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"
19 //#include "DwarfCompileUnit.h"
20 #include "llvm/ADT/ArrayRef.h"
21 #include "llvm/ADT/StringRef.h"
22 #include "llvm/Support/Debug.h"
23 #include "llvm/Support/Dwarf.h"
24 #include "llvm/Support/Endian.h"
25 #include "llvm/Support/MD5.h"
26 #include "llvm/Support/raw_ostream.h"
30 /// \brief Grabs the string in whichever attribute is passed in and returns
31 /// a reference to it.
32 static StringRef getDIEStringAttr(const DIE &Die, uint16_t Attr) {
33 const SmallVectorImpl<DIEValue *> &Values = Die.getValues();
34 const DIEAbbrev &Abbrevs = Die.getAbbrev();
36 // Iterate through all the attributes until we find the one we're
37 // looking for, if we can't find it return an empty string.
38 for (size_t i = 0; i < Values.size(); ++i) {
39 if (Abbrevs.getData()[i].getAttribute() == Attr) {
40 DIEValue *V = Values[i];
41 assert(isa<DIEString>(V) && "String requested. Not a string.");
42 DIEString *S = cast<DIEString>(V);
43 return S->getString();
49 /// \brief Adds the string in \p Str to the hash. This also hashes
50 /// a trailing NULL with the string.
51 void DIEHash::addString(StringRef Str) {
52 DEBUG(dbgs() << "Adding string " << Str << " to hash.\n");
54 Hash.update(makeArrayRef((uint8_t)'\0'));
57 // FIXME: The LEB128 routines are copied and only slightly modified out of
60 /// \brief Adds the unsigned in \p Value to the hash encoded as a ULEB128.
61 void DIEHash::addULEB128(uint64_t Value) {
62 DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n");
64 uint8_t Byte = Value & 0x7f;
67 Byte |= 0x80; // Mark this byte to show that more bytes will follow.
72 void DIEHash::addSLEB128(int64_t Value) {
73 DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n");
76 uint8_t Byte = Value & 0x7f;
78 More = !((((Value == 0 ) && ((Byte & 0x40) == 0)) ||
79 ((Value == -1) && ((Byte & 0x40) != 0))));
81 Byte |= 0x80; // Mark this byte to show that more bytes will follow.
86 /// \brief Including \p Parent adds the context of Parent to the hash..
87 void DIEHash::addParentContext(const DIE &Parent) {
89 DEBUG(dbgs() << "Adding parent context to hash...\n");
91 // [7.27.2] For each surrounding type or namespace beginning with the
92 // outermost such construct...
93 SmallVector<const DIE *, 1> Parents;
94 const DIE *Cur = &Parent;
95 while (Cur->getParent()) {
96 Parents.push_back(Cur);
97 Cur = Cur->getParent();
99 assert(Cur->getTag() == dwarf::DW_TAG_compile_unit ||
100 Cur->getTag() == dwarf::DW_TAG_type_unit);
102 // Reverse iterate over our list to go from the outermost construct to the
104 for (SmallVectorImpl<const DIE *>::reverse_iterator I = Parents.rbegin(),
107 const DIE &Die = **I;
109 // ... Append the letter "C" to the sequence...
112 // ... Followed by the DWARF tag of the construct...
113 addULEB128(Die.getTag());
115 // ... Then the name, taken from the DW_AT_name attribute.
116 StringRef Name = getDIEStringAttr(Die, dwarf::DW_AT_name);
117 DEBUG(dbgs() << "... adding context: " << Name << "\n");
123 // Collect all of the attributes for a particular DIE in single structure.
124 void DIEHash::collectAttributes(const DIE &Die, DIEAttrs &Attrs) {
125 const SmallVectorImpl<DIEValue *> &Values = Die.getValues();
126 const DIEAbbrev &Abbrevs = Die.getAbbrev();
128 #define COLLECT_ATTR(NAME) \
130 Attrs.NAME.Val = Values[i]; \
131 Attrs.NAME.Desc = &Abbrevs.getData()[i]; \
134 for (size_t i = 0, e = Values.size(); i != e; ++i) {
135 DEBUG(dbgs() << "Attribute: "
136 << dwarf::AttributeString(Abbrevs.getData()[i].getAttribute())
138 switch (Abbrevs.getData()[i].getAttribute()) {
139 COLLECT_ATTR(DW_AT_name);
140 COLLECT_ATTR(DW_AT_accessibility);
141 COLLECT_ATTR(DW_AT_address_class);
142 COLLECT_ATTR(DW_AT_allocated);
143 COLLECT_ATTR(DW_AT_artificial);
144 COLLECT_ATTR(DW_AT_associated);
145 COLLECT_ATTR(DW_AT_binary_scale);
146 COLLECT_ATTR(DW_AT_bit_offset);
147 COLLECT_ATTR(DW_AT_bit_size);
148 COLLECT_ATTR(DW_AT_bit_stride);
149 COLLECT_ATTR(DW_AT_byte_size);
150 COLLECT_ATTR(DW_AT_byte_stride);
151 COLLECT_ATTR(DW_AT_const_expr);
152 COLLECT_ATTR(DW_AT_const_value);
153 COLLECT_ATTR(DW_AT_containing_type);
154 COLLECT_ATTR(DW_AT_count);
155 COLLECT_ATTR(DW_AT_data_bit_offset);
156 COLLECT_ATTR(DW_AT_data_location);
157 COLLECT_ATTR(DW_AT_data_member_location);
158 COLLECT_ATTR(DW_AT_decimal_scale);
159 COLLECT_ATTR(DW_AT_decimal_sign);
160 COLLECT_ATTR(DW_AT_default_value);
161 COLLECT_ATTR(DW_AT_digit_count);
162 COLLECT_ATTR(DW_AT_discr);
163 COLLECT_ATTR(DW_AT_discr_list);
164 COLLECT_ATTR(DW_AT_discr_value);
165 COLLECT_ATTR(DW_AT_encoding);
166 COLLECT_ATTR(DW_AT_enum_class);
167 COLLECT_ATTR(DW_AT_endianity);
168 COLLECT_ATTR(DW_AT_explicit);
169 COLLECT_ATTR(DW_AT_is_optional);
170 COLLECT_ATTR(DW_AT_location);
171 COLLECT_ATTR(DW_AT_lower_bound);
172 COLLECT_ATTR(DW_AT_mutable);
173 COLLECT_ATTR(DW_AT_ordering);
174 COLLECT_ATTR(DW_AT_picture_string);
175 COLLECT_ATTR(DW_AT_prototyped);
176 COLLECT_ATTR(DW_AT_small);
177 COLLECT_ATTR(DW_AT_segment);
178 COLLECT_ATTR(DW_AT_string_length);
179 COLLECT_ATTR(DW_AT_threads_scaled);
180 COLLECT_ATTR(DW_AT_upper_bound);
181 COLLECT_ATTR(DW_AT_use_location);
182 COLLECT_ATTR(DW_AT_use_UTF8);
183 COLLECT_ATTR(DW_AT_variable_parameter);
184 COLLECT_ATTR(DW_AT_virtuality);
185 COLLECT_ATTR(DW_AT_visibility);
186 COLLECT_ATTR(DW_AT_vtable_elem_location);
187 COLLECT_ATTR(DW_AT_type);
194 void DIEHash::hashShallowTypeReference(dwarf::Attribute Attribute,
195 const DIE &Entry, StringRef Name) {
196 // append the letter 'N'
199 // the DWARF attribute code (DW_AT_type or DW_AT_friend),
200 addULEB128(Attribute);
202 // the context of the tag,
203 if (const DIE *Parent = Entry.getParent())
204 addParentContext(*Parent);
209 // and the name of the type.
212 // Currently DW_TAG_friends are not used by Clang, but if they do become so,
213 // here's the relevant spec text to implement:
215 // For DW_TAG_friend, if the referenced entry is the DW_TAG_subprogram,
216 // the context is omitted and the name to be used is the ABI-specific name
217 // of the subprogram (e.g., the mangled linker name).
220 void DIEHash::hashRepeatedTypeReference(dwarf::Attribute Attribute,
221 unsigned DieNumber) {
222 // a) If T is in the list of [previously hashed types], use the letter
226 addULEB128(Attribute);
228 // and use the unsigned LEB128 encoding of [the index of T in the
229 // list] as the attribute value;
230 addULEB128(DieNumber);
233 void DIEHash::hashDIEEntry(dwarf::Attribute Attribute, dwarf::Tag Tag,
235 assert(Tag != dwarf::DW_TAG_friend && "No current LLVM clients emit friend "
236 "tags. Add support here when there's "
239 // If the tag in Step 3 is one of [the below tags]
240 if ((Tag == dwarf::DW_TAG_pointer_type ||
241 Tag == dwarf::DW_TAG_reference_type ||
242 Tag == dwarf::DW_TAG_rvalue_reference_type ||
243 Tag == dwarf::DW_TAG_ptr_to_member_type) &&
244 // and the referenced type (via the [below attributes])
245 // FIXME: This seems overly restrictive, and causes hash mismatches
246 // there's a decl/def difference in the containing type of a
247 // ptr_to_member_type, but it's what DWARF says, for some reason.
248 Attribute == dwarf::DW_AT_type) {
249 // ... has a DW_AT_name attribute,
250 StringRef Name = getDIEStringAttr(Entry, dwarf::DW_AT_name);
252 hashShallowTypeReference(Attribute, Entry, Name);
257 unsigned &DieNumber = Numbering[&Entry];
259 hashRepeatedTypeReference(Attribute, DieNumber);
263 // otherwise, b) use the letter 'T' as a the marker, ...
266 addULEB128(Attribute);
268 // ... process the type T recursively by performing Steps 2 through 7, and
269 // use the result as the attribute value.
270 DieNumber = Numbering.size();
274 // Hash an individual attribute \param Attr based on the type of attribute and
276 void DIEHash::hashAttribute(AttrEntry Attr, dwarf::Tag Tag) {
277 const DIEValue *Value = Attr.Val;
278 const DIEAbbrevData *Desc = Attr.Desc;
279 dwarf::Attribute Attribute = Desc->getAttribute();
282 // ... An attribute that refers to another type entry T is processed as
284 if (const DIEEntry *EntryAttr = dyn_cast<DIEEntry>(Value)) {
285 hashDIEEntry(Attribute, Tag, *EntryAttr->getEntry());
289 // Other attribute values use the letter 'A' as the marker, ...
292 addULEB128(Attribute);
294 // ... and the value consists of the form code (encoded as an unsigned LEB128
295 // value) followed by the encoding of the value according to the form code. To
296 // ensure reproducibility of the signature, the set of forms used in the
297 // signature computation is limited to the following: DW_FORM_sdata,
298 // DW_FORM_flag, DW_FORM_string, and DW_FORM_block.
299 switch (Desc->getForm()) {
300 case dwarf::DW_FORM_string:
302 "Add support for DW_FORM_string if we ever start emitting them again");
303 case dwarf::DW_FORM_GNU_str_index:
304 case dwarf::DW_FORM_strp:
305 addULEB128(dwarf::DW_FORM_string);
306 addString(cast<DIEString>(Value)->getString());
308 case dwarf::DW_FORM_data1:
309 case dwarf::DW_FORM_data2:
310 case dwarf::DW_FORM_data4:
311 case dwarf::DW_FORM_data8:
312 case dwarf::DW_FORM_udata:
313 addULEB128(dwarf::DW_FORM_sdata);
314 addSLEB128((int64_t)cast<DIEInteger>(Value)->getValue());
317 llvm_unreachable("Add support for additional forms");
321 // Go through the attributes from \param Attrs in the order specified in 7.27.4
323 void DIEHash::hashAttributes(const DIEAttrs &Attrs, dwarf::Tag Tag) {
324 #define ADD_ATTR(ATTR) \
327 hashAttribute(ATTR, Tag); \
330 ADD_ATTR(Attrs.DW_AT_name);
331 ADD_ATTR(Attrs.DW_AT_accessibility);
332 ADD_ATTR(Attrs.DW_AT_address_class);
333 ADD_ATTR(Attrs.DW_AT_allocated);
334 ADD_ATTR(Attrs.DW_AT_artificial);
335 ADD_ATTR(Attrs.DW_AT_associated);
336 ADD_ATTR(Attrs.DW_AT_binary_scale);
337 ADD_ATTR(Attrs.DW_AT_bit_offset);
338 ADD_ATTR(Attrs.DW_AT_bit_size);
339 ADD_ATTR(Attrs.DW_AT_bit_stride);
340 ADD_ATTR(Attrs.DW_AT_byte_size);
341 ADD_ATTR(Attrs.DW_AT_byte_stride);
342 ADD_ATTR(Attrs.DW_AT_const_expr);
343 ADD_ATTR(Attrs.DW_AT_const_value);
344 ADD_ATTR(Attrs.DW_AT_containing_type);
345 ADD_ATTR(Attrs.DW_AT_count);
346 ADD_ATTR(Attrs.DW_AT_data_bit_offset);
347 ADD_ATTR(Attrs.DW_AT_data_location);
348 ADD_ATTR(Attrs.DW_AT_data_member_location);
349 ADD_ATTR(Attrs.DW_AT_decimal_scale);
350 ADD_ATTR(Attrs.DW_AT_decimal_sign);
351 ADD_ATTR(Attrs.DW_AT_default_value);
352 ADD_ATTR(Attrs.DW_AT_digit_count);
353 ADD_ATTR(Attrs.DW_AT_discr);
354 ADD_ATTR(Attrs.DW_AT_discr_list);
355 ADD_ATTR(Attrs.DW_AT_discr_value);
356 ADD_ATTR(Attrs.DW_AT_encoding);
357 ADD_ATTR(Attrs.DW_AT_enum_class);
358 ADD_ATTR(Attrs.DW_AT_endianity);
359 ADD_ATTR(Attrs.DW_AT_explicit);
360 ADD_ATTR(Attrs.DW_AT_is_optional);
361 ADD_ATTR(Attrs.DW_AT_location);
362 ADD_ATTR(Attrs.DW_AT_lower_bound);
363 ADD_ATTR(Attrs.DW_AT_mutable);
364 ADD_ATTR(Attrs.DW_AT_ordering);
365 ADD_ATTR(Attrs.DW_AT_picture_string);
366 ADD_ATTR(Attrs.DW_AT_prototyped);
367 ADD_ATTR(Attrs.DW_AT_small);
368 ADD_ATTR(Attrs.DW_AT_segment);
369 ADD_ATTR(Attrs.DW_AT_string_length);
370 ADD_ATTR(Attrs.DW_AT_threads_scaled);
371 ADD_ATTR(Attrs.DW_AT_upper_bound);
372 ADD_ATTR(Attrs.DW_AT_use_location);
373 ADD_ATTR(Attrs.DW_AT_use_UTF8);
374 ADD_ATTR(Attrs.DW_AT_variable_parameter);
375 ADD_ATTR(Attrs.DW_AT_virtuality);
376 ADD_ATTR(Attrs.DW_AT_visibility);
377 ADD_ATTR(Attrs.DW_AT_vtable_elem_location);
378 ADD_ATTR(Attrs.DW_AT_type);
380 // FIXME: Add the extended attributes.
383 // Add all of the attributes for \param Die to the hash.
384 void DIEHash::addAttributes(const DIE &Die) {
386 collectAttributes(Die, Attrs);
387 hashAttributes(Attrs, Die.getTag());
390 void DIEHash::hashNestedType(const DIE &Die, StringRef Name) {
392 // ... append the letter 'S',
396 addULEB128(Die.getTag());
402 // Compute the hash of a DIE. This is based on the type signature computation
403 // given in section 7.27 of the DWARF4 standard. It is the md5 hash of a
404 // flattened description of the DIE.
405 void DIEHash::computeHash(const DIE &Die) {
406 // Append the letter 'D', followed by the DWARF tag of the DIE.
408 addULEB128(Die.getTag());
410 // Add each of the attributes of the DIE.
413 // Then hash each of the children of the DIE.
414 for (std::vector<DIE *>::const_iterator I = Die.getChildren().begin(),
415 E = Die.getChildren().end();
418 // If C is a nested type entry or a member function entry, ...
419 if (isType((*I)->getTag()) || (*I)->getTag() == dwarf::DW_TAG_subprogram) {
420 StringRef Name = getDIEStringAttr(**I, dwarf::DW_AT_name);
421 // ... and has a DW_AT_name attribute
423 hashNestedType(**I, Name);
430 // Following the last (or if there are no children), append a zero byte.
431 Hash.update(makeArrayRef((uint8_t)'\0'));
434 /// This is based on the type signature computation given in section 7.27 of the
435 /// DWARF4 standard. It is the md5 hash of a flattened description of the DIE
436 /// with the exception that we are hashing only the context and the name of the
438 uint64_t DIEHash::computeDIEODRSignature(const DIE &Die) {
440 // Add the contexts to the hash. We won't be computing the ODR hash for
441 // function local types so it's safe to use the generic context hashing
443 // FIXME: If we figure out how to account for linkage in some way we could
444 // actually do this with a slight modification to the parent hash algorithm.
445 if (const DIE *Parent = Die.getParent())
446 addParentContext(*Parent);
448 // Add the current DIE information.
450 // Add the DWARF tag of the DIE.
451 addULEB128(Die.getTag());
453 // Add the name of the type to the hash.
454 addString(getDIEStringAttr(Die, dwarf::DW_AT_name));
456 // Now get the result.
457 MD5::MD5Result Result;
460 // ... take the least significant 8 bytes and return those. Our MD5
461 // implementation always returns its results in little endian, swap bytes
463 return *reinterpret_cast<support::ulittle64_t *>(Result + 8);
466 /// This is based on the type signature computation given in section 7.27 of the
467 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE
468 /// with the inclusion of the full CU and all top level CU entities.
469 // TODO: Initialize the type chain at 0 instead of 1 for CU signatures.
470 uint64_t DIEHash::computeCUSignature(const DIE &Die) {
477 // Now return the result.
478 MD5::MD5Result Result;
481 // ... take the least significant 8 bytes and return those. Our MD5
482 // implementation always returns its results in little endian, swap bytes
484 return *reinterpret_cast<support::ulittle64_t *>(Result + 8);
487 /// This is based on the type signature computation given in section 7.27 of the
488 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE
489 /// with the inclusion of additional forms not specifically called out in the
491 uint64_t DIEHash::computeTypeSignature(const DIE &Die) {
495 if (const DIE *Parent = Die.getParent())
496 addParentContext(*Parent);
501 // Now return the result.
502 MD5::MD5Result Result;
505 // ... take the least significant 8 bytes and return those. Our MD5
506 // implementation always returns its results in little endian, swap bytes
508 return *reinterpret_cast<support::ulittle64_t *>(Result + 8);