Re: device timeouts on hard disks no matter what..

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Fri, Dec 11, 2009 at 11:51:32AM +0200, Teemu Rinta-aho wrote:
> That's right.
> 
> I have now found out that I can reproduce those read timeouts when
> raidframe is running a parity check on the raid set. The read
> timeouts hit random sectors, so I don't think the drives are broken,
> or then all drives that they sell nowadays are broken.
> 
> One thing I tried was to run kernel with only one CPU, but it didn't
> change anything. I thought it would make a difference, as the problems
> started when I went multi-core.
> 
> I guess the next step is to move the disks to yet another PC with
> a bigger power supply.
> 
> But, could you please shortly describe, what does it actually mean
> when there is a read/write timeout, that is "soft corrected"?

The command didn't complete, so the driver did reset the drive and
retry the command. The retry did complete (the driver will retry 5
times before reporting it as an error to upper layer)

> Who timeouts,

The command sent to the device

> where is the timer, and who has set that timer?

The timer is in the ata driver, and is set by the ata driver.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--