target-arm/helper-a64.c

   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 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 "cpu.h"
  21 #include "exec/gdbstub.h"
  22 #include "exec/helper-proto.h"
  23 #include "qemu/host-utils.h"
  24 #include "sysemu/sysemu.h"
  25 #include "qemu/bitops.h"
  26 #include "internals.h"
  27 #include "qemu/crc32c.h"
  28 #include <zlib.h> /* For crc32 */
  29
  30 /* C2.4.7 Multiply and divide */
  31 /* special cases for 0 and LLONG_MIN are mandated by the standard */
  32 uint64_t HELPER(udiv64)(uint64_t num, uint64_t den)
  33 {
  34     if (den == 0) {
  35         return 0;
  36     }
  37     return num / den;
  38 }
  39
  40 int64_t HELPER(sdiv64)(int64_t num, int64_t den)
  41 {
  42     if (den == 0) {
  43         return 0;
  44     }
  45     if (num == LLONG_MIN && den == -1) {
  46         return LLONG_MIN;
  47     }
  48     return num / den;
  49 }
  50
  51 uint64_t HELPER(clz64)(uint64_t x)
  52 {
  53     return clz64(x);
  54 }
  55
  56 uint64_t HELPER(cls64)(uint64_t x)
  57 {
  58     return clrsb64(x);
  59 }
  60
  61 uint32_t HELPER(cls32)(uint32_t x)
  62 {
  63     return clrsb32(x);
  64 }
  65
  66 uint32_t HELPER(clz32)(uint32_t x)
  67 {
  68     return clz32(x);
  69 }
  70
  71 uint64_t HELPER(rbit64)(uint64_t x)
  72 {
  73     /* assign the correct byte position */
  74     x = bswap64(x);
  75
  76     /* assign the correct nibble position */
  77     x = ((x & 0xf0f0f0f0f0f0f0f0ULL) >> 4)
  78         | ((x & 0x0f0f0f0f0f0f0f0fULL) << 4);
  79
  80     /* assign the correct bit position */
  81     x = ((x & 0x8888888888888888ULL) >> 3)
  82         | ((x & 0x4444444444444444ULL) >> 1)
  83         | ((x & 0x2222222222222222ULL) << 1)
  84         | ((x & 0x1111111111111111ULL) << 3);
  85
  86     return x;
  87 }
  88
  89 /* Convert a softfloat float_relation_ (as returned by
  90  * the float*_compare functions) to the correct ARM
  91  * NZCV flag state.
  92  */
  93 static inline uint32_t float_rel_to_flags(int res)
  94 {
  95     uint64_t flags;
  96     switch (res) {
  97     case float_relation_equal:
  98         flags = PSTATE_Z | PSTATE_C;
  99         break;
 100     case float_relation_less:
 101         flags = PSTATE_N;
 102         break;
 103     case float_relation_greater:
 104         flags = PSTATE_C;
 105         break;
 106     case float_relation_unordered:
 107     default:
 108         flags = PSTATE_C | PSTATE_V;
 109         break;
 110     }
 111     return flags;
 112 }
 113
 114 uint64_t HELPER(vfp_cmps_a64)(float32 x, float32 y, void *fp_status)
 115 {
 116     return float_rel_to_flags(float32_compare_quiet(x, y, fp_status));
 117 }
 118
 119 uint64_t HELPER(vfp_cmpes_a64)(float32 x, float32 y, void *fp_status)
 120 {
 121     return float_rel_to_flags(float32_compare(x, y, fp_status));
 122 }
 123
 124 uint64_t HELPER(vfp_cmpd_a64)(float64 x, float64 y, void *fp_status)
 125 {
 126     return float_rel_to_flags(float64_compare_quiet(x, y, fp_status));
 127 }
 128
 129 uint64_t HELPER(vfp_cmped_a64)(float64 x, float64 y, void *fp_status)
 130 {
 131     return float_rel_to_flags(float64_compare(x, y, fp_status));
 132 }
 133
 134 float32 HELPER(vfp_mulxs)(float32 a, float32 b, void *fpstp)
 135 {
 136     float_status *fpst = fpstp;
 137
 138     a = float32_squash_input_denormal(a, fpst);
 139     b = float32_squash_input_denormal(b, fpst);
 140
 141     if ((float32_is_zero(a) && float32_is_infinity(b)) ||
 142         (float32_is_infinity(a) && float32_is_zero(b))) {
 143         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
 144         return make_float32((1U << 30) |
 145                             ((float32_val(a) ^ float32_val(b)) & (1U << 31)));
 146     }
 147     return float32_mul(a, b, fpst);
 148 }
 149
 150 float64 HELPER(vfp_mulxd)(float64 a, float64 b, void *fpstp)
 151 {
 152     float_status *fpst = fpstp;
 153
 154     a = float64_squash_input_denormal(a, fpst);
 155     b = float64_squash_input_denormal(b, fpst);
 156
 157     if ((float64_is_zero(a) && float64_is_infinity(b)) ||
 158         (float64_is_infinity(a) && float64_is_zero(b))) {
 159         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
 160         return make_float64((1ULL << 62) |
 161                             ((float64_val(a) ^ float64_val(b)) & (1ULL << 63)));
 162     }
 163     return float64_mul(a, b, fpst);
 164 }
 165
 166 uint64_t HELPER(simd_tbl)(CPUARMState *env, uint64_t result, uint64_t indices,
 167                           uint32_t rn, uint32_t numregs)
 168 {
 169     /* Helper function for SIMD TBL and TBX. We have to do the table
 170      * lookup part for the 64 bits worth of indices we're passed in.
 171      * result is the initial results vector (either zeroes for TBL
 172      * or some guest values for TBX), rn the register number where
 173      * the table starts, and numregs the number of registers in the table.
 174      * We return the results of the lookups.
 175      */
 176     int shift;
 177
 178     for (shift = 0; shift < 64; shift += 8) {
 179         int index = extract64(indices, shift, 8);
 180         if (index < 16 * numregs) {
 181             /* Convert index (a byte offset into the virtual table
 182              * which is a series of 128-bit vectors concatenated)
 183              * into the correct vfp.regs[] element plus a bit offset
 184              * into that element, bearing in mind that the table
 185              * can wrap around from V31 to V0.
 186              */
 187             int elt = (rn * 2 + (index >> 3)) % 64;
 188             int bitidx = (index & 7) * 8;
 189             uint64_t val = extract64(env->vfp.regs[elt], bitidx, 8);
 190
 191             result = deposit64(result, shift, 8, val);
 192         }
 193     }
 194     return result;
 195 }
 196
 197 /* 64bit/double versions of the neon float compare functions */
 198 uint64_t HELPER(neon_ceq_f64)(float64 a, float64 b, void *fpstp)
 199 {
 200     float_status *fpst = fpstp;
 201     return -float64_eq_quiet(a, b, fpst);
 202 }
 203
 204 uint64_t HELPER(neon_cge_f64)(float64 a, float64 b, void *fpstp)
 205 {
 206     float_status *fpst = fpstp;
 207     return -float64_le(b, a, fpst);
 208 }
 209
 210 uint64_t HELPER(neon_cgt_f64)(float64 a, float64 b, void *fpstp)
 211 {
 212     float_status *fpst = fpstp;
 213     return -float64_lt(b, a, fpst);
 214 }
 215
 216 /* Reciprocal step and sqrt step. Note that unlike the A32/T32
 217  * versions, these do a fully fused multiply-add or
 218  * multiply-add-and-halve.
 219  */
 220 #define float32_two make_float32(0x40000000)
 221 #define float32_three make_float32(0x40400000)
 222 #define float32_one_point_five make_float32(0x3fc00000)
 223
 224 #define float64_two make_float64(0x4000000000000000ULL)
 225 #define float64_three make_float64(0x4008000000000000ULL)
 226 #define float64_one_point_five make_float64(0x3FF8000000000000ULL)
 227
 228 float32 HELPER(recpsf_f32)(float32 a, float32 b, void *fpstp)
 229 {
 230     float_status *fpst = fpstp;
 231
 232     a = float32_squash_input_denormal(a, fpst);
 233     b = float32_squash_input_denormal(b, fpst);
 234
 235     a = float32_chs(a);
 236     if ((float32_is_infinity(a) && float32_is_zero(b)) ||
 237         (float32_is_infinity(b) && float32_is_zero(a))) {
 238         return float32_two;
 239     }
 240     return float32_muladd(a, b, float32_two, 0, fpst);
 241 }
 242
 243 float64 HELPER(recpsf_f64)(float64 a, float64 b, void *fpstp)
 244 {
 245     float_status *fpst = fpstp;
 246
 247     a = float64_squash_input_denormal(a, fpst);
 248     b = float64_squash_input_denormal(b, fpst);
 249
 250     a = float64_chs(a);
 251     if ((float64_is_infinity(a) && float64_is_zero(b)) ||
 252         (float64_is_infinity(b) && float64_is_zero(a))) {
 253         return float64_two;
 254     }
 255     return float64_muladd(a, b, float64_two, 0, fpst);
 256 }
 257
 258 float32 HELPER(rsqrtsf_f32)(float32 a, float32 b, void *fpstp)
 259 {
 260     float_status *fpst = fpstp;
 261
 262     a = float32_squash_input_denormal(a, fpst);
 263     b = float32_squash_input_denormal(b, fpst);
 264
 265     a = float32_chs(a);
 266     if ((float32_is_infinity(a) && float32_is_zero(b)) ||
 267         (float32_is_infinity(b) && float32_is_zero(a))) {
 268         return float32_one_point_five;
 269     }
 270     return float32_muladd(a, b, float32_three, float_muladd_halve_result, fpst);
 271 }
 272
 273 float64 HELPER(rsqrtsf_f64)(float64 a, float64 b, void *fpstp)
 274 {
 275     float_status *fpst = fpstp;
 276
 277     a = float64_squash_input_denormal(a, fpst);
 278     b = float64_squash_input_denormal(b, fpst);
 279
 280     a = float64_chs(a);
 281     if ((float64_is_infinity(a) && float64_is_zero(b)) ||
 282         (float64_is_infinity(b) && float64_is_zero(a))) {
 283         return float64_one_point_five;
 284     }
 285     return float64_muladd(a, b, float64_three, float_muladd_halve_result, fpst);
 286 }
 287
 288 /* Pairwise long add: add pairs of adjacent elements into
 289  * double-width elements in the result (eg _s8 is an 8x8->16 op)
 290  */
 291 uint64_t HELPER(neon_addlp_s8)(uint64_t a)
 292 {
 293     uint64_t nsignmask = 0x0080008000800080ULL;
 294     uint64_t wsignmask = 0x8000800080008000ULL;
 295     uint64_t elementmask = 0x00ff00ff00ff00ffULL;
 296     uint64_t tmp1, tmp2;
 297     uint64_t res, signres;
 298
 299     /* Extract odd elements, sign extend each to a 16 bit field */
 300     tmp1 = a & elementmask;
 301     tmp1 ^= nsignmask;
 302     tmp1 |= wsignmask;
 303     tmp1 = (tmp1 - nsignmask) ^ wsignmask;
 304     /* Ditto for the even elements */
 305     tmp2 = (a >> 8) & elementmask;
 306     tmp2 ^= nsignmask;
 307     tmp2 |= wsignmask;
 308     tmp2 = (tmp2 - nsignmask) ^ wsignmask;
 309
 310     /* calculate the result by summing bits 0..14, 16..22, etc,
 311      * and then adjusting the sign bits 15, 23, etc manually.
 312      * This ensures the addition can't overflow the 16 bit field.
 313      */
 314     signres = (tmp1 ^ tmp2) & wsignmask;
 315     res = (tmp1 & ~wsignmask) + (tmp2 & ~wsignmask);
 316     res ^= signres;
 317
 318     return res;
 319 }
 320
 321 uint64_t HELPER(neon_addlp_u8)(uint64_t a)
 322 {
 323     uint64_t tmp;
 324
 325     tmp = a & 0x00ff00ff00ff00ffULL;
 326     tmp += (a >> 8) & 0x00ff00ff00ff00ffULL;
 327     return tmp;
 328 }
 329
 330 uint64_t HELPER(neon_addlp_s16)(uint64_t a)
 331 {
 332     int32_t reslo, reshi;
 333
 334     reslo = (int32_t)(int16_t)a + (int32_t)(int16_t)(a >> 16);
 335     reshi = (int32_t)(int16_t)(a >> 32) + (int32_t)(int16_t)(a >> 48);
 336
 337     return (uint32_t)reslo | (((uint64_t)reshi) << 32);
 338 }
 339
 340 uint64_t HELPER(neon_addlp_u16)(uint64_t a)
 341 {
 342     uint64_t tmp;
 343
 344     tmp = a & 0x0000ffff0000ffffULL;
 345     tmp += (a >> 16) & 0x0000ffff0000ffffULL;
 346     return tmp;
 347 }
 348
 349 /* Floating-point reciprocal exponent - see FPRecpX in ARM ARM */
 350 float32 HELPER(frecpx_f32)(float32 a, void *fpstp)
 351 {
 352     float_status *fpst = fpstp;
 353     uint32_t val32, sbit;
 354     int32_t exp;
 355
 356     if (float32_is_any_nan(a)) {
 357         float32 nan = a;
 358         if (float32_is_signaling_nan(a)) {
 359             float_raise(float_flag_invalid, fpst);
 360             nan = float32_maybe_silence_nan(a);
 361         }
 362         if (fpst->default_nan_mode) {
 363             nan = float32_default_nan;
 364         }
 365         return nan;
 366     }
 367
 368     val32 = float32_val(a);
 369     sbit = 0x80000000ULL & val32;
 370     exp = extract32(val32, 23, 8);
 371
 372     if (exp == 0) {
 373         return make_float32(sbit | (0xfe << 23));
 374     } else {
 375         return make_float32(sbit | (~exp & 0xff) << 23);
 376     }
 377 }
 378
 379 float64 HELPER(frecpx_f64)(float64 a, void *fpstp)
 380 {
 381     float_status *fpst = fpstp;
 382     uint64_t val64, sbit;
 383     int64_t exp;
 384
 385     if (float64_is_any_nan(a)) {
 386         float64 nan = a;
 387         if (float64_is_signaling_nan(a)) {
 388             float_raise(float_flag_invalid, fpst);
 389             nan = float64_maybe_silence_nan(a);
 390         }
 391         if (fpst->default_nan_mode) {
 392             nan = float64_default_nan;
 393         }
 394         return nan;
 395     }
 396
 397     val64 = float64_val(a);
 398     sbit = 0x8000000000000000ULL & val64;
 399     exp = extract64(float64_val(a), 52, 11);
 400
 401     if (exp == 0) {
 402         return make_float64(sbit | (0x7feULL << 52));
 403     } else {
 404         return make_float64(sbit | (~exp & 0x7ffULL) << 52);
 405     }
 406 }
 407
 408 float32 HELPER(fcvtx_f64_to_f32)(float64 a, CPUARMState *env)
 409 {
 410     /* Von Neumann rounding is implemented by using round-to-zero
 411      * and then setting the LSB of the result if Inexact was raised.
 412      */
 413     float32 r;
 414     float_status *fpst = &env->vfp.fp_status;
 415     float_status tstat = *fpst;
 416     int exflags;
 417
 418     set_float_rounding_mode(float_round_to_zero, &tstat);
 419     set_float_exception_flags(0, &tstat);
 420     r = float64_to_float32(a, &tstat);
 421     r = float32_maybe_silence_nan(r);
 422     exflags = get_float_exception_flags(&tstat);
 423     if (exflags & float_flag_inexact) {
 424         r = make_float32(float32_val(r) | 1);
 425     }
 426     exflags |= get_float_exception_flags(fpst);
 427     set_float_exception_flags(exflags, fpst);
 428     return r;
 429 }
 430
 431 /* 64-bit versions of the CRC helpers. Note that although the operation
 432  * (and the prototypes of crc32c() and crc32() mean that only the bottom
 433  * 32 bits of the accumulator and result are used, we pass and return
 434  * uint64_t for convenience of the generated code. Unlike the 32-bit
 435  * instruction set versions, val may genuinely have 64 bits of data in it.
 436  * The upper bytes of val (above the number specified by 'bytes') must have
 437  * been zeroed out by the caller.
 438  */
 439 uint64_t HELPER(crc32_64)(uint64_t acc, uint64_t val, uint32_t bytes)
 440 {
 441     uint8_t buf[8];
 442
 443     stq_le_p(buf, val);
 444
 445     /* zlib crc32 converts the accumulator and output to one's complement.  */
 446     return crc32(acc ^ 0xffffffff, buf, bytes) ^ 0xffffffff;
 447 }
 448
 449 uint64_t HELPER(crc32c_64)(uint64_t acc, uint64_t val, uint32_t bytes)
 450 {
 451     uint8_t buf[8];
 452
 453     stq_le_p(buf, val);
 454
 455     /* Linux crc32c converts the output to one's complement.  */
 456     return crc32c(acc, buf, bytes) ^ 0xffffffff;
 457 }
 458
 459 #if !defined(CONFIG_USER_ONLY)
 460
 461 /* Handle a CPU exception.  */
 462 void aarch64_cpu_do_interrupt(CPUState *cs)
 463 {
 464     ARMCPU *cpu = ARM_CPU(cs);
 465     CPUARMState *env = &cpu->env;
 466     unsigned int new_el = arm_excp_target_el(cs, cs->exception_index);
 467     target_ulong addr = env->cp15.vbar_el[new_el];
 468     unsigned int new_mode = aarch64_pstate_mode(new_el, true);
 469
 470     if (arm_current_el(env) < new_el) {
 471         if (env->aarch64) {
 472             addr += 0x400;
 473         } else {
 474             addr += 0x600;
 475         }
 476     } else if (pstate_read(env) & PSTATE_SP) {
 477         addr += 0x200;
 478     }
 479
 480     arm_log_exception(cs->exception_index);
 481     qemu_log_mask(CPU_LOG_INT, "...from EL%d\n", arm_current_el(env));
 482     if (qemu_loglevel_mask(CPU_LOG_INT)
 483         && !excp_is_internal(cs->exception_index)) {
 484         qemu_log_mask(CPU_LOG_INT, "...with ESR 0x%" PRIx32 "\n",
 485                       env->exception.syndrome);
 486     }
 487
 488     if (arm_is_psci_call(cpu, cs->exception_index)) {
 489         arm_handle_psci_call(cpu);
 490         qemu_log_mask(CPU_LOG_INT, "...handled as PSCI call\n");
 491         return;
 492     }
 493
 494     switch (cs->exception_index) {
 495     case EXCP_PREFETCH_ABORT:
 496     case EXCP_DATA_ABORT:
 497         env->cp15.far_el[new_el] = env->exception.vaddress;
 498         qemu_log_mask(CPU_LOG_INT, "...with FAR 0x%" PRIx64 "\n",
 499                       env->cp15.far_el[new_el]);
 500         /* fall through */
 501     case EXCP_BKPT:
 502     case EXCP_UDEF:
 503     case EXCP_SWI:
 504     case EXCP_HVC:
 505     case EXCP_HYP_TRAP:
 506     case EXCP_SMC:
 507         env->cp15.esr_el[new_el] = env->exception.syndrome;
 508         break;
 509     case EXCP_IRQ:
 510     case EXCP_VIRQ:
 511         addr += 0x80;
 512         break;
 513     case EXCP_FIQ:
 514     case EXCP_VFIQ:
 515         addr += 0x100;
 516         break;
 517     default:
 518         cpu_abort(cs, "Unhandled exception 0x%x\n", cs->exception_index);
 519     }
 520
 521     if (is_a64(env)) {
 522         env->banked_spsr[aarch64_banked_spsr_index(new_el)] = pstate_read(env);
 523         aarch64_save_sp(env, arm_current_el(env));
 524         env->elr_el[new_el] = env->pc;
 525     } else {
 526         env->banked_spsr[0] = cpsr_read(env);
 527         if (!env->thumb) {
 528             env->cp15.esr_el[new_el] |= 1 << 25;
 529         }
 530         env->elr_el[new_el] = env->regs[15];
 531
 532         aarch64_sync_32_to_64(env);
 533
 534         env->condexec_bits = 0;
 535     }
 536
 537     pstate_write(env, PSTATE_DAIF | new_mode);
 538     env->aarch64 = 1;
 539     aarch64_restore_sp(env, new_el);
 540
 541     env->pc = addr;
 542     cs->interrupt_request |= CPU_INTERRUPT_EXITTB;
 543 }
 544 #endif