1 | /* -----------------------------------------------------------------------
|
---|
2 | ffi.c - Copyright (c) 2002 Bo Thorsen <bo@suse.de>
|
---|
3 |
|
---|
4 | x86-64 Foreign Function Interface
|
---|
5 |
|
---|
6 | Permission is hereby granted, free of charge, to any person obtaining
|
---|
7 | a copy of this software and associated documentation files (the
|
---|
8 | ``Software''), to deal in the Software without restriction, including
|
---|
9 | without limitation the rights to use, copy, modify, merge, publish,
|
---|
10 | distribute, sublicense, and/or sell copies of the Software, and to
|
---|
11 | permit persons to whom the Software is furnished to do so, subject to
|
---|
12 | the following conditions:
|
---|
13 |
|
---|
14 | The above copyright notice and this permission notice shall be included
|
---|
15 | in all copies or substantial portions of the Software.
|
---|
16 |
|
---|
17 | THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, EXPRESS
|
---|
18 | OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
|
---|
19 | MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
|
---|
20 | IN NO EVENT SHALL CYGNUS SOLUTIONS BE LIABLE FOR ANY CLAIM, DAMAGES OR
|
---|
21 | OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
|
---|
22 | ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
|
---|
23 | OTHER DEALINGS IN THE SOFTWARE.
|
---|
24 | ----------------------------------------------------------------------- */
|
---|
25 |
|
---|
26 | #include <ffi.h>
|
---|
27 | #include <ffi_common.h>
|
---|
28 |
|
---|
29 | #include <stdlib.h>
|
---|
30 | #include <stdarg.h>
|
---|
31 |
|
---|
32 | /* ffi_prep_args is called by the assembly routine once stack space
|
---|
33 | has been allocated for the function's arguments */
|
---|
34 |
|
---|
35 | #ifdef __x86_64__
|
---|
36 |
|
---|
37 | #define MAX_GPR_REGS 6
|
---|
38 | #define MAX_SSE_REGS 8
|
---|
39 | typedef struct
|
---|
40 | {
|
---|
41 | /* Registers for argument passing. */
|
---|
42 | long gpr[MAX_GPR_REGS];
|
---|
43 | __int128_t sse[MAX_SSE_REGS];
|
---|
44 |
|
---|
45 | /* Stack space for arguments. */
|
---|
46 | char argspace[0];
|
---|
47 | } stackLayout;
|
---|
48 |
|
---|
49 | /* All reference to register classes here is identical to the code in
|
---|
50 | gcc/config/i386/i386.c. Do *not* change one without the other. */
|
---|
51 |
|
---|
52 | /* Register class used for passing given 64bit part of the argument.
|
---|
53 | These represent classes as documented by the PS ABI, with the exception
|
---|
54 | of SSESF, SSEDF classes, that are basically SSE class, just gcc will
|
---|
55 | use SF or DFmode move instead of DImode to avoid reformating penalties.
|
---|
56 |
|
---|
57 | Similary we play games with INTEGERSI_CLASS to use cheaper SImode moves
|
---|
58 | whenever possible (upper half does contain padding).
|
---|
59 | */
|
---|
60 | enum x86_64_reg_class
|
---|
61 | {
|
---|
62 | X86_64_NO_CLASS,
|
---|
63 | X86_64_INTEGER_CLASS,
|
---|
64 | X86_64_INTEGERSI_CLASS,
|
---|
65 | X86_64_SSE_CLASS,
|
---|
66 | X86_64_SSESF_CLASS,
|
---|
67 | X86_64_SSEDF_CLASS,
|
---|
68 | X86_64_SSEUP_CLASS,
|
---|
69 | X86_64_X87_CLASS,
|
---|
70 | X86_64_X87UP_CLASS,
|
---|
71 | X86_64_MEMORY_CLASS
|
---|
72 | };
|
---|
73 |
|
---|
74 | #define MAX_CLASSES 4
|
---|
75 |
|
---|
76 | /* x86-64 register passing implementation. See x86-64 ABI for details. Goal
|
---|
77 | of this code is to classify each 8bytes of incoming argument by the register
|
---|
78 | class and assign registers accordingly. */
|
---|
79 |
|
---|
80 | /* Return the union class of CLASS1 and CLASS2.
|
---|
81 | See the x86-64 PS ABI for details. */
|
---|
82 |
|
---|
83 | static enum x86_64_reg_class
|
---|
84 | merge_classes (enum x86_64_reg_class class1, enum x86_64_reg_class class2)
|
---|
85 | {
|
---|
86 | /* Rule #1: If both classes are equal, this is the resulting class. */
|
---|
87 | if (class1 == class2)
|
---|
88 | return class1;
|
---|
89 |
|
---|
90 | /* Rule #2: If one of the classes is NO_CLASS, the resulting class is
|
---|
91 | the other class. */
|
---|
92 | if (class1 == X86_64_NO_CLASS)
|
---|
93 | return class2;
|
---|
94 | if (class2 == X86_64_NO_CLASS)
|
---|
95 | return class1;
|
---|
96 |
|
---|
97 | /* Rule #3: If one of the classes is MEMORY, the result is MEMORY. */
|
---|
98 | if (class1 == X86_64_MEMORY_CLASS || class2 == X86_64_MEMORY_CLASS)
|
---|
99 | return X86_64_MEMORY_CLASS;
|
---|
100 |
|
---|
101 | /* Rule #4: If one of the classes is INTEGER, the result is INTEGER. */
|
---|
102 | if ((class1 == X86_64_INTEGERSI_CLASS && class2 == X86_64_SSESF_CLASS)
|
---|
103 | || (class2 == X86_64_INTEGERSI_CLASS && class1 == X86_64_SSESF_CLASS))
|
---|
104 | return X86_64_INTEGERSI_CLASS;
|
---|
105 | if (class1 == X86_64_INTEGER_CLASS || class1 == X86_64_INTEGERSI_CLASS
|
---|
106 | || class2 == X86_64_INTEGER_CLASS || class2 == X86_64_INTEGERSI_CLASS)
|
---|
107 | return X86_64_INTEGER_CLASS;
|
---|
108 |
|
---|
109 | /* Rule #5: If one of the classes is X87 or X87UP class, MEMORY is used. */
|
---|
110 | if (class1 == X86_64_X87_CLASS || class1 == X86_64_X87UP_CLASS
|
---|
111 | || class2 == X86_64_X87_CLASS || class2 == X86_64_X87UP_CLASS)
|
---|
112 | return X86_64_MEMORY_CLASS;
|
---|
113 |
|
---|
114 | /* Rule #6: Otherwise class SSE is used. */
|
---|
115 | return X86_64_SSE_CLASS;
|
---|
116 | }
|
---|
117 |
|
---|
118 | /* Classify the argument of type TYPE and mode MODE.
|
---|
119 | CLASSES will be filled by the register class used to pass each word
|
---|
120 | of the operand. The number of words is returned. In case the parameter
|
---|
121 | should be passed in memory, 0 is returned. As a special case for zero
|
---|
122 | sized containers, classes[0] will be NO_CLASS and 1 is returned.
|
---|
123 |
|
---|
124 | See the x86-64 PS ABI for details.
|
---|
125 | */
|
---|
126 | static int
|
---|
127 | classify_argument (ffi_type *type, enum x86_64_reg_class classes[],
|
---|
128 | int *byte_offset)
|
---|
129 | {
|
---|
130 | /* First, align to the right place. */
|
---|
131 | *byte_offset = ALIGN(*byte_offset, type->alignment);
|
---|
132 |
|
---|
133 | switch (type->type)
|
---|
134 | {
|
---|
135 | case FFI_TYPE_UINT8:
|
---|
136 | case FFI_TYPE_SINT8:
|
---|
137 | case FFI_TYPE_UINT16:
|
---|
138 | case FFI_TYPE_SINT16:
|
---|
139 | case FFI_TYPE_UINT32:
|
---|
140 | case FFI_TYPE_SINT32:
|
---|
141 | case FFI_TYPE_UINT64:
|
---|
142 | case FFI_TYPE_SINT64:
|
---|
143 | case FFI_TYPE_POINTER:
|
---|
144 | if (((*byte_offset) % 8 + type->size) <= 4)
|
---|
145 | classes[0] = X86_64_INTEGERSI_CLASS;
|
---|
146 | else
|
---|
147 | classes[0] = X86_64_INTEGER_CLASS;
|
---|
148 | return 1;
|
---|
149 | case FFI_TYPE_FLOAT:
|
---|
150 | if (((*byte_offset) % 8) == 0)
|
---|
151 | classes[0] = X86_64_SSESF_CLASS;
|
---|
152 | else
|
---|
153 | classes[0] = X86_64_SSE_CLASS;
|
---|
154 | return 1;
|
---|
155 | case FFI_TYPE_DOUBLE:
|
---|
156 | classes[0] = X86_64_SSEDF_CLASS;
|
---|
157 | return 1;
|
---|
158 | case FFI_TYPE_LONGDOUBLE:
|
---|
159 | classes[0] = X86_64_X87_CLASS;
|
---|
160 | classes[1] = X86_64_X87UP_CLASS;
|
---|
161 | return 2;
|
---|
162 | case FFI_TYPE_STRUCT:
|
---|
163 | {
|
---|
164 | const int UNITS_PER_WORD = 8;
|
---|
165 | int words = (type->size + UNITS_PER_WORD - 1) / UNITS_PER_WORD;
|
---|
166 | ffi_type **ptr;
|
---|
167 | int i;
|
---|
168 | enum x86_64_reg_class subclasses[MAX_CLASSES];
|
---|
169 |
|
---|
170 | /* If the struct is larger than 16 bytes, pass it on the stack. */
|
---|
171 | if (type->size > 16)
|
---|
172 | return 0;
|
---|
173 |
|
---|
174 | for (i = 0; i < words; i++)
|
---|
175 | classes[i] = X86_64_NO_CLASS;
|
---|
176 |
|
---|
177 | /* Merge the fields of structure. */
|
---|
178 | for (ptr=type->elements; (*ptr)!=NULL; ptr++)
|
---|
179 | {
|
---|
180 | int num;
|
---|
181 |
|
---|
182 | num = classify_argument (*ptr, subclasses, byte_offset);
|
---|
183 | if (num == 0)
|
---|
184 | return 0;
|
---|
185 | for (i = 0; i < num; i++)
|
---|
186 | {
|
---|
187 | int pos = *byte_offset / 8;
|
---|
188 | classes[i + pos] =
|
---|
189 | merge_classes (subclasses[i], classes[i + pos]);
|
---|
190 | }
|
---|
191 |
|
---|
192 | if ((*ptr)->type != FFI_TYPE_STRUCT)
|
---|
193 | *byte_offset += (*ptr)->size;
|
---|
194 | }
|
---|
195 |
|
---|
196 | /* Final merger cleanup. */
|
---|
197 | for (i = 0; i < words; i++)
|
---|
198 | {
|
---|
199 | /* If one class is MEMORY, everything should be passed in
|
---|
200 | memory. */
|
---|
201 | if (classes[i] == X86_64_MEMORY_CLASS)
|
---|
202 | return 0;
|
---|
203 |
|
---|
204 | /* The X86_64_SSEUP_CLASS should be always preceded by
|
---|
205 | X86_64_SSE_CLASS. */
|
---|
206 | if (classes[i] == X86_64_SSEUP_CLASS
|
---|
207 | && (i == 0 || classes[i - 1] != X86_64_SSE_CLASS))
|
---|
208 | classes[i] = X86_64_SSE_CLASS;
|
---|
209 |
|
---|
210 | /* X86_64_X87UP_CLASS should be preceded by X86_64_X87_CLASS. */
|
---|
211 | if (classes[i] == X86_64_X87UP_CLASS
|
---|
212 | && (i == 0 || classes[i - 1] != X86_64_X87_CLASS))
|
---|
213 | classes[i] = X86_64_SSE_CLASS;
|
---|
214 | }
|
---|
215 | return words;
|
---|
216 | }
|
---|
217 |
|
---|
218 | default:
|
---|
219 | FFI_ASSERT(0);
|
---|
220 | }
|
---|
221 | return 0; /* Never reached. */
|
---|
222 | }
|
---|
223 |
|
---|
224 | /* Examine the argument and return set number of register required in each
|
---|
225 | class. Return 0 iff parameter should be passed in memory. */
|
---|
226 | static int
|
---|
227 | examine_argument (ffi_type *type, int in_return, int *int_nregs,int *sse_nregs)
|
---|
228 | {
|
---|
229 | enum x86_64_reg_class class[MAX_CLASSES];
|
---|
230 | int offset = 0;
|
---|
231 | int n;
|
---|
232 |
|
---|
233 | n = classify_argument (type, class, &offset);
|
---|
234 |
|
---|
235 | if (n == 0)
|
---|
236 | return 0;
|
---|
237 |
|
---|
238 | *int_nregs = 0;
|
---|
239 | *sse_nregs = 0;
|
---|
240 | for (n--; n>=0; n--)
|
---|
241 | switch (class[n])
|
---|
242 | {
|
---|
243 | case X86_64_INTEGER_CLASS:
|
---|
244 | case X86_64_INTEGERSI_CLASS:
|
---|
245 | (*int_nregs)++;
|
---|
246 | break;
|
---|
247 | case X86_64_SSE_CLASS:
|
---|
248 | case X86_64_SSESF_CLASS:
|
---|
249 | case X86_64_SSEDF_CLASS:
|
---|
250 | (*sse_nregs)++;
|
---|
251 | break;
|
---|
252 | case X86_64_NO_CLASS:
|
---|
253 | case X86_64_SSEUP_CLASS:
|
---|
254 | break;
|
---|
255 | case X86_64_X87_CLASS:
|
---|
256 | case X86_64_X87UP_CLASS:
|
---|
257 | if (!in_return)
|
---|
258 | return 0;
|
---|
259 | break;
|
---|
260 | default:
|
---|
261 | abort ();
|
---|
262 | }
|
---|
263 | return 1;
|
---|
264 | }
|
---|
265 |
|
---|
266 | /* Functions to load floats and double to an SSE register placeholder. */
|
---|
267 | extern void float2sse (float, __int128_t *);
|
---|
268 | extern void double2sse (double, __int128_t *);
|
---|
269 | extern void floatfloat2sse (void *, __int128_t *);
|
---|
270 |
|
---|
271 | /* Functions to put the floats and doubles back. */
|
---|
272 | extern float sse2float (__int128_t *);
|
---|
273 | extern double sse2double (__int128_t *);
|
---|
274 | extern void sse2floatfloat(__int128_t *, void *);
|
---|
275 |
|
---|
276 | /*@-exportheader@*/
|
---|
277 | void
|
---|
278 | ffi_prep_args (stackLayout *stack, extended_cif *ecif)
|
---|
279 | /*@=exportheader@*/
|
---|
280 | {
|
---|
281 | int gprcount, ssecount, i, g, s;
|
---|
282 | void **p_argv;
|
---|
283 | void *argp = &stack->argspace;
|
---|
284 | ffi_type **p_arg;
|
---|
285 |
|
---|
286 | /* First check if the return value should be passed in memory. If so,
|
---|
287 | pass the pointer as the first argument. */
|
---|
288 | gprcount = ssecount = 0;
|
---|
289 | if (ecif->cif->rtype->type != FFI_TYPE_VOID
|
---|
290 | && examine_argument (ecif->cif->rtype, 1, &g, &s) == 0)
|
---|
291 | (void *)stack->gpr[gprcount++] = ecif->rvalue;
|
---|
292 |
|
---|
293 | for (i=ecif->cif->nargs, p_arg=ecif->cif->arg_types, p_argv = ecif->avalue;
|
---|
294 | i!=0; i--, p_arg++, p_argv++)
|
---|
295 | {
|
---|
296 | int in_register = 0;
|
---|
297 |
|
---|
298 | switch ((*p_arg)->type)
|
---|
299 | {
|
---|
300 | case FFI_TYPE_SINT8:
|
---|
301 | case FFI_TYPE_SINT16:
|
---|
302 | case FFI_TYPE_SINT32:
|
---|
303 | case FFI_TYPE_SINT64:
|
---|
304 | case FFI_TYPE_UINT8:
|
---|
305 | case FFI_TYPE_UINT16:
|
---|
306 | case FFI_TYPE_UINT32:
|
---|
307 | case FFI_TYPE_UINT64:
|
---|
308 | case FFI_TYPE_POINTER:
|
---|
309 | if (gprcount < MAX_GPR_REGS)
|
---|
310 | {
|
---|
311 | stack->gpr[gprcount] = 0;
|
---|
312 | stack->gpr[gprcount++] = *(long long *)(*p_argv);
|
---|
313 | in_register = 1;
|
---|
314 | }
|
---|
315 | break;
|
---|
316 |
|
---|
317 | case FFI_TYPE_FLOAT:
|
---|
318 | if (ssecount < MAX_SSE_REGS)
|
---|
319 | {
|
---|
320 | float2sse (*(float *)(*p_argv), &stack->sse[ssecount++]);
|
---|
321 | in_register = 1;
|
---|
322 | }
|
---|
323 | break;
|
---|
324 |
|
---|
325 | case FFI_TYPE_DOUBLE:
|
---|
326 | if (ssecount < MAX_SSE_REGS)
|
---|
327 | {
|
---|
328 | double2sse (*(double *)(*p_argv), &stack->sse[ssecount++]);
|
---|
329 | in_register = 1;
|
---|
330 | }
|
---|
331 | break;
|
---|
332 | }
|
---|
333 |
|
---|
334 | if (in_register)
|
---|
335 | continue;
|
---|
336 |
|
---|
337 | /* Either all places in registers where filled, or this is a
|
---|
338 | type that potentially goes into a memory slot. */
|
---|
339 | if (examine_argument (*p_arg, 0, &g, &s) == 0
|
---|
340 | || gprcount + g > MAX_GPR_REGS || ssecount + s > MAX_SSE_REGS)
|
---|
341 | {
|
---|
342 | /* Pass this argument in memory. */
|
---|
343 | argp = (void *)ALIGN(argp, (*p_arg)->alignment);
|
---|
344 | memcpy (argp, *p_argv, (*p_arg)->size);
|
---|
345 | argp += (*p_arg)->size;
|
---|
346 | }
|
---|
347 | else
|
---|
348 | {
|
---|
349 | /* All easy cases are eliminated. Now fire the big guns. */
|
---|
350 |
|
---|
351 | enum x86_64_reg_class classes[MAX_CLASSES];
|
---|
352 | int offset = 0, j, num;
|
---|
353 | void *a;
|
---|
354 |
|
---|
355 | num = classify_argument (*p_arg, classes, &offset);
|
---|
356 | for (j=0, a=*p_argv; j<num; j++, a+=8)
|
---|
357 | {
|
---|
358 | switch (classes[j])
|
---|
359 | {
|
---|
360 | case X86_64_INTEGER_CLASS:
|
---|
361 | case X86_64_INTEGERSI_CLASS:
|
---|
362 | stack->gpr[gprcount++] = *(long long *)a;
|
---|
363 | break;
|
---|
364 | case X86_64_SSE_CLASS:
|
---|
365 | floatfloat2sse (a, &stack->sse[ssecount++]);
|
---|
366 | break;
|
---|
367 | case X86_64_SSESF_CLASS:
|
---|
368 | float2sse (*(float *)a, &stack->sse[ssecount++]);
|
---|
369 | break;
|
---|
370 | case X86_64_SSEDF_CLASS:
|
---|
371 | double2sse (*(double *)a, &stack->sse[ssecount++]);
|
---|
372 | break;
|
---|
373 | default:
|
---|
374 | abort();
|
---|
375 | }
|
---|
376 | }
|
---|
377 | }
|
---|
378 | }
|
---|
379 | }
|
---|
380 |
|
---|
381 | /* Perform machine dependent cif processing. */
|
---|
382 | ffi_status
|
---|
383 | ffi_prep_cif_machdep (ffi_cif *cif)
|
---|
384 | {
|
---|
385 | int gprcount, ssecount, i, g, s;
|
---|
386 |
|
---|
387 | gprcount = ssecount = 0;
|
---|
388 |
|
---|
389 | /* Reset the byte count. We handle this size estimation here. */
|
---|
390 | cif->bytes = 0;
|
---|
391 |
|
---|
392 | /* If the return value should be passed in memory, pass the pointer
|
---|
393 | as the first argument. The actual memory isn't allocated here. */
|
---|
394 | if (cif->rtype->type != FFI_TYPE_VOID
|
---|
395 | && examine_argument (cif->rtype, 1, &g, &s) == 0)
|
---|
396 | gprcount = 1;
|
---|
397 |
|
---|
398 | /* Go over all arguments and determine the way they should be passed.
|
---|
399 | If it's in a register and there is space for it, let that be so. If
|
---|
400 | not, add it's size to the stack byte count. */
|
---|
401 | for (i=0; i<cif->nargs; i++)
|
---|
402 | {
|
---|
403 | if (examine_argument (cif->arg_types[i], 0, &g, &s) == 0
|
---|
404 | || gprcount + g > MAX_GPR_REGS || ssecount + s > MAX_SSE_REGS)
|
---|
405 | {
|
---|
406 | /* This is passed in memory. First align to the basic type. */
|
---|
407 | cif->bytes = ALIGN(cif->bytes, cif->arg_types[i]->alignment);
|
---|
408 |
|
---|
409 | /* Stack arguments are *always* at least 8 byte aligned. */
|
---|
410 | cif->bytes = ALIGN(cif->bytes, 8);
|
---|
411 |
|
---|
412 | /* Now add the size of this argument. */
|
---|
413 | cif->bytes += cif->arg_types[i]->size;
|
---|
414 | }
|
---|
415 | else
|
---|
416 | {
|
---|
417 | gprcount += g;
|
---|
418 | ssecount += s;
|
---|
419 | }
|
---|
420 | }
|
---|
421 |
|
---|
422 | /* Set the flag for the closures return. */
|
---|
423 | switch (cif->rtype->type)
|
---|
424 | {
|
---|
425 | case FFI_TYPE_VOID:
|
---|
426 | case FFI_TYPE_STRUCT:
|
---|
427 | case FFI_TYPE_SINT64:
|
---|
428 | case FFI_TYPE_FLOAT:
|
---|
429 | case FFI_TYPE_DOUBLE:
|
---|
430 | case FFI_TYPE_LONGDOUBLE:
|
---|
431 | cif->flags = (unsigned) cif->rtype->type;
|
---|
432 | break;
|
---|
433 |
|
---|
434 | case FFI_TYPE_UINT64:
|
---|
435 | cif->flags = FFI_TYPE_SINT64;
|
---|
436 | break;
|
---|
437 |
|
---|
438 | default:
|
---|
439 | cif->flags = FFI_TYPE_INT;
|
---|
440 | break;
|
---|
441 | }
|
---|
442 |
|
---|
443 | return FFI_OK;
|
---|
444 | }
|
---|
445 |
|
---|
446 | typedef struct
|
---|
447 | {
|
---|
448 | long gpr[2];
|
---|
449 | __int128_t sse[2];
|
---|
450 | long double st0;
|
---|
451 | } return_value;
|
---|
452 |
|
---|
453 | void
|
---|
454 | ffi_fill_return_value (return_value *rv, extended_cif *ecif)
|
---|
455 | {
|
---|
456 | enum x86_64_reg_class classes[MAX_CLASSES];
|
---|
457 | int i = 0, num;
|
---|
458 | long *gpr = rv->gpr;
|
---|
459 | __int128_t *sse = rv->sse;
|
---|
460 | signed char sc;
|
---|
461 | signed short ss;
|
---|
462 |
|
---|
463 | /* This is needed because of the way x86-64 handles signed short
|
---|
464 | integers. */
|
---|
465 | switch (ecif->cif->rtype->type)
|
---|
466 | {
|
---|
467 | case FFI_TYPE_SINT8:
|
---|
468 | sc = *(signed char *)gpr;
|
---|
469 | *(long long *)ecif->rvalue = (long long)sc;
|
---|
470 | return;
|
---|
471 | case FFI_TYPE_SINT16:
|
---|
472 | ss = *(signed short *)gpr;
|
---|
473 | *(long long *)ecif->rvalue = (long long)ss;
|
---|
474 | return;
|
---|
475 | default:
|
---|
476 | /* Just continue. */
|
---|
477 | ;
|
---|
478 | }
|
---|
479 |
|
---|
480 | num = classify_argument (ecif->cif->rtype, classes, &i);
|
---|
481 |
|
---|
482 | if (num == 0)
|
---|
483 | /* Return in memory. */
|
---|
484 | ecif->rvalue = (void *) rv->gpr[0];
|
---|
485 | else if (num == 2 && classes[0] == X86_64_X87_CLASS &&
|
---|
486 | classes[1] == X86_64_X87UP_CLASS)
|
---|
487 | /* This is a long double (this is easiest to handle this way instead
|
---|
488 | of an eightbyte at a time as in the loop below. */
|
---|
489 | *((long double *)ecif->rvalue) = rv->st0;
|
---|
490 | else
|
---|
491 | {
|
---|
492 | void *a;
|
---|
493 |
|
---|
494 | for (i=0, a=ecif->rvalue; i<num; i++, a+=8)
|
---|
495 | {
|
---|
496 | switch (classes[i])
|
---|
497 | {
|
---|
498 | case X86_64_INTEGER_CLASS:
|
---|
499 | case X86_64_INTEGERSI_CLASS:
|
---|
500 | *(long long *)a = *gpr;
|
---|
501 | gpr++;
|
---|
502 | break;
|
---|
503 | case X86_64_SSE_CLASS:
|
---|
504 | sse2floatfloat (sse++, a);
|
---|
505 | break;
|
---|
506 | case X86_64_SSESF_CLASS:
|
---|
507 | *(float *)a = sse2float (sse++);
|
---|
508 | break;
|
---|
509 | case X86_64_SSEDF_CLASS:
|
---|
510 | *(double *)a = sse2double (sse++);
|
---|
511 | break;
|
---|
512 | default:
|
---|
513 | abort();
|
---|
514 | }
|
---|
515 | }
|
---|
516 | }
|
---|
517 | }
|
---|
518 |
|
---|
519 | /*@-declundef@*/
|
---|
520 | /*@-exportheader@*/
|
---|
521 | extern void ffi_call_UNIX64(void (*)(stackLayout *, extended_cif *),
|
---|
522 | void (*) (return_value *, extended_cif *),
|
---|
523 | /*@out@*/ extended_cif *,
|
---|
524 | unsigned, /*@out@*/ unsigned *, void (*fn)());
|
---|
525 | /*@=declundef@*/
|
---|
526 | /*@=exportheader@*/
|
---|
527 |
|
---|
528 | void ffi_call(/*@dependent@*/ ffi_cif *cif,
|
---|
529 | void (*fn)(),
|
---|
530 | /*@out@*/ void *rvalue,
|
---|
531 | /*@dependent@*/ void **avalue)
|
---|
532 | {
|
---|
533 | extended_cif ecif;
|
---|
534 | int dummy;
|
---|
535 |
|
---|
536 | ecif.cif = cif;
|
---|
537 | ecif.avalue = avalue;
|
---|
538 |
|
---|
539 | /* If the return value is a struct and we don't have a return */
|
---|
540 | /* value address then we need to make one */
|
---|
541 |
|
---|
542 | if ((rvalue == NULL) &&
|
---|
543 | (examine_argument (cif->rtype, 1, &dummy, &dummy) == 0))
|
---|
544 | {
|
---|
545 | /*@-sysunrecog@*/
|
---|
546 | ecif.rvalue = alloca(cif->rtype->size);
|
---|
547 | /*@=sysunrecog@*/
|
---|
548 | }
|
---|
549 | else
|
---|
550 | ecif.rvalue = rvalue;
|
---|
551 |
|
---|
552 | /* Stack must always be 16byte aligned. Make it so. */
|
---|
553 | cif->bytes = ALIGN(cif->bytes, 16);
|
---|
554 |
|
---|
555 | switch (cif->abi)
|
---|
556 | {
|
---|
557 | case FFI_SYSV:
|
---|
558 | /* Calling 32bit code from 64bit is not possible */
|
---|
559 | FFI_ASSERT(0);
|
---|
560 | break;
|
---|
561 |
|
---|
562 | case FFI_UNIX64:
|
---|
563 | /*@-usedef@*/
|
---|
564 | ffi_call_UNIX64 (ffi_prep_args, ffi_fill_return_value, &ecif,
|
---|
565 | cif->bytes, ecif.rvalue, fn);
|
---|
566 | /*@=usedef@*/
|
---|
567 | break;
|
---|
568 |
|
---|
569 | default:
|
---|
570 | FFI_ASSERT(0);
|
---|
571 | break;
|
---|
572 | }
|
---|
573 | }
|
---|
574 |
|
---|
575 | extern void ffi_closure_UNIX64(void);
|
---|
576 |
|
---|
577 | ffi_status
|
---|
578 | ffi_prep_closure (ffi_closure* closure,
|
---|
579 | ffi_cif* cif,
|
---|
580 | void (*fun)(ffi_cif*, void*, void**, void*),
|
---|
581 | void *user_data)
|
---|
582 | {
|
---|
583 | volatile unsigned short *tramp;
|
---|
584 |
|
---|
585 | /* FFI_ASSERT (cif->abi == FFI_OSF); */
|
---|
586 |
|
---|
587 | tramp = (volatile unsigned short *) &closure->tramp[0];
|
---|
588 | tramp[0] = 0xbb49; /* mov <code>, %r11 */
|
---|
589 | tramp[5] = 0xba49; /* mov <data>, %r10 */
|
---|
590 | tramp[10] = 0xff49; /* jmp *%r11 */
|
---|
591 | tramp[11] = 0x00e3;
|
---|
592 | *(void * volatile *) &tramp[1] = ffi_closure_UNIX64;
|
---|
593 | *(void * volatile *) &tramp[6] = closure;
|
---|
594 |
|
---|
595 | closure->cif = cif;
|
---|
596 | closure->fun = fun;
|
---|
597 | closure->user_data = user_data;
|
---|
598 |
|
---|
599 | return FFI_OK;
|
---|
600 | }
|
---|
601 |
|
---|
602 | int
|
---|
603 | ffi_closure_UNIX64_inner(ffi_closure *closure, va_list l, void *rp)
|
---|
604 | {
|
---|
605 | ffi_cif *cif;
|
---|
606 | void **avalue;
|
---|
607 | ffi_type **arg_types;
|
---|
608 | long i, avn, argn;
|
---|
609 |
|
---|
610 | cif = closure->cif;
|
---|
611 | avalue = alloca(cif->nargs * sizeof(void *));
|
---|
612 |
|
---|
613 | argn = 0;
|
---|
614 |
|
---|
615 | i = 0;
|
---|
616 | avn = cif->nargs;
|
---|
617 | arg_types = cif->arg_types;
|
---|
618 |
|
---|
619 | /* Grab the addresses of the arguments from the stack frame. */
|
---|
620 | while (i < avn)
|
---|
621 | {
|
---|
622 | switch (arg_types[i]->type)
|
---|
623 | {
|
---|
624 | case FFI_TYPE_SINT8:
|
---|
625 | case FFI_TYPE_UINT8:
|
---|
626 | case FFI_TYPE_SINT16:
|
---|
627 | case FFI_TYPE_UINT16:
|
---|
628 | case FFI_TYPE_SINT32:
|
---|
629 | case FFI_TYPE_UINT32:
|
---|
630 | case FFI_TYPE_SINT64:
|
---|
631 | case FFI_TYPE_UINT64:
|
---|
632 | case FFI_TYPE_POINTER:
|
---|
633 | {
|
---|
634 | if (l->gp_offset > 48-8)
|
---|
635 | {
|
---|
636 | avalue[i] = l->overflow_arg_area;
|
---|
637 | l->overflow_arg_area = (char *)l->overflow_arg_area + 8;
|
---|
638 | }
|
---|
639 | else
|
---|
640 | {
|
---|
641 | avalue[i] = (char *)l->reg_save_area + l->gp_offset;
|
---|
642 | l->gp_offset += 8;
|
---|
643 | }
|
---|
644 | }
|
---|
645 | break;
|
---|
646 |
|
---|
647 | case FFI_TYPE_STRUCT:
|
---|
648 | /* FIXME */
|
---|
649 | FFI_ASSERT(0);
|
---|
650 | break;
|
---|
651 |
|
---|
652 | case FFI_TYPE_DOUBLE:
|
---|
653 | {
|
---|
654 | if (l->fp_offset > 176-16)
|
---|
655 | {
|
---|
656 | avalue[i] = l->overflow_arg_area;
|
---|
657 | l->overflow_arg_area = (char *)l->overflow_arg_area + 8;
|
---|
658 | }
|
---|
659 | else
|
---|
660 | {
|
---|
661 | avalue[i] = (char *)l->reg_save_area + l->fp_offset;
|
---|
662 | l->fp_offset += 16;
|
---|
663 | }
|
---|
664 | }
|
---|
665 | #if DEBUG_FFI
|
---|
666 | fprintf (stderr, "double arg %d = %g\n", i, *(double *)avalue[i]);
|
---|
667 | #endif
|
---|
668 | break;
|
---|
669 |
|
---|
670 | case FFI_TYPE_FLOAT:
|
---|
671 | {
|
---|
672 | if (l->fp_offset > 176-16)
|
---|
673 | {
|
---|
674 | avalue[i] = l->overflow_arg_area;
|
---|
675 | l->overflow_arg_area = (char *)l->overflow_arg_area + 8;
|
---|
676 | }
|
---|
677 | else
|
---|
678 | {
|
---|
679 | avalue[i] = (char *)l->reg_save_area + l->fp_offset;
|
---|
680 | l->fp_offset += 16;
|
---|
681 | }
|
---|
682 | }
|
---|
683 | #if DEBUG_FFI
|
---|
684 | fprintf (stderr, "float arg %d = %g\n", i, *(float *)avalue[i]);
|
---|
685 | #endif
|
---|
686 | break;
|
---|
687 |
|
---|
688 | default:
|
---|
689 | FFI_ASSERT(0);
|
---|
690 | }
|
---|
691 |
|
---|
692 | argn += ALIGN(arg_types[i]->size, SIZEOF_ARG) / SIZEOF_ARG;
|
---|
693 | i++;
|
---|
694 | }
|
---|
695 |
|
---|
696 | /* Invoke the closure. */
|
---|
697 | (closure->fun) (cif, rp, avalue, closure->user_data);
|
---|
698 |
|
---|
699 | /* FIXME: Structs not supported. */
|
---|
700 | FFI_ASSERT(cif->rtype->type != FFI_TYPE_STRUCT);
|
---|
701 |
|
---|
702 | /* Tell ffi_closure_UNIX64 how to perform return type promotions. */
|
---|
703 |
|
---|
704 | return cif->rtype->type;
|
---|
705 | }
|
---|
706 | #endif /* ifndef __x86_64__ */
|
---|