Reader preference, writer preference, and task-fair readerwriter locks are shown to cause undue blocking in multiprocessor real-time systems. A new phase-fair reader-writer lock is proposed as an alternative that significantly reduces worstcase blocking for readers and an efficient local-spin implementation is provided. Both task- and phase-fair locks are evaluated and contrasted to mutex locks in terms of hard and soft real-time schedulability under consideration of runtime overheads on a multicore computer.
Björn B. Brandenburg, James H. Anderson