?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
linux-user: Fix calculation of auxv length
[qemu/ar7.git] / tcg / arm / tcg-target.inc.c
blob98a12535a5a330f10e7d17e0ff87ddc6d637e6c2
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 #include "elf.h"
26 #include "tcg-pool.inc.c"
27
28 int arm_arch = __ARM_ARCH;
29
30 #ifndef use_idiv_instructions
31 bool use_idiv_instructions;
32 #endif
33
34 /* ??? Ought to think about changing CONFIG_SOFTMMU to always defined. */
35 #ifdef CONFIG_SOFTMMU
36 # define USING_SOFTMMU 1
37 #else
38 # define USING_SOFTMMU 0
39 #endif
40
41 #ifdef CONFIG_DEBUG_TCG
42 static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
43 "%r0",
44 "%r1",
45 "%r2",
46 "%r3",
47 "%r4",
48 "%r5",
49 "%r6",
50 "%r7",
51 "%r8",
52 "%r9",
53 "%r10",
54 "%r11",
55 "%r12",
56 "%r13",
57 "%r14",
58 "%pc",
59 };
60 #endif
61
62 static const int tcg_target_reg_alloc_order[] = {
63 TCG_REG_R4,
64 TCG_REG_R5,
65 TCG_REG_R6,
66 TCG_REG_R7,
67 TCG_REG_R8,
68 TCG_REG_R9,
69 TCG_REG_R10,
70 TCG_REG_R11,
71 TCG_REG_R13,
72 TCG_REG_R0,
73 TCG_REG_R1,
74 TCG_REG_R2,
75 TCG_REG_R3,
76 TCG_REG_R12,
77 TCG_REG_R14,
78 };
79
80 static const int tcg_target_call_iarg_regs[4] = {
81 TCG_REG_R0, TCG_REG_R1, TCG_REG_R2, TCG_REG_R3
82 };
83 static const int tcg_target_call_oarg_regs[2] = {
84 TCG_REG_R0, TCG_REG_R1
85 };
86
87 #define TCG_REG_TMP TCG_REG_R12
88
89 enum arm_cond_code_e {
90 COND_EQ = 0x0,
91 COND_NE = 0x1,
92 COND_CS = 0x2, /* Unsigned greater or equal */
93 COND_CC = 0x3, /* Unsigned less than */
94 COND_MI = 0x4, /* Negative */
95 COND_PL = 0x5, /* Zero or greater */
96 COND_VS = 0x6, /* Overflow */
97 COND_VC = 0x7, /* No overflow */
98 COND_HI = 0x8, /* Unsigned greater than */
99 COND_LS = 0x9, /* Unsigned less or equal */
100 COND_GE = 0xa,
101 COND_LT = 0xb,
102 COND_GT = 0xc,
103 COND_LE = 0xd,
104 COND_AL = 0xe,
105 };
106
107 #define TO_CPSR (1 << 20)
108
109 #define SHIFT_IMM_LSL(im) (((im) << 7) | 0x00)
110 #define SHIFT_IMM_LSR(im) (((im) << 7) | 0x20)
111 #define SHIFT_IMM_ASR(im) (((im) << 7) | 0x40)
112 #define SHIFT_IMM_ROR(im) (((im) << 7) | 0x60)
113 #define SHIFT_REG_LSL(rs) (((rs) << 8) | 0x10)
114 #define SHIFT_REG_LSR(rs) (((rs) << 8) | 0x30)
115 #define SHIFT_REG_ASR(rs) (((rs) << 8) | 0x50)
116 #define SHIFT_REG_ROR(rs) (((rs) << 8) | 0x70)
117
118 typedef enum {
119 ARITH_AND = 0x0 << 21,
120 ARITH_EOR = 0x1 << 21,
121 ARITH_SUB = 0x2 << 21,
122 ARITH_RSB = 0x3 << 21,
123 ARITH_ADD = 0x4 << 21,
124 ARITH_ADC = 0x5 << 21,
125 ARITH_SBC = 0x6 << 21,
126 ARITH_RSC = 0x7 << 21,
127 ARITH_TST = 0x8 << 21 | TO_CPSR,
128 ARITH_CMP = 0xa << 21 | TO_CPSR,
129 ARITH_CMN = 0xb << 21 | TO_CPSR,
130 ARITH_ORR = 0xc << 21,
131 ARITH_MOV = 0xd << 21,
132 ARITH_BIC = 0xe << 21,
133 ARITH_MVN = 0xf << 21,
134
135 INSN_CLZ = 0x016f0f10,
136 INSN_RBIT = 0x06ff0f30,
137
138 INSN_LDR_IMM = 0x04100000,
139 INSN_LDR_REG = 0x06100000,
140 INSN_STR_IMM = 0x04000000,
141 INSN_STR_REG = 0x06000000,
142
143 INSN_LDRH_IMM = 0x005000b0,
144 INSN_LDRH_REG = 0x001000b0,
145 INSN_LDRSH_IMM = 0x005000f0,
146 INSN_LDRSH_REG = 0x001000f0,
147 INSN_STRH_IMM = 0x004000b0,
148 INSN_STRH_REG = 0x000000b0,
149
150 INSN_LDRB_IMM = 0x04500000,
151 INSN_LDRB_REG = 0x06500000,
152 INSN_LDRSB_IMM = 0x005000d0,
153 INSN_LDRSB_REG = 0x001000d0,
154 INSN_STRB_IMM = 0x04400000,
155 INSN_STRB_REG = 0x06400000,
156
157 INSN_LDRD_IMM = 0x004000d0,
158 INSN_LDRD_REG = 0x000000d0,
159 INSN_STRD_IMM = 0x004000f0,
160 INSN_STRD_REG = 0x000000f0,
161
162 INSN_DMB_ISH = 0x5bf07ff5,
163 INSN_DMB_MCR = 0xba0f07ee,
164
165 /* Architected nop introduced in v6k. */
166 /* ??? This is an MSR (imm) 0,0,0 insn. Anyone know if this
167 also Just So Happened to do nothing on pre-v6k so that we
168 don't need to conditionalize it? */
169 INSN_NOP_v6k = 0xe320f000,
170 /* Otherwise the assembler uses mov r0,r0 */
171 INSN_NOP_v4 = (COND_AL << 28) | ARITH_MOV,
172 } ARMInsn;
173
174 #define INSN_NOP (use_armv7_instructions ? INSN_NOP_v6k : INSN_NOP_v4)
175
176 static const uint8_t tcg_cond_to_arm_cond[] = {
177 [TCG_COND_EQ] = COND_EQ,
178 [TCG_COND_NE] = COND_NE,
179 [TCG_COND_LT] = COND_LT,
180 [TCG_COND_GE] = COND_GE,
181 [TCG_COND_LE] = COND_LE,
182 [TCG_COND_GT] = COND_GT,
183 /* unsigned */
184 [TCG_COND_LTU] = COND_CC,
185 [TCG_COND_GEU] = COND_CS,
186 [TCG_COND_LEU] = COND_LS,
187 [TCG_COND_GTU] = COND_HI,
188 };
189
190 static inline void reloc_pc24(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
191 {
192 ptrdiff_t offset = (tcg_ptr_byte_diff(target, code_ptr) - 8) >> 2;
193 *code_ptr = (*code_ptr & ~0xffffff) | (offset & 0xffffff);
194 }
195
196 static inline void reloc_pc24_atomic(tcg_insn_unit *code_ptr, tcg_insn_unit *target)
197 {
198 ptrdiff_t offset = (tcg_ptr_byte_diff(target, code_ptr) - 8) >> 2;
199 tcg_insn_unit insn = atomic_read(code_ptr);
200 tcg_debug_assert(offset == sextract32(offset, 0, 24));
201 atomic_set(code_ptr, deposit32(insn, 0, 24, offset));
202 }
203
204 static void patch_reloc(tcg_insn_unit *code_ptr, int type,
205 intptr_t value, intptr_t addend)
206 {
207 tcg_debug_assert(addend == 0);
208
209 if (type == R_ARM_PC24) {
210 reloc_pc24(code_ptr, (tcg_insn_unit *)value);
211 } else if (type == R_ARM_PC13) {
212 intptr_t diff = value - (uintptr_t)(code_ptr + 2);
213 tcg_insn_unit insn = *code_ptr;
214 bool u;
215
216 if (diff >= -0xfff && diff <= 0xfff) {
217 u = (diff >= 0);
218 if (!u) {
219 diff = -diff;
220 }
221 } else {
222 int rd = extract32(insn, 12, 4);
223 int rt = rd == TCG_REG_PC ? TCG_REG_TMP : rd;
224 assert(diff >= 0x1000 && diff < 0x100000);
225 /* add rt, pc, #high */
226 *code_ptr++ = ((insn & 0xf0000000) | (1 << 25) | ARITH_ADD
227 | (TCG_REG_PC << 16) | (rt << 12)
228 | (20 << 7) | (diff >> 12));
229 /* ldr rd, [rt, #low] */
230 insn = deposit32(insn, 12, 4, rt);
231 diff &= 0xfff;
232 u = 1;
233 }
234 insn = deposit32(insn, 23, 1, u);
235 insn = deposit32(insn, 0, 12, diff);
236 *code_ptr = insn;
237 } else {
238 g_assert_not_reached();
239 }
240 }
241
242 #define TCG_CT_CONST_ARM 0x100
243 #define TCG_CT_CONST_INV 0x200
244 #define TCG_CT_CONST_NEG 0x400
245 #define TCG_CT_CONST_ZERO 0x800
246
247 /* parse target specific constraints */
248 static const char *target_parse_constraint(TCGArgConstraint *ct,
249 const char *ct_str, TCGType type)
250 {
251 switch (*ct_str++) {
252 case 'I':
253 ct->ct |= TCG_CT_CONST_ARM;
254 break;
255 case 'K':
256 ct->ct |= TCG_CT_CONST_INV;
257 break;
258 case 'N': /* The gcc constraint letter is L, already used here. */
259 ct->ct |= TCG_CT_CONST_NEG;
260 break;
261 case 'Z':
262 ct->ct |= TCG_CT_CONST_ZERO;
263 break;
264
265 case 'r':
266 ct->ct |= TCG_CT_REG;
267 ct->u.regs = 0xffff;
268 break;
269
270 /* qemu_ld address */
271 case 'l':
272 ct->ct |= TCG_CT_REG;
273 ct->u.regs = 0xffff;
274 #ifdef CONFIG_SOFTMMU
275 /* r0-r2,lr will be overwritten when reading the tlb entry,
276 so don't use these. */
277 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0);
278 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1);
279 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R2);
280 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R14);
281 #endif
282 break;
283
284 /* qemu_st address & data */
285 case 's':
286 ct->ct |= TCG_CT_REG;
287 ct->u.regs = 0xffff;
288 /* r0-r2 will be overwritten when reading the tlb entry (softmmu only)
289 and r0-r1 doing the byte swapping, so don't use these. */
290 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0);
291 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1);
292 #if defined(CONFIG_SOFTMMU)
293 /* Avoid clashes with registers being used for helper args */
294 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R2);
295 #if TARGET_LONG_BITS == 64
296 /* Avoid clashes with registers being used for helper args */
297 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3);
298 #endif
299 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R14);
300 #endif
301 break;
302
303 default:
304 return NULL;
305 }
306 return ct_str;
307 }
308
309 static inline uint32_t rotl(uint32_t val, int n)
310 {
311 return (val << n) | (val >> (32 - n));
312 }
313
314 /* ARM immediates for ALU instructions are made of an unsigned 8-bit
315 right-rotated by an even amount between 0 and 30. */
316 static inline int encode_imm(uint32_t imm)
317 {
318 int shift;
319
320 /* simple case, only lower bits */
321 if ((imm & ~0xff) == 0)
322 return 0;
323 /* then try a simple even shift */
324 shift = ctz32(imm) & ~1;
325 if (((imm >> shift) & ~0xff) == 0)
326 return 32 - shift;
327 /* now try harder with rotations */
328 if ((rotl(imm, 2) & ~0xff) == 0)
329 return 2;
330 if ((rotl(imm, 4) & ~0xff) == 0)
331 return 4;
332 if ((rotl(imm, 6) & ~0xff) == 0)
333 return 6;
334 /* imm can't be encoded */
335 return -1;
336 }
337
338 static inline int check_fit_imm(uint32_t imm)
339 {
340 return encode_imm(imm) >= 0;
341 }
342
343 /* Test if a constant matches the constraint.
344 * TODO: define constraints for:
345 *
346 * ldr/str offset: between -0xfff and 0xfff
347 * ldrh/strh offset: between -0xff and 0xff
348 * mov operand2: values represented with x << (2 * y), x < 0x100
349 * add, sub, eor...: ditto
350 */
351 static inline int tcg_target_const_match(tcg_target_long val, TCGType type,
352 const TCGArgConstraint *arg_ct)
353 {
354 int ct;
355 ct = arg_ct->ct;
356 if (ct & TCG_CT_CONST) {
357 return 1;
358 } else if ((ct & TCG_CT_CONST_ARM) && check_fit_imm(val)) {
359 return 1;
360 } else if ((ct & TCG_CT_CONST_INV) && check_fit_imm(~val)) {
361 return 1;
362 } else if ((ct & TCG_CT_CONST_NEG) && check_fit_imm(-val)) {
363 return 1;
364 } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
365 return 1;
366 } else {
367 return 0;
368 }
369 }
370
371 static inline void tcg_out_b(TCGContext *s, int cond, int32_t offset)
372 {
373 tcg_out32(s, (cond << 28) | 0x0a000000 |
374 (((offset - 8) >> 2) & 0x00ffffff));
375 }
376
377 static inline void tcg_out_b_noaddr(TCGContext *s, int cond)
378 {
379 /* We pay attention here to not modify the branch target by masking
380 the corresponding bytes. This ensure that caches and memory are
381 kept coherent during retranslation. */
382 tcg_out32(s, deposit32(*s->code_ptr, 24, 8, (cond << 4) | 0x0a));
383 }
384
385 static inline void tcg_out_bl_noaddr(TCGContext *s, int cond)
386 {
387 /* We pay attention here to not modify the branch target by masking
388 the corresponding bytes. This ensure that caches and memory are
389 kept coherent during retranslation. */
390 tcg_out32(s, deposit32(*s->code_ptr, 24, 8, (cond << 4) | 0x0b));
391 }
392
393 static inline void tcg_out_bl(TCGContext *s, int cond, int32_t offset)
394 {
395 tcg_out32(s, (cond << 28) | 0x0b000000 |
396 (((offset - 8) >> 2) & 0x00ffffff));
397 }
398
399 static inline void tcg_out_blx(TCGContext *s, int cond, int rn)
400 {
401 tcg_out32(s, (cond << 28) | 0x012fff30 | rn);
402 }
403
404 static inline void tcg_out_blx_imm(TCGContext *s, int32_t offset)
405 {
406 tcg_out32(s, 0xfa000000 | ((offset & 2) << 23) |
407 (((offset - 8) >> 2) & 0x00ffffff));
408 }
409
410 static inline void tcg_out_dat_reg(TCGContext *s,
411 int cond, int opc, int rd, int rn, int rm, int shift)
412 {
413 tcg_out32(s, (cond << 28) | (0 << 25) | opc |
414 (rn << 16) | (rd << 12) | shift | rm);
415 }
416
417 static inline void tcg_out_nop(TCGContext *s)
418 {
419 tcg_out32(s, INSN_NOP);
420 }
421
422 static inline void tcg_out_mov_reg(TCGContext *s, int cond, int rd, int rm)
423 {
424 /* Simple reg-reg move, optimising out the 'do nothing' case */
425 if (rd != rm) {
426 tcg_out_dat_reg(s, cond, ARITH_MOV, rd, 0, rm, SHIFT_IMM_LSL(0));
427 }
428 }
429
430 static inline void tcg_out_bx(TCGContext *s, int cond, TCGReg rn)
431 {
432 /* Unless the C portion of QEMU is compiled as thumb, we don't
433 actually need true BX semantics; merely a branch to an address
434 held in a register. */
435 if (use_armv5t_instructions) {
436 tcg_out32(s, (cond << 28) | 0x012fff10 | rn);
437 } else {
438 tcg_out_mov_reg(s, cond, TCG_REG_PC, rn);
439 }
440 }
441
442 static inline void tcg_out_dat_imm(TCGContext *s,
443 int cond, int opc, int rd, int rn, int im)
444 {
445 tcg_out32(s, (cond << 28) | (1 << 25) | opc |
446 (rn << 16) | (rd << 12) | im);
447 }
448
449 /* Note that this routine is used for both LDR and LDRH formats, so we do
450 not wish to include an immediate shift at this point. */
451 static void tcg_out_memop_r(TCGContext *s, int cond, ARMInsn opc, TCGReg rt,
452 TCGReg rn, TCGReg rm, bool u, bool p, bool w)
453 {
454 tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24)
455 | (w << 21) | (rn << 16) | (rt << 12) | rm);
456 }
457
458 static void tcg_out_memop_8(TCGContext *s, int cond, ARMInsn opc, TCGReg rt,
459 TCGReg rn, int imm8, bool p, bool w)
460 {
461 bool u = 1;
462 if (imm8 < 0) {
463 imm8 = -imm8;
464 u = 0;
465 }
466 tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
467 (rn << 16) | (rt << 12) | ((imm8 & 0xf0) << 4) | (imm8 & 0xf));
468 }
469
470 static void tcg_out_memop_12(TCGContext *s, int cond, ARMInsn opc, TCGReg rt,
471 TCGReg rn, int imm12, bool p, bool w)
472 {
473 bool u = 1;
474 if (imm12 < 0) {
475 imm12 = -imm12;
476 u = 0;
477 }
478 tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
479 (rn << 16) | (rt << 12) | imm12);
480 }
481
482 static inline void tcg_out_ld32_12(TCGContext *s, int cond, TCGReg rt,
483 TCGReg rn, int imm12)
484 {
485 tcg_out_memop_12(s, cond, INSN_LDR_IMM, rt, rn, imm12, 1, 0);
486 }
487
488 static inline void tcg_out_st32_12(TCGContext *s, int cond, TCGReg rt,
489 TCGReg rn, int imm12)
490 {
491 tcg_out_memop_12(s, cond, INSN_STR_IMM, rt, rn, imm12, 1, 0);
492 }
493
494 static inline void tcg_out_ld32_r(TCGContext *s, int cond, TCGReg rt,
495 TCGReg rn, TCGReg rm)
496 {
497 tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 0);
498 }
499
500 static inline void tcg_out_st32_r(TCGContext *s, int cond, TCGReg rt,
501 TCGReg rn, TCGReg rm)
502 {
503 tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 0);
504 }
505
506 static inline void tcg_out_ldrd_8(TCGContext *s, int cond, TCGReg rt,
507 TCGReg rn, int imm8)
508 {
509 tcg_out_memop_8(s, cond, INSN_LDRD_IMM, rt, rn, imm8, 1, 0);
510 }
511
512 static inline void tcg_out_ldrd_r(TCGContext *s, int cond, TCGReg rt,
513 TCGReg rn, TCGReg rm)
514 {
515 tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 0);
516 }
517
518 static inline void tcg_out_strd_8(TCGContext *s, int cond, TCGReg rt,
519 TCGReg rn, int imm8)
520 {
521 tcg_out_memop_8(s, cond, INSN_STRD_IMM, rt, rn, imm8, 1, 0);
522 }
523
524 static inline void tcg_out_strd_r(TCGContext *s, int cond, TCGReg rt,
525 TCGReg rn, TCGReg rm)
526 {
527 tcg_out_memop_r(s, cond, INSN_STRD_REG, rt, rn, rm, 1, 1, 0);
528 }
529
530 /* Register pre-increment with base writeback. */
531 static inline void tcg_out_ld32_rwb(TCGContext *s, int cond, TCGReg rt,
532 TCGReg rn, TCGReg rm)
533 {
534 tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 1);
535 }
536
537 static inline void tcg_out_st32_rwb(TCGContext *s, int cond, TCGReg rt,
538 TCGReg rn, TCGReg rm)
539 {
540 tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 1);
541 }
542
543 static inline void tcg_out_ld16u_8(TCGContext *s, int cond, TCGReg rt,
544 TCGReg rn, int imm8)
545 {
546 tcg_out_memop_8(s, cond, INSN_LDRH_IMM, rt, rn, imm8, 1, 0);
547 }
548
549 static inline void tcg_out_st16_8(TCGContext *s, int cond, TCGReg rt,
550 TCGReg rn, int imm8)
551 {
552 tcg_out_memop_8(s, cond, INSN_STRH_IMM, rt, rn, imm8, 1, 0);
553 }
554
555 static inline void tcg_out_ld16u_r(TCGContext *s, int cond, TCGReg rt,
556 TCGReg rn, TCGReg rm)
557 {
558 tcg_out_memop_r(s, cond, INSN_LDRH_REG, rt, rn, rm, 1, 1, 0);
559 }
560
561 static inline void tcg_out_st16_r(TCGContext *s, int cond, TCGReg rt,
562 TCGReg rn, TCGReg rm)
563 {
564 tcg_out_memop_r(s, cond, INSN_STRH_REG, rt, rn, rm, 1, 1, 0);
565 }
566
567 static inline void tcg_out_ld16s_8(TCGContext *s, int cond, TCGReg rt,
568 TCGReg rn, int imm8)
569 {
570 tcg_out_memop_8(s, cond, INSN_LDRSH_IMM, rt, rn, imm8, 1, 0);
571 }
572
573 static inline void tcg_out_ld16s_r(TCGContext *s, int cond, TCGReg rt,
574 TCGReg rn, TCGReg rm)
575 {
576 tcg_out_memop_r(s, cond, INSN_LDRSH_REG, rt, rn, rm, 1, 1, 0);
577 }
578
579 static inline void tcg_out_ld8_12(TCGContext *s, int cond, TCGReg rt,
580 TCGReg rn, int imm12)
581 {
582 tcg_out_memop_12(s, cond, INSN_LDRB_IMM, rt, rn, imm12, 1, 0);
583 }
584
585 static inline void tcg_out_st8_12(TCGContext *s, int cond, TCGReg rt,
586 TCGReg rn, int imm12)
587 {
588 tcg_out_memop_12(s, cond, INSN_STRB_IMM, rt, rn, imm12, 1, 0);
589 }
590
591 static inline void tcg_out_ld8_r(TCGContext *s, int cond, TCGReg rt,
592 TCGReg rn, TCGReg rm)
593 {
594 tcg_out_memop_r(s, cond, INSN_LDRB_REG, rt, rn, rm, 1, 1, 0);
595 }
596
597 static inline void tcg_out_st8_r(TCGContext *s, int cond, TCGReg rt,
598 TCGReg rn, TCGReg rm)
599 {
600 tcg_out_memop_r(s, cond, INSN_STRB_REG, rt, rn, rm, 1, 1, 0);
601 }
602
603 static inline void tcg_out_ld8s_8(TCGContext *s, int cond, TCGReg rt,
604 TCGReg rn, int imm8)
605 {
606 tcg_out_memop_8(s, cond, INSN_LDRSB_IMM, rt, rn, imm8, 1, 0);
607 }
608
609 static inline void tcg_out_ld8s_r(TCGContext *s, int cond, TCGReg rt,
610 TCGReg rn, TCGReg rm)
611 {
612 tcg_out_memop_r(s, cond, INSN_LDRSB_REG, rt, rn, rm, 1, 1, 0);
613 }
614
615 static void tcg_out_movi_pool(TCGContext *s, int cond, int rd, uint32_t arg)
616 {
617 /* The 12-bit range on the ldr insn is sometimes a bit too small.
618 In order to get around that we require two insns, one of which
619 will usually be a nop, but may be replaced in patch_reloc. */
620 new_pool_label(s, arg, R_ARM_PC13, s->code_ptr, 0);
621 tcg_out_ld32_12(s, cond, rd, TCG_REG_PC, 0);
622 tcg_out_nop(s);
623 }
624
625 static void tcg_out_movi32(TCGContext *s, int cond, int rd, uint32_t arg)
626 {
627 int rot, diff, opc, sh1, sh2;
628 uint32_t tt0, tt1, tt2;
629
630 /* Check a single MOV/MVN before anything else. */
631 rot = encode_imm(arg);
632 if (rot >= 0) {
633 tcg_out_dat_imm(s, cond, ARITH_MOV, rd, 0,
634 rotl(arg, rot) | (rot << 7));
635 return;
636 }
637 rot = encode_imm(~arg);
638 if (rot >= 0) {
639 tcg_out_dat_imm(s, cond, ARITH_MVN, rd, 0,
640 rotl(~arg, rot) | (rot << 7));
641 return;
642 }
643
644 /* Check for a pc-relative address. This will usually be the TB,
645 or within the TB, which is immediately before the code block. */
646 diff = arg - ((intptr_t)s->code_ptr + 8);
647 if (diff >= 0) {
648 rot = encode_imm(diff);
649 if (rot >= 0) {
650 tcg_out_dat_imm(s, cond, ARITH_ADD, rd, TCG_REG_PC,
651 rotl(diff, rot) | (rot << 7));
652 return;
653 }
654 } else {
655 rot = encode_imm(-diff);
656 if (rot >= 0) {
657 tcg_out_dat_imm(s, cond, ARITH_SUB, rd, TCG_REG_PC,
658 rotl(-diff, rot) | (rot << 7));
659 return;
660 }
661 }
662
663 /* Use movw + movt. */
664 if (use_armv7_instructions) {
665 /* movw */
666 tcg_out32(s, (cond << 28) | 0x03000000 | (rd << 12)
667 | ((arg << 4) & 0x000f0000) | (arg & 0xfff));
668 if (arg & 0xffff0000) {
669 /* movt */
670 tcg_out32(s, (cond << 28) | 0x03400000 | (rd << 12)
671 | ((arg >> 12) & 0x000f0000) | ((arg >> 16) & 0xfff));
672 }
673 return;
674 }
675
676 /* Look for sequences of two insns. If we have lots of 1's, we can
677 shorten the sequence by beginning with mvn and then clearing
678 higher bits with eor. */
679 tt0 = arg;
680 opc = ARITH_MOV;
681 if (ctpop32(arg) > 16) {
682 tt0 = ~arg;
683 opc = ARITH_MVN;
684 }
685 sh1 = ctz32(tt0) & ~1;
686 tt1 = tt0 & ~(0xff << sh1);
687 sh2 = ctz32(tt1) & ~1;
688 tt2 = tt1 & ~(0xff << sh2);
689 if (tt2 == 0) {
690 rot = ((32 - sh1) << 7) & 0xf00;
691 tcg_out_dat_imm(s, cond, opc, rd, 0, ((tt0 >> sh1) & 0xff) | rot);
692 rot = ((32 - sh2) << 7) & 0xf00;
693 tcg_out_dat_imm(s, cond, ARITH_EOR, rd, rd,
694 ((tt0 >> sh2) & 0xff) | rot);
695 return;
696 }
697
698 /* Otherwise, drop it into the constant pool. */
699 tcg_out_movi_pool(s, cond, rd, arg);
700 }
701
702 static inline void tcg_out_dat_rI(TCGContext *s, int cond, int opc, TCGArg dst,
703 TCGArg lhs, TCGArg rhs, int rhs_is_const)
704 {
705 /* Emit either the reg,imm or reg,reg form of a data-processing insn.
706 * rhs must satisfy the "rI" constraint.
707 */
708 if (rhs_is_const) {
709 int rot = encode_imm(rhs);
710 tcg_debug_assert(rot >= 0);
711 tcg_out_dat_imm(s, cond, opc, dst, lhs, rotl(rhs, rot) | (rot << 7));
712 } else {
713 tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
714 }
715 }
716
717 static void tcg_out_dat_rIK(TCGContext *s, int cond, int opc, int opinv,
718 TCGReg dst, TCGReg lhs, TCGArg rhs,
719 bool rhs_is_const)
720 {
721 /* Emit either the reg,imm or reg,reg form of a data-processing insn.
722 * rhs must satisfy the "rIK" constraint.
723 */
724 if (rhs_is_const) {
725 int rot = encode_imm(rhs);
726 if (rot < 0) {
727 rhs = ~rhs;
728 rot = encode_imm(rhs);
729 tcg_debug_assert(rot >= 0);
730 opc = opinv;
731 }
732 tcg_out_dat_imm(s, cond, opc, dst, lhs, rotl(rhs, rot) | (rot << 7));
733 } else {
734 tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
735 }
736 }
737
738 static void tcg_out_dat_rIN(TCGContext *s, int cond, int opc, int opneg,
739 TCGArg dst, TCGArg lhs, TCGArg rhs,
740 bool rhs_is_const)
741 {
742 /* Emit either the reg,imm or reg,reg form of a data-processing insn.
743 * rhs must satisfy the "rIN" constraint.
744 */
745 if (rhs_is_const) {
746 int rot = encode_imm(rhs);
747 if (rot < 0) {
748 rhs = -rhs;
749 rot = encode_imm(rhs);
750 tcg_debug_assert(rot >= 0);
751 opc = opneg;
752 }
753 tcg_out_dat_imm(s, cond, opc, dst, lhs, rotl(rhs, rot) | (rot << 7));
754 } else {
755 tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
756 }
757 }
758
759 static inline void tcg_out_mul32(TCGContext *s, int cond, TCGReg rd,
760 TCGReg rn, TCGReg rm)
761 {
762 /* if ArchVersion() < 6 && d == n then UNPREDICTABLE; */
763 if (!use_armv6_instructions && rd == rn) {
764 if (rd == rm) {
765 /* rd == rn == rm; copy an input to tmp first. */
766 tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
767 rm = rn = TCG_REG_TMP;
768 } else {
769 rn = rm;
770 rm = rd;
771 }
772 }
773 /* mul */
774 tcg_out32(s, (cond << 28) | 0x90 | (rd << 16) | (rm << 8) | rn);
775 }
776
777 static inline void tcg_out_umull32(TCGContext *s, int cond, TCGReg rd0,
778 TCGReg rd1, TCGReg rn, TCGReg rm)
779 {
780 /* if ArchVersion() < 6 && (dHi == n || dLo == n) then UNPREDICTABLE; */
781 if (!use_armv6_instructions && (rd0 == rn || rd1 == rn)) {
782 if (rd0 == rm || rd1 == rm) {
783 tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
784 rn = TCG_REG_TMP;
785 } else {
786 TCGReg t = rn;
787 rn = rm;
788 rm = t;
789 }
790 }
791 /* umull */
792 tcg_out32(s, (cond << 28) | 0x00800090 |
793 (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
794 }
795
796 static inline void tcg_out_smull32(TCGContext *s, int cond, TCGReg rd0,
797 TCGReg rd1, TCGReg rn, TCGReg rm)
798 {
799 /* if ArchVersion() < 6 && (dHi == n || dLo == n) then UNPREDICTABLE; */
800 if (!use_armv6_instructions && (rd0 == rn || rd1 == rn)) {
801 if (rd0 == rm || rd1 == rm) {
802 tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
803 rn = TCG_REG_TMP;
804 } else {
805 TCGReg t = rn;
806 rn = rm;
807 rm = t;
808 }
809 }
810 /* smull */
811 tcg_out32(s, (cond << 28) | 0x00c00090 |
812 (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
813 }
814
815 static inline void tcg_out_sdiv(TCGContext *s, int cond, int rd, int rn, int rm)
816 {
817 tcg_out32(s, 0x0710f010 | (cond << 28) | (rd << 16) | rn | (rm << 8));
818 }
819
820 static inline void tcg_out_udiv(TCGContext *s, int cond, int rd, int rn, int rm)
821 {
822 tcg_out32(s, 0x0730f010 | (cond << 28) | (rd << 16) | rn | (rm << 8));
823 }
824
825 static inline void tcg_out_ext8s(TCGContext *s, int cond,
826 int rd, int rn)
827 {
828 if (use_armv6_instructions) {
829 /* sxtb */
830 tcg_out32(s, 0x06af0070 | (cond << 28) | (rd << 12) | rn);
831 } else {
832 tcg_out_dat_reg(s, cond, ARITH_MOV,
833 rd, 0, rn, SHIFT_IMM_LSL(24));
834 tcg_out_dat_reg(s, cond, ARITH_MOV,
835 rd, 0, rd, SHIFT_IMM_ASR(24));
836 }
837 }
838
839 static inline void tcg_out_ext8u(TCGContext *s, int cond,
840 int rd, int rn)
841 {
842 tcg_out_dat_imm(s, cond, ARITH_AND, rd, rn, 0xff);
843 }
844
845 static inline void tcg_out_ext16s(TCGContext *s, int cond,
846 int rd, int rn)
847 {
848 if (use_armv6_instructions) {
849 /* sxth */
850 tcg_out32(s, 0x06bf0070 | (cond << 28) | (rd << 12) | rn);
851 } else {
852 tcg_out_dat_reg(s, cond, ARITH_MOV,
853 rd, 0, rn, SHIFT_IMM_LSL(16));
854 tcg_out_dat_reg(s, cond, ARITH_MOV,
855 rd, 0, rd, SHIFT_IMM_ASR(16));
856 }
857 }
858
859 static inline void tcg_out_ext16u(TCGContext *s, int cond,
860 int rd, int rn)
861 {
862 if (use_armv6_instructions) {
863 /* uxth */
864 tcg_out32(s, 0x06ff0070 | (cond << 28) | (rd << 12) | rn);
865 } else {
866 tcg_out_dat_reg(s, cond, ARITH_MOV,
867 rd, 0, rn, SHIFT_IMM_LSL(16));
868 tcg_out_dat_reg(s, cond, ARITH_MOV,
869 rd, 0, rd, SHIFT_IMM_LSR(16));
870 }
871 }
872
873 static inline void tcg_out_bswap16s(TCGContext *s, int cond, int rd, int rn)
874 {
875 if (use_armv6_instructions) {
876 /* revsh */
877 tcg_out32(s, 0x06ff0fb0 | (cond << 28) | (rd << 12) | rn);
878 } else {
879 tcg_out_dat_reg(s, cond, ARITH_MOV,
880 TCG_REG_TMP, 0, rn, SHIFT_IMM_LSL(24));
881 tcg_out_dat_reg(s, cond, ARITH_MOV,
882 TCG_REG_TMP, 0, TCG_REG_TMP, SHIFT_IMM_ASR(16));
883 tcg_out_dat_reg(s, cond, ARITH_ORR,
884 rd, TCG_REG_TMP, rn, SHIFT_IMM_LSR(8));
885 }
886 }
887
888 static inline void tcg_out_bswap16(TCGContext *s, int cond, int rd, int rn)
889 {
890 if (use_armv6_instructions) {
891 /* rev16 */
892 tcg_out32(s, 0x06bf0fb0 | (cond << 28) | (rd << 12) | rn);
893 } else {
894 tcg_out_dat_reg(s, cond, ARITH_MOV,
895 TCG_REG_TMP, 0, rn, SHIFT_IMM_LSL(24));
896 tcg_out_dat_reg(s, cond, ARITH_MOV,
897 TCG_REG_TMP, 0, TCG_REG_TMP, SHIFT_IMM_LSR(16));
898 tcg_out_dat_reg(s, cond, ARITH_ORR,
899 rd, TCG_REG_TMP, rn, SHIFT_IMM_LSR(8));
900 }
901 }
902
903 /* swap the two low bytes assuming that the two high input bytes and the
904 two high output bit can hold any value. */
905 static inline void tcg_out_bswap16st(TCGContext *s, int cond, int rd, int rn)
906 {
907 if (use_armv6_instructions) {
908 /* rev16 */
909 tcg_out32(s, 0x06bf0fb0 | (cond << 28) | (rd << 12) | rn);
910 } else {
911 tcg_out_dat_reg(s, cond, ARITH_MOV,
912 TCG_REG_TMP, 0, rn, SHIFT_IMM_LSR(8));
913 tcg_out_dat_imm(s, cond, ARITH_AND, TCG_REG_TMP, TCG_REG_TMP, 0xff);
914 tcg_out_dat_reg(s, cond, ARITH_ORR,
915 rd, TCG_REG_TMP, rn, SHIFT_IMM_LSL(8));
916 }
917 }
918
919 static inline void tcg_out_bswap32(TCGContext *s, int cond, int rd, int rn)
920 {
921 if (use_armv6_instructions) {
922 /* rev */
923 tcg_out32(s, 0x06bf0f30 | (cond << 28) | (rd << 12) | rn);
924 } else {
925 tcg_out_dat_reg(s, cond, ARITH_EOR,
926 TCG_REG_TMP, rn, rn, SHIFT_IMM_ROR(16));
927 tcg_out_dat_imm(s, cond, ARITH_BIC,
928 TCG_REG_TMP, TCG_REG_TMP, 0xff | 0x800);
929 tcg_out_dat_reg(s, cond, ARITH_MOV,
930 rd, 0, rn, SHIFT_IMM_ROR(8));
931 tcg_out_dat_reg(s, cond, ARITH_EOR,
932 rd, rd, TCG_REG_TMP, SHIFT_IMM_LSR(8));
933 }
934 }
935
936 static inline void tcg_out_deposit(TCGContext *s, int cond, TCGReg rd,
937 TCGArg a1, int ofs, int len, bool const_a1)
938 {
939 if (const_a1) {
940 /* bfi becomes bfc with rn == 15. */
941 a1 = 15;
942 }
943 /* bfi/bfc */
944 tcg_out32(s, 0x07c00010 | (cond << 28) | (rd << 12) | a1
945 | (ofs << 7) | ((ofs + len - 1) << 16));
946 }
947
948 static inline void tcg_out_extract(TCGContext *s, int cond, TCGReg rd,
949 TCGArg a1, int ofs, int len)
950 {
951 /* ubfx */
952 tcg_out32(s, 0x07e00050 | (cond << 28) | (rd << 12) | a1
953 | (ofs << 7) | ((len - 1) << 16));
954 }
955
956 static inline void tcg_out_sextract(TCGContext *s, int cond, TCGReg rd,
957 TCGArg a1, int ofs, int len)
958 {
959 /* sbfx */
960 tcg_out32(s, 0x07a00050 | (cond << 28) | (rd << 12) | a1
961 | (ofs << 7) | ((len - 1) << 16));
962 }
963
964 static inline void tcg_out_ld32u(TCGContext *s, int cond,
965 int rd, int rn, int32_t offset)
966 {
967 if (offset > 0xfff || offset < -0xfff) {
968 tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
969 tcg_out_ld32_r(s, cond, rd, rn, TCG_REG_TMP);
970 } else
971 tcg_out_ld32_12(s, cond, rd, rn, offset);
972 }
973
974 static inline void tcg_out_st32(TCGContext *s, int cond,
975 int rd, int rn, int32_t offset)
976 {
977 if (offset > 0xfff || offset < -0xfff) {
978 tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
979 tcg_out_st32_r(s, cond, rd, rn, TCG_REG_TMP);
980 } else
981 tcg_out_st32_12(s, cond, rd, rn, offset);
982 }
983
984 static inline void tcg_out_ld16u(TCGContext *s, int cond,
985 int rd, int rn, int32_t offset)
986 {
987 if (offset > 0xff || offset < -0xff) {
988 tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
989 tcg_out_ld16u_r(s, cond, rd, rn, TCG_REG_TMP);
990 } else
991 tcg_out_ld16u_8(s, cond, rd, rn, offset);
992 }
993
994 static inline void tcg_out_ld16s(TCGContext *s, int cond,
995 int rd, int rn, int32_t offset)
996 {
997 if (offset > 0xff || offset < -0xff) {
998 tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
999 tcg_out_ld16s_r(s, cond, rd, rn, TCG_REG_TMP);
1000 } else
1001 tcg_out_ld16s_8(s, cond, rd, rn, offset);
1002 }
1003
1004 static inline void tcg_out_st16(TCGContext *s, int cond,
1005 int rd, int rn, int32_t offset)
1006 {
1007 if (offset > 0xff || offset < -0xff) {
1008 tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1009 tcg_out_st16_r(s, cond, rd, rn, TCG_REG_TMP);
1010 } else
1011 tcg_out_st16_8(s, cond, rd, rn, offset);
1012 }
1013
1014 static inline void tcg_out_ld8u(TCGContext *s, int cond,
1015 int rd, int rn, int32_t offset)
1016 {
1017 if (offset > 0xfff || offset < -0xfff) {
1018 tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1019 tcg_out_ld8_r(s, cond, rd, rn, TCG_REG_TMP);
1020 } else
1021 tcg_out_ld8_12(s, cond, rd, rn, offset);
1022 }
1023
1024 static inline void tcg_out_ld8s(TCGContext *s, int cond,
1025 int rd, int rn, int32_t offset)
1026 {
1027 if (offset > 0xff || offset < -0xff) {
1028 tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1029 tcg_out_ld8s_r(s, cond, rd, rn, TCG_REG_TMP);
1030 } else
1031 tcg_out_ld8s_8(s, cond, rd, rn, offset);
1032 }
1033
1034 static inline void tcg_out_st8(TCGContext *s, int cond,
1035 int rd, int rn, int32_t offset)
1036 {
1037 if (offset > 0xfff || offset < -0xfff) {
1038 tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1039 tcg_out_st8_r(s, cond, rd, rn, TCG_REG_TMP);
1040 } else
1041 tcg_out_st8_12(s, cond, rd, rn, offset);
1042 }
1043
1044 /* The _goto case is normally between TBs within the same code buffer, and
1045 * with the code buffer limited to 16MB we wouldn't need the long case.
1046 * But we also use it for the tail-call to the qemu_ld/st helpers, which does.
1047 */
1048 static void tcg_out_goto(TCGContext *s, int cond, tcg_insn_unit *addr)
1049 {
1050 intptr_t addri = (intptr_t)addr;
1051 ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1052
1053 if ((addri & 1) == 0 && disp - 8 < 0x01fffffd && disp - 8 > -0x01fffffd) {
1054 tcg_out_b(s, cond, disp);
1055 return;
1056 }
1057 tcg_out_movi_pool(s, cond, TCG_REG_PC, addri);
1058 }
1059
1060 /* The call case is mostly used for helpers - so it's not unreasonable
1061 * for them to be beyond branch range */
1062 static void tcg_out_call(TCGContext *s, tcg_insn_unit *addr)
1063 {
1064 intptr_t addri = (intptr_t)addr;
1065 ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1066
1067 if (disp - 8 < 0x02000000 && disp - 8 >= -0x02000000) {
1068 if (addri & 1) {
1069 /* Use BLX if the target is in Thumb mode */
1070 if (!use_armv5t_instructions) {
1071 tcg_abort();
1072 }
1073 tcg_out_blx_imm(s, disp);
1074 } else {
1075 tcg_out_bl(s, COND_AL, disp);
1076 }
1077 } else if (use_armv7_instructions) {
1078 tcg_out_movi32(s, COND_AL, TCG_REG_TMP, addri);
1079 tcg_out_blx(s, COND_AL, TCG_REG_TMP);
1080 } else {
1081 /* ??? Know that movi_pool emits exactly 2 insns. */
1082 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R14, TCG_REG_PC, 4);
1083 tcg_out_movi_pool(s, COND_AL, TCG_REG_PC, addri);
1084 }
1085 }
1086
1087 static inline void tcg_out_goto_label(TCGContext *s, int cond, TCGLabel *l)
1088 {
1089 if (l->has_value) {
1090 tcg_out_goto(s, cond, l->u.value_ptr);
1091 } else {
1092 tcg_out_reloc(s, s->code_ptr, R_ARM_PC24, l, 0);
1093 tcg_out_b_noaddr(s, cond);
1094 }
1095 }
1096
1097 static inline void tcg_out_mb(TCGContext *s, TCGArg a0)
1098 {
1099 if (use_armv7_instructions) {
1100 tcg_out32(s, INSN_DMB_ISH);
1101 } else if (use_armv6_instructions) {
1102 tcg_out32(s, INSN_DMB_MCR);
1103 }
1104 }
1105
1106 #ifdef CONFIG_SOFTMMU
1107 #include "tcg-ldst.inc.c"
1108
1109 /* helper signature: helper_ret_ld_mmu(CPUState *env, target_ulong addr,
1110 * int mmu_idx, uintptr_t ra)
1111 */
1112 static void * const qemu_ld_helpers[16] = {
1113 [MO_UB] = helper_ret_ldub_mmu,
1114 [MO_SB] = helper_ret_ldsb_mmu,
1115
1116 [MO_LEUW] = helper_le_lduw_mmu,
1117 [MO_LEUL] = helper_le_ldul_mmu,
1118 [MO_LEQ] = helper_le_ldq_mmu,
1119 [MO_LESW] = helper_le_ldsw_mmu,
1120 [MO_LESL] = helper_le_ldul_mmu,
1121
1122 [MO_BEUW] = helper_be_lduw_mmu,
1123 [MO_BEUL] = helper_be_ldul_mmu,
1124 [MO_BEQ] = helper_be_ldq_mmu,
1125 [MO_BESW] = helper_be_ldsw_mmu,
1126 [MO_BESL] = helper_be_ldul_mmu,
1127 };
1128
1129 /* helper signature: helper_ret_st_mmu(CPUState *env, target_ulong addr,
1130 * uintxx_t val, int mmu_idx, uintptr_t ra)
1131 */
1132 static void * const qemu_st_helpers[16] = {
1133 [MO_UB] = helper_ret_stb_mmu,
1134 [MO_LEUW] = helper_le_stw_mmu,
1135 [MO_LEUL] = helper_le_stl_mmu,
1136 [MO_LEQ] = helper_le_stq_mmu,
1137 [MO_BEUW] = helper_be_stw_mmu,
1138 [MO_BEUL] = helper_be_stl_mmu,
1139 [MO_BEQ] = helper_be_stq_mmu,
1140 };
1141
1142 /* Helper routines for marshalling helper function arguments into
1143 * the correct registers and stack.
1144 * argreg is where we want to put this argument, arg is the argument itself.
1145 * Return value is the updated argreg ready for the next call.
1146 * Note that argreg 0..3 is real registers, 4+ on stack.
1147 *
1148 * We provide routines for arguments which are: immediate, 32 bit
1149 * value in register, 16 and 8 bit values in register (which must be zero
1150 * extended before use) and 64 bit value in a lo:hi register pair.
1151 */
1152 #define DEFINE_TCG_OUT_ARG(NAME, ARGTYPE, MOV_ARG, EXT_ARG) \
1153 static TCGReg NAME(TCGContext *s, TCGReg argreg, ARGTYPE arg) \
1154 { \
1155 if (argreg < 4) { \
1156 MOV_ARG(s, COND_AL, argreg, arg); \
1157 } else { \
1158 int ofs = (argreg - 4) * 4; \
1159 EXT_ARG; \
1160 tcg_debug_assert(ofs + 4 <= TCG_STATIC_CALL_ARGS_SIZE); \
1161 tcg_out_st32_12(s, COND_AL, arg, TCG_REG_CALL_STACK, ofs); \
1162 } \
1163 return argreg + 1; \
1164 }
1165
1166 DEFINE_TCG_OUT_ARG(tcg_out_arg_imm32, uint32_t, tcg_out_movi32,
1167 (tcg_out_movi32(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1168 DEFINE_TCG_OUT_ARG(tcg_out_arg_reg8, TCGReg, tcg_out_ext8u,
1169 (tcg_out_ext8u(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1170 DEFINE_TCG_OUT_ARG(tcg_out_arg_reg16, TCGReg, tcg_out_ext16u,
1171 (tcg_out_ext16u(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1172 DEFINE_TCG_OUT_ARG(tcg_out_arg_reg32, TCGReg, tcg_out_mov_reg, )
1173
1174 static TCGReg tcg_out_arg_reg64(TCGContext *s, TCGReg argreg,
1175 TCGReg arglo, TCGReg arghi)
1176 {
1177 /* 64 bit arguments must go in even/odd register pairs
1178 * and in 8-aligned stack slots.
1179 */
1180 if (argreg & 1) {
1181 argreg++;
1182 }
1183 if (use_armv6_instructions && argreg >= 4
1184 && (arglo & 1) == 0 && arghi == arglo + 1) {
1185 tcg_out_strd_8(s, COND_AL, arglo,
1186 TCG_REG_CALL_STACK, (argreg - 4) * 4);
1187 return argreg + 2;
1188 } else {
1189 argreg = tcg_out_arg_reg32(s, argreg, arglo);
1190 argreg = tcg_out_arg_reg32(s, argreg, arghi);
1191 return argreg;
1192 }
1193 }
1194
1195 #define TLB_SHIFT (CPU_TLB_ENTRY_BITS + CPU_TLB_BITS)
1196
1197 /* We're expecting to use an 8-bit immediate and to mask. */
1198 QEMU_BUILD_BUG_ON(CPU_TLB_BITS > 8);
1199
1200 /* We're expecting to use an 8-bit immediate add + 8-bit ldrd offset.
1201 Using the offset of the second entry in the last tlb table ensures
1202 that we can index all of the elements of the first entry. */
1203 QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table[NB_MMU_MODES - 1][1])
1204 > 0xffff);
1205
1206 /* Load and compare a TLB entry, leaving the flags set. Returns the register
1207 containing the addend of the tlb entry. Clobbers R0, R1, R2, TMP. */
1208
1209 static TCGReg tcg_out_tlb_read(TCGContext *s, TCGReg addrlo, TCGReg addrhi,
1210 TCGMemOp opc, int mem_index, bool is_load)
1211 {
1212 TCGReg base = TCG_AREG0;
1213 int cmp_off =
1214 (is_load
1215 ? offsetof(CPUArchState, tlb_table[mem_index][0].addr_read)
1216 : offsetof(CPUArchState, tlb_table[mem_index][0].addr_write));
1217 int add_off = offsetof(CPUArchState, tlb_table[mem_index][0].addend);
1218 unsigned s_bits = opc & MO_SIZE;
1219 unsigned a_bits = get_alignment_bits(opc);
1220
1221 /* V7 generates the following:
1222 * ubfx r0, addrlo, #TARGET_PAGE_BITS, #CPU_TLB_BITS
1223 * add r2, env, #high
1224 * add r2, r2, r0, lsl #CPU_TLB_ENTRY_BITS
1225 * ldr r0, [r2, #cmp]
1226 * ldr r2, [r2, #add]
1227 * movw tmp, #page_align_mask
1228 * bic tmp, addrlo, tmp
1229 * cmp r0, tmp
1230 *
1231 * Otherwise we generate:
1232 * shr tmp, addrlo, #TARGET_PAGE_BITS
1233 * add r2, env, #high
1234 * and r0, tmp, #(CPU_TLB_SIZE - 1)
1235 * add r2, r2, r0, lsl #CPU_TLB_ENTRY_BITS
1236 * ldr r0, [r2, #cmp]
1237 * ldr r2, [r2, #add]
1238 * tst addrlo, #s_mask
1239 * cmpeq r0, tmp, lsl #TARGET_PAGE_BITS
1240 */
1241 if (use_armv7_instructions) {
1242 tcg_out_extract(s, COND_AL, TCG_REG_R0, addrlo,
1243 TARGET_PAGE_BITS, CPU_TLB_BITS);
1244 } else {
1245 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_TMP,
1246 0, addrlo, SHIFT_IMM_LSR(TARGET_PAGE_BITS));
1247 }
1248
1249 /* We checked that the offset is contained within 16 bits above. */
1250 if (add_off > 0xfff
1251 || (use_armv6_instructions && TARGET_LONG_BITS == 64
1252 && cmp_off > 0xff)) {
1253 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R2, base,
1254 (24 << 7) | (cmp_off >> 8));
1255 base = TCG_REG_R2;
1256 add_off -= cmp_off & 0xff00;
1257 cmp_off &= 0xff;
1258 }
1259 if (!use_armv7_instructions) {
1260 tcg_out_dat_imm(s, COND_AL, ARITH_AND,
1261 TCG_REG_R0, TCG_REG_TMP, CPU_TLB_SIZE - 1);
1262 }
1263 tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_R2, base,
1264 TCG_REG_R0, SHIFT_IMM_LSL(CPU_TLB_ENTRY_BITS));
1265
1266 /* Load the tlb comparator. Use ldrd if needed and available,
1267 but due to how the pointer needs setting up, ldm isn't useful.
1268 Base arm5 doesn't have ldrd, but armv5te does. */
1269 if (use_armv6_instructions && TARGET_LONG_BITS == 64) {
1270 tcg_out_ldrd_8(s, COND_AL, TCG_REG_R0, TCG_REG_R2, cmp_off);
1271 } else {
1272 tcg_out_ld32_12(s, COND_AL, TCG_REG_R0, TCG_REG_R2, cmp_off);
1273 if (TARGET_LONG_BITS == 64) {
1274 tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R2, cmp_off + 4);
1275 }
1276 }
1277
1278 /* Load the tlb addend. */
1279 tcg_out_ld32_12(s, COND_AL, TCG_REG_R2, TCG_REG_R2, add_off);
1280
1281 /* Check alignment. We don't support inline unaligned acceses,
1282 but we can easily support overalignment checks. */
1283 if (a_bits < s_bits) {
1284 a_bits = s_bits;
1285 }
1286
1287 if (use_armv7_instructions) {
1288 tcg_target_ulong mask = ~(TARGET_PAGE_MASK | ((1 << a_bits) - 1));
1289 int rot = encode_imm(mask);
1290
1291 if (rot >= 0) {
1292 tcg_out_dat_imm(s, COND_AL, ARITH_BIC, TCG_REG_TMP, addrlo,
1293 rotl(mask, rot) | (rot << 7));
1294 } else {
1295 tcg_out_movi32(s, COND_AL, TCG_REG_TMP, mask);
1296 tcg_out_dat_reg(s, COND_AL, ARITH_BIC, TCG_REG_TMP,
1297 addrlo, TCG_REG_TMP, 0);
1298 }
1299 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, TCG_REG_R0, TCG_REG_TMP, 0);
1300 } else {
1301 if (a_bits) {
1302 tcg_out_dat_imm(s, COND_AL, ARITH_TST, 0, addrlo,
1303 (1 << a_bits) - 1);
1304 }
1305 tcg_out_dat_reg(s, (a_bits ? COND_EQ : COND_AL), ARITH_CMP,
1306 0, TCG_REG_R0, TCG_REG_TMP,
1307 SHIFT_IMM_LSL(TARGET_PAGE_BITS));
1308 }
1309
1310 if (TARGET_LONG_BITS == 64) {
1311 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0, TCG_REG_R1, addrhi, 0);
1312 }
1313
1314 return TCG_REG_R2;
1315 }
1316
1317 /* Record the context of a call to the out of line helper code for the slow
1318 path for a load or store, so that we can later generate the correct
1319 helper code. */
1320 static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
1321 TCGReg datalo, TCGReg datahi, TCGReg addrlo,
1322 TCGReg addrhi, tcg_insn_unit *raddr,
1323 tcg_insn_unit *label_ptr)
1324 {
1325 TCGLabelQemuLdst *label = new_ldst_label(s);
1326
1327 label->is_ld = is_ld;
1328 label->oi = oi;
1329 label->datalo_reg = datalo;
1330 label->datahi_reg = datahi;
1331 label->addrlo_reg = addrlo;
1332 label->addrhi_reg = addrhi;
1333 label->raddr = raddr;
1334 label->label_ptr[0] = label_ptr;
1335 }
1336
1337 static void tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1338 {
1339 TCGReg argreg, datalo, datahi;
1340 TCGMemOpIdx oi = lb->oi;
1341 TCGMemOp opc = get_memop(oi);
1342 void *func;
1343
1344 reloc_pc24(lb->label_ptr[0], s->code_ptr);
1345
1346 argreg = tcg_out_arg_reg32(s, TCG_REG_R0, TCG_AREG0);
1347 if (TARGET_LONG_BITS == 64) {
1348 argreg = tcg_out_arg_reg64(s, argreg, lb->addrlo_reg, lb->addrhi_reg);
1349 } else {
1350 argreg = tcg_out_arg_reg32(s, argreg, lb->addrlo_reg);
1351 }
1352 argreg = tcg_out_arg_imm32(s, argreg, oi);
1353 argreg = tcg_out_arg_reg32(s, argreg, TCG_REG_R14);
1354
1355 /* For armv6 we can use the canonical unsigned helpers and minimize
1356 icache usage. For pre-armv6, use the signed helpers since we do
1357 not have a single insn sign-extend. */
1358 if (use_armv6_instructions) {
1359 func = qemu_ld_helpers[opc & (MO_BSWAP | MO_SIZE)];
1360 } else {
1361 func = qemu_ld_helpers[opc & (MO_BSWAP | MO_SSIZE)];
1362 if (opc & MO_SIGN) {
1363 opc = MO_UL;
1364 }
1365 }
1366 tcg_out_call(s, func);
1367
1368 datalo = lb->datalo_reg;
1369 datahi = lb->datahi_reg;
1370 switch (opc & MO_SSIZE) {
1371 case MO_SB:
1372 tcg_out_ext8s(s, COND_AL, datalo, TCG_REG_R0);
1373 break;
1374 case MO_SW:
1375 tcg_out_ext16s(s, COND_AL, datalo, TCG_REG_R0);
1376 break;
1377 default:
1378 tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1379 break;
1380 case MO_Q:
1381 if (datalo != TCG_REG_R1) {
1382 tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1383 tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1384 } else if (datahi != TCG_REG_R0) {
1385 tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1386 tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1387 } else {
1388 tcg_out_mov_reg(s, COND_AL, TCG_REG_TMP, TCG_REG_R0);
1389 tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1390 tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_TMP);
1391 }
1392 break;
1393 }
1394
1395 tcg_out_goto(s, COND_AL, lb->raddr);
1396 }
1397
1398 static void tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1399 {
1400 TCGReg argreg, datalo, datahi;
1401 TCGMemOpIdx oi = lb->oi;
1402 TCGMemOp opc = get_memop(oi);
1403
1404 reloc_pc24(lb->label_ptr[0], s->code_ptr);
1405
1406 argreg = TCG_REG_R0;
1407 argreg = tcg_out_arg_reg32(s, argreg, TCG_AREG0);
1408 if (TARGET_LONG_BITS == 64) {
1409 argreg = tcg_out_arg_reg64(s, argreg, lb->addrlo_reg, lb->addrhi_reg);
1410 } else {
1411 argreg = tcg_out_arg_reg32(s, argreg, lb->addrlo_reg);
1412 }
1413
1414 datalo = lb->datalo_reg;
1415 datahi = lb->datahi_reg;
1416 switch (opc & MO_SIZE) {
1417 case MO_8:
1418 argreg = tcg_out_arg_reg8(s, argreg, datalo);
1419 break;
1420 case MO_16:
1421 argreg = tcg_out_arg_reg16(s, argreg, datalo);
1422 break;
1423 case MO_32:
1424 default:
1425 argreg = tcg_out_arg_reg32(s, argreg, datalo);
1426 break;
1427 case MO_64:
1428 argreg = tcg_out_arg_reg64(s, argreg, datalo, datahi);
1429 break;
1430 }
1431
1432 argreg = tcg_out_arg_imm32(s, argreg, oi);
1433 argreg = tcg_out_arg_reg32(s, argreg, TCG_REG_R14);
1434
1435 /* Tail-call to the helper, which will return to the fast path. */
1436 tcg_out_goto(s, COND_AL, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)]);
1437 }
1438 #endif /* SOFTMMU */
1439
1440 static inline void tcg_out_qemu_ld_index(TCGContext *s, TCGMemOp opc,
1441 TCGReg datalo, TCGReg datahi,
1442 TCGReg addrlo, TCGReg addend)
1443 {
1444 TCGMemOp bswap = opc & MO_BSWAP;
1445
1446 switch (opc & MO_SSIZE) {
1447 case MO_UB:
1448 tcg_out_ld8_r(s, COND_AL, datalo, addrlo, addend);
1449 break;
1450 case MO_SB:
1451 tcg_out_ld8s_r(s, COND_AL, datalo, addrlo, addend);
1452 break;
1453 case MO_UW:
1454 tcg_out_ld16u_r(s, COND_AL, datalo, addrlo, addend);
1455 if (bswap) {
1456 tcg_out_bswap16(s, COND_AL, datalo, datalo);
1457 }
1458 break;
1459 case MO_SW:
1460 if (bswap) {
1461 tcg_out_ld16u_r(s, COND_AL, datalo, addrlo, addend);
1462 tcg_out_bswap16s(s, COND_AL, datalo, datalo);
1463 } else {
1464 tcg_out_ld16s_r(s, COND_AL, datalo, addrlo, addend);
1465 }
1466 break;
1467 case MO_UL:
1468 default:
1469 tcg_out_ld32_r(s, COND_AL, datalo, addrlo, addend);
1470 if (bswap) {
1471 tcg_out_bswap32(s, COND_AL, datalo, datalo);
1472 }
1473 break;
1474 case MO_Q:
1475 {
1476 TCGReg dl = (bswap ? datahi : datalo);
1477 TCGReg dh = (bswap ? datalo : datahi);
1478
1479 /* Avoid ldrd for user-only emulation, to handle unaligned. */
1480 if (USING_SOFTMMU && use_armv6_instructions
1481 && (dl & 1) == 0 && dh == dl + 1) {
1482 tcg_out_ldrd_r(s, COND_AL, dl, addrlo, addend);
1483 } else if (dl != addend) {
1484 tcg_out_ld32_rwb(s, COND_AL, dl, addend, addrlo);
1485 tcg_out_ld32_12(s, COND_AL, dh, addend, 4);
1486 } else {
1487 tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_TMP,
1488 addend, addrlo, SHIFT_IMM_LSL(0));
1489 tcg_out_ld32_12(s, COND_AL, dl, TCG_REG_TMP, 0);
1490 tcg_out_ld32_12(s, COND_AL, dh, TCG_REG_TMP, 4);
1491 }
1492 if (bswap) {
1493 tcg_out_bswap32(s, COND_AL, dl, dl);
1494 tcg_out_bswap32(s, COND_AL, dh, dh);
1495 }
1496 }
1497 break;
1498 }
1499 }
1500
1501 static inline void tcg_out_qemu_ld_direct(TCGContext *s, TCGMemOp opc,
1502 TCGReg datalo, TCGReg datahi,
1503 TCGReg addrlo)
1504 {
1505 TCGMemOp bswap = opc & MO_BSWAP;
1506
1507 switch (opc & MO_SSIZE) {
1508 case MO_UB:
1509 tcg_out_ld8_12(s, COND_AL, datalo, addrlo, 0);
1510 break;
1511 case MO_SB:
1512 tcg_out_ld8s_8(s, COND_AL, datalo, addrlo, 0);
1513 break;
1514 case MO_UW:
1515 tcg_out_ld16u_8(s, COND_AL, datalo, addrlo, 0);
1516 if (bswap) {
1517 tcg_out_bswap16(s, COND_AL, datalo, datalo);
1518 }
1519 break;
1520 case MO_SW:
1521 if (bswap) {
1522 tcg_out_ld16u_8(s, COND_AL, datalo, addrlo, 0);
1523 tcg_out_bswap16s(s, COND_AL, datalo, datalo);
1524 } else {
1525 tcg_out_ld16s_8(s, COND_AL, datalo, addrlo, 0);
1526 }
1527 break;
1528 case MO_UL:
1529 default:
1530 tcg_out_ld32_12(s, COND_AL, datalo, addrlo, 0);
1531 if (bswap) {
1532 tcg_out_bswap32(s, COND_AL, datalo, datalo);
1533 }
1534 break;
1535 case MO_Q:
1536 {
1537 TCGReg dl = (bswap ? datahi : datalo);
1538 TCGReg dh = (bswap ? datalo : datahi);
1539
1540 /* Avoid ldrd for user-only emulation, to handle unaligned. */
1541 if (USING_SOFTMMU && use_armv6_instructions
1542 && (dl & 1) == 0 && dh == dl + 1) {
1543 tcg_out_ldrd_8(s, COND_AL, dl, addrlo, 0);
1544 } else if (dl == addrlo) {
1545 tcg_out_ld32_12(s, COND_AL, dh, addrlo, bswap ? 0 : 4);
1546 tcg_out_ld32_12(s, COND_AL, dl, addrlo, bswap ? 4 : 0);
1547 } else {
1548 tcg_out_ld32_12(s, COND_AL, dl, addrlo, bswap ? 4 : 0);
1549 tcg_out_ld32_12(s, COND_AL, dh, addrlo, bswap ? 0 : 4);
1550 }
1551 if (bswap) {
1552 tcg_out_bswap32(s, COND_AL, dl, dl);
1553 tcg_out_bswap32(s, COND_AL, dh, dh);
1554 }
1555 }
1556 break;
1557 }
1558 }
1559
1560 static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is64)
1561 {
1562 TCGReg addrlo, datalo, datahi, addrhi __attribute__((unused));
1563 TCGMemOpIdx oi;
1564 TCGMemOp opc;
1565 #ifdef CONFIG_SOFTMMU
1566 int mem_index;
1567 TCGReg addend;
1568 tcg_insn_unit *label_ptr;
1569 #endif
1570
1571 datalo = *args++;
1572 datahi = (is64 ? *args++ : 0);
1573 addrlo = *args++;
1574 addrhi = (TARGET_LONG_BITS == 64 ? *args++ : 0);
1575 oi = *args++;
1576 opc = get_memop(oi);
1577
1578 #ifdef CONFIG_SOFTMMU
1579 mem_index = get_mmuidx(oi);
1580 addend = tcg_out_tlb_read(s, addrlo, addrhi, opc, mem_index, 1);
1581
1582 /* This a conditional BL only to load a pointer within this opcode into LR
1583 for the slow path. We will not be using the value for a tail call. */
1584 label_ptr = s->code_ptr;
1585 tcg_out_bl_noaddr(s, COND_NE);
1586
1587 tcg_out_qemu_ld_index(s, opc, datalo, datahi, addrlo, addend);
1588
1589 add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
1590 s->code_ptr, label_ptr);
1591 #else /* !CONFIG_SOFTMMU */
1592 if (guest_base) {
1593 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, guest_base);
1594 tcg_out_qemu_ld_index(s, opc, datalo, datahi, addrlo, TCG_REG_TMP);
1595 } else {
1596 tcg_out_qemu_ld_direct(s, opc, datalo, datahi, addrlo);
1597 }
1598 #endif
1599 }
1600
1601 static inline void tcg_out_qemu_st_index(TCGContext *s, int cond, TCGMemOp opc,
1602 TCGReg datalo, TCGReg datahi,
1603 TCGReg addrlo, TCGReg addend)
1604 {
1605 TCGMemOp bswap = opc & MO_BSWAP;
1606
1607 switch (opc & MO_SIZE) {
1608 case MO_8:
1609 tcg_out_st8_r(s, cond, datalo, addrlo, addend);
1610 break;
1611 case MO_16:
1612 if (bswap) {
1613 tcg_out_bswap16st(s, cond, TCG_REG_R0, datalo);
1614 tcg_out_st16_r(s, cond, TCG_REG_R0, addrlo, addend);
1615 } else {
1616 tcg_out_st16_r(s, cond, datalo, addrlo, addend);
1617 }
1618 break;
1619 case MO_32:
1620 default:
1621 if (bswap) {
1622 tcg_out_bswap32(s, cond, TCG_REG_R0, datalo);
1623 tcg_out_st32_r(s, cond, TCG_REG_R0, addrlo, addend);
1624 } else {
1625 tcg_out_st32_r(s, cond, datalo, addrlo, addend);
1626 }
1627 break;
1628 case MO_64:
1629 /* Avoid strd for user-only emulation, to handle unaligned. */
1630 if (bswap) {
1631 tcg_out_bswap32(s, cond, TCG_REG_R0, datahi);
1632 tcg_out_st32_rwb(s, cond, TCG_REG_R0, addend, addrlo);
1633 tcg_out_bswap32(s, cond, TCG_REG_R0, datalo);
1634 tcg_out_st32_12(s, cond, TCG_REG_R0, addend, 4);
1635 } else if (USING_SOFTMMU && use_armv6_instructions
1636 && (datalo & 1) == 0 && datahi == datalo + 1) {
1637 tcg_out_strd_r(s, cond, datalo, addrlo, addend);
1638 } else {
1639 tcg_out_st32_rwb(s, cond, datalo, addend, addrlo);
1640 tcg_out_st32_12(s, cond, datahi, addend, 4);
1641 }
1642 break;
1643 }
1644 }
1645
1646 static inline void tcg_out_qemu_st_direct(TCGContext *s, TCGMemOp opc,
1647 TCGReg datalo, TCGReg datahi,
1648 TCGReg addrlo)
1649 {
1650 TCGMemOp bswap = opc & MO_BSWAP;
1651
1652 switch (opc & MO_SIZE) {
1653 case MO_8:
1654 tcg_out_st8_12(s, COND_AL, datalo, addrlo, 0);
1655 break;
1656 case MO_16:
1657 if (bswap) {
1658 tcg_out_bswap16st(s, COND_AL, TCG_REG_R0, datalo);
1659 tcg_out_st16_8(s, COND_AL, TCG_REG_R0, addrlo, 0);
1660 } else {
1661 tcg_out_st16_8(s, COND_AL, datalo, addrlo, 0);
1662 }
1663 break;
1664 case MO_32:
1665 default:
1666 if (bswap) {
1667 tcg_out_bswap32(s, COND_AL, TCG_REG_R0, datalo);
1668 tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addrlo, 0);
1669 } else {
1670 tcg_out_st32_12(s, COND_AL, datalo, addrlo, 0);
1671 }
1672 break;
1673 case MO_64:
1674 /* Avoid strd for user-only emulation, to handle unaligned. */
1675 if (bswap) {
1676 tcg_out_bswap32(s, COND_AL, TCG_REG_R0, datahi);
1677 tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addrlo, 0);
1678 tcg_out_bswap32(s, COND_AL, TCG_REG_R0, datalo);
1679 tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addrlo, 4);
1680 } else if (USING_SOFTMMU && use_armv6_instructions
1681 && (datalo & 1) == 0 && datahi == datalo + 1) {
1682 tcg_out_strd_8(s, COND_AL, datalo, addrlo, 0);
1683 } else {
1684 tcg_out_st32_12(s, COND_AL, datalo, addrlo, 0);
1685 tcg_out_st32_12(s, COND_AL, datahi, addrlo, 4);
1686 }
1687 break;
1688 }
1689 }
1690
1691 static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is64)
1692 {
1693 TCGReg addrlo, datalo, datahi, addrhi __attribute__((unused));
1694 TCGMemOpIdx oi;
1695 TCGMemOp opc;
1696 #ifdef CONFIG_SOFTMMU
1697 int mem_index;
1698 TCGReg addend;
1699 tcg_insn_unit *label_ptr;
1700 #endif
1701
1702 datalo = *args++;
1703 datahi = (is64 ? *args++ : 0);
1704 addrlo = *args++;
1705 addrhi = (TARGET_LONG_BITS == 64 ? *args++ : 0);
1706 oi = *args++;
1707 opc = get_memop(oi);
1708
1709 #ifdef CONFIG_SOFTMMU
1710 mem_index = get_mmuidx(oi);
1711 addend = tcg_out_tlb_read(s, addrlo, addrhi, opc, mem_index, 0);
1712
1713 tcg_out_qemu_st_index(s, COND_EQ, opc, datalo, datahi, addrlo, addend);
1714
1715 /* The conditional call must come last, as we're going to return here. */
1716 label_ptr = s->code_ptr;
1717 tcg_out_bl_noaddr(s, COND_NE);
1718
1719 add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
1720 s->code_ptr, label_ptr);
1721 #else /* !CONFIG_SOFTMMU */
1722 if (guest_base) {
1723 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, guest_base);
1724 tcg_out_qemu_st_index(s, COND_AL, opc, datalo,
1725 datahi, addrlo, TCG_REG_TMP);
1726 } else {
1727 tcg_out_qemu_st_direct(s, opc, datalo, datahi, addrlo);
1728 }
1729 #endif
1730 }
1731
1732 static tcg_insn_unit *tb_ret_addr;
1733
1734 static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
1735 const TCGArg *args, const int *const_args)
1736 {
1737 TCGArg a0, a1, a2, a3, a4, a5;
1738 int c;
1739
1740 switch (opc) {
1741 case INDEX_op_exit_tb:
1742 /* Reuse the zeroing that exists for goto_ptr. */
1743 a0 = args[0];
1744 if (a0 == 0) {
1745 tcg_out_goto(s, COND_AL, s->code_gen_epilogue);
1746 } else {
1747 tcg_out_movi32(s, COND_AL, TCG_REG_R0, args[0]);
1748 tcg_out_goto(s, COND_AL, tb_ret_addr);
1749 }
1750 break;
1751 case INDEX_op_goto_tb:
1752 {
1753 /* Indirect jump method */
1754 intptr_t ptr, dif, dil;
1755 TCGReg base = TCG_REG_PC;
1756
1757 tcg_debug_assert(s->tb_jmp_insn_offset == 0);
1758 ptr = (intptr_t)(s->tb_jmp_target_addr + args[0]);
1759 dif = ptr - ((intptr_t)s->code_ptr + 8);
1760 dil = sextract32(dif, 0, 12);
1761 if (dif != dil) {
1762 /* The TB is close, but outside the 12 bits addressable by
1763 the load. We can extend this to 20 bits with a sub of a
1764 shifted immediate from pc. In the vastly unlikely event
1765 the code requires more than 1MB, we'll use 2 insns and
1766 be no worse off. */
1767 base = TCG_REG_R0;
1768 tcg_out_movi32(s, COND_AL, base, ptr - dil);
1769 }
1770 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, base, dil);
1771 s->tb_jmp_reset_offset[args[0]] = tcg_current_code_size(s);
1772 }
1773 break;
1774 case INDEX_op_goto_ptr:
1775 tcg_out_bx(s, COND_AL, args[0]);
1776 break;
1777 case INDEX_op_br:
1778 tcg_out_goto_label(s, COND_AL, arg_label(args[0]));
1779 break;
1780
1781 case INDEX_op_ld8u_i32:
1782 tcg_out_ld8u(s, COND_AL, args[0], args[1], args[2]);
1783 break;
1784 case INDEX_op_ld8s_i32:
1785 tcg_out_ld8s(s, COND_AL, args[0], args[1], args[2]);
1786 break;
1787 case INDEX_op_ld16u_i32:
1788 tcg_out_ld16u(s, COND_AL, args[0], args[1], args[2]);
1789 break;
1790 case INDEX_op_ld16s_i32:
1791 tcg_out_ld16s(s, COND_AL, args[0], args[1], args[2]);
1792 break;
1793 case INDEX_op_ld_i32:
1794 tcg_out_ld32u(s, COND_AL, args[0], args[1], args[2]);
1795 break;
1796 case INDEX_op_st8_i32:
1797 tcg_out_st8(s, COND_AL, args[0], args[1], args[2]);
1798 break;
1799 case INDEX_op_st16_i32:
1800 tcg_out_st16(s, COND_AL, args[0], args[1], args[2]);
1801 break;
1802 case INDEX_op_st_i32:
1803 tcg_out_st32(s, COND_AL, args[0], args[1], args[2]);
1804 break;
1805
1806 case INDEX_op_movcond_i32:
1807 /* Constraints mean that v2 is always in the same register as dest,
1808 * so we only need to do "if condition passed, move v1 to dest".
1809 */
1810 tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
1811 args[1], args[2], const_args[2]);
1812 tcg_out_dat_rIK(s, tcg_cond_to_arm_cond[args[5]], ARITH_MOV,
1813 ARITH_MVN, args[0], 0, args[3], const_args[3]);
1814 break;
1815 case INDEX_op_add_i32:
1816 tcg_out_dat_rIN(s, COND_AL, ARITH_ADD, ARITH_SUB,
1817 args[0], args[1], args[2], const_args[2]);
1818 break;
1819 case INDEX_op_sub_i32:
1820 if (const_args[1]) {
1821 if (const_args[2]) {
1822 tcg_out_movi32(s, COND_AL, args[0], args[1] - args[2]);
1823 } else {
1824 tcg_out_dat_rI(s, COND_AL, ARITH_RSB,
1825 args[0], args[2], args[1], 1);
1826 }
1827 } else {
1828 tcg_out_dat_rIN(s, COND_AL, ARITH_SUB, ARITH_ADD,
1829 args[0], args[1], args[2], const_args[2]);
1830 }
1831 break;
1832 case INDEX_op_and_i32:
1833 tcg_out_dat_rIK(s, COND_AL, ARITH_AND, ARITH_BIC,
1834 args[0], args[1], args[2], const_args[2]);
1835 break;
1836 case INDEX_op_andc_i32:
1837 tcg_out_dat_rIK(s, COND_AL, ARITH_BIC, ARITH_AND,
1838 args[0], args[1], args[2], const_args[2]);
1839 break;
1840 case INDEX_op_or_i32:
1841 c = ARITH_ORR;
1842 goto gen_arith;
1843 case INDEX_op_xor_i32:
1844 c = ARITH_EOR;
1845 /* Fall through. */
1846 gen_arith:
1847 tcg_out_dat_rI(s, COND_AL, c, args[0], args[1], args[2], const_args[2]);
1848 break;
1849 case INDEX_op_add2_i32:
1850 a0 = args[0], a1 = args[1], a2 = args[2];
1851 a3 = args[3], a4 = args[4], a5 = args[5];
1852 if (a0 == a3 || (a0 == a5 && !const_args[5])) {
1853 a0 = TCG_REG_TMP;
1854 }
1855 tcg_out_dat_rIN(s, COND_AL, ARITH_ADD | TO_CPSR, ARITH_SUB | TO_CPSR,
1856 a0, a2, a4, const_args[4]);
1857 tcg_out_dat_rIK(s, COND_AL, ARITH_ADC, ARITH_SBC,
1858 a1, a3, a5, const_args[5]);
1859 tcg_out_mov_reg(s, COND_AL, args[0], a0);
1860 break;
1861 case INDEX_op_sub2_i32:
1862 a0 = args[0], a1 = args[1], a2 = args[2];
1863 a3 = args[3], a4 = args[4], a5 = args[5];
1864 if ((a0 == a3 && !const_args[3]) || (a0 == a5 && !const_args[5])) {
1865 a0 = TCG_REG_TMP;
1866 }
1867 if (const_args[2]) {
1868 if (const_args[4]) {
1869 tcg_out_movi32(s, COND_AL, a0, a4);
1870 a4 = a0;
1871 }
1872 tcg_out_dat_rI(s, COND_AL, ARITH_RSB | TO_CPSR, a0, a4, a2, 1);
1873 } else {
1874 tcg_out_dat_rIN(s, COND_AL, ARITH_SUB | TO_CPSR,
1875 ARITH_ADD | TO_CPSR, a0, a2, a4, const_args[4]);
1876 }
1877 if (const_args[3]) {
1878 if (const_args[5]) {
1879 tcg_out_movi32(s, COND_AL, a1, a5);
1880 a5 = a1;
1881 }
1882 tcg_out_dat_rI(s, COND_AL, ARITH_RSC, a1, a5, a3, 1);
1883 } else {
1884 tcg_out_dat_rIK(s, COND_AL, ARITH_SBC, ARITH_ADC,
1885 a1, a3, a5, const_args[5]);
1886 }
1887 tcg_out_mov_reg(s, COND_AL, args[0], a0);
1888 break;
1889 case INDEX_op_neg_i32:
1890 tcg_out_dat_imm(s, COND_AL, ARITH_RSB, args[0], args[1], 0);
1891 break;
1892 case INDEX_op_not_i32:
1893 tcg_out_dat_reg(s, COND_AL,
1894 ARITH_MVN, args[0], 0, args[1], SHIFT_IMM_LSL(0));
1895 break;
1896 case INDEX_op_mul_i32:
1897 tcg_out_mul32(s, COND_AL, args[0], args[1], args[2]);
1898 break;
1899 case INDEX_op_mulu2_i32:
1900 tcg_out_umull32(s, COND_AL, args[0], args[1], args[2], args[3]);
1901 break;
1902 case INDEX_op_muls2_i32:
1903 tcg_out_smull32(s, COND_AL, args[0], args[1], args[2], args[3]);
1904 break;
1905 /* XXX: Perhaps args[2] & 0x1f is wrong */
1906 case INDEX_op_shl_i32:
1907 c = const_args[2] ?
1908 SHIFT_IMM_LSL(args[2] & 0x1f) : SHIFT_REG_LSL(args[2]);
1909 goto gen_shift32;
1910 case INDEX_op_shr_i32:
1911 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_LSR(args[2] & 0x1f) :
1912 SHIFT_IMM_LSL(0) : SHIFT_REG_LSR(args[2]);
1913 goto gen_shift32;
1914 case INDEX_op_sar_i32:
1915 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ASR(args[2] & 0x1f) :
1916 SHIFT_IMM_LSL(0) : SHIFT_REG_ASR(args[2]);
1917 goto gen_shift32;
1918 case INDEX_op_rotr_i32:
1919 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ROR(args[2] & 0x1f) :
1920 SHIFT_IMM_LSL(0) : SHIFT_REG_ROR(args[2]);
1921 /* Fall through. */
1922 gen_shift32:
1923 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1], c);
1924 break;
1925
1926 case INDEX_op_rotl_i32:
1927 if (const_args[2]) {
1928 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
1929 ((0x20 - args[2]) & 0x1f) ?
1930 SHIFT_IMM_ROR((0x20 - args[2]) & 0x1f) :
1931 SHIFT_IMM_LSL(0));
1932 } else {
1933 tcg_out_dat_imm(s, COND_AL, ARITH_RSB, TCG_REG_TMP, args[2], 0x20);
1934 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
1935 SHIFT_REG_ROR(TCG_REG_TMP));
1936 }
1937 break;
1938
1939 case INDEX_op_ctz_i32:
1940 tcg_out_dat_reg(s, COND_AL, INSN_RBIT, TCG_REG_TMP, 0, args[1], 0);
1941 a1 = TCG_REG_TMP;
1942 goto do_clz;
1943
1944 case INDEX_op_clz_i32:
1945 a1 = args[1];
1946 do_clz:
1947 a0 = args[0];
1948 a2 = args[2];
1949 c = const_args[2];
1950 if (c && a2 == 32) {
1951 tcg_out_dat_reg(s, COND_AL, INSN_CLZ, a0, 0, a1, 0);
1952 break;
1953 }
1954 tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0, a1, 0);
1955 tcg_out_dat_reg(s, COND_NE, INSN_CLZ, a0, 0, a1, 0);
1956 if (c || a0 != a2) {
1957 tcg_out_dat_rIK(s, COND_EQ, ARITH_MOV, ARITH_MVN, a0, 0, a2, c);
1958 }
1959 break;
1960
1961 case INDEX_op_brcond_i32:
1962 tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
1963 args[0], args[1], const_args[1]);
1964 tcg_out_goto_label(s, tcg_cond_to_arm_cond[args[2]],
1965 arg_label(args[3]));
1966 break;
1967 case INDEX_op_brcond2_i32:
1968 /* The resulting conditions are:
1969 * TCG_COND_EQ --> a0 == a2 && a1 == a3,
1970 * TCG_COND_NE --> (a0 != a2 && a1 == a3) || a1 != a3,
1971 * TCG_COND_LT(U) --> (a0 < a2 && a1 == a3) || a1 < a3,
1972 * TCG_COND_GE(U) --> (a0 >= a2 && a1 == a3) || (a1 >= a3 && a1 != a3),
1973 * TCG_COND_LE(U) --> (a0 <= a2 && a1 == a3) || (a1 <= a3 && a1 != a3),
1974 * TCG_COND_GT(U) --> (a0 > a2 && a1 == a3) || a1 > a3,
1975 */
1976 tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
1977 args[1], args[3], const_args[3]);
1978 tcg_out_dat_rIN(s, COND_EQ, ARITH_CMP, ARITH_CMN, 0,
1979 args[0], args[2], const_args[2]);
1980 tcg_out_goto_label(s, tcg_cond_to_arm_cond[args[4]],
1981 arg_label(args[5]));
1982 break;
1983 case INDEX_op_setcond_i32:
1984 tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
1985 args[1], args[2], const_args[2]);
1986 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[args[3]],
1987 ARITH_MOV, args[0], 0, 1);
1988 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(args[3])],
1989 ARITH_MOV, args[0], 0, 0);
1990 break;
1991 case INDEX_op_setcond2_i32:
1992 /* See brcond2_i32 comment */
1993 tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
1994 args[2], args[4], const_args[4]);
1995 tcg_out_dat_rIN(s, COND_EQ, ARITH_CMP, ARITH_CMN, 0,
1996 args[1], args[3], const_args[3]);
1997 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[args[5]],
1998 ARITH_MOV, args[0], 0, 1);
1999 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(args[5])],
2000 ARITH_MOV, args[0], 0, 0);
2001 break;
2002
2003 case INDEX_op_qemu_ld_i32:
2004 tcg_out_qemu_ld(s, args, 0);
2005 break;
2006 case INDEX_op_qemu_ld_i64:
2007 tcg_out_qemu_ld(s, args, 1);
2008 break;
2009 case INDEX_op_qemu_st_i32:
2010 tcg_out_qemu_st(s, args, 0);
2011 break;
2012 case INDEX_op_qemu_st_i64:
2013 tcg_out_qemu_st(s, args, 1);
2014 break;
2015
2016 case INDEX_op_bswap16_i32:
2017 tcg_out_bswap16(s, COND_AL, args[0], args[1]);
2018 break;
2019 case INDEX_op_bswap32_i32:
2020 tcg_out_bswap32(s, COND_AL, args[0], args[1]);
2021 break;
2022
2023 case INDEX_op_ext8s_i32:
2024 tcg_out_ext8s(s, COND_AL, args[0], args[1]);
2025 break;
2026 case INDEX_op_ext16s_i32:
2027 tcg_out_ext16s(s, COND_AL, args[0], args[1]);
2028 break;
2029 case INDEX_op_ext16u_i32:
2030 tcg_out_ext16u(s, COND_AL, args[0], args[1]);
2031 break;
2032
2033 case INDEX_op_deposit_i32:
2034 tcg_out_deposit(s, COND_AL, args[0], args[2],
2035 args[3], args[4], const_args[2]);
2036 break;
2037 case INDEX_op_extract_i32:
2038 tcg_out_extract(s, COND_AL, args[0], args[1], args[2], args[3]);
2039 break;
2040 case INDEX_op_sextract_i32:
2041 tcg_out_sextract(s, COND_AL, args[0], args[1], args[2], args[3]);
2042 break;
2043
2044 case INDEX_op_div_i32:
2045 tcg_out_sdiv(s, COND_AL, args[0], args[1], args[2]);
2046 break;
2047 case INDEX_op_divu_i32:
2048 tcg_out_udiv(s, COND_AL, args[0], args[1], args[2]);
2049 break;
2050
2051 case INDEX_op_mb:
2052 tcg_out_mb(s, args[0]);
2053 break;
2054
2055 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */
2056 case INDEX_op_movi_i32: /* Always emitted via tcg_out_movi. */
2057 case INDEX_op_call: /* Always emitted via tcg_out_call. */
2058 default:
2059 tcg_abort();
2060 }
2061 }
2062
2063 static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op)
2064 {
2065 static const TCGTargetOpDef r = { .args_ct_str = { "r" } };
2066 static const TCGTargetOpDef r_r = { .args_ct_str = { "r", "r" } };
2067 static const TCGTargetOpDef s_s = { .args_ct_str = { "s", "s" } };
2068 static const TCGTargetOpDef r_l = { .args_ct_str = { "r", "l" } };
2069 static const TCGTargetOpDef r_r_r = { .args_ct_str = { "r", "r", "r" } };
2070 static const TCGTargetOpDef r_r_l = { .args_ct_str = { "r", "r", "l" } };
2071 static const TCGTargetOpDef r_l_l = { .args_ct_str = { "r", "l", "l" } };
2072 static const TCGTargetOpDef s_s_s = { .args_ct_str = { "s", "s", "s" } };
2073 static const TCGTargetOpDef r_r_ri = { .args_ct_str = { "r", "r", "ri" } };
2074 static const TCGTargetOpDef r_r_rI = { .args_ct_str = { "r", "r", "rI" } };
2075 static const TCGTargetOpDef r_r_rIN
2076 = { .args_ct_str = { "r", "r", "rIN" } };
2077 static const TCGTargetOpDef r_r_rIK
2078 = { .args_ct_str = { "r", "r", "rIK" } };
2079 static const TCGTargetOpDef r_r_r_r
2080 = { .args_ct_str = { "r", "r", "r", "r" } };
2081 static const TCGTargetOpDef r_r_l_l
2082 = { .args_ct_str = { "r", "r", "l", "l" } };
2083 static const TCGTargetOpDef s_s_s_s
2084 = { .args_ct_str = { "s", "s", "s", "s" } };
2085 static const TCGTargetOpDef br
2086 = { .args_ct_str = { "r", "rIN" } };
2087 static const TCGTargetOpDef dep
2088 = { .args_ct_str = { "r", "0", "rZ" } };
2089 static const TCGTargetOpDef movc
2090 = { .args_ct_str = { "r", "r", "rIN", "rIK", "0" } };
2091 static const TCGTargetOpDef add2
2092 = { .args_ct_str = { "r", "r", "r", "r", "rIN", "rIK" } };
2093 static const TCGTargetOpDef sub2
2094 = { .args_ct_str = { "r", "r", "rI", "rI", "rIN", "rIK" } };
2095 static const TCGTargetOpDef br2
2096 = { .args_ct_str = { "r", "r", "rIN", "rIN" } };
2097 static const TCGTargetOpDef setc2
2098 = { .args_ct_str = { "r", "r", "r", "rIN", "rIN" } };
2099
2100 switch (op) {
2101 case INDEX_op_goto_ptr:
2102 return &r;
2103
2104 case INDEX_op_ld8u_i32:
2105 case INDEX_op_ld8s_i32:
2106 case INDEX_op_ld16u_i32:
2107 case INDEX_op_ld16s_i32:
2108 case INDEX_op_ld_i32:
2109 case INDEX_op_st8_i32:
2110 case INDEX_op_st16_i32:
2111 case INDEX_op_st_i32:
2112 case INDEX_op_neg_i32:
2113 case INDEX_op_not_i32:
2114 case INDEX_op_bswap16_i32:
2115 case INDEX_op_bswap32_i32:
2116 case INDEX_op_ext8s_i32:
2117 case INDEX_op_ext16s_i32:
2118 case INDEX_op_ext16u_i32:
2119 case INDEX_op_extract_i32:
2120 case INDEX_op_sextract_i32:
2121 return &r_r;
2122
2123 case INDEX_op_add_i32:
2124 case INDEX_op_sub_i32:
2125 case INDEX_op_setcond_i32:
2126 return &r_r_rIN;
2127 case INDEX_op_and_i32:
2128 case INDEX_op_andc_i32:
2129 case INDEX_op_clz_i32:
2130 case INDEX_op_ctz_i32:
2131 return &r_r_rIK;
2132 case INDEX_op_mul_i32:
2133 case INDEX_op_div_i32:
2134 case INDEX_op_divu_i32:
2135 return &r_r_r;
2136 case INDEX_op_mulu2_i32:
2137 case INDEX_op_muls2_i32:
2138 return &r_r_r_r;
2139 case INDEX_op_or_i32:
2140 case INDEX_op_xor_i32:
2141 return &r_r_rI;
2142 case INDEX_op_shl_i32:
2143 case INDEX_op_shr_i32:
2144 case INDEX_op_sar_i32:
2145 case INDEX_op_rotl_i32:
2146 case INDEX_op_rotr_i32:
2147 return &r_r_ri;
2148
2149 case INDEX_op_brcond_i32:
2150 return &br;
2151 case INDEX_op_deposit_i32:
2152 return &dep;
2153 case INDEX_op_movcond_i32:
2154 return &movc;
2155 case INDEX_op_add2_i32:
2156 return &add2;
2157 case INDEX_op_sub2_i32:
2158 return &sub2;
2159 case INDEX_op_brcond2_i32:
2160 return &br2;
2161 case INDEX_op_setcond2_i32:
2162 return &setc2;
2163
2164 case INDEX_op_qemu_ld_i32:
2165 return TARGET_LONG_BITS == 32 ? &r_l : &r_l_l;
2166 case INDEX_op_qemu_ld_i64:
2167 return TARGET_LONG_BITS == 32 ? &r_r_l : &r_r_l_l;
2168 case INDEX_op_qemu_st_i32:
2169 return TARGET_LONG_BITS == 32 ? &s_s : &s_s_s;
2170 case INDEX_op_qemu_st_i64:
2171 return TARGET_LONG_BITS == 32 ? &s_s_s : &s_s_s_s;
2172
2173 default:
2174 return NULL;
2175 }
2176 }
2177
2178 static void tcg_target_init(TCGContext *s)
2179 {
2180 /* Only probe for the platform and capabilities if we havn't already
2181 determined maximum values at compile time. */
2182 #ifndef use_idiv_instructions
2183 {
2184 unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2185 use_idiv_instructions = (hwcap & HWCAP_ARM_IDIVA) != 0;
2186 }
2187 #endif
2188 if (__ARM_ARCH < 7) {
2189 const char *pl = (const char *)qemu_getauxval(AT_PLATFORM);
2190 if (pl != NULL && pl[0] == 'v' && pl[1] >= '4' && pl[1] <= '9') {
2191 arm_arch = pl[1] - '0';
2192 }
2193 }
2194
2195 tcg_target_available_regs[TCG_TYPE_I32] = 0xffff;
2196
2197 tcg_target_call_clobber_regs = 0;
2198 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
2199 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
2200 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
2201 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
2202 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
2203 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
2204
2205 s->reserved_regs = 0;
2206 tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
2207 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP);
2208 tcg_regset_set_reg(s->reserved_regs, TCG_REG_PC);
2209 }
2210
2211 static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
2212 TCGReg arg1, intptr_t arg2)
2213 {
2214 tcg_out_ld32u(s, COND_AL, arg, arg1, arg2);
2215 }
2216
2217 static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
2218 TCGReg arg1, intptr_t arg2)
2219 {
2220 tcg_out_st32(s, COND_AL, arg, arg1, arg2);
2221 }
2222
2223 static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
2224 TCGReg base, intptr_t ofs)
2225 {
2226 return false;
2227 }
2228
2229 static inline void tcg_out_mov(TCGContext *s, TCGType type,
2230 TCGReg ret, TCGReg arg)
2231 {
2232 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, ret, 0, arg, SHIFT_IMM_LSL(0));
2233 }
2234
2235 static inline void tcg_out_movi(TCGContext *s, TCGType type,
2236 TCGReg ret, tcg_target_long arg)
2237 {
2238 tcg_out_movi32(s, COND_AL, ret, arg);
2239 }
2240
2241 static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
2242 {
2243 int i;
2244 for (i = 0; i < count; ++i) {
2245 p[i] = INSN_NOP;
2246 }
2247 }
2248
2249 /* Compute frame size via macros, to share between tcg_target_qemu_prologue
2250 and tcg_register_jit. */
2251
2252 #define PUSH_SIZE ((11 - 4 + 1 + 1) * sizeof(tcg_target_long))
2253
2254 #define FRAME_SIZE \
2255 ((PUSH_SIZE \
2256 + TCG_STATIC_CALL_ARGS_SIZE \
2257 + CPU_TEMP_BUF_NLONGS * sizeof(long) \
2258 + TCG_TARGET_STACK_ALIGN - 1) \
2259 & -TCG_TARGET_STACK_ALIGN)
2260
2261 static void tcg_target_qemu_prologue(TCGContext *s)
2262 {
2263 int stack_addend;
2264
2265 /* Calling convention requires us to save r4-r11 and lr. */
2266 /* stmdb sp!, { r4 - r11, lr } */
2267 tcg_out32(s, (COND_AL << 28) | 0x092d4ff0);
2268
2269 /* Reserve callee argument and tcg temp space. */
2270 stack_addend = FRAME_SIZE - PUSH_SIZE;
2271
2272 tcg_out_dat_rI(s, COND_AL, ARITH_SUB, TCG_REG_CALL_STACK,
2273 TCG_REG_CALL_STACK, stack_addend, 1);
2274 tcg_set_frame(s, TCG_REG_CALL_STACK, TCG_STATIC_CALL_ARGS_SIZE,
2275 CPU_TEMP_BUF_NLONGS * sizeof(long));
2276
2277 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2278
2279 tcg_out_bx(s, COND_AL, tcg_target_call_iarg_regs[1]);
2280
2281 /*
2282 * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
2283 * and fall through to the rest of the epilogue.
2284 */
2285 s->code_gen_epilogue = s->code_ptr;
2286 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, 0);
2287
2288 /* TB epilogue */
2289 tb_ret_addr = s->code_ptr;
2290 tcg_out_dat_rI(s, COND_AL, ARITH_ADD, TCG_REG_CALL_STACK,
2291 TCG_REG_CALL_STACK, stack_addend, 1);
2292
2293 /* ldmia sp!, { r4 - r11, pc } */
2294 tcg_out32(s, (COND_AL << 28) | 0x08bd8ff0);
2295 }
2296
2297 typedef struct {
2298 DebugFrameHeader h;
2299 uint8_t fde_def_cfa[4];
2300 uint8_t fde_reg_ofs[18];
2301 } DebugFrame;
2302
2303 #define ELF_HOST_MACHINE EM_ARM
2304
2305 /* We're expecting a 2 byte uleb128 encoded value. */
2306 QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
2307
2308 static const DebugFrame debug_frame = {
2309 .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
2310 .h.cie.id = -1,
2311 .h.cie.version = 1,
2312 .h.cie.code_align = 1,
2313 .h.cie.data_align = 0x7c, /* sleb128 -4 */
2314 .h.cie.return_column = 14,
2315
2316 /* Total FDE size does not include the "len" member. */
2317 .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
2318
2319 .fde_def_cfa = {
2320 12, 13, /* DW_CFA_def_cfa sp, ... */
2321 (FRAME_SIZE & 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */
2322 (FRAME_SIZE >> 7)
2323 },
2324 .fde_reg_ofs = {
2325 /* The following must match the stmdb in the prologue. */
2326 0x8e, 1, /* DW_CFA_offset, lr, -4 */
2327 0x8b, 2, /* DW_CFA_offset, r11, -8 */
2328 0x8a, 3, /* DW_CFA_offset, r10, -12 */
2329 0x89, 4, /* DW_CFA_offset, r9, -16 */
2330 0x88, 5, /* DW_CFA_offset, r8, -20 */
2331 0x87, 6, /* DW_CFA_offset, r7, -24 */
2332 0x86, 7, /* DW_CFA_offset, r6, -28 */
2333 0x85, 8, /* DW_CFA_offset, r5, -32 */
2334 0x84, 9, /* DW_CFA_offset, r4, -36 */
2335 }
2336 };
2337
2338 void tcg_register_jit(void *buf, size_t buf_size)
2339 {
2340 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
2341 }
AR7 emulation for QEMU