1 // Copyright 2012 the V8 project authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style license that can be 3 // found in the LICENSE file. 4 5 #ifndef V8_REGEXP_X64_REGEXP_MACRO_ASSEMBLER_X64_H_ 6 #define V8_REGEXP_X64_REGEXP_MACRO_ASSEMBLER_X64_H_ 7 8 #include "src/codegen/macro-assembler.h" 9 #include "src/regexp/regexp-macro-assembler.h" 10 #include "src/zone/zone-chunk-list.h" 11 12 namespace v8 { 13 namespace internal { 14 15 class V8_EXPORT_PRIVATE RegExpMacroAssemblerX64 16 : public NativeRegExpMacroAssembler { 17 public: 18 RegExpMacroAssemblerX64(Isolate* isolate, Zone* zone, Mode mode, 19 int registers_to_save); 20 ~RegExpMacroAssemblerX64() override; 21 int stack_limit_slack() override; 22 void AdvanceCurrentPosition(int by) override; 23 void AdvanceRegister(int reg, int by) override; 24 void Backtrack() override; 25 void Bind(Label* label) override; 26 void CheckAtStart(int cp_offset, Label* on_at_start) override; 27 void CheckCharacter(uint32_t c, Label* on_equal) override; 28 void CheckCharacterAfterAnd(uint32_t c, uint32_t mask, 29 Label* on_equal) override; 30 void CheckCharacterGT(base::uc16 limit, Label* on_greater) override; 31 void CheckCharacterLT(base::uc16 limit, Label* on_less) override; 32 // A "greedy loop" is a loop that is both greedy and with a simple 33 // body. It has a particularly simple implementation. 34 void CheckGreedyLoop(Label* on_tos_equals_current_position) override; 35 void CheckNotAtStart(int cp_offset, Label* on_not_at_start) override; 36 void CheckNotBackReference(int start_reg, bool read_backward, 37 Label* on_no_match) override; 38 void CheckNotBackReferenceIgnoreCase(int start_reg, bool read_backward, 39 bool unicode, 40 Label* on_no_match) override; 41 void CheckNotCharacter(uint32_t c, Label* on_not_equal) override; 42 void CheckNotCharacterAfterAnd(uint32_t c, uint32_t mask, 43 Label* on_not_equal) override; 44 void CheckNotCharacterAfterMinusAnd(base::uc16 c, base::uc16 minus, 45 base::uc16 mask, 46 Label* on_not_equal) override; 47 void CheckCharacterInRange(base::uc16 from, base::uc16 to, 48 Label* on_in_range) override; 49 void CheckCharacterNotInRange(base::uc16 from, base::uc16 to, 50 Label* on_not_in_range) override; 51 bool CheckCharacterInRangeArray(const ZoneList<CharacterRange>* ranges, 52 Label* on_in_range) override; 53 bool CheckCharacterNotInRangeArray(const ZoneList<CharacterRange>* ranges, 54 Label* on_not_in_range) override; 55 void CheckBitInTable(Handle<ByteArray> table, Label* on_bit_set) override; 56 57 // Checks whether the given offset from the current position is before 58 // the end of the string. 59 void CheckPosition(int cp_offset, Label* on_outside_input) override; 60 bool CheckSpecialCharacterClass(StandardCharacterSet type, 61 Label* on_no_match) override; 62 void Fail() override; 63 Handle<HeapObject> GetCode(Handle<String> source) override; 64 void GoTo(Label* label) override; 65 void IfRegisterGE(int reg, int comparand, Label* if_ge) override; 66 void IfRegisterLT(int reg, int comparand, Label* if_lt) override; 67 void IfRegisterEqPos(int reg, Label* if_eq) override; 68 IrregexpImplementation Implementation() override; 69 void LoadCurrentCharacterUnchecked(int cp_offset, 70 int character_count) override; 71 void PopCurrentPosition() override; 72 void PopRegister(int register_index) override; 73 void PushBacktrack(Label* label) override; 74 void PushCurrentPosition() override; 75 void PushRegister(int register_index, 76 StackCheckFlag check_stack_limit) override; 77 void ReadCurrentPositionFromRegister(int reg) override; 78 void ReadStackPointerFromRegister(int reg) override; 79 void SetCurrentPositionFromEnd(int by) override; 80 void SetRegister(int register_index, int to) override; 81 bool Succeed() override; 82 void WriteCurrentPositionToRegister(int reg, int cp_offset) override; 83 void ClearRegisters(int reg_from, int reg_to) override; 84 void WriteStackPointerToRegister(int reg) override; 85 86 // Called from RegExp if the stack-guard is triggered. 87 // If the code object is relocated, the return address is fixed before 88 // returning. 89 // {raw_code} is an Address because this is called via ExternalReference. 90 static int CheckStackGuardState(Address* return_address, Address raw_code, 91 Address re_frame); 92 93 private: 94 // Offsets from rbp of function parameters and stored registers. 95 static const int kFramePointer = 0; 96 // Above the frame pointer - function parameters and return address. 97 static const int kReturn_eip = kFramePointer + kSystemPointerSize; 98 static const int kFrameAlign = kReturn_eip + kSystemPointerSize; 99 100 #ifdef V8_TARGET_OS_WIN 101 // Parameters (first four passed as registers, but with room on stack). 102 // In Microsoft 64-bit Calling Convention, there is room on the callers 103 // stack (before the return address) to spill parameter registers. We 104 // use this space to store the register passed parameters. 105 static const int kInputString = kFrameAlign; 106 // StartIndex is passed as 32 bit int. 107 static const int kStartIndex = kInputString + kSystemPointerSize; 108 static const int kInputStart = kStartIndex + kSystemPointerSize; 109 static const int kInputEnd = kInputStart + kSystemPointerSize; 110 static const int kRegisterOutput = kInputEnd + kSystemPointerSize; 111 // For the case of global regular expression, we have room to store at least 112 // one set of capture results. For the case of non-global regexp, we ignore 113 // this value. NumOutputRegisters is passed as 32-bit value. The upper 114 // 32 bit of this 64-bit stack slot may contain garbage. 115 static const int kNumOutputRegisters = kRegisterOutput + kSystemPointerSize; 116 // DirectCall is passed as 32 bit int (values 0 or 1). 117 static const int kDirectCall = kNumOutputRegisters + kSystemPointerSize; 118 static const int kIsolate = kDirectCall + kSystemPointerSize; 119 #else 120 // In AMD64 ABI Calling Convention, the first six integer parameters 121 // are passed as registers, and caller must allocate space on the stack 122 // if it wants them stored. We push the parameters after the frame pointer. 123 static const int kInputString = kFramePointer - kSystemPointerSize; 124 static const int kStartIndex = kInputString - kSystemPointerSize; 125 static const int kInputStart = kStartIndex - kSystemPointerSize; 126 static const int kInputEnd = kInputStart - kSystemPointerSize; 127 static const int kRegisterOutput = kInputEnd - kSystemPointerSize; 128 // For the case of global regular expression, we have room to store at least 129 // one set of capture results. For the case of non-global regexp, we ignore 130 // this value. 131 static const int kNumOutputRegisters = kRegisterOutput - kSystemPointerSize; 132 133 static const int kDirectCall = kFrameAlign; 134 static const int kIsolate = kDirectCall + kSystemPointerSize; 135 #endif 136 137 // We push callee-save registers that we use after the frame pointer (and 138 // after the parameters). 139 #ifdef V8_TARGET_OS_WIN 140 static const int kBackup_rsi = kFramePointer - kSystemPointerSize; 141 static const int kBackup_rdi = kBackup_rsi - kSystemPointerSize; 142 static const int kBackup_rbx = kBackup_rdi - kSystemPointerSize; 143 static const int kNumCalleeSaveRegisters = 3; 144 static const int kLastCalleeSaveRegister = kBackup_rbx; 145 #else 146 static const int kBackup_rbx = kNumOutputRegisters - kSystemPointerSize; 147 static const int kNumCalleeSaveRegisters = 1; 148 static const int kLastCalleeSaveRegister = kBackup_rbx; 149 #endif 150 151 // When adding local variables remember to push space for them in 152 // the frame in GetCode. 153 static const int kSuccessfulCaptures = 154 kLastCalleeSaveRegister - kSystemPointerSize; 155 static const int kStringStartMinusOne = 156 kSuccessfulCaptures - kSystemPointerSize; 157 static const int kBacktrackCount = kStringStartMinusOne - kSystemPointerSize; 158 // Stores the initial value of the regexp stack pointer in a 159 // position-independent representation (in case the regexp stack grows and 160 // thus moves). 161 static const int kRegExpStackBasePointer = 162 kBacktrackCount - kSystemPointerSize; 163 164 // First register address. Following registers are below it on the stack. 165 static const int kRegisterZero = kRegExpStackBasePointer - kSystemPointerSize; 166 167 // Initial size of code buffer. 168 static const int kRegExpCodeSize = 1024; 169 170 void PushCallerSavedRegisters(); 171 void PopCallerSavedRegisters(); 172 173 // Check whether preemption has been requested. 174 void CheckPreemption(); 175 176 // Check whether we are exceeding the stack limit on the backtrack stack. 177 void CheckStackLimit(); 178 179 void CallCheckStackGuardState(); 180 void CallIsCharacterInRangeArray(const ZoneList<CharacterRange>* ranges); 181 182 // The rbp-relative location of a regexp register. 183 Operand register_location(int register_index); 184 185 // The register containing the current character after LoadCurrentCharacter. current_character()186 static constexpr Register current_character() { return rdx; } 187 188 // The register containing the backtrack stack top. Provides a meaningful 189 // name to the register. backtrack_stackpointer()190 static constexpr Register backtrack_stackpointer() { return rcx; } 191 192 // The registers containing a self pointer to this code's Code object. code_object_pointer()193 static constexpr Register code_object_pointer() { return r8; } 194 195 // Byte size of chars in the string to match (decided by the Mode argument) char_size()196 inline int char_size() { return static_cast<int>(mode_); } 197 198 // Equivalent to a conditional branch to the label, unless the label 199 // is nullptr, in which case it is a conditional Backtrack. 200 void BranchOrBacktrack(Condition condition, Label* to); 201 MarkPositionForCodeRelativeFixup()202 void MarkPositionForCodeRelativeFixup() { 203 code_relative_fixup_positions_.push_back(masm_.pc_offset()); 204 } 205 206 void FixupCodeRelativePositions(); 207 208 // Call and return internally in the generated code in a way that 209 // is GC-safe (i.e., doesn't leave absolute code addresses on the stack) 210 inline void SafeCall(Label* to); 211 inline void SafeCallTarget(Label* label); 212 inline void SafeReturn(); 213 214 // Pushes the value of a register on the backtrack stack. Decrements the 215 // stack pointer (rcx) by a word size and stores the register's value there. 216 inline void Push(Register source); 217 218 // Pushes a value on the backtrack stack. Decrements the stack pointer (rcx) 219 // by a word size and stores the value there. 220 inline void Push(Immediate value); 221 222 // Pushes the Code object relative offset of a label on the backtrack stack 223 // (i.e., a backtrack target). Decrements the stack pointer (rcx) 224 // by a word size and stores the value there. 225 inline void Push(Label* label); 226 227 // Pops a value from the backtrack stack. Reads the word at the stack pointer 228 // (rcx) and increments it by a word size. 229 inline void Pop(Register target); 230 231 // Drops the top value from the backtrack stack without reading it. 232 // Increments the stack pointer (rcx) by a word size. 233 inline void Drop(); 234 235 void LoadRegExpStackPointerFromMemory(Register dst); 236 void StoreRegExpStackPointerToMemory(Register src, Register scratch); 237 void PushRegExpBasePointer(Register scratch_pointer, Register scratch); 238 void PopRegExpBasePointer(Register scratch_pointer_out, Register scratch); 239 240 inline void ReadPositionFromRegister(Register dst, int reg); 241 isolate()242 Isolate* isolate() const { return masm_.isolate(); } 243 244 MacroAssembler masm_; 245 246 // On x64, there is no reason to keep the kRootRegister uninitialized; we 247 // could easily use it by 1. initializing it and 2. storing/restoring it 248 // as callee-save on entry/exit. 249 // But: on other platforms, specifically ia32, it would be tricky to enable 250 // the kRootRegister since it's currently used for other purposes. Thus, for 251 // consistency, we also keep it uninitialized here. 252 const NoRootArrayScope no_root_array_scope_; 253 254 ZoneChunkList<int> code_relative_fixup_positions_; 255 256 // Which mode to generate code for (LATIN1 or UC16). 257 const Mode mode_; 258 259 // One greater than maximal register index actually used. 260 int num_registers_; 261 262 // Number of registers to output at the end (the saved registers 263 // are always 0..num_saved_registers_-1) 264 const int num_saved_registers_; 265 266 // Labels used internally. 267 Label entry_label_; 268 Label start_label_; 269 Label success_label_; 270 Label backtrack_label_; 271 Label exit_label_; 272 Label check_preempt_label_; 273 Label stack_overflow_label_; 274 Label fallback_label_; 275 }; 276 277 } // namespace internal 278 } // namespace v8 279 280 #endif // V8_REGEXP_X64_REGEXP_MACRO_ASSEMBLER_X64_H_ 281