data corruption with Tekram controller
Hardware:
- Alphastation 255/300 (avanti), with on-board Symbios 810a.
- Tekram DC390U2B (wide fast-40 lvd based on Symios 895).
- IBM DDRS hard drive (wide fast-40 lvd).
Software:
- SRM 6.something (a 1996 version)
- Linux 2.2.12.
- Tried with both ncr and sym drivers, even in safe:y mode.
SRM:
- There's a setting named pci_parity; it can be set to on, off or sniff (?).
With this configuration, I get an error rate in the order of 10^-9 (1 wrong
bit out of 10^9 bits). Some errors seem quite reproducible, some are
transient.
With pci_parity set to on, the system crashes back to SRM on error, and
becomes wildly unstable: a new boot most of the times hangs the machine.
With pci_parity off or sniff, the system carries on the wrong data without
notice.
The same controller and hard disk, installed in an Intel-based PC, work
perfectly.
Substantial differences are: the on-board BIOS is NOT run on the Alpha
machine; the PCI parity setting is not available for configuration (and,
in fact, I don't know it).
My hypotheses are:
- some sort of initialization is performed by the controller BIOS but not
by the Linux driver
- some sort of PCI level incompatibility between the Alpha PCI bridge
(which is an Intel chip!) and the SCSI board
- some sort of hardware breakage in my workstation
Anybody can help?
Would a newer SRM execute the SCSI BIOS (which is Intel code, of course)?
Would that help?
Ciao, William
Reply to: