thanny n00b
Joined: 15 Aug 2005 Posts: 55
|
Posted: Sun Jan 13, 2013 10:41 am Post subject: megaraid_sas with Intel SRCSAS18E, kernel >2.6.31 |
|
|
One of my servers has a pair of Intel SRCSAS18E RAID controllers, and it has been running kernel 2.6.31 for quite some time. I decided to replace 320GB drives with 2TB drives, and planned on moving from jfs to either ext4 or xfs. To get the latest file system code, I emerged the sources for 3.6.11 and built that.
After a brand new ext4 volume failed to mount, I stripped away layers (lvm and software RAID) until I found something very peculiar. After formatting a partition, RAID device, or lvm volume with either ext4 or xfs, then immediately running fsck on it, the result was complete integrity failure. The first mount attempt would always fail. After it succeeded, attempts to write data to the file system, resulted in copious errors in the syslog.
After trying some different 3.6.11 kernel options to no avail, I rebooted the server back into 2.6.31 to try again. I had to rebuild it with GPT support, but after that, I was able to format a partition and get no errors afterwards. Recreated the software stripe of the two hardware arrays, added some lvm volumes, formatted them, and fsck reported no problems.
So I have to conclude that the megaraid_sas driver in 3.6.11 is broken, at least with respect to the Intel SRCSAS18E card.
Does anyone have any idea when this problem might have appeared? I figure there are probably fixes or enhancements to ext4 in later kernel versions that I'll want, but I'm not keen on building each available version in turn to see when the driver breaks. Especially since I'm an hour away from the server, and getting a clean kernel build to boot on the first try doesn't always happen. |
|