Summary: Efficiently Verifiable Sufficient Conditions
for Deadlock-freedom of Large Concurrent Programs
Paul C. Attie and Hana Chockler
College of Computer Science, Northeastern University,
Cullinane Hall, 360 Huntington Avenue,
Boston, Massachusetts 02115.
May 17, 2004
We present two polynomial-time algorithms for automatic verification of deadlock-freedom of
large finite-state concurrent programs. We consider shared-memory concurrent programs in which
a process can nondeterministically choose amongst several (enabled) actions at any step. Our al-
gorithms are sound but incomplete: if they return a positive answer, then the program is indeed
deadlock-free, while a negative answer conveys no information: the program might be deadlock free,
but "fails" the test that our algorithms apply.
Our algorithms apply to programs which are expressed in a particular syntactic form, in which
variables are shared between pairs of processes, and the synchronization code for each pair of inter-