1/* 2 * Tiny Code Generator for QEMU 3 * 4 * Copyright (c) 2008 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 25/* We only support generating code for 64-bit mode. */ 26#ifndef __arch64__ 27#error "unsupported code generation mode" 28#endif 29 30#include "../tcg-ldst.c.inc" 31#include "../tcg-pool.c.inc" 32 33#ifdef CONFIG_DEBUG_TCG 34static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = { 35 "%g0", 36 "%g1", 37 "%g2", 38 "%g3", 39 "%g4", 40 "%g5", 41 "%g6", 42 "%g7", 43 "%o0", 44 "%o1", 45 "%o2", 46 "%o3", 47 "%o4", 48 "%o5", 49 "%o6", 50 "%o7", 51 "%l0", 52 "%l1", 53 "%l2", 54 "%l3", 55 "%l4", 56 "%l5", 57 "%l6", 58 "%l7", 59 "%i0", 60 "%i1", 61 "%i2", 62 "%i3", 63 "%i4", 64 "%i5", 65 "%i6", 66 "%i7", 67}; 68#endif 69 70#define TCG_CT_CONST_S11 0x100 71#define TCG_CT_CONST_S13 0x200 72#define TCG_CT_CONST_ZERO 0x400 73 74#define ALL_GENERAL_REGS MAKE_64BIT_MASK(0, 32) 75 76/* Define some temporary registers. T3 is used for constant generation. */ 77#define TCG_REG_T1 TCG_REG_G1 78#define TCG_REG_T2 TCG_REG_G2 79#define TCG_REG_T3 TCG_REG_O7 80 81#ifndef CONFIG_SOFTMMU 82# define TCG_GUEST_BASE_REG TCG_REG_I5 83#endif 84 85#define TCG_REG_TB TCG_REG_I1 86 87static const int tcg_target_reg_alloc_order[] = { 88 TCG_REG_L0, 89 TCG_REG_L1, 90 TCG_REG_L2, 91 TCG_REG_L3, 92 TCG_REG_L4, 93 TCG_REG_L5, 94 TCG_REG_L6, 95 TCG_REG_L7, 96 97 TCG_REG_I0, 98 TCG_REG_I1, 99 TCG_REG_I2, 100 TCG_REG_I3, 101 TCG_REG_I4, 102 TCG_REG_I5, 103 104 TCG_REG_G3, 105 TCG_REG_G4, 106 TCG_REG_G5, 107 108 TCG_REG_O0, 109 TCG_REG_O1, 110 TCG_REG_O2, 111 TCG_REG_O3, 112 TCG_REG_O4, 113 TCG_REG_O5, 114}; 115 116static const int tcg_target_call_iarg_regs[6] = { 117 TCG_REG_O0, 118 TCG_REG_O1, 119 TCG_REG_O2, 120 TCG_REG_O3, 121 TCG_REG_O4, 122 TCG_REG_O5, 123}; 124 125static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot) 126{ 127 tcg_debug_assert(kind == TCG_CALL_RET_NORMAL); 128 tcg_debug_assert(slot >= 0 && slot <= 3); 129 return TCG_REG_O0 + slot; 130} 131 132#define INSN_OP(x) ((x) << 30) 133#define INSN_OP2(x) ((x) << 22) 134#define INSN_OP3(x) ((x) << 19) 135#define INSN_OPF(x) ((x) << 5) 136#define INSN_RD(x) ((x) << 25) 137#define INSN_RS1(x) ((x) << 14) 138#define INSN_RS2(x) (x) 139#define INSN_ASI(x) ((x) << 5) 140 141#define INSN_IMM10(x) ((1 << 13) | ((x) & 0x3ff)) 142#define INSN_IMM11(x) ((1 << 13) | ((x) & 0x7ff)) 143#define INSN_IMM13(x) ((1 << 13) | ((x) & 0x1fff)) 144#define INSN_OFF16(x) ((((x) >> 2) & 0x3fff) | ((((x) >> 16) & 3) << 20)) 145#define INSN_OFF19(x) (((x) >> 2) & 0x07ffff) 146#define INSN_COND(x) ((x) << 25) 147 148#define COND_N 0x0 149#define COND_E 0x1 150#define COND_LE 0x2 151#define COND_L 0x3 152#define COND_LEU 0x4 153#define COND_CS 0x5 154#define COND_NEG 0x6 155#define COND_VS 0x7 156#define COND_A 0x8 157#define COND_NE 0x9 158#define COND_G 0xa 159#define COND_GE 0xb 160#define COND_GU 0xc 161#define COND_CC 0xd 162#define COND_POS 0xe 163#define COND_VC 0xf 164#define BA (INSN_OP(0) | INSN_COND(COND_A) | INSN_OP2(0x2)) 165 166#define RCOND_Z 1 167#define RCOND_LEZ 2 168#define RCOND_LZ 3 169#define RCOND_NZ 5 170#define RCOND_GZ 6 171#define RCOND_GEZ 7 172 173#define MOVCC_ICC (1 << 18) 174#define MOVCC_XCC (1 << 18 | 1 << 12) 175 176#define BPCC_ICC 0 177#define BPCC_XCC (2 << 20) 178#define BPCC_PT (1 << 19) 179#define BPCC_PN 0 180#define BPCC_A (1 << 29) 181 182#define BPR_PT BPCC_PT 183 184#define ARITH_ADD (INSN_OP(2) | INSN_OP3(0x00)) 185#define ARITH_ADDCC (INSN_OP(2) | INSN_OP3(0x10)) 186#define ARITH_AND (INSN_OP(2) | INSN_OP3(0x01)) 187#define ARITH_ANDCC (INSN_OP(2) | INSN_OP3(0x11)) 188#define ARITH_ANDN (INSN_OP(2) | INSN_OP3(0x05)) 189#define ARITH_OR (INSN_OP(2) | INSN_OP3(0x02)) 190#define ARITH_ORCC (INSN_OP(2) | INSN_OP3(0x12)) 191#define ARITH_ORN (INSN_OP(2) | INSN_OP3(0x06)) 192#define ARITH_XOR (INSN_OP(2) | INSN_OP3(0x03)) 193#define ARITH_SUB (INSN_OP(2) | INSN_OP3(0x04)) 194#define ARITH_SUBCC (INSN_OP(2) | INSN_OP3(0x14)) 195#define ARITH_ADDC (INSN_OP(2) | INSN_OP3(0x08)) 196#define ARITH_SUBC (INSN_OP(2) | INSN_OP3(0x0c)) 197#define ARITH_UMUL (INSN_OP(2) | INSN_OP3(0x0a)) 198#define ARITH_SMUL (INSN_OP(2) | INSN_OP3(0x0b)) 199#define ARITH_UDIV (INSN_OP(2) | INSN_OP3(0x0e)) 200#define ARITH_SDIV (INSN_OP(2) | INSN_OP3(0x0f)) 201#define ARITH_MULX (INSN_OP(2) | INSN_OP3(0x09)) 202#define ARITH_UDIVX (INSN_OP(2) | INSN_OP3(0x0d)) 203#define ARITH_SDIVX (INSN_OP(2) | INSN_OP3(0x2d)) 204#define ARITH_MOVCC (INSN_OP(2) | INSN_OP3(0x2c)) 205#define ARITH_MOVR (INSN_OP(2) | INSN_OP3(0x2f)) 206 207#define ARITH_ADDXC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x11)) 208#define ARITH_UMULXHI (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x16)) 209 210#define SHIFT_SLL (INSN_OP(2) | INSN_OP3(0x25)) 211#define SHIFT_SRL (INSN_OP(2) | INSN_OP3(0x26)) 212#define SHIFT_SRA (INSN_OP(2) | INSN_OP3(0x27)) 213 214#define SHIFT_SLLX (INSN_OP(2) | INSN_OP3(0x25) | (1 << 12)) 215#define SHIFT_SRLX (INSN_OP(2) | INSN_OP3(0x26) | (1 << 12)) 216#define SHIFT_SRAX (INSN_OP(2) | INSN_OP3(0x27) | (1 << 12)) 217 218#define RDY (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(0)) 219#define WRY (INSN_OP(2) | INSN_OP3(0x30) | INSN_RD(0)) 220#define JMPL (INSN_OP(2) | INSN_OP3(0x38)) 221#define RETURN (INSN_OP(2) | INSN_OP3(0x39)) 222#define SAVE (INSN_OP(2) | INSN_OP3(0x3c)) 223#define RESTORE (INSN_OP(2) | INSN_OP3(0x3d)) 224#define SETHI (INSN_OP(0) | INSN_OP2(0x4)) 225#define CALL INSN_OP(1) 226#define LDUB (INSN_OP(3) | INSN_OP3(0x01)) 227#define LDSB (INSN_OP(3) | INSN_OP3(0x09)) 228#define LDUH (INSN_OP(3) | INSN_OP3(0x02)) 229#define LDSH (INSN_OP(3) | INSN_OP3(0x0a)) 230#define LDUW (INSN_OP(3) | INSN_OP3(0x00)) 231#define LDSW (INSN_OP(3) | INSN_OP3(0x08)) 232#define LDX (INSN_OP(3) | INSN_OP3(0x0b)) 233#define STB (INSN_OP(3) | INSN_OP3(0x05)) 234#define STH (INSN_OP(3) | INSN_OP3(0x06)) 235#define STW (INSN_OP(3) | INSN_OP3(0x04)) 236#define STX (INSN_OP(3) | INSN_OP3(0x0e)) 237#define LDUBA (INSN_OP(3) | INSN_OP3(0x11)) 238#define LDSBA (INSN_OP(3) | INSN_OP3(0x19)) 239#define LDUHA (INSN_OP(3) | INSN_OP3(0x12)) 240#define LDSHA (INSN_OP(3) | INSN_OP3(0x1a)) 241#define LDUWA (INSN_OP(3) | INSN_OP3(0x10)) 242#define LDSWA (INSN_OP(3) | INSN_OP3(0x18)) 243#define LDXA (INSN_OP(3) | INSN_OP3(0x1b)) 244#define STBA (INSN_OP(3) | INSN_OP3(0x15)) 245#define STHA (INSN_OP(3) | INSN_OP3(0x16)) 246#define STWA (INSN_OP(3) | INSN_OP3(0x14)) 247#define STXA (INSN_OP(3) | INSN_OP3(0x1e)) 248 249#define MEMBAR (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(15) | (1 << 13)) 250 251#define NOP (SETHI | INSN_RD(TCG_REG_G0) | 0) 252 253#ifndef ASI_PRIMARY_LITTLE 254#define ASI_PRIMARY_LITTLE 0x88 255#endif 256 257#define LDUH_LE (LDUHA | INSN_ASI(ASI_PRIMARY_LITTLE)) 258#define LDSH_LE (LDSHA | INSN_ASI(ASI_PRIMARY_LITTLE)) 259#define LDUW_LE (LDUWA | INSN_ASI(ASI_PRIMARY_LITTLE)) 260#define LDSW_LE (LDSWA | INSN_ASI(ASI_PRIMARY_LITTLE)) 261#define LDX_LE (LDXA | INSN_ASI(ASI_PRIMARY_LITTLE)) 262 263#define STH_LE (STHA | INSN_ASI(ASI_PRIMARY_LITTLE)) 264#define STW_LE (STWA | INSN_ASI(ASI_PRIMARY_LITTLE)) 265#define STX_LE (STXA | INSN_ASI(ASI_PRIMARY_LITTLE)) 266 267#ifndef use_vis3_instructions 268bool use_vis3_instructions; 269#endif 270 271static bool check_fit_i64(int64_t val, unsigned int bits) 272{ 273 return val == sextract64(val, 0, bits); 274} 275 276static bool check_fit_i32(int32_t val, unsigned int bits) 277{ 278 return val == sextract32(val, 0, bits); 279} 280 281#define check_fit_tl check_fit_i64 282#define check_fit_ptr check_fit_i64 283 284static bool patch_reloc(tcg_insn_unit *src_rw, int type, 285 intptr_t value, intptr_t addend) 286{ 287 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw); 288 uint32_t insn = *src_rw; 289 intptr_t pcrel; 290 291 value += addend; 292 pcrel = tcg_ptr_byte_diff((tcg_insn_unit *)value, src_rx); 293 294 switch (type) { 295 case R_SPARC_WDISP16: 296 if (!check_fit_ptr(pcrel >> 2, 16)) { 297 return false; 298 } 299 insn &= ~INSN_OFF16(-1); 300 insn |= INSN_OFF16(pcrel); 301 break; 302 case R_SPARC_WDISP19: 303 if (!check_fit_ptr(pcrel >> 2, 19)) { 304 return false; 305 } 306 insn &= ~INSN_OFF19(-1); 307 insn |= INSN_OFF19(pcrel); 308 break; 309 case R_SPARC_13: 310 if (!check_fit_ptr(value, 13)) { 311 return false; 312 } 313 insn &= ~INSN_IMM13(-1); 314 insn |= INSN_IMM13(value); 315 break; 316 default: 317 g_assert_not_reached(); 318 } 319 320 *src_rw = insn; 321 return true; 322} 323 324/* test if a constant matches the constraint */ 325static bool tcg_target_const_match(int64_t val, TCGType type, int ct) 326{ 327 if (ct & TCG_CT_CONST) { 328 return 1; 329 } 330 331 if (type == TCG_TYPE_I32) { 332 val = (int32_t)val; 333 } 334 335 if ((ct & TCG_CT_CONST_ZERO) && val == 0) { 336 return 1; 337 } else if ((ct & TCG_CT_CONST_S11) && check_fit_tl(val, 11)) { 338 return 1; 339 } else if ((ct & TCG_CT_CONST_S13) && check_fit_tl(val, 13)) { 340 return 1; 341 } else { 342 return 0; 343 } 344} 345 346static void tcg_out_nop(TCGContext *s) 347{ 348 tcg_out32(s, NOP); 349} 350 351static void tcg_out_arith(TCGContext *s, TCGReg rd, TCGReg rs1, 352 TCGReg rs2, int op) 353{ 354 tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_RS2(rs2)); 355} 356 357static void tcg_out_arithi(TCGContext *s, TCGReg rd, TCGReg rs1, 358 int32_t offset, int op) 359{ 360 tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_IMM13(offset)); 361} 362 363static void tcg_out_arithc(TCGContext *s, TCGReg rd, TCGReg rs1, 364 int32_t val2, int val2const, int op) 365{ 366 tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) 367 | (val2const ? INSN_IMM13(val2) : INSN_RS2(val2))); 368} 369 370static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg) 371{ 372 if (ret != arg) { 373 tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR); 374 } 375 return true; 376} 377 378static void tcg_out_mov_delay(TCGContext *s, TCGReg ret, TCGReg arg) 379{ 380 if (ret != arg) { 381 tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR); 382 } else { 383 tcg_out_nop(s); 384 } 385} 386 387static void tcg_out_sethi(TCGContext *s, TCGReg ret, uint32_t arg) 388{ 389 tcg_out32(s, SETHI | INSN_RD(ret) | ((arg & 0xfffffc00) >> 10)); 390} 391 392/* A 13-bit constant sign-extended to 64 bits. */ 393static void tcg_out_movi_s13(TCGContext *s, TCGReg ret, int32_t arg) 394{ 395 tcg_out_arithi(s, ret, TCG_REG_G0, arg, ARITH_OR); 396} 397 398/* A 32-bit constant sign-extended to 64 bits. */ 399static void tcg_out_movi_s32(TCGContext *s, TCGReg ret, int32_t arg) 400{ 401 tcg_out_sethi(s, ret, ~arg); 402 tcg_out_arithi(s, ret, ret, (arg & 0x3ff) | -0x400, ARITH_XOR); 403} 404 405/* A 32-bit constant zero-extended to 64 bits. */ 406static void tcg_out_movi_u32(TCGContext *s, TCGReg ret, uint32_t arg) 407{ 408 tcg_out_sethi(s, ret, arg); 409 if (arg & 0x3ff) { 410 tcg_out_arithi(s, ret, ret, arg & 0x3ff, ARITH_OR); 411 } 412} 413 414static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret, 415 tcg_target_long arg, bool in_prologue, 416 TCGReg scratch) 417{ 418 tcg_target_long hi, lo = (int32_t)arg; 419 tcg_target_long test, lsb; 420 421 /* A 13-bit constant sign-extended to 64-bits. */ 422 if (check_fit_tl(arg, 13)) { 423 tcg_out_movi_s13(s, ret, arg); 424 return; 425 } 426 427 /* A 32-bit constant, or 32-bit zero-extended to 64-bits. */ 428 if (type == TCG_TYPE_I32 || arg == (uint32_t)arg) { 429 tcg_out_movi_u32(s, ret, arg); 430 return; 431 } 432 433 /* A 13-bit constant relative to the TB. */ 434 if (!in_prologue) { 435 test = tcg_tbrel_diff(s, (void *)arg); 436 if (check_fit_ptr(test, 13)) { 437 tcg_out_arithi(s, ret, TCG_REG_TB, test, ARITH_ADD); 438 return; 439 } 440 } 441 442 /* A 32-bit constant sign-extended to 64-bits. */ 443 if (arg == lo) { 444 tcg_out_movi_s32(s, ret, arg); 445 return; 446 } 447 448 /* A 32-bit constant, shifted. */ 449 lsb = ctz64(arg); 450 test = (tcg_target_long)arg >> lsb; 451 if (lsb > 10 && test == extract64(test, 0, 21)) { 452 tcg_out_sethi(s, ret, test << 10); 453 tcg_out_arithi(s, ret, ret, lsb - 10, SHIFT_SLLX); 454 return; 455 } else if (test == (uint32_t)test || test == (int32_t)test) { 456 tcg_out_movi_int(s, TCG_TYPE_I64, ret, test, in_prologue, scratch); 457 tcg_out_arithi(s, ret, ret, lsb, SHIFT_SLLX); 458 return; 459 } 460 461 /* Use the constant pool, if possible. */ 462 if (!in_prologue) { 463 new_pool_label(s, arg, R_SPARC_13, s->code_ptr, 464 tcg_tbrel_diff(s, NULL)); 465 tcg_out32(s, LDX | INSN_RD(ret) | INSN_RS1(TCG_REG_TB)); 466 return; 467 } 468 469 /* A 64-bit constant decomposed into 2 32-bit pieces. */ 470 if (check_fit_i32(lo, 13)) { 471 hi = (arg - lo) >> 32; 472 tcg_out_movi_u32(s, ret, hi); 473 tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX); 474 tcg_out_arithi(s, ret, ret, lo, ARITH_ADD); 475 } else { 476 hi = arg >> 32; 477 tcg_out_movi_u32(s, ret, hi); 478 tcg_out_movi_u32(s, scratch, lo); 479 tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX); 480 tcg_out_arith(s, ret, ret, scratch, ARITH_OR); 481 } 482} 483 484static void tcg_out_movi(TCGContext *s, TCGType type, 485 TCGReg ret, tcg_target_long arg) 486{ 487 tcg_debug_assert(ret != TCG_REG_T3); 488 tcg_out_movi_int(s, type, ret, arg, false, TCG_REG_T3); 489} 490 491static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs) 492{ 493 g_assert_not_reached(); 494} 495 496static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs) 497{ 498 g_assert_not_reached(); 499} 500 501static void tcg_out_ext8u(TCGContext *s, TCGReg rd, TCGReg rs) 502{ 503 tcg_out_arithi(s, rd, rs, 0xff, ARITH_AND); 504} 505 506static void tcg_out_ext16u(TCGContext *s, TCGReg rd, TCGReg rs) 507{ 508 tcg_out_arithi(s, rd, rs, 16, SHIFT_SLL); 509 tcg_out_arithi(s, rd, rd, 16, SHIFT_SRL); 510} 511 512static void tcg_out_ext32s(TCGContext *s, TCGReg rd, TCGReg rs) 513{ 514 tcg_out_arithi(s, rd, rs, 0, SHIFT_SRA); 515} 516 517static void tcg_out_ext32u(TCGContext *s, TCGReg rd, TCGReg rs) 518{ 519 tcg_out_arithi(s, rd, rs, 0, SHIFT_SRL); 520} 521 522static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs) 523{ 524 tcg_out_ext32s(s, rd, rs); 525} 526 527static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs) 528{ 529 tcg_out_ext32u(s, rd, rs); 530} 531 532static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rs) 533{ 534 tcg_out_mov(s, TCG_TYPE_I32, rd, rs); 535} 536 537static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2) 538{ 539 return false; 540} 541 542static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs, 543 tcg_target_long imm) 544{ 545 /* This function is only used for passing structs by reference. */ 546 g_assert_not_reached(); 547} 548 549static void tcg_out_ldst_rr(TCGContext *s, TCGReg data, TCGReg a1, 550 TCGReg a2, int op) 551{ 552 tcg_out32(s, op | INSN_RD(data) | INSN_RS1(a1) | INSN_RS2(a2)); 553} 554 555static void tcg_out_ldst(TCGContext *s, TCGReg ret, TCGReg addr, 556 intptr_t offset, int op) 557{ 558 if (check_fit_ptr(offset, 13)) { 559 tcg_out32(s, op | INSN_RD(ret) | INSN_RS1(addr) | 560 INSN_IMM13(offset)); 561 } else { 562 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, offset); 563 tcg_out_ldst_rr(s, ret, addr, TCG_REG_T1, op); 564 } 565} 566 567static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret, 568 TCGReg arg1, intptr_t arg2) 569{ 570 tcg_out_ldst(s, ret, arg1, arg2, (type == TCG_TYPE_I32 ? LDUW : LDX)); 571} 572 573static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, 574 TCGReg arg1, intptr_t arg2) 575{ 576 tcg_out_ldst(s, arg, arg1, arg2, (type == TCG_TYPE_I32 ? STW : STX)); 577} 578 579static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val, 580 TCGReg base, intptr_t ofs) 581{ 582 if (val == 0) { 583 tcg_out_st(s, type, TCG_REG_G0, base, ofs); 584 return true; 585 } 586 return false; 587} 588 589static void tcg_out_sety(TCGContext *s, TCGReg rs) 590{ 591 tcg_out32(s, WRY | INSN_RS1(TCG_REG_G0) | INSN_RS2(rs)); 592} 593 594static void tcg_out_div32(TCGContext *s, TCGReg rd, TCGReg rs1, 595 int32_t val2, int val2const, int uns) 596{ 597 /* Load Y with the sign/zero extension of RS1 to 64-bits. */ 598 if (uns) { 599 tcg_out_sety(s, TCG_REG_G0); 600 } else { 601 tcg_out_arithi(s, TCG_REG_T1, rs1, 31, SHIFT_SRA); 602 tcg_out_sety(s, TCG_REG_T1); 603 } 604 605 tcg_out_arithc(s, rd, rs1, val2, val2const, 606 uns ? ARITH_UDIV : ARITH_SDIV); 607} 608 609static const uint8_t tcg_cond_to_bcond[] = { 610 [TCG_COND_EQ] = COND_E, 611 [TCG_COND_NE] = COND_NE, 612 [TCG_COND_LT] = COND_L, 613 [TCG_COND_GE] = COND_GE, 614 [TCG_COND_LE] = COND_LE, 615 [TCG_COND_GT] = COND_G, 616 [TCG_COND_LTU] = COND_CS, 617 [TCG_COND_GEU] = COND_CC, 618 [TCG_COND_LEU] = COND_LEU, 619 [TCG_COND_GTU] = COND_GU, 620}; 621 622static const uint8_t tcg_cond_to_rcond[] = { 623 [TCG_COND_EQ] = RCOND_Z, 624 [TCG_COND_NE] = RCOND_NZ, 625 [TCG_COND_LT] = RCOND_LZ, 626 [TCG_COND_GT] = RCOND_GZ, 627 [TCG_COND_LE] = RCOND_LEZ, 628 [TCG_COND_GE] = RCOND_GEZ 629}; 630 631static void tcg_out_bpcc0(TCGContext *s, int scond, int flags, int off19) 632{ 633 tcg_out32(s, INSN_OP(0) | INSN_OP2(1) | INSN_COND(scond) | flags | off19); 634} 635 636static void tcg_out_bpcc(TCGContext *s, int scond, int flags, TCGLabel *l) 637{ 638 int off19 = 0; 639 640 if (l->has_value) { 641 off19 = INSN_OFF19(tcg_pcrel_diff(s, l->u.value_ptr)); 642 } else { 643 tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP19, l, 0); 644 } 645 tcg_out_bpcc0(s, scond, flags, off19); 646} 647 648static void tcg_out_cmp(TCGContext *s, TCGReg c1, int32_t c2, int c2const) 649{ 650 tcg_out_arithc(s, TCG_REG_G0, c1, c2, c2const, ARITH_SUBCC); 651} 652 653static void tcg_out_brcond_i32(TCGContext *s, TCGCond cond, TCGReg arg1, 654 int32_t arg2, int const_arg2, TCGLabel *l) 655{ 656 tcg_out_cmp(s, arg1, arg2, const_arg2); 657 tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_ICC | BPCC_PT, l); 658 tcg_out_nop(s); 659} 660 661static void tcg_out_movcc(TCGContext *s, TCGCond cond, int cc, TCGReg ret, 662 int32_t v1, int v1const) 663{ 664 tcg_out32(s, ARITH_MOVCC | cc | INSN_RD(ret) 665 | INSN_RS1(tcg_cond_to_bcond[cond]) 666 | (v1const ? INSN_IMM11(v1) : INSN_RS2(v1))); 667} 668 669static void tcg_out_movcond_i32(TCGContext *s, TCGCond cond, TCGReg ret, 670 TCGReg c1, int32_t c2, int c2const, 671 int32_t v1, int v1const) 672{ 673 tcg_out_cmp(s, c1, c2, c2const); 674 tcg_out_movcc(s, cond, MOVCC_ICC, ret, v1, v1const); 675} 676 677static void tcg_out_brcond_i64(TCGContext *s, TCGCond cond, TCGReg arg1, 678 int32_t arg2, int const_arg2, TCGLabel *l) 679{ 680 /* For 64-bit signed comparisons vs zero, we can avoid the compare. */ 681 if (arg2 == 0 && !is_unsigned_cond(cond)) { 682 int off16 = 0; 683 684 if (l->has_value) { 685 off16 = INSN_OFF16(tcg_pcrel_diff(s, l->u.value_ptr)); 686 } else { 687 tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP16, l, 0); 688 } 689 tcg_out32(s, INSN_OP(0) | INSN_OP2(3) | BPR_PT | INSN_RS1(arg1) 690 | INSN_COND(tcg_cond_to_rcond[cond]) | off16); 691 } else { 692 tcg_out_cmp(s, arg1, arg2, const_arg2); 693 tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_XCC | BPCC_PT, l); 694 } 695 tcg_out_nop(s); 696} 697 698static void tcg_out_movr(TCGContext *s, TCGCond cond, TCGReg ret, TCGReg c1, 699 int32_t v1, int v1const) 700{ 701 tcg_out32(s, ARITH_MOVR | INSN_RD(ret) | INSN_RS1(c1) 702 | (tcg_cond_to_rcond[cond] << 10) 703 | (v1const ? INSN_IMM10(v1) : INSN_RS2(v1))); 704} 705 706static void tcg_out_movcond_i64(TCGContext *s, TCGCond cond, TCGReg ret, 707 TCGReg c1, int32_t c2, int c2const, 708 int32_t v1, int v1const) 709{ 710 /* For 64-bit signed comparisons vs zero, we can avoid the compare. 711 Note that the immediate range is one bit smaller, so we must check 712 for that as well. */ 713 if (c2 == 0 && !is_unsigned_cond(cond) 714 && (!v1const || check_fit_i32(v1, 10))) { 715 tcg_out_movr(s, cond, ret, c1, v1, v1const); 716 } else { 717 tcg_out_cmp(s, c1, c2, c2const); 718 tcg_out_movcc(s, cond, MOVCC_XCC, ret, v1, v1const); 719 } 720} 721 722static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret, 723 TCGReg c1, int32_t c2, int c2const) 724{ 725 /* For 32-bit comparisons, we can play games with ADDC/SUBC. */ 726 switch (cond) { 727 case TCG_COND_LTU: 728 case TCG_COND_GEU: 729 /* The result of the comparison is in the carry bit. */ 730 break; 731 732 case TCG_COND_EQ: 733 case TCG_COND_NE: 734 /* For equality, we can transform to inequality vs zero. */ 735 if (c2 != 0) { 736 tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_XOR); 737 c2 = TCG_REG_T1; 738 } else { 739 c2 = c1; 740 } 741 c1 = TCG_REG_G0, c2const = 0; 742 cond = (cond == TCG_COND_EQ ? TCG_COND_GEU : TCG_COND_LTU); 743 break; 744 745 case TCG_COND_GTU: 746 case TCG_COND_LEU: 747 /* If we don't need to load a constant into a register, we can 748 swap the operands on GTU/LEU. There's no benefit to loading 749 the constant into a temporary register. */ 750 if (!c2const || c2 == 0) { 751 TCGReg t = c1; 752 c1 = c2; 753 c2 = t; 754 c2const = 0; 755 cond = tcg_swap_cond(cond); 756 break; 757 } 758 /* FALLTHRU */ 759 760 default: 761 tcg_out_cmp(s, c1, c2, c2const); 762 tcg_out_movi_s13(s, ret, 0); 763 tcg_out_movcc(s, cond, MOVCC_ICC, ret, 1, 1); 764 return; 765 } 766 767 tcg_out_cmp(s, c1, c2, c2const); 768 if (cond == TCG_COND_LTU) { 769 tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_ADDC); 770 } else { 771 tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_SUBC); 772 } 773} 774 775static void tcg_out_setcond_i64(TCGContext *s, TCGCond cond, TCGReg ret, 776 TCGReg c1, int32_t c2, int c2const) 777{ 778 if (use_vis3_instructions) { 779 switch (cond) { 780 case TCG_COND_NE: 781 if (c2 != 0) { 782 break; 783 } 784 c2 = c1, c2const = 0, c1 = TCG_REG_G0; 785 /* FALLTHRU */ 786 case TCG_COND_LTU: 787 tcg_out_cmp(s, c1, c2, c2const); 788 tcg_out_arith(s, ret, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC); 789 return; 790 default: 791 break; 792 } 793 } 794 795 /* For 64-bit signed comparisons vs zero, we can avoid the compare 796 if the input does not overlap the output. */ 797 if (c2 == 0 && !is_unsigned_cond(cond) && c1 != ret) { 798 tcg_out_movi_s13(s, ret, 0); 799 tcg_out_movr(s, cond, ret, c1, 1, 1); 800 } else { 801 tcg_out_cmp(s, c1, c2, c2const); 802 tcg_out_movi_s13(s, ret, 0); 803 tcg_out_movcc(s, cond, MOVCC_XCC, ret, 1, 1); 804 } 805} 806 807static void tcg_out_addsub2_i32(TCGContext *s, TCGReg rl, TCGReg rh, 808 TCGReg al, TCGReg ah, int32_t bl, int blconst, 809 int32_t bh, int bhconst, int opl, int oph) 810{ 811 TCGReg tmp = TCG_REG_T1; 812 813 /* Note that the low parts are fully consumed before tmp is set. */ 814 if (rl != ah && (bhconst || rl != bh)) { 815 tmp = rl; 816 } 817 818 tcg_out_arithc(s, tmp, al, bl, blconst, opl); 819 tcg_out_arithc(s, rh, ah, bh, bhconst, oph); 820 tcg_out_mov(s, TCG_TYPE_I32, rl, tmp); 821} 822 823static void tcg_out_addsub2_i64(TCGContext *s, TCGReg rl, TCGReg rh, 824 TCGReg al, TCGReg ah, int32_t bl, int blconst, 825 int32_t bh, int bhconst, bool is_sub) 826{ 827 TCGReg tmp = TCG_REG_T1; 828 829 /* Note that the low parts are fully consumed before tmp is set. */ 830 if (rl != ah && (bhconst || rl != bh)) { 831 tmp = rl; 832 } 833 834 tcg_out_arithc(s, tmp, al, bl, blconst, is_sub ? ARITH_SUBCC : ARITH_ADDCC); 835 836 if (use_vis3_instructions && !is_sub) { 837 /* Note that ADDXC doesn't accept immediates. */ 838 if (bhconst && bh != 0) { 839 tcg_out_movi_s13(s, TCG_REG_T2, bh); 840 bh = TCG_REG_T2; 841 } 842 tcg_out_arith(s, rh, ah, bh, ARITH_ADDXC); 843 } else if (bh == TCG_REG_G0) { 844 /* If we have a zero, we can perform the operation in two insns, 845 with the arithmetic first, and a conditional move into place. */ 846 if (rh == ah) { 847 tcg_out_arithi(s, TCG_REG_T2, ah, 1, 848 is_sub ? ARITH_SUB : ARITH_ADD); 849 tcg_out_movcc(s, TCG_COND_LTU, MOVCC_XCC, rh, TCG_REG_T2, 0); 850 } else { 851 tcg_out_arithi(s, rh, ah, 1, is_sub ? ARITH_SUB : ARITH_ADD); 852 tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, rh, ah, 0); 853 } 854 } else { 855 /* 856 * Otherwise adjust BH as if there is carry into T2. 857 * Note that constant BH is constrained to 11 bits for the MOVCC, 858 * so the adjustment fits 12 bits. 859 */ 860 if (bhconst) { 861 tcg_out_movi_s13(s, TCG_REG_T2, bh + (is_sub ? -1 : 1)); 862 } else { 863 tcg_out_arithi(s, TCG_REG_T2, bh, 1, 864 is_sub ? ARITH_SUB : ARITH_ADD); 865 } 866 /* ... smoosh T2 back to original BH if carry is clear ... */ 867 tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, TCG_REG_T2, bh, bhconst); 868 /* ... and finally perform the arithmetic with the new operand. */ 869 tcg_out_arith(s, rh, ah, TCG_REG_T2, is_sub ? ARITH_SUB : ARITH_ADD); 870 } 871 872 tcg_out_mov(s, TCG_TYPE_I64, rl, tmp); 873} 874 875static void tcg_out_jmpl_const(TCGContext *s, const tcg_insn_unit *dest, 876 bool in_prologue, bool tail_call) 877{ 878 uintptr_t desti = (uintptr_t)dest; 879 880 tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_REG_T1, 881 desti & ~0xfff, in_prologue, TCG_REG_T2); 882 tcg_out_arithi(s, tail_call ? TCG_REG_G0 : TCG_REG_O7, 883 TCG_REG_T1, desti & 0xfff, JMPL); 884} 885 886static void tcg_out_call_nodelay(TCGContext *s, const tcg_insn_unit *dest, 887 bool in_prologue) 888{ 889 ptrdiff_t disp = tcg_pcrel_diff(s, dest); 890 891 if (disp == (int32_t)disp) { 892 tcg_out32(s, CALL | (uint32_t)disp >> 2); 893 } else { 894 tcg_out_jmpl_const(s, dest, in_prologue, false); 895 } 896} 897 898static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest, 899 const TCGHelperInfo *info) 900{ 901 tcg_out_call_nodelay(s, dest, false); 902 tcg_out_nop(s); 903} 904 905static void tcg_out_mb(TCGContext *s, TCGArg a0) 906{ 907 /* Note that the TCG memory order constants mirror the Sparc MEMBAR. */ 908 tcg_out32(s, MEMBAR | (a0 & TCG_MO_ALL)); 909} 910 911/* Generate global QEMU prologue and epilogue code */ 912static void tcg_target_qemu_prologue(TCGContext *s) 913{ 914 int tmp_buf_size, frame_size; 915 916 /* 917 * The TCG temp buffer is at the top of the frame, immediately 918 * below the frame pointer. Use the logical (aligned) offset here; 919 * the stack bias is applied in temp_allocate_frame(). 920 */ 921 tmp_buf_size = CPU_TEMP_BUF_NLONGS * (int)sizeof(long); 922 tcg_set_frame(s, TCG_REG_I6, -tmp_buf_size, tmp_buf_size); 923 924 /* 925 * TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is 926 * otherwise the minimal frame usable by callees. 927 */ 928 frame_size = TCG_TARGET_CALL_STACK_OFFSET - TCG_TARGET_STACK_BIAS; 929 frame_size += TCG_STATIC_CALL_ARGS_SIZE + tmp_buf_size; 930 frame_size += TCG_TARGET_STACK_ALIGN - 1; 931 frame_size &= -TCG_TARGET_STACK_ALIGN; 932 tcg_out32(s, SAVE | INSN_RD(TCG_REG_O6) | INSN_RS1(TCG_REG_O6) | 933 INSN_IMM13(-frame_size)); 934 935#ifndef CONFIG_SOFTMMU 936 if (guest_base != 0) { 937 tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, 938 guest_base, true, TCG_REG_T1); 939 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG); 940 } 941#endif 942 943 /* We choose TCG_REG_TB such that no move is required. */ 944 QEMU_BUILD_BUG_ON(TCG_REG_TB != TCG_REG_I1); 945 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB); 946 947 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I1, 0, JMPL); 948 /* delay slot */ 949 tcg_out_nop(s); 950 951 /* Epilogue for goto_ptr. */ 952 tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr); 953 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN); 954 /* delay slot */ 955 tcg_out_movi_s13(s, TCG_REG_O0, 0); 956} 957 958static void tcg_out_nop_fill(tcg_insn_unit *p, int count) 959{ 960 int i; 961 for (i = 0; i < count; ++i) { 962 p[i] = NOP; 963 } 964} 965 966static const TCGLdstHelperParam ldst_helper_param = { 967 .ntmp = 1, .tmp = { TCG_REG_T1 } 968}; 969 970static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb) 971{ 972 MemOp opc = get_memop(lb->oi); 973 MemOp sgn; 974 975 if (!patch_reloc(lb->label_ptr[0], R_SPARC_WDISP19, 976 (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 0)) { 977 return false; 978 } 979 980 /* Use inline tcg_out_ext32s; otherwise let the helper sign-extend. */ 981 sgn = (opc & MO_SIZE) < MO_32 ? MO_SIGN : 0; 982 983 tcg_out_ld_helper_args(s, lb, &ldst_helper_param); 984 tcg_out_call(s, qemu_ld_helpers[opc & (MO_SIZE | sgn)], NULL); 985 tcg_out_ld_helper_ret(s, lb, sgn, &ldst_helper_param); 986 987 tcg_out_bpcc0(s, COND_A, BPCC_A | BPCC_PT, 0); 988 return patch_reloc(s->code_ptr - 1, R_SPARC_WDISP19, 989 (intptr_t)lb->raddr, 0); 990} 991 992static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb) 993{ 994 MemOp opc = get_memop(lb->oi); 995 996 if (!patch_reloc(lb->label_ptr[0], R_SPARC_WDISP19, 997 (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 0)) { 998 return false; 999 } 1000 1001 tcg_out_st_helper_args(s, lb, &ldst_helper_param); 1002 tcg_out_call(s, qemu_st_helpers[opc & MO_SIZE], NULL); 1003 1004 tcg_out_bpcc0(s, COND_A, BPCC_A | BPCC_PT, 0); 1005 return patch_reloc(s->code_ptr - 1, R_SPARC_WDISP19, 1006 (intptr_t)lb->raddr, 0); 1007} 1008 1009typedef struct { 1010 TCGReg base; 1011 TCGReg index; 1012} HostAddress; 1013 1014bool tcg_target_has_memory_bswap(MemOp memop) 1015{ 1016 return true; 1017} 1018 1019/* 1020 * For softmmu, perform the TLB load and compare. 1021 * For useronly, perform any required alignment tests. 1022 * In both cases, return a TCGLabelQemuLdst structure if the slow path 1023 * is required and fill in @h with the host address for the fast path. 1024 */ 1025static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h, 1026 TCGReg addr_reg, MemOpIdx oi, 1027 bool is_ld) 1028{ 1029 TCGLabelQemuLdst *ldst = NULL; 1030 MemOp opc = get_memop(oi); 1031 unsigned a_bits = get_alignment_bits(opc); 1032 unsigned s_bits = opc & MO_SIZE; 1033 unsigned a_mask; 1034 1035 /* We don't support unaligned accesses. */ 1036 a_bits = MAX(a_bits, s_bits); 1037 a_mask = (1u << a_bits) - 1; 1038 1039#ifdef CONFIG_SOFTMMU 1040 int mem_index = get_mmuidx(oi); 1041 int fast_off = TLB_MASK_TABLE_OFS(mem_index); 1042 int mask_off = fast_off + offsetof(CPUTLBDescFast, mask); 1043 int table_off = fast_off + offsetof(CPUTLBDescFast, table); 1044 int cmp_off = is_ld ? offsetof(CPUTLBEntry, addr_read) 1045 : offsetof(CPUTLBEntry, addr_write); 1046 int add_off = offsetof(CPUTLBEntry, addend); 1047 int compare_mask; 1048 int cc; 1049 1050 /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx]. */ 1051 QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0); 1052 QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -(1 << 12)); 1053 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T2, TCG_AREG0, mask_off); 1054 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T3, TCG_AREG0, table_off); 1055 1056 /* Extract the page index, shifted into place for tlb index. */ 1057 tcg_out_arithi(s, TCG_REG_T1, addr_reg, 1058 TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS, SHIFT_SRL); 1059 tcg_out_arith(s, TCG_REG_T1, TCG_REG_T1, TCG_REG_T2, ARITH_AND); 1060 1061 /* Add the tlb_table pointer, creating the CPUTLBEntry address into R2. */ 1062 tcg_out_arith(s, TCG_REG_T1, TCG_REG_T1, TCG_REG_T3, ARITH_ADD); 1063 1064 /* Load the tlb comparator and the addend. */ 1065 tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_T2, TCG_REG_T1, cmp_off); 1066 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T1, TCG_REG_T1, add_off); 1067 h->base = TCG_REG_T1; 1068 1069 /* Mask out the page offset, except for the required alignment. */ 1070 compare_mask = TARGET_PAGE_MASK | a_mask; 1071 if (check_fit_tl(compare_mask, 13)) { 1072 tcg_out_arithi(s, TCG_REG_T3, addr_reg, compare_mask, ARITH_AND); 1073 } else { 1074 tcg_out_movi_s32(s, TCG_REG_T3, compare_mask); 1075 tcg_out_arith(s, TCG_REG_T3, addr_reg, TCG_REG_T3, ARITH_AND); 1076 } 1077 tcg_out_cmp(s, TCG_REG_T2, TCG_REG_T3, 0); 1078 1079 ldst = new_ldst_label(s); 1080 ldst->is_ld = is_ld; 1081 ldst->oi = oi; 1082 ldst->addrlo_reg = addr_reg; 1083 ldst->label_ptr[0] = s->code_ptr; 1084 1085 /* bne,pn %[xi]cc, label0 */ 1086 cc = TARGET_LONG_BITS == 64 ? BPCC_XCC : BPCC_ICC; 1087 tcg_out_bpcc0(s, COND_NE, BPCC_PN | cc, 0); 1088#else 1089 if (a_bits != s_bits) { 1090 /* 1091 * Test for at least natural alignment, and defer 1092 * everything else to the helper functions. 1093 */ 1094 tcg_debug_assert(check_fit_tl(a_mask, 13)); 1095 tcg_out_arithi(s, TCG_REG_G0, addr_reg, a_mask, ARITH_ANDCC); 1096 1097 ldst = new_ldst_label(s); 1098 ldst->is_ld = is_ld; 1099 ldst->oi = oi; 1100 ldst->addrlo_reg = addr_reg; 1101 ldst->label_ptr[0] = s->code_ptr; 1102 1103 /* bne,pn %icc, label0 */ 1104 tcg_out_bpcc0(s, COND_NE, BPCC_PN | BPCC_ICC, 0); 1105 } 1106 h->base = guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0; 1107#endif 1108 1109 /* If the guest address must be zero-extended, do in the delay slot. */ 1110 if (TARGET_LONG_BITS == 32) { 1111 tcg_out_ext32u(s, TCG_REG_T2, addr_reg); 1112 h->index = TCG_REG_T2; 1113 } else { 1114 if (ldst) { 1115 tcg_out_nop(s); 1116 } 1117 h->index = addr_reg; 1118 } 1119 return ldst; 1120} 1121 1122static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr, 1123 MemOpIdx oi, TCGType data_type) 1124{ 1125 static const int ld_opc[(MO_SSIZE | MO_BSWAP) + 1] = { 1126 [MO_UB] = LDUB, 1127 [MO_SB] = LDSB, 1128 [MO_UB | MO_LE] = LDUB, 1129 [MO_SB | MO_LE] = LDSB, 1130 1131 [MO_BEUW] = LDUH, 1132 [MO_BESW] = LDSH, 1133 [MO_BEUL] = LDUW, 1134 [MO_BESL] = LDSW, 1135 [MO_BEUQ] = LDX, 1136 [MO_BESQ] = LDX, 1137 1138 [MO_LEUW] = LDUH_LE, 1139 [MO_LESW] = LDSH_LE, 1140 [MO_LEUL] = LDUW_LE, 1141 [MO_LESL] = LDSW_LE, 1142 [MO_LEUQ] = LDX_LE, 1143 [MO_LESQ] = LDX_LE, 1144 }; 1145 1146 TCGLabelQemuLdst *ldst; 1147 HostAddress h; 1148 1149 ldst = prepare_host_addr(s, &h, addr, oi, true); 1150 1151 tcg_out_ldst_rr(s, data, h.base, h.index, 1152 ld_opc[get_memop(oi) & (MO_BSWAP | MO_SSIZE)]); 1153 1154 if (ldst) { 1155 ldst->type = data_type; 1156 ldst->datalo_reg = data; 1157 ldst->raddr = tcg_splitwx_to_rx(s->code_ptr); 1158 } 1159} 1160 1161static void tcg_out_qemu_st(TCGContext *s, TCGReg data, TCGReg addr, 1162 MemOpIdx oi, TCGType data_type) 1163{ 1164 static const int st_opc[(MO_SIZE | MO_BSWAP) + 1] = { 1165 [MO_UB] = STB, 1166 1167 [MO_BEUW] = STH, 1168 [MO_BEUL] = STW, 1169 [MO_BEUQ] = STX, 1170 1171 [MO_LEUW] = STH_LE, 1172 [MO_LEUL] = STW_LE, 1173 [MO_LEUQ] = STX_LE, 1174 }; 1175 1176 TCGLabelQemuLdst *ldst; 1177 HostAddress h; 1178 1179 ldst = prepare_host_addr(s, &h, addr, oi, false); 1180 1181 tcg_out_ldst_rr(s, data, h.base, h.index, 1182 st_opc[get_memop(oi) & (MO_BSWAP | MO_SIZE)]); 1183 1184 if (ldst) { 1185 ldst->type = data_type; 1186 ldst->datalo_reg = data; 1187 ldst->raddr = tcg_splitwx_to_rx(s->code_ptr); 1188 } 1189} 1190 1191static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0) 1192{ 1193 if (check_fit_ptr(a0, 13)) { 1194 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN); 1195 tcg_out_movi_s13(s, TCG_REG_O0, a0); 1196 return; 1197 } else { 1198 intptr_t tb_diff = tcg_tbrel_diff(s, (void *)a0); 1199 if (check_fit_ptr(tb_diff, 13)) { 1200 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN); 1201 /* Note that TCG_REG_TB has been unwound to O1. */ 1202 tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O1, tb_diff, ARITH_ADD); 1203 return; 1204 } 1205 } 1206 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_I0, a0 & ~0x3ff); 1207 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN); 1208 tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, a0 & 0x3ff, ARITH_OR); 1209} 1210 1211static void tcg_out_goto_tb(TCGContext *s, int which) 1212{ 1213 ptrdiff_t off = tcg_tbrel_diff(s, (void *)get_jmp_target_addr(s, which)); 1214 1215 /* Load link and indirect branch. */ 1216 set_jmp_insn_offset(s, which); 1217 tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, TCG_REG_TB, off); 1218 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_TB, 0, JMPL); 1219 /* delay slot */ 1220 tcg_out_nop(s); 1221 set_jmp_reset_offset(s, which); 1222 1223 /* 1224 * For the unlinked path of goto_tb, we need to reset TCG_REG_TB 1225 * to the beginning of this TB. 1226 */ 1227 off = -tcg_current_code_size(s); 1228 if (check_fit_i32(off, 13)) { 1229 tcg_out_arithi(s, TCG_REG_TB, TCG_REG_TB, off, ARITH_ADD); 1230 } else { 1231 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, off); 1232 tcg_out_arith(s, TCG_REG_TB, TCG_REG_TB, TCG_REG_T1, ARITH_ADD); 1233 } 1234} 1235 1236void tb_target_set_jmp_target(const TranslationBlock *tb, int n, 1237 uintptr_t jmp_rx, uintptr_t jmp_rw) 1238{ 1239} 1240 1241static void tcg_out_op(TCGContext *s, TCGOpcode opc, 1242 const TCGArg args[TCG_MAX_OP_ARGS], 1243 const int const_args[TCG_MAX_OP_ARGS]) 1244{ 1245 TCGArg a0, a1, a2; 1246 int c, c2; 1247 1248 /* Hoist the loads of the most common arguments. */ 1249 a0 = args[0]; 1250 a1 = args[1]; 1251 a2 = args[2]; 1252 c2 = const_args[2]; 1253 1254 switch (opc) { 1255 case INDEX_op_goto_ptr: 1256 tcg_out_arithi(s, TCG_REG_G0, a0, 0, JMPL); 1257 tcg_out_mov_delay(s, TCG_REG_TB, a0); 1258 break; 1259 case INDEX_op_br: 1260 tcg_out_bpcc(s, COND_A, BPCC_PT, arg_label(a0)); 1261 tcg_out_nop(s); 1262 break; 1263 1264#define OP_32_64(x) \ 1265 glue(glue(case INDEX_op_, x), _i32): \ 1266 glue(glue(case INDEX_op_, x), _i64) 1267 1268 OP_32_64(ld8u): 1269 tcg_out_ldst(s, a0, a1, a2, LDUB); 1270 break; 1271 OP_32_64(ld8s): 1272 tcg_out_ldst(s, a0, a1, a2, LDSB); 1273 break; 1274 OP_32_64(ld16u): 1275 tcg_out_ldst(s, a0, a1, a2, LDUH); 1276 break; 1277 OP_32_64(ld16s): 1278 tcg_out_ldst(s, a0, a1, a2, LDSH); 1279 break; 1280 case INDEX_op_ld_i32: 1281 case INDEX_op_ld32u_i64: 1282 tcg_out_ldst(s, a0, a1, a2, LDUW); 1283 break; 1284 OP_32_64(st8): 1285 tcg_out_ldst(s, a0, a1, a2, STB); 1286 break; 1287 OP_32_64(st16): 1288 tcg_out_ldst(s, a0, a1, a2, STH); 1289 break; 1290 case INDEX_op_st_i32: 1291 case INDEX_op_st32_i64: 1292 tcg_out_ldst(s, a0, a1, a2, STW); 1293 break; 1294 OP_32_64(add): 1295 c = ARITH_ADD; 1296 goto gen_arith; 1297 OP_32_64(sub): 1298 c = ARITH_SUB; 1299 goto gen_arith; 1300 OP_32_64(and): 1301 c = ARITH_AND; 1302 goto gen_arith; 1303 OP_32_64(andc): 1304 c = ARITH_ANDN; 1305 goto gen_arith; 1306 OP_32_64(or): 1307 c = ARITH_OR; 1308 goto gen_arith; 1309 OP_32_64(orc): 1310 c = ARITH_ORN; 1311 goto gen_arith; 1312 OP_32_64(xor): 1313 c = ARITH_XOR; 1314 goto gen_arith; 1315 case INDEX_op_shl_i32: 1316 c = SHIFT_SLL; 1317 do_shift32: 1318 /* Limit immediate shift count lest we create an illegal insn. */ 1319 tcg_out_arithc(s, a0, a1, a2 & 31, c2, c); 1320 break; 1321 case INDEX_op_shr_i32: 1322 c = SHIFT_SRL; 1323 goto do_shift32; 1324 case INDEX_op_sar_i32: 1325 c = SHIFT_SRA; 1326 goto do_shift32; 1327 case INDEX_op_mul_i32: 1328 c = ARITH_UMUL; 1329 goto gen_arith; 1330 1331 OP_32_64(neg): 1332 c = ARITH_SUB; 1333 goto gen_arith1; 1334 OP_32_64(not): 1335 c = ARITH_ORN; 1336 goto gen_arith1; 1337 1338 case INDEX_op_div_i32: 1339 tcg_out_div32(s, a0, a1, a2, c2, 0); 1340 break; 1341 case INDEX_op_divu_i32: 1342 tcg_out_div32(s, a0, a1, a2, c2, 1); 1343 break; 1344 1345 case INDEX_op_brcond_i32: 1346 tcg_out_brcond_i32(s, a2, a0, a1, const_args[1], arg_label(args[3])); 1347 break; 1348 case INDEX_op_setcond_i32: 1349 tcg_out_setcond_i32(s, args[3], a0, a1, a2, c2); 1350 break; 1351 case INDEX_op_movcond_i32: 1352 tcg_out_movcond_i32(s, args[5], a0, a1, a2, c2, args[3], const_args[3]); 1353 break; 1354 1355 case INDEX_op_add2_i32: 1356 tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3], 1357 args[4], const_args[4], args[5], const_args[5], 1358 ARITH_ADDCC, ARITH_ADDC); 1359 break; 1360 case INDEX_op_sub2_i32: 1361 tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3], 1362 args[4], const_args[4], args[5], const_args[5], 1363 ARITH_SUBCC, ARITH_SUBC); 1364 break; 1365 case INDEX_op_mulu2_i32: 1366 c = ARITH_UMUL; 1367 goto do_mul2; 1368 case INDEX_op_muls2_i32: 1369 c = ARITH_SMUL; 1370 do_mul2: 1371 /* The 32-bit multiply insns produce a full 64-bit result. */ 1372 tcg_out_arithc(s, a0, a2, args[3], const_args[3], c); 1373 tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX); 1374 break; 1375 1376 case INDEX_op_qemu_ld_i32: 1377 tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I32); 1378 break; 1379 case INDEX_op_qemu_ld_i64: 1380 tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I64); 1381 break; 1382 case INDEX_op_qemu_st_i32: 1383 tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I32); 1384 break; 1385 case INDEX_op_qemu_st_i64: 1386 tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I64); 1387 break; 1388 1389 case INDEX_op_ld32s_i64: 1390 tcg_out_ldst(s, a0, a1, a2, LDSW); 1391 break; 1392 case INDEX_op_ld_i64: 1393 tcg_out_ldst(s, a0, a1, a2, LDX); 1394 break; 1395 case INDEX_op_st_i64: 1396 tcg_out_ldst(s, a0, a1, a2, STX); 1397 break; 1398 case INDEX_op_shl_i64: 1399 c = SHIFT_SLLX; 1400 do_shift64: 1401 /* Limit immediate shift count lest we create an illegal insn. */ 1402 tcg_out_arithc(s, a0, a1, a2 & 63, c2, c); 1403 break; 1404 case INDEX_op_shr_i64: 1405 c = SHIFT_SRLX; 1406 goto do_shift64; 1407 case INDEX_op_sar_i64: 1408 c = SHIFT_SRAX; 1409 goto do_shift64; 1410 case INDEX_op_mul_i64: 1411 c = ARITH_MULX; 1412 goto gen_arith; 1413 case INDEX_op_div_i64: 1414 c = ARITH_SDIVX; 1415 goto gen_arith; 1416 case INDEX_op_divu_i64: 1417 c = ARITH_UDIVX; 1418 goto gen_arith; 1419 case INDEX_op_extrh_i64_i32: 1420 tcg_out_arithi(s, a0, a1, 32, SHIFT_SRLX); 1421 break; 1422 1423 case INDEX_op_brcond_i64: 1424 tcg_out_brcond_i64(s, a2, a0, a1, const_args[1], arg_label(args[3])); 1425 break; 1426 case INDEX_op_setcond_i64: 1427 tcg_out_setcond_i64(s, args[3], a0, a1, a2, c2); 1428 break; 1429 case INDEX_op_movcond_i64: 1430 tcg_out_movcond_i64(s, args[5], a0, a1, a2, c2, args[3], const_args[3]); 1431 break; 1432 case INDEX_op_add2_i64: 1433 tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4], 1434 const_args[4], args[5], const_args[5], false); 1435 break; 1436 case INDEX_op_sub2_i64: 1437 tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4], 1438 const_args[4], args[5], const_args[5], true); 1439 break; 1440 case INDEX_op_muluh_i64: 1441 tcg_out_arith(s, args[0], args[1], args[2], ARITH_UMULXHI); 1442 break; 1443 1444 gen_arith: 1445 tcg_out_arithc(s, a0, a1, a2, c2, c); 1446 break; 1447 1448 gen_arith1: 1449 tcg_out_arithc(s, a0, TCG_REG_G0, a1, const_args[1], c); 1450 break; 1451 1452 case INDEX_op_mb: 1453 tcg_out_mb(s, a0); 1454 break; 1455 1456 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */ 1457 case INDEX_op_mov_i64: 1458 case INDEX_op_call: /* Always emitted via tcg_out_call. */ 1459 case INDEX_op_exit_tb: /* Always emitted via tcg_out_exit_tb. */ 1460 case INDEX_op_goto_tb: /* Always emitted via tcg_out_goto_tb. */ 1461 case INDEX_op_ext8s_i32: /* Always emitted via tcg_reg_alloc_op. */ 1462 case INDEX_op_ext8s_i64: 1463 case INDEX_op_ext8u_i32: 1464 case INDEX_op_ext8u_i64: 1465 case INDEX_op_ext16s_i32: 1466 case INDEX_op_ext16s_i64: 1467 case INDEX_op_ext16u_i32: 1468 case INDEX_op_ext16u_i64: 1469 case INDEX_op_ext32s_i64: 1470 case INDEX_op_ext32u_i64: 1471 case INDEX_op_ext_i32_i64: 1472 case INDEX_op_extu_i32_i64: 1473 case INDEX_op_extrl_i64_i32: 1474 default: 1475 g_assert_not_reached(); 1476 } 1477} 1478 1479static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op) 1480{ 1481 switch (op) { 1482 case INDEX_op_goto_ptr: 1483 return C_O0_I1(r); 1484 1485 case INDEX_op_ld8u_i32: 1486 case INDEX_op_ld8u_i64: 1487 case INDEX_op_ld8s_i32: 1488 case INDEX_op_ld8s_i64: 1489 case INDEX_op_ld16u_i32: 1490 case INDEX_op_ld16u_i64: 1491 case INDEX_op_ld16s_i32: 1492 case INDEX_op_ld16s_i64: 1493 case INDEX_op_ld_i32: 1494 case INDEX_op_ld32u_i64: 1495 case INDEX_op_ld32s_i64: 1496 case INDEX_op_ld_i64: 1497 case INDEX_op_neg_i32: 1498 case INDEX_op_neg_i64: 1499 case INDEX_op_not_i32: 1500 case INDEX_op_not_i64: 1501 case INDEX_op_ext32s_i64: 1502 case INDEX_op_ext32u_i64: 1503 case INDEX_op_ext_i32_i64: 1504 case INDEX_op_extu_i32_i64: 1505 case INDEX_op_extrl_i64_i32: 1506 case INDEX_op_extrh_i64_i32: 1507 case INDEX_op_qemu_ld_i32: 1508 case INDEX_op_qemu_ld_i64: 1509 return C_O1_I1(r, r); 1510 1511 case INDEX_op_st8_i32: 1512 case INDEX_op_st8_i64: 1513 case INDEX_op_st16_i32: 1514 case INDEX_op_st16_i64: 1515 case INDEX_op_st_i32: 1516 case INDEX_op_st32_i64: 1517 case INDEX_op_st_i64: 1518 case INDEX_op_qemu_st_i32: 1519 case INDEX_op_qemu_st_i64: 1520 return C_O0_I2(rZ, r); 1521 1522 case INDEX_op_add_i32: 1523 case INDEX_op_add_i64: 1524 case INDEX_op_mul_i32: 1525 case INDEX_op_mul_i64: 1526 case INDEX_op_div_i32: 1527 case INDEX_op_div_i64: 1528 case INDEX_op_divu_i32: 1529 case INDEX_op_divu_i64: 1530 case INDEX_op_sub_i32: 1531 case INDEX_op_sub_i64: 1532 case INDEX_op_and_i32: 1533 case INDEX_op_and_i64: 1534 case INDEX_op_andc_i32: 1535 case INDEX_op_andc_i64: 1536 case INDEX_op_or_i32: 1537 case INDEX_op_or_i64: 1538 case INDEX_op_orc_i32: 1539 case INDEX_op_orc_i64: 1540 case INDEX_op_xor_i32: 1541 case INDEX_op_xor_i64: 1542 case INDEX_op_shl_i32: 1543 case INDEX_op_shl_i64: 1544 case INDEX_op_shr_i32: 1545 case INDEX_op_shr_i64: 1546 case INDEX_op_sar_i32: 1547 case INDEX_op_sar_i64: 1548 case INDEX_op_setcond_i32: 1549 case INDEX_op_setcond_i64: 1550 return C_O1_I2(r, rZ, rJ); 1551 1552 case INDEX_op_brcond_i32: 1553 case INDEX_op_brcond_i64: 1554 return C_O0_I2(rZ, rJ); 1555 case INDEX_op_movcond_i32: 1556 case INDEX_op_movcond_i64: 1557 return C_O1_I4(r, rZ, rJ, rI, 0); 1558 case INDEX_op_add2_i32: 1559 case INDEX_op_add2_i64: 1560 case INDEX_op_sub2_i32: 1561 case INDEX_op_sub2_i64: 1562 return C_O2_I4(r, r, rZ, rZ, rJ, rJ); 1563 case INDEX_op_mulu2_i32: 1564 case INDEX_op_muls2_i32: 1565 return C_O2_I2(r, r, rZ, rJ); 1566 case INDEX_op_muluh_i64: 1567 return C_O1_I2(r, r, r); 1568 1569 default: 1570 g_assert_not_reached(); 1571 } 1572} 1573 1574static void tcg_target_init(TCGContext *s) 1575{ 1576 /* 1577 * Only probe for the platform and capabilities if we haven't already 1578 * determined maximum values at compile time. 1579 */ 1580#ifndef use_vis3_instructions 1581 { 1582 unsigned long hwcap = qemu_getauxval(AT_HWCAP); 1583 use_vis3_instructions = (hwcap & HWCAP_SPARC_VIS3) != 0; 1584 } 1585#endif 1586 1587 tcg_target_available_regs[TCG_TYPE_I32] = ALL_GENERAL_REGS; 1588 tcg_target_available_regs[TCG_TYPE_I64] = ALL_GENERAL_REGS; 1589 1590 tcg_target_call_clobber_regs = 0; 1591 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G1); 1592 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G2); 1593 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G3); 1594 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G4); 1595 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G5); 1596 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G6); 1597 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G7); 1598 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O0); 1599 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O1); 1600 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O2); 1601 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O3); 1602 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O4); 1603 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O5); 1604 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O6); 1605 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O7); 1606 1607 s->reserved_regs = 0; 1608 tcg_regset_set_reg(s->reserved_regs, TCG_REG_G0); /* zero */ 1609 tcg_regset_set_reg(s->reserved_regs, TCG_REG_G6); /* reserved for os */ 1610 tcg_regset_set_reg(s->reserved_regs, TCG_REG_G7); /* thread pointer */ 1611 tcg_regset_set_reg(s->reserved_regs, TCG_REG_I6); /* frame pointer */ 1612 tcg_regset_set_reg(s->reserved_regs, TCG_REG_I7); /* return address */ 1613 tcg_regset_set_reg(s->reserved_regs, TCG_REG_O6); /* stack pointer */ 1614 tcg_regset_set_reg(s->reserved_regs, TCG_REG_T1); /* for internal use */ 1615 tcg_regset_set_reg(s->reserved_regs, TCG_REG_T2); /* for internal use */ 1616 tcg_regset_set_reg(s->reserved_regs, TCG_REG_T3); /* for internal use */ 1617} 1618 1619#define ELF_HOST_MACHINE EM_SPARCV9 1620 1621typedef struct { 1622 DebugFrameHeader h; 1623 uint8_t fde_def_cfa[4]; 1624 uint8_t fde_win_save; 1625 uint8_t fde_ret_save[3]; 1626} DebugFrame; 1627 1628static const DebugFrame debug_frame = { 1629 .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */ 1630 .h.cie.id = -1, 1631 .h.cie.version = 1, 1632 .h.cie.code_align = 1, 1633 .h.cie.data_align = -sizeof(void *) & 0x7f, 1634 .h.cie.return_column = 15, /* o7 */ 1635 1636 /* Total FDE size does not include the "len" member. */ 1637 .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset), 1638 1639 .fde_def_cfa = { 1640 12, 30, /* DW_CFA_def_cfa i6, 2047 */ 1641 (2047 & 0x7f) | 0x80, (2047 >> 7) 1642 }, 1643 .fde_win_save = 0x2d, /* DW_CFA_GNU_window_save */ 1644 .fde_ret_save = { 9, 15, 31 }, /* DW_CFA_register o7, i7 */ 1645}; 1646 1647void tcg_register_jit(const void *buf, size_t buf_size) 1648{ 1649 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame)); 1650} 1651