Re: VIA IDE driver, v1.5 (final)

From: Vojtech Pavlik (vojtech@suse.cz)
Date: Mon Jul 24 2000 - 10:53:20 EST


On Mon, Jul 24, 2000 at 03:10:32PM +0200, Matthias Andree wrote:
> Hi Vojtech!
>
> On Mon, 24 Jul 2000, Vojtech Pavlik wrote:
>
> > No. You're not supposed to get CRC errors with UDMA. Not a single one.
> > If you do, it'd mean you'd get data corruption without the CRCs.
> > MWDMA/16 runs at the same speed as UDMA/16 and doesn't have CRC
> > protection. Many people are using MWDMA/16. If your expectations (there
> > will be a bit error now and then) were true, many people would see
> > filesystem corruption.
>
> No, you're not noticing bit flips in data blocks unless you CRC that
> data; and data flips are there, and more likely through memory or busses
> flipping bits than on disk, since disks use serious error correction
> mechanisms.

Perhaps I didn't make myself clear enough: I don't say you can detect
bit errors without CRC or that CRC makes them stop happening.

What I tried to say is that under normal operation, when every bit of
hardware is in a good state, on an IDE transfer, that is, between
harddrive electronics and ide controller, there shouldn't be bit errors.

It's great that it's CRC protected with UDMA, but there shouldn't be bit
errors anyway. It's running on 5V, after all, and there would have to be
damn strong noise to cause errors like this.

So, if one is getting CRC errors with an UDMA drive, it means the
hardware isn't in a good state. Either there is a flaky contact, or the
cable is too long or he's communicating at a speed too high for the
cable to carry (and needs an 80-wire cable for that.) But once a slow
enough speed is selected, there shouldn't be any CRC errors any more.

So the point I was trying to make is that it is reasonable to lower UDMA
transfer speed if we see a CRC error on the drive and that we won't end
with the slowest transfer speed if we run the drive like this for long
enough time.

-- 
Vojtech Pavlik
SuSE Labs

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Mon Jul 31 2000 - 21:00:17 EST