4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2005-2007 CodeSourcery
6 * Copyright (c) 2007 OpenedHand, Ltd.
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.
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.
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/>.
21 #include "qemu/osdep.h"
24 #include "internals.h"
25 #include "disas/disas.h"
26 #include "exec/exec-all.h"
28 #include "tcg-op-gvec.h"
30 #include "qemu/bitops.h"
32 #include "exec/semihost.h"
34 #include "exec/helper-proto.h"
35 #include "exec/helper-gen.h"
37 #include "trace-tcg.h"
41 #define ENABLE_ARCH_4T arm_dc_feature(s, ARM_FEATURE_V4T)
42 #define ENABLE_ARCH_5 arm_dc_feature(s, ARM_FEATURE_V5)
43 /* currently all emulated v5 cores are also v5TE, so don't bother */
44 #define ENABLE_ARCH_5TE arm_dc_feature(s, ARM_FEATURE_V5)
45 #define ENABLE_ARCH_5J arm_dc_feature(s, ARM_FEATURE_JAZELLE)
46 #define ENABLE_ARCH_6 arm_dc_feature(s, ARM_FEATURE_V6)
47 #define ENABLE_ARCH_6K arm_dc_feature(s, ARM_FEATURE_V6K)
48 #define ENABLE_ARCH_6T2 arm_dc_feature(s, ARM_FEATURE_THUMB2)
49 #define ENABLE_ARCH_7 arm_dc_feature(s, ARM_FEATURE_V7)
50 #define ENABLE_ARCH_8 arm_dc_feature(s, ARM_FEATURE_V8)
52 #define ARCH(x) do { if (!ENABLE_ARCH_##x) goto illegal_op; } while(0)
54 #include "translate.h"
56 #if defined(CONFIG_USER_ONLY)
59 #define IS_USER(s) (s->user)
62 /* We reuse the same 64-bit temporaries for efficiency. */
63 static TCGv_i64 cpu_V0, cpu_V1, cpu_M0;
64 static TCGv_i32 cpu_R[16];
65 TCGv_i32 cpu_CF, cpu_NF, cpu_VF, cpu_ZF;
66 TCGv_i64 cpu_exclusive_addr;
67 TCGv_i64 cpu_exclusive_val;
69 /* FIXME: These should be removed. */
70 static TCGv_i32 cpu_F0s, cpu_F1s;
71 static TCGv_i64 cpu_F0d, cpu_F1d;
73 #include "exec/gen-icount.h"
75 static const char *regnames[] =
76 { "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7",
77 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "pc" };
79 /* Function prototypes for gen_ functions calling Neon helpers. */
80 typedef void NeonGenThreeOpEnvFn(TCGv_i32, TCGv_env, TCGv_i32,
83 /* initialize TCG globals. */
84 void arm_translate_init(void)
88 for (i = 0; i < 16; i++) {
89 cpu_R[i] = tcg_global_mem_new_i32(cpu_env,
90 offsetof(CPUARMState, regs[i]),
93 cpu_CF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, CF), "CF");
94 cpu_NF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, NF), "NF");
95 cpu_VF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, VF), "VF");
96 cpu_ZF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, ZF), "ZF");
98 cpu_exclusive_addr = tcg_global_mem_new_i64(cpu_env,
99 offsetof(CPUARMState, exclusive_addr), "exclusive_addr");
100 cpu_exclusive_val = tcg_global_mem_new_i64(cpu_env,
101 offsetof(CPUARMState, exclusive_val), "exclusive_val");
103 a64_translate_init();
106 /* Flags for the disas_set_da_iss info argument:
107 * lower bits hold the Rt register number, higher bits are flags.
109 typedef enum ISSInfo {
112 ISSInvalid = (1 << 5),
113 ISSIsAcqRel = (1 << 6),
114 ISSIsWrite = (1 << 7),
115 ISSIs16Bit = (1 << 8),
118 /* Save the syndrome information for a Data Abort */
119 static void disas_set_da_iss(DisasContext *s, TCGMemOp memop, ISSInfo issinfo)
122 int sas = memop & MO_SIZE;
123 bool sse = memop & MO_SIGN;
124 bool is_acqrel = issinfo & ISSIsAcqRel;
125 bool is_write = issinfo & ISSIsWrite;
126 bool is_16bit = issinfo & ISSIs16Bit;
127 int srt = issinfo & ISSRegMask;
129 if (issinfo & ISSInvalid) {
130 /* Some callsites want to conditionally provide ISS info,
131 * eg "only if this was not a writeback"
137 /* For AArch32, insns where the src/dest is R15 never generate
138 * ISS information. Catching that here saves checking at all
144 syn = syn_data_abort_with_iss(0, sas, sse, srt, 0, is_acqrel,
145 0, 0, 0, is_write, 0, is_16bit);
146 disas_set_insn_syndrome(s, syn);
149 static inline int get_a32_user_mem_index(DisasContext *s)
151 /* Return the core mmu_idx to use for A32/T32 "unprivileged load/store"
153 * if PL2, UNPREDICTABLE (we choose to implement as if PL0)
154 * otherwise, access as if at PL0.
156 switch (s->mmu_idx) {
157 case ARMMMUIdx_S1E2: /* this one is UNPREDICTABLE */
158 case ARMMMUIdx_S12NSE0:
159 case ARMMMUIdx_S12NSE1:
160 return arm_to_core_mmu_idx(ARMMMUIdx_S12NSE0);
162 case ARMMMUIdx_S1SE0:
163 case ARMMMUIdx_S1SE1:
164 return arm_to_core_mmu_idx(ARMMMUIdx_S1SE0);
165 case ARMMMUIdx_MUser:
166 case ARMMMUIdx_MPriv:
167 return arm_to_core_mmu_idx(ARMMMUIdx_MUser);
168 case ARMMMUIdx_MUserNegPri:
169 case ARMMMUIdx_MPrivNegPri:
170 return arm_to_core_mmu_idx(ARMMMUIdx_MUserNegPri);
171 case ARMMMUIdx_MSUser:
172 case ARMMMUIdx_MSPriv:
173 return arm_to_core_mmu_idx(ARMMMUIdx_MSUser);
174 case ARMMMUIdx_MSUserNegPri:
175 case ARMMMUIdx_MSPrivNegPri:
176 return arm_to_core_mmu_idx(ARMMMUIdx_MSUserNegPri);
179 g_assert_not_reached();
183 static inline TCGv_i32 load_cpu_offset(int offset)
185 TCGv_i32 tmp = tcg_temp_new_i32();
186 tcg_gen_ld_i32(tmp, cpu_env, offset);
190 #define load_cpu_field(name) load_cpu_offset(offsetof(CPUARMState, name))
192 static inline void store_cpu_offset(TCGv_i32 var, int offset)
194 tcg_gen_st_i32(var, cpu_env, offset);
195 tcg_temp_free_i32(var);
198 #define store_cpu_field(var, name) \
199 store_cpu_offset(var, offsetof(CPUARMState, name))
201 /* Set a variable to the value of a CPU register. */
202 static void load_reg_var(DisasContext *s, TCGv_i32 var, int reg)
206 /* normally, since we updated PC, we need only to add one insn */
208 addr = (long)s->pc + 2;
210 addr = (long)s->pc + 4;
211 tcg_gen_movi_i32(var, addr);
213 tcg_gen_mov_i32(var, cpu_R[reg]);
217 /* Create a new temporary and set it to the value of a CPU register. */
218 static inline TCGv_i32 load_reg(DisasContext *s, int reg)
220 TCGv_i32 tmp = tcg_temp_new_i32();
221 load_reg_var(s, tmp, reg);
225 /* Set a CPU register. The source must be a temporary and will be
227 static void store_reg(DisasContext *s, int reg, TCGv_i32 var)
230 /* In Thumb mode, we must ignore bit 0.
231 * In ARM mode, for ARMv4 and ARMv5, it is UNPREDICTABLE if bits [1:0]
232 * are not 0b00, but for ARMv6 and above, we must ignore bits [1:0].
233 * We choose to ignore [1:0] in ARM mode for all architecture versions.
235 tcg_gen_andi_i32(var, var, s->thumb ? ~1 : ~3);
236 s->base.is_jmp = DISAS_JUMP;
238 tcg_gen_mov_i32(cpu_R[reg], var);
239 tcg_temp_free_i32(var);
242 /* Value extensions. */
243 #define gen_uxtb(var) tcg_gen_ext8u_i32(var, var)
244 #define gen_uxth(var) tcg_gen_ext16u_i32(var, var)
245 #define gen_sxtb(var) tcg_gen_ext8s_i32(var, var)
246 #define gen_sxth(var) tcg_gen_ext16s_i32(var, var)
248 #define gen_sxtb16(var) gen_helper_sxtb16(var, var)
249 #define gen_uxtb16(var) gen_helper_uxtb16(var, var)
252 static inline void gen_set_cpsr(TCGv_i32 var, uint32_t mask)
254 TCGv_i32 tmp_mask = tcg_const_i32(mask);
255 gen_helper_cpsr_write(cpu_env, var, tmp_mask);
256 tcg_temp_free_i32(tmp_mask);
258 /* Set NZCV flags from the high 4 bits of var. */
259 #define gen_set_nzcv(var) gen_set_cpsr(var, CPSR_NZCV)
261 static void gen_exception_internal(int excp)
263 TCGv_i32 tcg_excp = tcg_const_i32(excp);
265 assert(excp_is_internal(excp));
266 gen_helper_exception_internal(cpu_env, tcg_excp);
267 tcg_temp_free_i32(tcg_excp);
270 static void gen_exception(int excp, uint32_t syndrome, uint32_t target_el)
272 TCGv_i32 tcg_excp = tcg_const_i32(excp);
273 TCGv_i32 tcg_syn = tcg_const_i32(syndrome);
274 TCGv_i32 tcg_el = tcg_const_i32(target_el);
276 gen_helper_exception_with_syndrome(cpu_env, tcg_excp,
279 tcg_temp_free_i32(tcg_el);
280 tcg_temp_free_i32(tcg_syn);
281 tcg_temp_free_i32(tcg_excp);
284 static void gen_ss_advance(DisasContext *s)
286 /* If the singlestep state is Active-not-pending, advance to
291 gen_helper_clear_pstate_ss(cpu_env);
295 static void gen_step_complete_exception(DisasContext *s)
297 /* We just completed step of an insn. Move from Active-not-pending
298 * to Active-pending, and then also take the swstep exception.
299 * This corresponds to making the (IMPDEF) choice to prioritize
300 * swstep exceptions over asynchronous exceptions taken to an exception
301 * level where debug is disabled. This choice has the advantage that
302 * we do not need to maintain internal state corresponding to the
303 * ISV/EX syndrome bits between completion of the step and generation
304 * of the exception, and our syndrome information is always correct.
307 gen_exception(EXCP_UDEF, syn_swstep(s->ss_same_el, 1, s->is_ldex),
308 default_exception_el(s));
309 s->base.is_jmp = DISAS_NORETURN;
312 static void gen_singlestep_exception(DisasContext *s)
314 /* Generate the right kind of exception for singlestep, which is
315 * either the architectural singlestep or EXCP_DEBUG for QEMU's
316 * gdb singlestepping.
319 gen_step_complete_exception(s);
321 gen_exception_internal(EXCP_DEBUG);
325 static inline bool is_singlestepping(DisasContext *s)
327 /* Return true if we are singlestepping either because of
328 * architectural singlestep or QEMU gdbstub singlestep. This does
329 * not include the command line '-singlestep' mode which is rather
330 * misnamed as it only means "one instruction per TB" and doesn't
331 * affect the code we generate.
333 return s->base.singlestep_enabled || s->ss_active;
336 static void gen_smul_dual(TCGv_i32 a, TCGv_i32 b)
338 TCGv_i32 tmp1 = tcg_temp_new_i32();
339 TCGv_i32 tmp2 = tcg_temp_new_i32();
340 tcg_gen_ext16s_i32(tmp1, a);
341 tcg_gen_ext16s_i32(tmp2, b);
342 tcg_gen_mul_i32(tmp1, tmp1, tmp2);
343 tcg_temp_free_i32(tmp2);
344 tcg_gen_sari_i32(a, a, 16);
345 tcg_gen_sari_i32(b, b, 16);
346 tcg_gen_mul_i32(b, b, a);
347 tcg_gen_mov_i32(a, tmp1);
348 tcg_temp_free_i32(tmp1);
351 /* Byteswap each halfword. */
352 static void gen_rev16(TCGv_i32 var)
354 TCGv_i32 tmp = tcg_temp_new_i32();
355 TCGv_i32 mask = tcg_const_i32(0x00ff00ff);
356 tcg_gen_shri_i32(tmp, var, 8);
357 tcg_gen_and_i32(tmp, tmp, mask);
358 tcg_gen_and_i32(var, var, mask);
359 tcg_gen_shli_i32(var, var, 8);
360 tcg_gen_or_i32(var, var, tmp);
361 tcg_temp_free_i32(mask);
362 tcg_temp_free_i32(tmp);
365 /* Byteswap low halfword and sign extend. */
366 static void gen_revsh(TCGv_i32 var)
368 tcg_gen_ext16u_i32(var, var);
369 tcg_gen_bswap16_i32(var, var);
370 tcg_gen_ext16s_i32(var, var);
373 /* Return (b << 32) + a. Mark inputs as dead */
374 static TCGv_i64 gen_addq_msw(TCGv_i64 a, TCGv_i32 b)
376 TCGv_i64 tmp64 = tcg_temp_new_i64();
378 tcg_gen_extu_i32_i64(tmp64, b);
379 tcg_temp_free_i32(b);
380 tcg_gen_shli_i64(tmp64, tmp64, 32);
381 tcg_gen_add_i64(a, tmp64, a);
383 tcg_temp_free_i64(tmp64);
387 /* Return (b << 32) - a. Mark inputs as dead. */
388 static TCGv_i64 gen_subq_msw(TCGv_i64 a, TCGv_i32 b)
390 TCGv_i64 tmp64 = tcg_temp_new_i64();
392 tcg_gen_extu_i32_i64(tmp64, b);
393 tcg_temp_free_i32(b);
394 tcg_gen_shli_i64(tmp64, tmp64, 32);
395 tcg_gen_sub_i64(a, tmp64, a);
397 tcg_temp_free_i64(tmp64);
401 /* 32x32->64 multiply. Marks inputs as dead. */
402 static TCGv_i64 gen_mulu_i64_i32(TCGv_i32 a, TCGv_i32 b)
404 TCGv_i32 lo = tcg_temp_new_i32();
405 TCGv_i32 hi = tcg_temp_new_i32();
408 tcg_gen_mulu2_i32(lo, hi, a, b);
409 tcg_temp_free_i32(a);
410 tcg_temp_free_i32(b);
412 ret = tcg_temp_new_i64();
413 tcg_gen_concat_i32_i64(ret, lo, hi);
414 tcg_temp_free_i32(lo);
415 tcg_temp_free_i32(hi);
420 static TCGv_i64 gen_muls_i64_i32(TCGv_i32 a, TCGv_i32 b)
422 TCGv_i32 lo = tcg_temp_new_i32();
423 TCGv_i32 hi = tcg_temp_new_i32();
426 tcg_gen_muls2_i32(lo, hi, a, b);
427 tcg_temp_free_i32(a);
428 tcg_temp_free_i32(b);
430 ret = tcg_temp_new_i64();
431 tcg_gen_concat_i32_i64(ret, lo, hi);
432 tcg_temp_free_i32(lo);
433 tcg_temp_free_i32(hi);
438 /* Swap low and high halfwords. */
439 static void gen_swap_half(TCGv_i32 var)
441 TCGv_i32 tmp = tcg_temp_new_i32();
442 tcg_gen_shri_i32(tmp, var, 16);
443 tcg_gen_shli_i32(var, var, 16);
444 tcg_gen_or_i32(var, var, tmp);
445 tcg_temp_free_i32(tmp);
448 /* Dual 16-bit add. Result placed in t0 and t1 is marked as dead.
449 tmp = (t0 ^ t1) & 0x8000;
452 t0 = (t0 + t1) ^ tmp;
455 static void gen_add16(TCGv_i32 t0, TCGv_i32 t1)
457 TCGv_i32 tmp = tcg_temp_new_i32();
458 tcg_gen_xor_i32(tmp, t0, t1);
459 tcg_gen_andi_i32(tmp, tmp, 0x8000);
460 tcg_gen_andi_i32(t0, t0, ~0x8000);
461 tcg_gen_andi_i32(t1, t1, ~0x8000);
462 tcg_gen_add_i32(t0, t0, t1);
463 tcg_gen_xor_i32(t0, t0, tmp);
464 tcg_temp_free_i32(tmp);
465 tcg_temp_free_i32(t1);
468 /* Set CF to the top bit of var. */
469 static void gen_set_CF_bit31(TCGv_i32 var)
471 tcg_gen_shri_i32(cpu_CF, var, 31);
474 /* Set N and Z flags from var. */
475 static inline void gen_logic_CC(TCGv_i32 var)
477 tcg_gen_mov_i32(cpu_NF, var);
478 tcg_gen_mov_i32(cpu_ZF, var);
482 static void gen_adc(TCGv_i32 t0, TCGv_i32 t1)
484 tcg_gen_add_i32(t0, t0, t1);
485 tcg_gen_add_i32(t0, t0, cpu_CF);
488 /* dest = T0 + T1 + CF. */
489 static void gen_add_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
491 tcg_gen_add_i32(dest, t0, t1);
492 tcg_gen_add_i32(dest, dest, cpu_CF);
495 /* dest = T0 - T1 + CF - 1. */
496 static void gen_sub_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
498 tcg_gen_sub_i32(dest, t0, t1);
499 tcg_gen_add_i32(dest, dest, cpu_CF);
500 tcg_gen_subi_i32(dest, dest, 1);
503 /* dest = T0 + T1. Compute C, N, V and Z flags */
504 static void gen_add_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
506 TCGv_i32 tmp = tcg_temp_new_i32();
507 tcg_gen_movi_i32(tmp, 0);
508 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, t1, tmp);
509 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
510 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
511 tcg_gen_xor_i32(tmp, t0, t1);
512 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
513 tcg_temp_free_i32(tmp);
514 tcg_gen_mov_i32(dest, cpu_NF);
517 /* dest = T0 + T1 + CF. Compute C, N, V and Z flags */
518 static void gen_adc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
520 TCGv_i32 tmp = tcg_temp_new_i32();
521 if (TCG_TARGET_HAS_add2_i32) {
522 tcg_gen_movi_i32(tmp, 0);
523 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, cpu_CF, tmp);
524 tcg_gen_add2_i32(cpu_NF, cpu_CF, cpu_NF, cpu_CF, t1, tmp);
526 TCGv_i64 q0 = tcg_temp_new_i64();
527 TCGv_i64 q1 = tcg_temp_new_i64();
528 tcg_gen_extu_i32_i64(q0, t0);
529 tcg_gen_extu_i32_i64(q1, t1);
530 tcg_gen_add_i64(q0, q0, q1);
531 tcg_gen_extu_i32_i64(q1, cpu_CF);
532 tcg_gen_add_i64(q0, q0, q1);
533 tcg_gen_extr_i64_i32(cpu_NF, cpu_CF, q0);
534 tcg_temp_free_i64(q0);
535 tcg_temp_free_i64(q1);
537 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
538 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
539 tcg_gen_xor_i32(tmp, t0, t1);
540 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
541 tcg_temp_free_i32(tmp);
542 tcg_gen_mov_i32(dest, cpu_NF);
545 /* dest = T0 - T1. Compute C, N, V and Z flags */
546 static void gen_sub_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
549 tcg_gen_sub_i32(cpu_NF, t0, t1);
550 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
551 tcg_gen_setcond_i32(TCG_COND_GEU, cpu_CF, t0, t1);
552 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
553 tmp = tcg_temp_new_i32();
554 tcg_gen_xor_i32(tmp, t0, t1);
555 tcg_gen_and_i32(cpu_VF, cpu_VF, tmp);
556 tcg_temp_free_i32(tmp);
557 tcg_gen_mov_i32(dest, cpu_NF);
560 /* dest = T0 + ~T1 + CF. Compute C, N, V and Z flags */
561 static void gen_sbc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
563 TCGv_i32 tmp = tcg_temp_new_i32();
564 tcg_gen_not_i32(tmp, t1);
565 gen_adc_CC(dest, t0, tmp);
566 tcg_temp_free_i32(tmp);
569 #define GEN_SHIFT(name) \
570 static void gen_##name(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1) \
572 TCGv_i32 tmp1, tmp2, tmp3; \
573 tmp1 = tcg_temp_new_i32(); \
574 tcg_gen_andi_i32(tmp1, t1, 0xff); \
575 tmp2 = tcg_const_i32(0); \
576 tmp3 = tcg_const_i32(0x1f); \
577 tcg_gen_movcond_i32(TCG_COND_GTU, tmp2, tmp1, tmp3, tmp2, t0); \
578 tcg_temp_free_i32(tmp3); \
579 tcg_gen_andi_i32(tmp1, tmp1, 0x1f); \
580 tcg_gen_##name##_i32(dest, tmp2, tmp1); \
581 tcg_temp_free_i32(tmp2); \
582 tcg_temp_free_i32(tmp1); \
588 static void gen_sar(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
591 tmp1 = tcg_temp_new_i32();
592 tcg_gen_andi_i32(tmp1, t1, 0xff);
593 tmp2 = tcg_const_i32(0x1f);
594 tcg_gen_movcond_i32(TCG_COND_GTU, tmp1, tmp1, tmp2, tmp2, tmp1);
595 tcg_temp_free_i32(tmp2);
596 tcg_gen_sar_i32(dest, t0, tmp1);
597 tcg_temp_free_i32(tmp1);
600 static void tcg_gen_abs_i32(TCGv_i32 dest, TCGv_i32 src)
602 TCGv_i32 c0 = tcg_const_i32(0);
603 TCGv_i32 tmp = tcg_temp_new_i32();
604 tcg_gen_neg_i32(tmp, src);
605 tcg_gen_movcond_i32(TCG_COND_GT, dest, src, c0, src, tmp);
606 tcg_temp_free_i32(c0);
607 tcg_temp_free_i32(tmp);
610 static void shifter_out_im(TCGv_i32 var, int shift)
613 tcg_gen_andi_i32(cpu_CF, var, 1);
615 tcg_gen_shri_i32(cpu_CF, var, shift);
617 tcg_gen_andi_i32(cpu_CF, cpu_CF, 1);
622 /* Shift by immediate. Includes special handling for shift == 0. */
623 static inline void gen_arm_shift_im(TCGv_i32 var, int shiftop,
624 int shift, int flags)
630 shifter_out_im(var, 32 - shift);
631 tcg_gen_shli_i32(var, var, shift);
637 tcg_gen_shri_i32(cpu_CF, var, 31);
639 tcg_gen_movi_i32(var, 0);
642 shifter_out_im(var, shift - 1);
643 tcg_gen_shri_i32(var, var, shift);
650 shifter_out_im(var, shift - 1);
653 tcg_gen_sari_i32(var, var, shift);
655 case 3: /* ROR/RRX */
658 shifter_out_im(var, shift - 1);
659 tcg_gen_rotri_i32(var, var, shift); break;
661 TCGv_i32 tmp = tcg_temp_new_i32();
662 tcg_gen_shli_i32(tmp, cpu_CF, 31);
664 shifter_out_im(var, 0);
665 tcg_gen_shri_i32(var, var, 1);
666 tcg_gen_or_i32(var, var, tmp);
667 tcg_temp_free_i32(tmp);
672 static inline void gen_arm_shift_reg(TCGv_i32 var, int shiftop,
673 TCGv_i32 shift, int flags)
677 case 0: gen_helper_shl_cc(var, cpu_env, var, shift); break;
678 case 1: gen_helper_shr_cc(var, cpu_env, var, shift); break;
679 case 2: gen_helper_sar_cc(var, cpu_env, var, shift); break;
680 case 3: gen_helper_ror_cc(var, cpu_env, var, shift); break;
685 gen_shl(var, var, shift);
688 gen_shr(var, var, shift);
691 gen_sar(var, var, shift);
693 case 3: tcg_gen_andi_i32(shift, shift, 0x1f);
694 tcg_gen_rotr_i32(var, var, shift); break;
697 tcg_temp_free_i32(shift);
700 #define PAS_OP(pfx) \
702 case 0: gen_pas_helper(glue(pfx,add16)); break; \
703 case 1: gen_pas_helper(glue(pfx,addsubx)); break; \
704 case 2: gen_pas_helper(glue(pfx,subaddx)); break; \
705 case 3: gen_pas_helper(glue(pfx,sub16)); break; \
706 case 4: gen_pas_helper(glue(pfx,add8)); break; \
707 case 7: gen_pas_helper(glue(pfx,sub8)); break; \
709 static void gen_arm_parallel_addsub(int op1, int op2, TCGv_i32 a, TCGv_i32 b)
714 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b, tmp)
716 tmp = tcg_temp_new_ptr();
717 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
719 tcg_temp_free_ptr(tmp);
722 tmp = tcg_temp_new_ptr();
723 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
725 tcg_temp_free_ptr(tmp);
727 #undef gen_pas_helper
728 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b)
741 #undef gen_pas_helper
746 /* For unknown reasons Arm and Thumb-2 use arbitrarily different encodings. */
747 #define PAS_OP(pfx) \
749 case 0: gen_pas_helper(glue(pfx,add8)); break; \
750 case 1: gen_pas_helper(glue(pfx,add16)); break; \
751 case 2: gen_pas_helper(glue(pfx,addsubx)); break; \
752 case 4: gen_pas_helper(glue(pfx,sub8)); break; \
753 case 5: gen_pas_helper(glue(pfx,sub16)); break; \
754 case 6: gen_pas_helper(glue(pfx,subaddx)); break; \
756 static void gen_thumb2_parallel_addsub(int op1, int op2, TCGv_i32 a, TCGv_i32 b)
761 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b, tmp)
763 tmp = tcg_temp_new_ptr();
764 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
766 tcg_temp_free_ptr(tmp);
769 tmp = tcg_temp_new_ptr();
770 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
772 tcg_temp_free_ptr(tmp);
774 #undef gen_pas_helper
775 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b)
788 #undef gen_pas_helper
794 * Generate a conditional based on ARM condition code cc.
795 * This is common between ARM and Aarch64 targets.
797 void arm_test_cc(DisasCompare *cmp, int cc)
828 case 8: /* hi: C && !Z */
829 case 9: /* ls: !C || Z -> !(C && !Z) */
831 value = tcg_temp_new_i32();
833 /* CF is 1 for C, so -CF is an all-bits-set mask for C;
834 ZF is non-zero for !Z; so AND the two subexpressions. */
835 tcg_gen_neg_i32(value, cpu_CF);
836 tcg_gen_and_i32(value, value, cpu_ZF);
839 case 10: /* ge: N == V -> N ^ V == 0 */
840 case 11: /* lt: N != V -> N ^ V != 0 */
841 /* Since we're only interested in the sign bit, == 0 is >= 0. */
843 value = tcg_temp_new_i32();
845 tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
848 case 12: /* gt: !Z && N == V */
849 case 13: /* le: Z || N != V */
851 value = tcg_temp_new_i32();
853 /* (N == V) is equal to the sign bit of ~(NF ^ VF). Propagate
854 * the sign bit then AND with ZF to yield the result. */
855 tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
856 tcg_gen_sari_i32(value, value, 31);
857 tcg_gen_andc_i32(value, cpu_ZF, value);
860 case 14: /* always */
861 case 15: /* always */
862 /* Use the ALWAYS condition, which will fold early.
863 * It doesn't matter what we use for the value. */
864 cond = TCG_COND_ALWAYS;
869 fprintf(stderr, "Bad condition code 0x%x\n", cc);
874 cond = tcg_invert_cond(cond);
880 cmp->value_global = global;
883 void arm_free_cc(DisasCompare *cmp)
885 if (!cmp->value_global) {
886 tcg_temp_free_i32(cmp->value);
890 void arm_jump_cc(DisasCompare *cmp, TCGLabel *label)
892 tcg_gen_brcondi_i32(cmp->cond, cmp->value, 0, label);
895 void arm_gen_test_cc(int cc, TCGLabel *label)
898 arm_test_cc(&cmp, cc);
899 arm_jump_cc(&cmp, label);
903 static const uint8_t table_logic_cc[16] = {
922 static inline void gen_set_condexec(DisasContext *s)
924 if (s->condexec_mask) {
925 uint32_t val = (s->condexec_cond << 4) | (s->condexec_mask >> 1);
926 TCGv_i32 tmp = tcg_temp_new_i32();
927 tcg_gen_movi_i32(tmp, val);
928 store_cpu_field(tmp, condexec_bits);
932 static inline void gen_set_pc_im(DisasContext *s, target_ulong val)
934 tcg_gen_movi_i32(cpu_R[15], val);
937 /* Set PC and Thumb state from an immediate address. */
938 static inline void gen_bx_im(DisasContext *s, uint32_t addr)
942 s->base.is_jmp = DISAS_JUMP;
943 if (s->thumb != (addr & 1)) {
944 tmp = tcg_temp_new_i32();
945 tcg_gen_movi_i32(tmp, addr & 1);
946 tcg_gen_st_i32(tmp, cpu_env, offsetof(CPUARMState, thumb));
947 tcg_temp_free_i32(tmp);
949 tcg_gen_movi_i32(cpu_R[15], addr & ~1);
952 /* Set PC and Thumb state from var. var is marked as dead. */
953 static inline void gen_bx(DisasContext *s, TCGv_i32 var)
955 s->base.is_jmp = DISAS_JUMP;
956 tcg_gen_andi_i32(cpu_R[15], var, ~1);
957 tcg_gen_andi_i32(var, var, 1);
958 store_cpu_field(var, thumb);
961 /* Set PC and Thumb state from var. var is marked as dead.
962 * For M-profile CPUs, include logic to detect exception-return
963 * branches and handle them. This is needed for Thumb POP/LDM to PC, LDR to PC,
964 * and BX reg, and no others, and happens only for code in Handler mode.
966 static inline void gen_bx_excret(DisasContext *s, TCGv_i32 var)
968 /* Generate the same code here as for a simple bx, but flag via
969 * s->base.is_jmp that we need to do the rest of the work later.
972 if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY) ||
973 (s->v7m_handler_mode && arm_dc_feature(s, ARM_FEATURE_M))) {
974 s->base.is_jmp = DISAS_BX_EXCRET;
978 static inline void gen_bx_excret_final_code(DisasContext *s)
980 /* Generate the code to finish possible exception return and end the TB */
981 TCGLabel *excret_label = gen_new_label();
984 if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY)) {
985 /* Covers FNC_RETURN and EXC_RETURN magic */
986 min_magic = FNC_RETURN_MIN_MAGIC;
988 /* EXC_RETURN magic only */
989 min_magic = EXC_RETURN_MIN_MAGIC;
992 /* Is the new PC value in the magic range indicating exception return? */
993 tcg_gen_brcondi_i32(TCG_COND_GEU, cpu_R[15], min_magic, excret_label);
994 /* No: end the TB as we would for a DISAS_JMP */
995 if (is_singlestepping(s)) {
996 gen_singlestep_exception(s);
998 tcg_gen_exit_tb(NULL, 0);
1000 gen_set_label(excret_label);
1001 /* Yes: this is an exception return.
1002 * At this point in runtime env->regs[15] and env->thumb will hold
1003 * the exception-return magic number, which do_v7m_exception_exit()
1004 * will read. Nothing else will be able to see those values because
1005 * the cpu-exec main loop guarantees that we will always go straight
1006 * from raising the exception to the exception-handling code.
1008 * gen_ss_advance(s) does nothing on M profile currently but
1009 * calling it is conceptually the right thing as we have executed
1010 * this instruction (compare SWI, HVC, SMC handling).
1013 gen_exception_internal(EXCP_EXCEPTION_EXIT);
1016 static inline void gen_bxns(DisasContext *s, int rm)
1018 TCGv_i32 var = load_reg(s, rm);
1020 /* The bxns helper may raise an EXCEPTION_EXIT exception, so in theory
1021 * we need to sync state before calling it, but:
1022 * - we don't need to do gen_set_pc_im() because the bxns helper will
1023 * always set the PC itself
1024 * - we don't need to do gen_set_condexec() because BXNS is UNPREDICTABLE
1025 * unless it's outside an IT block or the last insn in an IT block,
1026 * so we know that condexec == 0 (already set at the top of the TB)
1027 * is correct in the non-UNPREDICTABLE cases, and we can choose
1028 * "zeroes the IT bits" as our UNPREDICTABLE behaviour otherwise.
1030 gen_helper_v7m_bxns(cpu_env, var);
1031 tcg_temp_free_i32(var);
1032 s->base.is_jmp = DISAS_EXIT;
1035 static inline void gen_blxns(DisasContext *s, int rm)
1037 TCGv_i32 var = load_reg(s, rm);
1039 /* We don't need to sync condexec state, for the same reason as bxns.
1040 * We do however need to set the PC, because the blxns helper reads it.
1041 * The blxns helper may throw an exception.
1043 gen_set_pc_im(s, s->pc);
1044 gen_helper_v7m_blxns(cpu_env, var);
1045 tcg_temp_free_i32(var);
1046 s->base.is_jmp = DISAS_EXIT;
1049 /* Variant of store_reg which uses branch&exchange logic when storing
1050 to r15 in ARM architecture v7 and above. The source must be a temporary
1051 and will be marked as dead. */
1052 static inline void store_reg_bx(DisasContext *s, int reg, TCGv_i32 var)
1054 if (reg == 15 && ENABLE_ARCH_7) {
1057 store_reg(s, reg, var);
1061 /* Variant of store_reg which uses branch&exchange logic when storing
1062 * to r15 in ARM architecture v5T and above. This is used for storing
1063 * the results of a LDR/LDM/POP into r15, and corresponds to the cases
1064 * in the ARM ARM which use the LoadWritePC() pseudocode function. */
1065 static inline void store_reg_from_load(DisasContext *s, int reg, TCGv_i32 var)
1067 if (reg == 15 && ENABLE_ARCH_5) {
1068 gen_bx_excret(s, var);
1070 store_reg(s, reg, var);
1074 #ifdef CONFIG_USER_ONLY
1075 #define IS_USER_ONLY 1
1077 #define IS_USER_ONLY 0
1080 /* Abstractions of "generate code to do a guest load/store for
1081 * AArch32", where a vaddr is always 32 bits (and is zero
1082 * extended if we're a 64 bit core) and data is also
1083 * 32 bits unless specifically doing a 64 bit access.
1084 * These functions work like tcg_gen_qemu_{ld,st}* except
1085 * that the address argument is TCGv_i32 rather than TCGv.
1088 static inline TCGv gen_aa32_addr(DisasContext *s, TCGv_i32 a32, TCGMemOp op)
1090 TCGv addr = tcg_temp_new();
1091 tcg_gen_extu_i32_tl(addr, a32);
1093 /* Not needed for user-mode BE32, where we use MO_BE instead. */
1094 if (!IS_USER_ONLY && s->sctlr_b && (op & MO_SIZE) < MO_32) {
1095 tcg_gen_xori_tl(addr, addr, 4 - (1 << (op & MO_SIZE)));
1100 static void gen_aa32_ld_i32(DisasContext *s, TCGv_i32 val, TCGv_i32 a32,
1101 int index, TCGMemOp opc)
1105 if (arm_dc_feature(s, ARM_FEATURE_M) &&
1106 !arm_dc_feature(s, ARM_FEATURE_M_MAIN)) {
1110 addr = gen_aa32_addr(s, a32, opc);
1111 tcg_gen_qemu_ld_i32(val, addr, index, opc);
1112 tcg_temp_free(addr);
1115 static void gen_aa32_st_i32(DisasContext *s, TCGv_i32 val, TCGv_i32 a32,
1116 int index, TCGMemOp opc)
1120 if (arm_dc_feature(s, ARM_FEATURE_M) &&
1121 !arm_dc_feature(s, ARM_FEATURE_M_MAIN)) {
1125 addr = gen_aa32_addr(s, a32, opc);
1126 tcg_gen_qemu_st_i32(val, addr, index, opc);
1127 tcg_temp_free(addr);
1130 #define DO_GEN_LD(SUFF, OPC) \
1131 static inline void gen_aa32_ld##SUFF(DisasContext *s, TCGv_i32 val, \
1132 TCGv_i32 a32, int index) \
1134 gen_aa32_ld_i32(s, val, a32, index, OPC | s->be_data); \
1136 static inline void gen_aa32_ld##SUFF##_iss(DisasContext *s, \
1138 TCGv_i32 a32, int index, \
1141 gen_aa32_ld##SUFF(s, val, a32, index); \
1142 disas_set_da_iss(s, OPC, issinfo); \
1145 #define DO_GEN_ST(SUFF, OPC) \
1146 static inline void gen_aa32_st##SUFF(DisasContext *s, TCGv_i32 val, \
1147 TCGv_i32 a32, int index) \
1149 gen_aa32_st_i32(s, val, a32, index, OPC | s->be_data); \
1151 static inline void gen_aa32_st##SUFF##_iss(DisasContext *s, \
1153 TCGv_i32 a32, int index, \
1156 gen_aa32_st##SUFF(s, val, a32, index); \
1157 disas_set_da_iss(s, OPC, issinfo | ISSIsWrite); \
1160 static inline void gen_aa32_frob64(DisasContext *s, TCGv_i64 val)
1162 /* Not needed for user-mode BE32, where we use MO_BE instead. */
1163 if (!IS_USER_ONLY && s->sctlr_b) {
1164 tcg_gen_rotri_i64(val, val, 32);
1168 static void gen_aa32_ld_i64(DisasContext *s, TCGv_i64 val, TCGv_i32 a32,
1169 int index, TCGMemOp opc)
1171 TCGv addr = gen_aa32_addr(s, a32, opc);
1172 tcg_gen_qemu_ld_i64(val, addr, index, opc);
1173 gen_aa32_frob64(s, val);
1174 tcg_temp_free(addr);
1177 static inline void gen_aa32_ld64(DisasContext *s, TCGv_i64 val,
1178 TCGv_i32 a32, int index)
1180 gen_aa32_ld_i64(s, val, a32, index, MO_Q | s->be_data);
1183 static void gen_aa32_st_i64(DisasContext *s, TCGv_i64 val, TCGv_i32 a32,
1184 int index, TCGMemOp opc)
1186 TCGv addr = gen_aa32_addr(s, a32, opc);
1188 /* Not needed for user-mode BE32, where we use MO_BE instead. */
1189 if (!IS_USER_ONLY && s->sctlr_b) {
1190 TCGv_i64 tmp = tcg_temp_new_i64();
1191 tcg_gen_rotri_i64(tmp, val, 32);
1192 tcg_gen_qemu_st_i64(tmp, addr, index, opc);
1193 tcg_temp_free_i64(tmp);
1195 tcg_gen_qemu_st_i64(val, addr, index, opc);
1197 tcg_temp_free(addr);
1200 static inline void gen_aa32_st64(DisasContext *s, TCGv_i64 val,
1201 TCGv_i32 a32, int index)
1203 gen_aa32_st_i64(s, val, a32, index, MO_Q | s->be_data);
1206 DO_GEN_LD(8s, MO_SB)
1207 DO_GEN_LD(8u, MO_UB)
1208 DO_GEN_LD(16s, MO_SW)
1209 DO_GEN_LD(16u, MO_UW)
1210 DO_GEN_LD(32u, MO_UL)
1212 DO_GEN_ST(16, MO_UW)
1213 DO_GEN_ST(32, MO_UL)
1215 static inline void gen_hvc(DisasContext *s, int imm16)
1217 /* The pre HVC helper handles cases when HVC gets trapped
1218 * as an undefined insn by runtime configuration (ie before
1219 * the insn really executes).
1221 gen_set_pc_im(s, s->pc - 4);
1222 gen_helper_pre_hvc(cpu_env);
1223 /* Otherwise we will treat this as a real exception which
1224 * happens after execution of the insn. (The distinction matters
1225 * for the PC value reported to the exception handler and also
1226 * for single stepping.)
1229 gen_set_pc_im(s, s->pc);
1230 s->base.is_jmp = DISAS_HVC;
1233 static inline void gen_smc(DisasContext *s)
1235 /* As with HVC, we may take an exception either before or after
1236 * the insn executes.
1240 gen_set_pc_im(s, s->pc - 4);
1241 tmp = tcg_const_i32(syn_aa32_smc());
1242 gen_helper_pre_smc(cpu_env, tmp);
1243 tcg_temp_free_i32(tmp);
1244 gen_set_pc_im(s, s->pc);
1245 s->base.is_jmp = DISAS_SMC;
1248 static void gen_exception_internal_insn(DisasContext *s, int offset, int excp)
1250 gen_set_condexec(s);
1251 gen_set_pc_im(s, s->pc - offset);
1252 gen_exception_internal(excp);
1253 s->base.is_jmp = DISAS_NORETURN;
1256 static void gen_exception_insn(DisasContext *s, int offset, int excp,
1257 int syn, uint32_t target_el)
1259 gen_set_condexec(s);
1260 gen_set_pc_im(s, s->pc - offset);
1261 gen_exception(excp, syn, target_el);
1262 s->base.is_jmp = DISAS_NORETURN;
1265 static void gen_exception_bkpt_insn(DisasContext *s, int offset, uint32_t syn)
1269 gen_set_condexec(s);
1270 gen_set_pc_im(s, s->pc - offset);
1271 tcg_syn = tcg_const_i32(syn);
1272 gen_helper_exception_bkpt_insn(cpu_env, tcg_syn);
1273 tcg_temp_free_i32(tcg_syn);
1274 s->base.is_jmp = DISAS_NORETURN;
1277 /* Force a TB lookup after an instruction that changes the CPU state. */
1278 static inline void gen_lookup_tb(DisasContext *s)
1280 tcg_gen_movi_i32(cpu_R[15], s->pc & ~1);
1281 s->base.is_jmp = DISAS_EXIT;
1284 static inline void gen_hlt(DisasContext *s, int imm)
1286 /* HLT. This has two purposes.
1287 * Architecturally, it is an external halting debug instruction.
1288 * Since QEMU doesn't implement external debug, we treat this as
1289 * it is required for halting debug disabled: it will UNDEF.
1290 * Secondly, "HLT 0x3C" is a T32 semihosting trap instruction,
1291 * and "HLT 0xF000" is an A32 semihosting syscall. These traps
1292 * must trigger semihosting even for ARMv7 and earlier, where
1293 * HLT was an undefined encoding.
1294 * In system mode, we don't allow userspace access to
1295 * semihosting, to provide some semblance of security
1296 * (and for consistency with our 32-bit semihosting).
1298 if (semihosting_enabled() &&
1299 #ifndef CONFIG_USER_ONLY
1300 s->current_el != 0 &&
1302 (imm == (s->thumb ? 0x3c : 0xf000))) {
1303 gen_exception_internal_insn(s, 0, EXCP_SEMIHOST);
1307 gen_exception_insn(s, s->thumb ? 2 : 4, EXCP_UDEF, syn_uncategorized(),
1308 default_exception_el(s));
1311 static inline void gen_add_data_offset(DisasContext *s, unsigned int insn,
1314 int val, rm, shift, shiftop;
1317 if (!(insn & (1 << 25))) {
1320 if (!(insn & (1 << 23)))
1323 tcg_gen_addi_i32(var, var, val);
1325 /* shift/register */
1327 shift = (insn >> 7) & 0x1f;
1328 shiftop = (insn >> 5) & 3;
1329 offset = load_reg(s, rm);
1330 gen_arm_shift_im(offset, shiftop, shift, 0);
1331 if (!(insn & (1 << 23)))
1332 tcg_gen_sub_i32(var, var, offset);
1334 tcg_gen_add_i32(var, var, offset);
1335 tcg_temp_free_i32(offset);
1339 static inline void gen_add_datah_offset(DisasContext *s, unsigned int insn,
1340 int extra, TCGv_i32 var)
1345 if (insn & (1 << 22)) {
1347 val = (insn & 0xf) | ((insn >> 4) & 0xf0);
1348 if (!(insn & (1 << 23)))
1352 tcg_gen_addi_i32(var, var, val);
1356 tcg_gen_addi_i32(var, var, extra);
1358 offset = load_reg(s, rm);
1359 if (!(insn & (1 << 23)))
1360 tcg_gen_sub_i32(var, var, offset);
1362 tcg_gen_add_i32(var, var, offset);
1363 tcg_temp_free_i32(offset);
1367 static TCGv_ptr get_fpstatus_ptr(int neon)
1369 TCGv_ptr statusptr = tcg_temp_new_ptr();
1372 offset = offsetof(CPUARMState, vfp.standard_fp_status);
1374 offset = offsetof(CPUARMState, vfp.fp_status);
1376 tcg_gen_addi_ptr(statusptr, cpu_env, offset);
1380 #define VFP_OP2(name) \
1381 static inline void gen_vfp_##name(int dp) \
1383 TCGv_ptr fpst = get_fpstatus_ptr(0); \
1385 gen_helper_vfp_##name##d(cpu_F0d, cpu_F0d, cpu_F1d, fpst); \
1387 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, cpu_F1s, fpst); \
1389 tcg_temp_free_ptr(fpst); \
1399 static inline void gen_vfp_F1_mul(int dp)
1401 /* Like gen_vfp_mul() but put result in F1 */
1402 TCGv_ptr fpst = get_fpstatus_ptr(0);
1404 gen_helper_vfp_muld(cpu_F1d, cpu_F0d, cpu_F1d, fpst);
1406 gen_helper_vfp_muls(cpu_F1s, cpu_F0s, cpu_F1s, fpst);
1408 tcg_temp_free_ptr(fpst);
1411 static inline void gen_vfp_F1_neg(int dp)
1413 /* Like gen_vfp_neg() but put result in F1 */
1415 gen_helper_vfp_negd(cpu_F1d, cpu_F0d);
1417 gen_helper_vfp_negs(cpu_F1s, cpu_F0s);
1421 static inline void gen_vfp_abs(int dp)
1424 gen_helper_vfp_absd(cpu_F0d, cpu_F0d);
1426 gen_helper_vfp_abss(cpu_F0s, cpu_F0s);
1429 static inline void gen_vfp_neg(int dp)
1432 gen_helper_vfp_negd(cpu_F0d, cpu_F0d);
1434 gen_helper_vfp_negs(cpu_F0s, cpu_F0s);
1437 static inline void gen_vfp_sqrt(int dp)
1440 gen_helper_vfp_sqrtd(cpu_F0d, cpu_F0d, cpu_env);
1442 gen_helper_vfp_sqrts(cpu_F0s, cpu_F0s, cpu_env);
1445 static inline void gen_vfp_cmp(int dp)
1448 gen_helper_vfp_cmpd(cpu_F0d, cpu_F1d, cpu_env);
1450 gen_helper_vfp_cmps(cpu_F0s, cpu_F1s, cpu_env);
1453 static inline void gen_vfp_cmpe(int dp)
1456 gen_helper_vfp_cmped(cpu_F0d, cpu_F1d, cpu_env);
1458 gen_helper_vfp_cmpes(cpu_F0s, cpu_F1s, cpu_env);
1461 static inline void gen_vfp_F1_ld0(int dp)
1464 tcg_gen_movi_i64(cpu_F1d, 0);
1466 tcg_gen_movi_i32(cpu_F1s, 0);
1469 #define VFP_GEN_ITOF(name) \
1470 static inline void gen_vfp_##name(int dp, int neon) \
1472 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1474 gen_helper_vfp_##name##d(cpu_F0d, cpu_F0s, statusptr); \
1476 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, statusptr); \
1478 tcg_temp_free_ptr(statusptr); \
1485 #define VFP_GEN_FTOI(name) \
1486 static inline void gen_vfp_##name(int dp, int neon) \
1488 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1490 gen_helper_vfp_##name##d(cpu_F0s, cpu_F0d, statusptr); \
1492 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, statusptr); \
1494 tcg_temp_free_ptr(statusptr); \
1503 #define VFP_GEN_FIX(name, round) \
1504 static inline void gen_vfp_##name(int dp, int shift, int neon) \
1506 TCGv_i32 tmp_shift = tcg_const_i32(shift); \
1507 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1509 gen_helper_vfp_##name##d##round(cpu_F0d, cpu_F0d, tmp_shift, \
1512 gen_helper_vfp_##name##s##round(cpu_F0s, cpu_F0s, tmp_shift, \
1515 tcg_temp_free_i32(tmp_shift); \
1516 tcg_temp_free_ptr(statusptr); \
1518 VFP_GEN_FIX(tosh, _round_to_zero)
1519 VFP_GEN_FIX(tosl, _round_to_zero)
1520 VFP_GEN_FIX(touh, _round_to_zero)
1521 VFP_GEN_FIX(toul, _round_to_zero)
1528 static inline void gen_vfp_ld(DisasContext *s, int dp, TCGv_i32 addr)
1531 gen_aa32_ld64(s, cpu_F0d, addr, get_mem_index(s));
1533 gen_aa32_ld32u(s, cpu_F0s, addr, get_mem_index(s));
1537 static inline void gen_vfp_st(DisasContext *s, int dp, TCGv_i32 addr)
1540 gen_aa32_st64(s, cpu_F0d, addr, get_mem_index(s));
1542 gen_aa32_st32(s, cpu_F0s, addr, get_mem_index(s));
1546 static inline long vfp_reg_offset(bool dp, unsigned reg)
1549 return offsetof(CPUARMState, vfp.zregs[reg >> 1].d[reg & 1]);
1551 long ofs = offsetof(CPUARMState, vfp.zregs[reg >> 2].d[(reg >> 1) & 1]);
1553 ofs += offsetof(CPU_DoubleU, l.upper);
1555 ofs += offsetof(CPU_DoubleU, l.lower);
1561 /* Return the offset of a 32-bit piece of a NEON register.
1562 zero is the least significant end of the register. */
1564 neon_reg_offset (int reg, int n)
1568 return vfp_reg_offset(0, sreg);
1571 static TCGv_i32 neon_load_reg(int reg, int pass)
1573 TCGv_i32 tmp = tcg_temp_new_i32();
1574 tcg_gen_ld_i32(tmp, cpu_env, neon_reg_offset(reg, pass));
1578 static void neon_store_reg(int reg, int pass, TCGv_i32 var)
1580 tcg_gen_st_i32(var, cpu_env, neon_reg_offset(reg, pass));
1581 tcg_temp_free_i32(var);
1584 static inline void neon_load_reg64(TCGv_i64 var, int reg)
1586 tcg_gen_ld_i64(var, cpu_env, vfp_reg_offset(1, reg));
1589 static inline void neon_store_reg64(TCGv_i64 var, int reg)
1591 tcg_gen_st_i64(var, cpu_env, vfp_reg_offset(1, reg));
1594 static TCGv_ptr vfp_reg_ptr(bool dp, int reg)
1596 TCGv_ptr ret = tcg_temp_new_ptr();
1597 tcg_gen_addi_ptr(ret, cpu_env, vfp_reg_offset(dp, reg));
1601 #define tcg_gen_ld_f32 tcg_gen_ld_i32
1602 #define tcg_gen_ld_f64 tcg_gen_ld_i64
1603 #define tcg_gen_st_f32 tcg_gen_st_i32
1604 #define tcg_gen_st_f64 tcg_gen_st_i64
1606 static inline void gen_mov_F0_vreg(int dp, int reg)
1609 tcg_gen_ld_f64(cpu_F0d, cpu_env, vfp_reg_offset(dp, reg));
1611 tcg_gen_ld_f32(cpu_F0s, cpu_env, vfp_reg_offset(dp, reg));
1614 static inline void gen_mov_F1_vreg(int dp, int reg)
1617 tcg_gen_ld_f64(cpu_F1d, cpu_env, vfp_reg_offset(dp, reg));
1619 tcg_gen_ld_f32(cpu_F1s, cpu_env, vfp_reg_offset(dp, reg));
1622 static inline void gen_mov_vreg_F0(int dp, int reg)
1625 tcg_gen_st_f64(cpu_F0d, cpu_env, vfp_reg_offset(dp, reg));
1627 tcg_gen_st_f32(cpu_F0s, cpu_env, vfp_reg_offset(dp, reg));
1630 #define ARM_CP_RW_BIT (1 << 20)
1632 static inline void iwmmxt_load_reg(TCGv_i64 var, int reg)
1634 tcg_gen_ld_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1637 static inline void iwmmxt_store_reg(TCGv_i64 var, int reg)
1639 tcg_gen_st_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1642 static inline TCGv_i32 iwmmxt_load_creg(int reg)
1644 TCGv_i32 var = tcg_temp_new_i32();
1645 tcg_gen_ld_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1649 static inline void iwmmxt_store_creg(int reg, TCGv_i32 var)
1651 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1652 tcg_temp_free_i32(var);
1655 static inline void gen_op_iwmmxt_movq_wRn_M0(int rn)
1657 iwmmxt_store_reg(cpu_M0, rn);
1660 static inline void gen_op_iwmmxt_movq_M0_wRn(int rn)
1662 iwmmxt_load_reg(cpu_M0, rn);
1665 static inline void gen_op_iwmmxt_orq_M0_wRn(int rn)
1667 iwmmxt_load_reg(cpu_V1, rn);
1668 tcg_gen_or_i64(cpu_M0, cpu_M0, cpu_V1);
1671 static inline void gen_op_iwmmxt_andq_M0_wRn(int rn)
1673 iwmmxt_load_reg(cpu_V1, rn);
1674 tcg_gen_and_i64(cpu_M0, cpu_M0, cpu_V1);
1677 static inline void gen_op_iwmmxt_xorq_M0_wRn(int rn)
1679 iwmmxt_load_reg(cpu_V1, rn);
1680 tcg_gen_xor_i64(cpu_M0, cpu_M0, cpu_V1);
1683 #define IWMMXT_OP(name) \
1684 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1686 iwmmxt_load_reg(cpu_V1, rn); \
1687 gen_helper_iwmmxt_##name(cpu_M0, cpu_M0, cpu_V1); \
1690 #define IWMMXT_OP_ENV(name) \
1691 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1693 iwmmxt_load_reg(cpu_V1, rn); \
1694 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0, cpu_V1); \
1697 #define IWMMXT_OP_ENV_SIZE(name) \
1698 IWMMXT_OP_ENV(name##b) \
1699 IWMMXT_OP_ENV(name##w) \
1700 IWMMXT_OP_ENV(name##l)
1702 #define IWMMXT_OP_ENV1(name) \
1703 static inline void gen_op_iwmmxt_##name##_M0(void) \
1705 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0); \
1719 IWMMXT_OP_ENV_SIZE(unpackl)
1720 IWMMXT_OP_ENV_SIZE(unpackh)
1722 IWMMXT_OP_ENV1(unpacklub)
1723 IWMMXT_OP_ENV1(unpackluw)
1724 IWMMXT_OP_ENV1(unpacklul)
1725 IWMMXT_OP_ENV1(unpackhub)
1726 IWMMXT_OP_ENV1(unpackhuw)
1727 IWMMXT_OP_ENV1(unpackhul)
1728 IWMMXT_OP_ENV1(unpacklsb)
1729 IWMMXT_OP_ENV1(unpacklsw)
1730 IWMMXT_OP_ENV1(unpacklsl)
1731 IWMMXT_OP_ENV1(unpackhsb)
1732 IWMMXT_OP_ENV1(unpackhsw)
1733 IWMMXT_OP_ENV1(unpackhsl)
1735 IWMMXT_OP_ENV_SIZE(cmpeq)
1736 IWMMXT_OP_ENV_SIZE(cmpgtu)
1737 IWMMXT_OP_ENV_SIZE(cmpgts)
1739 IWMMXT_OP_ENV_SIZE(mins)
1740 IWMMXT_OP_ENV_SIZE(minu)
1741 IWMMXT_OP_ENV_SIZE(maxs)
1742 IWMMXT_OP_ENV_SIZE(maxu)
1744 IWMMXT_OP_ENV_SIZE(subn)
1745 IWMMXT_OP_ENV_SIZE(addn)
1746 IWMMXT_OP_ENV_SIZE(subu)
1747 IWMMXT_OP_ENV_SIZE(addu)
1748 IWMMXT_OP_ENV_SIZE(subs)
1749 IWMMXT_OP_ENV_SIZE(adds)
1751 IWMMXT_OP_ENV(avgb0)
1752 IWMMXT_OP_ENV(avgb1)
1753 IWMMXT_OP_ENV(avgw0)
1754 IWMMXT_OP_ENV(avgw1)
1756 IWMMXT_OP_ENV(packuw)
1757 IWMMXT_OP_ENV(packul)
1758 IWMMXT_OP_ENV(packuq)
1759 IWMMXT_OP_ENV(packsw)
1760 IWMMXT_OP_ENV(packsl)
1761 IWMMXT_OP_ENV(packsq)
1763 static void gen_op_iwmmxt_set_mup(void)
1766 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1767 tcg_gen_ori_i32(tmp, tmp, 2);
1768 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1771 static void gen_op_iwmmxt_set_cup(void)
1774 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1775 tcg_gen_ori_i32(tmp, tmp, 1);
1776 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1779 static void gen_op_iwmmxt_setpsr_nz(void)
1781 TCGv_i32 tmp = tcg_temp_new_i32();
1782 gen_helper_iwmmxt_setpsr_nz(tmp, cpu_M0);
1783 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCASF]);
1786 static inline void gen_op_iwmmxt_addl_M0_wRn(int rn)
1788 iwmmxt_load_reg(cpu_V1, rn);
1789 tcg_gen_ext32u_i64(cpu_V1, cpu_V1);
1790 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1793 static inline int gen_iwmmxt_address(DisasContext *s, uint32_t insn,
1800 rd = (insn >> 16) & 0xf;
1801 tmp = load_reg(s, rd);
1803 offset = (insn & 0xff) << ((insn >> 7) & 2);
1804 if (insn & (1 << 24)) {
1806 if (insn & (1 << 23))
1807 tcg_gen_addi_i32(tmp, tmp, offset);
1809 tcg_gen_addi_i32(tmp, tmp, -offset);
1810 tcg_gen_mov_i32(dest, tmp);
1811 if (insn & (1 << 21))
1812 store_reg(s, rd, tmp);
1814 tcg_temp_free_i32(tmp);
1815 } else if (insn & (1 << 21)) {
1817 tcg_gen_mov_i32(dest, tmp);
1818 if (insn & (1 << 23))
1819 tcg_gen_addi_i32(tmp, tmp, offset);
1821 tcg_gen_addi_i32(tmp, tmp, -offset);
1822 store_reg(s, rd, tmp);
1823 } else if (!(insn & (1 << 23)))
1828 static inline int gen_iwmmxt_shift(uint32_t insn, uint32_t mask, TCGv_i32 dest)
1830 int rd = (insn >> 0) & 0xf;
1833 if (insn & (1 << 8)) {
1834 if (rd < ARM_IWMMXT_wCGR0 || rd > ARM_IWMMXT_wCGR3) {
1837 tmp = iwmmxt_load_creg(rd);
1840 tmp = tcg_temp_new_i32();
1841 iwmmxt_load_reg(cpu_V0, rd);
1842 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
1844 tcg_gen_andi_i32(tmp, tmp, mask);
1845 tcg_gen_mov_i32(dest, tmp);
1846 tcg_temp_free_i32(tmp);
1850 /* Disassemble an iwMMXt instruction. Returns nonzero if an error occurred
1851 (ie. an undefined instruction). */
1852 static int disas_iwmmxt_insn(DisasContext *s, uint32_t insn)
1855 int rdhi, rdlo, rd0, rd1, i;
1857 TCGv_i32 tmp, tmp2, tmp3;
1859 if ((insn & 0x0e000e00) == 0x0c000000) {
1860 if ((insn & 0x0fe00ff0) == 0x0c400000) {
1862 rdlo = (insn >> 12) & 0xf;
1863 rdhi = (insn >> 16) & 0xf;
1864 if (insn & ARM_CP_RW_BIT) { /* TMRRC */
1865 iwmmxt_load_reg(cpu_V0, wrd);
1866 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
1867 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
1868 tcg_gen_extrl_i64_i32(cpu_R[rdhi], cpu_V0);
1869 } else { /* TMCRR */
1870 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
1871 iwmmxt_store_reg(cpu_V0, wrd);
1872 gen_op_iwmmxt_set_mup();
1877 wrd = (insn >> 12) & 0xf;
1878 addr = tcg_temp_new_i32();
1879 if (gen_iwmmxt_address(s, insn, addr)) {
1880 tcg_temp_free_i32(addr);
1883 if (insn & ARM_CP_RW_BIT) {
1884 if ((insn >> 28) == 0xf) { /* WLDRW wCx */
1885 tmp = tcg_temp_new_i32();
1886 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1887 iwmmxt_store_creg(wrd, tmp);
1890 if (insn & (1 << 8)) {
1891 if (insn & (1 << 22)) { /* WLDRD */
1892 gen_aa32_ld64(s, cpu_M0, addr, get_mem_index(s));
1894 } else { /* WLDRW wRd */
1895 tmp = tcg_temp_new_i32();
1896 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1899 tmp = tcg_temp_new_i32();
1900 if (insn & (1 << 22)) { /* WLDRH */
1901 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
1902 } else { /* WLDRB */
1903 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
1907 tcg_gen_extu_i32_i64(cpu_M0, tmp);
1908 tcg_temp_free_i32(tmp);
1910 gen_op_iwmmxt_movq_wRn_M0(wrd);
1913 if ((insn >> 28) == 0xf) { /* WSTRW wCx */
1914 tmp = iwmmxt_load_creg(wrd);
1915 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1917 gen_op_iwmmxt_movq_M0_wRn(wrd);
1918 tmp = tcg_temp_new_i32();
1919 if (insn & (1 << 8)) {
1920 if (insn & (1 << 22)) { /* WSTRD */
1921 gen_aa32_st64(s, cpu_M0, addr, get_mem_index(s));
1922 } else { /* WSTRW wRd */
1923 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1924 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1927 if (insn & (1 << 22)) { /* WSTRH */
1928 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1929 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
1930 } else { /* WSTRB */
1931 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1932 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
1936 tcg_temp_free_i32(tmp);
1938 tcg_temp_free_i32(addr);
1942 if ((insn & 0x0f000000) != 0x0e000000)
1945 switch (((insn >> 12) & 0xf00) | ((insn >> 4) & 0xff)) {
1946 case 0x000: /* WOR */
1947 wrd = (insn >> 12) & 0xf;
1948 rd0 = (insn >> 0) & 0xf;
1949 rd1 = (insn >> 16) & 0xf;
1950 gen_op_iwmmxt_movq_M0_wRn(rd0);
1951 gen_op_iwmmxt_orq_M0_wRn(rd1);
1952 gen_op_iwmmxt_setpsr_nz();
1953 gen_op_iwmmxt_movq_wRn_M0(wrd);
1954 gen_op_iwmmxt_set_mup();
1955 gen_op_iwmmxt_set_cup();
1957 case 0x011: /* TMCR */
1960 rd = (insn >> 12) & 0xf;
1961 wrd = (insn >> 16) & 0xf;
1963 case ARM_IWMMXT_wCID:
1964 case ARM_IWMMXT_wCASF:
1966 case ARM_IWMMXT_wCon:
1967 gen_op_iwmmxt_set_cup();
1969 case ARM_IWMMXT_wCSSF:
1970 tmp = iwmmxt_load_creg(wrd);
1971 tmp2 = load_reg(s, rd);
1972 tcg_gen_andc_i32(tmp, tmp, tmp2);
1973 tcg_temp_free_i32(tmp2);
1974 iwmmxt_store_creg(wrd, tmp);
1976 case ARM_IWMMXT_wCGR0:
1977 case ARM_IWMMXT_wCGR1:
1978 case ARM_IWMMXT_wCGR2:
1979 case ARM_IWMMXT_wCGR3:
1980 gen_op_iwmmxt_set_cup();
1981 tmp = load_reg(s, rd);
1982 iwmmxt_store_creg(wrd, tmp);
1988 case 0x100: /* WXOR */
1989 wrd = (insn >> 12) & 0xf;
1990 rd0 = (insn >> 0) & 0xf;
1991 rd1 = (insn >> 16) & 0xf;
1992 gen_op_iwmmxt_movq_M0_wRn(rd0);
1993 gen_op_iwmmxt_xorq_M0_wRn(rd1);
1994 gen_op_iwmmxt_setpsr_nz();
1995 gen_op_iwmmxt_movq_wRn_M0(wrd);
1996 gen_op_iwmmxt_set_mup();
1997 gen_op_iwmmxt_set_cup();
1999 case 0x111: /* TMRC */
2002 rd = (insn >> 12) & 0xf;
2003 wrd = (insn >> 16) & 0xf;
2004 tmp = iwmmxt_load_creg(wrd);
2005 store_reg(s, rd, tmp);
2007 case 0x300: /* WANDN */
2008 wrd = (insn >> 12) & 0xf;
2009 rd0 = (insn >> 0) & 0xf;
2010 rd1 = (insn >> 16) & 0xf;
2011 gen_op_iwmmxt_movq_M0_wRn(rd0);
2012 tcg_gen_neg_i64(cpu_M0, cpu_M0);
2013 gen_op_iwmmxt_andq_M0_wRn(rd1);
2014 gen_op_iwmmxt_setpsr_nz();
2015 gen_op_iwmmxt_movq_wRn_M0(wrd);
2016 gen_op_iwmmxt_set_mup();
2017 gen_op_iwmmxt_set_cup();
2019 case 0x200: /* WAND */
2020 wrd = (insn >> 12) & 0xf;
2021 rd0 = (insn >> 0) & 0xf;
2022 rd1 = (insn >> 16) & 0xf;
2023 gen_op_iwmmxt_movq_M0_wRn(rd0);
2024 gen_op_iwmmxt_andq_M0_wRn(rd1);
2025 gen_op_iwmmxt_setpsr_nz();
2026 gen_op_iwmmxt_movq_wRn_M0(wrd);
2027 gen_op_iwmmxt_set_mup();
2028 gen_op_iwmmxt_set_cup();
2030 case 0x810: case 0xa10: /* WMADD */
2031 wrd = (insn >> 12) & 0xf;
2032 rd0 = (insn >> 0) & 0xf;
2033 rd1 = (insn >> 16) & 0xf;
2034 gen_op_iwmmxt_movq_M0_wRn(rd0);
2035 if (insn & (1 << 21))
2036 gen_op_iwmmxt_maddsq_M0_wRn(rd1);
2038 gen_op_iwmmxt_madduq_M0_wRn(rd1);
2039 gen_op_iwmmxt_movq_wRn_M0(wrd);
2040 gen_op_iwmmxt_set_mup();
2042 case 0x10e: case 0x50e: case 0x90e: case 0xd0e: /* WUNPCKIL */
2043 wrd = (insn >> 12) & 0xf;
2044 rd0 = (insn >> 16) & 0xf;
2045 rd1 = (insn >> 0) & 0xf;
2046 gen_op_iwmmxt_movq_M0_wRn(rd0);
2047 switch ((insn >> 22) & 3) {
2049 gen_op_iwmmxt_unpacklb_M0_wRn(rd1);
2052 gen_op_iwmmxt_unpacklw_M0_wRn(rd1);
2055 gen_op_iwmmxt_unpackll_M0_wRn(rd1);
2060 gen_op_iwmmxt_movq_wRn_M0(wrd);
2061 gen_op_iwmmxt_set_mup();
2062 gen_op_iwmmxt_set_cup();
2064 case 0x10c: case 0x50c: case 0x90c: case 0xd0c: /* WUNPCKIH */
2065 wrd = (insn >> 12) & 0xf;
2066 rd0 = (insn >> 16) & 0xf;
2067 rd1 = (insn >> 0) & 0xf;
2068 gen_op_iwmmxt_movq_M0_wRn(rd0);
2069 switch ((insn >> 22) & 3) {
2071 gen_op_iwmmxt_unpackhb_M0_wRn(rd1);
2074 gen_op_iwmmxt_unpackhw_M0_wRn(rd1);
2077 gen_op_iwmmxt_unpackhl_M0_wRn(rd1);
2082 gen_op_iwmmxt_movq_wRn_M0(wrd);
2083 gen_op_iwmmxt_set_mup();
2084 gen_op_iwmmxt_set_cup();
2086 case 0x012: case 0x112: case 0x412: case 0x512: /* WSAD */
2087 wrd = (insn >> 12) & 0xf;
2088 rd0 = (insn >> 16) & 0xf;
2089 rd1 = (insn >> 0) & 0xf;
2090 gen_op_iwmmxt_movq_M0_wRn(rd0);
2091 if (insn & (1 << 22))
2092 gen_op_iwmmxt_sadw_M0_wRn(rd1);
2094 gen_op_iwmmxt_sadb_M0_wRn(rd1);
2095 if (!(insn & (1 << 20)))
2096 gen_op_iwmmxt_addl_M0_wRn(wrd);
2097 gen_op_iwmmxt_movq_wRn_M0(wrd);
2098 gen_op_iwmmxt_set_mup();
2100 case 0x010: case 0x110: case 0x210: case 0x310: /* WMUL */
2101 wrd = (insn >> 12) & 0xf;
2102 rd0 = (insn >> 16) & 0xf;
2103 rd1 = (insn >> 0) & 0xf;
2104 gen_op_iwmmxt_movq_M0_wRn(rd0);
2105 if (insn & (1 << 21)) {
2106 if (insn & (1 << 20))
2107 gen_op_iwmmxt_mulshw_M0_wRn(rd1);
2109 gen_op_iwmmxt_mulslw_M0_wRn(rd1);
2111 if (insn & (1 << 20))
2112 gen_op_iwmmxt_muluhw_M0_wRn(rd1);
2114 gen_op_iwmmxt_mululw_M0_wRn(rd1);
2116 gen_op_iwmmxt_movq_wRn_M0(wrd);
2117 gen_op_iwmmxt_set_mup();
2119 case 0x410: case 0x510: case 0x610: case 0x710: /* WMAC */
2120 wrd = (insn >> 12) & 0xf;
2121 rd0 = (insn >> 16) & 0xf;
2122 rd1 = (insn >> 0) & 0xf;
2123 gen_op_iwmmxt_movq_M0_wRn(rd0);
2124 if (insn & (1 << 21))
2125 gen_op_iwmmxt_macsw_M0_wRn(rd1);
2127 gen_op_iwmmxt_macuw_M0_wRn(rd1);
2128 if (!(insn & (1 << 20))) {
2129 iwmmxt_load_reg(cpu_V1, wrd);
2130 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
2132 gen_op_iwmmxt_movq_wRn_M0(wrd);
2133 gen_op_iwmmxt_set_mup();
2135 case 0x006: case 0x406: case 0x806: case 0xc06: /* WCMPEQ */
2136 wrd = (insn >> 12) & 0xf;
2137 rd0 = (insn >> 16) & 0xf;
2138 rd1 = (insn >> 0) & 0xf;
2139 gen_op_iwmmxt_movq_M0_wRn(rd0);
2140 switch ((insn >> 22) & 3) {
2142 gen_op_iwmmxt_cmpeqb_M0_wRn(rd1);
2145 gen_op_iwmmxt_cmpeqw_M0_wRn(rd1);
2148 gen_op_iwmmxt_cmpeql_M0_wRn(rd1);
2153 gen_op_iwmmxt_movq_wRn_M0(wrd);
2154 gen_op_iwmmxt_set_mup();
2155 gen_op_iwmmxt_set_cup();
2157 case 0x800: case 0x900: case 0xc00: case 0xd00: /* WAVG2 */
2158 wrd = (insn >> 12) & 0xf;
2159 rd0 = (insn >> 16) & 0xf;
2160 rd1 = (insn >> 0) & 0xf;
2161 gen_op_iwmmxt_movq_M0_wRn(rd0);
2162 if (insn & (1 << 22)) {
2163 if (insn & (1 << 20))
2164 gen_op_iwmmxt_avgw1_M0_wRn(rd1);
2166 gen_op_iwmmxt_avgw0_M0_wRn(rd1);
2168 if (insn & (1 << 20))
2169 gen_op_iwmmxt_avgb1_M0_wRn(rd1);
2171 gen_op_iwmmxt_avgb0_M0_wRn(rd1);
2173 gen_op_iwmmxt_movq_wRn_M0(wrd);
2174 gen_op_iwmmxt_set_mup();
2175 gen_op_iwmmxt_set_cup();
2177 case 0x802: case 0x902: case 0xa02: case 0xb02: /* WALIGNR */
2178 wrd = (insn >> 12) & 0xf;
2179 rd0 = (insn >> 16) & 0xf;
2180 rd1 = (insn >> 0) & 0xf;
2181 gen_op_iwmmxt_movq_M0_wRn(rd0);
2182 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCGR0 + ((insn >> 20) & 3));
2183 tcg_gen_andi_i32(tmp, tmp, 7);
2184 iwmmxt_load_reg(cpu_V1, rd1);
2185 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
2186 tcg_temp_free_i32(tmp);
2187 gen_op_iwmmxt_movq_wRn_M0(wrd);
2188 gen_op_iwmmxt_set_mup();
2190 case 0x601: case 0x605: case 0x609: case 0x60d: /* TINSR */
2191 if (((insn >> 6) & 3) == 3)
2193 rd = (insn >> 12) & 0xf;
2194 wrd = (insn >> 16) & 0xf;
2195 tmp = load_reg(s, rd);
2196 gen_op_iwmmxt_movq_M0_wRn(wrd);
2197 switch ((insn >> 6) & 3) {
2199 tmp2 = tcg_const_i32(0xff);
2200 tmp3 = tcg_const_i32((insn & 7) << 3);
2203 tmp2 = tcg_const_i32(0xffff);
2204 tmp3 = tcg_const_i32((insn & 3) << 4);
2207 tmp2 = tcg_const_i32(0xffffffff);
2208 tmp3 = tcg_const_i32((insn & 1) << 5);
2214 gen_helper_iwmmxt_insr(cpu_M0, cpu_M0, tmp, tmp2, tmp3);
2215 tcg_temp_free_i32(tmp3);
2216 tcg_temp_free_i32(tmp2);
2217 tcg_temp_free_i32(tmp);
2218 gen_op_iwmmxt_movq_wRn_M0(wrd);
2219 gen_op_iwmmxt_set_mup();
2221 case 0x107: case 0x507: case 0x907: case 0xd07: /* TEXTRM */
2222 rd = (insn >> 12) & 0xf;
2223 wrd = (insn >> 16) & 0xf;
2224 if (rd == 15 || ((insn >> 22) & 3) == 3)
2226 gen_op_iwmmxt_movq_M0_wRn(wrd);
2227 tmp = tcg_temp_new_i32();
2228 switch ((insn >> 22) & 3) {
2230 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 7) << 3);
2231 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
2233 tcg_gen_ext8s_i32(tmp, tmp);
2235 tcg_gen_andi_i32(tmp, tmp, 0xff);
2239 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 3) << 4);
2240 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
2242 tcg_gen_ext16s_i32(tmp, tmp);
2244 tcg_gen_andi_i32(tmp, tmp, 0xffff);
2248 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 1) << 5);
2249 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
2252 store_reg(s, rd, tmp);
2254 case 0x117: case 0x517: case 0x917: case 0xd17: /* TEXTRC */
2255 if ((insn & 0x000ff008) != 0x0003f000 || ((insn >> 22) & 3) == 3)
2257 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
2258 switch ((insn >> 22) & 3) {
2260 tcg_gen_shri_i32(tmp, tmp, ((insn & 7) << 2) + 0);
2263 tcg_gen_shri_i32(tmp, tmp, ((insn & 3) << 3) + 4);
2266 tcg_gen_shri_i32(tmp, tmp, ((insn & 1) << 4) + 12);
2269 tcg_gen_shli_i32(tmp, tmp, 28);
2271 tcg_temp_free_i32(tmp);
2273 case 0x401: case 0x405: case 0x409: case 0x40d: /* TBCST */
2274 if (((insn >> 6) & 3) == 3)
2276 rd = (insn >> 12) & 0xf;
2277 wrd = (insn >> 16) & 0xf;
2278 tmp = load_reg(s, rd);
2279 switch ((insn >> 6) & 3) {
2281 gen_helper_iwmmxt_bcstb(cpu_M0, tmp);
2284 gen_helper_iwmmxt_bcstw(cpu_M0, tmp);
2287 gen_helper_iwmmxt_bcstl(cpu_M0, tmp);
2290 tcg_temp_free_i32(tmp);
2291 gen_op_iwmmxt_movq_wRn_M0(wrd);
2292 gen_op_iwmmxt_set_mup();
2294 case 0x113: case 0x513: case 0x913: case 0xd13: /* TANDC */
2295 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
2297 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
2298 tmp2 = tcg_temp_new_i32();
2299 tcg_gen_mov_i32(tmp2, tmp);
2300 switch ((insn >> 22) & 3) {
2302 for (i = 0; i < 7; i ++) {
2303 tcg_gen_shli_i32(tmp2, tmp2, 4);
2304 tcg_gen_and_i32(tmp, tmp, tmp2);
2308 for (i = 0; i < 3; i ++) {
2309 tcg_gen_shli_i32(tmp2, tmp2, 8);
2310 tcg_gen_and_i32(tmp, tmp, tmp2);
2314 tcg_gen_shli_i32(tmp2, tmp2, 16);
2315 tcg_gen_and_i32(tmp, tmp, tmp2);
2319 tcg_temp_free_i32(tmp2);
2320 tcg_temp_free_i32(tmp);
2322 case 0x01c: case 0x41c: case 0x81c: case 0xc1c: /* WACC */
2323 wrd = (insn >> 12) & 0xf;
2324 rd0 = (insn >> 16) & 0xf;
2325 gen_op_iwmmxt_movq_M0_wRn(rd0);
2326 switch ((insn >> 22) & 3) {
2328 gen_helper_iwmmxt_addcb(cpu_M0, cpu_M0);
2331 gen_helper_iwmmxt_addcw(cpu_M0, cpu_M0);
2334 gen_helper_iwmmxt_addcl(cpu_M0, cpu_M0);
2339 gen_op_iwmmxt_movq_wRn_M0(wrd);
2340 gen_op_iwmmxt_set_mup();
2342 case 0x115: case 0x515: case 0x915: case 0xd15: /* TORC */
2343 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
2345 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
2346 tmp2 = tcg_temp_new_i32();
2347 tcg_gen_mov_i32(tmp2, tmp);
2348 switch ((insn >> 22) & 3) {
2350 for (i = 0; i < 7; i ++) {
2351 tcg_gen_shli_i32(tmp2, tmp2, 4);
2352 tcg_gen_or_i32(tmp, tmp, tmp2);
2356 for (i = 0; i < 3; i ++) {
2357 tcg_gen_shli_i32(tmp2, tmp2, 8);
2358 tcg_gen_or_i32(tmp, tmp, tmp2);
2362 tcg_gen_shli_i32(tmp2, tmp2, 16);
2363 tcg_gen_or_i32(tmp, tmp, tmp2);
2367 tcg_temp_free_i32(tmp2);
2368 tcg_temp_free_i32(tmp);
2370 case 0x103: case 0x503: case 0x903: case 0xd03: /* TMOVMSK */
2371 rd = (insn >> 12) & 0xf;
2372 rd0 = (insn >> 16) & 0xf;
2373 if ((insn & 0xf) != 0 || ((insn >> 22) & 3) == 3)
2375 gen_op_iwmmxt_movq_M0_wRn(rd0);
2376 tmp = tcg_temp_new_i32();
2377 switch ((insn >> 22) & 3) {
2379 gen_helper_iwmmxt_msbb(tmp, cpu_M0);
2382 gen_helper_iwmmxt_msbw(tmp, cpu_M0);
2385 gen_helper_iwmmxt_msbl(tmp, cpu_M0);
2388 store_reg(s, rd, tmp);
2390 case 0x106: case 0x306: case 0x506: case 0x706: /* WCMPGT */
2391 case 0x906: case 0xb06: case 0xd06: case 0xf06:
2392 wrd = (insn >> 12) & 0xf;
2393 rd0 = (insn >> 16) & 0xf;
2394 rd1 = (insn >> 0) & 0xf;
2395 gen_op_iwmmxt_movq_M0_wRn(rd0);
2396 switch ((insn >> 22) & 3) {
2398 if (insn & (1 << 21))
2399 gen_op_iwmmxt_cmpgtsb_M0_wRn(rd1);
2401 gen_op_iwmmxt_cmpgtub_M0_wRn(rd1);
2404 if (insn & (1 << 21))
2405 gen_op_iwmmxt_cmpgtsw_M0_wRn(rd1);
2407 gen_op_iwmmxt_cmpgtuw_M0_wRn(rd1);
2410 if (insn & (1 << 21))
2411 gen_op_iwmmxt_cmpgtsl_M0_wRn(rd1);
2413 gen_op_iwmmxt_cmpgtul_M0_wRn(rd1);
2418 gen_op_iwmmxt_movq_wRn_M0(wrd);
2419 gen_op_iwmmxt_set_mup();
2420 gen_op_iwmmxt_set_cup();
2422 case 0x00e: case 0x20e: case 0x40e: case 0x60e: /* WUNPCKEL */
2423 case 0x80e: case 0xa0e: case 0xc0e: case 0xe0e:
2424 wrd = (insn >> 12) & 0xf;
2425 rd0 = (insn >> 16) & 0xf;
2426 gen_op_iwmmxt_movq_M0_wRn(rd0);
2427 switch ((insn >> 22) & 3) {
2429 if (insn & (1 << 21))
2430 gen_op_iwmmxt_unpacklsb_M0();
2432 gen_op_iwmmxt_unpacklub_M0();
2435 if (insn & (1 << 21))
2436 gen_op_iwmmxt_unpacklsw_M0();
2438 gen_op_iwmmxt_unpackluw_M0();
2441 if (insn & (1 << 21))
2442 gen_op_iwmmxt_unpacklsl_M0();
2444 gen_op_iwmmxt_unpacklul_M0();
2449 gen_op_iwmmxt_movq_wRn_M0(wrd);
2450 gen_op_iwmmxt_set_mup();
2451 gen_op_iwmmxt_set_cup();
2453 case 0x00c: case 0x20c: case 0x40c: case 0x60c: /* WUNPCKEH */
2454 case 0x80c: case 0xa0c: case 0xc0c: case 0xe0c:
2455 wrd = (insn >> 12) & 0xf;
2456 rd0 = (insn >> 16) & 0xf;
2457 gen_op_iwmmxt_movq_M0_wRn(rd0);
2458 switch ((insn >> 22) & 3) {
2460 if (insn & (1 << 21))
2461 gen_op_iwmmxt_unpackhsb_M0();
2463 gen_op_iwmmxt_unpackhub_M0();
2466 if (insn & (1 << 21))
2467 gen_op_iwmmxt_unpackhsw_M0();
2469 gen_op_iwmmxt_unpackhuw_M0();
2472 if (insn & (1 << 21))
2473 gen_op_iwmmxt_unpackhsl_M0();
2475 gen_op_iwmmxt_unpackhul_M0();
2480 gen_op_iwmmxt_movq_wRn_M0(wrd);
2481 gen_op_iwmmxt_set_mup();
2482 gen_op_iwmmxt_set_cup();
2484 case 0x204: case 0x604: case 0xa04: case 0xe04: /* WSRL */
2485 case 0x214: case 0x614: case 0xa14: case 0xe14:
2486 if (((insn >> 22) & 3) == 0)
2488 wrd = (insn >> 12) & 0xf;
2489 rd0 = (insn >> 16) & 0xf;
2490 gen_op_iwmmxt_movq_M0_wRn(rd0);
2491 tmp = tcg_temp_new_i32();
2492 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2493 tcg_temp_free_i32(tmp);
2496 switch ((insn >> 22) & 3) {
2498 gen_helper_iwmmxt_srlw(cpu_M0, cpu_env, cpu_M0, tmp);
2501 gen_helper_iwmmxt_srll(cpu_M0, cpu_env, cpu_M0, tmp);
2504 gen_helper_iwmmxt_srlq(cpu_M0, cpu_env, cpu_M0, tmp);
2507 tcg_temp_free_i32(tmp);
2508 gen_op_iwmmxt_movq_wRn_M0(wrd);
2509 gen_op_iwmmxt_set_mup();
2510 gen_op_iwmmxt_set_cup();
2512 case 0x004: case 0x404: case 0x804: case 0xc04: /* WSRA */
2513 case 0x014: case 0x414: case 0x814: case 0xc14:
2514 if (((insn >> 22) & 3) == 0)
2516 wrd = (insn >> 12) & 0xf;
2517 rd0 = (insn >> 16) & 0xf;
2518 gen_op_iwmmxt_movq_M0_wRn(rd0);
2519 tmp = tcg_temp_new_i32();
2520 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2521 tcg_temp_free_i32(tmp);
2524 switch ((insn >> 22) & 3) {
2526 gen_helper_iwmmxt_sraw(cpu_M0, cpu_env, cpu_M0, tmp);
2529 gen_helper_iwmmxt_sral(cpu_M0, cpu_env, cpu_M0, tmp);
2532 gen_helper_iwmmxt_sraq(cpu_M0, cpu_env, cpu_M0, tmp);
2535 tcg_temp_free_i32(tmp);
2536 gen_op_iwmmxt_movq_wRn_M0(wrd);
2537 gen_op_iwmmxt_set_mup();
2538 gen_op_iwmmxt_set_cup();
2540 case 0x104: case 0x504: case 0x904: case 0xd04: /* WSLL */
2541 case 0x114: case 0x514: case 0x914: case 0xd14:
2542 if (((insn >> 22) & 3) == 0)
2544 wrd = (insn >> 12) & 0xf;
2545 rd0 = (insn >> 16) & 0xf;
2546 gen_op_iwmmxt_movq_M0_wRn(rd0);
2547 tmp = tcg_temp_new_i32();
2548 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2549 tcg_temp_free_i32(tmp);
2552 switch ((insn >> 22) & 3) {
2554 gen_helper_iwmmxt_sllw(cpu_M0, cpu_env, cpu_M0, tmp);
2557 gen_helper_iwmmxt_slll(cpu_M0, cpu_env, cpu_M0, tmp);
2560 gen_helper_iwmmxt_sllq(cpu_M0, cpu_env, cpu_M0, tmp);
2563 tcg_temp_free_i32(tmp);
2564 gen_op_iwmmxt_movq_wRn_M0(wrd);
2565 gen_op_iwmmxt_set_mup();
2566 gen_op_iwmmxt_set_cup();
2568 case 0x304: case 0x704: case 0xb04: case 0xf04: /* WROR */
2569 case 0x314: case 0x714: case 0xb14: case 0xf14:
2570 if (((insn >> 22) & 3) == 0)
2572 wrd = (insn >> 12) & 0xf;
2573 rd0 = (insn >> 16) & 0xf;
2574 gen_op_iwmmxt_movq_M0_wRn(rd0);
2575 tmp = tcg_temp_new_i32();
2576 switch ((insn >> 22) & 3) {
2578 if (gen_iwmmxt_shift(insn, 0xf, tmp)) {
2579 tcg_temp_free_i32(tmp);
2582 gen_helper_iwmmxt_rorw(cpu_M0, cpu_env, cpu_M0, tmp);
2585 if (gen_iwmmxt_shift(insn, 0x1f, tmp)) {
2586 tcg_temp_free_i32(tmp);
2589 gen_helper_iwmmxt_rorl(cpu_M0, cpu_env, cpu_M0, tmp);
2592 if (gen_iwmmxt_shift(insn, 0x3f, tmp)) {
2593 tcg_temp_free_i32(tmp);
2596 gen_helper_iwmmxt_rorq(cpu_M0, cpu_env, cpu_M0, tmp);
2599 tcg_temp_free_i32(tmp);
2600 gen_op_iwmmxt_movq_wRn_M0(wrd);
2601 gen_op_iwmmxt_set_mup();
2602 gen_op_iwmmxt_set_cup();
2604 case 0x116: case 0x316: case 0x516: case 0x716: /* WMIN */
2605 case 0x916: case 0xb16: case 0xd16: case 0xf16:
2606 wrd = (insn >> 12) & 0xf;
2607 rd0 = (insn >> 16) & 0xf;
2608 rd1 = (insn >> 0) & 0xf;
2609 gen_op_iwmmxt_movq_M0_wRn(rd0);
2610 switch ((insn >> 22) & 3) {
2612 if (insn & (1 << 21))
2613 gen_op_iwmmxt_minsb_M0_wRn(rd1);
2615 gen_op_iwmmxt_minub_M0_wRn(rd1);
2618 if (insn & (1 << 21))
2619 gen_op_iwmmxt_minsw_M0_wRn(rd1);
2621 gen_op_iwmmxt_minuw_M0_wRn(rd1);
2624 if (insn & (1 << 21))
2625 gen_op_iwmmxt_minsl_M0_wRn(rd1);
2627 gen_op_iwmmxt_minul_M0_wRn(rd1);
2632 gen_op_iwmmxt_movq_wRn_M0(wrd);
2633 gen_op_iwmmxt_set_mup();
2635 case 0x016: case 0x216: case 0x416: case 0x616: /* WMAX */
2636 case 0x816: case 0xa16: case 0xc16: case 0xe16:
2637 wrd = (insn >> 12) & 0xf;
2638 rd0 = (insn >> 16) & 0xf;
2639 rd1 = (insn >> 0) & 0xf;
2640 gen_op_iwmmxt_movq_M0_wRn(rd0);
2641 switch ((insn >> 22) & 3) {
2643 if (insn & (1 << 21))
2644 gen_op_iwmmxt_maxsb_M0_wRn(rd1);
2646 gen_op_iwmmxt_maxub_M0_wRn(rd1);
2649 if (insn & (1 << 21))
2650 gen_op_iwmmxt_maxsw_M0_wRn(rd1);
2652 gen_op_iwmmxt_maxuw_M0_wRn(rd1);
2655 if (insn & (1 << 21))
2656 gen_op_iwmmxt_maxsl_M0_wRn(rd1);
2658 gen_op_iwmmxt_maxul_M0_wRn(rd1);
2663 gen_op_iwmmxt_movq_wRn_M0(wrd);
2664 gen_op_iwmmxt_set_mup();
2666 case 0x002: case 0x102: case 0x202: case 0x302: /* WALIGNI */
2667 case 0x402: case 0x502: case 0x602: case 0x702:
2668 wrd = (insn >> 12) & 0xf;
2669 rd0 = (insn >> 16) & 0xf;
2670 rd1 = (insn >> 0) & 0xf;
2671 gen_op_iwmmxt_movq_M0_wRn(rd0);
2672 tmp = tcg_const_i32((insn >> 20) & 3);
2673 iwmmxt_load_reg(cpu_V1, rd1);
2674 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
2675 tcg_temp_free_i32(tmp);
2676 gen_op_iwmmxt_movq_wRn_M0(wrd);
2677 gen_op_iwmmxt_set_mup();
2679 case 0x01a: case 0x11a: case 0x21a: case 0x31a: /* WSUB */
2680 case 0x41a: case 0x51a: case 0x61a: case 0x71a:
2681 case 0x81a: case 0x91a: case 0xa1a: case 0xb1a:
2682 case 0xc1a: case 0xd1a: case 0xe1a: case 0xf1a:
2683 wrd = (insn >> 12) & 0xf;
2684 rd0 = (insn >> 16) & 0xf;
2685 rd1 = (insn >> 0) & 0xf;
2686 gen_op_iwmmxt_movq_M0_wRn(rd0);
2687 switch ((insn >> 20) & 0xf) {
2689 gen_op_iwmmxt_subnb_M0_wRn(rd1);
2692 gen_op_iwmmxt_subub_M0_wRn(rd1);
2695 gen_op_iwmmxt_subsb_M0_wRn(rd1);
2698 gen_op_iwmmxt_subnw_M0_wRn(rd1);
2701 gen_op_iwmmxt_subuw_M0_wRn(rd1);
2704 gen_op_iwmmxt_subsw_M0_wRn(rd1);
2707 gen_op_iwmmxt_subnl_M0_wRn(rd1);
2710 gen_op_iwmmxt_subul_M0_wRn(rd1);
2713 gen_op_iwmmxt_subsl_M0_wRn(rd1);
2718 gen_op_iwmmxt_movq_wRn_M0(wrd);
2719 gen_op_iwmmxt_set_mup();
2720 gen_op_iwmmxt_set_cup();
2722 case 0x01e: case 0x11e: case 0x21e: case 0x31e: /* WSHUFH */
2723 case 0x41e: case 0x51e: case 0x61e: case 0x71e:
2724 case 0x81e: case 0x91e: case 0xa1e: case 0xb1e:
2725 case 0xc1e: case 0xd1e: case 0xe1e: case 0xf1e:
2726 wrd = (insn >> 12) & 0xf;
2727 rd0 = (insn >> 16) & 0xf;
2728 gen_op_iwmmxt_movq_M0_wRn(rd0);
2729 tmp = tcg_const_i32(((insn >> 16) & 0xf0) | (insn & 0x0f));
2730 gen_helper_iwmmxt_shufh(cpu_M0, cpu_env, cpu_M0, tmp);
2731 tcg_temp_free_i32(tmp);
2732 gen_op_iwmmxt_movq_wRn_M0(wrd);
2733 gen_op_iwmmxt_set_mup();
2734 gen_op_iwmmxt_set_cup();
2736 case 0x018: case 0x118: case 0x218: case 0x318: /* WADD */
2737 case 0x418: case 0x518: case 0x618: case 0x718:
2738 case 0x818: case 0x918: case 0xa18: case 0xb18:
2739 case 0xc18: case 0xd18: case 0xe18: case 0xf18:
2740 wrd = (insn >> 12) & 0xf;
2741 rd0 = (insn >> 16) & 0xf;
2742 rd1 = (insn >> 0) & 0xf;
2743 gen_op_iwmmxt_movq_M0_wRn(rd0);
2744 switch ((insn >> 20) & 0xf) {
2746 gen_op_iwmmxt_addnb_M0_wRn(rd1);
2749 gen_op_iwmmxt_addub_M0_wRn(rd1);
2752 gen_op_iwmmxt_addsb_M0_wRn(rd1);
2755 gen_op_iwmmxt_addnw_M0_wRn(rd1);
2758 gen_op_iwmmxt_adduw_M0_wRn(rd1);
2761 gen_op_iwmmxt_addsw_M0_wRn(rd1);
2764 gen_op_iwmmxt_addnl_M0_wRn(rd1);
2767 gen_op_iwmmxt_addul_M0_wRn(rd1);
2770 gen_op_iwmmxt_addsl_M0_wRn(rd1);
2775 gen_op_iwmmxt_movq_wRn_M0(wrd);
2776 gen_op_iwmmxt_set_mup();
2777 gen_op_iwmmxt_set_cup();
2779 case 0x008: case 0x108: case 0x208: case 0x308: /* WPACK */
2780 case 0x408: case 0x508: case 0x608: case 0x708:
2781 case 0x808: case 0x908: case 0xa08: case 0xb08:
2782 case 0xc08: case 0xd08: case 0xe08: case 0xf08:
2783 if (!(insn & (1 << 20)) || ((insn >> 22) & 3) == 0)
2785 wrd = (insn >> 12) & 0xf;
2786 rd0 = (insn >> 16) & 0xf;
2787 rd1 = (insn >> 0) & 0xf;
2788 gen_op_iwmmxt_movq_M0_wRn(rd0);
2789 switch ((insn >> 22) & 3) {
2791 if (insn & (1 << 21))
2792 gen_op_iwmmxt_packsw_M0_wRn(rd1);
2794 gen_op_iwmmxt_packuw_M0_wRn(rd1);
2797 if (insn & (1 << 21))
2798 gen_op_iwmmxt_packsl_M0_wRn(rd1);
2800 gen_op_iwmmxt_packul_M0_wRn(rd1);
2803 if (insn & (1 << 21))
2804 gen_op_iwmmxt_packsq_M0_wRn(rd1);
2806 gen_op_iwmmxt_packuq_M0_wRn(rd1);
2809 gen_op_iwmmxt_movq_wRn_M0(wrd);
2810 gen_op_iwmmxt_set_mup();
2811 gen_op_iwmmxt_set_cup();
2813 case 0x201: case 0x203: case 0x205: case 0x207:
2814 case 0x209: case 0x20b: case 0x20d: case 0x20f:
2815 case 0x211: case 0x213: case 0x215: case 0x217:
2816 case 0x219: case 0x21b: case 0x21d: case 0x21f:
2817 wrd = (insn >> 5) & 0xf;
2818 rd0 = (insn >> 12) & 0xf;
2819 rd1 = (insn >> 0) & 0xf;
2820 if (rd0 == 0xf || rd1 == 0xf)
2822 gen_op_iwmmxt_movq_M0_wRn(wrd);
2823 tmp = load_reg(s, rd0);
2824 tmp2 = load_reg(s, rd1);
2825 switch ((insn >> 16) & 0xf) {
2826 case 0x0: /* TMIA */
2827 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2829 case 0x8: /* TMIAPH */
2830 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2832 case 0xc: case 0xd: case 0xe: case 0xf: /* TMIAxy */
2833 if (insn & (1 << 16))
2834 tcg_gen_shri_i32(tmp, tmp, 16);
2835 if (insn & (1 << 17))
2836 tcg_gen_shri_i32(tmp2, tmp2, 16);
2837 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2840 tcg_temp_free_i32(tmp2);
2841 tcg_temp_free_i32(tmp);
2844 tcg_temp_free_i32(tmp2);
2845 tcg_temp_free_i32(tmp);
2846 gen_op_iwmmxt_movq_wRn_M0(wrd);
2847 gen_op_iwmmxt_set_mup();
2856 /* Disassemble an XScale DSP instruction. Returns nonzero if an error occurred
2857 (ie. an undefined instruction). */
2858 static int disas_dsp_insn(DisasContext *s, uint32_t insn)
2860 int acc, rd0, rd1, rdhi, rdlo;
2863 if ((insn & 0x0ff00f10) == 0x0e200010) {
2864 /* Multiply with Internal Accumulate Format */
2865 rd0 = (insn >> 12) & 0xf;
2867 acc = (insn >> 5) & 7;
2872 tmp = load_reg(s, rd0);
2873 tmp2 = load_reg(s, rd1);
2874 switch ((insn >> 16) & 0xf) {
2876 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2878 case 0x8: /* MIAPH */
2879 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2881 case 0xc: /* MIABB */
2882 case 0xd: /* MIABT */
2883 case 0xe: /* MIATB */
2884 case 0xf: /* MIATT */
2885 if (insn & (1 << 16))
2886 tcg_gen_shri_i32(tmp, tmp, 16);
2887 if (insn & (1 << 17))
2888 tcg_gen_shri_i32(tmp2, tmp2, 16);
2889 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2894 tcg_temp_free_i32(tmp2);
2895 tcg_temp_free_i32(tmp);
2897 gen_op_iwmmxt_movq_wRn_M0(acc);
2901 if ((insn & 0x0fe00ff8) == 0x0c400000) {
2902 /* Internal Accumulator Access Format */
2903 rdhi = (insn >> 16) & 0xf;
2904 rdlo = (insn >> 12) & 0xf;
2910 if (insn & ARM_CP_RW_BIT) { /* MRA */
2911 iwmmxt_load_reg(cpu_V0, acc);
2912 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
2913 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
2914 tcg_gen_extrl_i64_i32(cpu_R[rdhi], cpu_V0);
2915 tcg_gen_andi_i32(cpu_R[rdhi], cpu_R[rdhi], (1 << (40 - 32)) - 1);
2917 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
2918 iwmmxt_store_reg(cpu_V0, acc);
2926 #define VFP_REG_SHR(x, n) (((n) > 0) ? (x) >> (n) : (x) << -(n))
2927 #define VFP_SREG(insn, bigbit, smallbit) \
2928 ((VFP_REG_SHR(insn, bigbit - 1) & 0x1e) | (((insn) >> (smallbit)) & 1))
2929 #define VFP_DREG(reg, insn, bigbit, smallbit) do { \
2930 if (arm_dc_feature(s, ARM_FEATURE_VFP3)) { \
2931 reg = (((insn) >> (bigbit)) & 0x0f) \
2932 | (((insn) >> ((smallbit) - 4)) & 0x10); \
2934 if (insn & (1 << (smallbit))) \
2936 reg = ((insn) >> (bigbit)) & 0x0f; \
2939 #define VFP_SREG_D(insn) VFP_SREG(insn, 12, 22)
2940 #define VFP_DREG_D(reg, insn) VFP_DREG(reg, insn, 12, 22)
2941 #define VFP_SREG_N(insn) VFP_SREG(insn, 16, 7)
2942 #define VFP_DREG_N(reg, insn) VFP_DREG(reg, insn, 16, 7)
2943 #define VFP_SREG_M(insn) VFP_SREG(insn, 0, 5)
2944 #define VFP_DREG_M(reg, insn) VFP_DREG(reg, insn, 0, 5)
2946 /* Move between integer and VFP cores. */
2947 static TCGv_i32 gen_vfp_mrs(void)
2949 TCGv_i32 tmp = tcg_temp_new_i32();
2950 tcg_gen_mov_i32(tmp, cpu_F0s);
2954 static void gen_vfp_msr(TCGv_i32 tmp)
2956 tcg_gen_mov_i32(cpu_F0s, tmp);
2957 tcg_temp_free_i32(tmp);
2960 static void gen_neon_dup_u8(TCGv_i32 var, int shift)
2962 TCGv_i32 tmp = tcg_temp_new_i32();
2964 tcg_gen_shri_i32(var, var, shift);
2965 tcg_gen_ext8u_i32(var, var);
2966 tcg_gen_shli_i32(tmp, var, 8);
2967 tcg_gen_or_i32(var, var, tmp);
2968 tcg_gen_shli_i32(tmp, var, 16);
2969 tcg_gen_or_i32(var, var, tmp);
2970 tcg_temp_free_i32(tmp);
2973 static void gen_neon_dup_low16(TCGv_i32 var)
2975 TCGv_i32 tmp = tcg_temp_new_i32();
2976 tcg_gen_ext16u_i32(var, var);
2977 tcg_gen_shli_i32(tmp, var, 16);
2978 tcg_gen_or_i32(var, var, tmp);
2979 tcg_temp_free_i32(tmp);
2982 static void gen_neon_dup_high16(TCGv_i32 var)
2984 TCGv_i32 tmp = tcg_temp_new_i32();
2985 tcg_gen_andi_i32(var, var, 0xffff0000);
2986 tcg_gen_shri_i32(tmp, var, 16);
2987 tcg_gen_or_i32(var, var, tmp);
2988 tcg_temp_free_i32(tmp);
2991 static TCGv_i32 gen_load_and_replicate(DisasContext *s, TCGv_i32 addr, int size)
2993 /* Load a single Neon element and replicate into a 32 bit TCG reg */
2994 TCGv_i32 tmp = tcg_temp_new_i32();
2997 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
2998 gen_neon_dup_u8(tmp, 0);
3001 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
3002 gen_neon_dup_low16(tmp);
3005 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
3007 default: /* Avoid compiler warnings. */
3013 static int handle_vsel(uint32_t insn, uint32_t rd, uint32_t rn, uint32_t rm,
3016 uint32_t cc = extract32(insn, 20, 2);
3019 TCGv_i64 frn, frm, dest;
3020 TCGv_i64 tmp, zero, zf, nf, vf;
3022 zero = tcg_const_i64(0);
3024 frn = tcg_temp_new_i64();
3025 frm = tcg_temp_new_i64();
3026 dest = tcg_temp_new_i64();
3028 zf = tcg_temp_new_i64();
3029 nf = tcg_temp_new_i64();
3030 vf = tcg_temp_new_i64();
3032 tcg_gen_extu_i32_i64(zf, cpu_ZF);
3033 tcg_gen_ext_i32_i64(nf, cpu_NF);
3034 tcg_gen_ext_i32_i64(vf, cpu_VF);
3036 tcg_gen_ld_f64(frn, cpu_env, vfp_reg_offset(dp, rn));
3037 tcg_gen_ld_f64(frm, cpu_env, vfp_reg_offset(dp, rm));
3040 tcg_gen_movcond_i64(TCG_COND_EQ, dest, zf, zero,
3044 tcg_gen_movcond_i64(TCG_COND_LT, dest, vf, zero,
3047 case 2: /* ge: N == V -> N ^ V == 0 */
3048 tmp = tcg_temp_new_i64();
3049 tcg_gen_xor_i64(tmp, vf, nf);
3050 tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero,
3052 tcg_temp_free_i64(tmp);
3054 case 3: /* gt: !Z && N == V */
3055 tcg_gen_movcond_i64(TCG_COND_NE, dest, zf, zero,
3057 tmp = tcg_temp_new_i64();
3058 tcg_gen_xor_i64(tmp, vf, nf);
3059 tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero,
3061 tcg_temp_free_i64(tmp);
3064 tcg_gen_st_f64(dest, cpu_env, vfp_reg_offset(dp, rd));
3065 tcg_temp_free_i64(frn);
3066 tcg_temp_free_i64(frm);
3067 tcg_temp_free_i64(dest);
3069 tcg_temp_free_i64(zf);
3070 tcg_temp_free_i64(nf);
3071 tcg_temp_free_i64(vf);
3073 tcg_temp_free_i64(zero);
3075 TCGv_i32 frn, frm, dest;
3078 zero = tcg_const_i32(0);
3080 frn = tcg_temp_new_i32();
3081 frm = tcg_temp_new_i32();
3082 dest = tcg_temp_new_i32();
3083 tcg_gen_ld_f32(frn, cpu_env, vfp_reg_offset(dp, rn));
3084 tcg_gen_ld_f32(frm, cpu_env, vfp_reg_offset(dp, rm));
3087 tcg_gen_movcond_i32(TCG_COND_EQ, dest, cpu_ZF, zero,
3091 tcg_gen_movcond_i32(TCG_COND_LT, dest, cpu_VF, zero,
3094 case 2: /* ge: N == V -> N ^ V == 0 */
3095 tmp = tcg_temp_new_i32();
3096 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
3097 tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero,
3099 tcg_temp_free_i32(tmp);
3101 case 3: /* gt: !Z && N == V */
3102 tcg_gen_movcond_i32(TCG_COND_NE, dest, cpu_ZF, zero,
3104 tmp = tcg_temp_new_i32();
3105 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
3106 tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero,
3108 tcg_temp_free_i32(tmp);
3111 tcg_gen_st_f32(dest, cpu_env, vfp_reg_offset(dp, rd));
3112 tcg_temp_free_i32(frn);
3113 tcg_temp_free_i32(frm);
3114 tcg_temp_free_i32(dest);
3116 tcg_temp_free_i32(zero);
3122 static int handle_vminmaxnm(uint32_t insn, uint32_t rd, uint32_t rn,
3123 uint32_t rm, uint32_t dp)
3125 uint32_t vmin = extract32(insn, 6, 1);
3126 TCGv_ptr fpst = get_fpstatus_ptr(0);
3129 TCGv_i64 frn, frm, dest;
3131 frn = tcg_temp_new_i64();
3132 frm = tcg_temp_new_i64();
3133 dest = tcg_temp_new_i64();
3135 tcg_gen_ld_f64(frn, cpu_env, vfp_reg_offset(dp, rn));
3136 tcg_gen_ld_f64(frm, cpu_env, vfp_reg_offset(dp, rm));
3138 gen_helper_vfp_minnumd(dest, frn, frm, fpst);
3140 gen_helper_vfp_maxnumd(dest, frn, frm, fpst);
3142 tcg_gen_st_f64(dest, cpu_env, vfp_reg_offset(dp, rd));
3143 tcg_temp_free_i64(frn);
3144 tcg_temp_free_i64(frm);
3145 tcg_temp_free_i64(dest);
3147 TCGv_i32 frn, frm, dest;
3149 frn = tcg_temp_new_i32();
3150 frm = tcg_temp_new_i32();
3151 dest = tcg_temp_new_i32();
3153 tcg_gen_ld_f32(frn, cpu_env, vfp_reg_offset(dp, rn));
3154 tcg_gen_ld_f32(frm, cpu_env, vfp_reg_offset(dp, rm));
3156 gen_helper_vfp_minnums(dest, frn, frm, fpst);
3158 gen_helper_vfp_maxnums(dest, frn, frm, fpst);
3160 tcg_gen_st_f32(dest, cpu_env, vfp_reg_offset(dp, rd));
3161 tcg_temp_free_i32(frn);
3162 tcg_temp_free_i32(frm);
3163 tcg_temp_free_i32(dest);
3166 tcg_temp_free_ptr(fpst);
3170 static int handle_vrint(uint32_t insn, uint32_t rd, uint32_t rm, uint32_t dp,
3173 TCGv_ptr fpst = get_fpstatus_ptr(0);
3176 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding));
3177 gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
3182 tcg_op = tcg_temp_new_i64();
3183 tcg_res = tcg_temp_new_i64();
3184 tcg_gen_ld_f64(tcg_op, cpu_env, vfp_reg_offset(dp, rm));
3185 gen_helper_rintd(tcg_res, tcg_op, fpst);
3186 tcg_gen_st_f64(tcg_res, cpu_env, vfp_reg_offset(dp, rd));
3187 tcg_temp_free_i64(tcg_op);
3188 tcg_temp_free_i64(tcg_res);
3192 tcg_op = tcg_temp_new_i32();
3193 tcg_res = tcg_temp_new_i32();
3194 tcg_gen_ld_f32(tcg_op, cpu_env, vfp_reg_offset(dp, rm));
3195 gen_helper_rints(tcg_res, tcg_op, fpst);
3196 tcg_gen_st_f32(tcg_res, cpu_env, vfp_reg_offset(dp, rd));
3197 tcg_temp_free_i32(tcg_op);
3198 tcg_temp_free_i32(tcg_res);
3201 gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
3202 tcg_temp_free_i32(tcg_rmode);
3204 tcg_temp_free_ptr(fpst);
3208 static int handle_vcvt(uint32_t insn, uint32_t rd, uint32_t rm, uint32_t dp,
3211 bool is_signed = extract32(insn, 7, 1);
3212 TCGv_ptr fpst = get_fpstatus_ptr(0);
3213 TCGv_i32 tcg_rmode, tcg_shift;
3215 tcg_shift = tcg_const_i32(0);
3217 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding));
3218 gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
3221 TCGv_i64 tcg_double, tcg_res;
3223 /* Rd is encoded as a single precision register even when the source
3224 * is double precision.
3226 rd = ((rd << 1) & 0x1e) | ((rd >> 4) & 0x1);
3227 tcg_double = tcg_temp_new_i64();
3228 tcg_res = tcg_temp_new_i64();
3229 tcg_tmp = tcg_temp_new_i32();
3230 tcg_gen_ld_f64(tcg_double, cpu_env, vfp_reg_offset(1, rm));
3232 gen_helper_vfp_tosld(tcg_res, tcg_double, tcg_shift, fpst);
3234 gen_helper_vfp_tould(tcg_res, tcg_double, tcg_shift, fpst);
3236 tcg_gen_extrl_i64_i32(tcg_tmp, tcg_res);
3237 tcg_gen_st_f32(tcg_tmp, cpu_env, vfp_reg_offset(0, rd));
3238 tcg_temp_free_i32(tcg_tmp);
3239 tcg_temp_free_i64(tcg_res);
3240 tcg_temp_free_i64(tcg_double);
3242 TCGv_i32 tcg_single, tcg_res;
3243 tcg_single = tcg_temp_new_i32();
3244 tcg_res = tcg_temp_new_i32();
3245 tcg_gen_ld_f32(tcg_single, cpu_env, vfp_reg_offset(0, rm));
3247 gen_helper_vfp_tosls(tcg_res, tcg_single, tcg_shift, fpst);
3249 gen_helper_vfp_touls(tcg_res, tcg_single, tcg_shift, fpst);
3251 tcg_gen_st_f32(tcg_res, cpu_env, vfp_reg_offset(0, rd));
3252 tcg_temp_free_i32(tcg_res);
3253 tcg_temp_free_i32(tcg_single);
3256 gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
3257 tcg_temp_free_i32(tcg_rmode);
3259 tcg_temp_free_i32(tcg_shift);
3261 tcg_temp_free_ptr(fpst);
3266 /* Table for converting the most common AArch32 encoding of
3267 * rounding mode to arm_fprounding order (which matches the
3268 * common AArch64 order); see ARM ARM pseudocode FPDecodeRM().
3270 static const uint8_t fp_decode_rm[] = {
3277 static int disas_vfp_v8_insn(DisasContext *s, uint32_t insn)
3279 uint32_t rd, rn, rm, dp = extract32(insn, 8, 1);
3281 if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
3286 VFP_DREG_D(rd, insn);
3287 VFP_DREG_N(rn, insn);
3288 VFP_DREG_M(rm, insn);
3290 rd = VFP_SREG_D(insn);
3291 rn = VFP_SREG_N(insn);
3292 rm = VFP_SREG_M(insn);
3295 if ((insn & 0x0f800e50) == 0x0e000a00) {
3296 return handle_vsel(insn, rd, rn, rm, dp);
3297 } else if ((insn & 0x0fb00e10) == 0x0e800a00) {
3298 return handle_vminmaxnm(insn, rd, rn, rm, dp);
3299 } else if ((insn & 0x0fbc0ed0) == 0x0eb80a40) {
3300 /* VRINTA, VRINTN, VRINTP, VRINTM */
3301 int rounding = fp_decode_rm[extract32(insn, 16, 2)];
3302 return handle_vrint(insn, rd, rm, dp, rounding);
3303 } else if ((insn & 0x0fbc0e50) == 0x0ebc0a40) {
3304 /* VCVTA, VCVTN, VCVTP, VCVTM */
3305 int rounding = fp_decode_rm[extract32(insn, 16, 2)];
3306 return handle_vcvt(insn, rd, rm, dp, rounding);
3311 /* Disassemble a VFP instruction. Returns nonzero if an error occurred
3312 (ie. an undefined instruction). */
3313 static int disas_vfp_insn(DisasContext *s, uint32_t insn)
3315 uint32_t rd, rn, rm, op, i, n, offset, delta_d, delta_m, bank_mask;
3321 if (!arm_dc_feature(s, ARM_FEATURE_VFP)) {
3325 /* FIXME: this access check should not take precedence over UNDEF
3326 * for invalid encodings; we will generate incorrect syndrome information
3327 * for attempts to execute invalid vfp/neon encodings with FP disabled.
3329 if (s->fp_excp_el) {
3330 gen_exception_insn(s, 4, EXCP_UDEF,
3331 syn_fp_access_trap(1, 0xe, false), s->fp_excp_el);
3335 if (!s->vfp_enabled) {
3336 /* VFP disabled. Only allow fmxr/fmrx to/from some control regs. */
3337 if ((insn & 0x0fe00fff) != 0x0ee00a10)
3339 rn = (insn >> 16) & 0xf;
3340 if (rn != ARM_VFP_FPSID && rn != ARM_VFP_FPEXC && rn != ARM_VFP_MVFR2
3341 && rn != ARM_VFP_MVFR1 && rn != ARM_VFP_MVFR0) {
3346 if (extract32(insn, 28, 4) == 0xf) {
3347 /* Encodings with T=1 (Thumb) or unconditional (ARM):
3348 * only used in v8 and above.
3350 return disas_vfp_v8_insn(s, insn);
3353 dp = ((insn & 0xf00) == 0xb00);
3354 switch ((insn >> 24) & 0xf) {
3356 if (insn & (1 << 4)) {
3357 /* single register transfer */
3358 rd = (insn >> 12) & 0xf;
3363 VFP_DREG_N(rn, insn);
3366 if (insn & 0x00c00060
3367 && !arm_dc_feature(s, ARM_FEATURE_NEON)) {
3371 pass = (insn >> 21) & 1;
3372 if (insn & (1 << 22)) {
3374 offset = ((insn >> 5) & 3) * 8;
3375 } else if (insn & (1 << 5)) {
3377 offset = (insn & (1 << 6)) ? 16 : 0;
3382 if (insn & ARM_CP_RW_BIT) {
3384 tmp = neon_load_reg(rn, pass);
3388 tcg_gen_shri_i32(tmp, tmp, offset);
3389 if (insn & (1 << 23))
3395 if (insn & (1 << 23)) {
3397 tcg_gen_shri_i32(tmp, tmp, 16);
3403 tcg_gen_sari_i32(tmp, tmp, 16);
3412 store_reg(s, rd, tmp);
3415 tmp = load_reg(s, rd);
3416 if (insn & (1 << 23)) {
3419 gen_neon_dup_u8(tmp, 0);
3420 } else if (size == 1) {
3421 gen_neon_dup_low16(tmp);
3423 for (n = 0; n <= pass * 2; n++) {
3424 tmp2 = tcg_temp_new_i32();
3425 tcg_gen_mov_i32(tmp2, tmp);
3426 neon_store_reg(rn, n, tmp2);
3428 neon_store_reg(rn, n, tmp);
3433 tmp2 = neon_load_reg(rn, pass);
3434 tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 8);
3435 tcg_temp_free_i32(tmp2);
3438 tmp2 = neon_load_reg(rn, pass);
3439 tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 16);
3440 tcg_temp_free_i32(tmp2);
3445 neon_store_reg(rn, pass, tmp);
3449 if ((insn & 0x6f) != 0x00)
3451 rn = VFP_SREG_N(insn);
3452 if (insn & ARM_CP_RW_BIT) {
3454 if (insn & (1 << 21)) {
3455 /* system register */
3460 /* VFP2 allows access to FSID from userspace.
3461 VFP3 restricts all id registers to privileged
3464 && arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3467 tmp = load_cpu_field(vfp.xregs[rn]);
3472 tmp = load_cpu_field(vfp.xregs[rn]);
3474 case ARM_VFP_FPINST:
3475 case ARM_VFP_FPINST2:
3476 /* Not present in VFP3. */
3478 || arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3481 tmp = load_cpu_field(vfp.xregs[rn]);
3485 tmp = load_cpu_field(vfp.xregs[ARM_VFP_FPSCR]);
3486 tcg_gen_andi_i32(tmp, tmp, 0xf0000000);
3488 tmp = tcg_temp_new_i32();
3489 gen_helper_vfp_get_fpscr(tmp, cpu_env);
3493 if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
3500 || !arm_dc_feature(s, ARM_FEATURE_MVFR)) {
3503 tmp = load_cpu_field(vfp.xregs[rn]);
3509 gen_mov_F0_vreg(0, rn);
3510 tmp = gen_vfp_mrs();
3513 /* Set the 4 flag bits in the CPSR. */
3515 tcg_temp_free_i32(tmp);
3517 store_reg(s, rd, tmp);
3521 if (insn & (1 << 21)) {
3523 /* system register */
3528 /* Writes are ignored. */
3531 tmp = load_reg(s, rd);
3532 gen_helper_vfp_set_fpscr(cpu_env, tmp);
3533 tcg_temp_free_i32(tmp);
3539 /* TODO: VFP subarchitecture support.
3540 * For now, keep the EN bit only */
3541 tmp = load_reg(s, rd);
3542 tcg_gen_andi_i32(tmp, tmp, 1 << 30);
3543 store_cpu_field(tmp, vfp.xregs[rn]);
3546 case ARM_VFP_FPINST:
3547 case ARM_VFP_FPINST2:
3551 tmp = load_reg(s, rd);
3552 store_cpu_field(tmp, vfp.xregs[rn]);
3558 tmp = load_reg(s, rd);
3560 gen_mov_vreg_F0(0, rn);
3565 /* data processing */
3566 /* The opcode is in bits 23, 21, 20 and 6. */
3567 op = ((insn >> 20) & 8) | ((insn >> 19) & 6) | ((insn >> 6) & 1);
3571 rn = ((insn >> 15) & 0x1e) | ((insn >> 7) & 1);
3573 /* rn is register number */
3574 VFP_DREG_N(rn, insn);
3577 if (op == 15 && (rn == 15 || ((rn & 0x1c) == 0x18) ||
3578 ((rn & 0x1e) == 0x6))) {
3579 /* Integer or single/half precision destination. */
3580 rd = VFP_SREG_D(insn);
3582 VFP_DREG_D(rd, insn);
3585 (((rn & 0x1c) == 0x10) || ((rn & 0x14) == 0x14) ||
3586 ((rn & 0x1e) == 0x4))) {
3587 /* VCVT from int or half precision is always from S reg
3588 * regardless of dp bit. VCVT with immediate frac_bits
3589 * has same format as SREG_M.
3591 rm = VFP_SREG_M(insn);
3593 VFP_DREG_M(rm, insn);
3596 rn = VFP_SREG_N(insn);
3597 if (op == 15 && rn == 15) {
3598 /* Double precision destination. */
3599 VFP_DREG_D(rd, insn);
3601 rd = VFP_SREG_D(insn);
3603 /* NB that we implicitly rely on the encoding for the frac_bits
3604 * in VCVT of fixed to float being the same as that of an SREG_M
3606 rm = VFP_SREG_M(insn);
3609 veclen = s->vec_len;
3610 if (op == 15 && rn > 3)
3613 /* Shut up compiler warnings. */
3624 /* Figure out what type of vector operation this is. */
3625 if ((rd & bank_mask) == 0) {
3630 delta_d = (s->vec_stride >> 1) + 1;
3632 delta_d = s->vec_stride + 1;
3634 if ((rm & bank_mask) == 0) {
3635 /* mixed scalar/vector */
3644 /* Load the initial operands. */
3649 /* Integer source */
3650 gen_mov_F0_vreg(0, rm);
3655 gen_mov_F0_vreg(dp, rd);
3656 gen_mov_F1_vreg(dp, rm);
3660 /* Compare with zero */
3661 gen_mov_F0_vreg(dp, rd);
3672 /* Source and destination the same. */
3673 gen_mov_F0_vreg(dp, rd);
3679 /* VCVTB, VCVTT: only present with the halfprec extension
3680 * UNPREDICTABLE if bit 8 is set prior to ARMv8
3681 * (we choose to UNDEF)
3683 if ((dp && !arm_dc_feature(s, ARM_FEATURE_V8)) ||
3684 !arm_dc_feature(s, ARM_FEATURE_VFP_FP16)) {
3687 if (!extract32(rn, 1, 1)) {
3688 /* Half precision source. */
3689 gen_mov_F0_vreg(0, rm);
3692 /* Otherwise fall through */
3694 /* One source operand. */
3695 gen_mov_F0_vreg(dp, rm);
3699 /* Two source operands. */
3700 gen_mov_F0_vreg(dp, rn);
3701 gen_mov_F1_vreg(dp, rm);
3705 /* Perform the calculation. */
3707 case 0: /* VMLA: fd + (fn * fm) */
3708 /* Note that order of inputs to the add matters for NaNs */
3710 gen_mov_F0_vreg(dp, rd);
3713 case 1: /* VMLS: fd + -(fn * fm) */
3716 gen_mov_F0_vreg(dp, rd);
3719 case 2: /* VNMLS: -fd + (fn * fm) */
3720 /* Note that it isn't valid to replace (-A + B) with (B - A)
3721 * or similar plausible looking simplifications
3722 * because this will give wrong results for NaNs.
3725 gen_mov_F0_vreg(dp, rd);
3729 case 3: /* VNMLA: -fd + -(fn * fm) */
3732 gen_mov_F0_vreg(dp, rd);
3736 case 4: /* mul: fn * fm */
3739 case 5: /* nmul: -(fn * fm) */
3743 case 6: /* add: fn + fm */
3746 case 7: /* sub: fn - fm */
3749 case 8: /* div: fn / fm */
3752 case 10: /* VFNMA : fd = muladd(-fd, fn, fm) */
3753 case 11: /* VFNMS : fd = muladd(-fd, -fn, fm) */
3754 case 12: /* VFMA : fd = muladd( fd, fn, fm) */
3755 case 13: /* VFMS : fd = muladd( fd, -fn, fm) */
3756 /* These are fused multiply-add, and must be done as one
3757 * floating point operation with no rounding between the
3758 * multiplication and addition steps.
3759 * NB that doing the negations here as separate steps is
3760 * correct : an input NaN should come out with its sign bit
3761 * flipped if it is a negated-input.
3763 if (!arm_dc_feature(s, ARM_FEATURE_VFP4)) {
3771 gen_helper_vfp_negd(cpu_F0d, cpu_F0d);
3773 frd = tcg_temp_new_i64();
3774 tcg_gen_ld_f64(frd, cpu_env, vfp_reg_offset(dp, rd));
3777 gen_helper_vfp_negd(frd, frd);
3779 fpst = get_fpstatus_ptr(0);
3780 gen_helper_vfp_muladdd(cpu_F0d, cpu_F0d,
3781 cpu_F1d, frd, fpst);
3782 tcg_temp_free_ptr(fpst);
3783 tcg_temp_free_i64(frd);
3789 gen_helper_vfp_negs(cpu_F0s, cpu_F0s);
3791 frd = tcg_temp_new_i32();
3792 tcg_gen_ld_f32(frd, cpu_env, vfp_reg_offset(dp, rd));
3794 gen_helper_vfp_negs(frd, frd);
3796 fpst = get_fpstatus_ptr(0);
3797 gen_helper_vfp_muladds(cpu_F0s, cpu_F0s,
3798 cpu_F1s, frd, fpst);
3799 tcg_temp_free_ptr(fpst);
3800 tcg_temp_free_i32(frd);
3803 case 14: /* fconst */
3804 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3808 n = (insn << 12) & 0x80000000;
3809 i = ((insn >> 12) & 0x70) | (insn & 0xf);
3816 tcg_gen_movi_i64(cpu_F0d, ((uint64_t)n) << 32);
3823 tcg_gen_movi_i32(cpu_F0s, n);
3826 case 15: /* extension space */
3840 case 4: /* vcvtb.f32.f16, vcvtb.f64.f16 */
3842 TCGv_ptr fpst = get_fpstatus_ptr(false);
3843 TCGv_i32 ahp_mode = get_ahp_flag();
3844 tmp = gen_vfp_mrs();
3845 tcg_gen_ext16u_i32(tmp, tmp);
3847 gen_helper_vfp_fcvt_f16_to_f64(cpu_F0d, tmp,
3850 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp,
3853 tcg_temp_free_i32(ahp_mode);
3854 tcg_temp_free_ptr(fpst);
3855 tcg_temp_free_i32(tmp);
3858 case 5: /* vcvtt.f32.f16, vcvtt.f64.f16 */
3860 TCGv_ptr fpst = get_fpstatus_ptr(false);
3861 TCGv_i32 ahp = get_ahp_flag();
3862 tmp = gen_vfp_mrs();
3863 tcg_gen_shri_i32(tmp, tmp, 16);
3865 gen_helper_vfp_fcvt_f16_to_f64(cpu_F0d, tmp,
3868 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp,
3871 tcg_temp_free_i32(tmp);
3872 tcg_temp_free_i32(ahp);
3873 tcg_temp_free_ptr(fpst);
3876 case 6: /* vcvtb.f16.f32, vcvtb.f16.f64 */
3878 TCGv_ptr fpst = get_fpstatus_ptr(false);
3879 TCGv_i32 ahp = get_ahp_flag();
3880 tmp = tcg_temp_new_i32();
3883 gen_helper_vfp_fcvt_f64_to_f16(tmp, cpu_F0d,
3886 gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s,
3889 tcg_temp_free_i32(ahp);
3890 tcg_temp_free_ptr(fpst);
3891 gen_mov_F0_vreg(0, rd);
3892 tmp2 = gen_vfp_mrs();
3893 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
3894 tcg_gen_or_i32(tmp, tmp, tmp2);
3895 tcg_temp_free_i32(tmp2);
3899 case 7: /* vcvtt.f16.f32, vcvtt.f16.f64 */
3901 TCGv_ptr fpst = get_fpstatus_ptr(false);
3902 TCGv_i32 ahp = get_ahp_flag();
3903 tmp = tcg_temp_new_i32();
3905 gen_helper_vfp_fcvt_f64_to_f16(tmp, cpu_F0d,
3908 gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s,
3911 tcg_temp_free_i32(ahp);
3912 tcg_temp_free_ptr(fpst);
3913 tcg_gen_shli_i32(tmp, tmp, 16);
3914 gen_mov_F0_vreg(0, rd);
3915 tmp2 = gen_vfp_mrs();
3916 tcg_gen_ext16u_i32(tmp2, tmp2);
3917 tcg_gen_or_i32(tmp, tmp, tmp2);
3918 tcg_temp_free_i32(tmp2);
3931 case 11: /* cmpez */
3935 case 12: /* vrintr */
3937 TCGv_ptr fpst = get_fpstatus_ptr(0);
3939 gen_helper_rintd(cpu_F0d, cpu_F0d, fpst);
3941 gen_helper_rints(cpu_F0s, cpu_F0s, fpst);
3943 tcg_temp_free_ptr(fpst);
3946 case 13: /* vrintz */
3948 TCGv_ptr fpst = get_fpstatus_ptr(0);
3950 tcg_rmode = tcg_const_i32(float_round_to_zero);
3951 gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
3953 gen_helper_rintd(cpu_F0d, cpu_F0d, fpst);
3955 gen_helper_rints(cpu_F0s, cpu_F0s, fpst);
3957 gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
3958 tcg_temp_free_i32(tcg_rmode);
3959 tcg_temp_free_ptr(fpst);
3962 case 14: /* vrintx */
3964 TCGv_ptr fpst = get_fpstatus_ptr(0);
3966 gen_helper_rintd_exact(cpu_F0d, cpu_F0d, fpst);
3968 gen_helper_rints_exact(cpu_F0s, cpu_F0s, fpst);
3970 tcg_temp_free_ptr(fpst);
3973 case 15: /* single<->double conversion */
3975 gen_helper_vfp_fcvtsd(cpu_F0s, cpu_F0d, cpu_env);
3977 gen_helper_vfp_fcvtds(cpu_F0d, cpu_F0s, cpu_env);
3979 case 16: /* fuito */
3980 gen_vfp_uito(dp, 0);
3982 case 17: /* fsito */
3983 gen_vfp_sito(dp, 0);
3985 case 20: /* fshto */
3986 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3989 gen_vfp_shto(dp, 16 - rm, 0);
3991 case 21: /* fslto */
3992 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3995 gen_vfp_slto(dp, 32 - rm, 0);
3997 case 22: /* fuhto */
3998 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
4001 gen_vfp_uhto(dp, 16 - rm, 0);
4003 case 23: /* fulto */
4004 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
4007 gen_vfp_ulto(dp, 32 - rm, 0);
4009 case 24: /* ftoui */
4010 gen_vfp_toui(dp, 0);
4012 case 25: /* ftouiz */
4013 gen_vfp_touiz(dp, 0);
4015 case 26: /* ftosi */
4016 gen_vfp_tosi(dp, 0);
4018 case 27: /* ftosiz */
4019 gen_vfp_tosiz(dp, 0);
4021 case 28: /* ftosh */
4022 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
4025 gen_vfp_tosh(dp, 16 - rm, 0);
4027 case 29: /* ftosl */
4028 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
4031 gen_vfp_tosl(dp, 32 - rm, 0);
4033 case 30: /* ftouh */
4034 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
4037 gen_vfp_touh(dp, 16 - rm, 0);
4039 case 31: /* ftoul */
4040 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
4043 gen_vfp_toul(dp, 32 - rm, 0);
4045 default: /* undefined */
4049 default: /* undefined */
4053 /* Write back the result. */
4054 if (op == 15 && (rn >= 8 && rn <= 11)) {
4055 /* Comparison, do nothing. */
4056 } else if (op == 15 && dp && ((rn & 0x1c) == 0x18 ||
4057 (rn & 0x1e) == 0x6)) {
4058 /* VCVT double to int: always integer result.
4059 * VCVT double to half precision is always a single
4062 gen_mov_vreg_F0(0, rd);
4063 } else if (op == 15 && rn == 15) {
4065 gen_mov_vreg_F0(!dp, rd);
4067 gen_mov_vreg_F0(dp, rd);
4070 /* break out of the loop if we have finished */
4074 if (op == 15 && delta_m == 0) {
4075 /* single source one-many */
4077 rd = ((rd + delta_d) & (bank_mask - 1))
4079 gen_mov_vreg_F0(dp, rd);
4083 /* Setup the next operands. */
4085 rd = ((rd + delta_d) & (bank_mask - 1))
4089 /* One source operand. */
4090 rm = ((rm + delta_m) & (bank_mask - 1))
4092 gen_mov_F0_vreg(dp, rm);
4094 /* Two source operands. */
4095 rn = ((rn + delta_d) & (bank_mask - 1))
4097 gen_mov_F0_vreg(dp, rn);
4099 rm = ((rm + delta_m) & (bank_mask - 1))
4101 gen_mov_F1_vreg(dp, rm);
4109 if ((insn & 0x03e00000) == 0x00400000) {
4110 /* two-register transfer */
4111 rn = (insn >> 16) & 0xf;
4112 rd = (insn >> 12) & 0xf;
4114 VFP_DREG_M(rm, insn);
4116 rm = VFP_SREG_M(insn);
4119 if (insn & ARM_CP_RW_BIT) {
4122 gen_mov_F0_vreg(0, rm * 2);
4123 tmp = gen_vfp_mrs();
4124 store_reg(s, rd, tmp);
4125 gen_mov_F0_vreg(0, rm * 2 + 1);
4126 tmp = gen_vfp_mrs();
4127 store_reg(s, rn, tmp);
4129 gen_mov_F0_vreg(0, rm);
4130 tmp = gen_vfp_mrs();
4131 store_reg(s, rd, tmp);
4132 gen_mov_F0_vreg(0, rm + 1);
4133 tmp = gen_vfp_mrs();
4134 store_reg(s, rn, tmp);
4139 tmp = load_reg(s, rd);
4141 gen_mov_vreg_F0(0, rm * 2);
4142 tmp = load_reg(s, rn);
4144 gen_mov_vreg_F0(0, rm * 2 + 1);
4146 tmp = load_reg(s, rd);
4148 gen_mov_vreg_F0(0, rm);
4149 tmp = load_reg(s, rn);
4151 gen_mov_vreg_F0(0, rm + 1);
4156 rn = (insn >> 16) & 0xf;
4158 VFP_DREG_D(rd, insn);
4160 rd = VFP_SREG_D(insn);
4161 if ((insn & 0x01200000) == 0x01000000) {
4162 /* Single load/store */
4163 offset = (insn & 0xff) << 2;
4164 if ((insn & (1 << 23)) == 0)
4166 if (s->thumb && rn == 15) {
4167 /* This is actually UNPREDICTABLE */
4168 addr = tcg_temp_new_i32();
4169 tcg_gen_movi_i32(addr, s->pc & ~2);
4171 addr = load_reg(s, rn);
4173 tcg_gen_addi_i32(addr, addr, offset);
4174 if (insn & (1 << 20)) {
4175 gen_vfp_ld(s, dp, addr);
4176 gen_mov_vreg_F0(dp, rd);
4178 gen_mov_F0_vreg(dp, rd);
4179 gen_vfp_st(s, dp, addr);
4181 tcg_temp_free_i32(addr);
4183 /* load/store multiple */
4184 int w = insn & (1 << 21);
4186 n = (insn >> 1) & 0x7f;
4190 if (w && !(((insn >> 23) ^ (insn >> 24)) & 1)) {
4191 /* P == U , W == 1 => UNDEF */
4194 if (n == 0 || (rd + n) > 32 || (dp && n > 16)) {
4195 /* UNPREDICTABLE cases for bad immediates: we choose to
4196 * UNDEF to avoid generating huge numbers of TCG ops
4200 if (rn == 15 && w) {
4201 /* writeback to PC is UNPREDICTABLE, we choose to UNDEF */
4205 if (s->thumb && rn == 15) {
4206 /* This is actually UNPREDICTABLE */
4207 addr = tcg_temp_new_i32();
4208 tcg_gen_movi_i32(addr, s->pc & ~2);
4210 addr = load_reg(s, rn);
4212 if (insn & (1 << 24)) /* pre-decrement */
4213 tcg_gen_addi_i32(addr, addr, -((insn & 0xff) << 2));
4219 for (i = 0; i < n; i++) {
4220 if (insn & ARM_CP_RW_BIT) {
4222 gen_vfp_ld(s, dp, addr);
4223 gen_mov_vreg_F0(dp, rd + i);
4226 gen_mov_F0_vreg(dp, rd + i);
4227 gen_vfp_st(s, dp, addr);
4229 tcg_gen_addi_i32(addr, addr, offset);
4233 if (insn & (1 << 24))
4234 offset = -offset * n;
4235 else if (dp && (insn & 1))
4241 tcg_gen_addi_i32(addr, addr, offset);
4242 store_reg(s, rn, addr);
4244 tcg_temp_free_i32(addr);
4250 /* Should never happen. */
4256 static inline bool use_goto_tb(DisasContext *s, target_ulong dest)
4258 #ifndef CONFIG_USER_ONLY
4259 return (s->base.tb->pc & TARGET_PAGE_MASK) == (dest & TARGET_PAGE_MASK) ||
4260 ((s->pc - 1) & TARGET_PAGE_MASK) == (dest & TARGET_PAGE_MASK);
4266 static void gen_goto_ptr(void)
4268 tcg_gen_lookup_and_goto_ptr();
4271 /* This will end the TB but doesn't guarantee we'll return to
4272 * cpu_loop_exec. Any live exit_requests will be processed as we
4273 * enter the next TB.
4275 static void gen_goto_tb(DisasContext *s, int n, target_ulong dest)
4277 if (use_goto_tb(s, dest)) {
4279 gen_set_pc_im(s, dest);
4280 tcg_gen_exit_tb(s->base.tb, n);
4282 gen_set_pc_im(s, dest);
4285 s->base.is_jmp = DISAS_NORETURN;
4288 static inline void gen_jmp (DisasContext *s, uint32_t dest)
4290 if (unlikely(is_singlestepping(s))) {
4291 /* An indirect jump so that we still trigger the debug exception. */
4296 gen_goto_tb(s, 0, dest);
4300 static inline void gen_mulxy(TCGv_i32 t0, TCGv_i32 t1, int x, int y)
4303 tcg_gen_sari_i32(t0, t0, 16);
4307 tcg_gen_sari_i32(t1, t1, 16);
4310 tcg_gen_mul_i32(t0, t0, t1);
4313 /* Return the mask of PSR bits set by a MSR instruction. */
4314 static uint32_t msr_mask(DisasContext *s, int flags, int spsr)
4319 if (flags & (1 << 0))
4321 if (flags & (1 << 1))
4323 if (flags & (1 << 2))
4325 if (flags & (1 << 3))
4328 /* Mask out undefined bits. */
4329 mask &= ~CPSR_RESERVED;
4330 if (!arm_dc_feature(s, ARM_FEATURE_V4T)) {
4333 if (!arm_dc_feature(s, ARM_FEATURE_V5)) {
4334 mask &= ~CPSR_Q; /* V5TE in reality*/
4336 if (!arm_dc_feature(s, ARM_FEATURE_V6)) {
4337 mask &= ~(CPSR_E | CPSR_GE);
4339 if (!arm_dc_feature(s, ARM_FEATURE_THUMB2)) {
4342 /* Mask out execution state and reserved bits. */
4344 mask &= ~(CPSR_EXEC | CPSR_RESERVED);
4346 /* Mask out privileged bits. */
4352 /* Returns nonzero if access to the PSR is not permitted. Marks t0 as dead. */
4353 static int gen_set_psr(DisasContext *s, uint32_t mask, int spsr, TCGv_i32 t0)
4357 /* ??? This is also undefined in system mode. */
4361 tmp = load_cpu_field(spsr);
4362 tcg_gen_andi_i32(tmp, tmp, ~mask);
4363 tcg_gen_andi_i32(t0, t0, mask);
4364 tcg_gen_or_i32(tmp, tmp, t0);
4365 store_cpu_field(tmp, spsr);
4367 gen_set_cpsr(t0, mask);
4369 tcg_temp_free_i32(t0);
4374 /* Returns nonzero if access to the PSR is not permitted. */
4375 static int gen_set_psr_im(DisasContext *s, uint32_t mask, int spsr, uint32_t val)
4378 tmp = tcg_temp_new_i32();
4379 tcg_gen_movi_i32(tmp, val);
4380 return gen_set_psr(s, mask, spsr, tmp);
4383 static bool msr_banked_access_decode(DisasContext *s, int r, int sysm, int rn,
4384 int *tgtmode, int *regno)
4386 /* Decode the r and sysm fields of MSR/MRS banked accesses into
4387 * the target mode and register number, and identify the various
4388 * unpredictable cases.
4389 * MSR (banked) and MRS (banked) are CONSTRAINED UNPREDICTABLE if:
4390 * + executed in user mode
4391 * + using R15 as the src/dest register
4392 * + accessing an unimplemented register
4393 * + accessing a register that's inaccessible at current PL/security state*
4394 * + accessing a register that you could access with a different insn
4395 * We choose to UNDEF in all these cases.
4396 * Since we don't know which of the various AArch32 modes we are in
4397 * we have to defer some checks to runtime.
4398 * Accesses to Monitor mode registers from Secure EL1 (which implies
4399 * that EL3 is AArch64) must trap to EL3.
4401 * If the access checks fail this function will emit code to take
4402 * an exception and return false. Otherwise it will return true,
4403 * and set *tgtmode and *regno appropriately.
4405 int exc_target = default_exception_el(s);
4407 /* These instructions are present only in ARMv8, or in ARMv7 with the
4408 * Virtualization Extensions.
4410 if (!arm_dc_feature(s, ARM_FEATURE_V8) &&
4411 !arm_dc_feature(s, ARM_FEATURE_EL2)) {
4415 if (IS_USER(s) || rn == 15) {
4419 /* The table in the v8 ARM ARM section F5.2.3 describes the encoding
4420 * of registers into (r, sysm).
4423 /* SPSRs for other modes */
4425 case 0xe: /* SPSR_fiq */
4426 *tgtmode = ARM_CPU_MODE_FIQ;
4428 case 0x10: /* SPSR_irq */
4429 *tgtmode = ARM_CPU_MODE_IRQ;
4431 case 0x12: /* SPSR_svc */
4432 *tgtmode = ARM_CPU_MODE_SVC;
4434 case 0x14: /* SPSR_abt */
4435 *tgtmode = ARM_CPU_MODE_ABT;
4437 case 0x16: /* SPSR_und */
4438 *tgtmode = ARM_CPU_MODE_UND;
4440 case 0x1c: /* SPSR_mon */
4441 *tgtmode = ARM_CPU_MODE_MON;
4443 case 0x1e: /* SPSR_hyp */
4444 *tgtmode = ARM_CPU_MODE_HYP;
4446 default: /* unallocated */
4449 /* We arbitrarily assign SPSR a register number of 16. */
4452 /* general purpose registers for other modes */
4454 case 0x0 ... 0x6: /* 0b00xxx : r8_usr ... r14_usr */
4455 *tgtmode = ARM_CPU_MODE_USR;
4458 case 0x8 ... 0xe: /* 0b01xxx : r8_fiq ... r14_fiq */
4459 *tgtmode = ARM_CPU_MODE_FIQ;
4462 case 0x10 ... 0x11: /* 0b1000x : r14_irq, r13_irq */
4463 *tgtmode = ARM_CPU_MODE_IRQ;
4464 *regno = sysm & 1 ? 13 : 14;
4466 case 0x12 ... 0x13: /* 0b1001x : r14_svc, r13_svc */
4467 *tgtmode = ARM_CPU_MODE_SVC;
4468 *regno = sysm & 1 ? 13 : 14;
4470 case 0x14 ... 0x15: /* 0b1010x : r14_abt, r13_abt */
4471 *tgtmode = ARM_CPU_MODE_ABT;
4472 *regno = sysm & 1 ? 13 : 14;
4474 case 0x16 ... 0x17: /* 0b1011x : r14_und, r13_und */
4475 *tgtmode = ARM_CPU_MODE_UND;
4476 *regno = sysm & 1 ? 13 : 14;
4478 case 0x1c ... 0x1d: /* 0b1110x : r14_mon, r13_mon */
4479 *tgtmode = ARM_CPU_MODE_MON;
4480 *regno = sysm & 1 ? 13 : 14;
4482 case 0x1e ... 0x1f: /* 0b1111x : elr_hyp, r13_hyp */
4483 *tgtmode = ARM_CPU_MODE_HYP;
4484 /* Arbitrarily pick 17 for ELR_Hyp (which is not a banked LR!) */
4485 *regno = sysm & 1 ? 13 : 17;
4487 default: /* unallocated */
4492 /* Catch the 'accessing inaccessible register' cases we can detect
4493 * at translate time.
4496 case ARM_CPU_MODE_MON:
4497 if (!arm_dc_feature(s, ARM_FEATURE_EL3) || s->ns) {
4500 if (s->current_el == 1) {
4501 /* If we're in Secure EL1 (which implies that EL3 is AArch64)
4502 * then accesses to Mon registers trap to EL3
4508 case ARM_CPU_MODE_HYP:
4509 /* Note that we can forbid accesses from EL2 here because they
4510 * must be from Hyp mode itself
4512 if (!arm_dc_feature(s, ARM_FEATURE_EL2) || s->current_el < 3) {
4523 /* If we get here then some access check did not pass */
4524 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(), exc_target);
4528 static void gen_msr_banked(DisasContext *s, int r, int sysm, int rn)
4530 TCGv_i32 tcg_reg, tcg_tgtmode, tcg_regno;
4531 int tgtmode = 0, regno = 0;
4533 if (!msr_banked_access_decode(s, r, sysm, rn, &tgtmode, ®no)) {
4537 /* Sync state because msr_banked() can raise exceptions */
4538 gen_set_condexec(s);
4539 gen_set_pc_im(s, s->pc - 4);
4540 tcg_reg = load_reg(s, rn);
4541 tcg_tgtmode = tcg_const_i32(tgtmode);
4542 tcg_regno = tcg_const_i32(regno);
4543 gen_helper_msr_banked(cpu_env, tcg_reg, tcg_tgtmode, tcg_regno);
4544 tcg_temp_free_i32(tcg_tgtmode);
4545 tcg_temp_free_i32(tcg_regno);
4546 tcg_temp_free_i32(tcg_reg);
4547 s->base.is_jmp = DISAS_UPDATE;
4550 static void gen_mrs_banked(DisasContext *s, int r, int sysm, int rn)
4552 TCGv_i32 tcg_reg, tcg_tgtmode, tcg_regno;
4553 int tgtmode = 0, regno = 0;
4555 if (!msr_banked_access_decode(s, r, sysm, rn, &tgtmode, ®no)) {
4559 /* Sync state because mrs_banked() can raise exceptions */
4560 gen_set_condexec(s);
4561 gen_set_pc_im(s, s->pc - 4);
4562 tcg_reg = tcg_temp_new_i32();
4563 tcg_tgtmode = tcg_const_i32(tgtmode);
4564 tcg_regno = tcg_const_i32(regno);
4565 gen_helper_mrs_banked(tcg_reg, cpu_env, tcg_tgtmode, tcg_regno);
4566 tcg_temp_free_i32(tcg_tgtmode);
4567 tcg_temp_free_i32(tcg_regno);
4568 store_reg(s, rn, tcg_reg);
4569 s->base.is_jmp = DISAS_UPDATE;
4572 /* Store value to PC as for an exception return (ie don't
4573 * mask bits). The subsequent call to gen_helper_cpsr_write_eret()
4574 * will do the masking based on the new value of the Thumb bit.
4576 static void store_pc_exc_ret(DisasContext *s, TCGv_i32 pc)
4578 tcg_gen_mov_i32(cpu_R[15], pc);
4579 tcg_temp_free_i32(pc);
4582 /* Generate a v6 exception return. Marks both values as dead. */
4583 static void gen_rfe(DisasContext *s, TCGv_i32 pc, TCGv_i32 cpsr)
4585 store_pc_exc_ret(s, pc);
4586 /* The cpsr_write_eret helper will mask the low bits of PC
4587 * appropriately depending on the new Thumb bit, so it must
4588 * be called after storing the new PC.
4590 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
4593 gen_helper_cpsr_write_eret(cpu_env, cpsr);
4594 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
4597 tcg_temp_free_i32(cpsr);
4598 /* Must exit loop to check un-masked IRQs */
4599 s->base.is_jmp = DISAS_EXIT;
4602 /* Generate an old-style exception return. Marks pc as dead. */
4603 static void gen_exception_return(DisasContext *s, TCGv_i32 pc)
4605 gen_rfe(s, pc, load_cpu_field(spsr));
4609 * For WFI we will halt the vCPU until an IRQ. For WFE and YIELD we
4610 * only call the helper when running single threaded TCG code to ensure
4611 * the next round-robin scheduled vCPU gets a crack. In MTTCG mode we
4612 * just skip this instruction. Currently the SEV/SEVL instructions
4613 * which are *one* of many ways to wake the CPU from WFE are not
4614 * implemented so we can't sleep like WFI does.
4616 static void gen_nop_hint(DisasContext *s, int val)
4619 /* When running in MTTCG we don't generate jumps to the yield and
4620 * WFE helpers as it won't affect the scheduling of other vCPUs.
4621 * If we wanted to more completely model WFE/SEV so we don't busy
4622 * spin unnecessarily we would need to do something more involved.
4625 if (!(tb_cflags(s->base.tb) & CF_PARALLEL)) {
4626 gen_set_pc_im(s, s->pc);
4627 s->base.is_jmp = DISAS_YIELD;
4631 gen_set_pc_im(s, s->pc);
4632 s->base.is_jmp = DISAS_WFI;
4635 if (!(tb_cflags(s->base.tb) & CF_PARALLEL)) {
4636 gen_set_pc_im(s, s->pc);
4637 s->base.is_jmp = DISAS_WFE;
4642 /* TODO: Implement SEV, SEVL and WFE. May help SMP performance. */
4648 #define CPU_V001 cpu_V0, cpu_V0, cpu_V1
4650 static inline void gen_neon_add(int size, TCGv_i32 t0, TCGv_i32 t1)
4653 case 0: gen_helper_neon_add_u8(t0, t0, t1); break;
4654 case 1: gen_helper_neon_add_u16(t0, t0, t1); break;
4655 case 2: tcg_gen_add_i32(t0, t0, t1); break;
4660 static inline void gen_neon_rsb(int size, TCGv_i32 t0, TCGv_i32 t1)
4663 case 0: gen_helper_neon_sub_u8(t0, t1, t0); break;
4664 case 1: gen_helper_neon_sub_u16(t0, t1, t0); break;
4665 case 2: tcg_gen_sub_i32(t0, t1, t0); break;
4670 /* 32-bit pairwise ops end up the same as the elementwise versions. */
4671 #define gen_helper_neon_pmax_s32 gen_helper_neon_max_s32
4672 #define gen_helper_neon_pmax_u32 gen_helper_neon_max_u32
4673 #define gen_helper_neon_pmin_s32 gen_helper_neon_min_s32
4674 #define gen_helper_neon_pmin_u32 gen_helper_neon_min_u32
4676 #define GEN_NEON_INTEGER_OP_ENV(name) do { \
4677 switch ((size << 1) | u) { \
4679 gen_helper_neon_##name##_s8(tmp, cpu_env, tmp, tmp2); \
4682 gen_helper_neon_##name##_u8(tmp, cpu_env, tmp, tmp2); \
4685 gen_helper_neon_##name##_s16(tmp, cpu_env, tmp, tmp2); \
4688 gen_helper_neon_##name##_u16(tmp, cpu_env, tmp, tmp2); \
4691 gen_helper_neon_##name##_s32(tmp, cpu_env, tmp, tmp2); \
4694 gen_helper_neon_##name##_u32(tmp, cpu_env, tmp, tmp2); \
4696 default: return 1; \
4699 #define GEN_NEON_INTEGER_OP(name) do { \
4700 switch ((size << 1) | u) { \
4702 gen_helper_neon_##name##_s8(tmp, tmp, tmp2); \
4705 gen_helper_neon_##name##_u8(tmp, tmp, tmp2); \
4708 gen_helper_neon_##name##_s16(tmp, tmp, tmp2); \
4711 gen_helper_neon_##name##_u16(tmp, tmp, tmp2); \
4714 gen_helper_neon_##name##_s32(tmp, tmp, tmp2); \
4717 gen_helper_neon_##name##_u32(tmp, tmp, tmp2); \
4719 default: return 1; \
4722 static TCGv_i32 neon_load_scratch(int scratch)
4724 TCGv_i32 tmp = tcg_temp_new_i32();
4725 tcg_gen_ld_i32(tmp, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
4729 static void neon_store_scratch(int scratch, TCGv_i32 var)
4731 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
4732 tcg_temp_free_i32(var);
4735 static inline TCGv_i32 neon_get_scalar(int size, int reg)
4739 tmp = neon_load_reg(reg & 7, reg >> 4);
4741 gen_neon_dup_high16(tmp);
4743 gen_neon_dup_low16(tmp);
4746 tmp = neon_load_reg(reg & 15, reg >> 4);
4751 static int gen_neon_unzip(int rd, int rm, int size, int q)
4755 if (!q && size == 2) {
4758 pd = vfp_reg_ptr(true, rd);
4759 pm = vfp_reg_ptr(true, rm);
4763 gen_helper_neon_qunzip8(pd, pm);
4766 gen_helper_neon_qunzip16(pd, pm);
4769 gen_helper_neon_qunzip32(pd, pm);
4777 gen_helper_neon_unzip8(pd, pm);
4780 gen_helper_neon_unzip16(pd, pm);
4786 tcg_temp_free_ptr(pd);
4787 tcg_temp_free_ptr(pm);
4791 static int gen_neon_zip(int rd, int rm, int size, int q)
4795 if (!q && size == 2) {
4798 pd = vfp_reg_ptr(true, rd);
4799 pm = vfp_reg_ptr(true, rm);
4803 gen_helper_neon_qzip8(pd, pm);
4806 gen_helper_neon_qzip16(pd, pm);
4809 gen_helper_neon_qzip32(pd, pm);
4817 gen_helper_neon_zip8(pd, pm);
4820 gen_helper_neon_zip16(pd, pm);
4826 tcg_temp_free_ptr(pd);
4827 tcg_temp_free_ptr(pm);
4831 static void gen_neon_trn_u8(TCGv_i32 t0, TCGv_i32 t1)
4835 rd = tcg_temp_new_i32();
4836 tmp = tcg_temp_new_i32();
4838 tcg_gen_shli_i32(rd, t0, 8);
4839 tcg_gen_andi_i32(rd, rd, 0xff00ff00);
4840 tcg_gen_andi_i32(tmp, t1, 0x00ff00ff);
4841 tcg_gen_or_i32(rd, rd, tmp);
4843 tcg_gen_shri_i32(t1, t1, 8);
4844 tcg_gen_andi_i32(t1, t1, 0x00ff00ff);
4845 tcg_gen_andi_i32(tmp, t0, 0xff00ff00);
4846 tcg_gen_or_i32(t1, t1, tmp);
4847 tcg_gen_mov_i32(t0, rd);
4849 tcg_temp_free_i32(tmp);
4850 tcg_temp_free_i32(rd);
4853 static void gen_neon_trn_u16(TCGv_i32 t0, TCGv_i32 t1)
4857 rd = tcg_temp_new_i32();
4858 tmp = tcg_temp_new_i32();
4860 tcg_gen_shli_i32(rd, t0, 16);
4861 tcg_gen_andi_i32(tmp, t1, 0xffff);
4862 tcg_gen_or_i32(rd, rd, tmp);
4863 tcg_gen_shri_i32(t1, t1, 16);
4864 tcg_gen_andi_i32(tmp, t0, 0xffff0000);
4865 tcg_gen_or_i32(t1, t1, tmp);
4866 tcg_gen_mov_i32(t0, rd);
4868 tcg_temp_free_i32(tmp);
4869 tcg_temp_free_i32(rd);
4877 } neon_ls_element_type[11] = {
4891 /* Translate a NEON load/store element instruction. Return nonzero if the
4892 instruction is invalid. */
4893 static int disas_neon_ls_insn(DisasContext *s, uint32_t insn)
4912 /* FIXME: this access check should not take precedence over UNDEF
4913 * for invalid encodings; we will generate incorrect syndrome information
4914 * for attempts to execute invalid vfp/neon encodings with FP disabled.
4916 if (s->fp_excp_el) {
4917 gen_exception_insn(s, 4, EXCP_UDEF,
4918 syn_fp_access_trap(1, 0xe, false), s->fp_excp_el);
4922 if (!s->vfp_enabled)
4924 VFP_DREG_D(rd, insn);
4925 rn = (insn >> 16) & 0xf;
4927 load = (insn & (1 << 21)) != 0;
4928 if ((insn & (1 << 23)) == 0) {
4929 /* Load store all elements. */
4930 op = (insn >> 8) & 0xf;
4931 size = (insn >> 6) & 3;
4934 /* Catch UNDEF cases for bad values of align field */
4937 if (((insn >> 5) & 1) == 1) {
4942 if (((insn >> 4) & 3) == 3) {
4949 nregs = neon_ls_element_type[op].nregs;
4950 interleave = neon_ls_element_type[op].interleave;
4951 spacing = neon_ls_element_type[op].spacing;
4952 if (size == 3 && (interleave | spacing) != 1)
4954 addr = tcg_temp_new_i32();
4955 load_reg_var(s, addr, rn);
4956 stride = (1 << size) * interleave;
4957 for (reg = 0; reg < nregs; reg++) {
4958 if (interleave > 2 || (interleave == 2 && nregs == 2)) {
4959 load_reg_var(s, addr, rn);
4960 tcg_gen_addi_i32(addr, addr, (1 << size) * reg);
4961 } else if (interleave == 2 && nregs == 4 && reg == 2) {
4962 load_reg_var(s, addr, rn);
4963 tcg_gen_addi_i32(addr, addr, 1 << size);
4966 tmp64 = tcg_temp_new_i64();
4968 gen_aa32_ld64(s, tmp64, addr, get_mem_index(s));
4969 neon_store_reg64(tmp64, rd);
4971 neon_load_reg64(tmp64, rd);
4972 gen_aa32_st64(s, tmp64, addr, get_mem_index(s));
4974 tcg_temp_free_i64(tmp64);
4975 tcg_gen_addi_i32(addr, addr, stride);
4977 for (pass = 0; pass < 2; pass++) {
4980 tmp = tcg_temp_new_i32();
4981 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
4982 neon_store_reg(rd, pass, tmp);
4984 tmp = neon_load_reg(rd, pass);
4985 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
4986 tcg_temp_free_i32(tmp);
4988 tcg_gen_addi_i32(addr, addr, stride);
4989 } else if (size == 1) {
4991 tmp = tcg_temp_new_i32();
4992 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
4993 tcg_gen_addi_i32(addr, addr, stride);
4994 tmp2 = tcg_temp_new_i32();
4995 gen_aa32_ld16u(s, tmp2, addr, get_mem_index(s));
4996 tcg_gen_addi_i32(addr, addr, stride);
4997 tcg_gen_shli_i32(tmp2, tmp2, 16);
4998 tcg_gen_or_i32(tmp, tmp, tmp2);
4999 tcg_temp_free_i32(tmp2);
5000 neon_store_reg(rd, pass, tmp);
5002 tmp = neon_load_reg(rd, pass);
5003 tmp2 = tcg_temp_new_i32();
5004 tcg_gen_shri_i32(tmp2, tmp, 16);
5005 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
5006 tcg_temp_free_i32(tmp);
5007 tcg_gen_addi_i32(addr, addr, stride);
5008 gen_aa32_st16(s, tmp2, addr, get_mem_index(s));
5009 tcg_temp_free_i32(tmp2);
5010 tcg_gen_addi_i32(addr, addr, stride);
5012 } else /* size == 0 */ {
5015 for (n = 0; n < 4; n++) {
5016 tmp = tcg_temp_new_i32();
5017 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
5018 tcg_gen_addi_i32(addr, addr, stride);
5022 tcg_gen_shli_i32(tmp, tmp, n * 8);
5023 tcg_gen_or_i32(tmp2, tmp2, tmp);
5024 tcg_temp_free_i32(tmp);
5027 neon_store_reg(rd, pass, tmp2);
5029 tmp2 = neon_load_reg(rd, pass);
5030 for (n = 0; n < 4; n++) {
5031 tmp = tcg_temp_new_i32();
5033 tcg_gen_mov_i32(tmp, tmp2);
5035 tcg_gen_shri_i32(tmp, tmp2, n * 8);
5037 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
5038 tcg_temp_free_i32(tmp);
5039 tcg_gen_addi_i32(addr, addr, stride);
5041 tcg_temp_free_i32(tmp2);
5048 tcg_temp_free_i32(addr);
5051 size = (insn >> 10) & 3;
5053 /* Load single element to all lanes. */
5054 int a = (insn >> 4) & 1;
5058 size = (insn >> 6) & 3;
5059 nregs = ((insn >> 8) & 3) + 1;
5062 if (nregs != 4 || a == 0) {
5065 /* For VLD4 size==3 a == 1 means 32 bits at 16 byte alignment */
5068 if (nregs == 1 && a == 1 && size == 0) {
5071 if (nregs == 3 && a == 1) {
5074 addr = tcg_temp_new_i32();
5075 load_reg_var(s, addr, rn);
5077 /* VLD1 to all lanes: bit 5 indicates how many Dregs to write */
5078 tmp = gen_load_and_replicate(s, addr, size);
5079 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 0));
5080 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 1));
5081 if (insn & (1 << 5)) {
5082 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd + 1, 0));
5083 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd + 1, 1));
5085 tcg_temp_free_i32(tmp);
5087 /* VLD2/3/4 to all lanes: bit 5 indicates register stride */
5088 stride = (insn & (1 << 5)) ? 2 : 1;
5089 for (reg = 0; reg < nregs; reg++) {
5090 tmp = gen_load_and_replicate(s, addr, size);
5091 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 0));
5092 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 1));
5093 tcg_temp_free_i32(tmp);
5094 tcg_gen_addi_i32(addr, addr, 1 << size);
5098 tcg_temp_free_i32(addr);
5099 stride = (1 << size) * nregs;
5101 /* Single element. */
5102 int idx = (insn >> 4) & 0xf;
5103 pass = (insn >> 7) & 1;
5106 shift = ((insn >> 5) & 3) * 8;
5110 shift = ((insn >> 6) & 1) * 16;
5111 stride = (insn & (1 << 5)) ? 2 : 1;
5115 stride = (insn & (1 << 6)) ? 2 : 1;
5120 nregs = ((insn >> 8) & 3) + 1;
5121 /* Catch the UNDEF cases. This is unavoidably a bit messy. */
5124 if (((idx & (1 << size)) != 0) ||
5125 (size == 2 && ((idx & 3) == 1 || (idx & 3) == 2))) {
5130 if ((idx & 1) != 0) {
5135 if (size == 2 && (idx & 2) != 0) {
5140 if ((size == 2) && ((idx & 3) == 3)) {
5147 if ((rd + stride * (nregs - 1)) > 31) {
5148 /* Attempts to write off the end of the register file
5149 * are UNPREDICTABLE; we choose to UNDEF because otherwise
5150 * the neon_load_reg() would write off the end of the array.
5154 addr = tcg_temp_new_i32();
5155 load_reg_var(s, addr, rn);
5156 for (reg = 0; reg < nregs; reg++) {
5158 tmp = tcg_temp_new_i32();
5161 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
5164 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
5167 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
5169 default: /* Avoid compiler warnings. */
5173 tmp2 = neon_load_reg(rd, pass);
5174 tcg_gen_deposit_i32(tmp, tmp2, tmp,
5175 shift, size ? 16 : 8);
5176 tcg_temp_free_i32(tmp2);
5178 neon_store_reg(rd, pass, tmp);
5179 } else { /* Store */
5180 tmp = neon_load_reg(rd, pass);
5182 tcg_gen_shri_i32(tmp, tmp, shift);
5185 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
5188 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
5191 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
5194 tcg_temp_free_i32(tmp);
5197 tcg_gen_addi_i32(addr, addr, 1 << size);
5199 tcg_temp_free_i32(addr);
5200 stride = nregs * (1 << size);
5206 base = load_reg(s, rn);
5208 tcg_gen_addi_i32(base, base, stride);
5211 index = load_reg(s, rm);
5212 tcg_gen_add_i32(base, base, index);
5213 tcg_temp_free_i32(index);
5215 store_reg(s, rn, base);
5220 /* Bitwise select. dest = c ? t : f. Clobbers T and F. */
5221 static void gen_neon_bsl(TCGv_i32 dest, TCGv_i32 t, TCGv_i32 f, TCGv_i32 c)
5223 tcg_gen_and_i32(t, t, c);
5224 tcg_gen_andc_i32(f, f, c);
5225 tcg_gen_or_i32(dest, t, f);
5228 static inline void gen_neon_narrow(int size, TCGv_i32 dest, TCGv_i64 src)
5231 case 0: gen_helper_neon_narrow_u8(dest, src); break;
5232 case 1: gen_helper_neon_narrow_u16(dest, src); break;
5233 case 2: tcg_gen_extrl_i64_i32(dest, src); break;
5238 static inline void gen_neon_narrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
5241 case 0: gen_helper_neon_narrow_sat_s8(dest, cpu_env, src); break;
5242 case 1: gen_helper_neon_narrow_sat_s16(dest, cpu_env, src); break;
5243 case 2: gen_helper_neon_narrow_sat_s32(dest, cpu_env, src); break;
5248 static inline void gen_neon_narrow_satu(int size, TCGv_i32 dest, TCGv_i64 src)
5251 case 0: gen_helper_neon_narrow_sat_u8(dest, cpu_env, src); break;
5252 case 1: gen_helper_neon_narrow_sat_u16(dest, cpu_env, src); break;
5253 case 2: gen_helper_neon_narrow_sat_u32(dest, cpu_env, src); break;
5258 static inline void gen_neon_unarrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
5261 case 0: gen_helper_neon_unarrow_sat8(dest, cpu_env, src); break;
5262 case 1: gen_helper_neon_unarrow_sat16(dest, cpu_env, src); break;
5263 case 2: gen_helper_neon_unarrow_sat32(dest, cpu_env, src); break;
5268 static inline void gen_neon_shift_narrow(int size, TCGv_i32 var, TCGv_i32 shift,
5274 case 1: gen_helper_neon_rshl_u16(var, var, shift); break;
5275 case 2: gen_helper_neon_rshl_u32(var, var, shift); break;
5280 case 1: gen_helper_neon_rshl_s16(var, var, shift); break;
5281 case 2: gen_helper_neon_rshl_s32(var, var, shift); break;
5288 case 1: gen_helper_neon_shl_u16(var, var, shift); break;
5289 case 2: gen_helper_neon_shl_u32(var, var, shift); break;
5294 case 1: gen_helper_neon_shl_s16(var, var, shift); break;
5295 case 2: gen_helper_neon_shl_s32(var, var, shift); break;
5302 static inline void gen_neon_widen(TCGv_i64 dest, TCGv_i32 src, int size, int u)
5306 case 0: gen_helper_neon_widen_u8(dest, src); break;
5307 case 1: gen_helper_neon_widen_u16(dest, src); break;
5308 case 2: tcg_gen_extu_i32_i64(dest, src); break;
5313 case 0: gen_helper_neon_widen_s8(dest, src); break;
5314 case 1: gen_helper_neon_widen_s16(dest, src); break;
5315 case 2: tcg_gen_ext_i32_i64(dest, src); break;
5319 tcg_temp_free_i32(src);
5322 static inline void gen_neon_addl(int size)
5325 case 0: gen_helper_neon_addl_u16(CPU_V001); break;
5326 case 1: gen_helper_neon_addl_u32(CPU_V001); break;
5327 case 2: tcg_gen_add_i64(CPU_V001); break;
5332 static inline void gen_neon_subl(int size)
5335 case 0: gen_helper_neon_subl_u16(CPU_V001); break;
5336 case 1: gen_helper_neon_subl_u32(CPU_V001); break;
5337 case 2: tcg_gen_sub_i64(CPU_V001); break;
5342 static inline void gen_neon_negl(TCGv_i64 var, int size)
5345 case 0: gen_helper_neon_negl_u16(var, var); break;
5346 case 1: gen_helper_neon_negl_u32(var, var); break;
5348 tcg_gen_neg_i64(var, var);
5354 static inline void gen_neon_addl_saturate(TCGv_i64 op0, TCGv_i64 op1, int size)
5357 case 1: gen_helper_neon_addl_saturate_s32(op0, cpu_env, op0, op1); break;
5358 case 2: gen_helper_neon_addl_saturate_s64(op0, cpu_env, op0, op1); break;
5363 static inline void gen_neon_mull(TCGv_i64 dest, TCGv_i32 a, TCGv_i32 b,
5368 switch ((size << 1) | u) {
5369 case 0: gen_helper_neon_mull_s8(dest, a, b); break;
5370 case 1: gen_helper_neon_mull_u8(dest, a, b); break;
5371 case 2: gen_helper_neon_mull_s16(dest, a, b); break;
5372 case 3: gen_helper_neon_mull_u16(dest, a, b); break;
5374 tmp = gen_muls_i64_i32(a, b);
5375 tcg_gen_mov_i64(dest, tmp);
5376 tcg_temp_free_i64(tmp);
5379 tmp = gen_mulu_i64_i32(a, b);
5380 tcg_gen_mov_i64(dest, tmp);
5381 tcg_temp_free_i64(tmp);
5386 /* gen_helper_neon_mull_[su]{8|16} do not free their parameters.
5387 Don't forget to clean them now. */
5389 tcg_temp_free_i32(a);
5390 tcg_temp_free_i32(b);
5394 static void gen_neon_narrow_op(int op, int u, int size,
5395 TCGv_i32 dest, TCGv_i64 src)
5399 gen_neon_unarrow_sats(size, dest, src);
5401 gen_neon_narrow(size, dest, src);
5405 gen_neon_narrow_satu(size, dest, src);
5407 gen_neon_narrow_sats(size, dest, src);
5412 /* Symbolic constants for op fields for Neon 3-register same-length.
5413 * The values correspond to bits [11:8,4]; see the ARM ARM DDI0406B
5416 #define NEON_3R_VHADD 0
5417 #define NEON_3R_VQADD 1
5418 #define NEON_3R_VRHADD 2
5419 #define NEON_3R_LOGIC 3 /* VAND,VBIC,VORR,VMOV,VORN,VEOR,VBIF,VBIT,VBSL */
5420 #define NEON_3R_VHSUB 4
5421 #define NEON_3R_VQSUB 5
5422 #define NEON_3R_VCGT 6
5423 #define NEON_3R_VCGE 7
5424 #define NEON_3R_VSHL 8
5425 #define NEON_3R_VQSHL 9
5426 #define NEON_3R_VRSHL 10
5427 #define NEON_3R_VQRSHL 11
5428 #define NEON_3R_VMAX 12
5429 #define NEON_3R_VMIN 13
5430 #define NEON_3R_VABD 14
5431 #define NEON_3R_VABA 15
5432 #define NEON_3R_VADD_VSUB 16
5433 #define NEON_3R_VTST_VCEQ 17
5434 #define NEON_3R_VML 18 /* VMLA, VMLAL, VMLS, VMLSL */
5435 #define NEON_3R_VMUL 19
5436 #define NEON_3R_VPMAX 20
5437 #define NEON_3R_VPMIN 21
5438 #define NEON_3R_VQDMULH_VQRDMULH 22
5439 #define NEON_3R_VPADD_VQRDMLAH 23
5440 #define NEON_3R_SHA 24 /* SHA1C,SHA1P,SHA1M,SHA1SU0,SHA256H{2},SHA256SU1 */
5441 #define NEON_3R_VFM_VQRDMLSH 25 /* VFMA, VFMS, VQRDMLSH */
5442 #define NEON_3R_FLOAT_ARITH 26 /* float VADD, VSUB, VPADD, VABD */
5443 #define NEON_3R_FLOAT_MULTIPLY 27 /* float VMLA, VMLS, VMUL */
5444 #define NEON_3R_FLOAT_CMP 28 /* float VCEQ, VCGE, VCGT */
5445 #define NEON_3R_FLOAT_ACMP 29 /* float VACGE, VACGT, VACLE, VACLT */
5446 #define NEON_3R_FLOAT_MINMAX 30 /* float VMIN, VMAX */
5447 #define NEON_3R_FLOAT_MISC 31 /* float VRECPS, VRSQRTS, VMAXNM/MINNM */
5449 static const uint8_t neon_3r_sizes[] = {
5450 [NEON_3R_VHADD] = 0x7,
5451 [NEON_3R_VQADD] = 0xf,
5452 [NEON_3R_VRHADD] = 0x7,
5453 [NEON_3R_LOGIC] = 0xf, /* size field encodes op type */
5454 [NEON_3R_VHSUB] = 0x7,
5455 [NEON_3R_VQSUB] = 0xf,
5456 [NEON_3R_VCGT] = 0x7,
5457 [NEON_3R_VCGE] = 0x7,
5458 [NEON_3R_VSHL] = 0xf,
5459 [NEON_3R_VQSHL] = 0xf,
5460 [NEON_3R_VRSHL] = 0xf,
5461 [NEON_3R_VQRSHL] = 0xf,
5462 [NEON_3R_VMAX] = 0x7,
5463 [NEON_3R_VMIN] = 0x7,
5464 [NEON_3R_VABD] = 0x7,
5465 [NEON_3R_VABA] = 0x7,
5466 [NEON_3R_VADD_VSUB] = 0xf,
5467 [NEON_3R_VTST_VCEQ] = 0x7,
5468 [NEON_3R_VML] = 0x7,
5469 [NEON_3R_VMUL] = 0x7,
5470 [NEON_3R_VPMAX] = 0x7,
5471 [NEON_3R_VPMIN] = 0x7,
5472 [NEON_3R_VQDMULH_VQRDMULH] = 0x6,
5473 [NEON_3R_VPADD_VQRDMLAH] = 0x7,
5474 [NEON_3R_SHA] = 0xf, /* size field encodes op type */
5475 [NEON_3R_VFM_VQRDMLSH] = 0x7, /* For VFM, size bit 1 encodes op */
5476 [NEON_3R_FLOAT_ARITH] = 0x5, /* size bit 1 encodes op */
5477 [NEON_3R_FLOAT_MULTIPLY] = 0x5, /* size bit 1 encodes op */
5478 [NEON_3R_FLOAT_CMP] = 0x5, /* size bit 1 encodes op */
5479 [NEON_3R_FLOAT_ACMP] = 0x5, /* size bit 1 encodes op */
5480 [NEON_3R_FLOAT_MINMAX] = 0x5, /* size bit 1 encodes op */
5481 [NEON_3R_FLOAT_MISC] = 0x5, /* size bit 1 encodes op */
5484 /* Symbolic constants for op fields for Neon 2-register miscellaneous.
5485 * The values correspond to bits [17:16,10:7]; see the ARM ARM DDI0406B
5488 #define NEON_2RM_VREV64 0
5489 #define NEON_2RM_VREV32 1
5490 #define NEON_2RM_VREV16 2
5491 #define NEON_2RM_VPADDL 4
5492 #define NEON_2RM_VPADDL_U 5
5493 #define NEON_2RM_AESE 6 /* Includes AESD */
5494 #define NEON_2RM_AESMC 7 /* Includes AESIMC */
5495 #define NEON_2RM_VCLS 8
5496 #define NEON_2RM_VCLZ 9
5497 #define NEON_2RM_VCNT 10
5498 #define NEON_2RM_VMVN 11
5499 #define NEON_2RM_VPADAL 12
5500 #define NEON_2RM_VPADAL_U 13
5501 #define NEON_2RM_VQABS 14
5502 #define NEON_2RM_VQNEG 15
5503 #define NEON_2RM_VCGT0 16
5504 #define NEON_2RM_VCGE0 17
5505 #define NEON_2RM_VCEQ0 18
5506 #define NEON_2RM_VCLE0 19
5507 #define NEON_2RM_VCLT0 20
5508 #define NEON_2RM_SHA1H 21
5509 #define NEON_2RM_VABS 22
5510 #define NEON_2RM_VNEG 23
5511 #define NEON_2RM_VCGT0_F 24
5512 #define NEON_2RM_VCGE0_F 25
5513 #define NEON_2RM_VCEQ0_F 26
5514 #define NEON_2RM_VCLE0_F 27
5515 #define NEON_2RM_VCLT0_F 28
5516 #define NEON_2RM_VABS_F 30
5517 #define NEON_2RM_VNEG_F 31
5518 #define NEON_2RM_VSWP 32
5519 #define NEON_2RM_VTRN 33
5520 #define NEON_2RM_VUZP 34
5521 #define NEON_2RM_VZIP 35
5522 #define NEON_2RM_VMOVN 36 /* Includes VQMOVN, VQMOVUN */
5523 #define NEON_2RM_VQMOVN 37 /* Includes VQMOVUN */
5524 #define NEON_2RM_VSHLL 38
5525 #define NEON_2RM_SHA1SU1 39 /* Includes SHA256SU0 */
5526 #define NEON_2RM_VRINTN 40
5527 #define NEON_2RM_VRINTX 41
5528 #define NEON_2RM_VRINTA 42
5529 #define NEON_2RM_VRINTZ 43
5530 #define NEON_2RM_VCVT_F16_F32 44
5531 #define NEON_2RM_VRINTM 45
5532 #define NEON_2RM_VCVT_F32_F16 46
5533 #define NEON_2RM_VRINTP 47
5534 #define NEON_2RM_VCVTAU 48
5535 #define NEON_2RM_VCVTAS 49
5536 #define NEON_2RM_VCVTNU 50
5537 #define NEON_2RM_VCVTNS 51
5538 #define NEON_2RM_VCVTPU 52
5539 #define NEON_2RM_VCVTPS 53
5540 #define NEON_2RM_VCVTMU 54
5541 #define NEON_2RM_VCVTMS 55
5542 #define NEON_2RM_VRECPE 56
5543 #define NEON_2RM_VRSQRTE 57
5544 #define NEON_2RM_VRECPE_F 58
5545 #define NEON_2RM_VRSQRTE_F 59
5546 #define NEON_2RM_VCVT_FS 60
5547 #define NEON_2RM_VCVT_FU 61
5548 #define NEON_2RM_VCVT_SF 62
5549 #define NEON_2RM_VCVT_UF 63
5551 static int neon_2rm_is_float_op(int op)
5553 /* Return true if this neon 2reg-misc op is float-to-float */
5554 return (op == NEON_2RM_VABS_F || op == NEON_2RM_VNEG_F ||
5555 (op >= NEON_2RM_VRINTN && op <= NEON_2RM_VRINTZ) ||
5556 op == NEON_2RM_VRINTM ||
5557 (op >= NEON_2RM_VRINTP && op <= NEON_2RM_VCVTMS) ||
5558 op >= NEON_2RM_VRECPE_F);
5561 static bool neon_2rm_is_v8_op(int op)
5563 /* Return true if this neon 2reg-misc op is ARMv8 and up */
5565 case NEON_2RM_VRINTN:
5566 case NEON_2RM_VRINTA:
5567 case NEON_2RM_VRINTM:
5568 case NEON_2RM_VRINTP:
5569 case NEON_2RM_VRINTZ:
5570 case NEON_2RM_VRINTX:
5571 case NEON_2RM_VCVTAU:
5572 case NEON_2RM_VCVTAS:
5573 case NEON_2RM_VCVTNU:
5574 case NEON_2RM_VCVTNS:
5575 case NEON_2RM_VCVTPU:
5576 case NEON_2RM_VCVTPS:
5577 case NEON_2RM_VCVTMU:
5578 case NEON_2RM_VCVTMS:
5585 /* Each entry in this array has bit n set if the insn allows
5586 * size value n (otherwise it will UNDEF). Since unallocated
5587 * op values will have no bits set they always UNDEF.
5589 static const uint8_t neon_2rm_sizes[] = {
5590 [NEON_2RM_VREV64] = 0x7,
5591 [NEON_2RM_VREV32] = 0x3,
5592 [NEON_2RM_VREV16] = 0x1,
5593 [NEON_2RM_VPADDL] = 0x7,
5594 [NEON_2RM_VPADDL_U] = 0x7,
5595 [NEON_2RM_AESE] = 0x1,
5596 [NEON_2RM_AESMC] = 0x1,
5597 [NEON_2RM_VCLS] = 0x7,
5598 [NEON_2RM_VCLZ] = 0x7,
5599 [NEON_2RM_VCNT] = 0x1,
5600 [NEON_2RM_VMVN] = 0x1,
5601 [NEON_2RM_VPADAL] = 0x7,
5602 [NEON_2RM_VPADAL_U] = 0x7,
5603 [NEON_2RM_VQABS] = 0x7,
5604 [NEON_2RM_VQNEG] = 0x7,
5605 [NEON_2RM_VCGT0] = 0x7,
5606 [NEON_2RM_VCGE0] = 0x7,
5607 [NEON_2RM_VCEQ0] = 0x7,
5608 [NEON_2RM_VCLE0] = 0x7,
5609 [NEON_2RM_VCLT0] = 0x7,
5610 [NEON_2RM_SHA1H] = 0x4,
5611 [NEON_2RM_VABS] = 0x7,
5612 [NEON_2RM_VNEG] = 0x7,
5613 [NEON_2RM_VCGT0_F] = 0x4,
5614 [NEON_2RM_VCGE0_F] = 0x4,
5615 [NEON_2RM_VCEQ0_F] = 0x4,
5616 [NEON_2RM_VCLE0_F] = 0x4,
5617 [NEON_2RM_VCLT0_F] = 0x4,
5618 [NEON_2RM_VABS_F] = 0x4,
5619 [NEON_2RM_VNEG_F] = 0x4,
5620 [NEON_2RM_VSWP] = 0x1,
5621 [NEON_2RM_VTRN] = 0x7,
5622 [NEON_2RM_VUZP] = 0x7,
5623 [NEON_2RM_VZIP] = 0x7,
5624 [NEON_2RM_VMOVN] = 0x7,
5625 [NEON_2RM_VQMOVN] = 0x7,
5626 [NEON_2RM_VSHLL] = 0x7,
5627 [NEON_2RM_SHA1SU1] = 0x4,
5628 [NEON_2RM_VRINTN] = 0x4,
5629 [NEON_2RM_VRINTX] = 0x4,
5630 [NEON_2RM_VRINTA] = 0x4,
5631 [NEON_2RM_VRINTZ] = 0x4,
5632 [NEON_2RM_VCVT_F16_F32] = 0x2,
5633 [NEON_2RM_VRINTM] = 0x4,
5634 [NEON_2RM_VCVT_F32_F16] = 0x2,
5635 [NEON_2RM_VRINTP] = 0x4,
5636 [NEON_2RM_VCVTAU] = 0x4,
5637 [NEON_2RM_VCVTAS] = 0x4,
5638 [NEON_2RM_VCVTNU] = 0x4,
5639 [NEON_2RM_VCVTNS] = 0x4,
5640 [NEON_2RM_VCVTPU] = 0x4,
5641 [NEON_2RM_VCVTPS] = 0x4,
5642 [NEON_2RM_VCVTMU] = 0x4,
5643 [NEON_2RM_VCVTMS] = 0x4,
5644 [NEON_2RM_VRECPE] = 0x4,
5645 [NEON_2RM_VRSQRTE] = 0x4,
5646 [NEON_2RM_VRECPE_F] = 0x4,
5647 [NEON_2RM_VRSQRTE_F] = 0x4,
5648 [NEON_2RM_VCVT_FS] = 0x4,
5649 [NEON_2RM_VCVT_FU] = 0x4,
5650 [NEON_2RM_VCVT_SF] = 0x4,
5651 [NEON_2RM_VCVT_UF] = 0x4,
5655 /* Expand v8.1 simd helper. */
5656 static int do_v81_helper(DisasContext *s, gen_helper_gvec_3_ptr *fn,
5657 int q, int rd, int rn, int rm)
5659 if (arm_dc_feature(s, ARM_FEATURE_V8_RDM)) {
5660 int opr_sz = (1 + q) * 8;
5661 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, rd),
5662 vfp_reg_offset(1, rn),
5663 vfp_reg_offset(1, rm), cpu_env,
5664 opr_sz, opr_sz, 0, fn);
5670 /* Translate a NEON data processing instruction. Return nonzero if the
5671 instruction is invalid.
5672 We process data in a mixture of 32-bit and 64-bit chunks.
5673 Mostly we use 32-bit chunks so we can use normal scalar instructions. */
5675 static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
5687 TCGv_i32 tmp, tmp2, tmp3, tmp4, tmp5;
5688 TCGv_ptr ptr1, ptr2, ptr3;
5691 /* FIXME: this access check should not take precedence over UNDEF
5692 * for invalid encodings; we will generate incorrect syndrome information
5693 * for attempts to execute invalid vfp/neon encodings with FP disabled.
5695 if (s->fp_excp_el) {
5696 gen_exception_insn(s, 4, EXCP_UDEF,
5697 syn_fp_access_trap(1, 0xe, false), s->fp_excp_el);
5701 if (!s->vfp_enabled)
5703 q = (insn & (1 << 6)) != 0;
5704 u = (insn >> 24) & 1;
5705 VFP_DREG_D(rd, insn);
5706 VFP_DREG_N(rn, insn);
5707 VFP_DREG_M(rm, insn);
5708 size = (insn >> 20) & 3;
5709 if ((insn & (1 << 23)) == 0) {
5710 /* Three register same length. */
5711 op = ((insn >> 7) & 0x1e) | ((insn >> 4) & 1);
5712 /* Catch invalid op and bad size combinations: UNDEF */
5713 if ((neon_3r_sizes[op] & (1 << size)) == 0) {
5716 /* All insns of this form UNDEF for either this condition or the
5717 * superset of cases "Q==1"; we catch the latter later.
5719 if (q && ((rd | rn | rm) & 1)) {
5724 /* The SHA-1/SHA-256 3-register instructions require special
5725 * treatment here, as their size field is overloaded as an
5726 * op type selector, and they all consume their input in a
5732 if (!u) { /* SHA-1 */
5733 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA1)) {
5736 ptr1 = vfp_reg_ptr(true, rd);
5737 ptr2 = vfp_reg_ptr(true, rn);
5738 ptr3 = vfp_reg_ptr(true, rm);
5739 tmp4 = tcg_const_i32(size);
5740 gen_helper_crypto_sha1_3reg(ptr1, ptr2, ptr3, tmp4);
5741 tcg_temp_free_i32(tmp4);
5742 } else { /* SHA-256 */
5743 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA256) || size == 3) {
5746 ptr1 = vfp_reg_ptr(true, rd);
5747 ptr2 = vfp_reg_ptr(true, rn);
5748 ptr3 = vfp_reg_ptr(true, rm);
5751 gen_helper_crypto_sha256h(ptr1, ptr2, ptr3);
5754 gen_helper_crypto_sha256h2(ptr1, ptr2, ptr3);
5757 gen_helper_crypto_sha256su1(ptr1, ptr2, ptr3);
5761 tcg_temp_free_ptr(ptr1);
5762 tcg_temp_free_ptr(ptr2);
5763 tcg_temp_free_ptr(ptr3);
5766 case NEON_3R_VPADD_VQRDMLAH:
5773 return do_v81_helper(s, gen_helper_gvec_qrdmlah_s16,
5776 return do_v81_helper(s, gen_helper_gvec_qrdmlah_s32,
5781 case NEON_3R_VFM_VQRDMLSH:
5792 return do_v81_helper(s, gen_helper_gvec_qrdmlsh_s16,
5795 return do_v81_helper(s, gen_helper_gvec_qrdmlsh_s32,
5800 if (size == 3 && op != NEON_3R_LOGIC) {
5801 /* 64-bit element instructions. */
5802 for (pass = 0; pass < (q ? 2 : 1); pass++) {
5803 neon_load_reg64(cpu_V0, rn + pass);
5804 neon_load_reg64(cpu_V1, rm + pass);
5808 gen_helper_neon_qadd_u64(cpu_V0, cpu_env,
5811 gen_helper_neon_qadd_s64(cpu_V0, cpu_env,
5817 gen_helper_neon_qsub_u64(cpu_V0, cpu_env,
5820 gen_helper_neon_qsub_s64(cpu_V0, cpu_env,
5826 gen_helper_neon_shl_u64(cpu_V0, cpu_V1, cpu_V0);
5828 gen_helper_neon_shl_s64(cpu_V0, cpu_V1, cpu_V0);
5833 gen_helper_neon_qshl_u64(cpu_V0, cpu_env,
5836 gen_helper_neon_qshl_s64(cpu_V0, cpu_env,
5842 gen_helper_neon_rshl_u64(cpu_V0, cpu_V1, cpu_V0);
5844 gen_helper_neon_rshl_s64(cpu_V0, cpu_V1, cpu_V0);
5847 case NEON_3R_VQRSHL:
5849 gen_helper_neon_qrshl_u64(cpu_V0, cpu_env,
5852 gen_helper_neon_qrshl_s64(cpu_V0, cpu_env,
5856 case NEON_3R_VADD_VSUB:
5858 tcg_gen_sub_i64(CPU_V001);
5860 tcg_gen_add_i64(CPU_V001);
5866 neon_store_reg64(cpu_V0, rd + pass);
5875 case NEON_3R_VQRSHL:
5878 /* Shift instruction operands are reversed. */
5884 case NEON_3R_VPADD_VQRDMLAH:
5889 case NEON_3R_FLOAT_ARITH:
5890 pairwise = (u && size < 2); /* if VPADD (float) */
5892 case NEON_3R_FLOAT_MINMAX:
5893 pairwise = u; /* if VPMIN/VPMAX (float) */
5895 case NEON_3R_FLOAT_CMP:
5897 /* no encoding for U=0 C=1x */
5901 case NEON_3R_FLOAT_ACMP:
5906 case NEON_3R_FLOAT_MISC:
5907 /* VMAXNM/VMINNM in ARMv8 */
5908 if (u && !arm_dc_feature(s, ARM_FEATURE_V8)) {
5913 if (u && (size != 0)) {
5914 /* UNDEF on invalid size for polynomial subcase */
5918 case NEON_3R_VFM_VQRDMLSH:
5919 if (!arm_dc_feature(s, ARM_FEATURE_VFP4)) {
5927 if (pairwise && q) {
5928 /* All the pairwise insns UNDEF if Q is set */
5932 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5937 tmp = neon_load_reg(rn, 0);
5938 tmp2 = neon_load_reg(rn, 1);
5940 tmp = neon_load_reg(rm, 0);
5941 tmp2 = neon_load_reg(rm, 1);
5945 tmp = neon_load_reg(rn, pass);
5946 tmp2 = neon_load_reg(rm, pass);
5950 GEN_NEON_INTEGER_OP(hadd);
5953 GEN_NEON_INTEGER_OP_ENV(qadd);
5955 case NEON_3R_VRHADD:
5956 GEN_NEON_INTEGER_OP(rhadd);
5958 case NEON_3R_LOGIC: /* Logic ops. */
5959 switch ((u << 2) | size) {
5961 tcg_gen_and_i32(tmp, tmp, tmp2);
5964 tcg_gen_andc_i32(tmp, tmp, tmp2);
5967 tcg_gen_or_i32(tmp, tmp, tmp2);
5970 tcg_gen_orc_i32(tmp, tmp, tmp2);
5973 tcg_gen_xor_i32(tmp, tmp, tmp2);
5976 tmp3 = neon_load_reg(rd, pass);
5977 gen_neon_bsl(tmp, tmp, tmp2, tmp3);
5978 tcg_temp_free_i32(tmp3);
5981 tmp3 = neon_load_reg(rd, pass);
5982 gen_neon_bsl(tmp, tmp, tmp3, tmp2);
5983 tcg_temp_free_i32(tmp3);
5986 tmp3 = neon_load_reg(rd, pass);
5987 gen_neon_bsl(tmp, tmp3, tmp, tmp2);
5988 tcg_temp_free_i32(tmp3);
5993 GEN_NEON_INTEGER_OP(hsub);
5996 GEN_NEON_INTEGER_OP_ENV(qsub);
5999 GEN_NEON_INTEGER_OP(cgt);
6002 GEN_NEON_INTEGER_OP(cge);
6005 GEN_NEON_INTEGER_OP(shl);
6008 GEN_NEON_INTEGER_OP_ENV(qshl);
6011 GEN_NEON_INTEGER_OP(rshl);
6013 case NEON_3R_VQRSHL:
6014 GEN_NEON_INTEGER_OP_ENV(qrshl);
6017 GEN_NEON_INTEGER_OP(max);
6020 GEN_NEON_INTEGER_OP(min);
6023 GEN_NEON_INTEGER_OP(abd);
6026 GEN_NEON_INTEGER_OP(abd);
6027 tcg_temp_free_i32(tmp2);
6028 tmp2 = neon_load_reg(rd, pass);
6029 gen_neon_add(size, tmp, tmp2);
6031 case NEON_3R_VADD_VSUB:
6032 if (!u) { /* VADD */
6033 gen_neon_add(size, tmp, tmp2);
6036 case 0: gen_helper_neon_sub_u8(tmp, tmp, tmp2); break;
6037 case 1: gen_helper_neon_sub_u16(tmp, tmp, tmp2); break;
6038 case 2: tcg_gen_sub_i32(tmp, tmp, tmp2); break;
6043 case NEON_3R_VTST_VCEQ:
6044 if (!u) { /* VTST */
6046 case 0: gen_helper_neon_tst_u8(tmp, tmp, tmp2); break;
6047 case 1: gen_helper_neon_tst_u16(tmp, tmp, tmp2); break;
6048 case 2: gen_helper_neon_tst_u32(tmp, tmp, tmp2); break;
6053 case 0: gen_helper_neon_ceq_u8(tmp, tmp, tmp2); break;
6054 case 1: gen_helper_neon_ceq_u16(tmp, tmp, tmp2); break;
6055 case 2: gen_helper_neon_ceq_u32(tmp, tmp, tmp2); break;
6060 case NEON_3R_VML: /* VMLA, VMLAL, VMLS,VMLSL */
6062 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
6063 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
6064 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
6067 tcg_temp_free_i32(tmp2);
6068 tmp2 = neon_load_reg(rd, pass);
6070 gen_neon_rsb(size, tmp, tmp2);
6072 gen_neon_add(size, tmp, tmp2);
6076 if (u) { /* polynomial */
6077 gen_helper_neon_mul_p8(tmp, tmp, tmp2);
6078 } else { /* Integer */
6080 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
6081 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
6082 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
6088 GEN_NEON_INTEGER_OP(pmax);
6091 GEN_NEON_INTEGER_OP(pmin);
6093 case NEON_3R_VQDMULH_VQRDMULH: /* Multiply high. */
6094 if (!u) { /* VQDMULH */
6097 gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2);
6100 gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2);
6104 } else { /* VQRDMULH */
6107 gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2);
6110 gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2);
6116 case NEON_3R_VPADD_VQRDMLAH:
6118 case 0: gen_helper_neon_padd_u8(tmp, tmp, tmp2); break;
6119 case 1: gen_helper_neon_padd_u16(tmp, tmp, tmp2); break;
6120 case 2: tcg_gen_add_i32(tmp, tmp, tmp2); break;
6124 case NEON_3R_FLOAT_ARITH: /* Floating point arithmetic. */
6126 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6127 switch ((u << 2) | size) {
6130 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
6133 gen_helper_vfp_subs(tmp, tmp, tmp2, fpstatus);
6136 gen_helper_neon_abd_f32(tmp, tmp, tmp2, fpstatus);
6141 tcg_temp_free_ptr(fpstatus);
6144 case NEON_3R_FLOAT_MULTIPLY:
6146 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6147 gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus);
6149 tcg_temp_free_i32(tmp2);
6150 tmp2 = neon_load_reg(rd, pass);
6152 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
6154 gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus);
6157 tcg_temp_free_ptr(fpstatus);
6160 case NEON_3R_FLOAT_CMP:
6162 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6164 gen_helper_neon_ceq_f32(tmp, tmp, tmp2, fpstatus);
6167 gen_helper_neon_cge_f32(tmp, tmp, tmp2, fpstatus);
6169 gen_helper_neon_cgt_f32(tmp, tmp, tmp2, fpstatus);
6172 tcg_temp_free_ptr(fpstatus);
6175 case NEON_3R_FLOAT_ACMP:
6177 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6179 gen_helper_neon_acge_f32(tmp, tmp, tmp2, fpstatus);
6181 gen_helper_neon_acgt_f32(tmp, tmp, tmp2, fpstatus);
6183 tcg_temp_free_ptr(fpstatus);
6186 case NEON_3R_FLOAT_MINMAX:
6188 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6190 gen_helper_vfp_maxs(tmp, tmp, tmp2, fpstatus);
6192 gen_helper_vfp_mins(tmp, tmp, tmp2, fpstatus);
6194 tcg_temp_free_ptr(fpstatus);
6197 case NEON_3R_FLOAT_MISC:
6200 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6202 gen_helper_vfp_maxnums(tmp, tmp, tmp2, fpstatus);
6204 gen_helper_vfp_minnums(tmp, tmp, tmp2, fpstatus);
6206 tcg_temp_free_ptr(fpstatus);
6209 gen_helper_recps_f32(tmp, tmp, tmp2, cpu_env);
6211 gen_helper_rsqrts_f32(tmp, tmp, tmp2, cpu_env);
6215 case NEON_3R_VFM_VQRDMLSH:
6217 /* VFMA, VFMS: fused multiply-add */
6218 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6219 TCGv_i32 tmp3 = neon_load_reg(rd, pass);
6222 gen_helper_vfp_negs(tmp, tmp);
6224 gen_helper_vfp_muladds(tmp, tmp, tmp2, tmp3, fpstatus);
6225 tcg_temp_free_i32(tmp3);
6226 tcg_temp_free_ptr(fpstatus);
6232 tcg_temp_free_i32(tmp2);
6234 /* Save the result. For elementwise operations we can put it
6235 straight into the destination register. For pairwise operations
6236 we have to be careful to avoid clobbering the source operands. */
6237 if (pairwise && rd == rm) {
6238 neon_store_scratch(pass, tmp);
6240 neon_store_reg(rd, pass, tmp);
6244 if (pairwise && rd == rm) {
6245 for (pass = 0; pass < (q ? 4 : 2); pass++) {
6246 tmp = neon_load_scratch(pass);
6247 neon_store_reg(rd, pass, tmp);
6250 /* End of 3 register same size operations. */
6251 } else if (insn & (1 << 4)) {
6252 if ((insn & 0x00380080) != 0) {
6253 /* Two registers and shift. */
6254 op = (insn >> 8) & 0xf;
6255 if (insn & (1 << 7)) {
6263 while ((insn & (1 << (size + 19))) == 0)
6266 shift = (insn >> 16) & ((1 << (3 + size)) - 1);
6267 /* To avoid excessive duplication of ops we implement shift
6268 by immediate using the variable shift operations. */
6270 /* Shift by immediate:
6271 VSHR, VSRA, VRSHR, VRSRA, VSRI, VSHL, VQSHL, VQSHLU. */
6272 if (q && ((rd | rm) & 1)) {
6275 if (!u && (op == 4 || op == 6)) {
6278 /* Right shifts are encoded as N - shift, where N is the
6279 element size in bits. */
6281 shift = shift - (1 << (size + 3));
6289 imm = (uint8_t) shift;
6294 imm = (uint16_t) shift;
6305 for (pass = 0; pass < count; pass++) {
6307 neon_load_reg64(cpu_V0, rm + pass);
6308 tcg_gen_movi_i64(cpu_V1, imm);
6313 gen_helper_neon_shl_u64(cpu_V0, cpu_V0, cpu_V1);
6315 gen_helper_neon_shl_s64(cpu_V0, cpu_V0, cpu_V1);
6320 gen_helper_neon_rshl_u64(cpu_V0, cpu_V0, cpu_V1);
6322 gen_helper_neon_rshl_s64(cpu_V0, cpu_V0, cpu_V1);
6325 case 5: /* VSHL, VSLI */
6326 gen_helper_neon_shl_u64(cpu_V0, cpu_V0, cpu_V1);
6328 case 6: /* VQSHLU */
6329 gen_helper_neon_qshlu_s64(cpu_V0, cpu_env,
6334 gen_helper_neon_qshl_u64(cpu_V0, cpu_env,
6337 gen_helper_neon_qshl_s64(cpu_V0, cpu_env,
6342 if (op == 1 || op == 3) {
6344 neon_load_reg64(cpu_V1, rd + pass);
6345 tcg_gen_add_i64(cpu_V0, cpu_V0, cpu_V1);
6346 } else if (op == 4 || (op == 5 && u)) {
6348 neon_load_reg64(cpu_V1, rd + pass);
6350 if (shift < -63 || shift > 63) {
6354 mask = 0xffffffffffffffffull >> -shift;
6356 mask = 0xffffffffffffffffull << shift;
6359 tcg_gen_andi_i64(cpu_V1, cpu_V1, ~mask);
6360 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
6362 neon_store_reg64(cpu_V0, rd + pass);
6363 } else { /* size < 3 */
6364 /* Operands in T0 and T1. */
6365 tmp = neon_load_reg(rm, pass);
6366 tmp2 = tcg_temp_new_i32();
6367 tcg_gen_movi_i32(tmp2, imm);
6371 GEN_NEON_INTEGER_OP(shl);
6375 GEN_NEON_INTEGER_OP(rshl);
6378 case 5: /* VSHL, VSLI */
6380 case 0: gen_helper_neon_shl_u8(tmp, tmp, tmp2); break;
6381 case 1: gen_helper_neon_shl_u16(tmp, tmp, tmp2); break;
6382 case 2: gen_helper_neon_shl_u32(tmp, tmp, tmp2); break;
6386 case 6: /* VQSHLU */
6389 gen_helper_neon_qshlu_s8(tmp, cpu_env,
6393 gen_helper_neon_qshlu_s16(tmp, cpu_env,
6397 gen_helper_neon_qshlu_s32(tmp, cpu_env,
6405 GEN_NEON_INTEGER_OP_ENV(qshl);
6408 tcg_temp_free_i32(tmp2);
6410 if (op == 1 || op == 3) {
6412 tmp2 = neon_load_reg(rd, pass);
6413 gen_neon_add(size, tmp, tmp2);
6414 tcg_temp_free_i32(tmp2);
6415 } else if (op == 4 || (op == 5 && u)) {
6420 mask = 0xff >> -shift;
6422 mask = (uint8_t)(0xff << shift);
6428 mask = 0xffff >> -shift;
6430 mask = (uint16_t)(0xffff << shift);
6434 if (shift < -31 || shift > 31) {
6438 mask = 0xffffffffu >> -shift;
6440 mask = 0xffffffffu << shift;
6446 tmp2 = neon_load_reg(rd, pass);
6447 tcg_gen_andi_i32(tmp, tmp, mask);
6448 tcg_gen_andi_i32(tmp2, tmp2, ~mask);
6449 tcg_gen_or_i32(tmp, tmp, tmp2);
6450 tcg_temp_free_i32(tmp2);
6452 neon_store_reg(rd, pass, tmp);
6455 } else if (op < 10) {
6456 /* Shift by immediate and narrow:
6457 VSHRN, VRSHRN, VQSHRN, VQRSHRN. */
6458 int input_unsigned = (op == 8) ? !u : u;
6462 shift = shift - (1 << (size + 3));
6465 tmp64 = tcg_const_i64(shift);
6466 neon_load_reg64(cpu_V0, rm);
6467 neon_load_reg64(cpu_V1, rm + 1);
6468 for (pass = 0; pass < 2; pass++) {
6476 if (input_unsigned) {
6477 gen_helper_neon_rshl_u64(cpu_V0, in, tmp64);
6479 gen_helper_neon_rshl_s64(cpu_V0, in, tmp64);
6482 if (input_unsigned) {
6483 gen_helper_neon_shl_u64(cpu_V0, in, tmp64);
6485 gen_helper_neon_shl_s64(cpu_V0, in, tmp64);
6488 tmp = tcg_temp_new_i32();
6489 gen_neon_narrow_op(op == 8, u, size - 1, tmp, cpu_V0);
6490 neon_store_reg(rd, pass, tmp);
6492 tcg_temp_free_i64(tmp64);
6495 imm = (uint16_t)shift;
6499 imm = (uint32_t)shift;
6501 tmp2 = tcg_const_i32(imm);
6502 tmp4 = neon_load_reg(rm + 1, 0);
6503 tmp5 = neon_load_reg(rm + 1, 1);
6504 for (pass = 0; pass < 2; pass++) {
6506 tmp = neon_load_reg(rm, 0);
6510 gen_neon_shift_narrow(size, tmp, tmp2, q,
6513 tmp3 = neon_load_reg(rm, 1);
6517 gen_neon_shift_narrow(size, tmp3, tmp2, q,
6519 tcg_gen_concat_i32_i64(cpu_V0, tmp, tmp3);
6520 tcg_temp_free_i32(tmp);
6521 tcg_temp_free_i32(tmp3);
6522 tmp = tcg_temp_new_i32();
6523 gen_neon_narrow_op(op == 8, u, size - 1, tmp, cpu_V0);
6524 neon_store_reg(rd, pass, tmp);
6526 tcg_temp_free_i32(tmp2);
6528 } else if (op == 10) {
6530 if (q || (rd & 1)) {
6533 tmp = neon_load_reg(rm, 0);
6534 tmp2 = neon_load_reg(rm, 1);
6535 for (pass = 0; pass < 2; pass++) {
6539 gen_neon_widen(cpu_V0, tmp, size, u);
6542 /* The shift is less than the width of the source
6543 type, so we can just shift the whole register. */
6544 tcg_gen_shli_i64(cpu_V0, cpu_V0, shift);
6545 /* Widen the result of shift: we need to clear
6546 * the potential overflow bits resulting from
6547 * left bits of the narrow input appearing as
6548 * right bits of left the neighbour narrow
6550 if (size < 2 || !u) {
6553 imm = (0xffu >> (8 - shift));
6555 } else if (size == 1) {
6556 imm = 0xffff >> (16 - shift);
6559 imm = 0xffffffff >> (32 - shift);
6562 imm64 = imm | (((uint64_t)imm) << 32);
6566 tcg_gen_andi_i64(cpu_V0, cpu_V0, ~imm64);
6569 neon_store_reg64(cpu_V0, rd + pass);
6571 } else if (op >= 14) {
6572 /* VCVT fixed-point. */
6573 if (!(insn & (1 << 21)) || (q && ((rd | rm) & 1))) {
6576 /* We have already masked out the must-be-1 top bit of imm6,
6577 * hence this 32-shift where the ARM ARM has 64-imm6.
6580 for (pass = 0; pass < (q ? 4 : 2); pass++) {
6581 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, pass));
6584 gen_vfp_ulto(0, shift, 1);
6586 gen_vfp_slto(0, shift, 1);
6589 gen_vfp_toul(0, shift, 1);
6591 gen_vfp_tosl(0, shift, 1);
6593 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, pass));
6598 } else { /* (insn & 0x00380080) == 0 */
6600 if (q && (rd & 1)) {
6604 op = (insn >> 8) & 0xf;
6605 /* One register and immediate. */
6606 imm = (u << 7) | ((insn >> 12) & 0x70) | (insn & 0xf);
6607 invert = (insn & (1 << 5)) != 0;
6608 /* Note that op = 2,3,4,5,6,7,10,11,12,13 imm=0 is UNPREDICTABLE.
6609 * We choose to not special-case this and will behave as if a
6610 * valid constant encoding of 0 had been given.
6629 imm = (imm << 8) | (imm << 24);
6632 imm = (imm << 8) | 0xff;
6635 imm = (imm << 16) | 0xffff;
6638 imm |= (imm << 8) | (imm << 16) | (imm << 24);
6646 imm = ((imm & 0x80) << 24) | ((imm & 0x3f) << 19)
6647 | ((imm & 0x40) ? (0x1f << 25) : (1 << 30));
6653 for (pass = 0; pass < (q ? 4 : 2); pass++) {
6654 if (op & 1 && op < 12) {
6655 tmp = neon_load_reg(rd, pass);
6657 /* The immediate value has already been inverted, so
6659 tcg_gen_andi_i32(tmp, tmp, imm);
6661 tcg_gen_ori_i32(tmp, tmp, imm);
6665 tmp = tcg_temp_new_i32();
6666 if (op == 14 && invert) {
6670 for (n = 0; n < 4; n++) {
6671 if (imm & (1 << (n + (pass & 1) * 4)))
6672 val |= 0xff << (n * 8);
6674 tcg_gen_movi_i32(tmp, val);
6676 tcg_gen_movi_i32(tmp, imm);
6679 neon_store_reg(rd, pass, tmp);
6682 } else { /* (insn & 0x00800010 == 0x00800000) */
6684 op = (insn >> 8) & 0xf;
6685 if ((insn & (1 << 6)) == 0) {
6686 /* Three registers of different lengths. */
6690 /* undefreq: bit 0 : UNDEF if size == 0
6691 * bit 1 : UNDEF if size == 1
6692 * bit 2 : UNDEF if size == 2
6693 * bit 3 : UNDEF if U == 1
6694 * Note that [2:0] set implies 'always UNDEF'
6697 /* prewiden, src1_wide, src2_wide, undefreq */
6698 static const int neon_3reg_wide[16][4] = {
6699 {1, 0, 0, 0}, /* VADDL */
6700 {1, 1, 0, 0}, /* VADDW */
6701 {1, 0, 0, 0}, /* VSUBL */
6702 {1, 1, 0, 0}, /* VSUBW */
6703 {0, 1, 1, 0}, /* VADDHN */
6704 {0, 0, 0, 0}, /* VABAL */
6705 {0, 1, 1, 0}, /* VSUBHN */
6706 {0, 0, 0, 0}, /* VABDL */
6707 {0, 0, 0, 0}, /* VMLAL */
6708 {0, 0, 0, 9}, /* VQDMLAL */
6709 {0, 0, 0, 0}, /* VMLSL */
6710 {0, 0, 0, 9}, /* VQDMLSL */
6711 {0, 0, 0, 0}, /* Integer VMULL */
6712 {0, 0, 0, 1}, /* VQDMULL */
6713 {0, 0, 0, 0xa}, /* Polynomial VMULL */
6714 {0, 0, 0, 7}, /* Reserved: always UNDEF */
6717 prewiden = neon_3reg_wide[op][0];
6718 src1_wide = neon_3reg_wide[op][1];
6719 src2_wide = neon_3reg_wide[op][2];
6720 undefreq = neon_3reg_wide[op][3];
6722 if ((undefreq & (1 << size)) ||
6723 ((undefreq & 8) && u)) {
6726 if ((src1_wide && (rn & 1)) ||
6727 (src2_wide && (rm & 1)) ||
6728 (!src2_wide && (rd & 1))) {
6732 /* Handle VMULL.P64 (Polynomial 64x64 to 128 bit multiply)
6733 * outside the loop below as it only performs a single pass.
6735 if (op == 14 && size == 2) {
6736 TCGv_i64 tcg_rn, tcg_rm, tcg_rd;
6738 if (!arm_dc_feature(s, ARM_FEATURE_V8_PMULL)) {
6741 tcg_rn = tcg_temp_new_i64();
6742 tcg_rm = tcg_temp_new_i64();
6743 tcg_rd = tcg_temp_new_i64();
6744 neon_load_reg64(tcg_rn, rn);
6745 neon_load_reg64(tcg_rm, rm);
6746 gen_helper_neon_pmull_64_lo(tcg_rd, tcg_rn, tcg_rm);
6747 neon_store_reg64(tcg_rd, rd);
6748 gen_helper_neon_pmull_64_hi(tcg_rd, tcg_rn, tcg_rm);
6749 neon_store_reg64(tcg_rd, rd + 1);
6750 tcg_temp_free_i64(tcg_rn);
6751 tcg_temp_free_i64(tcg_rm);
6752 tcg_temp_free_i64(tcg_rd);
6756 /* Avoid overlapping operands. Wide source operands are
6757 always aligned so will never overlap with wide
6758 destinations in problematic ways. */
6759 if (rd == rm && !src2_wide) {
6760 tmp = neon_load_reg(rm, 1);
6761 neon_store_scratch(2, tmp);
6762 } else if (rd == rn && !src1_wide) {
6763 tmp = neon_load_reg(rn, 1);
6764 neon_store_scratch(2, tmp);
6767 for (pass = 0; pass < 2; pass++) {
6769 neon_load_reg64(cpu_V0, rn + pass);
6772 if (pass == 1 && rd == rn) {
6773 tmp = neon_load_scratch(2);
6775 tmp = neon_load_reg(rn, pass);
6778 gen_neon_widen(cpu_V0, tmp, size, u);
6782 neon_load_reg64(cpu_V1, rm + pass);
6785 if (pass == 1 && rd == rm) {
6786 tmp2 = neon_load_scratch(2);
6788 tmp2 = neon_load_reg(rm, pass);
6791 gen_neon_widen(cpu_V1, tmp2, size, u);
6795 case 0: case 1: case 4: /* VADDL, VADDW, VADDHN, VRADDHN */
6796 gen_neon_addl(size);
6798 case 2: case 3: case 6: /* VSUBL, VSUBW, VSUBHN, VRSUBHN */
6799 gen_neon_subl(size);
6801 case 5: case 7: /* VABAL, VABDL */
6802 switch ((size << 1) | u) {
6804 gen_helper_neon_abdl_s16(cpu_V0, tmp, tmp2);
6807 gen_helper_neon_abdl_u16(cpu_V0, tmp, tmp2);
6810 gen_helper_neon_abdl_s32(cpu_V0, tmp, tmp2);
6813 gen_helper_neon_abdl_u32(cpu_V0, tmp, tmp2);
6816 gen_helper_neon_abdl_s64(cpu_V0, tmp, tmp2);
6819 gen_helper_neon_abdl_u64(cpu_V0, tmp, tmp2);
6823 tcg_temp_free_i32(tmp2);
6824 tcg_temp_free_i32(tmp);
6826 case 8: case 9: case 10: case 11: case 12: case 13:
6827 /* VMLAL, VQDMLAL, VMLSL, VQDMLSL, VMULL, VQDMULL */
6828 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
6830 case 14: /* Polynomial VMULL */
6831 gen_helper_neon_mull_p8(cpu_V0, tmp, tmp2);
6832 tcg_temp_free_i32(tmp2);
6833 tcg_temp_free_i32(tmp);
6835 default: /* 15 is RESERVED: caught earlier */
6840 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
6841 neon_store_reg64(cpu_V0, rd + pass);
6842 } else if (op == 5 || (op >= 8 && op <= 11)) {
6844 neon_load_reg64(cpu_V1, rd + pass);
6846 case 10: /* VMLSL */
6847 gen_neon_negl(cpu_V0, size);
6849 case 5: case 8: /* VABAL, VMLAL */
6850 gen_neon_addl(size);
6852 case 9: case 11: /* VQDMLAL, VQDMLSL */
6853 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
6855 gen_neon_negl(cpu_V0, size);
6857 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
6862 neon_store_reg64(cpu_V0, rd + pass);
6863 } else if (op == 4 || op == 6) {
6864 /* Narrowing operation. */
6865 tmp = tcg_temp_new_i32();
6869 gen_helper_neon_narrow_high_u8(tmp, cpu_V0);
6872 gen_helper_neon_narrow_high_u16(tmp, cpu_V0);
6875 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
6876 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
6883 gen_helper_neon_narrow_round_high_u8(tmp, cpu_V0);
6886 gen_helper_neon_narrow_round_high_u16(tmp, cpu_V0);
6889 tcg_gen_addi_i64(cpu_V0, cpu_V0, 1u << 31);
6890 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
6891 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
6899 neon_store_reg(rd, 0, tmp3);
6900 neon_store_reg(rd, 1, tmp);
6903 /* Write back the result. */
6904 neon_store_reg64(cpu_V0, rd + pass);
6908 /* Two registers and a scalar. NB that for ops of this form
6909 * the ARM ARM labels bit 24 as Q, but it is in our variable
6916 case 1: /* Float VMLA scalar */
6917 case 5: /* Floating point VMLS scalar */
6918 case 9: /* Floating point VMUL scalar */
6923 case 0: /* Integer VMLA scalar */
6924 case 4: /* Integer VMLS scalar */
6925 case 8: /* Integer VMUL scalar */
6926 case 12: /* VQDMULH scalar */
6927 case 13: /* VQRDMULH scalar */
6928 if (u && ((rd | rn) & 1)) {
6931 tmp = neon_get_scalar(size, rm);
6932 neon_store_scratch(0, tmp);
6933 for (pass = 0; pass < (u ? 4 : 2); pass++) {
6934 tmp = neon_load_scratch(0);
6935 tmp2 = neon_load_reg(rn, pass);
6938 gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2);
6940 gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2);
6942 } else if (op == 13) {
6944 gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2);
6946 gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2);
6948 } else if (op & 1) {
6949 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6950 gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus);
6951 tcg_temp_free_ptr(fpstatus);
6954 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
6955 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
6956 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
6960 tcg_temp_free_i32(tmp2);
6963 tmp2 = neon_load_reg(rd, pass);
6966 gen_neon_add(size, tmp, tmp2);
6970 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6971 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
6972 tcg_temp_free_ptr(fpstatus);
6976 gen_neon_rsb(size, tmp, tmp2);
6980 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6981 gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus);
6982 tcg_temp_free_ptr(fpstatus);
6988 tcg_temp_free_i32(tmp2);
6990 neon_store_reg(rd, pass, tmp);
6993 case 3: /* VQDMLAL scalar */
6994 case 7: /* VQDMLSL scalar */
6995 case 11: /* VQDMULL scalar */
7000 case 2: /* VMLAL sclar */
7001 case 6: /* VMLSL scalar */
7002 case 10: /* VMULL scalar */
7006 tmp2 = neon_get_scalar(size, rm);
7007 /* We need a copy of tmp2 because gen_neon_mull
7008 * deletes it during pass 0. */
7009 tmp4 = tcg_temp_new_i32();
7010 tcg_gen_mov_i32(tmp4, tmp2);
7011 tmp3 = neon_load_reg(rn, 1);
7013 for (pass = 0; pass < 2; pass++) {
7015 tmp = neon_load_reg(rn, 0);
7020 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
7022 neon_load_reg64(cpu_V1, rd + pass);
7026 gen_neon_negl(cpu_V0, size);
7029 gen_neon_addl(size);
7032 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
7034 gen_neon_negl(cpu_V0, size);
7036 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
7042 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
7047 neon_store_reg64(cpu_V0, rd + pass);
7050 case 14: /* VQRDMLAH scalar */
7051 case 15: /* VQRDMLSH scalar */
7053 NeonGenThreeOpEnvFn *fn;
7055 if (!arm_dc_feature(s, ARM_FEATURE_V8_RDM)) {
7058 if (u && ((rd | rn) & 1)) {
7063 fn = gen_helper_neon_qrdmlah_s16;
7065 fn = gen_helper_neon_qrdmlah_s32;
7069 fn = gen_helper_neon_qrdmlsh_s16;
7071 fn = gen_helper_neon_qrdmlsh_s32;
7075 tmp2 = neon_get_scalar(size, rm);
7076 for (pass = 0; pass < (u ? 4 : 2); pass++) {
7077 tmp = neon_load_reg(rn, pass);
7078 tmp3 = neon_load_reg(rd, pass);
7079 fn(tmp, cpu_env, tmp, tmp2, tmp3);
7080 tcg_temp_free_i32(tmp3);
7081 neon_store_reg(rd, pass, tmp);
7083 tcg_temp_free_i32(tmp2);
7087 g_assert_not_reached();
7090 } else { /* size == 3 */
7093 imm = (insn >> 8) & 0xf;
7098 if (q && ((rd | rn | rm) & 1)) {
7103 neon_load_reg64(cpu_V0, rn);
7105 neon_load_reg64(cpu_V1, rn + 1);
7107 } else if (imm == 8) {
7108 neon_load_reg64(cpu_V0, rn + 1);
7110 neon_load_reg64(cpu_V1, rm);
7113 tmp64 = tcg_temp_new_i64();
7115 neon_load_reg64(cpu_V0, rn);
7116 neon_load_reg64(tmp64, rn + 1);
7118 neon_load_reg64(cpu_V0, rn + 1);
7119 neon_load_reg64(tmp64, rm);
7121 tcg_gen_shri_i64(cpu_V0, cpu_V0, (imm & 7) * 8);
7122 tcg_gen_shli_i64(cpu_V1, tmp64, 64 - ((imm & 7) * 8));
7123 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
7125 neon_load_reg64(cpu_V1, rm);
7127 neon_load_reg64(cpu_V1, rm + 1);
7130 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
7131 tcg_gen_shri_i64(tmp64, tmp64, imm * 8);
7132 tcg_gen_or_i64(cpu_V1, cpu_V1, tmp64);
7133 tcg_temp_free_i64(tmp64);
7136 neon_load_reg64(cpu_V0, rn);
7137 tcg_gen_shri_i64(cpu_V0, cpu_V0, imm * 8);
7138 neon_load_reg64(cpu_V1, rm);
7139 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
7140 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
7142 neon_store_reg64(cpu_V0, rd);
7144 neon_store_reg64(cpu_V1, rd + 1);
7146 } else if ((insn & (1 << 11)) == 0) {
7147 /* Two register misc. */
7148 op = ((insn >> 12) & 0x30) | ((insn >> 7) & 0xf);
7149 size = (insn >> 18) & 3;
7150 /* UNDEF for unknown op values and bad op-size combinations */
7151 if ((neon_2rm_sizes[op] & (1 << size)) == 0) {
7154 if (neon_2rm_is_v8_op(op) &&
7155 !arm_dc_feature(s, ARM_FEATURE_V8)) {
7158 if ((op != NEON_2RM_VMOVN && op != NEON_2RM_VQMOVN) &&
7159 q && ((rm | rd) & 1)) {
7163 case NEON_2RM_VREV64:
7164 for (pass = 0; pass < (q ? 2 : 1); pass++) {
7165 tmp = neon_load_reg(rm, pass * 2);
7166 tmp2 = neon_load_reg(rm, pass * 2 + 1);
7168 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
7169 case 1: gen_swap_half(tmp); break;
7170 case 2: /* no-op */ break;
7173 neon_store_reg(rd, pass * 2 + 1, tmp);
7175 neon_store_reg(rd, pass * 2, tmp2);
7178 case 0: tcg_gen_bswap32_i32(tmp2, tmp2); break;
7179 case 1: gen_swap_half(tmp2); break;
7182 neon_store_reg(rd, pass * 2, tmp2);
7186 case NEON_2RM_VPADDL: case NEON_2RM_VPADDL_U:
7187 case NEON_2RM_VPADAL: case NEON_2RM_VPADAL_U:
7188 for (pass = 0; pass < q + 1; pass++) {
7189 tmp = neon_load_reg(rm, pass * 2);
7190 gen_neon_widen(cpu_V0, tmp, size, op & 1);
7191 tmp = neon_load_reg(rm, pass * 2 + 1);
7192 gen_neon_widen(cpu_V1, tmp, size, op & 1);
7194 case 0: gen_helper_neon_paddl_u16(CPU_V001); break;
7195 case 1: gen_helper_neon_paddl_u32(CPU_V001); break;
7196 case 2: tcg_gen_add_i64(CPU_V001); break;
7199 if (op >= NEON_2RM_VPADAL) {
7201 neon_load_reg64(cpu_V1, rd + pass);
7202 gen_neon_addl(size);
7204 neon_store_reg64(cpu_V0, rd + pass);
7210 for (n = 0; n < (q ? 4 : 2); n += 2) {
7211 tmp = neon_load_reg(rm, n);
7212 tmp2 = neon_load_reg(rd, n + 1);
7213 neon_store_reg(rm, n, tmp2);
7214 neon_store_reg(rd, n + 1, tmp);
7221 if (gen_neon_unzip(rd, rm, size, q)) {
7226 if (gen_neon_zip(rd, rm, size, q)) {
7230 case NEON_2RM_VMOVN: case NEON_2RM_VQMOVN:
7231 /* also VQMOVUN; op field and mnemonics don't line up */
7236 for (pass = 0; pass < 2; pass++) {
7237 neon_load_reg64(cpu_V0, rm + pass);
7238 tmp = tcg_temp_new_i32();
7239 gen_neon_narrow_op(op == NEON_2RM_VMOVN, q, size,
7244 neon_store_reg(rd, 0, tmp2);
7245 neon_store_reg(rd, 1, tmp);
7249 case NEON_2RM_VSHLL:
7250 if (q || (rd & 1)) {
7253 tmp = neon_load_reg(rm, 0);
7254 tmp2 = neon_load_reg(rm, 1);
7255 for (pass = 0; pass < 2; pass++) {
7258 gen_neon_widen(cpu_V0, tmp, size, 1);
7259 tcg_gen_shli_i64(cpu_V0, cpu_V0, 8 << size);
7260 neon_store_reg64(cpu_V0, rd + pass);
7263 case NEON_2RM_VCVT_F16_F32:
7268 if (!arm_dc_feature(s, ARM_FEATURE_VFP_FP16) ||
7272 tmp = tcg_temp_new_i32();
7273 tmp2 = tcg_temp_new_i32();
7274 fpst = get_fpstatus_ptr(true);
7275 ahp = get_ahp_flag();
7276 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 0));
7277 gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s, fpst, ahp);
7278 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 1));
7279 gen_helper_vfp_fcvt_f32_to_f16(tmp2, cpu_F0s, fpst, ahp);
7280 tcg_gen_shli_i32(tmp2, tmp2, 16);
7281 tcg_gen_or_i32(tmp2, tmp2, tmp);
7282 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 2));
7283 gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s, fpst, ahp);
7284 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 3));
7285 neon_store_reg(rd, 0, tmp2);
7286 tmp2 = tcg_temp_new_i32();
7287 gen_helper_vfp_fcvt_f32_to_f16(tmp2, cpu_F0s, fpst, ahp);
7288 tcg_gen_shli_i32(tmp2, tmp2, 16);
7289 tcg_gen_or_i32(tmp2, tmp2, tmp);
7290 neon_store_reg(rd, 1, tmp2);
7291 tcg_temp_free_i32(tmp);
7292 tcg_temp_free_i32(ahp);
7293 tcg_temp_free_ptr(fpst);
7296 case NEON_2RM_VCVT_F32_F16:
7300 if (!arm_dc_feature(s, ARM_FEATURE_VFP_FP16) ||
7304 fpst = get_fpstatus_ptr(true);
7305 ahp = get_ahp_flag();
7306 tmp3 = tcg_temp_new_i32();
7307 tmp = neon_load_reg(rm, 0);
7308 tmp2 = neon_load_reg(rm, 1);
7309 tcg_gen_ext16u_i32(tmp3, tmp);
7310 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp3, fpst, ahp);
7311 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 0));
7312 tcg_gen_shri_i32(tmp3, tmp, 16);
7313 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp3, fpst, ahp);
7314 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 1));
7315 tcg_temp_free_i32(tmp);
7316 tcg_gen_ext16u_i32(tmp3, tmp2);
7317 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp3, fpst, ahp);
7318 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 2));
7319 tcg_gen_shri_i32(tmp3, tmp2, 16);
7320 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp3, fpst, ahp);
7321 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 3));
7322 tcg_temp_free_i32(tmp2);
7323 tcg_temp_free_i32(tmp3);
7324 tcg_temp_free_i32(ahp);
7325 tcg_temp_free_ptr(fpst);
7328 case NEON_2RM_AESE: case NEON_2RM_AESMC:
7329 if (!arm_dc_feature(s, ARM_FEATURE_V8_AES)
7330 || ((rm | rd) & 1)) {
7333 ptr1 = vfp_reg_ptr(true, rd);
7334 ptr2 = vfp_reg_ptr(true, rm);
7336 /* Bit 6 is the lowest opcode bit; it distinguishes between
7337 * encryption (AESE/AESMC) and decryption (AESD/AESIMC)
7339 tmp3 = tcg_const_i32(extract32(insn, 6, 1));
7341 if (op == NEON_2RM_AESE) {
7342 gen_helper_crypto_aese(ptr1, ptr2, tmp3);
7344 gen_helper_crypto_aesmc(ptr1, ptr2, tmp3);
7346 tcg_temp_free_ptr(ptr1);
7347 tcg_temp_free_ptr(ptr2);
7348 tcg_temp_free_i32(tmp3);
7350 case NEON_2RM_SHA1H:
7351 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA1)
7352 || ((rm | rd) & 1)) {
7355 ptr1 = vfp_reg_ptr(true, rd);
7356 ptr2 = vfp_reg_ptr(true, rm);
7358 gen_helper_crypto_sha1h(ptr1, ptr2);
7360 tcg_temp_free_ptr(ptr1);
7361 tcg_temp_free_ptr(ptr2);
7363 case NEON_2RM_SHA1SU1:
7364 if ((rm | rd) & 1) {
7367 /* bit 6 (q): set -> SHA256SU0, cleared -> SHA1SU1 */
7369 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA256)) {
7372 } else if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA1)) {
7375 ptr1 = vfp_reg_ptr(true, rd);
7376 ptr2 = vfp_reg_ptr(true, rm);
7378 gen_helper_crypto_sha256su0(ptr1, ptr2);
7380 gen_helper_crypto_sha1su1(ptr1, ptr2);
7382 tcg_temp_free_ptr(ptr1);
7383 tcg_temp_free_ptr(ptr2);
7387 for (pass = 0; pass < (q ? 4 : 2); pass++) {
7388 if (neon_2rm_is_float_op(op)) {
7389 tcg_gen_ld_f32(cpu_F0s, cpu_env,
7390 neon_reg_offset(rm, pass));
7393 tmp = neon_load_reg(rm, pass);
7396 case NEON_2RM_VREV32:
7398 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
7399 case 1: gen_swap_half(tmp); break;
7403 case NEON_2RM_VREV16:
7408 case 0: gen_helper_neon_cls_s8(tmp, tmp); break;
7409 case 1: gen_helper_neon_cls_s16(tmp, tmp); break;
7410 case 2: gen_helper_neon_cls_s32(tmp, tmp); break;
7416 case 0: gen_helper_neon_clz_u8(tmp, tmp); break;
7417 case 1: gen_helper_neon_clz_u16(tmp, tmp); break;
7418 case 2: tcg_gen_clzi_i32(tmp, tmp, 32); break;
7423 gen_helper_neon_cnt_u8(tmp, tmp);
7426 tcg_gen_not_i32(tmp, tmp);
7428 case NEON_2RM_VQABS:
7431 gen_helper_neon_qabs_s8(tmp, cpu_env, tmp);
7434 gen_helper_neon_qabs_s16(tmp, cpu_env, tmp);
7437 gen_helper_neon_qabs_s32(tmp, cpu_env, tmp);
7442 case NEON_2RM_VQNEG:
7445 gen_helper_neon_qneg_s8(tmp, cpu_env, tmp);
7448 gen_helper_neon_qneg_s16(tmp, cpu_env, tmp);
7451 gen_helper_neon_qneg_s32(tmp, cpu_env, tmp);
7456 case NEON_2RM_VCGT0: case NEON_2RM_VCLE0:
7457 tmp2 = tcg_const_i32(0);
7459 case 0: gen_helper_neon_cgt_s8(tmp, tmp, tmp2); break;
7460 case 1: gen_helper_neon_cgt_s16(tmp, tmp, tmp2); break;
7461 case 2: gen_helper_neon_cgt_s32(tmp, tmp, tmp2); break;
7464 tcg_temp_free_i32(tmp2);
7465 if (op == NEON_2RM_VCLE0) {
7466 tcg_gen_not_i32(tmp, tmp);
7469 case NEON_2RM_VCGE0: case NEON_2RM_VCLT0:
7470 tmp2 = tcg_const_i32(0);
7472 case 0: gen_helper_neon_cge_s8(tmp, tmp, tmp2); break;
7473 case 1: gen_helper_neon_cge_s16(tmp, tmp, tmp2); break;
7474 case 2: gen_helper_neon_cge_s32(tmp, tmp, tmp2); break;
7477 tcg_temp_free_i32(tmp2);
7478 if (op == NEON_2RM_VCLT0) {
7479 tcg_gen_not_i32(tmp, tmp);
7482 case NEON_2RM_VCEQ0:
7483 tmp2 = tcg_const_i32(0);
7485 case 0: gen_helper_neon_ceq_u8(tmp, tmp, tmp2); break;
7486 case 1: gen_helper_neon_ceq_u16(tmp, tmp, tmp2); break;
7487 case 2: gen_helper_neon_ceq_u32(tmp, tmp, tmp2); break;
7490 tcg_temp_free_i32(tmp2);
7494 case 0: gen_helper_neon_abs_s8(tmp, tmp); break;
7495 case 1: gen_helper_neon_abs_s16(tmp, tmp); break;
7496 case 2: tcg_gen_abs_i32(tmp, tmp); break;
7501 tmp2 = tcg_const_i32(0);
7502 gen_neon_rsb(size, tmp, tmp2);
7503 tcg_temp_free_i32(tmp2);
7505 case NEON_2RM_VCGT0_F:
7507 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7508 tmp2 = tcg_const_i32(0);
7509 gen_helper_neon_cgt_f32(tmp, tmp, tmp2, fpstatus);
7510 tcg_temp_free_i32(tmp2);
7511 tcg_temp_free_ptr(fpstatus);
7514 case NEON_2RM_VCGE0_F:
7516 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7517 tmp2 = tcg_const_i32(0);
7518 gen_helper_neon_cge_f32(tmp, tmp, tmp2, fpstatus);
7519 tcg_temp_free_i32(tmp2);
7520 tcg_temp_free_ptr(fpstatus);
7523 case NEON_2RM_VCEQ0_F:
7525 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7526 tmp2 = tcg_const_i32(0);
7527 gen_helper_neon_ceq_f32(tmp, tmp, tmp2, fpstatus);
7528 tcg_temp_free_i32(tmp2);
7529 tcg_temp_free_ptr(fpstatus);
7532 case NEON_2RM_VCLE0_F:
7534 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7535 tmp2 = tcg_const_i32(0);
7536 gen_helper_neon_cge_f32(tmp, tmp2, tmp, fpstatus);
7537 tcg_temp_free_i32(tmp2);
7538 tcg_temp_free_ptr(fpstatus);
7541 case NEON_2RM_VCLT0_F:
7543 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7544 tmp2 = tcg_const_i32(0);
7545 gen_helper_neon_cgt_f32(tmp, tmp2, tmp, fpstatus);
7546 tcg_temp_free_i32(tmp2);
7547 tcg_temp_free_ptr(fpstatus);
7550 case NEON_2RM_VABS_F:
7553 case NEON_2RM_VNEG_F:
7557 tmp2 = neon_load_reg(rd, pass);
7558 neon_store_reg(rm, pass, tmp2);
7561 tmp2 = neon_load_reg(rd, pass);
7563 case 0: gen_neon_trn_u8(tmp, tmp2); break;
7564 case 1: gen_neon_trn_u16(tmp, tmp2); break;
7567 neon_store_reg(rm, pass, tmp2);
7569 case NEON_2RM_VRINTN:
7570 case NEON_2RM_VRINTA:
7571 case NEON_2RM_VRINTM:
7572 case NEON_2RM_VRINTP:
7573 case NEON_2RM_VRINTZ:
7576 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7579 if (op == NEON_2RM_VRINTZ) {
7580 rmode = FPROUNDING_ZERO;
7582 rmode = fp_decode_rm[((op & 0x6) >> 1) ^ 1];
7585 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
7586 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
7588 gen_helper_rints(cpu_F0s, cpu_F0s, fpstatus);
7589 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
7591 tcg_temp_free_ptr(fpstatus);
7592 tcg_temp_free_i32(tcg_rmode);
7595 case NEON_2RM_VRINTX:
7597 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7598 gen_helper_rints_exact(cpu_F0s, cpu_F0s, fpstatus);
7599 tcg_temp_free_ptr(fpstatus);
7602 case NEON_2RM_VCVTAU:
7603 case NEON_2RM_VCVTAS:
7604 case NEON_2RM_VCVTNU:
7605 case NEON_2RM_VCVTNS:
7606 case NEON_2RM_VCVTPU:
7607 case NEON_2RM_VCVTPS:
7608 case NEON_2RM_VCVTMU:
7609 case NEON_2RM_VCVTMS:
7611 bool is_signed = !extract32(insn, 7, 1);
7612 TCGv_ptr fpst = get_fpstatus_ptr(1);
7613 TCGv_i32 tcg_rmode, tcg_shift;
7614 int rmode = fp_decode_rm[extract32(insn, 8, 2)];
7616 tcg_shift = tcg_const_i32(0);
7617 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
7618 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
7622 gen_helper_vfp_tosls(cpu_F0s, cpu_F0s,
7625 gen_helper_vfp_touls(cpu_F0s, cpu_F0s,
7629 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
7631 tcg_temp_free_i32(tcg_rmode);
7632 tcg_temp_free_i32(tcg_shift);
7633 tcg_temp_free_ptr(fpst);
7636 case NEON_2RM_VRECPE:
7638 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7639 gen_helper_recpe_u32(tmp, tmp, fpstatus);
7640 tcg_temp_free_ptr(fpstatus);
7643 case NEON_2RM_VRSQRTE:
7645 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7646 gen_helper_rsqrte_u32(tmp, tmp, fpstatus);
7647 tcg_temp_free_ptr(fpstatus);
7650 case NEON_2RM_VRECPE_F:
7652 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7653 gen_helper_recpe_f32(cpu_F0s, cpu_F0s, fpstatus);
7654 tcg_temp_free_ptr(fpstatus);
7657 case NEON_2RM_VRSQRTE_F:
7659 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
7660 gen_helper_rsqrte_f32(cpu_F0s, cpu_F0s, fpstatus);
7661 tcg_temp_free_ptr(fpstatus);
7664 case NEON_2RM_VCVT_FS: /* VCVT.F32.S32 */
7667 case NEON_2RM_VCVT_FU: /* VCVT.F32.U32 */
7670 case NEON_2RM_VCVT_SF: /* VCVT.S32.F32 */
7671 gen_vfp_tosiz(0, 1);
7673 case NEON_2RM_VCVT_UF: /* VCVT.U32.F32 */
7674 gen_vfp_touiz(0, 1);
7677 /* Reserved op values were caught by the
7678 * neon_2rm_sizes[] check earlier.
7682 if (neon_2rm_is_float_op(op)) {
7683 tcg_gen_st_f32(cpu_F0s, cpu_env,
7684 neon_reg_offset(rd, pass));
7686 neon_store_reg(rd, pass, tmp);
7691 } else if ((insn & (1 << 10)) == 0) {
7693 int n = ((insn >> 8) & 3) + 1;
7694 if ((rn + n) > 32) {
7695 /* This is UNPREDICTABLE; we choose to UNDEF to avoid the
7696 * helper function running off the end of the register file.
7701 if (insn & (1 << 6)) {
7702 tmp = neon_load_reg(rd, 0);
7704 tmp = tcg_temp_new_i32();
7705 tcg_gen_movi_i32(tmp, 0);
7707 tmp2 = neon_load_reg(rm, 0);
7708 ptr1 = vfp_reg_ptr(true, rn);
7709 tmp5 = tcg_const_i32(n);
7710 gen_helper_neon_tbl(tmp2, tmp2, tmp, ptr1, tmp5);
7711 tcg_temp_free_i32(tmp);
7712 if (insn & (1 << 6)) {
7713 tmp = neon_load_reg(rd, 1);
7715 tmp = tcg_temp_new_i32();
7716 tcg_gen_movi_i32(tmp, 0);
7718 tmp3 = neon_load_reg(rm, 1);
7719 gen_helper_neon_tbl(tmp3, tmp3, tmp, ptr1, tmp5);
7720 tcg_temp_free_i32(tmp5);
7721 tcg_temp_free_ptr(ptr1);
7722 neon_store_reg(rd, 0, tmp2);
7723 neon_store_reg(rd, 1, tmp3);
7724 tcg_temp_free_i32(tmp);
7725 } else if ((insn & 0x380) == 0) {
7727 if ((insn & (7 << 16)) == 0 || (q && (rd & 1))) {
7730 if (insn & (1 << 19)) {
7731 tmp = neon_load_reg(rm, 1);
7733 tmp = neon_load_reg(rm, 0);
7735 if (insn & (1 << 16)) {
7736 gen_neon_dup_u8(tmp, ((insn >> 17) & 3) * 8);
7737 } else if (insn & (1 << 17)) {
7738 if ((insn >> 18) & 1)
7739 gen_neon_dup_high16(tmp);
7741 gen_neon_dup_low16(tmp);
7743 for (pass = 0; pass < (q ? 4 : 2); pass++) {
7744 tmp2 = tcg_temp_new_i32();
7745 tcg_gen_mov_i32(tmp2, tmp);
7746 neon_store_reg(rd, pass, tmp2);
7748 tcg_temp_free_i32(tmp);
7757 /* Advanced SIMD three registers of the same length extension.
7758 * 31 25 23 22 20 16 12 11 10 9 8 3 0
7759 * +---------------+-----+---+-----+----+----+---+----+---+----+---------+----+
7760 * | 1 1 1 1 1 1 0 | op1 | D | op2 | Vn | Vd | 1 | o3 | 0 | o4 | N Q M U | Vm |
7761 * +---------------+-----+---+-----+----+----+---+----+---+----+---------+----+
7763 static int disas_neon_insn_3same_ext(DisasContext *s, uint32_t insn)
7765 gen_helper_gvec_3_ptr *fn_gvec_ptr;
7766 int rd, rn, rm, rot, size, opr_sz;
7770 q = extract32(insn, 6, 1);
7771 VFP_DREG_D(rd, insn);
7772 VFP_DREG_N(rn, insn);
7773 VFP_DREG_M(rm, insn);
7774 if ((rd | rn | rm) & q) {
7778 if ((insn & 0xfe200f10) == 0xfc200800) {
7779 /* VCMLA -- 1111 110R R.1S .... .... 1000 ...0 .... */
7780 size = extract32(insn, 20, 1);
7781 rot = extract32(insn, 23, 2);
7782 if (!arm_dc_feature(s, ARM_FEATURE_V8_FCMA)
7783 || (!size && !arm_dc_feature(s, ARM_FEATURE_V8_FP16))) {
7786 fn_gvec_ptr = size ? gen_helper_gvec_fcmlas : gen_helper_gvec_fcmlah;
7787 } else if ((insn & 0xfea00f10) == 0xfc800800) {
7788 /* VCADD -- 1111 110R 1.0S .... .... 1000 ...0 .... */
7789 size = extract32(insn, 20, 1);
7790 rot = extract32(insn, 24, 1);
7791 if (!arm_dc_feature(s, ARM_FEATURE_V8_FCMA)
7792 || (!size && !arm_dc_feature(s, ARM_FEATURE_V8_FP16))) {
7795 fn_gvec_ptr = size ? gen_helper_gvec_fcadds : gen_helper_gvec_fcaddh;
7800 if (s->fp_excp_el) {
7801 gen_exception_insn(s, 4, EXCP_UDEF,
7802 syn_fp_access_trap(1, 0xe, false), s->fp_excp_el);
7805 if (!s->vfp_enabled) {
7809 opr_sz = (1 + q) * 8;
7810 fpst = get_fpstatus_ptr(1);
7811 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, rd),
7812 vfp_reg_offset(1, rn),
7813 vfp_reg_offset(1, rm), fpst,
7814 opr_sz, opr_sz, rot, fn_gvec_ptr);
7815 tcg_temp_free_ptr(fpst);
7819 /* Advanced SIMD two registers and a scalar extension.
7820 * 31 24 23 22 20 16 12 11 10 9 8 3 0
7821 * +-----------------+----+---+----+----+----+---+----+---+----+---------+----+
7822 * | 1 1 1 1 1 1 1 0 | o1 | D | o2 | Vn | Vd | 1 | o3 | 0 | o4 | N Q M U | Vm |
7823 * +-----------------+----+---+----+----+----+---+----+---+----+---------+----+
7827 static int disas_neon_insn_2reg_scalar_ext(DisasContext *s, uint32_t insn)
7829 int rd, rn, rm, rot, size, opr_sz;
7833 q = extract32(insn, 6, 1);
7834 VFP_DREG_D(rd, insn);
7835 VFP_DREG_N(rn, insn);
7836 VFP_DREG_M(rm, insn);
7837 if ((rd | rn) & q) {
7841 if ((insn & 0xff000f10) == 0xfe000800) {
7842 /* VCMLA (indexed) -- 1111 1110 S.RR .... .... 1000 ...0 .... */
7843 rot = extract32(insn, 20, 2);
7844 size = extract32(insn, 23, 1);
7845 if (!arm_dc_feature(s, ARM_FEATURE_V8_FCMA)
7846 || (!size && !arm_dc_feature(s, ARM_FEATURE_V8_FP16))) {
7853 if (s->fp_excp_el) {
7854 gen_exception_insn(s, 4, EXCP_UDEF,
7855 syn_fp_access_trap(1, 0xe, false), s->fp_excp_el);
7858 if (!s->vfp_enabled) {
7862 opr_sz = (1 + q) * 8;
7863 fpst = get_fpstatus_ptr(1);
7864 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, rd),
7865 vfp_reg_offset(1, rn),
7866 vfp_reg_offset(1, rm), fpst,
7867 opr_sz, opr_sz, rot,
7868 size ? gen_helper_gvec_fcmlas_idx
7869 : gen_helper_gvec_fcmlah_idx);
7870 tcg_temp_free_ptr(fpst);
7874 static int disas_coproc_insn(DisasContext *s, uint32_t insn)
7876 int cpnum, is64, crn, crm, opc1, opc2, isread, rt, rt2;
7877 const ARMCPRegInfo *ri;
7879 cpnum = (insn >> 8) & 0xf;
7881 /* First check for coprocessor space used for XScale/iwMMXt insns */
7882 if (arm_dc_feature(s, ARM_FEATURE_XSCALE) && (cpnum < 2)) {
7883 if (extract32(s->c15_cpar, cpnum, 1) == 0) {
7886 if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
7887 return disas_iwmmxt_insn(s, insn);
7888 } else if (arm_dc_feature(s, ARM_FEATURE_XSCALE)) {
7889 return disas_dsp_insn(s, insn);
7894 /* Otherwise treat as a generic register access */
7895 is64 = (insn & (1 << 25)) == 0;
7896 if (!is64 && ((insn & (1 << 4)) == 0)) {
7904 opc1 = (insn >> 4) & 0xf;
7906 rt2 = (insn >> 16) & 0xf;
7908 crn = (insn >> 16) & 0xf;
7909 opc1 = (insn >> 21) & 7;
7910 opc2 = (insn >> 5) & 7;
7913 isread = (insn >> 20) & 1;
7914 rt = (insn >> 12) & 0xf;
7916 ri = get_arm_cp_reginfo(s->cp_regs,
7917 ENCODE_CP_REG(cpnum, is64, s->ns, crn, crm, opc1, opc2));
7919 /* Check access permissions */
7920 if (!cp_access_ok(s->current_el, ri, isread)) {
7925 (arm_dc_feature(s, ARM_FEATURE_XSCALE) && cpnum < 14)) {
7926 /* Emit code to perform further access permissions checks at
7927 * runtime; this may result in an exception.
7928 * Note that on XScale all cp0..c13 registers do an access check
7929 * call in order to handle c15_cpar.
7932 TCGv_i32 tcg_syn, tcg_isread;
7935 /* Note that since we are an implementation which takes an
7936 * exception on a trapped conditional instruction only if the
7937 * instruction passes its condition code check, we can take
7938 * advantage of the clause in the ARM ARM that allows us to set
7939 * the COND field in the instruction to 0xE in all cases.
7940 * We could fish the actual condition out of the insn (ARM)
7941 * or the condexec bits (Thumb) but it isn't necessary.
7946 syndrome = syn_cp14_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
7949 syndrome = syn_cp14_rt_trap(1, 0xe, opc1, opc2, crn, crm,
7955 syndrome = syn_cp15_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
7958 syndrome = syn_cp15_rt_trap(1, 0xe, opc1, opc2, crn, crm,
7963 /* ARMv8 defines that only coprocessors 14 and 15 exist,
7964 * so this can only happen if this is an ARMv7 or earlier CPU,
7965 * in which case the syndrome information won't actually be
7968 assert(!arm_dc_feature(s, ARM_FEATURE_V8));
7969 syndrome = syn_uncategorized();
7973 gen_set_condexec(s);
7974 gen_set_pc_im(s, s->pc - 4);
7975 tmpptr = tcg_const_ptr(ri);
7976 tcg_syn = tcg_const_i32(syndrome);
7977 tcg_isread = tcg_const_i32(isread);
7978 gen_helper_access_check_cp_reg(cpu_env, tmpptr, tcg_syn,
7980 tcg_temp_free_ptr(tmpptr);
7981 tcg_temp_free_i32(tcg_syn);
7982 tcg_temp_free_i32(tcg_isread);
7985 /* Handle special cases first */
7986 switch (ri->type & ~(ARM_CP_FLAG_MASK & ~ARM_CP_SPECIAL)) {
7993 gen_set_pc_im(s, s->pc);
7994 s->base.is_jmp = DISAS_WFI;
8000 if ((tb_cflags(s->base.tb) & CF_USE_ICOUNT) && (ri->type & ARM_CP_IO)) {
8009 if (ri->type & ARM_CP_CONST) {
8010 tmp64 = tcg_const_i64(ri->resetvalue);
8011 } else if (ri->readfn) {
8013 tmp64 = tcg_temp_new_i64();
8014 tmpptr = tcg_const_ptr(ri);
8015 gen_helper_get_cp_reg64(tmp64, cpu_env, tmpptr);
8016 tcg_temp_free_ptr(tmpptr);
8018 tmp64 = tcg_temp_new_i64();
8019 tcg_gen_ld_i64(tmp64, cpu_env, ri->fieldoffset);
8021 tmp = tcg_temp_new_i32();
8022 tcg_gen_extrl_i64_i32(tmp, tmp64);
8023 store_reg(s, rt, tmp);
8024 tcg_gen_shri_i64(tmp64, tmp64, 32);
8025 tmp = tcg_temp_new_i32();
8026 tcg_gen_extrl_i64_i32(tmp, tmp64);
8027 tcg_temp_free_i64(tmp64);
8028 store_reg(s, rt2, tmp);
8031 if (ri->type & ARM_CP_CONST) {
8032 tmp = tcg_const_i32(ri->resetvalue);
8033 } else if (ri->readfn) {
8035 tmp = tcg_temp_new_i32();
8036 tmpptr = tcg_const_ptr(ri);
8037 gen_helper_get_cp_reg(tmp, cpu_env, tmpptr);
8038 tcg_temp_free_ptr(tmpptr);
8040 tmp = load_cpu_offset(ri->fieldoffset);
8043 /* Destination register of r15 for 32 bit loads sets
8044 * the condition codes from the high 4 bits of the value
8047 tcg_temp_free_i32(tmp);
8049 store_reg(s, rt, tmp);
8054 if (ri->type & ARM_CP_CONST) {
8055 /* If not forbidden by access permissions, treat as WI */
8060 TCGv_i32 tmplo, tmphi;
8061 TCGv_i64 tmp64 = tcg_temp_new_i64();
8062 tmplo = load_reg(s, rt);
8063 tmphi = load_reg(s, rt2);
8064 tcg_gen_concat_i32_i64(tmp64, tmplo, tmphi);
8065 tcg_temp_free_i32(tmplo);
8066 tcg_temp_free_i32(tmphi);
8068 TCGv_ptr tmpptr = tcg_const_ptr(ri);
8069 gen_helper_set_cp_reg64(cpu_env, tmpptr, tmp64);
8070 tcg_temp_free_ptr(tmpptr);
8072 tcg_gen_st_i64(tmp64, cpu_env, ri->fieldoffset);
8074 tcg_temp_free_i64(tmp64);
8079 tmp = load_reg(s, rt);
8080 tmpptr = tcg_const_ptr(ri);
8081 gen_helper_set_cp_reg(cpu_env, tmpptr, tmp);
8082 tcg_temp_free_ptr(tmpptr);
8083 tcg_temp_free_i32(tmp);
8085 TCGv_i32 tmp = load_reg(s, rt);
8086 store_cpu_offset(tmp, ri->fieldoffset);
8091 if ((tb_cflags(s->base.tb) & CF_USE_ICOUNT) && (ri->type & ARM_CP_IO)) {
8092 /* I/O operations must end the TB here (whether read or write) */
8095 } else if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
8096 /* We default to ending the TB on a coprocessor register write,
8097 * but allow this to be suppressed by the register definition
8098 * (usually only necessary to work around guest bugs).
8106 /* Unknown register; this might be a guest error or a QEMU
8107 * unimplemented feature.
8110 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
8111 "64 bit system register cp:%d opc1: %d crm:%d "
8113 isread ? "read" : "write", cpnum, opc1, crm,
8114 s->ns ? "non-secure" : "secure");
8116 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
8117 "system register cp:%d opc1:%d crn:%d crm:%d opc2:%d "
8119 isread ? "read" : "write", cpnum, opc1, crn, crm, opc2,
8120 s->ns ? "non-secure" : "secure");
8127 /* Store a 64-bit value to a register pair. Clobbers val. */
8128 static void gen_storeq_reg(DisasContext *s, int rlow, int rhigh, TCGv_i64 val)
8131 tmp = tcg_temp_new_i32();
8132 tcg_gen_extrl_i64_i32(tmp, val);
8133 store_reg(s, rlow, tmp);
8134 tmp = tcg_temp_new_i32();
8135 tcg_gen_shri_i64(val, val, 32);
8136 tcg_gen_extrl_i64_i32(tmp, val);
8137 store_reg(s, rhigh, tmp);
8140 /* load a 32-bit value from a register and perform a 64-bit accumulate. */
8141 static void gen_addq_lo(DisasContext *s, TCGv_i64 val, int rlow)
8146 /* Load value and extend to 64 bits. */
8147 tmp = tcg_temp_new_i64();
8148 tmp2 = load_reg(s, rlow);
8149 tcg_gen_extu_i32_i64(tmp, tmp2);
8150 tcg_temp_free_i32(tmp2);
8151 tcg_gen_add_i64(val, val, tmp);
8152 tcg_temp_free_i64(tmp);
8155 /* load and add a 64-bit value from a register pair. */
8156 static void gen_addq(DisasContext *s, TCGv_i64 val, int rlow, int rhigh)
8162 /* Load 64-bit value rd:rn. */
8163 tmpl = load_reg(s, rlow);
8164 tmph = load_reg(s, rhigh);
8165 tmp = tcg_temp_new_i64();
8166 tcg_gen_concat_i32_i64(tmp, tmpl, tmph);
8167 tcg_temp_free_i32(tmpl);
8168 tcg_temp_free_i32(tmph);
8169 tcg_gen_add_i64(val, val, tmp);
8170 tcg_temp_free_i64(tmp);
8173 /* Set N and Z flags from hi|lo. */
8174 static void gen_logicq_cc(TCGv_i32 lo, TCGv_i32 hi)
8176 tcg_gen_mov_i32(cpu_NF, hi);
8177 tcg_gen_or_i32(cpu_ZF, lo, hi);
8180 /* Load/Store exclusive instructions are implemented by remembering
8181 the value/address loaded, and seeing if these are the same
8182 when the store is performed. This should be sufficient to implement
8183 the architecturally mandated semantics, and avoids having to monitor
8184 regular stores. The compare vs the remembered value is done during
8185 the cmpxchg operation, but we must compare the addresses manually. */
8186 static void gen_load_exclusive(DisasContext *s, int rt, int rt2,
8187 TCGv_i32 addr, int size)
8189 TCGv_i32 tmp = tcg_temp_new_i32();
8190 TCGMemOp opc = size | MO_ALIGN | s->be_data;
8195 TCGv_i32 tmp2 = tcg_temp_new_i32();
8196 TCGv_i64 t64 = tcg_temp_new_i64();
8198 /* For AArch32, architecturally the 32-bit word at the lowest
8199 * address is always Rt and the one at addr+4 is Rt2, even if
8200 * the CPU is big-endian. That means we don't want to do a
8201 * gen_aa32_ld_i64(), which invokes gen_aa32_frob64() as if
8202 * for an architecturally 64-bit access, but instead do a
8203 * 64-bit access using MO_BE if appropriate and then split
8205 * This only makes a difference for BE32 user-mode, where
8206 * frob64() must not flip the two halves of the 64-bit data
8207 * but this code must treat BE32 user-mode like BE32 system.
8209 TCGv taddr = gen_aa32_addr(s, addr, opc);
8211 tcg_gen_qemu_ld_i64(t64, taddr, get_mem_index(s), opc);
8212 tcg_temp_free(taddr);
8213 tcg_gen_mov_i64(cpu_exclusive_val, t64);
8214 if (s->be_data == MO_BE) {
8215 tcg_gen_extr_i64_i32(tmp2, tmp, t64);
8217 tcg_gen_extr_i64_i32(tmp, tmp2, t64);
8219 tcg_temp_free_i64(t64);
8221 store_reg(s, rt2, tmp2);
8223 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), opc);
8224 tcg_gen_extu_i32_i64(cpu_exclusive_val, tmp);
8227 store_reg(s, rt, tmp);
8228 tcg_gen_extu_i32_i64(cpu_exclusive_addr, addr);
8231 static void gen_clrex(DisasContext *s)
8233 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
8236 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
8237 TCGv_i32 addr, int size)
8239 TCGv_i32 t0, t1, t2;
8242 TCGLabel *done_label;
8243 TCGLabel *fail_label;
8244 TCGMemOp opc = size | MO_ALIGN | s->be_data;
8246 /* if (env->exclusive_addr == addr && env->exclusive_val == [addr]) {
8252 fail_label = gen_new_label();
8253 done_label = gen_new_label();
8254 extaddr = tcg_temp_new_i64();
8255 tcg_gen_extu_i32_i64(extaddr, addr);
8256 tcg_gen_brcond_i64(TCG_COND_NE, extaddr, cpu_exclusive_addr, fail_label);
8257 tcg_temp_free_i64(extaddr);
8259 taddr = gen_aa32_addr(s, addr, opc);
8260 t0 = tcg_temp_new_i32();
8261 t1 = load_reg(s, rt);
8263 TCGv_i64 o64 = tcg_temp_new_i64();
8264 TCGv_i64 n64 = tcg_temp_new_i64();
8266 t2 = load_reg(s, rt2);
8267 /* For AArch32, architecturally the 32-bit word at the lowest
8268 * address is always Rt and the one at addr+4 is Rt2, even if
8269 * the CPU is big-endian. Since we're going to treat this as a
8270 * single 64-bit BE store, we need to put the two halves in the
8271 * opposite order for BE to LE, so that they end up in the right
8273 * We don't want gen_aa32_frob64() because that does the wrong
8274 * thing for BE32 usermode.
8276 if (s->be_data == MO_BE) {
8277 tcg_gen_concat_i32_i64(n64, t2, t1);
8279 tcg_gen_concat_i32_i64(n64, t1, t2);
8281 tcg_temp_free_i32(t2);
8283 tcg_gen_atomic_cmpxchg_i64(o64, taddr, cpu_exclusive_val, n64,
8284 get_mem_index(s), opc);
8285 tcg_temp_free_i64(n64);
8287 tcg_gen_setcond_i64(TCG_COND_NE, o64, o64, cpu_exclusive_val);
8288 tcg_gen_extrl_i64_i32(t0, o64);
8290 tcg_temp_free_i64(o64);
8292 t2 = tcg_temp_new_i32();
8293 tcg_gen_extrl_i64_i32(t2, cpu_exclusive_val);
8294 tcg_gen_atomic_cmpxchg_i32(t0, taddr, t2, t1, get_mem_index(s), opc);
8295 tcg_gen_setcond_i32(TCG_COND_NE, t0, t0, t2);
8296 tcg_temp_free_i32(t2);
8298 tcg_temp_free_i32(t1);
8299 tcg_temp_free(taddr);
8300 tcg_gen_mov_i32(cpu_R[rd], t0);
8301 tcg_temp_free_i32(t0);
8302 tcg_gen_br(done_label);
8304 gen_set_label(fail_label);
8305 tcg_gen_movi_i32(cpu_R[rd], 1);
8306 gen_set_label(done_label);
8307 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
8313 * @mode: mode field from insn (which stack to store to)
8314 * @amode: addressing mode (DA/IA/DB/IB), encoded as per P,U bits in ARM insn
8315 * @writeback: true if writeback bit set
8317 * Generate code for the SRS (Store Return State) insn.
8319 static void gen_srs(DisasContext *s,
8320 uint32_t mode, uint32_t amode, bool writeback)
8327 * - trapped to EL3 if EL3 is AArch64 and we are at Secure EL1
8328 * and specified mode is monitor mode
8329 * - UNDEFINED in Hyp mode
8330 * - UNPREDICTABLE in User or System mode
8331 * - UNPREDICTABLE if the specified mode is:
8332 * -- not implemented
8333 * -- not a valid mode number
8334 * -- a mode that's at a higher exception level
8335 * -- Monitor, if we are Non-secure
8336 * For the UNPREDICTABLE cases we choose to UNDEF.
8338 if (s->current_el == 1 && !s->ns && mode == ARM_CPU_MODE_MON) {
8339 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(), 3);
8343 if (s->current_el == 0 || s->current_el == 2) {
8348 case ARM_CPU_MODE_USR:
8349 case ARM_CPU_MODE_FIQ:
8350 case ARM_CPU_MODE_IRQ:
8351 case ARM_CPU_MODE_SVC:
8352 case ARM_CPU_MODE_ABT:
8353 case ARM_CPU_MODE_UND:
8354 case ARM_CPU_MODE_SYS:
8356 case ARM_CPU_MODE_HYP:
8357 if (s->current_el == 1 || !arm_dc_feature(s, ARM_FEATURE_EL2)) {
8361 case ARM_CPU_MODE_MON:
8362 /* No need to check specifically for "are we non-secure" because
8363 * we've already made EL0 UNDEF and handled the trap for S-EL1;
8364 * so if this isn't EL3 then we must be non-secure.
8366 if (s->current_el != 3) {
8375 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(),
8376 default_exception_el(s));
8380 addr = tcg_temp_new_i32();
8381 tmp = tcg_const_i32(mode);
8382 /* get_r13_banked() will raise an exception if called from System mode */
8383 gen_set_condexec(s);
8384 gen_set_pc_im(s, s->pc - 4);
8385 gen_helper_get_r13_banked(addr, cpu_env, tmp);
8386 tcg_temp_free_i32(tmp);
8403 tcg_gen_addi_i32(addr, addr, offset);
8404 tmp = load_reg(s, 14);
8405 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
8406 tcg_temp_free_i32(tmp);
8407 tmp = load_cpu_field(spsr);
8408 tcg_gen_addi_i32(addr, addr, 4);
8409 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
8410 tcg_temp_free_i32(tmp);
8428 tcg_gen_addi_i32(addr, addr, offset);
8429 tmp = tcg_const_i32(mode);
8430 gen_helper_set_r13_banked(cpu_env, tmp, addr);
8431 tcg_temp_free_i32(tmp);
8433 tcg_temp_free_i32(addr);
8434 s->base.is_jmp = DISAS_UPDATE;
8437 static void disas_arm_insn(DisasContext *s, unsigned int insn)
8439 unsigned int cond, val, op1, i, shift, rm, rs, rn, rd, sh;
8446 /* M variants do not implement ARM mode; this must raise the INVSTATE
8447 * UsageFault exception.
8449 if (arm_dc_feature(s, ARM_FEATURE_M)) {
8450 gen_exception_insn(s, 4, EXCP_INVSTATE, syn_uncategorized(),
8451 default_exception_el(s));
8456 /* In ARMv3 and v4 the NV condition is UNPREDICTABLE; we
8457 * choose to UNDEF. In ARMv5 and above the space is used
8458 * for miscellaneous unconditional instructions.
8462 /* Unconditional instructions. */
8463 if (((insn >> 25) & 7) == 1) {
8464 /* NEON Data processing. */
8465 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
8469 if (disas_neon_data_insn(s, insn)) {
8474 if ((insn & 0x0f100000) == 0x04000000) {
8475 /* NEON load/store. */
8476 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
8480 if (disas_neon_ls_insn(s, insn)) {
8485 if ((insn & 0x0f000e10) == 0x0e000a00) {
8487 if (disas_vfp_insn(s, insn)) {
8492 if (((insn & 0x0f30f000) == 0x0510f000) ||
8493 ((insn & 0x0f30f010) == 0x0710f000)) {
8494 if ((insn & (1 << 22)) == 0) {
8496 if (!arm_dc_feature(s, ARM_FEATURE_V7MP)) {
8500 /* Otherwise PLD; v5TE+ */
8504 if (((insn & 0x0f70f000) == 0x0450f000) ||
8505 ((insn & 0x0f70f010) == 0x0650f000)) {
8507 return; /* PLI; V7 */
8509 if (((insn & 0x0f700000) == 0x04100000) ||
8510 ((insn & 0x0f700010) == 0x06100000)) {
8511 if (!arm_dc_feature(s, ARM_FEATURE_V7MP)) {
8514 return; /* v7MP: Unallocated memory hint: must NOP */
8517 if ((insn & 0x0ffffdff) == 0x01010000) {
8520 if (((insn >> 9) & 1) != !!(s->be_data == MO_BE)) {
8521 gen_helper_setend(cpu_env);
8522 s->base.is_jmp = DISAS_UPDATE;
8525 } else if ((insn & 0x0fffff00) == 0x057ff000) {
8526 switch ((insn >> 4) & 0xf) {
8534 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
8537 /* We need to break the TB after this insn to execute
8538 * self-modifying code correctly and also to take
8539 * any pending interrupts immediately.
8541 gen_goto_tb(s, 0, s->pc & ~1);
8546 } else if ((insn & 0x0e5fffe0) == 0x084d0500) {
8549 gen_srs(s, (insn & 0x1f), (insn >> 23) & 3, insn & (1 << 21));
8551 } else if ((insn & 0x0e50ffe0) == 0x08100a00) {
8557 rn = (insn >> 16) & 0xf;
8558 addr = load_reg(s, rn);
8559 i = (insn >> 23) & 3;
8561 case 0: offset = -4; break; /* DA */
8562 case 1: offset = 0; break; /* IA */
8563 case 2: offset = -8; break; /* DB */
8564 case 3: offset = 4; break; /* IB */
8568 tcg_gen_addi_i32(addr, addr, offset);
8569 /* Load PC into tmp and CPSR into tmp2. */
8570 tmp = tcg_temp_new_i32();
8571 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
8572 tcg_gen_addi_i32(addr, addr, 4);
8573 tmp2 = tcg_temp_new_i32();
8574 gen_aa32_ld32u(s, tmp2, addr, get_mem_index(s));
8575 if (insn & (1 << 21)) {
8576 /* Base writeback. */
8578 case 0: offset = -8; break;
8579 case 1: offset = 4; break;
8580 case 2: offset = -4; break;
8581 case 3: offset = 0; break;
8585 tcg_gen_addi_i32(addr, addr, offset);
8586 store_reg(s, rn, addr);
8588 tcg_temp_free_i32(addr);
8590 gen_rfe(s, tmp, tmp2);
8592 } else if ((insn & 0x0e000000) == 0x0a000000) {
8593 /* branch link and change to thumb (blx <offset>) */
8596 val = (uint32_t)s->pc;
8597 tmp = tcg_temp_new_i32();
8598 tcg_gen_movi_i32(tmp, val);
8599 store_reg(s, 14, tmp);
8600 /* Sign-extend the 24-bit offset */
8601 offset = (((int32_t)insn) << 8) >> 8;
8602 /* offset * 4 + bit24 * 2 + (thumb bit) */
8603 val += (offset << 2) | ((insn >> 23) & 2) | 1;
8604 /* pipeline offset */
8606 /* protected by ARCH(5); above, near the start of uncond block */
8609 } else if ((insn & 0x0e000f00) == 0x0c000100) {
8610 if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
8611 /* iWMMXt register transfer. */
8612 if (extract32(s->c15_cpar, 1, 1)) {
8613 if (!disas_iwmmxt_insn(s, insn)) {
8618 } else if ((insn & 0x0e000a00) == 0x0c000800
8619 && arm_dc_feature(s, ARM_FEATURE_V8)) {
8620 if (disas_neon_insn_3same_ext(s, insn)) {
8624 } else if ((insn & 0x0f000a00) == 0x0e000800
8625 && arm_dc_feature(s, ARM_FEATURE_V8)) {
8626 if (disas_neon_insn_2reg_scalar_ext(s, insn)) {
8630 } else if ((insn & 0x0fe00000) == 0x0c400000) {
8631 /* Coprocessor double register transfer. */
8633 } else if ((insn & 0x0f000010) == 0x0e000010) {
8634 /* Additional coprocessor register transfer. */
8635 } else if ((insn & 0x0ff10020) == 0x01000000) {
8638 /* cps (privileged) */
8642 if (insn & (1 << 19)) {
8643 if (insn & (1 << 8))
8645 if (insn & (1 << 7))
8647 if (insn & (1 << 6))
8649 if (insn & (1 << 18))
8652 if (insn & (1 << 17)) {
8654 val |= (insn & 0x1f);
8657 gen_set_psr_im(s, mask, 0, val);
8664 /* if not always execute, we generate a conditional jump to
8666 s->condlabel = gen_new_label();
8667 arm_gen_test_cc(cond ^ 1, s->condlabel);
8670 if ((insn & 0x0f900000) == 0x03000000) {
8671 if ((insn & (1 << 21)) == 0) {
8673 rd = (insn >> 12) & 0xf;
8674 val = ((insn >> 4) & 0xf000) | (insn & 0xfff);
8675 if ((insn & (1 << 22)) == 0) {
8677 tmp = tcg_temp_new_i32();
8678 tcg_gen_movi_i32(tmp, val);
8681 tmp = load_reg(s, rd);
8682 tcg_gen_ext16u_i32(tmp, tmp);
8683 tcg_gen_ori_i32(tmp, tmp, val << 16);
8685 store_reg(s, rd, tmp);
8687 if (((insn >> 12) & 0xf) != 0xf)
8689 if (((insn >> 16) & 0xf) == 0) {
8690 gen_nop_hint(s, insn & 0xff);
8692 /* CPSR = immediate */
8694 shift = ((insn >> 8) & 0xf) * 2;
8696 val = (val >> shift) | (val << (32 - shift));
8697 i = ((insn & (1 << 22)) != 0);
8698 if (gen_set_psr_im(s, msr_mask(s, (insn >> 16) & 0xf, i),
8704 } else if ((insn & 0x0f900000) == 0x01000000
8705 && (insn & 0x00000090) != 0x00000090) {
8706 /* miscellaneous instructions */
8707 op1 = (insn >> 21) & 3;
8708 sh = (insn >> 4) & 0xf;
8711 case 0x0: /* MSR, MRS */
8712 if (insn & (1 << 9)) {
8713 /* MSR (banked) and MRS (banked) */
8714 int sysm = extract32(insn, 16, 4) |
8715 (extract32(insn, 8, 1) << 4);
8716 int r = extract32(insn, 22, 1);
8720 gen_msr_banked(s, r, sysm, rm);
8723 int rd = extract32(insn, 12, 4);
8725 gen_mrs_banked(s, r, sysm, rd);
8730 /* MSR, MRS (for PSRs) */
8733 tmp = load_reg(s, rm);
8734 i = ((op1 & 2) != 0);
8735 if (gen_set_psr(s, msr_mask(s, (insn >> 16) & 0xf, i), i, tmp))
8739 rd = (insn >> 12) & 0xf;
8743 tmp = load_cpu_field(spsr);
8745 tmp = tcg_temp_new_i32();
8746 gen_helper_cpsr_read(tmp, cpu_env);
8748 store_reg(s, rd, tmp);
8753 /* branch/exchange thumb (bx). */
8755 tmp = load_reg(s, rm);
8757 } else if (op1 == 3) {
8760 rd = (insn >> 12) & 0xf;
8761 tmp = load_reg(s, rm);
8762 tcg_gen_clzi_i32(tmp, tmp, 32);
8763 store_reg(s, rd, tmp);
8771 /* Trivial implementation equivalent to bx. */
8772 tmp = load_reg(s, rm);
8783 /* branch link/exchange thumb (blx) */
8784 tmp = load_reg(s, rm);
8785 tmp2 = tcg_temp_new_i32();
8786 tcg_gen_movi_i32(tmp2, s->pc);
8787 store_reg(s, 14, tmp2);
8793 uint32_t c = extract32(insn, 8, 4);
8795 /* Check this CPU supports ARMv8 CRC instructions.
8796 * op1 == 3 is UNPREDICTABLE but handle as UNDEFINED.
8797 * Bits 8, 10 and 11 should be zero.
8799 if (!arm_dc_feature(s, ARM_FEATURE_CRC) || op1 == 0x3 ||
8804 rn = extract32(insn, 16, 4);
8805 rd = extract32(insn, 12, 4);
8807 tmp = load_reg(s, rn);
8808 tmp2 = load_reg(s, rm);
8810 tcg_gen_andi_i32(tmp2, tmp2, 0xff);
8811 } else if (op1 == 1) {
8812 tcg_gen_andi_i32(tmp2, tmp2, 0xffff);
8814 tmp3 = tcg_const_i32(1 << op1);
8816 gen_helper_crc32c(tmp, tmp, tmp2, tmp3);
8818 gen_helper_crc32(tmp, tmp, tmp2, tmp3);
8820 tcg_temp_free_i32(tmp2);
8821 tcg_temp_free_i32(tmp3);
8822 store_reg(s, rd, tmp);
8825 case 0x5: /* saturating add/subtract */
8827 rd = (insn >> 12) & 0xf;
8828 rn = (insn >> 16) & 0xf;
8829 tmp = load_reg(s, rm);
8830 tmp2 = load_reg(s, rn);
8832 gen_helper_double_saturate(tmp2, cpu_env, tmp2);
8834 gen_helper_sub_saturate(tmp, cpu_env, tmp, tmp2);
8836 gen_helper_add_saturate(tmp, cpu_env, tmp, tmp2);
8837 tcg_temp_free_i32(tmp2);
8838 store_reg(s, rd, tmp);
8842 int imm16 = extract32(insn, 0, 4) | (extract32(insn, 8, 12) << 4);
8851 gen_exception_bkpt_insn(s, 4, syn_aa32_bkpt(imm16, false));
8854 /* Hypervisor call (v7) */
8862 /* Secure monitor call (v6+) */
8870 g_assert_not_reached();
8874 case 0x8: /* signed multiply */
8879 rs = (insn >> 8) & 0xf;
8880 rn = (insn >> 12) & 0xf;
8881 rd = (insn >> 16) & 0xf;
8883 /* (32 * 16) >> 16 */
8884 tmp = load_reg(s, rm);
8885 tmp2 = load_reg(s, rs);
8887 tcg_gen_sari_i32(tmp2, tmp2, 16);
8890 tmp64 = gen_muls_i64_i32(tmp, tmp2);
8891 tcg_gen_shri_i64(tmp64, tmp64, 16);
8892 tmp = tcg_temp_new_i32();
8893 tcg_gen_extrl_i64_i32(tmp, tmp64);
8894 tcg_temp_free_i64(tmp64);
8895 if ((sh & 2) == 0) {
8896 tmp2 = load_reg(s, rn);
8897 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8898 tcg_temp_free_i32(tmp2);
8900 store_reg(s, rd, tmp);
8903 tmp = load_reg(s, rm);
8904 tmp2 = load_reg(s, rs);
8905 gen_mulxy(tmp, tmp2, sh & 2, sh & 4);
8906 tcg_temp_free_i32(tmp2);
8908 tmp64 = tcg_temp_new_i64();
8909 tcg_gen_ext_i32_i64(tmp64, tmp);
8910 tcg_temp_free_i32(tmp);
8911 gen_addq(s, tmp64, rn, rd);
8912 gen_storeq_reg(s, rn, rd, tmp64);
8913 tcg_temp_free_i64(tmp64);
8916 tmp2 = load_reg(s, rn);
8917 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8918 tcg_temp_free_i32(tmp2);
8920 store_reg(s, rd, tmp);
8927 } else if (((insn & 0x0e000000) == 0 &&
8928 (insn & 0x00000090) != 0x90) ||
8929 ((insn & 0x0e000000) == (1 << 25))) {
8930 int set_cc, logic_cc, shiftop;
8932 op1 = (insn >> 21) & 0xf;
8933 set_cc = (insn >> 20) & 1;
8934 logic_cc = table_logic_cc[op1] & set_cc;
8936 /* data processing instruction */
8937 if (insn & (1 << 25)) {
8938 /* immediate operand */
8940 shift = ((insn >> 8) & 0xf) * 2;
8942 val = (val >> shift) | (val << (32 - shift));
8944 tmp2 = tcg_temp_new_i32();
8945 tcg_gen_movi_i32(tmp2, val);
8946 if (logic_cc && shift) {
8947 gen_set_CF_bit31(tmp2);
8952 tmp2 = load_reg(s, rm);
8953 shiftop = (insn >> 5) & 3;
8954 if (!(insn & (1 << 4))) {
8955 shift = (insn >> 7) & 0x1f;
8956 gen_arm_shift_im(tmp2, shiftop, shift, logic_cc);
8958 rs = (insn >> 8) & 0xf;
8959 tmp = load_reg(s, rs);
8960 gen_arm_shift_reg(tmp2, shiftop, tmp, logic_cc);
8963 if (op1 != 0x0f && op1 != 0x0d) {
8964 rn = (insn >> 16) & 0xf;
8965 tmp = load_reg(s, rn);
8969 rd = (insn >> 12) & 0xf;
8972 tcg_gen_and_i32(tmp, tmp, tmp2);
8976 store_reg_bx(s, rd, tmp);
8979 tcg_gen_xor_i32(tmp, tmp, tmp2);
8983 store_reg_bx(s, rd, tmp);
8986 if (set_cc && rd == 15) {
8987 /* SUBS r15, ... is used for exception return. */
8991 gen_sub_CC(tmp, tmp, tmp2);
8992 gen_exception_return(s, tmp);
8995 gen_sub_CC(tmp, tmp, tmp2);
8997 tcg_gen_sub_i32(tmp, tmp, tmp2);
8999 store_reg_bx(s, rd, tmp);
9004 gen_sub_CC(tmp, tmp2, tmp);
9006 tcg_gen_sub_i32(tmp, tmp2, tmp);
9008 store_reg_bx(s, rd, tmp);
9012 gen_add_CC(tmp, tmp, tmp2);
9014 tcg_gen_add_i32(tmp, tmp, tmp2);
9016 store_reg_bx(s, rd, tmp);
9020 gen_adc_CC(tmp, tmp, tmp2);
9022 gen_add_carry(tmp, tmp, tmp2);
9024 store_reg_bx(s, rd, tmp);
9028 gen_sbc_CC(tmp, tmp, tmp2);
9030 gen_sub_carry(tmp, tmp, tmp2);
9032 store_reg_bx(s, rd, tmp);
9036 gen_sbc_CC(tmp, tmp2, tmp);
9038 gen_sub_carry(tmp, tmp2, tmp);
9040 store_reg_bx(s, rd, tmp);
9044 tcg_gen_and_i32(tmp, tmp, tmp2);
9047 tcg_temp_free_i32(tmp);
9051 tcg_gen_xor_i32(tmp, tmp, tmp2);
9054 tcg_temp_free_i32(tmp);
9058 gen_sub_CC(tmp, tmp, tmp2);
9060 tcg_temp_free_i32(tmp);
9064 gen_add_CC(tmp, tmp, tmp2);
9066 tcg_temp_free_i32(tmp);
9069 tcg_gen_or_i32(tmp, tmp, tmp2);
9073 store_reg_bx(s, rd, tmp);
9076 if (logic_cc && rd == 15) {
9077 /* MOVS r15, ... is used for exception return. */
9081 gen_exception_return(s, tmp2);
9086 store_reg_bx(s, rd, tmp2);
9090 tcg_gen_andc_i32(tmp, tmp, tmp2);
9094 store_reg_bx(s, rd, tmp);
9098 tcg_gen_not_i32(tmp2, tmp2);
9102 store_reg_bx(s, rd, tmp2);
9105 if (op1 != 0x0f && op1 != 0x0d) {
9106 tcg_temp_free_i32(tmp2);
9109 /* other instructions */
9110 op1 = (insn >> 24) & 0xf;
9114 /* multiplies, extra load/stores */
9115 sh = (insn >> 5) & 3;
9118 rd = (insn >> 16) & 0xf;
9119 rn = (insn >> 12) & 0xf;
9120 rs = (insn >> 8) & 0xf;
9122 op1 = (insn >> 20) & 0xf;
9124 case 0: case 1: case 2: case 3: case 6:
9126 tmp = load_reg(s, rs);
9127 tmp2 = load_reg(s, rm);
9128 tcg_gen_mul_i32(tmp, tmp, tmp2);
9129 tcg_temp_free_i32(tmp2);
9130 if (insn & (1 << 22)) {
9131 /* Subtract (mls) */
9133 tmp2 = load_reg(s, rn);
9134 tcg_gen_sub_i32(tmp, tmp2, tmp);
9135 tcg_temp_free_i32(tmp2);
9136 } else if (insn & (1 << 21)) {
9138 tmp2 = load_reg(s, rn);
9139 tcg_gen_add_i32(tmp, tmp, tmp2);
9140 tcg_temp_free_i32(tmp2);
9142 if (insn & (1 << 20))
9144 store_reg(s, rd, tmp);
9147 /* 64 bit mul double accumulate (UMAAL) */
9149 tmp = load_reg(s, rs);
9150 tmp2 = load_reg(s, rm);
9151 tmp64 = gen_mulu_i64_i32(tmp, tmp2);
9152 gen_addq_lo(s, tmp64, rn);
9153 gen_addq_lo(s, tmp64, rd);
9154 gen_storeq_reg(s, rn, rd, tmp64);
9155 tcg_temp_free_i64(tmp64);
9157 case 8: case 9: case 10: case 11:
9158 case 12: case 13: case 14: case 15:
9159 /* 64 bit mul: UMULL, UMLAL, SMULL, SMLAL. */
9160 tmp = load_reg(s, rs);
9161 tmp2 = load_reg(s, rm);
9162 if (insn & (1 << 22)) {
9163 tcg_gen_muls2_i32(tmp, tmp2, tmp, tmp2);
9165 tcg_gen_mulu2_i32(tmp, tmp2, tmp, tmp2);
9167 if (insn & (1 << 21)) { /* mult accumulate */
9168 TCGv_i32 al = load_reg(s, rn);
9169 TCGv_i32 ah = load_reg(s, rd);
9170 tcg_gen_add2_i32(tmp, tmp2, tmp, tmp2, al, ah);
9171 tcg_temp_free_i32(al);
9172 tcg_temp_free_i32(ah);
9174 if (insn & (1 << 20)) {
9175 gen_logicq_cc(tmp, tmp2);
9177 store_reg(s, rn, tmp);
9178 store_reg(s, rd, tmp2);
9184 rn = (insn >> 16) & 0xf;
9185 rd = (insn >> 12) & 0xf;
9186 if (insn & (1 << 23)) {
9187 /* load/store exclusive */
9188 int op2 = (insn >> 8) & 3;
9189 op1 = (insn >> 21) & 0x3;
9192 case 0: /* lda/stl */
9198 case 1: /* reserved */
9200 case 2: /* ldaex/stlex */
9203 case 3: /* ldrex/strex */
9212 addr = tcg_temp_local_new_i32();
9213 load_reg_var(s, addr, rn);
9215 /* Since the emulation does not have barriers,
9216 the acquire/release semantics need no special
9219 if (insn & (1 << 20)) {
9220 tmp = tcg_temp_new_i32();
9223 gen_aa32_ld32u_iss(s, tmp, addr,
9228 gen_aa32_ld8u_iss(s, tmp, addr,
9233 gen_aa32_ld16u_iss(s, tmp, addr,
9240 store_reg(s, rd, tmp);
9243 tmp = load_reg(s, rm);
9246 gen_aa32_st32_iss(s, tmp, addr,
9251 gen_aa32_st8_iss(s, tmp, addr,
9256 gen_aa32_st16_iss(s, tmp, addr,
9263 tcg_temp_free_i32(tmp);
9265 } else if (insn & (1 << 20)) {
9268 gen_load_exclusive(s, rd, 15, addr, 2);
9270 case 1: /* ldrexd */
9271 gen_load_exclusive(s, rd, rd + 1, addr, 3);
9273 case 2: /* ldrexb */
9274 gen_load_exclusive(s, rd, 15, addr, 0);
9276 case 3: /* ldrexh */
9277 gen_load_exclusive(s, rd, 15, addr, 1);
9286 gen_store_exclusive(s, rd, rm, 15, addr, 2);
9288 case 1: /* strexd */
9289 gen_store_exclusive(s, rd, rm, rm + 1, addr, 3);
9291 case 2: /* strexb */
9292 gen_store_exclusive(s, rd, rm, 15, addr, 0);
9294 case 3: /* strexh */
9295 gen_store_exclusive(s, rd, rm, 15, addr, 1);
9301 tcg_temp_free_i32(addr);
9302 } else if ((insn & 0x00300f00) == 0) {
9303 /* 0bcccc_0001_0x00_xxxx_xxxx_0000_1001_xxxx
9308 TCGMemOp opc = s->be_data;
9312 if (insn & (1 << 22)) {
9315 opc |= MO_UL | MO_ALIGN;
9318 addr = load_reg(s, rn);
9319 taddr = gen_aa32_addr(s, addr, opc);
9320 tcg_temp_free_i32(addr);
9322 tmp = load_reg(s, rm);
9323 tcg_gen_atomic_xchg_i32(tmp, taddr, tmp,
9324 get_mem_index(s), opc);
9325 tcg_temp_free(taddr);
9326 store_reg(s, rd, tmp);
9333 bool load = insn & (1 << 20);
9334 bool wbit = insn & (1 << 21);
9335 bool pbit = insn & (1 << 24);
9336 bool doubleword = false;
9339 /* Misc load/store */
9340 rn = (insn >> 16) & 0xf;
9341 rd = (insn >> 12) & 0xf;
9343 /* ISS not valid if writeback */
9344 issinfo = (pbit & !wbit) ? rd : ISSInvalid;
9346 if (!load && (sh & 2)) {
9350 /* UNPREDICTABLE; we choose to UNDEF */
9353 load = (sh & 1) == 0;
9357 addr = load_reg(s, rn);
9359 gen_add_datah_offset(s, insn, 0, addr);
9366 tmp = load_reg(s, rd);
9367 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
9368 tcg_temp_free_i32(tmp);
9369 tcg_gen_addi_i32(addr, addr, 4);
9370 tmp = load_reg(s, rd + 1);
9371 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
9372 tcg_temp_free_i32(tmp);
9375 tmp = tcg_temp_new_i32();
9376 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
9377 store_reg(s, rd, tmp);
9378 tcg_gen_addi_i32(addr, addr, 4);
9379 tmp = tcg_temp_new_i32();
9380 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
9383 address_offset = -4;
9386 tmp = tcg_temp_new_i32();
9389 gen_aa32_ld16u_iss(s, tmp, addr, get_mem_index(s),
9393 gen_aa32_ld8s_iss(s, tmp, addr, get_mem_index(s),
9398 gen_aa32_ld16s_iss(s, tmp, addr, get_mem_index(s),
9404 tmp = load_reg(s, rd);
9405 gen_aa32_st16_iss(s, tmp, addr, get_mem_index(s), issinfo);
9406 tcg_temp_free_i32(tmp);
9408 /* Perform base writeback before the loaded value to
9409 ensure correct behavior with overlapping index registers.
9410 ldrd with base writeback is undefined if the
9411 destination and index registers overlap. */
9413 gen_add_datah_offset(s, insn, address_offset, addr);
9414 store_reg(s, rn, addr);
9417 tcg_gen_addi_i32(addr, addr, address_offset);
9418 store_reg(s, rn, addr);
9420 tcg_temp_free_i32(addr);
9423 /* Complete the load. */
9424 store_reg(s, rd, tmp);
9433 if (insn & (1 << 4)) {
9435 /* Armv6 Media instructions. */
9437 rn = (insn >> 16) & 0xf;
9438 rd = (insn >> 12) & 0xf;
9439 rs = (insn >> 8) & 0xf;
9440 switch ((insn >> 23) & 3) {
9441 case 0: /* Parallel add/subtract. */
9442 op1 = (insn >> 20) & 7;
9443 tmp = load_reg(s, rn);
9444 tmp2 = load_reg(s, rm);
9445 sh = (insn >> 5) & 7;
9446 if ((op1 & 3) == 0 || sh == 5 || sh == 6)
9448 gen_arm_parallel_addsub(op1, sh, tmp, tmp2);
9449 tcg_temp_free_i32(tmp2);
9450 store_reg(s, rd, tmp);
9453 if ((insn & 0x00700020) == 0) {
9454 /* Halfword pack. */
9455 tmp = load_reg(s, rn);
9456 tmp2 = load_reg(s, rm);
9457 shift = (insn >> 7) & 0x1f;
9458 if (insn & (1 << 6)) {
9462 tcg_gen_sari_i32(tmp2, tmp2, shift);
9463 tcg_gen_andi_i32(tmp, tmp, 0xffff0000);
9464 tcg_gen_ext16u_i32(tmp2, tmp2);
9468 tcg_gen_shli_i32(tmp2, tmp2, shift);
9469 tcg_gen_ext16u_i32(tmp, tmp);
9470 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
9472 tcg_gen_or_i32(tmp, tmp, tmp2);
9473 tcg_temp_free_i32(tmp2);
9474 store_reg(s, rd, tmp);
9475 } else if ((insn & 0x00200020) == 0x00200000) {
9477 tmp = load_reg(s, rm);
9478 shift = (insn >> 7) & 0x1f;
9479 if (insn & (1 << 6)) {
9482 tcg_gen_sari_i32(tmp, tmp, shift);
9484 tcg_gen_shli_i32(tmp, tmp, shift);
9486 sh = (insn >> 16) & 0x1f;
9487 tmp2 = tcg_const_i32(sh);
9488 if (insn & (1 << 22))
9489 gen_helper_usat(tmp, cpu_env, tmp, tmp2);
9491 gen_helper_ssat(tmp, cpu_env, tmp, tmp2);
9492 tcg_temp_free_i32(tmp2);
9493 store_reg(s, rd, tmp);
9494 } else if ((insn & 0x00300fe0) == 0x00200f20) {
9496 tmp = load_reg(s, rm);
9497 sh = (insn >> 16) & 0x1f;
9498 tmp2 = tcg_const_i32(sh);
9499 if (insn & (1 << 22))
9500 gen_helper_usat16(tmp, cpu_env, tmp, tmp2);
9502 gen_helper_ssat16(tmp, cpu_env, tmp, tmp2);
9503 tcg_temp_free_i32(tmp2);
9504 store_reg(s, rd, tmp);
9505 } else if ((insn & 0x00700fe0) == 0x00000fa0) {
9507 tmp = load_reg(s, rn);
9508 tmp2 = load_reg(s, rm);
9509 tmp3 = tcg_temp_new_i32();
9510 tcg_gen_ld_i32(tmp3, cpu_env, offsetof(CPUARMState, GE));
9511 gen_helper_sel_flags(tmp, tmp3, tmp, tmp2);
9512 tcg_temp_free_i32(tmp3);
9513 tcg_temp_free_i32(tmp2);
9514 store_reg(s, rd, tmp);
9515 } else if ((insn & 0x000003e0) == 0x00000060) {
9516 tmp = load_reg(s, rm);
9517 shift = (insn >> 10) & 3;
9518 /* ??? In many cases it's not necessary to do a
9519 rotate, a shift is sufficient. */
9521 tcg_gen_rotri_i32(tmp, tmp, shift * 8);
9522 op1 = (insn >> 20) & 7;
9524 case 0: gen_sxtb16(tmp); break;
9525 case 2: gen_sxtb(tmp); break;
9526 case 3: gen_sxth(tmp); break;
9527 case 4: gen_uxtb16(tmp); break;
9528 case 6: gen_uxtb(tmp); break;
9529 case 7: gen_uxth(tmp); break;
9530 default: goto illegal_op;
9533 tmp2 = load_reg(s, rn);
9534 if ((op1 & 3) == 0) {
9535 gen_add16(tmp, tmp2);
9537 tcg_gen_add_i32(tmp, tmp, tmp2);
9538 tcg_temp_free_i32(tmp2);
9541 store_reg(s, rd, tmp);
9542 } else if ((insn & 0x003f0f60) == 0x003f0f20) {
9544 tmp = load_reg(s, rm);
9545 if (insn & (1 << 22)) {
9546 if (insn & (1 << 7)) {
9550 gen_helper_rbit(tmp, tmp);
9553 if (insn & (1 << 7))
9556 tcg_gen_bswap32_i32(tmp, tmp);
9558 store_reg(s, rd, tmp);
9563 case 2: /* Multiplies (Type 3). */
9564 switch ((insn >> 20) & 0x7) {
9566 if (((insn >> 6) ^ (insn >> 7)) & 1) {
9567 /* op2 not 00x or 11x : UNDEF */
9570 /* Signed multiply most significant [accumulate].
9571 (SMMUL, SMMLA, SMMLS) */
9572 tmp = load_reg(s, rm);
9573 tmp2 = load_reg(s, rs);
9574 tmp64 = gen_muls_i64_i32(tmp, tmp2);
9577 tmp = load_reg(s, rd);
9578 if (insn & (1 << 6)) {
9579 tmp64 = gen_subq_msw(tmp64, tmp);
9581 tmp64 = gen_addq_msw(tmp64, tmp);
9584 if (insn & (1 << 5)) {
9585 tcg_gen_addi_i64(tmp64, tmp64, 0x80000000u);
9587 tcg_gen_shri_i64(tmp64, tmp64, 32);
9588 tmp = tcg_temp_new_i32();
9589 tcg_gen_extrl_i64_i32(tmp, tmp64);
9590 tcg_temp_free_i64(tmp64);
9591 store_reg(s, rn, tmp);
9595 /* SMLAD, SMUAD, SMLSD, SMUSD, SMLALD, SMLSLD */
9596 if (insn & (1 << 7)) {
9599 tmp = load_reg(s, rm);
9600 tmp2 = load_reg(s, rs);
9601 if (insn & (1 << 5))
9602 gen_swap_half(tmp2);
9603 gen_smul_dual(tmp, tmp2);
9604 if (insn & (1 << 22)) {
9605 /* smlald, smlsld */
9608 tmp64 = tcg_temp_new_i64();
9609 tmp64_2 = tcg_temp_new_i64();
9610 tcg_gen_ext_i32_i64(tmp64, tmp);
9611 tcg_gen_ext_i32_i64(tmp64_2, tmp2);
9612 tcg_temp_free_i32(tmp);
9613 tcg_temp_free_i32(tmp2);
9614 if (insn & (1 << 6)) {
9615 tcg_gen_sub_i64(tmp64, tmp64, tmp64_2);
9617 tcg_gen_add_i64(tmp64, tmp64, tmp64_2);
9619 tcg_temp_free_i64(tmp64_2);
9620 gen_addq(s, tmp64, rd, rn);
9621 gen_storeq_reg(s, rd, rn, tmp64);
9622 tcg_temp_free_i64(tmp64);
9624 /* smuad, smusd, smlad, smlsd */
9625 if (insn & (1 << 6)) {
9626 /* This subtraction cannot overflow. */
9627 tcg_gen_sub_i32(tmp, tmp, tmp2);
9629 /* This addition cannot overflow 32 bits;
9630 * however it may overflow considered as a
9631 * signed operation, in which case we must set
9634 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
9636 tcg_temp_free_i32(tmp2);
9639 tmp2 = load_reg(s, rd);
9640 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
9641 tcg_temp_free_i32(tmp2);
9643 store_reg(s, rn, tmp);
9649 if (!arm_dc_feature(s, ARM_FEATURE_ARM_DIV)) {
9652 if (((insn >> 5) & 7) || (rd != 15)) {
9655 tmp = load_reg(s, rm);
9656 tmp2 = load_reg(s, rs);
9657 if (insn & (1 << 21)) {
9658 gen_helper_udiv(tmp, tmp, tmp2);
9660 gen_helper_sdiv(tmp, tmp, tmp2);
9662 tcg_temp_free_i32(tmp2);
9663 store_reg(s, rn, tmp);
9670 op1 = ((insn >> 17) & 0x38) | ((insn >> 5) & 7);
9672 case 0: /* Unsigned sum of absolute differences. */
9674 tmp = load_reg(s, rm);
9675 tmp2 = load_reg(s, rs);
9676 gen_helper_usad8(tmp, tmp, tmp2);
9677 tcg_temp_free_i32(tmp2);
9679 tmp2 = load_reg(s, rd);
9680 tcg_gen_add_i32(tmp, tmp, tmp2);
9681 tcg_temp_free_i32(tmp2);
9683 store_reg(s, rn, tmp);
9685 case 0x20: case 0x24: case 0x28: case 0x2c:
9686 /* Bitfield insert/clear. */
9688 shift = (insn >> 7) & 0x1f;
9689 i = (insn >> 16) & 0x1f;
9691 /* UNPREDICTABLE; we choose to UNDEF */
9696 tmp = tcg_temp_new_i32();
9697 tcg_gen_movi_i32(tmp, 0);
9699 tmp = load_reg(s, rm);
9702 tmp2 = load_reg(s, rd);
9703 tcg_gen_deposit_i32(tmp, tmp2, tmp, shift, i);
9704 tcg_temp_free_i32(tmp2);
9706 store_reg(s, rd, tmp);
9708 case 0x12: case 0x16: case 0x1a: case 0x1e: /* sbfx */
9709 case 0x32: case 0x36: case 0x3a: case 0x3e: /* ubfx */
9711 tmp = load_reg(s, rm);
9712 shift = (insn >> 7) & 0x1f;
9713 i = ((insn >> 16) & 0x1f) + 1;
9718 tcg_gen_extract_i32(tmp, tmp, shift, i);
9720 tcg_gen_sextract_i32(tmp, tmp, shift, i);
9723 store_reg(s, rd, tmp);
9733 /* Check for undefined extension instructions
9734 * per the ARM Bible IE:
9735 * xxxx 0111 1111 xxxx xxxx xxxx 1111 xxxx
9737 sh = (0xf << 20) | (0xf << 4);
9738 if (op1 == 0x7 && ((insn & sh) == sh))
9742 /* load/store byte/word */
9743 rn = (insn >> 16) & 0xf;
9744 rd = (insn >> 12) & 0xf;
9745 tmp2 = load_reg(s, rn);
9746 if ((insn & 0x01200000) == 0x00200000) {
9748 i = get_a32_user_mem_index(s);
9750 i = get_mem_index(s);
9752 if (insn & (1 << 24))
9753 gen_add_data_offset(s, insn, tmp2);
9754 if (insn & (1 << 20)) {
9756 tmp = tcg_temp_new_i32();
9757 if (insn & (1 << 22)) {
9758 gen_aa32_ld8u_iss(s, tmp, tmp2, i, rd);
9760 gen_aa32_ld32u_iss(s, tmp, tmp2, i, rd);
9764 tmp = load_reg(s, rd);
9765 if (insn & (1 << 22)) {
9766 gen_aa32_st8_iss(s, tmp, tmp2, i, rd);
9768 gen_aa32_st32_iss(s, tmp, tmp2, i, rd);
9770 tcg_temp_free_i32(tmp);
9772 if (!(insn & (1 << 24))) {
9773 gen_add_data_offset(s, insn, tmp2);
9774 store_reg(s, rn, tmp2);
9775 } else if (insn & (1 << 21)) {
9776 store_reg(s, rn, tmp2);
9778 tcg_temp_free_i32(tmp2);
9780 if (insn & (1 << 20)) {
9781 /* Complete the load. */
9782 store_reg_from_load(s, rd, tmp);
9788 int j, n, loaded_base;
9789 bool exc_return = false;
9790 bool is_load = extract32(insn, 20, 1);
9792 TCGv_i32 loaded_var;
9793 /* load/store multiple words */
9794 /* XXX: store correct base if write back */
9795 if (insn & (1 << 22)) {
9796 /* LDM (user), LDM (exception return) and STM (user) */
9798 goto illegal_op; /* only usable in supervisor mode */
9800 if (is_load && extract32(insn, 15, 1)) {
9806 rn = (insn >> 16) & 0xf;
9807 addr = load_reg(s, rn);
9809 /* compute total size */
9814 if (insn & (1 << i))
9817 /* XXX: test invalid n == 0 case ? */
9818 if (insn & (1 << 23)) {
9819 if (insn & (1 << 24)) {
9821 tcg_gen_addi_i32(addr, addr, 4);
9823 /* post increment */
9826 if (insn & (1 << 24)) {
9828 tcg_gen_addi_i32(addr, addr, -(n * 4));
9830 /* post decrement */
9832 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
9837 if (insn & (1 << i)) {
9840 tmp = tcg_temp_new_i32();
9841 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
9843 tmp2 = tcg_const_i32(i);
9844 gen_helper_set_user_reg(cpu_env, tmp2, tmp);
9845 tcg_temp_free_i32(tmp2);
9846 tcg_temp_free_i32(tmp);
9847 } else if (i == rn) {
9850 } else if (rn == 15 && exc_return) {
9851 store_pc_exc_ret(s, tmp);
9853 store_reg_from_load(s, i, tmp);
9858 /* special case: r15 = PC + 8 */
9859 val = (long)s->pc + 4;
9860 tmp = tcg_temp_new_i32();
9861 tcg_gen_movi_i32(tmp, val);
9863 tmp = tcg_temp_new_i32();
9864 tmp2 = tcg_const_i32(i);
9865 gen_helper_get_user_reg(tmp, cpu_env, tmp2);
9866 tcg_temp_free_i32(tmp2);
9868 tmp = load_reg(s, i);
9870 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
9871 tcg_temp_free_i32(tmp);
9874 /* no need to add after the last transfer */
9876 tcg_gen_addi_i32(addr, addr, 4);
9879 if (insn & (1 << 21)) {
9881 if (insn & (1 << 23)) {
9882 if (insn & (1 << 24)) {
9885 /* post increment */
9886 tcg_gen_addi_i32(addr, addr, 4);
9889 if (insn & (1 << 24)) {
9892 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
9894 /* post decrement */
9895 tcg_gen_addi_i32(addr, addr, -(n * 4));
9898 store_reg(s, rn, addr);
9900 tcg_temp_free_i32(addr);
9903 store_reg(s, rn, loaded_var);
9906 /* Restore CPSR from SPSR. */
9907 tmp = load_cpu_field(spsr);
9908 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
9911 gen_helper_cpsr_write_eret(cpu_env, tmp);
9912 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
9915 tcg_temp_free_i32(tmp);
9916 /* Must exit loop to check un-masked IRQs */
9917 s->base.is_jmp = DISAS_EXIT;
9926 /* branch (and link) */
9927 val = (int32_t)s->pc;
9928 if (insn & (1 << 24)) {
9929 tmp = tcg_temp_new_i32();
9930 tcg_gen_movi_i32(tmp, val);
9931 store_reg(s, 14, tmp);
9933 offset = sextract32(insn << 2, 0, 26);
9941 if (((insn >> 8) & 0xe) == 10) {
9943 if (disas_vfp_insn(s, insn)) {
9946 } else if (disas_coproc_insn(s, insn)) {
9953 gen_set_pc_im(s, s->pc);
9954 s->svc_imm = extract32(insn, 0, 24);
9955 s->base.is_jmp = DISAS_SWI;
9959 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(),
9960 default_exception_el(s));
9966 static bool thumb_insn_is_16bit(DisasContext *s, uint32_t insn)
9968 /* Return true if this is a 16 bit instruction. We must be precise
9969 * about this (matching the decode). We assume that s->pc still
9970 * points to the first 16 bits of the insn.
9972 if ((insn >> 11) < 0x1d) {
9973 /* Definitely a 16-bit instruction */
9977 /* Top five bits 0b11101 / 0b11110 / 0b11111 : this is the
9978 * first half of a 32-bit Thumb insn. Thumb-1 cores might
9979 * end up actually treating this as two 16-bit insns, though,
9980 * if it's half of a bl/blx pair that might span a page boundary.
9982 if (arm_dc_feature(s, ARM_FEATURE_THUMB2) ||
9983 arm_dc_feature(s, ARM_FEATURE_M)) {
9984 /* Thumb2 cores (including all M profile ones) always treat
9985 * 32-bit insns as 32-bit.
9990 if ((insn >> 11) == 0x1e && s->pc - s->page_start < TARGET_PAGE_SIZE - 3) {
9991 /* 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix, and the suffix
9992 * is not on the next page; we merge this into a 32-bit
9997 /* 0b1110_1xxx_xxxx_xxxx : BLX suffix (or UNDEF);
9998 * 0b1111_1xxx_xxxx_xxxx : BL suffix;
9999 * 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix on the end of a page
10000 * -- handle as single 16 bit insn
10005 /* Return true if this is a Thumb-2 logical op. */
10007 thumb2_logic_op(int op)
10012 /* Generate code for a Thumb-2 data processing operation. If CONDS is nonzero
10013 then set condition code flags based on the result of the operation.
10014 If SHIFTER_OUT is nonzero then set the carry flag for logical operations
10015 to the high bit of T1.
10016 Returns zero if the opcode is valid. */
10019 gen_thumb2_data_op(DisasContext *s, int op, int conds, uint32_t shifter_out,
10020 TCGv_i32 t0, TCGv_i32 t1)
10027 tcg_gen_and_i32(t0, t0, t1);
10031 tcg_gen_andc_i32(t0, t0, t1);
10035 tcg_gen_or_i32(t0, t0, t1);
10039 tcg_gen_orc_i32(t0, t0, t1);
10043 tcg_gen_xor_i32(t0, t0, t1);
10048 gen_add_CC(t0, t0, t1);
10050 tcg_gen_add_i32(t0, t0, t1);
10054 gen_adc_CC(t0, t0, t1);
10060 gen_sbc_CC(t0, t0, t1);
10062 gen_sub_carry(t0, t0, t1);
10067 gen_sub_CC(t0, t0, t1);
10069 tcg_gen_sub_i32(t0, t0, t1);
10073 gen_sub_CC(t0, t1, t0);
10075 tcg_gen_sub_i32(t0, t1, t0);
10077 default: /* 5, 6, 7, 9, 12, 15. */
10083 gen_set_CF_bit31(t1);
10088 /* Translate a 32-bit thumb instruction. */
10089 static void disas_thumb2_insn(DisasContext *s, uint32_t insn)
10091 uint32_t imm, shift, offset;
10092 uint32_t rd, rn, rm, rs;
10104 * ARMv6-M supports a limited subset of Thumb2 instructions.
10105 * Other Thumb1 architectures allow only 32-bit
10106 * combined BL/BLX prefix and suffix.
10108 if (arm_dc_feature(s, ARM_FEATURE_M) &&
10109 !arm_dc_feature(s, ARM_FEATURE_V7)) {
10111 bool found = false;
10112 static const uint32_t armv6m_insn[] = {0xf3808000 /* msr */,
10113 0xf3b08040 /* dsb */,
10114 0xf3b08050 /* dmb */,
10115 0xf3b08060 /* isb */,
10116 0xf3e08000 /* mrs */,
10117 0xf000d000 /* bl */};
10118 static const uint32_t armv6m_mask[] = {0xffe0d000,
10125 for (i = 0; i < ARRAY_SIZE(armv6m_insn); i++) {
10126 if ((insn & armv6m_mask[i]) == armv6m_insn[i]) {
10134 } else if ((insn & 0xf800e800) != 0xf000e800) {
10138 rn = (insn >> 16) & 0xf;
10139 rs = (insn >> 12) & 0xf;
10140 rd = (insn >> 8) & 0xf;
10142 switch ((insn >> 25) & 0xf) {
10143 case 0: case 1: case 2: case 3:
10144 /* 16-bit instructions. Should never happen. */
10147 if (insn & (1 << 22)) {
10148 /* 0b1110_100x_x1xx_xxxx_xxxx_xxxx_xxxx_xxxx
10149 * - load/store doubleword, load/store exclusive, ldacq/strel,
10150 * table branch, TT.
10152 if (insn == 0xe97fe97f && arm_dc_feature(s, ARM_FEATURE_M) &&
10153 arm_dc_feature(s, ARM_FEATURE_V8)) {
10154 /* 0b1110_1001_0111_1111_1110_1001_0111_111
10156 * The bulk of the behaviour for this instruction is implemented
10157 * in v7m_handle_execute_nsc(), which deals with the insn when
10158 * it is executed by a CPU in non-secure state from memory
10159 * which is Secure & NonSecure-Callable.
10160 * Here we only need to handle the remaining cases:
10161 * * in NS memory (including the "security extension not
10162 * implemented" case) : NOP
10163 * * in S memory but CPU already secure (clear IT bits)
10164 * We know that the attribute for the memory this insn is
10165 * in must match the current CPU state, because otherwise
10166 * get_phys_addr_pmsav8 would have generated an exception.
10168 if (s->v8m_secure) {
10169 /* Like the IT insn, we don't need to generate any code */
10170 s->condexec_cond = 0;
10171 s->condexec_mask = 0;
10173 } else if (insn & 0x01200000) {
10174 /* 0b1110_1000_x11x_xxxx_xxxx_xxxx_xxxx_xxxx
10175 * - load/store dual (post-indexed)
10176 * 0b1111_1001_x10x_xxxx_xxxx_xxxx_xxxx_xxxx
10177 * - load/store dual (literal and immediate)
10178 * 0b1111_1001_x11x_xxxx_xxxx_xxxx_xxxx_xxxx
10179 * - load/store dual (pre-indexed)
10182 if (insn & (1 << 21)) {
10183 /* UNPREDICTABLE */
10186 addr = tcg_temp_new_i32();
10187 tcg_gen_movi_i32(addr, s->pc & ~3);
10189 addr = load_reg(s, rn);
10191 offset = (insn & 0xff) * 4;
10192 if ((insn & (1 << 23)) == 0)
10194 if (insn & (1 << 24)) {
10195 tcg_gen_addi_i32(addr, addr, offset);
10198 if (insn & (1 << 20)) {
10200 tmp = tcg_temp_new_i32();
10201 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
10202 store_reg(s, rs, tmp);
10203 tcg_gen_addi_i32(addr, addr, 4);
10204 tmp = tcg_temp_new_i32();
10205 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
10206 store_reg(s, rd, tmp);
10209 tmp = load_reg(s, rs);
10210 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
10211 tcg_temp_free_i32(tmp);
10212 tcg_gen_addi_i32(addr, addr, 4);
10213 tmp = load_reg(s, rd);
10214 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
10215 tcg_temp_free_i32(tmp);
10217 if (insn & (1 << 21)) {
10218 /* Base writeback. */
10219 tcg_gen_addi_i32(addr, addr, offset - 4);
10220 store_reg(s, rn, addr);
10222 tcg_temp_free_i32(addr);
10224 } else if ((insn & (1 << 23)) == 0) {
10225 /* 0b1110_1000_010x_xxxx_xxxx_xxxx_xxxx_xxxx
10226 * - load/store exclusive word
10230 if (!(insn & (1 << 20)) &&
10231 arm_dc_feature(s, ARM_FEATURE_M) &&
10232 arm_dc_feature(s, ARM_FEATURE_V8)) {
10233 /* 0b1110_1000_0100_xxxx_1111_xxxx_xxxx_xxxx
10236 bool alt = insn & (1 << 7);
10237 TCGv_i32 addr, op, ttresp;
10239 if ((insn & 0x3f) || rd == 13 || rd == 15 || rn == 15) {
10240 /* we UNDEF for these UNPREDICTABLE cases */
10244 if (alt && !s->v8m_secure) {
10248 addr = load_reg(s, rn);
10249 op = tcg_const_i32(extract32(insn, 6, 2));
10250 ttresp = tcg_temp_new_i32();
10251 gen_helper_v7m_tt(ttresp, cpu_env, addr, op);
10252 tcg_temp_free_i32(addr);
10253 tcg_temp_free_i32(op);
10254 store_reg(s, rd, ttresp);
10259 addr = tcg_temp_local_new_i32();
10260 load_reg_var(s, addr, rn);
10261 tcg_gen_addi_i32(addr, addr, (insn & 0xff) << 2);
10262 if (insn & (1 << 20)) {
10263 gen_load_exclusive(s, rs, 15, addr, 2);
10265 gen_store_exclusive(s, rd, rs, 15, addr, 2);
10267 tcg_temp_free_i32(addr);
10268 } else if ((insn & (7 << 5)) == 0) {
10269 /* Table Branch. */
10271 addr = tcg_temp_new_i32();
10272 tcg_gen_movi_i32(addr, s->pc);
10274 addr = load_reg(s, rn);
10276 tmp = load_reg(s, rm);
10277 tcg_gen_add_i32(addr, addr, tmp);
10278 if (insn & (1 << 4)) {
10280 tcg_gen_add_i32(addr, addr, tmp);
10281 tcg_temp_free_i32(tmp);
10282 tmp = tcg_temp_new_i32();
10283 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
10285 tcg_temp_free_i32(tmp);
10286 tmp = tcg_temp_new_i32();
10287 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
10289 tcg_temp_free_i32(addr);
10290 tcg_gen_shli_i32(tmp, tmp, 1);
10291 tcg_gen_addi_i32(tmp, tmp, s->pc);
10292 store_reg(s, 15, tmp);
10294 int op2 = (insn >> 6) & 0x3;
10295 op = (insn >> 4) & 0x3;
10300 /* Load/store exclusive byte/halfword/doubleword */
10307 /* Load-acquire/store-release */
10313 /* Load-acquire/store-release exclusive */
10317 addr = tcg_temp_local_new_i32();
10318 load_reg_var(s, addr, rn);
10320 if (insn & (1 << 20)) {
10321 tmp = tcg_temp_new_i32();
10324 gen_aa32_ld8u_iss(s, tmp, addr, get_mem_index(s),
10328 gen_aa32_ld16u_iss(s, tmp, addr, get_mem_index(s),
10332 gen_aa32_ld32u_iss(s, tmp, addr, get_mem_index(s),
10338 store_reg(s, rs, tmp);
10340 tmp = load_reg(s, rs);
10343 gen_aa32_st8_iss(s, tmp, addr, get_mem_index(s),
10347 gen_aa32_st16_iss(s, tmp, addr, get_mem_index(s),
10351 gen_aa32_st32_iss(s, tmp, addr, get_mem_index(s),
10357 tcg_temp_free_i32(tmp);
10359 } else if (insn & (1 << 20)) {
10360 gen_load_exclusive(s, rs, rd, addr, op);
10362 gen_store_exclusive(s, rm, rs, rd, addr, op);
10364 tcg_temp_free_i32(addr);
10367 /* Load/store multiple, RFE, SRS. */
10368 if (((insn >> 23) & 1) == ((insn >> 24) & 1)) {
10369 /* RFE, SRS: not available in user mode or on M profile */
10370 if (IS_USER(s) || arm_dc_feature(s, ARM_FEATURE_M)) {
10373 if (insn & (1 << 20)) {
10375 addr = load_reg(s, rn);
10376 if ((insn & (1 << 24)) == 0)
10377 tcg_gen_addi_i32(addr, addr, -8);
10378 /* Load PC into tmp and CPSR into tmp2. */
10379 tmp = tcg_temp_new_i32();
10380 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
10381 tcg_gen_addi_i32(addr, addr, 4);
10382 tmp2 = tcg_temp_new_i32();
10383 gen_aa32_ld32u(s, tmp2, addr, get_mem_index(s));
10384 if (insn & (1 << 21)) {
10385 /* Base writeback. */
10386 if (insn & (1 << 24)) {
10387 tcg_gen_addi_i32(addr, addr, 4);
10389 tcg_gen_addi_i32(addr, addr, -4);
10391 store_reg(s, rn, addr);
10393 tcg_temp_free_i32(addr);
10395 gen_rfe(s, tmp, tmp2);
10398 gen_srs(s, (insn & 0x1f), (insn & (1 << 24)) ? 1 : 2,
10402 int i, loaded_base = 0;
10403 TCGv_i32 loaded_var;
10404 /* Load/store multiple. */
10405 addr = load_reg(s, rn);
10407 for (i = 0; i < 16; i++) {
10408 if (insn & (1 << i))
10411 if (insn & (1 << 24)) {
10412 tcg_gen_addi_i32(addr, addr, -offset);
10416 for (i = 0; i < 16; i++) {
10417 if ((insn & (1 << i)) == 0)
10419 if (insn & (1 << 20)) {
10421 tmp = tcg_temp_new_i32();
10422 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
10424 gen_bx_excret(s, tmp);
10425 } else if (i == rn) {
10429 store_reg(s, i, tmp);
10433 tmp = load_reg(s, i);
10434 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
10435 tcg_temp_free_i32(tmp);
10437 tcg_gen_addi_i32(addr, addr, 4);
10440 store_reg(s, rn, loaded_var);
10442 if (insn & (1 << 21)) {
10443 /* Base register writeback. */
10444 if (insn & (1 << 24)) {
10445 tcg_gen_addi_i32(addr, addr, -offset);
10447 /* Fault if writeback register is in register list. */
10448 if (insn & (1 << rn))
10450 store_reg(s, rn, addr);
10452 tcg_temp_free_i32(addr);
10459 op = (insn >> 21) & 0xf;
10461 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10464 /* Halfword pack. */
10465 tmp = load_reg(s, rn);
10466 tmp2 = load_reg(s, rm);
10467 shift = ((insn >> 10) & 0x1c) | ((insn >> 6) & 0x3);
10468 if (insn & (1 << 5)) {
10472 tcg_gen_sari_i32(tmp2, tmp2, shift);
10473 tcg_gen_andi_i32(tmp, tmp, 0xffff0000);
10474 tcg_gen_ext16u_i32(tmp2, tmp2);
10478 tcg_gen_shli_i32(tmp2, tmp2, shift);
10479 tcg_gen_ext16u_i32(tmp, tmp);
10480 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
10482 tcg_gen_or_i32(tmp, tmp, tmp2);
10483 tcg_temp_free_i32(tmp2);
10484 store_reg(s, rd, tmp);
10486 /* Data processing register constant shift. */
10488 tmp = tcg_temp_new_i32();
10489 tcg_gen_movi_i32(tmp, 0);
10491 tmp = load_reg(s, rn);
10493 tmp2 = load_reg(s, rm);
10495 shiftop = (insn >> 4) & 3;
10496 shift = ((insn >> 6) & 3) | ((insn >> 10) & 0x1c);
10497 conds = (insn & (1 << 20)) != 0;
10498 logic_cc = (conds && thumb2_logic_op(op));
10499 gen_arm_shift_im(tmp2, shiftop, shift, logic_cc);
10500 if (gen_thumb2_data_op(s, op, conds, 0, tmp, tmp2))
10502 tcg_temp_free_i32(tmp2);
10504 store_reg(s, rd, tmp);
10506 tcg_temp_free_i32(tmp);
10510 case 13: /* Misc data processing. */
10511 op = ((insn >> 22) & 6) | ((insn >> 7) & 1);
10512 if (op < 4 && (insn & 0xf000) != 0xf000)
10515 case 0: /* Register controlled shift. */
10516 tmp = load_reg(s, rn);
10517 tmp2 = load_reg(s, rm);
10518 if ((insn & 0x70) != 0)
10520 op = (insn >> 21) & 3;
10521 logic_cc = (insn & (1 << 20)) != 0;
10522 gen_arm_shift_reg(tmp, op, tmp2, logic_cc);
10525 store_reg(s, rd, tmp);
10527 case 1: /* Sign/zero extend. */
10528 op = (insn >> 20) & 7;
10530 case 0: /* SXTAH, SXTH */
10531 case 1: /* UXTAH, UXTH */
10532 case 4: /* SXTAB, SXTB */
10533 case 5: /* UXTAB, UXTB */
10535 case 2: /* SXTAB16, SXTB16 */
10536 case 3: /* UXTAB16, UXTB16 */
10537 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10545 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10549 tmp = load_reg(s, rm);
10550 shift = (insn >> 4) & 3;
10551 /* ??? In many cases it's not necessary to do a
10552 rotate, a shift is sufficient. */
10554 tcg_gen_rotri_i32(tmp, tmp, shift * 8);
10555 op = (insn >> 20) & 7;
10557 case 0: gen_sxth(tmp); break;
10558 case 1: gen_uxth(tmp); break;
10559 case 2: gen_sxtb16(tmp); break;
10560 case 3: gen_uxtb16(tmp); break;
10561 case 4: gen_sxtb(tmp); break;
10562 case 5: gen_uxtb(tmp); break;
10564 g_assert_not_reached();
10567 tmp2 = load_reg(s, rn);
10568 if ((op >> 1) == 1) {
10569 gen_add16(tmp, tmp2);
10571 tcg_gen_add_i32(tmp, tmp, tmp2);
10572 tcg_temp_free_i32(tmp2);
10575 store_reg(s, rd, tmp);
10577 case 2: /* SIMD add/subtract. */
10578 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10581 op = (insn >> 20) & 7;
10582 shift = (insn >> 4) & 7;
10583 if ((op & 3) == 3 || (shift & 3) == 3)
10585 tmp = load_reg(s, rn);
10586 tmp2 = load_reg(s, rm);
10587 gen_thumb2_parallel_addsub(op, shift, tmp, tmp2);
10588 tcg_temp_free_i32(tmp2);
10589 store_reg(s, rd, tmp);
10591 case 3: /* Other data processing. */
10592 op = ((insn >> 17) & 0x38) | ((insn >> 4) & 7);
10594 /* Saturating add/subtract. */
10595 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10598 tmp = load_reg(s, rn);
10599 tmp2 = load_reg(s, rm);
10601 gen_helper_double_saturate(tmp, cpu_env, tmp);
10603 gen_helper_sub_saturate(tmp, cpu_env, tmp2, tmp);
10605 gen_helper_add_saturate(tmp, cpu_env, tmp, tmp2);
10606 tcg_temp_free_i32(tmp2);
10609 case 0x0a: /* rbit */
10610 case 0x08: /* rev */
10611 case 0x09: /* rev16 */
10612 case 0x0b: /* revsh */
10613 case 0x18: /* clz */
10615 case 0x10: /* sel */
10616 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10620 case 0x20: /* crc32/crc32c */
10626 if (!arm_dc_feature(s, ARM_FEATURE_CRC)) {
10633 tmp = load_reg(s, rn);
10635 case 0x0a: /* rbit */
10636 gen_helper_rbit(tmp, tmp);
10638 case 0x08: /* rev */
10639 tcg_gen_bswap32_i32(tmp, tmp);
10641 case 0x09: /* rev16 */
10644 case 0x0b: /* revsh */
10647 case 0x10: /* sel */
10648 tmp2 = load_reg(s, rm);
10649 tmp3 = tcg_temp_new_i32();
10650 tcg_gen_ld_i32(tmp3, cpu_env, offsetof(CPUARMState, GE));
10651 gen_helper_sel_flags(tmp, tmp3, tmp, tmp2);
10652 tcg_temp_free_i32(tmp3);
10653 tcg_temp_free_i32(tmp2);
10655 case 0x18: /* clz */
10656 tcg_gen_clzi_i32(tmp, tmp, 32);
10666 uint32_t sz = op & 0x3;
10667 uint32_t c = op & 0x8;
10669 tmp2 = load_reg(s, rm);
10671 tcg_gen_andi_i32(tmp2, tmp2, 0xff);
10672 } else if (sz == 1) {
10673 tcg_gen_andi_i32(tmp2, tmp2, 0xffff);
10675 tmp3 = tcg_const_i32(1 << sz);
10677 gen_helper_crc32c(tmp, tmp, tmp2, tmp3);
10679 gen_helper_crc32(tmp, tmp, tmp2, tmp3);
10681 tcg_temp_free_i32(tmp2);
10682 tcg_temp_free_i32(tmp3);
10686 g_assert_not_reached();
10689 store_reg(s, rd, tmp);
10691 case 4: case 5: /* 32-bit multiply. Sum of absolute differences. */
10692 switch ((insn >> 20) & 7) {
10693 case 0: /* 32 x 32 -> 32 */
10694 case 7: /* Unsigned sum of absolute differences. */
10696 case 1: /* 16 x 16 -> 32 */
10697 case 2: /* Dual multiply add. */
10698 case 3: /* 32 * 16 -> 32msb */
10699 case 4: /* Dual multiply subtract. */
10700 case 5: case 6: /* 32 * 32 -> 32msb (SMMUL, SMMLA, SMMLS) */
10701 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10706 op = (insn >> 4) & 0xf;
10707 tmp = load_reg(s, rn);
10708 tmp2 = load_reg(s, rm);
10709 switch ((insn >> 20) & 7) {
10710 case 0: /* 32 x 32 -> 32 */
10711 tcg_gen_mul_i32(tmp, tmp, tmp2);
10712 tcg_temp_free_i32(tmp2);
10714 tmp2 = load_reg(s, rs);
10716 tcg_gen_sub_i32(tmp, tmp2, tmp);
10718 tcg_gen_add_i32(tmp, tmp, tmp2);
10719 tcg_temp_free_i32(tmp2);
10722 case 1: /* 16 x 16 -> 32 */
10723 gen_mulxy(tmp, tmp2, op & 2, op & 1);
10724 tcg_temp_free_i32(tmp2);
10726 tmp2 = load_reg(s, rs);
10727 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
10728 tcg_temp_free_i32(tmp2);
10731 case 2: /* Dual multiply add. */
10732 case 4: /* Dual multiply subtract. */
10734 gen_swap_half(tmp2);
10735 gen_smul_dual(tmp, tmp2);
10736 if (insn & (1 << 22)) {
10737 /* This subtraction cannot overflow. */
10738 tcg_gen_sub_i32(tmp, tmp, tmp2);
10740 /* This addition cannot overflow 32 bits;
10741 * however it may overflow considered as a signed
10742 * operation, in which case we must set the Q flag.
10744 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
10746 tcg_temp_free_i32(tmp2);
10749 tmp2 = load_reg(s, rs);
10750 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
10751 tcg_temp_free_i32(tmp2);
10754 case 3: /* 32 * 16 -> 32msb */
10756 tcg_gen_sari_i32(tmp2, tmp2, 16);
10759 tmp64 = gen_muls_i64_i32(tmp, tmp2);
10760 tcg_gen_shri_i64(tmp64, tmp64, 16);
10761 tmp = tcg_temp_new_i32();
10762 tcg_gen_extrl_i64_i32(tmp, tmp64);
10763 tcg_temp_free_i64(tmp64);
10766 tmp2 = load_reg(s, rs);
10767 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
10768 tcg_temp_free_i32(tmp2);
10771 case 5: case 6: /* 32 * 32 -> 32msb (SMMUL, SMMLA, SMMLS) */
10772 tmp64 = gen_muls_i64_i32(tmp, tmp2);
10774 tmp = load_reg(s, rs);
10775 if (insn & (1 << 20)) {
10776 tmp64 = gen_addq_msw(tmp64, tmp);
10778 tmp64 = gen_subq_msw(tmp64, tmp);
10781 if (insn & (1 << 4)) {
10782 tcg_gen_addi_i64(tmp64, tmp64, 0x80000000u);
10784 tcg_gen_shri_i64(tmp64, tmp64, 32);
10785 tmp = tcg_temp_new_i32();
10786 tcg_gen_extrl_i64_i32(tmp, tmp64);
10787 tcg_temp_free_i64(tmp64);
10789 case 7: /* Unsigned sum of absolute differences. */
10790 gen_helper_usad8(tmp, tmp, tmp2);
10791 tcg_temp_free_i32(tmp2);
10793 tmp2 = load_reg(s, rs);
10794 tcg_gen_add_i32(tmp, tmp, tmp2);
10795 tcg_temp_free_i32(tmp2);
10799 store_reg(s, rd, tmp);
10801 case 6: case 7: /* 64-bit multiply, Divide. */
10802 op = ((insn >> 4) & 0xf) | ((insn >> 16) & 0x70);
10803 tmp = load_reg(s, rn);
10804 tmp2 = load_reg(s, rm);
10805 if ((op & 0x50) == 0x10) {
10807 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DIV)) {
10811 gen_helper_udiv(tmp, tmp, tmp2);
10813 gen_helper_sdiv(tmp, tmp, tmp2);
10814 tcg_temp_free_i32(tmp2);
10815 store_reg(s, rd, tmp);
10816 } else if ((op & 0xe) == 0xc) {
10817 /* Dual multiply accumulate long. */
10818 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10819 tcg_temp_free_i32(tmp);
10820 tcg_temp_free_i32(tmp2);
10824 gen_swap_half(tmp2);
10825 gen_smul_dual(tmp, tmp2);
10827 tcg_gen_sub_i32(tmp, tmp, tmp2);
10829 tcg_gen_add_i32(tmp, tmp, tmp2);
10831 tcg_temp_free_i32(tmp2);
10833 tmp64 = tcg_temp_new_i64();
10834 tcg_gen_ext_i32_i64(tmp64, tmp);
10835 tcg_temp_free_i32(tmp);
10836 gen_addq(s, tmp64, rs, rd);
10837 gen_storeq_reg(s, rs, rd, tmp64);
10838 tcg_temp_free_i64(tmp64);
10841 /* Unsigned 64-bit multiply */
10842 tmp64 = gen_mulu_i64_i32(tmp, tmp2);
10846 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10847 tcg_temp_free_i32(tmp2);
10848 tcg_temp_free_i32(tmp);
10851 gen_mulxy(tmp, tmp2, op & 2, op & 1);
10852 tcg_temp_free_i32(tmp2);
10853 tmp64 = tcg_temp_new_i64();
10854 tcg_gen_ext_i32_i64(tmp64, tmp);
10855 tcg_temp_free_i32(tmp);
10857 /* Signed 64-bit multiply */
10858 tmp64 = gen_muls_i64_i32(tmp, tmp2);
10863 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10864 tcg_temp_free_i64(tmp64);
10867 gen_addq_lo(s, tmp64, rs);
10868 gen_addq_lo(s, tmp64, rd);
10869 } else if (op & 0x40) {
10870 /* 64-bit accumulate. */
10871 gen_addq(s, tmp64, rs, rd);
10873 gen_storeq_reg(s, rs, rd, tmp64);
10874 tcg_temp_free_i64(tmp64);
10879 case 6: case 7: case 14: case 15:
10881 if (arm_dc_feature(s, ARM_FEATURE_M)) {
10882 /* We don't currently implement M profile FP support,
10883 * so this entire space should give a NOCP fault, with
10884 * the exception of the v8M VLLDM and VLSTM insns, which
10885 * must be NOPs in Secure state and UNDEF in Nonsecure state.
10887 if (arm_dc_feature(s, ARM_FEATURE_V8) &&
10888 (insn & 0xffa00f00) == 0xec200a00) {
10889 /* 0b1110_1100_0x1x_xxxx_xxxx_1010_xxxx_xxxx
10891 * We choose to UNDEF if the RAZ bits are non-zero.
10893 if (!s->v8m_secure || (insn & 0x0040f0ff)) {
10896 /* Just NOP since FP support is not implemented */
10899 /* All other insns: NOCP */
10900 gen_exception_insn(s, 4, EXCP_NOCP, syn_uncategorized(),
10901 default_exception_el(s));
10904 if ((insn & 0xfe000a00) == 0xfc000800
10905 && arm_dc_feature(s, ARM_FEATURE_V8)) {
10906 /* The Thumb2 and ARM encodings are identical. */
10907 if (disas_neon_insn_3same_ext(s, insn)) {
10910 } else if ((insn & 0xff000a00) == 0xfe000800
10911 && arm_dc_feature(s, ARM_FEATURE_V8)) {
10912 /* The Thumb2 and ARM encodings are identical. */
10913 if (disas_neon_insn_2reg_scalar_ext(s, insn)) {
10916 } else if (((insn >> 24) & 3) == 3) {
10917 /* Translate into the equivalent ARM encoding. */
10918 insn = (insn & 0xe2ffffff) | ((insn & (1 << 28)) >> 4) | (1 << 28);
10919 if (disas_neon_data_insn(s, insn)) {
10922 } else if (((insn >> 8) & 0xe) == 10) {
10923 if (disas_vfp_insn(s, insn)) {
10927 if (insn & (1 << 28))
10929 if (disas_coproc_insn(s, insn)) {
10934 case 8: case 9: case 10: case 11:
10935 if (insn & (1 << 15)) {
10936 /* Branches, misc control. */
10937 if (insn & 0x5000) {
10938 /* Unconditional branch. */
10939 /* signextend(hw1[10:0]) -> offset[:12]. */
10940 offset = ((int32_t)insn << 5) >> 9 & ~(int32_t)0xfff;
10941 /* hw1[10:0] -> offset[11:1]. */
10942 offset |= (insn & 0x7ff) << 1;
10943 /* (~hw2[13, 11] ^ offset[24]) -> offset[23,22]
10944 offset[24:22] already have the same value because of the
10945 sign extension above. */
10946 offset ^= ((~insn) & (1 << 13)) << 10;
10947 offset ^= ((~insn) & (1 << 11)) << 11;
10949 if (insn & (1 << 14)) {
10950 /* Branch and link. */
10951 tcg_gen_movi_i32(cpu_R[14], s->pc | 1);
10955 if (insn & (1 << 12)) {
10957 gen_jmp(s, offset);
10960 offset &= ~(uint32_t)2;
10961 /* thumb2 bx, no need to check */
10962 gen_bx_im(s, offset);
10964 } else if (((insn >> 23) & 7) == 7) {
10966 if (insn & (1 << 13))
10969 if (insn & (1 << 26)) {
10970 if (arm_dc_feature(s, ARM_FEATURE_M)) {
10973 if (!(insn & (1 << 20))) {
10974 /* Hypervisor call (v7) */
10975 int imm16 = extract32(insn, 16, 4) << 12
10976 | extract32(insn, 0, 12);
10983 /* Secure monitor call (v6+) */
10991 op = (insn >> 20) & 7;
10993 case 0: /* msr cpsr. */
10994 if (arm_dc_feature(s, ARM_FEATURE_M)) {
10995 tmp = load_reg(s, rn);
10996 /* the constant is the mask and SYSm fields */
10997 addr = tcg_const_i32(insn & 0xfff);
10998 gen_helper_v7m_msr(cpu_env, addr, tmp);
10999 tcg_temp_free_i32(addr);
11000 tcg_temp_free_i32(tmp);
11005 case 1: /* msr spsr. */
11006 if (arm_dc_feature(s, ARM_FEATURE_M)) {
11010 if (extract32(insn, 5, 1)) {
11012 int sysm = extract32(insn, 8, 4) |
11013 (extract32(insn, 4, 1) << 4);
11016 gen_msr_banked(s, r, sysm, rm);
11020 /* MSR (for PSRs) */
11021 tmp = load_reg(s, rn);
11023 msr_mask(s, (insn >> 8) & 0xf, op == 1),
11027 case 2: /* cps, nop-hint. */
11028 if (((insn >> 8) & 7) == 0) {
11029 gen_nop_hint(s, insn & 0xff);
11031 /* Implemented as NOP in user mode. */
11036 if (insn & (1 << 10)) {
11037 if (insn & (1 << 7))
11039 if (insn & (1 << 6))
11041 if (insn & (1 << 5))
11043 if (insn & (1 << 9))
11044 imm = CPSR_A | CPSR_I | CPSR_F;
11046 if (insn & (1 << 8)) {
11048 imm |= (insn & 0x1f);
11051 gen_set_psr_im(s, offset, 0, imm);
11054 case 3: /* Special control operations. */
11055 if (!arm_dc_feature(s, ARM_FEATURE_V7) &&
11056 !arm_dc_feature(s, ARM_FEATURE_M)) {
11059 op = (insn >> 4) & 0xf;
11061 case 2: /* clrex */
11066 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
11069 /* We need to break the TB after this insn
11070 * to execute self-modifying code correctly
11071 * and also to take any pending interrupts
11074 gen_goto_tb(s, 0, s->pc & ~1);
11081 /* Trivial implementation equivalent to bx.
11082 * This instruction doesn't exist at all for M-profile.
11084 if (arm_dc_feature(s, ARM_FEATURE_M)) {
11087 tmp = load_reg(s, rn);
11090 case 5: /* Exception return. */
11094 if (rn != 14 || rd != 15) {
11097 tmp = load_reg(s, rn);
11098 tcg_gen_subi_i32(tmp, tmp, insn & 0xff);
11099 gen_exception_return(s, tmp);
11102 if (extract32(insn, 5, 1) &&
11103 !arm_dc_feature(s, ARM_FEATURE_M)) {
11105 int sysm = extract32(insn, 16, 4) |
11106 (extract32(insn, 4, 1) << 4);
11108 gen_mrs_banked(s, 0, sysm, rd);
11112 if (extract32(insn, 16, 4) != 0xf) {
11115 if (!arm_dc_feature(s, ARM_FEATURE_M) &&
11116 extract32(insn, 0, 8) != 0) {
11121 tmp = tcg_temp_new_i32();
11122 if (arm_dc_feature(s, ARM_FEATURE_M)) {
11123 addr = tcg_const_i32(insn & 0xff);
11124 gen_helper_v7m_mrs(tmp, cpu_env, addr);
11125 tcg_temp_free_i32(addr);
11127 gen_helper_cpsr_read(tmp, cpu_env);
11129 store_reg(s, rd, tmp);
11132 if (extract32(insn, 5, 1) &&
11133 !arm_dc_feature(s, ARM_FEATURE_M)) {
11135 int sysm = extract32(insn, 16, 4) |
11136 (extract32(insn, 4, 1) << 4);
11138 gen_mrs_banked(s, 1, sysm, rd);
11143 /* Not accessible in user mode. */
11144 if (IS_USER(s) || arm_dc_feature(s, ARM_FEATURE_M)) {
11148 if (extract32(insn, 16, 4) != 0xf ||
11149 extract32(insn, 0, 8) != 0) {
11153 tmp = load_cpu_field(spsr);
11154 store_reg(s, rd, tmp);
11159 /* Conditional branch. */
11160 op = (insn >> 22) & 0xf;
11161 /* Generate a conditional jump to next instruction. */
11162 s->condlabel = gen_new_label();
11163 arm_gen_test_cc(op ^ 1, s->condlabel);
11166 /* offset[11:1] = insn[10:0] */
11167 offset = (insn & 0x7ff) << 1;
11168 /* offset[17:12] = insn[21:16]. */
11169 offset |= (insn & 0x003f0000) >> 4;
11170 /* offset[31:20] = insn[26]. */
11171 offset |= ((int32_t)((insn << 5) & 0x80000000)) >> 11;
11172 /* offset[18] = insn[13]. */
11173 offset |= (insn & (1 << 13)) << 5;
11174 /* offset[19] = insn[11]. */
11175 offset |= (insn & (1 << 11)) << 8;
11177 /* jump to the offset */
11178 gen_jmp(s, s->pc + offset);
11181 /* Data processing immediate. */
11182 if (insn & (1 << 25)) {
11183 if (insn & (1 << 24)) {
11184 if (insn & (1 << 20))
11186 /* Bitfield/Saturate. */
11187 op = (insn >> 21) & 7;
11189 shift = ((insn >> 6) & 3) | ((insn >> 10) & 0x1c);
11191 tmp = tcg_temp_new_i32();
11192 tcg_gen_movi_i32(tmp, 0);
11194 tmp = load_reg(s, rn);
11197 case 2: /* Signed bitfield extract. */
11199 if (shift + imm > 32)
11202 tcg_gen_sextract_i32(tmp, tmp, shift, imm);
11205 case 6: /* Unsigned bitfield extract. */
11207 if (shift + imm > 32)
11210 tcg_gen_extract_i32(tmp, tmp, shift, imm);
11213 case 3: /* Bitfield insert/clear. */
11216 imm = imm + 1 - shift;
11218 tmp2 = load_reg(s, rd);
11219 tcg_gen_deposit_i32(tmp, tmp2, tmp, shift, imm);
11220 tcg_temp_free_i32(tmp2);
11225 default: /* Saturate. */
11228 tcg_gen_sari_i32(tmp, tmp, shift);
11230 tcg_gen_shli_i32(tmp, tmp, shift);
11232 tmp2 = tcg_const_i32(imm);
11235 if ((op & 1) && shift == 0) {
11236 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
11237 tcg_temp_free_i32(tmp);
11238 tcg_temp_free_i32(tmp2);
11241 gen_helper_usat16(tmp, cpu_env, tmp, tmp2);
11243 gen_helper_usat(tmp, cpu_env, tmp, tmp2);
11247 if ((op & 1) && shift == 0) {
11248 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
11249 tcg_temp_free_i32(tmp);
11250 tcg_temp_free_i32(tmp2);
11253 gen_helper_ssat16(tmp, cpu_env, tmp, tmp2);
11255 gen_helper_ssat(tmp, cpu_env, tmp, tmp2);
11258 tcg_temp_free_i32(tmp2);
11261 store_reg(s, rd, tmp);
11263 imm = ((insn & 0x04000000) >> 15)
11264 | ((insn & 0x7000) >> 4) | (insn & 0xff);
11265 if (insn & (1 << 22)) {
11266 /* 16-bit immediate. */
11267 imm |= (insn >> 4) & 0xf000;
11268 if (insn & (1 << 23)) {
11270 tmp = load_reg(s, rd);
11271 tcg_gen_ext16u_i32(tmp, tmp);
11272 tcg_gen_ori_i32(tmp, tmp, imm << 16);
11275 tmp = tcg_temp_new_i32();
11276 tcg_gen_movi_i32(tmp, imm);
11279 /* Add/sub 12-bit immediate. */
11281 offset = s->pc & ~(uint32_t)3;
11282 if (insn & (1 << 23))
11286 tmp = tcg_temp_new_i32();
11287 tcg_gen_movi_i32(tmp, offset);
11289 tmp = load_reg(s, rn);
11290 if (insn & (1 << 23))
11291 tcg_gen_subi_i32(tmp, tmp, imm);
11293 tcg_gen_addi_i32(tmp, tmp, imm);
11296 store_reg(s, rd, tmp);
11299 int shifter_out = 0;
11300 /* modified 12-bit immediate. */
11301 shift = ((insn & 0x04000000) >> 23) | ((insn & 0x7000) >> 12);
11302 imm = (insn & 0xff);
11305 /* Nothing to do. */
11307 case 1: /* 00XY00XY */
11310 case 2: /* XY00XY00 */
11314 case 3: /* XYXYXYXY */
11318 default: /* Rotated constant. */
11319 shift = (shift << 1) | (imm >> 7);
11321 imm = imm << (32 - shift);
11325 tmp2 = tcg_temp_new_i32();
11326 tcg_gen_movi_i32(tmp2, imm);
11327 rn = (insn >> 16) & 0xf;
11329 tmp = tcg_temp_new_i32();
11330 tcg_gen_movi_i32(tmp, 0);
11332 tmp = load_reg(s, rn);
11334 op = (insn >> 21) & 0xf;
11335 if (gen_thumb2_data_op(s, op, (insn & (1 << 20)) != 0,
11336 shifter_out, tmp, tmp2))
11338 tcg_temp_free_i32(tmp2);
11339 rd = (insn >> 8) & 0xf;
11341 store_reg(s, rd, tmp);
11343 tcg_temp_free_i32(tmp);
11348 case 12: /* Load/store single data item. */
11355 if ((insn & 0x01100000) == 0x01000000) {
11356 if (disas_neon_ls_insn(s, insn)) {
11361 op = ((insn >> 21) & 3) | ((insn >> 22) & 4);
11363 if (!(insn & (1 << 20))) {
11367 /* Byte or halfword load space with dest == r15 : memory hints.
11368 * Catch them early so we don't emit pointless addressing code.
11369 * This space is a mix of:
11370 * PLD/PLDW/PLI, which we implement as NOPs (note that unlike
11371 * the ARM encodings, PLDW space doesn't UNDEF for non-v7MP
11373 * unallocated hints, which must be treated as NOPs
11374 * UNPREDICTABLE space, which we NOP or UNDEF depending on
11375 * which is easiest for the decoding logic
11376 * Some space which must UNDEF
11378 int op1 = (insn >> 23) & 3;
11379 int op2 = (insn >> 6) & 0x3f;
11384 /* UNPREDICTABLE, unallocated hint or
11385 * PLD/PLDW/PLI (literal)
11390 return; /* PLD/PLDW/PLI or unallocated hint */
11392 if ((op2 == 0) || ((op2 & 0x3c) == 0x30)) {
11393 return; /* PLD/PLDW/PLI or unallocated hint */
11395 /* UNDEF space, or an UNPREDICTABLE */
11399 memidx = get_mem_index(s);
11401 addr = tcg_temp_new_i32();
11403 /* s->pc has already been incremented by 4. */
11404 imm = s->pc & 0xfffffffc;
11405 if (insn & (1 << 23))
11406 imm += insn & 0xfff;
11408 imm -= insn & 0xfff;
11409 tcg_gen_movi_i32(addr, imm);
11411 addr = load_reg(s, rn);
11412 if (insn & (1 << 23)) {
11413 /* Positive offset. */
11414 imm = insn & 0xfff;
11415 tcg_gen_addi_i32(addr, addr, imm);
11418 switch ((insn >> 8) & 0xf) {
11419 case 0x0: /* Shifted Register. */
11420 shift = (insn >> 4) & 0xf;
11422 tcg_temp_free_i32(addr);
11425 tmp = load_reg(s, rm);
11427 tcg_gen_shli_i32(tmp, tmp, shift);
11428 tcg_gen_add_i32(addr, addr, tmp);
11429 tcg_temp_free_i32(tmp);
11431 case 0xc: /* Negative offset. */
11432 tcg_gen_addi_i32(addr, addr, -imm);
11434 case 0xe: /* User privilege. */
11435 tcg_gen_addi_i32(addr, addr, imm);
11436 memidx = get_a32_user_mem_index(s);
11438 case 0x9: /* Post-decrement. */
11440 /* Fall through. */
11441 case 0xb: /* Post-increment. */
11445 case 0xd: /* Pre-decrement. */
11447 /* Fall through. */
11448 case 0xf: /* Pre-increment. */
11449 tcg_gen_addi_i32(addr, addr, imm);
11453 tcg_temp_free_i32(addr);
11459 issinfo = writeback ? ISSInvalid : rs;
11461 if (insn & (1 << 20)) {
11463 tmp = tcg_temp_new_i32();
11466 gen_aa32_ld8u_iss(s, tmp, addr, memidx, issinfo);
11469 gen_aa32_ld8s_iss(s, tmp, addr, memidx, issinfo);
11472 gen_aa32_ld16u_iss(s, tmp, addr, memidx, issinfo);
11475 gen_aa32_ld16s_iss(s, tmp, addr, memidx, issinfo);
11478 gen_aa32_ld32u_iss(s, tmp, addr, memidx, issinfo);
11481 tcg_temp_free_i32(tmp);
11482 tcg_temp_free_i32(addr);
11486 gen_bx_excret(s, tmp);
11488 store_reg(s, rs, tmp);
11492 tmp = load_reg(s, rs);
11495 gen_aa32_st8_iss(s, tmp, addr, memidx, issinfo);
11498 gen_aa32_st16_iss(s, tmp, addr, memidx, issinfo);
11501 gen_aa32_st32_iss(s, tmp, addr, memidx, issinfo);
11504 tcg_temp_free_i32(tmp);
11505 tcg_temp_free_i32(addr);
11508 tcg_temp_free_i32(tmp);
11511 tcg_gen_addi_i32(addr, addr, imm);
11513 store_reg(s, rn, addr);
11515 tcg_temp_free_i32(addr);
11524 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(),
11525 default_exception_el(s));
11528 static void disas_thumb_insn(DisasContext *s, uint32_t insn)
11530 uint32_t val, op, rm, rn, rd, shift, cond;
11537 switch (insn >> 12) {
11541 op = (insn >> 11) & 3;
11544 rn = (insn >> 3) & 7;
11545 tmp = load_reg(s, rn);
11546 if (insn & (1 << 10)) {
11548 tmp2 = tcg_temp_new_i32();
11549 tcg_gen_movi_i32(tmp2, (insn >> 6) & 7);
11552 rm = (insn >> 6) & 7;
11553 tmp2 = load_reg(s, rm);
11555 if (insn & (1 << 9)) {
11556 if (s->condexec_mask)
11557 tcg_gen_sub_i32(tmp, tmp, tmp2);
11559 gen_sub_CC(tmp, tmp, tmp2);
11561 if (s->condexec_mask)
11562 tcg_gen_add_i32(tmp, tmp, tmp2);
11564 gen_add_CC(tmp, tmp, tmp2);
11566 tcg_temp_free_i32(tmp2);
11567 store_reg(s, rd, tmp);
11569 /* shift immediate */
11570 rm = (insn >> 3) & 7;
11571 shift = (insn >> 6) & 0x1f;
11572 tmp = load_reg(s, rm);
11573 gen_arm_shift_im(tmp, op, shift, s->condexec_mask == 0);
11574 if (!s->condexec_mask)
11576 store_reg(s, rd, tmp);
11580 /* arithmetic large immediate */
11581 op = (insn >> 11) & 3;
11582 rd = (insn >> 8) & 0x7;
11583 if (op == 0) { /* mov */
11584 tmp = tcg_temp_new_i32();
11585 tcg_gen_movi_i32(tmp, insn & 0xff);
11586 if (!s->condexec_mask)
11588 store_reg(s, rd, tmp);
11590 tmp = load_reg(s, rd);
11591 tmp2 = tcg_temp_new_i32();
11592 tcg_gen_movi_i32(tmp2, insn & 0xff);
11595 gen_sub_CC(tmp, tmp, tmp2);
11596 tcg_temp_free_i32(tmp);
11597 tcg_temp_free_i32(tmp2);
11600 if (s->condexec_mask)
11601 tcg_gen_add_i32(tmp, tmp, tmp2);
11603 gen_add_CC(tmp, tmp, tmp2);
11604 tcg_temp_free_i32(tmp2);
11605 store_reg(s, rd, tmp);
11608 if (s->condexec_mask)
11609 tcg_gen_sub_i32(tmp, tmp, tmp2);
11611 gen_sub_CC(tmp, tmp, tmp2);
11612 tcg_temp_free_i32(tmp2);
11613 store_reg(s, rd, tmp);
11619 if (insn & (1 << 11)) {
11620 rd = (insn >> 8) & 7;
11621 /* load pc-relative. Bit 1 of PC is ignored. */
11622 val = s->pc + 2 + ((insn & 0xff) * 4);
11623 val &= ~(uint32_t)2;
11624 addr = tcg_temp_new_i32();
11625 tcg_gen_movi_i32(addr, val);
11626 tmp = tcg_temp_new_i32();
11627 gen_aa32_ld32u_iss(s, tmp, addr, get_mem_index(s),
11629 tcg_temp_free_i32(addr);
11630 store_reg(s, rd, tmp);
11633 if (insn & (1 << 10)) {
11634 /* 0b0100_01xx_xxxx_xxxx
11635 * - data processing extended, branch and exchange
11637 rd = (insn & 7) | ((insn >> 4) & 8);
11638 rm = (insn >> 3) & 0xf;
11639 op = (insn >> 8) & 3;
11642 tmp = load_reg(s, rd);
11643 tmp2 = load_reg(s, rm);
11644 tcg_gen_add_i32(tmp, tmp, tmp2);
11645 tcg_temp_free_i32(tmp2);
11646 store_reg(s, rd, tmp);
11649 tmp = load_reg(s, rd);
11650 tmp2 = load_reg(s, rm);
11651 gen_sub_CC(tmp, tmp, tmp2);
11652 tcg_temp_free_i32(tmp2);
11653 tcg_temp_free_i32(tmp);
11655 case 2: /* mov/cpy */
11656 tmp = load_reg(s, rm);
11657 store_reg(s, rd, tmp);
11661 /* 0b0100_0111_xxxx_xxxx
11662 * - branch [and link] exchange thumb register
11664 bool link = insn & (1 << 7);
11673 /* BXNS/BLXNS: only exists for v8M with the
11674 * security extensions, and always UNDEF if NonSecure.
11675 * We don't implement these in the user-only mode
11676 * either (in theory you can use them from Secure User
11677 * mode but they are too tied in to system emulation.)
11679 if (!s->v8m_secure || IS_USER_ONLY) {
11690 tmp = load_reg(s, rm);
11692 val = (uint32_t)s->pc | 1;
11693 tmp2 = tcg_temp_new_i32();
11694 tcg_gen_movi_i32(tmp2, val);
11695 store_reg(s, 14, tmp2);
11698 /* Only BX works as exception-return, not BLX */
11699 gen_bx_excret(s, tmp);
11707 /* data processing register */
11709 rm = (insn >> 3) & 7;
11710 op = (insn >> 6) & 0xf;
11711 if (op == 2 || op == 3 || op == 4 || op == 7) {
11712 /* the shift/rotate ops want the operands backwards */
11721 if (op == 9) { /* neg */
11722 tmp = tcg_temp_new_i32();
11723 tcg_gen_movi_i32(tmp, 0);
11724 } else if (op != 0xf) { /* mvn doesn't read its first operand */
11725 tmp = load_reg(s, rd);
11730 tmp2 = load_reg(s, rm);
11732 case 0x0: /* and */
11733 tcg_gen_and_i32(tmp, tmp, tmp2);
11734 if (!s->condexec_mask)
11737 case 0x1: /* eor */
11738 tcg_gen_xor_i32(tmp, tmp, tmp2);
11739 if (!s->condexec_mask)
11742 case 0x2: /* lsl */
11743 if (s->condexec_mask) {
11744 gen_shl(tmp2, tmp2, tmp);
11746 gen_helper_shl_cc(tmp2, cpu_env, tmp2, tmp);
11747 gen_logic_CC(tmp2);
11750 case 0x3: /* lsr */
11751 if (s->condexec_mask) {
11752 gen_shr(tmp2, tmp2, tmp);
11754 gen_helper_shr_cc(tmp2, cpu_env, tmp2, tmp);
11755 gen_logic_CC(tmp2);
11758 case 0x4: /* asr */
11759 if (s->condexec_mask) {
11760 gen_sar(tmp2, tmp2, tmp);
11762 gen_helper_sar_cc(tmp2, cpu_env, tmp2, tmp);
11763 gen_logic_CC(tmp2);
11766 case 0x5: /* adc */
11767 if (s->condexec_mask) {
11768 gen_adc(tmp, tmp2);
11770 gen_adc_CC(tmp, tmp, tmp2);
11773 case 0x6: /* sbc */
11774 if (s->condexec_mask) {
11775 gen_sub_carry(tmp, tmp, tmp2);
11777 gen_sbc_CC(tmp, tmp, tmp2);
11780 case 0x7: /* ror */
11781 if (s->condexec_mask) {
11782 tcg_gen_andi_i32(tmp, tmp, 0x1f);
11783 tcg_gen_rotr_i32(tmp2, tmp2, tmp);
11785 gen_helper_ror_cc(tmp2, cpu_env, tmp2, tmp);
11786 gen_logic_CC(tmp2);
11789 case 0x8: /* tst */
11790 tcg_gen_and_i32(tmp, tmp, tmp2);
11794 case 0x9: /* neg */
11795 if (s->condexec_mask)
11796 tcg_gen_neg_i32(tmp, tmp2);
11798 gen_sub_CC(tmp, tmp, tmp2);
11800 case 0xa: /* cmp */
11801 gen_sub_CC(tmp, tmp, tmp2);
11804 case 0xb: /* cmn */
11805 gen_add_CC(tmp, tmp, tmp2);
11808 case 0xc: /* orr */
11809 tcg_gen_or_i32(tmp, tmp, tmp2);
11810 if (!s->condexec_mask)
11813 case 0xd: /* mul */
11814 tcg_gen_mul_i32(tmp, tmp, tmp2);
11815 if (!s->condexec_mask)
11818 case 0xe: /* bic */
11819 tcg_gen_andc_i32(tmp, tmp, tmp2);
11820 if (!s->condexec_mask)
11823 case 0xf: /* mvn */
11824 tcg_gen_not_i32(tmp2, tmp2);
11825 if (!s->condexec_mask)
11826 gen_logic_CC(tmp2);
11833 store_reg(s, rm, tmp2);
11835 tcg_temp_free_i32(tmp);
11837 store_reg(s, rd, tmp);
11838 tcg_temp_free_i32(tmp2);
11841 tcg_temp_free_i32(tmp);
11842 tcg_temp_free_i32(tmp2);
11847 /* load/store register offset. */
11849 rn = (insn >> 3) & 7;
11850 rm = (insn >> 6) & 7;
11851 op = (insn >> 9) & 7;
11852 addr = load_reg(s, rn);
11853 tmp = load_reg(s, rm);
11854 tcg_gen_add_i32(addr, addr, tmp);
11855 tcg_temp_free_i32(tmp);
11857 if (op < 3) { /* store */
11858 tmp = load_reg(s, rd);
11860 tmp = tcg_temp_new_i32();
11865 gen_aa32_st32_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11868 gen_aa32_st16_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11871 gen_aa32_st8_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11873 case 3: /* ldrsb */
11874 gen_aa32_ld8s_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11877 gen_aa32_ld32u_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11880 gen_aa32_ld16u_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11883 gen_aa32_ld8u_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11885 case 7: /* ldrsh */
11886 gen_aa32_ld16s_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11889 if (op >= 3) { /* load */
11890 store_reg(s, rd, tmp);
11892 tcg_temp_free_i32(tmp);
11894 tcg_temp_free_i32(addr);
11898 /* load/store word immediate offset */
11900 rn = (insn >> 3) & 7;
11901 addr = load_reg(s, rn);
11902 val = (insn >> 4) & 0x7c;
11903 tcg_gen_addi_i32(addr, addr, val);
11905 if (insn & (1 << 11)) {
11907 tmp = tcg_temp_new_i32();
11908 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
11909 store_reg(s, rd, tmp);
11912 tmp = load_reg(s, rd);
11913 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
11914 tcg_temp_free_i32(tmp);
11916 tcg_temp_free_i32(addr);
11920 /* load/store byte immediate offset */
11922 rn = (insn >> 3) & 7;
11923 addr = load_reg(s, rn);
11924 val = (insn >> 6) & 0x1f;
11925 tcg_gen_addi_i32(addr, addr, val);
11927 if (insn & (1 << 11)) {
11929 tmp = tcg_temp_new_i32();
11930 gen_aa32_ld8u_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11931 store_reg(s, rd, tmp);
11934 tmp = load_reg(s, rd);
11935 gen_aa32_st8_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11936 tcg_temp_free_i32(tmp);
11938 tcg_temp_free_i32(addr);
11942 /* load/store halfword immediate offset */
11944 rn = (insn >> 3) & 7;
11945 addr = load_reg(s, rn);
11946 val = (insn >> 5) & 0x3e;
11947 tcg_gen_addi_i32(addr, addr, val);
11949 if (insn & (1 << 11)) {
11951 tmp = tcg_temp_new_i32();
11952 gen_aa32_ld16u_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11953 store_reg(s, rd, tmp);
11956 tmp = load_reg(s, rd);
11957 gen_aa32_st16_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11958 tcg_temp_free_i32(tmp);
11960 tcg_temp_free_i32(addr);
11964 /* load/store from stack */
11965 rd = (insn >> 8) & 7;
11966 addr = load_reg(s, 13);
11967 val = (insn & 0xff) * 4;
11968 tcg_gen_addi_i32(addr, addr, val);
11970 if (insn & (1 << 11)) {
11972 tmp = tcg_temp_new_i32();
11973 gen_aa32_ld32u_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11974 store_reg(s, rd, tmp);
11977 tmp = load_reg(s, rd);
11978 gen_aa32_st32_iss(s, tmp, addr, get_mem_index(s), rd | ISSIs16Bit);
11979 tcg_temp_free_i32(tmp);
11981 tcg_temp_free_i32(addr);
11985 /* add to high reg */
11986 rd = (insn >> 8) & 7;
11987 if (insn & (1 << 11)) {
11989 tmp = load_reg(s, 13);
11991 /* PC. bit 1 is ignored. */
11992 tmp = tcg_temp_new_i32();
11993 tcg_gen_movi_i32(tmp, (s->pc + 2) & ~(uint32_t)2);
11995 val = (insn & 0xff) * 4;
11996 tcg_gen_addi_i32(tmp, tmp, val);
11997 store_reg(s, rd, tmp);
12002 op = (insn >> 8) & 0xf;
12005 /* adjust stack pointer */
12006 tmp = load_reg(s, 13);
12007 val = (insn & 0x7f) * 4;
12008 if (insn & (1 << 7))
12009 val = -(int32_t)val;
12010 tcg_gen_addi_i32(tmp, tmp, val);
12011 store_reg(s, 13, tmp);
12014 case 2: /* sign/zero extend. */
12017 rm = (insn >> 3) & 7;
12018 tmp = load_reg(s, rm);
12019 switch ((insn >> 6) & 3) {
12020 case 0: gen_sxth(tmp); break;
12021 case 1: gen_sxtb(tmp); break;
12022 case 2: gen_uxth(tmp); break;
12023 case 3: gen_uxtb(tmp); break;
12025 store_reg(s, rd, tmp);
12027 case 4: case 5: case 0xc: case 0xd:
12029 addr = load_reg(s, 13);
12030 if (insn & (1 << 8))
12034 for (i = 0; i < 8; i++) {
12035 if (insn & (1 << i))
12038 if ((insn & (1 << 11)) == 0) {
12039 tcg_gen_addi_i32(addr, addr, -offset);
12041 for (i = 0; i < 8; i++) {
12042 if (insn & (1 << i)) {
12043 if (insn & (1 << 11)) {
12045 tmp = tcg_temp_new_i32();
12046 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
12047 store_reg(s, i, tmp);
12050 tmp = load_reg(s, i);
12051 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
12052 tcg_temp_free_i32(tmp);
12054 /* advance to the next address. */
12055 tcg_gen_addi_i32(addr, addr, 4);
12059 if (insn & (1 << 8)) {
12060 if (insn & (1 << 11)) {
12062 tmp = tcg_temp_new_i32();
12063 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
12064 /* don't set the pc until the rest of the instruction
12068 tmp = load_reg(s, 14);
12069 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
12070 tcg_temp_free_i32(tmp);
12072 tcg_gen_addi_i32(addr, addr, 4);
12074 if ((insn & (1 << 11)) == 0) {
12075 tcg_gen_addi_i32(addr, addr, -offset);
12077 /* write back the new stack pointer */
12078 store_reg(s, 13, addr);
12079 /* set the new PC value */
12080 if ((insn & 0x0900) == 0x0900) {
12081 store_reg_from_load(s, 15, tmp);
12085 case 1: case 3: case 9: case 11: /* czb */
12087 tmp = load_reg(s, rm);
12088 s->condlabel = gen_new_label();
12090 if (insn & (1 << 11))
12091 tcg_gen_brcondi_i32(TCG_COND_EQ, tmp, 0, s->condlabel);
12093 tcg_gen_brcondi_i32(TCG_COND_NE, tmp, 0, s->condlabel);
12094 tcg_temp_free_i32(tmp);
12095 offset = ((insn & 0xf8) >> 2) | (insn & 0x200) >> 3;
12096 val = (uint32_t)s->pc + 2;
12101 case 15: /* IT, nop-hint. */
12102 if ((insn & 0xf) == 0) {
12103 gen_nop_hint(s, (insn >> 4) & 0xf);
12107 s->condexec_cond = (insn >> 4) & 0xe;
12108 s->condexec_mask = insn & 0x1f;
12109 /* No actual code generated for this insn, just setup state. */
12112 case 0xe: /* bkpt */
12114 int imm8 = extract32(insn, 0, 8);
12116 gen_exception_bkpt_insn(s, 2, syn_aa32_bkpt(imm8, true));
12120 case 0xa: /* rev, and hlt */
12122 int op1 = extract32(insn, 6, 2);
12126 int imm6 = extract32(insn, 0, 6);
12132 /* Otherwise this is rev */
12134 rn = (insn >> 3) & 0x7;
12136 tmp = load_reg(s, rn);
12138 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
12139 case 1: gen_rev16(tmp); break;
12140 case 3: gen_revsh(tmp); break;
12142 g_assert_not_reached();
12144 store_reg(s, rd, tmp);
12149 switch ((insn >> 5) & 7) {
12153 if (((insn >> 3) & 1) != !!(s->be_data == MO_BE)) {
12154 gen_helper_setend(cpu_env);
12155 s->base.is_jmp = DISAS_UPDATE;
12164 if (arm_dc_feature(s, ARM_FEATURE_M)) {
12165 tmp = tcg_const_i32((insn & (1 << 4)) != 0);
12168 addr = tcg_const_i32(19);
12169 gen_helper_v7m_msr(cpu_env, addr, tmp);
12170 tcg_temp_free_i32(addr);
12174 addr = tcg_const_i32(16);
12175 gen_helper_v7m_msr(cpu_env, addr, tmp);
12176 tcg_temp_free_i32(addr);
12178 tcg_temp_free_i32(tmp);
12181 if (insn & (1 << 4)) {
12182 shift = CPSR_A | CPSR_I | CPSR_F;
12186 gen_set_psr_im(s, ((insn & 7) << 6), 0, shift);
12201 /* load/store multiple */
12202 TCGv_i32 loaded_var = NULL;
12203 rn = (insn >> 8) & 0x7;
12204 addr = load_reg(s, rn);
12205 for (i = 0; i < 8; i++) {
12206 if (insn & (1 << i)) {
12207 if (insn & (1 << 11)) {
12209 tmp = tcg_temp_new_i32();
12210 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
12214 store_reg(s, i, tmp);
12218 tmp = load_reg(s, i);
12219 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
12220 tcg_temp_free_i32(tmp);
12222 /* advance to the next address */
12223 tcg_gen_addi_i32(addr, addr, 4);
12226 if ((insn & (1 << rn)) == 0) {
12227 /* base reg not in list: base register writeback */
12228 store_reg(s, rn, addr);
12230 /* base reg in list: if load, complete it now */
12231 if (insn & (1 << 11)) {
12232 store_reg(s, rn, loaded_var);
12234 tcg_temp_free_i32(addr);
12239 /* conditional branch or swi */
12240 cond = (insn >> 8) & 0xf;
12246 gen_set_pc_im(s, s->pc);
12247 s->svc_imm = extract32(insn, 0, 8);
12248 s->base.is_jmp = DISAS_SWI;
12251 /* generate a conditional jump to next instruction */
12252 s->condlabel = gen_new_label();
12253 arm_gen_test_cc(cond ^ 1, s->condlabel);
12256 /* jump to the offset */
12257 val = (uint32_t)s->pc + 2;
12258 offset = ((int32_t)insn << 24) >> 24;
12259 val += offset << 1;
12264 if (insn & (1 << 11)) {
12265 /* thumb_insn_is_16bit() ensures we can't get here for
12266 * a Thumb2 CPU, so this must be a thumb1 split BL/BLX:
12267 * 0b1110_1xxx_xxxx_xxxx : BLX suffix (or UNDEF)
12269 assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
12271 offset = ((insn & 0x7ff) << 1);
12272 tmp = load_reg(s, 14);
12273 tcg_gen_addi_i32(tmp, tmp, offset);
12274 tcg_gen_andi_i32(tmp, tmp, 0xfffffffc);
12276 tmp2 = tcg_temp_new_i32();
12277 tcg_gen_movi_i32(tmp2, s->pc | 1);
12278 store_reg(s, 14, tmp2);
12282 /* unconditional branch */
12283 val = (uint32_t)s->pc;
12284 offset = ((int32_t)insn << 21) >> 21;
12285 val += (offset << 1) + 2;
12290 /* thumb_insn_is_16bit() ensures we can't get here for
12291 * a Thumb2 CPU, so this must be a thumb1 split BL/BLX.
12293 assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
12295 if (insn & (1 << 11)) {
12296 /* 0b1111_1xxx_xxxx_xxxx : BL suffix */
12297 offset = ((insn & 0x7ff) << 1) | 1;
12298 tmp = load_reg(s, 14);
12299 tcg_gen_addi_i32(tmp, tmp, offset);
12301 tmp2 = tcg_temp_new_i32();
12302 tcg_gen_movi_i32(tmp2, s->pc | 1);
12303 store_reg(s, 14, tmp2);
12306 /* 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix */
12307 uint32_t uoffset = ((int32_t)insn << 21) >> 9;
12309 tcg_gen_movi_i32(cpu_R[14], s->pc + 2 + uoffset);
12316 gen_exception_insn(s, 2, EXCP_UDEF, syn_uncategorized(),
12317 default_exception_el(s));
12320 static bool insn_crosses_page(CPUARMState *env, DisasContext *s)
12322 /* Return true if the insn at dc->pc might cross a page boundary.
12323 * (False positives are OK, false negatives are not.)
12324 * We know this is a Thumb insn, and our caller ensures we are
12325 * only called if dc->pc is less than 4 bytes from the page
12326 * boundary, so we cross the page if the first 16 bits indicate
12327 * that this is a 32 bit insn.
12329 uint16_t insn = arm_lduw_code(env, s->pc, s->sctlr_b);
12331 return !thumb_insn_is_16bit(s, insn);
12334 static void arm_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cs)
12336 DisasContext *dc = container_of(dcbase, DisasContext, base);
12337 CPUARMState *env = cs->env_ptr;
12338 ARMCPU *cpu = arm_env_get_cpu(env);
12340 dc->pc = dc->base.pc_first;
12344 /* If we are coming from secure EL0 in a system with a 32-bit EL3, then
12345 * there is no secure EL1, so we route exceptions to EL3.
12347 dc->secure_routed_to_el3 = arm_feature(env, ARM_FEATURE_EL3) &&
12348 !arm_el_is_aa64(env, 3);
12349 dc->thumb = ARM_TBFLAG_THUMB(dc->base.tb->flags);
12350 dc->sctlr_b = ARM_TBFLAG_SCTLR_B(dc->base.tb->flags);
12351 dc->be_data = ARM_TBFLAG_BE_DATA(dc->base.tb->flags) ? MO_BE : MO_LE;
12352 dc->condexec_mask = (ARM_TBFLAG_CONDEXEC(dc->base.tb->flags) & 0xf) << 1;
12353 dc->condexec_cond = ARM_TBFLAG_CONDEXEC(dc->base.tb->flags) >> 4;
12354 dc->mmu_idx = core_to_arm_mmu_idx(env, ARM_TBFLAG_MMUIDX(dc->base.tb->flags));
12355 dc->current_el = arm_mmu_idx_to_el(dc->mmu_idx);
12356 #if !defined(CONFIG_USER_ONLY)
12357 dc->user = (dc->current_el == 0);
12359 dc->ns = ARM_TBFLAG_NS(dc->base.tb->flags);
12360 dc->fp_excp_el = ARM_TBFLAG_FPEXC_EL(dc->base.tb->flags);
12361 dc->vfp_enabled = ARM_TBFLAG_VFPEN(dc->base.tb->flags);
12362 dc->vec_len = ARM_TBFLAG_VECLEN(dc->base.tb->flags);
12363 dc->vec_stride = ARM_TBFLAG_VECSTRIDE(dc->base.tb->flags);
12364 dc->c15_cpar = ARM_TBFLAG_XSCALE_CPAR(dc->base.tb->flags);
12365 dc->v7m_handler_mode = ARM_TBFLAG_HANDLER(dc->base.tb->flags);
12366 dc->v8m_secure = arm_feature(env, ARM_FEATURE_M_SECURITY) &&
12367 regime_is_secure(env, dc->mmu_idx);
12368 dc->cp_regs = cpu->cp_regs;
12369 dc->features = env->features;
12371 /* Single step state. The code-generation logic here is:
12373 * generate code with no special handling for single-stepping (except
12374 * that anything that can make us go to SS_ACTIVE == 1 must end the TB;
12375 * this happens anyway because those changes are all system register or
12377 * SS_ACTIVE == 1, PSTATE.SS == 1: (active-not-pending)
12378 * emit code for one insn
12379 * emit code to clear PSTATE.SS
12380 * emit code to generate software step exception for completed step
12381 * end TB (as usual for having generated an exception)
12382 * SS_ACTIVE == 1, PSTATE.SS == 0: (active-pending)
12383 * emit code to generate a software step exception
12386 dc->ss_active = ARM_TBFLAG_SS_ACTIVE(dc->base.tb->flags);
12387 dc->pstate_ss = ARM_TBFLAG_PSTATE_SS(dc->base.tb->flags);
12388 dc->is_ldex = false;
12389 dc->ss_same_el = false; /* Can't be true since EL_d must be AArch64 */
12391 dc->page_start = dc->base.pc_first & TARGET_PAGE_MASK;
12393 /* If architectural single step active, limit to 1. */
12394 if (is_singlestepping(dc)) {
12395 dc->base.max_insns = 1;
12398 /* ARM is a fixed-length ISA. Bound the number of insns to execute
12399 to those left on the page. */
12401 int bound = -(dc->base.pc_first | TARGET_PAGE_MASK) / 4;
12402 dc->base.max_insns = MIN(dc->base.max_insns, bound);
12405 cpu_F0s = tcg_temp_new_i32();
12406 cpu_F1s = tcg_temp_new_i32();
12407 cpu_F0d = tcg_temp_new_i64();
12408 cpu_F1d = tcg_temp_new_i64();
12411 /* FIXME: cpu_M0 can probably be the same as cpu_V0. */
12412 cpu_M0 = tcg_temp_new_i64();
12415 static void arm_tr_tb_start(DisasContextBase *dcbase, CPUState *cpu)
12417 DisasContext *dc = container_of(dcbase, DisasContext, base);
12419 /* A note on handling of the condexec (IT) bits:
12421 * We want to avoid the overhead of having to write the updated condexec
12422 * bits back to the CPUARMState for every instruction in an IT block. So:
12423 * (1) if the condexec bits are not already zero then we write
12424 * zero back into the CPUARMState now. This avoids complications trying
12425 * to do it at the end of the block. (For example if we don't do this
12426 * it's hard to identify whether we can safely skip writing condexec
12427 * at the end of the TB, which we definitely want to do for the case
12428 * where a TB doesn't do anything with the IT state at all.)
12429 * (2) if we are going to leave the TB then we call gen_set_condexec()
12430 * which will write the correct value into CPUARMState if zero is wrong.
12431 * This is done both for leaving the TB at the end, and for leaving
12432 * it because of an exception we know will happen, which is done in
12433 * gen_exception_insn(). The latter is necessary because we need to
12434 * leave the TB with the PC/IT state just prior to execution of the
12435 * instruction which caused the exception.
12436 * (3) if we leave the TB unexpectedly (eg a data abort on a load)
12437 * then the CPUARMState will be wrong and we need to reset it.
12438 * This is handled in the same way as restoration of the
12439 * PC in these situations; we save the value of the condexec bits
12440 * for each PC via tcg_gen_insn_start(), and restore_state_to_opc()
12441 * then uses this to restore them after an exception.
12443 * Note that there are no instructions which can read the condexec
12444 * bits, and none which can write non-static values to them, so
12445 * we don't need to care about whether CPUARMState is correct in the
12449 /* Reset the conditional execution bits immediately. This avoids
12450 complications trying to do it at the end of the block. */
12451 if (dc->condexec_mask || dc->condexec_cond) {
12452 TCGv_i32 tmp = tcg_temp_new_i32();
12453 tcg_gen_movi_i32(tmp, 0);
12454 store_cpu_field(tmp, condexec_bits);
12456 tcg_clear_temp_count();
12459 static void arm_tr_insn_start(DisasContextBase *dcbase, CPUState *cpu)
12461 DisasContext *dc = container_of(dcbase, DisasContext, base);
12463 tcg_gen_insn_start(dc->pc,
12464 (dc->condexec_cond << 4) | (dc->condexec_mask >> 1),
12466 dc->insn_start = tcg_last_op();
12469 static bool arm_tr_breakpoint_check(DisasContextBase *dcbase, CPUState *cpu,
12470 const CPUBreakpoint *bp)
12472 DisasContext *dc = container_of(dcbase, DisasContext, base);
12474 if (bp->flags & BP_CPU) {
12475 gen_set_condexec(dc);
12476 gen_set_pc_im(dc, dc->pc);
12477 gen_helper_check_breakpoints(cpu_env);
12478 /* End the TB early; it's likely not going to be executed */
12479 dc->base.is_jmp = DISAS_TOO_MANY;
12481 gen_exception_internal_insn(dc, 0, EXCP_DEBUG);
12482 /* The address covered by the breakpoint must be
12483 included in [tb->pc, tb->pc + tb->size) in order
12484 to for it to be properly cleared -- thus we
12485 increment the PC here so that the logic setting
12486 tb->size below does the right thing. */
12487 /* TODO: Advance PC by correct instruction length to
12488 * avoid disassembler error messages */
12490 dc->base.is_jmp = DISAS_NORETURN;
12496 static bool arm_pre_translate_insn(DisasContext *dc)
12498 #ifdef CONFIG_USER_ONLY
12499 /* Intercept jump to the magic kernel page. */
12500 if (dc->pc >= 0xffff0000) {
12501 /* We always get here via a jump, so know we are not in a
12502 conditional execution block. */
12503 gen_exception_internal(EXCP_KERNEL_TRAP);
12504 dc->base.is_jmp = DISAS_NORETURN;
12509 if (dc->ss_active && !dc->pstate_ss) {
12510 /* Singlestep state is Active-pending.
12511 * If we're in this state at the start of a TB then either
12512 * a) we just took an exception to an EL which is being debugged
12513 * and this is the first insn in the exception handler
12514 * b) debug exceptions were masked and we just unmasked them
12515 * without changing EL (eg by clearing PSTATE.D)
12516 * In either case we're going to take a swstep exception in the
12517 * "did not step an insn" case, and so the syndrome ISV and EX
12518 * bits should be zero.
12520 assert(dc->base.num_insns == 1);
12521 gen_exception(EXCP_UDEF, syn_swstep(dc->ss_same_el, 0, 0),
12522 default_exception_el(dc));
12523 dc->base.is_jmp = DISAS_NORETURN;
12530 static void arm_post_translate_insn(DisasContext *dc)
12532 if (dc->condjmp && !dc->base.is_jmp) {
12533 gen_set_label(dc->condlabel);
12536 dc->base.pc_next = dc->pc;
12537 translator_loop_temp_check(&dc->base);
12540 static void arm_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
12542 DisasContext *dc = container_of(dcbase, DisasContext, base);
12543 CPUARMState *env = cpu->env_ptr;
12546 if (arm_pre_translate_insn(dc)) {
12550 insn = arm_ldl_code(env, dc->pc, dc->sctlr_b);
12553 disas_arm_insn(dc, insn);
12555 arm_post_translate_insn(dc);
12557 /* ARM is a fixed-length ISA. We performed the cross-page check
12558 in init_disas_context by adjusting max_insns. */
12561 static bool thumb_insn_is_unconditional(DisasContext *s, uint32_t insn)
12563 /* Return true if this Thumb insn is always unconditional,
12564 * even inside an IT block. This is true of only a very few
12565 * instructions: BKPT, HLT, and SG.
12567 * A larger class of instructions are UNPREDICTABLE if used
12568 * inside an IT block; we do not need to detect those here, because
12569 * what we do by default (perform the cc check and update the IT
12570 * bits state machine) is a permitted CONSTRAINED UNPREDICTABLE
12571 * choice for those situations.
12573 * insn is either a 16-bit or a 32-bit instruction; the two are
12574 * distinguishable because for the 16-bit case the top 16 bits
12575 * are zeroes, and that isn't a valid 32-bit encoding.
12577 if ((insn & 0xffffff00) == 0xbe00) {
12582 if ((insn & 0xffffffc0) == 0xba80 && arm_dc_feature(s, ARM_FEATURE_V8) &&
12583 !arm_dc_feature(s, ARM_FEATURE_M)) {
12584 /* HLT: v8A only. This is unconditional even when it is going to
12585 * UNDEF; see the v8A ARM ARM DDI0487B.a H3.3.
12586 * For v7 cores this was a plain old undefined encoding and so
12587 * honours its cc check. (We might be using the encoding as
12588 * a semihosting trap, but we don't change the cc check behaviour
12589 * on that account, because a debugger connected to a real v7A
12590 * core and emulating semihosting traps by catching the UNDEF
12591 * exception would also only see cases where the cc check passed.
12592 * No guest code should be trying to do a HLT semihosting trap
12593 * in an IT block anyway.
12598 if (insn == 0xe97fe97f && arm_dc_feature(s, ARM_FEATURE_V8) &&
12599 arm_dc_feature(s, ARM_FEATURE_M)) {
12607 static void thumb_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
12609 DisasContext *dc = container_of(dcbase, DisasContext, base);
12610 CPUARMState *env = cpu->env_ptr;
12614 if (arm_pre_translate_insn(dc)) {
12618 insn = arm_lduw_code(env, dc->pc, dc->sctlr_b);
12619 is_16bit = thumb_insn_is_16bit(dc, insn);
12622 uint32_t insn2 = arm_lduw_code(env, dc->pc, dc->sctlr_b);
12624 insn = insn << 16 | insn2;
12629 if (dc->condexec_mask && !thumb_insn_is_unconditional(dc, insn)) {
12630 uint32_t cond = dc->condexec_cond;
12632 if (cond != 0x0e) { /* Skip conditional when condition is AL. */
12633 dc->condlabel = gen_new_label();
12634 arm_gen_test_cc(cond ^ 1, dc->condlabel);
12640 disas_thumb_insn(dc, insn);
12642 disas_thumb2_insn(dc, insn);
12645 /* Advance the Thumb condexec condition. */
12646 if (dc->condexec_mask) {
12647 dc->condexec_cond = ((dc->condexec_cond & 0xe) |
12648 ((dc->condexec_mask >> 4) & 1));
12649 dc->condexec_mask = (dc->condexec_mask << 1) & 0x1f;
12650 if (dc->condexec_mask == 0) {
12651 dc->condexec_cond = 0;
12655 arm_post_translate_insn(dc);
12657 /* Thumb is a variable-length ISA. Stop translation when the next insn
12658 * will touch a new page. This ensures that prefetch aborts occur at
12661 * We want to stop the TB if the next insn starts in a new page,
12662 * or if it spans between this page and the next. This means that
12663 * if we're looking at the last halfword in the page we need to
12664 * see if it's a 16-bit Thumb insn (which will fit in this TB)
12665 * or a 32-bit Thumb insn (which won't).
12666 * This is to avoid generating a silly TB with a single 16-bit insn
12667 * in it at the end of this page (which would execute correctly
12668 * but isn't very efficient).
12670 if (dc->base.is_jmp == DISAS_NEXT
12671 && (dc->pc - dc->page_start >= TARGET_PAGE_SIZE
12672 || (dc->pc - dc->page_start >= TARGET_PAGE_SIZE - 3
12673 && insn_crosses_page(env, dc)))) {
12674 dc->base.is_jmp = DISAS_TOO_MANY;
12678 static void arm_tr_tb_stop(DisasContextBase *dcbase, CPUState *cpu)
12680 DisasContext *dc = container_of(dcbase, DisasContext, base);
12682 if (tb_cflags(dc->base.tb) & CF_LAST_IO && dc->condjmp) {
12683 /* FIXME: This can theoretically happen with self-modifying code. */
12684 cpu_abort(cpu, "IO on conditional branch instruction");
12687 /* At this stage dc->condjmp will only be set when the skipped
12688 instruction was a conditional branch or trap, and the PC has
12689 already been written. */
12690 gen_set_condexec(dc);
12691 if (dc->base.is_jmp == DISAS_BX_EXCRET) {
12692 /* Exception return branches need some special case code at the
12693 * end of the TB, which is complex enough that it has to
12694 * handle the single-step vs not and the condition-failed
12695 * insn codepath itself.
12697 gen_bx_excret_final_code(dc);
12698 } else if (unlikely(is_singlestepping(dc))) {
12699 /* Unconditional and "condition passed" instruction codepath. */
12700 switch (dc->base.is_jmp) {
12702 gen_ss_advance(dc);
12703 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
12704 default_exception_el(dc));
12707 gen_ss_advance(dc);
12708 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
12711 gen_ss_advance(dc);
12712 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
12715 case DISAS_TOO_MANY:
12717 gen_set_pc_im(dc, dc->pc);
12720 /* FIXME: Single stepping a WFI insn will not halt the CPU. */
12721 gen_singlestep_exception(dc);
12723 case DISAS_NORETURN:
12727 /* While branches must always occur at the end of an IT block,
12728 there are a few other things that can cause us to terminate
12729 the TB in the middle of an IT block:
12730 - Exception generating instructions (bkpt, swi, undefined).
12732 - Hardware watchpoints.
12733 Hardware breakpoints have already been handled and skip this code.
12735 switch(dc->base.is_jmp) {
12737 case DISAS_TOO_MANY:
12738 gen_goto_tb(dc, 1, dc->pc);
12744 gen_set_pc_im(dc, dc->pc);
12747 /* indicate that the hash table must be used to find the next TB */
12748 tcg_gen_exit_tb(NULL, 0);
12750 case DISAS_NORETURN:
12751 /* nothing more to generate */
12755 TCGv_i32 tmp = tcg_const_i32((dc->thumb &&
12756 !(dc->insn & (1U << 31))) ? 2 : 4);
12758 gen_helper_wfi(cpu_env, tmp);
12759 tcg_temp_free_i32(tmp);
12760 /* The helper doesn't necessarily throw an exception, but we
12761 * must go back to the main loop to check for interrupts anyway.
12763 tcg_gen_exit_tb(NULL, 0);
12767 gen_helper_wfe(cpu_env);
12770 gen_helper_yield(cpu_env);
12773 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
12774 default_exception_el(dc));
12777 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
12780 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
12786 /* "Condition failed" instruction codepath for the branch/trap insn */
12787 gen_set_label(dc->condlabel);
12788 gen_set_condexec(dc);
12789 if (unlikely(is_singlestepping(dc))) {
12790 gen_set_pc_im(dc, dc->pc);
12791 gen_singlestep_exception(dc);
12793 gen_goto_tb(dc, 1, dc->pc);
12797 /* Functions above can change dc->pc, so re-align db->pc_next */
12798 dc->base.pc_next = dc->pc;
12801 static void arm_tr_disas_log(const DisasContextBase *dcbase, CPUState *cpu)
12803 DisasContext *dc = container_of(dcbase, DisasContext, base);
12805 qemu_log("IN: %s\n", lookup_symbol(dc->base.pc_first));
12806 log_target_disas(cpu, dc->base.pc_first, dc->base.tb->size);
12809 static const TranslatorOps arm_translator_ops = {
12810 .init_disas_context = arm_tr_init_disas_context,
12811 .tb_start = arm_tr_tb_start,
12812 .insn_start = arm_tr_insn_start,
12813 .breakpoint_check = arm_tr_breakpoint_check,
12814 .translate_insn = arm_tr_translate_insn,
12815 .tb_stop = arm_tr_tb_stop,
12816 .disas_log = arm_tr_disas_log,
12819 static const TranslatorOps thumb_translator_ops = {
12820 .init_disas_context = arm_tr_init_disas_context,
12821 .tb_start = arm_tr_tb_start,
12822 .insn_start = arm_tr_insn_start,
12823 .breakpoint_check = arm_tr_breakpoint_check,
12824 .translate_insn = thumb_tr_translate_insn,
12825 .tb_stop = arm_tr_tb_stop,
12826 .disas_log = arm_tr_disas_log,
12829 /* generate intermediate code for basic block 'tb'. */
12830 void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb)
12833 const TranslatorOps *ops = &arm_translator_ops;
12835 if (ARM_TBFLAG_THUMB(tb->flags)) {
12836 ops = &thumb_translator_ops;
12838 #ifdef TARGET_AARCH64
12839 if (ARM_TBFLAG_AARCH64_STATE(tb->flags)) {
12840 ops = &aarch64_translator_ops;
12844 translator_loop(ops, &dc.base, cpu, tb);
12847 static const char *cpu_mode_names[16] = {
12848 "usr", "fiq", "irq", "svc", "???", "???", "mon", "abt",
12849 "???", "???", "hyp", "und", "???", "???", "???", "sys"
12852 void arm_cpu_dump_state(CPUState *cs, FILE *f, fprintf_function cpu_fprintf,
12855 ARMCPU *cpu = ARM_CPU(cs);
12856 CPUARMState *env = &cpu->env;
12860 aarch64_cpu_dump_state(cs, f, cpu_fprintf, flags);
12864 for(i=0;i<16;i++) {
12865 cpu_fprintf(f, "R%02d=%08x", i, env->regs[i]);
12867 cpu_fprintf(f, "\n");
12869 cpu_fprintf(f, " ");
12872 if (arm_feature(env, ARM_FEATURE_M)) {
12873 uint32_t xpsr = xpsr_read(env);
12875 const char *ns_status = "";
12877 if (arm_feature(env, ARM_FEATURE_M_SECURITY)) {
12878 ns_status = env->v7m.secure ? "S " : "NS ";
12881 if (xpsr & XPSR_EXCP) {
12884 if (env->v7m.control[env->v7m.secure] & R_V7M_CONTROL_NPRIV_MASK) {
12885 mode = "unpriv-thread";
12887 mode = "priv-thread";
12891 cpu_fprintf(f, "XPSR=%08x %c%c%c%c %c %s%s\n",
12893 xpsr & XPSR_N ? 'N' : '-',
12894 xpsr & XPSR_Z ? 'Z' : '-',
12895 xpsr & XPSR_C ? 'C' : '-',
12896 xpsr & XPSR_V ? 'V' : '-',
12897 xpsr & XPSR_T ? 'T' : 'A',
12901 uint32_t psr = cpsr_read(env);
12902 const char *ns_status = "";
12904 if (arm_feature(env, ARM_FEATURE_EL3) &&
12905 (psr & CPSR_M) != ARM_CPU_MODE_MON) {
12906 ns_status = env->cp15.scr_el3 & SCR_NS ? "NS " : "S ";
12909 cpu_fprintf(f, "PSR=%08x %c%c%c%c %c %s%s%d\n",
12911 psr & CPSR_N ? 'N' : '-',
12912 psr & CPSR_Z ? 'Z' : '-',
12913 psr & CPSR_C ? 'C' : '-',
12914 psr & CPSR_V ? 'V' : '-',
12915 psr & CPSR_T ? 'T' : 'A',
12917 cpu_mode_names[psr & 0xf], (psr & 0x10) ? 32 : 26);
12920 if (flags & CPU_DUMP_FPU) {
12921 int numvfpregs = 0;
12922 if (arm_feature(env, ARM_FEATURE_VFP)) {
12925 if (arm_feature(env, ARM_FEATURE_VFP3)) {
12928 for (i = 0; i < numvfpregs; i++) {
12929 uint64_t v = *aa32_vfp_dreg(env, i);
12930 cpu_fprintf(f, "s%02d=%08x s%02d=%08x d%02d=%016" PRIx64 "\n",
12931 i * 2, (uint32_t)v,
12932 i * 2 + 1, (uint32_t)(v >> 32),
12935 cpu_fprintf(f, "FPSCR: %08x\n", (int)env->vfp.xregs[ARM_VFP_FPSCR]);
12939 void restore_state_to_opc(CPUARMState *env, TranslationBlock *tb,
12940 target_ulong *data)
12944 env->condexec_bits = 0;
12945 env->exception.syndrome = data[2] << ARM_INSN_START_WORD2_SHIFT;
12947 env->regs[15] = data[0];
12948 env->condexec_bits = data[1];
12949 env->exception.syndrome = data[2] << ARM_INSN_START_WORD2_SHIFT;