Hi Andrew,
this is a Suse kernel bug and will be fixed in a future kernel (at the
latest in SP4). You have 2 possibilities:
- downgrade to a pre-SP3 version
- switch off async IO (DB param disk_asynch_io = FALSE)
BTW, it would also work with SLES10, but this is not supported.
Clemens
McAllister, Andrew wrote:
>Thanks to many of you for suggestions...
>
>We tried a patched kernel from SuSE that looked to have worked for about
>10 days, but alas, last Friday we apparently hit the same AIO bug in the
>kernel.
>
>Anyone have an older combination of kernels and lib aio for SLES9 that
>is stable with Oracle 9.2.0.8? I'm totally willing to go back to SP2 or
>even SP1 if I have to. The group I'm running this DB for is a windows
>only shop and they're wondering why I don't just put up Oracle on
>Windows instead. Their expectations are that the OS should only be
>rebooted once per month on patch Tuesday ;).
>
>Thanks
>Andy
>
>Oct 13 02:48:53 um-tia-01 kernel: ------------[ cut here ]------------
>Oct 13 02:48:53 um-tia-01 kernel: kernel BUG at fs/aio.c:733!
>Oct 13 02:48:53 um-tia-01 kernel: invalid operand: 0000 [#1]
>Oct 13 02:48:53 um-tia-01 kernel: SMP
>Oct 13 02:48:53 um-tia-01 kernel: CPU: 3
>Oct 13 02:48:53 um-tia-01 kernel: EIP: 0060:[<c019dbef>] Not
>tainted
>Oct 13 02:48:53 um-tia-01 kernel: EFLAGS: 00010287
>(2.6.5-7.257.PTF-smp SLES9_SP3_BRANCH-20060515141414)
>Oct 13 02:48:53 um-tia-01 kernel: EIP is at __aio_run_iocbs+0x22f/0x240
>Oct 13 02:48:53 um-tia-01 kernel: eax: ee3e07e0 ebx: ee3e0780 ecx:
>00000000 edx: 00000000
>Oct 13 02:48:53 um-tia-01 kernel: esi: f238e000 edi: eaf0e880 ebp:
>0000c000 esp: f238ff1c
>Oct 13 02:48:53 um-tia-01 kernel: ds: 007b es: 007b ss: 0068
>Oct 13 02:48:53 um-tia-01 kernel: Process oracle (pid: 7482,
>threadinfo=f238e000 task=f4125990)
>Oct 13 02:48:53 um-tia-01 kernel: Stack: 00000082 c5bfcc80 eaf0e880
>eaf0e8b0 f238ff2c f238ff2c 00000000 c5bfcc80
>Oct 13 02:48:53 um-tia-01 kernel: f238ff70 eaf0e880 c019e397
>00000000 0af87e68 0af87e68 eaf0e880 00000000
>Oct 13 02:48:53 um-tia-01 kernel: f238e000 c019e4e2 00000000
>00000000 fffffff2 0af87e68 00000000 00000000
>Oct 13 02:48:53 um-tia-01 kernel: Call Trace:
>Oct 13 02:48:53 um-tia-01 kernel: [<c019e397>]
>io_submit_one+0x1a7/0x210
>Oct 13 02:48:53 um-tia-01 kernel: [<c019e4e2>] sys_io_submit+0xe2/0x150
>Oct 13 02:48:53 um-tia-01 kernel: [<c0109257>] syscall_call+0x7/0xb
>
>
>
>>-----Original Message-----
>>From: Frank Westheider [mailto:frank.westheider@(protected)]
>>Sent: Sunday, October 01, 2006 6:47 AM
>>To: McAllister, Andrew; suse-oracle@(protected)
>>Subject: Re: [suse-oracle] Libaio kernel bug on SLES9 SP3 with 9.2.0.8
>>
>>Hi Andrew, long time no seen ;-)
>>And again with a kernel bug.
>>
>>We already opened a bug with suse/novell and oracle on this issue.
>>I scanned the aio.c in the current SP3 kernel and new
>>kernels...there are a
>>lot of smp/irq locking issues addressed in the new aio-version.
>>I don't meen the libaio, i mean the aio kernel layer.
>>
>>
>snip
>
>
>>----- Original Message -----
>>From: "McAllister, Andrew" <McAllisterA@(protected)>
>>To: <suse-oracle@(protected)>
>>Sent: Friday, September 29, 2006 11:18 PM
>>Subject: [suse-oracle] Libaio kernel bug on SLES9 SP3 with 9.2.0.8
>>
>>
>>I received a kernel bug in my /var/log/messages and 0 seconds later my
>>oracle database hung, all I could do was "shutdown abort"
>>even then the
>>database wouldn't restart. I'm rebooting right now.
>>
>>The bug info is below. I've got libaio-0.3.102-1.2 installed which
>>appears to be the latest available through yast updates. Does anyone
>>have a more stable libaio for use with Oracle 9.2.0.8 running
>>on SLES 9
>>with SP3?
>>
>>
>snip
>
>
>
