INTRODUCTION
Nowadays processors are composed of several subsystems (such as caches, pipelines, transfer lookaside buffers, etc.) which display, at any time-instant, an associated “state”. In the context of this work, what we understand by “state of a subsystem” is the history information enclosed in the subsystem, as well as its logical configuration, at a particular time-instant. For example, the state of a cache at a given time-instant can be seen as a snapshot of all the information stored in that cache at that instant. The objective of this state is to accelerate average-case execution times. All these processor subsystems quasi-continuously face state changes at run-time and we are concerned with state changes which affect
the temporal behavior of the executing tasks. In particular, it is the case for task preemptions: when a task resumes its execution(after being preempted), for example, the cache(s) will display a state which is different from its state at the time the task got interrupted. If its state has been substantially altered during the time the task was pending, it is likely that it might be needed to reconstruct at least partially its working set after the the task resumes execution.