Subject: Re: UDMA downgrading
To: Thomas Hertz <thomas@hz.se>
From: Manuel Bouyer <bouyer@antioche.eu.org>
List: current-users
Date: 10/15/2003 23:47:11
On Wed, Oct 15, 2003 at 08:14:13PM +0200, Thomas Hertz wrote:
> I'm running a bunch of IDE UDMA100 disks (4, to be exact) in a raid5 
> array. The disks have one controller per disk hanging off two Promise 
> Ultra100 cards (using pdcide(4), but this problem showed up on pciide(4) 
> as well). After some load on the array, and a few hours of time, the 
> disks start downgrading to UDMA/33 while giving out these messages:
> 
> wd1(pdcide0:0:0): using PIO mode 4, Ultra-DMA mode 2 (Ultra/33) (using 
> DMA data transfers)
> wd1a: DMA error reading fsbn 118858432 of 118858432-118858463 (wd1 bn 
> 118858432; cn 117915 tn 1 sn 49), retrying
> wd1: soft error (corrected)

There should be errors messages around this as well (here we know it
downgraded but we don't know why).


> 
> Strangely enough I haven't been able to reproduce even ONE of these 
> messages when the disks are not on a raid array! I realize that the 
> raidframe is putting heavier load on the disks, but it has a huge impact 
> on performance with the disks running on UDMA33!
> 
> What is the problem here? I've gotten all kinds of suggestions from a 
> bad mainboard to bad disks, but none of them seem likely. Would it be 
> better if I would put the disks in the end of the IDE cables instead of 

Yes, definitively.

> Secondly, is it possible to make the driver not downgrade the transfer 
> mode when it experiences these DMA errors once in a while? They really 
> only occur like once an hour or so, and that shouldn't be a problem.

Yes, with flags on the devices in the kenrel config file. See wd(4)

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 24 ans d'experience feront toujours la difference
--