xref: /qemu/target/arm/tcg/helper-a64.c (revision 35aae9d24c060f5de2cfb3511359818a41e383b1)
1 /*
2  *  AArch64 specific helpers
3  *
4  *  Copyright (c) 2013 Alexander Graf <agraf@suse.de>
5  *
6  * This library is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * This library is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "qemu/units.h"
22 #include "cpu.h"
23 #include "gdbstub/helpers.h"
24 #include "exec/helper-proto.h"
25 #include "qemu/host-utils.h"
26 #include "qemu/log.h"
27 #include "qemu/main-loop.h"
28 #include "qemu/bitops.h"
29 #include "internals.h"
30 #include "qemu/crc32c.h"
31 #include "exec/cpu-common.h"
32 #include "exec/exec-all.h"
33 #include "exec/cpu_ldst.h"
34 #include "qemu/int128.h"
35 #include "qemu/atomic128.h"
36 #include "fpu/softfloat.h"
37 #include <zlib.h> /* for crc32 */
38 #ifdef CONFIG_USER_ONLY
39 #include "user/page-protection.h"
40 #endif
41 
42 /* C2.4.7 Multiply and divide */
43 /* special cases for 0 and LLONG_MIN are mandated by the standard */
44 uint64_t HELPER(udiv64)(uint64_t num, uint64_t den)
45 {
46     if (den == 0) {
47         return 0;
48     }
49     return num / den;
50 }
51 
52 int64_t HELPER(sdiv64)(int64_t num, int64_t den)
53 {
54     if (den == 0) {
55         return 0;
56     }
57     if (num == LLONG_MIN && den == -1) {
58         return LLONG_MIN;
59     }
60     return num / den;
61 }
62 
63 uint64_t HELPER(rbit64)(uint64_t x)
64 {
65     return revbit64(x);
66 }
67 
68 void HELPER(msr_i_spsel)(CPUARMState *env, uint32_t imm)
69 {
70     update_spsel(env, imm);
71 }
72 
73 void HELPER(msr_set_allint_el1)(CPUARMState *env)
74 {
75     /* ALLINT update to PSTATE. */
76     if (arm_hcrx_el2_eff(env) & HCRX_TALLINT) {
77         raise_exception_ra(env, EXCP_UDEF,
78                            syn_aa64_sysregtrap(0, 1, 0, 4, 1, 0x1f, 0), 2,
79                            GETPC());
80     }
81 
82     env->pstate |= PSTATE_ALLINT;
83 }
84 
85 static void daif_check(CPUARMState *env, uint32_t op,
86                        uint32_t imm, uintptr_t ra)
87 {
88     /* DAIF update to PSTATE. This is OK from EL0 only if UMA is set.  */
89     if (arm_current_el(env) == 0 && !(arm_sctlr(env, 0) & SCTLR_UMA)) {
90         raise_exception_ra(env, EXCP_UDEF,
91                            syn_aa64_sysregtrap(0, extract32(op, 0, 3),
92                                                extract32(op, 3, 3), 4,
93                                                imm, 0x1f, 0),
94                            exception_target_el(env), ra);
95     }
96 }
97 
98 void HELPER(msr_i_daifset)(CPUARMState *env, uint32_t imm)
99 {
100     daif_check(env, 0x1e, imm, GETPC());
101     env->daif |= (imm << 6) & PSTATE_DAIF;
102     arm_rebuild_hflags(env);
103 }
104 
105 void HELPER(msr_i_daifclear)(CPUARMState *env, uint32_t imm)
106 {
107     daif_check(env, 0x1f, imm, GETPC());
108     env->daif &= ~((imm << 6) & PSTATE_DAIF);
109     arm_rebuild_hflags(env);
110 }
111 
112 /* Convert a softfloat float_relation_ (as returned by
113  * the float*_compare functions) to the correct ARM
114  * NZCV flag state.
115  */
116 static inline uint32_t float_rel_to_flags(int res)
117 {
118     uint64_t flags;
119     switch (res) {
120     case float_relation_equal:
121         flags = PSTATE_Z | PSTATE_C;
122         break;
123     case float_relation_less:
124         flags = PSTATE_N;
125         break;
126     case float_relation_greater:
127         flags = PSTATE_C;
128         break;
129     case float_relation_unordered:
130     default:
131         flags = PSTATE_C | PSTATE_V;
132         break;
133     }
134     return flags;
135 }
136 
137 uint64_t HELPER(vfp_cmph_a64)(uint32_t x, uint32_t y, float_status *fp_status)
138 {
139     return float_rel_to_flags(float16_compare_quiet(x, y, fp_status));
140 }
141 
142 uint64_t HELPER(vfp_cmpeh_a64)(uint32_t x, uint32_t y, float_status *fp_status)
143 {
144     return float_rel_to_flags(float16_compare(x, y, fp_status));
145 }
146 
147 uint64_t HELPER(vfp_cmps_a64)(float32 x, float32 y, float_status *fp_status)
148 {
149     return float_rel_to_flags(float32_compare_quiet(x, y, fp_status));
150 }
151 
152 uint64_t HELPER(vfp_cmpes_a64)(float32 x, float32 y, float_status *fp_status)
153 {
154     return float_rel_to_flags(float32_compare(x, y, fp_status));
155 }
156 
157 uint64_t HELPER(vfp_cmpd_a64)(float64 x, float64 y, float_status *fp_status)
158 {
159     return float_rel_to_flags(float64_compare_quiet(x, y, fp_status));
160 }
161 
162 uint64_t HELPER(vfp_cmped_a64)(float64 x, float64 y, float_status *fp_status)
163 {
164     return float_rel_to_flags(float64_compare(x, y, fp_status));
165 }
166 
167 float32 HELPER(vfp_mulxs)(float32 a, float32 b, float_status *fpst)
168 {
169     a = float32_squash_input_denormal(a, fpst);
170     b = float32_squash_input_denormal(b, fpst);
171 
172     if ((float32_is_zero(a) && float32_is_infinity(b)) ||
173         (float32_is_infinity(a) && float32_is_zero(b))) {
174         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
175         return make_float32((1U << 30) |
176                             ((float32_val(a) ^ float32_val(b)) & (1U << 31)));
177     }
178     return float32_mul(a, b, fpst);
179 }
180 
181 float64 HELPER(vfp_mulxd)(float64 a, float64 b, float_status *fpst)
182 {
183     a = float64_squash_input_denormal(a, fpst);
184     b = float64_squash_input_denormal(b, fpst);
185 
186     if ((float64_is_zero(a) && float64_is_infinity(b)) ||
187         (float64_is_infinity(a) && float64_is_zero(b))) {
188         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
189         return make_float64((1ULL << 62) |
190                             ((float64_val(a) ^ float64_val(b)) & (1ULL << 63)));
191     }
192     return float64_mul(a, b, fpst);
193 }
194 
195 /* 64bit/double versions of the neon float compare functions */
196 uint64_t HELPER(neon_ceq_f64)(float64 a, float64 b, float_status *fpst)
197 {
198     return -float64_eq_quiet(a, b, fpst);
199 }
200 
201 uint64_t HELPER(neon_cge_f64)(float64 a, float64 b, float_status *fpst)
202 {
203     return -float64_le(b, a, fpst);
204 }
205 
206 uint64_t HELPER(neon_cgt_f64)(float64 a, float64 b, float_status *fpst)
207 {
208     return -float64_lt(b, a, fpst);
209 }
210 
211 /* Reciprocal step and sqrt step. Note that unlike the A32/T32
212  * versions, these do a fully fused multiply-add or
213  * multiply-add-and-halve.
214  */
215 
216 uint32_t HELPER(recpsf_f16)(uint32_t a, uint32_t b, float_status *fpst)
217 {
218     a = float16_squash_input_denormal(a, fpst);
219     b = float16_squash_input_denormal(b, fpst);
220 
221     a = float16_chs(a);
222     if ((float16_is_infinity(a) && float16_is_zero(b)) ||
223         (float16_is_infinity(b) && float16_is_zero(a))) {
224         return float16_two;
225     }
226     return float16_muladd(a, b, float16_two, 0, fpst);
227 }
228 
229 float32 HELPER(recpsf_f32)(float32 a, float32 b, float_status *fpst)
230 {
231     a = float32_squash_input_denormal(a, fpst);
232     b = float32_squash_input_denormal(b, fpst);
233 
234     a = float32_chs(a);
235     if ((float32_is_infinity(a) && float32_is_zero(b)) ||
236         (float32_is_infinity(b) && float32_is_zero(a))) {
237         return float32_two;
238     }
239     return float32_muladd(a, b, float32_two, 0, fpst);
240 }
241 
242 float64 HELPER(recpsf_f64)(float64 a, float64 b, float_status *fpst)
243 {
244     a = float64_squash_input_denormal(a, fpst);
245     b = float64_squash_input_denormal(b, fpst);
246 
247     a = float64_chs(a);
248     if ((float64_is_infinity(a) && float64_is_zero(b)) ||
249         (float64_is_infinity(b) && float64_is_zero(a))) {
250         return float64_two;
251     }
252     return float64_muladd(a, b, float64_two, 0, fpst);
253 }
254 
255 uint32_t HELPER(rsqrtsf_f16)(uint32_t a, uint32_t b, float_status *fpst)
256 {
257     a = float16_squash_input_denormal(a, fpst);
258     b = float16_squash_input_denormal(b, fpst);
259 
260     a = float16_chs(a);
261     if ((float16_is_infinity(a) && float16_is_zero(b)) ||
262         (float16_is_infinity(b) && float16_is_zero(a))) {
263         return float16_one_point_five;
264     }
265     return float16_muladd_scalbn(a, b, float16_three, -1, 0, fpst);
266 }
267 
268 float32 HELPER(rsqrtsf_f32)(float32 a, float32 b, float_status *fpst)
269 {
270     a = float32_squash_input_denormal(a, fpst);
271     b = float32_squash_input_denormal(b, fpst);
272 
273     a = float32_chs(a);
274     if ((float32_is_infinity(a) && float32_is_zero(b)) ||
275         (float32_is_infinity(b) && float32_is_zero(a))) {
276         return float32_one_point_five;
277     }
278     return float32_muladd_scalbn(a, b, float32_three, -1, 0, fpst);
279 }
280 
281 float64 HELPER(rsqrtsf_f64)(float64 a, float64 b, float_status *fpst)
282 {
283     a = float64_squash_input_denormal(a, fpst);
284     b = float64_squash_input_denormal(b, fpst);
285 
286     a = float64_chs(a);
287     if ((float64_is_infinity(a) && float64_is_zero(b)) ||
288         (float64_is_infinity(b) && float64_is_zero(a))) {
289         return float64_one_point_five;
290     }
291     return float64_muladd_scalbn(a, b, float64_three, -1, 0, fpst);
292 }
293 
294 /* Floating-point reciprocal exponent - see FPRecpX in ARM ARM */
295 uint32_t HELPER(frecpx_f16)(uint32_t a, float_status *fpst)
296 {
297     uint16_t val16, sbit;
298     int16_t exp;
299 
300     if (float16_is_any_nan(a)) {
301         float16 nan = a;
302         if (float16_is_signaling_nan(a, fpst)) {
303             float_raise(float_flag_invalid, fpst);
304             if (!fpst->default_nan_mode) {
305                 nan = float16_silence_nan(a, fpst);
306             }
307         }
308         if (fpst->default_nan_mode) {
309             nan = float16_default_nan(fpst);
310         }
311         return nan;
312     }
313 
314     a = float16_squash_input_denormal(a, fpst);
315 
316     val16 = float16_val(a);
317     sbit = 0x8000 & val16;
318     exp = extract32(val16, 10, 5);
319 
320     if (exp == 0) {
321         return make_float16(deposit32(sbit, 10, 5, 0x1e));
322     } else {
323         return make_float16(deposit32(sbit, 10, 5, ~exp));
324     }
325 }
326 
327 float32 HELPER(frecpx_f32)(float32 a, float_status *fpst)
328 {
329     uint32_t val32, sbit;
330     int32_t exp;
331 
332     if (float32_is_any_nan(a)) {
333         float32 nan = a;
334         if (float32_is_signaling_nan(a, fpst)) {
335             float_raise(float_flag_invalid, fpst);
336             if (!fpst->default_nan_mode) {
337                 nan = float32_silence_nan(a, fpst);
338             }
339         }
340         if (fpst->default_nan_mode) {
341             nan = float32_default_nan(fpst);
342         }
343         return nan;
344     }
345 
346     a = float32_squash_input_denormal(a, fpst);
347 
348     val32 = float32_val(a);
349     sbit = 0x80000000ULL & val32;
350     exp = extract32(val32, 23, 8);
351 
352     if (exp == 0) {
353         return make_float32(sbit | (0xfe << 23));
354     } else {
355         return make_float32(sbit | (~exp & 0xff) << 23);
356     }
357 }
358 
359 float64 HELPER(frecpx_f64)(float64 a, float_status *fpst)
360 {
361     uint64_t val64, sbit;
362     int64_t exp;
363 
364     if (float64_is_any_nan(a)) {
365         float64 nan = a;
366         if (float64_is_signaling_nan(a, fpst)) {
367             float_raise(float_flag_invalid, fpst);
368             if (!fpst->default_nan_mode) {
369                 nan = float64_silence_nan(a, fpst);
370             }
371         }
372         if (fpst->default_nan_mode) {
373             nan = float64_default_nan(fpst);
374         }
375         return nan;
376     }
377 
378     a = float64_squash_input_denormal(a, fpst);
379 
380     val64 = float64_val(a);
381     sbit = 0x8000000000000000ULL & val64;
382     exp = extract64(float64_val(a), 52, 11);
383 
384     if (exp == 0) {
385         return make_float64(sbit | (0x7feULL << 52));
386     } else {
387         return make_float64(sbit | (~exp & 0x7ffULL) << 52);
388     }
389 }
390 
391 float32 HELPER(fcvtx_f64_to_f32)(float64 a, float_status *fpst)
392 {
393     float32 r;
394     int old = get_float_rounding_mode(fpst);
395 
396     set_float_rounding_mode(float_round_to_odd, fpst);
397     r = float64_to_float32(a, fpst);
398     set_float_rounding_mode(old, fpst);
399     return r;
400 }
401 
402 /* 64-bit versions of the CRC helpers. Note that although the operation
403  * (and the prototypes of crc32c() and crc32() mean that only the bottom
404  * 32 bits of the accumulator and result are used, we pass and return
405  * uint64_t for convenience of the generated code. Unlike the 32-bit
406  * instruction set versions, val may genuinely have 64 bits of data in it.
407  * The upper bytes of val (above the number specified by 'bytes') must have
408  * been zeroed out by the caller.
409  */
410 uint64_t HELPER(crc32_64)(uint64_t acc, uint64_t val, uint32_t bytes)
411 {
412     uint8_t buf[8];
413 
414     stq_le_p(buf, val);
415 
416     /* zlib crc32 converts the accumulator and output to one's complement.  */
417     return crc32(acc ^ 0xffffffff, buf, bytes) ^ 0xffffffff;
418 }
419 
420 uint64_t HELPER(crc32c_64)(uint64_t acc, uint64_t val, uint32_t bytes)
421 {
422     uint8_t buf[8];
423 
424     stq_le_p(buf, val);
425 
426     /* Linux crc32c converts the output to one's complement.  */
427     return crc32c(acc, buf, bytes) ^ 0xffffffff;
428 }
429 
430 /*
431  * AdvSIMD half-precision
432  */
433 
434 #define ADVSIMD_HELPER(name, suffix) HELPER(glue(glue(advsimd_, name), suffix))
435 
436 #define ADVSIMD_HALFOP(name) \
437 uint32_t ADVSIMD_HELPER(name, h)(uint32_t a, uint32_t b, float_status *fpst) \
438 { \
439     return float16_ ## name(a, b, fpst);    \
440 }
441 
442 #define ADVSIMD_TWOHALFOP(name)                                         \
443 uint32_t ADVSIMD_HELPER(name, 2h)(uint32_t two_a, uint32_t two_b,       \
444                                   float_status *fpst)                   \
445 { \
446     float16  a1, a2, b1, b2;                        \
447     uint32_t r1, r2;                                \
448     a1 = extract32(two_a, 0, 16);                   \
449     a2 = extract32(two_a, 16, 16);                  \
450     b1 = extract32(two_b, 0, 16);                   \
451     b2 = extract32(two_b, 16, 16);                  \
452     r1 = float16_ ## name(a1, b1, fpst);            \
453     r2 = float16_ ## name(a2, b2, fpst);            \
454     return deposit32(r1, 16, 16, r2);               \
455 }
456 
457 ADVSIMD_TWOHALFOP(add)
458 ADVSIMD_TWOHALFOP(sub)
459 ADVSIMD_TWOHALFOP(mul)
460 ADVSIMD_TWOHALFOP(div)
461 ADVSIMD_TWOHALFOP(min)
462 ADVSIMD_TWOHALFOP(max)
463 ADVSIMD_TWOHALFOP(minnum)
464 ADVSIMD_TWOHALFOP(maxnum)
465 
466 /* Data processing - scalar floating-point and advanced SIMD */
467 static float16 float16_mulx(float16 a, float16 b, float_status *fpst)
468 {
469     a = float16_squash_input_denormal(a, fpst);
470     b = float16_squash_input_denormal(b, fpst);
471 
472     if ((float16_is_zero(a) && float16_is_infinity(b)) ||
473         (float16_is_infinity(a) && float16_is_zero(b))) {
474         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
475         return make_float16((1U << 14) |
476                             ((float16_val(a) ^ float16_val(b)) & (1U << 15)));
477     }
478     return float16_mul(a, b, fpst);
479 }
480 
481 ADVSIMD_HALFOP(mulx)
482 ADVSIMD_TWOHALFOP(mulx)
483 
484 /* fused multiply-accumulate */
485 uint32_t HELPER(advsimd_muladdh)(uint32_t a, uint32_t b, uint32_t c,
486                                  float_status *fpst)
487 {
488     return float16_muladd(a, b, c, 0, fpst);
489 }
490 
491 uint32_t HELPER(advsimd_muladd2h)(uint32_t two_a, uint32_t two_b,
492                                   uint32_t two_c, float_status *fpst)
493 {
494     float16  a1, a2, b1, b2, c1, c2;
495     uint32_t r1, r2;
496     a1 = extract32(two_a, 0, 16);
497     a2 = extract32(two_a, 16, 16);
498     b1 = extract32(two_b, 0, 16);
499     b2 = extract32(two_b, 16, 16);
500     c1 = extract32(two_c, 0, 16);
501     c2 = extract32(two_c, 16, 16);
502     r1 = float16_muladd(a1, b1, c1, 0, fpst);
503     r2 = float16_muladd(a2, b2, c2, 0, fpst);
504     return deposit32(r1, 16, 16, r2);
505 }
506 
507 /*
508  * Floating point comparisons produce an integer result. Softfloat
509  * routines return float_relation types which we convert to the 0/-1
510  * Neon requires.
511  */
512 
513 #define ADVSIMD_CMPRES(test) (test) ? 0xffff : 0
514 
515 uint32_t HELPER(advsimd_ceq_f16)(uint32_t a, uint32_t b, float_status *fpst)
516 {
517     int compare = float16_compare_quiet(a, b, fpst);
518     return ADVSIMD_CMPRES(compare == float_relation_equal);
519 }
520 
521 uint32_t HELPER(advsimd_cge_f16)(uint32_t a, uint32_t b, float_status *fpst)
522 {
523     int compare = float16_compare(a, b, fpst);
524     return ADVSIMD_CMPRES(compare == float_relation_greater ||
525                           compare == float_relation_equal);
526 }
527 
528 uint32_t HELPER(advsimd_cgt_f16)(uint32_t a, uint32_t b, float_status *fpst)
529 {
530     int compare = float16_compare(a, b, fpst);
531     return ADVSIMD_CMPRES(compare == float_relation_greater);
532 }
533 
534 uint32_t HELPER(advsimd_acge_f16)(uint32_t a, uint32_t b, float_status *fpst)
535 {
536     float16 f0 = float16_abs(a);
537     float16 f1 = float16_abs(b);
538     int compare = float16_compare(f0, f1, fpst);
539     return ADVSIMD_CMPRES(compare == float_relation_greater ||
540                           compare == float_relation_equal);
541 }
542 
543 uint32_t HELPER(advsimd_acgt_f16)(uint32_t a, uint32_t b, float_status *fpst)
544 {
545     float16 f0 = float16_abs(a);
546     float16 f1 = float16_abs(b);
547     int compare = float16_compare(f0, f1, fpst);
548     return ADVSIMD_CMPRES(compare == float_relation_greater);
549 }
550 
551 /* round to integral */
552 uint32_t HELPER(advsimd_rinth_exact)(uint32_t x, float_status *fp_status)
553 {
554     return float16_round_to_int(x, fp_status);
555 }
556 
557 uint32_t HELPER(advsimd_rinth)(uint32_t x, float_status *fp_status)
558 {
559     int old_flags = get_float_exception_flags(fp_status), new_flags;
560     float16 ret;
561 
562     ret = float16_round_to_int(x, fp_status);
563 
564     /* Suppress any inexact exceptions the conversion produced */
565     if (!(old_flags & float_flag_inexact)) {
566         new_flags = get_float_exception_flags(fp_status);
567         set_float_exception_flags(new_flags & ~float_flag_inexact, fp_status);
568     }
569 
570     return ret;
571 }
572 
573 static int el_from_spsr(uint32_t spsr)
574 {
575     /* Return the exception level that this SPSR is requesting a return to,
576      * or -1 if it is invalid (an illegal return)
577      */
578     if (spsr & PSTATE_nRW) {
579         switch (spsr & CPSR_M) {
580         case ARM_CPU_MODE_USR:
581             return 0;
582         case ARM_CPU_MODE_HYP:
583             return 2;
584         case ARM_CPU_MODE_FIQ:
585         case ARM_CPU_MODE_IRQ:
586         case ARM_CPU_MODE_SVC:
587         case ARM_CPU_MODE_ABT:
588         case ARM_CPU_MODE_UND:
589         case ARM_CPU_MODE_SYS:
590             return 1;
591         case ARM_CPU_MODE_MON:
592             /* Returning to Mon from AArch64 is never possible,
593              * so this is an illegal return.
594              */
595         default:
596             return -1;
597         }
598     } else {
599         if (extract32(spsr, 1, 1)) {
600             /* Return with reserved M[1] bit set */
601             return -1;
602         }
603         if (extract32(spsr, 0, 4) == 1) {
604             /* return to EL0 with M[0] bit set */
605             return -1;
606         }
607         return extract32(spsr, 2, 2);
608     }
609 }
610 
611 static void cpsr_write_from_spsr_elx(CPUARMState *env,
612                                      uint32_t val)
613 {
614     uint32_t mask;
615 
616     /* Save SPSR_ELx.SS into PSTATE. */
617     env->pstate = (env->pstate & ~PSTATE_SS) | (val & PSTATE_SS);
618     val &= ~PSTATE_SS;
619 
620     /* Move DIT to the correct location for CPSR */
621     if (val & PSTATE_DIT) {
622         val &= ~PSTATE_DIT;
623         val |= CPSR_DIT;
624     }
625 
626     mask = aarch32_cpsr_valid_mask(env->features, \
627         &env_archcpu(env)->isar);
628     cpsr_write(env, val, mask, CPSRWriteRaw);
629 }
630 
631 void HELPER(exception_return)(CPUARMState *env, uint64_t new_pc)
632 {
633     int cur_el = arm_current_el(env);
634     unsigned int spsr_idx = aarch64_banked_spsr_index(cur_el);
635     uint32_t spsr = env->banked_spsr[spsr_idx];
636     int new_el;
637     bool return_to_aa64 = (spsr & PSTATE_nRW) == 0;
638 
639     aarch64_save_sp(env, cur_el);
640 
641     arm_clear_exclusive(env);
642 
643     /* We must squash the PSTATE.SS bit to zero unless both of the
644      * following hold:
645      *  1. debug exceptions are currently disabled
646      *  2. singlestep will be active in the EL we return to
647      * We check 1 here and 2 after we've done the pstate/cpsr write() to
648      * transition to the EL we're going to.
649      */
650     if (arm_generate_debug_exceptions(env)) {
651         spsr &= ~PSTATE_SS;
652     }
653 
654     /*
655      * FEAT_RME forbids return from EL3 with an invalid security state.
656      * We don't need an explicit check for FEAT_RME here because we enforce
657      * in scr_write() that you can't set the NSE bit without it.
658      */
659     if (cur_el == 3 && (env->cp15.scr_el3 & (SCR_NS | SCR_NSE)) == SCR_NSE) {
660         goto illegal_return;
661     }
662 
663     new_el = el_from_spsr(spsr);
664     if (new_el == -1) {
665         goto illegal_return;
666     }
667     if (new_el > cur_el || (new_el == 2 && !arm_is_el2_enabled(env))) {
668         /* Disallow return to an EL which is unimplemented or higher
669          * than the current one.
670          */
671         goto illegal_return;
672     }
673 
674     if (new_el != 0 && arm_el_is_aa64(env, new_el) != return_to_aa64) {
675         /* Return to an EL which is configured for a different register width */
676         goto illegal_return;
677     }
678 
679     if (new_el == 1 && (arm_hcr_el2_eff(env) & HCR_TGE)) {
680         goto illegal_return;
681     }
682 
683     bql_lock();
684     arm_call_pre_el_change_hook(env_archcpu(env));
685     bql_unlock();
686 
687     if (!return_to_aa64) {
688         env->aarch64 = false;
689         /* We do a raw CPSR write because aarch64_sync_64_to_32()
690          * will sort the register banks out for us, and we've already
691          * caught all the bad-mode cases in el_from_spsr().
692          */
693         cpsr_write_from_spsr_elx(env, spsr);
694         if (!arm_singlestep_active(env)) {
695             env->pstate &= ~PSTATE_SS;
696         }
697         aarch64_sync_64_to_32(env);
698 
699         if (spsr & CPSR_T) {
700             env->regs[15] = new_pc & ~0x1;
701         } else {
702             env->regs[15] = new_pc & ~0x3;
703         }
704         helper_rebuild_hflags_a32(env, new_el);
705         qemu_log_mask(CPU_LOG_INT, "Exception return from AArch64 EL%d to "
706                       "AArch32 EL%d PC 0x%" PRIx32 "\n",
707                       cur_el, new_el, env->regs[15]);
708     } else {
709         int tbii;
710 
711         env->aarch64 = true;
712         spsr &= aarch64_pstate_valid_mask(&env_archcpu(env)->isar);
713         pstate_write(env, spsr);
714         if (!arm_singlestep_active(env)) {
715             env->pstate &= ~PSTATE_SS;
716         }
717         aarch64_restore_sp(env, new_el);
718         helper_rebuild_hflags_a64(env, new_el);
719 
720         /*
721          * Apply TBI to the exception return address.  We had to delay this
722          * until after we selected the new EL, so that we could select the
723          * correct TBI+TBID bits.  This is made easier by waiting until after
724          * the hflags rebuild, since we can pull the composite TBII field
725          * from there.
726          */
727         tbii = EX_TBFLAG_A64(env->hflags, TBII);
728         if ((tbii >> extract64(new_pc, 55, 1)) & 1) {
729             /* TBI is enabled. */
730             int core_mmu_idx = arm_env_mmu_index(env);
731             if (regime_has_2_ranges(core_to_aa64_mmu_idx(core_mmu_idx))) {
732                 new_pc = sextract64(new_pc, 0, 56);
733             } else {
734                 new_pc = extract64(new_pc, 0, 56);
735             }
736         }
737         env->pc = new_pc;
738 
739         qemu_log_mask(CPU_LOG_INT, "Exception return from AArch64 EL%d to "
740                       "AArch64 EL%d PC 0x%" PRIx64 "\n",
741                       cur_el, new_el, env->pc);
742     }
743 
744     /*
745      * Note that cur_el can never be 0.  If new_el is 0, then
746      * el0_a64 is return_to_aa64, else el0_a64 is ignored.
747      */
748     aarch64_sve_change_el(env, cur_el, new_el, return_to_aa64);
749 
750     bql_lock();
751     arm_call_el_change_hook(env_archcpu(env));
752     bql_unlock();
753 
754     return;
755 
756 illegal_return:
757     /* Illegal return events of various kinds have architecturally
758      * mandated behaviour:
759      * restore NZCV and DAIF from SPSR_ELx
760      * set PSTATE.IL
761      * restore PC from ELR_ELx
762      * no change to exception level, execution state or stack pointer
763      */
764     env->pstate |= PSTATE_IL;
765     env->pc = new_pc;
766     spsr &= PSTATE_NZCV | PSTATE_DAIF | PSTATE_ALLINT;
767     spsr |= pstate_read(env) & ~(PSTATE_NZCV | PSTATE_DAIF | PSTATE_ALLINT);
768     pstate_write(env, spsr);
769     if (!arm_singlestep_active(env)) {
770         env->pstate &= ~PSTATE_SS;
771     }
772     helper_rebuild_hflags_a64(env, cur_el);
773     qemu_log_mask(LOG_GUEST_ERROR, "Illegal exception return at EL%d: "
774                   "resuming execution at 0x%" PRIx64 "\n", cur_el, env->pc);
775 }
776 
777 void HELPER(dc_zva)(CPUARMState *env, uint64_t vaddr_in)
778 {
779     uintptr_t ra = GETPC();
780 
781     /*
782      * Implement DC ZVA, which zeroes a fixed-length block of memory.
783      * Note that we do not implement the (architecturally mandated)
784      * alignment fault for attempts to use this on Device memory
785      * (which matches the usual QEMU behaviour of not implementing either
786      * alignment faults or any memory attribute handling).
787      */
788     int blocklen = 4 << env_archcpu(env)->dcz_blocksize;
789     uint64_t vaddr = vaddr_in & ~(blocklen - 1);
790     int mmu_idx = arm_env_mmu_index(env);
791     void *mem;
792 
793     /*
794      * Trapless lookup.  In addition to actual invalid page, may
795      * return NULL for I/O, watchpoints, clean pages, etc.
796      */
797     mem = tlb_vaddr_to_host(env, vaddr, MMU_DATA_STORE, mmu_idx);
798 
799 #ifndef CONFIG_USER_ONLY
800     if (unlikely(!mem)) {
801         /*
802          * Trap if accessing an invalid page.  DC_ZVA requires that we supply
803          * the original pointer for an invalid page.  But watchpoints require
804          * that we probe the actual space.  So do both.
805          */
806         (void) probe_write(env, vaddr_in, 1, mmu_idx, ra);
807         mem = probe_write(env, vaddr, blocklen, mmu_idx, ra);
808 
809         if (unlikely(!mem)) {
810             /*
811              * The only remaining reason for mem == NULL is I/O.
812              * Just do a series of byte writes as the architecture demands.
813              */
814             for (int i = 0; i < blocklen; i++) {
815                 cpu_stb_mmuidx_ra(env, vaddr + i, 0, mmu_idx, ra);
816             }
817             return;
818         }
819     }
820 #endif
821 
822     set_helper_retaddr(ra);
823     memset(mem, 0, blocklen);
824     clear_helper_retaddr();
825 }
826 
827 void HELPER(unaligned_access)(CPUARMState *env, uint64_t addr,
828                               uint32_t access_type, uint32_t mmu_idx)
829 {
830     arm_cpu_do_unaligned_access(env_cpu(env), addr, access_type,
831                                 mmu_idx, GETPC());
832 }
833 
834 /* Memory operations (memset, memmove, memcpy) */
835 
836 /*
837  * Return true if the CPY* and SET* insns can execute; compare
838  * pseudocode CheckMOPSEnabled(), though we refactor it a little.
839  */
840 static bool mops_enabled(CPUARMState *env)
841 {
842     int el = arm_current_el(env);
843 
844     if (el < 2 &&
845         (arm_hcr_el2_eff(env) & (HCR_E2H | HCR_TGE)) != (HCR_E2H | HCR_TGE) &&
846         !(arm_hcrx_el2_eff(env) & HCRX_MSCEN)) {
847         return false;
848     }
849 
850     if (el == 0) {
851         if (!el_is_in_host(env, 0)) {
852             return env->cp15.sctlr_el[1] & SCTLR_MSCEN;
853         } else {
854             return env->cp15.sctlr_el[2] & SCTLR_MSCEN;
855         }
856     }
857     return true;
858 }
859 
860 static void check_mops_enabled(CPUARMState *env, uintptr_t ra)
861 {
862     if (!mops_enabled(env)) {
863         raise_exception_ra(env, EXCP_UDEF, syn_uncategorized(),
864                            exception_target_el(env), ra);
865     }
866 }
867 
868 /*
869  * Return the target exception level for an exception due
870  * to mismatched arguments in a FEAT_MOPS copy or set.
871  * Compare pseudocode MismatchedCpySetTargetEL()
872  */
873 static int mops_mismatch_exception_target_el(CPUARMState *env)
874 {
875     int el = arm_current_el(env);
876 
877     if (el > 1) {
878         return el;
879     }
880     if (el == 0 && (arm_hcr_el2_eff(env) & HCR_TGE)) {
881         return 2;
882     }
883     if (el == 1 && (arm_hcrx_el2_eff(env) & HCRX_MCE2)) {
884         return 2;
885     }
886     return 1;
887 }
888 
889 /*
890  * Check whether an M or E instruction was executed with a CF value
891  * indicating the wrong option for this implementation.
892  * Assumes we are always Option A.
893  */
894 static void check_mops_wrong_option(CPUARMState *env, uint32_t syndrome,
895                                     uintptr_t ra)
896 {
897     if (env->CF != 0) {
898         syndrome |= 1 << 17; /* Set the wrong-option bit */
899         raise_exception_ra(env, EXCP_UDEF, syndrome,
900                            mops_mismatch_exception_target_el(env), ra);
901     }
902 }
903 
904 /*
905  * Return the maximum number of bytes we can transfer starting at addr
906  * without crossing a page boundary.
907  */
908 static uint64_t page_limit(uint64_t addr)
909 {
910     return TARGET_PAGE_ALIGN(addr + 1) - addr;
911 }
912 
913 /*
914  * Return the number of bytes we can copy starting from addr and working
915  * backwards without crossing a page boundary.
916  */
917 static uint64_t page_limit_rev(uint64_t addr)
918 {
919     return (addr & ~TARGET_PAGE_MASK) + 1;
920 }
921 
922 /*
923  * Perform part of a memory set on an area of guest memory starting at
924  * toaddr (a dirty address) and extending for setsize bytes.
925  *
926  * Returns the number of bytes actually set, which might be less than
927  * setsize; the caller should loop until the whole set has been done.
928  * The caller should ensure that the guest registers are correct
929  * for the possibility that the first byte of the set encounters
930  * an exception or watchpoint. We guarantee not to take any faults
931  * for bytes other than the first.
932  */
933 static uint64_t set_step(CPUARMState *env, uint64_t toaddr,
934                          uint64_t setsize, uint32_t data, int memidx,
935                          uint32_t *mtedesc, uintptr_t ra)
936 {
937     void *mem;
938 
939     setsize = MIN(setsize, page_limit(toaddr));
940     if (*mtedesc) {
941         uint64_t mtesize = mte_mops_probe(env, toaddr, setsize, *mtedesc);
942         if (mtesize == 0) {
943             /* Trap, or not. All CPU state is up to date */
944             mte_check_fail(env, *mtedesc, toaddr, ra);
945             /* Continue, with no further MTE checks required */
946             *mtedesc = 0;
947         } else {
948             /* Advance to the end, or to the tag mismatch */
949             setsize = MIN(setsize, mtesize);
950         }
951     }
952 
953     toaddr = useronly_clean_ptr(toaddr);
954     /*
955      * Trapless lookup: returns NULL for invalid page, I/O,
956      * watchpoints, clean pages, etc.
957      */
958     mem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, memidx);
959 
960 #ifndef CONFIG_USER_ONLY
961     if (unlikely(!mem)) {
962         /*
963          * Slow-path: just do one byte write. This will handle the
964          * watchpoint, invalid page, etc handling correctly.
965          * For clean code pages, the next iteration will see
966          * the page dirty and will use the fast path.
967          */
968         cpu_stb_mmuidx_ra(env, toaddr, data, memidx, ra);
969         return 1;
970     }
971 #endif
972     /* Easy case: just memset the host memory */
973     set_helper_retaddr(ra);
974     memset(mem, data, setsize);
975     clear_helper_retaddr();
976     return setsize;
977 }
978 
979 /*
980  * Similar, but setting tags. The architecture requires us to do this
981  * in 16-byte chunks. SETP accesses are not tag checked; they set
982  * the tags.
983  */
984 static uint64_t set_step_tags(CPUARMState *env, uint64_t toaddr,
985                               uint64_t setsize, uint32_t data, int memidx,
986                               uint32_t *mtedesc, uintptr_t ra)
987 {
988     void *mem;
989     uint64_t cleanaddr;
990 
991     setsize = MIN(setsize, page_limit(toaddr));
992 
993     cleanaddr = useronly_clean_ptr(toaddr);
994     /*
995      * Trapless lookup: returns NULL for invalid page, I/O,
996      * watchpoints, clean pages, etc.
997      */
998     mem = tlb_vaddr_to_host(env, cleanaddr, MMU_DATA_STORE, memidx);
999 
1000 #ifndef CONFIG_USER_ONLY
1001     if (unlikely(!mem)) {
1002         /*
1003          * Slow-path: just do one write. This will handle the
1004          * watchpoint, invalid page, etc handling correctly.
1005          * The architecture requires that we do 16 bytes at a time,
1006          * and we know both ptr and size are 16 byte aligned.
1007          * For clean code pages, the next iteration will see
1008          * the page dirty and will use the fast path.
1009          */
1010         uint64_t repldata = data * 0x0101010101010101ULL;
1011         MemOpIdx oi16 = make_memop_idx(MO_TE | MO_128, memidx);
1012         cpu_st16_mmu(env, toaddr, int128_make128(repldata, repldata), oi16, ra);
1013         mte_mops_set_tags(env, toaddr, 16, *mtedesc);
1014         return 16;
1015     }
1016 #endif
1017     /* Easy case: just memset the host memory */
1018     set_helper_retaddr(ra);
1019     memset(mem, data, setsize);
1020     clear_helper_retaddr();
1021     mte_mops_set_tags(env, toaddr, setsize, *mtedesc);
1022     return setsize;
1023 }
1024 
1025 typedef uint64_t StepFn(CPUARMState *env, uint64_t toaddr,
1026                         uint64_t setsize, uint32_t data,
1027                         int memidx, uint32_t *mtedesc, uintptr_t ra);
1028 
1029 /* Extract register numbers from a MOPS exception syndrome value */
1030 static int mops_destreg(uint32_t syndrome)
1031 {
1032     return extract32(syndrome, 10, 5);
1033 }
1034 
1035 static int mops_srcreg(uint32_t syndrome)
1036 {
1037     return extract32(syndrome, 5, 5);
1038 }
1039 
1040 static int mops_sizereg(uint32_t syndrome)
1041 {
1042     return extract32(syndrome, 0, 5);
1043 }
1044 
1045 /*
1046  * Return true if TCMA and TBI bits mean we need to do MTE checks.
1047  * We only need to do this once per MOPS insn, not for every page.
1048  */
1049 static bool mte_checks_needed(uint64_t ptr, uint32_t desc)
1050 {
1051     int bit55 = extract64(ptr, 55, 1);
1052 
1053     /*
1054      * Note that tbi_check() returns true for "access checked" but
1055      * tcma_check() returns true for "access unchecked".
1056      */
1057     if (!tbi_check(desc, bit55)) {
1058         return false;
1059     }
1060     return !tcma_check(desc, bit55, allocation_tag_from_addr(ptr));
1061 }
1062 
1063 /* Take an exception if the SETG addr/size are not granule aligned */
1064 static void check_setg_alignment(CPUARMState *env, uint64_t ptr, uint64_t size,
1065                                  uint32_t memidx, uintptr_t ra)
1066 {
1067     if ((size != 0 && !QEMU_IS_ALIGNED(ptr, TAG_GRANULE)) ||
1068         !QEMU_IS_ALIGNED(size, TAG_GRANULE)) {
1069         arm_cpu_do_unaligned_access(env_cpu(env), ptr, MMU_DATA_STORE,
1070                                     memidx, ra);
1071 
1072     }
1073 }
1074 
1075 static uint64_t arm_reg_or_xzr(CPUARMState *env, int reg)
1076 {
1077     /*
1078      * Runtime equivalent of cpu_reg() -- return the CPU register value,
1079      * for contexts when index 31 means XZR (not SP).
1080      */
1081     return reg == 31 ? 0 : env->xregs[reg];
1082 }
1083 
1084 /*
1085  * For the Memory Set operation, our implementation chooses
1086  * always to use "option A", where we update Xd to the final
1087  * address in the SETP insn, and set Xn to be -(bytes remaining).
1088  * On SETM and SETE insns we only need update Xn.
1089  *
1090  * @env: CPU
1091  * @syndrome: syndrome value for mismatch exceptions
1092  * (also contains the register numbers we need to use)
1093  * @mtedesc: MTE descriptor word
1094  * @stepfn: function which does a single part of the set operation
1095  * @is_setg: true if this is the tag-setting SETG variant
1096  */
1097 static void do_setp(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1098                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1099 {
1100     /* Prologue: we choose to do up to the next page boundary */
1101     int rd = mops_destreg(syndrome);
1102     int rs = mops_srcreg(syndrome);
1103     int rn = mops_sizereg(syndrome);
1104     uint8_t data = arm_reg_or_xzr(env, rs);
1105     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1106     uint64_t toaddr = env->xregs[rd];
1107     uint64_t setsize = env->xregs[rn];
1108     uint64_t stagesetsize, step;
1109 
1110     check_mops_enabled(env, ra);
1111 
1112     if (setsize > INT64_MAX) {
1113         setsize = INT64_MAX;
1114         if (is_setg) {
1115             setsize &= ~0xf;
1116         }
1117     }
1118 
1119     if (unlikely(is_setg)) {
1120         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1121     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1122         mtedesc = 0;
1123     }
1124 
1125     stagesetsize = MIN(setsize, page_limit(toaddr));
1126     while (stagesetsize) {
1127         env->xregs[rd] = toaddr;
1128         env->xregs[rn] = setsize;
1129         step = stepfn(env, toaddr, stagesetsize, data, memidx, &mtedesc, ra);
1130         toaddr += step;
1131         setsize -= step;
1132         stagesetsize -= step;
1133     }
1134     /* Insn completed, so update registers to the Option A format */
1135     env->xregs[rd] = toaddr + setsize;
1136     env->xregs[rn] = -setsize;
1137 
1138     /* Set NZCV = 0000 to indicate we are an Option A implementation */
1139     env->NF = 0;
1140     env->ZF = 1; /* our env->ZF encoding is inverted */
1141     env->CF = 0;
1142     env->VF = 0;
1143     return;
1144 }
1145 
1146 void HELPER(setp)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1147 {
1148     do_setp(env, syndrome, mtedesc, set_step, false, GETPC());
1149 }
1150 
1151 void HELPER(setgp)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1152 {
1153     do_setp(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1154 }
1155 
1156 static void do_setm(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1157                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1158 {
1159     /* Main: we choose to do all the full-page chunks */
1160     CPUState *cs = env_cpu(env);
1161     int rd = mops_destreg(syndrome);
1162     int rs = mops_srcreg(syndrome);
1163     int rn = mops_sizereg(syndrome);
1164     uint8_t data = arm_reg_or_xzr(env, rs);
1165     uint64_t toaddr = env->xregs[rd] + env->xregs[rn];
1166     uint64_t setsize = -env->xregs[rn];
1167     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1168     uint64_t step, stagesetsize;
1169 
1170     check_mops_enabled(env, ra);
1171 
1172     /*
1173      * We're allowed to NOP out "no data to copy" before the consistency
1174      * checks; we choose to do so.
1175      */
1176     if (env->xregs[rn] == 0) {
1177         return;
1178     }
1179 
1180     check_mops_wrong_option(env, syndrome, ra);
1181 
1182     /*
1183      * Our implementation will work fine even if we have an unaligned
1184      * destination address, and because we update Xn every time around
1185      * the loop below and the return value from stepfn() may be less
1186      * than requested, we might find toaddr is unaligned. So we don't
1187      * have an IMPDEF check for alignment here.
1188      */
1189 
1190     if (unlikely(is_setg)) {
1191         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1192     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1193         mtedesc = 0;
1194     }
1195 
1196     /* Do the actual memset: we leave the last partial page to SETE */
1197     stagesetsize = setsize & TARGET_PAGE_MASK;
1198     while (stagesetsize > 0) {
1199         step = stepfn(env, toaddr, stagesetsize, data, memidx, &mtedesc, ra);
1200         toaddr += step;
1201         setsize -= step;
1202         stagesetsize -= step;
1203         env->xregs[rn] = -setsize;
1204         if (stagesetsize > 0 && unlikely(cpu_loop_exit_requested(cs))) {
1205             cpu_loop_exit_restore(cs, ra);
1206         }
1207     }
1208 }
1209 
1210 void HELPER(setm)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1211 {
1212     do_setm(env, syndrome, mtedesc, set_step, false, GETPC());
1213 }
1214 
1215 void HELPER(setgm)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1216 {
1217     do_setm(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1218 }
1219 
1220 static void do_sete(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1221                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1222 {
1223     /* Epilogue: do the last partial page */
1224     int rd = mops_destreg(syndrome);
1225     int rs = mops_srcreg(syndrome);
1226     int rn = mops_sizereg(syndrome);
1227     uint8_t data = arm_reg_or_xzr(env, rs);
1228     uint64_t toaddr = env->xregs[rd] + env->xregs[rn];
1229     uint64_t setsize = -env->xregs[rn];
1230     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1231     uint64_t step;
1232 
1233     check_mops_enabled(env, ra);
1234 
1235     /*
1236      * We're allowed to NOP out "no data to copy" before the consistency
1237      * checks; we choose to do so.
1238      */
1239     if (setsize == 0) {
1240         return;
1241     }
1242 
1243     check_mops_wrong_option(env, syndrome, ra);
1244 
1245     /*
1246      * Our implementation has no address alignment requirements, but
1247      * we do want to enforce the "less than a page" size requirement,
1248      * so we don't need to have the "check for interrupts" here.
1249      */
1250     if (setsize >= TARGET_PAGE_SIZE) {
1251         raise_exception_ra(env, EXCP_UDEF, syndrome,
1252                            mops_mismatch_exception_target_el(env), ra);
1253     }
1254 
1255     if (unlikely(is_setg)) {
1256         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1257     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1258         mtedesc = 0;
1259     }
1260 
1261     /* Do the actual memset */
1262     while (setsize > 0) {
1263         step = stepfn(env, toaddr, setsize, data, memidx, &mtedesc, ra);
1264         toaddr += step;
1265         setsize -= step;
1266         env->xregs[rn] = -setsize;
1267     }
1268 }
1269 
1270 void HELPER(sete)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1271 {
1272     do_sete(env, syndrome, mtedesc, set_step, false, GETPC());
1273 }
1274 
1275 void HELPER(setge)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1276 {
1277     do_sete(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1278 }
1279 
1280 /*
1281  * Perform part of a memory copy from the guest memory at fromaddr
1282  * and extending for copysize bytes, to the guest memory at
1283  * toaddr. Both addresses are dirty.
1284  *
1285  * Returns the number of bytes actually set, which might be less than
1286  * copysize; the caller should loop until the whole copy has been done.
1287  * The caller should ensure that the guest registers are correct
1288  * for the possibility that the first byte of the copy encounters
1289  * an exception or watchpoint. We guarantee not to take any faults
1290  * for bytes other than the first.
1291  */
1292 static uint64_t copy_step(CPUARMState *env, uint64_t toaddr, uint64_t fromaddr,
1293                           uint64_t copysize, int wmemidx, int rmemidx,
1294                           uint32_t *wdesc, uint32_t *rdesc, uintptr_t ra)
1295 {
1296     void *rmem;
1297     void *wmem;
1298 
1299     /* Don't cross a page boundary on either source or destination */
1300     copysize = MIN(copysize, page_limit(toaddr));
1301     copysize = MIN(copysize, page_limit(fromaddr));
1302     /*
1303      * Handle MTE tag checks: either handle the tag mismatch for byte 0,
1304      * or else copy up to but not including the byte with the mismatch.
1305      */
1306     if (*rdesc) {
1307         uint64_t mtesize = mte_mops_probe(env, fromaddr, copysize, *rdesc);
1308         if (mtesize == 0) {
1309             mte_check_fail(env, *rdesc, fromaddr, ra);
1310             *rdesc = 0;
1311         } else {
1312             copysize = MIN(copysize, mtesize);
1313         }
1314     }
1315     if (*wdesc) {
1316         uint64_t mtesize = mte_mops_probe(env, toaddr, copysize, *wdesc);
1317         if (mtesize == 0) {
1318             mte_check_fail(env, *wdesc, toaddr, ra);
1319             *wdesc = 0;
1320         } else {
1321             copysize = MIN(copysize, mtesize);
1322         }
1323     }
1324 
1325     toaddr = useronly_clean_ptr(toaddr);
1326     fromaddr = useronly_clean_ptr(fromaddr);
1327     /* Trapless lookup of whether we can get a host memory pointer */
1328     wmem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, wmemidx);
1329     rmem = tlb_vaddr_to_host(env, fromaddr, MMU_DATA_LOAD, rmemidx);
1330 
1331 #ifndef CONFIG_USER_ONLY
1332     /*
1333      * If we don't have host memory for both source and dest then just
1334      * do a single byte copy. This will handle watchpoints, invalid pages,
1335      * etc correctly. For clean code pages, the next iteration will see
1336      * the page dirty and will use the fast path.
1337      */
1338     if (unlikely(!rmem || !wmem)) {
1339         uint8_t byte;
1340         if (rmem) {
1341             byte = *(uint8_t *)rmem;
1342         } else {
1343             byte = cpu_ldub_mmuidx_ra(env, fromaddr, rmemidx, ra);
1344         }
1345         if (wmem) {
1346             *(uint8_t *)wmem = byte;
1347         } else {
1348             cpu_stb_mmuidx_ra(env, toaddr, byte, wmemidx, ra);
1349         }
1350         return 1;
1351     }
1352 #endif
1353     /* Easy case: just memmove the host memory */
1354     set_helper_retaddr(ra);
1355     memmove(wmem, rmem, copysize);
1356     clear_helper_retaddr();
1357     return copysize;
1358 }
1359 
1360 /*
1361  * Do part of a backwards memory copy. Here toaddr and fromaddr point
1362  * to the *last* byte to be copied.
1363  */
1364 static uint64_t copy_step_rev(CPUARMState *env, uint64_t toaddr,
1365                               uint64_t fromaddr,
1366                               uint64_t copysize, int wmemidx, int rmemidx,
1367                               uint32_t *wdesc, uint32_t *rdesc, uintptr_t ra)
1368 {
1369     void *rmem;
1370     void *wmem;
1371 
1372     /* Don't cross a page boundary on either source or destination */
1373     copysize = MIN(copysize, page_limit_rev(toaddr));
1374     copysize = MIN(copysize, page_limit_rev(fromaddr));
1375 
1376     /*
1377      * Handle MTE tag checks: either handle the tag mismatch for byte 0,
1378      * or else copy up to but not including the byte with the mismatch.
1379      */
1380     if (*rdesc) {
1381         uint64_t mtesize = mte_mops_probe_rev(env, fromaddr, copysize, *rdesc);
1382         if (mtesize == 0) {
1383             mte_check_fail(env, *rdesc, fromaddr, ra);
1384             *rdesc = 0;
1385         } else {
1386             copysize = MIN(copysize, mtesize);
1387         }
1388     }
1389     if (*wdesc) {
1390         uint64_t mtesize = mte_mops_probe_rev(env, toaddr, copysize, *wdesc);
1391         if (mtesize == 0) {
1392             mte_check_fail(env, *wdesc, toaddr, ra);
1393             *wdesc = 0;
1394         } else {
1395             copysize = MIN(copysize, mtesize);
1396         }
1397     }
1398 
1399     toaddr = useronly_clean_ptr(toaddr);
1400     fromaddr = useronly_clean_ptr(fromaddr);
1401     /* Trapless lookup of whether we can get a host memory pointer */
1402     wmem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, wmemidx);
1403     rmem = tlb_vaddr_to_host(env, fromaddr, MMU_DATA_LOAD, rmemidx);
1404 
1405 #ifndef CONFIG_USER_ONLY
1406     /*
1407      * If we don't have host memory for both source and dest then just
1408      * do a single byte copy. This will handle watchpoints, invalid pages,
1409      * etc correctly. For clean code pages, the next iteration will see
1410      * the page dirty and will use the fast path.
1411      */
1412     if (unlikely(!rmem || !wmem)) {
1413         uint8_t byte;
1414         if (rmem) {
1415             byte = *(uint8_t *)rmem;
1416         } else {
1417             byte = cpu_ldub_mmuidx_ra(env, fromaddr, rmemidx, ra);
1418         }
1419         if (wmem) {
1420             *(uint8_t *)wmem = byte;
1421         } else {
1422             cpu_stb_mmuidx_ra(env, toaddr, byte, wmemidx, ra);
1423         }
1424         return 1;
1425     }
1426 #endif
1427     /*
1428      * Easy case: just memmove the host memory. Note that wmem and
1429      * rmem here point to the *last* byte to copy.
1430      */
1431     set_helper_retaddr(ra);
1432     memmove(wmem - (copysize - 1), rmem - (copysize - 1), copysize);
1433     clear_helper_retaddr();
1434     return copysize;
1435 }
1436 
1437 /*
1438  * for the Memory Copy operation, our implementation chooses always
1439  * to use "option A", where we update Xd and Xs to the final addresses
1440  * in the CPYP insn, and then in CPYM and CPYE only need to update Xn.
1441  *
1442  * @env: CPU
1443  * @syndrome: syndrome value for mismatch exceptions
1444  * (also contains the register numbers we need to use)
1445  * @wdesc: MTE descriptor for the writes (destination)
1446  * @rdesc: MTE descriptor for the reads (source)
1447  * @move: true if this is CPY (memmove), false for CPYF (memcpy forwards)
1448  */
1449 static void do_cpyp(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1450                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1451 {
1452     int rd = mops_destreg(syndrome);
1453     int rs = mops_srcreg(syndrome);
1454     int rn = mops_sizereg(syndrome);
1455     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1456     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1457     bool forwards = true;
1458     uint64_t toaddr = env->xregs[rd];
1459     uint64_t fromaddr = env->xregs[rs];
1460     uint64_t copysize = env->xregs[rn];
1461     uint64_t stagecopysize, step;
1462 
1463     check_mops_enabled(env, ra);
1464 
1465 
1466     if (move) {
1467         /*
1468          * Copy backwards if necessary. The direction for a non-overlapping
1469          * copy is IMPDEF; we choose forwards.
1470          */
1471         if (copysize > 0x007FFFFFFFFFFFFFULL) {
1472             copysize = 0x007FFFFFFFFFFFFFULL;
1473         }
1474         uint64_t fs = extract64(fromaddr, 0, 56);
1475         uint64_t ts = extract64(toaddr, 0, 56);
1476         uint64_t fe = extract64(fromaddr + copysize, 0, 56);
1477 
1478         if (fs < ts && fe > ts) {
1479             forwards = false;
1480         }
1481     } else {
1482         if (copysize > INT64_MAX) {
1483             copysize = INT64_MAX;
1484         }
1485     }
1486 
1487     if (!mte_checks_needed(fromaddr, rdesc)) {
1488         rdesc = 0;
1489     }
1490     if (!mte_checks_needed(toaddr, wdesc)) {
1491         wdesc = 0;
1492     }
1493 
1494     if (forwards) {
1495         stagecopysize = MIN(copysize, page_limit(toaddr));
1496         stagecopysize = MIN(stagecopysize, page_limit(fromaddr));
1497         while (stagecopysize) {
1498             env->xregs[rd] = toaddr;
1499             env->xregs[rs] = fromaddr;
1500             env->xregs[rn] = copysize;
1501             step = copy_step(env, toaddr, fromaddr, stagecopysize,
1502                              wmemidx, rmemidx, &wdesc, &rdesc, ra);
1503             toaddr += step;
1504             fromaddr += step;
1505             copysize -= step;
1506             stagecopysize -= step;
1507         }
1508         /* Insn completed, so update registers to the Option A format */
1509         env->xregs[rd] = toaddr + copysize;
1510         env->xregs[rs] = fromaddr + copysize;
1511         env->xregs[rn] = -copysize;
1512     } else {
1513         /*
1514          * In a reverse copy the to and from addrs in Xs and Xd are the start
1515          * of the range, but it's more convenient for us to work with pointers
1516          * to the last byte being copied.
1517          */
1518         toaddr += copysize - 1;
1519         fromaddr += copysize - 1;
1520         stagecopysize = MIN(copysize, page_limit_rev(toaddr));
1521         stagecopysize = MIN(stagecopysize, page_limit_rev(fromaddr));
1522         while (stagecopysize) {
1523             env->xregs[rn] = copysize;
1524             step = copy_step_rev(env, toaddr, fromaddr, stagecopysize,
1525                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1526             copysize -= step;
1527             stagecopysize -= step;
1528             toaddr -= step;
1529             fromaddr -= step;
1530         }
1531         /*
1532          * Insn completed, so update registers to the Option A format.
1533          * For a reverse copy this is no different to the CPYP input format.
1534          */
1535         env->xregs[rn] = copysize;
1536     }
1537 
1538     /* Set NZCV = 0000 to indicate we are an Option A implementation */
1539     env->NF = 0;
1540     env->ZF = 1; /* our env->ZF encoding is inverted */
1541     env->CF = 0;
1542     env->VF = 0;
1543     return;
1544 }
1545 
1546 void HELPER(cpyp)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1547                   uint32_t rdesc)
1548 {
1549     do_cpyp(env, syndrome, wdesc, rdesc, true, GETPC());
1550 }
1551 
1552 void HELPER(cpyfp)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1553                    uint32_t rdesc)
1554 {
1555     do_cpyp(env, syndrome, wdesc, rdesc, false, GETPC());
1556 }
1557 
1558 static void do_cpym(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1559                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1560 {
1561     /* Main: we choose to copy until less than a page remaining */
1562     CPUState *cs = env_cpu(env);
1563     int rd = mops_destreg(syndrome);
1564     int rs = mops_srcreg(syndrome);
1565     int rn = mops_sizereg(syndrome);
1566     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1567     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1568     bool forwards = true;
1569     uint64_t toaddr, fromaddr, copysize, step;
1570 
1571     check_mops_enabled(env, ra);
1572 
1573     /* We choose to NOP out "no data to copy" before consistency checks */
1574     if (env->xregs[rn] == 0) {
1575         return;
1576     }
1577 
1578     check_mops_wrong_option(env, syndrome, ra);
1579 
1580     if (move) {
1581         forwards = (int64_t)env->xregs[rn] < 0;
1582     }
1583 
1584     if (forwards) {
1585         toaddr = env->xregs[rd] + env->xregs[rn];
1586         fromaddr = env->xregs[rs] + env->xregs[rn];
1587         copysize = -env->xregs[rn];
1588     } else {
1589         copysize = env->xregs[rn];
1590         /* This toaddr and fromaddr point to the *last* byte to copy */
1591         toaddr = env->xregs[rd] + copysize - 1;
1592         fromaddr = env->xregs[rs] + copysize - 1;
1593     }
1594 
1595     if (!mte_checks_needed(fromaddr, rdesc)) {
1596         rdesc = 0;
1597     }
1598     if (!mte_checks_needed(toaddr, wdesc)) {
1599         wdesc = 0;
1600     }
1601 
1602     /* Our implementation has no particular parameter requirements for CPYM */
1603 
1604     /* Do the actual memmove */
1605     if (forwards) {
1606         while (copysize >= TARGET_PAGE_SIZE) {
1607             step = copy_step(env, toaddr, fromaddr, copysize,
1608                              wmemidx, rmemidx, &wdesc, &rdesc, ra);
1609             toaddr += step;
1610             fromaddr += step;
1611             copysize -= step;
1612             env->xregs[rn] = -copysize;
1613             if (copysize >= TARGET_PAGE_SIZE &&
1614                 unlikely(cpu_loop_exit_requested(cs))) {
1615                 cpu_loop_exit_restore(cs, ra);
1616             }
1617         }
1618     } else {
1619         while (copysize >= TARGET_PAGE_SIZE) {
1620             step = copy_step_rev(env, toaddr, fromaddr, copysize,
1621                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1622             toaddr -= step;
1623             fromaddr -= step;
1624             copysize -= step;
1625             env->xregs[rn] = copysize;
1626             if (copysize >= TARGET_PAGE_SIZE &&
1627                 unlikely(cpu_loop_exit_requested(cs))) {
1628                 cpu_loop_exit_restore(cs, ra);
1629             }
1630         }
1631     }
1632 }
1633 
1634 void HELPER(cpym)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1635                   uint32_t rdesc)
1636 {
1637     do_cpym(env, syndrome, wdesc, rdesc, true, GETPC());
1638 }
1639 
1640 void HELPER(cpyfm)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1641                    uint32_t rdesc)
1642 {
1643     do_cpym(env, syndrome, wdesc, rdesc, false, GETPC());
1644 }
1645 
1646 static void do_cpye(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1647                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1648 {
1649     /* Epilogue: do the last partial page */
1650     int rd = mops_destreg(syndrome);
1651     int rs = mops_srcreg(syndrome);
1652     int rn = mops_sizereg(syndrome);
1653     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1654     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1655     bool forwards = true;
1656     uint64_t toaddr, fromaddr, copysize, step;
1657 
1658     check_mops_enabled(env, ra);
1659 
1660     /* We choose to NOP out "no data to copy" before consistency checks */
1661     if (env->xregs[rn] == 0) {
1662         return;
1663     }
1664 
1665     check_mops_wrong_option(env, syndrome, ra);
1666 
1667     if (move) {
1668         forwards = (int64_t)env->xregs[rn] < 0;
1669     }
1670 
1671     if (forwards) {
1672         toaddr = env->xregs[rd] + env->xregs[rn];
1673         fromaddr = env->xregs[rs] + env->xregs[rn];
1674         copysize = -env->xregs[rn];
1675     } else {
1676         copysize = env->xregs[rn];
1677         /* This toaddr and fromaddr point to the *last* byte to copy */
1678         toaddr = env->xregs[rd] + copysize - 1;
1679         fromaddr = env->xregs[rs] + copysize - 1;
1680     }
1681 
1682     if (!mte_checks_needed(fromaddr, rdesc)) {
1683         rdesc = 0;
1684     }
1685     if (!mte_checks_needed(toaddr, wdesc)) {
1686         wdesc = 0;
1687     }
1688 
1689     /* Check the size; we don't want to have do a check-for-interrupts */
1690     if (copysize >= TARGET_PAGE_SIZE) {
1691         raise_exception_ra(env, EXCP_UDEF, syndrome,
1692                            mops_mismatch_exception_target_el(env), ra);
1693     }
1694 
1695     /* Do the actual memmove */
1696     if (forwards) {
1697         while (copysize > 0) {
1698             step = copy_step(env, toaddr, fromaddr, copysize,
1699                              wmemidx, rmemidx, &wdesc, &rdesc, ra);
1700             toaddr += step;
1701             fromaddr += step;
1702             copysize -= step;
1703             env->xregs[rn] = -copysize;
1704         }
1705     } else {
1706         while (copysize > 0) {
1707             step = copy_step_rev(env, toaddr, fromaddr, copysize,
1708                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1709             toaddr -= step;
1710             fromaddr -= step;
1711             copysize -= step;
1712             env->xregs[rn] = copysize;
1713         }
1714     }
1715 }
1716 
1717 void HELPER(cpye)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1718                   uint32_t rdesc)
1719 {
1720     do_cpye(env, syndrome, wdesc, rdesc, true, GETPC());
1721 }
1722 
1723 void HELPER(cpyfe)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1724                    uint32_t rdesc)
1725 {
1726     do_cpye(env, syndrome, wdesc, rdesc, false, GETPC());
1727 }
1728 
1729 static bool is_guarded_page(CPUARMState *env, target_ulong addr, uintptr_t ra)
1730 {
1731 #ifdef CONFIG_USER_ONLY
1732     return page_get_flags(addr) & PAGE_BTI;
1733 #else
1734     CPUTLBEntryFull *full;
1735     void *host;
1736     int mmu_idx = cpu_mmu_index(env_cpu(env), true);
1737     int flags = probe_access_full(env, addr, 0, MMU_INST_FETCH, mmu_idx,
1738                                   false, &host, &full, ra);
1739 
1740     assert(!(flags & TLB_INVALID_MASK));
1741     return full->extra.arm.guarded;
1742 #endif
1743 }
1744 
1745 void HELPER(guarded_page_check)(CPUARMState *env)
1746 {
1747     /*
1748      * We have already verified that bti is enabled, and that the
1749      * instruction at PC is not ok for BTYPE.  This is always at
1750      * the beginning of a block, so PC is always up-to-date and
1751      * no unwind is required.
1752      */
1753     if (is_guarded_page(env, env->pc, 0)) {
1754         raise_exception(env, EXCP_UDEF, syn_btitrap(env->btype),
1755                         exception_target_el(env));
1756     }
1757 }
1758 
1759 void HELPER(guarded_page_br)(CPUARMState *env, target_ulong pc)
1760 {
1761     /*
1762      * We have already checked for branch via x16 and x17.
1763      * What remains for choosing BTYPE is checking for a guarded page.
1764      */
1765     env->btype = is_guarded_page(env, pc, GETPC()) ? 3 : 1;
1766 }
1767