Description of problem: Unable to handle kernel paging request at virtual address 6b6b6b6b printing eip: 0f86fb5b *pde = 00000000 Oops: 0000 [#1] SMP Modules linked in: nfsd exportfs nfs lockd ipv6 parport_pc lp parport autofs4 rfcomm l2cap bluetooth sunrpc tlan floppy sg microcode dm_mod usb_storage ohci_hcd ehci_hcd ext3 jbd sym53c8xx scsi_transport_spi cpqarray sd_mod scsi_mod CPU: 0 EIP: 0060:[<0f86fb5b>] Not tainted EFLAGS: 00010246 (2.6.7-1.476smp) EIP is at do_get_write_access+0x28/0x54c [jbd] eax: 0e3e65b4 ebx: 02d05c94 ecx: 00000000 edx: 6b6b6b6b esi: 02d05c94 edi: 0e3e65b4 ebp: 0c93ff58 esp: 0d958ce4 ds: 007b es: 007b ss: 0068 Process diotest4 (pid: 10985, threadinfo=0d958000 task=0d0e6290) Stack: 00000000 00000000 00000000 fffe7000 03172b00 00000000 0e3e65b4 0e589510 00000246 0c93ff58 0c93ff58 00250056 00000000 0d958d20 022d3002 0e8c1b54 0215e7c6 00001000 0e5e9564 0e3e65b4 00000001 023ab190 0f8a1015 00001000 Call Trace: [<022d3002>] cond_resched+0x12/0x37 [<0215e7c6>] __getblk+0x29/0x47 [<0f8a1015>] ext3_get_inode_loc+0x4f/0x20e [ext3] [<0f87009b>] journal_get_write_access+0x1c/0x2c [jbd] [<0f8a1a8d>] ext3_reserve_inode_write+0x34/0x81 [ext3] [<0f8a44c4>] ext3_orphan_del+0xac/0x1d1 [ext3] [<0f8a0296>] ext3_direct_IO+0x11 Version-Release number of selected component (if applicable): 2.6.7-1.476smp - not seen in earlier kernels
... and the rest of the trace. DUH! [<0f8a0296>] ext3_direct_IO+0x115/0x191 [ext3] [<0213f322>] generic_file_direct_IO+0x3c/0x5c [<0213eb9b>] generic_file_aio_write_nolock+0x39e/0x8db [<02107c9b>] do_IRQ+0x1ce/0x22e [<0211d960>] find_busiest_group+0xe6/0x2bf [<0213f1ad>] generic_file_aio_write+0x6b/0x7f [<0f89d95b>] ext3_file_write+0x19/0x8a [ext3] [<0215bef1>] do_sync_write+0x6a/0x99 [<0211e43c>] scheduler_tick+0x3b9/0x3c1 [<0215bfd6>] vfs_write+0xb6/0xe2 [<0215c070>] sys_write+0x2c/0x42 Code: 8b 02 89 54 24 10 89 44 24 0c f6 00 02 74 0a b8 e2 ff ff ff
This was reported easily reproducible in .476, not present in .478. The only significant change I can see between those two was the dropping of the lowlatency cooperative-preempt patch, which could easily be related to odd problems like this. Rik, you mentioned you saw this 2 or 3 times --- just for completeness could you record the other oopses here in case there's a more worrying pattern underneath?
I'll throw .476 back in cerberus today, to reproduce the bug again ;)