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 "DwarfDebug.h" 16 #include "llvm/ADT/ArrayRef.h" 17 #include "llvm/ADT/StringRef.h" 18 #include "llvm/BinaryFormat/Dwarf.h" 19 #include "llvm/CodeGen/AsmPrinter.h" 20 #include "llvm/Support/Debug.h" 21 #include "llvm/Support/Endian.h" 22 #include "llvm/Support/raw_ostream.h" 23 24 using namespace llvm; 25 26 #define DEBUG_TYPE "dwarfdebug" 27 28 /// 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 // Iterate through all the attributes until we find the one we're 32 // looking for, if we can't find it return an empty string. 33 for (const auto &V : Die.values()) 34 if (V.getAttribute() == Attr) 35 return V.getDIEString().getString(); 36 37 return StringRef(""); 38 } 39 40 /// Adds the string in \p Str to the hash. This also hashes 41 /// a trailing NULL with the string. 42 void DIEHash::addString(StringRef Str) { 43 LLVM_DEBUG(dbgs() << "Adding string " << Str << " to hash.\n"); 44 Hash.update(Str); 45 Hash.update(makeArrayRef((uint8_t)'\0')); 46 } 47 48 // FIXME: The LEB128 routines are copied and only slightly modified out of 49 // LEB128.h. 50 51 /// Adds the unsigned in \p Value to the hash encoded as a ULEB128. 52 void DIEHash::addULEB128(uint64_t Value) { 53 LLVM_DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n"); 54 do { 55 uint8_t Byte = Value & 0x7f; 56 Value >>= 7; 57 if (Value != 0) 58 Byte |= 0x80; // Mark this byte to show that more bytes will follow. 59 Hash.update(Byte); 60 } while (Value != 0); 61 } 62 63 void DIEHash::addSLEB128(int64_t Value) { 64 LLVM_DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n"); 65 bool More; 66 do { 67 uint8_t Byte = Value & 0x7f; 68 Value >>= 7; 69 More = !((((Value == 0) && ((Byte & 0x40) == 0)) || 70 ((Value == -1) && ((Byte & 0x40) != 0)))); 71 if (More) 72 Byte |= 0x80; // Mark this byte to show that more bytes will follow. 73 Hash.update(Byte); 74 } while (More); 75 } 76 77 /// Including \p Parent adds the context of Parent to the hash.. 78 void DIEHash::addParentContext(const DIE &Parent) { 79 80 LLVM_DEBUG(dbgs() << "Adding parent context to hash...\n"); 81 82 // [7.27.2] For each surrounding type or namespace beginning with the 83 // outermost such construct... 84 SmallVector<const DIE *, 1> Parents; 85 const DIE *Cur = &Parent; 86 while (Cur->getParent()) { 87 Parents.push_back(Cur); 88 Cur = Cur->getParent(); 89 } 90 assert(Cur->getTag() == dwarf::DW_TAG_compile_unit || 91 Cur->getTag() == dwarf::DW_TAG_type_unit); 92 93 // Reverse iterate over our list to go from the outermost construct to the 94 // innermost. 95 for (SmallVectorImpl<const DIE *>::reverse_iterator I = Parents.rbegin(), 96 E = Parents.rend(); 97 I != E; ++I) { 98 const DIE &Die = **I; 99 100 // ... Append the letter "C" to the sequence... 101 addULEB128('C'); 102 103 // ... Followed by the DWARF tag of the construct... 104 addULEB128(Die.getTag()); 105 106 // ... Then the name, taken from the DW_AT_name attribute. 107 StringRef Name = getDIEStringAttr(Die, dwarf::DW_AT_name); 108 LLVM_DEBUG(dbgs() << "... adding context: " << Name << "\n"); 109 if (!Name.empty()) 110 addString(Name); 111 } 112 } 113 114 // Collect all of the attributes for a particular DIE in single structure. 115 void DIEHash::collectAttributes(const DIE &Die, DIEAttrs &Attrs) { 116 117 for (const auto &V : Die.values()) { 118 LLVM_DEBUG(dbgs() << "Attribute: " 119 << dwarf::AttributeString(V.getAttribute()) 120 << " added.\n"); 121 switch (V.getAttribute()) { 122 #define HANDLE_DIE_HASH_ATTR(NAME) \ 123 case dwarf::NAME: \ 124 Attrs.NAME = V; \ 125 break; 126 #include "DIEHashAttributes.def" 127 default: 128 break; 129 } 130 } 131 } 132 133 void DIEHash::hashShallowTypeReference(dwarf::Attribute Attribute, 134 const DIE &Entry, StringRef Name) { 135 // append the letter 'N' 136 addULEB128('N'); 137 138 // the DWARF attribute code (DW_AT_type or DW_AT_friend), 139 addULEB128(Attribute); 140 141 // the context of the tag, 142 if (const DIE *Parent = Entry.getParent()) 143 addParentContext(*Parent); 144 145 // the letter 'E', 146 addULEB128('E'); 147 148 // and the name of the type. 149 addString(Name); 150 151 // Currently DW_TAG_friends are not used by Clang, but if they do become so, 152 // here's the relevant spec text to implement: 153 // 154 // For DW_TAG_friend, if the referenced entry is the DW_TAG_subprogram, 155 // the context is omitted and the name to be used is the ABI-specific name 156 // of the subprogram (e.g., the mangled linker name). 157 } 158 159 void DIEHash::hashRepeatedTypeReference(dwarf::Attribute Attribute, 160 unsigned DieNumber) { 161 // a) If T is in the list of [previously hashed types], use the letter 162 // 'R' as the marker 163 addULEB128('R'); 164 165 addULEB128(Attribute); 166 167 // and use the unsigned LEB128 encoding of [the index of T in the 168 // list] as the attribute value; 169 addULEB128(DieNumber); 170 } 171 172 void DIEHash::hashDIEEntry(dwarf::Attribute Attribute, dwarf::Tag Tag, 173 const DIE &Entry) { 174 assert(Tag != dwarf::DW_TAG_friend && "No current LLVM clients emit friend " 175 "tags. Add support here when there's " 176 "a use case"); 177 // Step 5 178 // If the tag in Step 3 is one of [the below tags] 179 if ((Tag == dwarf::DW_TAG_pointer_type || 180 Tag == dwarf::DW_TAG_reference_type || 181 Tag == dwarf::DW_TAG_rvalue_reference_type || 182 Tag == dwarf::DW_TAG_ptr_to_member_type) && 183 // and the referenced type (via the [below attributes]) 184 // FIXME: This seems overly restrictive, and causes hash mismatches 185 // there's a decl/def difference in the containing type of a 186 // ptr_to_member_type, but it's what DWARF says, for some reason. 187 Attribute == dwarf::DW_AT_type) { 188 // ... has a DW_AT_name attribute, 189 StringRef Name = getDIEStringAttr(Entry, dwarf::DW_AT_name); 190 if (!Name.empty()) { 191 hashShallowTypeReference(Attribute, Entry, Name); 192 return; 193 } 194 } 195 196 unsigned &DieNumber = Numbering[&Entry]; 197 if (DieNumber) { 198 hashRepeatedTypeReference(Attribute, DieNumber); 199 return; 200 } 201 202 // otherwise, b) use the letter 'T' as the marker, ... 203 addULEB128('T'); 204 205 addULEB128(Attribute); 206 207 // ... process the type T recursively by performing Steps 2 through 7, and 208 // use the result as the attribute value. 209 DieNumber = Numbering.size(); 210 computeHash(Entry); 211 } 212 213 // Hash all of the values in a block like set of values. This assumes that 214 // all of the data is going to be added as integers. 215 void DIEHash::hashBlockData(const DIE::const_value_range &Values) { 216 for (const auto &V : Values) 217 Hash.update((uint64_t)V.getDIEInteger().getValue()); 218 } 219 220 // Hash the contents of a loclistptr class. 221 void DIEHash::hashLocList(const DIELocList &LocList) { 222 HashingByteStreamer Streamer(*this); 223 DwarfDebug &DD = *AP->getDwarfDebug(); 224 const DebugLocStream &Locs = DD.getDebugLocs(); 225 const DebugLocStream::List &List = Locs.getList(LocList.getValue()); 226 for (const DebugLocStream::Entry &Entry : Locs.getEntries(List)) 227 DD.emitDebugLocEntry(Streamer, Entry, List.CU); 228 } 229 230 // Hash an individual attribute \param Attr based on the type of attribute and 231 // the form. 232 void DIEHash::hashAttribute(const DIEValue &Value, dwarf::Tag Tag) { 233 dwarf::Attribute Attribute = Value.getAttribute(); 234 235 // Other attribute values use the letter 'A' as the marker, and the value 236 // consists of the form code (encoded as an unsigned LEB128 value) followed by 237 // the encoding of the value according to the form code. To ensure 238 // reproducibility of the signature, the set of forms used in the signature 239 // computation is limited to the following: DW_FORM_sdata, DW_FORM_flag, 240 // DW_FORM_string, and DW_FORM_block. 241 242 switch (Value.getType()) { 243 case DIEValue::isNone: 244 llvm_unreachable("Expected valid DIEValue"); 245 246 // 7.27 Step 3 247 // ... An attribute that refers to another type entry T is processed as 248 // follows: 249 case DIEValue::isEntry: 250 hashDIEEntry(Attribute, Tag, Value.getDIEEntry().getEntry()); 251 break; 252 case DIEValue::isInteger: { 253 addULEB128('A'); 254 addULEB128(Attribute); 255 switch (Value.getForm()) { 256 case dwarf::DW_FORM_data1: 257 case dwarf::DW_FORM_data2: 258 case dwarf::DW_FORM_data4: 259 case dwarf::DW_FORM_data8: 260 case dwarf::DW_FORM_udata: 261 case dwarf::DW_FORM_sdata: 262 addULEB128(dwarf::DW_FORM_sdata); 263 addSLEB128((int64_t)Value.getDIEInteger().getValue()); 264 break; 265 // DW_FORM_flag_present is just flag with a value of one. We still give it a 266 // value so just use the value. 267 case dwarf::DW_FORM_flag_present: 268 case dwarf::DW_FORM_flag: 269 addULEB128(dwarf::DW_FORM_flag); 270 addULEB128((int64_t)Value.getDIEInteger().getValue()); 271 break; 272 default: 273 llvm_unreachable("Unknown integer form!"); 274 } 275 break; 276 } 277 case DIEValue::isString: 278 addULEB128('A'); 279 addULEB128(Attribute); 280 addULEB128(dwarf::DW_FORM_string); 281 addString(Value.getDIEString().getString()); 282 break; 283 case DIEValue::isInlineString: 284 addULEB128('A'); 285 addULEB128(Attribute); 286 addULEB128(dwarf::DW_FORM_string); 287 addString(Value.getDIEInlineString().getString()); 288 break; 289 case DIEValue::isBlock: 290 case DIEValue::isLoc: 291 case DIEValue::isLocList: 292 addULEB128('A'); 293 addULEB128(Attribute); 294 addULEB128(dwarf::DW_FORM_block); 295 if (Value.getType() == DIEValue::isBlock) { 296 addULEB128(Value.getDIEBlock().ComputeSize(AP)); 297 hashBlockData(Value.getDIEBlock().values()); 298 } else if (Value.getType() == DIEValue::isLoc) { 299 addULEB128(Value.getDIELoc().ComputeSize(AP)); 300 hashBlockData(Value.getDIELoc().values()); 301 } else { 302 // We could add the block length, but that would take 303 // a bit of work and not add a lot of uniqueness 304 // to the hash in some way we could test. 305 hashLocList(Value.getDIELocList()); 306 } 307 break; 308 // FIXME: It's uncertain whether or not we should handle this at the moment. 309 case DIEValue::isExpr: 310 case DIEValue::isLabel: 311 case DIEValue::isBaseTypeRef: 312 case DIEValue::isDelta: 313 llvm_unreachable("Add support for additional value types."); 314 } 315 } 316 317 // Go through the attributes from \param Attrs in the order specified in 7.27.4 318 // and hash them. 319 void DIEHash::hashAttributes(const DIEAttrs &Attrs, dwarf::Tag Tag) { 320 #define HANDLE_DIE_HASH_ATTR(NAME) \ 321 { \ 322 if (Attrs.NAME) \ 323 hashAttribute(Attrs.NAME, Tag); \ 324 } 325 #include "DIEHashAttributes.def" 326 // FIXME: Add the extended attributes. 327 } 328 329 // Add all of the attributes for \param Die to the hash. 330 void DIEHash::addAttributes(const DIE &Die) { 331 DIEAttrs Attrs = {}; 332 collectAttributes(Die, Attrs); 333 hashAttributes(Attrs, Die.getTag()); 334 } 335 336 void DIEHash::hashNestedType(const DIE &Die, StringRef Name) { 337 // 7.27 Step 7 338 // ... append the letter 'S', 339 addULEB128('S'); 340 341 // the tag of C, 342 addULEB128(Die.getTag()); 343 344 // and the name. 345 addString(Name); 346 } 347 348 // Compute the hash of a DIE. This is based on the type signature computation 349 // given in section 7.27 of the DWARF4 standard. It is the md5 hash of a 350 // flattened description of the DIE. 351 void DIEHash::computeHash(const DIE &Die) { 352 // Append the letter 'D', followed by the DWARF tag of the DIE. 353 addULEB128('D'); 354 addULEB128(Die.getTag()); 355 356 // Add each of the attributes of the DIE. 357 addAttributes(Die); 358 359 // Then hash each of the children of the DIE. 360 for (auto &C : Die.children()) { 361 // 7.27 Step 7 362 // If C is a nested type entry or a member function entry, ... 363 if (isType(C.getTag()) || (C.getTag() == dwarf::DW_TAG_subprogram && isType(C.getParent()->getTag()))) { 364 StringRef Name = getDIEStringAttr(C, dwarf::DW_AT_name); 365 // ... and has a DW_AT_name attribute 366 if (!Name.empty()) { 367 hashNestedType(C, Name); 368 continue; 369 } 370 } 371 computeHash(C); 372 } 373 374 // Following the last (or if there are no children), append a zero byte. 375 Hash.update(makeArrayRef((uint8_t)'\0')); 376 } 377 378 /// This is based on the type signature computation given in section 7.27 of the 379 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE 380 /// with the inclusion of the full CU and all top level CU entities. 381 // TODO: Initialize the type chain at 0 instead of 1 for CU signatures. 382 uint64_t DIEHash::computeCUSignature(StringRef DWOName, const DIE &Die) { 383 Numbering.clear(); 384 Numbering[&Die] = 1; 385 386 if (!DWOName.empty()) 387 Hash.update(DWOName); 388 // Hash the DIE. 389 computeHash(Die); 390 391 // Now return the result. 392 MD5::MD5Result Result; 393 Hash.final(Result); 394 395 // ... take the least significant 8 bytes and return those. Our MD5 396 // implementation always returns its results in little endian, so we actually 397 // need the "high" word. 398 return Result.high(); 399 } 400 401 /// This is based on the type signature computation given in section 7.27 of the 402 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE 403 /// with the inclusion of additional forms not specifically called out in the 404 /// standard. 405 uint64_t DIEHash::computeTypeSignature(const DIE &Die) { 406 Numbering.clear(); 407 Numbering[&Die] = 1; 408 409 if (const DIE *Parent = Die.getParent()) 410 addParentContext(*Parent); 411 412 // Hash the DIE. 413 computeHash(Die); 414 415 // Now return the result. 416 MD5::MD5Result Result; 417 Hash.final(Result); 418 419 // ... take the least significant 8 bytes and return those. Our MD5 420 // implementation always returns its results in little endian, so we actually 421 // need the "high" word. 422 return Result.high(); 423 } 424