xref: /openbmc/qemu/target/arm/tcg/helper-a64.c (revision 5f69e42d)
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/exec-all.h"
32 #include "exec/cpu_ldst.h"
33 #include "qemu/int128.h"
34 #include "qemu/atomic128.h"
35 #include "fpu/softfloat.h"
36 #include <zlib.h> /* For crc32 */
37 
38 /* C2.4.7 Multiply and divide */
39 /* special cases for 0 and LLONG_MIN are mandated by the standard */
40 uint64_t HELPER(udiv64)(uint64_t num, uint64_t den)
41 {
42     if (den == 0) {
43         return 0;
44     }
45     return num / den;
46 }
47 
48 int64_t HELPER(sdiv64)(int64_t num, int64_t den)
49 {
50     if (den == 0) {
51         return 0;
52     }
53     if (num == LLONG_MIN && den == -1) {
54         return LLONG_MIN;
55     }
56     return num / den;
57 }
58 
59 uint64_t HELPER(rbit64)(uint64_t x)
60 {
61     return revbit64(x);
62 }
63 
64 void HELPER(msr_i_spsel)(CPUARMState *env, uint32_t imm)
65 {
66     update_spsel(env, imm);
67 }
68 
69 void HELPER(msr_set_allint_el1)(CPUARMState *env)
70 {
71     /* ALLINT update to PSTATE. */
72     if (arm_hcrx_el2_eff(env) & HCRX_TALLINT) {
73         raise_exception_ra(env, EXCP_UDEF,
74                            syn_aa64_sysregtrap(0, 1, 0, 4, 1, 0x1f, 0), 2,
75                            GETPC());
76     }
77 
78     env->pstate |= PSTATE_ALLINT;
79 }
80 
81 static void daif_check(CPUARMState *env, uint32_t op,
82                        uint32_t imm, uintptr_t ra)
83 {
84     /* DAIF update to PSTATE. This is OK from EL0 only if UMA is set.  */
85     if (arm_current_el(env) == 0 && !(arm_sctlr(env, 0) & SCTLR_UMA)) {
86         raise_exception_ra(env, EXCP_UDEF,
87                            syn_aa64_sysregtrap(0, extract32(op, 0, 3),
88                                                extract32(op, 3, 3), 4,
89                                                imm, 0x1f, 0),
90                            exception_target_el(env), ra);
91     }
92 }
93 
94 void HELPER(msr_i_daifset)(CPUARMState *env, uint32_t imm)
95 {
96     daif_check(env, 0x1e, imm, GETPC());
97     env->daif |= (imm << 6) & PSTATE_DAIF;
98     arm_rebuild_hflags(env);
99 }
100 
101 void HELPER(msr_i_daifclear)(CPUARMState *env, uint32_t imm)
102 {
103     daif_check(env, 0x1f, imm, GETPC());
104     env->daif &= ~((imm << 6) & PSTATE_DAIF);
105     arm_rebuild_hflags(env);
106 }
107 
108 /* Convert a softfloat float_relation_ (as returned by
109  * the float*_compare functions) to the correct ARM
110  * NZCV flag state.
111  */
112 static inline uint32_t float_rel_to_flags(int res)
113 {
114     uint64_t flags;
115     switch (res) {
116     case float_relation_equal:
117         flags = PSTATE_Z | PSTATE_C;
118         break;
119     case float_relation_less:
120         flags = PSTATE_N;
121         break;
122     case float_relation_greater:
123         flags = PSTATE_C;
124         break;
125     case float_relation_unordered:
126     default:
127         flags = PSTATE_C | PSTATE_V;
128         break;
129     }
130     return flags;
131 }
132 
133 uint64_t HELPER(vfp_cmph_a64)(uint32_t x, uint32_t y, void *fp_status)
134 {
135     return float_rel_to_flags(float16_compare_quiet(x, y, fp_status));
136 }
137 
138 uint64_t HELPER(vfp_cmpeh_a64)(uint32_t x, uint32_t y, void *fp_status)
139 {
140     return float_rel_to_flags(float16_compare(x, y, fp_status));
141 }
142 
143 uint64_t HELPER(vfp_cmps_a64)(float32 x, float32 y, void *fp_status)
144 {
145     return float_rel_to_flags(float32_compare_quiet(x, y, fp_status));
146 }
147 
148 uint64_t HELPER(vfp_cmpes_a64)(float32 x, float32 y, void *fp_status)
149 {
150     return float_rel_to_flags(float32_compare(x, y, fp_status));
151 }
152 
153 uint64_t HELPER(vfp_cmpd_a64)(float64 x, float64 y, void *fp_status)
154 {
155     return float_rel_to_flags(float64_compare_quiet(x, y, fp_status));
156 }
157 
158 uint64_t HELPER(vfp_cmped_a64)(float64 x, float64 y, void *fp_status)
159 {
160     return float_rel_to_flags(float64_compare(x, y, fp_status));
161 }
162 
163 float32 HELPER(vfp_mulxs)(float32 a, float32 b, void *fpstp)
164 {
165     float_status *fpst = fpstp;
166 
167     a = float32_squash_input_denormal(a, fpst);
168     b = float32_squash_input_denormal(b, fpst);
169 
170     if ((float32_is_zero(a) && float32_is_infinity(b)) ||
171         (float32_is_infinity(a) && float32_is_zero(b))) {
172         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
173         return make_float32((1U << 30) |
174                             ((float32_val(a) ^ float32_val(b)) & (1U << 31)));
175     }
176     return float32_mul(a, b, fpst);
177 }
178 
179 float64 HELPER(vfp_mulxd)(float64 a, float64 b, void *fpstp)
180 {
181     float_status *fpst = fpstp;
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, void *fpstp)
197 {
198     float_status *fpst = fpstp;
199     return -float64_eq_quiet(a, b, fpst);
200 }
201 
202 uint64_t HELPER(neon_cge_f64)(float64 a, float64 b, void *fpstp)
203 {
204     float_status *fpst = fpstp;
205     return -float64_le(b, a, fpst);
206 }
207 
208 uint64_t HELPER(neon_cgt_f64)(float64 a, float64 b, void *fpstp)
209 {
210     float_status *fpst = fpstp;
211     return -float64_lt(b, a, fpst);
212 }
213 
214 /* Reciprocal step and sqrt step. Note that unlike the A32/T32
215  * versions, these do a fully fused multiply-add or
216  * multiply-add-and-halve.
217  */
218 
219 uint32_t HELPER(recpsf_f16)(uint32_t a, uint32_t b, void *fpstp)
220 {
221     float_status *fpst = fpstp;
222 
223     a = float16_squash_input_denormal(a, fpst);
224     b = float16_squash_input_denormal(b, fpst);
225 
226     a = float16_chs(a);
227     if ((float16_is_infinity(a) && float16_is_zero(b)) ||
228         (float16_is_infinity(b) && float16_is_zero(a))) {
229         return float16_two;
230     }
231     return float16_muladd(a, b, float16_two, 0, fpst);
232 }
233 
234 float32 HELPER(recpsf_f32)(float32 a, float32 b, void *fpstp)
235 {
236     float_status *fpst = fpstp;
237 
238     a = float32_squash_input_denormal(a, fpst);
239     b = float32_squash_input_denormal(b, fpst);
240 
241     a = float32_chs(a);
242     if ((float32_is_infinity(a) && float32_is_zero(b)) ||
243         (float32_is_infinity(b) && float32_is_zero(a))) {
244         return float32_two;
245     }
246     return float32_muladd(a, b, float32_two, 0, fpst);
247 }
248 
249 float64 HELPER(recpsf_f64)(float64 a, float64 b, void *fpstp)
250 {
251     float_status *fpst = fpstp;
252 
253     a = float64_squash_input_denormal(a, fpst);
254     b = float64_squash_input_denormal(b, fpst);
255 
256     a = float64_chs(a);
257     if ((float64_is_infinity(a) && float64_is_zero(b)) ||
258         (float64_is_infinity(b) && float64_is_zero(a))) {
259         return float64_two;
260     }
261     return float64_muladd(a, b, float64_two, 0, fpst);
262 }
263 
264 uint32_t HELPER(rsqrtsf_f16)(uint32_t a, uint32_t b, void *fpstp)
265 {
266     float_status *fpst = fpstp;
267 
268     a = float16_squash_input_denormal(a, fpst);
269     b = float16_squash_input_denormal(b, fpst);
270 
271     a = float16_chs(a);
272     if ((float16_is_infinity(a) && float16_is_zero(b)) ||
273         (float16_is_infinity(b) && float16_is_zero(a))) {
274         return float16_one_point_five;
275     }
276     return float16_muladd(a, b, float16_three, float_muladd_halve_result, fpst);
277 }
278 
279 float32 HELPER(rsqrtsf_f32)(float32 a, float32 b, void *fpstp)
280 {
281     float_status *fpst = fpstp;
282 
283     a = float32_squash_input_denormal(a, fpst);
284     b = float32_squash_input_denormal(b, fpst);
285 
286     a = float32_chs(a);
287     if ((float32_is_infinity(a) && float32_is_zero(b)) ||
288         (float32_is_infinity(b) && float32_is_zero(a))) {
289         return float32_one_point_five;
290     }
291     return float32_muladd(a, b, float32_three, float_muladd_halve_result, fpst);
292 }
293 
294 float64 HELPER(rsqrtsf_f64)(float64 a, float64 b, void *fpstp)
295 {
296     float_status *fpst = fpstp;
297 
298     a = float64_squash_input_denormal(a, fpst);
299     b = float64_squash_input_denormal(b, fpst);
300 
301     a = float64_chs(a);
302     if ((float64_is_infinity(a) && float64_is_zero(b)) ||
303         (float64_is_infinity(b) && float64_is_zero(a))) {
304         return float64_one_point_five;
305     }
306     return float64_muladd(a, b, float64_three, float_muladd_halve_result, fpst);
307 }
308 
309 /* Pairwise long add: add pairs of adjacent elements into
310  * double-width elements in the result (eg _s8 is an 8x8->16 op)
311  */
312 uint64_t HELPER(neon_addlp_s8)(uint64_t a)
313 {
314     uint64_t nsignmask = 0x0080008000800080ULL;
315     uint64_t wsignmask = 0x8000800080008000ULL;
316     uint64_t elementmask = 0x00ff00ff00ff00ffULL;
317     uint64_t tmp1, tmp2;
318     uint64_t res, signres;
319 
320     /* Extract odd elements, sign extend each to a 16 bit field */
321     tmp1 = a & elementmask;
322     tmp1 ^= nsignmask;
323     tmp1 |= wsignmask;
324     tmp1 = (tmp1 - nsignmask) ^ wsignmask;
325     /* Ditto for the even elements */
326     tmp2 = (a >> 8) & elementmask;
327     tmp2 ^= nsignmask;
328     tmp2 |= wsignmask;
329     tmp2 = (tmp2 - nsignmask) ^ wsignmask;
330 
331     /* calculate the result by summing bits 0..14, 16..22, etc,
332      * and then adjusting the sign bits 15, 23, etc manually.
333      * This ensures the addition can't overflow the 16 bit field.
334      */
335     signres = (tmp1 ^ tmp2) & wsignmask;
336     res = (tmp1 & ~wsignmask) + (tmp2 & ~wsignmask);
337     res ^= signres;
338 
339     return res;
340 }
341 
342 uint64_t HELPER(neon_addlp_u8)(uint64_t a)
343 {
344     uint64_t tmp;
345 
346     tmp = a & 0x00ff00ff00ff00ffULL;
347     tmp += (a >> 8) & 0x00ff00ff00ff00ffULL;
348     return tmp;
349 }
350 
351 uint64_t HELPER(neon_addlp_s16)(uint64_t a)
352 {
353     int32_t reslo, reshi;
354 
355     reslo = (int32_t)(int16_t)a + (int32_t)(int16_t)(a >> 16);
356     reshi = (int32_t)(int16_t)(a >> 32) + (int32_t)(int16_t)(a >> 48);
357 
358     return (uint32_t)reslo | (((uint64_t)reshi) << 32);
359 }
360 
361 uint64_t HELPER(neon_addlp_u16)(uint64_t a)
362 {
363     uint64_t tmp;
364 
365     tmp = a & 0x0000ffff0000ffffULL;
366     tmp += (a >> 16) & 0x0000ffff0000ffffULL;
367     return tmp;
368 }
369 
370 /* Floating-point reciprocal exponent - see FPRecpX in ARM ARM */
371 uint32_t HELPER(frecpx_f16)(uint32_t a, void *fpstp)
372 {
373     float_status *fpst = fpstp;
374     uint16_t val16, sbit;
375     int16_t exp;
376 
377     if (float16_is_any_nan(a)) {
378         float16 nan = a;
379         if (float16_is_signaling_nan(a, fpst)) {
380             float_raise(float_flag_invalid, fpst);
381             if (!fpst->default_nan_mode) {
382                 nan = float16_silence_nan(a, fpst);
383             }
384         }
385         if (fpst->default_nan_mode) {
386             nan = float16_default_nan(fpst);
387         }
388         return nan;
389     }
390 
391     a = float16_squash_input_denormal(a, fpst);
392 
393     val16 = float16_val(a);
394     sbit = 0x8000 & val16;
395     exp = extract32(val16, 10, 5);
396 
397     if (exp == 0) {
398         return make_float16(deposit32(sbit, 10, 5, 0x1e));
399     } else {
400         return make_float16(deposit32(sbit, 10, 5, ~exp));
401     }
402 }
403 
404 float32 HELPER(frecpx_f32)(float32 a, void *fpstp)
405 {
406     float_status *fpst = fpstp;
407     uint32_t val32, sbit;
408     int32_t exp;
409 
410     if (float32_is_any_nan(a)) {
411         float32 nan = a;
412         if (float32_is_signaling_nan(a, fpst)) {
413             float_raise(float_flag_invalid, fpst);
414             if (!fpst->default_nan_mode) {
415                 nan = float32_silence_nan(a, fpst);
416             }
417         }
418         if (fpst->default_nan_mode) {
419             nan = float32_default_nan(fpst);
420         }
421         return nan;
422     }
423 
424     a = float32_squash_input_denormal(a, fpst);
425 
426     val32 = float32_val(a);
427     sbit = 0x80000000ULL & val32;
428     exp = extract32(val32, 23, 8);
429 
430     if (exp == 0) {
431         return make_float32(sbit | (0xfe << 23));
432     } else {
433         return make_float32(sbit | (~exp & 0xff) << 23);
434     }
435 }
436 
437 float64 HELPER(frecpx_f64)(float64 a, void *fpstp)
438 {
439     float_status *fpst = fpstp;
440     uint64_t val64, sbit;
441     int64_t exp;
442 
443     if (float64_is_any_nan(a)) {
444         float64 nan = a;
445         if (float64_is_signaling_nan(a, fpst)) {
446             float_raise(float_flag_invalid, fpst);
447             if (!fpst->default_nan_mode) {
448                 nan = float64_silence_nan(a, fpst);
449             }
450         }
451         if (fpst->default_nan_mode) {
452             nan = float64_default_nan(fpst);
453         }
454         return nan;
455     }
456 
457     a = float64_squash_input_denormal(a, fpst);
458 
459     val64 = float64_val(a);
460     sbit = 0x8000000000000000ULL & val64;
461     exp = extract64(float64_val(a), 52, 11);
462 
463     if (exp == 0) {
464         return make_float64(sbit | (0x7feULL << 52));
465     } else {
466         return make_float64(sbit | (~exp & 0x7ffULL) << 52);
467     }
468 }
469 
470 float32 HELPER(fcvtx_f64_to_f32)(float64 a, CPUARMState *env)
471 {
472     /* Von Neumann rounding is implemented by using round-to-zero
473      * and then setting the LSB of the result if Inexact was raised.
474      */
475     float32 r;
476     float_status *fpst = &env->vfp.fp_status;
477     float_status tstat = *fpst;
478     int exflags;
479 
480     set_float_rounding_mode(float_round_to_zero, &tstat);
481     set_float_exception_flags(0, &tstat);
482     r = float64_to_float32(a, &tstat);
483     exflags = get_float_exception_flags(&tstat);
484     if (exflags & float_flag_inexact) {
485         r = make_float32(float32_val(r) | 1);
486     }
487     exflags |= get_float_exception_flags(fpst);
488     set_float_exception_flags(exflags, fpst);
489     return r;
490 }
491 
492 /* 64-bit versions of the CRC helpers. Note that although the operation
493  * (and the prototypes of crc32c() and crc32() mean that only the bottom
494  * 32 bits of the accumulator and result are used, we pass and return
495  * uint64_t for convenience of the generated code. Unlike the 32-bit
496  * instruction set versions, val may genuinely have 64 bits of data in it.
497  * The upper bytes of val (above the number specified by 'bytes') must have
498  * been zeroed out by the caller.
499  */
500 uint64_t HELPER(crc32_64)(uint64_t acc, uint64_t val, uint32_t bytes)
501 {
502     uint8_t buf[8];
503 
504     stq_le_p(buf, val);
505 
506     /* zlib crc32 converts the accumulator and output to one's complement.  */
507     return crc32(acc ^ 0xffffffff, buf, bytes) ^ 0xffffffff;
508 }
509 
510 uint64_t HELPER(crc32c_64)(uint64_t acc, uint64_t val, uint32_t bytes)
511 {
512     uint8_t buf[8];
513 
514     stq_le_p(buf, val);
515 
516     /* Linux crc32c converts the output to one's complement.  */
517     return crc32c(acc, buf, bytes) ^ 0xffffffff;
518 }
519 
520 /*
521  * AdvSIMD half-precision
522  */
523 
524 #define ADVSIMD_HELPER(name, suffix) HELPER(glue(glue(advsimd_, name), suffix))
525 
526 #define ADVSIMD_HALFOP(name) \
527 uint32_t ADVSIMD_HELPER(name, h)(uint32_t a, uint32_t b, void *fpstp) \
528 { \
529     float_status *fpst = fpstp; \
530     return float16_ ## name(a, b, fpst);    \
531 }
532 
533 ADVSIMD_HALFOP(add)
534 ADVSIMD_HALFOP(sub)
535 ADVSIMD_HALFOP(mul)
536 ADVSIMD_HALFOP(div)
537 ADVSIMD_HALFOP(min)
538 ADVSIMD_HALFOP(max)
539 ADVSIMD_HALFOP(minnum)
540 ADVSIMD_HALFOP(maxnum)
541 
542 #define ADVSIMD_TWOHALFOP(name)                                         \
543 uint32_t ADVSIMD_HELPER(name, 2h)(uint32_t two_a, uint32_t two_b, void *fpstp) \
544 { \
545     float16  a1, a2, b1, b2;                        \
546     uint32_t r1, r2;                                \
547     float_status *fpst = fpstp;                     \
548     a1 = extract32(two_a, 0, 16);                   \
549     a2 = extract32(two_a, 16, 16);                  \
550     b1 = extract32(two_b, 0, 16);                   \
551     b2 = extract32(two_b, 16, 16);                  \
552     r1 = float16_ ## name(a1, b1, fpst);            \
553     r2 = float16_ ## name(a2, b2, fpst);            \
554     return deposit32(r1, 16, 16, r2);               \
555 }
556 
557 ADVSIMD_TWOHALFOP(add)
558 ADVSIMD_TWOHALFOP(sub)
559 ADVSIMD_TWOHALFOP(mul)
560 ADVSIMD_TWOHALFOP(div)
561 ADVSIMD_TWOHALFOP(min)
562 ADVSIMD_TWOHALFOP(max)
563 ADVSIMD_TWOHALFOP(minnum)
564 ADVSIMD_TWOHALFOP(maxnum)
565 
566 /* Data processing - scalar floating-point and advanced SIMD */
567 static float16 float16_mulx(float16 a, float16 b, void *fpstp)
568 {
569     float_status *fpst = fpstp;
570 
571     a = float16_squash_input_denormal(a, fpst);
572     b = float16_squash_input_denormal(b, fpst);
573 
574     if ((float16_is_zero(a) && float16_is_infinity(b)) ||
575         (float16_is_infinity(a) && float16_is_zero(b))) {
576         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
577         return make_float16((1U << 14) |
578                             ((float16_val(a) ^ float16_val(b)) & (1U << 15)));
579     }
580     return float16_mul(a, b, fpst);
581 }
582 
583 ADVSIMD_HALFOP(mulx)
584 ADVSIMD_TWOHALFOP(mulx)
585 
586 /* fused multiply-accumulate */
587 uint32_t HELPER(advsimd_muladdh)(uint32_t a, uint32_t b, uint32_t c,
588                                  void *fpstp)
589 {
590     float_status *fpst = fpstp;
591     return float16_muladd(a, b, c, 0, fpst);
592 }
593 
594 uint32_t HELPER(advsimd_muladd2h)(uint32_t two_a, uint32_t two_b,
595                                   uint32_t two_c, void *fpstp)
596 {
597     float_status *fpst = fpstp;
598     float16  a1, a2, b1, b2, c1, c2;
599     uint32_t r1, r2;
600     a1 = extract32(two_a, 0, 16);
601     a2 = extract32(two_a, 16, 16);
602     b1 = extract32(two_b, 0, 16);
603     b2 = extract32(two_b, 16, 16);
604     c1 = extract32(two_c, 0, 16);
605     c2 = extract32(two_c, 16, 16);
606     r1 = float16_muladd(a1, b1, c1, 0, fpst);
607     r2 = float16_muladd(a2, b2, c2, 0, fpst);
608     return deposit32(r1, 16, 16, r2);
609 }
610 
611 /*
612  * Floating point comparisons produce an integer result. Softfloat
613  * routines return float_relation types which we convert to the 0/-1
614  * Neon requires.
615  */
616 
617 #define ADVSIMD_CMPRES(test) (test) ? 0xffff : 0
618 
619 uint32_t HELPER(advsimd_ceq_f16)(uint32_t a, uint32_t b, void *fpstp)
620 {
621     float_status *fpst = fpstp;
622     int compare = float16_compare_quiet(a, b, fpst);
623     return ADVSIMD_CMPRES(compare == float_relation_equal);
624 }
625 
626 uint32_t HELPER(advsimd_cge_f16)(uint32_t a, uint32_t b, void *fpstp)
627 {
628     float_status *fpst = fpstp;
629     int compare = float16_compare(a, b, fpst);
630     return ADVSIMD_CMPRES(compare == float_relation_greater ||
631                           compare == float_relation_equal);
632 }
633 
634 uint32_t HELPER(advsimd_cgt_f16)(uint32_t a, uint32_t b, void *fpstp)
635 {
636     float_status *fpst = fpstp;
637     int compare = float16_compare(a, b, fpst);
638     return ADVSIMD_CMPRES(compare == float_relation_greater);
639 }
640 
641 uint32_t HELPER(advsimd_acge_f16)(uint32_t a, uint32_t b, void *fpstp)
642 {
643     float_status *fpst = fpstp;
644     float16 f0 = float16_abs(a);
645     float16 f1 = float16_abs(b);
646     int compare = float16_compare(f0, f1, fpst);
647     return ADVSIMD_CMPRES(compare == float_relation_greater ||
648                           compare == float_relation_equal);
649 }
650 
651 uint32_t HELPER(advsimd_acgt_f16)(uint32_t a, uint32_t b, void *fpstp)
652 {
653     float_status *fpst = fpstp;
654     float16 f0 = float16_abs(a);
655     float16 f1 = float16_abs(b);
656     int compare = float16_compare(f0, f1, fpst);
657     return ADVSIMD_CMPRES(compare == float_relation_greater);
658 }
659 
660 /* round to integral */
661 uint32_t HELPER(advsimd_rinth_exact)(uint32_t x, void *fp_status)
662 {
663     return float16_round_to_int(x, fp_status);
664 }
665 
666 uint32_t HELPER(advsimd_rinth)(uint32_t x, void *fp_status)
667 {
668     int old_flags = get_float_exception_flags(fp_status), new_flags;
669     float16 ret;
670 
671     ret = float16_round_to_int(x, fp_status);
672 
673     /* Suppress any inexact exceptions the conversion produced */
674     if (!(old_flags & float_flag_inexact)) {
675         new_flags = get_float_exception_flags(fp_status);
676         set_float_exception_flags(new_flags & ~float_flag_inexact, fp_status);
677     }
678 
679     return ret;
680 }
681 
682 /*
683  * Half-precision floating point conversion functions
684  *
685  * There are a multitude of conversion functions with various
686  * different rounding modes. This is dealt with by the calling code
687  * setting the mode appropriately before calling the helper.
688  */
689 
690 uint32_t HELPER(advsimd_f16tosinth)(uint32_t a, void *fpstp)
691 {
692     float_status *fpst = fpstp;
693 
694     /* Invalid if we are passed a NaN */
695     if (float16_is_any_nan(a)) {
696         float_raise(float_flag_invalid, fpst);
697         return 0;
698     }
699     return float16_to_int16(a, fpst);
700 }
701 
702 uint32_t HELPER(advsimd_f16touinth)(uint32_t a, void *fpstp)
703 {
704     float_status *fpst = fpstp;
705 
706     /* Invalid if we are passed a NaN */
707     if (float16_is_any_nan(a)) {
708         float_raise(float_flag_invalid, fpst);
709         return 0;
710     }
711     return float16_to_uint16(a, fpst);
712 }
713 
714 static int el_from_spsr(uint32_t spsr)
715 {
716     /* Return the exception level that this SPSR is requesting a return to,
717      * or -1 if it is invalid (an illegal return)
718      */
719     if (spsr & PSTATE_nRW) {
720         switch (spsr & CPSR_M) {
721         case ARM_CPU_MODE_USR:
722             return 0;
723         case ARM_CPU_MODE_HYP:
724             return 2;
725         case ARM_CPU_MODE_FIQ:
726         case ARM_CPU_MODE_IRQ:
727         case ARM_CPU_MODE_SVC:
728         case ARM_CPU_MODE_ABT:
729         case ARM_CPU_MODE_UND:
730         case ARM_CPU_MODE_SYS:
731             return 1;
732         case ARM_CPU_MODE_MON:
733             /* Returning to Mon from AArch64 is never possible,
734              * so this is an illegal return.
735              */
736         default:
737             return -1;
738         }
739     } else {
740         if (extract32(spsr, 1, 1)) {
741             /* Return with reserved M[1] bit set */
742             return -1;
743         }
744         if (extract32(spsr, 0, 4) == 1) {
745             /* return to EL0 with M[0] bit set */
746             return -1;
747         }
748         return extract32(spsr, 2, 2);
749     }
750 }
751 
752 static void cpsr_write_from_spsr_elx(CPUARMState *env,
753                                      uint32_t val)
754 {
755     uint32_t mask;
756 
757     /* Save SPSR_ELx.SS into PSTATE. */
758     env->pstate = (env->pstate & ~PSTATE_SS) | (val & PSTATE_SS);
759     val &= ~PSTATE_SS;
760 
761     /* Move DIT to the correct location for CPSR */
762     if (val & PSTATE_DIT) {
763         val &= ~PSTATE_DIT;
764         val |= CPSR_DIT;
765     }
766 
767     mask = aarch32_cpsr_valid_mask(env->features, \
768         &env_archcpu(env)->isar);
769     cpsr_write(env, val, mask, CPSRWriteRaw);
770 }
771 
772 void HELPER(exception_return)(CPUARMState *env, uint64_t new_pc)
773 {
774     int cur_el = arm_current_el(env);
775     unsigned int spsr_idx = aarch64_banked_spsr_index(cur_el);
776     uint32_t spsr = env->banked_spsr[spsr_idx];
777     int new_el;
778     bool return_to_aa64 = (spsr & PSTATE_nRW) == 0;
779 
780     aarch64_save_sp(env, cur_el);
781 
782     arm_clear_exclusive(env);
783 
784     /* We must squash the PSTATE.SS bit to zero unless both of the
785      * following hold:
786      *  1. debug exceptions are currently disabled
787      *  2. singlestep will be active in the EL we return to
788      * We check 1 here and 2 after we've done the pstate/cpsr write() to
789      * transition to the EL we're going to.
790      */
791     if (arm_generate_debug_exceptions(env)) {
792         spsr &= ~PSTATE_SS;
793     }
794 
795     /*
796      * FEAT_RME forbids return from EL3 with an invalid security state.
797      * We don't need an explicit check for FEAT_RME here because we enforce
798      * in scr_write() that you can't set the NSE bit without it.
799      */
800     if (cur_el == 3 && (env->cp15.scr_el3 & (SCR_NS | SCR_NSE)) == SCR_NSE) {
801         goto illegal_return;
802     }
803 
804     new_el = el_from_spsr(spsr);
805     if (new_el == -1) {
806         goto illegal_return;
807     }
808     if (new_el > cur_el || (new_el == 2 && !arm_is_el2_enabled(env))) {
809         /* Disallow return to an EL which is unimplemented or higher
810          * than the current one.
811          */
812         goto illegal_return;
813     }
814 
815     if (new_el != 0 && arm_el_is_aa64(env, new_el) != return_to_aa64) {
816         /* Return to an EL which is configured for a different register width */
817         goto illegal_return;
818     }
819 
820     if (new_el == 1 && (arm_hcr_el2_eff(env) & HCR_TGE)) {
821         goto illegal_return;
822     }
823 
824     bql_lock();
825     arm_call_pre_el_change_hook(env_archcpu(env));
826     bql_unlock();
827 
828     if (!return_to_aa64) {
829         env->aarch64 = false;
830         /* We do a raw CPSR write because aarch64_sync_64_to_32()
831          * will sort the register banks out for us, and we've already
832          * caught all the bad-mode cases in el_from_spsr().
833          */
834         cpsr_write_from_spsr_elx(env, spsr);
835         if (!arm_singlestep_active(env)) {
836             env->pstate &= ~PSTATE_SS;
837         }
838         aarch64_sync_64_to_32(env);
839 
840         if (spsr & CPSR_T) {
841             env->regs[15] = new_pc & ~0x1;
842         } else {
843             env->regs[15] = new_pc & ~0x3;
844         }
845         helper_rebuild_hflags_a32(env, new_el);
846         qemu_log_mask(CPU_LOG_INT, "Exception return from AArch64 EL%d to "
847                       "AArch32 EL%d PC 0x%" PRIx32 "\n",
848                       cur_el, new_el, env->regs[15]);
849     } else {
850         int tbii;
851 
852         env->aarch64 = true;
853         spsr &= aarch64_pstate_valid_mask(&env_archcpu(env)->isar);
854         pstate_write(env, spsr);
855         if (!arm_singlestep_active(env)) {
856             env->pstate &= ~PSTATE_SS;
857         }
858         aarch64_restore_sp(env, new_el);
859         helper_rebuild_hflags_a64(env, new_el);
860 
861         /*
862          * Apply TBI to the exception return address.  We had to delay this
863          * until after we selected the new EL, so that we could select the
864          * correct TBI+TBID bits.  This is made easier by waiting until after
865          * the hflags rebuild, since we can pull the composite TBII field
866          * from there.
867          */
868         tbii = EX_TBFLAG_A64(env->hflags, TBII);
869         if ((tbii >> extract64(new_pc, 55, 1)) & 1) {
870             /* TBI is enabled. */
871             int core_mmu_idx = arm_env_mmu_index(env);
872             if (regime_has_2_ranges(core_to_aa64_mmu_idx(core_mmu_idx))) {
873                 new_pc = sextract64(new_pc, 0, 56);
874             } else {
875                 new_pc = extract64(new_pc, 0, 56);
876             }
877         }
878         env->pc = new_pc;
879 
880         qemu_log_mask(CPU_LOG_INT, "Exception return from AArch64 EL%d to "
881                       "AArch64 EL%d PC 0x%" PRIx64 "\n",
882                       cur_el, new_el, env->pc);
883     }
884 
885     /*
886      * Note that cur_el can never be 0.  If new_el is 0, then
887      * el0_a64 is return_to_aa64, else el0_a64 is ignored.
888      */
889     aarch64_sve_change_el(env, cur_el, new_el, return_to_aa64);
890 
891     bql_lock();
892     arm_call_el_change_hook(env_archcpu(env));
893     bql_unlock();
894 
895     return;
896 
897 illegal_return:
898     /* Illegal return events of various kinds have architecturally
899      * mandated behaviour:
900      * restore NZCV and DAIF from SPSR_ELx
901      * set PSTATE.IL
902      * restore PC from ELR_ELx
903      * no change to exception level, execution state or stack pointer
904      */
905     env->pstate |= PSTATE_IL;
906     env->pc = new_pc;
907     spsr &= PSTATE_NZCV | PSTATE_DAIF | PSTATE_ALLINT;
908     spsr |= pstate_read(env) & ~(PSTATE_NZCV | PSTATE_DAIF | PSTATE_ALLINT);
909     pstate_write(env, spsr);
910     if (!arm_singlestep_active(env)) {
911         env->pstate &= ~PSTATE_SS;
912     }
913     helper_rebuild_hflags_a64(env, cur_el);
914     qemu_log_mask(LOG_GUEST_ERROR, "Illegal exception return at EL%d: "
915                   "resuming execution at 0x%" PRIx64 "\n", cur_el, env->pc);
916 }
917 
918 /*
919  * Square Root and Reciprocal square root
920  */
921 
922 uint32_t HELPER(sqrt_f16)(uint32_t a, void *fpstp)
923 {
924     float_status *s = fpstp;
925 
926     return float16_sqrt(a, s);
927 }
928 
929 void HELPER(dc_zva)(CPUARMState *env, uint64_t vaddr_in)
930 {
931     /*
932      * Implement DC ZVA, which zeroes a fixed-length block of memory.
933      * Note that we do not implement the (architecturally mandated)
934      * alignment fault for attempts to use this on Device memory
935      * (which matches the usual QEMU behaviour of not implementing either
936      * alignment faults or any memory attribute handling).
937      */
938     int blocklen = 4 << env_archcpu(env)->dcz_blocksize;
939     uint64_t vaddr = vaddr_in & ~(blocklen - 1);
940     int mmu_idx = arm_env_mmu_index(env);
941     void *mem;
942 
943     /*
944      * Trapless lookup.  In addition to actual invalid page, may
945      * return NULL for I/O, watchpoints, clean pages, etc.
946      */
947     mem = tlb_vaddr_to_host(env, vaddr, MMU_DATA_STORE, mmu_idx);
948 
949 #ifndef CONFIG_USER_ONLY
950     if (unlikely(!mem)) {
951         uintptr_t ra = GETPC();
952 
953         /*
954          * Trap if accessing an invalid page.  DC_ZVA requires that we supply
955          * the original pointer for an invalid page.  But watchpoints require
956          * that we probe the actual space.  So do both.
957          */
958         (void) probe_write(env, vaddr_in, 1, mmu_idx, ra);
959         mem = probe_write(env, vaddr, blocklen, mmu_idx, ra);
960 
961         if (unlikely(!mem)) {
962             /*
963              * The only remaining reason for mem == NULL is I/O.
964              * Just do a series of byte writes as the architecture demands.
965              */
966             for (int i = 0; i < blocklen; i++) {
967                 cpu_stb_mmuidx_ra(env, vaddr + i, 0, mmu_idx, ra);
968             }
969             return;
970         }
971     }
972 #endif
973 
974     memset(mem, 0, blocklen);
975 }
976 
977 void HELPER(unaligned_access)(CPUARMState *env, uint64_t addr,
978                               uint32_t access_type, uint32_t mmu_idx)
979 {
980     arm_cpu_do_unaligned_access(env_cpu(env), addr, access_type,
981                                 mmu_idx, GETPC());
982 }
983 
984 /* Memory operations (memset, memmove, memcpy) */
985 
986 /*
987  * Return true if the CPY* and SET* insns can execute; compare
988  * pseudocode CheckMOPSEnabled(), though we refactor it a little.
989  */
990 static bool mops_enabled(CPUARMState *env)
991 {
992     int el = arm_current_el(env);
993 
994     if (el < 2 &&
995         (arm_hcr_el2_eff(env) & (HCR_E2H | HCR_TGE)) != (HCR_E2H | HCR_TGE) &&
996         !(arm_hcrx_el2_eff(env) & HCRX_MSCEN)) {
997         return false;
998     }
999 
1000     if (el == 0) {
1001         if (!el_is_in_host(env, 0)) {
1002             return env->cp15.sctlr_el[1] & SCTLR_MSCEN;
1003         } else {
1004             return env->cp15.sctlr_el[2] & SCTLR_MSCEN;
1005         }
1006     }
1007     return true;
1008 }
1009 
1010 static void check_mops_enabled(CPUARMState *env, uintptr_t ra)
1011 {
1012     if (!mops_enabled(env)) {
1013         raise_exception_ra(env, EXCP_UDEF, syn_uncategorized(),
1014                            exception_target_el(env), ra);
1015     }
1016 }
1017 
1018 /*
1019  * Return the target exception level for an exception due
1020  * to mismatched arguments in a FEAT_MOPS copy or set.
1021  * Compare pseudocode MismatchedCpySetTargetEL()
1022  */
1023 static int mops_mismatch_exception_target_el(CPUARMState *env)
1024 {
1025     int el = arm_current_el(env);
1026 
1027     if (el > 1) {
1028         return el;
1029     }
1030     if (el == 0 && (arm_hcr_el2_eff(env) & HCR_TGE)) {
1031         return 2;
1032     }
1033     if (el == 1 && (arm_hcrx_el2_eff(env) & HCRX_MCE2)) {
1034         return 2;
1035     }
1036     return 1;
1037 }
1038 
1039 /*
1040  * Check whether an M or E instruction was executed with a CF value
1041  * indicating the wrong option for this implementation.
1042  * Assumes we are always Option A.
1043  */
1044 static void check_mops_wrong_option(CPUARMState *env, uint32_t syndrome,
1045                                     uintptr_t ra)
1046 {
1047     if (env->CF != 0) {
1048         syndrome |= 1 << 17; /* Set the wrong-option bit */
1049         raise_exception_ra(env, EXCP_UDEF, syndrome,
1050                            mops_mismatch_exception_target_el(env), ra);
1051     }
1052 }
1053 
1054 /*
1055  * Return the maximum number of bytes we can transfer starting at addr
1056  * without crossing a page boundary.
1057  */
1058 static uint64_t page_limit(uint64_t addr)
1059 {
1060     return TARGET_PAGE_ALIGN(addr + 1) - addr;
1061 }
1062 
1063 /*
1064  * Return the number of bytes we can copy starting from addr and working
1065  * backwards without crossing a page boundary.
1066  */
1067 static uint64_t page_limit_rev(uint64_t addr)
1068 {
1069     return (addr & ~TARGET_PAGE_MASK) + 1;
1070 }
1071 
1072 /*
1073  * Perform part of a memory set on an area of guest memory starting at
1074  * toaddr (a dirty address) and extending for setsize bytes.
1075  *
1076  * Returns the number of bytes actually set, which might be less than
1077  * setsize; the caller should loop until the whole set has been done.
1078  * The caller should ensure that the guest registers are correct
1079  * for the possibility that the first byte of the set encounters
1080  * an exception or watchpoint. We guarantee not to take any faults
1081  * for bytes other than the first.
1082  */
1083 static uint64_t set_step(CPUARMState *env, uint64_t toaddr,
1084                          uint64_t setsize, uint32_t data, int memidx,
1085                          uint32_t *mtedesc, uintptr_t ra)
1086 {
1087     void *mem;
1088 
1089     setsize = MIN(setsize, page_limit(toaddr));
1090     if (*mtedesc) {
1091         uint64_t mtesize = mte_mops_probe(env, toaddr, setsize, *mtedesc);
1092         if (mtesize == 0) {
1093             /* Trap, or not. All CPU state is up to date */
1094             mte_check_fail(env, *mtedesc, toaddr, ra);
1095             /* Continue, with no further MTE checks required */
1096             *mtedesc = 0;
1097         } else {
1098             /* Advance to the end, or to the tag mismatch */
1099             setsize = MIN(setsize, mtesize);
1100         }
1101     }
1102 
1103     toaddr = useronly_clean_ptr(toaddr);
1104     /*
1105      * Trapless lookup: returns NULL for invalid page, I/O,
1106      * watchpoints, clean pages, etc.
1107      */
1108     mem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, memidx);
1109 
1110 #ifndef CONFIG_USER_ONLY
1111     if (unlikely(!mem)) {
1112         /*
1113          * Slow-path: just do one byte write. This will handle the
1114          * watchpoint, invalid page, etc handling correctly.
1115          * For clean code pages, the next iteration will see
1116          * the page dirty and will use the fast path.
1117          */
1118         cpu_stb_mmuidx_ra(env, toaddr, data, memidx, ra);
1119         return 1;
1120     }
1121 #endif
1122     /* Easy case: just memset the host memory */
1123     memset(mem, data, setsize);
1124     return setsize;
1125 }
1126 
1127 /*
1128  * Similar, but setting tags. The architecture requires us to do this
1129  * in 16-byte chunks. SETP accesses are not tag checked; they set
1130  * the tags.
1131  */
1132 static uint64_t set_step_tags(CPUARMState *env, uint64_t toaddr,
1133                               uint64_t setsize, uint32_t data, int memidx,
1134                               uint32_t *mtedesc, uintptr_t ra)
1135 {
1136     void *mem;
1137     uint64_t cleanaddr;
1138 
1139     setsize = MIN(setsize, page_limit(toaddr));
1140 
1141     cleanaddr = useronly_clean_ptr(toaddr);
1142     /*
1143      * Trapless lookup: returns NULL for invalid page, I/O,
1144      * watchpoints, clean pages, etc.
1145      */
1146     mem = tlb_vaddr_to_host(env, cleanaddr, MMU_DATA_STORE, memidx);
1147 
1148 #ifndef CONFIG_USER_ONLY
1149     if (unlikely(!mem)) {
1150         /*
1151          * Slow-path: just do one write. This will handle the
1152          * watchpoint, invalid page, etc handling correctly.
1153          * The architecture requires that we do 16 bytes at a time,
1154          * and we know both ptr and size are 16 byte aligned.
1155          * For clean code pages, the next iteration will see
1156          * the page dirty and will use the fast path.
1157          */
1158         uint64_t repldata = data * 0x0101010101010101ULL;
1159         MemOpIdx oi16 = make_memop_idx(MO_TE | MO_128, memidx);
1160         cpu_st16_mmu(env, toaddr, int128_make128(repldata, repldata), oi16, ra);
1161         mte_mops_set_tags(env, toaddr, 16, *mtedesc);
1162         return 16;
1163     }
1164 #endif
1165     /* Easy case: just memset the host memory */
1166     memset(mem, data, setsize);
1167     mte_mops_set_tags(env, toaddr, setsize, *mtedesc);
1168     return setsize;
1169 }
1170 
1171 typedef uint64_t StepFn(CPUARMState *env, uint64_t toaddr,
1172                         uint64_t setsize, uint32_t data,
1173                         int memidx, uint32_t *mtedesc, uintptr_t ra);
1174 
1175 /* Extract register numbers from a MOPS exception syndrome value */
1176 static int mops_destreg(uint32_t syndrome)
1177 {
1178     return extract32(syndrome, 10, 5);
1179 }
1180 
1181 static int mops_srcreg(uint32_t syndrome)
1182 {
1183     return extract32(syndrome, 5, 5);
1184 }
1185 
1186 static int mops_sizereg(uint32_t syndrome)
1187 {
1188     return extract32(syndrome, 0, 5);
1189 }
1190 
1191 /*
1192  * Return true if TCMA and TBI bits mean we need to do MTE checks.
1193  * We only need to do this once per MOPS insn, not for every page.
1194  */
1195 static bool mte_checks_needed(uint64_t ptr, uint32_t desc)
1196 {
1197     int bit55 = extract64(ptr, 55, 1);
1198 
1199     /*
1200      * Note that tbi_check() returns true for "access checked" but
1201      * tcma_check() returns true for "access unchecked".
1202      */
1203     if (!tbi_check(desc, bit55)) {
1204         return false;
1205     }
1206     return !tcma_check(desc, bit55, allocation_tag_from_addr(ptr));
1207 }
1208 
1209 /* Take an exception if the SETG addr/size are not granule aligned */
1210 static void check_setg_alignment(CPUARMState *env, uint64_t ptr, uint64_t size,
1211                                  uint32_t memidx, uintptr_t ra)
1212 {
1213     if ((size != 0 && !QEMU_IS_ALIGNED(ptr, TAG_GRANULE)) ||
1214         !QEMU_IS_ALIGNED(size, TAG_GRANULE)) {
1215         arm_cpu_do_unaligned_access(env_cpu(env), ptr, MMU_DATA_STORE,
1216                                     memidx, ra);
1217 
1218     }
1219 }
1220 
1221 static uint64_t arm_reg_or_xzr(CPUARMState *env, int reg)
1222 {
1223     /*
1224      * Runtime equivalent of cpu_reg() -- return the CPU register value,
1225      * for contexts when index 31 means XZR (not SP).
1226      */
1227     return reg == 31 ? 0 : env->xregs[reg];
1228 }
1229 
1230 /*
1231  * For the Memory Set operation, our implementation chooses
1232  * always to use "option A", where we update Xd to the final
1233  * address in the SETP insn, and set Xn to be -(bytes remaining).
1234  * On SETM and SETE insns we only need update Xn.
1235  *
1236  * @env: CPU
1237  * @syndrome: syndrome value for mismatch exceptions
1238  * (also contains the register numbers we need to use)
1239  * @mtedesc: MTE descriptor word
1240  * @stepfn: function which does a single part of the set operation
1241  * @is_setg: true if this is the tag-setting SETG variant
1242  */
1243 static void do_setp(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1244                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1245 {
1246     /* Prologue: we choose to do up to the next page boundary */
1247     int rd = mops_destreg(syndrome);
1248     int rs = mops_srcreg(syndrome);
1249     int rn = mops_sizereg(syndrome);
1250     uint8_t data = arm_reg_or_xzr(env, rs);
1251     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1252     uint64_t toaddr = env->xregs[rd];
1253     uint64_t setsize = env->xregs[rn];
1254     uint64_t stagesetsize, step;
1255 
1256     check_mops_enabled(env, ra);
1257 
1258     if (setsize > INT64_MAX) {
1259         setsize = INT64_MAX;
1260         if (is_setg) {
1261             setsize &= ~0xf;
1262         }
1263     }
1264 
1265     if (unlikely(is_setg)) {
1266         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1267     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1268         mtedesc = 0;
1269     }
1270 
1271     stagesetsize = MIN(setsize, page_limit(toaddr));
1272     while (stagesetsize) {
1273         env->xregs[rd] = toaddr;
1274         env->xregs[rn] = setsize;
1275         step = stepfn(env, toaddr, stagesetsize, data, memidx, &mtedesc, ra);
1276         toaddr += step;
1277         setsize -= step;
1278         stagesetsize -= step;
1279     }
1280     /* Insn completed, so update registers to the Option A format */
1281     env->xregs[rd] = toaddr + setsize;
1282     env->xregs[rn] = -setsize;
1283 
1284     /* Set NZCV = 0000 to indicate we are an Option A implementation */
1285     env->NF = 0;
1286     env->ZF = 1; /* our env->ZF encoding is inverted */
1287     env->CF = 0;
1288     env->VF = 0;
1289     return;
1290 }
1291 
1292 void HELPER(setp)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1293 {
1294     do_setp(env, syndrome, mtedesc, set_step, false, GETPC());
1295 }
1296 
1297 void HELPER(setgp)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1298 {
1299     do_setp(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1300 }
1301 
1302 static void do_setm(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1303                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1304 {
1305     /* Main: we choose to do all the full-page chunks */
1306     CPUState *cs = env_cpu(env);
1307     int rd = mops_destreg(syndrome);
1308     int rs = mops_srcreg(syndrome);
1309     int rn = mops_sizereg(syndrome);
1310     uint8_t data = arm_reg_or_xzr(env, rs);
1311     uint64_t toaddr = env->xregs[rd] + env->xregs[rn];
1312     uint64_t setsize = -env->xregs[rn];
1313     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1314     uint64_t step, stagesetsize;
1315 
1316     check_mops_enabled(env, ra);
1317 
1318     /*
1319      * We're allowed to NOP out "no data to copy" before the consistency
1320      * checks; we choose to do so.
1321      */
1322     if (env->xregs[rn] == 0) {
1323         return;
1324     }
1325 
1326     check_mops_wrong_option(env, syndrome, ra);
1327 
1328     /*
1329      * Our implementation will work fine even if we have an unaligned
1330      * destination address, and because we update Xn every time around
1331      * the loop below and the return value from stepfn() may be less
1332      * than requested, we might find toaddr is unaligned. So we don't
1333      * have an IMPDEF check for alignment here.
1334      */
1335 
1336     if (unlikely(is_setg)) {
1337         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1338     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1339         mtedesc = 0;
1340     }
1341 
1342     /* Do the actual memset: we leave the last partial page to SETE */
1343     stagesetsize = setsize & TARGET_PAGE_MASK;
1344     while (stagesetsize > 0) {
1345         step = stepfn(env, toaddr, setsize, data, memidx, &mtedesc, ra);
1346         toaddr += step;
1347         setsize -= step;
1348         stagesetsize -= step;
1349         env->xregs[rn] = -setsize;
1350         if (stagesetsize > 0 && unlikely(cpu_loop_exit_requested(cs))) {
1351             cpu_loop_exit_restore(cs, ra);
1352         }
1353     }
1354 }
1355 
1356 void HELPER(setm)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1357 {
1358     do_setm(env, syndrome, mtedesc, set_step, false, GETPC());
1359 }
1360 
1361 void HELPER(setgm)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1362 {
1363     do_setm(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1364 }
1365 
1366 static void do_sete(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc,
1367                     StepFn *stepfn, bool is_setg, uintptr_t ra)
1368 {
1369     /* Epilogue: do the last partial page */
1370     int rd = mops_destreg(syndrome);
1371     int rs = mops_srcreg(syndrome);
1372     int rn = mops_sizereg(syndrome);
1373     uint8_t data = arm_reg_or_xzr(env, rs);
1374     uint64_t toaddr = env->xregs[rd] + env->xregs[rn];
1375     uint64_t setsize = -env->xregs[rn];
1376     uint32_t memidx = FIELD_EX32(mtedesc, MTEDESC, MIDX);
1377     uint64_t step;
1378 
1379     check_mops_enabled(env, ra);
1380 
1381     /*
1382      * We're allowed to NOP out "no data to copy" before the consistency
1383      * checks; we choose to do so.
1384      */
1385     if (setsize == 0) {
1386         return;
1387     }
1388 
1389     check_mops_wrong_option(env, syndrome, ra);
1390 
1391     /*
1392      * Our implementation has no address alignment requirements, but
1393      * we do want to enforce the "less than a page" size requirement,
1394      * so we don't need to have the "check for interrupts" here.
1395      */
1396     if (setsize >= TARGET_PAGE_SIZE) {
1397         raise_exception_ra(env, EXCP_UDEF, syndrome,
1398                            mops_mismatch_exception_target_el(env), ra);
1399     }
1400 
1401     if (unlikely(is_setg)) {
1402         check_setg_alignment(env, toaddr, setsize, memidx, ra);
1403     } else if (!mte_checks_needed(toaddr, mtedesc)) {
1404         mtedesc = 0;
1405     }
1406 
1407     /* Do the actual memset */
1408     while (setsize > 0) {
1409         step = stepfn(env, toaddr, setsize, data, memidx, &mtedesc, ra);
1410         toaddr += step;
1411         setsize -= step;
1412         env->xregs[rn] = -setsize;
1413     }
1414 }
1415 
1416 void HELPER(sete)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1417 {
1418     do_sete(env, syndrome, mtedesc, set_step, false, GETPC());
1419 }
1420 
1421 void HELPER(setge)(CPUARMState *env, uint32_t syndrome, uint32_t mtedesc)
1422 {
1423     do_sete(env, syndrome, mtedesc, set_step_tags, true, GETPC());
1424 }
1425 
1426 /*
1427  * Perform part of a memory copy from the guest memory at fromaddr
1428  * and extending for copysize bytes, to the guest memory at
1429  * toaddr. Both addresses are dirty.
1430  *
1431  * Returns the number of bytes actually set, which might be less than
1432  * copysize; the caller should loop until the whole copy has been done.
1433  * The caller should ensure that the guest registers are correct
1434  * for the possibility that the first byte of the copy encounters
1435  * an exception or watchpoint. We guarantee not to take any faults
1436  * for bytes other than the first.
1437  */
1438 static uint64_t copy_step(CPUARMState *env, uint64_t toaddr, uint64_t fromaddr,
1439                           uint64_t copysize, int wmemidx, int rmemidx,
1440                           uint32_t *wdesc, uint32_t *rdesc, uintptr_t ra)
1441 {
1442     void *rmem;
1443     void *wmem;
1444 
1445     /* Don't cross a page boundary on either source or destination */
1446     copysize = MIN(copysize, page_limit(toaddr));
1447     copysize = MIN(copysize, page_limit(fromaddr));
1448     /*
1449      * Handle MTE tag checks: either handle the tag mismatch for byte 0,
1450      * or else copy up to but not including the byte with the mismatch.
1451      */
1452     if (*rdesc) {
1453         uint64_t mtesize = mte_mops_probe(env, fromaddr, copysize, *rdesc);
1454         if (mtesize == 0) {
1455             mte_check_fail(env, *rdesc, fromaddr, ra);
1456             *rdesc = 0;
1457         } else {
1458             copysize = MIN(copysize, mtesize);
1459         }
1460     }
1461     if (*wdesc) {
1462         uint64_t mtesize = mte_mops_probe(env, toaddr, copysize, *wdesc);
1463         if (mtesize == 0) {
1464             mte_check_fail(env, *wdesc, toaddr, ra);
1465             *wdesc = 0;
1466         } else {
1467             copysize = MIN(copysize, mtesize);
1468         }
1469     }
1470 
1471     toaddr = useronly_clean_ptr(toaddr);
1472     fromaddr = useronly_clean_ptr(fromaddr);
1473     /* Trapless lookup of whether we can get a host memory pointer */
1474     wmem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, wmemidx);
1475     rmem = tlb_vaddr_to_host(env, fromaddr, MMU_DATA_LOAD, rmemidx);
1476 
1477 #ifndef CONFIG_USER_ONLY
1478     /*
1479      * If we don't have host memory for both source and dest then just
1480      * do a single byte copy. This will handle watchpoints, invalid pages,
1481      * etc correctly. For clean code pages, the next iteration will see
1482      * the page dirty and will use the fast path.
1483      */
1484     if (unlikely(!rmem || !wmem)) {
1485         uint8_t byte;
1486         if (rmem) {
1487             byte = *(uint8_t *)rmem;
1488         } else {
1489             byte = cpu_ldub_mmuidx_ra(env, fromaddr, rmemidx, ra);
1490         }
1491         if (wmem) {
1492             *(uint8_t *)wmem = byte;
1493         } else {
1494             cpu_stb_mmuidx_ra(env, toaddr, byte, wmemidx, ra);
1495         }
1496         return 1;
1497     }
1498 #endif
1499     /* Easy case: just memmove the host memory */
1500     memmove(wmem, rmem, copysize);
1501     return copysize;
1502 }
1503 
1504 /*
1505  * Do part of a backwards memory copy. Here toaddr and fromaddr point
1506  * to the *last* byte to be copied.
1507  */
1508 static uint64_t copy_step_rev(CPUARMState *env, uint64_t toaddr,
1509                               uint64_t fromaddr,
1510                               uint64_t copysize, int wmemidx, int rmemidx,
1511                               uint32_t *wdesc, uint32_t *rdesc, uintptr_t ra)
1512 {
1513     void *rmem;
1514     void *wmem;
1515 
1516     /* Don't cross a page boundary on either source or destination */
1517     copysize = MIN(copysize, page_limit_rev(toaddr));
1518     copysize = MIN(copysize, page_limit_rev(fromaddr));
1519 
1520     /*
1521      * Handle MTE tag checks: either handle the tag mismatch for byte 0,
1522      * or else copy up to but not including the byte with the mismatch.
1523      */
1524     if (*rdesc) {
1525         uint64_t mtesize = mte_mops_probe_rev(env, fromaddr, copysize, *rdesc);
1526         if (mtesize == 0) {
1527             mte_check_fail(env, *rdesc, fromaddr, ra);
1528             *rdesc = 0;
1529         } else {
1530             copysize = MIN(copysize, mtesize);
1531         }
1532     }
1533     if (*wdesc) {
1534         uint64_t mtesize = mte_mops_probe_rev(env, toaddr, copysize, *wdesc);
1535         if (mtesize == 0) {
1536             mte_check_fail(env, *wdesc, toaddr, ra);
1537             *wdesc = 0;
1538         } else {
1539             copysize = MIN(copysize, mtesize);
1540         }
1541     }
1542 
1543     toaddr = useronly_clean_ptr(toaddr);
1544     fromaddr = useronly_clean_ptr(fromaddr);
1545     /* Trapless lookup of whether we can get a host memory pointer */
1546     wmem = tlb_vaddr_to_host(env, toaddr, MMU_DATA_STORE, wmemidx);
1547     rmem = tlb_vaddr_to_host(env, fromaddr, MMU_DATA_LOAD, rmemidx);
1548 
1549 #ifndef CONFIG_USER_ONLY
1550     /*
1551      * If we don't have host memory for both source and dest then just
1552      * do a single byte copy. This will handle watchpoints, invalid pages,
1553      * etc correctly. For clean code pages, the next iteration will see
1554      * the page dirty and will use the fast path.
1555      */
1556     if (unlikely(!rmem || !wmem)) {
1557         uint8_t byte;
1558         if (rmem) {
1559             byte = *(uint8_t *)rmem;
1560         } else {
1561             byte = cpu_ldub_mmuidx_ra(env, fromaddr, rmemidx, ra);
1562         }
1563         if (wmem) {
1564             *(uint8_t *)wmem = byte;
1565         } else {
1566             cpu_stb_mmuidx_ra(env, toaddr, byte, wmemidx, ra);
1567         }
1568         return 1;
1569     }
1570 #endif
1571     /*
1572      * Easy case: just memmove the host memory. Note that wmem and
1573      * rmem here point to the *last* byte to copy.
1574      */
1575     memmove(wmem - (copysize - 1), rmem - (copysize - 1), copysize);
1576     return copysize;
1577 }
1578 
1579 /*
1580  * for the Memory Copy operation, our implementation chooses always
1581  * to use "option A", where we update Xd and Xs to the final addresses
1582  * in the CPYP insn, and then in CPYM and CPYE only need to update Xn.
1583  *
1584  * @env: CPU
1585  * @syndrome: syndrome value for mismatch exceptions
1586  * (also contains the register numbers we need to use)
1587  * @wdesc: MTE descriptor for the writes (destination)
1588  * @rdesc: MTE descriptor for the reads (source)
1589  * @move: true if this is CPY (memmove), false for CPYF (memcpy forwards)
1590  */
1591 static void do_cpyp(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1592                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1593 {
1594     int rd = mops_destreg(syndrome);
1595     int rs = mops_srcreg(syndrome);
1596     int rn = mops_sizereg(syndrome);
1597     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1598     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1599     bool forwards = true;
1600     uint64_t toaddr = env->xregs[rd];
1601     uint64_t fromaddr = env->xregs[rs];
1602     uint64_t copysize = env->xregs[rn];
1603     uint64_t stagecopysize, step;
1604 
1605     check_mops_enabled(env, ra);
1606 
1607 
1608     if (move) {
1609         /*
1610          * Copy backwards if necessary. The direction for a non-overlapping
1611          * copy is IMPDEF; we choose forwards.
1612          */
1613         if (copysize > 0x007FFFFFFFFFFFFFULL) {
1614             copysize = 0x007FFFFFFFFFFFFFULL;
1615         }
1616         uint64_t fs = extract64(fromaddr, 0, 56);
1617         uint64_t ts = extract64(toaddr, 0, 56);
1618         uint64_t fe = extract64(fromaddr + copysize, 0, 56);
1619 
1620         if (fs < ts && fe > ts) {
1621             forwards = false;
1622         }
1623     } else {
1624         if (copysize > INT64_MAX) {
1625             copysize = INT64_MAX;
1626         }
1627     }
1628 
1629     if (!mte_checks_needed(fromaddr, rdesc)) {
1630         rdesc = 0;
1631     }
1632     if (!mte_checks_needed(toaddr, wdesc)) {
1633         wdesc = 0;
1634     }
1635 
1636     if (forwards) {
1637         stagecopysize = MIN(copysize, page_limit(toaddr));
1638         stagecopysize = MIN(stagecopysize, page_limit(fromaddr));
1639         while (stagecopysize) {
1640             env->xregs[rd] = toaddr;
1641             env->xregs[rs] = fromaddr;
1642             env->xregs[rn] = copysize;
1643             step = copy_step(env, toaddr, fromaddr, stagecopysize,
1644                              wmemidx, rmemidx, &wdesc, &rdesc, ra);
1645             toaddr += step;
1646             fromaddr += step;
1647             copysize -= step;
1648             stagecopysize -= step;
1649         }
1650         /* Insn completed, so update registers to the Option A format */
1651         env->xregs[rd] = toaddr + copysize;
1652         env->xregs[rs] = fromaddr + copysize;
1653         env->xregs[rn] = -copysize;
1654     } else {
1655         /*
1656          * In a reverse copy the to and from addrs in Xs and Xd are the start
1657          * of the range, but it's more convenient for us to work with pointers
1658          * to the last byte being copied.
1659          */
1660         toaddr += copysize - 1;
1661         fromaddr += copysize - 1;
1662         stagecopysize = MIN(copysize, page_limit_rev(toaddr));
1663         stagecopysize = MIN(stagecopysize, page_limit_rev(fromaddr));
1664         while (stagecopysize) {
1665             env->xregs[rn] = copysize;
1666             step = copy_step_rev(env, toaddr, fromaddr, stagecopysize,
1667                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1668             copysize -= step;
1669             stagecopysize -= step;
1670             toaddr -= step;
1671             fromaddr -= step;
1672         }
1673         /*
1674          * Insn completed, so update registers to the Option A format.
1675          * For a reverse copy this is no different to the CPYP input format.
1676          */
1677         env->xregs[rn] = copysize;
1678     }
1679 
1680     /* Set NZCV = 0000 to indicate we are an Option A implementation */
1681     env->NF = 0;
1682     env->ZF = 1; /* our env->ZF encoding is inverted */
1683     env->CF = 0;
1684     env->VF = 0;
1685     return;
1686 }
1687 
1688 void HELPER(cpyp)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1689                   uint32_t rdesc)
1690 {
1691     do_cpyp(env, syndrome, wdesc, rdesc, true, GETPC());
1692 }
1693 
1694 void HELPER(cpyfp)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1695                    uint32_t rdesc)
1696 {
1697     do_cpyp(env, syndrome, wdesc, rdesc, false, GETPC());
1698 }
1699 
1700 static void do_cpym(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1701                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1702 {
1703     /* Main: we choose to copy until less than a page remaining */
1704     CPUState *cs = env_cpu(env);
1705     int rd = mops_destreg(syndrome);
1706     int rs = mops_srcreg(syndrome);
1707     int rn = mops_sizereg(syndrome);
1708     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1709     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1710     bool forwards = true;
1711     uint64_t toaddr, fromaddr, copysize, step;
1712 
1713     check_mops_enabled(env, ra);
1714 
1715     /* We choose to NOP out "no data to copy" before consistency checks */
1716     if (env->xregs[rn] == 0) {
1717         return;
1718     }
1719 
1720     check_mops_wrong_option(env, syndrome, ra);
1721 
1722     if (move) {
1723         forwards = (int64_t)env->xregs[rn] < 0;
1724     }
1725 
1726     if (forwards) {
1727         toaddr = env->xregs[rd] + env->xregs[rn];
1728         fromaddr = env->xregs[rs] + env->xregs[rn];
1729         copysize = -env->xregs[rn];
1730     } else {
1731         copysize = env->xregs[rn];
1732         /* This toaddr and fromaddr point to the *last* byte to copy */
1733         toaddr = env->xregs[rd] + copysize - 1;
1734         fromaddr = env->xregs[rs] + copysize - 1;
1735     }
1736 
1737     if (!mte_checks_needed(fromaddr, rdesc)) {
1738         rdesc = 0;
1739     }
1740     if (!mte_checks_needed(toaddr, wdesc)) {
1741         wdesc = 0;
1742     }
1743 
1744     /* Our implementation has no particular parameter requirements for CPYM */
1745 
1746     /* Do the actual memmove */
1747     if (forwards) {
1748         while (copysize >= TARGET_PAGE_SIZE) {
1749             step = copy_step(env, toaddr, fromaddr, copysize,
1750                              wmemidx, rmemidx, &wdesc, &rdesc, ra);
1751             toaddr += step;
1752             fromaddr += step;
1753             copysize -= step;
1754             env->xregs[rn] = -copysize;
1755             if (copysize >= TARGET_PAGE_SIZE &&
1756                 unlikely(cpu_loop_exit_requested(cs))) {
1757                 cpu_loop_exit_restore(cs, ra);
1758             }
1759         }
1760     } else {
1761         while (copysize >= TARGET_PAGE_SIZE) {
1762             step = copy_step_rev(env, toaddr, fromaddr, copysize,
1763                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1764             toaddr -= step;
1765             fromaddr -= step;
1766             copysize -= step;
1767             env->xregs[rn] = copysize;
1768             if (copysize >= TARGET_PAGE_SIZE &&
1769                 unlikely(cpu_loop_exit_requested(cs))) {
1770                 cpu_loop_exit_restore(cs, ra);
1771             }
1772         }
1773     }
1774 }
1775 
1776 void HELPER(cpym)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1777                   uint32_t rdesc)
1778 {
1779     do_cpym(env, syndrome, wdesc, rdesc, true, GETPC());
1780 }
1781 
1782 void HELPER(cpyfm)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1783                    uint32_t rdesc)
1784 {
1785     do_cpym(env, syndrome, wdesc, rdesc, false, GETPC());
1786 }
1787 
1788 static void do_cpye(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1789                     uint32_t rdesc, uint32_t move, uintptr_t ra)
1790 {
1791     /* Epilogue: do the last partial page */
1792     int rd = mops_destreg(syndrome);
1793     int rs = mops_srcreg(syndrome);
1794     int rn = mops_sizereg(syndrome);
1795     uint32_t rmemidx = FIELD_EX32(rdesc, MTEDESC, MIDX);
1796     uint32_t wmemidx = FIELD_EX32(wdesc, MTEDESC, MIDX);
1797     bool forwards = true;
1798     uint64_t toaddr, fromaddr, copysize, step;
1799 
1800     check_mops_enabled(env, ra);
1801 
1802     /* We choose to NOP out "no data to copy" before consistency checks */
1803     if (env->xregs[rn] == 0) {
1804         return;
1805     }
1806 
1807     check_mops_wrong_option(env, syndrome, ra);
1808 
1809     if (move) {
1810         forwards = (int64_t)env->xregs[rn] < 0;
1811     }
1812 
1813     if (forwards) {
1814         toaddr = env->xregs[rd] + env->xregs[rn];
1815         fromaddr = env->xregs[rs] + env->xregs[rn];
1816         copysize = -env->xregs[rn];
1817     } else {
1818         copysize = env->xregs[rn];
1819         /* This toaddr and fromaddr point to the *last* byte to copy */
1820         toaddr = env->xregs[rd] + copysize - 1;
1821         fromaddr = env->xregs[rs] + copysize - 1;
1822     }
1823 
1824     if (!mte_checks_needed(fromaddr, rdesc)) {
1825         rdesc = 0;
1826     }
1827     if (!mte_checks_needed(toaddr, wdesc)) {
1828         wdesc = 0;
1829     }
1830 
1831     /* Check the size; we don't want to have do a check-for-interrupts */
1832     if (copysize >= TARGET_PAGE_SIZE) {
1833         raise_exception_ra(env, EXCP_UDEF, syndrome,
1834                            mops_mismatch_exception_target_el(env), ra);
1835     }
1836 
1837     /* Do the actual memmove */
1838     if (forwards) {
1839         while (copysize > 0) {
1840             step = copy_step(env, toaddr, fromaddr, copysize,
1841                              wmemidx, rmemidx, &wdesc, &rdesc, ra);
1842             toaddr += step;
1843             fromaddr += step;
1844             copysize -= step;
1845             env->xregs[rn] = -copysize;
1846         }
1847     } else {
1848         while (copysize > 0) {
1849             step = copy_step_rev(env, toaddr, fromaddr, copysize,
1850                                  wmemidx, rmemidx, &wdesc, &rdesc, ra);
1851             toaddr -= step;
1852             fromaddr -= step;
1853             copysize -= step;
1854             env->xregs[rn] = copysize;
1855         }
1856     }
1857 }
1858 
1859 void HELPER(cpye)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1860                   uint32_t rdesc)
1861 {
1862     do_cpye(env, syndrome, wdesc, rdesc, true, GETPC());
1863 }
1864 
1865 void HELPER(cpyfe)(CPUARMState *env, uint32_t syndrome, uint32_t wdesc,
1866                    uint32_t rdesc)
1867 {
1868     do_cpye(env, syndrome, wdesc, rdesc, false, GETPC());
1869 }
1870