
> > >
> > > On Wed, 21 Jan 2004 16:46:27 +0100, GCS said:
> > >
> > > > > > CONFIG_IPV6_PRIVACY=y
> > > > Can you both try it without the above? At least it's solved my problem,
> and
> > > > I can have 'CONFIG_IPV6=y' and ipv6 netfilter options as modules.
> > >
> > > Confirm on that. Same config, turn off CONFIG_IPV6_PRIVACY, and the
> > > kernel boots just fine. I'm willing to test patches if needed....
> > >
> >
> > Which kernel fails to boot? There were ipv6 fixes applied to 2.6.2-rc1.
>
> 2.6.1-mm4 worked, 2.6.1-mm5 failed, haven't tried 2.6.2-rc1 (will do so this evening).
2.6.2-rc1 does _not_ have this problem, it is caused by a patch Andrew
applied between 2.6.1-mm4 and -mm5; only affects if CONFIG_IPV6_PRIVACY
is set.
GCS


Attachments:
(No filename) (933.00 B)
signature.asc (189.00 B)
Digital signature
Download all attachments
Subject: Re: 2.6.1-mm5 - oops during network initialization

On Wed, 21 Jan 2004, GCS wrote:

> On Wed, Jan 21, 2004 at 10:31:23AM +0200, Catalin BOIE <[email protected]> wrote:
> > On Wed, 21 Jan 2004, Valdis Kletnieks wrote:
> >
> > I can confirm I get this also.
> > > CONFIG_IPV6_PRIVACY=y
> Can you both try it without the above? At least it's solved my problem, and
> I can have 'CONFIG_IPV6=y' and ipv6 netfilter options as modules.
>
> Hope this helps,
> GCS
>

2.6.2-rc1 works good.

---
Catalin(ux) BOIE
[email protected]

2004-01-23 04:12:39

by Valdis Klētnieks

[permalink] [raw]
Subject: Re: 2.6.1-mm5 - oops during network initialization

On Wed, 21 Jan 2004 14:35:07 EST, [email protected] said:

> > > > > > CONFIG_IPV6_PRIVACY=y

> 2.6.1-mm4 worked, 2.6.1-mm5 failed, haven't tried 2.6.2-rc1 (will do so this
evening).

Just to follow up, 2.6.2-rc1-mm1 boots fine with IPV6_PRIVACY defined.


Attachments:
(No filename) (226.00 B)