xref: /openbmc/qemu/target/arm/tcg/translate.c (revision 7d87775f)
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 "translate.h"
24 #include "translate-a32.h"
25 #include "qemu/log.h"
26 #include "arm_ldst.h"
27 #include "semihosting/semihost.h"
28 #include "cpregs.h"
29 #include "exec/helper-proto.h"
30 
31 #define HELPER_H "helper.h"
32 #include "exec/helper-info.c.inc"
33 #undef  HELPER_H
34 
35 #define ENABLE_ARCH_4T    arm_dc_feature(s, ARM_FEATURE_V4T)
36 #define ENABLE_ARCH_5     arm_dc_feature(s, ARM_FEATURE_V5)
37 /* currently all emulated v5 cores are also v5TE, so don't bother */
38 #define ENABLE_ARCH_5TE   arm_dc_feature(s, ARM_FEATURE_V5)
39 #define ENABLE_ARCH_5J    dc_isar_feature(aa32_jazelle, s)
40 #define ENABLE_ARCH_6     arm_dc_feature(s, ARM_FEATURE_V6)
41 #define ENABLE_ARCH_6K    arm_dc_feature(s, ARM_FEATURE_V6K)
42 #define ENABLE_ARCH_6T2   arm_dc_feature(s, ARM_FEATURE_THUMB2)
43 #define ENABLE_ARCH_7     arm_dc_feature(s, ARM_FEATURE_V7)
44 #define ENABLE_ARCH_8     arm_dc_feature(s, ARM_FEATURE_V8)
45 
46 /* These are TCG temporaries used only by the legacy iwMMXt decoder */
47 static TCGv_i64 cpu_V0, cpu_V1, cpu_M0;
48 /* These are TCG globals which alias CPUARMState fields */
49 static TCGv_i32 cpu_R[16];
50 TCGv_i32 cpu_CF, cpu_NF, cpu_VF, cpu_ZF;
51 TCGv_i64 cpu_exclusive_addr;
52 TCGv_i64 cpu_exclusive_val;
53 
54 static const char * const regnames[] =
55     { "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7",
56       "r8", "r9", "r10", "r11", "r12", "r13", "r14", "pc" };
57 
58 
59 /* initialize TCG globals.  */
60 void arm_translate_init(void)
61 {
62     int i;
63 
64     for (i = 0; i < 16; i++) {
65         cpu_R[i] = tcg_global_mem_new_i32(tcg_env,
66                                           offsetof(CPUARMState, regs[i]),
67                                           regnames[i]);
68     }
69     cpu_CF = tcg_global_mem_new_i32(tcg_env, offsetof(CPUARMState, CF), "CF");
70     cpu_NF = tcg_global_mem_new_i32(tcg_env, offsetof(CPUARMState, NF), "NF");
71     cpu_VF = tcg_global_mem_new_i32(tcg_env, offsetof(CPUARMState, VF), "VF");
72     cpu_ZF = tcg_global_mem_new_i32(tcg_env, offsetof(CPUARMState, ZF), "ZF");
73 
74     cpu_exclusive_addr = tcg_global_mem_new_i64(tcg_env,
75         offsetof(CPUARMState, exclusive_addr), "exclusive_addr");
76     cpu_exclusive_val = tcg_global_mem_new_i64(tcg_env,
77         offsetof(CPUARMState, exclusive_val), "exclusive_val");
78 
79     a64_translate_init();
80 }
81 
82 uint64_t asimd_imm_const(uint32_t imm, int cmode, int op)
83 {
84     /* Expand the encoded constant as per AdvSIMDExpandImm pseudocode */
85     switch (cmode) {
86     case 0: case 1:
87         /* no-op */
88         break;
89     case 2: case 3:
90         imm <<= 8;
91         break;
92     case 4: case 5:
93         imm <<= 16;
94         break;
95     case 6: case 7:
96         imm <<= 24;
97         break;
98     case 8: case 9:
99         imm |= imm << 16;
100         break;
101     case 10: case 11:
102         imm = (imm << 8) | (imm << 24);
103         break;
104     case 12:
105         imm = (imm << 8) | 0xff;
106         break;
107     case 13:
108         imm = (imm << 16) | 0xffff;
109         break;
110     case 14:
111         if (op) {
112             /*
113              * This and cmode == 15 op == 1 are the only cases where
114              * the top and bottom 32 bits of the encoded constant differ.
115              */
116             uint64_t imm64 = 0;
117             int n;
118 
119             for (n = 0; n < 8; n++) {
120                 if (imm & (1 << n)) {
121                     imm64 |= (0xffULL << (n * 8));
122                 }
123             }
124             return imm64;
125         }
126         imm |= (imm << 8) | (imm << 16) | (imm << 24);
127         break;
128     case 15:
129         if (op) {
130             /* Reserved encoding for AArch32; valid for AArch64 */
131             uint64_t imm64 = (uint64_t)(imm & 0x3f) << 48;
132             if (imm & 0x80) {
133                 imm64 |= 0x8000000000000000ULL;
134             }
135             if (imm & 0x40) {
136                 imm64 |= 0x3fc0000000000000ULL;
137             } else {
138                 imm64 |= 0x4000000000000000ULL;
139             }
140             return imm64;
141         }
142         imm = ((imm & 0x80) << 24) | ((imm & 0x3f) << 19)
143             | ((imm & 0x40) ? (0x1f << 25) : (1 << 30));
144         break;
145     }
146     if (op) {
147         imm = ~imm;
148     }
149     return dup_const(MO_32, imm);
150 }
151 
152 /* Generate a label used for skipping this instruction */
153 void arm_gen_condlabel(DisasContext *s)
154 {
155     if (!s->condjmp) {
156         s->condlabel = gen_disas_label(s);
157         s->condjmp = 1;
158     }
159 }
160 
161 /* Flags for the disas_set_da_iss info argument:
162  * lower bits hold the Rt register number, higher bits are flags.
163  */
164 typedef enum ISSInfo {
165     ISSNone = 0,
166     ISSRegMask = 0x1f,
167     ISSInvalid = (1 << 5),
168     ISSIsAcqRel = (1 << 6),
169     ISSIsWrite = (1 << 7),
170     ISSIs16Bit = (1 << 8),
171 } ISSInfo;
172 
173 /*
174  * Store var into env + offset to a member with size bytes.
175  * Free var after use.
176  */
177 void store_cpu_offset(TCGv_i32 var, int offset, int size)
178 {
179     switch (size) {
180     case 1:
181         tcg_gen_st8_i32(var, tcg_env, offset);
182         break;
183     case 4:
184         tcg_gen_st_i32(var, tcg_env, offset);
185         break;
186     default:
187         g_assert_not_reached();
188     }
189 }
190 
191 /* Save the syndrome information for a Data Abort */
192 static void disas_set_da_iss(DisasContext *s, MemOp memop, ISSInfo issinfo)
193 {
194     uint32_t syn;
195     int sas = memop & MO_SIZE;
196     bool sse = memop & MO_SIGN;
197     bool is_acqrel = issinfo & ISSIsAcqRel;
198     bool is_write = issinfo & ISSIsWrite;
199     bool is_16bit = issinfo & ISSIs16Bit;
200     int srt = issinfo & ISSRegMask;
201 
202     if (issinfo & ISSInvalid) {
203         /* Some callsites want to conditionally provide ISS info,
204          * eg "only if this was not a writeback"
205          */
206         return;
207     }
208 
209     if (srt == 15) {
210         /* For AArch32, insns where the src/dest is R15 never generate
211          * ISS information. Catching that here saves checking at all
212          * the call sites.
213          */
214         return;
215     }
216 
217     syn = syn_data_abort_with_iss(0, sas, sse, srt, 0, is_acqrel,
218                                   0, 0, 0, is_write, 0, is_16bit);
219     disas_set_insn_syndrome(s, syn);
220 }
221 
222 static inline int get_a32_user_mem_index(DisasContext *s)
223 {
224     /* Return the core mmu_idx to use for A32/T32 "unprivileged load/store"
225      * insns:
226      *  if PL2, UNPREDICTABLE (we choose to implement as if PL0)
227      *  otherwise, access as if at PL0.
228      */
229     switch (s->mmu_idx) {
230     case ARMMMUIdx_E3:
231     case ARMMMUIdx_E30_0:
232     case ARMMMUIdx_E30_3_PAN:
233         return arm_to_core_mmu_idx(ARMMMUIdx_E30_0);
234     case ARMMMUIdx_E2:        /* this one is UNPREDICTABLE */
235     case ARMMMUIdx_E10_0:
236     case ARMMMUIdx_E10_1:
237     case ARMMMUIdx_E10_1_PAN:
238         return arm_to_core_mmu_idx(ARMMMUIdx_E10_0);
239     case ARMMMUIdx_MUser:
240     case ARMMMUIdx_MPriv:
241         return arm_to_core_mmu_idx(ARMMMUIdx_MUser);
242     case ARMMMUIdx_MUserNegPri:
243     case ARMMMUIdx_MPrivNegPri:
244         return arm_to_core_mmu_idx(ARMMMUIdx_MUserNegPri);
245     case ARMMMUIdx_MSUser:
246     case ARMMMUIdx_MSPriv:
247         return arm_to_core_mmu_idx(ARMMMUIdx_MSUser);
248     case ARMMMUIdx_MSUserNegPri:
249     case ARMMMUIdx_MSPrivNegPri:
250         return arm_to_core_mmu_idx(ARMMMUIdx_MSUserNegPri);
251     default:
252         g_assert_not_reached();
253     }
254 }
255 
256 /* The pc_curr difference for an architectural jump. */
257 static target_long jmp_diff(DisasContext *s, target_long diff)
258 {
259     return diff + (s->thumb ? 4 : 8);
260 }
261 
262 static void gen_pc_plus_diff(DisasContext *s, TCGv_i32 var, target_long diff)
263 {
264     assert(s->pc_save != -1);
265     if (tb_cflags(s->base.tb) & CF_PCREL) {
266         tcg_gen_addi_i32(var, cpu_R[15], (s->pc_curr - s->pc_save) + diff);
267     } else {
268         tcg_gen_movi_i32(var, s->pc_curr + diff);
269     }
270 }
271 
272 /* Set a variable to the value of a CPU register.  */
273 void load_reg_var(DisasContext *s, TCGv_i32 var, int reg)
274 {
275     if (reg == 15) {
276         gen_pc_plus_diff(s, var, jmp_diff(s, 0));
277     } else {
278         tcg_gen_mov_i32(var, cpu_R[reg]);
279     }
280 }
281 
282 /*
283  * Create a new temp, REG + OFS, except PC is ALIGN(PC, 4).
284  * This is used for load/store for which use of PC implies (literal),
285  * or ADD that implies ADR.
286  */
287 TCGv_i32 add_reg_for_lit(DisasContext *s, int reg, int ofs)
288 {
289     TCGv_i32 tmp = tcg_temp_new_i32();
290 
291     if (reg == 15) {
292         /*
293          * This address is computed from an aligned PC:
294          * subtract off the low bits.
295          */
296         gen_pc_plus_diff(s, tmp, jmp_diff(s, ofs - (s->pc_curr & 3)));
297     } else {
298         tcg_gen_addi_i32(tmp, cpu_R[reg], ofs);
299     }
300     return tmp;
301 }
302 
303 /* Set a CPU register.  The source must be a temporary and will be
304    marked as dead.  */
305 void store_reg(DisasContext *s, int reg, TCGv_i32 var)
306 {
307     if (reg == 15) {
308         /* In Thumb mode, we must ignore bit 0.
309          * In ARM mode, for ARMv4 and ARMv5, it is UNPREDICTABLE if bits [1:0]
310          * are not 0b00, but for ARMv6 and above, we must ignore bits [1:0].
311          * We choose to ignore [1:0] in ARM mode for all architecture versions.
312          */
313         tcg_gen_andi_i32(var, var, s->thumb ? ~1 : ~3);
314         s->base.is_jmp = DISAS_JUMP;
315         s->pc_save = -1;
316     } else if (reg == 13 && arm_dc_feature(s, ARM_FEATURE_M)) {
317         /* For M-profile SP bits [1:0] are always zero */
318         tcg_gen_andi_i32(var, var, ~3);
319     }
320     tcg_gen_mov_i32(cpu_R[reg], var);
321 }
322 
323 /*
324  * Variant of store_reg which applies v8M stack-limit checks before updating
325  * SP. If the check fails this will result in an exception being taken.
326  * We disable the stack checks for CONFIG_USER_ONLY because we have
327  * no idea what the stack limits should be in that case.
328  * If stack checking is not being done this just acts like store_reg().
329  */
330 static void store_sp_checked(DisasContext *s, TCGv_i32 var)
331 {
332 #ifndef CONFIG_USER_ONLY
333     if (s->v8m_stackcheck) {
334         gen_helper_v8m_stackcheck(tcg_env, var);
335     }
336 #endif
337     store_reg(s, 13, var);
338 }
339 
340 /* Value extensions.  */
341 #define gen_uxtb(var) tcg_gen_ext8u_i32(var, var)
342 #define gen_uxth(var) tcg_gen_ext16u_i32(var, var)
343 #define gen_sxtb(var) tcg_gen_ext8s_i32(var, var)
344 #define gen_sxth(var) tcg_gen_ext16s_i32(var, var)
345 
346 #define gen_sxtb16(var) gen_helper_sxtb16(var, var)
347 #define gen_uxtb16(var) gen_helper_uxtb16(var, var)
348 
349 void gen_set_cpsr(TCGv_i32 var, uint32_t mask)
350 {
351     gen_helper_cpsr_write(tcg_env, var, tcg_constant_i32(mask));
352 }
353 
354 static void gen_rebuild_hflags(DisasContext *s, bool new_el)
355 {
356     bool m_profile = arm_dc_feature(s, ARM_FEATURE_M);
357 
358     if (new_el) {
359         if (m_profile) {
360             gen_helper_rebuild_hflags_m32_newel(tcg_env);
361         } else {
362             gen_helper_rebuild_hflags_a32_newel(tcg_env);
363         }
364     } else {
365         TCGv_i32 tcg_el = tcg_constant_i32(s->current_el);
366         if (m_profile) {
367             gen_helper_rebuild_hflags_m32(tcg_env, tcg_el);
368         } else {
369             gen_helper_rebuild_hflags_a32(tcg_env, tcg_el);
370         }
371     }
372 }
373 
374 static void gen_exception_internal(int excp)
375 {
376     assert(excp_is_internal(excp));
377     gen_helper_exception_internal(tcg_env, tcg_constant_i32(excp));
378 }
379 
380 static void gen_singlestep_exception(DisasContext *s)
381 {
382     /* We just completed step of an insn. Move from Active-not-pending
383      * to Active-pending, and then also take the swstep exception.
384      * This corresponds to making the (IMPDEF) choice to prioritize
385      * swstep exceptions over asynchronous exceptions taken to an exception
386      * level where debug is disabled. This choice has the advantage that
387      * we do not need to maintain internal state corresponding to the
388      * ISV/EX syndrome bits between completion of the step and generation
389      * of the exception, and our syndrome information is always correct.
390      */
391     gen_ss_advance(s);
392     gen_swstep_exception(s, 1, s->is_ldex);
393     s->base.is_jmp = DISAS_NORETURN;
394 }
395 
396 void clear_eci_state(DisasContext *s)
397 {
398     /*
399      * Clear any ECI/ICI state: used when a load multiple/store
400      * multiple insn executes.
401      */
402     if (s->eci) {
403         store_cpu_field_constant(0, condexec_bits);
404         s->eci = 0;
405     }
406 }
407 
408 static void gen_smul_dual(TCGv_i32 a, TCGv_i32 b)
409 {
410     TCGv_i32 tmp1 = tcg_temp_new_i32();
411     TCGv_i32 tmp2 = tcg_temp_new_i32();
412     tcg_gen_ext16s_i32(tmp1, a);
413     tcg_gen_ext16s_i32(tmp2, b);
414     tcg_gen_mul_i32(tmp1, tmp1, tmp2);
415     tcg_gen_sari_i32(a, a, 16);
416     tcg_gen_sari_i32(b, b, 16);
417     tcg_gen_mul_i32(b, b, a);
418     tcg_gen_mov_i32(a, tmp1);
419 }
420 
421 /* Byteswap each halfword.  */
422 void gen_rev16(TCGv_i32 dest, TCGv_i32 var)
423 {
424     TCGv_i32 tmp = tcg_temp_new_i32();
425     TCGv_i32 mask = tcg_constant_i32(0x00ff00ff);
426     tcg_gen_shri_i32(tmp, var, 8);
427     tcg_gen_and_i32(tmp, tmp, mask);
428     tcg_gen_and_i32(var, var, mask);
429     tcg_gen_shli_i32(var, var, 8);
430     tcg_gen_or_i32(dest, var, tmp);
431 }
432 
433 /* Byteswap low halfword and sign extend.  */
434 static void gen_revsh(TCGv_i32 dest, TCGv_i32 var)
435 {
436     tcg_gen_bswap16_i32(var, var, TCG_BSWAP_OS);
437 }
438 
439 /* Dual 16-bit add.  Result placed in t0 and t1 is marked as dead.
440     tmp = (t0 ^ t1) & 0x8000;
441     t0 &= ~0x8000;
442     t1 &= ~0x8000;
443     t0 = (t0 + t1) ^ tmp;
444  */
445 
446 static void gen_add16(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
447 {
448     TCGv_i32 tmp = tcg_temp_new_i32();
449     tcg_gen_xor_i32(tmp, t0, t1);
450     tcg_gen_andi_i32(tmp, tmp, 0x8000);
451     tcg_gen_andi_i32(t0, t0, ~0x8000);
452     tcg_gen_andi_i32(t1, t1, ~0x8000);
453     tcg_gen_add_i32(t0, t0, t1);
454     tcg_gen_xor_i32(dest, t0, tmp);
455 }
456 
457 /* Set N and Z flags from var.  */
458 static inline void gen_logic_CC(TCGv_i32 var)
459 {
460     tcg_gen_mov_i32(cpu_NF, var);
461     tcg_gen_mov_i32(cpu_ZF, var);
462 }
463 
464 /* dest = T0 + T1 + CF. */
465 static void gen_add_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
466 {
467     tcg_gen_add_i32(dest, t0, t1);
468     tcg_gen_add_i32(dest, dest, cpu_CF);
469 }
470 
471 /* dest = T0 - T1 + CF - 1.  */
472 static void gen_sub_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
473 {
474     tcg_gen_sub_i32(dest, t0, t1);
475     tcg_gen_add_i32(dest, dest, cpu_CF);
476     tcg_gen_subi_i32(dest, dest, 1);
477 }
478 
479 /* dest = T0 + T1. Compute C, N, V and Z flags */
480 static void gen_add_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
481 {
482     TCGv_i32 tmp = tcg_temp_new_i32();
483     tcg_gen_movi_i32(tmp, 0);
484     tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, t1, tmp);
485     tcg_gen_mov_i32(cpu_ZF, cpu_NF);
486     tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
487     tcg_gen_xor_i32(tmp, t0, t1);
488     tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
489     tcg_gen_mov_i32(dest, cpu_NF);
490 }
491 
492 /* dest = T0 + T1 + CF.  Compute C, N, V and Z flags */
493 static void gen_adc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
494 {
495     TCGv_i32 tmp = tcg_temp_new_i32();
496     if (TCG_TARGET_HAS_add2_i32) {
497         tcg_gen_movi_i32(tmp, 0);
498         tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, cpu_CF, tmp);
499         tcg_gen_add2_i32(cpu_NF, cpu_CF, cpu_NF, cpu_CF, t1, tmp);
500     } else {
501         TCGv_i64 q0 = tcg_temp_new_i64();
502         TCGv_i64 q1 = tcg_temp_new_i64();
503         tcg_gen_extu_i32_i64(q0, t0);
504         tcg_gen_extu_i32_i64(q1, t1);
505         tcg_gen_add_i64(q0, q0, q1);
506         tcg_gen_extu_i32_i64(q1, cpu_CF);
507         tcg_gen_add_i64(q0, q0, q1);
508         tcg_gen_extr_i64_i32(cpu_NF, cpu_CF, q0);
509     }
510     tcg_gen_mov_i32(cpu_ZF, cpu_NF);
511     tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
512     tcg_gen_xor_i32(tmp, t0, t1);
513     tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
514     tcg_gen_mov_i32(dest, cpu_NF);
515 }
516 
517 /* dest = T0 - T1. Compute C, N, V and Z flags */
518 static void gen_sub_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
519 {
520     TCGv_i32 tmp;
521     tcg_gen_sub_i32(cpu_NF, t0, t1);
522     tcg_gen_mov_i32(cpu_ZF, cpu_NF);
523     tcg_gen_setcond_i32(TCG_COND_GEU, cpu_CF, t0, t1);
524     tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
525     tmp = tcg_temp_new_i32();
526     tcg_gen_xor_i32(tmp, t0, t1);
527     tcg_gen_and_i32(cpu_VF, cpu_VF, tmp);
528     tcg_gen_mov_i32(dest, cpu_NF);
529 }
530 
531 /* dest = T0 + ~T1 + CF.  Compute C, N, V and Z flags */
532 static void gen_sbc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
533 {
534     TCGv_i32 tmp = tcg_temp_new_i32();
535     tcg_gen_not_i32(tmp, t1);
536     gen_adc_CC(dest, t0, tmp);
537 }
538 
539 #define GEN_SHIFT(name)                                               \
540 static void gen_##name(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)       \
541 {                                                                     \
542     TCGv_i32 tmpd = tcg_temp_new_i32();                               \
543     TCGv_i32 tmp1 = tcg_temp_new_i32();                               \
544     TCGv_i32 zero = tcg_constant_i32(0);                              \
545     tcg_gen_andi_i32(tmp1, t1, 0x1f);                                 \
546     tcg_gen_##name##_i32(tmpd, t0, tmp1);                             \
547     tcg_gen_andi_i32(tmp1, t1, 0xe0);                                 \
548     tcg_gen_movcond_i32(TCG_COND_NE, dest, tmp1, zero, zero, tmpd);   \
549 }
550 GEN_SHIFT(shl)
551 GEN_SHIFT(shr)
552 #undef GEN_SHIFT
553 
554 static void gen_sar(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
555 {
556     TCGv_i32 tmp1 = tcg_temp_new_i32();
557 
558     tcg_gen_andi_i32(tmp1, t1, 0xff);
559     tcg_gen_umin_i32(tmp1, tmp1, tcg_constant_i32(31));
560     tcg_gen_sar_i32(dest, t0, tmp1);
561 }
562 
563 static void shifter_out_im(TCGv_i32 var, int shift)
564 {
565     tcg_gen_extract_i32(cpu_CF, var, shift, 1);
566 }
567 
568 /* Shift by immediate.  Includes special handling for shift == 0.  */
569 static inline void gen_arm_shift_im(TCGv_i32 var, int shiftop,
570                                     int shift, int flags)
571 {
572     switch (shiftop) {
573     case 0: /* LSL */
574         if (shift != 0) {
575             if (flags)
576                 shifter_out_im(var, 32 - shift);
577             tcg_gen_shli_i32(var, var, shift);
578         }
579         break;
580     case 1: /* LSR */
581         if (shift == 0) {
582             if (flags) {
583                 tcg_gen_shri_i32(cpu_CF, var, 31);
584             }
585             tcg_gen_movi_i32(var, 0);
586         } else {
587             if (flags)
588                 shifter_out_im(var, shift - 1);
589             tcg_gen_shri_i32(var, var, shift);
590         }
591         break;
592     case 2: /* ASR */
593         if (shift == 0)
594             shift = 32;
595         if (flags)
596             shifter_out_im(var, shift - 1);
597         if (shift == 32)
598           shift = 31;
599         tcg_gen_sari_i32(var, var, shift);
600         break;
601     case 3: /* ROR/RRX */
602         if (shift != 0) {
603             if (flags)
604                 shifter_out_im(var, shift - 1);
605             tcg_gen_rotri_i32(var, var, shift); break;
606         } else {
607             TCGv_i32 tmp = tcg_temp_new_i32();
608             tcg_gen_shli_i32(tmp, cpu_CF, 31);
609             if (flags)
610                 shifter_out_im(var, 0);
611             tcg_gen_shri_i32(var, var, 1);
612             tcg_gen_or_i32(var, var, tmp);
613         }
614     }
615 };
616 
617 static inline void gen_arm_shift_reg(TCGv_i32 var, int shiftop,
618                                      TCGv_i32 shift, int flags)
619 {
620     if (flags) {
621         switch (shiftop) {
622         case 0: gen_helper_shl_cc(var, tcg_env, var, shift); break;
623         case 1: gen_helper_shr_cc(var, tcg_env, var, shift); break;
624         case 2: gen_helper_sar_cc(var, tcg_env, var, shift); break;
625         case 3: gen_helper_ror_cc(var, tcg_env, var, shift); break;
626         }
627     } else {
628         switch (shiftop) {
629         case 0:
630             gen_shl(var, var, shift);
631             break;
632         case 1:
633             gen_shr(var, var, shift);
634             break;
635         case 2:
636             gen_sar(var, var, shift);
637             break;
638         case 3: tcg_gen_andi_i32(shift, shift, 0x1f);
639                 tcg_gen_rotr_i32(var, var, shift); break;
640         }
641     }
642 }
643 
644 /*
645  * Generate a conditional based on ARM condition code cc.
646  * This is common between ARM and Aarch64 targets.
647  */
648 void arm_test_cc(DisasCompare *cmp, int cc)
649 {
650     TCGv_i32 value;
651     TCGCond cond;
652 
653     switch (cc) {
654     case 0: /* eq: Z */
655     case 1: /* ne: !Z */
656         cond = TCG_COND_EQ;
657         value = cpu_ZF;
658         break;
659 
660     case 2: /* cs: C */
661     case 3: /* cc: !C */
662         cond = TCG_COND_NE;
663         value = cpu_CF;
664         break;
665 
666     case 4: /* mi: N */
667     case 5: /* pl: !N */
668         cond = TCG_COND_LT;
669         value = cpu_NF;
670         break;
671 
672     case 6: /* vs: V */
673     case 7: /* vc: !V */
674         cond = TCG_COND_LT;
675         value = cpu_VF;
676         break;
677 
678     case 8: /* hi: C && !Z */
679     case 9: /* ls: !C || Z -> !(C && !Z) */
680         cond = TCG_COND_NE;
681         value = tcg_temp_new_i32();
682         /* CF is 1 for C, so -CF is an all-bits-set mask for C;
683            ZF is non-zero for !Z; so AND the two subexpressions.  */
684         tcg_gen_neg_i32(value, cpu_CF);
685         tcg_gen_and_i32(value, value, cpu_ZF);
686         break;
687 
688     case 10: /* ge: N == V -> N ^ V == 0 */
689     case 11: /* lt: N != V -> N ^ V != 0 */
690         /* Since we're only interested in the sign bit, == 0 is >= 0.  */
691         cond = TCG_COND_GE;
692         value = tcg_temp_new_i32();
693         tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
694         break;
695 
696     case 12: /* gt: !Z && N == V */
697     case 13: /* le: Z || N != V */
698         cond = TCG_COND_NE;
699         value = tcg_temp_new_i32();
700         /* (N == V) is equal to the sign bit of ~(NF ^ VF).  Propagate
701          * the sign bit then AND with ZF to yield the result.  */
702         tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
703         tcg_gen_sari_i32(value, value, 31);
704         tcg_gen_andc_i32(value, cpu_ZF, value);
705         break;
706 
707     case 14: /* always */
708     case 15: /* always */
709         /* Use the ALWAYS condition, which will fold early.
710          * It doesn't matter what we use for the value.  */
711         cond = TCG_COND_ALWAYS;
712         value = cpu_ZF;
713         goto no_invert;
714 
715     default:
716         fprintf(stderr, "Bad condition code 0x%x\n", cc);
717         abort();
718     }
719 
720     if (cc & 1) {
721         cond = tcg_invert_cond(cond);
722     }
723 
724  no_invert:
725     cmp->cond = cond;
726     cmp->value = value;
727 }
728 
729 void arm_jump_cc(DisasCompare *cmp, TCGLabel *label)
730 {
731     tcg_gen_brcondi_i32(cmp->cond, cmp->value, 0, label);
732 }
733 
734 void arm_gen_test_cc(int cc, TCGLabel *label)
735 {
736     DisasCompare cmp;
737     arm_test_cc(&cmp, cc);
738     arm_jump_cc(&cmp, label);
739 }
740 
741 void gen_set_condexec(DisasContext *s)
742 {
743     if (s->condexec_mask) {
744         uint32_t val = (s->condexec_cond << 4) | (s->condexec_mask >> 1);
745 
746         store_cpu_field_constant(val, condexec_bits);
747     }
748 }
749 
750 void gen_update_pc(DisasContext *s, target_long diff)
751 {
752     gen_pc_plus_diff(s, cpu_R[15], diff);
753     s->pc_save = s->pc_curr + diff;
754 }
755 
756 /* Set PC and Thumb state from var.  var is marked as dead.  */
757 static inline void gen_bx(DisasContext *s, TCGv_i32 var)
758 {
759     s->base.is_jmp = DISAS_JUMP;
760     tcg_gen_andi_i32(cpu_R[15], var, ~1);
761     tcg_gen_andi_i32(var, var, 1);
762     store_cpu_field(var, thumb);
763     s->pc_save = -1;
764 }
765 
766 /*
767  * Set PC and Thumb state from var. var is marked as dead.
768  * For M-profile CPUs, include logic to detect exception-return
769  * branches and handle them. This is needed for Thumb POP/LDM to PC, LDR to PC,
770  * and BX reg, and no others, and happens only for code in Handler mode.
771  * The Security Extension also requires us to check for the FNC_RETURN
772  * which signals a function return from non-secure state; this can happen
773  * in both Handler and Thread mode.
774  * To avoid having to do multiple comparisons in inline generated code,
775  * we make the check we do here loose, so it will match for EXC_RETURN
776  * in Thread mode. For system emulation do_v7m_exception_exit() checks
777  * for these spurious cases and returns without doing anything (giving
778  * the same behaviour as for a branch to a non-magic address).
779  *
780  * In linux-user mode it is unclear what the right behaviour for an
781  * attempted FNC_RETURN should be, because in real hardware this will go
782  * directly to Secure code (ie not the Linux kernel) which will then treat
783  * the error in any way it chooses. For QEMU we opt to make the FNC_RETURN
784  * attempt behave the way it would on a CPU without the security extension,
785  * which is to say "like a normal branch". That means we can simply treat
786  * all branches as normal with no magic address behaviour.
787  */
788 static inline void gen_bx_excret(DisasContext *s, TCGv_i32 var)
789 {
790     /* Generate the same code here as for a simple bx, but flag via
791      * s->base.is_jmp that we need to do the rest of the work later.
792      */
793     gen_bx(s, var);
794 #ifndef CONFIG_USER_ONLY
795     if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY) ||
796         (s->v7m_handler_mode && arm_dc_feature(s, ARM_FEATURE_M))) {
797         s->base.is_jmp = DISAS_BX_EXCRET;
798     }
799 #endif
800 }
801 
802 static inline void gen_bx_excret_final_code(DisasContext *s)
803 {
804     /* Generate the code to finish possible exception return and end the TB */
805     DisasLabel excret_label = gen_disas_label(s);
806     uint32_t min_magic;
807 
808     if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY)) {
809         /* Covers FNC_RETURN and EXC_RETURN magic */
810         min_magic = FNC_RETURN_MIN_MAGIC;
811     } else {
812         /* EXC_RETURN magic only */
813         min_magic = EXC_RETURN_MIN_MAGIC;
814     }
815 
816     /* Is the new PC value in the magic range indicating exception return? */
817     tcg_gen_brcondi_i32(TCG_COND_GEU, cpu_R[15], min_magic, excret_label.label);
818     /* No: end the TB as we would for a DISAS_JMP */
819     if (s->ss_active) {
820         gen_singlestep_exception(s);
821     } else {
822         tcg_gen_exit_tb(NULL, 0);
823     }
824     set_disas_label(s, excret_label);
825     /* Yes: this is an exception return.
826      * At this point in runtime env->regs[15] and env->thumb will hold
827      * the exception-return magic number, which do_v7m_exception_exit()
828      * will read. Nothing else will be able to see those values because
829      * the cpu-exec main loop guarantees that we will always go straight
830      * from raising the exception to the exception-handling code.
831      *
832      * gen_ss_advance(s) does nothing on M profile currently but
833      * calling it is conceptually the right thing as we have executed
834      * this instruction (compare SWI, HVC, SMC handling).
835      */
836     gen_ss_advance(s);
837     gen_exception_internal(EXCP_EXCEPTION_EXIT);
838 }
839 
840 static inline void gen_bxns(DisasContext *s, int rm)
841 {
842     TCGv_i32 var = load_reg(s, rm);
843 
844     /* The bxns helper may raise an EXCEPTION_EXIT exception, so in theory
845      * we need to sync state before calling it, but:
846      *  - we don't need to do gen_update_pc() because the bxns helper will
847      *    always set the PC itself
848      *  - we don't need to do gen_set_condexec() because BXNS is UNPREDICTABLE
849      *    unless it's outside an IT block or the last insn in an IT block,
850      *    so we know that condexec == 0 (already set at the top of the TB)
851      *    is correct in the non-UNPREDICTABLE cases, and we can choose
852      *    "zeroes the IT bits" as our UNPREDICTABLE behaviour otherwise.
853      */
854     gen_helper_v7m_bxns(tcg_env, var);
855     s->base.is_jmp = DISAS_EXIT;
856 }
857 
858 static inline void gen_blxns(DisasContext *s, int rm)
859 {
860     TCGv_i32 var = load_reg(s, rm);
861 
862     /* We don't need to sync condexec state, for the same reason as bxns.
863      * We do however need to set the PC, because the blxns helper reads it.
864      * The blxns helper may throw an exception.
865      */
866     gen_update_pc(s, curr_insn_len(s));
867     gen_helper_v7m_blxns(tcg_env, var);
868     s->base.is_jmp = DISAS_EXIT;
869 }
870 
871 /* Variant of store_reg which uses branch&exchange logic when storing
872    to r15 in ARM architecture v7 and above. The source must be a temporary
873    and will be marked as dead. */
874 static inline void store_reg_bx(DisasContext *s, int reg, TCGv_i32 var)
875 {
876     if (reg == 15 && ENABLE_ARCH_7) {
877         gen_bx(s, var);
878     } else {
879         store_reg(s, reg, var);
880     }
881 }
882 
883 /* Variant of store_reg which uses branch&exchange logic when storing
884  * to r15 in ARM architecture v5T and above. This is used for storing
885  * the results of a LDR/LDM/POP into r15, and corresponds to the cases
886  * in the ARM ARM which use the LoadWritePC() pseudocode function. */
887 static inline void store_reg_from_load(DisasContext *s, int reg, TCGv_i32 var)
888 {
889     if (reg == 15 && ENABLE_ARCH_5) {
890         gen_bx_excret(s, var);
891     } else {
892         store_reg(s, reg, var);
893     }
894 }
895 
896 #ifdef CONFIG_USER_ONLY
897 #define IS_USER_ONLY 1
898 #else
899 #define IS_USER_ONLY 0
900 #endif
901 
902 MemOp pow2_align(unsigned i)
903 {
904     static const MemOp mop_align[] = {
905         0, MO_ALIGN_2, MO_ALIGN_4, MO_ALIGN_8, MO_ALIGN_16, MO_ALIGN_32
906     };
907     g_assert(i < ARRAY_SIZE(mop_align));
908     return mop_align[i];
909 }
910 
911 /*
912  * Abstractions of "generate code to do a guest load/store for
913  * AArch32", where a vaddr is always 32 bits (and is zero
914  * extended if we're a 64 bit core) and  data is also
915  * 32 bits unless specifically doing a 64 bit access.
916  * These functions work like tcg_gen_qemu_{ld,st}* except
917  * that the address argument is TCGv_i32 rather than TCGv.
918  */
919 
920 static TCGv gen_aa32_addr(DisasContext *s, TCGv_i32 a32, MemOp op)
921 {
922     TCGv addr = tcg_temp_new();
923     tcg_gen_extu_i32_tl(addr, a32);
924 
925     /* Not needed for user-mode BE32, where we use MO_BE instead.  */
926     if (!IS_USER_ONLY && s->sctlr_b && (op & MO_SIZE) < MO_32) {
927         tcg_gen_xori_tl(addr, addr, 4 - (1 << (op & MO_SIZE)));
928     }
929     return addr;
930 }
931 
932 /*
933  * Internal routines are used for NEON cases where the endianness
934  * and/or alignment has already been taken into account and manipulated.
935  */
936 void gen_aa32_ld_internal_i32(DisasContext *s, TCGv_i32 val,
937                               TCGv_i32 a32, int index, MemOp opc)
938 {
939     TCGv addr = gen_aa32_addr(s, a32, opc);
940     tcg_gen_qemu_ld_i32(val, addr, index, opc);
941 }
942 
943 void gen_aa32_st_internal_i32(DisasContext *s, TCGv_i32 val,
944                               TCGv_i32 a32, int index, MemOp opc)
945 {
946     TCGv addr = gen_aa32_addr(s, a32, opc);
947     tcg_gen_qemu_st_i32(val, addr, index, opc);
948 }
949 
950 void gen_aa32_ld_internal_i64(DisasContext *s, TCGv_i64 val,
951                               TCGv_i32 a32, int index, MemOp opc)
952 {
953     TCGv addr = gen_aa32_addr(s, a32, opc);
954 
955     tcg_gen_qemu_ld_i64(val, addr, index, opc);
956 
957     /* Not needed for user-mode BE32, where we use MO_BE instead.  */
958     if (!IS_USER_ONLY && s->sctlr_b && (opc & MO_SIZE) == MO_64) {
959         tcg_gen_rotri_i64(val, val, 32);
960     }
961 }
962 
963 void gen_aa32_st_internal_i64(DisasContext *s, TCGv_i64 val,
964                               TCGv_i32 a32, int index, MemOp opc)
965 {
966     TCGv addr = gen_aa32_addr(s, a32, opc);
967 
968     /* Not needed for user-mode BE32, where we use MO_BE instead.  */
969     if (!IS_USER_ONLY && s->sctlr_b && (opc & MO_SIZE) == MO_64) {
970         TCGv_i64 tmp = tcg_temp_new_i64();
971         tcg_gen_rotri_i64(tmp, val, 32);
972         tcg_gen_qemu_st_i64(tmp, addr, index, opc);
973     } else {
974         tcg_gen_qemu_st_i64(val, addr, index, opc);
975     }
976 }
977 
978 void gen_aa32_ld_i32(DisasContext *s, TCGv_i32 val, TCGv_i32 a32,
979                      int index, MemOp opc)
980 {
981     gen_aa32_ld_internal_i32(s, val, a32, index, finalize_memop(s, opc));
982 }
983 
984 void gen_aa32_st_i32(DisasContext *s, TCGv_i32 val, TCGv_i32 a32,
985                      int index, MemOp opc)
986 {
987     gen_aa32_st_internal_i32(s, val, a32, index, finalize_memop(s, opc));
988 }
989 
990 void gen_aa32_ld_i64(DisasContext *s, TCGv_i64 val, TCGv_i32 a32,
991                      int index, MemOp opc)
992 {
993     gen_aa32_ld_internal_i64(s, val, a32, index, finalize_memop(s, opc));
994 }
995 
996 void gen_aa32_st_i64(DisasContext *s, TCGv_i64 val, TCGv_i32 a32,
997                      int index, MemOp opc)
998 {
999     gen_aa32_st_internal_i64(s, val, a32, index, finalize_memop(s, opc));
1000 }
1001 
1002 #define DO_GEN_LD(SUFF, OPC)                                            \
1003     static inline void gen_aa32_ld##SUFF(DisasContext *s, TCGv_i32 val, \
1004                                          TCGv_i32 a32, int index)       \
1005     {                                                                   \
1006         gen_aa32_ld_i32(s, val, a32, index, OPC);                       \
1007     }
1008 
1009 #define DO_GEN_ST(SUFF, OPC)                                            \
1010     static inline void gen_aa32_st##SUFF(DisasContext *s, TCGv_i32 val, \
1011                                          TCGv_i32 a32, int index)       \
1012     {                                                                   \
1013         gen_aa32_st_i32(s, val, a32, index, OPC);                       \
1014     }
1015 
1016 static inline void gen_hvc(DisasContext *s, int imm16)
1017 {
1018     /* The pre HVC helper handles cases when HVC gets trapped
1019      * as an undefined insn by runtime configuration (ie before
1020      * the insn really executes).
1021      */
1022     gen_update_pc(s, 0);
1023     gen_helper_pre_hvc(tcg_env);
1024     /* Otherwise we will treat this as a real exception which
1025      * happens after execution of the insn. (The distinction matters
1026      * for the PC value reported to the exception handler and also
1027      * for single stepping.)
1028      */
1029     s->svc_imm = imm16;
1030     gen_update_pc(s, curr_insn_len(s));
1031     s->base.is_jmp = DISAS_HVC;
1032 }
1033 
1034 static inline void gen_smc(DisasContext *s)
1035 {
1036     /* As with HVC, we may take an exception either before or after
1037      * the insn executes.
1038      */
1039     gen_update_pc(s, 0);
1040     gen_helper_pre_smc(tcg_env, tcg_constant_i32(syn_aa32_smc()));
1041     gen_update_pc(s, curr_insn_len(s));
1042     s->base.is_jmp = DISAS_SMC;
1043 }
1044 
1045 static void gen_exception_internal_insn(DisasContext *s, int excp)
1046 {
1047     gen_set_condexec(s);
1048     gen_update_pc(s, 0);
1049     gen_exception_internal(excp);
1050     s->base.is_jmp = DISAS_NORETURN;
1051 }
1052 
1053 static void gen_exception_el_v(int excp, uint32_t syndrome, TCGv_i32 tcg_el)
1054 {
1055     gen_helper_exception_with_syndrome_el(tcg_env, tcg_constant_i32(excp),
1056                                           tcg_constant_i32(syndrome), tcg_el);
1057 }
1058 
1059 static void gen_exception_el(int excp, uint32_t syndrome, uint32_t target_el)
1060 {
1061     gen_exception_el_v(excp, syndrome, tcg_constant_i32(target_el));
1062 }
1063 
1064 static void gen_exception(int excp, uint32_t syndrome)
1065 {
1066     gen_helper_exception_with_syndrome(tcg_env, tcg_constant_i32(excp),
1067                                        tcg_constant_i32(syndrome));
1068 }
1069 
1070 static void gen_exception_insn_el_v(DisasContext *s, target_long pc_diff,
1071                                     int excp, uint32_t syn, TCGv_i32 tcg_el)
1072 {
1073     if (s->aarch64) {
1074         gen_a64_update_pc(s, pc_diff);
1075     } else {
1076         gen_set_condexec(s);
1077         gen_update_pc(s, pc_diff);
1078     }
1079     gen_exception_el_v(excp, syn, tcg_el);
1080     s->base.is_jmp = DISAS_NORETURN;
1081 }
1082 
1083 void gen_exception_insn_el(DisasContext *s, target_long pc_diff, int excp,
1084                            uint32_t syn, uint32_t target_el)
1085 {
1086     gen_exception_insn_el_v(s, pc_diff, excp, syn,
1087                             tcg_constant_i32(target_el));
1088 }
1089 
1090 void gen_exception_insn(DisasContext *s, target_long pc_diff,
1091                         int excp, uint32_t syn)
1092 {
1093     if (s->aarch64) {
1094         gen_a64_update_pc(s, pc_diff);
1095     } else {
1096         gen_set_condexec(s);
1097         gen_update_pc(s, pc_diff);
1098     }
1099     gen_exception(excp, syn);
1100     s->base.is_jmp = DISAS_NORETURN;
1101 }
1102 
1103 static void gen_exception_bkpt_insn(DisasContext *s, uint32_t syn)
1104 {
1105     gen_set_condexec(s);
1106     gen_update_pc(s, 0);
1107     gen_helper_exception_bkpt_insn(tcg_env, tcg_constant_i32(syn));
1108     s->base.is_jmp = DISAS_NORETURN;
1109 }
1110 
1111 void unallocated_encoding(DisasContext *s)
1112 {
1113     /* Unallocated and reserved encodings are uncategorized */
1114     gen_exception_insn(s, 0, EXCP_UDEF, syn_uncategorized());
1115 }
1116 
1117 /* Force a TB lookup after an instruction that changes the CPU state.  */
1118 void gen_lookup_tb(DisasContext *s)
1119 {
1120     gen_pc_plus_diff(s, cpu_R[15], curr_insn_len(s));
1121     s->base.is_jmp = DISAS_EXIT;
1122 }
1123 
1124 static inline void gen_hlt(DisasContext *s, int imm)
1125 {
1126     /* HLT. This has two purposes.
1127      * Architecturally, it is an external halting debug instruction.
1128      * Since QEMU doesn't implement external debug, we treat this as
1129      * it is required for halting debug disabled: it will UNDEF.
1130      * Secondly, "HLT 0x3C" is a T32 semihosting trap instruction,
1131      * and "HLT 0xF000" is an A32 semihosting syscall. These traps
1132      * must trigger semihosting even for ARMv7 and earlier, where
1133      * HLT was an undefined encoding.
1134      * In system mode, we don't allow userspace access to
1135      * semihosting, to provide some semblance of security
1136      * (and for consistency with our 32-bit semihosting).
1137      */
1138     if (semihosting_enabled(s->current_el == 0) &&
1139         (imm == (s->thumb ? 0x3c : 0xf000))) {
1140         gen_exception_internal_insn(s, EXCP_SEMIHOST);
1141         return;
1142     }
1143 
1144     unallocated_encoding(s);
1145 }
1146 
1147 /*
1148  * Return the offset of a "full" NEON Dreg.
1149  */
1150 long neon_full_reg_offset(unsigned reg)
1151 {
1152     return offsetof(CPUARMState, vfp.zregs[reg >> 1].d[reg & 1]);
1153 }
1154 
1155 /*
1156  * Return the offset of a 2**SIZE piece of a NEON register, at index ELE,
1157  * where 0 is the least significant end of the register.
1158  */
1159 long neon_element_offset(int reg, int element, MemOp memop)
1160 {
1161     int element_size = 1 << (memop & MO_SIZE);
1162     int ofs = element * element_size;
1163 #if HOST_BIG_ENDIAN
1164     /*
1165      * Calculate the offset assuming fully little-endian,
1166      * then XOR to account for the order of the 8-byte units.
1167      */
1168     if (element_size < 8) {
1169         ofs ^= 8 - element_size;
1170     }
1171 #endif
1172     return neon_full_reg_offset(reg) + ofs;
1173 }
1174 
1175 /* Return the offset of a VFP Dreg (dp = true) or VFP Sreg (dp = false). */
1176 long vfp_reg_offset(bool dp, unsigned reg)
1177 {
1178     if (dp) {
1179         return neon_element_offset(reg, 0, MO_64);
1180     } else {
1181         return neon_element_offset(reg >> 1, reg & 1, MO_32);
1182     }
1183 }
1184 
1185 void read_neon_element32(TCGv_i32 dest, int reg, int ele, MemOp memop)
1186 {
1187     long off = neon_element_offset(reg, ele, memop);
1188 
1189     switch (memop) {
1190     case MO_SB:
1191         tcg_gen_ld8s_i32(dest, tcg_env, off);
1192         break;
1193     case MO_UB:
1194         tcg_gen_ld8u_i32(dest, tcg_env, off);
1195         break;
1196     case MO_SW:
1197         tcg_gen_ld16s_i32(dest, tcg_env, off);
1198         break;
1199     case MO_UW:
1200         tcg_gen_ld16u_i32(dest, tcg_env, off);
1201         break;
1202     case MO_UL:
1203     case MO_SL:
1204         tcg_gen_ld_i32(dest, tcg_env, off);
1205         break;
1206     default:
1207         g_assert_not_reached();
1208     }
1209 }
1210 
1211 void read_neon_element64(TCGv_i64 dest, int reg, int ele, MemOp memop)
1212 {
1213     long off = neon_element_offset(reg, ele, memop);
1214 
1215     switch (memop) {
1216     case MO_SL:
1217         tcg_gen_ld32s_i64(dest, tcg_env, off);
1218         break;
1219     case MO_UL:
1220         tcg_gen_ld32u_i64(dest, tcg_env, off);
1221         break;
1222     case MO_UQ:
1223         tcg_gen_ld_i64(dest, tcg_env, off);
1224         break;
1225     default:
1226         g_assert_not_reached();
1227     }
1228 }
1229 
1230 void write_neon_element32(TCGv_i32 src, int reg, int ele, MemOp memop)
1231 {
1232     long off = neon_element_offset(reg, ele, memop);
1233 
1234     switch (memop) {
1235     case MO_8:
1236         tcg_gen_st8_i32(src, tcg_env, off);
1237         break;
1238     case MO_16:
1239         tcg_gen_st16_i32(src, tcg_env, off);
1240         break;
1241     case MO_32:
1242         tcg_gen_st_i32(src, tcg_env, off);
1243         break;
1244     default:
1245         g_assert_not_reached();
1246     }
1247 }
1248 
1249 void write_neon_element64(TCGv_i64 src, int reg, int ele, MemOp memop)
1250 {
1251     long off = neon_element_offset(reg, ele, memop);
1252 
1253     switch (memop) {
1254     case MO_32:
1255         tcg_gen_st32_i64(src, tcg_env, off);
1256         break;
1257     case MO_64:
1258         tcg_gen_st_i64(src, tcg_env, off);
1259         break;
1260     default:
1261         g_assert_not_reached();
1262     }
1263 }
1264 
1265 #define ARM_CP_RW_BIT   (1 << 20)
1266 
1267 static inline void iwmmxt_load_reg(TCGv_i64 var, int reg)
1268 {
1269     tcg_gen_ld_i64(var, tcg_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1270 }
1271 
1272 static inline void iwmmxt_store_reg(TCGv_i64 var, int reg)
1273 {
1274     tcg_gen_st_i64(var, tcg_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1275 }
1276 
1277 static inline TCGv_i32 iwmmxt_load_creg(int reg)
1278 {
1279     TCGv_i32 var = tcg_temp_new_i32();
1280     tcg_gen_ld_i32(var, tcg_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1281     return var;
1282 }
1283 
1284 static inline void iwmmxt_store_creg(int reg, TCGv_i32 var)
1285 {
1286     tcg_gen_st_i32(var, tcg_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1287 }
1288 
1289 static inline void gen_op_iwmmxt_movq_wRn_M0(int rn)
1290 {
1291     iwmmxt_store_reg(cpu_M0, rn);
1292 }
1293 
1294 static inline void gen_op_iwmmxt_movq_M0_wRn(int rn)
1295 {
1296     iwmmxt_load_reg(cpu_M0, rn);
1297 }
1298 
1299 static inline void gen_op_iwmmxt_orq_M0_wRn(int rn)
1300 {
1301     iwmmxt_load_reg(cpu_V1, rn);
1302     tcg_gen_or_i64(cpu_M0, cpu_M0, cpu_V1);
1303 }
1304 
1305 static inline void gen_op_iwmmxt_andq_M0_wRn(int rn)
1306 {
1307     iwmmxt_load_reg(cpu_V1, rn);
1308     tcg_gen_and_i64(cpu_M0, cpu_M0, cpu_V1);
1309 }
1310 
1311 static inline void gen_op_iwmmxt_xorq_M0_wRn(int rn)
1312 {
1313     iwmmxt_load_reg(cpu_V1, rn);
1314     tcg_gen_xor_i64(cpu_M0, cpu_M0, cpu_V1);
1315 }
1316 
1317 #define IWMMXT_OP(name) \
1318 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1319 { \
1320     iwmmxt_load_reg(cpu_V1, rn); \
1321     gen_helper_iwmmxt_##name(cpu_M0, cpu_M0, cpu_V1); \
1322 }
1323 
1324 #define IWMMXT_OP_ENV(name) \
1325 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1326 { \
1327     iwmmxt_load_reg(cpu_V1, rn); \
1328     gen_helper_iwmmxt_##name(cpu_M0, tcg_env, cpu_M0, cpu_V1); \
1329 }
1330 
1331 #define IWMMXT_OP_ENV_SIZE(name) \
1332 IWMMXT_OP_ENV(name##b) \
1333 IWMMXT_OP_ENV(name##w) \
1334 IWMMXT_OP_ENV(name##l)
1335 
1336 #define IWMMXT_OP_ENV1(name) \
1337 static inline void gen_op_iwmmxt_##name##_M0(void) \
1338 { \
1339     gen_helper_iwmmxt_##name(cpu_M0, tcg_env, cpu_M0); \
1340 }
1341 
1342 IWMMXT_OP(maddsq)
1343 IWMMXT_OP(madduq)
1344 IWMMXT_OP(sadb)
1345 IWMMXT_OP(sadw)
1346 IWMMXT_OP(mulslw)
1347 IWMMXT_OP(mulshw)
1348 IWMMXT_OP(mululw)
1349 IWMMXT_OP(muluhw)
1350 IWMMXT_OP(macsw)
1351 IWMMXT_OP(macuw)
1352 
1353 IWMMXT_OP_ENV_SIZE(unpackl)
1354 IWMMXT_OP_ENV_SIZE(unpackh)
1355 
1356 IWMMXT_OP_ENV1(unpacklub)
1357 IWMMXT_OP_ENV1(unpackluw)
1358 IWMMXT_OP_ENV1(unpacklul)
1359 IWMMXT_OP_ENV1(unpackhub)
1360 IWMMXT_OP_ENV1(unpackhuw)
1361 IWMMXT_OP_ENV1(unpackhul)
1362 IWMMXT_OP_ENV1(unpacklsb)
1363 IWMMXT_OP_ENV1(unpacklsw)
1364 IWMMXT_OP_ENV1(unpacklsl)
1365 IWMMXT_OP_ENV1(unpackhsb)
1366 IWMMXT_OP_ENV1(unpackhsw)
1367 IWMMXT_OP_ENV1(unpackhsl)
1368 
1369 IWMMXT_OP_ENV_SIZE(cmpeq)
1370 IWMMXT_OP_ENV_SIZE(cmpgtu)
1371 IWMMXT_OP_ENV_SIZE(cmpgts)
1372 
1373 IWMMXT_OP_ENV_SIZE(mins)
1374 IWMMXT_OP_ENV_SIZE(minu)
1375 IWMMXT_OP_ENV_SIZE(maxs)
1376 IWMMXT_OP_ENV_SIZE(maxu)
1377 
1378 IWMMXT_OP_ENV_SIZE(subn)
1379 IWMMXT_OP_ENV_SIZE(addn)
1380 IWMMXT_OP_ENV_SIZE(subu)
1381 IWMMXT_OP_ENV_SIZE(addu)
1382 IWMMXT_OP_ENV_SIZE(subs)
1383 IWMMXT_OP_ENV_SIZE(adds)
1384 
1385 IWMMXT_OP_ENV(avgb0)
1386 IWMMXT_OP_ENV(avgb1)
1387 IWMMXT_OP_ENV(avgw0)
1388 IWMMXT_OP_ENV(avgw1)
1389 
1390 IWMMXT_OP_ENV(packuw)
1391 IWMMXT_OP_ENV(packul)
1392 IWMMXT_OP_ENV(packuq)
1393 IWMMXT_OP_ENV(packsw)
1394 IWMMXT_OP_ENV(packsl)
1395 IWMMXT_OP_ENV(packsq)
1396 
1397 static void gen_op_iwmmxt_set_mup(void)
1398 {
1399     TCGv_i32 tmp;
1400     tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1401     tcg_gen_ori_i32(tmp, tmp, 2);
1402     store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1403 }
1404 
1405 static void gen_op_iwmmxt_set_cup(void)
1406 {
1407     TCGv_i32 tmp;
1408     tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1409     tcg_gen_ori_i32(tmp, tmp, 1);
1410     store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1411 }
1412 
1413 static void gen_op_iwmmxt_setpsr_nz(void)
1414 {
1415     TCGv_i32 tmp = tcg_temp_new_i32();
1416     gen_helper_iwmmxt_setpsr_nz(tmp, cpu_M0);
1417     store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCASF]);
1418 }
1419 
1420 static inline void gen_op_iwmmxt_addl_M0_wRn(int rn)
1421 {
1422     iwmmxt_load_reg(cpu_V1, rn);
1423     tcg_gen_ext32u_i64(cpu_V1, cpu_V1);
1424     tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1425 }
1426 
1427 static inline int gen_iwmmxt_address(DisasContext *s, uint32_t insn,
1428                                      TCGv_i32 dest)
1429 {
1430     int rd;
1431     uint32_t offset;
1432     TCGv_i32 tmp;
1433 
1434     rd = (insn >> 16) & 0xf;
1435     tmp = load_reg(s, rd);
1436 
1437     offset = (insn & 0xff) << ((insn >> 7) & 2);
1438     if (insn & (1 << 24)) {
1439         /* Pre indexed */
1440         if (insn & (1 << 23))
1441             tcg_gen_addi_i32(tmp, tmp, offset);
1442         else
1443             tcg_gen_addi_i32(tmp, tmp, -offset);
1444         tcg_gen_mov_i32(dest, tmp);
1445         if (insn & (1 << 21)) {
1446             store_reg(s, rd, tmp);
1447         }
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     return 0;
1480 }
1481 
1482 /* Disassemble an iwMMXt instruction.  Returns nonzero if an error occurred
1483    (ie. an undefined instruction).  */
1484 static int disas_iwmmxt_insn(DisasContext *s, uint32_t insn)
1485 {
1486     int rd, wrd;
1487     int rdhi, rdlo, rd0, rd1, i;
1488     TCGv_i32 addr;
1489     TCGv_i32 tmp, tmp2, tmp3;
1490 
1491     if ((insn & 0x0e000e00) == 0x0c000000) {
1492         if ((insn & 0x0fe00ff0) == 0x0c400000) {
1493             wrd = insn & 0xf;
1494             rdlo = (insn >> 12) & 0xf;
1495             rdhi = (insn >> 16) & 0xf;
1496             if (insn & ARM_CP_RW_BIT) {                         /* TMRRC */
1497                 iwmmxt_load_reg(cpu_V0, wrd);
1498                 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
1499                 tcg_gen_extrh_i64_i32(cpu_R[rdhi], cpu_V0);
1500             } else {                                    /* TMCRR */
1501                 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
1502                 iwmmxt_store_reg(cpu_V0, wrd);
1503                 gen_op_iwmmxt_set_mup();
1504             }
1505             return 0;
1506         }
1507 
1508         wrd = (insn >> 12) & 0xf;
1509         addr = tcg_temp_new_i32();
1510         if (gen_iwmmxt_address(s, insn, addr)) {
1511             return 1;
1512         }
1513         if (insn & ARM_CP_RW_BIT) {
1514             if ((insn >> 28) == 0xf) {                  /* WLDRW wCx */
1515                 tmp = tcg_temp_new_i32();
1516                 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1517                 iwmmxt_store_creg(wrd, tmp);
1518             } else {
1519                 i = 1;
1520                 if (insn & (1 << 8)) {
1521                     if (insn & (1 << 22)) {             /* WLDRD */
1522                         gen_aa32_ld64(s, cpu_M0, addr, get_mem_index(s));
1523                         i = 0;
1524                     } else {                            /* WLDRW wRd */
1525                         tmp = tcg_temp_new_i32();
1526                         gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1527                     }
1528                 } else {
1529                     tmp = tcg_temp_new_i32();
1530                     if (insn & (1 << 22)) {             /* WLDRH */
1531                         gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
1532                     } else {                            /* WLDRB */
1533                         gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
1534                     }
1535                 }
1536                 if (i) {
1537                     tcg_gen_extu_i32_i64(cpu_M0, tmp);
1538                 }
1539                 gen_op_iwmmxt_movq_wRn_M0(wrd);
1540             }
1541         } else {
1542             if ((insn >> 28) == 0xf) {                  /* WSTRW wCx */
1543                 tmp = iwmmxt_load_creg(wrd);
1544                 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1545             } else {
1546                 gen_op_iwmmxt_movq_M0_wRn(wrd);
1547                 tmp = tcg_temp_new_i32();
1548                 if (insn & (1 << 8)) {
1549                     if (insn & (1 << 22)) {             /* WSTRD */
1550                         gen_aa32_st64(s, cpu_M0, addr, get_mem_index(s));
1551                     } else {                            /* WSTRW wRd */
1552                         tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1553                         gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1554                     }
1555                 } else {
1556                     if (insn & (1 << 22)) {             /* WSTRH */
1557                         tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1558                         gen_aa32_st16(s, tmp, addr, get_mem_index(s));
1559                     } else {                            /* WSTRB */
1560                         tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1561                         gen_aa32_st8(s, tmp, addr, get_mem_index(s));
1562                     }
1563                 }
1564             }
1565         }
1566         return 0;
1567     }
1568 
1569     if ((insn & 0x0f000000) != 0x0e000000)
1570         return 1;
1571 
1572     switch (((insn >> 12) & 0xf00) | ((insn >> 4) & 0xff)) {
1573     case 0x000:                                                 /* WOR */
1574         wrd = (insn >> 12) & 0xf;
1575         rd0 = (insn >> 0) & 0xf;
1576         rd1 = (insn >> 16) & 0xf;
1577         gen_op_iwmmxt_movq_M0_wRn(rd0);
1578         gen_op_iwmmxt_orq_M0_wRn(rd1);
1579         gen_op_iwmmxt_setpsr_nz();
1580         gen_op_iwmmxt_movq_wRn_M0(wrd);
1581         gen_op_iwmmxt_set_mup();
1582         gen_op_iwmmxt_set_cup();
1583         break;
1584     case 0x011:                                                 /* TMCR */
1585         if (insn & 0xf)
1586             return 1;
1587         rd = (insn >> 12) & 0xf;
1588         wrd = (insn >> 16) & 0xf;
1589         switch (wrd) {
1590         case ARM_IWMMXT_wCID:
1591         case ARM_IWMMXT_wCASF:
1592             break;
1593         case ARM_IWMMXT_wCon:
1594             gen_op_iwmmxt_set_cup();
1595             /* Fall through.  */
1596         case ARM_IWMMXT_wCSSF:
1597             tmp = iwmmxt_load_creg(wrd);
1598             tmp2 = load_reg(s, rd);
1599             tcg_gen_andc_i32(tmp, tmp, tmp2);
1600             iwmmxt_store_creg(wrd, tmp);
1601             break;
1602         case ARM_IWMMXT_wCGR0:
1603         case ARM_IWMMXT_wCGR1:
1604         case ARM_IWMMXT_wCGR2:
1605         case ARM_IWMMXT_wCGR3:
1606             gen_op_iwmmxt_set_cup();
1607             tmp = load_reg(s, rd);
1608             iwmmxt_store_creg(wrd, tmp);
1609             break;
1610         default:
1611             return 1;
1612         }
1613         break;
1614     case 0x100:                                                 /* WXOR */
1615         wrd = (insn >> 12) & 0xf;
1616         rd0 = (insn >> 0) & 0xf;
1617         rd1 = (insn >> 16) & 0xf;
1618         gen_op_iwmmxt_movq_M0_wRn(rd0);
1619         gen_op_iwmmxt_xorq_M0_wRn(rd1);
1620         gen_op_iwmmxt_setpsr_nz();
1621         gen_op_iwmmxt_movq_wRn_M0(wrd);
1622         gen_op_iwmmxt_set_mup();
1623         gen_op_iwmmxt_set_cup();
1624         break;
1625     case 0x111:                                                 /* TMRC */
1626         if (insn & 0xf)
1627             return 1;
1628         rd = (insn >> 12) & 0xf;
1629         wrd = (insn >> 16) & 0xf;
1630         tmp = iwmmxt_load_creg(wrd);
1631         store_reg(s, rd, tmp);
1632         break;
1633     case 0x300:                                                 /* WANDN */
1634         wrd = (insn >> 12) & 0xf;
1635         rd0 = (insn >> 0) & 0xf;
1636         rd1 = (insn >> 16) & 0xf;
1637         gen_op_iwmmxt_movq_M0_wRn(rd0);
1638         tcg_gen_neg_i64(cpu_M0, cpu_M0);
1639         gen_op_iwmmxt_andq_M0_wRn(rd1);
1640         gen_op_iwmmxt_setpsr_nz();
1641         gen_op_iwmmxt_movq_wRn_M0(wrd);
1642         gen_op_iwmmxt_set_mup();
1643         gen_op_iwmmxt_set_cup();
1644         break;
1645     case 0x200:                                                 /* WAND */
1646         wrd = (insn >> 12) & 0xf;
1647         rd0 = (insn >> 0) & 0xf;
1648         rd1 = (insn >> 16) & 0xf;
1649         gen_op_iwmmxt_movq_M0_wRn(rd0);
1650         gen_op_iwmmxt_andq_M0_wRn(rd1);
1651         gen_op_iwmmxt_setpsr_nz();
1652         gen_op_iwmmxt_movq_wRn_M0(wrd);
1653         gen_op_iwmmxt_set_mup();
1654         gen_op_iwmmxt_set_cup();
1655         break;
1656     case 0x810: case 0xa10:                             /* WMADD */
1657         wrd = (insn >> 12) & 0xf;
1658         rd0 = (insn >> 0) & 0xf;
1659         rd1 = (insn >> 16) & 0xf;
1660         gen_op_iwmmxt_movq_M0_wRn(rd0);
1661         if (insn & (1 << 21))
1662             gen_op_iwmmxt_maddsq_M0_wRn(rd1);
1663         else
1664             gen_op_iwmmxt_madduq_M0_wRn(rd1);
1665         gen_op_iwmmxt_movq_wRn_M0(wrd);
1666         gen_op_iwmmxt_set_mup();
1667         break;
1668     case 0x10e: case 0x50e: case 0x90e: case 0xd0e:     /* WUNPCKIL */
1669         wrd = (insn >> 12) & 0xf;
1670         rd0 = (insn >> 16) & 0xf;
1671         rd1 = (insn >> 0) & 0xf;
1672         gen_op_iwmmxt_movq_M0_wRn(rd0);
1673         switch ((insn >> 22) & 3) {
1674         case 0:
1675             gen_op_iwmmxt_unpacklb_M0_wRn(rd1);
1676             break;
1677         case 1:
1678             gen_op_iwmmxt_unpacklw_M0_wRn(rd1);
1679             break;
1680         case 2:
1681             gen_op_iwmmxt_unpackll_M0_wRn(rd1);
1682             break;
1683         case 3:
1684             return 1;
1685         }
1686         gen_op_iwmmxt_movq_wRn_M0(wrd);
1687         gen_op_iwmmxt_set_mup();
1688         gen_op_iwmmxt_set_cup();
1689         break;
1690     case 0x10c: case 0x50c: case 0x90c: case 0xd0c:     /* WUNPCKIH */
1691         wrd = (insn >> 12) & 0xf;
1692         rd0 = (insn >> 16) & 0xf;
1693         rd1 = (insn >> 0) & 0xf;
1694         gen_op_iwmmxt_movq_M0_wRn(rd0);
1695         switch ((insn >> 22) & 3) {
1696         case 0:
1697             gen_op_iwmmxt_unpackhb_M0_wRn(rd1);
1698             break;
1699         case 1:
1700             gen_op_iwmmxt_unpackhw_M0_wRn(rd1);
1701             break;
1702         case 2:
1703             gen_op_iwmmxt_unpackhl_M0_wRn(rd1);
1704             break;
1705         case 3:
1706             return 1;
1707         }
1708         gen_op_iwmmxt_movq_wRn_M0(wrd);
1709         gen_op_iwmmxt_set_mup();
1710         gen_op_iwmmxt_set_cup();
1711         break;
1712     case 0x012: case 0x112: case 0x412: case 0x512:     /* WSAD */
1713         wrd = (insn >> 12) & 0xf;
1714         rd0 = (insn >> 16) & 0xf;
1715         rd1 = (insn >> 0) & 0xf;
1716         gen_op_iwmmxt_movq_M0_wRn(rd0);
1717         if (insn & (1 << 22))
1718             gen_op_iwmmxt_sadw_M0_wRn(rd1);
1719         else
1720             gen_op_iwmmxt_sadb_M0_wRn(rd1);
1721         if (!(insn & (1 << 20)))
1722             gen_op_iwmmxt_addl_M0_wRn(wrd);
1723         gen_op_iwmmxt_movq_wRn_M0(wrd);
1724         gen_op_iwmmxt_set_mup();
1725         break;
1726     case 0x010: case 0x110: case 0x210: case 0x310:     /* WMUL */
1727         wrd = (insn >> 12) & 0xf;
1728         rd0 = (insn >> 16) & 0xf;
1729         rd1 = (insn >> 0) & 0xf;
1730         gen_op_iwmmxt_movq_M0_wRn(rd0);
1731         if (insn & (1 << 21)) {
1732             if (insn & (1 << 20))
1733                 gen_op_iwmmxt_mulshw_M0_wRn(rd1);
1734             else
1735                 gen_op_iwmmxt_mulslw_M0_wRn(rd1);
1736         } else {
1737             if (insn & (1 << 20))
1738                 gen_op_iwmmxt_muluhw_M0_wRn(rd1);
1739             else
1740                 gen_op_iwmmxt_mululw_M0_wRn(rd1);
1741         }
1742         gen_op_iwmmxt_movq_wRn_M0(wrd);
1743         gen_op_iwmmxt_set_mup();
1744         break;
1745     case 0x410: case 0x510: case 0x610: case 0x710:     /* WMAC */
1746         wrd = (insn >> 12) & 0xf;
1747         rd0 = (insn >> 16) & 0xf;
1748         rd1 = (insn >> 0) & 0xf;
1749         gen_op_iwmmxt_movq_M0_wRn(rd0);
1750         if (insn & (1 << 21))
1751             gen_op_iwmmxt_macsw_M0_wRn(rd1);
1752         else
1753             gen_op_iwmmxt_macuw_M0_wRn(rd1);
1754         if (!(insn & (1 << 20))) {
1755             iwmmxt_load_reg(cpu_V1, wrd);
1756             tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1757         }
1758         gen_op_iwmmxt_movq_wRn_M0(wrd);
1759         gen_op_iwmmxt_set_mup();
1760         break;
1761     case 0x006: case 0x406: case 0x806: case 0xc06:     /* WCMPEQ */
1762         wrd = (insn >> 12) & 0xf;
1763         rd0 = (insn >> 16) & 0xf;
1764         rd1 = (insn >> 0) & 0xf;
1765         gen_op_iwmmxt_movq_M0_wRn(rd0);
1766         switch ((insn >> 22) & 3) {
1767         case 0:
1768             gen_op_iwmmxt_cmpeqb_M0_wRn(rd1);
1769             break;
1770         case 1:
1771             gen_op_iwmmxt_cmpeqw_M0_wRn(rd1);
1772             break;
1773         case 2:
1774             gen_op_iwmmxt_cmpeql_M0_wRn(rd1);
1775             break;
1776         case 3:
1777             return 1;
1778         }
1779         gen_op_iwmmxt_movq_wRn_M0(wrd);
1780         gen_op_iwmmxt_set_mup();
1781         gen_op_iwmmxt_set_cup();
1782         break;
1783     case 0x800: case 0x900: case 0xc00: case 0xd00:     /* WAVG2 */
1784         wrd = (insn >> 12) & 0xf;
1785         rd0 = (insn >> 16) & 0xf;
1786         rd1 = (insn >> 0) & 0xf;
1787         gen_op_iwmmxt_movq_M0_wRn(rd0);
1788         if (insn & (1 << 22)) {
1789             if (insn & (1 << 20))
1790                 gen_op_iwmmxt_avgw1_M0_wRn(rd1);
1791             else
1792                 gen_op_iwmmxt_avgw0_M0_wRn(rd1);
1793         } else {
1794             if (insn & (1 << 20))
1795                 gen_op_iwmmxt_avgb1_M0_wRn(rd1);
1796             else
1797                 gen_op_iwmmxt_avgb0_M0_wRn(rd1);
1798         }
1799         gen_op_iwmmxt_movq_wRn_M0(wrd);
1800         gen_op_iwmmxt_set_mup();
1801         gen_op_iwmmxt_set_cup();
1802         break;
1803     case 0x802: case 0x902: case 0xa02: case 0xb02:     /* WALIGNR */
1804         wrd = (insn >> 12) & 0xf;
1805         rd0 = (insn >> 16) & 0xf;
1806         rd1 = (insn >> 0) & 0xf;
1807         gen_op_iwmmxt_movq_M0_wRn(rd0);
1808         tmp = iwmmxt_load_creg(ARM_IWMMXT_wCGR0 + ((insn >> 20) & 3));
1809         tcg_gen_andi_i32(tmp, tmp, 7);
1810         iwmmxt_load_reg(cpu_V1, rd1);
1811         gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
1812         gen_op_iwmmxt_movq_wRn_M0(wrd);
1813         gen_op_iwmmxt_set_mup();
1814         break;
1815     case 0x601: case 0x605: case 0x609: case 0x60d:     /* TINSR */
1816         if (((insn >> 6) & 3) == 3)
1817             return 1;
1818         rd = (insn >> 12) & 0xf;
1819         wrd = (insn >> 16) & 0xf;
1820         tmp = load_reg(s, rd);
1821         gen_op_iwmmxt_movq_M0_wRn(wrd);
1822         switch ((insn >> 6) & 3) {
1823         case 0:
1824             tmp2 = tcg_constant_i32(0xff);
1825             tmp3 = tcg_constant_i32((insn & 7) << 3);
1826             break;
1827         case 1:
1828             tmp2 = tcg_constant_i32(0xffff);
1829             tmp3 = tcg_constant_i32((insn & 3) << 4);
1830             break;
1831         case 2:
1832             tmp2 = tcg_constant_i32(0xffffffff);
1833             tmp3 = tcg_constant_i32((insn & 1) << 5);
1834             break;
1835         default:
1836             g_assert_not_reached();
1837         }
1838         gen_helper_iwmmxt_insr(cpu_M0, cpu_M0, tmp, tmp2, tmp3);
1839         gen_op_iwmmxt_movq_wRn_M0(wrd);
1840         gen_op_iwmmxt_set_mup();
1841         break;
1842     case 0x107: case 0x507: case 0x907: case 0xd07:     /* TEXTRM */
1843         rd = (insn >> 12) & 0xf;
1844         wrd = (insn >> 16) & 0xf;
1845         if (rd == 15 || ((insn >> 22) & 3) == 3)
1846             return 1;
1847         gen_op_iwmmxt_movq_M0_wRn(wrd);
1848         tmp = tcg_temp_new_i32();
1849         switch ((insn >> 22) & 3) {
1850         case 0:
1851             tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 7) << 3);
1852             tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1853             if (insn & 8) {
1854                 tcg_gen_ext8s_i32(tmp, tmp);
1855             } else {
1856                 tcg_gen_andi_i32(tmp, tmp, 0xff);
1857             }
1858             break;
1859         case 1:
1860             tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 3) << 4);
1861             tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1862             if (insn & 8) {
1863                 tcg_gen_ext16s_i32(tmp, tmp);
1864             } else {
1865                 tcg_gen_andi_i32(tmp, tmp, 0xffff);
1866             }
1867             break;
1868         case 2:
1869             tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 1) << 5);
1870             tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1871             break;
1872         }
1873         store_reg(s, rd, tmp);
1874         break;
1875     case 0x117: case 0x517: case 0x917: case 0xd17:     /* TEXTRC */
1876         if ((insn & 0x000ff008) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1877             return 1;
1878         tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1879         switch ((insn >> 22) & 3) {
1880         case 0:
1881             tcg_gen_shri_i32(tmp, tmp, ((insn & 7) << 2) + 0);
1882             break;
1883         case 1:
1884             tcg_gen_shri_i32(tmp, tmp, ((insn & 3) << 3) + 4);
1885             break;
1886         case 2:
1887             tcg_gen_shri_i32(tmp, tmp, ((insn & 1) << 4) + 12);
1888             break;
1889         }
1890         tcg_gen_shli_i32(tmp, tmp, 28);
1891         gen_set_nzcv(tmp);
1892         break;
1893     case 0x401: case 0x405: case 0x409: case 0x40d:     /* TBCST */
1894         if (((insn >> 6) & 3) == 3)
1895             return 1;
1896         rd = (insn >> 12) & 0xf;
1897         wrd = (insn >> 16) & 0xf;
1898         tmp = load_reg(s, rd);
1899         switch ((insn >> 6) & 3) {
1900         case 0:
1901             gen_helper_iwmmxt_bcstb(cpu_M0, tmp);
1902             break;
1903         case 1:
1904             gen_helper_iwmmxt_bcstw(cpu_M0, tmp);
1905             break;
1906         case 2:
1907             gen_helper_iwmmxt_bcstl(cpu_M0, tmp);
1908             break;
1909         }
1910         gen_op_iwmmxt_movq_wRn_M0(wrd);
1911         gen_op_iwmmxt_set_mup();
1912         break;
1913     case 0x113: case 0x513: case 0x913: case 0xd13:     /* TANDC */
1914         if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1915             return 1;
1916         tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1917         tmp2 = tcg_temp_new_i32();
1918         tcg_gen_mov_i32(tmp2, tmp);
1919         switch ((insn >> 22) & 3) {
1920         case 0:
1921             for (i = 0; i < 7; i ++) {
1922                 tcg_gen_shli_i32(tmp2, tmp2, 4);
1923                 tcg_gen_and_i32(tmp, tmp, tmp2);
1924             }
1925             break;
1926         case 1:
1927             for (i = 0; i < 3; i ++) {
1928                 tcg_gen_shli_i32(tmp2, tmp2, 8);
1929                 tcg_gen_and_i32(tmp, tmp, tmp2);
1930             }
1931             break;
1932         case 2:
1933             tcg_gen_shli_i32(tmp2, tmp2, 16);
1934             tcg_gen_and_i32(tmp, tmp, tmp2);
1935             break;
1936         }
1937         gen_set_nzcv(tmp);
1938         break;
1939     case 0x01c: case 0x41c: case 0x81c: case 0xc1c:     /* WACC */
1940         wrd = (insn >> 12) & 0xf;
1941         rd0 = (insn >> 16) & 0xf;
1942         gen_op_iwmmxt_movq_M0_wRn(rd0);
1943         switch ((insn >> 22) & 3) {
1944         case 0:
1945             gen_helper_iwmmxt_addcb(cpu_M0, cpu_M0);
1946             break;
1947         case 1:
1948             gen_helper_iwmmxt_addcw(cpu_M0, cpu_M0);
1949             break;
1950         case 2:
1951             gen_helper_iwmmxt_addcl(cpu_M0, cpu_M0);
1952             break;
1953         case 3:
1954             return 1;
1955         }
1956         gen_op_iwmmxt_movq_wRn_M0(wrd);
1957         gen_op_iwmmxt_set_mup();
1958         break;
1959     case 0x115: case 0x515: case 0x915: case 0xd15:     /* TORC */
1960         if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1961             return 1;
1962         tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1963         tmp2 = tcg_temp_new_i32();
1964         tcg_gen_mov_i32(tmp2, tmp);
1965         switch ((insn >> 22) & 3) {
1966         case 0:
1967             for (i = 0; i < 7; i ++) {
1968                 tcg_gen_shli_i32(tmp2, tmp2, 4);
1969                 tcg_gen_or_i32(tmp, tmp, tmp2);
1970             }
1971             break;
1972         case 1:
1973             for (i = 0; i < 3; i ++) {
1974                 tcg_gen_shli_i32(tmp2, tmp2, 8);
1975                 tcg_gen_or_i32(tmp, tmp, tmp2);
1976             }
1977             break;
1978         case 2:
1979             tcg_gen_shli_i32(tmp2, tmp2, 16);
1980             tcg_gen_or_i32(tmp, tmp, tmp2);
1981             break;
1982         }
1983         gen_set_nzcv(tmp);
1984         break;
1985     case 0x103: case 0x503: case 0x903: case 0xd03:     /* TMOVMSK */
1986         rd = (insn >> 12) & 0xf;
1987         rd0 = (insn >> 16) & 0xf;
1988         if ((insn & 0xf) != 0 || ((insn >> 22) & 3) == 3)
1989             return 1;
1990         gen_op_iwmmxt_movq_M0_wRn(rd0);
1991         tmp = tcg_temp_new_i32();
1992         switch ((insn >> 22) & 3) {
1993         case 0:
1994             gen_helper_iwmmxt_msbb(tmp, cpu_M0);
1995             break;
1996         case 1:
1997             gen_helper_iwmmxt_msbw(tmp, cpu_M0);
1998             break;
1999         case 2:
2000             gen_helper_iwmmxt_msbl(tmp, cpu_M0);
2001             break;
2002         }
2003         store_reg(s, rd, tmp);
2004         break;
2005     case 0x106: case 0x306: case 0x506: case 0x706:     /* WCMPGT */
2006     case 0x906: case 0xb06: case 0xd06: case 0xf06:
2007         wrd = (insn >> 12) & 0xf;
2008         rd0 = (insn >> 16) & 0xf;
2009         rd1 = (insn >> 0) & 0xf;
2010         gen_op_iwmmxt_movq_M0_wRn(rd0);
2011         switch ((insn >> 22) & 3) {
2012         case 0:
2013             if (insn & (1 << 21))
2014                 gen_op_iwmmxt_cmpgtsb_M0_wRn(rd1);
2015             else
2016                 gen_op_iwmmxt_cmpgtub_M0_wRn(rd1);
2017             break;
2018         case 1:
2019             if (insn & (1 << 21))
2020                 gen_op_iwmmxt_cmpgtsw_M0_wRn(rd1);
2021             else
2022                 gen_op_iwmmxt_cmpgtuw_M0_wRn(rd1);
2023             break;
2024         case 2:
2025             if (insn & (1 << 21))
2026                 gen_op_iwmmxt_cmpgtsl_M0_wRn(rd1);
2027             else
2028                 gen_op_iwmmxt_cmpgtul_M0_wRn(rd1);
2029             break;
2030         case 3:
2031             return 1;
2032         }
2033         gen_op_iwmmxt_movq_wRn_M0(wrd);
2034         gen_op_iwmmxt_set_mup();
2035         gen_op_iwmmxt_set_cup();
2036         break;
2037     case 0x00e: case 0x20e: case 0x40e: case 0x60e:     /* WUNPCKEL */
2038     case 0x80e: case 0xa0e: case 0xc0e: case 0xe0e:
2039         wrd = (insn >> 12) & 0xf;
2040         rd0 = (insn >> 16) & 0xf;
2041         gen_op_iwmmxt_movq_M0_wRn(rd0);
2042         switch ((insn >> 22) & 3) {
2043         case 0:
2044             if (insn & (1 << 21))
2045                 gen_op_iwmmxt_unpacklsb_M0();
2046             else
2047                 gen_op_iwmmxt_unpacklub_M0();
2048             break;
2049         case 1:
2050             if (insn & (1 << 21))
2051                 gen_op_iwmmxt_unpacklsw_M0();
2052             else
2053                 gen_op_iwmmxt_unpackluw_M0();
2054             break;
2055         case 2:
2056             if (insn & (1 << 21))
2057                 gen_op_iwmmxt_unpacklsl_M0();
2058             else
2059                 gen_op_iwmmxt_unpacklul_M0();
2060             break;
2061         case 3:
2062             return 1;
2063         }
2064         gen_op_iwmmxt_movq_wRn_M0(wrd);
2065         gen_op_iwmmxt_set_mup();
2066         gen_op_iwmmxt_set_cup();
2067         break;
2068     case 0x00c: case 0x20c: case 0x40c: case 0x60c:     /* WUNPCKEH */
2069     case 0x80c: case 0xa0c: case 0xc0c: case 0xe0c:
2070         wrd = (insn >> 12) & 0xf;
2071         rd0 = (insn >> 16) & 0xf;
2072         gen_op_iwmmxt_movq_M0_wRn(rd0);
2073         switch ((insn >> 22) & 3) {
2074         case 0:
2075             if (insn & (1 << 21))
2076                 gen_op_iwmmxt_unpackhsb_M0();
2077             else
2078                 gen_op_iwmmxt_unpackhub_M0();
2079             break;
2080         case 1:
2081             if (insn & (1 << 21))
2082                 gen_op_iwmmxt_unpackhsw_M0();
2083             else
2084                 gen_op_iwmmxt_unpackhuw_M0();
2085             break;
2086         case 2:
2087             if (insn & (1 << 21))
2088                 gen_op_iwmmxt_unpackhsl_M0();
2089             else
2090                 gen_op_iwmmxt_unpackhul_M0();
2091             break;
2092         case 3:
2093             return 1;
2094         }
2095         gen_op_iwmmxt_movq_wRn_M0(wrd);
2096         gen_op_iwmmxt_set_mup();
2097         gen_op_iwmmxt_set_cup();
2098         break;
2099     case 0x204: case 0x604: case 0xa04: case 0xe04:     /* WSRL */
2100     case 0x214: case 0x614: case 0xa14: case 0xe14:
2101         if (((insn >> 22) & 3) == 0)
2102             return 1;
2103         wrd = (insn >> 12) & 0xf;
2104         rd0 = (insn >> 16) & 0xf;
2105         gen_op_iwmmxt_movq_M0_wRn(rd0);
2106         tmp = tcg_temp_new_i32();
2107         if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2108             return 1;
2109         }
2110         switch ((insn >> 22) & 3) {
2111         case 1:
2112             gen_helper_iwmmxt_srlw(cpu_M0, tcg_env, cpu_M0, tmp);
2113             break;
2114         case 2:
2115             gen_helper_iwmmxt_srll(cpu_M0, tcg_env, cpu_M0, tmp);
2116             break;
2117         case 3:
2118             gen_helper_iwmmxt_srlq(cpu_M0, tcg_env, cpu_M0, tmp);
2119             break;
2120         }
2121         gen_op_iwmmxt_movq_wRn_M0(wrd);
2122         gen_op_iwmmxt_set_mup();
2123         gen_op_iwmmxt_set_cup();
2124         break;
2125     case 0x004: case 0x404: case 0x804: case 0xc04:     /* WSRA */
2126     case 0x014: case 0x414: case 0x814: case 0xc14:
2127         if (((insn >> 22) & 3) == 0)
2128             return 1;
2129         wrd = (insn >> 12) & 0xf;
2130         rd0 = (insn >> 16) & 0xf;
2131         gen_op_iwmmxt_movq_M0_wRn(rd0);
2132         tmp = tcg_temp_new_i32();
2133         if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2134             return 1;
2135         }
2136         switch ((insn >> 22) & 3) {
2137         case 1:
2138             gen_helper_iwmmxt_sraw(cpu_M0, tcg_env, cpu_M0, tmp);
2139             break;
2140         case 2:
2141             gen_helper_iwmmxt_sral(cpu_M0, tcg_env, cpu_M0, tmp);
2142             break;
2143         case 3:
2144             gen_helper_iwmmxt_sraq(cpu_M0, tcg_env, cpu_M0, tmp);
2145             break;
2146         }
2147         gen_op_iwmmxt_movq_wRn_M0(wrd);
2148         gen_op_iwmmxt_set_mup();
2149         gen_op_iwmmxt_set_cup();
2150         break;
2151     case 0x104: case 0x504: case 0x904: case 0xd04:     /* WSLL */
2152     case 0x114: case 0x514: case 0x914: case 0xd14:
2153         if (((insn >> 22) & 3) == 0)
2154             return 1;
2155         wrd = (insn >> 12) & 0xf;
2156         rd0 = (insn >> 16) & 0xf;
2157         gen_op_iwmmxt_movq_M0_wRn(rd0);
2158         tmp = tcg_temp_new_i32();
2159         if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2160             return 1;
2161         }
2162         switch ((insn >> 22) & 3) {
2163         case 1:
2164             gen_helper_iwmmxt_sllw(cpu_M0, tcg_env, cpu_M0, tmp);
2165             break;
2166         case 2:
2167             gen_helper_iwmmxt_slll(cpu_M0, tcg_env, cpu_M0, tmp);
2168             break;
2169         case 3:
2170             gen_helper_iwmmxt_sllq(cpu_M0, tcg_env, cpu_M0, tmp);
2171             break;
2172         }
2173         gen_op_iwmmxt_movq_wRn_M0(wrd);
2174         gen_op_iwmmxt_set_mup();
2175         gen_op_iwmmxt_set_cup();
2176         break;
2177     case 0x304: case 0x704: case 0xb04: case 0xf04:     /* WROR */
2178     case 0x314: case 0x714: case 0xb14: case 0xf14:
2179         if (((insn >> 22) & 3) == 0)
2180             return 1;
2181         wrd = (insn >> 12) & 0xf;
2182         rd0 = (insn >> 16) & 0xf;
2183         gen_op_iwmmxt_movq_M0_wRn(rd0);
2184         tmp = tcg_temp_new_i32();
2185         switch ((insn >> 22) & 3) {
2186         case 1:
2187             if (gen_iwmmxt_shift(insn, 0xf, tmp)) {
2188                 return 1;
2189             }
2190             gen_helper_iwmmxt_rorw(cpu_M0, tcg_env, cpu_M0, tmp);
2191             break;
2192         case 2:
2193             if (gen_iwmmxt_shift(insn, 0x1f, tmp)) {
2194                 return 1;
2195             }
2196             gen_helper_iwmmxt_rorl(cpu_M0, tcg_env, cpu_M0, tmp);
2197             break;
2198         case 3:
2199             if (gen_iwmmxt_shift(insn, 0x3f, tmp)) {
2200                 return 1;
2201             }
2202             gen_helper_iwmmxt_rorq(cpu_M0, tcg_env, cpu_M0, tmp);
2203             break;
2204         }
2205         gen_op_iwmmxt_movq_wRn_M0(wrd);
2206         gen_op_iwmmxt_set_mup();
2207         gen_op_iwmmxt_set_cup();
2208         break;
2209     case 0x116: case 0x316: case 0x516: case 0x716:     /* WMIN */
2210     case 0x916: case 0xb16: case 0xd16: case 0xf16:
2211         wrd = (insn >> 12) & 0xf;
2212         rd0 = (insn >> 16) & 0xf;
2213         rd1 = (insn >> 0) & 0xf;
2214         gen_op_iwmmxt_movq_M0_wRn(rd0);
2215         switch ((insn >> 22) & 3) {
2216         case 0:
2217             if (insn & (1 << 21))
2218                 gen_op_iwmmxt_minsb_M0_wRn(rd1);
2219             else
2220                 gen_op_iwmmxt_minub_M0_wRn(rd1);
2221             break;
2222         case 1:
2223             if (insn & (1 << 21))
2224                 gen_op_iwmmxt_minsw_M0_wRn(rd1);
2225             else
2226                 gen_op_iwmmxt_minuw_M0_wRn(rd1);
2227             break;
2228         case 2:
2229             if (insn & (1 << 21))
2230                 gen_op_iwmmxt_minsl_M0_wRn(rd1);
2231             else
2232                 gen_op_iwmmxt_minul_M0_wRn(rd1);
2233             break;
2234         case 3:
2235             return 1;
2236         }
2237         gen_op_iwmmxt_movq_wRn_M0(wrd);
2238         gen_op_iwmmxt_set_mup();
2239         break;
2240     case 0x016: case 0x216: case 0x416: case 0x616:     /* WMAX */
2241     case 0x816: case 0xa16: case 0xc16: case 0xe16:
2242         wrd = (insn >> 12) & 0xf;
2243         rd0 = (insn >> 16) & 0xf;
2244         rd1 = (insn >> 0) & 0xf;
2245         gen_op_iwmmxt_movq_M0_wRn(rd0);
2246         switch ((insn >> 22) & 3) {
2247         case 0:
2248             if (insn & (1 << 21))
2249                 gen_op_iwmmxt_maxsb_M0_wRn(rd1);
2250             else
2251                 gen_op_iwmmxt_maxub_M0_wRn(rd1);
2252             break;
2253         case 1:
2254             if (insn & (1 << 21))
2255                 gen_op_iwmmxt_maxsw_M0_wRn(rd1);
2256             else
2257                 gen_op_iwmmxt_maxuw_M0_wRn(rd1);
2258             break;
2259         case 2:
2260             if (insn & (1 << 21))
2261                 gen_op_iwmmxt_maxsl_M0_wRn(rd1);
2262             else
2263                 gen_op_iwmmxt_maxul_M0_wRn(rd1);
2264             break;
2265         case 3:
2266             return 1;
2267         }
2268         gen_op_iwmmxt_movq_wRn_M0(wrd);
2269         gen_op_iwmmxt_set_mup();
2270         break;
2271     case 0x002: case 0x102: case 0x202: case 0x302:     /* WALIGNI */
2272     case 0x402: case 0x502: case 0x602: case 0x702:
2273         wrd = (insn >> 12) & 0xf;
2274         rd0 = (insn >> 16) & 0xf;
2275         rd1 = (insn >> 0) & 0xf;
2276         gen_op_iwmmxt_movq_M0_wRn(rd0);
2277         iwmmxt_load_reg(cpu_V1, rd1);
2278         gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1,
2279                                 tcg_constant_i32((insn >> 20) & 3));
2280         gen_op_iwmmxt_movq_wRn_M0(wrd);
2281         gen_op_iwmmxt_set_mup();
2282         break;
2283     case 0x01a: case 0x11a: case 0x21a: case 0x31a:     /* WSUB */
2284     case 0x41a: case 0x51a: case 0x61a: case 0x71a:
2285     case 0x81a: case 0x91a: case 0xa1a: case 0xb1a:
2286     case 0xc1a: case 0xd1a: case 0xe1a: case 0xf1a:
2287         wrd = (insn >> 12) & 0xf;
2288         rd0 = (insn >> 16) & 0xf;
2289         rd1 = (insn >> 0) & 0xf;
2290         gen_op_iwmmxt_movq_M0_wRn(rd0);
2291         switch ((insn >> 20) & 0xf) {
2292         case 0x0:
2293             gen_op_iwmmxt_subnb_M0_wRn(rd1);
2294             break;
2295         case 0x1:
2296             gen_op_iwmmxt_subub_M0_wRn(rd1);
2297             break;
2298         case 0x3:
2299             gen_op_iwmmxt_subsb_M0_wRn(rd1);
2300             break;
2301         case 0x4:
2302             gen_op_iwmmxt_subnw_M0_wRn(rd1);
2303             break;
2304         case 0x5:
2305             gen_op_iwmmxt_subuw_M0_wRn(rd1);
2306             break;
2307         case 0x7:
2308             gen_op_iwmmxt_subsw_M0_wRn(rd1);
2309             break;
2310         case 0x8:
2311             gen_op_iwmmxt_subnl_M0_wRn(rd1);
2312             break;
2313         case 0x9:
2314             gen_op_iwmmxt_subul_M0_wRn(rd1);
2315             break;
2316         case 0xb:
2317             gen_op_iwmmxt_subsl_M0_wRn(rd1);
2318             break;
2319         default:
2320             return 1;
2321         }
2322         gen_op_iwmmxt_movq_wRn_M0(wrd);
2323         gen_op_iwmmxt_set_mup();
2324         gen_op_iwmmxt_set_cup();
2325         break;
2326     case 0x01e: case 0x11e: case 0x21e: case 0x31e:     /* WSHUFH */
2327     case 0x41e: case 0x51e: case 0x61e: case 0x71e:
2328     case 0x81e: case 0x91e: case 0xa1e: case 0xb1e:
2329     case 0xc1e: case 0xd1e: case 0xe1e: case 0xf1e:
2330         wrd = (insn >> 12) & 0xf;
2331         rd0 = (insn >> 16) & 0xf;
2332         gen_op_iwmmxt_movq_M0_wRn(rd0);
2333         tmp = tcg_constant_i32(((insn >> 16) & 0xf0) | (insn & 0x0f));
2334         gen_helper_iwmmxt_shufh(cpu_M0, tcg_env, cpu_M0, tmp);
2335         gen_op_iwmmxt_movq_wRn_M0(wrd);
2336         gen_op_iwmmxt_set_mup();
2337         gen_op_iwmmxt_set_cup();
2338         break;
2339     case 0x018: case 0x118: case 0x218: case 0x318:     /* WADD */
2340     case 0x418: case 0x518: case 0x618: case 0x718:
2341     case 0x818: case 0x918: case 0xa18: case 0xb18:
2342     case 0xc18: case 0xd18: case 0xe18: case 0xf18:
2343         wrd = (insn >> 12) & 0xf;
2344         rd0 = (insn >> 16) & 0xf;
2345         rd1 = (insn >> 0) & 0xf;
2346         gen_op_iwmmxt_movq_M0_wRn(rd0);
2347         switch ((insn >> 20) & 0xf) {
2348         case 0x0:
2349             gen_op_iwmmxt_addnb_M0_wRn(rd1);
2350             break;
2351         case 0x1:
2352             gen_op_iwmmxt_addub_M0_wRn(rd1);
2353             break;
2354         case 0x3:
2355             gen_op_iwmmxt_addsb_M0_wRn(rd1);
2356             break;
2357         case 0x4:
2358             gen_op_iwmmxt_addnw_M0_wRn(rd1);
2359             break;
2360         case 0x5:
2361             gen_op_iwmmxt_adduw_M0_wRn(rd1);
2362             break;
2363         case 0x7:
2364             gen_op_iwmmxt_addsw_M0_wRn(rd1);
2365             break;
2366         case 0x8:
2367             gen_op_iwmmxt_addnl_M0_wRn(rd1);
2368             break;
2369         case 0x9:
2370             gen_op_iwmmxt_addul_M0_wRn(rd1);
2371             break;
2372         case 0xb:
2373             gen_op_iwmmxt_addsl_M0_wRn(rd1);
2374             break;
2375         default:
2376             return 1;
2377         }
2378         gen_op_iwmmxt_movq_wRn_M0(wrd);
2379         gen_op_iwmmxt_set_mup();
2380         gen_op_iwmmxt_set_cup();
2381         break;
2382     case 0x008: case 0x108: case 0x208: case 0x308:     /* WPACK */
2383     case 0x408: case 0x508: case 0x608: case 0x708:
2384     case 0x808: case 0x908: case 0xa08: case 0xb08:
2385     case 0xc08: case 0xd08: case 0xe08: case 0xf08:
2386         if (!(insn & (1 << 20)) || ((insn >> 22) & 3) == 0)
2387             return 1;
2388         wrd = (insn >> 12) & 0xf;
2389         rd0 = (insn >> 16) & 0xf;
2390         rd1 = (insn >> 0) & 0xf;
2391         gen_op_iwmmxt_movq_M0_wRn(rd0);
2392         switch ((insn >> 22) & 3) {
2393         case 1:
2394             if (insn & (1 << 21))
2395                 gen_op_iwmmxt_packsw_M0_wRn(rd1);
2396             else
2397                 gen_op_iwmmxt_packuw_M0_wRn(rd1);
2398             break;
2399         case 2:
2400             if (insn & (1 << 21))
2401                 gen_op_iwmmxt_packsl_M0_wRn(rd1);
2402             else
2403                 gen_op_iwmmxt_packul_M0_wRn(rd1);
2404             break;
2405         case 3:
2406             if (insn & (1 << 21))
2407                 gen_op_iwmmxt_packsq_M0_wRn(rd1);
2408             else
2409                 gen_op_iwmmxt_packuq_M0_wRn(rd1);
2410             break;
2411         }
2412         gen_op_iwmmxt_movq_wRn_M0(wrd);
2413         gen_op_iwmmxt_set_mup();
2414         gen_op_iwmmxt_set_cup();
2415         break;
2416     case 0x201: case 0x203: case 0x205: case 0x207:
2417     case 0x209: case 0x20b: case 0x20d: case 0x20f:
2418     case 0x211: case 0x213: case 0x215: case 0x217:
2419     case 0x219: case 0x21b: case 0x21d: case 0x21f:
2420         wrd = (insn >> 5) & 0xf;
2421         rd0 = (insn >> 12) & 0xf;
2422         rd1 = (insn >> 0) & 0xf;
2423         if (rd0 == 0xf || rd1 == 0xf)
2424             return 1;
2425         gen_op_iwmmxt_movq_M0_wRn(wrd);
2426         tmp = load_reg(s, rd0);
2427         tmp2 = load_reg(s, rd1);
2428         switch ((insn >> 16) & 0xf) {
2429         case 0x0:                                       /* TMIA */
2430             gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2431             break;
2432         case 0x8:                                       /* TMIAPH */
2433             gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2434             break;
2435         case 0xc: case 0xd: case 0xe: case 0xf:                 /* TMIAxy */
2436             if (insn & (1 << 16))
2437                 tcg_gen_shri_i32(tmp, tmp, 16);
2438             if (insn & (1 << 17))
2439                 tcg_gen_shri_i32(tmp2, tmp2, 16);
2440             gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2441             break;
2442         default:
2443             return 1;
2444         }
2445         gen_op_iwmmxt_movq_wRn_M0(wrd);
2446         gen_op_iwmmxt_set_mup();
2447         break;
2448     default:
2449         return 1;
2450     }
2451 
2452     return 0;
2453 }
2454 
2455 /* Disassemble an XScale DSP instruction.  Returns nonzero if an error occurred
2456    (ie. an undefined instruction).  */
2457 static int disas_dsp_insn(DisasContext *s, uint32_t insn)
2458 {
2459     int acc, rd0, rd1, rdhi, rdlo;
2460     TCGv_i32 tmp, tmp2;
2461 
2462     if ((insn & 0x0ff00f10) == 0x0e200010) {
2463         /* Multiply with Internal Accumulate Format */
2464         rd0 = (insn >> 12) & 0xf;
2465         rd1 = insn & 0xf;
2466         acc = (insn >> 5) & 7;
2467 
2468         if (acc != 0)
2469             return 1;
2470 
2471         tmp = load_reg(s, rd0);
2472         tmp2 = load_reg(s, rd1);
2473         switch ((insn >> 16) & 0xf) {
2474         case 0x0:                                       /* MIA */
2475             gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2476             break;
2477         case 0x8:                                       /* MIAPH */
2478             gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2479             break;
2480         case 0xc:                                       /* MIABB */
2481         case 0xd:                                       /* MIABT */
2482         case 0xe:                                       /* MIATB */
2483         case 0xf:                                       /* MIATT */
2484             if (insn & (1 << 16))
2485                 tcg_gen_shri_i32(tmp, tmp, 16);
2486             if (insn & (1 << 17))
2487                 tcg_gen_shri_i32(tmp2, tmp2, 16);
2488             gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2489             break;
2490         default:
2491             return 1;
2492         }
2493 
2494         gen_op_iwmmxt_movq_wRn_M0(acc);
2495         return 0;
2496     }
2497 
2498     if ((insn & 0x0fe00ff8) == 0x0c400000) {
2499         /* Internal Accumulator Access Format */
2500         rdhi = (insn >> 16) & 0xf;
2501         rdlo = (insn >> 12) & 0xf;
2502         acc = insn & 7;
2503 
2504         if (acc != 0)
2505             return 1;
2506 
2507         if (insn & ARM_CP_RW_BIT) {                     /* MRA */
2508             iwmmxt_load_reg(cpu_V0, acc);
2509             tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
2510             tcg_gen_extrh_i64_i32(cpu_R[rdhi], cpu_V0);
2511             tcg_gen_andi_i32(cpu_R[rdhi], cpu_R[rdhi], (1 << (40 - 32)) - 1);
2512         } else {                                        /* MAR */
2513             tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
2514             iwmmxt_store_reg(cpu_V0, acc);
2515         }
2516         return 0;
2517     }
2518 
2519     return 1;
2520 }
2521 
2522 static void gen_goto_ptr(void)
2523 {
2524     tcg_gen_lookup_and_goto_ptr();
2525 }
2526 
2527 /* This will end the TB but doesn't guarantee we'll return to
2528  * cpu_loop_exec. Any live exit_requests will be processed as we
2529  * enter the next TB.
2530  */
2531 static void gen_goto_tb(DisasContext *s, int n, target_long diff)
2532 {
2533     if (translator_use_goto_tb(&s->base, s->pc_curr + diff)) {
2534         /*
2535          * For pcrel, the pc must always be up-to-date on entry to
2536          * the linked TB, so that it can use simple additions for all
2537          * further adjustments.  For !pcrel, the linked TB is compiled
2538          * to know its full virtual address, so we can delay the
2539          * update to pc to the unlinked path.  A long chain of links
2540          * can thus avoid many updates to the PC.
2541          */
2542         if (tb_cflags(s->base.tb) & CF_PCREL) {
2543             gen_update_pc(s, diff);
2544             tcg_gen_goto_tb(n);
2545         } else {
2546             tcg_gen_goto_tb(n);
2547             gen_update_pc(s, diff);
2548         }
2549         tcg_gen_exit_tb(s->base.tb, n);
2550     } else {
2551         gen_update_pc(s, diff);
2552         gen_goto_ptr();
2553     }
2554     s->base.is_jmp = DISAS_NORETURN;
2555 }
2556 
2557 /* Jump, specifying which TB number to use if we gen_goto_tb() */
2558 static void gen_jmp_tb(DisasContext *s, target_long diff, int tbno)
2559 {
2560     if (unlikely(s->ss_active)) {
2561         /* An indirect jump so that we still trigger the debug exception.  */
2562         gen_update_pc(s, diff);
2563         s->base.is_jmp = DISAS_JUMP;
2564         return;
2565     }
2566     switch (s->base.is_jmp) {
2567     case DISAS_NEXT:
2568     case DISAS_TOO_MANY:
2569     case DISAS_NORETURN:
2570         /*
2571          * The normal case: just go to the destination TB.
2572          * NB: NORETURN happens if we generate code like
2573          *    gen_brcondi(l);
2574          *    gen_jmp();
2575          *    gen_set_label(l);
2576          *    gen_jmp();
2577          * on the second call to gen_jmp().
2578          */
2579         gen_goto_tb(s, tbno, diff);
2580         break;
2581     case DISAS_UPDATE_NOCHAIN:
2582     case DISAS_UPDATE_EXIT:
2583         /*
2584          * We already decided we're leaving the TB for some other reason.
2585          * Avoid using goto_tb so we really do exit back to the main loop
2586          * and don't chain to another TB.
2587          */
2588         gen_update_pc(s, diff);
2589         gen_goto_ptr();
2590         s->base.is_jmp = DISAS_NORETURN;
2591         break;
2592     default:
2593         /*
2594          * We shouldn't be emitting code for a jump and also have
2595          * is_jmp set to one of the special cases like DISAS_SWI.
2596          */
2597         g_assert_not_reached();
2598     }
2599 }
2600 
2601 static inline void gen_jmp(DisasContext *s, target_long diff)
2602 {
2603     gen_jmp_tb(s, diff, 0);
2604 }
2605 
2606 static inline void gen_mulxy(TCGv_i32 t0, TCGv_i32 t1, int x, int y)
2607 {
2608     if (x)
2609         tcg_gen_sari_i32(t0, t0, 16);
2610     else
2611         gen_sxth(t0);
2612     if (y)
2613         tcg_gen_sari_i32(t1, t1, 16);
2614     else
2615         gen_sxth(t1);
2616     tcg_gen_mul_i32(t0, t0, t1);
2617 }
2618 
2619 /* Return the mask of PSR bits set by a MSR instruction.  */
2620 static uint32_t msr_mask(DisasContext *s, int flags, int spsr)
2621 {
2622     uint32_t mask = 0;
2623 
2624     if (flags & (1 << 0)) {
2625         mask |= 0xff;
2626     }
2627     if (flags & (1 << 1)) {
2628         mask |= 0xff00;
2629     }
2630     if (flags & (1 << 2)) {
2631         mask |= 0xff0000;
2632     }
2633     if (flags & (1 << 3)) {
2634         mask |= 0xff000000;
2635     }
2636 
2637     /* Mask out undefined and reserved bits.  */
2638     mask &= aarch32_cpsr_valid_mask(s->features, s->isar);
2639 
2640     /* Mask out execution state.  */
2641     if (!spsr) {
2642         mask &= ~CPSR_EXEC;
2643     }
2644 
2645     /* Mask out privileged bits.  */
2646     if (IS_USER(s)) {
2647         mask &= CPSR_USER;
2648     }
2649     return mask;
2650 }
2651 
2652 /* Returns nonzero if access to the PSR is not permitted. Marks t0 as dead. */
2653 static int gen_set_psr(DisasContext *s, uint32_t mask, int spsr, TCGv_i32 t0)
2654 {
2655     TCGv_i32 tmp;
2656     if (spsr) {
2657         /* ??? This is also undefined in system mode.  */
2658         if (IS_USER(s))
2659             return 1;
2660 
2661         tmp = load_cpu_field(spsr);
2662         tcg_gen_andi_i32(tmp, tmp, ~mask);
2663         tcg_gen_andi_i32(t0, t0, mask);
2664         tcg_gen_or_i32(tmp, tmp, t0);
2665         store_cpu_field(tmp, spsr);
2666     } else {
2667         gen_set_cpsr(t0, mask);
2668     }
2669     gen_lookup_tb(s);
2670     return 0;
2671 }
2672 
2673 /* Returns nonzero if access to the PSR is not permitted.  */
2674 static int gen_set_psr_im(DisasContext *s, uint32_t mask, int spsr, uint32_t val)
2675 {
2676     TCGv_i32 tmp;
2677     tmp = tcg_temp_new_i32();
2678     tcg_gen_movi_i32(tmp, val);
2679     return gen_set_psr(s, mask, spsr, tmp);
2680 }
2681 
2682 static bool msr_banked_access_decode(DisasContext *s, int r, int sysm, int rn,
2683                                      int *tgtmode, int *regno)
2684 {
2685     /* Decode the r and sysm fields of MSR/MRS banked accesses into
2686      * the target mode and register number, and identify the various
2687      * unpredictable cases.
2688      * MSR (banked) and MRS (banked) are CONSTRAINED UNPREDICTABLE if:
2689      *  + executed in user mode
2690      *  + using R15 as the src/dest register
2691      *  + accessing an unimplemented register
2692      *  + accessing a register that's inaccessible at current PL/security state*
2693      *  + accessing a register that you could access with a different insn
2694      * We choose to UNDEF in all these cases.
2695      * Since we don't know which of the various AArch32 modes we are in
2696      * we have to defer some checks to runtime.
2697      * Accesses to Monitor mode registers from Secure EL1 (which implies
2698      * that EL3 is AArch64) must trap to EL3.
2699      *
2700      * If the access checks fail this function will emit code to take
2701      * an exception and return false. Otherwise it will return true,
2702      * and set *tgtmode and *regno appropriately.
2703      */
2704     /* These instructions are present only in ARMv8, or in ARMv7 with the
2705      * Virtualization Extensions.
2706      */
2707     if (!arm_dc_feature(s, ARM_FEATURE_V8) &&
2708         !arm_dc_feature(s, ARM_FEATURE_EL2)) {
2709         goto undef;
2710     }
2711 
2712     if (IS_USER(s) || rn == 15) {
2713         goto undef;
2714     }
2715 
2716     /* The table in the v8 ARM ARM section F5.2.3 describes the encoding
2717      * of registers into (r, sysm).
2718      */
2719     if (r) {
2720         /* SPSRs for other modes */
2721         switch (sysm) {
2722         case 0xe: /* SPSR_fiq */
2723             *tgtmode = ARM_CPU_MODE_FIQ;
2724             break;
2725         case 0x10: /* SPSR_irq */
2726             *tgtmode = ARM_CPU_MODE_IRQ;
2727             break;
2728         case 0x12: /* SPSR_svc */
2729             *tgtmode = ARM_CPU_MODE_SVC;
2730             break;
2731         case 0x14: /* SPSR_abt */
2732             *tgtmode = ARM_CPU_MODE_ABT;
2733             break;
2734         case 0x16: /* SPSR_und */
2735             *tgtmode = ARM_CPU_MODE_UND;
2736             break;
2737         case 0x1c: /* SPSR_mon */
2738             *tgtmode = ARM_CPU_MODE_MON;
2739             break;
2740         case 0x1e: /* SPSR_hyp */
2741             *tgtmode = ARM_CPU_MODE_HYP;
2742             break;
2743         default: /* unallocated */
2744             goto undef;
2745         }
2746         /* We arbitrarily assign SPSR a register number of 16. */
2747         *regno = 16;
2748     } else {
2749         /* general purpose registers for other modes */
2750         switch (sysm) {
2751         case 0x0 ... 0x6:   /* 0b00xxx : r8_usr ... r14_usr */
2752             *tgtmode = ARM_CPU_MODE_USR;
2753             *regno = sysm + 8;
2754             break;
2755         case 0x8 ... 0xe:   /* 0b01xxx : r8_fiq ... r14_fiq */
2756             *tgtmode = ARM_CPU_MODE_FIQ;
2757             *regno = sysm;
2758             break;
2759         case 0x10 ... 0x11: /* 0b1000x : r14_irq, r13_irq */
2760             *tgtmode = ARM_CPU_MODE_IRQ;
2761             *regno = sysm & 1 ? 13 : 14;
2762             break;
2763         case 0x12 ... 0x13: /* 0b1001x : r14_svc, r13_svc */
2764             *tgtmode = ARM_CPU_MODE_SVC;
2765             *regno = sysm & 1 ? 13 : 14;
2766             break;
2767         case 0x14 ... 0x15: /* 0b1010x : r14_abt, r13_abt */
2768             *tgtmode = ARM_CPU_MODE_ABT;
2769             *regno = sysm & 1 ? 13 : 14;
2770             break;
2771         case 0x16 ... 0x17: /* 0b1011x : r14_und, r13_und */
2772             *tgtmode = ARM_CPU_MODE_UND;
2773             *regno = sysm & 1 ? 13 : 14;
2774             break;
2775         case 0x1c ... 0x1d: /* 0b1110x : r14_mon, r13_mon */
2776             *tgtmode = ARM_CPU_MODE_MON;
2777             *regno = sysm & 1 ? 13 : 14;
2778             break;
2779         case 0x1e ... 0x1f: /* 0b1111x : elr_hyp, r13_hyp */
2780             *tgtmode = ARM_CPU_MODE_HYP;
2781             /* Arbitrarily pick 17 for ELR_Hyp (which is not a banked LR!) */
2782             *regno = sysm & 1 ? 13 : 17;
2783             break;
2784         default: /* unallocated */
2785             goto undef;
2786         }
2787     }
2788 
2789     /* Catch the 'accessing inaccessible register' cases we can detect
2790      * at translate time.
2791      */
2792     switch (*tgtmode) {
2793     case ARM_CPU_MODE_MON:
2794         if (!arm_dc_feature(s, ARM_FEATURE_EL3) || s->ns) {
2795             goto undef;
2796         }
2797         if (s->current_el == 1) {
2798             /* If we're in Secure EL1 (which implies that EL3 is AArch64)
2799              * then accesses to Mon registers trap to Secure EL2, if it exists,
2800              * otherwise EL3.
2801              */
2802             TCGv_i32 tcg_el;
2803 
2804             if (arm_dc_feature(s, ARM_FEATURE_AARCH64) &&
2805                 dc_isar_feature(aa64_sel2, s)) {
2806                 /* Target EL is EL<3 minus SCR_EL3.EEL2> */
2807                 tcg_el = load_cpu_field_low32(cp15.scr_el3);
2808                 tcg_gen_sextract_i32(tcg_el, tcg_el, ctz32(SCR_EEL2), 1);
2809                 tcg_gen_addi_i32(tcg_el, tcg_el, 3);
2810             } else {
2811                 tcg_el = tcg_constant_i32(3);
2812             }
2813 
2814             gen_exception_insn_el_v(s, 0, EXCP_UDEF,
2815                                     syn_uncategorized(), tcg_el);
2816             return false;
2817         }
2818         break;
2819     case ARM_CPU_MODE_HYP:
2820         /*
2821          * r13_hyp can only be accessed from Monitor mode, and so we
2822          * can forbid accesses from EL2 or below.
2823          * elr_hyp can be accessed also from Hyp mode, so forbid
2824          * accesses from EL0 or EL1.
2825          * SPSR_hyp is supposed to be in the same category as r13_hyp
2826          * and UNPREDICTABLE if accessed from anything except Monitor
2827          * mode. However there is some real-world code that will do
2828          * it because at least some hardware happens to permit the
2829          * access. (Notably a standard Cortex-R52 startup code fragment
2830          * does this.) So we permit SPSR_hyp from Hyp mode also, to allow
2831          * this (incorrect) guest code to run.
2832          */
2833         if (!arm_dc_feature(s, ARM_FEATURE_EL2) || s->current_el < 2
2834             || (s->current_el < 3 && *regno != 16 && *regno != 17)) {
2835             goto undef;
2836         }
2837         break;
2838     default:
2839         break;
2840     }
2841 
2842     return true;
2843 
2844 undef:
2845     /* If we get here then some access check did not pass */
2846     gen_exception_insn(s, 0, EXCP_UDEF, syn_uncategorized());
2847     return false;
2848 }
2849 
2850 static void gen_msr_banked(DisasContext *s, int r, int sysm, int rn)
2851 {
2852     TCGv_i32 tcg_reg;
2853     int tgtmode = 0, regno = 0;
2854 
2855     if (!msr_banked_access_decode(s, r, sysm, rn, &tgtmode, &regno)) {
2856         return;
2857     }
2858 
2859     /* Sync state because msr_banked() can raise exceptions */
2860     gen_set_condexec(s);
2861     gen_update_pc(s, 0);
2862     tcg_reg = load_reg(s, rn);
2863     gen_helper_msr_banked(tcg_env, tcg_reg,
2864                           tcg_constant_i32(tgtmode),
2865                           tcg_constant_i32(regno));
2866     s->base.is_jmp = DISAS_UPDATE_EXIT;
2867 }
2868 
2869 static void gen_mrs_banked(DisasContext *s, int r, int sysm, int rn)
2870 {
2871     TCGv_i32 tcg_reg;
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 mrs_banked() can raise exceptions */
2879     gen_set_condexec(s);
2880     gen_update_pc(s, 0);
2881     tcg_reg = tcg_temp_new_i32();
2882     gen_helper_mrs_banked(tcg_reg, tcg_env,
2883                           tcg_constant_i32(tgtmode),
2884                           tcg_constant_i32(regno));
2885     store_reg(s, rn, tcg_reg);
2886     s->base.is_jmp = DISAS_UPDATE_EXIT;
2887 }
2888 
2889 /* Store value to PC as for an exception return (ie don't
2890  * mask bits). The subsequent call to gen_helper_cpsr_write_eret()
2891  * will do the masking based on the new value of the Thumb bit.
2892  */
2893 static void store_pc_exc_ret(DisasContext *s, TCGv_i32 pc)
2894 {
2895     tcg_gen_mov_i32(cpu_R[15], pc);
2896 }
2897 
2898 /* Generate a v6 exception return.  Marks both values as dead.  */
2899 static void gen_rfe(DisasContext *s, TCGv_i32 pc, TCGv_i32 cpsr)
2900 {
2901     store_pc_exc_ret(s, pc);
2902     /* The cpsr_write_eret helper will mask the low bits of PC
2903      * appropriately depending on the new Thumb bit, so it must
2904      * be called after storing the new PC.
2905      */
2906     translator_io_start(&s->base);
2907     gen_helper_cpsr_write_eret(tcg_env, cpsr);
2908     /* Must exit loop to check un-masked IRQs */
2909     s->base.is_jmp = DISAS_EXIT;
2910 }
2911 
2912 /* Generate an old-style exception return. Marks pc as dead. */
2913 static void gen_exception_return(DisasContext *s, TCGv_i32 pc)
2914 {
2915     gen_rfe(s, pc, load_cpu_field(spsr));
2916 }
2917 
2918 static bool aa32_cpreg_encoding_in_impdef_space(uint8_t crn, uint8_t crm)
2919 {
2920     static const uint16_t mask[3] = {
2921         0b0000000111100111,  /* crn ==  9, crm == {c0-c2, c5-c8}   */
2922         0b0000000100010011,  /* crn == 10, crm == {c0, c1, c4, c8} */
2923         0b1000000111111111,  /* crn == 11, crm == {c0-c8, c15}     */
2924     };
2925 
2926     if (crn >= 9 && crn <= 11) {
2927         return (mask[crn - 9] >> crm) & 1;
2928     }
2929     return false;
2930 }
2931 
2932 static void do_coproc_insn(DisasContext *s, int cpnum, int is64,
2933                            int opc1, int crn, int crm, int opc2,
2934                            bool isread, int rt, int rt2)
2935 {
2936     uint32_t key = ENCODE_CP_REG(cpnum, is64, s->ns, crn, crm, opc1, opc2);
2937     const ARMCPRegInfo *ri = get_arm_cp_reginfo(s->cp_regs, key);
2938     TCGv_ptr tcg_ri = NULL;
2939     bool need_exit_tb = false;
2940     uint32_t syndrome;
2941 
2942     /*
2943      * Note that since we are an implementation which takes an
2944      * exception on a trapped conditional instruction only if the
2945      * instruction passes its condition code check, we can take
2946      * advantage of the clause in the ARM ARM that allows us to set
2947      * the COND field in the instruction to 0xE in all cases.
2948      * We could fish the actual condition out of the insn (ARM)
2949      * or the condexec bits (Thumb) but it isn't necessary.
2950      */
2951     switch (cpnum) {
2952     case 14:
2953         if (is64) {
2954             syndrome = syn_cp14_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
2955                                          isread, false);
2956         } else {
2957             syndrome = syn_cp14_rt_trap(1, 0xe, opc1, opc2, crn, crm,
2958                                         rt, isread, false);
2959         }
2960         break;
2961     case 15:
2962         if (is64) {
2963             syndrome = syn_cp15_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
2964                                          isread, false);
2965         } else {
2966             syndrome = syn_cp15_rt_trap(1, 0xe, opc1, opc2, crn, crm,
2967                                         rt, isread, false);
2968         }
2969         break;
2970     default:
2971         /*
2972          * ARMv8 defines that only coprocessors 14 and 15 exist,
2973          * so this can only happen if this is an ARMv7 or earlier CPU,
2974          * in which case the syndrome information won't actually be
2975          * guest visible.
2976          */
2977         assert(!arm_dc_feature(s, ARM_FEATURE_V8));
2978         syndrome = syn_uncategorized();
2979         break;
2980     }
2981 
2982     if (s->hstr_active && cpnum == 15 && s->current_el == 1) {
2983         /*
2984          * At EL1, check for a HSTR_EL2 trap, which must take precedence
2985          * over the UNDEF for "no such register" or the UNDEF for "access
2986          * permissions forbid this EL1 access". HSTR_EL2 traps from EL0
2987          * only happen if the cpreg doesn't UNDEF at EL0, so we do those in
2988          * access_check_cp_reg(), after the checks for whether the access
2989          * configurably trapped to EL1.
2990          */
2991         uint32_t maskbit = is64 ? crm : crn;
2992 
2993         if (maskbit != 4 && maskbit != 14) {
2994             /* T4 and T14 are RES0 so never cause traps */
2995             TCGv_i32 t;
2996             DisasLabel over = gen_disas_label(s);
2997 
2998             t = load_cpu_offset(offsetoflow32(CPUARMState, cp15.hstr_el2));
2999             tcg_gen_andi_i32(t, t, 1u << maskbit);
3000             tcg_gen_brcondi_i32(TCG_COND_EQ, t, 0, over.label);
3001 
3002             gen_exception_insn_el(s, 0, EXCP_UDEF, syndrome, 2);
3003             /*
3004              * gen_exception_insn() will set is_jmp to DISAS_NORETURN,
3005              * but since we're conditionally branching over it, we want
3006              * to assume continue-to-next-instruction.
3007              */
3008             s->base.is_jmp = DISAS_NEXT;
3009             set_disas_label(s, over);
3010         }
3011     }
3012 
3013     if (cpnum == 15 && aa32_cpreg_encoding_in_impdef_space(crn, crm)) {
3014         /*
3015          * Check for TIDCP trap, which must take precedence over the UNDEF
3016          * for "no such register" etc.  It shares precedence with HSTR,
3017          * but raises the same exception, so order doesn't matter.
3018          */
3019         switch (s->current_el) {
3020         case 0:
3021             if (arm_dc_feature(s, ARM_FEATURE_AARCH64)
3022                 && dc_isar_feature(aa64_tidcp1, s)) {
3023                 gen_helper_tidcp_el0(tcg_env, tcg_constant_i32(syndrome));
3024             }
3025             break;
3026         case 1:
3027             gen_helper_tidcp_el1(tcg_env, tcg_constant_i32(syndrome));
3028             break;
3029         }
3030     }
3031 
3032     if (!ri) {
3033         /*
3034          * Unknown register; this might be a guest error or a QEMU
3035          * unimplemented feature.
3036          */
3037         if (is64) {
3038             qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
3039                           "64 bit system register cp:%d opc1: %d crm:%d "
3040                           "(%s)\n",
3041                           isread ? "read" : "write", cpnum, opc1, crm,
3042                           s->ns ? "non-secure" : "secure");
3043         } else {
3044             qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
3045                           "system register cp:%d opc1:%d crn:%d crm:%d "
3046                           "opc2:%d (%s)\n",
3047                           isread ? "read" : "write", cpnum, opc1, crn,
3048                           crm, opc2, s->ns ? "non-secure" : "secure");
3049         }
3050         unallocated_encoding(s);
3051         return;
3052     }
3053 
3054     /* Check access permissions */
3055     if (!cp_access_ok(s->current_el, ri, isread)) {
3056         unallocated_encoding(s);
3057         return;
3058     }
3059 
3060     if ((s->hstr_active && s->current_el == 0) || ri->accessfn ||
3061         (ri->fgt && s->fgt_active) ||
3062         (arm_dc_feature(s, ARM_FEATURE_XSCALE) && cpnum < 14)) {
3063         /*
3064          * Emit code to perform further access permissions checks at
3065          * runtime; this may result in an exception.
3066          * Note that on XScale all cp0..c13 registers do an access check
3067          * call in order to handle c15_cpar.
3068          */
3069         gen_set_condexec(s);
3070         gen_update_pc(s, 0);
3071         tcg_ri = tcg_temp_new_ptr();
3072         gen_helper_access_check_cp_reg(tcg_ri, tcg_env,
3073                                        tcg_constant_i32(key),
3074                                        tcg_constant_i32(syndrome),
3075                                        tcg_constant_i32(isread));
3076     } else if (ri->type & ARM_CP_RAISES_EXC) {
3077         /*
3078          * The readfn or writefn might raise an exception;
3079          * synchronize the CPU state in case it does.
3080          */
3081         gen_set_condexec(s);
3082         gen_update_pc(s, 0);
3083     }
3084 
3085     /* Handle special cases first */
3086     switch (ri->type & ARM_CP_SPECIAL_MASK) {
3087     case 0:
3088         break;
3089     case ARM_CP_NOP:
3090         return;
3091     case ARM_CP_WFI:
3092         if (isread) {
3093             unallocated_encoding(s);
3094         } else {
3095             gen_update_pc(s, curr_insn_len(s));
3096             s->base.is_jmp = DISAS_WFI;
3097         }
3098         return;
3099     default:
3100         g_assert_not_reached();
3101     }
3102 
3103     if (ri->type & ARM_CP_IO) {
3104         /* I/O operations must end the TB here (whether read or write) */
3105         need_exit_tb = translator_io_start(&s->base);
3106     }
3107 
3108     if (isread) {
3109         /* Read */
3110         if (is64) {
3111             TCGv_i64 tmp64;
3112             TCGv_i32 tmp;
3113             if (ri->type & ARM_CP_CONST) {
3114                 tmp64 = tcg_constant_i64(ri->resetvalue);
3115             } else if (ri->readfn) {
3116                 if (!tcg_ri) {
3117                     tcg_ri = gen_lookup_cp_reg(key);
3118                 }
3119                 tmp64 = tcg_temp_new_i64();
3120                 gen_helper_get_cp_reg64(tmp64, tcg_env, tcg_ri);
3121             } else {
3122                 tmp64 = tcg_temp_new_i64();
3123                 tcg_gen_ld_i64(tmp64, tcg_env, ri->fieldoffset);
3124             }
3125             tmp = tcg_temp_new_i32();
3126             tcg_gen_extrl_i64_i32(tmp, tmp64);
3127             store_reg(s, rt, tmp);
3128             tmp = tcg_temp_new_i32();
3129             tcg_gen_extrh_i64_i32(tmp, tmp64);
3130             store_reg(s, rt2, tmp);
3131         } else {
3132             TCGv_i32 tmp;
3133             if (ri->type & ARM_CP_CONST) {
3134                 tmp = tcg_constant_i32(ri->resetvalue);
3135             } else if (ri->readfn) {
3136                 if (!tcg_ri) {
3137                     tcg_ri = gen_lookup_cp_reg(key);
3138                 }
3139                 tmp = tcg_temp_new_i32();
3140                 gen_helper_get_cp_reg(tmp, tcg_env, tcg_ri);
3141             } else {
3142                 tmp = load_cpu_offset(ri->fieldoffset);
3143             }
3144             if (rt == 15) {
3145                 /* Destination register of r15 for 32 bit loads sets
3146                  * the condition codes from the high 4 bits of the value
3147                  */
3148                 gen_set_nzcv(tmp);
3149             } else {
3150                 store_reg(s, rt, tmp);
3151             }
3152         }
3153     } else {
3154         /* Write */
3155         if (ri->type & ARM_CP_CONST) {
3156             /* If not forbidden by access permissions, treat as WI */
3157             return;
3158         }
3159 
3160         if (is64) {
3161             TCGv_i32 tmplo, tmphi;
3162             TCGv_i64 tmp64 = tcg_temp_new_i64();
3163             tmplo = load_reg(s, rt);
3164             tmphi = load_reg(s, rt2);
3165             tcg_gen_concat_i32_i64(tmp64, tmplo, tmphi);
3166             if (ri->writefn) {
3167                 if (!tcg_ri) {
3168                     tcg_ri = gen_lookup_cp_reg(key);
3169                 }
3170                 gen_helper_set_cp_reg64(tcg_env, tcg_ri, tmp64);
3171             } else {
3172                 tcg_gen_st_i64(tmp64, tcg_env, ri->fieldoffset);
3173             }
3174         } else {
3175             TCGv_i32 tmp = load_reg(s, rt);
3176             if (ri->writefn) {
3177                 if (!tcg_ri) {
3178                     tcg_ri = gen_lookup_cp_reg(key);
3179                 }
3180                 gen_helper_set_cp_reg(tcg_env, tcg_ri, tmp);
3181             } else {
3182                 store_cpu_offset(tmp, ri->fieldoffset, 4);
3183             }
3184         }
3185     }
3186 
3187     if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
3188         /*
3189          * A write to any coprocessor register that ends a TB
3190          * must rebuild the hflags for the next TB.
3191          */
3192         gen_rebuild_hflags(s, ri->type & ARM_CP_NEWEL);
3193         /*
3194          * We default to ending the TB on a coprocessor register write,
3195          * but allow this to be suppressed by the register definition
3196          * (usually only necessary to work around guest bugs).
3197          */
3198         need_exit_tb = true;
3199     }
3200     if (need_exit_tb) {
3201         gen_lookup_tb(s);
3202     }
3203 }
3204 
3205 /* Decode XScale DSP or iWMMXt insn (in the copro space, cp=0 or 1) */
3206 static void disas_xscale_insn(DisasContext *s, uint32_t insn)
3207 {
3208     int cpnum = (insn >> 8) & 0xf;
3209 
3210     if (extract32(s->c15_cpar, cpnum, 1) == 0) {
3211         unallocated_encoding(s);
3212     } else if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
3213         if (disas_iwmmxt_insn(s, insn)) {
3214             unallocated_encoding(s);
3215         }
3216     } else if (arm_dc_feature(s, ARM_FEATURE_XSCALE)) {
3217         if (disas_dsp_insn(s, insn)) {
3218             unallocated_encoding(s);
3219         }
3220     }
3221 }
3222 
3223 /* Store a 64-bit value to a register pair.  Clobbers val.  */
3224 static void gen_storeq_reg(DisasContext *s, int rlow, int rhigh, TCGv_i64 val)
3225 {
3226     TCGv_i32 tmp;
3227     tmp = tcg_temp_new_i32();
3228     tcg_gen_extrl_i64_i32(tmp, val);
3229     store_reg(s, rlow, tmp);
3230     tmp = tcg_temp_new_i32();
3231     tcg_gen_extrh_i64_i32(tmp, val);
3232     store_reg(s, rhigh, tmp);
3233 }
3234 
3235 /* load and add a 64-bit value from a register pair.  */
3236 static void gen_addq(DisasContext *s, TCGv_i64 val, int rlow, int rhigh)
3237 {
3238     TCGv_i64 tmp;
3239     TCGv_i32 tmpl;
3240     TCGv_i32 tmph;
3241 
3242     /* Load 64-bit value rd:rn.  */
3243     tmpl = load_reg(s, rlow);
3244     tmph = load_reg(s, rhigh);
3245     tmp = tcg_temp_new_i64();
3246     tcg_gen_concat_i32_i64(tmp, tmpl, tmph);
3247     tcg_gen_add_i64(val, val, tmp);
3248 }
3249 
3250 /* Set N and Z flags from hi|lo.  */
3251 static void gen_logicq_cc(TCGv_i32 lo, TCGv_i32 hi)
3252 {
3253     tcg_gen_mov_i32(cpu_NF, hi);
3254     tcg_gen_or_i32(cpu_ZF, lo, hi);
3255 }
3256 
3257 /* Load/Store exclusive instructions are implemented by remembering
3258    the value/address loaded, and seeing if these are the same
3259    when the store is performed.  This should be sufficient to implement
3260    the architecturally mandated semantics, and avoids having to monitor
3261    regular stores.  The compare vs the remembered value is done during
3262    the cmpxchg operation, but we must compare the addresses manually.  */
3263 static void gen_load_exclusive(DisasContext *s, int rt, int rt2,
3264                                TCGv_i32 addr, int size)
3265 {
3266     TCGv_i32 tmp = tcg_temp_new_i32();
3267     MemOp opc = size | MO_ALIGN | s->be_data;
3268 
3269     s->is_ldex = true;
3270 
3271     if (size == 3) {
3272         TCGv_i32 tmp2 = tcg_temp_new_i32();
3273         TCGv_i64 t64 = tcg_temp_new_i64();
3274 
3275         /*
3276          * For AArch32, architecturally the 32-bit word at the lowest
3277          * address is always Rt and the one at addr+4 is Rt2, even if
3278          * the CPU is big-endian. That means we don't want to do a
3279          * gen_aa32_ld_i64(), which checks SCTLR_B as if for an
3280          * architecturally 64-bit access, but instead do a 64-bit access
3281          * using MO_BE if appropriate and then split the two halves.
3282          */
3283         TCGv taddr = gen_aa32_addr(s, addr, opc);
3284 
3285         tcg_gen_qemu_ld_i64(t64, taddr, get_mem_index(s), opc);
3286         tcg_gen_mov_i64(cpu_exclusive_val, t64);
3287         if (s->be_data == MO_BE) {
3288             tcg_gen_extr_i64_i32(tmp2, tmp, t64);
3289         } else {
3290             tcg_gen_extr_i64_i32(tmp, tmp2, t64);
3291         }
3292         store_reg(s, rt2, tmp2);
3293     } else {
3294         gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), opc);
3295         tcg_gen_extu_i32_i64(cpu_exclusive_val, tmp);
3296     }
3297 
3298     store_reg(s, rt, tmp);
3299     tcg_gen_extu_i32_i64(cpu_exclusive_addr, addr);
3300 }
3301 
3302 static void gen_clrex(DisasContext *s)
3303 {
3304     tcg_gen_movi_i64(cpu_exclusive_addr, -1);
3305 }
3306 
3307 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
3308                                 TCGv_i32 addr, int size)
3309 {
3310     TCGv_i32 t0, t1, t2;
3311     TCGv_i64 extaddr;
3312     TCGv taddr;
3313     TCGLabel *done_label;
3314     TCGLabel *fail_label;
3315     MemOp opc = size | MO_ALIGN | s->be_data;
3316 
3317     /* if (env->exclusive_addr == addr && env->exclusive_val == [addr]) {
3318          [addr] = {Rt};
3319          {Rd} = 0;
3320        } else {
3321          {Rd} = 1;
3322        } */
3323     fail_label = gen_new_label();
3324     done_label = gen_new_label();
3325     extaddr = tcg_temp_new_i64();
3326     tcg_gen_extu_i32_i64(extaddr, addr);
3327     tcg_gen_brcond_i64(TCG_COND_NE, extaddr, cpu_exclusive_addr, fail_label);
3328 
3329     taddr = gen_aa32_addr(s, addr, opc);
3330     t0 = tcg_temp_new_i32();
3331     t1 = load_reg(s, rt);
3332     if (size == 3) {
3333         TCGv_i64 o64 = tcg_temp_new_i64();
3334         TCGv_i64 n64 = tcg_temp_new_i64();
3335 
3336         t2 = load_reg(s, rt2);
3337 
3338         /*
3339          * For AArch32, architecturally the 32-bit word at the lowest
3340          * address is always Rt and the one at addr+4 is Rt2, even if
3341          * the CPU is big-endian. Since we're going to treat this as a
3342          * single 64-bit BE store, we need to put the two halves in the
3343          * opposite order for BE to LE, so that they end up in the right
3344          * places.  We don't want gen_aa32_st_i64, because that checks
3345          * SCTLR_B as if for an architectural 64-bit access.
3346          */
3347         if (s->be_data == MO_BE) {
3348             tcg_gen_concat_i32_i64(n64, t2, t1);
3349         } else {
3350             tcg_gen_concat_i32_i64(n64, t1, t2);
3351         }
3352 
3353         tcg_gen_atomic_cmpxchg_i64(o64, taddr, cpu_exclusive_val, n64,
3354                                    get_mem_index(s), opc);
3355 
3356         tcg_gen_setcond_i64(TCG_COND_NE, o64, o64, cpu_exclusive_val);
3357         tcg_gen_extrl_i64_i32(t0, o64);
3358     } else {
3359         t2 = tcg_temp_new_i32();
3360         tcg_gen_extrl_i64_i32(t2, cpu_exclusive_val);
3361         tcg_gen_atomic_cmpxchg_i32(t0, taddr, t2, t1, get_mem_index(s), opc);
3362         tcg_gen_setcond_i32(TCG_COND_NE, t0, t0, t2);
3363     }
3364     tcg_gen_mov_i32(cpu_R[rd], t0);
3365     tcg_gen_br(done_label);
3366 
3367     gen_set_label(fail_label);
3368     tcg_gen_movi_i32(cpu_R[rd], 1);
3369     gen_set_label(done_label);
3370     tcg_gen_movi_i64(cpu_exclusive_addr, -1);
3371 }
3372 
3373 /* gen_srs:
3374  * @env: CPUARMState
3375  * @s: DisasContext
3376  * @mode: mode field from insn (which stack to store to)
3377  * @amode: addressing mode (DA/IA/DB/IB), encoded as per P,U bits in ARM insn
3378  * @writeback: true if writeback bit set
3379  *
3380  * Generate code for the SRS (Store Return State) insn.
3381  */
3382 static void gen_srs(DisasContext *s,
3383                     uint32_t mode, uint32_t amode, bool writeback)
3384 {
3385     int32_t offset;
3386     TCGv_i32 addr, tmp;
3387     bool undef = false;
3388 
3389     /* SRS is:
3390      * - trapped to EL3 if EL3 is AArch64 and we are at Secure EL1
3391      *   and specified mode is monitor mode
3392      * - UNDEFINED in Hyp mode
3393      * - UNPREDICTABLE in User or System mode
3394      * - UNPREDICTABLE if the specified mode is:
3395      * -- not implemented
3396      * -- not a valid mode number
3397      * -- a mode that's at a higher exception level
3398      * -- Monitor, if we are Non-secure
3399      * For the UNPREDICTABLE cases we choose to UNDEF.
3400      */
3401     if (s->current_el == 1 && !s->ns && mode == ARM_CPU_MODE_MON) {
3402         gen_exception_insn_el(s, 0, EXCP_UDEF, syn_uncategorized(), 3);
3403         return;
3404     }
3405 
3406     if (s->current_el == 0 || s->current_el == 2) {
3407         undef = true;
3408     }
3409 
3410     switch (mode) {
3411     case ARM_CPU_MODE_USR:
3412     case ARM_CPU_MODE_FIQ:
3413     case ARM_CPU_MODE_IRQ:
3414     case ARM_CPU_MODE_SVC:
3415     case ARM_CPU_MODE_ABT:
3416     case ARM_CPU_MODE_UND:
3417     case ARM_CPU_MODE_SYS:
3418         break;
3419     case ARM_CPU_MODE_HYP:
3420         if (s->current_el == 1 || !arm_dc_feature(s, ARM_FEATURE_EL2)) {
3421             undef = true;
3422         }
3423         break;
3424     case ARM_CPU_MODE_MON:
3425         /* No need to check specifically for "are we non-secure" because
3426          * we've already made EL0 UNDEF and handled the trap for S-EL1;
3427          * so if this isn't EL3 then we must be non-secure.
3428          */
3429         if (s->current_el != 3) {
3430             undef = true;
3431         }
3432         break;
3433     default:
3434         undef = true;
3435     }
3436 
3437     if (undef) {
3438         unallocated_encoding(s);
3439         return;
3440     }
3441 
3442     addr = tcg_temp_new_i32();
3443     /* get_r13_banked() will raise an exception if called from System mode */
3444     gen_set_condexec(s);
3445     gen_update_pc(s, 0);
3446     gen_helper_get_r13_banked(addr, tcg_env, tcg_constant_i32(mode));
3447     switch (amode) {
3448     case 0: /* DA */
3449         offset = -4;
3450         break;
3451     case 1: /* IA */
3452         offset = 0;
3453         break;
3454     case 2: /* DB */
3455         offset = -8;
3456         break;
3457     case 3: /* IB */
3458         offset = 4;
3459         break;
3460     default:
3461         g_assert_not_reached();
3462     }
3463     tcg_gen_addi_i32(addr, addr, offset);
3464     tmp = load_reg(s, 14);
3465     gen_aa32_st_i32(s, tmp, addr, get_mem_index(s), MO_UL | MO_ALIGN);
3466     tmp = load_cpu_field(spsr);
3467     tcg_gen_addi_i32(addr, addr, 4);
3468     gen_aa32_st_i32(s, tmp, addr, get_mem_index(s), MO_UL | MO_ALIGN);
3469     if (writeback) {
3470         switch (amode) {
3471         case 0:
3472             offset = -8;
3473             break;
3474         case 1:
3475             offset = 4;
3476             break;
3477         case 2:
3478             offset = -4;
3479             break;
3480         case 3:
3481             offset = 0;
3482             break;
3483         default:
3484             g_assert_not_reached();
3485         }
3486         tcg_gen_addi_i32(addr, addr, offset);
3487         gen_helper_set_r13_banked(tcg_env, tcg_constant_i32(mode), addr);
3488     }
3489     s->base.is_jmp = DISAS_UPDATE_EXIT;
3490 }
3491 
3492 /* Skip this instruction if the ARM condition is false */
3493 static void arm_skip_unless(DisasContext *s, uint32_t cond)
3494 {
3495     arm_gen_condlabel(s);
3496     arm_gen_test_cc(cond ^ 1, s->condlabel.label);
3497 }
3498 
3499 
3500 /*
3501  * Constant expanders used by T16/T32 decode
3502  */
3503 
3504 /* Return only the rotation part of T32ExpandImm.  */
3505 static int t32_expandimm_rot(DisasContext *s, int x)
3506 {
3507     return x & 0xc00 ? extract32(x, 7, 5) : 0;
3508 }
3509 
3510 /* Return the unrotated immediate from T32ExpandImm.  */
3511 static int t32_expandimm_imm(DisasContext *s, int x)
3512 {
3513     int imm = extract32(x, 0, 8);
3514 
3515     switch (extract32(x, 8, 4)) {
3516     case 0: /* XY */
3517         /* Nothing to do.  */
3518         break;
3519     case 1: /* 00XY00XY */
3520         imm *= 0x00010001;
3521         break;
3522     case 2: /* XY00XY00 */
3523         imm *= 0x01000100;
3524         break;
3525     case 3: /* XYXYXYXY */
3526         imm *= 0x01010101;
3527         break;
3528     default:
3529         /* Rotated constant.  */
3530         imm |= 0x80;
3531         break;
3532     }
3533     return imm;
3534 }
3535 
3536 static int t32_branch24(DisasContext *s, int x)
3537 {
3538     /* Convert J1:J2 at x[22:21] to I2:I1, which involves I=J^~S.  */
3539     x ^= !(x < 0) * (3 << 21);
3540     /* Append the final zero.  */
3541     return x << 1;
3542 }
3543 
3544 static int t16_setflags(DisasContext *s)
3545 {
3546     return s->condexec_mask == 0;
3547 }
3548 
3549 static int t16_push_list(DisasContext *s, int x)
3550 {
3551     return (x & 0xff) | (x & 0x100) << (14 - 8);
3552 }
3553 
3554 static int t16_pop_list(DisasContext *s, int x)
3555 {
3556     return (x & 0xff) | (x & 0x100) << (15 - 8);
3557 }
3558 
3559 /*
3560  * Include the generated decoders.
3561  */
3562 
3563 #include "decode-a32.c.inc"
3564 #include "decode-a32-uncond.c.inc"
3565 #include "decode-t32.c.inc"
3566 #include "decode-t16.c.inc"
3567 
3568 static bool valid_cp(DisasContext *s, int cp)
3569 {
3570     /*
3571      * Return true if this coprocessor field indicates something
3572      * that's really a possible coprocessor.
3573      * For v7 and earlier, coprocessors 8..15 were reserved for Arm use,
3574      * and of those only cp14 and cp15 were used for registers.
3575      * cp10 and cp11 were used for VFP and Neon, whose decode is
3576      * dealt with elsewhere. With the advent of fp16, cp9 is also
3577      * now part of VFP.
3578      * For v8A and later, the encoding has been tightened so that
3579      * only cp14 and cp15 are valid, and other values aren't considered
3580      * to be in the coprocessor-instruction space at all. v8M still
3581      * permits coprocessors 0..7.
3582      * For XScale, we must not decode the XScale cp0, cp1 space as
3583      * a standard coprocessor insn, because we want to fall through to
3584      * the legacy disas_xscale_insn() decoder after decodetree is done.
3585      */
3586     if (arm_dc_feature(s, ARM_FEATURE_XSCALE) && (cp == 0 || cp == 1)) {
3587         return false;
3588     }
3589 
3590     if (arm_dc_feature(s, ARM_FEATURE_V8) &&
3591         !arm_dc_feature(s, ARM_FEATURE_M)) {
3592         return cp >= 14;
3593     }
3594     return cp < 8 || cp >= 14;
3595 }
3596 
3597 static bool trans_MCR(DisasContext *s, arg_MCR *a)
3598 {
3599     if (!valid_cp(s, a->cp)) {
3600         return false;
3601     }
3602     do_coproc_insn(s, a->cp, false, a->opc1, a->crn, a->crm, a->opc2,
3603                    false, a->rt, 0);
3604     return true;
3605 }
3606 
3607 static bool trans_MRC(DisasContext *s, arg_MRC *a)
3608 {
3609     if (!valid_cp(s, a->cp)) {
3610         return false;
3611     }
3612     do_coproc_insn(s, a->cp, false, a->opc1, a->crn, a->crm, a->opc2,
3613                    true, a->rt, 0);
3614     return true;
3615 }
3616 
3617 static bool trans_MCRR(DisasContext *s, arg_MCRR *a)
3618 {
3619     if (!valid_cp(s, a->cp)) {
3620         return false;
3621     }
3622     do_coproc_insn(s, a->cp, true, a->opc1, 0, a->crm, 0,
3623                    false, a->rt, a->rt2);
3624     return true;
3625 }
3626 
3627 static bool trans_MRRC(DisasContext *s, arg_MRRC *a)
3628 {
3629     if (!valid_cp(s, a->cp)) {
3630         return false;
3631     }
3632     do_coproc_insn(s, a->cp, true, a->opc1, 0, a->crm, 0,
3633                    true, a->rt, a->rt2);
3634     return true;
3635 }
3636 
3637 /* Helpers to swap operands for reverse-subtract.  */
3638 static void gen_rsb(TCGv_i32 dst, TCGv_i32 a, TCGv_i32 b)
3639 {
3640     tcg_gen_sub_i32(dst, b, a);
3641 }
3642 
3643 static void gen_rsb_CC(TCGv_i32 dst, TCGv_i32 a, TCGv_i32 b)
3644 {
3645     gen_sub_CC(dst, b, a);
3646 }
3647 
3648 static void gen_rsc(TCGv_i32 dest, TCGv_i32 a, TCGv_i32 b)
3649 {
3650     gen_sub_carry(dest, b, a);
3651 }
3652 
3653 static void gen_rsc_CC(TCGv_i32 dest, TCGv_i32 a, TCGv_i32 b)
3654 {
3655     gen_sbc_CC(dest, b, a);
3656 }
3657 
3658 /*
3659  * Helpers for the data processing routines.
3660  *
3661  * After the computation store the results back.
3662  * This may be suppressed altogether (STREG_NONE), require a runtime
3663  * check against the stack limits (STREG_SP_CHECK), or generate an
3664  * exception return.  Oh, or store into a register.
3665  *
3666  * Always return true, indicating success for a trans_* function.
3667  */
3668 typedef enum {
3669    STREG_NONE,
3670    STREG_NORMAL,
3671    STREG_SP_CHECK,
3672    STREG_EXC_RET,
3673 } StoreRegKind;
3674 
3675 static bool store_reg_kind(DisasContext *s, int rd,
3676                             TCGv_i32 val, StoreRegKind kind)
3677 {
3678     switch (kind) {
3679     case STREG_NONE:
3680         return true;
3681     case STREG_NORMAL:
3682         /* See ALUWritePC: Interworking only from a32 mode. */
3683         if (s->thumb) {
3684             store_reg(s, rd, val);
3685         } else {
3686             store_reg_bx(s, rd, val);
3687         }
3688         return true;
3689     case STREG_SP_CHECK:
3690         store_sp_checked(s, val);
3691         return true;
3692     case STREG_EXC_RET:
3693         gen_exception_return(s, val);
3694         return true;
3695     }
3696     g_assert_not_reached();
3697 }
3698 
3699 /*
3700  * Data Processing (register)
3701  *
3702  * Operate, with set flags, one register source,
3703  * one immediate shifted register source, and a destination.
3704  */
3705 static bool op_s_rrr_shi(DisasContext *s, arg_s_rrr_shi *a,
3706                          void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
3707                          int logic_cc, StoreRegKind kind)
3708 {
3709     TCGv_i32 tmp1, tmp2;
3710 
3711     tmp2 = load_reg(s, a->rm);
3712     gen_arm_shift_im(tmp2, a->shty, a->shim, logic_cc);
3713     tmp1 = load_reg(s, a->rn);
3714 
3715     gen(tmp1, tmp1, tmp2);
3716 
3717     if (logic_cc) {
3718         gen_logic_CC(tmp1);
3719     }
3720     return store_reg_kind(s, a->rd, tmp1, kind);
3721 }
3722 
3723 static bool op_s_rxr_shi(DisasContext *s, arg_s_rrr_shi *a,
3724                          void (*gen)(TCGv_i32, TCGv_i32),
3725                          int logic_cc, StoreRegKind kind)
3726 {
3727     TCGv_i32 tmp;
3728 
3729     tmp = load_reg(s, a->rm);
3730     gen_arm_shift_im(tmp, a->shty, a->shim, logic_cc);
3731 
3732     gen(tmp, tmp);
3733     if (logic_cc) {
3734         gen_logic_CC(tmp);
3735     }
3736     return store_reg_kind(s, a->rd, tmp, kind);
3737 }
3738 
3739 /*
3740  * Data-processing (register-shifted register)
3741  *
3742  * Operate, with set flags, one register source,
3743  * one register shifted register source, and a destination.
3744  */
3745 static bool op_s_rrr_shr(DisasContext *s, arg_s_rrr_shr *a,
3746                          void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
3747                          int logic_cc, StoreRegKind kind)
3748 {
3749     TCGv_i32 tmp1, tmp2;
3750 
3751     tmp1 = load_reg(s, a->rs);
3752     tmp2 = load_reg(s, a->rm);
3753     gen_arm_shift_reg(tmp2, a->shty, tmp1, logic_cc);
3754     tmp1 = load_reg(s, a->rn);
3755 
3756     gen(tmp1, tmp1, tmp2);
3757 
3758     if (logic_cc) {
3759         gen_logic_CC(tmp1);
3760     }
3761     return store_reg_kind(s, a->rd, tmp1, kind);
3762 }
3763 
3764 static bool op_s_rxr_shr(DisasContext *s, arg_s_rrr_shr *a,
3765                          void (*gen)(TCGv_i32, TCGv_i32),
3766                          int logic_cc, StoreRegKind kind)
3767 {
3768     TCGv_i32 tmp1, tmp2;
3769 
3770     tmp1 = load_reg(s, a->rs);
3771     tmp2 = load_reg(s, a->rm);
3772     gen_arm_shift_reg(tmp2, a->shty, tmp1, logic_cc);
3773 
3774     gen(tmp2, tmp2);
3775     if (logic_cc) {
3776         gen_logic_CC(tmp2);
3777     }
3778     return store_reg_kind(s, a->rd, tmp2, kind);
3779 }
3780 
3781 /*
3782  * Data-processing (immediate)
3783  *
3784  * Operate, with set flags, one register source,
3785  * one rotated immediate, and a destination.
3786  *
3787  * Note that logic_cc && a->rot setting CF based on the msb of the
3788  * immediate is the reason why we must pass in the unrotated form
3789  * of the immediate.
3790  */
3791 static bool op_s_rri_rot(DisasContext *s, arg_s_rri_rot *a,
3792                          void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
3793                          int logic_cc, StoreRegKind kind)
3794 {
3795     TCGv_i32 tmp1;
3796     uint32_t imm;
3797 
3798     imm = ror32(a->imm, a->rot);
3799     if (logic_cc && a->rot) {
3800         tcg_gen_movi_i32(cpu_CF, imm >> 31);
3801     }
3802     tmp1 = load_reg(s, a->rn);
3803 
3804     gen(tmp1, tmp1, tcg_constant_i32(imm));
3805 
3806     if (logic_cc) {
3807         gen_logic_CC(tmp1);
3808     }
3809     return store_reg_kind(s, a->rd, tmp1, kind);
3810 }
3811 
3812 static bool op_s_rxi_rot(DisasContext *s, arg_s_rri_rot *a,
3813                          void (*gen)(TCGv_i32, TCGv_i32),
3814                          int logic_cc, StoreRegKind kind)
3815 {
3816     TCGv_i32 tmp;
3817     uint32_t imm;
3818 
3819     imm = ror32(a->imm, a->rot);
3820     if (logic_cc && a->rot) {
3821         tcg_gen_movi_i32(cpu_CF, imm >> 31);
3822     }
3823 
3824     tmp = tcg_temp_new_i32();
3825     gen(tmp, tcg_constant_i32(imm));
3826 
3827     if (logic_cc) {
3828         gen_logic_CC(tmp);
3829     }
3830     return store_reg_kind(s, a->rd, tmp, kind);
3831 }
3832 
3833 #define DO_ANY3(NAME, OP, L, K)                                         \
3834     static bool trans_##NAME##_rrri(DisasContext *s, arg_s_rrr_shi *a)  \
3835     { StoreRegKind k = (K); return op_s_rrr_shi(s, a, OP, L, k); }      \
3836     static bool trans_##NAME##_rrrr(DisasContext *s, arg_s_rrr_shr *a)  \
3837     { StoreRegKind k = (K); return op_s_rrr_shr(s, a, OP, L, k); }      \
3838     static bool trans_##NAME##_rri(DisasContext *s, arg_s_rri_rot *a)   \
3839     { StoreRegKind k = (K); return op_s_rri_rot(s, a, OP, L, k); }
3840 
3841 #define DO_ANY2(NAME, OP, L, K)                                         \
3842     static bool trans_##NAME##_rxri(DisasContext *s, arg_s_rrr_shi *a)  \
3843     { StoreRegKind k = (K); return op_s_rxr_shi(s, a, OP, L, k); }      \
3844     static bool trans_##NAME##_rxrr(DisasContext *s, arg_s_rrr_shr *a)  \
3845     { StoreRegKind k = (K); return op_s_rxr_shr(s, a, OP, L, k); }      \
3846     static bool trans_##NAME##_rxi(DisasContext *s, arg_s_rri_rot *a)   \
3847     { StoreRegKind k = (K); return op_s_rxi_rot(s, a, OP, L, k); }
3848 
3849 #define DO_CMP2(NAME, OP, L)                                            \
3850     static bool trans_##NAME##_xrri(DisasContext *s, arg_s_rrr_shi *a)  \
3851     { return op_s_rrr_shi(s, a, OP, L, STREG_NONE); }                   \
3852     static bool trans_##NAME##_xrrr(DisasContext *s, arg_s_rrr_shr *a)  \
3853     { return op_s_rrr_shr(s, a, OP, L, STREG_NONE); }                   \
3854     static bool trans_##NAME##_xri(DisasContext *s, arg_s_rri_rot *a)   \
3855     { return op_s_rri_rot(s, a, OP, L, STREG_NONE); }
3856 
3857 DO_ANY3(AND, tcg_gen_and_i32, a->s, STREG_NORMAL)
3858 DO_ANY3(EOR, tcg_gen_xor_i32, a->s, STREG_NORMAL)
3859 DO_ANY3(ORR, tcg_gen_or_i32, a->s, STREG_NORMAL)
3860 DO_ANY3(BIC, tcg_gen_andc_i32, a->s, STREG_NORMAL)
3861 
3862 DO_ANY3(RSB, a->s ? gen_rsb_CC : gen_rsb, false, STREG_NORMAL)
3863 DO_ANY3(ADC, a->s ? gen_adc_CC : gen_add_carry, false, STREG_NORMAL)
3864 DO_ANY3(SBC, a->s ? gen_sbc_CC : gen_sub_carry, false, STREG_NORMAL)
3865 DO_ANY3(RSC, a->s ? gen_rsc_CC : gen_rsc, false, STREG_NORMAL)
3866 
3867 DO_CMP2(TST, tcg_gen_and_i32, true)
3868 DO_CMP2(TEQ, tcg_gen_xor_i32, true)
3869 DO_CMP2(CMN, gen_add_CC, false)
3870 DO_CMP2(CMP, gen_sub_CC, false)
3871 
3872 DO_ANY3(ADD, a->s ? gen_add_CC : tcg_gen_add_i32, false,
3873         a->rd == 13 && a->rn == 13 ? STREG_SP_CHECK : STREG_NORMAL)
3874 
3875 /*
3876  * Note for the computation of StoreRegKind we return out of the
3877  * middle of the functions that are expanded by DO_ANY3, and that
3878  * we modify a->s via that parameter before it is used by OP.
3879  */
3880 DO_ANY3(SUB, a->s ? gen_sub_CC : tcg_gen_sub_i32, false,
3881         ({
3882             StoreRegKind ret = STREG_NORMAL;
3883             if (a->rd == 15 && a->s) {
3884                 /*
3885                  * See ALUExceptionReturn:
3886                  * In User mode, UNPREDICTABLE; we choose UNDEF.
3887                  * In Hyp mode, UNDEFINED.
3888                  */
3889                 if (IS_USER(s) || s->current_el == 2) {
3890                     unallocated_encoding(s);
3891                     return true;
3892                 }
3893                 /* There is no writeback of nzcv to PSTATE.  */
3894                 a->s = 0;
3895                 ret = STREG_EXC_RET;
3896             } else if (a->rd == 13 && a->rn == 13) {
3897                 ret = STREG_SP_CHECK;
3898             }
3899             ret;
3900         }))
3901 
3902 DO_ANY2(MOV, tcg_gen_mov_i32, a->s,
3903         ({
3904             StoreRegKind ret = STREG_NORMAL;
3905             if (a->rd == 15 && a->s) {
3906                 /*
3907                  * See ALUExceptionReturn:
3908                  * In User mode, UNPREDICTABLE; we choose UNDEF.
3909                  * In Hyp mode, UNDEFINED.
3910                  */
3911                 if (IS_USER(s) || s->current_el == 2) {
3912                     unallocated_encoding(s);
3913                     return true;
3914                 }
3915                 /* There is no writeback of nzcv to PSTATE.  */
3916                 a->s = 0;
3917                 ret = STREG_EXC_RET;
3918             } else if (a->rd == 13) {
3919                 ret = STREG_SP_CHECK;
3920             }
3921             ret;
3922         }))
3923 
3924 DO_ANY2(MVN, tcg_gen_not_i32, a->s, STREG_NORMAL)
3925 
3926 /*
3927  * ORN is only available with T32, so there is no register-shifted-register
3928  * form of the insn.  Using the DO_ANY3 macro would create an unused function.
3929  */
3930 static bool trans_ORN_rrri(DisasContext *s, arg_s_rrr_shi *a)
3931 {
3932     return op_s_rrr_shi(s, a, tcg_gen_orc_i32, a->s, STREG_NORMAL);
3933 }
3934 
3935 static bool trans_ORN_rri(DisasContext *s, arg_s_rri_rot *a)
3936 {
3937     return op_s_rri_rot(s, a, tcg_gen_orc_i32, a->s, STREG_NORMAL);
3938 }
3939 
3940 #undef DO_ANY3
3941 #undef DO_ANY2
3942 #undef DO_CMP2
3943 
3944 static bool trans_ADR(DisasContext *s, arg_ri *a)
3945 {
3946     store_reg_bx(s, a->rd, add_reg_for_lit(s, 15, a->imm));
3947     return true;
3948 }
3949 
3950 static bool trans_MOVW(DisasContext *s, arg_MOVW *a)
3951 {
3952     if (!ENABLE_ARCH_6T2) {
3953         return false;
3954     }
3955 
3956     store_reg(s, a->rd, tcg_constant_i32(a->imm));
3957     return true;
3958 }
3959 
3960 static bool trans_MOVT(DisasContext *s, arg_MOVW *a)
3961 {
3962     TCGv_i32 tmp;
3963 
3964     if (!ENABLE_ARCH_6T2) {
3965         return false;
3966     }
3967 
3968     tmp = load_reg(s, a->rd);
3969     tcg_gen_ext16u_i32(tmp, tmp);
3970     tcg_gen_ori_i32(tmp, tmp, a->imm << 16);
3971     store_reg(s, a->rd, tmp);
3972     return true;
3973 }
3974 
3975 /*
3976  * v8.1M MVE wide-shifts
3977  */
3978 static bool do_mve_shl_ri(DisasContext *s, arg_mve_shl_ri *a,
3979                           WideShiftImmFn *fn)
3980 {
3981     TCGv_i64 rda;
3982     TCGv_i32 rdalo, rdahi;
3983 
3984     if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
3985         /* Decode falls through to ORR/MOV UNPREDICTABLE handling */
3986         return false;
3987     }
3988     if (a->rdahi == 15) {
3989         /* These are a different encoding (SQSHL/SRSHR/UQSHL/URSHR) */
3990         return false;
3991     }
3992     if (!dc_isar_feature(aa32_mve, s) ||
3993         !arm_dc_feature(s, ARM_FEATURE_M_MAIN) ||
3994         a->rdahi == 13) {
3995         /* RdaHi == 13 is UNPREDICTABLE; we choose to UNDEF */
3996         unallocated_encoding(s);
3997         return true;
3998     }
3999 
4000     if (a->shim == 0) {
4001         a->shim = 32;
4002     }
4003 
4004     rda = tcg_temp_new_i64();
4005     rdalo = load_reg(s, a->rdalo);
4006     rdahi = load_reg(s, a->rdahi);
4007     tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
4008 
4009     fn(rda, rda, a->shim);
4010 
4011     tcg_gen_extrl_i64_i32(rdalo, rda);
4012     tcg_gen_extrh_i64_i32(rdahi, rda);
4013     store_reg(s, a->rdalo, rdalo);
4014     store_reg(s, a->rdahi, rdahi);
4015 
4016     return true;
4017 }
4018 
4019 static bool trans_ASRL_ri(DisasContext *s, arg_mve_shl_ri *a)
4020 {
4021     return do_mve_shl_ri(s, a, tcg_gen_sari_i64);
4022 }
4023 
4024 static bool trans_LSLL_ri(DisasContext *s, arg_mve_shl_ri *a)
4025 {
4026     return do_mve_shl_ri(s, a, tcg_gen_shli_i64);
4027 }
4028 
4029 static bool trans_LSRL_ri(DisasContext *s, arg_mve_shl_ri *a)
4030 {
4031     return do_mve_shl_ri(s, a, tcg_gen_shri_i64);
4032 }
4033 
4034 static void gen_mve_sqshll(TCGv_i64 r, TCGv_i64 n, int64_t shift)
4035 {
4036     gen_helper_mve_sqshll(r, tcg_env, n, tcg_constant_i32(shift));
4037 }
4038 
4039 static bool trans_SQSHLL_ri(DisasContext *s, arg_mve_shl_ri *a)
4040 {
4041     return do_mve_shl_ri(s, a, gen_mve_sqshll);
4042 }
4043 
4044 static void gen_mve_uqshll(TCGv_i64 r, TCGv_i64 n, int64_t shift)
4045 {
4046     gen_helper_mve_uqshll(r, tcg_env, n, tcg_constant_i32(shift));
4047 }
4048 
4049 static bool trans_UQSHLL_ri(DisasContext *s, arg_mve_shl_ri *a)
4050 {
4051     return do_mve_shl_ri(s, a, gen_mve_uqshll);
4052 }
4053 
4054 static bool trans_SRSHRL_ri(DisasContext *s, arg_mve_shl_ri *a)
4055 {
4056     return do_mve_shl_ri(s, a, gen_srshr64_i64);
4057 }
4058 
4059 static bool trans_URSHRL_ri(DisasContext *s, arg_mve_shl_ri *a)
4060 {
4061     return do_mve_shl_ri(s, a, gen_urshr64_i64);
4062 }
4063 
4064 static bool do_mve_shl_rr(DisasContext *s, arg_mve_shl_rr *a, WideShiftFn *fn)
4065 {
4066     TCGv_i64 rda;
4067     TCGv_i32 rdalo, rdahi;
4068 
4069     if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
4070         /* Decode falls through to ORR/MOV UNPREDICTABLE handling */
4071         return false;
4072     }
4073     if (a->rdahi == 15) {
4074         /* These are a different encoding (SQSHL/SRSHR/UQSHL/URSHR) */
4075         return false;
4076     }
4077     if (!dc_isar_feature(aa32_mve, s) ||
4078         !arm_dc_feature(s, ARM_FEATURE_M_MAIN) ||
4079         a->rdahi == 13 || a->rm == 13 || a->rm == 15 ||
4080         a->rm == a->rdahi || a->rm == a->rdalo) {
4081         /* These rdahi/rdalo/rm cases are UNPREDICTABLE; we choose to UNDEF */
4082         unallocated_encoding(s);
4083         return true;
4084     }
4085 
4086     rda = tcg_temp_new_i64();
4087     rdalo = load_reg(s, a->rdalo);
4088     rdahi = load_reg(s, a->rdahi);
4089     tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
4090 
4091     /* The helper takes care of the sign-extension of the low 8 bits of Rm */
4092     fn(rda, tcg_env, rda, cpu_R[a->rm]);
4093 
4094     tcg_gen_extrl_i64_i32(rdalo, rda);
4095     tcg_gen_extrh_i64_i32(rdahi, rda);
4096     store_reg(s, a->rdalo, rdalo);
4097     store_reg(s, a->rdahi, rdahi);
4098 
4099     return true;
4100 }
4101 
4102 static bool trans_LSLL_rr(DisasContext *s, arg_mve_shl_rr *a)
4103 {
4104     return do_mve_shl_rr(s, a, gen_helper_mve_ushll);
4105 }
4106 
4107 static bool trans_ASRL_rr(DisasContext *s, arg_mve_shl_rr *a)
4108 {
4109     return do_mve_shl_rr(s, a, gen_helper_mve_sshrl);
4110 }
4111 
4112 static bool trans_UQRSHLL64_rr(DisasContext *s, arg_mve_shl_rr *a)
4113 {
4114     return do_mve_shl_rr(s, a, gen_helper_mve_uqrshll);
4115 }
4116 
4117 static bool trans_SQRSHRL64_rr(DisasContext *s, arg_mve_shl_rr *a)
4118 {
4119     return do_mve_shl_rr(s, a, gen_helper_mve_sqrshrl);
4120 }
4121 
4122 static bool trans_UQRSHLL48_rr(DisasContext *s, arg_mve_shl_rr *a)
4123 {
4124     return do_mve_shl_rr(s, a, gen_helper_mve_uqrshll48);
4125 }
4126 
4127 static bool trans_SQRSHRL48_rr(DisasContext *s, arg_mve_shl_rr *a)
4128 {
4129     return do_mve_shl_rr(s, a, gen_helper_mve_sqrshrl48);
4130 }
4131 
4132 static bool do_mve_sh_ri(DisasContext *s, arg_mve_sh_ri *a, ShiftImmFn *fn)
4133 {
4134     if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
4135         /* Decode falls through to ORR/MOV UNPREDICTABLE handling */
4136         return false;
4137     }
4138     if (!dc_isar_feature(aa32_mve, s) ||
4139         !arm_dc_feature(s, ARM_FEATURE_M_MAIN) ||
4140         a->rda == 13 || a->rda == 15) {
4141         /* These rda cases are UNPREDICTABLE; we choose to UNDEF */
4142         unallocated_encoding(s);
4143         return true;
4144     }
4145 
4146     if (a->shim == 0) {
4147         a->shim = 32;
4148     }
4149     fn(cpu_R[a->rda], cpu_R[a->rda], a->shim);
4150 
4151     return true;
4152 }
4153 
4154 static bool trans_URSHR_ri(DisasContext *s, arg_mve_sh_ri *a)
4155 {
4156     return do_mve_sh_ri(s, a, gen_urshr32_i32);
4157 }
4158 
4159 static bool trans_SRSHR_ri(DisasContext *s, arg_mve_sh_ri *a)
4160 {
4161     return do_mve_sh_ri(s, a, gen_srshr32_i32);
4162 }
4163 
4164 static void gen_mve_sqshl(TCGv_i32 r, TCGv_i32 n, int32_t shift)
4165 {
4166     gen_helper_mve_sqshl(r, tcg_env, n, tcg_constant_i32(shift));
4167 }
4168 
4169 static bool trans_SQSHL_ri(DisasContext *s, arg_mve_sh_ri *a)
4170 {
4171     return do_mve_sh_ri(s, a, gen_mve_sqshl);
4172 }
4173 
4174 static void gen_mve_uqshl(TCGv_i32 r, TCGv_i32 n, int32_t shift)
4175 {
4176     gen_helper_mve_uqshl(r, tcg_env, n, tcg_constant_i32(shift));
4177 }
4178 
4179 static bool trans_UQSHL_ri(DisasContext *s, arg_mve_sh_ri *a)
4180 {
4181     return do_mve_sh_ri(s, a, gen_mve_uqshl);
4182 }
4183 
4184 static bool do_mve_sh_rr(DisasContext *s, arg_mve_sh_rr *a, ShiftFn *fn)
4185 {
4186     if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
4187         /* Decode falls through to ORR/MOV UNPREDICTABLE handling */
4188         return false;
4189     }
4190     if (!dc_isar_feature(aa32_mve, s) ||
4191         !arm_dc_feature(s, ARM_FEATURE_M_MAIN) ||
4192         a->rda == 13 || a->rda == 15 || a->rm == 13 || a->rm == 15 ||
4193         a->rm == a->rda) {
4194         /* These rda/rm cases are UNPREDICTABLE; we choose to UNDEF */
4195         unallocated_encoding(s);
4196         return true;
4197     }
4198 
4199     /* The helper takes care of the sign-extension of the low 8 bits of Rm */
4200     fn(cpu_R[a->rda], tcg_env, cpu_R[a->rda], cpu_R[a->rm]);
4201     return true;
4202 }
4203 
4204 static bool trans_SQRSHR_rr(DisasContext *s, arg_mve_sh_rr *a)
4205 {
4206     return do_mve_sh_rr(s, a, gen_helper_mve_sqrshr);
4207 }
4208 
4209 static bool trans_UQRSHL_rr(DisasContext *s, arg_mve_sh_rr *a)
4210 {
4211     return do_mve_sh_rr(s, a, gen_helper_mve_uqrshl);
4212 }
4213 
4214 /*
4215  * Multiply and multiply accumulate
4216  */
4217 
4218 static bool op_mla(DisasContext *s, arg_s_rrrr *a, bool add)
4219 {
4220     TCGv_i32 t1, t2;
4221 
4222     t1 = load_reg(s, a->rn);
4223     t2 = load_reg(s, a->rm);
4224     tcg_gen_mul_i32(t1, t1, t2);
4225     if (add) {
4226         t2 = load_reg(s, a->ra);
4227         tcg_gen_add_i32(t1, t1, t2);
4228     }
4229     if (a->s) {
4230         gen_logic_CC(t1);
4231     }
4232     store_reg(s, a->rd, t1);
4233     return true;
4234 }
4235 
4236 static bool trans_MUL(DisasContext *s, arg_MUL *a)
4237 {
4238     return op_mla(s, a, false);
4239 }
4240 
4241 static bool trans_MLA(DisasContext *s, arg_MLA *a)
4242 {
4243     return op_mla(s, a, true);
4244 }
4245 
4246 static bool trans_MLS(DisasContext *s, arg_MLS *a)
4247 {
4248     TCGv_i32 t1, t2;
4249 
4250     if (!ENABLE_ARCH_6T2) {
4251         return false;
4252     }
4253     t1 = load_reg(s, a->rn);
4254     t2 = load_reg(s, a->rm);
4255     tcg_gen_mul_i32(t1, t1, t2);
4256     t2 = load_reg(s, a->ra);
4257     tcg_gen_sub_i32(t1, t2, t1);
4258     store_reg(s, a->rd, t1);
4259     return true;
4260 }
4261 
4262 static bool op_mlal(DisasContext *s, arg_s_rrrr *a, bool uns, bool add)
4263 {
4264     TCGv_i32 t0, t1, t2, t3;
4265 
4266     t0 = load_reg(s, a->rm);
4267     t1 = load_reg(s, a->rn);
4268     if (uns) {
4269         tcg_gen_mulu2_i32(t0, t1, t0, t1);
4270     } else {
4271         tcg_gen_muls2_i32(t0, t1, t0, t1);
4272     }
4273     if (add) {
4274         t2 = load_reg(s, a->ra);
4275         t3 = load_reg(s, a->rd);
4276         tcg_gen_add2_i32(t0, t1, t0, t1, t2, t3);
4277     }
4278     if (a->s) {
4279         gen_logicq_cc(t0, t1);
4280     }
4281     store_reg(s, a->ra, t0);
4282     store_reg(s, a->rd, t1);
4283     return true;
4284 }
4285 
4286 static bool trans_UMULL(DisasContext *s, arg_UMULL *a)
4287 {
4288     return op_mlal(s, a, true, false);
4289 }
4290 
4291 static bool trans_SMULL(DisasContext *s, arg_SMULL *a)
4292 {
4293     return op_mlal(s, a, false, false);
4294 }
4295 
4296 static bool trans_UMLAL(DisasContext *s, arg_UMLAL *a)
4297 {
4298     return op_mlal(s, a, true, true);
4299 }
4300 
4301 static bool trans_SMLAL(DisasContext *s, arg_SMLAL *a)
4302 {
4303     return op_mlal(s, a, false, true);
4304 }
4305 
4306 static bool trans_UMAAL(DisasContext *s, arg_UMAAL *a)
4307 {
4308     TCGv_i32 t0, t1, t2, zero;
4309 
4310     if (s->thumb
4311         ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
4312         : !ENABLE_ARCH_6) {
4313         return false;
4314     }
4315 
4316     t0 = load_reg(s, a->rm);
4317     t1 = load_reg(s, a->rn);
4318     tcg_gen_mulu2_i32(t0, t1, t0, t1);
4319     zero = tcg_constant_i32(0);
4320     t2 = load_reg(s, a->ra);
4321     tcg_gen_add2_i32(t0, t1, t0, t1, t2, zero);
4322     t2 = load_reg(s, a->rd);
4323     tcg_gen_add2_i32(t0, t1, t0, t1, t2, zero);
4324     store_reg(s, a->ra, t0);
4325     store_reg(s, a->rd, t1);
4326     return true;
4327 }
4328 
4329 /*
4330  * Saturating addition and subtraction
4331  */
4332 
4333 static bool op_qaddsub(DisasContext *s, arg_rrr *a, bool add, bool doub)
4334 {
4335     TCGv_i32 t0, t1;
4336 
4337     if (s->thumb
4338         ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
4339         : !ENABLE_ARCH_5TE) {
4340         return false;
4341     }
4342 
4343     t0 = load_reg(s, a->rm);
4344     t1 = load_reg(s, a->rn);
4345     if (doub) {
4346         gen_helper_add_saturate(t1, tcg_env, t1, t1);
4347     }
4348     if (add) {
4349         gen_helper_add_saturate(t0, tcg_env, t0, t1);
4350     } else {
4351         gen_helper_sub_saturate(t0, tcg_env, t0, t1);
4352     }
4353     store_reg(s, a->rd, t0);
4354     return true;
4355 }
4356 
4357 #define DO_QADDSUB(NAME, ADD, DOUB) \
4358 static bool trans_##NAME(DisasContext *s, arg_rrr *a)    \
4359 {                                                        \
4360     return op_qaddsub(s, a, ADD, DOUB);                  \
4361 }
4362 
4363 DO_QADDSUB(QADD, true, false)
4364 DO_QADDSUB(QSUB, false, false)
4365 DO_QADDSUB(QDADD, true, true)
4366 DO_QADDSUB(QDSUB, false, true)
4367 
4368 #undef DO_QADDSUB
4369 
4370 /*
4371  * Halfword multiply and multiply accumulate
4372  */
4373 
4374 static bool op_smlaxxx(DisasContext *s, arg_rrrr *a,
4375                        int add_long, bool nt, bool mt)
4376 {
4377     TCGv_i32 t0, t1, tl, th;
4378 
4379     if (s->thumb
4380         ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
4381         : !ENABLE_ARCH_5TE) {
4382         return false;
4383     }
4384 
4385     t0 = load_reg(s, a->rn);
4386     t1 = load_reg(s, a->rm);
4387     gen_mulxy(t0, t1, nt, mt);
4388 
4389     switch (add_long) {
4390     case 0:
4391         store_reg(s, a->rd, t0);
4392         break;
4393     case 1:
4394         t1 = load_reg(s, a->ra);
4395         gen_helper_add_setq(t0, tcg_env, t0, t1);
4396         store_reg(s, a->rd, t0);
4397         break;
4398     case 2:
4399         tl = load_reg(s, a->ra);
4400         th = load_reg(s, a->rd);
4401         /* Sign-extend the 32-bit product to 64 bits.  */
4402         t1 = tcg_temp_new_i32();
4403         tcg_gen_sari_i32(t1, t0, 31);
4404         tcg_gen_add2_i32(tl, th, tl, th, t0, t1);
4405         store_reg(s, a->ra, tl);
4406         store_reg(s, a->rd, th);
4407         break;
4408     default:
4409         g_assert_not_reached();
4410     }
4411     return true;
4412 }
4413 
4414 #define DO_SMLAX(NAME, add, nt, mt) \
4415 static bool trans_##NAME(DisasContext *s, arg_rrrr *a)     \
4416 {                                                          \
4417     return op_smlaxxx(s, a, add, nt, mt);                  \
4418 }
4419 
4420 DO_SMLAX(SMULBB, 0, 0, 0)
4421 DO_SMLAX(SMULBT, 0, 0, 1)
4422 DO_SMLAX(SMULTB, 0, 1, 0)
4423 DO_SMLAX(SMULTT, 0, 1, 1)
4424 
4425 DO_SMLAX(SMLABB, 1, 0, 0)
4426 DO_SMLAX(SMLABT, 1, 0, 1)
4427 DO_SMLAX(SMLATB, 1, 1, 0)
4428 DO_SMLAX(SMLATT, 1, 1, 1)
4429 
4430 DO_SMLAX(SMLALBB, 2, 0, 0)
4431 DO_SMLAX(SMLALBT, 2, 0, 1)
4432 DO_SMLAX(SMLALTB, 2, 1, 0)
4433 DO_SMLAX(SMLALTT, 2, 1, 1)
4434 
4435 #undef DO_SMLAX
4436 
4437 static bool op_smlawx(DisasContext *s, arg_rrrr *a, bool add, bool mt)
4438 {
4439     TCGv_i32 t0, t1;
4440 
4441     if (!ENABLE_ARCH_5TE) {
4442         return false;
4443     }
4444 
4445     t0 = load_reg(s, a->rn);
4446     t1 = load_reg(s, a->rm);
4447     /*
4448      * Since the nominal result is product<47:16>, shift the 16-bit
4449      * input up by 16 bits, so that the result is at product<63:32>.
4450      */
4451     if (mt) {
4452         tcg_gen_andi_i32(t1, t1, 0xffff0000);
4453     } else {
4454         tcg_gen_shli_i32(t1, t1, 16);
4455     }
4456     tcg_gen_muls2_i32(t0, t1, t0, t1);
4457     if (add) {
4458         t0 = load_reg(s, a->ra);
4459         gen_helper_add_setq(t1, tcg_env, t1, t0);
4460     }
4461     store_reg(s, a->rd, t1);
4462     return true;
4463 }
4464 
4465 #define DO_SMLAWX(NAME, add, mt) \
4466 static bool trans_##NAME(DisasContext *s, arg_rrrr *a)     \
4467 {                                                          \
4468     return op_smlawx(s, a, add, mt);                       \
4469 }
4470 
4471 DO_SMLAWX(SMULWB, 0, 0)
4472 DO_SMLAWX(SMULWT, 0, 1)
4473 DO_SMLAWX(SMLAWB, 1, 0)
4474 DO_SMLAWX(SMLAWT, 1, 1)
4475 
4476 #undef DO_SMLAWX
4477 
4478 /*
4479  * MSR (immediate) and hints
4480  */
4481 
4482 static bool trans_YIELD(DisasContext *s, arg_YIELD *a)
4483 {
4484     /*
4485      * When running single-threaded TCG code, use the helper to ensure that
4486      * the next round-robin scheduled vCPU gets a crack.  When running in
4487      * MTTCG we don't generate jumps to the helper as it won't affect the
4488      * scheduling of other vCPUs.
4489      */
4490     if (!(tb_cflags(s->base.tb) & CF_PARALLEL)) {
4491         gen_update_pc(s, curr_insn_len(s));
4492         s->base.is_jmp = DISAS_YIELD;
4493     }
4494     return true;
4495 }
4496 
4497 static bool trans_WFE(DisasContext *s, arg_WFE *a)
4498 {
4499     /*
4500      * When running single-threaded TCG code, use the helper to ensure that
4501      * the next round-robin scheduled vCPU gets a crack.  In MTTCG mode we
4502      * just skip this instruction.  Currently the SEV/SEVL instructions,
4503      * which are *one* of many ways to wake the CPU from WFE, are not
4504      * implemented so we can't sleep like WFI does.
4505      */
4506     if (!(tb_cflags(s->base.tb) & CF_PARALLEL)) {
4507         gen_update_pc(s, curr_insn_len(s));
4508         s->base.is_jmp = DISAS_WFE;
4509     }
4510     return true;
4511 }
4512 
4513 static bool trans_WFI(DisasContext *s, arg_WFI *a)
4514 {
4515     /* For WFI, halt the vCPU until an IRQ. */
4516     gen_update_pc(s, curr_insn_len(s));
4517     s->base.is_jmp = DISAS_WFI;
4518     return true;
4519 }
4520 
4521 static bool trans_ESB(DisasContext *s, arg_ESB *a)
4522 {
4523     /*
4524      * For M-profile, minimal-RAS ESB can be a NOP.
4525      * Without RAS, we must implement this as NOP.
4526      */
4527     if (!arm_dc_feature(s, ARM_FEATURE_M) && dc_isar_feature(aa32_ras, s)) {
4528         /*
4529          * QEMU does not have a source of physical SErrors,
4530          * so we are only concerned with virtual SErrors.
4531          * The pseudocode in the ARM for this case is
4532          *   if PSTATE.EL IN {EL0, EL1} && EL2Enabled() then
4533          *      AArch32.vESBOperation();
4534          * Most of the condition can be evaluated at translation time.
4535          * Test for EL2 present, and defer test for SEL2 to runtime.
4536          */
4537         if (s->current_el <= 1 && arm_dc_feature(s, ARM_FEATURE_EL2)) {
4538             gen_helper_vesb(tcg_env);
4539         }
4540     }
4541     return true;
4542 }
4543 
4544 static bool trans_NOP(DisasContext *s, arg_NOP *a)
4545 {
4546     return true;
4547 }
4548 
4549 static bool trans_MSR_imm(DisasContext *s, arg_MSR_imm *a)
4550 {
4551     uint32_t val = ror32(a->imm, a->rot * 2);
4552     uint32_t mask = msr_mask(s, a->mask, a->r);
4553 
4554     if (gen_set_psr_im(s, mask, a->r, val)) {
4555         unallocated_encoding(s);
4556     }
4557     return true;
4558 }
4559 
4560 /*
4561  * Cyclic Redundancy Check
4562  */
4563 
4564 static bool op_crc32(DisasContext *s, arg_rrr *a, bool c, MemOp sz)
4565 {
4566     TCGv_i32 t1, t2, t3;
4567 
4568     if (!dc_isar_feature(aa32_crc32, s)) {
4569         return false;
4570     }
4571 
4572     t1 = load_reg(s, a->rn);
4573     t2 = load_reg(s, a->rm);
4574     switch (sz) {
4575     case MO_8:
4576         gen_uxtb(t2);
4577         break;
4578     case MO_16:
4579         gen_uxth(t2);
4580         break;
4581     case MO_32:
4582         break;
4583     default:
4584         g_assert_not_reached();
4585     }
4586     t3 = tcg_constant_i32(1 << sz);
4587     if (c) {
4588         gen_helper_crc32c(t1, t1, t2, t3);
4589     } else {
4590         gen_helper_crc32(t1, t1, t2, t3);
4591     }
4592     store_reg(s, a->rd, t1);
4593     return true;
4594 }
4595 
4596 #define DO_CRC32(NAME, c, sz) \
4597 static bool trans_##NAME(DisasContext *s, arg_rrr *a)  \
4598     { return op_crc32(s, a, c, sz); }
4599 
4600 DO_CRC32(CRC32B, false, MO_8)
4601 DO_CRC32(CRC32H, false, MO_16)
4602 DO_CRC32(CRC32W, false, MO_32)
4603 DO_CRC32(CRC32CB, true, MO_8)
4604 DO_CRC32(CRC32CH, true, MO_16)
4605 DO_CRC32(CRC32CW, true, MO_32)
4606 
4607 #undef DO_CRC32
4608 
4609 /*
4610  * Miscellaneous instructions
4611  */
4612 
4613 static bool trans_MRS_bank(DisasContext *s, arg_MRS_bank *a)
4614 {
4615     if (arm_dc_feature(s, ARM_FEATURE_M)) {
4616         return false;
4617     }
4618     gen_mrs_banked(s, a->r, a->sysm, a->rd);
4619     return true;
4620 }
4621 
4622 static bool trans_MSR_bank(DisasContext *s, arg_MSR_bank *a)
4623 {
4624     if (arm_dc_feature(s, ARM_FEATURE_M)) {
4625         return false;
4626     }
4627     gen_msr_banked(s, a->r, a->sysm, a->rn);
4628     return true;
4629 }
4630 
4631 static bool trans_MRS_reg(DisasContext *s, arg_MRS_reg *a)
4632 {
4633     TCGv_i32 tmp;
4634 
4635     if (arm_dc_feature(s, ARM_FEATURE_M)) {
4636         return false;
4637     }
4638     if (a->r) {
4639         if (IS_USER(s)) {
4640             unallocated_encoding(s);
4641             return true;
4642         }
4643         tmp = load_cpu_field(spsr);
4644     } else {
4645         tmp = tcg_temp_new_i32();
4646         gen_helper_cpsr_read(tmp, tcg_env);
4647     }
4648     store_reg(s, a->rd, tmp);
4649     return true;
4650 }
4651 
4652 static bool trans_MSR_reg(DisasContext *s, arg_MSR_reg *a)
4653 {
4654     TCGv_i32 tmp;
4655     uint32_t mask = msr_mask(s, a->mask, a->r);
4656 
4657     if (arm_dc_feature(s, ARM_FEATURE_M)) {
4658         return false;
4659     }
4660     tmp = load_reg(s, a->rn);
4661     if (gen_set_psr(s, mask, a->r, tmp)) {
4662         unallocated_encoding(s);
4663     }
4664     return true;
4665 }
4666 
4667 static bool trans_MRS_v7m(DisasContext *s, arg_MRS_v7m *a)
4668 {
4669     TCGv_i32 tmp;
4670 
4671     if (!arm_dc_feature(s, ARM_FEATURE_M)) {
4672         return false;
4673     }
4674     tmp = tcg_temp_new_i32();
4675     gen_helper_v7m_mrs(tmp, tcg_env, tcg_constant_i32(a->sysm));
4676     store_reg(s, a->rd, tmp);
4677     return true;
4678 }
4679 
4680 static bool trans_MSR_v7m(DisasContext *s, arg_MSR_v7m *a)
4681 {
4682     TCGv_i32 addr, reg;
4683 
4684     if (!arm_dc_feature(s, ARM_FEATURE_M)) {
4685         return false;
4686     }
4687     addr = tcg_constant_i32((a->mask << 10) | a->sysm);
4688     reg = load_reg(s, a->rn);
4689     gen_helper_v7m_msr(tcg_env, addr, reg);
4690     /* If we wrote to CONTROL, the EL might have changed */
4691     gen_rebuild_hflags(s, true);
4692     gen_lookup_tb(s);
4693     return true;
4694 }
4695 
4696 static bool trans_BX(DisasContext *s, arg_BX *a)
4697 {
4698     if (!ENABLE_ARCH_4T) {
4699         return false;
4700     }
4701     gen_bx_excret(s, load_reg(s, a->rm));
4702     return true;
4703 }
4704 
4705 static bool trans_BXJ(DisasContext *s, arg_BXJ *a)
4706 {
4707     if (!ENABLE_ARCH_5J || arm_dc_feature(s, ARM_FEATURE_M)) {
4708         return false;
4709     }
4710     /*
4711      * v7A allows BXJ to be trapped via HSTR.TJDBX. We don't waste a
4712      * TBFLAGS bit on a basically-never-happens case, so call a helper
4713      * function to check for the trap and raise the exception if needed
4714      * (passing it the register number for the syndrome value).
4715      * v8A doesn't have this HSTR bit.
4716      */
4717     if (!arm_dc_feature(s, ARM_FEATURE_V8) &&
4718         arm_dc_feature(s, ARM_FEATURE_EL2) &&
4719         s->current_el < 2 && s->ns) {
4720         gen_helper_check_bxj_trap(tcg_env, tcg_constant_i32(a->rm));
4721     }
4722     /* Trivial implementation equivalent to bx.  */
4723     gen_bx(s, load_reg(s, a->rm));
4724     return true;
4725 }
4726 
4727 static bool trans_BLX_r(DisasContext *s, arg_BLX_r *a)
4728 {
4729     TCGv_i32 tmp;
4730 
4731     if (!ENABLE_ARCH_5) {
4732         return false;
4733     }
4734     tmp = load_reg(s, a->rm);
4735     gen_pc_plus_diff(s, cpu_R[14], curr_insn_len(s) | s->thumb);
4736     gen_bx(s, tmp);
4737     return true;
4738 }
4739 
4740 /*
4741  * BXNS/BLXNS: only exist for v8M with the security extensions,
4742  * and always UNDEF if NonSecure.  We don't implement these in
4743  * the user-only mode either (in theory you can use them from
4744  * Secure User mode but they are too tied in to system emulation).
4745  */
4746 static bool trans_BXNS(DisasContext *s, arg_BXNS *a)
4747 {
4748     if (!s->v8m_secure || IS_USER_ONLY) {
4749         unallocated_encoding(s);
4750     } else {
4751         gen_bxns(s, a->rm);
4752     }
4753     return true;
4754 }
4755 
4756 static bool trans_BLXNS(DisasContext *s, arg_BLXNS *a)
4757 {
4758     if (!s->v8m_secure || IS_USER_ONLY) {
4759         unallocated_encoding(s);
4760     } else {
4761         gen_blxns(s, a->rm);
4762     }
4763     return true;
4764 }
4765 
4766 static bool trans_CLZ(DisasContext *s, arg_CLZ *a)
4767 {
4768     TCGv_i32 tmp;
4769 
4770     if (!ENABLE_ARCH_5) {
4771         return false;
4772     }
4773     tmp = load_reg(s, a->rm);
4774     tcg_gen_clzi_i32(tmp, tmp, 32);
4775     store_reg(s, a->rd, tmp);
4776     return true;
4777 }
4778 
4779 static bool trans_ERET(DisasContext *s, arg_ERET *a)
4780 {
4781     TCGv_i32 tmp;
4782 
4783     if (!arm_dc_feature(s, ARM_FEATURE_V7VE)) {
4784         return false;
4785     }
4786     if (IS_USER(s)) {
4787         unallocated_encoding(s);
4788         return true;
4789     }
4790     if (s->current_el == 2) {
4791         /* ERET from Hyp uses ELR_Hyp, not LR */
4792         tmp = load_cpu_field_low32(elr_el[2]);
4793     } else {
4794         tmp = load_reg(s, 14);
4795     }
4796     gen_exception_return(s, tmp);
4797     return true;
4798 }
4799 
4800 static bool trans_HLT(DisasContext *s, arg_HLT *a)
4801 {
4802     gen_hlt(s, a->imm);
4803     return true;
4804 }
4805 
4806 static bool trans_BKPT(DisasContext *s, arg_BKPT *a)
4807 {
4808     if (!ENABLE_ARCH_5) {
4809         return false;
4810     }
4811     /* BKPT is OK with ECI set and leaves it untouched */
4812     s->eci_handled = true;
4813     if (arm_dc_feature(s, ARM_FEATURE_M) &&
4814         semihosting_enabled(s->current_el == 0) &&
4815         (a->imm == 0xab)) {
4816         gen_exception_internal_insn(s, EXCP_SEMIHOST);
4817     } else {
4818         gen_exception_bkpt_insn(s, syn_aa32_bkpt(a->imm, false));
4819     }
4820     return true;
4821 }
4822 
4823 static bool trans_HVC(DisasContext *s, arg_HVC *a)
4824 {
4825     if (!ENABLE_ARCH_7 || arm_dc_feature(s, ARM_FEATURE_M)) {
4826         return false;
4827     }
4828     if (IS_USER(s)) {
4829         unallocated_encoding(s);
4830     } else {
4831         gen_hvc(s, a->imm);
4832     }
4833     return true;
4834 }
4835 
4836 static bool trans_SMC(DisasContext *s, arg_SMC *a)
4837 {
4838     if (!ENABLE_ARCH_6K || arm_dc_feature(s, ARM_FEATURE_M)) {
4839         return false;
4840     }
4841     if (IS_USER(s)) {
4842         unallocated_encoding(s);
4843     } else {
4844         gen_smc(s);
4845     }
4846     return true;
4847 }
4848 
4849 static bool trans_SG(DisasContext *s, arg_SG *a)
4850 {
4851     if (!arm_dc_feature(s, ARM_FEATURE_M) ||
4852         !arm_dc_feature(s, ARM_FEATURE_V8)) {
4853         return false;
4854     }
4855     /*
4856      * SG (v8M only)
4857      * The bulk of the behaviour for this instruction is implemented
4858      * in v7m_handle_execute_nsc(), which deals with the insn when
4859      * it is executed by a CPU in non-secure state from memory
4860      * which is Secure & NonSecure-Callable.
4861      * Here we only need to handle the remaining cases:
4862      *  * in NS memory (including the "security extension not
4863      *    implemented" case) : NOP
4864      *  * in S memory but CPU already secure (clear IT bits)
4865      * We know that the attribute for the memory this insn is
4866      * in must match the current CPU state, because otherwise
4867      * get_phys_addr_pmsav8 would have generated an exception.
4868      */
4869     if (s->v8m_secure) {
4870         /* Like the IT insn, we don't need to generate any code */
4871         s->condexec_cond = 0;
4872         s->condexec_mask = 0;
4873     }
4874     return true;
4875 }
4876 
4877 static bool trans_TT(DisasContext *s, arg_TT *a)
4878 {
4879     TCGv_i32 addr, tmp;
4880 
4881     if (!arm_dc_feature(s, ARM_FEATURE_M) ||
4882         !arm_dc_feature(s, ARM_FEATURE_V8)) {
4883         return false;
4884     }
4885     if (a->rd == 13 || a->rd == 15 || a->rn == 15) {
4886         /* We UNDEF for these UNPREDICTABLE cases */
4887         unallocated_encoding(s);
4888         return true;
4889     }
4890     if (a->A && !s->v8m_secure) {
4891         /* This case is UNDEFINED.  */
4892         unallocated_encoding(s);
4893         return true;
4894     }
4895 
4896     addr = load_reg(s, a->rn);
4897     tmp = tcg_temp_new_i32();
4898     gen_helper_v7m_tt(tmp, tcg_env, addr, tcg_constant_i32((a->A << 1) | a->T));
4899     store_reg(s, a->rd, tmp);
4900     return true;
4901 }
4902 
4903 /*
4904  * Load/store register index
4905  */
4906 
4907 static ISSInfo make_issinfo(DisasContext *s, int rd, bool p, bool w)
4908 {
4909     ISSInfo ret;
4910 
4911     /* ISS not valid if writeback */
4912     if (p && !w) {
4913         ret = rd;
4914         if (curr_insn_len(s) == 2) {
4915             ret |= ISSIs16Bit;
4916         }
4917     } else {
4918         ret = ISSInvalid;
4919     }
4920     return ret;
4921 }
4922 
4923 static TCGv_i32 op_addr_rr_pre(DisasContext *s, arg_ldst_rr *a)
4924 {
4925     TCGv_i32 addr = load_reg(s, a->rn);
4926 
4927     if (s->v8m_stackcheck && a->rn == 13 && a->w) {
4928         gen_helper_v8m_stackcheck(tcg_env, addr);
4929     }
4930 
4931     if (a->p) {
4932         TCGv_i32 ofs = load_reg(s, a->rm);
4933         gen_arm_shift_im(ofs, a->shtype, a->shimm, 0);
4934         if (a->u) {
4935             tcg_gen_add_i32(addr, addr, ofs);
4936         } else {
4937             tcg_gen_sub_i32(addr, addr, ofs);
4938         }
4939     }
4940     return addr;
4941 }
4942 
4943 static void op_addr_rr_post(DisasContext *s, arg_ldst_rr *a,
4944                             TCGv_i32 addr, int address_offset)
4945 {
4946     if (!a->p) {
4947         TCGv_i32 ofs = load_reg(s, a->rm);
4948         gen_arm_shift_im(ofs, a->shtype, a->shimm, 0);
4949         if (a->u) {
4950             tcg_gen_add_i32(addr, addr, ofs);
4951         } else {
4952             tcg_gen_sub_i32(addr, addr, ofs);
4953         }
4954     } else if (!a->w) {
4955         return;
4956     }
4957     tcg_gen_addi_i32(addr, addr, address_offset);
4958     store_reg(s, a->rn, addr);
4959 }
4960 
4961 static bool op_load_rr(DisasContext *s, arg_ldst_rr *a,
4962                        MemOp mop, int mem_idx)
4963 {
4964     ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w);
4965     TCGv_i32 addr, tmp;
4966 
4967     addr = op_addr_rr_pre(s, a);
4968 
4969     tmp = tcg_temp_new_i32();
4970     gen_aa32_ld_i32(s, tmp, addr, mem_idx, mop);
4971     disas_set_da_iss(s, mop, issinfo);
4972 
4973     /*
4974      * Perform base writeback before the loaded value to
4975      * ensure correct behavior with overlapping index registers.
4976      */
4977     op_addr_rr_post(s, a, addr, 0);
4978     store_reg_from_load(s, a->rt, tmp);
4979     return true;
4980 }
4981 
4982 static bool op_store_rr(DisasContext *s, arg_ldst_rr *a,
4983                         MemOp mop, int mem_idx)
4984 {
4985     ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w) | ISSIsWrite;
4986     TCGv_i32 addr, tmp;
4987 
4988     /*
4989      * In Thumb encodings of stores Rn=1111 is UNDEF; for Arm it
4990      * is either UNPREDICTABLE or has defined behaviour
4991      */
4992     if (s->thumb && a->rn == 15) {
4993         return false;
4994     }
4995 
4996     addr = op_addr_rr_pre(s, a);
4997 
4998     tmp = load_reg(s, a->rt);
4999     gen_aa32_st_i32(s, tmp, addr, mem_idx, mop);
5000     disas_set_da_iss(s, mop, issinfo);
5001 
5002     op_addr_rr_post(s, a, addr, 0);
5003     return true;
5004 }
5005 
5006 static bool trans_LDRD_rr(DisasContext *s, arg_ldst_rr *a)
5007 {
5008     int mem_idx = get_mem_index(s);
5009     TCGv_i32 addr, tmp;
5010 
5011     if (!ENABLE_ARCH_5TE) {
5012         return false;
5013     }
5014     if (a->rt & 1) {
5015         unallocated_encoding(s);
5016         return true;
5017     }
5018     addr = op_addr_rr_pre(s, a);
5019 
5020     tmp = tcg_temp_new_i32();
5021     gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
5022     store_reg(s, a->rt, tmp);
5023 
5024     tcg_gen_addi_i32(addr, addr, 4);
5025 
5026     tmp = tcg_temp_new_i32();
5027     gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
5028     store_reg(s, a->rt + 1, tmp);
5029 
5030     /* LDRD w/ base writeback is undefined if the registers overlap.  */
5031     op_addr_rr_post(s, a, addr, -4);
5032     return true;
5033 }
5034 
5035 static bool trans_STRD_rr(DisasContext *s, arg_ldst_rr *a)
5036 {
5037     int mem_idx = get_mem_index(s);
5038     TCGv_i32 addr, tmp;
5039 
5040     if (!ENABLE_ARCH_5TE) {
5041         return false;
5042     }
5043     if (a->rt & 1) {
5044         unallocated_encoding(s);
5045         return true;
5046     }
5047     addr = op_addr_rr_pre(s, a);
5048 
5049     tmp = load_reg(s, a->rt);
5050     gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
5051 
5052     tcg_gen_addi_i32(addr, addr, 4);
5053 
5054     tmp = load_reg(s, a->rt + 1);
5055     gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
5056 
5057     op_addr_rr_post(s, a, addr, -4);
5058     return true;
5059 }
5060 
5061 /*
5062  * Load/store immediate index
5063  */
5064 
5065 static TCGv_i32 op_addr_ri_pre(DisasContext *s, arg_ldst_ri *a)
5066 {
5067     int ofs = a->imm;
5068 
5069     if (!a->u) {
5070         ofs = -ofs;
5071     }
5072 
5073     if (s->v8m_stackcheck && a->rn == 13 && a->w) {
5074         /*
5075          * Stackcheck. Here we know 'addr' is the current SP;
5076          * U is set if we're moving SP up, else down. It is
5077          * UNKNOWN whether the limit check triggers when SP starts
5078          * below the limit and ends up above it; we chose to do so.
5079          */
5080         if (!a->u) {
5081             TCGv_i32 newsp = tcg_temp_new_i32();
5082             tcg_gen_addi_i32(newsp, cpu_R[13], ofs);
5083             gen_helper_v8m_stackcheck(tcg_env, newsp);
5084         } else {
5085             gen_helper_v8m_stackcheck(tcg_env, cpu_R[13]);
5086         }
5087     }
5088 
5089     return add_reg_for_lit(s, a->rn, a->p ? ofs : 0);
5090 }
5091 
5092 static void op_addr_ri_post(DisasContext *s, arg_ldst_ri *a,
5093                             TCGv_i32 addr, int address_offset)
5094 {
5095     if (!a->p) {
5096         if (a->u) {
5097             address_offset += a->imm;
5098         } else {
5099             address_offset -= a->imm;
5100         }
5101     } else if (!a->w) {
5102         return;
5103     }
5104     tcg_gen_addi_i32(addr, addr, address_offset);
5105     store_reg(s, a->rn, addr);
5106 }
5107 
5108 static bool op_load_ri(DisasContext *s, arg_ldst_ri *a,
5109                        MemOp mop, int mem_idx)
5110 {
5111     ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w);
5112     TCGv_i32 addr, tmp;
5113 
5114     addr = op_addr_ri_pre(s, a);
5115 
5116     tmp = tcg_temp_new_i32();
5117     gen_aa32_ld_i32(s, tmp, addr, mem_idx, mop);
5118     disas_set_da_iss(s, mop, issinfo);
5119 
5120     /*
5121      * Perform base writeback before the loaded value to
5122      * ensure correct behavior with overlapping index registers.
5123      */
5124     op_addr_ri_post(s, a, addr, 0);
5125     store_reg_from_load(s, a->rt, tmp);
5126     return true;
5127 }
5128 
5129 static bool op_store_ri(DisasContext *s, arg_ldst_ri *a,
5130                         MemOp mop, int mem_idx)
5131 {
5132     ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w) | ISSIsWrite;
5133     TCGv_i32 addr, tmp;
5134 
5135     /*
5136      * In Thumb encodings of stores Rn=1111 is UNDEF; for Arm it
5137      * is either UNPREDICTABLE or has defined behaviour
5138      */
5139     if (s->thumb && a->rn == 15) {
5140         return false;
5141     }
5142 
5143     addr = op_addr_ri_pre(s, a);
5144 
5145     tmp = load_reg(s, a->rt);
5146     gen_aa32_st_i32(s, tmp, addr, mem_idx, mop);
5147     disas_set_da_iss(s, mop, issinfo);
5148 
5149     op_addr_ri_post(s, a, addr, 0);
5150     return true;
5151 }
5152 
5153 static bool op_ldrd_ri(DisasContext *s, arg_ldst_ri *a, int rt2)
5154 {
5155     int mem_idx = get_mem_index(s);
5156     TCGv_i32 addr, tmp;
5157 
5158     addr = op_addr_ri_pre(s, a);
5159 
5160     tmp = tcg_temp_new_i32();
5161     gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
5162     store_reg(s, a->rt, tmp);
5163 
5164     tcg_gen_addi_i32(addr, addr, 4);
5165 
5166     tmp = tcg_temp_new_i32();
5167     gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
5168     store_reg(s, rt2, tmp);
5169 
5170     /* LDRD w/ base writeback is undefined if the registers overlap.  */
5171     op_addr_ri_post(s, a, addr, -4);
5172     return true;
5173 }
5174 
5175 static bool trans_LDRD_ri_a32(DisasContext *s, arg_ldst_ri *a)
5176 {
5177     if (!ENABLE_ARCH_5TE || (a->rt & 1)) {
5178         return false;
5179     }
5180     return op_ldrd_ri(s, a, a->rt + 1);
5181 }
5182 
5183 static bool trans_LDRD_ri_t32(DisasContext *s, arg_ldst_ri2 *a)
5184 {
5185     arg_ldst_ri b = {
5186         .u = a->u, .w = a->w, .p = a->p,
5187         .rn = a->rn, .rt = a->rt, .imm = a->imm
5188     };
5189     return op_ldrd_ri(s, &b, a->rt2);
5190 }
5191 
5192 static bool op_strd_ri(DisasContext *s, arg_ldst_ri *a, int rt2)
5193 {
5194     int mem_idx = get_mem_index(s);
5195     TCGv_i32 addr, tmp;
5196 
5197     addr = op_addr_ri_pre(s, a);
5198 
5199     tmp = load_reg(s, a->rt);
5200     gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
5201 
5202     tcg_gen_addi_i32(addr, addr, 4);
5203 
5204     tmp = load_reg(s, rt2);
5205     gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
5206 
5207     op_addr_ri_post(s, a, addr, -4);
5208     return true;
5209 }
5210 
5211 static bool trans_STRD_ri_a32(DisasContext *s, arg_ldst_ri *a)
5212 {
5213     if (!ENABLE_ARCH_5TE || (a->rt & 1)) {
5214         return false;
5215     }
5216     return op_strd_ri(s, a, a->rt + 1);
5217 }
5218 
5219 static bool trans_STRD_ri_t32(DisasContext *s, arg_ldst_ri2 *a)
5220 {
5221     arg_ldst_ri b = {
5222         .u = a->u, .w = a->w, .p = a->p,
5223         .rn = a->rn, .rt = a->rt, .imm = a->imm
5224     };
5225     return op_strd_ri(s, &b, a->rt2);
5226 }
5227 
5228 #define DO_LDST(NAME, WHICH, MEMOP) \
5229 static bool trans_##NAME##_ri(DisasContext *s, arg_ldst_ri *a)        \
5230 {                                                                     \
5231     return op_##WHICH##_ri(s, a, MEMOP, get_mem_index(s));            \
5232 }                                                                     \
5233 static bool trans_##NAME##T_ri(DisasContext *s, arg_ldst_ri *a)       \
5234 {                                                                     \
5235     return op_##WHICH##_ri(s, a, MEMOP, get_a32_user_mem_index(s));   \
5236 }                                                                     \
5237 static bool trans_##NAME##_rr(DisasContext *s, arg_ldst_rr *a)        \
5238 {                                                                     \
5239     return op_##WHICH##_rr(s, a, MEMOP, get_mem_index(s));            \
5240 }                                                                     \
5241 static bool trans_##NAME##T_rr(DisasContext *s, arg_ldst_rr *a)       \
5242 {                                                                     \
5243     return op_##WHICH##_rr(s, a, MEMOP, get_a32_user_mem_index(s));   \
5244 }
5245 
5246 DO_LDST(LDR, load, MO_UL)
5247 DO_LDST(LDRB, load, MO_UB)
5248 DO_LDST(LDRH, load, MO_UW)
5249 DO_LDST(LDRSB, load, MO_SB)
5250 DO_LDST(LDRSH, load, MO_SW)
5251 
5252 DO_LDST(STR, store, MO_UL)
5253 DO_LDST(STRB, store, MO_UB)
5254 DO_LDST(STRH, store, MO_UW)
5255 
5256 #undef DO_LDST
5257 
5258 /*
5259  * Synchronization primitives
5260  */
5261 
5262 static bool op_swp(DisasContext *s, arg_SWP *a, MemOp opc)
5263 {
5264     TCGv_i32 addr, tmp;
5265     TCGv taddr;
5266 
5267     opc |= s->be_data;
5268     addr = load_reg(s, a->rn);
5269     taddr = gen_aa32_addr(s, addr, opc);
5270 
5271     tmp = load_reg(s, a->rt2);
5272     tcg_gen_atomic_xchg_i32(tmp, taddr, tmp, get_mem_index(s), opc);
5273 
5274     store_reg(s, a->rt, tmp);
5275     return true;
5276 }
5277 
5278 static bool trans_SWP(DisasContext *s, arg_SWP *a)
5279 {
5280     return op_swp(s, a, MO_UL | MO_ALIGN);
5281 }
5282 
5283 static bool trans_SWPB(DisasContext *s, arg_SWP *a)
5284 {
5285     return op_swp(s, a, MO_UB);
5286 }
5287 
5288 /*
5289  * Load/Store Exclusive and Load-Acquire/Store-Release
5290  */
5291 
5292 static bool op_strex(DisasContext *s, arg_STREX *a, MemOp mop, bool rel)
5293 {
5294     TCGv_i32 addr;
5295     /* Some cases stopped being UNPREDICTABLE in v8A (but not v8M) */
5296     bool v8a = ENABLE_ARCH_8 && !arm_dc_feature(s, ARM_FEATURE_M);
5297 
5298     /* We UNDEF for these UNPREDICTABLE cases.  */
5299     if (a->rd == 15 || a->rn == 15 || a->rt == 15
5300         || a->rd == a->rn || a->rd == a->rt
5301         || (!v8a && s->thumb && (a->rd == 13 || a->rt == 13))
5302         || (mop == MO_64
5303             && (a->rt2 == 15
5304                 || a->rd == a->rt2
5305                 || (!v8a && s->thumb && a->rt2 == 13)))) {
5306         unallocated_encoding(s);
5307         return true;
5308     }
5309 
5310     if (rel) {
5311         tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
5312     }
5313 
5314     addr = tcg_temp_new_i32();
5315     load_reg_var(s, addr, a->rn);
5316     tcg_gen_addi_i32(addr, addr, a->imm);
5317 
5318     gen_store_exclusive(s, a->rd, a->rt, a->rt2, addr, mop);
5319     return true;
5320 }
5321 
5322 static bool trans_STREX(DisasContext *s, arg_STREX *a)
5323 {
5324     if (!ENABLE_ARCH_6) {
5325         return false;
5326     }
5327     return op_strex(s, a, MO_32, false);
5328 }
5329 
5330 static bool trans_STREXD_a32(DisasContext *s, arg_STREX *a)
5331 {
5332     if (!ENABLE_ARCH_6K) {
5333         return false;
5334     }
5335     /* We UNDEF for these UNPREDICTABLE cases.  */
5336     if (a->rt & 1) {
5337         unallocated_encoding(s);
5338         return true;
5339     }
5340     a->rt2 = a->rt + 1;
5341     return op_strex(s, a, MO_64, false);
5342 }
5343 
5344 static bool trans_STREXD_t32(DisasContext *s, arg_STREX *a)
5345 {
5346     return op_strex(s, a, MO_64, false);
5347 }
5348 
5349 static bool trans_STREXB(DisasContext *s, arg_STREX *a)
5350 {
5351     if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
5352         return false;
5353     }
5354     return op_strex(s, a, MO_8, false);
5355 }
5356 
5357 static bool trans_STREXH(DisasContext *s, arg_STREX *a)
5358 {
5359     if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
5360         return false;
5361     }
5362     return op_strex(s, a, MO_16, false);
5363 }
5364 
5365 static bool trans_STLEX(DisasContext *s, arg_STREX *a)
5366 {
5367     if (!ENABLE_ARCH_8) {
5368         return false;
5369     }
5370     return op_strex(s, a, MO_32, true);
5371 }
5372 
5373 static bool trans_STLEXD_a32(DisasContext *s, arg_STREX *a)
5374 {
5375     if (!ENABLE_ARCH_8) {
5376         return false;
5377     }
5378     /* We UNDEF for these UNPREDICTABLE cases.  */
5379     if (a->rt & 1) {
5380         unallocated_encoding(s);
5381         return true;
5382     }
5383     a->rt2 = a->rt + 1;
5384     return op_strex(s, a, MO_64, true);
5385 }
5386 
5387 static bool trans_STLEXD_t32(DisasContext *s, arg_STREX *a)
5388 {
5389     if (!ENABLE_ARCH_8) {
5390         return false;
5391     }
5392     return op_strex(s, a, MO_64, true);
5393 }
5394 
5395 static bool trans_STLEXB(DisasContext *s, arg_STREX *a)
5396 {
5397     if (!ENABLE_ARCH_8) {
5398         return false;
5399     }
5400     return op_strex(s, a, MO_8, true);
5401 }
5402 
5403 static bool trans_STLEXH(DisasContext *s, arg_STREX *a)
5404 {
5405     if (!ENABLE_ARCH_8) {
5406         return false;
5407     }
5408     return op_strex(s, a, MO_16, true);
5409 }
5410 
5411 static bool op_stl(DisasContext *s, arg_STL *a, MemOp mop)
5412 {
5413     TCGv_i32 addr, tmp;
5414 
5415     if (!ENABLE_ARCH_8) {
5416         return false;
5417     }
5418     /* We UNDEF for these UNPREDICTABLE cases.  */
5419     if (a->rn == 15 || a->rt == 15) {
5420         unallocated_encoding(s);
5421         return true;
5422     }
5423 
5424     addr = load_reg(s, a->rn);
5425     tmp = load_reg(s, a->rt);
5426     tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
5427     gen_aa32_st_i32(s, tmp, addr, get_mem_index(s), mop | MO_ALIGN);
5428     disas_set_da_iss(s, mop, a->rt | ISSIsAcqRel | ISSIsWrite);
5429 
5430     return true;
5431 }
5432 
5433 static bool trans_STL(DisasContext *s, arg_STL *a)
5434 {
5435     return op_stl(s, a, MO_UL);
5436 }
5437 
5438 static bool trans_STLB(DisasContext *s, arg_STL *a)
5439 {
5440     return op_stl(s, a, MO_UB);
5441 }
5442 
5443 static bool trans_STLH(DisasContext *s, arg_STL *a)
5444 {
5445     return op_stl(s, a, MO_UW);
5446 }
5447 
5448 static bool op_ldrex(DisasContext *s, arg_LDREX *a, MemOp mop, bool acq)
5449 {
5450     TCGv_i32 addr;
5451     /* Some cases stopped being UNPREDICTABLE in v8A (but not v8M) */
5452     bool v8a = ENABLE_ARCH_8 && !arm_dc_feature(s, ARM_FEATURE_M);
5453 
5454     /* We UNDEF for these UNPREDICTABLE cases.  */
5455     if (a->rn == 15 || a->rt == 15
5456         || (!v8a && s->thumb && a->rt == 13)
5457         || (mop == MO_64
5458             && (a->rt2 == 15 || a->rt == a->rt2
5459                 || (!v8a && s->thumb && a->rt2 == 13)))) {
5460         unallocated_encoding(s);
5461         return true;
5462     }
5463 
5464     addr = tcg_temp_new_i32();
5465     load_reg_var(s, addr, a->rn);
5466     tcg_gen_addi_i32(addr, addr, a->imm);
5467 
5468     gen_load_exclusive(s, a->rt, a->rt2, addr, mop);
5469 
5470     if (acq) {
5471         tcg_gen_mb(TCG_MO_ALL | TCG_BAR_LDAQ);
5472     }
5473     return true;
5474 }
5475 
5476 static bool trans_LDREX(DisasContext *s, arg_LDREX *a)
5477 {
5478     if (!ENABLE_ARCH_6) {
5479         return false;
5480     }
5481     return op_ldrex(s, a, MO_32, false);
5482 }
5483 
5484 static bool trans_LDREXD_a32(DisasContext *s, arg_LDREX *a)
5485 {
5486     if (!ENABLE_ARCH_6K) {
5487         return false;
5488     }
5489     /* We UNDEF for these UNPREDICTABLE cases.  */
5490     if (a->rt & 1) {
5491         unallocated_encoding(s);
5492         return true;
5493     }
5494     a->rt2 = a->rt + 1;
5495     return op_ldrex(s, a, MO_64, false);
5496 }
5497 
5498 static bool trans_LDREXD_t32(DisasContext *s, arg_LDREX *a)
5499 {
5500     return op_ldrex(s, a, MO_64, false);
5501 }
5502 
5503 static bool trans_LDREXB(DisasContext *s, arg_LDREX *a)
5504 {
5505     if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
5506         return false;
5507     }
5508     return op_ldrex(s, a, MO_8, false);
5509 }
5510 
5511 static bool trans_LDREXH(DisasContext *s, arg_LDREX *a)
5512 {
5513     if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
5514         return false;
5515     }
5516     return op_ldrex(s, a, MO_16, false);
5517 }
5518 
5519 static bool trans_LDAEX(DisasContext *s, arg_LDREX *a)
5520 {
5521     if (!ENABLE_ARCH_8) {
5522         return false;
5523     }
5524     return op_ldrex(s, a, MO_32, true);
5525 }
5526 
5527 static bool trans_LDAEXD_a32(DisasContext *s, arg_LDREX *a)
5528 {
5529     if (!ENABLE_ARCH_8) {
5530         return false;
5531     }
5532     /* We UNDEF for these UNPREDICTABLE cases.  */
5533     if (a->rt & 1) {
5534         unallocated_encoding(s);
5535         return true;
5536     }
5537     a->rt2 = a->rt + 1;
5538     return op_ldrex(s, a, MO_64, true);
5539 }
5540 
5541 static bool trans_LDAEXD_t32(DisasContext *s, arg_LDREX *a)
5542 {
5543     if (!ENABLE_ARCH_8) {
5544         return false;
5545     }
5546     return op_ldrex(s, a, MO_64, true);
5547 }
5548 
5549 static bool trans_LDAEXB(DisasContext *s, arg_LDREX *a)
5550 {
5551     if (!ENABLE_ARCH_8) {
5552         return false;
5553     }
5554     return op_ldrex(s, a, MO_8, true);
5555 }
5556 
5557 static bool trans_LDAEXH(DisasContext *s, arg_LDREX *a)
5558 {
5559     if (!ENABLE_ARCH_8) {
5560         return false;
5561     }
5562     return op_ldrex(s, a, MO_16, true);
5563 }
5564 
5565 static bool op_lda(DisasContext *s, arg_LDA *a, MemOp mop)
5566 {
5567     TCGv_i32 addr, tmp;
5568 
5569     if (!ENABLE_ARCH_8) {
5570         return false;
5571     }
5572     /* We UNDEF for these UNPREDICTABLE cases.  */
5573     if (a->rn == 15 || a->rt == 15) {
5574         unallocated_encoding(s);
5575         return true;
5576     }
5577 
5578     addr = load_reg(s, a->rn);
5579     tmp = tcg_temp_new_i32();
5580     gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), mop | MO_ALIGN);
5581     disas_set_da_iss(s, mop, a->rt | ISSIsAcqRel);
5582 
5583     store_reg(s, a->rt, tmp);
5584     tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
5585     return true;
5586 }
5587 
5588 static bool trans_LDA(DisasContext *s, arg_LDA *a)
5589 {
5590     return op_lda(s, a, MO_UL);
5591 }
5592 
5593 static bool trans_LDAB(DisasContext *s, arg_LDA *a)
5594 {
5595     return op_lda(s, a, MO_UB);
5596 }
5597 
5598 static bool trans_LDAH(DisasContext *s, arg_LDA *a)
5599 {
5600     return op_lda(s, a, MO_UW);
5601 }
5602 
5603 /*
5604  * Media instructions
5605  */
5606 
5607 static bool trans_USADA8(DisasContext *s, arg_USADA8 *a)
5608 {
5609     TCGv_i32 t1, t2;
5610 
5611     if (!ENABLE_ARCH_6) {
5612         return false;
5613     }
5614 
5615     t1 = load_reg(s, a->rn);
5616     t2 = load_reg(s, a->rm);
5617     gen_helper_usad8(t1, t1, t2);
5618     if (a->ra != 15) {
5619         t2 = load_reg(s, a->ra);
5620         tcg_gen_add_i32(t1, t1, t2);
5621     }
5622     store_reg(s, a->rd, t1);
5623     return true;
5624 }
5625 
5626 static bool op_bfx(DisasContext *s, arg_UBFX *a, bool u)
5627 {
5628     TCGv_i32 tmp;
5629     int width = a->widthm1 + 1;
5630     int shift = a->lsb;
5631 
5632     if (!ENABLE_ARCH_6T2) {
5633         return false;
5634     }
5635     if (shift + width > 32) {
5636         /* UNPREDICTABLE; we choose to UNDEF */
5637         unallocated_encoding(s);
5638         return true;
5639     }
5640 
5641     tmp = load_reg(s, a->rn);
5642     if (u) {
5643         tcg_gen_extract_i32(tmp, tmp, shift, width);
5644     } else {
5645         tcg_gen_sextract_i32(tmp, tmp, shift, width);
5646     }
5647     store_reg(s, a->rd, tmp);
5648     return true;
5649 }
5650 
5651 static bool trans_SBFX(DisasContext *s, arg_SBFX *a)
5652 {
5653     return op_bfx(s, a, false);
5654 }
5655 
5656 static bool trans_UBFX(DisasContext *s, arg_UBFX *a)
5657 {
5658     return op_bfx(s, a, true);
5659 }
5660 
5661 static bool trans_BFCI(DisasContext *s, arg_BFCI *a)
5662 {
5663     int msb = a->msb, lsb = a->lsb;
5664     TCGv_i32 t_in, t_rd;
5665     int width;
5666 
5667     if (!ENABLE_ARCH_6T2) {
5668         return false;
5669     }
5670     if (msb < lsb) {
5671         /* UNPREDICTABLE; we choose to UNDEF */
5672         unallocated_encoding(s);
5673         return true;
5674     }
5675 
5676     width = msb + 1 - lsb;
5677     if (a->rn == 15) {
5678         /* BFC */
5679         t_in = tcg_constant_i32(0);
5680     } else {
5681         /* BFI */
5682         t_in = load_reg(s, a->rn);
5683     }
5684     t_rd = load_reg(s, a->rd);
5685     tcg_gen_deposit_i32(t_rd, t_rd, t_in, lsb, width);
5686     store_reg(s, a->rd, t_rd);
5687     return true;
5688 }
5689 
5690 static bool trans_UDF(DisasContext *s, arg_UDF *a)
5691 {
5692     unallocated_encoding(s);
5693     return true;
5694 }
5695 
5696 /*
5697  * Parallel addition and subtraction
5698  */
5699 
5700 static bool op_par_addsub(DisasContext *s, arg_rrr *a,
5701                           void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32))
5702 {
5703     TCGv_i32 t0, t1;
5704 
5705     if (s->thumb
5706         ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
5707         : !ENABLE_ARCH_6) {
5708         return false;
5709     }
5710 
5711     t0 = load_reg(s, a->rn);
5712     t1 = load_reg(s, a->rm);
5713 
5714     gen(t0, t0, t1);
5715 
5716     store_reg(s, a->rd, t0);
5717     return true;
5718 }
5719 
5720 static bool op_par_addsub_ge(DisasContext *s, arg_rrr *a,
5721                              void (*gen)(TCGv_i32, TCGv_i32,
5722                                          TCGv_i32, TCGv_ptr))
5723 {
5724     TCGv_i32 t0, t1;
5725     TCGv_ptr ge;
5726 
5727     if (s->thumb
5728         ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
5729         : !ENABLE_ARCH_6) {
5730         return false;
5731     }
5732 
5733     t0 = load_reg(s, a->rn);
5734     t1 = load_reg(s, a->rm);
5735 
5736     ge = tcg_temp_new_ptr();
5737     tcg_gen_addi_ptr(ge, tcg_env, offsetof(CPUARMState, GE));
5738     gen(t0, t0, t1, ge);
5739 
5740     store_reg(s, a->rd, t0);
5741     return true;
5742 }
5743 
5744 #define DO_PAR_ADDSUB(NAME, helper) \
5745 static bool trans_##NAME(DisasContext *s, arg_rrr *a)   \
5746 {                                                       \
5747     return op_par_addsub(s, a, helper);                 \
5748 }
5749 
5750 #define DO_PAR_ADDSUB_GE(NAME, helper) \
5751 static bool trans_##NAME(DisasContext *s, arg_rrr *a)   \
5752 {                                                       \
5753     return op_par_addsub_ge(s, a, helper);              \
5754 }
5755 
5756 DO_PAR_ADDSUB_GE(SADD16, gen_helper_sadd16)
5757 DO_PAR_ADDSUB_GE(SASX, gen_helper_saddsubx)
5758 DO_PAR_ADDSUB_GE(SSAX, gen_helper_ssubaddx)
5759 DO_PAR_ADDSUB_GE(SSUB16, gen_helper_ssub16)
5760 DO_PAR_ADDSUB_GE(SADD8, gen_helper_sadd8)
5761 DO_PAR_ADDSUB_GE(SSUB8, gen_helper_ssub8)
5762 
5763 DO_PAR_ADDSUB_GE(UADD16, gen_helper_uadd16)
5764 DO_PAR_ADDSUB_GE(UASX, gen_helper_uaddsubx)
5765 DO_PAR_ADDSUB_GE(USAX, gen_helper_usubaddx)
5766 DO_PAR_ADDSUB_GE(USUB16, gen_helper_usub16)
5767 DO_PAR_ADDSUB_GE(UADD8, gen_helper_uadd8)
5768 DO_PAR_ADDSUB_GE(USUB8, gen_helper_usub8)
5769 
5770 DO_PAR_ADDSUB(QADD16, gen_helper_qadd16)
5771 DO_PAR_ADDSUB(QASX, gen_helper_qaddsubx)
5772 DO_PAR_ADDSUB(QSAX, gen_helper_qsubaddx)
5773 DO_PAR_ADDSUB(QSUB16, gen_helper_qsub16)
5774 DO_PAR_ADDSUB(QADD8, gen_helper_qadd8)
5775 DO_PAR_ADDSUB(QSUB8, gen_helper_qsub8)
5776 
5777 DO_PAR_ADDSUB(UQADD16, gen_helper_uqadd16)
5778 DO_PAR_ADDSUB(UQASX, gen_helper_uqaddsubx)
5779 DO_PAR_ADDSUB(UQSAX, gen_helper_uqsubaddx)
5780 DO_PAR_ADDSUB(UQSUB16, gen_helper_uqsub16)
5781 DO_PAR_ADDSUB(UQADD8, gen_helper_uqadd8)
5782 DO_PAR_ADDSUB(UQSUB8, gen_helper_uqsub8)
5783 
5784 DO_PAR_ADDSUB(SHADD16, gen_helper_shadd16)
5785 DO_PAR_ADDSUB(SHASX, gen_helper_shaddsubx)
5786 DO_PAR_ADDSUB(SHSAX, gen_helper_shsubaddx)
5787 DO_PAR_ADDSUB(SHSUB16, gen_helper_shsub16)
5788 DO_PAR_ADDSUB(SHADD8, gen_helper_shadd8)
5789 DO_PAR_ADDSUB(SHSUB8, gen_helper_shsub8)
5790 
5791 DO_PAR_ADDSUB(UHADD16, gen_helper_uhadd16)
5792 DO_PAR_ADDSUB(UHASX, gen_helper_uhaddsubx)
5793 DO_PAR_ADDSUB(UHSAX, gen_helper_uhsubaddx)
5794 DO_PAR_ADDSUB(UHSUB16, gen_helper_uhsub16)
5795 DO_PAR_ADDSUB(UHADD8, gen_helper_uhadd8)
5796 DO_PAR_ADDSUB(UHSUB8, gen_helper_uhsub8)
5797 
5798 #undef DO_PAR_ADDSUB
5799 #undef DO_PAR_ADDSUB_GE
5800 
5801 /*
5802  * Packing, unpacking, saturation, and reversal
5803  */
5804 
5805 static bool trans_PKH(DisasContext *s, arg_PKH *a)
5806 {
5807     TCGv_i32 tn, tm;
5808     int shift = a->imm;
5809 
5810     if (s->thumb
5811         ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
5812         : !ENABLE_ARCH_6) {
5813         return false;
5814     }
5815 
5816     tn = load_reg(s, a->rn);
5817     tm = load_reg(s, a->rm);
5818     if (a->tb) {
5819         /* PKHTB */
5820         if (shift == 0) {
5821             shift = 31;
5822         }
5823         tcg_gen_sari_i32(tm, tm, shift);
5824         tcg_gen_deposit_i32(tn, tn, tm, 0, 16);
5825     } else {
5826         /* PKHBT */
5827         tcg_gen_shli_i32(tm, tm, shift);
5828         tcg_gen_deposit_i32(tn, tm, tn, 0, 16);
5829     }
5830     store_reg(s, a->rd, tn);
5831     return true;
5832 }
5833 
5834 static bool op_sat(DisasContext *s, arg_sat *a,
5835                    void (*gen)(TCGv_i32, TCGv_env, TCGv_i32, TCGv_i32))
5836 {
5837     TCGv_i32 tmp;
5838     int shift = a->imm;
5839 
5840     if (!ENABLE_ARCH_6) {
5841         return false;
5842     }
5843 
5844     tmp = load_reg(s, a->rn);
5845     if (a->sh) {
5846         tcg_gen_sari_i32(tmp, tmp, shift ? shift : 31);
5847     } else {
5848         tcg_gen_shli_i32(tmp, tmp, shift);
5849     }
5850 
5851     gen(tmp, tcg_env, tmp, tcg_constant_i32(a->satimm));
5852 
5853     store_reg(s, a->rd, tmp);
5854     return true;
5855 }
5856 
5857 static bool trans_SSAT(DisasContext *s, arg_sat *a)
5858 {
5859     return op_sat(s, a, gen_helper_ssat);
5860 }
5861 
5862 static bool trans_USAT(DisasContext *s, arg_sat *a)
5863 {
5864     return op_sat(s, a, gen_helper_usat);
5865 }
5866 
5867 static bool trans_SSAT16(DisasContext *s, arg_sat *a)
5868 {
5869     if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
5870         return false;
5871     }
5872     return op_sat(s, a, gen_helper_ssat16);
5873 }
5874 
5875 static bool trans_USAT16(DisasContext *s, arg_sat *a)
5876 {
5877     if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
5878         return false;
5879     }
5880     return op_sat(s, a, gen_helper_usat16);
5881 }
5882 
5883 static bool op_xta(DisasContext *s, arg_rrr_rot *a,
5884                    void (*gen_extract)(TCGv_i32, TCGv_i32),
5885                    void (*gen_add)(TCGv_i32, TCGv_i32, TCGv_i32))
5886 {
5887     TCGv_i32 tmp;
5888 
5889     if (!ENABLE_ARCH_6) {
5890         return false;
5891     }
5892 
5893     tmp = load_reg(s, a->rm);
5894     /*
5895      * TODO: In many cases we could do a shift instead of a rotate.
5896      * Combined with a simple extend, that becomes an extract.
5897      */
5898     tcg_gen_rotri_i32(tmp, tmp, a->rot * 8);
5899     gen_extract(tmp, tmp);
5900 
5901     if (a->rn != 15) {
5902         TCGv_i32 tmp2 = load_reg(s, a->rn);
5903         gen_add(tmp, tmp, tmp2);
5904     }
5905     store_reg(s, a->rd, tmp);
5906     return true;
5907 }
5908 
5909 static bool trans_SXTAB(DisasContext *s, arg_rrr_rot *a)
5910 {
5911     return op_xta(s, a, tcg_gen_ext8s_i32, tcg_gen_add_i32);
5912 }
5913 
5914 static bool trans_SXTAH(DisasContext *s, arg_rrr_rot *a)
5915 {
5916     return op_xta(s, a, tcg_gen_ext16s_i32, tcg_gen_add_i32);
5917 }
5918 
5919 static bool trans_SXTAB16(DisasContext *s, arg_rrr_rot *a)
5920 {
5921     if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
5922         return false;
5923     }
5924     return op_xta(s, a, gen_helper_sxtb16, gen_add16);
5925 }
5926 
5927 static bool trans_UXTAB(DisasContext *s, arg_rrr_rot *a)
5928 {
5929     return op_xta(s, a, tcg_gen_ext8u_i32, tcg_gen_add_i32);
5930 }
5931 
5932 static bool trans_UXTAH(DisasContext *s, arg_rrr_rot *a)
5933 {
5934     return op_xta(s, a, tcg_gen_ext16u_i32, tcg_gen_add_i32);
5935 }
5936 
5937 static bool trans_UXTAB16(DisasContext *s, arg_rrr_rot *a)
5938 {
5939     if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
5940         return false;
5941     }
5942     return op_xta(s, a, gen_helper_uxtb16, gen_add16);
5943 }
5944 
5945 static bool trans_SEL(DisasContext *s, arg_rrr *a)
5946 {
5947     TCGv_i32 t1, t2, t3;
5948 
5949     if (s->thumb
5950         ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
5951         : !ENABLE_ARCH_6) {
5952         return false;
5953     }
5954 
5955     t1 = load_reg(s, a->rn);
5956     t2 = load_reg(s, a->rm);
5957     t3 = tcg_temp_new_i32();
5958     tcg_gen_ld_i32(t3, tcg_env, offsetof(CPUARMState, GE));
5959     gen_helper_sel_flags(t1, t3, t1, t2);
5960     store_reg(s, a->rd, t1);
5961     return true;
5962 }
5963 
5964 static bool op_rr(DisasContext *s, arg_rr *a,
5965                   void (*gen)(TCGv_i32, TCGv_i32))
5966 {
5967     TCGv_i32 tmp;
5968 
5969     tmp = load_reg(s, a->rm);
5970     gen(tmp, tmp);
5971     store_reg(s, a->rd, tmp);
5972     return true;
5973 }
5974 
5975 static bool trans_REV(DisasContext *s, arg_rr *a)
5976 {
5977     if (!ENABLE_ARCH_6) {
5978         return false;
5979     }
5980     return op_rr(s, a, tcg_gen_bswap32_i32);
5981 }
5982 
5983 static bool trans_REV16(DisasContext *s, arg_rr *a)
5984 {
5985     if (!ENABLE_ARCH_6) {
5986         return false;
5987     }
5988     return op_rr(s, a, gen_rev16);
5989 }
5990 
5991 static bool trans_REVSH(DisasContext *s, arg_rr *a)
5992 {
5993     if (!ENABLE_ARCH_6) {
5994         return false;
5995     }
5996     return op_rr(s, a, gen_revsh);
5997 }
5998 
5999 static bool trans_RBIT(DisasContext *s, arg_rr *a)
6000 {
6001     if (!ENABLE_ARCH_6T2) {
6002         return false;
6003     }
6004     return op_rr(s, a, gen_helper_rbit);
6005 }
6006 
6007 /*
6008  * Signed multiply, signed and unsigned divide
6009  */
6010 
6011 static bool op_smlad(DisasContext *s, arg_rrrr *a, bool m_swap, bool sub)
6012 {
6013     TCGv_i32 t1, t2;
6014 
6015     if (!ENABLE_ARCH_6) {
6016         return false;
6017     }
6018 
6019     t1 = load_reg(s, a->rn);
6020     t2 = load_reg(s, a->rm);
6021     if (m_swap) {
6022         gen_swap_half(t2, t2);
6023     }
6024     gen_smul_dual(t1, t2);
6025 
6026     if (sub) {
6027         /*
6028          * This subtraction cannot overflow, so we can do a simple
6029          * 32-bit subtraction and then a possible 32-bit saturating
6030          * addition of Ra.
6031          */
6032         tcg_gen_sub_i32(t1, t1, t2);
6033 
6034         if (a->ra != 15) {
6035             t2 = load_reg(s, a->ra);
6036             gen_helper_add_setq(t1, tcg_env, t1, t2);
6037         }
6038     } else if (a->ra == 15) {
6039         /* Single saturation-checking addition */
6040         gen_helper_add_setq(t1, tcg_env, t1, t2);
6041     } else {
6042         /*
6043          * We need to add the products and Ra together and then
6044          * determine whether the final result overflowed. Doing
6045          * this as two separate add-and-check-overflow steps incorrectly
6046          * sets Q for cases like (-32768 * -32768) + (-32768 * -32768) + -1.
6047          * Do all the arithmetic at 64-bits and then check for overflow.
6048          */
6049         TCGv_i64 p64, q64;
6050         TCGv_i32 t3, qf, one;
6051 
6052         p64 = tcg_temp_new_i64();
6053         q64 = tcg_temp_new_i64();
6054         tcg_gen_ext_i32_i64(p64, t1);
6055         tcg_gen_ext_i32_i64(q64, t2);
6056         tcg_gen_add_i64(p64, p64, q64);
6057         load_reg_var(s, t2, a->ra);
6058         tcg_gen_ext_i32_i64(q64, t2);
6059         tcg_gen_add_i64(p64, p64, q64);
6060 
6061         tcg_gen_extr_i64_i32(t1, t2, p64);
6062         /*
6063          * t1 is the low half of the result which goes into Rd.
6064          * We have overflow and must set Q if the high half (t2)
6065          * is different from the sign-extension of t1.
6066          */
6067         t3 = tcg_temp_new_i32();
6068         tcg_gen_sari_i32(t3, t1, 31);
6069         qf = load_cpu_field(QF);
6070         one = tcg_constant_i32(1);
6071         tcg_gen_movcond_i32(TCG_COND_NE, qf, t2, t3, one, qf);
6072         store_cpu_field(qf, QF);
6073     }
6074     store_reg(s, a->rd, t1);
6075     return true;
6076 }
6077 
6078 static bool trans_SMLAD(DisasContext *s, arg_rrrr *a)
6079 {
6080     return op_smlad(s, a, false, false);
6081 }
6082 
6083 static bool trans_SMLADX(DisasContext *s, arg_rrrr *a)
6084 {
6085     return op_smlad(s, a, true, false);
6086 }
6087 
6088 static bool trans_SMLSD(DisasContext *s, arg_rrrr *a)
6089 {
6090     return op_smlad(s, a, false, true);
6091 }
6092 
6093 static bool trans_SMLSDX(DisasContext *s, arg_rrrr *a)
6094 {
6095     return op_smlad(s, a, true, true);
6096 }
6097 
6098 static bool op_smlald(DisasContext *s, arg_rrrr *a, bool m_swap, bool sub)
6099 {
6100     TCGv_i32 t1, t2;
6101     TCGv_i64 l1, l2;
6102 
6103     if (!ENABLE_ARCH_6) {
6104         return false;
6105     }
6106 
6107     t1 = load_reg(s, a->rn);
6108     t2 = load_reg(s, a->rm);
6109     if (m_swap) {
6110         gen_swap_half(t2, t2);
6111     }
6112     gen_smul_dual(t1, t2);
6113 
6114     l1 = tcg_temp_new_i64();
6115     l2 = tcg_temp_new_i64();
6116     tcg_gen_ext_i32_i64(l1, t1);
6117     tcg_gen_ext_i32_i64(l2, t2);
6118 
6119     if (sub) {
6120         tcg_gen_sub_i64(l1, l1, l2);
6121     } else {
6122         tcg_gen_add_i64(l1, l1, l2);
6123     }
6124 
6125     gen_addq(s, l1, a->ra, a->rd);
6126     gen_storeq_reg(s, a->ra, a->rd, l1);
6127     return true;
6128 }
6129 
6130 static bool trans_SMLALD(DisasContext *s, arg_rrrr *a)
6131 {
6132     return op_smlald(s, a, false, false);
6133 }
6134 
6135 static bool trans_SMLALDX(DisasContext *s, arg_rrrr *a)
6136 {
6137     return op_smlald(s, a, true, false);
6138 }
6139 
6140 static bool trans_SMLSLD(DisasContext *s, arg_rrrr *a)
6141 {
6142     return op_smlald(s, a, false, true);
6143 }
6144 
6145 static bool trans_SMLSLDX(DisasContext *s, arg_rrrr *a)
6146 {
6147     return op_smlald(s, a, true, true);
6148 }
6149 
6150 static bool op_smmla(DisasContext *s, arg_rrrr *a, bool round, bool sub)
6151 {
6152     TCGv_i32 t1, t2;
6153 
6154     if (s->thumb
6155         ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
6156         : !ENABLE_ARCH_6) {
6157         return false;
6158     }
6159 
6160     t1 = load_reg(s, a->rn);
6161     t2 = load_reg(s, a->rm);
6162     tcg_gen_muls2_i32(t2, t1, t1, t2);
6163 
6164     if (a->ra != 15) {
6165         TCGv_i32 t3 = load_reg(s, a->ra);
6166         if (sub) {
6167             /*
6168              * For SMMLS, we need a 64-bit subtract.  Borrow caused by
6169              * a non-zero multiplicand lowpart, and the correct result
6170              * lowpart for rounding.
6171              */
6172             tcg_gen_sub2_i32(t2, t1, tcg_constant_i32(0), t3, t2, t1);
6173         } else {
6174             tcg_gen_add_i32(t1, t1, t3);
6175         }
6176     }
6177     if (round) {
6178         /*
6179          * Adding 0x80000000 to the 64-bit quantity means that we have
6180          * carry in to the high word when the low word has the msb set.
6181          */
6182         tcg_gen_shri_i32(t2, t2, 31);
6183         tcg_gen_add_i32(t1, t1, t2);
6184     }
6185     store_reg(s, a->rd, t1);
6186     return true;
6187 }
6188 
6189 static bool trans_SMMLA(DisasContext *s, arg_rrrr *a)
6190 {
6191     return op_smmla(s, a, false, false);
6192 }
6193 
6194 static bool trans_SMMLAR(DisasContext *s, arg_rrrr *a)
6195 {
6196     return op_smmla(s, a, true, false);
6197 }
6198 
6199 static bool trans_SMMLS(DisasContext *s, arg_rrrr *a)
6200 {
6201     return op_smmla(s, a, false, true);
6202 }
6203 
6204 static bool trans_SMMLSR(DisasContext *s, arg_rrrr *a)
6205 {
6206     return op_smmla(s, a, true, true);
6207 }
6208 
6209 static bool op_div(DisasContext *s, arg_rrr *a, bool u)
6210 {
6211     TCGv_i32 t1, t2;
6212 
6213     if (s->thumb
6214         ? !dc_isar_feature(aa32_thumb_div, s)
6215         : !dc_isar_feature(aa32_arm_div, s)) {
6216         return false;
6217     }
6218 
6219     t1 = load_reg(s, a->rn);
6220     t2 = load_reg(s, a->rm);
6221     if (u) {
6222         gen_helper_udiv(t1, tcg_env, t1, t2);
6223     } else {
6224         gen_helper_sdiv(t1, tcg_env, t1, t2);
6225     }
6226     store_reg(s, a->rd, t1);
6227     return true;
6228 }
6229 
6230 static bool trans_SDIV(DisasContext *s, arg_rrr *a)
6231 {
6232     return op_div(s, a, false);
6233 }
6234 
6235 static bool trans_UDIV(DisasContext *s, arg_rrr *a)
6236 {
6237     return op_div(s, a, true);
6238 }
6239 
6240 /*
6241  * Block data transfer
6242  */
6243 
6244 static TCGv_i32 op_addr_block_pre(DisasContext *s, arg_ldst_block *a, int n)
6245 {
6246     TCGv_i32 addr = load_reg(s, a->rn);
6247 
6248     if (a->b) {
6249         if (a->i) {
6250             /* pre increment */
6251             tcg_gen_addi_i32(addr, addr, 4);
6252         } else {
6253             /* pre decrement */
6254             tcg_gen_addi_i32(addr, addr, -(n * 4));
6255         }
6256     } else if (!a->i && n != 1) {
6257         /* post decrement */
6258         tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
6259     }
6260 
6261     if (s->v8m_stackcheck && a->rn == 13 && a->w) {
6262         /*
6263          * If the writeback is incrementing SP rather than
6264          * decrementing it, and the initial SP is below the
6265          * stack limit but the final written-back SP would
6266          * be above, then we must not perform any memory
6267          * accesses, but it is IMPDEF whether we generate
6268          * an exception. We choose to do so in this case.
6269          * At this point 'addr' is the lowest address, so
6270          * either the original SP (if incrementing) or our
6271          * final SP (if decrementing), so that's what we check.
6272          */
6273         gen_helper_v8m_stackcheck(tcg_env, addr);
6274     }
6275 
6276     return addr;
6277 }
6278 
6279 static void op_addr_block_post(DisasContext *s, arg_ldst_block *a,
6280                                TCGv_i32 addr, int n)
6281 {
6282     if (a->w) {
6283         /* write back */
6284         if (!a->b) {
6285             if (a->i) {
6286                 /* post increment */
6287                 tcg_gen_addi_i32(addr, addr, 4);
6288             } else {
6289                 /* post decrement */
6290                 tcg_gen_addi_i32(addr, addr, -(n * 4));
6291             }
6292         } else if (!a->i && n != 1) {
6293             /* pre decrement */
6294             tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
6295         }
6296         store_reg(s, a->rn, addr);
6297     }
6298 }
6299 
6300 static bool op_stm(DisasContext *s, arg_ldst_block *a)
6301 {
6302     int i, j, n, list, mem_idx;
6303     bool user = a->u;
6304     TCGv_i32 addr, tmp;
6305 
6306     if (user) {
6307         /* STM (user) */
6308         if (IS_USER(s)) {
6309             /* Only usable in supervisor mode.  */
6310             unallocated_encoding(s);
6311             return true;
6312         }
6313     }
6314 
6315     list = a->list;
6316     n = ctpop16(list);
6317     /*
6318      * This is UNPREDICTABLE for n < 1 in all encodings, and we choose
6319      * to UNDEF. In the T32 STM encoding n == 1 is also UNPREDICTABLE,
6320      * but hardware treats it like the A32 version and implements the
6321      * single-register-store, and some in-the-wild (buggy) software
6322      * assumes that, so we don't UNDEF on that case.
6323      */
6324     if (n < 1 || a->rn == 15) {
6325         unallocated_encoding(s);
6326         return true;
6327     }
6328 
6329     s->eci_handled = true;
6330 
6331     addr = op_addr_block_pre(s, a, n);
6332     mem_idx = get_mem_index(s);
6333 
6334     for (i = j = 0; i < 16; i++) {
6335         if (!(list & (1 << i))) {
6336             continue;
6337         }
6338 
6339         if (user && i != 15) {
6340             tmp = tcg_temp_new_i32();
6341             gen_helper_get_user_reg(tmp, tcg_env, tcg_constant_i32(i));
6342         } else {
6343             tmp = load_reg(s, i);
6344         }
6345         gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
6346 
6347         /* No need to add after the last transfer.  */
6348         if (++j != n) {
6349             tcg_gen_addi_i32(addr, addr, 4);
6350         }
6351     }
6352 
6353     op_addr_block_post(s, a, addr, n);
6354     clear_eci_state(s);
6355     return true;
6356 }
6357 
6358 static bool trans_STM(DisasContext *s, arg_ldst_block *a)
6359 {
6360     return op_stm(s, a);
6361 }
6362 
6363 static bool trans_STM_t32(DisasContext *s, arg_ldst_block *a)
6364 {
6365     /* Writeback register in register list is UNPREDICTABLE for T32.  */
6366     if (a->w && (a->list & (1 << a->rn))) {
6367         unallocated_encoding(s);
6368         return true;
6369     }
6370     return op_stm(s, a);
6371 }
6372 
6373 static bool do_ldm(DisasContext *s, arg_ldst_block *a)
6374 {
6375     int i, j, n, list, mem_idx;
6376     bool loaded_base;
6377     bool user = a->u;
6378     bool exc_return = false;
6379     TCGv_i32 addr, tmp, loaded_var;
6380 
6381     if (user) {
6382         /* LDM (user), LDM (exception return) */
6383         if (IS_USER(s)) {
6384             /* Only usable in supervisor mode.  */
6385             unallocated_encoding(s);
6386             return true;
6387         }
6388         if (extract32(a->list, 15, 1)) {
6389             exc_return = true;
6390             user = false;
6391         } else {
6392             /* LDM (user) does not allow writeback.  */
6393             if (a->w) {
6394                 unallocated_encoding(s);
6395                 return true;
6396             }
6397         }
6398     }
6399 
6400     list = a->list;
6401     n = ctpop16(list);
6402     /*
6403      * This is UNPREDICTABLE for n < 1 in all encodings, and we choose
6404      * to UNDEF. In the T32 LDM encoding n == 1 is also UNPREDICTABLE,
6405      * but hardware treats it like the A32 version and implements the
6406      * single-register-load, and some in-the-wild (buggy) software
6407      * assumes that, so we don't UNDEF on that case.
6408      */
6409     if (n < 1 || a->rn == 15) {
6410         unallocated_encoding(s);
6411         return true;
6412     }
6413 
6414     s->eci_handled = true;
6415 
6416     addr = op_addr_block_pre(s, a, n);
6417     mem_idx = get_mem_index(s);
6418     loaded_base = false;
6419     loaded_var = NULL;
6420 
6421     for (i = j = 0; i < 16; i++) {
6422         if (!(list & (1 << i))) {
6423             continue;
6424         }
6425 
6426         tmp = tcg_temp_new_i32();
6427         gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | MO_ALIGN);
6428         if (user) {
6429             gen_helper_set_user_reg(tcg_env, tcg_constant_i32(i), tmp);
6430         } else if (i == a->rn) {
6431             loaded_var = tmp;
6432             loaded_base = true;
6433         } else if (i == 15 && exc_return) {
6434             store_pc_exc_ret(s, tmp);
6435         } else {
6436             store_reg_from_load(s, i, tmp);
6437         }
6438 
6439         /* No need to add after the last transfer.  */
6440         if (++j != n) {
6441             tcg_gen_addi_i32(addr, addr, 4);
6442         }
6443     }
6444 
6445     op_addr_block_post(s, a, addr, n);
6446 
6447     if (loaded_base) {
6448         /* Note that we reject base == pc above.  */
6449         store_reg(s, a->rn, loaded_var);
6450     }
6451 
6452     if (exc_return) {
6453         /* Restore CPSR from SPSR.  */
6454         tmp = load_cpu_field(spsr);
6455         translator_io_start(&s->base);
6456         gen_helper_cpsr_write_eret(tcg_env, tmp);
6457         /* Must exit loop to check un-masked IRQs */
6458         s->base.is_jmp = DISAS_EXIT;
6459     }
6460     clear_eci_state(s);
6461     return true;
6462 }
6463 
6464 static bool trans_LDM_a32(DisasContext *s, arg_ldst_block *a)
6465 {
6466     /*
6467      * Writeback register in register list is UNPREDICTABLE
6468      * for ArchVersion() >= 7.  Prior to v7, A32 would write
6469      * an UNKNOWN value to the base register.
6470      */
6471     if (ENABLE_ARCH_7 && a->w && (a->list & (1 << a->rn))) {
6472         unallocated_encoding(s);
6473         return true;
6474     }
6475     return do_ldm(s, a);
6476 }
6477 
6478 static bool trans_LDM_t32(DisasContext *s, arg_ldst_block *a)
6479 {
6480     /* Writeback register in register list is UNPREDICTABLE for T32. */
6481     if (a->w && (a->list & (1 << a->rn))) {
6482         unallocated_encoding(s);
6483         return true;
6484     }
6485     return do_ldm(s, a);
6486 }
6487 
6488 static bool trans_LDM_t16(DisasContext *s, arg_ldst_block *a)
6489 {
6490     /* Writeback is conditional on the base register not being loaded.  */
6491     a->w = !(a->list & (1 << a->rn));
6492     return do_ldm(s, a);
6493 }
6494 
6495 static bool trans_CLRM(DisasContext *s, arg_CLRM *a)
6496 {
6497     int i;
6498     TCGv_i32 zero;
6499 
6500     if (!dc_isar_feature(aa32_m_sec_state, s)) {
6501         return false;
6502     }
6503 
6504     if (extract32(a->list, 13, 1)) {
6505         return false;
6506     }
6507 
6508     if (!a->list) {
6509         /* UNPREDICTABLE; we choose to UNDEF */
6510         return false;
6511     }
6512 
6513     s->eci_handled = true;
6514 
6515     zero = tcg_constant_i32(0);
6516     for (i = 0; i < 15; i++) {
6517         if (extract32(a->list, i, 1)) {
6518             /* Clear R[i] */
6519             tcg_gen_mov_i32(cpu_R[i], zero);
6520         }
6521     }
6522     if (extract32(a->list, 15, 1)) {
6523         /*
6524          * Clear APSR (by calling the MSR helper with the same argument
6525          * as for "MSR APSR_nzcvqg, Rn": mask = 0b1100, SYSM=0)
6526          */
6527         gen_helper_v7m_msr(tcg_env, tcg_constant_i32(0xc00), zero);
6528     }
6529     clear_eci_state(s);
6530     return true;
6531 }
6532 
6533 /*
6534  * Branch, branch with link
6535  */
6536 
6537 static bool trans_B(DisasContext *s, arg_i *a)
6538 {
6539     gen_jmp(s, jmp_diff(s, a->imm));
6540     return true;
6541 }
6542 
6543 static bool trans_B_cond_thumb(DisasContext *s, arg_ci *a)
6544 {
6545     /* This has cond from encoding, required to be outside IT block.  */
6546     if (a->cond >= 0xe) {
6547         return false;
6548     }
6549     if (s->condexec_mask) {
6550         unallocated_encoding(s);
6551         return true;
6552     }
6553     arm_skip_unless(s, a->cond);
6554     gen_jmp(s, jmp_diff(s, a->imm));
6555     return true;
6556 }
6557 
6558 static bool trans_BL(DisasContext *s, arg_i *a)
6559 {
6560     gen_pc_plus_diff(s, cpu_R[14], curr_insn_len(s) | s->thumb);
6561     gen_jmp(s, jmp_diff(s, a->imm));
6562     return true;
6563 }
6564 
6565 static bool trans_BLX_i(DisasContext *s, arg_BLX_i *a)
6566 {
6567     /*
6568      * BLX <imm> would be useless on M-profile; the encoding space
6569      * is used for other insns from v8.1M onward, and UNDEFs before that.
6570      */
6571     if (arm_dc_feature(s, ARM_FEATURE_M)) {
6572         return false;
6573     }
6574 
6575     /* For A32, ARM_FEATURE_V5 is checked near the start of the uncond block. */
6576     if (s->thumb && (a->imm & 2)) {
6577         return false;
6578     }
6579     gen_pc_plus_diff(s, cpu_R[14], curr_insn_len(s) | s->thumb);
6580     store_cpu_field_constant(!s->thumb, thumb);
6581     /* This jump is computed from an aligned PC: subtract off the low bits. */
6582     gen_jmp(s, jmp_diff(s, a->imm - (s->pc_curr & 3)));
6583     return true;
6584 }
6585 
6586 static bool trans_BL_BLX_prefix(DisasContext *s, arg_BL_BLX_prefix *a)
6587 {
6588     assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
6589     gen_pc_plus_diff(s, cpu_R[14], jmp_diff(s, a->imm << 12));
6590     return true;
6591 }
6592 
6593 static bool trans_BL_suffix(DisasContext *s, arg_BL_suffix *a)
6594 {
6595     TCGv_i32 tmp = tcg_temp_new_i32();
6596 
6597     assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
6598     tcg_gen_addi_i32(tmp, cpu_R[14], (a->imm << 1) | 1);
6599     gen_pc_plus_diff(s, cpu_R[14], curr_insn_len(s) | 1);
6600     gen_bx(s, tmp);
6601     return true;
6602 }
6603 
6604 static bool trans_BLX_suffix(DisasContext *s, arg_BLX_suffix *a)
6605 {
6606     TCGv_i32 tmp;
6607 
6608     assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
6609     if (!ENABLE_ARCH_5) {
6610         return false;
6611     }
6612     tmp = tcg_temp_new_i32();
6613     tcg_gen_addi_i32(tmp, cpu_R[14], a->imm << 1);
6614     tcg_gen_andi_i32(tmp, tmp, 0xfffffffc);
6615     gen_pc_plus_diff(s, cpu_R[14], curr_insn_len(s) | 1);
6616     gen_bx(s, tmp);
6617     return true;
6618 }
6619 
6620 static bool trans_BF(DisasContext *s, arg_BF *a)
6621 {
6622     /*
6623      * M-profile branch future insns. The architecture permits an
6624      * implementation to implement these as NOPs (equivalent to
6625      * discarding the LO_BRANCH_INFO cache immediately), and we
6626      * take that IMPDEF option because for QEMU a "real" implementation
6627      * would be complicated and wouldn't execute any faster.
6628      */
6629     if (!dc_isar_feature(aa32_lob, s)) {
6630         return false;
6631     }
6632     if (a->boff == 0) {
6633         /* SEE "Related encodings" (loop insns) */
6634         return false;
6635     }
6636     /* Handle as NOP */
6637     return true;
6638 }
6639 
6640 static bool trans_DLS(DisasContext *s, arg_DLS *a)
6641 {
6642     /* M-profile low-overhead loop start */
6643     TCGv_i32 tmp;
6644 
6645     if (!dc_isar_feature(aa32_lob, s)) {
6646         return false;
6647     }
6648     if (a->rn == 13 || a->rn == 15) {
6649         /*
6650          * For DLSTP rn == 15 is a related encoding (LCTP); the
6651          * other cases caught by this condition are all
6652          * CONSTRAINED UNPREDICTABLE: we choose to UNDEF
6653          */
6654         return false;
6655     }
6656 
6657     if (a->size != 4) {
6658         /* DLSTP */
6659         if (!dc_isar_feature(aa32_mve, s)) {
6660             return false;
6661         }
6662         if (!vfp_access_check(s)) {
6663             return true;
6664         }
6665     }
6666 
6667     /* Not a while loop: set LR to the count, and set LTPSIZE for DLSTP */
6668     tmp = load_reg(s, a->rn);
6669     store_reg(s, 14, tmp);
6670     if (a->size != 4) {
6671         /* DLSTP: set FPSCR.LTPSIZE */
6672         store_cpu_field(tcg_constant_i32(a->size), v7m.ltpsize);
6673         s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
6674     }
6675     return true;
6676 }
6677 
6678 static bool trans_WLS(DisasContext *s, arg_WLS *a)
6679 {
6680     /* M-profile low-overhead while-loop start */
6681     TCGv_i32 tmp;
6682     DisasLabel nextlabel;
6683 
6684     if (!dc_isar_feature(aa32_lob, s)) {
6685         return false;
6686     }
6687     if (a->rn == 13 || a->rn == 15) {
6688         /*
6689          * For WLSTP rn == 15 is a related encoding (LE); the
6690          * other cases caught by this condition are all
6691          * CONSTRAINED UNPREDICTABLE: we choose to UNDEF
6692          */
6693         return false;
6694     }
6695     if (s->condexec_mask) {
6696         /*
6697          * WLS in an IT block is CONSTRAINED UNPREDICTABLE;
6698          * we choose to UNDEF, because otherwise our use of
6699          * gen_goto_tb(1) would clash with the use of TB exit 1
6700          * in the dc->condjmp condition-failed codepath in
6701          * arm_tr_tb_stop() and we'd get an assertion.
6702          */
6703         return false;
6704     }
6705     if (a->size != 4) {
6706         /* WLSTP */
6707         if (!dc_isar_feature(aa32_mve, s)) {
6708             return false;
6709         }
6710         /*
6711          * We need to check that the FPU is enabled here, but mustn't
6712          * call vfp_access_check() to do that because we don't want to
6713          * do the lazy state preservation in the "loop count is zero" case.
6714          * Do the check-and-raise-exception by hand.
6715          */
6716         if (s->fp_excp_el) {
6717             gen_exception_insn_el(s, 0, EXCP_NOCP,
6718                                   syn_uncategorized(), s->fp_excp_el);
6719             return true;
6720         }
6721     }
6722 
6723     nextlabel = gen_disas_label(s);
6724     tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_R[a->rn], 0, nextlabel.label);
6725     tmp = load_reg(s, a->rn);
6726     store_reg(s, 14, tmp);
6727     if (a->size != 4) {
6728         /*
6729          * WLSTP: set FPSCR.LTPSIZE. This requires that we do the
6730          * lazy state preservation, new FP context creation, etc,
6731          * that vfp_access_check() does. We know that the actual
6732          * access check will succeed (ie it won't generate code that
6733          * throws an exception) because we did that check by hand earlier.
6734          */
6735         bool ok = vfp_access_check(s);
6736         assert(ok);
6737         store_cpu_field(tcg_constant_i32(a->size), v7m.ltpsize);
6738         /*
6739          * LTPSIZE updated, but MVE_NO_PRED will always be the same thing (0)
6740          * when we take this upcoming exit from this TB, so gen_jmp_tb() is OK.
6741          */
6742     }
6743     gen_jmp_tb(s, curr_insn_len(s), 1);
6744 
6745     set_disas_label(s, nextlabel);
6746     gen_jmp(s, jmp_diff(s, a->imm));
6747     return true;
6748 }
6749 
6750 static bool trans_LE(DisasContext *s, arg_LE *a)
6751 {
6752     /*
6753      * M-profile low-overhead loop end. The architecture permits an
6754      * implementation to discard the LO_BRANCH_INFO cache at any time,
6755      * and we take the IMPDEF option to never set it in the first place
6756      * (equivalent to always discarding it immediately), because for QEMU
6757      * a "real" implementation would be complicated and wouldn't execute
6758      * any faster.
6759      */
6760     TCGv_i32 tmp;
6761     DisasLabel loopend;
6762     bool fpu_active;
6763 
6764     if (!dc_isar_feature(aa32_lob, s)) {
6765         return false;
6766     }
6767     if (a->f && a->tp) {
6768         return false;
6769     }
6770     if (s->condexec_mask) {
6771         /*
6772          * LE in an IT block is CONSTRAINED UNPREDICTABLE;
6773          * we choose to UNDEF, because otherwise our use of
6774          * gen_goto_tb(1) would clash with the use of TB exit 1
6775          * in the dc->condjmp condition-failed codepath in
6776          * arm_tr_tb_stop() and we'd get an assertion.
6777          */
6778         return false;
6779     }
6780     if (a->tp) {
6781         /* LETP */
6782         if (!dc_isar_feature(aa32_mve, s)) {
6783             return false;
6784         }
6785         if (!vfp_access_check(s)) {
6786             s->eci_handled = true;
6787             return true;
6788         }
6789     }
6790 
6791     /* LE/LETP is OK with ECI set and leaves it untouched */
6792     s->eci_handled = true;
6793 
6794     /*
6795      * With MVE, LTPSIZE might not be 4, and we must emit an INVSTATE
6796      * UsageFault exception for the LE insn in that case. Note that we
6797      * are not directly checking FPSCR.LTPSIZE but instead check the
6798      * pseudocode LTPSIZE() function, which returns 4 if the FPU is
6799      * not currently active (ie ActiveFPState() returns false). We
6800      * can identify not-active purely from our TB state flags, as the
6801      * FPU is active only if:
6802      *  the FPU is enabled
6803      *  AND lazy state preservation is not active
6804      *  AND we do not need a new fp context (this is the ASPEN/FPCA check)
6805      *
6806      * Usually we don't need to care about this distinction between
6807      * LTPSIZE and FPSCR.LTPSIZE, because the code in vfp_access_check()
6808      * will either take an exception or clear the conditions that make
6809      * the FPU not active. But LE is an unusual case of a non-FP insn
6810      * that looks at LTPSIZE.
6811      */
6812     fpu_active = !s->fp_excp_el && !s->v7m_lspact && !s->v7m_new_fp_ctxt_needed;
6813 
6814     if (!a->tp && dc_isar_feature(aa32_mve, s) && fpu_active) {
6815         /* Need to do a runtime check for LTPSIZE != 4 */
6816         DisasLabel skipexc = gen_disas_label(s);
6817         tmp = load_cpu_field(v7m.ltpsize);
6818         tcg_gen_brcondi_i32(TCG_COND_EQ, tmp, 4, skipexc.label);
6819         gen_exception_insn(s, 0, EXCP_INVSTATE, syn_uncategorized());
6820         set_disas_label(s, skipexc);
6821     }
6822 
6823     if (a->f) {
6824         /* Loop-forever: just jump back to the loop start */
6825         gen_jmp(s, jmp_diff(s, -a->imm));
6826         return true;
6827     }
6828 
6829     /*
6830      * Not loop-forever. If LR <= loop-decrement-value this is the last loop.
6831      * For LE, we know at this point that LTPSIZE must be 4 and the
6832      * loop decrement value is 1. For LETP we need to calculate the decrement
6833      * value from LTPSIZE.
6834      */
6835     loopend = gen_disas_label(s);
6836     if (!a->tp) {
6837         tcg_gen_brcondi_i32(TCG_COND_LEU, cpu_R[14], 1, loopend.label);
6838         tcg_gen_addi_i32(cpu_R[14], cpu_R[14], -1);
6839     } else {
6840         /*
6841          * Decrement by 1 << (4 - LTPSIZE). We need to use a TCG local
6842          * so that decr stays live after the brcondi.
6843          */
6844         TCGv_i32 decr = tcg_temp_new_i32();
6845         TCGv_i32 ltpsize = load_cpu_field(v7m.ltpsize);
6846         tcg_gen_sub_i32(decr, tcg_constant_i32(4), ltpsize);
6847         tcg_gen_shl_i32(decr, tcg_constant_i32(1), decr);
6848 
6849         tcg_gen_brcond_i32(TCG_COND_LEU, cpu_R[14], decr, loopend.label);
6850 
6851         tcg_gen_sub_i32(cpu_R[14], cpu_R[14], decr);
6852     }
6853     /* Jump back to the loop start */
6854     gen_jmp(s, jmp_diff(s, -a->imm));
6855 
6856     set_disas_label(s, loopend);
6857     if (a->tp) {
6858         /* Exits from tail-pred loops must reset LTPSIZE to 4 */
6859         store_cpu_field(tcg_constant_i32(4), v7m.ltpsize);
6860     }
6861     /* End TB, continuing to following insn */
6862     gen_jmp_tb(s, curr_insn_len(s), 1);
6863     return true;
6864 }
6865 
6866 static bool trans_LCTP(DisasContext *s, arg_LCTP *a)
6867 {
6868     /*
6869      * M-profile Loop Clear with Tail Predication. Since our implementation
6870      * doesn't cache branch information, all we need to do is reset
6871      * FPSCR.LTPSIZE to 4.
6872      */
6873 
6874     if (!dc_isar_feature(aa32_lob, s) ||
6875         !dc_isar_feature(aa32_mve, s)) {
6876         return false;
6877     }
6878 
6879     if (!vfp_access_check(s)) {
6880         return true;
6881     }
6882 
6883     store_cpu_field_constant(4, v7m.ltpsize);
6884     return true;
6885 }
6886 
6887 static bool trans_VCTP(DisasContext *s, arg_VCTP *a)
6888 {
6889     /*
6890      * M-profile Create Vector Tail Predicate. This insn is itself
6891      * predicated and is subject to beatwise execution.
6892      */
6893     TCGv_i32 rn_shifted, masklen;
6894 
6895     if (!dc_isar_feature(aa32_mve, s) || a->rn == 13 || a->rn == 15) {
6896         return false;
6897     }
6898 
6899     if (!mve_eci_check(s) || !vfp_access_check(s)) {
6900         return true;
6901     }
6902 
6903     /*
6904      * We pre-calculate the mask length here to avoid having
6905      * to have multiple helpers specialized for size.
6906      * We pass the helper "rn <= (1 << (4 - size)) ? (rn << size) : 16".
6907      */
6908     rn_shifted = tcg_temp_new_i32();
6909     masklen = load_reg(s, a->rn);
6910     tcg_gen_shli_i32(rn_shifted, masklen, a->size);
6911     tcg_gen_movcond_i32(TCG_COND_LEU, masklen,
6912                         masklen, tcg_constant_i32(1 << (4 - a->size)),
6913                         rn_shifted, tcg_constant_i32(16));
6914     gen_helper_mve_vctp(tcg_env, masklen);
6915     /* This insn updates predication bits */
6916     s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
6917     mve_update_eci(s);
6918     return true;
6919 }
6920 
6921 static bool op_tbranch(DisasContext *s, arg_tbranch *a, bool half)
6922 {
6923     TCGv_i32 addr, tmp;
6924 
6925     tmp = load_reg(s, a->rm);
6926     if (half) {
6927         tcg_gen_add_i32(tmp, tmp, tmp);
6928     }
6929     addr = load_reg(s, a->rn);
6930     tcg_gen_add_i32(addr, addr, tmp);
6931 
6932     gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), half ? MO_UW : MO_UB);
6933 
6934     tcg_gen_add_i32(tmp, tmp, tmp);
6935     gen_pc_plus_diff(s, addr, jmp_diff(s, 0));
6936     tcg_gen_add_i32(tmp, tmp, addr);
6937     store_reg(s, 15, tmp);
6938     return true;
6939 }
6940 
6941 static bool trans_TBB(DisasContext *s, arg_tbranch *a)
6942 {
6943     return op_tbranch(s, a, false);
6944 }
6945 
6946 static bool trans_TBH(DisasContext *s, arg_tbranch *a)
6947 {
6948     return op_tbranch(s, a, true);
6949 }
6950 
6951 static bool trans_CBZ(DisasContext *s, arg_CBZ *a)
6952 {
6953     TCGv_i32 tmp = load_reg(s, a->rn);
6954 
6955     arm_gen_condlabel(s);
6956     tcg_gen_brcondi_i32(a->nz ? TCG_COND_EQ : TCG_COND_NE,
6957                         tmp, 0, s->condlabel.label);
6958     gen_jmp(s, jmp_diff(s, a->imm));
6959     return true;
6960 }
6961 
6962 /*
6963  * Supervisor call - both T32 & A32 come here so we need to check
6964  * which mode we are in when checking for semihosting.
6965  */
6966 
6967 static bool trans_SVC(DisasContext *s, arg_SVC *a)
6968 {
6969     const uint32_t semihost_imm = s->thumb ? 0xab : 0x123456;
6970 
6971     if (!arm_dc_feature(s, ARM_FEATURE_M) &&
6972         semihosting_enabled(s->current_el == 0) &&
6973         (a->imm == semihost_imm)) {
6974         gen_exception_internal_insn(s, EXCP_SEMIHOST);
6975     } else {
6976         if (s->fgt_svc) {
6977             uint32_t syndrome = syn_aa32_svc(a->imm, s->thumb);
6978             gen_exception_insn_el(s, 0, EXCP_UDEF, syndrome, 2);
6979         } else {
6980             gen_update_pc(s, curr_insn_len(s));
6981             s->svc_imm = a->imm;
6982             s->base.is_jmp = DISAS_SWI;
6983         }
6984     }
6985     return true;
6986 }
6987 
6988 /*
6989  * Unconditional system instructions
6990  */
6991 
6992 static bool trans_RFE(DisasContext *s, arg_RFE *a)
6993 {
6994     static const int8_t pre_offset[4] = {
6995         /* DA */ -4, /* IA */ 0, /* DB */ -8, /* IB */ 4
6996     };
6997     static const int8_t post_offset[4] = {
6998         /* DA */ -8, /* IA */ 4, /* DB */ -4, /* IB */ 0
6999     };
7000     TCGv_i32 addr, t1, t2;
7001 
7002     if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
7003         return false;
7004     }
7005     if (IS_USER(s)) {
7006         unallocated_encoding(s);
7007         return true;
7008     }
7009 
7010     addr = load_reg(s, a->rn);
7011     tcg_gen_addi_i32(addr, addr, pre_offset[a->pu]);
7012 
7013     /* Load PC into tmp and CPSR into tmp2.  */
7014     t1 = tcg_temp_new_i32();
7015     gen_aa32_ld_i32(s, t1, addr, get_mem_index(s), MO_UL | MO_ALIGN);
7016     tcg_gen_addi_i32(addr, addr, 4);
7017     t2 = tcg_temp_new_i32();
7018     gen_aa32_ld_i32(s, t2, addr, get_mem_index(s), MO_UL | MO_ALIGN);
7019 
7020     if (a->w) {
7021         /* Base writeback.  */
7022         tcg_gen_addi_i32(addr, addr, post_offset[a->pu]);
7023         store_reg(s, a->rn, addr);
7024     }
7025     gen_rfe(s, t1, t2);
7026     return true;
7027 }
7028 
7029 static bool trans_SRS(DisasContext *s, arg_SRS *a)
7030 {
7031     if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
7032         return false;
7033     }
7034     gen_srs(s, a->mode, a->pu, a->w);
7035     return true;
7036 }
7037 
7038 static bool trans_CPS(DisasContext *s, arg_CPS *a)
7039 {
7040     uint32_t mask, val;
7041 
7042     if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
7043         return false;
7044     }
7045     if (IS_USER(s)) {
7046         /* Implemented as NOP in user mode.  */
7047         return true;
7048     }
7049     /* TODO: There are quite a lot of UNPREDICTABLE argument combinations. */
7050 
7051     mask = val = 0;
7052     if (a->imod & 2) {
7053         if (a->A) {
7054             mask |= CPSR_A;
7055         }
7056         if (a->I) {
7057             mask |= CPSR_I;
7058         }
7059         if (a->F) {
7060             mask |= CPSR_F;
7061         }
7062         if (a->imod & 1) {
7063             val |= mask;
7064         }
7065     }
7066     if (a->M) {
7067         mask |= CPSR_M;
7068         val |= a->mode;
7069     }
7070     if (mask) {
7071         gen_set_psr_im(s, mask, 0, val);
7072     }
7073     return true;
7074 }
7075 
7076 static bool trans_CPS_v7m(DisasContext *s, arg_CPS_v7m *a)
7077 {
7078     TCGv_i32 tmp, addr;
7079 
7080     if (!arm_dc_feature(s, ARM_FEATURE_M)) {
7081         return false;
7082     }
7083     if (IS_USER(s)) {
7084         /* Implemented as NOP in user mode.  */
7085         return true;
7086     }
7087 
7088     tmp = tcg_constant_i32(a->im);
7089     /* FAULTMASK */
7090     if (a->F) {
7091         addr = tcg_constant_i32(19);
7092         gen_helper_v7m_msr(tcg_env, addr, tmp);
7093     }
7094     /* PRIMASK */
7095     if (a->I) {
7096         addr = tcg_constant_i32(16);
7097         gen_helper_v7m_msr(tcg_env, addr, tmp);
7098     }
7099     gen_rebuild_hflags(s, false);
7100     gen_lookup_tb(s);
7101     return true;
7102 }
7103 
7104 /*
7105  * Clear-Exclusive, Barriers
7106  */
7107 
7108 static bool trans_CLREX(DisasContext *s, arg_CLREX *a)
7109 {
7110     if (s->thumb
7111         ? !ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)
7112         : !ENABLE_ARCH_6K) {
7113         return false;
7114     }
7115     gen_clrex(s);
7116     return true;
7117 }
7118 
7119 static bool trans_DSB(DisasContext *s, arg_DSB *a)
7120 {
7121     if (!ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)) {
7122         return false;
7123     }
7124     tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
7125     return true;
7126 }
7127 
7128 static bool trans_DMB(DisasContext *s, arg_DMB *a)
7129 {
7130     return trans_DSB(s, NULL);
7131 }
7132 
7133 static bool trans_ISB(DisasContext *s, arg_ISB *a)
7134 {
7135     if (!ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)) {
7136         return false;
7137     }
7138     /*
7139      * We need to break the TB after this insn to execute
7140      * self-modifying code correctly and also to take
7141      * any pending interrupts immediately.
7142      */
7143     s->base.is_jmp = DISAS_TOO_MANY;
7144     return true;
7145 }
7146 
7147 static bool trans_SB(DisasContext *s, arg_SB *a)
7148 {
7149     if (!dc_isar_feature(aa32_sb, s)) {
7150         return false;
7151     }
7152     /*
7153      * TODO: There is no speculation barrier opcode
7154      * for TCG; MB and end the TB instead.
7155      */
7156     tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
7157     s->base.is_jmp = DISAS_TOO_MANY;
7158     return true;
7159 }
7160 
7161 static bool trans_SETEND(DisasContext *s, arg_SETEND *a)
7162 {
7163     if (!ENABLE_ARCH_6) {
7164         return false;
7165     }
7166     if (a->E != (s->be_data == MO_BE)) {
7167         gen_helper_setend(tcg_env);
7168         s->base.is_jmp = DISAS_UPDATE_EXIT;
7169     }
7170     return true;
7171 }
7172 
7173 /*
7174  * Preload instructions
7175  * All are nops, contingent on the appropriate arch level.
7176  */
7177 
7178 static bool trans_PLD(DisasContext *s, arg_PLD *a)
7179 {
7180     return ENABLE_ARCH_5TE;
7181 }
7182 
7183 static bool trans_PLDW(DisasContext *s, arg_PLDW *a)
7184 {
7185     return arm_dc_feature(s, ARM_FEATURE_V7MP);
7186 }
7187 
7188 static bool trans_PLI(DisasContext *s, arg_PLI *a)
7189 {
7190     return ENABLE_ARCH_7;
7191 }
7192 
7193 /*
7194  * If-then
7195  */
7196 
7197 static bool trans_IT(DisasContext *s, arg_IT *a)
7198 {
7199     int cond_mask = a->cond_mask;
7200 
7201     /*
7202      * No actual code generated for this insn, just setup state.
7203      *
7204      * Combinations of firstcond and mask which set up an 0b1111
7205      * condition are UNPREDICTABLE; we take the CONSTRAINED
7206      * UNPREDICTABLE choice to treat 0b1111 the same as 0b1110,
7207      * i.e. both meaning "execute always".
7208      */
7209     s->condexec_cond = (cond_mask >> 4) & 0xe;
7210     s->condexec_mask = cond_mask & 0x1f;
7211     return true;
7212 }
7213 
7214 /* v8.1M CSEL/CSINC/CSNEG/CSINV */
7215 static bool trans_CSEL(DisasContext *s, arg_CSEL *a)
7216 {
7217     TCGv_i32 rn, rm;
7218     DisasCompare c;
7219 
7220     if (!arm_dc_feature(s, ARM_FEATURE_V8_1M)) {
7221         return false;
7222     }
7223 
7224     if (a->rm == 13) {
7225         /* SEE "Related encodings" (MVE shifts) */
7226         return false;
7227     }
7228 
7229     if (a->rd == 13 || a->rd == 15 || a->rn == 13 || a->fcond >= 14) {
7230         /* CONSTRAINED UNPREDICTABLE: we choose to UNDEF */
7231         return false;
7232     }
7233 
7234     /* In this insn input reg fields of 0b1111 mean "zero", not "PC" */
7235     rn = tcg_temp_new_i32();
7236     rm = tcg_temp_new_i32();
7237     if (a->rn == 15) {
7238         tcg_gen_movi_i32(rn, 0);
7239     } else {
7240         load_reg_var(s, rn, a->rn);
7241     }
7242     if (a->rm == 15) {
7243         tcg_gen_movi_i32(rm, 0);
7244     } else {
7245         load_reg_var(s, rm, a->rm);
7246     }
7247 
7248     switch (a->op) {
7249     case 0: /* CSEL */
7250         break;
7251     case 1: /* CSINC */
7252         tcg_gen_addi_i32(rm, rm, 1);
7253         break;
7254     case 2: /* CSINV */
7255         tcg_gen_not_i32(rm, rm);
7256         break;
7257     case 3: /* CSNEG */
7258         tcg_gen_neg_i32(rm, rm);
7259         break;
7260     default:
7261         g_assert_not_reached();
7262     }
7263 
7264     arm_test_cc(&c, a->fcond);
7265     tcg_gen_movcond_i32(c.cond, rn, c.value, tcg_constant_i32(0), rn, rm);
7266 
7267     store_reg(s, a->rd, rn);
7268     return true;
7269 }
7270 
7271 /*
7272  * Legacy decoder.
7273  */
7274 
7275 static void disas_arm_insn(DisasContext *s, unsigned int insn)
7276 {
7277     unsigned int cond = insn >> 28;
7278 
7279     /* M variants do not implement ARM mode; this must raise the INVSTATE
7280      * UsageFault exception.
7281      */
7282     if (arm_dc_feature(s, ARM_FEATURE_M)) {
7283         gen_exception_insn(s, 0, EXCP_INVSTATE, syn_uncategorized());
7284         return;
7285     }
7286 
7287     if (s->pstate_il) {
7288         /*
7289          * Illegal execution state. This has priority over BTI
7290          * exceptions, but comes after instruction abort exceptions.
7291          */
7292         gen_exception_insn(s, 0, EXCP_UDEF, syn_illegalstate());
7293         return;
7294     }
7295 
7296     if (cond == 0xf) {
7297         /* In ARMv3 and v4 the NV condition is UNPREDICTABLE; we
7298          * choose to UNDEF. In ARMv5 and above the space is used
7299          * for miscellaneous unconditional instructions.
7300          */
7301         if (!arm_dc_feature(s, ARM_FEATURE_V5)) {
7302             unallocated_encoding(s);
7303             return;
7304         }
7305 
7306         /* Unconditional instructions.  */
7307         /* TODO: Perhaps merge these into one decodetree output file.  */
7308         if (disas_a32_uncond(s, insn) ||
7309             disas_vfp_uncond(s, insn) ||
7310             disas_neon_dp(s, insn) ||
7311             disas_neon_ls(s, insn) ||
7312             disas_neon_shared(s, insn)) {
7313             return;
7314         }
7315         /* fall back to legacy decoder */
7316 
7317         if ((insn & 0x0e000f00) == 0x0c000100) {
7318             if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
7319                 /* iWMMXt register transfer.  */
7320                 if (extract32(s->c15_cpar, 1, 1)) {
7321                     if (!disas_iwmmxt_insn(s, insn)) {
7322                         return;
7323                     }
7324                 }
7325             }
7326         }
7327         goto illegal_op;
7328     }
7329     if (cond != 0xe) {
7330         /* if not always execute, we generate a conditional jump to
7331            next instruction */
7332         arm_skip_unless(s, cond);
7333     }
7334 
7335     /* TODO: Perhaps merge these into one decodetree output file.  */
7336     if (disas_a32(s, insn) ||
7337         disas_vfp(s, insn)) {
7338         return;
7339     }
7340     /* fall back to legacy decoder */
7341     /* TODO: convert xscale/iwmmxt decoder to decodetree ?? */
7342     if (arm_dc_feature(s, ARM_FEATURE_XSCALE)) {
7343         if (((insn & 0x0c000e00) == 0x0c000000)
7344             && ((insn & 0x03000000) != 0x03000000)) {
7345             /* Coprocessor insn, coprocessor 0 or 1 */
7346             disas_xscale_insn(s, insn);
7347             return;
7348         }
7349     }
7350 
7351 illegal_op:
7352     unallocated_encoding(s);
7353 }
7354 
7355 static bool thumb_insn_is_16bit(DisasContext *s, uint32_t pc, uint32_t insn)
7356 {
7357     /*
7358      * Return true if this is a 16 bit instruction. We must be precise
7359      * about this (matching the decode).
7360      */
7361     if ((insn >> 11) < 0x1d) {
7362         /* Definitely a 16-bit instruction */
7363         return true;
7364     }
7365 
7366     /* Top five bits 0b11101 / 0b11110 / 0b11111 : this is the
7367      * first half of a 32-bit Thumb insn. Thumb-1 cores might
7368      * end up actually treating this as two 16-bit insns, though,
7369      * if it's half of a bl/blx pair that might span a page boundary.
7370      */
7371     if (arm_dc_feature(s, ARM_FEATURE_THUMB2) ||
7372         arm_dc_feature(s, ARM_FEATURE_M)) {
7373         /* Thumb2 cores (including all M profile ones) always treat
7374          * 32-bit insns as 32-bit.
7375          */
7376         return false;
7377     }
7378 
7379     if ((insn >> 11) == 0x1e && pc - s->page_start < TARGET_PAGE_SIZE - 3) {
7380         /* 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix, and the suffix
7381          * is not on the next page; we merge this into a 32-bit
7382          * insn.
7383          */
7384         return false;
7385     }
7386     /* 0b1110_1xxx_xxxx_xxxx : BLX suffix (or UNDEF);
7387      * 0b1111_1xxx_xxxx_xxxx : BL suffix;
7388      * 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix on the end of a page
7389      *  -- handle as single 16 bit insn
7390      */
7391     return true;
7392 }
7393 
7394 /* Translate a 32-bit thumb instruction. */
7395 static void disas_thumb2_insn(DisasContext *s, uint32_t insn)
7396 {
7397     /*
7398      * ARMv6-M supports a limited subset of Thumb2 instructions.
7399      * Other Thumb1 architectures allow only 32-bit
7400      * combined BL/BLX prefix and suffix.
7401      */
7402     if (arm_dc_feature(s, ARM_FEATURE_M) &&
7403         !arm_dc_feature(s, ARM_FEATURE_V7)) {
7404         int i;
7405         bool found = false;
7406         static const uint32_t armv6m_insn[] = {0xf3808000 /* msr */,
7407                                                0xf3b08040 /* dsb */,
7408                                                0xf3b08050 /* dmb */,
7409                                                0xf3b08060 /* isb */,
7410                                                0xf3e08000 /* mrs */,
7411                                                0xf000d000 /* bl */};
7412         static const uint32_t armv6m_mask[] = {0xffe0d000,
7413                                                0xfff0d0f0,
7414                                                0xfff0d0f0,
7415                                                0xfff0d0f0,
7416                                                0xffe0d000,
7417                                                0xf800d000};
7418 
7419         for (i = 0; i < ARRAY_SIZE(armv6m_insn); i++) {
7420             if ((insn & armv6m_mask[i]) == armv6m_insn[i]) {
7421                 found = true;
7422                 break;
7423             }
7424         }
7425         if (!found) {
7426             goto illegal_op;
7427         }
7428     } else if ((insn & 0xf800e800) != 0xf000e800)  {
7429         if (!arm_dc_feature(s, ARM_FEATURE_THUMB2)) {
7430             unallocated_encoding(s);
7431             return;
7432         }
7433     }
7434 
7435     if (arm_dc_feature(s, ARM_FEATURE_M)) {
7436         /*
7437          * NOCP takes precedence over any UNDEF for (almost) the
7438          * entire wide range of coprocessor-space encodings, so check
7439          * for it first before proceeding to actually decode eg VFP
7440          * insns. This decode also handles the few insns which are
7441          * in copro space but do not have NOCP checks (eg VLLDM, VLSTM).
7442          */
7443         if (disas_m_nocp(s, insn)) {
7444             return;
7445         }
7446     }
7447 
7448     if ((insn & 0xef000000) == 0xef000000) {
7449         /*
7450          * T32 encodings 0b111p_1111_qqqq_qqqq_qqqq_qqqq_qqqq_qqqq
7451          * transform into
7452          * A32 encodings 0b1111_001p_qqqq_qqqq_qqqq_qqqq_qqqq_qqqq
7453          */
7454         uint32_t a32_insn = (insn & 0xe2ffffff) |
7455             ((insn & (1 << 28)) >> 4) | (1 << 28);
7456 
7457         if (disas_neon_dp(s, a32_insn)) {
7458             return;
7459         }
7460     }
7461 
7462     if ((insn & 0xff100000) == 0xf9000000) {
7463         /*
7464          * T32 encodings 0b1111_1001_ppp0_qqqq_qqqq_qqqq_qqqq_qqqq
7465          * transform into
7466          * A32 encodings 0b1111_0100_ppp0_qqqq_qqqq_qqqq_qqqq_qqqq
7467          */
7468         uint32_t a32_insn = (insn & 0x00ffffff) | 0xf4000000;
7469 
7470         if (disas_neon_ls(s, a32_insn)) {
7471             return;
7472         }
7473     }
7474 
7475     /*
7476      * TODO: Perhaps merge these into one decodetree output file.
7477      * Note disas_vfp is written for a32 with cond field in the
7478      * top nibble.  The t32 encoding requires 0xe in the top nibble.
7479      */
7480     if (disas_t32(s, insn) ||
7481         disas_vfp_uncond(s, insn) ||
7482         disas_neon_shared(s, insn) ||
7483         disas_mve(s, insn) ||
7484         ((insn >> 28) == 0xe && disas_vfp(s, insn))) {
7485         return;
7486     }
7487 
7488 illegal_op:
7489     unallocated_encoding(s);
7490 }
7491 
7492 static void disas_thumb_insn(DisasContext *s, uint32_t insn)
7493 {
7494     if (!disas_t16(s, insn)) {
7495         unallocated_encoding(s);
7496     }
7497 }
7498 
7499 static bool insn_crosses_page(CPUARMState *env, DisasContext *s)
7500 {
7501     /* Return true if the insn at dc->base.pc_next might cross a page boundary.
7502      * (False positives are OK, false negatives are not.)
7503      * We know this is a Thumb insn, and our caller ensures we are
7504      * only called if dc->base.pc_next is less than 4 bytes from the page
7505      * boundary, so we cross the page if the first 16 bits indicate
7506      * that this is a 32 bit insn.
7507      */
7508     uint16_t insn = arm_lduw_code(env, &s->base, s->base.pc_next, s->sctlr_b);
7509 
7510     return !thumb_insn_is_16bit(s, s->base.pc_next, insn);
7511 }
7512 
7513 static void arm_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cs)
7514 {
7515     DisasContext *dc = container_of(dcbase, DisasContext, base);
7516     CPUARMState *env = cpu_env(cs);
7517     ARMCPU *cpu = env_archcpu(env);
7518     CPUARMTBFlags tb_flags = arm_tbflags_from_tb(dc->base.tb);
7519     uint32_t condexec, core_mmu_idx;
7520 
7521     dc->isar = &cpu->isar;
7522     dc->condjmp = 0;
7523     dc->pc_save = dc->base.pc_first;
7524     dc->aarch64 = false;
7525     dc->thumb = EX_TBFLAG_AM32(tb_flags, THUMB);
7526     dc->be_data = EX_TBFLAG_ANY(tb_flags, BE_DATA) ? MO_BE : MO_LE;
7527     condexec = EX_TBFLAG_AM32(tb_flags, CONDEXEC);
7528     /*
7529      * the CONDEXEC TB flags are CPSR bits [15:10][26:25]. On A-profile this
7530      * is always the IT bits. On M-profile, some of the reserved encodings
7531      * of IT are used instead to indicate either ICI or ECI, which
7532      * indicate partial progress of a restartable insn that was interrupted
7533      * partway through by an exception:
7534      *  * if CONDEXEC[3:0] != 0b0000 : CONDEXEC is IT bits
7535      *  * if CONDEXEC[3:0] == 0b0000 : CONDEXEC is ICI or ECI bits
7536      * In all cases CONDEXEC == 0 means "not in IT block or restartable
7537      * insn, behave normally".
7538      */
7539     dc->eci = dc->condexec_mask = dc->condexec_cond = 0;
7540     dc->eci_handled = false;
7541     if (condexec & 0xf) {
7542         dc->condexec_mask = (condexec & 0xf) << 1;
7543         dc->condexec_cond = condexec >> 4;
7544     } else {
7545         if (arm_feature(env, ARM_FEATURE_M)) {
7546             dc->eci = condexec >> 4;
7547         }
7548     }
7549 
7550     core_mmu_idx = EX_TBFLAG_ANY(tb_flags, MMUIDX);
7551     dc->mmu_idx = core_to_arm_mmu_idx(env, core_mmu_idx);
7552     dc->current_el = arm_mmu_idx_to_el(dc->mmu_idx);
7553 #if !defined(CONFIG_USER_ONLY)
7554     dc->user = (dc->current_el == 0);
7555 #endif
7556     dc->fp_excp_el = EX_TBFLAG_ANY(tb_flags, FPEXC_EL);
7557     dc->align_mem = EX_TBFLAG_ANY(tb_flags, ALIGN_MEM);
7558     dc->pstate_il = EX_TBFLAG_ANY(tb_flags, PSTATE__IL);
7559     dc->fgt_active = EX_TBFLAG_ANY(tb_flags, FGT_ACTIVE);
7560     dc->fgt_svc = EX_TBFLAG_ANY(tb_flags, FGT_SVC);
7561 
7562     if (arm_feature(env, ARM_FEATURE_M)) {
7563         dc->vfp_enabled = 1;
7564         dc->be_data = MO_TE;
7565         dc->v7m_handler_mode = EX_TBFLAG_M32(tb_flags, HANDLER);
7566         dc->v8m_secure = EX_TBFLAG_M32(tb_flags, SECURE);
7567         dc->v8m_stackcheck = EX_TBFLAG_M32(tb_flags, STACKCHECK);
7568         dc->v8m_fpccr_s_wrong = EX_TBFLAG_M32(tb_flags, FPCCR_S_WRONG);
7569         dc->v7m_new_fp_ctxt_needed =
7570             EX_TBFLAG_M32(tb_flags, NEW_FP_CTXT_NEEDED);
7571         dc->v7m_lspact = EX_TBFLAG_M32(tb_flags, LSPACT);
7572         dc->mve_no_pred = EX_TBFLAG_M32(tb_flags, MVE_NO_PRED);
7573     } else {
7574         dc->sctlr_b = EX_TBFLAG_A32(tb_flags, SCTLR__B);
7575         dc->hstr_active = EX_TBFLAG_A32(tb_flags, HSTR_ACTIVE);
7576         dc->ns = EX_TBFLAG_A32(tb_flags, NS);
7577         dc->vfp_enabled = EX_TBFLAG_A32(tb_flags, VFPEN);
7578         if (arm_feature(env, ARM_FEATURE_XSCALE)) {
7579             dc->c15_cpar = EX_TBFLAG_A32(tb_flags, XSCALE_CPAR);
7580         } else {
7581             dc->vec_len = EX_TBFLAG_A32(tb_flags, VECLEN);
7582             dc->vec_stride = EX_TBFLAG_A32(tb_flags, VECSTRIDE);
7583         }
7584         dc->sme_trap_nonstreaming =
7585             EX_TBFLAG_A32(tb_flags, SME_TRAP_NONSTREAMING);
7586     }
7587     dc->lse2 = false; /* applies only to aarch64 */
7588     dc->cp_regs = cpu->cp_regs;
7589     dc->features = env->features;
7590 
7591     /* Single step state. The code-generation logic here is:
7592      *  SS_ACTIVE == 0:
7593      *   generate code with no special handling for single-stepping (except
7594      *   that anything that can make us go to SS_ACTIVE == 1 must end the TB;
7595      *   this happens anyway because those changes are all system register or
7596      *   PSTATE writes).
7597      *  SS_ACTIVE == 1, PSTATE.SS == 1: (active-not-pending)
7598      *   emit code for one insn
7599      *   emit code to clear PSTATE.SS
7600      *   emit code to generate software step exception for completed step
7601      *   end TB (as usual for having generated an exception)
7602      *  SS_ACTIVE == 1, PSTATE.SS == 0: (active-pending)
7603      *   emit code to generate a software step exception
7604      *   end the TB
7605      */
7606     dc->ss_active = EX_TBFLAG_ANY(tb_flags, SS_ACTIVE);
7607     dc->pstate_ss = EX_TBFLAG_ANY(tb_flags, PSTATE__SS);
7608     dc->is_ldex = false;
7609 
7610     dc->page_start = dc->base.pc_first & TARGET_PAGE_MASK;
7611 
7612     /* If architectural single step active, limit to 1.  */
7613     if (dc->ss_active) {
7614         dc->base.max_insns = 1;
7615     }
7616 
7617     /* ARM is a fixed-length ISA.  Bound the number of insns to execute
7618        to those left on the page.  */
7619     if (!dc->thumb) {
7620         int bound = -(dc->base.pc_first | TARGET_PAGE_MASK) / 4;
7621         dc->base.max_insns = MIN(dc->base.max_insns, bound);
7622     }
7623 
7624     cpu_V0 = tcg_temp_new_i64();
7625     cpu_V1 = tcg_temp_new_i64();
7626     cpu_M0 = tcg_temp_new_i64();
7627 }
7628 
7629 static void arm_tr_tb_start(DisasContextBase *dcbase, CPUState *cpu)
7630 {
7631     DisasContext *dc = container_of(dcbase, DisasContext, base);
7632 
7633     /* A note on handling of the condexec (IT) bits:
7634      *
7635      * We want to avoid the overhead of having to write the updated condexec
7636      * bits back to the CPUARMState for every instruction in an IT block. So:
7637      * (1) if the condexec bits are not already zero then we write
7638      * zero back into the CPUARMState now. This avoids complications trying
7639      * to do it at the end of the block. (For example if we don't do this
7640      * it's hard to identify whether we can safely skip writing condexec
7641      * at the end of the TB, which we definitely want to do for the case
7642      * where a TB doesn't do anything with the IT state at all.)
7643      * (2) if we are going to leave the TB then we call gen_set_condexec()
7644      * which will write the correct value into CPUARMState if zero is wrong.
7645      * This is done both for leaving the TB at the end, and for leaving
7646      * it because of an exception we know will happen, which is done in
7647      * gen_exception_insn(). The latter is necessary because we need to
7648      * leave the TB with the PC/IT state just prior to execution of the
7649      * instruction which caused the exception.
7650      * (3) if we leave the TB unexpectedly (eg a data abort on a load)
7651      * then the CPUARMState will be wrong and we need to reset it.
7652      * This is handled in the same way as restoration of the
7653      * PC in these situations; we save the value of the condexec bits
7654      * for each PC via tcg_gen_insn_start(), and restore_state_to_opc()
7655      * then uses this to restore them after an exception.
7656      *
7657      * Note that there are no instructions which can read the condexec
7658      * bits, and none which can write non-static values to them, so
7659      * we don't need to care about whether CPUARMState is correct in the
7660      * middle of a TB.
7661      */
7662 
7663     /* Reset the conditional execution bits immediately. This avoids
7664        complications trying to do it at the end of the block.  */
7665     if (dc->condexec_mask || dc->condexec_cond) {
7666         store_cpu_field_constant(0, condexec_bits);
7667     }
7668 }
7669 
7670 static void arm_tr_insn_start(DisasContextBase *dcbase, CPUState *cpu)
7671 {
7672     DisasContext *dc = container_of(dcbase, DisasContext, base);
7673     /*
7674      * The ECI/ICI bits share PSR bits with the IT bits, so we
7675      * need to reconstitute the bits from the split-out DisasContext
7676      * fields here.
7677      */
7678     uint32_t condexec_bits;
7679     target_ulong pc_arg = dc->base.pc_next;
7680 
7681     if (tb_cflags(dcbase->tb) & CF_PCREL) {
7682         pc_arg &= ~TARGET_PAGE_MASK;
7683     }
7684     if (dc->eci) {
7685         condexec_bits = dc->eci << 4;
7686     } else {
7687         condexec_bits = (dc->condexec_cond << 4) | (dc->condexec_mask >> 1);
7688     }
7689     tcg_gen_insn_start(pc_arg, condexec_bits, 0);
7690     dc->insn_start_updated = false;
7691 }
7692 
7693 static bool arm_check_kernelpage(DisasContext *dc)
7694 {
7695 #ifdef CONFIG_USER_ONLY
7696     /* Intercept jump to the magic kernel page.  */
7697     if (dc->base.pc_next >= 0xffff0000) {
7698         /* We always get here via a jump, so know we are not in a
7699            conditional execution block.  */
7700         gen_exception_internal(EXCP_KERNEL_TRAP);
7701         dc->base.is_jmp = DISAS_NORETURN;
7702         return true;
7703     }
7704 #endif
7705     return false;
7706 }
7707 
7708 static bool arm_check_ss_active(DisasContext *dc)
7709 {
7710     if (dc->ss_active && !dc->pstate_ss) {
7711         /* Singlestep state is Active-pending.
7712          * If we're in this state at the start of a TB then either
7713          *  a) we just took an exception to an EL which is being debugged
7714          *     and this is the first insn in the exception handler
7715          *  b) debug exceptions were masked and we just unmasked them
7716          *     without changing EL (eg by clearing PSTATE.D)
7717          * In either case we're going to take a swstep exception in the
7718          * "did not step an insn" case, and so the syndrome ISV and EX
7719          * bits should be zero.
7720          */
7721         assert(dc->base.num_insns == 1);
7722         gen_swstep_exception(dc, 0, 0);
7723         dc->base.is_jmp = DISAS_NORETURN;
7724         return true;
7725     }
7726 
7727     return false;
7728 }
7729 
7730 static void arm_post_translate_insn(DisasContext *dc)
7731 {
7732     if (dc->condjmp && dc->base.is_jmp == DISAS_NEXT) {
7733         if (dc->pc_save != dc->condlabel.pc_save) {
7734             gen_update_pc(dc, dc->condlabel.pc_save - dc->pc_save);
7735         }
7736         gen_set_label(dc->condlabel.label);
7737         dc->condjmp = 0;
7738     }
7739 }
7740 
7741 static void arm_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
7742 {
7743     DisasContext *dc = container_of(dcbase, DisasContext, base);
7744     CPUARMState *env = cpu_env(cpu);
7745     uint32_t pc = dc->base.pc_next;
7746     unsigned int insn;
7747 
7748     /* Singlestep exceptions have the highest priority. */
7749     if (arm_check_ss_active(dc)) {
7750         dc->base.pc_next = pc + 4;
7751         return;
7752     }
7753 
7754     if (pc & 3) {
7755         /*
7756          * PC alignment fault.  This has priority over the instruction abort
7757          * that we would receive from a translation fault via arm_ldl_code
7758          * (or the execution of the kernelpage entrypoint). This should only
7759          * be possible after an indirect branch, at the start of the TB.
7760          */
7761         assert(dc->base.num_insns == 1);
7762         gen_helper_exception_pc_alignment(tcg_env, tcg_constant_tl(pc));
7763         dc->base.is_jmp = DISAS_NORETURN;
7764         dc->base.pc_next = QEMU_ALIGN_UP(pc, 4);
7765         return;
7766     }
7767 
7768     if (arm_check_kernelpage(dc)) {
7769         dc->base.pc_next = pc + 4;
7770         return;
7771     }
7772 
7773     dc->pc_curr = pc;
7774     insn = arm_ldl_code(env, &dc->base, pc, dc->sctlr_b);
7775     dc->insn = insn;
7776     dc->base.pc_next = pc + 4;
7777     disas_arm_insn(dc, insn);
7778 
7779     arm_post_translate_insn(dc);
7780 
7781     /* ARM is a fixed-length ISA.  We performed the cross-page check
7782        in init_disas_context by adjusting max_insns.  */
7783 }
7784 
7785 static bool thumb_insn_is_unconditional(DisasContext *s, uint32_t insn)
7786 {
7787     /* Return true if this Thumb insn is always unconditional,
7788      * even inside an IT block. This is true of only a very few
7789      * instructions: BKPT, HLT, and SG.
7790      *
7791      * A larger class of instructions are UNPREDICTABLE if used
7792      * inside an IT block; we do not need to detect those here, because
7793      * what we do by default (perform the cc check and update the IT
7794      * bits state machine) is a permitted CONSTRAINED UNPREDICTABLE
7795      * choice for those situations.
7796      *
7797      * insn is either a 16-bit or a 32-bit instruction; the two are
7798      * distinguishable because for the 16-bit case the top 16 bits
7799      * are zeroes, and that isn't a valid 32-bit encoding.
7800      */
7801     if ((insn & 0xffffff00) == 0xbe00) {
7802         /* BKPT */
7803         return true;
7804     }
7805 
7806     if ((insn & 0xffffffc0) == 0xba80 && arm_dc_feature(s, ARM_FEATURE_V8) &&
7807         !arm_dc_feature(s, ARM_FEATURE_M)) {
7808         /* HLT: v8A only. This is unconditional even when it is going to
7809          * UNDEF; see the v8A ARM ARM DDI0487B.a H3.3.
7810          * For v7 cores this was a plain old undefined encoding and so
7811          * honours its cc check. (We might be using the encoding as
7812          * a semihosting trap, but we don't change the cc check behaviour
7813          * on that account, because a debugger connected to a real v7A
7814          * core and emulating semihosting traps by catching the UNDEF
7815          * exception would also only see cases where the cc check passed.
7816          * No guest code should be trying to do a HLT semihosting trap
7817          * in an IT block anyway.
7818          */
7819         return true;
7820     }
7821 
7822     if (insn == 0xe97fe97f && arm_dc_feature(s, ARM_FEATURE_V8) &&
7823         arm_dc_feature(s, ARM_FEATURE_M)) {
7824         /* SG: v8M only */
7825         return true;
7826     }
7827 
7828     return false;
7829 }
7830 
7831 static void thumb_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
7832 {
7833     DisasContext *dc = container_of(dcbase, DisasContext, base);
7834     CPUARMState *env = cpu_env(cpu);
7835     uint32_t pc = dc->base.pc_next;
7836     uint32_t insn;
7837     bool is_16bit;
7838     /* TCG op to rewind to if this turns out to be an invalid ECI state */
7839     TCGOp *insn_eci_rewind = NULL;
7840     target_ulong insn_eci_pc_save = -1;
7841 
7842     /* Misaligned thumb PC is architecturally impossible. */
7843     assert((dc->base.pc_next & 1) == 0);
7844 
7845     if (arm_check_ss_active(dc) || arm_check_kernelpage(dc)) {
7846         dc->base.pc_next = pc + 2;
7847         return;
7848     }
7849 
7850     dc->pc_curr = pc;
7851     insn = arm_lduw_code(env, &dc->base, pc, dc->sctlr_b);
7852     is_16bit = thumb_insn_is_16bit(dc, dc->base.pc_next, insn);
7853     pc += 2;
7854     if (!is_16bit) {
7855         uint32_t insn2 = arm_lduw_code(env, &dc->base, pc, dc->sctlr_b);
7856         insn = insn << 16 | insn2;
7857         pc += 2;
7858     }
7859     dc->base.pc_next = pc;
7860     dc->insn = insn;
7861 
7862     if (dc->pstate_il) {
7863         /*
7864          * Illegal execution state. This has priority over BTI
7865          * exceptions, but comes after instruction abort exceptions.
7866          */
7867         gen_exception_insn(dc, 0, EXCP_UDEF, syn_illegalstate());
7868         return;
7869     }
7870 
7871     if (dc->eci) {
7872         /*
7873          * For M-profile continuable instructions, ECI/ICI handling
7874          * falls into these cases:
7875          *  - interrupt-continuable instructions
7876          *     These are the various load/store multiple insns (both
7877          *     integer and fp). The ICI bits indicate the register
7878          *     where the load/store can resume. We make the IMPDEF
7879          *     choice to always do "instruction restart", ie ignore
7880          *     the ICI value and always execute the ldm/stm from the
7881          *     start. So all we need to do is zero PSR.ICI if the
7882          *     insn executes.
7883          *  - MVE instructions subject to beat-wise execution
7884          *     Here the ECI bits indicate which beats have already been
7885          *     executed, and we must honour this. Each insn of this
7886          *     type will handle it correctly. We will update PSR.ECI
7887          *     in the helper function for the insn (some ECI values
7888          *     mean that the following insn also has been partially
7889          *     executed).
7890          *  - Special cases which don't advance ECI
7891          *     The insns LE, LETP and BKPT leave the ECI/ICI state
7892          *     bits untouched.
7893          *  - all other insns (the common case)
7894          *     Non-zero ECI/ICI means an INVSTATE UsageFault.
7895          *     We place a rewind-marker here. Insns in the previous
7896          *     three categories will set a flag in the DisasContext.
7897          *     If the flag isn't set after we call disas_thumb_insn()
7898          *     or disas_thumb2_insn() then we know we have a "some other
7899          *     insn" case. We will rewind to the marker (ie throwing away
7900          *     all the generated code) and instead emit "take exception".
7901          */
7902         insn_eci_rewind = tcg_last_op();
7903         insn_eci_pc_save = dc->pc_save;
7904     }
7905 
7906     if (dc->condexec_mask && !thumb_insn_is_unconditional(dc, insn)) {
7907         uint32_t cond = dc->condexec_cond;
7908 
7909         /*
7910          * Conditionally skip the insn. Note that both 0xe and 0xf mean
7911          * "always"; 0xf is not "never".
7912          */
7913         if (cond < 0x0e) {
7914             arm_skip_unless(dc, cond);
7915         }
7916     }
7917 
7918     if (is_16bit) {
7919         disas_thumb_insn(dc, insn);
7920     } else {
7921         disas_thumb2_insn(dc, insn);
7922     }
7923 
7924     /* Advance the Thumb condexec condition.  */
7925     if (dc->condexec_mask) {
7926         dc->condexec_cond = ((dc->condexec_cond & 0xe) |
7927                              ((dc->condexec_mask >> 4) & 1));
7928         dc->condexec_mask = (dc->condexec_mask << 1) & 0x1f;
7929         if (dc->condexec_mask == 0) {
7930             dc->condexec_cond = 0;
7931         }
7932     }
7933 
7934     if (dc->eci && !dc->eci_handled) {
7935         /*
7936          * Insn wasn't valid for ECI/ICI at all: undo what we
7937          * just generated and instead emit an exception
7938          */
7939         tcg_remove_ops_after(insn_eci_rewind);
7940         dc->pc_save = insn_eci_pc_save;
7941         dc->condjmp = 0;
7942         gen_exception_insn(dc, 0, EXCP_INVSTATE, syn_uncategorized());
7943     }
7944 
7945     arm_post_translate_insn(dc);
7946 
7947     /* Thumb is a variable-length ISA.  Stop translation when the next insn
7948      * will touch a new page.  This ensures that prefetch aborts occur at
7949      * the right place.
7950      *
7951      * We want to stop the TB if the next insn starts in a new page,
7952      * or if it spans between this page and the next. This means that
7953      * if we're looking at the last halfword in the page we need to
7954      * see if it's a 16-bit Thumb insn (which will fit in this TB)
7955      * or a 32-bit Thumb insn (which won't).
7956      * This is to avoid generating a silly TB with a single 16-bit insn
7957      * in it at the end of this page (which would execute correctly
7958      * but isn't very efficient).
7959      */
7960     if (dc->base.is_jmp == DISAS_NEXT
7961         && (dc->base.pc_next - dc->page_start >= TARGET_PAGE_SIZE
7962             || (dc->base.pc_next - dc->page_start >= TARGET_PAGE_SIZE - 3
7963                 && insn_crosses_page(env, dc)))) {
7964         dc->base.is_jmp = DISAS_TOO_MANY;
7965     }
7966 }
7967 
7968 static void arm_tr_tb_stop(DisasContextBase *dcbase, CPUState *cpu)
7969 {
7970     DisasContext *dc = container_of(dcbase, DisasContext, base);
7971 
7972     /* At this stage dc->condjmp will only be set when the skipped
7973        instruction was a conditional branch or trap, and the PC has
7974        already been written.  */
7975     gen_set_condexec(dc);
7976     if (dc->base.is_jmp == DISAS_BX_EXCRET) {
7977         /* Exception return branches need some special case code at the
7978          * end of the TB, which is complex enough that it has to
7979          * handle the single-step vs not and the condition-failed
7980          * insn codepath itself.
7981          */
7982         gen_bx_excret_final_code(dc);
7983     } else if (unlikely(dc->ss_active)) {
7984         /* Unconditional and "condition passed" instruction codepath. */
7985         switch (dc->base.is_jmp) {
7986         case DISAS_SWI:
7987             gen_ss_advance(dc);
7988             gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb));
7989             break;
7990         case DISAS_HVC:
7991             gen_ss_advance(dc);
7992             gen_exception_el(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
7993             break;
7994         case DISAS_SMC:
7995             gen_ss_advance(dc);
7996             gen_exception_el(EXCP_SMC, syn_aa32_smc(), 3);
7997             break;
7998         case DISAS_NEXT:
7999         case DISAS_TOO_MANY:
8000         case DISAS_UPDATE_EXIT:
8001         case DISAS_UPDATE_NOCHAIN:
8002             gen_update_pc(dc, curr_insn_len(dc));
8003             /* fall through */
8004         default:
8005             /* FIXME: Single stepping a WFI insn will not halt the CPU. */
8006             gen_singlestep_exception(dc);
8007             break;
8008         case DISAS_NORETURN:
8009             break;
8010         }
8011     } else {
8012         /* While branches must always occur at the end of an IT block,
8013            there are a few other things that can cause us to terminate
8014            the TB in the middle of an IT block:
8015             - Exception generating instructions (bkpt, swi, undefined).
8016             - Page boundaries.
8017             - Hardware watchpoints.
8018            Hardware breakpoints have already been handled and skip this code.
8019          */
8020         switch (dc->base.is_jmp) {
8021         case DISAS_NEXT:
8022         case DISAS_TOO_MANY:
8023             gen_goto_tb(dc, 1, curr_insn_len(dc));
8024             break;
8025         case DISAS_UPDATE_NOCHAIN:
8026             gen_update_pc(dc, curr_insn_len(dc));
8027             /* fall through */
8028         case DISAS_JUMP:
8029             gen_goto_ptr();
8030             break;
8031         case DISAS_UPDATE_EXIT:
8032             gen_update_pc(dc, curr_insn_len(dc));
8033             /* fall through */
8034         default:
8035             /* indicate that the hash table must be used to find the next TB */
8036             tcg_gen_exit_tb(NULL, 0);
8037             break;
8038         case DISAS_NORETURN:
8039             /* nothing more to generate */
8040             break;
8041         case DISAS_WFI:
8042             gen_helper_wfi(tcg_env, tcg_constant_i32(curr_insn_len(dc)));
8043             /*
8044              * The helper doesn't necessarily throw an exception, but we
8045              * must go back to the main loop to check for interrupts anyway.
8046              */
8047             tcg_gen_exit_tb(NULL, 0);
8048             break;
8049         case DISAS_WFE:
8050             gen_helper_wfe(tcg_env);
8051             break;
8052         case DISAS_YIELD:
8053             gen_helper_yield(tcg_env);
8054             break;
8055         case DISAS_SWI:
8056             gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb));
8057             break;
8058         case DISAS_HVC:
8059             gen_exception_el(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
8060             break;
8061         case DISAS_SMC:
8062             gen_exception_el(EXCP_SMC, syn_aa32_smc(), 3);
8063             break;
8064         }
8065     }
8066 
8067     if (dc->condjmp) {
8068         /* "Condition failed" instruction codepath for the branch/trap insn */
8069         set_disas_label(dc, dc->condlabel);
8070         gen_set_condexec(dc);
8071         if (unlikely(dc->ss_active)) {
8072             gen_update_pc(dc, curr_insn_len(dc));
8073             gen_singlestep_exception(dc);
8074         } else {
8075             gen_goto_tb(dc, 1, curr_insn_len(dc));
8076         }
8077     }
8078 }
8079 
8080 static const TranslatorOps arm_translator_ops = {
8081     .init_disas_context = arm_tr_init_disas_context,
8082     .tb_start           = arm_tr_tb_start,
8083     .insn_start         = arm_tr_insn_start,
8084     .translate_insn     = arm_tr_translate_insn,
8085     .tb_stop            = arm_tr_tb_stop,
8086 };
8087 
8088 static const TranslatorOps thumb_translator_ops = {
8089     .init_disas_context = arm_tr_init_disas_context,
8090     .tb_start           = arm_tr_tb_start,
8091     .insn_start         = arm_tr_insn_start,
8092     .translate_insn     = thumb_tr_translate_insn,
8093     .tb_stop            = arm_tr_tb_stop,
8094 };
8095 
8096 /* generate intermediate code for basic block 'tb'.  */
8097 void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int *max_insns,
8098                            vaddr pc, void *host_pc)
8099 {
8100     DisasContext dc = { };
8101     const TranslatorOps *ops = &arm_translator_ops;
8102     CPUARMTBFlags tb_flags = arm_tbflags_from_tb(tb);
8103 
8104     if (EX_TBFLAG_AM32(tb_flags, THUMB)) {
8105         ops = &thumb_translator_ops;
8106     }
8107 #ifdef TARGET_AARCH64
8108     if (EX_TBFLAG_ANY(tb_flags, AARCH64_STATE)) {
8109         ops = &aarch64_translator_ops;
8110     }
8111 #endif
8112 
8113     translator_loop(cpu, tb, max_insns, pc, host_pc, ops, &dc.base);
8114 }
8115