Kernel Oops when receiving a lot of UDP packets while booting

Kernel Oops when receiving a lot of UDP packets while booting

[LEROY Christophe]

If the board is receiving a lot of 172 bytes UDP packets (800 packets by
second) while booting, it oopses like that:

Linux version 2.4.16 (root@localhost.localdomain) (gcc version 2.95.3
20010315 (
release)) #215 jeu déc 27 15:37:42 CET 2001
On node 0 totalpages: 2048
zone(0): 2048 pages.
zone(1): 0 pages.
zone(2): 0 pages.
Kernel command line: nfsroot=/mcrlinux ip=192.168.2.50:192.168.2.51:::::
root=/d
ev/nfs init=/bin/init
Decrementer Frequency = 150000000/60
Calibrating delay loop... 39.73 BogoMIPS
Memory: 6852k available (848k kernel code, 296k data, 44k init, 0k
highmem)
Dentry-cache hash table entries: 1024 (order: 1, 8192 bytes)
Inode-cache hash table entries: 512 (order: 0, 4096 bytes)
Mount-cache hash table entries: 512 (order: 0, 4096 bytes)
Buffer-cache hash table entries: 1024 (order: 0, 4096 bytes)
Page-cache hash table entries: 2048 (order: 1, 8192 bytes)
POSIX conformance testing by UNIFIX
Linux NET4.0 for Linux 2.4
Based upon Swansea University Computer Society NET3.039
Starting kswapd
CPM UART driver version 0.03
ttyS00 at 0x0280 is a SMC
ttyS01 at 0x0380 is a SMC
block: 64 slots per queue, batch=16
eth0: CPM ENET Version 0.2 on SCC1, 08:00:3e:01:00:17
NET4: Linux TCP/IP 1.0 for NET4.0
IP Protocols: ICMP, UDP, TCP
IP: routing cache hash table of 512 buckets, 4Kbytes
Oops: kernel access of bad area, sig: 11
NIP: C0106838 XER: 00000000 LR: C01067C8 SP: C017FF70 REGS: c017fec0
TRAP: 0300
   Not tainted
MSR: 00009032 EE: 1 PR: 0 FP: 0 ME: 1 IR/DR: 11
DAR: C0162004, DSISR: 82000000
TASK = c017e000[1] 'swapper' Last syscall: 120
last math 00000000 last altivec 00000000
GPR00: 00000000 C017FF70 C017E000 00000035 00001032 00000001 00000001
C016A060
GPR08: 000004EF 000001FF 00000001 C0162000 0000000D 0009CA80 00000000
00000000
GPR16: 00000000 00000000 00000000 00000000 00000000 00000000 00410000
00002963
GPR24: 00000000 00001089 C0110000 C00F0000 C0110000 C0110000 C0110000
00001000
Call backtrace:
C01067C8 C0106A2C C0107228 C00FF790 C00FF7D8 C00022C4 C0004B68
CPM ENET: BSY can't happen.
CPM ENET: BSY can't happen.
CPM ENET: BSY can't happen.
CPM ENET: BSY can't happen.
CPM ENET: BSY can't happen.
............. This message until I stop the emission of the packets.
CPM ENET: BSY can't happen.
CPM ENET: BSY can't happen.
CPM ENET: BSY can't happen.
CPM ENET: BSY can't happen.
CPM ENET: BSY can't happen.
Oops: kernel access of bad area, sig: 11
NIP: C00956B4 XER: 00000000 LR: C009784C SP: C017FC50 REGS: c017fba0
TRAP: 0300
   Not tainted
MSR: 00009032 EE: 1 PR: 0 FP: 0 ME: 1 IR/DR: 11
DAR: 45000144, DSISR: 0000000A
TASK = c017e000[1] 'swapper' Last syscall: 120
last math 00000000 last altivec 00000000
GPR00: 55555555 C017FC50 C017E000 C0164120 C0A80232 C0A80233 00000000
00000002
GPR08: C0120000 C0110000 450000C8 C0110000 C0110000 0009CA80 00000000
00000000
GPR16: 00000000 00000000 00000000 00000000 00001032 00000274 0000012B
C00FE4D0
GPR24: 0000012C C0169000 C01690DC C0169000 C0A80233 00000000 C0164120
C0A80232
Call backtrace:
0009CA80 C009784C C008CD14 C0012354 C000382C C00026EC C0002874
C0009CA4 C0009B28 C00026F4 C01067C8 C0106A2C C0107228 C00FF790
C00FF7D8 C00022C4 C0004B68
Kernel panic: Aiee, killing interrupt handler!
In interrupt handler - not syncing


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Kernel Oops when receiving a lot of UDP packets while booting

[Peter Desnoyers]

You're using HHL2.0?  There's a bug in fec.c where it will pass packets
up the stack too early in initialization.  The tcp stack doesn't take
too kindly to this, as a number of data structures which are assumed to
exist by the code are actually null pointers at this point in time...

--
.....................................................................
 Peter Desnoyers            (781) 457-1165   pdesnoyers@chinook.com
 Chinook Communications     (617) 661-1979   pjd@fred.cambridge.ma.us
 100 Hayden Ave, Lexington MA 02421

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Kernel Oops when receiving a lot of UDP packets while booting

[LEROY Christophe]

What is HHL2.0 ?
I'm using the official 2.4.16 kernel from kernel.org with the ppc ports
patch for 2.4.15.
I'm not using FEC but MC68160A (that is enet.c). But that's probably the
same problem.
Any idea how to solve the problem ?

C.Leroy

Peter Desnoyers a écrit :

> You're using HHL2.0?  There's a bug in fec.c where it will pass packets
> up the stack too early in initialization.  The tcp stack doesn't take
> too kindly to this, as a number of data structures which are assumed to
> exist by the code are actually null pointers at this point in time...
>
> --
> .....................................................................
>  Peter Desnoyers            (781) 457-1165   pdesnoyers@chinook.com
>  Chinook Communications     (617) 661-1979   pjd@fred.cambridge.ma.us
>  100 Hayden Ave, Lexington MA 02421


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Kernel Oops when receiving a lot of UDP packets while booting

[Peter Desnoyers]

LEROY Christophe wrote:
>
> What is HHL2.0 ?

Montevista hardhat linux 2.0.  A couple of days ago someone had an
ethernet question about that, and I was confused.

> I'm using the official 2.4.16 kernel from kernel.org with the ppc ports
> patch for 2.4.15.
> I'm not using FEC but MC68160A (that is enet.c). But that's probably the
> same problem.
> Any idea how to solve the problem ?

Check to make sure that ethernet packet reception is disabled when you
leave the init() function, and doesn't get enabled until the _open()
function is called.  You might have to actually do work to disable it if
the bootloader already enabled things...

First you could use some flags and printks to see whether the crash is
really due to early packets, as I suspect.

--
.....................................................................
 Peter Desnoyers            (781) 457-1165   pdesnoyers@chinook.com
 Chinook Communications     (617) 661-1979   pjd@fred.cambridge.ma.us
 100 Hayden Ave, Lexington MA 02421

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

RE: Kernel Oops when receiving a lot of UDP packets while booting

[Kári Davíðsson]

Hi,

I have verified that I am getting packets over the ethernet
before the scc_enet_open() is called, I somehow suspect that
this might be causing my random crashes.

I inserted the following in enet.c

static int opened = 0

and in scc_enet_open() I put

opened = 1;

and in scc_enet_interrupt() I put

if(opened == 0)
	printk("ETHO IRQ before open\n");

How do you propose to that the initialization is done to prevent this?

I have tried simply to move

sccp->scc_gmrl |= (SCC_GSMRL_ENR | SCC_GSMRL_ENT);

from scc_enet_init() to scc_enet_open() both before and after the
netif_start_queue()
call. But it only makes thing worse.

Thanks,

K.D.

> -----Original Message-----
> From: Peter Desnoyers [mailto:pdesnoyers@chinook.com]
> Sent: 28. desember 2001 16:10
> To: LEROY Christophe
> Cc: linuxppc
> Subject: Re: Kernel Oops when receiving a lot of UDP packets while
> booting
>
>
>
> LEROY Christophe wrote:
> >
> > What is HHL2.0 ?
>
> Montevista hardhat linux 2.0.  A couple of days ago someone had an
> ethernet question about that, and I was confused.
>
> > I'm using the official 2.4.16 kernel from kernel.org with
> the ppc ports
> > patch for 2.4.15.
> > I'm not using FEC but MC68160A (that is enet.c). But that's
> probably the
> > same problem.
> > Any idea how to solve the problem ?
>
> Check to make sure that ethernet packet reception is disabled when you
> leave the init() function, and doesn't get enabled until the _open()
> function is called.  You might have to actually do work to
> disable it if
> the bootloader already enabled things...
>
> First you could use some flags and printks to see whether the crash is
> really due to early packets, as I suspect.
>
> --
> .....................................................................
>  Peter Desnoyers            (781) 457-1165   pdesnoyers@chinook.com
>  Chinook Communications     (617) 661-1979   pjd@fred.cambridge.ma.us
>  100 Hayden Ave, Lexington MA 02421
>
>
>

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Kernel Oops when receiving a lot of UDP packets while booting

[Peter Desnoyers]

> I have verified that I am getting packets over the ethernet
> before the scc_enet_open() is called, I somehow suspect that
> this might be causing my random crashes.

You just kludge it by looking at your 'opened' flag and dropping the
packet in scc_enet_rx if it's false.  If it stops the crashes, you'll be
OK until you find a version of the driver that doesn't have this
problem.

The version of the driver I use does all the hardware initialization in
scc_enet_init, and then all that scc_enet_open does is call
netif_start_queue.

--
.....................................................................
 Peter Desnoyers            (781) 457-1165   pdesnoyers@chinook.com
 Chinook Communications     (617) 661-1979   pjd@fred.cambridge.ma.us
 100 Hayden Ave, Lexington MA 02421

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

RE: Kernel Oops when receiving a lot of UDP packets while booting

[Kári Davíðsson]

> -----Original Message-----
> From: Peter Desnoyers [mailto:pdesnoyers@chinook.com]
> Sent: 7. janúar 2002 15:02
> To: Kári Davíðsson
> Cc: LEROY Christophe; linuxppc
> Subject: Re: Kernel Oops when receiving a lot of UDP packets while
> booting
>
>
> > I have verified that I am getting packets over the ethernet
> > before the scc_enet_open() is called, I somehow suspect that
> > this might be causing my random crashes.
>
> You just kludge it by looking at your 'opened' flag and dropping the
> packet in scc_enet_rx if it's false.  If it stops the

Yes I will try that too.

> crashes, you'll be
> OK until you find a version of the driver that doesn't have this
> problem.
>
> The version of the driver I use does all the hardware
> initialization in
> scc_enet_init, and then all that scc_enet_open does is call
> netif_start_queue.

My enet.c does excactly this. And it is deffinitly not doing the right
thing,
try e.g. to flood ping the board while it is booting you will see all
kinds
of wonderful strange behaviour.

Thanks,

K.D.

>
> --
> .....................................................................
>  Peter Desnoyers            (781) 457-1165   pdesnoyers@chinook.com
>  Chinook Communications     (617) 661-1979   pjd@fred.cambridge.ma.us
>  100 Hayden Ave, Lexington MA 02421
>

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/