SDB:SCSI Checklist

Version: 1.0 -

When troubles with the SCSI subsystem are encountered, check out the following list, which has proven to be very useful over time.

Even if the SCSI bus works in DOS mode, when running Linux the SCSI bus system must be SCSI 2-compliant.
Check whether the cable's red marked wire is connected to pin 1. Are the connectors plugged in tightly?
Maybe somewhere a pin is deformed or has broken. Also, the cable should not be twisted repeatedly. It should be kept as straight and flat as possible. Test all self-made connections and cables. Use a new cable when you are not sure. The distance between the SCSI adapter and the first SCSI device must not exceed 40 cm. Between all further SCSI devices, it may not exceed 30 cm.
Have both ends of the SCSI bus been terminated correctly? According to the SCSI 2 standard, the SCSI bus must have passive termination at the physical ends. It should even have active termination at one end. If the host adapter is located at the end of the bus, it should provide the termination power for active termination. Otherwise, another SCSI device can supply the power. If all devices are SCSI 2 standard compliant, even several devices may provide termination power.
When using SCSI-1, only one device may supply termination power. Adaptec cards configure this using the BIOS. NCRs set this directly on the card.
When using an NCR host adapter, enable parity checking and disable XOR for all devices. For the Adaptec host adapter, disable, in the BIOS, Sync Negotiation and allow disconnected for CD-ROM drives, slow hard drives, and SCSI tapes.

Note according to Adaptec: disabling allow disconnected is the safe configuration, especially for slow drives. Try this setup first and enable this feature later (allow disconnected).

The SCSI host adapter card might be plugged into the wrong slot. Usually the first PCI slot (opposite the ISA slots) is set to INTA. If the host adapter card is plugged in there, the BIOS PCI setup should be configured to use an interrupt the PCI bus is able to handle. To do this, consult the motherboard's manual, as each manufacturer uses different values. Problems may occur if the relation Slot to INT is set up automatically: check the assignment and correct the interrupts for INT manually.
An NCR card can be set to first or second controller (jumper). Usually (check it) the card in slot INTA must be set to second.
If a slot is set to INTD (usually none of the 3 PCI slots has this configuration), the card usually needs to be jumpered to first (try it).
The BIOS of NCR cards must be at least Rel. 3.06.00: NCR SDMS (TM) V3.0 PCI SCSI BIOS, PCI Rev. 2.0, NCRPCI-3.06.0
A multiple appearance of the 32-Bit-PCI-BIOS within the motherboard BIOS (kernel warning) usually means that the board is likely to work incorrectly. Linux is not to blame for this; the motherboard BIOS is.
The reason for problems with high transfer rates and drivers might be due to an incoherent second level cache. This can be checked by disabling the second level cache and enabling simultaneously all PCI features (e.g., ASUS PCI/I-486 SP3, do not confuse this with ASUS PVI/I-486 SP3).
For ASUS PCI/I-486 SP3G, disable the GAT option and set DRAM Refresh to normal.
If nointerrupt has been assigned to the SCSI host adapter card, it will run in polling mode (IRQ 0) when in DOS mode. Howeverm Linux uses the interrupt mode, so the IRQ 0 is used by the timer and the driver does not work (0 hosts). This can be determined in the directory /proc/ in interrupts, ioports, and pci (the files should be viewed with cat or more, but not with less).

<keyword>SCSI,adaptec,2940,ncr</keyword>