Re: [libata - SII3112] 'Virtual' bad blocks?

[Jeff Garzik <jgarzik@pobox.com>]

Peter Schaefer wrote:
> Hello!
> 
> I'm using an Epox EP-8HDA3+ Athlon64 Motherboard with
> VIA K8T800 chipset as a file server running 24/7.
> 
> The box is running since 2.6.1 and was a long time rock
> solid on 2.6.4. Problems with the VIA RHINE ethernet let
> me switch to 2.6.7 and now 2.6.8.
> 
> The board has four on-board SATA ports, two provided by
> the chipset and two provided by a Silicon Image 3112 chip.
> I'm running a RAID5 array with three disks - two on the
> chipset and one on the Silicon Image controller.
> 
> For the chipset ports the standard VIA IDE driver is used
> (with SATA support). For the SI disk i'm using the libata
> driver.
> 
> Unfortunately it's exactly this disk which give me headache
> now (it started with 2.6.7) - see logs below.
> 
> I might add that those errors aren't really bad blocks -
> after each occurence i made a full sector scan with
> Western Digital's MS-DOS-based "Data Live Guard" which
> revealed no errors. However, i had to plug the disk to
> one of the VIA ports for the tests (otherwise the tool
> didn't detect the disk).
> 
> Another thing to add is that i have activated CPU
> frequency scaling with powernowd in about the same
> timeframe (to use Cool&Quiet during times of no activity).
> But i doubt that's the reason for this?!
> 
> Do i have to worry (and yes, i know that SATA is dirt
> cheap crap)?

The version of libata you have always reports errors -- be they 
controller errors, cable errors, or disk errors -- as either a read or 
write error.

Alan Cox submitted code that makes the errors more specific, rather than 
treating them all as disk errors.

You have either a flaky controller, flaky cable, or flaky disk, one of 
the three.

	Jeff