| 1 | /* |
| 2 | ** FFI C call handling. |
| 3 | ** Copyright (C) 2005-2021 Mike Pall. See Copyright Notice in luajit.h |
| 4 | */ |
| 5 | |
| 6 | #include "lj_obj.h" |
| 7 | |
| 8 | #if LJ_HASFFI |
| 9 | |
| 10 | #include "lj_gc.h" |
| 11 | #include "lj_err.h" |
| 12 | #include "lj_tab.h" |
| 13 | #include "lj_ctype.h" |
| 14 | #include "lj_cconv.h" |
| 15 | #include "lj_cdata.h" |
| 16 | #include "lj_ccall.h" |
| 17 | #include "lj_trace.h" |
| 18 | |
| 19 | /* Target-specific handling of register arguments. */ |
| 20 | #if LJ_TARGET_X86 |
| 21 | /* -- x86 calling conventions --------------------------------------------- */ |
| 22 | |
| 23 | #if LJ_ABI_WIN |
| 24 | |
| 25 | #define CCALL_HANDLE_STRUCTRET \ |
| 26 | /* Return structs bigger than 8 by reference (on stack only). */ \ |
| 27 | cc->retref = (sz > 8); \ |
| 28 | if (cc->retref) cc->stack[nsp++] = (GPRArg)dp; |
| 29 | |
| 30 | #define CCALL_HANDLE_COMPLEXRET CCALL_HANDLE_STRUCTRET |
| 31 | |
| 32 | #else |
| 33 | |
| 34 | #if LJ_TARGET_OSX |
| 35 | |
| 36 | #define CCALL_HANDLE_STRUCTRET \ |
| 37 | /* Return structs of size 1, 2, 4 or 8 in registers. */ \ |
| 38 | cc->retref = !(sz == 1 || sz == 2 || sz == 4 || sz == 8); \ |
| 39 | if (cc->retref) { \ |
| 40 | if (ngpr < maxgpr) \ |
| 41 | cc->gpr[ngpr++] = (GPRArg)dp; \ |
| 42 | else \ |
| 43 | cc->stack[nsp++] = (GPRArg)dp; \ |
| 44 | } else { /* Struct with single FP field ends up in FPR. */ \ |
| 45 | cc->resx87 = ccall_classify_struct(cts, ctr); \ |
| 46 | } |
| 47 | |
| 48 | #define CCALL_HANDLE_STRUCTRET2 \ |
| 49 | if (cc->resx87) sp = (uint8_t *)&cc->fpr[0]; \ |
| 50 | memcpy(dp, sp, ctr->size); |
| 51 | |
| 52 | #else |
| 53 | |
| 54 | #define CCALL_HANDLE_STRUCTRET \ |
| 55 | cc->retref = 1; /* Return all structs by reference (in reg or on stack). */ \ |
| 56 | if (ngpr < maxgpr) \ |
| 57 | cc->gpr[ngpr++] = (GPRArg)dp; \ |
| 58 | else \ |
| 59 | cc->stack[nsp++] = (GPRArg)dp; |
| 60 | |
| 61 | #endif |
| 62 | |
| 63 | #define CCALL_HANDLE_COMPLEXRET \ |
| 64 | /* Return complex float in GPRs and complex double by reference. */ \ |
| 65 | cc->retref = (sz > 8); \ |
| 66 | if (cc->retref) { \ |
| 67 | if (ngpr < maxgpr) \ |
| 68 | cc->gpr[ngpr++] = (GPRArg)dp; \ |
| 69 | else \ |
| 70 | cc->stack[nsp++] = (GPRArg)dp; \ |
| 71 | } |
| 72 | |
| 73 | #endif |
| 74 | |
| 75 | #define CCALL_HANDLE_COMPLEXRET2 \ |
| 76 | if (!cc->retref) \ |
| 77 | *(int64_t *)dp = *(int64_t *)sp; /* Copy complex float from GPRs. */ |
| 78 | |
| 79 | #define CCALL_HANDLE_STRUCTARG \ |
| 80 | ngpr = maxgpr; /* Pass all structs by value on the stack. */ |
| 81 | |
| 82 | #define CCALL_HANDLE_COMPLEXARG \ |
| 83 | isfp = 1; /* Pass complex by value on stack. */ |
| 84 | |
| 85 | #define CCALL_HANDLE_REGARG \ |
| 86 | if (!isfp) { /* Only non-FP values may be passed in registers. */ \ |
| 87 | if (n > 1) { /* Anything > 32 bit is passed on the stack. */ \ |
| 88 | if (!LJ_ABI_WIN) ngpr = maxgpr; /* Prevent reordering. */ \ |
| 89 | } else if (ngpr + 1 <= maxgpr) { \ |
| 90 | dp = &cc->gpr[ngpr]; \ |
| 91 | ngpr += n; \ |
| 92 | goto done; \ |
| 93 | } \ |
| 94 | } |
| 95 | |
| 96 | #elif LJ_TARGET_X64 && LJ_ABI_WIN |
| 97 | /* -- Windows/x64 calling conventions ------------------------------------- */ |
| 98 | |
| 99 | #define CCALL_HANDLE_STRUCTRET \ |
| 100 | /* Return structs of size 1, 2, 4 or 8 in a GPR. */ \ |
| 101 | cc->retref = !(sz == 1 || sz == 2 || sz == 4 || sz == 8); \ |
| 102 | if (cc->retref) cc->gpr[ngpr++] = (GPRArg)dp; |
| 103 | |
| 104 | #define CCALL_HANDLE_COMPLEXRET CCALL_HANDLE_STRUCTRET |
| 105 | |
| 106 | #define CCALL_HANDLE_COMPLEXRET2 \ |
| 107 | if (!cc->retref) \ |
| 108 | *(int64_t *)dp = *(int64_t *)sp; /* Copy complex float from GPRs. */ |
| 109 | |
| 110 | #define CCALL_HANDLE_STRUCTARG \ |
| 111 | /* Pass structs of size 1, 2, 4 or 8 in a GPR by value. */ \ |
| 112 | if (!(sz == 1 || sz == 2 || sz == 4 || sz == 8)) { \ |
| 113 | rp = cdataptr(lj_cdata_new(cts, did, sz)); \ |
| 114 | sz = CTSIZE_PTR; /* Pass all other structs by reference. */ \ |
| 115 | } |
| 116 | |
| 117 | #define CCALL_HANDLE_COMPLEXARG \ |
| 118 | /* Pass complex float in a GPR and complex double by reference. */ \ |
| 119 | if (sz != 2*sizeof(float)) { \ |
| 120 | rp = cdataptr(lj_cdata_new(cts, did, sz)); \ |
| 121 | sz = CTSIZE_PTR; \ |
| 122 | } |
| 123 | |
| 124 | /* Windows/x64 argument registers are strictly positional (use ngpr). */ |
| 125 | #define CCALL_HANDLE_REGARG \ |
| 126 | if (isfp) { \ |
| 127 | if (ngpr < maxgpr) { dp = &cc->fpr[ngpr++]; nfpr = ngpr; goto done; } \ |
| 128 | } else { \ |
| 129 | if (ngpr < maxgpr) { dp = &cc->gpr[ngpr++]; goto done; } \ |
| 130 | } |
| 131 | |
| 132 | #elif LJ_TARGET_X64 |
| 133 | /* -- POSIX/x64 calling conventions --------------------------------------- */ |
| 134 | |
| 135 | #define CCALL_HANDLE_STRUCTRET \ |
| 136 | int rcl[2]; rcl[0] = rcl[1] = 0; \ |
| 137 | if (ccall_classify_struct(cts, ctr, rcl, 0)) { \ |
| 138 | cc->retref = 1; /* Return struct by reference. */ \ |
| 139 | cc->gpr[ngpr++] = (GPRArg)dp; \ |
| 140 | } else { \ |
| 141 | cc->retref = 0; /* Return small structs in registers. */ \ |
| 142 | } |
| 143 | |
| 144 | #define CCALL_HANDLE_STRUCTRET2 \ |
| 145 | int rcl[2]; rcl[0] = rcl[1] = 0; \ |
| 146 | ccall_classify_struct(cts, ctr, rcl, 0); \ |
| 147 | ccall_struct_ret(cc, rcl, dp, ctr->size); |
| 148 | |
| 149 | #define CCALL_HANDLE_COMPLEXRET \ |
| 150 | /* Complex values are returned in one or two FPRs. */ \ |
| 151 | cc->retref = 0; |
| 152 | |
| 153 | #define CCALL_HANDLE_COMPLEXRET2 \ |
| 154 | if (ctr->size == 2*sizeof(float)) { /* Copy complex float from FPR. */ \ |
| 155 | *(int64_t *)dp = cc->fpr[0].l[0]; \ |
| 156 | } else { /* Copy non-contiguous complex double from FPRs. */ \ |
| 157 | ((int64_t *)dp)[0] = cc->fpr[0].l[0]; \ |
| 158 | ((int64_t *)dp)[1] = cc->fpr[1].l[0]; \ |
| 159 | } |
| 160 | |
| 161 | #define CCALL_HANDLE_STRUCTARG \ |
| 162 | int rcl[2]; rcl[0] = rcl[1] = 0; \ |
| 163 | if (!ccall_classify_struct(cts, d, rcl, 0)) { \ |
| 164 | cc->nsp = nsp; cc->ngpr = ngpr; cc->nfpr = nfpr; \ |
| 165 | if (ccall_struct_arg(cc, cts, d, rcl, o, narg)) goto err_nyi; \ |
| 166 | nsp = cc->nsp; ngpr = cc->ngpr; nfpr = cc->nfpr; \ |
| 167 | continue; \ |
| 168 | } /* Pass all other structs by value on stack. */ |
| 169 | |
| 170 | #define CCALL_HANDLE_COMPLEXARG \ |
| 171 | isfp = 2; /* Pass complex in FPRs or on stack. Needs postprocessing. */ |
| 172 | |
| 173 | #define CCALL_HANDLE_REGARG \ |
| 174 | if (isfp) { /* Try to pass argument in FPRs. */ \ |
| 175 | int n2 = ctype_isvector(d->info) ? 1 : n; \ |
| 176 | if (nfpr + n2 <= CCALL_NARG_FPR) { \ |
| 177 | dp = &cc->fpr[nfpr]; \ |
| 178 | nfpr += n2; \ |
| 179 | goto done; \ |
| 180 | } \ |
| 181 | } else { /* Try to pass argument in GPRs. */ \ |
| 182 | /* Note that reordering is explicitly allowed in the x64 ABI. */ \ |
| 183 | if (n <= 2 && ngpr + n <= maxgpr) { \ |
| 184 | dp = &cc->gpr[ngpr]; \ |
| 185 | ngpr += n; \ |
| 186 | goto done; \ |
| 187 | } \ |
| 188 | } |
| 189 | |
| 190 | #elif LJ_TARGET_ARM |
| 191 | /* -- ARM calling conventions --------------------------------------------- */ |
| 192 | |
| 193 | #if LJ_ABI_SOFTFP |
| 194 | |
| 195 | #define CCALL_HANDLE_STRUCTRET \ |
| 196 | /* Return structs of size <= 4 in a GPR. */ \ |
| 197 | cc->retref = !(sz <= 4); \ |
| 198 | if (cc->retref) cc->gpr[ngpr++] = (GPRArg)dp; |
| 199 | |
| 200 | #define CCALL_HANDLE_COMPLEXRET \ |
| 201 | cc->retref = 1; /* Return all complex values by reference. */ \ |
| 202 | cc->gpr[ngpr++] = (GPRArg)dp; |
| 203 | |
| 204 | #define CCALL_HANDLE_COMPLEXRET2 \ |
| 205 | UNUSED(dp); /* Nothing to do. */ |
| 206 | |
| 207 | #define CCALL_HANDLE_STRUCTARG \ |
| 208 | /* Pass all structs by value in registers and/or on the stack. */ |
| 209 | |
| 210 | #define CCALL_HANDLE_COMPLEXARG \ |
| 211 | /* Pass complex by value in 2 or 4 GPRs. */ |
| 212 | |
| 213 | #define CCALL_HANDLE_REGARG_FP1 |
| 214 | #define CCALL_HANDLE_REGARG_FP2 |
| 215 | |
| 216 | #else |
| 217 | |
| 218 | #define CCALL_HANDLE_STRUCTRET \ |
| 219 | cc->retref = !ccall_classify_struct(cts, ctr, ct); \ |
| 220 | if (cc->retref) cc->gpr[ngpr++] = (GPRArg)dp; |
| 221 | |
| 222 | #define CCALL_HANDLE_STRUCTRET2 \ |
| 223 | if (ccall_classify_struct(cts, ctr, ct) > 1) sp = (uint8_t *)&cc->fpr[0]; \ |
| 224 | memcpy(dp, sp, ctr->size); |
| 225 | |
| 226 | #define CCALL_HANDLE_COMPLEXRET \ |
| 227 | if (!(ct->info & CTF_VARARG)) cc->retref = 0; /* Return complex in FPRs. */ |
| 228 | |
| 229 | #define CCALL_HANDLE_COMPLEXRET2 \ |
| 230 | if (!(ct->info & CTF_VARARG)) memcpy(dp, &cc->fpr[0], ctr->size); |
| 231 | |
| 232 | #define CCALL_HANDLE_STRUCTARG \ |
| 233 | isfp = (ccall_classify_struct(cts, d, ct) > 1); |
| 234 | /* Pass all structs by value in registers and/or on the stack. */ |
| 235 | |
| 236 | #define CCALL_HANDLE_COMPLEXARG \ |
| 237 | isfp = 1; /* Pass complex by value in FPRs or on stack. */ |
| 238 | |
| 239 | #define CCALL_HANDLE_REGARG_FP1 \ |
| 240 | if (isfp && !(ct->info & CTF_VARARG)) { \ |
| 241 | if ((d->info & CTF_ALIGN) > CTALIGN_PTR) { \ |
| 242 | if (nfpr + (n >> 1) <= CCALL_NARG_FPR) { \ |
| 243 | dp = &cc->fpr[nfpr]; \ |
| 244 | nfpr += (n >> 1); \ |
| 245 | goto done; \ |
| 246 | } \ |
| 247 | } else { \ |
| 248 | if (sz > 1 && fprodd != nfpr) fprodd = 0; \ |
| 249 | if (fprodd) { \ |
| 250 | if (2*nfpr+n <= 2*CCALL_NARG_FPR+1) { \ |
| 251 | dp = (void *)&cc->fpr[fprodd-1].f[1]; \ |
| 252 | nfpr += (n >> 1); \ |
| 253 | if ((n & 1)) fprodd = 0; else fprodd = nfpr-1; \ |
| 254 | goto done; \ |
| 255 | } \ |
| 256 | } else { \ |
| 257 | if (2*nfpr+n <= 2*CCALL_NARG_FPR) { \ |
| 258 | dp = (void *)&cc->fpr[nfpr]; \ |
| 259 | nfpr += (n >> 1); \ |
| 260 | if ((n & 1)) fprodd = ++nfpr; else fprodd = 0; \ |
| 261 | goto done; \ |
| 262 | } \ |
| 263 | } \ |
| 264 | } \ |
| 265 | fprodd = 0; /* No reordering after the first FP value is on stack. */ \ |
| 266 | } else { |
| 267 | |
| 268 | #define CCALL_HANDLE_REGARG_FP2 } |
| 269 | |
| 270 | #endif |
| 271 | |
| 272 | #define CCALL_HANDLE_REGARG \ |
| 273 | CCALL_HANDLE_REGARG_FP1 \ |
| 274 | if ((d->info & CTF_ALIGN) > CTALIGN_PTR) { \ |
| 275 | if (ngpr < maxgpr) \ |
| 276 | ngpr = (ngpr + 1u) & ~1u; /* Align to regpair. */ \ |
| 277 | } \ |
| 278 | if (ngpr < maxgpr) { \ |
| 279 | dp = &cc->gpr[ngpr]; \ |
| 280 | if (ngpr + n > maxgpr) { \ |
| 281 | nsp += ngpr + n - maxgpr; /* Assumes contiguous gpr/stack fields. */ \ |
| 282 | if (nsp > CCALL_MAXSTACK) goto err_nyi; /* Too many arguments. */ \ |
| 283 | ngpr = maxgpr; \ |
| 284 | } else { \ |
| 285 | ngpr += n; \ |
| 286 | } \ |
| 287 | goto done; \ |
| 288 | } CCALL_HANDLE_REGARG_FP2 |
| 289 | |
| 290 | #define CCALL_HANDLE_RET \ |
| 291 | if ((ct->info & CTF_VARARG)) sp = (uint8_t *)&cc->gpr[0]; |
| 292 | |
| 293 | #elif LJ_TARGET_ARM64 |
| 294 | /* -- ARM64 calling conventions ------------------------------------------- */ |
| 295 | |
| 296 | #define CCALL_HANDLE_STRUCTRET \ |
| 297 | cc->retref = !ccall_classify_struct(cts, ctr); \ |
| 298 | if (cc->retref) cc->retp = dp; |
| 299 | |
| 300 | #define CCALL_HANDLE_STRUCTRET2 \ |
| 301 | unsigned int cl = ccall_classify_struct(cts, ctr); \ |
| 302 | if ((cl & 4)) { /* Combine float HFA from separate registers. */ \ |
| 303 | CTSize i = (cl >> 8) - 1; \ |
| 304 | do { ((uint32_t *)dp)[i] = cc->fpr[i].lo; } while (i--); \ |
| 305 | } else { \ |
| 306 | if (cl > 1) sp = (uint8_t *)&cc->fpr[0]; \ |
| 307 | memcpy(dp, sp, ctr->size); \ |
| 308 | } |
| 309 | |
| 310 | #define CCALL_HANDLE_COMPLEXRET \ |
| 311 | /* Complex values are returned in one or two FPRs. */ \ |
| 312 | cc->retref = 0; |
| 313 | |
| 314 | #define CCALL_HANDLE_COMPLEXRET2 \ |
| 315 | if (ctr->size == 2*sizeof(float)) { /* Copy complex float from FPRs. */ \ |
| 316 | ((float *)dp)[0] = cc->fpr[0].f; \ |
| 317 | ((float *)dp)[1] = cc->fpr[1].f; \ |
| 318 | } else { /* Copy complex double from FPRs. */ \ |
| 319 | ((double *)dp)[0] = cc->fpr[0].d; \ |
| 320 | ((double *)dp)[1] = cc->fpr[1].d; \ |
| 321 | } |
| 322 | |
| 323 | #define CCALL_HANDLE_STRUCTARG \ |
| 324 | unsigned int cl = ccall_classify_struct(cts, d); \ |
| 325 | if (cl == 0) { /* Pass struct by reference. */ \ |
| 326 | rp = cdataptr(lj_cdata_new(cts, did, sz)); \ |
| 327 | sz = CTSIZE_PTR; \ |
| 328 | } else if (cl > 1) { /* Pass struct in FPRs or on stack. */ \ |
| 329 | isfp = (cl & 4) ? 2 : 1; \ |
| 330 | } /* else: Pass struct in GPRs or on stack. */ |
| 331 | |
| 332 | #define CCALL_HANDLE_COMPLEXARG \ |
| 333 | /* Pass complex by value in separate (!) FPRs or on stack. */ \ |
| 334 | isfp = sz == 2*sizeof(float) ? 2 : 1; |
| 335 | |
| 336 | #define CCALL_HANDLE_REGARG \ |
| 337 | if (LJ_TARGET_IOS && isva) { \ |
| 338 | /* IOS: All variadic arguments are on the stack. */ \ |
| 339 | } else if (isfp) { /* Try to pass argument in FPRs. */ \ |
| 340 | int n2 = ctype_isvector(d->info) ? 1 : \ |
| 341 | isfp == 1 ? n : (d->size >> (4-isfp)); \ |
| 342 | if (nfpr + n2 <= CCALL_NARG_FPR) { \ |
| 343 | dp = &cc->fpr[nfpr]; \ |
| 344 | nfpr += n2; \ |
| 345 | goto done; \ |
| 346 | } else { \ |
| 347 | nfpr = CCALL_NARG_FPR; /* Prevent reordering. */ \ |
| 348 | if (LJ_TARGET_IOS && d->size < 8) goto err_nyi; \ |
| 349 | } \ |
| 350 | } else { /* Try to pass argument in GPRs. */ \ |
| 351 | if (!LJ_TARGET_IOS && (d->info & CTF_ALIGN) > CTALIGN_PTR) \ |
| 352 | ngpr = (ngpr + 1u) & ~1u; /* Align to regpair. */ \ |
| 353 | if (ngpr + n <= maxgpr) { \ |
| 354 | dp = &cc->gpr[ngpr]; \ |
| 355 | ngpr += n; \ |
| 356 | goto done; \ |
| 357 | } else { \ |
| 358 | ngpr = maxgpr; /* Prevent reordering. */ \ |
| 359 | if (LJ_TARGET_IOS && d->size < 8) goto err_nyi; \ |
| 360 | } \ |
| 361 | } |
| 362 | |
| 363 | #if LJ_BE |
| 364 | #define CCALL_HANDLE_RET \ |
| 365 | if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \ |
| 366 | sp = (uint8_t *)&cc->fpr[0].f; |
| 367 | #endif |
| 368 | |
| 369 | |
| 370 | #elif LJ_TARGET_PPC |
| 371 | /* -- PPC calling conventions --------------------------------------------- */ |
| 372 | |
| 373 | #define CCALL_HANDLE_STRUCTRET \ |
| 374 | cc->retref = 1; /* Return all structs by reference. */ \ |
| 375 | cc->gpr[ngpr++] = (GPRArg)dp; |
| 376 | |
| 377 | #define CCALL_HANDLE_COMPLEXRET \ |
| 378 | /* Complex values are returned in 2 or 4 GPRs. */ \ |
| 379 | cc->retref = 0; |
| 380 | |
| 381 | #define CCALL_HANDLE_COMPLEXRET2 \ |
| 382 | memcpy(dp, sp, ctr->size); /* Copy complex from GPRs. */ |
| 383 | |
| 384 | #define CCALL_HANDLE_STRUCTARG \ |
| 385 | rp = cdataptr(lj_cdata_new(cts, did, sz)); \ |
| 386 | sz = CTSIZE_PTR; /* Pass all structs by reference. */ |
| 387 | |
| 388 | #define CCALL_HANDLE_COMPLEXARG \ |
| 389 | /* Pass complex by value in 2 or 4 GPRs. */ |
| 390 | |
| 391 | #define CCALL_HANDLE_GPR \ |
| 392 | /* Try to pass argument in GPRs. */ \ |
| 393 | if (n > 1) { \ |
| 394 | /* int64_t or complex (float). */ \ |
| 395 | lj_assertL(n == 2 || n == 4, "bad GPR size %d", n); \ |
| 396 | if (ctype_isinteger(d->info) || ctype_isfp(d->info)) \ |
| 397 | ngpr = (ngpr + 1u) & ~1u; /* Align int64_t to regpair. */ \ |
| 398 | else if (ngpr + n > maxgpr) \ |
| 399 | ngpr = maxgpr; /* Prevent reordering. */ \ |
| 400 | } \ |
| 401 | if (ngpr + n <= maxgpr) { \ |
| 402 | dp = &cc->gpr[ngpr]; \ |
| 403 | ngpr += n; \ |
| 404 | goto done; \ |
| 405 | } \ |
| 406 | |
| 407 | #if LJ_ABI_SOFTFP |
| 408 | #define CCALL_HANDLE_REGARG CCALL_HANDLE_GPR |
| 409 | #else |
| 410 | #define CCALL_HANDLE_REGARG \ |
| 411 | if (isfp) { /* Try to pass argument in FPRs. */ \ |
| 412 | if (nfpr + 1 <= CCALL_NARG_FPR) { \ |
| 413 | dp = &cc->fpr[nfpr]; \ |
| 414 | nfpr += 1; \ |
| 415 | d = ctype_get(cts, CTID_DOUBLE); /* FPRs always hold doubles. */ \ |
| 416 | goto done; \ |
| 417 | } \ |
| 418 | } else { \ |
| 419 | CCALL_HANDLE_GPR \ |
| 420 | } |
| 421 | #endif |
| 422 | |
| 423 | #if !LJ_ABI_SOFTFP |
| 424 | #define CCALL_HANDLE_RET \ |
| 425 | if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \ |
| 426 | ctr = ctype_get(cts, CTID_DOUBLE); /* FPRs always hold doubles. */ |
| 427 | #endif |
| 428 | |
| 429 | #elif LJ_TARGET_MIPS32 |
| 430 | /* -- MIPS o32 calling conventions ---------------------------------------- */ |
| 431 | |
| 432 | #define CCALL_HANDLE_STRUCTRET \ |
| 433 | cc->retref = 1; /* Return all structs by reference. */ \ |
| 434 | cc->gpr[ngpr++] = (GPRArg)dp; |
| 435 | |
| 436 | #define CCALL_HANDLE_COMPLEXRET \ |
| 437 | /* Complex values are returned in 1 or 2 FPRs. */ \ |
| 438 | cc->retref = 0; |
| 439 | |
| 440 | #if LJ_ABI_SOFTFP |
| 441 | #define CCALL_HANDLE_COMPLEXRET2 \ |
| 442 | if (ctr->size == 2*sizeof(float)) { /* Copy complex float from GPRs. */ \ |
| 443 | ((intptr_t *)dp)[0] = cc->gpr[0]; \ |
| 444 | ((intptr_t *)dp)[1] = cc->gpr[1]; \ |
| 445 | } else { /* Copy complex double from GPRs. */ \ |
| 446 | ((intptr_t *)dp)[0] = cc->gpr[0]; \ |
| 447 | ((intptr_t *)dp)[1] = cc->gpr[1]; \ |
| 448 | ((intptr_t *)dp)[2] = cc->gpr[2]; \ |
| 449 | ((intptr_t *)dp)[3] = cc->gpr[3]; \ |
| 450 | } |
| 451 | #else |
| 452 | #define CCALL_HANDLE_COMPLEXRET2 \ |
| 453 | if (ctr->size == 2*sizeof(float)) { /* Copy complex float from FPRs. */ \ |
| 454 | ((float *)dp)[0] = cc->fpr[0].f; \ |
| 455 | ((float *)dp)[1] = cc->fpr[1].f; \ |
| 456 | } else { /* Copy complex double from FPRs. */ \ |
| 457 | ((double *)dp)[0] = cc->fpr[0].d; \ |
| 458 | ((double *)dp)[1] = cc->fpr[1].d; \ |
| 459 | } |
| 460 | #endif |
| 461 | |
| 462 | #define CCALL_HANDLE_STRUCTARG \ |
| 463 | /* Pass all structs by value in registers and/or on the stack. */ |
| 464 | |
| 465 | #define CCALL_HANDLE_COMPLEXARG \ |
| 466 | /* Pass complex by value in 2 or 4 GPRs. */ |
| 467 | |
| 468 | #define CCALL_HANDLE_GPR \ |
| 469 | if ((d->info & CTF_ALIGN) > CTALIGN_PTR) \ |
| 470 | ngpr = (ngpr + 1u) & ~1u; /* Align to regpair. */ \ |
| 471 | if (ngpr < maxgpr) { \ |
| 472 | dp = &cc->gpr[ngpr]; \ |
| 473 | if (ngpr + n > maxgpr) { \ |
| 474 | nsp += ngpr + n - maxgpr; /* Assumes contiguous gpr/stack fields. */ \ |
| 475 | if (nsp > CCALL_MAXSTACK) goto err_nyi; /* Too many arguments. */ \ |
| 476 | ngpr = maxgpr; \ |
| 477 | } else { \ |
| 478 | ngpr += n; \ |
| 479 | } \ |
| 480 | goto done; \ |
| 481 | } |
| 482 | |
| 483 | #if !LJ_ABI_SOFTFP /* MIPS32 hard-float */ |
| 484 | #define CCALL_HANDLE_REGARG \ |
| 485 | if (isfp && nfpr < CCALL_NARG_FPR && !(ct->info & CTF_VARARG)) { \ |
| 486 | /* Try to pass argument in FPRs. */ \ |
| 487 | dp = n == 1 ? (void *)&cc->fpr[nfpr].f : (void *)&cc->fpr[nfpr].d; \ |
| 488 | nfpr++; ngpr += n; \ |
| 489 | goto done; \ |
| 490 | } else { /* Try to pass argument in GPRs. */ \ |
| 491 | nfpr = CCALL_NARG_FPR; \ |
| 492 | CCALL_HANDLE_GPR \ |
| 493 | } |
| 494 | #else /* MIPS32 soft-float */ |
| 495 | #define CCALL_HANDLE_REGARG CCALL_HANDLE_GPR |
| 496 | #endif |
| 497 | |
| 498 | #if !LJ_ABI_SOFTFP |
| 499 | /* On MIPS64 soft-float, position of float return values is endian-dependant. */ |
| 500 | #define CCALL_HANDLE_RET \ |
| 501 | if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \ |
| 502 | sp = (uint8_t *)&cc->fpr[0].f; |
| 503 | #endif |
| 504 | |
| 505 | #elif LJ_TARGET_MIPS64 |
| 506 | /* -- MIPS n64 calling conventions ---------------------------------------- */ |
| 507 | |
| 508 | #define CCALL_HANDLE_STRUCTRET \ |
| 509 | cc->retref = !(sz <= 16); \ |
| 510 | if (cc->retref) cc->gpr[ngpr++] = (GPRArg)dp; |
| 511 | |
| 512 | #define CCALL_HANDLE_STRUCTRET2 \ |
| 513 | ccall_copy_struct(cc, ctr, dp, sp, ccall_classify_struct(cts, ctr, ct)); |
| 514 | |
| 515 | #define CCALL_HANDLE_COMPLEXRET \ |
| 516 | /* Complex values are returned in 1 or 2 FPRs. */ \ |
| 517 | cc->retref = 0; |
| 518 | |
| 519 | #if LJ_ABI_SOFTFP /* MIPS64 soft-float */ |
| 520 | |
| 521 | #define CCALL_HANDLE_COMPLEXRET2 \ |
| 522 | if (ctr->size == 2*sizeof(float)) { /* Copy complex float from GPRs. */ \ |
| 523 | ((intptr_t *)dp)[0] = cc->gpr[0]; \ |
| 524 | } else { /* Copy complex double from GPRs. */ \ |
| 525 | ((intptr_t *)dp)[0] = cc->gpr[0]; \ |
| 526 | ((intptr_t *)dp)[1] = cc->gpr[1]; \ |
| 527 | } |
| 528 | |
| 529 | #define CCALL_HANDLE_COMPLEXARG \ |
| 530 | /* Pass complex by value in 2 or 4 GPRs. */ |
| 531 | |
| 532 | /* Position of soft-float 'float' return value depends on endianess. */ |
| 533 | #define CCALL_HANDLE_RET \ |
| 534 | if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \ |
| 535 | sp = (uint8_t *)cc->gpr + LJ_ENDIAN_SELECT(0, 4); |
| 536 | |
| 537 | #else /* MIPS64 hard-float */ |
| 538 | |
| 539 | #define CCALL_HANDLE_COMPLEXRET2 \ |
| 540 | if (ctr->size == 2*sizeof(float)) { /* Copy complex float from FPRs. */ \ |
| 541 | ((float *)dp)[0] = cc->fpr[0].f; \ |
| 542 | ((float *)dp)[1] = cc->fpr[1].f; \ |
| 543 | } else { /* Copy complex double from FPRs. */ \ |
| 544 | ((double *)dp)[0] = cc->fpr[0].d; \ |
| 545 | ((double *)dp)[1] = cc->fpr[1].d; \ |
| 546 | } |
| 547 | |
| 548 | #define CCALL_HANDLE_COMPLEXARG \ |
| 549 | if (sz == 2*sizeof(float)) { \ |
| 550 | isfp = 2; \ |
| 551 | if (ngpr < maxgpr) \ |
| 552 | sz *= 2; \ |
| 553 | } |
| 554 | |
| 555 | #define CCALL_HANDLE_RET \ |
| 556 | if (ctype_isfp(ctr->info) && ctr->size == sizeof(float)) \ |
| 557 | sp = (uint8_t *)&cc->fpr[0].f; |
| 558 | |
| 559 | #endif |
| 560 | |
| 561 | #define CCALL_HANDLE_STRUCTARG \ |
| 562 | /* Pass all structs by value in registers and/or on the stack. */ |
| 563 | |
| 564 | #define CCALL_HANDLE_REGARG \ |
| 565 | if (ngpr < maxgpr) { \ |
| 566 | dp = &cc->gpr[ngpr]; \ |
| 567 | if (ngpr + n > maxgpr) { \ |
| 568 | nsp += ngpr + n - maxgpr; /* Assumes contiguous gpr/stack fields. */ \ |
| 569 | if (nsp > CCALL_MAXSTACK) goto err_nyi; /* Too many arguments. */ \ |
| 570 | ngpr = maxgpr; \ |
| 571 | } else { \ |
| 572 | ngpr += n; \ |
| 573 | } \ |
| 574 | goto done; \ |
| 575 | } |
| 576 | |
| 577 | #else |
| 578 | #error "Missing calling convention definitions for this architecture" |
| 579 | #endif |
| 580 | |
| 581 | #ifndef CCALL_HANDLE_STRUCTRET2 |
| 582 | #define CCALL_HANDLE_STRUCTRET2 \ |
| 583 | memcpy(dp, sp, ctr->size); /* Copy struct return value from GPRs. */ |
| 584 | #endif |
| 585 | |
| 586 | /* -- x86 OSX ABI struct classification ----------------------------------- */ |
| 587 | |
| 588 | #if LJ_TARGET_X86 && LJ_TARGET_OSX |
| 589 | |
| 590 | /* Check for struct with single FP field. */ |
| 591 | static int ccall_classify_struct(CTState *cts, CType *ct) |
| 592 | { |
| 593 | CTSize sz = ct->size; |
| 594 | if (!(sz == sizeof(float) || sz == sizeof(double))) return 0; |
| 595 | if ((ct->info & CTF_UNION)) return 0; |
| 596 | while (ct->sib) { |
| 597 | ct = ctype_get(cts, ct->sib); |
| 598 | if (ctype_isfield(ct->info)) { |
| 599 | CType *sct = ctype_rawchild(cts, ct); |
| 600 | if (ctype_isfp(sct->info)) { |
| 601 | if (sct->size == sz) |
| 602 | return (sz >> 2); /* Return 1 for float or 2 for double. */ |
| 603 | } else if (ctype_isstruct(sct->info)) { |
| 604 | if (sct->size) |
| 605 | return ccall_classify_struct(cts, sct); |
| 606 | } else { |
| 607 | break; |
| 608 | } |
| 609 | } else if (ctype_isbitfield(ct->info)) { |
| 610 | break; |
| 611 | } else if (ctype_isxattrib(ct->info, CTA_SUBTYPE)) { |
| 612 | CType *sct = ctype_rawchild(cts, ct); |
| 613 | if (sct->size) |
| 614 | return ccall_classify_struct(cts, sct); |
| 615 | } |
| 616 | } |
| 617 | return 0; |
| 618 | } |
| 619 | |
| 620 | #endif |
| 621 | |
| 622 | /* -- x64 struct classification ------------------------------------------- */ |
| 623 | |
| 624 | #if LJ_TARGET_X64 && !LJ_ABI_WIN |
| 625 | |
| 626 | /* Register classes for x64 struct classification. */ |
| 627 | #define CCALL_RCL_INT 1 |
| 628 | #define CCALL_RCL_SSE 2 |
| 629 | #define CCALL_RCL_MEM 4 |
| 630 | /* NYI: classify vectors. */ |
| 631 | |
| 632 | static int ccall_classify_struct(CTState *cts, CType *ct, int *rcl, CTSize ofs); |
| 633 | |
| 634 | /* Classify a C type. */ |
| 635 | static void ccall_classify_ct(CTState *cts, CType *ct, int *rcl, CTSize ofs) |
| 636 | { |
| 637 | if (ctype_isarray(ct->info)) { |
| 638 | CType *cct = ctype_rawchild(cts, ct); |
| 639 | CTSize eofs, esz = cct->size, asz = ct->size; |
| 640 | for (eofs = 0; eofs < asz; eofs += esz) |
| 641 | ccall_classify_ct(cts, cct, rcl, ofs+eofs); |
| 642 | } else if (ctype_isstruct(ct->info)) { |
| 643 | ccall_classify_struct(cts, ct, rcl, ofs); |
| 644 | } else { |
| 645 | int cl = ctype_isfp(ct->info) ? CCALL_RCL_SSE : CCALL_RCL_INT; |
| 646 | lj_assertCTS(ctype_hassize(ct->info), |
| 647 | "classify ctype %08x without size" , ct->info); |
| 648 | if ((ofs & (ct->size-1))) cl = CCALL_RCL_MEM; /* Unaligned. */ |
| 649 | rcl[(ofs >= 8)] |= cl; |
| 650 | } |
| 651 | } |
| 652 | |
| 653 | /* Recursively classify a struct based on its fields. */ |
| 654 | static int ccall_classify_struct(CTState *cts, CType *ct, int *rcl, CTSize ofs) |
| 655 | { |
| 656 | if (ct->size > 16) return CCALL_RCL_MEM; /* Too big, gets memory class. */ |
| 657 | while (ct->sib) { |
| 658 | CTSize fofs; |
| 659 | ct = ctype_get(cts, ct->sib); |
| 660 | fofs = ofs+ct->size; |
| 661 | if (ctype_isfield(ct->info)) |
| 662 | ccall_classify_ct(cts, ctype_rawchild(cts, ct), rcl, fofs); |
| 663 | else if (ctype_isbitfield(ct->info)) |
| 664 | rcl[(fofs >= 8)] |= CCALL_RCL_INT; /* NYI: unaligned bitfields? */ |
| 665 | else if (ctype_isxattrib(ct->info, CTA_SUBTYPE)) |
| 666 | ccall_classify_struct(cts, ctype_rawchild(cts, ct), rcl, fofs); |
| 667 | } |
| 668 | return ((rcl[0]|rcl[1]) & CCALL_RCL_MEM); /* Memory class? */ |
| 669 | } |
| 670 | |
| 671 | /* Try to split up a small struct into registers. */ |
| 672 | static int ccall_struct_reg(CCallState *cc, CTState *cts, GPRArg *dp, int *rcl) |
| 673 | { |
| 674 | MSize ngpr = cc->ngpr, nfpr = cc->nfpr; |
| 675 | uint32_t i; |
| 676 | UNUSED(cts); |
| 677 | for (i = 0; i < 2; i++) { |
| 678 | lj_assertCTS(!(rcl[i] & CCALL_RCL_MEM), "pass mem struct in reg" ); |
| 679 | if ((rcl[i] & CCALL_RCL_INT)) { /* Integer class takes precedence. */ |
| 680 | if (ngpr >= CCALL_NARG_GPR) return 1; /* Register overflow. */ |
| 681 | cc->gpr[ngpr++] = dp[i]; |
| 682 | } else if ((rcl[i] & CCALL_RCL_SSE)) { |
| 683 | if (nfpr >= CCALL_NARG_FPR) return 1; /* Register overflow. */ |
| 684 | cc->fpr[nfpr++].l[0] = dp[i]; |
| 685 | } |
| 686 | } |
| 687 | cc->ngpr = ngpr; cc->nfpr = nfpr; |
| 688 | return 0; /* Ok. */ |
| 689 | } |
| 690 | |
| 691 | /* Pass a small struct argument. */ |
| 692 | static int ccall_struct_arg(CCallState *cc, CTState *cts, CType *d, int *rcl, |
| 693 | TValue *o, int narg) |
| 694 | { |
| 695 | GPRArg dp[2]; |
| 696 | dp[0] = dp[1] = 0; |
| 697 | /* Convert to temp. struct. */ |
| 698 | lj_cconv_ct_tv(cts, d, (uint8_t *)dp, o, CCF_ARG(narg)); |
| 699 | if (ccall_struct_reg(cc, cts, dp, rcl)) { |
| 700 | /* Register overflow? Pass on stack. */ |
| 701 | MSize nsp = cc->nsp, n = rcl[1] ? 2 : 1; |
| 702 | if (nsp + n > CCALL_MAXSTACK) return 1; /* Too many arguments. */ |
| 703 | cc->nsp = nsp + n; |
| 704 | memcpy(&cc->stack[nsp], dp, n*CTSIZE_PTR); |
| 705 | } |
| 706 | return 0; /* Ok. */ |
| 707 | } |
| 708 | |
| 709 | /* Combine returned small struct. */ |
| 710 | static void ccall_struct_ret(CCallState *cc, int *rcl, uint8_t *dp, CTSize sz) |
| 711 | { |
| 712 | GPRArg sp[2]; |
| 713 | MSize ngpr = 0, nfpr = 0; |
| 714 | uint32_t i; |
| 715 | for (i = 0; i < 2; i++) { |
| 716 | if ((rcl[i] & CCALL_RCL_INT)) { /* Integer class takes precedence. */ |
| 717 | sp[i] = cc->gpr[ngpr++]; |
| 718 | } else if ((rcl[i] & CCALL_RCL_SSE)) { |
| 719 | sp[i] = cc->fpr[nfpr++].l[0]; |
| 720 | } |
| 721 | } |
| 722 | memcpy(dp, sp, sz); |
| 723 | } |
| 724 | #endif |
| 725 | |
| 726 | /* -- ARM hard-float ABI struct classification ---------------------------- */ |
| 727 | |
| 728 | #if LJ_TARGET_ARM && !LJ_ABI_SOFTFP |
| 729 | |
| 730 | /* Classify a struct based on its fields. */ |
| 731 | static unsigned int ccall_classify_struct(CTState *cts, CType *ct, CType *ctf) |
| 732 | { |
| 733 | CTSize sz = ct->size; |
| 734 | unsigned int r = 0, n = 0, isu = (ct->info & CTF_UNION); |
| 735 | if ((ctf->info & CTF_VARARG)) goto noth; |
| 736 | while (ct->sib) { |
| 737 | CType *sct; |
| 738 | ct = ctype_get(cts, ct->sib); |
| 739 | if (ctype_isfield(ct->info)) { |
| 740 | sct = ctype_rawchild(cts, ct); |
| 741 | if (ctype_isfp(sct->info)) { |
| 742 | r |= sct->size; |
| 743 | if (!isu) n++; else if (n == 0) n = 1; |
| 744 | } else if (ctype_iscomplex(sct->info)) { |
| 745 | r |= (sct->size >> 1); |
| 746 | if (!isu) n += 2; else if (n < 2) n = 2; |
| 747 | } else if (ctype_isstruct(sct->info)) { |
| 748 | goto substruct; |
| 749 | } else { |
| 750 | goto noth; |
| 751 | } |
| 752 | } else if (ctype_isbitfield(ct->info)) { |
| 753 | goto noth; |
| 754 | } else if (ctype_isxattrib(ct->info, CTA_SUBTYPE)) { |
| 755 | sct = ctype_rawchild(cts, ct); |
| 756 | substruct: |
| 757 | if (sct->size > 0) { |
| 758 | unsigned int s = ccall_classify_struct(cts, sct, ctf); |
| 759 | if (s <= 1) goto noth; |
| 760 | r |= (s & 255); |
| 761 | if (!isu) n += (s >> 8); else if (n < (s >>8)) n = (s >> 8); |
| 762 | } |
| 763 | } |
| 764 | } |
| 765 | if ((r == 4 || r == 8) && n <= 4) |
| 766 | return r + (n << 8); |
| 767 | noth: /* Not a homogeneous float/double aggregate. */ |
| 768 | return (sz <= 4); /* Return structs of size <= 4 in a GPR. */ |
| 769 | } |
| 770 | |
| 771 | #endif |
| 772 | |
| 773 | /* -- ARM64 ABI struct classification ------------------------------------- */ |
| 774 | |
| 775 | #if LJ_TARGET_ARM64 |
| 776 | |
| 777 | /* Classify a struct based on its fields. */ |
| 778 | static unsigned int ccall_classify_struct(CTState *cts, CType *ct) |
| 779 | { |
| 780 | CTSize sz = ct->size; |
| 781 | unsigned int r = 0, n = 0, isu = (ct->info & CTF_UNION); |
| 782 | while (ct->sib) { |
| 783 | CType *sct; |
| 784 | ct = ctype_get(cts, ct->sib); |
| 785 | if (ctype_isfield(ct->info)) { |
| 786 | sct = ctype_rawchild(cts, ct); |
| 787 | if (ctype_isfp(sct->info)) { |
| 788 | r |= sct->size; |
| 789 | if (!isu) n++; else if (n == 0) n = 1; |
| 790 | } else if (ctype_iscomplex(sct->info)) { |
| 791 | r |= (sct->size >> 1); |
| 792 | if (!isu) n += 2; else if (n < 2) n = 2; |
| 793 | } else if (ctype_isstruct(sct->info)) { |
| 794 | goto substruct; |
| 795 | } else { |
| 796 | goto noth; |
| 797 | } |
| 798 | } else if (ctype_isbitfield(ct->info)) { |
| 799 | goto noth; |
| 800 | } else if (ctype_isxattrib(ct->info, CTA_SUBTYPE)) { |
| 801 | sct = ctype_rawchild(cts, ct); |
| 802 | substruct: |
| 803 | if (sct->size > 0) { |
| 804 | unsigned int s = ccall_classify_struct(cts, sct); |
| 805 | if (s <= 1) goto noth; |
| 806 | r |= (s & 255); |
| 807 | if (!isu) n += (s >> 8); else if (n < (s >>8)) n = (s >> 8); |
| 808 | } |
| 809 | } |
| 810 | } |
| 811 | if ((r == 4 || r == 8) && n <= 4) |
| 812 | return r + (n << 8); |
| 813 | noth: /* Not a homogeneous float/double aggregate. */ |
| 814 | return (sz <= 16); /* Return structs of size <= 16 in GPRs. */ |
| 815 | } |
| 816 | |
| 817 | #endif |
| 818 | |
| 819 | /* -- MIPS64 ABI struct classification ---------------------------- */ |
| 820 | |
| 821 | #if LJ_TARGET_MIPS64 |
| 822 | |
| 823 | #define FTYPE_FLOAT 1 |
| 824 | #define FTYPE_DOUBLE 2 |
| 825 | |
| 826 | /* Classify FP fields (max. 2) and their types. */ |
| 827 | static unsigned int ccall_classify_struct(CTState *cts, CType *ct, CType *ctf) |
| 828 | { |
| 829 | int n = 0, ft = 0; |
| 830 | if ((ctf->info & CTF_VARARG) || (ct->info & CTF_UNION)) |
| 831 | goto noth; |
| 832 | while (ct->sib) { |
| 833 | CType *sct; |
| 834 | ct = ctype_get(cts, ct->sib); |
| 835 | if (n == 2) { |
| 836 | goto noth; |
| 837 | } else if (ctype_isfield(ct->info)) { |
| 838 | sct = ctype_rawchild(cts, ct); |
| 839 | if (ctype_isfp(sct->info)) { |
| 840 | ft |= (sct->size == 4 ? FTYPE_FLOAT : FTYPE_DOUBLE) << 2*n; |
| 841 | n++; |
| 842 | } else { |
| 843 | goto noth; |
| 844 | } |
| 845 | } else if (ctype_isbitfield(ct->info) || |
| 846 | ctype_isxattrib(ct->info, CTA_SUBTYPE)) { |
| 847 | goto noth; |
| 848 | } |
| 849 | } |
| 850 | if (n <= 2) |
| 851 | return ft; |
| 852 | noth: /* Not a homogeneous float/double aggregate. */ |
| 853 | return 0; /* Struct is in GPRs. */ |
| 854 | } |
| 855 | |
| 856 | static void ccall_copy_struct(CCallState *cc, CType *ctr, void *dp, void *sp, |
| 857 | int ft) |
| 858 | { |
| 859 | if (LJ_ABI_SOFTFP ? ft : |
| 860 | ((ft & 3) == FTYPE_FLOAT || (ft >> 2) == FTYPE_FLOAT)) { |
| 861 | int i, ofs = 0; |
| 862 | for (i = 0; ft != 0; i++, ft >>= 2) { |
| 863 | if ((ft & 3) == FTYPE_FLOAT) { |
| 864 | #if LJ_ABI_SOFTFP |
| 865 | /* The 2nd FP struct result is in CARG1 (gpr[2]) and not CRET2. */ |
| 866 | memcpy((uint8_t *)dp + ofs, |
| 867 | (uint8_t *)&cc->gpr[2*i] + LJ_ENDIAN_SELECT(0, 4), 4); |
| 868 | #else |
| 869 | *(float *)((uint8_t *)dp + ofs) = cc->fpr[i].f; |
| 870 | #endif |
| 871 | ofs += 4; |
| 872 | } else { |
| 873 | ofs = (ofs + 7) & ~7; /* 64 bit alignment. */ |
| 874 | #if LJ_ABI_SOFTFP |
| 875 | *(intptr_t *)((uint8_t *)dp + ofs) = cc->gpr[2*i]; |
| 876 | #else |
| 877 | *(double *)((uint8_t *)dp + ofs) = cc->fpr[i].d; |
| 878 | #endif |
| 879 | ofs += 8; |
| 880 | } |
| 881 | } |
| 882 | } else { |
| 883 | #if !LJ_ABI_SOFTFP |
| 884 | if (ft) sp = (uint8_t *)&cc->fpr[0]; |
| 885 | #endif |
| 886 | memcpy(dp, sp, ctr->size); |
| 887 | } |
| 888 | } |
| 889 | |
| 890 | #endif |
| 891 | |
| 892 | /* -- Common C call handling ---------------------------------------------- */ |
| 893 | |
| 894 | /* Infer the destination CTypeID for a vararg argument. */ |
| 895 | CTypeID lj_ccall_ctid_vararg(CTState *cts, cTValue *o) |
| 896 | { |
| 897 | if (tvisnumber(o)) { |
| 898 | return CTID_DOUBLE; |
| 899 | } else if (tviscdata(o)) { |
| 900 | CTypeID id = cdataV(o)->ctypeid; |
| 901 | CType *s = ctype_get(cts, id); |
| 902 | if (ctype_isrefarray(s->info)) { |
| 903 | return lj_ctype_intern(cts, |
| 904 | CTINFO(CT_PTR, CTALIGN_PTR|ctype_cid(s->info)), CTSIZE_PTR); |
| 905 | } else if (ctype_isstruct(s->info) || ctype_isfunc(s->info)) { |
| 906 | /* NYI: how to pass a struct by value in a vararg argument? */ |
| 907 | return lj_ctype_intern(cts, CTINFO(CT_PTR, CTALIGN_PTR|id), CTSIZE_PTR); |
| 908 | } else if (ctype_isfp(s->info) && s->size == sizeof(float)) { |
| 909 | return CTID_DOUBLE; |
| 910 | } else { |
| 911 | return id; |
| 912 | } |
| 913 | } else if (tvisstr(o)) { |
| 914 | return CTID_P_CCHAR; |
| 915 | } else if (tvisbool(o)) { |
| 916 | return CTID_BOOL; |
| 917 | } else { |
| 918 | return CTID_P_VOID; |
| 919 | } |
| 920 | } |
| 921 | |
| 922 | /* Setup arguments for C call. */ |
| 923 | static int ccall_set_args(lua_State *L, CTState *cts, CType *ct, |
| 924 | CCallState *cc) |
| 925 | { |
| 926 | int gcsteps = 0; |
| 927 | TValue *o, *top = L->top; |
| 928 | CTypeID fid; |
| 929 | CType *ctr; |
| 930 | MSize maxgpr, ngpr = 0, nsp = 0, narg; |
| 931 | #if CCALL_NARG_FPR |
| 932 | MSize nfpr = 0; |
| 933 | #if LJ_TARGET_ARM |
| 934 | MSize fprodd = 0; |
| 935 | #endif |
| 936 | #endif |
| 937 | |
| 938 | /* Clear unused regs to get some determinism in case of misdeclaration. */ |
| 939 | memset(cc->gpr, 0, sizeof(cc->gpr)); |
| 940 | #if CCALL_NUM_FPR |
| 941 | memset(cc->fpr, 0, sizeof(cc->fpr)); |
| 942 | #endif |
| 943 | |
| 944 | #if LJ_TARGET_X86 |
| 945 | /* x86 has several different calling conventions. */ |
| 946 | cc->resx87 = 0; |
| 947 | switch (ctype_cconv(ct->info)) { |
| 948 | case CTCC_FASTCALL: maxgpr = 2; break; |
| 949 | case CTCC_THISCALL: maxgpr = 1; break; |
| 950 | default: maxgpr = 0; break; |
| 951 | } |
| 952 | #else |
| 953 | maxgpr = CCALL_NARG_GPR; |
| 954 | #endif |
| 955 | |
| 956 | /* Perform required setup for some result types. */ |
| 957 | ctr = ctype_rawchild(cts, ct); |
| 958 | if (ctype_isvector(ctr->info)) { |
| 959 | if (!(CCALL_VECTOR_REG && (ctr->size == 8 || ctr->size == 16))) |
| 960 | goto err_nyi; |
| 961 | } else if (ctype_iscomplex(ctr->info) || ctype_isstruct(ctr->info)) { |
| 962 | /* Preallocate cdata object and anchor it after arguments. */ |
| 963 | CTSize sz = ctr->size; |
| 964 | GCcdata *cd = lj_cdata_new(cts, ctype_cid(ct->info), sz); |
| 965 | void *dp = cdataptr(cd); |
| 966 | setcdataV(L, L->top++, cd); |
| 967 | if (ctype_isstruct(ctr->info)) { |
| 968 | CCALL_HANDLE_STRUCTRET |
| 969 | } else { |
| 970 | CCALL_HANDLE_COMPLEXRET |
| 971 | } |
| 972 | #if LJ_TARGET_X86 |
| 973 | } else if (ctype_isfp(ctr->info)) { |
| 974 | cc->resx87 = ctr->size == sizeof(float) ? 1 : 2; |
| 975 | #endif |
| 976 | } |
| 977 | |
| 978 | /* Skip initial attributes. */ |
| 979 | fid = ct->sib; |
| 980 | while (fid) { |
| 981 | CType *ctf = ctype_get(cts, fid); |
| 982 | if (!ctype_isattrib(ctf->info)) break; |
| 983 | fid = ctf->sib; |
| 984 | } |
| 985 | |
| 986 | /* Walk through all passed arguments. */ |
| 987 | for (o = L->base+1, narg = 1; o < top; o++, narg++) { |
| 988 | CTypeID did; |
| 989 | CType *d; |
| 990 | CTSize sz; |
| 991 | MSize n, isfp = 0, isva = 0; |
| 992 | void *dp, *rp = NULL; |
| 993 | |
| 994 | if (fid) { /* Get argument type from field. */ |
| 995 | CType *ctf = ctype_get(cts, fid); |
| 996 | fid = ctf->sib; |
| 997 | lj_assertL(ctype_isfield(ctf->info), "field expected" ); |
| 998 | did = ctype_cid(ctf->info); |
| 999 | } else { |
| 1000 | if (!(ct->info & CTF_VARARG)) |
| 1001 | lj_err_caller(L, LJ_ERR_FFI_NUMARG); /* Too many arguments. */ |
| 1002 | did = lj_ccall_ctid_vararg(cts, o); /* Infer vararg type. */ |
| 1003 | isva = 1; |
| 1004 | } |
| 1005 | d = ctype_raw(cts, did); |
| 1006 | sz = d->size; |
| 1007 | |
| 1008 | /* Find out how (by value/ref) and where (GPR/FPR) to pass an argument. */ |
| 1009 | if (ctype_isnum(d->info)) { |
| 1010 | if (sz > 8) goto err_nyi; |
| 1011 | if ((d->info & CTF_FP)) |
| 1012 | isfp = 1; |
| 1013 | } else if (ctype_isvector(d->info)) { |
| 1014 | if (CCALL_VECTOR_REG && (sz == 8 || sz == 16)) |
| 1015 | isfp = 1; |
| 1016 | else |
| 1017 | goto err_nyi; |
| 1018 | } else if (ctype_isstruct(d->info)) { |
| 1019 | CCALL_HANDLE_STRUCTARG |
| 1020 | } else if (ctype_iscomplex(d->info)) { |
| 1021 | CCALL_HANDLE_COMPLEXARG |
| 1022 | } else { |
| 1023 | sz = CTSIZE_PTR; |
| 1024 | } |
| 1025 | sz = (sz + CTSIZE_PTR-1) & ~(CTSIZE_PTR-1); |
| 1026 | n = sz / CTSIZE_PTR; /* Number of GPRs or stack slots needed. */ |
| 1027 | |
| 1028 | CCALL_HANDLE_REGARG /* Handle register arguments. */ |
| 1029 | |
| 1030 | /* Otherwise pass argument on stack. */ |
| 1031 | if (CCALL_ALIGN_STACKARG && !rp && (d->info & CTF_ALIGN) > CTALIGN_PTR) { |
| 1032 | MSize align = (1u << ctype_align(d->info-CTALIGN_PTR)) -1; |
| 1033 | nsp = (nsp + align) & ~align; /* Align argument on stack. */ |
| 1034 | } |
| 1035 | if (nsp + n > CCALL_MAXSTACK) { /* Too many arguments. */ |
| 1036 | err_nyi: |
| 1037 | lj_err_caller(L, LJ_ERR_FFI_NYICALL); |
| 1038 | } |
| 1039 | dp = &cc->stack[nsp]; |
| 1040 | nsp += n; |
| 1041 | isva = 0; |
| 1042 | |
| 1043 | done: |
| 1044 | if (rp) { /* Pass by reference. */ |
| 1045 | gcsteps++; |
| 1046 | *(void **)dp = rp; |
| 1047 | dp = rp; |
| 1048 | } |
| 1049 | lj_cconv_ct_tv(cts, d, (uint8_t *)dp, o, CCF_ARG(narg)); |
| 1050 | /* Extend passed integers to 32 bits at least. */ |
| 1051 | if (ctype_isinteger_or_bool(d->info) && d->size < 4) { |
| 1052 | if (d->info & CTF_UNSIGNED) |
| 1053 | *(uint32_t *)dp = d->size == 1 ? (uint32_t)*(uint8_t *)dp : |
| 1054 | (uint32_t)*(uint16_t *)dp; |
| 1055 | else |
| 1056 | *(int32_t *)dp = d->size == 1 ? (int32_t)*(int8_t *)dp : |
| 1057 | (int32_t)*(int16_t *)dp; |
| 1058 | } |
| 1059 | #if LJ_TARGET_ARM64 && LJ_BE |
| 1060 | if (isfp && d->size == sizeof(float)) |
| 1061 | ((float *)dp)[1] = ((float *)dp)[0]; /* Floats occupy high slot. */ |
| 1062 | #endif |
| 1063 | #if LJ_TARGET_MIPS64 || (LJ_TARGET_ARM64 && LJ_BE) |
| 1064 | if ((ctype_isinteger_or_bool(d->info) || ctype_isenum(d->info) |
| 1065 | #if LJ_TARGET_MIPS64 |
| 1066 | || (isfp && nsp == 0) |
| 1067 | #endif |
| 1068 | ) && d->size <= 4) { |
| 1069 | *(int64_t *)dp = (int64_t)*(int32_t *)dp; /* Sign-extend to 64 bit. */ |
| 1070 | } |
| 1071 | #endif |
| 1072 | #if LJ_TARGET_X64 && LJ_ABI_WIN |
| 1073 | if (isva) { /* Windows/x64 mirrors varargs in both register sets. */ |
| 1074 | if (nfpr == ngpr) |
| 1075 | cc->gpr[ngpr-1] = cc->fpr[ngpr-1].l[0]; |
| 1076 | else |
| 1077 | cc->fpr[ngpr-1].l[0] = cc->gpr[ngpr-1]; |
| 1078 | } |
| 1079 | #else |
| 1080 | UNUSED(isva); |
| 1081 | #endif |
| 1082 | #if LJ_TARGET_X64 && !LJ_ABI_WIN |
| 1083 | if (isfp == 2 && n == 2 && (uint8_t *)dp == (uint8_t *)&cc->fpr[nfpr-2]) { |
| 1084 | cc->fpr[nfpr-1].d[0] = cc->fpr[nfpr-2].d[1]; /* Split complex double. */ |
| 1085 | cc->fpr[nfpr-2].d[1] = 0; |
| 1086 | } |
| 1087 | #elif LJ_TARGET_ARM64 || (LJ_TARGET_MIPS64 && !LJ_ABI_SOFTFP) |
| 1088 | if (isfp == 2 && (uint8_t *)dp < (uint8_t *)cc->stack) { |
| 1089 | /* Split float HFA or complex float into separate registers. */ |
| 1090 | CTSize i = (sz >> 2) - 1; |
| 1091 | do { ((uint64_t *)dp)[i] = ((uint32_t *)dp)[i]; } while (i--); |
| 1092 | } |
| 1093 | #else |
| 1094 | UNUSED(isfp); |
| 1095 | #endif |
| 1096 | } |
| 1097 | if (fid) lj_err_caller(L, LJ_ERR_FFI_NUMARG); /* Too few arguments. */ |
| 1098 | |
| 1099 | #if LJ_TARGET_X64 || (LJ_TARGET_PPC && !LJ_ABI_SOFTFP) |
| 1100 | cc->nfpr = nfpr; /* Required for vararg functions. */ |
| 1101 | #endif |
| 1102 | cc->nsp = nsp; |
| 1103 | cc->spadj = (CCALL_SPS_FREE + CCALL_SPS_EXTRA)*CTSIZE_PTR; |
| 1104 | if (nsp > CCALL_SPS_FREE) |
| 1105 | cc->spadj += (((nsp-CCALL_SPS_FREE)*CTSIZE_PTR + 15u) & ~15u); |
| 1106 | return gcsteps; |
| 1107 | } |
| 1108 | |
| 1109 | /* Get results from C call. */ |
| 1110 | static int ccall_get_results(lua_State *L, CTState *cts, CType *ct, |
| 1111 | CCallState *cc, int *ret) |
| 1112 | { |
| 1113 | CType *ctr = ctype_rawchild(cts, ct); |
| 1114 | uint8_t *sp = (uint8_t *)&cc->gpr[0]; |
| 1115 | if (ctype_isvoid(ctr->info)) { |
| 1116 | *ret = 0; /* Zero results. */ |
| 1117 | return 0; /* No additional GC step. */ |
| 1118 | } |
| 1119 | *ret = 1; /* One result. */ |
| 1120 | if (ctype_isstruct(ctr->info)) { |
| 1121 | /* Return cdata object which is already on top of stack. */ |
| 1122 | if (!cc->retref) { |
| 1123 | void *dp = cdataptr(cdataV(L->top-1)); /* Use preallocated object. */ |
| 1124 | CCALL_HANDLE_STRUCTRET2 |
| 1125 | } |
| 1126 | return 1; /* One GC step. */ |
| 1127 | } |
| 1128 | if (ctype_iscomplex(ctr->info)) { |
| 1129 | /* Return cdata object which is already on top of stack. */ |
| 1130 | void *dp = cdataptr(cdataV(L->top-1)); /* Use preallocated object. */ |
| 1131 | CCALL_HANDLE_COMPLEXRET2 |
| 1132 | return 1; /* One GC step. */ |
| 1133 | } |
| 1134 | if (LJ_BE && ctr->size < CTSIZE_PTR && |
| 1135 | (ctype_isinteger_or_bool(ctr->info) || ctype_isenum(ctr->info))) |
| 1136 | sp += (CTSIZE_PTR - ctr->size); |
| 1137 | #if CCALL_NUM_FPR |
| 1138 | if (ctype_isfp(ctr->info) || ctype_isvector(ctr->info)) |
| 1139 | sp = (uint8_t *)&cc->fpr[0]; |
| 1140 | #endif |
| 1141 | #ifdef CCALL_HANDLE_RET |
| 1142 | CCALL_HANDLE_RET |
| 1143 | #endif |
| 1144 | /* No reference types end up here, so there's no need for the CTypeID. */ |
| 1145 | lj_assertL(!(ctype_isrefarray(ctr->info) || ctype_isstruct(ctr->info)), |
| 1146 | "unexpected reference ctype" ); |
| 1147 | return lj_cconv_tv_ct(cts, ctr, 0, L->top-1, sp); |
| 1148 | } |
| 1149 | |
| 1150 | /* Call C function. */ |
| 1151 | int lj_ccall_func(lua_State *L, GCcdata *cd) |
| 1152 | { |
| 1153 | CTState *cts = ctype_cts(L); |
| 1154 | CType *ct = ctype_raw(cts, cd->ctypeid); |
| 1155 | CTSize sz = CTSIZE_PTR; |
| 1156 | if (ctype_isptr(ct->info)) { |
| 1157 | sz = ct->size; |
| 1158 | ct = ctype_rawchild(cts, ct); |
| 1159 | } |
| 1160 | if (ctype_isfunc(ct->info)) { |
| 1161 | CCallState cc; |
| 1162 | int gcsteps, ret; |
| 1163 | cc.func = (void (*)(void))cdata_getptr(cdataptr(cd), sz); |
| 1164 | gcsteps = ccall_set_args(L, cts, ct, &cc); |
| 1165 | ct = (CType *)((intptr_t)ct-(intptr_t)cts->tab); |
| 1166 | cts->cb.slot = ~0u; |
| 1167 | lj_vm_ffi_call(&cc); |
| 1168 | if (cts->cb.slot != ~0u) { /* Blacklist function that called a callback. */ |
| 1169 | TValue tv; |
| 1170 | tv.u64 = ((uintptr_t)(void *)cc.func >> 2) | U64x(800000000, 00000000); |
| 1171 | setboolV(lj_tab_set(L, cts->miscmap, &tv), 1); |
| 1172 | } |
| 1173 | ct = (CType *)((intptr_t)ct+(intptr_t)cts->tab); /* May be reallocated. */ |
| 1174 | gcsteps += ccall_get_results(L, cts, ct, &cc, &ret); |
| 1175 | #if LJ_TARGET_X86 && LJ_ABI_WIN |
| 1176 | /* Automatically detect __stdcall and fix up C function declaration. */ |
| 1177 | if (cc.spadj && ctype_cconv(ct->info) == CTCC_CDECL) { |
| 1178 | CTF_INSERT(ct->info, CCONV, CTCC_STDCALL); |
| 1179 | lj_trace_abort(G(L)); |
| 1180 | } |
| 1181 | #endif |
| 1182 | while (gcsteps-- > 0) |
| 1183 | lj_gc_check(L); |
| 1184 | return ret; |
| 1185 | } |
| 1186 | return -1; /* Not a function. */ |
| 1187 | } |
| 1188 | |
| 1189 | #endif |
| 1190 | |