Re: VIA silent disk corruption - fixed for me

From: Petr Vandrovec (
Date: Wed Feb 07 2001 - 07:24:28 EST

I wrote on 6 Feb evening CET:
> On 6 Feb 01 at 15:24, Udo A. Steinberg wrote:
> >
> > > So for today I'm back on [UMS]DMA disabled. I'll try downgrading BIOS
> > > today, but it looks to me like that something is severely broken here.
> >
> > Are your drives connected to the VIA or the Promise controller? Mine
> > are both connected to the PDC20265 and running in UDMA-100 mode. There
> > have been several threads on lkml about corruption on disks connected
> > to Via chipset IDE controllers, although I didn't follow them in great
> > detail. Maybe your problem is not related to the host bridge, but to
> > the IDE controller?
> They are connected to Promise, I reserved VIA for CDROM drive.
> One HDD runs in UDMA5 mode, another in UDMA2. Corruption is often
> when I run md5sum in parallel on both HDDs - in that case almost
> no file generates same checksum which was generated using PIO4.
> When I run md5sum on only one HDD, there are about 4 checksum errors
> in 6GB of data. But I'm more and more inclined to throw this A7V away,
> as it is impossible to get datasheet from Promise, and for VIA host
> bridge I was just able to slow down normal system operation by factor
> of 3... but still with same corruption :-( Just if page could have
> 4092 and not 4096 bytes ;-)

After upgrading BIOS (it did not help) I decided to switch my secondary
harddisk to master. And voila - hde running UDMA5, hdg running UDMA2
(hdf/hdh does not exist), whole night stresstests, no corruption.

So at least for me it means that Promise Linux driver does not support
slave-only configuration. I did not checked whether master-slave pair
works, but master alone for sure works for me.

That is, for me problem was not in VIA bridge, but in PDC20265 driver...
                                            Best regards,
                                                Petr Vandrovec
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
Please read the FAQ at

This archive was generated by hypermail 2b29 : Wed Feb 07 2001 - 21:00:26 EST