On Sunday 15 August 2004 20:53, Vojtech Pavlik wrote:
> On Sun, Aug 15, 2004 at 07:50:53PM +0200, Arjen Runsink wrote:
> > Hi Vojtech,
> > On Sunday 15 August 2004 18:56, Vojtech Pavlik wrote:
> > > Not possible. If the sata_via or sata_sil drivers are used, it's always
> > > sda. There, however, ale the IDE drivers (VIA_IDE - vt82cxxx, "SiI IDE"
> > > - siimage.c), which can drive the SATA drives as well, and then the
> > > drive will appear as hda.
> > >
> > > So I suppose what happened is that the IDE driver got initialized first
> > > in the "hda" cases and the libata (SATA) driver in the "sda" cases.
> > And thus a "mistery" is explained. So I talked to a kernel hacker too and
> > he said that using the sata_* would be better and will be used in future
> > kernels, for TCQ etc.
It really works better for me, I had to use the "barrier=off" boot parameter
to be able to boot with the official kernel releases from
kernel-default-2.6.5-7.75 and onwards (my latest was 2.6.5-7.108). All these
kernels addressed the serial ATA disk as hda and delivered a very poor
performance of only 3-5 MB/sec (hdparm -t).
Now I'm using the latest KOTD which addresses the same disk (ATA SAMSUNG
SP1614C) as sda (no barrier parameter is needed) and it delivers a
performance of 55-150 MB/sec. This is like it should be.
> > How can I force the standard suse kernels to use sata_via instead of
> > blk_dev_via82cxxx ?
> You might be able to use something like "ide0=noprobe" (replace 0 with
> the proper number for your SATA interface). It might not work, though,
> since the IDE driver could still grab the controller, even if it doesn't
> touch the drives.
ide0=noprobe didn't work for me, installing a different kernel (the latest
KOTD) did the job
> Vojtech Pavlik
> SuSE Labs, SuSE CR