Finally! I was right when I suggested that it was DMA across page
boundaries.  I disabled DMA in pciide and everything works perfectly!  So
to summarize the caveats when using a DPWS500a we have the following:

CIA_PYXIS_BUG_BASE is incorrectly set to (1*128*1024) instead of
(128*1024*1024) in cia_dma.c. (This patch should be in -current)

The de driver for the 21143 has to be set in 100baseTX mode on a 10baseT
lan.

And if you have IDE disable dma. I just commented out a small block in
pciide.c.  Perhaps some one could add a config value to disable IDE DMA.
A note about this in the faq would be good too.

Thanks!
Jeff

On Sun, 30 Jan 2000, Jeff Roberson wrote:

> Date: Sun, 30 Jan 2000 22:17:12 -0800 (PST)
> From: Jeff Roberson <nomad@nop.aliensystems.com>
> To: Jason Thorpe <thorpej@nas.nasa.gov>
> Cc: port-alpha@netbsd.org
> Subject: Re: DPWS500/Pyxis bugs, corrupt memory etc. 
> 
> Could this be the DMA across page boundaries bug?  I'm guessing that these
> files weren't corrupted on transfer, because tcp checksums would have
> cought that.  Could it be the IDE drive using DMA that's causing the
> problems?  The IDE controller is on the first PCI bus, so it hasn't gone
> across the ppb.  If this is the case would a scsi controler fix it? I
> notice that all of the people who have reported sucess with the miatas
> have had scsi.  This still doesn't explain why it would work in
> FreeBSD/Linux/NT and not netbsd though.  There must be some software
> glitch at play here along with the bunk hardware.
> 
> Thanks,
> Jeff
> 
> On Fri, 28 Jan 2000, Jeff Roberson wrote:
> 
> > Date: Fri, 28 Jan 2000 21:44:17 -0800 (PST)
> > From: Jeff Roberson <nomad@nop.aliensystems.com>
> > To: Jason Thorpe <thorpej@nas.nasa.gov>
> > Cc: port-alpha@netbsd.org
> > Subject: Re: DPWS500/Pyxis bugs, corrupt memory etc. 
> > 
> > I just confirmed.  This does happen with or without the patch. All of the
> > archives are still being corrupted..
> > 
> > Jeff
> > 
> > On Fri, 28 Jan 2000, Jason Thorpe wrote:
> > 
> > > Date: Fri, 28 Jan 2000 20:30:47 -0800
> > > From: Jason Thorpe <thorpej@nas.nasa.gov>
> > > To: Jeff Roberson <nomad@nop.aliensystems.com>
> > > Cc: port-alpha@netbsd.org
> > > Subject: Re: DPWS500/Pyxis bugs, corrupt memory etc. 
> > > 
> > > On Fri, 28 Jan 2000 20:04:48 -0800 (PST) 
> > >  Jeff Roberson <nomad@nop.aliensystems.com> wrote:
> > > 
> > >  > I'm having even more problems with this machine now though. There are some
> > >  > serious corruption problems.  When I downloaded the install files they all
> > >  > had crc errors.  So nfs mounted another machine and copied it over.  After
> > >  > that occasionally when I recompile my kernel I get gcc errors.  Some data
> > >  > written to .depend was corrupted, and occasionally the assembler reports
> > >  > invalid input.  Each time after I rerun make it works.  Fsck is also
> > >  > constantly reporting problems with my disk.  Does anyone know what might
> > >  > cause this?  I've ruled out hardware errors as other operating systems
> > >  > have not had these problems.
> > > 
> > > Does this happen both with and without that other patch that I sent you?
> > > 
> > >         -- Jason R. Thorpe <thorpej@nas.nasa.gov>
> > > 
> > > 
> > 
> > 
> 
>