?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
target/arm: Fix sve_zcr_len_for_el for VHE mode running
[qemu/rayw.git] / target / arm / translate.c
blobbf2196b9e24c3398584ca66b1d7b4bc349f739d0
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 2.1 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/tcg-op.h"
28 #include "tcg/tcg-op-gvec.h"
29 #include "qemu/log.h"
30 #include "qemu/bitops.h"
31 #include "arm_ldst.h"
32 #include "semihosting/semihost.h"
33
34 #include "exec/helper-proto.h"
35 #include "exec/helper-gen.h"
36
37 #include "exec/log.h"
38
39
40 #define ENABLE_ARCH_4T arm_dc_feature(s, ARM_FEATURE_V4T)
41 #define ENABLE_ARCH_5 arm_dc_feature(s, ARM_FEATURE_V5)
42 /* currently all emulated v5 cores are also v5TE, so don't bother */
43 #define ENABLE_ARCH_5TE arm_dc_feature(s, ARM_FEATURE_V5)
44 #define ENABLE_ARCH_5J dc_isar_feature(aa32_jazelle, s)
45 #define ENABLE_ARCH_6 arm_dc_feature(s, ARM_FEATURE_V6)
46 #define ENABLE_ARCH_6K arm_dc_feature(s, ARM_FEATURE_V6K)
47 #define ENABLE_ARCH_6T2 arm_dc_feature(s, ARM_FEATURE_THUMB2)
48 #define ENABLE_ARCH_7 arm_dc_feature(s, ARM_FEATURE_V7)
49 #define ENABLE_ARCH_8 arm_dc_feature(s, ARM_FEATURE_V8)
50
51 #include "translate.h"
52 #include "translate-a32.h"
53
54 /* These are TCG temporaries used only by the legacy iwMMXt decoder */
55 static TCGv_i64 cpu_V0, cpu_V1, cpu_M0;
56 /* These are TCG globals which alias CPUARMState fields */
57 static TCGv_i32 cpu_R[16];
58 TCGv_i32 cpu_CF, cpu_NF, cpu_VF, cpu_ZF;
59 TCGv_i64 cpu_exclusive_addr;
60 TCGv_i64 cpu_exclusive_val;
61
62 #include "exec/gen-icount.h"
63
64 static const char * const regnames[] =
65 { "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7",
66 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "pc" };
67
68
69 /* initialize TCG globals. */
70 void arm_translate_init(void)
71 {
72 int i;
73
74 for (i = 0; i < 16; i++) {
75 cpu_R[i] = tcg_global_mem_new_i32(cpu_env,
76 offsetof(CPUARMState, regs[i]),
77 regnames[i]);
78 }
79 cpu_CF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, CF), "CF");
80 cpu_NF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, NF), "NF");
81 cpu_VF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, VF), "VF");
82 cpu_ZF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, ZF), "ZF");
83
84 cpu_exclusive_addr = tcg_global_mem_new_i64(cpu_env,
85 offsetof(CPUARMState, exclusive_addr), "exclusive_addr");
86 cpu_exclusive_val = tcg_global_mem_new_i64(cpu_env,
87 offsetof(CPUARMState, exclusive_val), "exclusive_val");
88
89 a64_translate_init();
90 }
91
92 uint64_t asimd_imm_const(uint32_t imm, int cmode, int op)
93 {
94 /* Expand the encoded constant as per AdvSIMDExpandImm pseudocode */
95 switch (cmode) {
96 case 0: case 1:
97 /* no-op */
98 break;
99 case 2: case 3:
100 imm <<= 8;
101 break;
102 case 4: case 5:
103 imm <<= 16;
104 break;
105 case 6: case 7:
106 imm <<= 24;
107 break;
108 case 8: case 9:
109 imm |= imm << 16;
110 break;
111 case 10: case 11:
112 imm = (imm << 8) | (imm << 24);
113 break;
114 case 12:
115 imm = (imm << 8) | 0xff;
116 break;
117 case 13:
118 imm = (imm << 16) | 0xffff;
119 break;
120 case 14:
121 if (op) {
122 /*
123 * This and cmode == 15 op == 1 are the only cases where
124 * the top and bottom 32 bits of the encoded constant differ.
125 */
126 uint64_t imm64 = 0;
127 int n;
128
129 for (n = 0; n < 8; n++) {
130 if (imm & (1 << n)) {
131 imm64 |= (0xffULL << (n * 8));
132 }
133 }
134 return imm64;
135 }
136 imm |= (imm << 8) | (imm << 16) | (imm << 24);
137 break;
138 case 15:
139 if (op) {
140 /* Reserved encoding for AArch32; valid for AArch64 */
141 uint64_t imm64 = (uint64_t)(imm & 0x3f) << 48;
142 if (imm & 0x80) {
143 imm64 |= 0x8000000000000000ULL;
144 }
145 if (imm & 0x40) {
146 imm64 |= 0x3fc0000000000000ULL;
147 } else {
148 imm64 |= 0x4000000000000000ULL;
149 }
150 return imm64;
151 }
152 imm = ((imm & 0x80) << 24) | ((imm & 0x3f) << 19)
153 | ((imm & 0x40) ? (0x1f << 25) : (1 << 30));
154 break;
155 }
156 if (op) {
157 imm = ~imm;
158 }
159 return dup_const(MO_32, imm);
160 }
161
162 /* Generate a label used for skipping this instruction */
163 void arm_gen_condlabel(DisasContext *s)
164 {
165 if (!s->condjmp) {
166 s->condlabel = gen_new_label();
167 s->condjmp = 1;
168 }
169 }
170
171 /* Flags for the disas_set_da_iss info argument:
172 * lower bits hold the Rt register number, higher bits are flags.
173 */
174 typedef enum ISSInfo {
175 ISSNone = 0,
176 ISSRegMask = 0x1f,
177 ISSInvalid = (1 << 5),
178 ISSIsAcqRel = (1 << 6),
179 ISSIsWrite = (1 << 7),
180 ISSIs16Bit = (1 << 8),
181 } ISSInfo;
182
183 /* Save the syndrome information for a Data Abort */
184 static void disas_set_da_iss(DisasContext *s, MemOp memop, ISSInfo issinfo)
185 {
186 uint32_t syn;
187 int sas = memop & MO_SIZE;
188 bool sse = memop & MO_SIGN;
189 bool is_acqrel = issinfo & ISSIsAcqRel;
190 bool is_write = issinfo & ISSIsWrite;
191 bool is_16bit = issinfo & ISSIs16Bit;
192 int srt = issinfo & ISSRegMask;
193
194 if (issinfo & ISSInvalid) {
195 /* Some callsites want to conditionally provide ISS info,
196 * eg "only if this was not a writeback"
197 */
198 return;
199 }
200
201 if (srt == 15) {
202 /* For AArch32, insns where the src/dest is R15 never generate
203 * ISS information. Catching that here saves checking at all
204 * the call sites.
205 */
206 return;
207 }
208
209 syn = syn_data_abort_with_iss(0, sas, sse, srt, 0, is_acqrel,
210 0, 0, 0, is_write, 0, is_16bit);
211 disas_set_insn_syndrome(s, syn);
212 }
213
214 static inline int get_a32_user_mem_index(DisasContext *s)
215 {
216 /* Return the core mmu_idx to use for A32/T32 "unprivileged load/store"
217 * insns:
218 * if PL2, UNPREDICTABLE (we choose to implement as if PL0)
219 * otherwise, access as if at PL0.
220 */
221 switch (s->mmu_idx) {
222 case ARMMMUIdx_E2: /* this one is UNPREDICTABLE */
223 case ARMMMUIdx_E10_0:
224 case ARMMMUIdx_E10_1:
225 case ARMMMUIdx_E10_1_PAN:
226 return arm_to_core_mmu_idx(ARMMMUIdx_E10_0);
227 case ARMMMUIdx_SE3:
228 case ARMMMUIdx_SE10_0:
229 case ARMMMUIdx_SE10_1:
230 case ARMMMUIdx_SE10_1_PAN:
231 return arm_to_core_mmu_idx(ARMMMUIdx_SE10_0);
232 case ARMMMUIdx_MUser:
233 case ARMMMUIdx_MPriv:
234 return arm_to_core_mmu_idx(ARMMMUIdx_MUser);
235 case ARMMMUIdx_MUserNegPri:
236 case ARMMMUIdx_MPrivNegPri:
237 return arm_to_core_mmu_idx(ARMMMUIdx_MUserNegPri);
238 case ARMMMUIdx_MSUser:
239 case ARMMMUIdx_MSPriv:
240 return arm_to_core_mmu_idx(ARMMMUIdx_MSUser);
241 case ARMMMUIdx_MSUserNegPri:
242 case ARMMMUIdx_MSPrivNegPri:
243 return arm_to_core_mmu_idx(ARMMMUIdx_MSUserNegPri);
244 default:
245 g_assert_not_reached();
246 }
247 }
248
249 /* The architectural value of PC. */
250 static uint32_t read_pc(DisasContext *s)
251 {
252 return s->pc_curr + (s->thumb ? 4 : 8);
253 }
254
255 /* Set a variable to the value of a CPU register. */
256 void load_reg_var(DisasContext *s, TCGv_i32 var, int reg)
257 {
258 if (reg == 15) {
259 tcg_gen_movi_i32(var, read_pc(s));
260 } else {
261 tcg_gen_mov_i32(var, cpu_R[reg]);
262 }
263 }
264
265 /*
266 * Create a new temp, REG + OFS, except PC is ALIGN(PC, 4).
267 * This is used for load/store for which use of PC implies (literal),
268 * or ADD that implies ADR.
269 */
270 TCGv_i32 add_reg_for_lit(DisasContext *s, int reg, int ofs)
271 {
272 TCGv_i32 tmp = tcg_temp_new_i32();
273
274 if (reg == 15) {
275 tcg_gen_movi_i32(tmp, (read_pc(s) & ~3) + ofs);
276 } else {
277 tcg_gen_addi_i32(tmp, cpu_R[reg], ofs);
278 }
279 return tmp;
280 }
281
282 /* Set a CPU register. The source must be a temporary and will be
283 marked as dead. */
284 void store_reg(DisasContext *s, int reg, TCGv_i32 var)
285 {
286 if (reg == 15) {
287 /* In Thumb mode, we must ignore bit 0.
288 * In ARM mode, for ARMv4 and ARMv5, it is UNPREDICTABLE if bits [1:0]
289 * are not 0b00, but for ARMv6 and above, we must ignore bits [1:0].
290 * We choose to ignore [1:0] in ARM mode for all architecture versions.
291 */
292 tcg_gen_andi_i32(var, var, s->thumb ? ~1 : ~3);
293 s->base.is_jmp = DISAS_JUMP;
294 } else if (reg == 13 && arm_dc_feature(s, ARM_FEATURE_M)) {
295 /* For M-profile SP bits [1:0] are always zero */
296 tcg_gen_andi_i32(var, var, ~3);
297 }
298 tcg_gen_mov_i32(cpu_R[reg], var);
299 tcg_temp_free_i32(var);
300 }
301
302 /*
303 * Variant of store_reg which applies v8M stack-limit checks before updating
304 * SP. If the check fails this will result in an exception being taken.
305 * We disable the stack checks for CONFIG_USER_ONLY because we have
306 * no idea what the stack limits should be in that case.
307 * If stack checking is not being done this just acts like store_reg().
308 */
309 static void store_sp_checked(DisasContext *s, TCGv_i32 var)
310 {
311 #ifndef CONFIG_USER_ONLY
312 if (s->v8m_stackcheck) {
313 gen_helper_v8m_stackcheck(cpu_env, var);
314 }
315 #endif
316 store_reg(s, 13, var);
317 }
318
319 /* Value extensions. */
320 #define gen_uxtb(var) tcg_gen_ext8u_i32(var, var)
321 #define gen_uxth(var) tcg_gen_ext16u_i32(var, var)
322 #define gen_sxtb(var) tcg_gen_ext8s_i32(var, var)
323 #define gen_sxth(var) tcg_gen_ext16s_i32(var, var)
324
325 #define gen_sxtb16(var) gen_helper_sxtb16(var, var)
326 #define gen_uxtb16(var) gen_helper_uxtb16(var, var)
327
328 void gen_set_cpsr(TCGv_i32 var, uint32_t mask)
329 {
330 TCGv_i32 tmp_mask = tcg_const_i32(mask);
331 gen_helper_cpsr_write(cpu_env, var, tmp_mask);
332 tcg_temp_free_i32(tmp_mask);
333 }
334
335 static void gen_exception_internal(int excp)
336 {
337 TCGv_i32 tcg_excp = tcg_const_i32(excp);
338
339 assert(excp_is_internal(excp));
340 gen_helper_exception_internal(cpu_env, tcg_excp);
341 tcg_temp_free_i32(tcg_excp);
342 }
343
344 static void gen_singlestep_exception(DisasContext *s)
345 {
346 /* We just completed step of an insn. Move from Active-not-pending
347 * to Active-pending, and then also take the swstep exception.
348 * This corresponds to making the (IMPDEF) choice to prioritize
349 * swstep exceptions over asynchronous exceptions taken to an exception
350 * level where debug is disabled. This choice has the advantage that
351 * we do not need to maintain internal state corresponding to the
352 * ISV/EX syndrome bits between completion of the step and generation
353 * of the exception, and our syndrome information is always correct.
354 */
355 gen_ss_advance(s);
356 gen_swstep_exception(s, 1, s->is_ldex);
357 s->base.is_jmp = DISAS_NORETURN;
358 }
359
360 void clear_eci_state(DisasContext *s)
361 {
362 /*
363 * Clear any ECI/ICI state: used when a load multiple/store
364 * multiple insn executes.
365 */
366 if (s->eci) {
367 store_cpu_field_constant(0, condexec_bits);
368 s->eci = 0;
369 }
370 }
371
372 static void gen_smul_dual(TCGv_i32 a, TCGv_i32 b)
373 {
374 TCGv_i32 tmp1 = tcg_temp_new_i32();
375 TCGv_i32 tmp2 = tcg_temp_new_i32();
376 tcg_gen_ext16s_i32(tmp1, a);
377 tcg_gen_ext16s_i32(tmp2, b);
378 tcg_gen_mul_i32(tmp1, tmp1, tmp2);
379 tcg_temp_free_i32(tmp2);
380 tcg_gen_sari_i32(a, a, 16);
381 tcg_gen_sari_i32(b, b, 16);
382 tcg_gen_mul_i32(b, b, a);
383 tcg_gen_mov_i32(a, tmp1);
384 tcg_temp_free_i32(tmp1);
385 }
386
387 /* Byteswap each halfword. */
388 void gen_rev16(TCGv_i32 dest, TCGv_i32 var)
389 {
390 TCGv_i32 tmp = tcg_temp_new_i32();
391 TCGv_i32 mask = tcg_constant_i32(0x00ff00ff);
392 tcg_gen_shri_i32(tmp, var, 8);
393 tcg_gen_and_i32(tmp, tmp, mask);
394 tcg_gen_and_i32(var, var, mask);
395 tcg_gen_shli_i32(var, var, 8);
396 tcg_gen_or_i32(dest, var, tmp);
397 tcg_temp_free_i32(tmp);
398 }
399
400 /* Byteswap low halfword and sign extend. */
401 static void gen_revsh(TCGv_i32 dest, TCGv_i32 var)
402 {
403 tcg_gen_bswap16_i32(var, var, TCG_BSWAP_OS);
404 }
405
406 /* Dual 16-bit add. Result placed in t0 and t1 is marked as dead.
407 tmp = (t0 ^ t1) & 0x8000;
408 t0 &= ~0x8000;
409 t1 &= ~0x8000;
410 t0 = (t0 + t1) ^ tmp;
411 */
412
413 static void gen_add16(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
414 {
415 TCGv_i32 tmp = tcg_temp_new_i32();
416 tcg_gen_xor_i32(tmp, t0, t1);
417 tcg_gen_andi_i32(tmp, tmp, 0x8000);
418 tcg_gen_andi_i32(t0, t0, ~0x8000);
419 tcg_gen_andi_i32(t1, t1, ~0x8000);
420 tcg_gen_add_i32(t0, t0, t1);
421 tcg_gen_xor_i32(dest, t0, tmp);
422 tcg_temp_free_i32(tmp);
423 }
424
425 /* Set N and Z flags from var. */
426 static inline void gen_logic_CC(TCGv_i32 var)
427 {
428 tcg_gen_mov_i32(cpu_NF, var);
429 tcg_gen_mov_i32(cpu_ZF, var);
430 }
431
432 /* dest = T0 + T1 + CF. */
433 static void gen_add_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
434 {
435 tcg_gen_add_i32(dest, t0, t1);
436 tcg_gen_add_i32(dest, dest, cpu_CF);
437 }
438
439 /* dest = T0 - T1 + CF - 1. */
440 static void gen_sub_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
441 {
442 tcg_gen_sub_i32(dest, t0, t1);
443 tcg_gen_add_i32(dest, dest, cpu_CF);
444 tcg_gen_subi_i32(dest, dest, 1);
445 }
446
447 /* dest = T0 + T1. Compute C, N, V and Z flags */
448 static void gen_add_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
449 {
450 TCGv_i32 tmp = tcg_temp_new_i32();
451 tcg_gen_movi_i32(tmp, 0);
452 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, t1, tmp);
453 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
454 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
455 tcg_gen_xor_i32(tmp, t0, t1);
456 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
457 tcg_temp_free_i32(tmp);
458 tcg_gen_mov_i32(dest, cpu_NF);
459 }
460
461 /* dest = T0 + T1 + CF. Compute C, N, V and Z flags */
462 static void gen_adc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
463 {
464 TCGv_i32 tmp = tcg_temp_new_i32();
465 if (TCG_TARGET_HAS_add2_i32) {
466 tcg_gen_movi_i32(tmp, 0);
467 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, cpu_CF, tmp);
468 tcg_gen_add2_i32(cpu_NF, cpu_CF, cpu_NF, cpu_CF, t1, tmp);
469 } else {
470 TCGv_i64 q0 = tcg_temp_new_i64();
471 TCGv_i64 q1 = tcg_temp_new_i64();
472 tcg_gen_extu_i32_i64(q0, t0);
473 tcg_gen_extu_i32_i64(q1, t1);
474 tcg_gen_add_i64(q0, q0, q1);
475 tcg_gen_extu_i32_i64(q1, cpu_CF);
476 tcg_gen_add_i64(q0, q0, q1);
477 tcg_gen_extr_i64_i32(cpu_NF, cpu_CF, q0);
478 tcg_temp_free_i64(q0);
479 tcg_temp_free_i64(q1);
480 }
481 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
482 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
483 tcg_gen_xor_i32(tmp, t0, t1);
484 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
485 tcg_temp_free_i32(tmp);
486 tcg_gen_mov_i32(dest, cpu_NF);
487 }
488
489 /* dest = T0 - T1. Compute C, N, V and Z flags */
490 static void gen_sub_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
491 {
492 TCGv_i32 tmp;
493 tcg_gen_sub_i32(cpu_NF, t0, t1);
494 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
495 tcg_gen_setcond_i32(TCG_COND_GEU, cpu_CF, t0, t1);
496 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
497 tmp = tcg_temp_new_i32();
498 tcg_gen_xor_i32(tmp, t0, t1);
499 tcg_gen_and_i32(cpu_VF, cpu_VF, tmp);
500 tcg_temp_free_i32(tmp);
501 tcg_gen_mov_i32(dest, cpu_NF);
502 }
503
504 /* dest = T0 + ~T1 + CF. Compute C, N, V and Z flags */
505 static void gen_sbc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
506 {
507 TCGv_i32 tmp = tcg_temp_new_i32();
508 tcg_gen_not_i32(tmp, t1);
509 gen_adc_CC(dest, t0, tmp);
510 tcg_temp_free_i32(tmp);
511 }
512
513 #define GEN_SHIFT(name) \
514 static void gen_##name(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1) \
515 { \
516 TCGv_i32 tmp1, tmp2, tmp3; \
517 tmp1 = tcg_temp_new_i32(); \
518 tcg_gen_andi_i32(tmp1, t1, 0xff); \
519 tmp2 = tcg_const_i32(0); \
520 tmp3 = tcg_const_i32(0x1f); \
521 tcg_gen_movcond_i32(TCG_COND_GTU, tmp2, tmp1, tmp3, tmp2, t0); \
522 tcg_temp_free_i32(tmp3); \
523 tcg_gen_andi_i32(tmp1, tmp1, 0x1f); \
524 tcg_gen_##name##_i32(dest, tmp2, tmp1); \
525 tcg_temp_free_i32(tmp2); \
526 tcg_temp_free_i32(tmp1); \
527 }
528 GEN_SHIFT(shl)
529 GEN_SHIFT(shr)
530 #undef GEN_SHIFT
531
532 static void gen_sar(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
533 {
534 TCGv_i32 tmp1, tmp2;
535 tmp1 = tcg_temp_new_i32();
536 tcg_gen_andi_i32(tmp1, t1, 0xff);
537 tmp2 = tcg_const_i32(0x1f);
538 tcg_gen_movcond_i32(TCG_COND_GTU, tmp1, tmp1, tmp2, tmp2, tmp1);
539 tcg_temp_free_i32(tmp2);
540 tcg_gen_sar_i32(dest, t0, tmp1);
541 tcg_temp_free_i32(tmp1);
542 }
543
544 static void shifter_out_im(TCGv_i32 var, int shift)
545 {
546 tcg_gen_extract_i32(cpu_CF, var, shift, 1);
547 }
548
549 /* Shift by immediate. Includes special handling for shift == 0. */
550 static inline void gen_arm_shift_im(TCGv_i32 var, int shiftop,
551 int shift, int flags)
552 {
553 switch (shiftop) {
554 case 0: /* LSL */
555 if (shift != 0) {
556 if (flags)
557 shifter_out_im(var, 32 - shift);
558 tcg_gen_shli_i32(var, var, shift);
559 }
560 break;
561 case 1: /* LSR */
562 if (shift == 0) {
563 if (flags) {
564 tcg_gen_shri_i32(cpu_CF, var, 31);
565 }
566 tcg_gen_movi_i32(var, 0);
567 } else {
568 if (flags)
569 shifter_out_im(var, shift - 1);
570 tcg_gen_shri_i32(var, var, shift);
571 }
572 break;
573 case 2: /* ASR */
574 if (shift == 0)
575 shift = 32;
576 if (flags)
577 shifter_out_im(var, shift - 1);
578 if (shift == 32)
579 shift = 31;
580 tcg_gen_sari_i32(var, var, shift);
581 break;
582 case 3: /* ROR/RRX */
583 if (shift != 0) {
584 if (flags)
585 shifter_out_im(var, shift - 1);
586 tcg_gen_rotri_i32(var, var, shift); break;
587 } else {
588 TCGv_i32 tmp = tcg_temp_new_i32();
589 tcg_gen_shli_i32(tmp, cpu_CF, 31);
590 if (flags)
591 shifter_out_im(var, 0);
592 tcg_gen_shri_i32(var, var, 1);
593 tcg_gen_or_i32(var, var, tmp);
594 tcg_temp_free_i32(tmp);
595 }
596 }
597 };
598
599 static inline void gen_arm_shift_reg(TCGv_i32 var, int shiftop,
600 TCGv_i32 shift, int flags)
601 {
602 if (flags) {
603 switch (shiftop) {
604 case 0: gen_helper_shl_cc(var, cpu_env, var, shift); break;
605 case 1: gen_helper_shr_cc(var, cpu_env, var, shift); break;
606 case 2: gen_helper_sar_cc(var, cpu_env, var, shift); break;
607 case 3: gen_helper_ror_cc(var, cpu_env, var, shift); break;
608 }
609 } else {
610 switch (shiftop) {
611 case 0:
612 gen_shl(var, var, shift);
613 break;
614 case 1:
615 gen_shr(var, var, shift);
616 break;
617 case 2:
618 gen_sar(var, var, shift);
619 break;
620 case 3: tcg_gen_andi_i32(shift, shift, 0x1f);
621 tcg_gen_rotr_i32(var, var, shift); break;
622 }
623 }
624 tcg_temp_free_i32(shift);
625 }
626
627 /*
628 * Generate a conditional based on ARM condition code cc.
629 * This is common between ARM and Aarch64 targets.
630 */
631 void arm_test_cc(DisasCompare *cmp, int cc)
632 {
633 TCGv_i32 value;
634 TCGCond cond;
635 bool global = true;
636
637 switch (cc) {
638 case 0: /* eq: Z */
639 case 1: /* ne: !Z */
640 cond = TCG_COND_EQ;
641 value = cpu_ZF;
642 break;
643
644 case 2: /* cs: C */
645 case 3: /* cc: !C */
646 cond = TCG_COND_NE;
647 value = cpu_CF;
648 break;
649
650 case 4: /* mi: N */
651 case 5: /* pl: !N */
652 cond = TCG_COND_LT;
653 value = cpu_NF;
654 break;
655
656 case 6: /* vs: V */
657 case 7: /* vc: !V */
658 cond = TCG_COND_LT;
659 value = cpu_VF;
660 break;
661
662 case 8: /* hi: C && !Z */
663 case 9: /* ls: !C || Z -> !(C && !Z) */
664 cond = TCG_COND_NE;
665 value = tcg_temp_new_i32();
666 global = false;
667 /* CF is 1 for C, so -CF is an all-bits-set mask for C;
668 ZF is non-zero for !Z; so AND the two subexpressions. */
669 tcg_gen_neg_i32(value, cpu_CF);
670 tcg_gen_and_i32(value, value, cpu_ZF);
671 break;
672
673 case 10: /* ge: N == V -> N ^ V == 0 */
674 case 11: /* lt: N != V -> N ^ V != 0 */
675 /* Since we're only interested in the sign bit, == 0 is >= 0. */
676 cond = TCG_COND_GE;
677 value = tcg_temp_new_i32();
678 global = false;
679 tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
680 break;
681
682 case 12: /* gt: !Z && N == V */
683 case 13: /* le: Z || N != V */
684 cond = TCG_COND_NE;
685 value = tcg_temp_new_i32();
686 global = false;
687 /* (N == V) is equal to the sign bit of ~(NF ^ VF). Propagate
688 * the sign bit then AND with ZF to yield the result. */
689 tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
690 tcg_gen_sari_i32(value, value, 31);
691 tcg_gen_andc_i32(value, cpu_ZF, value);
692 break;
693
694 case 14: /* always */
695 case 15: /* always */
696 /* Use the ALWAYS condition, which will fold early.
697 * It doesn't matter what we use for the value. */
698 cond = TCG_COND_ALWAYS;
699 value = cpu_ZF;
700 goto no_invert;
701
702 default:
703 fprintf(stderr, "Bad condition code 0x%x\n", cc);
704 abort();
705 }
706
707 if (cc & 1) {
708 cond = tcg_invert_cond(cond);
709 }
710
711 no_invert:
712 cmp->cond = cond;
713 cmp->value = value;
714 cmp->value_global = global;
715 }
716
717 void arm_free_cc(DisasCompare *cmp)
718 {
719 if (!cmp->value_global) {
720 tcg_temp_free_i32(cmp->value);
721 }
722 }
723
724 void arm_jump_cc(DisasCompare *cmp, TCGLabel *label)
725 {
726 tcg_gen_brcondi_i32(cmp->cond, cmp->value, 0, label);
727 }
728
729 void arm_gen_test_cc(int cc, TCGLabel *label)
730 {
731 DisasCompare cmp;
732 arm_test_cc(&cmp, cc);
733 arm_jump_cc(&cmp, label);
734 arm_free_cc(&cmp);
735 }
736
737 void gen_set_condexec(DisasContext *s)
738 {
739 if (s->condexec_mask) {
740 uint32_t val = (s->condexec_cond << 4) | (s->condexec_mask >> 1);
741
742 store_cpu_field_constant(val, condexec_bits);
743 }
744 }
745
746 void gen_set_pc_im(DisasContext *s, target_ulong val)
747 {
748 tcg_gen_movi_i32(cpu_R[15], val);
749 }
750
751 /* Set PC and Thumb state from var. var is marked as dead. */
752 static inline void gen_bx(DisasContext *s, TCGv_i32 var)
753 {
754 s->base.is_jmp = DISAS_JUMP;
755 tcg_gen_andi_i32(cpu_R[15], var, ~1);
756 tcg_gen_andi_i32(var, var, 1);
757 store_cpu_field(var, thumb);
758 }
759
760 /*
761 * Set PC and Thumb state from var. var is marked as dead.
762 * For M-profile CPUs, include logic to detect exception-return
763 * branches and handle them. This is needed for Thumb POP/LDM to PC, LDR to PC,
764 * and BX reg, and no others, and happens only for code in Handler mode.
765 * The Security Extension also requires us to check for the FNC_RETURN
766 * which signals a function return from non-secure state; this can happen
767 * in both Handler and Thread mode.
768 * To avoid having to do multiple comparisons in inline generated code,
769 * we make the check we do here loose, so it will match for EXC_RETURN
770 * in Thread mode. For system emulation do_v7m_exception_exit() checks
771 * for these spurious cases and returns without doing anything (giving
772 * the same behaviour as for a branch to a non-magic address).
773 *
774 * In linux-user mode it is unclear what the right behaviour for an
775 * attempted FNC_RETURN should be, because in real hardware this will go
776 * directly to Secure code (ie not the Linux kernel) which will then treat
777 * the error in any way it chooses. For QEMU we opt to make the FNC_RETURN
778 * attempt behave the way it would on a CPU without the security extension,
779 * which is to say "like a normal branch". That means we can simply treat
780 * all branches as normal with no magic address behaviour.
781 */
782 static inline void gen_bx_excret(DisasContext *s, TCGv_i32 var)
783 {
784 /* Generate the same code here as for a simple bx, but flag via
785 * s->base.is_jmp that we need to do the rest of the work later.
786 */
787 gen_bx(s, var);
788 #ifndef CONFIG_USER_ONLY
789 if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY) ||
790 (s->v7m_handler_mode && arm_dc_feature(s, ARM_FEATURE_M))) {
791 s->base.is_jmp = DISAS_BX_EXCRET;
792 }
793 #endif
794 }
795
796 static inline void gen_bx_excret_final_code(DisasContext *s)
797 {
798 /* Generate the code to finish possible exception return and end the TB */
799 TCGLabel *excret_label = gen_new_label();
800 uint32_t min_magic;
801
802 if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY)) {
803 /* Covers FNC_RETURN and EXC_RETURN magic */
804 min_magic = FNC_RETURN_MIN_MAGIC;
805 } else {
806 /* EXC_RETURN magic only */
807 min_magic = EXC_RETURN_MIN_MAGIC;
808 }
809
810 /* Is the new PC value in the magic range indicating exception return? */
811 tcg_gen_brcondi_i32(TCG_COND_GEU, cpu_R[15], min_magic, excret_label);
812 /* No: end the TB as we would for a DISAS_JMP */
813 if (s->ss_active) {
814 gen_singlestep_exception(s);
815 } else {
816 tcg_gen_exit_tb(NULL, 0);
817 }
818 gen_set_label(excret_label);
819 /* Yes: this is an exception return.
820 * At this point in runtime env->regs[15] and env->thumb will hold
821 * the exception-return magic number, which do_v7m_exception_exit()
822 * will read. Nothing else will be able to see those values because
823 * the cpu-exec main loop guarantees that we will always go straight
824 * from raising the exception to the exception-handling code.
825 *
826 * gen_ss_advance(s) does nothing on M profile currently but
827 * calling it is conceptually the right thing as we have executed
828 * this instruction (compare SWI, HVC, SMC handling).
829 */
830 gen_ss_advance(s);
831 gen_exception_internal(EXCP_EXCEPTION_EXIT);
832 }
833
834 static inline void gen_bxns(DisasContext *s, int rm)
835 {
836 TCGv_i32 var = load_reg(s, rm);
837
838 /* The bxns helper may raise an EXCEPTION_EXIT exception, so in theory
839 * we need to sync state before calling it, but:
840 * - we don't need to do gen_set_pc_im() because the bxns helper will
841 * always set the PC itself
842 * - we don't need to do gen_set_condexec() because BXNS is UNPREDICTABLE
843 * unless it's outside an IT block or the last insn in an IT block,
844 * so we know that condexec == 0 (already set at the top of the TB)
845 * is correct in the non-UNPREDICTABLE cases, and we can choose
846 * "zeroes the IT bits" as our UNPREDICTABLE behaviour otherwise.
847 */
848 gen_helper_v7m_bxns(cpu_env, var);
849 tcg_temp_free_i32(var);
850 s->base.is_jmp = DISAS_EXIT;
851 }
852
853 static inline void gen_blxns(DisasContext *s, int rm)
854 {
855 TCGv_i32 var = load_reg(s, rm);
856
857 /* We don't need to sync condexec state, for the same reason as bxns.
858 * We do however need to set the PC, because the blxns helper reads it.
859 * The blxns helper may throw an exception.
860 */
861 gen_set_pc_im(s, s->base.pc_next);
862 gen_helper_v7m_blxns(cpu_env, var);
863 tcg_temp_free_i32(var);
864 s->base.is_jmp = DISAS_EXIT;
865 }
866
867 /* Variant of store_reg which uses branch&exchange logic when storing
868 to r15 in ARM architecture v7 and above. The source must be a temporary
869 and will be marked as dead. */
870 static inline void store_reg_bx(DisasContext *s, int reg, TCGv_i32 var)
871 {
872 if (reg == 15 && ENABLE_ARCH_7) {
873 gen_bx(s, var);
874 } else {
875 store_reg(s, reg, var);
876 }
877 }
878
879 /* Variant of store_reg which uses branch&exchange logic when storing
880 * to r15 in ARM architecture v5T and above. This is used for storing
881 * the results of a LDR/LDM/POP into r15, and corresponds to the cases
882 * in the ARM ARM which use the LoadWritePC() pseudocode function. */
883 static inline void store_reg_from_load(DisasContext *s, int reg, TCGv_i32 var)
884 {
885 if (reg == 15 && ENABLE_ARCH_5) {
886 gen_bx_excret(s, var);
887 } else {
888 store_reg(s, reg, var);
889 }
890 }
891
892 #ifdef CONFIG_USER_ONLY
893 #define IS_USER_ONLY 1
894 #else
895 #define IS_USER_ONLY 0
896 #endif
897
898 MemOp pow2_align(unsigned i)
899 {
900 static const MemOp mop_align[] = {
901 0, MO_ALIGN_2, MO_ALIGN_4, MO_ALIGN_8, MO_ALIGN_16,
902 /*
903 * FIXME: TARGET_PAGE_BITS_MIN affects TLB_FLAGS_MASK such
904 * that 256-bit alignment (MO_ALIGN_32) cannot be supported:
905 * see get_alignment_bits(). Enforce only 128-bit alignment for now.
906 */
907 MO_ALIGN_16
908 };
909 g_assert(i < ARRAY_SIZE(mop_align));
910 return mop_align[i];
911 }
912
913 /*
914 * Abstractions of "generate code to do a guest load/store for
915 * AArch32", where a vaddr is always 32 bits (and is zero
916 * extended if we're a 64 bit core) and data is also
917 * 32 bits unless specifically doing a 64 bit access.
918 * These functions work like tcg_gen_qemu_{ld,st}* except
919 * that the address argument is TCGv_i32 rather than TCGv.
920 */
921
922 static TCGv gen_aa32_addr(DisasContext *s, TCGv_i32 a32, MemOp op)
923 {
924 TCGv addr = tcg_temp_new();
925 tcg_gen_extu_i32_tl(addr, a32);
926
927 /* Not needed for user-mode BE32, where we use MO_BE instead. */
928 if (!IS_USER_ONLY && s->sctlr_b && (op & MO_SIZE) < MO_32) {
929 tcg_gen_xori_tl(addr, addr, 4 - (1 << (op & MO_SIZE)));
930 }
931 return addr;
932 }
933
934 /*
935 * Internal routines are used for NEON cases where the endianness
936 * and/or alignment has already been taken into account and manipulated.
937 */
938 void gen_aa32_ld_internal_i32(DisasContext *s, TCGv_i32 val,
939 TCGv_i32 a32, int index, MemOp opc)
940 {
941 TCGv addr = gen_aa32_addr(s, a32, opc);
942 tcg_gen_qemu_ld_i32(val, addr, index, opc);
943 tcg_temp_free(addr);
944 }
945
946 void gen_aa32_st_internal_i32(DisasContext *s, TCGv_i32 val,
947 TCGv_i32 a32, int index, MemOp opc)
948 {
949 TCGv addr = gen_aa32_addr(s, a32, opc);
950 tcg_gen_qemu_st_i32(val, addr, index, opc);
951 tcg_temp_free(addr);
952 }
953
954 void gen_aa32_ld_internal_i64(DisasContext *s, TCGv_i64 val,
955 TCGv_i32 a32, int index, MemOp opc)
956 {
957 TCGv addr = gen_aa32_addr(s, a32, opc);
958
959 tcg_gen_qemu_ld_i64(val, addr, index, opc);
960
961 /* Not needed for user-mode BE32, where we use MO_BE instead. */
962 if (!IS_USER_ONLY && s->sctlr_b && (opc & MO_SIZE) == MO_64) {
963 tcg_gen_rotri_i64(val, val, 32);
964 }
965 tcg_temp_free(addr);
966 }
967
968 void gen_aa32_st_internal_i64(DisasContext *s, TCGv_i64 val,
969 TCGv_i32 a32, int index, MemOp opc)
970 {
971 TCGv addr = gen_aa32_addr(s, a32, opc);
972
973 /* Not needed for user-mode BE32, where we use MO_BE instead. */
974 if (!IS_USER_ONLY && s->sctlr_b && (opc & MO_SIZE) == MO_64) {
975 TCGv_i64 tmp = tcg_temp_new_i64();
976 tcg_gen_rotri_i64(tmp, val, 32);
977 tcg_gen_qemu_st_i64(tmp, addr, index, opc);
978 tcg_temp_free_i64(tmp);
979 } else {
980 tcg_gen_qemu_st_i64(val, addr, index, opc);
981 }
982 tcg_temp_free(addr);
983 }
984
985 void gen_aa32_ld_i32(DisasContext *s, TCGv_i32 val, TCGv_i32 a32,
986 int index, MemOp opc)
987 {
988 gen_aa32_ld_internal_i32(s, val, a32, index, finalize_memop(s, opc));
989 }
990
991 void gen_aa32_st_i32(DisasContext *s, TCGv_i32 val, TCGv_i32 a32,
992 int index, MemOp opc)
993 {
994 gen_aa32_st_internal_i32(s, val, a32, index, finalize_memop(s, opc));
995 }
996
997 void gen_aa32_ld_i64(DisasContext *s, TCGv_i64 val, TCGv_i32 a32,
998 int index, MemOp opc)
999 {
1000 gen_aa32_ld_internal_i64(s, val, a32, index, finalize_memop(s, opc));
1001 }
1002
1003 void gen_aa32_st_i64(DisasContext *s, TCGv_i64 val, TCGv_i32 a32,
1004 int index, MemOp opc)
1005 {
1006 gen_aa32_st_internal_i64(s, val, a32, index, finalize_memop(s, opc));
1007 }
1008
1009 #define DO_GEN_LD(SUFF, OPC) \
1010 static inline void gen_aa32_ld##SUFF(DisasContext *s, TCGv_i32 val, \
1011 TCGv_i32 a32, int index) \
1012 { \
1013 gen_aa32_ld_i32(s, val, a32, index, OPC); \
1014 }
1015
1016 #define DO_GEN_ST(SUFF, OPC) \
1017 static inline void gen_aa32_st##SUFF(DisasContext *s, TCGv_i32 val, \
1018 TCGv_i32 a32, int index) \
1019 { \
1020 gen_aa32_st_i32(s, val, a32, index, OPC); \
1021 }
1022
1023 static inline void gen_hvc(DisasContext *s, int imm16)
1024 {
1025 /* The pre HVC helper handles cases when HVC gets trapped
1026 * as an undefined insn by runtime configuration (ie before
1027 * the insn really executes).
1028 */
1029 gen_set_pc_im(s, s->pc_curr);
1030 gen_helper_pre_hvc(cpu_env);
1031 /* Otherwise we will treat this as a real exception which
1032 * happens after execution of the insn. (The distinction matters
1033 * for the PC value reported to the exception handler and also
1034 * for single stepping.)
1035 */
1036 s->svc_imm = imm16;
1037 gen_set_pc_im(s, s->base.pc_next);
1038 s->base.is_jmp = DISAS_HVC;
1039 }
1040
1041 static inline void gen_smc(DisasContext *s)
1042 {
1043 /* As with HVC, we may take an exception either before or after
1044 * the insn executes.
1045 */
1046 TCGv_i32 tmp;
1047
1048 gen_set_pc_im(s, s->pc_curr);
1049 tmp = tcg_const_i32(syn_aa32_smc());
1050 gen_helper_pre_smc(cpu_env, tmp);
1051 tcg_temp_free_i32(tmp);
1052 gen_set_pc_im(s, s->base.pc_next);
1053 s->base.is_jmp = DISAS_SMC;
1054 }
1055
1056 static void gen_exception_internal_insn(DisasContext *s, uint32_t pc, int excp)
1057 {
1058 gen_set_condexec(s);
1059 gen_set_pc_im(s, pc);
1060 gen_exception_internal(excp);
1061 s->base.is_jmp = DISAS_NORETURN;
1062 }
1063
1064 void gen_exception_insn(DisasContext *s, uint64_t pc, int excp,
1065 uint32_t syn, uint32_t target_el)
1066 {
1067 if (s->aarch64) {
1068 gen_a64_set_pc_im(pc);
1069 } else {
1070 gen_set_condexec(s);
1071 gen_set_pc_im(s, pc);
1072 }
1073 gen_exception(excp, syn, target_el);
1074 s->base.is_jmp = DISAS_NORETURN;
1075 }
1076
1077 static void gen_exception_bkpt_insn(DisasContext *s, uint32_t syn)
1078 {
1079 TCGv_i32 tcg_syn;
1080
1081 gen_set_condexec(s);
1082 gen_set_pc_im(s, s->pc_curr);
1083 tcg_syn = tcg_const_i32(syn);
1084 gen_helper_exception_bkpt_insn(cpu_env, tcg_syn);
1085 tcg_temp_free_i32(tcg_syn);
1086 s->base.is_jmp = DISAS_NORETURN;
1087 }
1088
1089 void unallocated_encoding(DisasContext *s)
1090 {
1091 /* Unallocated and reserved encodings are uncategorized */
1092 gen_exception_insn(s, s->pc_curr, EXCP_UDEF, syn_uncategorized(),
1093 default_exception_el(s));
1094 }
1095
1096 static void gen_exception_el(DisasContext *s, int excp, uint32_t syn,
1097 TCGv_i32 tcg_el)
1098 {
1099 TCGv_i32 tcg_excp;
1100 TCGv_i32 tcg_syn;
1101
1102 gen_set_condexec(s);
1103 gen_set_pc_im(s, s->pc_curr);
1104 tcg_excp = tcg_const_i32(excp);
1105 tcg_syn = tcg_const_i32(syn);
1106 gen_helper_exception_with_syndrome(cpu_env, tcg_excp, tcg_syn, tcg_el);
1107 tcg_temp_free_i32(tcg_syn);
1108 tcg_temp_free_i32(tcg_excp);
1109 s->base.is_jmp = DISAS_NORETURN;
1110 }
1111
1112 /* Force a TB lookup after an instruction that changes the CPU state. */
1113 void gen_lookup_tb(DisasContext *s)
1114 {
1115 tcg_gen_movi_i32(cpu_R[15], s->base.pc_next);
1116 s->base.is_jmp = DISAS_EXIT;
1117 }
1118
1119 static inline void gen_hlt(DisasContext *s, int imm)
1120 {
1121 /* HLT. This has two purposes.
1122 * Architecturally, it is an external halting debug instruction.
1123 * Since QEMU doesn't implement external debug, we treat this as
1124 * it is required for halting debug disabled: it will UNDEF.
1125 * Secondly, "HLT 0x3C" is a T32 semihosting trap instruction,
1126 * and "HLT 0xF000" is an A32 semihosting syscall. These traps
1127 * must trigger semihosting even for ARMv7 and earlier, where
1128 * HLT was an undefined encoding.
1129 * In system mode, we don't allow userspace access to
1130 * semihosting, to provide some semblance of security
1131 * (and for consistency with our 32-bit semihosting).
1132 */
1133 if (semihosting_enabled() &&
1134 #ifndef CONFIG_USER_ONLY
1135 s->current_el != 0 &&
1136 #endif
1137 (imm == (s->thumb ? 0x3c : 0xf000))) {
1138 gen_exception_internal_insn(s, s->pc_curr, EXCP_SEMIHOST);
1139 return;
1140 }
1141
1142 unallocated_encoding(s);
1143 }
1144
1145 /*
1146 * Return the offset of a "full" NEON Dreg.
1147 */
1148 long neon_full_reg_offset(unsigned reg)
1149 {
1150 return offsetof(CPUARMState, vfp.zregs[reg >> 1].d[reg & 1]);
1151 }
1152
1153 /*
1154 * Return the offset of a 2**SIZE piece of a NEON register, at index ELE,
1155 * where 0 is the least significant end of the register.
1156 */
1157 long neon_element_offset(int reg, int element, MemOp memop)
1158 {
1159 int element_size = 1 << (memop & MO_SIZE);
1160 int ofs = element * element_size;
1161 #ifdef HOST_WORDS_BIGENDIAN
1162 /*
1163 * Calculate the offset assuming fully little-endian,
1164 * then XOR to account for the order of the 8-byte units.
1165 */
1166 if (element_size < 8) {
1167 ofs ^= 8 - element_size;
1168 }
1169 #endif
1170 return neon_full_reg_offset(reg) + ofs;
1171 }
1172
1173 /* Return the offset of a VFP Dreg (dp = true) or VFP Sreg (dp = false). */
1174 long vfp_reg_offset(bool dp, unsigned reg)
1175 {
1176 if (dp) {
1177 return neon_element_offset(reg, 0, MO_64);
1178 } else {
1179 return neon_element_offset(reg >> 1, reg & 1, MO_32);
1180 }
1181 }
1182
1183 void read_neon_element32(TCGv_i32 dest, int reg, int ele, MemOp memop)
1184 {
1185 long off = neon_element_offset(reg, ele, memop);
1186
1187 switch (memop) {
1188 case MO_SB:
1189 tcg_gen_ld8s_i32(dest, cpu_env, off);
1190 break;
1191 case MO_UB:
1192 tcg_gen_ld8u_i32(dest, cpu_env, off);
1193 break;
1194 case MO_SW:
1195 tcg_gen_ld16s_i32(dest, cpu_env, off);
1196 break;
1197 case MO_UW:
1198 tcg_gen_ld16u_i32(dest, cpu_env, off);
1199 break;
1200 case MO_UL:
1201 case MO_SL:
1202 tcg_gen_ld_i32(dest, cpu_env, off);
1203 break;
1204 default:
1205 g_assert_not_reached();
1206 }
1207 }
1208
1209 void read_neon_element64(TCGv_i64 dest, int reg, int ele, MemOp memop)
1210 {
1211 long off = neon_element_offset(reg, ele, memop);
1212
1213 switch (memop) {
1214 case MO_SL:
1215 tcg_gen_ld32s_i64(dest, cpu_env, off);
1216 break;
1217 case MO_UL:
1218 tcg_gen_ld32u_i64(dest, cpu_env, off);
1219 break;
1220 case MO_UQ:
1221 tcg_gen_ld_i64(dest, cpu_env, off);
1222 break;
1223 default:
1224 g_assert_not_reached();
1225 }
1226 }
1227
1228 void write_neon_element32(TCGv_i32 src, int reg, int ele, MemOp memop)
1229 {
1230 long off = neon_element_offset(reg, ele, memop);
1231
1232 switch (memop) {
1233 case MO_8:
1234 tcg_gen_st8_i32(src, cpu_env, off);
1235 break;
1236 case MO_16:
1237 tcg_gen_st16_i32(src, cpu_env, off);
1238 break;
1239 case MO_32:
1240 tcg_gen_st_i32(src, cpu_env, off);
1241 break;
1242 default:
1243 g_assert_not_reached();
1244 }
1245 }
1246
1247 void write_neon_element64(TCGv_i64 src, int reg, int ele, MemOp memop)
1248 {
1249 long off = neon_element_offset(reg, ele, memop);
1250
1251 switch (memop) {
1252 case MO_32:
1253 tcg_gen_st32_i64(src, cpu_env, off);
1254 break;
1255 case MO_64:
1256 tcg_gen_st_i64(src, cpu_env, off);
1257 break;
1258 default:
1259 g_assert_not_reached();
1260 }
1261 }
1262
1263 #define ARM_CP_RW_BIT (1 << 20)
1264
1265 static inline void iwmmxt_load_reg(TCGv_i64 var, int reg)
1266 {
1267 tcg_gen_ld_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1268 }
1269
1270 static inline void iwmmxt_store_reg(TCGv_i64 var, int reg)
1271 {
1272 tcg_gen_st_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1273 }
1274
1275 static inline TCGv_i32 iwmmxt_load_creg(int reg)
1276 {
1277 TCGv_i32 var = tcg_temp_new_i32();
1278 tcg_gen_ld_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1279 return var;
1280 }
1281
1282 static inline void iwmmxt_store_creg(int reg, TCGv_i32 var)
1283 {
1284 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1285 tcg_temp_free_i32(var);
1286 }
1287
1288 static inline void gen_op_iwmmxt_movq_wRn_M0(int rn)
1289 {
1290 iwmmxt_store_reg(cpu_M0, rn);
1291 }
1292
1293 static inline void gen_op_iwmmxt_movq_M0_wRn(int rn)
1294 {
1295 iwmmxt_load_reg(cpu_M0, rn);
1296 }
1297
1298 static inline void gen_op_iwmmxt_orq_M0_wRn(int rn)
1299 {
1300 iwmmxt_load_reg(cpu_V1, rn);
1301 tcg_gen_or_i64(cpu_M0, cpu_M0, cpu_V1);
1302 }
1303
1304 static inline void gen_op_iwmmxt_andq_M0_wRn(int rn)
1305 {
1306 iwmmxt_load_reg(cpu_V1, rn);
1307 tcg_gen_and_i64(cpu_M0, cpu_M0, cpu_V1);
1308 }
1309
1310 static inline void gen_op_iwmmxt_xorq_M0_wRn(int rn)
1311 {
1312 iwmmxt_load_reg(cpu_V1, rn);
1313 tcg_gen_xor_i64(cpu_M0, cpu_M0, cpu_V1);
1314 }
1315
1316 #define IWMMXT_OP(name) \
1317 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1318 { \
1319 iwmmxt_load_reg(cpu_V1, rn); \
1320 gen_helper_iwmmxt_##name(cpu_M0, cpu_M0, cpu_V1); \
1321 }
1322
1323 #define IWMMXT_OP_ENV(name) \
1324 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1325 { \
1326 iwmmxt_load_reg(cpu_V1, rn); \
1327 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0, cpu_V1); \
1328 }
1329
1330 #define IWMMXT_OP_ENV_SIZE(name) \
1331 IWMMXT_OP_ENV(name##b) \
1332 IWMMXT_OP_ENV(name##w) \
1333 IWMMXT_OP_ENV(name##l)
1334
1335 #define IWMMXT_OP_ENV1(name) \
1336 static inline void gen_op_iwmmxt_##name##_M0(void) \
1337 { \
1338 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0); \
1339 }
1340
1341 IWMMXT_OP(maddsq)
1342 IWMMXT_OP(madduq)
1343 IWMMXT_OP(sadb)
1344 IWMMXT_OP(sadw)
1345 IWMMXT_OP(mulslw)
1346 IWMMXT_OP(mulshw)
1347 IWMMXT_OP(mululw)
1348 IWMMXT_OP(muluhw)
1349 IWMMXT_OP(macsw)
1350 IWMMXT_OP(macuw)
1351
1352 IWMMXT_OP_ENV_SIZE(unpackl)
1353 IWMMXT_OP_ENV_SIZE(unpackh)
1354
1355 IWMMXT_OP_ENV1(unpacklub)
1356 IWMMXT_OP_ENV1(unpackluw)
1357 IWMMXT_OP_ENV1(unpacklul)
1358 IWMMXT_OP_ENV1(unpackhub)
1359 IWMMXT_OP_ENV1(unpackhuw)
1360 IWMMXT_OP_ENV1(unpackhul)
1361 IWMMXT_OP_ENV1(unpacklsb)
1362 IWMMXT_OP_ENV1(unpacklsw)
1363 IWMMXT_OP_ENV1(unpacklsl)
1364 IWMMXT_OP_ENV1(unpackhsb)
1365 IWMMXT_OP_ENV1(unpackhsw)
1366 IWMMXT_OP_ENV1(unpackhsl)
1367
1368 IWMMXT_OP_ENV_SIZE(cmpeq)
1369 IWMMXT_OP_ENV_SIZE(cmpgtu)
1370 IWMMXT_OP_ENV_SIZE(cmpgts)
1371
1372 IWMMXT_OP_ENV_SIZE(mins)
1373 IWMMXT_OP_ENV_SIZE(minu)
1374 IWMMXT_OP_ENV_SIZE(maxs)
1375 IWMMXT_OP_ENV_SIZE(maxu)
1376
1377 IWMMXT_OP_ENV_SIZE(subn)
1378 IWMMXT_OP_ENV_SIZE(addn)
1379 IWMMXT_OP_ENV_SIZE(subu)
1380 IWMMXT_OP_ENV_SIZE(addu)
1381 IWMMXT_OP_ENV_SIZE(subs)
1382 IWMMXT_OP_ENV_SIZE(adds)
1383
1384 IWMMXT_OP_ENV(avgb0)
1385 IWMMXT_OP_ENV(avgb1)
1386 IWMMXT_OP_ENV(avgw0)
1387 IWMMXT_OP_ENV(avgw1)
1388
1389 IWMMXT_OP_ENV(packuw)
1390 IWMMXT_OP_ENV(packul)
1391 IWMMXT_OP_ENV(packuq)
1392 IWMMXT_OP_ENV(packsw)
1393 IWMMXT_OP_ENV(packsl)
1394 IWMMXT_OP_ENV(packsq)
1395
1396 static void gen_op_iwmmxt_set_mup(void)
1397 {
1398 TCGv_i32 tmp;
1399 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1400 tcg_gen_ori_i32(tmp, tmp, 2);
1401 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1402 }
1403
1404 static void gen_op_iwmmxt_set_cup(void)
1405 {
1406 TCGv_i32 tmp;
1407 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1408 tcg_gen_ori_i32(tmp, tmp, 1);
1409 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1410 }
1411
1412 static void gen_op_iwmmxt_setpsr_nz(void)
1413 {
1414 TCGv_i32 tmp = tcg_temp_new_i32();
1415 gen_helper_iwmmxt_setpsr_nz(tmp, cpu_M0);
1416 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCASF]);
1417 }
1418
1419 static inline void gen_op_iwmmxt_addl_M0_wRn(int rn)
1420 {
1421 iwmmxt_load_reg(cpu_V1, rn);
1422 tcg_gen_ext32u_i64(cpu_V1, cpu_V1);
1423 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1424 }
1425
1426 static inline int gen_iwmmxt_address(DisasContext *s, uint32_t insn,
1427 TCGv_i32 dest)
1428 {
1429 int rd;
1430 uint32_t offset;
1431 TCGv_i32 tmp;
1432
1433 rd = (insn >> 16) & 0xf;
1434 tmp = load_reg(s, rd);
1435
1436 offset = (insn & 0xff) << ((insn >> 7) & 2);
1437 if (insn & (1 << 24)) {
1438 /* Pre indexed */
1439 if (insn & (1 << 23))
1440 tcg_gen_addi_i32(tmp, tmp, offset);
1441 else
1442 tcg_gen_addi_i32(tmp, tmp, -offset);
1443 tcg_gen_mov_i32(dest, tmp);
1444 if (insn & (1 << 21))
1445 store_reg(s, rd, tmp);
1446 else
1447 tcg_temp_free_i32(tmp);
1448 } else if (insn & (1 << 21)) {
1449 /* Post indexed */
1450 tcg_gen_mov_i32(dest, tmp);
1451 if (insn & (1 << 23))
1452 tcg_gen_addi_i32(tmp, tmp, offset);
1453 else
1454 tcg_gen_addi_i32(tmp, tmp, -offset);
1455 store_reg(s, rd, tmp);
1456 } else if (!(insn & (1 << 23)))
1457 return 1;
1458 return 0;
1459 }
1460
1461 static inline int gen_iwmmxt_shift(uint32_t insn, uint32_t mask, TCGv_i32 dest)
1462 {
1463 int rd = (insn >> 0) & 0xf;
1464 TCGv_i32 tmp;
1465
1466 if (insn & (1 << 8)) {
1467 if (rd < ARM_IWMMXT_wCGR0 || rd > ARM_IWMMXT_wCGR3) {
1468 return 1;
1469 } else {
1470 tmp = iwmmxt_load_creg(rd);
1471 }
1472 } else {
1473 tmp = tcg_temp_new_i32();
1474 iwmmxt_load_reg(cpu_V0, rd);
1475 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
1476 }
1477 tcg_gen_andi_i32(tmp, tmp, mask);
1478 tcg_gen_mov_i32(dest, tmp);
1479 tcg_temp_free_i32(tmp);
1480 return 0;
1481 }
1482
1483 /* Disassemble an iwMMXt instruction. Returns nonzero if an error occurred
1484 (ie. an undefined instruction). */
1485 static int disas_iwmmxt_insn(DisasContext *s, uint32_t insn)
1486 {
1487 int rd, wrd;
1488 int rdhi, rdlo, rd0, rd1, i;
1489 TCGv_i32 addr;
1490 TCGv_i32 tmp, tmp2, tmp3;
1491
1492 if ((insn & 0x0e000e00) == 0x0c000000) {
1493 if ((insn & 0x0fe00ff0) == 0x0c400000) {
1494 wrd = insn & 0xf;
1495 rdlo = (insn >> 12) & 0xf;
1496 rdhi = (insn >> 16) & 0xf;
1497 if (insn & ARM_CP_RW_BIT) { /* TMRRC */
1498 iwmmxt_load_reg(cpu_V0, wrd);
1499 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
1500 tcg_gen_extrh_i64_i32(cpu_R[rdhi], cpu_V0);
1501 } else { /* TMCRR */
1502 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
1503 iwmmxt_store_reg(cpu_V0, wrd);
1504 gen_op_iwmmxt_set_mup();
1505 }
1506 return 0;
1507 }
1508
1509 wrd = (insn >> 12) & 0xf;
1510 addr = tcg_temp_new_i32();
1511 if (gen_iwmmxt_address(s, insn, addr)) {
1512 tcg_temp_free_i32(addr);
1513 return 1;
1514 }
1515 if (insn & ARM_CP_RW_BIT) {
1516 if ((insn >> 28) == 0xf) { /* WLDRW wCx */
1517 tmp = tcg_temp_new_i32();
1518 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1519 iwmmxt_store_creg(wrd, tmp);
1520 } else {
1521 i = 1;
1522 if (insn & (1 << 8)) {
1523 if (insn & (1 << 22)) { /* WLDRD */
1524 gen_aa32_ld64(s, cpu_M0, addr, get_mem_index(s));
1525 i = 0;
1526 } else { /* WLDRW wRd */
1527 tmp = tcg_temp_new_i32();
1528 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1529 }
1530 } else {
1531 tmp = tcg_temp_new_i32();
1532 if (insn & (1 << 22)) { /* WLDRH */
1533 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
1534 } else { /* WLDRB */
1535 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
1536 }
1537 }
1538 if (i) {
1539 tcg_gen_extu_i32_i64(cpu_M0, tmp);
1540 tcg_temp_free_i32(tmp);
1541 }
1542 gen_op_iwmmxt_movq_wRn_M0(wrd);
1543 }
1544 } else {
1545 if ((insn >> 28) == 0xf) { /* WSTRW wCx */
1546 tmp = iwmmxt_load_creg(wrd);
1547 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1548 } else {
1549 gen_op_iwmmxt_movq_M0_wRn(wrd);
1550 tmp = tcg_temp_new_i32();
1551 if (insn & (1 << 8)) {
1552 if (insn & (1 << 22)) { /* WSTRD */
1553 gen_aa32_st64(s, cpu_M0, addr, get_mem_index(s));
1554 } else { /* WSTRW wRd */
1555 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1556 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1557 }
1558 } else {
1559 if (insn & (1 << 22)) { /* WSTRH */
1560 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1561 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
1562 } else { /* WSTRB */
1563 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1564 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
1565 }
1566 }
1567 }
1568 tcg_temp_free_i32(tmp);
1569 }
1570 tcg_temp_free_i32(addr);
1571 return 0;
1572 }
1573
1574 if ((insn & 0x0f000000) != 0x0e000000)
1575 return 1;
1576
1577 switch (((insn >> 12) & 0xf00) | ((insn >> 4) & 0xff)) {
1578 case 0x000: /* WOR */
1579 wrd = (insn >> 12) & 0xf;
1580 rd0 = (insn >> 0) & 0xf;
1581 rd1 = (insn >> 16) & 0xf;
1582 gen_op_iwmmxt_movq_M0_wRn(rd0);
1583 gen_op_iwmmxt_orq_M0_wRn(rd1);
1584 gen_op_iwmmxt_setpsr_nz();
1585 gen_op_iwmmxt_movq_wRn_M0(wrd);
1586 gen_op_iwmmxt_set_mup();
1587 gen_op_iwmmxt_set_cup();
1588 break;
1589 case 0x011: /* TMCR */
1590 if (insn & 0xf)
1591 return 1;
1592 rd = (insn >> 12) & 0xf;
1593 wrd = (insn >> 16) & 0xf;
1594 switch (wrd) {
1595 case ARM_IWMMXT_wCID:
1596 case ARM_IWMMXT_wCASF:
1597 break;
1598 case ARM_IWMMXT_wCon:
1599 gen_op_iwmmxt_set_cup();
1600 /* Fall through. */
1601 case ARM_IWMMXT_wCSSF:
1602 tmp = iwmmxt_load_creg(wrd);
1603 tmp2 = load_reg(s, rd);
1604 tcg_gen_andc_i32(tmp, tmp, tmp2);
1605 tcg_temp_free_i32(tmp2);
1606 iwmmxt_store_creg(wrd, tmp);
1607 break;
1608 case ARM_IWMMXT_wCGR0:
1609 case ARM_IWMMXT_wCGR1:
1610 case ARM_IWMMXT_wCGR2:
1611 case ARM_IWMMXT_wCGR3:
1612 gen_op_iwmmxt_set_cup();
1613 tmp = load_reg(s, rd);
1614 iwmmxt_store_creg(wrd, tmp);
1615 break;
1616 default:
1617 return 1;
1618 }
1619 break;
1620 case 0x100: /* WXOR */
1621 wrd = (insn >> 12) & 0xf;
1622 rd0 = (insn >> 0) & 0xf;
1623 rd1 = (insn >> 16) & 0xf;
1624 gen_op_iwmmxt_movq_M0_wRn(rd0);
1625 gen_op_iwmmxt_xorq_M0_wRn(rd1);
1626 gen_op_iwmmxt_setpsr_nz();
1627 gen_op_iwmmxt_movq_wRn_M0(wrd);
1628 gen_op_iwmmxt_set_mup();
1629 gen_op_iwmmxt_set_cup();
1630 break;
1631 case 0x111: /* TMRC */
1632 if (insn & 0xf)
1633 return 1;
1634 rd = (insn >> 12) & 0xf;
1635 wrd = (insn >> 16) & 0xf;
1636 tmp = iwmmxt_load_creg(wrd);
1637 store_reg(s, rd, tmp);
1638 break;
1639 case 0x300: /* WANDN */
1640 wrd = (insn >> 12) & 0xf;
1641 rd0 = (insn >> 0) & 0xf;
1642 rd1 = (insn >> 16) & 0xf;
1643 gen_op_iwmmxt_movq_M0_wRn(rd0);
1644 tcg_gen_neg_i64(cpu_M0, cpu_M0);
1645 gen_op_iwmmxt_andq_M0_wRn(rd1);
1646 gen_op_iwmmxt_setpsr_nz();
1647 gen_op_iwmmxt_movq_wRn_M0(wrd);
1648 gen_op_iwmmxt_set_mup();
1649 gen_op_iwmmxt_set_cup();
1650 break;
1651 case 0x200: /* WAND */
1652 wrd = (insn >> 12) & 0xf;
1653 rd0 = (insn >> 0) & 0xf;
1654 rd1 = (insn >> 16) & 0xf;
1655 gen_op_iwmmxt_movq_M0_wRn(rd0);
1656 gen_op_iwmmxt_andq_M0_wRn(rd1);
1657 gen_op_iwmmxt_setpsr_nz();
1658 gen_op_iwmmxt_movq_wRn_M0(wrd);
1659 gen_op_iwmmxt_set_mup();
1660 gen_op_iwmmxt_set_cup();
1661 break;
1662 case 0x810: case 0xa10: /* WMADD */
1663 wrd = (insn >> 12) & 0xf;
1664 rd0 = (insn >> 0) & 0xf;
1665 rd1 = (insn >> 16) & 0xf;
1666 gen_op_iwmmxt_movq_M0_wRn(rd0);
1667 if (insn & (1 << 21))
1668 gen_op_iwmmxt_maddsq_M0_wRn(rd1);
1669 else
1670 gen_op_iwmmxt_madduq_M0_wRn(rd1);
1671 gen_op_iwmmxt_movq_wRn_M0(wrd);
1672 gen_op_iwmmxt_set_mup();
1673 break;
1674 case 0x10e: case 0x50e: case 0x90e: case 0xd0e: /* WUNPCKIL */
1675 wrd = (insn >> 12) & 0xf;
1676 rd0 = (insn >> 16) & 0xf;
1677 rd1 = (insn >> 0) & 0xf;
1678 gen_op_iwmmxt_movq_M0_wRn(rd0);
1679 switch ((insn >> 22) & 3) {
1680 case 0:
1681 gen_op_iwmmxt_unpacklb_M0_wRn(rd1);
1682 break;
1683 case 1:
1684 gen_op_iwmmxt_unpacklw_M0_wRn(rd1);
1685 break;
1686 case 2:
1687 gen_op_iwmmxt_unpackll_M0_wRn(rd1);
1688 break;
1689 case 3:
1690 return 1;
1691 }
1692 gen_op_iwmmxt_movq_wRn_M0(wrd);
1693 gen_op_iwmmxt_set_mup();
1694 gen_op_iwmmxt_set_cup();
1695 break;
1696 case 0x10c: case 0x50c: case 0x90c: case 0xd0c: /* WUNPCKIH */
1697 wrd = (insn >> 12) & 0xf;
1698 rd0 = (insn >> 16) & 0xf;
1699 rd1 = (insn >> 0) & 0xf;
1700 gen_op_iwmmxt_movq_M0_wRn(rd0);
1701 switch ((insn >> 22) & 3) {
1702 case 0:
1703 gen_op_iwmmxt_unpackhb_M0_wRn(rd1);
1704 break;
1705 case 1:
1706 gen_op_iwmmxt_unpackhw_M0_wRn(rd1);
1707 break;
1708 case 2:
1709 gen_op_iwmmxt_unpackhl_M0_wRn(rd1);
1710 break;
1711 case 3:
1712 return 1;
1713 }
1714 gen_op_iwmmxt_movq_wRn_M0(wrd);
1715 gen_op_iwmmxt_set_mup();
1716 gen_op_iwmmxt_set_cup();
1717 break;
1718 case 0x012: case 0x112: case 0x412: case 0x512: /* WSAD */
1719 wrd = (insn >> 12) & 0xf;
1720 rd0 = (insn >> 16) & 0xf;
1721 rd1 = (insn >> 0) & 0xf;
1722 gen_op_iwmmxt_movq_M0_wRn(rd0);
1723 if (insn & (1 << 22))
1724 gen_op_iwmmxt_sadw_M0_wRn(rd1);
1725 else
1726 gen_op_iwmmxt_sadb_M0_wRn(rd1);
1727 if (!(insn & (1 << 20)))
1728 gen_op_iwmmxt_addl_M0_wRn(wrd);
1729 gen_op_iwmmxt_movq_wRn_M0(wrd);
1730 gen_op_iwmmxt_set_mup();
1731 break;
1732 case 0x010: case 0x110: case 0x210: case 0x310: /* WMUL */
1733 wrd = (insn >> 12) & 0xf;
1734 rd0 = (insn >> 16) & 0xf;
1735 rd1 = (insn >> 0) & 0xf;
1736 gen_op_iwmmxt_movq_M0_wRn(rd0);
1737 if (insn & (1 << 21)) {
1738 if (insn & (1 << 20))
1739 gen_op_iwmmxt_mulshw_M0_wRn(rd1);
1740 else
1741 gen_op_iwmmxt_mulslw_M0_wRn(rd1);
1742 } else {
1743 if (insn & (1 << 20))
1744 gen_op_iwmmxt_muluhw_M0_wRn(rd1);
1745 else
1746 gen_op_iwmmxt_mululw_M0_wRn(rd1);
1747 }
1748 gen_op_iwmmxt_movq_wRn_M0(wrd);
1749 gen_op_iwmmxt_set_mup();
1750 break;
1751 case 0x410: case 0x510: case 0x610: case 0x710: /* WMAC */
1752 wrd = (insn >> 12) & 0xf;
1753 rd0 = (insn >> 16) & 0xf;
1754 rd1 = (insn >> 0) & 0xf;
1755 gen_op_iwmmxt_movq_M0_wRn(rd0);
1756 if (insn & (1 << 21))
1757 gen_op_iwmmxt_macsw_M0_wRn(rd1);
1758 else
1759 gen_op_iwmmxt_macuw_M0_wRn(rd1);
1760 if (!(insn & (1 << 20))) {
1761 iwmmxt_load_reg(cpu_V1, wrd);
1762 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1763 }
1764 gen_op_iwmmxt_movq_wRn_M0(wrd);
1765 gen_op_iwmmxt_set_mup();
1766 break;
1767 case 0x006: case 0x406: case 0x806: case 0xc06: /* WCMPEQ */
1768 wrd = (insn >> 12) & 0xf;
1769 rd0 = (insn >> 16) & 0xf;
1770 rd1 = (insn >> 0) & 0xf;
1771 gen_op_iwmmxt_movq_M0_wRn(rd0);
1772 switch ((insn >> 22) & 3) {
1773 case 0:
1774 gen_op_iwmmxt_cmpeqb_M0_wRn(rd1);
1775 break;
1776 case 1:
1777 gen_op_iwmmxt_cmpeqw_M0_wRn(rd1);
1778 break;
1779 case 2:
1780 gen_op_iwmmxt_cmpeql_M0_wRn(rd1);
1781 break;
1782 case 3:
1783 return 1;
1784 }
1785 gen_op_iwmmxt_movq_wRn_M0(wrd);
1786 gen_op_iwmmxt_set_mup();
1787 gen_op_iwmmxt_set_cup();
1788 break;
1789 case 0x800: case 0x900: case 0xc00: case 0xd00: /* WAVG2 */
1790 wrd = (insn >> 12) & 0xf;
1791 rd0 = (insn >> 16) & 0xf;
1792 rd1 = (insn >> 0) & 0xf;
1793 gen_op_iwmmxt_movq_M0_wRn(rd0);
1794 if (insn & (1 << 22)) {
1795 if (insn & (1 << 20))
1796 gen_op_iwmmxt_avgw1_M0_wRn(rd1);
1797 else
1798 gen_op_iwmmxt_avgw0_M0_wRn(rd1);
1799 } else {
1800 if (insn & (1 << 20))
1801 gen_op_iwmmxt_avgb1_M0_wRn(rd1);
1802 else
1803 gen_op_iwmmxt_avgb0_M0_wRn(rd1);
1804 }
1805 gen_op_iwmmxt_movq_wRn_M0(wrd);
1806 gen_op_iwmmxt_set_mup();
1807 gen_op_iwmmxt_set_cup();
1808 break;
1809 case 0x802: case 0x902: case 0xa02: case 0xb02: /* WALIGNR */
1810 wrd = (insn >> 12) & 0xf;
1811 rd0 = (insn >> 16) & 0xf;
1812 rd1 = (insn >> 0) & 0xf;
1813 gen_op_iwmmxt_movq_M0_wRn(rd0);
1814 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCGR0 + ((insn >> 20) & 3));
1815 tcg_gen_andi_i32(tmp, tmp, 7);
1816 iwmmxt_load_reg(cpu_V1, rd1);
1817 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
1818 tcg_temp_free_i32(tmp);
1819 gen_op_iwmmxt_movq_wRn_M0(wrd);
1820 gen_op_iwmmxt_set_mup();
1821 break;
1822 case 0x601: case 0x605: case 0x609: case 0x60d: /* TINSR */
1823 if (((insn >> 6) & 3) == 3)
1824 return 1;
1825 rd = (insn >> 12) & 0xf;
1826 wrd = (insn >> 16) & 0xf;
1827 tmp = load_reg(s, rd);
1828 gen_op_iwmmxt_movq_M0_wRn(wrd);
1829 switch ((insn >> 6) & 3) {
1830 case 0:
1831 tmp2 = tcg_const_i32(0xff);
1832 tmp3 = tcg_const_i32((insn & 7) << 3);
1833 break;
1834 case 1:
1835 tmp2 = tcg_const_i32(0xffff);
1836 tmp3 = tcg_const_i32((insn & 3) << 4);
1837 break;
1838 case 2:
1839 tmp2 = tcg_const_i32(0xffffffff);
1840 tmp3 = tcg_const_i32((insn & 1) << 5);
1841 break;
1842 default:
1843 tmp2 = NULL;
1844 tmp3 = NULL;
1845 }
1846 gen_helper_iwmmxt_insr(cpu_M0, cpu_M0, tmp, tmp2, tmp3);
1847 tcg_temp_free_i32(tmp3);
1848 tcg_temp_free_i32(tmp2);
1849 tcg_temp_free_i32(tmp);
1850 gen_op_iwmmxt_movq_wRn_M0(wrd);
1851 gen_op_iwmmxt_set_mup();
1852 break;
1853 case 0x107: case 0x507: case 0x907: case 0xd07: /* TEXTRM */
1854 rd = (insn >> 12) & 0xf;
1855 wrd = (insn >> 16) & 0xf;
1856 if (rd == 15 || ((insn >> 22) & 3) == 3)
1857 return 1;
1858 gen_op_iwmmxt_movq_M0_wRn(wrd);
1859 tmp = tcg_temp_new_i32();
1860 switch ((insn >> 22) & 3) {
1861 case 0:
1862 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 7) << 3);
1863 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1864 if (insn & 8) {
1865 tcg_gen_ext8s_i32(tmp, tmp);
1866 } else {
1867 tcg_gen_andi_i32(tmp, tmp, 0xff);
1868 }
1869 break;
1870 case 1:
1871 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 3) << 4);
1872 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1873 if (insn & 8) {
1874 tcg_gen_ext16s_i32(tmp, tmp);
1875 } else {
1876 tcg_gen_andi_i32(tmp, tmp, 0xffff);
1877 }
1878 break;
1879 case 2:
1880 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 1) << 5);
1881 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1882 break;
1883 }
1884 store_reg(s, rd, tmp);
1885 break;
1886 case 0x117: case 0x517: case 0x917: case 0xd17: /* TEXTRC */
1887 if ((insn & 0x000ff008) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1888 return 1;
1889 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1890 switch ((insn >> 22) & 3) {
1891 case 0:
1892 tcg_gen_shri_i32(tmp, tmp, ((insn & 7) << 2) + 0);
1893 break;
1894 case 1:
1895 tcg_gen_shri_i32(tmp, tmp, ((insn & 3) << 3) + 4);
1896 break;
1897 case 2:
1898 tcg_gen_shri_i32(tmp, tmp, ((insn & 1) << 4) + 12);
1899 break;
1900 }
1901 tcg_gen_shli_i32(tmp, tmp, 28);
1902 gen_set_nzcv(tmp);
1903 tcg_temp_free_i32(tmp);
1904 break;
1905 case 0x401: case 0x405: case 0x409: case 0x40d: /* TBCST */
1906 if (((insn >> 6) & 3) == 3)
1907 return 1;
1908 rd = (insn >> 12) & 0xf;
1909 wrd = (insn >> 16) & 0xf;
1910 tmp = load_reg(s, rd);
1911 switch ((insn >> 6) & 3) {
1912 case 0:
1913 gen_helper_iwmmxt_bcstb(cpu_M0, tmp);
1914 break;
1915 case 1:
1916 gen_helper_iwmmxt_bcstw(cpu_M0, tmp);
1917 break;
1918 case 2:
1919 gen_helper_iwmmxt_bcstl(cpu_M0, tmp);
1920 break;
1921 }
1922 tcg_temp_free_i32(tmp);
1923 gen_op_iwmmxt_movq_wRn_M0(wrd);
1924 gen_op_iwmmxt_set_mup();
1925 break;
1926 case 0x113: case 0x513: case 0x913: case 0xd13: /* TANDC */
1927 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1928 return 1;
1929 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1930 tmp2 = tcg_temp_new_i32();
1931 tcg_gen_mov_i32(tmp2, tmp);
1932 switch ((insn >> 22) & 3) {
1933 case 0:
1934 for (i = 0; i < 7; i ++) {
1935 tcg_gen_shli_i32(tmp2, tmp2, 4);
1936 tcg_gen_and_i32(tmp, tmp, tmp2);
1937 }
1938 break;
1939 case 1:
1940 for (i = 0; i < 3; i ++) {
1941 tcg_gen_shli_i32(tmp2, tmp2, 8);
1942 tcg_gen_and_i32(tmp, tmp, tmp2);
1943 }
1944 break;
1945 case 2:
1946 tcg_gen_shli_i32(tmp2, tmp2, 16);
1947 tcg_gen_and_i32(tmp, tmp, tmp2);
1948 break;
1949 }
1950 gen_set_nzcv(tmp);
1951 tcg_temp_free_i32(tmp2);
1952 tcg_temp_free_i32(tmp);
1953 break;
1954 case 0x01c: case 0x41c: case 0x81c: case 0xc1c: /* WACC */
1955 wrd = (insn >> 12) & 0xf;
1956 rd0 = (insn >> 16) & 0xf;
1957 gen_op_iwmmxt_movq_M0_wRn(rd0);
1958 switch ((insn >> 22) & 3) {
1959 case 0:
1960 gen_helper_iwmmxt_addcb(cpu_M0, cpu_M0);
1961 break;
1962 case 1:
1963 gen_helper_iwmmxt_addcw(cpu_M0, cpu_M0);
1964 break;
1965 case 2:
1966 gen_helper_iwmmxt_addcl(cpu_M0, cpu_M0);
1967 break;
1968 case 3:
1969 return 1;
1970 }
1971 gen_op_iwmmxt_movq_wRn_M0(wrd);
1972 gen_op_iwmmxt_set_mup();
1973 break;
1974 case 0x115: case 0x515: case 0x915: case 0xd15: /* TORC */
1975 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1976 return 1;
1977 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1978 tmp2 = tcg_temp_new_i32();
1979 tcg_gen_mov_i32(tmp2, tmp);
1980 switch ((insn >> 22) & 3) {
1981 case 0:
1982 for (i = 0; i < 7; i ++) {
1983 tcg_gen_shli_i32(tmp2, tmp2, 4);
1984 tcg_gen_or_i32(tmp, tmp, tmp2);
1985 }
1986 break;
1987 case 1:
1988 for (i = 0; i < 3; i ++) {
1989 tcg_gen_shli_i32(tmp2, tmp2, 8);
1990 tcg_gen_or_i32(tmp, tmp, tmp2);
1991 }
1992 break;
1993 case 2:
1994 tcg_gen_shli_i32(tmp2, tmp2, 16);
1995 tcg_gen_or_i32(tmp, tmp, tmp2);
1996 break;
1997 }
1998 gen_set_nzcv(tmp);
1999 tcg_temp_free_i32(tmp2);
2000 tcg_temp_free_i32(tmp);
2001 break;
2002 case 0x103: case 0x503: case 0x903: case 0xd03: /* TMOVMSK */
2003 rd = (insn >> 12) & 0xf;
2004 rd0 = (insn >> 16) & 0xf;
2005 if ((insn & 0xf) != 0 || ((insn >> 22) & 3) == 3)
2006 return 1;
2007 gen_op_iwmmxt_movq_M0_wRn(rd0);
2008 tmp = tcg_temp_new_i32();
2009 switch ((insn >> 22) & 3) {
2010 case 0:
2011 gen_helper_iwmmxt_msbb(tmp, cpu_M0);
2012 break;
2013 case 1:
2014 gen_helper_iwmmxt_msbw(tmp, cpu_M0);
2015 break;
2016 case 2:
2017 gen_helper_iwmmxt_msbl(tmp, cpu_M0);
2018 break;
2019 }
2020 store_reg(s, rd, tmp);
2021 break;
2022 case 0x106: case 0x306: case 0x506: case 0x706: /* WCMPGT */
2023 case 0x906: case 0xb06: case 0xd06: case 0xf06:
2024 wrd = (insn >> 12) & 0xf;
2025 rd0 = (insn >> 16) & 0xf;
2026 rd1 = (insn >> 0) & 0xf;
2027 gen_op_iwmmxt_movq_M0_wRn(rd0);
2028 switch ((insn >> 22) & 3) {
2029 case 0:
2030 if (insn & (1 << 21))
2031 gen_op_iwmmxt_cmpgtsb_M0_wRn(rd1);
2032 else
2033 gen_op_iwmmxt_cmpgtub_M0_wRn(rd1);
2034 break;
2035 case 1:
2036 if (insn & (1 << 21))
2037 gen_op_iwmmxt_cmpgtsw_M0_wRn(rd1);
2038 else
2039 gen_op_iwmmxt_cmpgtuw_M0_wRn(rd1);
2040 break;
2041 case 2:
2042 if (insn & (1 << 21))
2043 gen_op_iwmmxt_cmpgtsl_M0_wRn(rd1);
2044 else
2045 gen_op_iwmmxt_cmpgtul_M0_wRn(rd1);
2046 break;
2047 case 3:
2048 return 1;
2049 }
2050 gen_op_iwmmxt_movq_wRn_M0(wrd);
2051 gen_op_iwmmxt_set_mup();
2052 gen_op_iwmmxt_set_cup();
2053 break;
2054 case 0x00e: case 0x20e: case 0x40e: case 0x60e: /* WUNPCKEL */
2055 case 0x80e: case 0xa0e: case 0xc0e: case 0xe0e:
2056 wrd = (insn >> 12) & 0xf;
2057 rd0 = (insn >> 16) & 0xf;
2058 gen_op_iwmmxt_movq_M0_wRn(rd0);
2059 switch ((insn >> 22) & 3) {
2060 case 0:
2061 if (insn & (1 << 21))
2062 gen_op_iwmmxt_unpacklsb_M0();
2063 else
2064 gen_op_iwmmxt_unpacklub_M0();
2065 break;
2066 case 1:
2067 if (insn & (1 << 21))
2068 gen_op_iwmmxt_unpacklsw_M0();
2069 else
2070 gen_op_iwmmxt_unpackluw_M0();
2071 break;
2072 case 2:
2073 if (insn & (1 << 21))
2074 gen_op_iwmmxt_unpacklsl_M0();
2075 else
2076 gen_op_iwmmxt_unpacklul_M0();
2077 break;
2078 case 3:
2079 return 1;
2080 }
2081 gen_op_iwmmxt_movq_wRn_M0(wrd);
2082 gen_op_iwmmxt_set_mup();
2083 gen_op_iwmmxt_set_cup();
2084 break;
2085 case 0x00c: case 0x20c: case 0x40c: case 0x60c: /* WUNPCKEH */
2086 case 0x80c: case 0xa0c: case 0xc0c: case 0xe0c:
2087 wrd = (insn >> 12) & 0xf;
2088 rd0 = (insn >> 16) & 0xf;
2089 gen_op_iwmmxt_movq_M0_wRn(rd0);
2090 switch ((insn >> 22) & 3) {
2091 case 0:
2092 if (insn & (1 << 21))
2093 gen_op_iwmmxt_unpackhsb_M0();
2094 else
2095 gen_op_iwmmxt_unpackhub_M0();
2096 break;
2097 case 1:
2098 if (insn & (1 << 21))
2099 gen_op_iwmmxt_unpackhsw_M0();
2100 else
2101 gen_op_iwmmxt_unpackhuw_M0();
2102 break;
2103 case 2:
2104 if (insn & (1 << 21))
2105 gen_op_iwmmxt_unpackhsl_M0();
2106 else
2107 gen_op_iwmmxt_unpackhul_M0();
2108 break;
2109 case 3:
2110 return 1;
2111 }
2112 gen_op_iwmmxt_movq_wRn_M0(wrd);
2113 gen_op_iwmmxt_set_mup();
2114 gen_op_iwmmxt_set_cup();
2115 break;
2116 case 0x204: case 0x604: case 0xa04: case 0xe04: /* WSRL */
2117 case 0x214: case 0x614: case 0xa14: case 0xe14:
2118 if (((insn >> 22) & 3) == 0)
2119 return 1;
2120 wrd = (insn >> 12) & 0xf;
2121 rd0 = (insn >> 16) & 0xf;
2122 gen_op_iwmmxt_movq_M0_wRn(rd0);
2123 tmp = tcg_temp_new_i32();
2124 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2125 tcg_temp_free_i32(tmp);
2126 return 1;
2127 }
2128 switch ((insn >> 22) & 3) {
2129 case 1:
2130 gen_helper_iwmmxt_srlw(cpu_M0, cpu_env, cpu_M0, tmp);
2131 break;
2132 case 2:
2133 gen_helper_iwmmxt_srll(cpu_M0, cpu_env, cpu_M0, tmp);
2134 break;
2135 case 3:
2136 gen_helper_iwmmxt_srlq(cpu_M0, cpu_env, cpu_M0, tmp);
2137 break;
2138 }
2139 tcg_temp_free_i32(tmp);
2140 gen_op_iwmmxt_movq_wRn_M0(wrd);
2141 gen_op_iwmmxt_set_mup();
2142 gen_op_iwmmxt_set_cup();
2143 break;
2144 case 0x004: case 0x404: case 0x804: case 0xc04: /* WSRA */
2145 case 0x014: case 0x414: case 0x814: case 0xc14:
2146 if (((insn >> 22) & 3) == 0)
2147 return 1;
2148 wrd = (insn >> 12) & 0xf;
2149 rd0 = (insn >> 16) & 0xf;
2150 gen_op_iwmmxt_movq_M0_wRn(rd0);
2151 tmp = tcg_temp_new_i32();
2152 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2153 tcg_temp_free_i32(tmp);
2154 return 1;
2155 }
2156 switch ((insn >> 22) & 3) {
2157 case 1:
2158 gen_helper_iwmmxt_sraw(cpu_M0, cpu_env, cpu_M0, tmp);
2159 break;
2160 case 2:
2161 gen_helper_iwmmxt_sral(cpu_M0, cpu_env, cpu_M0, tmp);
2162 break;
2163 case 3:
2164 gen_helper_iwmmxt_sraq(cpu_M0, cpu_env, cpu_M0, tmp);
2165 break;
2166 }
2167 tcg_temp_free_i32(tmp);
2168 gen_op_iwmmxt_movq_wRn_M0(wrd);
2169 gen_op_iwmmxt_set_mup();
2170 gen_op_iwmmxt_set_cup();
2171 break;
2172 case 0x104: case 0x504: case 0x904: case 0xd04: /* WSLL */
2173 case 0x114: case 0x514: case 0x914: case 0xd14:
2174 if (((insn >> 22) & 3) == 0)
2175 return 1;
2176 wrd = (insn >> 12) & 0xf;
2177 rd0 = (insn >> 16) & 0xf;
2178 gen_op_iwmmxt_movq_M0_wRn(rd0);
2179 tmp = tcg_temp_new_i32();
2180 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2181 tcg_temp_free_i32(tmp);
2182 return 1;
2183 }
2184 switch ((insn >> 22) & 3) {
2185 case 1:
2186 gen_helper_iwmmxt_sllw(cpu_M0, cpu_env, cpu_M0, tmp);
2187 break;
2188 case 2:
2189 gen_helper_iwmmxt_slll(cpu_M0, cpu_env, cpu_M0, tmp);
2190 break;
2191 case 3:
2192 gen_helper_iwmmxt_sllq(cpu_M0, cpu_env, cpu_M0, tmp);
2193 break;
2194 }
2195 tcg_temp_free_i32(tmp);
2196 gen_op_iwmmxt_movq_wRn_M0(wrd);
2197 gen_op_iwmmxt_set_mup();
2198 gen_op_iwmmxt_set_cup();
2199 break;
2200 case 0x304: case 0x704: case 0xb04: case 0xf04: /* WROR */
2201 case 0x314: case 0x714: case 0xb14: case 0xf14:
2202 if (((insn >> 22) & 3) == 0)
2203 return 1;
2204 wrd = (insn >> 12) & 0xf;
2205 rd0 = (insn >> 16) & 0xf;
2206 gen_op_iwmmxt_movq_M0_wRn(rd0);
2207 tmp = tcg_temp_new_i32();
2208 switch ((insn >> 22) & 3) {
2209 case 1:
2210 if (gen_iwmmxt_shift(insn, 0xf, tmp)) {
2211 tcg_temp_free_i32(tmp);
2212 return 1;
2213 }
2214 gen_helper_iwmmxt_rorw(cpu_M0, cpu_env, cpu_M0, tmp);
2215 break;
2216 case 2:
2217 if (gen_iwmmxt_shift(insn, 0x1f, tmp)) {
2218 tcg_temp_free_i32(tmp);
2219 return 1;
2220 }
2221 gen_helper_iwmmxt_rorl(cpu_M0, cpu_env, cpu_M0, tmp);
2222 break;
2223 case 3:
2224 if (gen_iwmmxt_shift(insn, 0x3f, tmp)) {
2225 tcg_temp_free_i32(tmp);
2226 return 1;
2227 }
2228 gen_helper_iwmmxt_rorq(cpu_M0, cpu_env, cpu_M0, tmp);
2229 break;
2230 }
2231 tcg_temp_free_i32(tmp);
2232 gen_op_iwmmxt_movq_wRn_M0(wrd);
2233 gen_op_iwmmxt_set_mup();
2234 gen_op_iwmmxt_set_cup();
2235 break;
2236 case 0x116: case 0x316: case 0x516: case 0x716: /* WMIN */
2237 case 0x916: case 0xb16: case 0xd16: case 0xf16:
2238 wrd = (insn >> 12) & 0xf;
2239 rd0 = (insn >> 16) & 0xf;
2240 rd1 = (insn >> 0) & 0xf;
2241 gen_op_iwmmxt_movq_M0_wRn(rd0);
2242 switch ((insn >> 22) & 3) {
2243 case 0:
2244 if (insn & (1 << 21))
2245 gen_op_iwmmxt_minsb_M0_wRn(rd1);
2246 else
2247 gen_op_iwmmxt_minub_M0_wRn(rd1);
2248 break;
2249 case 1:
2250 if (insn & (1 << 21))
2251 gen_op_iwmmxt_minsw_M0_wRn(rd1);
2252 else
2253 gen_op_iwmmxt_minuw_M0_wRn(rd1);
2254 break;
2255 case 2:
2256 if (insn & (1 << 21))
2257 gen_op_iwmmxt_minsl_M0_wRn(rd1);
2258 else
2259 gen_op_iwmmxt_minul_M0_wRn(rd1);
2260 break;
2261 case 3:
2262 return 1;
2263 }
2264 gen_op_iwmmxt_movq_wRn_M0(wrd);
2265 gen_op_iwmmxt_set_mup();
2266 break;
2267 case 0x016: case 0x216: case 0x416: case 0x616: /* WMAX */
2268 case 0x816: case 0xa16: case 0xc16: case 0xe16:
2269 wrd = (insn >> 12) & 0xf;
2270 rd0 = (insn >> 16) & 0xf;
2271 rd1 = (insn >> 0) & 0xf;
2272 gen_op_iwmmxt_movq_M0_wRn(rd0);
2273 switch ((insn >> 22) & 3) {
2274 case 0:
2275 if (insn & (1 << 21))
2276 gen_op_iwmmxt_maxsb_M0_wRn(rd1);
2277 else
2278 gen_op_iwmmxt_maxub_M0_wRn(rd1);
2279 break;
2280 case 1:
2281 if (insn & (1 << 21))
2282 gen_op_iwmmxt_maxsw_M0_wRn(rd1);
2283 else
2284 gen_op_iwmmxt_maxuw_M0_wRn(rd1);
2285 break;
2286 case 2:
2287 if (insn & (1 << 21))
2288 gen_op_iwmmxt_maxsl_M0_wRn(rd1);
2289 else
2290 gen_op_iwmmxt_maxul_M0_wRn(rd1);
2291 break;
2292 case 3:
2293 return 1;
2294 }
2295 gen_op_iwmmxt_movq_wRn_M0(wrd);
2296 gen_op_iwmmxt_set_mup();
2297 break;
2298 case 0x002: case 0x102: case 0x202: case 0x302: /* WALIGNI */
2299 case 0x402: case 0x502: case 0x602: case 0x702:
2300 wrd = (insn >> 12) & 0xf;
2301 rd0 = (insn >> 16) & 0xf;
2302 rd1 = (insn >> 0) & 0xf;
2303 gen_op_iwmmxt_movq_M0_wRn(rd0);
2304 tmp = tcg_const_i32((insn >> 20) & 3);
2305 iwmmxt_load_reg(cpu_V1, rd1);
2306 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
2307 tcg_temp_free_i32(tmp);
2308 gen_op_iwmmxt_movq_wRn_M0(wrd);
2309 gen_op_iwmmxt_set_mup();
2310 break;
2311 case 0x01a: case 0x11a: case 0x21a: case 0x31a: /* WSUB */
2312 case 0x41a: case 0x51a: case 0x61a: case 0x71a:
2313 case 0x81a: case 0x91a: case 0xa1a: case 0xb1a:
2314 case 0xc1a: case 0xd1a: case 0xe1a: case 0xf1a:
2315 wrd = (insn >> 12) & 0xf;
2316 rd0 = (insn >> 16) & 0xf;
2317 rd1 = (insn >> 0) & 0xf;
2318 gen_op_iwmmxt_movq_M0_wRn(rd0);
2319 switch ((insn >> 20) & 0xf) {
2320 case 0x0:
2321 gen_op_iwmmxt_subnb_M0_wRn(rd1);
2322 break;
2323 case 0x1:
2324 gen_op_iwmmxt_subub_M0_wRn(rd1);
2325 break;
2326 case 0x3:
2327 gen_op_iwmmxt_subsb_M0_wRn(rd1);
2328 break;
2329 case 0x4:
2330 gen_op_iwmmxt_subnw_M0_wRn(rd1);
2331 break;
2332 case 0x5:
2333 gen_op_iwmmxt_subuw_M0_wRn(rd1);
2334 break;
2335 case 0x7:
2336 gen_op_iwmmxt_subsw_M0_wRn(rd1);
2337 break;
2338 case 0x8:
2339 gen_op_iwmmxt_subnl_M0_wRn(rd1);
2340 break;
2341 case 0x9:
2342 gen_op_iwmmxt_subul_M0_wRn(rd1);
2343 break;
2344 case 0xb:
2345 gen_op_iwmmxt_subsl_M0_wRn(rd1);
2346 break;
2347 default:
2348 return 1;
2349 }
2350 gen_op_iwmmxt_movq_wRn_M0(wrd);
2351 gen_op_iwmmxt_set_mup();
2352 gen_op_iwmmxt_set_cup();
2353 break;
2354 case 0x01e: case 0x11e: case 0x21e: case 0x31e: /* WSHUFH */
2355 case 0x41e: case 0x51e: case 0x61e: case 0x71e:
2356 case 0x81e: case 0x91e: case 0xa1e: case 0xb1e:
2357 case 0xc1e: case 0xd1e: case 0xe1e: case 0xf1e:
2358 wrd = (insn >> 12) & 0xf;
2359 rd0 = (insn >> 16) & 0xf;
2360 gen_op_iwmmxt_movq_M0_wRn(rd0);
2361 tmp = tcg_const_i32(((insn >> 16) & 0xf0) | (insn & 0x0f));
2362 gen_helper_iwmmxt_shufh(cpu_M0, cpu_env, cpu_M0, tmp);
2363 tcg_temp_free_i32(tmp);
2364 gen_op_iwmmxt_movq_wRn_M0(wrd);
2365 gen_op_iwmmxt_set_mup();
2366 gen_op_iwmmxt_set_cup();
2367 break;
2368 case 0x018: case 0x118: case 0x218: case 0x318: /* WADD */
2369 case 0x418: case 0x518: case 0x618: case 0x718:
2370 case 0x818: case 0x918: case 0xa18: case 0xb18:
2371 case 0xc18: case 0xd18: case 0xe18: case 0xf18:
2372 wrd = (insn >> 12) & 0xf;
2373 rd0 = (insn >> 16) & 0xf;
2374 rd1 = (insn >> 0) & 0xf;
2375 gen_op_iwmmxt_movq_M0_wRn(rd0);
2376 switch ((insn >> 20) & 0xf) {
2377 case 0x0:
2378 gen_op_iwmmxt_addnb_M0_wRn(rd1);
2379 break;
2380 case 0x1:
2381 gen_op_iwmmxt_addub_M0_wRn(rd1);
2382 break;
2383 case 0x3:
2384 gen_op_iwmmxt_addsb_M0_wRn(rd1);
2385 break;
2386 case 0x4:
2387 gen_op_iwmmxt_addnw_M0_wRn(rd1);
2388 break;
2389 case 0x5:
2390 gen_op_iwmmxt_adduw_M0_wRn(rd1);
2391 break;
2392 case 0x7:
2393 gen_op_iwmmxt_addsw_M0_wRn(rd1);
2394 break;
2395 case 0x8:
2396 gen_op_iwmmxt_addnl_M0_wRn(rd1);
2397 break;
2398 case 0x9:
2399 gen_op_iwmmxt_addul_M0_wRn(rd1);
2400 break;
2401 case 0xb:
2402 gen_op_iwmmxt_addsl_M0_wRn(rd1);
2403 break;
2404 default:
2405 return 1;
2406 }
2407 gen_op_iwmmxt_movq_wRn_M0(wrd);
2408 gen_op_iwmmxt_set_mup();
2409 gen_op_iwmmxt_set_cup();
2410 break;
2411 case 0x008: case 0x108: case 0x208: case 0x308: /* WPACK */
2412 case 0x408: case 0x508: case 0x608: case 0x708:
2413 case 0x808: case 0x908: case 0xa08: case 0xb08:
2414 case 0xc08: case 0xd08: case 0xe08: case 0xf08:
2415 if (!(insn & (1 << 20)) || ((insn >> 22) & 3) == 0)
2416 return 1;
2417 wrd = (insn >> 12) & 0xf;
2418 rd0 = (insn >> 16) & 0xf;
2419 rd1 = (insn >> 0) & 0xf;
2420 gen_op_iwmmxt_movq_M0_wRn(rd0);
2421 switch ((insn >> 22) & 3) {
2422 case 1:
2423 if (insn & (1 << 21))
2424 gen_op_iwmmxt_packsw_M0_wRn(rd1);
2425 else
2426 gen_op_iwmmxt_packuw_M0_wRn(rd1);
2427 break;
2428 case 2:
2429 if (insn & (1 << 21))
2430 gen_op_iwmmxt_packsl_M0_wRn(rd1);
2431 else
2432 gen_op_iwmmxt_packul_M0_wRn(rd1);
2433 break;
2434 case 3:
2435 if (insn & (1 << 21))
2436 gen_op_iwmmxt_packsq_M0_wRn(rd1);
2437 else
2438 gen_op_iwmmxt_packuq_M0_wRn(rd1);
2439 break;
2440 }
2441 gen_op_iwmmxt_movq_wRn_M0(wrd);
2442 gen_op_iwmmxt_set_mup();
2443 gen_op_iwmmxt_set_cup();
2444 break;
2445 case 0x201: case 0x203: case 0x205: case 0x207:
2446 case 0x209: case 0x20b: case 0x20d: case 0x20f:
2447 case 0x211: case 0x213: case 0x215: case 0x217:
2448 case 0x219: case 0x21b: case 0x21d: case 0x21f:
2449 wrd = (insn >> 5) & 0xf;
2450 rd0 = (insn >> 12) & 0xf;
2451 rd1 = (insn >> 0) & 0xf;
2452 if (rd0 == 0xf || rd1 == 0xf)
2453 return 1;
2454 gen_op_iwmmxt_movq_M0_wRn(wrd);
2455 tmp = load_reg(s, rd0);
2456 tmp2 = load_reg(s, rd1);
2457 switch ((insn >> 16) & 0xf) {
2458 case 0x0: /* TMIA */
2459 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2460 break;
2461 case 0x8: /* TMIAPH */
2462 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2463 break;
2464 case 0xc: case 0xd: case 0xe: case 0xf: /* TMIAxy */
2465 if (insn & (1 << 16))
2466 tcg_gen_shri_i32(tmp, tmp, 16);
2467 if (insn & (1 << 17))
2468 tcg_gen_shri_i32(tmp2, tmp2, 16);
2469 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2470 break;
2471 default:
2472 tcg_temp_free_i32(tmp2);
2473 tcg_temp_free_i32(tmp);
2474 return 1;
2475 }
2476 tcg_temp_free_i32(tmp2);
2477 tcg_temp_free_i32(tmp);
2478 gen_op_iwmmxt_movq_wRn_M0(wrd);
2479 gen_op_iwmmxt_set_mup();
2480 break;
2481 default:
2482 return 1;
2483 }
2484
2485 return 0;
2486 }
2487
2488 /* Disassemble an XScale DSP instruction. Returns nonzero if an error occurred
2489 (ie. an undefined instruction). */
2490 static int disas_dsp_insn(DisasContext *s, uint32_t insn)
2491 {
2492 int acc, rd0, rd1, rdhi, rdlo;
2493 TCGv_i32 tmp, tmp2;
2494
2495 if ((insn & 0x0ff00f10) == 0x0e200010) {
2496 /* Multiply with Internal Accumulate Format */
2497 rd0 = (insn >> 12) & 0xf;
2498 rd1 = insn & 0xf;
2499 acc = (insn >> 5) & 7;
2500
2501 if (acc != 0)
2502 return 1;
2503
2504 tmp = load_reg(s, rd0);
2505 tmp2 = load_reg(s, rd1);
2506 switch ((insn >> 16) & 0xf) {
2507 case 0x0: /* MIA */
2508 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2509 break;
2510 case 0x8: /* MIAPH */
2511 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2512 break;
2513 case 0xc: /* MIABB */
2514 case 0xd: /* MIABT */
2515 case 0xe: /* MIATB */
2516 case 0xf: /* MIATT */
2517 if (insn & (1 << 16))
2518 tcg_gen_shri_i32(tmp, tmp, 16);
2519 if (insn & (1 << 17))
2520 tcg_gen_shri_i32(tmp2, tmp2, 16);
2521 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2522 break;
2523 default:
2524 return 1;
2525 }
2526 tcg_temp_free_i32(tmp2);
2527 tcg_temp_free_i32(tmp);
2528
2529 gen_op_iwmmxt_movq_wRn_M0(acc);
2530 return 0;
2531 }
2532
2533 if ((insn & 0x0fe00ff8) == 0x0c400000) {
2534 /* Internal Accumulator Access Format */
2535 rdhi = (insn >> 16) & 0xf;
2536 rdlo = (insn >> 12) & 0xf;
2537 acc = insn & 7;
2538
2539 if (acc != 0)
2540 return 1;
2541
2542 if (insn & ARM_CP_RW_BIT) { /* MRA */
2543 iwmmxt_load_reg(cpu_V0, acc);
2544 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
2545 tcg_gen_extrh_i64_i32(cpu_R[rdhi], cpu_V0);
2546 tcg_gen_andi_i32(cpu_R[rdhi], cpu_R[rdhi], (1 << (40 - 32)) - 1);
2547 } else { /* MAR */
2548 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
2549 iwmmxt_store_reg(cpu_V0, acc);
2550 }
2551 return 0;
2552 }
2553
2554 return 1;
2555 }
2556
2557 static void gen_goto_ptr(void)
2558 {
2559 tcg_gen_lookup_and_goto_ptr();
2560 }
2561
2562 /* This will end the TB but doesn't guarantee we'll return to
2563 * cpu_loop_exec. Any live exit_requests will be processed as we
2564 * enter the next TB.
2565 */
2566 static void gen_goto_tb(DisasContext *s, int n, target_ulong dest)
2567 {
2568 if (translator_use_goto_tb(&s->base, dest)) {
2569 tcg_gen_goto_tb(n);
2570 gen_set_pc_im(s, dest);
2571 tcg_gen_exit_tb(s->base.tb, n);
2572 } else {
2573 gen_set_pc_im(s, dest);
2574 gen_goto_ptr();
2575 }
2576 s->base.is_jmp = DISAS_NORETURN;
2577 }
2578
2579 /* Jump, specifying which TB number to use if we gen_goto_tb() */
2580 static inline void gen_jmp_tb(DisasContext *s, uint32_t dest, int tbno)
2581 {
2582 if (unlikely(s->ss_active)) {
2583 /* An indirect jump so that we still trigger the debug exception. */
2584 gen_set_pc_im(s, dest);
2585 s->base.is_jmp = DISAS_JUMP;
2586 return;
2587 }
2588 switch (s->base.is_jmp) {
2589 case DISAS_NEXT:
2590 case DISAS_TOO_MANY:
2591 case DISAS_NORETURN:
2592 /*
2593 * The normal case: just go to the destination TB.
2594 * NB: NORETURN happens if we generate code like
2595 * gen_brcondi(l);
2596 * gen_jmp();
2597 * gen_set_label(l);
2598 * gen_jmp();
2599 * on the second call to gen_jmp().
2600 */
2601 gen_goto_tb(s, tbno, dest);
2602 break;
2603 case DISAS_UPDATE_NOCHAIN:
2604 case DISAS_UPDATE_EXIT:
2605 /*
2606 * We already decided we're leaving the TB for some other reason.
2607 * Avoid using goto_tb so we really do exit back to the main loop
2608 * and don't chain to another TB.
2609 */
2610 gen_set_pc_im(s, dest);
2611 gen_goto_ptr();
2612 s->base.is_jmp = DISAS_NORETURN;
2613 break;
2614 default:
2615 /*
2616 * We shouldn't be emitting code for a jump and also have
2617 * is_jmp set to one of the special cases like DISAS_SWI.
2618 */
2619 g_assert_not_reached();
2620 }
2621 }
2622
2623 static inline void gen_jmp(DisasContext *s, uint32_t dest)
2624 {
2625 gen_jmp_tb(s, dest, 0);
2626 }
2627
2628 static inline void gen_mulxy(TCGv_i32 t0, TCGv_i32 t1, int x, int y)
2629 {
2630 if (x)
2631 tcg_gen_sari_i32(t0, t0, 16);
2632 else
2633 gen_sxth(t0);
2634 if (y)
2635 tcg_gen_sari_i32(t1, t1, 16);
2636 else
2637 gen_sxth(t1);
2638 tcg_gen_mul_i32(t0, t0, t1);
2639 }
2640
2641 /* Return the mask of PSR bits set by a MSR instruction. */
2642 static uint32_t msr_mask(DisasContext *s, int flags, int spsr)
2643 {
2644 uint32_t mask = 0;
2645
2646 if (flags & (1 << 0)) {
2647 mask |= 0xff;
2648 }
2649 if (flags & (1 << 1)) {
2650 mask |= 0xff00;
2651 }
2652 if (flags & (1 << 2)) {
2653 mask |= 0xff0000;
2654 }
2655 if (flags & (1 << 3)) {
2656 mask |= 0xff000000;
2657 }
2658
2659 /* Mask out undefined and reserved bits. */
2660 mask &= aarch32_cpsr_valid_mask(s->features, s->isar);
2661
2662 /* Mask out execution state. */
2663 if (!spsr) {
2664 mask &= ~CPSR_EXEC;
2665 }
2666
2667 /* Mask out privileged bits. */
2668 if (IS_USER(s)) {
2669 mask &= CPSR_USER;
2670 }
2671 return mask;
2672 }
2673
2674 /* Returns nonzero if access to the PSR is not permitted. Marks t0 as dead. */
2675 static int gen_set_psr(DisasContext *s, uint32_t mask, int spsr, TCGv_i32 t0)
2676 {
2677 TCGv_i32 tmp;
2678 if (spsr) {
2679 /* ??? This is also undefined in system mode. */
2680 if (IS_USER(s))
2681 return 1;
2682
2683 tmp = load_cpu_field(spsr);
2684 tcg_gen_andi_i32(tmp, tmp, ~mask);
2685 tcg_gen_andi_i32(t0, t0, mask);
2686 tcg_gen_or_i32(tmp, tmp, t0);
2687 store_cpu_field(tmp, spsr);
2688 } else {
2689 gen_set_cpsr(t0, mask);
2690 }
2691 tcg_temp_free_i32(t0);
2692 gen_lookup_tb(s);
2693 return 0;
2694 }
2695
2696 /* Returns nonzero if access to the PSR is not permitted. */
2697 static int gen_set_psr_im(DisasContext *s, uint32_t mask, int spsr, uint32_t val)
2698 {
2699 TCGv_i32 tmp;
2700 tmp = tcg_temp_new_i32();
2701 tcg_gen_movi_i32(tmp, val);
2702 return gen_set_psr(s, mask, spsr, tmp);
2703 }
2704
2705 static bool msr_banked_access_decode(DisasContext *s, int r, int sysm, int rn,
2706 int *tgtmode, int *regno)
2707 {
2708 /* Decode the r and sysm fields of MSR/MRS banked accesses into
2709 * the target mode and register number, and identify the various
2710 * unpredictable cases.
2711 * MSR (banked) and MRS (banked) are CONSTRAINED UNPREDICTABLE if:
2712 * + executed in user mode
2713 * + using R15 as the src/dest register
2714 * + accessing an unimplemented register
2715 * + accessing a register that's inaccessible at current PL/security state*
2716 * + accessing a register that you could access with a different insn
2717 * We choose to UNDEF in all these cases.
2718 * Since we don't know which of the various AArch32 modes we are in
2719 * we have to defer some checks to runtime.
2720 * Accesses to Monitor mode registers from Secure EL1 (which implies
2721 * that EL3 is AArch64) must trap to EL3.
2722 *
2723 * If the access checks fail this function will emit code to take
2724 * an exception and return false. Otherwise it will return true,
2725 * and set *tgtmode and *regno appropriately.
2726 */
2727 int exc_target = default_exception_el(s);
2728
2729 /* These instructions are present only in ARMv8, or in ARMv7 with the
2730 * Virtualization Extensions.
2731 */
2732 if (!arm_dc_feature(s, ARM_FEATURE_V8) &&
2733 !arm_dc_feature(s, ARM_FEATURE_EL2)) {
2734 goto undef;
2735 }
2736
2737 if (IS_USER(s) || rn == 15) {
2738 goto undef;
2739 }
2740
2741 /* The table in the v8 ARM ARM section F5.2.3 describes the encoding
2742 * of registers into (r, sysm).
2743 */
2744 if (r) {
2745 /* SPSRs for other modes */
2746 switch (sysm) {
2747 case 0xe: /* SPSR_fiq */
2748 *tgtmode = ARM_CPU_MODE_FIQ;
2749 break;
2750 case 0x10: /* SPSR_irq */
2751 *tgtmode = ARM_CPU_MODE_IRQ;
2752 break;
2753 case 0x12: /* SPSR_svc */
2754 *tgtmode = ARM_CPU_MODE_SVC;
2755 break;
2756 case 0x14: /* SPSR_abt */
2757 *tgtmode = ARM_CPU_MODE_ABT;
2758 break;
2759 case 0x16: /* SPSR_und */
2760 *tgtmode = ARM_CPU_MODE_UND;
2761 break;
2762 case 0x1c: /* SPSR_mon */
2763 *tgtmode = ARM_CPU_MODE_MON;
2764 break;
2765 case 0x1e: /* SPSR_hyp */
2766 *tgtmode = ARM_CPU_MODE_HYP;
2767 break;
2768 default: /* unallocated */
2769 goto undef;
2770 }
2771 /* We arbitrarily assign SPSR a register number of 16. */
2772 *regno = 16;
2773 } else {
2774 /* general purpose registers for other modes */
2775 switch (sysm) {
2776 case 0x0 ... 0x6: /* 0b00xxx : r8_usr ... r14_usr */
2777 *tgtmode = ARM_CPU_MODE_USR;
2778 *regno = sysm + 8;
2779 break;
2780 case 0x8 ... 0xe: /* 0b01xxx : r8_fiq ... r14_fiq */
2781 *tgtmode = ARM_CPU_MODE_FIQ;
2782 *regno = sysm;
2783 break;
2784 case 0x10 ... 0x11: /* 0b1000x : r14_irq, r13_irq */
2785 *tgtmode = ARM_CPU_MODE_IRQ;
2786 *regno = sysm & 1 ? 13 : 14;
2787 break;
2788 case 0x12 ... 0x13: /* 0b1001x : r14_svc, r13_svc */
2789 *tgtmode = ARM_CPU_MODE_SVC;
2790 *regno = sysm & 1 ? 13 : 14;
2791 break;
2792 case 0x14 ... 0x15: /* 0b1010x : r14_abt, r13_abt */
2793 *tgtmode = ARM_CPU_MODE_ABT;
2794 *regno = sysm & 1 ? 13 : 14;
2795 break;
2796 case 0x16 ... 0x17: /* 0b1011x : r14_und, r13_und */
2797 *tgtmode = ARM_CPU_MODE_UND;
2798 *regno = sysm & 1 ? 13 : 14;
2799 break;
2800 case 0x1c ... 0x1d: /* 0b1110x : r14_mon, r13_mon */
2801 *tgtmode = ARM_CPU_MODE_MON;
2802 *regno = sysm & 1 ? 13 : 14;
2803 break;
2804 case 0x1e ... 0x1f: /* 0b1111x : elr_hyp, r13_hyp */
2805 *tgtmode = ARM_CPU_MODE_HYP;
2806 /* Arbitrarily pick 17 for ELR_Hyp (which is not a banked LR!) */
2807 *regno = sysm & 1 ? 13 : 17;
2808 break;
2809 default: /* unallocated */
2810 goto undef;
2811 }
2812 }
2813
2814 /* Catch the 'accessing inaccessible register' cases we can detect
2815 * at translate time.
2816 */
2817 switch (*tgtmode) {
2818 case ARM_CPU_MODE_MON:
2819 if (!arm_dc_feature(s, ARM_FEATURE_EL3) || s->ns) {
2820 goto undef;
2821 }
2822 if (s->current_el == 1) {
2823 /* If we're in Secure EL1 (which implies that EL3 is AArch64)
2824 * then accesses to Mon registers trap to Secure EL2, if it exists,
2825 * otherwise EL3.
2826 */
2827 TCGv_i32 tcg_el;
2828
2829 if (arm_dc_feature(s, ARM_FEATURE_AARCH64) &&
2830 dc_isar_feature(aa64_sel2, s)) {
2831 /* Target EL is EL<3 minus SCR_EL3.EEL2> */
2832 tcg_el = load_cpu_field(cp15.scr_el3);
2833 tcg_gen_sextract_i32(tcg_el, tcg_el, ctz32(SCR_EEL2), 1);
2834 tcg_gen_addi_i32(tcg_el, tcg_el, 3);
2835 } else {
2836 tcg_el = tcg_const_i32(3);
2837 }
2838
2839 gen_exception_el(s, EXCP_UDEF, syn_uncategorized(), tcg_el);
2840 tcg_temp_free_i32(tcg_el);
2841 return false;
2842 }
2843 break;
2844 case ARM_CPU_MODE_HYP:
2845 /*
2846 * SPSR_hyp and r13_hyp can only be accessed from Monitor mode
2847 * (and so we can forbid accesses from EL2 or below). elr_hyp
2848 * can be accessed also from Hyp mode, so forbid accesses from
2849 * EL0 or EL1.
2850 */
2851 if (!arm_dc_feature(s, ARM_FEATURE_EL2) || s->current_el < 2 ||
2852 (s->current_el < 3 && *regno != 17)) {
2853 goto undef;
2854 }
2855 break;
2856 default:
2857 break;
2858 }
2859
2860 return true;
2861
2862 undef:
2863 /* If we get here then some access check did not pass */
2864 gen_exception_insn(s, s->pc_curr, EXCP_UDEF,
2865 syn_uncategorized(), exc_target);
2866 return false;
2867 }
2868
2869 static void gen_msr_banked(DisasContext *s, int r, int sysm, int rn)
2870 {
2871 TCGv_i32 tcg_reg, tcg_tgtmode, tcg_regno;
2872 int tgtmode = 0, regno = 0;
2873
2874 if (!msr_banked_access_decode(s, r, sysm, rn, &tgtmode, &regno)) {
2875 return;
2876 }
2877
2878 /* Sync state because msr_banked() can raise exceptions */
2879 gen_set_condexec(s);
2880 gen_set_pc_im(s, s->pc_curr);
2881 tcg_reg = load_reg(s, rn);
2882 tcg_tgtmode = tcg_const_i32(tgtmode);
2883 tcg_regno = tcg_const_i32(regno);
2884 gen_helper_msr_banked(cpu_env, tcg_reg, tcg_tgtmode, tcg_regno);
2885 tcg_temp_free_i32(tcg_tgtmode);
2886 tcg_temp_free_i32(tcg_regno);
2887 tcg_temp_free_i32(tcg_reg);
2888 s->base.is_jmp = DISAS_UPDATE_EXIT;
2889 }
2890
2891 static void gen_mrs_banked(DisasContext *s, int r, int sysm, int rn)
2892 {
2893 TCGv_i32 tcg_reg, tcg_tgtmode, tcg_regno;
2894 int tgtmode = 0, regno = 0;
2895
2896 if (!msr_banked_access_decode(s, r, sysm, rn, &tgtmode, &regno)) {
2897 return;
2898 }
2899
2900 /* Sync state because mrs_banked() can raise exceptions */
2901 gen_set_condexec(s);
2902 gen_set_pc_im(s, s->pc_curr);
2903 tcg_reg = tcg_temp_new_i32();
2904 tcg_tgtmode = tcg_const_i32(tgtmode);
2905 tcg_regno = tcg_const_i32(regno);
2906 gen_helper_mrs_banked(tcg_reg, cpu_env, tcg_tgtmode, tcg_regno);
2907 tcg_temp_free_i32(tcg_tgtmode);
2908 tcg_temp_free_i32(tcg_regno);
2909 store_reg(s, rn, tcg_reg);
2910 s->base.is_jmp = DISAS_UPDATE_EXIT;
2911 }
2912
2913 /* Store value to PC as for an exception return (ie don't
2914 * mask bits). The subsequent call to gen_helper_cpsr_write_eret()
2915 * will do the masking based on the new value of the Thumb bit.
2916 */
2917 static void store_pc_exc_ret(DisasContext *s, TCGv_i32 pc)
2918 {
2919 tcg_gen_mov_i32(cpu_R[15], pc);
2920 tcg_temp_free_i32(pc);
2921 }
2922
2923 /* Generate a v6 exception return. Marks both values as dead. */
2924 static void gen_rfe(DisasContext *s, TCGv_i32 pc, TCGv_i32 cpsr)
2925 {
2926 store_pc_exc_ret(s, pc);
2927 /* The cpsr_write_eret helper will mask the low bits of PC
2928 * appropriately depending on the new Thumb bit, so it must
2929 * be called after storing the new PC.
2930 */
2931 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
2932 gen_io_start();
2933 }
2934 gen_helper_cpsr_write_eret(cpu_env, cpsr);
2935 tcg_temp_free_i32(cpsr);
2936 /* Must exit loop to check un-masked IRQs */
2937 s->base.is_jmp = DISAS_EXIT;
2938 }
2939
2940 /* Generate an old-style exception return. Marks pc as dead. */
2941 static void gen_exception_return(DisasContext *s, TCGv_i32 pc)
2942 {
2943 gen_rfe(s, pc, load_cpu_field(spsr));
2944 }
2945
2946 static void gen_gvec_fn3_qc(uint32_t rd_ofs, uint32_t rn_ofs, uint32_t rm_ofs,
2947 uint32_t opr_sz, uint32_t max_sz,
2948 gen_helper_gvec_3_ptr *fn)
2949 {
2950 TCGv_ptr qc_ptr = tcg_temp_new_ptr();
2951
2952 tcg_gen_addi_ptr(qc_ptr, cpu_env, offsetof(CPUARMState, vfp.qc));
2953 tcg_gen_gvec_3_ptr(rd_ofs, rn_ofs, rm_ofs, qc_ptr,
2954 opr_sz, max_sz, 0, fn);
2955 tcg_temp_free_ptr(qc_ptr);
2956 }
2957
2958 void gen_gvec_sqrdmlah_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
2959 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
2960 {
2961 static gen_helper_gvec_3_ptr * const fns[2] = {
2962 gen_helper_gvec_qrdmlah_s16, gen_helper_gvec_qrdmlah_s32
2963 };
2964 tcg_debug_assert(vece >= 1 && vece <= 2);
2965 gen_gvec_fn3_qc(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, fns[vece - 1]);
2966 }
2967
2968 void gen_gvec_sqrdmlsh_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
2969 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
2970 {
2971 static gen_helper_gvec_3_ptr * const fns[2] = {
2972 gen_helper_gvec_qrdmlsh_s16, gen_helper_gvec_qrdmlsh_s32
2973 };
2974 tcg_debug_assert(vece >= 1 && vece <= 2);
2975 gen_gvec_fn3_qc(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, fns[vece - 1]);
2976 }
2977
2978 #define GEN_CMP0(NAME, COND) \
2979 static void gen_##NAME##0_i32(TCGv_i32 d, TCGv_i32 a) \
2980 { \
2981 tcg_gen_setcondi_i32(COND, d, a, 0); \
2982 tcg_gen_neg_i32(d, d); \
2983 } \
2984 static void gen_##NAME##0_i64(TCGv_i64 d, TCGv_i64 a) \
2985 { \
2986 tcg_gen_setcondi_i64(COND, d, a, 0); \
2987 tcg_gen_neg_i64(d, d); \
2988 } \
2989 static void gen_##NAME##0_vec(unsigned vece, TCGv_vec d, TCGv_vec a) \
2990 { \
2991 TCGv_vec zero = tcg_const_zeros_vec_matching(d); \
2992 tcg_gen_cmp_vec(COND, vece, d, a, zero); \
2993 tcg_temp_free_vec(zero); \
2994 } \
2995 void gen_gvec_##NAME##0(unsigned vece, uint32_t d, uint32_t m, \
2996 uint32_t opr_sz, uint32_t max_sz) \
2997 { \
2998 const GVecGen2 op[4] = { \
2999 { .fno = gen_helper_gvec_##NAME##0_b, \
3000 .fniv = gen_##NAME##0_vec, \
3001 .opt_opc = vecop_list_cmp, \
3002 .vece = MO_8 }, \
3003 { .fno = gen_helper_gvec_##NAME##0_h, \
3004 .fniv = gen_##NAME##0_vec, \
3005 .opt_opc = vecop_list_cmp, \
3006 .vece = MO_16 }, \
3007 { .fni4 = gen_##NAME##0_i32, \
3008 .fniv = gen_##NAME##0_vec, \
3009 .opt_opc = vecop_list_cmp, \
3010 .vece = MO_32 }, \
3011 { .fni8 = gen_##NAME##0_i64, \
3012 .fniv = gen_##NAME##0_vec, \
3013 .opt_opc = vecop_list_cmp, \
3014 .prefer_i64 = TCG_TARGET_REG_BITS == 64, \
3015 .vece = MO_64 }, \
3016 }; \
3017 tcg_gen_gvec_2(d, m, opr_sz, max_sz, &op[vece]); \
3018 }
3019
3020 static const TCGOpcode vecop_list_cmp[] = {
3021 INDEX_op_cmp_vec, 0
3022 };
3023
3024 GEN_CMP0(ceq, TCG_COND_EQ)
3025 GEN_CMP0(cle, TCG_COND_LE)
3026 GEN_CMP0(cge, TCG_COND_GE)
3027 GEN_CMP0(clt, TCG_COND_LT)
3028 GEN_CMP0(cgt, TCG_COND_GT)
3029
3030 #undef GEN_CMP0
3031
3032 static void gen_ssra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3033 {
3034 tcg_gen_vec_sar8i_i64(a, a, shift);
3035 tcg_gen_vec_add8_i64(d, d, a);
3036 }
3037
3038 static void gen_ssra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3039 {
3040 tcg_gen_vec_sar16i_i64(a, a, shift);
3041 tcg_gen_vec_add16_i64(d, d, a);
3042 }
3043
3044 static void gen_ssra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
3045 {
3046 tcg_gen_sari_i32(a, a, shift);
3047 tcg_gen_add_i32(d, d, a);
3048 }
3049
3050 static void gen_ssra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3051 {
3052 tcg_gen_sari_i64(a, a, shift);
3053 tcg_gen_add_i64(d, d, a);
3054 }
3055
3056 static void gen_ssra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3057 {
3058 tcg_gen_sari_vec(vece, a, a, sh);
3059 tcg_gen_add_vec(vece, d, d, a);
3060 }
3061
3062 void gen_gvec_ssra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3063 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3064 {
3065 static const TCGOpcode vecop_list[] = {
3066 INDEX_op_sari_vec, INDEX_op_add_vec, 0
3067 };
3068 static const GVecGen2i ops[4] = {
3069 { .fni8 = gen_ssra8_i64,
3070 .fniv = gen_ssra_vec,
3071 .fno = gen_helper_gvec_ssra_b,
3072 .load_dest = true,
3073 .opt_opc = vecop_list,
3074 .vece = MO_8 },
3075 { .fni8 = gen_ssra16_i64,
3076 .fniv = gen_ssra_vec,
3077 .fno = gen_helper_gvec_ssra_h,
3078 .load_dest = true,
3079 .opt_opc = vecop_list,
3080 .vece = MO_16 },
3081 { .fni4 = gen_ssra32_i32,
3082 .fniv = gen_ssra_vec,
3083 .fno = gen_helper_gvec_ssra_s,
3084 .load_dest = true,
3085 .opt_opc = vecop_list,
3086 .vece = MO_32 },
3087 { .fni8 = gen_ssra64_i64,
3088 .fniv = gen_ssra_vec,
3089 .fno = gen_helper_gvec_ssra_b,
3090 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3091 .opt_opc = vecop_list,
3092 .load_dest = true,
3093 .vece = MO_64 },
3094 };
3095
3096 /* tszimm encoding produces immediates in the range [1..esize]. */
3097 tcg_debug_assert(shift > 0);
3098 tcg_debug_assert(shift <= (8 << vece));
3099
3100 /*
3101 * Shifts larger than the element size are architecturally valid.
3102 * Signed results in all sign bits.
3103 */
3104 shift = MIN(shift, (8 << vece) - 1);
3105 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3106 }
3107
3108 static void gen_usra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3109 {
3110 tcg_gen_vec_shr8i_i64(a, a, shift);
3111 tcg_gen_vec_add8_i64(d, d, a);
3112 }
3113
3114 static void gen_usra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3115 {
3116 tcg_gen_vec_shr16i_i64(a, a, shift);
3117 tcg_gen_vec_add16_i64(d, d, a);
3118 }
3119
3120 static void gen_usra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
3121 {
3122 tcg_gen_shri_i32(a, a, shift);
3123 tcg_gen_add_i32(d, d, a);
3124 }
3125
3126 static void gen_usra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3127 {
3128 tcg_gen_shri_i64(a, a, shift);
3129 tcg_gen_add_i64(d, d, a);
3130 }
3131
3132 static void gen_usra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3133 {
3134 tcg_gen_shri_vec(vece, a, a, sh);
3135 tcg_gen_add_vec(vece, d, d, a);
3136 }
3137
3138 void gen_gvec_usra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3139 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3140 {
3141 static const TCGOpcode vecop_list[] = {
3142 INDEX_op_shri_vec, INDEX_op_add_vec, 0
3143 };
3144 static const GVecGen2i ops[4] = {
3145 { .fni8 = gen_usra8_i64,
3146 .fniv = gen_usra_vec,
3147 .fno = gen_helper_gvec_usra_b,
3148 .load_dest = true,
3149 .opt_opc = vecop_list,
3150 .vece = MO_8, },
3151 { .fni8 = gen_usra16_i64,
3152 .fniv = gen_usra_vec,
3153 .fno = gen_helper_gvec_usra_h,
3154 .load_dest = true,
3155 .opt_opc = vecop_list,
3156 .vece = MO_16, },
3157 { .fni4 = gen_usra32_i32,
3158 .fniv = gen_usra_vec,
3159 .fno = gen_helper_gvec_usra_s,
3160 .load_dest = true,
3161 .opt_opc = vecop_list,
3162 .vece = MO_32, },
3163 { .fni8 = gen_usra64_i64,
3164 .fniv = gen_usra_vec,
3165 .fno = gen_helper_gvec_usra_d,
3166 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3167 .load_dest = true,
3168 .opt_opc = vecop_list,
3169 .vece = MO_64, },
3170 };
3171
3172 /* tszimm encoding produces immediates in the range [1..esize]. */
3173 tcg_debug_assert(shift > 0);
3174 tcg_debug_assert(shift <= (8 << vece));
3175
3176 /*
3177 * Shifts larger than the element size are architecturally valid.
3178 * Unsigned results in all zeros as input to accumulate: nop.
3179 */
3180 if (shift < (8 << vece)) {
3181 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3182 } else {
3183 /* Nop, but we do need to clear the tail. */
3184 tcg_gen_gvec_mov(vece, rd_ofs, rd_ofs, opr_sz, max_sz);
3185 }
3186 }
3187
3188 /*
3189 * Shift one less than the requested amount, and the low bit is
3190 * the rounding bit. For the 8 and 16-bit operations, because we
3191 * mask the low bit, we can perform a normal integer shift instead
3192 * of a vector shift.
3193 */
3194 static void gen_srshr8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3195 {
3196 TCGv_i64 t = tcg_temp_new_i64();
3197
3198 tcg_gen_shri_i64(t, a, sh - 1);
3199 tcg_gen_andi_i64(t, t, dup_const(MO_8, 1));
3200 tcg_gen_vec_sar8i_i64(d, a, sh);
3201 tcg_gen_vec_add8_i64(d, d, t);
3202 tcg_temp_free_i64(t);
3203 }
3204
3205 static void gen_srshr16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3206 {
3207 TCGv_i64 t = tcg_temp_new_i64();
3208
3209 tcg_gen_shri_i64(t, a, sh - 1);
3210 tcg_gen_andi_i64(t, t, dup_const(MO_16, 1));
3211 tcg_gen_vec_sar16i_i64(d, a, sh);
3212 tcg_gen_vec_add16_i64(d, d, t);
3213 tcg_temp_free_i64(t);
3214 }
3215
3216 static void gen_srshr32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
3217 {
3218 TCGv_i32 t;
3219
3220 /* Handle shift by the input size for the benefit of trans_SRSHR_ri */
3221 if (sh == 32) {
3222 tcg_gen_movi_i32(d, 0);
3223 return;
3224 }
3225 t = tcg_temp_new_i32();
3226 tcg_gen_extract_i32(t, a, sh - 1, 1);
3227 tcg_gen_sari_i32(d, a, sh);
3228 tcg_gen_add_i32(d, d, t);
3229 tcg_temp_free_i32(t);
3230 }
3231
3232 static void gen_srshr64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3233 {
3234 TCGv_i64 t = tcg_temp_new_i64();
3235
3236 tcg_gen_extract_i64(t, a, sh - 1, 1);
3237 tcg_gen_sari_i64(d, a, sh);
3238 tcg_gen_add_i64(d, d, t);
3239 tcg_temp_free_i64(t);
3240 }
3241
3242 static void gen_srshr_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3243 {
3244 TCGv_vec t = tcg_temp_new_vec_matching(d);
3245 TCGv_vec ones = tcg_temp_new_vec_matching(d);
3246
3247 tcg_gen_shri_vec(vece, t, a, sh - 1);
3248 tcg_gen_dupi_vec(vece, ones, 1);
3249 tcg_gen_and_vec(vece, t, t, ones);
3250 tcg_gen_sari_vec(vece, d, a, sh);
3251 tcg_gen_add_vec(vece, d, d, t);
3252
3253 tcg_temp_free_vec(t);
3254 tcg_temp_free_vec(ones);
3255 }
3256
3257 void gen_gvec_srshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3258 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3259 {
3260 static const TCGOpcode vecop_list[] = {
3261 INDEX_op_shri_vec, INDEX_op_sari_vec, INDEX_op_add_vec, 0
3262 };
3263 static const GVecGen2i ops[4] = {
3264 { .fni8 = gen_srshr8_i64,
3265 .fniv = gen_srshr_vec,
3266 .fno = gen_helper_gvec_srshr_b,
3267 .opt_opc = vecop_list,
3268 .vece = MO_8 },
3269 { .fni8 = gen_srshr16_i64,
3270 .fniv = gen_srshr_vec,
3271 .fno = gen_helper_gvec_srshr_h,
3272 .opt_opc = vecop_list,
3273 .vece = MO_16 },
3274 { .fni4 = gen_srshr32_i32,
3275 .fniv = gen_srshr_vec,
3276 .fno = gen_helper_gvec_srshr_s,
3277 .opt_opc = vecop_list,
3278 .vece = MO_32 },
3279 { .fni8 = gen_srshr64_i64,
3280 .fniv = gen_srshr_vec,
3281 .fno = gen_helper_gvec_srshr_d,
3282 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3283 .opt_opc = vecop_list,
3284 .vece = MO_64 },
3285 };
3286
3287 /* tszimm encoding produces immediates in the range [1..esize] */
3288 tcg_debug_assert(shift > 0);
3289 tcg_debug_assert(shift <= (8 << vece));
3290
3291 if (shift == (8 << vece)) {
3292 /*
3293 * Shifts larger than the element size are architecturally valid.
3294 * Signed results in all sign bits. With rounding, this produces
3295 * (-1 + 1) >> 1 == 0, or (0 + 1) >> 1 == 0.
3296 * I.e. always zero.
3297 */
3298 tcg_gen_gvec_dup_imm(vece, rd_ofs, opr_sz, max_sz, 0);
3299 } else {
3300 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3301 }
3302 }
3303
3304 static void gen_srsra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3305 {
3306 TCGv_i64 t = tcg_temp_new_i64();
3307
3308 gen_srshr8_i64(t, a, sh);
3309 tcg_gen_vec_add8_i64(d, d, t);
3310 tcg_temp_free_i64(t);
3311 }
3312
3313 static void gen_srsra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3314 {
3315 TCGv_i64 t = tcg_temp_new_i64();
3316
3317 gen_srshr16_i64(t, a, sh);
3318 tcg_gen_vec_add16_i64(d, d, t);
3319 tcg_temp_free_i64(t);
3320 }
3321
3322 static void gen_srsra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
3323 {
3324 TCGv_i32 t = tcg_temp_new_i32();
3325
3326 gen_srshr32_i32(t, a, sh);
3327 tcg_gen_add_i32(d, d, t);
3328 tcg_temp_free_i32(t);
3329 }
3330
3331 static void gen_srsra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3332 {
3333 TCGv_i64 t = tcg_temp_new_i64();
3334
3335 gen_srshr64_i64(t, a, sh);
3336 tcg_gen_add_i64(d, d, t);
3337 tcg_temp_free_i64(t);
3338 }
3339
3340 static void gen_srsra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3341 {
3342 TCGv_vec t = tcg_temp_new_vec_matching(d);
3343
3344 gen_srshr_vec(vece, t, a, sh);
3345 tcg_gen_add_vec(vece, d, d, t);
3346 tcg_temp_free_vec(t);
3347 }
3348
3349 void gen_gvec_srsra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3350 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3351 {
3352 static const TCGOpcode vecop_list[] = {
3353 INDEX_op_shri_vec, INDEX_op_sari_vec, INDEX_op_add_vec, 0
3354 };
3355 static const GVecGen2i ops[4] = {
3356 { .fni8 = gen_srsra8_i64,
3357 .fniv = gen_srsra_vec,
3358 .fno = gen_helper_gvec_srsra_b,
3359 .opt_opc = vecop_list,
3360 .load_dest = true,
3361 .vece = MO_8 },
3362 { .fni8 = gen_srsra16_i64,
3363 .fniv = gen_srsra_vec,
3364 .fno = gen_helper_gvec_srsra_h,
3365 .opt_opc = vecop_list,
3366 .load_dest = true,
3367 .vece = MO_16 },
3368 { .fni4 = gen_srsra32_i32,
3369 .fniv = gen_srsra_vec,
3370 .fno = gen_helper_gvec_srsra_s,
3371 .opt_opc = vecop_list,
3372 .load_dest = true,
3373 .vece = MO_32 },
3374 { .fni8 = gen_srsra64_i64,
3375 .fniv = gen_srsra_vec,
3376 .fno = gen_helper_gvec_srsra_d,
3377 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3378 .opt_opc = vecop_list,
3379 .load_dest = true,
3380 .vece = MO_64 },
3381 };
3382
3383 /* tszimm encoding produces immediates in the range [1..esize] */
3384 tcg_debug_assert(shift > 0);
3385 tcg_debug_assert(shift <= (8 << vece));
3386
3387 /*
3388 * Shifts larger than the element size are architecturally valid.
3389 * Signed results in all sign bits. With rounding, this produces
3390 * (-1 + 1) >> 1 == 0, or (0 + 1) >> 1 == 0.
3391 * I.e. always zero. With accumulation, this leaves D unchanged.
3392 */
3393 if (shift == (8 << vece)) {
3394 /* Nop, but we do need to clear the tail. */
3395 tcg_gen_gvec_mov(vece, rd_ofs, rd_ofs, opr_sz, max_sz);
3396 } else {
3397 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3398 }
3399 }
3400
3401 static void gen_urshr8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3402 {
3403 TCGv_i64 t = tcg_temp_new_i64();
3404
3405 tcg_gen_shri_i64(t, a, sh - 1);
3406 tcg_gen_andi_i64(t, t, dup_const(MO_8, 1));
3407 tcg_gen_vec_shr8i_i64(d, a, sh);
3408 tcg_gen_vec_add8_i64(d, d, t);
3409 tcg_temp_free_i64(t);
3410 }
3411
3412 static void gen_urshr16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3413 {
3414 TCGv_i64 t = tcg_temp_new_i64();
3415
3416 tcg_gen_shri_i64(t, a, sh - 1);
3417 tcg_gen_andi_i64(t, t, dup_const(MO_16, 1));
3418 tcg_gen_vec_shr16i_i64(d, a, sh);
3419 tcg_gen_vec_add16_i64(d, d, t);
3420 tcg_temp_free_i64(t);
3421 }
3422
3423 static void gen_urshr32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
3424 {
3425 TCGv_i32 t;
3426
3427 /* Handle shift by the input size for the benefit of trans_URSHR_ri */
3428 if (sh == 32) {
3429 tcg_gen_extract_i32(d, a, sh - 1, 1);
3430 return;
3431 }
3432 t = tcg_temp_new_i32();
3433 tcg_gen_extract_i32(t, a, sh - 1, 1);
3434 tcg_gen_shri_i32(d, a, sh);
3435 tcg_gen_add_i32(d, d, t);
3436 tcg_temp_free_i32(t);
3437 }
3438
3439 static void gen_urshr64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3440 {
3441 TCGv_i64 t = tcg_temp_new_i64();
3442
3443 tcg_gen_extract_i64(t, a, sh - 1, 1);
3444 tcg_gen_shri_i64(d, a, sh);
3445 tcg_gen_add_i64(d, d, t);
3446 tcg_temp_free_i64(t);
3447 }
3448
3449 static void gen_urshr_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t shift)
3450 {
3451 TCGv_vec t = tcg_temp_new_vec_matching(d);
3452 TCGv_vec ones = tcg_temp_new_vec_matching(d);
3453
3454 tcg_gen_shri_vec(vece, t, a, shift - 1);
3455 tcg_gen_dupi_vec(vece, ones, 1);
3456 tcg_gen_and_vec(vece, t, t, ones);
3457 tcg_gen_shri_vec(vece, d, a, shift);
3458 tcg_gen_add_vec(vece, d, d, t);
3459
3460 tcg_temp_free_vec(t);
3461 tcg_temp_free_vec(ones);
3462 }
3463
3464 void gen_gvec_urshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3465 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3466 {
3467 static const TCGOpcode vecop_list[] = {
3468 INDEX_op_shri_vec, INDEX_op_add_vec, 0
3469 };
3470 static const GVecGen2i ops[4] = {
3471 { .fni8 = gen_urshr8_i64,
3472 .fniv = gen_urshr_vec,
3473 .fno = gen_helper_gvec_urshr_b,
3474 .opt_opc = vecop_list,
3475 .vece = MO_8 },
3476 { .fni8 = gen_urshr16_i64,
3477 .fniv = gen_urshr_vec,
3478 .fno = gen_helper_gvec_urshr_h,
3479 .opt_opc = vecop_list,
3480 .vece = MO_16 },
3481 { .fni4 = gen_urshr32_i32,
3482 .fniv = gen_urshr_vec,
3483 .fno = gen_helper_gvec_urshr_s,
3484 .opt_opc = vecop_list,
3485 .vece = MO_32 },
3486 { .fni8 = gen_urshr64_i64,
3487 .fniv = gen_urshr_vec,
3488 .fno = gen_helper_gvec_urshr_d,
3489 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3490 .opt_opc = vecop_list,
3491 .vece = MO_64 },
3492 };
3493
3494 /* tszimm encoding produces immediates in the range [1..esize] */
3495 tcg_debug_assert(shift > 0);
3496 tcg_debug_assert(shift <= (8 << vece));
3497
3498 if (shift == (8 << vece)) {
3499 /*
3500 * Shifts larger than the element size are architecturally valid.
3501 * Unsigned results in zero. With rounding, this produces a
3502 * copy of the most significant bit.
3503 */
3504 tcg_gen_gvec_shri(vece, rd_ofs, rm_ofs, shift - 1, opr_sz, max_sz);
3505 } else {
3506 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3507 }
3508 }
3509
3510 static void gen_ursra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3511 {
3512 TCGv_i64 t = tcg_temp_new_i64();
3513
3514 if (sh == 8) {
3515 tcg_gen_vec_shr8i_i64(t, a, 7);
3516 } else {
3517 gen_urshr8_i64(t, a, sh);
3518 }
3519 tcg_gen_vec_add8_i64(d, d, t);
3520 tcg_temp_free_i64(t);
3521 }
3522
3523 static void gen_ursra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3524 {
3525 TCGv_i64 t = tcg_temp_new_i64();
3526
3527 if (sh == 16) {
3528 tcg_gen_vec_shr16i_i64(t, a, 15);
3529 } else {
3530 gen_urshr16_i64(t, a, sh);
3531 }
3532 tcg_gen_vec_add16_i64(d, d, t);
3533 tcg_temp_free_i64(t);
3534 }
3535
3536 static void gen_ursra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
3537 {
3538 TCGv_i32 t = tcg_temp_new_i32();
3539
3540 if (sh == 32) {
3541 tcg_gen_shri_i32(t, a, 31);
3542 } else {
3543 gen_urshr32_i32(t, a, sh);
3544 }
3545 tcg_gen_add_i32(d, d, t);
3546 tcg_temp_free_i32(t);
3547 }
3548
3549 static void gen_ursra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3550 {
3551 TCGv_i64 t = tcg_temp_new_i64();
3552
3553 if (sh == 64) {
3554 tcg_gen_shri_i64(t, a, 63);
3555 } else {
3556 gen_urshr64_i64(t, a, sh);
3557 }
3558 tcg_gen_add_i64(d, d, t);
3559 tcg_temp_free_i64(t);
3560 }
3561
3562 static void gen_ursra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3563 {
3564 TCGv_vec t = tcg_temp_new_vec_matching(d);
3565
3566 if (sh == (8 << vece)) {
3567 tcg_gen_shri_vec(vece, t, a, sh - 1);
3568 } else {
3569 gen_urshr_vec(vece, t, a, sh);
3570 }
3571 tcg_gen_add_vec(vece, d, d, t);
3572 tcg_temp_free_vec(t);
3573 }
3574
3575 void gen_gvec_ursra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3576 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3577 {
3578 static const TCGOpcode vecop_list[] = {
3579 INDEX_op_shri_vec, INDEX_op_add_vec, 0
3580 };
3581 static const GVecGen2i ops[4] = {
3582 { .fni8 = gen_ursra8_i64,
3583 .fniv = gen_ursra_vec,
3584 .fno = gen_helper_gvec_ursra_b,
3585 .opt_opc = vecop_list,
3586 .load_dest = true,
3587 .vece = MO_8 },
3588 { .fni8 = gen_ursra16_i64,
3589 .fniv = gen_ursra_vec,
3590 .fno = gen_helper_gvec_ursra_h,
3591 .opt_opc = vecop_list,
3592 .load_dest = true,
3593 .vece = MO_16 },
3594 { .fni4 = gen_ursra32_i32,
3595 .fniv = gen_ursra_vec,
3596 .fno = gen_helper_gvec_ursra_s,
3597 .opt_opc = vecop_list,
3598 .load_dest = true,
3599 .vece = MO_32 },
3600 { .fni8 = gen_ursra64_i64,
3601 .fniv = gen_ursra_vec,
3602 .fno = gen_helper_gvec_ursra_d,
3603 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3604 .opt_opc = vecop_list,
3605 .load_dest = true,
3606 .vece = MO_64 },
3607 };
3608
3609 /* tszimm encoding produces immediates in the range [1..esize] */
3610 tcg_debug_assert(shift > 0);
3611 tcg_debug_assert(shift <= (8 << vece));
3612
3613 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3614 }
3615
3616 static void gen_shr8_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3617 {
3618 uint64_t mask = dup_const(MO_8, 0xff >> shift);
3619 TCGv_i64 t = tcg_temp_new_i64();
3620
3621 tcg_gen_shri_i64(t, a, shift);
3622 tcg_gen_andi_i64(t, t, mask);
3623 tcg_gen_andi_i64(d, d, ~mask);
3624 tcg_gen_or_i64(d, d, t);
3625 tcg_temp_free_i64(t);
3626 }
3627
3628 static void gen_shr16_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3629 {
3630 uint64_t mask = dup_const(MO_16, 0xffff >> shift);
3631 TCGv_i64 t = tcg_temp_new_i64();
3632
3633 tcg_gen_shri_i64(t, a, shift);
3634 tcg_gen_andi_i64(t, t, mask);
3635 tcg_gen_andi_i64(d, d, ~mask);
3636 tcg_gen_or_i64(d, d, t);
3637 tcg_temp_free_i64(t);
3638 }
3639
3640 static void gen_shr32_ins_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
3641 {
3642 tcg_gen_shri_i32(a, a, shift);
3643 tcg_gen_deposit_i32(d, d, a, 0, 32 - shift);
3644 }
3645
3646 static void gen_shr64_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3647 {
3648 tcg_gen_shri_i64(a, a, shift);
3649 tcg_gen_deposit_i64(d, d, a, 0, 64 - shift);
3650 }
3651
3652 static void gen_shr_ins_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3653 {
3654 TCGv_vec t = tcg_temp_new_vec_matching(d);
3655 TCGv_vec m = tcg_temp_new_vec_matching(d);
3656
3657 tcg_gen_dupi_vec(vece, m, MAKE_64BIT_MASK((8 << vece) - sh, sh));
3658 tcg_gen_shri_vec(vece, t, a, sh);
3659 tcg_gen_and_vec(vece, d, d, m);
3660 tcg_gen_or_vec(vece, d, d, t);
3661
3662 tcg_temp_free_vec(t);
3663 tcg_temp_free_vec(m);
3664 }
3665
3666 void gen_gvec_sri(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3667 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3668 {
3669 static const TCGOpcode vecop_list[] = { INDEX_op_shri_vec, 0 };
3670 const GVecGen2i ops[4] = {
3671 { .fni8 = gen_shr8_ins_i64,
3672 .fniv = gen_shr_ins_vec,
3673 .fno = gen_helper_gvec_sri_b,
3674 .load_dest = true,
3675 .opt_opc = vecop_list,
3676 .vece = MO_8 },
3677 { .fni8 = gen_shr16_ins_i64,
3678 .fniv = gen_shr_ins_vec,
3679 .fno = gen_helper_gvec_sri_h,
3680 .load_dest = true,
3681 .opt_opc = vecop_list,
3682 .vece = MO_16 },
3683 { .fni4 = gen_shr32_ins_i32,
3684 .fniv = gen_shr_ins_vec,
3685 .fno = gen_helper_gvec_sri_s,
3686 .load_dest = true,
3687 .opt_opc = vecop_list,
3688 .vece = MO_32 },
3689 { .fni8 = gen_shr64_ins_i64,
3690 .fniv = gen_shr_ins_vec,
3691 .fno = gen_helper_gvec_sri_d,
3692 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3693 .load_dest = true,
3694 .opt_opc = vecop_list,
3695 .vece = MO_64 },
3696 };
3697
3698 /* tszimm encoding produces immediates in the range [1..esize]. */
3699 tcg_debug_assert(shift > 0);
3700 tcg_debug_assert(shift <= (8 << vece));
3701
3702 /* Shift of esize leaves destination unchanged. */
3703 if (shift < (8 << vece)) {
3704 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3705 } else {
3706 /* Nop, but we do need to clear the tail. */
3707 tcg_gen_gvec_mov(vece, rd_ofs, rd_ofs, opr_sz, max_sz);
3708 }
3709 }
3710
3711 static void gen_shl8_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3712 {
3713 uint64_t mask = dup_const(MO_8, 0xff << shift);
3714 TCGv_i64 t = tcg_temp_new_i64();
3715
3716 tcg_gen_shli_i64(t, a, shift);
3717 tcg_gen_andi_i64(t, t, mask);
3718 tcg_gen_andi_i64(d, d, ~mask);
3719 tcg_gen_or_i64(d, d, t);
3720 tcg_temp_free_i64(t);
3721 }
3722
3723 static void gen_shl16_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3724 {
3725 uint64_t mask = dup_const(MO_16, 0xffff << shift);
3726 TCGv_i64 t = tcg_temp_new_i64();
3727
3728 tcg_gen_shli_i64(t, a, shift);
3729 tcg_gen_andi_i64(t, t, mask);
3730 tcg_gen_andi_i64(d, d, ~mask);
3731 tcg_gen_or_i64(d, d, t);
3732 tcg_temp_free_i64(t);
3733 }
3734
3735 static void gen_shl32_ins_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
3736 {
3737 tcg_gen_deposit_i32(d, d, a, shift, 32 - shift);
3738 }
3739
3740 static void gen_shl64_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3741 {
3742 tcg_gen_deposit_i64(d, d, a, shift, 64 - shift);
3743 }
3744
3745 static void gen_shl_ins_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3746 {
3747 TCGv_vec t = tcg_temp_new_vec_matching(d);
3748 TCGv_vec m = tcg_temp_new_vec_matching(d);
3749
3750 tcg_gen_shli_vec(vece, t, a, sh);
3751 tcg_gen_dupi_vec(vece, m, MAKE_64BIT_MASK(0, sh));
3752 tcg_gen_and_vec(vece, d, d, m);
3753 tcg_gen_or_vec(vece, d, d, t);
3754
3755 tcg_temp_free_vec(t);
3756 tcg_temp_free_vec(m);
3757 }
3758
3759 void gen_gvec_sli(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3760 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3761 {
3762 static const TCGOpcode vecop_list[] = { INDEX_op_shli_vec, 0 };
3763 const GVecGen2i ops[4] = {
3764 { .fni8 = gen_shl8_ins_i64,
3765 .fniv = gen_shl_ins_vec,
3766 .fno = gen_helper_gvec_sli_b,
3767 .load_dest = true,
3768 .opt_opc = vecop_list,
3769 .vece = MO_8 },
3770 { .fni8 = gen_shl16_ins_i64,
3771 .fniv = gen_shl_ins_vec,
3772 .fno = gen_helper_gvec_sli_h,
3773 .load_dest = true,
3774 .opt_opc = vecop_list,
3775 .vece = MO_16 },
3776 { .fni4 = gen_shl32_ins_i32,
3777 .fniv = gen_shl_ins_vec,
3778 .fno = gen_helper_gvec_sli_s,
3779 .load_dest = true,
3780 .opt_opc = vecop_list,
3781 .vece = MO_32 },
3782 { .fni8 = gen_shl64_ins_i64,
3783 .fniv = gen_shl_ins_vec,
3784 .fno = gen_helper_gvec_sli_d,
3785 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3786 .load_dest = true,
3787 .opt_opc = vecop_list,
3788 .vece = MO_64 },
3789 };
3790
3791 /* tszimm encoding produces immediates in the range [0..esize-1]. */
3792 tcg_debug_assert(shift >= 0);
3793 tcg_debug_assert(shift < (8 << vece));
3794
3795 if (shift == 0) {
3796 tcg_gen_gvec_mov(vece, rd_ofs, rm_ofs, opr_sz, max_sz);
3797 } else {
3798 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3799 }
3800 }
3801
3802 static void gen_mla8_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3803 {
3804 gen_helper_neon_mul_u8(a, a, b);
3805 gen_helper_neon_add_u8(d, d, a);
3806 }
3807
3808 static void gen_mls8_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3809 {
3810 gen_helper_neon_mul_u8(a, a, b);
3811 gen_helper_neon_sub_u8(d, d, a);
3812 }
3813
3814 static void gen_mla16_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3815 {
3816 gen_helper_neon_mul_u16(a, a, b);
3817 gen_helper_neon_add_u16(d, d, a);
3818 }
3819
3820 static void gen_mls16_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3821 {
3822 gen_helper_neon_mul_u16(a, a, b);
3823 gen_helper_neon_sub_u16(d, d, a);
3824 }
3825
3826 static void gen_mla32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3827 {
3828 tcg_gen_mul_i32(a, a, b);
3829 tcg_gen_add_i32(d, d, a);
3830 }
3831
3832 static void gen_mls32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3833 {
3834 tcg_gen_mul_i32(a, a, b);
3835 tcg_gen_sub_i32(d, d, a);
3836 }
3837
3838 static void gen_mla64_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
3839 {
3840 tcg_gen_mul_i64(a, a, b);
3841 tcg_gen_add_i64(d, d, a);
3842 }
3843
3844 static void gen_mls64_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
3845 {
3846 tcg_gen_mul_i64(a, a, b);
3847 tcg_gen_sub_i64(d, d, a);
3848 }
3849
3850 static void gen_mla_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
3851 {
3852 tcg_gen_mul_vec(vece, a, a, b);
3853 tcg_gen_add_vec(vece, d, d, a);
3854 }
3855
3856 static void gen_mls_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
3857 {
3858 tcg_gen_mul_vec(vece, a, a, b);
3859 tcg_gen_sub_vec(vece, d, d, a);
3860 }
3861
3862 /* Note that while NEON does not support VMLA and VMLS as 64-bit ops,
3863 * these tables are shared with AArch64 which does support them.
3864 */
3865 void gen_gvec_mla(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
3866 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
3867 {
3868 static const TCGOpcode vecop_list[] = {
3869 INDEX_op_mul_vec, INDEX_op_add_vec, 0
3870 };
3871 static const GVecGen3 ops[4] = {
3872 { .fni4 = gen_mla8_i32,
3873 .fniv = gen_mla_vec,
3874 .load_dest = true,
3875 .opt_opc = vecop_list,
3876 .vece = MO_8 },
3877 { .fni4 = gen_mla16_i32,
3878 .fniv = gen_mla_vec,
3879 .load_dest = true,
3880 .opt_opc = vecop_list,
3881 .vece = MO_16 },
3882 { .fni4 = gen_mla32_i32,
3883 .fniv = gen_mla_vec,
3884 .load_dest = true,
3885 .opt_opc = vecop_list,
3886 .vece = MO_32 },
3887 { .fni8 = gen_mla64_i64,
3888 .fniv = gen_mla_vec,
3889 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3890 .load_dest = true,
3891 .opt_opc = vecop_list,
3892 .vece = MO_64 },
3893 };
3894 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
3895 }
3896
3897 void gen_gvec_mls(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
3898 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
3899 {
3900 static const TCGOpcode vecop_list[] = {
3901 INDEX_op_mul_vec, INDEX_op_sub_vec, 0
3902 };
3903 static const GVecGen3 ops[4] = {
3904 { .fni4 = gen_mls8_i32,
3905 .fniv = gen_mls_vec,
3906 .load_dest = true,
3907 .opt_opc = vecop_list,
3908 .vece = MO_8 },
3909 { .fni4 = gen_mls16_i32,
3910 .fniv = gen_mls_vec,
3911 .load_dest = true,
3912 .opt_opc = vecop_list,
3913 .vece = MO_16 },
3914 { .fni4 = gen_mls32_i32,
3915 .fniv = gen_mls_vec,
3916 .load_dest = true,
3917 .opt_opc = vecop_list,
3918 .vece = MO_32 },
3919 { .fni8 = gen_mls64_i64,
3920 .fniv = gen_mls_vec,
3921 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3922 .load_dest = true,
3923 .opt_opc = vecop_list,
3924 .vece = MO_64 },
3925 };
3926 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
3927 }
3928
3929 /* CMTST : test is "if (X & Y != 0)". */
3930 static void gen_cmtst_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3931 {
3932 tcg_gen_and_i32(d, a, b);
3933 tcg_gen_setcondi_i32(TCG_COND_NE, d, d, 0);
3934 tcg_gen_neg_i32(d, d);
3935 }
3936
3937 void gen_cmtst_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
3938 {
3939 tcg_gen_and_i64(d, a, b);
3940 tcg_gen_setcondi_i64(TCG_COND_NE, d, d, 0);
3941 tcg_gen_neg_i64(d, d);
3942 }
3943
3944 static void gen_cmtst_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
3945 {
3946 tcg_gen_and_vec(vece, d, a, b);
3947 tcg_gen_dupi_vec(vece, a, 0);
3948 tcg_gen_cmp_vec(TCG_COND_NE, vece, d, d, a);
3949 }
3950
3951 void gen_gvec_cmtst(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
3952 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
3953 {
3954 static const TCGOpcode vecop_list[] = { INDEX_op_cmp_vec, 0 };
3955 static const GVecGen3 ops[4] = {
3956 { .fni4 = gen_helper_neon_tst_u8,
3957 .fniv = gen_cmtst_vec,
3958 .opt_opc = vecop_list,
3959 .vece = MO_8 },
3960 { .fni4 = gen_helper_neon_tst_u16,
3961 .fniv = gen_cmtst_vec,
3962 .opt_opc = vecop_list,
3963 .vece = MO_16 },
3964 { .fni4 = gen_cmtst_i32,
3965 .fniv = gen_cmtst_vec,
3966 .opt_opc = vecop_list,
3967 .vece = MO_32 },
3968 { .fni8 = gen_cmtst_i64,
3969 .fniv = gen_cmtst_vec,
3970 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3971 .opt_opc = vecop_list,
3972 .vece = MO_64 },
3973 };
3974 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
3975 }
3976
3977 void gen_ushl_i32(TCGv_i32 dst, TCGv_i32 src, TCGv_i32 shift)
3978 {
3979 TCGv_i32 lval = tcg_temp_new_i32();
3980 TCGv_i32 rval = tcg_temp_new_i32();
3981 TCGv_i32 lsh = tcg_temp_new_i32();
3982 TCGv_i32 rsh = tcg_temp_new_i32();
3983 TCGv_i32 zero = tcg_const_i32(0);
3984 TCGv_i32 max = tcg_const_i32(32);
3985
3986 /*
3987 * Rely on the TCG guarantee that out of range shifts produce
3988 * unspecified results, not undefined behaviour (i.e. no trap).
3989 * Discard out-of-range results after the fact.
3990 */
3991 tcg_gen_ext8s_i32(lsh, shift);
3992 tcg_gen_neg_i32(rsh, lsh);
3993 tcg_gen_shl_i32(lval, src, lsh);
3994 tcg_gen_shr_i32(rval, src, rsh);
3995 tcg_gen_movcond_i32(TCG_COND_LTU, dst, lsh, max, lval, zero);
3996 tcg_gen_movcond_i32(TCG_COND_LTU, dst, rsh, max, rval, dst);
3997
3998 tcg_temp_free_i32(lval);
3999 tcg_temp_free_i32(rval);
4000 tcg_temp_free_i32(lsh);
4001 tcg_temp_free_i32(rsh);
4002 tcg_temp_free_i32(zero);
4003 tcg_temp_free_i32(max);
4004 }
4005
4006 void gen_ushl_i64(TCGv_i64 dst, TCGv_i64 src, TCGv_i64 shift)
4007 {
4008 TCGv_i64 lval = tcg_temp_new_i64();
4009 TCGv_i64 rval = tcg_temp_new_i64();
4010 TCGv_i64 lsh = tcg_temp_new_i64();
4011 TCGv_i64 rsh = tcg_temp_new_i64();
4012 TCGv_i64 zero = tcg_const_i64(0);
4013 TCGv_i64 max = tcg_const_i64(64);
4014
4015 /*
4016 * Rely on the TCG guarantee that out of range shifts produce
4017 * unspecified results, not undefined behaviour (i.e. no trap).
4018 * Discard out-of-range results after the fact.
4019 */
4020 tcg_gen_ext8s_i64(lsh, shift);
4021 tcg_gen_neg_i64(rsh, lsh);
4022 tcg_gen_shl_i64(lval, src, lsh);
4023 tcg_gen_shr_i64(rval, src, rsh);
4024 tcg_gen_movcond_i64(TCG_COND_LTU, dst, lsh, max, lval, zero);
4025 tcg_gen_movcond_i64(TCG_COND_LTU, dst, rsh, max, rval, dst);
4026
4027 tcg_temp_free_i64(lval);
4028 tcg_temp_free_i64(rval);
4029 tcg_temp_free_i64(lsh);
4030 tcg_temp_free_i64(rsh);
4031 tcg_temp_free_i64(zero);
4032 tcg_temp_free_i64(max);
4033 }
4034
4035 static void gen_ushl_vec(unsigned vece, TCGv_vec dst,
4036 TCGv_vec src, TCGv_vec shift)
4037 {
4038 TCGv_vec lval = tcg_temp_new_vec_matching(dst);
4039 TCGv_vec rval = tcg_temp_new_vec_matching(dst);
4040 TCGv_vec lsh = tcg_temp_new_vec_matching(dst);
4041 TCGv_vec rsh = tcg_temp_new_vec_matching(dst);
4042 TCGv_vec msk, max;
4043
4044 tcg_gen_neg_vec(vece, rsh, shift);
4045 if (vece == MO_8) {
4046 tcg_gen_mov_vec(lsh, shift);
4047 } else {
4048 msk = tcg_temp_new_vec_matching(dst);
4049 tcg_gen_dupi_vec(vece, msk, 0xff);
4050 tcg_gen_and_vec(vece, lsh, shift, msk);
4051 tcg_gen_and_vec(vece, rsh, rsh, msk);
4052 tcg_temp_free_vec(msk);
4053 }
4054
4055 /*
4056 * Rely on the TCG guarantee that out of range shifts produce
4057 * unspecified results, not undefined behaviour (i.e. no trap).
4058 * Discard out-of-range results after the fact.
4059 */
4060 tcg_gen_shlv_vec(vece, lval, src, lsh);
4061 tcg_gen_shrv_vec(vece, rval, src, rsh);
4062
4063 max = tcg_temp_new_vec_matching(dst);
4064 tcg_gen_dupi_vec(vece, max, 8 << vece);
4065
4066 /*
4067 * The choice of LT (signed) and GEU (unsigned) are biased toward
4068 * the instructions of the x86_64 host. For MO_8, the whole byte
4069 * is significant so we must use an unsigned compare; otherwise we
4070 * have already masked to a byte and so a signed compare works.
4071 * Other tcg hosts have a full set of comparisons and do not care.
4072 */
4073 if (vece == MO_8) {
4074 tcg_gen_cmp_vec(TCG_COND_GEU, vece, lsh, lsh, max);
4075 tcg_gen_cmp_vec(TCG_COND_GEU, vece, rsh, rsh, max);
4076 tcg_gen_andc_vec(vece, lval, lval, lsh);
4077 tcg_gen_andc_vec(vece, rval, rval, rsh);
4078 } else {
4079 tcg_gen_cmp_vec(TCG_COND_LT, vece, lsh, lsh, max);
4080 tcg_gen_cmp_vec(TCG_COND_LT, vece, rsh, rsh, max);
4081 tcg_gen_and_vec(vece, lval, lval, lsh);
4082 tcg_gen_and_vec(vece, rval, rval, rsh);
4083 }
4084 tcg_gen_or_vec(vece, dst, lval, rval);
4085
4086 tcg_temp_free_vec(max);
4087 tcg_temp_free_vec(lval);
4088 tcg_temp_free_vec(rval);
4089 tcg_temp_free_vec(lsh);
4090 tcg_temp_free_vec(rsh);
4091 }
4092
4093 void gen_gvec_ushl(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4094 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4095 {
4096 static const TCGOpcode vecop_list[] = {
4097 INDEX_op_neg_vec, INDEX_op_shlv_vec,
4098 INDEX_op_shrv_vec, INDEX_op_cmp_vec, 0
4099 };
4100 static const GVecGen3 ops[4] = {
4101 { .fniv = gen_ushl_vec,
4102 .fno = gen_helper_gvec_ushl_b,
4103 .opt_opc = vecop_list,
4104 .vece = MO_8 },
4105 { .fniv = gen_ushl_vec,
4106 .fno = gen_helper_gvec_ushl_h,
4107 .opt_opc = vecop_list,
4108 .vece = MO_16 },
4109 { .fni4 = gen_ushl_i32,
4110 .fniv = gen_ushl_vec,
4111 .opt_opc = vecop_list,
4112 .vece = MO_32 },
4113 { .fni8 = gen_ushl_i64,
4114 .fniv = gen_ushl_vec,
4115 .opt_opc = vecop_list,
4116 .vece = MO_64 },
4117 };
4118 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4119 }
4120
4121 void gen_sshl_i32(TCGv_i32 dst, TCGv_i32 src, TCGv_i32 shift)
4122 {
4123 TCGv_i32 lval = tcg_temp_new_i32();
4124 TCGv_i32 rval = tcg_temp_new_i32();
4125 TCGv_i32 lsh = tcg_temp_new_i32();
4126 TCGv_i32 rsh = tcg_temp_new_i32();
4127 TCGv_i32 zero = tcg_const_i32(0);
4128 TCGv_i32 max = tcg_const_i32(31);
4129
4130 /*
4131 * Rely on the TCG guarantee that out of range shifts produce
4132 * unspecified results, not undefined behaviour (i.e. no trap).
4133 * Discard out-of-range results after the fact.
4134 */
4135 tcg_gen_ext8s_i32(lsh, shift);
4136 tcg_gen_neg_i32(rsh, lsh);
4137 tcg_gen_shl_i32(lval, src, lsh);
4138 tcg_gen_umin_i32(rsh, rsh, max);
4139 tcg_gen_sar_i32(rval, src, rsh);
4140 tcg_gen_movcond_i32(TCG_COND_LEU, lval, lsh, max, lval, zero);
4141 tcg_gen_movcond_i32(TCG_COND_LT, dst, lsh, zero, rval, lval);
4142
4143 tcg_temp_free_i32(lval);
4144 tcg_temp_free_i32(rval);
4145 tcg_temp_free_i32(lsh);
4146 tcg_temp_free_i32(rsh);
4147 tcg_temp_free_i32(zero);
4148 tcg_temp_free_i32(max);
4149 }
4150
4151 void gen_sshl_i64(TCGv_i64 dst, TCGv_i64 src, TCGv_i64 shift)
4152 {
4153 TCGv_i64 lval = tcg_temp_new_i64();
4154 TCGv_i64 rval = tcg_temp_new_i64();
4155 TCGv_i64 lsh = tcg_temp_new_i64();
4156 TCGv_i64 rsh = tcg_temp_new_i64();
4157 TCGv_i64 zero = tcg_const_i64(0);
4158 TCGv_i64 max = tcg_const_i64(63);
4159
4160 /*
4161 * Rely on the TCG guarantee that out of range shifts produce
4162 * unspecified results, not undefined behaviour (i.e. no trap).
4163 * Discard out-of-range results after the fact.
4164 */
4165 tcg_gen_ext8s_i64(lsh, shift);
4166 tcg_gen_neg_i64(rsh, lsh);
4167 tcg_gen_shl_i64(lval, src, lsh);
4168 tcg_gen_umin_i64(rsh, rsh, max);
4169 tcg_gen_sar_i64(rval, src, rsh);
4170 tcg_gen_movcond_i64(TCG_COND_LEU, lval, lsh, max, lval, zero);
4171 tcg_gen_movcond_i64(TCG_COND_LT, dst, lsh, zero, rval, lval);
4172
4173 tcg_temp_free_i64(lval);
4174 tcg_temp_free_i64(rval);
4175 tcg_temp_free_i64(lsh);
4176 tcg_temp_free_i64(rsh);
4177 tcg_temp_free_i64(zero);
4178 tcg_temp_free_i64(max);
4179 }
4180
4181 static void gen_sshl_vec(unsigned vece, TCGv_vec dst,
4182 TCGv_vec src, TCGv_vec shift)
4183 {
4184 TCGv_vec lval = tcg_temp_new_vec_matching(dst);
4185 TCGv_vec rval = tcg_temp_new_vec_matching(dst);
4186 TCGv_vec lsh = tcg_temp_new_vec_matching(dst);
4187 TCGv_vec rsh = tcg_temp_new_vec_matching(dst);
4188 TCGv_vec tmp = tcg_temp_new_vec_matching(dst);
4189
4190 /*
4191 * Rely on the TCG guarantee that out of range shifts produce
4192 * unspecified results, not undefined behaviour (i.e. no trap).
4193 * Discard out-of-range results after the fact.
4194 */
4195 tcg_gen_neg_vec(vece, rsh, shift);
4196 if (vece == MO_8) {
4197 tcg_gen_mov_vec(lsh, shift);
4198 } else {
4199 tcg_gen_dupi_vec(vece, tmp, 0xff);
4200 tcg_gen_and_vec(vece, lsh, shift, tmp);
4201 tcg_gen_and_vec(vece, rsh, rsh, tmp);
4202 }
4203
4204 /* Bound rsh so out of bound right shift gets -1. */
4205 tcg_gen_dupi_vec(vece, tmp, (8 << vece) - 1);
4206 tcg_gen_umin_vec(vece, rsh, rsh, tmp);
4207 tcg_gen_cmp_vec(TCG_COND_GT, vece, tmp, lsh, tmp);
4208
4209 tcg_gen_shlv_vec(vece, lval, src, lsh);
4210 tcg_gen_sarv_vec(vece, rval, src, rsh);
4211
4212 /* Select in-bound left shift. */
4213 tcg_gen_andc_vec(vece, lval, lval, tmp);
4214
4215 /* Select between left and right shift. */
4216 if (vece == MO_8) {
4217 tcg_gen_dupi_vec(vece, tmp, 0);
4218 tcg_gen_cmpsel_vec(TCG_COND_LT, vece, dst, lsh, tmp, rval, lval);
4219 } else {
4220 tcg_gen_dupi_vec(vece, tmp, 0x80);
4221 tcg_gen_cmpsel_vec(TCG_COND_LT, vece, dst, lsh, tmp, lval, rval);
4222 }
4223
4224 tcg_temp_free_vec(lval);
4225 tcg_temp_free_vec(rval);
4226 tcg_temp_free_vec(lsh);
4227 tcg_temp_free_vec(rsh);
4228 tcg_temp_free_vec(tmp);
4229 }
4230
4231 void gen_gvec_sshl(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4232 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4233 {
4234 static const TCGOpcode vecop_list[] = {
4235 INDEX_op_neg_vec, INDEX_op_umin_vec, INDEX_op_shlv_vec,
4236 INDEX_op_sarv_vec, INDEX_op_cmp_vec, INDEX_op_cmpsel_vec, 0
4237 };
4238 static const GVecGen3 ops[4] = {
4239 { .fniv = gen_sshl_vec,
4240 .fno = gen_helper_gvec_sshl_b,
4241 .opt_opc = vecop_list,
4242 .vece = MO_8 },
4243 { .fniv = gen_sshl_vec,
4244 .fno = gen_helper_gvec_sshl_h,
4245 .opt_opc = vecop_list,
4246 .vece = MO_16 },
4247 { .fni4 = gen_sshl_i32,
4248 .fniv = gen_sshl_vec,
4249 .opt_opc = vecop_list,
4250 .vece = MO_32 },
4251 { .fni8 = gen_sshl_i64,
4252 .fniv = gen_sshl_vec,
4253 .opt_opc = vecop_list,
4254 .vece = MO_64 },
4255 };
4256 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4257 }
4258
4259 static void gen_uqadd_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4260 TCGv_vec a, TCGv_vec b)
4261 {
4262 TCGv_vec x = tcg_temp_new_vec_matching(t);
4263 tcg_gen_add_vec(vece, x, a, b);
4264 tcg_gen_usadd_vec(vece, t, a, b);
4265 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4266 tcg_gen_or_vec(vece, sat, sat, x);
4267 tcg_temp_free_vec(x);
4268 }
4269
4270 void gen_gvec_uqadd_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4271 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4272 {
4273 static const TCGOpcode vecop_list[] = {
4274 INDEX_op_usadd_vec, INDEX_op_cmp_vec, INDEX_op_add_vec, 0
4275 };
4276 static const GVecGen4 ops[4] = {
4277 { .fniv = gen_uqadd_vec,
4278 .fno = gen_helper_gvec_uqadd_b,
4279 .write_aofs = true,
4280 .opt_opc = vecop_list,
4281 .vece = MO_8 },
4282 { .fniv = gen_uqadd_vec,
4283 .fno = gen_helper_gvec_uqadd_h,
4284 .write_aofs = true,
4285 .opt_opc = vecop_list,
4286 .vece = MO_16 },
4287 { .fniv = gen_uqadd_vec,
4288 .fno = gen_helper_gvec_uqadd_s,
4289 .write_aofs = true,
4290 .opt_opc = vecop_list,
4291 .vece = MO_32 },
4292 { .fniv = gen_uqadd_vec,
4293 .fno = gen_helper_gvec_uqadd_d,
4294 .write_aofs = true,
4295 .opt_opc = vecop_list,
4296 .vece = MO_64 },
4297 };
4298 tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc),
4299 rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4300 }
4301
4302 static void gen_sqadd_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4303 TCGv_vec a, TCGv_vec b)
4304 {
4305 TCGv_vec x = tcg_temp_new_vec_matching(t);
4306 tcg_gen_add_vec(vece, x, a, b);
4307 tcg_gen_ssadd_vec(vece, t, a, b);
4308 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4309 tcg_gen_or_vec(vece, sat, sat, x);
4310 tcg_temp_free_vec(x);
4311 }
4312
4313 void gen_gvec_sqadd_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4314 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4315 {
4316 static const TCGOpcode vecop_list[] = {
4317 INDEX_op_ssadd_vec, INDEX_op_cmp_vec, INDEX_op_add_vec, 0
4318 };
4319 static const GVecGen4 ops[4] = {
4320 { .fniv = gen_sqadd_vec,
4321 .fno = gen_helper_gvec_sqadd_b,
4322 .opt_opc = vecop_list,
4323 .write_aofs = true,
4324 .vece = MO_8 },
4325 { .fniv = gen_sqadd_vec,
4326 .fno = gen_helper_gvec_sqadd_h,
4327 .opt_opc = vecop_list,
4328 .write_aofs = true,
4329 .vece = MO_16 },
4330 { .fniv = gen_sqadd_vec,
4331 .fno = gen_helper_gvec_sqadd_s,
4332 .opt_opc = vecop_list,
4333 .write_aofs = true,
4334 .vece = MO_32 },
4335 { .fniv = gen_sqadd_vec,
4336 .fno = gen_helper_gvec_sqadd_d,
4337 .opt_opc = vecop_list,
4338 .write_aofs = true,
4339 .vece = MO_64 },
4340 };
4341 tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc),
4342 rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4343 }
4344
4345 static void gen_uqsub_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4346 TCGv_vec a, TCGv_vec b)
4347 {
4348 TCGv_vec x = tcg_temp_new_vec_matching(t);
4349 tcg_gen_sub_vec(vece, x, a, b);
4350 tcg_gen_ussub_vec(vece, t, a, b);
4351 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4352 tcg_gen_or_vec(vece, sat, sat, x);
4353 tcg_temp_free_vec(x);
4354 }
4355
4356 void gen_gvec_uqsub_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4357 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4358 {
4359 static const TCGOpcode vecop_list[] = {
4360 INDEX_op_ussub_vec, INDEX_op_cmp_vec, INDEX_op_sub_vec, 0
4361 };
4362 static const GVecGen4 ops[4] = {
4363 { .fniv = gen_uqsub_vec,
4364 .fno = gen_helper_gvec_uqsub_b,
4365 .opt_opc = vecop_list,
4366 .write_aofs = true,
4367 .vece = MO_8 },
4368 { .fniv = gen_uqsub_vec,
4369 .fno = gen_helper_gvec_uqsub_h,
4370 .opt_opc = vecop_list,
4371 .write_aofs = true,
4372 .vece = MO_16 },
4373 { .fniv = gen_uqsub_vec,
4374 .fno = gen_helper_gvec_uqsub_s,
4375 .opt_opc = vecop_list,
4376 .write_aofs = true,
4377 .vece = MO_32 },
4378 { .fniv = gen_uqsub_vec,
4379 .fno = gen_helper_gvec_uqsub_d,
4380 .opt_opc = vecop_list,
4381 .write_aofs = true,
4382 .vece = MO_64 },
4383 };
4384 tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc),
4385 rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4386 }
4387
4388 static void gen_sqsub_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4389 TCGv_vec a, TCGv_vec b)
4390 {
4391 TCGv_vec x = tcg_temp_new_vec_matching(t);
4392 tcg_gen_sub_vec(vece, x, a, b);
4393 tcg_gen_sssub_vec(vece, t, a, b);
4394 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4395 tcg_gen_or_vec(vece, sat, sat, x);
4396 tcg_temp_free_vec(x);
4397 }
4398
4399 void gen_gvec_sqsub_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4400 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4401 {
4402 static const TCGOpcode vecop_list[] = {
4403 INDEX_op_sssub_vec, INDEX_op_cmp_vec, INDEX_op_sub_vec, 0
4404 };
4405 static const GVecGen4 ops[4] = {
4406 { .fniv = gen_sqsub_vec,
4407 .fno = gen_helper_gvec_sqsub_b,
4408 .opt_opc = vecop_list,
4409 .write_aofs = true,
4410 .vece = MO_8 },
4411 { .fniv = gen_sqsub_vec,
4412 .fno = gen_helper_gvec_sqsub_h,
4413 .opt_opc = vecop_list,
4414 .write_aofs = true,
4415 .vece = MO_16 },
4416 { .fniv = gen_sqsub_vec,
4417 .fno = gen_helper_gvec_sqsub_s,
4418 .opt_opc = vecop_list,
4419 .write_aofs = true,
4420 .vece = MO_32 },
4421 { .fniv = gen_sqsub_vec,
4422 .fno = gen_helper_gvec_sqsub_d,
4423 .opt_opc = vecop_list,
4424 .write_aofs = true,
4425 .vece = MO_64 },
4426 };
4427 tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc),
4428 rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4429 }
4430
4431 static void gen_sabd_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4432 {
4433 TCGv_i32 t = tcg_temp_new_i32();
4434
4435 tcg_gen_sub_i32(t, a, b);
4436 tcg_gen_sub_i32(d, b, a);
4437 tcg_gen_movcond_i32(TCG_COND_LT, d, a, b, d, t);
4438 tcg_temp_free_i32(t);
4439 }
4440
4441 static void gen_sabd_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
4442 {
4443 TCGv_i64 t = tcg_temp_new_i64();
4444
4445 tcg_gen_sub_i64(t, a, b);
4446 tcg_gen_sub_i64(d, b, a);
4447 tcg_gen_movcond_i64(TCG_COND_LT, d, a, b, d, t);
4448 tcg_temp_free_i64(t);
4449 }
4450
4451 static void gen_sabd_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
4452 {
4453 TCGv_vec t = tcg_temp_new_vec_matching(d);
4454
4455 tcg_gen_smin_vec(vece, t, a, b);
4456 tcg_gen_smax_vec(vece, d, a, b);
4457 tcg_gen_sub_vec(vece, d, d, t);
4458 tcg_temp_free_vec(t);
4459 }
4460
4461 void gen_gvec_sabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4462 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4463 {
4464 static const TCGOpcode vecop_list[] = {
4465 INDEX_op_sub_vec, INDEX_op_smin_vec, INDEX_op_smax_vec, 0
4466 };
4467 static const GVecGen3 ops[4] = {
4468 { .fniv = gen_sabd_vec,
4469 .fno = gen_helper_gvec_sabd_b,
4470 .opt_opc = vecop_list,
4471 .vece = MO_8 },
4472 { .fniv = gen_sabd_vec,
4473 .fno = gen_helper_gvec_sabd_h,
4474 .opt_opc = vecop_list,
4475 .vece = MO_16 },
4476 { .fni4 = gen_sabd_i32,
4477 .fniv = gen_sabd_vec,
4478 .fno = gen_helper_gvec_sabd_s,
4479 .opt_opc = vecop_list,
4480 .vece = MO_32 },
4481 { .fni8 = gen_sabd_i64,
4482 .fniv = gen_sabd_vec,
4483 .fno = gen_helper_gvec_sabd_d,
4484 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4485 .opt_opc = vecop_list,
4486 .vece = MO_64 },
4487 };
4488 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4489 }
4490
4491 static void gen_uabd_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4492 {
4493 TCGv_i32 t = tcg_temp_new_i32();
4494
4495 tcg_gen_sub_i32(t, a, b);
4496 tcg_gen_sub_i32(d, b, a);
4497 tcg_gen_movcond_i32(TCG_COND_LTU, d, a, b, d, t);
4498 tcg_temp_free_i32(t);
4499 }
4500
4501 static void gen_uabd_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
4502 {
4503 TCGv_i64 t = tcg_temp_new_i64();
4504
4505 tcg_gen_sub_i64(t, a, b);
4506 tcg_gen_sub_i64(d, b, a);
4507 tcg_gen_movcond_i64(TCG_COND_LTU, d, a, b, d, t);
4508 tcg_temp_free_i64(t);
4509 }
4510
4511 static void gen_uabd_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
4512 {
4513 TCGv_vec t = tcg_temp_new_vec_matching(d);
4514
4515 tcg_gen_umin_vec(vece, t, a, b);
4516 tcg_gen_umax_vec(vece, d, a, b);
4517 tcg_gen_sub_vec(vece, d, d, t);
4518 tcg_temp_free_vec(t);
4519 }
4520
4521 void gen_gvec_uabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4522 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4523 {
4524 static const TCGOpcode vecop_list[] = {
4525 INDEX_op_sub_vec, INDEX_op_umin_vec, INDEX_op_umax_vec, 0
4526 };
4527 static const GVecGen3 ops[4] = {
4528 { .fniv = gen_uabd_vec,
4529 .fno = gen_helper_gvec_uabd_b,
4530 .opt_opc = vecop_list,
4531 .vece = MO_8 },
4532 { .fniv = gen_uabd_vec,
4533 .fno = gen_helper_gvec_uabd_h,
4534 .opt_opc = vecop_list,
4535 .vece = MO_16 },
4536 { .fni4 = gen_uabd_i32,
4537 .fniv = gen_uabd_vec,
4538 .fno = gen_helper_gvec_uabd_s,
4539 .opt_opc = vecop_list,
4540 .vece = MO_32 },
4541 { .fni8 = gen_uabd_i64,
4542 .fniv = gen_uabd_vec,
4543 .fno = gen_helper_gvec_uabd_d,
4544 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4545 .opt_opc = vecop_list,
4546 .vece = MO_64 },
4547 };
4548 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4549 }
4550
4551 static void gen_saba_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4552 {
4553 TCGv_i32 t = tcg_temp_new_i32();
4554 gen_sabd_i32(t, a, b);
4555 tcg_gen_add_i32(d, d, t);
4556 tcg_temp_free_i32(t);
4557 }
4558
4559 static void gen_saba_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
4560 {
4561 TCGv_i64 t = tcg_temp_new_i64();
4562 gen_sabd_i64(t, a, b);
4563 tcg_gen_add_i64(d, d, t);
4564 tcg_temp_free_i64(t);
4565 }
4566
4567 static void gen_saba_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
4568 {
4569 TCGv_vec t = tcg_temp_new_vec_matching(d);
4570 gen_sabd_vec(vece, t, a, b);
4571 tcg_gen_add_vec(vece, d, d, t);
4572 tcg_temp_free_vec(t);
4573 }
4574
4575 void gen_gvec_saba(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4576 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4577 {
4578 static const TCGOpcode vecop_list[] = {
4579 INDEX_op_sub_vec, INDEX_op_add_vec,
4580 INDEX_op_smin_vec, INDEX_op_smax_vec, 0
4581 };
4582 static const GVecGen3 ops[4] = {
4583 { .fniv = gen_saba_vec,
4584 .fno = gen_helper_gvec_saba_b,
4585 .opt_opc = vecop_list,
4586 .load_dest = true,
4587 .vece = MO_8 },
4588 { .fniv = gen_saba_vec,
4589 .fno = gen_helper_gvec_saba_h,
4590 .opt_opc = vecop_list,
4591 .load_dest = true,
4592 .vece = MO_16 },
4593 { .fni4 = gen_saba_i32,
4594 .fniv = gen_saba_vec,
4595 .fno = gen_helper_gvec_saba_s,
4596 .opt_opc = vecop_list,
4597 .load_dest = true,
4598 .vece = MO_32 },
4599 { .fni8 = gen_saba_i64,
4600 .fniv = gen_saba_vec,
4601 .fno = gen_helper_gvec_saba_d,
4602 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4603 .opt_opc = vecop_list,
4604 .load_dest = true,
4605 .vece = MO_64 },
4606 };
4607 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4608 }
4609
4610 static void gen_uaba_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4611 {
4612 TCGv_i32 t = tcg_temp_new_i32();
4613 gen_uabd_i32(t, a, b);
4614 tcg_gen_add_i32(d, d, t);
4615 tcg_temp_free_i32(t);
4616 }
4617
4618 static void gen_uaba_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
4619 {
4620 TCGv_i64 t = tcg_temp_new_i64();
4621 gen_uabd_i64(t, a, b);
4622 tcg_gen_add_i64(d, d, t);
4623 tcg_temp_free_i64(t);
4624 }
4625
4626 static void gen_uaba_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
4627 {
4628 TCGv_vec t = tcg_temp_new_vec_matching(d);
4629 gen_uabd_vec(vece, t, a, b);
4630 tcg_gen_add_vec(vece, d, d, t);
4631 tcg_temp_free_vec(t);
4632 }
4633
4634 void gen_gvec_uaba(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4635 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4636 {
4637 static const TCGOpcode vecop_list[] = {
4638 INDEX_op_sub_vec, INDEX_op_add_vec,
4639 INDEX_op_umin_vec, INDEX_op_umax_vec, 0
4640 };
4641 static const GVecGen3 ops[4] = {
4642 { .fniv = gen_uaba_vec,
4643 .fno = gen_helper_gvec_uaba_b,
4644 .opt_opc = vecop_list,
4645 .load_dest = true,
4646 .vece = MO_8 },
4647 { .fniv = gen_uaba_vec,
4648 .fno = gen_helper_gvec_uaba_h,
4649 .opt_opc = vecop_list,
4650 .load_dest = true,
4651 .vece = MO_16 },
4652 { .fni4 = gen_uaba_i32,
4653 .fniv = gen_uaba_vec,
4654 .fno = gen_helper_gvec_uaba_s,
4655 .opt_opc = vecop_list,
4656 .load_dest = true,
4657 .vece = MO_32 },
4658 { .fni8 = gen_uaba_i64,
4659 .fniv = gen_uaba_vec,
4660 .fno = gen_helper_gvec_uaba_d,
4661 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4662 .opt_opc = vecop_list,
4663 .load_dest = true,
4664 .vece = MO_64 },
4665 };
4666 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4667 }
4668
4669 static void do_coproc_insn(DisasContext *s, int cpnum, int is64,
4670 int opc1, int crn, int crm, int opc2,
4671 bool isread, int rt, int rt2)
4672 {
4673 const ARMCPRegInfo *ri;
4674
4675 ri = get_arm_cp_reginfo(s->cp_regs,
4676 ENCODE_CP_REG(cpnum, is64, s->ns, crn, crm, opc1, opc2));
4677 if (ri) {
4678 bool need_exit_tb;
4679
4680 /* Check access permissions */
4681 if (!cp_access_ok(s->current_el, ri, isread)) {
4682 unallocated_encoding(s);
4683 return;
4684 }
4685
4686 if (s->hstr_active || ri->accessfn ||
4687 (arm_dc_feature(s, ARM_FEATURE_XSCALE) && cpnum < 14)) {
4688 /* Emit code to perform further access permissions checks at
4689 * runtime; this may result in an exception.
4690 * Note that on XScale all cp0..c13 registers do an access check
4691 * call in order to handle c15_cpar.
4692 */
4693 TCGv_ptr tmpptr;
4694 TCGv_i32 tcg_syn, tcg_isread;
4695 uint32_t syndrome;
4696
4697 /* Note that since we are an implementation which takes an
4698 * exception on a trapped conditional instruction only if the
4699 * instruction passes its condition code check, we can take
4700 * advantage of the clause in the ARM ARM that allows us to set
4701 * the COND field in the instruction to 0xE in all cases.
4702 * We could fish the actual condition out of the insn (ARM)
4703 * or the condexec bits (Thumb) but it isn't necessary.
4704 */
4705 switch (cpnum) {
4706 case 14:
4707 if (is64) {
4708 syndrome = syn_cp14_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
4709 isread, false);
4710 } else {
4711 syndrome = syn_cp14_rt_trap(1, 0xe, opc1, opc2, crn, crm,
4712 rt, isread, false);
4713 }
4714 break;
4715 case 15:
4716 if (is64) {
4717 syndrome = syn_cp15_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
4718 isread, false);
4719 } else {
4720 syndrome = syn_cp15_rt_trap(1, 0xe, opc1, opc2, crn, crm,
4721 rt, isread, false);
4722 }
4723 break;
4724 default:
4725 /* ARMv8 defines that only coprocessors 14 and 15 exist,
4726 * so this can only happen if this is an ARMv7 or earlier CPU,
4727 * in which case the syndrome information won't actually be
4728 * guest visible.
4729 */
4730 assert(!arm_dc_feature(s, ARM_FEATURE_V8));
4731 syndrome = syn_uncategorized();
4732 break;
4733 }
4734
4735 gen_set_condexec(s);
4736 gen_set_pc_im(s, s->pc_curr);
4737 tmpptr = tcg_const_ptr(ri);
4738 tcg_syn = tcg_const_i32(syndrome);
4739 tcg_isread = tcg_const_i32(isread);
4740 gen_helper_access_check_cp_reg(cpu_env, tmpptr, tcg_syn,
4741 tcg_isread);
4742 tcg_temp_free_ptr(tmpptr);
4743 tcg_temp_free_i32(tcg_syn);
4744 tcg_temp_free_i32(tcg_isread);
4745 } else if (ri->type & ARM_CP_RAISES_EXC) {
4746 /*
4747 * The readfn or writefn might raise an exception;
4748 * synchronize the CPU state in case it does.
4749 */
4750 gen_set_condexec(s);
4751 gen_set_pc_im(s, s->pc_curr);
4752 }
4753
4754 /* Handle special cases first */
4755 switch (ri->type & ~(ARM_CP_FLAG_MASK & ~ARM_CP_SPECIAL)) {
4756 case ARM_CP_NOP:
4757 return;
4758 case ARM_CP_WFI:
4759 if (isread) {
4760 unallocated_encoding(s);
4761 return;
4762 }
4763 gen_set_pc_im(s, s->base.pc_next);
4764 s->base.is_jmp = DISAS_WFI;
4765 return;
4766 default:
4767 break;
4768 }
4769
4770 if ((tb_cflags(s->base.tb) & CF_USE_ICOUNT) && (ri->type & ARM_CP_IO)) {
4771 gen_io_start();
4772 }
4773
4774 if (isread) {
4775 /* Read */
4776 if (is64) {
4777 TCGv_i64 tmp64;
4778 TCGv_i32 tmp;
4779 if (ri->type & ARM_CP_CONST) {
4780 tmp64 = tcg_const_i64(ri->resetvalue);
4781 } else if (ri->readfn) {
4782 TCGv_ptr tmpptr;
4783 tmp64 = tcg_temp_new_i64();
4784 tmpptr = tcg_const_ptr(ri);
4785 gen_helper_get_cp_reg64(tmp64, cpu_env, tmpptr);
4786 tcg_temp_free_ptr(tmpptr);
4787 } else {
4788 tmp64 = tcg_temp_new_i64();
4789 tcg_gen_ld_i64(tmp64, cpu_env, ri->fieldoffset);
4790 }
4791 tmp = tcg_temp_new_i32();
4792 tcg_gen_extrl_i64_i32(tmp, tmp64);
4793 store_reg(s, rt, tmp);
4794 tmp = tcg_temp_new_i32();
4795 tcg_gen_extrh_i64_i32(tmp, tmp64);
4796 tcg_temp_free_i64(tmp64);
4797 store_reg(s, rt2, tmp);
4798 } else {
4799 TCGv_i32 tmp;
4800 if (ri->type & ARM_CP_CONST) {
4801 tmp = tcg_const_i32(ri->resetvalue);
4802 } else if (ri->readfn) {
4803 TCGv_ptr tmpptr;
4804 tmp = tcg_temp_new_i32();
4805 tmpptr = tcg_const_ptr(ri);
4806 gen_helper_get_cp_reg(tmp, cpu_env, tmpptr);
4807 tcg_temp_free_ptr(tmpptr);
4808 } else {
4809 tmp = load_cpu_offset(ri->fieldoffset);
4810 }
4811 if (rt == 15) {
4812 /* Destination register of r15 for 32 bit loads sets
4813 * the condition codes from the high 4 bits of the value
4814 */
4815 gen_set_nzcv(tmp);
4816 tcg_temp_free_i32(tmp);
4817 } else {
4818 store_reg(s, rt, tmp);
4819 }
4820 }
4821 } else {
4822 /* Write */
4823 if (ri->type & ARM_CP_CONST) {
4824 /* If not forbidden by access permissions, treat as WI */
4825 return;
4826 }
4827
4828 if (is64) {
4829 TCGv_i32 tmplo, tmphi;
4830 TCGv_i64 tmp64 = tcg_temp_new_i64();
4831 tmplo = load_reg(s, rt);
4832 tmphi = load_reg(s, rt2);
4833 tcg_gen_concat_i32_i64(tmp64, tmplo, tmphi);
4834 tcg_temp_free_i32(tmplo);
4835 tcg_temp_free_i32(tmphi);
4836 if (ri->writefn) {
4837 TCGv_ptr tmpptr = tcg_const_ptr(ri);
4838 gen_helper_set_cp_reg64(cpu_env, tmpptr, tmp64);
4839 tcg_temp_free_ptr(tmpptr);
4840 } else {
4841 tcg_gen_st_i64(tmp64, cpu_env, ri->fieldoffset);
4842 }
4843 tcg_temp_free_i64(tmp64);
4844 } else {
4845 if (ri->writefn) {
4846 TCGv_i32 tmp;
4847 TCGv_ptr tmpptr;
4848 tmp = load_reg(s, rt);
4849 tmpptr = tcg_const_ptr(ri);
4850 gen_helper_set_cp_reg(cpu_env, tmpptr, tmp);
4851 tcg_temp_free_ptr(tmpptr);
4852 tcg_temp_free_i32(tmp);
4853 } else {
4854 TCGv_i32 tmp = load_reg(s, rt);
4855 store_cpu_offset(tmp, ri->fieldoffset);
4856 }
4857 }
4858 }
4859
4860 /* I/O operations must end the TB here (whether read or write) */
4861 need_exit_tb = ((tb_cflags(s->base.tb) & CF_USE_ICOUNT) &&
4862 (ri->type & ARM_CP_IO));
4863
4864 if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
4865 /*
4866 * A write to any coprocessor register that ends a TB
4867 * must rebuild the hflags for the next TB.
4868 */
4869 TCGv_i32 tcg_el = tcg_const_i32(s->current_el);
4870 if (arm_dc_feature(s, ARM_FEATURE_M)) {
4871 gen_helper_rebuild_hflags_m32(cpu_env, tcg_el);
4872 } else {
4873 if (ri->type & ARM_CP_NEWEL) {
4874 gen_helper_rebuild_hflags_a32_newel(cpu_env);
4875 } else {
4876 gen_helper_rebuild_hflags_a32(cpu_env, tcg_el);
4877 }
4878 }
4879 tcg_temp_free_i32(tcg_el);
4880 /*
4881 * We default to ending the TB on a coprocessor register write,
4882 * but allow this to be suppressed by the register definition
4883 * (usually only necessary to work around guest bugs).
4884 */
4885 need_exit_tb = true;
4886 }
4887 if (need_exit_tb) {
4888 gen_lookup_tb(s);
4889 }
4890
4891 return;
4892 }
4893
4894 /* Unknown register; this might be a guest error or a QEMU
4895 * unimplemented feature.
4896 */
4897 if (is64) {
4898 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
4899 "64 bit system register cp:%d opc1: %d crm:%d "
4900 "(%s)\n",
4901 isread ? "read" : "write", cpnum, opc1, crm,
4902 s->ns ? "non-secure" : "secure");
4903 } else {
4904 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
4905 "system register cp:%d opc1:%d crn:%d crm:%d opc2:%d "
4906 "(%s)\n",
4907 isread ? "read" : "write", cpnum, opc1, crn, crm, opc2,
4908 s->ns ? "non-secure" : "secure");
4909 }
4910
4911 unallocated_encoding(s);
4912 return;
4913 }
4914
4915 /* Decode XScale DSP or iWMMXt insn (in the copro space, cp=0 or 1) */
4916 static void disas_xscale_insn(DisasContext *s, uint32_t insn)
4917 {
4918 int cpnum = (insn >> 8) & 0xf;
4919
4920 if (extract32(s->c15_cpar, cpnum, 1) == 0) {
4921 unallocated_encoding(s);
4922 } else if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
4923 if (disas_iwmmxt_insn(s, insn)) {
4924 unallocated_encoding(s);
4925 }
4926 } else if (arm_dc_feature(s, ARM_FEATURE_XSCALE)) {
4927 if (disas_dsp_insn(s, insn)) {
4928 unallocated_encoding(s);
4929 }
4930 }
4931 }
4932
4933 /* Store a 64-bit value to a register pair. Clobbers val. */
4934 static void gen_storeq_reg(DisasContext *s, int rlow, int rhigh, TCGv_i64 val)
4935 {
4936 TCGv_i32 tmp;
4937 tmp = tcg_temp_new_i32();
4938 tcg_gen_extrl_i64_i32(tmp, val);
4939 store_reg(s, rlow, tmp);
4940 tmp = tcg_temp_new_i32();
4941 tcg_gen_extrh_i64_i32(tmp, val);
4942 store_reg(s, rhigh, tmp);
4943 }
4944
4945 /* load and add a 64-bit value from a register pair. */
4946 static void gen_addq(DisasContext *s, TCGv_i64 val, int rlow, int rhigh)
4947 {
4948 TCGv_i64 tmp;
4949 TCGv_i32 tmpl;
4950 TCGv_i32 tmph;
4951
4952 /* Load 64-bit value rd:rn. */
4953 tmpl = load_reg(s, rlow);
4954 tmph = load_reg(s, rhigh);
4955 tmp = tcg_temp_new_i64();
4956 tcg_gen_concat_i32_i64(tmp, tmpl, tmph);
4957 tcg_temp_free_i32(tmpl);
4958 tcg_temp_free_i32(tmph);
4959 tcg_gen_add_i64(val, val, tmp);
4960 tcg_temp_free_i64(tmp);
4961 }
4962
4963 /* Set N and Z flags from hi|lo. */
4964 static void gen_logicq_cc(TCGv_i32 lo, TCGv_i32 hi)
4965 {
4966 tcg_gen_mov_i32(cpu_NF, hi);
4967 tcg_gen_or_i32(cpu_ZF, lo, hi);
4968 }
4969
4970 /* Load/Store exclusive instructions are implemented by remembering
4971 the value/address loaded, and seeing if these are the same
4972 when the store is performed. This should be sufficient to implement
4973 the architecturally mandated semantics, and avoids having to monitor
4974 regular stores. The compare vs the remembered value is done during
4975 the cmpxchg operation, but we must compare the addresses manually. */
4976 static void gen_load_exclusive(DisasContext *s, int rt, int rt2,
4977 TCGv_i32 addr, int size)
4978 {
4979 TCGv_i32 tmp = tcg_temp_new_i32();
4980 MemOp opc = size | MO_ALIGN | s->be_data;
4981
4982 s->is_ldex = true;
4983
4984 if (size == 3) {
4985 TCGv_i32 tmp2 = tcg_temp_new_i32();
4986 TCGv_i64 t64 = tcg_temp_new_i64();
4987
4988 /*
4989 * For AArch32, architecturally the 32-bit word at the lowest
4990 * address is always Rt and the one at addr+4 is Rt2, even if
4991 * the CPU is big-endian. That means we don't want to do a
4992 * gen_aa32_ld_i64(), which checks SCTLR_B as if for an
4993 * architecturally 64-bit access, but instead do a 64-bit access
4994 * using MO_BE if appropriate and then split the two halves.
4995 */
4996 TCGv taddr = gen_aa32_addr(s, addr, opc);
4997
4998 tcg_gen_qemu_ld_i64(t64, taddr, get_mem_index(s), opc);
4999 tcg_temp_free(taddr);
5000 tcg_gen_mov_i64(cpu_exclusive_val, t64);
5001 if (s->be_data == MO_BE) {
5002 tcg_gen_extr_i64_i32(tmp2, tmp, t64);
5003 } else {
5004 tcg_gen_extr_i64_i32(tmp, tmp2, t64);
5005 }
5006 tcg_temp_free_i64(t64);
5007
5008 store_reg(s, rt2, tmp2);
5009 } else {
5010 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), opc);
5011 tcg_gen_extu_i32_i64(cpu_exclusive_val, tmp);
5012 }
5013
5014 store_reg(s, rt, tmp);
5015 tcg_gen_extu_i32_i64(cpu_exclusive_addr, addr);
5016 }
5017
5018 static void gen_clrex(DisasContext *s)
5019 {
5020 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
5021 }
5022
5023 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
5024 TCGv_i32 addr, int size)
5025 {
5026 TCGv_i32 t0, t1, t2;
5027 TCGv_i64 extaddr;
5028 TCGv taddr;
5029 TCGLabel *done_label;
5030 TCGLabel *fail_label;
5031 MemOp opc = size | MO_ALIGN | s->be_data;
5032
5033 /* if (env->exclusive_addr == addr && env->exclusive_val == [addr]) {
5034 [addr] = {Rt};
5035 {Rd} = 0;
5036 } else {
5037 {Rd} = 1;
5038 } */
5039 fail_label = gen_new_label();
5040 done_label = gen_new_label();
5041 extaddr = tcg_temp_new_i64();
5042 tcg_gen_extu_i32_i64(extaddr, addr);
5043 tcg_gen_brcond_i64(TCG_COND_NE, extaddr, cpu_exclusive_addr, fail_label);
5044 tcg_temp_free_i64(extaddr);
5045
5046 taddr = gen_aa32_addr(s, addr, opc);
5047 t0 = tcg_temp_new_i32();
5048 t1 = load_reg(s, rt);
5049 if (size == 3) {
5050 TCGv_i64 o64 = tcg_temp_new_i64();
5051 TCGv_i64 n64 = tcg_temp_new_i64();
5052
5053 t2 = load_reg(s, rt2);
5054
5055 /*
5056 * For AArch32, architecturally the 32-bit word at the lowest
5057 * address is always Rt and the one at addr+4 is Rt2, even if
5058 * the CPU is big-endian. Since we're going to treat this as a
5059 * single 64-bit BE store, we need to put the two halves in the
5060 * opposite order for BE to LE, so that they end up in the right
5061 * places. We don't want gen_aa32_st_i64, because that checks
5062 * SCTLR_B as if for an architectural 64-bit access.
5063 */
5064 if (s->be_data == MO_BE) {
5065 tcg_gen_concat_i32_i64(n64, t2, t1);
5066 } else {
5067 tcg_gen_concat_i32_i64(n64, t1, t2);
5068 }
5069 tcg_temp_free_i32(t2);
5070
5071 tcg_gen_atomic_cmpxchg_i64(o64, taddr, cpu_exclusive_val, n64,
5072 get_mem_index(s), opc);
5073 tcg_temp_free_i64(n64);
5074
5075 tcg_gen_setcond_i64(TCG_COND_NE, o64, o64, cpu_exclusive_val);
5076 tcg_gen_extrl_i64_i32(t0, o64);
5077
5078 tcg_temp_free_i64(o64);
5079 } else {
5080 t2 = tcg_temp_new_i32();
5081 tcg_gen_extrl_i64_i32(t2, cpu_exclusive_val);
5082 tcg_gen_atomic_cmpxchg_i32(t0, taddr, t2, t1, get_mem_index(s), opc);
5083 tcg_gen_setcond_i32(TCG_COND_NE, t0, t0, t2);
5084 tcg_temp_free_i32(t2);
5085 }
5086 tcg_temp_free_i32(t1);
5087 tcg_temp_free(taddr);
5088 tcg_gen_mov_i32(cpu_R[rd], t0);
5089 tcg_temp_free_i32(t0);
5090 tcg_gen_br(done_label);
5091
5092 gen_set_label(fail_label);
5093 tcg_gen_movi_i32(cpu_R[rd], 1);
5094 gen_set_label(done_label);
5095 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
5096 }
5097
5098 /* gen_srs:
5099 * @env: CPUARMState
5100 * @s: DisasContext
5101 * @mode: mode field from insn (which stack to store to)
5102 * @amode: addressing mode (DA/IA/DB/IB), encoded as per P,U bits in ARM insn
5103 * @writeback: true if writeback bit set
5104 *
5105 * Generate code for the SRS (Store Return State) insn.
5106 */
5107 static void gen_srs(DisasContext *s,
5108 uint32_t mode, uint32_t amode, bool writeback)
5109 {
5110 int32_t offset;
5111 TCGv_i32 addr, tmp;
5112 bool undef = false;
5113
5114 /* SRS is:
5115 * - trapped to EL3 if EL3 is AArch64 and we are at Secure EL1
5116 * and specified mode is monitor mode
5117 * - UNDEFINED in Hyp mode
5118 * - UNPREDICTABLE in User or System mode
5119 * - UNPREDICTABLE if the specified mode is:
5120 * -- not implemented
5121 * -- not a valid mode number
5122 * -- a mode that's at a higher exception level
5123 * -- Monitor, if we are Non-secure
5124 * For the UNPREDICTABLE cases we choose to UNDEF.
5125 */
5126 if (s->current_el == 1 && !s->ns && mode == ARM_CPU_MODE_MON) {
5127 gen_exception_insn(s, s->pc_curr, EXCP_UDEF, syn_uncategorized(), 3);
5128 return;
5129 }
5130
5131 if (s->current_el == 0 || s->current_el == 2) {
5132 undef = true;
5133 }
5134
5135 switch (mode) {
5136 case ARM_CPU_MODE_USR:
5137 case ARM_CPU_MODE_FIQ:
5138 case ARM_CPU_MODE_IRQ:
5139 case ARM_CPU_MODE_SVC:
5140 case ARM_CPU_MODE_ABT:
5141 case ARM_CPU_MODE_UND:
5142 case ARM_CPU_MODE_SYS:
5143 break;
5144 case ARM_CPU_MODE_HYP:
5145 if (s->current_el == 1 || !arm_dc_feature(s, ARM_FEATURE_EL2)) {
5146 undef = true;
5147 }
5148 break;
5149 case ARM_CPU_MODE_MON:
5150 /* No need to check specifically for "are we non-secure" because
5151 * we've already made EL0 UNDEF and handled the trap for S-EL1;
5152 * so if this isn't EL3 then we must be non-secure.
5153 */
5154 if (s->current_el != 3) {
5155 undef = true;
5156 }
5157 break;
5158 default:
5159 undef = true;
5160 }
5161
5162 if (undef) {
5163 unallocated_encoding(s);
5164 return;
5165 }
5166
5167 addr = tcg_temp_new_i32();
5168 tmp = tcg_const_i32(mode);
5169 /* get_r13_banked() will raise an exception if called from System mode */
5170 gen_set_condexec(s);
5171 gen_set_pc_im(s, s->pc_curr);
5172 gen_helper_get_r13_banked(addr, cpu_env, tmp);
5173 tcg_temp_free_i32(tmp);
5174 switch (amode) {
5175 case 0: /* DA */
5176 offset = -4;
5177 break;
5178 case 1: /* IA */
5179 offset = 0;
5180 break;
5181 case 2: /* DB */
5182 offset = -8;
5183 break;
5184 case 3: /* IB */
5185 offset = 4;
5186 break;
5187 default:
5188 abort();
5189 }
5190 tcg_gen_addi_i32(addr, addr, offset);
5191 tmp = load_reg(s, 14);
5192 gen_aa32_st_i32(s, tmp, addr, get_mem_index(s), MO_UL | MO_ALIGN);
5193 tcg_temp_free_i32(tmp);
5194 tmp = load_cpu_field(spsr);
5195 tcg_gen_addi_i32(addr, addr, 4);
5196 gen_aa32_st_i32(s, tmp, addr, get_mem_index(s), MO_UL | MO_ALIGN);
5197 tcg_temp_free_i32(tmp);
5198 if (writeback) {
5199 switch (amode) {
5200 case 0:
5201 offset = -8;
5202 break;
5203 case 1:
5204 offset = 4;
5205 break;
5206 case 2:
5207 offset = -4;
5208 break;
5209 case 3:
5210 offset = 0;
5211 break;
5212 default:
5213 abort();
5214 }
5215 tcg_gen_addi_i32(addr, addr, offset);
5216 tmp = tcg_const_i32(mode);
5217 gen_helper_set_r13_banked(cpu_env, tmp, addr);
5218 tcg_temp_free_i32(tmp);
5219 }
5220 tcg_temp_free_i32(addr);
5221 s->base.is_jmp = DISAS_UPDATE_EXIT;
5222 }
5223
5224 /* Skip this instruction if the ARM condition is false */
5225 static void arm_skip_unless(DisasContext *s, uint32_t cond)
5226 {
5227 arm_gen_condlabel(s);
5228 arm_gen_test_cc(cond ^ 1, s->condlabel);
5229 }
5230
5231
5232 /*
5233 * Constant expanders used by T16/T32 decode
5234 */
5235
5236 /* Return only the rotation part of T32ExpandImm. */
5237 static int t32_expandimm_rot(DisasContext *s, int x)
5238 {
5239 return x & 0xc00 ? extract32(x, 7, 5) : 0;
5240 }
5241
5242 /* Return the unrotated immediate from T32ExpandImm. */
5243 static int t32_expandimm_imm(DisasContext *s, int x)
5244 {
5245 int imm = extract32(x, 0, 8);
5246
5247 switch (extract32(x, 8, 4)) {
5248 case 0: /* XY */
5249 /* Nothing to do. */
5250 break;
5251 case 1: /* 00XY00XY */
5252 imm *= 0x00010001;
5253 break;
5254 case 2: /* XY00XY00 */
5255 imm *= 0x01000100;
5256 break;
5257 case 3: /* XYXYXYXY */
5258 imm *= 0x01010101;
5259 break;
5260 default:
5261 /* Rotated constant. */
5262 imm |= 0x80;
5263 break;
5264 }
5265 return imm;
5266 }
5267
5268 static int t32_branch24(DisasContext *s, int x)
5269 {
5270 /* Convert J1:J2 at x[22:21] to I2:I1, which involves I=J^~S. */
5271 x ^= !(x < 0) * (3 << 21);
5272 /* Append the final zero. */
5273 return x << 1;
5274 }
5275
5276 static int t16_setflags(DisasContext *s)
5277 {
5278 return s->condexec_mask == 0;
5279 }
5280
5281 static int t16_push_list(DisasContext *s, int x)
5282 {
5283 return (x & 0xff) | (x & 0x100) << (14 - 8);
5284 }
5285
5286 static int t16_pop_list(DisasContext *s, int x)
5287 {
5288 return (x & 0xff) | (x & 0x100) << (15 - 8);
5289 }
5290
5291 /*
5292 * Include the generated decoders.
5293 */
5294
5295 #include "decode-a32.c.inc"
5296 #include "decode-a32-uncond.c.inc"
5297 #include "decode-t32.c.inc"
5298 #include "decode-t16.c.inc"
5299
5300 static bool valid_cp(DisasContext *s, int cp)
5301 {
5302 /*
5303 * Return true if this coprocessor field indicates something
5304 * that's really a possible coprocessor.
5305 * For v7 and earlier, coprocessors 8..15 were reserved for Arm use,
5306 * and of those only cp14 and cp15 were used for registers.
5307 * cp10 and cp11 were used for VFP and Neon, whose decode is
5308 * dealt with elsewhere. With the advent of fp16, cp9 is also
5309 * now part of VFP.
5310 * For v8A and later, the encoding has been tightened so that
5311 * only cp14 and cp15 are valid, and other values aren't considered
5312 * to be in the coprocessor-instruction space at all. v8M still
5313 * permits coprocessors 0..7.
5314 * For XScale, we must not decode the XScale cp0, cp1 space as
5315 * a standard coprocessor insn, because we want to fall through to
5316 * the legacy disas_xscale_insn() decoder after decodetree is done.
5317 */
5318 if (arm_dc_feature(s, ARM_FEATURE_XSCALE) && (cp == 0 || cp == 1)) {
5319 return false;
5320 }
5321
5322 if (arm_dc_feature(s, ARM_FEATURE_V8) &&
5323 !arm_dc_feature(s, ARM_FEATURE_M)) {
5324 return cp >= 14;
5325 }
5326 return cp < 8 || cp >= 14;
5327 }
5328
5329 static bool trans_MCR(DisasContext *s, arg_MCR *a)
5330 {
5331 if (!valid_cp(s, a->cp)) {
5332 return false;
5333 }
5334 do_coproc_insn(s, a->cp, false, a->opc1, a->crn, a->crm, a->opc2,
5335 false, a->rt, 0);
5336 return true;
5337 }
5338
5339 static bool trans_MRC(DisasContext *s, arg_MRC *a)
5340 {
5341 if (!valid_cp(s, a->cp)) {
5342 return false;
5343 }
5344 do_coproc_insn(s, a->cp, false, a->opc1, a->crn, a->crm, a->opc2,
5345 true, a->rt, 0);
5346 return true;
5347 }
5348
5349 static bool trans_MCRR(DisasContext *s, arg_MCRR *a)
5350 {
5351 if (!valid_cp(s, a->cp)) {
5352 return false;
5353 }
5354 do_coproc_insn(s, a->cp, true, a->opc1, 0, a->crm, 0,
5355 false, a->rt, a->rt2);
5356 return true;
5357 }
5358
5359 static bool trans_MRRC(DisasContext *s, arg_MRRC *a)
5360 {
5361 if (!valid_cp(s, a->cp)) {
5362 return false;
5363 }
5364 do_coproc_insn(s, a->cp, true, a->opc1, 0, a->crm, 0,
5365 true, a->rt, a->rt2);
5366 return true;
5367 }
5368
5369 /* Helpers to swap operands for reverse-subtract. */
5370 static void gen_rsb(TCGv_i32 dst, TCGv_i32 a, TCGv_i32 b)
5371 {
5372 tcg_gen_sub_i32(dst, b, a);
5373 }
5374
5375 static void gen_rsb_CC(TCGv_i32 dst, TCGv_i32 a, TCGv_i32 b)
5376 {
5377 gen_sub_CC(dst, b, a);
5378 }
5379
5380 static void gen_rsc(TCGv_i32 dest, TCGv_i32 a, TCGv_i32 b)
5381 {
5382 gen_sub_carry(dest, b, a);
5383 }
5384
5385 static void gen_rsc_CC(TCGv_i32 dest, TCGv_i32 a, TCGv_i32 b)
5386 {
5387 gen_sbc_CC(dest, b, a);
5388 }
5389
5390 /*
5391 * Helpers for the data processing routines.
5392 *
5393 * After the computation store the results back.
5394 * This may be suppressed altogether (STREG_NONE), require a runtime
5395 * check against the stack limits (STREG_SP_CHECK), or generate an
5396 * exception return. Oh, or store into a register.
5397 *
5398 * Always return true, indicating success for a trans_* function.
5399 */
5400 typedef enum {
5401 STREG_NONE,
5402 STREG_NORMAL,
5403 STREG_SP_CHECK,
5404 STREG_EXC_RET,
5405 } StoreRegKind;
5406
5407 static bool store_reg_kind(DisasContext *s, int rd,
5408 TCGv_i32 val, StoreRegKind kind)
5409 {
5410 switch (kind) {
5411 case STREG_NONE:
5412 tcg_temp_free_i32(val);
5413 return true;
5414 case STREG_NORMAL:
5415 /* See ALUWritePC: Interworking only from a32 mode. */
5416 if (s->thumb) {
5417 store_reg(s, rd, val);
5418 } else {
5419 store_reg_bx(s, rd, val);
5420 }
5421 return true;
5422 case STREG_SP_CHECK:
5423 store_sp_checked(s, val);
5424 return true;
5425 case STREG_EXC_RET:
5426 gen_exception_return(s, val);
5427 return true;
5428 }
5429 g_assert_not_reached();
5430 }
5431
5432 /*
5433 * Data Processing (register)
5434 *
5435 * Operate, with set flags, one register source,
5436 * one immediate shifted register source, and a destination.
5437 */
5438 static bool op_s_rrr_shi(DisasContext *s, arg_s_rrr_shi *a,
5439 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
5440 int logic_cc, StoreRegKind kind)
5441 {
5442 TCGv_i32 tmp1, tmp2;
5443
5444 tmp2 = load_reg(s, a->rm);
5445 gen_arm_shift_im(tmp2, a->shty, a->shim, logic_cc);
5446 tmp1 = load_reg(s, a->rn);
5447
5448 gen(tmp1, tmp1, tmp2);
5449 tcg_temp_free_i32(tmp2);
5450
5451 if (logic_cc) {
5452 gen_logic_CC(tmp1);
5453 }
5454 return store_reg_kind(s, a->rd, tmp1, kind);
5455 }
5456
5457 static bool op_s_rxr_shi(DisasContext *s, arg_s_rrr_shi *a,
5458 void (*gen)(TCGv_i32, TCGv_i32),
5459 int logic_cc, StoreRegKind kind)
5460 {
5461 TCGv_i32 tmp;
5462
5463 tmp = load_reg(s, a->rm);
5464 gen_arm_shift_im(tmp, a->shty, a->shim, logic_cc);
5465
5466 gen(tmp, tmp);
5467 if (logic_cc) {
5468 gen_logic_CC(tmp);
5469 }
5470 return store_reg_kind(s, a->rd, tmp, kind);
5471 }
5472
5473 /*
5474 * Data-processing (register-shifted register)
5475 *
5476 * Operate, with set flags, one register source,
5477 * one register shifted register source, and a destination.
5478 */
5479 static bool op_s_rrr_shr(DisasContext *s, arg_s_rrr_shr *a,
5480 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
5481 int logic_cc, StoreRegKind kind)
5482 {
5483 TCGv_i32 tmp1, tmp2;
5484
5485 tmp1 = load_reg(s, a->rs);
5486 tmp2 = load_reg(s, a->rm);
5487 gen_arm_shift_reg(tmp2, a->shty, tmp1, logic_cc);
5488 tmp1 = load_reg(s, a->rn);
5489
5490 gen(tmp1, tmp1, tmp2);
5491 tcg_temp_free_i32(tmp2);
5492
5493 if (logic_cc) {
5494 gen_logic_CC(tmp1);
5495 }
5496 return store_reg_kind(s, a->rd, tmp1, kind);
5497 }
5498
5499 static bool op_s_rxr_shr(DisasContext *s, arg_s_rrr_shr *a,
5500 void (*gen)(TCGv_i32, TCGv_i32),
5501 int logic_cc, StoreRegKind kind)
5502 {
5503 TCGv_i32 tmp1, tmp2;
5504
5505 tmp1 = load_reg(s, a->rs);
5506 tmp2 = load_reg(s, a->rm);
5507 gen_arm_shift_reg(tmp2, a->shty, tmp1, logic_cc);
5508
5509 gen(tmp2, tmp2);
5510 if (logic_cc) {
5511 gen_logic_CC(tmp2);
5512 }
5513 return store_reg_kind(s, a->rd, tmp2, kind);
5514 }
5515
5516 /*
5517 * Data-processing (immediate)
5518 *
5519 * Operate, with set flags, one register source,
5520 * one rotated immediate, and a destination.
5521 *
5522 * Note that logic_cc && a->rot setting CF based on the msb of the
5523 * immediate is the reason why we must pass in the unrotated form
5524 * of the immediate.
5525 */
5526 static bool op_s_rri_rot(DisasContext *s, arg_s_rri_rot *a,
5527 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
5528 int logic_cc, StoreRegKind kind)
5529 {
5530 TCGv_i32 tmp1, tmp2;
5531 uint32_t imm;
5532
5533 imm = ror32(a->imm, a->rot);
5534 if (logic_cc && a->rot) {
5535 tcg_gen_movi_i32(cpu_CF, imm >> 31);
5536 }
5537 tmp2 = tcg_const_i32(imm);
5538 tmp1 = load_reg(s, a->rn);
5539
5540 gen(tmp1, tmp1, tmp2);
5541 tcg_temp_free_i32(tmp2);
5542
5543 if (logic_cc) {
5544 gen_logic_CC(tmp1);
5545 }
5546 return store_reg_kind(s, a->rd, tmp1, kind);
5547 }
5548
5549 static bool op_s_rxi_rot(DisasContext *s, arg_s_rri_rot *a,
5550 void (*gen)(TCGv_i32, TCGv_i32),
5551 int logic_cc, StoreRegKind kind)
5552 {
5553 TCGv_i32 tmp;
5554 uint32_t imm;
5555
5556 imm = ror32(a->imm, a->rot);
5557 if (logic_cc && a->rot) {
5558 tcg_gen_movi_i32(cpu_CF, imm >> 31);
5559 }
5560 tmp = tcg_const_i32(imm);
5561
5562 gen(tmp, tmp);
5563 if (logic_cc) {
5564 gen_logic_CC(tmp);
5565 }
5566 return store_reg_kind(s, a->rd, tmp, kind);
5567 }
5568
5569 #define DO_ANY3(NAME, OP, L, K) \
5570 static bool trans_##NAME##_rrri(DisasContext *s, arg_s_rrr_shi *a) \
5571 { StoreRegKind k = (K); return op_s_rrr_shi(s, a, OP, L, k); } \
5572 static bool trans_##NAME##_rrrr(DisasContext *s, arg_s_rrr_shr *a) \
5573 { StoreRegKind k = (K); return op_s_rrr_shr(s, a, OP, L, k); } \
5574 static bool trans_##NAME##_rri(DisasContext *s, arg_s_rri_rot *a) \
5575 { StoreRegKind k = (K); return op_s_rri_rot(s, a, OP, L, k); }
5576
5577 #define DO_ANY2(NAME, OP, L, K) \
5578 static bool trans_##NAME##_rxri(DisasContext *s, arg_s_rrr_shi *a) \
5579 { StoreRegKind k = (K); return op_s_rxr_shi(s, a, OP, L, k); } \
5580 static bool trans_##NAME##_rxrr(DisasContext *s, arg_s_rrr_shr *a) \
5581 { StoreRegKind k = (K); return op_s_rxr_shr(s, a, OP, L, k); } \
5582 static bool trans_##NAME##_rxi(DisasContext *s, arg_s_rri_rot *a) \
5583 { StoreRegKind k = (K); return op_s_rxi_rot(s, a, OP, L, k); }
5584
5585 #define DO_CMP2(NAME, OP, L) \
5586 static bool trans_##NAME##_xrri(DisasContext *s, arg_s_rrr_shi *a) \
5587 { return op_s_rrr_shi(s, a, OP, L, STREG_NONE); } \
5588 static bool trans_##NAME##_xrrr(DisasContext *s, arg_s_rrr_shr *a) \
5589 { return op_s_rrr_shr(s, a, OP, L, STREG_NONE); } \
5590 static bool trans_##NAME##_xri(DisasContext *s, arg_s_rri_rot *a) \
5591 { return op_s_rri_rot(s, a, OP, L, STREG_NONE); }
5592
5593 DO_ANY3(AND, tcg_gen_and_i32, a->s, STREG_NORMAL)
5594 DO_ANY3(EOR, tcg_gen_xor_i32, a->s, STREG_NORMAL)
5595 DO_ANY3(ORR, tcg_gen_or_i32, a->s, STREG_NORMAL)
5596 DO_ANY3(BIC, tcg_gen_andc_i32, a->s, STREG_NORMAL)
5597
5598 DO_ANY3(RSB, a->s ? gen_rsb_CC : gen_rsb, false, STREG_NORMAL)
5599 DO_ANY3(ADC, a->s ? gen_adc_CC : gen_add_carry, false, STREG_NORMAL)
5600 DO_ANY3(SBC, a->s ? gen_sbc_CC : gen_sub_carry, false, STREG_NORMAL)
5601 DO_ANY3(RSC, a->s ? gen_rsc_CC : gen_rsc, false, STREG_NORMAL)
5602
5603 DO_CMP2(TST, tcg_gen_and_i32, true)
5604 DO_CMP2(TEQ, tcg_gen_xor_i32, true)
5605 DO_CMP2(CMN, gen_add_CC, false)
5606 DO_CMP2(CMP, gen_sub_CC, false)
5607
5608 DO_ANY3(ADD, a->s ? gen_add_CC : tcg_gen_add_i32, false,
5609 a->rd == 13 && a->rn == 13 ? STREG_SP_CHECK : STREG_NORMAL)
5610
5611 /*
5612 * Note for the computation of StoreRegKind we return out of the
5613 * middle of the functions that are expanded by DO_ANY3, and that
5614 * we modify a->s via that parameter before it is used by OP.
5615 */
5616 DO_ANY3(SUB, a->s ? gen_sub_CC : tcg_gen_sub_i32, false,
5617 ({
5618 StoreRegKind ret = STREG_NORMAL;
5619 if (a->rd == 15 && a->s) {
5620 /*
5621 * See ALUExceptionReturn:
5622 * In User mode, UNPREDICTABLE; we choose UNDEF.
5623 * In Hyp mode, UNDEFINED.
5624 */
5625 if (IS_USER(s) || s->current_el == 2) {
5626 unallocated_encoding(s);
5627 return true;
5628 }
5629 /* There is no writeback of nzcv to PSTATE. */
5630 a->s = 0;
5631 ret = STREG_EXC_RET;
5632 } else if (a->rd == 13 && a->rn == 13) {
5633 ret = STREG_SP_CHECK;
5634 }
5635 ret;
5636 }))
5637
5638 DO_ANY2(MOV, tcg_gen_mov_i32, a->s,
5639 ({
5640 StoreRegKind ret = STREG_NORMAL;
5641 if (a->rd == 15 && a->s) {
5642 /*
5643 * See ALUExceptionReturn:
5644 * In User mode, UNPREDICTABLE; we choose UNDEF.
5645 * In Hyp mode, UNDEFINED.
5646 */
5647 if (IS_USER(s) || s->current_el == 2) {
5648 unallocated_encoding(s);
5649 return true;
5650 }
5651 /* There is no writeback of nzcv to PSTATE. */
5652 a->s = 0;
5653 ret = STREG_EXC_RET;
5654 } else if (a->rd == 13) {
5655 ret = STREG_SP_CHECK;
5656 }
5657 ret;
5658 }))
5659
5660 DO_ANY2(MVN, tcg_gen_not_i32, a->s, STREG_NORMAL)
5661
5662 /*
5663 * ORN is only available with T32, so there is no register-shifted-register
5664 * form of the insn. Using the DO_ANY3 macro would create an unused function.
5665 */
5666 static bool trans_ORN_rrri(DisasContext *s, arg_s_rrr_shi *a)
5667 {
5668 return op_s_rrr_shi(s, a, tcg_gen_orc_i32, a->s, STREG_NORMAL);
5669 }
5670
5671 static bool trans_ORN_rri(DisasContext *s, arg_s_rri_rot *a)
5672 {
5673 return op_s_rri_rot(s, a, tcg_gen_orc_i32, a->s, STREG_NORMAL);
5674 }
5675
5676 #undef DO_ANY3
5677 #undef DO_ANY2
5678 #undef DO_CMP2
5679
5680 static bool trans_ADR(DisasContext *s, arg_ri *a)
5681 {
5682 store_reg_bx(s, a->rd, add_reg_for_lit(s, 15, a->imm));
5683 return true;
5684 }
5685
5686 static bool trans_MOVW(DisasContext *s, arg_MOVW *a)
5687 {
5688 TCGv_i32 tmp;
5689
5690 if (!ENABLE_ARCH_6T2) {
5691 return false;
5692 }
5693
5694 tmp = tcg_const_i32(a->imm);
5695 store_reg(s, a->rd, tmp);
5696 return true;
5697 }
5698
5699 static bool trans_MOVT(DisasContext *s, arg_MOVW *a)
5700 {
5701 TCGv_i32 tmp;
5702
5703 if (!ENABLE_ARCH_6T2) {
5704 return false;
5705 }
5706
5707 tmp = load_reg(s, a->rd);
5708 tcg_gen_ext16u_i32(tmp, tmp);
5709 tcg_gen_ori_i32(tmp, tmp, a->imm << 16);
5710 store_reg(s, a->rd, tmp);
5711 return true;
5712 }
5713
5714 /*
5715 * v8.1M MVE wide-shifts
5716 */
5717 static bool do_mve_shl_ri(DisasContext *s, arg_mve_shl_ri *a,
5718 WideShiftImmFn *fn)
5719 {
5720 TCGv_i64 rda;
5721 TCGv_i32 rdalo, rdahi;
5722
5723 if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
5724 /* Decode falls through to ORR/MOV UNPREDICTABLE handling */
5725 return false;
5726 }
5727 if (a->rdahi == 15) {
5728 /* These are a different encoding (SQSHL/SRSHR/UQSHL/URSHR) */
5729 return false;
5730 }
5731 if (!dc_isar_feature(aa32_mve, s) ||
5732 !arm_dc_feature(s, ARM_FEATURE_M_MAIN) ||
5733 a->rdahi == 13) {
5734 /* RdaHi == 13 is UNPREDICTABLE; we choose to UNDEF */
5735 unallocated_encoding(s);
5736 return true;
5737 }
5738
5739 if (a->shim == 0) {
5740 a->shim = 32;
5741 }
5742
5743 rda = tcg_temp_new_i64();
5744 rdalo = load_reg(s, a->rdalo);
5745 rdahi = load_reg(s, a->rdahi);
5746 tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
5747
5748 fn(rda, rda, a->shim);
5749
5750 tcg_gen_extrl_i64_i32(rdalo, rda);
5751 tcg_gen_extrh_i64_i32(rdahi, rda);
5752 store_reg(s, a->rdalo, rdalo);
5753 store_reg(s, a->rdahi, rdahi);
5754 tcg_temp_free_i64(rda);
5755
5756 return true;
5757 }
5758
5759 static bool trans_ASRL_ri(DisasContext *s, arg_mve_shl_ri *a)
5760 {
5761 return do_mve_shl_ri(s, a, tcg_gen_sari_i64);
5762 }
5763
5764 static bool trans_LSLL_ri(DisasContext *s, arg_mve_shl_ri *a)
5765 {
5766 return do_mve_shl_ri(s, a, tcg_gen_shli_i64);
5767 }
5768
5769 static bool trans_LSRL_ri(DisasContext *s, arg_mve_shl_ri *a)
5770 {
5771 return do_mve_shl_ri(s, a, tcg_gen_shri_i64);
5772 }
5773
5774 static void gen_mve_sqshll(TCGv_i64 r, TCGv_i64 n, int64_t shift)
5775 {
5776 gen_helper_mve_sqshll(r, cpu_env, n, tcg_constant_i32(shift));
5777 }
5778
5779 static bool trans_SQSHLL_ri(DisasContext *s, arg_mve_shl_ri *a)
5780 {
5781 return do_mve_shl_ri(s, a, gen_mve_sqshll);
5782 }
5783
5784 static void gen_mve_uqshll(TCGv_i64 r, TCGv_i64 n, int64_t shift)
5785 {
5786 gen_helper_mve_uqshll(r, cpu_env, n, tcg_constant_i32(shift));
5787 }
5788
5789 static bool trans_UQSHLL_ri(DisasContext *s, arg_mve_shl_ri *a)
5790 {
5791 return do_mve_shl_ri(s, a, gen_mve_uqshll);
5792 }
5793
5794 static bool trans_SRSHRL_ri(DisasContext *s, arg_mve_shl_ri *a)
5795 {
5796 return do_mve_shl_ri(s, a, gen_srshr64_i64);
5797 }
5798
5799 static bool trans_URSHRL_ri(DisasContext *s, arg_mve_shl_ri *a)
5800 {
5801 return do_mve_shl_ri(s, a, gen_urshr64_i64);
5802 }
5803
5804 static bool do_mve_shl_rr(DisasContext *s, arg_mve_shl_rr *a, WideShiftFn *fn)
5805 {
5806 TCGv_i64 rda;
5807 TCGv_i32 rdalo, rdahi;
5808
5809 if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
5810 /* Decode falls through to ORR/MOV UNPREDICTABLE handling */
5811 return false;
5812 }
5813 if (a->rdahi == 15) {
5814 /* These are a different encoding (SQSHL/SRSHR/UQSHL/URSHR) */
5815 return false;
5816 }
5817 if (!dc_isar_feature(aa32_mve, s) ||
5818 !arm_dc_feature(s, ARM_FEATURE_M_MAIN) ||
5819 a->rdahi == 13 || a->rm == 13 || a->rm == 15 ||
5820 a->rm == a->rdahi || a->rm == a->rdalo) {
5821 /* These rdahi/rdalo/rm cases are UNPREDICTABLE; we choose to UNDEF */
5822 unallocated_encoding(s);
5823 return true;
5824 }
5825
5826 rda = tcg_temp_new_i64();
5827 rdalo = load_reg(s, a->rdalo);
5828 rdahi = load_reg(s, a->rdahi);
5829 tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
5830
5831 /* The helper takes care of the sign-extension of the low 8 bits of Rm */
5832 fn(rda, cpu_env, rda, cpu_R[a->rm]);
5833
5834 tcg_gen_extrl_i64_i32(rdalo, rda);
5835 tcg_gen_extrh_i64_i32(rdahi, rda);
5836 store_reg(s, a->rdalo, rdalo);
5837 store_reg(s, a->rdahi, rdahi);
5838 tcg_temp_free_i64(rda);
5839
5840 return true;
5841 }
5842
5843 static bool trans_LSLL_rr(DisasContext *s, arg_mve_shl_rr *a)
5844 {
5845 return do_mve_shl_rr(s, a, gen_helper_mve_ushll);
5846 }
5847
5848 static bool trans_ASRL_rr(DisasContext *s, arg_mve_shl_rr *a)
5849 {
5850 return do_mve_shl_rr(s, a, gen_helper_mve_sshrl);
5851 }
5852
5853 static bool trans_UQRSHLL64_rr(DisasContext *s, arg_mve_shl_rr *a)
5854 {
5855 return do_mve_shl_rr(s, a, gen_helper_mve_uqrshll);
5856 }
5857
5858 static bool trans_SQRSHRL64_rr(DisasContext *s, arg_mve_shl_rr *a)
5859 {
5860 return do_mve_shl_rr(s, a, gen_helper_mve_sqrshrl);
5861 }
5862
5863 static bool trans_UQRSHLL48_rr(DisasContext *s, arg_mve_shl_rr *a)
5864 {
5865 return do_mve_shl_rr(s, a, gen_helper_mve_uqrshll48);
5866 }
5867
5868 static bool trans_SQRSHRL48_rr(DisasContext *s, arg_mve_shl_rr *a)
5869 {
5870 return do_mve_shl_rr(s, a, gen_helper_mve_sqrshrl48);
5871 }
5872
5873 static bool do_mve_sh_ri(DisasContext *s, arg_mve_sh_ri *a, ShiftImmFn *fn)
5874 {
5875 if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
5876 /* Decode falls through to ORR/MOV UNPREDICTABLE handling */
5877 return false;
5878 }
5879 if (!dc_isar_feature(aa32_mve, s) ||
5880 !arm_dc_feature(s, ARM_FEATURE_M_MAIN) ||
5881 a->rda == 13 || a->rda == 15) {
5882 /* These rda cases are UNPREDICTABLE; we choose to UNDEF */
5883 unallocated_encoding(s);
5884 return true;
5885 }
5886
5887 if (a->shim == 0) {
5888 a->shim = 32;
5889 }
5890 fn(cpu_R[a->rda], cpu_R[a->rda], a->shim);
5891
5892 return true;
5893 }
5894
5895 static bool trans_URSHR_ri(DisasContext *s, arg_mve_sh_ri *a)
5896 {
5897 return do_mve_sh_ri(s, a, gen_urshr32_i32);
5898 }
5899
5900 static bool trans_SRSHR_ri(DisasContext *s, arg_mve_sh_ri *a)
5901 {
5902 return do_mve_sh_ri(s, a, gen_srshr32_i32);
5903 }
5904
5905 static void gen_mve_sqshl(TCGv_i32 r, TCGv_i32 n, int32_t shift)
5906 {
5907 gen_helper_mve_sqshl(r, cpu_env, n, tcg_constant_i32(shift));
5908 }
5909
5910 static bool trans_SQSHL_ri(DisasContext *s, arg_mve_sh_ri *a)
5911 {
5912 return do_mve_sh_ri(s, a, gen_mve_sqshl);
5913 }
5914
5915 static void gen_mve_uqshl(TCGv_i32 r, TCGv_i32 n, int32_t shift)
5916 {
5917 gen_helper_mve_uqshl(r, cpu_env, n, tcg_constant_i32(shift));
5918 }
5919
5920 static bool trans_UQSHL_ri(DisasContext *s, arg_mve_sh_ri *a)
5921 {
5922 return do_mve_sh_ri(s, a, gen_mve_uqshl);
5923 }
5924
5925 static bool do_mve_sh_rr(DisasContext *s, arg_mve_sh_rr *a, ShiftFn *fn)
5926 {
5927 if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
5928 /* Decode falls through to ORR/MOV UNPREDICTABLE handling */
5929 return false;
5930 }
5931 if (!dc_isar_feature(aa32_mve, s) ||
5932 !arm_dc_feature(s, ARM_FEATURE_M_MAIN) ||
5933 a->rda == 13 || a->rda == 15 || a->rm == 13 || a->rm == 15 ||
5934 a->rm == a->rda) {
5935 /* These rda/rm cases are UNPREDICTABLE; we choose to UNDEF */
5936 unallocated_encoding(s);
5937 return true;
5938 }
5939
5940 /* The helper takes care of the sign-extension of the low 8 bits of Rm */
5941 fn(cpu_R[a->rda], cpu_env, cpu_R[a->rda], cpu_R[a->rm]);
5942 return true;
5943 }
5944
5945 static bool trans_SQRSHR_rr(DisasContext *s, arg_mve_sh_rr *a)
5946 {
5947 return do_mve_sh_rr(s, a, gen_helper_mve_sqrshr);
5948 }
5949
5950 static bool trans_UQRSHL_rr(DisasContext *s, arg_mve_sh_rr *a)
5951 {
5952 return do_mve_sh_rr(s, a, gen_helper_mve_uqrshl);
5953 }
5954
5955 /*
5956 * Multiply and multiply accumulate
5957 */
5958
5959 static bool op_mla(DisasContext *s, arg_s_rrrr *a, bool add)
5960 {
5961 TCGv_i32 t1, t2;
5962
5963 t1 = load_reg(s, a->rn);
5964 t2 = load_reg(s, a->rm);
5965 tcg_gen_mul_i32(t1, t1, t2);
5966 tcg_temp_free_i32(t2);
5967 if (add) {
5968 t2 = load_reg(s, a->ra);
5969 tcg_gen_add_i32(t1, t1, t2);
5970 tcg_temp_free_i32(t2);
5971 }
5972 if (a->s) {
5973 gen_logic_CC(t1);
5974 }
5975 store_reg(s, a->rd, t1);
5976 return true;
5977 }
5978
5979 static bool trans_MUL(DisasContext *s, arg_MUL *a)
5980 {
5981 return op_mla(s, a, false);
5982 }
5983
5984 static bool trans_MLA(DisasContext *s, arg_MLA *a)
5985 {
5986 return op_mla(s, a, true);
5987 }
5988
5989 static bool trans_MLS(DisasContext *s, arg_MLS *a)
5990 {
5991 TCGv_i32 t1, t2;
5992
5993 if (!ENABLE_ARCH_6T2) {
5994 return false;
5995 }
5996 t1 = load_reg(s, a->rn);
5997 t2 = load_reg(s, a->rm);
5998 tcg_gen_mul_i32(t1, t1, t2);
5999 tcg_temp_free_i32(t2);
6000 t2 = load_reg(s, a->ra);
6001 tcg_gen_sub_i32(t1, t2, t1);
6002 tcg_temp_free_i32(t2);
6003 store_reg(s, a->rd, t1);
6004 return true;
6005 }
6006
6007 static bool op_mlal(DisasContext *s, arg_s_rrrr *a, bool uns, bool add)
6008 {
6009 TCGv_i32 t0, t1, t2, t3;
6010
6011 t0 = load_reg(s, a->rm);
6012 t1 = load_reg(s, a->rn);
6013 if (uns) {
6014 tcg_gen_mulu2_i32(t0, t1, t0, t1);
6015 } else {
6016 tcg_gen_muls2_i32(t0, t1, t0, t1);
6017 }
6018 if (add) {
6019 t2 = load_reg(s, a->ra);
6020 t3 = load_reg(s, a->rd);
6021 tcg_gen_add2_i32(t0, t1, t0, t1, t2, t3);
6022 tcg_temp_free_i32(t2);
6023 tcg_temp_free_i32(t3);
6024 }
6025 if (a->s) {
6026 gen_logicq_cc(t0, t1);
6027 }
6028 store_reg(s, a->ra, t0);
6029 store_reg(s, a->rd, t1);
6030 return true;
6031 }
6032
6033 static bool trans_UMULL(DisasContext *s, arg_UMULL *a)
6034 {
6035 return op_mlal(s, a, true, false);
6036 }
6037
6038 static bool trans_SMULL(DisasContext *s, arg_SMULL *a)
6039 {
6040 return op_mlal(s, a, false, false);
6041 }
6042
6043 static bool trans_UMLAL(DisasContext *s, arg_UMLAL *a)
6044 {
6045 return op_mlal(s, a, true, true);
6046 }
6047
6048 static bool trans_SMLAL(DisasContext *s, arg_SMLAL *a)
6049 {
6050 return op_mlal(s, a, false, true);
6051 }
6052
6053 static bool trans_UMAAL(DisasContext *s, arg_UMAAL *a)
6054 {
6055 TCGv_i32 t0, t1, t2, zero;
6056
6057 if (s->thumb
6058 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
6059 : !ENABLE_ARCH_6) {
6060 return false;
6061 }
6062
6063 t0 = load_reg(s, a->rm);
6064 t1 = load_reg(s, a->rn);
6065 tcg_gen_mulu2_i32(t0, t1, t0, t1);
6066 zero = tcg_const_i32(0);
6067 t2 = load_reg(s, a->ra);
6068 tcg_gen_add2_i32(t0, t1, t0, t1, t2, zero);
6069 tcg_temp_free_i32(t2);
6070 t2 = load_reg(s, a->rd);
6071 tcg_gen_add2_i32(t0, t1, t0, t1, t2, zero);
6072 tcg_temp_free_i32(t2);
6073 tcg_temp_free_i32(zero);
6074 store_reg(s, a->ra, t0);
6075 store_reg(s, a->rd, t1);
6076 return true;
6077 }
6078
6079 /*
6080 * Saturating addition and subtraction
6081 */
6082
6083 static bool op_qaddsub(DisasContext *s, arg_rrr *a, bool add, bool doub)
6084 {
6085 TCGv_i32 t0, t1;
6086
6087 if (s->thumb
6088 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
6089 : !ENABLE_ARCH_5TE) {
6090 return false;
6091 }
6092
6093 t0 = load_reg(s, a->rm);
6094 t1 = load_reg(s, a->rn);
6095 if (doub) {
6096 gen_helper_add_saturate(t1, cpu_env, t1, t1);
6097 }
6098 if (add) {
6099 gen_helper_add_saturate(t0, cpu_env, t0, t1);
6100 } else {
6101 gen_helper_sub_saturate(t0, cpu_env, t0, t1);
6102 }
6103 tcg_temp_free_i32(t1);
6104 store_reg(s, a->rd, t0);
6105 return true;
6106 }
6107
6108 #define DO_QADDSUB(NAME, ADD, DOUB) \
6109 static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
6110 { \
6111 return op_qaddsub(s, a, ADD, DOUB); \
6112 }
6113
6114 DO_QADDSUB(QADD, true, false)
6115 DO_QADDSUB(QSUB, false, false)
6116 DO_QADDSUB(QDADD, true, true)
6117 DO_QADDSUB(QDSUB, false, true)
6118
6119 #undef DO_QADDSUB
6120
6121 /*
6122 * Halfword multiply and multiply accumulate
6123 */
6124
6125 static bool op_smlaxxx(DisasContext *s, arg_rrrr *a,
6126 int add_long, bool nt, bool mt)
6127 {
6128 TCGv_i32 t0, t1, tl, th;
6129
6130 if (s->thumb
6131 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
6132 : !ENABLE_ARCH_5TE) {
6133 return false;
6134 }
6135
6136 t0 = load_reg(s, a->rn);
6137 t1 = load_reg(s, a->rm);
6138 gen_mulxy(t0, t1, nt, mt);
6139 tcg_temp_free_i32(t1);
6140
6141 switch (add_long) {
6142 case 0:
6143 store_reg(s, a->rd, t0);
6144 break;
6145 case 1:
6146 t1 = load_reg(s, a->ra);
6147 gen_helper_add_setq(t0, cpu_env, t0, t1);
6148 tcg_temp_free_i32(t1);
6149 store_reg(s, a->rd, t0);
6150 break;
6151 case 2:
6152 tl = load_reg(s, a->ra);
6153 th = load_reg(s, a->rd);
6154 /* Sign-extend the 32-bit product to 64 bits. */
6155 t1 = tcg_temp_new_i32();
6156 tcg_gen_sari_i32(t1, t0, 31);
6157 tcg_gen_add2_i32(tl, th, tl, th, t0, t1);
6158 tcg_temp_free_i32(t0);
6159 tcg_temp_free_i32(t1);
6160 store_reg(s, a->ra, tl);
6161 store_reg(s, a->rd, th);
6162 break;
6163 default:
6164 g_assert_not_reached();
6165 }
6166 return true;
6167 }
6168
6169 #define DO_SMLAX(NAME, add, nt, mt) \
6170 static bool trans_##NAME(DisasContext *s, arg_rrrr *a) \
6171 { \
6172 return op_smlaxxx(s, a, add, nt, mt); \
6173 }
6174
6175 DO_SMLAX(SMULBB, 0, 0, 0)
6176 DO_SMLAX(SMULBT, 0, 0, 1)
6177 DO_SMLAX(SMULTB, 0, 1, 0)
6178 DO_SMLAX(SMULTT, 0, 1, 1)
6179
6180 DO_SMLAX(SMLABB, 1, 0, 0)
6181 DO_SMLAX(SMLABT, 1, 0, 1)
6182 DO_SMLAX(SMLATB, 1, 1, 0)
6183 DO_SMLAX(SMLATT, 1, 1, 1)
6184
6185 DO_SMLAX(SMLALBB, 2, 0, 0)
6186 DO_SMLAX(SMLALBT, 2, 0, 1)
6187 DO_SMLAX(SMLALTB, 2, 1, 0)
6188 DO_SMLAX(SMLALTT, 2, 1, 1)
6189
6190 #undef DO_SMLAX
6191
6192 static bool op_smlawx(DisasContext *s, arg_rrrr *a, bool add, bool mt)
6193 {
6194 TCGv_i32 t0, t1;
6195
6196 if (!ENABLE_ARCH_5TE) {
6197 return false;
6198 }
6199
6200 t0 = load_reg(s, a->rn);
6201 t1 = load_reg(s, a->rm);
6202 /*
6203 * Since the nominal result is product<47:16>, shift the 16-bit
6204 * input up by 16 bits, so that the result is at product<63:32>.
6205 */
6206 if (mt) {
6207 tcg_gen_andi_i32(t1, t1, 0xffff0000);
6208 } else {
6209 tcg_gen_shli_i32(t1, t1, 16);
6210 }
6211 tcg_gen_muls2_i32(t0, t1, t0, t1);
6212 tcg_temp_free_i32(t0);
6213 if (add) {
6214 t0 = load_reg(s, a->ra);
6215 gen_helper_add_setq(t1, cpu_env, t1, t0);
6216 tcg_temp_free_i32(t0);
6217 }
6218 store_reg(s, a->rd, t1);
6219 return true;
6220 }
6221
6222 #define DO_SMLAWX(NAME, add, mt) \
6223 static bool trans_##NAME(DisasContext *s, arg_rrrr *a) \
6224 { \
6225 return op_smlawx(s, a, add, mt); \
6226 }
6227
6228 DO_SMLAWX(SMULWB, 0, 0)
6229 DO_SMLAWX(SMULWT, 0, 1)
6230 DO_SMLAWX(SMLAWB, 1, 0)
6231 DO_SMLAWX(SMLAWT, 1, 1)
6232
6233 #undef DO_SMLAWX
6234
6235 /*
6236 * MSR (immediate) and hints
6237 */
6238
6239 static bool trans_YIELD(DisasContext *s, arg_YIELD *a)
6240 {
6241 /*
6242 * When running single-threaded TCG code, use the helper to ensure that
6243 * the next round-robin scheduled vCPU gets a crack. When running in
6244 * MTTCG we don't generate jumps to the helper as it won't affect the
6245 * scheduling of other vCPUs.
6246 */
6247 if (!(tb_cflags(s->base.tb) & CF_PARALLEL)) {
6248 gen_set_pc_im(s, s->base.pc_next);
6249 s->base.is_jmp = DISAS_YIELD;
6250 }
6251 return true;
6252 }
6253
6254 static bool trans_WFE(DisasContext *s, arg_WFE *a)
6255 {
6256 /*
6257 * When running single-threaded TCG code, use the helper to ensure that
6258 * the next round-robin scheduled vCPU gets a crack. In MTTCG mode we
6259 * just skip this instruction. Currently the SEV/SEVL instructions,
6260 * which are *one* of many ways to wake the CPU from WFE, are not
6261 * implemented so we can't sleep like WFI does.
6262 */
6263 if (!(tb_cflags(s->base.tb) & CF_PARALLEL)) {
6264 gen_set_pc_im(s, s->base.pc_next);
6265 s->base.is_jmp = DISAS_WFE;
6266 }
6267 return true;
6268 }
6269
6270 static bool trans_WFI(DisasContext *s, arg_WFI *a)
6271 {
6272 /* For WFI, halt the vCPU until an IRQ. */
6273 gen_set_pc_im(s, s->base.pc_next);
6274 s->base.is_jmp = DISAS_WFI;
6275 return true;
6276 }
6277
6278 static bool trans_NOP(DisasContext *s, arg_NOP *a)
6279 {
6280 return true;
6281 }
6282
6283 static bool trans_MSR_imm(DisasContext *s, arg_MSR_imm *a)
6284 {
6285 uint32_t val = ror32(a->imm, a->rot * 2);
6286 uint32_t mask = msr_mask(s, a->mask, a->r);
6287
6288 if (gen_set_psr_im(s, mask, a->r, val)) {
6289 unallocated_encoding(s);
6290 }
6291 return true;
6292 }
6293
6294 /*
6295 * Cyclic Redundancy Check
6296 */
6297
6298 static bool op_crc32(DisasContext *s, arg_rrr *a, bool c, MemOp sz)
6299 {
6300 TCGv_i32 t1, t2, t3;
6301
6302 if (!dc_isar_feature(aa32_crc32, s)) {
6303 return false;
6304 }
6305
6306 t1 = load_reg(s, a->rn);
6307 t2 = load_reg(s, a->rm);
6308 switch (sz) {
6309 case MO_8:
6310 gen_uxtb(t2);
6311 break;
6312 case MO_16:
6313 gen_uxth(t2);
6314 break;
6315 case MO_32:
6316 break;
6317 default:
6318 g_assert_not_reached();
6319 }
6320 t3 = tcg_const_i32(1 << sz);
6321 if (c) {
6322 gen_helper_crc32c(t1, t1, t2, t3);
6323 } else {
6324 gen_helper_crc32(t1, t1, t2, t3);
6325 }
6326 tcg_temp_free_i32(t2);
6327 tcg_temp_free_i32(t3);
6328 store_reg(s, a->rd, t1);
6329 return true;
6330 }
6331
6332 #define DO_CRC32(NAME, c, sz) \
6333 static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
6334 { return op_crc32(s, a, c, sz); }
6335
6336 DO_CRC32(CRC32B, false, MO_8)
6337 DO_CRC32(CRC32H, false, MO_16)
6338 DO_CRC32(CRC32W, false, MO_32)
6339 DO_CRC32(CRC32CB, true, MO_8)
6340 DO_CRC32(CRC32CH, true, MO_16)
6341 DO_CRC32(CRC32CW, true, MO_32)
6342
6343 #undef DO_CRC32
6344
6345 /*
6346 * Miscellaneous instructions
6347 */
6348
6349 static bool trans_MRS_bank(DisasContext *s, arg_MRS_bank *a)
6350 {
6351 if (arm_dc_feature(s, ARM_FEATURE_M)) {
6352 return false;
6353 }
6354 gen_mrs_banked(s, a->r, a->sysm, a->rd);
6355 return true;
6356 }
6357
6358 static bool trans_MSR_bank(DisasContext *s, arg_MSR_bank *a)
6359 {
6360 if (arm_dc_feature(s, ARM_FEATURE_M)) {
6361 return false;
6362 }
6363 gen_msr_banked(s, a->r, a->sysm, a->rn);
6364 return true;
6365 }
6366
6367 static bool trans_MRS_reg(DisasContext *s, arg_MRS_reg *a)
6368 {
6369 TCGv_i32 tmp;
6370
6371 if (arm_dc_feature(s, ARM_FEATURE_M)) {
6372 return false;
6373 }
6374 if (a->r) {
6375 if (IS_USER(s)) {
6376 unallocated_encoding(s);
6377 return true;
6378 }
6379 tmp = load_cpu_field(spsr);
6380 } else {
6381 tmp = tcg_temp_new_i32();
6382 gen_helper_cpsr_read(tmp, cpu_env);
6383 }
6384 store_reg(s, a->rd, tmp);
6385 return true;
6386 }
6387
6388 static bool trans_MSR_reg(DisasContext *s, arg_MSR_reg *a)
6389 {
6390 TCGv_i32 tmp;
6391 uint32_t mask = msr_mask(s, a->mask, a->r);
6392
6393 if (arm_dc_feature(s, ARM_FEATURE_M)) {
6394 return false;
6395 }
6396 tmp = load_reg(s, a->rn);
6397 if (gen_set_psr(s, mask, a->r, tmp)) {
6398 unallocated_encoding(s);
6399 }
6400 return true;
6401 }
6402
6403 static bool trans_MRS_v7m(DisasContext *s, arg_MRS_v7m *a)
6404 {
6405 TCGv_i32 tmp;
6406
6407 if (!arm_dc_feature(s, ARM_FEATURE_M)) {
6408 return false;
6409 }
6410 tmp = tcg_const_i32(a->sysm);
6411 gen_helper_v7m_mrs(tmp, cpu_env, tmp);
6412 store_reg(s, a->rd, tmp);
6413 return true;
6414 }
6415
6416 static bool trans_MSR_v7m(DisasContext *s, arg_MSR_v7m *a)
6417 {
6418 TCGv_i32 addr, reg;
6419
6420 if (!arm_dc_feature(s, ARM_FEATURE_M)) {
6421 return false;
6422 }
6423 addr = tcg_const_i32((a->mask << 10) | a->sysm);
6424 reg = load_reg(s, a->rn);
6425 gen_helper_v7m_msr(cpu_env, addr, reg);
6426 tcg_temp_free_i32(addr);
6427 tcg_temp_free_i32(reg);
6428 /* If we wrote to CONTROL, the EL might have changed */
6429 gen_helper_rebuild_hflags_m32_newel(cpu_env);
6430 gen_lookup_tb(s);
6431 return true;
6432 }
6433
6434 static bool trans_BX(DisasContext *s, arg_BX *a)
6435 {
6436 if (!ENABLE_ARCH_4T) {
6437 return false;
6438 }
6439 gen_bx_excret(s, load_reg(s, a->rm));
6440 return true;
6441 }
6442
6443 static bool trans_BXJ(DisasContext *s, arg_BXJ *a)
6444 {
6445 if (!ENABLE_ARCH_5J || arm_dc_feature(s, ARM_FEATURE_M)) {
6446 return false;
6447 }
6448 /*
6449 * v7A allows BXJ to be trapped via HSTR.TJDBX. We don't waste a
6450 * TBFLAGS bit on a basically-never-happens case, so call a helper
6451 * function to check for the trap and raise the exception if needed
6452 * (passing it the register number for the syndrome value).
6453 * v8A doesn't have this HSTR bit.
6454 */
6455 if (!arm_dc_feature(s, ARM_FEATURE_V8) &&
6456 arm_dc_feature(s, ARM_FEATURE_EL2) &&
6457 s->current_el < 2 && s->ns) {
6458 gen_helper_check_bxj_trap(cpu_env, tcg_constant_i32(a->rm));
6459 }
6460 /* Trivial implementation equivalent to bx. */
6461 gen_bx(s, load_reg(s, a->rm));
6462 return true;
6463 }
6464
6465 static bool trans_BLX_r(DisasContext *s, arg_BLX_r *a)
6466 {
6467 TCGv_i32 tmp;
6468
6469 if (!ENABLE_ARCH_5) {
6470 return false;
6471 }
6472 tmp = load_reg(s, a->rm);
6473 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | s->thumb);
6474 gen_bx(s, tmp);
6475 return true;
6476 }
6477
6478 /*
6479 * BXNS/BLXNS: only exist for v8M with the security extensions,
6480 * and always UNDEF if NonSecure. We don't implement these in
6481 * the user-only mode either (in theory you can use them from
6482 * Secure User mode but they are too tied in to system emulation).
6483 */
6484 static bool trans_BXNS(DisasContext *s, arg_BXNS *a)
6485 {
6486 if (!s->v8m_secure || IS_USER_ONLY) {
6487 unallocated_encoding(s);
6488 } else {
6489 gen_bxns(s, a->rm);
6490 }
6491 return true;
6492 }
6493
6494 static bool trans_BLXNS(DisasContext *s, arg_BLXNS *a)
6495 {
6496 if (!s->v8m_secure || IS_USER_ONLY) {
6497 unallocated_encoding(s);
6498 } else {
6499 gen_blxns(s, a->rm);
6500 }
6501 return true;
6502 }
6503
6504 static bool trans_CLZ(DisasContext *s, arg_CLZ *a)
6505 {
6506 TCGv_i32 tmp;
6507
6508 if (!ENABLE_ARCH_5) {
6509 return false;
6510 }
6511 tmp = load_reg(s, a->rm);
6512 tcg_gen_clzi_i32(tmp, tmp, 32);
6513 store_reg(s, a->rd, tmp);
6514 return true;
6515 }
6516
6517 static bool trans_ERET(DisasContext *s, arg_ERET *a)
6518 {
6519 TCGv_i32 tmp;
6520
6521 if (!arm_dc_feature(s, ARM_FEATURE_V7VE)) {
6522 return false;
6523 }
6524 if (IS_USER(s)) {
6525 unallocated_encoding(s);
6526 return true;
6527 }
6528 if (s->current_el == 2) {
6529 /* ERET from Hyp uses ELR_Hyp, not LR */
6530 tmp = load_cpu_field(elr_el[2]);
6531 } else {
6532 tmp = load_reg(s, 14);
6533 }
6534 gen_exception_return(s, tmp);
6535 return true;
6536 }
6537
6538 static bool trans_HLT(DisasContext *s, arg_HLT *a)
6539 {
6540 gen_hlt(s, a->imm);
6541 return true;
6542 }
6543
6544 static bool trans_BKPT(DisasContext *s, arg_BKPT *a)
6545 {
6546 if (!ENABLE_ARCH_5) {
6547 return false;
6548 }
6549 /* BKPT is OK with ECI set and leaves it untouched */
6550 s->eci_handled = true;
6551 if (arm_dc_feature(s, ARM_FEATURE_M) &&
6552 semihosting_enabled() &&
6553 #ifndef CONFIG_USER_ONLY
6554 !IS_USER(s) &&
6555 #endif
6556 (a->imm == 0xab)) {
6557 gen_exception_internal_insn(s, s->pc_curr, EXCP_SEMIHOST);
6558 } else {
6559 gen_exception_bkpt_insn(s, syn_aa32_bkpt(a->imm, false));
6560 }
6561 return true;
6562 }
6563
6564 static bool trans_HVC(DisasContext *s, arg_HVC *a)
6565 {
6566 if (!ENABLE_ARCH_7 || arm_dc_feature(s, ARM_FEATURE_M)) {
6567 return false;
6568 }
6569 if (IS_USER(s)) {
6570 unallocated_encoding(s);
6571 } else {
6572 gen_hvc(s, a->imm);
6573 }
6574 return true;
6575 }
6576
6577 static bool trans_SMC(DisasContext *s, arg_SMC *a)
6578 {
6579 if (!ENABLE_ARCH_6K || arm_dc_feature(s, ARM_FEATURE_M)) {
6580 return false;
6581 }
6582 if (IS_USER(s)) {
6583 unallocated_encoding(s);
6584 } else {
6585 gen_smc(s);
6586 }
6587 return true;
6588 }
6589
6590 static bool trans_SG(DisasContext *s, arg_SG *a)
6591 {
6592 if (!arm_dc_feature(s, ARM_FEATURE_M) ||
6593 !arm_dc_feature(s, ARM_FEATURE_V8)) {
6594 return false;
6595 }
6596 /*
6597 * SG (v8M only)
6598 * The bulk of the behaviour for this instruction is implemented
6599 * in v7m_handle_execute_nsc(), which deals with the insn when
6600 * it is executed by a CPU in non-secure state from memory
6601 * which is Secure & NonSecure-Callable.
6602 * Here we only need to handle the remaining cases:
6603 * * in NS memory (including the "security extension not
6604 * implemented" case) : NOP
6605 * * in S memory but CPU already secure (clear IT bits)
6606 * We know that the attribute for the memory this insn is
6607 * in must match the current CPU state, because otherwise
6608 * get_phys_addr_pmsav8 would have generated an exception.
6609 */
6610 if (s->v8m_secure) {
6611 /* Like the IT insn, we don't need to generate any code */
6612 s->condexec_cond = 0;
6613 s->condexec_mask = 0;
6614 }
6615 return true;
6616 }
6617
6618 static bool trans_TT(DisasContext *s, arg_TT *a)
6619 {
6620 TCGv_i32 addr, tmp;
6621
6622 if (!arm_dc_feature(s, ARM_FEATURE_M) ||
6623 !arm_dc_feature(s, ARM_FEATURE_V8)) {
6624 return false;
6625 }
6626 if (a->rd == 13 || a->rd == 15 || a->rn == 15) {
6627 /* We UNDEF for these UNPREDICTABLE cases */
6628 unallocated_encoding(s);
6629 return true;
6630 }
6631 if (a->A && !s->v8m_secure) {
6632 /* This case is UNDEFINED. */
6633 unallocated_encoding(s);
6634 return true;
6635 }
6636
6637 addr = load_reg(s, a->rn);
6638 tmp = tcg_const_i32((a->A << 1) | a->T);
6639 gen_helper_v7m_tt(tmp, cpu_env, addr, tmp);
6640 tcg_temp_free_i32(addr);
6641 store_reg(s, a->rd, tmp);
6642 return true;
6643 }
6644
6645 /*
6646 * Load/store register index
6647 */
6648
6649 static ISSInfo make_issinfo(DisasContext *s, int rd, bool p, bool w)
6650 {
6651 ISSInfo ret;
6652
6653 /* ISS not valid if writeback */
6654 if (p && !w) {
6655 ret = rd;
6656 if (s->base.pc_next - s->pc_curr == 2) {
6657 ret |= ISSIs16Bit;
6658 }
6659 } else {
6660 ret = ISSInvalid;
6661 }
6662 return ret;
6663 }
6664
6665 static TCGv_i32 op_addr_rr_pre(DisasContext *s, arg_ldst_rr *a)
6666 {
6667 TCGv_i32 addr = load_reg(s, a->rn);
6668
6669 if (s->v8m_stackcheck && a->rn == 13 && a->w) {
6670 gen_helper_v8m_stackcheck(cpu_env, addr);
6671 }
6672
6673 if (a->p) {
6674 TCGv_i32 ofs = load_reg(s, a->rm);
6675 gen_arm_shift_im(ofs, a->shtype, a->shimm, 0);
6676 if (a->u) {
6677 tcg_gen_add_i32(addr, addr, ofs);
6678 } else {
6679 tcg_gen_sub_i32(addr, addr, ofs);
6680 }
6681 tcg_temp_free_i32(ofs);
6682 }
6683 return addr;
6684 }
6685
6686 static void op_addr_rr_post(DisasContext *s, arg_ldst_rr *a,
6687 TCGv_i32 addr, int address_offset)
6688 {
6689 if (!a->p) {
6690 TCGv_i32 ofs = load_reg(s, a->rm);
6691 gen_arm_shift_im(ofs, a->shtype, a->shimm, 0);
6692 if (a->u) {
6693 tcg_gen_add_i32(addr, addr, ofs);
6694 } else {
6695 tcg_gen_sub_i32(addr, addr, ofs);
6696 }
6697 tcg_temp_free_i32(ofs);
6698 } else if (!a->w) {
6699 tcg_temp_free_i32(addr);
6700 return;
6701 }
6702 tcg_gen_addi_i32(addr, addr, address_offset);
6703 store_reg(s, a->rn, addr);
6704 }
6705
6706 static bool op_load_rr(DisasContext *s, arg_ldst_rr *a,
6707 MemOp mop, int mem_idx)
6708 {
6709 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w);
6710 TCGv_i32 addr, tmp;
6711
6712 addr = op_addr_rr_pre(s, a);
6713
6714 tmp = tcg_temp_new_i32();
6715 gen_aa32_ld_i32(s, tmp, addr, mem_idx, mop);
6716 disas_set_da_iss(s, mop, issinfo);
6717
6718 /*
6719 * Perform base writeback before the loaded value to
6720 * ensure correct behavior with overlapping index registers.
6721 */
6722 op_addr_rr_post(s, a, addr, 0);
6723 store_reg_from_load(s, a->rt, tmp);
6724 return true;
6725 }
6726
6727 static bool op_store_rr(DisasContext *s, arg_ldst_rr *a,
6728 MemOp mop, int mem_idx)
6729 {
6730 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w) | ISSIsWrite;
6731 TCGv_i32 addr, tmp;
6732
6733 /*
6734 * In Thumb encodings of stores Rn=1111 is UNDEF; for Arm it
6735 * is either UNPREDICTABLE or has defined behaviour
6736 */
6737 if (s->thumb && a->rn == 15) {
6738 return false;
6739 }
6740
6741 addr = op_addr_rr_pre(s, a);
6742
6743 tmp = load_reg(s, a->rt);
6744 gen_aa32_st_i32(s, tmp, addr, mem_idx, mop);
6745 disas_set_da_iss(s, mop, issinfo);
6746 tcg_temp_free_i32(tmp);
6747
6748 op_addr_rr_post(s, a, addr, 0);
6749 return true;
6750 }
6751
6752 static bool trans_LDRD_rr(DisasContext *s, arg_ldst_rr *a)
6753 {
6754 int mem_idx = get_mem_index(s);
6755 TCGv_i32 addr, tmp;
6756
6757 if (!ENABLE_ARCH_5TE) {
6758 return false;
6759 }
6760 if (a->rt & 1) {
6761 unallocated_encoding(s);
6762 return true;
6763 }
6764 addr = op_addr_rr_pre(s, a);
6765
6766 tmp = tcg_temp_new_i32();
6767 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
6768 store_reg(s, a->rt, tmp);
6769
6770 tcg_gen_addi_i32(addr, addr, 4);
6771
6772 tmp = tcg_temp_new_i32();
6773 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
6774 store_reg(s, a->rt + 1, tmp);
6775
6776 /* LDRD w/ base writeback is undefined if the registers overlap. */
6777 op_addr_rr_post(s, a, addr, -4);
6778 return true;
6779 }
6780
6781 static bool trans_STRD_rr(DisasContext *s, arg_ldst_rr *a)
6782 {
6783 int mem_idx = get_mem_index(s);
6784 TCGv_i32 addr, tmp;
6785
6786 if (!ENABLE_ARCH_5TE) {
6787 return false;
6788 }
6789 if (a->rt & 1) {
6790 unallocated_encoding(s);
6791 return true;
6792 }
6793 addr = op_addr_rr_pre(s, a);
6794
6795 tmp = load_reg(s, a->rt);
6796 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
6797 tcg_temp_free_i32(tmp);
6798
6799 tcg_gen_addi_i32(addr, addr, 4);
6800
6801 tmp = load_reg(s, a->rt + 1);
6802 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
6803 tcg_temp_free_i32(tmp);
6804
6805 op_addr_rr_post(s, a, addr, -4);
6806 return true;
6807 }
6808
6809 /*
6810 * Load/store immediate index
6811 */
6812
6813 static TCGv_i32 op_addr_ri_pre(DisasContext *s, arg_ldst_ri *a)
6814 {
6815 int ofs = a->imm;
6816
6817 if (!a->u) {
6818 ofs = -ofs;
6819 }
6820
6821 if (s->v8m_stackcheck && a->rn == 13 && a->w) {
6822 /*
6823 * Stackcheck. Here we know 'addr' is the current SP;
6824 * U is set if we're moving SP up, else down. It is
6825 * UNKNOWN whether the limit check triggers when SP starts
6826 * below the limit and ends up above it; we chose to do so.
6827 */
6828 if (!a->u) {
6829 TCGv_i32 newsp = tcg_temp_new_i32();
6830 tcg_gen_addi_i32(newsp, cpu_R[13], ofs);
6831 gen_helper_v8m_stackcheck(cpu_env, newsp);
6832 tcg_temp_free_i32(newsp);
6833 } else {
6834 gen_helper_v8m_stackcheck(cpu_env, cpu_R[13]);
6835 }
6836 }
6837
6838 return add_reg_for_lit(s, a->rn, a->p ? ofs : 0);
6839 }
6840
6841 static void op_addr_ri_post(DisasContext *s, arg_ldst_ri *a,
6842 TCGv_i32 addr, int address_offset)
6843 {
6844 if (!a->p) {
6845 if (a->u) {
6846 address_offset += a->imm;
6847 } else {
6848 address_offset -= a->imm;
6849 }
6850 } else if (!a->w) {
6851 tcg_temp_free_i32(addr);
6852 return;
6853 }
6854 tcg_gen_addi_i32(addr, addr, address_offset);
6855 store_reg(s, a->rn, addr);
6856 }
6857
6858 static bool op_load_ri(DisasContext *s, arg_ldst_ri *a,
6859 MemOp mop, int mem_idx)
6860 {
6861 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w);
6862 TCGv_i32 addr, tmp;
6863
6864 addr = op_addr_ri_pre(s, a);
6865
6866 tmp = tcg_temp_new_i32();
6867 gen_aa32_ld_i32(s, tmp, addr, mem_idx, mop);
6868 disas_set_da_iss(s, mop, issinfo);
6869
6870 /*
6871 * Perform base writeback before the loaded value to
6872 * ensure correct behavior with overlapping index registers.
6873 */
6874 op_addr_ri_post(s, a, addr, 0);
6875 store_reg_from_load(s, a->rt, tmp);
6876 return true;
6877 }
6878
6879 static bool op_store_ri(DisasContext *s, arg_ldst_ri *a,
6880 MemOp mop, int mem_idx)
6881 {
6882 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w) | ISSIsWrite;
6883 TCGv_i32 addr, tmp;
6884
6885 /*
6886 * In Thumb encodings of stores Rn=1111 is UNDEF; for Arm it
6887 * is either UNPREDICTABLE or has defined behaviour
6888 */
6889 if (s->thumb && a->rn == 15) {
6890 return false;
6891 }
6892
6893 addr = op_addr_ri_pre(s, a);
6894
6895 tmp = load_reg(s, a->rt);
6896 gen_aa32_st_i32(s, tmp, addr, mem_idx, mop);
6897 disas_set_da_iss(s, mop, issinfo);
6898 tcg_temp_free_i32(tmp);
6899
6900 op_addr_ri_post(s, a, addr, 0);
6901 return true;
6902 }
6903
6904 static bool op_ldrd_ri(DisasContext *s, arg_ldst_ri *a, int rt2)
6905 {
6906 int mem_idx = get_mem_index(s);
6907 TCGv_i32 addr, tmp;
6908
6909 addr = op_addr_ri_pre(s, a);
6910
6911 tmp = tcg_temp_new_i32();
6912 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
6913 store_reg(s, a->rt, tmp);
6914
6915 tcg_gen_addi_i32(addr, addr, 4);
6916
6917 tmp = tcg_temp_new_i32();
6918 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
6919 store_reg(s, rt2, tmp);
6920
6921 /* LDRD w/ base writeback is undefined if the registers overlap. */
6922 op_addr_ri_post(s, a, addr, -4);
6923 return true;
6924 }
6925
6926 static bool trans_LDRD_ri_a32(DisasContext *s, arg_ldst_ri *a)
6927 {
6928 if (!ENABLE_ARCH_5TE || (a->rt & 1)) {
6929 return false;
6930 }
6931 return op_ldrd_ri(s, a, a->rt + 1);
6932 }
6933
6934 static bool trans_LDRD_ri_t32(DisasContext *s, arg_ldst_ri2 *a)
6935 {
6936 arg_ldst_ri b = {
6937 .u = a->u, .w = a->w, .p = a->p,
6938 .rn = a->rn, .rt = a->rt, .imm = a->imm
6939 };
6940 return op_ldrd_ri(s, &b, a->rt2);
6941 }
6942
6943 static bool op_strd_ri(DisasContext *s, arg_ldst_ri *a, int rt2)
6944 {
6945 int mem_idx = get_mem_index(s);
6946 TCGv_i32 addr, tmp;
6947
6948 addr = op_addr_ri_pre(s, a);
6949
6950 tmp = load_reg(s, a->rt);
6951 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
6952 tcg_temp_free_i32(tmp);
6953
6954 tcg_gen_addi_i32(addr, addr, 4);
6955
6956 tmp = load_reg(s, rt2);
6957 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
6958 tcg_temp_free_i32(tmp);
6959
6960 op_addr_ri_post(s, a, addr, -4);
6961 return true;
6962 }
6963
6964 static bool trans_STRD_ri_a32(DisasContext *s, arg_ldst_ri *a)
6965 {
6966 if (!ENABLE_ARCH_5TE || (a->rt & 1)) {
6967 return false;
6968 }
6969 return op_strd_ri(s, a, a->rt + 1);
6970 }
6971
6972 static bool trans_STRD_ri_t32(DisasContext *s, arg_ldst_ri2 *a)
6973 {
6974 arg_ldst_ri b = {
6975 .u = a->u, .w = a->w, .p = a->p,
6976 .rn = a->rn, .rt = a->rt, .imm = a->imm
6977 };
6978 return op_strd_ri(s, &b, a->rt2);
6979 }
6980
6981 #define DO_LDST(NAME, WHICH, MEMOP) \
6982 static bool trans_##NAME##_ri(DisasContext *s, arg_ldst_ri *a) \
6983 { \
6984 return op_##WHICH##_ri(s, a, MEMOP, get_mem_index(s)); \
6985 } \
6986 static bool trans_##NAME##T_ri(DisasContext *s, arg_ldst_ri *a) \
6987 { \
6988 return op_##WHICH##_ri(s, a, MEMOP, get_a32_user_mem_index(s)); \
6989 } \
6990 static bool trans_##NAME##_rr(DisasContext *s, arg_ldst_rr *a) \
6991 { \
6992 return op_##WHICH##_rr(s, a, MEMOP, get_mem_index(s)); \
6993 } \
6994 static bool trans_##NAME##T_rr(DisasContext *s, arg_ldst_rr *a) \
6995 { \
6996 return op_##WHICH##_rr(s, a, MEMOP, get_a32_user_mem_index(s)); \
6997 }
6998
6999 DO_LDST(LDR, load, MO_UL)
7000 DO_LDST(LDRB, load, MO_UB)
7001 DO_LDST(LDRH, load, MO_UW)
7002 DO_LDST(LDRSB, load, MO_SB)
7003 DO_LDST(LDRSH, load, MO_SW)
7004
7005 DO_LDST(STR, store, MO_UL)
7006 DO_LDST(STRB, store, MO_UB)
7007 DO_LDST(STRH, store, MO_UW)
7008
7009 #undef DO_LDST
7010
7011 /*
7012 * Synchronization primitives
7013 */
7014
7015 static bool op_swp(DisasContext *s, arg_SWP *a, MemOp opc)
7016 {
7017 TCGv_i32 addr, tmp;
7018 TCGv taddr;
7019
7020 opc |= s->be_data;
7021 addr = load_reg(s, a->rn);
7022 taddr = gen_aa32_addr(s, addr, opc);
7023 tcg_temp_free_i32(addr);
7024
7025 tmp = load_reg(s, a->rt2);
7026 tcg_gen_atomic_xchg_i32(tmp, taddr, tmp, get_mem_index(s), opc);
7027 tcg_temp_free(taddr);
7028
7029 store_reg(s, a->rt, tmp);
7030 return true;
7031 }
7032
7033 static bool trans_SWP(DisasContext *s, arg_SWP *a)
7034 {
7035 return op_swp(s, a, MO_UL | MO_ALIGN);
7036 }
7037
7038 static bool trans_SWPB(DisasContext *s, arg_SWP *a)
7039 {
7040 return op_swp(s, a, MO_UB);
7041 }
7042
7043 /*
7044 * Load/Store Exclusive and Load-Acquire/Store-Release
7045 */
7046
7047 static bool op_strex(DisasContext *s, arg_STREX *a, MemOp mop, bool rel)
7048 {
7049 TCGv_i32 addr;
7050 /* Some cases stopped being UNPREDICTABLE in v8A (but not v8M) */
7051 bool v8a = ENABLE_ARCH_8 && !arm_dc_feature(s, ARM_FEATURE_M);
7052
7053 /* We UNDEF for these UNPREDICTABLE cases. */
7054 if (a->rd == 15 || a->rn == 15 || a->rt == 15
7055 || a->rd == a->rn || a->rd == a->rt
7056 || (!v8a && s->thumb && (a->rd == 13 || a->rt == 13))
7057 || (mop == MO_64
7058 && (a->rt2 == 15
7059 || a->rd == a->rt2
7060 || (!v8a && s->thumb && a->rt2 == 13)))) {
7061 unallocated_encoding(s);
7062 return true;
7063 }
7064
7065 if (rel) {
7066 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
7067 }
7068
7069 addr = tcg_temp_local_new_i32();
7070 load_reg_var(s, addr, a->rn);
7071 tcg_gen_addi_i32(addr, addr, a->imm);
7072
7073 gen_store_exclusive(s, a->rd, a->rt, a->rt2, addr, mop);
7074 tcg_temp_free_i32(addr);
7075 return true;
7076 }
7077
7078 static bool trans_STREX(DisasContext *s, arg_STREX *a)
7079 {
7080 if (!ENABLE_ARCH_6) {
7081 return false;
7082 }
7083 return op_strex(s, a, MO_32, false);
7084 }
7085
7086 static bool trans_STREXD_a32(DisasContext *s, arg_STREX *a)
7087 {
7088 if (!ENABLE_ARCH_6K) {
7089 return false;
7090 }
7091 /* We UNDEF for these UNPREDICTABLE cases. */
7092 if (a->rt & 1) {
7093 unallocated_encoding(s);
7094 return true;
7095 }
7096 a->rt2 = a->rt + 1;
7097 return op_strex(s, a, MO_64, false);
7098 }
7099
7100 static bool trans_STREXD_t32(DisasContext *s, arg_STREX *a)
7101 {
7102 return op_strex(s, a, MO_64, false);
7103 }
7104
7105 static bool trans_STREXB(DisasContext *s, arg_STREX *a)
7106 {
7107 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
7108 return false;
7109 }
7110 return op_strex(s, a, MO_8, false);
7111 }
7112
7113 static bool trans_STREXH(DisasContext *s, arg_STREX *a)
7114 {
7115 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
7116 return false;
7117 }
7118 return op_strex(s, a, MO_16, false);
7119 }
7120
7121 static bool trans_STLEX(DisasContext *s, arg_STREX *a)
7122 {
7123 if (!ENABLE_ARCH_8) {
7124 return false;
7125 }
7126 return op_strex(s, a, MO_32, true);
7127 }
7128
7129 static bool trans_STLEXD_a32(DisasContext *s, arg_STREX *a)
7130 {
7131 if (!ENABLE_ARCH_8) {
7132 return false;
7133 }
7134 /* We UNDEF for these UNPREDICTABLE cases. */
7135 if (a->rt & 1) {
7136 unallocated_encoding(s);
7137 return true;
7138 }
7139 a->rt2 = a->rt + 1;
7140 return op_strex(s, a, MO_64, true);
7141 }
7142
7143 static bool trans_STLEXD_t32(DisasContext *s, arg_STREX *a)
7144 {
7145 if (!ENABLE_ARCH_8) {
7146 return false;
7147 }
7148 return op_strex(s, a, MO_64, true);
7149 }
7150
7151 static bool trans_STLEXB(DisasContext *s, arg_STREX *a)
7152 {
7153 if (!ENABLE_ARCH_8) {
7154 return false;
7155 }
7156 return op_strex(s, a, MO_8, true);
7157 }
7158
7159 static bool trans_STLEXH(DisasContext *s, arg_STREX *a)
7160 {
7161 if (!ENABLE_ARCH_8) {
7162 return false;
7163 }
7164 return op_strex(s, a, MO_16, true);
7165 }
7166
7167 static bool op_stl(DisasContext *s, arg_STL *a, MemOp mop)
7168 {
7169 TCGv_i32 addr, tmp;
7170
7171 if (!ENABLE_ARCH_8) {
7172 return false;
7173 }
7174 /* We UNDEF for these UNPREDICTABLE cases. */
7175 if (a->rn == 15 || a->rt == 15) {
7176 unallocated_encoding(s);
7177 return true;
7178 }
7179
7180 addr = load_reg(s, a->rn);
7181 tmp = load_reg(s, a->rt);
7182 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
7183 gen_aa32_st_i32(s, tmp, addr, get_mem_index(s), mop | MO_ALIGN);
7184 disas_set_da_iss(s, mop, a->rt | ISSIsAcqRel | ISSIsWrite);
7185
7186 tcg_temp_free_i32(tmp);
7187 tcg_temp_free_i32(addr);
7188 return true;
7189 }
7190
7191 static bool trans_STL(DisasContext *s, arg_STL *a)
7192 {
7193 return op_stl(s, a, MO_UL);
7194 }
7195
7196 static bool trans_STLB(DisasContext *s, arg_STL *a)
7197 {
7198 return op_stl(s, a, MO_UB);
7199 }
7200
7201 static bool trans_STLH(DisasContext *s, arg_STL *a)
7202 {
7203 return op_stl(s, a, MO_UW);
7204 }
7205
7206 static bool op_ldrex(DisasContext *s, arg_LDREX *a, MemOp mop, bool acq)
7207 {
7208 TCGv_i32 addr;
7209 /* Some cases stopped being UNPREDICTABLE in v8A (but not v8M) */
7210 bool v8a = ENABLE_ARCH_8 && !arm_dc_feature(s, ARM_FEATURE_M);
7211
7212 /* We UNDEF for these UNPREDICTABLE cases. */
7213 if (a->rn == 15 || a->rt == 15
7214 || (!v8a && s->thumb && a->rt == 13)
7215 || (mop == MO_64
7216 && (a->rt2 == 15 || a->rt == a->rt2
7217 || (!v8a && s->thumb && a->rt2 == 13)))) {
7218 unallocated_encoding(s);
7219 return true;
7220 }
7221
7222 addr = tcg_temp_local_new_i32();
7223 load_reg_var(s, addr, a->rn);
7224 tcg_gen_addi_i32(addr, addr, a->imm);
7225
7226 gen_load_exclusive(s, a->rt, a->rt2, addr, mop);
7227 tcg_temp_free_i32(addr);
7228
7229 if (acq) {
7230 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_LDAQ);
7231 }
7232 return true;
7233 }
7234
7235 static bool trans_LDREX(DisasContext *s, arg_LDREX *a)
7236 {
7237 if (!ENABLE_ARCH_6) {
7238 return false;
7239 }
7240 return op_ldrex(s, a, MO_32, false);
7241 }
7242
7243 static bool trans_LDREXD_a32(DisasContext *s, arg_LDREX *a)
7244 {
7245 if (!ENABLE_ARCH_6K) {
7246 return false;
7247 }
7248 /* We UNDEF for these UNPREDICTABLE cases. */
7249 if (a->rt & 1) {
7250 unallocated_encoding(s);
7251 return true;
7252 }
7253 a->rt2 = a->rt + 1;
7254 return op_ldrex(s, a, MO_64, false);
7255 }
7256
7257 static bool trans_LDREXD_t32(DisasContext *s, arg_LDREX *a)
7258 {
7259 return op_ldrex(s, a, MO_64, false);
7260 }
7261
7262 static bool trans_LDREXB(DisasContext *s, arg_LDREX *a)
7263 {
7264 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
7265 return false;
7266 }
7267 return op_ldrex(s, a, MO_8, false);
7268 }
7269
7270 static bool trans_LDREXH(DisasContext *s, arg_LDREX *a)
7271 {
7272 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
7273 return false;
7274 }
7275 return op_ldrex(s, a, MO_16, false);
7276 }
7277
7278 static bool trans_LDAEX(DisasContext *s, arg_LDREX *a)
7279 {
7280 if (!ENABLE_ARCH_8) {
7281 return false;
7282 }
7283 return op_ldrex(s, a, MO_32, true);
7284 }
7285
7286 static bool trans_LDAEXD_a32(DisasContext *s, arg_LDREX *a)
7287 {
7288 if (!ENABLE_ARCH_8) {
7289 return false;
7290 }
7291 /* We UNDEF for these UNPREDICTABLE cases. */
7292 if (a->rt & 1) {
7293 unallocated_encoding(s);
7294 return true;
7295 }
7296 a->rt2 = a->rt + 1;
7297 return op_ldrex(s, a, MO_64, true);
7298 }
7299
7300 static bool trans_LDAEXD_t32(DisasContext *s, arg_LDREX *a)
7301 {
7302 if (!ENABLE_ARCH_8) {
7303 return false;
7304 }
7305 return op_ldrex(s, a, MO_64, true);
7306 }
7307
7308 static bool trans_LDAEXB(DisasContext *s, arg_LDREX *a)
7309 {
7310 if (!ENABLE_ARCH_8) {
7311 return false;
7312 }
7313 return op_ldrex(s, a, MO_8, true);
7314 }
7315
7316 static bool trans_LDAEXH(DisasContext *s, arg_LDREX *a)
7317 {
7318 if (!ENABLE_ARCH_8) {
7319 return false;
7320 }
7321 return op_ldrex(s, a, MO_16, true);
7322 }
7323
7324 static bool op_lda(DisasContext *s, arg_LDA *a, MemOp mop)
7325 {
7326 TCGv_i32 addr, tmp;
7327
7328 if (!ENABLE_ARCH_8) {
7329 return false;
7330 }
7331 /* We UNDEF for these UNPREDICTABLE cases. */
7332 if (a->rn == 15 || a->rt == 15) {
7333 unallocated_encoding(s);
7334 return true;
7335 }
7336
7337 addr = load_reg(s, a->rn);
7338 tmp = tcg_temp_new_i32();
7339 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), mop | MO_ALIGN);
7340 disas_set_da_iss(s, mop, a->rt | ISSIsAcqRel);
7341 tcg_temp_free_i32(addr);
7342
7343 store_reg(s, a->rt, tmp);
7344 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
7345 return true;
7346 }
7347
7348 static bool trans_LDA(DisasContext *s, arg_LDA *a)
7349 {
7350 return op_lda(s, a, MO_UL);
7351 }
7352
7353 static bool trans_LDAB(DisasContext *s, arg_LDA *a)
7354 {
7355 return op_lda(s, a, MO_UB);
7356 }
7357
7358 static bool trans_LDAH(DisasContext *s, arg_LDA *a)
7359 {
7360 return op_lda(s, a, MO_UW);
7361 }
7362
7363 /*
7364 * Media instructions
7365 */
7366
7367 static bool trans_USADA8(DisasContext *s, arg_USADA8 *a)
7368 {
7369 TCGv_i32 t1, t2;
7370
7371 if (!ENABLE_ARCH_6) {
7372 return false;
7373 }
7374
7375 t1 = load_reg(s, a->rn);
7376 t2 = load_reg(s, a->rm);
7377 gen_helper_usad8(t1, t1, t2);
7378 tcg_temp_free_i32(t2);
7379 if (a->ra != 15) {
7380 t2 = load_reg(s, a->ra);
7381 tcg_gen_add_i32(t1, t1, t2);
7382 tcg_temp_free_i32(t2);
7383 }
7384 store_reg(s, a->rd, t1);
7385 return true;
7386 }
7387
7388 static bool op_bfx(DisasContext *s, arg_UBFX *a, bool u)
7389 {
7390 TCGv_i32 tmp;
7391 int width = a->widthm1 + 1;
7392 int shift = a->lsb;
7393
7394 if (!ENABLE_ARCH_6T2) {
7395 return false;
7396 }
7397 if (shift + width > 32) {
7398 /* UNPREDICTABLE; we choose to UNDEF */
7399 unallocated_encoding(s);
7400 return true;
7401 }
7402
7403 tmp = load_reg(s, a->rn);
7404 if (u) {
7405 tcg_gen_extract_i32(tmp, tmp, shift, width);
7406 } else {
7407 tcg_gen_sextract_i32(tmp, tmp, shift, width);
7408 }
7409 store_reg(s, a->rd, tmp);
7410 return true;
7411 }
7412
7413 static bool trans_SBFX(DisasContext *s, arg_SBFX *a)
7414 {
7415 return op_bfx(s, a, false);
7416 }
7417
7418 static bool trans_UBFX(DisasContext *s, arg_UBFX *a)
7419 {
7420 return op_bfx(s, a, true);
7421 }
7422
7423 static bool trans_BFCI(DisasContext *s, arg_BFCI *a)
7424 {
7425 TCGv_i32 tmp;
7426 int msb = a->msb, lsb = a->lsb;
7427 int width;
7428
7429 if (!ENABLE_ARCH_6T2) {
7430 return false;
7431 }
7432 if (msb < lsb) {
7433 /* UNPREDICTABLE; we choose to UNDEF */
7434 unallocated_encoding(s);
7435 return true;
7436 }
7437
7438 width = msb + 1 - lsb;
7439 if (a->rn == 15) {
7440 /* BFC */
7441 tmp = tcg_const_i32(0);
7442 } else {
7443 /* BFI */
7444 tmp = load_reg(s, a->rn);
7445 }
7446 if (width != 32) {
7447 TCGv_i32 tmp2 = load_reg(s, a->rd);
7448 tcg_gen_deposit_i32(tmp, tmp2, tmp, lsb, width);
7449 tcg_temp_free_i32(tmp2);
7450 }
7451 store_reg(s, a->rd, tmp);
7452 return true;
7453 }
7454
7455 static bool trans_UDF(DisasContext *s, arg_UDF *a)
7456 {
7457 unallocated_encoding(s);
7458 return true;
7459 }
7460
7461 /*
7462 * Parallel addition and subtraction
7463 */
7464
7465 static bool op_par_addsub(DisasContext *s, arg_rrr *a,
7466 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32))
7467 {
7468 TCGv_i32 t0, t1;
7469
7470 if (s->thumb
7471 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
7472 : !ENABLE_ARCH_6) {
7473 return false;
7474 }
7475
7476 t0 = load_reg(s, a->rn);
7477 t1 = load_reg(s, a->rm);
7478
7479 gen(t0, t0, t1);
7480
7481 tcg_temp_free_i32(t1);
7482 store_reg(s, a->rd, t0);
7483 return true;
7484 }
7485
7486 static bool op_par_addsub_ge(DisasContext *s, arg_rrr *a,
7487 void (*gen)(TCGv_i32, TCGv_i32,
7488 TCGv_i32, TCGv_ptr))
7489 {
7490 TCGv_i32 t0, t1;
7491 TCGv_ptr ge;
7492
7493 if (s->thumb
7494 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
7495 : !ENABLE_ARCH_6) {
7496 return false;
7497 }
7498
7499 t0 = load_reg(s, a->rn);
7500 t1 = load_reg(s, a->rm);
7501
7502 ge = tcg_temp_new_ptr();
7503 tcg_gen_addi_ptr(ge, cpu_env, offsetof(CPUARMState, GE));
7504 gen(t0, t0, t1, ge);
7505
7506 tcg_temp_free_ptr(ge);
7507 tcg_temp_free_i32(t1);
7508 store_reg(s, a->rd, t0);
7509 return true;
7510 }
7511
7512 #define DO_PAR_ADDSUB(NAME, helper) \
7513 static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
7514 { \
7515 return op_par_addsub(s, a, helper); \
7516 }
7517
7518 #define DO_PAR_ADDSUB_GE(NAME, helper) \
7519 static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
7520 { \
7521 return op_par_addsub_ge(s, a, helper); \
7522 }
7523
7524 DO_PAR_ADDSUB_GE(SADD16, gen_helper_sadd16)
7525 DO_PAR_ADDSUB_GE(SASX, gen_helper_saddsubx)
7526 DO_PAR_ADDSUB_GE(SSAX, gen_helper_ssubaddx)
7527 DO_PAR_ADDSUB_GE(SSUB16, gen_helper_ssub16)
7528 DO_PAR_ADDSUB_GE(SADD8, gen_helper_sadd8)
7529 DO_PAR_ADDSUB_GE(SSUB8, gen_helper_ssub8)
7530
7531 DO_PAR_ADDSUB_GE(UADD16, gen_helper_uadd16)
7532 DO_PAR_ADDSUB_GE(UASX, gen_helper_uaddsubx)
7533 DO_PAR_ADDSUB_GE(USAX, gen_helper_usubaddx)
7534 DO_PAR_ADDSUB_GE(USUB16, gen_helper_usub16)
7535 DO_PAR_ADDSUB_GE(UADD8, gen_helper_uadd8)
7536 DO_PAR_ADDSUB_GE(USUB8, gen_helper_usub8)
7537
7538 DO_PAR_ADDSUB(QADD16, gen_helper_qadd16)
7539 DO_PAR_ADDSUB(QASX, gen_helper_qaddsubx)
7540 DO_PAR_ADDSUB(QSAX, gen_helper_qsubaddx)
7541 DO_PAR_ADDSUB(QSUB16, gen_helper_qsub16)
7542 DO_PAR_ADDSUB(QADD8, gen_helper_qadd8)
7543 DO_PAR_ADDSUB(QSUB8, gen_helper_qsub8)
7544
7545 DO_PAR_ADDSUB(UQADD16, gen_helper_uqadd16)
7546 DO_PAR_ADDSUB(UQASX, gen_helper_uqaddsubx)
7547 DO_PAR_ADDSUB(UQSAX, gen_helper_uqsubaddx)
7548 DO_PAR_ADDSUB(UQSUB16, gen_helper_uqsub16)
7549 DO_PAR_ADDSUB(UQADD8, gen_helper_uqadd8)
7550 DO_PAR_ADDSUB(UQSUB8, gen_helper_uqsub8)
7551
7552 DO_PAR_ADDSUB(SHADD16, gen_helper_shadd16)
7553 DO_PAR_ADDSUB(SHASX, gen_helper_shaddsubx)
7554 DO_PAR_ADDSUB(SHSAX, gen_helper_shsubaddx)
7555 DO_PAR_ADDSUB(SHSUB16, gen_helper_shsub16)
7556 DO_PAR_ADDSUB(SHADD8, gen_helper_shadd8)
7557 DO_PAR_ADDSUB(SHSUB8, gen_helper_shsub8)
7558
7559 DO_PAR_ADDSUB(UHADD16, gen_helper_uhadd16)
7560 DO_PAR_ADDSUB(UHASX, gen_helper_uhaddsubx)
7561 DO_PAR_ADDSUB(UHSAX, gen_helper_uhsubaddx)
7562 DO_PAR_ADDSUB(UHSUB16, gen_helper_uhsub16)
7563 DO_PAR_ADDSUB(UHADD8, gen_helper_uhadd8)
7564 DO_PAR_ADDSUB(UHSUB8, gen_helper_uhsub8)
7565
7566 #undef DO_PAR_ADDSUB
7567 #undef DO_PAR_ADDSUB_GE
7568
7569 /*
7570 * Packing, unpacking, saturation, and reversal
7571 */
7572
7573 static bool trans_PKH(DisasContext *s, arg_PKH *a)
7574 {
7575 TCGv_i32 tn, tm;
7576 int shift = a->imm;
7577
7578 if (s->thumb
7579 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
7580 : !ENABLE_ARCH_6) {
7581 return false;
7582 }
7583
7584 tn = load_reg(s, a->rn);
7585 tm = load_reg(s, a->rm);
7586 if (a->tb) {
7587 /* PKHTB */
7588 if (shift == 0) {
7589 shift = 31;
7590 }
7591 tcg_gen_sari_i32(tm, tm, shift);
7592 tcg_gen_deposit_i32(tn, tn, tm, 0, 16);
7593 } else {
7594 /* PKHBT */
7595 tcg_gen_shli_i32(tm, tm, shift);
7596 tcg_gen_deposit_i32(tn, tm, tn, 0, 16);
7597 }
7598 tcg_temp_free_i32(tm);
7599 store_reg(s, a->rd, tn);
7600 return true;
7601 }
7602
7603 static bool op_sat(DisasContext *s, arg_sat *a,
7604 void (*gen)(TCGv_i32, TCGv_env, TCGv_i32, TCGv_i32))
7605 {
7606 TCGv_i32 tmp, satimm;
7607 int shift = a->imm;
7608
7609 if (!ENABLE_ARCH_6) {
7610 return false;
7611 }
7612
7613 tmp = load_reg(s, a->rn);
7614 if (a->sh) {
7615 tcg_gen_sari_i32(tmp, tmp, shift ? shift : 31);
7616 } else {
7617 tcg_gen_shli_i32(tmp, tmp, shift);
7618 }
7619
7620 satimm = tcg_const_i32(a->satimm);
7621 gen(tmp, cpu_env, tmp, satimm);
7622 tcg_temp_free_i32(satimm);
7623
7624 store_reg(s, a->rd, tmp);
7625 return true;
7626 }
7627
7628 static bool trans_SSAT(DisasContext *s, arg_sat *a)
7629 {
7630 return op_sat(s, a, gen_helper_ssat);
7631 }
7632
7633 static bool trans_USAT(DisasContext *s, arg_sat *a)
7634 {
7635 return op_sat(s, a, gen_helper_usat);
7636 }
7637
7638 static bool trans_SSAT16(DisasContext *s, arg_sat *a)
7639 {
7640 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
7641 return false;
7642 }
7643 return op_sat(s, a, gen_helper_ssat16);
7644 }
7645
7646 static bool trans_USAT16(DisasContext *s, arg_sat *a)
7647 {
7648 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
7649 return false;
7650 }
7651 return op_sat(s, a, gen_helper_usat16);
7652 }
7653
7654 static bool op_xta(DisasContext *s, arg_rrr_rot *a,
7655 void (*gen_extract)(TCGv_i32, TCGv_i32),
7656 void (*gen_add)(TCGv_i32, TCGv_i32, TCGv_i32))
7657 {
7658 TCGv_i32 tmp;
7659
7660 if (!ENABLE_ARCH_6) {
7661 return false;
7662 }
7663
7664 tmp = load_reg(s, a->rm);
7665 /*
7666 * TODO: In many cases we could do a shift instead of a rotate.
7667 * Combined with a simple extend, that becomes an extract.
7668 */
7669 tcg_gen_rotri_i32(tmp, tmp, a->rot * 8);
7670 gen_extract(tmp, tmp);
7671
7672 if (a->rn != 15) {
7673 TCGv_i32 tmp2 = load_reg(s, a->rn);
7674 gen_add(tmp, tmp, tmp2);
7675 tcg_temp_free_i32(tmp2);
7676 }
7677 store_reg(s, a->rd, tmp);
7678 return true;
7679 }
7680
7681 static bool trans_SXTAB(DisasContext *s, arg_rrr_rot *a)
7682 {
7683 return op_xta(s, a, tcg_gen_ext8s_i32, tcg_gen_add_i32);
7684 }
7685
7686 static bool trans_SXTAH(DisasContext *s, arg_rrr_rot *a)
7687 {
7688 return op_xta(s, a, tcg_gen_ext16s_i32, tcg_gen_add_i32);
7689 }
7690
7691 static bool trans_SXTAB16(DisasContext *s, arg_rrr_rot *a)
7692 {
7693 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
7694 return false;
7695 }
7696 return op_xta(s, a, gen_helper_sxtb16, gen_add16);
7697 }
7698
7699 static bool trans_UXTAB(DisasContext *s, arg_rrr_rot *a)
7700 {
7701 return op_xta(s, a, tcg_gen_ext8u_i32, tcg_gen_add_i32);
7702 }
7703
7704 static bool trans_UXTAH(DisasContext *s, arg_rrr_rot *a)
7705 {
7706 return op_xta(s, a, tcg_gen_ext16u_i32, tcg_gen_add_i32);
7707 }
7708
7709 static bool trans_UXTAB16(DisasContext *s, arg_rrr_rot *a)
7710 {
7711 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
7712 return false;
7713 }
7714 return op_xta(s, a, gen_helper_uxtb16, gen_add16);
7715 }
7716
7717 static bool trans_SEL(DisasContext *s, arg_rrr *a)
7718 {
7719 TCGv_i32 t1, t2, t3;
7720
7721 if (s->thumb
7722 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
7723 : !ENABLE_ARCH_6) {
7724 return false;
7725 }
7726
7727 t1 = load_reg(s, a->rn);
7728 t2 = load_reg(s, a->rm);
7729 t3 = tcg_temp_new_i32();
7730 tcg_gen_ld_i32(t3, cpu_env, offsetof(CPUARMState, GE));
7731 gen_helper_sel_flags(t1, t3, t1, t2);
7732 tcg_temp_free_i32(t3);
7733 tcg_temp_free_i32(t2);
7734 store_reg(s, a->rd, t1);
7735 return true;
7736 }
7737
7738 static bool op_rr(DisasContext *s, arg_rr *a,
7739 void (*gen)(TCGv_i32, TCGv_i32))
7740 {
7741 TCGv_i32 tmp;
7742
7743 tmp = load_reg(s, a->rm);
7744 gen(tmp, tmp);
7745 store_reg(s, a->rd, tmp);
7746 return true;
7747 }
7748
7749 static bool trans_REV(DisasContext *s, arg_rr *a)
7750 {
7751 if (!ENABLE_ARCH_6) {
7752 return false;
7753 }
7754 return op_rr(s, a, tcg_gen_bswap32_i32);
7755 }
7756
7757 static bool trans_REV16(DisasContext *s, arg_rr *a)
7758 {
7759 if (!ENABLE_ARCH_6) {
7760 return false;
7761 }
7762 return op_rr(s, a, gen_rev16);
7763 }
7764
7765 static bool trans_REVSH(DisasContext *s, arg_rr *a)
7766 {
7767 if (!ENABLE_ARCH_6) {
7768 return false;
7769 }
7770 return op_rr(s, a, gen_revsh);
7771 }
7772
7773 static bool trans_RBIT(DisasContext *s, arg_rr *a)
7774 {
7775 if (!ENABLE_ARCH_6T2) {
7776 return false;
7777 }
7778 return op_rr(s, a, gen_helper_rbit);
7779 }
7780
7781 /*
7782 * Signed multiply, signed and unsigned divide
7783 */
7784
7785 static bool op_smlad(DisasContext *s, arg_rrrr *a, bool m_swap, bool sub)
7786 {
7787 TCGv_i32 t1, t2;
7788
7789 if (!ENABLE_ARCH_6) {
7790 return false;
7791 }
7792
7793 t1 = load_reg(s, a->rn);
7794 t2 = load_reg(s, a->rm);
7795 if (m_swap) {
7796 gen_swap_half(t2, t2);
7797 }
7798 gen_smul_dual(t1, t2);
7799
7800 if (sub) {
7801 /*
7802 * This subtraction cannot overflow, so we can do a simple
7803 * 32-bit subtraction and then a possible 32-bit saturating
7804 * addition of Ra.
7805 */
7806 tcg_gen_sub_i32(t1, t1, t2);
7807 tcg_temp_free_i32(t2);
7808
7809 if (a->ra != 15) {
7810 t2 = load_reg(s, a->ra);
7811 gen_helper_add_setq(t1, cpu_env, t1, t2);
7812 tcg_temp_free_i32(t2);
7813 }
7814 } else if (a->ra == 15) {
7815 /* Single saturation-checking addition */
7816 gen_helper_add_setq(t1, cpu_env, t1, t2);
7817 tcg_temp_free_i32(t2);
7818 } else {
7819 /*
7820 * We need to add the products and Ra together and then
7821 * determine whether the final result overflowed. Doing
7822 * this as two separate add-and-check-overflow steps incorrectly
7823 * sets Q for cases like (-32768 * -32768) + (-32768 * -32768) + -1.
7824 * Do all the arithmetic at 64-bits and then check for overflow.
7825 */
7826 TCGv_i64 p64, q64;
7827 TCGv_i32 t3, qf, one;
7828
7829 p64 = tcg_temp_new_i64();
7830 q64 = tcg_temp_new_i64();
7831 tcg_gen_ext_i32_i64(p64, t1);
7832 tcg_gen_ext_i32_i64(q64, t2);
7833 tcg_gen_add_i64(p64, p64, q64);
7834 load_reg_var(s, t2, a->ra);
7835 tcg_gen_ext_i32_i64(q64, t2);
7836 tcg_gen_add_i64(p64, p64, q64);
7837 tcg_temp_free_i64(q64);
7838
7839 tcg_gen_extr_i64_i32(t1, t2, p64);
7840 tcg_temp_free_i64(p64);
7841 /*
7842 * t1 is the low half of the result which goes into Rd.
7843 * We have overflow and must set Q if the high half (t2)
7844 * is different from the sign-extension of t1.
7845 */
7846 t3 = tcg_temp_new_i32();
7847 tcg_gen_sari_i32(t3, t1, 31);
7848 qf = load_cpu_field(QF);
7849 one = tcg_constant_i32(1);
7850 tcg_gen_movcond_i32(TCG_COND_NE, qf, t2, t3, one, qf);
7851 store_cpu_field(qf, QF);
7852 tcg_temp_free_i32(t3);
7853 tcg_temp_free_i32(t2);
7854 }
7855 store_reg(s, a->rd, t1);
7856 return true;
7857 }
7858
7859 static bool trans_SMLAD(DisasContext *s, arg_rrrr *a)
7860 {
7861 return op_smlad(s, a, false, false);
7862 }
7863
7864 static bool trans_SMLADX(DisasContext *s, arg_rrrr *a)
7865 {
7866 return op_smlad(s, a, true, false);
7867 }
7868
7869 static bool trans_SMLSD(DisasContext *s, arg_rrrr *a)
7870 {
7871 return op_smlad(s, a, false, true);
7872 }
7873
7874 static bool trans_SMLSDX(DisasContext *s, arg_rrrr *a)
7875 {
7876 return op_smlad(s, a, true, true);
7877 }
7878
7879 static bool op_smlald(DisasContext *s, arg_rrrr *a, bool m_swap, bool sub)
7880 {
7881 TCGv_i32 t1, t2;
7882 TCGv_i64 l1, l2;
7883
7884 if (!ENABLE_ARCH_6) {
7885 return false;
7886 }
7887
7888 t1 = load_reg(s, a->rn);
7889 t2 = load_reg(s, a->rm);
7890 if (m_swap) {
7891 gen_swap_half(t2, t2);
7892 }
7893 gen_smul_dual(t1, t2);
7894
7895 l1 = tcg_temp_new_i64();
7896 l2 = tcg_temp_new_i64();
7897 tcg_gen_ext_i32_i64(l1, t1);
7898 tcg_gen_ext_i32_i64(l2, t2);
7899 tcg_temp_free_i32(t1);
7900 tcg_temp_free_i32(t2);
7901
7902 if (sub) {
7903 tcg_gen_sub_i64(l1, l1, l2);
7904 } else {
7905 tcg_gen_add_i64(l1, l1, l2);
7906 }
7907 tcg_temp_free_i64(l2);
7908
7909 gen_addq(s, l1, a->ra, a->rd);
7910 gen_storeq_reg(s, a->ra, a->rd, l1);
7911 tcg_temp_free_i64(l1);
7912 return true;
7913 }
7914
7915 static bool trans_SMLALD(DisasContext *s, arg_rrrr *a)
7916 {
7917 return op_smlald(s, a, false, false);
7918 }
7919
7920 static bool trans_SMLALDX(DisasContext *s, arg_rrrr *a)
7921 {
7922 return op_smlald(s, a, true, false);
7923 }
7924
7925 static bool trans_SMLSLD(DisasContext *s, arg_rrrr *a)
7926 {
7927 return op_smlald(s, a, false, true);
7928 }
7929
7930 static bool trans_SMLSLDX(DisasContext *s, arg_rrrr *a)
7931 {
7932 return op_smlald(s, a, true, true);
7933 }
7934
7935 static bool op_smmla(DisasContext *s, arg_rrrr *a, bool round, bool sub)
7936 {
7937 TCGv_i32 t1, t2;
7938
7939 if (s->thumb
7940 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
7941 : !ENABLE_ARCH_6) {
7942 return false;
7943 }
7944
7945 t1 = load_reg(s, a->rn);
7946 t2 = load_reg(s, a->rm);
7947 tcg_gen_muls2_i32(t2, t1, t1, t2);
7948
7949 if (a->ra != 15) {
7950 TCGv_i32 t3 = load_reg(s, a->ra);
7951 if (sub) {
7952 /*
7953 * For SMMLS, we need a 64-bit subtract. Borrow caused by
7954 * a non-zero multiplicand lowpart, and the correct result
7955 * lowpart for rounding.
7956 */
7957 TCGv_i32 zero = tcg_const_i32(0);
7958 tcg_gen_sub2_i32(t2, t1, zero, t3, t2, t1);
7959 tcg_temp_free_i32(zero);
7960 } else {
7961 tcg_gen_add_i32(t1, t1, t3);
7962 }
7963 tcg_temp_free_i32(t3);
7964 }
7965 if (round) {
7966 /*
7967 * Adding 0x80000000 to the 64-bit quantity means that we have
7968 * carry in to the high word when the low word has the msb set.
7969 */
7970 tcg_gen_shri_i32(t2, t2, 31);
7971 tcg_gen_add_i32(t1, t1, t2);
7972 }
7973 tcg_temp_free_i32(t2);
7974 store_reg(s, a->rd, t1);
7975 return true;
7976 }
7977
7978 static bool trans_SMMLA(DisasContext *s, arg_rrrr *a)
7979 {
7980 return op_smmla(s, a, false, false);
7981 }
7982
7983 static bool trans_SMMLAR(DisasContext *s, arg_rrrr *a)
7984 {
7985 return op_smmla(s, a, true, false);
7986 }
7987
7988 static bool trans_SMMLS(DisasContext *s, arg_rrrr *a)
7989 {
7990 return op_smmla(s, a, false, true);
7991 }
7992
7993 static bool trans_SMMLSR(DisasContext *s, arg_rrrr *a)
7994 {
7995 return op_smmla(s, a, true, true);
7996 }
7997
7998 static bool op_div(DisasContext *s, arg_rrr *a, bool u)
7999 {
8000 TCGv_i32 t1, t2;
8001
8002 if (s->thumb
8003 ? !dc_isar_feature(aa32_thumb_div, s)
8004 : !dc_isar_feature(aa32_arm_div, s)) {
8005 return false;
8006 }
8007
8008 t1 = load_reg(s, a->rn);
8009 t2 = load_reg(s, a->rm);
8010 if (u) {
8011 gen_helper_udiv(t1, cpu_env, t1, t2);
8012 } else {
8013 gen_helper_sdiv(t1, cpu_env, t1, t2);
8014 }
8015 tcg_temp_free_i32(t2);
8016 store_reg(s, a->rd, t1);
8017 return true;
8018 }
8019
8020 static bool trans_SDIV(DisasContext *s, arg_rrr *a)
8021 {
8022 return op_div(s, a, false);
8023 }
8024
8025 static bool trans_UDIV(DisasContext *s, arg_rrr *a)
8026 {
8027 return op_div(s, a, true);
8028 }
8029
8030 /*
8031 * Block data transfer
8032 */
8033
8034 static TCGv_i32 op_addr_block_pre(DisasContext *s, arg_ldst_block *a, int n)
8035 {
8036 TCGv_i32 addr = load_reg(s, a->rn);
8037
8038 if (a->b) {
8039 if (a->i) {
8040 /* pre increment */
8041 tcg_gen_addi_i32(addr, addr, 4);
8042 } else {
8043 /* pre decrement */
8044 tcg_gen_addi_i32(addr, addr, -(n * 4));
8045 }
8046 } else if (!a->i && n != 1) {
8047 /* post decrement */
8048 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
8049 }
8050
8051 if (s->v8m_stackcheck && a->rn == 13 && a->w) {
8052 /*
8053 * If the writeback is incrementing SP rather than
8054 * decrementing it, and the initial SP is below the
8055 * stack limit but the final written-back SP would
8056 * be above, then then we must not perform any memory
8057 * accesses, but it is IMPDEF whether we generate
8058 * an exception. We choose to do so in this case.
8059 * At this point 'addr' is the lowest address, so
8060 * either the original SP (if incrementing) or our
8061 * final SP (if decrementing), so that's what we check.
8062 */
8063 gen_helper_v8m_stackcheck(cpu_env, addr);
8064 }
8065
8066 return addr;
8067 }
8068
8069 static void op_addr_block_post(DisasContext *s, arg_ldst_block *a,
8070 TCGv_i32 addr, int n)
8071 {
8072 if (a->w) {
8073 /* write back */
8074 if (!a->b) {
8075 if (a->i) {
8076 /* post increment */
8077 tcg_gen_addi_i32(addr, addr, 4);
8078 } else {
8079 /* post decrement */
8080 tcg_gen_addi_i32(addr, addr, -(n * 4));
8081 }
8082 } else if (!a->i && n != 1) {
8083 /* pre decrement */
8084 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
8085 }
8086 store_reg(s, a->rn, addr);
8087 } else {
8088 tcg_temp_free_i32(addr);
8089 }
8090 }
8091
8092 static bool op_stm(DisasContext *s, arg_ldst_block *a, int min_n)
8093 {
8094 int i, j, n, list, mem_idx;
8095 bool user = a->u;
8096 TCGv_i32 addr, tmp, tmp2;
8097
8098 if (user) {
8099 /* STM (user) */
8100 if (IS_USER(s)) {
8101 /* Only usable in supervisor mode. */
8102 unallocated_encoding(s);
8103 return true;
8104 }
8105 }
8106
8107 list = a->list;
8108 n = ctpop16(list);
8109 if (n < min_n || a->rn == 15) {
8110 unallocated_encoding(s);
8111 return true;
8112 }
8113
8114 s->eci_handled = true;
8115
8116 addr = op_addr_block_pre(s, a, n);
8117 mem_idx = get_mem_index(s);
8118
8119 for (i = j = 0; i < 16; i++) {
8120 if (!(list & (1 << i))) {
8121 continue;
8122 }
8123
8124 if (user && i != 15) {
8125 tmp = tcg_temp_new_i32();
8126 tmp2 = tcg_const_i32(i);
8127 gen_helper_get_user_reg(tmp, cpu_env, tmp2);
8128 tcg_temp_free_i32(tmp2);
8129 } else {
8130 tmp = load_reg(s, i);
8131 }
8132 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
8133 tcg_temp_free_i32(tmp);
8134
8135 /* No need to add after the last transfer. */
8136 if (++j != n) {
8137 tcg_gen_addi_i32(addr, addr, 4);
8138 }
8139 }
8140
8141 op_addr_block_post(s, a, addr, n);
8142 clear_eci_state(s);
8143 return true;
8144 }
8145
8146 static bool trans_STM(DisasContext *s, arg_ldst_block *a)
8147 {
8148 /* BitCount(list) < 1 is UNPREDICTABLE */
8149 return op_stm(s, a, 1);
8150 }
8151
8152 static bool trans_STM_t32(DisasContext *s, arg_ldst_block *a)
8153 {
8154 /* Writeback register in register list is UNPREDICTABLE for T32. */
8155 if (a->w && (a->list & (1 << a->rn))) {
8156 unallocated_encoding(s);
8157 return true;
8158 }
8159 /* BitCount(list) < 2 is UNPREDICTABLE */
8160 return op_stm(s, a, 2);
8161 }
8162
8163 static bool do_ldm(DisasContext *s, arg_ldst_block *a, int min_n)
8164 {
8165 int i, j, n, list, mem_idx;
8166 bool loaded_base;
8167 bool user = a->u;
8168 bool exc_return = false;
8169 TCGv_i32 addr, tmp, tmp2, loaded_var;
8170
8171 if (user) {
8172 /* LDM (user), LDM (exception return) */
8173 if (IS_USER(s)) {
8174 /* Only usable in supervisor mode. */
8175 unallocated_encoding(s);
8176 return true;
8177 }
8178 if (extract32(a->list, 15, 1)) {
8179 exc_return = true;
8180 user = false;
8181 } else {
8182 /* LDM (user) does not allow writeback. */
8183 if (a->w) {
8184 unallocated_encoding(s);
8185 return true;
8186 }
8187 }
8188 }
8189
8190 list = a->list;
8191 n = ctpop16(list);
8192 if (n < min_n || a->rn == 15) {
8193 unallocated_encoding(s);
8194 return true;
8195 }
8196
8197 s->eci_handled = true;
8198
8199 addr = op_addr_block_pre(s, a, n);
8200 mem_idx = get_mem_index(s);
8201 loaded_base = false;
8202 loaded_var = NULL;
8203
8204 for (i = j = 0; i < 16; i++) {
8205 if (!(list & (1 << i))) {
8206 continue;
8207 }
8208
8209 tmp = tcg_temp_new_i32();
8210 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
8211 if (user) {
8212 tmp2 = tcg_const_i32(i);
8213 gen_helper_set_user_reg(cpu_env, tmp2, tmp);
8214 tcg_temp_free_i32(tmp2);
8215 tcg_temp_free_i32(tmp);
8216 } else if (i == a->rn) {
8217 loaded_var = tmp;
8218 loaded_base = true;
8219 } else if (i == 15 && exc_return) {
8220 store_pc_exc_ret(s, tmp);
8221 } else {
8222 store_reg_from_load(s, i, tmp);
8223 }
8224
8225 /* No need to add after the last transfer. */
8226 if (++j != n) {
8227 tcg_gen_addi_i32(addr, addr, 4);
8228 }
8229 }
8230
8231 op_addr_block_post(s, a, addr, n);
8232
8233 if (loaded_base) {
8234 /* Note that we reject base == pc above. */
8235 store_reg(s, a->rn, loaded_var);
8236 }
8237
8238 if (exc_return) {
8239 /* Restore CPSR from SPSR. */
8240 tmp = load_cpu_field(spsr);
8241 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
8242 gen_io_start();
8243 }
8244 gen_helper_cpsr_write_eret(cpu_env, tmp);
8245 tcg_temp_free_i32(tmp);
8246 /* Must exit loop to check un-masked IRQs */
8247 s->base.is_jmp = DISAS_EXIT;
8248 }
8249 clear_eci_state(s);
8250 return true;
8251 }
8252
8253 static bool trans_LDM_a32(DisasContext *s, arg_ldst_block *a)
8254 {
8255 /*
8256 * Writeback register in register list is UNPREDICTABLE
8257 * for ArchVersion() >= 7. Prior to v7, A32 would write
8258 * an UNKNOWN value to the base register.
8259 */
8260 if (ENABLE_ARCH_7 && a->w && (a->list & (1 << a->rn))) {
8261 unallocated_encoding(s);
8262 return true;
8263 }
8264 /* BitCount(list) < 1 is UNPREDICTABLE */
8265 return do_ldm(s, a, 1);
8266 }
8267
8268 static bool trans_LDM_t32(DisasContext *s, arg_ldst_block *a)
8269 {
8270 /* Writeback register in register list is UNPREDICTABLE for T32. */
8271 if (a->w && (a->list & (1 << a->rn))) {
8272 unallocated_encoding(s);
8273 return true;
8274 }
8275 /* BitCount(list) < 2 is UNPREDICTABLE */
8276 return do_ldm(s, a, 2);
8277 }
8278
8279 static bool trans_LDM_t16(DisasContext *s, arg_ldst_block *a)
8280 {
8281 /* Writeback is conditional on the base register not being loaded. */
8282 a->w = !(a->list & (1 << a->rn));
8283 /* BitCount(list) < 1 is UNPREDICTABLE */
8284 return do_ldm(s, a, 1);
8285 }
8286
8287 static bool trans_CLRM(DisasContext *s, arg_CLRM *a)
8288 {
8289 int i;
8290 TCGv_i32 zero;
8291
8292 if (!dc_isar_feature(aa32_m_sec_state, s)) {
8293 return false;
8294 }
8295
8296 if (extract32(a->list, 13, 1)) {
8297 return false;
8298 }
8299
8300 if (!a->list) {
8301 /* UNPREDICTABLE; we choose to UNDEF */
8302 return false;
8303 }
8304
8305 s->eci_handled = true;
8306
8307 zero = tcg_const_i32(0);
8308 for (i = 0; i < 15; i++) {
8309 if (extract32(a->list, i, 1)) {
8310 /* Clear R[i] */
8311 tcg_gen_mov_i32(cpu_R[i], zero);
8312 }
8313 }
8314 if (extract32(a->list, 15, 1)) {
8315 /*
8316 * Clear APSR (by calling the MSR helper with the same argument
8317 * as for "MSR APSR_nzcvqg, Rn": mask = 0b1100, SYSM=0)
8318 */
8319 TCGv_i32 maskreg = tcg_const_i32(0xc << 8);
8320 gen_helper_v7m_msr(cpu_env, maskreg, zero);
8321 tcg_temp_free_i32(maskreg);
8322 }
8323 tcg_temp_free_i32(zero);
8324 clear_eci_state(s);
8325 return true;
8326 }
8327
8328 /*
8329 * Branch, branch with link
8330 */
8331
8332 static bool trans_B(DisasContext *s, arg_i *a)
8333 {
8334 gen_jmp(s, read_pc(s) + a->imm);
8335 return true;
8336 }
8337
8338 static bool trans_B_cond_thumb(DisasContext *s, arg_ci *a)
8339 {
8340 /* This has cond from encoding, required to be outside IT block. */
8341 if (a->cond >= 0xe) {
8342 return false;
8343 }
8344 if (s->condexec_mask) {
8345 unallocated_encoding(s);
8346 return true;
8347 }
8348 arm_skip_unless(s, a->cond);
8349 gen_jmp(s, read_pc(s) + a->imm);
8350 return true;
8351 }
8352
8353 static bool trans_BL(DisasContext *s, arg_i *a)
8354 {
8355 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | s->thumb);
8356 gen_jmp(s, read_pc(s) + a->imm);
8357 return true;
8358 }
8359
8360 static bool trans_BLX_i(DisasContext *s, arg_BLX_i *a)
8361 {
8362 /*
8363 * BLX <imm> would be useless on M-profile; the encoding space
8364 * is used for other insns from v8.1M onward, and UNDEFs before that.
8365 */
8366 if (arm_dc_feature(s, ARM_FEATURE_M)) {
8367 return false;
8368 }
8369
8370 /* For A32, ARM_FEATURE_V5 is checked near the start of the uncond block. */
8371 if (s->thumb && (a->imm & 2)) {
8372 return false;
8373 }
8374 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | s->thumb);
8375 store_cpu_field_constant(!s->thumb, thumb);
8376 gen_jmp(s, (read_pc(s) & ~3) + a->imm);
8377 return true;
8378 }
8379
8380 static bool trans_BL_BLX_prefix(DisasContext *s, arg_BL_BLX_prefix *a)
8381 {
8382 assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
8383 tcg_gen_movi_i32(cpu_R[14], read_pc(s) + (a->imm << 12));
8384 return true;
8385 }
8386
8387 static bool trans_BL_suffix(DisasContext *s, arg_BL_suffix *a)
8388 {
8389 TCGv_i32 tmp = tcg_temp_new_i32();
8390
8391 assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
8392 tcg_gen_addi_i32(tmp, cpu_R[14], (a->imm << 1) | 1);
8393 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | 1);
8394 gen_bx(s, tmp);
8395 return true;
8396 }
8397
8398 static bool trans_BLX_suffix(DisasContext *s, arg_BLX_suffix *a)
8399 {
8400 TCGv_i32 tmp;
8401
8402 assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
8403 if (!ENABLE_ARCH_5) {
8404 return false;
8405 }
8406 tmp = tcg_temp_new_i32();
8407 tcg_gen_addi_i32(tmp, cpu_R[14], a->imm << 1);
8408 tcg_gen_andi_i32(tmp, tmp, 0xfffffffc);
8409 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | 1);
8410 gen_bx(s, tmp);
8411 return true;
8412 }
8413
8414 static bool trans_BF(DisasContext *s, arg_BF *a)
8415 {
8416 /*
8417 * M-profile branch future insns. The architecture permits an
8418 * implementation to implement these as NOPs (equivalent to
8419 * discarding the LO_BRANCH_INFO cache immediately), and we
8420 * take that IMPDEF option because for QEMU a "real" implementation
8421 * would be complicated and wouldn't execute any faster.
8422 */
8423 if (!dc_isar_feature(aa32_lob, s)) {
8424 return false;
8425 }
8426 if (a->boff == 0) {
8427 /* SEE "Related encodings" (loop insns) */
8428 return false;
8429 }
8430 /* Handle as NOP */
8431 return true;
8432 }
8433
8434 static bool trans_DLS(DisasContext *s, arg_DLS *a)
8435 {
8436 /* M-profile low-overhead loop start */
8437 TCGv_i32 tmp;
8438
8439 if (!dc_isar_feature(aa32_lob, s)) {
8440 return false;
8441 }
8442 if (a->rn == 13 || a->rn == 15) {
8443 /*
8444 * For DLSTP rn == 15 is a related encoding (LCTP); the
8445 * other cases caught by this condition are all
8446 * CONSTRAINED UNPREDICTABLE: we choose to UNDEF
8447 */
8448 return false;
8449 }
8450
8451 if (a->size != 4) {
8452 /* DLSTP */
8453 if (!dc_isar_feature(aa32_mve, s)) {
8454 return false;
8455 }
8456 if (!vfp_access_check(s)) {
8457 return true;
8458 }
8459 }
8460
8461 /* Not a while loop: set LR to the count, and set LTPSIZE for DLSTP */
8462 tmp = load_reg(s, a->rn);
8463 store_reg(s, 14, tmp);
8464 if (a->size != 4) {
8465 /* DLSTP: set FPSCR.LTPSIZE */
8466 tmp = tcg_const_i32(a->size);
8467 store_cpu_field(tmp, v7m.ltpsize);
8468 s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
8469 }
8470 return true;
8471 }
8472
8473 static bool trans_WLS(DisasContext *s, arg_WLS *a)
8474 {
8475 /* M-profile low-overhead while-loop start */
8476 TCGv_i32 tmp;
8477 TCGLabel *nextlabel;
8478
8479 if (!dc_isar_feature(aa32_lob, s)) {
8480 return false;
8481 }
8482 if (a->rn == 13 || a->rn == 15) {
8483 /*
8484 * For WLSTP rn == 15 is a related encoding (LE); the
8485 * other cases caught by this condition are all
8486 * CONSTRAINED UNPREDICTABLE: we choose to UNDEF
8487 */
8488 return false;
8489 }
8490 if (s->condexec_mask) {
8491 /*
8492 * WLS in an IT block is CONSTRAINED UNPREDICTABLE;
8493 * we choose to UNDEF, because otherwise our use of
8494 * gen_goto_tb(1) would clash with the use of TB exit 1
8495 * in the dc->condjmp condition-failed codepath in
8496 * arm_tr_tb_stop() and we'd get an assertion.
8497 */
8498 return false;
8499 }
8500 if (a->size != 4) {
8501 /* WLSTP */
8502 if (!dc_isar_feature(aa32_mve, s)) {
8503 return false;
8504 }
8505 /*
8506 * We need to check that the FPU is enabled here, but mustn't
8507 * call vfp_access_check() to do that because we don't want to
8508 * do the lazy state preservation in the "loop count is zero" case.
8509 * Do the check-and-raise-exception by hand.
8510 */
8511 if (s->fp_excp_el) {
8512 gen_exception_insn(s, s->pc_curr, EXCP_NOCP,
8513 syn_uncategorized(), s->fp_excp_el);
8514 return true;
8515 }
8516 }
8517
8518 nextlabel = gen_new_label();
8519 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_R[a->rn], 0, nextlabel);
8520 tmp = load_reg(s, a->rn);
8521 store_reg(s, 14, tmp);
8522 if (a->size != 4) {
8523 /*
8524 * WLSTP: set FPSCR.LTPSIZE. This requires that we do the
8525 * lazy state preservation, new FP context creation, etc,
8526 * that vfp_access_check() does. We know that the actual
8527 * access check will succeed (ie it won't generate code that
8528 * throws an exception) because we did that check by hand earlier.
8529 */
8530 bool ok = vfp_access_check(s);
8531 assert(ok);
8532 tmp = tcg_const_i32(a->size);
8533 store_cpu_field(tmp, v7m.ltpsize);
8534 /*
8535 * LTPSIZE updated, but MVE_NO_PRED will always be the same thing (0)
8536 * when we take this upcoming exit from this TB, so gen_jmp_tb() is OK.
8537 */
8538 }
8539 gen_jmp_tb(s, s->base.pc_next, 1);
8540
8541 gen_set_label(nextlabel);
8542 gen_jmp(s, read_pc(s) + a->imm);
8543 return true;
8544 }
8545
8546 static bool trans_LE(DisasContext *s, arg_LE *a)
8547 {
8548 /*
8549 * M-profile low-overhead loop end. The architecture permits an
8550 * implementation to discard the LO_BRANCH_INFO cache at any time,
8551 * and we take the IMPDEF option to never set it in the first place
8552 * (equivalent to always discarding it immediately), because for QEMU
8553 * a "real" implementation would be complicated and wouldn't execute
8554 * any faster.
8555 */
8556 TCGv_i32 tmp;
8557 TCGLabel *loopend;
8558 bool fpu_active;
8559
8560 if (!dc_isar_feature(aa32_lob, s)) {
8561 return false;
8562 }
8563 if (a->f && a->tp) {
8564 return false;
8565 }
8566 if (s->condexec_mask) {
8567 /*
8568 * LE in an IT block is CONSTRAINED UNPREDICTABLE;
8569 * we choose to UNDEF, because otherwise our use of
8570 * gen_goto_tb(1) would clash with the use of TB exit 1
8571 * in the dc->condjmp condition-failed codepath in
8572 * arm_tr_tb_stop() and we'd get an assertion.
8573 */
8574 return false;
8575 }
8576 if (a->tp) {
8577 /* LETP */
8578 if (!dc_isar_feature(aa32_mve, s)) {
8579 return false;
8580 }
8581 if (!vfp_access_check(s)) {
8582 s->eci_handled = true;
8583 return true;
8584 }
8585 }
8586
8587 /* LE/LETP is OK with ECI set and leaves it untouched */
8588 s->eci_handled = true;
8589
8590 /*
8591 * With MVE, LTPSIZE might not be 4, and we must emit an INVSTATE
8592 * UsageFault exception for the LE insn in that case. Note that we
8593 * are not directly checking FPSCR.LTPSIZE but instead check the
8594 * pseudocode LTPSIZE() function, which returns 4 if the FPU is
8595 * not currently active (ie ActiveFPState() returns false). We
8596 * can identify not-active purely from our TB state flags, as the
8597 * FPU is active only if:
8598 * the FPU is enabled
8599 * AND lazy state preservation is not active
8600 * AND we do not need a new fp context (this is the ASPEN/FPCA check)
8601 *
8602 * Usually we don't need to care about this distinction between
8603 * LTPSIZE and FPSCR.LTPSIZE, because the code in vfp_access_check()
8604 * will either take an exception or clear the conditions that make
8605 * the FPU not active. But LE is an unusual case of a non-FP insn
8606 * that looks at LTPSIZE.
8607 */
8608 fpu_active = !s->fp_excp_el && !s->v7m_lspact && !s->v7m_new_fp_ctxt_needed;
8609
8610 if (!a->tp && dc_isar_feature(aa32_mve, s) && fpu_active) {
8611 /* Need to do a runtime check for LTPSIZE != 4 */
8612 TCGLabel *skipexc = gen_new_label();
8613 tmp = load_cpu_field(v7m.ltpsize);
8614 tcg_gen_brcondi_i32(TCG_COND_EQ, tmp, 4, skipexc);
8615 tcg_temp_free_i32(tmp);
8616 gen_exception_insn(s, s->pc_curr, EXCP_INVSTATE, syn_uncategorized(),
8617 default_exception_el(s));
8618 gen_set_label(skipexc);
8619 }
8620
8621 if (a->f) {
8622 /* Loop-forever: just jump back to the loop start */
8623 gen_jmp(s, read_pc(s) - a->imm);
8624 return true;
8625 }
8626
8627 /*
8628 * Not loop-forever. If LR <= loop-decrement-value this is the last loop.
8629 * For LE, we know at this point that LTPSIZE must be 4 and the
8630 * loop decrement value is 1. For LETP we need to calculate the decrement
8631 * value from LTPSIZE.
8632 */
8633 loopend = gen_new_label();
8634 if (!a->tp) {
8635 tcg_gen_brcondi_i32(TCG_COND_LEU, cpu_R[14], 1, loopend);
8636 tcg_gen_addi_i32(cpu_R[14], cpu_R[14], -1);
8637 } else {
8638 /*
8639 * Decrement by 1 << (4 - LTPSIZE). We need to use a TCG local
8640 * so that decr stays live after the brcondi.
8641 */
8642 TCGv_i32 decr = tcg_temp_local_new_i32();
8643 TCGv_i32 ltpsize = load_cpu_field(v7m.ltpsize);
8644 tcg_gen_sub_i32(decr, tcg_constant_i32(4), ltpsize);
8645 tcg_gen_shl_i32(decr, tcg_constant_i32(1), decr);
8646 tcg_temp_free_i32(ltpsize);
8647
8648 tcg_gen_brcond_i32(TCG_COND_LEU, cpu_R[14], decr, loopend);
8649
8650 tcg_gen_sub_i32(cpu_R[14], cpu_R[14], decr);
8651 tcg_temp_free_i32(decr);
8652 }
8653 /* Jump back to the loop start */
8654 gen_jmp(s, read_pc(s) - a->imm);
8655
8656 gen_set_label(loopend);
8657 if (a->tp) {
8658 /* Exits from tail-pred loops must reset LTPSIZE to 4 */
8659 tmp = tcg_const_i32(4);
8660 store_cpu_field(tmp, v7m.ltpsize);
8661 }
8662 /* End TB, continuing to following insn */
8663 gen_jmp_tb(s, s->base.pc_next, 1);
8664 return true;
8665 }
8666
8667 static bool trans_LCTP(DisasContext *s, arg_LCTP *a)
8668 {
8669 /*
8670 * M-profile Loop Clear with Tail Predication. Since our implementation
8671 * doesn't cache branch information, all we need to do is reset
8672 * FPSCR.LTPSIZE to 4.
8673 */
8674
8675 if (!dc_isar_feature(aa32_lob, s) ||
8676 !dc_isar_feature(aa32_mve, s)) {
8677 return false;
8678 }
8679
8680 if (!vfp_access_check(s)) {
8681 return true;
8682 }
8683
8684 store_cpu_field_constant(4, v7m.ltpsize);
8685 return true;
8686 }
8687
8688 static bool trans_VCTP(DisasContext *s, arg_VCTP *a)
8689 {
8690 /*
8691 * M-profile Create Vector Tail Predicate. This insn is itself
8692 * predicated and is subject to beatwise execution.
8693 */
8694 TCGv_i32 rn_shifted, masklen;
8695
8696 if (!dc_isar_feature(aa32_mve, s) || a->rn == 13 || a->rn == 15) {
8697 return false;
8698 }
8699
8700 if (!mve_eci_check(s) || !vfp_access_check(s)) {
8701 return true;
8702 }
8703
8704 /*
8705 * We pre-calculate the mask length here to avoid having
8706 * to have multiple helpers specialized for size.
8707 * We pass the helper "rn <= (1 << (4 - size)) ? (rn << size) : 16".
8708 */
8709 rn_shifted = tcg_temp_new_i32();
8710 masklen = load_reg(s, a->rn);
8711 tcg_gen_shli_i32(rn_shifted, masklen, a->size);
8712 tcg_gen_movcond_i32(TCG_COND_LEU, masklen,
8713 masklen, tcg_constant_i32(1 << (4 - a->size)),
8714 rn_shifted, tcg_constant_i32(16));
8715 gen_helper_mve_vctp(cpu_env, masklen);
8716 tcg_temp_free_i32(masklen);
8717 tcg_temp_free_i32(rn_shifted);
8718 /* This insn updates predication bits */
8719 s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
8720 mve_update_eci(s);
8721 return true;
8722 }
8723
8724 static bool op_tbranch(DisasContext *s, arg_tbranch *a, bool half)
8725 {
8726 TCGv_i32 addr, tmp;
8727
8728 tmp = load_reg(s, a->rm);
8729 if (half) {
8730 tcg_gen_add_i32(tmp, tmp, tmp);
8731 }
8732 addr = load_reg(s, a->rn);
8733 tcg_gen_add_i32(addr, addr, tmp);
8734
8735 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), half ? MO_UW : MO_UB);
8736 tcg_temp_free_i32(addr);
8737
8738 tcg_gen_add_i32(tmp, tmp, tmp);
8739 tcg_gen_addi_i32(tmp, tmp, read_pc(s));
8740 store_reg(s, 15, tmp);
8741 return true;
8742 }
8743
8744 static bool trans_TBB(DisasContext *s, arg_tbranch *a)
8745 {
8746 return op_tbranch(s, a, false);
8747 }
8748
8749 static bool trans_TBH(DisasContext *s, arg_tbranch *a)
8750 {
8751 return op_tbranch(s, a, true);
8752 }
8753
8754 static bool trans_CBZ(DisasContext *s, arg_CBZ *a)
8755 {
8756 TCGv_i32 tmp = load_reg(s, a->rn);
8757
8758 arm_gen_condlabel(s);
8759 tcg_gen_brcondi_i32(a->nz ? TCG_COND_EQ : TCG_COND_NE,
8760 tmp, 0, s->condlabel);
8761 tcg_temp_free_i32(tmp);
8762 gen_jmp(s, read_pc(s) + a->imm);
8763 return true;
8764 }
8765
8766 /*
8767 * Supervisor call - both T32 & A32 come here so we need to check
8768 * which mode we are in when checking for semihosting.
8769 */
8770
8771 static bool trans_SVC(DisasContext *s, arg_SVC *a)
8772 {
8773 const uint32_t semihost_imm = s->thumb ? 0xab : 0x123456;
8774
8775 if (!arm_dc_feature(s, ARM_FEATURE_M) && semihosting_enabled() &&
8776 #ifndef CONFIG_USER_ONLY
8777 !IS_USER(s) &&
8778 #endif
8779 (a->imm == semihost_imm)) {
8780 gen_exception_internal_insn(s, s->pc_curr, EXCP_SEMIHOST);
8781 } else {
8782 gen_set_pc_im(s, s->base.pc_next);
8783 s->svc_imm = a->imm;
8784 s->base.is_jmp = DISAS_SWI;
8785 }
8786 return true;
8787 }
8788
8789 /*
8790 * Unconditional system instructions
8791 */
8792
8793 static bool trans_RFE(DisasContext *s, arg_RFE *a)
8794 {
8795 static const int8_t pre_offset[4] = {
8796 /* DA */ -4, /* IA */ 0, /* DB */ -8, /* IB */ 4
8797 };
8798 static const int8_t post_offset[4] = {
8799 /* DA */ -8, /* IA */ 4, /* DB */ -4, /* IB */ 0
8800 };
8801 TCGv_i32 addr, t1, t2;
8802
8803 if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
8804 return false;
8805 }
8806 if (IS_USER(s)) {
8807 unallocated_encoding(s);
8808 return true;
8809 }
8810
8811 addr = load_reg(s, a->rn);
8812 tcg_gen_addi_i32(addr, addr, pre_offset[a->pu]);
8813
8814 /* Load PC into tmp and CPSR into tmp2. */
8815 t1 = tcg_temp_new_i32();
8816 gen_aa32_ld_i32(s, t1, addr, get_mem_index(s), MO_UL | MO_ALIGN);
8817 tcg_gen_addi_i32(addr, addr, 4);
8818 t2 = tcg_temp_new_i32();
8819 gen_aa32_ld_i32(s, t2, addr, get_mem_index(s), MO_UL | MO_ALIGN);
8820
8821 if (a->w) {
8822 /* Base writeback. */
8823 tcg_gen_addi_i32(addr, addr, post_offset[a->pu]);
8824 store_reg(s, a->rn, addr);
8825 } else {
8826 tcg_temp_free_i32(addr);
8827 }
8828 gen_rfe(s, t1, t2);
8829 return true;
8830 }
8831
8832 static bool trans_SRS(DisasContext *s, arg_SRS *a)
8833 {
8834 if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
8835 return false;
8836 }
8837 gen_srs(s, a->mode, a->pu, a->w);
8838 return true;
8839 }
8840
8841 static bool trans_CPS(DisasContext *s, arg_CPS *a)
8842 {
8843 uint32_t mask, val;
8844
8845 if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
8846 return false;
8847 }
8848 if (IS_USER(s)) {
8849 /* Implemented as NOP in user mode. */
8850 return true;
8851 }
8852 /* TODO: There are quite a lot of UNPREDICTABLE argument combinations. */
8853
8854 mask = val = 0;
8855 if (a->imod & 2) {
8856 if (a->A) {
8857 mask |= CPSR_A;
8858 }
8859 if (a->I) {
8860 mask |= CPSR_I;
8861 }
8862 if (a->F) {
8863 mask |= CPSR_F;
8864 }
8865 if (a->imod & 1) {
8866 val |= mask;
8867 }
8868 }
8869 if (a->M) {
8870 mask |= CPSR_M;
8871 val |= a->mode;
8872 }
8873 if (mask) {
8874 gen_set_psr_im(s, mask, 0, val);
8875 }
8876 return true;
8877 }
8878
8879 static bool trans_CPS_v7m(DisasContext *s, arg_CPS_v7m *a)
8880 {
8881 TCGv_i32 tmp, addr, el;
8882
8883 if (!arm_dc_feature(s, ARM_FEATURE_M)) {
8884 return false;
8885 }
8886 if (IS_USER(s)) {
8887 /* Implemented as NOP in user mode. */
8888 return true;
8889 }
8890
8891 tmp = tcg_const_i32(a->im);
8892 /* FAULTMASK */
8893 if (a->F) {
8894 addr = tcg_const_i32(19);
8895 gen_helper_v7m_msr(cpu_env, addr, tmp);
8896 tcg_temp_free_i32(addr);
8897 }
8898 /* PRIMASK */
8899 if (a->I) {
8900 addr = tcg_const_i32(16);
8901 gen_helper_v7m_msr(cpu_env, addr, tmp);
8902 tcg_temp_free_i32(addr);
8903 }
8904 el = tcg_const_i32(s->current_el);
8905 gen_helper_rebuild_hflags_m32(cpu_env, el);
8906 tcg_temp_free_i32(el);
8907 tcg_temp_free_i32(tmp);
8908 gen_lookup_tb(s);
8909 return true;
8910 }
8911
8912 /*
8913 * Clear-Exclusive, Barriers
8914 */
8915
8916 static bool trans_CLREX(DisasContext *s, arg_CLREX *a)
8917 {
8918 if (s->thumb
8919 ? !ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)
8920 : !ENABLE_ARCH_6K) {
8921 return false;
8922 }
8923 gen_clrex(s);
8924 return true;
8925 }
8926
8927 static bool trans_DSB(DisasContext *s, arg_DSB *a)
8928 {
8929 if (!ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)) {
8930 return false;
8931 }
8932 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
8933 return true;
8934 }
8935
8936 static bool trans_DMB(DisasContext *s, arg_DMB *a)
8937 {
8938 return trans_DSB(s, NULL);
8939 }
8940
8941 static bool trans_ISB(DisasContext *s, arg_ISB *a)
8942 {
8943 if (!ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)) {
8944 return false;
8945 }
8946 /*
8947 * We need to break the TB after this insn to execute
8948 * self-modifying code correctly and also to take
8949 * any pending interrupts immediately.
8950 */
8951 s->base.is_jmp = DISAS_TOO_MANY;
8952 return true;
8953 }
8954
8955 static bool trans_SB(DisasContext *s, arg_SB *a)
8956 {
8957 if (!dc_isar_feature(aa32_sb, s)) {
8958 return false;
8959 }
8960 /*
8961 * TODO: There is no speculation barrier opcode
8962 * for TCG; MB and end the TB instead.
8963 */
8964 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
8965 s->base.is_jmp = DISAS_TOO_MANY;
8966 return true;
8967 }
8968
8969 static bool trans_SETEND(DisasContext *s, arg_SETEND *a)
8970 {
8971 if (!ENABLE_ARCH_6) {
8972 return false;
8973 }
8974 if (a->E != (s->be_data == MO_BE)) {
8975 gen_helper_setend(cpu_env);
8976 s->base.is_jmp = DISAS_UPDATE_EXIT;
8977 }
8978 return true;
8979 }
8980
8981 /*
8982 * Preload instructions
8983 * All are nops, contingent on the appropriate arch level.
8984 */
8985
8986 static bool trans_PLD(DisasContext *s, arg_PLD *a)
8987 {
8988 return ENABLE_ARCH_5TE;
8989 }
8990
8991 static bool trans_PLDW(DisasContext *s, arg_PLD *a)
8992 {
8993 return arm_dc_feature(s, ARM_FEATURE_V7MP);
8994 }
8995
8996 static bool trans_PLI(DisasContext *s, arg_PLD *a)
8997 {
8998 return ENABLE_ARCH_7;
8999 }
9000
9001 /*
9002 * If-then
9003 */
9004
9005 static bool trans_IT(DisasContext *s, arg_IT *a)
9006 {
9007 int cond_mask = a->cond_mask;
9008
9009 /*
9010 * No actual code generated for this insn, just setup state.
9011 *
9012 * Combinations of firstcond and mask which set up an 0b1111
9013 * condition are UNPREDICTABLE; we take the CONSTRAINED
9014 * UNPREDICTABLE choice to treat 0b1111 the same as 0b1110,
9015 * i.e. both meaning "execute always".
9016 */
9017 s->condexec_cond = (cond_mask >> 4) & 0xe;
9018 s->condexec_mask = cond_mask & 0x1f;
9019 return true;
9020 }
9021
9022 /* v8.1M CSEL/CSINC/CSNEG/CSINV */
9023 static bool trans_CSEL(DisasContext *s, arg_CSEL *a)
9024 {
9025 TCGv_i32 rn, rm, zero;
9026 DisasCompare c;
9027
9028 if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
9029 return false;
9030 }
9031
9032 if (a->rm == 13) {
9033 /* SEE "Related encodings" (MVE shifts) */
9034 return false;
9035 }
9036
9037 if (a->rd == 13 || a->rd == 15 || a->rn == 13 || a->fcond >= 14) {
9038 /* CONSTRAINED UNPREDICTABLE: we choose to UNDEF */
9039 return false;
9040 }
9041
9042 /* In this insn input reg fields of 0b1111 mean "zero", not "PC" */
9043 if (a->rn == 15) {
9044 rn = tcg_const_i32(0);
9045 } else {
9046 rn = load_reg(s, a->rn);
9047 }
9048 if (a->rm == 15) {
9049 rm = tcg_const_i32(0);
9050 } else {
9051 rm = load_reg(s, a->rm);
9052 }
9053
9054 switch (a->op) {
9055 case 0: /* CSEL */
9056 break;
9057 case 1: /* CSINC */
9058 tcg_gen_addi_i32(rm, rm, 1);
9059 break;
9060 case 2: /* CSINV */
9061 tcg_gen_not_i32(rm, rm);
9062 break;
9063 case 3: /* CSNEG */
9064 tcg_gen_neg_i32(rm, rm);
9065 break;
9066 default:
9067 g_assert_not_reached();
9068 }
9069
9070 arm_test_cc(&c, a->fcond);
9071 zero = tcg_const_i32(0);
9072 tcg_gen_movcond_i32(c.cond, rn, c.value, zero, rn, rm);
9073 arm_free_cc(&c);
9074 tcg_temp_free_i32(zero);
9075
9076 store_reg(s, a->rd, rn);
9077 tcg_temp_free_i32(rm);
9078
9079 return true;
9080 }
9081
9082 /*
9083 * Legacy decoder.
9084 */
9085
9086 static void disas_arm_insn(DisasContext *s, unsigned int insn)
9087 {
9088 unsigned int cond = insn >> 28;
9089
9090 /* M variants do not implement ARM mode; this must raise the INVSTATE
9091 * UsageFault exception.
9092 */
9093 if (arm_dc_feature(s, ARM_FEATURE_M)) {
9094 gen_exception_insn(s, s->pc_curr, EXCP_INVSTATE, syn_uncategorized(),
9095 default_exception_el(s));
9096 return;
9097 }
9098
9099 if (s->pstate_il) {
9100 /*
9101 * Illegal execution state. This has priority over BTI
9102 * exceptions, but comes after instruction abort exceptions.
9103 */
9104 gen_exception_insn(s, s->pc_curr, EXCP_UDEF,
9105 syn_illegalstate(), default_exception_el(s));
9106 return;
9107 }
9108
9109 if (cond == 0xf) {
9110 /* In ARMv3 and v4 the NV condition is UNPREDICTABLE; we
9111 * choose to UNDEF. In ARMv5 and above the space is used
9112 * for miscellaneous unconditional instructions.
9113 */
9114 if (!arm_dc_feature(s, ARM_FEATURE_V5)) {
9115 unallocated_encoding(s);
9116 return;
9117 }
9118
9119 /* Unconditional instructions. */
9120 /* TODO: Perhaps merge these into one decodetree output file. */
9121 if (disas_a32_uncond(s, insn) ||
9122 disas_vfp_uncond(s, insn) ||
9123 disas_neon_dp(s, insn) ||
9124 disas_neon_ls(s, insn) ||
9125 disas_neon_shared(s, insn)) {
9126 return;
9127 }
9128 /* fall back to legacy decoder */
9129
9130 if ((insn & 0x0e000f00) == 0x0c000100) {
9131 if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
9132 /* iWMMXt register transfer. */
9133 if (extract32(s->c15_cpar, 1, 1)) {
9134 if (!disas_iwmmxt_insn(s, insn)) {
9135 return;
9136 }
9137 }
9138 }
9139 }
9140 goto illegal_op;
9141 }
9142 if (cond != 0xe) {
9143 /* if not always execute, we generate a conditional jump to
9144 next instruction */
9145 arm_skip_unless(s, cond);
9146 }
9147
9148 /* TODO: Perhaps merge these into one decodetree output file. */
9149 if (disas_a32(s, insn) ||
9150 disas_vfp(s, insn)) {
9151 return;
9152 }
9153 /* fall back to legacy decoder */
9154 /* TODO: convert xscale/iwmmxt decoder to decodetree ?? */
9155 if (arm_dc_feature(s, ARM_FEATURE_XSCALE)) {
9156 if (((insn & 0x0c000e00) == 0x0c000000)
9157 && ((insn & 0x03000000) != 0x03000000)) {
9158 /* Coprocessor insn, coprocessor 0 or 1 */
9159 disas_xscale_insn(s, insn);
9160 return;
9161 }
9162 }
9163
9164 illegal_op:
9165 unallocated_encoding(s);
9166 }
9167
9168 static bool thumb_insn_is_16bit(DisasContext *s, uint32_t pc, uint32_t insn)
9169 {
9170 /*
9171 * Return true if this is a 16 bit instruction. We must be precise
9172 * about this (matching the decode).
9173 */
9174 if ((insn >> 11) < 0x1d) {
9175 /* Definitely a 16-bit instruction */
9176 return true;
9177 }
9178
9179 /* Top five bits 0b11101 / 0b11110 / 0b11111 : this is the
9180 * first half of a 32-bit Thumb insn. Thumb-1 cores might
9181 * end up actually treating this as two 16-bit insns, though,
9182 * if it's half of a bl/blx pair that might span a page boundary.
9183 */
9184 if (arm_dc_feature(s, ARM_FEATURE_THUMB2) ||
9185 arm_dc_feature(s, ARM_FEATURE_M)) {
9186 /* Thumb2 cores (including all M profile ones) always treat
9187 * 32-bit insns as 32-bit.
9188 */
9189 return false;
9190 }
9191
9192 if ((insn >> 11) == 0x1e && pc - s->page_start < TARGET_PAGE_SIZE - 3) {
9193 /* 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix, and the suffix
9194 * is not on the next page; we merge this into a 32-bit
9195 * insn.
9196 */
9197 return false;
9198 }
9199 /* 0b1110_1xxx_xxxx_xxxx : BLX suffix (or UNDEF);
9200 * 0b1111_1xxx_xxxx_xxxx : BL suffix;
9201 * 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix on the end of a page
9202 * -- handle as single 16 bit insn
9203 */
9204 return true;
9205 }
9206
9207 /* Translate a 32-bit thumb instruction. */
9208 static void disas_thumb2_insn(DisasContext *s, uint32_t insn)
9209 {
9210 /*
9211 * ARMv6-M supports a limited subset of Thumb2 instructions.
9212 * Other Thumb1 architectures allow only 32-bit
9213 * combined BL/BLX prefix and suffix.
9214 */
9215 if (arm_dc_feature(s, ARM_FEATURE_M) &&
9216 !arm_dc_feature(s, ARM_FEATURE_V7)) {
9217 int i;
9218 bool found = false;
9219 static const uint32_t armv6m_insn[] = {0xf3808000 /* msr */,
9220 0xf3b08040 /* dsb */,
9221 0xf3b08050 /* dmb */,
9222 0xf3b08060 /* isb */,
9223 0xf3e08000 /* mrs */,
9224 0xf000d000 /* bl */};
9225 static const uint32_t armv6m_mask[] = {0xffe0d000,
9226 0xfff0d0f0,
9227 0xfff0d0f0,
9228 0xfff0d0f0,
9229 0xffe0d000,
9230 0xf800d000};
9231
9232 for (i = 0; i < ARRAY_SIZE(armv6m_insn); i++) {
9233 if ((insn & armv6m_mask[i]) == armv6m_insn[i]) {
9234 found = true;
9235 break;
9236 }
9237 }
9238 if (!found) {
9239 goto illegal_op;
9240 }
9241 } else if ((insn & 0xf800e800) != 0xf000e800) {
9242 if (!arm_dc_feature(s, ARM_FEATURE_THUMB2)) {
9243 unallocated_encoding(s);
9244 return;
9245 }
9246 }
9247
9248 if (arm_dc_feature(s, ARM_FEATURE_M)) {
9249 /*
9250 * NOCP takes precedence over any UNDEF for (almost) the
9251 * entire wide range of coprocessor-space encodings, so check
9252 * for it first before proceeding to actually decode eg VFP
9253 * insns. This decode also handles the few insns which are
9254 * in copro space but do not have NOCP checks (eg VLLDM, VLSTM).
9255 */
9256 if (disas_m_nocp(s, insn)) {
9257 return;
9258 }
9259 }
9260
9261 if ((insn & 0xef000000) == 0xef000000) {
9262 /*
9263 * T32 encodings 0b111p_1111_qqqq_qqqq_qqqq_qqqq_qqqq_qqqq
9264 * transform into
9265 * A32 encodings 0b1111_001p_qqqq_qqqq_qqqq_qqqq_qqqq_qqqq
9266 */
9267 uint32_t a32_insn = (insn & 0xe2ffffff) |
9268 ((insn & (1 << 28)) >> 4) | (1 << 28);
9269
9270 if (disas_neon_dp(s, a32_insn)) {
9271 return;
9272 }
9273 }
9274
9275 if ((insn & 0xff100000) == 0xf9000000) {
9276 /*
9277 * T32 encodings 0b1111_1001_ppp0_qqqq_qqqq_qqqq_qqqq_qqqq
9278 * transform into
9279 * A32 encodings 0b1111_0100_ppp0_qqqq_qqqq_qqqq_qqqq_qqqq
9280 */
9281 uint32_t a32_insn = (insn & 0x00ffffff) | 0xf4000000;
9282
9283 if (disas_neon_ls(s, a32_insn)) {
9284 return;
9285 }
9286 }
9287
9288 /*
9289 * TODO: Perhaps merge these into one decodetree output file.
9290 * Note disas_vfp is written for a32 with cond field in the
9291 * top nibble. The t32 encoding requires 0xe in the top nibble.
9292 */
9293 if (disas_t32(s, insn) ||
9294 disas_vfp_uncond(s, insn) ||
9295 disas_neon_shared(s, insn) ||
9296 disas_mve(s, insn) ||
9297 ((insn >> 28) == 0xe && disas_vfp(s, insn))) {
9298 return;
9299 }
9300
9301 illegal_op:
9302 unallocated_encoding(s);
9303 }
9304
9305 static void disas_thumb_insn(DisasContext *s, uint32_t insn)
9306 {
9307 if (!disas_t16(s, insn)) {
9308 unallocated_encoding(s);
9309 }
9310 }
9311
9312 static bool insn_crosses_page(CPUARMState *env, DisasContext *s)
9313 {
9314 /* Return true if the insn at dc->base.pc_next might cross a page boundary.
9315 * (False positives are OK, false negatives are not.)
9316 * We know this is a Thumb insn, and our caller ensures we are
9317 * only called if dc->base.pc_next is less than 4 bytes from the page
9318 * boundary, so we cross the page if the first 16 bits indicate
9319 * that this is a 32 bit insn.
9320 */
9321 uint16_t insn = arm_lduw_code(env, &s->base, s->base.pc_next, s->sctlr_b);
9322
9323 return !thumb_insn_is_16bit(s, s->base.pc_next, insn);
9324 }
9325
9326 static void arm_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cs)
9327 {
9328 DisasContext *dc = container_of(dcbase, DisasContext, base);
9329 CPUARMState *env = cs->env_ptr;
9330 ARMCPU *cpu = env_archcpu(env);
9331 CPUARMTBFlags tb_flags = arm_tbflags_from_tb(dc->base.tb);
9332 uint32_t condexec, core_mmu_idx;
9333
9334 dc->isar = &cpu->isar;
9335 dc->condjmp = 0;
9336
9337 dc->aarch64 = 0;
9338 /* If we are coming from secure EL0 in a system with a 32-bit EL3, then
9339 * there is no secure EL1, so we route exceptions to EL3.
9340 */
9341 dc->secure_routed_to_el3 = arm_feature(env, ARM_FEATURE_EL3) &&
9342 !arm_el_is_aa64(env, 3);
9343 dc->thumb = EX_TBFLAG_AM32(tb_flags, THUMB);
9344 dc->be_data = EX_TBFLAG_ANY(tb_flags, BE_DATA) ? MO_BE : MO_LE;
9345 condexec = EX_TBFLAG_AM32(tb_flags, CONDEXEC);
9346 /*
9347 * the CONDEXEC TB flags are CPSR bits [15:10][26:25]. On A-profile this
9348 * is always the IT bits. On M-profile, some of the reserved encodings
9349 * of IT are used instead to indicate either ICI or ECI, which
9350 * indicate partial progress of a restartable insn that was interrupted
9351 * partway through by an exception:
9352 * * if CONDEXEC[3:0] != 0b0000 : CONDEXEC is IT bits
9353 * * if CONDEXEC[3:0] == 0b0000 : CONDEXEC is ICI or ECI bits
9354 * In all cases CONDEXEC == 0 means "not in IT block or restartable
9355 * insn, behave normally".
9356 */
9357 dc->eci = dc->condexec_mask = dc->condexec_cond = 0;
9358 dc->eci_handled = false;
9359 dc->insn_eci_rewind = NULL;
9360 if (condexec & 0xf) {
9361 dc->condexec_mask = (condexec & 0xf) << 1;
9362 dc->condexec_cond = condexec >> 4;
9363 } else {
9364 if (arm_feature(env, ARM_FEATURE_M)) {
9365 dc->eci = condexec >> 4;
9366 }
9367 }
9368
9369 core_mmu_idx = EX_TBFLAG_ANY(tb_flags, MMUIDX);
9370 dc->mmu_idx = core_to_arm_mmu_idx(env, core_mmu_idx);
9371 dc->current_el = arm_mmu_idx_to_el(dc->mmu_idx);
9372 #if !defined(CONFIG_USER_ONLY)
9373 dc->user = (dc->current_el == 0);
9374 #endif
9375 dc->fp_excp_el = EX_TBFLAG_ANY(tb_flags, FPEXC_EL);
9376 dc->align_mem = EX_TBFLAG_ANY(tb_flags, ALIGN_MEM);
9377 dc->pstate_il = EX_TBFLAG_ANY(tb_flags, PSTATE__IL);
9378
9379 if (arm_feature(env, ARM_FEATURE_M)) {
9380 dc->vfp_enabled = 1;
9381 dc->be_data = MO_TE;
9382 dc->v7m_handler_mode = EX_TBFLAG_M32(tb_flags, HANDLER);
9383 dc->v8m_secure = arm_feature(env, ARM_FEATURE_M_SECURITY) &&
9384 regime_is_secure(env, dc->mmu_idx);
9385 dc->v8m_stackcheck = EX_TBFLAG_M32(tb_flags, STACKCHECK);
9386 dc->v8m_fpccr_s_wrong = EX_TBFLAG_M32(tb_flags, FPCCR_S_WRONG);
9387 dc->v7m_new_fp_ctxt_needed =
9388 EX_TBFLAG_M32(tb_flags, NEW_FP_CTXT_NEEDED);
9389 dc->v7m_lspact = EX_TBFLAG_M32(tb_flags, LSPACT);
9390 dc->mve_no_pred = EX_TBFLAG_M32(tb_flags, MVE_NO_PRED);
9391 } else {
9392 dc->debug_target_el = EX_TBFLAG_ANY(tb_flags, DEBUG_TARGET_EL);
9393 dc->sctlr_b = EX_TBFLAG_A32(tb_flags, SCTLR__B);
9394 dc->hstr_active = EX_TBFLAG_A32(tb_flags, HSTR_ACTIVE);
9395 dc->ns = EX_TBFLAG_A32(tb_flags, NS);
9396 dc->vfp_enabled = EX_TBFLAG_A32(tb_flags, VFPEN);
9397 if (arm_feature(env, ARM_FEATURE_XSCALE)) {
9398 dc->c15_cpar = EX_TBFLAG_A32(tb_flags, XSCALE_CPAR);
9399 } else {
9400 dc->vec_len = EX_TBFLAG_A32(tb_flags, VECLEN);
9401 dc->vec_stride = EX_TBFLAG_A32(tb_flags, VECSTRIDE);
9402 }
9403 }
9404 dc->cp_regs = cpu->cp_regs;
9405 dc->features = env->features;
9406
9407 /* Single step state. The code-generation logic here is:
9408 * SS_ACTIVE == 0:
9409 * generate code with no special handling for single-stepping (except
9410 * that anything that can make us go to SS_ACTIVE == 1 must end the TB;
9411 * this happens anyway because those changes are all system register or
9412 * PSTATE writes).
9413 * SS_ACTIVE == 1, PSTATE.SS == 1: (active-not-pending)
9414 * emit code for one insn
9415 * emit code to clear PSTATE.SS
9416 * emit code to generate software step exception for completed step
9417 * end TB (as usual for having generated an exception)
9418 * SS_ACTIVE == 1, PSTATE.SS == 0: (active-pending)
9419 * emit code to generate a software step exception
9420 * end the TB
9421 */
9422 dc->ss_active = EX_TBFLAG_ANY(tb_flags, SS_ACTIVE);
9423 dc->pstate_ss = EX_TBFLAG_ANY(tb_flags, PSTATE__SS);
9424 dc->is_ldex = false;
9425
9426 dc->page_start = dc->base.pc_first & TARGET_PAGE_MASK;
9427
9428 /* If architectural single step active, limit to 1. */
9429 if (dc->ss_active) {
9430 dc->base.max_insns = 1;
9431 }
9432
9433 /* ARM is a fixed-length ISA. Bound the number of insns to execute
9434 to those left on the page. */
9435 if (!dc->thumb) {
9436 int bound = -(dc->base.pc_first | TARGET_PAGE_MASK) / 4;
9437 dc->base.max_insns = MIN(dc->base.max_insns, bound);
9438 }
9439
9440 cpu_V0 = tcg_temp_new_i64();
9441 cpu_V1 = tcg_temp_new_i64();
9442 cpu_M0 = tcg_temp_new_i64();
9443 }
9444
9445 static void arm_tr_tb_start(DisasContextBase *dcbase, CPUState *cpu)
9446 {
9447 DisasContext *dc = container_of(dcbase, DisasContext, base);
9448
9449 /* A note on handling of the condexec (IT) bits:
9450 *
9451 * We want to avoid the overhead of having to write the updated condexec
9452 * bits back to the CPUARMState for every instruction in an IT block. So:
9453 * (1) if the condexec bits are not already zero then we write
9454 * zero back into the CPUARMState now. This avoids complications trying
9455 * to do it at the end of the block. (For example if we don't do this
9456 * it's hard to identify whether we can safely skip writing condexec
9457 * at the end of the TB, which we definitely want to do for the case
9458 * where a TB doesn't do anything with the IT state at all.)
9459 * (2) if we are going to leave the TB then we call gen_set_condexec()
9460 * which will write the correct value into CPUARMState if zero is wrong.
9461 * This is done both for leaving the TB at the end, and for leaving
9462 * it because of an exception we know will happen, which is done in
9463 * gen_exception_insn(). The latter is necessary because we need to
9464 * leave the TB with the PC/IT state just prior to execution of the
9465 * instruction which caused the exception.
9466 * (3) if we leave the TB unexpectedly (eg a data abort on a load)
9467 * then the CPUARMState will be wrong and we need to reset it.
9468 * This is handled in the same way as restoration of the
9469 * PC in these situations; we save the value of the condexec bits
9470 * for each PC via tcg_gen_insn_start(), and restore_state_to_opc()
9471 * then uses this to restore them after an exception.
9472 *
9473 * Note that there are no instructions which can read the condexec
9474 * bits, and none which can write non-static values to them, so
9475 * we don't need to care about whether CPUARMState is correct in the
9476 * middle of a TB.
9477 */
9478
9479 /* Reset the conditional execution bits immediately. This avoids
9480 complications trying to do it at the end of the block. */
9481 if (dc->condexec_mask || dc->condexec_cond) {
9482 store_cpu_field_constant(0, condexec_bits);
9483 }
9484 }
9485
9486 static void arm_tr_insn_start(DisasContextBase *dcbase, CPUState *cpu)
9487 {
9488 DisasContext *dc = container_of(dcbase, DisasContext, base);
9489 /*
9490 * The ECI/ICI bits share PSR bits with the IT bits, so we
9491 * need to reconstitute the bits from the split-out DisasContext
9492 * fields here.
9493 */
9494 uint32_t condexec_bits;
9495
9496 if (dc->eci) {
9497 condexec_bits = dc->eci << 4;
9498 } else {
9499 condexec_bits = (dc->condexec_cond << 4) | (dc->condexec_mask >> 1);
9500 }
9501 tcg_gen_insn_start(dc->base.pc_next, condexec_bits, 0);
9502 dc->insn_start = tcg_last_op();
9503 }
9504
9505 static bool arm_check_kernelpage(DisasContext *dc)
9506 {
9507 #ifdef CONFIG_USER_ONLY
9508 /* Intercept jump to the magic kernel page. */
9509 if (dc->base.pc_next >= 0xffff0000) {
9510 /* We always get here via a jump, so know we are not in a
9511 conditional execution block. */
9512 gen_exception_internal(EXCP_KERNEL_TRAP);
9513 dc->base.is_jmp = DISAS_NORETURN;
9514 return true;
9515 }
9516 #endif
9517 return false;
9518 }
9519
9520 static bool arm_check_ss_active(DisasContext *dc)
9521 {
9522 if (dc->ss_active && !dc->pstate_ss) {
9523 /* Singlestep state is Active-pending.
9524 * If we're in this state at the start of a TB then either
9525 * a) we just took an exception to an EL which is being debugged
9526 * and this is the first insn in the exception handler
9527 * b) debug exceptions were masked and we just unmasked them
9528 * without changing EL (eg by clearing PSTATE.D)
9529 * In either case we're going to take a swstep exception in the
9530 * "did not step an insn" case, and so the syndrome ISV and EX
9531 * bits should be zero.
9532 */
9533 assert(dc->base.num_insns == 1);
9534 gen_swstep_exception(dc, 0, 0);
9535 dc->base.is_jmp = DISAS_NORETURN;
9536 return true;
9537 }
9538
9539 return false;
9540 }
9541
9542 static void arm_post_translate_insn(DisasContext *dc)
9543 {
9544 if (dc->condjmp && !dc->base.is_jmp) {
9545 gen_set_label(dc->condlabel);
9546 dc->condjmp = 0;
9547 }
9548 translator_loop_temp_check(&dc->base);
9549 }
9550
9551 static void arm_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
9552 {
9553 DisasContext *dc = container_of(dcbase, DisasContext, base);
9554 CPUARMState *env = cpu->env_ptr;
9555 uint32_t pc = dc->base.pc_next;
9556 unsigned int insn;
9557
9558 /* Singlestep exceptions have the highest priority. */
9559 if (arm_check_ss_active(dc)) {
9560 dc->base.pc_next = pc + 4;
9561 return;
9562 }
9563
9564 if (pc & 3) {
9565 /*
9566 * PC alignment fault. This has priority over the instruction abort
9567 * that we would receive from a translation fault via arm_ldl_code
9568 * (or the execution of the kernelpage entrypoint). This should only
9569 * be possible after an indirect branch, at the start of the TB.
9570 */
9571 assert(dc->base.num_insns == 1);
9572 gen_helper_exception_pc_alignment(cpu_env, tcg_constant_tl(pc));
9573 dc->base.is_jmp = DISAS_NORETURN;
9574 dc->base.pc_next = QEMU_ALIGN_UP(pc, 4);
9575 return;
9576 }
9577
9578 if (arm_check_kernelpage(dc)) {
9579 dc->base.pc_next = pc + 4;
9580 return;
9581 }
9582
9583 dc->pc_curr = pc;
9584 insn = arm_ldl_code(env, &dc->base, pc, dc->sctlr_b);
9585 dc->insn = insn;
9586 dc->base.pc_next = pc + 4;
9587 disas_arm_insn(dc, insn);
9588
9589 arm_post_translate_insn(dc);
9590
9591 /* ARM is a fixed-length ISA. We performed the cross-page check
9592 in init_disas_context by adjusting max_insns. */
9593 }
9594
9595 static bool thumb_insn_is_unconditional(DisasContext *s, uint32_t insn)
9596 {
9597 /* Return true if this Thumb insn is always unconditional,
9598 * even inside an IT block. This is true of only a very few
9599 * instructions: BKPT, HLT, and SG.
9600 *
9601 * A larger class of instructions are UNPREDICTABLE if used
9602 * inside an IT block; we do not need to detect those here, because
9603 * what we do by default (perform the cc check and update the IT
9604 * bits state machine) is a permitted CONSTRAINED UNPREDICTABLE
9605 * choice for those situations.
9606 *
9607 * insn is either a 16-bit or a 32-bit instruction; the two are
9608 * distinguishable because for the 16-bit case the top 16 bits
9609 * are zeroes, and that isn't a valid 32-bit encoding.
9610 */
9611 if ((insn & 0xffffff00) == 0xbe00) {
9612 /* BKPT */
9613 return true;
9614 }
9615
9616 if ((insn & 0xffffffc0) == 0xba80 && arm_dc_feature(s, ARM_FEATURE_V8) &&
9617 !arm_dc_feature(s, ARM_FEATURE_M)) {
9618 /* HLT: v8A only. This is unconditional even when it is going to
9619 * UNDEF; see the v8A ARM ARM DDI0487B.a H3.3.
9620 * For v7 cores this was a plain old undefined encoding and so
9621 * honours its cc check. (We might be using the encoding as
9622 * a semihosting trap, but we don't change the cc check behaviour
9623 * on that account, because a debugger connected to a real v7A
9624 * core and emulating semihosting traps by catching the UNDEF
9625 * exception would also only see cases where the cc check passed.
9626 * No guest code should be trying to do a HLT semihosting trap
9627 * in an IT block anyway.
9628 */
9629 return true;
9630 }
9631
9632 if (insn == 0xe97fe97f && arm_dc_feature(s, ARM_FEATURE_V8) &&
9633 arm_dc_feature(s, ARM_FEATURE_M)) {
9634 /* SG: v8M only */
9635 return true;
9636 }
9637
9638 return false;
9639 }
9640
9641 static void thumb_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
9642 {
9643 DisasContext *dc = container_of(dcbase, DisasContext, base);
9644 CPUARMState *env = cpu->env_ptr;
9645 uint32_t pc = dc->base.pc_next;
9646 uint32_t insn;
9647 bool is_16bit;
9648
9649 /* Misaligned thumb PC is architecturally impossible. */
9650 assert((dc->base.pc_next & 1) == 0);
9651
9652 if (arm_check_ss_active(dc) || arm_check_kernelpage(dc)) {
9653 dc->base.pc_next = pc + 2;
9654 return;
9655 }
9656
9657 dc->pc_curr = pc;
9658 insn = arm_lduw_code(env, &dc->base, pc, dc->sctlr_b);
9659 is_16bit = thumb_insn_is_16bit(dc, dc->base.pc_next, insn);
9660 pc += 2;
9661 if (!is_16bit) {
9662 uint32_t insn2 = arm_lduw_code(env, &dc->base, pc, dc->sctlr_b);
9663 insn = insn << 16 | insn2;
9664 pc += 2;
9665 }
9666 dc->base.pc_next = pc;
9667 dc->insn = insn;
9668
9669 if (dc->pstate_il) {
9670 /*
9671 * Illegal execution state. This has priority over BTI
9672 * exceptions, but comes after instruction abort exceptions.
9673 */
9674 gen_exception_insn(dc, dc->pc_curr, EXCP_UDEF,
9675 syn_illegalstate(), default_exception_el(dc));
9676 return;
9677 }
9678
9679 if (dc->eci) {
9680 /*
9681 * For M-profile continuable instructions, ECI/ICI handling
9682 * falls into these cases:
9683 * - interrupt-continuable instructions
9684 * These are the various load/store multiple insns (both
9685 * integer and fp). The ICI bits indicate the register
9686 * where the load/store can resume. We make the IMPDEF
9687 * choice to always do "instruction restart", ie ignore
9688 * the ICI value and always execute the ldm/stm from the
9689 * start. So all we need to do is zero PSR.ICI if the
9690 * insn executes.
9691 * - MVE instructions subject to beat-wise execution
9692 * Here the ECI bits indicate which beats have already been
9693 * executed, and we must honour this. Each insn of this
9694 * type will handle it correctly. We will update PSR.ECI
9695 * in the helper function for the insn (some ECI values
9696 * mean that the following insn also has been partially
9697 * executed).
9698 * - Special cases which don't advance ECI
9699 * The insns LE, LETP and BKPT leave the ECI/ICI state
9700 * bits untouched.
9701 * - all other insns (the common case)
9702 * Non-zero ECI/ICI means an INVSTATE UsageFault.
9703 * We place a rewind-marker here. Insns in the previous
9704 * three categories will set a flag in the DisasContext.
9705 * If the flag isn't set after we call disas_thumb_insn()
9706 * or disas_thumb2_insn() then we know we have a "some other
9707 * insn" case. We will rewind to the marker (ie throwing away
9708 * all the generated code) and instead emit "take exception".
9709 */
9710 dc->insn_eci_rewind = tcg_last_op();
9711 }
9712
9713 if (dc->condexec_mask && !thumb_insn_is_unconditional(dc, insn)) {
9714 uint32_t cond = dc->condexec_cond;
9715
9716 /*
9717 * Conditionally skip the insn. Note that both 0xe and 0xf mean
9718 * "always"; 0xf is not "never".
9719 */
9720 if (cond < 0x0e) {
9721 arm_skip_unless(dc, cond);
9722 }
9723 }
9724
9725 if (is_16bit) {
9726 disas_thumb_insn(dc, insn);
9727 } else {
9728 disas_thumb2_insn(dc, insn);
9729 }
9730
9731 /* Advance the Thumb condexec condition. */
9732 if (dc->condexec_mask) {
9733 dc->condexec_cond = ((dc->condexec_cond & 0xe) |
9734 ((dc->condexec_mask >> 4) & 1));
9735 dc->condexec_mask = (dc->condexec_mask << 1) & 0x1f;
9736 if (dc->condexec_mask == 0) {
9737 dc->condexec_cond = 0;
9738 }
9739 }
9740
9741 if (dc->eci && !dc->eci_handled) {
9742 /*
9743 * Insn wasn't valid for ECI/ICI at all: undo what we
9744 * just generated and instead emit an exception
9745 */
9746 tcg_remove_ops_after(dc->insn_eci_rewind);
9747 dc->condjmp = 0;
9748 gen_exception_insn(dc, dc->pc_curr, EXCP_INVSTATE, syn_uncategorized(),
9749 default_exception_el(dc));
9750 }
9751
9752 arm_post_translate_insn(dc);
9753
9754 /* Thumb is a variable-length ISA. Stop translation when the next insn
9755 * will touch a new page. This ensures that prefetch aborts occur at
9756 * the right place.
9757 *
9758 * We want to stop the TB if the next insn starts in a new page,
9759 * or if it spans between this page and the next. This means that
9760 * if we're looking at the last halfword in the page we need to
9761 * see if it's a 16-bit Thumb insn (which will fit in this TB)
9762 * or a 32-bit Thumb insn (which won't).
9763 * This is to avoid generating a silly TB with a single 16-bit insn
9764 * in it at the end of this page (which would execute correctly
9765 * but isn't very efficient).
9766 */
9767 if (dc->base.is_jmp == DISAS_NEXT
9768 && (dc->base.pc_next - dc->page_start >= TARGET_PAGE_SIZE
9769 || (dc->base.pc_next - dc->page_start >= TARGET_PAGE_SIZE - 3
9770 && insn_crosses_page(env, dc)))) {
9771 dc->base.is_jmp = DISAS_TOO_MANY;
9772 }
9773 }
9774
9775 static void arm_tr_tb_stop(DisasContextBase *dcbase, CPUState *cpu)
9776 {
9777 DisasContext *dc = container_of(dcbase, DisasContext, base);
9778
9779 /* At this stage dc->condjmp will only be set when the skipped
9780 instruction was a conditional branch or trap, and the PC has
9781 already been written. */
9782 gen_set_condexec(dc);
9783 if (dc->base.is_jmp == DISAS_BX_EXCRET) {
9784 /* Exception return branches need some special case code at the
9785 * end of the TB, which is complex enough that it has to
9786 * handle the single-step vs not and the condition-failed
9787 * insn codepath itself.
9788 */
9789 gen_bx_excret_final_code(dc);
9790 } else if (unlikely(dc->ss_active)) {
9791 /* Unconditional and "condition passed" instruction codepath. */
9792 switch (dc->base.is_jmp) {
9793 case DISAS_SWI:
9794 gen_ss_advance(dc);
9795 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
9796 default_exception_el(dc));
9797 break;
9798 case DISAS_HVC:
9799 gen_ss_advance(dc);
9800 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
9801 break;
9802 case DISAS_SMC:
9803 gen_ss_advance(dc);
9804 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
9805 break;
9806 case DISAS_NEXT:
9807 case DISAS_TOO_MANY:
9808 case DISAS_UPDATE_EXIT:
9809 case DISAS_UPDATE_NOCHAIN:
9810 gen_set_pc_im(dc, dc->base.pc_next);
9811 /* fall through */
9812 default:
9813 /* FIXME: Single stepping a WFI insn will not halt the CPU. */
9814 gen_singlestep_exception(dc);
9815 break;
9816 case DISAS_NORETURN:
9817 break;
9818 }
9819 } else {
9820 /* While branches must always occur at the end of an IT block,
9821 there are a few other things that can cause us to terminate
9822 the TB in the middle of an IT block:
9823 - Exception generating instructions (bkpt, swi, undefined).
9824 - Page boundaries.
9825 - Hardware watchpoints.
9826 Hardware breakpoints have already been handled and skip this code.
9827 */
9828 switch (dc->base.is_jmp) {
9829 case DISAS_NEXT:
9830 case DISAS_TOO_MANY:
9831 gen_goto_tb(dc, 1, dc->base.pc_next);
9832 break;
9833 case DISAS_UPDATE_NOCHAIN:
9834 gen_set_pc_im(dc, dc->base.pc_next);
9835 /* fall through */
9836 case DISAS_JUMP:
9837 gen_goto_ptr();
9838 break;
9839 case DISAS_UPDATE_EXIT:
9840 gen_set_pc_im(dc, dc->base.pc_next);
9841 /* fall through */
9842 default:
9843 /* indicate that the hash table must be used to find the next TB */
9844 tcg_gen_exit_tb(NULL, 0);
9845 break;
9846 case DISAS_NORETURN:
9847 /* nothing more to generate */
9848 break;
9849 case DISAS_WFI:
9850 {
9851 TCGv_i32 tmp = tcg_const_i32((dc->thumb &&
9852 !(dc->insn & (1U << 31))) ? 2 : 4);
9853
9854 gen_helper_wfi(cpu_env, tmp);
9855 tcg_temp_free_i32(tmp);
9856 /* The helper doesn't necessarily throw an exception, but we
9857 * must go back to the main loop to check for interrupts anyway.
9858 */
9859 tcg_gen_exit_tb(NULL, 0);
9860 break;
9861 }
9862 case DISAS_WFE:
9863 gen_helper_wfe(cpu_env);
9864 break;
9865 case DISAS_YIELD:
9866 gen_helper_yield(cpu_env);
9867 break;
9868 case DISAS_SWI:
9869 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
9870 default_exception_el(dc));
9871 break;
9872 case DISAS_HVC:
9873 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
9874 break;
9875 case DISAS_SMC:
9876 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
9877 break;
9878 }
9879 }
9880
9881 if (dc->condjmp) {
9882 /* "Condition failed" instruction codepath for the branch/trap insn */
9883 gen_set_label(dc->condlabel);
9884 gen_set_condexec(dc);
9885 if (unlikely(dc->ss_active)) {
9886 gen_set_pc_im(dc, dc->base.pc_next);
9887 gen_singlestep_exception(dc);
9888 } else {
9889 gen_goto_tb(dc, 1, dc->base.pc_next);
9890 }
9891 }
9892 }
9893
9894 static void arm_tr_disas_log(const DisasContextBase *dcbase, CPUState *cpu)
9895 {
9896 DisasContext *dc = container_of(dcbase, DisasContext, base);
9897
9898 qemu_log("IN: %s\n", lookup_symbol(dc->base.pc_first));
9899 log_target_disas(cpu, dc->base.pc_first, dc->base.tb->size);
9900 }
9901
9902 static const TranslatorOps arm_translator_ops = {
9903 .init_disas_context = arm_tr_init_disas_context,
9904 .tb_start = arm_tr_tb_start,
9905 .insn_start = arm_tr_insn_start,
9906 .translate_insn = arm_tr_translate_insn,
9907 .tb_stop = arm_tr_tb_stop,
9908 .disas_log = arm_tr_disas_log,
9909 };
9910
9911 static const TranslatorOps thumb_translator_ops = {
9912 .init_disas_context = arm_tr_init_disas_context,
9913 .tb_start = arm_tr_tb_start,
9914 .insn_start = arm_tr_insn_start,
9915 .translate_insn = thumb_tr_translate_insn,
9916 .tb_stop = arm_tr_tb_stop,
9917 .disas_log = arm_tr_disas_log,
9918 };
9919
9920 /* generate intermediate code for basic block 'tb'. */
9921 void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int max_insns)
9922 {
9923 DisasContext dc = { };
9924 const TranslatorOps *ops = &arm_translator_ops;
9925 CPUARMTBFlags tb_flags = arm_tbflags_from_tb(tb);
9926
9927 if (EX_TBFLAG_AM32(tb_flags, THUMB)) {
9928 ops = &thumb_translator_ops;
9929 }
9930 #ifdef TARGET_AARCH64
9931 if (EX_TBFLAG_ANY(tb_flags, AARCH64_STATE)) {
9932 ops = &aarch64_translator_ops;
9933 }
9934 #endif
9935
9936 translator_loop(ops, &dc.base, cpu, tb, max_insns);
9937 }
9938
9939 void restore_state_to_opc(CPUARMState *env, TranslationBlock *tb,
9940 target_ulong *data)
9941 {
9942 if (is_a64(env)) {
9943 env->pc = data[0];
9944 env->condexec_bits = 0;
9945 env->exception.syndrome = data[2] << ARM_INSN_START_WORD2_SHIFT;
9946 } else {
9947 env->regs[15] = data[0];
9948 env->condexec_bits = data[1];
9949 env->exception.syndrome = data[2] << ARM_INSN_START_WORD2_SHIFT;
9950 }
9951 }
Ray Wang's QEMU Repository