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 #define DEBUG_TYPE "dwarfdebug"
18 #include "llvm/ADT/ArrayRef.h"
19 #include "llvm/ADT/StringRef.h"
20 #include "llvm/Support/Debug.h"
21 #include "llvm/Support/Dwarf.h"
22 #include "llvm/Support/Endian.h"
23 #include "llvm/Support/MD5.h"
24 #include "llvm/Support/raw_ostream.h"
28 /// \brief Grabs the string in whichever attribute is passed in and returns
29 /// a reference to it.
30 static StringRef getDIEStringAttr(const DIE &Die, uint16_t Attr) {
31 const SmallVectorImpl<DIEValue *> &Values = Die.getValues();
32 const DIEAbbrev &Abbrevs = Die.getAbbrev();
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 (size_t i = 0; i < Values.size(); ++i) {
37 if (Abbrevs.getData()[i].getAttribute() == Attr) {
38 DIEValue *V = Values[i];
39 assert(isa<DIEString>(V) && "String requested. Not a string.");
40 DIEString *S = cast<DIEString>(V);
41 return S->getString();
47 /// \brief Adds the string in \p Str to the hash. This also hashes
48 /// a trailing NULL with the string.
49 void DIEHash::addString(StringRef Str) {
50 DEBUG(dbgs() << "Adding string " << Str << " to hash.\n");
52 Hash.update(makeArrayRef((uint8_t)'\0'));
55 // FIXME: The LEB128 routines are copied and only slightly modified out of
58 /// \brief Adds the unsigned in \p Value to the hash encoded as a ULEB128.
59 void DIEHash::addULEB128(uint64_t Value) {
60 DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n");
62 uint8_t Byte = Value & 0x7f;
65 Byte |= 0x80; // Mark this byte to show that more bytes will follow.
70 void DIEHash::addSLEB128(int64_t Value) {
71 DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n");
74 uint8_t Byte = Value & 0x7f;
76 More = !((((Value == 0 ) && ((Byte & 0x40) == 0)) ||
77 ((Value == -1) && ((Byte & 0x40) != 0))));
79 Byte |= 0x80; // Mark this byte to show that more bytes will follow.
84 /// \brief Including \p Parent adds the context of Parent to the hash..
85 void DIEHash::addParentContext(const DIE &Parent) {
87 DEBUG(dbgs() << "Adding parent context to hash...\n");
89 // [7.27.2] For each surrounding type or namespace beginning with the
90 // outermost such construct...
91 SmallVector<const DIE *, 1> Parents;
92 const DIE *Cur = &Parent;
93 while (Cur->getParent()) {
94 Parents.push_back(Cur);
95 Cur = Cur->getParent();
97 assert(Cur->getTag() == dwarf::DW_TAG_compile_unit ||
98 Cur->getTag() == dwarf::DW_TAG_type_unit);
100 // Reverse iterate over our list to go from the outermost construct to the
102 for (SmallVectorImpl<const DIE *>::reverse_iterator I = Parents.rbegin(),
105 const DIE &Die = **I;
107 // ... Append the letter "C" to the sequence...
110 // ... Followed by the DWARF tag of the construct...
111 addULEB128(Die.getTag());
113 // ... Then the name, taken from the DW_AT_name attribute.
114 StringRef Name = getDIEStringAttr(Die, dwarf::DW_AT_name);
115 DEBUG(dbgs() << "... adding context: " << Name << "\n");
121 // Collect all of the attributes for a particular DIE in single structure.
122 void DIEHash::collectAttributes(const DIE &Die, DIEAttrs &Attrs) {
123 const SmallVectorImpl<DIEValue *> &Values = Die.getValues();
124 const DIEAbbrev &Abbrevs = Die.getAbbrev();
126 #define COLLECT_ATTR(NAME) \
128 Attrs.NAME.Val = Values[i]; \
129 Attrs.NAME.Desc = &Abbrevs.getData()[i]; \
132 for (size_t i = 0, e = Values.size(); i != e; ++i) {
133 DEBUG(dbgs() << "Attribute: "
134 << dwarf::AttributeString(Abbrevs.getData()[i].getAttribute())
136 switch (Abbrevs.getData()[i].getAttribute()) {
137 COLLECT_ATTR(DW_AT_name);
138 COLLECT_ATTR(DW_AT_accessibility);
139 COLLECT_ATTR(DW_AT_address_class);
140 COLLECT_ATTR(DW_AT_allocated);
141 COLLECT_ATTR(DW_AT_artificial);
142 COLLECT_ATTR(DW_AT_associated);
143 COLLECT_ATTR(DW_AT_binary_scale);
144 COLLECT_ATTR(DW_AT_bit_offset);
145 COLLECT_ATTR(DW_AT_bit_size);
146 COLLECT_ATTR(DW_AT_bit_stride);
147 COLLECT_ATTR(DW_AT_byte_size);
148 COLLECT_ATTR(DW_AT_byte_stride);
149 COLLECT_ATTR(DW_AT_const_expr);
150 COLLECT_ATTR(DW_AT_const_value);
151 COLLECT_ATTR(DW_AT_containing_type);
152 COLLECT_ATTR(DW_AT_count);
153 COLLECT_ATTR(DW_AT_data_bit_offset);
154 COLLECT_ATTR(DW_AT_data_location);
155 COLLECT_ATTR(DW_AT_data_member_location);
156 COLLECT_ATTR(DW_AT_decimal_scale);
157 COLLECT_ATTR(DW_AT_decimal_sign);
158 COLLECT_ATTR(DW_AT_default_value);
159 COLLECT_ATTR(DW_AT_digit_count);
160 COLLECT_ATTR(DW_AT_discr);
161 COLLECT_ATTR(DW_AT_discr_list);
162 COLLECT_ATTR(DW_AT_discr_value);
163 COLLECT_ATTR(DW_AT_encoding);
164 COLLECT_ATTR(DW_AT_enum_class);
165 COLLECT_ATTR(DW_AT_endianity);
166 COLLECT_ATTR(DW_AT_explicit);
167 COLLECT_ATTR(DW_AT_is_optional);
168 COLLECT_ATTR(DW_AT_location);
169 COLLECT_ATTR(DW_AT_lower_bound);
170 COLLECT_ATTR(DW_AT_mutable);
171 COLLECT_ATTR(DW_AT_ordering);
172 COLLECT_ATTR(DW_AT_picture_string);
173 COLLECT_ATTR(DW_AT_prototyped);
174 COLLECT_ATTR(DW_AT_small);
175 COLLECT_ATTR(DW_AT_segment);
176 COLLECT_ATTR(DW_AT_string_length);
177 COLLECT_ATTR(DW_AT_threads_scaled);
178 COLLECT_ATTR(DW_AT_upper_bound);
179 COLLECT_ATTR(DW_AT_use_location);
180 COLLECT_ATTR(DW_AT_use_UTF8);
181 COLLECT_ATTR(DW_AT_variable_parameter);
182 COLLECT_ATTR(DW_AT_virtuality);
183 COLLECT_ATTR(DW_AT_visibility);
184 COLLECT_ATTR(DW_AT_vtable_elem_location);
185 COLLECT_ATTR(DW_AT_type);
192 void DIEHash::hashShallowTypeReference(dwarf::Attribute Attribute,
193 const DIE &Entry, StringRef Name) {
194 // append the letter 'N'
197 // the DWARF attribute code (DW_AT_type or DW_AT_friend),
198 addULEB128(Attribute);
200 // the context of the tag,
201 if (const DIE *Parent = Entry.getParent())
202 addParentContext(*Parent);
207 // and the name of the type.
210 // Currently DW_TAG_friends are not used by Clang, but if they do become so,
211 // here's the relevant spec text to implement:
213 // For DW_TAG_friend, if the referenced entry is the DW_TAG_subprogram,
214 // the context is omitted and the name to be used is the ABI-specific name
215 // of the subprogram (e.g., the mangled linker name).
218 void DIEHash::hashRepeatedTypeReference(dwarf::Attribute Attribute,
219 unsigned DieNumber) {
220 // a) If T is in the list of [previously hashed types], use the letter
224 addULEB128(Attribute);
226 // and use the unsigned LEB128 encoding of [the index of T in the
227 // list] as the attribute value;
228 addULEB128(DieNumber);
231 void DIEHash::hashDIEEntry(dwarf::Attribute Attribute, dwarf::Tag Tag,
233 assert(Tag != dwarf::DW_TAG_friend && "No current LLVM clients emit friend "
234 "tags. Add support here when there's "
237 // If the tag in Step 3 is one of [the below tags]
238 if ((Tag == dwarf::DW_TAG_pointer_type ||
239 Tag == dwarf::DW_TAG_reference_type ||
240 Tag == dwarf::DW_TAG_rvalue_reference_type ||
241 Tag == dwarf::DW_TAG_ptr_to_member_type) &&
242 // and the referenced type (via the [below attributes])
243 // FIXME: This seems overly restrictive, and causes hash mismatches
244 // there's a decl/def difference in the containing type of a
245 // ptr_to_member_type, but it's what DWARF says, for some reason.
246 Attribute == dwarf::DW_AT_type) {
247 // ... has a DW_AT_name attribute,
248 StringRef Name = getDIEStringAttr(Entry, dwarf::DW_AT_name);
250 hashShallowTypeReference(Attribute, Entry, Name);
255 unsigned &DieNumber = Numbering[&Entry];
257 hashRepeatedTypeReference(Attribute, DieNumber);
261 // otherwise, b) use the letter 'T' as a the marker, ...
264 addULEB128(Attribute);
266 // ... process the type T recursively by performing Steps 2 through 7, and
267 // use the result as the attribute value.
268 DieNumber = Numbering.size();
272 // Hash an individual attribute \param Attr based on the type of attribute and
274 void DIEHash::hashAttribute(AttrEntry Attr, dwarf::Tag Tag) {
275 const DIEValue *Value = Attr.Val;
276 const DIEAbbrevData *Desc = Attr.Desc;
277 dwarf::Attribute Attribute = Desc->getAttribute();
280 // ... An attribute that refers to another type entry T is processed as
282 if (const DIEEntry *EntryAttr = dyn_cast<DIEEntry>(Value)) {
283 hashDIEEntry(Attribute, Tag, *EntryAttr->getEntry());
287 // Other attribute values use the letter 'A' as the marker, and the value
288 // consists of the form code (encoded as an unsigned LEB128 value) followed by
289 // the encoding of the value according to the form code. To ensure
290 // reproducibility of the signature, the set of forms used in the signature
291 // computation is limited to the following: DW_FORM_sdata, DW_FORM_flag,
292 // DW_FORM_string, and DW_FORM_block.
293 switch (Desc->getForm()) {
294 case dwarf::DW_FORM_string:
296 "Add support for DW_FORM_string if we ever start emitting them again");
297 case dwarf::DW_FORM_GNU_str_index:
298 case dwarf::DW_FORM_strp:
300 addULEB128(Attribute);
301 addULEB128(dwarf::DW_FORM_string);
302 addString(cast<DIEString>(Value)->getString());
304 case dwarf::DW_FORM_data1:
305 case dwarf::DW_FORM_data2:
306 case dwarf::DW_FORM_data4:
307 case dwarf::DW_FORM_data8:
308 case dwarf::DW_FORM_udata:
310 addULEB128(Attribute);
311 addULEB128(dwarf::DW_FORM_sdata);
312 addSLEB128((int64_t)cast<DIEInteger>(Value)->getValue());
314 // DW_FORM_flag_present is just flag with a value of one. We still give it a
315 // value so just use the value.
316 case dwarf::DW_FORM_flag_present:
317 case dwarf::DW_FORM_flag:
319 addULEB128(Attribute);
320 addULEB128(dwarf::DW_FORM_flag);
321 addULEB128((int64_t)cast<DIEInteger>(Value)->getValue());
324 llvm_unreachable("Add support for additional forms");
328 // Go through the attributes from \param Attrs in the order specified in 7.27.4
330 void DIEHash::hashAttributes(const DIEAttrs &Attrs, dwarf::Tag Tag) {
331 #define ADD_ATTR(ATTR) \
334 hashAttribute(ATTR, Tag); \
337 ADD_ATTR(Attrs.DW_AT_name);
338 ADD_ATTR(Attrs.DW_AT_accessibility);
339 ADD_ATTR(Attrs.DW_AT_address_class);
340 ADD_ATTR(Attrs.DW_AT_allocated);
341 ADD_ATTR(Attrs.DW_AT_artificial);
342 ADD_ATTR(Attrs.DW_AT_associated);
343 ADD_ATTR(Attrs.DW_AT_binary_scale);
344 ADD_ATTR(Attrs.DW_AT_bit_offset);
345 ADD_ATTR(Attrs.DW_AT_bit_size);
346 ADD_ATTR(Attrs.DW_AT_bit_stride);
347 ADD_ATTR(Attrs.DW_AT_byte_size);
348 ADD_ATTR(Attrs.DW_AT_byte_stride);
349 ADD_ATTR(Attrs.DW_AT_const_expr);
350 ADD_ATTR(Attrs.DW_AT_const_value);
351 ADD_ATTR(Attrs.DW_AT_containing_type);
352 ADD_ATTR(Attrs.DW_AT_count);
353 ADD_ATTR(Attrs.DW_AT_data_bit_offset);
354 ADD_ATTR(Attrs.DW_AT_data_location);
355 ADD_ATTR(Attrs.DW_AT_data_member_location);
356 ADD_ATTR(Attrs.DW_AT_decimal_scale);
357 ADD_ATTR(Attrs.DW_AT_decimal_sign);
358 ADD_ATTR(Attrs.DW_AT_default_value);
359 ADD_ATTR(Attrs.DW_AT_digit_count);
360 ADD_ATTR(Attrs.DW_AT_discr);
361 ADD_ATTR(Attrs.DW_AT_discr_list);
362 ADD_ATTR(Attrs.DW_AT_discr_value);
363 ADD_ATTR(Attrs.DW_AT_encoding);
364 ADD_ATTR(Attrs.DW_AT_enum_class);
365 ADD_ATTR(Attrs.DW_AT_endianity);
366 ADD_ATTR(Attrs.DW_AT_explicit);
367 ADD_ATTR(Attrs.DW_AT_is_optional);
368 ADD_ATTR(Attrs.DW_AT_location);
369 ADD_ATTR(Attrs.DW_AT_lower_bound);
370 ADD_ATTR(Attrs.DW_AT_mutable);
371 ADD_ATTR(Attrs.DW_AT_ordering);
372 ADD_ATTR(Attrs.DW_AT_picture_string);
373 ADD_ATTR(Attrs.DW_AT_prototyped);
374 ADD_ATTR(Attrs.DW_AT_small);
375 ADD_ATTR(Attrs.DW_AT_segment);
376 ADD_ATTR(Attrs.DW_AT_string_length);
377 ADD_ATTR(Attrs.DW_AT_threads_scaled);
378 ADD_ATTR(Attrs.DW_AT_upper_bound);
379 ADD_ATTR(Attrs.DW_AT_use_location);
380 ADD_ATTR(Attrs.DW_AT_use_UTF8);
381 ADD_ATTR(Attrs.DW_AT_variable_parameter);
382 ADD_ATTR(Attrs.DW_AT_virtuality);
383 ADD_ATTR(Attrs.DW_AT_visibility);
384 ADD_ATTR(Attrs.DW_AT_vtable_elem_location);
385 ADD_ATTR(Attrs.DW_AT_type);
387 // FIXME: Add the extended attributes.
390 // Add all of the attributes for \param Die to the hash.
391 void DIEHash::addAttributes(const DIE &Die) {
393 collectAttributes(Die, Attrs);
394 hashAttributes(Attrs, Die.getTag());
397 void DIEHash::hashNestedType(const DIE &Die, StringRef Name) {
399 // ... append the letter 'S',
403 addULEB128(Die.getTag());
409 // Compute the hash of a DIE. This is based on the type signature computation
410 // given in section 7.27 of the DWARF4 standard. It is the md5 hash of a
411 // flattened description of the DIE.
412 void DIEHash::computeHash(const DIE &Die) {
413 // Append the letter 'D', followed by the DWARF tag of the DIE.
415 addULEB128(Die.getTag());
417 // Add each of the attributes of the DIE.
420 // Then hash each of the children of the DIE.
421 for (std::vector<DIE *>::const_iterator I = Die.getChildren().begin(),
422 E = Die.getChildren().end();
425 // If C is a nested type entry or a member function entry, ...
426 if (isType((*I)->getTag()) || (*I)->getTag() == dwarf::DW_TAG_subprogram) {
427 StringRef Name = getDIEStringAttr(**I, dwarf::DW_AT_name);
428 // ... and has a DW_AT_name attribute
430 hashNestedType(**I, Name);
437 // Following the last (or if there are no children), append a zero byte.
438 Hash.update(makeArrayRef((uint8_t)'\0'));
441 /// This is based on the type signature computation given in section 7.27 of the
442 /// DWARF4 standard. It is the md5 hash of a flattened description of the DIE
443 /// with the exception that we are hashing only the context and the name of the
445 uint64_t DIEHash::computeDIEODRSignature(const DIE &Die) {
447 // Add the contexts to the hash. We won't be computing the ODR hash for
448 // function local types so it's safe to use the generic context hashing
450 // FIXME: If we figure out how to account for linkage in some way we could
451 // actually do this with a slight modification to the parent hash algorithm.
452 if (const DIE *Parent = Die.getParent())
453 addParentContext(*Parent);
455 // Add the current DIE information.
457 // Add the DWARF tag of the DIE.
458 addULEB128(Die.getTag());
460 // Add the name of the type to the hash.
461 addString(getDIEStringAttr(Die, dwarf::DW_AT_name));
463 // Now get the result.
464 MD5::MD5Result Result;
467 // ... take the least significant 8 bytes and return those. Our MD5
468 // implementation always returns its results in little endian, swap bytes
470 return *reinterpret_cast<support::ulittle64_t *>(Result + 8);
473 /// This is based on the type signature computation given in section 7.27 of the
474 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE
475 /// with the inclusion of the full CU and all top level CU entities.
476 // TODO: Initialize the type chain at 0 instead of 1 for CU signatures.
477 uint64_t DIEHash::computeCUSignature(const DIE &Die) {
484 // Now return the result.
485 MD5::MD5Result Result;
488 // ... take the least significant 8 bytes and return those. Our MD5
489 // implementation always returns its results in little endian, swap bytes
491 return *reinterpret_cast<support::ulittle64_t *>(Result + 8);
494 /// This is based on the type signature computation given in section 7.27 of the
495 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE
496 /// with the inclusion of additional forms not specifically called out in the
498 uint64_t DIEHash::computeTypeSignature(const DIE &Die) {
502 if (const DIE *Parent = Die.getParent())
503 addParentContext(*Parent);
508 // Now return the result.
509 MD5::MD5Result Result;
512 // ... take the least significant 8 bytes and return those. Our MD5
513 // implementation always returns its results in little endian, swap bytes
515 return *reinterpret_cast<support::ulittle64_t *>(Result + 8);