The Control Structure of an Operating System

CAL is an operating system based on the concepts of capabilities and of implementation via machine extension. We first present some brief comments on our design philosophy and our experience with this approach. Extensions to the capability (descriptor) mechanism are described. The remainder of the paper concerns the control structure for (a) intra-process communication: creation and display of processes, domains and gates, various forms of domain activation, and the fielding of traps; and (b) inter-process communication: messages, events, interrupts, and locks. 1. Motherhood The CAL operating system was begun in the fall of 1968. Within nine months the basic system had been implemented on an off-the-shelf CDC 6400 with extended core storage. Since that time it has been in daily use for further development and experimentation. Since the fall of 1970 it has been available to the Berkeley campus community. During this time we have had considerable experience with the system and are now in a position to judge its virtues and flaws. One of the richest aspects of CAL is its control structure. This paper describes those aspects of the control structure which we have found to be particularly useful. In doing this we have freely done violence to the realities of CAL. Obvious (but unimplemented) generalizations have been included. Mistakes made in the implementation are not repeated here. Concessions to the hardware are obscured. The design of an operating system is like a Chinese puzzle spread out on a table: because of its size it is impossible to tell what it will be when assembled and whether there are too many pieces or too few. For this reason we have been careful not to depart too far from reality for fear of losing a piece or two. We have simply shaved off a few rough edges.