Combining ModelCheckin g and Abstra ct Interpret a t ion P - - PowerPoint PPT Presentation

combining model checkin g and abstra ct interpret a t ion
SMART_READER_LITE
LIVE PREVIEW

Combining ModelCheckin g and Abstra ct Interpret a t ion P - - PowerPoint PPT Presentation

Combining ModelCheckin g and Abstra ct Interpret a t ion P arallel Combina tio n of Abstra ct Interpret a ti on Ho w and ModelBased A utoma tic Anal ysis of Softw are Abstract symb olic metho


slide-1
SLIDE 1 P arallel Combina tio n
  • f
Abstra ct Interpret a ti
  • n
and ModelBased A utoma tic Anal ysis
  • f
Softw are P atrick COUSOT
  • Ecole
No rmale Sup
  • erieure
DMI
  • rue
dUlm
  • P
a ris cedex
  • F
rance cousotd mi en s fr httpw ww e ns fr
  • cousot
Radhia COUSOT CNRS
  • Ecole
P
  • lytechnique
LIX
  • P
alaiseau cedex F rance rcousot li x pol yt ec hn iq ue fr httpli x pol yt ec hn iq ue fr
  • radhia
AAS P a ris Janua ry
  • Combining
ModelCheckin g and Abstra ct Interpret a ti
  • n
Why
  • Mo
delchecking
  • Finite
state space
  • Sound
and complete p rop ert y verication
  • Abstract
Interp retation
  • Innite
state space
  • Sound
but uncomplete p rop ert y determination Combining ModelCheckin g and Abstra ct Interpret a t ion Ho w
  • Abstract
symb
  • lic
metho ds
  • Use
symb
  • lic
rep resent ations
  • f
p rop erties BDDs convex p
  • ly
hedra
  • One
can mak e app ro ximations eg widenings
  • Appr
  • xima
t e pr
  • per
ties
  • f
an exa ct model
  • Mo
del abstraction
  • The
nite mo del is an abstraction
  • f
the system
  • Exa
ct pr
  • per
ties
  • f
an appr
  • xima
te model
  • In
this p aper
  • P
a rallel combination
  • f
mo delchecking and abstract interp retation
  • Mo
delchecking
  • Exact
symb
  • lic
rep resentat ion
  • f
p rop erties
  • The
mo del is an exact rep resent ation
  • f
the system
  • Exact
p rop erties
  • f
exact mo del
  • Abstract
interp retation
  • Prelimina
rypa rallel analysis
  • f
the mo del b y abstract in terp retation
  • Limit
the state sea rch space
  • Exa
ct pr
  • per
ties
  • f
an exa ct submodel P Cousot
  • R
Cousot
  • AAS
Jan
slide-2
SLIDE 2 Example Maximum Dela y Pr
  • blem
  • Find
the maximum dela y to reach a nal state sta rting from some initial state
  • Maximum
Dela y Algorithm maximum
  • pro
cedure maximum I
  • F
  • R
  • S
  • n
  • R
  • S
  • F
  • while
R
  • R
  • R
  • I
  • do
R
  • R
n
  • n
  • R
  • p
re t R
  • S
  • F
  • d
return if R
  • R
then
  • else
n
  • Halb
w achs N Dela ys analysis in synchronous p rograms CA V
  • LNCS
  • pp
  • Camp
  • s
S Cla rk e E Ma rrero W and Minea M V erus A to
  • l
fo r quantitative analysis
  • f
nitestate realtime systems Pro c A CM SIGPLAN
  • W
  • rkshop
  • n
Languages Compilers
  • T
  • ls
fo r RealTime Systems La Jolla Calif jun
  • pp
  • Execution
tra ce
  • f
the maximum algorithm It is useless to explo re the states which a re not
  • descendants
  • f
the initial states
  • ascendants
  • f
the initial states
  • Maximum
Dela y Algorithm maximum
  • with
st a te sear ch sp a ce restriction pro cedure maximum I
  • F
  • R
  • S
  • n
  • R
  • U
  • F
  • while
R
  • R
  • R
  • I
  • do
R
  • R
n
  • n
  • R
  • p
re t R
  • U
n
  • F
  • d
return if R
  • R
then
  • else
n where n
  • U
n
  • U
def
  • p
  • stt
  • I
  • p
re t
  • F
P Cousot
  • R
Cousot
  • AAS
Jan
slide-3
SLIDE 3 Execution tra ce
  • f
the maximum algorithm
  • Any
upp erapp ro ximations U
  • U
  • U
n
  • f
U can b e used
  • In
the w
  • rst
case U n
  • S
all states hence maximum
  • maximum
  • Anal
ysis
  • f
the model by abstra ct interpret a t ion
  • W
e can compute U
  • U
  • U
n
  • U
def
  • p
  • stt
  • I
  • p
re t
  • F
b y abstract interp retation
  • The
abstract interp retation can b e done in pa rallel with the mo del checking at almost no supplementa ry cost
  • The
abstract interp retation results a re used
  • n
the y fo r U n as they b ecome available to restrict the state sea rch space
  • Several
restriction
  • p
erato rs have b een p rop
  • sed
fo r symb
  • lic
mo del checking with BDDs
  • convex
p
  • lyhedra
  • Halb
w achs N and Ra ymond P
  • On
the use
  • f
app ro ximations in symb
  • lic
mo del checking T ech rep SPECTRE L jan
  • verima
g lab
  • rato
ry
  • Grenoble
F rance Upper appr
  • xima
tio n D
  • f
post t
  • I
  • lfp
  • X
  • I
  • post
t X by abstra ct interpret a ti
  • n
  • Consider
an abstract domain hL vi app ro ximating sets
  • f
states hS
  • i
  • dene
a co rresp
  • ndence
hS
  • i
  • hL
vi which is a Galois connection P
  • S
  • Q
  • L
  • P
  • v
Q
  • P
  • Q
  • The
abstract value P
  • is
the app ro ximation
  • f
P
  • S
  • P
  • P
  • Dene
an abstract p
  • stimage
transfo rmer F
  • L
  • m
  • L
Q
  • L
  • X
  • I
  • p
  • st
t X
  • Q
v F Q
  • Dene
a widening
  • p
erato r
  • L
  • L
  • L
  • it
is an upp er app ro ximation
  • it
enfo rces nite convergence
  • f
F up w a rd iterates
  • The
up w a rd fo rw a rd iteration sequence with widening
  • F
  • def
  • F
i def
  • F
i if F
  • F
i
  • v
  • F
i
  • F
i def
  • F
i
  • F
  • F
i
  • therwise
is ultimately stationa ry its limit
  • F
is a sound upp er app ro ximation
  • f
p
  • stt
  • I
in that p
  • st
t
  • I
  • lfp
v F
  • F
  • Cousot
P
  • and
Cousot R Abstract interp retation a unied lattice mo del fo r static analysis
  • f
p rograms b y construction
  • r
app ro ximation
  • f
xp
  • ints
th POPL Los Angeles
  • pp
  • x
y
  • L
  • x
v x
  • y
and x y
  • L
  • y
v x
  • y
  • fo
r all increasing chains x
  • v
x
  • v
  • v
x i v
  • the
increasing chain dened b y y
  • x
  • y
i
  • y
i
  • x
i
  • is
not strictly increasing P Cousot
  • R
Cousot
  • AAS
Jan
slide-4
SLIDE 4
  • Dene
a na rro wing
  • p
erato r
  • L
  • L
  • L
such that
  • it
is an upp er app ro ximation
  • it
enfo rces nite convergence
  • f
F do wnw a rd iterates
  • the
do wnw a rd fo rw a rd iteration sequence with na rro wing
  • F
  • def
  • F
  • F
i def
  • F
i if F
  • F
i
  • F
i
  • F
i def
  • F
i
  • F
  • F
i
  • therwise
is ultimately stationa ry its limit
  • F
is a b etter sound upp er app ro ximation p
  • stt
  • I
in that p
  • stt
  • I
  • lfp
v F
  • F
  • F
  • Abstra
ct interpret a t io n design
  • The
design
  • f
  • the
abstract algeb ra hL v
  • t
u
  • f
  • f
n i
  • the
transfo rmer F usually comp
  • sed
  • ut
  • f
the p rimitives f
  • f
n
  • a
re p roblem dep endent
  • Natural
choices in the mo delchecking context a re
  • BDDs
discrete systems
  • Convex
p
  • lyhedra
hyb rid systems fo r which widening
  • p
erato rs have b een dened
  • x
y
  • L
  • x
v y
  • x
v x
  • y
v y
  • F
  • r
all decreasing chains x
  • w
x
  • w
  • the
decreasing chain dened b y y
  • x
  • y
i
  • y
i
  • x
i
  • is
not strictly decreasing
  • Maub
  • rgne
L Abstract interp retation using TDGs In SAS
  • sep
  • LNCS
  • pp
  • Cousot
P
  • and
Halb w achs N Automatic discovery
  • f
linea r restraints among va riables
  • f
a p rogram In th POPL T ucson
  • pp
  • Upper
appr
  • xima
ti
  • n
A
  • f
pre t
  • F
  • lfp
  • X
  • F
  • pret
X by abstra ct interpret a t ion
  • Use
the same abstract algeb ra hL v
  • t
u
  • f
  • f
n i
  • Dene
an abstract p reimage transfo rmer F
  • L
  • m
  • L
Q
  • L
  • X
  • F
  • p
re t X
  • Q
v B Q
  • First
use an up w a rd backw a rd iteration sequence with widening nitely converging to
  • B
  • Imp
rove b y a do wnw a rd iteration sequence with na rro wing nitely con verging to
  • B
such that p re t
  • F
  • lfp
  • X
  • F
  • p
re t X
  • lfp
v B
  • B
  • B
  • Sequence
  • f
upper appr
  • xima
t ion s U
  • U
  • U
n
  • f
U
  • post
t
  • I
  • pret
  • F
by abstra ct interpret a ti
  • n
  • U
  • S
  • all
states
  • U
  • is
the
  • concretization
  • f
the limit
  • f
the up w a rd fo rw a rd iteration sequence with widening fo r F
  • U
  • is
the
  • concretization
  • f
the limit
  • f
the co rresp
  • nding
do wnw a rd fo rw a rd iteration sequence with na rro wing fo r F sta rting from U
  • Cousot
P
  • and
Cousot R Systematic design
  • f
p rogram analysis framew
  • rks
In th POPL San Antonio
  • pp
  • Cousot
P
  • M
  • etho
des it
  • eratives
de construction et dapp ro ximation de p
  • ints
xes dop
  • erateurs
monotones sur un treillis analyse s
  • emantique
de p rogrammes Ph D thesis Universit
  • e
scientique et m
  • edicale
de Grenoble
  • Cousot
P
  • and
Cousot R Abstract interp retation and application to logic p rograms J Logic Prog
  • The
edito r
  • f
JLP has mistak enly published the unreadable galley p ro
  • f
F
  • r
a co rrect version
  • f
this pap er see httpwwwensfrcousot P Cousot
  • R
Cousot
  • AAS
Jan
slide-5
SLIDE 5
  • U
n is the
  • concretization
  • f
the limit
  • f
the up w a rd backw a rd iteration sequence with widening fo r X
  • U
n u B X
  • U
n is the
  • concretization
  • f
the limit
  • f
the co rresp
  • nding
do wn w a rd backw a rd iteration sequence with na rro wing fo r X
  • U
n u B X
  • sta
rting from U n
  • U
n is the
  • concretization
  • f
the limit
  • f
the up w a rd fo rw a rd iter ation sequence with widening fo r X
  • U
n u F X
  • U
n is the
  • concretization
  • f
the limit
  • f
the co rresp
  • nding
do wn w a rd fo rw a rd iteration sequence with na rro wing fo r X
  • U
n u F X
  • sta
rting from U n
  • Correctness
  • The
sequence U
  • U
  • U
  • U
n
  • U
n
  • U
n
  • U
n
  • is
a descending chain
  • The
restriction is mo re and mo re p recise as the mo delchecking go es
  • n
  • All
elements U k is the sequence a re sound U k
  • p
  • stt
  • I
  • p
re t
  • F
  • Stop
the abstract interp retation computation with a na rro wing
  • r
when the pa rallel mo delchecking terminates Pr
  • blema
tic termina tio n
  • The
abstract interp retation alw a ys terminate
  • The
abstract interp retation is app ro ximate so the statespace restric tion ma y not b e nite
  • The
pa rallel combination
  • f
abstract interp retation and mo delchecking is incomplete since it ma y not terminate
  • In
case
  • f
nontermination the info rmation gathered b y abstract inter p retation is reusable fo r verication b y
  • abstract
symb
  • lic
metho ds
  • mo
del abstraction which a re also incomplete but gua rantee termination
  • Conclusion
  • W
e have p rop
  • sed
a metho d fo r the pa rallel combination
  • f
mo del analysis b y abstract interp retation and verication b y mo delchecking where the verication
  • mak
es no app ro ximation
  • n
states and transitions
  • explo
res an hop efully nite subgraph
  • Semialgo
rithm since there is no gua rantee that the explo red subgraph will b e nite
  • classical
mo delchecking w
  • uld
have failed anyw a y
  • case
b y case exp erimentation is needed
  • The
metho d should b e used b efo re reso rting to mo delchecking
  • f
a mo re abstract mo del the info rmation gathered ab
  • ut
the exact mo del b eing reusable P Cousot
  • R
Cousot
  • AAS
Jan