SemaChecking.cpp revision d69ec16b1b03b4a97c571ff14f15769fe13c1e5a
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 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 implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "Sema.h" 16#include "clang/Analysis/Analyses/PrintfFormatString.h" 17#include "clang/AST/ASTContext.h" 18#include "clang/AST/CharUnits.h" 19#include "clang/AST/DeclObjC.h" 20#include "clang/AST/ExprCXX.h" 21#include "clang/AST/ExprObjC.h" 22#include "clang/AST/DeclObjC.h" 23#include "clang/AST/StmtCXX.h" 24#include "clang/AST/StmtObjC.h" 25#include "clang/Lex/LiteralSupport.h" 26#include "clang/Lex/Preprocessor.h" 27#include "llvm/ADT/BitVector.h" 28#include "llvm/ADT/STLExtras.h" 29#include "llvm/ADT/StringExtras.h" 30#include "llvm/Support/raw_ostream.h" 31#include "clang/Basic/TargetBuiltins.h" 32#include "clang/Basic/TargetInfo.h" 33#include <limits> 34using namespace clang; 35 36/// getLocationOfStringLiteralByte - Return a source location that points to the 37/// specified byte of the specified string literal. 38/// 39/// Strings are amazingly complex. They can be formed from multiple tokens and 40/// can have escape sequences in them in addition to the usual trigraph and 41/// escaped newline business. This routine handles this complexity. 42/// 43SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 44 unsigned ByteNo) const { 45 assert(!SL->isWide() && "This doesn't work for wide strings yet"); 46 47 // Loop over all of the tokens in this string until we find the one that 48 // contains the byte we're looking for. 49 unsigned TokNo = 0; 50 while (1) { 51 assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!"); 52 SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo); 53 54 // Get the spelling of the string so that we can get the data that makes up 55 // the string literal, not the identifier for the macro it is potentially 56 // expanded through. 57 SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc); 58 59 // Re-lex the token to get its length and original spelling. 60 std::pair<FileID, unsigned> LocInfo = 61 SourceMgr.getDecomposedLoc(StrTokSpellingLoc); 62 bool Invalid = false; 63 llvm::StringRef Buffer = SourceMgr.getBufferData(LocInfo.first, &Invalid); 64 if (Invalid) 65 return StrTokSpellingLoc; 66 67 const char *StrData = Buffer.data()+LocInfo.second; 68 69 // Create a langops struct and enable trigraphs. This is sufficient for 70 // relexing tokens. 71 LangOptions LangOpts; 72 LangOpts.Trigraphs = true; 73 74 // Create a lexer starting at the beginning of this token. 75 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.begin(), StrData, 76 Buffer.end()); 77 Token TheTok; 78 TheLexer.LexFromRawLexer(TheTok); 79 80 // Use the StringLiteralParser to compute the length of the string in bytes. 81 StringLiteralParser SLP(&TheTok, 1, PP, /*Complain=*/false); 82 unsigned TokNumBytes = SLP.GetStringLength(); 83 84 // If the byte is in this token, return the location of the byte. 85 if (ByteNo < TokNumBytes || 86 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) { 87 unsigned Offset = 88 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP, 89 /*Complain=*/false); 90 91 // Now that we know the offset of the token in the spelling, use the 92 // preprocessor to get the offset in the original source. 93 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset); 94 } 95 96 // Move to the next string token. 97 ++TokNo; 98 ByteNo -= TokNumBytes; 99 } 100} 101 102/// CheckablePrintfAttr - does a function call have a "printf" attribute 103/// and arguments that merit checking? 104bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 105 if (Format->getType() == "printf") return true; 106 if (Format->getType() == "printf0") { 107 // printf0 allows null "format" string; if so don't check format/args 108 unsigned format_idx = Format->getFormatIdx() - 1; 109 // Does the index refer to the implicit object argument? 110 if (isa<CXXMemberCallExpr>(TheCall)) { 111 if (format_idx == 0) 112 return false; 113 --format_idx; 114 } 115 if (format_idx < TheCall->getNumArgs()) { 116 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 117 if (!Format->isNullPointerConstant(Context, 118 Expr::NPC_ValueDependentIsNull)) 119 return true; 120 } 121 } 122 return false; 123} 124 125Action::OwningExprResult 126Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 127 OwningExprResult TheCallResult(Owned(TheCall)); 128 129 switch (BuiltinID) { 130 case Builtin::BI__builtin___CFStringMakeConstantString: 131 assert(TheCall->getNumArgs() == 1 && 132 "Wrong # arguments to builtin CFStringMakeConstantString"); 133 if (CheckObjCString(TheCall->getArg(0))) 134 return ExprError(); 135 break; 136 case Builtin::BI__builtin_stdarg_start: 137 case Builtin::BI__builtin_va_start: 138 if (SemaBuiltinVAStart(TheCall)) 139 return ExprError(); 140 break; 141 case Builtin::BI__builtin_isgreater: 142 case Builtin::BI__builtin_isgreaterequal: 143 case Builtin::BI__builtin_isless: 144 case Builtin::BI__builtin_islessequal: 145 case Builtin::BI__builtin_islessgreater: 146 case Builtin::BI__builtin_isunordered: 147 if (SemaBuiltinUnorderedCompare(TheCall)) 148 return ExprError(); 149 break; 150 case Builtin::BI__builtin_fpclassify: 151 if (SemaBuiltinFPClassification(TheCall, 6)) 152 return ExprError(); 153 break; 154 case Builtin::BI__builtin_isfinite: 155 case Builtin::BI__builtin_isinf: 156 case Builtin::BI__builtin_isinf_sign: 157 case Builtin::BI__builtin_isnan: 158 case Builtin::BI__builtin_isnormal: 159 if (SemaBuiltinFPClassification(TheCall, 1)) 160 return ExprError(); 161 break; 162 case Builtin::BI__builtin_return_address: 163 case Builtin::BI__builtin_frame_address: { 164 llvm::APSInt Result; 165 if (SemaBuiltinConstantArg(TheCall, 0, Result)) 166 return ExprError(); 167 break; 168 } 169 case Builtin::BI__builtin_eh_return_data_regno: { 170 llvm::APSInt Result; 171 if (SemaBuiltinConstantArg(TheCall, 0, Result)) 172 return ExprError(); 173 break; 174 } 175 case Builtin::BI__builtin_shufflevector: 176 return SemaBuiltinShuffleVector(TheCall); 177 // TheCall will be freed by the smart pointer here, but that's fine, since 178 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 179 case Builtin::BI__builtin_prefetch: 180 if (SemaBuiltinPrefetch(TheCall)) 181 return ExprError(); 182 break; 183 case Builtin::BI__builtin_object_size: 184 if (SemaBuiltinObjectSize(TheCall)) 185 return ExprError(); 186 break; 187 case Builtin::BI__builtin_longjmp: 188 if (SemaBuiltinLongjmp(TheCall)) 189 return ExprError(); 190 break; 191 case Builtin::BI__sync_fetch_and_add: 192 case Builtin::BI__sync_fetch_and_sub: 193 case Builtin::BI__sync_fetch_and_or: 194 case Builtin::BI__sync_fetch_and_and: 195 case Builtin::BI__sync_fetch_and_xor: 196 case Builtin::BI__sync_add_and_fetch: 197 case Builtin::BI__sync_sub_and_fetch: 198 case Builtin::BI__sync_and_and_fetch: 199 case Builtin::BI__sync_or_and_fetch: 200 case Builtin::BI__sync_xor_and_fetch: 201 case Builtin::BI__sync_val_compare_and_swap: 202 case Builtin::BI__sync_bool_compare_and_swap: 203 case Builtin::BI__sync_lock_test_and_set: 204 case Builtin::BI__sync_lock_release: 205 if (SemaBuiltinAtomicOverloaded(TheCall)) 206 return ExprError(); 207 break; 208 } 209 210 // Since the target specific builtins for each arch overlap, only check those 211 // of the arch we are compiling for. 212 if (BuiltinID >= Builtin::FirstTSBuiltin) { 213 switch (Context.Target.getTriple().getArch()) { 214 case llvm::Triple::arm: 215 case llvm::Triple::thumb: 216 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 217 return ExprError(); 218 break; 219 case llvm::Triple::x86: 220 case llvm::Triple::x86_64: 221 if (CheckX86BuiltinFunctionCall(BuiltinID, TheCall)) 222 return ExprError(); 223 break; 224 default: 225 break; 226 } 227 } 228 229 return move(TheCallResult); 230} 231 232bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 233 switch (BuiltinID) { 234 case X86::BI__builtin_ia32_palignr128: 235 case X86::BI__builtin_ia32_palignr: { 236 llvm::APSInt Result; 237 if (SemaBuiltinConstantArg(TheCall, 2, Result)) 238 return true; 239 break; 240 } 241 } 242 return false; 243} 244 245// Get the valid immediate range for the specified NEON type code. 246static unsigned RFT(unsigned t, bool shift = false) { 247 bool quad = t & 0x10; 248 249 switch (t & 0x7) { 250 case 0: // i8 251 return shift ? 7 : (8 << (int)quad) - 1; 252 case 1: // i16 253 return shift ? 15 : (4 << (int)quad) - 1; 254 case 2: // i32 255 return shift ? 31 : (2 << (int)quad) - 1; 256 case 3: // i64 257 return shift ? 63 : (1 << (int)quad) - 1; 258 case 4: // f32 259 assert(!shift && "cannot shift float types!"); 260 return (2 << (int)quad) - 1; 261 case 5: // poly8 262 assert(!shift && "cannot shift polynomial types!"); 263 return (8 << (int)quad) - 1; 264 case 6: // poly16 265 assert(!shift && "cannot shift polynomial types!"); 266 return (4 << (int)quad) - 1; 267 case 7: // float16 268 assert(!shift && "cannot shift float types!"); 269 return (4 << (int)quad) - 1; 270 } 271 return 0; 272} 273 274bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 275 llvm::APSInt Result; 276 277 unsigned mask = 0; 278 unsigned TV = 0; 279 switch (BuiltinID) { 280 case ARM::BI__builtin_neon_vaba_v: mask = 0x707; break; 281 case ARM::BI__builtin_neon_vabaq_v: mask = 0x7070000; break; 282 case ARM::BI__builtin_neon_vabal_v: mask = 0xE0E0000; break; 283 case ARM::BI__builtin_neon_vabd_v: mask = 0x717; break; 284 case ARM::BI__builtin_neon_vabdq_v: mask = 0x7170000; break; 285 case ARM::BI__builtin_neon_vabdl_v: mask = 0xE0E0000; break; 286 case ARM::BI__builtin_neon_vabs_v: mask = 0x17; break; 287 case ARM::BI__builtin_neon_vabsq_v: mask = 0x170000; break; 288 case ARM::BI__builtin_neon_vaddhn_v: mask = 0x707; break; 289 case ARM::BI__builtin_neon_vaddl_v: mask = 0xE0E0000; break; 290 case ARM::BI__builtin_neon_vaddw_v: mask = 0xE0E0000; break; 291 case ARM::BI__builtin_neon_vcage_v: mask = 0x400; break; 292 case ARM::BI__builtin_neon_vcageq_v: mask = 0x4000000; break; 293 case ARM::BI__builtin_neon_vcagt_v: mask = 0x400; break; 294 case ARM::BI__builtin_neon_vcagtq_v: mask = 0x4000000; break; 295 case ARM::BI__builtin_neon_vcale_v: mask = 0x400; break; 296 case ARM::BI__builtin_neon_vcaleq_v: mask = 0x4000000; break; 297 case ARM::BI__builtin_neon_vcalt_v: mask = 0x400; break; 298 case ARM::BI__builtin_neon_vcaltq_v: mask = 0x4000000; break; 299 case ARM::BI__builtin_neon_vcls_v: mask = 0x7; break; 300 case ARM::BI__builtin_neon_vclsq_v: mask = 0x70000; break; 301 case ARM::BI__builtin_neon_vclz_v: mask = 0x707; break; 302 case ARM::BI__builtin_neon_vclzq_v: mask = 0x7070000; break; 303 case ARM::BI__builtin_neon_vcnt_v: mask = 0x121; break; 304 case ARM::BI__builtin_neon_vcntq_v: mask = 0x1210000; break; 305 case ARM::BI__builtin_neon_vcvt_f16_v: mask = 0x80; break; 306 case ARM::BI__builtin_neon_vcvt_f32_v: mask = 0x404; break; 307 case ARM::BI__builtin_neon_vcvtq_f32_v: mask = 0x4040000; break; 308 case ARM::BI__builtin_neon_vcvt_f32_f16: mask = 0x100000; break; 309 case ARM::BI__builtin_neon_vcvt_n_f32_v: mask = 0x404; break; 310 case ARM::BI__builtin_neon_vcvtq_n_f32_v: mask = 0x4040000; break; 311 case ARM::BI__builtin_neon_vcvt_n_s32_v: mask = 0x4; break; 312 case ARM::BI__builtin_neon_vcvtq_n_s32_v: mask = 0x40000; break; 313 case ARM::BI__builtin_neon_vcvt_n_u32_v: mask = 0x400; break; 314 case ARM::BI__builtin_neon_vcvtq_n_u32_v: mask = 0x4000000; break; 315 case ARM::BI__builtin_neon_vcvt_s32_v: mask = 0x4; break; 316 case ARM::BI__builtin_neon_vcvtq_s32_v: mask = 0x40000; break; 317 case ARM::BI__builtin_neon_vcvt_u32_v: mask = 0x400; break; 318 case ARM::BI__builtin_neon_vcvtq_u32_v: mask = 0x4000000; break; 319 case ARM::BI__builtin_neon_vext_v: mask = 0xF6F; break; 320 case ARM::BI__builtin_neon_vextq_v: mask = 0xF6F0000; break; 321 case ARM::BI__builtin_neon_vhadd_v: mask = 0x707; break; 322 case ARM::BI__builtin_neon_vhaddq_v: mask = 0x7070000; break; 323 case ARM::BI__builtin_neon_vhsub_v: mask = 0x707; break; 324 case ARM::BI__builtin_neon_vhsubq_v: mask = 0x7070000; break; 325 case ARM::BI__builtin_neon_vld1_v: mask = 0xFFF; break; 326 case ARM::BI__builtin_neon_vld1q_v: mask = 0xFFF0000; break; 327 case ARM::BI__builtin_neon_vld1_dup_v: mask = 0xFFF; break; 328 case ARM::BI__builtin_neon_vld1q_dup_v: mask = 0xFFF0000; break; 329 case ARM::BI__builtin_neon_vld1_lane_v: mask = 0xFFF; break; 330 case ARM::BI__builtin_neon_vld1q_lane_v: mask = 0xFFF0000; break; 331 case ARM::BI__builtin_neon_vld2_v: mask = 0xFFF; break; 332 case ARM::BI__builtin_neon_vld2q_v: mask = 0x7F70000; break; 333 case ARM::BI__builtin_neon_vld2_dup_v: mask = 0xFFF; break; 334 case ARM::BI__builtin_neon_vld2_lane_v: mask = 0x7F7; break; 335 case ARM::BI__builtin_neon_vld2q_lane_v: mask = 0x6D60000; break; 336 case ARM::BI__builtin_neon_vld3_v: mask = 0xFFF; break; 337 case ARM::BI__builtin_neon_vld3q_v: mask = 0x7F70000; break; 338 case ARM::BI__builtin_neon_vld3_dup_v: mask = 0xFFF; break; 339 case ARM::BI__builtin_neon_vld3_lane_v: mask = 0x7F7; break; 340 case ARM::BI__builtin_neon_vld3q_lane_v: mask = 0x6D60000; break; 341 case ARM::BI__builtin_neon_vld4_v: mask = 0xFFF; break; 342 case ARM::BI__builtin_neon_vld4q_v: mask = 0x7F70000; break; 343 case ARM::BI__builtin_neon_vld4_dup_v: mask = 0xFFF; break; 344 case ARM::BI__builtin_neon_vld4_lane_v: mask = 0x7F7; break; 345 case ARM::BI__builtin_neon_vld4q_lane_v: mask = 0x6D60000; break; 346 case ARM::BI__builtin_neon_vmax_v: mask = 0x717; break; 347 case ARM::BI__builtin_neon_vmaxq_v: mask = 0x7170000; break; 348 case ARM::BI__builtin_neon_vmin_v: mask = 0x717; break; 349 case ARM::BI__builtin_neon_vminq_v: mask = 0x7170000; break; 350 case ARM::BI__builtin_neon_vmlal_v: mask = 0xE0E0000; break; 351 case ARM::BI__builtin_neon_vmlal_lane_v: mask = 0xC0C0000; break; 352 case ARM::BI__builtin_neon_vmla_lane_v: mask = 0x616; break; 353 case ARM::BI__builtin_neon_vmlaq_lane_v: mask = 0x6160000; break; 354 case ARM::BI__builtin_neon_vmlsl_v: mask = 0xE0E0000; break; 355 case ARM::BI__builtin_neon_vmlsl_lane_v: mask = 0xC0C0000; break; 356 case ARM::BI__builtin_neon_vmls_lane_v: mask = 0x616; break; 357 case ARM::BI__builtin_neon_vmlsq_lane_v: mask = 0x6160000; break; 358 case ARM::BI__builtin_neon_vmovl_v: mask = 0xE0E0000; break; 359 case ARM::BI__builtin_neon_vmovn_v: mask = 0x707; break; 360 case ARM::BI__builtin_neon_vmull_v: mask = 0xE4E0000; break; 361 case ARM::BI__builtin_neon_vmull_lane_v: mask = 0xC0C0000; break; 362 case ARM::BI__builtin_neon_vpadal_v: mask = 0xE0E; break; 363 case ARM::BI__builtin_neon_vpadalq_v: mask = 0xE0E0000; break; 364 case ARM::BI__builtin_neon_vpadd_v: mask = 0x717; break; 365 case ARM::BI__builtin_neon_vpaddl_v: mask = 0xE0E; break; 366 case ARM::BI__builtin_neon_vpaddlq_v: mask = 0xE0E0000; break; 367 case ARM::BI__builtin_neon_vpmax_v: mask = 0x717; break; 368 case ARM::BI__builtin_neon_vpmin_v: mask = 0x717; break; 369 case ARM::BI__builtin_neon_vqabs_v: mask = 0x7; break; 370 case ARM::BI__builtin_neon_vqabsq_v: mask = 0x70000; break; 371 case ARM::BI__builtin_neon_vqadd_v: mask = 0xF0F; break; 372 case ARM::BI__builtin_neon_vqaddq_v: mask = 0xF0F0000; break; 373 case ARM::BI__builtin_neon_vqdmlal_v: mask = 0xC0000; break; 374 case ARM::BI__builtin_neon_vqdmlal_lane_v: mask = 0xC0000; break; 375 case ARM::BI__builtin_neon_vqdmlsl_v: mask = 0xC0000; break; 376 case ARM::BI__builtin_neon_vqdmlsl_lane_v: mask = 0xC0000; break; 377 case ARM::BI__builtin_neon_vqdmulh_v: mask = 0x6; break; 378 case ARM::BI__builtin_neon_vqdmulhq_v: mask = 0x60000; break; 379 case ARM::BI__builtin_neon_vqdmulh_lane_v: mask = 0x6; break; 380 case ARM::BI__builtin_neon_vqdmulhq_lane_v: mask = 0x60000; break; 381 case ARM::BI__builtin_neon_vqdmull_v: mask = 0xC0000; break; 382 case ARM::BI__builtin_neon_vqdmull_lane_v: mask = 0xC0000; break; 383 case ARM::BI__builtin_neon_vqmovn_v: mask = 0x707; break; 384 case ARM::BI__builtin_neon_vqmovun_v: mask = 0x700; break; 385 case ARM::BI__builtin_neon_vqneg_v: mask = 0x7; break; 386 case ARM::BI__builtin_neon_vqnegq_v: mask = 0x70000; break; 387 case ARM::BI__builtin_neon_vqrdmulh_v: mask = 0x6; break; 388 case ARM::BI__builtin_neon_vqrdmulhq_v: mask = 0x60000; break; 389 case ARM::BI__builtin_neon_vqrdmulh_lane_v: mask = 0x6; break; 390 case ARM::BI__builtin_neon_vqrdmulhq_lane_v: mask = 0x60000; break; 391 case ARM::BI__builtin_neon_vqrshl_v: mask = 0xF0F; break; 392 case ARM::BI__builtin_neon_vqrshlq_v: mask = 0xF0F0000; break; 393 case ARM::BI__builtin_neon_vqrshrn_n_v: mask = 0x707; break; 394 case ARM::BI__builtin_neon_vqrshrun_n_v: mask = 0x700; break; 395 case ARM::BI__builtin_neon_vqshl_v: mask = 0xF0F; break; 396 case ARM::BI__builtin_neon_vqshlq_v: mask = 0xF0F0000; break; 397 case ARM::BI__builtin_neon_vqshlu_n_v: mask = 0xF00; break; 398 case ARM::BI__builtin_neon_vqshluq_n_v: mask = 0xF000000; break; 399 case ARM::BI__builtin_neon_vqshl_n_v: mask = 0xF0F; break; 400 case ARM::BI__builtin_neon_vqshlq_n_v: mask = 0xF0F0000; break; 401 case ARM::BI__builtin_neon_vqshrn_n_v: mask = 0x707; break; 402 case ARM::BI__builtin_neon_vqshrun_n_v: mask = 0x700; break; 403 case ARM::BI__builtin_neon_vqsub_v: mask = 0xF0F; break; 404 case ARM::BI__builtin_neon_vqsubq_v: mask = 0xF0F0000; break; 405 case ARM::BI__builtin_neon_vraddhn_v: mask = 0x707; break; 406 case ARM::BI__builtin_neon_vrecpe_v: mask = 0x410; break; 407 case ARM::BI__builtin_neon_vrecpeq_v: mask = 0x4100000; break; 408 case ARM::BI__builtin_neon_vrecps_v: mask = 0x10; break; 409 case ARM::BI__builtin_neon_vrecpsq_v: mask = 0x100000; break; 410 case ARM::BI__builtin_neon_vrhadd_v: mask = 0x707; break; 411 case ARM::BI__builtin_neon_vrhaddq_v: mask = 0x7070000; break; 412 case ARM::BI__builtin_neon_vrshl_v: mask = 0xF0F; break; 413 case ARM::BI__builtin_neon_vrshlq_v: mask = 0xF0F0000; break; 414 case ARM::BI__builtin_neon_vrshrn_n_v: mask = 0x707; break; 415 case ARM::BI__builtin_neon_vrshr_n_v: mask = 0xF0F; break; 416 case ARM::BI__builtin_neon_vrshrq_n_v: mask = 0xF0F0000; break; 417 case ARM::BI__builtin_neon_vrsqrte_v: mask = 0x410; break; 418 case ARM::BI__builtin_neon_vrsqrteq_v: mask = 0x4100000; break; 419 case ARM::BI__builtin_neon_vrsqrts_v: mask = 0x10; break; 420 case ARM::BI__builtin_neon_vrsqrtsq_v: mask = 0x100000; break; 421 case ARM::BI__builtin_neon_vrsra_n_v: mask = 0xF0F; break; 422 case ARM::BI__builtin_neon_vrsraq_n_v: mask = 0xF0F0000; break; 423 case ARM::BI__builtin_neon_vrsubhn_v: mask = 0x707; break; 424 case ARM::BI__builtin_neon_vshl_v: mask = 0xF0F; break; 425 case ARM::BI__builtin_neon_vshlq_v: mask = 0xF0F0000; break; 426 case ARM::BI__builtin_neon_vshll_n_v: mask = 0xE0E0000; break; 427 case ARM::BI__builtin_neon_vshl_n_v: mask = 0xF0F; break; 428 case ARM::BI__builtin_neon_vshlq_n_v: mask = 0xF0F0000; break; 429 case ARM::BI__builtin_neon_vshrn_n_v: mask = 0x707; break; 430 case ARM::BI__builtin_neon_vshr_n_v: mask = 0xF0F; break; 431 case ARM::BI__builtin_neon_vshrq_n_v: mask = 0xF0F0000; break; 432 case ARM::BI__builtin_neon_vsli_n_v: mask = 0xF6F; break; 433 case ARM::BI__builtin_neon_vsliq_n_v: mask = 0xF6F0000; break; 434 case ARM::BI__builtin_neon_vsra_n_v: mask = 0xF0F; break; 435 case ARM::BI__builtin_neon_vsraq_n_v: mask = 0xF0F0000; break; 436 case ARM::BI__builtin_neon_vsri_n_v: mask = 0xF6F; break; 437 case ARM::BI__builtin_neon_vsriq_n_v: mask = 0xF6F0000; break; 438 case ARM::BI__builtin_neon_vst1_v: mask = 0x9F; break; 439 case ARM::BI__builtin_neon_vst1q_v: mask = 0x9F0000; break; 440 case ARM::BI__builtin_neon_vst1_lane_v: mask = 0x9F; break; 441 case ARM::BI__builtin_neon_vst1q_lane_v: mask = 0x9F0000; break; 442 case ARM::BI__builtin_neon_vst2_v: mask = 0x9F; break; 443 case ARM::BI__builtin_neon_vst2q_v: mask = 0x970000; break; 444 case ARM::BI__builtin_neon_vst2_lane_v: mask = 0x97; break; 445 case ARM::BI__builtin_neon_vst2q_lane_v: mask = 0x960000; break; 446 case ARM::BI__builtin_neon_vst3_v: mask = 0x9F; break; 447 case ARM::BI__builtin_neon_vst3q_v: mask = 0x970000; break; 448 case ARM::BI__builtin_neon_vst3_lane_v: mask = 0x97; break; 449 case ARM::BI__builtin_neon_vst3q_lane_v: mask = 0x960000; break; 450 case ARM::BI__builtin_neon_vst4_v: mask = 0x9F; break; 451 case ARM::BI__builtin_neon_vst4q_v: mask = 0x970000; break; 452 case ARM::BI__builtin_neon_vst4_lane_v: mask = 0x97; break; 453 case ARM::BI__builtin_neon_vst4q_lane_v: mask = 0x960000; break; 454 case ARM::BI__builtin_neon_vsubhn_v: mask = 0x707; break; 455 case ARM::BI__builtin_neon_vsubl_v: mask = 0xE0E0000; break; 456 case ARM::BI__builtin_neon_vsubw_v: mask = 0xE0E0000; break; 457 case ARM::BI__builtin_neon_vtbl1_v: mask = 0x121; break; 458 case ARM::BI__builtin_neon_vtbl2_v: mask = 0x121; break; 459 case ARM::BI__builtin_neon_vtbl3_v: mask = 0x121; break; 460 case ARM::BI__builtin_neon_vtbl4_v: mask = 0x121; break; 461 case ARM::BI__builtin_neon_vtbx1_v: mask = 0x121; break; 462 case ARM::BI__builtin_neon_vtbx2_v: mask = 0x121; break; 463 case ARM::BI__builtin_neon_vtbx3_v: mask = 0x121; break; 464 case ARM::BI__builtin_neon_vtbx4_v: mask = 0x121; break; 465 case ARM::BI__builtin_neon_vtrn_v: mask = 0x777; break; 466 case ARM::BI__builtin_neon_vtrnq_v: mask = 0x7770000; break; 467 case ARM::BI__builtin_neon_vtst_v: mask = 0x700; break; 468 case ARM::BI__builtin_neon_vtstq_v: mask = 0x7000000; break; 469 case ARM::BI__builtin_neon_vuzp_v: mask = 0x777; break; 470 case ARM::BI__builtin_neon_vuzpq_v: mask = 0x7770000; break; 471 case ARM::BI__builtin_neon_vzip_v: mask = 0x373; break; 472 case ARM::BI__builtin_neon_vzipq_v: mask = 0x7770000; break; 473 } 474 475 // For NEON intrinsics which are overloaded on vector element type, validate 476 // the immediate which specifies which variant to emit. 477 if (mask) { 478 unsigned ArgNo = TheCall->getNumArgs()-1; 479 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 480 return true; 481 482 TV = Result.getLimitedValue(32); 483 if ((TV > 31) || (mask & (1 << TV)) == 0) 484 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 485 << TheCall->getArg(ArgNo)->getSourceRange(); 486 } 487 488 // For NEON intrinsics which take an immediate value as part of the 489 // instruction, range check them here. 490 unsigned i = 0, l = 0, u = 0; 491 switch (BuiltinID) { 492 default: return false; 493 case ARM::BI__builtin_neon_vcvt_n_f32_v: i = 1; u = RFT(TV); break; 494 case ARM::BI__builtin_neon_vcvtq_n_f32_v: i = 1; u = RFT(TV); break; 495 case ARM::BI__builtin_neon_vcvt_n_s32_v: i = 1; u = RFT(TV); break; 496 case ARM::BI__builtin_neon_vcvtq_n_s32_v: i = 1; u = RFT(TV); break; 497 case ARM::BI__builtin_neon_vcvt_n_u32_v: i = 1; u = RFT(TV); break; 498 case ARM::BI__builtin_neon_vcvtq_n_u32_v: i = 1; u = RFT(TV); break; 499 case ARM::BI__builtin_neon_vext_v: i = 2; u = RFT(TV); break; 500 case ARM::BI__builtin_neon_vextq_v: i = 2; u = RFT(TV); break; 501 case ARM::BI__builtin_neon_vget_lane_i8: i = 1; u = 7; break; 502 case ARM::BI__builtin_neon_vget_lane_i16: i = 1; u = 3; break; 503 case ARM::BI__builtin_neon_vget_lane_i32: i = 1; u = 1; break; 504 case ARM::BI__builtin_neon_vget_lane_f32: i = 1; u = 1; break; 505 case ARM::BI__builtin_neon_vgetq_lane_i8: i = 1; u = 15; break; 506 case ARM::BI__builtin_neon_vgetq_lane_i16: i = 1; u = 7; break; 507 case ARM::BI__builtin_neon_vgetq_lane_i32: i = 1; u = 3; break; 508 case ARM::BI__builtin_neon_vgetq_lane_f32: i = 1; u = 3; break; 509 case ARM::BI__builtin_neon_vget_lane_i64: i = 1; u = 0; break; 510 case ARM::BI__builtin_neon_vgetq_lane_i64: i = 1; u = 1; break; 511 case ARM::BI__builtin_neon_vld1q_lane_v: i = 1; u = RFT(TV); break; 512 case ARM::BI__builtin_neon_vld1_lane_v: i = 1; u = RFT(TV); break; 513 case ARM::BI__builtin_neon_vld2q_lane_v: i = 1; u = RFT(TV); break; 514 case ARM::BI__builtin_neon_vld2_lane_v: i = 1; u = RFT(TV); break; 515 case ARM::BI__builtin_neon_vld3q_lane_v: i = 1; u = RFT(TV); break; 516 case ARM::BI__builtin_neon_vld3_lane_v: i = 1; u = RFT(TV); break; 517 case ARM::BI__builtin_neon_vld4q_lane_v: i = 1; u = RFT(TV); break; 518 case ARM::BI__builtin_neon_vld4_lane_v: i = 1; u = RFT(TV); break; 519 case ARM::BI__builtin_neon_vmlal_lane_v: i = 3; u = RFT(TV); break; 520 case ARM::BI__builtin_neon_vmla_lane_v: i = 3; u = RFT(TV); break; 521 case ARM::BI__builtin_neon_vmlaq_lane_v: i = 3; u = RFT(TV); break; 522 case ARM::BI__builtin_neon_vmlsl_lane_v: i = 3; u = RFT(TV); break; 523 case ARM::BI__builtin_neon_vmls_lane_v: i = 3; u = RFT(TV); break; 524 case ARM::BI__builtin_neon_vmlsq_lane_v: i = 3; u = RFT(TV); break; 525 case ARM::BI__builtin_neon_vmull_lane_v: i = 2; u = RFT(TV); break; 526 case ARM::BI__builtin_neon_vqdmlal_lane_v: i = 3; u = RFT(TV); break; 527 case ARM::BI__builtin_neon_vqdmlsl_lane_v: i = 3; u = RFT(TV); break; 528 case ARM::BI__builtin_neon_vqdmulh_lane_v: i = 2; u = RFT(TV); break; 529 case ARM::BI__builtin_neon_vqdmulhq_lane_v: i = 2; u = RFT(TV); break; 530 case ARM::BI__builtin_neon_vqdmull_lane_v: i = 2; u = RFT(TV); break; 531 case ARM::BI__builtin_neon_vqrdmulh_lane_v: i = 2; u = RFT(TV); break; 532 case ARM::BI__builtin_neon_vqrdmulhq_lane_v: i = 2; u = RFT(TV); break; 533 case ARM::BI__builtin_neon_vqrshrn_n_v: i = 1; l = 1; u = RFT(TV, true); break; 534 case ARM::BI__builtin_neon_vqrshrun_n_v: i = 1; l = 1; u = RFT(TV, true); break; 535 case ARM::BI__builtin_neon_vqshlu_n_v: i = 1; u = RFT(TV, true); break; 536 case ARM::BI__builtin_neon_vqshluq_n_v: i = 1; u = RFT(TV, true); break; 537 case ARM::BI__builtin_neon_vqshl_n_v: i = 1; u = RFT(TV, true); break; 538 case ARM::BI__builtin_neon_vqshlq_n_v: i = 1; u = RFT(TV, true); break; 539 case ARM::BI__builtin_neon_vqshrn_n_v: i = 1; l = 1; u = RFT(TV, true); break; 540 case ARM::BI__builtin_neon_vqshrun_n_v: i = 1; l = 1; u = RFT(TV, true); break; 541 case ARM::BI__builtin_neon_vrshrn_n_v: i = 1; l = 1; u = RFT(TV, true); break; 542 case ARM::BI__builtin_neon_vrshr_n_v: i = 1; l = 1; u = RFT(TV, true); break; 543 case ARM::BI__builtin_neon_vrshrq_n_v: i = 1; l = 1; u = RFT(TV, true); break; 544 case ARM::BI__builtin_neon_vrsra_n_v: i = 2; l = 1; u = RFT(TV, true); break; 545 case ARM::BI__builtin_neon_vrsraq_n_v: i = 2; l = 1; u = RFT(TV, true); break; 546 case ARM::BI__builtin_neon_vset_lane_i8: i = 2; u = 7; break; 547 case ARM::BI__builtin_neon_vset_lane_i16: i = 2; u = 3; break; 548 case ARM::BI__builtin_neon_vset_lane_i32: i = 2; u = 1; break; 549 case ARM::BI__builtin_neon_vset_lane_f32: i = 2; u = 1; break; 550 case ARM::BI__builtin_neon_vsetq_lane_i8: i = 2; u = 15; break; 551 case ARM::BI__builtin_neon_vsetq_lane_i16: i = 2; u = 7; break; 552 case ARM::BI__builtin_neon_vsetq_lane_i32: i = 2; u = 3; break; 553 case ARM::BI__builtin_neon_vsetq_lane_f32: i = 2; u = 3; break; 554 case ARM::BI__builtin_neon_vset_lane_i64: i = 2; u = 0; break; 555 case ARM::BI__builtin_neon_vsetq_lane_i64: i = 2; u = 1; break; 556 case ARM::BI__builtin_neon_vshll_n_v: i = 1; u = RFT(TV, true); break; 557 case ARM::BI__builtin_neon_vshl_n_v: i = 1; u = RFT(TV, true); break; 558 case ARM::BI__builtin_neon_vshlq_n_v: i = 1; u = RFT(TV, true); break; 559 case ARM::BI__builtin_neon_vshrn_n_v: i = 1; l = 1; u = RFT(TV, true); break; 560 case ARM::BI__builtin_neon_vshr_n_v: i = 1; l = 1; u = RFT(TV, true); break; 561 case ARM::BI__builtin_neon_vshrq_n_v: i = 1; l = 1; u = RFT(TV, true); break; 562 case ARM::BI__builtin_neon_vsli_n_v: i = 2; u = RFT(TV, true); break; 563 case ARM::BI__builtin_neon_vsliq_n_v: i = 2; u = RFT(TV, true); break; 564 case ARM::BI__builtin_neon_vsra_n_v: i = 2; l = 1; u = RFT(TV, true); break; 565 case ARM::BI__builtin_neon_vsraq_n_v: i = 2; l = 1; u = RFT(TV, true); break; 566 case ARM::BI__builtin_neon_vsri_n_v: i = 2; l = 1; u = RFT(TV, true); break; 567 case ARM::BI__builtin_neon_vsriq_n_v: i = 2; l = 1; u = RFT(TV, true); break; 568 case ARM::BI__builtin_neon_vst1q_lane_v: i = 2; u = RFT(TV); break; 569 case ARM::BI__builtin_neon_vst1_lane_v: i = 2; u = RFT(TV); break; 570 case ARM::BI__builtin_neon_vst2q_lane_v: i = 2; u = RFT(TV); break; 571 case ARM::BI__builtin_neon_vst2_lane_v: i = 2; u = RFT(TV); break; 572 case ARM::BI__builtin_neon_vst3q_lane_v: i = 2; u = RFT(TV); break; 573 case ARM::BI__builtin_neon_vst3_lane_v: i = 2; u = RFT(TV); break; 574 case ARM::BI__builtin_neon_vst4q_lane_v: i = 2; u = RFT(TV); break; 575 case ARM::BI__builtin_neon_vst4_lane_v: i = 2; u = RFT(TV); break; 576 }; 577 578 // Check that the immediate argument is actually a constant. 579 if (SemaBuiltinConstantArg(TheCall, i, Result)) 580 return true; 581 582 // Range check against the upper/lower values for this isntruction. 583 unsigned Val = Result.getZExtValue(); 584 if (Val < l || Val > (u + l)) 585 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 586 << llvm::utostr(l) << llvm::utostr(u+l) 587 << TheCall->getArg(i)->getSourceRange(); 588 589 return false; 590} 591 592/// CheckFunctionCall - Check a direct function call for various correctness 593/// and safety properties not strictly enforced by the C type system. 594bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 595 // Get the IdentifierInfo* for the called function. 596 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 597 598 // None of the checks below are needed for functions that don't have 599 // simple names (e.g., C++ conversion functions). 600 if (!FnInfo) 601 return false; 602 603 // FIXME: This mechanism should be abstracted to be less fragile and 604 // more efficient. For example, just map function ids to custom 605 // handlers. 606 607 // Printf checking. 608 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 609 if (CheckablePrintfAttr(Format, TheCall)) { 610 bool HasVAListArg = Format->getFirstArg() == 0; 611 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 612 HasVAListArg ? 0 : Format->getFirstArg() - 1); 613 } 614 } 615 616 for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull; 617 NonNull = NonNull->getNext<NonNullAttr>()) 618 CheckNonNullArguments(NonNull, TheCall); 619 620 return false; 621} 622 623bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 624 // Printf checking. 625 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 626 if (!Format) 627 return false; 628 629 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 630 if (!V) 631 return false; 632 633 QualType Ty = V->getType(); 634 if (!Ty->isBlockPointerType()) 635 return false; 636 637 if (!CheckablePrintfAttr(Format, TheCall)) 638 return false; 639 640 bool HasVAListArg = Format->getFirstArg() == 0; 641 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 642 HasVAListArg ? 0 : Format->getFirstArg() - 1); 643 644 return false; 645} 646 647/// SemaBuiltinAtomicOverloaded - We have a call to a function like 648/// __sync_fetch_and_add, which is an overloaded function based on the pointer 649/// type of its first argument. The main ActOnCallExpr routines have already 650/// promoted the types of arguments because all of these calls are prototyped as 651/// void(...). 652/// 653/// This function goes through and does final semantic checking for these 654/// builtins, 655bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) { 656 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 657 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 658 659 // Ensure that we have at least one argument to do type inference from. 660 if (TheCall->getNumArgs() < 1) 661 return Diag(TheCall->getLocEnd(), 662 diag::err_typecheck_call_too_few_args_at_least) 663 << 0 << 1 << TheCall->getNumArgs() 664 << TheCall->getCallee()->getSourceRange(); 665 666 // Inspect the first argument of the atomic builtin. This should always be 667 // a pointer type, whose element is an integral scalar or pointer type. 668 // Because it is a pointer type, we don't have to worry about any implicit 669 // casts here. 670 Expr *FirstArg = TheCall->getArg(0); 671 if (!FirstArg->getType()->isPointerType()) 672 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 673 << FirstArg->getType() << FirstArg->getSourceRange(); 674 675 QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 676 if (!ValType->isIntegerType() && !ValType->isPointerType() && 677 !ValType->isBlockPointerType()) 678 return Diag(DRE->getLocStart(), 679 diag::err_atomic_builtin_must_be_pointer_intptr) 680 << FirstArg->getType() << FirstArg->getSourceRange(); 681 682 // We need to figure out which concrete builtin this maps onto. For example, 683 // __sync_fetch_and_add with a 2 byte object turns into 684 // __sync_fetch_and_add_2. 685#define BUILTIN_ROW(x) \ 686 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 687 Builtin::BI##x##_8, Builtin::BI##x##_16 } 688 689 static const unsigned BuiltinIndices[][5] = { 690 BUILTIN_ROW(__sync_fetch_and_add), 691 BUILTIN_ROW(__sync_fetch_and_sub), 692 BUILTIN_ROW(__sync_fetch_and_or), 693 BUILTIN_ROW(__sync_fetch_and_and), 694 BUILTIN_ROW(__sync_fetch_and_xor), 695 696 BUILTIN_ROW(__sync_add_and_fetch), 697 BUILTIN_ROW(__sync_sub_and_fetch), 698 BUILTIN_ROW(__sync_and_and_fetch), 699 BUILTIN_ROW(__sync_or_and_fetch), 700 BUILTIN_ROW(__sync_xor_and_fetch), 701 702 BUILTIN_ROW(__sync_val_compare_and_swap), 703 BUILTIN_ROW(__sync_bool_compare_and_swap), 704 BUILTIN_ROW(__sync_lock_test_and_set), 705 BUILTIN_ROW(__sync_lock_release) 706 }; 707#undef BUILTIN_ROW 708 709 // Determine the index of the size. 710 unsigned SizeIndex; 711 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 712 case 1: SizeIndex = 0; break; 713 case 2: SizeIndex = 1; break; 714 case 4: SizeIndex = 2; break; 715 case 8: SizeIndex = 3; break; 716 case 16: SizeIndex = 4; break; 717 default: 718 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 719 << FirstArg->getType() << FirstArg->getSourceRange(); 720 } 721 722 // Each of these builtins has one pointer argument, followed by some number of 723 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 724 // that we ignore. Find out which row of BuiltinIndices to read from as well 725 // as the number of fixed args. 726 unsigned BuiltinID = FDecl->getBuiltinID(); 727 unsigned BuiltinIndex, NumFixed = 1; 728 switch (BuiltinID) { 729 default: assert(0 && "Unknown overloaded atomic builtin!"); 730 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 731 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 732 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 733 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 734 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 735 736 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 737 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 738 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 739 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 740 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 741 742 case Builtin::BI__sync_val_compare_and_swap: 743 BuiltinIndex = 10; 744 NumFixed = 2; 745 break; 746 case Builtin::BI__sync_bool_compare_and_swap: 747 BuiltinIndex = 11; 748 NumFixed = 2; 749 break; 750 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 751 case Builtin::BI__sync_lock_release: 752 BuiltinIndex = 13; 753 NumFixed = 0; 754 break; 755 } 756 757 // Now that we know how many fixed arguments we expect, first check that we 758 // have at least that many. 759 if (TheCall->getNumArgs() < 1+NumFixed) 760 return Diag(TheCall->getLocEnd(), 761 diag::err_typecheck_call_too_few_args_at_least) 762 << 0 << 1+NumFixed << TheCall->getNumArgs() 763 << TheCall->getCallee()->getSourceRange(); 764 765 766 // Get the decl for the concrete builtin from this, we can tell what the 767 // concrete integer type we should convert to is. 768 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 769 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 770 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 771 FunctionDecl *NewBuiltinDecl = 772 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 773 TUScope, false, DRE->getLocStart())); 774 const FunctionProtoType *BuiltinFT = 775 NewBuiltinDecl->getType()->getAs<FunctionProtoType>(); 776 ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType(); 777 778 // If the first type needs to be converted (e.g. void** -> int*), do it now. 779 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 780 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_BitCast); 781 TheCall->setArg(0, FirstArg); 782 } 783 784 // Next, walk the valid ones promoting to the right type. 785 for (unsigned i = 0; i != NumFixed; ++i) { 786 Expr *Arg = TheCall->getArg(i+1); 787 788 // If the argument is an implicit cast, then there was a promotion due to 789 // "...", just remove it now. 790 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 791 Arg = ICE->getSubExpr(); 792 ICE->setSubExpr(0); 793 ICE->Destroy(Context); 794 TheCall->setArg(i+1, Arg); 795 } 796 797 // GCC does an implicit conversion to the pointer or integer ValType. This 798 // can fail in some cases (1i -> int**), check for this error case now. 799 CastExpr::CastKind Kind = CastExpr::CK_Unknown; 800 CXXBaseSpecifierArray BasePath; 801 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, BasePath)) 802 return true; 803 804 // Okay, we have something that *can* be converted to the right type. Check 805 // to see if there is a potentially weird extension going on here. This can 806 // happen when you do an atomic operation on something like an char* and 807 // pass in 42. The 42 gets converted to char. This is even more strange 808 // for things like 45.123 -> char, etc. 809 // FIXME: Do this check. 810 ImpCastExprToType(Arg, ValType, Kind); 811 TheCall->setArg(i+1, Arg); 812 } 813 814 // Switch the DeclRefExpr to refer to the new decl. 815 DRE->setDecl(NewBuiltinDecl); 816 DRE->setType(NewBuiltinDecl->getType()); 817 818 // Set the callee in the CallExpr. 819 // FIXME: This leaks the original parens and implicit casts. 820 Expr *PromotedCall = DRE; 821 UsualUnaryConversions(PromotedCall); 822 TheCall->setCallee(PromotedCall); 823 824 825 // Change the result type of the call to match the result type of the decl. 826 TheCall->setType(NewBuiltinDecl->getResultType()); 827 return false; 828} 829 830 831/// CheckObjCString - Checks that the argument to the builtin 832/// CFString constructor is correct 833/// FIXME: GCC currently emits the following warning: 834/// "warning: input conversion stopped due to an input byte that does not 835/// belong to the input codeset UTF-8" 836/// Note: It might also make sense to do the UTF-16 conversion here (would 837/// simplify the backend). 838bool Sema::CheckObjCString(Expr *Arg) { 839 Arg = Arg->IgnoreParenCasts(); 840 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 841 842 if (!Literal || Literal->isWide()) { 843 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 844 << Arg->getSourceRange(); 845 return true; 846 } 847 848 const char *Data = Literal->getStrData(); 849 unsigned Length = Literal->getByteLength(); 850 851 for (unsigned i = 0; i < Length; ++i) { 852 if (!Data[i]) { 853 Diag(getLocationOfStringLiteralByte(Literal, i), 854 diag::warn_cfstring_literal_contains_nul_character) 855 << Arg->getSourceRange(); 856 break; 857 } 858 } 859 860 return false; 861} 862 863/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 864/// Emit an error and return true on failure, return false on success. 865bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 866 Expr *Fn = TheCall->getCallee(); 867 if (TheCall->getNumArgs() > 2) { 868 Diag(TheCall->getArg(2)->getLocStart(), 869 diag::err_typecheck_call_too_many_args) 870 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 871 << Fn->getSourceRange() 872 << SourceRange(TheCall->getArg(2)->getLocStart(), 873 (*(TheCall->arg_end()-1))->getLocEnd()); 874 return true; 875 } 876 877 if (TheCall->getNumArgs() < 2) { 878 return Diag(TheCall->getLocEnd(), 879 diag::err_typecheck_call_too_few_args_at_least) 880 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 881 } 882 883 // Determine whether the current function is variadic or not. 884 BlockScopeInfo *CurBlock = getCurBlock(); 885 bool isVariadic; 886 if (CurBlock) 887 isVariadic = CurBlock->TheDecl->isVariadic(); 888 else if (FunctionDecl *FD = getCurFunctionDecl()) 889 isVariadic = FD->isVariadic(); 890 else 891 isVariadic = getCurMethodDecl()->isVariadic(); 892 893 if (!isVariadic) { 894 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 895 return true; 896 } 897 898 // Verify that the second argument to the builtin is the last argument of the 899 // current function or method. 900 bool SecondArgIsLastNamedArgument = false; 901 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 902 903 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 904 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 905 // FIXME: This isn't correct for methods (results in bogus warning). 906 // Get the last formal in the current function. 907 const ParmVarDecl *LastArg; 908 if (CurBlock) 909 LastArg = *(CurBlock->TheDecl->param_end()-1); 910 else if (FunctionDecl *FD = getCurFunctionDecl()) 911 LastArg = *(FD->param_end()-1); 912 else 913 LastArg = *(getCurMethodDecl()->param_end()-1); 914 SecondArgIsLastNamedArgument = PV == LastArg; 915 } 916 } 917 918 if (!SecondArgIsLastNamedArgument) 919 Diag(TheCall->getArg(1)->getLocStart(), 920 diag::warn_second_parameter_of_va_start_not_last_named_argument); 921 return false; 922} 923 924/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 925/// friends. This is declared to take (...), so we have to check everything. 926bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 927 if (TheCall->getNumArgs() < 2) 928 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 929 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 930 if (TheCall->getNumArgs() > 2) 931 return Diag(TheCall->getArg(2)->getLocStart(), 932 diag::err_typecheck_call_too_many_args) 933 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 934 << SourceRange(TheCall->getArg(2)->getLocStart(), 935 (*(TheCall->arg_end()-1))->getLocEnd()); 936 937 Expr *OrigArg0 = TheCall->getArg(0); 938 Expr *OrigArg1 = TheCall->getArg(1); 939 940 // Do standard promotions between the two arguments, returning their common 941 // type. 942 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 943 944 // Make sure any conversions are pushed back into the call; this is 945 // type safe since unordered compare builtins are declared as "_Bool 946 // foo(...)". 947 TheCall->setArg(0, OrigArg0); 948 TheCall->setArg(1, OrigArg1); 949 950 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 951 return false; 952 953 // If the common type isn't a real floating type, then the arguments were 954 // invalid for this operation. 955 if (!Res->isRealFloatingType()) 956 return Diag(OrigArg0->getLocStart(), 957 diag::err_typecheck_call_invalid_ordered_compare) 958 << OrigArg0->getType() << OrigArg1->getType() 959 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 960 961 return false; 962} 963 964/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 965/// __builtin_isnan and friends. This is declared to take (...), so we have 966/// to check everything. We expect the last argument to be a floating point 967/// value. 968bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 969 if (TheCall->getNumArgs() < NumArgs) 970 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 971 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 972 if (TheCall->getNumArgs() > NumArgs) 973 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 974 diag::err_typecheck_call_too_many_args) 975 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 976 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 977 (*(TheCall->arg_end()-1))->getLocEnd()); 978 979 Expr *OrigArg = TheCall->getArg(NumArgs-1); 980 981 if (OrigArg->isTypeDependent()) 982 return false; 983 984 // This operation requires a non-_Complex floating-point number. 985 if (!OrigArg->getType()->isRealFloatingType()) 986 return Diag(OrigArg->getLocStart(), 987 diag::err_typecheck_call_invalid_unary_fp) 988 << OrigArg->getType() << OrigArg->getSourceRange(); 989 990 // If this is an implicit conversion from float -> double, remove it. 991 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 992 Expr *CastArg = Cast->getSubExpr(); 993 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 994 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 995 "promotion from float to double is the only expected cast here"); 996 Cast->setSubExpr(0); 997 Cast->Destroy(Context); 998 TheCall->setArg(NumArgs-1, CastArg); 999 OrigArg = CastArg; 1000 } 1001 } 1002 1003 return false; 1004} 1005 1006/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1007// This is declared to take (...), so we have to check everything. 1008Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1009 if (TheCall->getNumArgs() < 2) 1010 return ExprError(Diag(TheCall->getLocEnd(), 1011 diag::err_typecheck_call_too_few_args_at_least) 1012 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1013 << TheCall->getSourceRange()); 1014 1015 // Determine which of the following types of shufflevector we're checking: 1016 // 1) unary, vector mask: (lhs, mask) 1017 // 2) binary, vector mask: (lhs, rhs, mask) 1018 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1019 QualType resType = TheCall->getArg(0)->getType(); 1020 unsigned numElements = 0; 1021 1022 if (!TheCall->getArg(0)->isTypeDependent() && 1023 !TheCall->getArg(1)->isTypeDependent()) { 1024 QualType LHSType = TheCall->getArg(0)->getType(); 1025 QualType RHSType = TheCall->getArg(1)->getType(); 1026 1027 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1028 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1029 << SourceRange(TheCall->getArg(0)->getLocStart(), 1030 TheCall->getArg(1)->getLocEnd()); 1031 return ExprError(); 1032 } 1033 1034 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1035 unsigned numResElements = TheCall->getNumArgs() - 2; 1036 1037 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1038 // with mask. If so, verify that RHS is an integer vector type with the 1039 // same number of elts as lhs. 1040 if (TheCall->getNumArgs() == 2) { 1041 if (!RHSType->isIntegerType() || 1042 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1043 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1044 << SourceRange(TheCall->getArg(1)->getLocStart(), 1045 TheCall->getArg(1)->getLocEnd()); 1046 numResElements = numElements; 1047 } 1048 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1049 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1050 << SourceRange(TheCall->getArg(0)->getLocStart(), 1051 TheCall->getArg(1)->getLocEnd()); 1052 return ExprError(); 1053 } else if (numElements != numResElements) { 1054 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1055 resType = Context.getVectorType(eltType, numResElements, false, false); 1056 } 1057 } 1058 1059 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1060 if (TheCall->getArg(i)->isTypeDependent() || 1061 TheCall->getArg(i)->isValueDependent()) 1062 continue; 1063 1064 llvm::APSInt Result(32); 1065 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1066 return ExprError(Diag(TheCall->getLocStart(), 1067 diag::err_shufflevector_nonconstant_argument) 1068 << TheCall->getArg(i)->getSourceRange()); 1069 1070 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1071 return ExprError(Diag(TheCall->getLocStart(), 1072 diag::err_shufflevector_argument_too_large) 1073 << TheCall->getArg(i)->getSourceRange()); 1074 } 1075 1076 llvm::SmallVector<Expr*, 32> exprs; 1077 1078 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1079 exprs.push_back(TheCall->getArg(i)); 1080 TheCall->setArg(i, 0); 1081 } 1082 1083 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1084 exprs.size(), resType, 1085 TheCall->getCallee()->getLocStart(), 1086 TheCall->getRParenLoc())); 1087} 1088 1089/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1090// This is declared to take (const void*, ...) and can take two 1091// optional constant int args. 1092bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1093 unsigned NumArgs = TheCall->getNumArgs(); 1094 1095 if (NumArgs > 3) 1096 return Diag(TheCall->getLocEnd(), 1097 diag::err_typecheck_call_too_many_args_at_most) 1098 << 0 /*function call*/ << 3 << NumArgs 1099 << TheCall->getSourceRange(); 1100 1101 // Argument 0 is checked for us and the remaining arguments must be 1102 // constant integers. 1103 for (unsigned i = 1; i != NumArgs; ++i) { 1104 Expr *Arg = TheCall->getArg(i); 1105 1106 llvm::APSInt Result; 1107 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1108 return true; 1109 1110 // FIXME: gcc issues a warning and rewrites these to 0. These 1111 // seems especially odd for the third argument since the default 1112 // is 3. 1113 if (i == 1) { 1114 if (Result.getLimitedValue() > 1) 1115 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1116 << "0" << "1" << Arg->getSourceRange(); 1117 } else { 1118 if (Result.getLimitedValue() > 3) 1119 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1120 << "0" << "3" << Arg->getSourceRange(); 1121 } 1122 } 1123 1124 return false; 1125} 1126 1127/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1128/// TheCall is a constant expression. 1129bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1130 llvm::APSInt &Result) { 1131 Expr *Arg = TheCall->getArg(ArgNum); 1132 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1133 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1134 1135 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1136 1137 if (!Arg->isIntegerConstantExpr(Result, Context)) 1138 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1139 << FDecl->getDeclName() << Arg->getSourceRange(); 1140 1141 return false; 1142} 1143 1144/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1145/// int type). This simply type checks that type is one of the defined 1146/// constants (0-3). 1147// For compatability check 0-3, llvm only handles 0 and 2. 1148bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1149 llvm::APSInt Result; 1150 1151 // Check constant-ness first. 1152 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1153 return true; 1154 1155 Expr *Arg = TheCall->getArg(1); 1156 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1157 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1158 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1159 } 1160 1161 return false; 1162} 1163 1164/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1165/// This checks that val is a constant 1. 1166bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1167 Expr *Arg = TheCall->getArg(1); 1168 llvm::APSInt Result; 1169 1170 // TODO: This is less than ideal. Overload this to take a value. 1171 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1172 return true; 1173 1174 if (Result != 1) 1175 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1176 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1177 1178 return false; 1179} 1180 1181// Handle i > 1 ? "x" : "y", recursivelly 1182bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 1183 bool HasVAListArg, 1184 unsigned format_idx, unsigned firstDataArg) { 1185 if (E->isTypeDependent() || E->isValueDependent()) 1186 return false; 1187 1188 switch (E->getStmtClass()) { 1189 case Stmt::ConditionalOperatorClass: { 1190 const ConditionalOperator *C = cast<ConditionalOperator>(E); 1191 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, 1192 HasVAListArg, format_idx, firstDataArg) 1193 && SemaCheckStringLiteral(C->getRHS(), TheCall, 1194 HasVAListArg, format_idx, firstDataArg); 1195 } 1196 1197 case Stmt::ImplicitCastExprClass: { 1198 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 1199 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 1200 format_idx, firstDataArg); 1201 } 1202 1203 case Stmt::ParenExprClass: { 1204 const ParenExpr *Expr = cast<ParenExpr>(E); 1205 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 1206 format_idx, firstDataArg); 1207 } 1208 1209 case Stmt::DeclRefExprClass: { 1210 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1211 1212 // As an exception, do not flag errors for variables binding to 1213 // const string literals. 1214 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1215 bool isConstant = false; 1216 QualType T = DR->getType(); 1217 1218 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1219 isConstant = AT->getElementType().isConstant(Context); 1220 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1221 isConstant = T.isConstant(Context) && 1222 PT->getPointeeType().isConstant(Context); 1223 } 1224 1225 if (isConstant) { 1226 if (const Expr *Init = VD->getAnyInitializer()) 1227 return SemaCheckStringLiteral(Init, TheCall, 1228 HasVAListArg, format_idx, firstDataArg); 1229 } 1230 1231 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1232 // special check to see if the format string is a function parameter 1233 // of the function calling the printf function. If the function 1234 // has an attribute indicating it is a printf-like function, then we 1235 // should suppress warnings concerning non-literals being used in a call 1236 // to a vprintf function. For example: 1237 // 1238 // void 1239 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1240 // va_list ap; 1241 // va_start(ap, fmt); 1242 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1243 // ... 1244 // 1245 // 1246 // FIXME: We don't have full attribute support yet, so just check to see 1247 // if the argument is a DeclRefExpr that references a parameter. We'll 1248 // add proper support for checking the attribute later. 1249 if (HasVAListArg) 1250 if (isa<ParmVarDecl>(VD)) 1251 return true; 1252 } 1253 1254 return false; 1255 } 1256 1257 case Stmt::CallExprClass: { 1258 const CallExpr *CE = cast<CallExpr>(E); 1259 if (const ImplicitCastExpr *ICE 1260 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1261 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1262 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1263 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1264 unsigned ArgIndex = FA->getFormatIdx(); 1265 const Expr *Arg = CE->getArg(ArgIndex - 1); 1266 1267 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1268 format_idx, firstDataArg); 1269 } 1270 } 1271 } 1272 } 1273 1274 return false; 1275 } 1276 case Stmt::ObjCStringLiteralClass: 1277 case Stmt::StringLiteralClass: { 1278 const StringLiteral *StrE = NULL; 1279 1280 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1281 StrE = ObjCFExpr->getString(); 1282 else 1283 StrE = cast<StringLiteral>(E); 1284 1285 if (StrE) { 1286 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 1287 firstDataArg); 1288 return true; 1289 } 1290 1291 return false; 1292 } 1293 1294 default: 1295 return false; 1296 } 1297} 1298 1299void 1300Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1301 const CallExpr *TheCall) { 1302 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 1303 i != e; ++i) { 1304 const Expr *ArgExpr = TheCall->getArg(*i); 1305 if (ArgExpr->isNullPointerConstant(Context, 1306 Expr::NPC_ValueDependentIsNotNull)) 1307 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 1308 << ArgExpr->getSourceRange(); 1309 } 1310} 1311 1312/// CheckPrintfArguments - Check calls to printf (and similar functions) for 1313/// correct use of format strings. 1314/// 1315/// HasVAListArg - A predicate indicating whether the printf-like 1316/// function is passed an explicit va_arg argument (e.g., vprintf) 1317/// 1318/// format_idx - The index into Args for the format string. 1319/// 1320/// Improper format strings to functions in the printf family can be 1321/// the source of bizarre bugs and very serious security holes. A 1322/// good source of information is available in the following paper 1323/// (which includes additional references): 1324/// 1325/// FormatGuard: Automatic Protection From printf Format String 1326/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 1327/// 1328/// TODO: 1329/// Functionality implemented: 1330/// 1331/// We can statically check the following properties for string 1332/// literal format strings for non v.*printf functions (where the 1333/// arguments are passed directly): 1334// 1335/// (1) Are the number of format conversions equal to the number of 1336/// data arguments? 1337/// 1338/// (2) Does each format conversion correctly match the type of the 1339/// corresponding data argument? 1340/// 1341/// Moreover, for all printf functions we can: 1342/// 1343/// (3) Check for a missing format string (when not caught by type checking). 1344/// 1345/// (4) Check for no-operation flags; e.g. using "#" with format 1346/// conversion 'c' (TODO) 1347/// 1348/// (5) Check the use of '%n', a major source of security holes. 1349/// 1350/// (6) Check for malformed format conversions that don't specify anything. 1351/// 1352/// (7) Check for empty format strings. e.g: printf(""); 1353/// 1354/// (8) Check that the format string is a wide literal. 1355/// 1356/// All of these checks can be done by parsing the format string. 1357/// 1358void 1359Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 1360 unsigned format_idx, unsigned firstDataArg) { 1361 const Expr *Fn = TheCall->getCallee(); 1362 1363 // The way the format attribute works in GCC, the implicit this argument 1364 // of member functions is counted. However, it doesn't appear in our own 1365 // lists, so decrement format_idx in that case. 1366 if (isa<CXXMemberCallExpr>(TheCall)) { 1367 // Catch a format attribute mistakenly referring to the object argument. 1368 if (format_idx == 0) 1369 return; 1370 --format_idx; 1371 if(firstDataArg != 0) 1372 --firstDataArg; 1373 } 1374 1375 // CHECK: printf-like function is called with no format string. 1376 if (format_idx >= TheCall->getNumArgs()) { 1377 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 1378 << Fn->getSourceRange(); 1379 return; 1380 } 1381 1382 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1383 1384 // CHECK: format string is not a string literal. 1385 // 1386 // Dynamically generated format strings are difficult to 1387 // automatically vet at compile time. Requiring that format strings 1388 // are string literals: (1) permits the checking of format strings by 1389 // the compiler and thereby (2) can practically remove the source of 1390 // many format string exploits. 1391 1392 // Format string can be either ObjC string (e.g. @"%d") or 1393 // C string (e.g. "%d") 1394 // ObjC string uses the same format specifiers as C string, so we can use 1395 // the same format string checking logic for both ObjC and C strings. 1396 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1397 firstDataArg)) 1398 return; // Literal format string found, check done! 1399 1400 // If there are no arguments specified, warn with -Wformat-security, otherwise 1401 // warn only with -Wformat-nonliteral. 1402 if (TheCall->getNumArgs() == format_idx+1) 1403 Diag(TheCall->getArg(format_idx)->getLocStart(), 1404 diag::warn_printf_nonliteral_noargs) 1405 << OrigFormatExpr->getSourceRange(); 1406 else 1407 Diag(TheCall->getArg(format_idx)->getLocStart(), 1408 diag::warn_printf_nonliteral) 1409 << OrigFormatExpr->getSourceRange(); 1410} 1411 1412namespace { 1413class CheckPrintfHandler : public analyze_printf::FormatStringHandler { 1414 Sema &S; 1415 const StringLiteral *FExpr; 1416 const Expr *OrigFormatExpr; 1417 const unsigned FirstDataArg; 1418 const unsigned NumDataArgs; 1419 const bool IsObjCLiteral; 1420 const char *Beg; // Start of format string. 1421 const bool HasVAListArg; 1422 const CallExpr *TheCall; 1423 unsigned FormatIdx; 1424 llvm::BitVector CoveredArgs; 1425 bool usesPositionalArgs; 1426 bool atFirstArg; 1427public: 1428 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1429 const Expr *origFormatExpr, unsigned firstDataArg, 1430 unsigned numDataArgs, bool isObjCLiteral, 1431 const char *beg, bool hasVAListArg, 1432 const CallExpr *theCall, unsigned formatIdx) 1433 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1434 FirstDataArg(firstDataArg), 1435 NumDataArgs(numDataArgs), 1436 IsObjCLiteral(isObjCLiteral), Beg(beg), 1437 HasVAListArg(hasVAListArg), 1438 TheCall(theCall), FormatIdx(formatIdx), 1439 usesPositionalArgs(false), atFirstArg(true) { 1440 CoveredArgs.resize(numDataArgs); 1441 CoveredArgs.reset(); 1442 } 1443 1444 void DoneProcessing(); 1445 1446 void HandleIncompleteFormatSpecifier(const char *startSpecifier, 1447 unsigned specifierLen); 1448 1449 bool 1450 HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1451 const char *startSpecifier, 1452 unsigned specifierLen); 1453 1454 virtual void HandleInvalidPosition(const char *startSpecifier, 1455 unsigned specifierLen, 1456 analyze_printf::PositionContext p); 1457 1458 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1459 1460 void HandleNullChar(const char *nullCharacter); 1461 1462 bool HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1463 const char *startSpecifier, 1464 unsigned specifierLen); 1465private: 1466 SourceRange getFormatStringRange(); 1467 SourceRange getFormatSpecifierRange(const char *startSpecifier, 1468 unsigned specifierLen); 1469 SourceLocation getLocationOfByte(const char *x); 1470 1471 bool HandleAmount(const analyze_printf::OptionalAmount &Amt, unsigned k, 1472 const char *startSpecifier, unsigned specifierLen); 1473 void HandleFlags(const analyze_printf::FormatSpecifier &FS, 1474 llvm::StringRef flag, llvm::StringRef cspec, 1475 const char *startSpecifier, unsigned specifierLen); 1476 1477 const Expr *getDataArg(unsigned i) const; 1478}; 1479} 1480 1481SourceRange CheckPrintfHandler::getFormatStringRange() { 1482 return OrigFormatExpr->getSourceRange(); 1483} 1484 1485SourceRange CheckPrintfHandler:: 1486getFormatSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1487 return SourceRange(getLocationOfByte(startSpecifier), 1488 getLocationOfByte(startSpecifier+specifierLen-1)); 1489} 1490 1491SourceLocation CheckPrintfHandler::getLocationOfByte(const char *x) { 1492 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1493} 1494 1495void CheckPrintfHandler:: 1496HandleIncompleteFormatSpecifier(const char *startSpecifier, 1497 unsigned specifierLen) { 1498 SourceLocation Loc = getLocationOfByte(startSpecifier); 1499 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1500 << getFormatSpecifierRange(startSpecifier, specifierLen); 1501} 1502 1503void 1504CheckPrintfHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1505 analyze_printf::PositionContext p) { 1506 SourceLocation Loc = getLocationOfByte(startPos); 1507 S.Diag(Loc, diag::warn_printf_invalid_positional_specifier) 1508 << (unsigned) p << getFormatSpecifierRange(startPos, posLen); 1509} 1510 1511void CheckPrintfHandler::HandleZeroPosition(const char *startPos, 1512 unsigned posLen) { 1513 SourceLocation Loc = getLocationOfByte(startPos); 1514 S.Diag(Loc, diag::warn_printf_zero_positional_specifier) 1515 << getFormatSpecifierRange(startPos, posLen); 1516} 1517 1518bool CheckPrintfHandler:: 1519HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1520 const char *startSpecifier, 1521 unsigned specifierLen) { 1522 1523 unsigned argIndex = FS.getArgIndex(); 1524 bool keepGoing = true; 1525 if (argIndex < NumDataArgs) { 1526 // Consider the argument coverered, even though the specifier doesn't 1527 // make sense. 1528 CoveredArgs.set(argIndex); 1529 } 1530 else { 1531 // If argIndex exceeds the number of data arguments we 1532 // don't issue a warning because that is just a cascade of warnings (and 1533 // they may have intended '%%' anyway). We don't want to continue processing 1534 // the format string after this point, however, as we will like just get 1535 // gibberish when trying to match arguments. 1536 keepGoing = false; 1537 } 1538 1539 const analyze_printf::ConversionSpecifier &CS = 1540 FS.getConversionSpecifier(); 1541 SourceLocation Loc = getLocationOfByte(CS.getStart()); 1542 S.Diag(Loc, diag::warn_printf_invalid_conversion) 1543 << llvm::StringRef(CS.getStart(), CS.getLength()) 1544 << getFormatSpecifierRange(startSpecifier, specifierLen); 1545 1546 return keepGoing; 1547} 1548 1549void CheckPrintfHandler::HandleNullChar(const char *nullCharacter) { 1550 // The presence of a null character is likely an error. 1551 S.Diag(getLocationOfByte(nullCharacter), 1552 diag::warn_printf_format_string_contains_null_char) 1553 << getFormatStringRange(); 1554} 1555 1556const Expr *CheckPrintfHandler::getDataArg(unsigned i) const { 1557 return TheCall->getArg(FirstDataArg + i); 1558} 1559 1560void CheckPrintfHandler::HandleFlags(const analyze_printf::FormatSpecifier &FS, 1561 llvm::StringRef flag, 1562 llvm::StringRef cspec, 1563 const char *startSpecifier, 1564 unsigned specifierLen) { 1565 const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier(); 1566 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_nonsensical_flag) 1567 << flag << cspec << getFormatSpecifierRange(startSpecifier, specifierLen); 1568} 1569 1570bool 1571CheckPrintfHandler::HandleAmount(const analyze_printf::OptionalAmount &Amt, 1572 unsigned k, const char *startSpecifier, 1573 unsigned specifierLen) { 1574 1575 if (Amt.hasDataArgument()) { 1576 if (!HasVAListArg) { 1577 unsigned argIndex = Amt.getArgIndex(); 1578 if (argIndex >= NumDataArgs) { 1579 S.Diag(getLocationOfByte(Amt.getStart()), 1580 diag::warn_printf_asterisk_missing_arg) 1581 << k << getFormatSpecifierRange(startSpecifier, specifierLen); 1582 // Don't do any more checking. We will just emit 1583 // spurious errors. 1584 return false; 1585 } 1586 1587 // Type check the data argument. It should be an 'int'. 1588 // Although not in conformance with C99, we also allow the argument to be 1589 // an 'unsigned int' as that is a reasonably safe case. GCC also 1590 // doesn't emit a warning for that case. 1591 CoveredArgs.set(argIndex); 1592 const Expr *Arg = getDataArg(argIndex); 1593 QualType T = Arg->getType(); 1594 1595 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1596 assert(ATR.isValid()); 1597 1598 if (!ATR.matchesType(S.Context, T)) { 1599 S.Diag(getLocationOfByte(Amt.getStart()), 1600 diag::warn_printf_asterisk_wrong_type) 1601 << k 1602 << ATR.getRepresentativeType(S.Context) << T 1603 << getFormatSpecifierRange(startSpecifier, specifierLen) 1604 << Arg->getSourceRange(); 1605 // Don't do any more checking. We will just emit 1606 // spurious errors. 1607 return false; 1608 } 1609 } 1610 } 1611 return true; 1612} 1613 1614bool 1615CheckPrintfHandler::HandleFormatSpecifier(const analyze_printf::FormatSpecifier 1616 &FS, 1617 const char *startSpecifier, 1618 unsigned specifierLen) { 1619 1620 using namespace analyze_printf; 1621 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1622 1623 if (atFirstArg) { 1624 atFirstArg = false; 1625 usesPositionalArgs = FS.usesPositionalArg(); 1626 } 1627 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1628 // Cannot mix-and-match positional and non-positional arguments. 1629 S.Diag(getLocationOfByte(CS.getStart()), 1630 diag::warn_printf_mix_positional_nonpositional_args) 1631 << getFormatSpecifierRange(startSpecifier, specifierLen); 1632 return false; 1633 } 1634 1635 // First check if the field width, precision, and conversion specifier 1636 // have matching data arguments. 1637 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1638 startSpecifier, specifierLen)) { 1639 return false; 1640 } 1641 1642 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1643 startSpecifier, specifierLen)) { 1644 return false; 1645 } 1646 1647 if (!CS.consumesDataArgument()) { 1648 // FIXME: Technically specifying a precision or field width here 1649 // makes no sense. Worth issuing a warning at some point. 1650 return true; 1651 } 1652 1653 // Consume the argument. 1654 unsigned argIndex = FS.getArgIndex(); 1655 if (argIndex < NumDataArgs) { 1656 // The check to see if the argIndex is valid will come later. 1657 // We set the bit here because we may exit early from this 1658 // function if we encounter some other error. 1659 CoveredArgs.set(argIndex); 1660 } 1661 1662 // Check for using an Objective-C specific conversion specifier 1663 // in a non-ObjC literal. 1664 if (!IsObjCLiteral && CS.isObjCArg()) { 1665 return HandleInvalidConversionSpecifier(FS, startSpecifier, specifierLen); 1666 } 1667 1668 // Are we using '%n'? Issue a warning about this being 1669 // a possible security issue. 1670 if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) { 1671 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1672 << getFormatSpecifierRange(startSpecifier, specifierLen); 1673 // Continue checking the other format specifiers. 1674 return true; 1675 } 1676 1677 if (CS.getKind() == ConversionSpecifier::VoidPtrArg) { 1678 if (FS.getPrecision().getHowSpecified() != OptionalAmount::NotSpecified) 1679 S.Diag(getLocationOfByte(CS.getStart()), 1680 diag::warn_printf_nonsensical_precision) 1681 << CS.getCharacters() 1682 << getFormatSpecifierRange(startSpecifier, specifierLen); 1683 } 1684 if (CS.getKind() == ConversionSpecifier::VoidPtrArg || 1685 CS.getKind() == ConversionSpecifier::CStrArg) { 1686 // FIXME: Instead of using "0", "+", etc., eventually get them from 1687 // the FormatSpecifier. 1688 if (FS.hasLeadingZeros()) 1689 HandleFlags(FS, "0", CS.getCharacters(), startSpecifier, specifierLen); 1690 if (FS.hasPlusPrefix()) 1691 HandleFlags(FS, "+", CS.getCharacters(), startSpecifier, specifierLen); 1692 if (FS.hasSpacePrefix()) 1693 HandleFlags(FS, " ", CS.getCharacters(), startSpecifier, specifierLen); 1694 } 1695 1696 // The remaining checks depend on the data arguments. 1697 if (HasVAListArg) 1698 return true; 1699 1700 if (argIndex >= NumDataArgs) { 1701 if (FS.usesPositionalArg()) { 1702 S.Diag(getLocationOfByte(CS.getStart()), 1703 diag::warn_printf_positional_arg_exceeds_data_args) 1704 << (argIndex+1) << NumDataArgs 1705 << getFormatSpecifierRange(startSpecifier, specifierLen); 1706 } 1707 else { 1708 S.Diag(getLocationOfByte(CS.getStart()), 1709 diag::warn_printf_insufficient_data_args) 1710 << getFormatSpecifierRange(startSpecifier, specifierLen); 1711 } 1712 1713 // Don't do any more checking. 1714 return false; 1715 } 1716 1717 // Now type check the data expression that matches the 1718 // format specifier. 1719 const Expr *Ex = getDataArg(argIndex); 1720 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1721 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1722 // Check if we didn't match because of an implicit cast from a 'char' 1723 // or 'short' to an 'int'. This is done because printf is a varargs 1724 // function. 1725 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1726 if (ICE->getType() == S.Context.IntTy) 1727 if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType())) 1728 return true; 1729 1730 // We may be able to offer a FixItHint if it is a supported type. 1731 FormatSpecifier fixedFS = FS; 1732 bool success = fixedFS.fixType(Ex->getType()); 1733 1734 if (success) { 1735 // Get the fix string from the fixed format specifier 1736 llvm::SmallString<128> buf; 1737 llvm::raw_svector_ostream os(buf); 1738 fixedFS.toString(os); 1739 1740 S.Diag(getLocationOfByte(CS.getStart()), 1741 diag::warn_printf_conversion_argument_type_mismatch) 1742 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1743 << getFormatSpecifierRange(startSpecifier, specifierLen) 1744 << Ex->getSourceRange() 1745 << FixItHint::CreateReplacement( 1746 getFormatSpecifierRange(startSpecifier, specifierLen), 1747 os.str()); 1748 } 1749 else { 1750 S.Diag(getLocationOfByte(CS.getStart()), 1751 diag::warn_printf_conversion_argument_type_mismatch) 1752 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1753 << getFormatSpecifierRange(startSpecifier, specifierLen) 1754 << Ex->getSourceRange(); 1755 } 1756 } 1757 1758 return true; 1759} 1760 1761void CheckPrintfHandler::DoneProcessing() { 1762 // Does the number of data arguments exceed the number of 1763 // format conversions in the format string? 1764 if (!HasVAListArg) { 1765 // Find any arguments that weren't covered. 1766 CoveredArgs.flip(); 1767 signed notCoveredArg = CoveredArgs.find_first(); 1768 if (notCoveredArg >= 0) { 1769 assert((unsigned)notCoveredArg < NumDataArgs); 1770 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1771 diag::warn_printf_data_arg_not_used) 1772 << getFormatStringRange(); 1773 } 1774 } 1775} 1776 1777void Sema::CheckPrintfString(const StringLiteral *FExpr, 1778 const Expr *OrigFormatExpr, 1779 const CallExpr *TheCall, bool HasVAListArg, 1780 unsigned format_idx, unsigned firstDataArg) { 1781 1782 // CHECK: is the format string a wide literal? 1783 if (FExpr->isWide()) { 1784 Diag(FExpr->getLocStart(), 1785 diag::warn_printf_format_string_is_wide_literal) 1786 << OrigFormatExpr->getSourceRange(); 1787 return; 1788 } 1789 1790 // Str - The format string. NOTE: this is NOT null-terminated! 1791 const char *Str = FExpr->getStrData(); 1792 1793 // CHECK: empty format string? 1794 unsigned StrLen = FExpr->getByteLength(); 1795 1796 if (StrLen == 0) { 1797 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1798 << OrigFormatExpr->getSourceRange(); 1799 return; 1800 } 1801 1802 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1803 TheCall->getNumArgs() - firstDataArg, 1804 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1805 HasVAListArg, TheCall, format_idx); 1806 1807 if (!analyze_printf::ParseFormatString(H, Str, Str + StrLen)) 1808 H.DoneProcessing(); 1809} 1810 1811//===--- CHECK: Return Address of Stack Variable --------------------------===// 1812 1813static DeclRefExpr* EvalVal(Expr *E); 1814static DeclRefExpr* EvalAddr(Expr* E); 1815 1816/// CheckReturnStackAddr - Check if a return statement returns the address 1817/// of a stack variable. 1818void 1819Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1820 SourceLocation ReturnLoc) { 1821 1822 // Perform checking for returned stack addresses. 1823 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1824 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1825 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1826 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1827 1828 // Skip over implicit cast expressions when checking for block expressions. 1829 RetValExp = RetValExp->IgnoreParenCasts(); 1830 1831 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1832 if (C->hasBlockDeclRefExprs()) 1833 Diag(C->getLocStart(), diag::err_ret_local_block) 1834 << C->getSourceRange(); 1835 1836 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1837 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1838 << ALE->getSourceRange(); 1839 1840 } else if (lhsType->isReferenceType()) { 1841 // Perform checking for stack values returned by reference. 1842 // Check for a reference to the stack 1843 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1844 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1845 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1846 } 1847} 1848 1849/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1850/// check if the expression in a return statement evaluates to an address 1851/// to a location on the stack. The recursion is used to traverse the 1852/// AST of the return expression, with recursion backtracking when we 1853/// encounter a subexpression that (1) clearly does not lead to the address 1854/// of a stack variable or (2) is something we cannot determine leads to 1855/// the address of a stack variable based on such local checking. 1856/// 1857/// EvalAddr processes expressions that are pointers that are used as 1858/// references (and not L-values). EvalVal handles all other values. 1859/// At the base case of the recursion is a check for a DeclRefExpr* in 1860/// the refers to a stack variable. 1861/// 1862/// This implementation handles: 1863/// 1864/// * pointer-to-pointer casts 1865/// * implicit conversions from array references to pointers 1866/// * taking the address of fields 1867/// * arbitrary interplay between "&" and "*" operators 1868/// * pointer arithmetic from an address of a stack variable 1869/// * taking the address of an array element where the array is on the stack 1870static DeclRefExpr* EvalAddr(Expr *E) { 1871 // We should only be called for evaluating pointer expressions. 1872 assert((E->getType()->isAnyPointerType() || 1873 E->getType()->isBlockPointerType() || 1874 E->getType()->isObjCQualifiedIdType()) && 1875 "EvalAddr only works on pointers"); 1876 1877 // Our "symbolic interpreter" is just a dispatch off the currently 1878 // viewed AST node. We then recursively traverse the AST by calling 1879 // EvalAddr and EvalVal appropriately. 1880 switch (E->getStmtClass()) { 1881 case Stmt::ParenExprClass: 1882 // Ignore parentheses. 1883 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1884 1885 case Stmt::UnaryOperatorClass: { 1886 // The only unary operator that make sense to handle here 1887 // is AddrOf. All others don't make sense as pointers. 1888 UnaryOperator *U = cast<UnaryOperator>(E); 1889 1890 if (U->getOpcode() == UnaryOperator::AddrOf) 1891 return EvalVal(U->getSubExpr()); 1892 else 1893 return NULL; 1894 } 1895 1896 case Stmt::BinaryOperatorClass: { 1897 // Handle pointer arithmetic. All other binary operators are not valid 1898 // in this context. 1899 BinaryOperator *B = cast<BinaryOperator>(E); 1900 BinaryOperator::Opcode op = B->getOpcode(); 1901 1902 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1903 return NULL; 1904 1905 Expr *Base = B->getLHS(); 1906 1907 // Determine which argument is the real pointer base. It could be 1908 // the RHS argument instead of the LHS. 1909 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1910 1911 assert (Base->getType()->isPointerType()); 1912 return EvalAddr(Base); 1913 } 1914 1915 // For conditional operators we need to see if either the LHS or RHS are 1916 // valid DeclRefExpr*s. If one of them is valid, we return it. 1917 case Stmt::ConditionalOperatorClass: { 1918 ConditionalOperator *C = cast<ConditionalOperator>(E); 1919 1920 // Handle the GNU extension for missing LHS. 1921 if (Expr *lhsExpr = C->getLHS()) 1922 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1923 return LHS; 1924 1925 return EvalAddr(C->getRHS()); 1926 } 1927 1928 // For casts, we need to handle conversions from arrays to 1929 // pointer values, and pointer-to-pointer conversions. 1930 case Stmt::ImplicitCastExprClass: 1931 case Stmt::CStyleCastExprClass: 1932 case Stmt::CXXFunctionalCastExprClass: { 1933 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1934 QualType T = SubExpr->getType(); 1935 1936 if (SubExpr->getType()->isPointerType() || 1937 SubExpr->getType()->isBlockPointerType() || 1938 SubExpr->getType()->isObjCQualifiedIdType()) 1939 return EvalAddr(SubExpr); 1940 else if (T->isArrayType()) 1941 return EvalVal(SubExpr); 1942 else 1943 return 0; 1944 } 1945 1946 // C++ casts. For dynamic casts, static casts, and const casts, we 1947 // are always converting from a pointer-to-pointer, so we just blow 1948 // through the cast. In the case the dynamic cast doesn't fail (and 1949 // return NULL), we take the conservative route and report cases 1950 // where we return the address of a stack variable. For Reinterpre 1951 // FIXME: The comment about is wrong; we're not always converting 1952 // from pointer to pointer. I'm guessing that this code should also 1953 // handle references to objects. 1954 case Stmt::CXXStaticCastExprClass: 1955 case Stmt::CXXDynamicCastExprClass: 1956 case Stmt::CXXConstCastExprClass: 1957 case Stmt::CXXReinterpretCastExprClass: { 1958 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1959 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1960 return EvalAddr(S); 1961 else 1962 return NULL; 1963 } 1964 1965 // Everything else: we simply don't reason about them. 1966 default: 1967 return NULL; 1968 } 1969} 1970 1971 1972/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1973/// See the comments for EvalAddr for more details. 1974static DeclRefExpr* EvalVal(Expr *E) { 1975 1976 // We should only be called for evaluating non-pointer expressions, or 1977 // expressions with a pointer type that are not used as references but instead 1978 // are l-values (e.g., DeclRefExpr with a pointer type). 1979 1980 // Our "symbolic interpreter" is just a dispatch off the currently 1981 // viewed AST node. We then recursively traverse the AST by calling 1982 // EvalAddr and EvalVal appropriately. 1983 switch (E->getStmtClass()) { 1984 case Stmt::DeclRefExprClass: { 1985 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1986 // at code that refers to a variable's name. We check if it has local 1987 // storage within the function, and if so, return the expression. 1988 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1989 1990 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1991 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1992 1993 return NULL; 1994 } 1995 1996 case Stmt::ParenExprClass: 1997 // Ignore parentheses. 1998 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1999 2000 case Stmt::UnaryOperatorClass: { 2001 // The only unary operator that make sense to handle here 2002 // is Deref. All others don't resolve to a "name." This includes 2003 // handling all sorts of rvalues passed to a unary operator. 2004 UnaryOperator *U = cast<UnaryOperator>(E); 2005 2006 if (U->getOpcode() == UnaryOperator::Deref) 2007 return EvalAddr(U->getSubExpr()); 2008 2009 return NULL; 2010 } 2011 2012 case Stmt::ArraySubscriptExprClass: { 2013 // Array subscripts are potential references to data on the stack. We 2014 // retrieve the DeclRefExpr* for the array variable if it indeed 2015 // has local storage. 2016 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 2017 } 2018 2019 case Stmt::ConditionalOperatorClass: { 2020 // For conditional operators we need to see if either the LHS or RHS are 2021 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 2022 ConditionalOperator *C = cast<ConditionalOperator>(E); 2023 2024 // Handle the GNU extension for missing LHS. 2025 if (Expr *lhsExpr = C->getLHS()) 2026 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 2027 return LHS; 2028 2029 return EvalVal(C->getRHS()); 2030 } 2031 2032 // Accesses to members are potential references to data on the stack. 2033 case Stmt::MemberExprClass: { 2034 MemberExpr *M = cast<MemberExpr>(E); 2035 2036 // Check for indirect access. We only want direct field accesses. 2037 if (!M->isArrow()) 2038 return EvalVal(M->getBase()); 2039 else 2040 return NULL; 2041 } 2042 2043 // Everything else: we simply don't reason about them. 2044 default: 2045 return NULL; 2046 } 2047} 2048 2049//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2050 2051/// Check for comparisons of floating point operands using != and ==. 2052/// Issue a warning if these are no self-comparisons, as they are not likely 2053/// to do what the programmer intended. 2054void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2055 bool EmitWarning = true; 2056 2057 Expr* LeftExprSansParen = lex->IgnoreParens(); 2058 Expr* RightExprSansParen = rex->IgnoreParens(); 2059 2060 // Special case: check for x == x (which is OK). 2061 // Do not emit warnings for such cases. 2062 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2063 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2064 if (DRL->getDecl() == DRR->getDecl()) 2065 EmitWarning = false; 2066 2067 2068 // Special case: check for comparisons against literals that can be exactly 2069 // represented by APFloat. In such cases, do not emit a warning. This 2070 // is a heuristic: often comparison against such literals are used to 2071 // detect if a value in a variable has not changed. This clearly can 2072 // lead to false negatives. 2073 if (EmitWarning) { 2074 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2075 if (FLL->isExact()) 2076 EmitWarning = false; 2077 } else 2078 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2079 if (FLR->isExact()) 2080 EmitWarning = false; 2081 } 2082 } 2083 2084 // Check for comparisons with builtin types. 2085 if (EmitWarning) 2086 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2087 if (CL->isBuiltinCall(Context)) 2088 EmitWarning = false; 2089 2090 if (EmitWarning) 2091 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2092 if (CR->isBuiltinCall(Context)) 2093 EmitWarning = false; 2094 2095 // Emit the diagnostic. 2096 if (EmitWarning) 2097 Diag(loc, diag::warn_floatingpoint_eq) 2098 << lex->getSourceRange() << rex->getSourceRange(); 2099} 2100 2101//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2102//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2103 2104namespace { 2105 2106/// Structure recording the 'active' range of an integer-valued 2107/// expression. 2108struct IntRange { 2109 /// The number of bits active in the int. 2110 unsigned Width; 2111 2112 /// True if the int is known not to have negative values. 2113 bool NonNegative; 2114 2115 IntRange() {} 2116 IntRange(unsigned Width, bool NonNegative) 2117 : Width(Width), NonNegative(NonNegative) 2118 {} 2119 2120 // Returns the range of the bool type. 2121 static IntRange forBoolType() { 2122 return IntRange(1, true); 2123 } 2124 2125 // Returns the range of an integral type. 2126 static IntRange forType(ASTContext &C, QualType T) { 2127 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 2128 } 2129 2130 // Returns the range of an integeral type based on its canonical 2131 // representation. 2132 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 2133 assert(T->isCanonicalUnqualified()); 2134 2135 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2136 T = VT->getElementType().getTypePtr(); 2137 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2138 T = CT->getElementType().getTypePtr(); 2139 2140 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2141 EnumDecl *Enum = ET->getDecl(); 2142 unsigned NumPositive = Enum->getNumPositiveBits(); 2143 unsigned NumNegative = Enum->getNumNegativeBits(); 2144 2145 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2146 } 2147 2148 const BuiltinType *BT = cast<BuiltinType>(T); 2149 assert(BT->isInteger()); 2150 2151 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2152 } 2153 2154 // Returns the supremum of two ranges: i.e. their conservative merge. 2155 static IntRange join(IntRange L, IntRange R) { 2156 return IntRange(std::max(L.Width, R.Width), 2157 L.NonNegative && R.NonNegative); 2158 } 2159 2160 // Returns the infinum of two ranges: i.e. their aggressive merge. 2161 static IntRange meet(IntRange L, IntRange R) { 2162 return IntRange(std::min(L.Width, R.Width), 2163 L.NonNegative || R.NonNegative); 2164 } 2165}; 2166 2167IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2168 if (value.isSigned() && value.isNegative()) 2169 return IntRange(value.getMinSignedBits(), false); 2170 2171 if (value.getBitWidth() > MaxWidth) 2172 value.trunc(MaxWidth); 2173 2174 // isNonNegative() just checks the sign bit without considering 2175 // signedness. 2176 return IntRange(value.getActiveBits(), true); 2177} 2178 2179IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2180 unsigned MaxWidth) { 2181 if (result.isInt()) 2182 return GetValueRange(C, result.getInt(), MaxWidth); 2183 2184 if (result.isVector()) { 2185 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2186 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2187 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2188 R = IntRange::join(R, El); 2189 } 2190 return R; 2191 } 2192 2193 if (result.isComplexInt()) { 2194 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2195 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2196 return IntRange::join(R, I); 2197 } 2198 2199 // This can happen with lossless casts to intptr_t of "based" lvalues. 2200 // Assume it might use arbitrary bits. 2201 // FIXME: The only reason we need to pass the type in here is to get 2202 // the sign right on this one case. It would be nice if APValue 2203 // preserved this. 2204 assert(result.isLValue()); 2205 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 2206} 2207 2208/// Pseudo-evaluate the given integer expression, estimating the 2209/// range of values it might take. 2210/// 2211/// \param MaxWidth - the width to which the value will be truncated 2212IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2213 E = E->IgnoreParens(); 2214 2215 // Try a full evaluation first. 2216 Expr::EvalResult result; 2217 if (E->Evaluate(result, C)) 2218 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2219 2220 // I think we only want to look through implicit casts here; if the 2221 // user has an explicit widening cast, we should treat the value as 2222 // being of the new, wider type. 2223 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2224 if (CE->getCastKind() == CastExpr::CK_NoOp) 2225 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2226 2227 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 2228 2229 bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast); 2230 if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown) 2231 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 2232 2233 // Assume that non-integer casts can span the full range of the type. 2234 if (!isIntegerCast) 2235 return OutputTypeRange; 2236 2237 IntRange SubRange 2238 = GetExprRange(C, CE->getSubExpr(), 2239 std::min(MaxWidth, OutputTypeRange.Width)); 2240 2241 // Bail out if the subexpr's range is as wide as the cast type. 2242 if (SubRange.Width >= OutputTypeRange.Width) 2243 return OutputTypeRange; 2244 2245 // Otherwise, we take the smaller width, and we're non-negative if 2246 // either the output type or the subexpr is. 2247 return IntRange(SubRange.Width, 2248 SubRange.NonNegative || OutputTypeRange.NonNegative); 2249 } 2250 2251 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2252 // If we can fold the condition, just take that operand. 2253 bool CondResult; 2254 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2255 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2256 : CO->getFalseExpr(), 2257 MaxWidth); 2258 2259 // Otherwise, conservatively merge. 2260 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2261 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2262 return IntRange::join(L, R); 2263 } 2264 2265 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2266 switch (BO->getOpcode()) { 2267 2268 // Boolean-valued operations are single-bit and positive. 2269 case BinaryOperator::LAnd: 2270 case BinaryOperator::LOr: 2271 case BinaryOperator::LT: 2272 case BinaryOperator::GT: 2273 case BinaryOperator::LE: 2274 case BinaryOperator::GE: 2275 case BinaryOperator::EQ: 2276 case BinaryOperator::NE: 2277 return IntRange::forBoolType(); 2278 2279 // The type of these compound assignments is the type of the LHS, 2280 // so the RHS is not necessarily an integer. 2281 case BinaryOperator::MulAssign: 2282 case BinaryOperator::DivAssign: 2283 case BinaryOperator::RemAssign: 2284 case BinaryOperator::AddAssign: 2285 case BinaryOperator::SubAssign: 2286 return IntRange::forType(C, E->getType()); 2287 2288 // Operations with opaque sources are black-listed. 2289 case BinaryOperator::PtrMemD: 2290 case BinaryOperator::PtrMemI: 2291 return IntRange::forType(C, E->getType()); 2292 2293 // Bitwise-and uses the *infinum* of the two source ranges. 2294 case BinaryOperator::And: 2295 case BinaryOperator::AndAssign: 2296 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2297 GetExprRange(C, BO->getRHS(), MaxWidth)); 2298 2299 // Left shift gets black-listed based on a judgement call. 2300 case BinaryOperator::Shl: 2301 // ...except that we want to treat '1 << (blah)' as logically 2302 // positive. It's an important idiom. 2303 if (IntegerLiteral *I 2304 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2305 if (I->getValue() == 1) { 2306 IntRange R = IntRange::forType(C, E->getType()); 2307 return IntRange(R.Width, /*NonNegative*/ true); 2308 } 2309 } 2310 // fallthrough 2311 2312 case BinaryOperator::ShlAssign: 2313 return IntRange::forType(C, E->getType()); 2314 2315 // Right shift by a constant can narrow its left argument. 2316 case BinaryOperator::Shr: 2317 case BinaryOperator::ShrAssign: { 2318 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2319 2320 // If the shift amount is a positive constant, drop the width by 2321 // that much. 2322 llvm::APSInt shift; 2323 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2324 shift.isNonNegative()) { 2325 unsigned zext = shift.getZExtValue(); 2326 if (zext >= L.Width) 2327 L.Width = (L.NonNegative ? 0 : 1); 2328 else 2329 L.Width -= zext; 2330 } 2331 2332 return L; 2333 } 2334 2335 // Comma acts as its right operand. 2336 case BinaryOperator::Comma: 2337 return GetExprRange(C, BO->getRHS(), MaxWidth); 2338 2339 // Black-list pointer subtractions. 2340 case BinaryOperator::Sub: 2341 if (BO->getLHS()->getType()->isPointerType()) 2342 return IntRange::forType(C, E->getType()); 2343 // fallthrough 2344 2345 default: 2346 break; 2347 } 2348 2349 // Treat every other operator as if it were closed on the 2350 // narrowest type that encompasses both operands. 2351 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2352 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2353 return IntRange::join(L, R); 2354 } 2355 2356 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2357 switch (UO->getOpcode()) { 2358 // Boolean-valued operations are white-listed. 2359 case UnaryOperator::LNot: 2360 return IntRange::forBoolType(); 2361 2362 // Operations with opaque sources are black-listed. 2363 case UnaryOperator::Deref: 2364 case UnaryOperator::AddrOf: // should be impossible 2365 case UnaryOperator::OffsetOf: 2366 return IntRange::forType(C, E->getType()); 2367 2368 default: 2369 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2370 } 2371 } 2372 2373 if (dyn_cast<OffsetOfExpr>(E)) { 2374 IntRange::forType(C, E->getType()); 2375 } 2376 2377 FieldDecl *BitField = E->getBitField(); 2378 if (BitField) { 2379 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2380 unsigned BitWidth = BitWidthAP.getZExtValue(); 2381 2382 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 2383 } 2384 2385 return IntRange::forType(C, E->getType()); 2386} 2387 2388IntRange GetExprRange(ASTContext &C, Expr *E) { 2389 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2390} 2391 2392/// Checks whether the given value, which currently has the given 2393/// source semantics, has the same value when coerced through the 2394/// target semantics. 2395bool IsSameFloatAfterCast(const llvm::APFloat &value, 2396 const llvm::fltSemantics &Src, 2397 const llvm::fltSemantics &Tgt) { 2398 llvm::APFloat truncated = value; 2399 2400 bool ignored; 2401 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2402 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2403 2404 return truncated.bitwiseIsEqual(value); 2405} 2406 2407/// Checks whether the given value, which currently has the given 2408/// source semantics, has the same value when coerced through the 2409/// target semantics. 2410/// 2411/// The value might be a vector of floats (or a complex number). 2412bool IsSameFloatAfterCast(const APValue &value, 2413 const llvm::fltSemantics &Src, 2414 const llvm::fltSemantics &Tgt) { 2415 if (value.isFloat()) 2416 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2417 2418 if (value.isVector()) { 2419 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2420 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2421 return false; 2422 return true; 2423 } 2424 2425 assert(value.isComplexFloat()); 2426 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2427 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2428} 2429 2430void AnalyzeImplicitConversions(Sema &S, Expr *E); 2431 2432bool IsZero(Sema &S, Expr *E) { 2433 llvm::APSInt Value; 2434 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2435} 2436 2437void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2438 BinaryOperator::Opcode op = E->getOpcode(); 2439 if (op == BinaryOperator::LT && IsZero(S, E->getRHS())) { 2440 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2441 << "< 0" << "false" 2442 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2443 } else if (op == BinaryOperator::GE && IsZero(S, E->getRHS())) { 2444 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2445 << ">= 0" << "true" 2446 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2447 } else if (op == BinaryOperator::GT && IsZero(S, E->getLHS())) { 2448 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2449 << "0 >" << "false" 2450 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2451 } else if (op == BinaryOperator::LE && IsZero(S, E->getLHS())) { 2452 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2453 << "0 <=" << "true" 2454 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2455 } 2456} 2457 2458/// Analyze the operands of the given comparison. Implements the 2459/// fallback case from AnalyzeComparison. 2460void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2461 AnalyzeImplicitConversions(S, E->getLHS()); 2462 AnalyzeImplicitConversions(S, E->getRHS()); 2463} 2464 2465/// \brief Implements -Wsign-compare. 2466/// 2467/// \param lex the left-hand expression 2468/// \param rex the right-hand expression 2469/// \param OpLoc the location of the joining operator 2470/// \param BinOpc binary opcode or 0 2471void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2472 // The type the comparison is being performed in. 2473 QualType T = E->getLHS()->getType(); 2474 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2475 && "comparison with mismatched types"); 2476 2477 // We don't do anything special if this isn't an unsigned integral 2478 // comparison: we're only interested in integral comparisons, and 2479 // signed comparisons only happen in cases we don't care to warn about. 2480 if (!T->isUnsignedIntegerType()) 2481 return AnalyzeImpConvsInComparison(S, E); 2482 2483 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2484 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2485 2486 // Check to see if one of the (unmodified) operands is of different 2487 // signedness. 2488 Expr *signedOperand, *unsignedOperand; 2489 if (lex->getType()->isSignedIntegerType()) { 2490 assert(!rex->getType()->isSignedIntegerType() && 2491 "unsigned comparison between two signed integer expressions?"); 2492 signedOperand = lex; 2493 unsignedOperand = rex; 2494 } else if (rex->getType()->isSignedIntegerType()) { 2495 signedOperand = rex; 2496 unsignedOperand = lex; 2497 } else { 2498 CheckTrivialUnsignedComparison(S, E); 2499 return AnalyzeImpConvsInComparison(S, E); 2500 } 2501 2502 // Otherwise, calculate the effective range of the signed operand. 2503 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2504 2505 // Go ahead and analyze implicit conversions in the operands. Note 2506 // that we skip the implicit conversions on both sides. 2507 AnalyzeImplicitConversions(S, lex); 2508 AnalyzeImplicitConversions(S, rex); 2509 2510 // If the signed range is non-negative, -Wsign-compare won't fire, 2511 // but we should still check for comparisons which are always true 2512 // or false. 2513 if (signedRange.NonNegative) 2514 return CheckTrivialUnsignedComparison(S, E); 2515 2516 // For (in)equality comparisons, if the unsigned operand is a 2517 // constant which cannot collide with a overflowed signed operand, 2518 // then reinterpreting the signed operand as unsigned will not 2519 // change the result of the comparison. 2520 if (E->isEqualityOp()) { 2521 unsigned comparisonWidth = S.Context.getIntWidth(T); 2522 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2523 2524 // We should never be unable to prove that the unsigned operand is 2525 // non-negative. 2526 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2527 2528 if (unsignedRange.Width < comparisonWidth) 2529 return; 2530 } 2531 2532 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2533 << lex->getType() << rex->getType() 2534 << lex->getSourceRange() << rex->getSourceRange(); 2535} 2536 2537/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2538void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2539 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2540} 2541 2542void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 2543 bool *ICContext = 0) { 2544 if (E->isTypeDependent() || E->isValueDependent()) return; 2545 2546 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 2547 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 2548 if (Source == Target) return; 2549 if (Target->isDependentType()) return; 2550 2551 // Never diagnose implicit casts to bool. 2552 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2553 return; 2554 2555 // Strip vector types. 2556 if (isa<VectorType>(Source)) { 2557 if (!isa<VectorType>(Target)) 2558 return DiagnoseImpCast(S, E, T, diag::warn_impcast_vector_scalar); 2559 2560 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2561 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2562 } 2563 2564 // Strip complex types. 2565 if (isa<ComplexType>(Source)) { 2566 if (!isa<ComplexType>(Target)) 2567 return DiagnoseImpCast(S, E, T, diag::warn_impcast_complex_scalar); 2568 2569 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2570 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2571 } 2572 2573 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2574 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2575 2576 // If the source is floating point... 2577 if (SourceBT && SourceBT->isFloatingPoint()) { 2578 // ...and the target is floating point... 2579 if (TargetBT && TargetBT->isFloatingPoint()) { 2580 // ...then warn if we're dropping FP rank. 2581 2582 // Builtin FP kinds are ordered by increasing FP rank. 2583 if (SourceBT->getKind() > TargetBT->getKind()) { 2584 // Don't warn about float constants that are precisely 2585 // representable in the target type. 2586 Expr::EvalResult result; 2587 if (E->Evaluate(result, S.Context)) { 2588 // Value might be a float, a float vector, or a float complex. 2589 if (IsSameFloatAfterCast(result.Val, 2590 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2591 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2592 return; 2593 } 2594 2595 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_precision); 2596 } 2597 return; 2598 } 2599 2600 // If the target is integral, always warn. 2601 if ((TargetBT && TargetBT->isInteger())) 2602 // TODO: don't warn for integer values? 2603 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_integer); 2604 2605 return; 2606 } 2607 2608 if (!Source->isIntegerType() || !Target->isIntegerType()) 2609 return; 2610 2611 IntRange SourceRange = GetExprRange(S.Context, E); 2612 IntRange TargetRange = IntRange::forCanonicalType(S.Context, Target); 2613 2614 if (SourceRange.Width > TargetRange.Width) { 2615 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2616 // and by god we'll let them. 2617 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2618 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_64_32); 2619 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_precision); 2620 } 2621 2622 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 2623 (!TargetRange.NonNegative && SourceRange.NonNegative && 2624 SourceRange.Width == TargetRange.Width)) { 2625 unsigned DiagID = diag::warn_impcast_integer_sign; 2626 2627 // Traditionally, gcc has warned about this under -Wsign-compare. 2628 // We also want to warn about it in -Wconversion. 2629 // So if -Wconversion is off, use a completely identical diagnostic 2630 // in the sign-compare group. 2631 // The conditional-checking code will 2632 if (ICContext) { 2633 DiagID = diag::warn_impcast_integer_sign_conditional; 2634 *ICContext = true; 2635 } 2636 2637 return DiagnoseImpCast(S, E, T, DiagID); 2638 } 2639 2640 return; 2641} 2642 2643void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 2644 2645void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 2646 bool &ICContext) { 2647 E = E->IgnoreParenImpCasts(); 2648 2649 if (isa<ConditionalOperator>(E)) 2650 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 2651 2652 AnalyzeImplicitConversions(S, E); 2653 if (E->getType() != T) 2654 return CheckImplicitConversion(S, E, T, &ICContext); 2655 return; 2656} 2657 2658void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 2659 AnalyzeImplicitConversions(S, E->getCond()); 2660 2661 bool Suspicious = false; 2662 CheckConditionalOperand(S, E->getTrueExpr(), T, Suspicious); 2663 CheckConditionalOperand(S, E->getFalseExpr(), T, Suspicious); 2664 2665 // If -Wconversion would have warned about either of the candidates 2666 // for a signedness conversion to the context type... 2667 if (!Suspicious) return; 2668 2669 // ...but it's currently ignored... 2670 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional)) 2671 return; 2672 2673 // ...and -Wsign-compare isn't... 2674 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional)) 2675 return; 2676 2677 // ...then check whether it would have warned about either of the 2678 // candidates for a signedness conversion to the condition type. 2679 if (E->getType() != T) { 2680 Suspicious = false; 2681 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 2682 E->getType(), &Suspicious); 2683 if (!Suspicious) 2684 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 2685 E->getType(), &Suspicious); 2686 if (!Suspicious) 2687 return; 2688 } 2689 2690 // If so, emit a diagnostic under -Wsign-compare. 2691 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 2692 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 2693 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 2694 << lex->getType() << rex->getType() 2695 << lex->getSourceRange() << rex->getSourceRange(); 2696} 2697 2698/// AnalyzeImplicitConversions - Find and report any interesting 2699/// implicit conversions in the given expression. There are a couple 2700/// of competing diagnostics here, -Wconversion and -Wsign-compare. 2701void AnalyzeImplicitConversions(Sema &S, Expr *OrigE) { 2702 QualType T = OrigE->getType(); 2703 Expr *E = OrigE->IgnoreParenImpCasts(); 2704 2705 // For conditional operators, we analyze the arguments as if they 2706 // were being fed directly into the output. 2707 if (isa<ConditionalOperator>(E)) { 2708 ConditionalOperator *CO = cast<ConditionalOperator>(E); 2709 CheckConditionalOperator(S, CO, T); 2710 return; 2711 } 2712 2713 // Go ahead and check any implicit conversions we might have skipped. 2714 // The non-canonical typecheck is just an optimization; 2715 // CheckImplicitConversion will filter out dead implicit conversions. 2716 if (E->getType() != T) 2717 CheckImplicitConversion(S, E, T); 2718 2719 // Now continue drilling into this expression. 2720 2721 // Skip past explicit casts. 2722 if (isa<ExplicitCastExpr>(E)) { 2723 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 2724 return AnalyzeImplicitConversions(S, E); 2725 } 2726 2727 // Do a somewhat different check with comparison operators. 2728 if (isa<BinaryOperator>(E) && cast<BinaryOperator>(E)->isComparisonOp()) 2729 return AnalyzeComparison(S, cast<BinaryOperator>(E)); 2730 2731 // These break the otherwise-useful invariant below. Fortunately, 2732 // we don't really need to recurse into them, because any internal 2733 // expressions should have been analyzed already when they were 2734 // built into statements. 2735 if (isa<StmtExpr>(E)) return; 2736 2737 // Don't descend into unevaluated contexts. 2738 if (isa<SizeOfAlignOfExpr>(E)) return; 2739 2740 // Now just recurse over the expression's children. 2741 for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end(); 2742 I != IE; ++I) 2743 AnalyzeImplicitConversions(S, cast<Expr>(*I)); 2744} 2745 2746} // end anonymous namespace 2747 2748/// Diagnoses "dangerous" implicit conversions within the given 2749/// expression (which is a full expression). Implements -Wconversion 2750/// and -Wsign-compare. 2751void Sema::CheckImplicitConversions(Expr *E) { 2752 // Don't diagnose in unevaluated contexts. 2753 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2754 return; 2755 2756 // Don't diagnose for value- or type-dependent expressions. 2757 if (E->isTypeDependent() || E->isValueDependent()) 2758 return; 2759 2760 AnalyzeImplicitConversions(*this, E); 2761} 2762 2763/// CheckParmsForFunctionDef - Check that the parameters of the given 2764/// function are appropriate for the definition of a function. This 2765/// takes care of any checks that cannot be performed on the 2766/// declaration itself, e.g., that the types of each of the function 2767/// parameters are complete. 2768bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2769 bool HasInvalidParm = false; 2770 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2771 ParmVarDecl *Param = FD->getParamDecl(p); 2772 2773 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2774 // function declarator that is part of a function definition of 2775 // that function shall not have incomplete type. 2776 // 2777 // This is also C++ [dcl.fct]p6. 2778 if (!Param->isInvalidDecl() && 2779 RequireCompleteType(Param->getLocation(), Param->getType(), 2780 diag::err_typecheck_decl_incomplete_type)) { 2781 Param->setInvalidDecl(); 2782 HasInvalidParm = true; 2783 } 2784 2785 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2786 // declaration of each parameter shall include an identifier. 2787 if (Param->getIdentifier() == 0 && 2788 !Param->isImplicit() && 2789 !getLangOptions().CPlusPlus) 2790 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2791 2792 // C99 6.7.5.3p12: 2793 // If the function declarator is not part of a definition of that 2794 // function, parameters may have incomplete type and may use the [*] 2795 // notation in their sequences of declarator specifiers to specify 2796 // variable length array types. 2797 QualType PType = Param->getOriginalType(); 2798 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2799 if (AT->getSizeModifier() == ArrayType::Star) { 2800 // FIXME: This diagnosic should point the the '[*]' if source-location 2801 // information is added for it. 2802 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2803 } 2804 } 2805 } 2806 2807 return HasInvalidParm; 2808} 2809