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 "DwarfCompileUnit.h"
19 #include "llvm/ADT/ArrayRef.h"
20 #include "llvm/ADT/StringRef.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 /// \brief Grabs the string in whichever attribute is passed in and returns
30 /// a reference to it.
31 static StringRef getDIEStringAttr(const DIE &Die, uint16_t Attr) {
32 const SmallVectorImpl<DIEValue *> &Values = Die.getValues();
33 const DIEAbbrev &Abbrevs = Die.getAbbrev();
35 // Iterate through all the attributes until we find the one we're
36 // looking for, if we can't find it return an empty string.
37 for (size_t i = 0; i < Values.size(); ++i) {
38 if (Abbrevs.getData()[i].getAttribute() == Attr) {
39 DIEValue *V = Values[i];
40 assert(isa<DIEString>(V) && "String requested. Not a string.");
41 DIEString *S = cast<DIEString>(V);
42 return S->getString();
48 /// \brief Adds the string in \p Str to the hash. This also hashes
49 /// a trailing NULL with the string.
50 void DIEHash::addString(StringRef Str) {
51 DEBUG(dbgs() << "Adding string " << Str << " to hash.\n");
53 Hash.update(makeArrayRef((uint8_t)'\0'));
56 // FIXME: The LEB128 routines are copied and only slightly modified out of
59 /// \brief Adds the unsigned in \p Value to the hash encoded as a ULEB128.
60 void DIEHash::addULEB128(uint64_t Value) {
61 DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n");
63 uint8_t Byte = Value & 0x7f;
66 Byte |= 0x80; // Mark this byte to show that more bytes will follow.
71 void DIEHash::addSLEB128(int64_t Value) {
72 DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n");
75 uint8_t Byte = Value & 0x7f;
77 More = !((((Value == 0 ) && ((Byte & 0x40) == 0)) ||
78 ((Value == -1) && ((Byte & 0x40) != 0))));
80 Byte |= 0x80; // Mark this byte to show that more bytes will follow.
85 /// \brief Including \p Parent adds the context of Parent to the hash..
86 void DIEHash::addParentContext(const DIE &Parent) {
88 DEBUG(dbgs() << "Adding parent context to hash...\n");
90 // [7.27.2] For each surrounding type or namespace beginning with the
91 // outermost such construct...
92 SmallVector<const DIE *, 1> Parents;
93 const DIE *Cur = &Parent;
94 while (Cur->getTag() != dwarf::DW_TAG_compile_unit) {
95 Parents.push_back(Cur);
96 Cur = Cur->getParent();
99 // Reverse iterate over our list to go from the outermost construct to the
101 for (SmallVectorImpl<const DIE *>::reverse_iterator I = Parents.rbegin(),
104 const DIE &Die = **I;
106 // ... Append the letter "C" to the sequence...
109 // ... Followed by the DWARF tag of the construct...
110 addULEB128(Die.getTag());
112 // ... Then the name, taken from the DW_AT_name attribute.
113 StringRef Name = getDIEStringAttr(Die, dwarf::DW_AT_name);
114 DEBUG(dbgs() << "... adding context: " << Name << "\n");
120 // Collect all of the attributes for a particular DIE in single structure.
121 void DIEHash::collectAttributes(const DIE &Die, DIEAttrs &Attrs) {
122 const SmallVectorImpl<DIEValue *> &Values = Die.getValues();
123 const DIEAbbrev &Abbrevs = Die.getAbbrev();
125 #define COLLECT_ATTR(NAME) \
127 Attrs.NAME.Val = Values[i]; \
128 Attrs.NAME.Desc = &Abbrevs.getData()[i]; \
131 for (size_t i = 0, e = Values.size(); i != e; ++i) {
132 DEBUG(dbgs() << "Attribute: "
133 << dwarf::AttributeString(Abbrevs.getData()[i].getAttribute())
135 switch (Abbrevs.getData()[i].getAttribute()) {
136 COLLECT_ATTR(DW_AT_name);
137 COLLECT_ATTR(DW_AT_accessibility);
138 COLLECT_ATTR(DW_AT_address_class);
139 COLLECT_ATTR(DW_AT_allocated);
140 COLLECT_ATTR(DW_AT_artificial);
141 COLLECT_ATTR(DW_AT_associated);
142 COLLECT_ATTR(DW_AT_binary_scale);
143 COLLECT_ATTR(DW_AT_bit_offset);
144 COLLECT_ATTR(DW_AT_bit_size);
145 COLLECT_ATTR(DW_AT_bit_stride);
146 COLLECT_ATTR(DW_AT_byte_size);
147 COLLECT_ATTR(DW_AT_byte_stride);
148 COLLECT_ATTR(DW_AT_const_expr);
149 COLLECT_ATTR(DW_AT_const_value);
150 COLLECT_ATTR(DW_AT_containing_type);
151 COLLECT_ATTR(DW_AT_count);
152 COLLECT_ATTR(DW_AT_data_bit_offset);
153 COLLECT_ATTR(DW_AT_data_location);
154 COLLECT_ATTR(DW_AT_data_member_location);
155 COLLECT_ATTR(DW_AT_decimal_scale);
156 COLLECT_ATTR(DW_AT_decimal_sign);
157 COLLECT_ATTR(DW_AT_default_value);
158 COLLECT_ATTR(DW_AT_digit_count);
159 COLLECT_ATTR(DW_AT_discr);
160 COLLECT_ATTR(DW_AT_discr_list);
161 COLLECT_ATTR(DW_AT_discr_value);
162 COLLECT_ATTR(DW_AT_encoding);
163 COLLECT_ATTR(DW_AT_enum_class);
164 COLLECT_ATTR(DW_AT_endianity);
165 COLLECT_ATTR(DW_AT_explicit);
166 COLLECT_ATTR(DW_AT_is_optional);
167 COLLECT_ATTR(DW_AT_location);
168 COLLECT_ATTR(DW_AT_lower_bound);
169 COLLECT_ATTR(DW_AT_mutable);
170 COLLECT_ATTR(DW_AT_ordering);
171 COLLECT_ATTR(DW_AT_picture_string);
172 COLLECT_ATTR(DW_AT_prototyped);
173 COLLECT_ATTR(DW_AT_small);
174 COLLECT_ATTR(DW_AT_segment);
175 COLLECT_ATTR(DW_AT_string_length);
176 COLLECT_ATTR(DW_AT_threads_scaled);
177 COLLECT_ATTR(DW_AT_upper_bound);
178 COLLECT_ATTR(DW_AT_use_location);
179 COLLECT_ATTR(DW_AT_use_UTF8);
180 COLLECT_ATTR(DW_AT_variable_parameter);
181 COLLECT_ATTR(DW_AT_virtuality);
182 COLLECT_ATTR(DW_AT_visibility);
183 COLLECT_ATTR(DW_AT_vtable_elem_location);
184 COLLECT_ATTR(DW_AT_type);
191 void DIEHash::hashShallowTypeReference(dwarf::Attribute Attribute,
192 const DIE &Entry, StringRef Name) {
193 // append the letter 'N'
196 // the DWARF attribute code (DW_AT_type or DW_AT_friend),
197 addULEB128(Attribute);
199 // the context of the tag,
200 if (const DIE *Parent = Entry.getParent())
201 addParentContext(*Parent);
206 // and the name of the type.
209 // Currently DW_TAG_friends are not used by Clang, but if they do become so,
210 // here's the relevant spec text to implement:
212 // For DW_TAG_friend, if the referenced entry is the DW_TAG_subprogram,
213 // the context is omitted and the name to be used is the ABI-specific name
214 // of the subprogram (e.g., the mangled linker name).
217 void DIEHash::hashRepeatedTypeReference(dwarf::Attribute Attribute,
218 unsigned DieNumber) {
219 // a) If T is in the list of [previously hashed types], use the letter
223 addULEB128(Attribute);
225 // and use the unsigned LEB128 encoding of [the index of T in the
226 // list] as the attribute value;
227 addULEB128(DieNumber);
230 void DIEHash::hashDIEEntry(dwarf::Attribute Attribute, dwarf::Tag Tag,
232 assert(Tag != dwarf::DW_TAG_friend && "No current LLVM clients emit friend "
233 "tags. Add support here when there's "
236 // If the tag in Step 3 is one of [the below tags]
237 if ((Tag == dwarf::DW_TAG_pointer_type ||
238 Tag == dwarf::DW_TAG_reference_type ||
239 Tag == dwarf::DW_TAG_rvalue_reference_type ||
240 Tag == dwarf::DW_TAG_ptr_to_member_type) &&
241 // and the referenced type (via the [below attributes])
242 // FIXME: This seems overly restrictive, and causes hash mismatches
243 // there's a decl/def difference in the containing type of a
244 // ptr_to_member_type, but it's what DWARF says, for some reason.
245 Attribute == dwarf::DW_AT_type) {
246 // ... has a DW_AT_name attribute,
247 StringRef Name = getDIEStringAttr(Entry, dwarf::DW_AT_name);
249 hashShallowTypeReference(Attribute, Entry, Name);
254 unsigned &DieNumber = Numbering[&Entry];
256 hashRepeatedTypeReference(Attribute, DieNumber);
260 // otherwise, b) use the letter 'T' as a the marker, ...
263 addULEB128(Attribute);
265 // ... process the type T recursively by performing Steps 2 through 7, and
266 // use the result as the attribute value.
267 DieNumber = Numbering.size();
271 // Hash an individual attribute \param Attr based on the type of attribute and
273 void DIEHash::hashAttribute(AttrEntry Attr, dwarf::Tag Tag) {
274 const DIEValue *Value = Attr.Val;
275 const DIEAbbrevData *Desc = Attr.Desc;
276 dwarf::Attribute Attribute = Desc->getAttribute();
279 // ... An attribute that refers to another type entry T is processed as
281 if (const DIEEntry *EntryAttr = dyn_cast<DIEEntry>(Value)) {
282 hashDIEEntry(Attribute, Tag, *EntryAttr->getEntry());
286 // Other attribute values use the letter 'A' as the marker, ...
289 addULEB128(Attribute);
291 // ... and the value consists of the form code (encoded as an unsigned LEB128
292 // value) followed by the encoding of the value according to the form code. To
293 // ensure reproducibility of the signature, the set of forms used in the
294 // signature computation is limited to the following: DW_FORM_sdata,
295 // DW_FORM_flag, DW_FORM_string, and DW_FORM_block.
296 switch (Desc->getForm()) {
297 case dwarf::DW_FORM_string:
299 "Add support for DW_FORM_string if we ever start emitting them again");
300 case dwarf::DW_FORM_GNU_str_index:
301 case dwarf::DW_FORM_strp:
302 addULEB128(dwarf::DW_FORM_string);
303 addString(cast<DIEString>(Value)->getString());
305 case dwarf::DW_FORM_data1:
306 case dwarf::DW_FORM_data2:
307 case dwarf::DW_FORM_data4:
308 case dwarf::DW_FORM_data8:
309 case dwarf::DW_FORM_udata:
310 addULEB128(dwarf::DW_FORM_sdata);
311 addSLEB128((int64_t)cast<DIEInteger>(Value)->getValue());
314 llvm_unreachable("Add support for additional forms");
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 ADD_ATTR(ATTR) \
324 hashAttribute(ATTR, Tag); \
327 ADD_ATTR(Attrs.DW_AT_name);
328 ADD_ATTR(Attrs.DW_AT_accessibility);
329 ADD_ATTR(Attrs.DW_AT_address_class);
330 ADD_ATTR(Attrs.DW_AT_allocated);
331 ADD_ATTR(Attrs.DW_AT_artificial);
332 ADD_ATTR(Attrs.DW_AT_associated);
333 ADD_ATTR(Attrs.DW_AT_binary_scale);
334 ADD_ATTR(Attrs.DW_AT_bit_offset);
335 ADD_ATTR(Attrs.DW_AT_bit_size);
336 ADD_ATTR(Attrs.DW_AT_bit_stride);
337 ADD_ATTR(Attrs.DW_AT_byte_size);
338 ADD_ATTR(Attrs.DW_AT_byte_stride);
339 ADD_ATTR(Attrs.DW_AT_const_expr);
340 ADD_ATTR(Attrs.DW_AT_const_value);
341 ADD_ATTR(Attrs.DW_AT_containing_type);
342 ADD_ATTR(Attrs.DW_AT_count);
343 ADD_ATTR(Attrs.DW_AT_data_bit_offset);
344 ADD_ATTR(Attrs.DW_AT_data_location);
345 ADD_ATTR(Attrs.DW_AT_data_member_location);
346 ADD_ATTR(Attrs.DW_AT_decimal_scale);
347 ADD_ATTR(Attrs.DW_AT_decimal_sign);
348 ADD_ATTR(Attrs.DW_AT_default_value);
349 ADD_ATTR(Attrs.DW_AT_digit_count);
350 ADD_ATTR(Attrs.DW_AT_discr);
351 ADD_ATTR(Attrs.DW_AT_discr_list);
352 ADD_ATTR(Attrs.DW_AT_discr_value);
353 ADD_ATTR(Attrs.DW_AT_encoding);
354 ADD_ATTR(Attrs.DW_AT_enum_class);
355 ADD_ATTR(Attrs.DW_AT_endianity);
356 ADD_ATTR(Attrs.DW_AT_explicit);
357 ADD_ATTR(Attrs.DW_AT_is_optional);
358 ADD_ATTR(Attrs.DW_AT_location);
359 ADD_ATTR(Attrs.DW_AT_lower_bound);
360 ADD_ATTR(Attrs.DW_AT_mutable);
361 ADD_ATTR(Attrs.DW_AT_ordering);
362 ADD_ATTR(Attrs.DW_AT_picture_string);
363 ADD_ATTR(Attrs.DW_AT_prototyped);
364 ADD_ATTR(Attrs.DW_AT_small);
365 ADD_ATTR(Attrs.DW_AT_segment);
366 ADD_ATTR(Attrs.DW_AT_string_length);
367 ADD_ATTR(Attrs.DW_AT_threads_scaled);
368 ADD_ATTR(Attrs.DW_AT_upper_bound);
369 ADD_ATTR(Attrs.DW_AT_use_location);
370 ADD_ATTR(Attrs.DW_AT_use_UTF8);
371 ADD_ATTR(Attrs.DW_AT_variable_parameter);
372 ADD_ATTR(Attrs.DW_AT_virtuality);
373 ADD_ATTR(Attrs.DW_AT_visibility);
374 ADD_ATTR(Attrs.DW_AT_vtable_elem_location);
375 ADD_ATTR(Attrs.DW_AT_type);
377 // FIXME: Add the extended attributes.
380 // Add all of the attributes for \param Die to the hash.
381 void DIEHash::addAttributes(const DIE &Die) {
383 collectAttributes(Die, Attrs);
384 hashAttributes(Attrs, Die.getTag());
387 // Compute the hash of a DIE. This is based on the type signature computation
388 // given in section 7.27 of the DWARF4 standard. It is the md5 hash of a
389 // flattened description of the DIE.
390 void DIEHash::computeHash(const DIE &Die) {
391 // Append the letter 'D', followed by the DWARF tag of the DIE.
393 addULEB128(Die.getTag());
395 // Add each of the attributes of the DIE.
398 // Then hash each of the children of the DIE.
399 for (std::vector<DIE *>::const_iterator I = Die.getChildren().begin(),
400 E = Die.getChildren().end();
404 // Following the last (or if there are no children), append a zero byte.
405 Hash.update(makeArrayRef((uint8_t)'\0'));
408 /// This is based on the type signature computation given in section 7.27 of the
409 /// DWARF4 standard. It is the md5 hash of a flattened description of the DIE
410 /// with the exception that we are hashing only the context and the name of the
412 uint64_t DIEHash::computeDIEODRSignature(const DIE &Die) {
414 // Add the contexts to the hash. We won't be computing the ODR hash for
415 // function local types so it's safe to use the generic context hashing
417 // FIXME: If we figure out how to account for linkage in some way we could
418 // actually do this with a slight modification to the parent hash algorithm.
419 if (const DIE *Parent = Die.getParent())
420 addParentContext(*Parent);
422 // Add the current DIE information.
424 // Add the DWARF tag of the DIE.
425 addULEB128(Die.getTag());
427 // Add the name of the type to the hash.
428 addString(getDIEStringAttr(Die, dwarf::DW_AT_name));
430 // Now get the result.
431 MD5::MD5Result Result;
434 // ... take the least significant 8 bytes and return those. Our MD5
435 // implementation always returns its results in little endian, swap bytes
437 return *reinterpret_cast<support::ulittle64_t *>(Result + 8);
440 /// This is based on the type signature computation given in section 7.27 of the
441 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE
442 /// with the inclusion of the full CU and all top level CU entities.
443 // TODO: Initialize the type chain at 0 instead of 1 for CU signatures.
444 uint64_t DIEHash::computeCUSignature(const DIE &Die) {
451 // Now return the result.
452 MD5::MD5Result Result;
455 // ... take the least significant 8 bytes and return those. Our MD5
456 // implementation always returns its results in little endian, swap bytes
458 return *reinterpret_cast<support::ulittle64_t *>(Result + 8);
461 /// This is based on the type signature computation given in section 7.27 of the
462 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE
463 /// with the inclusion of additional forms not specifically called out in the
465 uint64_t DIEHash::computeTypeSignature(const DIE &Die) {
469 if (const DIE *Parent = Die.getParent())
470 addParentContext(*Parent);
475 // Now return the result.
476 MD5::MD5Result Result;
479 // ... take the least significant 8 bytes and return those. Our MD5
480 // implementation always returns its results in little endian, swap bytes
482 return *reinterpret_cast<support::ulittle64_t *>(Result + 8);