1 //===-- llvm/CodeGen/DIEHash.cpp - Dwarf Hashing Framework ----------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file contains support for DWARF4 hashing of DIEs.
12 //===----------------------------------------------------------------------===//
14 #include "ByteStreamer.h"
16 #include "DwarfDebug.h"
17 #include "llvm/ADT/ArrayRef.h"
18 #include "llvm/ADT/StringRef.h"
19 #include "llvm/CodeGen/AsmPrinter.h"
20 #include "llvm/CodeGen/DIE.h"
21 #include "llvm/Support/Debug.h"
22 #include "llvm/Support/Dwarf.h"
23 #include "llvm/Support/Endian.h"
24 #include "llvm/Support/MD5.h"
25 #include "llvm/Support/raw_ostream.h"
29 #define DEBUG_TYPE "dwarfdebug"
31 /// \brief Grabs the string in whichever attribute is passed in and returns
32 /// a reference to it.
33 static StringRef getDIEStringAttr(const DIE &Die, uint16_t Attr) {
34 // Iterate through all the attributes until we find the one we're
35 // looking for, if we can't find it return an empty string.
36 for (const auto &V : Die.values())
37 if (V.getAttribute() == Attr)
38 return V.getDIEString().getString();
43 /// \brief Adds the string in \p Str to the hash. This also hashes
44 /// a trailing NULL with the string.
45 void DIEHash::addString(StringRef Str) {
46 DEBUG(dbgs() << "Adding string " << Str << " to hash.\n");
48 Hash.update(makeArrayRef((uint8_t)'\0'));
51 // FIXME: The LEB128 routines are copied and only slightly modified out of
54 /// \brief Adds the unsigned in \p Value to the hash encoded as a ULEB128.
55 void DIEHash::addULEB128(uint64_t Value) {
56 DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n");
58 uint8_t Byte = Value & 0x7f;
61 Byte |= 0x80; // Mark this byte to show that more bytes will follow.
66 void DIEHash::addSLEB128(int64_t Value) {
67 DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n");
70 uint8_t Byte = Value & 0x7f;
72 More = !((((Value == 0) && ((Byte & 0x40) == 0)) ||
73 ((Value == -1) && ((Byte & 0x40) != 0))));
75 Byte |= 0x80; // Mark this byte to show that more bytes will follow.
80 /// \brief Including \p Parent adds the context of Parent to the hash..
81 void DIEHash::addParentContext(const DIE &Parent) {
83 DEBUG(dbgs() << "Adding parent context to hash...\n");
85 // [7.27.2] For each surrounding type or namespace beginning with the
86 // outermost such construct...
87 SmallVector<const DIE *, 1> Parents;
88 const DIE *Cur = &Parent;
89 while (Cur->getParent()) {
90 Parents.push_back(Cur);
91 Cur = Cur->getParent();
93 assert(Cur->getTag() == dwarf::DW_TAG_compile_unit ||
94 Cur->getTag() == dwarf::DW_TAG_type_unit);
96 // Reverse iterate over our list to go from the outermost construct to the
98 for (SmallVectorImpl<const DIE *>::reverse_iterator I = Parents.rbegin(),
101 const DIE &Die = **I;
103 // ... Append the letter "C" to the sequence...
106 // ... Followed by the DWARF tag of the construct...
107 addULEB128(Die.getTag());
109 // ... Then the name, taken from the DW_AT_name attribute.
110 StringRef Name = getDIEStringAttr(Die, dwarf::DW_AT_name);
111 DEBUG(dbgs() << "... adding context: " << Name << "\n");
117 // Collect all of the attributes for a particular DIE in single structure.
118 void DIEHash::collectAttributes(const DIE &Die, DIEAttrs &Attrs) {
120 for (const auto &V : Die.values()) {
121 DEBUG(dbgs() << "Attribute: "
122 << dwarf::AttributeString(V.getAttribute())
124 switch (V.getAttribute()) {
125 #define HANDLE_DIE_HASH_ATTR(NAME) \
129 #include "DIEHashAttributes.def"
136 void DIEHash::hashShallowTypeReference(dwarf::Attribute Attribute,
137 const DIE &Entry, StringRef Name) {
138 // append the letter 'N'
141 // the DWARF attribute code (DW_AT_type or DW_AT_friend),
142 addULEB128(Attribute);
144 // the context of the tag,
145 if (const DIE *Parent = Entry.getParent())
146 addParentContext(*Parent);
151 // and the name of the type.
154 // Currently DW_TAG_friends are not used by Clang, but if they do become so,
155 // here's the relevant spec text to implement:
157 // For DW_TAG_friend, if the referenced entry is the DW_TAG_subprogram,
158 // the context is omitted and the name to be used is the ABI-specific name
159 // of the subprogram (e.g., the mangled linker name).
162 void DIEHash::hashRepeatedTypeReference(dwarf::Attribute Attribute,
163 unsigned DieNumber) {
164 // a) If T is in the list of [previously hashed types], use the letter
168 addULEB128(Attribute);
170 // and use the unsigned LEB128 encoding of [the index of T in the
171 // list] as the attribute value;
172 addULEB128(DieNumber);
175 void DIEHash::hashDIEEntry(dwarf::Attribute Attribute, dwarf::Tag Tag,
177 assert(Tag != dwarf::DW_TAG_friend && "No current LLVM clients emit friend "
178 "tags. Add support here when there's "
181 // If the tag in Step 3 is one of [the below tags]
182 if ((Tag == dwarf::DW_TAG_pointer_type ||
183 Tag == dwarf::DW_TAG_reference_type ||
184 Tag == dwarf::DW_TAG_rvalue_reference_type ||
185 Tag == dwarf::DW_TAG_ptr_to_member_type) &&
186 // and the referenced type (via the [below attributes])
187 // FIXME: This seems overly restrictive, and causes hash mismatches
188 // there's a decl/def difference in the containing type of a
189 // ptr_to_member_type, but it's what DWARF says, for some reason.
190 Attribute == dwarf::DW_AT_type) {
191 // ... has a DW_AT_name attribute,
192 StringRef Name = getDIEStringAttr(Entry, dwarf::DW_AT_name);
194 hashShallowTypeReference(Attribute, Entry, Name);
199 unsigned &DieNumber = Numbering[&Entry];
201 hashRepeatedTypeReference(Attribute, DieNumber);
205 // otherwise, b) use the letter 'T' as the marker, ...
208 addULEB128(Attribute);
210 // ... process the type T recursively by performing Steps 2 through 7, and
211 // use the result as the attribute value.
212 DieNumber = Numbering.size();
216 // Hash all of the values in a block like set of values. This assumes that
217 // all of the data is going to be added as integers.
218 void DIEHash::hashBlockData(const DIE::const_value_range &Values) {
219 for (const auto &V : Values)
220 Hash.update((uint64_t)V.getDIEInteger().getValue());
223 // Hash the contents of a loclistptr class.
224 void DIEHash::hashLocList(const DIELocList &LocList) {
225 HashingByteStreamer Streamer(*this);
226 DwarfDebug &DD = *AP->getDwarfDebug();
227 const DebugLocStream &Locs = DD.getDebugLocs();
228 for (const auto &Entry : Locs.getEntries(Locs.getList(LocList.getValue())))
229 DD.emitDebugLocEntry(Streamer, Entry);
232 // Hash an individual attribute \param Attr based on the type of attribute and
234 void DIEHash::hashAttribute(const DIEValue &Value, dwarf::Tag Tag) {
235 dwarf::Attribute Attribute = Value.getAttribute();
237 // Other attribute values use the letter 'A' as the marker, and the value
238 // consists of the form code (encoded as an unsigned LEB128 value) followed by
239 // the encoding of the value according to the form code. To ensure
240 // reproducibility of the signature, the set of forms used in the signature
241 // computation is limited to the following: DW_FORM_sdata, DW_FORM_flag,
242 // DW_FORM_string, and DW_FORM_block.
244 switch (Value.getType()) {
245 case DIEValue::isNone:
246 llvm_unreachable("Expected valid DIEValue");
249 // ... An attribute that refers to another type entry T is processed as
251 case DIEValue::isEntry:
252 hashDIEEntry(Attribute, Tag, Value.getDIEEntry().getEntry());
254 case DIEValue::isInteger: {
256 addULEB128(Attribute);
257 switch (Value.getForm()) {
258 case dwarf::DW_FORM_data1:
259 case dwarf::DW_FORM_data2:
260 case dwarf::DW_FORM_data4:
261 case dwarf::DW_FORM_data8:
262 case dwarf::DW_FORM_udata:
263 case dwarf::DW_FORM_sdata:
264 addULEB128(dwarf::DW_FORM_sdata);
265 addSLEB128((int64_t)Value.getDIEInteger().getValue());
267 // DW_FORM_flag_present is just flag with a value of one. We still give it a
268 // value so just use the value.
269 case dwarf::DW_FORM_flag_present:
270 case dwarf::DW_FORM_flag:
271 addULEB128(dwarf::DW_FORM_flag);
272 addULEB128((int64_t)Value.getDIEInteger().getValue());
275 llvm_unreachable("Unknown integer form!");
279 case DIEValue::isString:
281 addULEB128(Attribute);
282 addULEB128(dwarf::DW_FORM_string);
283 addString(Value.getDIEString().getString());
285 case DIEValue::isInlineString:
287 addULEB128(Attribute);
288 addULEB128(dwarf::DW_FORM_string);
289 addString(Value.getDIEInlineString().getString());
291 case DIEValue::isBlock:
292 case DIEValue::isLoc:
293 case DIEValue::isLocList:
295 addULEB128(Attribute);
296 addULEB128(dwarf::DW_FORM_block);
297 if (Value.getType() == DIEValue::isBlock) {
298 addULEB128(Value.getDIEBlock().ComputeSize(AP));
299 hashBlockData(Value.getDIEBlock().values());
300 } else if (Value.getType() == DIEValue::isLoc) {
301 addULEB128(Value.getDIELoc().ComputeSize(AP));
302 hashBlockData(Value.getDIELoc().values());
304 // We could add the block length, but that would take
305 // a bit of work and not add a lot of uniqueness
306 // to the hash in some way we could test.
307 hashLocList(Value.getDIELocList());
310 // FIXME: It's uncertain whether or not we should handle this at the moment.
311 case DIEValue::isExpr:
312 case DIEValue::isLabel:
313 case DIEValue::isDelta:
314 llvm_unreachable("Add support for additional value types.");
318 // Go through the attributes from \param Attrs in the order specified in 7.27.4
320 void DIEHash::hashAttributes(const DIEAttrs &Attrs, dwarf::Tag Tag) {
321 #define HANDLE_DIE_HASH_ATTR(NAME) \
324 hashAttribute(Attrs.NAME, Tag); \
326 #include "DIEHashAttributes.def"
327 // FIXME: Add the extended attributes.
330 // Add all of the attributes for \param Die to the hash.
331 void DIEHash::addAttributes(const DIE &Die) {
333 collectAttributes(Die, Attrs);
334 hashAttributes(Attrs, Die.getTag());
337 void DIEHash::hashNestedType(const DIE &Die, StringRef Name) {
339 // ... append the letter 'S',
343 addULEB128(Die.getTag());
349 // Compute the hash of a DIE. This is based on the type signature computation
350 // given in section 7.27 of the DWARF4 standard. It is the md5 hash of a
351 // flattened description of the DIE.
352 void DIEHash::computeHash(const DIE &Die) {
353 // Append the letter 'D', followed by the DWARF tag of the DIE.
355 addULEB128(Die.getTag());
357 // Add each of the attributes of the DIE.
360 // Then hash each of the children of the DIE.
361 for (auto &C : Die.children()) {
363 // If C is a nested type entry or a member function entry, ...
364 if (isType(C.getTag()) || C.getTag() == dwarf::DW_TAG_subprogram) {
365 StringRef Name = getDIEStringAttr(C, dwarf::DW_AT_name);
366 // ... and has a DW_AT_name attribute
368 hashNestedType(C, Name);
375 // Following the last (or if there are no children), append a zero byte.
376 Hash.update(makeArrayRef((uint8_t)'\0'));
379 /// This is based on the type signature computation given in section 7.27 of the
380 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE
381 /// with the inclusion of the full CU and all top level CU entities.
382 // TODO: Initialize the type chain at 0 instead of 1 for CU signatures.
383 uint64_t DIEHash::computeCUSignature(StringRef DWOName, const DIE &Die) {
387 if (!DWOName.empty())
388 Hash.update(DWOName);
392 // Now return the result.
393 MD5::MD5Result Result;
396 // ... take the least significant 8 bytes and return those. Our MD5
397 // implementation always returns its results in little endian, so we actually
398 // need the "high" word.
399 return Result.high();
402 /// This is based on the type signature computation given in section 7.27 of the
403 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE
404 /// with the inclusion of additional forms not specifically called out in the
406 uint64_t DIEHash::computeTypeSignature(const DIE &Die) {
410 if (const DIE *Parent = Die.getParent())
411 addParentContext(*Parent);
416 // Now return the result.
417 MD5::MD5Result Result;
420 // ... take the least significant 8 bytes and return those. Our MD5
421 // implementation always returns its results in little endian, so we actually
422 // need the "high" word.
423 return Result.high();