1/*
2 * ARM translation
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2005-2007 CodeSourcery
6 * Copyright (c) 2007 OpenedHand, Ltd.
7 *
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2 of the License, or (at your option) any later version.
12 *
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
20 */
21#include "qemu/osdep.h"
22
23#include "cpu.h"
24#include "internals.h"
25#include "disas/disas.h"
26#include "exec/exec-all.h"
27#include "tcg-op.h"
28#include "tcg-op-gvec.h"
29#include "qemu/log.h"
30#include "qemu/bitops.h"
31#include "arm_ldst.h"
32#include "hw/semihosting/semihost.h"
33
34#include "exec/helper-proto.h"
35#include "exec/helper-gen.h"
36
37#include "trace-tcg.h"
38#include "exec/log.h"
39
40
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 dc_isar_feature(jazelle, s)
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)
51
52#define ARCH(x) do { if (!ENABLE_ARCH_##x) goto illegal_op; } while(0)
53
54#include "translate.h"
55
56#if defined(CONFIG_USER_ONLY)
57#define IS_USER(s) 1
58#else
59#define IS_USER(s) (s->user)
60#endif
61
62/* We reuse the same 64-bit temporaries for efficiency. */
63static TCGv_i64 cpu_V0, cpu_V1, cpu_M0;
64static TCGv_i32 cpu_R[16];
65TCGv_i32 cpu_CF, cpu_NF, cpu_VF, cpu_ZF;
66TCGv_i64 cpu_exclusive_addr;
67TCGv_i64 cpu_exclusive_val;
68
69#include "exec/gen-icount.h"
70
71static const char * const regnames[] =
72 { "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7",
73 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "pc" };
74
75/* Function prototypes for gen_ functions calling Neon helpers. */
76typedef void NeonGenThreeOpEnvFn(TCGv_i32, TCGv_env, TCGv_i32,
77 TCGv_i32, TCGv_i32);
78/* Function prototypes for gen_ functions for fix point conversions */
79typedef void VFPGenFixPointFn(TCGv_i32, TCGv_i32, TCGv_i32, TCGv_ptr);
80
81/* initialize TCG globals. */
82void arm_translate_init(void)
83{
84 int i;
85
86 for (i = 0; i < 16; i++) {
87 cpu_R[i] = tcg_global_mem_new_i32(cpu_env,
88 offsetof(CPUARMState, regs[i]),
89 regnames[i]);
90 }
91 cpu_CF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, CF), "CF");
92 cpu_NF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, NF), "NF");
93 cpu_VF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, VF), "VF");
94 cpu_ZF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, ZF), "ZF");
95
96 cpu_exclusive_addr = tcg_global_mem_new_i64(cpu_env,
97 offsetof(CPUARMState, exclusive_addr), "exclusive_addr");
98 cpu_exclusive_val = tcg_global_mem_new_i64(cpu_env,
99 offsetof(CPUARMState, exclusive_val), "exclusive_val");
100
101 a64_translate_init();
102}
103
104/* Flags for the disas_set_da_iss info argument:
105 * lower bits hold the Rt register number, higher bits are flags.
106 */
107typedef enum ISSInfo {
108 ISSNone = 0,
109 ISSRegMask = 0x1f,
110 ISSInvalid = (1 << 5),
111 ISSIsAcqRel = (1 << 6),
112 ISSIsWrite = (1 << 7),
113 ISSIs16Bit = (1 << 8),
114} ISSInfo;
115
116/* Save the syndrome information for a Data Abort */
117static void disas_set_da_iss(DisasContext *s, MemOp memop, ISSInfo issinfo)
118{
119 uint32_t syn;
120 int sas = memop & MO_SIZE;
121 bool sse = memop & MO_SIGN;
122 bool is_acqrel = issinfo & ISSIsAcqRel;
123 bool is_write = issinfo & ISSIsWrite;
124 bool is_16bit = issinfo & ISSIs16Bit;
125 int srt = issinfo & ISSRegMask;
126
127 if (issinfo & ISSInvalid) {
128 /* Some callsites want to conditionally provide ISS info,
129 * eg "only if this was not a writeback"
130 */
131 return;
132 }
133
134 if (srt == 15) {
135 /* For AArch32, insns where the src/dest is R15 never generate
136 * ISS information. Catching that here saves checking at all
137 * the call sites.
138 */
139 return;
140 }
141
142 syn = syn_data_abort_with_iss(0, sas, sse, srt, 0, is_acqrel,
143 0, 0, 0, is_write, 0, is_16bit);
144 disas_set_insn_syndrome(s, syn);
145}
146
147static inline int get_a32_user_mem_index(DisasContext *s)
148{
149 /* Return the core mmu_idx to use for A32/T32 "unprivileged load/store"
150 * insns:
151 * if PL2, UNPREDICTABLE (we choose to implement as if PL0)
152 * otherwise, access as if at PL0.
153 */
154 switch (s->mmu_idx) {
155 case ARMMMUIdx_S1E2: /* this one is UNPREDICTABLE */
156 case ARMMMUIdx_S12NSE0:
157 case ARMMMUIdx_S12NSE1:
158 return arm_to_core_mmu_idx(ARMMMUIdx_S12NSE0);
159 case ARMMMUIdx_S1E3:
160 case ARMMMUIdx_S1SE0:
161 case ARMMMUIdx_S1SE1:
162 return arm_to_core_mmu_idx(ARMMMUIdx_S1SE0);
163 case ARMMMUIdx_MUser:
164 case ARMMMUIdx_MPriv:
165 return arm_to_core_mmu_idx(ARMMMUIdx_MUser);
166 case ARMMMUIdx_MUserNegPri:
167 case ARMMMUIdx_MPrivNegPri:
168 return arm_to_core_mmu_idx(ARMMMUIdx_MUserNegPri);
169 case ARMMMUIdx_MSUser:
170 case ARMMMUIdx_MSPriv:
171 return arm_to_core_mmu_idx(ARMMMUIdx_MSUser);
172 case ARMMMUIdx_MSUserNegPri:
173 case ARMMMUIdx_MSPrivNegPri:
174 return arm_to_core_mmu_idx(ARMMMUIdx_MSUserNegPri);
175 case ARMMMUIdx_S2NS:
176 default:
177 g_assert_not_reached();
178 }
179}
180
181static inline TCGv_i32 load_cpu_offset(int offset)
182{
183 TCGv_i32 tmp = tcg_temp_new_i32();
184 tcg_gen_ld_i32(tmp, cpu_env, offset);
185 return tmp;
186}
187
188#define load_cpu_field(name) load_cpu_offset(offsetof(CPUARMState, name))
189
190static inline void store_cpu_offset(TCGv_i32 var, int offset)
191{
192 tcg_gen_st_i32(var, cpu_env, offset);
193 tcg_temp_free_i32(var);
194}
195
196#define store_cpu_field(var, name) \
197 store_cpu_offset(var, offsetof(CPUARMState, name))
198
199/* The architectural value of PC. */
200static uint32_t read_pc(DisasContext *s)
201{
202 return s->pc_curr + (s->thumb ? 4 : 8);
203}
204
205/* Set a variable to the value of a CPU register. */
206static void load_reg_var(DisasContext *s, TCGv_i32 var, int reg)
207{
208 if (reg == 15) {
209 tcg_gen_movi_i32(var, read_pc(s));
210 } else {
211 tcg_gen_mov_i32(var, cpu_R[reg]);
212 }
213}
214
215/* Create a new temporary and set it to the value of a CPU register. */
216static inline TCGv_i32 load_reg(DisasContext *s, int reg)
217{
218 TCGv_i32 tmp = tcg_temp_new_i32();
219 load_reg_var(s, tmp, reg);
220 return tmp;
221}
222
223/*
224 * Create a new temp, REG + OFS, except PC is ALIGN(PC, 4).
225 * This is used for load/store for which use of PC implies (literal),
226 * or ADD that implies ADR.
227 */
228static TCGv_i32 add_reg_for_lit(DisasContext *s, int reg, int ofs)
229{
230 TCGv_i32 tmp = tcg_temp_new_i32();
231
232 if (reg == 15) {
233 tcg_gen_movi_i32(tmp, (read_pc(s) & ~3) + ofs);
234 } else {
235 tcg_gen_addi_i32(tmp, cpu_R[reg], ofs);
236 }
237 return tmp;
238}
239
240/* Set a CPU register. The source must be a temporary and will be
241 marked as dead. */
242static void store_reg(DisasContext *s, int reg, TCGv_i32 var)
243{
244 if (reg == 15) {
245 /* In Thumb mode, we must ignore bit 0.
246 * In ARM mode, for ARMv4 and ARMv5, it is UNPREDICTABLE if bits [1:0]
247 * are not 0b00, but for ARMv6 and above, we must ignore bits [1:0].
248 * We choose to ignore [1:0] in ARM mode for all architecture versions.
249 */
250 tcg_gen_andi_i32(var, var, s->thumb ? ~1 : ~3);
251 s->base.is_jmp = DISAS_JUMP;
252 }
253 tcg_gen_mov_i32(cpu_R[reg], var);
254 tcg_temp_free_i32(var);
255}
256
257/*
258 * Variant of store_reg which applies v8M stack-limit checks before updating
259 * SP. If the check fails this will result in an exception being taken.
260 * We disable the stack checks for CONFIG_USER_ONLY because we have
261 * no idea what the stack limits should be in that case.
262 * If stack checking is not being done this just acts like store_reg().
263 */
264static void store_sp_checked(DisasContext *s, TCGv_i32 var)
265{
266#ifndef CONFIG_USER_ONLY
267 if (s->v8m_stackcheck) {
268 gen_helper_v8m_stackcheck(cpu_env, var);
269 }
270#endif
271 store_reg(s, 13, var);
272}
273
274/* Value extensions. */
275#define gen_uxtb(var) tcg_gen_ext8u_i32(var, var)
276#define gen_uxth(var) tcg_gen_ext16u_i32(var, var)
277#define gen_sxtb(var) tcg_gen_ext8s_i32(var, var)
278#define gen_sxth(var) tcg_gen_ext16s_i32(var, var)
279
280#define gen_sxtb16(var) gen_helper_sxtb16(var, var)
281#define gen_uxtb16(var) gen_helper_uxtb16(var, var)
282
283
284static inline void gen_set_cpsr(TCGv_i32 var, uint32_t mask)
285{
286 TCGv_i32 tmp_mask = tcg_const_i32(mask);
287 gen_helper_cpsr_write(cpu_env, var, tmp_mask);
288 tcg_temp_free_i32(tmp_mask);
289}
290/* Set NZCV flags from the high 4 bits of var. */
291#define gen_set_nzcv(var) gen_set_cpsr(var, CPSR_NZCV)
292
293static void gen_exception_internal(int excp)
294{
295 TCGv_i32 tcg_excp = tcg_const_i32(excp);
296
297 assert(excp_is_internal(excp));
298 gen_helper_exception_internal(cpu_env, tcg_excp);
299 tcg_temp_free_i32(tcg_excp);
300}
301
302static void gen_step_complete_exception(DisasContext *s)
303{
304 /* We just completed step of an insn. Move from Active-not-pending
305 * to Active-pending, and then also take the swstep exception.
306 * This corresponds to making the (IMPDEF) choice to prioritize
307 * swstep exceptions over asynchronous exceptions taken to an exception
308 * level where debug is disabled. This choice has the advantage that
309 * we do not need to maintain internal state corresponding to the
310 * ISV/EX syndrome bits between completion of the step and generation
311 * of the exception, and our syndrome information is always correct.
312 */
313 gen_ss_advance(s);
314 gen_swstep_exception(s, 1, s->is_ldex);
315 s->base.is_jmp = DISAS_NORETURN;
316}
317
318static void gen_singlestep_exception(DisasContext *s)
319{
320 /* Generate the right kind of exception for singlestep, which is
321 * either the architectural singlestep or EXCP_DEBUG for QEMU's
322 * gdb singlestepping.
323 */
324 if (s->ss_active) {
325 gen_step_complete_exception(s);
326 } else {
327 gen_exception_internal(EXCP_DEBUG);
328 }
329}
330
331static inline bool is_singlestepping(DisasContext *s)
332{
333 /* Return true if we are singlestepping either because of
334 * architectural singlestep or QEMU gdbstub singlestep. This does
335 * not include the command line '-singlestep' mode which is rather
336 * misnamed as it only means "one instruction per TB" and doesn't
337 * affect the code we generate.
338 */
339 return s->base.singlestep_enabled || s->ss_active;
340}
341
342static void gen_smul_dual(TCGv_i32 a, TCGv_i32 b)
343{
344 TCGv_i32 tmp1 = tcg_temp_new_i32();
345 TCGv_i32 tmp2 = tcg_temp_new_i32();
346 tcg_gen_ext16s_i32(tmp1, a);
347 tcg_gen_ext16s_i32(tmp2, b);
348 tcg_gen_mul_i32(tmp1, tmp1, tmp2);
349 tcg_temp_free_i32(tmp2);
350 tcg_gen_sari_i32(a, a, 16);
351 tcg_gen_sari_i32(b, b, 16);
352 tcg_gen_mul_i32(b, b, a);
353 tcg_gen_mov_i32(a, tmp1);
354 tcg_temp_free_i32(tmp1);
355}
356
357/* Byteswap each halfword. */
358static void gen_rev16(TCGv_i32 dest, TCGv_i32 var)
359{
360 TCGv_i32 tmp = tcg_temp_new_i32();
361 TCGv_i32 mask = tcg_const_i32(0x00ff00ff);
362 tcg_gen_shri_i32(tmp, var, 8);
363 tcg_gen_and_i32(tmp, tmp, mask);
364 tcg_gen_and_i32(var, var, mask);
365 tcg_gen_shli_i32(var, var, 8);
366 tcg_gen_or_i32(dest, var, tmp);
367 tcg_temp_free_i32(mask);
368 tcg_temp_free_i32(tmp);
369}
370
371/* Byteswap low halfword and sign extend. */
372static void gen_revsh(TCGv_i32 dest, TCGv_i32 var)
373{
374 tcg_gen_ext16u_i32(var, var);
375 tcg_gen_bswap16_i32(var, var);
376 tcg_gen_ext16s_i32(dest, var);
377}
378
379/* 32x32->64 multiply. Marks inputs as dead. */
380static TCGv_i64 gen_mulu_i64_i32(TCGv_i32 a, TCGv_i32 b)
381{
382 TCGv_i32 lo = tcg_temp_new_i32();
383 TCGv_i32 hi = tcg_temp_new_i32();
384 TCGv_i64 ret;
385
386 tcg_gen_mulu2_i32(lo, hi, a, b);
387 tcg_temp_free_i32(a);
388 tcg_temp_free_i32(b);
389
390 ret = tcg_temp_new_i64();
391 tcg_gen_concat_i32_i64(ret, lo, hi);
392 tcg_temp_free_i32(lo);
393 tcg_temp_free_i32(hi);
394
395 return ret;
396}
397
398static TCGv_i64 gen_muls_i64_i32(TCGv_i32 a, TCGv_i32 b)
399{
400 TCGv_i32 lo = tcg_temp_new_i32();
401 TCGv_i32 hi = tcg_temp_new_i32();
402 TCGv_i64 ret;
403
404 tcg_gen_muls2_i32(lo, hi, a, b);
405 tcg_temp_free_i32(a);
406 tcg_temp_free_i32(b);
407
408 ret = tcg_temp_new_i64();
409 tcg_gen_concat_i32_i64(ret, lo, hi);
410 tcg_temp_free_i32(lo);
411 tcg_temp_free_i32(hi);
412
413 return ret;
414}
415
416/* Swap low and high halfwords. */
417static void gen_swap_half(TCGv_i32 var)
418{
419 tcg_gen_rotri_i32(var, var, 16);
420}
421
422/* Dual 16-bit add. Result placed in t0 and t1 is marked as dead.
423 tmp = (t0 ^ t1) & 0x8000;
424 t0 &= ~0x8000;
425 t1 &= ~0x8000;
426 t0 = (t0 + t1) ^ tmp;
427 */
428
429static void gen_add16(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
430{
431 TCGv_i32 tmp = tcg_temp_new_i32();
432 tcg_gen_xor_i32(tmp, t0, t1);
433 tcg_gen_andi_i32(tmp, tmp, 0x8000);
434 tcg_gen_andi_i32(t0, t0, ~0x8000);
435 tcg_gen_andi_i32(t1, t1, ~0x8000);
436 tcg_gen_add_i32(t0, t0, t1);
437 tcg_gen_xor_i32(dest, t0, tmp);
438 tcg_temp_free_i32(tmp);
439}
440
441/* Set N and Z flags from var. */
442static inline void gen_logic_CC(TCGv_i32 var)
443{
444 tcg_gen_mov_i32(cpu_NF, var);
445 tcg_gen_mov_i32(cpu_ZF, var);
446}
447
448/* dest = T0 + T1 + CF. */
449static void gen_add_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
450{
451 tcg_gen_add_i32(dest, t0, t1);
452 tcg_gen_add_i32(dest, dest, cpu_CF);
453}
454
455/* dest = T0 - T1 + CF - 1. */
456static void gen_sub_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
457{
458 tcg_gen_sub_i32(dest, t0, t1);
459 tcg_gen_add_i32(dest, dest, cpu_CF);
460 tcg_gen_subi_i32(dest, dest, 1);
461}
462
463/* dest = T0 + T1. Compute C, N, V and Z flags */
464static void gen_add_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
465{
466 TCGv_i32 tmp = tcg_temp_new_i32();
467 tcg_gen_movi_i32(tmp, 0);
468 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, t1, tmp);
469 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
470 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
471 tcg_gen_xor_i32(tmp, t0, t1);
472 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
473 tcg_temp_free_i32(tmp);
474 tcg_gen_mov_i32(dest, cpu_NF);
475}
476
477/* dest = T0 + T1 + CF. Compute C, N, V and Z flags */
478static void gen_adc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
479{
480 TCGv_i32 tmp = tcg_temp_new_i32();
481 if (TCG_TARGET_HAS_add2_i32) {
482 tcg_gen_movi_i32(tmp, 0);
483 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, cpu_CF, tmp);
484 tcg_gen_add2_i32(cpu_NF, cpu_CF, cpu_NF, cpu_CF, t1, tmp);
485 } else {
486 TCGv_i64 q0 = tcg_temp_new_i64();
487 TCGv_i64 q1 = tcg_temp_new_i64();
488 tcg_gen_extu_i32_i64(q0, t0);
489 tcg_gen_extu_i32_i64(q1, t1);
490 tcg_gen_add_i64(q0, q0, q1);
491 tcg_gen_extu_i32_i64(q1, cpu_CF);
492 tcg_gen_add_i64(q0, q0, q1);
493 tcg_gen_extr_i64_i32(cpu_NF, cpu_CF, q0);
494 tcg_temp_free_i64(q0);
495 tcg_temp_free_i64(q1);
496 }
497 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
498 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
499 tcg_gen_xor_i32(tmp, t0, t1);
500 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
501 tcg_temp_free_i32(tmp);
502 tcg_gen_mov_i32(dest, cpu_NF);
503}
504
505/* dest = T0 - T1. Compute C, N, V and Z flags */
506static void gen_sub_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
507{
508 TCGv_i32 tmp;
509 tcg_gen_sub_i32(cpu_NF, t0, t1);
510 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
511 tcg_gen_setcond_i32(TCG_COND_GEU, cpu_CF, t0, t1);
512 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
513 tmp = tcg_temp_new_i32();
514 tcg_gen_xor_i32(tmp, t0, t1);
515 tcg_gen_and_i32(cpu_VF, cpu_VF, tmp);
516 tcg_temp_free_i32(tmp);
517 tcg_gen_mov_i32(dest, cpu_NF);
518}
519
520/* dest = T0 + ~T1 + CF. Compute C, N, V and Z flags */
521static void gen_sbc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
522{
523 TCGv_i32 tmp = tcg_temp_new_i32();
524 tcg_gen_not_i32(tmp, t1);
525 gen_adc_CC(dest, t0, tmp);
526 tcg_temp_free_i32(tmp);
527}
528
529#define GEN_SHIFT(name) \
530static void gen_##name(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1) \
531{ \
532 TCGv_i32 tmp1, tmp2, tmp3; \
533 tmp1 = tcg_temp_new_i32(); \
534 tcg_gen_andi_i32(tmp1, t1, 0xff); \
535 tmp2 = tcg_const_i32(0); \
536 tmp3 = tcg_const_i32(0x1f); \
537 tcg_gen_movcond_i32(TCG_COND_GTU, tmp2, tmp1, tmp3, tmp2, t0); \
538 tcg_temp_free_i32(tmp3); \
539 tcg_gen_andi_i32(tmp1, tmp1, 0x1f); \
540 tcg_gen_##name##_i32(dest, tmp2, tmp1); \
541 tcg_temp_free_i32(tmp2); \
542 tcg_temp_free_i32(tmp1); \
543}
544GEN_SHIFT(shl)
545GEN_SHIFT(shr)
546#undef GEN_SHIFT
547
548static void gen_sar(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
549{
550 TCGv_i32 tmp1, tmp2;
551 tmp1 = tcg_temp_new_i32();
552 tcg_gen_andi_i32(tmp1, t1, 0xff);
553 tmp2 = tcg_const_i32(0x1f);
554 tcg_gen_movcond_i32(TCG_COND_GTU, tmp1, tmp1, tmp2, tmp2, tmp1);
555 tcg_temp_free_i32(tmp2);
556 tcg_gen_sar_i32(dest, t0, tmp1);
557 tcg_temp_free_i32(tmp1);
558}
559
560static void shifter_out_im(TCGv_i32 var, int shift)
561{
562 tcg_gen_extract_i32(cpu_CF, var, shift, 1);
563}
564
565/* Shift by immediate. Includes special handling for shift == 0. */
566static inline void gen_arm_shift_im(TCGv_i32 var, int shiftop,
567 int shift, int flags)
568{
569 switch (shiftop) {
570 case 0: /* LSL */
571 if (shift != 0) {
572 if (flags)
573 shifter_out_im(var, 32 - shift);
574 tcg_gen_shli_i32(var, var, shift);
575 }
576 break;
577 case 1: /* LSR */
578 if (shift == 0) {
579 if (flags) {
580 tcg_gen_shri_i32(cpu_CF, var, 31);
581 }
582 tcg_gen_movi_i32(var, 0);
583 } else {
584 if (flags)
585 shifter_out_im(var, shift - 1);
586 tcg_gen_shri_i32(var, var, shift);
587 }
588 break;
589 case 2: /* ASR */
590 if (shift == 0)
591 shift = 32;
592 if (flags)
593 shifter_out_im(var, shift - 1);
594 if (shift == 32)
595 shift = 31;
596 tcg_gen_sari_i32(var, var, shift);
597 break;
598 case 3: /* ROR/RRX */
599 if (shift != 0) {
600 if (flags)
601 shifter_out_im(var, shift - 1);
602 tcg_gen_rotri_i32(var, var, shift); break;
603 } else {
604 TCGv_i32 tmp = tcg_temp_new_i32();
605 tcg_gen_shli_i32(tmp, cpu_CF, 31);
606 if (flags)
607 shifter_out_im(var, 0);
608 tcg_gen_shri_i32(var, var, 1);
609 tcg_gen_or_i32(var, var, tmp);
610 tcg_temp_free_i32(tmp);
611 }
612 }
613};
614
615static inline void gen_arm_shift_reg(TCGv_i32 var, int shiftop,
616 TCGv_i32 shift, int flags)
617{
618 if (flags) {
619 switch (shiftop) {
620 case 0: gen_helper_shl_cc(var, cpu_env, var, shift); break;
621 case 1: gen_helper_shr_cc(var, cpu_env, var, shift); break;
622 case 2: gen_helper_sar_cc(var, cpu_env, var, shift); break;
623 case 3: gen_helper_ror_cc(var, cpu_env, var, shift); break;
624 }
625 } else {
626 switch (shiftop) {
627 case 0:
628 gen_shl(var, var, shift);
629 break;
630 case 1:
631 gen_shr(var, var, shift);
632 break;
633 case 2:
634 gen_sar(var, var, shift);
635 break;
636 case 3: tcg_gen_andi_i32(shift, shift, 0x1f);
637 tcg_gen_rotr_i32(var, var, shift); break;
638 }
639 }
640 tcg_temp_free_i32(shift);
641}
642
643/*
644 * Generate a conditional based on ARM condition code cc.
645 * This is common between ARM and Aarch64 targets.
646 */
647void arm_test_cc(DisasCompare *cmp, int cc)
648{
649 TCGv_i32 value;
650 TCGCond cond;
651 bool global = true;
652
653 switch (cc) {
654 case 0: /* eq: Z */
655 case 1: /* ne: !Z */
656 cond = TCG_COND_EQ;
657 value = cpu_ZF;
658 break;
659
660 case 2: /* cs: C */
661 case 3: /* cc: !C */
662 cond = TCG_COND_NE;
663 value = cpu_CF;
664 break;
665
666 case 4: /* mi: N */
667 case 5: /* pl: !N */
668 cond = TCG_COND_LT;
669 value = cpu_NF;
670 break;
671
672 case 6: /* vs: V */
673 case 7: /* vc: !V */
674 cond = TCG_COND_LT;
675 value = cpu_VF;
676 break;
677
678 case 8: /* hi: C && !Z */
679 case 9: /* ls: !C || Z -> !(C && !Z) */
680 cond = TCG_COND_NE;
681 value = tcg_temp_new_i32();
682 global = false;
683 /* CF is 1 for C, so -CF is an all-bits-set mask for C;
684 ZF is non-zero for !Z; so AND the two subexpressions. */
685 tcg_gen_neg_i32(value, cpu_CF);
686 tcg_gen_and_i32(value, value, cpu_ZF);
687 break;
688
689 case 10: /* ge: N == V -> N ^ V == 0 */
690 case 11: /* lt: N != V -> N ^ V != 0 */
691 /* Since we're only interested in the sign bit, == 0 is >= 0. */
692 cond = TCG_COND_GE;
693 value = tcg_temp_new_i32();
694 global = false;
695 tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
696 break;
697
698 case 12: /* gt: !Z && N == V */
699 case 13: /* le: Z || N != V */
700 cond = TCG_COND_NE;
701 value = tcg_temp_new_i32();
702 global = false;
703 /* (N == V) is equal to the sign bit of ~(NF ^ VF). Propagate
704 * the sign bit then AND with ZF to yield the result. */
705 tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
706 tcg_gen_sari_i32(value, value, 31);
707 tcg_gen_andc_i32(value, cpu_ZF, value);
708 break;
709
710 case 14: /* always */
711 case 15: /* always */
712 /* Use the ALWAYS condition, which will fold early.
713 * It doesn't matter what we use for the value. */
714 cond = TCG_COND_ALWAYS;
715 value = cpu_ZF;
716 goto no_invert;
717
718 default:
719 fprintf(stderr, "Bad condition code 0x%x\n", cc);
720 abort();
721 }
722
723 if (cc & 1) {
724 cond = tcg_invert_cond(cond);
725 }
726
727 no_invert:
728 cmp->cond = cond;
729 cmp->value = value;
730 cmp->value_global = global;
731}
732
733void arm_free_cc(DisasCompare *cmp)
734{
735 if (!cmp->value_global) {
736 tcg_temp_free_i32(cmp->value);
737 }
738}
739
740void arm_jump_cc(DisasCompare *cmp, TCGLabel *label)
741{
742 tcg_gen_brcondi_i32(cmp->cond, cmp->value, 0, label);
743}
744
745void arm_gen_test_cc(int cc, TCGLabel *label)
746{
747 DisasCompare cmp;
748 arm_test_cc(&cmp, cc);
749 arm_jump_cc(&cmp, label);
750 arm_free_cc(&cmp);
751}
752
753static inline void gen_set_condexec(DisasContext *s)
754{
755 if (s->condexec_mask) {
756 uint32_t val = (s->condexec_cond << 4) | (s->condexec_mask >> 1);
757 TCGv_i32 tmp = tcg_temp_new_i32();
758 tcg_gen_movi_i32(tmp, val);
759 store_cpu_field(tmp, condexec_bits);
760 }
761}
762
763static inline void gen_set_pc_im(DisasContext *s, target_ulong val)
764{
765 tcg_gen_movi_i32(cpu_R[15], val);
766}
767
768/* Set PC and Thumb state from var. var is marked as dead. */
769static inline void gen_bx(DisasContext *s, TCGv_i32 var)
770{
771 s->base.is_jmp = DISAS_JUMP;
772 tcg_gen_andi_i32(cpu_R[15], var, ~1);
773 tcg_gen_andi_i32(var, var, 1);
774 store_cpu_field(var, thumb);
775}
776
777/*
778 * Set PC and Thumb state from var. var is marked as dead.
779 * For M-profile CPUs, include logic to detect exception-return
780 * branches and handle them. This is needed for Thumb POP/LDM to PC, LDR to PC,
781 * and BX reg, and no others, and happens only for code in Handler mode.
782 * The Security Extension also requires us to check for the FNC_RETURN
783 * which signals a function return from non-secure state; this can happen
784 * in both Handler and Thread mode.
785 * To avoid having to do multiple comparisons in inline generated code,
786 * we make the check we do here loose, so it will match for EXC_RETURN
787 * in Thread mode. For system emulation do_v7m_exception_exit() checks
788 * for these spurious cases and returns without doing anything (giving
789 * the same behaviour as for a branch to a non-magic address).
790 *
791 * In linux-user mode it is unclear what the right behaviour for an
792 * attempted FNC_RETURN should be, because in real hardware this will go
793 * directly to Secure code (ie not the Linux kernel) which will then treat
794 * the error in any way it chooses. For QEMU we opt to make the FNC_RETURN
795 * attempt behave the way it would on a CPU without the security extension,
796 * which is to say "like a normal branch". That means we can simply treat
797 * all branches as normal with no magic address behaviour.
798 */
799static inline void gen_bx_excret(DisasContext *s, TCGv_i32 var)
800{
801 /* Generate the same code here as for a simple bx, but flag via
802 * s->base.is_jmp that we need to do the rest of the work later.
803 */
804 gen_bx(s, var);
805#ifndef CONFIG_USER_ONLY
806 if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY) ||
807 (s->v7m_handler_mode && arm_dc_feature(s, ARM_FEATURE_M))) {
808 s->base.is_jmp = DISAS_BX_EXCRET;
809 }
810#endif
811}
812
813static inline void gen_bx_excret_final_code(DisasContext *s)
814{
815 /* Generate the code to finish possible exception return and end the TB */
816 TCGLabel *excret_label = gen_new_label();
817 uint32_t min_magic;
818
819 if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY)) {
820 /* Covers FNC_RETURN and EXC_RETURN magic */
821 min_magic = FNC_RETURN_MIN_MAGIC;
822 } else {
823 /* EXC_RETURN magic only */
824 min_magic = EXC_RETURN_MIN_MAGIC;
825 }
826
827 /* Is the new PC value in the magic range indicating exception return? */
828 tcg_gen_brcondi_i32(TCG_COND_GEU, cpu_R[15], min_magic, excret_label);
829 /* No: end the TB as we would for a DISAS_JMP */
830 if (is_singlestepping(s)) {
831 gen_singlestep_exception(s);
832 } else {
833 tcg_gen_exit_tb(NULL, 0);
834 }
835 gen_set_label(excret_label);
836 /* Yes: this is an exception return.
837 * At this point in runtime env->regs[15] and env->thumb will hold
838 * the exception-return magic number, which do_v7m_exception_exit()
839 * will read. Nothing else will be able to see those values because
840 * the cpu-exec main loop guarantees that we will always go straight
841 * from raising the exception to the exception-handling code.
842 *
843 * gen_ss_advance(s) does nothing on M profile currently but
844 * calling it is conceptually the right thing as we have executed
845 * this instruction (compare SWI, HVC, SMC handling).
846 */
847 gen_ss_advance(s);
848 gen_exception_internal(EXCP_EXCEPTION_EXIT);
849}
850
851static inline void gen_bxns(DisasContext *s, int rm)
852{
853 TCGv_i32 var = load_reg(s, rm);
854
855 /* The bxns helper may raise an EXCEPTION_EXIT exception, so in theory
856 * we need to sync state before calling it, but:
857 * - we don't need to do gen_set_pc_im() because the bxns helper will
858 * always set the PC itself
859 * - we don't need to do gen_set_condexec() because BXNS is UNPREDICTABLE
860 * unless it's outside an IT block or the last insn in an IT block,
861 * so we know that condexec == 0 (already set at the top of the TB)
862 * is correct in the non-UNPREDICTABLE cases, and we can choose
863 * "zeroes the IT bits" as our UNPREDICTABLE behaviour otherwise.
864 */
865 gen_helper_v7m_bxns(cpu_env, var);
866 tcg_temp_free_i32(var);
867 s->base.is_jmp = DISAS_EXIT;
868}
869
870static inline void gen_blxns(DisasContext *s, int rm)
871{
872 TCGv_i32 var = load_reg(s, rm);
873
874 /* We don't need to sync condexec state, for the same reason as bxns.
875 * We do however need to set the PC, because the blxns helper reads it.
876 * The blxns helper may throw an exception.
877 */
878 gen_set_pc_im(s, s->base.pc_next);
879 gen_helper_v7m_blxns(cpu_env, var);
880 tcg_temp_free_i32(var);
881 s->base.is_jmp = DISAS_EXIT;
882}
883
884/* Variant of store_reg which uses branch&exchange logic when storing
885 to r15 in ARM architecture v7 and above. The source must be a temporary
886 and will be marked as dead. */
887static inline void store_reg_bx(DisasContext *s, int reg, TCGv_i32 var)
888{
889 if (reg == 15 && ENABLE_ARCH_7) {
890 gen_bx(s, var);
891 } else {
892 store_reg(s, reg, var);
893 }
894}
895
896/* Variant of store_reg which uses branch&exchange logic when storing
897 * to r15 in ARM architecture v5T and above. This is used for storing
898 * the results of a LDR/LDM/POP into r15, and corresponds to the cases
899 * in the ARM ARM which use the LoadWritePC() pseudocode function. */
900static inline void store_reg_from_load(DisasContext *s, int reg, TCGv_i32 var)
901{
902 if (reg == 15 && ENABLE_ARCH_5) {
903 gen_bx_excret(s, var);
904 } else {
905 store_reg(s, reg, var);
906 }
907}
908
909#ifdef CONFIG_USER_ONLY
910#define IS_USER_ONLY 1
911#else
912#define IS_USER_ONLY 0
913#endif
914
915/* Abstractions of "generate code to do a guest load/store for
916 * AArch32", where a vaddr is always 32 bits (and is zero
917 * extended if we're a 64 bit core) and data is also
918 * 32 bits unless specifically doing a 64 bit access.
919 * These functions work like tcg_gen_qemu_{ld,st}* except
920 * that the address argument is TCGv_i32 rather than TCGv.
921 */
922
923static inline TCGv gen_aa32_addr(DisasContext *s, TCGv_i32 a32, MemOp op)
924{
925 TCGv addr = tcg_temp_new();
926 tcg_gen_extu_i32_tl(addr, a32);
927
928 /* Not needed for user-mode BE32, where we use MO_BE instead. */
929 if (!IS_USER_ONLY && s->sctlr_b && (op & MO_SIZE) < MO_32) {
930 tcg_gen_xori_tl(addr, addr, 4 - (1 << (op & MO_SIZE)));
931 }
932 return addr;
933}
934
935static void gen_aa32_ld_i32(DisasContext *s, TCGv_i32 val, TCGv_i32 a32,
936 int index, MemOp opc)
937{
938 TCGv addr;
939
940 if (arm_dc_feature(s, ARM_FEATURE_M) &&
941 !arm_dc_feature(s, ARM_FEATURE_M_MAIN)) {
942 opc |= MO_ALIGN;
943 }
944
945 addr = gen_aa32_addr(s, a32, opc);
946 tcg_gen_qemu_ld_i32(val, addr, index, opc);
947 tcg_temp_free(addr);
948}
949
950static void gen_aa32_st_i32(DisasContext *s, TCGv_i32 val, TCGv_i32 a32,
951 int index, MemOp opc)
952{
953 TCGv addr;
954
955 if (arm_dc_feature(s, ARM_FEATURE_M) &&
956 !arm_dc_feature(s, ARM_FEATURE_M_MAIN)) {
957 opc |= MO_ALIGN;
958 }
959
960 addr = gen_aa32_addr(s, a32, opc);
961 tcg_gen_qemu_st_i32(val, addr, index, opc);
962 tcg_temp_free(addr);
963}
964
965#define DO_GEN_LD(SUFF, OPC) \
966static inline void gen_aa32_ld##SUFF(DisasContext *s, TCGv_i32 val, \
967 TCGv_i32 a32, int index) \
968{ \
969 gen_aa32_ld_i32(s, val, a32, index, OPC | s->be_data); \
970}
971
972#define DO_GEN_ST(SUFF, OPC) \
973static inline void gen_aa32_st##SUFF(DisasContext *s, TCGv_i32 val, \
974 TCGv_i32 a32, int index) \
975{ \
976 gen_aa32_st_i32(s, val, a32, index, OPC | s->be_data); \
977}
978
979static inline void gen_aa32_frob64(DisasContext *s, TCGv_i64 val)
980{
981 /* Not needed for user-mode BE32, where we use MO_BE instead. */
982 if (!IS_USER_ONLY && s->sctlr_b) {
983 tcg_gen_rotri_i64(val, val, 32);
984 }
985}
986
987static void gen_aa32_ld_i64(DisasContext *s, TCGv_i64 val, TCGv_i32 a32,
988 int index, MemOp opc)
989{
990 TCGv addr = gen_aa32_addr(s, a32, opc);
991 tcg_gen_qemu_ld_i64(val, addr, index, opc);
992 gen_aa32_frob64(s, val);
993 tcg_temp_free(addr);
994}
995
996static inline void gen_aa32_ld64(DisasContext *s, TCGv_i64 val,
997 TCGv_i32 a32, int index)
998{
999 gen_aa32_ld_i64(s, val, a32, index, MO_Q | s->be_data);
1000}
1001
1002static void gen_aa32_st_i64(DisasContext *s, TCGv_i64 val, TCGv_i32 a32,
1003 int index, MemOp opc)
1004{
1005 TCGv addr = gen_aa32_addr(s, a32, opc);
1006
1007 /* Not needed for user-mode BE32, where we use MO_BE instead. */
1008 if (!IS_USER_ONLY && s->sctlr_b) {
1009 TCGv_i64 tmp = tcg_temp_new_i64();
1010 tcg_gen_rotri_i64(tmp, val, 32);
1011 tcg_gen_qemu_st_i64(tmp, addr, index, opc);
1012 tcg_temp_free_i64(tmp);
1013 } else {
1014 tcg_gen_qemu_st_i64(val, addr, index, opc);
1015 }
1016 tcg_temp_free(addr);
1017}
1018
1019static inline void gen_aa32_st64(DisasContext *s, TCGv_i64 val,
1020 TCGv_i32 a32, int index)
1021{
1022 gen_aa32_st_i64(s, val, a32, index, MO_Q | s->be_data);
1023}
1024
1025DO_GEN_LD(8u, MO_UB)
1026DO_GEN_LD(16u, MO_UW)
1027DO_GEN_LD(32u, MO_UL)
1028DO_GEN_ST(8, MO_UB)
1029DO_GEN_ST(16, MO_UW)
1030DO_GEN_ST(32, MO_UL)
1031
1032static inline void gen_hvc(DisasContext *s, int imm16)
1033{
1034 /* The pre HVC helper handles cases when HVC gets trapped
1035 * as an undefined insn by runtime configuration (ie before
1036 * the insn really executes).
1037 */
1038 gen_set_pc_im(s, s->pc_curr);
1039 gen_helper_pre_hvc(cpu_env);
1040 /* Otherwise we will treat this as a real exception which
1041 * happens after execution of the insn. (The distinction matters
1042 * for the PC value reported to the exception handler and also
1043 * for single stepping.)
1044 */
1045 s->svc_imm = imm16;
1046 gen_set_pc_im(s, s->base.pc_next);
1047 s->base.is_jmp = DISAS_HVC;
1048}
1049
1050static inline void gen_smc(DisasContext *s)
1051{
1052 /* As with HVC, we may take an exception either before or after
1053 * the insn executes.
1054 */
1055 TCGv_i32 tmp;
1056
1057 gen_set_pc_im(s, s->pc_curr);
1058 tmp = tcg_const_i32(syn_aa32_smc());
1059 gen_helper_pre_smc(cpu_env, tmp);
1060 tcg_temp_free_i32(tmp);
1061 gen_set_pc_im(s, s->base.pc_next);
1062 s->base.is_jmp = DISAS_SMC;
1063}
1064
1065static void gen_exception_internal_insn(DisasContext *s, uint32_t pc, int excp)
1066{
1067 gen_set_condexec(s);
1068 gen_set_pc_im(s, pc);
1069 gen_exception_internal(excp);
1070 s->base.is_jmp = DISAS_NORETURN;
1071}
1072
1073static void gen_exception_insn(DisasContext *s, uint32_t pc, int excp,
1074 int syn, uint32_t target_el)
1075{
1076 gen_set_condexec(s);
1077 gen_set_pc_im(s, pc);
1078 gen_exception(excp, syn, target_el);
1079 s->base.is_jmp = DISAS_NORETURN;
1080}
1081
1082static void gen_exception_bkpt_insn(DisasContext *s, uint32_t syn)
1083{
1084 TCGv_i32 tcg_syn;
1085
1086 gen_set_condexec(s);
1087 gen_set_pc_im(s, s->pc_curr);
1088 tcg_syn = tcg_const_i32(syn);
1089 gen_helper_exception_bkpt_insn(cpu_env, tcg_syn);
1090 tcg_temp_free_i32(tcg_syn);
1091 s->base.is_jmp = DISAS_NORETURN;
1092}
1093
1094static void unallocated_encoding(DisasContext *s)
1095{
1096 /* Unallocated and reserved encodings are uncategorized */
1097 gen_exception_insn(s, s->pc_curr, EXCP_UDEF, syn_uncategorized(),
1098 default_exception_el(s));
1099}
1100
1101/* Force a TB lookup after an instruction that changes the CPU state. */
1102static inline void gen_lookup_tb(DisasContext *s)
1103{
1104 tcg_gen_movi_i32(cpu_R[15], s->base.pc_next);
1105 s->base.is_jmp = DISAS_EXIT;
1106}
1107
1108static inline void gen_hlt(DisasContext *s, int imm)
1109{
1110 /* HLT. This has two purposes.
1111 * Architecturally, it is an external halting debug instruction.
1112 * Since QEMU doesn't implement external debug, we treat this as
1113 * it is required for halting debug disabled: it will UNDEF.
1114 * Secondly, "HLT 0x3C" is a T32 semihosting trap instruction,
1115 * and "HLT 0xF000" is an A32 semihosting syscall. These traps
1116 * must trigger semihosting even for ARMv7 and earlier, where
1117 * HLT was an undefined encoding.
1118 * In system mode, we don't allow userspace access to
1119 * semihosting, to provide some semblance of security
1120 * (and for consistency with our 32-bit semihosting).
1121 */
1122 if (semihosting_enabled() &&
1123#ifndef CONFIG_USER_ONLY
1124 s->current_el != 0 &&
1125#endif
1126 (imm == (s->thumb ? 0x3c : 0xf000))) {
1127 gen_exception_internal_insn(s, s->base.pc_next, EXCP_SEMIHOST);
1128 return;
1129 }
1130
1131 unallocated_encoding(s);
1132}
1133
1134static TCGv_ptr get_fpstatus_ptr(int neon)
1135{
1136 TCGv_ptr statusptr = tcg_temp_new_ptr();
1137 int offset;
1138 if (neon) {
1139 offset = offsetof(CPUARMState, vfp.standard_fp_status);
1140 } else {
1141 offset = offsetof(CPUARMState, vfp.fp_status);
1142 }
1143 tcg_gen_addi_ptr(statusptr, cpu_env, offset);
1144 return statusptr;
1145}
1146
1147static inline long vfp_reg_offset(bool dp, unsigned reg)
1148{
1149 if (dp) {
1150 return offsetof(CPUARMState, vfp.zregs[reg >> 1].d[reg & 1]);
1151 } else {
1152 long ofs = offsetof(CPUARMState, vfp.zregs[reg >> 2].d[(reg >> 1) & 1]);
1153 if (reg & 1) {
1154 ofs += offsetof(CPU_DoubleU, l.upper);
1155 } else {
1156 ofs += offsetof(CPU_DoubleU, l.lower);
1157 }
1158 return ofs;
1159 }
1160}
1161
1162/* Return the offset of a 32-bit piece of a NEON register.
1163 zero is the least significant end of the register. */
1164static inline long
1165neon_reg_offset (int reg, int n)
1166{
1167 int sreg;
1168 sreg = reg * 2 + n;
1169 return vfp_reg_offset(0, sreg);
1170}
1171
1172/* Return the offset of a 2**SIZE piece of a NEON register, at index ELE,
1173 * where 0 is the least significant end of the register.
1174 */
1175static inline long
1176neon_element_offset(int reg, int element, MemOp size)
1177{
1178 int element_size = 1 << size;
1179 int ofs = element * element_size;
1180#ifdef HOST_WORDS_BIGENDIAN
1181 /* Calculate the offset assuming fully little-endian,
1182 * then XOR to account for the order of the 8-byte units.
1183 */
1184 if (element_size < 8) {
1185 ofs ^= 8 - element_size;
1186 }
1187#endif
1188 return neon_reg_offset(reg, 0) + ofs;
1189}
1190
1191static TCGv_i32 neon_load_reg(int reg, int pass)
1192{
1193 TCGv_i32 tmp = tcg_temp_new_i32();
1194 tcg_gen_ld_i32(tmp, cpu_env, neon_reg_offset(reg, pass));
1195 return tmp;
1196}
1197
1198static void neon_load_element(TCGv_i32 var, int reg, int ele, MemOp mop)
1199{
1200 long offset = neon_element_offset(reg, ele, mop & MO_SIZE);
1201
1202 switch (mop) {
1203 case MO_UB:
1204 tcg_gen_ld8u_i32(var, cpu_env, offset);
1205 break;
1206 case MO_UW:
1207 tcg_gen_ld16u_i32(var, cpu_env, offset);
1208 break;
1209 case MO_UL:
1210 tcg_gen_ld_i32(var, cpu_env, offset);
1211 break;
1212 default:
1213 g_assert_not_reached();
1214 }
1215}
1216
1217static void neon_load_element64(TCGv_i64 var, int reg, int ele, MemOp mop)
1218{
1219 long offset = neon_element_offset(reg, ele, mop & MO_SIZE);
1220
1221 switch (mop) {
1222 case MO_UB:
1223 tcg_gen_ld8u_i64(var, cpu_env, offset);
1224 break;
1225 case MO_UW:
1226 tcg_gen_ld16u_i64(var, cpu_env, offset);
1227 break;
1228 case MO_UL:
1229 tcg_gen_ld32u_i64(var, cpu_env, offset);
1230 break;
1231 case MO_Q:
1232 tcg_gen_ld_i64(var, cpu_env, offset);
1233 break;
1234 default:
1235 g_assert_not_reached();
1236 }
1237}
1238
1239static void neon_store_reg(int reg, int pass, TCGv_i32 var)
1240{
1241 tcg_gen_st_i32(var, cpu_env, neon_reg_offset(reg, pass));
1242 tcg_temp_free_i32(var);
1243}
1244
1245static void neon_store_element(int reg, int ele, MemOp size, TCGv_i32 var)
1246{
1247 long offset = neon_element_offset(reg, ele, size);
1248
1249 switch (size) {
1250 case MO_8:
1251 tcg_gen_st8_i32(var, cpu_env, offset);
1252 break;
1253 case MO_16:
1254 tcg_gen_st16_i32(var, cpu_env, offset);
1255 break;
1256 case MO_32:
1257 tcg_gen_st_i32(var, cpu_env, offset);
1258 break;
1259 default:
1260 g_assert_not_reached();
1261 }
1262}
1263
1264static void neon_store_element64(int reg, int ele, MemOp size, TCGv_i64 var)
1265{
1266 long offset = neon_element_offset(reg, ele, size);
1267
1268 switch (size) {
1269 case MO_8:
1270 tcg_gen_st8_i64(var, cpu_env, offset);
1271 break;
1272 case MO_16:
1273 tcg_gen_st16_i64(var, cpu_env, offset);
1274 break;
1275 case MO_32:
1276 tcg_gen_st32_i64(var, cpu_env, offset);
1277 break;
1278 case MO_64:
1279 tcg_gen_st_i64(var, cpu_env, offset);
1280 break;
1281 default:
1282 g_assert_not_reached();
1283 }
1284}
1285
1286static inline void neon_load_reg64(TCGv_i64 var, int reg)
1287{
1288 tcg_gen_ld_i64(var, cpu_env, vfp_reg_offset(1, reg));
1289}
1290
1291static inline void neon_store_reg64(TCGv_i64 var, int reg)
1292{
1293 tcg_gen_st_i64(var, cpu_env, vfp_reg_offset(1, reg));
1294}
1295
1296static inline void neon_load_reg32(TCGv_i32 var, int reg)
1297{
1298 tcg_gen_ld_i32(var, cpu_env, vfp_reg_offset(false, reg));
1299}
1300
1301static inline void neon_store_reg32(TCGv_i32 var, int reg)
1302{
1303 tcg_gen_st_i32(var, cpu_env, vfp_reg_offset(false, reg));
1304}
1305
1306static TCGv_ptr vfp_reg_ptr(bool dp, int reg)
1307{
1308 TCGv_ptr ret = tcg_temp_new_ptr();
1309 tcg_gen_addi_ptr(ret, cpu_env, vfp_reg_offset(dp, reg));
1310 return ret;
1311}
1312
1313#define ARM_CP_RW_BIT (1 << 20)
1314
1315/* Include the VFP decoder */
1316#include "translate-vfp.inc.c"
1317
1318static inline void iwmmxt_load_reg(TCGv_i64 var, int reg)
1319{
1320 tcg_gen_ld_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1321}
1322
1323static inline void iwmmxt_store_reg(TCGv_i64 var, int reg)
1324{
1325 tcg_gen_st_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1326}
1327
1328static inline TCGv_i32 iwmmxt_load_creg(int reg)
1329{
1330 TCGv_i32 var = tcg_temp_new_i32();
1331 tcg_gen_ld_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1332 return var;
1333}
1334
1335static inline void iwmmxt_store_creg(int reg, TCGv_i32 var)
1336{
1337 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1338 tcg_temp_free_i32(var);
1339}
1340
1341static inline void gen_op_iwmmxt_movq_wRn_M0(int rn)
1342{
1343 iwmmxt_store_reg(cpu_M0, rn);
1344}
1345
1346static inline void gen_op_iwmmxt_movq_M0_wRn(int rn)
1347{
1348 iwmmxt_load_reg(cpu_M0, rn);
1349}
1350
1351static inline void gen_op_iwmmxt_orq_M0_wRn(int rn)
1352{
1353 iwmmxt_load_reg(cpu_V1, rn);
1354 tcg_gen_or_i64(cpu_M0, cpu_M0, cpu_V1);
1355}
1356
1357static inline void gen_op_iwmmxt_andq_M0_wRn(int rn)
1358{
1359 iwmmxt_load_reg(cpu_V1, rn);
1360 tcg_gen_and_i64(cpu_M0, cpu_M0, cpu_V1);
1361}
1362
1363static inline void gen_op_iwmmxt_xorq_M0_wRn(int rn)
1364{
1365 iwmmxt_load_reg(cpu_V1, rn);
1366 tcg_gen_xor_i64(cpu_M0, cpu_M0, cpu_V1);
1367}
1368
1369#define IWMMXT_OP(name) \
1370static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1371{ \
1372 iwmmxt_load_reg(cpu_V1, rn); \
1373 gen_helper_iwmmxt_##name(cpu_M0, cpu_M0, cpu_V1); \
1374}
1375
1376#define IWMMXT_OP_ENV(name) \
1377static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1378{ \
1379 iwmmxt_load_reg(cpu_V1, rn); \
1380 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0, cpu_V1); \
1381}
1382
1383#define IWMMXT_OP_ENV_SIZE(name) \
1384IWMMXT_OP_ENV(name##b) \
1385IWMMXT_OP_ENV(name##w) \
1386IWMMXT_OP_ENV(name##l)
1387
1388#define IWMMXT_OP_ENV1(name) \
1389static inline void gen_op_iwmmxt_##name##_M0(void) \
1390{ \
1391 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0); \
1392}
1393
1394IWMMXT_OP(maddsq)
1395IWMMXT_OP(madduq)
1396IWMMXT_OP(sadb)
1397IWMMXT_OP(sadw)
1398IWMMXT_OP(mulslw)
1399IWMMXT_OP(mulshw)
1400IWMMXT_OP(mululw)
1401IWMMXT_OP(muluhw)
1402IWMMXT_OP(macsw)
1403IWMMXT_OP(macuw)
1404
1405IWMMXT_OP_ENV_SIZE(unpackl)
1406IWMMXT_OP_ENV_SIZE(unpackh)
1407
1408IWMMXT_OP_ENV1(unpacklub)
1409IWMMXT_OP_ENV1(unpackluw)
1410IWMMXT_OP_ENV1(unpacklul)
1411IWMMXT_OP_ENV1(unpackhub)
1412IWMMXT_OP_ENV1(unpackhuw)
1413IWMMXT_OP_ENV1(unpackhul)
1414IWMMXT_OP_ENV1(unpacklsb)
1415IWMMXT_OP_ENV1(unpacklsw)
1416IWMMXT_OP_ENV1(unpacklsl)
1417IWMMXT_OP_ENV1(unpackhsb)
1418IWMMXT_OP_ENV1(unpackhsw)
1419IWMMXT_OP_ENV1(unpackhsl)
1420
1421IWMMXT_OP_ENV_SIZE(cmpeq)
1422IWMMXT_OP_ENV_SIZE(cmpgtu)
1423IWMMXT_OP_ENV_SIZE(cmpgts)
1424
1425IWMMXT_OP_ENV_SIZE(mins)
1426IWMMXT_OP_ENV_SIZE(minu)
1427IWMMXT_OP_ENV_SIZE(maxs)
1428IWMMXT_OP_ENV_SIZE(maxu)
1429
1430IWMMXT_OP_ENV_SIZE(subn)
1431IWMMXT_OP_ENV_SIZE(addn)
1432IWMMXT_OP_ENV_SIZE(subu)
1433IWMMXT_OP_ENV_SIZE(addu)
1434IWMMXT_OP_ENV_SIZE(subs)
1435IWMMXT_OP_ENV_SIZE(adds)
1436
1437IWMMXT_OP_ENV(avgb0)
1438IWMMXT_OP_ENV(avgb1)
1439IWMMXT_OP_ENV(avgw0)
1440IWMMXT_OP_ENV(avgw1)
1441
1442IWMMXT_OP_ENV(packuw)
1443IWMMXT_OP_ENV(packul)
1444IWMMXT_OP_ENV(packuq)
1445IWMMXT_OP_ENV(packsw)
1446IWMMXT_OP_ENV(packsl)
1447IWMMXT_OP_ENV(packsq)
1448
1449static void gen_op_iwmmxt_set_mup(void)
1450{
1451 TCGv_i32 tmp;
1452 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1453 tcg_gen_ori_i32(tmp, tmp, 2);
1454 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1455}
1456
1457static void gen_op_iwmmxt_set_cup(void)
1458{
1459 TCGv_i32 tmp;
1460 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1461 tcg_gen_ori_i32(tmp, tmp, 1);
1462 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1463}
1464
1465static void gen_op_iwmmxt_setpsr_nz(void)
1466{
1467 TCGv_i32 tmp = tcg_temp_new_i32();
1468 gen_helper_iwmmxt_setpsr_nz(tmp, cpu_M0);
1469 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCASF]);
1470}
1471
1472static inline void gen_op_iwmmxt_addl_M0_wRn(int rn)
1473{
1474 iwmmxt_load_reg(cpu_V1, rn);
1475 tcg_gen_ext32u_i64(cpu_V1, cpu_V1);
1476 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1477}
1478
1479static inline int gen_iwmmxt_address(DisasContext *s, uint32_t insn,
1480 TCGv_i32 dest)
1481{
1482 int rd;
1483 uint32_t offset;
1484 TCGv_i32 tmp;
1485
1486 rd = (insn >> 16) & 0xf;
1487 tmp = load_reg(s, rd);
1488
1489 offset = (insn & 0xff) << ((insn >> 7) & 2);
1490 if (insn & (1 << 24)) {
1491 /* Pre indexed */
1492 if (insn & (1 << 23))
1493 tcg_gen_addi_i32(tmp, tmp, offset);
1494 else
1495 tcg_gen_addi_i32(tmp, tmp, -offset);
1496 tcg_gen_mov_i32(dest, tmp);
1497 if (insn & (1 << 21))
1498 store_reg(s, rd, tmp);
1499 else
1500 tcg_temp_free_i32(tmp);
1501 } else if (insn & (1 << 21)) {
1502 /* Post indexed */
1503 tcg_gen_mov_i32(dest, tmp);
1504 if (insn & (1 << 23))
1505 tcg_gen_addi_i32(tmp, tmp, offset);
1506 else
1507 tcg_gen_addi_i32(tmp, tmp, -offset);
1508 store_reg(s, rd, tmp);
1509 } else if (!(insn & (1 << 23)))
1510 return 1;
1511 return 0;
1512}
1513
1514static inline int gen_iwmmxt_shift(uint32_t insn, uint32_t mask, TCGv_i32 dest)
1515{
1516 int rd = (insn >> 0) & 0xf;
1517 TCGv_i32 tmp;
1518
1519 if (insn & (1 << 8)) {
1520 if (rd < ARM_IWMMXT_wCGR0 || rd > ARM_IWMMXT_wCGR3) {
1521 return 1;
1522 } else {
1523 tmp = iwmmxt_load_creg(rd);
1524 }
1525 } else {
1526 tmp = tcg_temp_new_i32();
1527 iwmmxt_load_reg(cpu_V0, rd);
1528 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
1529 }
1530 tcg_gen_andi_i32(tmp, tmp, mask);
1531 tcg_gen_mov_i32(dest, tmp);
1532 tcg_temp_free_i32(tmp);
1533 return 0;
1534}
1535
1536/* Disassemble an iwMMXt instruction. Returns nonzero if an error occurred
1537 (ie. an undefined instruction). */
1538static int disas_iwmmxt_insn(DisasContext *s, uint32_t insn)
1539{
1540 int rd, wrd;
1541 int rdhi, rdlo, rd0, rd1, i;
1542 TCGv_i32 addr;
1543 TCGv_i32 tmp, tmp2, tmp3;
1544
1545 if ((insn & 0x0e000e00) == 0x0c000000) {
1546 if ((insn & 0x0fe00ff0) == 0x0c400000) {
1547 wrd = insn & 0xf;
1548 rdlo = (insn >> 12) & 0xf;
1549 rdhi = (insn >> 16) & 0xf;
1550 if (insn & ARM_CP_RW_BIT) { /* TMRRC */
1551 iwmmxt_load_reg(cpu_V0, wrd);
1552 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
1553 tcg_gen_extrh_i64_i32(cpu_R[rdhi], cpu_V0);
1554 } else { /* TMCRR */
1555 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
1556 iwmmxt_store_reg(cpu_V0, wrd);
1557 gen_op_iwmmxt_set_mup();
1558 }
1559 return 0;
1560 }
1561
1562 wrd = (insn >> 12) & 0xf;
1563 addr = tcg_temp_new_i32();
1564 if (gen_iwmmxt_address(s, insn, addr)) {
1565 tcg_temp_free_i32(addr);
1566 return 1;
1567 }
1568 if (insn & ARM_CP_RW_BIT) {
1569 if ((insn >> 28) == 0xf) { /* WLDRW wCx */
1570 tmp = tcg_temp_new_i32();
1571 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1572 iwmmxt_store_creg(wrd, tmp);
1573 } else {
1574 i = 1;
1575 if (insn & (1 << 8)) {
1576 if (insn & (1 << 22)) { /* WLDRD */
1577 gen_aa32_ld64(s, cpu_M0, addr, get_mem_index(s));
1578 i = 0;
1579 } else { /* WLDRW wRd */
1580 tmp = tcg_temp_new_i32();
1581 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1582 }
1583 } else {
1584 tmp = tcg_temp_new_i32();
1585 if (insn & (1 << 22)) { /* WLDRH */
1586 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
1587 } else { /* WLDRB */
1588 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
1589 }
1590 }
1591 if (i) {
1592 tcg_gen_extu_i32_i64(cpu_M0, tmp);
1593 tcg_temp_free_i32(tmp);
1594 }
1595 gen_op_iwmmxt_movq_wRn_M0(wrd);
1596 }
1597 } else {
1598 if ((insn >> 28) == 0xf) { /* WSTRW wCx */
1599 tmp = iwmmxt_load_creg(wrd);
1600 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1601 } else {
1602 gen_op_iwmmxt_movq_M0_wRn(wrd);
1603 tmp = tcg_temp_new_i32();
1604 if (insn & (1 << 8)) {
1605 if (insn & (1 << 22)) { /* WSTRD */
1606 gen_aa32_st64(s, cpu_M0, addr, get_mem_index(s));
1607 } else { /* WSTRW wRd */
1608 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1609 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1610 }
1611 } else {
1612 if (insn & (1 << 22)) { /* WSTRH */
1613 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1614 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
1615 } else { /* WSTRB */
1616 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1617 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
1618 }
1619 }
1620 }
1621 tcg_temp_free_i32(tmp);
1622 }
1623 tcg_temp_free_i32(addr);
1624 return 0;
1625 }
1626
1627 if ((insn & 0x0f000000) != 0x0e000000)
1628 return 1;
1629
1630 switch (((insn >> 12) & 0xf00) | ((insn >> 4) & 0xff)) {
1631 case 0x000: /* WOR */
1632 wrd = (insn >> 12) & 0xf;
1633 rd0 = (insn >> 0) & 0xf;
1634 rd1 = (insn >> 16) & 0xf;
1635 gen_op_iwmmxt_movq_M0_wRn(rd0);
1636 gen_op_iwmmxt_orq_M0_wRn(rd1);
1637 gen_op_iwmmxt_setpsr_nz();
1638 gen_op_iwmmxt_movq_wRn_M0(wrd);
1639 gen_op_iwmmxt_set_mup();
1640 gen_op_iwmmxt_set_cup();
1641 break;
1642 case 0x011: /* TMCR */
1643 if (insn & 0xf)
1644 return 1;
1645 rd = (insn >> 12) & 0xf;
1646 wrd = (insn >> 16) & 0xf;
1647 switch (wrd) {
1648 case ARM_IWMMXT_wCID:
1649 case ARM_IWMMXT_wCASF:
1650 break;
1651 case ARM_IWMMXT_wCon:
1652 gen_op_iwmmxt_set_cup();
1653 /* Fall through. */
1654 case ARM_IWMMXT_wCSSF:
1655 tmp = iwmmxt_load_creg(wrd);
1656 tmp2 = load_reg(s, rd);
1657 tcg_gen_andc_i32(tmp, tmp, tmp2);
1658 tcg_temp_free_i32(tmp2);
1659 iwmmxt_store_creg(wrd, tmp);
1660 break;
1661 case ARM_IWMMXT_wCGR0:
1662 case ARM_IWMMXT_wCGR1:
1663 case ARM_IWMMXT_wCGR2:
1664 case ARM_IWMMXT_wCGR3:
1665 gen_op_iwmmxt_set_cup();
1666 tmp = load_reg(s, rd);
1667 iwmmxt_store_creg(wrd, tmp);
1668 break;
1669 default:
1670 return 1;
1671 }
1672 break;
1673 case 0x100: /* WXOR */
1674 wrd = (insn >> 12) & 0xf;
1675 rd0 = (insn >> 0) & 0xf;
1676 rd1 = (insn >> 16) & 0xf;
1677 gen_op_iwmmxt_movq_M0_wRn(rd0);
1678 gen_op_iwmmxt_xorq_M0_wRn(rd1);
1679 gen_op_iwmmxt_setpsr_nz();
1680 gen_op_iwmmxt_movq_wRn_M0(wrd);
1681 gen_op_iwmmxt_set_mup();
1682 gen_op_iwmmxt_set_cup();
1683 break;
1684 case 0x111: /* TMRC */
1685 if (insn & 0xf)
1686 return 1;
1687 rd = (insn >> 12) & 0xf;
1688 wrd = (insn >> 16) & 0xf;
1689 tmp = iwmmxt_load_creg(wrd);
1690 store_reg(s, rd, tmp);
1691 break;
1692 case 0x300: /* WANDN */
1693 wrd = (insn >> 12) & 0xf;
1694 rd0 = (insn >> 0) & 0xf;
1695 rd1 = (insn >> 16) & 0xf;
1696 gen_op_iwmmxt_movq_M0_wRn(rd0);
1697 tcg_gen_neg_i64(cpu_M0, cpu_M0);
1698 gen_op_iwmmxt_andq_M0_wRn(rd1);
1699 gen_op_iwmmxt_setpsr_nz();
1700 gen_op_iwmmxt_movq_wRn_M0(wrd);
1701 gen_op_iwmmxt_set_mup();
1702 gen_op_iwmmxt_set_cup();
1703 break;
1704 case 0x200: /* WAND */
1705 wrd = (insn >> 12) & 0xf;
1706 rd0 = (insn >> 0) & 0xf;
1707 rd1 = (insn >> 16) & 0xf;
1708 gen_op_iwmmxt_movq_M0_wRn(rd0);
1709 gen_op_iwmmxt_andq_M0_wRn(rd1);
1710 gen_op_iwmmxt_setpsr_nz();
1711 gen_op_iwmmxt_movq_wRn_M0(wrd);
1712 gen_op_iwmmxt_set_mup();
1713 gen_op_iwmmxt_set_cup();
1714 break;
1715 case 0x810: case 0xa10: /* WMADD */
1716 wrd = (insn >> 12) & 0xf;
1717 rd0 = (insn >> 0) & 0xf;
1718 rd1 = (insn >> 16) & 0xf;
1719 gen_op_iwmmxt_movq_M0_wRn(rd0);
1720 if (insn & (1 << 21))
1721 gen_op_iwmmxt_maddsq_M0_wRn(rd1);
1722 else
1723 gen_op_iwmmxt_madduq_M0_wRn(rd1);
1724 gen_op_iwmmxt_movq_wRn_M0(wrd);
1725 gen_op_iwmmxt_set_mup();
1726 break;
1727 case 0x10e: case 0x50e: case 0x90e: case 0xd0e: /* WUNPCKIL */
1728 wrd = (insn >> 12) & 0xf;
1729 rd0 = (insn >> 16) & 0xf;
1730 rd1 = (insn >> 0) & 0xf;
1731 gen_op_iwmmxt_movq_M0_wRn(rd0);
1732 switch ((insn >> 22) & 3) {
1733 case 0:
1734 gen_op_iwmmxt_unpacklb_M0_wRn(rd1);
1735 break;
1736 case 1:
1737 gen_op_iwmmxt_unpacklw_M0_wRn(rd1);
1738 break;
1739 case 2:
1740 gen_op_iwmmxt_unpackll_M0_wRn(rd1);
1741 break;
1742 case 3:
1743 return 1;
1744 }
1745 gen_op_iwmmxt_movq_wRn_M0(wrd);
1746 gen_op_iwmmxt_set_mup();
1747 gen_op_iwmmxt_set_cup();
1748 break;
1749 case 0x10c: case 0x50c: case 0x90c: case 0xd0c: /* WUNPCKIH */
1750 wrd = (insn >> 12) & 0xf;
1751 rd0 = (insn >> 16) & 0xf;
1752 rd1 = (insn >> 0) & 0xf;
1753 gen_op_iwmmxt_movq_M0_wRn(rd0);
1754 switch ((insn >> 22) & 3) {
1755 case 0:
1756 gen_op_iwmmxt_unpackhb_M0_wRn(rd1);
1757 break;
1758 case 1:
1759 gen_op_iwmmxt_unpackhw_M0_wRn(rd1);
1760 break;
1761 case 2:
1762 gen_op_iwmmxt_unpackhl_M0_wRn(rd1);
1763 break;
1764 case 3:
1765 return 1;
1766 }
1767 gen_op_iwmmxt_movq_wRn_M0(wrd);
1768 gen_op_iwmmxt_set_mup();
1769 gen_op_iwmmxt_set_cup();
1770 break;
1771 case 0x012: case 0x112: case 0x412: case 0x512: /* WSAD */
1772 wrd = (insn >> 12) & 0xf;
1773 rd0 = (insn >> 16) & 0xf;
1774 rd1 = (insn >> 0) & 0xf;
1775 gen_op_iwmmxt_movq_M0_wRn(rd0);
1776 if (insn & (1 << 22))
1777 gen_op_iwmmxt_sadw_M0_wRn(rd1);
1778 else
1779 gen_op_iwmmxt_sadb_M0_wRn(rd1);
1780 if (!(insn & (1 << 20)))
1781 gen_op_iwmmxt_addl_M0_wRn(wrd);
1782 gen_op_iwmmxt_movq_wRn_M0(wrd);
1783 gen_op_iwmmxt_set_mup();
1784 break;
1785 case 0x010: case 0x110: case 0x210: case 0x310: /* WMUL */
1786 wrd = (insn >> 12) & 0xf;
1787 rd0 = (insn >> 16) & 0xf;
1788 rd1 = (insn >> 0) & 0xf;
1789 gen_op_iwmmxt_movq_M0_wRn(rd0);
1790 if (insn & (1 << 21)) {
1791 if (insn & (1 << 20))
1792 gen_op_iwmmxt_mulshw_M0_wRn(rd1);
1793 else
1794 gen_op_iwmmxt_mulslw_M0_wRn(rd1);
1795 } else {
1796 if (insn & (1 << 20))
1797 gen_op_iwmmxt_muluhw_M0_wRn(rd1);
1798 else
1799 gen_op_iwmmxt_mululw_M0_wRn(rd1);
1800 }
1801 gen_op_iwmmxt_movq_wRn_M0(wrd);
1802 gen_op_iwmmxt_set_mup();
1803 break;
1804 case 0x410: case 0x510: case 0x610: case 0x710: /* WMAC */
1805 wrd = (insn >> 12) & 0xf;
1806 rd0 = (insn >> 16) & 0xf;
1807 rd1 = (insn >> 0) & 0xf;
1808 gen_op_iwmmxt_movq_M0_wRn(rd0);
1809 if (insn & (1 << 21))
1810 gen_op_iwmmxt_macsw_M0_wRn(rd1);
1811 else
1812 gen_op_iwmmxt_macuw_M0_wRn(rd1);
1813 if (!(insn & (1 << 20))) {
1814 iwmmxt_load_reg(cpu_V1, wrd);
1815 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1816 }
1817 gen_op_iwmmxt_movq_wRn_M0(wrd);
1818 gen_op_iwmmxt_set_mup();
1819 break;
1820 case 0x006: case 0x406: case 0x806: case 0xc06: /* WCMPEQ */
1821 wrd = (insn >> 12) & 0xf;
1822 rd0 = (insn >> 16) & 0xf;
1823 rd1 = (insn >> 0) & 0xf;
1824 gen_op_iwmmxt_movq_M0_wRn(rd0);
1825 switch ((insn >> 22) & 3) {
1826 case 0:
1827 gen_op_iwmmxt_cmpeqb_M0_wRn(rd1);
1828 break;
1829 case 1:
1830 gen_op_iwmmxt_cmpeqw_M0_wRn(rd1);
1831 break;
1832 case 2:
1833 gen_op_iwmmxt_cmpeql_M0_wRn(rd1);
1834 break;
1835 case 3:
1836 return 1;
1837 }
1838 gen_op_iwmmxt_movq_wRn_M0(wrd);
1839 gen_op_iwmmxt_set_mup();
1840 gen_op_iwmmxt_set_cup();
1841 break;
1842 case 0x800: case 0x900: case 0xc00: case 0xd00: /* WAVG2 */
1843 wrd = (insn >> 12) & 0xf;
1844 rd0 = (insn >> 16) & 0xf;
1845 rd1 = (insn >> 0) & 0xf;
1846 gen_op_iwmmxt_movq_M0_wRn(rd0);
1847 if (insn & (1 << 22)) {
1848 if (insn & (1 << 20))
1849 gen_op_iwmmxt_avgw1_M0_wRn(rd1);
1850 else
1851 gen_op_iwmmxt_avgw0_M0_wRn(rd1);
1852 } else {
1853 if (insn & (1 << 20))
1854 gen_op_iwmmxt_avgb1_M0_wRn(rd1);
1855 else
1856 gen_op_iwmmxt_avgb0_M0_wRn(rd1);
1857 }
1858 gen_op_iwmmxt_movq_wRn_M0(wrd);
1859 gen_op_iwmmxt_set_mup();
1860 gen_op_iwmmxt_set_cup();
1861 break;
1862 case 0x802: case 0x902: case 0xa02: case 0xb02: /* WALIGNR */
1863 wrd = (insn >> 12) & 0xf;
1864 rd0 = (insn >> 16) & 0xf;
1865 rd1 = (insn >> 0) & 0xf;
1866 gen_op_iwmmxt_movq_M0_wRn(rd0);
1867 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCGR0 + ((insn >> 20) & 3));
1868 tcg_gen_andi_i32(tmp, tmp, 7);
1869 iwmmxt_load_reg(cpu_V1, rd1);
1870 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
1871 tcg_temp_free_i32(tmp);
1872 gen_op_iwmmxt_movq_wRn_M0(wrd);
1873 gen_op_iwmmxt_set_mup();
1874 break;
1875 case 0x601: case 0x605: case 0x609: case 0x60d: /* TINSR */
1876 if (((insn >> 6) & 3) == 3)
1877 return 1;
1878 rd = (insn >> 12) & 0xf;
1879 wrd = (insn >> 16) & 0xf;
1880 tmp = load_reg(s, rd);
1881 gen_op_iwmmxt_movq_M0_wRn(wrd);
1882 switch ((insn >> 6) & 3) {
1883 case 0:
1884 tmp2 = tcg_const_i32(0xff);
1885 tmp3 = tcg_const_i32((insn & 7) << 3);
1886 break;
1887 case 1:
1888 tmp2 = tcg_const_i32(0xffff);
1889 tmp3 = tcg_const_i32((insn & 3) << 4);
1890 break;
1891 case 2:
1892 tmp2 = tcg_const_i32(0xffffffff);
1893 tmp3 = tcg_const_i32((insn & 1) << 5);
1894 break;
1895 default:
1896 tmp2 = NULL;
1897 tmp3 = NULL;
1898 }
1899 gen_helper_iwmmxt_insr(cpu_M0, cpu_M0, tmp, tmp2, tmp3);
1900 tcg_temp_free_i32(tmp3);
1901 tcg_temp_free_i32(tmp2);
1902 tcg_temp_free_i32(tmp);
1903 gen_op_iwmmxt_movq_wRn_M0(wrd);
1904 gen_op_iwmmxt_set_mup();
1905 break;
1906 case 0x107: case 0x507: case 0x907: case 0xd07: /* TEXTRM */
1907 rd = (insn >> 12) & 0xf;
1908 wrd = (insn >> 16) & 0xf;
1909 if (rd == 15 || ((insn >> 22) & 3) == 3)
1910 return 1;
1911 gen_op_iwmmxt_movq_M0_wRn(wrd);
1912 tmp = tcg_temp_new_i32();
1913 switch ((insn >> 22) & 3) {
1914 case 0:
1915 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 7) << 3);
1916 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1917 if (insn & 8) {
1918 tcg_gen_ext8s_i32(tmp, tmp);
1919 } else {
1920 tcg_gen_andi_i32(tmp, tmp, 0xff);
1921 }
1922 break;
1923 case 1:
1924 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 3) << 4);
1925 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1926 if (insn & 8) {
1927 tcg_gen_ext16s_i32(tmp, tmp);
1928 } else {
1929 tcg_gen_andi_i32(tmp, tmp, 0xffff);
1930 }
1931 break;
1932 case 2:
1933 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 1) << 5);
1934 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1935 break;
1936 }
1937 store_reg(s, rd, tmp);
1938 break;
1939 case 0x117: case 0x517: case 0x917: case 0xd17: /* TEXTRC */
1940 if ((insn & 0x000ff008) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1941 return 1;
1942 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1943 switch ((insn >> 22) & 3) {
1944 case 0:
1945 tcg_gen_shri_i32(tmp, tmp, ((insn & 7) << 2) + 0);
1946 break;
1947 case 1:
1948 tcg_gen_shri_i32(tmp, tmp, ((insn & 3) << 3) + 4);
1949 break;
1950 case 2:
1951 tcg_gen_shri_i32(tmp, tmp, ((insn & 1) << 4) + 12);
1952 break;
1953 }
1954 tcg_gen_shli_i32(tmp, tmp, 28);
1955 gen_set_nzcv(tmp);
1956 tcg_temp_free_i32(tmp);
1957 break;
1958 case 0x401: case 0x405: case 0x409: case 0x40d: /* TBCST */
1959 if (((insn >> 6) & 3) == 3)
1960 return 1;
1961 rd = (insn >> 12) & 0xf;
1962 wrd = (insn >> 16) & 0xf;
1963 tmp = load_reg(s, rd);
1964 switch ((insn >> 6) & 3) {
1965 case 0:
1966 gen_helper_iwmmxt_bcstb(cpu_M0, tmp);
1967 break;
1968 case 1:
1969 gen_helper_iwmmxt_bcstw(cpu_M0, tmp);
1970 break;
1971 case 2:
1972 gen_helper_iwmmxt_bcstl(cpu_M0, tmp);
1973 break;
1974 }
1975 tcg_temp_free_i32(tmp);
1976 gen_op_iwmmxt_movq_wRn_M0(wrd);
1977 gen_op_iwmmxt_set_mup();
1978 break;
1979 case 0x113: case 0x513: case 0x913: case 0xd13: /* TANDC */
1980 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1981 return 1;
1982 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1983 tmp2 = tcg_temp_new_i32();
1984 tcg_gen_mov_i32(tmp2, tmp);
1985 switch ((insn >> 22) & 3) {
1986 case 0:
1987 for (i = 0; i < 7; i ++) {
1988 tcg_gen_shli_i32(tmp2, tmp2, 4);
1989 tcg_gen_and_i32(tmp, tmp, tmp2);
1990 }
1991 break;
1992 case 1:
1993 for (i = 0; i < 3; i ++) {
1994 tcg_gen_shli_i32(tmp2, tmp2, 8);
1995 tcg_gen_and_i32(tmp, tmp, tmp2);
1996 }
1997 break;
1998 case 2:
1999 tcg_gen_shli_i32(tmp2, tmp2, 16);
2000 tcg_gen_and_i32(tmp, tmp, tmp2);
2001 break;
2002 }
2003 gen_set_nzcv(tmp);
2004 tcg_temp_free_i32(tmp2);
2005 tcg_temp_free_i32(tmp);
2006 break;
2007 case 0x01c: case 0x41c: case 0x81c: case 0xc1c: /* WACC */
2008 wrd = (insn >> 12) & 0xf;
2009 rd0 = (insn >> 16) & 0xf;
2010 gen_op_iwmmxt_movq_M0_wRn(rd0);
2011 switch ((insn >> 22) & 3) {
2012 case 0:
2013 gen_helper_iwmmxt_addcb(cpu_M0, cpu_M0);
2014 break;
2015 case 1:
2016 gen_helper_iwmmxt_addcw(cpu_M0, cpu_M0);
2017 break;
2018 case 2:
2019 gen_helper_iwmmxt_addcl(cpu_M0, cpu_M0);
2020 break;
2021 case 3:
2022 return 1;
2023 }
2024 gen_op_iwmmxt_movq_wRn_M0(wrd);
2025 gen_op_iwmmxt_set_mup();
2026 break;
2027 case 0x115: case 0x515: case 0x915: case 0xd15: /* TORC */
2028 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
2029 return 1;
2030 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
2031 tmp2 = tcg_temp_new_i32();
2032 tcg_gen_mov_i32(tmp2, tmp);
2033 switch ((insn >> 22) & 3) {
2034 case 0:
2035 for (i = 0; i < 7; i ++) {
2036 tcg_gen_shli_i32(tmp2, tmp2, 4);
2037 tcg_gen_or_i32(tmp, tmp, tmp2);
2038 }
2039 break;
2040 case 1:
2041 for (i = 0; i < 3; i ++) {
2042 tcg_gen_shli_i32(tmp2, tmp2, 8);
2043 tcg_gen_or_i32(tmp, tmp, tmp2);
2044 }
2045 break;
2046 case 2:
2047 tcg_gen_shli_i32(tmp2, tmp2, 16);
2048 tcg_gen_or_i32(tmp, tmp, tmp2);
2049 break;
2050 }
2051 gen_set_nzcv(tmp);
2052 tcg_temp_free_i32(tmp2);
2053 tcg_temp_free_i32(tmp);
2054 break;
2055 case 0x103: case 0x503: case 0x903: case 0xd03: /* TMOVMSK */
2056 rd = (insn >> 12) & 0xf;
2057 rd0 = (insn >> 16) & 0xf;
2058 if ((insn & 0xf) != 0 || ((insn >> 22) & 3) == 3)
2059 return 1;
2060 gen_op_iwmmxt_movq_M0_wRn(rd0);
2061 tmp = tcg_temp_new_i32();
2062 switch ((insn >> 22) & 3) {
2063 case 0:
2064 gen_helper_iwmmxt_msbb(tmp, cpu_M0);
2065 break;
2066 case 1:
2067 gen_helper_iwmmxt_msbw(tmp, cpu_M0);
2068 break;
2069 case 2:
2070 gen_helper_iwmmxt_msbl(tmp, cpu_M0);
2071 break;
2072 }
2073 store_reg(s, rd, tmp);
2074 break;
2075 case 0x106: case 0x306: case 0x506: case 0x706: /* WCMPGT */
2076 case 0x906: case 0xb06: case 0xd06: case 0xf06:
2077 wrd = (insn >> 12) & 0xf;
2078 rd0 = (insn >> 16) & 0xf;
2079 rd1 = (insn >> 0) & 0xf;
2080 gen_op_iwmmxt_movq_M0_wRn(rd0);
2081 switch ((insn >> 22) & 3) {
2082 case 0:
2083 if (insn & (1 << 21))
2084 gen_op_iwmmxt_cmpgtsb_M0_wRn(rd1);
2085 else
2086 gen_op_iwmmxt_cmpgtub_M0_wRn(rd1);
2087 break;
2088 case 1:
2089 if (insn & (1 << 21))
2090 gen_op_iwmmxt_cmpgtsw_M0_wRn(rd1);
2091 else
2092 gen_op_iwmmxt_cmpgtuw_M0_wRn(rd1);
2093 break;
2094 case 2:
2095 if (insn & (1 << 21))
2096 gen_op_iwmmxt_cmpgtsl_M0_wRn(rd1);
2097 else
2098 gen_op_iwmmxt_cmpgtul_M0_wRn(rd1);
2099 break;
2100 case 3:
2101 return 1;
2102 }
2103 gen_op_iwmmxt_movq_wRn_M0(wrd);
2104 gen_op_iwmmxt_set_mup();
2105 gen_op_iwmmxt_set_cup();
2106 break;
2107 case 0x00e: case 0x20e: case 0x40e: case 0x60e: /* WUNPCKEL */
2108 case 0x80e: case 0xa0e: case 0xc0e: case 0xe0e:
2109 wrd = (insn >> 12) & 0xf;
2110 rd0 = (insn >> 16) & 0xf;
2111 gen_op_iwmmxt_movq_M0_wRn(rd0);
2112 switch ((insn >> 22) & 3) {
2113 case 0:
2114 if (insn & (1 << 21))
2115 gen_op_iwmmxt_unpacklsb_M0();
2116 else
2117 gen_op_iwmmxt_unpacklub_M0();
2118 break;
2119 case 1:
2120 if (insn & (1 << 21))
2121 gen_op_iwmmxt_unpacklsw_M0();
2122 else
2123 gen_op_iwmmxt_unpackluw_M0();
2124 break;
2125 case 2:
2126 if (insn & (1 << 21))
2127 gen_op_iwmmxt_unpacklsl_M0();
2128 else
2129 gen_op_iwmmxt_unpacklul_M0();
2130 break;
2131 case 3:
2132 return 1;
2133 }
2134 gen_op_iwmmxt_movq_wRn_M0(wrd);
2135 gen_op_iwmmxt_set_mup();
2136 gen_op_iwmmxt_set_cup();
2137 break;
2138 case 0x00c: case 0x20c: case 0x40c: case 0x60c: /* WUNPCKEH */
2139 case 0x80c: case 0xa0c: case 0xc0c: case 0xe0c:
2140 wrd = (insn >> 12) & 0xf;
2141 rd0 = (insn >> 16) & 0xf;
2142 gen_op_iwmmxt_movq_M0_wRn(rd0);
2143 switch ((insn >> 22) & 3) {
2144 case 0:
2145 if (insn & (1 << 21))
2146 gen_op_iwmmxt_unpackhsb_M0();
2147 else
2148 gen_op_iwmmxt_unpackhub_M0();
2149 break;
2150 case 1:
2151 if (insn & (1 << 21))
2152 gen_op_iwmmxt_unpackhsw_M0();
2153 else
2154 gen_op_iwmmxt_unpackhuw_M0();
2155 break;
2156 case 2:
2157 if (insn & (1 << 21))
2158 gen_op_iwmmxt_unpackhsl_M0();
2159 else
2160 gen_op_iwmmxt_unpackhul_M0();
2161 break;
2162 case 3:
2163 return 1;
2164 }
2165 gen_op_iwmmxt_movq_wRn_M0(wrd);
2166 gen_op_iwmmxt_set_mup();
2167 gen_op_iwmmxt_set_cup();
2168 break;
2169 case 0x204: case 0x604: case 0xa04: case 0xe04: /* WSRL */
2170 case 0x214: case 0x614: case 0xa14: case 0xe14:
2171 if (((insn >> 22) & 3) == 0)
2172 return 1;
2173 wrd = (insn >> 12) & 0xf;
2174 rd0 = (insn >> 16) & 0xf;
2175 gen_op_iwmmxt_movq_M0_wRn(rd0);
2176 tmp = tcg_temp_new_i32();
2177 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2178 tcg_temp_free_i32(tmp);
2179 return 1;
2180 }
2181 switch ((insn >> 22) & 3) {
2182 case 1:
2183 gen_helper_iwmmxt_srlw(cpu_M0, cpu_env, cpu_M0, tmp);
2184 break;
2185 case 2:
2186 gen_helper_iwmmxt_srll(cpu_M0, cpu_env, cpu_M0, tmp);
2187 break;
2188 case 3:
2189 gen_helper_iwmmxt_srlq(cpu_M0, cpu_env, cpu_M0, tmp);
2190 break;
2191 }
2192 tcg_temp_free_i32(tmp);
2193 gen_op_iwmmxt_movq_wRn_M0(wrd);
2194 gen_op_iwmmxt_set_mup();
2195 gen_op_iwmmxt_set_cup();
2196 break;
2197 case 0x004: case 0x404: case 0x804: case 0xc04: /* WSRA */
2198 case 0x014: case 0x414: case 0x814: case 0xc14:
2199 if (((insn >> 22) & 3) == 0)
2200 return 1;
2201 wrd = (insn >> 12) & 0xf;
2202 rd0 = (insn >> 16) & 0xf;
2203 gen_op_iwmmxt_movq_M0_wRn(rd0);
2204 tmp = tcg_temp_new_i32();
2205 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2206 tcg_temp_free_i32(tmp);
2207 return 1;
2208 }
2209 switch ((insn >> 22) & 3) {
2210 case 1:
2211 gen_helper_iwmmxt_sraw(cpu_M0, cpu_env, cpu_M0, tmp);
2212 break;
2213 case 2:
2214 gen_helper_iwmmxt_sral(cpu_M0, cpu_env, cpu_M0, tmp);
2215 break;
2216 case 3:
2217 gen_helper_iwmmxt_sraq(cpu_M0, cpu_env, cpu_M0, tmp);
2218 break;
2219 }
2220 tcg_temp_free_i32(tmp);
2221 gen_op_iwmmxt_movq_wRn_M0(wrd);
2222 gen_op_iwmmxt_set_mup();
2223 gen_op_iwmmxt_set_cup();
2224 break;
2225 case 0x104: case 0x504: case 0x904: case 0xd04: /* WSLL */
2226 case 0x114: case 0x514: case 0x914: case 0xd14:
2227 if (((insn >> 22) & 3) == 0)
2228 return 1;
2229 wrd = (insn >> 12) & 0xf;
2230 rd0 = (insn >> 16) & 0xf;
2231 gen_op_iwmmxt_movq_M0_wRn(rd0);
2232 tmp = tcg_temp_new_i32();
2233 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2234 tcg_temp_free_i32(tmp);
2235 return 1;
2236 }
2237 switch ((insn >> 22) & 3) {
2238 case 1:
2239 gen_helper_iwmmxt_sllw(cpu_M0, cpu_env, cpu_M0, tmp);
2240 break;
2241 case 2:
2242 gen_helper_iwmmxt_slll(cpu_M0, cpu_env, cpu_M0, tmp);
2243 break;
2244 case 3:
2245 gen_helper_iwmmxt_sllq(cpu_M0, cpu_env, cpu_M0, tmp);
2246 break;
2247 }
2248 tcg_temp_free_i32(tmp);
2249 gen_op_iwmmxt_movq_wRn_M0(wrd);
2250 gen_op_iwmmxt_set_mup();
2251 gen_op_iwmmxt_set_cup();
2252 break;
2253 case 0x304: case 0x704: case 0xb04: case 0xf04: /* WROR */
2254 case 0x314: case 0x714: case 0xb14: case 0xf14:
2255 if (((insn >> 22) & 3) == 0)
2256 return 1;
2257 wrd = (insn >> 12) & 0xf;
2258 rd0 = (insn >> 16) & 0xf;
2259 gen_op_iwmmxt_movq_M0_wRn(rd0);
2260 tmp = tcg_temp_new_i32();
2261 switch ((insn >> 22) & 3) {
2262 case 1:
2263 if (gen_iwmmxt_shift(insn, 0xf, tmp)) {
2264 tcg_temp_free_i32(tmp);
2265 return 1;
2266 }
2267 gen_helper_iwmmxt_rorw(cpu_M0, cpu_env, cpu_M0, tmp);
2268 break;
2269 case 2:
2270 if (gen_iwmmxt_shift(insn, 0x1f, tmp)) {
2271 tcg_temp_free_i32(tmp);
2272 return 1;
2273 }
2274 gen_helper_iwmmxt_rorl(cpu_M0, cpu_env, cpu_M0, tmp);
2275 break;
2276 case 3:
2277 if (gen_iwmmxt_shift(insn, 0x3f, tmp)) {
2278 tcg_temp_free_i32(tmp);
2279 return 1;
2280 }
2281 gen_helper_iwmmxt_rorq(cpu_M0, cpu_env, cpu_M0, tmp);
2282 break;
2283 }
2284 tcg_temp_free_i32(tmp);
2285 gen_op_iwmmxt_movq_wRn_M0(wrd);
2286 gen_op_iwmmxt_set_mup();
2287 gen_op_iwmmxt_set_cup();
2288 break;
2289 case 0x116: case 0x316: case 0x516: case 0x716: /* WMIN */
2290 case 0x916: case 0xb16: case 0xd16: case 0xf16:
2291 wrd = (insn >> 12) & 0xf;
2292 rd0 = (insn >> 16) & 0xf;
2293 rd1 = (insn >> 0) & 0xf;
2294 gen_op_iwmmxt_movq_M0_wRn(rd0);
2295 switch ((insn >> 22) & 3) {
2296 case 0:
2297 if (insn & (1 << 21))
2298 gen_op_iwmmxt_minsb_M0_wRn(rd1);
2299 else
2300 gen_op_iwmmxt_minub_M0_wRn(rd1);
2301 break;
2302 case 1:
2303 if (insn & (1 << 21))
2304 gen_op_iwmmxt_minsw_M0_wRn(rd1);
2305 else
2306 gen_op_iwmmxt_minuw_M0_wRn(rd1);
2307 break;
2308 case 2:
2309 if (insn & (1 << 21))
2310 gen_op_iwmmxt_minsl_M0_wRn(rd1);
2311 else
2312 gen_op_iwmmxt_minul_M0_wRn(rd1);
2313 break;
2314 case 3:
2315 return 1;
2316 }
2317 gen_op_iwmmxt_movq_wRn_M0(wrd);
2318 gen_op_iwmmxt_set_mup();
2319 break;
2320 case 0x016: case 0x216: case 0x416: case 0x616: /* WMAX */
2321 case 0x816: case 0xa16: case 0xc16: case 0xe16:
2322 wrd = (insn >> 12) & 0xf;
2323 rd0 = (insn >> 16) & 0xf;
2324 rd1 = (insn >> 0) & 0xf;
2325 gen_op_iwmmxt_movq_M0_wRn(rd0);
2326 switch ((insn >> 22) & 3) {
2327 case 0:
2328 if (insn & (1 << 21))
2329 gen_op_iwmmxt_maxsb_M0_wRn(rd1);
2330 else
2331 gen_op_iwmmxt_maxub_M0_wRn(rd1);
2332 break;
2333 case 1:
2334 if (insn & (1 << 21))
2335 gen_op_iwmmxt_maxsw_M0_wRn(rd1);
2336 else
2337 gen_op_iwmmxt_maxuw_M0_wRn(rd1);
2338 break;
2339 case 2:
2340 if (insn & (1 << 21))
2341 gen_op_iwmmxt_maxsl_M0_wRn(rd1);
2342 else
2343 gen_op_iwmmxt_maxul_M0_wRn(rd1);
2344 break;
2345 case 3:
2346 return 1;
2347 }
2348 gen_op_iwmmxt_movq_wRn_M0(wrd);
2349 gen_op_iwmmxt_set_mup();
2350 break;
2351 case 0x002: case 0x102: case 0x202: case 0x302: /* WALIGNI */
2352 case 0x402: case 0x502: case 0x602: case 0x702:
2353 wrd = (insn >> 12) & 0xf;
2354 rd0 = (insn >> 16) & 0xf;
2355 rd1 = (insn >> 0) & 0xf;
2356 gen_op_iwmmxt_movq_M0_wRn(rd0);
2357 tmp = tcg_const_i32((insn >> 20) & 3);
2358 iwmmxt_load_reg(cpu_V1, rd1);
2359 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
2360 tcg_temp_free_i32(tmp);
2361 gen_op_iwmmxt_movq_wRn_M0(wrd);
2362 gen_op_iwmmxt_set_mup();
2363 break;
2364 case 0x01a: case 0x11a: case 0x21a: case 0x31a: /* WSUB */
2365 case 0x41a: case 0x51a: case 0x61a: case 0x71a:
2366 case 0x81a: case 0x91a: case 0xa1a: case 0xb1a:
2367 case 0xc1a: case 0xd1a: case 0xe1a: case 0xf1a:
2368 wrd = (insn >> 12) & 0xf;
2369 rd0 = (insn >> 16) & 0xf;
2370 rd1 = (insn >> 0) & 0xf;
2371 gen_op_iwmmxt_movq_M0_wRn(rd0);
2372 switch ((insn >> 20) & 0xf) {
2373 case 0x0:
2374 gen_op_iwmmxt_subnb_M0_wRn(rd1);
2375 break;
2376 case 0x1:
2377 gen_op_iwmmxt_subub_M0_wRn(rd1);
2378 break;
2379 case 0x3:
2380 gen_op_iwmmxt_subsb_M0_wRn(rd1);
2381 break;
2382 case 0x4:
2383 gen_op_iwmmxt_subnw_M0_wRn(rd1);
2384 break;
2385 case 0x5:
2386 gen_op_iwmmxt_subuw_M0_wRn(rd1);
2387 break;
2388 case 0x7:
2389 gen_op_iwmmxt_subsw_M0_wRn(rd1);
2390 break;
2391 case 0x8:
2392 gen_op_iwmmxt_subnl_M0_wRn(rd1);
2393 break;
2394 case 0x9:
2395 gen_op_iwmmxt_subul_M0_wRn(rd1);
2396 break;
2397 case 0xb:
2398 gen_op_iwmmxt_subsl_M0_wRn(rd1);
2399 break;
2400 default:
2401 return 1;
2402 }
2403 gen_op_iwmmxt_movq_wRn_M0(wrd);
2404 gen_op_iwmmxt_set_mup();
2405 gen_op_iwmmxt_set_cup();
2406 break;
2407 case 0x01e: case 0x11e: case 0x21e: case 0x31e: /* WSHUFH */
2408 case 0x41e: case 0x51e: case 0x61e: case 0x71e:
2409 case 0x81e: case 0x91e: case 0xa1e: case 0xb1e:
2410 case 0xc1e: case 0xd1e: case 0xe1e: case 0xf1e:
2411 wrd = (insn >> 12) & 0xf;
2412 rd0 = (insn >> 16) & 0xf;
2413 gen_op_iwmmxt_movq_M0_wRn(rd0);
2414 tmp = tcg_const_i32(((insn >> 16) & 0xf0) | (insn & 0x0f));
2415 gen_helper_iwmmxt_shufh(cpu_M0, cpu_env, cpu_M0, tmp);
2416 tcg_temp_free_i32(tmp);
2417 gen_op_iwmmxt_movq_wRn_M0(wrd);
2418 gen_op_iwmmxt_set_mup();
2419 gen_op_iwmmxt_set_cup();
2420 break;
2421 case 0x018: case 0x118: case 0x218: case 0x318: /* WADD */
2422 case 0x418: case 0x518: case 0x618: case 0x718:
2423 case 0x818: case 0x918: case 0xa18: case 0xb18:
2424 case 0xc18: case 0xd18: case 0xe18: case 0xf18:
2425 wrd = (insn >> 12) & 0xf;
2426 rd0 = (insn >> 16) & 0xf;
2427 rd1 = (insn >> 0) & 0xf;
2428 gen_op_iwmmxt_movq_M0_wRn(rd0);
2429 switch ((insn >> 20) & 0xf) {
2430 case 0x0:
2431 gen_op_iwmmxt_addnb_M0_wRn(rd1);
2432 break;
2433 case 0x1:
2434 gen_op_iwmmxt_addub_M0_wRn(rd1);
2435 break;
2436 case 0x3:
2437 gen_op_iwmmxt_addsb_M0_wRn(rd1);
2438 break;
2439 case 0x4:
2440 gen_op_iwmmxt_addnw_M0_wRn(rd1);
2441 break;
2442 case 0x5:
2443 gen_op_iwmmxt_adduw_M0_wRn(rd1);
2444 break;
2445 case 0x7:
2446 gen_op_iwmmxt_addsw_M0_wRn(rd1);
2447 break;
2448 case 0x8:
2449 gen_op_iwmmxt_addnl_M0_wRn(rd1);
2450 break;
2451 case 0x9:
2452 gen_op_iwmmxt_addul_M0_wRn(rd1);
2453 break;
2454 case 0xb:
2455 gen_op_iwmmxt_addsl_M0_wRn(rd1);
2456 break;
2457 default:
2458 return 1;
2459 }
2460 gen_op_iwmmxt_movq_wRn_M0(wrd);
2461 gen_op_iwmmxt_set_mup();
2462 gen_op_iwmmxt_set_cup();
2463 break;
2464 case 0x008: case 0x108: case 0x208: case 0x308: /* WPACK */
2465 case 0x408: case 0x508: case 0x608: case 0x708:
2466 case 0x808: case 0x908: case 0xa08: case 0xb08:
2467 case 0xc08: case 0xd08: case 0xe08: case 0xf08:
2468 if (!(insn & (1 << 20)) || ((insn >> 22) & 3) == 0)
2469 return 1;
2470 wrd = (insn >> 12) & 0xf;
2471 rd0 = (insn >> 16) & 0xf;
2472 rd1 = (insn >> 0) & 0xf;
2473 gen_op_iwmmxt_movq_M0_wRn(rd0);
2474 switch ((insn >> 22) & 3) {
2475 case 1:
2476 if (insn & (1 << 21))
2477 gen_op_iwmmxt_packsw_M0_wRn(rd1);
2478 else
2479 gen_op_iwmmxt_packuw_M0_wRn(rd1);
2480 break;
2481 case 2:
2482 if (insn & (1 << 21))
2483 gen_op_iwmmxt_packsl_M0_wRn(rd1);
2484 else
2485 gen_op_iwmmxt_packul_M0_wRn(rd1);
2486 break;
2487 case 3:
2488 if (insn & (1 << 21))
2489 gen_op_iwmmxt_packsq_M0_wRn(rd1);
2490 else
2491 gen_op_iwmmxt_packuq_M0_wRn(rd1);
2492 break;
2493 }
2494 gen_op_iwmmxt_movq_wRn_M0(wrd);
2495 gen_op_iwmmxt_set_mup();
2496 gen_op_iwmmxt_set_cup();
2497 break;
2498 case 0x201: case 0x203: case 0x205: case 0x207:
2499 case 0x209: case 0x20b: case 0x20d: case 0x20f:
2500 case 0x211: case 0x213: case 0x215: case 0x217:
2501 case 0x219: case 0x21b: case 0x21d: case 0x21f:
2502 wrd = (insn >> 5) & 0xf;
2503 rd0 = (insn >> 12) & 0xf;
2504 rd1 = (insn >> 0) & 0xf;
2505 if (rd0 == 0xf || rd1 == 0xf)
2506 return 1;
2507 gen_op_iwmmxt_movq_M0_wRn(wrd);
2508 tmp = load_reg(s, rd0);
2509 tmp2 = load_reg(s, rd1);
2510 switch ((insn >> 16) & 0xf) {
2511 case 0x0: /* TMIA */
2512 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2513 break;
2514 case 0x8: /* TMIAPH */
2515 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2516 break;
2517 case 0xc: case 0xd: case 0xe: case 0xf: /* TMIAxy */
2518 if (insn & (1 << 16))
2519 tcg_gen_shri_i32(tmp, tmp, 16);
2520 if (insn & (1 << 17))
2521 tcg_gen_shri_i32(tmp2, tmp2, 16);
2522 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2523 break;
2524 default:
2525 tcg_temp_free_i32(tmp2);
2526 tcg_temp_free_i32(tmp);
2527 return 1;
2528 }
2529 tcg_temp_free_i32(tmp2);
2530 tcg_temp_free_i32(tmp);
2531 gen_op_iwmmxt_movq_wRn_M0(wrd);
2532 gen_op_iwmmxt_set_mup();
2533 break;
2534 default:
2535 return 1;
2536 }
2537
2538 return 0;
2539}
2540
2541/* Disassemble an XScale DSP instruction. Returns nonzero if an error occurred
2542 (ie. an undefined instruction). */
2543static int disas_dsp_insn(DisasContext *s, uint32_t insn)
2544{
2545 int acc, rd0, rd1, rdhi, rdlo;
2546 TCGv_i32 tmp, tmp2;
2547
2548 if ((insn & 0x0ff00f10) == 0x0e200010) {
2549 /* Multiply with Internal Accumulate Format */
2550 rd0 = (insn >> 12) & 0xf;
2551 rd1 = insn & 0xf;
2552 acc = (insn >> 5) & 7;
2553
2554 if (acc != 0)
2555 return 1;
2556
2557 tmp = load_reg(s, rd0);
2558 tmp2 = load_reg(s, rd1);
2559 switch ((insn >> 16) & 0xf) {
2560 case 0x0: /* MIA */
2561 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2562 break;
2563 case 0x8: /* MIAPH */
2564 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2565 break;
2566 case 0xc: /* MIABB */
2567 case 0xd: /* MIABT */
2568 case 0xe: /* MIATB */
2569 case 0xf: /* MIATT */
2570 if (insn & (1 << 16))
2571 tcg_gen_shri_i32(tmp, tmp, 16);
2572 if (insn & (1 << 17))
2573 tcg_gen_shri_i32(tmp2, tmp2, 16);
2574 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2575 break;
2576 default:
2577 return 1;
2578 }
2579 tcg_temp_free_i32(tmp2);
2580 tcg_temp_free_i32(tmp);
2581
2582 gen_op_iwmmxt_movq_wRn_M0(acc);
2583 return 0;
2584 }
2585
2586 if ((insn & 0x0fe00ff8) == 0x0c400000) {
2587 /* Internal Accumulator Access Format */
2588 rdhi = (insn >> 16) & 0xf;
2589 rdlo = (insn >> 12) & 0xf;
2590 acc = insn & 7;
2591
2592 if (acc != 0)
2593 return 1;
2594
2595 if (insn & ARM_CP_RW_BIT) { /* MRA */
2596 iwmmxt_load_reg(cpu_V0, acc);
2597 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
2598 tcg_gen_extrh_i64_i32(cpu_R[rdhi], cpu_V0);
2599 tcg_gen_andi_i32(cpu_R[rdhi], cpu_R[rdhi], (1 << (40 - 32)) - 1);
2600 } else { /* MAR */
2601 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
2602 iwmmxt_store_reg(cpu_V0, acc);
2603 }
2604 return 0;
2605 }
2606
2607 return 1;
2608}
2609
2610#define VFP_REG_SHR(x, n) (((n) > 0) ? (x) >> (n) : (x) << -(n))
2611#define VFP_SREG(insn, bigbit, smallbit) \
2612 ((VFP_REG_SHR(insn, bigbit - 1) & 0x1e) | (((insn) >> (smallbit)) & 1))
2613#define VFP_DREG(reg, insn, bigbit, smallbit) do { \
2614 if (arm_dc_feature(s, ARM_FEATURE_VFP3)) { \
2615 reg = (((insn) >> (bigbit)) & 0x0f) \
2616 | (((insn) >> ((smallbit) - 4)) & 0x10); \
2617 } else { \
2618 if (insn & (1 << (smallbit))) \
2619 return 1; \
2620 reg = ((insn) >> (bigbit)) & 0x0f; \
2621 }} while (0)
2622
2623#define VFP_SREG_D(insn) VFP_SREG(insn, 12, 22)
2624#define VFP_DREG_D(reg, insn) VFP_DREG(reg, insn, 12, 22)
2625#define VFP_SREG_N(insn) VFP_SREG(insn, 16, 7)
2626#define VFP_DREG_N(reg, insn) VFP_DREG(reg, insn, 16, 7)
2627#define VFP_SREG_M(insn) VFP_SREG(insn, 0, 5)
2628#define VFP_DREG_M(reg, insn) VFP_DREG(reg, insn, 0, 5)
2629
2630static void gen_neon_dup_low16(TCGv_i32 var)
2631{
2632 TCGv_i32 tmp = tcg_temp_new_i32();
2633 tcg_gen_ext16u_i32(var, var);
2634 tcg_gen_shli_i32(tmp, var, 16);
2635 tcg_gen_or_i32(var, var, tmp);
2636 tcg_temp_free_i32(tmp);
2637}
2638
2639static void gen_neon_dup_high16(TCGv_i32 var)
2640{
2641 TCGv_i32 tmp = tcg_temp_new_i32();
2642 tcg_gen_andi_i32(var, var, 0xffff0000);
2643 tcg_gen_shri_i32(tmp, var, 16);
2644 tcg_gen_or_i32(var, var, tmp);
2645 tcg_temp_free_i32(tmp);
2646}
2647
2648/*
2649 * Disassemble a VFP instruction. Returns nonzero if an error occurred
2650 * (ie. an undefined instruction).
2651 */
2652static int disas_vfp_insn(DisasContext *s, uint32_t insn)
2653{
2654 if (!arm_dc_feature(s, ARM_FEATURE_VFP)) {
2655 return 1;
2656 }
2657
2658 /*
2659 * If the decodetree decoder handles this insn it will always
2660 * emit code to either execute the insn or generate an appropriate
2661 * exception; so we don't need to ever return non-zero to tell
2662 * the calling code to emit an UNDEF exception.
2663 */
2664 if (extract32(insn, 28, 4) == 0xf) {
2665 if (disas_vfp_uncond(s, insn)) {
2666 return 0;
2667 }
2668 } else {
2669 if (disas_vfp(s, insn)) {
2670 return 0;
2671 }
2672 }
2673 /* If the decodetree decoder didn't handle this insn, it must be UNDEF */
2674 return 1;
2675}
2676
2677static inline bool use_goto_tb(DisasContext *s, target_ulong dest)
2678{
2679#ifndef CONFIG_USER_ONLY
2680 return (s->base.tb->pc & TARGET_PAGE_MASK) == (dest & TARGET_PAGE_MASK) ||
2681 ((s->base.pc_next - 1) & TARGET_PAGE_MASK) == (dest & TARGET_PAGE_MASK);
2682#else
2683 return true;
2684#endif
2685}
2686
2687static void gen_goto_ptr(void)
2688{
2689 tcg_gen_lookup_and_goto_ptr();
2690}
2691
2692/* This will end the TB but doesn't guarantee we'll return to
2693 * cpu_loop_exec. Any live exit_requests will be processed as we
2694 * enter the next TB.
2695 */
2696static void gen_goto_tb(DisasContext *s, int n, target_ulong dest)
2697{
2698 if (use_goto_tb(s, dest)) {
2699 tcg_gen_goto_tb(n);
2700 gen_set_pc_im(s, dest);
2701 tcg_gen_exit_tb(s->base.tb, n);
2702 } else {
2703 gen_set_pc_im(s, dest);
2704 gen_goto_ptr();
2705 }
2706 s->base.is_jmp = DISAS_NORETURN;
2707}
2708
2709static inline void gen_jmp (DisasContext *s, uint32_t dest)
2710{
2711 if (unlikely(is_singlestepping(s))) {
2712 /* An indirect jump so that we still trigger the debug exception. */
2713 gen_set_pc_im(s, dest);
2714 s->base.is_jmp = DISAS_JUMP;
2715 } else {
2716 gen_goto_tb(s, 0, dest);
2717 }
2718}
2719
2720static inline void gen_mulxy(TCGv_i32 t0, TCGv_i32 t1, int x, int y)
2721{
2722 if (x)
2723 tcg_gen_sari_i32(t0, t0, 16);
2724 else
2725 gen_sxth(t0);
2726 if (y)
2727 tcg_gen_sari_i32(t1, t1, 16);
2728 else
2729 gen_sxth(t1);
2730 tcg_gen_mul_i32(t0, t0, t1);
2731}
2732
2733/* Return the mask of PSR bits set by a MSR instruction. */
2734static uint32_t msr_mask(DisasContext *s, int flags, int spsr)
2735{
2736 uint32_t mask;
2737
2738 mask = 0;
2739 if (flags & (1 << 0))
2740 mask |= 0xff;
2741 if (flags & (1 << 1))
2742 mask |= 0xff00;
2743 if (flags & (1 << 2))
2744 mask |= 0xff0000;
2745 if (flags & (1 << 3))
2746 mask |= 0xff000000;
2747
2748 /* Mask out undefined bits. */
2749 mask &= ~CPSR_RESERVED;
2750 if (!arm_dc_feature(s, ARM_FEATURE_V4T)) {
2751 mask &= ~CPSR_T;
2752 }
2753 if (!arm_dc_feature(s, ARM_FEATURE_V5)) {
2754 mask &= ~CPSR_Q; /* V5TE in reality*/
2755 }
2756 if (!arm_dc_feature(s, ARM_FEATURE_V6)) {
2757 mask &= ~(CPSR_E | CPSR_GE);
2758 }
2759 if (!arm_dc_feature(s, ARM_FEATURE_THUMB2)) {
2760 mask &= ~CPSR_IT;
2761 }
2762 /* Mask out execution state and reserved bits. */
2763 if (!spsr) {
2764 mask &= ~(CPSR_EXEC | CPSR_RESERVED);
2765 }
2766 /* Mask out privileged bits. */
2767 if (IS_USER(s))
2768 mask &= CPSR_USER;
2769 return mask;
2770}
2771
2772/* Returns nonzero if access to the PSR is not permitted. Marks t0 as dead. */
2773static int gen_set_psr(DisasContext *s, uint32_t mask, int spsr, TCGv_i32 t0)
2774{
2775 TCGv_i32 tmp;
2776 if (spsr) {
2777 /* ??? This is also undefined in system mode. */
2778 if (IS_USER(s))
2779 return 1;
2780
2781 tmp = load_cpu_field(spsr);
2782 tcg_gen_andi_i32(tmp, tmp, ~mask);
2783 tcg_gen_andi_i32(t0, t0, mask);
2784 tcg_gen_or_i32(tmp, tmp, t0);
2785 store_cpu_field(tmp, spsr);
2786 } else {
2787 gen_set_cpsr(t0, mask);
2788 }
2789 tcg_temp_free_i32(t0);
2790 gen_lookup_tb(s);
2791 return 0;
2792}
2793
2794/* Returns nonzero if access to the PSR is not permitted. */
2795static int gen_set_psr_im(DisasContext *s, uint32_t mask, int spsr, uint32_t val)
2796{
2797 TCGv_i32 tmp;
2798 tmp = tcg_temp_new_i32();
2799 tcg_gen_movi_i32(tmp, val);
2800 return gen_set_psr(s, mask, spsr, tmp);
2801}
2802
2803static bool msr_banked_access_decode(DisasContext *s, int r, int sysm, int rn,
2804 int *tgtmode, int *regno)
2805{
2806 /* Decode the r and sysm fields of MSR/MRS banked accesses into
2807 * the target mode and register number, and identify the various
2808 * unpredictable cases.
2809 * MSR (banked) and MRS (banked) are CONSTRAINED UNPREDICTABLE if:
2810 * + executed in user mode
2811 * + using R15 as the src/dest register
2812 * + accessing an unimplemented register
2813 * + accessing a register that's inaccessible at current PL/security state*
2814 * + accessing a register that you could access with a different insn
2815 * We choose to UNDEF in all these cases.
2816 * Since we don't know which of the various AArch32 modes we are in
2817 * we have to defer some checks to runtime.
2818 * Accesses to Monitor mode registers from Secure EL1 (which implies
2819 * that EL3 is AArch64) must trap to EL3.
2820 *
2821 * If the access checks fail this function will emit code to take
2822 * an exception and return false. Otherwise it will return true,
2823 * and set *tgtmode and *regno appropriately.
2824 */
2825 int exc_target = default_exception_el(s);
2826
2827 /* These instructions are present only in ARMv8, or in ARMv7 with the
2828 * Virtualization Extensions.
2829 */
2830 if (!arm_dc_feature(s, ARM_FEATURE_V8) &&
2831 !arm_dc_feature(s, ARM_FEATURE_EL2)) {
2832 goto undef;
2833 }
2834
2835 if (IS_USER(s) || rn == 15) {
2836 goto undef;
2837 }
2838
2839 /* The table in the v8 ARM ARM section F5.2.3 describes the encoding
2840 * of registers into (r, sysm).
2841 */
2842 if (r) {
2843 /* SPSRs for other modes */
2844 switch (sysm) {
2845 case 0xe: /* SPSR_fiq */
2846 *tgtmode = ARM_CPU_MODE_FIQ;
2847 break;
2848 case 0x10: /* SPSR_irq */
2849 *tgtmode = ARM_CPU_MODE_IRQ;
2850 break;
2851 case 0x12: /* SPSR_svc */
2852 *tgtmode = ARM_CPU_MODE_SVC;
2853 break;
2854 case 0x14: /* SPSR_abt */
2855 *tgtmode = ARM_CPU_MODE_ABT;
2856 break;
2857 case 0x16: /* SPSR_und */
2858 *tgtmode = ARM_CPU_MODE_UND;
2859 break;
2860 case 0x1c: /* SPSR_mon */
2861 *tgtmode = ARM_CPU_MODE_MON;
2862 break;
2863 case 0x1e: /* SPSR_hyp */
2864 *tgtmode = ARM_CPU_MODE_HYP;
2865 break;
2866 default: /* unallocated */
2867 goto undef;
2868 }
2869 /* We arbitrarily assign SPSR a register number of 16. */
2870 *regno = 16;
2871 } else {
2872 /* general purpose registers for other modes */
2873 switch (sysm) {
2874 case 0x0 ... 0x6: /* 0b00xxx : r8_usr ... r14_usr */
2875 *tgtmode = ARM_CPU_MODE_USR;
2876 *regno = sysm + 8;
2877 break;
2878 case 0x8 ... 0xe: /* 0b01xxx : r8_fiq ... r14_fiq */
2879 *tgtmode = ARM_CPU_MODE_FIQ;
2880 *regno = sysm;
2881 break;
2882 case 0x10 ... 0x11: /* 0b1000x : r14_irq, r13_irq */
2883 *tgtmode = ARM_CPU_MODE_IRQ;
2884 *regno = sysm & 1 ? 13 : 14;
2885 break;
2886 case 0x12 ... 0x13: /* 0b1001x : r14_svc, r13_svc */
2887 *tgtmode = ARM_CPU_MODE_SVC;
2888 *regno = sysm & 1 ? 13 : 14;
2889 break;
2890 case 0x14 ... 0x15: /* 0b1010x : r14_abt, r13_abt */
2891 *tgtmode = ARM_CPU_MODE_ABT;
2892 *regno = sysm & 1 ? 13 : 14;
2893 break;
2894 case 0x16 ... 0x17: /* 0b1011x : r14_und, r13_und */
2895 *tgtmode = ARM_CPU_MODE_UND;
2896 *regno = sysm & 1 ? 13 : 14;
2897 break;
2898 case 0x1c ... 0x1d: /* 0b1110x : r14_mon, r13_mon */
2899 *tgtmode = ARM_CPU_MODE_MON;
2900 *regno = sysm & 1 ? 13 : 14;
2901 break;
2902 case 0x1e ... 0x1f: /* 0b1111x : elr_hyp, r13_hyp */
2903 *tgtmode = ARM_CPU_MODE_HYP;
2904 /* Arbitrarily pick 17 for ELR_Hyp (which is not a banked LR!) */
2905 *regno = sysm & 1 ? 13 : 17;
2906 break;
2907 default: /* unallocated */
2908 goto undef;
2909 }
2910 }
2911
2912 /* Catch the 'accessing inaccessible register' cases we can detect
2913 * at translate time.
2914 */
2915 switch (*tgtmode) {
2916 case ARM_CPU_MODE_MON:
2917 if (!arm_dc_feature(s, ARM_FEATURE_EL3) || s->ns) {
2918 goto undef;
2919 }
2920 if (s->current_el == 1) {
2921 /* If we're in Secure EL1 (which implies that EL3 is AArch64)
2922 * then accesses to Mon registers trap to EL3
2923 */
2924 exc_target = 3;
2925 goto undef;
2926 }
2927 break;
2928 case ARM_CPU_MODE_HYP:
2929 /*
2930 * SPSR_hyp and r13_hyp can only be accessed from Monitor mode
2931 * (and so we can forbid accesses from EL2 or below). elr_hyp
2932 * can be accessed also from Hyp mode, so forbid accesses from
2933 * EL0 or EL1.
2934 */
2935 if (!arm_dc_feature(s, ARM_FEATURE_EL2) || s->current_el < 2 ||
2936 (s->current_el < 3 && *regno != 17)) {
2937 goto undef;
2938 }
2939 break;
2940 default:
2941 break;
2942 }
2943
2944 return true;
2945
2946undef:
2947 /* If we get here then some access check did not pass */
2948 gen_exception_insn(s, s->pc_curr, EXCP_UDEF,
2949 syn_uncategorized(), exc_target);
2950 return false;
2951}
2952
2953static void gen_msr_banked(DisasContext *s, int r, int sysm, int rn)
2954{
2955 TCGv_i32 tcg_reg, tcg_tgtmode, tcg_regno;
2956 int tgtmode = 0, regno = 0;
2957
2958 if (!msr_banked_access_decode(s, r, sysm, rn, &tgtmode, &regno)) {
2959 return;
2960 }
2961
2962 /* Sync state because msr_banked() can raise exceptions */
2963 gen_set_condexec(s);
2964 gen_set_pc_im(s, s->pc_curr);
2965 tcg_reg = load_reg(s, rn);
2966 tcg_tgtmode = tcg_const_i32(tgtmode);
2967 tcg_regno = tcg_const_i32(regno);
2968 gen_helper_msr_banked(cpu_env, tcg_reg, tcg_tgtmode, tcg_regno);
2969 tcg_temp_free_i32(tcg_tgtmode);
2970 tcg_temp_free_i32(tcg_regno);
2971 tcg_temp_free_i32(tcg_reg);
2972 s->base.is_jmp = DISAS_UPDATE;
2973}
2974
2975static void gen_mrs_banked(DisasContext *s, int r, int sysm, int rn)
2976{
2977 TCGv_i32 tcg_reg, tcg_tgtmode, tcg_regno;
2978 int tgtmode = 0, regno = 0;
2979
2980 if (!msr_banked_access_decode(s, r, sysm, rn, &tgtmode, &regno)) {
2981 return;
2982 }
2983
2984 /* Sync state because mrs_banked() can raise exceptions */
2985 gen_set_condexec(s);
2986 gen_set_pc_im(s, s->pc_curr);
2987 tcg_reg = tcg_temp_new_i32();
2988 tcg_tgtmode = tcg_const_i32(tgtmode);
2989 tcg_regno = tcg_const_i32(regno);
2990 gen_helper_mrs_banked(tcg_reg, cpu_env, tcg_tgtmode, tcg_regno);
2991 tcg_temp_free_i32(tcg_tgtmode);
2992 tcg_temp_free_i32(tcg_regno);
2993 store_reg(s, rn, tcg_reg);
2994 s->base.is_jmp = DISAS_UPDATE;
2995}
2996
2997/* Store value to PC as for an exception return (ie don't
2998 * mask bits). The subsequent call to gen_helper_cpsr_write_eret()
2999 * will do the masking based on the new value of the Thumb bit.
3000 */
3001static void store_pc_exc_ret(DisasContext *s, TCGv_i32 pc)
3002{
3003 tcg_gen_mov_i32(cpu_R[15], pc);
3004 tcg_temp_free_i32(pc);
3005}
3006
3007/* Generate a v6 exception return. Marks both values as dead. */
3008static void gen_rfe(DisasContext *s, TCGv_i32 pc, TCGv_i32 cpsr)
3009{
3010 store_pc_exc_ret(s, pc);
3011 /* The cpsr_write_eret helper will mask the low bits of PC
3012 * appropriately depending on the new Thumb bit, so it must
3013 * be called after storing the new PC.
3014 */
3015 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
3016 gen_io_start();
3017 }
3018 gen_helper_cpsr_write_eret(cpu_env, cpsr);
3019 tcg_temp_free_i32(cpsr);
3020 /* Must exit loop to check un-masked IRQs */
3021 s->base.is_jmp = DISAS_EXIT;
3022}
3023
3024/* Generate an old-style exception return. Marks pc as dead. */
3025static void gen_exception_return(DisasContext *s, TCGv_i32 pc)
3026{
3027 gen_rfe(s, pc, load_cpu_field(spsr));
3028}
3029
3030#define CPU_V001 cpu_V0, cpu_V0, cpu_V1
3031
3032static inline void gen_neon_add(int size, TCGv_i32 t0, TCGv_i32 t1)
3033{
3034 switch (size) {
3035 case 0: gen_helper_neon_add_u8(t0, t0, t1); break;
3036 case 1: gen_helper_neon_add_u16(t0, t0, t1); break;
3037 case 2: tcg_gen_add_i32(t0, t0, t1); break;
3038 default: abort();
3039 }
3040}
3041
3042static inline void gen_neon_rsb(int size, TCGv_i32 t0, TCGv_i32 t1)
3043{
3044 switch (size) {
3045 case 0: gen_helper_neon_sub_u8(t0, t1, t0); break;
3046 case 1: gen_helper_neon_sub_u16(t0, t1, t0); break;
3047 case 2: tcg_gen_sub_i32(t0, t1, t0); break;
3048 default: return;
3049 }
3050}
3051
3052/* 32-bit pairwise ops end up the same as the elementwise versions. */
3053#define gen_helper_neon_pmax_s32 tcg_gen_smax_i32
3054#define gen_helper_neon_pmax_u32 tcg_gen_umax_i32
3055#define gen_helper_neon_pmin_s32 tcg_gen_smin_i32
3056#define gen_helper_neon_pmin_u32 tcg_gen_umin_i32
3057
3058#define GEN_NEON_INTEGER_OP_ENV(name) do { \
3059 switch ((size << 1) | u) { \
3060 case 0: \
3061 gen_helper_neon_##name##_s8(tmp, cpu_env, tmp, tmp2); \
3062 break; \
3063 case 1: \
3064 gen_helper_neon_##name##_u8(tmp, cpu_env, tmp, tmp2); \
3065 break; \
3066 case 2: \
3067 gen_helper_neon_##name##_s16(tmp, cpu_env, tmp, tmp2); \
3068 break; \
3069 case 3: \
3070 gen_helper_neon_##name##_u16(tmp, cpu_env, tmp, tmp2); \
3071 break; \
3072 case 4: \
3073 gen_helper_neon_##name##_s32(tmp, cpu_env, tmp, tmp2); \
3074 break; \
3075 case 5: \
3076 gen_helper_neon_##name##_u32(tmp, cpu_env, tmp, tmp2); \
3077 break; \
3078 default: return 1; \
3079 }} while (0)
3080
3081#define GEN_NEON_INTEGER_OP(name) do { \
3082 switch ((size << 1) | u) { \
3083 case 0: \
3084 gen_helper_neon_##name##_s8(tmp, tmp, tmp2); \
3085 break; \
3086 case 1: \
3087 gen_helper_neon_##name##_u8(tmp, tmp, tmp2); \
3088 break; \
3089 case 2: \
3090 gen_helper_neon_##name##_s16(tmp, tmp, tmp2); \
3091 break; \
3092 case 3: \
3093 gen_helper_neon_##name##_u16(tmp, tmp, tmp2); \
3094 break; \
3095 case 4: \
3096 gen_helper_neon_##name##_s32(tmp, tmp, tmp2); \
3097 break; \
3098 case 5: \
3099 gen_helper_neon_##name##_u32(tmp, tmp, tmp2); \
3100 break; \
3101 default: return 1; \
3102 }} while (0)
3103
3104static TCGv_i32 neon_load_scratch(int scratch)
3105{
3106 TCGv_i32 tmp = tcg_temp_new_i32();
3107 tcg_gen_ld_i32(tmp, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
3108 return tmp;
3109}
3110
3111static void neon_store_scratch(int scratch, TCGv_i32 var)
3112{
3113 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
3114 tcg_temp_free_i32(var);
3115}
3116
3117static inline TCGv_i32 neon_get_scalar(int size, int reg)
3118{
3119 TCGv_i32 tmp;
3120 if (size == 1) {
3121 tmp = neon_load_reg(reg & 7, reg >> 4);
3122 if (reg & 8) {
3123 gen_neon_dup_high16(tmp);
3124 } else {
3125 gen_neon_dup_low16(tmp);
3126 }
3127 } else {
3128 tmp = neon_load_reg(reg & 15, reg >> 4);
3129 }
3130 return tmp;
3131}
3132
3133static int gen_neon_unzip(int rd, int rm, int size, int q)
3134{
3135 TCGv_ptr pd, pm;
3136
3137 if (!q && size == 2) {
3138 return 1;
3139 }
3140 pd = vfp_reg_ptr(true, rd);
3141 pm = vfp_reg_ptr(true, rm);
3142 if (q) {
3143 switch (size) {
3144 case 0:
3145 gen_helper_neon_qunzip8(pd, pm);
3146 break;
3147 case 1:
3148 gen_helper_neon_qunzip16(pd, pm);
3149 break;
3150 case 2:
3151 gen_helper_neon_qunzip32(pd, pm);
3152 break;
3153 default:
3154 abort();
3155 }
3156 } else {
3157 switch (size) {
3158 case 0:
3159 gen_helper_neon_unzip8(pd, pm);
3160 break;
3161 case 1:
3162 gen_helper_neon_unzip16(pd, pm);
3163 break;
3164 default:
3165 abort();
3166 }
3167 }
3168 tcg_temp_free_ptr(pd);
3169 tcg_temp_free_ptr(pm);
3170 return 0;
3171}
3172
3173static int gen_neon_zip(int rd, int rm, int size, int q)
3174{
3175 TCGv_ptr pd, pm;
3176
3177 if (!q && size == 2) {
3178 return 1;
3179 }
3180 pd = vfp_reg_ptr(true, rd);
3181 pm = vfp_reg_ptr(true, rm);
3182 if (q) {
3183 switch (size) {
3184 case 0:
3185 gen_helper_neon_qzip8(pd, pm);
3186 break;
3187 case 1:
3188 gen_helper_neon_qzip16(pd, pm);
3189 break;
3190 case 2:
3191 gen_helper_neon_qzip32(pd, pm);
3192 break;
3193 default:
3194 abort();
3195 }
3196 } else {
3197 switch (size) {
3198 case 0:
3199 gen_helper_neon_zip8(pd, pm);
3200 break;
3201 case 1:
3202 gen_helper_neon_zip16(pd, pm);
3203 break;
3204 default:
3205 abort();
3206 }
3207 }
3208 tcg_temp_free_ptr(pd);
3209 tcg_temp_free_ptr(pm);
3210 return 0;
3211}
3212
3213static void gen_neon_trn_u8(TCGv_i32 t0, TCGv_i32 t1)
3214{
3215 TCGv_i32 rd, tmp;
3216
3217 rd = tcg_temp_new_i32();
3218 tmp = tcg_temp_new_i32();
3219
3220 tcg_gen_shli_i32(rd, t0, 8);
3221 tcg_gen_andi_i32(rd, rd, 0xff00ff00);
3222 tcg_gen_andi_i32(tmp, t1, 0x00ff00ff);
3223 tcg_gen_or_i32(rd, rd, tmp);
3224
3225 tcg_gen_shri_i32(t1, t1, 8);
3226 tcg_gen_andi_i32(t1, t1, 0x00ff00ff);
3227 tcg_gen_andi_i32(tmp, t0, 0xff00ff00);
3228 tcg_gen_or_i32(t1, t1, tmp);
3229 tcg_gen_mov_i32(t0, rd);
3230
3231 tcg_temp_free_i32(tmp);
3232 tcg_temp_free_i32(rd);
3233}
3234
3235static void gen_neon_trn_u16(TCGv_i32 t0, TCGv_i32 t1)
3236{
3237 TCGv_i32 rd, tmp;
3238
3239 rd = tcg_temp_new_i32();
3240 tmp = tcg_temp_new_i32();
3241
3242 tcg_gen_shli_i32(rd, t0, 16);
3243 tcg_gen_andi_i32(tmp, t1, 0xffff);
3244 tcg_gen_or_i32(rd, rd, tmp);
3245 tcg_gen_shri_i32(t1, t1, 16);
3246 tcg_gen_andi_i32(tmp, t0, 0xffff0000);
3247 tcg_gen_or_i32(t1, t1, tmp);
3248 tcg_gen_mov_i32(t0, rd);
3249
3250 tcg_temp_free_i32(tmp);
3251 tcg_temp_free_i32(rd);
3252}
3253
3254
3255static struct {
3256 int nregs;
3257 int interleave;
3258 int spacing;
3259} const neon_ls_element_type[11] = {
3260 {1, 4, 1},
3261 {1, 4, 2},
3262 {4, 1, 1},
3263 {2, 2, 2},
3264 {1, 3, 1},
3265 {1, 3, 2},
3266 {3, 1, 1},
3267 {1, 1, 1},
3268 {1, 2, 1},
3269 {1, 2, 2},
3270 {2, 1, 1}
3271};
3272
3273/* Translate a NEON load/store element instruction. Return nonzero if the
3274 instruction is invalid. */
3275static int disas_neon_ls_insn(DisasContext *s, uint32_t insn)
3276{
3277 int rd, rn, rm;
3278 int op;
3279 int nregs;
3280 int interleave;
3281 int spacing;
3282 int stride;
3283 int size;
3284 int reg;
3285 int load;
3286 int n;
3287 int vec_size;
3288 int mmu_idx;
3289 MemOp endian;
3290 TCGv_i32 addr;
3291 TCGv_i32 tmp;
3292 TCGv_i32 tmp2;
3293 TCGv_i64 tmp64;
3294
3295 /* FIXME: this access check should not take precedence over UNDEF
3296 * for invalid encodings; we will generate incorrect syndrome information
3297 * for attempts to execute invalid vfp/neon encodings with FP disabled.
3298 */
3299 if (s->fp_excp_el) {
3300 gen_exception_insn(s, s->pc_curr, EXCP_UDEF,
3301 syn_simd_access_trap(1, 0xe, false), s->fp_excp_el);
3302 return 0;
3303 }
3304
3305 if (!s->vfp_enabled)
3306 return 1;
3307 VFP_DREG_D(rd, insn);
3308 rn = (insn >> 16) & 0xf;
3309 rm = insn & 0xf;
3310 load = (insn & (1 << 21)) != 0;
3311 endian = s->be_data;
3312 mmu_idx = get_mem_index(s);
3313 if ((insn & (1 << 23)) == 0) {
3314 /* Load store all elements. */
3315 op = (insn >> 8) & 0xf;
3316 size = (insn >> 6) & 3;
3317 if (op > 10)
3318 return 1;
3319 /* Catch UNDEF cases for bad values of align field */
3320 switch (op & 0xc) {
3321 case 4:
3322 if (((insn >> 5) & 1) == 1) {
3323 return 1;
3324 }
3325 break;
3326 case 8:
3327 if (((insn >> 4) & 3) == 3) {
3328 return 1;
3329 }
3330 break;
3331 default:
3332 break;
3333 }
3334 nregs = neon_ls_element_type[op].nregs;
3335 interleave = neon_ls_element_type[op].interleave;
3336 spacing = neon_ls_element_type[op].spacing;
3337 if (size == 3 && (interleave | spacing) != 1) {
3338 return 1;
3339 }
3340 /* For our purposes, bytes are always little-endian. */
3341 if (size == 0) {
3342 endian = MO_LE;
3343 }
3344 /* Consecutive little-endian elements from a single register
3345 * can be promoted to a larger little-endian operation.
3346 */
3347 if (interleave == 1 && endian == MO_LE) {
3348 size = 3;
3349 }
3350 tmp64 = tcg_temp_new_i64();
3351 addr = tcg_temp_new_i32();
3352 tmp2 = tcg_const_i32(1 << size);
3353 load_reg_var(s, addr, rn);
3354 for (reg = 0; reg < nregs; reg++) {
3355 for (n = 0; n < 8 >> size; n++) {
3356 int xs;
3357 for (xs = 0; xs < interleave; xs++) {
3358 int tt = rd + reg + spacing * xs;
3359
3360 if (load) {
3361 gen_aa32_ld_i64(s, tmp64, addr, mmu_idx, endian | size);
3362 neon_store_element64(tt, n, size, tmp64);
3363 } else {
3364 neon_load_element64(tmp64, tt, n, size);
3365 gen_aa32_st_i64(s, tmp64, addr, mmu_idx, endian | size);
3366 }
3367 tcg_gen_add_i32(addr, addr, tmp2);
3368 }
3369 }
3370 }
3371 tcg_temp_free_i32(addr);
3372 tcg_temp_free_i32(tmp2);
3373 tcg_temp_free_i64(tmp64);
3374 stride = nregs * interleave * 8;
3375 } else {
3376 size = (insn >> 10) & 3;
3377 if (size == 3) {
3378 /* Load single element to all lanes. */
3379 int a = (insn >> 4) & 1;
3380 if (!load) {
3381 return 1;
3382 }
3383 size = (insn >> 6) & 3;
3384 nregs = ((insn >> 8) & 3) + 1;
3385
3386 if (size == 3) {
3387 if (nregs != 4 || a == 0) {
3388 return 1;
3389 }
3390 /* For VLD4 size==3 a == 1 means 32 bits at 16 byte alignment */
3391 size = 2;
3392 }
3393 if (nregs == 1 && a == 1 && size == 0) {
3394 return 1;
3395 }
3396 if (nregs == 3 && a == 1) {
3397 return 1;
3398 }
3399 addr = tcg_temp_new_i32();
3400 load_reg_var(s, addr, rn);
3401
3402 /* VLD1 to all lanes: bit 5 indicates how many Dregs to write.
3403 * VLD2/3/4 to all lanes: bit 5 indicates register stride.
3404 */
3405 stride = (insn & (1 << 5)) ? 2 : 1;
3406 vec_size = nregs == 1 ? stride * 8 : 8;
3407
3408 tmp = tcg_temp_new_i32();
3409 for (reg = 0; reg < nregs; reg++) {
3410 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s),
3411 s->be_data | size);
3412 if ((rd & 1) && vec_size == 16) {
3413 /* We cannot write 16 bytes at once because the
3414 * destination is unaligned.
3415 */
3416 tcg_gen_gvec_dup_i32(size, neon_reg_offset(rd, 0),
3417 8, 8, tmp);
3418 tcg_gen_gvec_mov(0, neon_reg_offset(rd + 1, 0),
3419 neon_reg_offset(rd, 0), 8, 8);
3420 } else {
3421 tcg_gen_gvec_dup_i32(size, neon_reg_offset(rd, 0),
3422 vec_size, vec_size, tmp);
3423 }
3424 tcg_gen_addi_i32(addr, addr, 1 << size);
3425 rd += stride;
3426 }
3427 tcg_temp_free_i32(tmp);
3428 tcg_temp_free_i32(addr);
3429 stride = (1 << size) * nregs;
3430 } else {
3431 /* Single element. */
3432 int idx = (insn >> 4) & 0xf;
3433 int reg_idx;
3434 switch (size) {
3435 case 0:
3436 reg_idx = (insn >> 5) & 7;
3437 stride = 1;
3438 break;
3439 case 1:
3440 reg_idx = (insn >> 6) & 3;
3441 stride = (insn & (1 << 5)) ? 2 : 1;
3442 break;
3443 case 2:
3444 reg_idx = (insn >> 7) & 1;
3445 stride = (insn & (1 << 6)) ? 2 : 1;
3446 break;
3447 default:
3448 abort();
3449 }
3450 nregs = ((insn >> 8) & 3) + 1;
3451 /* Catch the UNDEF cases. This is unavoidably a bit messy. */
3452 switch (nregs) {
3453 case 1:
3454 if (((idx & (1 << size)) != 0) ||
3455 (size == 2 && ((idx & 3) == 1 || (idx & 3) == 2))) {
3456 return 1;
3457 }
3458 break;
3459 case 3:
3460 if ((idx & 1) != 0) {
3461 return 1;
3462 }
3463 /* fall through */
3464 case 2:
3465 if (size == 2 && (idx & 2) != 0) {
3466 return 1;
3467 }
3468 break;
3469 case 4:
3470 if ((size == 2) && ((idx & 3) == 3)) {
3471 return 1;
3472 }
3473 break;
3474 default:
3475 abort();
3476 }
3477 if ((rd + stride * (nregs - 1)) > 31) {
3478 /* Attempts to write off the end of the register file
3479 * are UNPREDICTABLE; we choose to UNDEF because otherwise
3480 * the neon_load_reg() would write off the end of the array.
3481 */
3482 return 1;
3483 }
3484 tmp = tcg_temp_new_i32();
3485 addr = tcg_temp_new_i32();
3486 load_reg_var(s, addr, rn);
3487 for (reg = 0; reg < nregs; reg++) {
3488 if (load) {
3489 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s),
3490 s->be_data | size);
3491 neon_store_element(rd, reg_idx, size, tmp);
3492 } else { /* Store */
3493 neon_load_element(tmp, rd, reg_idx, size);
3494 gen_aa32_st_i32(s, tmp, addr, get_mem_index(s),
3495 s->be_data | size);
3496 }
3497 rd += stride;
3498 tcg_gen_addi_i32(addr, addr, 1 << size);
3499 }
3500 tcg_temp_free_i32(addr);
3501 tcg_temp_free_i32(tmp);
3502 stride = nregs * (1 << size);
3503 }
3504 }
3505 if (rm != 15) {
3506 TCGv_i32 base;
3507
3508 base = load_reg(s, rn);
3509 if (rm == 13) {
3510 tcg_gen_addi_i32(base, base, stride);
3511 } else {
3512 TCGv_i32 index;
3513 index = load_reg(s, rm);
3514 tcg_gen_add_i32(base, base, index);
3515 tcg_temp_free_i32(index);
3516 }
3517 store_reg(s, rn, base);
3518 }
3519 return 0;
3520}
3521
3522static inline void gen_neon_narrow(int size, TCGv_i32 dest, TCGv_i64 src)
3523{
3524 switch (size) {
3525 case 0: gen_helper_neon_narrow_u8(dest, src); break;
3526 case 1: gen_helper_neon_narrow_u16(dest, src); break;
3527 case 2: tcg_gen_extrl_i64_i32(dest, src); break;
3528 default: abort();
3529 }
3530}
3531
3532static inline void gen_neon_narrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
3533{
3534 switch (size) {
3535 case 0: gen_helper_neon_narrow_sat_s8(dest, cpu_env, src); break;
3536 case 1: gen_helper_neon_narrow_sat_s16(dest, cpu_env, src); break;
3537 case 2: gen_helper_neon_narrow_sat_s32(dest, cpu_env, src); break;
3538 default: abort();
3539 }
3540}
3541
3542static inline void gen_neon_narrow_satu(int size, TCGv_i32 dest, TCGv_i64 src)
3543{
3544 switch (size) {
3545 case 0: gen_helper_neon_narrow_sat_u8(dest, cpu_env, src); break;
3546 case 1: gen_helper_neon_narrow_sat_u16(dest, cpu_env, src); break;
3547 case 2: gen_helper_neon_narrow_sat_u32(dest, cpu_env, src); break;
3548 default: abort();
3549 }
3550}
3551
3552static inline void gen_neon_unarrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
3553{
3554 switch (size) {
3555 case 0: gen_helper_neon_unarrow_sat8(dest, cpu_env, src); break;
3556 case 1: gen_helper_neon_unarrow_sat16(dest, cpu_env, src); break;
3557 case 2: gen_helper_neon_unarrow_sat32(dest, cpu_env, src); break;
3558 default: abort();
3559 }
3560}
3561
3562static inline void gen_neon_shift_narrow(int size, TCGv_i32 var, TCGv_i32 shift,
3563 int q, int u)
3564{
3565 if (q) {
3566 if (u) {
3567 switch (size) {
3568 case 1: gen_helper_neon_rshl_u16(var, var, shift); break;
3569 case 2: gen_helper_neon_rshl_u32(var, var, shift); break;
3570 default: abort();
3571 }
3572 } else {
3573 switch (size) {
3574 case 1: gen_helper_neon_rshl_s16(var, var, shift); break;
3575 case 2: gen_helper_neon_rshl_s32(var, var, shift); break;
3576 default: abort();
3577 }
3578 }
3579 } else {
3580 if (u) {
3581 switch (size) {
3582 case 1: gen_helper_neon_shl_u16(var, var, shift); break;
3583 case 2: gen_helper_neon_shl_u32(var, var, shift); break;
3584 default: abort();
3585 }
3586 } else {
3587 switch (size) {
3588 case 1: gen_helper_neon_shl_s16(var, var, shift); break;
3589 case 2: gen_helper_neon_shl_s32(var, var, shift); break;
3590 default: abort();
3591 }
3592 }
3593 }
3594}
3595
3596static inline void gen_neon_widen(TCGv_i64 dest, TCGv_i32 src, int size, int u)
3597{
3598 if (u) {
3599 switch (size) {
3600 case 0: gen_helper_neon_widen_u8(dest, src); break;
3601 case 1: gen_helper_neon_widen_u16(dest, src); break;
3602 case 2: tcg_gen_extu_i32_i64(dest, src); break;
3603 default: abort();
3604 }
3605 } else {
3606 switch (size) {
3607 case 0: gen_helper_neon_widen_s8(dest, src); break;
3608 case 1: gen_helper_neon_widen_s16(dest, src); break;
3609 case 2: tcg_gen_ext_i32_i64(dest, src); break;
3610 default: abort();
3611 }
3612 }
3613 tcg_temp_free_i32(src);
3614}
3615
3616static inline void gen_neon_addl(int size)
3617{
3618 switch (size) {
3619 case 0: gen_helper_neon_addl_u16(CPU_V001); break;
3620 case 1: gen_helper_neon_addl_u32(CPU_V001); break;
3621 case 2: tcg_gen_add_i64(CPU_V001); break;
3622 default: abort();
3623 }
3624}
3625
3626static inline void gen_neon_subl(int size)
3627{
3628 switch (size) {
3629 case 0: gen_helper_neon_subl_u16(CPU_V001); break;
3630 case 1: gen_helper_neon_subl_u32(CPU_V001); break;
3631 case 2: tcg_gen_sub_i64(CPU_V001); break;
3632 default: abort();
3633 }
3634}
3635
3636static inline void gen_neon_negl(TCGv_i64 var, int size)
3637{
3638 switch (size) {
3639 case 0: gen_helper_neon_negl_u16(var, var); break;
3640 case 1: gen_helper_neon_negl_u32(var, var); break;
3641 case 2:
3642 tcg_gen_neg_i64(var, var);
3643 break;
3644 default: abort();
3645 }
3646}
3647
3648static inline void gen_neon_addl_saturate(TCGv_i64 op0, TCGv_i64 op1, int size)
3649{
3650 switch (size) {
3651 case 1: gen_helper_neon_addl_saturate_s32(op0, cpu_env, op0, op1); break;
3652 case 2: gen_helper_neon_addl_saturate_s64(op0, cpu_env, op0, op1); break;
3653 default: abort();
3654 }
3655}
3656
3657static inline void gen_neon_mull(TCGv_i64 dest, TCGv_i32 a, TCGv_i32 b,
3658 int size, int u)
3659{
3660 TCGv_i64 tmp;
3661
3662 switch ((size << 1) | u) {
3663 case 0: gen_helper_neon_mull_s8(dest, a, b); break;
3664 case 1: gen_helper_neon_mull_u8(dest, a, b); break;
3665 case 2: gen_helper_neon_mull_s16(dest, a, b); break;
3666 case 3: gen_helper_neon_mull_u16(dest, a, b); break;
3667 case 4:
3668 tmp = gen_muls_i64_i32(a, b);
3669 tcg_gen_mov_i64(dest, tmp);
3670 tcg_temp_free_i64(tmp);
3671 break;
3672 case 5:
3673 tmp = gen_mulu_i64_i32(a, b);
3674 tcg_gen_mov_i64(dest, tmp);
3675 tcg_temp_free_i64(tmp);
3676 break;
3677 default: abort();
3678 }
3679
3680 /* gen_helper_neon_mull_[su]{8|16} do not free their parameters.
3681 Don't forget to clean them now. */
3682 if (size < 2) {
3683 tcg_temp_free_i32(a);
3684 tcg_temp_free_i32(b);
3685 }
3686}
3687
3688static void gen_neon_narrow_op(int op, int u, int size,
3689 TCGv_i32 dest, TCGv_i64 src)
3690{
3691 if (op) {
3692 if (u) {
3693 gen_neon_unarrow_sats(size, dest, src);
3694 } else {
3695 gen_neon_narrow(size, dest, src);
3696 }
3697 } else {
3698 if (u) {
3699 gen_neon_narrow_satu(size, dest, src);
3700 } else {
3701 gen_neon_narrow_sats(size, dest, src);
3702 }
3703 }
3704}
3705
3706/* Symbolic constants for op fields for Neon 3-register same-length.
3707 * The values correspond to bits [11:8,4]; see the ARM ARM DDI0406B
3708 * table A7-9.
3709 */
3710#define NEON_3R_VHADD 0
3711#define NEON_3R_VQADD 1
3712#define NEON_3R_VRHADD 2
3713#define NEON_3R_LOGIC 3 /* VAND,VBIC,VORR,VMOV,VORN,VEOR,VBIF,VBIT,VBSL */
3714#define NEON_3R_VHSUB 4
3715#define NEON_3R_VQSUB 5
3716#define NEON_3R_VCGT 6
3717#define NEON_3R_VCGE 7
3718#define NEON_3R_VSHL 8
3719#define NEON_3R_VQSHL 9
3720#define NEON_3R_VRSHL 10
3721#define NEON_3R_VQRSHL 11
3722#define NEON_3R_VMAX 12
3723#define NEON_3R_VMIN 13
3724#define NEON_3R_VABD 14
3725#define NEON_3R_VABA 15
3726#define NEON_3R_VADD_VSUB 16
3727#define NEON_3R_VTST_VCEQ 17
3728#define NEON_3R_VML 18 /* VMLA, VMLS */
3729#define NEON_3R_VMUL 19
3730#define NEON_3R_VPMAX 20
3731#define NEON_3R_VPMIN 21
3732#define NEON_3R_VQDMULH_VQRDMULH 22
3733#define NEON_3R_VPADD_VQRDMLAH 23
3734#define NEON_3R_SHA 24 /* SHA1C,SHA1P,SHA1M,SHA1SU0,SHA256H{2},SHA256SU1 */
3735#define NEON_3R_VFM_VQRDMLSH 25 /* VFMA, VFMS, VQRDMLSH */
3736#define NEON_3R_FLOAT_ARITH 26 /* float VADD, VSUB, VPADD, VABD */
3737#define NEON_3R_FLOAT_MULTIPLY 27 /* float VMLA, VMLS, VMUL */
3738#define NEON_3R_FLOAT_CMP 28 /* float VCEQ, VCGE, VCGT */
3739#define NEON_3R_FLOAT_ACMP 29 /* float VACGE, VACGT, VACLE, VACLT */
3740#define NEON_3R_FLOAT_MINMAX 30 /* float VMIN, VMAX */
3741#define NEON_3R_FLOAT_MISC 31 /* float VRECPS, VRSQRTS, VMAXNM/MINNM */
3742
3743static const uint8_t neon_3r_sizes[] = {
3744 [NEON_3R_VHADD] = 0x7,
3745 [NEON_3R_VQADD] = 0xf,
3746 [NEON_3R_VRHADD] = 0x7,
3747 [NEON_3R_LOGIC] = 0xf, /* size field encodes op type */
3748 [NEON_3R_VHSUB] = 0x7,
3749 [NEON_3R_VQSUB] = 0xf,
3750 [NEON_3R_VCGT] = 0x7,
3751 [NEON_3R_VCGE] = 0x7,
3752 [NEON_3R_VSHL] = 0xf,
3753 [NEON_3R_VQSHL] = 0xf,
3754 [NEON_3R_VRSHL] = 0xf,
3755 [NEON_3R_VQRSHL] = 0xf,
3756 [NEON_3R_VMAX] = 0x7,
3757 [NEON_3R_VMIN] = 0x7,
3758 [NEON_3R_VABD] = 0x7,
3759 [NEON_3R_VABA] = 0x7,
3760 [NEON_3R_VADD_VSUB] = 0xf,
3761 [NEON_3R_VTST_VCEQ] = 0x7,
3762 [NEON_3R_VML] = 0x7,
3763 [NEON_3R_VMUL] = 0x7,
3764 [NEON_3R_VPMAX] = 0x7,
3765 [NEON_3R_VPMIN] = 0x7,
3766 [NEON_3R_VQDMULH_VQRDMULH] = 0x6,
3767 [NEON_3R_VPADD_VQRDMLAH] = 0x7,
3768 [NEON_3R_SHA] = 0xf, /* size field encodes op type */
3769 [NEON_3R_VFM_VQRDMLSH] = 0x7, /* For VFM, size bit 1 encodes op */
3770 [NEON_3R_FLOAT_ARITH] = 0x5, /* size bit 1 encodes op */
3771 [NEON_3R_FLOAT_MULTIPLY] = 0x5, /* size bit 1 encodes op */
3772 [NEON_3R_FLOAT_CMP] = 0x5, /* size bit 1 encodes op */
3773 [NEON_3R_FLOAT_ACMP] = 0x5, /* size bit 1 encodes op */
3774 [NEON_3R_FLOAT_MINMAX] = 0x5, /* size bit 1 encodes op */
3775 [NEON_3R_FLOAT_MISC] = 0x5, /* size bit 1 encodes op */
3776};
3777
3778/* Symbolic constants for op fields for Neon 2-register miscellaneous.
3779 * The values correspond to bits [17:16,10:7]; see the ARM ARM DDI0406B
3780 * table A7-13.
3781 */
3782#define NEON_2RM_VREV64 0
3783#define NEON_2RM_VREV32 1
3784#define NEON_2RM_VREV16 2
3785#define NEON_2RM_VPADDL 4
3786#define NEON_2RM_VPADDL_U 5
3787#define NEON_2RM_AESE 6 /* Includes AESD */
3788#define NEON_2RM_AESMC 7 /* Includes AESIMC */
3789#define NEON_2RM_VCLS 8
3790#define NEON_2RM_VCLZ 9
3791#define NEON_2RM_VCNT 10
3792#define NEON_2RM_VMVN 11
3793#define NEON_2RM_VPADAL 12
3794#define NEON_2RM_VPADAL_U 13
3795#define NEON_2RM_VQABS 14
3796#define NEON_2RM_VQNEG 15
3797#define NEON_2RM_VCGT0 16
3798#define NEON_2RM_VCGE0 17
3799#define NEON_2RM_VCEQ0 18
3800#define NEON_2RM_VCLE0 19
3801#define NEON_2RM_VCLT0 20
3802#define NEON_2RM_SHA1H 21
3803#define NEON_2RM_VABS 22
3804#define NEON_2RM_VNEG 23
3805#define NEON_2RM_VCGT0_F 24
3806#define NEON_2RM_VCGE0_F 25
3807#define NEON_2RM_VCEQ0_F 26
3808#define NEON_2RM_VCLE0_F 27
3809#define NEON_2RM_VCLT0_F 28
3810#define NEON_2RM_VABS_F 30
3811#define NEON_2RM_VNEG_F 31
3812#define NEON_2RM_VSWP 32
3813#define NEON_2RM_VTRN 33
3814#define NEON_2RM_VUZP 34
3815#define NEON_2RM_VZIP 35
3816#define NEON_2RM_VMOVN 36 /* Includes VQMOVN, VQMOVUN */
3817#define NEON_2RM_VQMOVN 37 /* Includes VQMOVUN */
3818#define NEON_2RM_VSHLL 38
3819#define NEON_2RM_SHA1SU1 39 /* Includes SHA256SU0 */
3820#define NEON_2RM_VRINTN 40
3821#define NEON_2RM_VRINTX 41
3822#define NEON_2RM_VRINTA 42
3823#define NEON_2RM_VRINTZ 43
3824#define NEON_2RM_VCVT_F16_F32 44
3825#define NEON_2RM_VRINTM 45
3826#define NEON_2RM_VCVT_F32_F16 46
3827#define NEON_2RM_VRINTP 47
3828#define NEON_2RM_VCVTAU 48
3829#define NEON_2RM_VCVTAS 49
3830#define NEON_2RM_VCVTNU 50
3831#define NEON_2RM_VCVTNS 51
3832#define NEON_2RM_VCVTPU 52
3833#define NEON_2RM_VCVTPS 53
3834#define NEON_2RM_VCVTMU 54
3835#define NEON_2RM_VCVTMS 55
3836#define NEON_2RM_VRECPE 56
3837#define NEON_2RM_VRSQRTE 57
3838#define NEON_2RM_VRECPE_F 58
3839#define NEON_2RM_VRSQRTE_F 59
3840#define NEON_2RM_VCVT_FS 60
3841#define NEON_2RM_VCVT_FU 61
3842#define NEON_2RM_VCVT_SF 62
3843#define NEON_2RM_VCVT_UF 63
3844
3845static bool neon_2rm_is_v8_op(int op)
3846{
3847 /* Return true if this neon 2reg-misc op is ARMv8 and up */
3848 switch (op) {
3849 case NEON_2RM_VRINTN:
3850 case NEON_2RM_VRINTA:
3851 case NEON_2RM_VRINTM:
3852 case NEON_2RM_VRINTP:
3853 case NEON_2RM_VRINTZ:
3854 case NEON_2RM_VRINTX:
3855 case NEON_2RM_VCVTAU:
3856 case NEON_2RM_VCVTAS:
3857 case NEON_2RM_VCVTNU:
3858 case NEON_2RM_VCVTNS:
3859 case NEON_2RM_VCVTPU:
3860 case NEON_2RM_VCVTPS:
3861 case NEON_2RM_VCVTMU:
3862 case NEON_2RM_VCVTMS:
3863 return true;
3864 default:
3865 return false;
3866 }
3867}
3868
3869/* Each entry in this array has bit n set if the insn allows
3870 * size value n (otherwise it will UNDEF). Since unallocated
3871 * op values will have no bits set they always UNDEF.
3872 */
3873static const uint8_t neon_2rm_sizes[] = {
3874 [NEON_2RM_VREV64] = 0x7,
3875 [NEON_2RM_VREV32] = 0x3,
3876 [NEON_2RM_VREV16] = 0x1,
3877 [NEON_2RM_VPADDL] = 0x7,
3878 [NEON_2RM_VPADDL_U] = 0x7,
3879 [NEON_2RM_AESE] = 0x1,
3880 [NEON_2RM_AESMC] = 0x1,
3881 [NEON_2RM_VCLS] = 0x7,
3882 [NEON_2RM_VCLZ] = 0x7,
3883 [NEON_2RM_VCNT] = 0x1,
3884 [NEON_2RM_VMVN] = 0x1,
3885 [NEON_2RM_VPADAL] = 0x7,
3886 [NEON_2RM_VPADAL_U] = 0x7,
3887 [NEON_2RM_VQABS] = 0x7,
3888 [NEON_2RM_VQNEG] = 0x7,
3889 [NEON_2RM_VCGT0] = 0x7,
3890 [NEON_2RM_VCGE0] = 0x7,
3891 [NEON_2RM_VCEQ0] = 0x7,
3892 [NEON_2RM_VCLE0] = 0x7,
3893 [NEON_2RM_VCLT0] = 0x7,
3894 [NEON_2RM_SHA1H] = 0x4,
3895 [NEON_2RM_VABS] = 0x7,
3896 [NEON_2RM_VNEG] = 0x7,
3897 [NEON_2RM_VCGT0_F] = 0x4,
3898 [NEON_2RM_VCGE0_F] = 0x4,
3899 [NEON_2RM_VCEQ0_F] = 0x4,
3900 [NEON_2RM_VCLE0_F] = 0x4,
3901 [NEON_2RM_VCLT0_F] = 0x4,
3902 [NEON_2RM_VABS_F] = 0x4,
3903 [NEON_2RM_VNEG_F] = 0x4,
3904 [NEON_2RM_VSWP] = 0x1,
3905 [NEON_2RM_VTRN] = 0x7,
3906 [NEON_2RM_VUZP] = 0x7,
3907 [NEON_2RM_VZIP] = 0x7,
3908 [NEON_2RM_VMOVN] = 0x7,
3909 [NEON_2RM_VQMOVN] = 0x7,
3910 [NEON_2RM_VSHLL] = 0x7,
3911 [NEON_2RM_SHA1SU1] = 0x4,
3912 [NEON_2RM_VRINTN] = 0x4,
3913 [NEON_2RM_VRINTX] = 0x4,
3914 [NEON_2RM_VRINTA] = 0x4,
3915 [NEON_2RM_VRINTZ] = 0x4,
3916 [NEON_2RM_VCVT_F16_F32] = 0x2,
3917 [NEON_2RM_VRINTM] = 0x4,
3918 [NEON_2RM_VCVT_F32_F16] = 0x2,
3919 [NEON_2RM_VRINTP] = 0x4,
3920 [NEON_2RM_VCVTAU] = 0x4,
3921 [NEON_2RM_VCVTAS] = 0x4,
3922 [NEON_2RM_VCVTNU] = 0x4,
3923 [NEON_2RM_VCVTNS] = 0x4,
3924 [NEON_2RM_VCVTPU] = 0x4,
3925 [NEON_2RM_VCVTPS] = 0x4,
3926 [NEON_2RM_VCVTMU] = 0x4,
3927 [NEON_2RM_VCVTMS] = 0x4,
3928 [NEON_2RM_VRECPE] = 0x4,
3929 [NEON_2RM_VRSQRTE] = 0x4,
3930 [NEON_2RM_VRECPE_F] = 0x4,
3931 [NEON_2RM_VRSQRTE_F] = 0x4,
3932 [NEON_2RM_VCVT_FS] = 0x4,
3933 [NEON_2RM_VCVT_FU] = 0x4,
3934 [NEON_2RM_VCVT_SF] = 0x4,
3935 [NEON_2RM_VCVT_UF] = 0x4,
3936};
3937
3938
3939/* Expand v8.1 simd helper. */
3940static int do_v81_helper(DisasContext *s, gen_helper_gvec_3_ptr *fn,
3941 int q, int rd, int rn, int rm)
3942{
3943 if (dc_isar_feature(aa32_rdm, s)) {
3944 int opr_sz = (1 + q) * 8;
3945 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, rd),
3946 vfp_reg_offset(1, rn),
3947 vfp_reg_offset(1, rm), cpu_env,
3948 opr_sz, opr_sz, 0, fn);
3949 return 0;
3950 }
3951 return 1;
3952}
3953
3954static void gen_ssra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3955{
3956 tcg_gen_vec_sar8i_i64(a, a, shift);
3957 tcg_gen_vec_add8_i64(d, d, a);
3958}
3959
3960static void gen_ssra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3961{
3962 tcg_gen_vec_sar16i_i64(a, a, shift);
3963 tcg_gen_vec_add16_i64(d, d, a);
3964}
3965
3966static void gen_ssra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
3967{
3968 tcg_gen_sari_i32(a, a, shift);
3969 tcg_gen_add_i32(d, d, a);
3970}
3971
3972static void gen_ssra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3973{
3974 tcg_gen_sari_i64(a, a, shift);
3975 tcg_gen_add_i64(d, d, a);
3976}
3977
3978static void gen_ssra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3979{
3980 tcg_gen_sari_vec(vece, a, a, sh);
3981 tcg_gen_add_vec(vece, d, d, a);
3982}
3983
3984static const TCGOpcode vecop_list_ssra[] = {
3985 INDEX_op_sari_vec, INDEX_op_add_vec, 0
3986};
3987
3988const GVecGen2i ssra_op[4] = {
3989 { .fni8 = gen_ssra8_i64,
3990 .fniv = gen_ssra_vec,
3991 .load_dest = true,
3992 .opt_opc = vecop_list_ssra,
3993 .vece = MO_8 },
3994 { .fni8 = gen_ssra16_i64,
3995 .fniv = gen_ssra_vec,
3996 .load_dest = true,
3997 .opt_opc = vecop_list_ssra,
3998 .vece = MO_16 },
3999 { .fni4 = gen_ssra32_i32,
4000 .fniv = gen_ssra_vec,
4001 .load_dest = true,
4002 .opt_opc = vecop_list_ssra,
4003 .vece = MO_32 },
4004 { .fni8 = gen_ssra64_i64,
4005 .fniv = gen_ssra_vec,
4006 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4007 .opt_opc = vecop_list_ssra,
4008 .load_dest = true,
4009 .vece = MO_64 },
4010};
4011
4012static void gen_usra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4013{
4014 tcg_gen_vec_shr8i_i64(a, a, shift);
4015 tcg_gen_vec_add8_i64(d, d, a);
4016}
4017
4018static void gen_usra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4019{
4020 tcg_gen_vec_shr16i_i64(a, a, shift);
4021 tcg_gen_vec_add16_i64(d, d, a);
4022}
4023
4024static void gen_usra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
4025{
4026 tcg_gen_shri_i32(a, a, shift);
4027 tcg_gen_add_i32(d, d, a);
4028}
4029
4030static void gen_usra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4031{
4032 tcg_gen_shri_i64(a, a, shift);
4033 tcg_gen_add_i64(d, d, a);
4034}
4035
4036static void gen_usra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
4037{
4038 tcg_gen_shri_vec(vece, a, a, sh);
4039 tcg_gen_add_vec(vece, d, d, a);
4040}
4041
4042static const TCGOpcode vecop_list_usra[] = {
4043 INDEX_op_shri_vec, INDEX_op_add_vec, 0
4044};
4045
4046const GVecGen2i usra_op[4] = {
4047 { .fni8 = gen_usra8_i64,
4048 .fniv = gen_usra_vec,
4049 .load_dest = true,
4050 .opt_opc = vecop_list_usra,
4051 .vece = MO_8, },
4052 { .fni8 = gen_usra16_i64,
4053 .fniv = gen_usra_vec,
4054 .load_dest = true,
4055 .opt_opc = vecop_list_usra,
4056 .vece = MO_16, },
4057 { .fni4 = gen_usra32_i32,
4058 .fniv = gen_usra_vec,
4059 .load_dest = true,
4060 .opt_opc = vecop_list_usra,
4061 .vece = MO_32, },
4062 { .fni8 = gen_usra64_i64,
4063 .fniv = gen_usra_vec,
4064 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4065 .load_dest = true,
4066 .opt_opc = vecop_list_usra,
4067 .vece = MO_64, },
4068};
4069
4070static void gen_shr8_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4071{
4072 uint64_t mask = dup_const(MO_8, 0xff >> shift);
4073 TCGv_i64 t = tcg_temp_new_i64();
4074
4075 tcg_gen_shri_i64(t, a, shift);
4076 tcg_gen_andi_i64(t, t, mask);
4077 tcg_gen_andi_i64(d, d, ~mask);
4078 tcg_gen_or_i64(d, d, t);
4079 tcg_temp_free_i64(t);
4080}
4081
4082static void gen_shr16_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4083{
4084 uint64_t mask = dup_const(MO_16, 0xffff >> shift);
4085 TCGv_i64 t = tcg_temp_new_i64();
4086
4087 tcg_gen_shri_i64(t, a, shift);
4088 tcg_gen_andi_i64(t, t, mask);
4089 tcg_gen_andi_i64(d, d, ~mask);
4090 tcg_gen_or_i64(d, d, t);
4091 tcg_temp_free_i64(t);
4092}
4093
4094static void gen_shr32_ins_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
4095{
4096 tcg_gen_shri_i32(a, a, shift);
4097 tcg_gen_deposit_i32(d, d, a, 0, 32 - shift);
4098}
4099
4100static void gen_shr64_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4101{
4102 tcg_gen_shri_i64(a, a, shift);
4103 tcg_gen_deposit_i64(d, d, a, 0, 64 - shift);
4104}
4105
4106static void gen_shr_ins_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
4107{
4108 if (sh == 0) {
4109 tcg_gen_mov_vec(d, a);
4110 } else {
4111 TCGv_vec t = tcg_temp_new_vec_matching(d);
4112 TCGv_vec m = tcg_temp_new_vec_matching(d);
4113
4114 tcg_gen_dupi_vec(vece, m, MAKE_64BIT_MASK((8 << vece) - sh, sh));
4115 tcg_gen_shri_vec(vece, t, a, sh);
4116 tcg_gen_and_vec(vece, d, d, m);
4117 tcg_gen_or_vec(vece, d, d, t);
4118
4119 tcg_temp_free_vec(t);
4120 tcg_temp_free_vec(m);
4121 }
4122}
4123
4124static const TCGOpcode vecop_list_sri[] = { INDEX_op_shri_vec, 0 };
4125
4126const GVecGen2i sri_op[4] = {
4127 { .fni8 = gen_shr8_ins_i64,
4128 .fniv = gen_shr_ins_vec,
4129 .load_dest = true,
4130 .opt_opc = vecop_list_sri,
4131 .vece = MO_8 },
4132 { .fni8 = gen_shr16_ins_i64,
4133 .fniv = gen_shr_ins_vec,
4134 .load_dest = true,
4135 .opt_opc = vecop_list_sri,
4136 .vece = MO_16 },
4137 { .fni4 = gen_shr32_ins_i32,
4138 .fniv = gen_shr_ins_vec,
4139 .load_dest = true,
4140 .opt_opc = vecop_list_sri,
4141 .vece = MO_32 },
4142 { .fni8 = gen_shr64_ins_i64,
4143 .fniv = gen_shr_ins_vec,
4144 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4145 .load_dest = true,
4146 .opt_opc = vecop_list_sri,
4147 .vece = MO_64 },
4148};
4149
4150static void gen_shl8_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4151{
4152 uint64_t mask = dup_const(MO_8, 0xff << shift);
4153 TCGv_i64 t = tcg_temp_new_i64();
4154
4155 tcg_gen_shli_i64(t, a, shift);
4156 tcg_gen_andi_i64(t, t, mask);
4157 tcg_gen_andi_i64(d, d, ~mask);
4158 tcg_gen_or_i64(d, d, t);
4159 tcg_temp_free_i64(t);
4160}
4161
4162static void gen_shl16_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4163{
4164 uint64_t mask = dup_const(MO_16, 0xffff << shift);
4165 TCGv_i64 t = tcg_temp_new_i64();
4166
4167 tcg_gen_shli_i64(t, a, shift);
4168 tcg_gen_andi_i64(t, t, mask);
4169 tcg_gen_andi_i64(d, d, ~mask);
4170 tcg_gen_or_i64(d, d, t);
4171 tcg_temp_free_i64(t);
4172}
4173
4174static void gen_shl32_ins_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
4175{
4176 tcg_gen_deposit_i32(d, d, a, shift, 32 - shift);
4177}
4178
4179static void gen_shl64_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4180{
4181 tcg_gen_deposit_i64(d, d, a, shift, 64 - shift);
4182}
4183
4184static void gen_shl_ins_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
4185{
4186 if (sh == 0) {
4187 tcg_gen_mov_vec(d, a);
4188 } else {
4189 TCGv_vec t = tcg_temp_new_vec_matching(d);
4190 TCGv_vec m = tcg_temp_new_vec_matching(d);
4191
4192 tcg_gen_dupi_vec(vece, m, MAKE_64BIT_MASK(0, sh));
4193 tcg_gen_shli_vec(vece, t, a, sh);
4194 tcg_gen_and_vec(vece, d, d, m);
4195 tcg_gen_or_vec(vece, d, d, t);
4196
4197 tcg_temp_free_vec(t);
4198 tcg_temp_free_vec(m);
4199 }
4200}
4201
4202static const TCGOpcode vecop_list_sli[] = { INDEX_op_shli_vec, 0 };
4203
4204const GVecGen2i sli_op[4] = {
4205 { .fni8 = gen_shl8_ins_i64,
4206 .fniv = gen_shl_ins_vec,
4207 .load_dest = true,
4208 .opt_opc = vecop_list_sli,
4209 .vece = MO_8 },
4210 { .fni8 = gen_shl16_ins_i64,
4211 .fniv = gen_shl_ins_vec,
4212 .load_dest = true,
4213 .opt_opc = vecop_list_sli,
4214 .vece = MO_16 },
4215 { .fni4 = gen_shl32_ins_i32,
4216 .fniv = gen_shl_ins_vec,
4217 .load_dest = true,
4218 .opt_opc = vecop_list_sli,
4219 .vece = MO_32 },
4220 { .fni8 = gen_shl64_ins_i64,
4221 .fniv = gen_shl_ins_vec,
4222 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4223 .load_dest = true,
4224 .opt_opc = vecop_list_sli,
4225 .vece = MO_64 },
4226};
4227
4228static void gen_mla8_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4229{
4230 gen_helper_neon_mul_u8(a, a, b);
4231 gen_helper_neon_add_u8(d, d, a);
4232}
4233
4234static void gen_mls8_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4235{
4236 gen_helper_neon_mul_u8(a, a, b);
4237 gen_helper_neon_sub_u8(d, d, a);
4238}
4239
4240static void gen_mla16_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4241{
4242 gen_helper_neon_mul_u16(a, a, b);
4243 gen_helper_neon_add_u16(d, d, a);
4244}
4245
4246static void gen_mls16_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4247{
4248 gen_helper_neon_mul_u16(a, a, b);
4249 gen_helper_neon_sub_u16(d, d, a);
4250}
4251
4252static void gen_mla32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4253{
4254 tcg_gen_mul_i32(a, a, b);
4255 tcg_gen_add_i32(d, d, a);
4256}
4257
4258static void gen_mls32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4259{
4260 tcg_gen_mul_i32(a, a, b);
4261 tcg_gen_sub_i32(d, d, a);
4262}
4263
4264static void gen_mla64_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
4265{
4266 tcg_gen_mul_i64(a, a, b);
4267 tcg_gen_add_i64(d, d, a);
4268}
4269
4270static void gen_mls64_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
4271{
4272 tcg_gen_mul_i64(a, a, b);
4273 tcg_gen_sub_i64(d, d, a);
4274}
4275
4276static void gen_mla_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
4277{
4278 tcg_gen_mul_vec(vece, a, a, b);
4279 tcg_gen_add_vec(vece, d, d, a);
4280}
4281
4282static void gen_mls_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
4283{
4284 tcg_gen_mul_vec(vece, a, a, b);
4285 tcg_gen_sub_vec(vece, d, d, a);
4286}
4287
4288/* Note that while NEON does not support VMLA and VMLS as 64-bit ops,
4289 * these tables are shared with AArch64 which does support them.
4290 */
4291
4292static const TCGOpcode vecop_list_mla[] = {
4293 INDEX_op_mul_vec, INDEX_op_add_vec, 0
4294};
4295
4296static const TCGOpcode vecop_list_mls[] = {
4297 INDEX_op_mul_vec, INDEX_op_sub_vec, 0
4298};
4299
4300const GVecGen3 mla_op[4] = {
4301 { .fni4 = gen_mla8_i32,
4302 .fniv = gen_mla_vec,
4303 .load_dest = true,
4304 .opt_opc = vecop_list_mla,
4305 .vece = MO_8 },
4306 { .fni4 = gen_mla16_i32,
4307 .fniv = gen_mla_vec,
4308 .load_dest = true,
4309 .opt_opc = vecop_list_mla,
4310 .vece = MO_16 },
4311 { .fni4 = gen_mla32_i32,
4312 .fniv = gen_mla_vec,
4313 .load_dest = true,
4314 .opt_opc = vecop_list_mla,
4315 .vece = MO_32 },
4316 { .fni8 = gen_mla64_i64,
4317 .fniv = gen_mla_vec,
4318 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4319 .load_dest = true,
4320 .opt_opc = vecop_list_mla,
4321 .vece = MO_64 },
4322};
4323
4324const GVecGen3 mls_op[4] = {
4325 { .fni4 = gen_mls8_i32,
4326 .fniv = gen_mls_vec,
4327 .load_dest = true,
4328 .opt_opc = vecop_list_mls,
4329 .vece = MO_8 },
4330 { .fni4 = gen_mls16_i32,
4331 .fniv = gen_mls_vec,
4332 .load_dest = true,
4333 .opt_opc = vecop_list_mls,
4334 .vece = MO_16 },
4335 { .fni4 = gen_mls32_i32,
4336 .fniv = gen_mls_vec,
4337 .load_dest = true,
4338 .opt_opc = vecop_list_mls,
4339 .vece = MO_32 },
4340 { .fni8 = gen_mls64_i64,
4341 .fniv = gen_mls_vec,
4342 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4343 .load_dest = true,
4344 .opt_opc = vecop_list_mls,
4345 .vece = MO_64 },
4346};
4347
4348/* CMTST : test is "if (X & Y != 0)". */
4349static void gen_cmtst_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4350{
4351 tcg_gen_and_i32(d, a, b);
4352 tcg_gen_setcondi_i32(TCG_COND_NE, d, d, 0);
4353 tcg_gen_neg_i32(d, d);
4354}
4355
4356void gen_cmtst_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
4357{
4358 tcg_gen_and_i64(d, a, b);
4359 tcg_gen_setcondi_i64(TCG_COND_NE, d, d, 0);
4360 tcg_gen_neg_i64(d, d);
4361}
4362
4363static void gen_cmtst_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
4364{
4365 tcg_gen_and_vec(vece, d, a, b);
4366 tcg_gen_dupi_vec(vece, a, 0);
4367 tcg_gen_cmp_vec(TCG_COND_NE, vece, d, d, a);
4368}
4369
4370static const TCGOpcode vecop_list_cmtst[] = { INDEX_op_cmp_vec, 0 };
4371
4372const GVecGen3 cmtst_op[4] = {
4373 { .fni4 = gen_helper_neon_tst_u8,
4374 .fniv = gen_cmtst_vec,
4375 .opt_opc = vecop_list_cmtst,
4376 .vece = MO_8 },
4377 { .fni4 = gen_helper_neon_tst_u16,
4378 .fniv = gen_cmtst_vec,
4379 .opt_opc = vecop_list_cmtst,
4380 .vece = MO_16 },
4381 { .fni4 = gen_cmtst_i32,
4382 .fniv = gen_cmtst_vec,
4383 .opt_opc = vecop_list_cmtst,
4384 .vece = MO_32 },
4385 { .fni8 = gen_cmtst_i64,
4386 .fniv = gen_cmtst_vec,
4387 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4388 .opt_opc = vecop_list_cmtst,
4389 .vece = MO_64 },
4390};
4391
4392static void gen_uqadd_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4393 TCGv_vec a, TCGv_vec b)
4394{
4395 TCGv_vec x = tcg_temp_new_vec_matching(t);
4396 tcg_gen_add_vec(vece, x, a, b);
4397 tcg_gen_usadd_vec(vece, t, a, b);
4398 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4399 tcg_gen_or_vec(vece, sat, sat, x);
4400 tcg_temp_free_vec(x);
4401}
4402
4403static const TCGOpcode vecop_list_uqadd[] = {
4404 INDEX_op_usadd_vec, INDEX_op_cmp_vec, INDEX_op_add_vec, 0
4405};
4406
4407const GVecGen4 uqadd_op[4] = {
4408 { .fniv = gen_uqadd_vec,
4409 .fno = gen_helper_gvec_uqadd_b,
4410 .write_aofs = true,
4411 .opt_opc = vecop_list_uqadd,
4412 .vece = MO_8 },
4413 { .fniv = gen_uqadd_vec,
4414 .fno = gen_helper_gvec_uqadd_h,
4415 .write_aofs = true,
4416 .opt_opc = vecop_list_uqadd,
4417 .vece = MO_16 },
4418 { .fniv = gen_uqadd_vec,
4419 .fno = gen_helper_gvec_uqadd_s,
4420 .write_aofs = true,
4421 .opt_opc = vecop_list_uqadd,
4422 .vece = MO_32 },
4423 { .fniv = gen_uqadd_vec,
4424 .fno = gen_helper_gvec_uqadd_d,
4425 .write_aofs = true,
4426 .opt_opc = vecop_list_uqadd,
4427 .vece = MO_64 },
4428};
4429
4430static void gen_sqadd_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4431 TCGv_vec a, TCGv_vec b)
4432{
4433 TCGv_vec x = tcg_temp_new_vec_matching(t);
4434 tcg_gen_add_vec(vece, x, a, b);
4435 tcg_gen_ssadd_vec(vece, t, a, b);
4436 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4437 tcg_gen_or_vec(vece, sat, sat, x);
4438 tcg_temp_free_vec(x);
4439}
4440
4441static const TCGOpcode vecop_list_sqadd[] = {
4442 INDEX_op_ssadd_vec, INDEX_op_cmp_vec, INDEX_op_add_vec, 0
4443};
4444
4445const GVecGen4 sqadd_op[4] = {
4446 { .fniv = gen_sqadd_vec,
4447 .fno = gen_helper_gvec_sqadd_b,
4448 .opt_opc = vecop_list_sqadd,
4449 .write_aofs = true,
4450 .vece = MO_8 },
4451 { .fniv = gen_sqadd_vec,
4452 .fno = gen_helper_gvec_sqadd_h,
4453 .opt_opc = vecop_list_sqadd,
4454 .write_aofs = true,
4455 .vece = MO_16 },
4456 { .fniv = gen_sqadd_vec,
4457 .fno = gen_helper_gvec_sqadd_s,
4458 .opt_opc = vecop_list_sqadd,
4459 .write_aofs = true,
4460 .vece = MO_32 },
4461 { .fniv = gen_sqadd_vec,
4462 .fno = gen_helper_gvec_sqadd_d,
4463 .opt_opc = vecop_list_sqadd,
4464 .write_aofs = true,
4465 .vece = MO_64 },
4466};
4467
4468static void gen_uqsub_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4469 TCGv_vec a, TCGv_vec b)
4470{
4471 TCGv_vec x = tcg_temp_new_vec_matching(t);
4472 tcg_gen_sub_vec(vece, x, a, b);
4473 tcg_gen_ussub_vec(vece, t, a, b);
4474 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4475 tcg_gen_or_vec(vece, sat, sat, x);
4476 tcg_temp_free_vec(x);
4477}
4478
4479static const TCGOpcode vecop_list_uqsub[] = {
4480 INDEX_op_ussub_vec, INDEX_op_cmp_vec, INDEX_op_sub_vec, 0
4481};
4482
4483const GVecGen4 uqsub_op[4] = {
4484 { .fniv = gen_uqsub_vec,
4485 .fno = gen_helper_gvec_uqsub_b,
4486 .opt_opc = vecop_list_uqsub,
4487 .write_aofs = true,
4488 .vece = MO_8 },
4489 { .fniv = gen_uqsub_vec,
4490 .fno = gen_helper_gvec_uqsub_h,
4491 .opt_opc = vecop_list_uqsub,
4492 .write_aofs = true,
4493 .vece = MO_16 },
4494 { .fniv = gen_uqsub_vec,
4495 .fno = gen_helper_gvec_uqsub_s,
4496 .opt_opc = vecop_list_uqsub,
4497 .write_aofs = true,
4498 .vece = MO_32 },
4499 { .fniv = gen_uqsub_vec,
4500 .fno = gen_helper_gvec_uqsub_d,
4501 .opt_opc = vecop_list_uqsub,
4502 .write_aofs = true,
4503 .vece = MO_64 },
4504};
4505
4506static void gen_sqsub_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4507 TCGv_vec a, TCGv_vec b)
4508{
4509 TCGv_vec x = tcg_temp_new_vec_matching(t);
4510 tcg_gen_sub_vec(vece, x, a, b);
4511 tcg_gen_sssub_vec(vece, t, a, b);
4512 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4513 tcg_gen_or_vec(vece, sat, sat, x);
4514 tcg_temp_free_vec(x);
4515}
4516
4517static const TCGOpcode vecop_list_sqsub[] = {
4518 INDEX_op_sssub_vec, INDEX_op_cmp_vec, INDEX_op_sub_vec, 0
4519};
4520
4521const GVecGen4 sqsub_op[4] = {
4522 { .fniv = gen_sqsub_vec,
4523 .fno = gen_helper_gvec_sqsub_b,
4524 .opt_opc = vecop_list_sqsub,
4525 .write_aofs = true,
4526 .vece = MO_8 },
4527 { .fniv = gen_sqsub_vec,
4528 .fno = gen_helper_gvec_sqsub_h,
4529 .opt_opc = vecop_list_sqsub,
4530 .write_aofs = true,
4531 .vece = MO_16 },
4532 { .fniv = gen_sqsub_vec,
4533 .fno = gen_helper_gvec_sqsub_s,
4534 .opt_opc = vecop_list_sqsub,
4535 .write_aofs = true,
4536 .vece = MO_32 },
4537 { .fniv = gen_sqsub_vec,
4538 .fno = gen_helper_gvec_sqsub_d,
4539 .opt_opc = vecop_list_sqsub,
4540 .write_aofs = true,
4541 .vece = MO_64 },
4542};
4543
4544/* Translate a NEON data processing instruction. Return nonzero if the
4545 instruction is invalid.
4546 We process data in a mixture of 32-bit and 64-bit chunks.
4547 Mostly we use 32-bit chunks so we can use normal scalar instructions. */
4548
4549static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
4550{
4551 int op;
4552 int q;
4553 int rd, rn, rm, rd_ofs, rn_ofs, rm_ofs;
4554 int size;
4555 int shift;
4556 int pass;
4557 int count;
4558 int pairwise;
4559 int u;
4560 int vec_size;
4561 uint32_t imm;
4562 TCGv_i32 tmp, tmp2, tmp3, tmp4, tmp5;
4563 TCGv_ptr ptr1, ptr2, ptr3;
4564 TCGv_i64 tmp64;
4565
4566 /* FIXME: this access check should not take precedence over UNDEF
4567 * for invalid encodings; we will generate incorrect syndrome information
4568 * for attempts to execute invalid vfp/neon encodings with FP disabled.
4569 */
4570 if (s->fp_excp_el) {
4571 gen_exception_insn(s, s->pc_curr, EXCP_UDEF,
4572 syn_simd_access_trap(1, 0xe, false), s->fp_excp_el);
4573 return 0;
4574 }
4575
4576 if (!s->vfp_enabled)
4577 return 1;
4578 q = (insn & (1 << 6)) != 0;
4579 u = (insn >> 24) & 1;
4580 VFP_DREG_D(rd, insn);
4581 VFP_DREG_N(rn, insn);
4582 VFP_DREG_M(rm, insn);
4583 size = (insn >> 20) & 3;
4584 vec_size = q ? 16 : 8;
4585 rd_ofs = neon_reg_offset(rd, 0);
4586 rn_ofs = neon_reg_offset(rn, 0);
4587 rm_ofs = neon_reg_offset(rm, 0);
4588
4589 if ((insn & (1 << 23)) == 0) {
4590 /* Three register same length. */
4591 op = ((insn >> 7) & 0x1e) | ((insn >> 4) & 1);
4592 /* Catch invalid op and bad size combinations: UNDEF */
4593 if ((neon_3r_sizes[op] & (1 << size)) == 0) {
4594 return 1;
4595 }
4596 /* All insns of this form UNDEF for either this condition or the
4597 * superset of cases "Q==1"; we catch the latter later.
4598 */
4599 if (q && ((rd | rn | rm) & 1)) {
4600 return 1;
4601 }
4602 switch (op) {
4603 case NEON_3R_SHA:
4604 /* The SHA-1/SHA-256 3-register instructions require special
4605 * treatment here, as their size field is overloaded as an
4606 * op type selector, and they all consume their input in a
4607 * single pass.
4608 */
4609 if (!q) {
4610 return 1;
4611 }
4612 if (!u) { /* SHA-1 */
4613 if (!dc_isar_feature(aa32_sha1, s)) {
4614 return 1;
4615 }
4616 ptr1 = vfp_reg_ptr(true, rd);
4617 ptr2 = vfp_reg_ptr(true, rn);
4618 ptr3 = vfp_reg_ptr(true, rm);
4619 tmp4 = tcg_const_i32(size);
4620 gen_helper_crypto_sha1_3reg(ptr1, ptr2, ptr3, tmp4);
4621 tcg_temp_free_i32(tmp4);
4622 } else { /* SHA-256 */
4623 if (!dc_isar_feature(aa32_sha2, s) || size == 3) {
4624 return 1;
4625 }
4626 ptr1 = vfp_reg_ptr(true, rd);
4627 ptr2 = vfp_reg_ptr(true, rn);
4628 ptr3 = vfp_reg_ptr(true, rm);
4629 switch (size) {
4630 case 0:
4631 gen_helper_crypto_sha256h(ptr1, ptr2, ptr3);
4632 break;
4633 case 1:
4634 gen_helper_crypto_sha256h2(ptr1, ptr2, ptr3);
4635 break;
4636 case 2:
4637 gen_helper_crypto_sha256su1(ptr1, ptr2, ptr3);
4638 break;
4639 }
4640 }
4641 tcg_temp_free_ptr(ptr1);
4642 tcg_temp_free_ptr(ptr2);
4643 tcg_temp_free_ptr(ptr3);
4644 return 0;
4645
4646 case NEON_3R_VPADD_VQRDMLAH:
4647 if (!u) {
4648 break; /* VPADD */
4649 }
4650 /* VQRDMLAH */
4651 switch (size) {
4652 case 1:
4653 return do_v81_helper(s, gen_helper_gvec_qrdmlah_s16,
4654 q, rd, rn, rm);
4655 case 2:
4656 return do_v81_helper(s, gen_helper_gvec_qrdmlah_s32,
4657 q, rd, rn, rm);
4658 }
4659 return 1;
4660
4661 case NEON_3R_VFM_VQRDMLSH:
4662 if (!u) {
4663 /* VFM, VFMS */
4664 if (size == 1) {
4665 return 1;
4666 }
4667 break;
4668 }
4669 /* VQRDMLSH */
4670 switch (size) {
4671 case 1:
4672 return do_v81_helper(s, gen_helper_gvec_qrdmlsh_s16,
4673 q, rd, rn, rm);
4674 case 2:
4675 return do_v81_helper(s, gen_helper_gvec_qrdmlsh_s32,
4676 q, rd, rn, rm);
4677 }
4678 return 1;
4679
4680 case NEON_3R_LOGIC: /* Logic ops. */
4681 switch ((u << 2) | size) {
4682 case 0: /* VAND */
4683 tcg_gen_gvec_and(0, rd_ofs, rn_ofs, rm_ofs,
4684 vec_size, vec_size);
4685 break;
4686 case 1: /* VBIC */
4687 tcg_gen_gvec_andc(0, rd_ofs, rn_ofs, rm_ofs,
4688 vec_size, vec_size);
4689 break;
4690 case 2: /* VORR */
4691 tcg_gen_gvec_or(0, rd_ofs, rn_ofs, rm_ofs,
4692 vec_size, vec_size);
4693 break;
4694 case 3: /* VORN */
4695 tcg_gen_gvec_orc(0, rd_ofs, rn_ofs, rm_ofs,
4696 vec_size, vec_size);
4697 break;
4698 case 4: /* VEOR */
4699 tcg_gen_gvec_xor(0, rd_ofs, rn_ofs, rm_ofs,
4700 vec_size, vec_size);
4701 break;
4702 case 5: /* VBSL */
4703 tcg_gen_gvec_bitsel(MO_8, rd_ofs, rd_ofs, rn_ofs, rm_ofs,
4704 vec_size, vec_size);
4705 break;
4706 case 6: /* VBIT */
4707 tcg_gen_gvec_bitsel(MO_8, rd_ofs, rm_ofs, rn_ofs, rd_ofs,
4708 vec_size, vec_size);
4709 break;
4710 case 7: /* VBIF */
4711 tcg_gen_gvec_bitsel(MO_8, rd_ofs, rm_ofs, rd_ofs, rn_ofs,
4712 vec_size, vec_size);
4713 break;
4714 }
4715 return 0;
4716
4717 case NEON_3R_VADD_VSUB:
4718 if (u) {
4719 tcg_gen_gvec_sub(size, rd_ofs, rn_ofs, rm_ofs,
4720 vec_size, vec_size);
4721 } else {
4722 tcg_gen_gvec_add(size, rd_ofs, rn_ofs, rm_ofs,
4723 vec_size, vec_size);
4724 }
4725 return 0;
4726
4727 case NEON_3R_VQADD:
4728 tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc),
4729 rn_ofs, rm_ofs, vec_size, vec_size,
4730 (u ? uqadd_op : sqadd_op) + size);
4731 return 0;
4732
4733 case NEON_3R_VQSUB:
4734 tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc),
4735 rn_ofs, rm_ofs, vec_size, vec_size,
4736 (u ? uqsub_op : sqsub_op) + size);
4737 return 0;
4738
4739 case NEON_3R_VMUL: /* VMUL */
4740 if (u) {
4741 /* Polynomial case allows only P8 and is handled below. */
4742 if (size != 0) {
4743 return 1;
4744 }
4745 } else {
4746 tcg_gen_gvec_mul(size, rd_ofs, rn_ofs, rm_ofs,
4747 vec_size, vec_size);
4748 return 0;
4749 }
4750 break;
4751
4752 case NEON_3R_VML: /* VMLA, VMLS */
4753 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, vec_size, vec_size,
4754 u ? &mls_op[size] : &mla_op[size]);
4755 return 0;
4756
4757 case NEON_3R_VTST_VCEQ:
4758 if (u) { /* VCEQ */
4759 tcg_gen_gvec_cmp(TCG_COND_EQ, size, rd_ofs, rn_ofs, rm_ofs,
4760 vec_size, vec_size);
4761 } else { /* VTST */
4762 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs,
4763 vec_size, vec_size, &cmtst_op[size]);
4764 }
4765 return 0;
4766
4767 case NEON_3R_VCGT:
4768 tcg_gen_gvec_cmp(u ? TCG_COND_GTU : TCG_COND_GT, size,
4769 rd_ofs, rn_ofs, rm_ofs, vec_size, vec_size);
4770 return 0;
4771
4772 case NEON_3R_VCGE:
4773 tcg_gen_gvec_cmp(u ? TCG_COND_GEU : TCG_COND_GE, size,
4774 rd_ofs, rn_ofs, rm_ofs, vec_size, vec_size);
4775 return 0;
4776
4777 case NEON_3R_VMAX:
4778 if (u) {
4779 tcg_gen_gvec_umax(size, rd_ofs, rn_ofs, rm_ofs,
4780 vec_size, vec_size);
4781 } else {
4782 tcg_gen_gvec_smax(size, rd_ofs, rn_ofs, rm_ofs,
4783 vec_size, vec_size);
4784 }
4785 return 0;
4786 case NEON_3R_VMIN:
4787 if (u) {
4788 tcg_gen_gvec_umin(size, rd_ofs, rn_ofs, rm_ofs,
4789 vec_size, vec_size);
4790 } else {
4791 tcg_gen_gvec_smin(size, rd_ofs, rn_ofs, rm_ofs,
4792 vec_size, vec_size);
4793 }
4794 return 0;
4795 }
4796
4797 if (size == 3) {
4798 /* 64-bit element instructions. */
4799 for (pass = 0; pass < (q ? 2 : 1); pass++) {
4800 neon_load_reg64(cpu_V0, rn + pass);
4801 neon_load_reg64(cpu_V1, rm + pass);
4802 switch (op) {
4803 case NEON_3R_VSHL:
4804 if (u) {
4805 gen_helper_neon_shl_u64(cpu_V0, cpu_V1, cpu_V0);
4806 } else {
4807 gen_helper_neon_shl_s64(cpu_V0, cpu_V1, cpu_V0);
4808 }
4809 break;
4810 case NEON_3R_VQSHL:
4811 if (u) {
4812 gen_helper_neon_qshl_u64(cpu_V0, cpu_env,
4813 cpu_V1, cpu_V0);
4814 } else {
4815 gen_helper_neon_qshl_s64(cpu_V0, cpu_env,
4816 cpu_V1, cpu_V0);
4817 }
4818 break;
4819 case NEON_3R_VRSHL:
4820 if (u) {
4821 gen_helper_neon_rshl_u64(cpu_V0, cpu_V1, cpu_V0);
4822 } else {
4823 gen_helper_neon_rshl_s64(cpu_V0, cpu_V1, cpu_V0);
4824 }
4825 break;
4826 case NEON_3R_VQRSHL:
4827 if (u) {
4828 gen_helper_neon_qrshl_u64(cpu_V0, cpu_env,
4829 cpu_V1, cpu_V0);
4830 } else {
4831 gen_helper_neon_qrshl_s64(cpu_V0, cpu_env,
4832 cpu_V1, cpu_V0);
4833 }
4834 break;
4835 default:
4836 abort();
4837 }
4838 neon_store_reg64(cpu_V0, rd + pass);
4839 }
4840 return 0;
4841 }
4842 pairwise = 0;
4843 switch (op) {
4844 case NEON_3R_VSHL:
4845 case NEON_3R_VQSHL:
4846 case NEON_3R_VRSHL:
4847 case NEON_3R_VQRSHL:
4848 {
4849 int rtmp;
4850 /* Shift instruction operands are reversed. */
4851 rtmp = rn;
4852 rn = rm;
4853 rm = rtmp;
4854 }
4855 break;
4856 case NEON_3R_VPADD_VQRDMLAH:
4857 case NEON_3R_VPMAX:
4858 case NEON_3R_VPMIN:
4859 pairwise = 1;
4860 break;
4861 case NEON_3R_FLOAT_ARITH:
4862 pairwise = (u && size < 2); /* if VPADD (float) */
4863 break;
4864 case NEON_3R_FLOAT_MINMAX:
4865 pairwise = u; /* if VPMIN/VPMAX (float) */
4866 break;
4867 case NEON_3R_FLOAT_CMP:
4868 if (!u && size) {
4869 /* no encoding for U=0 C=1x */
4870 return 1;
4871 }
4872 break;
4873 case NEON_3R_FLOAT_ACMP:
4874 if (!u) {
4875 return 1;
4876 }
4877 break;
4878 case NEON_3R_FLOAT_MISC:
4879 /* VMAXNM/VMINNM in ARMv8 */
4880 if (u && !arm_dc_feature(s, ARM_FEATURE_V8)) {
4881 return 1;
4882 }
4883 break;
4884 case NEON_3R_VFM_VQRDMLSH:
4885 if (!arm_dc_feature(s, ARM_FEATURE_VFP4)) {
4886 return 1;
4887 }
4888 break;
4889 default:
4890 break;
4891 }
4892
4893 if (pairwise && q) {
4894 /* All the pairwise insns UNDEF if Q is set */
4895 return 1;
4896 }
4897
4898 for (pass = 0; pass < (q ? 4 : 2); pass++) {
4899
4900 if (pairwise) {
4901 /* Pairwise. */
4902 if (pass < 1) {
4903 tmp = neon_load_reg(rn, 0);
4904 tmp2 = neon_load_reg(rn, 1);
4905 } else {
4906 tmp = neon_load_reg(rm, 0);
4907 tmp2 = neon_load_reg(rm, 1);
4908 }
4909 } else {
4910 /* Elementwise. */
4911 tmp = neon_load_reg(rn, pass);
4912 tmp2 = neon_load_reg(rm, pass);
4913 }
4914 switch (op) {
4915 case NEON_3R_VHADD:
4916 GEN_NEON_INTEGER_OP(hadd);
4917 break;
4918 case NEON_3R_VRHADD:
4919 GEN_NEON_INTEGER_OP(rhadd);
4920 break;
4921 case NEON_3R_VHSUB:
4922 GEN_NEON_INTEGER_OP(hsub);
4923 break;
4924 case NEON_3R_VSHL:
4925 GEN_NEON_INTEGER_OP(shl);
4926 break;
4927 case NEON_3R_VQSHL:
4928 GEN_NEON_INTEGER_OP_ENV(qshl);
4929 break;
4930 case NEON_3R_VRSHL:
4931 GEN_NEON_INTEGER_OP(rshl);
4932 break;
4933 case NEON_3R_VQRSHL:
4934 GEN_NEON_INTEGER_OP_ENV(qrshl);
4935 break;
4936 case NEON_3R_VABD:
4937 GEN_NEON_INTEGER_OP(abd);
4938 break;
4939 case NEON_3R_VABA:
4940 GEN_NEON_INTEGER_OP(abd);
4941 tcg_temp_free_i32(tmp2);
4942 tmp2 = neon_load_reg(rd, pass);
4943 gen_neon_add(size, tmp, tmp2);
4944 break;
4945 case NEON_3R_VMUL:
4946 /* VMUL.P8; other cases already eliminated. */
4947 gen_helper_neon_mul_p8(tmp, tmp, tmp2);
4948 break;
4949 case NEON_3R_VPMAX:
4950 GEN_NEON_INTEGER_OP(pmax);
4951 break;
4952 case NEON_3R_VPMIN:
4953 GEN_NEON_INTEGER_OP(pmin);
4954 break;
4955 case NEON_3R_VQDMULH_VQRDMULH: /* Multiply high. */
4956 if (!u) { /* VQDMULH */
4957 switch (size) {
4958 case 1:
4959 gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2);
4960 break;
4961 case 2:
4962 gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2);
4963 break;
4964 default: abort();
4965 }
4966 } else { /* VQRDMULH */
4967 switch (size) {
4968 case 1:
4969 gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2);
4970 break;
4971 case 2:
4972 gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2);
4973 break;
4974 default: abort();
4975 }
4976 }
4977 break;
4978 case NEON_3R_VPADD_VQRDMLAH:
4979 switch (size) {
4980 case 0: gen_helper_neon_padd_u8(tmp, tmp, tmp2); break;
4981 case 1: gen_helper_neon_padd_u16(tmp, tmp, tmp2); break;
4982 case 2: tcg_gen_add_i32(tmp, tmp, tmp2); break;
4983 default: abort();
4984 }
4985 break;
4986 case NEON_3R_FLOAT_ARITH: /* Floating point arithmetic. */
4987 {
4988 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
4989 switch ((u << 2) | size) {
4990 case 0: /* VADD */
4991 case 4: /* VPADD */
4992 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
4993 break;
4994 case 2: /* VSUB */
4995 gen_helper_vfp_subs(tmp, tmp, tmp2, fpstatus);
4996 break;
4997 case 6: /* VABD */
4998 gen_helper_neon_abd_f32(tmp, tmp, tmp2, fpstatus);
4999 break;
5000 default:
5001 abort();
5002 }
5003 tcg_temp_free_ptr(fpstatus);
5004 break;
5005 }
5006 case NEON_3R_FLOAT_MULTIPLY:
5007 {
5008 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5009 gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus);
5010 if (!u) {
5011 tcg_temp_free_i32(tmp2);
5012 tmp2 = neon_load_reg(rd, pass);
5013 if (size == 0) {
5014 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
5015 } else {
5016 gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus);
5017 }
5018 }
5019 tcg_temp_free_ptr(fpstatus);
5020 break;
5021 }
5022 case NEON_3R_FLOAT_CMP:
5023 {
5024 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5025 if (!u) {
5026 gen_helper_neon_ceq_f32(tmp, tmp, tmp2, fpstatus);
5027 } else {
5028 if (size == 0) {
5029 gen_helper_neon_cge_f32(tmp, tmp, tmp2, fpstatus);
5030 } else {
5031 gen_helper_neon_cgt_f32(tmp, tmp, tmp2, fpstatus);
5032 }
5033 }
5034 tcg_temp_free_ptr(fpstatus);
5035 break;
5036 }
5037 case NEON_3R_FLOAT_ACMP:
5038 {
5039 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5040 if (size == 0) {
5041 gen_helper_neon_acge_f32(tmp, tmp, tmp2, fpstatus);
5042 } else {
5043 gen_helper_neon_acgt_f32(tmp, tmp, tmp2, fpstatus);
5044 }
5045 tcg_temp_free_ptr(fpstatus);
5046 break;
5047 }
5048 case NEON_3R_FLOAT_MINMAX:
5049 {
5050 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5051 if (size == 0) {
5052 gen_helper_vfp_maxs(tmp, tmp, tmp2, fpstatus);
5053 } else {
5054 gen_helper_vfp_mins(tmp, tmp, tmp2, fpstatus);
5055 }
5056 tcg_temp_free_ptr(fpstatus);
5057 break;
5058 }
5059 case NEON_3R_FLOAT_MISC:
5060 if (u) {
5061 /* VMAXNM/VMINNM */
5062 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5063 if (size == 0) {
5064 gen_helper_vfp_maxnums(tmp, tmp, tmp2, fpstatus);
5065 } else {
5066 gen_helper_vfp_minnums(tmp, tmp, tmp2, fpstatus);
5067 }
5068 tcg_temp_free_ptr(fpstatus);
5069 } else {
5070 if (size == 0) {
5071 gen_helper_recps_f32(tmp, tmp, tmp2, cpu_env);
5072 } else {
5073 gen_helper_rsqrts_f32(tmp, tmp, tmp2, cpu_env);
5074 }
5075 }
5076 break;
5077 case NEON_3R_VFM_VQRDMLSH:
5078 {
5079 /* VFMA, VFMS: fused multiply-add */
5080 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5081 TCGv_i32 tmp3 = neon_load_reg(rd, pass);
5082 if (size) {
5083 /* VFMS */
5084 gen_helper_vfp_negs(tmp, tmp);
5085 }
5086 gen_helper_vfp_muladds(tmp, tmp, tmp2, tmp3, fpstatus);
5087 tcg_temp_free_i32(tmp3);
5088 tcg_temp_free_ptr(fpstatus);
5089 break;
5090 }
5091 default:
5092 abort();
5093 }
5094 tcg_temp_free_i32(tmp2);
5095
5096 /* Save the result. For elementwise operations we can put it
5097 straight into the destination register. For pairwise operations
5098 we have to be careful to avoid clobbering the source operands. */
5099 if (pairwise && rd == rm) {
5100 neon_store_scratch(pass, tmp);
5101 } else {
5102 neon_store_reg(rd, pass, tmp);
5103 }
5104
5105 } /* for pass */
5106 if (pairwise && rd == rm) {
5107 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5108 tmp = neon_load_scratch(pass);
5109 neon_store_reg(rd, pass, tmp);
5110 }
5111 }
5112 /* End of 3 register same size operations. */
5113 } else if (insn & (1 << 4)) {
5114 if ((insn & 0x00380080) != 0) {
5115 /* Two registers and shift. */
5116 op = (insn >> 8) & 0xf;
5117 if (insn & (1 << 7)) {
5118 /* 64-bit shift. */
5119 if (op > 7) {
5120 return 1;
5121 }
5122 size = 3;
5123 } else {
5124 size = 2;
5125 while ((insn & (1 << (size + 19))) == 0)
5126 size--;
5127 }
5128 shift = (insn >> 16) & ((1 << (3 + size)) - 1);
5129 if (op < 8) {
5130 /* Shift by immediate:
5131 VSHR, VSRA, VRSHR, VRSRA, VSRI, VSHL, VQSHL, VQSHLU. */
5132 if (q && ((rd | rm) & 1)) {
5133 return 1;
5134 }
5135 if (!u && (op == 4 || op == 6)) {
5136 return 1;
5137 }
5138 /* Right shifts are encoded as N - shift, where N is the
5139 element size in bits. */
5140 if (op <= 4) {
5141 shift = shift - (1 << (size + 3));
5142 }
5143
5144 switch (op) {
5145 case 0: /* VSHR */
5146 /* Right shift comes here negative. */
5147 shift = -shift;
5148 /* Shifts larger than the element size are architecturally
5149 * valid. Unsigned results in all zeros; signed results
5150 * in all sign bits.
5151 */
5152 if (!u) {
5153 tcg_gen_gvec_sari(size, rd_ofs, rm_ofs,
5154 MIN(shift, (8 << size) - 1),
5155 vec_size, vec_size);
5156 } else if (shift >= 8 << size) {
5157 tcg_gen_gvec_dup8i(rd_ofs, vec_size, vec_size, 0);
5158 } else {
5159 tcg_gen_gvec_shri(size, rd_ofs, rm_ofs, shift,
5160 vec_size, vec_size);
5161 }
5162 return 0;
5163
5164 case 1: /* VSRA */
5165 /* Right shift comes here negative. */
5166 shift = -shift;
5167 /* Shifts larger than the element size are architecturally
5168 * valid. Unsigned results in all zeros; signed results
5169 * in all sign bits.
5170 */
5171 if (!u) {
5172 tcg_gen_gvec_2i(rd_ofs, rm_ofs, vec_size, vec_size,
5173 MIN(shift, (8 << size) - 1),
5174 &ssra_op[size]);
5175 } else if (shift >= 8 << size) {
5176 /* rd += 0 */
5177 } else {
5178 tcg_gen_gvec_2i(rd_ofs, rm_ofs, vec_size, vec_size,
5179 shift, &usra_op[size]);
5180 }
5181 return 0;
5182
5183 case 4: /* VSRI */
5184 if (!u) {
5185 return 1;
5186 }
5187 /* Right shift comes here negative. */
5188 shift = -shift;
5189 /* Shift out of range leaves destination unchanged. */
5190 if (shift < 8 << size) {
5191 tcg_gen_gvec_2i(rd_ofs, rm_ofs, vec_size, vec_size,
5192 shift, &sri_op[size]);
5193 }
5194 return 0;
5195
5196 case 5: /* VSHL, VSLI */
5197 if (u) { /* VSLI */
5198 /* Shift out of range leaves destination unchanged. */
5199 if (shift < 8 << size) {
5200 tcg_gen_gvec_2i(rd_ofs, rm_ofs, vec_size,
5201 vec_size, shift, &sli_op[size]);
5202 }
5203 } else { /* VSHL */
5204 /* Shifts larger than the element size are
5205 * architecturally valid and results in zero.
5206 */
5207 if (shift >= 8 << size) {
5208 tcg_gen_gvec_dup8i(rd_ofs, vec_size, vec_size, 0);
5209 } else {
5210 tcg_gen_gvec_shli(size, rd_ofs, rm_ofs, shift,
5211 vec_size, vec_size);
5212 }
5213 }
5214 return 0;
5215 }
5216
5217 if (size == 3) {
5218 count = q + 1;
5219 } else {
5220 count = q ? 4: 2;
5221 }
5222
5223 /* To avoid excessive duplication of ops we implement shift
5224 * by immediate using the variable shift operations.
5225 */
5226 imm = dup_const(size, shift);
5227
5228 for (pass = 0; pass < count; pass++) {
5229 if (size == 3) {
5230 neon_load_reg64(cpu_V0, rm + pass);
5231 tcg_gen_movi_i64(cpu_V1, imm);
5232 switch (op) {
5233 case 2: /* VRSHR */
5234 case 3: /* VRSRA */
5235 if (u)
5236 gen_helper_neon_rshl_u64(cpu_V0, cpu_V0, cpu_V1);
5237 else
5238 gen_helper_neon_rshl_s64(cpu_V0, cpu_V0, cpu_V1);
5239 break;
5240 case 6: /* VQSHLU */
5241 gen_helper_neon_qshlu_s64(cpu_V0, cpu_env,
5242 cpu_V0, cpu_V1);
5243 break;
5244 case 7: /* VQSHL */
5245 if (u) {
5246 gen_helper_neon_qshl_u64(cpu_V0, cpu_env,
5247 cpu_V0, cpu_V1);
5248 } else {
5249 gen_helper_neon_qshl_s64(cpu_V0, cpu_env,
5250 cpu_V0, cpu_V1);
5251 }
5252 break;
5253 default:
5254 g_assert_not_reached();
5255 }
5256 if (op == 3) {
5257 /* Accumulate. */
5258 neon_load_reg64(cpu_V1, rd + pass);
5259 tcg_gen_add_i64(cpu_V0, cpu_V0, cpu_V1);
5260 }
5261 neon_store_reg64(cpu_V0, rd + pass);
5262 } else { /* size < 3 */
5263 /* Operands in T0 and T1. */
5264 tmp = neon_load_reg(rm, pass);
5265 tmp2 = tcg_temp_new_i32();
5266 tcg_gen_movi_i32(tmp2, imm);
5267 switch (op) {
5268 case 2: /* VRSHR */
5269 case 3: /* VRSRA */
5270 GEN_NEON_INTEGER_OP(rshl);
5271 break;
5272 case 6: /* VQSHLU */
5273 switch (size) {
5274 case 0:
5275 gen_helper_neon_qshlu_s8(tmp, cpu_env,
5276 tmp, tmp2);
5277 break;
5278 case 1:
5279 gen_helper_neon_qshlu_s16(tmp, cpu_env,
5280 tmp, tmp2);
5281 break;
5282 case 2:
5283 gen_helper_neon_qshlu_s32(tmp, cpu_env,
5284 tmp, tmp2);
5285 break;
5286 default:
5287 abort();
5288 }
5289 break;
5290 case 7: /* VQSHL */
5291 GEN_NEON_INTEGER_OP_ENV(qshl);
5292 break;
5293 default:
5294 g_assert_not_reached();
5295 }
5296 tcg_temp_free_i32(tmp2);
5297
5298 if (op == 3) {
5299 /* Accumulate. */
5300 tmp2 = neon_load_reg(rd, pass);
5301 gen_neon_add(size, tmp, tmp2);
5302 tcg_temp_free_i32(tmp2);
5303 }
5304 neon_store_reg(rd, pass, tmp);
5305 }
5306 } /* for pass */
5307 } else if (op < 10) {
5308 /* Shift by immediate and narrow:
5309 VSHRN, VRSHRN, VQSHRN, VQRSHRN. */
5310 int input_unsigned = (op == 8) ? !u : u;
5311 if (rm & 1) {
5312 return 1;
5313 }
5314 shift = shift - (1 << (size + 3));
5315 size++;
5316 if (size == 3) {
5317 tmp64 = tcg_const_i64(shift);
5318 neon_load_reg64(cpu_V0, rm);
5319 neon_load_reg64(cpu_V1, rm + 1);
5320 for (pass = 0; pass < 2; pass++) {
5321 TCGv_i64 in;
5322 if (pass == 0) {
5323 in = cpu_V0;
5324 } else {
5325 in = cpu_V1;
5326 }
5327 if (q) {
5328 if (input_unsigned) {
5329 gen_helper_neon_rshl_u64(cpu_V0, in, tmp64);
5330 } else {
5331 gen_helper_neon_rshl_s64(cpu_V0, in, tmp64);
5332 }
5333 } else {
5334 if (input_unsigned) {
5335 gen_helper_neon_shl_u64(cpu_V0, in, tmp64);
5336 } else {
5337 gen_helper_neon_shl_s64(cpu_V0, in, tmp64);
5338 }
5339 }
5340 tmp = tcg_temp_new_i32();
5341 gen_neon_narrow_op(op == 8, u, size - 1, tmp, cpu_V0);
5342 neon_store_reg(rd, pass, tmp);
5343 } /* for pass */
5344 tcg_temp_free_i64(tmp64);
5345 } else {
5346 if (size == 1) {
5347 imm = (uint16_t)shift;
5348 imm |= imm << 16;
5349 } else {
5350 /* size == 2 */
5351 imm = (uint32_t)shift;
5352 }
5353 tmp2 = tcg_const_i32(imm);
5354 tmp4 = neon_load_reg(rm + 1, 0);
5355 tmp5 = neon_load_reg(rm + 1, 1);
5356 for (pass = 0; pass < 2; pass++) {
5357 if (pass == 0) {
5358 tmp = neon_load_reg(rm, 0);
5359 } else {
5360 tmp = tmp4;
5361 }
5362 gen_neon_shift_narrow(size, tmp, tmp2, q,
5363 input_unsigned);
5364 if (pass == 0) {
5365 tmp3 = neon_load_reg(rm, 1);
5366 } else {
5367 tmp3 = tmp5;
5368 }
5369 gen_neon_shift_narrow(size, tmp3, tmp2, q,
5370 input_unsigned);
5371 tcg_gen_concat_i32_i64(cpu_V0, tmp, tmp3);
5372 tcg_temp_free_i32(tmp);
5373 tcg_temp_free_i32(tmp3);
5374 tmp = tcg_temp_new_i32();
5375 gen_neon_narrow_op(op == 8, u, size - 1, tmp, cpu_V0);
5376 neon_store_reg(rd, pass, tmp);
5377 } /* for pass */
5378 tcg_temp_free_i32(tmp2);
5379 }
5380 } else if (op == 10) {
5381 /* VSHLL, VMOVL */
5382 if (q || (rd & 1)) {
5383 return 1;
5384 }
5385 tmp = neon_load_reg(rm, 0);
5386 tmp2 = neon_load_reg(rm, 1);
5387 for (pass = 0; pass < 2; pass++) {
5388 if (pass == 1)
5389 tmp = tmp2;
5390
5391 gen_neon_widen(cpu_V0, tmp, size, u);
5392
5393 if (shift != 0) {
5394 /* The shift is less than the width of the source
5395 type, so we can just shift the whole register. */
5396 tcg_gen_shli_i64(cpu_V0, cpu_V0, shift);
5397 /* Widen the result of shift: we need to clear
5398 * the potential overflow bits resulting from
5399 * left bits of the narrow input appearing as
5400 * right bits of left the neighbour narrow
5401 * input. */
5402 if (size < 2 || !u) {
5403 uint64_t imm64;
5404 if (size == 0) {
5405 imm = (0xffu >> (8 - shift));
5406 imm |= imm << 16;
5407 } else if (size == 1) {
5408 imm = 0xffff >> (16 - shift);
5409 } else {
5410 /* size == 2 */
5411 imm = 0xffffffff >> (32 - shift);
5412 }
5413 if (size < 2) {
5414 imm64 = imm | (((uint64_t)imm) << 32);
5415 } else {
5416 imm64 = imm;
5417 }
5418 tcg_gen_andi_i64(cpu_V0, cpu_V0, ~imm64);
5419 }
5420 }
5421 neon_store_reg64(cpu_V0, rd + pass);
5422 }
5423 } else if (op >= 14) {
5424 /* VCVT fixed-point. */
5425 TCGv_ptr fpst;
5426 TCGv_i32 shiftv;
5427 VFPGenFixPointFn *fn;
5428
5429 if (!(insn & (1 << 21)) || (q && ((rd | rm) & 1))) {
5430 return 1;
5431 }
5432
5433 if (!(op & 1)) {
5434 if (u) {
5435 fn = gen_helper_vfp_ultos;
5436 } else {
5437 fn = gen_helper_vfp_sltos;
5438 }
5439 } else {
5440 if (u) {
5441 fn = gen_helper_vfp_touls_round_to_zero;
5442 } else {
5443 fn = gen_helper_vfp_tosls_round_to_zero;
5444 }
5445 }
5446
5447 /* We have already masked out the must-be-1 top bit of imm6,
5448 * hence this 32-shift where the ARM ARM has 64-imm6.
5449 */
5450 shift = 32 - shift;
5451 fpst = get_fpstatus_ptr(1);
5452 shiftv = tcg_const_i32(shift);
5453 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5454 TCGv_i32 tmpf = neon_load_reg(rm, pass);
5455 fn(tmpf, tmpf, shiftv, fpst);
5456 neon_store_reg(rd, pass, tmpf);
5457 }
5458 tcg_temp_free_ptr(fpst);
5459 tcg_temp_free_i32(shiftv);
5460 } else {
5461 return 1;
5462 }
5463 } else { /* (insn & 0x00380080) == 0 */
5464 int invert, reg_ofs, vec_size;
5465
5466 if (q && (rd & 1)) {
5467 return 1;
5468 }
5469
5470 op = (insn >> 8) & 0xf;
5471 /* One register and immediate. */
5472 imm = (u << 7) | ((insn >> 12) & 0x70) | (insn & 0xf);
5473 invert = (insn & (1 << 5)) != 0;
5474 /* Note that op = 2,3,4,5,6,7,10,11,12,13 imm=0 is UNPREDICTABLE.
5475 * We choose to not special-case this and will behave as if a
5476 * valid constant encoding of 0 had been given.
5477 */
5478 switch (op) {
5479 case 0: case 1:
5480 /* no-op */
5481 break;
5482 case 2: case 3:
5483 imm <<= 8;
5484 break;
5485 case 4: case 5:
5486 imm <<= 16;
5487 break;
5488 case 6: case 7:
5489 imm <<= 24;
5490 break;
5491 case 8: case 9:
5492 imm |= imm << 16;
5493 break;
5494 case 10: case 11:
5495 imm = (imm << 8) | (imm << 24);
5496 break;
5497 case 12:
5498 imm = (imm << 8) | 0xff;
5499 break;
5500 case 13:
5501 imm = (imm << 16) | 0xffff;
5502 break;
5503 case 14:
5504 imm |= (imm << 8) | (imm << 16) | (imm << 24);
5505 if (invert) {
5506 imm = ~imm;
5507 }
5508 break;
5509 case 15:
5510 if (invert) {
5511 return 1;
5512 }
5513 imm = ((imm & 0x80) << 24) | ((imm & 0x3f) << 19)
5514 | ((imm & 0x40) ? (0x1f << 25) : (1 << 30));
5515 break;
5516 }
5517 if (invert) {
5518 imm = ~imm;
5519 }
5520
5521 reg_ofs = neon_reg_offset(rd, 0);
5522 vec_size = q ? 16 : 8;
5523
5524 if (op & 1 && op < 12) {
5525 if (invert) {
5526 /* The immediate value has already been inverted,
5527 * so BIC becomes AND.
5528 */
5529 tcg_gen_gvec_andi(MO_32, reg_ofs, reg_ofs, imm,
5530 vec_size, vec_size);
5531 } else {
5532 tcg_gen_gvec_ori(MO_32, reg_ofs, reg_ofs, imm,
5533 vec_size, vec_size);
5534 }
5535 } else {
5536 /* VMOV, VMVN. */
5537 if (op == 14 && invert) {
5538 TCGv_i64 t64 = tcg_temp_new_i64();
5539
5540 for (pass = 0; pass <= q; ++pass) {
5541 uint64_t val = 0;
5542 int n;
5543
5544 for (n = 0; n < 8; n++) {
5545 if (imm & (1 << (n + pass * 8))) {
5546 val |= 0xffull << (n * 8);
5547 }
5548 }
5549 tcg_gen_movi_i64(t64, val);
5550 neon_store_reg64(t64, rd + pass);
5551 }
5552 tcg_temp_free_i64(t64);
5553 } else {
5554 tcg_gen_gvec_dup32i(reg_ofs, vec_size, vec_size, imm);
5555 }
5556 }
5557 }
5558 } else { /* (insn & 0x00800010 == 0x00800000) */
5559 if (size != 3) {
5560 op = (insn >> 8) & 0xf;
5561 if ((insn & (1 << 6)) == 0) {
5562 /* Three registers of different lengths. */
5563 int src1_wide;
5564 int src2_wide;
5565 int prewiden;
5566 /* undefreq: bit 0 : UNDEF if size == 0
5567 * bit 1 : UNDEF if size == 1
5568 * bit 2 : UNDEF if size == 2
5569 * bit 3 : UNDEF if U == 1
5570 * Note that [2:0] set implies 'always UNDEF'
5571 */
5572 int undefreq;
5573 /* prewiden, src1_wide, src2_wide, undefreq */
5574 static const int neon_3reg_wide[16][4] = {
5575 {1, 0, 0, 0}, /* VADDL */
5576 {1, 1, 0, 0}, /* VADDW */
5577 {1, 0, 0, 0}, /* VSUBL */
5578 {1, 1, 0, 0}, /* VSUBW */
5579 {0, 1, 1, 0}, /* VADDHN */
5580 {0, 0, 0, 0}, /* VABAL */
5581 {0, 1, 1, 0}, /* VSUBHN */
5582 {0, 0, 0, 0}, /* VABDL */
5583 {0, 0, 0, 0}, /* VMLAL */
5584 {0, 0, 0, 9}, /* VQDMLAL */
5585 {0, 0, 0, 0}, /* VMLSL */
5586 {0, 0, 0, 9}, /* VQDMLSL */
5587 {0, 0, 0, 0}, /* Integer VMULL */
5588 {0, 0, 0, 1}, /* VQDMULL */
5589 {0, 0, 0, 0xa}, /* Polynomial VMULL */
5590 {0, 0, 0, 7}, /* Reserved: always UNDEF */
5591 };
5592
5593 prewiden = neon_3reg_wide[op][0];
5594 src1_wide = neon_3reg_wide[op][1];
5595 src2_wide = neon_3reg_wide[op][2];
5596 undefreq = neon_3reg_wide[op][3];
5597
5598 if ((undefreq & (1 << size)) ||
5599 ((undefreq & 8) && u)) {
5600 return 1;
5601 }
5602 if ((src1_wide && (rn & 1)) ||
5603 (src2_wide && (rm & 1)) ||
5604 (!src2_wide && (rd & 1))) {
5605 return 1;
5606 }
5607
5608 /* Handle VMULL.P64 (Polynomial 64x64 to 128 bit multiply)
5609 * outside the loop below as it only performs a single pass.
5610 */
5611 if (op == 14 && size == 2) {
5612 TCGv_i64 tcg_rn, tcg_rm, tcg_rd;
5613
5614 if (!dc_isar_feature(aa32_pmull, s)) {
5615 return 1;
5616 }
5617 tcg_rn = tcg_temp_new_i64();
5618 tcg_rm = tcg_temp_new_i64();
5619 tcg_rd = tcg_temp_new_i64();
5620 neon_load_reg64(tcg_rn, rn);
5621 neon_load_reg64(tcg_rm, rm);
5622 gen_helper_neon_pmull_64_lo(tcg_rd, tcg_rn, tcg_rm);
5623 neon_store_reg64(tcg_rd, rd);
5624 gen_helper_neon_pmull_64_hi(tcg_rd, tcg_rn, tcg_rm);
5625 neon_store_reg64(tcg_rd, rd + 1);
5626 tcg_temp_free_i64(tcg_rn);
5627 tcg_temp_free_i64(tcg_rm);
5628 tcg_temp_free_i64(tcg_rd);
5629 return 0;
5630 }
5631
5632 /* Avoid overlapping operands. Wide source operands are
5633 always aligned so will never overlap with wide
5634 destinations in problematic ways. */
5635 if (rd == rm && !src2_wide) {
5636 tmp = neon_load_reg(rm, 1);
5637 neon_store_scratch(2, tmp);
5638 } else if (rd == rn && !src1_wide) {
5639 tmp = neon_load_reg(rn, 1);
5640 neon_store_scratch(2, tmp);
5641 }
5642 tmp3 = NULL;
5643 for (pass = 0; pass < 2; pass++) {
5644 if (src1_wide) {
5645 neon_load_reg64(cpu_V0, rn + pass);
5646 tmp = NULL;
5647 } else {
5648 if (pass == 1 && rd == rn) {
5649 tmp = neon_load_scratch(2);
5650 } else {
5651 tmp = neon_load_reg(rn, pass);
5652 }
5653 if (prewiden) {
5654 gen_neon_widen(cpu_V0, tmp, size, u);
5655 }
5656 }
5657 if (src2_wide) {
5658 neon_load_reg64(cpu_V1, rm + pass);
5659 tmp2 = NULL;
5660 } else {
5661 if (pass == 1 && rd == rm) {
5662 tmp2 = neon_load_scratch(2);
5663 } else {
5664 tmp2 = neon_load_reg(rm, pass);
5665 }
5666 if (prewiden) {
5667 gen_neon_widen(cpu_V1, tmp2, size, u);
5668 }
5669 }
5670 switch (op) {
5671 case 0: case 1: case 4: /* VADDL, VADDW, VADDHN, VRADDHN */
5672 gen_neon_addl(size);
5673 break;
5674 case 2: case 3: case 6: /* VSUBL, VSUBW, VSUBHN, VRSUBHN */
5675 gen_neon_subl(size);
5676 break;
5677 case 5: case 7: /* VABAL, VABDL */
5678 switch ((size << 1) | u) {
5679 case 0:
5680 gen_helper_neon_abdl_s16(cpu_V0, tmp, tmp2);
5681 break;
5682 case 1:
5683 gen_helper_neon_abdl_u16(cpu_V0, tmp, tmp2);
5684 break;
5685 case 2:
5686 gen_helper_neon_abdl_s32(cpu_V0, tmp, tmp2);
5687 break;
5688 case 3:
5689 gen_helper_neon_abdl_u32(cpu_V0, tmp, tmp2);
5690 break;
5691 case 4:
5692 gen_helper_neon_abdl_s64(cpu_V0, tmp, tmp2);
5693 break;
5694 case 5:
5695 gen_helper_neon_abdl_u64(cpu_V0, tmp, tmp2);
5696 break;
5697 default: abort();
5698 }
5699 tcg_temp_free_i32(tmp2);
5700 tcg_temp_free_i32(tmp);
5701 break;
5702 case 8: case 9: case 10: case 11: case 12: case 13:
5703 /* VMLAL, VQDMLAL, VMLSL, VQDMLSL, VMULL, VQDMULL */
5704 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
5705 break;
5706 case 14: /* Polynomial VMULL */
5707 gen_helper_neon_mull_p8(cpu_V0, tmp, tmp2);
5708 tcg_temp_free_i32(tmp2);
5709 tcg_temp_free_i32(tmp);
5710 break;
5711 default: /* 15 is RESERVED: caught earlier */
5712 abort();
5713 }
5714 if (op == 13) {
5715 /* VQDMULL */
5716 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5717 neon_store_reg64(cpu_V0, rd + pass);
5718 } else if (op == 5 || (op >= 8 && op <= 11)) {
5719 /* Accumulate. */
5720 neon_load_reg64(cpu_V1, rd + pass);
5721 switch (op) {
5722 case 10: /* VMLSL */
5723 gen_neon_negl(cpu_V0, size);
5724 /* Fall through */
5725 case 5: case 8: /* VABAL, VMLAL */
5726 gen_neon_addl(size);
5727 break;
5728 case 9: case 11: /* VQDMLAL, VQDMLSL */
5729 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5730 if (op == 11) {
5731 gen_neon_negl(cpu_V0, size);
5732 }
5733 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
5734 break;
5735 default:
5736 abort();
5737 }
5738 neon_store_reg64(cpu_V0, rd + pass);
5739 } else if (op == 4 || op == 6) {
5740 /* Narrowing operation. */
5741 tmp = tcg_temp_new_i32();
5742 if (!u) {
5743 switch (size) {
5744 case 0:
5745 gen_helper_neon_narrow_high_u8(tmp, cpu_V0);
5746 break;
5747 case 1:
5748 gen_helper_neon_narrow_high_u16(tmp, cpu_V0);
5749 break;
5750 case 2:
5751 tcg_gen_extrh_i64_i32(tmp, cpu_V0);
5752 break;
5753 default: abort();
5754 }
5755 } else {
5756 switch (size) {
5757 case 0:
5758 gen_helper_neon_narrow_round_high_u8(tmp, cpu_V0);
5759 break;
5760 case 1:
5761 gen_helper_neon_narrow_round_high_u16(tmp, cpu_V0);
5762 break;
5763 case 2:
5764 tcg_gen_addi_i64(cpu_V0, cpu_V0, 1u << 31);
5765 tcg_gen_extrh_i64_i32(tmp, cpu_V0);
5766 break;
5767 default: abort();
5768 }
5769 }
5770 if (pass == 0) {
5771 tmp3 = tmp;
5772 } else {
5773 neon_store_reg(rd, 0, tmp3);
5774 neon_store_reg(rd, 1, tmp);
5775 }
5776 } else {
5777 /* Write back the result. */
5778 neon_store_reg64(cpu_V0, rd + pass);
5779 }
5780 }
5781 } else {
5782 /* Two registers and a scalar. NB that for ops of this form
5783 * the ARM ARM labels bit 24 as Q, but it is in our variable
5784 * 'u', not 'q'.
5785 */
5786 if (size == 0) {
5787 return 1;
5788 }
5789 switch (op) {
5790 case 1: /* Float VMLA scalar */
5791 case 5: /* Floating point VMLS scalar */
5792 case 9: /* Floating point VMUL scalar */
5793 if (size == 1) {
5794 return 1;
5795 }
5796 /* fall through */
5797 case 0: /* Integer VMLA scalar */
5798 case 4: /* Integer VMLS scalar */
5799 case 8: /* Integer VMUL scalar */
5800 case 12: /* VQDMULH scalar */
5801 case 13: /* VQRDMULH scalar */
5802 if (u && ((rd | rn) & 1)) {
5803 return 1;
5804 }
5805 tmp = neon_get_scalar(size, rm);
5806 neon_store_scratch(0, tmp);
5807 for (pass = 0; pass < (u ? 4 : 2); pass++) {
5808 tmp = neon_load_scratch(0);
5809 tmp2 = neon_load_reg(rn, pass);
5810 if (op == 12) {
5811 if (size == 1) {
5812 gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2);
5813 } else {
5814 gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2);
5815 }
5816 } else if (op == 13) {
5817 if (size == 1) {
5818 gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2);
5819 } else {
5820 gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2);
5821 }
5822 } else if (op & 1) {
5823 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5824 gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus);
5825 tcg_temp_free_ptr(fpstatus);
5826 } else {
5827 switch (size) {
5828 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
5829 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
5830 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
5831 default: abort();
5832 }
5833 }
5834 tcg_temp_free_i32(tmp2);
5835 if (op < 8) {
5836 /* Accumulate. */
5837 tmp2 = neon_load_reg(rd, pass);
5838 switch (op) {
5839 case 0:
5840 gen_neon_add(size, tmp, tmp2);
5841 break;
5842 case 1:
5843 {
5844 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5845 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
5846 tcg_temp_free_ptr(fpstatus);
5847 break;
5848 }
5849 case 4:
5850 gen_neon_rsb(size, tmp, tmp2);
5851 break;
5852 case 5:
5853 {
5854 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5855 gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus);
5856 tcg_temp_free_ptr(fpstatus);
5857 break;
5858 }
5859 default:
5860 abort();
5861 }
5862 tcg_temp_free_i32(tmp2);
5863 }
5864 neon_store_reg(rd, pass, tmp);
5865 }
5866 break;
5867 case 3: /* VQDMLAL scalar */
5868 case 7: /* VQDMLSL scalar */
5869 case 11: /* VQDMULL scalar */
5870 if (u == 1) {
5871 return 1;
5872 }
5873 /* fall through */
5874 case 2: /* VMLAL sclar */
5875 case 6: /* VMLSL scalar */
5876 case 10: /* VMULL scalar */
5877 if (rd & 1) {
5878 return 1;
5879 }
5880 tmp2 = neon_get_scalar(size, rm);
5881 /* We need a copy of tmp2 because gen_neon_mull
5882 * deletes it during pass 0. */
5883 tmp4 = tcg_temp_new_i32();
5884 tcg_gen_mov_i32(tmp4, tmp2);
5885 tmp3 = neon_load_reg(rn, 1);
5886
5887 for (pass = 0; pass < 2; pass++) {
5888 if (pass == 0) {
5889 tmp = neon_load_reg(rn, 0);
5890 } else {
5891 tmp = tmp3;
5892 tmp2 = tmp4;
5893 }
5894 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
5895 if (op != 11) {
5896 neon_load_reg64(cpu_V1, rd + pass);
5897 }
5898 switch (op) {
5899 case 6:
5900 gen_neon_negl(cpu_V0, size);
5901 /* Fall through */
5902 case 2:
5903 gen_neon_addl(size);
5904 break;
5905 case 3: case 7:
5906 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5907 if (op == 7) {
5908 gen_neon_negl(cpu_V0, size);
5909 }
5910 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
5911 break;
5912 case 10:
5913 /* no-op */
5914 break;
5915 case 11:
5916 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5917 break;
5918 default:
5919 abort();
5920 }
5921 neon_store_reg64(cpu_V0, rd + pass);
5922 }
5923 break;
5924 case 14: /* VQRDMLAH scalar */
5925 case 15: /* VQRDMLSH scalar */
5926 {
5927 NeonGenThreeOpEnvFn *fn;
5928
5929 if (!dc_isar_feature(aa32_rdm, s)) {
5930 return 1;
5931 }
5932 if (u && ((rd | rn) & 1)) {
5933 return 1;
5934 }
5935 if (op == 14) {
5936 if (size == 1) {
5937 fn = gen_helper_neon_qrdmlah_s16;
5938 } else {
5939 fn = gen_helper_neon_qrdmlah_s32;
5940 }
5941 } else {
5942 if (size == 1) {
5943 fn = gen_helper_neon_qrdmlsh_s16;
5944 } else {
5945 fn = gen_helper_neon_qrdmlsh_s32;
5946 }
5947 }
5948
5949 tmp2 = neon_get_scalar(size, rm);
5950 for (pass = 0; pass < (u ? 4 : 2); pass++) {
5951 tmp = neon_load_reg(rn, pass);
5952 tmp3 = neon_load_reg(rd, pass);
5953 fn(tmp, cpu_env, tmp, tmp2, tmp3);
5954 tcg_temp_free_i32(tmp3);
5955 neon_store_reg(rd, pass, tmp);
5956 }
5957 tcg_temp_free_i32(tmp2);
5958 }
5959 break;
5960 default:
5961 g_assert_not_reached();
5962 }
5963 }
5964 } else { /* size == 3 */
5965 if (!u) {
5966 /* Extract. */
5967 imm = (insn >> 8) & 0xf;
5968
5969 if (imm > 7 && !q)
5970 return 1;
5971
5972 if (q && ((rd | rn | rm) & 1)) {
5973 return 1;
5974 }
5975
5976 if (imm == 0) {
5977 neon_load_reg64(cpu_V0, rn);
5978 if (q) {
5979 neon_load_reg64(cpu_V1, rn + 1);
5980 }
5981 } else if (imm == 8) {
5982 neon_load_reg64(cpu_V0, rn + 1);
5983 if (q) {
5984 neon_load_reg64(cpu_V1, rm);
5985 }
5986 } else if (q) {
5987 tmp64 = tcg_temp_new_i64();
5988 if (imm < 8) {
5989 neon_load_reg64(cpu_V0, rn);
5990 neon_load_reg64(tmp64, rn + 1);
5991 } else {
5992 neon_load_reg64(cpu_V0, rn + 1);
5993 neon_load_reg64(tmp64, rm);
5994 }
5995 tcg_gen_shri_i64(cpu_V0, cpu_V0, (imm & 7) * 8);
5996 tcg_gen_shli_i64(cpu_V1, tmp64, 64 - ((imm & 7) * 8));
5997 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
5998 if (imm < 8) {
5999 neon_load_reg64(cpu_V1, rm);
6000 } else {
6001 neon_load_reg64(cpu_V1, rm + 1);
6002 imm -= 8;
6003 }
6004 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
6005 tcg_gen_shri_i64(tmp64, tmp64, imm * 8);
6006 tcg_gen_or_i64(cpu_V1, cpu_V1, tmp64);
6007 tcg_temp_free_i64(tmp64);
6008 } else {
6009 /* BUGFIX */
6010 neon_load_reg64(cpu_V0, rn);
6011 tcg_gen_shri_i64(cpu_V0, cpu_V0, imm * 8);
6012 neon_load_reg64(cpu_V1, rm);
6013 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
6014 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
6015 }
6016 neon_store_reg64(cpu_V0, rd);
6017 if (q) {
6018 neon_store_reg64(cpu_V1, rd + 1);
6019 }
6020 } else if ((insn & (1 << 11)) == 0) {
6021 /* Two register misc. */
6022 op = ((insn >> 12) & 0x30) | ((insn >> 7) & 0xf);
6023 size = (insn >> 18) & 3;
6024 /* UNDEF for unknown op values and bad op-size combinations */
6025 if ((neon_2rm_sizes[op] & (1 << size)) == 0) {
6026 return 1;
6027 }
6028 if (neon_2rm_is_v8_op(op) &&
6029 !arm_dc_feature(s, ARM_FEATURE_V8)) {
6030 return 1;
6031 }
6032 if ((op != NEON_2RM_VMOVN && op != NEON_2RM_VQMOVN) &&
6033 q && ((rm | rd) & 1)) {
6034 return 1;
6035 }
6036 switch (op) {
6037 case NEON_2RM_VREV64:
6038 for (pass = 0; pass < (q ? 2 : 1); pass++) {
6039 tmp = neon_load_reg(rm, pass * 2);
6040 tmp2 = neon_load_reg(rm, pass * 2 + 1);
6041 switch (size) {
6042 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
6043 case 1: gen_swap_half(tmp); break;
6044 case 2: /* no-op */ break;
6045 default: abort();
6046 }
6047 neon_store_reg(rd, pass * 2 + 1, tmp);
6048 if (size == 2) {
6049 neon_store_reg(rd, pass * 2, tmp2);
6050 } else {
6051 switch (size) {
6052 case 0: tcg_gen_bswap32_i32(tmp2, tmp2); break;
6053 case 1: gen_swap_half(tmp2); break;
6054 default: abort();
6055 }
6056 neon_store_reg(rd, pass * 2, tmp2);
6057 }
6058 }
6059 break;
6060 case NEON_2RM_VPADDL: case NEON_2RM_VPADDL_U:
6061 case NEON_2RM_VPADAL: case NEON_2RM_VPADAL_U:
6062 for (pass = 0; pass < q + 1; pass++) {
6063 tmp = neon_load_reg(rm, pass * 2);
6064 gen_neon_widen(cpu_V0, tmp, size, op & 1);
6065 tmp = neon_load_reg(rm, pass * 2 + 1);
6066 gen_neon_widen(cpu_V1, tmp, size, op & 1);
6067 switch (size) {
6068 case 0: gen_helper_neon_paddl_u16(CPU_V001); break;
6069 case 1: gen_helper_neon_paddl_u32(CPU_V001); break;
6070 case 2: tcg_gen_add_i64(CPU_V001); break;
6071 default: abort();
6072 }
6073 if (op >= NEON_2RM_VPADAL) {
6074 /* Accumulate. */
6075 neon_load_reg64(cpu_V1, rd + pass);
6076 gen_neon_addl(size);
6077 }
6078 neon_store_reg64(cpu_V0, rd + pass);
6079 }
6080 break;
6081 case NEON_2RM_VTRN:
6082 if (size == 2) {
6083 int n;
6084 for (n = 0; n < (q ? 4 : 2); n += 2) {
6085 tmp = neon_load_reg(rm, n);
6086 tmp2 = neon_load_reg(rd, n + 1);
6087 neon_store_reg(rm, n, tmp2);
6088 neon_store_reg(rd, n + 1, tmp);
6089 }
6090 } else {
6091 goto elementwise;
6092 }
6093 break;
6094 case NEON_2RM_VUZP:
6095 if (gen_neon_unzip(rd, rm, size, q)) {
6096 return 1;
6097 }
6098 break;
6099 case NEON_2RM_VZIP:
6100 if (gen_neon_zip(rd, rm, size, q)) {
6101 return 1;
6102 }
6103 break;
6104 case NEON_2RM_VMOVN: case NEON_2RM_VQMOVN:
6105 /* also VQMOVUN; op field and mnemonics don't line up */
6106 if (rm & 1) {
6107 return 1;
6108 }
6109 tmp2 = NULL;
6110 for (pass = 0; pass < 2; pass++) {
6111 neon_load_reg64(cpu_V0, rm + pass);
6112 tmp = tcg_temp_new_i32();
6113 gen_neon_narrow_op(op == NEON_2RM_VMOVN, q, size,
6114 tmp, cpu_V0);
6115 if (pass == 0) {
6116 tmp2 = tmp;
6117 } else {
6118 neon_store_reg(rd, 0, tmp2);
6119 neon_store_reg(rd, 1, tmp);
6120 }
6121 }
6122 break;
6123 case NEON_2RM_VSHLL:
6124 if (q || (rd & 1)) {
6125 return 1;
6126 }
6127 tmp = neon_load_reg(rm, 0);
6128 tmp2 = neon_load_reg(rm, 1);
6129 for (pass = 0; pass < 2; pass++) {
6130 if (pass == 1)
6131 tmp = tmp2;
6132 gen_neon_widen(cpu_V0, tmp, size, 1);
6133 tcg_gen_shli_i64(cpu_V0, cpu_V0, 8 << size);
6134 neon_store_reg64(cpu_V0, rd + pass);
6135 }
6136 break;
6137 case NEON_2RM_VCVT_F16_F32:
6138 {
6139 TCGv_ptr fpst;
6140 TCGv_i32 ahp;
6141
6142 if (!dc_isar_feature(aa32_fp16_spconv, s) ||
6143 q || (rm & 1)) {
6144 return 1;
6145 }
6146 fpst = get_fpstatus_ptr(true);
6147 ahp = get_ahp_flag();
6148 tmp = neon_load_reg(rm, 0);
6149 gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp);
6150 tmp2 = neon_load_reg(rm, 1);
6151 gen_helper_vfp_fcvt_f32_to_f16(tmp2, tmp2, fpst, ahp);
6152 tcg_gen_shli_i32(tmp2, tmp2, 16);
6153 tcg_gen_or_i32(tmp2, tmp2, tmp);
6154 tcg_temp_free_i32(tmp);
6155 tmp = neon_load_reg(rm, 2);
6156 gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp);
6157 tmp3 = neon_load_reg(rm, 3);
6158 neon_store_reg(rd, 0, tmp2);
6159 gen_helper_vfp_fcvt_f32_to_f16(tmp3, tmp3, fpst, ahp);
6160 tcg_gen_shli_i32(tmp3, tmp3, 16);
6161 tcg_gen_or_i32(tmp3, tmp3, tmp);
6162 neon_store_reg(rd, 1, tmp3);
6163 tcg_temp_free_i32(tmp);
6164 tcg_temp_free_i32(ahp);
6165 tcg_temp_free_ptr(fpst);
6166 break;
6167 }
6168 case NEON_2RM_VCVT_F32_F16:
6169 {
6170 TCGv_ptr fpst;
6171 TCGv_i32 ahp;
6172 if (!dc_isar_feature(aa32_fp16_spconv, s) ||
6173 q || (rd & 1)) {
6174 return 1;
6175 }
6176 fpst = get_fpstatus_ptr(true);
6177 ahp = get_ahp_flag();
6178 tmp3 = tcg_temp_new_i32();
6179 tmp = neon_load_reg(rm, 0);
6180 tmp2 = neon_load_reg(rm, 1);
6181 tcg_gen_ext16u_i32(tmp3, tmp);
6182 gen_helper_vfp_fcvt_f16_to_f32(tmp3, tmp3, fpst, ahp);
6183 neon_store_reg(rd, 0, tmp3);
6184 tcg_gen_shri_i32(tmp, tmp, 16);
6185 gen_helper_vfp_fcvt_f16_to_f32(tmp, tmp, fpst, ahp);
6186 neon_store_reg(rd, 1, tmp);
6187 tmp3 = tcg_temp_new_i32();
6188 tcg_gen_ext16u_i32(tmp3, tmp2);
6189 gen_helper_vfp_fcvt_f16_to_f32(tmp3, tmp3, fpst, ahp);
6190 neon_store_reg(rd, 2, tmp3);
6191 tcg_gen_shri_i32(tmp2, tmp2, 16);
6192 gen_helper_vfp_fcvt_f16_to_f32(tmp2, tmp2, fpst, ahp);
6193 neon_store_reg(rd, 3, tmp2);
6194 tcg_temp_free_i32(ahp);
6195 tcg_temp_free_ptr(fpst);
6196 break;
6197 }
6198 case NEON_2RM_AESE: case NEON_2RM_AESMC:
6199 if (!dc_isar_feature(aa32_aes, s) || ((rm | rd) & 1)) {
6200 return 1;
6201 }
6202 ptr1 = vfp_reg_ptr(true, rd);
6203 ptr2 = vfp_reg_ptr(true, rm);
6204
6205 /* Bit 6 is the lowest opcode bit; it distinguishes between
6206 * encryption (AESE/AESMC) and decryption (AESD/AESIMC)
6207 */
6208 tmp3 = tcg_const_i32(extract32(insn, 6, 1));
6209
6210 if (op == NEON_2RM_AESE) {
6211 gen_helper_crypto_aese(ptr1, ptr2, tmp3);
6212 } else {
6213 gen_helper_crypto_aesmc(ptr1, ptr2, tmp3);
6214 }
6215 tcg_temp_free_ptr(ptr1);
6216 tcg_temp_free_ptr(ptr2);
6217 tcg_temp_free_i32(tmp3);
6218 break;
6219 case NEON_2RM_SHA1H:
6220 if (!dc_isar_feature(aa32_sha1, s) || ((rm | rd) & 1)) {
6221 return 1;
6222 }
6223 ptr1 = vfp_reg_ptr(true, rd);
6224 ptr2 = vfp_reg_ptr(true, rm);
6225
6226 gen_helper_crypto_sha1h(ptr1, ptr2);
6227
6228 tcg_temp_free_ptr(ptr1);
6229 tcg_temp_free_ptr(ptr2);
6230 break;
6231 case NEON_2RM_SHA1SU1:
6232 if ((rm | rd) & 1) {
6233 return 1;
6234 }
6235 /* bit 6 (q): set -> SHA256SU0, cleared -> SHA1SU1 */
6236 if (q) {
6237 if (!dc_isar_feature(aa32_sha2, s)) {
6238 return 1;
6239 }
6240 } else if (!dc_isar_feature(aa32_sha1, s)) {
6241 return 1;
6242 }
6243 ptr1 = vfp_reg_ptr(true, rd);
6244 ptr2 = vfp_reg_ptr(true, rm);
6245 if (q) {
6246 gen_helper_crypto_sha256su0(ptr1, ptr2);
6247 } else {
6248 gen_helper_crypto_sha1su1(ptr1, ptr2);
6249 }
6250 tcg_temp_free_ptr(ptr1);
6251 tcg_temp_free_ptr(ptr2);
6252 break;
6253
6254 case NEON_2RM_VMVN:
6255 tcg_gen_gvec_not(0, rd_ofs, rm_ofs, vec_size, vec_size);
6256 break;
6257 case NEON_2RM_VNEG:
6258 tcg_gen_gvec_neg(size, rd_ofs, rm_ofs, vec_size, vec_size);
6259 break;
6260 case NEON_2RM_VABS:
6261 tcg_gen_gvec_abs(size, rd_ofs, rm_ofs, vec_size, vec_size);
6262 break;
6263
6264 default:
6265 elementwise:
6266 for (pass = 0; pass < (q ? 4 : 2); pass++) {
6267 tmp = neon_load_reg(rm, pass);
6268 switch (op) {
6269 case NEON_2RM_VREV32:
6270 switch (size) {
6271 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
6272 case 1: gen_swap_half(tmp); break;
6273 default: abort();
6274 }
6275 break;
6276 case NEON_2RM_VREV16:
6277 gen_rev16(tmp, tmp);
6278 break;
6279 case NEON_2RM_VCLS:
6280 switch (size) {
6281 case 0: gen_helper_neon_cls_s8(tmp, tmp); break;
6282 case 1: gen_helper_neon_cls_s16(tmp, tmp); break;
6283 case 2: gen_helper_neon_cls_s32(tmp, tmp); break;
6284 default: abort();
6285 }
6286 break;
6287 case NEON_2RM_VCLZ:
6288 switch (size) {
6289 case 0: gen_helper_neon_clz_u8(tmp, tmp); break;
6290 case 1: gen_helper_neon_clz_u16(tmp, tmp); break;
6291 case 2: tcg_gen_clzi_i32(tmp, tmp, 32); break;
6292 default: abort();
6293 }
6294 break;
6295 case NEON_2RM_VCNT:
6296 gen_helper_neon_cnt_u8(tmp, tmp);
6297 break;
6298 case NEON_2RM_VQABS:
6299 switch (size) {
6300 case 0:
6301 gen_helper_neon_qabs_s8(tmp, cpu_env, tmp);
6302 break;
6303 case 1:
6304 gen_helper_neon_qabs_s16(tmp, cpu_env, tmp);
6305 break;
6306 case 2:
6307 gen_helper_neon_qabs_s32(tmp, cpu_env, tmp);
6308 break;
6309 default: abort();
6310 }
6311 break;
6312 case NEON_2RM_VQNEG:
6313 switch (size) {
6314 case 0:
6315 gen_helper_neon_qneg_s8(tmp, cpu_env, tmp);
6316 break;
6317 case 1:
6318 gen_helper_neon_qneg_s16(tmp, cpu_env, tmp);
6319 break;
6320 case 2:
6321 gen_helper_neon_qneg_s32(tmp, cpu_env, tmp);
6322 break;
6323 default: abort();
6324 }
6325 break;
6326 case NEON_2RM_VCGT0: case NEON_2RM_VCLE0:
6327 tmp2 = tcg_const_i32(0);
6328 switch(size) {
6329 case 0: gen_helper_neon_cgt_s8(tmp, tmp, tmp2); break;
6330 case 1: gen_helper_neon_cgt_s16(tmp, tmp, tmp2); break;
6331 case 2: gen_helper_neon_cgt_s32(tmp, tmp, tmp2); break;
6332 default: abort();
6333 }
6334 tcg_temp_free_i32(tmp2);
6335 if (op == NEON_2RM_VCLE0) {
6336 tcg_gen_not_i32(tmp, tmp);
6337 }
6338 break;
6339 case NEON_2RM_VCGE0: case NEON_2RM_VCLT0:
6340 tmp2 = tcg_const_i32(0);
6341 switch(size) {
6342 case 0: gen_helper_neon_cge_s8(tmp, tmp, tmp2); break;
6343 case 1: gen_helper_neon_cge_s16(tmp, tmp, tmp2); break;
6344 case 2: gen_helper_neon_cge_s32(tmp, tmp, tmp2); break;
6345 default: abort();
6346 }
6347 tcg_temp_free_i32(tmp2);
6348 if (op == NEON_2RM_VCLT0) {
6349 tcg_gen_not_i32(tmp, tmp);
6350 }
6351 break;
6352 case NEON_2RM_VCEQ0:
6353 tmp2 = tcg_const_i32(0);
6354 switch(size) {
6355 case 0: gen_helper_neon_ceq_u8(tmp, tmp, tmp2); break;
6356 case 1: gen_helper_neon_ceq_u16(tmp, tmp, tmp2); break;
6357 case 2: gen_helper_neon_ceq_u32(tmp, tmp, tmp2); break;
6358 default: abort();
6359 }
6360 tcg_temp_free_i32(tmp2);
6361 break;
6362 case NEON_2RM_VCGT0_F:
6363 {
6364 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6365 tmp2 = tcg_const_i32(0);
6366 gen_helper_neon_cgt_f32(tmp, tmp, tmp2, fpstatus);
6367 tcg_temp_free_i32(tmp2);
6368 tcg_temp_free_ptr(fpstatus);
6369 break;
6370 }
6371 case NEON_2RM_VCGE0_F:
6372 {
6373 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6374 tmp2 = tcg_const_i32(0);
6375 gen_helper_neon_cge_f32(tmp, tmp, tmp2, fpstatus);
6376 tcg_temp_free_i32(tmp2);
6377 tcg_temp_free_ptr(fpstatus);
6378 break;
6379 }
6380 case NEON_2RM_VCEQ0_F:
6381 {
6382 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6383 tmp2 = tcg_const_i32(0);
6384 gen_helper_neon_ceq_f32(tmp, tmp, tmp2, fpstatus);
6385 tcg_temp_free_i32(tmp2);
6386 tcg_temp_free_ptr(fpstatus);
6387 break;
6388 }
6389 case NEON_2RM_VCLE0_F:
6390 {
6391 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6392 tmp2 = tcg_const_i32(0);
6393 gen_helper_neon_cge_f32(tmp, tmp2, tmp, fpstatus);
6394 tcg_temp_free_i32(tmp2);
6395 tcg_temp_free_ptr(fpstatus);
6396 break;
6397 }
6398 case NEON_2RM_VCLT0_F:
6399 {
6400 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6401 tmp2 = tcg_const_i32(0);
6402 gen_helper_neon_cgt_f32(tmp, tmp2, tmp, fpstatus);
6403 tcg_temp_free_i32(tmp2);
6404 tcg_temp_free_ptr(fpstatus);
6405 break;
6406 }
6407 case NEON_2RM_VABS_F:
6408 gen_helper_vfp_abss(tmp, tmp);
6409 break;
6410 case NEON_2RM_VNEG_F:
6411 gen_helper_vfp_negs(tmp, tmp);
6412 break;
6413 case NEON_2RM_VSWP:
6414 tmp2 = neon_load_reg(rd, pass);
6415 neon_store_reg(rm, pass, tmp2);
6416 break;
6417 case NEON_2RM_VTRN:
6418 tmp2 = neon_load_reg(rd, pass);
6419 switch (size) {
6420 case 0: gen_neon_trn_u8(tmp, tmp2); break;
6421 case 1: gen_neon_trn_u16(tmp, tmp2); break;
6422 default: abort();
6423 }
6424 neon_store_reg(rm, pass, tmp2);
6425 break;
6426 case NEON_2RM_VRINTN:
6427 case NEON_2RM_VRINTA:
6428 case NEON_2RM_VRINTM:
6429 case NEON_2RM_VRINTP:
6430 case NEON_2RM_VRINTZ:
6431 {
6432 TCGv_i32 tcg_rmode;
6433 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6434 int rmode;
6435
6436 if (op == NEON_2RM_VRINTZ) {
6437 rmode = FPROUNDING_ZERO;
6438 } else {
6439 rmode = fp_decode_rm[((op & 0x6) >> 1) ^ 1];
6440 }
6441
6442 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
6443 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6444 cpu_env);
6445 gen_helper_rints(tmp, tmp, fpstatus);
6446 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6447 cpu_env);
6448 tcg_temp_free_ptr(fpstatus);
6449 tcg_temp_free_i32(tcg_rmode);
6450 break;
6451 }
6452 case NEON_2RM_VRINTX:
6453 {
6454 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6455 gen_helper_rints_exact(tmp, tmp, fpstatus);
6456 tcg_temp_free_ptr(fpstatus);
6457 break;
6458 }
6459 case NEON_2RM_VCVTAU:
6460 case NEON_2RM_VCVTAS:
6461 case NEON_2RM_VCVTNU:
6462 case NEON_2RM_VCVTNS:
6463 case NEON_2RM_VCVTPU:
6464 case NEON_2RM_VCVTPS:
6465 case NEON_2RM_VCVTMU:
6466 case NEON_2RM_VCVTMS:
6467 {
6468 bool is_signed = !extract32(insn, 7, 1);
6469 TCGv_ptr fpst = get_fpstatus_ptr(1);
6470 TCGv_i32 tcg_rmode, tcg_shift;
6471 int rmode = fp_decode_rm[extract32(insn, 8, 2)];
6472
6473 tcg_shift = tcg_const_i32(0);
6474 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
6475 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6476 cpu_env);
6477
6478 if (is_signed) {
6479 gen_helper_vfp_tosls(tmp, tmp,
6480 tcg_shift, fpst);
6481 } else {
6482 gen_helper_vfp_touls(tmp, tmp,
6483 tcg_shift, fpst);
6484 }
6485
6486 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6487 cpu_env);
6488 tcg_temp_free_i32(tcg_rmode);
6489 tcg_temp_free_i32(tcg_shift);
6490 tcg_temp_free_ptr(fpst);
6491 break;
6492 }
6493 case NEON_2RM_VRECPE:
6494 {
6495 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6496 gen_helper_recpe_u32(tmp, tmp, fpstatus);
6497 tcg_temp_free_ptr(fpstatus);
6498 break;
6499 }
6500 case NEON_2RM_VRSQRTE:
6501 {
6502 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6503 gen_helper_rsqrte_u32(tmp, tmp, fpstatus);
6504 tcg_temp_free_ptr(fpstatus);
6505 break;
6506 }
6507 case NEON_2RM_VRECPE_F:
6508 {
6509 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6510 gen_helper_recpe_f32(tmp, tmp, fpstatus);
6511 tcg_temp_free_ptr(fpstatus);
6512 break;
6513 }
6514 case NEON_2RM_VRSQRTE_F:
6515 {
6516 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6517 gen_helper_rsqrte_f32(tmp, tmp, fpstatus);
6518 tcg_temp_free_ptr(fpstatus);
6519 break;
6520 }
6521 case NEON_2RM_VCVT_FS: /* VCVT.F32.S32 */
6522 {
6523 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6524 gen_helper_vfp_sitos(tmp, tmp, fpstatus);
6525 tcg_temp_free_ptr(fpstatus);
6526 break;
6527 }
6528 case NEON_2RM_VCVT_FU: /* VCVT.F32.U32 */
6529 {
6530 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6531 gen_helper_vfp_uitos(tmp, tmp, fpstatus);
6532 tcg_temp_free_ptr(fpstatus);
6533 break;
6534 }
6535 case NEON_2RM_VCVT_SF: /* VCVT.S32.F32 */
6536 {
6537 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6538 gen_helper_vfp_tosizs(tmp, tmp, fpstatus);
6539 tcg_temp_free_ptr(fpstatus);
6540 break;
6541 }
6542 case NEON_2RM_VCVT_UF: /* VCVT.U32.F32 */
6543 {
6544 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6545 gen_helper_vfp_touizs(tmp, tmp, fpstatus);
6546 tcg_temp_free_ptr(fpstatus);
6547 break;
6548 }
6549 default:
6550 /* Reserved op values were caught by the
6551 * neon_2rm_sizes[] check earlier.
6552 */
6553 abort();
6554 }
6555 neon_store_reg(rd, pass, tmp);
6556 }
6557 break;
6558 }
6559 } else if ((insn & (1 << 10)) == 0) {
6560 /* VTBL, VTBX. */
6561 int n = ((insn >> 8) & 3) + 1;
6562 if ((rn + n) > 32) {
6563 /* This is UNPREDICTABLE; we choose to UNDEF to avoid the
6564 * helper function running off the end of the register file.
6565 */
6566 return 1;
6567 }
6568 n <<= 3;
6569 if (insn & (1 << 6)) {
6570 tmp = neon_load_reg(rd, 0);
6571 } else {
6572 tmp = tcg_temp_new_i32();
6573 tcg_gen_movi_i32(tmp, 0);
6574 }
6575 tmp2 = neon_load_reg(rm, 0);
6576 ptr1 = vfp_reg_ptr(true, rn);
6577 tmp5 = tcg_const_i32(n);
6578 gen_helper_neon_tbl(tmp2, tmp2, tmp, ptr1, tmp5);
6579 tcg_temp_free_i32(tmp);
6580 if (insn & (1 << 6)) {
6581 tmp = neon_load_reg(rd, 1);
6582 } else {
6583 tmp = tcg_temp_new_i32();
6584 tcg_gen_movi_i32(tmp, 0);
6585 }
6586 tmp3 = neon_load_reg(rm, 1);
6587 gen_helper_neon_tbl(tmp3, tmp3, tmp, ptr1, tmp5);
6588 tcg_temp_free_i32(tmp5);
6589 tcg_temp_free_ptr(ptr1);
6590 neon_store_reg(rd, 0, tmp2);
6591 neon_store_reg(rd, 1, tmp3);
6592 tcg_temp_free_i32(tmp);
6593 } else if ((insn & 0x380) == 0) {
6594 /* VDUP */
6595 int element;
6596 MemOp size;
6597
6598 if ((insn & (7 << 16)) == 0 || (q && (rd & 1))) {
6599 return 1;
6600 }
6601 if (insn & (1 << 16)) {
6602 size = MO_8;
6603 element = (insn >> 17) & 7;
6604 } else if (insn & (1 << 17)) {
6605 size = MO_16;
6606 element = (insn >> 18) & 3;
6607 } else {
6608 size = MO_32;
6609 element = (insn >> 19) & 1;
6610 }
6611 tcg_gen_gvec_dup_mem(size, neon_reg_offset(rd, 0),
6612 neon_element_offset(rm, element, size),
6613 q ? 16 : 8, q ? 16 : 8);
6614 } else {
6615 return 1;
6616 }
6617 }
6618 }
6619 return 0;
6620}
6621
6622/* Advanced SIMD three registers of the same length extension.
6623 * 31 25 23 22 20 16 12 11 10 9 8 3 0
6624 * +---------------+-----+---+-----+----+----+---+----+---+----+---------+----+
6625 * | 1 1 1 1 1 1 0 | op1 | D | op2 | Vn | Vd | 1 | o3 | 0 | o4 | N Q M U | Vm |
6626 * +---------------+-----+---+-----+----+----+---+----+---+----+---------+----+
6627 */
6628static int disas_neon_insn_3same_ext(DisasContext *s, uint32_t insn)
6629{
6630 gen_helper_gvec_3 *fn_gvec = NULL;
6631 gen_helper_gvec_3_ptr *fn_gvec_ptr = NULL;
6632 int rd, rn, rm, opr_sz;
6633 int data = 0;
6634 int off_rn, off_rm;
6635 bool is_long = false, q = extract32(insn, 6, 1);
6636 bool ptr_is_env = false;
6637
6638 if ((insn & 0xfe200f10) == 0xfc200800) {
6639 /* VCMLA -- 1111 110R R.1S .... .... 1000 ...0 .... */
6640 int size = extract32(insn, 20, 1);
6641 data = extract32(insn, 23, 2); /* rot */
6642 if (!dc_isar_feature(aa32_vcma, s)
6643 || (!size && !dc_isar_feature(aa32_fp16_arith, s))) {
6644 return 1;
6645 }
6646 fn_gvec_ptr = size ? gen_helper_gvec_fcmlas : gen_helper_gvec_fcmlah;
6647 } else if ((insn & 0xfea00f10) == 0xfc800800) {
6648 /* VCADD -- 1111 110R 1.0S .... .... 1000 ...0 .... */
6649 int size = extract32(insn, 20, 1);
6650 data = extract32(insn, 24, 1); /* rot */
6651 if (!dc_isar_feature(aa32_vcma, s)
6652 || (!size && !dc_isar_feature(aa32_fp16_arith, s))) {
6653 return 1;
6654 }
6655 fn_gvec_ptr = size ? gen_helper_gvec_fcadds : gen_helper_gvec_fcaddh;
6656 } else if ((insn & 0xfeb00f00) == 0xfc200d00) {
6657 /* V[US]DOT -- 1111 1100 0.10 .... .... 1101 .Q.U .... */
6658 bool u = extract32(insn, 4, 1);
6659 if (!dc_isar_feature(aa32_dp, s)) {
6660 return 1;
6661 }
6662 fn_gvec = u ? gen_helper_gvec_udot_b : gen_helper_gvec_sdot_b;
6663 } else if ((insn & 0xff300f10) == 0xfc200810) {
6664 /* VFM[AS]L -- 1111 1100 S.10 .... .... 1000 .Q.1 .... */
6665 int is_s = extract32(insn, 23, 1);
6666 if (!dc_isar_feature(aa32_fhm, s)) {
6667 return 1;
6668 }
6669 is_long = true;
6670 data = is_s; /* is_2 == 0 */
6671 fn_gvec_ptr = gen_helper_gvec_fmlal_a32;
6672 ptr_is_env = true;
6673 } else {
6674 return 1;
6675 }
6676
6677 VFP_DREG_D(rd, insn);
6678 if (rd & q) {
6679 return 1;
6680 }
6681 if (q || !is_long) {
6682 VFP_DREG_N(rn, insn);
6683 VFP_DREG_M(rm, insn);
6684 if ((rn | rm) & q & !is_long) {
6685 return 1;
6686 }
6687 off_rn = vfp_reg_offset(1, rn);
6688 off_rm = vfp_reg_offset(1, rm);
6689 } else {
6690 rn = VFP_SREG_N(insn);
6691 rm = VFP_SREG_M(insn);
6692 off_rn = vfp_reg_offset(0, rn);
6693 off_rm = vfp_reg_offset(0, rm);
6694 }
6695
6696 if (s->fp_excp_el) {
6697 gen_exception_insn(s, s->pc_curr, EXCP_UDEF,
6698 syn_simd_access_trap(1, 0xe, false), s->fp_excp_el);
6699 return 0;
6700 }
6701 if (!s->vfp_enabled) {
6702 return 1;
6703 }
6704
6705 opr_sz = (1 + q) * 8;
6706 if (fn_gvec_ptr) {
6707 TCGv_ptr ptr;
6708 if (ptr_is_env) {
6709 ptr = cpu_env;
6710 } else {
6711 ptr = get_fpstatus_ptr(1);
6712 }
6713 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, rd), off_rn, off_rm, ptr,
6714 opr_sz, opr_sz, data, fn_gvec_ptr);
6715 if (!ptr_is_env) {
6716 tcg_temp_free_ptr(ptr);
6717 }
6718 } else {
6719 tcg_gen_gvec_3_ool(vfp_reg_offset(1, rd), off_rn, off_rm,
6720 opr_sz, opr_sz, data, fn_gvec);
6721 }
6722 return 0;
6723}
6724
6725/* Advanced SIMD two registers and a scalar extension.
6726 * 31 24 23 22 20 16 12 11 10 9 8 3 0
6727 * +-----------------+----+---+----+----+----+---+----+---+----+---------+----+
6728 * | 1 1 1 1 1 1 1 0 | o1 | D | o2 | Vn | Vd | 1 | o3 | 0 | o4 | N Q M U | Vm |
6729 * +-----------------+----+---+----+----+----+---+----+---+----+---------+----+
6730 *
6731 */
6732
6733static int disas_neon_insn_2reg_scalar_ext(DisasContext *s, uint32_t insn)
6734{
6735 gen_helper_gvec_3 *fn_gvec = NULL;
6736 gen_helper_gvec_3_ptr *fn_gvec_ptr = NULL;
6737 int rd, rn, rm, opr_sz, data;
6738 int off_rn, off_rm;
6739 bool is_long = false, q = extract32(insn, 6, 1);
6740 bool ptr_is_env = false;
6741
6742 if ((insn & 0xff000f10) == 0xfe000800) {
6743 /* VCMLA (indexed) -- 1111 1110 S.RR .... .... 1000 ...0 .... */
6744 int rot = extract32(insn, 20, 2);
6745 int size = extract32(insn, 23, 1);
6746 int index;
6747
6748 if (!dc_isar_feature(aa32_vcma, s)) {
6749 return 1;
6750 }
6751 if (size == 0) {
6752 if (!dc_isar_feature(aa32_fp16_arith, s)) {
6753 return 1;
6754 }
6755 /* For fp16, rm is just Vm, and index is M. */
6756 rm = extract32(insn, 0, 4);
6757 index = extract32(insn, 5, 1);
6758 } else {
6759 /* For fp32, rm is the usual M:Vm, and index is 0. */
6760 VFP_DREG_M(rm, insn);
6761 index = 0;
6762 }
6763 data = (index << 2) | rot;
6764 fn_gvec_ptr = (size ? gen_helper_gvec_fcmlas_idx
6765 : gen_helper_gvec_fcmlah_idx);
6766 } else if ((insn & 0xffb00f00) == 0xfe200d00) {
6767 /* V[US]DOT -- 1111 1110 0.10 .... .... 1101 .Q.U .... */
6768 int u = extract32(insn, 4, 1);
6769
6770 if (!dc_isar_feature(aa32_dp, s)) {
6771 return 1;
6772 }
6773 fn_gvec = u ? gen_helper_gvec_udot_idx_b : gen_helper_gvec_sdot_idx_b;
6774 /* rm is just Vm, and index is M. */
6775 data = extract32(insn, 5, 1); /* index */
6776 rm = extract32(insn, 0, 4);
6777 } else if ((insn & 0xffa00f10) == 0xfe000810) {
6778 /* VFM[AS]L -- 1111 1110 0.0S .... .... 1000 .Q.1 .... */
6779 int is_s = extract32(insn, 20, 1);
6780 int vm20 = extract32(insn, 0, 3);
6781 int vm3 = extract32(insn, 3, 1);
6782 int m = extract32(insn, 5, 1);
6783 int index;
6784
6785 if (!dc_isar_feature(aa32_fhm, s)) {
6786 return 1;
6787 }
6788 if (q) {
6789 rm = vm20;
6790 index = m * 2 + vm3;
6791 } else {
6792 rm = vm20 * 2 + m;
6793 index = vm3;
6794 }
6795 is_long = true;
6796 data = (index << 2) | is_s; /* is_2 == 0 */
6797 fn_gvec_ptr = gen_helper_gvec_fmlal_idx_a32;
6798 ptr_is_env = true;
6799 } else {
6800 return 1;
6801 }
6802
6803 VFP_DREG_D(rd, insn);
6804 if (rd & q) {
6805 return 1;
6806 }
6807 if (q || !is_long) {
6808 VFP_DREG_N(rn, insn);
6809 if (rn & q & !is_long) {
6810 return 1;
6811 }
6812 off_rn = vfp_reg_offset(1, rn);
6813 off_rm = vfp_reg_offset(1, rm);
6814 } else {
6815 rn = VFP_SREG_N(insn);
6816 off_rn = vfp_reg_offset(0, rn);
6817 off_rm = vfp_reg_offset(0, rm);
6818 }
6819 if (s->fp_excp_el) {
6820 gen_exception_insn(s, s->pc_curr, EXCP_UDEF,
6821 syn_simd_access_trap(1, 0xe, false), s->fp_excp_el);
6822 return 0;
6823 }
6824 if (!s->vfp_enabled) {
6825 return 1;
6826 }
6827
6828 opr_sz = (1 + q) * 8;
6829 if (fn_gvec_ptr) {
6830 TCGv_ptr ptr;
6831 if (ptr_is_env) {
6832 ptr = cpu_env;
6833 } else {
6834 ptr = get_fpstatus_ptr(1);
6835 }
6836 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, rd), off_rn, off_rm, ptr,
6837 opr_sz, opr_sz, data, fn_gvec_ptr);
6838 if (!ptr_is_env) {
6839 tcg_temp_free_ptr(ptr);
6840 }
6841 } else {
6842 tcg_gen_gvec_3_ool(vfp_reg_offset(1, rd), off_rn, off_rm,
6843 opr_sz, opr_sz, data, fn_gvec);
6844 }
6845 return 0;
6846}
6847
6848static int disas_coproc_insn(DisasContext *s, uint32_t insn)
6849{
6850 int cpnum, is64, crn, crm, opc1, opc2, isread, rt, rt2;
6851 const ARMCPRegInfo *ri;
6852
6853 cpnum = (insn >> 8) & 0xf;
6854
6855 /* First check for coprocessor space used for XScale/iwMMXt insns */
6856 if (arm_dc_feature(s, ARM_FEATURE_XSCALE) && (cpnum < 2)) {
6857 if (extract32(s->c15_cpar, cpnum, 1) == 0) {
6858 return 1;
6859 }
6860 if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
6861 return disas_iwmmxt_insn(s, insn);
6862 } else if (arm_dc_feature(s, ARM_FEATURE_XSCALE)) {
6863 return disas_dsp_insn(s, insn);
6864 }
6865 return 1;
6866 }
6867
6868 /* Otherwise treat as a generic register access */
6869 is64 = (insn & (1 << 25)) == 0;
6870 if (!is64 && ((insn & (1 << 4)) == 0)) {
6871 /* cdp */
6872 return 1;
6873 }
6874
6875 crm = insn & 0xf;
6876 if (is64) {
6877 crn = 0;
6878 opc1 = (insn >> 4) & 0xf;
6879 opc2 = 0;
6880 rt2 = (insn >> 16) & 0xf;
6881 } else {
6882 crn = (insn >> 16) & 0xf;
6883 opc1 = (insn >> 21) & 7;
6884 opc2 = (insn >> 5) & 7;
6885 rt2 = 0;
6886 }
6887 isread = (insn >> 20) & 1;
6888 rt = (insn >> 12) & 0xf;
6889
6890 ri = get_arm_cp_reginfo(s->cp_regs,
6891 ENCODE_CP_REG(cpnum, is64, s->ns, crn, crm, opc1, opc2));
6892 if (ri) {
6893 /* Check access permissions */
6894 if (!cp_access_ok(s->current_el, ri, isread)) {
6895 return 1;
6896 }
6897
6898 if (ri->accessfn ||
6899 (arm_dc_feature(s, ARM_FEATURE_XSCALE) && cpnum < 14)) {
6900 /* Emit code to perform further access permissions checks at
6901 * runtime; this may result in an exception.
6902 * Note that on XScale all cp0..c13 registers do an access check
6903 * call in order to handle c15_cpar.
6904 */
6905 TCGv_ptr tmpptr;
6906 TCGv_i32 tcg_syn, tcg_isread;
6907 uint32_t syndrome;
6908
6909 /* Note that since we are an implementation which takes an
6910 * exception on a trapped conditional instruction only if the
6911 * instruction passes its condition code check, we can take
6912 * advantage of the clause in the ARM ARM that allows us to set
6913 * the COND field in the instruction to 0xE in all cases.
6914 * We could fish the actual condition out of the insn (ARM)
6915 * or the condexec bits (Thumb) but it isn't necessary.
6916 */
6917 switch (cpnum) {
6918 case 14:
6919 if (is64) {
6920 syndrome = syn_cp14_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
6921 isread, false);
6922 } else {
6923 syndrome = syn_cp14_rt_trap(1, 0xe, opc1, opc2, crn, crm,
6924 rt, isread, false);
6925 }
6926 break;
6927 case 15:
6928 if (is64) {
6929 syndrome = syn_cp15_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
6930 isread, false);
6931 } else {
6932 syndrome = syn_cp15_rt_trap(1, 0xe, opc1, opc2, crn, crm,
6933 rt, isread, false);
6934 }
6935 break;
6936 default:
6937 /* ARMv8 defines that only coprocessors 14 and 15 exist,
6938 * so this can only happen if this is an ARMv7 or earlier CPU,
6939 * in which case the syndrome information won't actually be
6940 * guest visible.
6941 */
6942 assert(!arm_dc_feature(s, ARM_FEATURE_V8));
6943 syndrome = syn_uncategorized();
6944 break;
6945 }
6946
6947 gen_set_condexec(s);
6948 gen_set_pc_im(s, s->pc_curr);
6949 tmpptr = tcg_const_ptr(ri);
6950 tcg_syn = tcg_const_i32(syndrome);
6951 tcg_isread = tcg_const_i32(isread);
6952 gen_helper_access_check_cp_reg(cpu_env, tmpptr, tcg_syn,
6953 tcg_isread);
6954 tcg_temp_free_ptr(tmpptr);
6955 tcg_temp_free_i32(tcg_syn);
6956 tcg_temp_free_i32(tcg_isread);
6957 } else if (ri->type & ARM_CP_RAISES_EXC) {
6958 /*
6959 * The readfn or writefn might raise an exception;
6960 * synchronize the CPU state in case it does.
6961 */
6962 gen_set_condexec(s);
6963 gen_set_pc_im(s, s->pc_curr);
6964 }
6965
6966 /* Handle special cases first */
6967 switch (ri->type & ~(ARM_CP_FLAG_MASK & ~ARM_CP_SPECIAL)) {
6968 case ARM_CP_NOP:
6969 return 0;
6970 case ARM_CP_WFI:
6971 if (isread) {
6972 return 1;
6973 }
6974 gen_set_pc_im(s, s->base.pc_next);
6975 s->base.is_jmp = DISAS_WFI;
6976 return 0;
6977 default:
6978 break;
6979 }
6980
6981 if ((tb_cflags(s->base.tb) & CF_USE_ICOUNT) && (ri->type & ARM_CP_IO)) {
6982 gen_io_start();
6983 }
6984
6985 if (isread) {
6986 /* Read */
6987 if (is64) {
6988 TCGv_i64 tmp64;
6989 TCGv_i32 tmp;
6990 if (ri->type & ARM_CP_CONST) {
6991 tmp64 = tcg_const_i64(ri->resetvalue);
6992 } else if (ri->readfn) {
6993 TCGv_ptr tmpptr;
6994 tmp64 = tcg_temp_new_i64();
6995 tmpptr = tcg_const_ptr(ri);
6996 gen_helper_get_cp_reg64(tmp64, cpu_env, tmpptr);
6997 tcg_temp_free_ptr(tmpptr);
6998 } else {
6999 tmp64 = tcg_temp_new_i64();
7000 tcg_gen_ld_i64(tmp64, cpu_env, ri->fieldoffset);
7001 }
7002 tmp = tcg_temp_new_i32();
7003 tcg_gen_extrl_i64_i32(tmp, tmp64);
7004 store_reg(s, rt, tmp);
7005 tmp = tcg_temp_new_i32();
7006 tcg_gen_extrh_i64_i32(tmp, tmp64);
7007 tcg_temp_free_i64(tmp64);
7008 store_reg(s, rt2, tmp);
7009 } else {
7010 TCGv_i32 tmp;
7011 if (ri->type & ARM_CP_CONST) {
7012 tmp = tcg_const_i32(ri->resetvalue);
7013 } else if (ri->readfn) {
7014 TCGv_ptr tmpptr;
7015 tmp = tcg_temp_new_i32();
7016 tmpptr = tcg_const_ptr(ri);
7017 gen_helper_get_cp_reg(tmp, cpu_env, tmpptr);
7018 tcg_temp_free_ptr(tmpptr);
7019 } else {
7020 tmp = load_cpu_offset(ri->fieldoffset);
7021 }
7022 if (rt == 15) {
7023 /* Destination register of r15 for 32 bit loads sets
7024 * the condition codes from the high 4 bits of the value
7025 */
7026 gen_set_nzcv(tmp);
7027 tcg_temp_free_i32(tmp);
7028 } else {
7029 store_reg(s, rt, tmp);
7030 }
7031 }
7032 } else {
7033 /* Write */
7034 if (ri->type & ARM_CP_CONST) {
7035 /* If not forbidden by access permissions, treat as WI */
7036 return 0;
7037 }
7038
7039 if (is64) {
7040 TCGv_i32 tmplo, tmphi;
7041 TCGv_i64 tmp64 = tcg_temp_new_i64();
7042 tmplo = load_reg(s, rt);
7043 tmphi = load_reg(s, rt2);
7044 tcg_gen_concat_i32_i64(tmp64, tmplo, tmphi);
7045 tcg_temp_free_i32(tmplo);
7046 tcg_temp_free_i32(tmphi);
7047 if (ri->writefn) {
7048 TCGv_ptr tmpptr = tcg_const_ptr(ri);
7049 gen_helper_set_cp_reg64(cpu_env, tmpptr, tmp64);
7050 tcg_temp_free_ptr(tmpptr);
7051 } else {
7052 tcg_gen_st_i64(tmp64, cpu_env, ri->fieldoffset);
7053 }
7054 tcg_temp_free_i64(tmp64);
7055 } else {
7056 if (ri->writefn) {
7057 TCGv_i32 tmp;
7058 TCGv_ptr tmpptr;
7059 tmp = load_reg(s, rt);
7060 tmpptr = tcg_const_ptr(ri);
7061 gen_helper_set_cp_reg(cpu_env, tmpptr, tmp);
7062 tcg_temp_free_ptr(tmpptr);
7063 tcg_temp_free_i32(tmp);
7064 } else {
7065 TCGv_i32 tmp = load_reg(s, rt);
7066 store_cpu_offset(tmp, ri->fieldoffset);
7067 }
7068 }
7069 }
7070
7071 if ((tb_cflags(s->base.tb) & CF_USE_ICOUNT) && (ri->type & ARM_CP_IO)) {
7072 /* I/O operations must end the TB here (whether read or write) */
7073 gen_lookup_tb(s);
7074 } else if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
7075 /* We default to ending the TB on a coprocessor register write,
7076 * but allow this to be suppressed by the register definition
7077 * (usually only necessary to work around guest bugs).
7078 */
7079 gen_lookup_tb(s);
7080 }
7081
7082 return 0;
7083 }
7084
7085 /* Unknown register; this might be a guest error or a QEMU
7086 * unimplemented feature.
7087 */
7088 if (is64) {
7089 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
7090 "64 bit system register cp:%d opc1: %d crm:%d "
7091 "(%s)\n",
7092 isread ? "read" : "write", cpnum, opc1, crm,
7093 s->ns ? "non-secure" : "secure");
7094 } else {
7095 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
7096 "system register cp:%d opc1:%d crn:%d crm:%d opc2:%d "
7097 "(%s)\n",
7098 isread ? "read" : "write", cpnum, opc1, crn, crm, opc2,
7099 s->ns ? "non-secure" : "secure");
7100 }
7101
7102 return 1;
7103}
7104
7105
7106/* Store a 64-bit value to a register pair. Clobbers val. */
7107static void gen_storeq_reg(DisasContext *s, int rlow, int rhigh, TCGv_i64 val)
7108{
7109 TCGv_i32 tmp;
7110 tmp = tcg_temp_new_i32();
7111 tcg_gen_extrl_i64_i32(tmp, val);
7112 store_reg(s, rlow, tmp);
7113 tmp = tcg_temp_new_i32();
7114 tcg_gen_extrh_i64_i32(tmp, val);
7115 store_reg(s, rhigh, tmp);
7116}
7117
7118/* load and add a 64-bit value from a register pair. */
7119static void gen_addq(DisasContext *s, TCGv_i64 val, int rlow, int rhigh)
7120{
7121 TCGv_i64 tmp;
7122 TCGv_i32 tmpl;
7123 TCGv_i32 tmph;
7124
7125 /* Load 64-bit value rd:rn. */
7126 tmpl = load_reg(s, rlow);
7127 tmph = load_reg(s, rhigh);
7128 tmp = tcg_temp_new_i64();
7129 tcg_gen_concat_i32_i64(tmp, tmpl, tmph);
7130 tcg_temp_free_i32(tmpl);
7131 tcg_temp_free_i32(tmph);
7132 tcg_gen_add_i64(val, val, tmp);
7133 tcg_temp_free_i64(tmp);
7134}
7135
7136/* Set N and Z flags from hi|lo. */
7137static void gen_logicq_cc(TCGv_i32 lo, TCGv_i32 hi)
7138{
7139 tcg_gen_mov_i32(cpu_NF, hi);
7140 tcg_gen_or_i32(cpu_ZF, lo, hi);
7141}
7142
7143/* Load/Store exclusive instructions are implemented by remembering
7144 the value/address loaded, and seeing if these are the same
7145 when the store is performed. This should be sufficient to implement
7146 the architecturally mandated semantics, and avoids having to monitor
7147 regular stores. The compare vs the remembered value is done during
7148 the cmpxchg operation, but we must compare the addresses manually. */
7149static void gen_load_exclusive(DisasContext *s, int rt, int rt2,
7150 TCGv_i32 addr, int size)
7151{
7152 TCGv_i32 tmp = tcg_temp_new_i32();
7153 MemOp opc = size | MO_ALIGN | s->be_data;
7154
7155 s->is_ldex = true;
7156
7157 if (size == 3) {
7158 TCGv_i32 tmp2 = tcg_temp_new_i32();
7159 TCGv_i64 t64 = tcg_temp_new_i64();
7160
7161 /* For AArch32, architecturally the 32-bit word at the lowest
7162 * address is always Rt and the one at addr+4 is Rt2, even if
7163 * the CPU is big-endian. That means we don't want to do a
7164 * gen_aa32_ld_i64(), which invokes gen_aa32_frob64() as if
7165 * for an architecturally 64-bit access, but instead do a
7166 * 64-bit access using MO_BE if appropriate and then split
7167 * the two halves.
7168 * This only makes a difference for BE32 user-mode, where
7169 * frob64() must not flip the two halves of the 64-bit data
7170 * but this code must treat BE32 user-mode like BE32 system.
7171 */
7172 TCGv taddr = gen_aa32_addr(s, addr, opc);
7173
7174 tcg_gen_qemu_ld_i64(t64, taddr, get_mem_index(s), opc);
7175 tcg_temp_free(taddr);
7176 tcg_gen_mov_i64(cpu_exclusive_val, t64);
7177 if (s->be_data == MO_BE) {
7178 tcg_gen_extr_i64_i32(tmp2, tmp, t64);
7179 } else {
7180 tcg_gen_extr_i64_i32(tmp, tmp2, t64);
7181 }
7182 tcg_temp_free_i64(t64);
7183
7184 store_reg(s, rt2, tmp2);
7185 } else {
7186 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), opc);
7187 tcg_gen_extu_i32_i64(cpu_exclusive_val, tmp);
7188 }
7189
7190 store_reg(s, rt, tmp);
7191 tcg_gen_extu_i32_i64(cpu_exclusive_addr, addr);
7192}
7193
7194static void gen_clrex(DisasContext *s)
7195{
7196 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
7197}
7198
7199static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
7200 TCGv_i32 addr, int size)
7201{
7202 TCGv_i32 t0, t1, t2;
7203 TCGv_i64 extaddr;
7204 TCGv taddr;
7205 TCGLabel *done_label;
7206 TCGLabel *fail_label;
7207 MemOp opc = size | MO_ALIGN | s->be_data;
7208
7209 /* if (env->exclusive_addr == addr && env->exclusive_val == [addr]) {
7210 [addr] = {Rt};
7211 {Rd} = 0;
7212 } else {
7213 {Rd} = 1;
7214 } */
7215 fail_label = gen_new_label();
7216 done_label = gen_new_label();
7217 extaddr = tcg_temp_new_i64();
7218 tcg_gen_extu_i32_i64(extaddr, addr);
7219 tcg_gen_brcond_i64(TCG_COND_NE, extaddr, cpu_exclusive_addr, fail_label);
7220 tcg_temp_free_i64(extaddr);
7221
7222 taddr = gen_aa32_addr(s, addr, opc);
7223 t0 = tcg_temp_new_i32();
7224 t1 = load_reg(s, rt);
7225 if (size == 3) {
7226 TCGv_i64 o64 = tcg_temp_new_i64();
7227 TCGv_i64 n64 = tcg_temp_new_i64();
7228
7229 t2 = load_reg(s, rt2);
7230 /* For AArch32, architecturally the 32-bit word at the lowest
7231 * address is always Rt and the one at addr+4 is Rt2, even if
7232 * the CPU is big-endian. Since we're going to treat this as a
7233 * single 64-bit BE store, we need to put the two halves in the
7234 * opposite order for BE to LE, so that they end up in the right
7235 * places.
7236 * We don't want gen_aa32_frob64() because that does the wrong
7237 * thing for BE32 usermode.
7238 */
7239 if (s->be_data == MO_BE) {
7240 tcg_gen_concat_i32_i64(n64, t2, t1);
7241 } else {
7242 tcg_gen_concat_i32_i64(n64, t1, t2);
7243 }
7244 tcg_temp_free_i32(t2);
7245
7246 tcg_gen_atomic_cmpxchg_i64(o64, taddr, cpu_exclusive_val, n64,
7247 get_mem_index(s), opc);
7248 tcg_temp_free_i64(n64);
7249
7250 tcg_gen_setcond_i64(TCG_COND_NE, o64, o64, cpu_exclusive_val);
7251 tcg_gen_extrl_i64_i32(t0, o64);
7252
7253 tcg_temp_free_i64(o64);
7254 } else {
7255 t2 = tcg_temp_new_i32();
7256 tcg_gen_extrl_i64_i32(t2, cpu_exclusive_val);
7257 tcg_gen_atomic_cmpxchg_i32(t0, taddr, t2, t1, get_mem_index(s), opc);
7258 tcg_gen_setcond_i32(TCG_COND_NE, t0, t0, t2);
7259 tcg_temp_free_i32(t2);
7260 }
7261 tcg_temp_free_i32(t1);
7262 tcg_temp_free(taddr);
7263 tcg_gen_mov_i32(cpu_R[rd], t0);
7264 tcg_temp_free_i32(t0);
7265 tcg_gen_br(done_label);
7266
7267 gen_set_label(fail_label);
7268 tcg_gen_movi_i32(cpu_R[rd], 1);
7269 gen_set_label(done_label);
7270 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
7271}
7272
7273/* gen_srs:
7274 * @env: CPUARMState
7275 * @s: DisasContext
7276 * @mode: mode field from insn (which stack to store to)
7277 * @amode: addressing mode (DA/IA/DB/IB), encoded as per P,U bits in ARM insn
7278 * @writeback: true if writeback bit set
7279 *
7280 * Generate code for the SRS (Store Return State) insn.
7281 */
7282static void gen_srs(DisasContext *s,
7283 uint32_t mode, uint32_t amode, bool writeback)
7284{
7285 int32_t offset;
7286 TCGv_i32 addr, tmp;
7287 bool undef = false;
7288
7289 /* SRS is:
7290 * - trapped to EL3 if EL3 is AArch64 and we are at Secure EL1
7291 * and specified mode is monitor mode
7292 * - UNDEFINED in Hyp mode
7293 * - UNPREDICTABLE in User or System mode
7294 * - UNPREDICTABLE if the specified mode is:
7295 * -- not implemented
7296 * -- not a valid mode number
7297 * -- a mode that's at a higher exception level
7298 * -- Monitor, if we are Non-secure
7299 * For the UNPREDICTABLE cases we choose to UNDEF.
7300 */
7301 if (s->current_el == 1 && !s->ns && mode == ARM_CPU_MODE_MON) {
7302 gen_exception_insn(s, s->pc_curr, EXCP_UDEF, syn_uncategorized(), 3);
7303 return;
7304 }
7305
7306 if (s->current_el == 0 || s->current_el == 2) {
7307 undef = true;
7308 }
7309
7310 switch (mode) {
7311 case ARM_CPU_MODE_USR:
7312 case ARM_CPU_MODE_FIQ:
7313 case ARM_CPU_MODE_IRQ:
7314 case ARM_CPU_MODE_SVC:
7315 case ARM_CPU_MODE_ABT:
7316 case ARM_CPU_MODE_UND:
7317 case ARM_CPU_MODE_SYS:
7318 break;
7319 case ARM_CPU_MODE_HYP:
7320 if (s->current_el == 1 || !arm_dc_feature(s, ARM_FEATURE_EL2)) {
7321 undef = true;
7322 }
7323 break;
7324 case ARM_CPU_MODE_MON:
7325 /* No need to check specifically for "are we non-secure" because
7326 * we've already made EL0 UNDEF and handled the trap for S-EL1;
7327 * so if this isn't EL3 then we must be non-secure.
7328 */
7329 if (s->current_el != 3) {
7330 undef = true;
7331 }
7332 break;
7333 default:
7334 undef = true;
7335 }
7336
7337 if (undef) {
7338 unallocated_encoding(s);
7339 return;
7340 }
7341
7342 addr = tcg_temp_new_i32();
7343 tmp = tcg_const_i32(mode);
7344 /* get_r13_banked() will raise an exception if called from System mode */
7345 gen_set_condexec(s);
7346 gen_set_pc_im(s, s->pc_curr);
7347 gen_helper_get_r13_banked(addr, cpu_env, tmp);
7348 tcg_temp_free_i32(tmp);
7349 switch (amode) {
7350 case 0: /* DA */
7351 offset = -4;
7352 break;
7353 case 1: /* IA */
7354 offset = 0;
7355 break;
7356 case 2: /* DB */
7357 offset = -8;
7358 break;
7359 case 3: /* IB */
7360 offset = 4;
7361 break;
7362 default:
7363 abort();
7364 }
7365 tcg_gen_addi_i32(addr, addr, offset);
7366 tmp = load_reg(s, 14);
7367 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
7368 tcg_temp_free_i32(tmp);
7369 tmp = load_cpu_field(spsr);
7370 tcg_gen_addi_i32(addr, addr, 4);
7371 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
7372 tcg_temp_free_i32(tmp);
7373 if (writeback) {
7374 switch (amode) {
7375 case 0:
7376 offset = -8;
7377 break;
7378 case 1:
7379 offset = 4;
7380 break;
7381 case 2:
7382 offset = -4;
7383 break;
7384 case 3:
7385 offset = 0;
7386 break;
7387 default:
7388 abort();
7389 }
7390 tcg_gen_addi_i32(addr, addr, offset);
7391 tmp = tcg_const_i32(mode);
7392 gen_helper_set_r13_banked(cpu_env, tmp, addr);
7393 tcg_temp_free_i32(tmp);
7394 }
7395 tcg_temp_free_i32(addr);
7396 s->base.is_jmp = DISAS_UPDATE;
7397}
7398
7399/* Generate a label used for skipping this instruction */
7400static void arm_gen_condlabel(DisasContext *s)
7401{
7402 if (!s->condjmp) {
7403 s->condlabel = gen_new_label();
7404 s->condjmp = 1;
7405 }
7406}
7407
7408/* Skip this instruction if the ARM condition is false */
7409static void arm_skip_unless(DisasContext *s, uint32_t cond)
7410{
7411 arm_gen_condlabel(s);
7412 arm_gen_test_cc(cond ^ 1, s->condlabel);
7413}
7414
7415
7416/*
7417 * Constant expanders for the decoders.
7418 */
7419
7420static int negate(DisasContext *s, int x)
7421{
7422 return -x;
7423}
7424
7425static int plus_2(DisasContext *s, int x)
7426{
7427 return x + 2;
7428}
7429
7430static int times_2(DisasContext *s, int x)
7431{
7432 return x * 2;
7433}
7434
7435static int times_4(DisasContext *s, int x)
7436{
7437 return x * 4;
7438}
7439
7440/* Return only the rotation part of T32ExpandImm. */
7441static int t32_expandimm_rot(DisasContext *s, int x)
7442{
7443 return x & 0xc00 ? extract32(x, 7, 5) : 0;
7444}
7445
7446/* Return the unrotated immediate from T32ExpandImm. */
7447static int t32_expandimm_imm(DisasContext *s, int x)
7448{
7449 int imm = extract32(x, 0, 8);
7450
7451 switch (extract32(x, 8, 4)) {
7452 case 0: /* XY */
7453 /* Nothing to do. */
7454 break;
7455 case 1: /* 00XY00XY */
7456 imm *= 0x00010001;
7457 break;
7458 case 2: /* XY00XY00 */
7459 imm *= 0x01000100;
7460 break;
7461 case 3: /* XYXYXYXY */
7462 imm *= 0x01010101;
7463 break;
7464 default:
7465 /* Rotated constant. */
7466 imm |= 0x80;
7467 break;
7468 }
7469 return imm;
7470}
7471
7472static int t32_branch24(DisasContext *s, int x)
7473{
7474 /* Convert J1:J2 at x[22:21] to I2:I1, which involves I=J^~S. */
7475 x ^= !(x < 0) * (3 << 21);
7476 /* Append the final zero. */
7477 return x << 1;
7478}
7479
7480static int t16_setflags(DisasContext *s)
7481{
7482 return s->condexec_mask == 0;
7483}
7484
7485static int t16_push_list(DisasContext *s, int x)
7486{
7487 return (x & 0xff) | (x & 0x100) << (14 - 8);
7488}
7489
7490static int t16_pop_list(DisasContext *s, int x)
7491{
7492 return (x & 0xff) | (x & 0x100) << (15 - 8);
7493}
7494
7495/*
7496 * Include the generated decoders.
7497 */
7498
7499#include "decode-a32.inc.c"
7500#include "decode-a32-uncond.inc.c"
7501#include "decode-t32.inc.c"
7502#include "decode-t16.inc.c"
7503
7504/* Helpers to swap operands for reverse-subtract. */
7505static void gen_rsb(TCGv_i32 dst, TCGv_i32 a, TCGv_i32 b)
7506{
7507 tcg_gen_sub_i32(dst, b, a);
7508}
7509
7510static void gen_rsb_CC(TCGv_i32 dst, TCGv_i32 a, TCGv_i32 b)
7511{
7512 gen_sub_CC(dst, b, a);
7513}
7514
7515static void gen_rsc(TCGv_i32 dest, TCGv_i32 a, TCGv_i32 b)
7516{
7517 gen_sub_carry(dest, b, a);
7518}
7519
7520static void gen_rsc_CC(TCGv_i32 dest, TCGv_i32 a, TCGv_i32 b)
7521{
7522 gen_sbc_CC(dest, b, a);
7523}
7524
7525/*
7526 * Helpers for the data processing routines.
7527 *
7528 * After the computation store the results back.
7529 * This may be suppressed altogether (STREG_NONE), require a runtime
7530 * check against the stack limits (STREG_SP_CHECK), or generate an
7531 * exception return. Oh, or store into a register.
7532 *
7533 * Always return true, indicating success for a trans_* function.
7534 */
7535typedef enum {
7536 STREG_NONE,
7537 STREG_NORMAL,
7538 STREG_SP_CHECK,
7539 STREG_EXC_RET,
7540} StoreRegKind;
7541
7542static bool store_reg_kind(DisasContext *s, int rd,
7543 TCGv_i32 val, StoreRegKind kind)
7544{
7545 switch (kind) {
7546 case STREG_NONE:
7547 tcg_temp_free_i32(val);
7548 return true;
7549 case STREG_NORMAL:
7550 /* See ALUWritePC: Interworking only from a32 mode. */
7551 if (s->thumb) {
7552 store_reg(s, rd, val);
7553 } else {
7554 store_reg_bx(s, rd, val);
7555 }
7556 return true;
7557 case STREG_SP_CHECK:
7558 store_sp_checked(s, val);
7559 return true;
7560 case STREG_EXC_RET:
7561 gen_exception_return(s, val);
7562 return true;
7563 }
7564 g_assert_not_reached();
7565}
7566
7567/*
7568 * Data Processing (register)
7569 *
7570 * Operate, with set flags, one register source,
7571 * one immediate shifted register source, and a destination.
7572 */
7573static bool op_s_rrr_shi(DisasContext *s, arg_s_rrr_shi *a,
7574 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
7575 int logic_cc, StoreRegKind kind)
7576{
7577 TCGv_i32 tmp1, tmp2;
7578
7579 tmp2 = load_reg(s, a->rm);
7580 gen_arm_shift_im(tmp2, a->shty, a->shim, logic_cc);
7581 tmp1 = load_reg(s, a->rn);
7582
7583 gen(tmp1, tmp1, tmp2);
7584 tcg_temp_free_i32(tmp2);
7585
7586 if (logic_cc) {
7587 gen_logic_CC(tmp1);
7588 }
7589 return store_reg_kind(s, a->rd, tmp1, kind);
7590}
7591
7592static bool op_s_rxr_shi(DisasContext *s, arg_s_rrr_shi *a,
7593 void (*gen)(TCGv_i32, TCGv_i32),
7594 int logic_cc, StoreRegKind kind)
7595{
7596 TCGv_i32 tmp;
7597
7598 tmp = load_reg(s, a->rm);
7599 gen_arm_shift_im(tmp, a->shty, a->shim, logic_cc);
7600
7601 gen(tmp, tmp);
7602 if (logic_cc) {
7603 gen_logic_CC(tmp);
7604 }
7605 return store_reg_kind(s, a->rd, tmp, kind);
7606}
7607
7608/*
7609 * Data-processing (register-shifted register)
7610 *
7611 * Operate, with set flags, one register source,
7612 * one register shifted register source, and a destination.
7613 */
7614static bool op_s_rrr_shr(DisasContext *s, arg_s_rrr_shr *a,
7615 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
7616 int logic_cc, StoreRegKind kind)
7617{
7618 TCGv_i32 tmp1, tmp2;
7619
7620 tmp1 = load_reg(s, a->rs);
7621 tmp2 = load_reg(s, a->rm);
7622 gen_arm_shift_reg(tmp2, a->shty, tmp1, logic_cc);
7623 tmp1 = load_reg(s, a->rn);
7624
7625 gen(tmp1, tmp1, tmp2);
7626 tcg_temp_free_i32(tmp2);
7627
7628 if (logic_cc) {
7629 gen_logic_CC(tmp1);
7630 }
7631 return store_reg_kind(s, a->rd, tmp1, kind);
7632}
7633
7634static bool op_s_rxr_shr(DisasContext *s, arg_s_rrr_shr *a,
7635 void (*gen)(TCGv_i32, TCGv_i32),
7636 int logic_cc, StoreRegKind kind)
7637{
7638 TCGv_i32 tmp1, tmp2;
7639
7640 tmp1 = load_reg(s, a->rs);
7641 tmp2 = load_reg(s, a->rm);
7642 gen_arm_shift_reg(tmp2, a->shty, tmp1, logic_cc);
7643
7644 gen(tmp2, tmp2);
7645 if (logic_cc) {
7646 gen_logic_CC(tmp2);
7647 }
7648 return store_reg_kind(s, a->rd, tmp2, kind);
7649}
7650
7651/*
7652 * Data-processing (immediate)
7653 *
7654 * Operate, with set flags, one register source,
7655 * one rotated immediate, and a destination.
7656 *
7657 * Note that logic_cc && a->rot setting CF based on the msb of the
7658 * immediate is the reason why we must pass in the unrotated form
7659 * of the immediate.
7660 */
7661static bool op_s_rri_rot(DisasContext *s, arg_s_rri_rot *a,
7662 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
7663 int logic_cc, StoreRegKind kind)
7664{
7665 TCGv_i32 tmp1, tmp2;
7666 uint32_t imm;
7667
7668 imm = ror32(a->imm, a->rot);
7669 if (logic_cc && a->rot) {
7670 tcg_gen_movi_i32(cpu_CF, imm >> 31);
7671 }
7672 tmp2 = tcg_const_i32(imm);
7673 tmp1 = load_reg(s, a->rn);
7674
7675 gen(tmp1, tmp1, tmp2);
7676 tcg_temp_free_i32(tmp2);
7677
7678 if (logic_cc) {
7679 gen_logic_CC(tmp1);
7680 }
7681 return store_reg_kind(s, a->rd, tmp1, kind);
7682}
7683
7684static bool op_s_rxi_rot(DisasContext *s, arg_s_rri_rot *a,
7685 void (*gen)(TCGv_i32, TCGv_i32),
7686 int logic_cc, StoreRegKind kind)
7687{
7688 TCGv_i32 tmp;
7689 uint32_t imm;
7690
7691 imm = ror32(a->imm, a->rot);
7692 if (logic_cc && a->rot) {
7693 tcg_gen_movi_i32(cpu_CF, imm >> 31);
7694 }
7695 tmp = tcg_const_i32(imm);
7696
7697 gen(tmp, tmp);
7698 if (logic_cc) {
7699 gen_logic_CC(tmp);
7700 }
7701 return store_reg_kind(s, a->rd, tmp, kind);
7702}
7703
7704#define DO_ANY3(NAME, OP, L, K) \
7705 static bool trans_##NAME##_rrri(DisasContext *s, arg_s_rrr_shi *a) \
7706 { StoreRegKind k = (K); return op_s_rrr_shi(s, a, OP, L, k); } \
7707 static bool trans_##NAME##_rrrr(DisasContext *s, arg_s_rrr_shr *a) \
7708 { StoreRegKind k = (K); return op_s_rrr_shr(s, a, OP, L, k); } \
7709 static bool trans_##NAME##_rri(DisasContext *s, arg_s_rri_rot *a) \
7710 { StoreRegKind k = (K); return op_s_rri_rot(s, a, OP, L, k); }
7711
7712#define DO_ANY2(NAME, OP, L, K) \
7713 static bool trans_##NAME##_rxri(DisasContext *s, arg_s_rrr_shi *a) \
7714 { StoreRegKind k = (K); return op_s_rxr_shi(s, a, OP, L, k); } \
7715 static bool trans_##NAME##_rxrr(DisasContext *s, arg_s_rrr_shr *a) \
7716 { StoreRegKind k = (K); return op_s_rxr_shr(s, a, OP, L, k); } \
7717 static bool trans_##NAME##_rxi(DisasContext *s, arg_s_rri_rot *a) \
7718 { StoreRegKind k = (K); return op_s_rxi_rot(s, a, OP, L, k); }
7719
7720#define DO_CMP2(NAME, OP, L) \
7721 static bool trans_##NAME##_xrri(DisasContext *s, arg_s_rrr_shi *a) \
7722 { return op_s_rrr_shi(s, a, OP, L, STREG_NONE); } \
7723 static bool trans_##NAME##_xrrr(DisasContext *s, arg_s_rrr_shr *a) \
7724 { return op_s_rrr_shr(s, a, OP, L, STREG_NONE); } \
7725 static bool trans_##NAME##_xri(DisasContext *s, arg_s_rri_rot *a) \
7726 { return op_s_rri_rot(s, a, OP, L, STREG_NONE); }
7727
7728DO_ANY3(AND, tcg_gen_and_i32, a->s, STREG_NORMAL)
7729DO_ANY3(EOR, tcg_gen_xor_i32, a->s, STREG_NORMAL)
7730DO_ANY3(ORR, tcg_gen_or_i32, a->s, STREG_NORMAL)
7731DO_ANY3(BIC, tcg_gen_andc_i32, a->s, STREG_NORMAL)
7732
7733DO_ANY3(RSB, a->s ? gen_rsb_CC : gen_rsb, false, STREG_NORMAL)
7734DO_ANY3(ADC, a->s ? gen_adc_CC : gen_add_carry, false, STREG_NORMAL)
7735DO_ANY3(SBC, a->s ? gen_sbc_CC : gen_sub_carry, false, STREG_NORMAL)
7736DO_ANY3(RSC, a->s ? gen_rsc_CC : gen_rsc, false, STREG_NORMAL)
7737
7738DO_CMP2(TST, tcg_gen_and_i32, true)
7739DO_CMP2(TEQ, tcg_gen_xor_i32, true)
7740DO_CMP2(CMN, gen_add_CC, false)
7741DO_CMP2(CMP, gen_sub_CC, false)
7742
7743DO_ANY3(ADD, a->s ? gen_add_CC : tcg_gen_add_i32, false,
7744 a->rd == 13 && a->rn == 13 ? STREG_SP_CHECK : STREG_NORMAL)
7745
7746/*
7747 * Note for the computation of StoreRegKind we return out of the
7748 * middle of the functions that are expanded by DO_ANY3, and that
7749 * we modify a->s via that parameter before it is used by OP.
7750 */
7751DO_ANY3(SUB, a->s ? gen_sub_CC : tcg_gen_sub_i32, false,
7752 ({
7753 StoreRegKind ret = STREG_NORMAL;
7754 if (a->rd == 15 && a->s) {
7755 /*
7756 * See ALUExceptionReturn:
7757 * In User mode, UNPREDICTABLE; we choose UNDEF.
7758 * In Hyp mode, UNDEFINED.
7759 */
7760 if (IS_USER(s) || s->current_el == 2) {
7761 unallocated_encoding(s);
7762 return true;
7763 }
7764 /* There is no writeback of nzcv to PSTATE. */
7765 a->s = 0;
7766 ret = STREG_EXC_RET;
7767 } else if (a->rd == 13 && a->rn == 13) {
7768 ret = STREG_SP_CHECK;
7769 }
7770 ret;
7771 }))
7772
7773DO_ANY2(MOV, tcg_gen_mov_i32, a->s,
7774 ({
7775 StoreRegKind ret = STREG_NORMAL;
7776 if (a->rd == 15 && a->s) {
7777 /*
7778 * See ALUExceptionReturn:
7779 * In User mode, UNPREDICTABLE; we choose UNDEF.
7780 * In Hyp mode, UNDEFINED.
7781 */
7782 if (IS_USER(s) || s->current_el == 2) {
7783 unallocated_encoding(s);
7784 return true;
7785 }
7786 /* There is no writeback of nzcv to PSTATE. */
7787 a->s = 0;
7788 ret = STREG_EXC_RET;
7789 } else if (a->rd == 13) {
7790 ret = STREG_SP_CHECK;
7791 }
7792 ret;
7793 }))
7794
7795DO_ANY2(MVN, tcg_gen_not_i32, a->s, STREG_NORMAL)
7796
7797/*
7798 * ORN is only available with T32, so there is no register-shifted-register
7799 * form of the insn. Using the DO_ANY3 macro would create an unused function.
7800 */
7801static bool trans_ORN_rrri(DisasContext *s, arg_s_rrr_shi *a)
7802{
7803 return op_s_rrr_shi(s, a, tcg_gen_orc_i32, a->s, STREG_NORMAL);
7804}
7805
7806static bool trans_ORN_rri(DisasContext *s, arg_s_rri_rot *a)
7807{
7808 return op_s_rri_rot(s, a, tcg_gen_orc_i32, a->s, STREG_NORMAL);
7809}
7810
7811#undef DO_ANY3
7812#undef DO_ANY2
7813#undef DO_CMP2
7814
7815static bool trans_ADR(DisasContext *s, arg_ri *a)
7816{
7817 store_reg_bx(s, a->rd, add_reg_for_lit(s, 15, a->imm));
7818 return true;
7819}
7820
7821static bool trans_MOVW(DisasContext *s, arg_MOVW *a)
7822{
7823 TCGv_i32 tmp;
7824
7825 if (!ENABLE_ARCH_6T2) {
7826 return false;
7827 }
7828
7829 tmp = tcg_const_i32(a->imm);
7830 store_reg(s, a->rd, tmp);
7831 return true;
7832}
7833
7834static bool trans_MOVT(DisasContext *s, arg_MOVW *a)
7835{
7836 TCGv_i32 tmp;
7837
7838 if (!ENABLE_ARCH_6T2) {
7839 return false;
7840 }
7841
7842 tmp = load_reg(s, a->rd);
7843 tcg_gen_ext16u_i32(tmp, tmp);
7844 tcg_gen_ori_i32(tmp, tmp, a->imm << 16);
7845 store_reg(s, a->rd, tmp);
7846 return true;
7847}
7848
7849/*
7850 * Multiply and multiply accumulate
7851 */
7852
7853static bool op_mla(DisasContext *s, arg_s_rrrr *a, bool add)
7854{
7855 TCGv_i32 t1, t2;
7856
7857 t1 = load_reg(s, a->rn);
7858 t2 = load_reg(s, a->rm);
7859 tcg_gen_mul_i32(t1, t1, t2);
7860 tcg_temp_free_i32(t2);
7861 if (add) {
7862 t2 = load_reg(s, a->ra);
7863 tcg_gen_add_i32(t1, t1, t2);
7864 tcg_temp_free_i32(t2);
7865 }
7866 if (a->s) {
7867 gen_logic_CC(t1);
7868 }
7869 store_reg(s, a->rd, t1);
7870 return true;
7871}
7872
7873static bool trans_MUL(DisasContext *s, arg_MUL *a)
7874{
7875 return op_mla(s, a, false);
7876}
7877
7878static bool trans_MLA(DisasContext *s, arg_MLA *a)
7879{
7880 return op_mla(s, a, true);
7881}
7882
7883static bool trans_MLS(DisasContext *s, arg_MLS *a)
7884{
7885 TCGv_i32 t1, t2;
7886
7887 if (!ENABLE_ARCH_6T2) {
7888 return false;
7889 }
7890 t1 = load_reg(s, a->rn);
7891 t2 = load_reg(s, a->rm);
7892 tcg_gen_mul_i32(t1, t1, t2);
7893 tcg_temp_free_i32(t2);
7894 t2 = load_reg(s, a->ra);
7895 tcg_gen_sub_i32(t1, t2, t1);
7896 tcg_temp_free_i32(t2);
7897 store_reg(s, a->rd, t1);
7898 return true;
7899}
7900
7901static bool op_mlal(DisasContext *s, arg_s_rrrr *a, bool uns, bool add)
7902{
7903 TCGv_i32 t0, t1, t2, t3;
7904
7905 t0 = load_reg(s, a->rm);
7906 t1 = load_reg(s, a->rn);
7907 if (uns) {
7908 tcg_gen_mulu2_i32(t0, t1, t0, t1);
7909 } else {
7910 tcg_gen_muls2_i32(t0, t1, t0, t1);
7911 }
7912 if (add) {
7913 t2 = load_reg(s, a->ra);
7914 t3 = load_reg(s, a->rd);
7915 tcg_gen_add2_i32(t0, t1, t0, t1, t2, t3);
7916 tcg_temp_free_i32(t2);
7917 tcg_temp_free_i32(t3);
7918 }
7919 if (a->s) {
7920 gen_logicq_cc(t0, t1);
7921 }
7922 store_reg(s, a->ra, t0);
7923 store_reg(s, a->rd, t1);
7924 return true;
7925}
7926
7927static bool trans_UMULL(DisasContext *s, arg_UMULL *a)
7928{
7929 return op_mlal(s, a, true, false);
7930}
7931
7932static bool trans_SMULL(DisasContext *s, arg_SMULL *a)
7933{
7934 return op_mlal(s, a, false, false);
7935}
7936
7937static bool trans_UMLAL(DisasContext *s, arg_UMLAL *a)
7938{
7939 return op_mlal(s, a, true, true);
7940}
7941
7942static bool trans_SMLAL(DisasContext *s, arg_SMLAL *a)
7943{
7944 return op_mlal(s, a, false, true);
7945}
7946
7947static bool trans_UMAAL(DisasContext *s, arg_UMAAL *a)
7948{
7949 TCGv_i32 t0, t1, t2, zero;
7950
7951 if (s->thumb
7952 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
7953 : !ENABLE_ARCH_6) {
7954 return false;
7955 }
7956
7957 t0 = load_reg(s, a->rm);
7958 t1 = load_reg(s, a->rn);
7959 tcg_gen_mulu2_i32(t0, t1, t0, t1);
7960 zero = tcg_const_i32(0);
7961 t2 = load_reg(s, a->ra);
7962 tcg_gen_add2_i32(t0, t1, t0, t1, t2, zero);
7963 tcg_temp_free_i32(t2);
7964 t2 = load_reg(s, a->rd);
7965 tcg_gen_add2_i32(t0, t1, t0, t1, t2, zero);
7966 tcg_temp_free_i32(t2);
7967 tcg_temp_free_i32(zero);
7968 store_reg(s, a->ra, t0);
7969 store_reg(s, a->rd, t1);
7970 return true;
7971}
7972
7973/*
7974 * Saturating addition and subtraction
7975 */
7976
7977static bool op_qaddsub(DisasContext *s, arg_rrr *a, bool add, bool doub)
7978{
7979 TCGv_i32 t0, t1;
7980
7981 if (s->thumb
7982 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
7983 : !ENABLE_ARCH_5TE) {
7984 return false;
7985 }
7986
7987 t0 = load_reg(s, a->rm);
7988 t1 = load_reg(s, a->rn);
7989 if (doub) {
7990 gen_helper_add_saturate(t1, cpu_env, t1, t1);
7991 }
7992 if (add) {
7993 gen_helper_add_saturate(t0, cpu_env, t0, t1);
7994 } else {
7995 gen_helper_sub_saturate(t0, cpu_env, t0, t1);
7996 }
7997 tcg_temp_free_i32(t1);
7998 store_reg(s, a->rd, t0);
7999 return true;
8000}
8001
8002#define DO_QADDSUB(NAME, ADD, DOUB) \
8003static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
8004{ \
8005 return op_qaddsub(s, a, ADD, DOUB); \
8006}
8007
8008DO_QADDSUB(QADD, true, false)
8009DO_QADDSUB(QSUB, false, false)
8010DO_QADDSUB(QDADD, true, true)
8011DO_QADDSUB(QDSUB, false, true)
8012
8013#undef DO_QADDSUB
8014
8015/*
8016 * Halfword multiply and multiply accumulate
8017 */
8018
8019static bool op_smlaxxx(DisasContext *s, arg_rrrr *a,
8020 int add_long, bool nt, bool mt)
8021{
8022 TCGv_i32 t0, t1, tl, th;
8023
8024 if (s->thumb
8025 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
8026 : !ENABLE_ARCH_5TE) {
8027 return false;
8028 }
8029
8030 t0 = load_reg(s, a->rn);
8031 t1 = load_reg(s, a->rm);
8032 gen_mulxy(t0, t1, nt, mt);
8033 tcg_temp_free_i32(t1);
8034
8035 switch (add_long) {
8036 case 0:
8037 store_reg(s, a->rd, t0);
8038 break;
8039 case 1:
8040 t1 = load_reg(s, a->ra);
8041 gen_helper_add_setq(t0, cpu_env, t0, t1);
8042 tcg_temp_free_i32(t1);
8043 store_reg(s, a->rd, t0);
8044 break;
8045 case 2:
8046 tl = load_reg(s, a->ra);
8047 th = load_reg(s, a->rd);
8048 t1 = tcg_const_i32(0);
8049 tcg_gen_add2_i32(tl, th, tl, th, t0, t1);
8050 tcg_temp_free_i32(t0);
8051 tcg_temp_free_i32(t1);
8052 store_reg(s, a->ra, tl);
8053 store_reg(s, a->rd, th);
8054 break;
8055 default:
8056 g_assert_not_reached();
8057 }
8058 return true;
8059}
8060
8061#define DO_SMLAX(NAME, add, nt, mt) \
8062static bool trans_##NAME(DisasContext *s, arg_rrrr *a) \
8063{ \
8064 return op_smlaxxx(s, a, add, nt, mt); \
8065}
8066
8067DO_SMLAX(SMULBB, 0, 0, 0)
8068DO_SMLAX(SMULBT, 0, 0, 1)
8069DO_SMLAX(SMULTB, 0, 1, 0)
8070DO_SMLAX(SMULTT, 0, 1, 1)
8071
8072DO_SMLAX(SMLABB, 1, 0, 0)
8073DO_SMLAX(SMLABT, 1, 0, 1)
8074DO_SMLAX(SMLATB, 1, 1, 0)
8075DO_SMLAX(SMLATT, 1, 1, 1)
8076
8077DO_SMLAX(SMLALBB, 2, 0, 0)
8078DO_SMLAX(SMLALBT, 2, 0, 1)
8079DO_SMLAX(SMLALTB, 2, 1, 0)
8080DO_SMLAX(SMLALTT, 2, 1, 1)
8081
8082#undef DO_SMLAX
8083
8084static bool op_smlawx(DisasContext *s, arg_rrrr *a, bool add, bool mt)
8085{
8086 TCGv_i32 t0, t1;
8087
8088 if (!ENABLE_ARCH_5TE) {
8089 return false;
8090 }
8091
8092 t0 = load_reg(s, a->rn);
8093 t1 = load_reg(s, a->rm);
8094 /*
8095 * Since the nominal result is product<47:16>, shift the 16-bit
8096 * input up by 16 bits, so that the result is at product<63:32>.
8097 */
8098 if (mt) {
8099 tcg_gen_andi_i32(t1, t1, 0xffff0000);
8100 } else {
8101 tcg_gen_shli_i32(t1, t1, 16);
8102 }
8103 tcg_gen_muls2_i32(t0, t1, t0, t1);
8104 tcg_temp_free_i32(t0);
8105 if (add) {
8106 t0 = load_reg(s, a->ra);
8107 gen_helper_add_setq(t1, cpu_env, t1, t0);
8108 tcg_temp_free_i32(t0);
8109 }
8110 store_reg(s, a->rd, t1);
8111 return true;
8112}
8113
8114#define DO_SMLAWX(NAME, add, mt) \
8115static bool trans_##NAME(DisasContext *s, arg_rrrr *a) \
8116{ \
8117 return op_smlawx(s, a, add, mt); \
8118}
8119
8120DO_SMLAWX(SMULWB, 0, 0)
8121DO_SMLAWX(SMULWT, 0, 1)
8122DO_SMLAWX(SMLAWB, 1, 0)
8123DO_SMLAWX(SMLAWT, 1, 1)
8124
8125#undef DO_SMLAWX
8126
8127/*
8128 * MSR (immediate) and hints
8129 */
8130
8131static bool trans_YIELD(DisasContext *s, arg_YIELD *a)
8132{
8133 /*
8134 * When running single-threaded TCG code, use the helper to ensure that
8135 * the next round-robin scheduled vCPU gets a crack. When running in
8136 * MTTCG we don't generate jumps to the helper as it won't affect the
8137 * scheduling of other vCPUs.
8138 */
8139 if (!(tb_cflags(s->base.tb) & CF_PARALLEL)) {
8140 gen_set_pc_im(s, s->base.pc_next);
8141 s->base.is_jmp = DISAS_YIELD;
8142 }
8143 return true;
8144}
8145
8146static bool trans_WFE(DisasContext *s, arg_WFE *a)
8147{
8148 /*
8149 * When running single-threaded TCG code, use the helper to ensure that
8150 * the next round-robin scheduled vCPU gets a crack. In MTTCG mode we
8151 * just skip this instruction. Currently the SEV/SEVL instructions,
8152 * which are *one* of many ways to wake the CPU from WFE, are not
8153 * implemented so we can't sleep like WFI does.
8154 */
8155 if (!(tb_cflags(s->base.tb) & CF_PARALLEL)) {
8156 gen_set_pc_im(s, s->base.pc_next);
8157 s->base.is_jmp = DISAS_WFE;
8158 }
8159 return true;
8160}
8161
8162static bool trans_WFI(DisasContext *s, arg_WFI *a)
8163{
8164 /* For WFI, halt the vCPU until an IRQ. */
8165 gen_set_pc_im(s, s->base.pc_next);
8166 s->base.is_jmp = DISAS_WFI;
8167 return true;
8168}
8169
8170static bool trans_NOP(DisasContext *s, arg_NOP *a)
8171{
8172 return true;
8173}
8174
8175static bool trans_MSR_imm(DisasContext *s, arg_MSR_imm *a)
8176{
8177 uint32_t val = ror32(a->imm, a->rot * 2);
8178 uint32_t mask = msr_mask(s, a->mask, a->r);
8179
8180 if (gen_set_psr_im(s, mask, a->r, val)) {
8181 unallocated_encoding(s);
8182 }
8183 return true;
8184}
8185
8186/*
8187 * Cyclic Redundancy Check
8188 */
8189
8190static bool op_crc32(DisasContext *s, arg_rrr *a, bool c, MemOp sz)
8191{
8192 TCGv_i32 t1, t2, t3;
8193
8194 if (!dc_isar_feature(aa32_crc32, s)) {
8195 return false;
8196 }
8197
8198 t1 = load_reg(s, a->rn);
8199 t2 = load_reg(s, a->rm);
8200 switch (sz) {
8201 case MO_8:
8202 gen_uxtb(t2);
8203 break;
8204 case MO_16:
8205 gen_uxth(t2);
8206 break;
8207 case MO_32:
8208 break;
8209 default:
8210 g_assert_not_reached();
8211 }
8212 t3 = tcg_const_i32(1 << sz);
8213 if (c) {
8214 gen_helper_crc32c(t1, t1, t2, t3);
8215 } else {
8216 gen_helper_crc32(t1, t1, t2, t3);
8217 }
8218 tcg_temp_free_i32(t2);
8219 tcg_temp_free_i32(t3);
8220 store_reg(s, a->rd, t1);
8221 return true;
8222}
8223
8224#define DO_CRC32(NAME, c, sz) \
8225static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
8226 { return op_crc32(s, a, c, sz); }
8227
8228DO_CRC32(CRC32B, false, MO_8)
8229DO_CRC32(CRC32H, false, MO_16)
8230DO_CRC32(CRC32W, false, MO_32)
8231DO_CRC32(CRC32CB, true, MO_8)
8232DO_CRC32(CRC32CH, true, MO_16)
8233DO_CRC32(CRC32CW, true, MO_32)
8234
8235#undef DO_CRC32
8236
8237/*
8238 * Miscellaneous instructions
8239 */
8240
8241static bool trans_MRS_bank(DisasContext *s, arg_MRS_bank *a)
8242{
8243 if (arm_dc_feature(s, ARM_FEATURE_M)) {
8244 return false;
8245 }
8246 gen_mrs_banked(s, a->r, a->sysm, a->rd);
8247 return true;
8248}
8249
8250static bool trans_MSR_bank(DisasContext *s, arg_MSR_bank *a)
8251{
8252 if (arm_dc_feature(s, ARM_FEATURE_M)) {
8253 return false;
8254 }
8255 gen_msr_banked(s, a->r, a->sysm, a->rn);
8256 return true;
8257}
8258
8259static bool trans_MRS_reg(DisasContext *s, arg_MRS_reg *a)
8260{
8261 TCGv_i32 tmp;
8262
8263 if (arm_dc_feature(s, ARM_FEATURE_M)) {
8264 return false;
8265 }
8266 if (a->r) {
8267 if (IS_USER(s)) {
8268 unallocated_encoding(s);
8269 return true;
8270 }
8271 tmp = load_cpu_field(spsr);
8272 } else {
8273 tmp = tcg_temp_new_i32();
8274 gen_helper_cpsr_read(tmp, cpu_env);
8275 }
8276 store_reg(s, a->rd, tmp);
8277 return true;
8278}
8279
8280static bool trans_MSR_reg(DisasContext *s, arg_MSR_reg *a)
8281{
8282 TCGv_i32 tmp;
8283 uint32_t mask = msr_mask(s, a->mask, a->r);
8284
8285 if (arm_dc_feature(s, ARM_FEATURE_M)) {
8286 return false;
8287 }
8288 tmp = load_reg(s, a->rn);
8289 if (gen_set_psr(s, mask, a->r, tmp)) {
8290 unallocated_encoding(s);
8291 }
8292 return true;
8293}
8294
8295static bool trans_MRS_v7m(DisasContext *s, arg_MRS_v7m *a)
8296{
8297 TCGv_i32 tmp;
8298
8299 if (!arm_dc_feature(s, ARM_FEATURE_M)) {
8300 return false;
8301 }
8302 tmp = tcg_const_i32(a->sysm);
8303 gen_helper_v7m_mrs(tmp, cpu_env, tmp);
8304 store_reg(s, a->rd, tmp);
8305 return true;
8306}
8307
8308static bool trans_MSR_v7m(DisasContext *s, arg_MSR_v7m *a)
8309{
8310 TCGv_i32 addr, reg;
8311
8312 if (!arm_dc_feature(s, ARM_FEATURE_M)) {
8313 return false;
8314 }
8315 addr = tcg_const_i32((a->mask << 10) | a->sysm);
8316 reg = load_reg(s, a->rn);
8317 gen_helper_v7m_msr(cpu_env, addr, reg);
8318 tcg_temp_free_i32(addr);
8319 tcg_temp_free_i32(reg);
8320 gen_lookup_tb(s);
8321 return true;
8322}
8323
8324static bool trans_BX(DisasContext *s, arg_BX *a)
8325{
8326 if (!ENABLE_ARCH_4T) {
8327 return false;
8328 }
8329 gen_bx_excret(s, load_reg(s, a->rm));
8330 return true;
8331}
8332
8333static bool trans_BXJ(DisasContext *s, arg_BXJ *a)
8334{
8335 if (!ENABLE_ARCH_5J || arm_dc_feature(s, ARM_FEATURE_M)) {
8336 return false;
8337 }
8338 /* Trivial implementation equivalent to bx. */
8339 gen_bx(s, load_reg(s, a->rm));
8340 return true;
8341}
8342
8343static bool trans_BLX_r(DisasContext *s, arg_BLX_r *a)
8344{
8345 TCGv_i32 tmp;
8346
8347 if (!ENABLE_ARCH_5) {
8348 return false;
8349 }
8350 tmp = load_reg(s, a->rm);
8351 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | s->thumb);
8352 gen_bx(s, tmp);
8353 return true;
8354}
8355
8356/*
8357 * BXNS/BLXNS: only exist for v8M with the security extensions,
8358 * and always UNDEF if NonSecure. We don't implement these in
8359 * the user-only mode either (in theory you can use them from
8360 * Secure User mode but they are too tied in to system emulation).
8361 */
8362static bool trans_BXNS(DisasContext *s, arg_BXNS *a)
8363{
8364 if (!s->v8m_secure || IS_USER_ONLY) {
8365 unallocated_encoding(s);
8366 } else {
8367 gen_bxns(s, a->rm);
8368 }
8369 return true;
8370}
8371
8372static bool trans_BLXNS(DisasContext *s, arg_BLXNS *a)
8373{
8374 if (!s->v8m_secure || IS_USER_ONLY) {
8375 unallocated_encoding(s);
8376 } else {
8377 gen_blxns(s, a->rm);
8378 }
8379 return true;
8380}
8381
8382static bool trans_CLZ(DisasContext *s, arg_CLZ *a)
8383{
8384 TCGv_i32 tmp;
8385
8386 if (!ENABLE_ARCH_5) {
8387 return false;
8388 }
8389 tmp = load_reg(s, a->rm);
8390 tcg_gen_clzi_i32(tmp, tmp, 32);
8391 store_reg(s, a->rd, tmp);
8392 return true;
8393}
8394
8395static bool trans_ERET(DisasContext *s, arg_ERET *a)
8396{
8397 TCGv_i32 tmp;
8398
8399 if (!arm_dc_feature(s, ARM_FEATURE_V7VE)) {
8400 return false;
8401 }
8402 if (IS_USER(s)) {
8403 unallocated_encoding(s);
8404 return true;
8405 }
8406 if (s->current_el == 2) {
8407 /* ERET from Hyp uses ELR_Hyp, not LR */
8408 tmp = load_cpu_field(elr_el[2]);
8409 } else {
8410 tmp = load_reg(s, 14);
8411 }
8412 gen_exception_return(s, tmp);
8413 return true;
8414}
8415
8416static bool trans_HLT(DisasContext *s, arg_HLT *a)
8417{
8418 gen_hlt(s, a->imm);
8419 return true;
8420}
8421
8422static bool trans_BKPT(DisasContext *s, arg_BKPT *a)
8423{
8424 if (!ENABLE_ARCH_5) {
8425 return false;
8426 }
8427 gen_exception_bkpt_insn(s, syn_aa32_bkpt(a->imm, false));
8428 return true;
8429}
8430
8431static bool trans_HVC(DisasContext *s, arg_HVC *a)
8432{
8433 if (!ENABLE_ARCH_7 || arm_dc_feature(s, ARM_FEATURE_M)) {
8434 return false;
8435 }
8436 if (IS_USER(s)) {
8437 unallocated_encoding(s);
8438 } else {
8439 gen_hvc(s, a->imm);
8440 }
8441 return true;
8442}
8443
8444static bool trans_SMC(DisasContext *s, arg_SMC *a)
8445{
8446 if (!ENABLE_ARCH_6K || arm_dc_feature(s, ARM_FEATURE_M)) {
8447 return false;
8448 }
8449 if (IS_USER(s)) {
8450 unallocated_encoding(s);
8451 } else {
8452 gen_smc(s);
8453 }
8454 return true;
8455}
8456
8457static bool trans_SG(DisasContext *s, arg_SG *a)
8458{
8459 if (!arm_dc_feature(s, ARM_FEATURE_M) ||
8460 !arm_dc_feature(s, ARM_FEATURE_V8)) {
8461 return false;
8462 }
8463 /*
8464 * SG (v8M only)
8465 * The bulk of the behaviour for this instruction is implemented
8466 * in v7m_handle_execute_nsc(), which deals with the insn when
8467 * it is executed by a CPU in non-secure state from memory
8468 * which is Secure & NonSecure-Callable.
8469 * Here we only need to handle the remaining cases:
8470 * * in NS memory (including the "security extension not
8471 * implemented" case) : NOP
8472 * * in S memory but CPU already secure (clear IT bits)
8473 * We know that the attribute for the memory this insn is
8474 * in must match the current CPU state, because otherwise
8475 * get_phys_addr_pmsav8 would have generated an exception.
8476 */
8477 if (s->v8m_secure) {
8478 /* Like the IT insn, we don't need to generate any code */
8479 s->condexec_cond = 0;
8480 s->condexec_mask = 0;
8481 }
8482 return true;
8483}
8484
8485static bool trans_TT(DisasContext *s, arg_TT *a)
8486{
8487 TCGv_i32 addr, tmp;
8488
8489 if (!arm_dc_feature(s, ARM_FEATURE_M) ||
8490 !arm_dc_feature(s, ARM_FEATURE_V8)) {
8491 return false;
8492 }
8493 if (a->rd == 13 || a->rd == 15 || a->rn == 15) {
8494 /* We UNDEF for these UNPREDICTABLE cases */
8495 unallocated_encoding(s);
8496 return true;
8497 }
8498 if (a->A && !s->v8m_secure) {
8499 /* This case is UNDEFINED. */
8500 unallocated_encoding(s);
8501 return true;
8502 }
8503
8504 addr = load_reg(s, a->rn);
8505 tmp = tcg_const_i32((a->A << 1) | a->T);
8506 gen_helper_v7m_tt(tmp, cpu_env, addr, tmp);
8507 tcg_temp_free_i32(addr);
8508 store_reg(s, a->rd, tmp);
8509 return true;
8510}
8511
8512/*
8513 * Load/store register index
8514 */
8515
8516static ISSInfo make_issinfo(DisasContext *s, int rd, bool p, bool w)
8517{
8518 ISSInfo ret;
8519
8520 /* ISS not valid if writeback */
8521 if (p && !w) {
8522 ret = rd;
8523 } else {
8524 ret = ISSInvalid;
8525 }
8526 return ret;
8527}
8528
8529static TCGv_i32 op_addr_rr_pre(DisasContext *s, arg_ldst_rr *a)
8530{
8531 TCGv_i32 addr = load_reg(s, a->rn);
8532
8533 if (s->v8m_stackcheck && a->rn == 13 && a->w) {
8534 gen_helper_v8m_stackcheck(cpu_env, addr);
8535 }
8536
8537 if (a->p) {
8538 TCGv_i32 ofs = load_reg(s, a->rm);
8539 gen_arm_shift_im(ofs, a->shtype, a->shimm, 0);
8540 if (a->u) {
8541 tcg_gen_add_i32(addr, addr, ofs);
8542 } else {
8543 tcg_gen_sub_i32(addr, addr, ofs);
8544 }
8545 tcg_temp_free_i32(ofs);
8546 }
8547 return addr;
8548}
8549
8550static void op_addr_rr_post(DisasContext *s, arg_ldst_rr *a,
8551 TCGv_i32 addr, int address_offset)
8552{
8553 if (!a->p) {
8554 TCGv_i32 ofs = load_reg(s, a->rm);
8555 gen_arm_shift_im(ofs, a->shtype, a->shimm, 0);
8556 if (a->u) {
8557 tcg_gen_add_i32(addr, addr, ofs);
8558 } else {
8559 tcg_gen_sub_i32(addr, addr, ofs);
8560 }
8561 tcg_temp_free_i32(ofs);
8562 } else if (!a->w) {
8563 tcg_temp_free_i32(addr);
8564 return;
8565 }
8566 tcg_gen_addi_i32(addr, addr, address_offset);
8567 store_reg(s, a->rn, addr);
8568}
8569
8570static bool op_load_rr(DisasContext *s, arg_ldst_rr *a,
8571 MemOp mop, int mem_idx)
8572{
8573 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w);
8574 TCGv_i32 addr, tmp;
8575
8576 addr = op_addr_rr_pre(s, a);
8577
8578 tmp = tcg_temp_new_i32();
8579 gen_aa32_ld_i32(s, tmp, addr, mem_idx, mop | s->be_data);
8580 disas_set_da_iss(s, mop, issinfo);
8581
8582 /*
8583 * Perform base writeback before the loaded value to
8584 * ensure correct behavior with overlapping index registers.
8585 */
8586 op_addr_rr_post(s, a, addr, 0);
8587 store_reg_from_load(s, a->rt, tmp);
8588 return true;
8589}
8590
8591static bool op_store_rr(DisasContext *s, arg_ldst_rr *a,
8592 MemOp mop, int mem_idx)
8593{
8594 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w) | ISSIsWrite;
8595 TCGv_i32 addr, tmp;
8596
8597 addr = op_addr_rr_pre(s, a);
8598
8599 tmp = load_reg(s, a->rt);
8600 gen_aa32_st_i32(s, tmp, addr, mem_idx, mop | s->be_data);
8601 disas_set_da_iss(s, mop, issinfo);
8602 tcg_temp_free_i32(tmp);
8603
8604 op_addr_rr_post(s, a, addr, 0);
8605 return true;
8606}
8607
8608static bool trans_LDRD_rr(DisasContext *s, arg_ldst_rr *a)
8609{
8610 int mem_idx = get_mem_index(s);
8611 TCGv_i32 addr, tmp;
8612
8613 if (!ENABLE_ARCH_5TE) {
8614 return false;
8615 }
8616 if (a->rt & 1) {
8617 unallocated_encoding(s);
8618 return true;
8619 }
8620 addr = op_addr_rr_pre(s, a);
8621
8622 tmp = tcg_temp_new_i32();
8623 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8624 store_reg(s, a->rt, tmp);
8625
8626 tcg_gen_addi_i32(addr, addr, 4);
8627
8628 tmp = tcg_temp_new_i32();
8629 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8630 store_reg(s, a->rt + 1, tmp);
8631
8632 /* LDRD w/ base writeback is undefined if the registers overlap. */
8633 op_addr_rr_post(s, a, addr, -4);
8634 return true;
8635}
8636
8637static bool trans_STRD_rr(DisasContext *s, arg_ldst_rr *a)
8638{
8639 int mem_idx = get_mem_index(s);
8640 TCGv_i32 addr, tmp;
8641
8642 if (!ENABLE_ARCH_5TE) {
8643 return false;
8644 }
8645 if (a->rt & 1) {
8646 unallocated_encoding(s);
8647 return true;
8648 }
8649 addr = op_addr_rr_pre(s, a);
8650
8651 tmp = load_reg(s, a->rt);
8652 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8653 tcg_temp_free_i32(tmp);
8654
8655 tcg_gen_addi_i32(addr, addr, 4);
8656
8657 tmp = load_reg(s, a->rt + 1);
8658 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8659 tcg_temp_free_i32(tmp);
8660
8661 op_addr_rr_post(s, a, addr, -4);
8662 return true;
8663}
8664
8665/*
8666 * Load/store immediate index
8667 */
8668
8669static TCGv_i32 op_addr_ri_pre(DisasContext *s, arg_ldst_ri *a)
8670{
8671 int ofs = a->imm;
8672
8673 if (!a->u) {
8674 ofs = -ofs;
8675 }
8676
8677 if (s->v8m_stackcheck && a->rn == 13 && a->w) {
8678 /*
8679 * Stackcheck. Here we know 'addr' is the current SP;
8680 * U is set if we're moving SP up, else down. It is
8681 * UNKNOWN whether the limit check triggers when SP starts
8682 * below the limit and ends up above it; we chose to do so.
8683 */
8684 if (!a->u) {
8685 TCGv_i32 newsp = tcg_temp_new_i32();
8686 tcg_gen_addi_i32(newsp, cpu_R[13], ofs);
8687 gen_helper_v8m_stackcheck(cpu_env, newsp);
8688 tcg_temp_free_i32(newsp);
8689 } else {
8690 gen_helper_v8m_stackcheck(cpu_env, cpu_R[13]);
8691 }
8692 }
8693
8694 return add_reg_for_lit(s, a->rn, a->p ? ofs : 0);
8695}
8696
8697static void op_addr_ri_post(DisasContext *s, arg_ldst_ri *a,
8698 TCGv_i32 addr, int address_offset)
8699{
8700 if (!a->p) {
8701 if (a->u) {
8702 address_offset += a->imm;
8703 } else {
8704 address_offset -= a->imm;
8705 }
8706 } else if (!a->w) {
8707 tcg_temp_free_i32(addr);
8708 return;
8709 }
8710 tcg_gen_addi_i32(addr, addr, address_offset);
8711 store_reg(s, a->rn, addr);
8712}
8713
8714static bool op_load_ri(DisasContext *s, arg_ldst_ri *a,
8715 MemOp mop, int mem_idx)
8716{
8717 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w);
8718 TCGv_i32 addr, tmp;
8719
8720 addr = op_addr_ri_pre(s, a);
8721
8722 tmp = tcg_temp_new_i32();
8723 gen_aa32_ld_i32(s, tmp, addr, mem_idx, mop | s->be_data);
8724 disas_set_da_iss(s, mop, issinfo);
8725
8726 /*
8727 * Perform base writeback before the loaded value to
8728 * ensure correct behavior with overlapping index registers.
8729 */
8730 op_addr_ri_post(s, a, addr, 0);
8731 store_reg_from_load(s, a->rt, tmp);
8732 return true;
8733}
8734
8735static bool op_store_ri(DisasContext *s, arg_ldst_ri *a,
8736 MemOp mop, int mem_idx)
8737{
8738 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w) | ISSIsWrite;
8739 TCGv_i32 addr, tmp;
8740
8741 addr = op_addr_ri_pre(s, a);
8742
8743 tmp = load_reg(s, a->rt);
8744 gen_aa32_st_i32(s, tmp, addr, mem_idx, mop | s->be_data);
8745 disas_set_da_iss(s, mop, issinfo);
8746 tcg_temp_free_i32(tmp);
8747
8748 op_addr_ri_post(s, a, addr, 0);
8749 return true;
8750}
8751
8752static bool op_ldrd_ri(DisasContext *s, arg_ldst_ri *a, int rt2)
8753{
8754 int mem_idx = get_mem_index(s);
8755 TCGv_i32 addr, tmp;
8756
8757 addr = op_addr_ri_pre(s, a);
8758
8759 tmp = tcg_temp_new_i32();
8760 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8761 store_reg(s, a->rt, tmp);
8762
8763 tcg_gen_addi_i32(addr, addr, 4);
8764
8765 tmp = tcg_temp_new_i32();
8766 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8767 store_reg(s, rt2, tmp);
8768
8769 /* LDRD w/ base writeback is undefined if the registers overlap. */
8770 op_addr_ri_post(s, a, addr, -4);
8771 return true;
8772}
8773
8774static bool trans_LDRD_ri_a32(DisasContext *s, arg_ldst_ri *a)
8775{
8776 if (!ENABLE_ARCH_5TE || (a->rt & 1)) {
8777 return false;
8778 }
8779 return op_ldrd_ri(s, a, a->rt + 1);
8780}
8781
8782static bool trans_LDRD_ri_t32(DisasContext *s, arg_ldst_ri2 *a)
8783{
8784 arg_ldst_ri b = {
8785 .u = a->u, .w = a->w, .p = a->p,
8786 .rn = a->rn, .rt = a->rt, .imm = a->imm
8787 };
8788 return op_ldrd_ri(s, &b, a->rt2);
8789}
8790
8791static bool op_strd_ri(DisasContext *s, arg_ldst_ri *a, int rt2)
8792{
8793 int mem_idx = get_mem_index(s);
8794 TCGv_i32 addr, tmp;
8795
8796 addr = op_addr_ri_pre(s, a);
8797
8798 tmp = load_reg(s, a->rt);
8799 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8800 tcg_temp_free_i32(tmp);
8801
8802 tcg_gen_addi_i32(addr, addr, 4);
8803
8804 tmp = load_reg(s, rt2);
8805 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8806 tcg_temp_free_i32(tmp);
8807
8808 op_addr_ri_post(s, a, addr, -4);
8809 return true;
8810}
8811
8812static bool trans_STRD_ri_a32(DisasContext *s, arg_ldst_ri *a)
8813{
8814 if (!ENABLE_ARCH_5TE || (a->rt & 1)) {
8815 return false;
8816 }
8817 return op_strd_ri(s, a, a->rt + 1);
8818}
8819
8820static bool trans_STRD_ri_t32(DisasContext *s, arg_ldst_ri2 *a)
8821{
8822 arg_ldst_ri b = {
8823 .u = a->u, .w = a->w, .p = a->p,
8824 .rn = a->rn, .rt = a->rt, .imm = a->imm
8825 };
8826 return op_strd_ri(s, &b, a->rt2);
8827}
8828
8829#define DO_LDST(NAME, WHICH, MEMOP) \
8830static bool trans_##NAME##_ri(DisasContext *s, arg_ldst_ri *a) \
8831{ \
8832 return op_##WHICH##_ri(s, a, MEMOP, get_mem_index(s)); \
8833} \
8834static bool trans_##NAME##T_ri(DisasContext *s, arg_ldst_ri *a) \
8835{ \
8836 return op_##WHICH##_ri(s, a, MEMOP, get_a32_user_mem_index(s)); \
8837} \
8838static bool trans_##NAME##_rr(DisasContext *s, arg_ldst_rr *a) \
8839{ \
8840 return op_##WHICH##_rr(s, a, MEMOP, get_mem_index(s)); \
8841} \
8842static bool trans_##NAME##T_rr(DisasContext *s, arg_ldst_rr *a) \
8843{ \
8844 return op_##WHICH##_rr(s, a, MEMOP, get_a32_user_mem_index(s)); \
8845}
8846
8847DO_LDST(LDR, load, MO_UL)
8848DO_LDST(LDRB, load, MO_UB)
8849DO_LDST(LDRH, load, MO_UW)
8850DO_LDST(LDRSB, load, MO_SB)
8851DO_LDST(LDRSH, load, MO_SW)
8852
8853DO_LDST(STR, store, MO_UL)
8854DO_LDST(STRB, store, MO_UB)
8855DO_LDST(STRH, store, MO_UW)
8856
8857#undef DO_LDST
8858
8859/*
8860 * Synchronization primitives
8861 */
8862
8863static bool op_swp(DisasContext *s, arg_SWP *a, MemOp opc)
8864{
8865 TCGv_i32 addr, tmp;
8866 TCGv taddr;
8867
8868 opc |= s->be_data;
8869 addr = load_reg(s, a->rn);
8870 taddr = gen_aa32_addr(s, addr, opc);
8871 tcg_temp_free_i32(addr);
8872
8873 tmp = load_reg(s, a->rt2);
8874 tcg_gen_atomic_xchg_i32(tmp, taddr, tmp, get_mem_index(s), opc);
8875 tcg_temp_free(taddr);
8876
8877 store_reg(s, a->rt, tmp);
8878 return true;
8879}
8880
8881static bool trans_SWP(DisasContext *s, arg_SWP *a)
8882{
8883 return op_swp(s, a, MO_UL | MO_ALIGN);
8884}
8885
8886static bool trans_SWPB(DisasContext *s, arg_SWP *a)
8887{
8888 return op_swp(s, a, MO_UB);
8889}
8890
8891/*
8892 * Load/Store Exclusive and Load-Acquire/Store-Release
8893 */
8894
8895static bool op_strex(DisasContext *s, arg_STREX *a, MemOp mop, bool rel)
8896{
8897 TCGv_i32 addr;
8898
8899 /* We UNDEF for these UNPREDICTABLE cases. */
8900 if (a->rd == 15 || a->rn == 15 || a->rt == 15
8901 || a->rd == a->rn || a->rd == a->rt
8902 || (s->thumb && (a->rd == 13 || a->rt == 13))
8903 || (mop == MO_64
8904 && (a->rt2 == 15
8905 || a->rd == a->rt2 || a->rt == a->rt2
8906 || (s->thumb && a->rt2 == 13)))) {
8907 unallocated_encoding(s);
8908 return true;
8909 }
8910
8911 if (rel) {
8912 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
8913 }
8914
8915 addr = tcg_temp_local_new_i32();
8916 load_reg_var(s, addr, a->rn);
8917 tcg_gen_addi_i32(addr, addr, a->imm);
8918
8919 gen_store_exclusive(s, a->rd, a->rt, a->rt2, addr, mop);
8920 tcg_temp_free_i32(addr);
8921 return true;
8922}
8923
8924static bool trans_STREX(DisasContext *s, arg_STREX *a)
8925{
8926 if (!ENABLE_ARCH_6) {
8927 return false;
8928 }
8929 return op_strex(s, a, MO_32, false);
8930}
8931
8932static bool trans_STREXD_a32(DisasContext *s, arg_STREX *a)
8933{
8934 if (!ENABLE_ARCH_6K) {
8935 return false;
8936 }
8937 /* We UNDEF for these UNPREDICTABLE cases. */
8938 if (a->rt & 1) {
8939 unallocated_encoding(s);
8940 return true;
8941 }
8942 a->rt2 = a->rt + 1;
8943 return op_strex(s, a, MO_64, false);
8944}
8945
8946static bool trans_STREXD_t32(DisasContext *s, arg_STREX *a)
8947{
8948 return op_strex(s, a, MO_64, false);
8949}
8950
8951static bool trans_STREXB(DisasContext *s, arg_STREX *a)
8952{
8953 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
8954 return false;
8955 }
8956 return op_strex(s, a, MO_8, false);
8957}
8958
8959static bool trans_STREXH(DisasContext *s, arg_STREX *a)
8960{
8961 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
8962 return false;
8963 }
8964 return op_strex(s, a, MO_16, false);
8965}
8966
8967static bool trans_STLEX(DisasContext *s, arg_STREX *a)
8968{
8969 if (!ENABLE_ARCH_8) {
8970 return false;
8971 }
8972 return op_strex(s, a, MO_32, true);
8973}
8974
8975static bool trans_STLEXD_a32(DisasContext *s, arg_STREX *a)
8976{
8977 if (!ENABLE_ARCH_8) {
8978 return false;
8979 }
8980 /* We UNDEF for these UNPREDICTABLE cases. */
8981 if (a->rt & 1) {
8982 unallocated_encoding(s);
8983 return true;
8984 }
8985 a->rt2 = a->rt + 1;
8986 return op_strex(s, a, MO_64, true);
8987}
8988
8989static bool trans_STLEXD_t32(DisasContext *s, arg_STREX *a)
8990{
8991 if (!ENABLE_ARCH_8) {
8992 return false;
8993 }
8994 return op_strex(s, a, MO_64, true);
8995}
8996
8997static bool trans_STLEXB(DisasContext *s, arg_STREX *a)
8998{
8999 if (!ENABLE_ARCH_8) {
9000 return false;
9001 }
9002 return op_strex(s, a, MO_8, true);
9003}
9004
9005static bool trans_STLEXH(DisasContext *s, arg_STREX *a)
9006{
9007 if (!ENABLE_ARCH_8) {
9008 return false;
9009 }
9010 return op_strex(s, a, MO_16, true);
9011}
9012
9013static bool op_stl(DisasContext *s, arg_STL *a, MemOp mop)
9014{
9015 TCGv_i32 addr, tmp;
9016
9017 if (!ENABLE_ARCH_8) {
9018 return false;
9019 }
9020 /* We UNDEF for these UNPREDICTABLE cases. */
9021 if (a->rn == 15 || a->rt == 15) {
9022 unallocated_encoding(s);
9023 return true;
9024 }
9025
9026 addr = load_reg(s, a->rn);
9027 tmp = load_reg(s, a->rt);
9028 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
9029 gen_aa32_st_i32(s, tmp, addr, get_mem_index(s), mop | s->be_data);
9030 disas_set_da_iss(s, mop, a->rt | ISSIsAcqRel | ISSIsWrite);
9031
9032 tcg_temp_free_i32(tmp);
9033 tcg_temp_free_i32(addr);
9034 return true;
9035}
9036
9037static bool trans_STL(DisasContext *s, arg_STL *a)
9038{
9039 return op_stl(s, a, MO_UL);
9040}
9041
9042static bool trans_STLB(DisasContext *s, arg_STL *a)
9043{
9044 return op_stl(s, a, MO_UB);
9045}
9046
9047static bool trans_STLH(DisasContext *s, arg_STL *a)
9048{
9049 return op_stl(s, a, MO_UW);
9050}
9051
9052static bool op_ldrex(DisasContext *s, arg_LDREX *a, MemOp mop, bool acq)
9053{
9054 TCGv_i32 addr;
9055
9056 /* We UNDEF for these UNPREDICTABLE cases. */
9057 if (a->rn == 15 || a->rt == 15
9058 || (s->thumb && a->rt == 13)
9059 || (mop == MO_64
9060 && (a->rt2 == 15 || a->rt == a->rt2
9061 || (s->thumb && a->rt2 == 13)))) {
9062 unallocated_encoding(s);
9063 return true;
9064 }
9065
9066 addr = tcg_temp_local_new_i32();
9067 load_reg_var(s, addr, a->rn);
9068 tcg_gen_addi_i32(addr, addr, a->imm);
9069
9070 gen_load_exclusive(s, a->rt, a->rt2, addr, mop);
9071 tcg_temp_free_i32(addr);
9072
9073 if (acq) {
9074 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_LDAQ);
9075 }
9076 return true;
9077}
9078
9079static bool trans_LDREX(DisasContext *s, arg_LDREX *a)
9080{
9081 if (!ENABLE_ARCH_6) {
9082 return false;
9083 }
9084 return op_ldrex(s, a, MO_32, false);
9085}
9086
9087static bool trans_LDREXD_a32(DisasContext *s, arg_LDREX *a)
9088{
9089 if (!ENABLE_ARCH_6K) {
9090 return false;
9091 }
9092 /* We UNDEF for these UNPREDICTABLE cases. */
9093 if (a->rt & 1) {
9094 unallocated_encoding(s);
9095 return true;
9096 }
9097 a->rt2 = a->rt + 1;
9098 return op_ldrex(s, a, MO_64, false);
9099}
9100
9101static bool trans_LDREXD_t32(DisasContext *s, arg_LDREX *a)
9102{
9103 return op_ldrex(s, a, MO_64, false);
9104}
9105
9106static bool trans_LDREXB(DisasContext *s, arg_LDREX *a)
9107{
9108 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
9109 return false;
9110 }
9111 return op_ldrex(s, a, MO_8, false);
9112}
9113
9114static bool trans_LDREXH(DisasContext *s, arg_LDREX *a)
9115{
9116 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
9117 return false;
9118 }
9119 return op_ldrex(s, a, MO_16, false);
9120}
9121
9122static bool trans_LDAEX(DisasContext *s, arg_LDREX *a)
9123{
9124 if (!ENABLE_ARCH_8) {
9125 return false;
9126 }
9127 return op_ldrex(s, a, MO_32, true);
9128}
9129
9130static bool trans_LDAEXD_a32(DisasContext *s, arg_LDREX *a)
9131{
9132 if (!ENABLE_ARCH_8) {
9133 return false;
9134 }
9135 /* We UNDEF for these UNPREDICTABLE cases. */
9136 if (a->rt & 1) {
9137 unallocated_encoding(s);
9138 return true;
9139 }
9140 a->rt2 = a->rt + 1;
9141 return op_ldrex(s, a, MO_64, true);
9142}
9143
9144static bool trans_LDAEXD_t32(DisasContext *s, arg_LDREX *a)
9145{
9146 if (!ENABLE_ARCH_8) {
9147 return false;
9148 }
9149 return op_ldrex(s, a, MO_64, true);
9150}
9151
9152static bool trans_LDAEXB(DisasContext *s, arg_LDREX *a)
9153{
9154 if (!ENABLE_ARCH_8) {
9155 return false;
9156 }
9157 return op_ldrex(s, a, MO_8, true);
9158}
9159
9160static bool trans_LDAEXH(DisasContext *s, arg_LDREX *a)
9161{
9162 if (!ENABLE_ARCH_8) {
9163 return false;
9164 }
9165 return op_ldrex(s, a, MO_16, true);
9166}
9167
9168static bool op_lda(DisasContext *s, arg_LDA *a, MemOp mop)
9169{
9170 TCGv_i32 addr, tmp;
9171
9172 if (!ENABLE_ARCH_8) {
9173 return false;
9174 }
9175 /* We UNDEF for these UNPREDICTABLE cases. */
9176 if (a->rn == 15 || a->rt == 15) {
9177 unallocated_encoding(s);
9178 return true;
9179 }
9180
9181 addr = load_reg(s, a->rn);
9182 tmp = tcg_temp_new_i32();
9183 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), mop | s->be_data);
9184 disas_set_da_iss(s, mop, a->rt | ISSIsAcqRel);
9185 tcg_temp_free_i32(addr);
9186
9187 store_reg(s, a->rt, tmp);
9188 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
9189 return true;
9190}
9191
9192static bool trans_LDA(DisasContext *s, arg_LDA *a)
9193{
9194 return op_lda(s, a, MO_UL);
9195}
9196
9197static bool trans_LDAB(DisasContext *s, arg_LDA *a)
9198{
9199 return op_lda(s, a, MO_UB);
9200}
9201
9202static bool trans_LDAH(DisasContext *s, arg_LDA *a)
9203{
9204 return op_lda(s, a, MO_UW);
9205}
9206
9207/*
9208 * Media instructions
9209 */
9210
9211static bool trans_USADA8(DisasContext *s, arg_USADA8 *a)
9212{
9213 TCGv_i32 t1, t2;
9214
9215 if (!ENABLE_ARCH_6) {
9216 return false;
9217 }
9218
9219 t1 = load_reg(s, a->rn);
9220 t2 = load_reg(s, a->rm);
9221 gen_helper_usad8(t1, t1, t2);
9222 tcg_temp_free_i32(t2);
9223 if (a->ra != 15) {
9224 t2 = load_reg(s, a->ra);
9225 tcg_gen_add_i32(t1, t1, t2);
9226 tcg_temp_free_i32(t2);
9227 }
9228 store_reg(s, a->rd, t1);
9229 return true;
9230}
9231
9232static bool op_bfx(DisasContext *s, arg_UBFX *a, bool u)
9233{
9234 TCGv_i32 tmp;
9235 int width = a->widthm1 + 1;
9236 int shift = a->lsb;
9237
9238 if (!ENABLE_ARCH_6T2) {
9239 return false;
9240 }
9241 if (shift + width > 32) {
9242 /* UNPREDICTABLE; we choose to UNDEF */
9243 unallocated_encoding(s);
9244 return true;
9245 }
9246
9247 tmp = load_reg(s, a->rn);
9248 if (u) {
9249 tcg_gen_extract_i32(tmp, tmp, shift, width);
9250 } else {
9251 tcg_gen_sextract_i32(tmp, tmp, shift, width);
9252 }
9253 store_reg(s, a->rd, tmp);
9254 return true;
9255}
9256
9257static bool trans_SBFX(DisasContext *s, arg_SBFX *a)
9258{
9259 return op_bfx(s, a, false);
9260}
9261
9262static bool trans_UBFX(DisasContext *s, arg_UBFX *a)
9263{
9264 return op_bfx(s, a, true);
9265}
9266
9267static bool trans_BFCI(DisasContext *s, arg_BFCI *a)
9268{
9269 TCGv_i32 tmp;
9270 int msb = a->msb, lsb = a->lsb;
9271 int width;
9272
9273 if (!ENABLE_ARCH_6T2) {
9274 return false;
9275 }
9276 if (msb < lsb) {
9277 /* UNPREDICTABLE; we choose to UNDEF */
9278 unallocated_encoding(s);
9279 return true;
9280 }
9281
9282 width = msb + 1 - lsb;
9283 if (a->rn == 15) {
9284 /* BFC */
9285 tmp = tcg_const_i32(0);
9286 } else {
9287 /* BFI */
9288 tmp = load_reg(s, a->rn);
9289 }
9290 if (width != 32) {
9291 TCGv_i32 tmp2 = load_reg(s, a->rd);
9292 tcg_gen_deposit_i32(tmp, tmp2, tmp, lsb, width);
9293 tcg_temp_free_i32(tmp2);
9294 }
9295 store_reg(s, a->rd, tmp);
9296 return true;
9297}
9298
9299static bool trans_UDF(DisasContext *s, arg_UDF *a)
9300{
9301 unallocated_encoding(s);
9302 return true;
9303}
9304
9305/*
9306 * Parallel addition and subtraction
9307 */
9308
9309static bool op_par_addsub(DisasContext *s, arg_rrr *a,
9310 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32))
9311{
9312 TCGv_i32 t0, t1;
9313
9314 if (s->thumb
9315 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
9316 : !ENABLE_ARCH_6) {
9317 return false;
9318 }
9319
9320 t0 = load_reg(s, a->rn);
9321 t1 = load_reg(s, a->rm);
9322
9323 gen(t0, t0, t1);
9324
9325 tcg_temp_free_i32(t1);
9326 store_reg(s, a->rd, t0);
9327 return true;
9328}
9329
9330static bool op_par_addsub_ge(DisasContext *s, arg_rrr *a,
9331 void (*gen)(TCGv_i32, TCGv_i32,
9332 TCGv_i32, TCGv_ptr))
9333{
9334 TCGv_i32 t0, t1;
9335 TCGv_ptr ge;
9336
9337 if (s->thumb
9338 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
9339 : !ENABLE_ARCH_6) {
9340 return false;
9341 }
9342
9343 t0 = load_reg(s, a->rn);
9344 t1 = load_reg(s, a->rm);
9345
9346 ge = tcg_temp_new_ptr();
9347 tcg_gen_addi_ptr(ge, cpu_env, offsetof(CPUARMState, GE));
9348 gen(t0, t0, t1, ge);
9349
9350 tcg_temp_free_ptr(ge);
9351 tcg_temp_free_i32(t1);
9352 store_reg(s, a->rd, t0);
9353 return true;
9354}
9355
9356#define DO_PAR_ADDSUB(NAME, helper) \
9357static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
9358{ \
9359 return op_par_addsub(s, a, helper); \
9360}
9361
9362#define DO_PAR_ADDSUB_GE(NAME, helper) \
9363static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
9364{ \
9365 return op_par_addsub_ge(s, a, helper); \
9366}
9367
9368DO_PAR_ADDSUB_GE(SADD16, gen_helper_sadd16)
9369DO_PAR_ADDSUB_GE(SASX, gen_helper_saddsubx)
9370DO_PAR_ADDSUB_GE(SSAX, gen_helper_ssubaddx)
9371DO_PAR_ADDSUB_GE(SSUB16, gen_helper_ssub16)
9372DO_PAR_ADDSUB_GE(SADD8, gen_helper_sadd8)
9373DO_PAR_ADDSUB_GE(SSUB8, gen_helper_ssub8)
9374
9375DO_PAR_ADDSUB_GE(UADD16, gen_helper_uadd16)
9376DO_PAR_ADDSUB_GE(UASX, gen_helper_uaddsubx)
9377DO_PAR_ADDSUB_GE(USAX, gen_helper_usubaddx)
9378DO_PAR_ADDSUB_GE(USUB16, gen_helper_usub16)
9379DO_PAR_ADDSUB_GE(UADD8, gen_helper_uadd8)
9380DO_PAR_ADDSUB_GE(USUB8, gen_helper_usub8)
9381
9382DO_PAR_ADDSUB(QADD16, gen_helper_qadd16)
9383DO_PAR_ADDSUB(QASX, gen_helper_qaddsubx)
9384DO_PAR_ADDSUB(QSAX, gen_helper_qsubaddx)
9385DO_PAR_ADDSUB(QSUB16, gen_helper_qsub16)
9386DO_PAR_ADDSUB(QADD8, gen_helper_qadd8)
9387DO_PAR_ADDSUB(QSUB8, gen_helper_qsub8)
9388
9389DO_PAR_ADDSUB(UQADD16, gen_helper_uqadd16)
9390DO_PAR_ADDSUB(UQASX, gen_helper_uqaddsubx)
9391DO_PAR_ADDSUB(UQSAX, gen_helper_uqsubaddx)
9392DO_PAR_ADDSUB(UQSUB16, gen_helper_uqsub16)
9393DO_PAR_ADDSUB(UQADD8, gen_helper_uqadd8)
9394DO_PAR_ADDSUB(UQSUB8, gen_helper_uqsub8)
9395
9396DO_PAR_ADDSUB(SHADD16, gen_helper_shadd16)
9397DO_PAR_ADDSUB(SHASX, gen_helper_shaddsubx)
9398DO_PAR_ADDSUB(SHSAX, gen_helper_shsubaddx)
9399DO_PAR_ADDSUB(SHSUB16, gen_helper_shsub16)
9400DO_PAR_ADDSUB(SHADD8, gen_helper_shadd8)
9401DO_PAR_ADDSUB(SHSUB8, gen_helper_shsub8)
9402
9403DO_PAR_ADDSUB(UHADD16, gen_helper_uhadd16)
9404DO_PAR_ADDSUB(UHASX, gen_helper_uhaddsubx)
9405DO_PAR_ADDSUB(UHSAX, gen_helper_uhsubaddx)
9406DO_PAR_ADDSUB(UHSUB16, gen_helper_uhsub16)
9407DO_PAR_ADDSUB(UHADD8, gen_helper_uhadd8)
9408DO_PAR_ADDSUB(UHSUB8, gen_helper_uhsub8)
9409
9410#undef DO_PAR_ADDSUB
9411#undef DO_PAR_ADDSUB_GE
9412
9413/*
9414 * Packing, unpacking, saturation, and reversal
9415 */
9416
9417static bool trans_PKH(DisasContext *s, arg_PKH *a)
9418{
9419 TCGv_i32 tn, tm;
9420 int shift = a->imm;
9421
9422 if (s->thumb
9423 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
9424 : !ENABLE_ARCH_6) {
9425 return false;
9426 }
9427
9428 tn = load_reg(s, a->rn);
9429 tm = load_reg(s, a->rm);
9430 if (a->tb) {
9431 /* PKHTB */
9432 if (shift == 0) {
9433 shift = 31;
9434 }
9435 tcg_gen_sari_i32(tm, tm, shift);
9436 tcg_gen_deposit_i32(tn, tn, tm, 0, 16);
9437 } else {
9438 /* PKHBT */
9439 tcg_gen_shli_i32(tm, tm, shift);
9440 tcg_gen_deposit_i32(tn, tm, tn, 0, 16);
9441 }
9442 tcg_temp_free_i32(tm);
9443 store_reg(s, a->rd, tn);
9444 return true;
9445}
9446
9447static bool op_sat(DisasContext *s, arg_sat *a,
9448 void (*gen)(TCGv_i32, TCGv_env, TCGv_i32, TCGv_i32))
9449{
9450 TCGv_i32 tmp, satimm;
9451 int shift = a->imm;
9452
9453 if (!ENABLE_ARCH_6) {
9454 return false;
9455 }
9456
9457 tmp = load_reg(s, a->rn);
9458 if (a->sh) {
9459 tcg_gen_sari_i32(tmp, tmp, shift ? shift : 31);
9460 } else {
9461 tcg_gen_shli_i32(tmp, tmp, shift);
9462 }
9463
9464 satimm = tcg_const_i32(a->satimm);
9465 gen(tmp, cpu_env, tmp, satimm);
9466 tcg_temp_free_i32(satimm);
9467
9468 store_reg(s, a->rd, tmp);
9469 return true;
9470}
9471
9472static bool trans_SSAT(DisasContext *s, arg_sat *a)
9473{
9474 return op_sat(s, a, gen_helper_ssat);
9475}
9476
9477static bool trans_USAT(DisasContext *s, arg_sat *a)
9478{
9479 return op_sat(s, a, gen_helper_usat);
9480}
9481
9482static bool trans_SSAT16(DisasContext *s, arg_sat *a)
9483{
9484 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9485 return false;
9486 }
9487 return op_sat(s, a, gen_helper_ssat16);
9488}
9489
9490static bool trans_USAT16(DisasContext *s, arg_sat *a)
9491{
9492 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9493 return false;
9494 }
9495 return op_sat(s, a, gen_helper_usat16);
9496}
9497
9498static bool op_xta(DisasContext *s, arg_rrr_rot *a,
9499 void (*gen_extract)(TCGv_i32, TCGv_i32),
9500 void (*gen_add)(TCGv_i32, TCGv_i32, TCGv_i32))
9501{
9502 TCGv_i32 tmp;
9503
9504 if (!ENABLE_ARCH_6) {
9505 return false;
9506 }
9507
9508 tmp = load_reg(s, a->rm);
9509 /*
9510 * TODO: In many cases we could do a shift instead of a rotate.
9511 * Combined with a simple extend, that becomes an extract.
9512 */
9513 tcg_gen_rotri_i32(tmp, tmp, a->rot * 8);
9514 gen_extract(tmp, tmp);
9515
9516 if (a->rn != 15) {
9517 TCGv_i32 tmp2 = load_reg(s, a->rn);
9518 gen_add(tmp, tmp, tmp2);
9519 tcg_temp_free_i32(tmp2);
9520 }
9521 store_reg(s, a->rd, tmp);
9522 return true;
9523}
9524
9525static bool trans_SXTAB(DisasContext *s, arg_rrr_rot *a)
9526{
9527 return op_xta(s, a, tcg_gen_ext8s_i32, tcg_gen_add_i32);
9528}
9529
9530static bool trans_SXTAH(DisasContext *s, arg_rrr_rot *a)
9531{
9532 return op_xta(s, a, tcg_gen_ext16s_i32, tcg_gen_add_i32);
9533}
9534
9535static bool trans_SXTAB16(DisasContext *s, arg_rrr_rot *a)
9536{
9537 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9538 return false;
9539 }
9540 return op_xta(s, a, gen_helper_sxtb16, gen_add16);
9541}
9542
9543static bool trans_UXTAB(DisasContext *s, arg_rrr_rot *a)
9544{
9545 return op_xta(s, a, tcg_gen_ext8u_i32, tcg_gen_add_i32);
9546}
9547
9548static bool trans_UXTAH(DisasContext *s, arg_rrr_rot *a)
9549{
9550 return op_xta(s, a, tcg_gen_ext16u_i32, tcg_gen_add_i32);
9551}
9552
9553static bool trans_UXTAB16(DisasContext *s, arg_rrr_rot *a)
9554{
9555 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9556 return false;
9557 }
9558 return op_xta(s, a, gen_helper_uxtb16, gen_add16);
9559}
9560
9561static bool trans_SEL(DisasContext *s, arg_rrr *a)
9562{
9563 TCGv_i32 t1, t2, t3;
9564
9565 if (s->thumb
9566 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
9567 : !ENABLE_ARCH_6) {
9568 return false;
9569 }
9570
9571 t1 = load_reg(s, a->rn);
9572 t2 = load_reg(s, a->rm);
9573 t3 = tcg_temp_new_i32();
9574 tcg_gen_ld_i32(t3, cpu_env, offsetof(CPUARMState, GE));
9575 gen_helper_sel_flags(t1, t3, t1, t2);
9576 tcg_temp_free_i32(t3);
9577 tcg_temp_free_i32(t2);
9578 store_reg(s, a->rd, t1);
9579 return true;
9580}
9581
9582static bool op_rr(DisasContext *s, arg_rr *a,
9583 void (*gen)(TCGv_i32, TCGv_i32))
9584{
9585 TCGv_i32 tmp;
9586
9587 tmp = load_reg(s, a->rm);
9588 gen(tmp, tmp);
9589 store_reg(s, a->rd, tmp);
9590 return true;
9591}
9592
9593static bool trans_REV(DisasContext *s, arg_rr *a)
9594{
9595 if (!ENABLE_ARCH_6) {
9596 return false;
9597 }
9598 return op_rr(s, a, tcg_gen_bswap32_i32);
9599}
9600
9601static bool trans_REV16(DisasContext *s, arg_rr *a)
9602{
9603 if (!ENABLE_ARCH_6) {
9604 return false;
9605 }
9606 return op_rr(s, a, gen_rev16);
9607}
9608
9609static bool trans_REVSH(DisasContext *s, arg_rr *a)
9610{
9611 if (!ENABLE_ARCH_6) {
9612 return false;
9613 }
9614 return op_rr(s, a, gen_revsh);
9615}
9616
9617static bool trans_RBIT(DisasContext *s, arg_rr *a)
9618{
9619 if (!ENABLE_ARCH_6T2) {
9620 return false;
9621 }
9622 return op_rr(s, a, gen_helper_rbit);
9623}
9624
9625/*
9626 * Signed multiply, signed and unsigned divide
9627 */
9628
9629static bool op_smlad(DisasContext *s, arg_rrrr *a, bool m_swap, bool sub)
9630{
9631 TCGv_i32 t1, t2;
9632
9633 if (!ENABLE_ARCH_6) {
9634 return false;
9635 }
9636
9637 t1 = load_reg(s, a->rn);
9638 t2 = load_reg(s, a->rm);
9639 if (m_swap) {
9640 gen_swap_half(t2);
9641 }
9642 gen_smul_dual(t1, t2);
9643
9644 if (sub) {
9645 /* This subtraction cannot overflow. */
9646 tcg_gen_sub_i32(t1, t1, t2);
9647 } else {
9648 /*
9649 * This addition cannot overflow 32 bits; however it may
9650 * overflow considered as a signed operation, in which case
9651 * we must set the Q flag.
9652 */
9653 gen_helper_add_setq(t1, cpu_env, t1, t2);
9654 }
9655 tcg_temp_free_i32(t2);
9656
9657 if (a->ra != 15) {
9658 t2 = load_reg(s, a->ra);
9659 gen_helper_add_setq(t1, cpu_env, t1, t2);
9660 tcg_temp_free_i32(t2);
9661 }
9662 store_reg(s, a->rd, t1);
9663 return true;
9664}
9665
9666static bool trans_SMLAD(DisasContext *s, arg_rrrr *a)
9667{
9668 return op_smlad(s, a, false, false);
9669}
9670
9671static bool trans_SMLADX(DisasContext *s, arg_rrrr *a)
9672{
9673 return op_smlad(s, a, true, false);
9674}
9675
9676static bool trans_SMLSD(DisasContext *s, arg_rrrr *a)
9677{
9678 return op_smlad(s, a, false, true);
9679}
9680
9681static bool trans_SMLSDX(DisasContext *s, arg_rrrr *a)
9682{
9683 return op_smlad(s, a, true, true);
9684}
9685
9686static bool op_smlald(DisasContext *s, arg_rrrr *a, bool m_swap, bool sub)
9687{
9688 TCGv_i32 t1, t2;
9689 TCGv_i64 l1, l2;
9690
9691 if (!ENABLE_ARCH_6) {
9692 return false;
9693 }
9694
9695 t1 = load_reg(s, a->rn);
9696 t2 = load_reg(s, a->rm);
9697 if (m_swap) {
9698 gen_swap_half(t2);
9699 }
9700 gen_smul_dual(t1, t2);
9701
9702 l1 = tcg_temp_new_i64();
9703 l2 = tcg_temp_new_i64();
9704 tcg_gen_ext_i32_i64(l1, t1);
9705 tcg_gen_ext_i32_i64(l2, t2);
9706 tcg_temp_free_i32(t1);
9707 tcg_temp_free_i32(t2);
9708
9709 if (sub) {
9710 tcg_gen_sub_i64(l1, l1, l2);
9711 } else {
9712 tcg_gen_add_i64(l1, l1, l2);
9713 }
9714 tcg_temp_free_i64(l2);
9715
9716 gen_addq(s, l1, a->ra, a->rd);
9717 gen_storeq_reg(s, a->ra, a->rd, l1);
9718 tcg_temp_free_i64(l1);
9719 return true;
9720}
9721
9722static bool trans_SMLALD(DisasContext *s, arg_rrrr *a)
9723{
9724 return op_smlald(s, a, false, false);
9725}
9726
9727static bool trans_SMLALDX(DisasContext *s, arg_rrrr *a)
9728{
9729 return op_smlald(s, a, true, false);
9730}
9731
9732static bool trans_SMLSLD(DisasContext *s, arg_rrrr *a)
9733{
9734 return op_smlald(s, a, false, true);
9735}
9736
9737static bool trans_SMLSLDX(DisasContext *s, arg_rrrr *a)
9738{
9739 return op_smlald(s, a, true, true);
9740}
9741
9742static bool op_smmla(DisasContext *s, arg_rrrr *a, bool round, bool sub)
9743{
9744 TCGv_i32 t1, t2;
9745
9746 if (s->thumb
9747 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
9748 : !ENABLE_ARCH_6) {
9749 return false;
9750 }
9751
9752 t1 = load_reg(s, a->rn);
9753 t2 = load_reg(s, a->rm);
9754 tcg_gen_muls2_i32(t2, t1, t1, t2);
9755
9756 if (a->ra != 15) {
9757 TCGv_i32 t3 = load_reg(s, a->ra);
9758 if (sub) {
9759 /*
9760 * For SMMLS, we need a 64-bit subtract. Borrow caused by
9761 * a non-zero multiplicand lowpart, and the correct result
9762 * lowpart for rounding.
9763 */
9764 TCGv_i32 zero = tcg_const_i32(0);
9765 tcg_gen_sub2_i32(t2, t1, zero, t3, t2, t1);
9766 tcg_temp_free_i32(zero);
9767 } else {
9768 tcg_gen_add_i32(t1, t1, t3);
9769 }
9770 tcg_temp_free_i32(t3);
9771 }
9772 if (round) {
9773 /*
9774 * Adding 0x80000000 to the 64-bit quantity means that we have
9775 * carry in to the high word when the low word has the msb set.
9776 */
9777 tcg_gen_shri_i32(t2, t2, 31);
9778 tcg_gen_add_i32(t1, t1, t2);
9779 }
9780 tcg_temp_free_i32(t2);
9781 store_reg(s, a->rd, t1);
9782 return true;
9783}
9784
9785static bool trans_SMMLA(DisasContext *s, arg_rrrr *a)
9786{
9787 return op_smmla(s, a, false, false);
9788}
9789
9790static bool trans_SMMLAR(DisasContext *s, arg_rrrr *a)
9791{
9792 return op_smmla(s, a, true, false);
9793}
9794
9795static bool trans_SMMLS(DisasContext *s, arg_rrrr *a)
9796{
9797 return op_smmla(s, a, false, true);
9798}
9799
9800static bool trans_SMMLSR(DisasContext *s, arg_rrrr *a)
9801{
9802 return op_smmla(s, a, true, true);
9803}
9804
9805static bool op_div(DisasContext *s, arg_rrr *a, bool u)
9806{
9807 TCGv_i32 t1, t2;
9808
9809 if (s->thumb
9810 ? !dc_isar_feature(thumb_div, s)
9811 : !dc_isar_feature(arm_div, s)) {
9812 return false;
9813 }
9814
9815 t1 = load_reg(s, a->rn);
9816 t2 = load_reg(s, a->rm);
9817 if (u) {
9818 gen_helper_udiv(t1, t1, t2);
9819 } else {
9820 gen_helper_sdiv(t1, t1, t2);
9821 }
9822 tcg_temp_free_i32(t2);
9823 store_reg(s, a->rd, t1);
9824 return true;
9825}
9826
9827static bool trans_SDIV(DisasContext *s, arg_rrr *a)
9828{
9829 return op_div(s, a, false);
9830}
9831
9832static bool trans_UDIV(DisasContext *s, arg_rrr *a)
9833{
9834 return op_div(s, a, true);
9835}
9836
9837/*
9838 * Block data transfer
9839 */
9840
9841static TCGv_i32 op_addr_block_pre(DisasContext *s, arg_ldst_block *a, int n)
9842{
9843 TCGv_i32 addr = load_reg(s, a->rn);
9844
9845 if (a->b) {
9846 if (a->i) {
9847 /* pre increment */
9848 tcg_gen_addi_i32(addr, addr, 4);
9849 } else {
9850 /* pre decrement */
9851 tcg_gen_addi_i32(addr, addr, -(n * 4));
9852 }
9853 } else if (!a->i && n != 1) {
9854 /* post decrement */
9855 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
9856 }
9857
9858 if (s->v8m_stackcheck && a->rn == 13 && a->w) {
9859 /*
9860 * If the writeback is incrementing SP rather than
9861 * decrementing it, and the initial SP is below the
9862 * stack limit but the final written-back SP would
9863 * be above, then then we must not perform any memory
9864 * accesses, but it is IMPDEF whether we generate
9865 * an exception. We choose to do so in this case.
9866 * At this point 'addr' is the lowest address, so
9867 * either the original SP (if incrementing) or our
9868 * final SP (if decrementing), so that's what we check.
9869 */
9870 gen_helper_v8m_stackcheck(cpu_env, addr);
9871 }
9872
9873 return addr;
9874}
9875
9876static void op_addr_block_post(DisasContext *s, arg_ldst_block *a,
9877 TCGv_i32 addr, int n)
9878{
9879 if (a->w) {
9880 /* write back */
9881 if (!a->b) {
9882 if (a->i) {
9883 /* post increment */
9884 tcg_gen_addi_i32(addr, addr, 4);
9885 } else {
9886 /* post decrement */
9887 tcg_gen_addi_i32(addr, addr, -(n * 4));
9888 }
9889 } else if (!a->i && n != 1) {
9890 /* pre decrement */
9891 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
9892 }
9893 store_reg(s, a->rn, addr);
9894 } else {
9895 tcg_temp_free_i32(addr);
9896 }
9897}
9898
9899static bool op_stm(DisasContext *s, arg_ldst_block *a, int min_n)
9900{
9901 int i, j, n, list, mem_idx;
9902 bool user = a->u;
9903 TCGv_i32 addr, tmp, tmp2;
9904
9905 if (user) {
9906 /* STM (user) */
9907 if (IS_USER(s)) {
9908 /* Only usable in supervisor mode. */
9909 unallocated_encoding(s);
9910 return true;
9911 }
9912 }
9913
9914 list = a->list;
9915 n = ctpop16(list);
9916 if (n < min_n || a->rn == 15) {
9917 unallocated_encoding(s);
9918 return true;
9919 }
9920
9921 addr = op_addr_block_pre(s, a, n);
9922 mem_idx = get_mem_index(s);
9923
9924 for (i = j = 0; i < 16; i++) {
9925 if (!(list & (1 << i))) {
9926 continue;
9927 }
9928
9929 if (user && i != 15) {
9930 tmp = tcg_temp_new_i32();
9931 tmp2 = tcg_const_i32(i);
9932 gen_helper_get_user_reg(tmp, cpu_env, tmp2);
9933 tcg_temp_free_i32(tmp2);
9934 } else {
9935 tmp = load_reg(s, i);
9936 }
9937 gen_aa32_st32(s, tmp, addr, mem_idx);
9938 tcg_temp_free_i32(tmp);
9939
9940 /* No need to add after the last transfer. */
9941 if (++j != n) {
9942 tcg_gen_addi_i32(addr, addr, 4);
9943 }
9944 }
9945
9946 op_addr_block_post(s, a, addr, n);
9947 return true;
9948}
9949
9950static bool trans_STM(DisasContext *s, arg_ldst_block *a)
9951{
9952 /* BitCount(list) < 1 is UNPREDICTABLE */
9953 return op_stm(s, a, 1);
9954}
9955
9956static bool trans_STM_t32(DisasContext *s, arg_ldst_block *a)
9957{
9958 /* Writeback register in register list is UNPREDICTABLE for T32. */
9959 if (a->w && (a->list & (1 << a->rn))) {
9960 unallocated_encoding(s);
9961 return true;
9962 }
9963 /* BitCount(list) < 2 is UNPREDICTABLE */
9964 return op_stm(s, a, 2);
9965}
9966
9967static bool do_ldm(DisasContext *s, arg_ldst_block *a, int min_n)
9968{
9969 int i, j, n, list, mem_idx;
9970 bool loaded_base;
9971 bool user = a->u;
9972 bool exc_return = false;
9973 TCGv_i32 addr, tmp, tmp2, loaded_var;
9974
9975 if (user) {
9976 /* LDM (user), LDM (exception return) */
9977 if (IS_USER(s)) {
9978 /* Only usable in supervisor mode. */
9979 unallocated_encoding(s);
9980 return true;
9981 }
9982 if (extract32(a->list, 15, 1)) {
9983 exc_return = true;
9984 user = false;
9985 } else {
9986 /* LDM (user) does not allow writeback. */
9987 if (a->w) {
9988 unallocated_encoding(s);
9989 return true;
9990 }
9991 }
9992 }
9993
9994 list = a->list;
9995 n = ctpop16(list);
9996 if (n < min_n || a->rn == 15) {
9997 unallocated_encoding(s);
9998 return true;
9999 }
10000
10001 addr = op_addr_block_pre(s, a, n);
10002 mem_idx = get_mem_index(s);
10003 loaded_base = false;
10004 loaded_var = NULL;
10005
10006 for (i = j = 0; i < 16; i++) {
10007 if (!(list & (1 << i))) {
10008 continue;
10009 }
10010
10011 tmp = tcg_temp_new_i32();
10012 gen_aa32_ld32u(s, tmp, addr, mem_idx);
10013 if (user) {
10014 tmp2 = tcg_const_i32(i);
10015 gen_helper_set_user_reg(cpu_env, tmp2, tmp);
10016 tcg_temp_free_i32(tmp2);
10017 tcg_temp_free_i32(tmp);
10018 } else if (i == a->rn) {
10019 loaded_var = tmp;
10020 loaded_base = true;
10021 } else if (i == 15 && exc_return) {
10022 store_pc_exc_ret(s, tmp);
10023 } else {
10024 store_reg_from_load(s, i, tmp);
10025 }
10026
10027 /* No need to add after the last transfer. */
10028 if (++j != n) {
10029 tcg_gen_addi_i32(addr, addr, 4);
10030 }
10031 }
10032
10033 op_addr_block_post(s, a, addr, n);
10034
10035 if (loaded_base) {
10036 /* Note that we reject base == pc above. */
10037 store_reg(s, a->rn, loaded_var);
10038 }
10039
10040 if (exc_return) {
10041 /* Restore CPSR from SPSR. */
10042 tmp = load_cpu_field(spsr);
10043 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
10044 gen_io_start();
10045 }
10046 gen_helper_cpsr_write_eret(cpu_env, tmp);
10047 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
10048 gen_io_end();
10049 }
10050 tcg_temp_free_i32(tmp);
10051 /* Must exit loop to check un-masked IRQs */
10052 s->base.is_jmp = DISAS_EXIT;
10053 }
10054 return true;
10055}
10056
10057static bool trans_LDM_a32(DisasContext *s, arg_ldst_block *a)
10058{
10059 /*
10060 * Writeback register in register list is UNPREDICTABLE
10061 * for ArchVersion() >= 7. Prior to v7, A32 would write
10062 * an UNKNOWN value to the base register.
10063 */
10064 if (ENABLE_ARCH_7 && a->w && (a->list & (1 << a->rn))) {
10065 unallocated_encoding(s);
10066 return true;
10067 }
10068 /* BitCount(list) < 1 is UNPREDICTABLE */
10069 return do_ldm(s, a, 1);
10070}
10071
10072static bool trans_LDM_t32(DisasContext *s, arg_ldst_block *a)
10073{
10074 /* Writeback register in register list is UNPREDICTABLE for T32. */
10075 if (a->w && (a->list & (1 << a->rn))) {
10076 unallocated_encoding(s);
10077 return true;
10078 }
10079 /* BitCount(list) < 2 is UNPREDICTABLE */
10080 return do_ldm(s, a, 2);
10081}
10082
10083static bool trans_LDM_t16(DisasContext *s, arg_ldst_block *a)
10084{
10085 /* Writeback is conditional on the base register not being loaded. */
10086 a->w = !(a->list & (1 << a->rn));
10087 /* BitCount(list) < 1 is UNPREDICTABLE */
10088 return do_ldm(s, a, 1);
10089}
10090
10091/*
10092 * Branch, branch with link
10093 */
10094
10095static bool trans_B(DisasContext *s, arg_i *a)
10096{
10097 gen_jmp(s, read_pc(s) + a->imm);
10098 return true;
10099}
10100
10101static bool trans_B_cond_thumb(DisasContext *s, arg_ci *a)
10102{
10103 /* This has cond from encoding, required to be outside IT block. */
10104 if (a->cond >= 0xe) {
10105 return false;
10106 }
10107 if (s->condexec_mask) {
10108 unallocated_encoding(s);
10109 return true;
10110 }
10111 arm_skip_unless(s, a->cond);
10112 gen_jmp(s, read_pc(s) + a->imm);
10113 return true;
10114}
10115
10116static bool trans_BL(DisasContext *s, arg_i *a)
10117{
10118 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | s->thumb);
10119 gen_jmp(s, read_pc(s) + a->imm);
10120 return true;
10121}
10122
10123static bool trans_BLX_i(DisasContext *s, arg_BLX_i *a)
10124{
10125 TCGv_i32 tmp;
10126
10127 /* For A32, ARCH(5) is checked near the start of the uncond block. */
10128 if (s->thumb && (a->imm & 2)) {
10129 return false;
10130 }
10131 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | s->thumb);
10132 tmp = tcg_const_i32(!s->thumb);
10133 store_cpu_field(tmp, thumb);
10134 gen_jmp(s, (read_pc(s) & ~3) + a->imm);
10135 return true;
10136}
10137
10138static bool trans_BL_BLX_prefix(DisasContext *s, arg_BL_BLX_prefix *a)
10139{
10140 assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
10141 tcg_gen_movi_i32(cpu_R[14], read_pc(s) + (a->imm << 12));
10142 return true;
10143}
10144
10145static bool trans_BL_suffix(DisasContext *s, arg_BL_suffix *a)
10146{
10147 TCGv_i32 tmp = tcg_temp_new_i32();
10148
10149 assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
10150 tcg_gen_addi_i32(tmp, cpu_R[14], (a->imm << 1) | 1);
10151 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | 1);
10152 gen_bx(s, tmp);
10153 return true;
10154}
10155
10156static bool trans_BLX_suffix(DisasContext *s, arg_BLX_suffix *a)
10157{
10158 TCGv_i32 tmp;
10159
10160 assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
10161 if (!ENABLE_ARCH_5) {
10162 return false;
10163 }
10164 tmp = tcg_temp_new_i32();
10165 tcg_gen_addi_i32(tmp, cpu_R[14], a->imm << 1);
10166 tcg_gen_andi_i32(tmp, tmp, 0xfffffffc);
10167 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | 1);
10168 gen_bx(s, tmp);
10169 return true;
10170}
10171
10172static bool op_tbranch(DisasContext *s, arg_tbranch *a, bool half)
10173{
10174 TCGv_i32 addr, tmp;
10175
10176 tmp = load_reg(s, a->rm);
10177 if (half) {
10178 tcg_gen_add_i32(tmp, tmp, tmp);
10179 }
10180 addr = load_reg(s, a->rn);
10181 tcg_gen_add_i32(addr, addr, tmp);
10182
10183 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s),
10184 half ? MO_UW | s->be_data : MO_UB);
10185 tcg_temp_free_i32(addr);
10186
10187 tcg_gen_add_i32(tmp, tmp, tmp);
10188 tcg_gen_addi_i32(tmp, tmp, read_pc(s));
10189 store_reg(s, 15, tmp);
10190 return true;
10191}
10192
10193static bool trans_TBB(DisasContext *s, arg_tbranch *a)
10194{
10195 return op_tbranch(s, a, false);
10196}
10197
10198static bool trans_TBH(DisasContext *s, arg_tbranch *a)
10199{
10200 return op_tbranch(s, a, true);
10201}
10202
10203static bool trans_CBZ(DisasContext *s, arg_CBZ *a)
10204{
10205 TCGv_i32 tmp = load_reg(s, a->rn);
10206
10207 arm_gen_condlabel(s);
10208 tcg_gen_brcondi_i32(a->nz ? TCG_COND_EQ : TCG_COND_NE,
10209 tmp, 0, s->condlabel);
10210 tcg_temp_free_i32(tmp);
10211 gen_jmp(s, read_pc(s) + a->imm);
10212 return true;
10213}
10214
10215/*
10216 * Supervisor call
10217 */
10218
10219static bool trans_SVC(DisasContext *s, arg_SVC *a)
10220{
10221 gen_set_pc_im(s, s->base.pc_next);
10222 s->svc_imm = a->imm;
10223 s->base.is_jmp = DISAS_SWI;
10224 return true;
10225}
10226
10227/*
10228 * Unconditional system instructions
10229 */
10230
10231static bool trans_RFE(DisasContext *s, arg_RFE *a)
10232{
10233 static const int8_t pre_offset[4] = {
10234 /* DA */ -4, /* IA */ 0, /* DB */ -8, /* IB */ 4
10235 };
10236 static const int8_t post_offset[4] = {
10237 /* DA */ -8, /* IA */ 4, /* DB */ -4, /* IB */ 0
10238 };
10239 TCGv_i32 addr, t1, t2;
10240
10241 if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
10242 return false;
10243 }
10244 if (IS_USER(s)) {
10245 unallocated_encoding(s);
10246 return true;
10247 }
10248
10249 addr = load_reg(s, a->rn);
10250 tcg_gen_addi_i32(addr, addr, pre_offset[a->pu]);
10251
10252 /* Load PC into tmp and CPSR into tmp2. */
10253 t1 = tcg_temp_new_i32();
10254 gen_aa32_ld32u(s, t1, addr, get_mem_index(s));
10255 tcg_gen_addi_i32(addr, addr, 4);
10256 t2 = tcg_temp_new_i32();
10257 gen_aa32_ld32u(s, t2, addr, get_mem_index(s));
10258
10259 if (a->w) {
10260 /* Base writeback. */
10261 tcg_gen_addi_i32(addr, addr, post_offset[a->pu]);
10262 store_reg(s, a->rn, addr);
10263 } else {
10264 tcg_temp_free_i32(addr);
10265 }
10266 gen_rfe(s, t1, t2);
10267 return true;
10268}
10269
10270static bool trans_SRS(DisasContext *s, arg_SRS *a)
10271{
10272 if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
10273 return false;
10274 }
10275 gen_srs(s, a->mode, a->pu, a->w);
10276 return true;
10277}
10278
10279static bool trans_CPS(DisasContext *s, arg_CPS *a)
10280{
10281 uint32_t mask, val;
10282
10283 if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
10284 return false;
10285 }
10286 if (IS_USER(s)) {
10287 /* Implemented as NOP in user mode. */
10288 return true;
10289 }
10290 /* TODO: There are quite a lot of UNPREDICTABLE argument combinations. */
10291
10292 mask = val = 0;
10293 if (a->imod & 2) {
10294 if (a->A) {
10295 mask |= CPSR_A;
10296 }
10297 if (a->I) {
10298 mask |= CPSR_I;
10299 }
10300 if (a->F) {
10301 mask |= CPSR_F;
10302 }
10303 if (a->imod & 1) {
10304 val |= mask;
10305 }
10306 }
10307 if (a->M) {
10308 mask |= CPSR_M;
10309 val |= a->mode;
10310 }
10311 if (mask) {
10312 gen_set_psr_im(s, mask, 0, val);
10313 }
10314 return true;
10315}
10316
10317static bool trans_CPS_v7m(DisasContext *s, arg_CPS_v7m *a)
10318{
10319 TCGv_i32 tmp, addr;
10320
10321 if (!arm_dc_feature(s, ARM_FEATURE_M)) {
10322 return false;
10323 }
10324 if (IS_USER(s)) {
10325 /* Implemented as NOP in user mode. */
10326 return true;
10327 }
10328
10329 tmp = tcg_const_i32(a->im);
10330 /* FAULTMASK */
10331 if (a->F) {
10332 addr = tcg_const_i32(19);
10333 gen_helper_v7m_msr(cpu_env, addr, tmp);
10334 tcg_temp_free_i32(addr);
10335 }
10336 /* PRIMASK */
10337 if (a->I) {
10338 addr = tcg_const_i32(16);
10339 gen_helper_v7m_msr(cpu_env, addr, tmp);
10340 tcg_temp_free_i32(addr);
10341 }
10342 tcg_temp_free_i32(tmp);
10343 gen_lookup_tb(s);
10344 return true;
10345}
10346
10347/*
10348 * Clear-Exclusive, Barriers
10349 */
10350
10351static bool trans_CLREX(DisasContext *s, arg_CLREX *a)
10352{
10353 if (s->thumb
10354 ? !ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)
10355 : !ENABLE_ARCH_6K) {
10356 return false;
10357 }
10358 gen_clrex(s);
10359 return true;
10360}
10361
10362static bool trans_DSB(DisasContext *s, arg_DSB *a)
10363{
10364 if (!ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)) {
10365 return false;
10366 }
10367 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
10368 return true;
10369}
10370
10371static bool trans_DMB(DisasContext *s, arg_DMB *a)
10372{
10373 return trans_DSB(s, NULL);
10374}
10375
10376static bool trans_ISB(DisasContext *s, arg_ISB *a)
10377{
10378 if (!ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)) {
10379 return false;
10380 }
10381 /*
10382 * We need to break the TB after this insn to execute
10383 * self-modifying code correctly and also to take
10384 * any pending interrupts immediately.
10385 */
10386 gen_goto_tb(s, 0, s->base.pc_next);
10387 return true;
10388}
10389
10390static bool trans_SB(DisasContext *s, arg_SB *a)
10391{
10392 if (!dc_isar_feature(aa32_sb, s)) {
10393 return false;
10394 }
10395 /*
10396 * TODO: There is no speculation barrier opcode
10397 * for TCG; MB and end the TB instead.
10398 */
10399 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
10400 gen_goto_tb(s, 0, s->base.pc_next);
10401 return true;
10402}
10403
10404static bool trans_SETEND(DisasContext *s, arg_SETEND *a)
10405{
10406 if (!ENABLE_ARCH_6) {
10407 return false;
10408 }
10409 if (a->E != (s->be_data == MO_BE)) {
10410 gen_helper_setend(cpu_env);
10411 s->base.is_jmp = DISAS_UPDATE;
10412 }
10413 return true;
10414}
10415
10416/*
10417 * Preload instructions
10418 * All are nops, contingent on the appropriate arch level.
10419 */
10420
10421static bool trans_PLD(DisasContext *s, arg_PLD *a)
10422{
10423 return ENABLE_ARCH_5TE;
10424}
10425
10426static bool trans_PLDW(DisasContext *s, arg_PLD *a)
10427{
10428 return arm_dc_feature(s, ARM_FEATURE_V7MP);
10429}
10430
10431static bool trans_PLI(DisasContext *s, arg_PLD *a)
10432{
10433 return ENABLE_ARCH_7;
10434}
10435
10436/*
10437 * If-then
10438 */
10439
10440static bool trans_IT(DisasContext *s, arg_IT *a)
10441{
10442 int cond_mask = a->cond_mask;
10443
10444 /*
10445 * No actual code generated for this insn, just setup state.
10446 *
10447 * Combinations of firstcond and mask which set up an 0b1111
10448 * condition are UNPREDICTABLE; we take the CONSTRAINED
10449 * UNPREDICTABLE choice to treat 0b1111 the same as 0b1110,
10450 * i.e. both meaning "execute always".
10451 */
10452 s->condexec_cond = (cond_mask >> 4) & 0xe;
10453 s->condexec_mask = cond_mask & 0x1f;
10454 return true;
10455}
10456
10457/*
10458 * Legacy decoder.
10459 */
10460
10461static void disas_arm_insn(DisasContext *s, unsigned int insn)
10462{
10463 unsigned int cond = insn >> 28;
10464
10465 /* M variants do not implement ARM mode; this must raise the INVSTATE
10466 * UsageFault exception.
10467 */
10468 if (arm_dc_feature(s, ARM_FEATURE_M)) {
10469 gen_exception_insn(s, s->pc_curr, EXCP_INVSTATE, syn_uncategorized(),
10470 default_exception_el(s));
10471 return;
10472 }
10473
10474 if (cond == 0xf) {
10475 /* In ARMv3 and v4 the NV condition is UNPREDICTABLE; we
10476 * choose to UNDEF. In ARMv5 and above the space is used
10477 * for miscellaneous unconditional instructions.
10478 */
10479 ARCH(5);
10480
10481 /* Unconditional instructions. */
10482 if (disas_a32_uncond(s, insn)) {
10483 return;
10484 }
10485 /* fall back to legacy decoder */
10486
10487 if (((insn >> 25) & 7) == 1) {
10488 /* NEON Data processing. */
10489 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
10490 goto illegal_op;
10491 }
10492
10493 if (disas_neon_data_insn(s, insn)) {
10494 goto illegal_op;
10495 }
10496 return;
10497 }
10498 if ((insn & 0x0f100000) == 0x04000000) {
10499 /* NEON load/store. */
10500 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
10501 goto illegal_op;
10502 }
10503
10504 if (disas_neon_ls_insn(s, insn)) {
10505 goto illegal_op;
10506 }
10507 return;
10508 }
10509 if ((insn & 0x0f000e10) == 0x0e000a00) {
10510 /* VFP. */
10511 if (disas_vfp_insn(s, insn)) {
10512 goto illegal_op;
10513 }
10514 return;
10515 }
10516 if ((insn & 0x0e000f00) == 0x0c000100) {
10517 if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
10518 /* iWMMXt register transfer. */
10519 if (extract32(s->c15_cpar, 1, 1)) {
10520 if (!disas_iwmmxt_insn(s, insn)) {
10521 return;
10522 }
10523 }
10524 }
10525 } else if ((insn & 0x0e000a00) == 0x0c000800
10526 && arm_dc_feature(s, ARM_FEATURE_V8)) {
10527 if (disas_neon_insn_3same_ext(s, insn)) {
10528 goto illegal_op;
10529 }
10530 return;
10531 } else if ((insn & 0x0f000a00) == 0x0e000800
10532 && arm_dc_feature(s, ARM_FEATURE_V8)) {
10533 if (disas_neon_insn_2reg_scalar_ext(s, insn)) {
10534 goto illegal_op;
10535 }
10536 return;
10537 }
10538 goto illegal_op;
10539 }
10540 if (cond != 0xe) {
10541 /* if not always execute, we generate a conditional jump to
10542 next instruction */
10543 arm_skip_unless(s, cond);
10544 }
10545
10546 if (disas_a32(s, insn)) {
10547 return;
10548 }
10549 /* fall back to legacy decoder */
10550
10551 switch ((insn >> 24) & 0xf) {
10552 case 0xc:
10553 case 0xd:
10554 case 0xe:
10555 if (((insn >> 8) & 0xe) == 10) {
10556 /* VFP. */
10557 if (disas_vfp_insn(s, insn)) {
10558 goto illegal_op;
10559 }
10560 } else if (disas_coproc_insn(s, insn)) {
10561 /* Coprocessor. */
10562 goto illegal_op;
10563 }
10564 break;
10565 default:
10566 illegal_op:
10567 unallocated_encoding(s);
10568 break;
10569 }
10570}
10571
10572static bool thumb_insn_is_16bit(DisasContext *s, uint32_t pc, uint32_t insn)
10573{
10574 /*
10575 * Return true if this is a 16 bit instruction. We must be precise
10576 * about this (matching the decode).
10577 */
10578 if ((insn >> 11) < 0x1d) {
10579 /* Definitely a 16-bit instruction */
10580 return true;
10581 }
10582
10583 /* Top five bits 0b11101 / 0b11110 / 0b11111 : this is the
10584 * first half of a 32-bit Thumb insn. Thumb-1 cores might
10585 * end up actually treating this as two 16-bit insns, though,
10586 * if it's half of a bl/blx pair that might span a page boundary.
10587 */
10588 if (arm_dc_feature(s, ARM_FEATURE_THUMB2) ||
10589 arm_dc_feature(s, ARM_FEATURE_M)) {
10590 /* Thumb2 cores (including all M profile ones) always treat
10591 * 32-bit insns as 32-bit.
10592 */
10593 return false;
10594 }
10595
10596 if ((insn >> 11) == 0x1e && pc - s->page_start < TARGET_PAGE_SIZE - 3) {
10597 /* 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix, and the suffix
10598 * is not on the next page; we merge this into a 32-bit
10599 * insn.
10600 */
10601 return false;
10602 }
10603 /* 0b1110_1xxx_xxxx_xxxx : BLX suffix (or UNDEF);
10604 * 0b1111_1xxx_xxxx_xxxx : BL suffix;
10605 * 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix on the end of a page
10606 * -- handle as single 16 bit insn
10607 */
10608 return true;
10609}
10610
10611/* Translate a 32-bit thumb instruction. */
10612static void disas_thumb2_insn(DisasContext *s, uint32_t insn)
10613{
10614 /*
10615 * ARMv6-M supports a limited subset of Thumb2 instructions.
10616 * Other Thumb1 architectures allow only 32-bit
10617 * combined BL/BLX prefix and suffix.
10618 */
10619 if (arm_dc_feature(s, ARM_FEATURE_M) &&
10620 !arm_dc_feature(s, ARM_FEATURE_V7)) {
10621 int i;
10622 bool found = false;
10623 static const uint32_t armv6m_insn[] = {0xf3808000 /* msr */,
10624 0xf3b08040 /* dsb */,
10625 0xf3b08050 /* dmb */,
10626 0xf3b08060 /* isb */,
10627 0xf3e08000 /* mrs */,
10628 0xf000d000 /* bl */};
10629 static const uint32_t armv6m_mask[] = {0xffe0d000,
10630 0xfff0d0f0,
10631 0xfff0d0f0,
10632 0xfff0d0f0,
10633 0xffe0d000,
10634 0xf800d000};
10635
10636 for (i = 0; i < ARRAY_SIZE(armv6m_insn); i++) {
10637 if ((insn & armv6m_mask[i]) == armv6m_insn[i]) {
10638 found = true;
10639 break;
10640 }
10641 }
10642 if (!found) {
10643 goto illegal_op;
10644 }
10645 } else if ((insn & 0xf800e800) != 0xf000e800) {
10646 ARCH(6T2);
10647 }
10648
10649 if (disas_t32(s, insn)) {
10650 return;
10651 }
10652 /* fall back to legacy decoder */
10653
10654 switch ((insn >> 25) & 0xf) {
10655 case 0: case 1: case 2: case 3:
10656 /* 16-bit instructions. Should never happen. */
10657 abort();
10658 case 6: case 7: case 14: case 15:
10659 /* Coprocessor. */
10660 if (arm_dc_feature(s, ARM_FEATURE_M)) {
10661 /* 0b111x_11xx_xxxx_xxxx_xxxx_xxxx_xxxx_xxxx */
10662 if (extract32(insn, 24, 2) == 3) {
10663 goto illegal_op; /* op0 = 0b11 : unallocated */
10664 }
10665
10666 /*
10667 * Decode VLLDM and VLSTM first: these are nonstandard because:
10668 * * if there is no FPU then these insns must NOP in
10669 * Secure state and UNDEF in Nonsecure state
10670 * * if there is an FPU then these insns do not have
10671 * the usual behaviour that disas_vfp_insn() provides of
10672 * being controlled by CPACR/NSACR enable bits or the
10673 * lazy-stacking logic.
10674 */
10675 if (arm_dc_feature(s, ARM_FEATURE_V8) &&
10676 (insn & 0xffa00f00) == 0xec200a00) {
10677 /* 0b1110_1100_0x1x_xxxx_xxxx_1010_xxxx_xxxx
10678 * - VLLDM, VLSTM
10679 * We choose to UNDEF if the RAZ bits are non-zero.
10680 */
10681 if (!s->v8m_secure || (insn & 0x0040f0ff)) {
10682 goto illegal_op;
10683 }
10684
10685 if (arm_dc_feature(s, ARM_FEATURE_VFP)) {
10686 uint32_t rn = (insn >> 16) & 0xf;
10687 TCGv_i32 fptr = load_reg(s, rn);
10688
10689 if (extract32(insn, 20, 1)) {
10690 gen_helper_v7m_vlldm(cpu_env, fptr);
10691 } else {
10692 gen_helper_v7m_vlstm(cpu_env, fptr);
10693 }
10694 tcg_temp_free_i32(fptr);
10695
10696 /* End the TB, because we have updated FP control bits */
10697 s->base.is_jmp = DISAS_UPDATE;
10698 }
10699 break;
10700 }
10701 if (arm_dc_feature(s, ARM_FEATURE_VFP) &&
10702 ((insn >> 8) & 0xe) == 10) {
10703 /* FP, and the CPU supports it */
10704 if (disas_vfp_insn(s, insn)) {
10705 goto illegal_op;
10706 }
10707 break;
10708 }
10709
10710 /* All other insns: NOCP */
10711 gen_exception_insn(s, s->pc_curr, EXCP_NOCP, syn_uncategorized(),
10712 default_exception_el(s));
10713 break;
10714 }
10715 if ((insn & 0xfe000a00) == 0xfc000800
10716 && arm_dc_feature(s, ARM_FEATURE_V8)) {
10717 /* The Thumb2 and ARM encodings are identical. */
10718 if (disas_neon_insn_3same_ext(s, insn)) {
10719 goto illegal_op;
10720 }
10721 } else if ((insn & 0xff000a00) == 0xfe000800
10722 && arm_dc_feature(s, ARM_FEATURE_V8)) {
10723 /* The Thumb2 and ARM encodings are identical. */
10724 if (disas_neon_insn_2reg_scalar_ext(s, insn)) {
10725 goto illegal_op;
10726 }
10727 } else if (((insn >> 24) & 3) == 3) {
10728 /* Translate into the equivalent ARM encoding. */
10729 insn = (insn & 0xe2ffffff) | ((insn & (1 << 28)) >> 4) | (1 << 28);
10730 if (disas_neon_data_insn(s, insn)) {
10731 goto illegal_op;
10732 }
10733 } else if (((insn >> 8) & 0xe) == 10) {
10734 if (disas_vfp_insn(s, insn)) {
10735 goto illegal_op;
10736 }
10737 } else {
10738 if (insn & (1 << 28))
10739 goto illegal_op;
10740 if (disas_coproc_insn(s, insn)) {
10741 goto illegal_op;
10742 }
10743 }
10744 break;
10745 case 12:
10746 if ((insn & 0x01100000) == 0x01000000) {
10747 if (disas_neon_ls_insn(s, insn)) {
10748 goto illegal_op;
10749 }
10750 break;
10751 }
10752 goto illegal_op;
10753 default:
10754 illegal_op:
10755 unallocated_encoding(s);
10756 }
10757}
10758
10759static void disas_thumb_insn(DisasContext *s, uint32_t insn)
10760{
10761 if (!disas_t16(s, insn)) {
10762 unallocated_encoding(s);
10763 }
10764}
10765
10766static bool insn_crosses_page(CPUARMState *env, DisasContext *s)
10767{
10768 /* Return true if the insn at dc->base.pc_next might cross a page boundary.
10769 * (False positives are OK, false negatives are not.)
10770 * We know this is a Thumb insn, and our caller ensures we are
10771 * only called if dc->base.pc_next is less than 4 bytes from the page
10772 * boundary, so we cross the page if the first 16 bits indicate
10773 * that this is a 32 bit insn.
10774 */
10775 uint16_t insn = arm_lduw_code(env, s->base.pc_next, s->sctlr_b);
10776
10777 return !thumb_insn_is_16bit(s, s->base.pc_next, insn);
10778}
10779
10780static void arm_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cs)
10781{
10782 DisasContext *dc = container_of(dcbase, DisasContext, base);
10783 CPUARMState *env = cs->env_ptr;
10784 ARMCPU *cpu = env_archcpu(env);
10785 uint32_t tb_flags = dc->base.tb->flags;
10786 uint32_t condexec, core_mmu_idx;
10787
10788 dc->isar = &cpu->isar;
10789 dc->condjmp = 0;
10790
10791 dc->aarch64 = 0;
10792 /* If we are coming from secure EL0 in a system with a 32-bit EL3, then
10793 * there is no secure EL1, so we route exceptions to EL3.
10794 */
10795 dc->secure_routed_to_el3 = arm_feature(env, ARM_FEATURE_EL3) &&
10796 !arm_el_is_aa64(env, 3);
10797 dc->thumb = FIELD_EX32(tb_flags, TBFLAG_A32, THUMB);
10798 dc->sctlr_b = FIELD_EX32(tb_flags, TBFLAG_A32, SCTLR_B);
10799 dc->be_data = FIELD_EX32(tb_flags, TBFLAG_ANY, BE_DATA) ? MO_BE : MO_LE;
10800 condexec = FIELD_EX32(tb_flags, TBFLAG_A32, CONDEXEC);
10801 dc->condexec_mask = (condexec & 0xf) << 1;
10802 dc->condexec_cond = condexec >> 4;
10803 core_mmu_idx = FIELD_EX32(tb_flags, TBFLAG_ANY, MMUIDX);
10804 dc->mmu_idx = core_to_arm_mmu_idx(env, core_mmu_idx);
10805 dc->current_el = arm_mmu_idx_to_el(dc->mmu_idx);
10806#if !defined(CONFIG_USER_ONLY)
10807 dc->user = (dc->current_el == 0);
10808#endif
10809 dc->ns = FIELD_EX32(tb_flags, TBFLAG_A32, NS);
10810 dc->fp_excp_el = FIELD_EX32(tb_flags, TBFLAG_ANY, FPEXC_EL);
10811 dc->vfp_enabled = FIELD_EX32(tb_flags, TBFLAG_A32, VFPEN);
10812 dc->vec_len = FIELD_EX32(tb_flags, TBFLAG_A32, VECLEN);
10813 if (arm_feature(env, ARM_FEATURE_XSCALE)) {
10814 dc->c15_cpar = FIELD_EX32(tb_flags, TBFLAG_A32, XSCALE_CPAR);
10815 dc->vec_stride = 0;
10816 } else {
10817 dc->vec_stride = FIELD_EX32(tb_flags, TBFLAG_A32, VECSTRIDE);
10818 dc->c15_cpar = 0;
10819 }
10820 dc->v7m_handler_mode = FIELD_EX32(tb_flags, TBFLAG_A32, HANDLER);
10821 dc->v8m_secure = arm_feature(env, ARM_FEATURE_M_SECURITY) &&
10822 regime_is_secure(env, dc->mmu_idx);
10823 dc->v8m_stackcheck = FIELD_EX32(tb_flags, TBFLAG_A32, STACKCHECK);
10824 dc->v8m_fpccr_s_wrong = FIELD_EX32(tb_flags, TBFLAG_A32, FPCCR_S_WRONG);
10825 dc->v7m_new_fp_ctxt_needed =
10826 FIELD_EX32(tb_flags, TBFLAG_A32, NEW_FP_CTXT_NEEDED);
10827 dc->v7m_lspact = FIELD_EX32(tb_flags, TBFLAG_A32, LSPACT);
10828 dc->cp_regs = cpu->cp_regs;
10829 dc->features = env->features;
10830
10831 /* Single step state. The code-generation logic here is:
10832 * SS_ACTIVE == 0:
10833 * generate code with no special handling for single-stepping (except
10834 * that anything that can make us go to SS_ACTIVE == 1 must end the TB;
10835 * this happens anyway because those changes are all system register or
10836 * PSTATE writes).
10837 * SS_ACTIVE == 1, PSTATE.SS == 1: (active-not-pending)
10838 * emit code for one insn
10839 * emit code to clear PSTATE.SS
10840 * emit code to generate software step exception for completed step
10841 * end TB (as usual for having generated an exception)
10842 * SS_ACTIVE == 1, PSTATE.SS == 0: (active-pending)
10843 * emit code to generate a software step exception
10844 * end the TB
10845 */
10846 dc->ss_active = FIELD_EX32(tb_flags, TBFLAG_ANY, SS_ACTIVE);
10847 dc->pstate_ss = FIELD_EX32(tb_flags, TBFLAG_ANY, PSTATE_SS);
10848 dc->is_ldex = false;
10849 if (!arm_feature(env, ARM_FEATURE_M)) {
10850 dc->debug_target_el = FIELD_EX32(tb_flags, TBFLAG_ANY, DEBUG_TARGET_EL);
10851 }
10852
10853 dc->page_start = dc->base.pc_first & TARGET_PAGE_MASK;
10854
10855 /* If architectural single step active, limit to 1. */
10856 if (is_singlestepping(dc)) {
10857 dc->base.max_insns = 1;
10858 }
10859
10860 /* ARM is a fixed-length ISA. Bound the number of insns to execute
10861 to those left on the page. */
10862 if (!dc->thumb) {
10863 int bound = -(dc->base.pc_first | TARGET_PAGE_MASK) / 4;
10864 dc->base.max_insns = MIN(dc->base.max_insns, bound);
10865 }
10866
10867 cpu_V0 = tcg_temp_new_i64();
10868 cpu_V1 = tcg_temp_new_i64();
10869 /* FIXME: cpu_M0 can probably be the same as cpu_V0. */
10870 cpu_M0 = tcg_temp_new_i64();
10871}
10872
10873static void arm_tr_tb_start(DisasContextBase *dcbase, CPUState *cpu)
10874{
10875 DisasContext *dc = container_of(dcbase, DisasContext, base);
10876
10877 /* A note on handling of the condexec (IT) bits:
10878 *
10879 * We want to avoid the overhead of having to write the updated condexec
10880 * bits back to the CPUARMState for every instruction in an IT block. So:
10881 * (1) if the condexec bits are not already zero then we write
10882 * zero back into the CPUARMState now. This avoids complications trying
10883 * to do it at the end of the block. (For example if we don't do this
10884 * it's hard to identify whether we can safely skip writing condexec
10885 * at the end of the TB, which we definitely want to do for the case
10886 * where a TB doesn't do anything with the IT state at all.)
10887 * (2) if we are going to leave the TB then we call gen_set_condexec()
10888 * which will write the correct value into CPUARMState if zero is wrong.
10889 * This is done both for leaving the TB at the end, and for leaving
10890 * it because of an exception we know will happen, which is done in
10891 * gen_exception_insn(). The latter is necessary because we need to
10892 * leave the TB with the PC/IT state just prior to execution of the
10893 * instruction which caused the exception.
10894 * (3) if we leave the TB unexpectedly (eg a data abort on a load)
10895 * then the CPUARMState will be wrong and we need to reset it.
10896 * This is handled in the same way as restoration of the
10897 * PC in these situations; we save the value of the condexec bits
10898 * for each PC via tcg_gen_insn_start(), and restore_state_to_opc()
10899 * then uses this to restore them after an exception.
10900 *
10901 * Note that there are no instructions which can read the condexec
10902 * bits, and none which can write non-static values to them, so
10903 * we don't need to care about whether CPUARMState is correct in the
10904 * middle of a TB.
10905 */
10906
10907 /* Reset the conditional execution bits immediately. This avoids
10908 complications trying to do it at the end of the block. */
10909 if (dc->condexec_mask || dc->condexec_cond) {
10910 TCGv_i32 tmp = tcg_temp_new_i32();
10911 tcg_gen_movi_i32(tmp, 0);
10912 store_cpu_field(tmp, condexec_bits);
10913 }
10914}
10915
10916static void arm_tr_insn_start(DisasContextBase *dcbase, CPUState *cpu)
10917{
10918 DisasContext *dc = container_of(dcbase, DisasContext, base);
10919
10920 tcg_gen_insn_start(dc->base.pc_next,
10921 (dc->condexec_cond << 4) | (dc->condexec_mask >> 1),
10922 0);
10923 dc->insn_start = tcg_last_op();
10924}
10925
10926static bool arm_tr_breakpoint_check(DisasContextBase *dcbase, CPUState *cpu,
10927 const CPUBreakpoint *bp)
10928{
10929 DisasContext *dc = container_of(dcbase, DisasContext, base);
10930
10931 if (bp->flags & BP_CPU) {
10932 gen_set_condexec(dc);
10933 gen_set_pc_im(dc, dc->base.pc_next);
10934 gen_helper_check_breakpoints(cpu_env);
10935 /* End the TB early; it's likely not going to be executed */
10936 dc->base.is_jmp = DISAS_TOO_MANY;
10937 } else {
10938 gen_exception_internal_insn(dc, dc->base.pc_next, EXCP_DEBUG);
10939 /* The address covered by the breakpoint must be
10940 included in [tb->pc, tb->pc + tb->size) in order
10941 to for it to be properly cleared -- thus we
10942 increment the PC here so that the logic setting
10943 tb->size below does the right thing. */
10944 /* TODO: Advance PC by correct instruction length to
10945 * avoid disassembler error messages */
10946 dc->base.pc_next += 2;
10947 dc->base.is_jmp = DISAS_NORETURN;
10948 }
10949
10950 return true;
10951}
10952
10953static bool arm_pre_translate_insn(DisasContext *dc)
10954{
10955#ifdef CONFIG_USER_ONLY
10956 /* Intercept jump to the magic kernel page. */
10957 if (dc->base.pc_next >= 0xffff0000) {
10958 /* We always get here via a jump, so know we are not in a
10959 conditional execution block. */
10960 gen_exception_internal(EXCP_KERNEL_TRAP);
10961 dc->base.is_jmp = DISAS_NORETURN;
10962 return true;
10963 }
10964#endif
10965
10966 if (dc->ss_active && !dc->pstate_ss) {
10967 /* Singlestep state is Active-pending.
10968 * If we're in this state at the start of a TB then either
10969 * a) we just took an exception to an EL which is being debugged
10970 * and this is the first insn in the exception handler
10971 * b) debug exceptions were masked and we just unmasked them
10972 * without changing EL (eg by clearing PSTATE.D)
10973 * In either case we're going to take a swstep exception in the
10974 * "did not step an insn" case, and so the syndrome ISV and EX
10975 * bits should be zero.
10976 */
10977 assert(dc->base.num_insns == 1);
10978 gen_swstep_exception(dc, 0, 0);
10979 dc->base.is_jmp = DISAS_NORETURN;
10980 return true;
10981 }
10982
10983 return false;
10984}
10985
10986static void arm_post_translate_insn(DisasContext *dc)
10987{
10988 if (dc->condjmp && !dc->base.is_jmp) {
10989 gen_set_label(dc->condlabel);
10990 dc->condjmp = 0;
10991 }
10992 translator_loop_temp_check(&dc->base);
10993}
10994
10995static void arm_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
10996{
10997 DisasContext *dc = container_of(dcbase, DisasContext, base);
10998 CPUARMState *env = cpu->env_ptr;
10999 unsigned int insn;
11000
11001 if (arm_pre_translate_insn(dc)) {
11002 return;
11003 }
11004
11005 dc->pc_curr = dc->base.pc_next;
11006 insn = arm_ldl_code(env, dc->base.pc_next, dc->sctlr_b);
11007 dc->insn = insn;
11008 dc->base.pc_next += 4;
11009 disas_arm_insn(dc, insn);
11010
11011 arm_post_translate_insn(dc);
11012
11013 /* ARM is a fixed-length ISA. We performed the cross-page check
11014 in init_disas_context by adjusting max_insns. */
11015}
11016
11017static bool thumb_insn_is_unconditional(DisasContext *s, uint32_t insn)
11018{
11019 /* Return true if this Thumb insn is always unconditional,
11020 * even inside an IT block. This is true of only a very few
11021 * instructions: BKPT, HLT, and SG.
11022 *
11023 * A larger class of instructions are UNPREDICTABLE if used
11024 * inside an IT block; we do not need to detect those here, because
11025 * what we do by default (perform the cc check and update the IT
11026 * bits state machine) is a permitted CONSTRAINED UNPREDICTABLE
11027 * choice for those situations.
11028 *
11029 * insn is either a 16-bit or a 32-bit instruction; the two are
11030 * distinguishable because for the 16-bit case the top 16 bits
11031 * are zeroes, and that isn't a valid 32-bit encoding.
11032 */
11033 if ((insn & 0xffffff00) == 0xbe00) {
11034 /* BKPT */
11035 return true;
11036 }
11037
11038 if ((insn & 0xffffffc0) == 0xba80 && arm_dc_feature(s, ARM_FEATURE_V8) &&
11039 !arm_dc_feature(s, ARM_FEATURE_M)) {
11040 /* HLT: v8A only. This is unconditional even when it is going to
11041 * UNDEF; see the v8A ARM ARM DDI0487B.a H3.3.
11042 * For v7 cores this was a plain old undefined encoding and so
11043 * honours its cc check. (We might be using the encoding as
11044 * a semihosting trap, but we don't change the cc check behaviour
11045 * on that account, because a debugger connected to a real v7A
11046 * core and emulating semihosting traps by catching the UNDEF
11047 * exception would also only see cases where the cc check passed.
11048 * No guest code should be trying to do a HLT semihosting trap
11049 * in an IT block anyway.
11050 */
11051 return true;
11052 }
11053
11054 if (insn == 0xe97fe97f && arm_dc_feature(s, ARM_FEATURE_V8) &&
11055 arm_dc_feature(s, ARM_FEATURE_M)) {
11056 /* SG: v8M only */
11057 return true;
11058 }
11059
11060 return false;
11061}
11062
11063static void thumb_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
11064{
11065 DisasContext *dc = container_of(dcbase, DisasContext, base);
11066 CPUARMState *env = cpu->env_ptr;
11067 uint32_t insn;
11068 bool is_16bit;
11069
11070 if (arm_pre_translate_insn(dc)) {
11071 return;
11072 }
11073
11074 dc->pc_curr = dc->base.pc_next;
11075 insn = arm_lduw_code(env, dc->base.pc_next, dc->sctlr_b);
11076 is_16bit = thumb_insn_is_16bit(dc, dc->base.pc_next, insn);
11077 dc->base.pc_next += 2;
11078 if (!is_16bit) {
11079 uint32_t insn2 = arm_lduw_code(env, dc->base.pc_next, dc->sctlr_b);
11080
11081 insn = insn << 16 | insn2;
11082 dc->base.pc_next += 2;
11083 }
11084 dc->insn = insn;
11085
11086 if (dc->condexec_mask && !thumb_insn_is_unconditional(dc, insn)) {
11087 uint32_t cond = dc->condexec_cond;
11088
11089 /*
11090 * Conditionally skip the insn. Note that both 0xe and 0xf mean
11091 * "always"; 0xf is not "never".
11092 */
11093 if (cond < 0x0e) {
11094 arm_skip_unless(dc, cond);
11095 }
11096 }
11097
11098 if (is_16bit) {
11099 disas_thumb_insn(dc, insn);
11100 } else {
11101 disas_thumb2_insn(dc, insn);
11102 }
11103
11104 /* Advance the Thumb condexec condition. */
11105 if (dc->condexec_mask) {
11106 dc->condexec_cond = ((dc->condexec_cond & 0xe) |
11107 ((dc->condexec_mask >> 4) & 1));
11108 dc->condexec_mask = (dc->condexec_mask << 1) & 0x1f;
11109 if (dc->condexec_mask == 0) {
11110 dc->condexec_cond = 0;
11111 }
11112 }
11113
11114 arm_post_translate_insn(dc);
11115
11116 /* Thumb is a variable-length ISA. Stop translation when the next insn
11117 * will touch a new page. This ensures that prefetch aborts occur at
11118 * the right place.
11119 *
11120 * We want to stop the TB if the next insn starts in a new page,
11121 * or if it spans between this page and the next. This means that
11122 * if we're looking at the last halfword in the page we need to
11123 * see if it's a 16-bit Thumb insn (which will fit in this TB)
11124 * or a 32-bit Thumb insn (which won't).
11125 * This is to avoid generating a silly TB with a single 16-bit insn
11126 * in it at the end of this page (which would execute correctly
11127 * but isn't very efficient).
11128 */
11129 if (dc->base.is_jmp == DISAS_NEXT
11130 && (dc->base.pc_next - dc->page_start >= TARGET_PAGE_SIZE
11131 || (dc->base.pc_next - dc->page_start >= TARGET_PAGE_SIZE - 3
11132 && insn_crosses_page(env, dc)))) {
11133 dc->base.is_jmp = DISAS_TOO_MANY;
11134 }
11135}
11136
11137static void arm_tr_tb_stop(DisasContextBase *dcbase, CPUState *cpu)
11138{
11139 DisasContext *dc = container_of(dcbase, DisasContext, base);
11140
11141 if (tb_cflags(dc->base.tb) & CF_LAST_IO && dc->condjmp) {
11142 /* FIXME: This can theoretically happen with self-modifying code. */
11143 cpu_abort(cpu, "IO on conditional branch instruction");
11144 }
11145
11146 /* At this stage dc->condjmp will only be set when the skipped
11147 instruction was a conditional branch or trap, and the PC has
11148 already been written. */
11149 gen_set_condexec(dc);
11150 if (dc->base.is_jmp == DISAS_BX_EXCRET) {
11151 /* Exception return branches need some special case code at the
11152 * end of the TB, which is complex enough that it has to
11153 * handle the single-step vs not and the condition-failed
11154 * insn codepath itself.
11155 */
11156 gen_bx_excret_final_code(dc);
11157 } else if (unlikely(is_singlestepping(dc))) {
11158 /* Unconditional and "condition passed" instruction codepath. */
11159 switch (dc->base.is_jmp) {
11160 case DISAS_SWI:
11161 gen_ss_advance(dc);
11162 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
11163 default_exception_el(dc));
11164 break;
11165 case DISAS_HVC:
11166 gen_ss_advance(dc);
11167 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
11168 break;
11169 case DISAS_SMC:
11170 gen_ss_advance(dc);
11171 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
11172 break;
11173 case DISAS_NEXT:
11174 case DISAS_TOO_MANY:
11175 case DISAS_UPDATE:
11176 gen_set_pc_im(dc, dc->base.pc_next);
11177 /* fall through */
11178 default:
11179 /* FIXME: Single stepping a WFI insn will not halt the CPU. */
11180 gen_singlestep_exception(dc);
11181 break;
11182 case DISAS_NORETURN:
11183 break;
11184 }
11185 } else {
11186 /* While branches must always occur at the end of an IT block,
11187 there are a few other things that can cause us to terminate
11188 the TB in the middle of an IT block:
11189 - Exception generating instructions (bkpt, swi, undefined).
11190 - Page boundaries.
11191 - Hardware watchpoints.
11192 Hardware breakpoints have already been handled and skip this code.
11193 */
11194 switch(dc->base.is_jmp) {
11195 case DISAS_NEXT:
11196 case DISAS_TOO_MANY:
11197 gen_goto_tb(dc, 1, dc->base.pc_next);
11198 break;
11199 case DISAS_JUMP:
11200 gen_goto_ptr();
11201 break;
11202 case DISAS_UPDATE:
11203 gen_set_pc_im(dc, dc->base.pc_next);
11204 /* fall through */
11205 default:
11206 /* indicate that the hash table must be used to find the next TB */
11207 tcg_gen_exit_tb(NULL, 0);
11208 break;
11209 case DISAS_NORETURN:
11210 /* nothing more to generate */
11211 break;
11212 case DISAS_WFI:
11213 {
11214 TCGv_i32 tmp = tcg_const_i32((dc->thumb &&
11215 !(dc->insn & (1U << 31))) ? 2 : 4);
11216
11217 gen_helper_wfi(cpu_env, tmp);
11218 tcg_temp_free_i32(tmp);
11219 /* The helper doesn't necessarily throw an exception, but we
11220 * must go back to the main loop to check for interrupts anyway.
11221 */
11222 tcg_gen_exit_tb(NULL, 0);
11223 break;
11224 }
11225 case DISAS_WFE:
11226 gen_helper_wfe(cpu_env);
11227 break;
11228 case DISAS_YIELD:
11229 gen_helper_yield(cpu_env);
11230 break;
11231 case DISAS_SWI:
11232 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
11233 default_exception_el(dc));
11234 break;
11235 case DISAS_HVC:
11236 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
11237 break;
11238 case DISAS_SMC:
11239 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
11240 break;
11241 }
11242 }
11243
11244 if (dc->condjmp) {
11245 /* "Condition failed" instruction codepath for the branch/trap insn */
11246 gen_set_label(dc->condlabel);
11247 gen_set_condexec(dc);
11248 if (unlikely(is_singlestepping(dc))) {
11249 gen_set_pc_im(dc, dc->base.pc_next);
11250 gen_singlestep_exception(dc);
11251 } else {
11252 gen_goto_tb(dc, 1, dc->base.pc_next);
11253 }
11254 }
11255}
11256
11257static void arm_tr_disas_log(const DisasContextBase *dcbase, CPUState *cpu)
11258{
11259 DisasContext *dc = container_of(dcbase, DisasContext, base);
11260
11261 qemu_log("IN: %s\n", lookup_symbol(dc->base.pc_first));
11262 log_target_disas(cpu, dc->base.pc_first, dc->base.tb->size);
11263}
11264
11265static const TranslatorOps arm_translator_ops = {
11266 .init_disas_context = arm_tr_init_disas_context,
11267 .tb_start = arm_tr_tb_start,
11268 .insn_start = arm_tr_insn_start,
11269 .breakpoint_check = arm_tr_breakpoint_check,
11270 .translate_insn = arm_tr_translate_insn,
11271 .tb_stop = arm_tr_tb_stop,
11272 .disas_log = arm_tr_disas_log,
11273};
11274
11275static const TranslatorOps thumb_translator_ops = {
11276 .init_disas_context = arm_tr_init_disas_context,
11277 .tb_start = arm_tr_tb_start,
11278 .insn_start = arm_tr_insn_start,
11279 .breakpoint_check = arm_tr_breakpoint_check,
11280 .translate_insn = thumb_tr_translate_insn,
11281 .tb_stop = arm_tr_tb_stop,
11282 .disas_log = arm_tr_disas_log,
11283};
11284
11285/* generate intermediate code for basic block 'tb'. */
11286void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int max_insns)
11287{
11288 DisasContext dc;
11289 const TranslatorOps *ops = &arm_translator_ops;
11290
11291 if (FIELD_EX32(tb->flags, TBFLAG_A32, THUMB)) {
11292 ops = &thumb_translator_ops;
11293 }
11294#ifdef TARGET_AARCH64
11295 if (FIELD_EX32(tb->flags, TBFLAG_ANY, AARCH64_STATE)) {
11296 ops = &aarch64_translator_ops;
11297 }
11298#endif
11299
11300 translator_loop(ops, &dc.base, cpu, tb, max_insns);
11301}
11302
11303void restore_state_to_opc(CPUARMState *env, TranslationBlock *tb,
11304 target_ulong *data)
11305{
11306 if (is_a64(env)) {
11307 env->pc = data[0];
11308 env->condexec_bits = 0;
11309 env->exception.syndrome = data[2] << ARM_INSN_START_WORD2_SHIFT;
11310 } else {
11311 env->regs[15] = data[0];
11312 env->condexec_bits = data[1];
11313 env->exception.syndrome = data[2] << ARM_INSN_START_WORD2_SHIFT;
11314 }
11315}
11316