?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
memory: move tcg flush into a tcg memory listener
[qemu/ar7.git] / tcg / arm / tcg-target.c
blob737200e5e68c80f5013f3214971990487b99d6da
1 /*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2008 Andrzej Zaborowski
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #if defined(__ARM_ARCH_7__) || \
26 defined(__ARM_ARCH_7A__) || \
27 defined(__ARM_ARCH_7EM__) || \
28 defined(__ARM_ARCH_7M__) || \
29 defined(__ARM_ARCH_7R__)
30 #define USE_ARMV7_INSTRUCTIONS
31 #endif
32
33 #if defined(USE_ARMV7_INSTRUCTIONS) || \
34 defined(__ARM_ARCH_6J__) || \
35 defined(__ARM_ARCH_6K__) || \
36 defined(__ARM_ARCH_6T2__) || \
37 defined(__ARM_ARCH_6Z__) || \
38 defined(__ARM_ARCH_6ZK__)
39 #define USE_ARMV6_INSTRUCTIONS
40 #endif
41
42 #if defined(USE_ARMV6_INSTRUCTIONS) || \
43 defined(__ARM_ARCH_5T__) || \
44 defined(__ARM_ARCH_5TE__) || \
45 defined(__ARM_ARCH_5TEJ__)
46 #define USE_ARMV5_INSTRUCTIONS
47 #endif
48
49 #ifdef USE_ARMV5_INSTRUCTIONS
50 static const int use_armv5_instructions = 1;
51 #else
52 static const int use_armv5_instructions = 0;
53 #endif
54 #undef USE_ARMV5_INSTRUCTIONS
55
56 #ifdef USE_ARMV6_INSTRUCTIONS
57 static const int use_armv6_instructions = 1;
58 #else
59 static const int use_armv6_instructions = 0;
60 #endif
61 #undef USE_ARMV6_INSTRUCTIONS
62
63 #ifdef USE_ARMV7_INSTRUCTIONS
64 static const int use_armv7_instructions = 1;
65 #else
66 static const int use_armv7_instructions = 0;
67 #endif
68 #undef USE_ARMV7_INSTRUCTIONS
69
70 #ifndef NDEBUG
71 static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
72 "%r0",
73 "%r1",
74 "%r2",
75 "%r3",
76 "%r4",
77 "%r5",
78 "%r6",
79 "%r7",
80 "%r8",
81 "%r9",
82 "%r10",
83 "%r11",
84 "%r12",
85 "%r13",
86 "%r14",
87 "%pc",
88 };
89 #endif
90
91 static const int tcg_target_reg_alloc_order[] = {
92 TCG_REG_R4,
93 TCG_REG_R5,
94 TCG_REG_R6,
95 TCG_REG_R7,
96 TCG_REG_R8,
97 TCG_REG_R9,
98 TCG_REG_R10,
99 TCG_REG_R11,
100 TCG_REG_R13,
101 TCG_REG_R0,
102 TCG_REG_R1,
103 TCG_REG_R2,
104 TCG_REG_R3,
105 TCG_REG_R12,
106 TCG_REG_R14,
107 };
108
109 static const int tcg_target_call_iarg_regs[4] = {
110 TCG_REG_R0, TCG_REG_R1, TCG_REG_R2, TCG_REG_R3
111 };
112 static const int tcg_target_call_oarg_regs[2] = {
113 TCG_REG_R0, TCG_REG_R1
114 };
115
116 static inline void reloc_abs32(void *code_ptr, tcg_target_long target)
117 {
118 *(uint32_t *) code_ptr = target;
119 }
120
121 static inline void reloc_pc24(void *code_ptr, tcg_target_long target)
122 {
123 uint32_t offset = ((target - ((tcg_target_long) code_ptr + 8)) >> 2);
124
125 *(uint32_t *) code_ptr = ((*(uint32_t *) code_ptr) & ~0xffffff)
126 | (offset & 0xffffff);
127 }
128
129 static void patch_reloc(uint8_t *code_ptr, int type,
130 tcg_target_long value, tcg_target_long addend)
131 {
132 switch (type) {
133 case R_ARM_ABS32:
134 reloc_abs32(code_ptr, value);
135 break;
136
137 case R_ARM_CALL:
138 case R_ARM_JUMP24:
139 default:
140 tcg_abort();
141
142 case R_ARM_PC24:
143 reloc_pc24(code_ptr, value);
144 break;
145 }
146 }
147
148 /* parse target specific constraints */
149 static int target_parse_constraint(TCGArgConstraint *ct, const char **pct_str)
150 {
151 const char *ct_str;
152
153 ct_str = *pct_str;
154 switch (ct_str[0]) {
155 case 'I':
156 ct->ct |= TCG_CT_CONST_ARM;
157 break;
158
159 case 'r':
160 ct->ct |= TCG_CT_REG;
161 tcg_regset_set32(ct->u.regs, 0, (1 << TCG_TARGET_NB_REGS) - 1);
162 break;
163
164 /* qemu_ld address */
165 case 'l':
166 ct->ct |= TCG_CT_REG;
167 tcg_regset_set32(ct->u.regs, 0, (1 << TCG_TARGET_NB_REGS) - 1);
168 #ifdef CONFIG_SOFTMMU
169 /* r0 and r1 will be overwritten when reading the tlb entry,
170 so don't use these. */
171 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0);
172 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1);
173 #if TARGET_LONG_BITS == 64
174 /* If we're passing env to the helper as r0 and need a regpair
175 * for the address then r2 will be overwritten as we're setting
176 * up the args to the helper.
177 */
178 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R2);
179 #endif
180 #endif
181 break;
182 case 'L':
183 ct->ct |= TCG_CT_REG;
184 tcg_regset_set32(ct->u.regs, 0, (1 << TCG_TARGET_NB_REGS) - 1);
185 #ifdef CONFIG_SOFTMMU
186 /* r1 is still needed to load data_reg or data_reg2,
187 so don't use it. */
188 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1);
189 #endif
190 break;
191
192 /* qemu_st address & data_reg */
193 case 's':
194 ct->ct |= TCG_CT_REG;
195 tcg_regset_set32(ct->u.regs, 0, (1 << TCG_TARGET_NB_REGS) - 1);
196 /* r0 and r1 will be overwritten when reading the tlb entry
197 (softmmu only) and doing the byte swapping, so don't
198 use these. */
199 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0);
200 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1);
201 #if defined(CONFIG_SOFTMMU) && (TARGET_LONG_BITS == 64)
202 /* Avoid clashes with registers being used for helper args */
203 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R2);
204 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3);
205 #endif
206 break;
207 /* qemu_st64 data_reg2 */
208 case 'S':
209 ct->ct |= TCG_CT_REG;
210 tcg_regset_set32(ct->u.regs, 0, (1 << TCG_TARGET_NB_REGS) - 1);
211 /* r0 and r1 will be overwritten when reading the tlb entry
212 (softmmu only) and doing the byte swapping, so don't
213 use these. */
214 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0);
215 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1);
216 #ifdef CONFIG_SOFTMMU
217 /* r2 is still needed to load data_reg, so don't use it. */
218 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R2);
219 #if TARGET_LONG_BITS == 64
220 /* Avoid clashes with registers being used for helper args */
221 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3);
222 #endif
223 #endif
224 break;
225
226 default:
227 return -1;
228 }
229 ct_str++;
230 *pct_str = ct_str;
231
232 return 0;
233 }
234
235 static inline uint32_t rotl(uint32_t val, int n)
236 {
237 return (val << n) | (val >> (32 - n));
238 }
239
240 /* ARM immediates for ALU instructions are made of an unsigned 8-bit
241 right-rotated by an even amount between 0 and 30. */
242 static inline int encode_imm(uint32_t imm)
243 {
244 int shift;
245
246 /* simple case, only lower bits */
247 if ((imm & ~0xff) == 0)
248 return 0;
249 /* then try a simple even shift */
250 shift = ctz32(imm) & ~1;
251 if (((imm >> shift) & ~0xff) == 0)
252 return 32 - shift;
253 /* now try harder with rotations */
254 if ((rotl(imm, 2) & ~0xff) == 0)
255 return 2;
256 if ((rotl(imm, 4) & ~0xff) == 0)
257 return 4;
258 if ((rotl(imm, 6) & ~0xff) == 0)
259 return 6;
260 /* imm can't be encoded */
261 return -1;
262 }
263
264 static inline int check_fit_imm(uint32_t imm)
265 {
266 return encode_imm(imm) >= 0;
267 }
268
269 /* Test if a constant matches the constraint.
270 * TODO: define constraints for:
271 *
272 * ldr/str offset: between -0xfff and 0xfff
273 * ldrh/strh offset: between -0xff and 0xff
274 * mov operand2: values represented with x << (2 * y), x < 0x100
275 * add, sub, eor...: ditto
276 */
277 static inline int tcg_target_const_match(tcg_target_long val,
278 const TCGArgConstraint *arg_ct)
279 {
280 int ct;
281 ct = arg_ct->ct;
282 if (ct & TCG_CT_CONST)
283 return 1;
284 else if ((ct & TCG_CT_CONST_ARM) && check_fit_imm(val))
285 return 1;
286 else
287 return 0;
288 }
289
290 enum arm_data_opc_e {
291 ARITH_AND = 0x0,
292 ARITH_EOR = 0x1,
293 ARITH_SUB = 0x2,
294 ARITH_RSB = 0x3,
295 ARITH_ADD = 0x4,
296 ARITH_ADC = 0x5,
297 ARITH_SBC = 0x6,
298 ARITH_RSC = 0x7,
299 ARITH_TST = 0x8,
300 ARITH_CMP = 0xa,
301 ARITH_CMN = 0xb,
302 ARITH_ORR = 0xc,
303 ARITH_MOV = 0xd,
304 ARITH_BIC = 0xe,
305 ARITH_MVN = 0xf,
306 };
307
308 #define TO_CPSR(opc) \
309 ((opc == ARITH_CMP || opc == ARITH_CMN || opc == ARITH_TST) << 20)
310
311 #define SHIFT_IMM_LSL(im) (((im) << 7) | 0x00)
312 #define SHIFT_IMM_LSR(im) (((im) << 7) | 0x20)
313 #define SHIFT_IMM_ASR(im) (((im) << 7) | 0x40)
314 #define SHIFT_IMM_ROR(im) (((im) << 7) | 0x60)
315 #define SHIFT_REG_LSL(rs) (((rs) << 8) | 0x10)
316 #define SHIFT_REG_LSR(rs) (((rs) << 8) | 0x30)
317 #define SHIFT_REG_ASR(rs) (((rs) << 8) | 0x50)
318 #define SHIFT_REG_ROR(rs) (((rs) << 8) | 0x70)
319
320 enum arm_cond_code_e {
321 COND_EQ = 0x0,
322 COND_NE = 0x1,
323 COND_CS = 0x2, /* Unsigned greater or equal */
324 COND_CC = 0x3, /* Unsigned less than */
325 COND_MI = 0x4, /* Negative */
326 COND_PL = 0x5, /* Zero or greater */
327 COND_VS = 0x6, /* Overflow */
328 COND_VC = 0x7, /* No overflow */
329 COND_HI = 0x8, /* Unsigned greater than */
330 COND_LS = 0x9, /* Unsigned less or equal */
331 COND_GE = 0xa,
332 COND_LT = 0xb,
333 COND_GT = 0xc,
334 COND_LE = 0xd,
335 COND_AL = 0xe,
336 };
337
338 static const uint8_t tcg_cond_to_arm_cond[] = {
339 [TCG_COND_EQ] = COND_EQ,
340 [TCG_COND_NE] = COND_NE,
341 [TCG_COND_LT] = COND_LT,
342 [TCG_COND_GE] = COND_GE,
343 [TCG_COND_LE] = COND_LE,
344 [TCG_COND_GT] = COND_GT,
345 /* unsigned */
346 [TCG_COND_LTU] = COND_CC,
347 [TCG_COND_GEU] = COND_CS,
348 [TCG_COND_LEU] = COND_LS,
349 [TCG_COND_GTU] = COND_HI,
350 };
351
352 static inline void tcg_out_bx(TCGContext *s, int cond, int rn)
353 {
354 tcg_out32(s, (cond << 28) | 0x012fff10 | rn);
355 }
356
357 static inline void tcg_out_b(TCGContext *s, int cond, int32_t offset)
358 {
359 tcg_out32(s, (cond << 28) | 0x0a000000 |
360 (((offset - 8) >> 2) & 0x00ffffff));
361 }
362
363 static inline void tcg_out_b_noaddr(TCGContext *s, int cond)
364 {
365 /* We pay attention here to not modify the branch target by skipping
366 the corresponding bytes. This ensure that caches and memory are
367 kept coherent during retranslation. */
368 #ifdef HOST_WORDS_BIGENDIAN
369 tcg_out8(s, (cond << 4) | 0x0a);
370 s->code_ptr += 3;
371 #else
372 s->code_ptr += 3;
373 tcg_out8(s, (cond << 4) | 0x0a);
374 #endif
375 }
376
377 static inline void tcg_out_bl(TCGContext *s, int cond, int32_t offset)
378 {
379 tcg_out32(s, (cond << 28) | 0x0b000000 |
380 (((offset - 8) >> 2) & 0x00ffffff));
381 }
382
383 static inline void tcg_out_blx(TCGContext *s, int cond, int rn)
384 {
385 tcg_out32(s, (cond << 28) | 0x012fff30 | rn);
386 }
387
388 static inline void tcg_out_blx_imm(TCGContext *s, int32_t offset)
389 {
390 tcg_out32(s, 0xfa000000 | ((offset & 2) << 23) |
391 (((offset - 8) >> 2) & 0x00ffffff));
392 }
393
394 static inline void tcg_out_dat_reg(TCGContext *s,
395 int cond, int opc, int rd, int rn, int rm, int shift)
396 {
397 tcg_out32(s, (cond << 28) | (0 << 25) | (opc << 21) | TO_CPSR(opc) |
398 (rn << 16) | (rd << 12) | shift | rm);
399 }
400
401 static inline void tcg_out_mov_reg(TCGContext *s, int cond, int rd, int rm)
402 {
403 /* Simple reg-reg move, optimising out the 'do nothing' case */
404 if (rd != rm) {
405 tcg_out_dat_reg(s, cond, ARITH_MOV, rd, 0, rm, SHIFT_IMM_LSL(0));
406 }
407 }
408
409 static inline void tcg_out_dat_reg2(TCGContext *s,
410 int cond, int opc0, int opc1, int rd0, int rd1,
411 int rn0, int rn1, int rm0, int rm1, int shift)
412 {
413 if (rd0 == rn1 || rd0 == rm1) {
414 tcg_out32(s, (cond << 28) | (0 << 25) | (opc0 << 21) | (1 << 20) |
415 (rn0 << 16) | (8 << 12) | shift | rm0);
416 tcg_out32(s, (cond << 28) | (0 << 25) | (opc1 << 21) |
417 (rn1 << 16) | (rd1 << 12) | shift | rm1);
418 tcg_out_dat_reg(s, cond, ARITH_MOV,
419 rd0, 0, TCG_REG_R8, SHIFT_IMM_LSL(0));
420 } else {
421 tcg_out32(s, (cond << 28) | (0 << 25) | (opc0 << 21) | (1 << 20) |
422 (rn0 << 16) | (rd0 << 12) | shift | rm0);
423 tcg_out32(s, (cond << 28) | (0 << 25) | (opc1 << 21) |
424 (rn1 << 16) | (rd1 << 12) | shift | rm1);
425 }
426 }
427
428 static inline void tcg_out_dat_imm(TCGContext *s,
429 int cond, int opc, int rd, int rn, int im)
430 {
431 tcg_out32(s, (cond << 28) | (1 << 25) | (opc << 21) | TO_CPSR(opc) |
432 (rn << 16) | (rd << 12) | im);
433 }
434
435 static inline void tcg_out_movi32(TCGContext *s,
436 int cond, int rd, uint32_t arg)
437 {
438 /* TODO: This is very suboptimal, we can easily have a constant
439 * pool somewhere after all the instructions. */
440 if ((int)arg < 0 && (int)arg >= -0x100) {
441 tcg_out_dat_imm(s, cond, ARITH_MVN, rd, 0, (~arg) & 0xff);
442 } else if (use_armv7_instructions) {
443 /* use movw/movt */
444 /* movw */
445 tcg_out32(s, (cond << 28) | 0x03000000 | (rd << 12)
446 | ((arg << 4) & 0x000f0000) | (arg & 0xfff));
447 if (arg & 0xffff0000) {
448 /* movt */
449 tcg_out32(s, (cond << 28) | 0x03400000 | (rd << 12)
450 | ((arg >> 12) & 0x000f0000) | ((arg >> 16) & 0xfff));
451 }
452 } else {
453 int opc = ARITH_MOV;
454 int rn = 0;
455
456 do {
457 int i, rot;
458
459 i = ctz32(arg) & ~1;
460 rot = ((32 - i) << 7) & 0xf00;
461 tcg_out_dat_imm(s, cond, opc, rd, rn, ((arg >> i) & 0xff) | rot);
462 arg &= ~(0xff << i);
463
464 opc = ARITH_ORR;
465 rn = rd;
466 } while (arg);
467 }
468 }
469
470 static inline void tcg_out_mul32(TCGContext *s,
471 int cond, int rd, int rs, int rm)
472 {
473 if (rd != rm)
474 tcg_out32(s, (cond << 28) | (rd << 16) | (0 << 12) |
475 (rs << 8) | 0x90 | rm);
476 else if (rd != rs)
477 tcg_out32(s, (cond << 28) | (rd << 16) | (0 << 12) |
478 (rm << 8) | 0x90 | rs);
479 else {
480 tcg_out32(s, (cond << 28) | ( 8 << 16) | (0 << 12) |
481 (rs << 8) | 0x90 | rm);
482 tcg_out_dat_reg(s, cond, ARITH_MOV,
483 rd, 0, TCG_REG_R8, SHIFT_IMM_LSL(0));
484 }
485 }
486
487 static inline void tcg_out_umull32(TCGContext *s,
488 int cond, int rd0, int rd1, int rs, int rm)
489 {
490 if (rd0 != rm && rd1 != rm)
491 tcg_out32(s, (cond << 28) | 0x800090 |
492 (rd1 << 16) | (rd0 << 12) | (rs << 8) | rm);
493 else if (rd0 != rs && rd1 != rs)
494 tcg_out32(s, (cond << 28) | 0x800090 |
495 (rd1 << 16) | (rd0 << 12) | (rm << 8) | rs);
496 else {
497 tcg_out_dat_reg(s, cond, ARITH_MOV,
498 TCG_REG_R8, 0, rm, SHIFT_IMM_LSL(0));
499 tcg_out32(s, (cond << 28) | 0x800098 |
500 (rd1 << 16) | (rd0 << 12) | (rs << 8));
501 }
502 }
503
504 static inline void tcg_out_smull32(TCGContext *s,
505 int cond, int rd0, int rd1, int rs, int rm)
506 {
507 if (rd0 != rm && rd1 != rm)
508 tcg_out32(s, (cond << 28) | 0xc00090 |
509 (rd1 << 16) | (rd0 << 12) | (rs << 8) | rm);
510 else if (rd0 != rs && rd1 != rs)
511 tcg_out32(s, (cond << 28) | 0xc00090 |
512 (rd1 << 16) | (rd0 << 12) | (rm << 8) | rs);
513 else {
514 tcg_out_dat_reg(s, cond, ARITH_MOV,
515 TCG_REG_R8, 0, rm, SHIFT_IMM_LSL(0));
516 tcg_out32(s, (cond << 28) | 0xc00098 |
517 (rd1 << 16) | (rd0 << 12) | (rs << 8));
518 }
519 }
520
521 static inline void tcg_out_ext8s(TCGContext *s, int cond,
522 int rd, int rn)
523 {
524 if (use_armv6_instructions) {
525 /* sxtb */
526 tcg_out32(s, 0x06af0070 | (cond << 28) | (rd << 12) | rn);
527 } else {
528 tcg_out_dat_reg(s, cond, ARITH_MOV,
529 rd, 0, rn, SHIFT_IMM_LSL(24));
530 tcg_out_dat_reg(s, cond, ARITH_MOV,
531 rd, 0, rd, SHIFT_IMM_ASR(24));
532 }
533 }
534
535 static inline void tcg_out_ext8u(TCGContext *s, int cond,
536 int rd, int rn)
537 {
538 tcg_out_dat_imm(s, cond, ARITH_AND, rd, rn, 0xff);
539 }
540
541 static inline void tcg_out_ext16s(TCGContext *s, int cond,
542 int rd, int rn)
543 {
544 if (use_armv6_instructions) {
545 /* sxth */
546 tcg_out32(s, 0x06bf0070 | (cond << 28) | (rd << 12) | rn);
547 } else {
548 tcg_out_dat_reg(s, cond, ARITH_MOV,
549 rd, 0, rn, SHIFT_IMM_LSL(16));
550 tcg_out_dat_reg(s, cond, ARITH_MOV,
551 rd, 0, rd, SHIFT_IMM_ASR(16));
552 }
553 }
554
555 static inline void tcg_out_ext16u(TCGContext *s, int cond,
556 int rd, int rn)
557 {
558 if (use_armv6_instructions) {
559 /* uxth */
560 tcg_out32(s, 0x06ff0070 | (cond << 28) | (rd << 12) | rn);
561 } else {
562 tcg_out_dat_reg(s, cond, ARITH_MOV,
563 rd, 0, rn, SHIFT_IMM_LSL(16));
564 tcg_out_dat_reg(s, cond, ARITH_MOV,
565 rd, 0, rd, SHIFT_IMM_LSR(16));
566 }
567 }
568
569 static inline void tcg_out_bswap16s(TCGContext *s, int cond, int rd, int rn)
570 {
571 if (use_armv6_instructions) {
572 /* revsh */
573 tcg_out32(s, 0x06ff0fb0 | (cond << 28) | (rd << 12) | rn);
574 } else {
575 tcg_out_dat_reg(s, cond, ARITH_MOV,
576 TCG_REG_R8, 0, rn, SHIFT_IMM_LSL(24));
577 tcg_out_dat_reg(s, cond, ARITH_MOV,
578 TCG_REG_R8, 0, TCG_REG_R8, SHIFT_IMM_ASR(16));
579 tcg_out_dat_reg(s, cond, ARITH_ORR,
580 rd, TCG_REG_R8, rn, SHIFT_IMM_LSR(8));
581 }
582 }
583
584 static inline void tcg_out_bswap16(TCGContext *s, int cond, int rd, int rn)
585 {
586 if (use_armv6_instructions) {
587 /* rev16 */
588 tcg_out32(s, 0x06bf0fb0 | (cond << 28) | (rd << 12) | rn);
589 } else {
590 tcg_out_dat_reg(s, cond, ARITH_MOV,
591 TCG_REG_R8, 0, rn, SHIFT_IMM_LSL(24));
592 tcg_out_dat_reg(s, cond, ARITH_MOV,
593 TCG_REG_R8, 0, TCG_REG_R8, SHIFT_IMM_LSR(16));
594 tcg_out_dat_reg(s, cond, ARITH_ORR,
595 rd, TCG_REG_R8, rn, SHIFT_IMM_LSR(8));
596 }
597 }
598
599 static inline void tcg_out_bswap32(TCGContext *s, int cond, int rd, int rn)
600 {
601 if (use_armv6_instructions) {
602 /* rev */
603 tcg_out32(s, 0x06bf0f30 | (cond << 28) | (rd << 12) | rn);
604 } else {
605 tcg_out_dat_reg(s, cond, ARITH_EOR,
606 TCG_REG_R8, rn, rn, SHIFT_IMM_ROR(16));
607 tcg_out_dat_imm(s, cond, ARITH_BIC,
608 TCG_REG_R8, TCG_REG_R8, 0xff | 0x800);
609 tcg_out_dat_reg(s, cond, ARITH_MOV,
610 rd, 0, rn, SHIFT_IMM_ROR(8));
611 tcg_out_dat_reg(s, cond, ARITH_EOR,
612 rd, rd, TCG_REG_R8, SHIFT_IMM_LSR(8));
613 }
614 }
615
616 static inline void tcg_out_ld32_12(TCGContext *s, int cond,
617 int rd, int rn, tcg_target_long im)
618 {
619 if (im >= 0)
620 tcg_out32(s, (cond << 28) | 0x05900000 |
621 (rn << 16) | (rd << 12) | (im & 0xfff));
622 else
623 tcg_out32(s, (cond << 28) | 0x05100000 |
624 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
625 }
626
627 static inline void tcg_out_st32_12(TCGContext *s, int cond,
628 int rd, int rn, tcg_target_long im)
629 {
630 if (im >= 0)
631 tcg_out32(s, (cond << 28) | 0x05800000 |
632 (rn << 16) | (rd << 12) | (im & 0xfff));
633 else
634 tcg_out32(s, (cond << 28) | 0x05000000 |
635 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
636 }
637
638 static inline void tcg_out_ld32_r(TCGContext *s, int cond,
639 int rd, int rn, int rm)
640 {
641 tcg_out32(s, (cond << 28) | 0x07900000 |
642 (rn << 16) | (rd << 12) | rm);
643 }
644
645 static inline void tcg_out_st32_r(TCGContext *s, int cond,
646 int rd, int rn, int rm)
647 {
648 tcg_out32(s, (cond << 28) | 0x07800000 |
649 (rn << 16) | (rd << 12) | rm);
650 }
651
652 /* Register pre-increment with base writeback. */
653 static inline void tcg_out_ld32_rwb(TCGContext *s, int cond,
654 int rd, int rn, int rm)
655 {
656 tcg_out32(s, (cond << 28) | 0x07b00000 |
657 (rn << 16) | (rd << 12) | rm);
658 }
659
660 static inline void tcg_out_st32_rwb(TCGContext *s, int cond,
661 int rd, int rn, int rm)
662 {
663 tcg_out32(s, (cond << 28) | 0x07a00000 |
664 (rn << 16) | (rd << 12) | rm);
665 }
666
667 static inline void tcg_out_ld16u_8(TCGContext *s, int cond,
668 int rd, int rn, tcg_target_long im)
669 {
670 if (im >= 0)
671 tcg_out32(s, (cond << 28) | 0x01d000b0 |
672 (rn << 16) | (rd << 12) |
673 ((im & 0xf0) << 4) | (im & 0xf));
674 else
675 tcg_out32(s, (cond << 28) | 0x015000b0 |
676 (rn << 16) | (rd << 12) |
677 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
678 }
679
680 static inline void tcg_out_st16_8(TCGContext *s, int cond,
681 int rd, int rn, tcg_target_long im)
682 {
683 if (im >= 0)
684 tcg_out32(s, (cond << 28) | 0x01c000b0 |
685 (rn << 16) | (rd << 12) |
686 ((im & 0xf0) << 4) | (im & 0xf));
687 else
688 tcg_out32(s, (cond << 28) | 0x014000b0 |
689 (rn << 16) | (rd << 12) |
690 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
691 }
692
693 static inline void tcg_out_ld16u_r(TCGContext *s, int cond,
694 int rd, int rn, int rm)
695 {
696 tcg_out32(s, (cond << 28) | 0x019000b0 |
697 (rn << 16) | (rd << 12) | rm);
698 }
699
700 static inline void tcg_out_st16_r(TCGContext *s, int cond,
701 int rd, int rn, int rm)
702 {
703 tcg_out32(s, (cond << 28) | 0x018000b0 |
704 (rn << 16) | (rd << 12) | rm);
705 }
706
707 static inline void tcg_out_ld16s_8(TCGContext *s, int cond,
708 int rd, int rn, tcg_target_long im)
709 {
710 if (im >= 0)
711 tcg_out32(s, (cond << 28) | 0x01d000f0 |
712 (rn << 16) | (rd << 12) |
713 ((im & 0xf0) << 4) | (im & 0xf));
714 else
715 tcg_out32(s, (cond << 28) | 0x015000f0 |
716 (rn << 16) | (rd << 12) |
717 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
718 }
719
720 static inline void tcg_out_ld16s_r(TCGContext *s, int cond,
721 int rd, int rn, int rm)
722 {
723 tcg_out32(s, (cond << 28) | 0x019000f0 |
724 (rn << 16) | (rd << 12) | rm);
725 }
726
727 static inline void tcg_out_ld8_12(TCGContext *s, int cond,
728 int rd, int rn, tcg_target_long im)
729 {
730 if (im >= 0)
731 tcg_out32(s, (cond << 28) | 0x05d00000 |
732 (rn << 16) | (rd << 12) | (im & 0xfff));
733 else
734 tcg_out32(s, (cond << 28) | 0x05500000 |
735 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
736 }
737
738 static inline void tcg_out_st8_12(TCGContext *s, int cond,
739 int rd, int rn, tcg_target_long im)
740 {
741 if (im >= 0)
742 tcg_out32(s, (cond << 28) | 0x05c00000 |
743 (rn << 16) | (rd << 12) | (im & 0xfff));
744 else
745 tcg_out32(s, (cond << 28) | 0x05400000 |
746 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
747 }
748
749 static inline void tcg_out_ld8_r(TCGContext *s, int cond,
750 int rd, int rn, int rm)
751 {
752 tcg_out32(s, (cond << 28) | 0x07d00000 |
753 (rn << 16) | (rd << 12) | rm);
754 }
755
756 static inline void tcg_out_st8_r(TCGContext *s, int cond,
757 int rd, int rn, int rm)
758 {
759 tcg_out32(s, (cond << 28) | 0x07c00000 |
760 (rn << 16) | (rd << 12) | rm);
761 }
762
763 static inline void tcg_out_ld8s_8(TCGContext *s, int cond,
764 int rd, int rn, tcg_target_long im)
765 {
766 if (im >= 0)
767 tcg_out32(s, (cond << 28) | 0x01d000d0 |
768 (rn << 16) | (rd << 12) |
769 ((im & 0xf0) << 4) | (im & 0xf));
770 else
771 tcg_out32(s, (cond << 28) | 0x015000d0 |
772 (rn << 16) | (rd << 12) |
773 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
774 }
775
776 static inline void tcg_out_ld8s_r(TCGContext *s, int cond,
777 int rd, int rn, int rm)
778 {
779 tcg_out32(s, (cond << 28) | 0x019000d0 |
780 (rn << 16) | (rd << 12) | rm);
781 }
782
783 static inline void tcg_out_ld32u(TCGContext *s, int cond,
784 int rd, int rn, int32_t offset)
785 {
786 if (offset > 0xfff || offset < -0xfff) {
787 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
788 tcg_out_ld32_r(s, cond, rd, rn, TCG_REG_R8);
789 } else
790 tcg_out_ld32_12(s, cond, rd, rn, offset);
791 }
792
793 static inline void tcg_out_st32(TCGContext *s, int cond,
794 int rd, int rn, int32_t offset)
795 {
796 if (offset > 0xfff || offset < -0xfff) {
797 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
798 tcg_out_st32_r(s, cond, rd, rn, TCG_REG_R8);
799 } else
800 tcg_out_st32_12(s, cond, rd, rn, offset);
801 }
802
803 static inline void tcg_out_ld16u(TCGContext *s, int cond,
804 int rd, int rn, int32_t offset)
805 {
806 if (offset > 0xff || offset < -0xff) {
807 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
808 tcg_out_ld16u_r(s, cond, rd, rn, TCG_REG_R8);
809 } else
810 tcg_out_ld16u_8(s, cond, rd, rn, offset);
811 }
812
813 static inline void tcg_out_ld16s(TCGContext *s, int cond,
814 int rd, int rn, int32_t offset)
815 {
816 if (offset > 0xff || offset < -0xff) {
817 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
818 tcg_out_ld16s_r(s, cond, rd, rn, TCG_REG_R8);
819 } else
820 tcg_out_ld16s_8(s, cond, rd, rn, offset);
821 }
822
823 static inline void tcg_out_st16(TCGContext *s, int cond,
824 int rd, int rn, int32_t offset)
825 {
826 if (offset > 0xff || offset < -0xff) {
827 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
828 tcg_out_st16_r(s, cond, rd, rn, TCG_REG_R8);
829 } else
830 tcg_out_st16_8(s, cond, rd, rn, offset);
831 }
832
833 static inline void tcg_out_ld8u(TCGContext *s, int cond,
834 int rd, int rn, int32_t offset)
835 {
836 if (offset > 0xfff || offset < -0xfff) {
837 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
838 tcg_out_ld8_r(s, cond, rd, rn, TCG_REG_R8);
839 } else
840 tcg_out_ld8_12(s, cond, rd, rn, offset);
841 }
842
843 static inline void tcg_out_ld8s(TCGContext *s, int cond,
844 int rd, int rn, int32_t offset)
845 {
846 if (offset > 0xff || offset < -0xff) {
847 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
848 tcg_out_ld8s_r(s, cond, rd, rn, TCG_REG_R8);
849 } else
850 tcg_out_ld8s_8(s, cond, rd, rn, offset);
851 }
852
853 static inline void tcg_out_st8(TCGContext *s, int cond,
854 int rd, int rn, int32_t offset)
855 {
856 if (offset > 0xfff || offset < -0xfff) {
857 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
858 tcg_out_st8_r(s, cond, rd, rn, TCG_REG_R8);
859 } else
860 tcg_out_st8_12(s, cond, rd, rn, offset);
861 }
862
863 /* The _goto case is normally between TBs within the same code buffer,
864 * and with the code buffer limited to 16MB we shouldn't need the long
865 * case.
866 *
867 * .... except to the prologue that is in its own buffer.
868 */
869 static inline void tcg_out_goto(TCGContext *s, int cond, uint32_t addr)
870 {
871 int32_t val;
872
873 if (addr & 1) {
874 /* goto to a Thumb destination isn't supported */
875 tcg_abort();
876 }
877
878 val = addr - (tcg_target_long) s->code_ptr;
879 if (val - 8 < 0x01fffffd && val - 8 > -0x01fffffd)
880 tcg_out_b(s, cond, val);
881 else {
882 if (cond == COND_AL) {
883 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
884 tcg_out32(s, addr);
885 } else {
886 tcg_out_movi32(s, cond, TCG_REG_R8, val - 8);
887 tcg_out_dat_reg(s, cond, ARITH_ADD,
888 TCG_REG_PC, TCG_REG_PC,
889 TCG_REG_R8, SHIFT_IMM_LSL(0));
890 }
891 }
892 }
893
894 /* The call case is mostly used for helpers - so it's not unreasonable
895 * for them to be beyond branch range */
896 static inline void tcg_out_call(TCGContext *s, uint32_t addr)
897 {
898 int32_t val;
899
900 val = addr - (tcg_target_long) s->code_ptr;
901 if (val - 8 < 0x02000000 && val - 8 >= -0x02000000) {
902 if (addr & 1) {
903 /* Use BLX if the target is in Thumb mode */
904 if (!use_armv5_instructions) {
905 tcg_abort();
906 }
907 tcg_out_blx_imm(s, val);
908 } else {
909 tcg_out_bl(s, COND_AL, val);
910 }
911 } else {
912 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R14, TCG_REG_PC, 4);
913 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
914 tcg_out32(s, addr);
915 }
916 }
917
918 static inline void tcg_out_callr(TCGContext *s, int cond, int arg)
919 {
920 if (use_armv5_instructions) {
921 tcg_out_blx(s, cond, arg);
922 } else {
923 tcg_out_dat_reg(s, cond, ARITH_MOV, TCG_REG_R14, 0,
924 TCG_REG_PC, SHIFT_IMM_LSL(0));
925 tcg_out_bx(s, cond, arg);
926 }
927 }
928
929 static inline void tcg_out_goto_label(TCGContext *s, int cond, int label_index)
930 {
931 TCGLabel *l = &s->labels[label_index];
932
933 if (l->has_value)
934 tcg_out_goto(s, cond, l->u.value);
935 else if (cond == COND_AL) {
936 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
937 tcg_out_reloc(s, s->code_ptr, R_ARM_ABS32, label_index, 31337);
938 s->code_ptr += 4;
939 } else {
940 /* Probably this should be preferred even for COND_AL... */
941 tcg_out_reloc(s, s->code_ptr, R_ARM_PC24, label_index, 31337);
942 tcg_out_b_noaddr(s, cond);
943 }
944 }
945
946 #ifdef CONFIG_SOFTMMU
947
948 #include "../../softmmu_defs.h"
949
950 /* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
951 int mmu_idx) */
952 static const void * const qemu_ld_helpers[4] = {
953 helper_ldb_mmu,
954 helper_ldw_mmu,
955 helper_ldl_mmu,
956 helper_ldq_mmu,
957 };
958
959 /* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
960 uintxx_t val, int mmu_idx) */
961 static const void * const qemu_st_helpers[4] = {
962 helper_stb_mmu,
963 helper_stw_mmu,
964 helper_stl_mmu,
965 helper_stq_mmu,
966 };
967
968 /* Helper routines for marshalling helper function arguments into
969 * the correct registers and stack.
970 * argreg is where we want to put this argument, arg is the argument itself.
971 * Return value is the updated argreg ready for the next call.
972 * Note that argreg 0..3 is real registers, 4+ on stack.
973 * When we reach the first stacked argument, we allocate space for it
974 * and the following stacked arguments using "str r8, [sp, #-0x10]!".
975 * Following arguments are filled in with "str r8, [sp, #0xNN]".
976 * For more than 4 stacked arguments we'd need to know how much
977 * space to allocate when we pushed the first stacked argument.
978 * We don't need this, so don't implement it (and will assert if you try it.)
979 *
980 * We provide routines for arguments which are: immediate, 32 bit
981 * value in register, 16 and 8 bit values in register (which must be zero
982 * extended before use) and 64 bit value in a lo:hi register pair.
983 */
984 #define DEFINE_TCG_OUT_ARG(NAME, ARGPARAM) \
985 static TCGReg NAME(TCGContext *s, TCGReg argreg, ARGPARAM) \
986 { \
987 if (argreg < 4) { \
988 TCG_OUT_ARG_GET_ARG(argreg); \
989 } else if (argreg == 4) { \
990 TCG_OUT_ARG_GET_ARG(TCG_REG_R8); \
991 tcg_out32(s, (COND_AL << 28) | 0x052d8010); \
992 } else { \
993 assert(argreg < 8); \
994 TCG_OUT_ARG_GET_ARG(TCG_REG_R8); \
995 tcg_out32(s, (COND_AL << 28) | 0x058d8000 | (argreg - 4) * 4); \
996 } \
997 return argreg + 1; \
998 }
999
1000 #define TCG_OUT_ARG_GET_ARG(A) tcg_out_dat_imm(s, COND_AL, ARITH_MOV, A, 0, arg)
1001 DEFINE_TCG_OUT_ARG(tcg_out_arg_imm32, uint32_t arg)
1002 #undef TCG_OUT_ARG_GET_ARG
1003 #define TCG_OUT_ARG_GET_ARG(A) tcg_out_ext8u(s, COND_AL, A, arg)
1004 DEFINE_TCG_OUT_ARG(tcg_out_arg_reg8, TCGReg arg)
1005 #undef TCG_OUT_ARG_GET_ARG
1006 #define TCG_OUT_ARG_GET_ARG(A) tcg_out_ext16u(s, COND_AL, A, arg)
1007 DEFINE_TCG_OUT_ARG(tcg_out_arg_reg16, TCGReg arg)
1008 #undef TCG_OUT_ARG_GET_ARG
1009
1010 /* We don't use the macro for this one to avoid an unnecessary reg-reg
1011 * move when storing to the stack.
1012 */
1013 static TCGReg tcg_out_arg_reg32(TCGContext *s, TCGReg argreg, TCGReg arg)
1014 {
1015 if (argreg < 4) {
1016 tcg_out_mov_reg(s, COND_AL, argreg, arg);
1017 } else if (argreg == 4) {
1018 /* str arg, [sp, #-0x10]! */
1019 tcg_out32(s, (COND_AL << 28) | 0x052d0010 | (arg << 12));
1020 } else {
1021 assert(argreg < 8);
1022 /* str arg, [sp, #0xNN] */
1023 tcg_out32(s, (COND_AL << 28) | 0x058d0000 |
1024 (arg << 12) | (argreg - 4) * 4);
1025 }
1026 return argreg + 1;
1027 }
1028
1029 static inline TCGReg tcg_out_arg_reg64(TCGContext *s, TCGReg argreg,
1030 TCGReg arglo, TCGReg arghi)
1031 {
1032 /* 64 bit arguments must go in even/odd register pairs
1033 * and in 8-aligned stack slots.
1034 */
1035 if (argreg & 1) {
1036 argreg++;
1037 }
1038 argreg = tcg_out_arg_reg32(s, argreg, arglo);
1039 argreg = tcg_out_arg_reg32(s, argreg, arghi);
1040 return argreg;
1041 }
1042
1043 static inline void tcg_out_arg_stacktidy(TCGContext *s, TCGReg argreg)
1044 {
1045 /* Output any necessary post-call cleanup of the stack */
1046 if (argreg > 4) {
1047 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R13, TCG_REG_R13, 0x10);
1048 }
1049 }
1050
1051 #endif
1052
1053 #define TLB_SHIFT (CPU_TLB_ENTRY_BITS + CPU_TLB_BITS)
1054
1055 static inline void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, int opc)
1056 {
1057 int addr_reg, data_reg, data_reg2, bswap;
1058 #ifdef CONFIG_SOFTMMU
1059 int mem_index, s_bits;
1060 TCGReg argreg;
1061 # if TARGET_LONG_BITS == 64
1062 int addr_reg2;
1063 # endif
1064 uint32_t *label_ptr;
1065 #endif
1066
1067 #ifdef TARGET_WORDS_BIGENDIAN
1068 bswap = 1;
1069 #else
1070 bswap = 0;
1071 #endif
1072 data_reg = *args++;
1073 if (opc == 3)
1074 data_reg2 = *args++;
1075 else
1076 data_reg2 = 0; /* suppress warning */
1077 addr_reg = *args++;
1078 #ifdef CONFIG_SOFTMMU
1079 # if TARGET_LONG_BITS == 64
1080 addr_reg2 = *args++;
1081 # endif
1082 mem_index = *args;
1083 s_bits = opc & 3;
1084
1085 /* Should generate something like the following:
1086 * shr r8, addr_reg, #TARGET_PAGE_BITS
1087 * and r0, r8, #(CPU_TLB_SIZE - 1) @ Assumption: CPU_TLB_BITS <= 8
1088 * add r0, env, r0 lsl #CPU_TLB_ENTRY_BITS
1089 */
1090 # if CPU_TLB_BITS > 8
1091 # error
1092 # endif
1093 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_R8,
1094 0, addr_reg, SHIFT_IMM_LSR(TARGET_PAGE_BITS));
1095 tcg_out_dat_imm(s, COND_AL, ARITH_AND,
1096 TCG_REG_R0, TCG_REG_R8, CPU_TLB_SIZE - 1);
1097 tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_R0, TCG_AREG0,
1098 TCG_REG_R0, SHIFT_IMM_LSL(CPU_TLB_ENTRY_BITS));
1099 /* In the
1100 * ldr r1 [r0, #(offsetof(CPUArchState, tlb_table[mem_index][0].addr_read))]
1101 * below, the offset is likely to exceed 12 bits if mem_index != 0 and
1102 * not exceed otherwise, so use an
1103 * add r0, r0, #(mem_index * sizeof *CPUArchState.tlb_table)
1104 * before.
1105 */
1106 if (mem_index)
1107 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R0, TCG_REG_R0,
1108 (mem_index << (TLB_SHIFT & 1)) |
1109 ((16 - (TLB_SHIFT >> 1)) << 8));
1110 tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R0,
1111 offsetof(CPUArchState, tlb_table[0][0].addr_read));
1112 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, TCG_REG_R1,
1113 TCG_REG_R8, SHIFT_IMM_LSL(TARGET_PAGE_BITS));
1114 /* Check alignment. */
1115 if (s_bits)
1116 tcg_out_dat_imm(s, COND_EQ, ARITH_TST,
1117 0, addr_reg, (1 << s_bits) - 1);
1118 # if TARGET_LONG_BITS == 64
1119 /* XXX: possibly we could use a block data load or writeback in
1120 * the first access. */
1121 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0,
1122 offsetof(CPUArchState, tlb_table[0][0].addr_read) + 4);
1123 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
1124 TCG_REG_R1, addr_reg2, SHIFT_IMM_LSL(0));
1125 # endif
1126 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0,
1127 offsetof(CPUArchState, tlb_table[0][0].addend));
1128
1129 switch (opc) {
1130 case 0:
1131 tcg_out_ld8_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1132 break;
1133 case 0 | 4:
1134 tcg_out_ld8s_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1135 break;
1136 case 1:
1137 tcg_out_ld16u_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1138 if (bswap) {
1139 tcg_out_bswap16(s, COND_EQ, data_reg, data_reg);
1140 }
1141 break;
1142 case 1 | 4:
1143 if (bswap) {
1144 tcg_out_ld16u_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1145 tcg_out_bswap16s(s, COND_EQ, data_reg, data_reg);
1146 } else {
1147 tcg_out_ld16s_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1148 }
1149 break;
1150 case 2:
1151 default:
1152 tcg_out_ld32_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1153 if (bswap) {
1154 tcg_out_bswap32(s, COND_EQ, data_reg, data_reg);
1155 }
1156 break;
1157 case 3:
1158 if (bswap) {
1159 tcg_out_ld32_rwb(s, COND_EQ, data_reg2, TCG_REG_R1, addr_reg);
1160 tcg_out_ld32_12(s, COND_EQ, data_reg, TCG_REG_R1, 4);
1161 tcg_out_bswap32(s, COND_EQ, data_reg2, data_reg2);
1162 tcg_out_bswap32(s, COND_EQ, data_reg, data_reg);
1163 } else {
1164 tcg_out_ld32_rwb(s, COND_EQ, data_reg, TCG_REG_R1, addr_reg);
1165 tcg_out_ld32_12(s, COND_EQ, data_reg2, TCG_REG_R1, 4);
1166 }
1167 break;
1168 }
1169
1170 label_ptr = (void *) s->code_ptr;
1171 tcg_out_b_noaddr(s, COND_EQ);
1172
1173 /* TODO: move this code to where the constants pool will be */
1174 /* Note that this code relies on the constraints we set in arm_op_defs[]
1175 * to ensure that later arguments are not passed to us in registers we
1176 * trash by moving the earlier arguments into them.
1177 */
1178 argreg = TCG_REG_R0;
1179 argreg = tcg_out_arg_reg32(s, argreg, TCG_AREG0);
1180 #if TARGET_LONG_BITS == 64
1181 argreg = tcg_out_arg_reg64(s, argreg, addr_reg, addr_reg2);
1182 #else
1183 argreg = tcg_out_arg_reg32(s, argreg, addr_reg);
1184 #endif
1185 argreg = tcg_out_arg_imm32(s, argreg, mem_index);
1186 tcg_out_call(s, (tcg_target_long) qemu_ld_helpers[s_bits]);
1187 tcg_out_arg_stacktidy(s, argreg);
1188
1189 switch (opc) {
1190 case 0 | 4:
1191 tcg_out_ext8s(s, COND_AL, data_reg, TCG_REG_R0);
1192 break;
1193 case 1 | 4:
1194 tcg_out_ext16s(s, COND_AL, data_reg, TCG_REG_R0);
1195 break;
1196 case 0:
1197 case 1:
1198 case 2:
1199 default:
1200 tcg_out_mov_reg(s, COND_AL, data_reg, TCG_REG_R0);
1201 break;
1202 case 3:
1203 tcg_out_mov_reg(s, COND_AL, data_reg, TCG_REG_R0);
1204 tcg_out_mov_reg(s, COND_AL, data_reg2, TCG_REG_R1);
1205 break;
1206 }
1207
1208 reloc_pc24(label_ptr, (tcg_target_long)s->code_ptr);
1209 #else /* !CONFIG_SOFTMMU */
1210 if (GUEST_BASE) {
1211 uint32_t offset = GUEST_BASE;
1212 int i;
1213 int rot;
1214
1215 while (offset) {
1216 i = ctz32(offset) & ~1;
1217 rot = ((32 - i) << 7) & 0xf00;
1218
1219 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R8, addr_reg,
1220 ((offset >> i) & 0xff) | rot);
1221 addr_reg = TCG_REG_R8;
1222 offset &= ~(0xff << i);
1223 }
1224 }
1225 switch (opc) {
1226 case 0:
1227 tcg_out_ld8_12(s, COND_AL, data_reg, addr_reg, 0);
1228 break;
1229 case 0 | 4:
1230 tcg_out_ld8s_8(s, COND_AL, data_reg, addr_reg, 0);
1231 break;
1232 case 1:
1233 tcg_out_ld16u_8(s, COND_AL, data_reg, addr_reg, 0);
1234 if (bswap) {
1235 tcg_out_bswap16(s, COND_AL, data_reg, data_reg);
1236 }
1237 break;
1238 case 1 | 4:
1239 if (bswap) {
1240 tcg_out_ld16u_8(s, COND_AL, data_reg, addr_reg, 0);
1241 tcg_out_bswap16s(s, COND_AL, data_reg, data_reg);
1242 } else {
1243 tcg_out_ld16s_8(s, COND_AL, data_reg, addr_reg, 0);
1244 }
1245 break;
1246 case 2:
1247 default:
1248 tcg_out_ld32_12(s, COND_AL, data_reg, addr_reg, 0);
1249 if (bswap) {
1250 tcg_out_bswap32(s, COND_AL, data_reg, data_reg);
1251 }
1252 break;
1253 case 3:
1254 /* TODO: use block load -
1255 * check that data_reg2 > data_reg or the other way */
1256 if (data_reg == addr_reg) {
1257 tcg_out_ld32_12(s, COND_AL, data_reg2, addr_reg, bswap ? 0 : 4);
1258 tcg_out_ld32_12(s, COND_AL, data_reg, addr_reg, bswap ? 4 : 0);
1259 } else {
1260 tcg_out_ld32_12(s, COND_AL, data_reg, addr_reg, bswap ? 4 : 0);
1261 tcg_out_ld32_12(s, COND_AL, data_reg2, addr_reg, bswap ? 0 : 4);
1262 }
1263 if (bswap) {
1264 tcg_out_bswap32(s, COND_AL, data_reg, data_reg);
1265 tcg_out_bswap32(s, COND_AL, data_reg2, data_reg2);
1266 }
1267 break;
1268 }
1269 #endif
1270 }
1271
1272 static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)
1273 {
1274 int addr_reg, data_reg, data_reg2, bswap;
1275 #ifdef CONFIG_SOFTMMU
1276 int mem_index, s_bits;
1277 TCGReg argreg;
1278 # if TARGET_LONG_BITS == 64
1279 int addr_reg2;
1280 # endif
1281 uint32_t *label_ptr;
1282 #endif
1283
1284 #ifdef TARGET_WORDS_BIGENDIAN
1285 bswap = 1;
1286 #else
1287 bswap = 0;
1288 #endif
1289 data_reg = *args++;
1290 if (opc == 3)
1291 data_reg2 = *args++;
1292 else
1293 data_reg2 = 0; /* suppress warning */
1294 addr_reg = *args++;
1295 #ifdef CONFIG_SOFTMMU
1296 # if TARGET_LONG_BITS == 64
1297 addr_reg2 = *args++;
1298 # endif
1299 mem_index = *args;
1300 s_bits = opc & 3;
1301
1302 /* Should generate something like the following:
1303 * shr r8, addr_reg, #TARGET_PAGE_BITS
1304 * and r0, r8, #(CPU_TLB_SIZE - 1) @ Assumption: CPU_TLB_BITS <= 8
1305 * add r0, env, r0 lsl #CPU_TLB_ENTRY_BITS
1306 */
1307 tcg_out_dat_reg(s, COND_AL, ARITH_MOV,
1308 TCG_REG_R8, 0, addr_reg, SHIFT_IMM_LSR(TARGET_PAGE_BITS));
1309 tcg_out_dat_imm(s, COND_AL, ARITH_AND,
1310 TCG_REG_R0, TCG_REG_R8, CPU_TLB_SIZE - 1);
1311 tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_R0,
1312 TCG_AREG0, TCG_REG_R0, SHIFT_IMM_LSL(CPU_TLB_ENTRY_BITS));
1313 /* In the
1314 * ldr r1 [r0, #(offsetof(CPUArchState, tlb_table[mem_index][0].addr_write))]
1315 * below, the offset is likely to exceed 12 bits if mem_index != 0 and
1316 * not exceed otherwise, so use an
1317 * add r0, r0, #(mem_index * sizeof *CPUArchState.tlb_table)
1318 * before.
1319 */
1320 if (mem_index)
1321 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R0, TCG_REG_R0,
1322 (mem_index << (TLB_SHIFT & 1)) |
1323 ((16 - (TLB_SHIFT >> 1)) << 8));
1324 tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R0,
1325 offsetof(CPUArchState, tlb_table[0][0].addr_write));
1326 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, TCG_REG_R1,
1327 TCG_REG_R8, SHIFT_IMM_LSL(TARGET_PAGE_BITS));
1328 /* Check alignment. */
1329 if (s_bits)
1330 tcg_out_dat_imm(s, COND_EQ, ARITH_TST,
1331 0, addr_reg, (1 << s_bits) - 1);
1332 # if TARGET_LONG_BITS == 64
1333 /* XXX: possibly we could use a block data load or writeback in
1334 * the first access. */
1335 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0,
1336 offsetof(CPUArchState, tlb_table[0][0].addr_write) + 4);
1337 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
1338 TCG_REG_R1, addr_reg2, SHIFT_IMM_LSL(0));
1339 # endif
1340 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0,
1341 offsetof(CPUArchState, tlb_table[0][0].addend));
1342
1343 switch (opc) {
1344 case 0:
1345 tcg_out_st8_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1346 break;
1347 case 1:
1348 if (bswap) {
1349 tcg_out_bswap16(s, COND_EQ, TCG_REG_R0, data_reg);
1350 tcg_out_st16_r(s, COND_EQ, TCG_REG_R0, addr_reg, TCG_REG_R1);
1351 } else {
1352 tcg_out_st16_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1353 }
1354 break;
1355 case 2:
1356 default:
1357 if (bswap) {
1358 tcg_out_bswap32(s, COND_EQ, TCG_REG_R0, data_reg);
1359 tcg_out_st32_r(s, COND_EQ, TCG_REG_R0, addr_reg, TCG_REG_R1);
1360 } else {
1361 tcg_out_st32_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1362 }
1363 break;
1364 case 3:
1365 if (bswap) {
1366 tcg_out_bswap32(s, COND_EQ, TCG_REG_R0, data_reg2);
1367 tcg_out_st32_rwb(s, COND_EQ, TCG_REG_R0, TCG_REG_R1, addr_reg);
1368 tcg_out_bswap32(s, COND_EQ, TCG_REG_R0, data_reg);
1369 tcg_out_st32_12(s, COND_EQ, TCG_REG_R0, TCG_REG_R1, 4);
1370 } else {
1371 tcg_out_st32_rwb(s, COND_EQ, data_reg, TCG_REG_R1, addr_reg);
1372 tcg_out_st32_12(s, COND_EQ, data_reg2, TCG_REG_R1, 4);
1373 }
1374 break;
1375 }
1376
1377 label_ptr = (void *) s->code_ptr;
1378 tcg_out_b_noaddr(s, COND_EQ);
1379
1380 /* TODO: move this code to where the constants pool will be */
1381 /* Note that this code relies on the constraints we set in arm_op_defs[]
1382 * to ensure that later arguments are not passed to us in registers we
1383 * trash by moving the earlier arguments into them.
1384 */
1385 argreg = TCG_REG_R0;
1386 argreg = tcg_out_arg_reg32(s, argreg, TCG_AREG0);
1387 #if TARGET_LONG_BITS == 64
1388 argreg = tcg_out_arg_reg64(s, argreg, addr_reg, addr_reg2);
1389 #else
1390 argreg = tcg_out_arg_reg32(s, argreg, addr_reg);
1391 #endif
1392
1393 switch (opc) {
1394 case 0:
1395 argreg = tcg_out_arg_reg8(s, argreg, data_reg);
1396 break;
1397 case 1:
1398 argreg = tcg_out_arg_reg16(s, argreg, data_reg);
1399 break;
1400 case 2:
1401 argreg = tcg_out_arg_reg32(s, argreg, data_reg);
1402 break;
1403 case 3:
1404 argreg = tcg_out_arg_reg64(s, argreg, data_reg, data_reg2);
1405 break;
1406 }
1407
1408 argreg = tcg_out_arg_imm32(s, argreg, mem_index);
1409 tcg_out_call(s, (tcg_target_long) qemu_st_helpers[s_bits]);
1410 tcg_out_arg_stacktidy(s, argreg);
1411
1412 reloc_pc24(label_ptr, (tcg_target_long)s->code_ptr);
1413 #else /* !CONFIG_SOFTMMU */
1414 if (GUEST_BASE) {
1415 uint32_t offset = GUEST_BASE;
1416 int i;
1417 int rot;
1418
1419 while (offset) {
1420 i = ctz32(offset) & ~1;
1421 rot = ((32 - i) << 7) & 0xf00;
1422
1423 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R1, addr_reg,
1424 ((offset >> i) & 0xff) | rot);
1425 addr_reg = TCG_REG_R1;
1426 offset &= ~(0xff << i);
1427 }
1428 }
1429 switch (opc) {
1430 case 0:
1431 tcg_out_st8_12(s, COND_AL, data_reg, addr_reg, 0);
1432 break;
1433 case 1:
1434 if (bswap) {
1435 tcg_out_bswap16(s, COND_AL, TCG_REG_R0, data_reg);
1436 tcg_out_st16_8(s, COND_AL, TCG_REG_R0, addr_reg, 0);
1437 } else {
1438 tcg_out_st16_8(s, COND_AL, data_reg, addr_reg, 0);
1439 }
1440 break;
1441 case 2:
1442 default:
1443 if (bswap) {
1444 tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg);
1445 tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addr_reg, 0);
1446 } else {
1447 tcg_out_st32_12(s, COND_AL, data_reg, addr_reg, 0);
1448 }
1449 break;
1450 case 3:
1451 /* TODO: use block store -
1452 * check that data_reg2 > data_reg or the other way */
1453 if (bswap) {
1454 tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg2);
1455 tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addr_reg, 0);
1456 tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg);
1457 tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addr_reg, 4);
1458 } else {
1459 tcg_out_st32_12(s, COND_AL, data_reg, addr_reg, 0);
1460 tcg_out_st32_12(s, COND_AL, data_reg2, addr_reg, 4);
1461 }
1462 break;
1463 }
1464 #endif
1465 }
1466
1467 static uint8_t *tb_ret_addr;
1468
1469 static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
1470 const TCGArg *args, const int *const_args)
1471 {
1472 int c;
1473
1474 switch (opc) {
1475 case INDEX_op_exit_tb:
1476 {
1477 uint8_t *ld_ptr = s->code_ptr;
1478 if (args[0] >> 8)
1479 tcg_out_ld32_12(s, COND_AL, TCG_REG_R0, TCG_REG_PC, 0);
1480 else
1481 tcg_out_dat_imm(s, COND_AL, ARITH_MOV, TCG_REG_R0, 0, args[0]);
1482 tcg_out_goto(s, COND_AL, (tcg_target_ulong) tb_ret_addr);
1483 if (args[0] >> 8) {
1484 *ld_ptr = (uint8_t) (s->code_ptr - ld_ptr) - 8;
1485 tcg_out32(s, args[0]);
1486 }
1487 }
1488 break;
1489 case INDEX_op_goto_tb:
1490 if (s->tb_jmp_offset) {
1491 /* Direct jump method */
1492 #if defined(USE_DIRECT_JUMP)
1493 s->tb_jmp_offset[args[0]] = s->code_ptr - s->code_buf;
1494 tcg_out_b_noaddr(s, COND_AL);
1495 #else
1496 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
1497 s->tb_jmp_offset[args[0]] = s->code_ptr - s->code_buf;
1498 tcg_out32(s, 0);
1499 #endif
1500 } else {
1501 /* Indirect jump method */
1502 #if 1
1503 c = (int) (s->tb_next + args[0]) - ((int) s->code_ptr + 8);
1504 if (c > 0xfff || c < -0xfff) {
1505 tcg_out_movi32(s, COND_AL, TCG_REG_R0,
1506 (tcg_target_long) (s->tb_next + args[0]));
1507 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, 0);
1508 } else
1509 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, c);
1510 #else
1511 tcg_out_ld32_12(s, COND_AL, TCG_REG_R0, TCG_REG_PC, 0);
1512 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, 0);
1513 tcg_out32(s, (tcg_target_long) (s->tb_next + args[0]));
1514 #endif
1515 }
1516 s->tb_next_offset[args[0]] = s->code_ptr - s->code_buf;
1517 break;
1518 case INDEX_op_call:
1519 if (const_args[0])
1520 tcg_out_call(s, args[0]);
1521 else
1522 tcg_out_callr(s, COND_AL, args[0]);
1523 break;
1524 case INDEX_op_br:
1525 tcg_out_goto_label(s, COND_AL, args[0]);
1526 break;
1527
1528 case INDEX_op_ld8u_i32:
1529 tcg_out_ld8u(s, COND_AL, args[0], args[1], args[2]);
1530 break;
1531 case INDEX_op_ld8s_i32:
1532 tcg_out_ld8s(s, COND_AL, args[0], args[1], args[2]);
1533 break;
1534 case INDEX_op_ld16u_i32:
1535 tcg_out_ld16u(s, COND_AL, args[0], args[1], args[2]);
1536 break;
1537 case INDEX_op_ld16s_i32:
1538 tcg_out_ld16s(s, COND_AL, args[0], args[1], args[2]);
1539 break;
1540 case INDEX_op_ld_i32:
1541 tcg_out_ld32u(s, COND_AL, args[0], args[1], args[2]);
1542 break;
1543 case INDEX_op_st8_i32:
1544 tcg_out_st8(s, COND_AL, args[0], args[1], args[2]);
1545 break;
1546 case INDEX_op_st16_i32:
1547 tcg_out_st16(s, COND_AL, args[0], args[1], args[2]);
1548 break;
1549 case INDEX_op_st_i32:
1550 tcg_out_st32(s, COND_AL, args[0], args[1], args[2]);
1551 break;
1552
1553 case INDEX_op_mov_i32:
1554 tcg_out_dat_reg(s, COND_AL, ARITH_MOV,
1555 args[0], 0, args[1], SHIFT_IMM_LSL(0));
1556 break;
1557 case INDEX_op_movi_i32:
1558 tcg_out_movi32(s, COND_AL, args[0], args[1]);
1559 break;
1560 case INDEX_op_add_i32:
1561 c = ARITH_ADD;
1562 goto gen_arith;
1563 case INDEX_op_sub_i32:
1564 c = ARITH_SUB;
1565 goto gen_arith;
1566 case INDEX_op_and_i32:
1567 c = ARITH_AND;
1568 goto gen_arith;
1569 case INDEX_op_andc_i32:
1570 c = ARITH_BIC;
1571 goto gen_arith;
1572 case INDEX_op_or_i32:
1573 c = ARITH_ORR;
1574 goto gen_arith;
1575 case INDEX_op_xor_i32:
1576 c = ARITH_EOR;
1577 /* Fall through. */
1578 gen_arith:
1579 if (const_args[2]) {
1580 int rot;
1581 rot = encode_imm(args[2]);
1582 tcg_out_dat_imm(s, COND_AL, c,
1583 args[0], args[1], rotl(args[2], rot) | (rot << 7));
1584 } else
1585 tcg_out_dat_reg(s, COND_AL, c,
1586 args[0], args[1], args[2], SHIFT_IMM_LSL(0));
1587 break;
1588 case INDEX_op_add2_i32:
1589 tcg_out_dat_reg2(s, COND_AL, ARITH_ADD, ARITH_ADC,
1590 args[0], args[1], args[2], args[3],
1591 args[4], args[5], SHIFT_IMM_LSL(0));
1592 break;
1593 case INDEX_op_sub2_i32:
1594 tcg_out_dat_reg2(s, COND_AL, ARITH_SUB, ARITH_SBC,
1595 args[0], args[1], args[2], args[3],
1596 args[4], args[5], SHIFT_IMM_LSL(0));
1597 break;
1598 case INDEX_op_neg_i32:
1599 tcg_out_dat_imm(s, COND_AL, ARITH_RSB, args[0], args[1], 0);
1600 break;
1601 case INDEX_op_not_i32:
1602 tcg_out_dat_reg(s, COND_AL,
1603 ARITH_MVN, args[0], 0, args[1], SHIFT_IMM_LSL(0));
1604 break;
1605 case INDEX_op_mul_i32:
1606 tcg_out_mul32(s, COND_AL, args[0], args[1], args[2]);
1607 break;
1608 case INDEX_op_mulu2_i32:
1609 tcg_out_umull32(s, COND_AL, args[0], args[1], args[2], args[3]);
1610 break;
1611 /* XXX: Perhaps args[2] & 0x1f is wrong */
1612 case INDEX_op_shl_i32:
1613 c = const_args[2] ?
1614 SHIFT_IMM_LSL(args[2] & 0x1f) : SHIFT_REG_LSL(args[2]);
1615 goto gen_shift32;
1616 case INDEX_op_shr_i32:
1617 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_LSR(args[2] & 0x1f) :
1618 SHIFT_IMM_LSL(0) : SHIFT_REG_LSR(args[2]);
1619 goto gen_shift32;
1620 case INDEX_op_sar_i32:
1621 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ASR(args[2] & 0x1f) :
1622 SHIFT_IMM_LSL(0) : SHIFT_REG_ASR(args[2]);
1623 goto gen_shift32;
1624 case INDEX_op_rotr_i32:
1625 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ROR(args[2] & 0x1f) :
1626 SHIFT_IMM_LSL(0) : SHIFT_REG_ROR(args[2]);
1627 /* Fall through. */
1628 gen_shift32:
1629 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1], c);
1630 break;
1631
1632 case INDEX_op_rotl_i32:
1633 if (const_args[2]) {
1634 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
1635 ((0x20 - args[2]) & 0x1f) ?
1636 SHIFT_IMM_ROR((0x20 - args[2]) & 0x1f) :
1637 SHIFT_IMM_LSL(0));
1638 } else {
1639 tcg_out_dat_imm(s, COND_AL, ARITH_RSB, TCG_REG_R8, args[1], 0x20);
1640 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
1641 SHIFT_REG_ROR(TCG_REG_R8));
1642 }
1643 break;
1644
1645 case INDEX_op_brcond_i32:
1646 if (const_args[1]) {
1647 int rot;
1648 rot = encode_imm(args[1]);
1649 tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0,
1650 args[0], rotl(args[1], rot) | (rot << 7));
1651 } else {
1652 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1653 args[0], args[1], SHIFT_IMM_LSL(0));
1654 }
1655 tcg_out_goto_label(s, tcg_cond_to_arm_cond[args[2]], args[3]);
1656 break;
1657 case INDEX_op_brcond2_i32:
1658 /* The resulting conditions are:
1659 * TCG_COND_EQ --> a0 == a2 && a1 == a3,
1660 * TCG_COND_NE --> (a0 != a2 && a1 == a3) || a1 != a3,
1661 * TCG_COND_LT(U) --> (a0 < a2 && a1 == a3) || a1 < a3,
1662 * TCG_COND_GE(U) --> (a0 >= a2 && a1 == a3) || (a1 >= a3 && a1 != a3),
1663 * TCG_COND_LE(U) --> (a0 <= a2 && a1 == a3) || (a1 <= a3 && a1 != a3),
1664 * TCG_COND_GT(U) --> (a0 > a2 && a1 == a3) || a1 > a3,
1665 */
1666 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1667 args[1], args[3], SHIFT_IMM_LSL(0));
1668 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
1669 args[0], args[2], SHIFT_IMM_LSL(0));
1670 tcg_out_goto_label(s, tcg_cond_to_arm_cond[args[4]], args[5]);
1671 break;
1672 case INDEX_op_setcond_i32:
1673 if (const_args[2]) {
1674 int rot;
1675 rot = encode_imm(args[2]);
1676 tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0,
1677 args[1], rotl(args[2], rot) | (rot << 7));
1678 } else {
1679 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1680 args[1], args[2], SHIFT_IMM_LSL(0));
1681 }
1682 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[args[3]],
1683 ARITH_MOV, args[0], 0, 1);
1684 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(args[3])],
1685 ARITH_MOV, args[0], 0, 0);
1686 break;
1687 case INDEX_op_setcond2_i32:
1688 /* See brcond2_i32 comment */
1689 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1690 args[2], args[4], SHIFT_IMM_LSL(0));
1691 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
1692 args[1], args[3], SHIFT_IMM_LSL(0));
1693 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[args[5]],
1694 ARITH_MOV, args[0], 0, 1);
1695 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(args[5])],
1696 ARITH_MOV, args[0], 0, 0);
1697 break;
1698
1699 case INDEX_op_qemu_ld8u:
1700 tcg_out_qemu_ld(s, args, 0);
1701 break;
1702 case INDEX_op_qemu_ld8s:
1703 tcg_out_qemu_ld(s, args, 0 | 4);
1704 break;
1705 case INDEX_op_qemu_ld16u:
1706 tcg_out_qemu_ld(s, args, 1);
1707 break;
1708 case INDEX_op_qemu_ld16s:
1709 tcg_out_qemu_ld(s, args, 1 | 4);
1710 break;
1711 case INDEX_op_qemu_ld32:
1712 tcg_out_qemu_ld(s, args, 2);
1713 break;
1714 case INDEX_op_qemu_ld64:
1715 tcg_out_qemu_ld(s, args, 3);
1716 break;
1717
1718 case INDEX_op_qemu_st8:
1719 tcg_out_qemu_st(s, args, 0);
1720 break;
1721 case INDEX_op_qemu_st16:
1722 tcg_out_qemu_st(s, args, 1);
1723 break;
1724 case INDEX_op_qemu_st32:
1725 tcg_out_qemu_st(s, args, 2);
1726 break;
1727 case INDEX_op_qemu_st64:
1728 tcg_out_qemu_st(s, args, 3);
1729 break;
1730
1731 case INDEX_op_bswap16_i32:
1732 tcg_out_bswap16(s, COND_AL, args[0], args[1]);
1733 break;
1734 case INDEX_op_bswap32_i32:
1735 tcg_out_bswap32(s, COND_AL, args[0], args[1]);
1736 break;
1737
1738 case INDEX_op_ext8s_i32:
1739 tcg_out_ext8s(s, COND_AL, args[0], args[1]);
1740 break;
1741 case INDEX_op_ext16s_i32:
1742 tcg_out_ext16s(s, COND_AL, args[0], args[1]);
1743 break;
1744 case INDEX_op_ext16u_i32:
1745 tcg_out_ext16u(s, COND_AL, args[0], args[1]);
1746 break;
1747
1748 default:
1749 tcg_abort();
1750 }
1751 }
1752
1753 static const TCGTargetOpDef arm_op_defs[] = {
1754 { INDEX_op_exit_tb, { } },
1755 { INDEX_op_goto_tb, { } },
1756 { INDEX_op_call, { "ri" } },
1757 { INDEX_op_br, { } },
1758
1759 { INDEX_op_mov_i32, { "r", "r" } },
1760 { INDEX_op_movi_i32, { "r" } },
1761
1762 { INDEX_op_ld8u_i32, { "r", "r" } },
1763 { INDEX_op_ld8s_i32, { "r", "r" } },
1764 { INDEX_op_ld16u_i32, { "r", "r" } },
1765 { INDEX_op_ld16s_i32, { "r", "r" } },
1766 { INDEX_op_ld_i32, { "r", "r" } },
1767 { INDEX_op_st8_i32, { "r", "r" } },
1768 { INDEX_op_st16_i32, { "r", "r" } },
1769 { INDEX_op_st_i32, { "r", "r" } },
1770
1771 /* TODO: "r", "r", "ri" */
1772 { INDEX_op_add_i32, { "r", "r", "rI" } },
1773 { INDEX_op_sub_i32, { "r", "r", "rI" } },
1774 { INDEX_op_mul_i32, { "r", "r", "r" } },
1775 { INDEX_op_mulu2_i32, { "r", "r", "r", "r" } },
1776 { INDEX_op_and_i32, { "r", "r", "rI" } },
1777 { INDEX_op_andc_i32, { "r", "r", "rI" } },
1778 { INDEX_op_or_i32, { "r", "r", "rI" } },
1779 { INDEX_op_xor_i32, { "r", "r", "rI" } },
1780 { INDEX_op_neg_i32, { "r", "r" } },
1781 { INDEX_op_not_i32, { "r", "r" } },
1782
1783 { INDEX_op_shl_i32, { "r", "r", "ri" } },
1784 { INDEX_op_shr_i32, { "r", "r", "ri" } },
1785 { INDEX_op_sar_i32, { "r", "r", "ri" } },
1786 { INDEX_op_rotl_i32, { "r", "r", "ri" } },
1787 { INDEX_op_rotr_i32, { "r", "r", "ri" } },
1788
1789 { INDEX_op_brcond_i32, { "r", "rI" } },
1790 { INDEX_op_setcond_i32, { "r", "r", "rI" } },
1791
1792 /* TODO: "r", "r", "r", "r", "ri", "ri" */
1793 { INDEX_op_add2_i32, { "r", "r", "r", "r", "r", "r" } },
1794 { INDEX_op_sub2_i32, { "r", "r", "r", "r", "r", "r" } },
1795 { INDEX_op_brcond2_i32, { "r", "r", "r", "r" } },
1796 { INDEX_op_setcond2_i32, { "r", "r", "r", "r", "r" } },
1797
1798 #if TARGET_LONG_BITS == 32
1799 { INDEX_op_qemu_ld8u, { "r", "l" } },
1800 { INDEX_op_qemu_ld8s, { "r", "l" } },
1801 { INDEX_op_qemu_ld16u, { "r", "l" } },
1802 { INDEX_op_qemu_ld16s, { "r", "l" } },
1803 { INDEX_op_qemu_ld32, { "r", "l" } },
1804 { INDEX_op_qemu_ld64, { "L", "L", "l" } },
1805
1806 { INDEX_op_qemu_st8, { "s", "s" } },
1807 { INDEX_op_qemu_st16, { "s", "s" } },
1808 { INDEX_op_qemu_st32, { "s", "s" } },
1809 { INDEX_op_qemu_st64, { "S", "S", "s" } },
1810 #else
1811 { INDEX_op_qemu_ld8u, { "r", "l", "l" } },
1812 { INDEX_op_qemu_ld8s, { "r", "l", "l" } },
1813 { INDEX_op_qemu_ld16u, { "r", "l", "l" } },
1814 { INDEX_op_qemu_ld16s, { "r", "l", "l" } },
1815 { INDEX_op_qemu_ld32, { "r", "l", "l" } },
1816 { INDEX_op_qemu_ld64, { "L", "L", "l", "l" } },
1817
1818 { INDEX_op_qemu_st8, { "s", "s", "s" } },
1819 { INDEX_op_qemu_st16, { "s", "s", "s" } },
1820 { INDEX_op_qemu_st32, { "s", "s", "s" } },
1821 { INDEX_op_qemu_st64, { "S", "S", "s", "s" } },
1822 #endif
1823
1824 { INDEX_op_bswap16_i32, { "r", "r" } },
1825 { INDEX_op_bswap32_i32, { "r", "r" } },
1826
1827 { INDEX_op_ext8s_i32, { "r", "r" } },
1828 { INDEX_op_ext16s_i32, { "r", "r" } },
1829 { INDEX_op_ext16u_i32, { "r", "r" } },
1830
1831 { -1 },
1832 };
1833
1834 static void tcg_target_init(TCGContext *s)
1835 {
1836 #if !defined(CONFIG_USER_ONLY)
1837 /* fail safe */
1838 if ((1 << CPU_TLB_ENTRY_BITS) != sizeof(CPUTLBEntry))
1839 tcg_abort();
1840 #endif
1841
1842 tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I32], 0, 0xffff);
1843 tcg_regset_set32(tcg_target_call_clobber_regs, 0,
1844 (1 << TCG_REG_R0) |
1845 (1 << TCG_REG_R1) |
1846 (1 << TCG_REG_R2) |
1847 (1 << TCG_REG_R3) |
1848 (1 << TCG_REG_R12) |
1849 (1 << TCG_REG_R14));
1850
1851 tcg_regset_clear(s->reserved_regs);
1852 tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
1853 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R8);
1854 tcg_regset_set_reg(s->reserved_regs, TCG_REG_PC);
1855
1856 tcg_add_target_add_op_defs(arm_op_defs);
1857 tcg_set_frame(s, TCG_AREG0, offsetof(CPUArchState, temp_buf),
1858 CPU_TEMP_BUF_NLONGS * sizeof(long));
1859 }
1860
1861 static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
1862 TCGReg arg1, tcg_target_long arg2)
1863 {
1864 tcg_out_ld32u(s, COND_AL, arg, arg1, arg2);
1865 }
1866
1867 static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
1868 TCGReg arg1, tcg_target_long arg2)
1869 {
1870 tcg_out_st32(s, COND_AL, arg, arg1, arg2);
1871 }
1872
1873 static inline void tcg_out_mov(TCGContext *s, TCGType type,
1874 TCGReg ret, TCGReg arg)
1875 {
1876 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, ret, 0, arg, SHIFT_IMM_LSL(0));
1877 }
1878
1879 static inline void tcg_out_movi(TCGContext *s, TCGType type,
1880 TCGReg ret, tcg_target_long arg)
1881 {
1882 tcg_out_movi32(s, COND_AL, ret, arg);
1883 }
1884
1885 static void tcg_target_qemu_prologue(TCGContext *s)
1886 {
1887 /* Calling convention requires us to save r4-r11 and lr;
1888 * save also r12 to maintain stack 8-alignment.
1889 */
1890
1891 /* stmdb sp!, { r4 - r12, lr } */
1892 tcg_out32(s, (COND_AL << 28) | 0x092d5ff0);
1893
1894 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
1895
1896 tcg_out_bx(s, COND_AL, tcg_target_call_iarg_regs[1]);
1897 tb_ret_addr = s->code_ptr;
1898
1899 /* ldmia sp!, { r4 - r12, pc } */
1900 tcg_out32(s, (COND_AL << 28) | 0x08bd9ff0);
1901 }
AR7 emulation for QEMU