[Heikki Kultala] This patch contains the ABI changes for the TCE target.
[clang.git] / docs / AnalyzerRegions.html
blob35708d57c9701f189e622ff9956d602f38139823
1 <html>
2 <head>
3 <title>Static Analyzer Design Document: Memory Regions</title>
4 </head>
5 <body>
7 <h1>Static Analyzer Design Document: Memory Regions</h1>
9 <h3>Authors</h3>
11 <p>Ted Kremenek, <tt>kremenek at apple</tt><br>
12 Zhongxing Xu, <tt>xuzhongzhing at gmail</tt></p>
14 <h2 id="intro">Introduction</h2>
16 <p>The path-sensitive analysis engine in libAnalysis employs an extensible API
17 for abstractly modeling the memory of an analyzed program. This API employs the
18 concept of "memory regions" to abstractly model chunks of program memory such as
19 program variables and dynamically allocated memory such as those returned from
20 'malloc' and 'alloca'. Regions are hierarchical, with subregions modeling
21 subtyping relationships, field and array offsets into larger chunks of memory,
22 and so on.</p>
24 <p>The region API consists of two components:</p>
26 <ul> <li>A taxonomy and representation of regions themselves within the analyzer
27 engine. The primary definitions and interfaces are described in <tt><a
28 href="http://clang.llvm.org/doxygen/MemRegion_8h-source.html">MemRegion.h</a></tt>.
29 At the root of the region hierarchy is the class <tt>MemRegion</tt> with
30 specific subclasses refining the region concept for variables, heap allocated
31 memory, and so forth.</li> <li>The modeling of binding of values to regions. For
32 example, modeling the value stored to a local variable <tt>x</tt> consists of
33 recording the binding between the region for <tt>x</tt> (which represents the
34 raw memory associated with <tt>x</tt>) and the value stored to <tt>x</tt>. This
35 binding relationship is captured with the notion of &quot;symbolic
36 stores.&quot;</li> </ul>
38 <p>Symbolic stores, which can be thought of as representing the relation
39 <tt>regions -> values</tt>, are implemented by subclasses of the
40 <tt>StoreManager</tt> class (<tt><a
41 href="http://clang.llvm.org/doxygen/Store_8h-source.html">Store.h</a></tt>). A
42 particular StoreManager implementation has complete flexibility concerning the
43 following:
45 <ul>
46 <li><em>How</em> to model the binding between regions and values</li>
47 <li><em>What</em> bindings are recorded
48 </ul>
50 <p>Together, both points allow different StoreManagers to tradeoff between
51 different levels of analysis precision and scalability concerning the reasoning
52 of program memory. Meanwhile, the core path-sensitive engine makes no
53 assumptions about either points, and queries a StoreManager about the bindings
54 to a memory region through a generic interface that all StoreManagers share. If
55 a particular StoreManager cannot reason about the potential bindings of a given
56 memory region (e.g., '<tt>BasicStoreManager</tt>' does not reason about fields
57 of structures) then the StoreManager can simply return 'unknown' (represented by
58 '<tt>UnknownVal</tt>') for a particular region-binding. This separation of
59 concerns not only isolates the core analysis engine from the details of
60 reasoning about program memory but also facilities the option of a client of the
61 path-sensitive engine to easily swap in different StoreManager implementations
62 that internally reason about program memory in very different ways.</pp>
64 <p>The rest of this document is divided into two parts. We first discuss region
65 taxonomy and the semantics of regions. We then discuss the StoreManager
66 interface, and details of how the currently available StoreManager classes
67 implement region bindings.</p>
69 <h2 id="regions">Memory Regions and Region Taxonomy</h2>
71 <h3>Pointers</h3>
73 <p>Before talking about the memory regions, we would talk about the pointers
74 since memory regions are essentially used to represent pointer values.</p>
76 <p>The pointer is a type of values. Pointer values have two semantic aspects.
77 One is its physical value, which is an address or location. The other is the
78 type of the memory object residing in the address.</p>
80 <p>Memory regions are designed to abstract these two properties of the pointer.
81 The physical value of a pointer is represented by MemRegion pointers. The rvalue
82 type of the region corresponds to the type of the pointee object.</p>
84 <p>One complication is that we could have different view regions on the same
85 memory chunk. They represent the same memory location, but have different
86 abstract location, i.e., MemRegion pointers. Thus we need to canonicalize the
87 abstract locations to get a unique abstract location for one physical
88 location.</p>
90 <p>Furthermore, these different view regions may or may not represent memory
91 objects of different types. Some different types are semantically the same,
92 for example, 'struct s' and 'my_type' are the same type.</p>
94 <pre>
95 struct s;
96 typedef struct s my_type;
97 </pre>
99 <p>But <tt>char</tt> and <tt>int</tt> are not the same type in the code below:</p>
101 <pre>
102 void *p;
103 int *q = (int*) p;
104 char *r = (char*) p;
105 </pre
107 <p>Thus we need to canonicalize the MemRegion which is used in binding and
108 retrieving.</p>
110 <h3>Regions</h3>
111 <p>Region is the entity used to model pointer values. A Region has the following
112 properties:</p>
114 <ul>
115 <li>Kind</li>
117 <li>ObjectType: the type of the object residing on the region.</li>
119 <li>LocationType: the type of the pointer value that the region corresponds to.
120 Usually this is the pointer to the ObjectType. But sometimes we want to cache
121 this type explicitly, for example, for a CodeTextRegion.</li>
123 <li>StartLocation</li>
125 <li>EndLocation</li>
126 </ul>
128 <h3>Symbolic Regions</h3>
130 <p>A symbolic region is a map of the concept of symbolic values into the domain
131 of regions. It is the way that we represent symbolic pointers. Whenever a
132 symbolic pointer value is needed, a symbolic region is created to represent
133 it.</p>
135 <p>A symbolic region has no type. It wraps a SymbolData. But sometimes we have
136 type information associated with a symbolic region. For this case, a
137 TypedViewRegion is created to layer the type information on top of the symbolic
138 region. The reason we do not carry type information with the symbolic region is
139 that the symbolic regions can have no type. To be consistent, we don't let them
140 to carry type information.</p>
142 <p>Like a symbolic pointer, a symbolic region may be NULL, has unknown extent,
143 and represents a generic chunk of memory.</p>
145 <p><em><b>NOTE</b>: We plan not to use loc::SymbolVal in RegionStore and remove it
146 gradually.</em></p>
148 <p>Symbolic regions get their rvalue types through the following ways:</p>
150 <ul>
151 <li>Through the parameter or global variable that points to it, e.g.:
152 <pre>
153 void f(struct s* p) {
156 </pre>
158 <p>The symbolic region pointed to by <tt>p</tt> has type <tt>struct
159 s</tt>.</p></li>
161 <li>Through explicit or implicit casts, e.g.:
162 <pre>
163 void f(void* p) {
164 struct s* q = (struct s*) p;
167 </pre>
168 </li>
169 </ul>
171 <p>We attach the type information to the symbolic region lazily. For the first
172 case above, we create the <tt>TypedViewRegion</tt> only when the pointer is
173 actually used to access the pointee memory object, that is when the element or
174 field region is created. For the cast case, the <tt>TypedViewRegion</tt> is
175 created when visiting the <tt>CastExpr</tt>.</p>
177 <p>The reason for doing lazy typing is that symbolic regions are sometimes only
178 used to do location comparison.</p>
180 <h3>Pointer Casts</h3>
182 <p>Pointer casts allow people to impose different 'views' onto a chunk of
183 memory.</p>
185 <p>Usually we have two kinds of casts. One kind of casts cast down with in the
186 type hierarchy. It imposes more specific views onto more generic memory regions.
187 The other kind of casts cast up with in the type hierarchy. It strips away more
188 specific views on top of the more generic memory regions.</p>
190 <p>We simulate the down casts by layering another <tt>TypedViewRegion</tt> on
191 top of the original region. We simulate the up casts by striping away the top
192 <tt>TypedViewRegion</tt>. Down casts is usually simple. For up casts, if the
193 there is no <tt>TypedViewRegion</tt> to be stripped, we return the original
194 region. If the underlying region is of the different type than the cast-to type,
195 we flag an error state.</p>
197 <p>For toll-free bridging casts, we return the original region.</p>
199 <p>We can set up a partial order for pointer types, with the most general type
200 <tt>void*</tt> at the top. The partial order forms a tree with <tt>void*</tt> as
201 its root node.</p>
203 <p>Every <tt>MemRegion</tt> has a root position in the type tree. For example,
204 the pointee region of <tt>void *p</tt> has its root position at the root node of
205 the tree. <tt>VarRegion</tt> of <tt>int x</tt> has its root position at the 'int
206 type' node.</p>
208 <p><tt>TypedViewRegion</tt> is used to move the region down or up in the tree.
209 Moving down in the tree adds a <tt>TypedViewRegion</tt>. Moving up in the tree
210 removes a <Tt>TypedViewRegion</tt>.</p>
212 <p>Do we want to allow moving up beyond the root position? This happens
213 when:</p> <pre> int x; void *p = &amp;x; </pre>
215 <p>The region of <tt>x</tt> has its root position at 'int*' node. the cast to
216 void* moves that region up to the 'void*' node. I propose to not allow such
217 casts, and assign the region of <tt>x</tt> for <tt>p</tt>.</p>
219 <p>Another non-ideal case is that people might cast to a non-generic pointer
220 from another non-generic pointer instead of first casting it back to the generic
221 pointer. Direct handling of this case would result in multiple layers of
222 TypedViewRegions. This enforces an incorrect semantic view to the region,
223 because we can only have one typed view on a region at a time. To avoid this
224 inconsistency, before casting the region, we strip the TypedViewRegion, then do
225 the cast. In summary, we only allow one layer of TypedViewRegion.</p>
227 <h3>Region Bindings</h3>
229 <p>The following region kinds are boundable: VarRegion, CompoundLiteralRegion,
230 StringRegion, ElementRegion, FieldRegion, and ObjCIvarRegion.</p>
232 <p>When binding regions, we perform canonicalization on element regions and field
233 regions. This is because we can have different views on the same region, some
234 of which are essentially the same view with different sugar type names.</p>
236 <p>To canonicalize a region, we get the canonical types for all TypedViewRegions
237 along the way up to the root region, and make new TypedViewRegions with those
238 canonical types.</p>
240 <p>For Objective-C and C++, perhaps another canonicalization rule should be
241 added: for FieldRegion, the least derived class that has the field is used as
242 the type of the super region of the FieldRegion.</p>
244 <p>All bindings and retrievings are done on the canonicalized regions.</p>
246 <p>Canonicalization is transparent outside the region store manager, and more
247 specifically, unaware outside the Bind() and Retrieve() method. We don't need to
248 consider region canonicalization when doing pointer cast.</p>
250 <h3>Constraint Manager</h3>
252 <p>The constraint manager reasons about the abstract location of memory objects.
253 We can have different views on a region, but none of these views changes the
254 location of that object. Thus we should get the same abstract location for those
255 regions.</p>
257 </body>
258 </html>