Silicon Image SATA woes

Silicon Image SATA woes

[Lasse Kärkkäinen / Tronic]

[-- Attachment #1: Type: text/plain, Size: 1798 bytes --]

Running 2.6.7-cko1 on Athlon64, VIA K8T800, Epox 8hda3+. All the SATA 
disks are Samsung SP1614C. There's a nine disk (5xSP1614C, 4xSP1614N) 
RAID-5, and ReiserFS on top of that.

Q-TEC SiI3112ATC144 two port PCI SATA card - shown as devices hde and 
hdg - works fine and never gives any errors in kernel log. Well, almost. 
The drawback is that it corrupts data randomly, about once in every ten 
gigabytes of data read. I don't know if this affects writes. Testing was 
done by repeatedly dd'ing a swap partition (that was not in use) to a 
file and then diffing it with the known contents. Initially it was found 
by unrar getting unexpected and random CRC errors, then PAR2 finding 
errors and md5sums not matching.

The on-board SiI3114 is having different kind on problem. After few 
minutes of heavy use I get the following in kernel log:
ata2: DMA timeout, stat 0x61

After this happens, trying to access the filesystem, /dev/md0, 
/proc/mdstat or /dev/sdX (where X is either a, b or c; this seems to 
vary from crash to crash) halts the accessing process. The log entry 
always seems to be about ata2. I have tried reordering the disks without 
help. However, this problem didn't surface with only one disk on the 
controller.

I tried 2.6.8-rc1, but that was so unstable (*) that I could hardly boot 
the system and couldn't keep running on it. However, I confirmed that 
the 3112 data corruption also occurs there.

*) for instance, running ifconfig (by the startup scripts) OOPSed the 
kernel and halted the system until ifconfig was killed; unfortunately, I 
didn't get the log of this one.

I need quick answer: are these hardware faults and should I take the 
hardware back for warranty? I could also get some other hardware instead 
of that 3112... Would that be wise?

- Tronic -


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 256 bytes --]