You must have a supported SCSI controller, obviously. Furthermore, your kernel must have controller support as well as generic support compiled in. Configuring the Linux kernel (via make config under /usr/src/linux) typically looks like the following:
... * * SCSI support * SCSI support? (CONFIG_SCSI) [n] y * * SCSI support type (disk, tape, CDrom) * ... Scsi generic support (CONFIG_CHR_DEV_SG) [n] y * * SCSI low-level drivers * ...
If available, modules can of course be build instead.
The generic SCSI driver uses its own device files, separate from those used by the other SCSI device drivers. They can be generated using the MAKEDEV script, typically found in the /dev directory. Running MAKEDEV sg produces these files:
crw------- 1 root system 21, 0 Aug 20 20:09 /dev/sga crw------- 1 root system 21, 1 Aug 20 20:09 /dev/sgb crw------- 1 root system 21, 2 Aug 20 20:09 /dev/sgc crw------- 1 root system 21, 3 Aug 20 20:09 /dev/sgd crw------- 1 root system 21, 4 Aug 20 20:09 /dev/sge crw------- 1 root system 21, 5 Aug 20 20:09 /dev/sgf crw------- 1 root system 21, 6 Aug 20 20:09 /dev/sgg crw------- 1 root system 21, 7 Aug 20 20:09 /dev/sgh | | major, minor device numbers
Note that these are character devices for raw access. On some systems these devices may be called /dev/{sg0,sg1,...}, depending on your installation, so adjust the following examples accordingly.
These device files are dynamically mapped to SCSI id/LUNs on your SCSI bus (LUN = logical unit). The mapping allocates devices consecutively for each LUN of each device on each SCSI bus found at time of the SCSI scan, beginning at the lower LUNs/ids/buses. It starts with the first SCSI controller and continues without interruption with all following controllers. This is currently done in the initialisation of the SCSI driver.
For example, assuming you had three SCSI devices hooked up with ids 1, 3, and 5 on the first SCSI bus (each having one LUN), then the following mapping would be in effect:
/dev/sga -> SCSI id 1 /dev/sgb -> SCSI id 3 /dev/sgc -> SCSI id 5
If you now add a new device with id 4, then the mapping (after the next rescan) will be:
/dev/sga -> SCSI id 1 /dev/sgb -> SCSI id 3 /dev/sgc -> SCSI id 4 /dev/sgd -> SCSI id 5
Notice the change for id 5 -- the corresponding device is no longer mapped to /dev/sgc but is now under /dev/sgd.
Luckily newer kernels allow for changing this order.
If a newer kernel and the /proc file system is running, a non-busy device can be removed and installed 'on the fly'.
To remove a SCSI device:
echo "scsi remove-single-device a b c d" > /proc/scsi/scsi
and similar, to add a SCSI device, do
echo "scsi add-single-device a b c d" > /proc/scsi/scsi
where
a == hostadapter id (first one being 0) b == SCSI channel on hostadapter (first one being 0) c == ID d == LUN (first one being 0)
So in order to swap the /dev/sgc and /dev/sgd mappings from the previous example, we could do
echo "scsi remove-single-device 0 0 4 0" > /proc/scsi/scsi echo "scsi remove-single-device 0 0 5 0" > /proc/scsi/scsi echo "scsi add-single-device 0 0 5 0" > /proc/scsi/scsi echo "scsi add-single-device 0 0 4 0" > /proc/scsi/scsi
since generic devices are mapped in the order of their insertion.
When adding more devices to the scsi bus keep in mind there are limited spare entries for new devices. The memory has been allocated at boot time and has room for 2 more devices.