AIO together with SMPtimers-A0 oops and freezing

Hi Ben, Hi Ingo,

Ben, I am using your AIO 20020619 patch + relevant fixes from the AIO mailinglist together with your patch Ingo, SMPtimers-A0.

As almost anything for WOLK is selectable via kernel configuration I was able to track the issue down we've experienced with both together. If I use AIO without SMPtimers, Oracle 9i works horribly fine :), no oops, no panic, no freeze, just fast as light ;) ... Unfortunately if I use them both, AIO + SMPtimers, doing some heavy traffic to the Oracle 9i database, the system oops() and after that it's almost freezed (only sysrq works).

If I use either AIO _or_ SMPtimers, no problem occurs, all works fine.

Is there any chance to give both a cooperation with each other? :) Otherwise I have to disable SMPtimers completely from the kernel config if AIO gets selected as I am not able to fix that issue :|

Kernel is  :  2.4.18-wolk3.5-rc4 (sf.net/projects/wolk)
Hardware is:  Compaq ML570, Quad Xeon 900MHz, 16GB RAM, u2w scsi raid.
System is  :  Debian Woody 3.0r0

Many thanks for your help and your time!!

Error follows:
--------------

Unable to handle kernel paging request at virtual address  00v03ab9
printing eip:
c01221e1
*pde = 00000000
Oops: 0002
CPU:    1
EIP:    0010:[<c01221e1>]       Not tainted
EFLAGS: 00010002
eax: 08c03ab5   ebx:  d1008000  ecx:  c03ab501  edx:  00c03ab5
esi: c03ab520   edi:  10c03ab5  ebp:  10x03ab5  esp:  d1009f38
ds:  0018       es  0018        ss: 0018
Process swapper (pid: 0, stackpage=d1009000)
Stack:  d1008000 00000080 00001020 c03ab520 00000086 c0122844 c03ab520 
00000080
                00000001 00000000 00000000 c011252d d1008000 d1008000 c0105470 
c02af3a8
                d1009f7c c0105470 d1008000 d1008000 d1008000 c0105470 00000000 
00000000
Call Trace: [<c0122884>] [<c011252d>] [<c0105470>] [<c105470>] [<c105470>]
        [<c0105499>] [<c0105502>] [<c0119326>] [<c01191c4>]

Code: 89 42 04 89 10 c7 41 04 00 00 00 00 c7 01 00 00 00 00 89 4e

Trace; c0122884 <del_timer_sync+9d0/a44>
Trace; c011252d <smp_call_function+7e9/1d24>
Trace; c0105470 <enable_hlt+8/190>
Trace; 0c105470 Before first symbol
Trace; 0c105470 Before first symbol
Trace; c0105499 <enable_hlt+31/190>
Trace; c0105502 <enable_hlt+9a/190>
Trace; c0119326 <acquire_console_sem+136/164>
Trace; c01191c4 <printk+144/170>

Code; c01221e1 <del_timer_sync+32d/a44>
00000000 <_EIP>:
Code; c01221e1 <del_timer_sync+32d/a44> <=====
0: 89 42 04 mov %eax,0x4(%edx) <=====
Code; c01221e4 <del_timer_sync+330/a44>
3: 89 10 mov %edx,(%eax)
Code; c01221e6 <del_timer_sync+332/a44>
5: c7 41 04 00 00 00 00 movl $0x0,0x4(%ecx)
Code; c01221ed <del_timer_sync+339/a44>
c: c7 01 00 00 00 00 movl $0x0,(%ecx)
Code; c01221f3 <del_timer_sync+33f/a44>
12: 89 4e 00 mov %ecx,0x0(%esi)

Kind regards
        Marc-Christian Petersen

http://sourceforge.net/projects/wolk

PGP/GnuPG Key: 1024D/408B2D54947750EC
Fingerprint: 8602 69E0 A9C2 A509 8661 2B0B 408B 2D54 9477 50EC
Key available at www.keyserver.net. Encrypted e-mail preferred.

Hmmm, the only problem I can see in the aio code wrt timer usage is the following. Does this patch make a difference? If not, I'm guessing that the problem is something in SMPtimers-A0 that aio happens to trigger. The only timer aio uses is for the timeout when waiting for an event, and the structure for that is put on the stack.

-ben

Index: aio.c
===================================================================
RCS file: /bcrl/cvs/CVSROOT/net-aio/linux/fs/aio.c,v retrieving revision 1.13

diff -u -u -r1.13 aio.c
@@ -774,8 +774,10 @@
 			goto out;
 
 		set_timeout(&to, &ts);
-		if (to.timed_out)
+		if (to.timed_out) {
 			timeout = 0;
+			clear_timeout(&to);
+		}
 	}
 
 	while (likely(i < nr)) {

Hmm, I forgot to comment, but I apply smptimers on top of latest -aa, that includes aio (is it different implementation?), and the kernel works fine.

J.A. Magallon             \   Software is like sex: It's better when it's free
mailto:[email protected]  \                    -- Linus Torvalds, FSF T-shirt
Linux werewolf 2.4.19-jam0, Mandrake Linux 9.0 (Cooker) for i586
gcc (GCC) 3.2 (Mandrake Linux 9.0 3.2-0.2mdk)

That would point to a merge error, or one of the changes that -aa made as being relevant. Someone has to extract the differences to track it down.

-ben

"You will be reincarnated as a toad; and you will be much happier."

Latest thing I run is here:

http://giga.cps.unizar.es/~magallon/linux/kernel/2.4.19-jam0/30-smptimers-A0.bz2

I applies on top of 00-aa0.bz2 (in the same location), which is a ported version of latest -aa (-rc5-aa1) to 2.4.19-final.
But aio patches from -aa are accesible directly.

J.A. Magallon             \   Software is like sex: It's better when it's free
mailto:[email protected]  \                    -- Linus Torvalds, FSF T-shirt
Linux werewolf 2.4.19-jam0, Mandrake Linux 9.0 (Cooker) for i586
gcc (GCC) 3.2 (Mandrake Linux 9.0 3.2-0.2mdk)

On Wednesday 07 August 2002 01:14, Benjamin LaHaise wrote:

Hi Benjamin,

Perfect!! That, really small, patch makes it working perfectly. The system is up for some hours now with stress testing Oracle without any oops() or any other problem.

Thanks alot Ben! :-)

I'll let you know, after some days of stress testing Oracle, if it still works. I expect it will do! :)

Kind regards
        Marc-Christian Petersen

http://sourceforge.net/projects/wolk

PGP/GnuPG Key: 1024D/569DE2E3DB441A16
Fingerprint: 3469 0CF8 CA7E 0042 7824 080A 569D E2E3 DB44 1A16
Key available at www.keyserver.net. Encrypted e-mail preferred.