1/* 2 * Tiny Code Generator for QEMU 3 * 4 * Copyright (c) 2008-2009 Arnaud Patard <arnaud.patard@rtp-net.org> 5 * Copyright (c) 2009 Aurelien Jarno <aurelien@aurel32.net> 6 * Based on i386/tcg-target.c - Copyright (c) 2008 Fabrice Bellard 7 * 8 * Permission is hereby granted, free of charge, to any person obtaining a copy 9 * of this software and associated documentation files (the "Software"), to deal 10 * in the Software without restriction, including without limitation the rights 11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 12 * copies of the Software, and to permit persons to whom the Software is 13 * furnished to do so, subject to the following conditions: 14 * 15 * The above copyright notice and this permission notice shall be included in 16 * all copies or substantial portions of the Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 24 * THE SOFTWARE. 25 */ 26 27#include "../tcg-ldst.c.inc" 28 29#if HOST_BIG_ENDIAN 30# define MIPS_BE 1 31#else 32# define MIPS_BE 0 33#endif 34 35#if TCG_TARGET_REG_BITS == 32 36# define LO_OFF (MIPS_BE * 4) 37# define HI_OFF (4 - LO_OFF) 38#else 39/* To assert at compile-time that these values are never used 40 for TCG_TARGET_REG_BITS == 64. */ 41int link_error(void); 42# define LO_OFF link_error() 43# define HI_OFF link_error() 44#endif 45 46#ifdef CONFIG_DEBUG_TCG 47static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = { 48 "zero", 49 "at", 50 "v0", 51 "v1", 52 "a0", 53 "a1", 54 "a2", 55 "a3", 56 "t0", 57 "t1", 58 "t2", 59 "t3", 60 "t4", 61 "t5", 62 "t6", 63 "t7", 64 "s0", 65 "s1", 66 "s2", 67 "s3", 68 "s4", 69 "s5", 70 "s6", 71 "s7", 72 "t8", 73 "t9", 74 "k0", 75 "k1", 76 "gp", 77 "sp", 78 "s8", 79 "ra", 80}; 81#endif 82 83#define TCG_TMP0 TCG_REG_AT 84#define TCG_TMP1 TCG_REG_T9 85#define TCG_TMP2 TCG_REG_T8 86#define TCG_TMP3 TCG_REG_T7 87 88#ifndef CONFIG_SOFTMMU 89#define TCG_GUEST_BASE_REG TCG_REG_S1 90#endif 91 92/* check if we really need so many registers :P */ 93static const int tcg_target_reg_alloc_order[] = { 94 /* Call saved registers. */ 95 TCG_REG_S0, 96 TCG_REG_S1, 97 TCG_REG_S2, 98 TCG_REG_S3, 99 TCG_REG_S4, 100 TCG_REG_S5, 101 TCG_REG_S6, 102 TCG_REG_S7, 103 TCG_REG_S8, 104 105 /* Call clobbered registers. */ 106 TCG_REG_T4, 107 TCG_REG_T5, 108 TCG_REG_T6, 109 TCG_REG_T7, 110 TCG_REG_T8, 111 TCG_REG_T9, 112 TCG_REG_V1, 113 TCG_REG_V0, 114 115 /* Argument registers, opposite order of allocation. */ 116 TCG_REG_T3, 117 TCG_REG_T2, 118 TCG_REG_T1, 119 TCG_REG_T0, 120 TCG_REG_A3, 121 TCG_REG_A2, 122 TCG_REG_A1, 123 TCG_REG_A0, 124}; 125 126static const TCGReg tcg_target_call_iarg_regs[] = { 127 TCG_REG_A0, 128 TCG_REG_A1, 129 TCG_REG_A2, 130 TCG_REG_A3, 131#if _MIPS_SIM == _ABIN32 || _MIPS_SIM == _ABI64 132 TCG_REG_T0, 133 TCG_REG_T1, 134 TCG_REG_T2, 135 TCG_REG_T3, 136#endif 137}; 138 139static const TCGReg tcg_target_call_oarg_regs[2] = { 140 TCG_REG_V0, 141 TCG_REG_V1 142}; 143 144static const tcg_insn_unit *tb_ret_addr; 145static const tcg_insn_unit *bswap32_addr; 146static const tcg_insn_unit *bswap32u_addr; 147static const tcg_insn_unit *bswap64_addr; 148 149static bool reloc_pc16(tcg_insn_unit *src_rw, const tcg_insn_unit *target) 150{ 151 /* Let the compiler perform the right-shift as part of the arithmetic. */ 152 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw); 153 ptrdiff_t disp = target - (src_rx + 1); 154 if (disp == (int16_t)disp) { 155 *src_rw = deposit32(*src_rw, 0, 16, disp); 156 return true; 157 } 158 return false; 159} 160 161static bool patch_reloc(tcg_insn_unit *code_ptr, int type, 162 intptr_t value, intptr_t addend) 163{ 164 tcg_debug_assert(type == R_MIPS_PC16); 165 tcg_debug_assert(addend == 0); 166 return reloc_pc16(code_ptr, (const tcg_insn_unit *)value); 167} 168 169#define TCG_CT_CONST_ZERO 0x100 170#define TCG_CT_CONST_U16 0x200 /* Unsigned 16-bit: 0 - 0xffff. */ 171#define TCG_CT_CONST_S16 0x400 /* Signed 16-bit: -32768 - 32767 */ 172#define TCG_CT_CONST_P2M1 0x800 /* Power of 2 minus 1. */ 173#define TCG_CT_CONST_N16 0x1000 /* "Negatable" 16-bit: -32767 - 32767 */ 174#define TCG_CT_CONST_WSZ 0x2000 /* word size */ 175 176#define ALL_GENERAL_REGS 0xffffffffu 177#define NOA0_REGS (ALL_GENERAL_REGS & ~(1 << TCG_REG_A0)) 178 179#ifdef CONFIG_SOFTMMU 180#define ALL_QLOAD_REGS \ 181 (NOA0_REGS & ~((TCG_TARGET_REG_BITS < TARGET_LONG_BITS) << TCG_REG_A2)) 182#define ALL_QSTORE_REGS \ 183 (NOA0_REGS & ~(TCG_TARGET_REG_BITS < TARGET_LONG_BITS \ 184 ? (1 << TCG_REG_A2) | (1 << TCG_REG_A3) \ 185 : (1 << TCG_REG_A1))) 186#else 187#define ALL_QLOAD_REGS NOA0_REGS 188#define ALL_QSTORE_REGS NOA0_REGS 189#endif 190 191 192static bool is_p2m1(tcg_target_long val) 193{ 194 return val && ((val + 1) & val) == 0; 195} 196 197/* test if a constant matches the constraint */ 198static bool tcg_target_const_match(int64_t val, TCGType type, int ct) 199{ 200 if (ct & TCG_CT_CONST) { 201 return 1; 202 } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) { 203 return 1; 204 } else if ((ct & TCG_CT_CONST_U16) && val == (uint16_t)val) { 205 return 1; 206 } else if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) { 207 return 1; 208 } else if ((ct & TCG_CT_CONST_N16) && val >= -32767 && val <= 32767) { 209 return 1; 210 } else if ((ct & TCG_CT_CONST_P2M1) 211 && use_mips32r2_instructions && is_p2m1(val)) { 212 return 1; 213 } else if ((ct & TCG_CT_CONST_WSZ) 214 && val == (type == TCG_TYPE_I32 ? 32 : 64)) { 215 return 1; 216 } 217 return 0; 218} 219 220/* instruction opcodes */ 221typedef enum { 222 OPC_J = 002 << 26, 223 OPC_JAL = 003 << 26, 224 OPC_BEQ = 004 << 26, 225 OPC_BNE = 005 << 26, 226 OPC_BLEZ = 006 << 26, 227 OPC_BGTZ = 007 << 26, 228 OPC_ADDIU = 011 << 26, 229 OPC_SLTI = 012 << 26, 230 OPC_SLTIU = 013 << 26, 231 OPC_ANDI = 014 << 26, 232 OPC_ORI = 015 << 26, 233 OPC_XORI = 016 << 26, 234 OPC_LUI = 017 << 26, 235 OPC_BNEL = 025 << 26, 236 OPC_BNEZALC_R6 = 030 << 26, 237 OPC_DADDIU = 031 << 26, 238 OPC_LDL = 032 << 26, 239 OPC_LDR = 033 << 26, 240 OPC_LB = 040 << 26, 241 OPC_LH = 041 << 26, 242 OPC_LWL = 042 << 26, 243 OPC_LW = 043 << 26, 244 OPC_LBU = 044 << 26, 245 OPC_LHU = 045 << 26, 246 OPC_LWR = 046 << 26, 247 OPC_LWU = 047 << 26, 248 OPC_SB = 050 << 26, 249 OPC_SH = 051 << 26, 250 OPC_SWL = 052 << 26, 251 OPC_SW = 053 << 26, 252 OPC_SDL = 054 << 26, 253 OPC_SDR = 055 << 26, 254 OPC_SWR = 056 << 26, 255 OPC_LD = 067 << 26, 256 OPC_SD = 077 << 26, 257 258 OPC_SPECIAL = 000 << 26, 259 OPC_SLL = OPC_SPECIAL | 000, 260 OPC_SRL = OPC_SPECIAL | 002, 261 OPC_ROTR = OPC_SPECIAL | 002 | (1 << 21), 262 OPC_SRA = OPC_SPECIAL | 003, 263 OPC_SLLV = OPC_SPECIAL | 004, 264 OPC_SRLV = OPC_SPECIAL | 006, 265 OPC_ROTRV = OPC_SPECIAL | 006 | 0100, 266 OPC_SRAV = OPC_SPECIAL | 007, 267 OPC_JR_R5 = OPC_SPECIAL | 010, 268 OPC_JALR = OPC_SPECIAL | 011, 269 OPC_MOVZ = OPC_SPECIAL | 012, 270 OPC_MOVN = OPC_SPECIAL | 013, 271 OPC_SYNC = OPC_SPECIAL | 017, 272 OPC_MFHI = OPC_SPECIAL | 020, 273 OPC_MFLO = OPC_SPECIAL | 022, 274 OPC_DSLLV = OPC_SPECIAL | 024, 275 OPC_DSRLV = OPC_SPECIAL | 026, 276 OPC_DROTRV = OPC_SPECIAL | 026 | 0100, 277 OPC_DSRAV = OPC_SPECIAL | 027, 278 OPC_MULT = OPC_SPECIAL | 030, 279 OPC_MUL_R6 = OPC_SPECIAL | 030 | 0200, 280 OPC_MUH = OPC_SPECIAL | 030 | 0300, 281 OPC_MULTU = OPC_SPECIAL | 031, 282 OPC_MULU = OPC_SPECIAL | 031 | 0200, 283 OPC_MUHU = OPC_SPECIAL | 031 | 0300, 284 OPC_DIV = OPC_SPECIAL | 032, 285 OPC_DIV_R6 = OPC_SPECIAL | 032 | 0200, 286 OPC_MOD = OPC_SPECIAL | 032 | 0300, 287 OPC_DIVU = OPC_SPECIAL | 033, 288 OPC_DIVU_R6 = OPC_SPECIAL | 033 | 0200, 289 OPC_MODU = OPC_SPECIAL | 033 | 0300, 290 OPC_DMULT = OPC_SPECIAL | 034, 291 OPC_DMUL = OPC_SPECIAL | 034 | 0200, 292 OPC_DMUH = OPC_SPECIAL | 034 | 0300, 293 OPC_DMULTU = OPC_SPECIAL | 035, 294 OPC_DMULU = OPC_SPECIAL | 035 | 0200, 295 OPC_DMUHU = OPC_SPECIAL | 035 | 0300, 296 OPC_DDIV = OPC_SPECIAL | 036, 297 OPC_DDIV_R6 = OPC_SPECIAL | 036 | 0200, 298 OPC_DMOD = OPC_SPECIAL | 036 | 0300, 299 OPC_DDIVU = OPC_SPECIAL | 037, 300 OPC_DDIVU_R6 = OPC_SPECIAL | 037 | 0200, 301 OPC_DMODU = OPC_SPECIAL | 037 | 0300, 302 OPC_ADDU = OPC_SPECIAL | 041, 303 OPC_SUBU = OPC_SPECIAL | 043, 304 OPC_AND = OPC_SPECIAL | 044, 305 OPC_OR = OPC_SPECIAL | 045, 306 OPC_XOR = OPC_SPECIAL | 046, 307 OPC_NOR = OPC_SPECIAL | 047, 308 OPC_SLT = OPC_SPECIAL | 052, 309 OPC_SLTU = OPC_SPECIAL | 053, 310 OPC_DADDU = OPC_SPECIAL | 055, 311 OPC_DSUBU = OPC_SPECIAL | 057, 312 OPC_SELEQZ = OPC_SPECIAL | 065, 313 OPC_SELNEZ = OPC_SPECIAL | 067, 314 OPC_DSLL = OPC_SPECIAL | 070, 315 OPC_DSRL = OPC_SPECIAL | 072, 316 OPC_DROTR = OPC_SPECIAL | 072 | (1 << 21), 317 OPC_DSRA = OPC_SPECIAL | 073, 318 OPC_DSLL32 = OPC_SPECIAL | 074, 319 OPC_DSRL32 = OPC_SPECIAL | 076, 320 OPC_DROTR32 = OPC_SPECIAL | 076 | (1 << 21), 321 OPC_DSRA32 = OPC_SPECIAL | 077, 322 OPC_CLZ_R6 = OPC_SPECIAL | 0120, 323 OPC_DCLZ_R6 = OPC_SPECIAL | 0122, 324 325 OPC_REGIMM = 001 << 26, 326 OPC_BLTZ = OPC_REGIMM | (000 << 16), 327 OPC_BGEZ = OPC_REGIMM | (001 << 16), 328 329 OPC_SPECIAL2 = 034 << 26, 330 OPC_MUL_R5 = OPC_SPECIAL2 | 002, 331 OPC_CLZ = OPC_SPECIAL2 | 040, 332 OPC_DCLZ = OPC_SPECIAL2 | 044, 333 334 OPC_SPECIAL3 = 037 << 26, 335 OPC_EXT = OPC_SPECIAL3 | 000, 336 OPC_DEXTM = OPC_SPECIAL3 | 001, 337 OPC_DEXTU = OPC_SPECIAL3 | 002, 338 OPC_DEXT = OPC_SPECIAL3 | 003, 339 OPC_INS = OPC_SPECIAL3 | 004, 340 OPC_DINSM = OPC_SPECIAL3 | 005, 341 OPC_DINSU = OPC_SPECIAL3 | 006, 342 OPC_DINS = OPC_SPECIAL3 | 007, 343 OPC_WSBH = OPC_SPECIAL3 | 00240, 344 OPC_DSBH = OPC_SPECIAL3 | 00244, 345 OPC_DSHD = OPC_SPECIAL3 | 00544, 346 OPC_SEB = OPC_SPECIAL3 | 02040, 347 OPC_SEH = OPC_SPECIAL3 | 03040, 348 349 /* MIPS r6 doesn't have JR, JALR should be used instead */ 350 OPC_JR = use_mips32r6_instructions ? OPC_JALR : OPC_JR_R5, 351 352 /* 353 * MIPS r6 replaces MUL with an alternative encoding which is 354 * backwards-compatible at the assembly level. 355 */ 356 OPC_MUL = use_mips32r6_instructions ? OPC_MUL_R6 : OPC_MUL_R5, 357 358 /* MIPS r6 introduced names for weaker variants of SYNC. These are 359 backward compatible to previous architecture revisions. */ 360 OPC_SYNC_WMB = OPC_SYNC | 0x04 << 6, 361 OPC_SYNC_MB = OPC_SYNC | 0x10 << 6, 362 OPC_SYNC_ACQUIRE = OPC_SYNC | 0x11 << 6, 363 OPC_SYNC_RELEASE = OPC_SYNC | 0x12 << 6, 364 OPC_SYNC_RMB = OPC_SYNC | 0x13 << 6, 365 366 /* Aliases for convenience. */ 367 ALIAS_PADD = sizeof(void *) == 4 ? OPC_ADDU : OPC_DADDU, 368 ALIAS_PADDI = sizeof(void *) == 4 ? OPC_ADDIU : OPC_DADDIU, 369 ALIAS_TSRL = TARGET_LONG_BITS == 32 || TCG_TARGET_REG_BITS == 32 370 ? OPC_SRL : OPC_DSRL, 371} MIPSInsn; 372 373/* 374 * Type reg 375 */ 376static void tcg_out_opc_reg(TCGContext *s, MIPSInsn opc, 377 TCGReg rd, TCGReg rs, TCGReg rt) 378{ 379 int32_t inst; 380 381 inst = opc; 382 inst |= (rs & 0x1F) << 21; 383 inst |= (rt & 0x1F) << 16; 384 inst |= (rd & 0x1F) << 11; 385 tcg_out32(s, inst); 386} 387 388/* 389 * Type immediate 390 */ 391static void tcg_out_opc_imm(TCGContext *s, MIPSInsn opc, 392 TCGReg rt, TCGReg rs, TCGArg imm) 393{ 394 int32_t inst; 395 396 inst = opc; 397 inst |= (rs & 0x1F) << 21; 398 inst |= (rt & 0x1F) << 16; 399 inst |= (imm & 0xffff); 400 tcg_out32(s, inst); 401} 402 403/* 404 * Type bitfield 405 */ 406static void tcg_out_opc_bf(TCGContext *s, MIPSInsn opc, TCGReg rt, 407 TCGReg rs, int msb, int lsb) 408{ 409 int32_t inst; 410 411 inst = opc; 412 inst |= (rs & 0x1F) << 21; 413 inst |= (rt & 0x1F) << 16; 414 inst |= (msb & 0x1F) << 11; 415 inst |= (lsb & 0x1F) << 6; 416 tcg_out32(s, inst); 417} 418 419static void tcg_out_opc_bf64(TCGContext *s, MIPSInsn opc, MIPSInsn opm, 420 MIPSInsn oph, TCGReg rt, TCGReg rs, 421 int msb, int lsb) 422{ 423 if (lsb >= 32) { 424 opc = oph; 425 msb -= 32; 426 lsb -= 32; 427 } else if (msb >= 32) { 428 opc = opm; 429 msb -= 32; 430 } 431 tcg_out_opc_bf(s, opc, rt, rs, msb, lsb); 432} 433 434/* 435 * Type branch 436 */ 437static void tcg_out_opc_br(TCGContext *s, MIPSInsn opc, TCGReg rt, TCGReg rs) 438{ 439 tcg_out_opc_imm(s, opc, rt, rs, 0); 440} 441 442/* 443 * Type sa 444 */ 445static void tcg_out_opc_sa(TCGContext *s, MIPSInsn opc, 446 TCGReg rd, TCGReg rt, TCGArg sa) 447{ 448 int32_t inst; 449 450 inst = opc; 451 inst |= (rt & 0x1F) << 16; 452 inst |= (rd & 0x1F) << 11; 453 inst |= (sa & 0x1F) << 6; 454 tcg_out32(s, inst); 455 456} 457 458static void tcg_out_opc_sa64(TCGContext *s, MIPSInsn opc1, MIPSInsn opc2, 459 TCGReg rd, TCGReg rt, TCGArg sa) 460{ 461 int32_t inst; 462 463 inst = (sa & 32 ? opc2 : opc1); 464 inst |= (rt & 0x1F) << 16; 465 inst |= (rd & 0x1F) << 11; 466 inst |= (sa & 0x1F) << 6; 467 tcg_out32(s, inst); 468} 469 470/* 471 * Type jump. 472 * Returns true if the branch was in range and the insn was emitted. 473 */ 474static bool tcg_out_opc_jmp(TCGContext *s, MIPSInsn opc, const void *target) 475{ 476 uintptr_t dest = (uintptr_t)target; 477 uintptr_t from = (uintptr_t)tcg_splitwx_to_rx(s->code_ptr) + 4; 478 int32_t inst; 479 480 /* The pc-region branch happens within the 256MB region of 481 the delay slot (thus the +4). */ 482 if ((from ^ dest) & -(1 << 28)) { 483 return false; 484 } 485 tcg_debug_assert((dest & 3) == 0); 486 487 inst = opc; 488 inst |= (dest >> 2) & 0x3ffffff; 489 tcg_out32(s, inst); 490 return true; 491} 492 493static void tcg_out_nop(TCGContext *s) 494{ 495 tcg_out32(s, 0); 496} 497 498static void tcg_out_dsll(TCGContext *s, TCGReg rd, TCGReg rt, TCGArg sa) 499{ 500 tcg_out_opc_sa64(s, OPC_DSLL, OPC_DSLL32, rd, rt, sa); 501} 502 503static void tcg_out_dsrl(TCGContext *s, TCGReg rd, TCGReg rt, TCGArg sa) 504{ 505 tcg_out_opc_sa64(s, OPC_DSRL, OPC_DSRL32, rd, rt, sa); 506} 507 508static void tcg_out_dsra(TCGContext *s, TCGReg rd, TCGReg rt, TCGArg sa) 509{ 510 tcg_out_opc_sa64(s, OPC_DSRA, OPC_DSRA32, rd, rt, sa); 511} 512 513static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg) 514{ 515 /* Simple reg-reg move, optimising out the 'do nothing' case */ 516 if (ret != arg) { 517 tcg_out_opc_reg(s, OPC_OR, ret, arg, TCG_REG_ZERO); 518 } 519 return true; 520} 521 522static void tcg_out_movi(TCGContext *s, TCGType type, 523 TCGReg ret, tcg_target_long arg) 524{ 525 if (TCG_TARGET_REG_BITS == 64 && type == TCG_TYPE_I32) { 526 arg = (int32_t)arg; 527 } 528 if (arg == (int16_t)arg) { 529 tcg_out_opc_imm(s, OPC_ADDIU, ret, TCG_REG_ZERO, arg); 530 return; 531 } 532 if (arg == (uint16_t)arg) { 533 tcg_out_opc_imm(s, OPC_ORI, ret, TCG_REG_ZERO, arg); 534 return; 535 } 536 if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) { 537 tcg_out_opc_imm(s, OPC_LUI, ret, TCG_REG_ZERO, arg >> 16); 538 } else { 539 tcg_out_movi(s, TCG_TYPE_I32, ret, arg >> 31 >> 1); 540 if (arg & 0xffff0000ull) { 541 tcg_out_dsll(s, ret, ret, 16); 542 tcg_out_opc_imm(s, OPC_ORI, ret, ret, arg >> 16); 543 tcg_out_dsll(s, ret, ret, 16); 544 } else { 545 tcg_out_dsll(s, ret, ret, 32); 546 } 547 } 548 if (arg & 0xffff) { 549 tcg_out_opc_imm(s, OPC_ORI, ret, ret, arg & 0xffff); 550 } 551} 552 553static void tcg_out_bswap16(TCGContext *s, TCGReg ret, TCGReg arg, int flags) 554{ 555 /* ret and arg can't be register tmp0 */ 556 tcg_debug_assert(ret != TCG_TMP0); 557 tcg_debug_assert(arg != TCG_TMP0); 558 559 /* With arg = abcd: */ 560 if (use_mips32r2_instructions) { 561 tcg_out_opc_reg(s, OPC_WSBH, ret, 0, arg); /* badc */ 562 if (flags & TCG_BSWAP_OS) { 563 tcg_out_opc_reg(s, OPC_SEH, ret, 0, ret); /* ssdc */ 564 } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) { 565 tcg_out_opc_imm(s, OPC_ANDI, ret, ret, 0xffff); /* 00dc */ 566 } 567 return; 568 } 569 570 tcg_out_opc_sa(s, OPC_SRL, TCG_TMP0, arg, 8); /* 0abc */ 571 if (!(flags & TCG_BSWAP_IZ)) { 572 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP0, TCG_TMP0, 0x00ff); /* 000c */ 573 } 574 if (flags & TCG_BSWAP_OS) { 575 tcg_out_opc_sa(s, OPC_SLL, ret, arg, 24); /* d000 */ 576 tcg_out_opc_sa(s, OPC_SRA, ret, ret, 16); /* ssd0 */ 577 } else { 578 tcg_out_opc_sa(s, OPC_SLL, ret, arg, 8); /* bcd0 */ 579 if (flags & TCG_BSWAP_OZ) { 580 tcg_out_opc_imm(s, OPC_ANDI, ret, ret, 0xff00); /* 00d0 */ 581 } 582 } 583 tcg_out_opc_reg(s, OPC_OR, ret, ret, TCG_TMP0); /* ssdc */ 584} 585 586static void tcg_out_bswap_subr(TCGContext *s, const tcg_insn_unit *sub) 587{ 588 if (!tcg_out_opc_jmp(s, OPC_JAL, sub)) { 589 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP1, (uintptr_t)sub); 590 tcg_out_opc_reg(s, OPC_JALR, TCG_REG_RA, TCG_TMP1, 0); 591 } 592} 593 594static void tcg_out_bswap32(TCGContext *s, TCGReg ret, TCGReg arg, int flags) 595{ 596 if (use_mips32r2_instructions) { 597 tcg_out_opc_reg(s, OPC_WSBH, ret, 0, arg); 598 tcg_out_opc_sa(s, OPC_ROTR, ret, ret, 16); 599 if (flags & TCG_BSWAP_OZ) { 600 tcg_out_opc_bf(s, OPC_DEXT, ret, ret, 31, 0); 601 } 602 } else { 603 if (flags & TCG_BSWAP_OZ) { 604 tcg_out_bswap_subr(s, bswap32u_addr); 605 } else { 606 tcg_out_bswap_subr(s, bswap32_addr); 607 } 608 /* delay slot -- never omit the insn, like tcg_out_mov might. */ 609 tcg_out_opc_reg(s, OPC_OR, TCG_TMP0, arg, TCG_REG_ZERO); 610 tcg_out_mov(s, TCG_TYPE_I32, ret, TCG_TMP3); 611 } 612} 613 614static void tcg_out_bswap64(TCGContext *s, TCGReg ret, TCGReg arg) 615{ 616 if (use_mips32r2_instructions) { 617 tcg_out_opc_reg(s, OPC_DSBH, ret, 0, arg); 618 tcg_out_opc_reg(s, OPC_DSHD, ret, 0, ret); 619 } else { 620 tcg_out_bswap_subr(s, bswap64_addr); 621 /* delay slot -- never omit the insn, like tcg_out_mov might. */ 622 tcg_out_opc_reg(s, OPC_OR, TCG_TMP0, arg, TCG_REG_ZERO); 623 tcg_out_mov(s, TCG_TYPE_I32, ret, TCG_TMP3); 624 } 625} 626 627static void tcg_out_ext32u(TCGContext *s, TCGReg ret, TCGReg arg) 628{ 629 if (use_mips32r2_instructions) { 630 tcg_out_opc_bf(s, OPC_DEXT, ret, arg, 31, 0); 631 } else { 632 tcg_out_dsll(s, ret, arg, 32); 633 tcg_out_dsrl(s, ret, ret, 32); 634 } 635} 636 637static void tcg_out_ldst(TCGContext *s, MIPSInsn opc, TCGReg data, 638 TCGReg addr, intptr_t ofs) 639{ 640 int16_t lo = ofs; 641 if (ofs != lo) { 642 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - lo); 643 if (addr != TCG_REG_ZERO) { 644 tcg_out_opc_reg(s, ALIAS_PADD, TCG_TMP0, TCG_TMP0, addr); 645 } 646 addr = TCG_TMP0; 647 } 648 tcg_out_opc_imm(s, opc, data, addr, lo); 649} 650 651static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg, 652 TCGReg arg1, intptr_t arg2) 653{ 654 MIPSInsn opc = OPC_LD; 655 if (TCG_TARGET_REG_BITS == 32 || type == TCG_TYPE_I32) { 656 opc = OPC_LW; 657 } 658 tcg_out_ldst(s, opc, arg, arg1, arg2); 659} 660 661static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, 662 TCGReg arg1, intptr_t arg2) 663{ 664 MIPSInsn opc = OPC_SD; 665 if (TCG_TARGET_REG_BITS == 32 || type == TCG_TYPE_I32) { 666 opc = OPC_SW; 667 } 668 tcg_out_ldst(s, opc, arg, arg1, arg2); 669} 670 671static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val, 672 TCGReg base, intptr_t ofs) 673{ 674 if (val == 0) { 675 tcg_out_st(s, type, TCG_REG_ZERO, base, ofs); 676 return true; 677 } 678 return false; 679} 680 681static void tcg_out_addsub2(TCGContext *s, TCGReg rl, TCGReg rh, TCGReg al, 682 TCGReg ah, TCGArg bl, TCGArg bh, bool cbl, 683 bool cbh, bool is_sub) 684{ 685 TCGReg th = TCG_TMP1; 686 687 /* If we have a negative constant such that negating it would 688 make the high part zero, we can (usually) eliminate one insn. */ 689 if (cbl && cbh && bh == -1 && bl != 0) { 690 bl = -bl; 691 bh = 0; 692 is_sub = !is_sub; 693 } 694 695 /* By operating on the high part first, we get to use the final 696 carry operation to move back from the temporary. */ 697 if (!cbh) { 698 tcg_out_opc_reg(s, (is_sub ? OPC_SUBU : OPC_ADDU), th, ah, bh); 699 } else if (bh != 0 || ah == rl) { 700 tcg_out_opc_imm(s, OPC_ADDIU, th, ah, (is_sub ? -bh : bh)); 701 } else { 702 th = ah; 703 } 704 705 /* Note that tcg optimization should eliminate the bl == 0 case. */ 706 if (is_sub) { 707 if (cbl) { 708 tcg_out_opc_imm(s, OPC_SLTIU, TCG_TMP0, al, bl); 709 tcg_out_opc_imm(s, OPC_ADDIU, rl, al, -bl); 710 } else { 711 tcg_out_opc_reg(s, OPC_SLTU, TCG_TMP0, al, bl); 712 tcg_out_opc_reg(s, OPC_SUBU, rl, al, bl); 713 } 714 tcg_out_opc_reg(s, OPC_SUBU, rh, th, TCG_TMP0); 715 } else { 716 if (cbl) { 717 tcg_out_opc_imm(s, OPC_ADDIU, rl, al, bl); 718 tcg_out_opc_imm(s, OPC_SLTIU, TCG_TMP0, rl, bl); 719 } else if (rl == al && rl == bl) { 720 tcg_out_opc_sa(s, OPC_SRL, TCG_TMP0, al, TCG_TARGET_REG_BITS - 1); 721 tcg_out_opc_reg(s, OPC_ADDU, rl, al, bl); 722 } else { 723 tcg_out_opc_reg(s, OPC_ADDU, rl, al, bl); 724 tcg_out_opc_reg(s, OPC_SLTU, TCG_TMP0, rl, (rl == bl ? al : bl)); 725 } 726 tcg_out_opc_reg(s, OPC_ADDU, rh, th, TCG_TMP0); 727 } 728} 729 730/* Bit 0 set if inversion required; bit 1 set if swapping required. */ 731#define MIPS_CMP_INV 1 732#define MIPS_CMP_SWAP 2 733 734static const uint8_t mips_cmp_map[16] = { 735 [TCG_COND_LT] = 0, 736 [TCG_COND_LTU] = 0, 737 [TCG_COND_GE] = MIPS_CMP_INV, 738 [TCG_COND_GEU] = MIPS_CMP_INV, 739 [TCG_COND_LE] = MIPS_CMP_INV | MIPS_CMP_SWAP, 740 [TCG_COND_LEU] = MIPS_CMP_INV | MIPS_CMP_SWAP, 741 [TCG_COND_GT] = MIPS_CMP_SWAP, 742 [TCG_COND_GTU] = MIPS_CMP_SWAP, 743}; 744 745static void tcg_out_setcond(TCGContext *s, TCGCond cond, TCGReg ret, 746 TCGReg arg1, TCGReg arg2) 747{ 748 MIPSInsn s_opc = OPC_SLTU; 749 int cmp_map; 750 751 switch (cond) { 752 case TCG_COND_EQ: 753 if (arg2 != 0) { 754 tcg_out_opc_reg(s, OPC_XOR, ret, arg1, arg2); 755 arg1 = ret; 756 } 757 tcg_out_opc_imm(s, OPC_SLTIU, ret, arg1, 1); 758 break; 759 760 case TCG_COND_NE: 761 if (arg2 != 0) { 762 tcg_out_opc_reg(s, OPC_XOR, ret, arg1, arg2); 763 arg1 = ret; 764 } 765 tcg_out_opc_reg(s, OPC_SLTU, ret, TCG_REG_ZERO, arg1); 766 break; 767 768 case TCG_COND_LT: 769 case TCG_COND_GE: 770 case TCG_COND_LE: 771 case TCG_COND_GT: 772 s_opc = OPC_SLT; 773 /* FALLTHRU */ 774 775 case TCG_COND_LTU: 776 case TCG_COND_GEU: 777 case TCG_COND_LEU: 778 case TCG_COND_GTU: 779 cmp_map = mips_cmp_map[cond]; 780 if (cmp_map & MIPS_CMP_SWAP) { 781 TCGReg t = arg1; 782 arg1 = arg2; 783 arg2 = t; 784 } 785 tcg_out_opc_reg(s, s_opc, ret, arg1, arg2); 786 if (cmp_map & MIPS_CMP_INV) { 787 tcg_out_opc_imm(s, OPC_XORI, ret, ret, 1); 788 } 789 break; 790 791 default: 792 tcg_abort(); 793 break; 794 } 795} 796 797static void tcg_out_brcond(TCGContext *s, TCGCond cond, TCGReg arg1, 798 TCGReg arg2, TCGLabel *l) 799{ 800 static const MIPSInsn b_zero[16] = { 801 [TCG_COND_LT] = OPC_BLTZ, 802 [TCG_COND_GT] = OPC_BGTZ, 803 [TCG_COND_LE] = OPC_BLEZ, 804 [TCG_COND_GE] = OPC_BGEZ, 805 }; 806 807 MIPSInsn s_opc = OPC_SLTU; 808 MIPSInsn b_opc; 809 int cmp_map; 810 811 switch (cond) { 812 case TCG_COND_EQ: 813 b_opc = OPC_BEQ; 814 break; 815 case TCG_COND_NE: 816 b_opc = OPC_BNE; 817 break; 818 819 case TCG_COND_LT: 820 case TCG_COND_GT: 821 case TCG_COND_LE: 822 case TCG_COND_GE: 823 if (arg2 == 0) { 824 b_opc = b_zero[cond]; 825 arg2 = arg1; 826 arg1 = 0; 827 break; 828 } 829 s_opc = OPC_SLT; 830 /* FALLTHRU */ 831 832 case TCG_COND_LTU: 833 case TCG_COND_GTU: 834 case TCG_COND_LEU: 835 case TCG_COND_GEU: 836 cmp_map = mips_cmp_map[cond]; 837 if (cmp_map & MIPS_CMP_SWAP) { 838 TCGReg t = arg1; 839 arg1 = arg2; 840 arg2 = t; 841 } 842 tcg_out_opc_reg(s, s_opc, TCG_TMP0, arg1, arg2); 843 b_opc = (cmp_map & MIPS_CMP_INV ? OPC_BEQ : OPC_BNE); 844 arg1 = TCG_TMP0; 845 arg2 = TCG_REG_ZERO; 846 break; 847 848 default: 849 tcg_abort(); 850 break; 851 } 852 853 tcg_out_opc_br(s, b_opc, arg1, arg2); 854 tcg_out_reloc(s, s->code_ptr - 1, R_MIPS_PC16, l, 0); 855 tcg_out_nop(s); 856} 857 858static TCGReg tcg_out_reduce_eq2(TCGContext *s, TCGReg tmp0, TCGReg tmp1, 859 TCGReg al, TCGReg ah, 860 TCGReg bl, TCGReg bh) 861{ 862 /* Merge highpart comparison into AH. */ 863 if (bh != 0) { 864 if (ah != 0) { 865 tcg_out_opc_reg(s, OPC_XOR, tmp0, ah, bh); 866 ah = tmp0; 867 } else { 868 ah = bh; 869 } 870 } 871 /* Merge lowpart comparison into AL. */ 872 if (bl != 0) { 873 if (al != 0) { 874 tcg_out_opc_reg(s, OPC_XOR, tmp1, al, bl); 875 al = tmp1; 876 } else { 877 al = bl; 878 } 879 } 880 /* Merge high and low part comparisons into AL. */ 881 if (ah != 0) { 882 if (al != 0) { 883 tcg_out_opc_reg(s, OPC_OR, tmp0, ah, al); 884 al = tmp0; 885 } else { 886 al = ah; 887 } 888 } 889 return al; 890} 891 892static void tcg_out_setcond2(TCGContext *s, TCGCond cond, TCGReg ret, 893 TCGReg al, TCGReg ah, TCGReg bl, TCGReg bh) 894{ 895 TCGReg tmp0 = TCG_TMP0; 896 TCGReg tmp1 = ret; 897 898 tcg_debug_assert(ret != TCG_TMP0); 899 if (ret == ah || ret == bh) { 900 tcg_debug_assert(ret != TCG_TMP1); 901 tmp1 = TCG_TMP1; 902 } 903 904 switch (cond) { 905 case TCG_COND_EQ: 906 case TCG_COND_NE: 907 tmp1 = tcg_out_reduce_eq2(s, tmp0, tmp1, al, ah, bl, bh); 908 tcg_out_setcond(s, cond, ret, tmp1, TCG_REG_ZERO); 909 break; 910 911 default: 912 tcg_out_setcond(s, TCG_COND_EQ, tmp0, ah, bh); 913 tcg_out_setcond(s, tcg_unsigned_cond(cond), tmp1, al, bl); 914 tcg_out_opc_reg(s, OPC_AND, tmp1, tmp1, tmp0); 915 tcg_out_setcond(s, tcg_high_cond(cond), tmp0, ah, bh); 916 tcg_out_opc_reg(s, OPC_OR, ret, tmp1, tmp0); 917 break; 918 } 919} 920 921static void tcg_out_brcond2(TCGContext *s, TCGCond cond, TCGReg al, TCGReg ah, 922 TCGReg bl, TCGReg bh, TCGLabel *l) 923{ 924 TCGCond b_cond = TCG_COND_NE; 925 TCGReg tmp = TCG_TMP1; 926 927 /* With branches, we emit between 4 and 9 insns with 2 or 3 branches. 928 With setcond, we emit between 3 and 10 insns and only 1 branch, 929 which ought to get better branch prediction. */ 930 switch (cond) { 931 case TCG_COND_EQ: 932 case TCG_COND_NE: 933 b_cond = cond; 934 tmp = tcg_out_reduce_eq2(s, TCG_TMP0, TCG_TMP1, al, ah, bl, bh); 935 break; 936 937 default: 938 /* Minimize code size by preferring a compare not requiring INV. */ 939 if (mips_cmp_map[cond] & MIPS_CMP_INV) { 940 cond = tcg_invert_cond(cond); 941 b_cond = TCG_COND_EQ; 942 } 943 tcg_out_setcond2(s, cond, tmp, al, ah, bl, bh); 944 break; 945 } 946 947 tcg_out_brcond(s, b_cond, tmp, TCG_REG_ZERO, l); 948} 949 950static void tcg_out_movcond(TCGContext *s, TCGCond cond, TCGReg ret, 951 TCGReg c1, TCGReg c2, TCGReg v1, TCGReg v2) 952{ 953 bool eqz = false; 954 955 /* If one of the values is zero, put it last to match SEL*Z instructions */ 956 if (use_mips32r6_instructions && v1 == 0) { 957 v1 = v2; 958 v2 = 0; 959 cond = tcg_invert_cond(cond); 960 } 961 962 switch (cond) { 963 case TCG_COND_EQ: 964 eqz = true; 965 /* FALLTHRU */ 966 case TCG_COND_NE: 967 if (c2 != 0) { 968 tcg_out_opc_reg(s, OPC_XOR, TCG_TMP0, c1, c2); 969 c1 = TCG_TMP0; 970 } 971 break; 972 973 default: 974 /* Minimize code size by preferring a compare not requiring INV. */ 975 if (mips_cmp_map[cond] & MIPS_CMP_INV) { 976 cond = tcg_invert_cond(cond); 977 eqz = true; 978 } 979 tcg_out_setcond(s, cond, TCG_TMP0, c1, c2); 980 c1 = TCG_TMP0; 981 break; 982 } 983 984 if (use_mips32r6_instructions) { 985 MIPSInsn m_opc_t = eqz ? OPC_SELEQZ : OPC_SELNEZ; 986 MIPSInsn m_opc_f = eqz ? OPC_SELNEZ : OPC_SELEQZ; 987 988 if (v2 != 0) { 989 tcg_out_opc_reg(s, m_opc_f, TCG_TMP1, v2, c1); 990 } 991 tcg_out_opc_reg(s, m_opc_t, ret, v1, c1); 992 if (v2 != 0) { 993 tcg_out_opc_reg(s, OPC_OR, ret, ret, TCG_TMP1); 994 } 995 } else { 996 MIPSInsn m_opc = eqz ? OPC_MOVZ : OPC_MOVN; 997 998 tcg_out_opc_reg(s, m_opc, ret, v1, c1); 999 1000 /* This should be guaranteed via constraints */ 1001 tcg_debug_assert(v2 == ret); 1002 } 1003} 1004 1005static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail) 1006{ 1007 /* Note that the ABI requires the called function's address to be 1008 loaded into T9, even if a direct branch is in range. */ 1009 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T9, (uintptr_t)arg); 1010 1011 /* But do try a direct branch, allowing the cpu better insn prefetch. */ 1012 if (tail) { 1013 if (!tcg_out_opc_jmp(s, OPC_J, arg)) { 1014 tcg_out_opc_reg(s, OPC_JR, 0, TCG_REG_T9, 0); 1015 } 1016 } else { 1017 if (!tcg_out_opc_jmp(s, OPC_JAL, arg)) { 1018 tcg_out_opc_reg(s, OPC_JALR, TCG_REG_RA, TCG_REG_T9, 0); 1019 } 1020 } 1021} 1022 1023static void tcg_out_call(TCGContext *s, const tcg_insn_unit *arg, 1024 const TCGHelperInfo *info) 1025{ 1026 tcg_out_call_int(s, arg, false); 1027 tcg_out_nop(s); 1028} 1029 1030#if defined(CONFIG_SOFTMMU) 1031static void * const qemu_ld_helpers[(MO_SSIZE | MO_BSWAP) + 1] = { 1032 [MO_UB] = helper_ret_ldub_mmu, 1033 [MO_SB] = helper_ret_ldsb_mmu, 1034 [MO_LEUW] = helper_le_lduw_mmu, 1035 [MO_LESW] = helper_le_ldsw_mmu, 1036 [MO_LEUL] = helper_le_ldul_mmu, 1037 [MO_LEUQ] = helper_le_ldq_mmu, 1038 [MO_BEUW] = helper_be_lduw_mmu, 1039 [MO_BESW] = helper_be_ldsw_mmu, 1040 [MO_BEUL] = helper_be_ldul_mmu, 1041 [MO_BEUQ] = helper_be_ldq_mmu, 1042#if TCG_TARGET_REG_BITS == 64 1043 [MO_LESL] = helper_le_ldsl_mmu, 1044 [MO_BESL] = helper_be_ldsl_mmu, 1045#endif 1046}; 1047 1048static void * const qemu_st_helpers[(MO_SIZE | MO_BSWAP) + 1] = { 1049 [MO_UB] = helper_ret_stb_mmu, 1050 [MO_LEUW] = helper_le_stw_mmu, 1051 [MO_LEUL] = helper_le_stl_mmu, 1052 [MO_LEUQ] = helper_le_stq_mmu, 1053 [MO_BEUW] = helper_be_stw_mmu, 1054 [MO_BEUL] = helper_be_stl_mmu, 1055 [MO_BEUQ] = helper_be_stq_mmu, 1056}; 1057 1058/* Helper routines for marshalling helper function arguments into 1059 * the correct registers and stack. 1060 * I is where we want to put this argument, and is updated and returned 1061 * for the next call. ARG is the argument itself. 1062 * 1063 * We provide routines for arguments which are: immediate, 32 bit 1064 * value in register, 16 and 8 bit values in register (which must be zero 1065 * extended before use) and 64 bit value in a lo:hi register pair. 1066 */ 1067 1068static int tcg_out_call_iarg_reg(TCGContext *s, int i, TCGReg arg) 1069{ 1070 if (i < ARRAY_SIZE(tcg_target_call_iarg_regs)) { 1071 tcg_out_mov(s, TCG_TYPE_REG, tcg_target_call_iarg_regs[i], arg); 1072 } else { 1073 /* For N32 and N64, the initial offset is different. But there 1074 we also have 8 argument register so we don't run out here. */ 1075 tcg_debug_assert(TCG_TARGET_REG_BITS == 32); 1076 tcg_out_st(s, TCG_TYPE_REG, arg, TCG_REG_SP, 4 * i); 1077 } 1078 return i + 1; 1079} 1080 1081static int tcg_out_call_iarg_reg8(TCGContext *s, int i, TCGReg arg) 1082{ 1083 TCGReg tmp = TCG_TMP0; 1084 if (i < ARRAY_SIZE(tcg_target_call_iarg_regs)) { 1085 tmp = tcg_target_call_iarg_regs[i]; 1086 } 1087 tcg_out_opc_imm(s, OPC_ANDI, tmp, arg, 0xff); 1088 return tcg_out_call_iarg_reg(s, i, tmp); 1089} 1090 1091static int tcg_out_call_iarg_reg16(TCGContext *s, int i, TCGReg arg) 1092{ 1093 TCGReg tmp = TCG_TMP0; 1094 if (i < ARRAY_SIZE(tcg_target_call_iarg_regs)) { 1095 tmp = tcg_target_call_iarg_regs[i]; 1096 } 1097 tcg_out_opc_imm(s, OPC_ANDI, tmp, arg, 0xffff); 1098 return tcg_out_call_iarg_reg(s, i, tmp); 1099} 1100 1101static int tcg_out_call_iarg_imm(TCGContext *s, int i, TCGArg arg) 1102{ 1103 TCGReg tmp = TCG_TMP0; 1104 if (arg == 0) { 1105 tmp = TCG_REG_ZERO; 1106 } else { 1107 if (i < ARRAY_SIZE(tcg_target_call_iarg_regs)) { 1108 tmp = tcg_target_call_iarg_regs[i]; 1109 } 1110 tcg_out_movi(s, TCG_TYPE_REG, tmp, arg); 1111 } 1112 return tcg_out_call_iarg_reg(s, i, tmp); 1113} 1114 1115static int tcg_out_call_iarg_reg2(TCGContext *s, int i, TCGReg al, TCGReg ah) 1116{ 1117 tcg_debug_assert(TCG_TARGET_REG_BITS == 32); 1118 i = (i + 1) & ~1; 1119 i = tcg_out_call_iarg_reg(s, i, (MIPS_BE ? ah : al)); 1120 i = tcg_out_call_iarg_reg(s, i, (MIPS_BE ? al : ah)); 1121 return i; 1122} 1123 1124/* We expect to use a 16-bit negative offset from ENV. */ 1125QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0); 1126QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -32768); 1127 1128/* 1129 * Perform the tlb comparison operation. 1130 * The complete host address is placed in BASE. 1131 * Clobbers TMP0, TMP1, TMP2, TMP3. 1132 */ 1133static void tcg_out_tlb_load(TCGContext *s, TCGReg base, TCGReg addrl, 1134 TCGReg addrh, MemOpIdx oi, 1135 tcg_insn_unit *label_ptr[2], bool is_load) 1136{ 1137 MemOp opc = get_memop(oi); 1138 unsigned a_bits = get_alignment_bits(opc); 1139 unsigned s_bits = opc & MO_SIZE; 1140 unsigned a_mask = (1 << a_bits) - 1; 1141 unsigned s_mask = (1 << s_bits) - 1; 1142 int mem_index = get_mmuidx(oi); 1143 int fast_off = TLB_MASK_TABLE_OFS(mem_index); 1144 int mask_off = fast_off + offsetof(CPUTLBDescFast, mask); 1145 int table_off = fast_off + offsetof(CPUTLBDescFast, table); 1146 int add_off = offsetof(CPUTLBEntry, addend); 1147 int cmp_off = (is_load ? offsetof(CPUTLBEntry, addr_read) 1148 : offsetof(CPUTLBEntry, addr_write)); 1149 target_ulong tlb_mask; 1150 1151 /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx]. */ 1152 tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP0, TCG_AREG0, mask_off); 1153 tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP1, TCG_AREG0, table_off); 1154 1155 /* Extract the TLB index from the address into TMP3. */ 1156 tcg_out_opc_sa(s, ALIAS_TSRL, TCG_TMP3, addrl, 1157 TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS); 1158 tcg_out_opc_reg(s, OPC_AND, TCG_TMP3, TCG_TMP3, TCG_TMP0); 1159 1160 /* Add the tlb_table pointer, creating the CPUTLBEntry address in TMP3. */ 1161 tcg_out_opc_reg(s, ALIAS_PADD, TCG_TMP3, TCG_TMP3, TCG_TMP1); 1162 1163 /* Load the (low-half) tlb comparator. */ 1164 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) { 1165 tcg_out_ldst(s, OPC_LW, TCG_TMP0, TCG_TMP3, cmp_off + LO_OFF); 1166 } else { 1167 tcg_out_ldst(s, (TARGET_LONG_BITS == 64 ? OPC_LD 1168 : TCG_TARGET_REG_BITS == 64 ? OPC_LWU : OPC_LW), 1169 TCG_TMP0, TCG_TMP3, cmp_off); 1170 } 1171 1172 /* Zero extend a 32-bit guest address for a 64-bit host. */ 1173 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) { 1174 tcg_out_ext32u(s, base, addrl); 1175 addrl = base; 1176 } 1177 1178 /* 1179 * Mask the page bits, keeping the alignment bits to compare against. 1180 * For unaligned accesses, compare against the end of the access to 1181 * verify that it does not cross a page boundary. 1182 */ 1183 tlb_mask = (target_ulong)TARGET_PAGE_MASK | a_mask; 1184 tcg_out_movi(s, TCG_TYPE_I32, TCG_TMP1, tlb_mask); 1185 if (a_mask >= s_mask) { 1186 tcg_out_opc_reg(s, OPC_AND, TCG_TMP1, TCG_TMP1, addrl); 1187 } else { 1188 tcg_out_opc_imm(s, ALIAS_PADDI, TCG_TMP2, addrl, s_mask - a_mask); 1189 tcg_out_opc_reg(s, OPC_AND, TCG_TMP1, TCG_TMP1, TCG_TMP2); 1190 } 1191 1192 if (TCG_TARGET_REG_BITS >= TARGET_LONG_BITS) { 1193 /* Load the tlb addend for the fast path. */ 1194 tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP2, TCG_TMP3, add_off); 1195 } 1196 1197 label_ptr[0] = s->code_ptr; 1198 tcg_out_opc_br(s, OPC_BNE, TCG_TMP1, TCG_TMP0); 1199 1200 /* Load and test the high half tlb comparator. */ 1201 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) { 1202 /* delay slot */ 1203 tcg_out_ldst(s, OPC_LW, TCG_TMP0, TCG_TMP3, cmp_off + HI_OFF); 1204 1205 /* Load the tlb addend for the fast path. */ 1206 tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP2, TCG_TMP3, add_off); 1207 1208 label_ptr[1] = s->code_ptr; 1209 tcg_out_opc_br(s, OPC_BNE, addrh, TCG_TMP0); 1210 } 1211 1212 /* delay slot */ 1213 tcg_out_opc_reg(s, ALIAS_PADD, base, TCG_TMP2, addrl); 1214} 1215 1216static void add_qemu_ldst_label(TCGContext *s, int is_ld, MemOpIdx oi, 1217 TCGType ext, 1218 TCGReg datalo, TCGReg datahi, 1219 TCGReg addrlo, TCGReg addrhi, 1220 void *raddr, tcg_insn_unit *label_ptr[2]) 1221{ 1222 TCGLabelQemuLdst *label = new_ldst_label(s); 1223 1224 label->is_ld = is_ld; 1225 label->oi = oi; 1226 label->type = ext; 1227 label->datalo_reg = datalo; 1228 label->datahi_reg = datahi; 1229 label->addrlo_reg = addrlo; 1230 label->addrhi_reg = addrhi; 1231 label->raddr = tcg_splitwx_to_rx(raddr); 1232 label->label_ptr[0] = label_ptr[0]; 1233 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) { 1234 label->label_ptr[1] = label_ptr[1]; 1235 } 1236} 1237 1238static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l) 1239{ 1240 const tcg_insn_unit *tgt_rx = tcg_splitwx_to_rx(s->code_ptr); 1241 MemOpIdx oi = l->oi; 1242 MemOp opc = get_memop(oi); 1243 TCGReg v0; 1244 int i; 1245 1246 /* resolve label address */ 1247 if (!reloc_pc16(l->label_ptr[0], tgt_rx) 1248 || (TCG_TARGET_REG_BITS < TARGET_LONG_BITS 1249 && !reloc_pc16(l->label_ptr[1], tgt_rx))) { 1250 return false; 1251 } 1252 1253 i = 1; 1254 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) { 1255 i = tcg_out_call_iarg_reg2(s, i, l->addrlo_reg, l->addrhi_reg); 1256 } else { 1257 i = tcg_out_call_iarg_reg(s, i, l->addrlo_reg); 1258 } 1259 i = tcg_out_call_iarg_imm(s, i, oi); 1260 i = tcg_out_call_iarg_imm(s, i, (intptr_t)l->raddr); 1261 tcg_out_call_int(s, qemu_ld_helpers[opc & (MO_BSWAP | MO_SSIZE)], false); 1262 /* delay slot */ 1263 tcg_out_mov(s, TCG_TYPE_PTR, tcg_target_call_iarg_regs[0], TCG_AREG0); 1264 1265 v0 = l->datalo_reg; 1266 if (TCG_TARGET_REG_BITS == 32 && (opc & MO_SIZE) == MO_64) { 1267 /* We eliminated V0 from the possible output registers, so it 1268 cannot be clobbered here. So we must move V1 first. */ 1269 if (MIPS_BE) { 1270 tcg_out_mov(s, TCG_TYPE_I32, v0, TCG_REG_V1); 1271 v0 = l->datahi_reg; 1272 } else { 1273 tcg_out_mov(s, TCG_TYPE_I32, l->datahi_reg, TCG_REG_V1); 1274 } 1275 } 1276 1277 tcg_out_opc_br(s, OPC_BEQ, TCG_REG_ZERO, TCG_REG_ZERO); 1278 if (!reloc_pc16(s->code_ptr - 1, l->raddr)) { 1279 return false; 1280 } 1281 1282 /* delay slot */ 1283 if (TCG_TARGET_REG_BITS == 64 && l->type == TCG_TYPE_I32) { 1284 /* we always sign-extend 32-bit loads */ 1285 tcg_out_opc_sa(s, OPC_SLL, v0, TCG_REG_V0, 0); 1286 } else { 1287 tcg_out_opc_reg(s, OPC_OR, v0, TCG_REG_V0, TCG_REG_ZERO); 1288 } 1289 return true; 1290} 1291 1292static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l) 1293{ 1294 const tcg_insn_unit *tgt_rx = tcg_splitwx_to_rx(s->code_ptr); 1295 MemOpIdx oi = l->oi; 1296 MemOp opc = get_memop(oi); 1297 MemOp s_bits = opc & MO_SIZE; 1298 int i; 1299 1300 /* resolve label address */ 1301 if (!reloc_pc16(l->label_ptr[0], tgt_rx) 1302 || (TCG_TARGET_REG_BITS < TARGET_LONG_BITS 1303 && !reloc_pc16(l->label_ptr[1], tgt_rx))) { 1304 return false; 1305 } 1306 1307 i = 1; 1308 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) { 1309 i = tcg_out_call_iarg_reg2(s, i, l->addrlo_reg, l->addrhi_reg); 1310 } else { 1311 i = tcg_out_call_iarg_reg(s, i, l->addrlo_reg); 1312 } 1313 switch (s_bits) { 1314 case MO_8: 1315 i = tcg_out_call_iarg_reg8(s, i, l->datalo_reg); 1316 break; 1317 case MO_16: 1318 i = tcg_out_call_iarg_reg16(s, i, l->datalo_reg); 1319 break; 1320 case MO_32: 1321 i = tcg_out_call_iarg_reg(s, i, l->datalo_reg); 1322 break; 1323 case MO_64: 1324 if (TCG_TARGET_REG_BITS == 32) { 1325 i = tcg_out_call_iarg_reg2(s, i, l->datalo_reg, l->datahi_reg); 1326 } else { 1327 i = tcg_out_call_iarg_reg(s, i, l->datalo_reg); 1328 } 1329 break; 1330 default: 1331 tcg_abort(); 1332 } 1333 i = tcg_out_call_iarg_imm(s, i, oi); 1334 1335 /* Tail call to the store helper. Thus force the return address 1336 computation to take place in the return address register. */ 1337 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_RA, (intptr_t)l->raddr); 1338 i = tcg_out_call_iarg_reg(s, i, TCG_REG_RA); 1339 tcg_out_call_int(s, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)], true); 1340 /* delay slot */ 1341 tcg_out_mov(s, TCG_TYPE_PTR, tcg_target_call_iarg_regs[0], TCG_AREG0); 1342 return true; 1343} 1344 1345#else 1346 1347static void tcg_out_test_alignment(TCGContext *s, bool is_ld, TCGReg addrlo, 1348 TCGReg addrhi, unsigned a_bits) 1349{ 1350 unsigned a_mask = (1 << a_bits) - 1; 1351 TCGLabelQemuLdst *l = new_ldst_label(s); 1352 1353 l->is_ld = is_ld; 1354 l->addrlo_reg = addrlo; 1355 l->addrhi_reg = addrhi; 1356 1357 /* We are expecting a_bits to max out at 7, much lower than ANDI. */ 1358 tcg_debug_assert(a_bits < 16); 1359 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP0, addrlo, a_mask); 1360 1361 l->label_ptr[0] = s->code_ptr; 1362 if (use_mips32r6_instructions) { 1363 tcg_out_opc_br(s, OPC_BNEZALC_R6, TCG_REG_ZERO, TCG_TMP0); 1364 } else { 1365 tcg_out_opc_br(s, OPC_BNEL, TCG_TMP0, TCG_REG_ZERO); 1366 tcg_out_nop(s); 1367 } 1368 1369 l->raddr = tcg_splitwx_to_rx(s->code_ptr); 1370} 1371 1372static bool tcg_out_fail_alignment(TCGContext *s, TCGLabelQemuLdst *l) 1373{ 1374 void *target; 1375 1376 if (!reloc_pc16(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) { 1377 return false; 1378 } 1379 1380 if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) { 1381 /* A0 is env, A1 is skipped, A2:A3 is the uint64_t address. */ 1382 TCGReg a2 = MIPS_BE ? l->addrhi_reg : l->addrlo_reg; 1383 TCGReg a3 = MIPS_BE ? l->addrlo_reg : l->addrhi_reg; 1384 1385 if (a3 != TCG_REG_A2) { 1386 tcg_out_mov(s, TCG_TYPE_I32, TCG_REG_A2, a2); 1387 tcg_out_mov(s, TCG_TYPE_I32, TCG_REG_A3, a3); 1388 } else if (a2 != TCG_REG_A3) { 1389 tcg_out_mov(s, TCG_TYPE_I32, TCG_REG_A3, a3); 1390 tcg_out_mov(s, TCG_TYPE_I32, TCG_REG_A2, a2); 1391 } else { 1392 tcg_out_mov(s, TCG_TYPE_I32, TCG_TMP0, TCG_REG_A2); 1393 tcg_out_mov(s, TCG_TYPE_I32, TCG_REG_A2, TCG_REG_A3); 1394 tcg_out_mov(s, TCG_TYPE_I32, TCG_REG_A3, TCG_TMP0); 1395 } 1396 } else { 1397 tcg_out_mov(s, TCG_TYPE_TL, TCG_REG_A1, l->addrlo_reg); 1398 } 1399 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_A0, TCG_AREG0); 1400 1401 /* 1402 * Tail call to the helper, with the return address back inline. 1403 * We have arrived here via BNEL, so $31 is already set. 1404 */ 1405 target = (l->is_ld ? helper_unaligned_ld : helper_unaligned_st); 1406 tcg_out_call_int(s, target, true); 1407 return true; 1408} 1409 1410static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l) 1411{ 1412 return tcg_out_fail_alignment(s, l); 1413} 1414 1415static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l) 1416{ 1417 return tcg_out_fail_alignment(s, l); 1418} 1419#endif /* SOFTMMU */ 1420 1421static void tcg_out_qemu_ld_direct(TCGContext *s, TCGReg lo, TCGReg hi, 1422 TCGReg base, MemOp opc, bool is_64) 1423{ 1424 switch (opc & (MO_SSIZE | MO_BSWAP)) { 1425 case MO_UB: 1426 tcg_out_opc_imm(s, OPC_LBU, lo, base, 0); 1427 break; 1428 case MO_SB: 1429 tcg_out_opc_imm(s, OPC_LB, lo, base, 0); 1430 break; 1431 case MO_UW | MO_BSWAP: 1432 tcg_out_opc_imm(s, OPC_LHU, TCG_TMP1, base, 0); 1433 tcg_out_bswap16(s, lo, TCG_TMP1, TCG_BSWAP_IZ | TCG_BSWAP_OZ); 1434 break; 1435 case MO_UW: 1436 tcg_out_opc_imm(s, OPC_LHU, lo, base, 0); 1437 break; 1438 case MO_SW | MO_BSWAP: 1439 tcg_out_opc_imm(s, OPC_LHU, TCG_TMP1, base, 0); 1440 tcg_out_bswap16(s, lo, TCG_TMP1, TCG_BSWAP_IZ | TCG_BSWAP_OS); 1441 break; 1442 case MO_SW: 1443 tcg_out_opc_imm(s, OPC_LH, lo, base, 0); 1444 break; 1445 case MO_UL | MO_BSWAP: 1446 if (TCG_TARGET_REG_BITS == 64 && is_64) { 1447 if (use_mips32r2_instructions) { 1448 tcg_out_opc_imm(s, OPC_LWU, lo, base, 0); 1449 tcg_out_bswap32(s, lo, lo, TCG_BSWAP_IZ | TCG_BSWAP_OZ); 1450 } else { 1451 tcg_out_bswap_subr(s, bswap32u_addr); 1452 /* delay slot */ 1453 tcg_out_opc_imm(s, OPC_LWU, TCG_TMP0, base, 0); 1454 tcg_out_mov(s, TCG_TYPE_I64, lo, TCG_TMP3); 1455 } 1456 break; 1457 } 1458 /* FALLTHRU */ 1459 case MO_SL | MO_BSWAP: 1460 if (use_mips32r2_instructions) { 1461 tcg_out_opc_imm(s, OPC_LW, lo, base, 0); 1462 tcg_out_bswap32(s, lo, lo, 0); 1463 } else { 1464 tcg_out_bswap_subr(s, bswap32_addr); 1465 /* delay slot */ 1466 tcg_out_opc_imm(s, OPC_LW, TCG_TMP0, base, 0); 1467 tcg_out_mov(s, TCG_TYPE_I32, lo, TCG_TMP3); 1468 } 1469 break; 1470 case MO_UL: 1471 if (TCG_TARGET_REG_BITS == 64 && is_64) { 1472 tcg_out_opc_imm(s, OPC_LWU, lo, base, 0); 1473 break; 1474 } 1475 /* FALLTHRU */ 1476 case MO_SL: 1477 tcg_out_opc_imm(s, OPC_LW, lo, base, 0); 1478 break; 1479 case MO_UQ | MO_BSWAP: 1480 if (TCG_TARGET_REG_BITS == 64) { 1481 if (use_mips32r2_instructions) { 1482 tcg_out_opc_imm(s, OPC_LD, lo, base, 0); 1483 tcg_out_bswap64(s, lo, lo); 1484 } else { 1485 tcg_out_bswap_subr(s, bswap64_addr); 1486 /* delay slot */ 1487 tcg_out_opc_imm(s, OPC_LD, TCG_TMP0, base, 0); 1488 tcg_out_mov(s, TCG_TYPE_I64, lo, TCG_TMP3); 1489 } 1490 } else if (use_mips32r2_instructions) { 1491 tcg_out_opc_imm(s, OPC_LW, TCG_TMP0, base, 0); 1492 tcg_out_opc_imm(s, OPC_LW, TCG_TMP1, base, 4); 1493 tcg_out_opc_reg(s, OPC_WSBH, TCG_TMP0, 0, TCG_TMP0); 1494 tcg_out_opc_reg(s, OPC_WSBH, TCG_TMP1, 0, TCG_TMP1); 1495 tcg_out_opc_sa(s, OPC_ROTR, MIPS_BE ? lo : hi, TCG_TMP0, 16); 1496 tcg_out_opc_sa(s, OPC_ROTR, MIPS_BE ? hi : lo, TCG_TMP1, 16); 1497 } else { 1498 tcg_out_bswap_subr(s, bswap32_addr); 1499 /* delay slot */ 1500 tcg_out_opc_imm(s, OPC_LW, TCG_TMP0, base, 0); 1501 tcg_out_opc_imm(s, OPC_LW, TCG_TMP0, base, 4); 1502 tcg_out_bswap_subr(s, bswap32_addr); 1503 /* delay slot */ 1504 tcg_out_mov(s, TCG_TYPE_I32, MIPS_BE ? lo : hi, TCG_TMP3); 1505 tcg_out_mov(s, TCG_TYPE_I32, MIPS_BE ? hi : lo, TCG_TMP3); 1506 } 1507 break; 1508 case MO_UQ: 1509 /* Prefer to load from offset 0 first, but allow for overlap. */ 1510 if (TCG_TARGET_REG_BITS == 64) { 1511 tcg_out_opc_imm(s, OPC_LD, lo, base, 0); 1512 } else if (MIPS_BE ? hi != base : lo == base) { 1513 tcg_out_opc_imm(s, OPC_LW, hi, base, HI_OFF); 1514 tcg_out_opc_imm(s, OPC_LW, lo, base, LO_OFF); 1515 } else { 1516 tcg_out_opc_imm(s, OPC_LW, lo, base, LO_OFF); 1517 tcg_out_opc_imm(s, OPC_LW, hi, base, HI_OFF); 1518 } 1519 break; 1520 default: 1521 tcg_abort(); 1522 } 1523} 1524 1525static void tcg_out_qemu_ld_unalign(TCGContext *s, TCGReg lo, TCGReg hi, 1526 TCGReg base, MemOp opc, bool is_64) 1527{ 1528 const MIPSInsn lw1 = MIPS_BE ? OPC_LWL : OPC_LWR; 1529 const MIPSInsn lw2 = MIPS_BE ? OPC_LWR : OPC_LWL; 1530 const MIPSInsn ld1 = MIPS_BE ? OPC_LDL : OPC_LDR; 1531 const MIPSInsn ld2 = MIPS_BE ? OPC_LDR : OPC_LDL; 1532 1533 bool sgn = (opc & MO_SIGN); 1534 1535 switch (opc & (MO_SSIZE | MO_BSWAP)) { 1536 case MO_SW | MO_BE: 1537 case MO_UW | MO_BE: 1538 tcg_out_opc_imm(s, sgn ? OPC_LB : OPC_LBU, TCG_TMP0, base, 0); 1539 tcg_out_opc_imm(s, OPC_LBU, lo, base, 1); 1540 if (use_mips32r2_instructions) { 1541 tcg_out_opc_bf(s, OPC_INS, lo, TCG_TMP0, 31, 8); 1542 } else { 1543 tcg_out_opc_sa(s, OPC_SLL, TCG_TMP0, TCG_TMP0, 8); 1544 tcg_out_opc_reg(s, OPC_OR, lo, TCG_TMP0, TCG_TMP1); 1545 } 1546 break; 1547 1548 case MO_SW | MO_LE: 1549 case MO_UW | MO_LE: 1550 if (use_mips32r2_instructions && lo != base) { 1551 tcg_out_opc_imm(s, OPC_LBU, lo, base, 0); 1552 tcg_out_opc_imm(s, sgn ? OPC_LB : OPC_LBU, TCG_TMP0, base, 1); 1553 tcg_out_opc_bf(s, OPC_INS, lo, TCG_TMP0, 31, 8); 1554 } else { 1555 tcg_out_opc_imm(s, OPC_LBU, TCG_TMP0, base, 0); 1556 tcg_out_opc_imm(s, sgn ? OPC_LB : OPC_LBU, TCG_TMP1, base, 1); 1557 tcg_out_opc_sa(s, OPC_SLL, TCG_TMP1, TCG_TMP1, 8); 1558 tcg_out_opc_reg(s, OPC_OR, lo, TCG_TMP0, TCG_TMP1); 1559 } 1560 break; 1561 1562 case MO_SL: 1563 case MO_UL: 1564 tcg_out_opc_imm(s, lw1, lo, base, 0); 1565 tcg_out_opc_imm(s, lw2, lo, base, 3); 1566 if (TCG_TARGET_REG_BITS == 64 && is_64 && !sgn) { 1567 tcg_out_ext32u(s, lo, lo); 1568 } 1569 break; 1570 1571 case MO_UL | MO_BSWAP: 1572 case MO_SL | MO_BSWAP: 1573 if (use_mips32r2_instructions) { 1574 tcg_out_opc_imm(s, lw1, lo, base, 0); 1575 tcg_out_opc_imm(s, lw2, lo, base, 3); 1576 tcg_out_bswap32(s, lo, lo, 1577 TCG_TARGET_REG_BITS == 64 && is_64 1578 ? (sgn ? TCG_BSWAP_OS : TCG_BSWAP_OZ) : 0); 1579 } else { 1580 const tcg_insn_unit *subr = 1581 (TCG_TARGET_REG_BITS == 64 && is_64 && !sgn 1582 ? bswap32u_addr : bswap32_addr); 1583 1584 tcg_out_opc_imm(s, lw1, TCG_TMP0, base, 0); 1585 tcg_out_bswap_subr(s, subr); 1586 /* delay slot */ 1587 tcg_out_opc_imm(s, lw2, TCG_TMP0, base, 3); 1588 tcg_out_mov(s, is_64 ? TCG_TYPE_I64 : TCG_TYPE_I32, lo, TCG_TMP3); 1589 } 1590 break; 1591 1592 case MO_UQ: 1593 if (TCG_TARGET_REG_BITS == 64) { 1594 tcg_out_opc_imm(s, ld1, lo, base, 0); 1595 tcg_out_opc_imm(s, ld2, lo, base, 7); 1596 } else { 1597 tcg_out_opc_imm(s, lw1, MIPS_BE ? hi : lo, base, 0 + 0); 1598 tcg_out_opc_imm(s, lw2, MIPS_BE ? hi : lo, base, 0 + 3); 1599 tcg_out_opc_imm(s, lw1, MIPS_BE ? lo : hi, base, 4 + 0); 1600 tcg_out_opc_imm(s, lw2, MIPS_BE ? lo : hi, base, 4 + 3); 1601 } 1602 break; 1603 1604 case MO_UQ | MO_BSWAP: 1605 if (TCG_TARGET_REG_BITS == 64) { 1606 if (use_mips32r2_instructions) { 1607 tcg_out_opc_imm(s, ld1, lo, base, 0); 1608 tcg_out_opc_imm(s, ld2, lo, base, 7); 1609 tcg_out_bswap64(s, lo, lo); 1610 } else { 1611 tcg_out_opc_imm(s, ld1, TCG_TMP0, base, 0); 1612 tcg_out_bswap_subr(s, bswap64_addr); 1613 /* delay slot */ 1614 tcg_out_opc_imm(s, ld2, TCG_TMP0, base, 7); 1615 tcg_out_mov(s, TCG_TYPE_I64, lo, TCG_TMP3); 1616 } 1617 } else if (use_mips32r2_instructions) { 1618 tcg_out_opc_imm(s, lw1, TCG_TMP0, base, 0 + 0); 1619 tcg_out_opc_imm(s, lw2, TCG_TMP0, base, 0 + 3); 1620 tcg_out_opc_imm(s, lw1, TCG_TMP1, base, 4 + 0); 1621 tcg_out_opc_imm(s, lw2, TCG_TMP1, base, 4 + 3); 1622 tcg_out_opc_reg(s, OPC_WSBH, TCG_TMP0, 0, TCG_TMP0); 1623 tcg_out_opc_reg(s, OPC_WSBH, TCG_TMP1, 0, TCG_TMP1); 1624 tcg_out_opc_sa(s, OPC_ROTR, MIPS_BE ? lo : hi, TCG_TMP0, 16); 1625 tcg_out_opc_sa(s, OPC_ROTR, MIPS_BE ? hi : lo, TCG_TMP1, 16); 1626 } else { 1627 tcg_out_opc_imm(s, lw1, TCG_TMP0, base, 0 + 0); 1628 tcg_out_bswap_subr(s, bswap32_addr); 1629 /* delay slot */ 1630 tcg_out_opc_imm(s, lw2, TCG_TMP0, base, 0 + 3); 1631 tcg_out_opc_imm(s, lw1, TCG_TMP0, base, 4 + 0); 1632 tcg_out_mov(s, TCG_TYPE_I32, MIPS_BE ? lo : hi, TCG_TMP3); 1633 tcg_out_bswap_subr(s, bswap32_addr); 1634 /* delay slot */ 1635 tcg_out_opc_imm(s, lw2, TCG_TMP0, base, 4 + 3); 1636 tcg_out_mov(s, TCG_TYPE_I32, MIPS_BE ? hi : lo, TCG_TMP3); 1637 } 1638 break; 1639 1640 default: 1641 g_assert_not_reached(); 1642 } 1643} 1644 1645static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64) 1646{ 1647 TCGReg addr_regl, addr_regh __attribute__((unused)); 1648 TCGReg data_regl, data_regh; 1649 MemOpIdx oi; 1650 MemOp opc; 1651#if defined(CONFIG_SOFTMMU) 1652 tcg_insn_unit *label_ptr[2]; 1653#else 1654#endif 1655 unsigned a_bits, s_bits; 1656 TCGReg base = TCG_REG_A0; 1657 1658 data_regl = *args++; 1659 data_regh = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0); 1660 addr_regl = *args++; 1661 addr_regh = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0); 1662 oi = *args++; 1663 opc = get_memop(oi); 1664 a_bits = get_alignment_bits(opc); 1665 s_bits = opc & MO_SIZE; 1666 1667 /* 1668 * R6 removes the left/right instructions but requires the 1669 * system to support misaligned memory accesses. 1670 */ 1671#if defined(CONFIG_SOFTMMU) 1672 tcg_out_tlb_load(s, base, addr_regl, addr_regh, oi, label_ptr, 1); 1673 if (use_mips32r6_instructions || a_bits >= s_bits) { 1674 tcg_out_qemu_ld_direct(s, data_regl, data_regh, base, opc, is_64); 1675 } else { 1676 tcg_out_qemu_ld_unalign(s, data_regl, data_regh, base, opc, is_64); 1677 } 1678 add_qemu_ldst_label(s, 1, oi, 1679 (is_64 ? TCG_TYPE_I64 : TCG_TYPE_I32), 1680 data_regl, data_regh, addr_regl, addr_regh, 1681 s->code_ptr, label_ptr); 1682#else 1683 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) { 1684 tcg_out_ext32u(s, base, addr_regl); 1685 addr_regl = base; 1686 } 1687 if (guest_base == 0 && data_regl != addr_regl) { 1688 base = addr_regl; 1689 } else if (guest_base == (int16_t)guest_base) { 1690 tcg_out_opc_imm(s, ALIAS_PADDI, base, addr_regl, guest_base); 1691 } else { 1692 tcg_out_opc_reg(s, ALIAS_PADD, base, TCG_GUEST_BASE_REG, addr_regl); 1693 } 1694 if (use_mips32r6_instructions) { 1695 if (a_bits) { 1696 tcg_out_test_alignment(s, true, addr_regl, addr_regh, a_bits); 1697 } 1698 tcg_out_qemu_ld_direct(s, data_regl, data_regh, base, opc, is_64); 1699 } else { 1700 if (a_bits && a_bits != s_bits) { 1701 tcg_out_test_alignment(s, true, addr_regl, addr_regh, a_bits); 1702 } 1703 if (a_bits >= s_bits) { 1704 tcg_out_qemu_ld_direct(s, data_regl, data_regh, base, opc, is_64); 1705 } else { 1706 tcg_out_qemu_ld_unalign(s, data_regl, data_regh, base, opc, is_64); 1707 } 1708 } 1709#endif 1710} 1711 1712static void tcg_out_qemu_st_direct(TCGContext *s, TCGReg lo, TCGReg hi, 1713 TCGReg base, MemOp opc) 1714{ 1715 /* Don't clutter the code below with checks to avoid bswapping ZERO. */ 1716 if ((lo | hi) == 0) { 1717 opc &= ~MO_BSWAP; 1718 } 1719 1720 switch (opc & (MO_SIZE | MO_BSWAP)) { 1721 case MO_8: 1722 tcg_out_opc_imm(s, OPC_SB, lo, base, 0); 1723 break; 1724 1725 case MO_16 | MO_BSWAP: 1726 tcg_out_bswap16(s, TCG_TMP1, lo, 0); 1727 lo = TCG_TMP1; 1728 /* FALLTHRU */ 1729 case MO_16: 1730 tcg_out_opc_imm(s, OPC_SH, lo, base, 0); 1731 break; 1732 1733 case MO_32 | MO_BSWAP: 1734 tcg_out_bswap32(s, TCG_TMP3, lo, 0); 1735 lo = TCG_TMP3; 1736 /* FALLTHRU */ 1737 case MO_32: 1738 tcg_out_opc_imm(s, OPC_SW, lo, base, 0); 1739 break; 1740 1741 case MO_64 | MO_BSWAP: 1742 if (TCG_TARGET_REG_BITS == 64) { 1743 tcg_out_bswap64(s, TCG_TMP3, lo); 1744 tcg_out_opc_imm(s, OPC_SD, TCG_TMP3, base, 0); 1745 } else if (use_mips32r2_instructions) { 1746 tcg_out_opc_reg(s, OPC_WSBH, TCG_TMP0, 0, MIPS_BE ? lo : hi); 1747 tcg_out_opc_reg(s, OPC_WSBH, TCG_TMP1, 0, MIPS_BE ? hi : lo); 1748 tcg_out_opc_sa(s, OPC_ROTR, TCG_TMP0, TCG_TMP0, 16); 1749 tcg_out_opc_sa(s, OPC_ROTR, TCG_TMP1, TCG_TMP1, 16); 1750 tcg_out_opc_imm(s, OPC_SW, TCG_TMP0, base, 0); 1751 tcg_out_opc_imm(s, OPC_SW, TCG_TMP1, base, 4); 1752 } else { 1753 tcg_out_bswap32(s, TCG_TMP3, MIPS_BE ? lo : hi, 0); 1754 tcg_out_opc_imm(s, OPC_SW, TCG_TMP3, base, 0); 1755 tcg_out_bswap32(s, TCG_TMP3, MIPS_BE ? hi : lo, 0); 1756 tcg_out_opc_imm(s, OPC_SW, TCG_TMP3, base, 4); 1757 } 1758 break; 1759 case MO_64: 1760 if (TCG_TARGET_REG_BITS == 64) { 1761 tcg_out_opc_imm(s, OPC_SD, lo, base, 0); 1762 } else { 1763 tcg_out_opc_imm(s, OPC_SW, MIPS_BE ? hi : lo, base, 0); 1764 tcg_out_opc_imm(s, OPC_SW, MIPS_BE ? lo : hi, base, 4); 1765 } 1766 break; 1767 1768 default: 1769 tcg_abort(); 1770 } 1771} 1772 1773static void tcg_out_qemu_st_unalign(TCGContext *s, TCGReg lo, TCGReg hi, 1774 TCGReg base, MemOp opc) 1775{ 1776 const MIPSInsn sw1 = MIPS_BE ? OPC_SWL : OPC_SWR; 1777 const MIPSInsn sw2 = MIPS_BE ? OPC_SWR : OPC_SWL; 1778 const MIPSInsn sd1 = MIPS_BE ? OPC_SDL : OPC_SDR; 1779 const MIPSInsn sd2 = MIPS_BE ? OPC_SDR : OPC_SDL; 1780 1781 /* Don't clutter the code below with checks to avoid bswapping ZERO. */ 1782 if ((lo | hi) == 0) { 1783 opc &= ~MO_BSWAP; 1784 } 1785 1786 switch (opc & (MO_SIZE | MO_BSWAP)) { 1787 case MO_16 | MO_BE: 1788 tcg_out_opc_sa(s, OPC_SRL, TCG_TMP0, lo, 8); 1789 tcg_out_opc_imm(s, OPC_SB, TCG_TMP0, base, 0); 1790 tcg_out_opc_imm(s, OPC_SB, lo, base, 1); 1791 break; 1792 1793 case MO_16 | MO_LE: 1794 tcg_out_opc_sa(s, OPC_SRL, TCG_TMP0, lo, 8); 1795 tcg_out_opc_imm(s, OPC_SB, lo, base, 0); 1796 tcg_out_opc_imm(s, OPC_SB, TCG_TMP0, base, 1); 1797 break; 1798 1799 case MO_32 | MO_BSWAP: 1800 tcg_out_bswap32(s, TCG_TMP3, lo, 0); 1801 lo = TCG_TMP3; 1802 /* fall through */ 1803 case MO_32: 1804 tcg_out_opc_imm(s, sw1, lo, base, 0); 1805 tcg_out_opc_imm(s, sw2, lo, base, 3); 1806 break; 1807 1808 case MO_64 | MO_BSWAP: 1809 if (TCG_TARGET_REG_BITS == 64) { 1810 tcg_out_bswap64(s, TCG_TMP3, lo); 1811 lo = TCG_TMP3; 1812 } else if (use_mips32r2_instructions) { 1813 tcg_out_opc_reg(s, OPC_WSBH, TCG_TMP0, 0, MIPS_BE ? hi : lo); 1814 tcg_out_opc_reg(s, OPC_WSBH, TCG_TMP1, 0, MIPS_BE ? lo : hi); 1815 tcg_out_opc_sa(s, OPC_ROTR, TCG_TMP0, TCG_TMP0, 16); 1816 tcg_out_opc_sa(s, OPC_ROTR, TCG_TMP1, TCG_TMP1, 16); 1817 hi = MIPS_BE ? TCG_TMP0 : TCG_TMP1; 1818 lo = MIPS_BE ? TCG_TMP1 : TCG_TMP0; 1819 } else { 1820 tcg_out_bswap32(s, TCG_TMP3, MIPS_BE ? lo : hi, 0); 1821 tcg_out_opc_imm(s, sw1, TCG_TMP3, base, 0 + 0); 1822 tcg_out_opc_imm(s, sw2, TCG_TMP3, base, 0 + 3); 1823 tcg_out_bswap32(s, TCG_TMP3, MIPS_BE ? hi : lo, 0); 1824 tcg_out_opc_imm(s, sw1, TCG_TMP3, base, 4 + 0); 1825 tcg_out_opc_imm(s, sw2, TCG_TMP3, base, 4 + 3); 1826 break; 1827 } 1828 /* fall through */ 1829 case MO_64: 1830 if (TCG_TARGET_REG_BITS == 64) { 1831 tcg_out_opc_imm(s, sd1, lo, base, 0); 1832 tcg_out_opc_imm(s, sd2, lo, base, 7); 1833 } else { 1834 tcg_out_opc_imm(s, sw1, MIPS_BE ? hi : lo, base, 0 + 0); 1835 tcg_out_opc_imm(s, sw2, MIPS_BE ? hi : lo, base, 0 + 3); 1836 tcg_out_opc_imm(s, sw1, MIPS_BE ? lo : hi, base, 4 + 0); 1837 tcg_out_opc_imm(s, sw2, MIPS_BE ? lo : hi, base, 4 + 3); 1838 } 1839 break; 1840 1841 default: 1842 tcg_abort(); 1843 } 1844} 1845static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64) 1846{ 1847 TCGReg addr_regl, addr_regh __attribute__((unused)); 1848 TCGReg data_regl, data_regh; 1849 MemOpIdx oi; 1850 MemOp opc; 1851#if defined(CONFIG_SOFTMMU) 1852 tcg_insn_unit *label_ptr[2]; 1853#endif 1854 unsigned a_bits, s_bits; 1855 TCGReg base = TCG_REG_A0; 1856 1857 data_regl = *args++; 1858 data_regh = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0); 1859 addr_regl = *args++; 1860 addr_regh = (TCG_TARGET_REG_BITS < TARGET_LONG_BITS ? *args++ : 0); 1861 oi = *args++; 1862 opc = get_memop(oi); 1863 a_bits = get_alignment_bits(opc); 1864 s_bits = opc & MO_SIZE; 1865 1866 /* 1867 * R6 removes the left/right instructions but requires the 1868 * system to support misaligned memory accesses. 1869 */ 1870#if defined(CONFIG_SOFTMMU) 1871 tcg_out_tlb_load(s, base, addr_regl, addr_regh, oi, label_ptr, 0); 1872 if (use_mips32r6_instructions || a_bits >= s_bits) { 1873 tcg_out_qemu_st_direct(s, data_regl, data_regh, base, opc); 1874 } else { 1875 tcg_out_qemu_st_unalign(s, data_regl, data_regh, base, opc); 1876 } 1877 add_qemu_ldst_label(s, 0, oi, 1878 (is_64 ? TCG_TYPE_I64 : TCG_TYPE_I32), 1879 data_regl, data_regh, addr_regl, addr_regh, 1880 s->code_ptr, label_ptr); 1881#else 1882 if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) { 1883 tcg_out_ext32u(s, base, addr_regl); 1884 addr_regl = base; 1885 } 1886 if (guest_base == 0) { 1887 base = addr_regl; 1888 } else if (guest_base == (int16_t)guest_base) { 1889 tcg_out_opc_imm(s, ALIAS_PADDI, base, addr_regl, guest_base); 1890 } else { 1891 tcg_out_opc_reg(s, ALIAS_PADD, base, TCG_GUEST_BASE_REG, addr_regl); 1892 } 1893 if (use_mips32r6_instructions) { 1894 if (a_bits) { 1895 tcg_out_test_alignment(s, true, addr_regl, addr_regh, a_bits); 1896 } 1897 tcg_out_qemu_st_direct(s, data_regl, data_regh, base, opc); 1898 } else { 1899 if (a_bits && a_bits != s_bits) { 1900 tcg_out_test_alignment(s, true, addr_regl, addr_regh, a_bits); 1901 } 1902 if (a_bits >= s_bits) { 1903 tcg_out_qemu_st_direct(s, data_regl, data_regh, base, opc); 1904 } else { 1905 tcg_out_qemu_st_unalign(s, data_regl, data_regh, base, opc); 1906 } 1907 } 1908#endif 1909} 1910 1911static void tcg_out_mb(TCGContext *s, TCGArg a0) 1912{ 1913 static const MIPSInsn sync[] = { 1914 /* Note that SYNC_MB is a slightly weaker than SYNC 0, 1915 as the former is an ordering barrier and the latter 1916 is a completion barrier. */ 1917 [0 ... TCG_MO_ALL] = OPC_SYNC_MB, 1918 [TCG_MO_LD_LD] = OPC_SYNC_RMB, 1919 [TCG_MO_ST_ST] = OPC_SYNC_WMB, 1920 [TCG_MO_LD_ST] = OPC_SYNC_RELEASE, 1921 [TCG_MO_LD_ST | TCG_MO_ST_ST] = OPC_SYNC_RELEASE, 1922 [TCG_MO_LD_ST | TCG_MO_LD_LD] = OPC_SYNC_ACQUIRE, 1923 }; 1924 tcg_out32(s, sync[a0 & TCG_MO_ALL]); 1925} 1926 1927static void tcg_out_clz(TCGContext *s, MIPSInsn opcv2, MIPSInsn opcv6, 1928 int width, TCGReg a0, TCGReg a1, TCGArg a2) 1929{ 1930 if (use_mips32r6_instructions) { 1931 if (a2 == width) { 1932 tcg_out_opc_reg(s, opcv6, a0, a1, 0); 1933 } else { 1934 tcg_out_opc_reg(s, opcv6, TCG_TMP0, a1, 0); 1935 tcg_out_movcond(s, TCG_COND_EQ, a0, a1, 0, a2, TCG_TMP0); 1936 } 1937 } else { 1938 if (a2 == width) { 1939 tcg_out_opc_reg(s, opcv2, a0, a1, a1); 1940 } else if (a0 == a2) { 1941 tcg_out_opc_reg(s, opcv2, TCG_TMP0, a1, a1); 1942 tcg_out_opc_reg(s, OPC_MOVN, a0, TCG_TMP0, a1); 1943 } else if (a0 != a1) { 1944 tcg_out_opc_reg(s, opcv2, a0, a1, a1); 1945 tcg_out_opc_reg(s, OPC_MOVZ, a0, a2, a1); 1946 } else { 1947 tcg_out_opc_reg(s, opcv2, TCG_TMP0, a1, a1); 1948 tcg_out_opc_reg(s, OPC_MOVZ, TCG_TMP0, a2, a1); 1949 tcg_out_mov(s, TCG_TYPE_REG, a0, TCG_TMP0); 1950 } 1951 } 1952} 1953 1954static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0) 1955{ 1956 TCGReg b0 = TCG_REG_ZERO; 1957 1958 if (a0 & ~0xffff) { 1959 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_V0, a0 & ~0xffff); 1960 b0 = TCG_REG_V0; 1961 } 1962 if (!tcg_out_opc_jmp(s, OPC_J, tb_ret_addr)) { 1963 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)tb_ret_addr); 1964 tcg_out_opc_reg(s, OPC_JR, 0, TCG_TMP0, 0); 1965 } 1966 tcg_out_opc_imm(s, OPC_ORI, TCG_REG_V0, b0, a0 & 0xffff); 1967} 1968 1969static void tcg_out_goto_tb(TCGContext *s, int which) 1970{ 1971 /* indirect jump method */ 1972 tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP0, TCG_REG_ZERO, 1973 get_jmp_target_addr(s, which)); 1974 tcg_out_opc_reg(s, OPC_JR, 0, TCG_TMP0, 0); 1975 tcg_out_nop(s); 1976 set_jmp_reset_offset(s, which); 1977} 1978 1979void tb_target_set_jmp_target(const TranslationBlock *tb, int n, 1980 uintptr_t jmp_rx, uintptr_t jmp_rw) 1981{ 1982 /* Always indirect, nothing to do */ 1983} 1984 1985static void tcg_out_op(TCGContext *s, TCGOpcode opc, 1986 const TCGArg args[TCG_MAX_OP_ARGS], 1987 const int const_args[TCG_MAX_OP_ARGS]) 1988{ 1989 MIPSInsn i1, i2; 1990 TCGArg a0, a1, a2; 1991 int c2; 1992 1993 /* 1994 * Note that many operands use the constraint set "rZ". 1995 * We make use of the fact that 0 is the ZERO register, 1996 * and hence such cases need not check for const_args. 1997 */ 1998 a0 = args[0]; 1999 a1 = args[1]; 2000 a2 = args[2]; 2001 c2 = const_args[2]; 2002 2003 switch (opc) { 2004 case INDEX_op_goto_ptr: 2005 /* jmp to the given host address (could be epilogue) */ 2006 tcg_out_opc_reg(s, OPC_JR, 0, a0, 0); 2007 tcg_out_nop(s); 2008 break; 2009 case INDEX_op_br: 2010 tcg_out_brcond(s, TCG_COND_EQ, TCG_REG_ZERO, TCG_REG_ZERO, 2011 arg_label(a0)); 2012 break; 2013 2014 case INDEX_op_ld8u_i32: 2015 case INDEX_op_ld8u_i64: 2016 i1 = OPC_LBU; 2017 goto do_ldst; 2018 case INDEX_op_ld8s_i32: 2019 case INDEX_op_ld8s_i64: 2020 i1 = OPC_LB; 2021 goto do_ldst; 2022 case INDEX_op_ld16u_i32: 2023 case INDEX_op_ld16u_i64: 2024 i1 = OPC_LHU; 2025 goto do_ldst; 2026 case INDEX_op_ld16s_i32: 2027 case INDEX_op_ld16s_i64: 2028 i1 = OPC_LH; 2029 goto do_ldst; 2030 case INDEX_op_ld_i32: 2031 case INDEX_op_ld32s_i64: 2032 i1 = OPC_LW; 2033 goto do_ldst; 2034 case INDEX_op_ld32u_i64: 2035 i1 = OPC_LWU; 2036 goto do_ldst; 2037 case INDEX_op_ld_i64: 2038 i1 = OPC_LD; 2039 goto do_ldst; 2040 case INDEX_op_st8_i32: 2041 case INDEX_op_st8_i64: 2042 i1 = OPC_SB; 2043 goto do_ldst; 2044 case INDEX_op_st16_i32: 2045 case INDEX_op_st16_i64: 2046 i1 = OPC_SH; 2047 goto do_ldst; 2048 case INDEX_op_st_i32: 2049 case INDEX_op_st32_i64: 2050 i1 = OPC_SW; 2051 goto do_ldst; 2052 case INDEX_op_st_i64: 2053 i1 = OPC_SD; 2054 do_ldst: 2055 tcg_out_ldst(s, i1, a0, a1, a2); 2056 break; 2057 2058 case INDEX_op_add_i32: 2059 i1 = OPC_ADDU, i2 = OPC_ADDIU; 2060 goto do_binary; 2061 case INDEX_op_add_i64: 2062 i1 = OPC_DADDU, i2 = OPC_DADDIU; 2063 goto do_binary; 2064 case INDEX_op_or_i32: 2065 case INDEX_op_or_i64: 2066 i1 = OPC_OR, i2 = OPC_ORI; 2067 goto do_binary; 2068 case INDEX_op_xor_i32: 2069 case INDEX_op_xor_i64: 2070 i1 = OPC_XOR, i2 = OPC_XORI; 2071 do_binary: 2072 if (c2) { 2073 tcg_out_opc_imm(s, i2, a0, a1, a2); 2074 break; 2075 } 2076 do_binaryv: 2077 tcg_out_opc_reg(s, i1, a0, a1, a2); 2078 break; 2079 2080 case INDEX_op_sub_i32: 2081 i1 = OPC_SUBU, i2 = OPC_ADDIU; 2082 goto do_subtract; 2083 case INDEX_op_sub_i64: 2084 i1 = OPC_DSUBU, i2 = OPC_DADDIU; 2085 do_subtract: 2086 if (c2) { 2087 tcg_out_opc_imm(s, i2, a0, a1, -a2); 2088 break; 2089 } 2090 goto do_binaryv; 2091 case INDEX_op_and_i32: 2092 if (c2 && a2 != (uint16_t)a2) { 2093 int msb = ctz32(~a2) - 1; 2094 tcg_debug_assert(use_mips32r2_instructions); 2095 tcg_debug_assert(is_p2m1(a2)); 2096 tcg_out_opc_bf(s, OPC_EXT, a0, a1, msb, 0); 2097 break; 2098 } 2099 i1 = OPC_AND, i2 = OPC_ANDI; 2100 goto do_binary; 2101 case INDEX_op_and_i64: 2102 if (c2 && a2 != (uint16_t)a2) { 2103 int msb = ctz64(~a2) - 1; 2104 tcg_debug_assert(use_mips32r2_instructions); 2105 tcg_debug_assert(is_p2m1(a2)); 2106 tcg_out_opc_bf64(s, OPC_DEXT, OPC_DEXTM, OPC_DEXTU, a0, a1, msb, 0); 2107 break; 2108 } 2109 i1 = OPC_AND, i2 = OPC_ANDI; 2110 goto do_binary; 2111 case INDEX_op_nor_i32: 2112 case INDEX_op_nor_i64: 2113 i1 = OPC_NOR; 2114 goto do_binaryv; 2115 2116 case INDEX_op_mul_i32: 2117 if (use_mips32_instructions) { 2118 tcg_out_opc_reg(s, OPC_MUL, a0, a1, a2); 2119 break; 2120 } 2121 i1 = OPC_MULT, i2 = OPC_MFLO; 2122 goto do_hilo1; 2123 case INDEX_op_mulsh_i32: 2124 if (use_mips32r6_instructions) { 2125 tcg_out_opc_reg(s, OPC_MUH, a0, a1, a2); 2126 break; 2127 } 2128 i1 = OPC_MULT, i2 = OPC_MFHI; 2129 goto do_hilo1; 2130 case INDEX_op_muluh_i32: 2131 if (use_mips32r6_instructions) { 2132 tcg_out_opc_reg(s, OPC_MUHU, a0, a1, a2); 2133 break; 2134 } 2135 i1 = OPC_MULTU, i2 = OPC_MFHI; 2136 goto do_hilo1; 2137 case INDEX_op_div_i32: 2138 if (use_mips32r6_instructions) { 2139 tcg_out_opc_reg(s, OPC_DIV_R6, a0, a1, a2); 2140 break; 2141 } 2142 i1 = OPC_DIV, i2 = OPC_MFLO; 2143 goto do_hilo1; 2144 case INDEX_op_divu_i32: 2145 if (use_mips32r6_instructions) { 2146 tcg_out_opc_reg(s, OPC_DIVU_R6, a0, a1, a2); 2147 break; 2148 } 2149 i1 = OPC_DIVU, i2 = OPC_MFLO; 2150 goto do_hilo1; 2151 case INDEX_op_rem_i32: 2152 if (use_mips32r6_instructions) { 2153 tcg_out_opc_reg(s, OPC_MOD, a0, a1, a2); 2154 break; 2155 } 2156 i1 = OPC_DIV, i2 = OPC_MFHI; 2157 goto do_hilo1; 2158 case INDEX_op_remu_i32: 2159 if (use_mips32r6_instructions) { 2160 tcg_out_opc_reg(s, OPC_MODU, a0, a1, a2); 2161 break; 2162 } 2163 i1 = OPC_DIVU, i2 = OPC_MFHI; 2164 goto do_hilo1; 2165 case INDEX_op_mul_i64: 2166 if (use_mips32r6_instructions) { 2167 tcg_out_opc_reg(s, OPC_DMUL, a0, a1, a2); 2168 break; 2169 } 2170 i1 = OPC_DMULT, i2 = OPC_MFLO; 2171 goto do_hilo1; 2172 case INDEX_op_mulsh_i64: 2173 if (use_mips32r6_instructions) { 2174 tcg_out_opc_reg(s, OPC_DMUH, a0, a1, a2); 2175 break; 2176 } 2177 i1 = OPC_DMULT, i2 = OPC_MFHI; 2178 goto do_hilo1; 2179 case INDEX_op_muluh_i64: 2180 if (use_mips32r6_instructions) { 2181 tcg_out_opc_reg(s, OPC_DMUHU, a0, a1, a2); 2182 break; 2183 } 2184 i1 = OPC_DMULTU, i2 = OPC_MFHI; 2185 goto do_hilo1; 2186 case INDEX_op_div_i64: 2187 if (use_mips32r6_instructions) { 2188 tcg_out_opc_reg(s, OPC_DDIV_R6, a0, a1, a2); 2189 break; 2190 } 2191 i1 = OPC_DDIV, i2 = OPC_MFLO; 2192 goto do_hilo1; 2193 case INDEX_op_divu_i64: 2194 if (use_mips32r6_instructions) { 2195 tcg_out_opc_reg(s, OPC_DDIVU_R6, a0, a1, a2); 2196 break; 2197 } 2198 i1 = OPC_DDIVU, i2 = OPC_MFLO; 2199 goto do_hilo1; 2200 case INDEX_op_rem_i64: 2201 if (use_mips32r6_instructions) { 2202 tcg_out_opc_reg(s, OPC_DMOD, a0, a1, a2); 2203 break; 2204 } 2205 i1 = OPC_DDIV, i2 = OPC_MFHI; 2206 goto do_hilo1; 2207 case INDEX_op_remu_i64: 2208 if (use_mips32r6_instructions) { 2209 tcg_out_opc_reg(s, OPC_DMODU, a0, a1, a2); 2210 break; 2211 } 2212 i1 = OPC_DDIVU, i2 = OPC_MFHI; 2213 do_hilo1: 2214 tcg_out_opc_reg(s, i1, 0, a1, a2); 2215 tcg_out_opc_reg(s, i2, a0, 0, 0); 2216 break; 2217 2218 case INDEX_op_muls2_i32: 2219 i1 = OPC_MULT; 2220 goto do_hilo2; 2221 case INDEX_op_mulu2_i32: 2222 i1 = OPC_MULTU; 2223 goto do_hilo2; 2224 case INDEX_op_muls2_i64: 2225 i1 = OPC_DMULT; 2226 goto do_hilo2; 2227 case INDEX_op_mulu2_i64: 2228 i1 = OPC_DMULTU; 2229 do_hilo2: 2230 tcg_out_opc_reg(s, i1, 0, a2, args[3]); 2231 tcg_out_opc_reg(s, OPC_MFLO, a0, 0, 0); 2232 tcg_out_opc_reg(s, OPC_MFHI, a1, 0, 0); 2233 break; 2234 2235 case INDEX_op_not_i32: 2236 case INDEX_op_not_i64: 2237 i1 = OPC_NOR; 2238 goto do_unary; 2239 case INDEX_op_ext8s_i32: 2240 case INDEX_op_ext8s_i64: 2241 i1 = OPC_SEB; 2242 goto do_unary; 2243 case INDEX_op_ext16s_i32: 2244 case INDEX_op_ext16s_i64: 2245 i1 = OPC_SEH; 2246 do_unary: 2247 tcg_out_opc_reg(s, i1, a0, TCG_REG_ZERO, a1); 2248 break; 2249 2250 case INDEX_op_bswap16_i32: 2251 case INDEX_op_bswap16_i64: 2252 tcg_out_bswap16(s, a0, a1, a2); 2253 break; 2254 case INDEX_op_bswap32_i32: 2255 tcg_out_bswap32(s, a0, a1, 0); 2256 break; 2257 case INDEX_op_bswap32_i64: 2258 tcg_out_bswap32(s, a0, a1, a2); 2259 break; 2260 case INDEX_op_bswap64_i64: 2261 tcg_out_bswap64(s, a0, a1); 2262 break; 2263 case INDEX_op_extrh_i64_i32: 2264 tcg_out_dsra(s, a0, a1, 32); 2265 break; 2266 case INDEX_op_ext32s_i64: 2267 case INDEX_op_ext_i32_i64: 2268 case INDEX_op_extrl_i64_i32: 2269 tcg_out_opc_sa(s, OPC_SLL, a0, a1, 0); 2270 break; 2271 case INDEX_op_ext32u_i64: 2272 case INDEX_op_extu_i32_i64: 2273 tcg_out_ext32u(s, a0, a1); 2274 break; 2275 2276 case INDEX_op_sar_i32: 2277 i1 = OPC_SRAV, i2 = OPC_SRA; 2278 goto do_shift; 2279 case INDEX_op_shl_i32: 2280 i1 = OPC_SLLV, i2 = OPC_SLL; 2281 goto do_shift; 2282 case INDEX_op_shr_i32: 2283 i1 = OPC_SRLV, i2 = OPC_SRL; 2284 goto do_shift; 2285 case INDEX_op_rotr_i32: 2286 i1 = OPC_ROTRV, i2 = OPC_ROTR; 2287 do_shift: 2288 if (c2) { 2289 tcg_out_opc_sa(s, i2, a0, a1, a2); 2290 break; 2291 } 2292 do_shiftv: 2293 tcg_out_opc_reg(s, i1, a0, a2, a1); 2294 break; 2295 case INDEX_op_rotl_i32: 2296 if (c2) { 2297 tcg_out_opc_sa(s, OPC_ROTR, a0, a1, 32 - a2); 2298 } else { 2299 tcg_out_opc_reg(s, OPC_SUBU, TCG_TMP0, TCG_REG_ZERO, a2); 2300 tcg_out_opc_reg(s, OPC_ROTRV, a0, TCG_TMP0, a1); 2301 } 2302 break; 2303 case INDEX_op_sar_i64: 2304 if (c2) { 2305 tcg_out_dsra(s, a0, a1, a2); 2306 break; 2307 } 2308 i1 = OPC_DSRAV; 2309 goto do_shiftv; 2310 case INDEX_op_shl_i64: 2311 if (c2) { 2312 tcg_out_dsll(s, a0, a1, a2); 2313 break; 2314 } 2315 i1 = OPC_DSLLV; 2316 goto do_shiftv; 2317 case INDEX_op_shr_i64: 2318 if (c2) { 2319 tcg_out_dsrl(s, a0, a1, a2); 2320 break; 2321 } 2322 i1 = OPC_DSRLV; 2323 goto do_shiftv; 2324 case INDEX_op_rotr_i64: 2325 if (c2) { 2326 tcg_out_opc_sa64(s, OPC_DROTR, OPC_DROTR32, a0, a1, a2); 2327 break; 2328 } 2329 i1 = OPC_DROTRV; 2330 goto do_shiftv; 2331 case INDEX_op_rotl_i64: 2332 if (c2) { 2333 tcg_out_opc_sa64(s, OPC_DROTR, OPC_DROTR32, a0, a1, 64 - a2); 2334 } else { 2335 tcg_out_opc_reg(s, OPC_DSUBU, TCG_TMP0, TCG_REG_ZERO, a2); 2336 tcg_out_opc_reg(s, OPC_DROTRV, a0, TCG_TMP0, a1); 2337 } 2338 break; 2339 2340 case INDEX_op_clz_i32: 2341 tcg_out_clz(s, OPC_CLZ, OPC_CLZ_R6, 32, a0, a1, a2); 2342 break; 2343 case INDEX_op_clz_i64: 2344 tcg_out_clz(s, OPC_DCLZ, OPC_DCLZ_R6, 64, a0, a1, a2); 2345 break; 2346 2347 case INDEX_op_deposit_i32: 2348 tcg_out_opc_bf(s, OPC_INS, a0, a2, args[3] + args[4] - 1, args[3]); 2349 break; 2350 case INDEX_op_deposit_i64: 2351 tcg_out_opc_bf64(s, OPC_DINS, OPC_DINSM, OPC_DINSU, a0, a2, 2352 args[3] + args[4] - 1, args[3]); 2353 break; 2354 case INDEX_op_extract_i32: 2355 tcg_out_opc_bf(s, OPC_EXT, a0, a1, args[3] - 1, a2); 2356 break; 2357 case INDEX_op_extract_i64: 2358 tcg_out_opc_bf64(s, OPC_DEXT, OPC_DEXTM, OPC_DEXTU, a0, a1, 2359 args[3] - 1, a2); 2360 break; 2361 2362 case INDEX_op_brcond_i32: 2363 case INDEX_op_brcond_i64: 2364 tcg_out_brcond(s, a2, a0, a1, arg_label(args[3])); 2365 break; 2366 case INDEX_op_brcond2_i32: 2367 tcg_out_brcond2(s, args[4], a0, a1, a2, args[3], arg_label(args[5])); 2368 break; 2369 2370 case INDEX_op_movcond_i32: 2371 case INDEX_op_movcond_i64: 2372 tcg_out_movcond(s, args[5], a0, a1, a2, args[3], args[4]); 2373 break; 2374 2375 case INDEX_op_setcond_i32: 2376 case INDEX_op_setcond_i64: 2377 tcg_out_setcond(s, args[3], a0, a1, a2); 2378 break; 2379 case INDEX_op_setcond2_i32: 2380 tcg_out_setcond2(s, args[5], a0, a1, a2, args[3], args[4]); 2381 break; 2382 2383 case INDEX_op_qemu_ld_i32: 2384 tcg_out_qemu_ld(s, args, false); 2385 break; 2386 case INDEX_op_qemu_ld_i64: 2387 tcg_out_qemu_ld(s, args, true); 2388 break; 2389 case INDEX_op_qemu_st_i32: 2390 tcg_out_qemu_st(s, args, false); 2391 break; 2392 case INDEX_op_qemu_st_i64: 2393 tcg_out_qemu_st(s, args, true); 2394 break; 2395 2396 case INDEX_op_add2_i32: 2397 tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5], 2398 const_args[4], const_args[5], false); 2399 break; 2400 case INDEX_op_sub2_i32: 2401 tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5], 2402 const_args[4], const_args[5], true); 2403 break; 2404 2405 case INDEX_op_mb: 2406 tcg_out_mb(s, a0); 2407 break; 2408 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */ 2409 case INDEX_op_mov_i64: 2410 case INDEX_op_call: /* Always emitted via tcg_out_call. */ 2411 case INDEX_op_exit_tb: /* Always emitted via tcg_out_exit_tb. */ 2412 case INDEX_op_goto_tb: /* Always emitted via tcg_out_goto_tb. */ 2413 default: 2414 tcg_abort(); 2415 } 2416} 2417 2418static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op) 2419{ 2420 switch (op) { 2421 case INDEX_op_goto_ptr: 2422 return C_O0_I1(r); 2423 2424 case INDEX_op_ld8u_i32: 2425 case INDEX_op_ld8s_i32: 2426 case INDEX_op_ld16u_i32: 2427 case INDEX_op_ld16s_i32: 2428 case INDEX_op_ld_i32: 2429 case INDEX_op_not_i32: 2430 case INDEX_op_bswap16_i32: 2431 case INDEX_op_bswap32_i32: 2432 case INDEX_op_ext8s_i32: 2433 case INDEX_op_ext16s_i32: 2434 case INDEX_op_extract_i32: 2435 case INDEX_op_ld8u_i64: 2436 case INDEX_op_ld8s_i64: 2437 case INDEX_op_ld16u_i64: 2438 case INDEX_op_ld16s_i64: 2439 case INDEX_op_ld32s_i64: 2440 case INDEX_op_ld32u_i64: 2441 case INDEX_op_ld_i64: 2442 case INDEX_op_not_i64: 2443 case INDEX_op_bswap16_i64: 2444 case INDEX_op_bswap32_i64: 2445 case INDEX_op_bswap64_i64: 2446 case INDEX_op_ext8s_i64: 2447 case INDEX_op_ext16s_i64: 2448 case INDEX_op_ext32s_i64: 2449 case INDEX_op_ext32u_i64: 2450 case INDEX_op_ext_i32_i64: 2451 case INDEX_op_extu_i32_i64: 2452 case INDEX_op_extrl_i64_i32: 2453 case INDEX_op_extrh_i64_i32: 2454 case INDEX_op_extract_i64: 2455 return C_O1_I1(r, r); 2456 2457 case INDEX_op_st8_i32: 2458 case INDEX_op_st16_i32: 2459 case INDEX_op_st_i32: 2460 case INDEX_op_st8_i64: 2461 case INDEX_op_st16_i64: 2462 case INDEX_op_st32_i64: 2463 case INDEX_op_st_i64: 2464 return C_O0_I2(rZ, r); 2465 2466 case INDEX_op_add_i32: 2467 case INDEX_op_add_i64: 2468 return C_O1_I2(r, r, rJ); 2469 case INDEX_op_sub_i32: 2470 case INDEX_op_sub_i64: 2471 return C_O1_I2(r, rZ, rN); 2472 case INDEX_op_mul_i32: 2473 case INDEX_op_mulsh_i32: 2474 case INDEX_op_muluh_i32: 2475 case INDEX_op_div_i32: 2476 case INDEX_op_divu_i32: 2477 case INDEX_op_rem_i32: 2478 case INDEX_op_remu_i32: 2479 case INDEX_op_nor_i32: 2480 case INDEX_op_setcond_i32: 2481 case INDEX_op_mul_i64: 2482 case INDEX_op_mulsh_i64: 2483 case INDEX_op_muluh_i64: 2484 case INDEX_op_div_i64: 2485 case INDEX_op_divu_i64: 2486 case INDEX_op_rem_i64: 2487 case INDEX_op_remu_i64: 2488 case INDEX_op_nor_i64: 2489 case INDEX_op_setcond_i64: 2490 return C_O1_I2(r, rZ, rZ); 2491 case INDEX_op_muls2_i32: 2492 case INDEX_op_mulu2_i32: 2493 case INDEX_op_muls2_i64: 2494 case INDEX_op_mulu2_i64: 2495 return C_O2_I2(r, r, r, r); 2496 case INDEX_op_and_i32: 2497 case INDEX_op_and_i64: 2498 return C_O1_I2(r, r, rIK); 2499 case INDEX_op_or_i32: 2500 case INDEX_op_xor_i32: 2501 case INDEX_op_or_i64: 2502 case INDEX_op_xor_i64: 2503 return C_O1_I2(r, r, rI); 2504 case INDEX_op_shl_i32: 2505 case INDEX_op_shr_i32: 2506 case INDEX_op_sar_i32: 2507 case INDEX_op_rotr_i32: 2508 case INDEX_op_rotl_i32: 2509 case INDEX_op_shl_i64: 2510 case INDEX_op_shr_i64: 2511 case INDEX_op_sar_i64: 2512 case INDEX_op_rotr_i64: 2513 case INDEX_op_rotl_i64: 2514 return C_O1_I2(r, r, ri); 2515 case INDEX_op_clz_i32: 2516 case INDEX_op_clz_i64: 2517 return C_O1_I2(r, r, rWZ); 2518 2519 case INDEX_op_deposit_i32: 2520 case INDEX_op_deposit_i64: 2521 return C_O1_I2(r, 0, rZ); 2522 case INDEX_op_brcond_i32: 2523 case INDEX_op_brcond_i64: 2524 return C_O0_I2(rZ, rZ); 2525 case INDEX_op_movcond_i32: 2526 case INDEX_op_movcond_i64: 2527 return (use_mips32r6_instructions 2528 ? C_O1_I4(r, rZ, rZ, rZ, rZ) 2529 : C_O1_I4(r, rZ, rZ, rZ, 0)); 2530 case INDEX_op_add2_i32: 2531 case INDEX_op_sub2_i32: 2532 return C_O2_I4(r, r, rZ, rZ, rN, rN); 2533 case INDEX_op_setcond2_i32: 2534 return C_O1_I4(r, rZ, rZ, rZ, rZ); 2535 case INDEX_op_brcond2_i32: 2536 return C_O0_I4(rZ, rZ, rZ, rZ); 2537 2538 case INDEX_op_qemu_ld_i32: 2539 return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32 2540 ? C_O1_I1(r, L) : C_O1_I2(r, L, L)); 2541 case INDEX_op_qemu_st_i32: 2542 return (TCG_TARGET_REG_BITS == 64 || TARGET_LONG_BITS == 32 2543 ? C_O0_I2(SZ, S) : C_O0_I3(SZ, S, S)); 2544 case INDEX_op_qemu_ld_i64: 2545 return (TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, L) 2546 : TARGET_LONG_BITS == 32 ? C_O2_I1(r, r, L) 2547 : C_O2_I2(r, r, L, L)); 2548 case INDEX_op_qemu_st_i64: 2549 return (TCG_TARGET_REG_BITS == 64 ? C_O0_I2(SZ, S) 2550 : TARGET_LONG_BITS == 32 ? C_O0_I3(SZ, SZ, S) 2551 : C_O0_I4(SZ, SZ, S, S)); 2552 2553 default: 2554 g_assert_not_reached(); 2555 } 2556} 2557 2558static const int tcg_target_callee_save_regs[] = { 2559 TCG_REG_S0, /* used for the global env (TCG_AREG0) */ 2560 TCG_REG_S1, 2561 TCG_REG_S2, 2562 TCG_REG_S3, 2563 TCG_REG_S4, 2564 TCG_REG_S5, 2565 TCG_REG_S6, 2566 TCG_REG_S7, 2567 TCG_REG_S8, 2568 TCG_REG_RA, /* should be last for ABI compliance */ 2569}; 2570 2571/* The Linux kernel doesn't provide any information about the available 2572 instruction set. Probe it using a signal handler. */ 2573 2574 2575#ifndef use_movnz_instructions 2576bool use_movnz_instructions = false; 2577#endif 2578 2579#ifndef use_mips32_instructions 2580bool use_mips32_instructions = false; 2581#endif 2582 2583#ifndef use_mips32r2_instructions 2584bool use_mips32r2_instructions = false; 2585#endif 2586 2587static volatile sig_atomic_t got_sigill; 2588 2589static void sigill_handler(int signo, siginfo_t *si, void *data) 2590{ 2591 /* Skip the faulty instruction */ 2592 ucontext_t *uc = (ucontext_t *)data; 2593 uc->uc_mcontext.pc += 4; 2594 2595 got_sigill = 1; 2596} 2597 2598static void tcg_target_detect_isa(void) 2599{ 2600 struct sigaction sa_old, sa_new; 2601 2602 memset(&sa_new, 0, sizeof(sa_new)); 2603 sa_new.sa_flags = SA_SIGINFO; 2604 sa_new.sa_sigaction = sigill_handler; 2605 sigaction(SIGILL, &sa_new, &sa_old); 2606 2607 /* Probe for movn/movz, necessary to implement movcond. */ 2608#ifndef use_movnz_instructions 2609 got_sigill = 0; 2610 asm volatile(".set push\n" 2611 ".set mips32\n" 2612 "movn $zero, $zero, $zero\n" 2613 "movz $zero, $zero, $zero\n" 2614 ".set pop\n" 2615 : : : ); 2616 use_movnz_instructions = !got_sigill; 2617#endif 2618 2619 /* Probe for MIPS32 instructions. As no subsetting is allowed 2620 by the specification, it is only necessary to probe for one 2621 of the instructions. */ 2622#ifndef use_mips32_instructions 2623 got_sigill = 0; 2624 asm volatile(".set push\n" 2625 ".set mips32\n" 2626 "mul $zero, $zero\n" 2627 ".set pop\n" 2628 : : : ); 2629 use_mips32_instructions = !got_sigill; 2630#endif 2631 2632 /* Probe for MIPS32r2 instructions if MIPS32 instructions are 2633 available. As no subsetting is allowed by the specification, 2634 it is only necessary to probe for one of the instructions. */ 2635#ifndef use_mips32r2_instructions 2636 if (use_mips32_instructions) { 2637 got_sigill = 0; 2638 asm volatile(".set push\n" 2639 ".set mips32r2\n" 2640 "seb $zero, $zero\n" 2641 ".set pop\n" 2642 : : : ); 2643 use_mips32r2_instructions = !got_sigill; 2644 } 2645#endif 2646 2647 sigaction(SIGILL, &sa_old, NULL); 2648} 2649 2650static tcg_insn_unit *align_code_ptr(TCGContext *s) 2651{ 2652 uintptr_t p = (uintptr_t)s->code_ptr; 2653 if (p & 15) { 2654 p = (p + 15) & -16; 2655 s->code_ptr = (void *)p; 2656 } 2657 return s->code_ptr; 2658} 2659 2660/* Stack frame parameters. */ 2661#define REG_SIZE (TCG_TARGET_REG_BITS / 8) 2662#define SAVE_SIZE ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * REG_SIZE) 2663#define TEMP_SIZE (CPU_TEMP_BUF_NLONGS * (int)sizeof(long)) 2664 2665#define FRAME_SIZE ((TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE + SAVE_SIZE \ 2666 + TCG_TARGET_STACK_ALIGN - 1) \ 2667 & -TCG_TARGET_STACK_ALIGN) 2668#define SAVE_OFS (TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE) 2669 2670/* We're expecting to be able to use an immediate for frame allocation. */ 2671QEMU_BUILD_BUG_ON(FRAME_SIZE > 0x7fff); 2672 2673/* Generate global QEMU prologue and epilogue code */ 2674static void tcg_target_qemu_prologue(TCGContext *s) 2675{ 2676 int i; 2677 2678 tcg_set_frame(s, TCG_REG_SP, TCG_STATIC_CALL_ARGS_SIZE, TEMP_SIZE); 2679 2680 /* TB prologue */ 2681 tcg_out_opc_imm(s, ALIAS_PADDI, TCG_REG_SP, TCG_REG_SP, -FRAME_SIZE); 2682 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) { 2683 tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i], 2684 TCG_REG_SP, SAVE_OFS + i * REG_SIZE); 2685 } 2686 2687#ifndef CONFIG_SOFTMMU 2688 if (guest_base) { 2689 tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base); 2690 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG); 2691 } 2692#endif 2693 2694 /* Call generated code */ 2695 tcg_out_opc_reg(s, OPC_JR, 0, tcg_target_call_iarg_regs[1], 0); 2696 /* delay slot */ 2697 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]); 2698 2699 /* 2700 * Return path for goto_ptr. Set return value to 0, a-la exit_tb, 2701 * and fall through to the rest of the epilogue. 2702 */ 2703 tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr); 2704 tcg_out_mov(s, TCG_TYPE_REG, TCG_REG_V0, TCG_REG_ZERO); 2705 2706 /* TB epilogue */ 2707 tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr); 2708 for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) { 2709 tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i], 2710 TCG_REG_SP, SAVE_OFS + i * REG_SIZE); 2711 } 2712 2713 tcg_out_opc_reg(s, OPC_JR, 0, TCG_REG_RA, 0); 2714 /* delay slot */ 2715 tcg_out_opc_imm(s, ALIAS_PADDI, TCG_REG_SP, TCG_REG_SP, FRAME_SIZE); 2716 2717 if (use_mips32r2_instructions) { 2718 return; 2719 } 2720 2721 /* Bswap subroutines: Input in TCG_TMP0, output in TCG_TMP3; 2722 clobbers TCG_TMP1, TCG_TMP2. */ 2723 2724 /* 2725 * bswap32 -- 32-bit swap (signed result for mips64). a0 = abcd. 2726 */ 2727 bswap32_addr = tcg_splitwx_to_rx(align_code_ptr(s)); 2728 /* t3 = (ssss)d000 */ 2729 tcg_out_opc_sa(s, OPC_SLL, TCG_TMP3, TCG_TMP0, 24); 2730 /* t1 = 000a */ 2731 tcg_out_opc_sa(s, OPC_SRL, TCG_TMP1, TCG_TMP0, 24); 2732 /* t2 = 00c0 */ 2733 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP2, TCG_TMP0, 0xff00); 2734 /* t3 = d00a */ 2735 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); 2736 /* t1 = 0abc */ 2737 tcg_out_opc_sa(s, OPC_SRL, TCG_TMP1, TCG_TMP0, 8); 2738 /* t2 = 0c00 */ 2739 tcg_out_opc_sa(s, OPC_SLL, TCG_TMP2, TCG_TMP2, 8); 2740 /* t1 = 00b0 */ 2741 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, TCG_TMP1, 0xff00); 2742 /* t3 = dc0a */ 2743 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP2); 2744 tcg_out_opc_reg(s, OPC_JR, 0, TCG_REG_RA, 0); 2745 /* t3 = dcba -- delay slot */ 2746 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); 2747 2748 if (TCG_TARGET_REG_BITS == 32) { 2749 return; 2750 } 2751 2752 /* 2753 * bswap32u -- unsigned 32-bit swap. a0 = ....abcd. 2754 */ 2755 bswap32u_addr = tcg_splitwx_to_rx(align_code_ptr(s)); 2756 /* t1 = (0000)000d */ 2757 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, TCG_TMP0, 0xff); 2758 /* t3 = 000a */ 2759 tcg_out_opc_sa(s, OPC_SRL, TCG_TMP3, TCG_TMP0, 24); 2760 /* t1 = (0000)d000 */ 2761 tcg_out_dsll(s, TCG_TMP1, TCG_TMP1, 24); 2762 /* t2 = 00c0 */ 2763 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP2, TCG_TMP0, 0xff00); 2764 /* t3 = d00a */ 2765 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); 2766 /* t1 = 0abc */ 2767 tcg_out_opc_sa(s, OPC_SRL, TCG_TMP1, TCG_TMP0, 8); 2768 /* t2 = 0c00 */ 2769 tcg_out_opc_sa(s, OPC_SLL, TCG_TMP2, TCG_TMP2, 8); 2770 /* t1 = 00b0 */ 2771 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, TCG_TMP1, 0xff00); 2772 /* t3 = dc0a */ 2773 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP2); 2774 tcg_out_opc_reg(s, OPC_JR, 0, TCG_REG_RA, 0); 2775 /* t3 = dcba -- delay slot */ 2776 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); 2777 2778 /* 2779 * bswap64 -- 64-bit swap. a0 = abcdefgh 2780 */ 2781 bswap64_addr = tcg_splitwx_to_rx(align_code_ptr(s)); 2782 /* t3 = h0000000 */ 2783 tcg_out_dsll(s, TCG_TMP3, TCG_TMP0, 56); 2784 /* t1 = 0000000a */ 2785 tcg_out_dsrl(s, TCG_TMP1, TCG_TMP0, 56); 2786 2787 /* t2 = 000000g0 */ 2788 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP2, TCG_TMP0, 0xff00); 2789 /* t3 = h000000a */ 2790 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); 2791 /* t1 = 00000abc */ 2792 tcg_out_dsrl(s, TCG_TMP1, TCG_TMP0, 40); 2793 /* t2 = 0g000000 */ 2794 tcg_out_dsll(s, TCG_TMP2, TCG_TMP2, 40); 2795 /* t1 = 000000b0 */ 2796 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, TCG_TMP1, 0xff00); 2797 2798 /* t3 = hg00000a */ 2799 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP2); 2800 /* t2 = 0000abcd */ 2801 tcg_out_dsrl(s, TCG_TMP2, TCG_TMP0, 32); 2802 /* t3 = hg0000ba */ 2803 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); 2804 2805 /* t1 = 000000c0 */ 2806 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, TCG_TMP2, 0xff00); 2807 /* t2 = 0000000d */ 2808 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP2, TCG_TMP2, 0x00ff); 2809 /* t1 = 00000c00 */ 2810 tcg_out_dsll(s, TCG_TMP1, TCG_TMP1, 8); 2811 /* t2 = 0000d000 */ 2812 tcg_out_dsll(s, TCG_TMP2, TCG_TMP2, 24); 2813 2814 /* t3 = hg000cba */ 2815 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); 2816 /* t1 = 00abcdef */ 2817 tcg_out_dsrl(s, TCG_TMP1, TCG_TMP0, 16); 2818 /* t3 = hg00dcba */ 2819 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP2); 2820 2821 /* t2 = 0000000f */ 2822 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP2, TCG_TMP1, 0x00ff); 2823 /* t1 = 000000e0 */ 2824 tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, TCG_TMP1, 0xff00); 2825 /* t2 = 00f00000 */ 2826 tcg_out_dsll(s, TCG_TMP2, TCG_TMP2, 40); 2827 /* t1 = 000e0000 */ 2828 tcg_out_dsll(s, TCG_TMP1, TCG_TMP1, 24); 2829 2830 /* t3 = hgf0dcba */ 2831 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP2); 2832 tcg_out_opc_reg(s, OPC_JR, 0, TCG_REG_RA, 0); 2833 /* t3 = hgfedcba -- delay slot */ 2834 tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); 2835} 2836 2837static void tcg_target_init(TCGContext *s) 2838{ 2839 tcg_target_detect_isa(); 2840 tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff; 2841 if (TCG_TARGET_REG_BITS == 64) { 2842 tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff; 2843 } 2844 2845 tcg_target_call_clobber_regs = 0; 2846 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0); 2847 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1); 2848 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_A0); 2849 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_A1); 2850 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_A2); 2851 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_A3); 2852 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_T0); 2853 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_T1); 2854 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_T2); 2855 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_T3); 2856 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_T4); 2857 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_T5); 2858 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_T6); 2859 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_T7); 2860 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_T8); 2861 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_T9); 2862 2863 s->reserved_regs = 0; 2864 tcg_regset_set_reg(s->reserved_regs, TCG_REG_ZERO); /* zero register */ 2865 tcg_regset_set_reg(s->reserved_regs, TCG_REG_K0); /* kernel use only */ 2866 tcg_regset_set_reg(s->reserved_regs, TCG_REG_K1); /* kernel use only */ 2867 tcg_regset_set_reg(s->reserved_regs, TCG_TMP0); /* internal use */ 2868 tcg_regset_set_reg(s->reserved_regs, TCG_TMP1); /* internal use */ 2869 tcg_regset_set_reg(s->reserved_regs, TCG_TMP2); /* internal use */ 2870 tcg_regset_set_reg(s->reserved_regs, TCG_TMP3); /* internal use */ 2871 tcg_regset_set_reg(s->reserved_regs, TCG_REG_RA); /* return address */ 2872 tcg_regset_set_reg(s->reserved_regs, TCG_REG_SP); /* stack pointer */ 2873 tcg_regset_set_reg(s->reserved_regs, TCG_REG_GP); /* global pointer */ 2874} 2875 2876typedef struct { 2877 DebugFrameHeader h; 2878 uint8_t fde_def_cfa[4]; 2879 uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2]; 2880} DebugFrame; 2881 2882#define ELF_HOST_MACHINE EM_MIPS 2883/* GDB doesn't appear to require proper setting of ELF_HOST_FLAGS, 2884 which is good because they're really quite complicated for MIPS. */ 2885 2886static const DebugFrame debug_frame = { 2887 .h.cie.len = sizeof(DebugFrameCIE) - 4, /* length after .len member */ 2888 .h.cie.id = -1, 2889 .h.cie.version = 1, 2890 .h.cie.code_align = 1, 2891 .h.cie.data_align = -(TCG_TARGET_REG_BITS / 8) & 0x7f, /* sleb128 */ 2892 .h.cie.return_column = TCG_REG_RA, 2893 2894 /* Total FDE size does not include the "len" member. */ 2895 .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset), 2896 2897 .fde_def_cfa = { 2898 12, TCG_REG_SP, /* DW_CFA_def_cfa sp, ... */ 2899 (FRAME_SIZE & 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */ 2900 (FRAME_SIZE >> 7) 2901 }, 2902 .fde_reg_ofs = { 2903 0x80 + 16, 9, /* DW_CFA_offset, s0, -72 */ 2904 0x80 + 17, 8, /* DW_CFA_offset, s2, -64 */ 2905 0x80 + 18, 7, /* DW_CFA_offset, s3, -56 */ 2906 0x80 + 19, 6, /* DW_CFA_offset, s4, -48 */ 2907 0x80 + 20, 5, /* DW_CFA_offset, s5, -40 */ 2908 0x80 + 21, 4, /* DW_CFA_offset, s6, -32 */ 2909 0x80 + 22, 3, /* DW_CFA_offset, s7, -24 */ 2910 0x80 + 30, 2, /* DW_CFA_offset, s8, -16 */ 2911 0x80 + 31, 1, /* DW_CFA_offset, ra, -8 */ 2912 } 2913}; 2914 2915void tcg_register_jit(const void *buf, size_t buf_size) 2916{ 2917 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame)); 2918} 2919