DIEHash.cpp revision 263508
1//===-- llvm/CodeGen/DIEHash.cpp - Dwarf Hashing Framework ----------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file contains support for DWARF4 hashing of DIEs. 11// 12//===----------------------------------------------------------------------===// 13 14#define DEBUG_TYPE "dwarfdebug" 15 16#include "DIEHash.h" 17 18#include "DIE.h" 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" 27 28using namespace llvm; 29 30/// \brief Grabs the string in whichever attribute is passed in and returns 31/// a reference to it. 32static StringRef getDIEStringAttr(const DIE &Die, uint16_t Attr) { 33 const SmallVectorImpl<DIEValue *> &Values = Die.getValues(); 34 const DIEAbbrev &Abbrevs = Die.getAbbrev(); 35 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(); 44 } 45 } 46 return StringRef(""); 47} 48 49/// \brief Adds the string in \p Str to the hash. This also hashes 50/// a trailing NULL with the string. 51void DIEHash::addString(StringRef Str) { 52 DEBUG(dbgs() << "Adding string " << Str << " to hash.\n"); 53 Hash.update(Str); 54 Hash.update(makeArrayRef((uint8_t)'\0')); 55} 56 57// FIXME: The LEB128 routines are copied and only slightly modified out of 58// LEB128.h. 59 60/// \brief Adds the unsigned in \p Value to the hash encoded as a ULEB128. 61void DIEHash::addULEB128(uint64_t Value) { 62 DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n"); 63 do { 64 uint8_t Byte = Value & 0x7f; 65 Value >>= 7; 66 if (Value != 0) 67 Byte |= 0x80; // Mark this byte to show that more bytes will follow. 68 Hash.update(Byte); 69 } while (Value != 0); 70} 71 72void DIEHash::addSLEB128(int64_t Value) { 73 DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n"); 74 bool More; 75 do { 76 uint8_t Byte = Value & 0x7f; 77 Value >>= 7; 78 More = !((((Value == 0 ) && ((Byte & 0x40) == 0)) || 79 ((Value == -1) && ((Byte & 0x40) != 0)))); 80 if (More) 81 Byte |= 0x80; // Mark this byte to show that more bytes will follow. 82 Hash.update(Byte); 83 } while (More); 84} 85 86/// \brief Including \p Parent adds the context of Parent to the hash.. 87void DIEHash::addParentContext(const DIE &Parent) { 88 89 DEBUG(dbgs() << "Adding parent context to hash...\n"); 90 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->getTag() != dwarf::DW_TAG_compile_unit) { 96 Parents.push_back(Cur); 97 Cur = Cur->getParent(); 98 } 99 100 // Reverse iterate over our list to go from the outermost construct to the 101 // innermost. 102 for (SmallVectorImpl<const DIE *>::reverse_iterator I = Parents.rbegin(), 103 E = Parents.rend(); 104 I != E; ++I) { 105 const DIE &Die = **I; 106 107 // ... Append the letter "C" to the sequence... 108 addULEB128('C'); 109 110 // ... Followed by the DWARF tag of the construct... 111 addULEB128(Die.getTag()); 112 113 // ... 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"); 116 if (!Name.empty()) 117 addString(Name); 118 } 119} 120 121// Collect all of the attributes for a particular DIE in single structure. 122void DIEHash::collectAttributes(const DIE &Die, DIEAttrs &Attrs) { 123 const SmallVectorImpl<DIEValue *> &Values = Die.getValues(); 124 const DIEAbbrev &Abbrevs = Die.getAbbrev(); 125 126#define COLLECT_ATTR(NAME) \ 127 case dwarf::NAME: \ 128 Attrs.NAME.Val = Values[i]; \ 129 Attrs.NAME.Desc = &Abbrevs.getData()[i]; \ 130 break 131 132 for (size_t i = 0, e = Values.size(); i != e; ++i) { 133 DEBUG(dbgs() << "Attribute: " 134 << dwarf::AttributeString(Abbrevs.getData()[i].getAttribute()) 135 << " added.\n"); 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); 186 default: 187 break; 188 } 189 } 190} 191 192void DIEHash::hashShallowTypeReference(dwarf::Attribute Attribute, 193 const DIE &Entry, StringRef Name) { 194 // append the letter 'N' 195 addULEB128('N'); 196 197 // the DWARF attribute code (DW_AT_type or DW_AT_friend), 198 addULEB128(Attribute); 199 200 // the context of the tag, 201 if (const DIE *Parent = Entry.getParent()) 202 addParentContext(*Parent); 203 204 // the letter 'E', 205 addULEB128('E'); 206 207 // and the name of the type. 208 addString(Name); 209 210 // Currently DW_TAG_friends are not used by Clang, but if they do become so, 211 // here's the relevant spec text to implement: 212 // 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). 216} 217 218void DIEHash::hashRepeatedTypeReference(dwarf::Attribute Attribute, 219 unsigned DieNumber) { 220 // a) If T is in the list of [previously hashed types], use the letter 221 // 'R' as the marker 222 addULEB128('R'); 223 224 addULEB128(Attribute); 225 226 // and use the unsigned LEB128 encoding of [the index of T in the 227 // list] as the attribute value; 228 addULEB128(DieNumber); 229} 230 231void DIEHash::hashDIEEntry(dwarf::Attribute Attribute, dwarf::Tag Tag, 232 const DIE &Entry) { 233 assert(Tag != dwarf::DW_TAG_friend && "No current LLVM clients emit friend " 234 "tags. Add support here when there's " 235 "a use case"); 236 // Step 5 237 // 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); 249 if (!Name.empty()) { 250 hashShallowTypeReference(Attribute, Entry, Name); 251 return; 252 } 253 } 254 255 unsigned &DieNumber = Numbering[&Entry]; 256 if (DieNumber) { 257 hashRepeatedTypeReference(Attribute, DieNumber); 258 return; 259 } 260 261 // otherwise, b) use the letter 'T' as a the marker, ... 262 addULEB128('T'); 263 264 addULEB128(Attribute); 265 266 // ... process the type T recursively by performing Steps 2 through 7, and 267 // use the result as the attribute value. 268 DieNumber = Numbering.size(); 269 computeHash(Entry); 270} 271 272// Hash an individual attribute \param Attr based on the type of attribute and 273// the form. 274void 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(); 278 279 // 7.27 Step 3 280 // ... An attribute that refers to another type entry T is processed as 281 // follows: 282 if (const DIEEntry *EntryAttr = dyn_cast<DIEEntry>(Value)) { 283 hashDIEEntry(Attribute, Tag, *EntryAttr->getEntry()); 284 return; 285 } 286 287 // Other attribute values use the letter 'A' as the marker, ... 288 addULEB128('A'); 289 290 addULEB128(Attribute); 291 292 // ... and the value consists of the form code (encoded as an unsigned LEB128 293 // value) followed by the encoding of the value according to the form code. To 294 // ensure reproducibility of the signature, the set of forms used in the 295 // signature computation is limited to the following: DW_FORM_sdata, 296 // DW_FORM_flag, DW_FORM_string, and DW_FORM_block. 297 switch (Desc->getForm()) { 298 case dwarf::DW_FORM_string: 299 llvm_unreachable( 300 "Add support for DW_FORM_string if we ever start emitting them again"); 301 case dwarf::DW_FORM_GNU_str_index: 302 case dwarf::DW_FORM_strp: 303 addULEB128(dwarf::DW_FORM_string); 304 addString(cast<DIEString>(Value)->getString()); 305 break; 306 case dwarf::DW_FORM_data1: 307 case dwarf::DW_FORM_data2: 308 case dwarf::DW_FORM_data4: 309 case dwarf::DW_FORM_data8: 310 case dwarf::DW_FORM_udata: 311 addULEB128(dwarf::DW_FORM_sdata); 312 addSLEB128((int64_t)cast<DIEInteger>(Value)->getValue()); 313 break; 314 default: 315 llvm_unreachable("Add support for additional forms"); 316 } 317} 318 319// Go through the attributes from \param Attrs in the order specified in 7.27.4 320// and hash them. 321void DIEHash::hashAttributes(const DIEAttrs &Attrs, dwarf::Tag Tag) { 322#define ADD_ATTR(ATTR) \ 323 { \ 324 if (ATTR.Val != 0) \ 325 hashAttribute(ATTR, Tag); \ 326 } 327 328 ADD_ATTR(Attrs.DW_AT_name); 329 ADD_ATTR(Attrs.DW_AT_accessibility); 330 ADD_ATTR(Attrs.DW_AT_address_class); 331 ADD_ATTR(Attrs.DW_AT_allocated); 332 ADD_ATTR(Attrs.DW_AT_artificial); 333 ADD_ATTR(Attrs.DW_AT_associated); 334 ADD_ATTR(Attrs.DW_AT_binary_scale); 335 ADD_ATTR(Attrs.DW_AT_bit_offset); 336 ADD_ATTR(Attrs.DW_AT_bit_size); 337 ADD_ATTR(Attrs.DW_AT_bit_stride); 338 ADD_ATTR(Attrs.DW_AT_byte_size); 339 ADD_ATTR(Attrs.DW_AT_byte_stride); 340 ADD_ATTR(Attrs.DW_AT_const_expr); 341 ADD_ATTR(Attrs.DW_AT_const_value); 342 ADD_ATTR(Attrs.DW_AT_containing_type); 343 ADD_ATTR(Attrs.DW_AT_count); 344 ADD_ATTR(Attrs.DW_AT_data_bit_offset); 345 ADD_ATTR(Attrs.DW_AT_data_location); 346 ADD_ATTR(Attrs.DW_AT_data_member_location); 347 ADD_ATTR(Attrs.DW_AT_decimal_scale); 348 ADD_ATTR(Attrs.DW_AT_decimal_sign); 349 ADD_ATTR(Attrs.DW_AT_default_value); 350 ADD_ATTR(Attrs.DW_AT_digit_count); 351 ADD_ATTR(Attrs.DW_AT_discr); 352 ADD_ATTR(Attrs.DW_AT_discr_list); 353 ADD_ATTR(Attrs.DW_AT_discr_value); 354 ADD_ATTR(Attrs.DW_AT_encoding); 355 ADD_ATTR(Attrs.DW_AT_enum_class); 356 ADD_ATTR(Attrs.DW_AT_endianity); 357 ADD_ATTR(Attrs.DW_AT_explicit); 358 ADD_ATTR(Attrs.DW_AT_is_optional); 359 ADD_ATTR(Attrs.DW_AT_location); 360 ADD_ATTR(Attrs.DW_AT_lower_bound); 361 ADD_ATTR(Attrs.DW_AT_mutable); 362 ADD_ATTR(Attrs.DW_AT_ordering); 363 ADD_ATTR(Attrs.DW_AT_picture_string); 364 ADD_ATTR(Attrs.DW_AT_prototyped); 365 ADD_ATTR(Attrs.DW_AT_small); 366 ADD_ATTR(Attrs.DW_AT_segment); 367 ADD_ATTR(Attrs.DW_AT_string_length); 368 ADD_ATTR(Attrs.DW_AT_threads_scaled); 369 ADD_ATTR(Attrs.DW_AT_upper_bound); 370 ADD_ATTR(Attrs.DW_AT_use_location); 371 ADD_ATTR(Attrs.DW_AT_use_UTF8); 372 ADD_ATTR(Attrs.DW_AT_variable_parameter); 373 ADD_ATTR(Attrs.DW_AT_virtuality); 374 ADD_ATTR(Attrs.DW_AT_visibility); 375 ADD_ATTR(Attrs.DW_AT_vtable_elem_location); 376 ADD_ATTR(Attrs.DW_AT_type); 377 378 // FIXME: Add the extended attributes. 379} 380 381// Add all of the attributes for \param Die to the hash. 382void DIEHash::addAttributes(const DIE &Die) { 383 DIEAttrs Attrs = {}; 384 collectAttributes(Die, Attrs); 385 hashAttributes(Attrs, Die.getTag()); 386} 387 388void DIEHash::hashNestedType(const DIE &Die, StringRef Name) { 389 // 7.27 Step 7 390 // ... append the letter 'S', 391 addULEB128('S'); 392 393 // the tag of C, 394 addULEB128(Die.getTag()); 395 396 // and the name. 397 addString(Name); 398} 399 400// Compute the hash of a DIE. This is based on the type signature computation 401// given in section 7.27 of the DWARF4 standard. It is the md5 hash of a 402// flattened description of the DIE. 403void DIEHash::computeHash(const DIE &Die) { 404 // Append the letter 'D', followed by the DWARF tag of the DIE. 405 addULEB128('D'); 406 addULEB128(Die.getTag()); 407 408 // Add each of the attributes of the DIE. 409 addAttributes(Die); 410 411 // Then hash each of the children of the DIE. 412 for (std::vector<DIE *>::const_iterator I = Die.getChildren().begin(), 413 E = Die.getChildren().end(); 414 I != E; ++I) { 415 // 7.27 Step 7 416 // If C is a nested type entry or a member function entry, ... 417 if (isType((*I)->getTag()) || (*I)->getTag() == dwarf::DW_TAG_subprogram) { 418 StringRef Name = getDIEStringAttr(**I, dwarf::DW_AT_name); 419 // ... and has a DW_AT_name attribute 420 if (!Name.empty()) { 421 hashNestedType(**I, Name); 422 continue; 423 } 424 } 425 computeHash(**I); 426 } 427 428 // Following the last (or if there are no children), append a zero byte. 429 Hash.update(makeArrayRef((uint8_t)'\0')); 430} 431 432/// This is based on the type signature computation given in section 7.27 of the 433/// DWARF4 standard. It is the md5 hash of a flattened description of the DIE 434/// with the exception that we are hashing only the context and the name of the 435/// type. 436uint64_t DIEHash::computeDIEODRSignature(const DIE &Die) { 437 438 // Add the contexts to the hash. We won't be computing the ODR hash for 439 // function local types so it's safe to use the generic context hashing 440 // algorithm here. 441 // FIXME: If we figure out how to account for linkage in some way we could 442 // actually do this with a slight modification to the parent hash algorithm. 443 if (const DIE *Parent = Die.getParent()) 444 addParentContext(*Parent); 445 446 // Add the current DIE information. 447 448 // Add the DWARF tag of the DIE. 449 addULEB128(Die.getTag()); 450 451 // Add the name of the type to the hash. 452 addString(getDIEStringAttr(Die, dwarf::DW_AT_name)); 453 454 // Now get the result. 455 MD5::MD5Result Result; 456 Hash.final(Result); 457 458 // ... take the least significant 8 bytes and return those. Our MD5 459 // implementation always returns its results in little endian, swap bytes 460 // appropriately. 461 return *reinterpret_cast<support::ulittle64_t *>(Result + 8); 462} 463 464/// This is based on the type signature computation given in section 7.27 of the 465/// DWARF4 standard. It is an md5 hash of the flattened description of the DIE 466/// with the inclusion of the full CU and all top level CU entities. 467// TODO: Initialize the type chain at 0 instead of 1 for CU signatures. 468uint64_t DIEHash::computeCUSignature(const DIE &Die) { 469 Numbering.clear(); 470 Numbering[&Die] = 1; 471 472 // Hash the DIE. 473 computeHash(Die); 474 475 // Now return the result. 476 MD5::MD5Result Result; 477 Hash.final(Result); 478 479 // ... take the least significant 8 bytes and return those. Our MD5 480 // implementation always returns its results in little endian, swap bytes 481 // appropriately. 482 return *reinterpret_cast<support::ulittle64_t *>(Result + 8); 483} 484 485/// This is based on the type signature computation given in section 7.27 of the 486/// DWARF4 standard. It is an md5 hash of the flattened description of the DIE 487/// with the inclusion of additional forms not specifically called out in the 488/// standard. 489uint64_t DIEHash::computeTypeSignature(const DIE &Die) { 490 Numbering.clear(); 491 Numbering[&Die] = 1; 492 493 if (const DIE *Parent = Die.getParent()) 494 addParentContext(*Parent); 495 496 // Hash the DIE. 497 computeHash(Die); 498 499 // Now return the result. 500 MD5::MD5Result Result; 501 Hash.final(Result); 502 503 // ... take the least significant 8 bytes and return those. Our MD5 504 // implementation always returns its results in little endian, swap bytes 505 // appropriately. 506 return *reinterpret_cast<support::ulittle64_t *>(Result + 8); 507} 508