NuSMV 2.6 Tutorial
Roberto Cavada, Alessandro Cimatti,
Gavin Keighren, Emanuele Olivetti,
Marco Pistore and Marco Roveri
FBK-irst - Via Sommarive 18, 38055 Povo (Trento) – Italy
Email: nusmv@fbk.eu
Contents
1 Introduction 2
2 Examples 3
2.1 Synchronous Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.1 Single Process Example . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.2 Binary Counter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Asynchronous Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.1 Inverter Ring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.2 Mutual Exclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Direct Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3 Simulation 8
3.1 Trace Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Interactive Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2.1 Choosing an Initial State . . . . . . . . . . . . . . . . . . . . . . . 9
3.2.2 Starting a New Simulation . . . . . . . . . . . . . . . . . . . . . . 10
3.2.3 Specifying Constraints . . . . . . . . . . . . . . . . . . . . . . . . 12
4 CTL Model Checking 13
4.1 Computation Tree Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2 Semaphore Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5 LTL Model Checking 17
5.1 Linear Temporal Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2 Semaphore Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.3 Past Temporal Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6 Bounded Model Checking 20
6.1 Checking LTL Specifications . . . . . . . . . . . . . . . . . . . . . . . . . 20
6.2 Finding Counterexamples . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6.3 Checking Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
6.3.1 2-Step Inductive Reasoning . . . . . . . . . . . . . . . . . . . . . 26
6.3.2 Complete Invariant Checking . . . . . . . . . . . . . . . . . . . . 27
1
Chapter 1
Introduction
In this tutorial we give a short introduction to the usage of the main functionalities of
NUSMV. In Chapter 2 [Examples], page 3 we describe the input language of NUSMV by
presenting some examples of NUSMV models. Chapter 3 [Simulation], page 8 shows how
the user can get familiar with the behavior of a NUSMV model by exploring its possible
executions. Chapter 4 [CTL Model Checking], page 13 and Chapter 5 [LTL Model Checking],
page 17 give an overview of BDD-based model checking, while Chapter 6 [Bounded
Model Checking], page 20 presents SAT-based model checking in NUSMV.
2
Chapter 2
Examples
In this section we describe the input language of NUSMV by presenting some examples
of NUSMV models. A complete description of the NUSMV language can be found in the
NuSMV 2.6 User Manual. Also, all mentioned example files can be found in the distributed
archive of NuSMV 2.6, or can be individually downloaded from the NUSMV web pages,
at the URL <http://nusmv.fbk.eu/examples/examples.html>.
The input language of NUSMV is designed to allow for the description of Finite State
Machines (FSMs from now on) which range from completely synchronous to completely
asynchronous, and from the detailed to the abstract. One can specify a system as a synchronous
Mealy machine, or as an asynchronous network of nondeterministic processes.
The language provides for modular hierarchical descriptions, and for the definition of
reusable components. Since it is intended to describe finite state machines, the only data
types in the language are finite ones – booleans, scalars and fixed arrays. Static data types
can also be constructed.
The primary purpose of the NUSMV input is to describe the transition relation of the
FSM; this relation describes the valid evolutions of the state of the FSM. In general, any
propositional expression in the propositional calculus can be used to define the transition
relation. This provides a great deal of flexibility, and at the same time a certain danger of
inconsistency. For example, the presence of a logical contradiction can result in a deadlock
– a state or states with no successor. This can make some specifications vacuously true, and
makes the description unimplementable. While the model checking process can be used to
check for deadlocks, it is best to avoid the problem when possible by using a restricted
description style. The NUSMV system supports this by providing a parallel-assignment
syntax. The semantics of assignment in NUSMV is similar to that of single assignment
data flow language. By checking programs for multiple parallel assignments to the same
variable, circular assignments, and type errors, the interpreter insures that a program using
only the assignment mechanism is implementable. Consequently, this fragment of the
language can be viewed as a description language, or a programming language.
2.1 Synchronous Systems
2.1.1 Single Process Example
Consider the following simple program in the NUSMV language:
MODULE main
VAR
request : boolean;
state : {ready, busy};
3
ASSIGN
init(state) := ready;
next(state) := case
state = ready & request = TRUE : busy;
TRUE : {ready, busy};
esac;
The space of states of the FSM is determined by the declarations of the state variables
(in the above example request and state). The variable request is declared to be of
(predefined) type boolean. This means that it can assume the (boolean) values FALSE
and TRUE. The variable state is a scalar variable, which can take the symbolic values
ready or busy. The following assignment sets the initial value of the variable state
to ready. The initial value of request is completely unspecified, i.e. it can be either
FALSE or TRUE. The transition relation of the FSM is expressed by defining the value
of variables in the next state (i.e. after each transition), given the value of variables in
the current states (i.e. before the transition). The case segment sets the next value of
the variable state to the value busy (after the colon) if its current value is ready and
request is TRUE. Otherwise (the TRUE before the colon) the next value for state can
be any in the set {ready, busy}. The variable request is not assigned. This means
that there are no constraints on its values, and thus it can assume any value. request is
thus an unconstrained input to the system.
2.1.2 Binary Counter
The following program illustrates the definition of reusable modules and expressions. It is
a model of a three bit binary counter circuit. The order of module definitions in the input
file is not relevant.
MODULE counter_cell(carry_in)
VAR
value : boolean;
ASSIGN
init(value) := FALSE;
next(value) := value xor carry_in;
DEFINE
carry_out := value & carry_in;
MODULE main
VAR
bit0 : counter_cell(TRUE);
bit1 : counter_cell(bit0.carry_out);
bit2 : counter_cell(bit1.carry_out);
The FSM is defined by instantiating three times the module type counter cell in
the module main, with the names bit0, bit1 and bit2 respectively. The counter cell
module has one formal parameter carry in. In the instance bit0, this parameter is given
the actual value TRUE. In the instance bit1, carry in is given the value of the expression
bit0.carry out. This expression is evaluated in the context of the main module.
However, an expression of the form ‘a.b’ denotes component ‘b’ of module ‘a’, just as
if the module ‘a’ were a data structure in a standard programming language. Hence, the
carry in of module bit1 is the carry out of module bit0.
The keyword ‘DEFINE’ is used to assign the expression value & carry in to the
symbol carry out. A definition can be thought of as a variable with value (functionally)
depending on the current values of other variables. The same effect could have been obtained
as follows (notice that the current value of the variable is assigned, rather than the
next value.):
4
VAR
carry_out : boolean;
ASSIGN
carry_out := value & carry_in;
Defined symbols do not require introducing a new variable, and hence do not increase the
state space of the FSM. On the other hand, it is not possible to assign to a defined symbol a
value non-deterministically. Another difference between defined symbols and variables is
that while the type of variables is declared a priori, for definitions this is not the case.
2.2 Asynchronous Systems
Important!
Since NUSMV version 2.5.0 processes are deprecated. In future versions of NUSMV
processes may be no longer supported, and only synchronous systems will be supported by
the input language. Modeling of asynchronous systems will have to be resolved at higher
level.
The previous examples describe synchronous systems, where the assignments statements
are taken into account in parallel and simultaneously. NUSMV allows to model
asynchronous systems. It is possible to define a collection of parallel processes, whose
actions are interleaved, following an asynchronous model of concurrency. This is useful
for describing communication protocols, or asynchronous circuits, or other systems whose
actions are not synchronized (including synchronous circuits with more than one clock
region).
2.2.1 Inverter Ring
The following program represents a ring of three asynchronous inverting gates.
MODULE inverter(input)
VAR
output : boolean;
ASSIGN
init(output) := FALSE;
next(output) := !input;
MODULE main
VAR
gate1 : process inverter(gate3.output);
gate2 : process inverter(gate1.output);
gate3 : process inverter(gate2.output);
Among all the modules instantiated with the process keyword, one is nondeterministically
chosen, and the assignment statements declared in that process are executed in parallel.
It is implicit that if a given variable is not assigned by the process, then its value
remains unchanged. Because the choice of the next process to execute is non-deterministic,
this program models the ring of inverters independently of the speed of the gates.
We remark that the system is not forced to eventually choose a given process to execute.
As a consequence the output of a given gate may remain constant, regardless of its input. In
order to force a given process to execute infinitely often, we can use a fairness constraint. A
fairness constraint restricts the attention of the model checker to only those execution paths
along which a given formula is true infinitely often. Each process has a special variable
called running which is TRUE if and only if that process is currently executing.
5
By adding the declaration:
FAIRNESS
running
to the module inverter, we can effectively force every instance of inverter to execute
infinitely often.
An alternative to using processes to model an asynchronous circuit is to allow all gates
to execute simultaneously, but to allow each gate to choose non-deterministically to reevaluate
its output or to keep the same output value. Such a model of the inverter ring
would look like the following:
MODULE inverter(input)
VAR
output : boolean;
ASSIGN
init(output) := FALSE;
next(output) := (!input) union output;
MODULE main
VAR
gate1 : inverter(gate3.output);
gate2 : inverter(gate1.output);
gate3 : inverter(gate2.output);
The union operator (set union) coerces its arguments to singleton sets as necessary. Thus,
the next output of each gate can be either its current output, or the negation of its
current input – each gate can choose non-deterministically whether to delay or not. As
a result, the number of possible transitions from a given state can be as 2n
, where n is the
number of gates. This sometimes (but not always) makes it more expensive to represent
the transition relation. We remark that in this case we cannot force the inverters to be
effectively active infinitely often using a fairness declaration. In fact, a valid scenario for
the synchronous model is the one where all the inverters are idle and assign to the next
output the current value of output.
2.2.2 Mutual Exclusion
The following program is another example of asynchronous model. It uses a variable
semaphore to implement mutual exclusion between two asynchronous processes. Each
process has four states: idle, entering, critical and exiting. The entering
state indicates that the process wants to enter its critical region. If the variable semaphore
is FALSE, it goes to the critical state, and sets semaphore to TRUE. On exiting its
critical region, the process sets semaphore to FALSE again.
MODULE main
VAR
semaphore : boolean;
proc1 : process user(semaphore);
proc2 : process user(semaphore);
ASSIGN
init(semaphore) := FALSE;
MODULE user(semaphore)
VAR
state : {idle, entering, critical, exiting};
6
ASSIGN
init(state) := idle;
next(state) :=
case
state = idle : {idle, entering};
state = entering & !semaphore : critical;
state = critical : {critical, exiting};
state = exiting : idle;
TRUE : state;
esac;
next(semaphore) :=
case
state = entering : TRUE;
state = exiting : FALSE;
TRUE : semaphore;
esac;
FAIRNESS
running
2.3 Direct Specification
NUSMV allows to specify the FSM directly in terms of propositional formulas. The set of
possible initial states is specified as a formula in the current state variables. A state is initial
if it satisfies the formula. The transition relation is directly specified as a propositional
formula in terms of the current and next values of the state variables. Any current state/next
state pair is in the transition relation if and only if it satisfies the formula.
These two functions are accomplished by the ‘INIT’ and ‘TRANS’ keywords. As an
example, here is a description of the three inverter ring using only TRANS and INIT:
MODULE main
VAR
gate1 : inverter(gate3.output);
gate2 : inverter(gate1.output);
gate3 : inverter(gate2.output);
MODULE inverter(input)
VAR
output : boolean;
INIT
output = FALSE
TRANS
next(output) = !input | next(output) = output
According to the TRANS declaration, for each inverter, the next value of the output
is equal either to the negation of the input, or to the current value of the output. Thus,
in effect, each gate can choose non-deterministically whether or not to delay.
Using TRANS and INIT it is possible to specify inadmissible FSMs, where the set
of initial states is empty or the transition relation is not total. This may result in logical
absurdities.
7
Chapter 3
Simulation
Simulation offers to the user the possibility of exploring the possible executions (traces
from now on) of a NUSMV model. In this way, the user can get familiar with a model
and can acquire confidence with its correctness before the actual verification of properties.
This section describes the basic features of simulation in NUSMV. Further details on the
simulation commands can be found in the NuSMV 2.6 User Manual.
3.1 Trace Strategies
In order to achieve maximum flexibility and degrees of freedom in a simulation session,
NUSMV permits three different trace generation strategies: deterministic, random and
interactive. Each of them corresponds to a different way a state is picked from a set of
possible choices. In deterministic simulation mode the first state of a set (whatever it is)
is chosen, while in the random one the choice is performed nondeterministically. In these
two first modes traces are automatically generated by NUSMV: the user obtains the whole
of the trace in a time without control over the generation itself (except for the simulation
mode and the number of states entered via command line).
In the third simulation mode, the user has a complete control over traces generation by
interactively building the trace. During an interactive simulation session, the system stops
at every step, showing a list of possible future states: the user is requested to choose one
of the items. This feature is particularly useful when one wants to inspect some particular
reactions of the model to be checked. When the number of possible future states exceeds
an internal limit, rather than “confusing” the user with a choice from a high number of
possible evolutions, the system asks the user to “guide” the simulation via the insertion
of some further constraints over the possible future states. The system will continue to
ask for constraints insertion until the number of future states will be under the predefined
threshold. The constraints entered during this phase are accumulated (in a logical product)
in a single big constraint. This constraint is used only for the current step of the simulation
and is discarded before the next step. The system checks the expressions entered by the
user and does not accept them whenever an inconsistency arises. Cases of inconsistency
(i.e. empty set of states) may be caused by:
• the entered expressions (i.e. a & ˜ a);
• the result of the entered expressions conjoined with previous accumulated ones;
• the result of accumulated constraints conjoined with the set of possible future states.
8
3.2 Interactive Mode
A typical execution sequence of a simulation session could be as follows. Suppose we use
the model described below.
MODULE main
VAR
request : boolean;
state : {ready,busy};
ASSIGN
init(state) := ready;
next(state) := case
state = ready & request : busy;
TRUE : {ready,busy};
esac;
As a preliminary step, this model has to read into the NUSMV system. This can be
obtained by executing the following commands (we assume that the model is saved in file
short.smv): 1
system prompt> NuSMV -int short.smv
NuSMV> go
NuSMV>
3.2.1 Choosing an Initial State
In order to start the simulation, an initial state has to be chosen. This can be done in three
ways:
• by default, the simulator uses the current state as a starting point of every new simulation;
this behavior if possible only if a current state is defined (e.g., if we are
exploring a trace);
• if command goto state is used, the user can select any state of an already existent
trace as the current state;
• if pick state is used, then the user can choose the starting state of the simulation
among the initial states of the model; this command has to be used when a current
state does not exist yet (that is when the model has not yet been processed or when
the system has been reset).
At this point of the example current state does not exist, and there is no trace currently
stored in the system. Therefore, an item from the set of initial states has to be picked using
command pick state. A simulation session can be started now, using the simulate
command. Consider for instance the following simulation session:
system prompt> NuSMV -int short.smv
NuSMV> go
NuSMV> pick state -r
NuSMV> print current state -v
Current state is 1.1
request = FALSE
state = ready
1We assume that every NUSMV command is followed by a <RET> keystroke. In the following examples,
NUSMV commands are written like this to distinguish them from system output messages.
9
NuSMV> simulate -r -k 3
********* Starting Simulation From State 1.1 *********
NuSMV> show traces -t
There is 1 trace currently available.
NuSMV> show traces -v
#################### Trace number: 1 ####################
Trace Description: Simulation Trace
Trace Type: Simulation
-> State: 1.1 <request
= FALSE
state = ready
-> State: 1.2 <request
= TRUE
state = busy
-> State: 1.3 <request
= TRUE
state = ready
-> State: 1.4 <request
= TRUE
state = busy
Command pick state -r requires to pick the starting state of the simulation randomly
from the set of initial states of the model. Command simulate -r -k 3 asks to build a threesteps
simulation by picking randomly the next states of the steps. As shown by command
show traces -v, the resulting trace contains 4 states (the initial one, and the three ones
that have been added by the random simulation). We remark that the generated traces are
numbered: every trace is identified by an integer number, while every state belonging to
a trace is identified by a dot notation: for example state 1.3 is the third state of the first
generated trace.
3.2.2 Starting a New Simulation
Now the user can start a new simulation by choosing a new starting state. In the next
example, for instance, the user extends trace 1 by first choosing state 1.4 as the current
state and by then running a random simulation of length 3.
NuSMV> goto state 1.4
The starting state for new trace is:
-> State 2.4 <request
= TRUE
state = busy
NuSMV> simulate -r -k 3
******** Simulation Starting From State 2.4 ********
NuSMV> show traces 2
################### Trace number: 2 ###################
Trace Description: Simulation Trace
Trace Type: Simulation
-> State: 2.1 <request
= TRUE
state = ready
-> State: 2.2 <state
= busy
-> State: 2.3 <request
= FALSE
-> State: 2.4 <-
10
request = TRUE
-> State: 2.5 <request
= FALSE
-> State: 2.6 <state
= ready
-> State: 2.7 <-
NuSMV>
As the reader can see from the previous example, the new trace is stored as trace 2. The
user is also able to interactively choose the states of the trace he wants to build: an example
of an interactive simulation is shown below:
NuSMV> pick state -i
*************** AVAILABLE STATES *************
================= State =================
0) -------------------------
request = TRUE
state = ready
================= State =================
1) -------------------------
request = FALSE
state = ready
Choose a state from the above (0-1): 1 <RET>
Chosen state is: 1
NuSMV> simulate -i -k 1
******** Simulation Starting From State 3.1 ********
*************** AVAILABLE FUTURE STATES *************
================= State =================
0) -------------------------
request = TRUE
state = ready
================= State =================
1) -------------------------
request = TRUE
state = busy
================= State =================
2) -------------------------
request = FALSE
state = ready
================= State =================
3) -------------------------
request = FALSE
state = busy
11
Choose a state from the above (0-3): 0 <RET>
Chosen state is: 0
NuSMV> show traces 3
################### Trace number: 3 ###################
Trace Description: Simulation Trace
Trace Type: Simulation
-> State: 3.1 <request
= FALSE
state = ready
-> State: 3.2 <request
= TRUE
3.2.3 Specifying Constraints
The user can also specify some constraints to restrict the set of states from which the simulator
will pick out. Constraints can be set for both the pick state command and the
simulate command using option -c. For example the following command picks an
initial state by defining a simple constraint:
NuSMV> pick state -c "request = TRUE" -i
*************** AVAILABLE STATES ***************
================= State =================
0) -------------------------
request = TRUE
state = ready
There’s only one future state. Press Return to Proceed. <RET>
Chosen state is: 0
NuSMV> quit
system prompt>
Note how the set of possible states to choose has being restricted (in this case there is
only one future state, so the system will automatically pick it, waiting for the user to press
the <RET> key). We remark that, in the case of command simulate, the constraints defined
using option -c are “global” for the actual trace to be generated, in the sense that they
are always included in every step of the simulation. They are hence complementary to the
constraints entered with the pick state command, or during an interactive simulation
session when the number of future states to be displayed is too high, since these are “local”
only to a single simulation step and are “forgotten” in the next one.
12
Chapter 4
CTL Model Checking
The main purpose of a model checker is to verify that a model satisfies a set of desired properties
specified by the user. In NUSMV, the specifications to be checked can be expressed
in two different temporal logics: the Computation Tree Logic CTL, and the Linear Temporal
Logic LTL extended with Past Operators. CTL and LTL specifications are evaluated
by NUSMV in order to determine their truth or falsity in the FSM. When a specification
is discovered to be false, NUSMV constructs and prints a counterexample, i.e. a trace
of the FSM that falsifies the property. In this section we will describe model checking
of specifications expressed in CTL, while the next section we consider the case of LTL
specifications.
4.1 Computation Tree Logic
CTL is a branching-time logic: its formulas allow for specifying properties that take into
account the non-deterministic, branching evolution of a FSM. More precisely, the evolution
of a FSM from a given state can be described as an infinite tree, where the nodes are the
states of the FSM and the branching in due to the non-determinism in the transition relation.
The paths in the tree that start in a given state are the possible alternative evolutions of the
FSM from that state. In CTL one can express properties that should hold for all the paths
that start in a state, as well as for properties that should hold just for some of the paths.
Consider for instance CTL formula AF p. It expresses the condition that, for all the
paths (A) stating from a state, eventually in the future (F) condition p must hold. That is,
all the possible evolutions of the system will eventually reach a state satisfying condition
p. CTL formula EF p, on the other hand, requires than there exists some path (E) that
eventually in the future satisfies p.
Similarly, formula AG p requires that condition p is always, or globally, true in all the
states of all the possible paths, while formula EG p requires that there is some path along
which condition p is continuously true.
Other CTL operators are:
• A [p U q] and E [p U q], requiring condition p to be true until a state is
reached that satisfies condition q;
• AX p and EX p, requiring that condition p is true in all or in some of the next states
reachable from the current state.
CTL operators can be nested in an arbitrary way and can be combined using logic operators
(!, &, |, ->, <->, ...). Typical examples of CTL formulas are AG ! p (“condition p is
absent in all the evolutions”), AG EF p (“it is always possible to reach a state where
p holds”), and AG (p -> AF q) (“each occurrence of condition p is followed by an
occurrence of condition q”).
13
In NUSMV a CTL specification is given as CTL formula introduced by the keyword
“SPEC”. Whenever a CTL specification is processed, NUSMV checks whether the CTL
formula is true in all the initial states of the model. If this is not a case, then NUSMV
generates a counter-example, that is, a (finite or infinite) trace that exhibits a valid behavior
of the model that does not satisfy the specification. Traces are very useful for identifying the
error in the specification that leads to the wrong behavior. We remark that the generation of
a counter-example trace is not always possible for CTL specifications. Temporal operators
corresponding to existential path quantifiers cannot be proved false by a showing of a single
execution path. Similarly, sub-formulas preceded by universal path quantifier cannot be
proved true by a showing of a single execution path.
4.2 Semaphore Example
Consider the case of the semaphore program described in Chapter 2 [Examples], page 3. A
desired property for this program is that it should never be the case that the two processes
proc1 and proc2 are at the same time in the critical state (this is an example of a
“safety” property). This property can be expressed by the following CTL formula:
AG ! (proc1.state = critical & proc2.state = critical)
Another desired property is that, if proc1 wants to enter its critical state, it eventually
does (this is an example of a “liveness” property). This property can be expressed by the
following CTL formula:
AG (proc1.state = entering -> AF proc1.state = critical)
In order to verify the two formulas on the semaphore model, we add the two corresponding
CTL specification to the program, as follows:
MODULE main
VAR
semaphore : boolean;
proc1 : process user(semaphore);
proc2 : process user(semaphore);
ASSIGN
init(semaphore) := FALSE;
SPEC AG ! (proc1.state = critical & proc2.state = critical)
SPEC AG (proc1.state = entering -> AF proc1.state = critical)
MODULE user(semaphore)
VAR
state : {idle, entering, critical, exiting};
ASSIGN
init(state) := idle;
next(state) :=
case
state = idle : {idle, entering};
state = entering & !semaphore : critical;
state = critical : {critical, exiting};
state = exiting : idle;
TRUE : state;
esac;
next(semaphore) :=
case
state = entering : TRUE;
state = exiting : FALSE;
14
TRUE : semaphore;
esac;
FAIRNESS
running
By running NUSMV with the command
system prompt> NuSMV semaphore.smv
we obtain the following output:
-- specification AG (!(proc1.state = critical & proc2.state = critical))
-- is true
-- specification AG (proc1.state = entering -> AF proc1.state = critical)
-- is false
-- as demonstrated by the following execution sequence
-> State: 1.1 <semaphore
= FALSE
proc1.state = idle
proc2.state = idle
-> Input: 1.2 <_process_selector_
= proc1
-- Loop starts here
-> State: 1.2 <proc1.state
= entering
-> Input: 1.3 <_process_selector_
= proc2
-> State: 1.3 <proc2.state
= entering
-> Input: 1.4 <_process_selector_
= proc2
-> State: 1.4 <semaphore
= FALSE
proc2.state = critical
-> Input: 1.5 <_process_selector_
= proc1
-> State: 1.5 <->
Input: 1.6 <_process_selector_
= proc2
-> State 1.6 <proc2.state
= exiting
-> Input: 1.7 <_process_selector_
= proc2
-> State 1.7 <semaphore
= FALSE
proc2.state = idle
Note that process selector is a special variable which contains the name of the process
that will execute to cause a transition to the next state. The ’Input’ section displays
the values of variables that the model has no control over, that is it cannot change their
value. Since processes are chosen nondeterministically in this model, it has no control over
the value of process selector .
NUSMV tells us that the first CTL specification is true: it is never the case that the two
processes will be at the same time in the critical region. On the other hand, the second specification
is false. NUSMV produces a counter-example path where initially proc1 goes
to state entering (state 1.2), and then a loop starts in which proc2 repeatedly enters its
15
critical region (state 1.4) and then returns to its idle state (state 1.7); in the loop, proc1
is activated only when proc2 is in the critical region (input 1.5), and is therefore not able
to enter its critical region (state 1.5). This path not only shows that the specification is false,
it also points out why can it happen that proc1 never enters its critical region.
Note that in the printout of a cyclic, infinite counter-example the starting point of the
loop is marked by -- loop starts here. Moreover, in order to make it easier to
follow the action in systems with a large number of variables, only the values of variables
that have changed in the last step are printed in the states of the trace.
16
Chapter 5
LTL Model Checking
5.1 Linear Temporal Logic
NUSMV allows for specifications expressed in LTL. Intuitively, while CTL specifications
express properties over the computation tree of the FSM (branching-time approach), LTL
characterizes each linear path induced by the FSM (linear-time approach). The two logics
have in general different expressive power, but also share a significant intersection that
includes most of the common properties used in practice. Typical LTL operators are:
• F p (read “in the future p”), stating that a certain condition p holds in one of the
future time instants;
• G p (read “globally p”), stating that a certain condition p holds in all future time
instants;
• p U q (read “p until q”), stating that condition p holds until a state is reached where
condition q holds;
• X p (read “next p”), stating that condition p is true in the next state.
We remark that, differently from CTL, LTL temporal operators do not have path quantifiers.
In fact, LTL formulas are evaluated on linear paths, and a formula is considered true in a
given state if it is true for all the paths starting in that state.
5.2 Semaphore Example
Consider the case of the semaphore program and of the safety and liveness properties already
described in Chapter 4 [CTL Model Checking], page 13. These properties correspond
to LTL formulas
G ! (proc1.state = critical & proc2.state = critical)
expressing that the two processes cannot be in the critical region at the same time, and
G (proc1.state = entering -> F proc1.state = critical)
expressing that whenever a process wants to enter its critical session, it eventually does.
If we add the two corresponding LTL specification to the program, as follows:1
1In NUSMV a LTL specification are introduced by the keyword “LTLSPEC”.
17
MODULE main
VAR
semaphore : boolean;
proc1 : process user(semaphore);
proc2 : process user(semaphore);
ASSIGN
init(semaphore) := FALSE;
LTLSPEC G ! (proc1.state = critical & proc2.state = critical)
LTLSPEC G (proc1.state = entering -> F proc1.state = critical)
MODULE user(semaphore)
VAR
state : {idle, entering, critical, exiting};
ASSIGN
init(state) := idle;
next(state) :=
case
state = idle : {idle, entering};
state = entering & !semaphore : critical;
state = critical : {critical, exiting};
state = exiting : idle;
TRUE : state;
esac;
next(semaphore) :=
case
state = entering : TRUE;
state = exiting : FALSE;
TRUE : semaphore;
esac;
FAIRNESS
running
NUSMV produces the following output:
-- specification G (!(proc1.state = critical & proc2.state = critical))
-- is true
-- specification G (proc1.state = entering -> F proc1.state = critical)
-- is false
-- as demonstrated by the following execution sequence
Trace Description: CTL Counterexample
Trace Type: Counterexample
-> State: 1.1 <semaphore
= FALSE
proc1.state = idle
proc2.state = idle
-> Input: 1.2 <_process_selector_
= proc2
-- Loop starts here
-> State 1.2 <-
[...]
That is, the first specification is true, while the second is false and a counter-example path
is generated.
18
5.3 Past Temporal Operators
In NUSMV, LTL properties can also include past temporal operators. Differently from
standard temporal operators, that allow to express properties over the future evolution of
the FSM, past temporal operators allow to characterize properties of the path that lead to
the current situation. The typical past operators are:
• O p (read ”once p ”), stating that a certain condition p holds in one of the past time
instants;
• H p (read ”historically p ”), stating that a certain condition p holds in all previous
time instants;
• p S q (read ”p since q ”), stating that condition p holds since a previous state
where condition q holds;
• Y p (read ”yesterday p ”), stating that condition p holds in the previous time instant.
Past temporal operators can be combined with future temporal operators, and allow for the
compact characterization of complex properties.
A detailed description of the syntax of LTL formulas can be found in the NuSMV 2.6
User Manual.
19
Chapter 6
Bounded Model Checking
In this section we give a short introduction to the use of Bounded Model Checking (BMC)
in NUSMV. For a more in-depth introduction to the theory underlying BMC please refer to
[BCCZ99]. To perform SAT based bounded model checking NUSMV needs to be linked
to an external Sat Solver. If none is provided, all Bounded Model Checking routines in
NUSMV will not be available. Currently NUSMV supports two SAT solvers: MiniSat and
Zchaff.
Consider the following model, representing a simple, deterministic counter modulo 8
(we assume that the following specification is contained in file bmc tutorial.smv):
MODULE main
VAR
y : 0..15;
ASSIGN
init(y) := 0;
TRANS
case
y = 7 : next(y) = 0;
TRUE : next(y) = ((y + 1) mod 16);
esac
This slightly artificial model has only the state variable y, ranging from 0 to 15. The
values of y are limited by the transition relation to the [0, 7] interval. The counter starts
from 0, deterministically increments by one the value of y at each transition up to 7, and
then restarts from zero.
6.1 Checking LTL Specifications
We would like to check with BMC the LTL specification G ( y=4 -> X y=6 ) expressing
that “each time the counter value is 4, the next counter value will be 6”. This
specification is obviously false, and our first step is to use NUSMV BMC to demonstrate its
falsity. To this purpose, we add the following specification to file bmc tutorial.smv:
LTLSPEC G ( y=4 -> X y=6 )
and we instruct NUSMV to run in BMC by using command-line option -bmc:
system prompt> NuSMV -bmc bmc tutorial.smv
-- no counterexample found with bound 0 for specification
G(y = 4 -> X y = 6)
-- no counterexample found with bound 1 for ...
20
-- no counterexample found with bound 2 for ...
-- no counterexample found with bound 3 for ...
-- no counterexample found with bound 4 for ...
-- specification G (y = 4 -> X y = 6) is false
-- as demonstrated by the following execution sequence
Trace Description: BMC Counterexample
Trace Type: Counterexample
-> State: 1.1 <y
= 0
-> State: 1.2 <y
= 1
-> State: 1.3 <y
= 2
-> State: 1.4 <y
= 3
-> State: 1.5 <y
= 4
-> State: 1.6 <y
= 5
system prompt>
NUSMV has found that the specification is false, and is showing us a counterexample, i.e.
a trace where the value of y becomes 4 (at time 4) and at the next step is not 6.
bound: 0 1 2 3 4 5
o--->o--->o--->o--->o--->o
state: y=0 y=1 y=2 y=3 y=4 y=5
The output produced by NUSMV shows that, before the counterexample of length 5 is
found, NUSMV also tried to finds counterexamples of lengths 0 to 4. However, there are
no such counterexamples. For instance, in the case of bound 4, the traces of the model have
the following form:
bound: 0 1 2 3 4
o--->o--->o--->o--->o
state: y=0 y=1 y=2 y=3 y=4
In this situation, y gets the value 4, but it is impossible for NUSMV to say anything about
the following state.
In general, in BMC mode NUSMV tries to find a counterexample of increasing length,
and immediately stops when it succeeds, declaring that the formula is false. The maximum
number of iterations can be controlled by using command-line option -bmc length. The
default value is 10. If the maximum number of iterations is reached and no counter-example
is found, then NUSMV exits, and the truth of the formula is not decided. We remark that
in this case we cannot conclude that the formula is true, but only that any counter-example
should be longer than the maximum length.
system prompt> NuSMV -bmc -bmc length 4 bmc tutorial.smv
-- no counterexample found with bound 0 for ...
-- no counterexample found with bound 1 for ...
-- no counterexample found with bound 2 for ...
-- no counterexample found with bound 3 for ...
-- no counterexample found with bound 4 for ...
system prompt>
Let us consider now another property, !G F (y = 2), stating that y gets the value 2
only a finite number of times. Again, this is a false property due to the cyclic nature of the
model. Let us modify the specification of file bmc tutorial.smv as follows:
21
LTLSPEC !G F (y = 2)
and let us run NUSMV in BMC mode:
system prompt> NuSMV -bmc bmc tutorial.smv
-- no counterexample found with bound 0 for specification ! G F y = 2
-- no counterexample found with bound 1 for ...
-- no counterexample found with bound 2 for ...
-- no counterexample found with bound 3 for ...
-- no counterexample found with bound 4 for ...
-- no counterexample found with bound 5 for ...
-- no counterexample found with bound 6 for ...
-- no counterexample found with bound 7 for ...
-- specification ! G F y = 2 is false
-- as demonstrated by the following execution sequence
Trace Description: BMC Counterexample
Trace Type: Counterexample
-- Loop starts here
-> State: 1.1 <y
= 0
-> State: 1.2 <y
= 1
-> State: 1.3 <y
= 2
-> State: 1.4 <y
= 3
-> State: 1.5 <y
= 4
-> State: 1.6 <y
= 5
-> State: 1.7 <y
= 6
-> State: 1.8 <y
= 7
-> State: 1.9 <y
= 0
system prompt>
In this example NUSMV has increased the problem bound until a cyclic behavior of length
8 is found that contains a state where y value is 2. Since the behavior is cyclic, state 1.3 is
entered infinitely often and the property is false.
=
+---------------------------------------+
| |
| |
| |
o--->o--->o--->o--->o--->o--->o--->o--->o
bound: 0 1 2 3 4 5 6 7 8
y value: 0 1 2 3 4 5 6 7 0
6.2 Finding Counterexamples
In general, BMC can find two kinds of counterexamples, depending on the property being
analyzed. For safety properties (e.g. like the first one used in this tutorial), a counterexample
is a finite sequence of transitions through different states. For liveness properties,
22
counterexamples are infinite but periodic sequences, and can be represented in a bounded
setting as a finite prefix followed by a loop, i.e. a finite sequence of states ending with a
loop back to some previous state. So a counterexample which demonstrates the falsity of a
liveness property as “! G F p” cannot be a finite sequence of transitions. It must contain
a loop which makes the infinite sequence of transitions as well as we expected.
=
+---------------------+
| |
| |
| |
o--->o-...->o--->o--->o-...->o--->o--->o--->o--->
time: 0 1 l-1 l l+1 k-2 k-1 k k+1 ...
Consider the above figure. It represents an examples of a generic infinite counterexample,
with its two parts: the prefix part (times from 0 to l − 1), and the loop part (indefinitely
from l to k − 1). Because the loop always jumps to a previous time it is called loopback.
The loopback condition requires that state k is identical to state l. As a consequence, state
k + 1 is forced to be equal to state l + 1, state k + 2 to be equal to state l + 2, and so on.
A fine-grained control of the length and of the loopback condition for the counterexample
can be specified by using command check ltlspec bmc onepb in interactive
mode. This command accepts options -k, that specifies the length of the counter-example
we are looking for, and -l, that defines the loopback condition. Consider the following
interactive session:
system prompt> NuSMV -int bmc tutorial.smv
NuSMV> go bmc
NuSMV> check ltlspec bmc onepb -k 9 -l 0
-- no counterexample found with bound 9 and loop at 0 for specification
! G F y = 2
NuSMV> check ltlspec bmc onepb -k 8 -l 1
-- no counterexample found with bound 8 and loop at 1 for specification
! G F y = 2
NuSMV> check ltlspec bmc onepb -k 9 -l 1
-- specification ! G F y = 2 is false
-- as demonstrated by the following execution sequence
Trace Description: BMC Counterexample
Trace Type: Counterexample
-> State: 1.1 <y
= 0
-- Loop starts here
-> State: 1.2 <y
= 1
-> State: 1.3 <y
= 2
-> State: 1.4 <y
= 3
-> State: 1.5 <y
= 4
-> State: 1.6 <y
= 5
-> State: 1.7 <y
= 6
-> State: 1.8 <y
= 7
23
-> State: 1.9 <y
= 0
-> State: 1.10 <y
= 1
NuSMV> quit
system prompt>
NUSMV did not find a counterexample for cases (k = 9, l = 0) and (k = 8, l = 1).
The following figures show that these case look for counterexamples that do not match with
the model of the counter, so it is not possible for NUSMV to satisfy them.
k = 9, l = 0:
=
+--------------------------------------------+
| |
| |
| |
o--->o--->o--->o--->o--->o--->o--->o--->o--->o
bound: 0 1 2 3 4 5 6 7 8 9
y value: 0 1 2 3 4 5 6 7 0 1
k = 8, l = 1:
=
+----------------------------------+
| |
| |
| |
o--->o--->o--->o--->o--->o--->o--->o--->o
bound: 0 1 2 3 4 5 6 7 8
y value: 0 1 2 3 4 5 6 7 0
Case (k = 9, l = 1), instead allows for a counter-example:
k = 9, l = 1:
=
+---------------------------------------+
| |
| |
| |
o--->o--->o--->o--->o--->o--->o--->o--->o--->o
bound: 0 1 2 3 4 5 6 7 8 9
y value: 0 1 2 3 4 5 6 7 0 1
In NUSMV it is possible to specify the loopback condition in four different ways:
• The loop as a precise time-point. Use a natural number as the argument of option
-l.
• The loop length. Use a negative number as the argument of option -l. The negative
number is the loop length, and you can also imagine it as a precise time-point loop
relative to the path bound.
• No loopback. Use symbol ‘X’ as the argument of option -l. In this case NUSMV
will not find infinite counterexamples.
24
• All possible loops. Use symbol ‘*’ as the argument of option -l. In this case
NUSMV will search counterexamples for paths with any possible loopback structure.
A counterexample with no loop will be also searched. This is the default value
for option -l.
In the following example we look for a counter-example of length 12 with a loop of
length 8:
system prompt> NuSMV -int bmc tutorial.smv
NuSMV> go bmc
NuSMV> check ltlspec bmc onepb -k 12 -l -8
-- specification ! G F y = 2 is false
-- as demonstrated by the following execution sequence
Trace Description: BMC Counterexample
Trace Type: Counterexample
-> State: 1.1 <y
= 0
-> State: 1.2 <y
= 1
-> State: 1.3 <y
= 2
-> State: 1.4 <y
= 3
-- Loop starts here
-> State: 1.5 <y
= 4
-> State: 1.6 <y
= 5
-> State: 1.7 <y
= 6
-> State: 1.8 <y
= 7
-> State: 1.9 <y
= 0
-> State: 1.10 <y
= 1
-> State: 1.11 <y
= 2
-> State: 1.12 <y
= 3
-> State: 1.13 <y
= 4
NuSMV>
This picture illustrates the produced counterexample in a more effective way:
=
+-------------------------------+
| |
| |
| |
o-->o-->o-->o-->o-->o-->o-->o-->o-->o-->o-->o-->o
bound: 0 1 2 3 4 5 6 7 8 9 10 11 12
y value: 0 1 2 3 4 5 6 7 0 1 2 3 4
25
If no loopback is specified, NUSMV is not able to find a counterexample for the given
liveness property:
system prompt> NuSMV -int bmc tutorial.smv
NuSMV> go bmc
NuSMV> check ltlspec bmc onepb -k 12 -l X
-- no counterexample found with bound 12 and no loop for ...
NuSMV>
6.3 Checking Invariants
Ahead of version 2.2.2, NUSMV supported only the 2-step inductive reasoning algorithm
for invariant checking. As will become clear from this tutorial, this algorithm is not complete,
so in certain cases it cannot be used to state whether an invariant specification is
actually true or false.
Since version 2.2.2, NUSMV supports total inductive reasoning, which might be heavier
than the 2-step approach but can make invariant specifications provable even when the
latter fails.
Please refer to [ES04] for a more in-depth explanation of the theory underlying the
algorithms for total temporal induction.
6.3.1 2-Step Inductive Reasoning
Bounded Model Checking in NUSMV can be used not only for checking LTL specification,
but also for checking invariants. An invariant is a propositional property which must always
hold. BMC tries to prove the truth of invariants via a process of inductive reasoning, by
checking if (i) the property holds in every initial state, and (ii) if it holds in any state
reachable from any state where it holds.
Let us modify file bmc tutorial.smv by replacing the LTL specification with the
following invariant specification:
INVARSPEC y in (0..12)
and let us run NUSMV in BMC mode:
system prompt> NuSMV -bmc bmc tutorial.smv
-- cannot prove the invariant y in (0 .. 12) : the induction fails
-- as demonstrated by the following execution sequence
Trace Description: BMC Counterexample
Trace Type: Counterexample
-> State: 1.1 <y
= 12
-> State: 1.2 <y
= 13
system prompt>
NUSMV reports that the given invariant cannot be proved, and it shows a state satisfying
“y in (0..12)” that has a successor state not satisfying ”y in (0..12)”. This two-steps
sequence of assignments shows why the induction fails. Note that NUSMV does not state
the given formula is really false, but only that it cannot be proven to be true using the 2-step
inductive reasoning described previously.
If we try to prove the stronger invariant y in (0..7) we obtain:
system prompt> NuSMV -bmc bmc tutorial.smv
-- invariant y in (0 .. 7) is true
system prompt>
26
In this case NUSMV is able to prove that y in (0..7) is true. As a consequence, also
the weaker invariant y in (0..12) is true, even if NUSMV is not able to prove it in
BMC mode. On the other hand, the returned counter-example can be used to strengthen
the invariant, until NUSMV is able to prove it.
Now we check the false invariant y in (0..6):
-- cannot prove the invariant y in (0 .. 6) : the induction fails
-- as demonstrated by the following execution sequence
Trace Description: BMC Counterexample
Trace Type: Counterexample
-> State: 1.1 <y
= 6
-> State: 1.2 <y
= 7
NuSMV>
As for property y in (0..12), NUSMV returns a two steps sequence showing that the
induction fails. The difference is that, in the former case state ’y=12’ is NOT reachable,
while in the latter case the state ’y=6’ can be reached.
6.3.2 Complete Invariant Checking
Since version 2.2.2, complete invariant checking can be obtained by running the command
check invar bmc in interactive mode, and specifying the algorithm ‘‘een-sorensson’’
using the option -a. If an incremental sat solver is available, the command
check invar bmc inc may also be used.
The classic 2-step algorithm was not able to prove directly the truth of the invariant
y in (0..12). This invariant can now be easily checked by the complete invariant
checking algorithm.
system prompt> NuSMV -int bmc tutorial.smv
NuSMV> go bmc
NuSMV> check invar bmc -a een-sorensson -p "y in (0..12)"
-- no proof or counterexample found with bound 0 ...
-- no proof or counterexample found with bound 1 ...
-- no proof or counterexample found with bound 2 ...
-- no proof or counterexample found with bound 3 ...
-- no proof or counterexample found with bound 4 ...
-- no proof or counterexample found with bound 5 ...
-- invariant y in (0 .. 12) is true
NuSMV>
As can be inferred from this example, NUSMV proved that the invariant actually holds,
requiring a length of 6 to prove it.
Complete invariant checking can also prove that an invariant does not hold, and provide
a convincing counter-example for it. For example property y in (0..6) that the
‘‘classic’’ algorithm failed to check is now proved to be false:
NuSMV> check invar bmc -a een-sorensson -p "y in (0..6)"
-- no proof or counterexample found with bound 0 ...
-- no proof or counterexample found with bound 1 ...
-- no proof or counterexample found with bound 2 ...
-- no proof or counterexample found with bound 3 ...
-- no proof or counterexample found with bound 4 ...
-- no proof or counterexample found with bound 5 ...
27
-- no proof or counterexample found with bound 6 ...
-- invariant y in (0 .. 6) is false
-- as demonstrated by the following execution sequence
Trace Description: BMC Counterexample
Trace Type: Counterexample
-> State: 1.1 <y
= 0
-> State: 1.2 <y
= 1
-> State: 1.3 <y
= 2
-> State: 1.4 <y
= 3
-> State: 1.5 <y
= 4
-> State: 1.6 <y
= 5
-> State: 1.7 <y
= 6
-> State: 1.8 <y
= 7
NuSMV>
The provided counter-example shows that y actually can reach a value out of the set
(0..6).
28
Bibliography
[BCCZ99] A. Biere, A. Cimatti, E. Clarke, and Y. Zhu. Symbolic model checking without
bdds. In Tools and Algorithms for Construction and Analysis of Systems, In
TACAS’99, March 1999.
[ES04] Niklas E´en and Niklas S¨orensson. Temporal induction by incremental sat solving.
In Ofer Strichman and Armin Biere, editors, Electronic Notes in Theoretical
Computer Science, volume 89. Elsevier, 2004.
29
Command Index
-bmc length, 21
check ltlspec bmc onepb, 23
goto state, 9
pick state, 9, 10, 12
print current state, 9
show traces, 10
simulate, 10, 12
30
Index
A
Asynchronous Systems, 5
B
Bounded Model Checking, 20
Checking Invariants, 26
Checking LTL Specifications, 20
Finding Counterexamples, 22
C
Checking Invariants (BMC), 26
2-step Inductive Reasoning, 26
Complete Invariant Checking, 27
Checking LTL Specifications (BMC), 20
Choosing an Initial State, 9
Computation Tree Logic, 13
CTL Model Checking, 13
CTL Operators, 13
D
DEFINE keyword, 4
Direct Specification, 7
E
Examples, 3
Binary Counter, 4
Inverter Ring, 5
Semaphore
Asynchronous, 6
CTL, 14
LTL, 17
Single Process, 3
F
FAIRNESS keyword, 5
Finding Counterexamples (BMC), 22
I
Inductive Reasoning (2-Step), 26
INIT keyword, 7
Input Language, 3
Interactive Mode, 9
Introduction, 2
L
Linear Temporal Logic, 17
Liveness Property, 14
Loopback Condition, 24
LTL Model Checking, 17
LTL Operators, 17
LTLSPEC keyword, 17
M
Model Checking
BMC, 20
CTL, 13
LTL, 17
Modules, 4
P
Past Temporal Operators (LTL), 19
process keyword, 5
S
Safety Property, 14
Simulation, 8
Specification
CTL, 13
LTL, 17
Specifying Constraints, 12
SPEC keyword, 13
Starting a New Simulation, 10
Synchronous Systems, 3
T
Trace Strategies, 8
Tracing
Choosing an Initial State, 9
Interactive Mode, 9
Specifying Constraints, 12
Starting a New Simulation, 10
Strategies, 8
TRANS keyword, 7
U
union operator, 6
31