| 1 |
/* ----------------------------------------------------------------------- |
| 2 |
ffi.c - Copyright (c) 2011 Timothy Wall |
| 3 |
Copyright (c) 2011 Plausible Labs Cooperative, Inc. |
| 4 |
Copyright (c) 2011 Anthony Green |
| 5 |
Copyright (c) 2011 Free Software Foundation |
| 6 |
Copyright (c) 1998, 2008, 2011 Red Hat, Inc. |
| 7 |
|
| 8 |
ARM Foreign Function Interface |
| 9 |
|
| 10 |
Permission is hereby granted, free of charge, to any person obtaining |
| 11 |
a copy of this software and associated documentation files (the |
| 12 |
``Software''), to deal in the Software without restriction, including |
| 13 |
without limitation the rights to use, copy, modify, merge, publish, |
| 14 |
distribute, sublicense, and/or sell copies of the Software, and to |
| 15 |
permit persons to whom the Software is furnished to do so, subject to |
| 16 |
the following conditions: |
| 17 |
|
| 18 |
The above copyright notice and this permission notice shall be included |
| 19 |
in all copies or substantial portions of the Software. |
| 20 |
|
| 21 |
THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, |
| 22 |
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF |
| 23 |
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND |
| 24 |
NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT |
| 25 |
HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, |
| 26 |
WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 27 |
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER |
| 28 |
DEALINGS IN THE SOFTWARE. |
| 29 |
----------------------------------------------------------------------- */ |
| 30 |
|
| 31 |
#if defined(__arm__) || defined(_M_ARM) |
| 32 |
#include <fficonfig.h> |
| 33 |
#include <ffi.h> |
| 34 |
#include <ffi_common.h> |
| 35 |
#include <stdint.h> |
| 36 |
#include <stdlib.h> |
| 37 |
#include <tramp.h> |
| 38 |
#include "internal.h" |
| 39 |
|
| 40 |
#if defined(_WIN32) |
| 41 |
#define WIN32_LEAN_AND_MEAN |
| 42 |
#include <windows.h> |
| 43 |
#endif |
| 44 |
|
| 45 |
#if FFI_EXEC_TRAMPOLINE_TABLE |
| 46 |
|
| 47 |
#ifdef __MACH__ |
| 48 |
#include <mach/machine/vm_param.h> |
| 49 |
#endif |
| 50 |
|
| 51 |
#else |
| 52 |
#ifndef _WIN32 |
| 53 |
extern unsigned int ffi_arm_trampoline[2] FFI_HIDDEN; |
| 54 |
#else |
| 55 |
// Declare this as an array of char, instead of array of int, |
| 56 |
// otherwise Clang optimizes out the "& 0xFFFFFFFE" for clearing |
| 57 |
// the thumb bit. |
| 58 |
extern unsigned char ffi_arm_trampoline[12] FFI_HIDDEN; |
| 59 |
#endif |
| 60 |
#endif |
| 61 |
|
| 62 |
#if defined(__FreeBSD__) && defined(__arm__) |
| 63 |
#include <sys/types.h> |
| 64 |
#include <machine/sysarch.h> |
| 65 |
#endif |
| 66 |
|
| 67 |
#if defined(__QNX__) |
| 68 |
#include <sys/mman.h> |
| 69 |
#endif |
| 70 |
|
| 71 |
/* Forward declares. */ |
| 72 |
static int vfp_type_p (const ffi_type *); |
| 73 |
static void layout_vfp_args (ffi_cif *); |
| 74 |
|
| 75 |
static void * |
| 76 |
ffi_align (ffi_type *ty, void *p) |
| 77 |
{ |
| 78 |
/* Align if necessary */ |
| 79 |
size_t alignment; |
| 80 |
#ifdef _WIN32_WCE |
| 81 |
alignment = 4; |
| 82 |
#else |
| 83 |
alignment = ty->alignment; |
| 84 |
if (alignment < 4) |
| 85 |
alignment = 4; |
| 86 |
#endif |
| 87 |
return (void *) FFI_ALIGN (p, alignment); |
| 88 |
} |
| 89 |
|
| 90 |
static size_t |
| 91 |
ffi_put_arg (ffi_type *ty, void *src, void *dst) |
| 92 |
{ |
| 93 |
size_t z = ty->size; |
| 94 |
|
| 95 |
switch (ty->type) |
| 96 |
{ |
| 97 |
case FFI_TYPE_SINT8: |
| 98 |
*(UINT32 *)dst = *(SINT8 *)src; |
| 99 |
break; |
| 100 |
case FFI_TYPE_UINT8: |
| 101 |
*(UINT32 *)dst = *(UINT8 *)src; |
| 102 |
break; |
| 103 |
case FFI_TYPE_SINT16: |
| 104 |
*(UINT32 *)dst = *(SINT16 *)src; |
| 105 |
break; |
| 106 |
case FFI_TYPE_UINT16: |
| 107 |
*(UINT32 *)dst = *(UINT16 *)src; |
| 108 |
break; |
| 109 |
|
| 110 |
case FFI_TYPE_INT: |
| 111 |
case FFI_TYPE_SINT32: |
| 112 |
case FFI_TYPE_UINT32: |
| 113 |
case FFI_TYPE_POINTER: |
| 114 |
#ifndef _WIN32 |
| 115 |
case FFI_TYPE_FLOAT: |
| 116 |
#endif |
| 117 |
*(UINT32 *)dst = *(UINT32 *)src; |
| 118 |
break; |
| 119 |
|
| 120 |
#ifdef _WIN32 |
| 121 |
// casting a float* to a UINT32* doesn't work on Windows |
| 122 |
case FFI_TYPE_FLOAT: |
| 123 |
*(uintptr_t *)dst = 0; |
| 124 |
*(float *)dst = *(float *)src; |
| 125 |
break; |
| 126 |
#endif |
| 127 |
|
| 128 |
case FFI_TYPE_SINT64: |
| 129 |
case FFI_TYPE_UINT64: |
| 130 |
case FFI_TYPE_DOUBLE: |
| 131 |
*(UINT64 *)dst = *(UINT64 *)src; |
| 132 |
break; |
| 133 |
|
| 134 |
case FFI_TYPE_STRUCT: |
| 135 |
case FFI_TYPE_COMPLEX: |
| 136 |
memcpy (dst, src, z); |
| 137 |
break; |
| 138 |
|
| 139 |
default: |
| 140 |
abort(); |
| 141 |
} |
| 142 |
|
| 143 |
return FFI_ALIGN (z, 4); |
| 144 |
} |
| 145 |
|
| 146 |
/* ffi_prep_args is called once stack space has been allocated |
| 147 |
for the function's arguments. |
| 148 |
|
| 149 |
The vfp_space parameter is the load area for VFP regs, the return |
| 150 |
value is cif->vfp_used (word bitset of VFP regs used for passing |
| 151 |
arguments). These are only used for the VFP hard-float ABI. |
| 152 |
*/ |
| 153 |
static void |
| 154 |
ffi_prep_args_SYSV (ffi_cif *cif, int flags, void *rvalue, |
| 155 |
void **avalue, char *argp) |
| 156 |
{ |
| 157 |
ffi_type **arg_types = cif->arg_types; |
| 158 |
int i, n; |
| 159 |
|
| 160 |
if (flags == ARM_TYPE_STRUCT) |
| 161 |
{ |
| 162 |
*(void **) argp = rvalue; |
| 163 |
argp += 4; |
| 164 |
} |
| 165 |
|
| 166 |
for (i = 0, n = cif->nargs; i < n; i++) |
| 167 |
{ |
| 168 |
ffi_type *ty = arg_types[i]; |
| 169 |
argp = ffi_align (ty, argp); |
| 170 |
argp += ffi_put_arg (ty, avalue[i], argp); |
| 171 |
} |
| 172 |
} |
| 173 |
|
| 174 |
static void |
| 175 |
ffi_prep_args_VFP (ffi_cif *cif, int flags, void *rvalue, |
| 176 |
void **avalue, char *stack, char *vfp_space) |
| 177 |
{ |
| 178 |
ffi_type **arg_types = cif->arg_types; |
| 179 |
int i, n, vi = 0; |
| 180 |
char *argp, *regp, *eo_regp; |
| 181 |
char stack_used = 0; |
| 182 |
char done_with_regs = 0; |
| 183 |
|
| 184 |
/* The first 4 words on the stack are used for values |
| 185 |
passed in core registers. */ |
| 186 |
regp = stack; |
| 187 |
eo_regp = argp = regp + 16; |
| 188 |
|
| 189 |
/* If the function returns an FFI_TYPE_STRUCT in memory, |
| 190 |
that address is passed in r0 to the function. */ |
| 191 |
if (flags == ARM_TYPE_STRUCT) |
| 192 |
{ |
| 193 |
*(void **) regp = rvalue; |
| 194 |
regp += 4; |
| 195 |
} |
| 196 |
|
| 197 |
for (i = 0, n = cif->nargs; i < n; i++) |
| 198 |
{ |
| 199 |
ffi_type *ty = arg_types[i]; |
| 200 |
void *a = avalue[i]; |
| 201 |
int is_vfp_type = vfp_type_p (ty); |
| 202 |
|
| 203 |
/* Allocated in VFP registers. */ |
| 204 |
if (vi < cif->vfp_nargs && is_vfp_type) |
| 205 |
{ |
| 206 |
char *vfp_slot = vfp_space + cif->vfp_args[vi++] * 4; |
| 207 |
ffi_put_arg (ty, a, vfp_slot); |
| 208 |
continue; |
| 209 |
} |
| 210 |
/* Try allocating in core registers. */ |
| 211 |
else if (!done_with_regs && !is_vfp_type) |
| 212 |
{ |
| 213 |
char *tregp = ffi_align (ty, regp); |
| 214 |
size_t size = ty->size; |
| 215 |
size = (size < 4) ? 4 : size; // pad |
| 216 |
/* Check if there is space left in the aligned register |
| 217 |
area to place the argument. */ |
| 218 |
if (tregp + size <= eo_regp) |
| 219 |
{ |
| 220 |
regp = tregp + ffi_put_arg (ty, a, tregp); |
| 221 |
done_with_regs = (regp == argp); |
| 222 |
// ensure we did not write into the stack area |
| 223 |
FFI_ASSERT (regp <= argp); |
| 224 |
continue; |
| 225 |
} |
| 226 |
/* In case there are no arguments in the stack area yet, |
| 227 |
the argument is passed in the remaining core registers |
| 228 |
and on the stack. */ |
| 229 |
else if (!stack_used) |
| 230 |
{ |
| 231 |
stack_used = 1; |
| 232 |
done_with_regs = 1; |
| 233 |
argp = tregp + ffi_put_arg (ty, a, tregp); |
| 234 |
FFI_ASSERT (eo_regp < argp); |
| 235 |
continue; |
| 236 |
} |
| 237 |
} |
| 238 |
/* Base case, arguments are passed on the stack */ |
| 239 |
stack_used = 1; |
| 240 |
argp = ffi_align (ty, argp); |
| 241 |
argp += ffi_put_arg (ty, a, argp); |
| 242 |
} |
| 243 |
} |
| 244 |
|
| 245 |
/* Perform machine dependent cif processing */ |
| 246 |
ffi_status FFI_HIDDEN |
| 247 |
ffi_prep_cif_machdep (ffi_cif *cif) |
| 248 |
{ |
| 249 |
int flags = 0, cabi = cif->abi; |
| 250 |
size_t bytes = cif->bytes; |
| 251 |
|
| 252 |
/* Map out the register placements of VFP register args. The VFP |
| 253 |
hard-float calling conventions are slightly more sophisticated |
| 254 |
than the base calling conventions, so we do it here instead of |
| 255 |
in ffi_prep_args(). */ |
| 256 |
if (cabi == FFI_VFP) |
| 257 |
layout_vfp_args (cif); |
| 258 |
|
| 259 |
/* Set the return type flag */ |
| 260 |
switch (cif->rtype->type) |
| 261 |
{ |
| 262 |
case FFI_TYPE_VOID: |
| 263 |
flags = ARM_TYPE_VOID; |
| 264 |
break; |
| 265 |
|
| 266 |
case FFI_TYPE_INT: |
| 267 |
case FFI_TYPE_UINT8: |
| 268 |
case FFI_TYPE_SINT8: |
| 269 |
case FFI_TYPE_UINT16: |
| 270 |
case FFI_TYPE_SINT16: |
| 271 |
case FFI_TYPE_UINT32: |
| 272 |
case FFI_TYPE_SINT32: |
| 273 |
case FFI_TYPE_POINTER: |
| 274 |
flags = ARM_TYPE_INT; |
| 275 |
break; |
| 276 |
|
| 277 |
case FFI_TYPE_SINT64: |
| 278 |
case FFI_TYPE_UINT64: |
| 279 |
flags = ARM_TYPE_INT64; |
| 280 |
break; |
| 281 |
|
| 282 |
case FFI_TYPE_FLOAT: |
| 283 |
flags = (cabi == FFI_VFP ? ARM_TYPE_VFP_S : ARM_TYPE_INT); |
| 284 |
break; |
| 285 |
case FFI_TYPE_DOUBLE: |
| 286 |
flags = (cabi == FFI_VFP ? ARM_TYPE_VFP_D : ARM_TYPE_INT64); |
| 287 |
break; |
| 288 |
|
| 289 |
case FFI_TYPE_STRUCT: |
| 290 |
case FFI_TYPE_COMPLEX: |
| 291 |
if (cabi == FFI_VFP) |
| 292 |
{ |
| 293 |
int h = vfp_type_p (cif->rtype); |
| 294 |
|
| 295 |
flags = ARM_TYPE_VFP_N; |
| 296 |
if (h == 0x100 + FFI_TYPE_FLOAT) |
| 297 |
flags = ARM_TYPE_VFP_S; |
| 298 |
if (h == 0x100 + FFI_TYPE_DOUBLE) |
| 299 |
flags = ARM_TYPE_VFP_D; |
| 300 |
if (h != 0) |
| 301 |
break; |
| 302 |
} |
| 303 |
|
| 304 |
/* A Composite Type not larger than 4 bytes is returned in r0. |
| 305 |
A Composite Type larger than 4 bytes, or whose size cannot |
| 306 |
be determined statically ... is stored in memory at an |
| 307 |
address passed [in r0]. */ |
| 308 |
if (cif->rtype->size <= 4) |
| 309 |
flags = ARM_TYPE_INT; |
| 310 |
else |
| 311 |
{ |
| 312 |
flags = ARM_TYPE_STRUCT; |
| 313 |
bytes += 4; |
| 314 |
} |
| 315 |
break; |
| 316 |
|
| 317 |
default: |
| 318 |
abort(); |
| 319 |
} |
| 320 |
|
| 321 |
/* Round the stack up to a multiple of 8 bytes. This isn't needed |
| 322 |
everywhere, but it is on some platforms, and it doesn't harm anything |
| 323 |
when it isn't needed. */ |
| 324 |
bytes = FFI_ALIGN (bytes, 8); |
| 325 |
|
| 326 |
/* Minimum stack space is the 4 register arguments that we pop. */ |
| 327 |
if (bytes < 4*4) |
| 328 |
bytes = 4*4; |
| 329 |
|
| 330 |
cif->bytes = bytes; |
| 331 |
cif->flags = flags; |
| 332 |
|
| 333 |
return FFI_OK; |
| 334 |
} |
| 335 |
|
| 336 |
/* Perform machine dependent cif processing for variadic calls */ |
| 337 |
ffi_status FFI_HIDDEN |
| 338 |
ffi_prep_cif_machdep_var (ffi_cif * cif, |
| 339 |
unsigned int nfixedargs, unsigned int ntotalargs) |
| 340 |
{ |
| 341 |
/* VFP variadic calls actually use the SYSV ABI */ |
| 342 |
if (cif->abi == FFI_VFP) |
| 343 |
cif->abi = FFI_SYSV; |
| 344 |
|
| 345 |
return ffi_prep_cif_machdep (cif); |
| 346 |
} |
| 347 |
|
| 348 |
/* Prototypes for assembly functions, in sysv.S. */ |
| 349 |
|
| 350 |
struct call_frame |
| 351 |
{ |
| 352 |
void *fp; |
| 353 |
void *lr; |
| 354 |
void *rvalue; |
| 355 |
int flags; |
| 356 |
void *closure; |
| 357 |
}; |
| 358 |
|
| 359 |
extern void ffi_call_SYSV (void *stack, struct call_frame *, |
| 360 |
void (*fn) (void)) FFI_HIDDEN; |
| 361 |
extern void ffi_call_VFP (void *vfp_space, struct call_frame *, |
| 362 |
void (*fn) (void), unsigned vfp_used) FFI_HIDDEN; |
| 363 |
|
| 364 |
static void |
| 365 |
ffi_call_int (ffi_cif * cif, void (*fn) (void), void *rvalue, |
| 366 |
void **avalue, void *closure) |
| 367 |
{ |
| 368 |
int flags = cif->flags; |
| 369 |
ffi_type *rtype = cif->rtype; |
| 370 |
size_t bytes, rsize, vfp_size; |
| 371 |
char *stack, *vfp_space, *new_rvalue; |
| 372 |
struct call_frame *frame; |
| 373 |
|
| 374 |
rsize = 0; |
| 375 |
if (rvalue == NULL) |
| 376 |
{ |
| 377 |
/* If the return value is a struct and we don't have a return |
| 378 |
value address then we need to make one. Otherwise the return |
| 379 |
value is in registers and we can ignore them. */ |
| 380 |
if (flags == ARM_TYPE_STRUCT) |
| 381 |
rsize = rtype->size; |
| 382 |
else |
| 383 |
flags = ARM_TYPE_VOID; |
| 384 |
} |
| 385 |
else if (flags == ARM_TYPE_VFP_N) |
| 386 |
{ |
| 387 |
/* Largest case is double x 4. */ |
| 388 |
rsize = 32; |
| 389 |
} |
| 390 |
else if (flags == ARM_TYPE_INT && rtype->type == FFI_TYPE_STRUCT) |
| 391 |
rsize = 4; |
| 392 |
|
| 393 |
/* Largest case. */ |
| 394 |
vfp_size = (cif->abi == FFI_VFP && cif->vfp_used ? 8*8: 0); |
| 395 |
|
| 396 |
bytes = cif->bytes; |
| 397 |
stack = alloca (vfp_size + bytes + sizeof(struct call_frame) + rsize); |
| 398 |
|
| 399 |
vfp_space = NULL; |
| 400 |
if (vfp_size) |
| 401 |
{ |
| 402 |
vfp_space = stack; |
| 403 |
stack += vfp_size; |
| 404 |
} |
| 405 |
|
| 406 |
frame = (struct call_frame *)(stack + bytes); |
| 407 |
|
| 408 |
new_rvalue = rvalue; |
| 409 |
if (rsize) |
| 410 |
new_rvalue = (void *)(frame + 1); |
| 411 |
|
| 412 |
frame->rvalue = new_rvalue; |
| 413 |
frame->flags = flags; |
| 414 |
frame->closure = closure; |
| 415 |
|
| 416 |
if (vfp_space) |
| 417 |
{ |
| 418 |
ffi_prep_args_VFP (cif, flags, new_rvalue, avalue, stack, vfp_space); |
| 419 |
ffi_call_VFP (vfp_space, frame, fn, cif->vfp_used); |
| 420 |
} |
| 421 |
else |
| 422 |
{ |
| 423 |
ffi_prep_args_SYSV (cif, flags, new_rvalue, avalue, stack); |
| 424 |
ffi_call_SYSV (stack, frame, fn); |
| 425 |
} |
| 426 |
|
| 427 |
if (rvalue && rvalue != new_rvalue) |
| 428 |
memcpy (rvalue, new_rvalue, rtype->size); |
| 429 |
} |
| 430 |
|
| 431 |
void |
| 432 |
ffi_call (ffi_cif *cif, void (*fn) (void), void *rvalue, void **avalue) |
| 433 |
{ |
| 434 |
ffi_call_int (cif, fn, rvalue, avalue, NULL); |
| 435 |
} |
| 436 |
|
| 437 |
#ifdef FFI_GO_CLOSURES |
| 438 |
void |
| 439 |
ffi_call_go (ffi_cif *cif, void (*fn) (void), void *rvalue, |
| 440 |
void **avalue, void *closure) |
| 441 |
{ |
| 442 |
ffi_call_int (cif, fn, rvalue, avalue, closure); |
| 443 |
} |
| 444 |
#endif |
| 445 |
|
| 446 |
static void * |
| 447 |
ffi_prep_incoming_args_SYSV (ffi_cif *cif, void *rvalue, |
| 448 |
char *argp, void **avalue) |
| 449 |
{ |
| 450 |
ffi_type **arg_types = cif->arg_types; |
| 451 |
int i, n; |
| 452 |
|
| 453 |
if (cif->flags == ARM_TYPE_STRUCT) |
| 454 |
{ |
| 455 |
rvalue = *(void **) argp; |
| 456 |
argp += 4; |
| 457 |
} |
| 458 |
else |
| 459 |
{ |
| 460 |
if (cif->rtype->size && cif->rtype->size < 4) |
| 461 |
*(uint32_t *) rvalue = 0; |
| 462 |
} |
| 463 |
|
| 464 |
for (i = 0, n = cif->nargs; i < n; i++) |
| 465 |
{ |
| 466 |
ffi_type *ty = arg_types[i]; |
| 467 |
size_t z = ty->size; |
| 468 |
|
| 469 |
argp = ffi_align (ty, argp); |
| 470 |
avalue[i] = (void *) argp; |
| 471 |
argp += z; |
| 472 |
} |
| 473 |
|
| 474 |
return rvalue; |
| 475 |
} |
| 476 |
|
| 477 |
static void * |
| 478 |
ffi_prep_incoming_args_VFP (ffi_cif *cif, void *rvalue, char *stack, |
| 479 |
char *vfp_space, void **avalue) |
| 480 |
{ |
| 481 |
ffi_type **arg_types = cif->arg_types; |
| 482 |
int i, n, vi = 0; |
| 483 |
char *argp, *regp, *eo_regp; |
| 484 |
char done_with_regs = 0; |
| 485 |
char stack_used = 0; |
| 486 |
|
| 487 |
regp = stack; |
| 488 |
eo_regp = argp = regp + 16; |
| 489 |
|
| 490 |
if (cif->flags == ARM_TYPE_STRUCT) |
| 491 |
{ |
| 492 |
rvalue = *(void **) regp; |
| 493 |
regp += 4; |
| 494 |
} |
| 495 |
|
| 496 |
for (i = 0, n = cif->nargs; i < n; i++) |
| 497 |
{ |
| 498 |
ffi_type *ty = arg_types[i]; |
| 499 |
int is_vfp_type = vfp_type_p (ty); |
| 500 |
size_t z = ty->size; |
| 501 |
|
| 502 |
if (vi < cif->vfp_nargs && is_vfp_type) |
| 503 |
{ |
| 504 |
avalue[i] = vfp_space + cif->vfp_args[vi++] * 4; |
| 505 |
continue; |
| 506 |
} |
| 507 |
else if (!done_with_regs && !is_vfp_type) |
| 508 |
{ |
| 509 |
char *tregp = ffi_align (ty, regp); |
| 510 |
|
| 511 |
z = (z < 4) ? 4 : z; // pad |
| 512 |
|
| 513 |
/* If the arguments either fits into the registers or uses registers |
| 514 |
and stack, while we haven't read other things from the stack */ |
| 515 |
if (tregp + z <= eo_regp || !stack_used) |
| 516 |
{ |
| 517 |
/* Because we're little endian, this is what it turns into. */ |
| 518 |
avalue[i] = (void *) tregp; |
| 519 |
regp = tregp + z; |
| 520 |
|
| 521 |
/* If we read past the last core register, make sure we |
| 522 |
have not read from the stack before and continue |
| 523 |
reading after regp. */ |
| 524 |
if (regp > eo_regp) |
| 525 |
{ |
| 526 |
FFI_ASSERT (!stack_used); |
| 527 |
argp = regp; |
| 528 |
} |
| 529 |
if (regp >= eo_regp) |
| 530 |
{ |
| 531 |
done_with_regs = 1; |
| 532 |
stack_used = 1; |
| 533 |
} |
| 534 |
continue; |
| 535 |
} |
| 536 |
} |
| 537 |
|
| 538 |
stack_used = 1; |
| 539 |
argp = ffi_align (ty, argp); |
| 540 |
avalue[i] = (void *) argp; |
| 541 |
argp += z; |
| 542 |
} |
| 543 |
|
| 544 |
return rvalue; |
| 545 |
} |
| 546 |
|
| 547 |
#if FFI_CLOSURES |
| 548 |
|
| 549 |
struct closure_frame |
| 550 |
{ |
| 551 |
char vfp_space[8*8] __attribute__((aligned(8))); |
| 552 |
char result[8*4]; |
| 553 |
char argp[]; |
| 554 |
}; |
| 555 |
|
| 556 |
int FFI_HIDDEN |
| 557 |
ffi_closure_inner_SYSV (ffi_cif *cif, |
| 558 |
void (*fun) (ffi_cif *, void *, void **, void *), |
| 559 |
void *user_data, |
| 560 |
struct closure_frame *frame) |
| 561 |
{ |
| 562 |
void **avalue = (void **) alloca (cif->nargs * sizeof (void *)); |
| 563 |
void *rvalue = ffi_prep_incoming_args_SYSV (cif, frame->result, |
| 564 |
frame->argp, avalue); |
| 565 |
fun (cif, rvalue, avalue, user_data); |
| 566 |
return cif->flags; |
| 567 |
} |
| 568 |
|
| 569 |
int FFI_HIDDEN |
| 570 |
ffi_closure_inner_VFP (ffi_cif *cif, |
| 571 |
void (*fun) (ffi_cif *, void *, void **, void *), |
| 572 |
void *user_data, |
| 573 |
struct closure_frame *frame) |
| 574 |
{ |
| 575 |
void **avalue = (void **) alloca (cif->nargs * sizeof (void *)); |
| 576 |
void *rvalue = ffi_prep_incoming_args_VFP (cif, frame->result, frame->argp, |
| 577 |
frame->vfp_space, avalue); |
| 578 |
fun (cif, rvalue, avalue, user_data); |
| 579 |
return cif->flags; |
| 580 |
} |
| 581 |
|
| 582 |
void ffi_closure_SYSV (void) FFI_HIDDEN; |
| 583 |
void ffi_closure_VFP (void) FFI_HIDDEN; |
| 584 |
#if defined(FFI_EXEC_STATIC_TRAMP) |
| 585 |
void ffi_closure_SYSV_alt (void) FFI_HIDDEN; |
| 586 |
void ffi_closure_VFP_alt (void) FFI_HIDDEN; |
| 587 |
#endif |
| 588 |
|
| 589 |
#ifdef FFI_GO_CLOSURES |
| 590 |
void ffi_go_closure_SYSV (void) FFI_HIDDEN; |
| 591 |
void ffi_go_closure_VFP (void) FFI_HIDDEN; |
| 592 |
#endif |
| 593 |
|
| 594 |
/* the cif must already be prep'ed */ |
| 595 |
|
| 596 |
#if defined(__FreeBSD__) && defined(__arm__) |
| 597 |
#define __clear_cache(start, end) do { \ |
| 598 |
struct arm_sync_icache_args ua; \ |
| 599 |
\ |
| 600 |
ua.addr = (uintptr_t)(start); \ |
| 601 |
ua.len = (char *)(end) - (char *)start; \ |
| 602 |
sysarch(ARM_SYNC_ICACHE, &ua); \ |
| 603 |
} while (0); |
| 604 |
#endif |
| 605 |
|
| 606 |
ffi_status |
| 607 |
ffi_prep_closure_loc (ffi_closure * closure, |
| 608 |
ffi_cif * cif, |
| 609 |
void (*fun) (ffi_cif *, void *, void **, void *), |
| 610 |
void *user_data, void *codeloc) |
| 611 |
{ |
| 612 |
void (*closure_func) (void) = ffi_closure_SYSV; |
| 613 |
|
| 614 |
if (cif->abi == FFI_VFP) |
| 615 |
{ |
| 616 |
/* We only need take the vfp path if there are vfp arguments. */ |
| 617 |
if (cif->vfp_used) |
| 618 |
closure_func = ffi_closure_VFP; |
| 619 |
} |
| 620 |
else if (cif->abi != FFI_SYSV) |
| 621 |
return FFI_BAD_ABI; |
| 622 |
|
| 623 |
#if FFI_EXEC_TRAMPOLINE_TABLE |
| 624 |
void **config = (void **)((uint8_t *)codeloc - PAGE_MAX_SIZE); |
| 625 |
config[0] = closure; |
| 626 |
config[1] = closure_func; |
| 627 |
#else |
| 628 |
|
| 629 |
# if defined(FFI_EXEC_STATIC_TRAMP) |
| 630 |
if (ffi_tramp_is_present(closure)) |
| 631 |
{ |
| 632 |
/* Initialize the static trampoline's parameters. */ |
| 633 |
if (closure_func == ffi_closure_SYSV) |
| 634 |
closure_func = ffi_closure_SYSV_alt; |
| 635 |
else |
| 636 |
closure_func = ffi_closure_VFP_alt; |
| 637 |
ffi_tramp_set_parms (closure->ftramp, closure_func, closure); |
| 638 |
goto out; |
| 639 |
} |
| 640 |
# endif |
| 641 |
|
| 642 |
/* Initialize the dynamic trampoline. */ |
| 643 |
# ifndef _WIN32 |
| 644 |
memcpy(closure->tramp, ffi_arm_trampoline, 8); |
| 645 |
# else |
| 646 |
// cast away function type so MSVC doesn't set the lower bit of the function pointer |
| 647 |
memcpy(closure->tramp, (void*)((uintptr_t)ffi_arm_trampoline & 0xFFFFFFFE), FFI_TRAMPOLINE_CLOSURE_OFFSET); |
| 648 |
# endif |
| 649 |
|
| 650 |
# if defined(__QNX__) |
| 651 |
msync (closure->tramp, 8, MS_INVALIDATE_ICACHE); /* clear data map */ |
| 652 |
msync (codeloc, 8, MS_INVALIDATE_ICACHE); /* clear insn map */ |
| 653 |
# elif defined(_WIN32) |
| 654 |
FlushInstructionCache(GetCurrentProcess(), closure->tramp, FFI_TRAMPOLINE_SIZE); |
| 655 |
# else |
| 656 |
__clear_cache(closure->tramp, closure->tramp + 8); /* clear data map */ |
| 657 |
__clear_cache(codeloc, codeloc + 8); /* clear insn map */ |
| 658 |
# endif |
| 659 |
# ifdef _WIN32 |
| 660 |
*(void(**)(void))(closure->tramp + FFI_TRAMPOLINE_CLOSURE_FUNCTION) = closure_func; |
| 661 |
# else |
| 662 |
*(void (**)(void))(closure->tramp + 8) = closure_func; |
| 663 |
# endif |
| 664 |
# if defined(FFI_EXEC_STATIC_TRAMP) |
| 665 |
out: |
| 666 |
# endif |
| 667 |
#endif |
| 668 |
|
| 669 |
closure->cif = cif; |
| 670 |
closure->fun = fun; |
| 671 |
closure->user_data = user_data; |
| 672 |
|
| 673 |
return FFI_OK; |
| 674 |
} |
| 675 |
|
| 676 |
#ifdef FFI_GO_CLOSURES |
| 677 |
ffi_status |
| 678 |
ffi_prep_go_closure (ffi_go_closure *closure, ffi_cif *cif, |
| 679 |
void (*fun) (ffi_cif *, void *, void **, void *)) |
| 680 |
{ |
| 681 |
void (*closure_func) (void) = ffi_go_closure_SYSV; |
| 682 |
|
| 683 |
if (cif->abi == FFI_VFP) |
| 684 |
{ |
| 685 |
/* We only need take the vfp path if there are vfp arguments. */ |
| 686 |
if (cif->vfp_used) |
| 687 |
closure_func = ffi_go_closure_VFP; |
| 688 |
} |
| 689 |
else if (cif->abi != FFI_SYSV) |
| 690 |
return FFI_BAD_ABI; |
| 691 |
|
| 692 |
closure->tramp = closure_func; |
| 693 |
closure->cif = cif; |
| 694 |
closure->fun = fun; |
| 695 |
|
| 696 |
return FFI_OK; |
| 697 |
} |
| 698 |
#endif |
| 699 |
|
| 700 |
#endif /* FFI_CLOSURES */ |
| 701 |
|
| 702 |
/* Below are routines for VFP hard-float support. */ |
| 703 |
|
| 704 |
/* A subroutine of vfp_type_p. Given a structure type, return the type code |
| 705 |
of the first non-structure element. Recurse for structure elements. |
| 706 |
Return -1 if the structure is in fact empty, i.e. no nested elements. */ |
| 707 |
|
| 708 |
static int |
| 709 |
is_hfa0 (const ffi_type *ty) |
| 710 |
{ |
| 711 |
ffi_type **elements = ty->elements; |
| 712 |
int i, ret = -1; |
| 713 |
|
| 714 |
if (elements != NULL) |
| 715 |
for (i = 0; elements[i]; ++i) |
| 716 |
{ |
| 717 |
ret = elements[i]->type; |
| 718 |
if (ret == FFI_TYPE_STRUCT || ret == FFI_TYPE_COMPLEX) |
| 719 |
{ |
| 720 |
ret = is_hfa0 (elements[i]); |
| 721 |
if (ret < 0) |
| 722 |
continue; |
| 723 |
} |
| 724 |
break; |
| 725 |
} |
| 726 |
|
| 727 |
return ret; |
| 728 |
} |
| 729 |
|
| 730 |
/* A subroutine of vfp_type_p. Given a structure type, return true if all |
| 731 |
of the non-structure elements are the same as CANDIDATE. */ |
| 732 |
|
| 733 |
static int |
| 734 |
is_hfa1 (const ffi_type *ty, int candidate) |
| 735 |
{ |
| 736 |
ffi_type **elements = ty->elements; |
| 737 |
int i; |
| 738 |
|
| 739 |
if (elements != NULL) |
| 740 |
for (i = 0; elements[i]; ++i) |
| 741 |
{ |
| 742 |
int t = elements[i]->type; |
| 743 |
if (t == FFI_TYPE_STRUCT || t == FFI_TYPE_COMPLEX) |
| 744 |
{ |
| 745 |
if (!is_hfa1 (elements[i], candidate)) |
| 746 |
return 0; |
| 747 |
} |
| 748 |
else if (t != candidate) |
| 749 |
return 0; |
| 750 |
} |
| 751 |
|
| 752 |
return 1; |
| 753 |
} |
| 754 |
|
| 755 |
/* Determine if TY is an homogenous floating point aggregate (HFA). |
| 756 |
That is, a structure consisting of 1 to 4 members of all the same type, |
| 757 |
where that type is a floating point scalar. |
| 758 |
|
| 759 |
Returns non-zero iff TY is an HFA. The result is an encoded value where |
| 760 |
bits 0-7 contain the type code, and bits 8-10 contain the element count. */ |
| 761 |
|
| 762 |
static int |
| 763 |
vfp_type_p (const ffi_type *ty) |
| 764 |
{ |
| 765 |
ffi_type **elements; |
| 766 |
int candidate, i; |
| 767 |
size_t size, ele_count; |
| 768 |
|
| 769 |
/* Quickest tests first. */ |
| 770 |
candidate = ty->type; |
| 771 |
switch (ty->type) |
| 772 |
{ |
| 773 |
default: |
| 774 |
return 0; |
| 775 |
case FFI_TYPE_FLOAT: |
| 776 |
case FFI_TYPE_DOUBLE: |
| 777 |
ele_count = 1; |
| 778 |
goto done; |
| 779 |
case FFI_TYPE_COMPLEX: |
| 780 |
candidate = ty->elements[0]->type; |
| 781 |
if (candidate != FFI_TYPE_FLOAT && candidate != FFI_TYPE_DOUBLE) |
| 782 |
return 0; |
| 783 |
ele_count = 2; |
| 784 |
goto done; |
| 785 |
case FFI_TYPE_STRUCT: |
| 786 |
break; |
| 787 |
} |
| 788 |
|
| 789 |
/* No HFA types are smaller than 4 bytes, or larger than 32 bytes. */ |
| 790 |
size = ty->size; |
| 791 |
if (size < 4 || size > 32) |
| 792 |
return 0; |
| 793 |
|
| 794 |
/* Find the type of the first non-structure member. */ |
| 795 |
elements = ty->elements; |
| 796 |
candidate = elements[0]->type; |
| 797 |
if (candidate == FFI_TYPE_STRUCT || candidate == FFI_TYPE_COMPLEX) |
| 798 |
{ |
| 799 |
for (i = 0; ; ++i) |
| 800 |
{ |
| 801 |
candidate = is_hfa0 (elements[i]); |
| 802 |
if (candidate >= 0) |
| 803 |
break; |
| 804 |
} |
| 805 |
} |
| 806 |
|
| 807 |
/* If the first member is not a floating point type, it's not an HFA. |
| 808 |
Also quickly re-check the size of the structure. */ |
| 809 |
switch (candidate) |
| 810 |
{ |
| 811 |
case FFI_TYPE_FLOAT: |
| 812 |
ele_count = size / sizeof(float); |
| 813 |
if (size != ele_count * sizeof(float)) |
| 814 |
return 0; |
| 815 |
break; |
| 816 |
case FFI_TYPE_DOUBLE: |
| 817 |
ele_count = size / sizeof(double); |
| 818 |
if (size != ele_count * sizeof(double)) |
| 819 |
return 0; |
| 820 |
break; |
| 821 |
default: |
| 822 |
return 0; |
| 823 |
} |
| 824 |
if (ele_count > 4) |
| 825 |
return 0; |
| 826 |
|
| 827 |
/* Finally, make sure that all scalar elements are the same type. */ |
| 828 |
for (i = 0; elements[i]; ++i) |
| 829 |
{ |
| 830 |
int t = elements[i]->type; |
| 831 |
if (t == FFI_TYPE_STRUCT || t == FFI_TYPE_COMPLEX) |
| 832 |
{ |
| 833 |
if (!is_hfa1 (elements[i], candidate)) |
| 834 |
return 0; |
| 835 |
} |
| 836 |
else if (t != candidate) |
| 837 |
return 0; |
| 838 |
} |
| 839 |
|
| 840 |
/* All tests succeeded. Encode the result. */ |
| 841 |
done: |
| 842 |
return (ele_count << 8) | candidate; |
| 843 |
} |
| 844 |
|
| 845 |
static int |
| 846 |
place_vfp_arg (ffi_cif *cif, int h) |
| 847 |
{ |
| 848 |
unsigned short reg = cif->vfp_reg_free; |
| 849 |
int align = 1, nregs = h >> 8; |
| 850 |
|
| 851 |
if ((h & 0xff) == FFI_TYPE_DOUBLE) |
| 852 |
align = 2, nregs *= 2; |
| 853 |
|
| 854 |
/* Align register number. */ |
| 855 |
if ((reg & 1) && align == 2) |
| 856 |
reg++; |
| 857 |
|
| 858 |
while (reg + nregs <= 16) |
| 859 |
{ |
| 860 |
int s, new_used = 0; |
| 861 |
for (s = reg; s < reg + nregs; s++) |
| 862 |
{ |
| 863 |
new_used |= (1 << s); |
| 864 |
if (cif->vfp_used & (1 << s)) |
| 865 |
{ |
| 866 |
reg += align; |
| 867 |
goto next_reg; |
| 868 |
} |
| 869 |
} |
| 870 |
/* Found regs to allocate. */ |
| 871 |
cif->vfp_used |= new_used; |
| 872 |
cif->vfp_args[cif->vfp_nargs++] = (signed char)reg; |
| 873 |
|
| 874 |
/* Update vfp_reg_free. */ |
| 875 |
if (cif->vfp_used & (1 << cif->vfp_reg_free)) |
| 876 |
{ |
| 877 |
reg += nregs; |
| 878 |
while (cif->vfp_used & (1 << reg)) |
| 879 |
reg += 1; |
| 880 |
cif->vfp_reg_free = reg; |
| 881 |
} |
| 882 |
return 0; |
| 883 |
next_reg:; |
| 884 |
} |
| 885 |
// done, mark all regs as used |
| 886 |
cif->vfp_reg_free = 16; |
| 887 |
cif->vfp_used = 0xFFFF; |
| 888 |
return 1; |
| 889 |
} |
| 890 |
|
| 891 |
static void |
| 892 |
layout_vfp_args (ffi_cif * cif) |
| 893 |
{ |
| 894 |
unsigned int i; |
| 895 |
/* Init VFP fields */ |
| 896 |
cif->vfp_used = 0; |
| 897 |
cif->vfp_nargs = 0; |
| 898 |
cif->vfp_reg_free = 0; |
| 899 |
memset (cif->vfp_args, -1, 16); /* Init to -1. */ |
| 900 |
|
| 901 |
for (i = 0; i < cif->nargs; i++) |
| 902 |
{ |
| 903 |
int h = vfp_type_p (cif->arg_types[i]); |
| 904 |
if (h && place_vfp_arg (cif, h) == 1) |
| 905 |
break; |
| 906 |
} |
| 907 |
} |
| 908 |
|
| 909 |
#if defined(FFI_EXEC_STATIC_TRAMP) |
| 910 |
void * |
| 911 |
ffi_tramp_arch (size_t *tramp_size, size_t *map_size) |
| 912 |
{ |
| 913 |
extern void *trampoline_code_table; |
| 914 |
|
| 915 |
*tramp_size = ARM_TRAMP_SIZE; |
| 916 |
*map_size = ARM_TRAMP_MAP_SIZE; |
| 917 |
return &trampoline_code_table; |
| 918 |
} |
| 919 |
#endif |
| 920 |
|
| 921 |
#endif /* __arm__ or _M_ARM */ |