|
Had this regularity been true for the UNIX operating system, all 80x86 UNIX systems would be alike, and the development of a specification would be a simple task. However, in exploiting the power and flexibility of UNIX, one is faced with a grander specification. The kernel architect is now faced with competing alternatives. With UNIX, the choices are no longer "cut and dried." Adding to this dilemma, the 386 is at least two generations beyond its simple ancestor. The enhanced features the 386 now offers allow us many competing ways to satisfy a UNIX system design. Continuing our example, instead of using the INT $XX instruction, we can use the intersegment LCALL instruction to call the operating system through call gate segments. We can use some powerful features of the 386, but at the cost of a more elaborate mechanism. Is it worth it? In this case, the LCALL instruction can be used to support reverse compatibility with other versions of UNIX in the form of an applications package rather than within the operating systems kernel, and thus may be worth the effort. However, choosing among the myriad, often conflicting, alternatives is typically a task fraught with peril. For the 386BSD project, we first determined our priorities: 100 percent BSD kernel and user functionality. The system must contain all important underlying mechanisms of the Berkeley UNIX system. Any extensive modification pertaining to how Berkeley UNIX functions on other extant platforms can result in incompatibility. Incompatibility is like an irritating insect that bites in many places -- and tends to lay hidden until after extensive distribution. As such, we did not exploit some features of the 386, such as its elaborate segmented architecture, at the expense of incompatibility. Efficient use of the native processor architecture. We would like to use the system in ways to obtain the highest performance and greatest functionality possible. Interoperability with existing commercial standards. We would like to use the system in ways which maintains compliance with extant commercial standards. We do not intend to unnecessarily create arbitrary new standards if current standards are acceptable. Rapid implementation of the basic operating system. One maxim of any UNIX development effort is "the best tool to build a UNIX system IS a UNIX system." We needed to bootstrap ourselves rapidly into operation and leverage 386BSD itself to complete the project.
|