Subject: Re: pciide performance on alpha
To: Ken Wellsch <email@example.com>
From: Manuel Bouyer <firstname.lastname@example.org>
Date: 08/27/1999 19:10:08
On Fri, Aug 27, 1999 at 09:26:11AM -0400, Ken Wellsch wrote:
> | I think I found where the problem is with the cypress. I didn't read the
> | doc carefuly enouth, and the DMA modes were not set (so it was using DMA
> | mode 0). By the way, DMA mode config registers are in isa space, not pci
> | space. Why did they do this this way ? Bah !
> | Could you try the new sources I've put in
> | ftp://antioche.lip6.fr/pub/tmp/bouyer/pciide.tgz ?
> I'm sure Matt has provided tons of feedback, but just as an extra data-point,
Unfortunably no, not yet. I guess he's busy doing other things ...
> I too tried this last night on my PC164SX. Using bonnie as a metric I found
> that the write-performance improved by 2 to 3 times (from 2-3Mb/s to 5-6Mb/s)
> while the read-performance was basically unchanged. Oh, I also noted the
> CPU time figures for writing changed dramatically too, from 2-3% up to 98%
> with your changes. I don't know whether this helps at all, but there it is.
> -- Ken
> P.S. Sorry I forgot to bring the actual dmesg and output from home, but I
> also recall seeing an error about not being able to map DMA registers.
This is interesting. This certainly mean that the driver was using PIO
then (which should be setup properly :)
Manuel Bouyer, LIP6, Universite Paris VI. Manuel.Bouyer@lip6.fr