xref: /openbmc/qemu/target/arm/tcg/helper-a64.c (revision 8a2b516ba2855c4530388051de2b8d17bc780ea8)
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 #include "vec_internal.h"
42 
43 /* C2.4.7 Multiply and divide */
44 /* special cases for 0 and LLONG_MIN are mandated by the standard */
45 uint64_t HELPER(udiv64)(uint64_t num, uint64_t den)
46 {
47     if (den == 0) {
48         return 0;
49     }
50     return num / den;
51 }
52 
53 int64_t HELPER(sdiv64)(int64_t num, int64_t den)
54 {
55     if (den == 0) {
56         return 0;
57     }
58     if (num == LLONG_MIN && den == -1) {
59         return LLONG_MIN;
60     }
61     return num / den;
62 }
63 
64 uint64_t HELPER(rbit64)(uint64_t x)
65 {
66     return revbit64(x);
67 }
68 
69 void HELPER(msr_i_spsel)(CPUARMState *env, uint32_t imm)
70 {
71     update_spsel(env, imm);
72 }
73 
74 void HELPER(msr_set_allint_el1)(CPUARMState *env)
75 {
76     /* ALLINT update to PSTATE. */
77     if (arm_hcrx_el2_eff(env) & HCRX_TALLINT) {
78         raise_exception_ra(env, EXCP_UDEF,
79                            syn_aa64_sysregtrap(0, 1, 0, 4, 1, 0x1f, 0), 2,
80                            GETPC());
81     }
82 
83     env->pstate |= PSTATE_ALLINT;
84 }
85 
86 static void daif_check(CPUARMState *env, uint32_t op,
87                        uint32_t imm, uintptr_t ra)
88 {
89     /* DAIF update to PSTATE. This is OK from EL0 only if UMA is set.  */
90     if (arm_current_el(env) == 0 && !(arm_sctlr(env, 0) & SCTLR_UMA)) {
91         raise_exception_ra(env, EXCP_UDEF,
92                            syn_aa64_sysregtrap(0, extract32(op, 0, 3),
93                                                extract32(op, 3, 3), 4,
94                                                imm, 0x1f, 0),
95                            exception_target_el(env), ra);
96     }
97 }
98 
99 void HELPER(msr_i_daifset)(CPUARMState *env, uint32_t imm)
100 {
101     daif_check(env, 0x1e, imm, GETPC());
102     env->daif |= (imm << 6) & PSTATE_DAIF;
103     arm_rebuild_hflags(env);
104 }
105 
106 void HELPER(msr_i_daifclear)(CPUARMState *env, uint32_t imm)
107 {
108     daif_check(env, 0x1f, imm, GETPC());
109     env->daif &= ~((imm << 6) & PSTATE_DAIF);
110     arm_rebuild_hflags(env);
111 }
112 
113 /* Convert a softfloat float_relation_ (as returned by
114  * the float*_compare functions) to the correct ARM
115  * NZCV flag state.
116  */
117 static inline uint32_t float_rel_to_flags(int res)
118 {
119     uint64_t flags;
120     switch (res) {
121     case float_relation_equal:
122         flags = PSTATE_Z | PSTATE_C;
123         break;
124     case float_relation_less:
125         flags = PSTATE_N;
126         break;
127     case float_relation_greater:
128         flags = PSTATE_C;
129         break;
130     case float_relation_unordered:
131     default:
132         flags = PSTATE_C | PSTATE_V;
133         break;
134     }
135     return flags;
136 }
137 
138 uint64_t HELPER(vfp_cmph_a64)(uint32_t x, uint32_t y, float_status *fp_status)
139 {
140     return float_rel_to_flags(float16_compare_quiet(x, y, fp_status));
141 }
142 
143 uint64_t HELPER(vfp_cmpeh_a64)(uint32_t x, uint32_t y, float_status *fp_status)
144 {
145     return float_rel_to_flags(float16_compare(x, y, fp_status));
146 }
147 
148 uint64_t HELPER(vfp_cmps_a64)(float32 x, float32 y, float_status *fp_status)
149 {
150     return float_rel_to_flags(float32_compare_quiet(x, y, fp_status));
151 }
152 
153 uint64_t HELPER(vfp_cmpes_a64)(float32 x, float32 y, float_status *fp_status)
154 {
155     return float_rel_to_flags(float32_compare(x, y, fp_status));
156 }
157 
158 uint64_t HELPER(vfp_cmpd_a64)(float64 x, float64 y, float_status *fp_status)
159 {
160     return float_rel_to_flags(float64_compare_quiet(x, y, fp_status));
161 }
162 
163 uint64_t HELPER(vfp_cmped_a64)(float64 x, float64 y, float_status *fp_status)
164 {
165     return float_rel_to_flags(float64_compare(x, y, fp_status));
166 }
167 
168 float32 HELPER(vfp_mulxs)(float32 a, float32 b, float_status *fpst)
169 {
170     a = float32_squash_input_denormal(a, fpst);
171     b = float32_squash_input_denormal(b, fpst);
172 
173     if ((float32_is_zero(a) && float32_is_infinity(b)) ||
174         (float32_is_infinity(a) && float32_is_zero(b))) {
175         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
176         return make_float32((1U << 30) |
177                             ((float32_val(a) ^ float32_val(b)) & (1U << 31)));
178     }
179     return float32_mul(a, b, fpst);
180 }
181 
182 float64 HELPER(vfp_mulxd)(float64 a, float64 b, float_status *fpst)
183 {
184     a = float64_squash_input_denormal(a, fpst);
185     b = float64_squash_input_denormal(b, fpst);
186 
187     if ((float64_is_zero(a) && float64_is_infinity(b)) ||
188         (float64_is_infinity(a) && float64_is_zero(b))) {
189         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
190         return make_float64((1ULL << 62) |
191                             ((float64_val(a) ^ float64_val(b)) & (1ULL << 63)));
192     }
193     return float64_mul(a, b, fpst);
194 }
195 
196 /* 64bit/double versions of the neon float compare functions */
197 uint64_t HELPER(neon_ceq_f64)(float64 a, float64 b, float_status *fpst)
198 {
199     return -float64_eq_quiet(a, b, fpst);
200 }
201 
202 uint64_t HELPER(neon_cge_f64)(float64 a, float64 b, float_status *fpst)
203 {
204     return -float64_le(b, a, fpst);
205 }
206 
207 uint64_t HELPER(neon_cgt_f64)(float64 a, float64 b, float_status *fpst)
208 {
209     return -float64_lt(b, a, fpst);
210 }
211 
212 /*
213  * Reciprocal step and sqrt step. Note that unlike the A32/T32
214  * versions, these do a fully fused multiply-add or
215  * multiply-add-and-halve.
216  * The FPCR.AH == 1 versions need to avoid flipping the sign of NaN.
217  */
218 #define DO_RECPS(NAME, CTYPE, FLOATTYPE, CHSFN)                         \
219     CTYPE HELPER(NAME)(CTYPE a, CTYPE b, float_status *fpst)            \
220     {                                                                   \
221         a = FLOATTYPE ## _squash_input_denormal(a, fpst);               \
222         b = FLOATTYPE ## _squash_input_denormal(b, fpst);               \
223         a = FLOATTYPE ## _ ## CHSFN(a);                                 \
224         if ((FLOATTYPE ## _is_infinity(a) && FLOATTYPE ## _is_zero(b)) || \
225             (FLOATTYPE ## _is_infinity(b) && FLOATTYPE ## _is_zero(a))) { \
226             return FLOATTYPE ## _two;                                   \
227         }                                                               \
228         return FLOATTYPE ## _muladd(a, b, FLOATTYPE ## _two, 0, fpst);  \
229     }
230 
231 DO_RECPS(recpsf_f16, uint32_t, float16, chs)
232 DO_RECPS(recpsf_f32, float32, float32, chs)
233 DO_RECPS(recpsf_f64, float64, float64, chs)
234 DO_RECPS(recpsf_ah_f16, uint32_t, float16, ah_chs)
235 DO_RECPS(recpsf_ah_f32, float32, float32, ah_chs)
236 DO_RECPS(recpsf_ah_f64, float64, float64, ah_chs)
237 
238 #define DO_RSQRTSF(NAME, CTYPE, FLOATTYPE, CHSFN)                       \
239     CTYPE HELPER(NAME)(CTYPE a, CTYPE b, float_status *fpst)            \
240     {                                                                   \
241         a = FLOATTYPE ## _squash_input_denormal(a, fpst);               \
242         b = FLOATTYPE ## _squash_input_denormal(b, fpst);               \
243         a = FLOATTYPE ## _ ## CHSFN(a);                                 \
244         if ((FLOATTYPE ## _is_infinity(a) && FLOATTYPE ## _is_zero(b)) || \
245             (FLOATTYPE ## _is_infinity(b) && FLOATTYPE ## _is_zero(a))) { \
246             return FLOATTYPE ## _one_point_five;                        \
247         }                                                               \
248         return FLOATTYPE ## _muladd_scalbn(a, b, FLOATTYPE ## _three,   \
249                                            -1, 0, fpst);                \
250     }                                                                   \
251 
252 DO_RSQRTSF(rsqrtsf_f16, uint32_t, float16, chs)
253 DO_RSQRTSF(rsqrtsf_f32, float32, float32, chs)
254 DO_RSQRTSF(rsqrtsf_f64, float64, float64, chs)
255 DO_RSQRTSF(rsqrtsf_ah_f16, uint32_t, float16, ah_chs)
256 DO_RSQRTSF(rsqrtsf_ah_f32, float32, float32, ah_chs)
257 DO_RSQRTSF(rsqrtsf_ah_f64, float64, float64, ah_chs)
258 
259 /* Floating-point reciprocal exponent - see FPRecpX in ARM ARM */
260 uint32_t HELPER(frecpx_f16)(uint32_t a, float_status *fpst)
261 {
262     uint16_t val16, sbit;
263     int16_t exp;
264 
265     if (float16_is_any_nan(a)) {
266         float16 nan = a;
267         if (float16_is_signaling_nan(a, fpst)) {
268             float_raise(float_flag_invalid, fpst);
269             if (!fpst->default_nan_mode) {
270                 nan = float16_silence_nan(a, fpst);
271             }
272         }
273         if (fpst->default_nan_mode) {
274             nan = float16_default_nan(fpst);
275         }
276         return nan;
277     }
278 
279     a = float16_squash_input_denormal(a, fpst);
280 
281     val16 = float16_val(a);
282     sbit = 0x8000 & val16;
283     exp = extract32(val16, 10, 5);
284 
285     if (exp == 0) {
286         return make_float16(deposit32(sbit, 10, 5, 0x1e));
287     } else {
288         return make_float16(deposit32(sbit, 10, 5, ~exp));
289     }
290 }
291 
292 float32 HELPER(frecpx_f32)(float32 a, float_status *fpst)
293 {
294     uint32_t val32, sbit;
295     int32_t exp;
296 
297     if (float32_is_any_nan(a)) {
298         float32 nan = a;
299         if (float32_is_signaling_nan(a, fpst)) {
300             float_raise(float_flag_invalid, fpst);
301             if (!fpst->default_nan_mode) {
302                 nan = float32_silence_nan(a, fpst);
303             }
304         }
305         if (fpst->default_nan_mode) {
306             nan = float32_default_nan(fpst);
307         }
308         return nan;
309     }
310 
311     a = float32_squash_input_denormal(a, fpst);
312 
313     val32 = float32_val(a);
314     sbit = 0x80000000ULL & val32;
315     exp = extract32(val32, 23, 8);
316 
317     if (exp == 0) {
318         return make_float32(sbit | (0xfe << 23));
319     } else {
320         return make_float32(sbit | (~exp & 0xff) << 23);
321     }
322 }
323 
324 float64 HELPER(frecpx_f64)(float64 a, float_status *fpst)
325 {
326     uint64_t val64, sbit;
327     int64_t exp;
328 
329     if (float64_is_any_nan(a)) {
330         float64 nan = a;
331         if (float64_is_signaling_nan(a, fpst)) {
332             float_raise(float_flag_invalid, fpst);
333             if (!fpst->default_nan_mode) {
334                 nan = float64_silence_nan(a, fpst);
335             }
336         }
337         if (fpst->default_nan_mode) {
338             nan = float64_default_nan(fpst);
339         }
340         return nan;
341     }
342 
343     a = float64_squash_input_denormal(a, fpst);
344 
345     val64 = float64_val(a);
346     sbit = 0x8000000000000000ULL & val64;
347     exp = extract64(float64_val(a), 52, 11);
348 
349     if (exp == 0) {
350         return make_float64(sbit | (0x7feULL << 52));
351     } else {
352         return make_float64(sbit | (~exp & 0x7ffULL) << 52);
353     }
354 }
355 
356 float32 HELPER(fcvtx_f64_to_f32)(float64 a, float_status *fpst)
357 {
358     float32 r;
359     int old = get_float_rounding_mode(fpst);
360 
361     set_float_rounding_mode(float_round_to_odd, fpst);
362     r = float64_to_float32(a, fpst);
363     set_float_rounding_mode(old, fpst);
364     return r;
365 }
366 
367 /*
368  * AH=1 min/max have some odd special cases:
369  * comparing two zeroes (regardless of sign), (NaN, anything),
370  * or (anything, NaN) should return the second argument (possibly
371  * squashed to zero).
372  * Also, denormal outputs are not squashed to zero regardless of FZ or FZ16.
373  */
374 #define AH_MINMAX_HELPER(NAME, CTYPE, FLOATTYPE, MINMAX)                \
375     CTYPE HELPER(NAME)(CTYPE a, CTYPE b, float_status *fpst)            \
376     {                                                                   \
377         bool save;                                                      \
378         CTYPE r;                                                        \
379         a = FLOATTYPE ## _squash_input_denormal(a, fpst);               \
380         b = FLOATTYPE ## _squash_input_denormal(b, fpst);               \
381         if (FLOATTYPE ## _is_zero(a) && FLOATTYPE ## _is_zero(b)) {     \
382             return b;                                                   \
383         }                                                               \
384         if (FLOATTYPE ## _is_any_nan(a) ||                              \
385             FLOATTYPE ## _is_any_nan(b)) {                              \
386             float_raise(float_flag_invalid, fpst);                      \
387             return b;                                                   \
388         }                                                               \
389         save = get_flush_to_zero(fpst);                                 \
390         set_flush_to_zero(false, fpst);                                 \
391         r = FLOATTYPE ## _ ## MINMAX(a, b, fpst);                       \
392         set_flush_to_zero(save, fpst);                                  \
393         return r;                                                       \
394     }
395 
396 AH_MINMAX_HELPER(vfp_ah_minh, dh_ctype_f16, float16, min)
397 AH_MINMAX_HELPER(vfp_ah_mins, float32, float32, min)
398 AH_MINMAX_HELPER(vfp_ah_mind, float64, float64, min)
399 AH_MINMAX_HELPER(vfp_ah_maxh, dh_ctype_f16, float16, max)
400 AH_MINMAX_HELPER(vfp_ah_maxs, float32, float32, max)
401 AH_MINMAX_HELPER(vfp_ah_maxd, float64, float64, max)
402 
403 /* 64-bit versions of the CRC helpers. Note that although the operation
404  * (and the prototypes of crc32c() and crc32() mean that only the bottom
405  * 32 bits of the accumulator and result are used, we pass and return
406  * uint64_t for convenience of the generated code. Unlike the 32-bit
407  * instruction set versions, val may genuinely have 64 bits of data in it.
408  * The upper bytes of val (above the number specified by 'bytes') must have
409  * been zeroed out by the caller.
410  */
411 uint64_t HELPER(crc32_64)(uint64_t acc, uint64_t val, uint32_t bytes)
412 {
413     uint8_t buf[8];
414 
415     stq_le_p(buf, val);
416 
417     /* zlib crc32 converts the accumulator and output to one's complement.  */
418     return crc32(acc ^ 0xffffffff, buf, bytes) ^ 0xffffffff;
419 }
420 
421 uint64_t HELPER(crc32c_64)(uint64_t acc, uint64_t val, uint32_t bytes)
422 {
423     uint8_t buf[8];
424 
425     stq_le_p(buf, val);
426 
427     /* Linux crc32c converts the output to one's complement.  */
428     return crc32c(acc, buf, bytes) ^ 0xffffffff;
429 }
430 
431 /*
432  * AdvSIMD half-precision
433  */
434 
435 #define ADVSIMD_HELPER(name, suffix) HELPER(glue(glue(advsimd_, name), suffix))
436 
437 #define ADVSIMD_HALFOP(name) \
438 uint32_t ADVSIMD_HELPER(name, h)(uint32_t a, uint32_t b, float_status *fpst) \
439 { \
440     return float16_ ## name(a, b, fpst);    \
441 }
442 
443 #define ADVSIMD_TWOHALFOP(name)                                         \
444 uint32_t ADVSIMD_HELPER(name, 2h)(uint32_t two_a, uint32_t two_b,       \
445                                   float_status *fpst)                   \
446 { \
447     float16  a1, a2, b1, b2;                        \
448     uint32_t r1, r2;                                \
449     a1 = extract32(two_a, 0, 16);                   \
450     a2 = extract32(two_a, 16, 16);                  \
451     b1 = extract32(two_b, 0, 16);                   \
452     b2 = extract32(two_b, 16, 16);                  \
453     r1 = float16_ ## name(a1, b1, fpst);            \
454     r2 = float16_ ## name(a2, b2, fpst);            \
455     return deposit32(r1, 16, 16, r2);               \
456 }
457 
458 ADVSIMD_TWOHALFOP(add)
459 ADVSIMD_TWOHALFOP(sub)
460 ADVSIMD_TWOHALFOP(mul)
461 ADVSIMD_TWOHALFOP(div)
462 ADVSIMD_TWOHALFOP(min)
463 ADVSIMD_TWOHALFOP(max)
464 ADVSIMD_TWOHALFOP(minnum)
465 ADVSIMD_TWOHALFOP(maxnum)
466 
467 /* Data processing - scalar floating-point and advanced SIMD */
468 static float16 float16_mulx(float16 a, float16 b, float_status *fpst)
469 {
470     a = float16_squash_input_denormal(a, fpst);
471     b = float16_squash_input_denormal(b, fpst);
472 
473     if ((float16_is_zero(a) && float16_is_infinity(b)) ||
474         (float16_is_infinity(a) && float16_is_zero(b))) {
475         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
476         return make_float16((1U << 14) |
477                             ((float16_val(a) ^ float16_val(b)) & (1U << 15)));
478     }
479     return float16_mul(a, b, fpst);
480 }
481 
482 ADVSIMD_HALFOP(mulx)
483 ADVSIMD_TWOHALFOP(mulx)
484 
485 /* fused multiply-accumulate */
486 uint32_t HELPER(advsimd_muladdh)(uint32_t a, uint32_t b, uint32_t c,
487                                  float_status *fpst)
488 {
489     return float16_muladd(a, b, c, 0, fpst);
490 }
491 
492 uint32_t HELPER(advsimd_muladd2h)(uint32_t two_a, uint32_t two_b,
493                                   uint32_t two_c, float_status *fpst)
494 {
495     float16  a1, a2, b1, b2, c1, c2;
496     uint32_t r1, r2;
497     a1 = extract32(two_a, 0, 16);
498     a2 = extract32(two_a, 16, 16);
499     b1 = extract32(two_b, 0, 16);
500     b2 = extract32(two_b, 16, 16);
501     c1 = extract32(two_c, 0, 16);
502     c2 = extract32(two_c, 16, 16);
503     r1 = float16_muladd(a1, b1, c1, 0, fpst);
504     r2 = float16_muladd(a2, b2, c2, 0, fpst);
505     return deposit32(r1, 16, 16, r2);
506 }
507 
508 /*
509  * Floating point comparisons produce an integer result. Softfloat
510  * routines return float_relation types which we convert to the 0/-1
511  * Neon requires.
512  */
513 
514 #define ADVSIMD_CMPRES(test) (test) ? 0xffff : 0
515 
516 uint32_t HELPER(advsimd_ceq_f16)(uint32_t a, uint32_t b, float_status *fpst)
517 {
518     int compare = float16_compare_quiet(a, b, fpst);
519     return ADVSIMD_CMPRES(compare == float_relation_equal);
520 }
521 
522 uint32_t HELPER(advsimd_cge_f16)(uint32_t a, uint32_t b, float_status *fpst)
523 {
524     int compare = float16_compare(a, b, fpst);
525     return ADVSIMD_CMPRES(compare == float_relation_greater ||
526                           compare == float_relation_equal);
527 }
528 
529 uint32_t HELPER(advsimd_cgt_f16)(uint32_t a, uint32_t b, float_status *fpst)
530 {
531     int compare = float16_compare(a, b, fpst);
532     return ADVSIMD_CMPRES(compare == float_relation_greater);
533 }
534 
535 uint32_t HELPER(advsimd_acge_f16)(uint32_t a, uint32_t b, float_status *fpst)
536 {
537     float16 f0 = float16_abs(a);
538     float16 f1 = float16_abs(b);
539     int compare = float16_compare(f0, f1, fpst);
540     return ADVSIMD_CMPRES(compare == float_relation_greater ||
541                           compare == float_relation_equal);
542 }
543 
544 uint32_t HELPER(advsimd_acgt_f16)(uint32_t a, uint32_t b, float_status *fpst)
545 {
546     float16 f0 = float16_abs(a);
547     float16 f1 = float16_abs(b);
548     int compare = float16_compare(f0, f1, fpst);
549     return ADVSIMD_CMPRES(compare == float_relation_greater);
550 }
551 
552 /* round to integral */
553 uint32_t HELPER(advsimd_rinth_exact)(uint32_t x, float_status *fp_status)
554 {
555     return float16_round_to_int(x, fp_status);
556 }
557 
558 uint32_t HELPER(advsimd_rinth)(uint32_t x, float_status *fp_status)
559 {
560     int old_flags = get_float_exception_flags(fp_status), new_flags;
561     float16 ret;
562 
563     ret = float16_round_to_int(x, fp_status);
564 
565     /* Suppress any inexact exceptions the conversion produced */
566     if (!(old_flags & float_flag_inexact)) {
567         new_flags = get_float_exception_flags(fp_status);
568         set_float_exception_flags(new_flags & ~float_flag_inexact, fp_status);
569     }
570 
571     return ret;
572 }
573 
574 static int el_from_spsr(uint32_t spsr)
575 {
576     /* Return the exception level that this SPSR is requesting a return to,
577      * or -1 if it is invalid (an illegal return)
578      */
579     if (spsr & PSTATE_nRW) {
580         switch (spsr & CPSR_M) {
581         case ARM_CPU_MODE_USR:
582             return 0;
583         case ARM_CPU_MODE_HYP:
584             return 2;
585         case ARM_CPU_MODE_FIQ:
586         case ARM_CPU_MODE_IRQ:
587         case ARM_CPU_MODE_SVC:
588         case ARM_CPU_MODE_ABT:
589         case ARM_CPU_MODE_UND:
590         case ARM_CPU_MODE_SYS:
591             return 1;
592         case ARM_CPU_MODE_MON:
593             /* Returning to Mon from AArch64 is never possible,
594              * so this is an illegal return.
595              */
596         default:
597             return -1;
598         }
599     } else {
600         if (extract32(spsr, 1, 1)) {
601             /* Return with reserved M[1] bit set */
602             return -1;
603         }
604         if (extract32(spsr, 0, 4) == 1) {
605             /* return to EL0 with M[0] bit set */
606             return -1;
607         }
608         return extract32(spsr, 2, 2);
609     }
610 }
611 
612 static void cpsr_write_from_spsr_elx(CPUARMState *env,
613                                      uint32_t val)
614 {
615     uint32_t mask;
616 
617     /* Save SPSR_ELx.SS into PSTATE. */
618     env->pstate = (env->pstate & ~PSTATE_SS) | (val & PSTATE_SS);
619     val &= ~PSTATE_SS;
620 
621     /* Move DIT to the correct location for CPSR */
622     if (val & PSTATE_DIT) {
623         val &= ~PSTATE_DIT;
624         val |= CPSR_DIT;
625     }
626 
627     mask = aarch32_cpsr_valid_mask(env->features, \
628         &env_archcpu(env)->isar);
629     cpsr_write(env, val, mask, CPSRWriteRaw);
630 }
631 
632 void HELPER(exception_return)(CPUARMState *env, uint64_t new_pc)
633 {
634     ARMCPU *cpu = env_archcpu(env);
635     int cur_el = arm_current_el(env);
636     unsigned int spsr_idx = aarch64_banked_spsr_index(cur_el);
637     uint32_t spsr = env->banked_spsr[spsr_idx];
638     int new_el;
639     bool return_to_aa64 = (spsr & PSTATE_nRW) == 0;
640 
641     aarch64_save_sp(env, cur_el);
642 
643     arm_clear_exclusive(env);
644 
645     /* We must squash the PSTATE.SS bit to zero unless both of the
646      * following hold:
647      *  1. debug exceptions are currently disabled
648      *  2. singlestep will be active in the EL we return to
649      * We check 1 here and 2 after we've done the pstate/cpsr write() to
650      * transition to the EL we're going to.
651      */
652     if (arm_generate_debug_exceptions(env)) {
653         spsr &= ~PSTATE_SS;
654     }
655 
656     /*
657      * FEAT_RME forbids return from EL3 with an invalid security state.
658      * We don't need an explicit check for FEAT_RME here because we enforce
659      * in scr_write() that you can't set the NSE bit without it.
660      */
661     if (cur_el == 3 && (env->cp15.scr_el3 & (SCR_NS | SCR_NSE)) == SCR_NSE) {
662         goto illegal_return;
663     }
664 
665     new_el = el_from_spsr(spsr);
666     if (new_el == -1) {
667         goto illegal_return;
668     }
669     if (new_el > cur_el || (new_el == 2 && !arm_is_el2_enabled(env))) {
670         /* Disallow return to an EL which is unimplemented or higher
671          * than the current one.
672          */
673         goto illegal_return;
674     }
675 
676     if (new_el != 0 && arm_el_is_aa64(env, new_el) != return_to_aa64) {
677         /* Return to an EL which is configured for a different register width */
678         goto illegal_return;
679     }
680 
681     if (!return_to_aa64 && !cpu_isar_feature(aa64_aa32, cpu)) {
682         /* Return to AArch32 when CPU is AArch64-only */
683         goto illegal_return;
684     }
685 
686     if (new_el == 1 && (arm_hcr_el2_eff(env) & HCR_TGE)) {
687         goto illegal_return;
688     }
689 
690     bql_lock();
691     arm_call_pre_el_change_hook(cpu);
692     bql_unlock();
693 
694     if (!return_to_aa64) {
695         env->aarch64 = false;
696         /* We do a raw CPSR write because aarch64_sync_64_to_32()
697          * will sort the register banks out for us, and we've already
698          * caught all the bad-mode cases in el_from_spsr().
699          */
700         cpsr_write_from_spsr_elx(env, spsr);
701         if (!arm_singlestep_active(env)) {
702             env->pstate &= ~PSTATE_SS;
703         }
704         aarch64_sync_64_to_32(env);
705 
706         if (spsr & CPSR_T) {
707             env->regs[15] = new_pc & ~0x1;
708         } else {
709             env->regs[15] = new_pc & ~0x3;
710         }
711         helper_rebuild_hflags_a32(env, new_el);
712         qemu_log_mask(CPU_LOG_INT, "Exception return from AArch64 EL%d to "
713                       "AArch32 EL%d PC 0x%" PRIx32 "\n",
714                       cur_el, new_el, env->regs[15]);
715     } else {
716         int tbii;
717 
718         env->aarch64 = true;
719         spsr &= aarch64_pstate_valid_mask(&cpu->isar);
720         pstate_write(env, spsr);
721         if (!arm_singlestep_active(env)) {
722             env->pstate &= ~PSTATE_SS;
723         }
724         aarch64_restore_sp(env, new_el);
725         helper_rebuild_hflags_a64(env, new_el);
726 
727         /*
728          * Apply TBI to the exception return address.  We had to delay this
729          * until after we selected the new EL, so that we could select the
730          * correct TBI+TBID bits.  This is made easier by waiting until after
731          * the hflags rebuild, since we can pull the composite TBII field
732          * from there.
733          */
734         tbii = EX_TBFLAG_A64(env->hflags, TBII);
735         if ((tbii >> extract64(new_pc, 55, 1)) & 1) {
736             /* TBI is enabled. */
737             int core_mmu_idx = arm_env_mmu_index(env);
738             if (regime_has_2_ranges(core_to_aa64_mmu_idx(core_mmu_idx))) {
739                 new_pc = sextract64(new_pc, 0, 56);
740             } else {
741                 new_pc = extract64(new_pc, 0, 56);
742             }
743         }
744         env->pc = new_pc;
745 
746         qemu_log_mask(CPU_LOG_INT, "Exception return from AArch64 EL%d to "
747                       "AArch64 EL%d PC 0x%" PRIx64 "\n",
748                       cur_el, new_el, env->pc);
749     }
750 
751     /*
752      * Note that cur_el can never be 0.  If new_el is 0, then
753      * el0_a64 is return_to_aa64, else el0_a64 is ignored.
754      */
755     aarch64_sve_change_el(env, cur_el, new_el, return_to_aa64);
756 
757     bql_lock();
758     arm_call_el_change_hook(cpu);
759     bql_unlock();
760 
761     return;
762 
763 illegal_return:
764     /* Illegal return events of various kinds have architecturally
765      * mandated behaviour:
766      * restore NZCV and DAIF from SPSR_ELx
767      * set PSTATE.IL
768      * restore PC from ELR_ELx
769      * no change to exception level, execution state or stack pointer
770      */
771     env->pstate |= PSTATE_IL;
772     env->pc = new_pc;
773     spsr &= PSTATE_NZCV | PSTATE_DAIF | PSTATE_ALLINT;
774     spsr |= pstate_read(env) & ~(PSTATE_NZCV | PSTATE_DAIF | PSTATE_ALLINT);
775     pstate_write(env, spsr);
776     if (!arm_singlestep_active(env)) {
777         env->pstate &= ~PSTATE_SS;
778     }
779     helper_rebuild_hflags_a64(env, cur_el);
780     qemu_log_mask(LOG_GUEST_ERROR, "Illegal exception return at EL%d: "
781                   "resuming execution at 0x%" PRIx64 "\n", cur_el, env->pc);
782 }
783 
784 void HELPER(dc_zva)(CPUARMState *env, uint64_t vaddr_in)
785 {
786     uintptr_t ra = GETPC();
787 
788     /*
789      * Implement DC ZVA, which zeroes a fixed-length block of memory.
790      * Note that we do not implement the (architecturally mandated)
791      * alignment fault for attempts to use this on Device memory
792      * (which matches the usual QEMU behaviour of not implementing either
793      * alignment faults or any memory attribute handling).
794      */
795     int blocklen = 4 << env_archcpu(env)->dcz_blocksize;
796     uint64_t vaddr = vaddr_in & ~(blocklen - 1);
797     int mmu_idx = arm_env_mmu_index(env);
798     void *mem;
799 
800     /*
801      * Trapless lookup.  In addition to actual invalid page, may
802      * return NULL for I/O, watchpoints, clean pages, etc.
803      */
804     mem = tlb_vaddr_to_host(env, vaddr, MMU_DATA_STORE, mmu_idx);
805 
806 #ifndef CONFIG_USER_ONLY
807     if (unlikely(!mem)) {
808         /*
809          * Trap if accessing an invalid page.  DC_ZVA requires that we supply
810          * the original pointer for an invalid page.  But watchpoints require
811          * that we probe the actual space.  So do both.
812          */
813         (void) probe_write(env, vaddr_in, 1, mmu_idx, ra);
814         mem = probe_write(env, vaddr, blocklen, mmu_idx, ra);
815 
816         if (unlikely(!mem)) {
817             /*
818              * The only remaining reason for mem == NULL is I/O.
819              * Just do a series of byte writes as the architecture demands.
820              */
821             for (int i = 0; i < blocklen; i++) {
822                 cpu_stb_mmuidx_ra(env, vaddr + i, 0, mmu_idx, ra);
823             }
824             return;
825         }
826     }
827 #endif
828 
829     set_helper_retaddr(ra);
830     memset(mem, 0, blocklen);
831     clear_helper_retaddr();
832 }
833 
834 void HELPER(unaligned_access)(CPUARMState *env, uint64_t addr,
835                               uint32_t access_type, uint32_t mmu_idx)
836 {
837     arm_cpu_do_unaligned_access(env_cpu(env), addr, access_type,
838                                 mmu_idx, GETPC());
839 }
840 
841 /* Memory operations (memset, memmove, memcpy) */
842 
843 /*
844  * Return true if the CPY* and SET* insns can execute; compare
845  * pseudocode CheckMOPSEnabled(), though we refactor it a little.
846  */
847 static bool mops_enabled(CPUARMState *env)
848 {
849     int el = arm_current_el(env);
850 
851     if (el < 2 &&
852         (arm_hcr_el2_eff(env) & (HCR_E2H | HCR_TGE)) != (HCR_E2H | HCR_TGE) &&
853         !(arm_hcrx_el2_eff(env) & HCRX_MSCEN)) {
854         return false;
855     }
856 
857     if (el == 0) {
858         if (!el_is_in_host(env, 0)) {
859             return env->cp15.sctlr_el[1] & SCTLR_MSCEN;
860         } else {
861             return env->cp15.sctlr_el[2] & SCTLR_MSCEN;
862         }
863     }
864     return true;
865 }
866 
867 static void check_mops_enabled(CPUARMState *env, uintptr_t ra)
868 {
869     if (!mops_enabled(env)) {
870         raise_exception_ra(env, EXCP_UDEF, syn_uncategorized(),
871                            exception_target_el(env), ra);
872     }
873 }
874 
875 /*
876  * Return the target exception level for an exception due
877  * to mismatched arguments in a FEAT_MOPS copy or set.
878  * Compare pseudocode MismatchedCpySetTargetEL()
879  */
880 static int mops_mismatch_exception_target_el(CPUARMState *env)
881 {
882     int el = arm_current_el(env);
883 
884     if (el > 1) {
885         return el;
886     }
887     if (el == 0 && (arm_hcr_el2_eff(env) & HCR_TGE)) {
888         return 2;
889     }
890     if (el == 1 && (arm_hcrx_el2_eff(env) & HCRX_MCE2)) {
891         return 2;
892     }
893     return 1;
894 }
895 
896 /*
897  * Check whether an M or E instruction was executed with a CF value
898  * indicating the wrong option for this implementation.
899  * Assumes we are always Option A.
900  */
901 static void check_mops_wrong_option(CPUARMState *env, uint32_t syndrome,
902                                     uintptr_t ra)
903 {
904     if (env->CF != 0) {
905         syndrome |= 1 << 17; /* Set the wrong-option bit */
906         raise_exception_ra(env, EXCP_UDEF, syndrome,
907                            mops_mismatch_exception_target_el(env), ra);
908     }
909 }
910 
911 /*
912  * Return the maximum number of bytes we can transfer starting at addr
913  * without crossing a page boundary.
914  */
915 static uint64_t page_limit(uint64_t addr)
916 {
917     return TARGET_PAGE_ALIGN(addr + 1) - addr;
918 }
919 
920 /*
921  * Return the number of bytes we can copy starting from addr and working
922  * backwards without crossing a page boundary.
923  */
924 static uint64_t page_limit_rev(uint64_t addr)
925 {
926     return (addr & ~TARGET_PAGE_MASK) + 1;
927 }
928 
929 /*
930  * Perform part of a memory set on an area of guest memory starting at
931  * toaddr (a dirty address) and extending for setsize bytes.
932  *
933  * Returns the number of bytes actually set, which might be less than
934  * setsize; the caller should loop until the whole set has been done.
935  * The caller should ensure that the guest registers are correct
936  * for the possibility that the first byte of the set encounters
937  * an exception or watchpoint. We guarantee not to take any faults
938  * for bytes other than the first.
939  */
940 static uint64_t set_step(CPUARMState *env, uint64_t toaddr,
941                          uint64_t setsize, uint32_t data, int memidx,
942                          uint32_t *mtedesc, uintptr_t ra)
943 {
944     void *mem;
945 
946     setsize = MIN(setsize, page_limit(toaddr));
947     if (*mtedesc) {
948         uint64_t mtesize = mte_mops_probe(env, toaddr, setsize, *mtedesc);
949         if (mtesize == 0) {
950             /* Trap, or not. All CPU state is up to date */
951             mte_check_fail(env, *mtedesc, toaddr, ra);
952             /* Continue, with no further MTE checks required */
953             *mtedesc = 0;
954         } else {
955             /* Advance to the end, or to the tag mismatch */
956             setsize = MIN(setsize, mtesize);
957         }
958     }
959 
960     toaddr = useronly_clean_ptr(toaddr);
961     /*
962      * Trapless lookup: returns NULL for invalid page, I/O,
963      * watchpoints, clean pages, etc.
964      */
965     mem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, memidx);
966 
967 #ifndef CONFIG_USER_ONLY
968     if (unlikely(!mem)) {
969         /*
970          * Slow-path: just do one byte write. This will handle the
971          * watchpoint, invalid page, etc handling correctly.
972          * For clean code pages, the next iteration will see
973          * the page dirty and will use the fast path.
974          */
975         cpu_stb_mmuidx_ra(env, toaddr, data, memidx, ra);
976         return 1;
977     }
978 #endif
979     /* Easy case: just memset the host memory */
980     set_helper_retaddr(ra);
981     memset(mem, data, setsize);
982     clear_helper_retaddr();
983     return setsize;
984 }
985 
986 /*
987  * Similar, but setting tags. The architecture requires us to do this
988  * in 16-byte chunks. SETP accesses are not tag checked; they set
989  * the tags.
990  */
991 static uint64_t set_step_tags(CPUARMState *env, uint64_t toaddr,
992                               uint64_t setsize, uint32_t data, int memidx,
993                               uint32_t *mtedesc, uintptr_t ra)
994 {
995     void *mem;
996     uint64_t cleanaddr;
997 
998     setsize = MIN(setsize, page_limit(toaddr));
999 
1000     cleanaddr = useronly_clean_ptr(toaddr);
1001     /*
1002      * Trapless lookup: returns NULL for invalid page, I/O,
1003      * watchpoints, clean pages, etc.
1004      */
1005     mem = tlb_vaddr_to_host(env, cleanaddr, MMU_DATA_STORE, memidx);
1006 
1007 #ifndef CONFIG_USER_ONLY
1008     if (unlikely(!mem)) {
1009         /*
1010          * Slow-path: just do one write. This will handle the
1011          * watchpoint, invalid page, etc handling correctly.
1012          * The architecture requires that we do 16 bytes at a time,
1013          * and we know both ptr and size are 16 byte aligned.
1014          * For clean code pages, the next iteration will see
1015          * the page dirty and will use the fast path.
1016          */
1017         uint64_t repldata = data * 0x0101010101010101ULL;
1018         MemOpIdx oi16 = make_memop_idx(MO_TE | MO_128, memidx);
1019         cpu_st16_mmu(env, toaddr, int128_make128(repldata, repldata), oi16, ra);
1020         mte_mops_set_tags(env, toaddr, 16, *mtedesc);
1021         return 16;
1022     }
1023 #endif
1024     /* Easy case: just memset the host memory */
1025     set_helper_retaddr(ra);
1026     memset(mem, data, setsize);
1027     clear_helper_retaddr();
1028     mte_mops_set_tags(env, toaddr, setsize, *mtedesc);
1029     return setsize;
1030 }
1031 
1032 typedef uint64_t StepFn(CPUARMState *env, uint64_t toaddr,
1033                         uint64_t setsize, uint32_t data,
1034                         int memidx, uint32_t *mtedesc, uintptr_t ra);
1035 
1036 /* Extract register numbers from a MOPS exception syndrome value */
1037 static int mops_destreg(uint32_t syndrome)
1038 {
1039     return extract32(syndrome, 10, 5);
1040 }
1041 
1042 static int mops_srcreg(uint32_t syndrome)
1043 {
1044     return extract32(syndrome, 5, 5);
1045 }
1046 
1047 static int mops_sizereg(uint32_t syndrome)
1048 {
1049     return extract32(syndrome, 0, 5);
1050 }
1051 
1052 /*
1053  * Return true if TCMA and TBI bits mean we need to do MTE checks.
1054  * We only need to do this once per MOPS insn, not for every page.
1055  */
1056 static bool mte_checks_needed(uint64_t ptr, uint32_t desc)
1057 {
1058     int bit55 = extract64(ptr, 55, 1);
1059 
1060     /*
1061      * Note that tbi_check() returns true for "access checked" but
1062      * tcma_check() returns true for "access unchecked".
1063      */
1064     if (!tbi_check(desc, bit55)) {
1065         return false;
1066     }
1067     return !tcma_check(desc, bit55, allocation_tag_from_addr(ptr));
1068 }
1069 
1070 /* Take an exception if the SETG addr/size are not granule aligned */
1071 static void check_setg_alignment(CPUARMState *env, uint64_t ptr, uint64_t size,
1072                                  uint32_t memidx, uintptr_t ra)
1073 {
1074     if ((size != 0 && !QEMU_IS_ALIGNED(ptr, TAG_GRANULE)) ||
1075         !QEMU_IS_ALIGNED(size, TAG_GRANULE)) {
1076         arm_cpu_do_unaligned_access(env_cpu(env), ptr, MMU_DATA_STORE,
1077                                     memidx, ra);
1078 
1079     }
1080 }
1081 
1082 static uint64_t arm_reg_or_xzr(CPUARMState *env, int reg)
1083 {
1084     /*
1085      * Runtime equivalent of cpu_reg() -- return the CPU register value,
1086      * for contexts when index 31 means XZR (not SP).
1087      */
1088     return reg == 31 ? 0 : env->xregs[reg];
1089 }
1090 
1091 /*
1092  * For the Memory Set operation, our implementation chooses
1093  * always to use "option A", where we update Xd to the final
1094  * address in the SETP insn, and set Xn to be -(bytes remaining).
1095  * On SETM and SETE insns we only need update Xn.
1096  *
1097  * @env: CPU
1098  * @syndrome: syndrome value for mismatch exceptions
1099  * (also contains the register numbers we need to use)
1100  * @mtedesc: MTE descriptor word
1101  * @stepfn: function which does a single part of the set operation
1102  * @is_setg: true if this is the tag-setting SETG variant
1103  */
1104 static void do_setp(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1105                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1106 {
1107     /* Prologue: we choose to do up to the next page boundary */
1108     int rd = mops_destreg(syndrome);
1109     int rs = mops_srcreg(syndrome);
1110     int rn = mops_sizereg(syndrome);
1111     uint8_t data = arm_reg_or_xzr(env, rs);
1112     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1113     uint64_t toaddr = env->xregs[rd];
1114     uint64_t setsize = env->xregs[rn];
1115     uint64_t stagesetsize, step;
1116 
1117     check_mops_enabled(env, ra);
1118 
1119     if (setsize > INT64_MAX) {
1120         setsize = INT64_MAX;
1121         if (is_setg) {
1122             setsize &= ~0xf;
1123         }
1124     }
1125 
1126     if (unlikely(is_setg)) {
1127         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1128     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1129         mtedesc = 0;
1130     }
1131 
1132     stagesetsize = MIN(setsize, page_limit(toaddr));
1133     while (stagesetsize) {
1134         env->xregs[rd] = toaddr;
1135         env->xregs[rn] = setsize;
1136         step = stepfn(env, toaddr, stagesetsize, data, memidx, &mtedesc, ra);
1137         toaddr += step;
1138         setsize -= step;
1139         stagesetsize -= step;
1140     }
1141     /* Insn completed, so update registers to the Option A format */
1142     env->xregs[rd] = toaddr + setsize;
1143     env->xregs[rn] = -setsize;
1144 
1145     /* Set NZCV = 0000 to indicate we are an Option A implementation */
1146     env->NF = 0;
1147     env->ZF = 1; /* our env->ZF encoding is inverted */
1148     env->CF = 0;
1149     env->VF = 0;
1150 }
1151 
1152 void HELPER(setp)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1153 {
1154     do_setp(env, syndrome, mtedesc, set_step, false, GETPC());
1155 }
1156 
1157 void HELPER(setgp)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1158 {
1159     do_setp(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1160 }
1161 
1162 static void do_setm(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1163                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1164 {
1165     /* Main: we choose to do all the full-page chunks */
1166     CPUState *cs = env_cpu(env);
1167     int rd = mops_destreg(syndrome);
1168     int rs = mops_srcreg(syndrome);
1169     int rn = mops_sizereg(syndrome);
1170     uint8_t data = arm_reg_or_xzr(env, rs);
1171     uint64_t toaddr = env->xregs[rd] + env->xregs[rn];
1172     uint64_t setsize = -env->xregs[rn];
1173     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1174     uint64_t step, stagesetsize;
1175 
1176     check_mops_enabled(env, ra);
1177 
1178     /*
1179      * We're allowed to NOP out "no data to copy" before the consistency
1180      * checks; we choose to do so.
1181      */
1182     if (env->xregs[rn] == 0) {
1183         return;
1184     }
1185 
1186     check_mops_wrong_option(env, syndrome, ra);
1187 
1188     /*
1189      * Our implementation will work fine even if we have an unaligned
1190      * destination address, and because we update Xn every time around
1191      * the loop below and the return value from stepfn() may be less
1192      * than requested, we might find toaddr is unaligned. So we don't
1193      * have an IMPDEF check for alignment here.
1194      */
1195 
1196     if (unlikely(is_setg)) {
1197         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1198     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1199         mtedesc = 0;
1200     }
1201 
1202     /* Do the actual memset: we leave the last partial page to SETE */
1203     stagesetsize = setsize & TARGET_PAGE_MASK;
1204     while (stagesetsize > 0) {
1205         step = stepfn(env, toaddr, stagesetsize, data, memidx, &mtedesc, ra);
1206         toaddr += step;
1207         setsize -= step;
1208         stagesetsize -= step;
1209         env->xregs[rn] = -setsize;
1210         if (stagesetsize > 0 && unlikely(cpu_loop_exit_requested(cs))) {
1211             cpu_loop_exit_restore(cs, ra);
1212         }
1213     }
1214 }
1215 
1216 void HELPER(setm)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1217 {
1218     do_setm(env, syndrome, mtedesc, set_step, false, GETPC());
1219 }
1220 
1221 void HELPER(setgm)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1222 {
1223     do_setm(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1224 }
1225 
1226 static void do_sete(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1227                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1228 {
1229     /* Epilogue: do the last partial page */
1230     int rd = mops_destreg(syndrome);
1231     int rs = mops_srcreg(syndrome);
1232     int rn = mops_sizereg(syndrome);
1233     uint8_t data = arm_reg_or_xzr(env, rs);
1234     uint64_t toaddr = env->xregs[rd] + env->xregs[rn];
1235     uint64_t setsize = -env->xregs[rn];
1236     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1237     uint64_t step;
1238 
1239     check_mops_enabled(env, ra);
1240 
1241     /*
1242      * We're allowed to NOP out "no data to copy" before the consistency
1243      * checks; we choose to do so.
1244      */
1245     if (setsize == 0) {
1246         return;
1247     }
1248 
1249     check_mops_wrong_option(env, syndrome, ra);
1250 
1251     /*
1252      * Our implementation has no address alignment requirements, but
1253      * we do want to enforce the "less than a page" size requirement,
1254      * so we don't need to have the "check for interrupts" here.
1255      */
1256     if (setsize >= TARGET_PAGE_SIZE) {
1257         raise_exception_ra(env, EXCP_UDEF, syndrome,
1258                            mops_mismatch_exception_target_el(env), ra);
1259     }
1260 
1261     if (unlikely(is_setg)) {
1262         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1263     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1264         mtedesc = 0;
1265     }
1266 
1267     /* Do the actual memset */
1268     while (setsize > 0) {
1269         step = stepfn(env, toaddr, setsize, data, memidx, &mtedesc, ra);
1270         toaddr += step;
1271         setsize -= step;
1272         env->xregs[rn] = -setsize;
1273     }
1274 }
1275 
1276 void HELPER(sete)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1277 {
1278     do_sete(env, syndrome, mtedesc, set_step, false, GETPC());
1279 }
1280 
1281 void HELPER(setge)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1282 {
1283     do_sete(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1284 }
1285 
1286 /*
1287  * Perform part of a memory copy from the guest memory at fromaddr
1288  * and extending for copysize bytes, to the guest memory at
1289  * toaddr. Both addresses are dirty.
1290  *
1291  * Returns the number of bytes actually set, which might be less than
1292  * copysize; the caller should loop until the whole copy has been done.
1293  * The caller should ensure that the guest registers are correct
1294  * for the possibility that the first byte of the copy encounters
1295  * an exception or watchpoint. We guarantee not to take any faults
1296  * for bytes other than the first.
1297  */
1298 static uint64_t copy_step(CPUARMState *env, uint64_t toaddr, uint64_t fromaddr,
1299                           uint64_t copysize, int wmemidx, int rmemidx,
1300                           uint32_t *wdesc, uint32_t *rdesc, uintptr_t ra)
1301 {
1302     void *rmem;
1303     void *wmem;
1304 
1305     /* Don't cross a page boundary on either source or destination */
1306     copysize = MIN(copysize, page_limit(toaddr));
1307     copysize = MIN(copysize, page_limit(fromaddr));
1308     /*
1309      * Handle MTE tag checks: either handle the tag mismatch for byte 0,
1310      * or else copy up to but not including the byte with the mismatch.
1311      */
1312     if (*rdesc) {
1313         uint64_t mtesize = mte_mops_probe(env, fromaddr, copysize, *rdesc);
1314         if (mtesize == 0) {
1315             mte_check_fail(env, *rdesc, fromaddr, ra);
1316             *rdesc = 0;
1317         } else {
1318             copysize = MIN(copysize, mtesize);
1319         }
1320     }
1321     if (*wdesc) {
1322         uint64_t mtesize = mte_mops_probe(env, toaddr, copysize, *wdesc);
1323         if (mtesize == 0) {
1324             mte_check_fail(env, *wdesc, toaddr, ra);
1325             *wdesc = 0;
1326         } else {
1327             copysize = MIN(copysize, mtesize);
1328         }
1329     }
1330 
1331     toaddr = useronly_clean_ptr(toaddr);
1332     fromaddr = useronly_clean_ptr(fromaddr);
1333     /* Trapless lookup of whether we can get a host memory pointer */
1334     wmem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, wmemidx);
1335     rmem = tlb_vaddr_to_host(env, fromaddr, MMU_DATA_LOAD, rmemidx);
1336 
1337 #ifndef CONFIG_USER_ONLY
1338     /*
1339      * If we don't have host memory for both source and dest then just
1340      * do a single byte copy. This will handle watchpoints, invalid pages,
1341      * etc correctly. For clean code pages, the next iteration will see
1342      * the page dirty and will use the fast path.
1343      */
1344     if (unlikely(!rmem || !wmem)) {
1345         uint8_t byte;
1346         if (rmem) {
1347             byte = *(uint8_t *)rmem;
1348         } else {
1349             byte = cpu_ldub_mmuidx_ra(env, fromaddr, rmemidx, ra);
1350         }
1351         if (wmem) {
1352             *(uint8_t *)wmem = byte;
1353         } else {
1354             cpu_stb_mmuidx_ra(env, toaddr, byte, wmemidx, ra);
1355         }
1356         return 1;
1357     }
1358 #endif
1359     /* Easy case: just memmove the host memory */
1360     set_helper_retaddr(ra);
1361     memmove(wmem, rmem, copysize);
1362     clear_helper_retaddr();
1363     return copysize;
1364 }
1365 
1366 /*
1367  * Do part of a backwards memory copy. Here toaddr and fromaddr point
1368  * to the *last* byte to be copied.
1369  */
1370 static uint64_t copy_step_rev(CPUARMState *env, uint64_t toaddr,
1371                               uint64_t fromaddr,
1372                               uint64_t copysize, int wmemidx, int rmemidx,
1373                               uint32_t *wdesc, uint32_t *rdesc, uintptr_t ra)
1374 {
1375     void *rmem;
1376     void *wmem;
1377 
1378     /* Don't cross a page boundary on either source or destination */
1379     copysize = MIN(copysize, page_limit_rev(toaddr));
1380     copysize = MIN(copysize, page_limit_rev(fromaddr));
1381 
1382     /*
1383      * Handle MTE tag checks: either handle the tag mismatch for byte 0,
1384      * or else copy up to but not including the byte with the mismatch.
1385      */
1386     if (*rdesc) {
1387         uint64_t mtesize = mte_mops_probe_rev(env, fromaddr, copysize, *rdesc);
1388         if (mtesize == 0) {
1389             mte_check_fail(env, *rdesc, fromaddr, ra);
1390             *rdesc = 0;
1391         } else {
1392             copysize = MIN(copysize, mtesize);
1393         }
1394     }
1395     if (*wdesc) {
1396         uint64_t mtesize = mte_mops_probe_rev(env, toaddr, copysize, *wdesc);
1397         if (mtesize == 0) {
1398             mte_check_fail(env, *wdesc, toaddr, ra);
1399             *wdesc = 0;
1400         } else {
1401             copysize = MIN(copysize, mtesize);
1402         }
1403     }
1404 
1405     toaddr = useronly_clean_ptr(toaddr);
1406     fromaddr = useronly_clean_ptr(fromaddr);
1407     /* Trapless lookup of whether we can get a host memory pointer */
1408     wmem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, wmemidx);
1409     rmem = tlb_vaddr_to_host(env, fromaddr, MMU_DATA_LOAD, rmemidx);
1410 
1411 #ifndef CONFIG_USER_ONLY
1412     /*
1413      * If we don't have host memory for both source and dest then just
1414      * do a single byte copy. This will handle watchpoints, invalid pages,
1415      * etc correctly. For clean code pages, the next iteration will see
1416      * the page dirty and will use the fast path.
1417      */
1418     if (unlikely(!rmem || !wmem)) {
1419         uint8_t byte;
1420         if (rmem) {
1421             byte = *(uint8_t *)rmem;
1422         } else {
1423             byte = cpu_ldub_mmuidx_ra(env, fromaddr, rmemidx, ra);
1424         }
1425         if (wmem) {
1426             *(uint8_t *)wmem = byte;
1427         } else {
1428             cpu_stb_mmuidx_ra(env, toaddr, byte, wmemidx, ra);
1429         }
1430         return 1;
1431     }
1432 #endif
1433     /*
1434      * Easy case: just memmove the host memory. Note that wmem and
1435      * rmem here point to the *last* byte to copy.
1436      */
1437     set_helper_retaddr(ra);
1438     memmove(wmem - (copysize - 1), rmem - (copysize - 1), copysize);
1439     clear_helper_retaddr();
1440     return copysize;
1441 }
1442 
1443 /*
1444  * for the Memory Copy operation, our implementation chooses always
1445  * to use "option A", where we update Xd and Xs to the final addresses
1446  * in the CPYP insn, and then in CPYM and CPYE only need to update Xn.
1447  *
1448  * @env: CPU
1449  * @syndrome: syndrome value for mismatch exceptions
1450  * (also contains the register numbers we need to use)
1451  * @wdesc: MTE descriptor for the writes (destination)
1452  * @rdesc: MTE descriptor for the reads (source)
1453  * @move: true if this is CPY (memmove), false for CPYF (memcpy forwards)
1454  */
1455 static void do_cpyp(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1456                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1457 {
1458     int rd = mops_destreg(syndrome);
1459     int rs = mops_srcreg(syndrome);
1460     int rn = mops_sizereg(syndrome);
1461     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1462     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1463     bool forwards = true;
1464     uint64_t toaddr = env->xregs[rd];
1465     uint64_t fromaddr = env->xregs[rs];
1466     uint64_t copysize = env->xregs[rn];
1467     uint64_t stagecopysize, step;
1468 
1469     check_mops_enabled(env, ra);
1470 
1471 
1472     if (move) {
1473         /*
1474          * Copy backwards if necessary. The direction for a non-overlapping
1475          * copy is IMPDEF; we choose forwards.
1476          */
1477         if (copysize > 0x007FFFFFFFFFFFFFULL) {
1478             copysize = 0x007FFFFFFFFFFFFFULL;
1479         }
1480         uint64_t fs = extract64(fromaddr, 0, 56);
1481         uint64_t ts = extract64(toaddr, 0, 56);
1482         uint64_t fe = extract64(fromaddr + copysize, 0, 56);
1483 
1484         if (fs < ts && fe > ts) {
1485             forwards = false;
1486         }
1487     } else {
1488         if (copysize > INT64_MAX) {
1489             copysize = INT64_MAX;
1490         }
1491     }
1492 
1493     if (!mte_checks_needed(fromaddr, rdesc)) {
1494         rdesc = 0;
1495     }
1496     if (!mte_checks_needed(toaddr, wdesc)) {
1497         wdesc = 0;
1498     }
1499 
1500     if (forwards) {
1501         stagecopysize = MIN(copysize, page_limit(toaddr));
1502         stagecopysize = MIN(stagecopysize, page_limit(fromaddr));
1503         while (stagecopysize) {
1504             env->xregs[rd] = toaddr;
1505             env->xregs[rs] = fromaddr;
1506             env->xregs[rn] = copysize;
1507             step = copy_step(env, toaddr, fromaddr, stagecopysize,
1508                              wmemidx, rmemidx, &wdesc, &rdesc, ra);
1509             toaddr += step;
1510             fromaddr += step;
1511             copysize -= step;
1512             stagecopysize -= step;
1513         }
1514         /* Insn completed, so update registers to the Option A format */
1515         env->xregs[rd] = toaddr + copysize;
1516         env->xregs[rs] = fromaddr + copysize;
1517         env->xregs[rn] = -copysize;
1518     } else {
1519         /*
1520          * In a reverse copy the to and from addrs in Xs and Xd are the start
1521          * of the range, but it's more convenient for us to work with pointers
1522          * to the last byte being copied.
1523          */
1524         toaddr += copysize - 1;
1525         fromaddr += copysize - 1;
1526         stagecopysize = MIN(copysize, page_limit_rev(toaddr));
1527         stagecopysize = MIN(stagecopysize, page_limit_rev(fromaddr));
1528         while (stagecopysize) {
1529             env->xregs[rn] = copysize;
1530             step = copy_step_rev(env, toaddr, fromaddr, stagecopysize,
1531                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1532             copysize -= step;
1533             stagecopysize -= step;
1534             toaddr -= step;
1535             fromaddr -= step;
1536         }
1537         /*
1538          * Insn completed, so update registers to the Option A format.
1539          * For a reverse copy this is no different to the CPYP input format.
1540          */
1541         env->xregs[rn] = copysize;
1542     }
1543 
1544     /* Set NZCV = 0000 to indicate we are an Option A implementation */
1545     env->NF = 0;
1546     env->ZF = 1; /* our env->ZF encoding is inverted */
1547     env->CF = 0;
1548     env->VF = 0;
1549 }
1550 
1551 void HELPER(cpyp)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1552                   uint32_t rdesc)
1553 {
1554     do_cpyp(env, syndrome, wdesc, rdesc, true, GETPC());
1555 }
1556 
1557 void HELPER(cpyfp)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1558                    uint32_t rdesc)
1559 {
1560     do_cpyp(env, syndrome, wdesc, rdesc, false, GETPC());
1561 }
1562 
1563 static void do_cpym(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1564                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1565 {
1566     /* Main: we choose to copy until less than a page remaining */
1567     CPUState *cs = env_cpu(env);
1568     int rd = mops_destreg(syndrome);
1569     int rs = mops_srcreg(syndrome);
1570     int rn = mops_sizereg(syndrome);
1571     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1572     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1573     bool forwards = true;
1574     uint64_t toaddr, fromaddr, copysize, step;
1575 
1576     check_mops_enabled(env, ra);
1577 
1578     /* We choose to NOP out "no data to copy" before consistency checks */
1579     if (env->xregs[rn] == 0) {
1580         return;
1581     }
1582 
1583     check_mops_wrong_option(env, syndrome, ra);
1584 
1585     if (move) {
1586         forwards = (int64_t)env->xregs[rn] < 0;
1587     }
1588 
1589     if (forwards) {
1590         toaddr = env->xregs[rd] + env->xregs[rn];
1591         fromaddr = env->xregs[rs] + env->xregs[rn];
1592         copysize = -env->xregs[rn];
1593     } else {
1594         copysize = env->xregs[rn];
1595         /* This toaddr and fromaddr point to the *last* byte to copy */
1596         toaddr = env->xregs[rd] + copysize - 1;
1597         fromaddr = env->xregs[rs] + copysize - 1;
1598     }
1599 
1600     if (!mte_checks_needed(fromaddr, rdesc)) {
1601         rdesc = 0;
1602     }
1603     if (!mte_checks_needed(toaddr, wdesc)) {
1604         wdesc = 0;
1605     }
1606 
1607     /* Our implementation has no particular parameter requirements for CPYM */
1608 
1609     /* Do the actual memmove */
1610     if (forwards) {
1611         while (copysize >= TARGET_PAGE_SIZE) {
1612             step = copy_step(env, toaddr, fromaddr, copysize,
1613                              wmemidx, rmemidx, &wdesc, &rdesc, ra);
1614             toaddr += step;
1615             fromaddr += step;
1616             copysize -= step;
1617             env->xregs[rn] = -copysize;
1618             if (copysize >= TARGET_PAGE_SIZE &&
1619                 unlikely(cpu_loop_exit_requested(cs))) {
1620                 cpu_loop_exit_restore(cs, ra);
1621             }
1622         }
1623     } else {
1624         while (copysize >= TARGET_PAGE_SIZE) {
1625             step = copy_step_rev(env, toaddr, fromaddr, copysize,
1626                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1627             toaddr -= step;
1628             fromaddr -= step;
1629             copysize -= step;
1630             env->xregs[rn] = copysize;
1631             if (copysize >= TARGET_PAGE_SIZE &&
1632                 unlikely(cpu_loop_exit_requested(cs))) {
1633                 cpu_loop_exit_restore(cs, ra);
1634             }
1635         }
1636     }
1637 }
1638 
1639 void HELPER(cpym)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1640                   uint32_t rdesc)
1641 {
1642     do_cpym(env, syndrome, wdesc, rdesc, true, GETPC());
1643 }
1644 
1645 void HELPER(cpyfm)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1646                    uint32_t rdesc)
1647 {
1648     do_cpym(env, syndrome, wdesc, rdesc, false, GETPC());
1649 }
1650 
1651 static void do_cpye(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1652                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1653 {
1654     /* Epilogue: do the last partial page */
1655     int rd = mops_destreg(syndrome);
1656     int rs = mops_srcreg(syndrome);
1657     int rn = mops_sizereg(syndrome);
1658     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1659     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1660     bool forwards = true;
1661     uint64_t toaddr, fromaddr, copysize, step;
1662 
1663     check_mops_enabled(env, ra);
1664 
1665     /* We choose to NOP out "no data to copy" before consistency checks */
1666     if (env->xregs[rn] == 0) {
1667         return;
1668     }
1669 
1670     check_mops_wrong_option(env, syndrome, ra);
1671 
1672     if (move) {
1673         forwards = (int64_t)env->xregs[rn] < 0;
1674     }
1675 
1676     if (forwards) {
1677         toaddr = env->xregs[rd] + env->xregs[rn];
1678         fromaddr = env->xregs[rs] + env->xregs[rn];
1679         copysize = -env->xregs[rn];
1680     } else {
1681         copysize = env->xregs[rn];
1682         /* This toaddr and fromaddr point to the *last* byte to copy */
1683         toaddr = env->xregs[rd] + copysize - 1;
1684         fromaddr = env->xregs[rs] + copysize - 1;
1685     }
1686 
1687     if (!mte_checks_needed(fromaddr, rdesc)) {
1688         rdesc = 0;
1689     }
1690     if (!mte_checks_needed(toaddr, wdesc)) {
1691         wdesc = 0;
1692     }
1693 
1694     /* Check the size; we don't want to have do a check-for-interrupts */
1695     if (copysize >= TARGET_PAGE_SIZE) {
1696         raise_exception_ra(env, EXCP_UDEF, syndrome,
1697                            mops_mismatch_exception_target_el(env), ra);
1698     }
1699 
1700     /* Do the actual memmove */
1701     if (forwards) {
1702         while (copysize > 0) {
1703             step = copy_step(env, toaddr, fromaddr, copysize,
1704                              wmemidx, rmemidx, &wdesc, &rdesc, ra);
1705             toaddr += step;
1706             fromaddr += step;
1707             copysize -= step;
1708             env->xregs[rn] = -copysize;
1709         }
1710     } else {
1711         while (copysize > 0) {
1712             step = copy_step_rev(env, toaddr, fromaddr, copysize,
1713                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1714             toaddr -= step;
1715             fromaddr -= step;
1716             copysize -= step;
1717             env->xregs[rn] = copysize;
1718         }
1719     }
1720 }
1721 
1722 void HELPER(cpye)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1723                   uint32_t rdesc)
1724 {
1725     do_cpye(env, syndrome, wdesc, rdesc, true, GETPC());
1726 }
1727 
1728 void HELPER(cpyfe)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1729                    uint32_t rdesc)
1730 {
1731     do_cpye(env, syndrome, wdesc, rdesc, false, GETPC());
1732 }
1733 
1734 static bool is_guarded_page(CPUARMState *env, target_ulong addr, uintptr_t ra)
1735 {
1736 #ifdef CONFIG_USER_ONLY
1737     return page_get_flags(addr) & PAGE_BTI;
1738 #else
1739     CPUTLBEntryFull *full;
1740     void *host;
1741     int mmu_idx = cpu_mmu_index(env_cpu(env), true);
1742     int flags = probe_access_full(env, addr, 0, MMU_INST_FETCH, mmu_idx,
1743                                   false, &host, &full, ra);
1744 
1745     assert(!(flags & TLB_INVALID_MASK));
1746     return full->extra.arm.guarded;
1747 #endif
1748 }
1749 
1750 void HELPER(guarded_page_check)(CPUARMState *env)
1751 {
1752     /*
1753      * We have already verified that bti is enabled, and that the
1754      * instruction at PC is not ok for BTYPE.  This is always at
1755      * the beginning of a block, so PC is always up-to-date and
1756      * no unwind is required.
1757      */
1758     if (is_guarded_page(env, env->pc, 0)) {
1759         raise_exception(env, EXCP_UDEF, syn_btitrap(env->btype),
1760                         exception_target_el(env));
1761     }
1762 }
1763 
1764 void HELPER(guarded_page_br)(CPUARMState *env, target_ulong pc)
1765 {
1766     /*
1767      * We have already checked for branch via x16 and x17.
1768      * What remains for choosing BTYPE is checking for a guarded page.
1769      */
1770     env->btype = is_guarded_page(env, pc, GETPC()) ? 3 : 1;
1771 }
1772