Johnny Hughes wrote:
>On Tue, 2005-03-15 at 11:54 +0100, Ivo Panacek wrote:
>
>
>>Hi,
>>
>>I have recently installed CentOS4 (i386 version) on server.
>>
>>After I run
>>
>>mdadm --create /dev/md0 --level=raid5 --raid-devices=6
>>--spare-devices=1 $DEVLIST
>>
>>($DEVLIST is /dev/sda1 ... 7 same SATA disks on three different
>>sata controllers, two on-board, one pci-card)
>>
>>it starts to build raid but at the end (> 3 hours) it produces kernel bug.
>>
>>
>>
>>is it a real dual processor machine or HyperThreading.
>>
>>I am assuming that the drivers work for the drives, that you see them
>>all via fdisk -l, and that they are the correct size, etc.
>>
>>Maybe try booting to the non-smp (regular) kernel for building the RAID.
>>
>>should your device list entry be:
>>
>>/dev/sda[1-7]
>>
>>Just a couple thoughts
>>
>>
I would also recommend building the RAID without the SMP kernel and see
if the bug goes away. I'm not a kernel developer, but I follow the list
and the spinlock issue seems to be a recurring theme with multiprocessor
machines. Also, as Johnny mentions, the behaviour may be quite
different between a "genuine" SMP machine and a virtual SMP setup with a
hyperthreading P4. You might also want to report the bug to the kernel
gods and see if they have a more detailed plan for either fixing it or
avoiding it in the future.
Cheers,
C
