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     if ((float32_is_zero(a) && float32_is_infinity(b)) ||
 139         (float32_is_infinity(a) && float32_is_zero(b))) {
 140         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
 141         return make_float32((1U << 30) |
 142                             ((float32_val(a) ^ float32_val(b)) & (1U << 31)));
 143     }
 144     return float32_mul(a, b, fpst);
 145 }
 146
 147 float64 HELPER(vfp_mulxd)(float64 a, float64 b, void *fpstp)
 148 {
 149     float_status *fpst = fpstp;
 150
 151     if ((float64_is_zero(a) && float64_is_infinity(b)) ||
 152         (float64_is_infinity(a) && float64_is_zero(b))) {
 153         /* 2.0 with the sign bit set to sign(A) XOR sign(B) */
 154         return make_float64((1ULL << 62) |
 155                             ((float64_val(a) ^ float64_val(b)) & (1ULL << 63)));
 156     }
 157     return float64_mul(a, b, fpst);
 158 }
 159
 160 uint64_t HELPER(simd_tbl)(CPUARMState *env, uint64_t result, uint64_t indices,
 161                           uint32_t rn, uint32_t numregs)
 162 {
 163     /* Helper function for SIMD TBL and TBX. We have to do the table
 164      * lookup part for the 64 bits worth of indices we're passed in.
 165      * result is the initial results vector (either zeroes for TBL
 166      * or some guest values for TBX), rn the register number where
 167      * the table starts, and numregs the number of registers in the table.
 168      * We return the results of the lookups.
 169      */
 170     int shift;
 171
 172     for (shift = 0; shift < 64; shift += 8) {
 173         int index = extract64(indices, shift, 8);
 174         if (index < 16 * numregs) {
 175             /* Convert index (a byte offset into the virtual table
 176              * which is a series of 128-bit vectors concatenated)
 177              * into the correct vfp.regs[] element plus a bit offset
 178              * into that element, bearing in mind that the table
 179              * can wrap around from V31 to V0.
 180              */
 181             int elt = (rn * 2 + (index >> 3)) % 64;
 182             int bitidx = (index & 7) * 8;
 183             uint64_t val = extract64(env->vfp.regs[elt], bitidx, 8);
 184
 185             result = deposit64(result, shift, 8, val);
 186         }
 187     }
 188     return result;
 189 }
 190
 191 /* 64bit/double versions of the neon float compare functions */
 192 uint64_t HELPER(neon_ceq_f64)(float64 a, float64 b, void *fpstp)
 193 {
 194     float_status *fpst = fpstp;
 195     return -float64_eq_quiet(a, b, fpst);
 196 }
 197
 198 uint64_t HELPER(neon_cge_f64)(float64 a, float64 b, void *fpstp)
 199 {
 200     float_status *fpst = fpstp;
 201     return -float64_le(b, a, fpst);
 202 }
 203
 204 uint64_t HELPER(neon_cgt_f64)(float64 a, float64 b, void *fpstp)
 205 {
 206     float_status *fpst = fpstp;
 207     return -float64_lt(b, a, fpst);
 208 }
 209
 210 /* Reciprocal step and sqrt step. Note that unlike the A32/T32
 211  * versions, these do a fully fused multiply-add or
 212  * multiply-add-and-halve.
 213  */
 214 #define float32_two make_float32(0x40000000)
 215 #define float32_three make_float32(0x40400000)
 216 #define float32_one_point_five make_float32(0x3fc00000)
 217
 218 #define float64_two make_float64(0x4000000000000000ULL)
 219 #define float64_three make_float64(0x4008000000000000ULL)
 220 #define float64_one_point_five make_float64(0x3FF8000000000000ULL)
 221
 222 float32 HELPER(recpsf_f32)(float32 a, float32 b, void *fpstp)
 223 {
 224     float_status *fpst = fpstp;
 225
 226     a = float32_chs(a);
 227     if ((float32_is_infinity(a) && float32_is_zero(b)) ||
 228         (float32_is_infinity(b) && float32_is_zero(a))) {
 229         return float32_two;
 230     }
 231     return float32_muladd(a, b, float32_two, 0, fpst);
 232 }
 233
 234 float64 HELPER(recpsf_f64)(float64 a, float64 b, void *fpstp)
 235 {
 236     float_status *fpst = fpstp;
 237
 238     a = float64_chs(a);
 239     if ((float64_is_infinity(a) && float64_is_zero(b)) ||
 240         (float64_is_infinity(b) && float64_is_zero(a))) {
 241         return float64_two;
 242     }
 243     return float64_muladd(a, b, float64_two, 0, fpst);
 244 }
 245
 246 float32 HELPER(rsqrtsf_f32)(float32 a, float32 b, void *fpstp)
 247 {
 248     float_status *fpst = fpstp;
 249
 250     a = float32_chs(a);
 251     if ((float32_is_infinity(a) && float32_is_zero(b)) ||
 252         (float32_is_infinity(b) && float32_is_zero(a))) {
 253         return float32_one_point_five;
 254     }
 255     return float32_muladd(a, b, float32_three, float_muladd_halve_result, fpst);
 256 }
 257
 258 float64 HELPER(rsqrtsf_f64)(float64 a, float64 b, void *fpstp)
 259 {
 260     float_status *fpst = fpstp;
 261
 262     a = float64_chs(a);
 263     if ((float64_is_infinity(a) && float64_is_zero(b)) ||
 264         (float64_is_infinity(b) && float64_is_zero(a))) {
 265         return float64_one_point_five;
 266     }
 267     return float64_muladd(a, b, float64_three, float_muladd_halve_result, fpst);
 268 }
 269
 270 /* Pairwise long add: add pairs of adjacent elements into
 271  * double-width elements in the result (eg _s8 is an 8x8->16 op)
 272  */
 273 uint64_t HELPER(neon_addlp_s8)(uint64_t a)
 274 {
 275     uint64_t nsignmask = 0x0080008000800080ULL;
 276     uint64_t wsignmask = 0x8000800080008000ULL;
 277     uint64_t elementmask = 0x00ff00ff00ff00ffULL;
 278     uint64_t tmp1, tmp2;
 279     uint64_t res, signres;
 280
 281     /* Extract odd elements, sign extend each to a 16 bit field */
 282     tmp1 = a & elementmask;
 283     tmp1 ^= nsignmask;
 284     tmp1 |= wsignmask;
 285     tmp1 = (tmp1 - nsignmask) ^ wsignmask;
 286     /* Ditto for the even elements */
 287     tmp2 = (a >> 8) & elementmask;
 288     tmp2 ^= nsignmask;
 289     tmp2 |= wsignmask;
 290     tmp2 = (tmp2 - nsignmask) ^ wsignmask;
 291
 292     /* calculate the result by summing bits 0..14, 16..22, etc,
 293      * and then adjusting the sign bits 15, 23, etc manually.
 294      * This ensures the addition can't overflow the 16 bit field.
 295      */
 296     signres = (tmp1 ^ tmp2) & wsignmask;
 297     res = (tmp1 & ~wsignmask) + (tmp2 & ~wsignmask);
 298     res ^= signres;
 299
 300     return res;
 301 }
 302
 303 uint64_t HELPER(neon_addlp_u8)(uint64_t a)
 304 {
 305     uint64_t tmp;
 306
 307     tmp = a & 0x00ff00ff00ff00ffULL;
 308     tmp += (a >> 8) & 0x00ff00ff00ff00ffULL;
 309     return tmp;
 310 }
 311
 312 uint64_t HELPER(neon_addlp_s16)(uint64_t a)
 313 {
 314     int32_t reslo, reshi;
 315
 316     reslo = (int32_t)(int16_t)a + (int32_t)(int16_t)(a >> 16);
 317     reshi = (int32_t)(int16_t)(a >> 32) + (int32_t)(int16_t)(a >> 48);
 318
 319     return (uint32_t)reslo | (((uint64_t)reshi) << 32);
 320 }
 321
 322 uint64_t HELPER(neon_addlp_u16)(uint64_t a)
 323 {
 324     uint64_t tmp;
 325
 326     tmp = a & 0x0000ffff0000ffffULL;
 327     tmp += (a >> 16) & 0x0000ffff0000ffffULL;
 328     return tmp;
 329 }
 330
 331 /* Floating-point reciprocal exponent - see FPRecpX in ARM ARM */
 332 float32 HELPER(frecpx_f32)(float32 a, void *fpstp)
 333 {
 334     float_status *fpst = fpstp;
 335     uint32_t val32, sbit;
 336     int32_t exp;
 337
 338     if (float32_is_any_nan(a)) {
 339         float32 nan = a;
 340         if (float32_is_signaling_nan(a)) {
 341             float_raise(float_flag_invalid, fpst);
 342             nan = float32_maybe_silence_nan(a);
 343         }
 344         if (fpst->default_nan_mode) {
 345             nan = float32_default_nan;
 346         }
 347         return nan;
 348     }
 349
 350     val32 = float32_val(a);
 351     sbit = 0x80000000ULL & val32;
 352     exp = extract32(val32, 23, 8);
 353
 354     if (exp == 0) {
 355         return make_float32(sbit | (0xfe << 23));
 356     } else {
 357         return make_float32(sbit | (~exp & 0xff) << 23);
 358     }
 359 }
 360
 361 float64 HELPER(frecpx_f64)(float64 a, void *fpstp)
 362 {
 363     float_status *fpst = fpstp;
 364     uint64_t val64, sbit;
 365     int64_t exp;
 366
 367     if (float64_is_any_nan(a)) {
 368         float64 nan = a;
 369         if (float64_is_signaling_nan(a)) {
 370             float_raise(float_flag_invalid, fpst);
 371             nan = float64_maybe_silence_nan(a);
 372         }
 373         if (fpst->default_nan_mode) {
 374             nan = float64_default_nan;
 375         }
 376         return nan;
 377     }
 378
 379     val64 = float64_val(a);
 380     sbit = 0x8000000000000000ULL & val64;
 381     exp = extract64(float64_val(a), 52, 11);
 382
 383     if (exp == 0) {
 384         return make_float64(sbit | (0x7feULL << 52));
 385     } else {
 386         return make_float64(sbit | (~exp & 0x7ffULL) << 52);
 387     }
 388 }
 389
 390 float32 HELPER(fcvtx_f64_to_f32)(float64 a, CPUARMState *env)
 391 {
 392     /* Von Neumann rounding is implemented by using round-to-zero
 393      * and then setting the LSB of the result if Inexact was raised.
 394      */
 395     float32 r;
 396     float_status *fpst = &env->vfp.fp_status;
 397     float_status tstat = *fpst;
 398     int exflags;
 399
 400     set_float_rounding_mode(float_round_to_zero, &tstat);
 401     set_float_exception_flags(0, &tstat);
 402     r = float64_to_float32(a, &tstat);
 403     r = float32_maybe_silence_nan(r);
 404     exflags = get_float_exception_flags(&tstat);
 405     if (exflags & float_flag_inexact) {
 406         r = make_float32(float32_val(r) | 1);
 407     }
 408     exflags |= get_float_exception_flags(fpst);
 409     set_float_exception_flags(exflags, fpst);
 410     return r;
 411 }
 412
 413 /* 64-bit versions of the CRC helpers. Note that although the operation
 414  * (and the prototypes of crc32c() and crc32() mean that only the bottom
 415  * 32 bits of the accumulator and result are used, we pass and return
 416  * uint64_t for convenience of the generated code. Unlike the 32-bit
 417  * instruction set versions, val may genuinely have 64 bits of data in it.
 418  * The upper bytes of val (above the number specified by 'bytes') must have
 419  * been zeroed out by the caller.
 420  */
 421 uint64_t HELPER(crc32_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     /* zlib crc32 converts the accumulator and output to one's complement.  */
 428     return crc32(acc ^ 0xffffffff, buf, bytes) ^ 0xffffffff;
 429 }
 430
 431 uint64_t HELPER(crc32c_64)(uint64_t acc, uint64_t val, uint32_t bytes)
 432 {
 433     uint8_t buf[8];
 434
 435     stq_le_p(buf, val);
 436
 437     /* Linux crc32c converts the output to one's complement.  */
 438     return crc32c(acc, buf, bytes) ^ 0xffffffff;
 439 }
 440
 441 /* Handle a CPU exception.  */
 442 void aarch64_cpu_do_interrupt(CPUState *cs)
 443 {
 444     ARMCPU *cpu = ARM_CPU(cs);
 445     CPUARMState *env = &cpu->env;
 446     unsigned int new_el = arm_excp_target_el(cs, cs->exception_index);
 447     target_ulong addr = env->cp15.vbar_el[new_el];
 448     unsigned int new_mode = aarch64_pstate_mode(new_el, true);
 449     int i;
 450
 451     if (arm_current_pl(env) < new_el) {
 452         if (env->aarch64) {
 453             addr += 0x400;
 454         } else {
 455             addr += 0x600;
 456         }
 457     } else if (pstate_read(env) & PSTATE_SP) {
 458         addr += 0x200;
 459     }
 460
 461     arm_log_exception(cs->exception_index);
 462     qemu_log_mask(CPU_LOG_INT, "...from EL%d\n", arm_current_pl(env));
 463     if (qemu_loglevel_mask(CPU_LOG_INT)
 464         && !excp_is_internal(cs->exception_index)) {
 465         qemu_log_mask(CPU_LOG_INT, "...with ESR 0x%" PRIx32 "\n",
 466                       env->exception.syndrome);
 467     }
 468
 469     switch (cs->exception_index) {
 470     case EXCP_PREFETCH_ABORT:
 471     case EXCP_DATA_ABORT:
 472         env->cp15.far_el[new_el] = env->exception.vaddress;
 473         qemu_log_mask(CPU_LOG_INT, "...with FAR 0x%" PRIx64 "\n",
 474                       env->cp15.far_el[new_el]);
 475         /* fall through */
 476     case EXCP_BKPT:
 477     case EXCP_UDEF:
 478     case EXCP_SWI:
 479     case EXCP_HVC:
 480     case EXCP_HYP_TRAP:
 481     case EXCP_SMC:
 482         env->cp15.esr_el[new_el] = env->exception.syndrome;
 483         break;
 484     case EXCP_IRQ:
 485     case EXCP_VIRQ:
 486         addr += 0x80;
 487         break;
 488     case EXCP_FIQ:
 489     case EXCP_VFIQ:
 490         addr += 0x100;
 491         break;
 492     default:
 493         cpu_abort(cs, "Unhandled exception 0x%x\n", cs->exception_index);
 494     }
 495
 496     if (is_a64(env)) {
 497         env->banked_spsr[aarch64_banked_spsr_index(new_el)] = pstate_read(env);
 498         aarch64_save_sp(env, arm_current_pl(env));
 499         env->elr_el[new_el] = env->pc;
 500     } else {
 501         env->banked_spsr[0] = cpsr_read(env);
 502         if (!env->thumb) {
 503             env->cp15.esr_el[new_el] |= 1 << 25;
 504         }
 505         env->elr_el[new_el] = env->regs[15];
 506
 507         for (i = 0; i < 15; i++) {
 508             env->xregs[i] = env->regs[i];
 509         }
 510
 511         env->condexec_bits = 0;
 512     }
 513
 514     pstate_write(env, PSTATE_DAIF | new_mode);
 515     env->aarch64 = 1;
 516     aarch64_restore_sp(env, new_el);
 517
 518     env->pc = addr;
 519     cs->interrupt_request |= CPU_INTERRUPT_EXITTB;
 520 }