A hardware implementation of capability-based addressing

Abstract

The SWARD architecture, an experimental higher-level architecture, contains the naming and protection concept of capability-based addressing. After discussing the merits of capability-based addressing, its general representation in the SWARD architecture is discussed. The initial representation of capability-based addressing in the architecture led to a set of problems; these problems are described, as well as their solutions. Finally, the implementation of capabilities by the processor is discussed.

Password systems: Design and implementation

“…In this approach, the representation of an object is complicated by the necessity to reserve a capability list for storage of the capabilities for the data segments containing the object value. Alternatively, in a system featuring hardware support for memory tagging, a specific tag configuration will indicate that the corresponding cell contains a capability [23], [24], [25]. Contrary to hardware standardization, this approach requires an ad hoc memory system.…”

Customized Message Passing

“…The reference can define what permissions the invoking object has with respect to the object on which it is operating. These references can be as simple as direct pointers to other objects and in effect provide no protection or as complex capabilities in full capability based systems [29,33]. Object-based systems allow both multiple implementations and interfaces but favor multiple implementations hidden by a single interface.…”

A VLSI architecture for enhancing software reliability

“…Most of the prior capability-based computer systems incorporate capabili^-based addressing into the architecture. Architectures in this class, in approximate chronological order, are as follows: the Plessey PP 250 system [15], the Magic Number Computer (abandoned) [40], the CAP systems [40], [66], the C.mmp/Hydra system [67], the Cm*/Star OS system [20], die IBM S/38 [28], SWARD [46], [47] and the iAPX 432 system [29], [40] (scrapped) from Intel.…”

“…Our scheme for generating uids for active capabilities is similar to the approach taken in the implementation of SWARD [47] and process id generations in some versions of Unix. It guarantees fast capability translation, a veiy frequent task.…”

“…1) The revocation problem: After granting a capability to an object, it may be necessary to withdraw it later. This is the revocation problem [47]. 2) The confinement problem [67]: once a capability is granted to a user, it may be neces sary to prevent the propagation of this capability and information accessible using this capability from the user to other users.…”