1 //===-- llvm/CodeGen/DIEHash.cpp - Dwarf Hashing Framework ----------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file contains support for DWARF4 hashing of DIEs. 10 // 11 //===----------------------------------------------------------------------===// 12 13 #include "DIEHash.h" 14 #include "ByteStreamer.h" 15 #include "DwarfCompileUnit.h" 16 #include "DwarfDebug.h" 17 #include "llvm/ADT/ArrayRef.h" 18 #include "llvm/ADT/StringRef.h" 19 #include "llvm/BinaryFormat/Dwarf.h" 20 #include "llvm/CodeGen/AsmPrinter.h" 21 #include "llvm/Support/Debug.h" 22 #include "llvm/Support/Endian.h" 23 #include "llvm/Support/raw_ostream.h" 24 25 using namespace llvm; 26 27 #define DEBUG_TYPE "dwarfdebug" 28 29 /// 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 // Iterate through all the attributes until we find the one we're 33 // looking for, if we can't find it return an empty string. 34 for (const auto &V : Die.values()) 35 if (V.getAttribute() == Attr) 36 return V.getDIEString().getString(); 37 38 return StringRef(""); 39 } 40 41 /// Adds the string in \p Str to the hash. This also hashes 42 /// a trailing NULL with the string. 43 void DIEHash::addString(StringRef Str) { 44 LLVM_DEBUG(dbgs() << "Adding string " << Str << " to hash.\n"); 45 Hash.update(Str); 46 Hash.update(makeArrayRef((uint8_t)'\0')); 47 } 48 49 // FIXME: The LEB128 routines are copied and only slightly modified out of 50 // LEB128.h. 51 52 /// Adds the unsigned in \p Value to the hash encoded as a ULEB128. 53 void DIEHash::addULEB128(uint64_t Value) { 54 LLVM_DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n"); 55 do { 56 uint8_t Byte = Value & 0x7f; 57 Value >>= 7; 58 if (Value != 0) 59 Byte |= 0x80; // Mark this byte to show that more bytes will follow. 60 Hash.update(Byte); 61 } while (Value != 0); 62 } 63 64 void DIEHash::addSLEB128(int64_t Value) { 65 LLVM_DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n"); 66 bool More; 67 do { 68 uint8_t Byte = Value & 0x7f; 69 Value >>= 7; 70 More = !((((Value == 0) && ((Byte & 0x40) == 0)) || 71 ((Value == -1) && ((Byte & 0x40) != 0)))); 72 if (More) 73 Byte |= 0x80; // Mark this byte to show that more bytes will follow. 74 Hash.update(Byte); 75 } while (More); 76 } 77 78 /// Including \p Parent adds the context of Parent to the hash.. 79 void DIEHash::addParentContext(const DIE &Parent) { 80 81 LLVM_DEBUG(dbgs() << "Adding parent context to hash...\n"); 82 83 // [7.27.2] For each surrounding type or namespace beginning with the 84 // outermost such construct... 85 SmallVector<const DIE *, 1> Parents; 86 const DIE *Cur = &Parent; 87 while (Cur->getParent()) { 88 Parents.push_back(Cur); 89 Cur = Cur->getParent(); 90 } 91 assert(Cur->getTag() == dwarf::DW_TAG_compile_unit || 92 Cur->getTag() == dwarf::DW_TAG_type_unit); 93 94 // Reverse iterate over our list to go from the outermost construct to the 95 // innermost. 96 for (SmallVectorImpl<const DIE *>::reverse_iterator I = Parents.rbegin(), 97 E = Parents.rend(); 98 I != E; ++I) { 99 const DIE &Die = **I; 100 101 // ... Append the letter "C" to the sequence... 102 addULEB128('C'); 103 104 // ... Followed by the DWARF tag of the construct... 105 addULEB128(Die.getTag()); 106 107 // ... Then the name, taken from the DW_AT_name attribute. 108 StringRef Name = getDIEStringAttr(Die, dwarf::DW_AT_name); 109 LLVM_DEBUG(dbgs() << "... adding context: " << Name << "\n"); 110 if (!Name.empty()) 111 addString(Name); 112 } 113 } 114 115 // Collect all of the attributes for a particular DIE in single structure. 116 void DIEHash::collectAttributes(const DIE &Die, DIEAttrs &Attrs) { 117 118 for (const auto &V : Die.values()) { 119 LLVM_DEBUG(dbgs() << "Attribute: " 120 << dwarf::AttributeString(V.getAttribute()) 121 << " added.\n"); 122 switch (V.getAttribute()) { 123 #define HANDLE_DIE_HASH_ATTR(NAME) \ 124 case dwarf::NAME: \ 125 Attrs.NAME = V; \ 126 break; 127 #include "DIEHashAttributes.def" 128 default: 129 break; 130 } 131 } 132 } 133 134 void DIEHash::hashShallowTypeReference(dwarf::Attribute Attribute, 135 const DIE &Entry, StringRef Name) { 136 // append the letter 'N' 137 addULEB128('N'); 138 139 // the DWARF attribute code (DW_AT_type or DW_AT_friend), 140 addULEB128(Attribute); 141 142 // the context of the tag, 143 if (const DIE *Parent = Entry.getParent()) 144 addParentContext(*Parent); 145 146 // the letter 'E', 147 addULEB128('E'); 148 149 // and the name of the type. 150 addString(Name); 151 152 // Currently DW_TAG_friends are not used by Clang, but if they do become so, 153 // here's the relevant spec text to implement: 154 // 155 // For DW_TAG_friend, if the referenced entry is the DW_TAG_subprogram, 156 // the context is omitted and the name to be used is the ABI-specific name 157 // of the subprogram (e.g., the mangled linker name). 158 } 159 160 void DIEHash::hashRepeatedTypeReference(dwarf::Attribute Attribute, 161 unsigned DieNumber) { 162 // a) If T is in the list of [previously hashed types], use the letter 163 // 'R' as the marker 164 addULEB128('R'); 165 166 addULEB128(Attribute); 167 168 // and use the unsigned LEB128 encoding of [the index of T in the 169 // list] as the attribute value; 170 addULEB128(DieNumber); 171 } 172 173 void DIEHash::hashDIEEntry(dwarf::Attribute Attribute, dwarf::Tag Tag, 174 const DIE &Entry) { 175 assert(Tag != dwarf::DW_TAG_friend && "No current LLVM clients emit friend " 176 "tags. Add support here when there's " 177 "a use case"); 178 // Step 5 179 // If the tag in Step 3 is one of [the below tags] 180 if ((Tag == dwarf::DW_TAG_pointer_type || 181 Tag == dwarf::DW_TAG_reference_type || 182 Tag == dwarf::DW_TAG_rvalue_reference_type || 183 Tag == dwarf::DW_TAG_ptr_to_member_type) && 184 // and the referenced type (via the [below attributes]) 185 // FIXME: This seems overly restrictive, and causes hash mismatches 186 // there's a decl/def difference in the containing type of a 187 // ptr_to_member_type, but it's what DWARF says, for some reason. 188 Attribute == dwarf::DW_AT_type) { 189 // ... has a DW_AT_name attribute, 190 StringRef Name = getDIEStringAttr(Entry, dwarf::DW_AT_name); 191 if (!Name.empty()) { 192 hashShallowTypeReference(Attribute, Entry, Name); 193 return; 194 } 195 } 196 197 unsigned &DieNumber = Numbering[&Entry]; 198 if (DieNumber) { 199 hashRepeatedTypeReference(Attribute, DieNumber); 200 return; 201 } 202 203 // otherwise, b) use the letter 'T' as the marker, ... 204 addULEB128('T'); 205 206 addULEB128(Attribute); 207 208 // ... process the type T recursively by performing Steps 2 through 7, and 209 // use the result as the attribute value. 210 DieNumber = Numbering.size(); 211 computeHash(Entry); 212 } 213 214 // Hash all of the values in a block like set of values. This assumes that 215 // all of the data is going to be added as integers. 216 void DIEHash::hashBlockData(const DIE::const_value_range &Values) { 217 for (const auto &V : Values) 218 if (V.getType() == DIEValue::isBaseTypeRef) { 219 const DIE &C = 220 *CU->ExprRefedBaseTypes[V.getDIEBaseTypeRef().getIndex()].Die; 221 StringRef Name = getDIEStringAttr(C, dwarf::DW_AT_name); 222 assert(!Name.empty() && 223 "Base types referenced from DW_OP_convert should have a name"); 224 hashNestedType(C, Name); 225 } else 226 Hash.update((uint64_t)V.getDIEInteger().getValue()); 227 } 228 229 // Hash the contents of a loclistptr class. 230 void DIEHash::hashLocList(const DIELocList &LocList) { 231 HashingByteStreamer Streamer(*this); 232 DwarfDebug &DD = *AP->getDwarfDebug(); 233 const DebugLocStream &Locs = DD.getDebugLocs(); 234 const DebugLocStream::List &List = Locs.getList(LocList.getValue()); 235 for (const DebugLocStream::Entry &Entry : Locs.getEntries(List)) 236 DD.emitDebugLocEntry(Streamer, Entry, List.CU); 237 } 238 239 // Hash an individual attribute \param Attr based on the type of attribute and 240 // the form. 241 void DIEHash::hashAttribute(const DIEValue &Value, dwarf::Tag Tag) { 242 dwarf::Attribute Attribute = Value.getAttribute(); 243 244 // Other attribute values use the letter 'A' as the marker, and the value 245 // consists of the form code (encoded as an unsigned LEB128 value) followed by 246 // the encoding of the value according to the form code. To ensure 247 // reproducibility of the signature, the set of forms used in the signature 248 // computation is limited to the following: DW_FORM_sdata, DW_FORM_flag, 249 // DW_FORM_string, and DW_FORM_block. 250 251 switch (Value.getType()) { 252 case DIEValue::isNone: 253 llvm_unreachable("Expected valid DIEValue"); 254 255 // 7.27 Step 3 256 // ... An attribute that refers to another type entry T is processed as 257 // follows: 258 case DIEValue::isEntry: 259 hashDIEEntry(Attribute, Tag, Value.getDIEEntry().getEntry()); 260 break; 261 case DIEValue::isInteger: { 262 addULEB128('A'); 263 addULEB128(Attribute); 264 switch (Value.getForm()) { 265 case dwarf::DW_FORM_data1: 266 case dwarf::DW_FORM_data2: 267 case dwarf::DW_FORM_data4: 268 case dwarf::DW_FORM_data8: 269 case dwarf::DW_FORM_udata: 270 case dwarf::DW_FORM_sdata: 271 addULEB128(dwarf::DW_FORM_sdata); 272 addSLEB128((int64_t)Value.getDIEInteger().getValue()); 273 break; 274 // DW_FORM_flag_present is just flag with a value of one. We still give it a 275 // value so just use the value. 276 case dwarf::DW_FORM_flag_present: 277 case dwarf::DW_FORM_flag: 278 addULEB128(dwarf::DW_FORM_flag); 279 addULEB128((int64_t)Value.getDIEInteger().getValue()); 280 break; 281 default: 282 llvm_unreachable("Unknown integer form!"); 283 } 284 break; 285 } 286 case DIEValue::isString: 287 addULEB128('A'); 288 addULEB128(Attribute); 289 addULEB128(dwarf::DW_FORM_string); 290 addString(Value.getDIEString().getString()); 291 break; 292 case DIEValue::isInlineString: 293 addULEB128('A'); 294 addULEB128(Attribute); 295 addULEB128(dwarf::DW_FORM_string); 296 addString(Value.getDIEInlineString().getString()); 297 break; 298 case DIEValue::isBlock: 299 case DIEValue::isLoc: 300 case DIEValue::isLocList: 301 addULEB128('A'); 302 addULEB128(Attribute); 303 addULEB128(dwarf::DW_FORM_block); 304 if (Value.getType() == DIEValue::isBlock) { 305 addULEB128(Value.getDIEBlock().ComputeSize(AP)); 306 hashBlockData(Value.getDIEBlock().values()); 307 } else if (Value.getType() == DIEValue::isLoc) { 308 addULEB128(Value.getDIELoc().ComputeSize(AP)); 309 hashBlockData(Value.getDIELoc().values()); 310 } else { 311 // We could add the block length, but that would take 312 // a bit of work and not add a lot of uniqueness 313 // to the hash in some way we could test. 314 hashLocList(Value.getDIELocList()); 315 } 316 break; 317 // FIXME: It's uncertain whether or not we should handle this at the moment. 318 case DIEValue::isExpr: 319 case DIEValue::isLabel: 320 case DIEValue::isBaseTypeRef: 321 case DIEValue::isDelta: 322 llvm_unreachable("Add support for additional value types."); 323 } 324 } 325 326 // Go through the attributes from \param Attrs in the order specified in 7.27.4 327 // and hash them. 328 void DIEHash::hashAttributes(const DIEAttrs &Attrs, dwarf::Tag Tag) { 329 #define HANDLE_DIE_HASH_ATTR(NAME) \ 330 { \ 331 if (Attrs.NAME) \ 332 hashAttribute(Attrs.NAME, Tag); \ 333 } 334 #include "DIEHashAttributes.def" 335 // FIXME: Add the extended attributes. 336 } 337 338 // Add all of the attributes for \param Die to the hash. 339 void DIEHash::addAttributes(const DIE &Die) { 340 DIEAttrs Attrs = {}; 341 collectAttributes(Die, Attrs); 342 hashAttributes(Attrs, Die.getTag()); 343 } 344 345 void DIEHash::hashNestedType(const DIE &Die, StringRef Name) { 346 // 7.27 Step 7 347 // ... append the letter 'S', 348 addULEB128('S'); 349 350 // the tag of C, 351 addULEB128(Die.getTag()); 352 353 // and the name. 354 addString(Name); 355 } 356 357 // Compute the hash of a DIE. This is based on the type signature computation 358 // given in section 7.27 of the DWARF4 standard. It is the md5 hash of a 359 // flattened description of the DIE. 360 void DIEHash::computeHash(const DIE &Die) { 361 // Append the letter 'D', followed by the DWARF tag of the DIE. 362 addULEB128('D'); 363 addULEB128(Die.getTag()); 364 365 // Add each of the attributes of the DIE. 366 addAttributes(Die); 367 368 // Then hash each of the children of the DIE. 369 for (auto &C : Die.children()) { 370 // 7.27 Step 7 371 // If C is a nested type entry or a member function entry, ... 372 if (isType(C.getTag()) || (C.getTag() == dwarf::DW_TAG_subprogram && isType(C.getParent()->getTag()))) { 373 StringRef Name = getDIEStringAttr(C, dwarf::DW_AT_name); 374 // ... and has a DW_AT_name attribute 375 if (!Name.empty()) { 376 hashNestedType(C, Name); 377 continue; 378 } 379 } 380 computeHash(C); 381 } 382 383 // Following the last (or if there are no children), append a zero byte. 384 Hash.update(makeArrayRef((uint8_t)'\0')); 385 } 386 387 /// This is based on the type signature computation given in section 7.27 of the 388 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE 389 /// with the inclusion of the full CU and all top level CU entities. 390 // TODO: Initialize the type chain at 0 instead of 1 for CU signatures. 391 uint64_t DIEHash::computeCUSignature(StringRef DWOName, const DIE &Die) { 392 Numbering.clear(); 393 Numbering[&Die] = 1; 394 395 if (!DWOName.empty()) 396 Hash.update(DWOName); 397 // Hash the DIE. 398 computeHash(Die); 399 400 // Now return the result. 401 MD5::MD5Result Result; 402 Hash.final(Result); 403 404 // ... take the least significant 8 bytes and return those. Our MD5 405 // implementation always returns its results in little endian, so we actually 406 // need the "high" word. 407 return Result.high(); 408 } 409 410 /// This is based on the type signature computation given in section 7.27 of the 411 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE 412 /// with the inclusion of additional forms not specifically called out in the 413 /// standard. 414 uint64_t DIEHash::computeTypeSignature(const DIE &Die) { 415 Numbering.clear(); 416 Numbering[&Die] = 1; 417 418 if (const DIE *Parent = Die.getParent()) 419 addParentContext(*Parent); 420 421 // Hash the DIE. 422 computeHash(Die); 423 424 // Now return the result. 425 MD5::MD5Result Result; 426 Hash.final(Result); 427 428 // ... take the least significant 8 bytes and return those. Our MD5 429 // implementation always returns its results in little endian, so we actually 430 // need the "high" word. 431 return Result.high(); 432 } 433