The DiVinE 2 model checker
[http://web.archive.org/web/20120723095042/http://divine.fi.muni.cz/index.html]
used to have a specification language called DVE, for modelling
processes synchonizing through channels
[http://web.archive.org/web/20120723095115/http://divine.fi.muni.cz/language.html].
A lot of models can be found in the BEEM database at
http://paradise.fi.muni.cz/beem/
The LTSmin group [https://ltsmin.utwente.nl/] patched
DiVinE and to compile models as dynamic libraries. This dynamic
library provides a very simple C interface (no C++) and extra
information about state variables (name, type, possible values).
They also distribute SpinS, a compiler for PROMELA models generating
dynamic libraries with the same interface.
Spot uses this interface so you will need to install their version of
these tools to use Spot with DVE or PROMELA models.
The source code for our interface is in spot/ltsmin/ and generates a
separate library, libspotltsmin.so, that has to be linked in addition
to libspot.so. The current directory contains some testing code based
on a toy modelchecker built upon the above interface: using it require
an installation of DiVinE or SpinS (preferably both for testing
purpose).
Installation of DiVinE
======================
Use the following commands to compile and install the patched version
of DiVinE.
git clone https://gitlab.lrde.epita.fr/spot/divine-ltsmin-deb
cd divine-ltsmin-deb
mkdir _build && cd _build
cmake .. -DMURPHI=OFF -DHOARD=OFF -DGUI=OFF -DRX_PATH= -DCMAKE_INSTALL_PREFIX=$HOME/usr
make
make install
The CMAKE_INSTALL_PREFIX variable is the equivalent of the --prefix
option of configure scripts. If you decide to install in $HOME/usr
like I do, make sure that $HOME/usr/bin is in your PATH. If you omit
the CMAKE_INSTALL_PREFIX setting, it will default to /usr/local.
If you are using MacOS, you must add option -DHOARD=OF to the cmake
command line in order to make it compile without errors. Also,
DiVinE 2 only compiles with the GNU std C++ library; as a consequence,
you must provide the option -DCMAKE_CXX_FLAGS="-stdlib=libstdc++" to
the cmake command line.
The above git repository is our own copy of the LTSmin fork of Divine,
that we patched to generate Debian packages for amd64. If you use
our Debian repository [https://spot.lrde.epita.fr/install.html#Debian]
you can actually install this version of divine with just:
apt-get install divine-ltsmin
After installation, you can check that compilation works by running
the following command on any DVE model. It should create a file
model.dve2C (which is a dynamic library).
divine compile --ltsmin model.dve
Installation of SpinS
======================
The extended version of SpinJa is called SpinS and should be included
with LTSmin. You can download LTSmin from their website
[http://ltsmin.utwente.nl/] and install it following the INSTALL
instructions.
To compile a promela model, simply run the following command:
spins model.pm
It should create a dynamic library called model.pm.spins in the
current directory.
Usage with Spot
===============
The function load_dve2() defined in dve2.hh in this directory will
accept either a model or its compiled version as file argument. In
the former case, it will call "divine compile --ltsmin model.dve" or
"spins model.pm" depending on the file extension, only if a compiled
model with the corresponding file extension (.dve2C or .spins) does
not exist or is older. Then it will load the compiled model
dynamically.
load_dve2() also requires a set of atomic propositions that should
be observed in the model. These are usually the atomic propositions
that occur in the formula to verify, but it might be a larger set.
There are two kinds of atomic propositions, those that refer to the
state of a process, and those that compare the value of a variable.
Let's have some example on an excerpt of the beem-peterson.4.dve
model included in this directory:
byte pos[4];
byte step[4];
process P_0 {
byte j=0, k=0;
state NCS, CS, wait ,q2,q3;
init NCS;
trans
NCS -> wait { effect j = 1; },
wait -> q2 { guard j < 4; effect pos[0] = j;},
q2 -> q3 { effect step[j-1] = 0, k = 0; },
q3 -> q3 { guard k < 4 && (k == 0 || pos[k] < j); effect k = k+1;},
q3 -> wait { guard step[j-1] != 0 || k == 4; effect j = j+1;},
wait -> CS { guard j == 4; },
CS -> NCS { effect pos[0] = 0;};
}
The following atomic propositions could be used in LTL formula:
P_0.CS Process P_0 is in state CS.
"pos[3] < 3" Global variable pos[3] is less than 3.
"P_0.j >= 2" Process P_0's variable j is greater or equal to 2.
P_0.j This is equivalent to "P_0.j != 0".
Comparison operators available are "<", ">", ">=", "<=", "==", and
"!=". The left operand should always be a variable and the right
operand should always be a number, so you cannot write something
like "P_0.j <= P_0.i".
Because the LTL parser knows nothing about the details of the
languages we interface with, every atomic proposition that cannot be
expressed using only alphanumeric characters (plus `_' and `.')
should be enclosed in double quote.
Caveat: "P_0.j >= 2" and " P_0.j>=2" (watch the spaces!) are
considered to be two distinct atomic propositions with the same
semantics.
Examples
========
Using the modelcheck program built into this directory, we can verify
that the critical section is accessed infinitely often by some
processes using:
% ./modelcheck --model beem-peterson.4.dve --formula '!GF(P_0.CS|P_1.CS|P_2.CS|P_3.CS)' --is-empty
2239039 unique states visited
0 strongly connected components in search stack
11449204 transitions explored
1024245 items max in DFS search stack
111081 pages allocated for emptiness check
no accepting run found
Process P_0 can starve, waiting to enter in critical section:
% ./modelcheck --model beem-peterson.4.dve --formula '!G(P_0.wait -> F P_0.CS)' --is-empty
3978 unique states visited
31 strongly connected components in search stack
4723 transitions explored
3302 items max in DFS search stack
1099 pages allocated for emptiness check
an accepting run exists (use -c to print it)
Variable pos[1] is not always < 3 (this formula makes no sense, it
is just to demonstrate the use of double quote).
% ./modelcheck --model beem-peterson.4.dve --formula '!G("pos[1] < 3")' --is-empty
130 unique states visited
61 strongly connected components in search stack
132 transitions explored
130 items max in DFS search stack
512 pages allocated for emptiness check
an accepting run exists (use -c to print it)
Two state-compression techniques have been implemented as experiments.
Prefer the -Z option if your model use only non-negative value less
than 2^28, it is way faster than -z (which will work for all values).
Activating state compression will often reduce runtime. Compare:
$ ./modelcheck --model beem-peterson.4.dve --formula '!GF(P_0.CS|P_1.CS|P_2.CS|P_3.CS)' --is-empty --timer
2239039 unique states visited
0 strongly connected components in search stack
11449204 transitions explored
1024245 items max in DFS search stack
111081 pages allocated for emptiness check
no accepting run found | user time | sys. time | total |
name | ticks % | ticks % | ticks % | n
-------------------------------------------------------------------------------
loading ltsmin model | 0 0.0 | 0 0.0 | 0 0.0 | 1
parsing formula | 0 0.0 | 0 0.0 | 0 0.0 | 1
running emptiness chec | 672 100.0 | 13 100.0 | 685 100.0 | 1
translating formula | 0 0.0 | 0 0.0 | 0 0.0 | 1
-------------------------------------------------------------------------------
TOTAL | 672 100.0 | 13 100.0 | 685 100.0 |
$ ./modelcheck --model beem-peterson.4.dve --formula '!GF(P_0.CS|P_1.CS|P_2.CS|P_3.CS)' --is-empty --timer -z 2
2239039 unique states visited
0 strongly connected components in search stack
11449204 transitions explored
1024245 items max in DFS search stack
85991 pages allocated for emptiness check
no accepting run found
| user time | sys. time | total |
name | ticks % | ticks % | ticks % | n
-------------------------------------------------------------------------------
loading ltsmin model | 40 6.1 | 2 16.7 | 42 6.2 | 1
parsing formula | 0 0.0 | 0 0.0 | 0 0.0 | 1
running emptiness chec | 620 93.8 | 10 83.3 | 630 93.6 | 1
translating formula | 1 0.2 | 0 0.0 | 1 0.1 | 1
-------------------------------------------------------------------------------
TOTAL | 661 100.0 | 12 100.0 | 673 100.0 |
It's a 14% speedup in this case, be the improvement can be more
important on larger models.
The parallel deadlock detection has also been implemented is this tool:
% ./modelcheck --model beem-peterson.4.dve --has-deadlock --csv -p 1
Thread #0: on CPU 0
---- Thread number : 0
1119560 unique states visited
3864896 transitions explored
78157 items max in DFS search stack
1316 milliseconds
Find following the csv: thread_id,walltimems,type,states,transitions
@th_0,1316,NO-DEADLOCK,1119560,3864896
Summary :
No no deadlock found!
Find following the csv: model,walltimems,memused,type,states,transitions
#beem-peterson.4.dve,1317,103681,NO-DEADLOCK,1119560,3864896
Running the same algorithm with 3 threads save 40% of the computation time:
$ ./modelcheck --model beem-peterson.4.dve --has-deadlock --csv -p 3
Thread #1: on CPU 1
Thread #2: on CPU 2
Thread #0: on CPU 0
---- Thread number : 0
417923 unique states visited
1418775 transitions explored
56403 items max in DFS search stack
819 milliseconds
Find following the csv: thread_id,walltimems,type,states,transitions
@th_0,819,NO-DEADLOCK,417923,1418775
---- Thread number : 1
526175 unique states visited
1813440 transitions explored
69322 items max in DFS search stack
819 milliseconds
Find following the csv: thread_id,walltimems,type,states,transitions
@th_1,819,NO-DEADLOCK,526175,1813440
---- Thread number : 2
404501 unique states visited
1411645 transitions explored
61888 items max in DFS search stack
819 milliseconds
Find following the csv: thread_id,walltimems,type,states,transitions
@th_2,819,NO-DEADLOCK,404501,1411645
Summary :
No no deadlock found!
Find following the csv: model,walltimems,memused,type,states,transitions
#beem-peterson.4.dve,820,158211,NO-DEADLOCK,404501,1411645
One can observe that when possible (i.e., if the OS allows it) we try
has much as possible to pin threads to a CPU.
