On Thu, May 01, 2008 at 06:18:34PM +0200, Nele Custers wrote: > I'm in the middle of the installation procedure (wrote installation > files to CDROM, booted the Alpha from the CD with SRM and successfully > passed the first 10 steps or so of the installation, including > partitioning). > > However, when now trying to perform the "Install the base system" step, > after a few minuts the installer starts to check packages (in alphabetic > order) and troubles start when at the f with findutils. It gives an > error "Debootstrap warning: > ftp.freenet.de/debian/pool/main/f/findutils/findutils_4.2.28-1etch1_alpha.deb > was corrupt". After this a lot of other packages give the same issue and it > ends with a Debootstrap error (return value 1). > > What can I do about this?
Are you using an IDE or a SCSI harddisk? Many PWS systems have a defective PCI controller that makes the IDE controller not work correctly in DMA mode which means you get corrupted transfers (on writes only, reads are fine). This makes the install imposible to do. If you disable DMA access to the HD (I forget exactly how I did that when I installed mine), then it works fine. Reading back the data is always fine (which is why reading from the IDE cdrom works fine). SCSI drives should always be fine since the scsi controller and driver apparently avoid the PCI bug. The PCI bug is that any DMA transfer that crosses a page boundary is corrupted, but any that don't cross a page boundary are fine, if I recall the problem correctly. The newer PWS machines were fixed, but as far as I can tell the majority of them were broken. Any device behind the bridge (that is in a 32bit PCI slot) is immune from the problem, so an add in controller would also work fine in DMA mode, although of course the SRM won't boot from it. I have thought of using an add in card and using a very small IDE disk as a boot drive. Or maybe I can find a small scsi drive to use as a boot drive, and use an add in controller to run a large IDE drive from the 32bit PCI bus. I just haven't got around to doing that yet. -- Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
