Re: FS Corruption
I've been using a motherboard with the VIA chipset and a UDMA drive and
haven't had any problems with any of the 2.2 kernels (IDEDMA and
VIA82C586 both enabled).
Bob
On Sat, Aug 28, 1999 at 06:00:13AM -0700, tony mancill wrote:
> On Sat, 28 Aug 1999, Hamish Moffatt wrote:
>
> > On Fri, Aug 27, 1999 at 11:13:16PM -0700, tony mancill wrote:
> > > I'm having the same problem, did some research, and there seems to be a
> > > problem with DMA for the VIA chipset. This jives well with the fact that
> >
> > Do you mean UltraDMA 33 (or 66)? I have one drive here which has that
> > and I haven't seen a problem with 2.2.10ac12, although since the UDMA
> > drive does not contain any linux partitions (only Windows) I might not
> > have noticed.
>
> UDMA 33. I've been hacking on it all night, and have found out that I can
> kill the filesystem everytime I use hdparm to turn DMA on, and cannot kill
> it if DMA is off. (I've been copying a slink source CD onto the drive and
> the doing "find . -exec cksum {} \;" on the contents. :) If DMA is on,
> nasty things start popping up in syslog almost immediately. Even when
> just reading from the fs (well, that involves modifying the atimes in the
> inodes), I experience data loss.
>
> I feel certain that it's related to the warning in the 2.2.x kernel
> sources about the VIA chipset. I haven't had a chance to reboot with the
> experimental support for the VIA, but that's next. The danger is that by
> default, the 2.2.x kernels will enable DMA for the IDE controller (the
> VIA_VP included), and corruption can happen quickly. I've been mondo
> lucky, because there must be something wrong with the drive I've been
> using for the past year (it may be jumpered incorrectly). It supports
> UDMA, but DMA requests to it time out immediately, so the kernel turns off
> DMA in the driver for it. The problem occurred for real when I added a
> new drive...
>
> You probably wouldn't have Windows problems since most motherboard
> manufacturers include drivers for there IDE controller which fix crap like
> this. I checked the Soyo page, and they have one, even for my sorry
> motherboard.
>
> Anyway, sorry for the off-topic mail. Send me mail directly if you want
> any details about the problem.
>
> tony@mancill.com | Boy afraid... prudence never pays.
> http://www.debian.org | And everything she wants costs money. (The Smiths)
>
>
> --
> To UNSUBSCRIBE, email to debian-devel-request@lists.debian.org
> with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
>
--
Bob Nielsen Internet: nielsen@primenet.com
Tucson, AZ AMPRnet: w6swe@w6swe.ampr.org
DM42nh http://www.primenet.com/~nielsen
Reply to: