?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Add linux-headers to QEMU_INCLUDES
[qemu/ar7.git] / tcg / arm / tcg-target.c
blob94c6ca43aa71cc9de3d912e24f35fd3d5d986d12
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_dat_rI(TCGContext *s, int cond, int opc, TCGArg dst,
471 TCGArg lhs, TCGArg rhs, int rhs_is_const)
472 {
473 /* Emit either the reg,imm or reg,reg form of a data-processing insn.
474 * rhs must satisfy the "rI" constraint.
475 */
476 if (rhs_is_const) {
477 int rot = encode_imm(rhs);
478 assert(rot >= 0);
479 tcg_out_dat_imm(s, cond, opc, dst, lhs, rotl(rhs, rot) | (rot << 7));
480 } else {
481 tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
482 }
483 }
484
485 static inline void tcg_out_mul32(TCGContext *s,
486 int cond, int rd, int rs, int rm)
487 {
488 if (rd != rm)
489 tcg_out32(s, (cond << 28) | (rd << 16) | (0 << 12) |
490 (rs << 8) | 0x90 | rm);
491 else if (rd != rs)
492 tcg_out32(s, (cond << 28) | (rd << 16) | (0 << 12) |
493 (rm << 8) | 0x90 | rs);
494 else {
495 tcg_out32(s, (cond << 28) | ( 8 << 16) | (0 << 12) |
496 (rs << 8) | 0x90 | rm);
497 tcg_out_dat_reg(s, cond, ARITH_MOV,
498 rd, 0, TCG_REG_R8, SHIFT_IMM_LSL(0));
499 }
500 }
501
502 static inline void tcg_out_umull32(TCGContext *s,
503 int cond, int rd0, int rd1, int rs, int rm)
504 {
505 if (rd0 != rm && rd1 != rm)
506 tcg_out32(s, (cond << 28) | 0x800090 |
507 (rd1 << 16) | (rd0 << 12) | (rs << 8) | rm);
508 else if (rd0 != rs && rd1 != rs)
509 tcg_out32(s, (cond << 28) | 0x800090 |
510 (rd1 << 16) | (rd0 << 12) | (rm << 8) | rs);
511 else {
512 tcg_out_dat_reg(s, cond, ARITH_MOV,
513 TCG_REG_R8, 0, rm, SHIFT_IMM_LSL(0));
514 tcg_out32(s, (cond << 28) | 0x800098 |
515 (rd1 << 16) | (rd0 << 12) | (rs << 8));
516 }
517 }
518
519 static inline void tcg_out_smull32(TCGContext *s,
520 int cond, int rd0, int rd1, int rs, int rm)
521 {
522 if (rd0 != rm && rd1 != rm)
523 tcg_out32(s, (cond << 28) | 0xc00090 |
524 (rd1 << 16) | (rd0 << 12) | (rs << 8) | rm);
525 else if (rd0 != rs && rd1 != rs)
526 tcg_out32(s, (cond << 28) | 0xc00090 |
527 (rd1 << 16) | (rd0 << 12) | (rm << 8) | rs);
528 else {
529 tcg_out_dat_reg(s, cond, ARITH_MOV,
530 TCG_REG_R8, 0, rm, SHIFT_IMM_LSL(0));
531 tcg_out32(s, (cond << 28) | 0xc00098 |
532 (rd1 << 16) | (rd0 << 12) | (rs << 8));
533 }
534 }
535
536 static inline void tcg_out_ext8s(TCGContext *s, int cond,
537 int rd, int rn)
538 {
539 if (use_armv6_instructions) {
540 /* sxtb */
541 tcg_out32(s, 0x06af0070 | (cond << 28) | (rd << 12) | rn);
542 } else {
543 tcg_out_dat_reg(s, cond, ARITH_MOV,
544 rd, 0, rn, SHIFT_IMM_LSL(24));
545 tcg_out_dat_reg(s, cond, ARITH_MOV,
546 rd, 0, rd, SHIFT_IMM_ASR(24));
547 }
548 }
549
550 static inline void tcg_out_ext8u(TCGContext *s, int cond,
551 int rd, int rn)
552 {
553 tcg_out_dat_imm(s, cond, ARITH_AND, rd, rn, 0xff);
554 }
555
556 static inline void tcg_out_ext16s(TCGContext *s, int cond,
557 int rd, int rn)
558 {
559 if (use_armv6_instructions) {
560 /* sxth */
561 tcg_out32(s, 0x06bf0070 | (cond << 28) | (rd << 12) | rn);
562 } else {
563 tcg_out_dat_reg(s, cond, ARITH_MOV,
564 rd, 0, rn, SHIFT_IMM_LSL(16));
565 tcg_out_dat_reg(s, cond, ARITH_MOV,
566 rd, 0, rd, SHIFT_IMM_ASR(16));
567 }
568 }
569
570 static inline void tcg_out_ext16u(TCGContext *s, int cond,
571 int rd, int rn)
572 {
573 if (use_armv6_instructions) {
574 /* uxth */
575 tcg_out32(s, 0x06ff0070 | (cond << 28) | (rd << 12) | rn);
576 } else {
577 tcg_out_dat_reg(s, cond, ARITH_MOV,
578 rd, 0, rn, SHIFT_IMM_LSL(16));
579 tcg_out_dat_reg(s, cond, ARITH_MOV,
580 rd, 0, rd, SHIFT_IMM_LSR(16));
581 }
582 }
583
584 static inline void tcg_out_bswap16s(TCGContext *s, int cond, int rd, int rn)
585 {
586 if (use_armv6_instructions) {
587 /* revsh */
588 tcg_out32(s, 0x06ff0fb0 | (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_ASR(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_bswap16(TCGContext *s, int cond, int rd, int rn)
600 {
601 if (use_armv6_instructions) {
602 /* rev16 */
603 tcg_out32(s, 0x06bf0fb0 | (cond << 28) | (rd << 12) | rn);
604 } else {
605 tcg_out_dat_reg(s, cond, ARITH_MOV,
606 TCG_REG_R8, 0, rn, SHIFT_IMM_LSL(24));
607 tcg_out_dat_reg(s, cond, ARITH_MOV,
608 TCG_REG_R8, 0, TCG_REG_R8, SHIFT_IMM_LSR(16));
609 tcg_out_dat_reg(s, cond, ARITH_ORR,
610 rd, TCG_REG_R8, rn, SHIFT_IMM_LSR(8));
611 }
612 }
613
614 /* swap the two low bytes assuming that the two high input bytes and the
615 two high output bit can hold any value. */
616 static inline void tcg_out_bswap16st(TCGContext *s, int cond, int rd, int rn)
617 {
618 if (use_armv6_instructions) {
619 /* rev16 */
620 tcg_out32(s, 0x06bf0fb0 | (cond << 28) | (rd << 12) | rn);
621 } else {
622 tcg_out_dat_reg(s, cond, ARITH_MOV,
623 TCG_REG_R8, 0, rn, SHIFT_IMM_LSR(8));
624 tcg_out_dat_imm(s, cond, ARITH_AND, TCG_REG_R8, TCG_REG_R8, 0xff);
625 tcg_out_dat_reg(s, cond, ARITH_ORR,
626 rd, TCG_REG_R8, rn, SHIFT_IMM_LSL(8));
627 }
628 }
629
630 static inline void tcg_out_bswap32(TCGContext *s, int cond, int rd, int rn)
631 {
632 if (use_armv6_instructions) {
633 /* rev */
634 tcg_out32(s, 0x06bf0f30 | (cond << 28) | (rd << 12) | rn);
635 } else {
636 tcg_out_dat_reg(s, cond, ARITH_EOR,
637 TCG_REG_R8, rn, rn, SHIFT_IMM_ROR(16));
638 tcg_out_dat_imm(s, cond, ARITH_BIC,
639 TCG_REG_R8, TCG_REG_R8, 0xff | 0x800);
640 tcg_out_dat_reg(s, cond, ARITH_MOV,
641 rd, 0, rn, SHIFT_IMM_ROR(8));
642 tcg_out_dat_reg(s, cond, ARITH_EOR,
643 rd, rd, TCG_REG_R8, SHIFT_IMM_LSR(8));
644 }
645 }
646
647 static inline void tcg_out_ld32_12(TCGContext *s, int cond,
648 int rd, int rn, tcg_target_long im)
649 {
650 if (im >= 0)
651 tcg_out32(s, (cond << 28) | 0x05900000 |
652 (rn << 16) | (rd << 12) | (im & 0xfff));
653 else
654 tcg_out32(s, (cond << 28) | 0x05100000 |
655 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
656 }
657
658 /* Offset pre-increment with base writeback. */
659 static inline void tcg_out_ld32_12wb(TCGContext *s, int cond,
660 int rd, int rn, tcg_target_long im)
661 {
662 /* ldr with writeback and both register equals is UNPREDICTABLE */
663 assert(rd != rn);
664
665 if (im >= 0) {
666 tcg_out32(s, (cond << 28) | 0x05b00000 |
667 (rn << 16) | (rd << 12) | (im & 0xfff));
668 } else {
669 tcg_out32(s, (cond << 28) | 0x05300000 |
670 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
671 }
672 }
673
674 static inline void tcg_out_st32_12(TCGContext *s, int cond,
675 int rd, int rn, tcg_target_long im)
676 {
677 if (im >= 0)
678 tcg_out32(s, (cond << 28) | 0x05800000 |
679 (rn << 16) | (rd << 12) | (im & 0xfff));
680 else
681 tcg_out32(s, (cond << 28) | 0x05000000 |
682 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
683 }
684
685 static inline void tcg_out_ld32_r(TCGContext *s, int cond,
686 int rd, int rn, int rm)
687 {
688 tcg_out32(s, (cond << 28) | 0x07900000 |
689 (rn << 16) | (rd << 12) | rm);
690 }
691
692 static inline void tcg_out_st32_r(TCGContext *s, int cond,
693 int rd, int rn, int rm)
694 {
695 tcg_out32(s, (cond << 28) | 0x07800000 |
696 (rn << 16) | (rd << 12) | rm);
697 }
698
699 /* Register pre-increment with base writeback. */
700 static inline void tcg_out_ld32_rwb(TCGContext *s, int cond,
701 int rd, int rn, int rm)
702 {
703 tcg_out32(s, (cond << 28) | 0x07b00000 |
704 (rn << 16) | (rd << 12) | rm);
705 }
706
707 static inline void tcg_out_st32_rwb(TCGContext *s, int cond,
708 int rd, int rn, int rm)
709 {
710 tcg_out32(s, (cond << 28) | 0x07a00000 |
711 (rn << 16) | (rd << 12) | rm);
712 }
713
714 static inline void tcg_out_ld16u_8(TCGContext *s, int cond,
715 int rd, int rn, tcg_target_long im)
716 {
717 if (im >= 0)
718 tcg_out32(s, (cond << 28) | 0x01d000b0 |
719 (rn << 16) | (rd << 12) |
720 ((im & 0xf0) << 4) | (im & 0xf));
721 else
722 tcg_out32(s, (cond << 28) | 0x015000b0 |
723 (rn << 16) | (rd << 12) |
724 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
725 }
726
727 static inline void tcg_out_st16_8(TCGContext *s, int cond,
728 int rd, int rn, tcg_target_long im)
729 {
730 if (im >= 0)
731 tcg_out32(s, (cond << 28) | 0x01c000b0 |
732 (rn << 16) | (rd << 12) |
733 ((im & 0xf0) << 4) | (im & 0xf));
734 else
735 tcg_out32(s, (cond << 28) | 0x014000b0 |
736 (rn << 16) | (rd << 12) |
737 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
738 }
739
740 static inline void tcg_out_ld16u_r(TCGContext *s, int cond,
741 int rd, int rn, int rm)
742 {
743 tcg_out32(s, (cond << 28) | 0x019000b0 |
744 (rn << 16) | (rd << 12) | rm);
745 }
746
747 static inline void tcg_out_st16_r(TCGContext *s, int cond,
748 int rd, int rn, int rm)
749 {
750 tcg_out32(s, (cond << 28) | 0x018000b0 |
751 (rn << 16) | (rd << 12) | rm);
752 }
753
754 static inline void tcg_out_ld16s_8(TCGContext *s, int cond,
755 int rd, int rn, tcg_target_long im)
756 {
757 if (im >= 0)
758 tcg_out32(s, (cond << 28) | 0x01d000f0 |
759 (rn << 16) | (rd << 12) |
760 ((im & 0xf0) << 4) | (im & 0xf));
761 else
762 tcg_out32(s, (cond << 28) | 0x015000f0 |
763 (rn << 16) | (rd << 12) |
764 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
765 }
766
767 static inline void tcg_out_ld16s_r(TCGContext *s, int cond,
768 int rd, int rn, int rm)
769 {
770 tcg_out32(s, (cond << 28) | 0x019000f0 |
771 (rn << 16) | (rd << 12) | rm);
772 }
773
774 static inline void tcg_out_ld8_12(TCGContext *s, int cond,
775 int rd, int rn, tcg_target_long im)
776 {
777 if (im >= 0)
778 tcg_out32(s, (cond << 28) | 0x05d00000 |
779 (rn << 16) | (rd << 12) | (im & 0xfff));
780 else
781 tcg_out32(s, (cond << 28) | 0x05500000 |
782 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
783 }
784
785 static inline void tcg_out_st8_12(TCGContext *s, int cond,
786 int rd, int rn, tcg_target_long im)
787 {
788 if (im >= 0)
789 tcg_out32(s, (cond << 28) | 0x05c00000 |
790 (rn << 16) | (rd << 12) | (im & 0xfff));
791 else
792 tcg_out32(s, (cond << 28) | 0x05400000 |
793 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
794 }
795
796 static inline void tcg_out_ld8_r(TCGContext *s, int cond,
797 int rd, int rn, int rm)
798 {
799 tcg_out32(s, (cond << 28) | 0x07d00000 |
800 (rn << 16) | (rd << 12) | rm);
801 }
802
803 static inline void tcg_out_st8_r(TCGContext *s, int cond,
804 int rd, int rn, int rm)
805 {
806 tcg_out32(s, (cond << 28) | 0x07c00000 |
807 (rn << 16) | (rd << 12) | rm);
808 }
809
810 static inline void tcg_out_ld8s_8(TCGContext *s, int cond,
811 int rd, int rn, tcg_target_long im)
812 {
813 if (im >= 0)
814 tcg_out32(s, (cond << 28) | 0x01d000d0 |
815 (rn << 16) | (rd << 12) |
816 ((im & 0xf0) << 4) | (im & 0xf));
817 else
818 tcg_out32(s, (cond << 28) | 0x015000d0 |
819 (rn << 16) | (rd << 12) |
820 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
821 }
822
823 static inline void tcg_out_ld8s_r(TCGContext *s, int cond,
824 int rd, int rn, int rm)
825 {
826 tcg_out32(s, (cond << 28) | 0x019000d0 |
827 (rn << 16) | (rd << 12) | rm);
828 }
829
830 static inline void tcg_out_ld32u(TCGContext *s, int cond,
831 int rd, int rn, int32_t offset)
832 {
833 if (offset > 0xfff || offset < -0xfff) {
834 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
835 tcg_out_ld32_r(s, cond, rd, rn, TCG_REG_R8);
836 } else
837 tcg_out_ld32_12(s, cond, rd, rn, offset);
838 }
839
840 static inline void tcg_out_st32(TCGContext *s, int cond,
841 int rd, int rn, int32_t offset)
842 {
843 if (offset > 0xfff || offset < -0xfff) {
844 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
845 tcg_out_st32_r(s, cond, rd, rn, TCG_REG_R8);
846 } else
847 tcg_out_st32_12(s, cond, rd, rn, offset);
848 }
849
850 static inline void tcg_out_ld16u(TCGContext *s, int cond,
851 int rd, int rn, int32_t offset)
852 {
853 if (offset > 0xff || offset < -0xff) {
854 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
855 tcg_out_ld16u_r(s, cond, rd, rn, TCG_REG_R8);
856 } else
857 tcg_out_ld16u_8(s, cond, rd, rn, offset);
858 }
859
860 static inline void tcg_out_ld16s(TCGContext *s, int cond,
861 int rd, int rn, int32_t offset)
862 {
863 if (offset > 0xff || offset < -0xff) {
864 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
865 tcg_out_ld16s_r(s, cond, rd, rn, TCG_REG_R8);
866 } else
867 tcg_out_ld16s_8(s, cond, rd, rn, offset);
868 }
869
870 static inline void tcg_out_st16(TCGContext *s, int cond,
871 int rd, int rn, int32_t offset)
872 {
873 if (offset > 0xff || offset < -0xff) {
874 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
875 tcg_out_st16_r(s, cond, rd, rn, TCG_REG_R8);
876 } else
877 tcg_out_st16_8(s, cond, rd, rn, offset);
878 }
879
880 static inline void tcg_out_ld8u(TCGContext *s, int cond,
881 int rd, int rn, int32_t offset)
882 {
883 if (offset > 0xfff || offset < -0xfff) {
884 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
885 tcg_out_ld8_r(s, cond, rd, rn, TCG_REG_R8);
886 } else
887 tcg_out_ld8_12(s, cond, rd, rn, offset);
888 }
889
890 static inline void tcg_out_ld8s(TCGContext *s, int cond,
891 int rd, int rn, int32_t offset)
892 {
893 if (offset > 0xff || offset < -0xff) {
894 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
895 tcg_out_ld8s_r(s, cond, rd, rn, TCG_REG_R8);
896 } else
897 tcg_out_ld8s_8(s, cond, rd, rn, offset);
898 }
899
900 static inline void tcg_out_st8(TCGContext *s, int cond,
901 int rd, int rn, int32_t offset)
902 {
903 if (offset > 0xfff || offset < -0xfff) {
904 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
905 tcg_out_st8_r(s, cond, rd, rn, TCG_REG_R8);
906 } else
907 tcg_out_st8_12(s, cond, rd, rn, offset);
908 }
909
910 /* The _goto case is normally between TBs within the same code buffer,
911 * and with the code buffer limited to 16MB we shouldn't need the long
912 * case.
913 *
914 * .... except to the prologue that is in its own buffer.
915 */
916 static inline void tcg_out_goto(TCGContext *s, int cond, uint32_t addr)
917 {
918 int32_t val;
919
920 if (addr & 1) {
921 /* goto to a Thumb destination isn't supported */
922 tcg_abort();
923 }
924
925 val = addr - (tcg_target_long) s->code_ptr;
926 if (val - 8 < 0x01fffffd && val - 8 > -0x01fffffd)
927 tcg_out_b(s, cond, val);
928 else {
929 if (cond == COND_AL) {
930 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
931 tcg_out32(s, addr);
932 } else {
933 tcg_out_movi32(s, cond, TCG_REG_R8, val - 8);
934 tcg_out_dat_reg(s, cond, ARITH_ADD,
935 TCG_REG_PC, TCG_REG_PC,
936 TCG_REG_R8, SHIFT_IMM_LSL(0));
937 }
938 }
939 }
940
941 /* The call case is mostly used for helpers - so it's not unreasonable
942 * for them to be beyond branch range */
943 static inline void tcg_out_call(TCGContext *s, uint32_t addr)
944 {
945 int32_t val;
946
947 val = addr - (tcg_target_long) s->code_ptr;
948 if (val - 8 < 0x02000000 && val - 8 >= -0x02000000) {
949 if (addr & 1) {
950 /* Use BLX if the target is in Thumb mode */
951 if (!use_armv5_instructions) {
952 tcg_abort();
953 }
954 tcg_out_blx_imm(s, val);
955 } else {
956 tcg_out_bl(s, COND_AL, val);
957 }
958 } else {
959 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R14, TCG_REG_PC, 4);
960 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
961 tcg_out32(s, addr);
962 }
963 }
964
965 static inline void tcg_out_callr(TCGContext *s, int cond, int arg)
966 {
967 if (use_armv5_instructions) {
968 tcg_out_blx(s, cond, arg);
969 } else {
970 tcg_out_dat_reg(s, cond, ARITH_MOV, TCG_REG_R14, 0,
971 TCG_REG_PC, SHIFT_IMM_LSL(0));
972 tcg_out_bx(s, cond, arg);
973 }
974 }
975
976 static inline void tcg_out_goto_label(TCGContext *s, int cond, int label_index)
977 {
978 TCGLabel *l = &s->labels[label_index];
979
980 if (l->has_value)
981 tcg_out_goto(s, cond, l->u.value);
982 else if (cond == COND_AL) {
983 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
984 tcg_out_reloc(s, s->code_ptr, R_ARM_ABS32, label_index, 31337);
985 s->code_ptr += 4;
986 } else {
987 /* Probably this should be preferred even for COND_AL... */
988 tcg_out_reloc(s, s->code_ptr, R_ARM_PC24, label_index, 31337);
989 tcg_out_b_noaddr(s, cond);
990 }
991 }
992
993 #ifdef CONFIG_SOFTMMU
994
995 #include "exec/softmmu_defs.h"
996
997 /* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
998 int mmu_idx) */
999 static const void * const qemu_ld_helpers[4] = {
1000 helper_ldb_mmu,
1001 helper_ldw_mmu,
1002 helper_ldl_mmu,
1003 helper_ldq_mmu,
1004 };
1005
1006 /* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
1007 uintxx_t val, int mmu_idx) */
1008 static const void * const qemu_st_helpers[4] = {
1009 helper_stb_mmu,
1010 helper_stw_mmu,
1011 helper_stl_mmu,
1012 helper_stq_mmu,
1013 };
1014
1015 /* Helper routines for marshalling helper function arguments into
1016 * the correct registers and stack.
1017 * argreg is where we want to put this argument, arg is the argument itself.
1018 * Return value is the updated argreg ready for the next call.
1019 * Note that argreg 0..3 is real registers, 4+ on stack.
1020 * When we reach the first stacked argument, we allocate space for it
1021 * and the following stacked arguments using "str r8, [sp, #-0x10]!".
1022 * Following arguments are filled in with "str r8, [sp, #0xNN]".
1023 * For more than 4 stacked arguments we'd need to know how much
1024 * space to allocate when we pushed the first stacked argument.
1025 * We don't need this, so don't implement it (and will assert if you try it.)
1026 *
1027 * We provide routines for arguments which are: immediate, 32 bit
1028 * value in register, 16 and 8 bit values in register (which must be zero
1029 * extended before use) and 64 bit value in a lo:hi register pair.
1030 */
1031 #define DEFINE_TCG_OUT_ARG(NAME, ARGPARAM) \
1032 static TCGReg NAME(TCGContext *s, TCGReg argreg, ARGPARAM) \
1033 { \
1034 if (argreg < 4) { \
1035 TCG_OUT_ARG_GET_ARG(argreg); \
1036 } else if (argreg == 4) { \
1037 TCG_OUT_ARG_GET_ARG(TCG_REG_R8); \
1038 tcg_out32(s, (COND_AL << 28) | 0x052d8010); \
1039 } else { \
1040 assert(argreg < 8); \
1041 TCG_OUT_ARG_GET_ARG(TCG_REG_R8); \
1042 tcg_out32(s, (COND_AL << 28) | 0x058d8000 | (argreg - 4) * 4); \
1043 } \
1044 return argreg + 1; \
1045 }
1046
1047 #define TCG_OUT_ARG_GET_ARG(A) tcg_out_dat_imm(s, COND_AL, ARITH_MOV, A, 0, arg)
1048 DEFINE_TCG_OUT_ARG(tcg_out_arg_imm32, uint32_t arg)
1049 #undef TCG_OUT_ARG_GET_ARG
1050 #define TCG_OUT_ARG_GET_ARG(A) tcg_out_ext8u(s, COND_AL, A, arg)
1051 DEFINE_TCG_OUT_ARG(tcg_out_arg_reg8, TCGReg arg)
1052 #undef TCG_OUT_ARG_GET_ARG
1053 #define TCG_OUT_ARG_GET_ARG(A) tcg_out_ext16u(s, COND_AL, A, arg)
1054 DEFINE_TCG_OUT_ARG(tcg_out_arg_reg16, TCGReg arg)
1055 #undef TCG_OUT_ARG_GET_ARG
1056
1057 /* We don't use the macro for this one to avoid an unnecessary reg-reg
1058 * move when storing to the stack.
1059 */
1060 static TCGReg tcg_out_arg_reg32(TCGContext *s, TCGReg argreg, TCGReg arg)
1061 {
1062 if (argreg < 4) {
1063 tcg_out_mov_reg(s, COND_AL, argreg, arg);
1064 } else if (argreg == 4) {
1065 /* str arg, [sp, #-0x10]! */
1066 tcg_out32(s, (COND_AL << 28) | 0x052d0010 | (arg << 12));
1067 } else {
1068 assert(argreg < 8);
1069 /* str arg, [sp, #0xNN] */
1070 tcg_out32(s, (COND_AL << 28) | 0x058d0000 |
1071 (arg << 12) | (argreg - 4) * 4);
1072 }
1073 return argreg + 1;
1074 }
1075
1076 static inline TCGReg tcg_out_arg_reg64(TCGContext *s, TCGReg argreg,
1077 TCGReg arglo, TCGReg arghi)
1078 {
1079 /* 64 bit arguments must go in even/odd register pairs
1080 * and in 8-aligned stack slots.
1081 */
1082 if (argreg & 1) {
1083 argreg++;
1084 }
1085 argreg = tcg_out_arg_reg32(s, argreg, arglo);
1086 argreg = tcg_out_arg_reg32(s, argreg, arghi);
1087 return argreg;
1088 }
1089
1090 static inline void tcg_out_arg_stacktidy(TCGContext *s, TCGReg argreg)
1091 {
1092 /* Output any necessary post-call cleanup of the stack */
1093 if (argreg > 4) {
1094 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R13, TCG_REG_R13, 0x10);
1095 }
1096 }
1097
1098 #endif
1099
1100 #define TLB_SHIFT (CPU_TLB_ENTRY_BITS + CPU_TLB_BITS)
1101
1102 static inline void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, int opc)
1103 {
1104 int addr_reg, data_reg, data_reg2, bswap;
1105 #ifdef CONFIG_SOFTMMU
1106 int mem_index, s_bits, tlb_offset;
1107 TCGReg argreg;
1108 # if TARGET_LONG_BITS == 64
1109 int addr_reg2;
1110 # endif
1111 uint32_t *label_ptr;
1112 #endif
1113
1114 #ifdef TARGET_WORDS_BIGENDIAN
1115 bswap = 1;
1116 #else
1117 bswap = 0;
1118 #endif
1119 data_reg = *args++;
1120 if (opc == 3)
1121 data_reg2 = *args++;
1122 else
1123 data_reg2 = 0; /* suppress warning */
1124 addr_reg = *args++;
1125 #ifdef CONFIG_SOFTMMU
1126 # if TARGET_LONG_BITS == 64
1127 addr_reg2 = *args++;
1128 # endif
1129 mem_index = *args;
1130 s_bits = opc & 3;
1131
1132 /* Should generate something like the following:
1133 * shr r8, addr_reg, #TARGET_PAGE_BITS
1134 * and r0, r8, #(CPU_TLB_SIZE - 1) @ Assumption: CPU_TLB_BITS <= 8
1135 * add r0, env, r0 lsl #CPU_TLB_ENTRY_BITS
1136 */
1137 # if CPU_TLB_BITS > 8
1138 # error
1139 # endif
1140 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_R8,
1141 0, addr_reg, SHIFT_IMM_LSR(TARGET_PAGE_BITS));
1142 tcg_out_dat_imm(s, COND_AL, ARITH_AND,
1143 TCG_REG_R0, TCG_REG_R8, CPU_TLB_SIZE - 1);
1144 tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_R0, TCG_AREG0,
1145 TCG_REG_R0, SHIFT_IMM_LSL(CPU_TLB_ENTRY_BITS));
1146 /* We assume that the offset is contained within 20 bits. */
1147 tlb_offset = offsetof(CPUArchState, tlb_table[mem_index][0].addr_read);
1148 assert((tlb_offset & ~0xfffff) == 0);
1149 if (tlb_offset > 0xfff) {
1150 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R0, TCG_REG_R0,
1151 0xa00 | (tlb_offset >> 12));
1152 tlb_offset &= 0xfff;
1153 }
1154 tcg_out_ld32_12wb(s, COND_AL, TCG_REG_R1, TCG_REG_R0, tlb_offset);
1155 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, TCG_REG_R1,
1156 TCG_REG_R8, SHIFT_IMM_LSL(TARGET_PAGE_BITS));
1157 /* Check alignment. */
1158 if (s_bits)
1159 tcg_out_dat_imm(s, COND_EQ, ARITH_TST,
1160 0, addr_reg, (1 << s_bits) - 1);
1161 # if TARGET_LONG_BITS == 64
1162 /* XXX: possibly we could use a block data load in the first access. */
1163 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0, 4);
1164 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
1165 TCG_REG_R1, addr_reg2, SHIFT_IMM_LSL(0));
1166 # endif
1167 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0,
1168 offsetof(CPUTLBEntry, addend)
1169 - offsetof(CPUTLBEntry, addr_read));
1170
1171 switch (opc) {
1172 case 0:
1173 tcg_out_ld8_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1174 break;
1175 case 0 | 4:
1176 tcg_out_ld8s_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1177 break;
1178 case 1:
1179 tcg_out_ld16u_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1180 if (bswap) {
1181 tcg_out_bswap16(s, COND_EQ, data_reg, data_reg);
1182 }
1183 break;
1184 case 1 | 4:
1185 if (bswap) {
1186 tcg_out_ld16u_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1187 tcg_out_bswap16s(s, COND_EQ, data_reg, data_reg);
1188 } else {
1189 tcg_out_ld16s_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1190 }
1191 break;
1192 case 2:
1193 default:
1194 tcg_out_ld32_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1195 if (bswap) {
1196 tcg_out_bswap32(s, COND_EQ, data_reg, data_reg);
1197 }
1198 break;
1199 case 3:
1200 if (bswap) {
1201 tcg_out_ld32_rwb(s, COND_EQ, data_reg2, TCG_REG_R1, addr_reg);
1202 tcg_out_ld32_12(s, COND_EQ, data_reg, TCG_REG_R1, 4);
1203 tcg_out_bswap32(s, COND_EQ, data_reg2, data_reg2);
1204 tcg_out_bswap32(s, COND_EQ, data_reg, data_reg);
1205 } else {
1206 tcg_out_ld32_rwb(s, COND_EQ, data_reg, TCG_REG_R1, addr_reg);
1207 tcg_out_ld32_12(s, COND_EQ, data_reg2, TCG_REG_R1, 4);
1208 }
1209 break;
1210 }
1211
1212 label_ptr = (void *) s->code_ptr;
1213 tcg_out_b_noaddr(s, COND_EQ);
1214
1215 /* TODO: move this code to where the constants pool will be */
1216 /* Note that this code relies on the constraints we set in arm_op_defs[]
1217 * to ensure that later arguments are not passed to us in registers we
1218 * trash by moving the earlier arguments into them.
1219 */
1220 argreg = TCG_REG_R0;
1221 argreg = tcg_out_arg_reg32(s, argreg, TCG_AREG0);
1222 #if TARGET_LONG_BITS == 64
1223 argreg = tcg_out_arg_reg64(s, argreg, addr_reg, addr_reg2);
1224 #else
1225 argreg = tcg_out_arg_reg32(s, argreg, addr_reg);
1226 #endif
1227 argreg = tcg_out_arg_imm32(s, argreg, mem_index);
1228 tcg_out_call(s, (tcg_target_long) qemu_ld_helpers[s_bits]);
1229 tcg_out_arg_stacktidy(s, argreg);
1230
1231 switch (opc) {
1232 case 0 | 4:
1233 tcg_out_ext8s(s, COND_AL, data_reg, TCG_REG_R0);
1234 break;
1235 case 1 | 4:
1236 tcg_out_ext16s(s, COND_AL, data_reg, TCG_REG_R0);
1237 break;
1238 case 0:
1239 case 1:
1240 case 2:
1241 default:
1242 tcg_out_mov_reg(s, COND_AL, data_reg, TCG_REG_R0);
1243 break;
1244 case 3:
1245 tcg_out_mov_reg(s, COND_AL, data_reg, TCG_REG_R0);
1246 tcg_out_mov_reg(s, COND_AL, data_reg2, TCG_REG_R1);
1247 break;
1248 }
1249
1250 reloc_pc24(label_ptr, (tcg_target_long)s->code_ptr);
1251 #else /* !CONFIG_SOFTMMU */
1252 if (GUEST_BASE) {
1253 uint32_t offset = GUEST_BASE;
1254 int i;
1255 int rot;
1256
1257 while (offset) {
1258 i = ctz32(offset) & ~1;
1259 rot = ((32 - i) << 7) & 0xf00;
1260
1261 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R8, addr_reg,
1262 ((offset >> i) & 0xff) | rot);
1263 addr_reg = TCG_REG_R8;
1264 offset &= ~(0xff << i);
1265 }
1266 }
1267 switch (opc) {
1268 case 0:
1269 tcg_out_ld8_12(s, COND_AL, data_reg, addr_reg, 0);
1270 break;
1271 case 0 | 4:
1272 tcg_out_ld8s_8(s, COND_AL, data_reg, addr_reg, 0);
1273 break;
1274 case 1:
1275 tcg_out_ld16u_8(s, COND_AL, data_reg, addr_reg, 0);
1276 if (bswap) {
1277 tcg_out_bswap16(s, COND_AL, data_reg, data_reg);
1278 }
1279 break;
1280 case 1 | 4:
1281 if (bswap) {
1282 tcg_out_ld16u_8(s, COND_AL, data_reg, addr_reg, 0);
1283 tcg_out_bswap16s(s, COND_AL, data_reg, data_reg);
1284 } else {
1285 tcg_out_ld16s_8(s, COND_AL, data_reg, addr_reg, 0);
1286 }
1287 break;
1288 case 2:
1289 default:
1290 tcg_out_ld32_12(s, COND_AL, data_reg, addr_reg, 0);
1291 if (bswap) {
1292 tcg_out_bswap32(s, COND_AL, data_reg, data_reg);
1293 }
1294 break;
1295 case 3:
1296 /* TODO: use block load -
1297 * check that data_reg2 > data_reg or the other way */
1298 if (data_reg == addr_reg) {
1299 tcg_out_ld32_12(s, COND_AL, data_reg2, addr_reg, bswap ? 0 : 4);
1300 tcg_out_ld32_12(s, COND_AL, data_reg, addr_reg, bswap ? 4 : 0);
1301 } else {
1302 tcg_out_ld32_12(s, COND_AL, data_reg, addr_reg, bswap ? 4 : 0);
1303 tcg_out_ld32_12(s, COND_AL, data_reg2, addr_reg, bswap ? 0 : 4);
1304 }
1305 if (bswap) {
1306 tcg_out_bswap32(s, COND_AL, data_reg, data_reg);
1307 tcg_out_bswap32(s, COND_AL, data_reg2, data_reg2);
1308 }
1309 break;
1310 }
1311 #endif
1312 }
1313
1314 static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)
1315 {
1316 int addr_reg, data_reg, data_reg2, bswap;
1317 #ifdef CONFIG_SOFTMMU
1318 int mem_index, s_bits, tlb_offset;
1319 TCGReg argreg;
1320 # if TARGET_LONG_BITS == 64
1321 int addr_reg2;
1322 # endif
1323 uint32_t *label_ptr;
1324 #endif
1325
1326 #ifdef TARGET_WORDS_BIGENDIAN
1327 bswap = 1;
1328 #else
1329 bswap = 0;
1330 #endif
1331 data_reg = *args++;
1332 if (opc == 3)
1333 data_reg2 = *args++;
1334 else
1335 data_reg2 = 0; /* suppress warning */
1336 addr_reg = *args++;
1337 #ifdef CONFIG_SOFTMMU
1338 # if TARGET_LONG_BITS == 64
1339 addr_reg2 = *args++;
1340 # endif
1341 mem_index = *args;
1342 s_bits = opc & 3;
1343
1344 /* Should generate something like the following:
1345 * shr r8, addr_reg, #TARGET_PAGE_BITS
1346 * and r0, r8, #(CPU_TLB_SIZE - 1) @ Assumption: CPU_TLB_BITS <= 8
1347 * add r0, env, r0 lsl #CPU_TLB_ENTRY_BITS
1348 */
1349 tcg_out_dat_reg(s, COND_AL, ARITH_MOV,
1350 TCG_REG_R8, 0, addr_reg, SHIFT_IMM_LSR(TARGET_PAGE_BITS));
1351 tcg_out_dat_imm(s, COND_AL, ARITH_AND,
1352 TCG_REG_R0, TCG_REG_R8, CPU_TLB_SIZE - 1);
1353 tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_R0,
1354 TCG_AREG0, TCG_REG_R0, SHIFT_IMM_LSL(CPU_TLB_ENTRY_BITS));
1355 /* We assume that the offset is contained within 20 bits. */
1356 tlb_offset = offsetof(CPUArchState, tlb_table[mem_index][0].addr_write);
1357 assert((tlb_offset & ~0xfffff) == 0);
1358 if (tlb_offset > 0xfff) {
1359 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R0, TCG_REG_R0,
1360 0xa00 | (tlb_offset >> 12));
1361 tlb_offset &= 0xfff;
1362 }
1363 tcg_out_ld32_12wb(s, COND_AL, TCG_REG_R1, TCG_REG_R0, tlb_offset);
1364 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, TCG_REG_R1,
1365 TCG_REG_R8, SHIFT_IMM_LSL(TARGET_PAGE_BITS));
1366 /* Check alignment. */
1367 if (s_bits)
1368 tcg_out_dat_imm(s, COND_EQ, ARITH_TST,
1369 0, addr_reg, (1 << s_bits) - 1);
1370 # if TARGET_LONG_BITS == 64
1371 /* XXX: possibly we could use a block data load in the first access. */
1372 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0, 4);
1373 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
1374 TCG_REG_R1, addr_reg2, SHIFT_IMM_LSL(0));
1375 # endif
1376 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0,
1377 offsetof(CPUTLBEntry, addend)
1378 - offsetof(CPUTLBEntry, addr_write));
1379
1380 switch (opc) {
1381 case 0:
1382 tcg_out_st8_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1383 break;
1384 case 1:
1385 if (bswap) {
1386 tcg_out_bswap16st(s, COND_EQ, TCG_REG_R0, data_reg);
1387 tcg_out_st16_r(s, COND_EQ, TCG_REG_R0, addr_reg, TCG_REG_R1);
1388 } else {
1389 tcg_out_st16_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1390 }
1391 break;
1392 case 2:
1393 default:
1394 if (bswap) {
1395 tcg_out_bswap32(s, COND_EQ, TCG_REG_R0, data_reg);
1396 tcg_out_st32_r(s, COND_EQ, TCG_REG_R0, addr_reg, TCG_REG_R1);
1397 } else {
1398 tcg_out_st32_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1399 }
1400 break;
1401 case 3:
1402 if (bswap) {
1403 tcg_out_bswap32(s, COND_EQ, TCG_REG_R0, data_reg2);
1404 tcg_out_st32_rwb(s, COND_EQ, TCG_REG_R0, TCG_REG_R1, addr_reg);
1405 tcg_out_bswap32(s, COND_EQ, TCG_REG_R0, data_reg);
1406 tcg_out_st32_12(s, COND_EQ, TCG_REG_R0, TCG_REG_R1, 4);
1407 } else {
1408 tcg_out_st32_rwb(s, COND_EQ, data_reg, TCG_REG_R1, addr_reg);
1409 tcg_out_st32_12(s, COND_EQ, data_reg2, TCG_REG_R1, 4);
1410 }
1411 break;
1412 }
1413
1414 label_ptr = (void *) s->code_ptr;
1415 tcg_out_b_noaddr(s, COND_EQ);
1416
1417 /* TODO: move this code to where the constants pool will be */
1418 /* Note that this code relies on the constraints we set in arm_op_defs[]
1419 * to ensure that later arguments are not passed to us in registers we
1420 * trash by moving the earlier arguments into them.
1421 */
1422 argreg = TCG_REG_R0;
1423 argreg = tcg_out_arg_reg32(s, argreg, TCG_AREG0);
1424 #if TARGET_LONG_BITS == 64
1425 argreg = tcg_out_arg_reg64(s, argreg, addr_reg, addr_reg2);
1426 #else
1427 argreg = tcg_out_arg_reg32(s, argreg, addr_reg);
1428 #endif
1429
1430 switch (opc) {
1431 case 0:
1432 argreg = tcg_out_arg_reg8(s, argreg, data_reg);
1433 break;
1434 case 1:
1435 argreg = tcg_out_arg_reg16(s, argreg, data_reg);
1436 break;
1437 case 2:
1438 argreg = tcg_out_arg_reg32(s, argreg, data_reg);
1439 break;
1440 case 3:
1441 argreg = tcg_out_arg_reg64(s, argreg, data_reg, data_reg2);
1442 break;
1443 }
1444
1445 argreg = tcg_out_arg_imm32(s, argreg, mem_index);
1446 tcg_out_call(s, (tcg_target_long) qemu_st_helpers[s_bits]);
1447 tcg_out_arg_stacktidy(s, argreg);
1448
1449 reloc_pc24(label_ptr, (tcg_target_long)s->code_ptr);
1450 #else /* !CONFIG_SOFTMMU */
1451 if (GUEST_BASE) {
1452 uint32_t offset = GUEST_BASE;
1453 int i;
1454 int rot;
1455
1456 while (offset) {
1457 i = ctz32(offset) & ~1;
1458 rot = ((32 - i) << 7) & 0xf00;
1459
1460 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R1, addr_reg,
1461 ((offset >> i) & 0xff) | rot);
1462 addr_reg = TCG_REG_R1;
1463 offset &= ~(0xff << i);
1464 }
1465 }
1466 switch (opc) {
1467 case 0:
1468 tcg_out_st8_12(s, COND_AL, data_reg, addr_reg, 0);
1469 break;
1470 case 1:
1471 if (bswap) {
1472 tcg_out_bswap16st(s, COND_AL, TCG_REG_R0, data_reg);
1473 tcg_out_st16_8(s, COND_AL, TCG_REG_R0, addr_reg, 0);
1474 } else {
1475 tcg_out_st16_8(s, COND_AL, data_reg, addr_reg, 0);
1476 }
1477 break;
1478 case 2:
1479 default:
1480 if (bswap) {
1481 tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg);
1482 tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addr_reg, 0);
1483 } else {
1484 tcg_out_st32_12(s, COND_AL, data_reg, addr_reg, 0);
1485 }
1486 break;
1487 case 3:
1488 /* TODO: use block store -
1489 * check that data_reg2 > data_reg or the other way */
1490 if (bswap) {
1491 tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg2);
1492 tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addr_reg, 0);
1493 tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg);
1494 tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addr_reg, 4);
1495 } else {
1496 tcg_out_st32_12(s, COND_AL, data_reg, addr_reg, 0);
1497 tcg_out_st32_12(s, COND_AL, data_reg2, addr_reg, 4);
1498 }
1499 break;
1500 }
1501 #endif
1502 }
1503
1504 static uint8_t *tb_ret_addr;
1505
1506 static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
1507 const TCGArg *args, const int *const_args)
1508 {
1509 int c;
1510
1511 switch (opc) {
1512 case INDEX_op_exit_tb:
1513 {
1514 uint8_t *ld_ptr = s->code_ptr;
1515 if (args[0] >> 8)
1516 tcg_out_ld32_12(s, COND_AL, TCG_REG_R0, TCG_REG_PC, 0);
1517 else
1518 tcg_out_dat_imm(s, COND_AL, ARITH_MOV, TCG_REG_R0, 0, args[0]);
1519 tcg_out_goto(s, COND_AL, (tcg_target_ulong) tb_ret_addr);
1520 if (args[0] >> 8) {
1521 *ld_ptr = (uint8_t) (s->code_ptr - ld_ptr) - 8;
1522 tcg_out32(s, args[0]);
1523 }
1524 }
1525 break;
1526 case INDEX_op_goto_tb:
1527 if (s->tb_jmp_offset) {
1528 /* Direct jump method */
1529 #if defined(USE_DIRECT_JUMP)
1530 s->tb_jmp_offset[args[0]] = s->code_ptr - s->code_buf;
1531 tcg_out_b_noaddr(s, COND_AL);
1532 #else
1533 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
1534 s->tb_jmp_offset[args[0]] = s->code_ptr - s->code_buf;
1535 tcg_out32(s, 0);
1536 #endif
1537 } else {
1538 /* Indirect jump method */
1539 #if 1
1540 c = (int) (s->tb_next + args[0]) - ((int) s->code_ptr + 8);
1541 if (c > 0xfff || c < -0xfff) {
1542 tcg_out_movi32(s, COND_AL, TCG_REG_R0,
1543 (tcg_target_long) (s->tb_next + args[0]));
1544 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, 0);
1545 } else
1546 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, c);
1547 #else
1548 tcg_out_ld32_12(s, COND_AL, TCG_REG_R0, TCG_REG_PC, 0);
1549 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, 0);
1550 tcg_out32(s, (tcg_target_long) (s->tb_next + args[0]));
1551 #endif
1552 }
1553 s->tb_next_offset[args[0]] = s->code_ptr - s->code_buf;
1554 break;
1555 case INDEX_op_call:
1556 if (const_args[0])
1557 tcg_out_call(s, args[0]);
1558 else
1559 tcg_out_callr(s, COND_AL, args[0]);
1560 break;
1561 case INDEX_op_br:
1562 tcg_out_goto_label(s, COND_AL, args[0]);
1563 break;
1564
1565 case INDEX_op_ld8u_i32:
1566 tcg_out_ld8u(s, COND_AL, args[0], args[1], args[2]);
1567 break;
1568 case INDEX_op_ld8s_i32:
1569 tcg_out_ld8s(s, COND_AL, args[0], args[1], args[2]);
1570 break;
1571 case INDEX_op_ld16u_i32:
1572 tcg_out_ld16u(s, COND_AL, args[0], args[1], args[2]);
1573 break;
1574 case INDEX_op_ld16s_i32:
1575 tcg_out_ld16s(s, COND_AL, args[0], args[1], args[2]);
1576 break;
1577 case INDEX_op_ld_i32:
1578 tcg_out_ld32u(s, COND_AL, args[0], args[1], args[2]);
1579 break;
1580 case INDEX_op_st8_i32:
1581 tcg_out_st8(s, COND_AL, args[0], args[1], args[2]);
1582 break;
1583 case INDEX_op_st16_i32:
1584 tcg_out_st16(s, COND_AL, args[0], args[1], args[2]);
1585 break;
1586 case INDEX_op_st_i32:
1587 tcg_out_st32(s, COND_AL, args[0], args[1], args[2]);
1588 break;
1589
1590 case INDEX_op_mov_i32:
1591 tcg_out_dat_reg(s, COND_AL, ARITH_MOV,
1592 args[0], 0, args[1], SHIFT_IMM_LSL(0));
1593 break;
1594 case INDEX_op_movi_i32:
1595 tcg_out_movi32(s, COND_AL, args[0], args[1]);
1596 break;
1597 case INDEX_op_movcond_i32:
1598 /* Constraints mean that v2 is always in the same register as dest,
1599 * so we only need to do "if condition passed, move v1 to dest".
1600 */
1601 tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0,
1602 args[1], args[2], const_args[2]);
1603 tcg_out_dat_rI(s, tcg_cond_to_arm_cond[args[5]],
1604 ARITH_MOV, args[0], 0, args[3], const_args[3]);
1605 break;
1606 case INDEX_op_add_i32:
1607 c = ARITH_ADD;
1608 goto gen_arith;
1609 case INDEX_op_sub_i32:
1610 c = ARITH_SUB;
1611 goto gen_arith;
1612 case INDEX_op_and_i32:
1613 c = ARITH_AND;
1614 goto gen_arith;
1615 case INDEX_op_andc_i32:
1616 c = ARITH_BIC;
1617 goto gen_arith;
1618 case INDEX_op_or_i32:
1619 c = ARITH_ORR;
1620 goto gen_arith;
1621 case INDEX_op_xor_i32:
1622 c = ARITH_EOR;
1623 /* Fall through. */
1624 gen_arith:
1625 tcg_out_dat_rI(s, COND_AL, c, args[0], args[1], args[2], const_args[2]);
1626 break;
1627 case INDEX_op_add2_i32:
1628 tcg_out_dat_reg2(s, COND_AL, ARITH_ADD, ARITH_ADC,
1629 args[0], args[1], args[2], args[3],
1630 args[4], args[5], SHIFT_IMM_LSL(0));
1631 break;
1632 case INDEX_op_sub2_i32:
1633 tcg_out_dat_reg2(s, COND_AL, ARITH_SUB, ARITH_SBC,
1634 args[0], args[1], args[2], args[3],
1635 args[4], args[5], SHIFT_IMM_LSL(0));
1636 break;
1637 case INDEX_op_neg_i32:
1638 tcg_out_dat_imm(s, COND_AL, ARITH_RSB, args[0], args[1], 0);
1639 break;
1640 case INDEX_op_not_i32:
1641 tcg_out_dat_reg(s, COND_AL,
1642 ARITH_MVN, args[0], 0, args[1], SHIFT_IMM_LSL(0));
1643 break;
1644 case INDEX_op_mul_i32:
1645 tcg_out_mul32(s, COND_AL, args[0], args[1], args[2]);
1646 break;
1647 case INDEX_op_mulu2_i32:
1648 tcg_out_umull32(s, COND_AL, args[0], args[1], args[2], args[3]);
1649 break;
1650 case INDEX_op_muls2_i32:
1651 tcg_out_smull32(s, COND_AL, args[0], args[1], args[2], args[3]);
1652 break;
1653 /* XXX: Perhaps args[2] & 0x1f is wrong */
1654 case INDEX_op_shl_i32:
1655 c = const_args[2] ?
1656 SHIFT_IMM_LSL(args[2] & 0x1f) : SHIFT_REG_LSL(args[2]);
1657 goto gen_shift32;
1658 case INDEX_op_shr_i32:
1659 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_LSR(args[2] & 0x1f) :
1660 SHIFT_IMM_LSL(0) : SHIFT_REG_LSR(args[2]);
1661 goto gen_shift32;
1662 case INDEX_op_sar_i32:
1663 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ASR(args[2] & 0x1f) :
1664 SHIFT_IMM_LSL(0) : SHIFT_REG_ASR(args[2]);
1665 goto gen_shift32;
1666 case INDEX_op_rotr_i32:
1667 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ROR(args[2] & 0x1f) :
1668 SHIFT_IMM_LSL(0) : SHIFT_REG_ROR(args[2]);
1669 /* Fall through. */
1670 gen_shift32:
1671 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1], c);
1672 break;
1673
1674 case INDEX_op_rotl_i32:
1675 if (const_args[2]) {
1676 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
1677 ((0x20 - args[2]) & 0x1f) ?
1678 SHIFT_IMM_ROR((0x20 - args[2]) & 0x1f) :
1679 SHIFT_IMM_LSL(0));
1680 } else {
1681 tcg_out_dat_imm(s, COND_AL, ARITH_RSB, TCG_REG_R8, args[1], 0x20);
1682 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
1683 SHIFT_REG_ROR(TCG_REG_R8));
1684 }
1685 break;
1686
1687 case INDEX_op_brcond_i32:
1688 tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0,
1689 args[0], args[1], const_args[1]);
1690 tcg_out_goto_label(s, tcg_cond_to_arm_cond[args[2]], args[3]);
1691 break;
1692 case INDEX_op_brcond2_i32:
1693 /* The resulting conditions are:
1694 * TCG_COND_EQ --> a0 == a2 && a1 == a3,
1695 * TCG_COND_NE --> (a0 != a2 && a1 == a3) || a1 != a3,
1696 * TCG_COND_LT(U) --> (a0 < a2 && a1 == a3) || a1 < a3,
1697 * TCG_COND_GE(U) --> (a0 >= a2 && a1 == a3) || (a1 >= a3 && a1 != a3),
1698 * TCG_COND_LE(U) --> (a0 <= a2 && a1 == a3) || (a1 <= a3 && a1 != a3),
1699 * TCG_COND_GT(U) --> (a0 > a2 && a1 == a3) || a1 > a3,
1700 */
1701 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1702 args[1], args[3], SHIFT_IMM_LSL(0));
1703 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
1704 args[0], args[2], SHIFT_IMM_LSL(0));
1705 tcg_out_goto_label(s, tcg_cond_to_arm_cond[args[4]], args[5]);
1706 break;
1707 case INDEX_op_setcond_i32:
1708 tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0,
1709 args[1], args[2], const_args[2]);
1710 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[args[3]],
1711 ARITH_MOV, args[0], 0, 1);
1712 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(args[3])],
1713 ARITH_MOV, args[0], 0, 0);
1714 break;
1715 case INDEX_op_setcond2_i32:
1716 /* See brcond2_i32 comment */
1717 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1718 args[2], args[4], SHIFT_IMM_LSL(0));
1719 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
1720 args[1], args[3], SHIFT_IMM_LSL(0));
1721 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[args[5]],
1722 ARITH_MOV, args[0], 0, 1);
1723 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(args[5])],
1724 ARITH_MOV, args[0], 0, 0);
1725 break;
1726
1727 case INDEX_op_qemu_ld8u:
1728 tcg_out_qemu_ld(s, args, 0);
1729 break;
1730 case INDEX_op_qemu_ld8s:
1731 tcg_out_qemu_ld(s, args, 0 | 4);
1732 break;
1733 case INDEX_op_qemu_ld16u:
1734 tcg_out_qemu_ld(s, args, 1);
1735 break;
1736 case INDEX_op_qemu_ld16s:
1737 tcg_out_qemu_ld(s, args, 1 | 4);
1738 break;
1739 case INDEX_op_qemu_ld32:
1740 tcg_out_qemu_ld(s, args, 2);
1741 break;
1742 case INDEX_op_qemu_ld64:
1743 tcg_out_qemu_ld(s, args, 3);
1744 break;
1745
1746 case INDEX_op_qemu_st8:
1747 tcg_out_qemu_st(s, args, 0);
1748 break;
1749 case INDEX_op_qemu_st16:
1750 tcg_out_qemu_st(s, args, 1);
1751 break;
1752 case INDEX_op_qemu_st32:
1753 tcg_out_qemu_st(s, args, 2);
1754 break;
1755 case INDEX_op_qemu_st64:
1756 tcg_out_qemu_st(s, args, 3);
1757 break;
1758
1759 case INDEX_op_bswap16_i32:
1760 tcg_out_bswap16(s, COND_AL, args[0], args[1]);
1761 break;
1762 case INDEX_op_bswap32_i32:
1763 tcg_out_bswap32(s, COND_AL, args[0], args[1]);
1764 break;
1765
1766 case INDEX_op_ext8s_i32:
1767 tcg_out_ext8s(s, COND_AL, args[0], args[1]);
1768 break;
1769 case INDEX_op_ext16s_i32:
1770 tcg_out_ext16s(s, COND_AL, args[0], args[1]);
1771 break;
1772 case INDEX_op_ext16u_i32:
1773 tcg_out_ext16u(s, COND_AL, args[0], args[1]);
1774 break;
1775
1776 default:
1777 tcg_abort();
1778 }
1779 }
1780
1781 static const TCGTargetOpDef arm_op_defs[] = {
1782 { INDEX_op_exit_tb, { } },
1783 { INDEX_op_goto_tb, { } },
1784 { INDEX_op_call, { "ri" } },
1785 { INDEX_op_br, { } },
1786
1787 { INDEX_op_mov_i32, { "r", "r" } },
1788 { INDEX_op_movi_i32, { "r" } },
1789
1790 { INDEX_op_ld8u_i32, { "r", "r" } },
1791 { INDEX_op_ld8s_i32, { "r", "r" } },
1792 { INDEX_op_ld16u_i32, { "r", "r" } },
1793 { INDEX_op_ld16s_i32, { "r", "r" } },
1794 { INDEX_op_ld_i32, { "r", "r" } },
1795 { INDEX_op_st8_i32, { "r", "r" } },
1796 { INDEX_op_st16_i32, { "r", "r" } },
1797 { INDEX_op_st_i32, { "r", "r" } },
1798
1799 /* TODO: "r", "r", "ri" */
1800 { INDEX_op_add_i32, { "r", "r", "rI" } },
1801 { INDEX_op_sub_i32, { "r", "r", "rI" } },
1802 { INDEX_op_mul_i32, { "r", "r", "r" } },
1803 { INDEX_op_mulu2_i32, { "r", "r", "r", "r" } },
1804 { INDEX_op_muls2_i32, { "r", "r", "r", "r" } },
1805 { INDEX_op_and_i32, { "r", "r", "rI" } },
1806 { INDEX_op_andc_i32, { "r", "r", "rI" } },
1807 { INDEX_op_or_i32, { "r", "r", "rI" } },
1808 { INDEX_op_xor_i32, { "r", "r", "rI" } },
1809 { INDEX_op_neg_i32, { "r", "r" } },
1810 { INDEX_op_not_i32, { "r", "r" } },
1811
1812 { INDEX_op_shl_i32, { "r", "r", "ri" } },
1813 { INDEX_op_shr_i32, { "r", "r", "ri" } },
1814 { INDEX_op_sar_i32, { "r", "r", "ri" } },
1815 { INDEX_op_rotl_i32, { "r", "r", "ri" } },
1816 { INDEX_op_rotr_i32, { "r", "r", "ri" } },
1817
1818 { INDEX_op_brcond_i32, { "r", "rI" } },
1819 { INDEX_op_setcond_i32, { "r", "r", "rI" } },
1820 { INDEX_op_movcond_i32, { "r", "r", "rI", "rI", "0" } },
1821
1822 /* TODO: "r", "r", "r", "r", "ri", "ri" */
1823 { INDEX_op_add2_i32, { "r", "r", "r", "r", "r", "r" } },
1824 { INDEX_op_sub2_i32, { "r", "r", "r", "r", "r", "r" } },
1825 { INDEX_op_brcond2_i32, { "r", "r", "r", "r" } },
1826 { INDEX_op_setcond2_i32, { "r", "r", "r", "r", "r" } },
1827
1828 #if TARGET_LONG_BITS == 32
1829 { INDEX_op_qemu_ld8u, { "r", "l" } },
1830 { INDEX_op_qemu_ld8s, { "r", "l" } },
1831 { INDEX_op_qemu_ld16u, { "r", "l" } },
1832 { INDEX_op_qemu_ld16s, { "r", "l" } },
1833 { INDEX_op_qemu_ld32, { "r", "l" } },
1834 { INDEX_op_qemu_ld64, { "L", "L", "l" } },
1835
1836 { INDEX_op_qemu_st8, { "s", "s" } },
1837 { INDEX_op_qemu_st16, { "s", "s" } },
1838 { INDEX_op_qemu_st32, { "s", "s" } },
1839 { INDEX_op_qemu_st64, { "S", "S", "s" } },
1840 #else
1841 { INDEX_op_qemu_ld8u, { "r", "l", "l" } },
1842 { INDEX_op_qemu_ld8s, { "r", "l", "l" } },
1843 { INDEX_op_qemu_ld16u, { "r", "l", "l" } },
1844 { INDEX_op_qemu_ld16s, { "r", "l", "l" } },
1845 { INDEX_op_qemu_ld32, { "r", "l", "l" } },
1846 { INDEX_op_qemu_ld64, { "L", "L", "l", "l" } },
1847
1848 { INDEX_op_qemu_st8, { "s", "s", "s" } },
1849 { INDEX_op_qemu_st16, { "s", "s", "s" } },
1850 { INDEX_op_qemu_st32, { "s", "s", "s" } },
1851 { INDEX_op_qemu_st64, { "S", "S", "s", "s" } },
1852 #endif
1853
1854 { INDEX_op_bswap16_i32, { "r", "r" } },
1855 { INDEX_op_bswap32_i32, { "r", "r" } },
1856
1857 { INDEX_op_ext8s_i32, { "r", "r" } },
1858 { INDEX_op_ext16s_i32, { "r", "r" } },
1859 { INDEX_op_ext16u_i32, { "r", "r" } },
1860
1861 { -1 },
1862 };
1863
1864 static void tcg_target_init(TCGContext *s)
1865 {
1866 #if !defined(CONFIG_USER_ONLY)
1867 /* fail safe */
1868 if ((1 << CPU_TLB_ENTRY_BITS) != sizeof(CPUTLBEntry))
1869 tcg_abort();
1870 #endif
1871
1872 tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I32], 0, 0xffff);
1873 tcg_regset_set32(tcg_target_call_clobber_regs, 0,
1874 (1 << TCG_REG_R0) |
1875 (1 << TCG_REG_R1) |
1876 (1 << TCG_REG_R2) |
1877 (1 << TCG_REG_R3) |
1878 (1 << TCG_REG_R12) |
1879 (1 << TCG_REG_R14));
1880
1881 tcg_regset_clear(s->reserved_regs);
1882 tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
1883 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R8);
1884 tcg_regset_set_reg(s->reserved_regs, TCG_REG_PC);
1885
1886 tcg_add_target_add_op_defs(arm_op_defs);
1887 tcg_set_frame(s, TCG_AREG0, offsetof(CPUArchState, temp_buf),
1888 CPU_TEMP_BUF_NLONGS * sizeof(long));
1889 }
1890
1891 static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
1892 TCGReg arg1, tcg_target_long arg2)
1893 {
1894 tcg_out_ld32u(s, COND_AL, arg, arg1, arg2);
1895 }
1896
1897 static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
1898 TCGReg arg1, tcg_target_long arg2)
1899 {
1900 tcg_out_st32(s, COND_AL, arg, arg1, arg2);
1901 }
1902
1903 static inline void tcg_out_mov(TCGContext *s, TCGType type,
1904 TCGReg ret, TCGReg arg)
1905 {
1906 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, ret, 0, arg, SHIFT_IMM_LSL(0));
1907 }
1908
1909 static inline void tcg_out_movi(TCGContext *s, TCGType type,
1910 TCGReg ret, tcg_target_long arg)
1911 {
1912 tcg_out_movi32(s, COND_AL, ret, arg);
1913 }
1914
1915 static void tcg_target_qemu_prologue(TCGContext *s)
1916 {
1917 /* Calling convention requires us to save r4-r11 and lr;
1918 * save also r12 to maintain stack 8-alignment.
1919 */
1920
1921 /* stmdb sp!, { r4 - r12, lr } */
1922 tcg_out32(s, (COND_AL << 28) | 0x092d5ff0);
1923
1924 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
1925
1926 tcg_out_bx(s, COND_AL, tcg_target_call_iarg_regs[1]);
1927 tb_ret_addr = s->code_ptr;
1928
1929 /* ldmia sp!, { r4 - r12, pc } */
1930 tcg_out32(s, (COND_AL << 28) | 0x08bd9ff0);
1931 }
AR7 emulation for QEMU