Chapter 10. Lower Level drivers

There are too many SCSI low level drivers to detail in this document. As an alternative to giving any superficial overview here, the reader is given suggestions of places to look for further information.

The source directory for the SCSI subsystem in the Linux kernel is a good place to start: /usr/src/linux/drivers/scsi. Several drivers have information in a "readme" file: README.<driver_name> . Others have extensive information at the top of their ".c" file This information often includes a version number, change logs and kernel boot time and module load time options. Often the latter information can be found in the installation guides of the various Linux distributions. Sometimes the driver maintainer will have a web site containing the most recent bug fix information. Official maintainers are listed in the /usr/src/linux/MAINTAINERS file. If there is nothing there, look in the relevant ".c" file in the SCSI subsystem directory. Some old drivers have no active maintainers. In such cases posting to the linux-scsi newsgroup may help [see N1 ].

For an overview of the drivers supplied with the kernel source tree, use one of the kernel configuration programs (e.g. cd /usr/src/linux; make menuconfig). The help information associated with each selection can be found together in one (large) flat file at /usr/src/linux/Documentation/Configure.help. Drivers can be obtained from other places. It is unlikely that a SCSI driver made for the lk 2.2 series (or before) will build or operate successfully in the lk 2.4 series. [From a programmatic viewpoint there are not a lot of things that need changing.] Drivers may even be only available in binary form, in which case make sure that you trust the provider and follow their instructions closely.

Lower level drivers can support either of 2 error handling strategies. The older one is considered obsolete while the newer one is often called "new_eh". The advantage of "new_eh" is that it uses a separate kernel thread per host (named "scsi_eh_<n>" where <n> is the host number) to facilitate error recovery. Both error handling strategies were also available in the lk 2.2 series in which very few adapter drivers used "new_eh". In the lk 2.4 series, more drivers are using it and the plan for the forthcoming lk 2.5 development series is to drop mid level support for the older, obsolete error strategy.

Drew Eckhardt's SCSI-HOWTO document [see reference W7 ] goes into much more detail about lower level (adapter) drivers than this document. Since that SCSI-HOWTO is 5 years old, many things have changed and more drivers have been added.

There is a lower level driver called scsi_debug that simulates one or more "direct access" devices (i.e. disk(s)) using the computer's memory. From lk 2.4.17 it acts as a "ram disk". While there are many ram disk implementations available in Linux (e.g. ramfs), scsi_debug may help to isolate a defective scsi driver in a problematic installation. See scsi_debug.c for further information.