xref: /qemu/target/arm/tcg/helper-a64.c (revision 68df8c8dba57f539d24f1a92a8699a179d9bb6fb)
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(a, b, float16_three, float_muladd_halve_result, 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(a, b, float32_three, float_muladd_halve_result, 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(a, b, float64_three, float_muladd_halve_result, 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 ADVSIMD_HALFOP(add)
443 ADVSIMD_HALFOP(sub)
444 ADVSIMD_HALFOP(mul)
445 ADVSIMD_HALFOP(div)
446 ADVSIMD_HALFOP(min)
447 ADVSIMD_HALFOP(max)
448 ADVSIMD_HALFOP(minnum)
449 ADVSIMD_HALFOP(maxnum)
450 
451 #define ADVSIMD_TWOHALFOP(name)                                         \
452 uint32_t ADVSIMD_HELPER(name, 2h)(uint32_t two_a, uint32_t two_b,       \
453                                   float_status *fpst)                   \
454 { \
455     float16  a1, a2, b1, b2;                        \
456     uint32_t r1, r2;                                \
457     a1 = extract32(two_a, 0, 16);                   \
458     a2 = extract32(two_a, 16, 16);                  \
459     b1 = extract32(two_b, 0, 16);                   \
460     b2 = extract32(two_b, 16, 16);                  \
461     r1 = float16_ ## name(a1, b1, fpst);            \
462     r2 = float16_ ## name(a2, b2, fpst);            \
463     return deposit32(r1, 16, 16, r2);               \
464 }
465 
466 ADVSIMD_TWOHALFOP(add)
467 ADVSIMD_TWOHALFOP(sub)
468 ADVSIMD_TWOHALFOP(mul)
469 ADVSIMD_TWOHALFOP(div)
470 ADVSIMD_TWOHALFOP(min)
471 ADVSIMD_TWOHALFOP(max)
472 ADVSIMD_TWOHALFOP(minnum)
473 ADVSIMD_TWOHALFOP(maxnum)
474 
475 /* Data processing - scalar floating-point and advanced SIMD */
476 static float16 float16_mulx(float16 a, float16 b, float_status *fpst)
477 {
478     a = float16_squash_input_denormal(a, fpst);
479     b = float16_squash_input_denormal(b, fpst);
480 
481     if ((float16_is_zero(a) && float16_is_infinity(b)) ||
482         (float16_is_infinity(a) && float16_is_zero(b))) {
483         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
484         return make_float16((1U << 14) |
485                             ((float16_val(a) ^ float16_val(b)) & (1U << 15)));
486     }
487     return float16_mul(a, b, fpst);
488 }
489 
490 ADVSIMD_HALFOP(mulx)
491 ADVSIMD_TWOHALFOP(mulx)
492 
493 /* fused multiply-accumulate */
494 uint32_t HELPER(advsimd_muladdh)(uint32_t a, uint32_t b, uint32_t c,
495                                  float_status *fpst)
496 {
497     return float16_muladd(a, b, c, 0, fpst);
498 }
499 
500 uint32_t HELPER(advsimd_muladd2h)(uint32_t two_a, uint32_t two_b,
501                                   uint32_t two_c, float_status *fpst)
502 {
503     float16  a1, a2, b1, b2, c1, c2;
504     uint32_t r1, r2;
505     a1 = extract32(two_a, 0, 16);
506     a2 = extract32(two_a, 16, 16);
507     b1 = extract32(two_b, 0, 16);
508     b2 = extract32(two_b, 16, 16);
509     c1 = extract32(two_c, 0, 16);
510     c2 = extract32(two_c, 16, 16);
511     r1 = float16_muladd(a1, b1, c1, 0, fpst);
512     r2 = float16_muladd(a2, b2, c2, 0, fpst);
513     return deposit32(r1, 16, 16, r2);
514 }
515 
516 /*
517  * Floating point comparisons produce an integer result. Softfloat
518  * routines return float_relation types which we convert to the 0/-1
519  * Neon requires.
520  */
521 
522 #define ADVSIMD_CMPRES(test) (test) ? 0xffff : 0
523 
524 uint32_t HELPER(advsimd_ceq_f16)(uint32_t a, uint32_t b, float_status *fpst)
525 {
526     int compare = float16_compare_quiet(a, b, fpst);
527     return ADVSIMD_CMPRES(compare == float_relation_equal);
528 }
529 
530 uint32_t HELPER(advsimd_cge_f16)(uint32_t a, uint32_t b, float_status *fpst)
531 {
532     int compare = float16_compare(a, b, fpst);
533     return ADVSIMD_CMPRES(compare == float_relation_greater ||
534                           compare == float_relation_equal);
535 }
536 
537 uint32_t HELPER(advsimd_cgt_f16)(uint32_t a, uint32_t b, float_status *fpst)
538 {
539     int compare = float16_compare(a, b, fpst);
540     return ADVSIMD_CMPRES(compare == float_relation_greater);
541 }
542 
543 uint32_t HELPER(advsimd_acge_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                           compare == float_relation_equal);
550 }
551 
552 uint32_t HELPER(advsimd_acgt_f16)(uint32_t a, uint32_t b, float_status *fpst)
553 {
554     float16 f0 = float16_abs(a);
555     float16 f1 = float16_abs(b);
556     int compare = float16_compare(f0, f1, fpst);
557     return ADVSIMD_CMPRES(compare == float_relation_greater);
558 }
559 
560 /* round to integral */
561 uint32_t HELPER(advsimd_rinth_exact)(uint32_t x, float_status *fp_status)
562 {
563     return float16_round_to_int(x, fp_status);
564 }
565 
566 uint32_t HELPER(advsimd_rinth)(uint32_t x, float_status *fp_status)
567 {
568     int old_flags = get_float_exception_flags(fp_status), new_flags;
569     float16 ret;
570 
571     ret = float16_round_to_int(x, fp_status);
572 
573     /* Suppress any inexact exceptions the conversion produced */
574     if (!(old_flags & float_flag_inexact)) {
575         new_flags = get_float_exception_flags(fp_status);
576         set_float_exception_flags(new_flags & ~float_flag_inexact, fp_status);
577     }
578 
579     return ret;
580 }
581 
582 static int el_from_spsr(uint32_t spsr)
583 {
584     /* Return the exception level that this SPSR is requesting a return to,
585      * or -1 if it is invalid (an illegal return)
586      */
587     if (spsr & PSTATE_nRW) {
588         switch (spsr & CPSR_M) {
589         case ARM_CPU_MODE_USR:
590             return 0;
591         case ARM_CPU_MODE_HYP:
592             return 2;
593         case ARM_CPU_MODE_FIQ:
594         case ARM_CPU_MODE_IRQ:
595         case ARM_CPU_MODE_SVC:
596         case ARM_CPU_MODE_ABT:
597         case ARM_CPU_MODE_UND:
598         case ARM_CPU_MODE_SYS:
599             return 1;
600         case ARM_CPU_MODE_MON:
601             /* Returning to Mon from AArch64 is never possible,
602              * so this is an illegal return.
603              */
604         default:
605             return -1;
606         }
607     } else {
608         if (extract32(spsr, 1, 1)) {
609             /* Return with reserved M[1] bit set */
610             return -1;
611         }
612         if (extract32(spsr, 0, 4) == 1) {
613             /* return to EL0 with M[0] bit set */
614             return -1;
615         }
616         return extract32(spsr, 2, 2);
617     }
618 }
619 
620 static void cpsr_write_from_spsr_elx(CPUARMState *env,
621                                      uint32_t val)
622 {
623     uint32_t mask;
624 
625     /* Save SPSR_ELx.SS into PSTATE. */
626     env->pstate = (env->pstate & ~PSTATE_SS) | (val & PSTATE_SS);
627     val &= ~PSTATE_SS;
628 
629     /* Move DIT to the correct location for CPSR */
630     if (val & PSTATE_DIT) {
631         val &= ~PSTATE_DIT;
632         val |= CPSR_DIT;
633     }
634 
635     mask = aarch32_cpsr_valid_mask(env->features, \
636         &env_archcpu(env)->isar);
637     cpsr_write(env, val, mask, CPSRWriteRaw);
638 }
639 
640 void HELPER(exception_return)(CPUARMState *env, uint64_t new_pc)
641 {
642     int cur_el = arm_current_el(env);
643     unsigned int spsr_idx = aarch64_banked_spsr_index(cur_el);
644     uint32_t spsr = env->banked_spsr[spsr_idx];
645     int new_el;
646     bool return_to_aa64 = (spsr & PSTATE_nRW) == 0;
647 
648     aarch64_save_sp(env, cur_el);
649 
650     arm_clear_exclusive(env);
651 
652     /* We must squash the PSTATE.SS bit to zero unless both of the
653      * following hold:
654      *  1. debug exceptions are currently disabled
655      *  2. singlestep will be active in the EL we return to
656      * We check 1 here and 2 after we've done the pstate/cpsr write() to
657      * transition to the EL we're going to.
658      */
659     if (arm_generate_debug_exceptions(env)) {
660         spsr &= ~PSTATE_SS;
661     }
662 
663     /*
664      * FEAT_RME forbids return from EL3 with an invalid security state.
665      * We don't need an explicit check for FEAT_RME here because we enforce
666      * in scr_write() that you can't set the NSE bit without it.
667      */
668     if (cur_el == 3 && (env->cp15.scr_el3 & (SCR_NS | SCR_NSE)) == SCR_NSE) {
669         goto illegal_return;
670     }
671 
672     new_el = el_from_spsr(spsr);
673     if (new_el == -1) {
674         goto illegal_return;
675     }
676     if (new_el > cur_el || (new_el == 2 && !arm_is_el2_enabled(env))) {
677         /* Disallow return to an EL which is unimplemented or higher
678          * than the current one.
679          */
680         goto illegal_return;
681     }
682 
683     if (new_el != 0 && arm_el_is_aa64(env, new_el) != return_to_aa64) {
684         /* Return to an EL which is configured for a different register width */
685         goto illegal_return;
686     }
687 
688     if (new_el == 1 && (arm_hcr_el2_eff(env) & HCR_TGE)) {
689         goto illegal_return;
690     }
691 
692     bql_lock();
693     arm_call_pre_el_change_hook(env_archcpu(env));
694     bql_unlock();
695 
696     if (!return_to_aa64) {
697         env->aarch64 = false;
698         /* We do a raw CPSR write because aarch64_sync_64_to_32()
699          * will sort the register banks out for us, and we've already
700          * caught all the bad-mode cases in el_from_spsr().
701          */
702         cpsr_write_from_spsr_elx(env, spsr);
703         if (!arm_singlestep_active(env)) {
704             env->pstate &= ~PSTATE_SS;
705         }
706         aarch64_sync_64_to_32(env);
707 
708         if (spsr & CPSR_T) {
709             env->regs[15] = new_pc & ~0x1;
710         } else {
711             env->regs[15] = new_pc & ~0x3;
712         }
713         helper_rebuild_hflags_a32(env, new_el);
714         qemu_log_mask(CPU_LOG_INT, "Exception return from AArch64 EL%d to "
715                       "AArch32 EL%d PC 0x%" PRIx32 "\n",
716                       cur_el, new_el, env->regs[15]);
717     } else {
718         int tbii;
719 
720         env->aarch64 = true;
721         spsr &= aarch64_pstate_valid_mask(&env_archcpu(env)->isar);
722         pstate_write(env, spsr);
723         if (!arm_singlestep_active(env)) {
724             env->pstate &= ~PSTATE_SS;
725         }
726         aarch64_restore_sp(env, new_el);
727         helper_rebuild_hflags_a64(env, new_el);
728 
729         /*
730          * Apply TBI to the exception return address.  We had to delay this
731          * until after we selected the new EL, so that we could select the
732          * correct TBI+TBID bits.  This is made easier by waiting until after
733          * the hflags rebuild, since we can pull the composite TBII field
734          * from there.
735          */
736         tbii = EX_TBFLAG_A64(env->hflags, TBII);
737         if ((tbii >> extract64(new_pc, 55, 1)) & 1) {
738             /* TBI is enabled. */
739             int core_mmu_idx = arm_env_mmu_index(env);
740             if (regime_has_2_ranges(core_to_aa64_mmu_idx(core_mmu_idx))) {
741                 new_pc = sextract64(new_pc, 0, 56);
742             } else {
743                 new_pc = extract64(new_pc, 0, 56);
744             }
745         }
746         env->pc = new_pc;
747 
748         qemu_log_mask(CPU_LOG_INT, "Exception return from AArch64 EL%d to "
749                       "AArch64 EL%d PC 0x%" PRIx64 "\n",
750                       cur_el, new_el, env->pc);
751     }
752 
753     /*
754      * Note that cur_el can never be 0.  If new_el is 0, then
755      * el0_a64 is return_to_aa64, else el0_a64 is ignored.
756      */
757     aarch64_sve_change_el(env, cur_el, new_el, return_to_aa64);
758 
759     bql_lock();
760     arm_call_el_change_hook(env_archcpu(env));
761     bql_unlock();
762 
763     return;
764 
765 illegal_return:
766     /* Illegal return events of various kinds have architecturally
767      * mandated behaviour:
768      * restore NZCV and DAIF from SPSR_ELx
769      * set PSTATE.IL
770      * restore PC from ELR_ELx
771      * no change to exception level, execution state or stack pointer
772      */
773     env->pstate |= PSTATE_IL;
774     env->pc = new_pc;
775     spsr &= PSTATE_NZCV | PSTATE_DAIF | PSTATE_ALLINT;
776     spsr |= pstate_read(env) & ~(PSTATE_NZCV | PSTATE_DAIF | PSTATE_ALLINT);
777     pstate_write(env, spsr);
778     if (!arm_singlestep_active(env)) {
779         env->pstate &= ~PSTATE_SS;
780     }
781     helper_rebuild_hflags_a64(env, cur_el);
782     qemu_log_mask(LOG_GUEST_ERROR, "Illegal exception return at EL%d: "
783                   "resuming execution at 0x%" PRIx64 "\n", cur_el, env->pc);
784 }
785 
786 void HELPER(dc_zva)(CPUARMState *env, uint64_t vaddr_in)
787 {
788     uintptr_t ra = GETPC();
789 
790     /*
791      * Implement DC ZVA, which zeroes a fixed-length block of memory.
792      * Note that we do not implement the (architecturally mandated)
793      * alignment fault for attempts to use this on Device memory
794      * (which matches the usual QEMU behaviour of not implementing either
795      * alignment faults or any memory attribute handling).
796      */
797     int blocklen = 4 << env_archcpu(env)->dcz_blocksize;
798     uint64_t vaddr = vaddr_in & ~(blocklen - 1);
799     int mmu_idx = arm_env_mmu_index(env);
800     void *mem;
801 
802     /*
803      * Trapless lookup.  In addition to actual invalid page, may
804      * return NULL for I/O, watchpoints, clean pages, etc.
805      */
806     mem = tlb_vaddr_to_host(env, vaddr, MMU_DATA_STORE, mmu_idx);
807 
808 #ifndef CONFIG_USER_ONLY
809     if (unlikely(!mem)) {
810         /*
811          * Trap if accessing an invalid page.  DC_ZVA requires that we supply
812          * the original pointer for an invalid page.  But watchpoints require
813          * that we probe the actual space.  So do both.
814          */
815         (void) probe_write(env, vaddr_in, 1, mmu_idx, ra);
816         mem = probe_write(env, vaddr, blocklen, mmu_idx, ra);
817 
818         if (unlikely(!mem)) {
819             /*
820              * The only remaining reason for mem == NULL is I/O.
821              * Just do a series of byte writes as the architecture demands.
822              */
823             for (int i = 0; i < blocklen; i++) {
824                 cpu_stb_mmuidx_ra(env, vaddr + i, 0, mmu_idx, ra);
825             }
826             return;
827         }
828     }
829 #endif
830 
831     set_helper_retaddr(ra);
832     memset(mem, 0, blocklen);
833     clear_helper_retaddr();
834 }
835 
836 void HELPER(unaligned_access)(CPUARMState *env, uint64_t addr,
837                               uint32_t access_type, uint32_t mmu_idx)
838 {
839     arm_cpu_do_unaligned_access(env_cpu(env), addr, access_type,
840                                 mmu_idx, GETPC());
841 }
842 
843 /* Memory operations (memset, memmove, memcpy) */
844 
845 /*
846  * Return true if the CPY* and SET* insns can execute; compare
847  * pseudocode CheckMOPSEnabled(), though we refactor it a little.
848  */
849 static bool mops_enabled(CPUARMState *env)
850 {
851     int el = arm_current_el(env);
852 
853     if (el < 2 &&
854         (arm_hcr_el2_eff(env) & (HCR_E2H | HCR_TGE)) != (HCR_E2H | HCR_TGE) &&
855         !(arm_hcrx_el2_eff(env) & HCRX_MSCEN)) {
856         return false;
857     }
858 
859     if (el == 0) {
860         if (!el_is_in_host(env, 0)) {
861             return env->cp15.sctlr_el[1] & SCTLR_MSCEN;
862         } else {
863             return env->cp15.sctlr_el[2] & SCTLR_MSCEN;
864         }
865     }
866     return true;
867 }
868 
869 static void check_mops_enabled(CPUARMState *env, uintptr_t ra)
870 {
871     if (!mops_enabled(env)) {
872         raise_exception_ra(env, EXCP_UDEF, syn_uncategorized(),
873                            exception_target_el(env), ra);
874     }
875 }
876 
877 /*
878  * Return the target exception level for an exception due
879  * to mismatched arguments in a FEAT_MOPS copy or set.
880  * Compare pseudocode MismatchedCpySetTargetEL()
881  */
882 static int mops_mismatch_exception_target_el(CPUARMState *env)
883 {
884     int el = arm_current_el(env);
885 
886     if (el > 1) {
887         return el;
888     }
889     if (el == 0 && (arm_hcr_el2_eff(env) & HCR_TGE)) {
890         return 2;
891     }
892     if (el == 1 && (arm_hcrx_el2_eff(env) & HCRX_MCE2)) {
893         return 2;
894     }
895     return 1;
896 }
897 
898 /*
899  * Check whether an M or E instruction was executed with a CF value
900  * indicating the wrong option for this implementation.
901  * Assumes we are always Option A.
902  */
903 static void check_mops_wrong_option(CPUARMState *env, uint32_t syndrome,
904                                     uintptr_t ra)
905 {
906     if (env->CF != 0) {
907         syndrome |= 1 << 17; /* Set the wrong-option bit */
908         raise_exception_ra(env, EXCP_UDEF, syndrome,
909                            mops_mismatch_exception_target_el(env), ra);
910     }
911 }
912 
913 /*
914  * Return the maximum number of bytes we can transfer starting at addr
915  * without crossing a page boundary.
916  */
917 static uint64_t page_limit(uint64_t addr)
918 {
919     return TARGET_PAGE_ALIGN(addr + 1) - addr;
920 }
921 
922 /*
923  * Return the number of bytes we can copy starting from addr and working
924  * backwards without crossing a page boundary.
925  */
926 static uint64_t page_limit_rev(uint64_t addr)
927 {
928     return (addr & ~TARGET_PAGE_MASK) + 1;
929 }
930 
931 /*
932  * Perform part of a memory set on an area of guest memory starting at
933  * toaddr (a dirty address) and extending for setsize bytes.
934  *
935  * Returns the number of bytes actually set, which might be less than
936  * setsize; the caller should loop until the whole set has been done.
937  * The caller should ensure that the guest registers are correct
938  * for the possibility that the first byte of the set encounters
939  * an exception or watchpoint. We guarantee not to take any faults
940  * for bytes other than the first.
941  */
942 static uint64_t set_step(CPUARMState *env, uint64_t toaddr,
943                          uint64_t setsize, uint32_t data, int memidx,
944                          uint32_t *mtedesc, uintptr_t ra)
945 {
946     void *mem;
947 
948     setsize = MIN(setsize, page_limit(toaddr));
949     if (*mtedesc) {
950         uint64_t mtesize = mte_mops_probe(env, toaddr, setsize, *mtedesc);
951         if (mtesize == 0) {
952             /* Trap, or not. All CPU state is up to date */
953             mte_check_fail(env, *mtedesc, toaddr, ra);
954             /* Continue, with no further MTE checks required */
955             *mtedesc = 0;
956         } else {
957             /* Advance to the end, or to the tag mismatch */
958             setsize = MIN(setsize, mtesize);
959         }
960     }
961 
962     toaddr = useronly_clean_ptr(toaddr);
963     /*
964      * Trapless lookup: returns NULL for invalid page, I/O,
965      * watchpoints, clean pages, etc.
966      */
967     mem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, memidx);
968 
969 #ifndef CONFIG_USER_ONLY
970     if (unlikely(!mem)) {
971         /*
972          * Slow-path: just do one byte write. This will handle the
973          * watchpoint, invalid page, etc handling correctly.
974          * For clean code pages, the next iteration will see
975          * the page dirty and will use the fast path.
976          */
977         cpu_stb_mmuidx_ra(env, toaddr, data, memidx, ra);
978         return 1;
979     }
980 #endif
981     /* Easy case: just memset the host memory */
982     set_helper_retaddr(ra);
983     memset(mem, data, setsize);
984     clear_helper_retaddr();
985     return setsize;
986 }
987 
988 /*
989  * Similar, but setting tags. The architecture requires us to do this
990  * in 16-byte chunks. SETP accesses are not tag checked; they set
991  * the tags.
992  */
993 static uint64_t set_step_tags(CPUARMState *env, uint64_t toaddr,
994                               uint64_t setsize, uint32_t data, int memidx,
995                               uint32_t *mtedesc, uintptr_t ra)
996 {
997     void *mem;
998     uint64_t cleanaddr;
999 
1000     setsize = MIN(setsize, page_limit(toaddr));
1001 
1002     cleanaddr = useronly_clean_ptr(toaddr);
1003     /*
1004      * Trapless lookup: returns NULL for invalid page, I/O,
1005      * watchpoints, clean pages, etc.
1006      */
1007     mem = tlb_vaddr_to_host(env, cleanaddr, MMU_DATA_STORE, memidx);
1008 
1009 #ifndef CONFIG_USER_ONLY
1010     if (unlikely(!mem)) {
1011         /*
1012          * Slow-path: just do one write. This will handle the
1013          * watchpoint, invalid page, etc handling correctly.
1014          * The architecture requires that we do 16 bytes at a time,
1015          * and we know both ptr and size are 16 byte aligned.
1016          * For clean code pages, the next iteration will see
1017          * the page dirty and will use the fast path.
1018          */
1019         uint64_t repldata = data * 0x0101010101010101ULL;
1020         MemOpIdx oi16 = make_memop_idx(MO_TE | MO_128, memidx);
1021         cpu_st16_mmu(env, toaddr, int128_make128(repldata, repldata), oi16, ra);
1022         mte_mops_set_tags(env, toaddr, 16, *mtedesc);
1023         return 16;
1024     }
1025 #endif
1026     /* Easy case: just memset the host memory */
1027     set_helper_retaddr(ra);
1028     memset(mem, data, setsize);
1029     clear_helper_retaddr();
1030     mte_mops_set_tags(env, toaddr, setsize, *mtedesc);
1031     return setsize;
1032 }
1033 
1034 typedef uint64_t StepFn(CPUARMState *env, uint64_t toaddr,
1035                         uint64_t setsize, uint32_t data,
1036                         int memidx, uint32_t *mtedesc, uintptr_t ra);
1037 
1038 /* Extract register numbers from a MOPS exception syndrome value */
1039 static int mops_destreg(uint32_t syndrome)
1040 {
1041     return extract32(syndrome, 10, 5);
1042 }
1043 
1044 static int mops_srcreg(uint32_t syndrome)
1045 {
1046     return extract32(syndrome, 5, 5);
1047 }
1048 
1049 static int mops_sizereg(uint32_t syndrome)
1050 {
1051     return extract32(syndrome, 0, 5);
1052 }
1053 
1054 /*
1055  * Return true if TCMA and TBI bits mean we need to do MTE checks.
1056  * We only need to do this once per MOPS insn, not for every page.
1057  */
1058 static bool mte_checks_needed(uint64_t ptr, uint32_t desc)
1059 {
1060     int bit55 = extract64(ptr, 55, 1);
1061 
1062     /*
1063      * Note that tbi_check() returns true for "access checked" but
1064      * tcma_check() returns true for "access unchecked".
1065      */
1066     if (!tbi_check(desc, bit55)) {
1067         return false;
1068     }
1069     return !tcma_check(desc, bit55, allocation_tag_from_addr(ptr));
1070 }
1071 
1072 /* Take an exception if the SETG addr/size are not granule aligned */
1073 static void check_setg_alignment(CPUARMState *env, uint64_t ptr, uint64_t size,
1074                                  uint32_t memidx, uintptr_t ra)
1075 {
1076     if ((size != 0 && !QEMU_IS_ALIGNED(ptr, TAG_GRANULE)) ||
1077         !QEMU_IS_ALIGNED(size, TAG_GRANULE)) {
1078         arm_cpu_do_unaligned_access(env_cpu(env), ptr, MMU_DATA_STORE,
1079                                     memidx, ra);
1080 
1081     }
1082 }
1083 
1084 static uint64_t arm_reg_or_xzr(CPUARMState *env, int reg)
1085 {
1086     /*
1087      * Runtime equivalent of cpu_reg() -- return the CPU register value,
1088      * for contexts when index 31 means XZR (not SP).
1089      */
1090     return reg == 31 ? 0 : env->xregs[reg];
1091 }
1092 
1093 /*
1094  * For the Memory Set operation, our implementation chooses
1095  * always to use "option A", where we update Xd to the final
1096  * address in the SETP insn, and set Xn to be -(bytes remaining).
1097  * On SETM and SETE insns we only need update Xn.
1098  *
1099  * @env: CPU
1100  * @syndrome: syndrome value for mismatch exceptions
1101  * (also contains the register numbers we need to use)
1102  * @mtedesc: MTE descriptor word
1103  * @stepfn: function which does a single part of the set operation
1104  * @is_setg: true if this is the tag-setting SETG variant
1105  */
1106 static void do_setp(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1107                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1108 {
1109     /* Prologue: we choose to do up to the next page boundary */
1110     int rd = mops_destreg(syndrome);
1111     int rs = mops_srcreg(syndrome);
1112     int rn = mops_sizereg(syndrome);
1113     uint8_t data = arm_reg_or_xzr(env, rs);
1114     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1115     uint64_t toaddr = env->xregs[rd];
1116     uint64_t setsize = env->xregs[rn];
1117     uint64_t stagesetsize, step;
1118 
1119     check_mops_enabled(env, ra);
1120 
1121     if (setsize > INT64_MAX) {
1122         setsize = INT64_MAX;
1123         if (is_setg) {
1124             setsize &= ~0xf;
1125         }
1126     }
1127 
1128     if (unlikely(is_setg)) {
1129         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1130     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1131         mtedesc = 0;
1132     }
1133 
1134     stagesetsize = MIN(setsize, page_limit(toaddr));
1135     while (stagesetsize) {
1136         env->xregs[rd] = toaddr;
1137         env->xregs[rn] = setsize;
1138         step = stepfn(env, toaddr, stagesetsize, data, memidx, &mtedesc, ra);
1139         toaddr += step;
1140         setsize -= step;
1141         stagesetsize -= step;
1142     }
1143     /* Insn completed, so update registers to the Option A format */
1144     env->xregs[rd] = toaddr + setsize;
1145     env->xregs[rn] = -setsize;
1146 
1147     /* Set NZCV = 0000 to indicate we are an Option A implementation */
1148     env->NF = 0;
1149     env->ZF = 1; /* our env->ZF encoding is inverted */
1150     env->CF = 0;
1151     env->VF = 0;
1152     return;
1153 }
1154 
1155 void HELPER(setp)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1156 {
1157     do_setp(env, syndrome, mtedesc, set_step, false, GETPC());
1158 }
1159 
1160 void HELPER(setgp)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1161 {
1162     do_setp(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1163 }
1164 
1165 static void do_setm(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1166                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1167 {
1168     /* Main: we choose to do all the full-page chunks */
1169     CPUState *cs = env_cpu(env);
1170     int rd = mops_destreg(syndrome);
1171     int rs = mops_srcreg(syndrome);
1172     int rn = mops_sizereg(syndrome);
1173     uint8_t data = arm_reg_or_xzr(env, rs);
1174     uint64_t toaddr = env->xregs[rd] + env->xregs[rn];
1175     uint64_t setsize = -env->xregs[rn];
1176     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1177     uint64_t step, stagesetsize;
1178 
1179     check_mops_enabled(env, ra);
1180 
1181     /*
1182      * We're allowed to NOP out "no data to copy" before the consistency
1183      * checks; we choose to do so.
1184      */
1185     if (env->xregs[rn] == 0) {
1186         return;
1187     }
1188 
1189     check_mops_wrong_option(env, syndrome, ra);
1190 
1191     /*
1192      * Our implementation will work fine even if we have an unaligned
1193      * destination address, and because we update Xn every time around
1194      * the loop below and the return value from stepfn() may be less
1195      * than requested, we might find toaddr is unaligned. So we don't
1196      * have an IMPDEF check for alignment here.
1197      */
1198 
1199     if (unlikely(is_setg)) {
1200         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1201     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1202         mtedesc = 0;
1203     }
1204 
1205     /* Do the actual memset: we leave the last partial page to SETE */
1206     stagesetsize = setsize & TARGET_PAGE_MASK;
1207     while (stagesetsize > 0) {
1208         step = stepfn(env, toaddr, stagesetsize, data, memidx, &mtedesc, ra);
1209         toaddr += step;
1210         setsize -= step;
1211         stagesetsize -= step;
1212         env->xregs[rn] = -setsize;
1213         if (stagesetsize > 0 && unlikely(cpu_loop_exit_requested(cs))) {
1214             cpu_loop_exit_restore(cs, ra);
1215         }
1216     }
1217 }
1218 
1219 void HELPER(setm)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1220 {
1221     do_setm(env, syndrome, mtedesc, set_step, false, GETPC());
1222 }
1223 
1224 void HELPER(setgm)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1225 {
1226     do_setm(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1227 }
1228 
1229 static void do_sete(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1230                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1231 {
1232     /* Epilogue: do the last partial page */
1233     int rd = mops_destreg(syndrome);
1234     int rs = mops_srcreg(syndrome);
1235     int rn = mops_sizereg(syndrome);
1236     uint8_t data = arm_reg_or_xzr(env, rs);
1237     uint64_t toaddr = env->xregs[rd] + env->xregs[rn];
1238     uint64_t setsize = -env->xregs[rn];
1239     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1240     uint64_t step;
1241 
1242     check_mops_enabled(env, ra);
1243 
1244     /*
1245      * We're allowed to NOP out "no data to copy" before the consistency
1246      * checks; we choose to do so.
1247      */
1248     if (setsize == 0) {
1249         return;
1250     }
1251 
1252     check_mops_wrong_option(env, syndrome, ra);
1253 
1254     /*
1255      * Our implementation has no address alignment requirements, but
1256      * we do want to enforce the "less than a page" size requirement,
1257      * so we don't need to have the "check for interrupts" here.
1258      */
1259     if (setsize >= TARGET_PAGE_SIZE) {
1260         raise_exception_ra(env, EXCP_UDEF, syndrome,
1261                            mops_mismatch_exception_target_el(env), ra);
1262     }
1263 
1264     if (unlikely(is_setg)) {
1265         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1266     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1267         mtedesc = 0;
1268     }
1269 
1270     /* Do the actual memset */
1271     while (setsize > 0) {
1272         step = stepfn(env, toaddr, setsize, data, memidx, &mtedesc, ra);
1273         toaddr += step;
1274         setsize -= step;
1275         env->xregs[rn] = -setsize;
1276     }
1277 }
1278 
1279 void HELPER(sete)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1280 {
1281     do_sete(env, syndrome, mtedesc, set_step, false, GETPC());
1282 }
1283 
1284 void HELPER(setge)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1285 {
1286     do_sete(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1287 }
1288 
1289 /*
1290  * Perform part of a memory copy from the guest memory at fromaddr
1291  * and extending for copysize bytes, to the guest memory at
1292  * toaddr. Both addresses are dirty.
1293  *
1294  * Returns the number of bytes actually set, which might be less than
1295  * copysize; the caller should loop until the whole copy has been done.
1296  * The caller should ensure that the guest registers are correct
1297  * for the possibility that the first byte of the copy encounters
1298  * an exception or watchpoint. We guarantee not to take any faults
1299  * for bytes other than the first.
1300  */
1301 static uint64_t copy_step(CPUARMState *env, uint64_t toaddr, uint64_t fromaddr,
1302                           uint64_t copysize, int wmemidx, int rmemidx,
1303                           uint32_t *wdesc, uint32_t *rdesc, uintptr_t ra)
1304 {
1305     void *rmem;
1306     void *wmem;
1307 
1308     /* Don't cross a page boundary on either source or destination */
1309     copysize = MIN(copysize, page_limit(toaddr));
1310     copysize = MIN(copysize, page_limit(fromaddr));
1311     /*
1312      * Handle MTE tag checks: either handle the tag mismatch for byte 0,
1313      * or else copy up to but not including the byte with the mismatch.
1314      */
1315     if (*rdesc) {
1316         uint64_t mtesize = mte_mops_probe(env, fromaddr, copysize, *rdesc);
1317         if (mtesize == 0) {
1318             mte_check_fail(env, *rdesc, fromaddr, ra);
1319             *rdesc = 0;
1320         } else {
1321             copysize = MIN(copysize, mtesize);
1322         }
1323     }
1324     if (*wdesc) {
1325         uint64_t mtesize = mte_mops_probe(env, toaddr, copysize, *wdesc);
1326         if (mtesize == 0) {
1327             mte_check_fail(env, *wdesc, toaddr, ra);
1328             *wdesc = 0;
1329         } else {
1330             copysize = MIN(copysize, mtesize);
1331         }
1332     }
1333 
1334     toaddr = useronly_clean_ptr(toaddr);
1335     fromaddr = useronly_clean_ptr(fromaddr);
1336     /* Trapless lookup of whether we can get a host memory pointer */
1337     wmem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, wmemidx);
1338     rmem = tlb_vaddr_to_host(env, fromaddr, MMU_DATA_LOAD, rmemidx);
1339 
1340 #ifndef CONFIG_USER_ONLY
1341     /*
1342      * If we don't have host memory for both source and dest then just
1343      * do a single byte copy. This will handle watchpoints, invalid pages,
1344      * etc correctly. For clean code pages, the next iteration will see
1345      * the page dirty and will use the fast path.
1346      */
1347     if (unlikely(!rmem || !wmem)) {
1348         uint8_t byte;
1349         if (rmem) {
1350             byte = *(uint8_t *)rmem;
1351         } else {
1352             byte = cpu_ldub_mmuidx_ra(env, fromaddr, rmemidx, ra);
1353         }
1354         if (wmem) {
1355             *(uint8_t *)wmem = byte;
1356         } else {
1357             cpu_stb_mmuidx_ra(env, toaddr, byte, wmemidx, ra);
1358         }
1359         return 1;
1360     }
1361 #endif
1362     /* Easy case: just memmove the host memory */
1363     set_helper_retaddr(ra);
1364     memmove(wmem, rmem, copysize);
1365     clear_helper_retaddr();
1366     return copysize;
1367 }
1368 
1369 /*
1370  * Do part of a backwards memory copy. Here toaddr and fromaddr point
1371  * to the *last* byte to be copied.
1372  */
1373 static uint64_t copy_step_rev(CPUARMState *env, uint64_t toaddr,
1374                               uint64_t fromaddr,
1375                               uint64_t copysize, int wmemidx, int rmemidx,
1376                               uint32_t *wdesc, uint32_t *rdesc, uintptr_t ra)
1377 {
1378     void *rmem;
1379     void *wmem;
1380 
1381     /* Don't cross a page boundary on either source or destination */
1382     copysize = MIN(copysize, page_limit_rev(toaddr));
1383     copysize = MIN(copysize, page_limit_rev(fromaddr));
1384 
1385     /*
1386      * Handle MTE tag checks: either handle the tag mismatch for byte 0,
1387      * or else copy up to but not including the byte with the mismatch.
1388      */
1389     if (*rdesc) {
1390         uint64_t mtesize = mte_mops_probe_rev(env, fromaddr, copysize, *rdesc);
1391         if (mtesize == 0) {
1392             mte_check_fail(env, *rdesc, fromaddr, ra);
1393             *rdesc = 0;
1394         } else {
1395             copysize = MIN(copysize, mtesize);
1396         }
1397     }
1398     if (*wdesc) {
1399         uint64_t mtesize = mte_mops_probe_rev(env, toaddr, copysize, *wdesc);
1400         if (mtesize == 0) {
1401             mte_check_fail(env, *wdesc, toaddr, ra);
1402             *wdesc = 0;
1403         } else {
1404             copysize = MIN(copysize, mtesize);
1405         }
1406     }
1407 
1408     toaddr = useronly_clean_ptr(toaddr);
1409     fromaddr = useronly_clean_ptr(fromaddr);
1410     /* Trapless lookup of whether we can get a host memory pointer */
1411     wmem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, wmemidx);
1412     rmem = tlb_vaddr_to_host(env, fromaddr, MMU_DATA_LOAD, rmemidx);
1413 
1414 #ifndef CONFIG_USER_ONLY
1415     /*
1416      * If we don't have host memory for both source and dest then just
1417      * do a single byte copy. This will handle watchpoints, invalid pages,
1418      * etc correctly. For clean code pages, the next iteration will see
1419      * the page dirty and will use the fast path.
1420      */
1421     if (unlikely(!rmem || !wmem)) {
1422         uint8_t byte;
1423         if (rmem) {
1424             byte = *(uint8_t *)rmem;
1425         } else {
1426             byte = cpu_ldub_mmuidx_ra(env, fromaddr, rmemidx, ra);
1427         }
1428         if (wmem) {
1429             *(uint8_t *)wmem = byte;
1430         } else {
1431             cpu_stb_mmuidx_ra(env, toaddr, byte, wmemidx, ra);
1432         }
1433         return 1;
1434     }
1435 #endif
1436     /*
1437      * Easy case: just memmove the host memory. Note that wmem and
1438      * rmem here point to the *last* byte to copy.
1439      */
1440     set_helper_retaddr(ra);
1441     memmove(wmem - (copysize - 1), rmem - (copysize - 1), copysize);
1442     clear_helper_retaddr();
1443     return copysize;
1444 }
1445 
1446 /*
1447  * for the Memory Copy operation, our implementation chooses always
1448  * to use "option A", where we update Xd and Xs to the final addresses
1449  * in the CPYP insn, and then in CPYM and CPYE only need to update Xn.
1450  *
1451  * @env: CPU
1452  * @syndrome: syndrome value for mismatch exceptions
1453  * (also contains the register numbers we need to use)
1454  * @wdesc: MTE descriptor for the writes (destination)
1455  * @rdesc: MTE descriptor for the reads (source)
1456  * @move: true if this is CPY (memmove), false for CPYF (memcpy forwards)
1457  */
1458 static void do_cpyp(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1459                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1460 {
1461     int rd = mops_destreg(syndrome);
1462     int rs = mops_srcreg(syndrome);
1463     int rn = mops_sizereg(syndrome);
1464     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1465     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1466     bool forwards = true;
1467     uint64_t toaddr = env->xregs[rd];
1468     uint64_t fromaddr = env->xregs[rs];
1469     uint64_t copysize = env->xregs[rn];
1470     uint64_t stagecopysize, step;
1471 
1472     check_mops_enabled(env, ra);
1473 
1474 
1475     if (move) {
1476         /*
1477          * Copy backwards if necessary. The direction for a non-overlapping
1478          * copy is IMPDEF; we choose forwards.
1479          */
1480         if (copysize > 0x007FFFFFFFFFFFFFULL) {
1481             copysize = 0x007FFFFFFFFFFFFFULL;
1482         }
1483         uint64_t fs = extract64(fromaddr, 0, 56);
1484         uint64_t ts = extract64(toaddr, 0, 56);
1485         uint64_t fe = extract64(fromaddr + copysize, 0, 56);
1486 
1487         if (fs < ts && fe > ts) {
1488             forwards = false;
1489         }
1490     } else {
1491         if (copysize > INT64_MAX) {
1492             copysize = INT64_MAX;
1493         }
1494     }
1495 
1496     if (!mte_checks_needed(fromaddr, rdesc)) {
1497         rdesc = 0;
1498     }
1499     if (!mte_checks_needed(toaddr, wdesc)) {
1500         wdesc = 0;
1501     }
1502 
1503     if (forwards) {
1504         stagecopysize = MIN(copysize, page_limit(toaddr));
1505         stagecopysize = MIN(stagecopysize, page_limit(fromaddr));
1506         while (stagecopysize) {
1507             env->xregs[rd] = toaddr;
1508             env->xregs[rs] = fromaddr;
1509             env->xregs[rn] = copysize;
1510             step = copy_step(env, toaddr, fromaddr, stagecopysize,
1511                              wmemidx, rmemidx, &wdesc, &rdesc, ra);
1512             toaddr += step;
1513             fromaddr += step;
1514             copysize -= step;
1515             stagecopysize -= step;
1516         }
1517         /* Insn completed, so update registers to the Option A format */
1518         env->xregs[rd] = toaddr + copysize;
1519         env->xregs[rs] = fromaddr + copysize;
1520         env->xregs[rn] = -copysize;
1521     } else {
1522         /*
1523          * In a reverse copy the to and from addrs in Xs and Xd are the start
1524          * of the range, but it's more convenient for us to work with pointers
1525          * to the last byte being copied.
1526          */
1527         toaddr += copysize - 1;
1528         fromaddr += copysize - 1;
1529         stagecopysize = MIN(copysize, page_limit_rev(toaddr));
1530         stagecopysize = MIN(stagecopysize, page_limit_rev(fromaddr));
1531         while (stagecopysize) {
1532             env->xregs[rn] = copysize;
1533             step = copy_step_rev(env, toaddr, fromaddr, stagecopysize,
1534                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1535             copysize -= step;
1536             stagecopysize -= step;
1537             toaddr -= step;
1538             fromaddr -= step;
1539         }
1540         /*
1541          * Insn completed, so update registers to the Option A format.
1542          * For a reverse copy this is no different to the CPYP input format.
1543          */
1544         env->xregs[rn] = copysize;
1545     }
1546 
1547     /* Set NZCV = 0000 to indicate we are an Option A implementation */
1548     env->NF = 0;
1549     env->ZF = 1; /* our env->ZF encoding is inverted */
1550     env->CF = 0;
1551     env->VF = 0;
1552     return;
1553 }
1554 
1555 void HELPER(cpyp)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1556                   uint32_t rdesc)
1557 {
1558     do_cpyp(env, syndrome, wdesc, rdesc, true, GETPC());
1559 }
1560 
1561 void HELPER(cpyfp)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1562                    uint32_t rdesc)
1563 {
1564     do_cpyp(env, syndrome, wdesc, rdesc, false, GETPC());
1565 }
1566 
1567 static void do_cpym(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1568                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1569 {
1570     /* Main: we choose to copy until less than a page remaining */
1571     CPUState *cs = env_cpu(env);
1572     int rd = mops_destreg(syndrome);
1573     int rs = mops_srcreg(syndrome);
1574     int rn = mops_sizereg(syndrome);
1575     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1576     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1577     bool forwards = true;
1578     uint64_t toaddr, fromaddr, copysize, step;
1579 
1580     check_mops_enabled(env, ra);
1581 
1582     /* We choose to NOP out "no data to copy" before consistency checks */
1583     if (env->xregs[rn] == 0) {
1584         return;
1585     }
1586 
1587     check_mops_wrong_option(env, syndrome, ra);
1588 
1589     if (move) {
1590         forwards = (int64_t)env->xregs[rn] < 0;
1591     }
1592 
1593     if (forwards) {
1594         toaddr = env->xregs[rd] + env->xregs[rn];
1595         fromaddr = env->xregs[rs] + env->xregs[rn];
1596         copysize = -env->xregs[rn];
1597     } else {
1598         copysize = env->xregs[rn];
1599         /* This toaddr and fromaddr point to the *last* byte to copy */
1600         toaddr = env->xregs[rd] + copysize - 1;
1601         fromaddr = env->xregs[rs] + copysize - 1;
1602     }
1603 
1604     if (!mte_checks_needed(fromaddr, rdesc)) {
1605         rdesc = 0;
1606     }
1607     if (!mte_checks_needed(toaddr, wdesc)) {
1608         wdesc = 0;
1609     }
1610 
1611     /* Our implementation has no particular parameter requirements for CPYM */
1612 
1613     /* Do the actual memmove */
1614     if (forwards) {
1615         while (copysize >= TARGET_PAGE_SIZE) {
1616             step = copy_step(env, toaddr, fromaddr, copysize,
1617                              wmemidx, rmemidx, &wdesc, &rdesc, ra);
1618             toaddr += step;
1619             fromaddr += step;
1620             copysize -= step;
1621             env->xregs[rn] = -copysize;
1622             if (copysize >= TARGET_PAGE_SIZE &&
1623                 unlikely(cpu_loop_exit_requested(cs))) {
1624                 cpu_loop_exit_restore(cs, ra);
1625             }
1626         }
1627     } else {
1628         while (copysize >= TARGET_PAGE_SIZE) {
1629             step = copy_step_rev(env, toaddr, fromaddr, copysize,
1630                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1631             toaddr -= step;
1632             fromaddr -= step;
1633             copysize -= step;
1634             env->xregs[rn] = copysize;
1635             if (copysize >= TARGET_PAGE_SIZE &&
1636                 unlikely(cpu_loop_exit_requested(cs))) {
1637                 cpu_loop_exit_restore(cs, ra);
1638             }
1639         }
1640     }
1641 }
1642 
1643 void HELPER(cpym)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1644                   uint32_t rdesc)
1645 {
1646     do_cpym(env, syndrome, wdesc, rdesc, true, GETPC());
1647 }
1648 
1649 void HELPER(cpyfm)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1650                    uint32_t rdesc)
1651 {
1652     do_cpym(env, syndrome, wdesc, rdesc, false, GETPC());
1653 }
1654 
1655 static void do_cpye(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1656                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1657 {
1658     /* Epilogue: do the last partial page */
1659     int rd = mops_destreg(syndrome);
1660     int rs = mops_srcreg(syndrome);
1661     int rn = mops_sizereg(syndrome);
1662     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1663     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1664     bool forwards = true;
1665     uint64_t toaddr, fromaddr, copysize, step;
1666 
1667     check_mops_enabled(env, ra);
1668 
1669     /* We choose to NOP out "no data to copy" before consistency checks */
1670     if (env->xregs[rn] == 0) {
1671         return;
1672     }
1673 
1674     check_mops_wrong_option(env, syndrome, ra);
1675 
1676     if (move) {
1677         forwards = (int64_t)env->xregs[rn] < 0;
1678     }
1679 
1680     if (forwards) {
1681         toaddr = env->xregs[rd] + env->xregs[rn];
1682         fromaddr = env->xregs[rs] + env->xregs[rn];
1683         copysize = -env->xregs[rn];
1684     } else {
1685         copysize = env->xregs[rn];
1686         /* This toaddr and fromaddr point to the *last* byte to copy */
1687         toaddr = env->xregs[rd] + copysize - 1;
1688         fromaddr = env->xregs[rs] + copysize - 1;
1689     }
1690 
1691     if (!mte_checks_needed(fromaddr, rdesc)) {
1692         rdesc = 0;
1693     }
1694     if (!mte_checks_needed(toaddr, wdesc)) {
1695         wdesc = 0;
1696     }
1697 
1698     /* Check the size; we don't want to have do a check-for-interrupts */
1699     if (copysize >= TARGET_PAGE_SIZE) {
1700         raise_exception_ra(env, EXCP_UDEF, syndrome,
1701                            mops_mismatch_exception_target_el(env), ra);
1702     }
1703 
1704     /* Do the actual memmove */
1705     if (forwards) {
1706         while (copysize > 0) {
1707             step = copy_step(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     } else {
1715         while (copysize > 0) {
1716             step = copy_step_rev(env, toaddr, fromaddr, copysize,
1717                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1718             toaddr -= step;
1719             fromaddr -= step;
1720             copysize -= step;
1721             env->xregs[rn] = copysize;
1722         }
1723     }
1724 }
1725 
1726 void HELPER(cpye)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1727                   uint32_t rdesc)
1728 {
1729     do_cpye(env, syndrome, wdesc, rdesc, true, GETPC());
1730 }
1731 
1732 void HELPER(cpyfe)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1733                    uint32_t rdesc)
1734 {
1735     do_cpye(env, syndrome, wdesc, rdesc, false, GETPC());
1736 }
1737 
1738 static bool is_guarded_page(CPUARMState *env, target_ulong addr, uintptr_t ra)
1739 {
1740 #ifdef CONFIG_USER_ONLY
1741     return page_get_flags(addr) & PAGE_BTI;
1742 #else
1743     CPUTLBEntryFull *full;
1744     void *host;
1745     int mmu_idx = cpu_mmu_index(env_cpu(env), true);
1746     int flags = probe_access_full(env, addr, 0, MMU_INST_FETCH, mmu_idx,
1747                                   false, &host, &full, ra);
1748 
1749     assert(!(flags & TLB_INVALID_MASK));
1750     return full->extra.arm.guarded;
1751 #endif
1752 }
1753 
1754 void HELPER(guarded_page_check)(CPUARMState *env)
1755 {
1756     /*
1757      * We have already verified that bti is enabled, and that the
1758      * instruction at PC is not ok for BTYPE.  This is always at
1759      * the beginning of a block, so PC is always up-to-date and
1760      * no unwind is required.
1761      */
1762     if (is_guarded_page(env, env->pc, 0)) {
1763         raise_exception(env, EXCP_UDEF, syn_btitrap(env->btype),
1764                         exception_target_el(env));
1765     }
1766 }
1767 
1768 void HELPER(guarded_page_br)(CPUARMState *env, target_ulong pc)
1769 {
1770     /*
1771      * We have already checked for branch via x16 and x17.
1772      * What remains for choosing BTYPE is checking for a guarded page.
1773      */
1774     env->btype = is_guarded_page(env, pc, GETPC()) ? 3 : 1;
1775 }
1776