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/CodeGen/DIE.h" 21 #include "llvm/Support/Debug.h" 22 #include "llvm/Support/Endian.h" 23 #include "llvm/Support/MD5.h" 24 #include "llvm/Support/raw_ostream.h" 25 26 using namespace llvm; 27 28 #define DEBUG_TYPE "dwarfdebug" 29 30 /// 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 // Iterate through all the attributes until we find the one we're 34 // looking for, if we can't find it return an empty string. 35 for (const auto &V : Die.values()) 36 if (V.getAttribute() == Attr) 37 return V.getDIEString().getString(); 38 39 return StringRef(""); 40 } 41 42 /// Adds the string in \p Str to the hash. This also hashes 43 /// a trailing NULL with the string. 44 void DIEHash::addString(StringRef Str) { 45 LLVM_DEBUG(dbgs() << "Adding string " << Str << " to hash.\n"); 46 Hash.update(Str); 47 Hash.update(makeArrayRef((uint8_t)'\0')); 48 } 49 50 // FIXME: The LEB128 routines are copied and only slightly modified out of 51 // LEB128.h. 52 53 /// Adds the unsigned in \p Value to the hash encoded as a ULEB128. 54 void DIEHash::addULEB128(uint64_t Value) { 55 LLVM_DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n"); 56 do { 57 uint8_t Byte = Value & 0x7f; 58 Value >>= 7; 59 if (Value != 0) 60 Byte |= 0x80; // Mark this byte to show that more bytes will follow. 61 Hash.update(Byte); 62 } while (Value != 0); 63 } 64 65 void DIEHash::addSLEB128(int64_t Value) { 66 LLVM_DEBUG(dbgs() << "Adding ULEB128 " << Value << " to hash.\n"); 67 bool More; 68 do { 69 uint8_t Byte = Value & 0x7f; 70 Value >>= 7; 71 More = !((((Value == 0) && ((Byte & 0x40) == 0)) || 72 ((Value == -1) && ((Byte & 0x40) != 0)))); 73 if (More) 74 Byte |= 0x80; // Mark this byte to show that more bytes will follow. 75 Hash.update(Byte); 76 } while (More); 77 } 78 79 /// Including \p Parent adds the context of Parent to the hash.. 80 void DIEHash::addParentContext(const DIE &Parent) { 81 82 LLVM_DEBUG(dbgs() << "Adding parent context to hash...\n"); 83 84 // [7.27.2] For each surrounding type or namespace beginning with the 85 // outermost such construct... 86 SmallVector<const DIE *, 1> Parents; 87 const DIE *Cur = &Parent; 88 while (Cur->getParent()) { 89 Parents.push_back(Cur); 90 Cur = Cur->getParent(); 91 } 92 assert(Cur->getTag() == dwarf::DW_TAG_compile_unit || 93 Cur->getTag() == dwarf::DW_TAG_type_unit); 94 95 // Reverse iterate over our list to go from the outermost construct to the 96 // innermost. 97 for (SmallVectorImpl<const DIE *>::reverse_iterator I = Parents.rbegin(), 98 E = Parents.rend(); 99 I != E; ++I) { 100 const DIE &Die = **I; 101 102 // ... Append the letter "C" to the sequence... 103 addULEB128('C'); 104 105 // ... Followed by the DWARF tag of the construct... 106 addULEB128(Die.getTag()); 107 108 // ... Then the name, taken from the DW_AT_name attribute. 109 StringRef Name = getDIEStringAttr(Die, dwarf::DW_AT_name); 110 LLVM_DEBUG(dbgs() << "... adding context: " << Name << "\n"); 111 if (!Name.empty()) 112 addString(Name); 113 } 114 } 115 116 // Collect all of the attributes for a particular DIE in single structure. 117 void DIEHash::collectAttributes(const DIE &Die, DIEAttrs &Attrs) { 118 119 for (const auto &V : Die.values()) { 120 LLVM_DEBUG(dbgs() << "Attribute: " 121 << dwarf::AttributeString(V.getAttribute()) 122 << " added.\n"); 123 switch (V.getAttribute()) { 124 #define HANDLE_DIE_HASH_ATTR(NAME) \ 125 case dwarf::NAME: \ 126 Attrs.NAME = V; \ 127 break; 128 #include "DIEHashAttributes.def" 129 default: 130 break; 131 } 132 } 133 } 134 135 void DIEHash::hashShallowTypeReference(dwarf::Attribute Attribute, 136 const DIE &Entry, StringRef Name) { 137 // append the letter 'N' 138 addULEB128('N'); 139 140 // the DWARF attribute code (DW_AT_type or DW_AT_friend), 141 addULEB128(Attribute); 142 143 // the context of the tag, 144 if (const DIE *Parent = Entry.getParent()) 145 addParentContext(*Parent); 146 147 // the letter 'E', 148 addULEB128('E'); 149 150 // and the name of the type. 151 addString(Name); 152 153 // Currently DW_TAG_friends are not used by Clang, but if they do become so, 154 // here's the relevant spec text to implement: 155 // 156 // For DW_TAG_friend, if the referenced entry is the DW_TAG_subprogram, 157 // the context is omitted and the name to be used is the ABI-specific name 158 // of the subprogram (e.g., the mangled linker name). 159 } 160 161 void DIEHash::hashRepeatedTypeReference(dwarf::Attribute Attribute, 162 unsigned DieNumber) { 163 // a) If T is in the list of [previously hashed types], use the letter 164 // 'R' as the marker 165 addULEB128('R'); 166 167 addULEB128(Attribute); 168 169 // and use the unsigned LEB128 encoding of [the index of T in the 170 // list] as the attribute value; 171 addULEB128(DieNumber); 172 } 173 174 void DIEHash::hashDIEEntry(dwarf::Attribute Attribute, dwarf::Tag Tag, 175 const DIE &Entry) { 176 assert(Tag != dwarf::DW_TAG_friend && "No current LLVM clients emit friend " 177 "tags. Add support here when there's " 178 "a use case"); 179 // Step 5 180 // If the tag in Step 3 is one of [the below tags] 181 if ((Tag == dwarf::DW_TAG_pointer_type || 182 Tag == dwarf::DW_TAG_reference_type || 183 Tag == dwarf::DW_TAG_rvalue_reference_type || 184 Tag == dwarf::DW_TAG_ptr_to_member_type) && 185 // and the referenced type (via the [below attributes]) 186 // FIXME: This seems overly restrictive, and causes hash mismatches 187 // there's a decl/def difference in the containing type of a 188 // ptr_to_member_type, but it's what DWARF says, for some reason. 189 Attribute == dwarf::DW_AT_type) { 190 // ... has a DW_AT_name attribute, 191 StringRef Name = getDIEStringAttr(Entry, dwarf::DW_AT_name); 192 if (!Name.empty()) { 193 hashShallowTypeReference(Attribute, Entry, Name); 194 return; 195 } 196 } 197 198 unsigned &DieNumber = Numbering[&Entry]; 199 if (DieNumber) { 200 hashRepeatedTypeReference(Attribute, DieNumber); 201 return; 202 } 203 204 // otherwise, b) use the letter 'T' as the marker, ... 205 addULEB128('T'); 206 207 addULEB128(Attribute); 208 209 // ... process the type T recursively by performing Steps 2 through 7, and 210 // use the result as the attribute value. 211 DieNumber = Numbering.size(); 212 computeHash(Entry); 213 } 214 215 // Hash all of the values in a block like set of values. This assumes that 216 // all of the data is going to be added as integers. 217 void DIEHash::hashBlockData(const DIE::const_value_range &Values) { 218 for (const auto &V : Values) 219 Hash.update((uint64_t)V.getDIEInteger().getValue()); 220 } 221 222 // Hash the contents of a loclistptr class. 223 void DIEHash::hashLocList(const DIELocList &LocList) { 224 HashingByteStreamer Streamer(*this); 225 DwarfDebug &DD = *AP->getDwarfDebug(); 226 const DebugLocStream &Locs = DD.getDebugLocs(); 227 for (const auto &Entry : Locs.getEntries(Locs.getList(LocList.getValue()))) 228 DD.emitDebugLocEntry(Streamer, Entry, nullptr); 229 } 230 231 // Hash an individual attribute \param Attr based on the type of attribute and 232 // the form. 233 void DIEHash::hashAttribute(const DIEValue &Value, dwarf::Tag Tag) { 234 dwarf::Attribute Attribute = Value.getAttribute(); 235 236 // Other attribute values use the letter 'A' as the marker, and the value 237 // consists of the form code (encoded as an unsigned LEB128 value) followed by 238 // the encoding of the value according to the form code. To ensure 239 // reproducibility of the signature, the set of forms used in the signature 240 // computation is limited to the following: DW_FORM_sdata, DW_FORM_flag, 241 // DW_FORM_string, and DW_FORM_block. 242 243 switch (Value.getType()) { 244 case DIEValue::isNone: 245 llvm_unreachable("Expected valid DIEValue"); 246 247 // 7.27 Step 3 248 // ... An attribute that refers to another type entry T is processed as 249 // follows: 250 case DIEValue::isEntry: 251 hashDIEEntry(Attribute, Tag, Value.getDIEEntry().getEntry()); 252 break; 253 case DIEValue::isInteger: { 254 addULEB128('A'); 255 addULEB128(Attribute); 256 switch (Value.getForm()) { 257 case dwarf::DW_FORM_data1: 258 case dwarf::DW_FORM_data2: 259 case dwarf::DW_FORM_data4: 260 case dwarf::DW_FORM_data8: 261 case dwarf::DW_FORM_udata: 262 case dwarf::DW_FORM_sdata: 263 addULEB128(dwarf::DW_FORM_sdata); 264 addSLEB128((int64_t)Value.getDIEInteger().getValue()); 265 break; 266 // DW_FORM_flag_present is just flag with a value of one. We still give it a 267 // value so just use the value. 268 case dwarf::DW_FORM_flag_present: 269 case dwarf::DW_FORM_flag: 270 addULEB128(dwarf::DW_FORM_flag); 271 addULEB128((int64_t)Value.getDIEInteger().getValue()); 272 break; 273 default: 274 llvm_unreachable("Unknown integer form!"); 275 } 276 break; 277 } 278 case DIEValue::isString: 279 addULEB128('A'); 280 addULEB128(Attribute); 281 addULEB128(dwarf::DW_FORM_string); 282 addString(Value.getDIEString().getString()); 283 break; 284 case DIEValue::isInlineString: 285 addULEB128('A'); 286 addULEB128(Attribute); 287 addULEB128(dwarf::DW_FORM_string); 288 addString(Value.getDIEInlineString().getString()); 289 break; 290 case DIEValue::isBlock: 291 case DIEValue::isLoc: 292 case DIEValue::isLocList: 293 addULEB128('A'); 294 addULEB128(Attribute); 295 addULEB128(dwarf::DW_FORM_block); 296 if (Value.getType() == DIEValue::isBlock) { 297 addULEB128(Value.getDIEBlock().ComputeSize(AP)); 298 hashBlockData(Value.getDIEBlock().values()); 299 } else if (Value.getType() == DIEValue::isLoc) { 300 addULEB128(Value.getDIELoc().ComputeSize(AP)); 301 hashBlockData(Value.getDIELoc().values()); 302 } else { 303 // We could add the block length, but that would take 304 // a bit of work and not add a lot of uniqueness 305 // to the hash in some way we could test. 306 hashLocList(Value.getDIELocList()); 307 } 308 break; 309 // FIXME: It's uncertain whether or not we should handle this at the moment. 310 case DIEValue::isExpr: 311 case DIEValue::isLabel: 312 case DIEValue::isBaseTypeRef: 313 case DIEValue::isDelta: 314 llvm_unreachable("Add support for additional value types."); 315 } 316 } 317 318 // Go through the attributes from \param Attrs in the order specified in 7.27.4 319 // and hash them. 320 void DIEHash::hashAttributes(const DIEAttrs &Attrs, dwarf::Tag Tag) { 321 #define HANDLE_DIE_HASH_ATTR(NAME) \ 322 { \ 323 if (Attrs.NAME) \ 324 hashAttribute(Attrs.NAME, Tag); \ 325 } 326 #include "DIEHashAttributes.def" 327 // FIXME: Add the extended attributes. 328 } 329 330 // Add all of the attributes for \param Die to the hash. 331 void DIEHash::addAttributes(const DIE &Die) { 332 DIEAttrs Attrs = {}; 333 collectAttributes(Die, Attrs); 334 hashAttributes(Attrs, Die.getTag()); 335 } 336 337 void DIEHash::hashNestedType(const DIE &Die, StringRef Name) { 338 // 7.27 Step 7 339 // ... append the letter 'S', 340 addULEB128('S'); 341 342 // the tag of C, 343 addULEB128(Die.getTag()); 344 345 // and the name. 346 addString(Name); 347 } 348 349 // Compute the hash of a DIE. This is based on the type signature computation 350 // given in section 7.27 of the DWARF4 standard. It is the md5 hash of a 351 // flattened description of the DIE. 352 void DIEHash::computeHash(const DIE &Die) { 353 // Append the letter 'D', followed by the DWARF tag of the DIE. 354 addULEB128('D'); 355 addULEB128(Die.getTag()); 356 357 // Add each of the attributes of the DIE. 358 addAttributes(Die); 359 360 // Then hash each of the children of the DIE. 361 for (auto &C : Die.children()) { 362 // 7.27 Step 7 363 // If C is a nested type entry or a member function entry, ... 364 if (isType(C.getTag()) || C.getTag() == dwarf::DW_TAG_subprogram) { 365 StringRef Name = getDIEStringAttr(C, dwarf::DW_AT_name); 366 // ... and has a DW_AT_name attribute 367 if (!Name.empty()) { 368 hashNestedType(C, Name); 369 continue; 370 } 371 } 372 computeHash(C); 373 } 374 375 // Following the last (or if there are no children), append a zero byte. 376 Hash.update(makeArrayRef((uint8_t)'\0')); 377 } 378 379 /// This is based on the type signature computation given in section 7.27 of the 380 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE 381 /// with the inclusion of the full CU and all top level CU entities. 382 // TODO: Initialize the type chain at 0 instead of 1 for CU signatures. 383 uint64_t DIEHash::computeCUSignature(StringRef DWOName, const DIE &Die) { 384 Numbering.clear(); 385 Numbering[&Die] = 1; 386 387 if (!DWOName.empty()) 388 Hash.update(DWOName); 389 // Hash the DIE. 390 computeHash(Die); 391 392 // Now return the result. 393 MD5::MD5Result Result; 394 Hash.final(Result); 395 396 // ... take the least significant 8 bytes and return those. Our MD5 397 // implementation always returns its results in little endian, so we actually 398 // need the "high" word. 399 return Result.high(); 400 } 401 402 /// This is based on the type signature computation given in section 7.27 of the 403 /// DWARF4 standard. It is an md5 hash of the flattened description of the DIE 404 /// with the inclusion of additional forms not specifically called out in the 405 /// standard. 406 uint64_t DIEHash::computeTypeSignature(const DIE &Die) { 407 Numbering.clear(); 408 Numbering[&Die] = 1; 409 410 if (const DIE *Parent = Die.getParent()) 411 addParentContext(*Parent); 412 413 // Hash the DIE. 414 computeHash(Die); 415 416 // Now return the result. 417 MD5::MD5Result Result; 418 Hash.final(Result); 419 420 // ... take the least significant 8 bytes and return those. Our MD5 421 // implementation always returns its results in little endian, so we actually 422 // need the "high" word. 423 return Result.high(); 424 } 425