Iain D. Craig - Böcker

The work that this book represents is something I have wanted to do since 1979. While in Ireland, probably in 2001, I sketched some parts of a small operating system speci?cation in Z but left it because of other duties. In 2002, I worked on the sketches again but was interrupted. Finally, in April, 2005, I decided to devote some time to it and produced what amounted to a ?rst version of the kernel to be found in Chapter 3 of this book. I even produced a few proofs, just to show that I was not on a completely insane tack. I decided to suggest the material as the subject of a book to Beverley Ford. The material was sent on a Thursday (I think). The following Monday, I received an email from her saying that it had gone out for review. The review process took less than 2 weeks; the response was as surprising as it was encouraging: a de?nite acceptance. So I got on with it. This book is intended as a new way to approach operating systems - sign in general, and kernel design in particular. It was partly driven by the old ambition mentioned above, by the need for greater clarity where it comes to kernels and by the need, as I see it, for a better foundation for operating systemsdesign.Securityaspects,too,playedapart—asnotedintheintrod- tory chapter, if a system’s kernel is insecure or unreliable, it will undermine attemptstoconstructsecuresoftwareontopofit.Securitydoesnototherwise play a part in this book.

1.1 Introduction Object-oriented programming has opened a great many perspectives on the concept of software and has been hailed as part of the solution to the so-called “software crisis”. It has given the possibility that software components can be constructedandreusedwithconsiderablymorecredibility.Therearenowmany case studies in which the reuse of object-oriented components has been made and analysed. Object-oriented programming relates the programming activity to that of modelling or simulation; objects are identi?ed by a correspondence with the objects found in the application area of the program and are used to model those domain operations. Object-oriented programming also opens the prospect of more ?exible software that is able to respond dynamically to the needs of the application at runtime. It is very easy to think that object-oriented programming can be performed in only one way. The prevalence of C++ and Java suggests that they are the onlywaytoapproachtheproblemofwhatanobject-orientedprogrammingl- guage should look like. There are many approaches to this way of programming andC++andJavaexemplifyjustoneofthesedi?erentapproaches.Indeed,the wayinwhichtheconceptoftheobjectisinterpreteddi?ersbetweenapproaches and between languages. The two main approaches found in object-oriented programming languages are, respectively, class-based and prototype-based languages. Class-based l- guages are exempli?ed by Smalltalk [34], C++ [75, 74] and Java [47]. This 2 1. Introduction approach is based upon the identi?cation of common properties of objects and their description in terms of a de?nitional structure called a class. The objects manipulated by class-based programs are the result of instantiating classes.

This book was written as a companion to my book on modelling operating system kernels. It is intended to demonstrate that the formal derivation of kernels is possible (and, actually, quite easy, or so I have found thus far). Itisimportantforthereadertounderstandthatthere?nementscontained in this book are not the only ones I have performed of microkernels. To date, I have re?ned four microkernels down to executable code and have now p- duced a kit of formally speci?ed components that can be composed to form kernels. The ?rst kernel included in this book is just one example of this work. The second kernel, the Separation Kernel, is new and was partly constructed out of the kit of parts (and the reader will see reuse in its speci?cation and re?nement) and was included for speci?c reasons that will become clear anon. Bothkernelstooklessthanthreemonths’workingtimetoproduce(theactual time is rather hard to calculate because of frequent interruptions). Previous experience in re?ning kernels also paid o? in the sense that there was l- tle revision involved in their speci?cation or re?nement; the usual process of yo-yoing between levels of the derivation was absent. This appears to be an inevitable consequence of experience.

Operating systems kernels are central to the functioning of computers. Security of the overall system, as well as its reliability and responsiveness, depend upon the correct functioning of the kernel. This unique approach -- presenting a formal specification of a kernel -- starts with basic constructs and develops a set of kernels; proofs are included as part of the text.

This book was written as a companion to my book on modelling operating system kernels. It is intended to demonstrate that the formal derivation of kernels is possible (and, actually, quite easy, or so I have found thus far). Itisimportantforthereadertounderstandthatthere?nementscontained in this book are not the only ones I have performed of microkernels. To date, I have re?ned four microkernels down to executable code and have now p- duced a kit of formally speci?ed components that can be composed to form kernels. The ?rst kernel included in this book is just one example of this work. The second kernel, the Separation Kernel, is new and was partly constructed out of the kit of parts (and the reader will see reuse in its speci?cation and re?nement) and was included for speci?c reasons that will become clear anon. Bothkernelstooklessthanthreemonths’workingtimetoproduce(theactual time is rather hard to calculate because of frequent interruptions). Previous experience in re?ning kernels also paid o? in the sense that there was l- tle revision involved in their speci?cation or re?nement; the usual process of yo-yoing between levels of the derivation was absent. This appears to be an inevitable consequence of experience.

This book offers a unique insight into the building of virtual machines for sequential, object-oriented and parallel languages, and uses comparisons of different VMs to provide actual, practical examples. State transitions are used as a formal technique for the specification of virtual machines throughout and -- in addition -- transitions and state transitions relating to the general operation are included for specification of the virtual machine for the event-based system. Two virtual machines are defined using a simple sequential language, which is then generalised to include object and parallelism. Other themes explored include the implementation of VMs and proposals for future work. Appendices contain high-level specifications of two compilers: one for the simple language that serves as the basic example throughout the book, the other for the event-based language specified in Chapter 6. This book is an essential reference tool for academic and industrial researchers as well as postgraduates in this area.

This book offers a unique insight into the building of virtual machines for sequential, object-oriented & parallel languages, & uses comparisons of different VMs to provide actual, practical examples.State transitions are used as a formal technique for the specification of virtual machines throughout and - in addition - transitions & state transitions relating to the general operation are included for specification of the virtual machine for the event-based system. Two virtual machines are defined using a simple sequential language, which is then generalised to include object & parallelism. Other themes explored include the implementation of VMs & proposals for future work. Appendices contain high-level specifications of two compilers: one for the simple language that serves as the basic example throughout the book, the other for the event-based language specified in Chapter 6.This book is an essential reference tool for academic & industrial researchers as well as postgraduates in this area.