1 /* ----------------------------------------------------------------------- 2 ffi.c - Copyright (c) 1996, 1998, 1999, 2001 Red Hat, Inc. 3 Copyright (c) 2002 Ranjit Mathew 4 Copyright (c) 2002 Bo Thorsen 5 Copyright (c) 2002 Roger Sayle 6 7 x86 Foreign Function Interface 8 9 Permission is hereby granted, free of charge, to any person obtaining 10 a copy of this software and associated documentation files (the 11 ``Software''), to deal in the Software without restriction, including 12 without limitation the rights to use, copy, modify, merge, publish, 13 distribute, sublicense, and/or sell copies of the Software, and to 14 permit persons to whom the Software is furnished to do so, subject to 15 the following conditions: 16 17 The above copyright notice and this permission notice shall be included 18 in all copies or substantial portions of the Software. 19 20 THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, EXPRESS 21 OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 22 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 23 IN NO EVENT SHALL CYGNUS SOLUTIONS BE LIABLE FOR ANY CLAIM, DAMAGES OR 24 OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 25 ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 26 OTHER DEALINGS IN THE SOFTWARE. 27 ----------------------------------------------------------------------- */ 28 29 #include <ffi.h> 30 #include <ffi_common.h> 31 32 #include <stdlib.h> 33 34 /* ffi_prep_args is called by the assembly routine once stack space 35 has been allocated for the function's arguments */ 36 37 extern void Py_FatalError(const char *msg); 38 39 /*@-exportheader@*/ 40 void ffi_prep_args(char *stack, extended_cif *ecif) 41 /*@=exportheader@*/ 42 { 43 register unsigned int i; 44 register void **p_argv; 45 register char *argp; 46 register ffi_type **p_arg; 47 48 argp = stack; 49 if (ecif->cif->rtype->type == FFI_TYPE_STRUCT) 50 { 51 *(void **) argp = ecif->rvalue; 52 argp += sizeof(void *); 53 } 54 55 p_argv = ecif->avalue; 56 57 for (i = ecif->cif->nargs, p_arg = ecif->cif->arg_types; 58 i != 0; 59 i--, p_arg++) 60 { 61 size_t z; 62 63 /* Align if necessary */ 64 if ((sizeof(void *) - 1) & (size_t) argp) 65 argp = (char *) ALIGN(argp, sizeof(void *)); 66 67 z = (*p_arg)->size; 68 if (z < sizeof(intptr_t)) 69 { 70 z = sizeof(intptr_t); 71 switch ((*p_arg)->type) 72 { 73 case FFI_TYPE_SINT8: 74 *(intptr_t *) argp = (intptr_t)*(SINT8 *)(* p_argv); 75 break; 76 77 case FFI_TYPE_UINT8: 78 *(uintptr_t *) argp = (uintptr_t)*(UINT8 *)(* p_argv); 79 break; 80 81 case FFI_TYPE_SINT16: 82 *(intptr_t *) argp = (intptr_t)*(SINT16 *)(* p_argv); 83 break; 84 85 case FFI_TYPE_UINT16: 86 *(uintptr_t *) argp = (uintptr_t)*(UINT16 *)(* p_argv); 87 break; 88 89 case FFI_TYPE_SINT32: 90 *(intptr_t *) argp = (intptr_t)*(SINT32 *)(* p_argv); 91 break; 92 93 case FFI_TYPE_UINT32: 94 *(uintptr_t *) argp = (uintptr_t)*(UINT32 *)(* p_argv); 95 break; 96 97 case FFI_TYPE_FLOAT: 98 *(uintptr_t *) argp = 0; 99 *(float *) argp = *(float *)(* p_argv); 100 break; 101 102 // 64-bit value cases should never be used for x86 and AMD64 builds 103 case FFI_TYPE_SINT64: 104 *(intptr_t *) argp = (intptr_t)*(SINT64 *)(* p_argv); 105 break; 106 107 case FFI_TYPE_UINT64: 108 *(uintptr_t *) argp = (uintptr_t)*(UINT64 *)(* p_argv); 109 break; 110 111 case FFI_TYPE_STRUCT: 112 *(uintptr_t *) argp = (uintptr_t)*(UINT32 *)(* p_argv); 113 break; 114 115 case FFI_TYPE_DOUBLE: 116 *(uintptr_t *) argp = 0; 117 *(double *) argp = *(double *)(* p_argv); 118 break; 119 120 default: 121 FFI_ASSERT(0); 122 } 123 } 124 #ifdef _WIN64 125 else if (z > 8) 126 { 127 /* On Win64, if a single argument takes more than 8 bytes, 128 then it is always passed by reference. */ 129 *(void **)argp = *p_argv; 130 z = 8; 131 } 132 #endif 133 else 134 { 135 memcpy(argp, *p_argv, z); 136 } 137 p_argv++; 138 argp += z; 139 } 140 141 if (argp >= stack && (unsigned)(argp - stack) > ecif->cif->bytes) 142 { 143 Py_FatalError("FFI BUG: not enough stack space for arguments"); 144 } 145 return; 146 } 147 148 /* Perform machine dependent cif processing */ 149 ffi_status ffi_prep_cif_machdep(ffi_cif *cif) 150 { 151 /* Set the return type flag */ 152 switch (cif->rtype->type) 153 { 154 case FFI_TYPE_VOID: 155 case FFI_TYPE_SINT64: 156 case FFI_TYPE_FLOAT: 157 case FFI_TYPE_DOUBLE: 158 case FFI_TYPE_LONGDOUBLE: 159 cif->flags = (unsigned) cif->rtype->type; 160 break; 161 162 case FFI_TYPE_STRUCT: 163 /* MSVC returns small structures in registers. Put in cif->flags 164 the value FFI_TYPE_STRUCT only if the structure is big enough; 165 otherwise, put the 4- or 8-bytes integer type. */ 166 if (cif->rtype->size <= 4) 167 cif->flags = FFI_TYPE_INT; 168 else if (cif->rtype->size <= 8) 169 cif->flags = FFI_TYPE_SINT64; 170 else 171 cif->flags = FFI_TYPE_STRUCT; 172 break; 173 174 case FFI_TYPE_UINT64: 175 #ifdef _WIN64 176 case FFI_TYPE_POINTER: 177 #endif 178 cif->flags = FFI_TYPE_SINT64; 179 break; 180 181 default: 182 cif->flags = FFI_TYPE_INT; 183 break; 184 } 185 186 return FFI_OK; 187 } 188 189 #ifdef _WIN32 190 extern int 191 ffi_call_x86(void (*)(char *, extended_cif *), 192 /*@out@*/ extended_cif *, 193 unsigned, unsigned, 194 /*@out@*/ unsigned *, 195 void (*fn)()); 196 #endif 197 198 #ifdef _WIN64 199 extern int 200 ffi_call_AMD64(void (*)(char *, extended_cif *), 201 /*@out@*/ extended_cif *, 202 unsigned, unsigned, 203 /*@out@*/ unsigned *, 204 void (*fn)()); 205 #endif 206 207 int 208 ffi_call(/*@dependent@*/ ffi_cif *cif, 209 void (*fn)(), 210 /*@out@*/ void *rvalue, 211 /*@dependent@*/ void **avalue) 212 { 213 extended_cif ecif; 214 215 ecif.cif = cif; 216 ecif.avalue = avalue; 217 218 /* If the return value is a struct and we don't have a return */ 219 /* value address then we need to make one */ 220 221 if ((rvalue == NULL) && 222 (cif->rtype->type == FFI_TYPE_STRUCT)) 223 { 224 /*@-sysunrecog@*/ 225 ecif.rvalue = alloca(cif->rtype->size); 226 /*@=sysunrecog@*/ 227 } 228 else 229 ecif.rvalue = rvalue; 230 231 232 switch (cif->abi) 233 { 234 #if !defined(_WIN64) 235 case FFI_SYSV: 236 case FFI_STDCALL: 237 return ffi_call_x86(ffi_prep_args, &ecif, cif->bytes, 238 cif->flags, ecif.rvalue, fn); 239 break; 240 #else 241 case FFI_SYSV: 242 /* If a single argument takes more than 8 bytes, 243 then a copy is passed by reference. */ 244 for (unsigned i = 0; i < cif->nargs; i++) { 245 size_t z = cif->arg_types[i]->size; 246 if (z > 8) { 247 void *temp = alloca(z); 248 memcpy(temp, avalue[i], z); 249 avalue[i] = temp; 250 } 251 } 252 /*@-usedef@*/ 253 return ffi_call_AMD64(ffi_prep_args, &ecif, cif->bytes, 254 cif->flags, ecif.rvalue, fn); 255 /*@=usedef@*/ 256 break; 257 #endif 258 259 default: 260 FFI_ASSERT(0); 261 break; 262 } 263 return -1; /* theller: Hrm. */ 264 } 265 266 267 /** private members **/ 268 269 static void ffi_prep_incoming_args_SYSV (char *stack, void **ret, 270 void** args, ffi_cif* cif); 271 /* This function is jumped to by the trampoline */ 272 273 #ifdef _WIN64 274 void * 275 #else 276 static void __fastcall 277 #endif 278 ffi_closure_SYSV (ffi_closure *closure, char *argp) 279 { 280 // this is our return value storage 281 long double res; 282 283 // our various things... 284 ffi_cif *cif; 285 void **arg_area; 286 unsigned short rtype; 287 void *resp = (void*)&res; 288 void *args = argp + sizeof(void*); 289 290 cif = closure->cif; 291 arg_area = (void**) alloca (cif->nargs * sizeof (void*)); 292 293 /* this call will initialize ARG_AREA, such that each 294 * element in that array points to the corresponding 295 * value on the stack; and if the function returns 296 * a structure, it will re-set RESP to point to the 297 * structure return address. */ 298 299 ffi_prep_incoming_args_SYSV(args, (void**)&resp, arg_area, cif); 300 301 (closure->fun) (cif, resp, arg_area, closure->user_data); 302 303 rtype = cif->flags; 304 305 #if defined(_WIN32) && !defined(_WIN64) 306 #ifdef _MSC_VER 307 /* now, do a generic return based on the value of rtype */ 308 if (rtype == FFI_TYPE_INT) 309 { 310 _asm mov eax, resp ; 311 _asm mov eax, [eax] ; 312 } 313 else if (rtype == FFI_TYPE_FLOAT) 314 { 315 _asm mov eax, resp ; 316 _asm fld DWORD PTR [eax] ; 317 // asm ("flds (%0)" : : "r" (resp) : "st" ); 318 } 319 else if (rtype == FFI_TYPE_DOUBLE) 320 { 321 _asm mov eax, resp ; 322 _asm fld QWORD PTR [eax] ; 323 // asm ("fldl (%0)" : : "r" (resp) : "st", "st(1)" ); 324 } 325 else if (rtype == FFI_TYPE_LONGDOUBLE) 326 { 327 // asm ("fldt (%0)" : : "r" (resp) : "st", "st(1)" ); 328 } 329 else if (rtype == FFI_TYPE_SINT64) 330 { 331 _asm mov edx, resp ; 332 _asm mov eax, [edx] ; 333 _asm mov edx, [edx + 4] ; 334 // asm ("movl 0(%0),%%eax;" 335 // "movl 4(%0),%%edx" 336 // : : "r"(resp) 337 // : "eax", "edx"); 338 } 339 #else 340 /* now, do a generic return based on the value of rtype */ 341 if (rtype == FFI_TYPE_INT) 342 { 343 asm ("movl (%0),%%eax" : : "r" (resp) : "eax"); 344 } 345 else if (rtype == FFI_TYPE_FLOAT) 346 { 347 asm ("flds (%0)" : : "r" (resp) : "st" ); 348 } 349 else if (rtype == FFI_TYPE_DOUBLE) 350 { 351 asm ("fldl (%0)" : : "r" (resp) : "st", "st(1)" ); 352 } 353 else if (rtype == FFI_TYPE_LONGDOUBLE) 354 { 355 asm ("fldt (%0)" : : "r" (resp) : "st", "st(1)" ); 356 } 357 else if (rtype == FFI_TYPE_SINT64) 358 { 359 asm ("movl 0(%0),%%eax;" 360 "movl 4(%0),%%edx" 361 : : "r"(resp) 362 : "eax", "edx"); 363 } 364 #endif 365 #endif 366 367 #ifdef _WIN64 368 /* The result is returned in rax. This does the right thing for 369 result types except for floats; we have to 'mov xmm0, rax' in the 370 caller to correct this. 371 */ 372 return *(void **)resp; 373 #endif 374 } 375 376 /*@-exportheader@*/ 377 static void 378 ffi_prep_incoming_args_SYSV(char *stack, void **rvalue, 379 void **avalue, ffi_cif *cif) 380 /*@=exportheader@*/ 381 { 382 register unsigned int i; 383 register void **p_argv; 384 register char *argp; 385 register ffi_type **p_arg; 386 387 argp = stack; 388 389 if ( cif->rtype->type == FFI_TYPE_STRUCT ) { 390 *rvalue = *(void **) argp; 391 argp += sizeof(void *); 392 } 393 394 p_argv = avalue; 395 396 for (i = cif->nargs, p_arg = cif->arg_types; (i != 0); i--, p_arg++) 397 { 398 size_t z; 399 400 /* Align if necessary */ 401 if ((sizeof(char *) - 1) & (size_t) argp) { 402 argp = (char *) ALIGN(argp, sizeof(char*)); 403 } 404 405 z = (*p_arg)->size; 406 407 /* because we're little endian, this is what it turns into. */ 408 409 #ifdef _WIN64 410 if (z > 8) { 411 /* On Win64, if a single argument takes more than 8 bytes, 412 * then it is always passed by reference. 413 */ 414 *p_argv = *((void**) argp); 415 z = 8; 416 } 417 else 418 #endif 419 *p_argv = (void*) argp; 420 421 p_argv++; 422 argp += z; 423 } 424 425 return; 426 } 427 428 /* the cif must already be prep'ed */ 429 extern void ffi_closure_OUTER(); 430 431 ffi_status 432 ffi_prep_closure_loc (ffi_closure* closure, 433 ffi_cif* cif, 434 void (*fun)(ffi_cif*,void*,void**,void*), 435 void *user_data, 436 void *codeloc) 437 { 438 short bytes; 439 char *tramp; 440 #ifdef _WIN64 441 int mask = 0; 442 #endif 443 FFI_ASSERT (cif->abi == FFI_SYSV); 444 445 if (cif->abi == FFI_SYSV) 446 bytes = 0; 447 #if !defined(_WIN64) 448 else if (cif->abi == FFI_STDCALL) 449 bytes = cif->bytes; 450 #endif 451 else 452 return FFI_BAD_ABI; 453 454 tramp = &closure->tramp[0]; 455 456 #define BYTES(text) memcpy(tramp, text, sizeof(text)), tramp += sizeof(text)-1 457 #define POINTER(x) *(void**)tramp = (void*)(x), tramp += sizeof(void*) 458 #define SHORT(x) *(short*)tramp = x, tramp += sizeof(short) 459 #define INT(x) *(int*)tramp = x, tramp += sizeof(int) 460 461 #ifdef _WIN64 462 if (cif->nargs >= 1 && 463 (cif->arg_types[0]->type == FFI_TYPE_FLOAT 464 || cif->arg_types[0]->type == FFI_TYPE_DOUBLE)) 465 mask |= 1; 466 if (cif->nargs >= 2 && 467 (cif->arg_types[1]->type == FFI_TYPE_FLOAT 468 || cif->arg_types[1]->type == FFI_TYPE_DOUBLE)) 469 mask |= 2; 470 if (cif->nargs >= 3 && 471 (cif->arg_types[2]->type == FFI_TYPE_FLOAT 472 || cif->arg_types[2]->type == FFI_TYPE_DOUBLE)) 473 mask |= 4; 474 if (cif->nargs >= 4 && 475 (cif->arg_types[3]->type == FFI_TYPE_FLOAT 476 || cif->arg_types[3]->type == FFI_TYPE_DOUBLE)) 477 mask |= 8; 478 479 /* 41 BB ---- mov r11d,mask */ 480 BYTES("\x41\xBB"); INT(mask); 481 482 /* 48 B8 -------- mov rax, closure */ 483 BYTES("\x48\xB8"); POINTER(closure); 484 485 /* 49 BA -------- mov r10, ffi_closure_OUTER */ 486 BYTES("\x49\xBA"); POINTER(ffi_closure_OUTER); 487 488 /* 41 FF E2 jmp r10 */ 489 BYTES("\x41\xFF\xE2"); 490 491 #else 492 493 /* mov ecx, closure */ 494 BYTES("\xb9"); POINTER(closure); 495 496 /* mov edx, esp */ 497 BYTES("\x8b\xd4"); 498 499 /* call ffi_closure_SYSV */ 500 BYTES("\xe8"); POINTER((char*)&ffi_closure_SYSV - (tramp + 4)); 501 502 /* ret bytes */ 503 BYTES("\xc2"); 504 SHORT(bytes); 505 506 #endif 507 508 if (tramp - &closure->tramp[0] > FFI_TRAMPOLINE_SIZE) 509 Py_FatalError("FFI_TRAMPOLINE_SIZE too small in " __FILE__); 510 511 closure->cif = cif; 512 closure->user_data = user_data; 513 closure->fun = fun; 514 return FFI_OK; 515 } 516