?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge remote-tracking branch 'remotes/kraxel/tags/pull-input-20160928-1' into staging
[qemu/kevin.git] / target-arm / translate.c
blob693d4bc6a245a4cb32477289f51357dfe7ebd742
1 /*
2 * ARM translation
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2005-2007 CodeSourcery
6 * Copyright (c) 2007 OpenedHand, Ltd.
7 *
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2 of the License, or (at your option) any later version.
12 *
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
20 */
21 #include "qemu/osdep.h"
22
23 #include "cpu.h"
24 #include "internals.h"
25 #include "disas/disas.h"
26 #include "exec/exec-all.h"
27 #include "tcg-op.h"
28 #include "qemu/log.h"
29 #include "qemu/bitops.h"
30 #include "arm_ldst.h"
31
32 #include "exec/helper-proto.h"
33 #include "exec/helper-gen.h"
34
35 #include "trace-tcg.h"
36 #include "exec/log.h"
37
38
39 #define ENABLE_ARCH_4T arm_dc_feature(s, ARM_FEATURE_V4T)
40 #define ENABLE_ARCH_5 arm_dc_feature(s, ARM_FEATURE_V5)
41 /* currently all emulated v5 cores are also v5TE, so don't bother */
42 #define ENABLE_ARCH_5TE arm_dc_feature(s, ARM_FEATURE_V5)
43 #define ENABLE_ARCH_5J 0
44 #define ENABLE_ARCH_6 arm_dc_feature(s, ARM_FEATURE_V6)
45 #define ENABLE_ARCH_6K arm_dc_feature(s, ARM_FEATURE_V6K)
46 #define ENABLE_ARCH_6T2 arm_dc_feature(s, ARM_FEATURE_THUMB2)
47 #define ENABLE_ARCH_7 arm_dc_feature(s, ARM_FEATURE_V7)
48 #define ENABLE_ARCH_8 arm_dc_feature(s, ARM_FEATURE_V8)
49
50 #define ARCH(x) do { if (!ENABLE_ARCH_##x) goto illegal_op; } while(0)
51
52 #include "translate.h"
53
54 #if defined(CONFIG_USER_ONLY)
55 #define IS_USER(s) 1
56 #else
57 #define IS_USER(s) (s->user)
58 #endif
59
60 TCGv_env cpu_env;
61 /* We reuse the same 64-bit temporaries for efficiency. */
62 static TCGv_i64 cpu_V0, cpu_V1, cpu_M0;
63 static TCGv_i32 cpu_R[16];
64 TCGv_i32 cpu_CF, cpu_NF, cpu_VF, cpu_ZF;
65 TCGv_i64 cpu_exclusive_addr;
66 TCGv_i64 cpu_exclusive_val;
67 #ifdef CONFIG_USER_ONLY
68 TCGv_i64 cpu_exclusive_test;
69 TCGv_i32 cpu_exclusive_info;
70 #endif
71
72 /* FIXME: These should be removed. */
73 static TCGv_i32 cpu_F0s, cpu_F1s;
74 static TCGv_i64 cpu_F0d, cpu_F1d;
75
76 #include "exec/gen-icount.h"
77
78 static const char *regnames[] =
79 { "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7",
80 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "pc" };
81
82 /* initialize TCG globals. */
83 void arm_translate_init(void)
84 {
85 int i;
86
87 cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env");
88 tcg_ctx.tcg_env = cpu_env;
89
90 for (i = 0; i < 16; i++) {
91 cpu_R[i] = tcg_global_mem_new_i32(cpu_env,
92 offsetof(CPUARMState, regs[i]),
93 regnames[i]);
94 }
95 cpu_CF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, CF), "CF");
96 cpu_NF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, NF), "NF");
97 cpu_VF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, VF), "VF");
98 cpu_ZF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, ZF), "ZF");
99
100 cpu_exclusive_addr = tcg_global_mem_new_i64(cpu_env,
101 offsetof(CPUARMState, exclusive_addr), "exclusive_addr");
102 cpu_exclusive_val = tcg_global_mem_new_i64(cpu_env,
103 offsetof(CPUARMState, exclusive_val), "exclusive_val");
104 #ifdef CONFIG_USER_ONLY
105 cpu_exclusive_test = tcg_global_mem_new_i64(cpu_env,
106 offsetof(CPUARMState, exclusive_test), "exclusive_test");
107 cpu_exclusive_info = tcg_global_mem_new_i32(cpu_env,
108 offsetof(CPUARMState, exclusive_info), "exclusive_info");
109 #endif
110
111 a64_translate_init();
112 }
113
114 static inline ARMMMUIdx get_a32_user_mem_index(DisasContext *s)
115 {
116 /* Return the mmu_idx to use for A32/T32 "unprivileged load/store"
117 * insns:
118 * if PL2, UNPREDICTABLE (we choose to implement as if PL0)
119 * otherwise, access as if at PL0.
120 */
121 switch (s->mmu_idx) {
122 case ARMMMUIdx_S1E2: /* this one is UNPREDICTABLE */
123 case ARMMMUIdx_S12NSE0:
124 case ARMMMUIdx_S12NSE1:
125 return ARMMMUIdx_S12NSE0;
126 case ARMMMUIdx_S1E3:
127 case ARMMMUIdx_S1SE0:
128 case ARMMMUIdx_S1SE1:
129 return ARMMMUIdx_S1SE0;
130 case ARMMMUIdx_S2NS:
131 default:
132 g_assert_not_reached();
133 }
134 }
135
136 static inline TCGv_i32 load_cpu_offset(int offset)
137 {
138 TCGv_i32 tmp = tcg_temp_new_i32();
139 tcg_gen_ld_i32(tmp, cpu_env, offset);
140 return tmp;
141 }
142
143 #define load_cpu_field(name) load_cpu_offset(offsetof(CPUARMState, name))
144
145 static inline void store_cpu_offset(TCGv_i32 var, int offset)
146 {
147 tcg_gen_st_i32(var, cpu_env, offset);
148 tcg_temp_free_i32(var);
149 }
150
151 #define store_cpu_field(var, name) \
152 store_cpu_offset(var, offsetof(CPUARMState, name))
153
154 /* Set a variable to the value of a CPU register. */
155 static void load_reg_var(DisasContext *s, TCGv_i32 var, int reg)
156 {
157 if (reg == 15) {
158 uint32_t addr;
159 /* normally, since we updated PC, we need only to add one insn */
160 if (s->thumb)
161 addr = (long)s->pc + 2;
162 else
163 addr = (long)s->pc + 4;
164 tcg_gen_movi_i32(var, addr);
165 } else {
166 tcg_gen_mov_i32(var, cpu_R[reg]);
167 }
168 }
169
170 /* Create a new temporary and set it to the value of a CPU register. */
171 static inline TCGv_i32 load_reg(DisasContext *s, int reg)
172 {
173 TCGv_i32 tmp = tcg_temp_new_i32();
174 load_reg_var(s, tmp, reg);
175 return tmp;
176 }
177
178 /* Set a CPU register. The source must be a temporary and will be
179 marked as dead. */
180 static void store_reg(DisasContext *s, int reg, TCGv_i32 var)
181 {
182 if (reg == 15) {
183 tcg_gen_andi_i32(var, var, ~1);
184 s->is_jmp = DISAS_JUMP;
185 }
186 tcg_gen_mov_i32(cpu_R[reg], var);
187 tcg_temp_free_i32(var);
188 }
189
190 /* Value extensions. */
191 #define gen_uxtb(var) tcg_gen_ext8u_i32(var, var)
192 #define gen_uxth(var) tcg_gen_ext16u_i32(var, var)
193 #define gen_sxtb(var) tcg_gen_ext8s_i32(var, var)
194 #define gen_sxth(var) tcg_gen_ext16s_i32(var, var)
195
196 #define gen_sxtb16(var) gen_helper_sxtb16(var, var)
197 #define gen_uxtb16(var) gen_helper_uxtb16(var, var)
198
199
200 static inline void gen_set_cpsr(TCGv_i32 var, uint32_t mask)
201 {
202 TCGv_i32 tmp_mask = tcg_const_i32(mask);
203 gen_helper_cpsr_write(cpu_env, var, tmp_mask);
204 tcg_temp_free_i32(tmp_mask);
205 }
206 /* Set NZCV flags from the high 4 bits of var. */
207 #define gen_set_nzcv(var) gen_set_cpsr(var, CPSR_NZCV)
208
209 static void gen_exception_internal(int excp)
210 {
211 TCGv_i32 tcg_excp = tcg_const_i32(excp);
212
213 assert(excp_is_internal(excp));
214 gen_helper_exception_internal(cpu_env, tcg_excp);
215 tcg_temp_free_i32(tcg_excp);
216 }
217
218 static void gen_exception(int excp, uint32_t syndrome, uint32_t target_el)
219 {
220 TCGv_i32 tcg_excp = tcg_const_i32(excp);
221 TCGv_i32 tcg_syn = tcg_const_i32(syndrome);
222 TCGv_i32 tcg_el = tcg_const_i32(target_el);
223
224 gen_helper_exception_with_syndrome(cpu_env, tcg_excp,
225 tcg_syn, tcg_el);
226
227 tcg_temp_free_i32(tcg_el);
228 tcg_temp_free_i32(tcg_syn);
229 tcg_temp_free_i32(tcg_excp);
230 }
231
232 static void gen_ss_advance(DisasContext *s)
233 {
234 /* If the singlestep state is Active-not-pending, advance to
235 * Active-pending.
236 */
237 if (s->ss_active) {
238 s->pstate_ss = 0;
239 gen_helper_clear_pstate_ss(cpu_env);
240 }
241 }
242
243 static void gen_step_complete_exception(DisasContext *s)
244 {
245 /* We just completed step of an insn. Move from Active-not-pending
246 * to Active-pending, and then also take the swstep exception.
247 * This corresponds to making the (IMPDEF) choice to prioritize
248 * swstep exceptions over asynchronous exceptions taken to an exception
249 * level where debug is disabled. This choice has the advantage that
250 * we do not need to maintain internal state corresponding to the
251 * ISV/EX syndrome bits between completion of the step and generation
252 * of the exception, and our syndrome information is always correct.
253 */
254 gen_ss_advance(s);
255 gen_exception(EXCP_UDEF, syn_swstep(s->ss_same_el, 1, s->is_ldex),
256 default_exception_el(s));
257 s->is_jmp = DISAS_EXC;
258 }
259
260 static void gen_smul_dual(TCGv_i32 a, TCGv_i32 b)
261 {
262 TCGv_i32 tmp1 = tcg_temp_new_i32();
263 TCGv_i32 tmp2 = tcg_temp_new_i32();
264 tcg_gen_ext16s_i32(tmp1, a);
265 tcg_gen_ext16s_i32(tmp2, b);
266 tcg_gen_mul_i32(tmp1, tmp1, tmp2);
267 tcg_temp_free_i32(tmp2);
268 tcg_gen_sari_i32(a, a, 16);
269 tcg_gen_sari_i32(b, b, 16);
270 tcg_gen_mul_i32(b, b, a);
271 tcg_gen_mov_i32(a, tmp1);
272 tcg_temp_free_i32(tmp1);
273 }
274
275 /* Byteswap each halfword. */
276 static void gen_rev16(TCGv_i32 var)
277 {
278 TCGv_i32 tmp = tcg_temp_new_i32();
279 tcg_gen_shri_i32(tmp, var, 8);
280 tcg_gen_andi_i32(tmp, tmp, 0x00ff00ff);
281 tcg_gen_shli_i32(var, var, 8);
282 tcg_gen_andi_i32(var, var, 0xff00ff00);
283 tcg_gen_or_i32(var, var, tmp);
284 tcg_temp_free_i32(tmp);
285 }
286
287 /* Byteswap low halfword and sign extend. */
288 static void gen_revsh(TCGv_i32 var)
289 {
290 tcg_gen_ext16u_i32(var, var);
291 tcg_gen_bswap16_i32(var, var);
292 tcg_gen_ext16s_i32(var, var);
293 }
294
295 /* Unsigned bitfield extract. */
296 static void gen_ubfx(TCGv_i32 var, int shift, uint32_t mask)
297 {
298 if (shift)
299 tcg_gen_shri_i32(var, var, shift);
300 tcg_gen_andi_i32(var, var, mask);
301 }
302
303 /* Signed bitfield extract. */
304 static void gen_sbfx(TCGv_i32 var, int shift, int width)
305 {
306 uint32_t signbit;
307
308 if (shift)
309 tcg_gen_sari_i32(var, var, shift);
310 if (shift + width < 32) {
311 signbit = 1u << (width - 1);
312 tcg_gen_andi_i32(var, var, (1u << width) - 1);
313 tcg_gen_xori_i32(var, var, signbit);
314 tcg_gen_subi_i32(var, var, signbit);
315 }
316 }
317
318 /* Return (b << 32) + a. Mark inputs as dead */
319 static TCGv_i64 gen_addq_msw(TCGv_i64 a, TCGv_i32 b)
320 {
321 TCGv_i64 tmp64 = tcg_temp_new_i64();
322
323 tcg_gen_extu_i32_i64(tmp64, b);
324 tcg_temp_free_i32(b);
325 tcg_gen_shli_i64(tmp64, tmp64, 32);
326 tcg_gen_add_i64(a, tmp64, a);
327
328 tcg_temp_free_i64(tmp64);
329 return a;
330 }
331
332 /* Return (b << 32) - a. Mark inputs as dead. */
333 static TCGv_i64 gen_subq_msw(TCGv_i64 a, TCGv_i32 b)
334 {
335 TCGv_i64 tmp64 = tcg_temp_new_i64();
336
337 tcg_gen_extu_i32_i64(tmp64, b);
338 tcg_temp_free_i32(b);
339 tcg_gen_shli_i64(tmp64, tmp64, 32);
340 tcg_gen_sub_i64(a, tmp64, a);
341
342 tcg_temp_free_i64(tmp64);
343 return a;
344 }
345
346 /* 32x32->64 multiply. Marks inputs as dead. */
347 static TCGv_i64 gen_mulu_i64_i32(TCGv_i32 a, TCGv_i32 b)
348 {
349 TCGv_i32 lo = tcg_temp_new_i32();
350 TCGv_i32 hi = tcg_temp_new_i32();
351 TCGv_i64 ret;
352
353 tcg_gen_mulu2_i32(lo, hi, a, b);
354 tcg_temp_free_i32(a);
355 tcg_temp_free_i32(b);
356
357 ret = tcg_temp_new_i64();
358 tcg_gen_concat_i32_i64(ret, lo, hi);
359 tcg_temp_free_i32(lo);
360 tcg_temp_free_i32(hi);
361
362 return ret;
363 }
364
365 static TCGv_i64 gen_muls_i64_i32(TCGv_i32 a, TCGv_i32 b)
366 {
367 TCGv_i32 lo = tcg_temp_new_i32();
368 TCGv_i32 hi = tcg_temp_new_i32();
369 TCGv_i64 ret;
370
371 tcg_gen_muls2_i32(lo, hi, a, b);
372 tcg_temp_free_i32(a);
373 tcg_temp_free_i32(b);
374
375 ret = tcg_temp_new_i64();
376 tcg_gen_concat_i32_i64(ret, lo, hi);
377 tcg_temp_free_i32(lo);
378 tcg_temp_free_i32(hi);
379
380 return ret;
381 }
382
383 /* Swap low and high halfwords. */
384 static void gen_swap_half(TCGv_i32 var)
385 {
386 TCGv_i32 tmp = tcg_temp_new_i32();
387 tcg_gen_shri_i32(tmp, var, 16);
388 tcg_gen_shli_i32(var, var, 16);
389 tcg_gen_or_i32(var, var, tmp);
390 tcg_temp_free_i32(tmp);
391 }
392
393 /* Dual 16-bit add. Result placed in t0 and t1 is marked as dead.
394 tmp = (t0 ^ t1) & 0x8000;
395 t0 &= ~0x8000;
396 t1 &= ~0x8000;
397 t0 = (t0 + t1) ^ tmp;
398 */
399
400 static void gen_add16(TCGv_i32 t0, TCGv_i32 t1)
401 {
402 TCGv_i32 tmp = tcg_temp_new_i32();
403 tcg_gen_xor_i32(tmp, t0, t1);
404 tcg_gen_andi_i32(tmp, tmp, 0x8000);
405 tcg_gen_andi_i32(t0, t0, ~0x8000);
406 tcg_gen_andi_i32(t1, t1, ~0x8000);
407 tcg_gen_add_i32(t0, t0, t1);
408 tcg_gen_xor_i32(t0, t0, tmp);
409 tcg_temp_free_i32(tmp);
410 tcg_temp_free_i32(t1);
411 }
412
413 /* Set CF to the top bit of var. */
414 static void gen_set_CF_bit31(TCGv_i32 var)
415 {
416 tcg_gen_shri_i32(cpu_CF, var, 31);
417 }
418
419 /* Set N and Z flags from var. */
420 static inline void gen_logic_CC(TCGv_i32 var)
421 {
422 tcg_gen_mov_i32(cpu_NF, var);
423 tcg_gen_mov_i32(cpu_ZF, var);
424 }
425
426 /* T0 += T1 + CF. */
427 static void gen_adc(TCGv_i32 t0, TCGv_i32 t1)
428 {
429 tcg_gen_add_i32(t0, t0, t1);
430 tcg_gen_add_i32(t0, t0, cpu_CF);
431 }
432
433 /* dest = T0 + T1 + CF. */
434 static void gen_add_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
435 {
436 tcg_gen_add_i32(dest, t0, t1);
437 tcg_gen_add_i32(dest, dest, cpu_CF);
438 }
439
440 /* dest = T0 - T1 + CF - 1. */
441 static void gen_sub_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
442 {
443 tcg_gen_sub_i32(dest, t0, t1);
444 tcg_gen_add_i32(dest, dest, cpu_CF);
445 tcg_gen_subi_i32(dest, dest, 1);
446 }
447
448 /* dest = T0 + T1. Compute C, N, V and Z flags */
449 static void gen_add_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
450 {
451 TCGv_i32 tmp = tcg_temp_new_i32();
452 tcg_gen_movi_i32(tmp, 0);
453 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, t1, tmp);
454 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
455 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
456 tcg_gen_xor_i32(tmp, t0, t1);
457 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
458 tcg_temp_free_i32(tmp);
459 tcg_gen_mov_i32(dest, cpu_NF);
460 }
461
462 /* dest = T0 + T1 + CF. Compute C, N, V and Z flags */
463 static void gen_adc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
464 {
465 TCGv_i32 tmp = tcg_temp_new_i32();
466 if (TCG_TARGET_HAS_add2_i32) {
467 tcg_gen_movi_i32(tmp, 0);
468 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, cpu_CF, tmp);
469 tcg_gen_add2_i32(cpu_NF, cpu_CF, cpu_NF, cpu_CF, t1, tmp);
470 } else {
471 TCGv_i64 q0 = tcg_temp_new_i64();
472 TCGv_i64 q1 = tcg_temp_new_i64();
473 tcg_gen_extu_i32_i64(q0, t0);
474 tcg_gen_extu_i32_i64(q1, t1);
475 tcg_gen_add_i64(q0, q0, q1);
476 tcg_gen_extu_i32_i64(q1, cpu_CF);
477 tcg_gen_add_i64(q0, q0, q1);
478 tcg_gen_extr_i64_i32(cpu_NF, cpu_CF, q0);
479 tcg_temp_free_i64(q0);
480 tcg_temp_free_i64(q1);
481 }
482 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
483 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
484 tcg_gen_xor_i32(tmp, t0, t1);
485 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
486 tcg_temp_free_i32(tmp);
487 tcg_gen_mov_i32(dest, cpu_NF);
488 }
489
490 /* dest = T0 - T1. Compute C, N, V and Z flags */
491 static void gen_sub_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
492 {
493 TCGv_i32 tmp;
494 tcg_gen_sub_i32(cpu_NF, t0, t1);
495 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
496 tcg_gen_setcond_i32(TCG_COND_GEU, cpu_CF, t0, t1);
497 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
498 tmp = tcg_temp_new_i32();
499 tcg_gen_xor_i32(tmp, t0, t1);
500 tcg_gen_and_i32(cpu_VF, cpu_VF, tmp);
501 tcg_temp_free_i32(tmp);
502 tcg_gen_mov_i32(dest, cpu_NF);
503 }
504
505 /* dest = T0 + ~T1 + CF. Compute C, N, V and Z flags */
506 static void gen_sbc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
507 {
508 TCGv_i32 tmp = tcg_temp_new_i32();
509 tcg_gen_not_i32(tmp, t1);
510 gen_adc_CC(dest, t0, tmp);
511 tcg_temp_free_i32(tmp);
512 }
513
514 #define GEN_SHIFT(name) \
515 static void gen_##name(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1) \
516 { \
517 TCGv_i32 tmp1, tmp2, tmp3; \
518 tmp1 = tcg_temp_new_i32(); \
519 tcg_gen_andi_i32(tmp1, t1, 0xff); \
520 tmp2 = tcg_const_i32(0); \
521 tmp3 = tcg_const_i32(0x1f); \
522 tcg_gen_movcond_i32(TCG_COND_GTU, tmp2, tmp1, tmp3, tmp2, t0); \
523 tcg_temp_free_i32(tmp3); \
524 tcg_gen_andi_i32(tmp1, tmp1, 0x1f); \
525 tcg_gen_##name##_i32(dest, tmp2, tmp1); \
526 tcg_temp_free_i32(tmp2); \
527 tcg_temp_free_i32(tmp1); \
528 }
529 GEN_SHIFT(shl)
530 GEN_SHIFT(shr)
531 #undef GEN_SHIFT
532
533 static void gen_sar(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
534 {
535 TCGv_i32 tmp1, tmp2;
536 tmp1 = tcg_temp_new_i32();
537 tcg_gen_andi_i32(tmp1, t1, 0xff);
538 tmp2 = tcg_const_i32(0x1f);
539 tcg_gen_movcond_i32(TCG_COND_GTU, tmp1, tmp1, tmp2, tmp2, tmp1);
540 tcg_temp_free_i32(tmp2);
541 tcg_gen_sar_i32(dest, t0, tmp1);
542 tcg_temp_free_i32(tmp1);
543 }
544
545 static void tcg_gen_abs_i32(TCGv_i32 dest, TCGv_i32 src)
546 {
547 TCGv_i32 c0 = tcg_const_i32(0);
548 TCGv_i32 tmp = tcg_temp_new_i32();
549 tcg_gen_neg_i32(tmp, src);
550 tcg_gen_movcond_i32(TCG_COND_GT, dest, src, c0, src, tmp);
551 tcg_temp_free_i32(c0);
552 tcg_temp_free_i32(tmp);
553 }
554
555 static void shifter_out_im(TCGv_i32 var, int shift)
556 {
557 if (shift == 0) {
558 tcg_gen_andi_i32(cpu_CF, var, 1);
559 } else {
560 tcg_gen_shri_i32(cpu_CF, var, shift);
561 if (shift != 31) {
562 tcg_gen_andi_i32(cpu_CF, cpu_CF, 1);
563 }
564 }
565 }
566
567 /* Shift by immediate. Includes special handling for shift == 0. */
568 static inline void gen_arm_shift_im(TCGv_i32 var, int shiftop,
569 int shift, int flags)
570 {
571 switch (shiftop) {
572 case 0: /* LSL */
573 if (shift != 0) {
574 if (flags)
575 shifter_out_im(var, 32 - shift);
576 tcg_gen_shli_i32(var, var, shift);
577 }
578 break;
579 case 1: /* LSR */
580 if (shift == 0) {
581 if (flags) {
582 tcg_gen_shri_i32(cpu_CF, var, 31);
583 }
584 tcg_gen_movi_i32(var, 0);
585 } else {
586 if (flags)
587 shifter_out_im(var, shift - 1);
588 tcg_gen_shri_i32(var, var, shift);
589 }
590 break;
591 case 2: /* ASR */
592 if (shift == 0)
593 shift = 32;
594 if (flags)
595 shifter_out_im(var, shift - 1);
596 if (shift == 32)
597 shift = 31;
598 tcg_gen_sari_i32(var, var, shift);
599 break;
600 case 3: /* ROR/RRX */
601 if (shift != 0) {
602 if (flags)
603 shifter_out_im(var, shift - 1);
604 tcg_gen_rotri_i32(var, var, shift); break;
605 } else {
606 TCGv_i32 tmp = tcg_temp_new_i32();
607 tcg_gen_shli_i32(tmp, cpu_CF, 31);
608 if (flags)
609 shifter_out_im(var, 0);
610 tcg_gen_shri_i32(var, var, 1);
611 tcg_gen_or_i32(var, var, tmp);
612 tcg_temp_free_i32(tmp);
613 }
614 }
615 };
616
617 static inline void gen_arm_shift_reg(TCGv_i32 var, int shiftop,
618 TCGv_i32 shift, int flags)
619 {
620 if (flags) {
621 switch (shiftop) {
622 case 0: gen_helper_shl_cc(var, cpu_env, var, shift); break;
623 case 1: gen_helper_shr_cc(var, cpu_env, var, shift); break;
624 case 2: gen_helper_sar_cc(var, cpu_env, var, shift); break;
625 case 3: gen_helper_ror_cc(var, cpu_env, var, shift); break;
626 }
627 } else {
628 switch (shiftop) {
629 case 0:
630 gen_shl(var, var, shift);
631 break;
632 case 1:
633 gen_shr(var, var, shift);
634 break;
635 case 2:
636 gen_sar(var, var, shift);
637 break;
638 case 3: tcg_gen_andi_i32(shift, shift, 0x1f);
639 tcg_gen_rotr_i32(var, var, shift); break;
640 }
641 }
642 tcg_temp_free_i32(shift);
643 }
644
645 #define PAS_OP(pfx) \
646 switch (op2) { \
647 case 0: gen_pas_helper(glue(pfx,add16)); break; \
648 case 1: gen_pas_helper(glue(pfx,addsubx)); break; \
649 case 2: gen_pas_helper(glue(pfx,subaddx)); break; \
650 case 3: gen_pas_helper(glue(pfx,sub16)); break; \
651 case 4: gen_pas_helper(glue(pfx,add8)); break; \
652 case 7: gen_pas_helper(glue(pfx,sub8)); break; \
653 }
654 static void gen_arm_parallel_addsub(int op1, int op2, TCGv_i32 a, TCGv_i32 b)
655 {
656 TCGv_ptr tmp;
657
658 switch (op1) {
659 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b, tmp)
660 case 1:
661 tmp = tcg_temp_new_ptr();
662 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
663 PAS_OP(s)
664 tcg_temp_free_ptr(tmp);
665 break;
666 case 5:
667 tmp = tcg_temp_new_ptr();
668 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
669 PAS_OP(u)
670 tcg_temp_free_ptr(tmp);
671 break;
672 #undef gen_pas_helper
673 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b)
674 case 2:
675 PAS_OP(q);
676 break;
677 case 3:
678 PAS_OP(sh);
679 break;
680 case 6:
681 PAS_OP(uq);
682 break;
683 case 7:
684 PAS_OP(uh);
685 break;
686 #undef gen_pas_helper
687 }
688 }
689 #undef PAS_OP
690
691 /* For unknown reasons Arm and Thumb-2 use arbitrarily different encodings. */
692 #define PAS_OP(pfx) \
693 switch (op1) { \
694 case 0: gen_pas_helper(glue(pfx,add8)); break; \
695 case 1: gen_pas_helper(glue(pfx,add16)); break; \
696 case 2: gen_pas_helper(glue(pfx,addsubx)); break; \
697 case 4: gen_pas_helper(glue(pfx,sub8)); break; \
698 case 5: gen_pas_helper(glue(pfx,sub16)); break; \
699 case 6: gen_pas_helper(glue(pfx,subaddx)); break; \
700 }
701 static void gen_thumb2_parallel_addsub(int op1, int op2, TCGv_i32 a, TCGv_i32 b)
702 {
703 TCGv_ptr tmp;
704
705 switch (op2) {
706 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b, tmp)
707 case 0:
708 tmp = tcg_temp_new_ptr();
709 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
710 PAS_OP(s)
711 tcg_temp_free_ptr(tmp);
712 break;
713 case 4:
714 tmp = tcg_temp_new_ptr();
715 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
716 PAS_OP(u)
717 tcg_temp_free_ptr(tmp);
718 break;
719 #undef gen_pas_helper
720 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b)
721 case 1:
722 PAS_OP(q);
723 break;
724 case 2:
725 PAS_OP(sh);
726 break;
727 case 5:
728 PAS_OP(uq);
729 break;
730 case 6:
731 PAS_OP(uh);
732 break;
733 #undef gen_pas_helper
734 }
735 }
736 #undef PAS_OP
737
738 /*
739 * Generate a conditional based on ARM condition code cc.
740 * This is common between ARM and Aarch64 targets.
741 */
742 void arm_test_cc(DisasCompare *cmp, int cc)
743 {
744 TCGv_i32 value;
745 TCGCond cond;
746 bool global = true;
747
748 switch (cc) {
749 case 0: /* eq: Z */
750 case 1: /* ne: !Z */
751 cond = TCG_COND_EQ;
752 value = cpu_ZF;
753 break;
754
755 case 2: /* cs: C */
756 case 3: /* cc: !C */
757 cond = TCG_COND_NE;
758 value = cpu_CF;
759 break;
760
761 case 4: /* mi: N */
762 case 5: /* pl: !N */
763 cond = TCG_COND_LT;
764 value = cpu_NF;
765 break;
766
767 case 6: /* vs: V */
768 case 7: /* vc: !V */
769 cond = TCG_COND_LT;
770 value = cpu_VF;
771 break;
772
773 case 8: /* hi: C && !Z */
774 case 9: /* ls: !C || Z -> !(C && !Z) */
775 cond = TCG_COND_NE;
776 value = tcg_temp_new_i32();
777 global = false;
778 /* CF is 1 for C, so -CF is an all-bits-set mask for C;
779 ZF is non-zero for !Z; so AND the two subexpressions. */
780 tcg_gen_neg_i32(value, cpu_CF);
781 tcg_gen_and_i32(value, value, cpu_ZF);
782 break;
783
784 case 10: /* ge: N == V -> N ^ V == 0 */
785 case 11: /* lt: N != V -> N ^ V != 0 */
786 /* Since we're only interested in the sign bit, == 0 is >= 0. */
787 cond = TCG_COND_GE;
788 value = tcg_temp_new_i32();
789 global = false;
790 tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
791 break;
792
793 case 12: /* gt: !Z && N == V */
794 case 13: /* le: Z || N != V */
795 cond = TCG_COND_NE;
796 value = tcg_temp_new_i32();
797 global = false;
798 /* (N == V) is equal to the sign bit of ~(NF ^ VF). Propagate
799 * the sign bit then AND with ZF to yield the result. */
800 tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
801 tcg_gen_sari_i32(value, value, 31);
802 tcg_gen_andc_i32(value, cpu_ZF, value);
803 break;
804
805 case 14: /* always */
806 case 15: /* always */
807 /* Use the ALWAYS condition, which will fold early.
808 * It doesn't matter what we use for the value. */
809 cond = TCG_COND_ALWAYS;
810 value = cpu_ZF;
811 goto no_invert;
812
813 default:
814 fprintf(stderr, "Bad condition code 0x%x\n", cc);
815 abort();
816 }
817
818 if (cc & 1) {
819 cond = tcg_invert_cond(cond);
820 }
821
822 no_invert:
823 cmp->cond = cond;
824 cmp->value = value;
825 cmp->value_global = global;
826 }
827
828 void arm_free_cc(DisasCompare *cmp)
829 {
830 if (!cmp->value_global) {
831 tcg_temp_free_i32(cmp->value);
832 }
833 }
834
835 void arm_jump_cc(DisasCompare *cmp, TCGLabel *label)
836 {
837 tcg_gen_brcondi_i32(cmp->cond, cmp->value, 0, label);
838 }
839
840 void arm_gen_test_cc(int cc, TCGLabel *label)
841 {
842 DisasCompare cmp;
843 arm_test_cc(&cmp, cc);
844 arm_jump_cc(&cmp, label);
845 arm_free_cc(&cmp);
846 }
847
848 static const uint8_t table_logic_cc[16] = {
849 1, /* and */
850 1, /* xor */
851 0, /* sub */
852 0, /* rsb */
853 0, /* add */
854 0, /* adc */
855 0, /* sbc */
856 0, /* rsc */
857 1, /* andl */
858 1, /* xorl */
859 0, /* cmp */
860 0, /* cmn */
861 1, /* orr */
862 1, /* mov */
863 1, /* bic */
864 1, /* mvn */
865 };
866
867 /* Set PC and Thumb state from an immediate address. */
868 static inline void gen_bx_im(DisasContext *s, uint32_t addr)
869 {
870 TCGv_i32 tmp;
871
872 s->is_jmp = DISAS_JUMP;
873 if (s->thumb != (addr & 1)) {
874 tmp = tcg_temp_new_i32();
875 tcg_gen_movi_i32(tmp, addr & 1);
876 tcg_gen_st_i32(tmp, cpu_env, offsetof(CPUARMState, thumb));
877 tcg_temp_free_i32(tmp);
878 }
879 tcg_gen_movi_i32(cpu_R[15], addr & ~1);
880 }
881
882 /* Set PC and Thumb state from var. var is marked as dead. */
883 static inline void gen_bx(DisasContext *s, TCGv_i32 var)
884 {
885 s->is_jmp = DISAS_JUMP;
886 tcg_gen_andi_i32(cpu_R[15], var, ~1);
887 tcg_gen_andi_i32(var, var, 1);
888 store_cpu_field(var, thumb);
889 }
890
891 /* Variant of store_reg which uses branch&exchange logic when storing
892 to r15 in ARM architecture v7 and above. The source must be a temporary
893 and will be marked as dead. */
894 static inline void store_reg_bx(DisasContext *s, int reg, TCGv_i32 var)
895 {
896 if (reg == 15 && ENABLE_ARCH_7) {
897 gen_bx(s, var);
898 } else {
899 store_reg(s, reg, var);
900 }
901 }
902
903 /* Variant of store_reg which uses branch&exchange logic when storing
904 * to r15 in ARM architecture v5T and above. This is used for storing
905 * the results of a LDR/LDM/POP into r15, and corresponds to the cases
906 * in the ARM ARM which use the LoadWritePC() pseudocode function. */
907 static inline void store_reg_from_load(DisasContext *s, int reg, TCGv_i32 var)
908 {
909 if (reg == 15 && ENABLE_ARCH_5) {
910 gen_bx(s, var);
911 } else {
912 store_reg(s, reg, var);
913 }
914 }
915
916 #ifdef CONFIG_USER_ONLY
917 #define IS_USER_ONLY 1
918 #else
919 #define IS_USER_ONLY 0
920 #endif
921
922 /* Abstractions of "generate code to do a guest load/store for
923 * AArch32", where a vaddr is always 32 bits (and is zero
924 * extended if we're a 64 bit core) and data is also
925 * 32 bits unless specifically doing a 64 bit access.
926 * These functions work like tcg_gen_qemu_{ld,st}* except
927 * that the address argument is TCGv_i32 rather than TCGv.
928 */
929 #if TARGET_LONG_BITS == 32
930
931 #define DO_GEN_LD(SUFF, OPC, BE32_XOR) \
932 static inline void gen_aa32_ld##SUFF(DisasContext *s, TCGv_i32 val, \
933 TCGv_i32 addr, int index) \
934 { \
935 TCGMemOp opc = (OPC) | s->be_data; \
936 /* Not needed for user-mode BE32, where we use MO_BE instead. */ \
937 if (!IS_USER_ONLY && s->sctlr_b && BE32_XOR) { \
938 TCGv addr_be = tcg_temp_new(); \
939 tcg_gen_xori_i32(addr_be, addr, BE32_XOR); \
940 tcg_gen_qemu_ld_i32(val, addr_be, index, opc); \
941 tcg_temp_free(addr_be); \
942 return; \
943 } \
944 tcg_gen_qemu_ld_i32(val, addr, index, opc); \
945 }
946
947 #define DO_GEN_ST(SUFF, OPC, BE32_XOR) \
948 static inline void gen_aa32_st##SUFF(DisasContext *s, TCGv_i32 val, \
949 TCGv_i32 addr, int index) \
950 { \
951 TCGMemOp opc = (OPC) | s->be_data; \
952 /* Not needed for user-mode BE32, where we use MO_BE instead. */ \
953 if (!IS_USER_ONLY && s->sctlr_b && BE32_XOR) { \
954 TCGv addr_be = tcg_temp_new(); \
955 tcg_gen_xori_i32(addr_be, addr, BE32_XOR); \
956 tcg_gen_qemu_st_i32(val, addr_be, index, opc); \
957 tcg_temp_free(addr_be); \
958 return; \
959 } \
960 tcg_gen_qemu_st_i32(val, addr, index, opc); \
961 }
962
963 static inline void gen_aa32_ld64(DisasContext *s, TCGv_i64 val,
964 TCGv_i32 addr, int index)
965 {
966 TCGMemOp opc = MO_Q | s->be_data;
967 tcg_gen_qemu_ld_i64(val, addr, index, opc);
968 /* Not needed for user-mode BE32, where we use MO_BE instead. */
969 if (!IS_USER_ONLY && s->sctlr_b) {
970 tcg_gen_rotri_i64(val, val, 32);
971 }
972 }
973
974 static inline void gen_aa32_st64(DisasContext *s, TCGv_i64 val,
975 TCGv_i32 addr, int index)
976 {
977 TCGMemOp opc = MO_Q | s->be_data;
978 /* Not needed for user-mode BE32, where we use MO_BE instead. */
979 if (!IS_USER_ONLY && s->sctlr_b) {
980 TCGv_i64 tmp = tcg_temp_new_i64();
981 tcg_gen_rotri_i64(tmp, val, 32);
982 tcg_gen_qemu_st_i64(tmp, addr, index, opc);
983 tcg_temp_free_i64(tmp);
984 return;
985 }
986 tcg_gen_qemu_st_i64(val, addr, index, opc);
987 }
988
989 #else
990
991 #define DO_GEN_LD(SUFF, OPC, BE32_XOR) \
992 static inline void gen_aa32_ld##SUFF(DisasContext *s, TCGv_i32 val, \
993 TCGv_i32 addr, int index) \
994 { \
995 TCGMemOp opc = (OPC) | s->be_data; \
996 TCGv addr64 = tcg_temp_new(); \
997 tcg_gen_extu_i32_i64(addr64, addr); \
998 /* Not needed for user-mode BE32, where we use MO_BE instead. */ \
999 if (!IS_USER_ONLY && s->sctlr_b && BE32_XOR) { \
1000 tcg_gen_xori_i64(addr64, addr64, BE32_XOR); \
1001 } \
1002 tcg_gen_qemu_ld_i32(val, addr64, index, opc); \
1003 tcg_temp_free(addr64); \
1004 }
1005
1006 #define DO_GEN_ST(SUFF, OPC, BE32_XOR) \
1007 static inline void gen_aa32_st##SUFF(DisasContext *s, TCGv_i32 val, \
1008 TCGv_i32 addr, int index) \
1009 { \
1010 TCGMemOp opc = (OPC) | s->be_data; \
1011 TCGv addr64 = tcg_temp_new(); \
1012 tcg_gen_extu_i32_i64(addr64, addr); \
1013 /* Not needed for user-mode BE32, where we use MO_BE instead. */ \
1014 if (!IS_USER_ONLY && s->sctlr_b && BE32_XOR) { \
1015 tcg_gen_xori_i64(addr64, addr64, BE32_XOR); \
1016 } \
1017 tcg_gen_qemu_st_i32(val, addr64, index, opc); \
1018 tcg_temp_free(addr64); \
1019 }
1020
1021 static inline void gen_aa32_ld64(DisasContext *s, TCGv_i64 val,
1022 TCGv_i32 addr, int index)
1023 {
1024 TCGMemOp opc = MO_Q | s->be_data;
1025 TCGv addr64 = tcg_temp_new();
1026 tcg_gen_extu_i32_i64(addr64, addr);
1027 tcg_gen_qemu_ld_i64(val, addr64, index, opc);
1028
1029 /* Not needed for user-mode BE32, where we use MO_BE instead. */
1030 if (!IS_USER_ONLY && s->sctlr_b) {
1031 tcg_gen_rotri_i64(val, val, 32);
1032 }
1033 tcg_temp_free(addr64);
1034 }
1035
1036 static inline void gen_aa32_st64(DisasContext *s, TCGv_i64 val,
1037 TCGv_i32 addr, int index)
1038 {
1039 TCGMemOp opc = MO_Q | s->be_data;
1040 TCGv addr64 = tcg_temp_new();
1041 tcg_gen_extu_i32_i64(addr64, addr);
1042
1043 /* Not needed for user-mode BE32, where we use MO_BE instead. */
1044 if (!IS_USER_ONLY && s->sctlr_b) {
1045 TCGv tmp = tcg_temp_new();
1046 tcg_gen_rotri_i64(tmp, val, 32);
1047 tcg_gen_qemu_st_i64(tmp, addr64, index, opc);
1048 tcg_temp_free(tmp);
1049 } else {
1050 tcg_gen_qemu_st_i64(val, addr64, index, opc);
1051 }
1052 tcg_temp_free(addr64);
1053 }
1054
1055 #endif
1056
1057 DO_GEN_LD(8s, MO_SB, 3)
1058 DO_GEN_LD(8u, MO_UB, 3)
1059 DO_GEN_LD(16s, MO_SW, 2)
1060 DO_GEN_LD(16u, MO_UW, 2)
1061 DO_GEN_LD(32u, MO_UL, 0)
1062 /* 'a' variants include an alignment check */
1063 DO_GEN_LD(16ua, MO_UW | MO_ALIGN, 2)
1064 DO_GEN_LD(32ua, MO_UL | MO_ALIGN, 0)
1065 DO_GEN_ST(8, MO_UB, 3)
1066 DO_GEN_ST(16, MO_UW, 2)
1067 DO_GEN_ST(32, MO_UL, 0)
1068
1069 static inline void gen_set_pc_im(DisasContext *s, target_ulong val)
1070 {
1071 tcg_gen_movi_i32(cpu_R[15], val);
1072 }
1073
1074 static inline void gen_hvc(DisasContext *s, int imm16)
1075 {
1076 /* The pre HVC helper handles cases when HVC gets trapped
1077 * as an undefined insn by runtime configuration (ie before
1078 * the insn really executes).
1079 */
1080 gen_set_pc_im(s, s->pc - 4);
1081 gen_helper_pre_hvc(cpu_env);
1082 /* Otherwise we will treat this as a real exception which
1083 * happens after execution of the insn. (The distinction matters
1084 * for the PC value reported to the exception handler and also
1085 * for single stepping.)
1086 */
1087 s->svc_imm = imm16;
1088 gen_set_pc_im(s, s->pc);
1089 s->is_jmp = DISAS_HVC;
1090 }
1091
1092 static inline void gen_smc(DisasContext *s)
1093 {
1094 /* As with HVC, we may take an exception either before or after
1095 * the insn executes.
1096 */
1097 TCGv_i32 tmp;
1098
1099 gen_set_pc_im(s, s->pc - 4);
1100 tmp = tcg_const_i32(syn_aa32_smc());
1101 gen_helper_pre_smc(cpu_env, tmp);
1102 tcg_temp_free_i32(tmp);
1103 gen_set_pc_im(s, s->pc);
1104 s->is_jmp = DISAS_SMC;
1105 }
1106
1107 static inline void
1108 gen_set_condexec (DisasContext *s)
1109 {
1110 if (s->condexec_mask) {
1111 uint32_t val = (s->condexec_cond << 4) | (s->condexec_mask >> 1);
1112 TCGv_i32 tmp = tcg_temp_new_i32();
1113 tcg_gen_movi_i32(tmp, val);
1114 store_cpu_field(tmp, condexec_bits);
1115 }
1116 }
1117
1118 static void gen_exception_internal_insn(DisasContext *s, int offset, int excp)
1119 {
1120 gen_set_condexec(s);
1121 gen_set_pc_im(s, s->pc - offset);
1122 gen_exception_internal(excp);
1123 s->is_jmp = DISAS_JUMP;
1124 }
1125
1126 static void gen_exception_insn(DisasContext *s, int offset, int excp,
1127 int syn, uint32_t target_el)
1128 {
1129 gen_set_condexec(s);
1130 gen_set_pc_im(s, s->pc - offset);
1131 gen_exception(excp, syn, target_el);
1132 s->is_jmp = DISAS_JUMP;
1133 }
1134
1135 /* Force a TB lookup after an instruction that changes the CPU state. */
1136 static inline void gen_lookup_tb(DisasContext *s)
1137 {
1138 tcg_gen_movi_i32(cpu_R[15], s->pc & ~1);
1139 s->is_jmp = DISAS_JUMP;
1140 }
1141
1142 static inline void gen_add_data_offset(DisasContext *s, unsigned int insn,
1143 TCGv_i32 var)
1144 {
1145 int val, rm, shift, shiftop;
1146 TCGv_i32 offset;
1147
1148 if (!(insn & (1 << 25))) {
1149 /* immediate */
1150 val = insn & 0xfff;
1151 if (!(insn & (1 << 23)))
1152 val = -val;
1153 if (val != 0)
1154 tcg_gen_addi_i32(var, var, val);
1155 } else {
1156 /* shift/register */
1157 rm = (insn) & 0xf;
1158 shift = (insn >> 7) & 0x1f;
1159 shiftop = (insn >> 5) & 3;
1160 offset = load_reg(s, rm);
1161 gen_arm_shift_im(offset, shiftop, shift, 0);
1162 if (!(insn & (1 << 23)))
1163 tcg_gen_sub_i32(var, var, offset);
1164 else
1165 tcg_gen_add_i32(var, var, offset);
1166 tcg_temp_free_i32(offset);
1167 }
1168 }
1169
1170 static inline void gen_add_datah_offset(DisasContext *s, unsigned int insn,
1171 int extra, TCGv_i32 var)
1172 {
1173 int val, rm;
1174 TCGv_i32 offset;
1175
1176 if (insn & (1 << 22)) {
1177 /* immediate */
1178 val = (insn & 0xf) | ((insn >> 4) & 0xf0);
1179 if (!(insn & (1 << 23)))
1180 val = -val;
1181 val += extra;
1182 if (val != 0)
1183 tcg_gen_addi_i32(var, var, val);
1184 } else {
1185 /* register */
1186 if (extra)
1187 tcg_gen_addi_i32(var, var, extra);
1188 rm = (insn) & 0xf;
1189 offset = load_reg(s, rm);
1190 if (!(insn & (1 << 23)))
1191 tcg_gen_sub_i32(var, var, offset);
1192 else
1193 tcg_gen_add_i32(var, var, offset);
1194 tcg_temp_free_i32(offset);
1195 }
1196 }
1197
1198 static TCGv_ptr get_fpstatus_ptr(int neon)
1199 {
1200 TCGv_ptr statusptr = tcg_temp_new_ptr();
1201 int offset;
1202 if (neon) {
1203 offset = offsetof(CPUARMState, vfp.standard_fp_status);
1204 } else {
1205 offset = offsetof(CPUARMState, vfp.fp_status);
1206 }
1207 tcg_gen_addi_ptr(statusptr, cpu_env, offset);
1208 return statusptr;
1209 }
1210
1211 #define VFP_OP2(name) \
1212 static inline void gen_vfp_##name(int dp) \
1213 { \
1214 TCGv_ptr fpst = get_fpstatus_ptr(0); \
1215 if (dp) { \
1216 gen_helper_vfp_##name##d(cpu_F0d, cpu_F0d, cpu_F1d, fpst); \
1217 } else { \
1218 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, cpu_F1s, fpst); \
1219 } \
1220 tcg_temp_free_ptr(fpst); \
1221 }
1222
1223 VFP_OP2(add)
1224 VFP_OP2(sub)
1225 VFP_OP2(mul)
1226 VFP_OP2(div)
1227
1228 #undef VFP_OP2
1229
1230 static inline void gen_vfp_F1_mul(int dp)
1231 {
1232 /* Like gen_vfp_mul() but put result in F1 */
1233 TCGv_ptr fpst = get_fpstatus_ptr(0);
1234 if (dp) {
1235 gen_helper_vfp_muld(cpu_F1d, cpu_F0d, cpu_F1d, fpst);
1236 } else {
1237 gen_helper_vfp_muls(cpu_F1s, cpu_F0s, cpu_F1s, fpst);
1238 }
1239 tcg_temp_free_ptr(fpst);
1240 }
1241
1242 static inline void gen_vfp_F1_neg(int dp)
1243 {
1244 /* Like gen_vfp_neg() but put result in F1 */
1245 if (dp) {
1246 gen_helper_vfp_negd(cpu_F1d, cpu_F0d);
1247 } else {
1248 gen_helper_vfp_negs(cpu_F1s, cpu_F0s);
1249 }
1250 }
1251
1252 static inline void gen_vfp_abs(int dp)
1253 {
1254 if (dp)
1255 gen_helper_vfp_absd(cpu_F0d, cpu_F0d);
1256 else
1257 gen_helper_vfp_abss(cpu_F0s, cpu_F0s);
1258 }
1259
1260 static inline void gen_vfp_neg(int dp)
1261 {
1262 if (dp)
1263 gen_helper_vfp_negd(cpu_F0d, cpu_F0d);
1264 else
1265 gen_helper_vfp_negs(cpu_F0s, cpu_F0s);
1266 }
1267
1268 static inline void gen_vfp_sqrt(int dp)
1269 {
1270 if (dp)
1271 gen_helper_vfp_sqrtd(cpu_F0d, cpu_F0d, cpu_env);
1272 else
1273 gen_helper_vfp_sqrts(cpu_F0s, cpu_F0s, cpu_env);
1274 }
1275
1276 static inline void gen_vfp_cmp(int dp)
1277 {
1278 if (dp)
1279 gen_helper_vfp_cmpd(cpu_F0d, cpu_F1d, cpu_env);
1280 else
1281 gen_helper_vfp_cmps(cpu_F0s, cpu_F1s, cpu_env);
1282 }
1283
1284 static inline void gen_vfp_cmpe(int dp)
1285 {
1286 if (dp)
1287 gen_helper_vfp_cmped(cpu_F0d, cpu_F1d, cpu_env);
1288 else
1289 gen_helper_vfp_cmpes(cpu_F0s, cpu_F1s, cpu_env);
1290 }
1291
1292 static inline void gen_vfp_F1_ld0(int dp)
1293 {
1294 if (dp)
1295 tcg_gen_movi_i64(cpu_F1d, 0);
1296 else
1297 tcg_gen_movi_i32(cpu_F1s, 0);
1298 }
1299
1300 #define VFP_GEN_ITOF(name) \
1301 static inline void gen_vfp_##name(int dp, int neon) \
1302 { \
1303 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1304 if (dp) { \
1305 gen_helper_vfp_##name##d(cpu_F0d, cpu_F0s, statusptr); \
1306 } else { \
1307 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, statusptr); \
1308 } \
1309 tcg_temp_free_ptr(statusptr); \
1310 }
1311
1312 VFP_GEN_ITOF(uito)
1313 VFP_GEN_ITOF(sito)
1314 #undef VFP_GEN_ITOF
1315
1316 #define VFP_GEN_FTOI(name) \
1317 static inline void gen_vfp_##name(int dp, int neon) \
1318 { \
1319 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1320 if (dp) { \
1321 gen_helper_vfp_##name##d(cpu_F0s, cpu_F0d, statusptr); \
1322 } else { \
1323 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, statusptr); \
1324 } \
1325 tcg_temp_free_ptr(statusptr); \
1326 }
1327
1328 VFP_GEN_FTOI(toui)
1329 VFP_GEN_FTOI(touiz)
1330 VFP_GEN_FTOI(tosi)
1331 VFP_GEN_FTOI(tosiz)
1332 #undef VFP_GEN_FTOI
1333
1334 #define VFP_GEN_FIX(name, round) \
1335 static inline void gen_vfp_##name(int dp, int shift, int neon) \
1336 { \
1337 TCGv_i32 tmp_shift = tcg_const_i32(shift); \
1338 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1339 if (dp) { \
1340 gen_helper_vfp_##name##d##round(cpu_F0d, cpu_F0d, tmp_shift, \
1341 statusptr); \
1342 } else { \
1343 gen_helper_vfp_##name##s##round(cpu_F0s, cpu_F0s, tmp_shift, \
1344 statusptr); \
1345 } \
1346 tcg_temp_free_i32(tmp_shift); \
1347 tcg_temp_free_ptr(statusptr); \
1348 }
1349 VFP_GEN_FIX(tosh, _round_to_zero)
1350 VFP_GEN_FIX(tosl, _round_to_zero)
1351 VFP_GEN_FIX(touh, _round_to_zero)
1352 VFP_GEN_FIX(toul, _round_to_zero)
1353 VFP_GEN_FIX(shto, )
1354 VFP_GEN_FIX(slto, )
1355 VFP_GEN_FIX(uhto, )
1356 VFP_GEN_FIX(ulto, )
1357 #undef VFP_GEN_FIX
1358
1359 static inline void gen_vfp_ld(DisasContext *s, int dp, TCGv_i32 addr)
1360 {
1361 if (dp) {
1362 gen_aa32_ld64(s, cpu_F0d, addr, get_mem_index(s));
1363 } else {
1364 gen_aa32_ld32u(s, cpu_F0s, addr, get_mem_index(s));
1365 }
1366 }
1367
1368 static inline void gen_vfp_st(DisasContext *s, int dp, TCGv_i32 addr)
1369 {
1370 if (dp) {
1371 gen_aa32_st64(s, cpu_F0d, addr, get_mem_index(s));
1372 } else {
1373 gen_aa32_st32(s, cpu_F0s, addr, get_mem_index(s));
1374 }
1375 }
1376
1377 static inline long
1378 vfp_reg_offset (int dp, int reg)
1379 {
1380 if (dp)
1381 return offsetof(CPUARMState, vfp.regs[reg]);
1382 else if (reg & 1) {
1383 return offsetof(CPUARMState, vfp.regs[reg >> 1])
1384 + offsetof(CPU_DoubleU, l.upper);
1385 } else {
1386 return offsetof(CPUARMState, vfp.regs[reg >> 1])
1387 + offsetof(CPU_DoubleU, l.lower);
1388 }
1389 }
1390
1391 /* Return the offset of a 32-bit piece of a NEON register.
1392 zero is the least significant end of the register. */
1393 static inline long
1394 neon_reg_offset (int reg, int n)
1395 {
1396 int sreg;
1397 sreg = reg * 2 + n;
1398 return vfp_reg_offset(0, sreg);
1399 }
1400
1401 static TCGv_i32 neon_load_reg(int reg, int pass)
1402 {
1403 TCGv_i32 tmp = tcg_temp_new_i32();
1404 tcg_gen_ld_i32(tmp, cpu_env, neon_reg_offset(reg, pass));
1405 return tmp;
1406 }
1407
1408 static void neon_store_reg(int reg, int pass, TCGv_i32 var)
1409 {
1410 tcg_gen_st_i32(var, cpu_env, neon_reg_offset(reg, pass));
1411 tcg_temp_free_i32(var);
1412 }
1413
1414 static inline void neon_load_reg64(TCGv_i64 var, int reg)
1415 {
1416 tcg_gen_ld_i64(var, cpu_env, vfp_reg_offset(1, reg));
1417 }
1418
1419 static inline void neon_store_reg64(TCGv_i64 var, int reg)
1420 {
1421 tcg_gen_st_i64(var, cpu_env, vfp_reg_offset(1, reg));
1422 }
1423
1424 #define tcg_gen_ld_f32 tcg_gen_ld_i32
1425 #define tcg_gen_ld_f64 tcg_gen_ld_i64
1426 #define tcg_gen_st_f32 tcg_gen_st_i32
1427 #define tcg_gen_st_f64 tcg_gen_st_i64
1428
1429 static inline void gen_mov_F0_vreg(int dp, int reg)
1430 {
1431 if (dp)
1432 tcg_gen_ld_f64(cpu_F0d, cpu_env, vfp_reg_offset(dp, reg));
1433 else
1434 tcg_gen_ld_f32(cpu_F0s, cpu_env, vfp_reg_offset(dp, reg));
1435 }
1436
1437 static inline void gen_mov_F1_vreg(int dp, int reg)
1438 {
1439 if (dp)
1440 tcg_gen_ld_f64(cpu_F1d, cpu_env, vfp_reg_offset(dp, reg));
1441 else
1442 tcg_gen_ld_f32(cpu_F1s, cpu_env, vfp_reg_offset(dp, reg));
1443 }
1444
1445 static inline void gen_mov_vreg_F0(int dp, int reg)
1446 {
1447 if (dp)
1448 tcg_gen_st_f64(cpu_F0d, cpu_env, vfp_reg_offset(dp, reg));
1449 else
1450 tcg_gen_st_f32(cpu_F0s, cpu_env, vfp_reg_offset(dp, reg));
1451 }
1452
1453 #define ARM_CP_RW_BIT (1 << 20)
1454
1455 static inline void iwmmxt_load_reg(TCGv_i64 var, int reg)
1456 {
1457 tcg_gen_ld_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1458 }
1459
1460 static inline void iwmmxt_store_reg(TCGv_i64 var, int reg)
1461 {
1462 tcg_gen_st_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1463 }
1464
1465 static inline TCGv_i32 iwmmxt_load_creg(int reg)
1466 {
1467 TCGv_i32 var = tcg_temp_new_i32();
1468 tcg_gen_ld_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1469 return var;
1470 }
1471
1472 static inline void iwmmxt_store_creg(int reg, TCGv_i32 var)
1473 {
1474 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1475 tcg_temp_free_i32(var);
1476 }
1477
1478 static inline void gen_op_iwmmxt_movq_wRn_M0(int rn)
1479 {
1480 iwmmxt_store_reg(cpu_M0, rn);
1481 }
1482
1483 static inline void gen_op_iwmmxt_movq_M0_wRn(int rn)
1484 {
1485 iwmmxt_load_reg(cpu_M0, rn);
1486 }
1487
1488 static inline void gen_op_iwmmxt_orq_M0_wRn(int rn)
1489 {
1490 iwmmxt_load_reg(cpu_V1, rn);
1491 tcg_gen_or_i64(cpu_M0, cpu_M0, cpu_V1);
1492 }
1493
1494 static inline void gen_op_iwmmxt_andq_M0_wRn(int rn)
1495 {
1496 iwmmxt_load_reg(cpu_V1, rn);
1497 tcg_gen_and_i64(cpu_M0, cpu_M0, cpu_V1);
1498 }
1499
1500 static inline void gen_op_iwmmxt_xorq_M0_wRn(int rn)
1501 {
1502 iwmmxt_load_reg(cpu_V1, rn);
1503 tcg_gen_xor_i64(cpu_M0, cpu_M0, cpu_V1);
1504 }
1505
1506 #define IWMMXT_OP(name) \
1507 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1508 { \
1509 iwmmxt_load_reg(cpu_V1, rn); \
1510 gen_helper_iwmmxt_##name(cpu_M0, cpu_M0, cpu_V1); \
1511 }
1512
1513 #define IWMMXT_OP_ENV(name) \
1514 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1515 { \
1516 iwmmxt_load_reg(cpu_V1, rn); \
1517 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0, cpu_V1); \
1518 }
1519
1520 #define IWMMXT_OP_ENV_SIZE(name) \
1521 IWMMXT_OP_ENV(name##b) \
1522 IWMMXT_OP_ENV(name##w) \
1523 IWMMXT_OP_ENV(name##l)
1524
1525 #define IWMMXT_OP_ENV1(name) \
1526 static inline void gen_op_iwmmxt_##name##_M0(void) \
1527 { \
1528 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0); \
1529 }
1530
1531 IWMMXT_OP(maddsq)
1532 IWMMXT_OP(madduq)
1533 IWMMXT_OP(sadb)
1534 IWMMXT_OP(sadw)
1535 IWMMXT_OP(mulslw)
1536 IWMMXT_OP(mulshw)
1537 IWMMXT_OP(mululw)
1538 IWMMXT_OP(muluhw)
1539 IWMMXT_OP(macsw)
1540 IWMMXT_OP(macuw)
1541
1542 IWMMXT_OP_ENV_SIZE(unpackl)
1543 IWMMXT_OP_ENV_SIZE(unpackh)
1544
1545 IWMMXT_OP_ENV1(unpacklub)
1546 IWMMXT_OP_ENV1(unpackluw)
1547 IWMMXT_OP_ENV1(unpacklul)
1548 IWMMXT_OP_ENV1(unpackhub)
1549 IWMMXT_OP_ENV1(unpackhuw)
1550 IWMMXT_OP_ENV1(unpackhul)
1551 IWMMXT_OP_ENV1(unpacklsb)
1552 IWMMXT_OP_ENV1(unpacklsw)
1553 IWMMXT_OP_ENV1(unpacklsl)
1554 IWMMXT_OP_ENV1(unpackhsb)
1555 IWMMXT_OP_ENV1(unpackhsw)
1556 IWMMXT_OP_ENV1(unpackhsl)
1557
1558 IWMMXT_OP_ENV_SIZE(cmpeq)
1559 IWMMXT_OP_ENV_SIZE(cmpgtu)
1560 IWMMXT_OP_ENV_SIZE(cmpgts)
1561
1562 IWMMXT_OP_ENV_SIZE(mins)
1563 IWMMXT_OP_ENV_SIZE(minu)
1564 IWMMXT_OP_ENV_SIZE(maxs)
1565 IWMMXT_OP_ENV_SIZE(maxu)
1566
1567 IWMMXT_OP_ENV_SIZE(subn)
1568 IWMMXT_OP_ENV_SIZE(addn)
1569 IWMMXT_OP_ENV_SIZE(subu)
1570 IWMMXT_OP_ENV_SIZE(addu)
1571 IWMMXT_OP_ENV_SIZE(subs)
1572 IWMMXT_OP_ENV_SIZE(adds)
1573
1574 IWMMXT_OP_ENV(avgb0)
1575 IWMMXT_OP_ENV(avgb1)
1576 IWMMXT_OP_ENV(avgw0)
1577 IWMMXT_OP_ENV(avgw1)
1578
1579 IWMMXT_OP_ENV(packuw)
1580 IWMMXT_OP_ENV(packul)
1581 IWMMXT_OP_ENV(packuq)
1582 IWMMXT_OP_ENV(packsw)
1583 IWMMXT_OP_ENV(packsl)
1584 IWMMXT_OP_ENV(packsq)
1585
1586 static void gen_op_iwmmxt_set_mup(void)
1587 {
1588 TCGv_i32 tmp;
1589 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1590 tcg_gen_ori_i32(tmp, tmp, 2);
1591 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1592 }
1593
1594 static void gen_op_iwmmxt_set_cup(void)
1595 {
1596 TCGv_i32 tmp;
1597 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1598 tcg_gen_ori_i32(tmp, tmp, 1);
1599 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1600 }
1601
1602 static void gen_op_iwmmxt_setpsr_nz(void)
1603 {
1604 TCGv_i32 tmp = tcg_temp_new_i32();
1605 gen_helper_iwmmxt_setpsr_nz(tmp, cpu_M0);
1606 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCASF]);
1607 }
1608
1609 static inline void gen_op_iwmmxt_addl_M0_wRn(int rn)
1610 {
1611 iwmmxt_load_reg(cpu_V1, rn);
1612 tcg_gen_ext32u_i64(cpu_V1, cpu_V1);
1613 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1614 }
1615
1616 static inline int gen_iwmmxt_address(DisasContext *s, uint32_t insn,
1617 TCGv_i32 dest)
1618 {
1619 int rd;
1620 uint32_t offset;
1621 TCGv_i32 tmp;
1622
1623 rd = (insn >> 16) & 0xf;
1624 tmp = load_reg(s, rd);
1625
1626 offset = (insn & 0xff) << ((insn >> 7) & 2);
1627 if (insn & (1 << 24)) {
1628 /* Pre indexed */
1629 if (insn & (1 << 23))
1630 tcg_gen_addi_i32(tmp, tmp, offset);
1631 else
1632 tcg_gen_addi_i32(tmp, tmp, -offset);
1633 tcg_gen_mov_i32(dest, tmp);
1634 if (insn & (1 << 21))
1635 store_reg(s, rd, tmp);
1636 else
1637 tcg_temp_free_i32(tmp);
1638 } else if (insn & (1 << 21)) {
1639 /* Post indexed */
1640 tcg_gen_mov_i32(dest, tmp);
1641 if (insn & (1 << 23))
1642 tcg_gen_addi_i32(tmp, tmp, offset);
1643 else
1644 tcg_gen_addi_i32(tmp, tmp, -offset);
1645 store_reg(s, rd, tmp);
1646 } else if (!(insn & (1 << 23)))
1647 return 1;
1648 return 0;
1649 }
1650
1651 static inline int gen_iwmmxt_shift(uint32_t insn, uint32_t mask, TCGv_i32 dest)
1652 {
1653 int rd = (insn >> 0) & 0xf;
1654 TCGv_i32 tmp;
1655
1656 if (insn & (1 << 8)) {
1657 if (rd < ARM_IWMMXT_wCGR0 || rd > ARM_IWMMXT_wCGR3) {
1658 return 1;
1659 } else {
1660 tmp = iwmmxt_load_creg(rd);
1661 }
1662 } else {
1663 tmp = tcg_temp_new_i32();
1664 iwmmxt_load_reg(cpu_V0, rd);
1665 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
1666 }
1667 tcg_gen_andi_i32(tmp, tmp, mask);
1668 tcg_gen_mov_i32(dest, tmp);
1669 tcg_temp_free_i32(tmp);
1670 return 0;
1671 }
1672
1673 /* Disassemble an iwMMXt instruction. Returns nonzero if an error occurred
1674 (ie. an undefined instruction). */
1675 static int disas_iwmmxt_insn(DisasContext *s, uint32_t insn)
1676 {
1677 int rd, wrd;
1678 int rdhi, rdlo, rd0, rd1, i;
1679 TCGv_i32 addr;
1680 TCGv_i32 tmp, tmp2, tmp3;
1681
1682 if ((insn & 0x0e000e00) == 0x0c000000) {
1683 if ((insn & 0x0fe00ff0) == 0x0c400000) {
1684 wrd = insn & 0xf;
1685 rdlo = (insn >> 12) & 0xf;
1686 rdhi = (insn >> 16) & 0xf;
1687 if (insn & ARM_CP_RW_BIT) { /* TMRRC */
1688 iwmmxt_load_reg(cpu_V0, wrd);
1689 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
1690 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
1691 tcg_gen_extrl_i64_i32(cpu_R[rdhi], cpu_V0);
1692 } else { /* TMCRR */
1693 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
1694 iwmmxt_store_reg(cpu_V0, wrd);
1695 gen_op_iwmmxt_set_mup();
1696 }
1697 return 0;
1698 }
1699
1700 wrd = (insn >> 12) & 0xf;
1701 addr = tcg_temp_new_i32();
1702 if (gen_iwmmxt_address(s, insn, addr)) {
1703 tcg_temp_free_i32(addr);
1704 return 1;
1705 }
1706 if (insn & ARM_CP_RW_BIT) {
1707 if ((insn >> 28) == 0xf) { /* WLDRW wCx */
1708 tmp = tcg_temp_new_i32();
1709 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1710 iwmmxt_store_creg(wrd, tmp);
1711 } else {
1712 i = 1;
1713 if (insn & (1 << 8)) {
1714 if (insn & (1 << 22)) { /* WLDRD */
1715 gen_aa32_ld64(s, cpu_M0, addr, get_mem_index(s));
1716 i = 0;
1717 } else { /* WLDRW wRd */
1718 tmp = tcg_temp_new_i32();
1719 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1720 }
1721 } else {
1722 tmp = tcg_temp_new_i32();
1723 if (insn & (1 << 22)) { /* WLDRH */
1724 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
1725 } else { /* WLDRB */
1726 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
1727 }
1728 }
1729 if (i) {
1730 tcg_gen_extu_i32_i64(cpu_M0, tmp);
1731 tcg_temp_free_i32(tmp);
1732 }
1733 gen_op_iwmmxt_movq_wRn_M0(wrd);
1734 }
1735 } else {
1736 if ((insn >> 28) == 0xf) { /* WSTRW wCx */
1737 tmp = iwmmxt_load_creg(wrd);
1738 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1739 } else {
1740 gen_op_iwmmxt_movq_M0_wRn(wrd);
1741 tmp = tcg_temp_new_i32();
1742 if (insn & (1 << 8)) {
1743 if (insn & (1 << 22)) { /* WSTRD */
1744 gen_aa32_st64(s, cpu_M0, addr, get_mem_index(s));
1745 } else { /* WSTRW wRd */
1746 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1747 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1748 }
1749 } else {
1750 if (insn & (1 << 22)) { /* WSTRH */
1751 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1752 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
1753 } else { /* WSTRB */
1754 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1755 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
1756 }
1757 }
1758 }
1759 tcg_temp_free_i32(tmp);
1760 }
1761 tcg_temp_free_i32(addr);
1762 return 0;
1763 }
1764
1765 if ((insn & 0x0f000000) != 0x0e000000)
1766 return 1;
1767
1768 switch (((insn >> 12) & 0xf00) | ((insn >> 4) & 0xff)) {
1769 case 0x000: /* WOR */
1770 wrd = (insn >> 12) & 0xf;
1771 rd0 = (insn >> 0) & 0xf;
1772 rd1 = (insn >> 16) & 0xf;
1773 gen_op_iwmmxt_movq_M0_wRn(rd0);
1774 gen_op_iwmmxt_orq_M0_wRn(rd1);
1775 gen_op_iwmmxt_setpsr_nz();
1776 gen_op_iwmmxt_movq_wRn_M0(wrd);
1777 gen_op_iwmmxt_set_mup();
1778 gen_op_iwmmxt_set_cup();
1779 break;
1780 case 0x011: /* TMCR */
1781 if (insn & 0xf)
1782 return 1;
1783 rd = (insn >> 12) & 0xf;
1784 wrd = (insn >> 16) & 0xf;
1785 switch (wrd) {
1786 case ARM_IWMMXT_wCID:
1787 case ARM_IWMMXT_wCASF:
1788 break;
1789 case ARM_IWMMXT_wCon:
1790 gen_op_iwmmxt_set_cup();
1791 /* Fall through. */
1792 case ARM_IWMMXT_wCSSF:
1793 tmp = iwmmxt_load_creg(wrd);
1794 tmp2 = load_reg(s, rd);
1795 tcg_gen_andc_i32(tmp, tmp, tmp2);
1796 tcg_temp_free_i32(tmp2);
1797 iwmmxt_store_creg(wrd, tmp);
1798 break;
1799 case ARM_IWMMXT_wCGR0:
1800 case ARM_IWMMXT_wCGR1:
1801 case ARM_IWMMXT_wCGR2:
1802 case ARM_IWMMXT_wCGR3:
1803 gen_op_iwmmxt_set_cup();
1804 tmp = load_reg(s, rd);
1805 iwmmxt_store_creg(wrd, tmp);
1806 break;
1807 default:
1808 return 1;
1809 }
1810 break;
1811 case 0x100: /* WXOR */
1812 wrd = (insn >> 12) & 0xf;
1813 rd0 = (insn >> 0) & 0xf;
1814 rd1 = (insn >> 16) & 0xf;
1815 gen_op_iwmmxt_movq_M0_wRn(rd0);
1816 gen_op_iwmmxt_xorq_M0_wRn(rd1);
1817 gen_op_iwmmxt_setpsr_nz();
1818 gen_op_iwmmxt_movq_wRn_M0(wrd);
1819 gen_op_iwmmxt_set_mup();
1820 gen_op_iwmmxt_set_cup();
1821 break;
1822 case 0x111: /* TMRC */
1823 if (insn & 0xf)
1824 return 1;
1825 rd = (insn >> 12) & 0xf;
1826 wrd = (insn >> 16) & 0xf;
1827 tmp = iwmmxt_load_creg(wrd);
1828 store_reg(s, rd, tmp);
1829 break;
1830 case 0x300: /* WANDN */
1831 wrd = (insn >> 12) & 0xf;
1832 rd0 = (insn >> 0) & 0xf;
1833 rd1 = (insn >> 16) & 0xf;
1834 gen_op_iwmmxt_movq_M0_wRn(rd0);
1835 tcg_gen_neg_i64(cpu_M0, cpu_M0);
1836 gen_op_iwmmxt_andq_M0_wRn(rd1);
1837 gen_op_iwmmxt_setpsr_nz();
1838 gen_op_iwmmxt_movq_wRn_M0(wrd);
1839 gen_op_iwmmxt_set_mup();
1840 gen_op_iwmmxt_set_cup();
1841 break;
1842 case 0x200: /* WAND */
1843 wrd = (insn >> 12) & 0xf;
1844 rd0 = (insn >> 0) & 0xf;
1845 rd1 = (insn >> 16) & 0xf;
1846 gen_op_iwmmxt_movq_M0_wRn(rd0);
1847 gen_op_iwmmxt_andq_M0_wRn(rd1);
1848 gen_op_iwmmxt_setpsr_nz();
1849 gen_op_iwmmxt_movq_wRn_M0(wrd);
1850 gen_op_iwmmxt_set_mup();
1851 gen_op_iwmmxt_set_cup();
1852 break;
1853 case 0x810: case 0xa10: /* WMADD */
1854 wrd = (insn >> 12) & 0xf;
1855 rd0 = (insn >> 0) & 0xf;
1856 rd1 = (insn >> 16) & 0xf;
1857 gen_op_iwmmxt_movq_M0_wRn(rd0);
1858 if (insn & (1 << 21))
1859 gen_op_iwmmxt_maddsq_M0_wRn(rd1);
1860 else
1861 gen_op_iwmmxt_madduq_M0_wRn(rd1);
1862 gen_op_iwmmxt_movq_wRn_M0(wrd);
1863 gen_op_iwmmxt_set_mup();
1864 break;
1865 case 0x10e: case 0x50e: case 0x90e: case 0xd0e: /* WUNPCKIL */
1866 wrd = (insn >> 12) & 0xf;
1867 rd0 = (insn >> 16) & 0xf;
1868 rd1 = (insn >> 0) & 0xf;
1869 gen_op_iwmmxt_movq_M0_wRn(rd0);
1870 switch ((insn >> 22) & 3) {
1871 case 0:
1872 gen_op_iwmmxt_unpacklb_M0_wRn(rd1);
1873 break;
1874 case 1:
1875 gen_op_iwmmxt_unpacklw_M0_wRn(rd1);
1876 break;
1877 case 2:
1878 gen_op_iwmmxt_unpackll_M0_wRn(rd1);
1879 break;
1880 case 3:
1881 return 1;
1882 }
1883 gen_op_iwmmxt_movq_wRn_M0(wrd);
1884 gen_op_iwmmxt_set_mup();
1885 gen_op_iwmmxt_set_cup();
1886 break;
1887 case 0x10c: case 0x50c: case 0x90c: case 0xd0c: /* WUNPCKIH */
1888 wrd = (insn >> 12) & 0xf;
1889 rd0 = (insn >> 16) & 0xf;
1890 rd1 = (insn >> 0) & 0xf;
1891 gen_op_iwmmxt_movq_M0_wRn(rd0);
1892 switch ((insn >> 22) & 3) {
1893 case 0:
1894 gen_op_iwmmxt_unpackhb_M0_wRn(rd1);
1895 break;
1896 case 1:
1897 gen_op_iwmmxt_unpackhw_M0_wRn(rd1);
1898 break;
1899 case 2:
1900 gen_op_iwmmxt_unpackhl_M0_wRn(rd1);
1901 break;
1902 case 3:
1903 return 1;
1904 }
1905 gen_op_iwmmxt_movq_wRn_M0(wrd);
1906 gen_op_iwmmxt_set_mup();
1907 gen_op_iwmmxt_set_cup();
1908 break;
1909 case 0x012: case 0x112: case 0x412: case 0x512: /* WSAD */
1910 wrd = (insn >> 12) & 0xf;
1911 rd0 = (insn >> 16) & 0xf;
1912 rd1 = (insn >> 0) & 0xf;
1913 gen_op_iwmmxt_movq_M0_wRn(rd0);
1914 if (insn & (1 << 22))
1915 gen_op_iwmmxt_sadw_M0_wRn(rd1);
1916 else
1917 gen_op_iwmmxt_sadb_M0_wRn(rd1);
1918 if (!(insn & (1 << 20)))
1919 gen_op_iwmmxt_addl_M0_wRn(wrd);
1920 gen_op_iwmmxt_movq_wRn_M0(wrd);
1921 gen_op_iwmmxt_set_mup();
1922 break;
1923 case 0x010: case 0x110: case 0x210: case 0x310: /* WMUL */
1924 wrd = (insn >> 12) & 0xf;
1925 rd0 = (insn >> 16) & 0xf;
1926 rd1 = (insn >> 0) & 0xf;
1927 gen_op_iwmmxt_movq_M0_wRn(rd0);
1928 if (insn & (1 << 21)) {
1929 if (insn & (1 << 20))
1930 gen_op_iwmmxt_mulshw_M0_wRn(rd1);
1931 else
1932 gen_op_iwmmxt_mulslw_M0_wRn(rd1);
1933 } else {
1934 if (insn & (1 << 20))
1935 gen_op_iwmmxt_muluhw_M0_wRn(rd1);
1936 else
1937 gen_op_iwmmxt_mululw_M0_wRn(rd1);
1938 }
1939 gen_op_iwmmxt_movq_wRn_M0(wrd);
1940 gen_op_iwmmxt_set_mup();
1941 break;
1942 case 0x410: case 0x510: case 0x610: case 0x710: /* WMAC */
1943 wrd = (insn >> 12) & 0xf;
1944 rd0 = (insn >> 16) & 0xf;
1945 rd1 = (insn >> 0) & 0xf;
1946 gen_op_iwmmxt_movq_M0_wRn(rd0);
1947 if (insn & (1 << 21))
1948 gen_op_iwmmxt_macsw_M0_wRn(rd1);
1949 else
1950 gen_op_iwmmxt_macuw_M0_wRn(rd1);
1951 if (!(insn & (1 << 20))) {
1952 iwmmxt_load_reg(cpu_V1, wrd);
1953 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1954 }
1955 gen_op_iwmmxt_movq_wRn_M0(wrd);
1956 gen_op_iwmmxt_set_mup();
1957 break;
1958 case 0x006: case 0x406: case 0x806: case 0xc06: /* WCMPEQ */
1959 wrd = (insn >> 12) & 0xf;
1960 rd0 = (insn >> 16) & 0xf;
1961 rd1 = (insn >> 0) & 0xf;
1962 gen_op_iwmmxt_movq_M0_wRn(rd0);
1963 switch ((insn >> 22) & 3) {
1964 case 0:
1965 gen_op_iwmmxt_cmpeqb_M0_wRn(rd1);
1966 break;
1967 case 1:
1968 gen_op_iwmmxt_cmpeqw_M0_wRn(rd1);
1969 break;
1970 case 2:
1971 gen_op_iwmmxt_cmpeql_M0_wRn(rd1);
1972 break;
1973 case 3:
1974 return 1;
1975 }
1976 gen_op_iwmmxt_movq_wRn_M0(wrd);
1977 gen_op_iwmmxt_set_mup();
1978 gen_op_iwmmxt_set_cup();
1979 break;
1980 case 0x800: case 0x900: case 0xc00: case 0xd00: /* WAVG2 */
1981 wrd = (insn >> 12) & 0xf;
1982 rd0 = (insn >> 16) & 0xf;
1983 rd1 = (insn >> 0) & 0xf;
1984 gen_op_iwmmxt_movq_M0_wRn(rd0);
1985 if (insn & (1 << 22)) {
1986 if (insn & (1 << 20))
1987 gen_op_iwmmxt_avgw1_M0_wRn(rd1);
1988 else
1989 gen_op_iwmmxt_avgw0_M0_wRn(rd1);
1990 } else {
1991 if (insn & (1 << 20))
1992 gen_op_iwmmxt_avgb1_M0_wRn(rd1);
1993 else
1994 gen_op_iwmmxt_avgb0_M0_wRn(rd1);
1995 }
1996 gen_op_iwmmxt_movq_wRn_M0(wrd);
1997 gen_op_iwmmxt_set_mup();
1998 gen_op_iwmmxt_set_cup();
1999 break;
2000 case 0x802: case 0x902: case 0xa02: case 0xb02: /* WALIGNR */
2001 wrd = (insn >> 12) & 0xf;
2002 rd0 = (insn >> 16) & 0xf;
2003 rd1 = (insn >> 0) & 0xf;
2004 gen_op_iwmmxt_movq_M0_wRn(rd0);
2005 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCGR0 + ((insn >> 20) & 3));
2006 tcg_gen_andi_i32(tmp, tmp, 7);
2007 iwmmxt_load_reg(cpu_V1, rd1);
2008 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
2009 tcg_temp_free_i32(tmp);
2010 gen_op_iwmmxt_movq_wRn_M0(wrd);
2011 gen_op_iwmmxt_set_mup();
2012 break;
2013 case 0x601: case 0x605: case 0x609: case 0x60d: /* TINSR */
2014 if (((insn >> 6) & 3) == 3)
2015 return 1;
2016 rd = (insn >> 12) & 0xf;
2017 wrd = (insn >> 16) & 0xf;
2018 tmp = load_reg(s, rd);
2019 gen_op_iwmmxt_movq_M0_wRn(wrd);
2020 switch ((insn >> 6) & 3) {
2021 case 0:
2022 tmp2 = tcg_const_i32(0xff);
2023 tmp3 = tcg_const_i32((insn & 7) << 3);
2024 break;
2025 case 1:
2026 tmp2 = tcg_const_i32(0xffff);
2027 tmp3 = tcg_const_i32((insn & 3) << 4);
2028 break;
2029 case 2:
2030 tmp2 = tcg_const_i32(0xffffffff);
2031 tmp3 = tcg_const_i32((insn & 1) << 5);
2032 break;
2033 default:
2034 TCGV_UNUSED_I32(tmp2);
2035 TCGV_UNUSED_I32(tmp3);
2036 }
2037 gen_helper_iwmmxt_insr(cpu_M0, cpu_M0, tmp, tmp2, tmp3);
2038 tcg_temp_free_i32(tmp3);
2039 tcg_temp_free_i32(tmp2);
2040 tcg_temp_free_i32(tmp);
2041 gen_op_iwmmxt_movq_wRn_M0(wrd);
2042 gen_op_iwmmxt_set_mup();
2043 break;
2044 case 0x107: case 0x507: case 0x907: case 0xd07: /* TEXTRM */
2045 rd = (insn >> 12) & 0xf;
2046 wrd = (insn >> 16) & 0xf;
2047 if (rd == 15 || ((insn >> 22) & 3) == 3)
2048 return 1;
2049 gen_op_iwmmxt_movq_M0_wRn(wrd);
2050 tmp = tcg_temp_new_i32();
2051 switch ((insn >> 22) & 3) {
2052 case 0:
2053 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 7) << 3);
2054 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
2055 if (insn & 8) {
2056 tcg_gen_ext8s_i32(tmp, tmp);
2057 } else {
2058 tcg_gen_andi_i32(tmp, tmp, 0xff);
2059 }
2060 break;
2061 case 1:
2062 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 3) << 4);
2063 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
2064 if (insn & 8) {
2065 tcg_gen_ext16s_i32(tmp, tmp);
2066 } else {
2067 tcg_gen_andi_i32(tmp, tmp, 0xffff);
2068 }
2069 break;
2070 case 2:
2071 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 1) << 5);
2072 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
2073 break;
2074 }
2075 store_reg(s, rd, tmp);
2076 break;
2077 case 0x117: case 0x517: case 0x917: case 0xd17: /* TEXTRC */
2078 if ((insn & 0x000ff008) != 0x0003f000 || ((insn >> 22) & 3) == 3)
2079 return 1;
2080 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
2081 switch ((insn >> 22) & 3) {
2082 case 0:
2083 tcg_gen_shri_i32(tmp, tmp, ((insn & 7) << 2) + 0);
2084 break;
2085 case 1:
2086 tcg_gen_shri_i32(tmp, tmp, ((insn & 3) << 3) + 4);
2087 break;
2088 case 2:
2089 tcg_gen_shri_i32(tmp, tmp, ((insn & 1) << 4) + 12);
2090 break;
2091 }
2092 tcg_gen_shli_i32(tmp, tmp, 28);
2093 gen_set_nzcv(tmp);
2094 tcg_temp_free_i32(tmp);
2095 break;
2096 case 0x401: case 0x405: case 0x409: case 0x40d: /* TBCST */
2097 if (((insn >> 6) & 3) == 3)
2098 return 1;
2099 rd = (insn >> 12) & 0xf;
2100 wrd = (insn >> 16) & 0xf;
2101 tmp = load_reg(s, rd);
2102 switch ((insn >> 6) & 3) {
2103 case 0:
2104 gen_helper_iwmmxt_bcstb(cpu_M0, tmp);
2105 break;
2106 case 1:
2107 gen_helper_iwmmxt_bcstw(cpu_M0, tmp);
2108 break;
2109 case 2:
2110 gen_helper_iwmmxt_bcstl(cpu_M0, tmp);
2111 break;
2112 }
2113 tcg_temp_free_i32(tmp);
2114 gen_op_iwmmxt_movq_wRn_M0(wrd);
2115 gen_op_iwmmxt_set_mup();
2116 break;
2117 case 0x113: case 0x513: case 0x913: case 0xd13: /* TANDC */
2118 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
2119 return 1;
2120 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
2121 tmp2 = tcg_temp_new_i32();
2122 tcg_gen_mov_i32(tmp2, tmp);
2123 switch ((insn >> 22) & 3) {
2124 case 0:
2125 for (i = 0; i < 7; i ++) {
2126 tcg_gen_shli_i32(tmp2, tmp2, 4);
2127 tcg_gen_and_i32(tmp, tmp, tmp2);
2128 }
2129 break;
2130 case 1:
2131 for (i = 0; i < 3; i ++) {
2132 tcg_gen_shli_i32(tmp2, tmp2, 8);
2133 tcg_gen_and_i32(tmp, tmp, tmp2);
2134 }
2135 break;
2136 case 2:
2137 tcg_gen_shli_i32(tmp2, tmp2, 16);
2138 tcg_gen_and_i32(tmp, tmp, tmp2);
2139 break;
2140 }
2141 gen_set_nzcv(tmp);
2142 tcg_temp_free_i32(tmp2);
2143 tcg_temp_free_i32(tmp);
2144 break;
2145 case 0x01c: case 0x41c: case 0x81c: case 0xc1c: /* WACC */
2146 wrd = (insn >> 12) & 0xf;
2147 rd0 = (insn >> 16) & 0xf;
2148 gen_op_iwmmxt_movq_M0_wRn(rd0);
2149 switch ((insn >> 22) & 3) {
2150 case 0:
2151 gen_helper_iwmmxt_addcb(cpu_M0, cpu_M0);
2152 break;
2153 case 1:
2154 gen_helper_iwmmxt_addcw(cpu_M0, cpu_M0);
2155 break;
2156 case 2:
2157 gen_helper_iwmmxt_addcl(cpu_M0, cpu_M0);
2158 break;
2159 case 3:
2160 return 1;
2161 }
2162 gen_op_iwmmxt_movq_wRn_M0(wrd);
2163 gen_op_iwmmxt_set_mup();
2164 break;
2165 case 0x115: case 0x515: case 0x915: case 0xd15: /* TORC */
2166 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
2167 return 1;
2168 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
2169 tmp2 = tcg_temp_new_i32();
2170 tcg_gen_mov_i32(tmp2, tmp);
2171 switch ((insn >> 22) & 3) {
2172 case 0:
2173 for (i = 0; i < 7; i ++) {
2174 tcg_gen_shli_i32(tmp2, tmp2, 4);
2175 tcg_gen_or_i32(tmp, tmp, tmp2);
2176 }
2177 break;
2178 case 1:
2179 for (i = 0; i < 3; i ++) {
2180 tcg_gen_shli_i32(tmp2, tmp2, 8);
2181 tcg_gen_or_i32(tmp, tmp, tmp2);
2182 }
2183 break;
2184 case 2:
2185 tcg_gen_shli_i32(tmp2, tmp2, 16);
2186 tcg_gen_or_i32(tmp, tmp, tmp2);
2187 break;
2188 }
2189 gen_set_nzcv(tmp);
2190 tcg_temp_free_i32(tmp2);
2191 tcg_temp_free_i32(tmp);
2192 break;
2193 case 0x103: case 0x503: case 0x903: case 0xd03: /* TMOVMSK */
2194 rd = (insn >> 12) & 0xf;
2195 rd0 = (insn >> 16) & 0xf;
2196 if ((insn & 0xf) != 0 || ((insn >> 22) & 3) == 3)
2197 return 1;
2198 gen_op_iwmmxt_movq_M0_wRn(rd0);
2199 tmp = tcg_temp_new_i32();
2200 switch ((insn >> 22) & 3) {
2201 case 0:
2202 gen_helper_iwmmxt_msbb(tmp, cpu_M0);
2203 break;
2204 case 1:
2205 gen_helper_iwmmxt_msbw(tmp, cpu_M0);
2206 break;
2207 case 2:
2208 gen_helper_iwmmxt_msbl(tmp, cpu_M0);
2209 break;
2210 }
2211 store_reg(s, rd, tmp);
2212 break;
2213 case 0x106: case 0x306: case 0x506: case 0x706: /* WCMPGT */
2214 case 0x906: case 0xb06: case 0xd06: case 0xf06:
2215 wrd = (insn >> 12) & 0xf;
2216 rd0 = (insn >> 16) & 0xf;
2217 rd1 = (insn >> 0) & 0xf;
2218 gen_op_iwmmxt_movq_M0_wRn(rd0);
2219 switch ((insn >> 22) & 3) {
2220 case 0:
2221 if (insn & (1 << 21))
2222 gen_op_iwmmxt_cmpgtsb_M0_wRn(rd1);
2223 else
2224 gen_op_iwmmxt_cmpgtub_M0_wRn(rd1);
2225 break;
2226 case 1:
2227 if (insn & (1 << 21))
2228 gen_op_iwmmxt_cmpgtsw_M0_wRn(rd1);
2229 else
2230 gen_op_iwmmxt_cmpgtuw_M0_wRn(rd1);
2231 break;
2232 case 2:
2233 if (insn & (1 << 21))
2234 gen_op_iwmmxt_cmpgtsl_M0_wRn(rd1);
2235 else
2236 gen_op_iwmmxt_cmpgtul_M0_wRn(rd1);
2237 break;
2238 case 3:
2239 return 1;
2240 }
2241 gen_op_iwmmxt_movq_wRn_M0(wrd);
2242 gen_op_iwmmxt_set_mup();
2243 gen_op_iwmmxt_set_cup();
2244 break;
2245 case 0x00e: case 0x20e: case 0x40e: case 0x60e: /* WUNPCKEL */
2246 case 0x80e: case 0xa0e: case 0xc0e: case 0xe0e:
2247 wrd = (insn >> 12) & 0xf;
2248 rd0 = (insn >> 16) & 0xf;
2249 gen_op_iwmmxt_movq_M0_wRn(rd0);
2250 switch ((insn >> 22) & 3) {
2251 case 0:
2252 if (insn & (1 << 21))
2253 gen_op_iwmmxt_unpacklsb_M0();
2254 else
2255 gen_op_iwmmxt_unpacklub_M0();
2256 break;
2257 case 1:
2258 if (insn & (1 << 21))
2259 gen_op_iwmmxt_unpacklsw_M0();
2260 else
2261 gen_op_iwmmxt_unpackluw_M0();
2262 break;
2263 case 2:
2264 if (insn & (1 << 21))
2265 gen_op_iwmmxt_unpacklsl_M0();
2266 else
2267 gen_op_iwmmxt_unpacklul_M0();
2268 break;
2269 case 3:
2270 return 1;
2271 }
2272 gen_op_iwmmxt_movq_wRn_M0(wrd);
2273 gen_op_iwmmxt_set_mup();
2274 gen_op_iwmmxt_set_cup();
2275 break;
2276 case 0x00c: case 0x20c: case 0x40c: case 0x60c: /* WUNPCKEH */
2277 case 0x80c: case 0xa0c: case 0xc0c: case 0xe0c:
2278 wrd = (insn >> 12) & 0xf;
2279 rd0 = (insn >> 16) & 0xf;
2280 gen_op_iwmmxt_movq_M0_wRn(rd0);
2281 switch ((insn >> 22) & 3) {
2282 case 0:
2283 if (insn & (1 << 21))
2284 gen_op_iwmmxt_unpackhsb_M0();
2285 else
2286 gen_op_iwmmxt_unpackhub_M0();
2287 break;
2288 case 1:
2289 if (insn & (1 << 21))
2290 gen_op_iwmmxt_unpackhsw_M0();
2291 else
2292 gen_op_iwmmxt_unpackhuw_M0();
2293 break;
2294 case 2:
2295 if (insn & (1 << 21))
2296 gen_op_iwmmxt_unpackhsl_M0();
2297 else
2298 gen_op_iwmmxt_unpackhul_M0();
2299 break;
2300 case 3:
2301 return 1;
2302 }
2303 gen_op_iwmmxt_movq_wRn_M0(wrd);
2304 gen_op_iwmmxt_set_mup();
2305 gen_op_iwmmxt_set_cup();
2306 break;
2307 case 0x204: case 0x604: case 0xa04: case 0xe04: /* WSRL */
2308 case 0x214: case 0x614: case 0xa14: case 0xe14:
2309 if (((insn >> 22) & 3) == 0)
2310 return 1;
2311 wrd = (insn >> 12) & 0xf;
2312 rd0 = (insn >> 16) & 0xf;
2313 gen_op_iwmmxt_movq_M0_wRn(rd0);
2314 tmp = tcg_temp_new_i32();
2315 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2316 tcg_temp_free_i32(tmp);
2317 return 1;
2318 }
2319 switch ((insn >> 22) & 3) {
2320 case 1:
2321 gen_helper_iwmmxt_srlw(cpu_M0, cpu_env, cpu_M0, tmp);
2322 break;
2323 case 2:
2324 gen_helper_iwmmxt_srll(cpu_M0, cpu_env, cpu_M0, tmp);
2325 break;
2326 case 3:
2327 gen_helper_iwmmxt_srlq(cpu_M0, cpu_env, cpu_M0, tmp);
2328 break;
2329 }
2330 tcg_temp_free_i32(tmp);
2331 gen_op_iwmmxt_movq_wRn_M0(wrd);
2332 gen_op_iwmmxt_set_mup();
2333 gen_op_iwmmxt_set_cup();
2334 break;
2335 case 0x004: case 0x404: case 0x804: case 0xc04: /* WSRA */
2336 case 0x014: case 0x414: case 0x814: case 0xc14:
2337 if (((insn >> 22) & 3) == 0)
2338 return 1;
2339 wrd = (insn >> 12) & 0xf;
2340 rd0 = (insn >> 16) & 0xf;
2341 gen_op_iwmmxt_movq_M0_wRn(rd0);
2342 tmp = tcg_temp_new_i32();
2343 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2344 tcg_temp_free_i32(tmp);
2345 return 1;
2346 }
2347 switch ((insn >> 22) & 3) {
2348 case 1:
2349 gen_helper_iwmmxt_sraw(cpu_M0, cpu_env, cpu_M0, tmp);
2350 break;
2351 case 2:
2352 gen_helper_iwmmxt_sral(cpu_M0, cpu_env, cpu_M0, tmp);
2353 break;
2354 case 3:
2355 gen_helper_iwmmxt_sraq(cpu_M0, cpu_env, cpu_M0, tmp);
2356 break;
2357 }
2358 tcg_temp_free_i32(tmp);
2359 gen_op_iwmmxt_movq_wRn_M0(wrd);
2360 gen_op_iwmmxt_set_mup();
2361 gen_op_iwmmxt_set_cup();
2362 break;
2363 case 0x104: case 0x504: case 0x904: case 0xd04: /* WSLL */
2364 case 0x114: case 0x514: case 0x914: case 0xd14:
2365 if (((insn >> 22) & 3) == 0)
2366 return 1;
2367 wrd = (insn >> 12) & 0xf;
2368 rd0 = (insn >> 16) & 0xf;
2369 gen_op_iwmmxt_movq_M0_wRn(rd0);
2370 tmp = tcg_temp_new_i32();
2371 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2372 tcg_temp_free_i32(tmp);
2373 return 1;
2374 }
2375 switch ((insn >> 22) & 3) {
2376 case 1:
2377 gen_helper_iwmmxt_sllw(cpu_M0, cpu_env, cpu_M0, tmp);
2378 break;
2379 case 2:
2380 gen_helper_iwmmxt_slll(cpu_M0, cpu_env, cpu_M0, tmp);
2381 break;
2382 case 3:
2383 gen_helper_iwmmxt_sllq(cpu_M0, cpu_env, cpu_M0, tmp);
2384 break;
2385 }
2386 tcg_temp_free_i32(tmp);
2387 gen_op_iwmmxt_movq_wRn_M0(wrd);
2388 gen_op_iwmmxt_set_mup();
2389 gen_op_iwmmxt_set_cup();
2390 break;
2391 case 0x304: case 0x704: case 0xb04: case 0xf04: /* WROR */
2392 case 0x314: case 0x714: case 0xb14: case 0xf14:
2393 if (((insn >> 22) & 3) == 0)
2394 return 1;
2395 wrd = (insn >> 12) & 0xf;
2396 rd0 = (insn >> 16) & 0xf;
2397 gen_op_iwmmxt_movq_M0_wRn(rd0);
2398 tmp = tcg_temp_new_i32();
2399 switch ((insn >> 22) & 3) {
2400 case 1:
2401 if (gen_iwmmxt_shift(insn, 0xf, tmp)) {
2402 tcg_temp_free_i32(tmp);
2403 return 1;
2404 }
2405 gen_helper_iwmmxt_rorw(cpu_M0, cpu_env, cpu_M0, tmp);
2406 break;
2407 case 2:
2408 if (gen_iwmmxt_shift(insn, 0x1f, tmp)) {
2409 tcg_temp_free_i32(tmp);
2410 return 1;
2411 }
2412 gen_helper_iwmmxt_rorl(cpu_M0, cpu_env, cpu_M0, tmp);
2413 break;
2414 case 3:
2415 if (gen_iwmmxt_shift(insn, 0x3f, tmp)) {
2416 tcg_temp_free_i32(tmp);
2417 return 1;
2418 }
2419 gen_helper_iwmmxt_rorq(cpu_M0, cpu_env, cpu_M0, tmp);
2420 break;
2421 }
2422 tcg_temp_free_i32(tmp);
2423 gen_op_iwmmxt_movq_wRn_M0(wrd);
2424 gen_op_iwmmxt_set_mup();
2425 gen_op_iwmmxt_set_cup();
2426 break;
2427 case 0x116: case 0x316: case 0x516: case 0x716: /* WMIN */
2428 case 0x916: case 0xb16: case 0xd16: case 0xf16:
2429 wrd = (insn >> 12) & 0xf;
2430 rd0 = (insn >> 16) & 0xf;
2431 rd1 = (insn >> 0) & 0xf;
2432 gen_op_iwmmxt_movq_M0_wRn(rd0);
2433 switch ((insn >> 22) & 3) {
2434 case 0:
2435 if (insn & (1 << 21))
2436 gen_op_iwmmxt_minsb_M0_wRn(rd1);
2437 else
2438 gen_op_iwmmxt_minub_M0_wRn(rd1);
2439 break;
2440 case 1:
2441 if (insn & (1 << 21))
2442 gen_op_iwmmxt_minsw_M0_wRn(rd1);
2443 else
2444 gen_op_iwmmxt_minuw_M0_wRn(rd1);
2445 break;
2446 case 2:
2447 if (insn & (1 << 21))
2448 gen_op_iwmmxt_minsl_M0_wRn(rd1);
2449 else
2450 gen_op_iwmmxt_minul_M0_wRn(rd1);
2451 break;
2452 case 3:
2453 return 1;
2454 }
2455 gen_op_iwmmxt_movq_wRn_M0(wrd);
2456 gen_op_iwmmxt_set_mup();
2457 break;
2458 case 0x016: case 0x216: case 0x416: case 0x616: /* WMAX */
2459 case 0x816: case 0xa16: case 0xc16: case 0xe16:
2460 wrd = (insn >> 12) & 0xf;
2461 rd0 = (insn >> 16) & 0xf;
2462 rd1 = (insn >> 0) & 0xf;
2463 gen_op_iwmmxt_movq_M0_wRn(rd0);
2464 switch ((insn >> 22) & 3) {
2465 case 0:
2466 if (insn & (1 << 21))
2467 gen_op_iwmmxt_maxsb_M0_wRn(rd1);
2468 else
2469 gen_op_iwmmxt_maxub_M0_wRn(rd1);
2470 break;
2471 case 1:
2472 if (insn & (1 << 21))
2473 gen_op_iwmmxt_maxsw_M0_wRn(rd1);
2474 else
2475 gen_op_iwmmxt_maxuw_M0_wRn(rd1);
2476 break;
2477 case 2:
2478 if (insn & (1 << 21))
2479 gen_op_iwmmxt_maxsl_M0_wRn(rd1);
2480 else
2481 gen_op_iwmmxt_maxul_M0_wRn(rd1);
2482 break;
2483 case 3:
2484 return 1;
2485 }
2486 gen_op_iwmmxt_movq_wRn_M0(wrd);
2487 gen_op_iwmmxt_set_mup();
2488 break;
2489 case 0x002: case 0x102: case 0x202: case 0x302: /* WALIGNI */
2490 case 0x402: case 0x502: case 0x602: case 0x702:
2491 wrd = (insn >> 12) & 0xf;
2492 rd0 = (insn >> 16) & 0xf;
2493 rd1 = (insn >> 0) & 0xf;
2494 gen_op_iwmmxt_movq_M0_wRn(rd0);
2495 tmp = tcg_const_i32((insn >> 20) & 3);
2496 iwmmxt_load_reg(cpu_V1, rd1);
2497 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
2498 tcg_temp_free_i32(tmp);
2499 gen_op_iwmmxt_movq_wRn_M0(wrd);
2500 gen_op_iwmmxt_set_mup();
2501 break;
2502 case 0x01a: case 0x11a: case 0x21a: case 0x31a: /* WSUB */
2503 case 0x41a: case 0x51a: case 0x61a: case 0x71a:
2504 case 0x81a: case 0x91a: case 0xa1a: case 0xb1a:
2505 case 0xc1a: case 0xd1a: case 0xe1a: case 0xf1a:
2506 wrd = (insn >> 12) & 0xf;
2507 rd0 = (insn >> 16) & 0xf;
2508 rd1 = (insn >> 0) & 0xf;
2509 gen_op_iwmmxt_movq_M0_wRn(rd0);
2510 switch ((insn >> 20) & 0xf) {
2511 case 0x0:
2512 gen_op_iwmmxt_subnb_M0_wRn(rd1);
2513 break;
2514 case 0x1:
2515 gen_op_iwmmxt_subub_M0_wRn(rd1);
2516 break;
2517 case 0x3:
2518 gen_op_iwmmxt_subsb_M0_wRn(rd1);
2519 break;
2520 case 0x4:
2521 gen_op_iwmmxt_subnw_M0_wRn(rd1);
2522 break;
2523 case 0x5:
2524 gen_op_iwmmxt_subuw_M0_wRn(rd1);
2525 break;
2526 case 0x7:
2527 gen_op_iwmmxt_subsw_M0_wRn(rd1);
2528 break;
2529 case 0x8:
2530 gen_op_iwmmxt_subnl_M0_wRn(rd1);
2531 break;
2532 case 0x9:
2533 gen_op_iwmmxt_subul_M0_wRn(rd1);
2534 break;
2535 case 0xb:
2536 gen_op_iwmmxt_subsl_M0_wRn(rd1);
2537 break;
2538 default:
2539 return 1;
2540 }
2541 gen_op_iwmmxt_movq_wRn_M0(wrd);
2542 gen_op_iwmmxt_set_mup();
2543 gen_op_iwmmxt_set_cup();
2544 break;
2545 case 0x01e: case 0x11e: case 0x21e: case 0x31e: /* WSHUFH */
2546 case 0x41e: case 0x51e: case 0x61e: case 0x71e:
2547 case 0x81e: case 0x91e: case 0xa1e: case 0xb1e:
2548 case 0xc1e: case 0xd1e: case 0xe1e: case 0xf1e:
2549 wrd = (insn >> 12) & 0xf;
2550 rd0 = (insn >> 16) & 0xf;
2551 gen_op_iwmmxt_movq_M0_wRn(rd0);
2552 tmp = tcg_const_i32(((insn >> 16) & 0xf0) | (insn & 0x0f));
2553 gen_helper_iwmmxt_shufh(cpu_M0, cpu_env, cpu_M0, tmp);
2554 tcg_temp_free_i32(tmp);
2555 gen_op_iwmmxt_movq_wRn_M0(wrd);
2556 gen_op_iwmmxt_set_mup();
2557 gen_op_iwmmxt_set_cup();
2558 break;
2559 case 0x018: case 0x118: case 0x218: case 0x318: /* WADD */
2560 case 0x418: case 0x518: case 0x618: case 0x718:
2561 case 0x818: case 0x918: case 0xa18: case 0xb18:
2562 case 0xc18: case 0xd18: case 0xe18: case 0xf18:
2563 wrd = (insn >> 12) & 0xf;
2564 rd0 = (insn >> 16) & 0xf;
2565 rd1 = (insn >> 0) & 0xf;
2566 gen_op_iwmmxt_movq_M0_wRn(rd0);
2567 switch ((insn >> 20) & 0xf) {
2568 case 0x0:
2569 gen_op_iwmmxt_addnb_M0_wRn(rd1);
2570 break;
2571 case 0x1:
2572 gen_op_iwmmxt_addub_M0_wRn(rd1);
2573 break;
2574 case 0x3:
2575 gen_op_iwmmxt_addsb_M0_wRn(rd1);
2576 break;
2577 case 0x4:
2578 gen_op_iwmmxt_addnw_M0_wRn(rd1);
2579 break;
2580 case 0x5:
2581 gen_op_iwmmxt_adduw_M0_wRn(rd1);
2582 break;
2583 case 0x7:
2584 gen_op_iwmmxt_addsw_M0_wRn(rd1);
2585 break;
2586 case 0x8:
2587 gen_op_iwmmxt_addnl_M0_wRn(rd1);
2588 break;
2589 case 0x9:
2590 gen_op_iwmmxt_addul_M0_wRn(rd1);
2591 break;
2592 case 0xb:
2593 gen_op_iwmmxt_addsl_M0_wRn(rd1);
2594 break;
2595 default:
2596 return 1;
2597 }
2598 gen_op_iwmmxt_movq_wRn_M0(wrd);
2599 gen_op_iwmmxt_set_mup();
2600 gen_op_iwmmxt_set_cup();
2601 break;
2602 case 0x008: case 0x108: case 0x208: case 0x308: /* WPACK */
2603 case 0x408: case 0x508: case 0x608: case 0x708:
2604 case 0x808: case 0x908: case 0xa08: case 0xb08:
2605 case 0xc08: case 0xd08: case 0xe08: case 0xf08:
2606 if (!(insn & (1 << 20)) || ((insn >> 22) & 3) == 0)
2607 return 1;
2608 wrd = (insn >> 12) & 0xf;
2609 rd0 = (insn >> 16) & 0xf;
2610 rd1 = (insn >> 0) & 0xf;
2611 gen_op_iwmmxt_movq_M0_wRn(rd0);
2612 switch ((insn >> 22) & 3) {
2613 case 1:
2614 if (insn & (1 << 21))
2615 gen_op_iwmmxt_packsw_M0_wRn(rd1);
2616 else
2617 gen_op_iwmmxt_packuw_M0_wRn(rd1);
2618 break;
2619 case 2:
2620 if (insn & (1 << 21))
2621 gen_op_iwmmxt_packsl_M0_wRn(rd1);
2622 else
2623 gen_op_iwmmxt_packul_M0_wRn(rd1);
2624 break;
2625 case 3:
2626 if (insn & (1 << 21))
2627 gen_op_iwmmxt_packsq_M0_wRn(rd1);
2628 else
2629 gen_op_iwmmxt_packuq_M0_wRn(rd1);
2630 break;
2631 }
2632 gen_op_iwmmxt_movq_wRn_M0(wrd);
2633 gen_op_iwmmxt_set_mup();
2634 gen_op_iwmmxt_set_cup();
2635 break;
2636 case 0x201: case 0x203: case 0x205: case 0x207:
2637 case 0x209: case 0x20b: case 0x20d: case 0x20f:
2638 case 0x211: case 0x213: case 0x215: case 0x217:
2639 case 0x219: case 0x21b: case 0x21d: case 0x21f:
2640 wrd = (insn >> 5) & 0xf;
2641 rd0 = (insn >> 12) & 0xf;
2642 rd1 = (insn >> 0) & 0xf;
2643 if (rd0 == 0xf || rd1 == 0xf)
2644 return 1;
2645 gen_op_iwmmxt_movq_M0_wRn(wrd);
2646 tmp = load_reg(s, rd0);
2647 tmp2 = load_reg(s, rd1);
2648 switch ((insn >> 16) & 0xf) {
2649 case 0x0: /* TMIA */
2650 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2651 break;
2652 case 0x8: /* TMIAPH */
2653 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2654 break;
2655 case 0xc: case 0xd: case 0xe: case 0xf: /* TMIAxy */
2656 if (insn & (1 << 16))
2657 tcg_gen_shri_i32(tmp, tmp, 16);
2658 if (insn & (1 << 17))
2659 tcg_gen_shri_i32(tmp2, tmp2, 16);
2660 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2661 break;
2662 default:
2663 tcg_temp_free_i32(tmp2);
2664 tcg_temp_free_i32(tmp);
2665 return 1;
2666 }
2667 tcg_temp_free_i32(tmp2);
2668 tcg_temp_free_i32(tmp);
2669 gen_op_iwmmxt_movq_wRn_M0(wrd);
2670 gen_op_iwmmxt_set_mup();
2671 break;
2672 default:
2673 return 1;
2674 }
2675
2676 return 0;
2677 }
2678
2679 /* Disassemble an XScale DSP instruction. Returns nonzero if an error occurred
2680 (ie. an undefined instruction). */
2681 static int disas_dsp_insn(DisasContext *s, uint32_t insn)
2682 {
2683 int acc, rd0, rd1, rdhi, rdlo;
2684 TCGv_i32 tmp, tmp2;
2685
2686 if ((insn & 0x0ff00f10) == 0x0e200010) {
2687 /* Multiply with Internal Accumulate Format */
2688 rd0 = (insn >> 12) & 0xf;
2689 rd1 = insn & 0xf;
2690 acc = (insn >> 5) & 7;
2691
2692 if (acc != 0)
2693 return 1;
2694
2695 tmp = load_reg(s, rd0);
2696 tmp2 = load_reg(s, rd1);
2697 switch ((insn >> 16) & 0xf) {
2698 case 0x0: /* MIA */
2699 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2700 break;
2701 case 0x8: /* MIAPH */
2702 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2703 break;
2704 case 0xc: /* MIABB */
2705 case 0xd: /* MIABT */
2706 case 0xe: /* MIATB */
2707 case 0xf: /* MIATT */
2708 if (insn & (1 << 16))
2709 tcg_gen_shri_i32(tmp, tmp, 16);
2710 if (insn & (1 << 17))
2711 tcg_gen_shri_i32(tmp2, tmp2, 16);
2712 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2713 break;
2714 default:
2715 return 1;
2716 }
2717 tcg_temp_free_i32(tmp2);
2718 tcg_temp_free_i32(tmp);
2719
2720 gen_op_iwmmxt_movq_wRn_M0(acc);
2721 return 0;
2722 }
2723
2724 if ((insn & 0x0fe00ff8) == 0x0c400000) {
2725 /* Internal Accumulator Access Format */
2726 rdhi = (insn >> 16) & 0xf;
2727 rdlo = (insn >> 12) & 0xf;
2728 acc = insn & 7;
2729
2730 if (acc != 0)
2731 return 1;
2732
2733 if (insn & ARM_CP_RW_BIT) { /* MRA */
2734 iwmmxt_load_reg(cpu_V0, acc);
2735 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
2736 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
2737 tcg_gen_extrl_i64_i32(cpu_R[rdhi], cpu_V0);
2738 tcg_gen_andi_i32(cpu_R[rdhi], cpu_R[rdhi], (1 << (40 - 32)) - 1);
2739 } else { /* MAR */
2740 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
2741 iwmmxt_store_reg(cpu_V0, acc);
2742 }
2743 return 0;
2744 }
2745
2746 return 1;
2747 }
2748
2749 #define VFP_REG_SHR(x, n) (((n) > 0) ? (x) >> (n) : (x) << -(n))
2750 #define VFP_SREG(insn, bigbit, smallbit) \
2751 ((VFP_REG_SHR(insn, bigbit - 1) & 0x1e) | (((insn) >> (smallbit)) & 1))
2752 #define VFP_DREG(reg, insn, bigbit, smallbit) do { \
2753 if (arm_dc_feature(s, ARM_FEATURE_VFP3)) { \
2754 reg = (((insn) >> (bigbit)) & 0x0f) \
2755 | (((insn) >> ((smallbit) - 4)) & 0x10); \
2756 } else { \
2757 if (insn & (1 << (smallbit))) \
2758 return 1; \
2759 reg = ((insn) >> (bigbit)) & 0x0f; \
2760 }} while (0)
2761
2762 #define VFP_SREG_D(insn) VFP_SREG(insn, 12, 22)
2763 #define VFP_DREG_D(reg, insn) VFP_DREG(reg, insn, 12, 22)
2764 #define VFP_SREG_N(insn) VFP_SREG(insn, 16, 7)
2765 #define VFP_DREG_N(reg, insn) VFP_DREG(reg, insn, 16, 7)
2766 #define VFP_SREG_M(insn) VFP_SREG(insn, 0, 5)
2767 #define VFP_DREG_M(reg, insn) VFP_DREG(reg, insn, 0, 5)
2768
2769 /* Move between integer and VFP cores. */
2770 static TCGv_i32 gen_vfp_mrs(void)
2771 {
2772 TCGv_i32 tmp = tcg_temp_new_i32();
2773 tcg_gen_mov_i32(tmp, cpu_F0s);
2774 return tmp;
2775 }
2776
2777 static void gen_vfp_msr(TCGv_i32 tmp)
2778 {
2779 tcg_gen_mov_i32(cpu_F0s, tmp);
2780 tcg_temp_free_i32(tmp);
2781 }
2782
2783 static void gen_neon_dup_u8(TCGv_i32 var, int shift)
2784 {
2785 TCGv_i32 tmp = tcg_temp_new_i32();
2786 if (shift)
2787 tcg_gen_shri_i32(var, var, shift);
2788 tcg_gen_ext8u_i32(var, var);
2789 tcg_gen_shli_i32(tmp, var, 8);
2790 tcg_gen_or_i32(var, var, tmp);
2791 tcg_gen_shli_i32(tmp, var, 16);
2792 tcg_gen_or_i32(var, var, tmp);
2793 tcg_temp_free_i32(tmp);
2794 }
2795
2796 static void gen_neon_dup_low16(TCGv_i32 var)
2797 {
2798 TCGv_i32 tmp = tcg_temp_new_i32();
2799 tcg_gen_ext16u_i32(var, var);
2800 tcg_gen_shli_i32(tmp, var, 16);
2801 tcg_gen_or_i32(var, var, tmp);
2802 tcg_temp_free_i32(tmp);
2803 }
2804
2805 static void gen_neon_dup_high16(TCGv_i32 var)
2806 {
2807 TCGv_i32 tmp = tcg_temp_new_i32();
2808 tcg_gen_andi_i32(var, var, 0xffff0000);
2809 tcg_gen_shri_i32(tmp, var, 16);
2810 tcg_gen_or_i32(var, var, tmp);
2811 tcg_temp_free_i32(tmp);
2812 }
2813
2814 static TCGv_i32 gen_load_and_replicate(DisasContext *s, TCGv_i32 addr, int size)
2815 {
2816 /* Load a single Neon element and replicate into a 32 bit TCG reg */
2817 TCGv_i32 tmp = tcg_temp_new_i32();
2818 switch (size) {
2819 case 0:
2820 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
2821 gen_neon_dup_u8(tmp, 0);
2822 break;
2823 case 1:
2824 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
2825 gen_neon_dup_low16(tmp);
2826 break;
2827 case 2:
2828 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
2829 break;
2830 default: /* Avoid compiler warnings. */
2831 abort();
2832 }
2833 return tmp;
2834 }
2835
2836 static int handle_vsel(uint32_t insn, uint32_t rd, uint32_t rn, uint32_t rm,
2837 uint32_t dp)
2838 {
2839 uint32_t cc = extract32(insn, 20, 2);
2840
2841 if (dp) {
2842 TCGv_i64 frn, frm, dest;
2843 TCGv_i64 tmp, zero, zf, nf, vf;
2844
2845 zero = tcg_const_i64(0);
2846
2847 frn = tcg_temp_new_i64();
2848 frm = tcg_temp_new_i64();
2849 dest = tcg_temp_new_i64();
2850
2851 zf = tcg_temp_new_i64();
2852 nf = tcg_temp_new_i64();
2853 vf = tcg_temp_new_i64();
2854
2855 tcg_gen_extu_i32_i64(zf, cpu_ZF);
2856 tcg_gen_ext_i32_i64(nf, cpu_NF);
2857 tcg_gen_ext_i32_i64(vf, cpu_VF);
2858
2859 tcg_gen_ld_f64(frn, cpu_env, vfp_reg_offset(dp, rn));
2860 tcg_gen_ld_f64(frm, cpu_env, vfp_reg_offset(dp, rm));
2861 switch (cc) {
2862 case 0: /* eq: Z */
2863 tcg_gen_movcond_i64(TCG_COND_EQ, dest, zf, zero,
2864 frn, frm);
2865 break;
2866 case 1: /* vs: V */
2867 tcg_gen_movcond_i64(TCG_COND_LT, dest, vf, zero,
2868 frn, frm);
2869 break;
2870 case 2: /* ge: N == V -> N ^ V == 0 */
2871 tmp = tcg_temp_new_i64();
2872 tcg_gen_xor_i64(tmp, vf, nf);
2873 tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero,
2874 frn, frm);
2875 tcg_temp_free_i64(tmp);
2876 break;
2877 case 3: /* gt: !Z && N == V */
2878 tcg_gen_movcond_i64(TCG_COND_NE, dest, zf, zero,
2879 frn, frm);
2880 tmp = tcg_temp_new_i64();
2881 tcg_gen_xor_i64(tmp, vf, nf);
2882 tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero,
2883 dest, frm);
2884 tcg_temp_free_i64(tmp);
2885 break;
2886 }
2887 tcg_gen_st_f64(dest, cpu_env, vfp_reg_offset(dp, rd));
2888 tcg_temp_free_i64(frn);
2889 tcg_temp_free_i64(frm);
2890 tcg_temp_free_i64(dest);
2891
2892 tcg_temp_free_i64(zf);
2893 tcg_temp_free_i64(nf);
2894 tcg_temp_free_i64(vf);
2895
2896 tcg_temp_free_i64(zero);
2897 } else {
2898 TCGv_i32 frn, frm, dest;
2899 TCGv_i32 tmp, zero;
2900
2901 zero = tcg_const_i32(0);
2902
2903 frn = tcg_temp_new_i32();
2904 frm = tcg_temp_new_i32();
2905 dest = tcg_temp_new_i32();
2906 tcg_gen_ld_f32(frn, cpu_env, vfp_reg_offset(dp, rn));
2907 tcg_gen_ld_f32(frm, cpu_env, vfp_reg_offset(dp, rm));
2908 switch (cc) {
2909 case 0: /* eq: Z */
2910 tcg_gen_movcond_i32(TCG_COND_EQ, dest, cpu_ZF, zero,
2911 frn, frm);
2912 break;
2913 case 1: /* vs: V */
2914 tcg_gen_movcond_i32(TCG_COND_LT, dest, cpu_VF, zero,
2915 frn, frm);
2916 break;
2917 case 2: /* ge: N == V -> N ^ V == 0 */
2918 tmp = tcg_temp_new_i32();
2919 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
2920 tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero,
2921 frn, frm);
2922 tcg_temp_free_i32(tmp);
2923 break;
2924 case 3: /* gt: !Z && N == V */
2925 tcg_gen_movcond_i32(TCG_COND_NE, dest, cpu_ZF, zero,
2926 frn, frm);
2927 tmp = tcg_temp_new_i32();
2928 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
2929 tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero,
2930 dest, frm);
2931 tcg_temp_free_i32(tmp);
2932 break;
2933 }
2934 tcg_gen_st_f32(dest, cpu_env, vfp_reg_offset(dp, rd));
2935 tcg_temp_free_i32(frn);
2936 tcg_temp_free_i32(frm);
2937 tcg_temp_free_i32(dest);
2938
2939 tcg_temp_free_i32(zero);
2940 }
2941
2942 return 0;
2943 }
2944
2945 static int handle_vminmaxnm(uint32_t insn, uint32_t rd, uint32_t rn,
2946 uint32_t rm, uint32_t dp)
2947 {
2948 uint32_t vmin = extract32(insn, 6, 1);
2949 TCGv_ptr fpst = get_fpstatus_ptr(0);
2950
2951 if (dp) {
2952 TCGv_i64 frn, frm, dest;
2953
2954 frn = tcg_temp_new_i64();
2955 frm = tcg_temp_new_i64();
2956 dest = tcg_temp_new_i64();
2957
2958 tcg_gen_ld_f64(frn, cpu_env, vfp_reg_offset(dp, rn));
2959 tcg_gen_ld_f64(frm, cpu_env, vfp_reg_offset(dp, rm));
2960 if (vmin) {
2961 gen_helper_vfp_minnumd(dest, frn, frm, fpst);
2962 } else {
2963 gen_helper_vfp_maxnumd(dest, frn, frm, fpst);
2964 }
2965 tcg_gen_st_f64(dest, cpu_env, vfp_reg_offset(dp, rd));
2966 tcg_temp_free_i64(frn);
2967 tcg_temp_free_i64(frm);
2968 tcg_temp_free_i64(dest);
2969 } else {
2970 TCGv_i32 frn, frm, dest;
2971
2972 frn = tcg_temp_new_i32();
2973 frm = tcg_temp_new_i32();
2974 dest = tcg_temp_new_i32();
2975
2976 tcg_gen_ld_f32(frn, cpu_env, vfp_reg_offset(dp, rn));
2977 tcg_gen_ld_f32(frm, cpu_env, vfp_reg_offset(dp, rm));
2978 if (vmin) {
2979 gen_helper_vfp_minnums(dest, frn, frm, fpst);
2980 } else {
2981 gen_helper_vfp_maxnums(dest, frn, frm, fpst);
2982 }
2983 tcg_gen_st_f32(dest, cpu_env, vfp_reg_offset(dp, rd));
2984 tcg_temp_free_i32(frn);
2985 tcg_temp_free_i32(frm);
2986 tcg_temp_free_i32(dest);
2987 }
2988
2989 tcg_temp_free_ptr(fpst);
2990 return 0;
2991 }
2992
2993 static int handle_vrint(uint32_t insn, uint32_t rd, uint32_t rm, uint32_t dp,
2994 int rounding)
2995 {
2996 TCGv_ptr fpst = get_fpstatus_ptr(0);
2997 TCGv_i32 tcg_rmode;
2998
2999 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding));
3000 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
3001
3002 if (dp) {
3003 TCGv_i64 tcg_op;
3004 TCGv_i64 tcg_res;
3005 tcg_op = tcg_temp_new_i64();
3006 tcg_res = tcg_temp_new_i64();
3007 tcg_gen_ld_f64(tcg_op, cpu_env, vfp_reg_offset(dp, rm));
3008 gen_helper_rintd(tcg_res, tcg_op, fpst);
3009 tcg_gen_st_f64(tcg_res, cpu_env, vfp_reg_offset(dp, rd));
3010 tcg_temp_free_i64(tcg_op);
3011 tcg_temp_free_i64(tcg_res);
3012 } else {
3013 TCGv_i32 tcg_op;
3014 TCGv_i32 tcg_res;
3015 tcg_op = tcg_temp_new_i32();
3016 tcg_res = tcg_temp_new_i32();
3017 tcg_gen_ld_f32(tcg_op, cpu_env, vfp_reg_offset(dp, rm));
3018 gen_helper_rints(tcg_res, tcg_op, fpst);
3019 tcg_gen_st_f32(tcg_res, cpu_env, vfp_reg_offset(dp, rd));
3020 tcg_temp_free_i32(tcg_op);
3021 tcg_temp_free_i32(tcg_res);
3022 }
3023
3024 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
3025 tcg_temp_free_i32(tcg_rmode);
3026
3027 tcg_temp_free_ptr(fpst);
3028 return 0;
3029 }
3030
3031 static int handle_vcvt(uint32_t insn, uint32_t rd, uint32_t rm, uint32_t dp,
3032 int rounding)
3033 {
3034 bool is_signed = extract32(insn, 7, 1);
3035 TCGv_ptr fpst = get_fpstatus_ptr(0);
3036 TCGv_i32 tcg_rmode, tcg_shift;
3037
3038 tcg_shift = tcg_const_i32(0);
3039
3040 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding));
3041 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
3042
3043 if (dp) {
3044 TCGv_i64 tcg_double, tcg_res;
3045 TCGv_i32 tcg_tmp;
3046 /* Rd is encoded as a single precision register even when the source
3047 * is double precision.
3048 */
3049 rd = ((rd << 1) & 0x1e) | ((rd >> 4) & 0x1);
3050 tcg_double = tcg_temp_new_i64();
3051 tcg_res = tcg_temp_new_i64();
3052 tcg_tmp = tcg_temp_new_i32();
3053 tcg_gen_ld_f64(tcg_double, cpu_env, vfp_reg_offset(1, rm));
3054 if (is_signed) {
3055 gen_helper_vfp_tosld(tcg_res, tcg_double, tcg_shift, fpst);
3056 } else {
3057 gen_helper_vfp_tould(tcg_res, tcg_double, tcg_shift, fpst);
3058 }
3059 tcg_gen_extrl_i64_i32(tcg_tmp, tcg_res);
3060 tcg_gen_st_f32(tcg_tmp, cpu_env, vfp_reg_offset(0, rd));
3061 tcg_temp_free_i32(tcg_tmp);
3062 tcg_temp_free_i64(tcg_res);
3063 tcg_temp_free_i64(tcg_double);
3064 } else {
3065 TCGv_i32 tcg_single, tcg_res;
3066 tcg_single = tcg_temp_new_i32();
3067 tcg_res = tcg_temp_new_i32();
3068 tcg_gen_ld_f32(tcg_single, cpu_env, vfp_reg_offset(0, rm));
3069 if (is_signed) {
3070 gen_helper_vfp_tosls(tcg_res, tcg_single, tcg_shift, fpst);
3071 } else {
3072 gen_helper_vfp_touls(tcg_res, tcg_single, tcg_shift, fpst);
3073 }
3074 tcg_gen_st_f32(tcg_res, cpu_env, vfp_reg_offset(0, rd));
3075 tcg_temp_free_i32(tcg_res);
3076 tcg_temp_free_i32(tcg_single);
3077 }
3078
3079 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
3080 tcg_temp_free_i32(tcg_rmode);
3081
3082 tcg_temp_free_i32(tcg_shift);
3083
3084 tcg_temp_free_ptr(fpst);
3085
3086 return 0;
3087 }
3088
3089 /* Table for converting the most common AArch32 encoding of
3090 * rounding mode to arm_fprounding order (which matches the
3091 * common AArch64 order); see ARM ARM pseudocode FPDecodeRM().
3092 */
3093 static const uint8_t fp_decode_rm[] = {
3094 FPROUNDING_TIEAWAY,
3095 FPROUNDING_TIEEVEN,
3096 FPROUNDING_POSINF,
3097 FPROUNDING_NEGINF,
3098 };
3099
3100 static int disas_vfp_v8_insn(DisasContext *s, uint32_t insn)
3101 {
3102 uint32_t rd, rn, rm, dp = extract32(insn, 8, 1);
3103
3104 if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
3105 return 1;
3106 }
3107
3108 if (dp) {
3109 VFP_DREG_D(rd, insn);
3110 VFP_DREG_N(rn, insn);
3111 VFP_DREG_M(rm, insn);
3112 } else {
3113 rd = VFP_SREG_D(insn);
3114 rn = VFP_SREG_N(insn);
3115 rm = VFP_SREG_M(insn);
3116 }
3117
3118 if ((insn & 0x0f800e50) == 0x0e000a00) {
3119 return handle_vsel(insn, rd, rn, rm, dp);
3120 } else if ((insn & 0x0fb00e10) == 0x0e800a00) {
3121 return handle_vminmaxnm(insn, rd, rn, rm, dp);
3122 } else if ((insn & 0x0fbc0ed0) == 0x0eb80a40) {
3123 /* VRINTA, VRINTN, VRINTP, VRINTM */
3124 int rounding = fp_decode_rm[extract32(insn, 16, 2)];
3125 return handle_vrint(insn, rd, rm, dp, rounding);
3126 } else if ((insn & 0x0fbc0e50) == 0x0ebc0a40) {
3127 /* VCVTA, VCVTN, VCVTP, VCVTM */
3128 int rounding = fp_decode_rm[extract32(insn, 16, 2)];
3129 return handle_vcvt(insn, rd, rm, dp, rounding);
3130 }
3131 return 1;
3132 }
3133
3134 /* Disassemble a VFP instruction. Returns nonzero if an error occurred
3135 (ie. an undefined instruction). */
3136 static int disas_vfp_insn(DisasContext *s, uint32_t insn)
3137 {
3138 uint32_t rd, rn, rm, op, i, n, offset, delta_d, delta_m, bank_mask;
3139 int dp, veclen;
3140 TCGv_i32 addr;
3141 TCGv_i32 tmp;
3142 TCGv_i32 tmp2;
3143
3144 if (!arm_dc_feature(s, ARM_FEATURE_VFP)) {
3145 return 1;
3146 }
3147
3148 /* FIXME: this access check should not take precedence over UNDEF
3149 * for invalid encodings; we will generate incorrect syndrome information
3150 * for attempts to execute invalid vfp/neon encodings with FP disabled.
3151 */
3152 if (s->fp_excp_el) {
3153 gen_exception_insn(s, 4, EXCP_UDEF,
3154 syn_fp_access_trap(1, 0xe, false), s->fp_excp_el);
3155 return 0;
3156 }
3157
3158 if (!s->vfp_enabled) {
3159 /* VFP disabled. Only allow fmxr/fmrx to/from some control regs. */
3160 if ((insn & 0x0fe00fff) != 0x0ee00a10)
3161 return 1;
3162 rn = (insn >> 16) & 0xf;
3163 if (rn != ARM_VFP_FPSID && rn != ARM_VFP_FPEXC && rn != ARM_VFP_MVFR2
3164 && rn != ARM_VFP_MVFR1 && rn != ARM_VFP_MVFR0) {
3165 return 1;
3166 }
3167 }
3168
3169 if (extract32(insn, 28, 4) == 0xf) {
3170 /* Encodings with T=1 (Thumb) or unconditional (ARM):
3171 * only used in v8 and above.
3172 */
3173 return disas_vfp_v8_insn(s, insn);
3174 }
3175
3176 dp = ((insn & 0xf00) == 0xb00);
3177 switch ((insn >> 24) & 0xf) {
3178 case 0xe:
3179 if (insn & (1 << 4)) {
3180 /* single register transfer */
3181 rd = (insn >> 12) & 0xf;
3182 if (dp) {
3183 int size;
3184 int pass;
3185
3186 VFP_DREG_N(rn, insn);
3187 if (insn & 0xf)
3188 return 1;
3189 if (insn & 0x00c00060
3190 && !arm_dc_feature(s, ARM_FEATURE_NEON)) {
3191 return 1;
3192 }
3193
3194 pass = (insn >> 21) & 1;
3195 if (insn & (1 << 22)) {
3196 size = 0;
3197 offset = ((insn >> 5) & 3) * 8;
3198 } else if (insn & (1 << 5)) {
3199 size = 1;
3200 offset = (insn & (1 << 6)) ? 16 : 0;
3201 } else {
3202 size = 2;
3203 offset = 0;
3204 }
3205 if (insn & ARM_CP_RW_BIT) {
3206 /* vfp->arm */
3207 tmp = neon_load_reg(rn, pass);
3208 switch (size) {
3209 case 0:
3210 if (offset)
3211 tcg_gen_shri_i32(tmp, tmp, offset);
3212 if (insn & (1 << 23))
3213 gen_uxtb(tmp);
3214 else
3215 gen_sxtb(tmp);
3216 break;
3217 case 1:
3218 if (insn & (1 << 23)) {
3219 if (offset) {
3220 tcg_gen_shri_i32(tmp, tmp, 16);
3221 } else {
3222 gen_uxth(tmp);
3223 }
3224 } else {
3225 if (offset) {
3226 tcg_gen_sari_i32(tmp, tmp, 16);
3227 } else {
3228 gen_sxth(tmp);
3229 }
3230 }
3231 break;
3232 case 2:
3233 break;
3234 }
3235 store_reg(s, rd, tmp);
3236 } else {
3237 /* arm->vfp */
3238 tmp = load_reg(s, rd);
3239 if (insn & (1 << 23)) {
3240 /* VDUP */
3241 if (size == 0) {
3242 gen_neon_dup_u8(tmp, 0);
3243 } else if (size == 1) {
3244 gen_neon_dup_low16(tmp);
3245 }
3246 for (n = 0; n <= pass * 2; n++) {
3247 tmp2 = tcg_temp_new_i32();
3248 tcg_gen_mov_i32(tmp2, tmp);
3249 neon_store_reg(rn, n, tmp2);
3250 }
3251 neon_store_reg(rn, n, tmp);
3252 } else {
3253 /* VMOV */
3254 switch (size) {
3255 case 0:
3256 tmp2 = neon_load_reg(rn, pass);
3257 tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 8);
3258 tcg_temp_free_i32(tmp2);
3259 break;
3260 case 1:
3261 tmp2 = neon_load_reg(rn, pass);
3262 tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 16);
3263 tcg_temp_free_i32(tmp2);
3264 break;
3265 case 2:
3266 break;
3267 }
3268 neon_store_reg(rn, pass, tmp);
3269 }
3270 }
3271 } else { /* !dp */
3272 if ((insn & 0x6f) != 0x00)
3273 return 1;
3274 rn = VFP_SREG_N(insn);
3275 if (insn & ARM_CP_RW_BIT) {
3276 /* vfp->arm */
3277 if (insn & (1 << 21)) {
3278 /* system register */
3279 rn >>= 1;
3280
3281 switch (rn) {
3282 case ARM_VFP_FPSID:
3283 /* VFP2 allows access to FSID from userspace.
3284 VFP3 restricts all id registers to privileged
3285 accesses. */
3286 if (IS_USER(s)
3287 && arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3288 return 1;
3289 }
3290 tmp = load_cpu_field(vfp.xregs[rn]);
3291 break;
3292 case ARM_VFP_FPEXC:
3293 if (IS_USER(s))
3294 return 1;
3295 tmp = load_cpu_field(vfp.xregs[rn]);
3296 break;
3297 case ARM_VFP_FPINST:
3298 case ARM_VFP_FPINST2:
3299 /* Not present in VFP3. */
3300 if (IS_USER(s)
3301 || arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3302 return 1;
3303 }
3304 tmp = load_cpu_field(vfp.xregs[rn]);
3305 break;
3306 case ARM_VFP_FPSCR:
3307 if (rd == 15) {
3308 tmp = load_cpu_field(vfp.xregs[ARM_VFP_FPSCR]);
3309 tcg_gen_andi_i32(tmp, tmp, 0xf0000000);
3310 } else {
3311 tmp = tcg_temp_new_i32();
3312 gen_helper_vfp_get_fpscr(tmp, cpu_env);
3313 }
3314 break;
3315 case ARM_VFP_MVFR2:
3316 if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
3317 return 1;
3318 }
3319 /* fall through */
3320 case ARM_VFP_MVFR0:
3321 case ARM_VFP_MVFR1:
3322 if (IS_USER(s)
3323 || !arm_dc_feature(s, ARM_FEATURE_MVFR)) {
3324 return 1;
3325 }
3326 tmp = load_cpu_field(vfp.xregs[rn]);
3327 break;
3328 default:
3329 return 1;
3330 }
3331 } else {
3332 gen_mov_F0_vreg(0, rn);
3333 tmp = gen_vfp_mrs();
3334 }
3335 if (rd == 15) {
3336 /* Set the 4 flag bits in the CPSR. */
3337 gen_set_nzcv(tmp);
3338 tcg_temp_free_i32(tmp);
3339 } else {
3340 store_reg(s, rd, tmp);
3341 }
3342 } else {
3343 /* arm->vfp */
3344 if (insn & (1 << 21)) {
3345 rn >>= 1;
3346 /* system register */
3347 switch (rn) {
3348 case ARM_VFP_FPSID:
3349 case ARM_VFP_MVFR0:
3350 case ARM_VFP_MVFR1:
3351 /* Writes are ignored. */
3352 break;
3353 case ARM_VFP_FPSCR:
3354 tmp = load_reg(s, rd);
3355 gen_helper_vfp_set_fpscr(cpu_env, tmp);
3356 tcg_temp_free_i32(tmp);
3357 gen_lookup_tb(s);
3358 break;
3359 case ARM_VFP_FPEXC:
3360 if (IS_USER(s))
3361 return 1;
3362 /* TODO: VFP subarchitecture support.
3363 * For now, keep the EN bit only */
3364 tmp = load_reg(s, rd);
3365 tcg_gen_andi_i32(tmp, tmp, 1 << 30);
3366 store_cpu_field(tmp, vfp.xregs[rn]);
3367 gen_lookup_tb(s);
3368 break;
3369 case ARM_VFP_FPINST:
3370 case ARM_VFP_FPINST2:
3371 if (IS_USER(s)) {
3372 return 1;
3373 }
3374 tmp = load_reg(s, rd);
3375 store_cpu_field(tmp, vfp.xregs[rn]);
3376 break;
3377 default:
3378 return 1;
3379 }
3380 } else {
3381 tmp = load_reg(s, rd);
3382 gen_vfp_msr(tmp);
3383 gen_mov_vreg_F0(0, rn);
3384 }
3385 }
3386 }
3387 } else {
3388 /* data processing */
3389 /* The opcode is in bits 23, 21, 20 and 6. */
3390 op = ((insn >> 20) & 8) | ((insn >> 19) & 6) | ((insn >> 6) & 1);
3391 if (dp) {
3392 if (op == 15) {
3393 /* rn is opcode */
3394 rn = ((insn >> 15) & 0x1e) | ((insn >> 7) & 1);
3395 } else {
3396 /* rn is register number */
3397 VFP_DREG_N(rn, insn);
3398 }
3399
3400 if (op == 15 && (rn == 15 || ((rn & 0x1c) == 0x18) ||
3401 ((rn & 0x1e) == 0x6))) {
3402 /* Integer or single/half precision destination. */
3403 rd = VFP_SREG_D(insn);
3404 } else {
3405 VFP_DREG_D(rd, insn);
3406 }
3407 if (op == 15 &&
3408 (((rn & 0x1c) == 0x10) || ((rn & 0x14) == 0x14) ||
3409 ((rn & 0x1e) == 0x4))) {
3410 /* VCVT from int or half precision is always from S reg
3411 * regardless of dp bit. VCVT with immediate frac_bits
3412 * has same format as SREG_M.
3413 */
3414 rm = VFP_SREG_M(insn);
3415 } else {
3416 VFP_DREG_M(rm, insn);
3417 }
3418 } else {
3419 rn = VFP_SREG_N(insn);
3420 if (op == 15 && rn == 15) {
3421 /* Double precision destination. */
3422 VFP_DREG_D(rd, insn);
3423 } else {
3424 rd = VFP_SREG_D(insn);
3425 }
3426 /* NB that we implicitly rely on the encoding for the frac_bits
3427 * in VCVT of fixed to float being the same as that of an SREG_M
3428 */
3429 rm = VFP_SREG_M(insn);
3430 }
3431
3432 veclen = s->vec_len;
3433 if (op == 15 && rn > 3)
3434 veclen = 0;
3435
3436 /* Shut up compiler warnings. */
3437 delta_m = 0;
3438 delta_d = 0;
3439 bank_mask = 0;
3440
3441 if (veclen > 0) {
3442 if (dp)
3443 bank_mask = 0xc;
3444 else
3445 bank_mask = 0x18;
3446
3447 /* Figure out what type of vector operation this is. */
3448 if ((rd & bank_mask) == 0) {
3449 /* scalar */
3450 veclen = 0;
3451 } else {
3452 if (dp)
3453 delta_d = (s->vec_stride >> 1) + 1;
3454 else
3455 delta_d = s->vec_stride + 1;
3456
3457 if ((rm & bank_mask) == 0) {
3458 /* mixed scalar/vector */
3459 delta_m = 0;
3460 } else {
3461 /* vector */
3462 delta_m = delta_d;
3463 }
3464 }
3465 }
3466
3467 /* Load the initial operands. */
3468 if (op == 15) {
3469 switch (rn) {
3470 case 16:
3471 case 17:
3472 /* Integer source */
3473 gen_mov_F0_vreg(0, rm);
3474 break;
3475 case 8:
3476 case 9:
3477 /* Compare */
3478 gen_mov_F0_vreg(dp, rd);
3479 gen_mov_F1_vreg(dp, rm);
3480 break;
3481 case 10:
3482 case 11:
3483 /* Compare with zero */
3484 gen_mov_F0_vreg(dp, rd);
3485 gen_vfp_F1_ld0(dp);
3486 break;
3487 case 20:
3488 case 21:
3489 case 22:
3490 case 23:
3491 case 28:
3492 case 29:
3493 case 30:
3494 case 31:
3495 /* Source and destination the same. */
3496 gen_mov_F0_vreg(dp, rd);
3497 break;
3498 case 4:
3499 case 5:
3500 case 6:
3501 case 7:
3502 /* VCVTB, VCVTT: only present with the halfprec extension
3503 * UNPREDICTABLE if bit 8 is set prior to ARMv8
3504 * (we choose to UNDEF)
3505 */
3506 if ((dp && !arm_dc_feature(s, ARM_FEATURE_V8)) ||
3507 !arm_dc_feature(s, ARM_FEATURE_VFP_FP16)) {
3508 return 1;
3509 }
3510 if (!extract32(rn, 1, 1)) {
3511 /* Half precision source. */
3512 gen_mov_F0_vreg(0, rm);
3513 break;
3514 }
3515 /* Otherwise fall through */
3516 default:
3517 /* One source operand. */
3518 gen_mov_F0_vreg(dp, rm);
3519 break;
3520 }
3521 } else {
3522 /* Two source operands. */
3523 gen_mov_F0_vreg(dp, rn);
3524 gen_mov_F1_vreg(dp, rm);
3525 }
3526
3527 for (;;) {
3528 /* Perform the calculation. */
3529 switch (op) {
3530 case 0: /* VMLA: fd + (fn * fm) */
3531 /* Note that order of inputs to the add matters for NaNs */
3532 gen_vfp_F1_mul(dp);
3533 gen_mov_F0_vreg(dp, rd);
3534 gen_vfp_add(dp);
3535 break;
3536 case 1: /* VMLS: fd + -(fn * fm) */
3537 gen_vfp_mul(dp);
3538 gen_vfp_F1_neg(dp);
3539 gen_mov_F0_vreg(dp, rd);
3540 gen_vfp_add(dp);
3541 break;
3542 case 2: /* VNMLS: -fd + (fn * fm) */
3543 /* Note that it isn't valid to replace (-A + B) with (B - A)
3544 * or similar plausible looking simplifications
3545 * because this will give wrong results for NaNs.
3546 */
3547 gen_vfp_F1_mul(dp);
3548 gen_mov_F0_vreg(dp, rd);
3549 gen_vfp_neg(dp);
3550 gen_vfp_add(dp);
3551 break;
3552 case 3: /* VNMLA: -fd + -(fn * fm) */
3553 gen_vfp_mul(dp);
3554 gen_vfp_F1_neg(dp);
3555 gen_mov_F0_vreg(dp, rd);
3556 gen_vfp_neg(dp);
3557 gen_vfp_add(dp);
3558 break;
3559 case 4: /* mul: fn * fm */
3560 gen_vfp_mul(dp);
3561 break;
3562 case 5: /* nmul: -(fn * fm) */
3563 gen_vfp_mul(dp);
3564 gen_vfp_neg(dp);
3565 break;
3566 case 6: /* add: fn + fm */
3567 gen_vfp_add(dp);
3568 break;
3569 case 7: /* sub: fn - fm */
3570 gen_vfp_sub(dp);
3571 break;
3572 case 8: /* div: fn / fm */
3573 gen_vfp_div(dp);
3574 break;
3575 case 10: /* VFNMA : fd = muladd(-fd, fn, fm) */
3576 case 11: /* VFNMS : fd = muladd(-fd, -fn, fm) */
3577 case 12: /* VFMA : fd = muladd( fd, fn, fm) */
3578 case 13: /* VFMS : fd = muladd( fd, -fn, fm) */
3579 /* These are fused multiply-add, and must be done as one
3580 * floating point operation with no rounding between the
3581 * multiplication and addition steps.
3582 * NB that doing the negations here as separate steps is
3583 * correct : an input NaN should come out with its sign bit
3584 * flipped if it is a negated-input.
3585 */
3586 if (!arm_dc_feature(s, ARM_FEATURE_VFP4)) {
3587 return 1;
3588 }
3589 if (dp) {
3590 TCGv_ptr fpst;
3591 TCGv_i64 frd;
3592 if (op & 1) {
3593 /* VFNMS, VFMS */
3594 gen_helper_vfp_negd(cpu_F0d, cpu_F0d);
3595 }
3596 frd = tcg_temp_new_i64();
3597 tcg_gen_ld_f64(frd, cpu_env, vfp_reg_offset(dp, rd));
3598 if (op & 2) {
3599 /* VFNMA, VFNMS */
3600 gen_helper_vfp_negd(frd, frd);
3601 }
3602 fpst = get_fpstatus_ptr(0);
3603 gen_helper_vfp_muladdd(cpu_F0d, cpu_F0d,
3604 cpu_F1d, frd, fpst);
3605 tcg_temp_free_ptr(fpst);
3606 tcg_temp_free_i64(frd);
3607 } else {
3608 TCGv_ptr fpst;
3609 TCGv_i32 frd;
3610 if (op & 1) {
3611 /* VFNMS, VFMS */
3612 gen_helper_vfp_negs(cpu_F0s, cpu_F0s);
3613 }
3614 frd = tcg_temp_new_i32();
3615 tcg_gen_ld_f32(frd, cpu_env, vfp_reg_offset(dp, rd));
3616 if (op & 2) {
3617 gen_helper_vfp_negs(frd, frd);
3618 }
3619 fpst = get_fpstatus_ptr(0);
3620 gen_helper_vfp_muladds(cpu_F0s, cpu_F0s,
3621 cpu_F1s, frd, fpst);
3622 tcg_temp_free_ptr(fpst);
3623 tcg_temp_free_i32(frd);
3624 }
3625 break;
3626 case 14: /* fconst */
3627 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3628 return 1;
3629 }
3630
3631 n = (insn << 12) & 0x80000000;
3632 i = ((insn >> 12) & 0x70) | (insn & 0xf);
3633 if (dp) {
3634 if (i & 0x40)
3635 i |= 0x3f80;
3636 else
3637 i |= 0x4000;
3638 n |= i << 16;
3639 tcg_gen_movi_i64(cpu_F0d, ((uint64_t)n) << 32);
3640 } else {
3641 if (i & 0x40)
3642 i |= 0x780;
3643 else
3644 i |= 0x800;
3645 n |= i << 19;
3646 tcg_gen_movi_i32(cpu_F0s, n);
3647 }
3648 break;
3649 case 15: /* extension space */
3650 switch (rn) {
3651 case 0: /* cpy */
3652 /* no-op */
3653 break;
3654 case 1: /* abs */
3655 gen_vfp_abs(dp);
3656 break;
3657 case 2: /* neg */
3658 gen_vfp_neg(dp);
3659 break;
3660 case 3: /* sqrt */
3661 gen_vfp_sqrt(dp);
3662 break;
3663 case 4: /* vcvtb.f32.f16, vcvtb.f64.f16 */
3664 tmp = gen_vfp_mrs();
3665 tcg_gen_ext16u_i32(tmp, tmp);
3666 if (dp) {
3667 gen_helper_vfp_fcvt_f16_to_f64(cpu_F0d, tmp,
3668 cpu_env);
3669 } else {
3670 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp,
3671 cpu_env);
3672 }
3673 tcg_temp_free_i32(tmp);
3674 break;
3675 case 5: /* vcvtt.f32.f16, vcvtt.f64.f16 */
3676 tmp = gen_vfp_mrs();
3677 tcg_gen_shri_i32(tmp, tmp, 16);
3678 if (dp) {
3679 gen_helper_vfp_fcvt_f16_to_f64(cpu_F0d, tmp,
3680 cpu_env);
3681 } else {
3682 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp,
3683 cpu_env);
3684 }
3685 tcg_temp_free_i32(tmp);
3686 break;
3687 case 6: /* vcvtb.f16.f32, vcvtb.f16.f64 */
3688 tmp = tcg_temp_new_i32();
3689 if (dp) {
3690 gen_helper_vfp_fcvt_f64_to_f16(tmp, cpu_F0d,
3691 cpu_env);
3692 } else {
3693 gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s,
3694 cpu_env);
3695 }
3696 gen_mov_F0_vreg(0, rd);
3697 tmp2 = gen_vfp_mrs();
3698 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
3699 tcg_gen_or_i32(tmp, tmp, tmp2);
3700 tcg_temp_free_i32(tmp2);
3701 gen_vfp_msr(tmp);
3702 break;
3703 case 7: /* vcvtt.f16.f32, vcvtt.f16.f64 */
3704 tmp = tcg_temp_new_i32();
3705 if (dp) {
3706 gen_helper_vfp_fcvt_f64_to_f16(tmp, cpu_F0d,
3707 cpu_env);
3708 } else {
3709 gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s,
3710 cpu_env);
3711 }
3712 tcg_gen_shli_i32(tmp, tmp, 16);
3713 gen_mov_F0_vreg(0, rd);
3714 tmp2 = gen_vfp_mrs();
3715 tcg_gen_ext16u_i32(tmp2, tmp2);
3716 tcg_gen_or_i32(tmp, tmp, tmp2);
3717 tcg_temp_free_i32(tmp2);
3718 gen_vfp_msr(tmp);
3719 break;
3720 case 8: /* cmp */
3721 gen_vfp_cmp(dp);
3722 break;
3723 case 9: /* cmpe */
3724 gen_vfp_cmpe(dp);
3725 break;
3726 case 10: /* cmpz */
3727 gen_vfp_cmp(dp);
3728 break;
3729 case 11: /* cmpez */
3730 gen_vfp_F1_ld0(dp);
3731 gen_vfp_cmpe(dp);
3732 break;
3733 case 12: /* vrintr */
3734 {
3735 TCGv_ptr fpst = get_fpstatus_ptr(0);
3736 if (dp) {
3737 gen_helper_rintd(cpu_F0d, cpu_F0d, fpst);
3738 } else {
3739 gen_helper_rints(cpu_F0s, cpu_F0s, fpst);
3740 }
3741 tcg_temp_free_ptr(fpst);
3742 break;
3743 }
3744 case 13: /* vrintz */
3745 {
3746 TCGv_ptr fpst = get_fpstatus_ptr(0);
3747 TCGv_i32 tcg_rmode;
3748 tcg_rmode = tcg_const_i32(float_round_to_zero);
3749 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
3750 if (dp) {
3751 gen_helper_rintd(cpu_F0d, cpu_F0d, fpst);
3752 } else {
3753 gen_helper_rints(cpu_F0s, cpu_F0s, fpst);
3754 }
3755 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
3756 tcg_temp_free_i32(tcg_rmode);
3757 tcg_temp_free_ptr(fpst);
3758 break;
3759 }
3760 case 14: /* vrintx */
3761 {
3762 TCGv_ptr fpst = get_fpstatus_ptr(0);
3763 if (dp) {
3764 gen_helper_rintd_exact(cpu_F0d, cpu_F0d, fpst);
3765 } else {
3766 gen_helper_rints_exact(cpu_F0s, cpu_F0s, fpst);
3767 }
3768 tcg_temp_free_ptr(fpst);
3769 break;
3770 }
3771 case 15: /* single<->double conversion */
3772 if (dp)
3773 gen_helper_vfp_fcvtsd(cpu_F0s, cpu_F0d, cpu_env);
3774 else
3775 gen_helper_vfp_fcvtds(cpu_F0d, cpu_F0s, cpu_env);
3776 break;
3777 case 16: /* fuito */
3778 gen_vfp_uito(dp, 0);
3779 break;
3780 case 17: /* fsito */
3781 gen_vfp_sito(dp, 0);
3782 break;
3783 case 20: /* fshto */
3784 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3785 return 1;
3786 }
3787 gen_vfp_shto(dp, 16 - rm, 0);
3788 break;
3789 case 21: /* fslto */
3790 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3791 return 1;
3792 }
3793 gen_vfp_slto(dp, 32 - rm, 0);
3794 break;
3795 case 22: /* fuhto */
3796 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3797 return 1;
3798 }
3799 gen_vfp_uhto(dp, 16 - rm, 0);
3800 break;
3801 case 23: /* fulto */
3802 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3803 return 1;
3804 }
3805 gen_vfp_ulto(dp, 32 - rm, 0);
3806 break;
3807 case 24: /* ftoui */
3808 gen_vfp_toui(dp, 0);
3809 break;
3810 case 25: /* ftouiz */
3811 gen_vfp_touiz(dp, 0);
3812 break;
3813 case 26: /* ftosi */
3814 gen_vfp_tosi(dp, 0);
3815 break;
3816 case 27: /* ftosiz */
3817 gen_vfp_tosiz(dp, 0);
3818 break;
3819 case 28: /* ftosh */
3820 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3821 return 1;
3822 }
3823 gen_vfp_tosh(dp, 16 - rm, 0);
3824 break;
3825 case 29: /* ftosl */
3826 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3827 return 1;
3828 }
3829 gen_vfp_tosl(dp, 32 - rm, 0);
3830 break;
3831 case 30: /* ftouh */
3832 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3833 return 1;
3834 }
3835 gen_vfp_touh(dp, 16 - rm, 0);
3836 break;
3837 case 31: /* ftoul */
3838 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3839 return 1;
3840 }
3841 gen_vfp_toul(dp, 32 - rm, 0);
3842 break;
3843 default: /* undefined */
3844 return 1;
3845 }
3846 break;
3847 default: /* undefined */
3848 return 1;
3849 }
3850
3851 /* Write back the result. */
3852 if (op == 15 && (rn >= 8 && rn <= 11)) {
3853 /* Comparison, do nothing. */
3854 } else if (op == 15 && dp && ((rn & 0x1c) == 0x18 ||
3855 (rn & 0x1e) == 0x6)) {
3856 /* VCVT double to int: always integer result.
3857 * VCVT double to half precision is always a single
3858 * precision result.
3859 */
3860 gen_mov_vreg_F0(0, rd);
3861 } else if (op == 15 && rn == 15) {
3862 /* conversion */
3863 gen_mov_vreg_F0(!dp, rd);
3864 } else {
3865 gen_mov_vreg_F0(dp, rd);
3866 }
3867
3868 /* break out of the loop if we have finished */
3869 if (veclen == 0)
3870 break;
3871
3872 if (op == 15 && delta_m == 0) {
3873 /* single source one-many */
3874 while (veclen--) {
3875 rd = ((rd + delta_d) & (bank_mask - 1))
3876 | (rd & bank_mask);
3877 gen_mov_vreg_F0(dp, rd);
3878 }
3879 break;
3880 }
3881 /* Setup the next operands. */
3882 veclen--;
3883 rd = ((rd + delta_d) & (bank_mask - 1))
3884 | (rd & bank_mask);
3885
3886 if (op == 15) {
3887 /* One source operand. */
3888 rm = ((rm + delta_m) & (bank_mask - 1))
3889 | (rm & bank_mask);
3890 gen_mov_F0_vreg(dp, rm);
3891 } else {
3892 /* Two source operands. */
3893 rn = ((rn + delta_d) & (bank_mask - 1))
3894 | (rn & bank_mask);
3895 gen_mov_F0_vreg(dp, rn);
3896 if (delta_m) {
3897 rm = ((rm + delta_m) & (bank_mask - 1))
3898 | (rm & bank_mask);
3899 gen_mov_F1_vreg(dp, rm);
3900 }
3901 }
3902 }
3903 }
3904 break;
3905 case 0xc:
3906 case 0xd:
3907 if ((insn & 0x03e00000) == 0x00400000) {
3908 /* two-register transfer */
3909 rn = (insn >> 16) & 0xf;
3910 rd = (insn >> 12) & 0xf;
3911 if (dp) {
3912 VFP_DREG_M(rm, insn);
3913 } else {
3914 rm = VFP_SREG_M(insn);
3915 }
3916
3917 if (insn & ARM_CP_RW_BIT) {
3918 /* vfp->arm */
3919 if (dp) {
3920 gen_mov_F0_vreg(0, rm * 2);
3921 tmp = gen_vfp_mrs();
3922 store_reg(s, rd, tmp);
3923 gen_mov_F0_vreg(0, rm * 2 + 1);
3924 tmp = gen_vfp_mrs();
3925 store_reg(s, rn, tmp);
3926 } else {
3927 gen_mov_F0_vreg(0, rm);
3928 tmp = gen_vfp_mrs();
3929 store_reg(s, rd, tmp);
3930 gen_mov_F0_vreg(0, rm + 1);
3931 tmp = gen_vfp_mrs();
3932 store_reg(s, rn, tmp);
3933 }
3934 } else {
3935 /* arm->vfp */
3936 if (dp) {
3937 tmp = load_reg(s, rd);
3938 gen_vfp_msr(tmp);
3939 gen_mov_vreg_F0(0, rm * 2);
3940 tmp = load_reg(s, rn);
3941 gen_vfp_msr(tmp);
3942 gen_mov_vreg_F0(0, rm * 2 + 1);
3943 } else {
3944 tmp = load_reg(s, rd);
3945 gen_vfp_msr(tmp);
3946 gen_mov_vreg_F0(0, rm);
3947 tmp = load_reg(s, rn);
3948 gen_vfp_msr(tmp);
3949 gen_mov_vreg_F0(0, rm + 1);
3950 }
3951 }
3952 } else {
3953 /* Load/store */
3954 rn = (insn >> 16) & 0xf;
3955 if (dp)
3956 VFP_DREG_D(rd, insn);
3957 else
3958 rd = VFP_SREG_D(insn);
3959 if ((insn & 0x01200000) == 0x01000000) {
3960 /* Single load/store */
3961 offset = (insn & 0xff) << 2;
3962 if ((insn & (1 << 23)) == 0)
3963 offset = -offset;
3964 if (s->thumb && rn == 15) {
3965 /* This is actually UNPREDICTABLE */
3966 addr = tcg_temp_new_i32();
3967 tcg_gen_movi_i32(addr, s->pc & ~2);
3968 } else {
3969 addr = load_reg(s, rn);
3970 }
3971 tcg_gen_addi_i32(addr, addr, offset);
3972 if (insn & (1 << 20)) {
3973 gen_vfp_ld(s, dp, addr);
3974 gen_mov_vreg_F0(dp, rd);
3975 } else {
3976 gen_mov_F0_vreg(dp, rd);
3977 gen_vfp_st(s, dp, addr);
3978 }
3979 tcg_temp_free_i32(addr);
3980 } else {
3981 /* load/store multiple */
3982 int w = insn & (1 << 21);
3983 if (dp)
3984 n = (insn >> 1) & 0x7f;
3985 else
3986 n = insn & 0xff;
3987
3988 if (w && !(((insn >> 23) ^ (insn >> 24)) & 1)) {
3989 /* P == U , W == 1 => UNDEF */
3990 return 1;
3991 }
3992 if (n == 0 || (rd + n) > 32 || (dp && n > 16)) {
3993 /* UNPREDICTABLE cases for bad immediates: we choose to
3994 * UNDEF to avoid generating huge numbers of TCG ops
3995 */
3996 return 1;
3997 }
3998 if (rn == 15 && w) {
3999 /* writeback to PC is UNPREDICTABLE, we choose to UNDEF */
4000 return 1;
4001 }
4002
4003 if (s->thumb && rn == 15) {
4004 /* This is actually UNPREDICTABLE */
4005 addr = tcg_temp_new_i32();
4006 tcg_gen_movi_i32(addr, s->pc & ~2);
4007 } else {
4008 addr = load_reg(s, rn);
4009 }
4010 if (insn & (1 << 24)) /* pre-decrement */
4011 tcg_gen_addi_i32(addr, addr, -((insn & 0xff) << 2));
4012
4013 if (dp)
4014 offset = 8;
4015 else
4016 offset = 4;
4017 for (i = 0; i < n; i++) {
4018 if (insn & ARM_CP_RW_BIT) {
4019 /* load */
4020 gen_vfp_ld(s, dp, addr);
4021 gen_mov_vreg_F0(dp, rd + i);
4022 } else {
4023 /* store */
4024 gen_mov_F0_vreg(dp, rd + i);
4025 gen_vfp_st(s, dp, addr);
4026 }
4027 tcg_gen_addi_i32(addr, addr, offset);
4028 }
4029 if (w) {
4030 /* writeback */
4031 if (insn & (1 << 24))
4032 offset = -offset * n;
4033 else if (dp && (insn & 1))
4034 offset = 4;
4035 else
4036 offset = 0;
4037
4038 if (offset != 0)
4039 tcg_gen_addi_i32(addr, addr, offset);
4040 store_reg(s, rn, addr);
4041 } else {
4042 tcg_temp_free_i32(addr);
4043 }
4044 }
4045 }
4046 break;
4047 default:
4048 /* Should never happen. */
4049 return 1;
4050 }
4051 return 0;
4052 }
4053
4054 static inline bool use_goto_tb(DisasContext *s, target_ulong dest)
4055 {
4056 #ifndef CONFIG_USER_ONLY
4057 return (s->tb->pc & TARGET_PAGE_MASK) == (dest & TARGET_PAGE_MASK) ||
4058 ((s->pc - 1) & TARGET_PAGE_MASK) == (dest & TARGET_PAGE_MASK);
4059 #else
4060 return true;
4061 #endif
4062 }
4063
4064 static inline void gen_goto_tb(DisasContext *s, int n, target_ulong dest)
4065 {
4066 if (use_goto_tb(s, dest)) {
4067 tcg_gen_goto_tb(n);
4068 gen_set_pc_im(s, dest);
4069 tcg_gen_exit_tb((uintptr_t)s->tb + n);
4070 } else {
4071 gen_set_pc_im(s, dest);
4072 tcg_gen_exit_tb(0);
4073 }
4074 }
4075
4076 static inline void gen_jmp (DisasContext *s, uint32_t dest)
4077 {
4078 if (unlikely(s->singlestep_enabled || s->ss_active)) {
4079 /* An indirect jump so that we still trigger the debug exception. */
4080 if (s->thumb)
4081 dest |= 1;
4082 gen_bx_im(s, dest);
4083 } else {
4084 gen_goto_tb(s, 0, dest);
4085 s->is_jmp = DISAS_TB_JUMP;
4086 }
4087 }
4088
4089 static inline void gen_mulxy(TCGv_i32 t0, TCGv_i32 t1, int x, int y)
4090 {
4091 if (x)
4092 tcg_gen_sari_i32(t0, t0, 16);
4093 else
4094 gen_sxth(t0);
4095 if (y)
4096 tcg_gen_sari_i32(t1, t1, 16);
4097 else
4098 gen_sxth(t1);
4099 tcg_gen_mul_i32(t0, t0, t1);
4100 }
4101
4102 /* Return the mask of PSR bits set by a MSR instruction. */
4103 static uint32_t msr_mask(DisasContext *s, int flags, int spsr)
4104 {
4105 uint32_t mask;
4106
4107 mask = 0;
4108 if (flags & (1 << 0))
4109 mask |= 0xff;
4110 if (flags & (1 << 1))
4111 mask |= 0xff00;
4112 if (flags & (1 << 2))
4113 mask |= 0xff0000;
4114 if (flags & (1 << 3))
4115 mask |= 0xff000000;
4116
4117 /* Mask out undefined bits. */
4118 mask &= ~CPSR_RESERVED;
4119 if (!arm_dc_feature(s, ARM_FEATURE_V4T)) {
4120 mask &= ~CPSR_T;
4121 }
4122 if (!arm_dc_feature(s, ARM_FEATURE_V5)) {
4123 mask &= ~CPSR_Q; /* V5TE in reality*/
4124 }
4125 if (!arm_dc_feature(s, ARM_FEATURE_V6)) {
4126 mask &= ~(CPSR_E | CPSR_GE);
4127 }
4128 if (!arm_dc_feature(s, ARM_FEATURE_THUMB2)) {
4129 mask &= ~CPSR_IT;
4130 }
4131 /* Mask out execution state and reserved bits. */
4132 if (!spsr) {
4133 mask &= ~(CPSR_EXEC | CPSR_RESERVED);
4134 }
4135 /* Mask out privileged bits. */
4136 if (IS_USER(s))
4137 mask &= CPSR_USER;
4138 return mask;
4139 }
4140
4141 /* Returns nonzero if access to the PSR is not permitted. Marks t0 as dead. */
4142 static int gen_set_psr(DisasContext *s, uint32_t mask, int spsr, TCGv_i32 t0)
4143 {
4144 TCGv_i32 tmp;
4145 if (spsr) {
4146 /* ??? This is also undefined in system mode. */
4147 if (IS_USER(s))
4148 return 1;
4149
4150 tmp = load_cpu_field(spsr);
4151 tcg_gen_andi_i32(tmp, tmp, ~mask);
4152 tcg_gen_andi_i32(t0, t0, mask);
4153 tcg_gen_or_i32(tmp, tmp, t0);
4154 store_cpu_field(tmp, spsr);
4155 } else {
4156 gen_set_cpsr(t0, mask);
4157 }
4158 tcg_temp_free_i32(t0);
4159 gen_lookup_tb(s);
4160 return 0;
4161 }
4162
4163 /* Returns nonzero if access to the PSR is not permitted. */
4164 static int gen_set_psr_im(DisasContext *s, uint32_t mask, int spsr, uint32_t val)
4165 {
4166 TCGv_i32 tmp;
4167 tmp = tcg_temp_new_i32();
4168 tcg_gen_movi_i32(tmp, val);
4169 return gen_set_psr(s, mask, spsr, tmp);
4170 }
4171
4172 static bool msr_banked_access_decode(DisasContext *s, int r, int sysm, int rn,
4173 int *tgtmode, int *regno)
4174 {
4175 /* Decode the r and sysm fields of MSR/MRS banked accesses into
4176 * the target mode and register number, and identify the various
4177 * unpredictable cases.
4178 * MSR (banked) and MRS (banked) are CONSTRAINED UNPREDICTABLE if:
4179 * + executed in user mode
4180 * + using R15 as the src/dest register
4181 * + accessing an unimplemented register
4182 * + accessing a register that's inaccessible at current PL/security state*
4183 * + accessing a register that you could access with a different insn
4184 * We choose to UNDEF in all these cases.
4185 * Since we don't know which of the various AArch32 modes we are in
4186 * we have to defer some checks to runtime.
4187 * Accesses to Monitor mode registers from Secure EL1 (which implies
4188 * that EL3 is AArch64) must trap to EL3.
4189 *
4190 * If the access checks fail this function will emit code to take
4191 * an exception and return false. Otherwise it will return true,
4192 * and set *tgtmode and *regno appropriately.
4193 */
4194 int exc_target = default_exception_el(s);
4195
4196 /* These instructions are present only in ARMv8, or in ARMv7 with the
4197 * Virtualization Extensions.
4198 */
4199 if (!arm_dc_feature(s, ARM_FEATURE_V8) &&
4200 !arm_dc_feature(s, ARM_FEATURE_EL2)) {
4201 goto undef;
4202 }
4203
4204 if (IS_USER(s) || rn == 15) {
4205 goto undef;
4206 }
4207
4208 /* The table in the v8 ARM ARM section F5.2.3 describes the encoding
4209 * of registers into (r, sysm).
4210 */
4211 if (r) {
4212 /* SPSRs for other modes */
4213 switch (sysm) {
4214 case 0xe: /* SPSR_fiq */
4215 *tgtmode = ARM_CPU_MODE_FIQ;
4216 break;
4217 case 0x10: /* SPSR_irq */
4218 *tgtmode = ARM_CPU_MODE_IRQ;
4219 break;
4220 case 0x12: /* SPSR_svc */
4221 *tgtmode = ARM_CPU_MODE_SVC;
4222 break;
4223 case 0x14: /* SPSR_abt */
4224 *tgtmode = ARM_CPU_MODE_ABT;
4225 break;
4226 case 0x16: /* SPSR_und */
4227 *tgtmode = ARM_CPU_MODE_UND;
4228 break;
4229 case 0x1c: /* SPSR_mon */
4230 *tgtmode = ARM_CPU_MODE_MON;
4231 break;
4232 case 0x1e: /* SPSR_hyp */
4233 *tgtmode = ARM_CPU_MODE_HYP;
4234 break;
4235 default: /* unallocated */
4236 goto undef;
4237 }
4238 /* We arbitrarily assign SPSR a register number of 16. */
4239 *regno = 16;
4240 } else {
4241 /* general purpose registers for other modes */
4242 switch (sysm) {
4243 case 0x0 ... 0x6: /* 0b00xxx : r8_usr ... r14_usr */
4244 *tgtmode = ARM_CPU_MODE_USR;
4245 *regno = sysm + 8;
4246 break;
4247 case 0x8 ... 0xe: /* 0b01xxx : r8_fiq ... r14_fiq */
4248 *tgtmode = ARM_CPU_MODE_FIQ;
4249 *regno = sysm;
4250 break;
4251 case 0x10 ... 0x11: /* 0b1000x : r14_irq, r13_irq */
4252 *tgtmode = ARM_CPU_MODE_IRQ;
4253 *regno = sysm & 1 ? 13 : 14;
4254 break;
4255 case 0x12 ... 0x13: /* 0b1001x : r14_svc, r13_svc */
4256 *tgtmode = ARM_CPU_MODE_SVC;
4257 *regno = sysm & 1 ? 13 : 14;
4258 break;
4259 case 0x14 ... 0x15: /* 0b1010x : r14_abt, r13_abt */
4260 *tgtmode = ARM_CPU_MODE_ABT;
4261 *regno = sysm & 1 ? 13 : 14;
4262 break;
4263 case 0x16 ... 0x17: /* 0b1011x : r14_und, r13_und */
4264 *tgtmode = ARM_CPU_MODE_UND;
4265 *regno = sysm & 1 ? 13 : 14;
4266 break;
4267 case 0x1c ... 0x1d: /* 0b1110x : r14_mon, r13_mon */
4268 *tgtmode = ARM_CPU_MODE_MON;
4269 *regno = sysm & 1 ? 13 : 14;
4270 break;
4271 case 0x1e ... 0x1f: /* 0b1111x : elr_hyp, r13_hyp */
4272 *tgtmode = ARM_CPU_MODE_HYP;
4273 /* Arbitrarily pick 17 for ELR_Hyp (which is not a banked LR!) */
4274 *regno = sysm & 1 ? 13 : 17;
4275 break;
4276 default: /* unallocated */
4277 goto undef;
4278 }
4279 }
4280
4281 /* Catch the 'accessing inaccessible register' cases we can detect
4282 * at translate time.
4283 */
4284 switch (*tgtmode) {
4285 case ARM_CPU_MODE_MON:
4286 if (!arm_dc_feature(s, ARM_FEATURE_EL3) || s->ns) {
4287 goto undef;
4288 }
4289 if (s->current_el == 1) {
4290 /* If we're in Secure EL1 (which implies that EL3 is AArch64)
4291 * then accesses to Mon registers trap to EL3
4292 */
4293 exc_target = 3;
4294 goto undef;
4295 }
4296 break;
4297 case ARM_CPU_MODE_HYP:
4298 /* Note that we can forbid accesses from EL2 here because they
4299 * must be from Hyp mode itself
4300 */
4301 if (!arm_dc_feature(s, ARM_FEATURE_EL2) || s->current_el < 3) {
4302 goto undef;
4303 }
4304 break;
4305 default:
4306 break;
4307 }
4308
4309 return true;
4310
4311 undef:
4312 /* If we get here then some access check did not pass */
4313 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(), exc_target);
4314 return false;
4315 }
4316
4317 static void gen_msr_banked(DisasContext *s, int r, int sysm, int rn)
4318 {
4319 TCGv_i32 tcg_reg, tcg_tgtmode, tcg_regno;
4320 int tgtmode = 0, regno = 0;
4321
4322 if (!msr_banked_access_decode(s, r, sysm, rn, &tgtmode, &regno)) {
4323 return;
4324 }
4325
4326 /* Sync state because msr_banked() can raise exceptions */
4327 gen_set_condexec(s);
4328 gen_set_pc_im(s, s->pc - 4);
4329 tcg_reg = load_reg(s, rn);
4330 tcg_tgtmode = tcg_const_i32(tgtmode);
4331 tcg_regno = tcg_const_i32(regno);
4332 gen_helper_msr_banked(cpu_env, tcg_reg, tcg_tgtmode, tcg_regno);
4333 tcg_temp_free_i32(tcg_tgtmode);
4334 tcg_temp_free_i32(tcg_regno);
4335 tcg_temp_free_i32(tcg_reg);
4336 s->is_jmp = DISAS_UPDATE;
4337 }
4338
4339 static void gen_mrs_banked(DisasContext *s, int r, int sysm, int rn)
4340 {
4341 TCGv_i32 tcg_reg, tcg_tgtmode, tcg_regno;
4342 int tgtmode = 0, regno = 0;
4343
4344 if (!msr_banked_access_decode(s, r, sysm, rn, &tgtmode, &regno)) {
4345 return;
4346 }
4347
4348 /* Sync state because mrs_banked() can raise exceptions */
4349 gen_set_condexec(s);
4350 gen_set_pc_im(s, s->pc - 4);
4351 tcg_reg = tcg_temp_new_i32();
4352 tcg_tgtmode = tcg_const_i32(tgtmode);
4353 tcg_regno = tcg_const_i32(regno);
4354 gen_helper_mrs_banked(tcg_reg, cpu_env, tcg_tgtmode, tcg_regno);
4355 tcg_temp_free_i32(tcg_tgtmode);
4356 tcg_temp_free_i32(tcg_regno);
4357 store_reg(s, rn, tcg_reg);
4358 s->is_jmp = DISAS_UPDATE;
4359 }
4360
4361 /* Generate an old-style exception return. Marks pc as dead. */
4362 static void gen_exception_return(DisasContext *s, TCGv_i32 pc)
4363 {
4364 TCGv_i32 tmp;
4365 store_reg(s, 15, pc);
4366 tmp = load_cpu_field(spsr);
4367 gen_helper_cpsr_write_eret(cpu_env, tmp);
4368 tcg_temp_free_i32(tmp);
4369 s->is_jmp = DISAS_JUMP;
4370 }
4371
4372 /* Generate a v6 exception return. Marks both values as dead. */
4373 static void gen_rfe(DisasContext *s, TCGv_i32 pc, TCGv_i32 cpsr)
4374 {
4375 gen_helper_cpsr_write_eret(cpu_env, cpsr);
4376 tcg_temp_free_i32(cpsr);
4377 store_reg(s, 15, pc);
4378 s->is_jmp = DISAS_JUMP;
4379 }
4380
4381 static void gen_nop_hint(DisasContext *s, int val)
4382 {
4383 switch (val) {
4384 case 1: /* yield */
4385 gen_set_pc_im(s, s->pc);
4386 s->is_jmp = DISAS_YIELD;
4387 break;
4388 case 3: /* wfi */
4389 gen_set_pc_im(s, s->pc);
4390 s->is_jmp = DISAS_WFI;
4391 break;
4392 case 2: /* wfe */
4393 gen_set_pc_im(s, s->pc);
4394 s->is_jmp = DISAS_WFE;
4395 break;
4396 case 4: /* sev */
4397 case 5: /* sevl */
4398 /* TODO: Implement SEV, SEVL and WFE. May help SMP performance. */
4399 default: /* nop */
4400 break;
4401 }
4402 }
4403
4404 #define CPU_V001 cpu_V0, cpu_V0, cpu_V1
4405
4406 static inline void gen_neon_add(int size, TCGv_i32 t0, TCGv_i32 t1)
4407 {
4408 switch (size) {
4409 case 0: gen_helper_neon_add_u8(t0, t0, t1); break;
4410 case 1: gen_helper_neon_add_u16(t0, t0, t1); break;
4411 case 2: tcg_gen_add_i32(t0, t0, t1); break;
4412 default: abort();
4413 }
4414 }
4415
4416 static inline void gen_neon_rsb(int size, TCGv_i32 t0, TCGv_i32 t1)
4417 {
4418 switch (size) {
4419 case 0: gen_helper_neon_sub_u8(t0, t1, t0); break;
4420 case 1: gen_helper_neon_sub_u16(t0, t1, t0); break;
4421 case 2: tcg_gen_sub_i32(t0, t1, t0); break;
4422 default: return;
4423 }
4424 }
4425
4426 /* 32-bit pairwise ops end up the same as the elementwise versions. */
4427 #define gen_helper_neon_pmax_s32 gen_helper_neon_max_s32
4428 #define gen_helper_neon_pmax_u32 gen_helper_neon_max_u32
4429 #define gen_helper_neon_pmin_s32 gen_helper_neon_min_s32
4430 #define gen_helper_neon_pmin_u32 gen_helper_neon_min_u32
4431
4432 #define GEN_NEON_INTEGER_OP_ENV(name) do { \
4433 switch ((size << 1) | u) { \
4434 case 0: \
4435 gen_helper_neon_##name##_s8(tmp, cpu_env, tmp, tmp2); \
4436 break; \
4437 case 1: \
4438 gen_helper_neon_##name##_u8(tmp, cpu_env, tmp, tmp2); \
4439 break; \
4440 case 2: \
4441 gen_helper_neon_##name##_s16(tmp, cpu_env, tmp, tmp2); \
4442 break; \
4443 case 3: \
4444 gen_helper_neon_##name##_u16(tmp, cpu_env, tmp, tmp2); \
4445 break; \
4446 case 4: \
4447 gen_helper_neon_##name##_s32(tmp, cpu_env, tmp, tmp2); \
4448 break; \
4449 case 5: \
4450 gen_helper_neon_##name##_u32(tmp, cpu_env, tmp, tmp2); \
4451 break; \
4452 default: return 1; \
4453 }} while (0)
4454
4455 #define GEN_NEON_INTEGER_OP(name) do { \
4456 switch ((size << 1) | u) { \
4457 case 0: \
4458 gen_helper_neon_##name##_s8(tmp, tmp, tmp2); \
4459 break; \
4460 case 1: \
4461 gen_helper_neon_##name##_u8(tmp, tmp, tmp2); \
4462 break; \
4463 case 2: \
4464 gen_helper_neon_##name##_s16(tmp, tmp, tmp2); \
4465 break; \
4466 case 3: \
4467 gen_helper_neon_##name##_u16(tmp, tmp, tmp2); \
4468 break; \
4469 case 4: \
4470 gen_helper_neon_##name##_s32(tmp, tmp, tmp2); \
4471 break; \
4472 case 5: \
4473 gen_helper_neon_##name##_u32(tmp, tmp, tmp2); \
4474 break; \
4475 default: return 1; \
4476 }} while (0)
4477
4478 static TCGv_i32 neon_load_scratch(int scratch)
4479 {
4480 TCGv_i32 tmp = tcg_temp_new_i32();
4481 tcg_gen_ld_i32(tmp, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
4482 return tmp;
4483 }
4484
4485 static void neon_store_scratch(int scratch, TCGv_i32 var)
4486 {
4487 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
4488 tcg_temp_free_i32(var);
4489 }
4490
4491 static inline TCGv_i32 neon_get_scalar(int size, int reg)
4492 {
4493 TCGv_i32 tmp;
4494 if (size == 1) {
4495 tmp = neon_load_reg(reg & 7, reg >> 4);
4496 if (reg & 8) {
4497 gen_neon_dup_high16(tmp);
4498 } else {
4499 gen_neon_dup_low16(tmp);
4500 }
4501 } else {
4502 tmp = neon_load_reg(reg & 15, reg >> 4);
4503 }
4504 return tmp;
4505 }
4506
4507 static int gen_neon_unzip(int rd, int rm, int size, int q)
4508 {
4509 TCGv_i32 tmp, tmp2;
4510 if (!q && size == 2) {
4511 return 1;
4512 }
4513 tmp = tcg_const_i32(rd);
4514 tmp2 = tcg_const_i32(rm);
4515 if (q) {
4516 switch (size) {
4517 case 0:
4518 gen_helper_neon_qunzip8(cpu_env, tmp, tmp2);
4519 break;
4520 case 1:
4521 gen_helper_neon_qunzip16(cpu_env, tmp, tmp2);
4522 break;
4523 case 2:
4524 gen_helper_neon_qunzip32(cpu_env, tmp, tmp2);
4525 break;
4526 default:
4527 abort();
4528 }
4529 } else {
4530 switch (size) {
4531 case 0:
4532 gen_helper_neon_unzip8(cpu_env, tmp, tmp2);
4533 break;
4534 case 1:
4535 gen_helper_neon_unzip16(cpu_env, tmp, tmp2);
4536 break;
4537 default:
4538 abort();
4539 }
4540 }
4541 tcg_temp_free_i32(tmp);
4542 tcg_temp_free_i32(tmp2);
4543 return 0;
4544 }
4545
4546 static int gen_neon_zip(int rd, int rm, int size, int q)
4547 {
4548 TCGv_i32 tmp, tmp2;
4549 if (!q && size == 2) {
4550 return 1;
4551 }
4552 tmp = tcg_const_i32(rd);
4553 tmp2 = tcg_const_i32(rm);
4554 if (q) {
4555 switch (size) {
4556 case 0:
4557 gen_helper_neon_qzip8(cpu_env, tmp, tmp2);
4558 break;
4559 case 1:
4560 gen_helper_neon_qzip16(cpu_env, tmp, tmp2);
4561 break;
4562 case 2:
4563 gen_helper_neon_qzip32(cpu_env, tmp, tmp2);
4564 break;
4565 default:
4566 abort();
4567 }
4568 } else {
4569 switch (size) {
4570 case 0:
4571 gen_helper_neon_zip8(cpu_env, tmp, tmp2);
4572 break;
4573 case 1:
4574 gen_helper_neon_zip16(cpu_env, tmp, tmp2);
4575 break;
4576 default:
4577 abort();
4578 }
4579 }
4580 tcg_temp_free_i32(tmp);
4581 tcg_temp_free_i32(tmp2);
4582 return 0;
4583 }
4584
4585 static void gen_neon_trn_u8(TCGv_i32 t0, TCGv_i32 t1)
4586 {
4587 TCGv_i32 rd, tmp;
4588
4589 rd = tcg_temp_new_i32();
4590 tmp = tcg_temp_new_i32();
4591
4592 tcg_gen_shli_i32(rd, t0, 8);
4593 tcg_gen_andi_i32(rd, rd, 0xff00ff00);
4594 tcg_gen_andi_i32(tmp, t1, 0x00ff00ff);
4595 tcg_gen_or_i32(rd, rd, tmp);
4596
4597 tcg_gen_shri_i32(t1, t1, 8);
4598 tcg_gen_andi_i32(t1, t1, 0x00ff00ff);
4599 tcg_gen_andi_i32(tmp, t0, 0xff00ff00);
4600 tcg_gen_or_i32(t1, t1, tmp);
4601 tcg_gen_mov_i32(t0, rd);
4602
4603 tcg_temp_free_i32(tmp);
4604 tcg_temp_free_i32(rd);
4605 }
4606
4607 static void gen_neon_trn_u16(TCGv_i32 t0, TCGv_i32 t1)
4608 {
4609 TCGv_i32 rd, tmp;
4610
4611 rd = tcg_temp_new_i32();
4612 tmp = tcg_temp_new_i32();
4613
4614 tcg_gen_shli_i32(rd, t0, 16);
4615 tcg_gen_andi_i32(tmp, t1, 0xffff);
4616 tcg_gen_or_i32(rd, rd, tmp);
4617 tcg_gen_shri_i32(t1, t1, 16);
4618 tcg_gen_andi_i32(tmp, t0, 0xffff0000);
4619 tcg_gen_or_i32(t1, t1, tmp);
4620 tcg_gen_mov_i32(t0, rd);
4621
4622 tcg_temp_free_i32(tmp);
4623 tcg_temp_free_i32(rd);
4624 }
4625
4626
4627 static struct {
4628 int nregs;
4629 int interleave;
4630 int spacing;
4631 } neon_ls_element_type[11] = {
4632 {4, 4, 1},
4633 {4, 4, 2},
4634 {4, 1, 1},
4635 {4, 2, 1},
4636 {3, 3, 1},
4637 {3, 3, 2},
4638 {3, 1, 1},
4639 {1, 1, 1},
4640 {2, 2, 1},
4641 {2, 2, 2},
4642 {2, 1, 1}
4643 };
4644
4645 /* Translate a NEON load/store element instruction. Return nonzero if the
4646 instruction is invalid. */
4647 static int disas_neon_ls_insn(DisasContext *s, uint32_t insn)
4648 {
4649 int rd, rn, rm;
4650 int op;
4651 int nregs;
4652 int interleave;
4653 int spacing;
4654 int stride;
4655 int size;
4656 int reg;
4657 int pass;
4658 int load;
4659 int shift;
4660 int n;
4661 TCGv_i32 addr;
4662 TCGv_i32 tmp;
4663 TCGv_i32 tmp2;
4664 TCGv_i64 tmp64;
4665
4666 /* FIXME: this access check should not take precedence over UNDEF
4667 * for invalid encodings; we will generate incorrect syndrome information
4668 * for attempts to execute invalid vfp/neon encodings with FP disabled.
4669 */
4670 if (s->fp_excp_el) {
4671 gen_exception_insn(s, 4, EXCP_UDEF,
4672 syn_fp_access_trap(1, 0xe, false), s->fp_excp_el);
4673 return 0;
4674 }
4675
4676 if (!s->vfp_enabled)
4677 return 1;
4678 VFP_DREG_D(rd, insn);
4679 rn = (insn >> 16) & 0xf;
4680 rm = insn & 0xf;
4681 load = (insn & (1 << 21)) != 0;
4682 if ((insn & (1 << 23)) == 0) {
4683 /* Load store all elements. */
4684 op = (insn >> 8) & 0xf;
4685 size = (insn >> 6) & 3;
4686 if (op > 10)
4687 return 1;
4688 /* Catch UNDEF cases for bad values of align field */
4689 switch (op & 0xc) {
4690 case 4:
4691 if (((insn >> 5) & 1) == 1) {
4692 return 1;
4693 }
4694 break;
4695 case 8:
4696 if (((insn >> 4) & 3) == 3) {
4697 return 1;
4698 }
4699 break;
4700 default:
4701 break;
4702 }
4703 nregs = neon_ls_element_type[op].nregs;
4704 interleave = neon_ls_element_type[op].interleave;
4705 spacing = neon_ls_element_type[op].spacing;
4706 if (size == 3 && (interleave | spacing) != 1)
4707 return 1;
4708 addr = tcg_temp_new_i32();
4709 load_reg_var(s, addr, rn);
4710 stride = (1 << size) * interleave;
4711 for (reg = 0; reg < nregs; reg++) {
4712 if (interleave > 2 || (interleave == 2 && nregs == 2)) {
4713 load_reg_var(s, addr, rn);
4714 tcg_gen_addi_i32(addr, addr, (1 << size) * reg);
4715 } else if (interleave == 2 && nregs == 4 && reg == 2) {
4716 load_reg_var(s, addr, rn);
4717 tcg_gen_addi_i32(addr, addr, 1 << size);
4718 }
4719 if (size == 3) {
4720 tmp64 = tcg_temp_new_i64();
4721 if (load) {
4722 gen_aa32_ld64(s, tmp64, addr, get_mem_index(s));
4723 neon_store_reg64(tmp64, rd);
4724 } else {
4725 neon_load_reg64(tmp64, rd);
4726 gen_aa32_st64(s, tmp64, addr, get_mem_index(s));
4727 }
4728 tcg_temp_free_i64(tmp64);
4729 tcg_gen_addi_i32(addr, addr, stride);
4730 } else {
4731 for (pass = 0; pass < 2; pass++) {
4732 if (size == 2) {
4733 if (load) {
4734 tmp = tcg_temp_new_i32();
4735 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
4736 neon_store_reg(rd, pass, tmp);
4737 } else {
4738 tmp = neon_load_reg(rd, pass);
4739 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
4740 tcg_temp_free_i32(tmp);
4741 }
4742 tcg_gen_addi_i32(addr, addr, stride);
4743 } else if (size == 1) {
4744 if (load) {
4745 tmp = tcg_temp_new_i32();
4746 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
4747 tcg_gen_addi_i32(addr, addr, stride);
4748 tmp2 = tcg_temp_new_i32();
4749 gen_aa32_ld16u(s, tmp2, addr, get_mem_index(s));
4750 tcg_gen_addi_i32(addr, addr, stride);
4751 tcg_gen_shli_i32(tmp2, tmp2, 16);
4752 tcg_gen_or_i32(tmp, tmp, tmp2);
4753 tcg_temp_free_i32(tmp2);
4754 neon_store_reg(rd, pass, tmp);
4755 } else {
4756 tmp = neon_load_reg(rd, pass);
4757 tmp2 = tcg_temp_new_i32();
4758 tcg_gen_shri_i32(tmp2, tmp, 16);
4759 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
4760 tcg_temp_free_i32(tmp);
4761 tcg_gen_addi_i32(addr, addr, stride);
4762 gen_aa32_st16(s, tmp2, addr, get_mem_index(s));
4763 tcg_temp_free_i32(tmp2);
4764 tcg_gen_addi_i32(addr, addr, stride);
4765 }
4766 } else /* size == 0 */ {
4767 if (load) {
4768 TCGV_UNUSED_I32(tmp2);
4769 for (n = 0; n < 4; n++) {
4770 tmp = tcg_temp_new_i32();
4771 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
4772 tcg_gen_addi_i32(addr, addr, stride);
4773 if (n == 0) {
4774 tmp2 = tmp;
4775 } else {
4776 tcg_gen_shli_i32(tmp, tmp, n * 8);
4777 tcg_gen_or_i32(tmp2, tmp2, tmp);
4778 tcg_temp_free_i32(tmp);
4779 }
4780 }
4781 neon_store_reg(rd, pass, tmp2);
4782 } else {
4783 tmp2 = neon_load_reg(rd, pass);
4784 for (n = 0; n < 4; n++) {
4785 tmp = tcg_temp_new_i32();
4786 if (n == 0) {
4787 tcg_gen_mov_i32(tmp, tmp2);
4788 } else {
4789 tcg_gen_shri_i32(tmp, tmp2, n * 8);
4790 }
4791 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
4792 tcg_temp_free_i32(tmp);
4793 tcg_gen_addi_i32(addr, addr, stride);
4794 }
4795 tcg_temp_free_i32(tmp2);
4796 }
4797 }
4798 }
4799 }
4800 rd += spacing;
4801 }
4802 tcg_temp_free_i32(addr);
4803 stride = nregs * 8;
4804 } else {
4805 size = (insn >> 10) & 3;
4806 if (size == 3) {
4807 /* Load single element to all lanes. */
4808 int a = (insn >> 4) & 1;
4809 if (!load) {
4810 return 1;
4811 }
4812 size = (insn >> 6) & 3;
4813 nregs = ((insn >> 8) & 3) + 1;
4814
4815 if (size == 3) {
4816 if (nregs != 4 || a == 0) {
4817 return 1;
4818 }
4819 /* For VLD4 size==3 a == 1 means 32 bits at 16 byte alignment */
4820 size = 2;
4821 }
4822 if (nregs == 1 && a == 1 && size == 0) {
4823 return 1;
4824 }
4825 if (nregs == 3 && a == 1) {
4826 return 1;
4827 }
4828 addr = tcg_temp_new_i32();
4829 load_reg_var(s, addr, rn);
4830 if (nregs == 1) {
4831 /* VLD1 to all lanes: bit 5 indicates how many Dregs to write */
4832 tmp = gen_load_and_replicate(s, addr, size);
4833 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 0));
4834 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 1));
4835 if (insn & (1 << 5)) {
4836 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd + 1, 0));
4837 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd + 1, 1));
4838 }
4839 tcg_temp_free_i32(tmp);
4840 } else {
4841 /* VLD2/3/4 to all lanes: bit 5 indicates register stride */
4842 stride = (insn & (1 << 5)) ? 2 : 1;
4843 for (reg = 0; reg < nregs; reg++) {
4844 tmp = gen_load_and_replicate(s, addr, size);
4845 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 0));
4846 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 1));
4847 tcg_temp_free_i32(tmp);
4848 tcg_gen_addi_i32(addr, addr, 1 << size);
4849 rd += stride;
4850 }
4851 }
4852 tcg_temp_free_i32(addr);
4853 stride = (1 << size) * nregs;
4854 } else {
4855 /* Single element. */
4856 int idx = (insn >> 4) & 0xf;
4857 pass = (insn >> 7) & 1;
4858 switch (size) {
4859 case 0:
4860 shift = ((insn >> 5) & 3) * 8;
4861 stride = 1;
4862 break;
4863 case 1:
4864 shift = ((insn >> 6) & 1) * 16;
4865 stride = (insn & (1 << 5)) ? 2 : 1;
4866 break;
4867 case 2:
4868 shift = 0;
4869 stride = (insn & (1 << 6)) ? 2 : 1;
4870 break;
4871 default:
4872 abort();
4873 }
4874 nregs = ((insn >> 8) & 3) + 1;
4875 /* Catch the UNDEF cases. This is unavoidably a bit messy. */
4876 switch (nregs) {
4877 case 1:
4878 if (((idx & (1 << size)) != 0) ||
4879 (size == 2 && ((idx & 3) == 1 || (idx & 3) == 2))) {
4880 return 1;
4881 }
4882 break;
4883 case 3:
4884 if ((idx & 1) != 0) {
4885 return 1;
4886 }
4887 /* fall through */
4888 case 2:
4889 if (size == 2 && (idx & 2) != 0) {
4890 return 1;
4891 }
4892 break;
4893 case 4:
4894 if ((size == 2) && ((idx & 3) == 3)) {
4895 return 1;
4896 }
4897 break;
4898 default:
4899 abort();
4900 }
4901 if ((rd + stride * (nregs - 1)) > 31) {
4902 /* Attempts to write off the end of the register file
4903 * are UNPREDICTABLE; we choose to UNDEF because otherwise
4904 * the neon_load_reg() would write off the end of the array.
4905 */
4906 return 1;
4907 }
4908 addr = tcg_temp_new_i32();
4909 load_reg_var(s, addr, rn);
4910 for (reg = 0; reg < nregs; reg++) {
4911 if (load) {
4912 tmp = tcg_temp_new_i32();
4913 switch (size) {
4914 case 0:
4915 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
4916 break;
4917 case 1:
4918 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
4919 break;
4920 case 2:
4921 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
4922 break;
4923 default: /* Avoid compiler warnings. */
4924 abort();
4925 }
4926 if (size != 2) {
4927 tmp2 = neon_load_reg(rd, pass);
4928 tcg_gen_deposit_i32(tmp, tmp2, tmp,
4929 shift, size ? 16 : 8);
4930 tcg_temp_free_i32(tmp2);
4931 }
4932 neon_store_reg(rd, pass, tmp);
4933 } else { /* Store */
4934 tmp = neon_load_reg(rd, pass);
4935 if (shift)
4936 tcg_gen_shri_i32(tmp, tmp, shift);
4937 switch (size) {
4938 case 0:
4939 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
4940 break;
4941 case 1:
4942 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
4943 break;
4944 case 2:
4945 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
4946 break;
4947 }
4948 tcg_temp_free_i32(tmp);
4949 }
4950 rd += stride;
4951 tcg_gen_addi_i32(addr, addr, 1 << size);
4952 }
4953 tcg_temp_free_i32(addr);
4954 stride = nregs * (1 << size);
4955 }
4956 }
4957 if (rm != 15) {
4958 TCGv_i32 base;
4959
4960 base = load_reg(s, rn);
4961 if (rm == 13) {
4962 tcg_gen_addi_i32(base, base, stride);
4963 } else {
4964 TCGv_i32 index;
4965 index = load_reg(s, rm);
4966 tcg_gen_add_i32(base, base, index);
4967 tcg_temp_free_i32(index);
4968 }
4969 store_reg(s, rn, base);
4970 }
4971 return 0;
4972 }
4973
4974 /* Bitwise select. dest = c ? t : f. Clobbers T and F. */
4975 static void gen_neon_bsl(TCGv_i32 dest, TCGv_i32 t, TCGv_i32 f, TCGv_i32 c)
4976 {
4977 tcg_gen_and_i32(t, t, c);
4978 tcg_gen_andc_i32(f, f, c);
4979 tcg_gen_or_i32(dest, t, f);
4980 }
4981
4982 static inline void gen_neon_narrow(int size, TCGv_i32 dest, TCGv_i64 src)
4983 {
4984 switch (size) {
4985 case 0: gen_helper_neon_narrow_u8(dest, src); break;
4986 case 1: gen_helper_neon_narrow_u16(dest, src); break;
4987 case 2: tcg_gen_extrl_i64_i32(dest, src); break;
4988 default: abort();
4989 }
4990 }
4991
4992 static inline void gen_neon_narrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
4993 {
4994 switch (size) {
4995 case 0: gen_helper_neon_narrow_sat_s8(dest, cpu_env, src); break;
4996 case 1: gen_helper_neon_narrow_sat_s16(dest, cpu_env, src); break;
4997 case 2: gen_helper_neon_narrow_sat_s32(dest, cpu_env, src); break;
4998 default: abort();
4999 }
5000 }
5001
5002 static inline void gen_neon_narrow_satu(int size, TCGv_i32 dest, TCGv_i64 src)
5003 {
5004 switch (size) {
5005 case 0: gen_helper_neon_narrow_sat_u8(dest, cpu_env, src); break;
5006 case 1: gen_helper_neon_narrow_sat_u16(dest, cpu_env, src); break;
5007 case 2: gen_helper_neon_narrow_sat_u32(dest, cpu_env, src); break;
5008 default: abort();
5009 }
5010 }
5011
5012 static inline void gen_neon_unarrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
5013 {
5014 switch (size) {
5015 case 0: gen_helper_neon_unarrow_sat8(dest, cpu_env, src); break;
5016 case 1: gen_helper_neon_unarrow_sat16(dest, cpu_env, src); break;
5017 case 2: gen_helper_neon_unarrow_sat32(dest, cpu_env, src); break;
5018 default: abort();
5019 }
5020 }
5021
5022 static inline void gen_neon_shift_narrow(int size, TCGv_i32 var, TCGv_i32 shift,
5023 int q, int u)
5024 {
5025 if (q) {
5026 if (u) {
5027 switch (size) {
5028 case 1: gen_helper_neon_rshl_u16(var, var, shift); break;
5029 case 2: gen_helper_neon_rshl_u32(var, var, shift); break;
5030 default: abort();
5031 }
5032 } else {
5033 switch (size) {
5034 case 1: gen_helper_neon_rshl_s16(var, var, shift); break;
5035 case 2: gen_helper_neon_rshl_s32(var, var, shift); break;
5036 default: abort();
5037 }
5038 }
5039 } else {
5040 if (u) {
5041 switch (size) {
5042 case 1: gen_helper_neon_shl_u16(var, var, shift); break;
5043 case 2: gen_helper_neon_shl_u32(var, var, shift); break;
5044 default: abort();
5045 }
5046 } else {
5047 switch (size) {
5048 case 1: gen_helper_neon_shl_s16(var, var, shift); break;
5049 case 2: gen_helper_neon_shl_s32(var, var, shift); break;
5050 default: abort();
5051 }
5052 }
5053 }
5054 }
5055
5056 static inline void gen_neon_widen(TCGv_i64 dest, TCGv_i32 src, int size, int u)
5057 {
5058 if (u) {
5059 switch (size) {
5060 case 0: gen_helper_neon_widen_u8(dest, src); break;
5061 case 1: gen_helper_neon_widen_u16(dest, src); break;
5062 case 2: tcg_gen_extu_i32_i64(dest, src); break;
5063 default: abort();
5064 }
5065 } else {
5066 switch (size) {
5067 case 0: gen_helper_neon_widen_s8(dest, src); break;
5068 case 1: gen_helper_neon_widen_s16(dest, src); break;
5069 case 2: tcg_gen_ext_i32_i64(dest, src); break;
5070 default: abort();
5071 }
5072 }
5073 tcg_temp_free_i32(src);
5074 }
5075
5076 static inline void gen_neon_addl(int size)
5077 {
5078 switch (size) {
5079 case 0: gen_helper_neon_addl_u16(CPU_V001); break;
5080 case 1: gen_helper_neon_addl_u32(CPU_V001); break;
5081 case 2: tcg_gen_add_i64(CPU_V001); break;
5082 default: abort();
5083 }
5084 }
5085
5086 static inline void gen_neon_subl(int size)
5087 {
5088 switch (size) {
5089 case 0: gen_helper_neon_subl_u16(CPU_V001); break;
5090 case 1: gen_helper_neon_subl_u32(CPU_V001); break;
5091 case 2: tcg_gen_sub_i64(CPU_V001); break;
5092 default: abort();
5093 }
5094 }
5095
5096 static inline void gen_neon_negl(TCGv_i64 var, int size)
5097 {
5098 switch (size) {
5099 case 0: gen_helper_neon_negl_u16(var, var); break;
5100 case 1: gen_helper_neon_negl_u32(var, var); break;
5101 case 2:
5102 tcg_gen_neg_i64(var, var);
5103 break;
5104 default: abort();
5105 }
5106 }
5107
5108 static inline void gen_neon_addl_saturate(TCGv_i64 op0, TCGv_i64 op1, int size)
5109 {
5110 switch (size) {
5111 case 1: gen_helper_neon_addl_saturate_s32(op0, cpu_env, op0, op1); break;
5112 case 2: gen_helper_neon_addl_saturate_s64(op0, cpu_env, op0, op1); break;
5113 default: abort();
5114 }
5115 }
5116
5117 static inline void gen_neon_mull(TCGv_i64 dest, TCGv_i32 a, TCGv_i32 b,
5118 int size, int u)
5119 {
5120 TCGv_i64 tmp;
5121
5122 switch ((size << 1) | u) {
5123 case 0: gen_helper_neon_mull_s8(dest, a, b); break;
5124 case 1: gen_helper_neon_mull_u8(dest, a, b); break;
5125 case 2: gen_helper_neon_mull_s16(dest, a, b); break;
5126 case 3: gen_helper_neon_mull_u16(dest, a, b); break;
5127 case 4:
5128 tmp = gen_muls_i64_i32(a, b);
5129 tcg_gen_mov_i64(dest, tmp);
5130 tcg_temp_free_i64(tmp);
5131 break;
5132 case 5:
5133 tmp = gen_mulu_i64_i32(a, b);
5134 tcg_gen_mov_i64(dest, tmp);
5135 tcg_temp_free_i64(tmp);
5136 break;
5137 default: abort();
5138 }
5139
5140 /* gen_helper_neon_mull_[su]{8|16} do not free their parameters.
5141 Don't forget to clean them now. */
5142 if (size < 2) {
5143 tcg_temp_free_i32(a);
5144 tcg_temp_free_i32(b);
5145 }
5146 }
5147
5148 static void gen_neon_narrow_op(int op, int u, int size,
5149 TCGv_i32 dest, TCGv_i64 src)
5150 {
5151 if (op) {
5152 if (u) {
5153 gen_neon_unarrow_sats(size, dest, src);
5154 } else {
5155 gen_neon_narrow(size, dest, src);
5156 }
5157 } else {
5158 if (u) {
5159 gen_neon_narrow_satu(size, dest, src);
5160 } else {
5161 gen_neon_narrow_sats(size, dest, src);
5162 }
5163 }
5164 }
5165
5166 /* Symbolic constants for op fields for Neon 3-register same-length.
5167 * The values correspond to bits [11:8,4]; see the ARM ARM DDI0406B
5168 * table A7-9.
5169 */
5170 #define NEON_3R_VHADD 0
5171 #define NEON_3R_VQADD 1
5172 #define NEON_3R_VRHADD 2
5173 #define NEON_3R_LOGIC 3 /* VAND,VBIC,VORR,VMOV,VORN,VEOR,VBIF,VBIT,VBSL */
5174 #define NEON_3R_VHSUB 4
5175 #define NEON_3R_VQSUB 5
5176 #define NEON_3R_VCGT 6
5177 #define NEON_3R_VCGE 7
5178 #define NEON_3R_VSHL 8
5179 #define NEON_3R_VQSHL 9
5180 #define NEON_3R_VRSHL 10
5181 #define NEON_3R_VQRSHL 11
5182 #define NEON_3R_VMAX 12
5183 #define NEON_3R_VMIN 13
5184 #define NEON_3R_VABD 14
5185 #define NEON_3R_VABA 15
5186 #define NEON_3R_VADD_VSUB 16
5187 #define NEON_3R_VTST_VCEQ 17
5188 #define NEON_3R_VML 18 /* VMLA, VMLAL, VMLS, VMLSL */
5189 #define NEON_3R_VMUL 19
5190 #define NEON_3R_VPMAX 20
5191 #define NEON_3R_VPMIN 21
5192 #define NEON_3R_VQDMULH_VQRDMULH 22
5193 #define NEON_3R_VPADD 23
5194 #define NEON_3R_SHA 24 /* SHA1C,SHA1P,SHA1M,SHA1SU0,SHA256H{2},SHA256SU1 */
5195 #define NEON_3R_VFM 25 /* VFMA, VFMS : float fused multiply-add */
5196 #define NEON_3R_FLOAT_ARITH 26 /* float VADD, VSUB, VPADD, VABD */
5197 #define NEON_3R_FLOAT_MULTIPLY 27 /* float VMLA, VMLS, VMUL */
5198 #define NEON_3R_FLOAT_CMP 28 /* float VCEQ, VCGE, VCGT */
5199 #define NEON_3R_FLOAT_ACMP 29 /* float VACGE, VACGT, VACLE, VACLT */
5200 #define NEON_3R_FLOAT_MINMAX 30 /* float VMIN, VMAX */
5201 #define NEON_3R_FLOAT_MISC 31 /* float VRECPS, VRSQRTS, VMAXNM/MINNM */
5202
5203 static const uint8_t neon_3r_sizes[] = {
5204 [NEON_3R_VHADD] = 0x7,
5205 [NEON_3R_VQADD] = 0xf,
5206 [NEON_3R_VRHADD] = 0x7,
5207 [NEON_3R_LOGIC] = 0xf, /* size field encodes op type */
5208 [NEON_3R_VHSUB] = 0x7,
5209 [NEON_3R_VQSUB] = 0xf,
5210 [NEON_3R_VCGT] = 0x7,
5211 [NEON_3R_VCGE] = 0x7,
5212 [NEON_3R_VSHL] = 0xf,
5213 [NEON_3R_VQSHL] = 0xf,
5214 [NEON_3R_VRSHL] = 0xf,
5215 [NEON_3R_VQRSHL] = 0xf,
5216 [NEON_3R_VMAX] = 0x7,
5217 [NEON_3R_VMIN] = 0x7,
5218 [NEON_3R_VABD] = 0x7,
5219 [NEON_3R_VABA] = 0x7,
5220 [NEON_3R_VADD_VSUB] = 0xf,
5221 [NEON_3R_VTST_VCEQ] = 0x7,
5222 [NEON_3R_VML] = 0x7,
5223 [NEON_3R_VMUL] = 0x7,
5224 [NEON_3R_VPMAX] = 0x7,
5225 [NEON_3R_VPMIN] = 0x7,
5226 [NEON_3R_VQDMULH_VQRDMULH] = 0x6,
5227 [NEON_3R_VPADD] = 0x7,
5228 [NEON_3R_SHA] = 0xf, /* size field encodes op type */
5229 [NEON_3R_VFM] = 0x5, /* size bit 1 encodes op */
5230 [NEON_3R_FLOAT_ARITH] = 0x5, /* size bit 1 encodes op */
5231 [NEON_3R_FLOAT_MULTIPLY] = 0x5, /* size bit 1 encodes op */
5232 [NEON_3R_FLOAT_CMP] = 0x5, /* size bit 1 encodes op */
5233 [NEON_3R_FLOAT_ACMP] = 0x5, /* size bit 1 encodes op */
5234 [NEON_3R_FLOAT_MINMAX] = 0x5, /* size bit 1 encodes op */
5235 [NEON_3R_FLOAT_MISC] = 0x5, /* size bit 1 encodes op */
5236 };
5237
5238 /* Symbolic constants for op fields for Neon 2-register miscellaneous.
5239 * The values correspond to bits [17:16,10:7]; see the ARM ARM DDI0406B
5240 * table A7-13.
5241 */
5242 #define NEON_2RM_VREV64 0
5243 #define NEON_2RM_VREV32 1
5244 #define NEON_2RM_VREV16 2
5245 #define NEON_2RM_VPADDL 4
5246 #define NEON_2RM_VPADDL_U 5
5247 #define NEON_2RM_AESE 6 /* Includes AESD */
5248 #define NEON_2RM_AESMC 7 /* Includes AESIMC */
5249 #define NEON_2RM_VCLS 8
5250 #define NEON_2RM_VCLZ 9
5251 #define NEON_2RM_VCNT 10
5252 #define NEON_2RM_VMVN 11
5253 #define NEON_2RM_VPADAL 12
5254 #define NEON_2RM_VPADAL_U 13
5255 #define NEON_2RM_VQABS 14
5256 #define NEON_2RM_VQNEG 15
5257 #define NEON_2RM_VCGT0 16
5258 #define NEON_2RM_VCGE0 17
5259 #define NEON_2RM_VCEQ0 18
5260 #define NEON_2RM_VCLE0 19
5261 #define NEON_2RM_VCLT0 20
5262 #define NEON_2RM_SHA1H 21
5263 #define NEON_2RM_VABS 22
5264 #define NEON_2RM_VNEG 23
5265 #define NEON_2RM_VCGT0_F 24
5266 #define NEON_2RM_VCGE0_F 25
5267 #define NEON_2RM_VCEQ0_F 26
5268 #define NEON_2RM_VCLE0_F 27
5269 #define NEON_2RM_VCLT0_F 28
5270 #define NEON_2RM_VABS_F 30
5271 #define NEON_2RM_VNEG_F 31
5272 #define NEON_2RM_VSWP 32
5273 #define NEON_2RM_VTRN 33
5274 #define NEON_2RM_VUZP 34
5275 #define NEON_2RM_VZIP 35
5276 #define NEON_2RM_VMOVN 36 /* Includes VQMOVN, VQMOVUN */
5277 #define NEON_2RM_VQMOVN 37 /* Includes VQMOVUN */
5278 #define NEON_2RM_VSHLL 38
5279 #define NEON_2RM_SHA1SU1 39 /* Includes SHA256SU0 */
5280 #define NEON_2RM_VRINTN 40
5281 #define NEON_2RM_VRINTX 41
5282 #define NEON_2RM_VRINTA 42
5283 #define NEON_2RM_VRINTZ 43
5284 #define NEON_2RM_VCVT_F16_F32 44
5285 #define NEON_2RM_VRINTM 45
5286 #define NEON_2RM_VCVT_F32_F16 46
5287 #define NEON_2RM_VRINTP 47
5288 #define NEON_2RM_VCVTAU 48
5289 #define NEON_2RM_VCVTAS 49
5290 #define NEON_2RM_VCVTNU 50
5291 #define NEON_2RM_VCVTNS 51
5292 #define NEON_2RM_VCVTPU 52
5293 #define NEON_2RM_VCVTPS 53
5294 #define NEON_2RM_VCVTMU 54
5295 #define NEON_2RM_VCVTMS 55
5296 #define NEON_2RM_VRECPE 56
5297 #define NEON_2RM_VRSQRTE 57
5298 #define NEON_2RM_VRECPE_F 58
5299 #define NEON_2RM_VRSQRTE_F 59
5300 #define NEON_2RM_VCVT_FS 60
5301 #define NEON_2RM_VCVT_FU 61
5302 #define NEON_2RM_VCVT_SF 62
5303 #define NEON_2RM_VCVT_UF 63
5304
5305 static int neon_2rm_is_float_op(int op)
5306 {
5307 /* Return true if this neon 2reg-misc op is float-to-float */
5308 return (op == NEON_2RM_VABS_F || op == NEON_2RM_VNEG_F ||
5309 (op >= NEON_2RM_VRINTN && op <= NEON_2RM_VRINTZ) ||
5310 op == NEON_2RM_VRINTM ||
5311 (op >= NEON_2RM_VRINTP && op <= NEON_2RM_VCVTMS) ||
5312 op >= NEON_2RM_VRECPE_F);
5313 }
5314
5315 static bool neon_2rm_is_v8_op(int op)
5316 {
5317 /* Return true if this neon 2reg-misc op is ARMv8 and up */
5318 switch (op) {
5319 case NEON_2RM_VRINTN:
5320 case NEON_2RM_VRINTA:
5321 case NEON_2RM_VRINTM:
5322 case NEON_2RM_VRINTP:
5323 case NEON_2RM_VRINTZ:
5324 case NEON_2RM_VRINTX:
5325 case NEON_2RM_VCVTAU:
5326 case NEON_2RM_VCVTAS:
5327 case NEON_2RM_VCVTNU:
5328 case NEON_2RM_VCVTNS:
5329 case NEON_2RM_VCVTPU:
5330 case NEON_2RM_VCVTPS:
5331 case NEON_2RM_VCVTMU:
5332 case NEON_2RM_VCVTMS:
5333 return true;
5334 default:
5335 return false;
5336 }
5337 }
5338
5339 /* Each entry in this array has bit n set if the insn allows
5340 * size value n (otherwise it will UNDEF). Since unallocated
5341 * op values will have no bits set they always UNDEF.
5342 */
5343 static const uint8_t neon_2rm_sizes[] = {
5344 [NEON_2RM_VREV64] = 0x7,
5345 [NEON_2RM_VREV32] = 0x3,
5346 [NEON_2RM_VREV16] = 0x1,
5347 [NEON_2RM_VPADDL] = 0x7,
5348 [NEON_2RM_VPADDL_U] = 0x7,
5349 [NEON_2RM_AESE] = 0x1,
5350 [NEON_2RM_AESMC] = 0x1,
5351 [NEON_2RM_VCLS] = 0x7,
5352 [NEON_2RM_VCLZ] = 0x7,
5353 [NEON_2RM_VCNT] = 0x1,
5354 [NEON_2RM_VMVN] = 0x1,
5355 [NEON_2RM_VPADAL] = 0x7,
5356 [NEON_2RM_VPADAL_U] = 0x7,
5357 [NEON_2RM_VQABS] = 0x7,
5358 [NEON_2RM_VQNEG] = 0x7,
5359 [NEON_2RM_VCGT0] = 0x7,
5360 [NEON_2RM_VCGE0] = 0x7,
5361 [NEON_2RM_VCEQ0] = 0x7,
5362 [NEON_2RM_VCLE0] = 0x7,
5363 [NEON_2RM_VCLT0] = 0x7,
5364 [NEON_2RM_SHA1H] = 0x4,
5365 [NEON_2RM_VABS] = 0x7,
5366 [NEON_2RM_VNEG] = 0x7,
5367 [NEON_2RM_VCGT0_F] = 0x4,
5368 [NEON_2RM_VCGE0_F] = 0x4,
5369 [NEON_2RM_VCEQ0_F] = 0x4,
5370 [NEON_2RM_VCLE0_F] = 0x4,
5371 [NEON_2RM_VCLT0_F] = 0x4,
5372 [NEON_2RM_VABS_F] = 0x4,
5373 [NEON_2RM_VNEG_F] = 0x4,
5374 [NEON_2RM_VSWP] = 0x1,
5375 [NEON_2RM_VTRN] = 0x7,
5376 [NEON_2RM_VUZP] = 0x7,
5377 [NEON_2RM_VZIP] = 0x7,
5378 [NEON_2RM_VMOVN] = 0x7,
5379 [NEON_2RM_VQMOVN] = 0x7,
5380 [NEON_2RM_VSHLL] = 0x7,
5381 [NEON_2RM_SHA1SU1] = 0x4,
5382 [NEON_2RM_VRINTN] = 0x4,
5383 [NEON_2RM_VRINTX] = 0x4,
5384 [NEON_2RM_VRINTA] = 0x4,
5385 [NEON_2RM_VRINTZ] = 0x4,
5386 [NEON_2RM_VCVT_F16_F32] = 0x2,
5387 [NEON_2RM_VRINTM] = 0x4,
5388 [NEON_2RM_VCVT_F32_F16] = 0x2,
5389 [NEON_2RM_VRINTP] = 0x4,
5390 [NEON_2RM_VCVTAU] = 0x4,
5391 [NEON_2RM_VCVTAS] = 0x4,
5392 [NEON_2RM_VCVTNU] = 0x4,
5393 [NEON_2RM_VCVTNS] = 0x4,
5394 [NEON_2RM_VCVTPU] = 0x4,
5395 [NEON_2RM_VCVTPS] = 0x4,
5396 [NEON_2RM_VCVTMU] = 0x4,
5397 [NEON_2RM_VCVTMS] = 0x4,
5398 [NEON_2RM_VRECPE] = 0x4,
5399 [NEON_2RM_VRSQRTE] = 0x4,
5400 [NEON_2RM_VRECPE_F] = 0x4,
5401 [NEON_2RM_VRSQRTE_F] = 0x4,
5402 [NEON_2RM_VCVT_FS] = 0x4,
5403 [NEON_2RM_VCVT_FU] = 0x4,
5404 [NEON_2RM_VCVT_SF] = 0x4,
5405 [NEON_2RM_VCVT_UF] = 0x4,
5406 };
5407
5408 /* Translate a NEON data processing instruction. Return nonzero if the
5409 instruction is invalid.
5410 We process data in a mixture of 32-bit and 64-bit chunks.
5411 Mostly we use 32-bit chunks so we can use normal scalar instructions. */
5412
5413 static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
5414 {
5415 int op;
5416 int q;
5417 int rd, rn, rm;
5418 int size;
5419 int shift;
5420 int pass;
5421 int count;
5422 int pairwise;
5423 int u;
5424 uint32_t imm, mask;
5425 TCGv_i32 tmp, tmp2, tmp3, tmp4, tmp5;
5426 TCGv_i64 tmp64;
5427
5428 /* FIXME: this access check should not take precedence over UNDEF
5429 * for invalid encodings; we will generate incorrect syndrome information
5430 * for attempts to execute invalid vfp/neon encodings with FP disabled.
5431 */
5432 if (s->fp_excp_el) {
5433 gen_exception_insn(s, 4, EXCP_UDEF,
5434 syn_fp_access_trap(1, 0xe, false), s->fp_excp_el);
5435 return 0;
5436 }
5437
5438 if (!s->vfp_enabled)
5439 return 1;
5440 q = (insn & (1 << 6)) != 0;
5441 u = (insn >> 24) & 1;
5442 VFP_DREG_D(rd, insn);
5443 VFP_DREG_N(rn, insn);
5444 VFP_DREG_M(rm, insn);
5445 size = (insn >> 20) & 3;
5446 if ((insn & (1 << 23)) == 0) {
5447 /* Three register same length. */
5448 op = ((insn >> 7) & 0x1e) | ((insn >> 4) & 1);
5449 /* Catch invalid op and bad size combinations: UNDEF */
5450 if ((neon_3r_sizes[op] & (1 << size)) == 0) {
5451 return 1;
5452 }
5453 /* All insns of this form UNDEF for either this condition or the
5454 * superset of cases "Q==1"; we catch the latter later.
5455 */
5456 if (q && ((rd | rn | rm) & 1)) {
5457 return 1;
5458 }
5459 /*
5460 * The SHA-1/SHA-256 3-register instructions require special treatment
5461 * here, as their size field is overloaded as an op type selector, and
5462 * they all consume their input in a single pass.
5463 */
5464 if (op == NEON_3R_SHA) {
5465 if (!q) {
5466 return 1;
5467 }
5468 if (!u) { /* SHA-1 */
5469 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA1)) {
5470 return 1;
5471 }
5472 tmp = tcg_const_i32(rd);
5473 tmp2 = tcg_const_i32(rn);
5474 tmp3 = tcg_const_i32(rm);
5475 tmp4 = tcg_const_i32(size);
5476 gen_helper_crypto_sha1_3reg(cpu_env, tmp, tmp2, tmp3, tmp4);
5477 tcg_temp_free_i32(tmp4);
5478 } else { /* SHA-256 */
5479 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA256) || size == 3) {
5480 return 1;
5481 }
5482 tmp = tcg_const_i32(rd);
5483 tmp2 = tcg_const_i32(rn);
5484 tmp3 = tcg_const_i32(rm);
5485 switch (size) {
5486 case 0:
5487 gen_helper_crypto_sha256h(cpu_env, tmp, tmp2, tmp3);
5488 break;
5489 case 1:
5490 gen_helper_crypto_sha256h2(cpu_env, tmp, tmp2, tmp3);
5491 break;
5492 case 2:
5493 gen_helper_crypto_sha256su1(cpu_env, tmp, tmp2, tmp3);
5494 break;
5495 }
5496 }
5497 tcg_temp_free_i32(tmp);
5498 tcg_temp_free_i32(tmp2);
5499 tcg_temp_free_i32(tmp3);
5500 return 0;
5501 }
5502 if (size == 3 && op != NEON_3R_LOGIC) {
5503 /* 64-bit element instructions. */
5504 for (pass = 0; pass < (q ? 2 : 1); pass++) {
5505 neon_load_reg64(cpu_V0, rn + pass);
5506 neon_load_reg64(cpu_V1, rm + pass);
5507 switch (op) {
5508 case NEON_3R_VQADD:
5509 if (u) {
5510 gen_helper_neon_qadd_u64(cpu_V0, cpu_env,
5511 cpu_V0, cpu_V1);
5512 } else {
5513 gen_helper_neon_qadd_s64(cpu_V0, cpu_env,
5514 cpu_V0, cpu_V1);
5515 }
5516 break;
5517 case NEON_3R_VQSUB:
5518 if (u) {
5519 gen_helper_neon_qsub_u64(cpu_V0, cpu_env,
5520 cpu_V0, cpu_V1);
5521 } else {
5522 gen_helper_neon_qsub_s64(cpu_V0, cpu_env,
5523 cpu_V0, cpu_V1);
5524 }
5525 break;
5526 case NEON_3R_VSHL:
5527 if (u) {
5528 gen_helper_neon_shl_u64(cpu_V0, cpu_V1, cpu_V0);
5529 } else {
5530 gen_helper_neon_shl_s64(cpu_V0, cpu_V1, cpu_V0);
5531 }
5532 break;
5533 case NEON_3R_VQSHL:
5534 if (u) {
5535 gen_helper_neon_qshl_u64(cpu_V0, cpu_env,
5536 cpu_V1, cpu_V0);
5537 } else {
5538 gen_helper_neon_qshl_s64(cpu_V0, cpu_env,
5539 cpu_V1, cpu_V0);
5540 }
5541 break;
5542 case NEON_3R_VRSHL:
5543 if (u) {
5544 gen_helper_neon_rshl_u64(cpu_V0, cpu_V1, cpu_V0);
5545 } else {
5546 gen_helper_neon_rshl_s64(cpu_V0, cpu_V1, cpu_V0);
5547 }
5548 break;
5549 case NEON_3R_VQRSHL:
5550 if (u) {
5551 gen_helper_neon_qrshl_u64(cpu_V0, cpu_env,
5552 cpu_V1, cpu_V0);
5553 } else {
5554 gen_helper_neon_qrshl_s64(cpu_V0, cpu_env,
5555 cpu_V1, cpu_V0);
5556 }
5557 break;
5558 case NEON_3R_VADD_VSUB:
5559 if (u) {
5560 tcg_gen_sub_i64(CPU_V001);
5561 } else {
5562 tcg_gen_add_i64(CPU_V001);
5563 }
5564 break;
5565 default:
5566 abort();
5567 }
5568 neon_store_reg64(cpu_V0, rd + pass);
5569 }
5570 return 0;
5571 }
5572 pairwise = 0;
5573 switch (op) {
5574 case NEON_3R_VSHL:
5575 case NEON_3R_VQSHL:
5576 case NEON_3R_VRSHL:
5577 case NEON_3R_VQRSHL:
5578 {
5579 int rtmp;
5580 /* Shift instruction operands are reversed. */
5581 rtmp = rn;
5582 rn = rm;
5583 rm = rtmp;
5584 }
5585 break;
5586 case NEON_3R_VPADD:
5587 if (u) {
5588 return 1;
5589 }
5590 /* Fall through */
5591 case NEON_3R_VPMAX:
5592 case NEON_3R_VPMIN:
5593 pairwise = 1;
5594 break;
5595 case NEON_3R_FLOAT_ARITH:
5596 pairwise = (u && size < 2); /* if VPADD (float) */
5597 break;
5598 case NEON_3R_FLOAT_MINMAX:
5599 pairwise = u; /* if VPMIN/VPMAX (float) */
5600 break;
5601 case NEON_3R_FLOAT_CMP:
5602 if (!u && size) {
5603 /* no encoding for U=0 C=1x */
5604 return 1;
5605 }
5606 break;
5607 case NEON_3R_FLOAT_ACMP:
5608 if (!u) {
5609 return 1;
5610 }
5611 break;
5612 case NEON_3R_FLOAT_MISC:
5613 /* VMAXNM/VMINNM in ARMv8 */
5614 if (u && !arm_dc_feature(s, ARM_FEATURE_V8)) {
5615 return 1;
5616 }
5617 break;
5618 case NEON_3R_VMUL:
5619 if (u && (size != 0)) {
5620 /* UNDEF on invalid size for polynomial subcase */
5621 return 1;
5622 }
5623 break;
5624 case NEON_3R_VFM:
5625 if (!arm_dc_feature(s, ARM_FEATURE_VFP4) || u) {
5626 return 1;
5627 }
5628 break;
5629 default:
5630 break;
5631 }
5632
5633 if (pairwise && q) {
5634 /* All the pairwise insns UNDEF if Q is set */
5635 return 1;
5636 }
5637
5638 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5639
5640 if (pairwise) {
5641 /* Pairwise. */
5642 if (pass < 1) {
5643 tmp = neon_load_reg(rn, 0);
5644 tmp2 = neon_load_reg(rn, 1);
5645 } else {
5646 tmp = neon_load_reg(rm, 0);
5647 tmp2 = neon_load_reg(rm, 1);
5648 }
5649 } else {
5650 /* Elementwise. */
5651 tmp = neon_load_reg(rn, pass);
5652 tmp2 = neon_load_reg(rm, pass);
5653 }
5654 switch (op) {
5655 case NEON_3R_VHADD:
5656 GEN_NEON_INTEGER_OP(hadd);
5657 break;
5658 case NEON_3R_VQADD:
5659 GEN_NEON_INTEGER_OP_ENV(qadd);
5660 break;
5661 case NEON_3R_VRHADD:
5662 GEN_NEON_INTEGER_OP(rhadd);
5663 break;
5664 case NEON_3R_LOGIC: /* Logic ops. */
5665 switch ((u << 2) | size) {
5666 case 0: /* VAND */
5667 tcg_gen_and_i32(tmp, tmp, tmp2);
5668 break;
5669 case 1: /* BIC */
5670 tcg_gen_andc_i32(tmp, tmp, tmp2);
5671 break;
5672 case 2: /* VORR */
5673 tcg_gen_or_i32(tmp, tmp, tmp2);
5674 break;
5675 case 3: /* VORN */
5676 tcg_gen_orc_i32(tmp, tmp, tmp2);
5677 break;
5678 case 4: /* VEOR */
5679 tcg_gen_xor_i32(tmp, tmp, tmp2);
5680 break;
5681 case 5: /* VBSL */
5682 tmp3 = neon_load_reg(rd, pass);
5683 gen_neon_bsl(tmp, tmp, tmp2, tmp3);
5684 tcg_temp_free_i32(tmp3);
5685 break;
5686 case 6: /* VBIT */
5687 tmp3 = neon_load_reg(rd, pass);
5688 gen_neon_bsl(tmp, tmp, tmp3, tmp2);
5689 tcg_temp_free_i32(tmp3);
5690 break;
5691 case 7: /* VBIF */
5692 tmp3 = neon_load_reg(rd, pass);
5693 gen_neon_bsl(tmp, tmp3, tmp, tmp2);
5694 tcg_temp_free_i32(tmp3);
5695 break;
5696 }
5697 break;
5698 case NEON_3R_VHSUB:
5699 GEN_NEON_INTEGER_OP(hsub);
5700 break;
5701 case NEON_3R_VQSUB:
5702 GEN_NEON_INTEGER_OP_ENV(qsub);
5703 break;
5704 case NEON_3R_VCGT:
5705 GEN_NEON_INTEGER_OP(cgt);
5706 break;
5707 case NEON_3R_VCGE:
5708 GEN_NEON_INTEGER_OP(cge);
5709 break;
5710 case NEON_3R_VSHL:
5711 GEN_NEON_INTEGER_OP(shl);
5712 break;
5713 case NEON_3R_VQSHL:
5714 GEN_NEON_INTEGER_OP_ENV(qshl);
5715 break;
5716 case NEON_3R_VRSHL:
5717 GEN_NEON_INTEGER_OP(rshl);
5718 break;
5719 case NEON_3R_VQRSHL:
5720 GEN_NEON_INTEGER_OP_ENV(qrshl);
5721 break;
5722 case NEON_3R_VMAX:
5723 GEN_NEON_INTEGER_OP(max);
5724 break;
5725 case NEON_3R_VMIN:
5726 GEN_NEON_INTEGER_OP(min);
5727 break;
5728 case NEON_3R_VABD:
5729 GEN_NEON_INTEGER_OP(abd);
5730 break;
5731 case NEON_3R_VABA:
5732 GEN_NEON_INTEGER_OP(abd);
5733 tcg_temp_free_i32(tmp2);
5734 tmp2 = neon_load_reg(rd, pass);
5735 gen_neon_add(size, tmp, tmp2);
5736 break;
5737 case NEON_3R_VADD_VSUB:
5738 if (!u) { /* VADD */
5739 gen_neon_add(size, tmp, tmp2);
5740 } else { /* VSUB */
5741 switch (size) {
5742 case 0: gen_helper_neon_sub_u8(tmp, tmp, tmp2); break;
5743 case 1: gen_helper_neon_sub_u16(tmp, tmp, tmp2); break;
5744 case 2: tcg_gen_sub_i32(tmp, tmp, tmp2); break;
5745 default: abort();
5746 }
5747 }
5748 break;
5749 case NEON_3R_VTST_VCEQ:
5750 if (!u) { /* VTST */
5751 switch (size) {
5752 case 0: gen_helper_neon_tst_u8(tmp, tmp, tmp2); break;
5753 case 1: gen_helper_neon_tst_u16(tmp, tmp, tmp2); break;
5754 case 2: gen_helper_neon_tst_u32(tmp, tmp, tmp2); break;
5755 default: abort();
5756 }
5757 } else { /* VCEQ */
5758 switch (size) {
5759 case 0: gen_helper_neon_ceq_u8(tmp, tmp, tmp2); break;
5760 case 1: gen_helper_neon_ceq_u16(tmp, tmp, tmp2); break;
5761 case 2: gen_helper_neon_ceq_u32(tmp, tmp, tmp2); break;
5762 default: abort();
5763 }
5764 }
5765 break;
5766 case NEON_3R_VML: /* VMLA, VMLAL, VMLS,VMLSL */
5767 switch (size) {
5768 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
5769 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
5770 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
5771 default: abort();
5772 }
5773 tcg_temp_free_i32(tmp2);
5774 tmp2 = neon_load_reg(rd, pass);
5775 if (u) { /* VMLS */
5776 gen_neon_rsb(size, tmp, tmp2);
5777 } else { /* VMLA */
5778 gen_neon_add(size, tmp, tmp2);
5779 }
5780 break;
5781 case NEON_3R_VMUL:
5782 if (u) { /* polynomial */
5783 gen_helper_neon_mul_p8(tmp, tmp, tmp2);
5784 } else { /* Integer */
5785 switch (size) {
5786 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
5787 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
5788 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
5789 default: abort();
5790 }
5791 }
5792 break;
5793 case NEON_3R_VPMAX:
5794 GEN_NEON_INTEGER_OP(pmax);
5795 break;
5796 case NEON_3R_VPMIN:
5797 GEN_NEON_INTEGER_OP(pmin);
5798 break;
5799 case NEON_3R_VQDMULH_VQRDMULH: /* Multiply high. */
5800 if (!u) { /* VQDMULH */
5801 switch (size) {
5802 case 1:
5803 gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2);
5804 break;
5805 case 2:
5806 gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2);
5807 break;
5808 default: abort();
5809 }
5810 } else { /* VQRDMULH */
5811 switch (size) {
5812 case 1:
5813 gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2);
5814 break;
5815 case 2:
5816 gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2);
5817 break;
5818 default: abort();
5819 }
5820 }
5821 break;
5822 case NEON_3R_VPADD:
5823 switch (size) {
5824 case 0: gen_helper_neon_padd_u8(tmp, tmp, tmp2); break;
5825 case 1: gen_helper_neon_padd_u16(tmp, tmp, tmp2); break;
5826 case 2: tcg_gen_add_i32(tmp, tmp, tmp2); break;
5827 default: abort();
5828 }
5829 break;
5830 case NEON_3R_FLOAT_ARITH: /* Floating point arithmetic. */
5831 {
5832 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5833 switch ((u << 2) | size) {
5834 case 0: /* VADD */
5835 case 4: /* VPADD */
5836 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
5837 break;
5838 case 2: /* VSUB */
5839 gen_helper_vfp_subs(tmp, tmp, tmp2, fpstatus);
5840 break;
5841 case 6: /* VABD */
5842 gen_helper_neon_abd_f32(tmp, tmp, tmp2, fpstatus);
5843 break;
5844 default:
5845 abort();
5846 }
5847 tcg_temp_free_ptr(fpstatus);
5848 break;
5849 }
5850 case NEON_3R_FLOAT_MULTIPLY:
5851 {
5852 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5853 gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus);
5854 if (!u) {
5855 tcg_temp_free_i32(tmp2);
5856 tmp2 = neon_load_reg(rd, pass);
5857 if (size == 0) {
5858 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
5859 } else {
5860 gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus);
5861 }
5862 }
5863 tcg_temp_free_ptr(fpstatus);
5864 break;
5865 }
5866 case NEON_3R_FLOAT_CMP:
5867 {
5868 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5869 if (!u) {
5870 gen_helper_neon_ceq_f32(tmp, tmp, tmp2, fpstatus);
5871 } else {
5872 if (size == 0) {
5873 gen_helper_neon_cge_f32(tmp, tmp, tmp2, fpstatus);
5874 } else {
5875 gen_helper_neon_cgt_f32(tmp, tmp, tmp2, fpstatus);
5876 }
5877 }
5878 tcg_temp_free_ptr(fpstatus);
5879 break;
5880 }
5881 case NEON_3R_FLOAT_ACMP:
5882 {
5883 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5884 if (size == 0) {
5885 gen_helper_neon_acge_f32(tmp, tmp, tmp2, fpstatus);
5886 } else {
5887 gen_helper_neon_acgt_f32(tmp, tmp, tmp2, fpstatus);
5888 }
5889 tcg_temp_free_ptr(fpstatus);
5890 break;
5891 }
5892 case NEON_3R_FLOAT_MINMAX:
5893 {
5894 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5895 if (size == 0) {
5896 gen_helper_vfp_maxs(tmp, tmp, tmp2, fpstatus);
5897 } else {
5898 gen_helper_vfp_mins(tmp, tmp, tmp2, fpstatus);
5899 }
5900 tcg_temp_free_ptr(fpstatus);
5901 break;
5902 }
5903 case NEON_3R_FLOAT_MISC:
5904 if (u) {
5905 /* VMAXNM/VMINNM */
5906 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5907 if (size == 0) {
5908 gen_helper_vfp_maxnums(tmp, tmp, tmp2, fpstatus);
5909 } else {
5910 gen_helper_vfp_minnums(tmp, tmp, tmp2, fpstatus);
5911 }
5912 tcg_temp_free_ptr(fpstatus);
5913 } else {
5914 if (size == 0) {
5915 gen_helper_recps_f32(tmp, tmp, tmp2, cpu_env);
5916 } else {
5917 gen_helper_rsqrts_f32(tmp, tmp, tmp2, cpu_env);
5918 }
5919 }
5920 break;
5921 case NEON_3R_VFM:
5922 {
5923 /* VFMA, VFMS: fused multiply-add */
5924 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5925 TCGv_i32 tmp3 = neon_load_reg(rd, pass);
5926 if (size) {
5927 /* VFMS */
5928 gen_helper_vfp_negs(tmp, tmp);
5929 }
5930 gen_helper_vfp_muladds(tmp, tmp, tmp2, tmp3, fpstatus);
5931 tcg_temp_free_i32(tmp3);
5932 tcg_temp_free_ptr(fpstatus);
5933 break;
5934 }
5935 default:
5936 abort();
5937 }
5938 tcg_temp_free_i32(tmp2);
5939
5940 /* Save the result. For elementwise operations we can put it
5941 straight into the destination register. For pairwise operations
5942 we have to be careful to avoid clobbering the source operands. */
5943 if (pairwise && rd == rm) {
5944 neon_store_scratch(pass, tmp);
5945 } else {
5946 neon_store_reg(rd, pass, tmp);
5947 }
5948
5949 } /* for pass */
5950 if (pairwise && rd == rm) {
5951 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5952 tmp = neon_load_scratch(pass);
5953 neon_store_reg(rd, pass, tmp);
5954 }
5955 }
5956 /* End of 3 register same size operations. */
5957 } else if (insn & (1 << 4)) {
5958 if ((insn & 0x00380080) != 0) {
5959 /* Two registers and shift. */
5960 op = (insn >> 8) & 0xf;
5961 if (insn & (1 << 7)) {
5962 /* 64-bit shift. */
5963 if (op > 7) {
5964 return 1;
5965 }
5966 size = 3;
5967 } else {
5968 size = 2;
5969 while ((insn & (1 << (size + 19))) == 0)
5970 size--;
5971 }
5972 shift = (insn >> 16) & ((1 << (3 + size)) - 1);
5973 /* To avoid excessive duplication of ops we implement shift
5974 by immediate using the variable shift operations. */
5975 if (op < 8) {
5976 /* Shift by immediate:
5977 VSHR, VSRA, VRSHR, VRSRA, VSRI, VSHL, VQSHL, VQSHLU. */
5978 if (q && ((rd | rm) & 1)) {
5979 return 1;
5980 }
5981 if (!u && (op == 4 || op == 6)) {
5982 return 1;
5983 }
5984 /* Right shifts are encoded as N - shift, where N is the
5985 element size in bits. */
5986 if (op <= 4)
5987 shift = shift - (1 << (size + 3));
5988 if (size == 3) {
5989 count = q + 1;
5990 } else {
5991 count = q ? 4: 2;
5992 }
5993 switch (size) {
5994 case 0:
5995 imm = (uint8_t) shift;
5996 imm |= imm << 8;
5997 imm |= imm << 16;
5998 break;
5999 case 1:
6000 imm = (uint16_t) shift;
6001 imm |= imm << 16;
6002 break;
6003 case 2:
6004 case 3:
6005 imm = shift;
6006 break;
6007 default:
6008 abort();
6009 }
6010
6011 for (pass = 0; pass < count; pass++) {
6012 if (size == 3) {
6013 neon_load_reg64(cpu_V0, rm + pass);
6014 tcg_gen_movi_i64(cpu_V1, imm);
6015 switch (op) {
6016 case 0: /* VSHR */
6017 case 1: /* VSRA */
6018 if (u)
6019 gen_helper_neon_shl_u64(cpu_V0, cpu_V0, cpu_V1);
6020 else
6021 gen_helper_neon_shl_s64(cpu_V0, cpu_V0, cpu_V1);
6022 break;
6023 case 2: /* VRSHR */
6024 case 3: /* VRSRA */
6025 if (u)
6026 gen_helper_neon_rshl_u64(cpu_V0, cpu_V0, cpu_V1);
6027 else
6028 gen_helper_neon_rshl_s64(cpu_V0, cpu_V0, cpu_V1);
6029 break;
6030 case 4: /* VSRI */
6031 case 5: /* VSHL, VSLI */
6032 gen_helper_neon_shl_u64(cpu_V0, cpu_V0, cpu_V1);
6033 break;
6034 case 6: /* VQSHLU */
6035 gen_helper_neon_qshlu_s64(cpu_V0, cpu_env,
6036 cpu_V0, cpu_V1);
6037 break;
6038 case 7: /* VQSHL */
6039 if (u) {
6040 gen_helper_neon_qshl_u64(cpu_V0, cpu_env,
6041 cpu_V0, cpu_V1);
6042 } else {
6043 gen_helper_neon_qshl_s64(cpu_V0, cpu_env,
6044 cpu_V0, cpu_V1);
6045 }
6046 break;
6047 }
6048 if (op == 1 || op == 3) {
6049 /* Accumulate. */
6050 neon_load_reg64(cpu_V1, rd + pass);
6051 tcg_gen_add_i64(cpu_V0, cpu_V0, cpu_V1);
6052 } else if (op == 4 || (op == 5 && u)) {
6053 /* Insert */
6054 neon_load_reg64(cpu_V1, rd + pass);
6055 uint64_t mask;
6056 if (shift < -63 || shift > 63) {
6057 mask = 0;
6058 } else {
6059 if (op == 4) {
6060 mask = 0xffffffffffffffffull >> -shift;
6061 } else {
6062 mask = 0xffffffffffffffffull << shift;
6063 }
6064 }
6065 tcg_gen_andi_i64(cpu_V1, cpu_V1, ~mask);
6066 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
6067 }
6068 neon_store_reg64(cpu_V0, rd + pass);
6069 } else { /* size < 3 */
6070 /* Operands in T0 and T1. */
6071 tmp = neon_load_reg(rm, pass);
6072 tmp2 = tcg_temp_new_i32();
6073 tcg_gen_movi_i32(tmp2, imm);
6074 switch (op) {
6075 case 0: /* VSHR */
6076 case 1: /* VSRA */
6077 GEN_NEON_INTEGER_OP(shl);
6078 break;
6079 case 2: /* VRSHR */
6080 case 3: /* VRSRA */
6081 GEN_NEON_INTEGER_OP(rshl);
6082 break;
6083 case 4: /* VSRI */
6084 case 5: /* VSHL, VSLI */
6085 switch (size) {
6086 case 0: gen_helper_neon_shl_u8(tmp, tmp, tmp2); break;
6087 case 1: gen_helper_neon_shl_u16(tmp, tmp, tmp2); break;
6088 case 2: gen_helper_neon_shl_u32(tmp, tmp, tmp2); break;
6089 default: abort();
6090 }
6091 break;
6092 case 6: /* VQSHLU */
6093 switch (size) {
6094 case 0:
6095 gen_helper_neon_qshlu_s8(tmp, cpu_env,
6096 tmp, tmp2);
6097 break;
6098 case 1:
6099 gen_helper_neon_qshlu_s16(tmp, cpu_env,
6100 tmp, tmp2);
6101 break;
6102 case 2:
6103 gen_helper_neon_qshlu_s32(tmp, cpu_env,
6104 tmp, tmp2);
6105 break;
6106 default:
6107 abort();
6108 }
6109 break;
6110 case 7: /* VQSHL */
6111 GEN_NEON_INTEGER_OP_ENV(qshl);
6112 break;
6113 }
6114 tcg_temp_free_i32(tmp2);
6115
6116 if (op == 1 || op == 3) {
6117 /* Accumulate. */
6118 tmp2 = neon_load_reg(rd, pass);
6119 gen_neon_add(size, tmp, tmp2);
6120 tcg_temp_free_i32(tmp2);
6121 } else if (op == 4 || (op == 5 && u)) {
6122 /* Insert */
6123 switch (size) {
6124 case 0:
6125 if (op == 4)
6126 mask = 0xff >> -shift;
6127 else
6128 mask = (uint8_t)(0xff << shift);
6129 mask |= mask << 8;
6130 mask |= mask << 16;
6131 break;
6132 case 1:
6133 if (op == 4)
6134 mask = 0xffff >> -shift;
6135 else
6136 mask = (uint16_t)(0xffff << shift);
6137 mask |= mask << 16;
6138 break;
6139 case 2:
6140 if (shift < -31 || shift > 31) {
6141 mask = 0;
6142 } else {
6143 if (op == 4)
6144 mask = 0xffffffffu >> -shift;
6145 else
6146 mask = 0xffffffffu << shift;
6147 }
6148 break;
6149 default:
6150 abort();
6151 }
6152 tmp2 = neon_load_reg(rd, pass);
6153 tcg_gen_andi_i32(tmp, tmp, mask);
6154 tcg_gen_andi_i32(tmp2, tmp2, ~mask);
6155 tcg_gen_or_i32(tmp, tmp, tmp2);
6156 tcg_temp_free_i32(tmp2);
6157 }
6158 neon_store_reg(rd, pass, tmp);
6159 }
6160 } /* for pass */
6161 } else if (op < 10) {
6162 /* Shift by immediate and narrow:
6163 VSHRN, VRSHRN, VQSHRN, VQRSHRN. */
6164 int input_unsigned = (op == 8) ? !u : u;
6165 if (rm & 1) {
6166 return 1;
6167 }
6168 shift = shift - (1 << (size + 3));
6169 size++;
6170 if (size == 3) {
6171 tmp64 = tcg_const_i64(shift);
6172 neon_load_reg64(cpu_V0, rm);
6173 neon_load_reg64(cpu_V1, rm + 1);
6174 for (pass = 0; pass < 2; pass++) {
6175 TCGv_i64 in;
6176 if (pass == 0) {
6177 in = cpu_V0;
6178 } else {
6179 in = cpu_V1;
6180 }
6181 if (q) {
6182 if (input_unsigned) {
6183 gen_helper_neon_rshl_u64(cpu_V0, in, tmp64);
6184 } else {
6185 gen_helper_neon_rshl_s64(cpu_V0, in, tmp64);
6186 }
6187 } else {
6188 if (input_unsigned) {
6189 gen_helper_neon_shl_u64(cpu_V0, in, tmp64);
6190 } else {
6191 gen_helper_neon_shl_s64(cpu_V0, in, tmp64);
6192 }
6193 }
6194 tmp = tcg_temp_new_i32();
6195 gen_neon_narrow_op(op == 8, u, size - 1, tmp, cpu_V0);
6196 neon_store_reg(rd, pass, tmp);
6197 } /* for pass */
6198 tcg_temp_free_i64(tmp64);
6199 } else {
6200 if (size == 1) {
6201 imm = (uint16_t)shift;
6202 imm |= imm << 16;
6203 } else {
6204 /* size == 2 */
6205 imm = (uint32_t)shift;
6206 }
6207 tmp2 = tcg_const_i32(imm);
6208 tmp4 = neon_load_reg(rm + 1, 0);
6209 tmp5 = neon_load_reg(rm + 1, 1);
6210 for (pass = 0; pass < 2; pass++) {
6211 if (pass == 0) {
6212 tmp = neon_load_reg(rm, 0);
6213 } else {
6214 tmp = tmp4;
6215 }
6216 gen_neon_shift_narrow(size, tmp, tmp2, q,
6217 input_unsigned);
6218 if (pass == 0) {
6219 tmp3 = neon_load_reg(rm, 1);
6220 } else {
6221 tmp3 = tmp5;
6222 }
6223 gen_neon_shift_narrow(size, tmp3, tmp2, q,
6224 input_unsigned);
6225 tcg_gen_concat_i32_i64(cpu_V0, tmp, tmp3);
6226 tcg_temp_free_i32(tmp);
6227 tcg_temp_free_i32(tmp3);
6228 tmp = tcg_temp_new_i32();
6229 gen_neon_narrow_op(op == 8, u, size - 1, tmp, cpu_V0);
6230 neon_store_reg(rd, pass, tmp);
6231 } /* for pass */
6232 tcg_temp_free_i32(tmp2);
6233 }
6234 } else if (op == 10) {
6235 /* VSHLL, VMOVL */
6236 if (q || (rd & 1)) {
6237 return 1;
6238 }
6239 tmp = neon_load_reg(rm, 0);
6240 tmp2 = neon_load_reg(rm, 1);
6241 for (pass = 0; pass < 2; pass++) {
6242 if (pass == 1)
6243 tmp = tmp2;
6244
6245 gen_neon_widen(cpu_V0, tmp, size, u);
6246
6247 if (shift != 0) {
6248 /* The shift is less than the width of the source
6249 type, so we can just shift the whole register. */
6250 tcg_gen_shli_i64(cpu_V0, cpu_V0, shift);
6251 /* Widen the result of shift: we need to clear
6252 * the potential overflow bits resulting from
6253 * left bits of the narrow input appearing as
6254 * right bits of left the neighbour narrow
6255 * input. */
6256 if (size < 2 || !u) {
6257 uint64_t imm64;
6258 if (size == 0) {
6259 imm = (0xffu >> (8 - shift));
6260 imm |= imm << 16;
6261 } else if (size == 1) {
6262 imm = 0xffff >> (16 - shift);
6263 } else {
6264 /* size == 2 */
6265 imm = 0xffffffff >> (32 - shift);
6266 }
6267 if (size < 2) {
6268 imm64 = imm | (((uint64_t)imm) << 32);
6269 } else {
6270 imm64 = imm;
6271 }
6272 tcg_gen_andi_i64(cpu_V0, cpu_V0, ~imm64);
6273 }
6274 }
6275 neon_store_reg64(cpu_V0, rd + pass);
6276 }
6277 } else if (op >= 14) {
6278 /* VCVT fixed-point. */
6279 if (!(insn & (1 << 21)) || (q && ((rd | rm) & 1))) {
6280 return 1;
6281 }
6282 /* We have already masked out the must-be-1 top bit of imm6,
6283 * hence this 32-shift where the ARM ARM has 64-imm6.
6284 */
6285 shift = 32 - shift;
6286 for (pass = 0; pass < (q ? 4 : 2); pass++) {
6287 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, pass));
6288 if (!(op & 1)) {
6289 if (u)
6290 gen_vfp_ulto(0, shift, 1);
6291 else
6292 gen_vfp_slto(0, shift, 1);
6293 } else {
6294 if (u)
6295 gen_vfp_toul(0, shift, 1);
6296 else
6297 gen_vfp_tosl(0, shift, 1);
6298 }
6299 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, pass));
6300 }
6301 } else {
6302 return 1;
6303 }
6304 } else { /* (insn & 0x00380080) == 0 */
6305 int invert;
6306 if (q && (rd & 1)) {
6307 return 1;
6308 }
6309
6310 op = (insn >> 8) & 0xf;
6311 /* One register and immediate. */
6312 imm = (u << 7) | ((insn >> 12) & 0x70) | (insn & 0xf);
6313 invert = (insn & (1 << 5)) != 0;
6314 /* Note that op = 2,3,4,5,6,7,10,11,12,13 imm=0 is UNPREDICTABLE.
6315 * We choose to not special-case this and will behave as if a
6316 * valid constant encoding of 0 had been given.
6317 */
6318 switch (op) {
6319 case 0: case 1:
6320 /* no-op */
6321 break;
6322 case 2: case 3:
6323 imm <<= 8;
6324 break;
6325 case 4: case 5:
6326 imm <<= 16;
6327 break;
6328 case 6: case 7:
6329 imm <<= 24;
6330 break;
6331 case 8: case 9:
6332 imm |= imm << 16;
6333 break;
6334 case 10: case 11:
6335 imm = (imm << 8) | (imm << 24);
6336 break;
6337 case 12:
6338 imm = (imm << 8) | 0xff;
6339 break;
6340 case 13:
6341 imm = (imm << 16) | 0xffff;
6342 break;
6343 case 14:
6344 imm |= (imm << 8) | (imm << 16) | (imm << 24);
6345 if (invert)
6346 imm = ~imm;
6347 break;
6348 case 15:
6349 if (invert) {
6350 return 1;
6351 }
6352 imm = ((imm & 0x80) << 24) | ((imm & 0x3f) << 19)
6353 | ((imm & 0x40) ? (0x1f << 25) : (1 << 30));
6354 break;
6355 }
6356 if (invert)
6357 imm = ~imm;
6358
6359 for (pass = 0; pass < (q ? 4 : 2); pass++) {
6360 if (op & 1 && op < 12) {
6361 tmp = neon_load_reg(rd, pass);
6362 if (invert) {
6363 /* The immediate value has already been inverted, so
6364 BIC becomes AND. */
6365 tcg_gen_andi_i32(tmp, tmp, imm);
6366 } else {
6367 tcg_gen_ori_i32(tmp, tmp, imm);
6368 }
6369 } else {
6370 /* VMOV, VMVN. */
6371 tmp = tcg_temp_new_i32();
6372 if (op == 14 && invert) {
6373 int n;
6374 uint32_t val;
6375 val = 0;
6376 for (n = 0; n < 4; n++) {
6377 if (imm & (1 << (n + (pass & 1) * 4)))
6378 val |= 0xff << (n * 8);
6379 }
6380 tcg_gen_movi_i32(tmp, val);
6381 } else {
6382 tcg_gen_movi_i32(tmp, imm);
6383 }
6384 }
6385 neon_store_reg(rd, pass, tmp);
6386 }
6387 }
6388 } else { /* (insn & 0x00800010 == 0x00800000) */
6389 if (size != 3) {
6390 op = (insn >> 8) & 0xf;
6391 if ((insn & (1 << 6)) == 0) {
6392 /* Three registers of different lengths. */
6393 int src1_wide;
6394 int src2_wide;
6395 int prewiden;
6396 /* undefreq: bit 0 : UNDEF if size == 0
6397 * bit 1 : UNDEF if size == 1
6398 * bit 2 : UNDEF if size == 2
6399 * bit 3 : UNDEF if U == 1
6400 * Note that [2:0] set implies 'always UNDEF'
6401 */
6402 int undefreq;
6403 /* prewiden, src1_wide, src2_wide, undefreq */
6404 static const int neon_3reg_wide[16][4] = {
6405 {1, 0, 0, 0}, /* VADDL */
6406 {1, 1, 0, 0}, /* VADDW */
6407 {1, 0, 0, 0}, /* VSUBL */
6408 {1, 1, 0, 0}, /* VSUBW */
6409 {0, 1, 1, 0}, /* VADDHN */
6410 {0, 0, 0, 0}, /* VABAL */
6411 {0, 1, 1, 0}, /* VSUBHN */
6412 {0, 0, 0, 0}, /* VABDL */
6413 {0, 0, 0, 0}, /* VMLAL */
6414 {0, 0, 0, 9}, /* VQDMLAL */
6415 {0, 0, 0, 0}, /* VMLSL */
6416 {0, 0, 0, 9}, /* VQDMLSL */
6417 {0, 0, 0, 0}, /* Integer VMULL */
6418 {0, 0, 0, 1}, /* VQDMULL */
6419 {0, 0, 0, 0xa}, /* Polynomial VMULL */
6420 {0, 0, 0, 7}, /* Reserved: always UNDEF */
6421 };
6422
6423 prewiden = neon_3reg_wide[op][0];
6424 src1_wide = neon_3reg_wide[op][1];
6425 src2_wide = neon_3reg_wide[op][2];
6426 undefreq = neon_3reg_wide[op][3];
6427
6428 if ((undefreq & (1 << size)) ||
6429 ((undefreq & 8) && u)) {
6430 return 1;
6431 }
6432 if ((src1_wide && (rn & 1)) ||
6433 (src2_wide && (rm & 1)) ||
6434 (!src2_wide && (rd & 1))) {
6435 return 1;
6436 }
6437
6438 /* Handle VMULL.P64 (Polynomial 64x64 to 128 bit multiply)
6439 * outside the loop below as it only performs a single pass.
6440 */
6441 if (op == 14 && size == 2) {
6442 TCGv_i64 tcg_rn, tcg_rm, tcg_rd;
6443
6444 if (!arm_dc_feature(s, ARM_FEATURE_V8_PMULL)) {
6445 return 1;
6446 }
6447 tcg_rn = tcg_temp_new_i64();
6448 tcg_rm = tcg_temp_new_i64();
6449 tcg_rd = tcg_temp_new_i64();
6450 neon_load_reg64(tcg_rn, rn);
6451 neon_load_reg64(tcg_rm, rm);
6452 gen_helper_neon_pmull_64_lo(tcg_rd, tcg_rn, tcg_rm);
6453 neon_store_reg64(tcg_rd, rd);
6454 gen_helper_neon_pmull_64_hi(tcg_rd, tcg_rn, tcg_rm);
6455 neon_store_reg64(tcg_rd, rd + 1);
6456 tcg_temp_free_i64(tcg_rn);
6457 tcg_temp_free_i64(tcg_rm);
6458 tcg_temp_free_i64(tcg_rd);
6459 return 0;
6460 }
6461
6462 /* Avoid overlapping operands. Wide source operands are
6463 always aligned so will never overlap with wide
6464 destinations in problematic ways. */
6465 if (rd == rm && !src2_wide) {
6466 tmp = neon_load_reg(rm, 1);
6467 neon_store_scratch(2, tmp);
6468 } else if (rd == rn && !src1_wide) {
6469 tmp = neon_load_reg(rn, 1);
6470 neon_store_scratch(2, tmp);
6471 }
6472 TCGV_UNUSED_I32(tmp3);
6473 for (pass = 0; pass < 2; pass++) {
6474 if (src1_wide) {
6475 neon_load_reg64(cpu_V0, rn + pass);
6476 TCGV_UNUSED_I32(tmp);
6477 } else {
6478 if (pass == 1 && rd == rn) {
6479 tmp = neon_load_scratch(2);
6480 } else {
6481 tmp = neon_load_reg(rn, pass);
6482 }
6483 if (prewiden) {
6484 gen_neon_widen(cpu_V0, tmp, size, u);
6485 }
6486 }
6487 if (src2_wide) {
6488 neon_load_reg64(cpu_V1, rm + pass);
6489 TCGV_UNUSED_I32(tmp2);
6490 } else {
6491 if (pass == 1 && rd == rm) {
6492 tmp2 = neon_load_scratch(2);
6493 } else {
6494 tmp2 = neon_load_reg(rm, pass);
6495 }
6496 if (prewiden) {
6497 gen_neon_widen(cpu_V1, tmp2, size, u);
6498 }
6499 }
6500 switch (op) {
6501 case 0: case 1: case 4: /* VADDL, VADDW, VADDHN, VRADDHN */
6502 gen_neon_addl(size);
6503 break;
6504 case 2: case 3: case 6: /* VSUBL, VSUBW, VSUBHN, VRSUBHN */
6505 gen_neon_subl(size);
6506 break;
6507 case 5: case 7: /* VABAL, VABDL */
6508 switch ((size << 1) | u) {
6509 case 0:
6510 gen_helper_neon_abdl_s16(cpu_V0, tmp, tmp2);
6511 break;
6512 case 1:
6513 gen_helper_neon_abdl_u16(cpu_V0, tmp, tmp2);
6514 break;
6515 case 2:
6516 gen_helper_neon_abdl_s32(cpu_V0, tmp, tmp2);
6517 break;
6518 case 3:
6519 gen_helper_neon_abdl_u32(cpu_V0, tmp, tmp2);
6520 break;
6521 case 4:
6522 gen_helper_neon_abdl_s64(cpu_V0, tmp, tmp2);
6523 break;
6524 case 5:
6525 gen_helper_neon_abdl_u64(cpu_V0, tmp, tmp2);
6526 break;
6527 default: abort();
6528 }
6529 tcg_temp_free_i32(tmp2);
6530 tcg_temp_free_i32(tmp);
6531 break;
6532 case 8: case 9: case 10: case 11: case 12: case 13:
6533 /* VMLAL, VQDMLAL, VMLSL, VQDMLSL, VMULL, VQDMULL */
6534 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
6535 break;
6536 case 14: /* Polynomial VMULL */
6537 gen_helper_neon_mull_p8(cpu_V0, tmp, tmp2);
6538 tcg_temp_free_i32(tmp2);
6539 tcg_temp_free_i32(tmp);
6540 break;
6541 default: /* 15 is RESERVED: caught earlier */
6542 abort();
6543 }
6544 if (op == 13) {
6545 /* VQDMULL */
6546 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
6547 neon_store_reg64(cpu_V0, rd + pass);
6548 } else if (op == 5 || (op >= 8 && op <= 11)) {
6549 /* Accumulate. */
6550 neon_load_reg64(cpu_V1, rd + pass);
6551 switch (op) {
6552 case 10: /* VMLSL */
6553 gen_neon_negl(cpu_V0, size);
6554 /* Fall through */
6555 case 5: case 8: /* VABAL, VMLAL */
6556 gen_neon_addl(size);
6557 break;
6558 case 9: case 11: /* VQDMLAL, VQDMLSL */
6559 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
6560 if (op == 11) {
6561 gen_neon_negl(cpu_V0, size);
6562 }
6563 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
6564 break;
6565 default:
6566 abort();
6567 }
6568 neon_store_reg64(cpu_V0, rd + pass);
6569 } else if (op == 4 || op == 6) {
6570 /* Narrowing operation. */
6571 tmp = tcg_temp_new_i32();
6572 if (!u) {
6573 switch (size) {
6574 case 0:
6575 gen_helper_neon_narrow_high_u8(tmp, cpu_V0);
6576 break;
6577 case 1:
6578 gen_helper_neon_narrow_high_u16(tmp, cpu_V0);
6579 break;
6580 case 2:
6581 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
6582 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
6583 break;
6584 default: abort();
6585 }
6586 } else {
6587 switch (size) {
6588 case 0:
6589 gen_helper_neon_narrow_round_high_u8(tmp, cpu_V0);
6590 break;
6591 case 1:
6592 gen_helper_neon_narrow_round_high_u16(tmp, cpu_V0);
6593 break;
6594 case 2:
6595 tcg_gen_addi_i64(cpu_V0, cpu_V0, 1u << 31);
6596 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
6597 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
6598 break;
6599 default: abort();
6600 }
6601 }
6602 if (pass == 0) {
6603 tmp3 = tmp;
6604 } else {
6605 neon_store_reg(rd, 0, tmp3);
6606 neon_store_reg(rd, 1, tmp);
6607 }
6608 } else {
6609 /* Write back the result. */
6610 neon_store_reg64(cpu_V0, rd + pass);
6611 }
6612 }
6613 } else {
6614 /* Two registers and a scalar. NB that for ops of this form
6615 * the ARM ARM labels bit 24 as Q, but it is in our variable
6616 * 'u', not 'q'.
6617 */
6618 if (size == 0) {
6619 return 1;
6620 }
6621 switch (op) {
6622 case 1: /* Float VMLA scalar */
6623 case 5: /* Floating point VMLS scalar */
6624 case 9: /* Floating point VMUL scalar */
6625 if (size == 1) {
6626 return 1;
6627 }
6628 /* fall through */
6629 case 0: /* Integer VMLA scalar */
6630 case 4: /* Integer VMLS scalar */
6631 case 8: /* Integer VMUL scalar */
6632 case 12: /* VQDMULH scalar */
6633 case 13: /* VQRDMULH scalar */
6634 if (u && ((rd | rn) & 1)) {
6635 return 1;
6636 }
6637 tmp = neon_get_scalar(size, rm);
6638 neon_store_scratch(0, tmp);
6639 for (pass = 0; pass < (u ? 4 : 2); pass++) {
6640 tmp = neon_load_scratch(0);
6641 tmp2 = neon_load_reg(rn, pass);
6642 if (op == 12) {
6643 if (size == 1) {
6644 gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2);
6645 } else {
6646 gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2);
6647 }
6648 } else if (op == 13) {
6649 if (size == 1) {
6650 gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2);
6651 } else {
6652 gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2);
6653 }
6654 } else if (op & 1) {
6655 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6656 gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus);
6657 tcg_temp_free_ptr(fpstatus);
6658 } else {
6659 switch (size) {
6660 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
6661 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
6662 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
6663 default: abort();
6664 }
6665 }
6666 tcg_temp_free_i32(tmp2);
6667 if (op < 8) {
6668 /* Accumulate. */
6669 tmp2 = neon_load_reg(rd, pass);
6670 switch (op) {
6671 case 0:
6672 gen_neon_add(size, tmp, tmp2);
6673 break;
6674 case 1:
6675 {
6676 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6677 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
6678 tcg_temp_free_ptr(fpstatus);
6679 break;
6680 }
6681 case 4:
6682 gen_neon_rsb(size, tmp, tmp2);
6683 break;
6684 case 5:
6685 {
6686 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6687 gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus);
6688 tcg_temp_free_ptr(fpstatus);
6689 break;
6690 }
6691 default:
6692 abort();
6693 }
6694 tcg_temp_free_i32(tmp2);
6695 }
6696 neon_store_reg(rd, pass, tmp);
6697 }
6698 break;
6699 case 3: /* VQDMLAL scalar */
6700 case 7: /* VQDMLSL scalar */
6701 case 11: /* VQDMULL scalar */
6702 if (u == 1) {
6703 return 1;
6704 }
6705 /* fall through */
6706 case 2: /* VMLAL sclar */
6707 case 6: /* VMLSL scalar */
6708 case 10: /* VMULL scalar */
6709 if (rd & 1) {
6710 return 1;
6711 }
6712 tmp2 = neon_get_scalar(size, rm);
6713 /* We need a copy of tmp2 because gen_neon_mull
6714 * deletes it during pass 0. */
6715 tmp4 = tcg_temp_new_i32();
6716 tcg_gen_mov_i32(tmp4, tmp2);
6717 tmp3 = neon_load_reg(rn, 1);
6718
6719 for (pass = 0; pass < 2; pass++) {
6720 if (pass == 0) {
6721 tmp = neon_load_reg(rn, 0);
6722 } else {
6723 tmp = tmp3;
6724 tmp2 = tmp4;
6725 }
6726 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
6727 if (op != 11) {
6728 neon_load_reg64(cpu_V1, rd + pass);
6729 }
6730 switch (op) {
6731 case 6:
6732 gen_neon_negl(cpu_V0, size);
6733 /* Fall through */
6734 case 2:
6735 gen_neon_addl(size);
6736 break;
6737 case 3: case 7:
6738 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
6739 if (op == 7) {
6740 gen_neon_negl(cpu_V0, size);
6741 }
6742 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
6743 break;
6744 case 10:
6745 /* no-op */
6746 break;
6747 case 11:
6748 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
6749 break;
6750 default:
6751 abort();
6752 }
6753 neon_store_reg64(cpu_V0, rd + pass);
6754 }
6755
6756
6757 break;
6758 default: /* 14 and 15 are RESERVED */
6759 return 1;
6760 }
6761 }
6762 } else { /* size == 3 */
6763 if (!u) {
6764 /* Extract. */
6765 imm = (insn >> 8) & 0xf;
6766
6767 if (imm > 7 && !q)
6768 return 1;
6769
6770 if (q && ((rd | rn | rm) & 1)) {
6771 return 1;
6772 }
6773
6774 if (imm == 0) {
6775 neon_load_reg64(cpu_V0, rn);
6776 if (q) {
6777 neon_load_reg64(cpu_V1, rn + 1);
6778 }
6779 } else if (imm == 8) {
6780 neon_load_reg64(cpu_V0, rn + 1);
6781 if (q) {
6782 neon_load_reg64(cpu_V1, rm);
6783 }
6784 } else if (q) {
6785 tmp64 = tcg_temp_new_i64();
6786 if (imm < 8) {
6787 neon_load_reg64(cpu_V0, rn);
6788 neon_load_reg64(tmp64, rn + 1);
6789 } else {
6790 neon_load_reg64(cpu_V0, rn + 1);
6791 neon_load_reg64(tmp64, rm);
6792 }
6793 tcg_gen_shri_i64(cpu_V0, cpu_V0, (imm & 7) * 8);
6794 tcg_gen_shli_i64(cpu_V1, tmp64, 64 - ((imm & 7) * 8));
6795 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
6796 if (imm < 8) {
6797 neon_load_reg64(cpu_V1, rm);
6798 } else {
6799 neon_load_reg64(cpu_V1, rm + 1);
6800 imm -= 8;
6801 }
6802 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
6803 tcg_gen_shri_i64(tmp64, tmp64, imm * 8);
6804 tcg_gen_or_i64(cpu_V1, cpu_V1, tmp64);
6805 tcg_temp_free_i64(tmp64);
6806 } else {
6807 /* BUGFIX */
6808 neon_load_reg64(cpu_V0, rn);
6809 tcg_gen_shri_i64(cpu_V0, cpu_V0, imm * 8);
6810 neon_load_reg64(cpu_V1, rm);
6811 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
6812 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
6813 }
6814 neon_store_reg64(cpu_V0, rd);
6815 if (q) {
6816 neon_store_reg64(cpu_V1, rd + 1);
6817 }
6818 } else if ((insn & (1 << 11)) == 0) {
6819 /* Two register misc. */
6820 op = ((insn >> 12) & 0x30) | ((insn >> 7) & 0xf);
6821 size = (insn >> 18) & 3;
6822 /* UNDEF for unknown op values and bad op-size combinations */
6823 if ((neon_2rm_sizes[op] & (1 << size)) == 0) {
6824 return 1;
6825 }
6826 if (neon_2rm_is_v8_op(op) &&
6827 !arm_dc_feature(s, ARM_FEATURE_V8)) {
6828 return 1;
6829 }
6830 if ((op != NEON_2RM_VMOVN && op != NEON_2RM_VQMOVN) &&
6831 q && ((rm | rd) & 1)) {
6832 return 1;
6833 }
6834 switch (op) {
6835 case NEON_2RM_VREV64:
6836 for (pass = 0; pass < (q ? 2 : 1); pass++) {
6837 tmp = neon_load_reg(rm, pass * 2);
6838 tmp2 = neon_load_reg(rm, pass * 2 + 1);
6839 switch (size) {
6840 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
6841 case 1: gen_swap_half(tmp); break;
6842 case 2: /* no-op */ break;
6843 default: abort();
6844 }
6845 neon_store_reg(rd, pass * 2 + 1, tmp);
6846 if (size == 2) {
6847 neon_store_reg(rd, pass * 2, tmp2);
6848 } else {
6849 switch (size) {
6850 case 0: tcg_gen_bswap32_i32(tmp2, tmp2); break;
6851 case 1: gen_swap_half(tmp2); break;
6852 default: abort();
6853 }
6854 neon_store_reg(rd, pass * 2, tmp2);
6855 }
6856 }
6857 break;
6858 case NEON_2RM_VPADDL: case NEON_2RM_VPADDL_U:
6859 case NEON_2RM_VPADAL: case NEON_2RM_VPADAL_U:
6860 for (pass = 0; pass < q + 1; pass++) {
6861 tmp = neon_load_reg(rm, pass * 2);
6862 gen_neon_widen(cpu_V0, tmp, size, op & 1);
6863 tmp = neon_load_reg(rm, pass * 2 + 1);
6864 gen_neon_widen(cpu_V1, tmp, size, op & 1);
6865 switch (size) {
6866 case 0: gen_helper_neon_paddl_u16(CPU_V001); break;
6867 case 1: gen_helper_neon_paddl_u32(CPU_V001); break;
6868 case 2: tcg_gen_add_i64(CPU_V001); break;
6869 default: abort();
6870 }
6871 if (op >= NEON_2RM_VPADAL) {
6872 /* Accumulate. */
6873 neon_load_reg64(cpu_V1, rd + pass);
6874 gen_neon_addl(size);
6875 }
6876 neon_store_reg64(cpu_V0, rd + pass);
6877 }
6878 break;
6879 case NEON_2RM_VTRN:
6880 if (size == 2) {
6881 int n;
6882 for (n = 0; n < (q ? 4 : 2); n += 2) {
6883 tmp = neon_load_reg(rm, n);
6884 tmp2 = neon_load_reg(rd, n + 1);
6885 neon_store_reg(rm, n, tmp2);
6886 neon_store_reg(rd, n + 1, tmp);
6887 }
6888 } else {
6889 goto elementwise;
6890 }
6891 break;
6892 case NEON_2RM_VUZP:
6893 if (gen_neon_unzip(rd, rm, size, q)) {
6894 return 1;
6895 }
6896 break;
6897 case NEON_2RM_VZIP:
6898 if (gen_neon_zip(rd, rm, size, q)) {
6899 return 1;
6900 }
6901 break;
6902 case NEON_2RM_VMOVN: case NEON_2RM_VQMOVN:
6903 /* also VQMOVUN; op field and mnemonics don't line up */
6904 if (rm & 1) {
6905 return 1;
6906 }
6907 TCGV_UNUSED_I32(tmp2);
6908 for (pass = 0; pass < 2; pass++) {
6909 neon_load_reg64(cpu_V0, rm + pass);
6910 tmp = tcg_temp_new_i32();
6911 gen_neon_narrow_op(op == NEON_2RM_VMOVN, q, size,
6912 tmp, cpu_V0);
6913 if (pass == 0) {
6914 tmp2 = tmp;
6915 } else {
6916 neon_store_reg(rd, 0, tmp2);
6917 neon_store_reg(rd, 1, tmp);
6918 }
6919 }
6920 break;
6921 case NEON_2RM_VSHLL:
6922 if (q || (rd & 1)) {
6923 return 1;
6924 }
6925 tmp = neon_load_reg(rm, 0);
6926 tmp2 = neon_load_reg(rm, 1);
6927 for (pass = 0; pass < 2; pass++) {
6928 if (pass == 1)
6929 tmp = tmp2;
6930 gen_neon_widen(cpu_V0, tmp, size, 1);
6931 tcg_gen_shli_i64(cpu_V0, cpu_V0, 8 << size);
6932 neon_store_reg64(cpu_V0, rd + pass);
6933 }
6934 break;
6935 case NEON_2RM_VCVT_F16_F32:
6936 if (!arm_dc_feature(s, ARM_FEATURE_VFP_FP16) ||
6937 q || (rm & 1)) {
6938 return 1;
6939 }
6940 tmp = tcg_temp_new_i32();
6941 tmp2 = tcg_temp_new_i32();
6942 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 0));
6943 gen_helper_neon_fcvt_f32_to_f16(tmp, cpu_F0s, cpu_env);
6944 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 1));
6945 gen_helper_neon_fcvt_f32_to_f16(tmp2, cpu_F0s, cpu_env);
6946 tcg_gen_shli_i32(tmp2, tmp2, 16);
6947 tcg_gen_or_i32(tmp2, tmp2, tmp);
6948 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 2));
6949 gen_helper_neon_fcvt_f32_to_f16(tmp, cpu_F0s, cpu_env);
6950 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 3));
6951 neon_store_reg(rd, 0, tmp2);
6952 tmp2 = tcg_temp_new_i32();
6953 gen_helper_neon_fcvt_f32_to_f16(tmp2, cpu_F0s, cpu_env);
6954 tcg_gen_shli_i32(tmp2, tmp2, 16);
6955 tcg_gen_or_i32(tmp2, tmp2, tmp);
6956 neon_store_reg(rd, 1, tmp2);
6957 tcg_temp_free_i32(tmp);
6958 break;
6959 case NEON_2RM_VCVT_F32_F16:
6960 if (!arm_dc_feature(s, ARM_FEATURE_VFP_FP16) ||
6961 q || (rd & 1)) {
6962 return 1;
6963 }
6964 tmp3 = tcg_temp_new_i32();
6965 tmp = neon_load_reg(rm, 0);
6966 tmp2 = neon_load_reg(rm, 1);
6967 tcg_gen_ext16u_i32(tmp3, tmp);
6968 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
6969 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 0));
6970 tcg_gen_shri_i32(tmp3, tmp, 16);
6971 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
6972 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 1));
6973 tcg_temp_free_i32(tmp);
6974 tcg_gen_ext16u_i32(tmp3, tmp2);
6975 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
6976 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 2));
6977 tcg_gen_shri_i32(tmp3, tmp2, 16);
6978 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
6979 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 3));
6980 tcg_temp_free_i32(tmp2);
6981 tcg_temp_free_i32(tmp3);
6982 break;
6983 case NEON_2RM_AESE: case NEON_2RM_AESMC:
6984 if (!arm_dc_feature(s, ARM_FEATURE_V8_AES)
6985 || ((rm | rd) & 1)) {
6986 return 1;
6987 }
6988 tmp = tcg_const_i32(rd);
6989 tmp2 = tcg_const_i32(rm);
6990
6991 /* Bit 6 is the lowest opcode bit; it distinguishes between
6992 * encryption (AESE/AESMC) and decryption (AESD/AESIMC)
6993 */
6994 tmp3 = tcg_const_i32(extract32(insn, 6, 1));
6995
6996 if (op == NEON_2RM_AESE) {
6997 gen_helper_crypto_aese(cpu_env, tmp, tmp2, tmp3);
6998 } else {
6999 gen_helper_crypto_aesmc(cpu_env, tmp, tmp2, tmp3);
7000 }
7001 tcg_temp_free_i32(tmp);
7002 tcg_temp_free_i32(tmp2);
7003 tcg_temp_free_i32(tmp3);
7004 break;
7005 case NEON_2RM_SHA1H:
7006 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA1)
7007 || ((rm | rd) & 1)) {
7008 return 1;
7009 }
7010 tmp = tcg_const_i32(rd);
7011 tmp2 = tcg_const_i32(rm);
7012
7013 gen_helper_crypto_sha1h(cpu_env, tmp, tmp2);
7014
7015 tcg_temp_free_i32(tmp);
7016 tcg_temp_free_i32(tmp2);
7017 break;
7018 case NEON_2RM_SHA1SU1:
7019 if ((rm | rd) & 1) {
7020 return 1;
7021 }
7022 /* bit 6 (q): set -> SHA256SU0, cleared -> SHA1SU1 */
7023 if (q) {
7024 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA256)) {
7025 return 1;
7026 }
7027 } else if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA1)) {
7028 return 1;
7029 }
7030 tmp = tcg_const_i32(rd);
7031 tmp2 = tcg_const_i32(rm);
7032 if (q) {
7033 gen_helper_crypto_sha256su0(cpu_env, tmp, tmp2);
7034 } else {
7035 gen_helper_crypto_sha1su1(cpu_env, tmp, tmp2);
7036 }
7037 tcg_temp_free_i32(tmp);
7038 tcg_temp_free_i32(tmp2);
7039 break;
7040 default:
7041 elementwise:
7042 for (pass = 0; pass < (q ? 4 : 2); pass++) {
7043 if (neon_2rm_is_float_op(op)) {
7044 tcg_gen_ld_f32(cpu_F0s, cpu_env,
7045 neon_reg_offset(rm, pass));
7046 TCGV_UNUSED_I32(tmp);
7047 } else {
7048 tmp = neon_load_reg(rm, pass);
7049 }
7050 switch (op) {
7051 case NEON_2RM_VREV32:
7052 switch (size) {
7053 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
7054 case 1: gen_swap_half(tmp); break;
7055 default: abort();
7056 }
7057 break;
7058 case NEON_2RM_VREV16:
7059 gen_rev16(tmp);
7060 break;
7061 case NEON_2RM_VCLS:
7062 switch (size) {
7063 case 0: gen_helper_neon_cls_s8(tmp, tmp); break;
7064 case 1: gen_helper_neon_cls_s16(tmp, tmp); break;
7065 case 2: gen_helper_neon_cls_s32(tmp, tmp); break;
7066 default: abort();
7067 }
7068 break;
7069 case NEON_2RM_VCLZ:
7070 switch (size) {
7071 case 0: gen_helper_neon_clz_u8(tmp, tmp); break;
7072 case 1: gen_helper_neon_clz_u16(tmp, tmp); break;
7073 case 2: gen_helper_clz(tmp, tmp); break;
7074 default: abort();
7075 }
7076 break;
7077 case NEON_2RM_VCNT:
7078 gen_helper_neon_cnt_u8(tmp, tmp);
7079 break;
7080 case NEON_2RM_VMVN:
7081 tcg_gen_not_i32(tmp, tmp);
7082 break;
7083 case NEON_2RM_VQABS:
7084 switch (size) {
7085 case 0:
7086 gen_helper_neon_qabs_s8(tmp, cpu_env, tmp);
7087 break;
7088 case 1:
7089 gen_helper_neon_qabs_s16(tmp, cpu_env, tmp);
7090 break;
7091 case 2:
7092 gen_helper_neon_qabs_s32(tmp, cpu_env, tmp);
7093 break;
7094 default: abort();
7095 }
7096 break;
7097 case NEON_2RM_VQNEG:
7098 switch (size) {
7099 case 0:
7100 gen_helper_neon_qneg_s8(tmp, cpu_env, tmp);
7101 break;
7102 case 1:
7103 gen_helper_neon_qneg_s16(tmp, cpu_env, tmp);
7104 break;
7105 case 2:
7106 gen_helper_neon_qneg_s32(tmp, cpu_env, tmp);
7107 break;
7108 default: abort();
7109 }
7110 break;
7111 case NEON_2RM_VCGT0: case NEON_2RM_VCLE0:
7112 tmp2 = tcg_const_i32(0);
7113 switch(size) {
7114 case 0: gen_helper_neon_cgt_s8(tmp, tmp, tmp2); break;
7115 case 1: gen_helper_neon_cgt_s16(tmp, tmp, tmp2); break;
7116 case 2: gen_helper_neon_cgt_s32(tmp, tmp, tmp2); break;
7117 default: abort();
7118 }
7119 tcg_temp_free_i32(tmp2);
7120 if (op == NEON_2RM_VCLE0) {
7121 tcg_gen_not_i32(tmp, tmp);
7122 }
7123 break;
7124 case NEON_2RM_VCGE0: case NEON_2RM_VCLT0:
7125 tmp2 = tcg_const_i32(0);
7126 switch(size) {
7127 case 0: gen_helper_neon_cge_s8(tmp, tmp, tmp2); break;
7128 case 1: gen_helper_neon_cge_s16(tmp, tmp, tmp2); break;
7129 case 2: gen_helper_neon_cge_s32(tmp, tmp, tmp2); break;
7130 default: abort();
7131 }
7132 tcg_temp_free_i32(tmp2);
7133 if (op == NEON_2RM_VCLT0) {
7134 tcg_gen_not_i32(tmp, tmp);
7135 }
7136 break;
7137 case NEON_2RM_VCEQ0:
7138 tmp2 = tcg_const_i32(0);
7139 switch(size) {
7140 case 0: gen_helper_neon_ceq_u8(tmp, tmp, tmp2); break;
7141 case 1: gen_helper_neon_ceq_u16(tmp, tmp, tmp2); break;
7142 case 2: gen_helper_neon_ceq_u32(tmp, tmp, tmp2); break;
7143 default: abort();
7144 }
7145 tcg_temp_free_i32(tmp2);
7146 break;
7147 case NEON_2RM_VABS:
7148 switch(size) {
7149 case 0: gen_helper_neon_abs_s8(tmp, tmp); break;
7150 case 1: gen_helper_neon_abs_s16(tmp, tmp); break;
7151 case 2: tcg_gen_abs_i32(tmp, tmp); break;
7152 default: abort();
7153 }
7154 break;
7155 case NEON_2RM_VNEG:
7156 tmp2 = tcg_const_i32(0);
7157 gen_neon_rsb(size, tmp, tmp2);
7158 tcg_temp_free_i32(tmp2);
7159 break;
7160 case NEON_2RM_VCGT0_F:
7161 {
7162 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7163 tmp2 = tcg_const_i32(0);
7164 gen_helper_neon_cgt_f32(tmp, tmp, tmp2, fpstatus);
7165 tcg_temp_free_i32(tmp2);
7166 tcg_temp_free_ptr(fpstatus);
7167 break;
7168 }
7169 case NEON_2RM_VCGE0_F:
7170 {
7171 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7172 tmp2 = tcg_const_i32(0);
7173 gen_helper_neon_cge_f32(tmp, tmp, tmp2, fpstatus);
7174 tcg_temp_free_i32(tmp2);
7175 tcg_temp_free_ptr(fpstatus);
7176 break;
7177 }
7178 case NEON_2RM_VCEQ0_F:
7179 {
7180 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7181 tmp2 = tcg_const_i32(0);
7182 gen_helper_neon_ceq_f32(tmp, tmp, tmp2, fpstatus);
7183 tcg_temp_free_i32(tmp2);
7184 tcg_temp_free_ptr(fpstatus);
7185 break;
7186 }
7187 case NEON_2RM_VCLE0_F:
7188 {
7189 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7190 tmp2 = tcg_const_i32(0);
7191 gen_helper_neon_cge_f32(tmp, tmp2, tmp, fpstatus);
7192 tcg_temp_free_i32(tmp2);
7193 tcg_temp_free_ptr(fpstatus);
7194 break;
7195 }
7196 case NEON_2RM_VCLT0_F:
7197 {
7198 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7199 tmp2 = tcg_const_i32(0);
7200 gen_helper_neon_cgt_f32(tmp, tmp2, tmp, fpstatus);
7201 tcg_temp_free_i32(tmp2);
7202 tcg_temp_free_ptr(fpstatus);
7203 break;
7204 }
7205 case NEON_2RM_VABS_F:
7206 gen_vfp_abs(0);
7207 break;
7208 case NEON_2RM_VNEG_F:
7209 gen_vfp_neg(0);
7210 break;
7211 case NEON_2RM_VSWP:
7212 tmp2 = neon_load_reg(rd, pass);
7213 neon_store_reg(rm, pass, tmp2);
7214 break;
7215 case NEON_2RM_VTRN:
7216 tmp2 = neon_load_reg(rd, pass);
7217 switch (size) {
7218 case 0: gen_neon_trn_u8(tmp, tmp2); break;
7219 case 1: gen_neon_trn_u16(tmp, tmp2); break;
7220 default: abort();
7221 }
7222 neon_store_reg(rm, pass, tmp2);
7223 break;
7224 case NEON_2RM_VRINTN:
7225 case NEON_2RM_VRINTA:
7226 case NEON_2RM_VRINTM:
7227 case NEON_2RM_VRINTP:
7228 case NEON_2RM_VRINTZ:
7229 {
7230 TCGv_i32 tcg_rmode;
7231 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7232 int rmode;
7233
7234 if (op == NEON_2RM_VRINTZ) {
7235 rmode = FPROUNDING_ZERO;
7236 } else {
7237 rmode = fp_decode_rm[((op & 0x6) >> 1) ^ 1];
7238 }
7239
7240 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
7241 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
7242 cpu_env);
7243 gen_helper_rints(cpu_F0s, cpu_F0s, fpstatus);
7244 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
7245 cpu_env);
7246 tcg_temp_free_ptr(fpstatus);
7247 tcg_temp_free_i32(tcg_rmode);
7248 break;
7249 }
7250 case NEON_2RM_VRINTX:
7251 {
7252 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7253 gen_helper_rints_exact(cpu_F0s, cpu_F0s, fpstatus);
7254 tcg_temp_free_ptr(fpstatus);
7255 break;
7256 }
7257 case NEON_2RM_VCVTAU:
7258 case NEON_2RM_VCVTAS:
7259 case NEON_2RM_VCVTNU:
7260 case NEON_2RM_VCVTNS:
7261 case NEON_2RM_VCVTPU:
7262 case NEON_2RM_VCVTPS:
7263 case NEON_2RM_VCVTMU:
7264 case NEON_2RM_VCVTMS:
7265 {
7266 bool is_signed = !extract32(insn, 7, 1);
7267 TCGv_ptr fpst = get_fpstatus_ptr(1);
7268 TCGv_i32 tcg_rmode, tcg_shift;
7269 int rmode = fp_decode_rm[extract32(insn, 8, 2)];
7270
7271 tcg_shift = tcg_const_i32(0);
7272 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
7273 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
7274 cpu_env);
7275
7276 if (is_signed) {
7277 gen_helper_vfp_tosls(cpu_F0s, cpu_F0s,
7278 tcg_shift, fpst);
7279 } else {
7280 gen_helper_vfp_touls(cpu_F0s, cpu_F0s,
7281 tcg_shift, fpst);
7282 }
7283
7284 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
7285 cpu_env);
7286 tcg_temp_free_i32(tcg_rmode);
7287 tcg_temp_free_i32(tcg_shift);
7288 tcg_temp_free_ptr(fpst);
7289 break;
7290 }
7291 case NEON_2RM_VRECPE:
7292 {
7293 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7294 gen_helper_recpe_u32(tmp, tmp, fpstatus);
7295 tcg_temp_free_ptr(fpstatus);
7296 break;
7297 }
7298 case NEON_2RM_VRSQRTE:
7299 {
7300 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7301 gen_helper_rsqrte_u32(tmp, tmp, fpstatus);
7302 tcg_temp_free_ptr(fpstatus);
7303 break;
7304 }
7305 case NEON_2RM_VRECPE_F:
7306 {
7307 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7308 gen_helper_recpe_f32(cpu_F0s, cpu_F0s, fpstatus);
7309 tcg_temp_free_ptr(fpstatus);
7310 break;
7311 }
7312 case NEON_2RM_VRSQRTE_F:
7313 {
7314 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7315 gen_helper_rsqrte_f32(cpu_F0s, cpu_F0s, fpstatus);
7316 tcg_temp_free_ptr(fpstatus);
7317 break;
7318 }
7319 case NEON_2RM_VCVT_FS: /* VCVT.F32.S32 */
7320 gen_vfp_sito(0, 1);
7321 break;
7322 case NEON_2RM_VCVT_FU: /* VCVT.F32.U32 */
7323 gen_vfp_uito(0, 1);
7324 break;
7325 case NEON_2RM_VCVT_SF: /* VCVT.S32.F32 */
7326 gen_vfp_tosiz(0, 1);
7327 break;
7328 case NEON_2RM_VCVT_UF: /* VCVT.U32.F32 */
7329 gen_vfp_touiz(0, 1);
7330 break;
7331 default:
7332 /* Reserved op values were caught by the
7333 * neon_2rm_sizes[] check earlier.
7334 */
7335 abort();
7336 }
7337 if (neon_2rm_is_float_op(op)) {
7338 tcg_gen_st_f32(cpu_F0s, cpu_env,
7339 neon_reg_offset(rd, pass));
7340 } else {
7341 neon_store_reg(rd, pass, tmp);
7342 }
7343 }
7344 break;
7345 }
7346 } else if ((insn & (1 << 10)) == 0) {
7347 /* VTBL, VTBX. */
7348 int n = ((insn >> 8) & 3) + 1;
7349 if ((rn + n) > 32) {
7350 /* This is UNPREDICTABLE; we choose to UNDEF to avoid the
7351 * helper function running off the end of the register file.
7352 */
7353 return 1;
7354 }
7355 n <<= 3;
7356 if (insn & (1 << 6)) {
7357 tmp = neon_load_reg(rd, 0);
7358 } else {
7359 tmp = tcg_temp_new_i32();
7360 tcg_gen_movi_i32(tmp, 0);
7361 }
7362 tmp2 = neon_load_reg(rm, 0);
7363 tmp4 = tcg_const_i32(rn);
7364 tmp5 = tcg_const_i32(n);
7365 gen_helper_neon_tbl(tmp2, cpu_env, tmp2, tmp, tmp4, tmp5);
7366 tcg_temp_free_i32(tmp);
7367 if (insn & (1 << 6)) {
7368 tmp = neon_load_reg(rd, 1);
7369 } else {
7370 tmp = tcg_temp_new_i32();
7371 tcg_gen_movi_i32(tmp, 0);
7372 }
7373 tmp3 = neon_load_reg(rm, 1);
7374 gen_helper_neon_tbl(tmp3, cpu_env, tmp3, tmp, tmp4, tmp5);
7375 tcg_temp_free_i32(tmp5);
7376 tcg_temp_free_i32(tmp4);
7377 neon_store_reg(rd, 0, tmp2);
7378 neon_store_reg(rd, 1, tmp3);
7379 tcg_temp_free_i32(tmp);
7380 } else if ((insn & 0x380) == 0) {
7381 /* VDUP */
7382 if ((insn & (7 << 16)) == 0 || (q && (rd & 1))) {
7383 return 1;
7384 }
7385 if (insn & (1 << 19)) {
7386 tmp = neon_load_reg(rm, 1);
7387 } else {
7388 tmp = neon_load_reg(rm, 0);
7389 }
7390 if (insn & (1 << 16)) {
7391 gen_neon_dup_u8(tmp, ((insn >> 17) & 3) * 8);
7392 } else if (insn & (1 << 17)) {
7393 if ((insn >> 18) & 1)
7394 gen_neon_dup_high16(tmp);
7395 else
7396 gen_neon_dup_low16(tmp);
7397 }
7398 for (pass = 0; pass < (q ? 4 : 2); pass++) {
7399 tmp2 = tcg_temp_new_i32();
7400 tcg_gen_mov_i32(tmp2, tmp);
7401 neon_store_reg(rd, pass, tmp2);
7402 }
7403 tcg_temp_free_i32(tmp);
7404 } else {
7405 return 1;
7406 }
7407 }
7408 }
7409 return 0;
7410 }
7411
7412 static int disas_coproc_insn(DisasContext *s, uint32_t insn)
7413 {
7414 int cpnum, is64, crn, crm, opc1, opc2, isread, rt, rt2;
7415 const ARMCPRegInfo *ri;
7416
7417 cpnum = (insn >> 8) & 0xf;
7418
7419 /* First check for coprocessor space used for XScale/iwMMXt insns */
7420 if (arm_dc_feature(s, ARM_FEATURE_XSCALE) && (cpnum < 2)) {
7421 if (extract32(s->c15_cpar, cpnum, 1) == 0) {
7422 return 1;
7423 }
7424 if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
7425 return disas_iwmmxt_insn(s, insn);
7426 } else if (arm_dc_feature(s, ARM_FEATURE_XSCALE)) {
7427 return disas_dsp_insn(s, insn);
7428 }
7429 return 1;
7430 }
7431
7432 /* Otherwise treat as a generic register access */
7433 is64 = (insn & (1 << 25)) == 0;
7434 if (!is64 && ((insn & (1 << 4)) == 0)) {
7435 /* cdp */
7436 return 1;
7437 }
7438
7439 crm = insn & 0xf;
7440 if (is64) {
7441 crn = 0;
7442 opc1 = (insn >> 4) & 0xf;
7443 opc2 = 0;
7444 rt2 = (insn >> 16) & 0xf;
7445 } else {
7446 crn = (insn >> 16) & 0xf;
7447 opc1 = (insn >> 21) & 7;
7448 opc2 = (insn >> 5) & 7;
7449 rt2 = 0;
7450 }
7451 isread = (insn >> 20) & 1;
7452 rt = (insn >> 12) & 0xf;
7453
7454 ri = get_arm_cp_reginfo(s->cp_regs,
7455 ENCODE_CP_REG(cpnum, is64, s->ns, crn, crm, opc1, opc2));
7456 if (ri) {
7457 /* Check access permissions */
7458 if (!cp_access_ok(s->current_el, ri, isread)) {
7459 return 1;
7460 }
7461
7462 if (ri->accessfn ||
7463 (arm_dc_feature(s, ARM_FEATURE_XSCALE) && cpnum < 14)) {
7464 /* Emit code to perform further access permissions checks at
7465 * runtime; this may result in an exception.
7466 * Note that on XScale all cp0..c13 registers do an access check
7467 * call in order to handle c15_cpar.
7468 */
7469 TCGv_ptr tmpptr;
7470 TCGv_i32 tcg_syn, tcg_isread;
7471 uint32_t syndrome;
7472
7473 /* Note that since we are an implementation which takes an
7474 * exception on a trapped conditional instruction only if the
7475 * instruction passes its condition code check, we can take
7476 * advantage of the clause in the ARM ARM that allows us to set
7477 * the COND field in the instruction to 0xE in all cases.
7478 * We could fish the actual condition out of the insn (ARM)
7479 * or the condexec bits (Thumb) but it isn't necessary.
7480 */
7481 switch (cpnum) {
7482 case 14:
7483 if (is64) {
7484 syndrome = syn_cp14_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
7485 isread, false);
7486 } else {
7487 syndrome = syn_cp14_rt_trap(1, 0xe, opc1, opc2, crn, crm,
7488 rt, isread, false);
7489 }
7490 break;
7491 case 15:
7492 if (is64) {
7493 syndrome = syn_cp15_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
7494 isread, false);
7495 } else {
7496 syndrome = syn_cp15_rt_trap(1, 0xe, opc1, opc2, crn, crm,
7497 rt, isread, false);
7498 }
7499 break;
7500 default:
7501 /* ARMv8 defines that only coprocessors 14 and 15 exist,
7502 * so this can only happen if this is an ARMv7 or earlier CPU,
7503 * in which case the syndrome information won't actually be
7504 * guest visible.
7505 */
7506 assert(!arm_dc_feature(s, ARM_FEATURE_V8));
7507 syndrome = syn_uncategorized();
7508 break;
7509 }
7510
7511 gen_set_condexec(s);
7512 gen_set_pc_im(s, s->pc - 4);
7513 tmpptr = tcg_const_ptr(ri);
7514 tcg_syn = tcg_const_i32(syndrome);
7515 tcg_isread = tcg_const_i32(isread);
7516 gen_helper_access_check_cp_reg(cpu_env, tmpptr, tcg_syn,
7517 tcg_isread);
7518 tcg_temp_free_ptr(tmpptr);
7519 tcg_temp_free_i32(tcg_syn);
7520 tcg_temp_free_i32(tcg_isread);
7521 }
7522
7523 /* Handle special cases first */
7524 switch (ri->type & ~(ARM_CP_FLAG_MASK & ~ARM_CP_SPECIAL)) {
7525 case ARM_CP_NOP:
7526 return 0;
7527 case ARM_CP_WFI:
7528 if (isread) {
7529 return 1;
7530 }
7531 gen_set_pc_im(s, s->pc);
7532 s->is_jmp = DISAS_WFI;
7533 return 0;
7534 default:
7535 break;
7536 }
7537
7538 if ((s->tb->cflags & CF_USE_ICOUNT) && (ri->type & ARM_CP_IO)) {
7539 gen_io_start();
7540 }
7541
7542 if (isread) {
7543 /* Read */
7544 if (is64) {
7545 TCGv_i64 tmp64;
7546 TCGv_i32 tmp;
7547 if (ri->type & ARM_CP_CONST) {
7548 tmp64 = tcg_const_i64(ri->resetvalue);
7549 } else if (ri->readfn) {
7550 TCGv_ptr tmpptr;
7551 tmp64 = tcg_temp_new_i64();
7552 tmpptr = tcg_const_ptr(ri);
7553 gen_helper_get_cp_reg64(tmp64, cpu_env, tmpptr);
7554 tcg_temp_free_ptr(tmpptr);
7555 } else {
7556 tmp64 = tcg_temp_new_i64();
7557 tcg_gen_ld_i64(tmp64, cpu_env, ri->fieldoffset);
7558 }
7559 tmp = tcg_temp_new_i32();
7560 tcg_gen_extrl_i64_i32(tmp, tmp64);
7561 store_reg(s, rt, tmp);
7562 tcg_gen_shri_i64(tmp64, tmp64, 32);
7563 tmp = tcg_temp_new_i32();
7564 tcg_gen_extrl_i64_i32(tmp, tmp64);
7565 tcg_temp_free_i64(tmp64);
7566 store_reg(s, rt2, tmp);
7567 } else {
7568 TCGv_i32 tmp;
7569 if (ri->type & ARM_CP_CONST) {
7570 tmp = tcg_const_i32(ri->resetvalue);
7571 } else if (ri->readfn) {
7572 TCGv_ptr tmpptr;
7573 tmp = tcg_temp_new_i32();
7574 tmpptr = tcg_const_ptr(ri);
7575 gen_helper_get_cp_reg(tmp, cpu_env, tmpptr);
7576 tcg_temp_free_ptr(tmpptr);
7577 } else {
7578 tmp = load_cpu_offset(ri->fieldoffset);
7579 }
7580 if (rt == 15) {
7581 /* Destination register of r15 for 32 bit loads sets
7582 * the condition codes from the high 4 bits of the value
7583 */
7584 gen_set_nzcv(tmp);
7585 tcg_temp_free_i32(tmp);
7586 } else {
7587 store_reg(s, rt, tmp);
7588 }
7589 }
7590 } else {
7591 /* Write */
7592 if (ri->type & ARM_CP_CONST) {
7593 /* If not forbidden by access permissions, treat as WI */
7594 return 0;
7595 }
7596
7597 if (is64) {
7598 TCGv_i32 tmplo, tmphi;
7599 TCGv_i64 tmp64 = tcg_temp_new_i64();
7600 tmplo = load_reg(s, rt);
7601 tmphi = load_reg(s, rt2);
7602 tcg_gen_concat_i32_i64(tmp64, tmplo, tmphi);
7603 tcg_temp_free_i32(tmplo);
7604 tcg_temp_free_i32(tmphi);
7605 if (ri->writefn) {
7606 TCGv_ptr tmpptr = tcg_const_ptr(ri);
7607 gen_helper_set_cp_reg64(cpu_env, tmpptr, tmp64);
7608 tcg_temp_free_ptr(tmpptr);
7609 } else {
7610 tcg_gen_st_i64(tmp64, cpu_env, ri->fieldoffset);
7611 }
7612 tcg_temp_free_i64(tmp64);
7613 } else {
7614 if (ri->writefn) {
7615 TCGv_i32 tmp;
7616 TCGv_ptr tmpptr;
7617 tmp = load_reg(s, rt);
7618 tmpptr = tcg_const_ptr(ri);
7619 gen_helper_set_cp_reg(cpu_env, tmpptr, tmp);
7620 tcg_temp_free_ptr(tmpptr);
7621 tcg_temp_free_i32(tmp);
7622 } else {
7623 TCGv_i32 tmp = load_reg(s, rt);
7624 store_cpu_offset(tmp, ri->fieldoffset);
7625 }
7626 }
7627 }
7628
7629 if ((s->tb->cflags & CF_USE_ICOUNT) && (ri->type & ARM_CP_IO)) {
7630 /* I/O operations must end the TB here (whether read or write) */
7631 gen_io_end();
7632 gen_lookup_tb(s);
7633 } else if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
7634 /* We default to ending the TB on a coprocessor register write,
7635 * but allow this to be suppressed by the register definition
7636 * (usually only necessary to work around guest bugs).
7637 */
7638 gen_lookup_tb(s);
7639 }
7640
7641 return 0;
7642 }
7643
7644 /* Unknown register; this might be a guest error or a QEMU
7645 * unimplemented feature.
7646 */
7647 if (is64) {
7648 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
7649 "64 bit system register cp:%d opc1: %d crm:%d "
7650 "(%s)\n",
7651 isread ? "read" : "write", cpnum, opc1, crm,
7652 s->ns ? "non-secure" : "secure");
7653 } else {
7654 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
7655 "system register cp:%d opc1:%d crn:%d crm:%d opc2:%d "
7656 "(%s)\n",
7657 isread ? "read" : "write", cpnum, opc1, crn, crm, opc2,
7658 s->ns ? "non-secure" : "secure");
7659 }
7660
7661 return 1;
7662 }
7663
7664
7665 /* Store a 64-bit value to a register pair. Clobbers val. */
7666 static void gen_storeq_reg(DisasContext *s, int rlow, int rhigh, TCGv_i64 val)
7667 {
7668 TCGv_i32 tmp;
7669 tmp = tcg_temp_new_i32();
7670 tcg_gen_extrl_i64_i32(tmp, val);
7671 store_reg(s, rlow, tmp);
7672 tmp = tcg_temp_new_i32();
7673 tcg_gen_shri_i64(val, val, 32);
7674 tcg_gen_extrl_i64_i32(tmp, val);
7675 store_reg(s, rhigh, tmp);
7676 }
7677
7678 /* load a 32-bit value from a register and perform a 64-bit accumulate. */
7679 static void gen_addq_lo(DisasContext *s, TCGv_i64 val, int rlow)
7680 {
7681 TCGv_i64 tmp;
7682 TCGv_i32 tmp2;
7683
7684 /* Load value and extend to 64 bits. */
7685 tmp = tcg_temp_new_i64();
7686 tmp2 = load_reg(s, rlow);
7687 tcg_gen_extu_i32_i64(tmp, tmp2);
7688 tcg_temp_free_i32(tmp2);
7689 tcg_gen_add_i64(val, val, tmp);
7690 tcg_temp_free_i64(tmp);
7691 }
7692
7693 /* load and add a 64-bit value from a register pair. */
7694 static void gen_addq(DisasContext *s, TCGv_i64 val, int rlow, int rhigh)
7695 {
7696 TCGv_i64 tmp;
7697 TCGv_i32 tmpl;
7698 TCGv_i32 tmph;
7699
7700 /* Load 64-bit value rd:rn. */
7701 tmpl = load_reg(s, rlow);
7702 tmph = load_reg(s, rhigh);
7703 tmp = tcg_temp_new_i64();
7704 tcg_gen_concat_i32_i64(tmp, tmpl, tmph);
7705 tcg_temp_free_i32(tmpl);
7706 tcg_temp_free_i32(tmph);
7707 tcg_gen_add_i64(val, val, tmp);
7708 tcg_temp_free_i64(tmp);
7709 }
7710
7711 /* Set N and Z flags from hi|lo. */
7712 static void gen_logicq_cc(TCGv_i32 lo, TCGv_i32 hi)
7713 {
7714 tcg_gen_mov_i32(cpu_NF, hi);
7715 tcg_gen_or_i32(cpu_ZF, lo, hi);
7716 }
7717
7718 /* Load/Store exclusive instructions are implemented by remembering
7719 the value/address loaded, and seeing if these are the same
7720 when the store is performed. This should be sufficient to implement
7721 the architecturally mandated semantics, and avoids having to monitor
7722 regular stores.
7723
7724 In system emulation mode only one CPU will be running at once, so
7725 this sequence is effectively atomic. In user emulation mode we
7726 throw an exception and handle the atomic operation elsewhere. */
7727 static void gen_load_exclusive(DisasContext *s, int rt, int rt2,
7728 TCGv_i32 addr, int size)
7729 {
7730 TCGv_i32 tmp = tcg_temp_new_i32();
7731
7732 s->is_ldex = true;
7733
7734 switch (size) {
7735 case 0:
7736 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
7737 break;
7738 case 1:
7739 gen_aa32_ld16ua(s, tmp, addr, get_mem_index(s));
7740 break;
7741 case 2:
7742 case 3:
7743 gen_aa32_ld32ua(s, tmp, addr, get_mem_index(s));
7744 break;
7745 default:
7746 abort();
7747 }
7748
7749 if (size == 3) {
7750 TCGv_i32 tmp2 = tcg_temp_new_i32();
7751 TCGv_i32 tmp3 = tcg_temp_new_i32();
7752
7753 tcg_gen_addi_i32(tmp2, addr, 4);
7754 gen_aa32_ld32u(s, tmp3, tmp2, get_mem_index(s));
7755 tcg_temp_free_i32(tmp2);
7756 tcg_gen_concat_i32_i64(cpu_exclusive_val, tmp, tmp3);
7757 store_reg(s, rt2, tmp3);
7758 } else {
7759 tcg_gen_extu_i32_i64(cpu_exclusive_val, tmp);
7760 }
7761
7762 store_reg(s, rt, tmp);
7763 tcg_gen_extu_i32_i64(cpu_exclusive_addr, addr);
7764 }
7765
7766 static void gen_clrex(DisasContext *s)
7767 {
7768 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
7769 }
7770
7771 #ifdef CONFIG_USER_ONLY
7772 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
7773 TCGv_i32 addr, int size)
7774 {
7775 tcg_gen_extu_i32_i64(cpu_exclusive_test, addr);
7776 tcg_gen_movi_i32(cpu_exclusive_info,
7777 size | (rd << 4) | (rt << 8) | (rt2 << 12));
7778 gen_exception_internal_insn(s, 4, EXCP_STREX);
7779 }
7780 #else
7781 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
7782 TCGv_i32 addr, int size)
7783 {
7784 TCGv_i32 tmp;
7785 TCGv_i64 val64, extaddr;
7786 TCGLabel *done_label;
7787 TCGLabel *fail_label;
7788
7789 /* if (env->exclusive_addr == addr && env->exclusive_val == [addr]) {
7790 [addr] = {Rt};
7791 {Rd} = 0;
7792 } else {
7793 {Rd} = 1;
7794 } */
7795 fail_label = gen_new_label();
7796 done_label = gen_new_label();
7797 extaddr = tcg_temp_new_i64();
7798 tcg_gen_extu_i32_i64(extaddr, addr);
7799 tcg_gen_brcond_i64(TCG_COND_NE, extaddr, cpu_exclusive_addr, fail_label);
7800 tcg_temp_free_i64(extaddr);
7801
7802 tmp = tcg_temp_new_i32();
7803 switch (size) {
7804 case 0:
7805 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
7806 break;
7807 case 1:
7808 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
7809 break;
7810 case 2:
7811 case 3:
7812 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
7813 break;
7814 default:
7815 abort();
7816 }
7817
7818 val64 = tcg_temp_new_i64();
7819 if (size == 3) {
7820 TCGv_i32 tmp2 = tcg_temp_new_i32();
7821 TCGv_i32 tmp3 = tcg_temp_new_i32();
7822 tcg_gen_addi_i32(tmp2, addr, 4);
7823 gen_aa32_ld32u(s, tmp3, tmp2, get_mem_index(s));
7824 tcg_temp_free_i32(tmp2);
7825 tcg_gen_concat_i32_i64(val64, tmp, tmp3);
7826 tcg_temp_free_i32(tmp3);
7827 } else {
7828 tcg_gen_extu_i32_i64(val64, tmp);
7829 }
7830 tcg_temp_free_i32(tmp);
7831
7832 tcg_gen_brcond_i64(TCG_COND_NE, val64, cpu_exclusive_val, fail_label);
7833 tcg_temp_free_i64(val64);
7834
7835 tmp = load_reg(s, rt);
7836 switch (size) {
7837 case 0:
7838 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
7839 break;
7840 case 1:
7841 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
7842 break;
7843 case 2:
7844 case 3:
7845 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
7846 break;
7847 default:
7848 abort();
7849 }
7850 tcg_temp_free_i32(tmp);
7851 if (size == 3) {
7852 tcg_gen_addi_i32(addr, addr, 4);
7853 tmp = load_reg(s, rt2);
7854 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
7855 tcg_temp_free_i32(tmp);
7856 }
7857 tcg_gen_movi_i32(cpu_R[rd], 0);
7858 tcg_gen_br(done_label);
7859 gen_set_label(fail_label);
7860 tcg_gen_movi_i32(cpu_R[rd], 1);
7861 gen_set_label(done_label);
7862 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
7863 }
7864 #endif
7865
7866 /* gen_srs:
7867 * @env: CPUARMState
7868 * @s: DisasContext
7869 * @mode: mode field from insn (which stack to store to)
7870 * @amode: addressing mode (DA/IA/DB/IB), encoded as per P,U bits in ARM insn
7871 * @writeback: true if writeback bit set
7872 *
7873 * Generate code for the SRS (Store Return State) insn.
7874 */
7875 static void gen_srs(DisasContext *s,
7876 uint32_t mode, uint32_t amode, bool writeback)
7877 {
7878 int32_t offset;
7879 TCGv_i32 addr, tmp;
7880 bool undef = false;
7881
7882 /* SRS is:
7883 * - trapped to EL3 if EL3 is AArch64 and we are at Secure EL1
7884 * and specified mode is monitor mode
7885 * - UNDEFINED in Hyp mode
7886 * - UNPREDICTABLE in User or System mode
7887 * - UNPREDICTABLE if the specified mode is:
7888 * -- not implemented
7889 * -- not a valid mode number
7890 * -- a mode that's at a higher exception level
7891 * -- Monitor, if we are Non-secure
7892 * For the UNPREDICTABLE cases we choose to UNDEF.
7893 */
7894 if (s->current_el == 1 && !s->ns && mode == ARM_CPU_MODE_MON) {
7895 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(), 3);
7896 return;
7897 }
7898
7899 if (s->current_el == 0 || s->current_el == 2) {
7900 undef = true;
7901 }
7902
7903 switch (mode) {
7904 case ARM_CPU_MODE_USR:
7905 case ARM_CPU_MODE_FIQ:
7906 case ARM_CPU_MODE_IRQ:
7907 case ARM_CPU_MODE_SVC:
7908 case ARM_CPU_MODE_ABT:
7909 case ARM_CPU_MODE_UND:
7910 case ARM_CPU_MODE_SYS:
7911 break;
7912 case ARM_CPU_MODE_HYP:
7913 if (s->current_el == 1 || !arm_dc_feature(s, ARM_FEATURE_EL2)) {
7914 undef = true;
7915 }
7916 break;
7917 case ARM_CPU_MODE_MON:
7918 /* No need to check specifically for "are we non-secure" because
7919 * we've already made EL0 UNDEF and handled the trap for S-EL1;
7920 * so if this isn't EL3 then we must be non-secure.
7921 */
7922 if (s->current_el != 3) {
7923 undef = true;
7924 }
7925 break;
7926 default:
7927 undef = true;
7928 }
7929
7930 if (undef) {
7931 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(),
7932 default_exception_el(s));
7933 return;
7934 }
7935
7936 addr = tcg_temp_new_i32();
7937 tmp = tcg_const_i32(mode);
7938 /* get_r13_banked() will raise an exception if called from System mode */
7939 gen_set_condexec(s);
7940 gen_set_pc_im(s, s->pc - 4);
7941 gen_helper_get_r13_banked(addr, cpu_env, tmp);
7942 tcg_temp_free_i32(tmp);
7943 switch (amode) {
7944 case 0: /* DA */
7945 offset = -4;
7946 break;
7947 case 1: /* IA */
7948 offset = 0;
7949 break;
7950 case 2: /* DB */
7951 offset = -8;
7952 break;
7953 case 3: /* IB */
7954 offset = 4;
7955 break;
7956 default:
7957 abort();
7958 }
7959 tcg_gen_addi_i32(addr, addr, offset);
7960 tmp = load_reg(s, 14);
7961 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
7962 tcg_temp_free_i32(tmp);
7963 tmp = load_cpu_field(spsr);
7964 tcg_gen_addi_i32(addr, addr, 4);
7965 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
7966 tcg_temp_free_i32(tmp);
7967 if (writeback) {
7968 switch (amode) {
7969 case 0:
7970 offset = -8;
7971 break;
7972 case 1:
7973 offset = 4;
7974 break;
7975 case 2:
7976 offset = -4;
7977 break;
7978 case 3:
7979 offset = 0;
7980 break;
7981 default:
7982 abort();
7983 }
7984 tcg_gen_addi_i32(addr, addr, offset);
7985 tmp = tcg_const_i32(mode);
7986 gen_helper_set_r13_banked(cpu_env, tmp, addr);
7987 tcg_temp_free_i32(tmp);
7988 }
7989 tcg_temp_free_i32(addr);
7990 s->is_jmp = DISAS_UPDATE;
7991 }
7992
7993 static void disas_arm_insn(DisasContext *s, unsigned int insn)
7994 {
7995 unsigned int cond, val, op1, i, shift, rm, rs, rn, rd, sh;
7996 TCGv_i32 tmp;
7997 TCGv_i32 tmp2;
7998 TCGv_i32 tmp3;
7999 TCGv_i32 addr;
8000 TCGv_i64 tmp64;
8001
8002 /* M variants do not implement ARM mode. */
8003 if (arm_dc_feature(s, ARM_FEATURE_M)) {
8004 goto illegal_op;
8005 }
8006 cond = insn >> 28;
8007 if (cond == 0xf){
8008 /* In ARMv3 and v4 the NV condition is UNPREDICTABLE; we
8009 * choose to UNDEF. In ARMv5 and above the space is used
8010 * for miscellaneous unconditional instructions.
8011 */
8012 ARCH(5);
8013
8014 /* Unconditional instructions. */
8015 if (((insn >> 25) & 7) == 1) {
8016 /* NEON Data processing. */
8017 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
8018 goto illegal_op;
8019 }
8020
8021 if (disas_neon_data_insn(s, insn)) {
8022 goto illegal_op;
8023 }
8024 return;
8025 }
8026 if ((insn & 0x0f100000) == 0x04000000) {
8027 /* NEON load/store. */
8028 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
8029 goto illegal_op;
8030 }
8031
8032 if (disas_neon_ls_insn(s, insn)) {
8033 goto illegal_op;
8034 }
8035 return;
8036 }
8037 if ((insn & 0x0f000e10) == 0x0e000a00) {
8038 /* VFP. */
8039 if (disas_vfp_insn(s, insn)) {
8040 goto illegal_op;
8041 }
8042 return;
8043 }
8044 if (((insn & 0x0f30f000) == 0x0510f000) ||
8045 ((insn & 0x0f30f010) == 0x0710f000)) {
8046 if ((insn & (1 << 22)) == 0) {
8047 /* PLDW; v7MP */
8048 if (!arm_dc_feature(s, ARM_FEATURE_V7MP)) {
8049 goto illegal_op;
8050 }
8051 }
8052 /* Otherwise PLD; v5TE+ */
8053 ARCH(5TE);
8054 return;
8055 }
8056 if (((insn & 0x0f70f000) == 0x0450f000) ||
8057 ((insn & 0x0f70f010) == 0x0650f000)) {
8058 ARCH(7);
8059 return; /* PLI; V7 */
8060 }
8061 if (((insn & 0x0f700000) == 0x04100000) ||
8062 ((insn & 0x0f700010) == 0x06100000)) {
8063 if (!arm_dc_feature(s, ARM_FEATURE_V7MP)) {
8064 goto illegal_op;
8065 }
8066 return; /* v7MP: Unallocated memory hint: must NOP */
8067 }
8068
8069 if ((insn & 0x0ffffdff) == 0x01010000) {
8070 ARCH(6);
8071 /* setend */
8072 if (((insn >> 9) & 1) != !!(s->be_data == MO_BE)) {
8073 gen_helper_setend(cpu_env);
8074 s->is_jmp = DISAS_UPDATE;
8075 }
8076 return;
8077 } else if ((insn & 0x0fffff00) == 0x057ff000) {
8078 switch ((insn >> 4) & 0xf) {
8079 case 1: /* clrex */
8080 ARCH(6K);
8081 gen_clrex(s);
8082 return;
8083 case 4: /* dsb */
8084 case 5: /* dmb */
8085 ARCH(7);
8086 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
8087 return;
8088 case 6: /* isb */
8089 /* We need to break the TB after this insn to execute
8090 * self-modifying code correctly and also to take
8091 * any pending interrupts immediately.
8092 */
8093 gen_lookup_tb(s);
8094 return;
8095 default:
8096 goto illegal_op;
8097 }
8098 } else if ((insn & 0x0e5fffe0) == 0x084d0500) {
8099 /* srs */
8100 ARCH(6);
8101 gen_srs(s, (insn & 0x1f), (insn >> 23) & 3, insn & (1 << 21));
8102 return;
8103 } else if ((insn & 0x0e50ffe0) == 0x08100a00) {
8104 /* rfe */
8105 int32_t offset;
8106 if (IS_USER(s))
8107 goto illegal_op;
8108 ARCH(6);
8109 rn = (insn >> 16) & 0xf;
8110 addr = load_reg(s, rn);
8111 i = (insn >> 23) & 3;
8112 switch (i) {
8113 case 0: offset = -4; break; /* DA */
8114 case 1: offset = 0; break; /* IA */
8115 case 2: offset = -8; break; /* DB */
8116 case 3: offset = 4; break; /* IB */
8117 default: abort();
8118 }
8119 if (offset)
8120 tcg_gen_addi_i32(addr, addr, offset);
8121 /* Load PC into tmp and CPSR into tmp2. */
8122 tmp = tcg_temp_new_i32();
8123 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
8124 tcg_gen_addi_i32(addr, addr, 4);
8125 tmp2 = tcg_temp_new_i32();
8126 gen_aa32_ld32u(s, tmp2, addr, get_mem_index(s));
8127 if (insn & (1 << 21)) {
8128 /* Base writeback. */
8129 switch (i) {
8130 case 0: offset = -8; break;
8131 case 1: offset = 4; break;
8132 case 2: offset = -4; break;
8133 case 3: offset = 0; break;
8134 default: abort();
8135 }
8136 if (offset)
8137 tcg_gen_addi_i32(addr, addr, offset);
8138 store_reg(s, rn, addr);
8139 } else {
8140 tcg_temp_free_i32(addr);
8141 }
8142 gen_rfe(s, tmp, tmp2);
8143 return;
8144 } else if ((insn & 0x0e000000) == 0x0a000000) {
8145 /* branch link and change to thumb (blx <offset>) */
8146 int32_t offset;
8147
8148 val = (uint32_t)s->pc;
8149 tmp = tcg_temp_new_i32();
8150 tcg_gen_movi_i32(tmp, val);
8151 store_reg(s, 14, tmp);
8152 /* Sign-extend the 24-bit offset */
8153 offset = (((int32_t)insn) << 8) >> 8;
8154 /* offset * 4 + bit24 * 2 + (thumb bit) */
8155 val += (offset << 2) | ((insn >> 23) & 2) | 1;
8156 /* pipeline offset */
8157 val += 4;
8158 /* protected by ARCH(5); above, near the start of uncond block */
8159 gen_bx_im(s, val);
8160 return;
8161 } else if ((insn & 0x0e000f00) == 0x0c000100) {
8162 if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
8163 /* iWMMXt register transfer. */
8164 if (extract32(s->c15_cpar, 1, 1)) {
8165 if (!disas_iwmmxt_insn(s, insn)) {
8166 return;
8167 }
8168 }
8169 }
8170 } else if ((insn & 0x0fe00000) == 0x0c400000) {
8171 /* Coprocessor double register transfer. */
8172 ARCH(5TE);
8173 } else if ((insn & 0x0f000010) == 0x0e000010) {
8174 /* Additional coprocessor register transfer. */
8175 } else if ((insn & 0x0ff10020) == 0x01000000) {
8176 uint32_t mask;
8177 uint32_t val;
8178 /* cps (privileged) */
8179 if (IS_USER(s))
8180 return;
8181 mask = val = 0;
8182 if (insn & (1 << 19)) {
8183 if (insn & (1 << 8))
8184 mask |= CPSR_A;
8185 if (insn & (1 << 7))
8186 mask |= CPSR_I;
8187 if (insn & (1 << 6))
8188 mask |= CPSR_F;
8189 if (insn & (1 << 18))
8190 val |= mask;
8191 }
8192 if (insn & (1 << 17)) {
8193 mask |= CPSR_M;
8194 val |= (insn & 0x1f);
8195 }
8196 if (mask) {
8197 gen_set_psr_im(s, mask, 0, val);
8198 }
8199 return;
8200 }
8201 goto illegal_op;
8202 }
8203 if (cond != 0xe) {
8204 /* if not always execute, we generate a conditional jump to
8205 next instruction */
8206 s->condlabel = gen_new_label();
8207 arm_gen_test_cc(cond ^ 1, s->condlabel);
8208 s->condjmp = 1;
8209 }
8210 if ((insn & 0x0f900000) == 0x03000000) {
8211 if ((insn & (1 << 21)) == 0) {
8212 ARCH(6T2);
8213 rd = (insn >> 12) & 0xf;
8214 val = ((insn >> 4) & 0xf000) | (insn & 0xfff);
8215 if ((insn & (1 << 22)) == 0) {
8216 /* MOVW */
8217 tmp = tcg_temp_new_i32();
8218 tcg_gen_movi_i32(tmp, val);
8219 } else {
8220 /* MOVT */
8221 tmp = load_reg(s, rd);
8222 tcg_gen_ext16u_i32(tmp, tmp);
8223 tcg_gen_ori_i32(tmp, tmp, val << 16);
8224 }
8225 store_reg(s, rd, tmp);
8226 } else {
8227 if (((insn >> 12) & 0xf) != 0xf)
8228 goto illegal_op;
8229 if (((insn >> 16) & 0xf) == 0) {
8230 gen_nop_hint(s, insn & 0xff);
8231 } else {
8232 /* CPSR = immediate */
8233 val = insn & 0xff;
8234 shift = ((insn >> 8) & 0xf) * 2;
8235 if (shift)
8236 val = (val >> shift) | (val << (32 - shift));
8237 i = ((insn & (1 << 22)) != 0);
8238 if (gen_set_psr_im(s, msr_mask(s, (insn >> 16) & 0xf, i),
8239 i, val)) {
8240 goto illegal_op;
8241 }
8242 }
8243 }
8244 } else if ((insn & 0x0f900000) == 0x01000000
8245 && (insn & 0x00000090) != 0x00000090) {
8246 /* miscellaneous instructions */
8247 op1 = (insn >> 21) & 3;
8248 sh = (insn >> 4) & 0xf;
8249 rm = insn & 0xf;
8250 switch (sh) {
8251 case 0x0: /* MSR, MRS */
8252 if (insn & (1 << 9)) {
8253 /* MSR (banked) and MRS (banked) */
8254 int sysm = extract32(insn, 16, 4) |
8255 (extract32(insn, 8, 1) << 4);
8256 int r = extract32(insn, 22, 1);
8257
8258 if (op1 & 1) {
8259 /* MSR (banked) */
8260 gen_msr_banked(s, r, sysm, rm);
8261 } else {
8262 /* MRS (banked) */
8263 int rd = extract32(insn, 12, 4);
8264
8265 gen_mrs_banked(s, r, sysm, rd);
8266 }
8267 break;
8268 }
8269
8270 /* MSR, MRS (for PSRs) */
8271 if (op1 & 1) {
8272 /* PSR = reg */
8273 tmp = load_reg(s, rm);
8274 i = ((op1 & 2) != 0);
8275 if (gen_set_psr(s, msr_mask(s, (insn >> 16) & 0xf, i), i, tmp))
8276 goto illegal_op;
8277 } else {
8278 /* reg = PSR */
8279 rd = (insn >> 12) & 0xf;
8280 if (op1 & 2) {
8281 if (IS_USER(s))
8282 goto illegal_op;
8283 tmp = load_cpu_field(spsr);
8284 } else {
8285 tmp = tcg_temp_new_i32();
8286 gen_helper_cpsr_read(tmp, cpu_env);
8287 }
8288 store_reg(s, rd, tmp);
8289 }
8290 break;
8291 case 0x1:
8292 if (op1 == 1) {
8293 /* branch/exchange thumb (bx). */
8294 ARCH(4T);
8295 tmp = load_reg(s, rm);
8296 gen_bx(s, tmp);
8297 } else if (op1 == 3) {
8298 /* clz */
8299 ARCH(5);
8300 rd = (insn >> 12) & 0xf;
8301 tmp = load_reg(s, rm);
8302 gen_helper_clz(tmp, tmp);
8303 store_reg(s, rd, tmp);
8304 } else {
8305 goto illegal_op;
8306 }
8307 break;
8308 case 0x2:
8309 if (op1 == 1) {
8310 ARCH(5J); /* bxj */
8311 /* Trivial implementation equivalent to bx. */
8312 tmp = load_reg(s, rm);
8313 gen_bx(s, tmp);
8314 } else {
8315 goto illegal_op;
8316 }
8317 break;
8318 case 0x3:
8319 if (op1 != 1)
8320 goto illegal_op;
8321
8322 ARCH(5);
8323 /* branch link/exchange thumb (blx) */
8324 tmp = load_reg(s, rm);
8325 tmp2 = tcg_temp_new_i32();
8326 tcg_gen_movi_i32(tmp2, s->pc);
8327 store_reg(s, 14, tmp2);
8328 gen_bx(s, tmp);
8329 break;
8330 case 0x4:
8331 {
8332 /* crc32/crc32c */
8333 uint32_t c = extract32(insn, 8, 4);
8334
8335 /* Check this CPU supports ARMv8 CRC instructions.
8336 * op1 == 3 is UNPREDICTABLE but handle as UNDEFINED.
8337 * Bits 8, 10 and 11 should be zero.
8338 */
8339 if (!arm_dc_feature(s, ARM_FEATURE_CRC) || op1 == 0x3 ||
8340 (c & 0xd) != 0) {
8341 goto illegal_op;
8342 }
8343
8344 rn = extract32(insn, 16, 4);
8345 rd = extract32(insn, 12, 4);
8346
8347 tmp = load_reg(s, rn);
8348 tmp2 = load_reg(s, rm);
8349 if (op1 == 0) {
8350 tcg_gen_andi_i32(tmp2, tmp2, 0xff);
8351 } else if (op1 == 1) {
8352 tcg_gen_andi_i32(tmp2, tmp2, 0xffff);
8353 }
8354 tmp3 = tcg_const_i32(1 << op1);
8355 if (c & 0x2) {
8356 gen_helper_crc32c(tmp, tmp, tmp2, tmp3);
8357 } else {
8358 gen_helper_crc32(tmp, tmp, tmp2, tmp3);
8359 }
8360 tcg_temp_free_i32(tmp2);
8361 tcg_temp_free_i32(tmp3);
8362 store_reg(s, rd, tmp);
8363 break;
8364 }
8365 case 0x5: /* saturating add/subtract */
8366 ARCH(5TE);
8367 rd = (insn >> 12) & 0xf;
8368 rn = (insn >> 16) & 0xf;
8369 tmp = load_reg(s, rm);
8370 tmp2 = load_reg(s, rn);
8371 if (op1 & 2)
8372 gen_helper_double_saturate(tmp2, cpu_env, tmp2);
8373 if (op1 & 1)
8374 gen_helper_sub_saturate(tmp, cpu_env, tmp, tmp2);
8375 else
8376 gen_helper_add_saturate(tmp, cpu_env, tmp, tmp2);
8377 tcg_temp_free_i32(tmp2);
8378 store_reg(s, rd, tmp);
8379 break;
8380 case 7:
8381 {
8382 int imm16 = extract32(insn, 0, 4) | (extract32(insn, 8, 12) << 4);
8383 switch (op1) {
8384 case 1:
8385 /* bkpt */
8386 ARCH(5);
8387 gen_exception_insn(s, 4, EXCP_BKPT,
8388 syn_aa32_bkpt(imm16, false),
8389 default_exception_el(s));
8390 break;
8391 case 2:
8392 /* Hypervisor call (v7) */
8393 ARCH(7);
8394 if (IS_USER(s)) {
8395 goto illegal_op;
8396 }
8397 gen_hvc(s, imm16);
8398 break;
8399 case 3:
8400 /* Secure monitor call (v6+) */
8401 ARCH(6K);
8402 if (IS_USER(s)) {
8403 goto illegal_op;
8404 }
8405 gen_smc(s);
8406 break;
8407 default:
8408 goto illegal_op;
8409 }
8410 break;
8411 }
8412 case 0x8: /* signed multiply */
8413 case 0xa:
8414 case 0xc:
8415 case 0xe:
8416 ARCH(5TE);
8417 rs = (insn >> 8) & 0xf;
8418 rn = (insn >> 12) & 0xf;
8419 rd = (insn >> 16) & 0xf;
8420 if (op1 == 1) {
8421 /* (32 * 16) >> 16 */
8422 tmp = load_reg(s, rm);
8423 tmp2 = load_reg(s, rs);
8424 if (sh & 4)
8425 tcg_gen_sari_i32(tmp2, tmp2, 16);
8426 else
8427 gen_sxth(tmp2);
8428 tmp64 = gen_muls_i64_i32(tmp, tmp2);
8429 tcg_gen_shri_i64(tmp64, tmp64, 16);
8430 tmp = tcg_temp_new_i32();
8431 tcg_gen_extrl_i64_i32(tmp, tmp64);
8432 tcg_temp_free_i64(tmp64);
8433 if ((sh & 2) == 0) {
8434 tmp2 = load_reg(s, rn);
8435 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8436 tcg_temp_free_i32(tmp2);
8437 }
8438 store_reg(s, rd, tmp);
8439 } else {
8440 /* 16 * 16 */
8441 tmp = load_reg(s, rm);
8442 tmp2 = load_reg(s, rs);
8443 gen_mulxy(tmp, tmp2, sh & 2, sh & 4);
8444 tcg_temp_free_i32(tmp2);
8445 if (op1 == 2) {
8446 tmp64 = tcg_temp_new_i64();
8447 tcg_gen_ext_i32_i64(tmp64, tmp);
8448 tcg_temp_free_i32(tmp);
8449 gen_addq(s, tmp64, rn, rd);
8450 gen_storeq_reg(s, rn, rd, tmp64);
8451 tcg_temp_free_i64(tmp64);
8452 } else {
8453 if (op1 == 0) {
8454 tmp2 = load_reg(s, rn);
8455 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8456 tcg_temp_free_i32(tmp2);
8457 }
8458 store_reg(s, rd, tmp);
8459 }
8460 }
8461 break;
8462 default:
8463 goto illegal_op;
8464 }
8465 } else if (((insn & 0x0e000000) == 0 &&
8466 (insn & 0x00000090) != 0x90) ||
8467 ((insn & 0x0e000000) == (1 << 25))) {
8468 int set_cc, logic_cc, shiftop;
8469
8470 op1 = (insn >> 21) & 0xf;
8471 set_cc = (insn >> 20) & 1;
8472 logic_cc = table_logic_cc[op1] & set_cc;
8473
8474 /* data processing instruction */
8475 if (insn & (1 << 25)) {
8476 /* immediate operand */
8477 val = insn & 0xff;
8478 shift = ((insn >> 8) & 0xf) * 2;
8479 if (shift) {
8480 val = (val >> shift) | (val << (32 - shift));
8481 }
8482 tmp2 = tcg_temp_new_i32();
8483 tcg_gen_movi_i32(tmp2, val);
8484 if (logic_cc && shift) {
8485 gen_set_CF_bit31(tmp2);
8486 }
8487 } else {
8488 /* register */
8489 rm = (insn) & 0xf;
8490 tmp2 = load_reg(s, rm);
8491 shiftop = (insn >> 5) & 3;
8492 if (!(insn & (1 << 4))) {
8493 shift = (insn >> 7) & 0x1f;
8494 gen_arm_shift_im(tmp2, shiftop, shift, logic_cc);
8495 } else {
8496 rs = (insn >> 8) & 0xf;
8497 tmp = load_reg(s, rs);
8498 gen_arm_shift_reg(tmp2, shiftop, tmp, logic_cc);
8499 }
8500 }
8501 if (op1 != 0x0f && op1 != 0x0d) {
8502 rn = (insn >> 16) & 0xf;
8503 tmp = load_reg(s, rn);
8504 } else {
8505 TCGV_UNUSED_I32(tmp);
8506 }
8507 rd = (insn >> 12) & 0xf;
8508 switch(op1) {
8509 case 0x00:
8510 tcg_gen_and_i32(tmp, tmp, tmp2);
8511 if (logic_cc) {
8512 gen_logic_CC(tmp);
8513 }
8514 store_reg_bx(s, rd, tmp);
8515 break;
8516 case 0x01:
8517 tcg_gen_xor_i32(tmp, tmp, tmp2);
8518 if (logic_cc) {
8519 gen_logic_CC(tmp);
8520 }
8521 store_reg_bx(s, rd, tmp);
8522 break;
8523 case 0x02:
8524 if (set_cc && rd == 15) {
8525 /* SUBS r15, ... is used for exception return. */
8526 if (IS_USER(s)) {
8527 goto illegal_op;
8528 }
8529 gen_sub_CC(tmp, tmp, tmp2);
8530 gen_exception_return(s, tmp);
8531 } else {
8532 if (set_cc) {
8533 gen_sub_CC(tmp, tmp, tmp2);
8534 } else {
8535 tcg_gen_sub_i32(tmp, tmp, tmp2);
8536 }
8537 store_reg_bx(s, rd, tmp);
8538 }
8539 break;
8540 case 0x03:
8541 if (set_cc) {
8542 gen_sub_CC(tmp, tmp2, tmp);
8543 } else {
8544 tcg_gen_sub_i32(tmp, tmp2, tmp);
8545 }
8546 store_reg_bx(s, rd, tmp);
8547 break;
8548 case 0x04:
8549 if (set_cc) {
8550 gen_add_CC(tmp, tmp, tmp2);
8551 } else {
8552 tcg_gen_add_i32(tmp, tmp, tmp2);
8553 }
8554 store_reg_bx(s, rd, tmp);
8555 break;
8556 case 0x05:
8557 if (set_cc) {
8558 gen_adc_CC(tmp, tmp, tmp2);
8559 } else {
8560 gen_add_carry(tmp, tmp, tmp2);
8561 }
8562 store_reg_bx(s, rd, tmp);
8563 break;
8564 case 0x06:
8565 if (set_cc) {
8566 gen_sbc_CC(tmp, tmp, tmp2);
8567 } else {
8568 gen_sub_carry(tmp, tmp, tmp2);
8569 }
8570 store_reg_bx(s, rd, tmp);
8571 break;
8572 case 0x07:
8573 if (set_cc) {
8574 gen_sbc_CC(tmp, tmp2, tmp);
8575 } else {
8576 gen_sub_carry(tmp, tmp2, tmp);
8577 }
8578 store_reg_bx(s, rd, tmp);
8579 break;
8580 case 0x08:
8581 if (set_cc) {
8582 tcg_gen_and_i32(tmp, tmp, tmp2);
8583 gen_logic_CC(tmp);
8584 }
8585 tcg_temp_free_i32(tmp);
8586 break;
8587 case 0x09:
8588 if (set_cc) {
8589 tcg_gen_xor_i32(tmp, tmp, tmp2);
8590 gen_logic_CC(tmp);
8591 }
8592 tcg_temp_free_i32(tmp);
8593 break;
8594 case 0x0a:
8595 if (set_cc) {
8596 gen_sub_CC(tmp, tmp, tmp2);
8597 }
8598 tcg_temp_free_i32(tmp);
8599 break;
8600 case 0x0b:
8601 if (set_cc) {
8602 gen_add_CC(tmp, tmp, tmp2);
8603 }
8604 tcg_temp_free_i32(tmp);
8605 break;
8606 case 0x0c:
8607 tcg_gen_or_i32(tmp, tmp, tmp2);
8608 if (logic_cc) {
8609 gen_logic_CC(tmp);
8610 }
8611 store_reg_bx(s, rd, tmp);
8612 break;
8613 case 0x0d:
8614 if (logic_cc && rd == 15) {
8615 /* MOVS r15, ... is used for exception return. */
8616 if (IS_USER(s)) {
8617 goto illegal_op;
8618 }
8619 gen_exception_return(s, tmp2);
8620 } else {
8621 if (logic_cc) {
8622 gen_logic_CC(tmp2);
8623 }
8624 store_reg_bx(s, rd, tmp2);
8625 }
8626 break;
8627 case 0x0e:
8628 tcg_gen_andc_i32(tmp, tmp, tmp2);
8629 if (logic_cc) {
8630 gen_logic_CC(tmp);
8631 }
8632 store_reg_bx(s, rd, tmp);
8633 break;
8634 default:
8635 case 0x0f:
8636 tcg_gen_not_i32(tmp2, tmp2);
8637 if (logic_cc) {
8638 gen_logic_CC(tmp2);
8639 }
8640 store_reg_bx(s, rd, tmp2);
8641 break;
8642 }
8643 if (op1 != 0x0f && op1 != 0x0d) {
8644 tcg_temp_free_i32(tmp2);
8645 }
8646 } else {
8647 /* other instructions */
8648 op1 = (insn >> 24) & 0xf;
8649 switch(op1) {
8650 case 0x0:
8651 case 0x1:
8652 /* multiplies, extra load/stores */
8653 sh = (insn >> 5) & 3;
8654 if (sh == 0) {
8655 if (op1 == 0x0) {
8656 rd = (insn >> 16) & 0xf;
8657 rn = (insn >> 12) & 0xf;
8658 rs = (insn >> 8) & 0xf;
8659 rm = (insn) & 0xf;
8660 op1 = (insn >> 20) & 0xf;
8661 switch (op1) {
8662 case 0: case 1: case 2: case 3: case 6:
8663 /* 32 bit mul */
8664 tmp = load_reg(s, rs);
8665 tmp2 = load_reg(s, rm);
8666 tcg_gen_mul_i32(tmp, tmp, tmp2);
8667 tcg_temp_free_i32(tmp2);
8668 if (insn & (1 << 22)) {
8669 /* Subtract (mls) */
8670 ARCH(6T2);
8671 tmp2 = load_reg(s, rn);
8672 tcg_gen_sub_i32(tmp, tmp2, tmp);
8673 tcg_temp_free_i32(tmp2);
8674 } else if (insn & (1 << 21)) {
8675 /* Add */
8676 tmp2 = load_reg(s, rn);
8677 tcg_gen_add_i32(tmp, tmp, tmp2);
8678 tcg_temp_free_i32(tmp2);
8679 }
8680 if (insn & (1 << 20))
8681 gen_logic_CC(tmp);
8682 store_reg(s, rd, tmp);
8683 break;
8684 case 4:
8685 /* 64 bit mul double accumulate (UMAAL) */
8686 ARCH(6);
8687 tmp = load_reg(s, rs);
8688 tmp2 = load_reg(s, rm);
8689 tmp64 = gen_mulu_i64_i32(tmp, tmp2);
8690 gen_addq_lo(s, tmp64, rn);
8691 gen_addq_lo(s, tmp64, rd);
8692 gen_storeq_reg(s, rn, rd, tmp64);
8693 tcg_temp_free_i64(tmp64);
8694 break;
8695 case 8: case 9: case 10: case 11:
8696 case 12: case 13: case 14: case 15:
8697 /* 64 bit mul: UMULL, UMLAL, SMULL, SMLAL. */
8698 tmp = load_reg(s, rs);
8699 tmp2 = load_reg(s, rm);
8700 if (insn & (1 << 22)) {
8701 tcg_gen_muls2_i32(tmp, tmp2, tmp, tmp2);
8702 } else {
8703 tcg_gen_mulu2_i32(tmp, tmp2, tmp, tmp2);
8704 }
8705 if (insn & (1 << 21)) { /* mult accumulate */
8706 TCGv_i32 al = load_reg(s, rn);
8707 TCGv_i32 ah = load_reg(s, rd);
8708 tcg_gen_add2_i32(tmp, tmp2, tmp, tmp2, al, ah);
8709 tcg_temp_free_i32(al);
8710 tcg_temp_free_i32(ah);
8711 }
8712 if (insn & (1 << 20)) {
8713 gen_logicq_cc(tmp, tmp2);
8714 }
8715 store_reg(s, rn, tmp);
8716 store_reg(s, rd, tmp2);
8717 break;
8718 default:
8719 goto illegal_op;
8720 }
8721 } else {
8722 rn = (insn >> 16) & 0xf;
8723 rd = (insn >> 12) & 0xf;
8724 if (insn & (1 << 23)) {
8725 /* load/store exclusive */
8726 int op2 = (insn >> 8) & 3;
8727 op1 = (insn >> 21) & 0x3;
8728
8729 switch (op2) {
8730 case 0: /* lda/stl */
8731 if (op1 == 1) {
8732 goto illegal_op;
8733 }
8734 ARCH(8);
8735 break;
8736 case 1: /* reserved */
8737 goto illegal_op;
8738 case 2: /* ldaex/stlex */
8739 ARCH(8);
8740 break;
8741 case 3: /* ldrex/strex */
8742 if (op1) {
8743 ARCH(6K);
8744 } else {
8745 ARCH(6);
8746 }
8747 break;
8748 }
8749
8750 addr = tcg_temp_local_new_i32();
8751 load_reg_var(s, addr, rn);
8752
8753 /* Since the emulation does not have barriers,
8754 the acquire/release semantics need no special
8755 handling */
8756 if (op2 == 0) {
8757 if (insn & (1 << 20)) {
8758 tmp = tcg_temp_new_i32();
8759 switch (op1) {
8760 case 0: /* lda */
8761 gen_aa32_ld32u(s, tmp, addr,
8762 get_mem_index(s));
8763 break;
8764 case 2: /* ldab */
8765 gen_aa32_ld8u(s, tmp, addr,
8766 get_mem_index(s));
8767 break;
8768 case 3: /* ldah */
8769 gen_aa32_ld16u(s, tmp, addr,
8770 get_mem_index(s));
8771 break;
8772 default:
8773 abort();
8774 }
8775 store_reg(s, rd, tmp);
8776 } else {
8777 rm = insn & 0xf;
8778 tmp = load_reg(s, rm);
8779 switch (op1) {
8780 case 0: /* stl */
8781 gen_aa32_st32(s, tmp, addr,
8782 get_mem_index(s));
8783 break;
8784 case 2: /* stlb */
8785 gen_aa32_st8(s, tmp, addr,
8786 get_mem_index(s));
8787 break;
8788 case 3: /* stlh */
8789 gen_aa32_st16(s, tmp, addr,
8790 get_mem_index(s));
8791 break;
8792 default:
8793 abort();
8794 }
8795 tcg_temp_free_i32(tmp);
8796 }
8797 } else if (insn & (1 << 20)) {
8798 switch (op1) {
8799 case 0: /* ldrex */
8800 gen_load_exclusive(s, rd, 15, addr, 2);
8801 break;
8802 case 1: /* ldrexd */
8803 gen_load_exclusive(s, rd, rd + 1, addr, 3);
8804 break;
8805 case 2: /* ldrexb */
8806 gen_load_exclusive(s, rd, 15, addr, 0);
8807 break;
8808 case 3: /* ldrexh */
8809 gen_load_exclusive(s, rd, 15, addr, 1);
8810 break;
8811 default:
8812 abort();
8813 }
8814 } else {
8815 rm = insn & 0xf;
8816 switch (op1) {
8817 case 0: /* strex */
8818 gen_store_exclusive(s, rd, rm, 15, addr, 2);
8819 break;
8820 case 1: /* strexd */
8821 gen_store_exclusive(s, rd, rm, rm + 1, addr, 3);
8822 break;
8823 case 2: /* strexb */
8824 gen_store_exclusive(s, rd, rm, 15, addr, 0);
8825 break;
8826 case 3: /* strexh */
8827 gen_store_exclusive(s, rd, rm, 15, addr, 1);
8828 break;
8829 default:
8830 abort();
8831 }
8832 }
8833 tcg_temp_free_i32(addr);
8834 } else {
8835 /* SWP instruction */
8836 rm = (insn) & 0xf;
8837
8838 /* ??? This is not really atomic. However we know
8839 we never have multiple CPUs running in parallel,
8840 so it is good enough. */
8841 addr = load_reg(s, rn);
8842 tmp = load_reg(s, rm);
8843 tmp2 = tcg_temp_new_i32();
8844 if (insn & (1 << 22)) {
8845 gen_aa32_ld8u(s, tmp2, addr, get_mem_index(s));
8846 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
8847 } else {
8848 gen_aa32_ld32u(s, tmp2, addr, get_mem_index(s));
8849 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
8850 }
8851 tcg_temp_free_i32(tmp);
8852 tcg_temp_free_i32(addr);
8853 store_reg(s, rd, tmp2);
8854 }
8855 }
8856 } else {
8857 int address_offset;
8858 bool load = insn & (1 << 20);
8859 bool doubleword = false;
8860 /* Misc load/store */
8861 rn = (insn >> 16) & 0xf;
8862 rd = (insn >> 12) & 0xf;
8863
8864 if (!load && (sh & 2)) {
8865 /* doubleword */
8866 ARCH(5TE);
8867 if (rd & 1) {
8868 /* UNPREDICTABLE; we choose to UNDEF */
8869 goto illegal_op;
8870 }
8871 load = (sh & 1) == 0;
8872 doubleword = true;
8873 }
8874
8875 addr = load_reg(s, rn);
8876 if (insn & (1 << 24))
8877 gen_add_datah_offset(s, insn, 0, addr);
8878 address_offset = 0;
8879
8880 if (doubleword) {
8881 if (!load) {
8882 /* store */
8883 tmp = load_reg(s, rd);
8884 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
8885 tcg_temp_free_i32(tmp);
8886 tcg_gen_addi_i32(addr, addr, 4);
8887 tmp = load_reg(s, rd + 1);
8888 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
8889 tcg_temp_free_i32(tmp);
8890 } else {
8891 /* load */
8892 tmp = tcg_temp_new_i32();
8893 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
8894 store_reg(s, rd, tmp);
8895 tcg_gen_addi_i32(addr, addr, 4);
8896 tmp = tcg_temp_new_i32();
8897 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
8898 rd++;
8899 }
8900 address_offset = -4;
8901 } else if (load) {
8902 /* load */
8903 tmp = tcg_temp_new_i32();
8904 switch (sh) {
8905 case 1:
8906 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
8907 break;
8908 case 2:
8909 gen_aa32_ld8s(s, tmp, addr, get_mem_index(s));
8910 break;
8911 default:
8912 case 3:
8913 gen_aa32_ld16s(s, tmp, addr, get_mem_index(s));
8914 break;
8915 }
8916 } else {
8917 /* store */
8918 tmp = load_reg(s, rd);
8919 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
8920 tcg_temp_free_i32(tmp);
8921 }
8922 /* Perform base writeback before the loaded value to
8923 ensure correct behavior with overlapping index registers.
8924 ldrd with base writeback is undefined if the
8925 destination and index registers overlap. */
8926 if (!(insn & (1 << 24))) {
8927 gen_add_datah_offset(s, insn, address_offset, addr);
8928 store_reg(s, rn, addr);
8929 } else if (insn & (1 << 21)) {
8930 if (address_offset)
8931 tcg_gen_addi_i32(addr, addr, address_offset);
8932 store_reg(s, rn, addr);
8933 } else {
8934 tcg_temp_free_i32(addr);
8935 }
8936 if (load) {
8937 /* Complete the load. */
8938 store_reg(s, rd, tmp);
8939 }
8940 }
8941 break;
8942 case 0x4:
8943 case 0x5:
8944 goto do_ldst;
8945 case 0x6:
8946 case 0x7:
8947 if (insn & (1 << 4)) {
8948 ARCH(6);
8949 /* Armv6 Media instructions. */
8950 rm = insn & 0xf;
8951 rn = (insn >> 16) & 0xf;
8952 rd = (insn >> 12) & 0xf;
8953 rs = (insn >> 8) & 0xf;
8954 switch ((insn >> 23) & 3) {
8955 case 0: /* Parallel add/subtract. */
8956 op1 = (insn >> 20) & 7;
8957 tmp = load_reg(s, rn);
8958 tmp2 = load_reg(s, rm);
8959 sh = (insn >> 5) & 7;
8960 if ((op1 & 3) == 0 || sh == 5 || sh == 6)
8961 goto illegal_op;
8962 gen_arm_parallel_addsub(op1, sh, tmp, tmp2);
8963 tcg_temp_free_i32(tmp2);
8964 store_reg(s, rd, tmp);
8965 break;
8966 case 1:
8967 if ((insn & 0x00700020) == 0) {
8968 /* Halfword pack. */
8969 tmp = load_reg(s, rn);
8970 tmp2 = load_reg(s, rm);
8971 shift = (insn >> 7) & 0x1f;
8972 if (insn & (1 << 6)) {
8973 /* pkhtb */
8974 if (shift == 0)
8975 shift = 31;
8976 tcg_gen_sari_i32(tmp2, tmp2, shift);
8977 tcg_gen_andi_i32(tmp, tmp, 0xffff0000);
8978 tcg_gen_ext16u_i32(tmp2, tmp2);
8979 } else {
8980 /* pkhbt */
8981 if (shift)
8982 tcg_gen_shli_i32(tmp2, tmp2, shift);
8983 tcg_gen_ext16u_i32(tmp, tmp);
8984 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
8985 }
8986 tcg_gen_or_i32(tmp, tmp, tmp2);
8987 tcg_temp_free_i32(tmp2);
8988 store_reg(s, rd, tmp);
8989 } else if ((insn & 0x00200020) == 0x00200000) {
8990 /* [us]sat */
8991 tmp = load_reg(s, rm);
8992 shift = (insn >> 7) & 0x1f;
8993 if (insn & (1 << 6)) {
8994 if (shift == 0)
8995 shift = 31;
8996 tcg_gen_sari_i32(tmp, tmp, shift);
8997 } else {
8998 tcg_gen_shli_i32(tmp, tmp, shift);
8999 }
9000 sh = (insn >> 16) & 0x1f;
9001 tmp2 = tcg_const_i32(sh);
9002 if (insn & (1 << 22))
9003 gen_helper_usat(tmp, cpu_env, tmp, tmp2);
9004 else
9005 gen_helper_ssat(tmp, cpu_env, tmp, tmp2);
9006 tcg_temp_free_i32(tmp2);
9007 store_reg(s, rd, tmp);
9008 } else if ((insn & 0x00300fe0) == 0x00200f20) {
9009 /* [us]sat16 */
9010 tmp = load_reg(s, rm);
9011 sh = (insn >> 16) & 0x1f;
9012 tmp2 = tcg_const_i32(sh);
9013 if (insn & (1 << 22))
9014 gen_helper_usat16(tmp, cpu_env, tmp, tmp2);
9015 else
9016 gen_helper_ssat16(tmp, cpu_env, tmp, tmp2);
9017 tcg_temp_free_i32(tmp2);
9018 store_reg(s, rd, tmp);
9019 } else if ((insn & 0x00700fe0) == 0x00000fa0) {
9020 /* Select bytes. */
9021 tmp = load_reg(s, rn);
9022 tmp2 = load_reg(s, rm);
9023 tmp3 = tcg_temp_new_i32();
9024 tcg_gen_ld_i32(tmp3, cpu_env, offsetof(CPUARMState, GE));
9025 gen_helper_sel_flags(tmp, tmp3, tmp, tmp2);
9026 tcg_temp_free_i32(tmp3);
9027 tcg_temp_free_i32(tmp2);
9028 store_reg(s, rd, tmp);
9029 } else if ((insn & 0x000003e0) == 0x00000060) {
9030 tmp = load_reg(s, rm);
9031 shift = (insn >> 10) & 3;
9032 /* ??? In many cases it's not necessary to do a
9033 rotate, a shift is sufficient. */
9034 if (shift != 0)
9035 tcg_gen_rotri_i32(tmp, tmp, shift * 8);
9036 op1 = (insn >> 20) & 7;
9037 switch (op1) {
9038 case 0: gen_sxtb16(tmp); break;
9039 case 2: gen_sxtb(tmp); break;
9040 case 3: gen_sxth(tmp); break;
9041 case 4: gen_uxtb16(tmp); break;
9042 case 6: gen_uxtb(tmp); break;
9043 case 7: gen_uxth(tmp); break;
9044 default: goto illegal_op;
9045 }
9046 if (rn != 15) {
9047 tmp2 = load_reg(s, rn);
9048 if ((op1 & 3) == 0) {
9049 gen_add16(tmp, tmp2);
9050 } else {
9051 tcg_gen_add_i32(tmp, tmp, tmp2);
9052 tcg_temp_free_i32(tmp2);
9053 }
9054 }
9055 store_reg(s, rd, tmp);
9056 } else if ((insn & 0x003f0f60) == 0x003f0f20) {
9057 /* rev */
9058 tmp = load_reg(s, rm);
9059 if (insn & (1 << 22)) {
9060 if (insn & (1 << 7)) {
9061 gen_revsh(tmp);
9062 } else {
9063 ARCH(6T2);
9064 gen_helper_rbit(tmp, tmp);
9065 }
9066 } else {
9067 if (insn & (1 << 7))
9068 gen_rev16(tmp);
9069 else
9070 tcg_gen_bswap32_i32(tmp, tmp);
9071 }
9072 store_reg(s, rd, tmp);
9073 } else {
9074 goto illegal_op;
9075 }
9076 break;
9077 case 2: /* Multiplies (Type 3). */
9078 switch ((insn >> 20) & 0x7) {
9079 case 5:
9080 if (((insn >> 6) ^ (insn >> 7)) & 1) {
9081 /* op2 not 00x or 11x : UNDEF */
9082 goto illegal_op;
9083 }
9084 /* Signed multiply most significant [accumulate].
9085 (SMMUL, SMMLA, SMMLS) */
9086 tmp = load_reg(s, rm);
9087 tmp2 = load_reg(s, rs);
9088 tmp64 = gen_muls_i64_i32(tmp, tmp2);
9089
9090 if (rd != 15) {
9091 tmp = load_reg(s, rd);
9092 if (insn & (1 << 6)) {
9093 tmp64 = gen_subq_msw(tmp64, tmp);
9094 } else {
9095 tmp64 = gen_addq_msw(tmp64, tmp);
9096 }
9097 }
9098 if (insn & (1 << 5)) {
9099 tcg_gen_addi_i64(tmp64, tmp64, 0x80000000u);
9100 }
9101 tcg_gen_shri_i64(tmp64, tmp64, 32);
9102 tmp = tcg_temp_new_i32();
9103 tcg_gen_extrl_i64_i32(tmp, tmp64);
9104 tcg_temp_free_i64(tmp64);
9105 store_reg(s, rn, tmp);
9106 break;
9107 case 0:
9108 case 4:
9109 /* SMLAD, SMUAD, SMLSD, SMUSD, SMLALD, SMLSLD */
9110 if (insn & (1 << 7)) {
9111 goto illegal_op;
9112 }
9113 tmp = load_reg(s, rm);
9114 tmp2 = load_reg(s, rs);
9115 if (insn & (1 << 5))
9116 gen_swap_half(tmp2);
9117 gen_smul_dual(tmp, tmp2);
9118 if (insn & (1 << 22)) {
9119 /* smlald, smlsld */
9120 TCGv_i64 tmp64_2;
9121
9122 tmp64 = tcg_temp_new_i64();
9123 tmp64_2 = tcg_temp_new_i64();
9124 tcg_gen_ext_i32_i64(tmp64, tmp);
9125 tcg_gen_ext_i32_i64(tmp64_2, tmp2);
9126 tcg_temp_free_i32(tmp);
9127 tcg_temp_free_i32(tmp2);
9128 if (insn & (1 << 6)) {
9129 tcg_gen_sub_i64(tmp64, tmp64, tmp64_2);
9130 } else {
9131 tcg_gen_add_i64(tmp64, tmp64, tmp64_2);
9132 }
9133 tcg_temp_free_i64(tmp64_2);
9134 gen_addq(s, tmp64, rd, rn);
9135 gen_storeq_reg(s, rd, rn, tmp64);
9136 tcg_temp_free_i64(tmp64);
9137 } else {
9138 /* smuad, smusd, smlad, smlsd */
9139 if (insn & (1 << 6)) {
9140 /* This subtraction cannot overflow. */
9141 tcg_gen_sub_i32(tmp, tmp, tmp2);
9142 } else {
9143 /* This addition cannot overflow 32 bits;
9144 * however it may overflow considered as a
9145 * signed operation, in which case we must set
9146 * the Q flag.
9147 */
9148 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
9149 }
9150 tcg_temp_free_i32(tmp2);
9151 if (rd != 15)
9152 {
9153 tmp2 = load_reg(s, rd);
9154 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
9155 tcg_temp_free_i32(tmp2);
9156 }
9157 store_reg(s, rn, tmp);
9158 }
9159 break;
9160 case 1:
9161 case 3:
9162 /* SDIV, UDIV */
9163 if (!arm_dc_feature(s, ARM_FEATURE_ARM_DIV)) {
9164 goto illegal_op;
9165 }
9166 if (((insn >> 5) & 7) || (rd != 15)) {
9167 goto illegal_op;
9168 }
9169 tmp = load_reg(s, rm);
9170 tmp2 = load_reg(s, rs);
9171 if (insn & (1 << 21)) {
9172 gen_helper_udiv(tmp, tmp, tmp2);
9173 } else {
9174 gen_helper_sdiv(tmp, tmp, tmp2);
9175 }
9176 tcg_temp_free_i32(tmp2);
9177 store_reg(s, rn, tmp);
9178 break;
9179 default:
9180 goto illegal_op;
9181 }
9182 break;
9183 case 3:
9184 op1 = ((insn >> 17) & 0x38) | ((insn >> 5) & 7);
9185 switch (op1) {
9186 case 0: /* Unsigned sum of absolute differences. */
9187 ARCH(6);
9188 tmp = load_reg(s, rm);
9189 tmp2 = load_reg(s, rs);
9190 gen_helper_usad8(tmp, tmp, tmp2);
9191 tcg_temp_free_i32(tmp2);
9192 if (rd != 15) {
9193 tmp2 = load_reg(s, rd);
9194 tcg_gen_add_i32(tmp, tmp, tmp2);
9195 tcg_temp_free_i32(tmp2);
9196 }
9197 store_reg(s, rn, tmp);
9198 break;
9199 case 0x20: case 0x24: case 0x28: case 0x2c:
9200 /* Bitfield insert/clear. */
9201 ARCH(6T2);
9202 shift = (insn >> 7) & 0x1f;
9203 i = (insn >> 16) & 0x1f;
9204 if (i < shift) {
9205 /* UNPREDICTABLE; we choose to UNDEF */
9206 goto illegal_op;
9207 }
9208 i = i + 1 - shift;
9209 if (rm == 15) {
9210 tmp = tcg_temp_new_i32();
9211 tcg_gen_movi_i32(tmp, 0);
9212 } else {
9213 tmp = load_reg(s, rm);
9214 }
9215 if (i != 32) {
9216 tmp2 = load_reg(s, rd);
9217 tcg_gen_deposit_i32(tmp, tmp2, tmp, shift, i);
9218 tcg_temp_free_i32(tmp2);
9219 }
9220 store_reg(s, rd, tmp);
9221 break;
9222 case 0x12: case 0x16: case 0x1a: case 0x1e: /* sbfx */
9223 case 0x32: case 0x36: case 0x3a: case 0x3e: /* ubfx */
9224 ARCH(6T2);
9225 tmp = load_reg(s, rm);
9226 shift = (insn >> 7) & 0x1f;
9227 i = ((insn >> 16) & 0x1f) + 1;
9228 if (shift + i > 32)
9229 goto illegal_op;
9230 if (i < 32) {
9231 if (op1 & 0x20) {
9232 gen_ubfx(tmp, shift, (1u << i) - 1);
9233 } else {
9234 gen_sbfx(tmp, shift, i);
9235 }
9236 }
9237 store_reg(s, rd, tmp);
9238 break;
9239 default:
9240 goto illegal_op;
9241 }
9242 break;
9243 }
9244 break;
9245 }
9246 do_ldst:
9247 /* Check for undefined extension instructions
9248 * per the ARM Bible IE:
9249 * xxxx 0111 1111 xxxx xxxx xxxx 1111 xxxx
9250 */
9251 sh = (0xf << 20) | (0xf << 4);
9252 if (op1 == 0x7 && ((insn & sh) == sh))
9253 {
9254 goto illegal_op;
9255 }
9256 /* load/store byte/word */
9257 rn = (insn >> 16) & 0xf;
9258 rd = (insn >> 12) & 0xf;
9259 tmp2 = load_reg(s, rn);
9260 if ((insn & 0x01200000) == 0x00200000) {
9261 /* ldrt/strt */
9262 i = get_a32_user_mem_index(s);
9263 } else {
9264 i = get_mem_index(s);
9265 }
9266 if (insn & (1 << 24))
9267 gen_add_data_offset(s, insn, tmp2);
9268 if (insn & (1 << 20)) {
9269 /* load */
9270 tmp = tcg_temp_new_i32();
9271 if (insn & (1 << 22)) {
9272 gen_aa32_ld8u(s, tmp, tmp2, i);
9273 } else {
9274 gen_aa32_ld32u(s, tmp, tmp2, i);
9275 }
9276 } else {
9277 /* store */
9278 tmp = load_reg(s, rd);
9279 if (insn & (1 << 22)) {
9280 gen_aa32_st8(s, tmp, tmp2, i);
9281 } else {
9282 gen_aa32_st32(s, tmp, tmp2, i);
9283 }
9284 tcg_temp_free_i32(tmp);
9285 }
9286 if (!(insn & (1 << 24))) {
9287 gen_add_data_offset(s, insn, tmp2);
9288 store_reg(s, rn, tmp2);
9289 } else if (insn & (1 << 21)) {
9290 store_reg(s, rn, tmp2);
9291 } else {
9292 tcg_temp_free_i32(tmp2);
9293 }
9294 if (insn & (1 << 20)) {
9295 /* Complete the load. */
9296 store_reg_from_load(s, rd, tmp);
9297 }
9298 break;
9299 case 0x08:
9300 case 0x09:
9301 {
9302 int j, n, loaded_base;
9303 bool exc_return = false;
9304 bool is_load = extract32(insn, 20, 1);
9305 bool user = false;
9306 TCGv_i32 loaded_var;
9307 /* load/store multiple words */
9308 /* XXX: store correct base if write back */
9309 if (insn & (1 << 22)) {
9310 /* LDM (user), LDM (exception return) and STM (user) */
9311 if (IS_USER(s))
9312 goto illegal_op; /* only usable in supervisor mode */
9313
9314 if (is_load && extract32(insn, 15, 1)) {
9315 exc_return = true;
9316 } else {
9317 user = true;
9318 }
9319 }
9320 rn = (insn >> 16) & 0xf;
9321 addr = load_reg(s, rn);
9322
9323 /* compute total size */
9324 loaded_base = 0;
9325 TCGV_UNUSED_I32(loaded_var);
9326 n = 0;
9327 for(i=0;i<16;i++) {
9328 if (insn & (1 << i))
9329 n++;
9330 }
9331 /* XXX: test invalid n == 0 case ? */
9332 if (insn & (1 << 23)) {
9333 if (insn & (1 << 24)) {
9334 /* pre increment */
9335 tcg_gen_addi_i32(addr, addr, 4);
9336 } else {
9337 /* post increment */
9338 }
9339 } else {
9340 if (insn & (1 << 24)) {
9341 /* pre decrement */
9342 tcg_gen_addi_i32(addr, addr, -(n * 4));
9343 } else {
9344 /* post decrement */
9345 if (n != 1)
9346 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
9347 }
9348 }
9349 j = 0;
9350 for(i=0;i<16;i++) {
9351 if (insn & (1 << i)) {
9352 if (is_load) {
9353 /* load */
9354 tmp = tcg_temp_new_i32();
9355 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
9356 if (user) {
9357 tmp2 = tcg_const_i32(i);
9358 gen_helper_set_user_reg(cpu_env, tmp2, tmp);
9359 tcg_temp_free_i32(tmp2);
9360 tcg_temp_free_i32(tmp);
9361 } else if (i == rn) {
9362 loaded_var = tmp;
9363 loaded_base = 1;
9364 } else {
9365 store_reg_from_load(s, i, tmp);
9366 }
9367 } else {
9368 /* store */
9369 if (i == 15) {
9370 /* special case: r15 = PC + 8 */
9371 val = (long)s->pc + 4;
9372 tmp = tcg_temp_new_i32();
9373 tcg_gen_movi_i32(tmp, val);
9374 } else if (user) {
9375 tmp = tcg_temp_new_i32();
9376 tmp2 = tcg_const_i32(i);
9377 gen_helper_get_user_reg(tmp, cpu_env, tmp2);
9378 tcg_temp_free_i32(tmp2);
9379 } else {
9380 tmp = load_reg(s, i);
9381 }
9382 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
9383 tcg_temp_free_i32(tmp);
9384 }
9385 j++;
9386 /* no need to add after the last transfer */
9387 if (j != n)
9388 tcg_gen_addi_i32(addr, addr, 4);
9389 }
9390 }
9391 if (insn & (1 << 21)) {
9392 /* write back */
9393 if (insn & (1 << 23)) {
9394 if (insn & (1 << 24)) {
9395 /* pre increment */
9396 } else {
9397 /* post increment */
9398 tcg_gen_addi_i32(addr, addr, 4);
9399 }
9400 } else {
9401 if (insn & (1 << 24)) {
9402 /* pre decrement */
9403 if (n != 1)
9404 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
9405 } else {
9406 /* post decrement */
9407 tcg_gen_addi_i32(addr, addr, -(n * 4));
9408 }
9409 }
9410 store_reg(s, rn, addr);
9411 } else {
9412 tcg_temp_free_i32(addr);
9413 }
9414 if (loaded_base) {
9415 store_reg(s, rn, loaded_var);
9416 }
9417 if (exc_return) {
9418 /* Restore CPSR from SPSR. */
9419 tmp = load_cpu_field(spsr);
9420 gen_helper_cpsr_write_eret(cpu_env, tmp);
9421 tcg_temp_free_i32(tmp);
9422 s->is_jmp = DISAS_JUMP;
9423 }
9424 }
9425 break;
9426 case 0xa:
9427 case 0xb:
9428 {
9429 int32_t offset;
9430
9431 /* branch (and link) */
9432 val = (int32_t)s->pc;
9433 if (insn & (1 << 24)) {
9434 tmp = tcg_temp_new_i32();
9435 tcg_gen_movi_i32(tmp, val);
9436 store_reg(s, 14, tmp);
9437 }
9438 offset = sextract32(insn << 2, 0, 26);
9439 val += offset + 4;
9440 gen_jmp(s, val);
9441 }
9442 break;
9443 case 0xc:
9444 case 0xd:
9445 case 0xe:
9446 if (((insn >> 8) & 0xe) == 10) {
9447 /* VFP. */
9448 if (disas_vfp_insn(s, insn)) {
9449 goto illegal_op;
9450 }
9451 } else if (disas_coproc_insn(s, insn)) {
9452 /* Coprocessor. */
9453 goto illegal_op;
9454 }
9455 break;
9456 case 0xf:
9457 /* swi */
9458 gen_set_pc_im(s, s->pc);
9459 s->svc_imm = extract32(insn, 0, 24);
9460 s->is_jmp = DISAS_SWI;
9461 break;
9462 default:
9463 illegal_op:
9464 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(),
9465 default_exception_el(s));
9466 break;
9467 }
9468 }
9469 }
9470
9471 /* Return true if this is a Thumb-2 logical op. */
9472 static int
9473 thumb2_logic_op(int op)
9474 {
9475 return (op < 8);
9476 }
9477
9478 /* Generate code for a Thumb-2 data processing operation. If CONDS is nonzero
9479 then set condition code flags based on the result of the operation.
9480 If SHIFTER_OUT is nonzero then set the carry flag for logical operations
9481 to the high bit of T1.
9482 Returns zero if the opcode is valid. */
9483
9484 static int
9485 gen_thumb2_data_op(DisasContext *s, int op, int conds, uint32_t shifter_out,
9486 TCGv_i32 t0, TCGv_i32 t1)
9487 {
9488 int logic_cc;
9489
9490 logic_cc = 0;
9491 switch (op) {
9492 case 0: /* and */
9493 tcg_gen_and_i32(t0, t0, t1);
9494 logic_cc = conds;
9495 break;
9496 case 1: /* bic */
9497 tcg_gen_andc_i32(t0, t0, t1);
9498 logic_cc = conds;
9499 break;
9500 case 2: /* orr */
9501 tcg_gen_or_i32(t0, t0, t1);
9502 logic_cc = conds;
9503 break;
9504 case 3: /* orn */
9505 tcg_gen_orc_i32(t0, t0, t1);
9506 logic_cc = conds;
9507 break;
9508 case 4: /* eor */
9509 tcg_gen_xor_i32(t0, t0, t1);
9510 logic_cc = conds;
9511 break;
9512 case 8: /* add */
9513 if (conds)
9514 gen_add_CC(t0, t0, t1);
9515 else
9516 tcg_gen_add_i32(t0, t0, t1);
9517 break;
9518 case 10: /* adc */
9519 if (conds)
9520 gen_adc_CC(t0, t0, t1);
9521 else
9522 gen_adc(t0, t1);
9523 break;
9524 case 11: /* sbc */
9525 if (conds) {
9526 gen_sbc_CC(t0, t0, t1);
9527 } else {
9528 gen_sub_carry(t0, t0, t1);
9529 }
9530 break;
9531 case 13: /* sub */
9532 if (conds)
9533 gen_sub_CC(t0, t0, t1);
9534 else
9535 tcg_gen_sub_i32(t0, t0, t1);
9536 break;
9537 case 14: /* rsb */
9538 if (conds)
9539 gen_sub_CC(t0, t1, t0);
9540 else
9541 tcg_gen_sub_i32(t0, t1, t0);
9542 break;
9543 default: /* 5, 6, 7, 9, 12, 15. */
9544 return 1;
9545 }
9546 if (logic_cc) {
9547 gen_logic_CC(t0);
9548 if (shifter_out)
9549 gen_set_CF_bit31(t1);
9550 }
9551 return 0;
9552 }
9553
9554 /* Translate a 32-bit thumb instruction. Returns nonzero if the instruction
9555 is not legal. */
9556 static int disas_thumb2_insn(CPUARMState *env, DisasContext *s, uint16_t insn_hw1)
9557 {
9558 uint32_t insn, imm, shift, offset;
9559 uint32_t rd, rn, rm, rs;
9560 TCGv_i32 tmp;
9561 TCGv_i32 tmp2;
9562 TCGv_i32 tmp3;
9563 TCGv_i32 addr;
9564 TCGv_i64 tmp64;
9565 int op;
9566 int shiftop;
9567 int conds;
9568 int logic_cc;
9569
9570 if (!(arm_dc_feature(s, ARM_FEATURE_THUMB2)
9571 || arm_dc_feature(s, ARM_FEATURE_M))) {
9572 /* Thumb-1 cores may need to treat bl and blx as a pair of
9573 16-bit instructions to get correct prefetch abort behavior. */
9574 insn = insn_hw1;
9575 if ((insn & (1 << 12)) == 0) {
9576 ARCH(5);
9577 /* Second half of blx. */
9578 offset = ((insn & 0x7ff) << 1);
9579 tmp = load_reg(s, 14);
9580 tcg_gen_addi_i32(tmp, tmp, offset);
9581 tcg_gen_andi_i32(tmp, tmp, 0xfffffffc);
9582
9583 tmp2 = tcg_temp_new_i32();
9584 tcg_gen_movi_i32(tmp2, s->pc | 1);
9585 store_reg(s, 14, tmp2);
9586 gen_bx(s, tmp);
9587 return 0;
9588 }
9589 if (insn & (1 << 11)) {
9590 /* Second half of bl. */
9591 offset = ((insn & 0x7ff) << 1) | 1;
9592 tmp = load_reg(s, 14);
9593 tcg_gen_addi_i32(tmp, tmp, offset);
9594
9595 tmp2 = tcg_temp_new_i32();
9596 tcg_gen_movi_i32(tmp2, s->pc | 1);
9597 store_reg(s, 14, tmp2);
9598 gen_bx(s, tmp);
9599 return 0;
9600 }
9601 if ((s->pc & ~TARGET_PAGE_MASK) == 0) {
9602 /* Instruction spans a page boundary. Implement it as two
9603 16-bit instructions in case the second half causes an
9604 prefetch abort. */
9605 offset = ((int32_t)insn << 21) >> 9;
9606 tcg_gen_movi_i32(cpu_R[14], s->pc + 2 + offset);
9607 return 0;
9608 }
9609 /* Fall through to 32-bit decode. */
9610 }
9611
9612 insn = arm_lduw_code(env, s->pc, s->sctlr_b);
9613 s->pc += 2;
9614 insn |= (uint32_t)insn_hw1 << 16;
9615
9616 if ((insn & 0xf800e800) != 0xf000e800) {
9617 ARCH(6T2);
9618 }
9619
9620 rn = (insn >> 16) & 0xf;
9621 rs = (insn >> 12) & 0xf;
9622 rd = (insn >> 8) & 0xf;
9623 rm = insn & 0xf;
9624 switch ((insn >> 25) & 0xf) {
9625 case 0: case 1: case 2: case 3:
9626 /* 16-bit instructions. Should never happen. */
9627 abort();
9628 case 4:
9629 if (insn & (1 << 22)) {
9630 /* Other load/store, table branch. */
9631 if (insn & 0x01200000) {
9632 /* Load/store doubleword. */
9633 if (rn == 15) {
9634 addr = tcg_temp_new_i32();
9635 tcg_gen_movi_i32(addr, s->pc & ~3);
9636 } else {
9637 addr = load_reg(s, rn);
9638 }
9639 offset = (insn & 0xff) * 4;
9640 if ((insn & (1 << 23)) == 0)
9641 offset = -offset;
9642 if (insn & (1 << 24)) {
9643 tcg_gen_addi_i32(addr, addr, offset);
9644 offset = 0;
9645 }
9646 if (insn & (1 << 20)) {
9647 /* ldrd */
9648 tmp = tcg_temp_new_i32();
9649 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
9650 store_reg(s, rs, tmp);
9651 tcg_gen_addi_i32(addr, addr, 4);
9652 tmp = tcg_temp_new_i32();
9653 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
9654 store_reg(s, rd, tmp);
9655 } else {
9656 /* strd */
9657 tmp = load_reg(s, rs);
9658 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
9659 tcg_temp_free_i32(tmp);
9660 tcg_gen_addi_i32(addr, addr, 4);
9661 tmp = load_reg(s, rd);
9662 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
9663 tcg_temp_free_i32(tmp);
9664 }
9665 if (insn & (1 << 21)) {
9666 /* Base writeback. */
9667 if (rn == 15)
9668 goto illegal_op;
9669 tcg_gen_addi_i32(addr, addr, offset - 4);
9670 store_reg(s, rn, addr);
9671 } else {
9672 tcg_temp_free_i32(addr);
9673 }
9674 } else if ((insn & (1 << 23)) == 0) {
9675 /* Load/store exclusive word. */
9676 addr = tcg_temp_local_new_i32();
9677 load_reg_var(s, addr, rn);
9678 tcg_gen_addi_i32(addr, addr, (insn & 0xff) << 2);
9679 if (insn & (1 << 20)) {
9680 gen_load_exclusive(s, rs, 15, addr, 2);
9681 } else {
9682 gen_store_exclusive(s, rd, rs, 15, addr, 2);
9683 }
9684 tcg_temp_free_i32(addr);
9685 } else if ((insn & (7 << 5)) == 0) {
9686 /* Table Branch. */
9687 if (rn == 15) {
9688 addr = tcg_temp_new_i32();
9689 tcg_gen_movi_i32(addr, s->pc);
9690 } else {
9691 addr = load_reg(s, rn);
9692 }
9693 tmp = load_reg(s, rm);
9694 tcg_gen_add_i32(addr, addr, tmp);
9695 if (insn & (1 << 4)) {
9696 /* tbh */
9697 tcg_gen_add_i32(addr, addr, tmp);
9698 tcg_temp_free_i32(tmp);
9699 tmp = tcg_temp_new_i32();
9700 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
9701 } else { /* tbb */
9702 tcg_temp_free_i32(tmp);
9703 tmp = tcg_temp_new_i32();
9704 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
9705 }
9706 tcg_temp_free_i32(addr);
9707 tcg_gen_shli_i32(tmp, tmp, 1);
9708 tcg_gen_addi_i32(tmp, tmp, s->pc);
9709 store_reg(s, 15, tmp);
9710 } else {
9711 int op2 = (insn >> 6) & 0x3;
9712 op = (insn >> 4) & 0x3;
9713 switch (op2) {
9714 case 0:
9715 goto illegal_op;
9716 case 1:
9717 /* Load/store exclusive byte/halfword/doubleword */
9718 if (op == 2) {
9719 goto illegal_op;
9720 }
9721 ARCH(7);
9722 break;
9723 case 2:
9724 /* Load-acquire/store-release */
9725 if (op == 3) {
9726 goto illegal_op;
9727 }
9728 /* Fall through */
9729 case 3:
9730 /* Load-acquire/store-release exclusive */
9731 ARCH(8);
9732 break;
9733 }
9734 addr = tcg_temp_local_new_i32();
9735 load_reg_var(s, addr, rn);
9736 if (!(op2 & 1)) {
9737 if (insn & (1 << 20)) {
9738 tmp = tcg_temp_new_i32();
9739 switch (op) {
9740 case 0: /* ldab */
9741 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
9742 break;
9743 case 1: /* ldah */
9744 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
9745 break;
9746 case 2: /* lda */
9747 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
9748 break;
9749 default:
9750 abort();
9751 }
9752 store_reg(s, rs, tmp);
9753 } else {
9754 tmp = load_reg(s, rs);
9755 switch (op) {
9756 case 0: /* stlb */
9757 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
9758 break;
9759 case 1: /* stlh */
9760 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
9761 break;
9762 case 2: /* stl */
9763 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
9764 break;
9765 default:
9766 abort();
9767 }
9768 tcg_temp_free_i32(tmp);
9769 }
9770 } else if (insn & (1 << 20)) {
9771 gen_load_exclusive(s, rs, rd, addr, op);
9772 } else {
9773 gen_store_exclusive(s, rm, rs, rd, addr, op);
9774 }
9775 tcg_temp_free_i32(addr);
9776 }
9777 } else {
9778 /* Load/store multiple, RFE, SRS. */
9779 if (((insn >> 23) & 1) == ((insn >> 24) & 1)) {
9780 /* RFE, SRS: not available in user mode or on M profile */
9781 if (IS_USER(s) || arm_dc_feature(s, ARM_FEATURE_M)) {
9782 goto illegal_op;
9783 }
9784 if (insn & (1 << 20)) {
9785 /* rfe */
9786 addr = load_reg(s, rn);
9787 if ((insn & (1 << 24)) == 0)
9788 tcg_gen_addi_i32(addr, addr, -8);
9789 /* Load PC into tmp and CPSR into tmp2. */
9790 tmp = tcg_temp_new_i32();
9791 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
9792 tcg_gen_addi_i32(addr, addr, 4);
9793 tmp2 = tcg_temp_new_i32();
9794 gen_aa32_ld32u(s, tmp2, addr, get_mem_index(s));
9795 if (insn & (1 << 21)) {
9796 /* Base writeback. */
9797 if (insn & (1 << 24)) {
9798 tcg_gen_addi_i32(addr, addr, 4);
9799 } else {
9800 tcg_gen_addi_i32(addr, addr, -4);
9801 }
9802 store_reg(s, rn, addr);
9803 } else {
9804 tcg_temp_free_i32(addr);
9805 }
9806 gen_rfe(s, tmp, tmp2);
9807 } else {
9808 /* srs */
9809 gen_srs(s, (insn & 0x1f), (insn & (1 << 24)) ? 1 : 2,
9810 insn & (1 << 21));
9811 }
9812 } else {
9813 int i, loaded_base = 0;
9814 TCGv_i32 loaded_var;
9815 /* Load/store multiple. */
9816 addr = load_reg(s, rn);
9817 offset = 0;
9818 for (i = 0; i < 16; i++) {
9819 if (insn & (1 << i))
9820 offset += 4;
9821 }
9822 if (insn & (1 << 24)) {
9823 tcg_gen_addi_i32(addr, addr, -offset);
9824 }
9825
9826 TCGV_UNUSED_I32(loaded_var);
9827 for (i = 0; i < 16; i++) {
9828 if ((insn & (1 << i)) == 0)
9829 continue;
9830 if (insn & (1 << 20)) {
9831 /* Load. */
9832 tmp = tcg_temp_new_i32();
9833 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
9834 if (i == 15) {
9835 gen_bx(s, tmp);
9836 } else if (i == rn) {
9837 loaded_var = tmp;
9838 loaded_base = 1;
9839 } else {
9840 store_reg(s, i, tmp);
9841 }
9842 } else {
9843 /* Store. */
9844 tmp = load_reg(s, i);
9845 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
9846 tcg_temp_free_i32(tmp);
9847 }
9848 tcg_gen_addi_i32(addr, addr, 4);
9849 }
9850 if (loaded_base) {
9851 store_reg(s, rn, loaded_var);
9852 }
9853 if (insn & (1 << 21)) {
9854 /* Base register writeback. */
9855 if (insn & (1 << 24)) {
9856 tcg_gen_addi_i32(addr, addr, -offset);
9857 }
9858 /* Fault if writeback register is in register list. */
9859 if (insn & (1 << rn))
9860 goto illegal_op;
9861 store_reg(s, rn, addr);
9862 } else {
9863 tcg_temp_free_i32(addr);
9864 }
9865 }
9866 }
9867 break;
9868 case 5:
9869
9870 op = (insn >> 21) & 0xf;
9871 if (op == 6) {
9872 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9873 goto illegal_op;
9874 }
9875 /* Halfword pack. */
9876 tmp = load_reg(s, rn);
9877 tmp2 = load_reg(s, rm);
9878 shift = ((insn >> 10) & 0x1c) | ((insn >> 6) & 0x3);
9879 if (insn & (1 << 5)) {
9880 /* pkhtb */
9881 if (shift == 0)
9882 shift = 31;
9883 tcg_gen_sari_i32(tmp2, tmp2, shift);
9884 tcg_gen_andi_i32(tmp, tmp, 0xffff0000);
9885 tcg_gen_ext16u_i32(tmp2, tmp2);
9886 } else {
9887 /* pkhbt */
9888 if (shift)
9889 tcg_gen_shli_i32(tmp2, tmp2, shift);
9890 tcg_gen_ext16u_i32(tmp, tmp);
9891 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
9892 }
9893 tcg_gen_or_i32(tmp, tmp, tmp2);
9894 tcg_temp_free_i32(tmp2);
9895 store_reg(s, rd, tmp);
9896 } else {
9897 /* Data processing register constant shift. */
9898 if (rn == 15) {
9899 tmp = tcg_temp_new_i32();
9900 tcg_gen_movi_i32(tmp, 0);
9901 } else {
9902 tmp = load_reg(s, rn);
9903 }
9904 tmp2 = load_reg(s, rm);
9905
9906 shiftop = (insn >> 4) & 3;
9907 shift = ((insn >> 6) & 3) | ((insn >> 10) & 0x1c);
9908 conds = (insn & (1 << 20)) != 0;
9909 logic_cc = (conds && thumb2_logic_op(op));
9910 gen_arm_shift_im(tmp2, shiftop, shift, logic_cc);
9911 if (gen_thumb2_data_op(s, op, conds, 0, tmp, tmp2))
9912 goto illegal_op;
9913 tcg_temp_free_i32(tmp2);
9914 if (rd != 15) {
9915 store_reg(s, rd, tmp);
9916 } else {
9917 tcg_temp_free_i32(tmp);
9918 }
9919 }
9920 break;
9921 case 13: /* Misc data processing. */
9922 op = ((insn >> 22) & 6) | ((insn >> 7) & 1);
9923 if (op < 4 && (insn & 0xf000) != 0xf000)
9924 goto illegal_op;
9925 switch (op) {
9926 case 0: /* Register controlled shift. */
9927 tmp = load_reg(s, rn);
9928 tmp2 = load_reg(s, rm);
9929 if ((insn & 0x70) != 0)
9930 goto illegal_op;
9931 op = (insn >> 21) & 3;
9932 logic_cc = (insn & (1 << 20)) != 0;
9933 gen_arm_shift_reg(tmp, op, tmp2, logic_cc);
9934 if (logic_cc)
9935 gen_logic_CC(tmp);
9936 store_reg_bx(s, rd, tmp);
9937 break;
9938 case 1: /* Sign/zero extend. */
9939 op = (insn >> 20) & 7;
9940 switch (op) {
9941 case 0: /* SXTAH, SXTH */
9942 case 1: /* UXTAH, UXTH */
9943 case 4: /* SXTAB, SXTB */
9944 case 5: /* UXTAB, UXTB */
9945 break;
9946 case 2: /* SXTAB16, SXTB16 */
9947 case 3: /* UXTAB16, UXTB16 */
9948 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9949 goto illegal_op;
9950 }
9951 break;
9952 default:
9953 goto illegal_op;
9954 }
9955 if (rn != 15) {
9956 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9957 goto illegal_op;
9958 }
9959 }
9960 tmp = load_reg(s, rm);
9961 shift = (insn >> 4) & 3;
9962 /* ??? In many cases it's not necessary to do a
9963 rotate, a shift is sufficient. */
9964 if (shift != 0)
9965 tcg_gen_rotri_i32(tmp, tmp, shift * 8);
9966 op = (insn >> 20) & 7;
9967 switch (op) {
9968 case 0: gen_sxth(tmp); break;
9969 case 1: gen_uxth(tmp); break;
9970 case 2: gen_sxtb16(tmp); break;
9971 case 3: gen_uxtb16(tmp); break;
9972 case 4: gen_sxtb(tmp); break;
9973 case 5: gen_uxtb(tmp); break;
9974 default:
9975 g_assert_not_reached();
9976 }
9977 if (rn != 15) {
9978 tmp2 = load_reg(s, rn);
9979 if ((op >> 1) == 1) {
9980 gen_add16(tmp, tmp2);
9981 } else {
9982 tcg_gen_add_i32(tmp, tmp, tmp2);
9983 tcg_temp_free_i32(tmp2);
9984 }
9985 }
9986 store_reg(s, rd, tmp);
9987 break;
9988 case 2: /* SIMD add/subtract. */
9989 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9990 goto illegal_op;
9991 }
9992 op = (insn >> 20) & 7;
9993 shift = (insn >> 4) & 7;
9994 if ((op & 3) == 3 || (shift & 3) == 3)
9995 goto illegal_op;
9996 tmp = load_reg(s, rn);
9997 tmp2 = load_reg(s, rm);
9998 gen_thumb2_parallel_addsub(op, shift, tmp, tmp2);
9999 tcg_temp_free_i32(tmp2);
10000 store_reg(s, rd, tmp);
10001 break;
10002 case 3: /* Other data processing. */
10003 op = ((insn >> 17) & 0x38) | ((insn >> 4) & 7);
10004 if (op < 4) {
10005 /* Saturating add/subtract. */
10006 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10007 goto illegal_op;
10008 }
10009 tmp = load_reg(s, rn);
10010 tmp2 = load_reg(s, rm);
10011 if (op & 1)
10012 gen_helper_double_saturate(tmp, cpu_env, tmp);
10013 if (op & 2)
10014 gen_helper_sub_saturate(tmp, cpu_env, tmp2, tmp);
10015 else
10016 gen_helper_add_saturate(tmp, cpu_env, tmp, tmp2);
10017 tcg_temp_free_i32(tmp2);
10018 } else {
10019 switch (op) {
10020 case 0x0a: /* rbit */
10021 case 0x08: /* rev */
10022 case 0x09: /* rev16 */
10023 case 0x0b: /* revsh */
10024 case 0x18: /* clz */
10025 break;
10026 case 0x10: /* sel */
10027 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10028 goto illegal_op;
10029 }
10030 break;
10031 case 0x20: /* crc32/crc32c */
10032 case 0x21:
10033 case 0x22:
10034 case 0x28:
10035 case 0x29:
10036 case 0x2a:
10037 if (!arm_dc_feature(s, ARM_FEATURE_CRC)) {
10038 goto illegal_op;
10039 }
10040 break;
10041 default:
10042 goto illegal_op;
10043 }
10044 tmp = load_reg(s, rn);
10045 switch (op) {
10046 case 0x0a: /* rbit */
10047 gen_helper_rbit(tmp, tmp);
10048 break;
10049 case 0x08: /* rev */
10050 tcg_gen_bswap32_i32(tmp, tmp);
10051 break;
10052 case 0x09: /* rev16 */
10053 gen_rev16(tmp);
10054 break;
10055 case 0x0b: /* revsh */
10056 gen_revsh(tmp);
10057 break;
10058 case 0x10: /* sel */
10059 tmp2 = load_reg(s, rm);
10060 tmp3 = tcg_temp_new_i32();
10061 tcg_gen_ld_i32(tmp3, cpu_env, offsetof(CPUARMState, GE));
10062 gen_helper_sel_flags(tmp, tmp3, tmp, tmp2);
10063 tcg_temp_free_i32(tmp3);
10064 tcg_temp_free_i32(tmp2);
10065 break;
10066 case 0x18: /* clz */
10067 gen_helper_clz(tmp, tmp);
10068 break;
10069 case 0x20:
10070 case 0x21:
10071 case 0x22:
10072 case 0x28:
10073 case 0x29:
10074 case 0x2a:
10075 {
10076 /* crc32/crc32c */
10077 uint32_t sz = op & 0x3;
10078 uint32_t c = op & 0x8;
10079
10080 tmp2 = load_reg(s, rm);
10081 if (sz == 0) {
10082 tcg_gen_andi_i32(tmp2, tmp2, 0xff);
10083 } else if (sz == 1) {
10084 tcg_gen_andi_i32(tmp2, tmp2, 0xffff);
10085 }
10086 tmp3 = tcg_const_i32(1 << sz);
10087 if (c) {
10088 gen_helper_crc32c(tmp, tmp, tmp2, tmp3);
10089 } else {
10090 gen_helper_crc32(tmp, tmp, tmp2, tmp3);
10091 }
10092 tcg_temp_free_i32(tmp2);
10093 tcg_temp_free_i32(tmp3);
10094 break;
10095 }
10096 default:
10097 g_assert_not_reached();
10098 }
10099 }
10100 store_reg(s, rd, tmp);
10101 break;
10102 case 4: case 5: /* 32-bit multiply. Sum of absolute differences. */
10103 switch ((insn >> 20) & 7) {
10104 case 0: /* 32 x 32 -> 32 */
10105 case 7: /* Unsigned sum of absolute differences. */
10106 break;
10107 case 1: /* 16 x 16 -> 32 */
10108 case 2: /* Dual multiply add. */
10109 case 3: /* 32 * 16 -> 32msb */
10110 case 4: /* Dual multiply subtract. */
10111 case 5: case 6: /* 32 * 32 -> 32msb (SMMUL, SMMLA, SMMLS) */
10112 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10113 goto illegal_op;
10114 }
10115 break;
10116 }
10117 op = (insn >> 4) & 0xf;
10118 tmp = load_reg(s, rn);
10119 tmp2 = load_reg(s, rm);
10120 switch ((insn >> 20) & 7) {
10121 case 0: /* 32 x 32 -> 32 */
10122 tcg_gen_mul_i32(tmp, tmp, tmp2);
10123 tcg_temp_free_i32(tmp2);
10124 if (rs != 15) {
10125 tmp2 = load_reg(s, rs);
10126 if (op)
10127 tcg_gen_sub_i32(tmp, tmp2, tmp);
10128 else
10129 tcg_gen_add_i32(tmp, tmp, tmp2);
10130 tcg_temp_free_i32(tmp2);
10131 }
10132 break;
10133 case 1: /* 16 x 16 -> 32 */
10134 gen_mulxy(tmp, tmp2, op & 2, op & 1);
10135 tcg_temp_free_i32(tmp2);
10136 if (rs != 15) {
10137 tmp2 = load_reg(s, rs);
10138 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
10139 tcg_temp_free_i32(tmp2);
10140 }
10141 break;
10142 case 2: /* Dual multiply add. */
10143 case 4: /* Dual multiply subtract. */
10144 if (op)
10145 gen_swap_half(tmp2);
10146 gen_smul_dual(tmp, tmp2);
10147 if (insn & (1 << 22)) {
10148 /* This subtraction cannot overflow. */
10149 tcg_gen_sub_i32(tmp, tmp, tmp2);
10150 } else {
10151 /* This addition cannot overflow 32 bits;
10152 * however it may overflow considered as a signed
10153 * operation, in which case we must set the Q flag.
10154 */
10155 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
10156 }
10157 tcg_temp_free_i32(tmp2);
10158 if (rs != 15)
10159 {
10160 tmp2 = load_reg(s, rs);
10161 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
10162 tcg_temp_free_i32(tmp2);
10163 }
10164 break;
10165 case 3: /* 32 * 16 -> 32msb */
10166 if (op)
10167 tcg_gen_sari_i32(tmp2, tmp2, 16);
10168 else
10169 gen_sxth(tmp2);
10170 tmp64 = gen_muls_i64_i32(tmp, tmp2);
10171 tcg_gen_shri_i64(tmp64, tmp64, 16);
10172 tmp = tcg_temp_new_i32();
10173 tcg_gen_extrl_i64_i32(tmp, tmp64);
10174 tcg_temp_free_i64(tmp64);
10175 if (rs != 15)
10176 {
10177 tmp2 = load_reg(s, rs);
10178 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
10179 tcg_temp_free_i32(tmp2);
10180 }
10181 break;
10182 case 5: case 6: /* 32 * 32 -> 32msb (SMMUL, SMMLA, SMMLS) */
10183 tmp64 = gen_muls_i64_i32(tmp, tmp2);
10184 if (rs != 15) {
10185 tmp = load_reg(s, rs);
10186 if (insn & (1 << 20)) {
10187 tmp64 = gen_addq_msw(tmp64, tmp);
10188 } else {
10189 tmp64 = gen_subq_msw(tmp64, tmp);
10190 }
10191 }
10192 if (insn & (1 << 4)) {
10193 tcg_gen_addi_i64(tmp64, tmp64, 0x80000000u);
10194 }
10195 tcg_gen_shri_i64(tmp64, tmp64, 32);
10196 tmp = tcg_temp_new_i32();
10197 tcg_gen_extrl_i64_i32(tmp, tmp64);
10198 tcg_temp_free_i64(tmp64);
10199 break;
10200 case 7: /* Unsigned sum of absolute differences. */
10201 gen_helper_usad8(tmp, tmp, tmp2);
10202 tcg_temp_free_i32(tmp2);
10203 if (rs != 15) {
10204 tmp2 = load_reg(s, rs);
10205 tcg_gen_add_i32(tmp, tmp, tmp2);
10206 tcg_temp_free_i32(tmp2);
10207 }
10208 break;
10209 }
10210 store_reg(s, rd, tmp);
10211 break;
10212 case 6: case 7: /* 64-bit multiply, Divide. */
10213 op = ((insn >> 4) & 0xf) | ((insn >> 16) & 0x70);
10214 tmp = load_reg(s, rn);
10215 tmp2 = load_reg(s, rm);
10216 if ((op & 0x50) == 0x10) {
10217 /* sdiv, udiv */
10218 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DIV)) {
10219 goto illegal_op;
10220 }
10221 if (op & 0x20)
10222 gen_helper_udiv(tmp, tmp, tmp2);
10223 else
10224 gen_helper_sdiv(tmp, tmp, tmp2);
10225 tcg_temp_free_i32(tmp2);
10226 store_reg(s, rd, tmp);
10227 } else if ((op & 0xe) == 0xc) {
10228 /* Dual multiply accumulate long. */
10229 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10230 tcg_temp_free_i32(tmp);
10231 tcg_temp_free_i32(tmp2);
10232 goto illegal_op;
10233 }
10234 if (op & 1)
10235 gen_swap_half(tmp2);
10236 gen_smul_dual(tmp, tmp2);
10237 if (op & 0x10) {
10238 tcg_gen_sub_i32(tmp, tmp, tmp2);
10239 } else {
10240 tcg_gen_add_i32(tmp, tmp, tmp2);
10241 }
10242 tcg_temp_free_i32(tmp2);
10243 /* BUGFIX */
10244 tmp64 = tcg_temp_new_i64();
10245 tcg_gen_ext_i32_i64(tmp64, tmp);
10246 tcg_temp_free_i32(tmp);
10247 gen_addq(s, tmp64, rs, rd);
10248 gen_storeq_reg(s, rs, rd, tmp64);
10249 tcg_temp_free_i64(tmp64);
10250 } else {
10251 if (op & 0x20) {
10252 /* Unsigned 64-bit multiply */
10253 tmp64 = gen_mulu_i64_i32(tmp, tmp2);
10254 } else {
10255 if (op & 8) {
10256 /* smlalxy */
10257 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10258 tcg_temp_free_i32(tmp2);
10259 tcg_temp_free_i32(tmp);
10260 goto illegal_op;
10261 }
10262 gen_mulxy(tmp, tmp2, op & 2, op & 1);
10263 tcg_temp_free_i32(tmp2);
10264 tmp64 = tcg_temp_new_i64();
10265 tcg_gen_ext_i32_i64(tmp64, tmp);
10266 tcg_temp_free_i32(tmp);
10267 } else {
10268 /* Signed 64-bit multiply */
10269 tmp64 = gen_muls_i64_i32(tmp, tmp2);
10270 }
10271 }
10272 if (op & 4) {
10273 /* umaal */
10274 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10275 tcg_temp_free_i64(tmp64);
10276 goto illegal_op;
10277 }
10278 gen_addq_lo(s, tmp64, rs);
10279 gen_addq_lo(s, tmp64, rd);
10280 } else if (op & 0x40) {
10281 /* 64-bit accumulate. */
10282 gen_addq(s, tmp64, rs, rd);
10283 }
10284 gen_storeq_reg(s, rs, rd, tmp64);
10285 tcg_temp_free_i64(tmp64);
10286 }
10287 break;
10288 }
10289 break;
10290 case 6: case 7: case 14: case 15:
10291 /* Coprocessor. */
10292 if (((insn >> 24) & 3) == 3) {
10293 /* Translate into the equivalent ARM encoding. */
10294 insn = (insn & 0xe2ffffff) | ((insn & (1 << 28)) >> 4) | (1 << 28);
10295 if (disas_neon_data_insn(s, insn)) {
10296 goto illegal_op;
10297 }
10298 } else if (((insn >> 8) & 0xe) == 10) {
10299 if (disas_vfp_insn(s, insn)) {
10300 goto illegal_op;
10301 }
10302 } else {
10303 if (insn & (1 << 28))
10304 goto illegal_op;
10305 if (disas_coproc_insn(s, insn)) {
10306 goto illegal_op;
10307 }
10308 }
10309 break;
10310 case 8: case 9: case 10: case 11:
10311 if (insn & (1 << 15)) {
10312 /* Branches, misc control. */
10313 if (insn & 0x5000) {
10314 /* Unconditional branch. */
10315 /* signextend(hw1[10:0]) -> offset[:12]. */
10316 offset = ((int32_t)insn << 5) >> 9 & ~(int32_t)0xfff;
10317 /* hw1[10:0] -> offset[11:1]. */
10318 offset |= (insn & 0x7ff) << 1;
10319 /* (~hw2[13, 11] ^ offset[24]) -> offset[23,22]
10320 offset[24:22] already have the same value because of the
10321 sign extension above. */
10322 offset ^= ((~insn) & (1 << 13)) << 10;
10323 offset ^= ((~insn) & (1 << 11)) << 11;
10324
10325 if (insn & (1 << 14)) {
10326 /* Branch and link. */
10327 tcg_gen_movi_i32(cpu_R[14], s->pc | 1);
10328 }
10329
10330 offset += s->pc;
10331 if (insn & (1 << 12)) {
10332 /* b/bl */
10333 gen_jmp(s, offset);
10334 } else {
10335 /* blx */
10336 offset &= ~(uint32_t)2;
10337 /* thumb2 bx, no need to check */
10338 gen_bx_im(s, offset);
10339 }
10340 } else if (((insn >> 23) & 7) == 7) {
10341 /* Misc control */
10342 if (insn & (1 << 13))
10343 goto illegal_op;
10344
10345 if (insn & (1 << 26)) {
10346 if (!(insn & (1 << 20))) {
10347 /* Hypervisor call (v7) */
10348 int imm16 = extract32(insn, 16, 4) << 12
10349 | extract32(insn, 0, 12);
10350 ARCH(7);
10351 if (IS_USER(s)) {
10352 goto illegal_op;
10353 }
10354 gen_hvc(s, imm16);
10355 } else {
10356 /* Secure monitor call (v6+) */
10357 ARCH(6K);
10358 if (IS_USER(s)) {
10359 goto illegal_op;
10360 }
10361 gen_smc(s);
10362 }
10363 } else {
10364 op = (insn >> 20) & 7;
10365 switch (op) {
10366 case 0: /* msr cpsr. */
10367 if (arm_dc_feature(s, ARM_FEATURE_M)) {
10368 tmp = load_reg(s, rn);
10369 addr = tcg_const_i32(insn & 0xff);
10370 gen_helper_v7m_msr(cpu_env, addr, tmp);
10371 tcg_temp_free_i32(addr);
10372 tcg_temp_free_i32(tmp);
10373 gen_lookup_tb(s);
10374 break;
10375 }
10376 /* fall through */
10377 case 1: /* msr spsr. */
10378 if (arm_dc_feature(s, ARM_FEATURE_M)) {
10379 goto illegal_op;
10380 }
10381
10382 if (extract32(insn, 5, 1)) {
10383 /* MSR (banked) */
10384 int sysm = extract32(insn, 8, 4) |
10385 (extract32(insn, 4, 1) << 4);
10386 int r = op & 1;
10387
10388 gen_msr_banked(s, r, sysm, rm);
10389 break;
10390 }
10391
10392 /* MSR (for PSRs) */
10393 tmp = load_reg(s, rn);
10394 if (gen_set_psr(s,
10395 msr_mask(s, (insn >> 8) & 0xf, op == 1),
10396 op == 1, tmp))
10397 goto illegal_op;
10398 break;
10399 case 2: /* cps, nop-hint. */
10400 if (((insn >> 8) & 7) == 0) {
10401 gen_nop_hint(s, insn & 0xff);
10402 }
10403 /* Implemented as NOP in user mode. */
10404 if (IS_USER(s))
10405 break;
10406 offset = 0;
10407 imm = 0;
10408 if (insn & (1 << 10)) {
10409 if (insn & (1 << 7))
10410 offset |= CPSR_A;
10411 if (insn & (1 << 6))
10412 offset |= CPSR_I;
10413 if (insn & (1 << 5))
10414 offset |= CPSR_F;
10415 if (insn & (1 << 9))
10416 imm = CPSR_A | CPSR_I | CPSR_F;
10417 }
10418 if (insn & (1 << 8)) {
10419 offset |= 0x1f;
10420 imm |= (insn & 0x1f);
10421 }
10422 if (offset) {
10423 gen_set_psr_im(s, offset, 0, imm);
10424 }
10425 break;
10426 case 3: /* Special control operations. */
10427 ARCH(7);
10428 op = (insn >> 4) & 0xf;
10429 switch (op) {
10430 case 2: /* clrex */
10431 gen_clrex(s);
10432 break;
10433 case 4: /* dsb */
10434 case 5: /* dmb */
10435 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
10436 break;
10437 case 6: /* isb */
10438 /* We need to break the TB after this insn
10439 * to execute self-modifying code correctly
10440 * and also to take any pending interrupts
10441 * immediately.
10442 */
10443 gen_lookup_tb(s);
10444 break;
10445 default:
10446 goto illegal_op;
10447 }
10448 break;
10449 case 4: /* bxj */
10450 /* Trivial implementation equivalent to bx. */
10451 tmp = load_reg(s, rn);
10452 gen_bx(s, tmp);
10453 break;
10454 case 5: /* Exception return. */
10455 if (IS_USER(s)) {
10456 goto illegal_op;
10457 }
10458 if (rn != 14 || rd != 15) {
10459 goto illegal_op;
10460 }
10461 tmp = load_reg(s, rn);
10462 tcg_gen_subi_i32(tmp, tmp, insn & 0xff);
10463 gen_exception_return(s, tmp);
10464 break;
10465 case 6: /* MRS */
10466 if (extract32(insn, 5, 1)) {
10467 /* MRS (banked) */
10468 int sysm = extract32(insn, 16, 4) |
10469 (extract32(insn, 4, 1) << 4);
10470
10471 gen_mrs_banked(s, 0, sysm, rd);
10472 break;
10473 }
10474
10475 /* mrs cpsr */
10476 tmp = tcg_temp_new_i32();
10477 if (arm_dc_feature(s, ARM_FEATURE_M)) {
10478 addr = tcg_const_i32(insn & 0xff);
10479 gen_helper_v7m_mrs(tmp, cpu_env, addr);
10480 tcg_temp_free_i32(addr);
10481 } else {
10482 gen_helper_cpsr_read(tmp, cpu_env);
10483 }
10484 store_reg(s, rd, tmp);
10485 break;
10486 case 7: /* MRS */
10487 if (extract32(insn, 5, 1)) {
10488 /* MRS (banked) */
10489 int sysm = extract32(insn, 16, 4) |
10490 (extract32(insn, 4, 1) << 4);
10491
10492 gen_mrs_banked(s, 1, sysm, rd);
10493 break;
10494 }
10495
10496 /* mrs spsr. */
10497 /* Not accessible in user mode. */
10498 if (IS_USER(s) || arm_dc_feature(s, ARM_FEATURE_M)) {
10499 goto illegal_op;
10500 }
10501 tmp = load_cpu_field(spsr);
10502 store_reg(s, rd, tmp);
10503 break;
10504 }
10505 }
10506 } else {
10507 /* Conditional branch. */
10508 op = (insn >> 22) & 0xf;
10509 /* Generate a conditional jump to next instruction. */
10510 s->condlabel = gen_new_label();
10511 arm_gen_test_cc(op ^ 1, s->condlabel);
10512 s->condjmp = 1;
10513
10514 /* offset[11:1] = insn[10:0] */
10515 offset = (insn & 0x7ff) << 1;
10516 /* offset[17:12] = insn[21:16]. */
10517 offset |= (insn & 0x003f0000) >> 4;
10518 /* offset[31:20] = insn[26]. */
10519 offset |= ((int32_t)((insn << 5) & 0x80000000)) >> 11;
10520 /* offset[18] = insn[13]. */
10521 offset |= (insn & (1 << 13)) << 5;
10522 /* offset[19] = insn[11]. */
10523 offset |= (insn & (1 << 11)) << 8;
10524
10525 /* jump to the offset */
10526 gen_jmp(s, s->pc + offset);
10527 }
10528 } else {
10529 /* Data processing immediate. */
10530 if (insn & (1 << 25)) {
10531 if (insn & (1 << 24)) {
10532 if (insn & (1 << 20))
10533 goto illegal_op;
10534 /* Bitfield/Saturate. */
10535 op = (insn >> 21) & 7;
10536 imm = insn & 0x1f;
10537 shift = ((insn >> 6) & 3) | ((insn >> 10) & 0x1c);
10538 if (rn == 15) {
10539 tmp = tcg_temp_new_i32();
10540 tcg_gen_movi_i32(tmp, 0);
10541 } else {
10542 tmp = load_reg(s, rn);
10543 }
10544 switch (op) {
10545 case 2: /* Signed bitfield extract. */
10546 imm++;
10547 if (shift + imm > 32)
10548 goto illegal_op;
10549 if (imm < 32)
10550 gen_sbfx(tmp, shift, imm);
10551 break;
10552 case 6: /* Unsigned bitfield extract. */
10553 imm++;
10554 if (shift + imm > 32)
10555 goto illegal_op;
10556 if (imm < 32)
10557 gen_ubfx(tmp, shift, (1u << imm) - 1);
10558 break;
10559 case 3: /* Bitfield insert/clear. */
10560 if (imm < shift)
10561 goto illegal_op;
10562 imm = imm + 1 - shift;
10563 if (imm != 32) {
10564 tmp2 = load_reg(s, rd);
10565 tcg_gen_deposit_i32(tmp, tmp2, tmp, shift, imm);
10566 tcg_temp_free_i32(tmp2);
10567 }
10568 break;
10569 case 7:
10570 goto illegal_op;
10571 default: /* Saturate. */
10572 if (shift) {
10573 if (op & 1)
10574 tcg_gen_sari_i32(tmp, tmp, shift);
10575 else
10576 tcg_gen_shli_i32(tmp, tmp, shift);
10577 }
10578 tmp2 = tcg_const_i32(imm);
10579 if (op & 4) {
10580 /* Unsigned. */
10581 if ((op & 1) && shift == 0) {
10582 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10583 tcg_temp_free_i32(tmp);
10584 tcg_temp_free_i32(tmp2);
10585 goto illegal_op;
10586 }
10587 gen_helper_usat16(tmp, cpu_env, tmp, tmp2);
10588 } else {
10589 gen_helper_usat(tmp, cpu_env, tmp, tmp2);
10590 }
10591 } else {
10592 /* Signed. */
10593 if ((op & 1) && shift == 0) {
10594 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10595 tcg_temp_free_i32(tmp);
10596 tcg_temp_free_i32(tmp2);
10597 goto illegal_op;
10598 }
10599 gen_helper_ssat16(tmp, cpu_env, tmp, tmp2);
10600 } else {
10601 gen_helper_ssat(tmp, cpu_env, tmp, tmp2);
10602 }
10603 }
10604 tcg_temp_free_i32(tmp2);
10605 break;
10606 }
10607 store_reg(s, rd, tmp);
10608 } else {
10609 imm = ((insn & 0x04000000) >> 15)
10610 | ((insn & 0x7000) >> 4) | (insn & 0xff);
10611 if (insn & (1 << 22)) {
10612 /* 16-bit immediate. */
10613 imm |= (insn >> 4) & 0xf000;
10614 if (insn & (1 << 23)) {
10615 /* movt */
10616 tmp = load_reg(s, rd);
10617 tcg_gen_ext16u_i32(tmp, tmp);
10618 tcg_gen_ori_i32(tmp, tmp, imm << 16);
10619 } else {
10620 /* movw */
10621 tmp = tcg_temp_new_i32();
10622 tcg_gen_movi_i32(tmp, imm);
10623 }
10624 } else {
10625 /* Add/sub 12-bit immediate. */
10626 if (rn == 15) {
10627 offset = s->pc & ~(uint32_t)3;
10628 if (insn & (1 << 23))
10629 offset -= imm;
10630 else
10631 offset += imm;
10632 tmp = tcg_temp_new_i32();
10633 tcg_gen_movi_i32(tmp, offset);
10634 } else {
10635 tmp = load_reg(s, rn);
10636 if (insn & (1 << 23))
10637 tcg_gen_subi_i32(tmp, tmp, imm);
10638 else
10639 tcg_gen_addi_i32(tmp, tmp, imm);
10640 }
10641 }
10642 store_reg(s, rd, tmp);
10643 }
10644 } else {
10645 int shifter_out = 0;
10646 /* modified 12-bit immediate. */
10647 shift = ((insn & 0x04000000) >> 23) | ((insn & 0x7000) >> 12);
10648 imm = (insn & 0xff);
10649 switch (shift) {
10650 case 0: /* XY */
10651 /* Nothing to do. */
10652 break;
10653 case 1: /* 00XY00XY */
10654 imm |= imm << 16;
10655 break;
10656 case 2: /* XY00XY00 */
10657 imm |= imm << 16;
10658 imm <<= 8;
10659 break;
10660 case 3: /* XYXYXYXY */
10661 imm |= imm << 16;
10662 imm |= imm << 8;
10663 break;
10664 default: /* Rotated constant. */
10665 shift = (shift << 1) | (imm >> 7);
10666 imm |= 0x80;
10667 imm = imm << (32 - shift);
10668 shifter_out = 1;
10669 break;
10670 }
10671 tmp2 = tcg_temp_new_i32();
10672 tcg_gen_movi_i32(tmp2, imm);
10673 rn = (insn >> 16) & 0xf;
10674 if (rn == 15) {
10675 tmp = tcg_temp_new_i32();
10676 tcg_gen_movi_i32(tmp, 0);
10677 } else {
10678 tmp = load_reg(s, rn);
10679 }
10680 op = (insn >> 21) & 0xf;
10681 if (gen_thumb2_data_op(s, op, (insn & (1 << 20)) != 0,
10682 shifter_out, tmp, tmp2))
10683 goto illegal_op;
10684 tcg_temp_free_i32(tmp2);
10685 rd = (insn >> 8) & 0xf;
10686 if (rd != 15) {
10687 store_reg(s, rd, tmp);
10688 } else {
10689 tcg_temp_free_i32(tmp);
10690 }
10691 }
10692 }
10693 break;
10694 case 12: /* Load/store single data item. */
10695 {
10696 int postinc = 0;
10697 int writeback = 0;
10698 int memidx;
10699 if ((insn & 0x01100000) == 0x01000000) {
10700 if (disas_neon_ls_insn(s, insn)) {
10701 goto illegal_op;
10702 }
10703 break;
10704 }
10705 op = ((insn >> 21) & 3) | ((insn >> 22) & 4);
10706 if (rs == 15) {
10707 if (!(insn & (1 << 20))) {
10708 goto illegal_op;
10709 }
10710 if (op != 2) {
10711 /* Byte or halfword load space with dest == r15 : memory hints.
10712 * Catch them early so we don't emit pointless addressing code.
10713 * This space is a mix of:
10714 * PLD/PLDW/PLI, which we implement as NOPs (note that unlike
10715 * the ARM encodings, PLDW space doesn't UNDEF for non-v7MP
10716 * cores)
10717 * unallocated hints, which must be treated as NOPs
10718 * UNPREDICTABLE space, which we NOP or UNDEF depending on
10719 * which is easiest for the decoding logic
10720 * Some space which must UNDEF
10721 */
10722 int op1 = (insn >> 23) & 3;
10723 int op2 = (insn >> 6) & 0x3f;
10724 if (op & 2) {
10725 goto illegal_op;
10726 }
10727 if (rn == 15) {
10728 /* UNPREDICTABLE, unallocated hint or
10729 * PLD/PLDW/PLI (literal)
10730 */
10731 return 0;
10732 }
10733 if (op1 & 1) {
10734 return 0; /* PLD/PLDW/PLI or unallocated hint */
10735 }
10736 if ((op2 == 0) || ((op2 & 0x3c) == 0x30)) {
10737 return 0; /* PLD/PLDW/PLI or unallocated hint */
10738 }
10739 /* UNDEF space, or an UNPREDICTABLE */
10740 return 1;
10741 }
10742 }
10743 memidx = get_mem_index(s);
10744 if (rn == 15) {
10745 addr = tcg_temp_new_i32();
10746 /* PC relative. */
10747 /* s->pc has already been incremented by 4. */
10748 imm = s->pc & 0xfffffffc;
10749 if (insn & (1 << 23))
10750 imm += insn & 0xfff;
10751 else
10752 imm -= insn & 0xfff;
10753 tcg_gen_movi_i32(addr, imm);
10754 } else {
10755 addr = load_reg(s, rn);
10756 if (insn & (1 << 23)) {
10757 /* Positive offset. */
10758 imm = insn & 0xfff;
10759 tcg_gen_addi_i32(addr, addr, imm);
10760 } else {
10761 imm = insn & 0xff;
10762 switch ((insn >> 8) & 0xf) {
10763 case 0x0: /* Shifted Register. */
10764 shift = (insn >> 4) & 0xf;
10765 if (shift > 3) {
10766 tcg_temp_free_i32(addr);
10767 goto illegal_op;
10768 }
10769 tmp = load_reg(s, rm);
10770 if (shift)
10771 tcg_gen_shli_i32(tmp, tmp, shift);
10772 tcg_gen_add_i32(addr, addr, tmp);
10773 tcg_temp_free_i32(tmp);
10774 break;
10775 case 0xc: /* Negative offset. */
10776 tcg_gen_addi_i32(addr, addr, -imm);
10777 break;
10778 case 0xe: /* User privilege. */
10779 tcg_gen_addi_i32(addr, addr, imm);
10780 memidx = get_a32_user_mem_index(s);
10781 break;
10782 case 0x9: /* Post-decrement. */
10783 imm = -imm;
10784 /* Fall through. */
10785 case 0xb: /* Post-increment. */
10786 postinc = 1;
10787 writeback = 1;
10788 break;
10789 case 0xd: /* Pre-decrement. */
10790 imm = -imm;
10791 /* Fall through. */
10792 case 0xf: /* Pre-increment. */
10793 tcg_gen_addi_i32(addr, addr, imm);
10794 writeback = 1;
10795 break;
10796 default:
10797 tcg_temp_free_i32(addr);
10798 goto illegal_op;
10799 }
10800 }
10801 }
10802 if (insn & (1 << 20)) {
10803 /* Load. */
10804 tmp = tcg_temp_new_i32();
10805 switch (op) {
10806 case 0:
10807 gen_aa32_ld8u(s, tmp, addr, memidx);
10808 break;
10809 case 4:
10810 gen_aa32_ld8s(s, tmp, addr, memidx);
10811 break;
10812 case 1:
10813 gen_aa32_ld16u(s, tmp, addr, memidx);
10814 break;
10815 case 5:
10816 gen_aa32_ld16s(s, tmp, addr, memidx);
10817 break;
10818 case 2:
10819 gen_aa32_ld32u(s, tmp, addr, memidx);
10820 break;
10821 default:
10822 tcg_temp_free_i32(tmp);
10823 tcg_temp_free_i32(addr);
10824 goto illegal_op;
10825 }
10826 if (rs == 15) {
10827 gen_bx(s, tmp);
10828 } else {
10829 store_reg(s, rs, tmp);
10830 }
10831 } else {
10832 /* Store. */
10833 tmp = load_reg(s, rs);
10834 switch (op) {
10835 case 0:
10836 gen_aa32_st8(s, tmp, addr, memidx);
10837 break;
10838 case 1:
10839 gen_aa32_st16(s, tmp, addr, memidx);
10840 break;
10841 case 2:
10842 gen_aa32_st32(s, tmp, addr, memidx);
10843 break;
10844 default:
10845 tcg_temp_free_i32(tmp);
10846 tcg_temp_free_i32(addr);
10847 goto illegal_op;
10848 }
10849 tcg_temp_free_i32(tmp);
10850 }
10851 if (postinc)
10852 tcg_gen_addi_i32(addr, addr, imm);
10853 if (writeback) {
10854 store_reg(s, rn, addr);
10855 } else {
10856 tcg_temp_free_i32(addr);
10857 }
10858 }
10859 break;
10860 default:
10861 goto illegal_op;
10862 }
10863 return 0;
10864 illegal_op:
10865 return 1;
10866 }
10867
10868 static void disas_thumb_insn(CPUARMState *env, DisasContext *s)
10869 {
10870 uint32_t val, insn, op, rm, rn, rd, shift, cond;
10871 int32_t offset;
10872 int i;
10873 TCGv_i32 tmp;
10874 TCGv_i32 tmp2;
10875 TCGv_i32 addr;
10876
10877 if (s->condexec_mask) {
10878 cond = s->condexec_cond;
10879 if (cond != 0x0e) { /* Skip conditional when condition is AL. */
10880 s->condlabel = gen_new_label();
10881 arm_gen_test_cc(cond ^ 1, s->condlabel);
10882 s->condjmp = 1;
10883 }
10884 }
10885
10886 insn = arm_lduw_code(env, s->pc, s->sctlr_b);
10887 s->pc += 2;
10888
10889 switch (insn >> 12) {
10890 case 0: case 1:
10891
10892 rd = insn & 7;
10893 op = (insn >> 11) & 3;
10894 if (op == 3) {
10895 /* add/subtract */
10896 rn = (insn >> 3) & 7;
10897 tmp = load_reg(s, rn);
10898 if (insn & (1 << 10)) {
10899 /* immediate */
10900 tmp2 = tcg_temp_new_i32();
10901 tcg_gen_movi_i32(tmp2, (insn >> 6) & 7);
10902 } else {
10903 /* reg */
10904 rm = (insn >> 6) & 7;
10905 tmp2 = load_reg(s, rm);
10906 }
10907 if (insn & (1 << 9)) {
10908 if (s->condexec_mask)
10909 tcg_gen_sub_i32(tmp, tmp, tmp2);
10910 else
10911 gen_sub_CC(tmp, tmp, tmp2);
10912 } else {
10913 if (s->condexec_mask)
10914 tcg_gen_add_i32(tmp, tmp, tmp2);
10915 else
10916 gen_add_CC(tmp, tmp, tmp2);
10917 }
10918 tcg_temp_free_i32(tmp2);
10919 store_reg(s, rd, tmp);
10920 } else {
10921 /* shift immediate */
10922 rm = (insn >> 3) & 7;
10923 shift = (insn >> 6) & 0x1f;
10924 tmp = load_reg(s, rm);
10925 gen_arm_shift_im(tmp, op, shift, s->condexec_mask == 0);
10926 if (!s->condexec_mask)
10927 gen_logic_CC(tmp);
10928 store_reg(s, rd, tmp);
10929 }
10930 break;
10931 case 2: case 3:
10932 /* arithmetic large immediate */
10933 op = (insn >> 11) & 3;
10934 rd = (insn >> 8) & 0x7;
10935 if (op == 0) { /* mov */
10936 tmp = tcg_temp_new_i32();
10937 tcg_gen_movi_i32(tmp, insn & 0xff);
10938 if (!s->condexec_mask)
10939 gen_logic_CC(tmp);
10940 store_reg(s, rd, tmp);
10941 } else {
10942 tmp = load_reg(s, rd);
10943 tmp2 = tcg_temp_new_i32();
10944 tcg_gen_movi_i32(tmp2, insn & 0xff);
10945 switch (op) {
10946 case 1: /* cmp */
10947 gen_sub_CC(tmp, tmp, tmp2);
10948 tcg_temp_free_i32(tmp);
10949 tcg_temp_free_i32(tmp2);
10950 break;
10951 case 2: /* add */
10952 if (s->condexec_mask)
10953 tcg_gen_add_i32(tmp, tmp, tmp2);
10954 else
10955 gen_add_CC(tmp, tmp, tmp2);
10956 tcg_temp_free_i32(tmp2);
10957 store_reg(s, rd, tmp);
10958 break;
10959 case 3: /* sub */
10960 if (s->condexec_mask)
10961 tcg_gen_sub_i32(tmp, tmp, tmp2);
10962 else
10963 gen_sub_CC(tmp, tmp, tmp2);
10964 tcg_temp_free_i32(tmp2);
10965 store_reg(s, rd, tmp);
10966 break;
10967 }
10968 }
10969 break;
10970 case 4:
10971 if (insn & (1 << 11)) {
10972 rd = (insn >> 8) & 7;
10973 /* load pc-relative. Bit 1 of PC is ignored. */
10974 val = s->pc + 2 + ((insn & 0xff) * 4);
10975 val &= ~(uint32_t)2;
10976 addr = tcg_temp_new_i32();
10977 tcg_gen_movi_i32(addr, val);
10978 tmp = tcg_temp_new_i32();
10979 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
10980 tcg_temp_free_i32(addr);
10981 store_reg(s, rd, tmp);
10982 break;
10983 }
10984 if (insn & (1 << 10)) {
10985 /* data processing extended or blx */
10986 rd = (insn & 7) | ((insn >> 4) & 8);
10987 rm = (insn >> 3) & 0xf;
10988 op = (insn >> 8) & 3;
10989 switch (op) {
10990 case 0: /* add */
10991 tmp = load_reg(s, rd);
10992 tmp2 = load_reg(s, rm);
10993 tcg_gen_add_i32(tmp, tmp, tmp2);
10994 tcg_temp_free_i32(tmp2);
10995 store_reg(s, rd, tmp);
10996 break;
10997 case 1: /* cmp */
10998 tmp = load_reg(s, rd);
10999 tmp2 = load_reg(s, rm);
11000 gen_sub_CC(tmp, tmp, tmp2);
11001 tcg_temp_free_i32(tmp2);
11002 tcg_temp_free_i32(tmp);
11003 break;
11004 case 2: /* mov/cpy */
11005 tmp = load_reg(s, rm);
11006 store_reg(s, rd, tmp);
11007 break;
11008 case 3:/* branch [and link] exchange thumb register */
11009 tmp = load_reg(s, rm);
11010 if (insn & (1 << 7)) {
11011 ARCH(5);
11012 val = (uint32_t)s->pc | 1;
11013 tmp2 = tcg_temp_new_i32();
11014 tcg_gen_movi_i32(tmp2, val);
11015 store_reg(s, 14, tmp2);
11016 }
11017 /* already thumb, no need to check */
11018 gen_bx(s, tmp);
11019 break;
11020 }
11021 break;
11022 }
11023
11024 /* data processing register */
11025 rd = insn & 7;
11026 rm = (insn >> 3) & 7;
11027 op = (insn >> 6) & 0xf;
11028 if (op == 2 || op == 3 || op == 4 || op == 7) {
11029 /* the shift/rotate ops want the operands backwards */
11030 val = rm;
11031 rm = rd;
11032 rd = val;
11033 val = 1;
11034 } else {
11035 val = 0;
11036 }
11037
11038 if (op == 9) { /* neg */
11039 tmp = tcg_temp_new_i32();
11040 tcg_gen_movi_i32(tmp, 0);
11041 } else if (op != 0xf) { /* mvn doesn't read its first operand */
11042 tmp = load_reg(s, rd);
11043 } else {
11044 TCGV_UNUSED_I32(tmp);
11045 }
11046
11047 tmp2 = load_reg(s, rm);
11048 switch (op) {
11049 case 0x0: /* and */
11050 tcg_gen_and_i32(tmp, tmp, tmp2);
11051 if (!s->condexec_mask)
11052 gen_logic_CC(tmp);
11053 break;
11054 case 0x1: /* eor */
11055 tcg_gen_xor_i32(tmp, tmp, tmp2);
11056 if (!s->condexec_mask)
11057 gen_logic_CC(tmp);
11058 break;
11059 case 0x2: /* lsl */
11060 if (s->condexec_mask) {
11061 gen_shl(tmp2, tmp2, tmp);
11062 } else {
11063 gen_helper_shl_cc(tmp2, cpu_env, tmp2, tmp);
11064 gen_logic_CC(tmp2);
11065 }
11066 break;
11067 case 0x3: /* lsr */
11068 if (s->condexec_mask) {
11069 gen_shr(tmp2, tmp2, tmp);
11070 } else {
11071 gen_helper_shr_cc(tmp2, cpu_env, tmp2, tmp);
11072 gen_logic_CC(tmp2);
11073 }
11074 break;
11075 case 0x4: /* asr */
11076 if (s->condexec_mask) {
11077 gen_sar(tmp2, tmp2, tmp);
11078 } else {
11079 gen_helper_sar_cc(tmp2, cpu_env, tmp2, tmp);
11080 gen_logic_CC(tmp2);
11081 }
11082 break;
11083 case 0x5: /* adc */
11084 if (s->condexec_mask) {
11085 gen_adc(tmp, tmp2);
11086 } else {
11087 gen_adc_CC(tmp, tmp, tmp2);
11088 }
11089 break;
11090 case 0x6: /* sbc */
11091 if (s->condexec_mask) {
11092 gen_sub_carry(tmp, tmp, tmp2);
11093 } else {
11094 gen_sbc_CC(tmp, tmp, tmp2);
11095 }
11096 break;
11097 case 0x7: /* ror */
11098 if (s->condexec_mask) {
11099 tcg_gen_andi_i32(tmp, tmp, 0x1f);
11100 tcg_gen_rotr_i32(tmp2, tmp2, tmp);
11101 } else {
11102 gen_helper_ror_cc(tmp2, cpu_env, tmp2, tmp);
11103 gen_logic_CC(tmp2);
11104 }
11105 break;
11106 case 0x8: /* tst */
11107 tcg_gen_and_i32(tmp, tmp, tmp2);
11108 gen_logic_CC(tmp);
11109 rd = 16;
11110 break;
11111 case 0x9: /* neg */
11112 if (s->condexec_mask)
11113 tcg_gen_neg_i32(tmp, tmp2);
11114 else
11115 gen_sub_CC(tmp, tmp, tmp2);
11116 break;
11117 case 0xa: /* cmp */
11118 gen_sub_CC(tmp, tmp, tmp2);
11119 rd = 16;
11120 break;
11121 case 0xb: /* cmn */
11122 gen_add_CC(tmp, tmp, tmp2);
11123 rd = 16;
11124 break;
11125 case 0xc: /* orr */
11126 tcg_gen_or_i32(tmp, tmp, tmp2);
11127 if (!s->condexec_mask)
11128 gen_logic_CC(tmp);
11129 break;
11130 case 0xd: /* mul */
11131 tcg_gen_mul_i32(tmp, tmp, tmp2);
11132 if (!s->condexec_mask)
11133 gen_logic_CC(tmp);
11134 break;
11135 case 0xe: /* bic */
11136 tcg_gen_andc_i32(tmp, tmp, tmp2);
11137 if (!s->condexec_mask)
11138 gen_logic_CC(tmp);
11139 break;
11140 case 0xf: /* mvn */
11141 tcg_gen_not_i32(tmp2, tmp2);
11142 if (!s->condexec_mask)
11143 gen_logic_CC(tmp2);
11144 val = 1;
11145 rm = rd;
11146 break;
11147 }
11148 if (rd != 16) {
11149 if (val) {
11150 store_reg(s, rm, tmp2);
11151 if (op != 0xf)
11152 tcg_temp_free_i32(tmp);
11153 } else {
11154 store_reg(s, rd, tmp);
11155 tcg_temp_free_i32(tmp2);
11156 }
11157 } else {
11158 tcg_temp_free_i32(tmp);
11159 tcg_temp_free_i32(tmp2);
11160 }
11161 break;
11162
11163 case 5:
11164 /* load/store register offset. */
11165 rd = insn & 7;
11166 rn = (insn >> 3) & 7;
11167 rm = (insn >> 6) & 7;
11168 op = (insn >> 9) & 7;
11169 addr = load_reg(s, rn);
11170 tmp = load_reg(s, rm);
11171 tcg_gen_add_i32(addr, addr, tmp);
11172 tcg_temp_free_i32(tmp);
11173
11174 if (op < 3) { /* store */
11175 tmp = load_reg(s, rd);
11176 } else {
11177 tmp = tcg_temp_new_i32();
11178 }
11179
11180 switch (op) {
11181 case 0: /* str */
11182 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
11183 break;
11184 case 1: /* strh */
11185 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
11186 break;
11187 case 2: /* strb */
11188 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
11189 break;
11190 case 3: /* ldrsb */
11191 gen_aa32_ld8s(s, tmp, addr, get_mem_index(s));
11192 break;
11193 case 4: /* ldr */
11194 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
11195 break;
11196 case 5: /* ldrh */
11197 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
11198 break;
11199 case 6: /* ldrb */
11200 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
11201 break;
11202 case 7: /* ldrsh */
11203 gen_aa32_ld16s(s, tmp, addr, get_mem_index(s));
11204 break;
11205 }
11206 if (op >= 3) { /* load */
11207 store_reg(s, rd, tmp);
11208 } else {
11209 tcg_temp_free_i32(tmp);
11210 }
11211 tcg_temp_free_i32(addr);
11212 break;
11213
11214 case 6:
11215 /* load/store word immediate offset */
11216 rd = insn & 7;
11217 rn = (insn >> 3) & 7;
11218 addr = load_reg(s, rn);
11219 val = (insn >> 4) & 0x7c;
11220 tcg_gen_addi_i32(addr, addr, val);
11221
11222 if (insn & (1 << 11)) {
11223 /* load */
11224 tmp = tcg_temp_new_i32();
11225 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
11226 store_reg(s, rd, tmp);
11227 } else {
11228 /* store */
11229 tmp = load_reg(s, rd);
11230 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
11231 tcg_temp_free_i32(tmp);
11232 }
11233 tcg_temp_free_i32(addr);
11234 break;
11235
11236 case 7:
11237 /* load/store byte immediate offset */
11238 rd = insn & 7;
11239 rn = (insn >> 3) & 7;
11240 addr = load_reg(s, rn);
11241 val = (insn >> 6) & 0x1f;
11242 tcg_gen_addi_i32(addr, addr, val);
11243
11244 if (insn & (1 << 11)) {
11245 /* load */
11246 tmp = tcg_temp_new_i32();
11247 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
11248 store_reg(s, rd, tmp);
11249 } else {
11250 /* store */
11251 tmp = load_reg(s, rd);
11252 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
11253 tcg_temp_free_i32(tmp);
11254 }
11255 tcg_temp_free_i32(addr);
11256 break;
11257
11258 case 8:
11259 /* load/store halfword immediate offset */
11260 rd = insn & 7;
11261 rn = (insn >> 3) & 7;
11262 addr = load_reg(s, rn);
11263 val = (insn >> 5) & 0x3e;
11264 tcg_gen_addi_i32(addr, addr, val);
11265
11266 if (insn & (1 << 11)) {
11267 /* load */
11268 tmp = tcg_temp_new_i32();
11269 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
11270 store_reg(s, rd, tmp);
11271 } else {
11272 /* store */
11273 tmp = load_reg(s, rd);
11274 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
11275 tcg_temp_free_i32(tmp);
11276 }
11277 tcg_temp_free_i32(addr);
11278 break;
11279
11280 case 9:
11281 /* load/store from stack */
11282 rd = (insn >> 8) & 7;
11283 addr = load_reg(s, 13);
11284 val = (insn & 0xff) * 4;
11285 tcg_gen_addi_i32(addr, addr, val);
11286
11287 if (insn & (1 << 11)) {
11288 /* load */
11289 tmp = tcg_temp_new_i32();
11290 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
11291 store_reg(s, rd, tmp);
11292 } else {
11293 /* store */
11294 tmp = load_reg(s, rd);
11295 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
11296 tcg_temp_free_i32(tmp);
11297 }
11298 tcg_temp_free_i32(addr);
11299 break;
11300
11301 case 10:
11302 /* add to high reg */
11303 rd = (insn >> 8) & 7;
11304 if (insn & (1 << 11)) {
11305 /* SP */
11306 tmp = load_reg(s, 13);
11307 } else {
11308 /* PC. bit 1 is ignored. */
11309 tmp = tcg_temp_new_i32();
11310 tcg_gen_movi_i32(tmp, (s->pc + 2) & ~(uint32_t)2);
11311 }
11312 val = (insn & 0xff) * 4;
11313 tcg_gen_addi_i32(tmp, tmp, val);
11314 store_reg(s, rd, tmp);
11315 break;
11316
11317 case 11:
11318 /* misc */
11319 op = (insn >> 8) & 0xf;
11320 switch (op) {
11321 case 0:
11322 /* adjust stack pointer */
11323 tmp = load_reg(s, 13);
11324 val = (insn & 0x7f) * 4;
11325 if (insn & (1 << 7))
11326 val = -(int32_t)val;
11327 tcg_gen_addi_i32(tmp, tmp, val);
11328 store_reg(s, 13, tmp);
11329 break;
11330
11331 case 2: /* sign/zero extend. */
11332 ARCH(6);
11333 rd = insn & 7;
11334 rm = (insn >> 3) & 7;
11335 tmp = load_reg(s, rm);
11336 switch ((insn >> 6) & 3) {
11337 case 0: gen_sxth(tmp); break;
11338 case 1: gen_sxtb(tmp); break;
11339 case 2: gen_uxth(tmp); break;
11340 case 3: gen_uxtb(tmp); break;
11341 }
11342 store_reg(s, rd, tmp);
11343 break;
11344 case 4: case 5: case 0xc: case 0xd:
11345 /* push/pop */
11346 addr = load_reg(s, 13);
11347 if (insn & (1 << 8))
11348 offset = 4;
11349 else
11350 offset = 0;
11351 for (i = 0; i < 8; i++) {
11352 if (insn & (1 << i))
11353 offset += 4;
11354 }
11355 if ((insn & (1 << 11)) == 0) {
11356 tcg_gen_addi_i32(addr, addr, -offset);
11357 }
11358 for (i = 0; i < 8; i++) {
11359 if (insn & (1 << i)) {
11360 if (insn & (1 << 11)) {
11361 /* pop */
11362 tmp = tcg_temp_new_i32();
11363 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
11364 store_reg(s, i, tmp);
11365 } else {
11366 /* push */
11367 tmp = load_reg(s, i);
11368 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
11369 tcg_temp_free_i32(tmp);
11370 }
11371 /* advance to the next address. */
11372 tcg_gen_addi_i32(addr, addr, 4);
11373 }
11374 }
11375 TCGV_UNUSED_I32(tmp);
11376 if (insn & (1 << 8)) {
11377 if (insn & (1 << 11)) {
11378 /* pop pc */
11379 tmp = tcg_temp_new_i32();
11380 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
11381 /* don't set the pc until the rest of the instruction
11382 has completed */
11383 } else {
11384 /* push lr */
11385 tmp = load_reg(s, 14);
11386 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
11387 tcg_temp_free_i32(tmp);
11388 }
11389 tcg_gen_addi_i32(addr, addr, 4);
11390 }
11391 if ((insn & (1 << 11)) == 0) {
11392 tcg_gen_addi_i32(addr, addr, -offset);
11393 }
11394 /* write back the new stack pointer */
11395 store_reg(s, 13, addr);
11396 /* set the new PC value */
11397 if ((insn & 0x0900) == 0x0900) {
11398 store_reg_from_load(s, 15, tmp);
11399 }
11400 break;
11401
11402 case 1: case 3: case 9: case 11: /* czb */
11403 rm = insn & 7;
11404 tmp = load_reg(s, rm);
11405 s->condlabel = gen_new_label();
11406 s->condjmp = 1;
11407 if (insn & (1 << 11))
11408 tcg_gen_brcondi_i32(TCG_COND_EQ, tmp, 0, s->condlabel);
11409 else
11410 tcg_gen_brcondi_i32(TCG_COND_NE, tmp, 0, s->condlabel);
11411 tcg_temp_free_i32(tmp);
11412 offset = ((insn & 0xf8) >> 2) | (insn & 0x200) >> 3;
11413 val = (uint32_t)s->pc + 2;
11414 val += offset;
11415 gen_jmp(s, val);
11416 break;
11417
11418 case 15: /* IT, nop-hint. */
11419 if ((insn & 0xf) == 0) {
11420 gen_nop_hint(s, (insn >> 4) & 0xf);
11421 break;
11422 }
11423 /* If Then. */
11424 s->condexec_cond = (insn >> 4) & 0xe;
11425 s->condexec_mask = insn & 0x1f;
11426 /* No actual code generated for this insn, just setup state. */
11427 break;
11428
11429 case 0xe: /* bkpt */
11430 {
11431 int imm8 = extract32(insn, 0, 8);
11432 ARCH(5);
11433 gen_exception_insn(s, 2, EXCP_BKPT, syn_aa32_bkpt(imm8, true),
11434 default_exception_el(s));
11435 break;
11436 }
11437
11438 case 0xa: /* rev */
11439 ARCH(6);
11440 rn = (insn >> 3) & 0x7;
11441 rd = insn & 0x7;
11442 tmp = load_reg(s, rn);
11443 switch ((insn >> 6) & 3) {
11444 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
11445 case 1: gen_rev16(tmp); break;
11446 case 3: gen_revsh(tmp); break;
11447 default: goto illegal_op;
11448 }
11449 store_reg(s, rd, tmp);
11450 break;
11451
11452 case 6:
11453 switch ((insn >> 5) & 7) {
11454 case 2:
11455 /* setend */
11456 ARCH(6);
11457 if (((insn >> 3) & 1) != !!(s->be_data == MO_BE)) {
11458 gen_helper_setend(cpu_env);
11459 s->is_jmp = DISAS_UPDATE;
11460 }
11461 break;
11462 case 3:
11463 /* cps */
11464 ARCH(6);
11465 if (IS_USER(s)) {
11466 break;
11467 }
11468 if (arm_dc_feature(s, ARM_FEATURE_M)) {
11469 tmp = tcg_const_i32((insn & (1 << 4)) != 0);
11470 /* FAULTMASK */
11471 if (insn & 1) {
11472 addr = tcg_const_i32(19);
11473 gen_helper_v7m_msr(cpu_env, addr, tmp);
11474 tcg_temp_free_i32(addr);
11475 }
11476 /* PRIMASK */
11477 if (insn & 2) {
11478 addr = tcg_const_i32(16);
11479 gen_helper_v7m_msr(cpu_env, addr, tmp);
11480 tcg_temp_free_i32(addr);
11481 }
11482 tcg_temp_free_i32(tmp);
11483 gen_lookup_tb(s);
11484 } else {
11485 if (insn & (1 << 4)) {
11486 shift = CPSR_A | CPSR_I | CPSR_F;
11487 } else {
11488 shift = 0;
11489 }
11490 gen_set_psr_im(s, ((insn & 7) << 6), 0, shift);
11491 }
11492 break;
11493 default:
11494 goto undef;
11495 }
11496 break;
11497
11498 default:
11499 goto undef;
11500 }
11501 break;
11502
11503 case 12:
11504 {
11505 /* load/store multiple */
11506 TCGv_i32 loaded_var;
11507 TCGV_UNUSED_I32(loaded_var);
11508 rn = (insn >> 8) & 0x7;
11509 addr = load_reg(s, rn);
11510 for (i = 0; i < 8; i++) {
11511 if (insn & (1 << i)) {
11512 if (insn & (1 << 11)) {
11513 /* load */
11514 tmp = tcg_temp_new_i32();
11515 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
11516 if (i == rn) {
11517 loaded_var = tmp;
11518 } else {
11519 store_reg(s, i, tmp);
11520 }
11521 } else {
11522 /* store */
11523 tmp = load_reg(s, i);
11524 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
11525 tcg_temp_free_i32(tmp);
11526 }
11527 /* advance to the next address */
11528 tcg_gen_addi_i32(addr, addr, 4);
11529 }
11530 }
11531 if ((insn & (1 << rn)) == 0) {
11532 /* base reg not in list: base register writeback */
11533 store_reg(s, rn, addr);
11534 } else {
11535 /* base reg in list: if load, complete it now */
11536 if (insn & (1 << 11)) {
11537 store_reg(s, rn, loaded_var);
11538 }
11539 tcg_temp_free_i32(addr);
11540 }
11541 break;
11542 }
11543 case 13:
11544 /* conditional branch or swi */
11545 cond = (insn >> 8) & 0xf;
11546 if (cond == 0xe)
11547 goto undef;
11548
11549 if (cond == 0xf) {
11550 /* swi */
11551 gen_set_pc_im(s, s->pc);
11552 s->svc_imm = extract32(insn, 0, 8);
11553 s->is_jmp = DISAS_SWI;
11554 break;
11555 }
11556 /* generate a conditional jump to next instruction */
11557 s->condlabel = gen_new_label();
11558 arm_gen_test_cc(cond ^ 1, s->condlabel);
11559 s->condjmp = 1;
11560
11561 /* jump to the offset */
11562 val = (uint32_t)s->pc + 2;
11563 offset = ((int32_t)insn << 24) >> 24;
11564 val += offset << 1;
11565 gen_jmp(s, val);
11566 break;
11567
11568 case 14:
11569 if (insn & (1 << 11)) {
11570 if (disas_thumb2_insn(env, s, insn))
11571 goto undef32;
11572 break;
11573 }
11574 /* unconditional branch */
11575 val = (uint32_t)s->pc;
11576 offset = ((int32_t)insn << 21) >> 21;
11577 val += (offset << 1) + 2;
11578 gen_jmp(s, val);
11579 break;
11580
11581 case 15:
11582 if (disas_thumb2_insn(env, s, insn))
11583 goto undef32;
11584 break;
11585 }
11586 return;
11587 undef32:
11588 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(),
11589 default_exception_el(s));
11590 return;
11591 illegal_op:
11592 undef:
11593 gen_exception_insn(s, 2, EXCP_UDEF, syn_uncategorized(),
11594 default_exception_el(s));
11595 }
11596
11597 static bool insn_crosses_page(CPUARMState *env, DisasContext *s)
11598 {
11599 /* Return true if the insn at dc->pc might cross a page boundary.
11600 * (False positives are OK, false negatives are not.)
11601 */
11602 uint16_t insn;
11603
11604 if ((s->pc & 3) == 0) {
11605 /* At a 4-aligned address we can't be crossing a page */
11606 return false;
11607 }
11608
11609 /* This must be a Thumb insn */
11610 insn = arm_lduw_code(env, s->pc, s->sctlr_b);
11611
11612 if ((insn >> 11) >= 0x1d) {
11613 /* Top five bits 0b11101 / 0b11110 / 0b11111 : this is the
11614 * First half of a 32-bit Thumb insn. Thumb-1 cores might
11615 * end up actually treating this as two 16-bit insns (see the
11616 * code at the start of disas_thumb2_insn()) but we don't bother
11617 * to check for that as it is unlikely, and false positives here
11618 * are harmless.
11619 */
11620 return true;
11621 }
11622 /* Definitely a 16-bit insn, can't be crossing a page. */
11623 return false;
11624 }
11625
11626 /* generate intermediate code for basic block 'tb'. */
11627 void gen_intermediate_code(CPUARMState *env, TranslationBlock *tb)
11628 {
11629 ARMCPU *cpu = arm_env_get_cpu(env);
11630 CPUState *cs = CPU(cpu);
11631 DisasContext dc1, *dc = &dc1;
11632 target_ulong pc_start;
11633 target_ulong next_page_start;
11634 int num_insns;
11635 int max_insns;
11636 bool end_of_page;
11637
11638 /* generate intermediate code */
11639
11640 /* The A64 decoder has its own top level loop, because it doesn't need
11641 * the A32/T32 complexity to do with conditional execution/IT blocks/etc.
11642 */
11643 if (ARM_TBFLAG_AARCH64_STATE(tb->flags)) {
11644 gen_intermediate_code_a64(cpu, tb);
11645 return;
11646 }
11647
11648 pc_start = tb->pc;
11649
11650 dc->tb = tb;
11651
11652 dc->is_jmp = DISAS_NEXT;
11653 dc->pc = pc_start;
11654 dc->singlestep_enabled = cs->singlestep_enabled;
11655 dc->condjmp = 0;
11656
11657 dc->aarch64 = 0;
11658 /* If we are coming from secure EL0 in a system with a 32-bit EL3, then
11659 * there is no secure EL1, so we route exceptions to EL3.
11660 */
11661 dc->secure_routed_to_el3 = arm_feature(env, ARM_FEATURE_EL3) &&
11662 !arm_el_is_aa64(env, 3);
11663 dc->thumb = ARM_TBFLAG_THUMB(tb->flags);
11664 dc->sctlr_b = ARM_TBFLAG_SCTLR_B(tb->flags);
11665 dc->be_data = ARM_TBFLAG_BE_DATA(tb->flags) ? MO_BE : MO_LE;
11666 dc->condexec_mask = (ARM_TBFLAG_CONDEXEC(tb->flags) & 0xf) << 1;
11667 dc->condexec_cond = ARM_TBFLAG_CONDEXEC(tb->flags) >> 4;
11668 dc->mmu_idx = ARM_TBFLAG_MMUIDX(tb->flags);
11669 dc->current_el = arm_mmu_idx_to_el(dc->mmu_idx);
11670 #if !defined(CONFIG_USER_ONLY)
11671 dc->user = (dc->current_el == 0);
11672 #endif
11673 dc->ns = ARM_TBFLAG_NS(tb->flags);
11674 dc->fp_excp_el = ARM_TBFLAG_FPEXC_EL(tb->flags);
11675 dc->vfp_enabled = ARM_TBFLAG_VFPEN(tb->flags);
11676 dc->vec_len = ARM_TBFLAG_VECLEN(tb->flags);
11677 dc->vec_stride = ARM_TBFLAG_VECSTRIDE(tb->flags);
11678 dc->c15_cpar = ARM_TBFLAG_XSCALE_CPAR(tb->flags);
11679 dc->cp_regs = cpu->cp_regs;
11680 dc->features = env->features;
11681
11682 /* Single step state. The code-generation logic here is:
11683 * SS_ACTIVE == 0:
11684 * generate code with no special handling for single-stepping (except
11685 * that anything that can make us go to SS_ACTIVE == 1 must end the TB;
11686 * this happens anyway because those changes are all system register or
11687 * PSTATE writes).
11688 * SS_ACTIVE == 1, PSTATE.SS == 1: (active-not-pending)
11689 * emit code for one insn
11690 * emit code to clear PSTATE.SS
11691 * emit code to generate software step exception for completed step
11692 * end TB (as usual for having generated an exception)
11693 * SS_ACTIVE == 1, PSTATE.SS == 0: (active-pending)
11694 * emit code to generate a software step exception
11695 * end the TB
11696 */
11697 dc->ss_active = ARM_TBFLAG_SS_ACTIVE(tb->flags);
11698 dc->pstate_ss = ARM_TBFLAG_PSTATE_SS(tb->flags);
11699 dc->is_ldex = false;
11700 dc->ss_same_el = false; /* Can't be true since EL_d must be AArch64 */
11701
11702 cpu_F0s = tcg_temp_new_i32();
11703 cpu_F1s = tcg_temp_new_i32();
11704 cpu_F0d = tcg_temp_new_i64();
11705 cpu_F1d = tcg_temp_new_i64();
11706 cpu_V0 = cpu_F0d;
11707 cpu_V1 = cpu_F1d;
11708 /* FIXME: cpu_M0 can probably be the same as cpu_V0. */
11709 cpu_M0 = tcg_temp_new_i64();
11710 next_page_start = (pc_start & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE;
11711 num_insns = 0;
11712 max_insns = tb->cflags & CF_COUNT_MASK;
11713 if (max_insns == 0) {
11714 max_insns = CF_COUNT_MASK;
11715 }
11716 if (max_insns > TCG_MAX_INSNS) {
11717 max_insns = TCG_MAX_INSNS;
11718 }
11719
11720 gen_tb_start(tb);
11721
11722 tcg_clear_temp_count();
11723
11724 /* A note on handling of the condexec (IT) bits:
11725 *
11726 * We want to avoid the overhead of having to write the updated condexec
11727 * bits back to the CPUARMState for every instruction in an IT block. So:
11728 * (1) if the condexec bits are not already zero then we write
11729 * zero back into the CPUARMState now. This avoids complications trying
11730 * to do it at the end of the block. (For example if we don't do this
11731 * it's hard to identify whether we can safely skip writing condexec
11732 * at the end of the TB, which we definitely want to do for the case
11733 * where a TB doesn't do anything with the IT state at all.)
11734 * (2) if we are going to leave the TB then we call gen_set_condexec()
11735 * which will write the correct value into CPUARMState if zero is wrong.
11736 * This is done both for leaving the TB at the end, and for leaving
11737 * it because of an exception we know will happen, which is done in
11738 * gen_exception_insn(). The latter is necessary because we need to
11739 * leave the TB with the PC/IT state just prior to execution of the
11740 * instruction which caused the exception.
11741 * (3) if we leave the TB unexpectedly (eg a data abort on a load)
11742 * then the CPUARMState will be wrong and we need to reset it.
11743 * This is handled in the same way as restoration of the
11744 * PC in these situations; we save the value of the condexec bits
11745 * for each PC via tcg_gen_insn_start(), and restore_state_to_opc()
11746 * then uses this to restore them after an exception.
11747 *
11748 * Note that there are no instructions which can read the condexec
11749 * bits, and none which can write non-static values to them, so
11750 * we don't need to care about whether CPUARMState is correct in the
11751 * middle of a TB.
11752 */
11753
11754 /* Reset the conditional execution bits immediately. This avoids
11755 complications trying to do it at the end of the block. */
11756 if (dc->condexec_mask || dc->condexec_cond)
11757 {
11758 TCGv_i32 tmp = tcg_temp_new_i32();
11759 tcg_gen_movi_i32(tmp, 0);
11760 store_cpu_field(tmp, condexec_bits);
11761 }
11762 do {
11763 tcg_gen_insn_start(dc->pc,
11764 (dc->condexec_cond << 4) | (dc->condexec_mask >> 1),
11765 0);
11766 num_insns++;
11767
11768 #ifdef CONFIG_USER_ONLY
11769 /* Intercept jump to the magic kernel page. */
11770 if (dc->pc >= 0xffff0000) {
11771 /* We always get here via a jump, so know we are not in a
11772 conditional execution block. */
11773 gen_exception_internal(EXCP_KERNEL_TRAP);
11774 dc->is_jmp = DISAS_EXC;
11775 break;
11776 }
11777 #else
11778 if (dc->pc >= 0xfffffff0 && arm_dc_feature(dc, ARM_FEATURE_M)) {
11779 /* We always get here via a jump, so know we are not in a
11780 conditional execution block. */
11781 gen_exception_internal(EXCP_EXCEPTION_EXIT);
11782 dc->is_jmp = DISAS_EXC;
11783 break;
11784 }
11785 #endif
11786
11787 if (unlikely(!QTAILQ_EMPTY(&cs->breakpoints))) {
11788 CPUBreakpoint *bp;
11789 QTAILQ_FOREACH(bp, &cs->breakpoints, entry) {
11790 if (bp->pc == dc->pc) {
11791 if (bp->flags & BP_CPU) {
11792 gen_set_condexec(dc);
11793 gen_set_pc_im(dc, dc->pc);
11794 gen_helper_check_breakpoints(cpu_env);
11795 /* End the TB early; it's likely not going to be executed */
11796 dc->is_jmp = DISAS_UPDATE;
11797 } else {
11798 gen_exception_internal_insn(dc, 0, EXCP_DEBUG);
11799 /* The address covered by the breakpoint must be
11800 included in [tb->pc, tb->pc + tb->size) in order
11801 to for it to be properly cleared -- thus we
11802 increment the PC here so that the logic setting
11803 tb->size below does the right thing. */
11804 /* TODO: Advance PC by correct instruction length to
11805 * avoid disassembler error messages */
11806 dc->pc += 2;
11807 goto done_generating;
11808 }
11809 break;
11810 }
11811 }
11812 }
11813
11814 if (num_insns == max_insns && (tb->cflags & CF_LAST_IO)) {
11815 gen_io_start();
11816 }
11817
11818 if (dc->ss_active && !dc->pstate_ss) {
11819 /* Singlestep state is Active-pending.
11820 * If we're in this state at the start of a TB then either
11821 * a) we just took an exception to an EL which is being debugged
11822 * and this is the first insn in the exception handler
11823 * b) debug exceptions were masked and we just unmasked them
11824 * without changing EL (eg by clearing PSTATE.D)
11825 * In either case we're going to take a swstep exception in the
11826 * "did not step an insn" case, and so the syndrome ISV and EX
11827 * bits should be zero.
11828 */
11829 assert(num_insns == 1);
11830 gen_exception(EXCP_UDEF, syn_swstep(dc->ss_same_el, 0, 0),
11831 default_exception_el(dc));
11832 goto done_generating;
11833 }
11834
11835 if (dc->thumb) {
11836 disas_thumb_insn(env, dc);
11837 if (dc->condexec_mask) {
11838 dc->condexec_cond = (dc->condexec_cond & 0xe)
11839 | ((dc->condexec_mask >> 4) & 1);
11840 dc->condexec_mask = (dc->condexec_mask << 1) & 0x1f;
11841 if (dc->condexec_mask == 0) {
11842 dc->condexec_cond = 0;
11843 }
11844 }
11845 } else {
11846 unsigned int insn = arm_ldl_code(env, dc->pc, dc->sctlr_b);
11847 dc->pc += 4;
11848 disas_arm_insn(dc, insn);
11849 }
11850
11851 if (dc->condjmp && !dc->is_jmp) {
11852 gen_set_label(dc->condlabel);
11853 dc->condjmp = 0;
11854 }
11855
11856 if (tcg_check_temp_count()) {
11857 fprintf(stderr, "TCG temporary leak before "TARGET_FMT_lx"\n",
11858 dc->pc);
11859 }
11860
11861 /* Translation stops when a conditional branch is encountered.
11862 * Otherwise the subsequent code could get translated several times.
11863 * Also stop translation when a page boundary is reached. This
11864 * ensures prefetch aborts occur at the right place. */
11865
11866 /* We want to stop the TB if the next insn starts in a new page,
11867 * or if it spans between this page and the next. This means that
11868 * if we're looking at the last halfword in the page we need to
11869 * see if it's a 16-bit Thumb insn (which will fit in this TB)
11870 * or a 32-bit Thumb insn (which won't).
11871 * This is to avoid generating a silly TB with a single 16-bit insn
11872 * in it at the end of this page (which would execute correctly
11873 * but isn't very efficient).
11874 */
11875 end_of_page = (dc->pc >= next_page_start) ||
11876 ((dc->pc >= next_page_start - 3) && insn_crosses_page(env, dc));
11877
11878 } while (!dc->is_jmp && !tcg_op_buf_full() &&
11879 !cs->singlestep_enabled &&
11880 !singlestep &&
11881 !dc->ss_active &&
11882 !end_of_page &&
11883 num_insns < max_insns);
11884
11885 if (tb->cflags & CF_LAST_IO) {
11886 if (dc->condjmp) {
11887 /* FIXME: This can theoretically happen with self-modifying
11888 code. */
11889 cpu_abort(cs, "IO on conditional branch instruction");
11890 }
11891 gen_io_end();
11892 }
11893
11894 /* At this stage dc->condjmp will only be set when the skipped
11895 instruction was a conditional branch or trap, and the PC has
11896 already been written. */
11897 if (unlikely(cs->singlestep_enabled || dc->ss_active)) {
11898 /* Unconditional and "condition passed" instruction codepath. */
11899 gen_set_condexec(dc);
11900 switch (dc->is_jmp) {
11901 case DISAS_SWI:
11902 gen_ss_advance(dc);
11903 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
11904 default_exception_el(dc));
11905 break;
11906 case DISAS_HVC:
11907 gen_ss_advance(dc);
11908 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
11909 break;
11910 case DISAS_SMC:
11911 gen_ss_advance(dc);
11912 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
11913 break;
11914 case DISAS_NEXT:
11915 case DISAS_UPDATE:
11916 gen_set_pc_im(dc, dc->pc);
11917 /* fall through */
11918 default:
11919 if (dc->ss_active) {
11920 gen_step_complete_exception(dc);
11921 } else {
11922 /* FIXME: Single stepping a WFI insn will not halt
11923 the CPU. */
11924 gen_exception_internal(EXCP_DEBUG);
11925 }
11926 }
11927 if (dc->condjmp) {
11928 /* "Condition failed" instruction codepath. */
11929 gen_set_label(dc->condlabel);
11930 gen_set_condexec(dc);
11931 gen_set_pc_im(dc, dc->pc);
11932 if (dc->ss_active) {
11933 gen_step_complete_exception(dc);
11934 } else {
11935 gen_exception_internal(EXCP_DEBUG);
11936 }
11937 }
11938 } else {
11939 /* While branches must always occur at the end of an IT block,
11940 there are a few other things that can cause us to terminate
11941 the TB in the middle of an IT block:
11942 - Exception generating instructions (bkpt, swi, undefined).
11943 - Page boundaries.
11944 - Hardware watchpoints.
11945 Hardware breakpoints have already been handled and skip this code.
11946 */
11947 gen_set_condexec(dc);
11948 switch(dc->is_jmp) {
11949 case DISAS_NEXT:
11950 gen_goto_tb(dc, 1, dc->pc);
11951 break;
11952 case DISAS_UPDATE:
11953 gen_set_pc_im(dc, dc->pc);
11954 /* fall through */
11955 case DISAS_JUMP:
11956 default:
11957 /* indicate that the hash table must be used to find the next TB */
11958 tcg_gen_exit_tb(0);
11959 break;
11960 case DISAS_TB_JUMP:
11961 /* nothing more to generate */
11962 break;
11963 case DISAS_WFI:
11964 gen_helper_wfi(cpu_env);
11965 /* The helper doesn't necessarily throw an exception, but we
11966 * must go back to the main loop to check for interrupts anyway.
11967 */
11968 tcg_gen_exit_tb(0);
11969 break;
11970 case DISAS_WFE:
11971 gen_helper_wfe(cpu_env);
11972 break;
11973 case DISAS_YIELD:
11974 gen_helper_yield(cpu_env);
11975 break;
11976 case DISAS_SWI:
11977 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
11978 default_exception_el(dc));
11979 break;
11980 case DISAS_HVC:
11981 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
11982 break;
11983 case DISAS_SMC:
11984 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
11985 break;
11986 }
11987 if (dc->condjmp) {
11988 gen_set_label(dc->condlabel);
11989 gen_set_condexec(dc);
11990 gen_goto_tb(dc, 1, dc->pc);
11991 dc->condjmp = 0;
11992 }
11993 }
11994
11995 done_generating:
11996 gen_tb_end(tb, num_insns);
11997
11998 #ifdef DEBUG_DISAS
11999 if (qemu_loglevel_mask(CPU_LOG_TB_IN_ASM) &&
12000 qemu_log_in_addr_range(pc_start)) {
12001 qemu_log("----------------\n");
12002 qemu_log("IN: %s\n", lookup_symbol(pc_start));
12003 log_target_disas(cs, pc_start, dc->pc - pc_start,
12004 dc->thumb | (dc->sctlr_b << 1));
12005 qemu_log("\n");
12006 }
12007 #endif
12008 tb->size = dc->pc - pc_start;
12009 tb->icount = num_insns;
12010 }
12011
12012 static const char *cpu_mode_names[16] = {
12013 "usr", "fiq", "irq", "svc", "???", "???", "mon", "abt",
12014 "???", "???", "hyp", "und", "???", "???", "???", "sys"
12015 };
12016
12017 void arm_cpu_dump_state(CPUState *cs, FILE *f, fprintf_function cpu_fprintf,
12018 int flags)
12019 {
12020 ARMCPU *cpu = ARM_CPU(cs);
12021 CPUARMState *env = &cpu->env;
12022 int i;
12023 uint32_t psr;
12024 const char *ns_status;
12025
12026 if (is_a64(env)) {
12027 aarch64_cpu_dump_state(cs, f, cpu_fprintf, flags);
12028 return;
12029 }
12030
12031 for(i=0;i<16;i++) {
12032 cpu_fprintf(f, "R%02d=%08x", i, env->regs[i]);
12033 if ((i % 4) == 3)
12034 cpu_fprintf(f, "\n");
12035 else
12036 cpu_fprintf(f, " ");
12037 }
12038 psr = cpsr_read(env);
12039
12040 if (arm_feature(env, ARM_FEATURE_EL3) &&
12041 (psr & CPSR_M) != ARM_CPU_MODE_MON) {
12042 ns_status = env->cp15.scr_el3 & SCR_NS ? "NS " : "S ";
12043 } else {
12044 ns_status = "";
12045 }
12046
12047 cpu_fprintf(f, "PSR=%08x %c%c%c%c %c %s%s%d\n",
12048 psr,
12049 psr & (1 << 31) ? 'N' : '-',
12050 psr & (1 << 30) ? 'Z' : '-',
12051 psr & (1 << 29) ? 'C' : '-',
12052 psr & (1 << 28) ? 'V' : '-',
12053 psr & CPSR_T ? 'T' : 'A',
12054 ns_status,
12055 cpu_mode_names[psr & 0xf], (psr & 0x10) ? 32 : 26);
12056
12057 if (flags & CPU_DUMP_FPU) {
12058 int numvfpregs = 0;
12059 if (arm_feature(env, ARM_FEATURE_VFP)) {
12060 numvfpregs += 16;
12061 }
12062 if (arm_feature(env, ARM_FEATURE_VFP3)) {
12063 numvfpregs += 16;
12064 }
12065 for (i = 0; i < numvfpregs; i++) {
12066 uint64_t v = float64_val(env->vfp.regs[i]);
12067 cpu_fprintf(f, "s%02d=%08x s%02d=%08x d%02d=%016" PRIx64 "\n",
12068 i * 2, (uint32_t)v,
12069 i * 2 + 1, (uint32_t)(v >> 32),
12070 i, v);
12071 }
12072 cpu_fprintf(f, "FPSCR: %08x\n", (int)env->vfp.xregs[ARM_VFP_FPSCR]);
12073 }
12074 }
12075
12076 void restore_state_to_opc(CPUARMState *env, TranslationBlock *tb,
12077 target_ulong *data)
12078 {
12079 if (is_a64(env)) {
12080 env->pc = data[0];
12081 env->condexec_bits = 0;
12082 env->exception.syndrome = data[2] << ARM_INSN_START_WORD2_SHIFT;
12083 } else {
12084 env->regs[15] = data[0];
12085 env->condexec_bits = data[1];
12086 env->exception.syndrome = data[2] << ARM_INSN_START_WORD2_SHIFT;
12087 }
12088 }
Random patches