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

On Sunday 15 August 2004 20:53, Vojtech Pavlik wrote: 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

Roel Vestjens