1/* 2 * Tiny Code Generator for QEMU 3 * 4 * Copyright (c) 2009 Ulrich Hecht <uli@suse.de> 5 * Copyright (c) 2009 Alexander Graf <agraf@suse.de> 6 * Copyright (c) 2010 Richard Henderson <rth@twiddle.net> 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 "elf.h" 28 29/* Used for function call generation. */ 30#define TCG_TARGET_STACK_ALIGN 8 31#define TCG_TARGET_CALL_STACK_OFFSET 160 32#define TCG_TARGET_CALL_ARG_I32 TCG_CALL_ARG_EXTEND 33#define TCG_TARGET_CALL_ARG_I64 TCG_CALL_ARG_NORMAL 34#define TCG_TARGET_CALL_ARG_I128 TCG_CALL_ARG_BY_REF 35#define TCG_TARGET_CALL_RET_I128 TCG_CALL_RET_BY_REF 36 37#define TCG_CT_CONST_S16 (1 << 8) 38#define TCG_CT_CONST_S32 (1 << 9) 39#define TCG_CT_CONST_U32 (1 << 10) 40#define TCG_CT_CONST_ZERO (1 << 11) 41#define TCG_CT_CONST_P32 (1 << 12) 42#define TCG_CT_CONST_INV (1 << 13) 43#define TCG_CT_CONST_INVRISBG (1 << 14) 44#define TCG_CT_CONST_CMP (1 << 15) 45#define TCG_CT_CONST_M1 (1 << 16) 46 47#define ALL_GENERAL_REGS MAKE_64BIT_MASK(0, 16) 48#define ALL_VECTOR_REGS MAKE_64BIT_MASK(32, 32) 49 50/* Several places within the instruction set 0 means "no register" 51 rather than TCG_REG_R0. */ 52#define TCG_REG_NONE 0 53 54/* A scratch register that may be be used throughout the backend. */ 55#define TCG_TMP0 TCG_REG_R1 56#define TCG_VEC_TMP0 TCG_REG_V31 57 58#define TCG_GUEST_BASE_REG TCG_REG_R13 59 60/* All of the following instructions are prefixed with their instruction 61 format, and are defined as 8- or 16-bit quantities, even when the two 62 halves of the 16-bit quantity may appear 32 bits apart in the insn. 63 This makes it easy to copy the values from the tables in Appendix B. */ 64typedef enum S390Opcode { 65 RIL_AFI = 0xc209, 66 RIL_AGFI = 0xc208, 67 RIL_ALFI = 0xc20b, 68 RIL_ALGFI = 0xc20a, 69 RIL_BRASL = 0xc005, 70 RIL_BRCL = 0xc004, 71 RIL_CFI = 0xc20d, 72 RIL_CGFI = 0xc20c, 73 RIL_CLFI = 0xc20f, 74 RIL_CLGFI = 0xc20e, 75 RIL_CLRL = 0xc60f, 76 RIL_CLGRL = 0xc60a, 77 RIL_CRL = 0xc60d, 78 RIL_CGRL = 0xc608, 79 RIL_IIHF = 0xc008, 80 RIL_IILF = 0xc009, 81 RIL_LARL = 0xc000, 82 RIL_LGFI = 0xc001, 83 RIL_LGRL = 0xc408, 84 RIL_LLIHF = 0xc00e, 85 RIL_LLILF = 0xc00f, 86 RIL_LRL = 0xc40d, 87 RIL_MSFI = 0xc201, 88 RIL_MSGFI = 0xc200, 89 RIL_NIHF = 0xc00a, 90 RIL_NILF = 0xc00b, 91 RIL_OIHF = 0xc00c, 92 RIL_OILF = 0xc00d, 93 RIL_SLFI = 0xc205, 94 RIL_SLGFI = 0xc204, 95 RIL_XIHF = 0xc006, 96 RIL_XILF = 0xc007, 97 98 RI_AGHI = 0xa70b, 99 RI_AHI = 0xa70a, 100 RI_BRC = 0xa704, 101 RI_CHI = 0xa70e, 102 RI_CGHI = 0xa70f, 103 RI_IIHH = 0xa500, 104 RI_IIHL = 0xa501, 105 RI_IILH = 0xa502, 106 RI_IILL = 0xa503, 107 RI_LGHI = 0xa709, 108 RI_LLIHH = 0xa50c, 109 RI_LLIHL = 0xa50d, 110 RI_LLILH = 0xa50e, 111 RI_LLILL = 0xa50f, 112 RI_MGHI = 0xa70d, 113 RI_MHI = 0xa70c, 114 RI_NIHH = 0xa504, 115 RI_NIHL = 0xa505, 116 RI_NILH = 0xa506, 117 RI_NILL = 0xa507, 118 RI_OIHH = 0xa508, 119 RI_OIHL = 0xa509, 120 RI_OILH = 0xa50a, 121 RI_OILL = 0xa50b, 122 RI_TMLL = 0xa701, 123 RI_TMLH = 0xa700, 124 RI_TMHL = 0xa703, 125 RI_TMHH = 0xa702, 126 127 RIEb_CGRJ = 0xec64, 128 RIEb_CLGRJ = 0xec65, 129 RIEb_CLRJ = 0xec77, 130 RIEb_CRJ = 0xec76, 131 132 RIEc_CGIJ = 0xec7c, 133 RIEc_CIJ = 0xec7e, 134 RIEc_CLGIJ = 0xec7d, 135 RIEc_CLIJ = 0xec7f, 136 137 RIEf_RISBG = 0xec55, 138 139 RIEg_LOCGHI = 0xec46, 140 141 RRE_AGR = 0xb908, 142 RRE_ALGR = 0xb90a, 143 RRE_ALCR = 0xb998, 144 RRE_ALCGR = 0xb988, 145 RRE_ALGFR = 0xb91a, 146 RRE_CGR = 0xb920, 147 RRE_CLGR = 0xb921, 148 RRE_DLGR = 0xb987, 149 RRE_DLR = 0xb997, 150 RRE_DSGFR = 0xb91d, 151 RRE_DSGR = 0xb90d, 152 RRE_FLOGR = 0xb983, 153 RRE_LGBR = 0xb906, 154 RRE_LCGR = 0xb903, 155 RRE_LGFR = 0xb914, 156 RRE_LGHR = 0xb907, 157 RRE_LGR = 0xb904, 158 RRE_LLGCR = 0xb984, 159 RRE_LLGFR = 0xb916, 160 RRE_LLGHR = 0xb985, 161 RRE_LRVR = 0xb91f, 162 RRE_LRVGR = 0xb90f, 163 RRE_LTGR = 0xb902, 164 RRE_MLGR = 0xb986, 165 RRE_MSGR = 0xb90c, 166 RRE_MSR = 0xb252, 167 RRE_NGR = 0xb980, 168 RRE_OGR = 0xb981, 169 RRE_SGR = 0xb909, 170 RRE_SLGR = 0xb90b, 171 RRE_SLBR = 0xb999, 172 RRE_SLBGR = 0xb989, 173 RRE_XGR = 0xb982, 174 175 RRFa_MGRK = 0xb9ec, 176 RRFa_MSRKC = 0xb9fd, 177 RRFa_MSGRKC = 0xb9ed, 178 RRFa_NCRK = 0xb9f5, 179 RRFa_NCGRK = 0xb9e5, 180 RRFa_NNRK = 0xb974, 181 RRFa_NNGRK = 0xb964, 182 RRFa_NORK = 0xb976, 183 RRFa_NOGRK = 0xb966, 184 RRFa_NRK = 0xb9f4, 185 RRFa_NGRK = 0xb9e4, 186 RRFa_NXRK = 0xb977, 187 RRFa_NXGRK = 0xb967, 188 RRFa_OCRK = 0xb975, 189 RRFa_OCGRK = 0xb965, 190 RRFa_ORK = 0xb9f6, 191 RRFa_OGRK = 0xb9e6, 192 RRFa_SRK = 0xb9f9, 193 RRFa_SGRK = 0xb9e9, 194 RRFa_SLRK = 0xb9fb, 195 RRFa_SLGRK = 0xb9eb, 196 RRFa_XRK = 0xb9f7, 197 RRFa_XGRK = 0xb9e7, 198 199 RRFam_SELGR = 0xb9e3, 200 201 RRFc_LOCR = 0xb9f2, 202 RRFc_LOCGR = 0xb9e2, 203 RRFc_POPCNT = 0xb9e1, 204 205 RR_AR = 0x1a, 206 RR_ALR = 0x1e, 207 RR_BASR = 0x0d, 208 RR_BCR = 0x07, 209 RR_CLR = 0x15, 210 RR_CR = 0x19, 211 RR_DR = 0x1d, 212 RR_LCR = 0x13, 213 RR_LR = 0x18, 214 RR_LTR = 0x12, 215 RR_NR = 0x14, 216 RR_OR = 0x16, 217 RR_SR = 0x1b, 218 RR_SLR = 0x1f, 219 RR_XR = 0x17, 220 221 RSY_RLL = 0xeb1d, 222 RSY_RLLG = 0xeb1c, 223 RSY_SLLG = 0xeb0d, 224 RSY_SLLK = 0xebdf, 225 RSY_SRAG = 0xeb0a, 226 RSY_SRAK = 0xebdc, 227 RSY_SRLG = 0xeb0c, 228 RSY_SRLK = 0xebde, 229 230 RS_SLL = 0x89, 231 RS_SRA = 0x8a, 232 RS_SRL = 0x88, 233 234 RXY_AG = 0xe308, 235 RXY_AY = 0xe35a, 236 RXY_CG = 0xe320, 237 RXY_CLG = 0xe321, 238 RXY_CLY = 0xe355, 239 RXY_CY = 0xe359, 240 RXY_LAY = 0xe371, 241 RXY_LB = 0xe376, 242 RXY_LG = 0xe304, 243 RXY_LGB = 0xe377, 244 RXY_LGF = 0xe314, 245 RXY_LGH = 0xe315, 246 RXY_LHY = 0xe378, 247 RXY_LLGC = 0xe390, 248 RXY_LLGF = 0xe316, 249 RXY_LLGH = 0xe391, 250 RXY_LMG = 0xeb04, 251 RXY_LPQ = 0xe38f, 252 RXY_LRV = 0xe31e, 253 RXY_LRVG = 0xe30f, 254 RXY_LRVH = 0xe31f, 255 RXY_LY = 0xe358, 256 RXY_NG = 0xe380, 257 RXY_OG = 0xe381, 258 RXY_STCY = 0xe372, 259 RXY_STG = 0xe324, 260 RXY_STHY = 0xe370, 261 RXY_STMG = 0xeb24, 262 RXY_STPQ = 0xe38e, 263 RXY_STRV = 0xe33e, 264 RXY_STRVG = 0xe32f, 265 RXY_STRVH = 0xe33f, 266 RXY_STY = 0xe350, 267 RXY_XG = 0xe382, 268 269 RX_A = 0x5a, 270 RX_C = 0x59, 271 RX_L = 0x58, 272 RX_LA = 0x41, 273 RX_LH = 0x48, 274 RX_ST = 0x50, 275 RX_STC = 0x42, 276 RX_STH = 0x40, 277 278 VRIa_VGBM = 0xe744, 279 VRIa_VREPI = 0xe745, 280 VRIb_VGM = 0xe746, 281 VRIc_VREP = 0xe74d, 282 283 VRRa_VLC = 0xe7de, 284 VRRa_VLP = 0xe7df, 285 VRRa_VLR = 0xe756, 286 VRRc_VA = 0xe7f3, 287 VRRc_VCEQ = 0xe7f8, /* we leave the m5 cs field 0 */ 288 VRRc_VCH = 0xe7fb, /* " */ 289 VRRc_VCHL = 0xe7f9, /* " */ 290 VRRc_VERLLV = 0xe773, 291 VRRc_VESLV = 0xe770, 292 VRRc_VESRAV = 0xe77a, 293 VRRc_VESRLV = 0xe778, 294 VRRc_VML = 0xe7a2, 295 VRRc_VMN = 0xe7fe, 296 VRRc_VMNL = 0xe7fc, 297 VRRc_VMX = 0xe7ff, 298 VRRc_VMXL = 0xe7fd, 299 VRRc_VN = 0xe768, 300 VRRc_VNC = 0xe769, 301 VRRc_VNN = 0xe76e, 302 VRRc_VNO = 0xe76b, 303 VRRc_VNX = 0xe76c, 304 VRRc_VO = 0xe76a, 305 VRRc_VOC = 0xe76f, 306 VRRc_VPKS = 0xe797, /* we leave the m5 cs field 0 */ 307 VRRc_VS = 0xe7f7, 308 VRRa_VUPH = 0xe7d7, 309 VRRa_VUPL = 0xe7d6, 310 VRRc_VX = 0xe76d, 311 VRRe_VSEL = 0xe78d, 312 VRRf_VLVGP = 0xe762, 313 314 VRSa_VERLL = 0xe733, 315 VRSa_VESL = 0xe730, 316 VRSa_VESRA = 0xe73a, 317 VRSa_VESRL = 0xe738, 318 VRSb_VLVG = 0xe722, 319 VRSc_VLGV = 0xe721, 320 321 VRX_VL = 0xe706, 322 VRX_VLLEZ = 0xe704, 323 VRX_VLREP = 0xe705, 324 VRX_VST = 0xe70e, 325 VRX_VSTEF = 0xe70b, 326 VRX_VSTEG = 0xe70a, 327 328 NOP = 0x0707, 329} S390Opcode; 330 331#ifdef CONFIG_DEBUG_TCG 332static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = { 333 "%r0", "%r1", "%r2", "%r3", "%r4", "%r5", "%r6", "%r7", 334 "%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15", 335 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 336 "%v0", "%v1", "%v2", "%v3", "%v4", "%v5", "%v6", "%v7", 337 "%v8", "%v9", "%v10", "%v11", "%v12", "%v13", "%v14", "%v15", 338 "%v16", "%v17", "%v18", "%v19", "%v20", "%v21", "%v22", "%v23", 339 "%v24", "%v25", "%v26", "%v27", "%v28", "%v29", "%v30", "%v31", 340}; 341#endif 342 343/* Since R6 is a potential argument register, choose it last of the 344 call-saved registers. Likewise prefer the call-clobbered registers 345 in reverse order to maximize the chance of avoiding the arguments. */ 346static const int tcg_target_reg_alloc_order[] = { 347 /* Call saved registers. */ 348 TCG_REG_R13, 349 TCG_REG_R12, 350 TCG_REG_R11, 351 TCG_REG_R10, 352 TCG_REG_R9, 353 TCG_REG_R8, 354 TCG_REG_R7, 355 TCG_REG_R6, 356 /* Call clobbered registers. */ 357 TCG_REG_R14, 358 TCG_REG_R0, 359 TCG_REG_R1, 360 /* Argument registers, in reverse order of allocation. */ 361 TCG_REG_R5, 362 TCG_REG_R4, 363 TCG_REG_R3, 364 TCG_REG_R2, 365 366 /* V8-V15 are call saved, and omitted. */ 367 TCG_REG_V0, 368 TCG_REG_V1, 369 TCG_REG_V2, 370 TCG_REG_V3, 371 TCG_REG_V4, 372 TCG_REG_V5, 373 TCG_REG_V6, 374 TCG_REG_V7, 375 TCG_REG_V16, 376 TCG_REG_V17, 377 TCG_REG_V18, 378 TCG_REG_V19, 379 TCG_REG_V20, 380 TCG_REG_V21, 381 TCG_REG_V22, 382 TCG_REG_V23, 383 TCG_REG_V24, 384 TCG_REG_V25, 385 TCG_REG_V26, 386 TCG_REG_V27, 387 TCG_REG_V28, 388 TCG_REG_V29, 389 TCG_REG_V30, 390 TCG_REG_V31, 391}; 392 393static const int tcg_target_call_iarg_regs[] = { 394 TCG_REG_R2, 395 TCG_REG_R3, 396 TCG_REG_R4, 397 TCG_REG_R5, 398 TCG_REG_R6, 399}; 400 401static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot) 402{ 403 tcg_debug_assert(kind == TCG_CALL_RET_NORMAL); 404 tcg_debug_assert(slot == 0); 405 return TCG_REG_R2; 406} 407 408#define S390_CC_EQ 8 409#define S390_CC_LT 4 410#define S390_CC_GT 2 411#define S390_CC_OV 1 412#define S390_CC_NE (S390_CC_LT | S390_CC_GT) 413#define S390_CC_LE (S390_CC_LT | S390_CC_EQ) 414#define S390_CC_GE (S390_CC_GT | S390_CC_EQ) 415#define S390_CC_NEVER 0 416#define S390_CC_ALWAYS 15 417 418#define S390_TM_EQ 8 /* CC == 0 */ 419#define S390_TM_NE 7 /* CC in {1,2,3} */ 420 421/* Condition codes that result from a COMPARE and COMPARE LOGICAL. */ 422static const uint8_t tcg_cond_to_s390_cond[16] = { 423 [TCG_COND_EQ] = S390_CC_EQ, 424 [TCG_COND_NE] = S390_CC_NE, 425 [TCG_COND_TSTEQ] = S390_CC_EQ, 426 [TCG_COND_TSTNE] = S390_CC_NE, 427 [TCG_COND_LT] = S390_CC_LT, 428 [TCG_COND_LE] = S390_CC_LE, 429 [TCG_COND_GT] = S390_CC_GT, 430 [TCG_COND_GE] = S390_CC_GE, 431 [TCG_COND_LTU] = S390_CC_LT, 432 [TCG_COND_LEU] = S390_CC_LE, 433 [TCG_COND_GTU] = S390_CC_GT, 434 [TCG_COND_GEU] = S390_CC_GE, 435}; 436 437/* Condition codes that result from a LOAD AND TEST. Here, we have no 438 unsigned instruction variation, however since the test is vs zero we 439 can re-map the outcomes appropriately. */ 440static const uint8_t tcg_cond_to_ltr_cond[16] = { 441 [TCG_COND_EQ] = S390_CC_EQ, 442 [TCG_COND_NE] = S390_CC_NE, 443 [TCG_COND_TSTEQ] = S390_CC_ALWAYS, 444 [TCG_COND_TSTNE] = S390_CC_NEVER, 445 [TCG_COND_LT] = S390_CC_LT, 446 [TCG_COND_LE] = S390_CC_LE, 447 [TCG_COND_GT] = S390_CC_GT, 448 [TCG_COND_GE] = S390_CC_GE, 449 [TCG_COND_LTU] = S390_CC_NEVER, 450 [TCG_COND_LEU] = S390_CC_EQ, 451 [TCG_COND_GTU] = S390_CC_NE, 452 [TCG_COND_GEU] = S390_CC_ALWAYS, 453}; 454 455static const tcg_insn_unit *tb_ret_addr; 456uint64_t s390_facilities[3]; 457 458static inline bool is_general_reg(TCGReg r) 459{ 460 return r <= TCG_REG_R15; 461} 462 463static inline bool is_vector_reg(TCGReg r) 464{ 465 return r >= TCG_REG_V0 && r <= TCG_REG_V31; 466} 467 468static bool patch_reloc(tcg_insn_unit *src_rw, int type, 469 intptr_t value, intptr_t addend) 470{ 471 const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw); 472 intptr_t pcrel2; 473 uint32_t old; 474 475 value += addend; 476 pcrel2 = (tcg_insn_unit *)value - src_rx; 477 478 switch (type) { 479 case R_390_PC16DBL: 480 if (pcrel2 == (int16_t)pcrel2) { 481 tcg_patch16(src_rw, pcrel2); 482 return true; 483 } 484 break; 485 case R_390_PC32DBL: 486 if (pcrel2 == (int32_t)pcrel2) { 487 tcg_patch32(src_rw, pcrel2); 488 return true; 489 } 490 break; 491 case R_390_20: 492 if (value == sextract64(value, 0, 20)) { 493 old = *(uint32_t *)src_rw & 0xf00000ff; 494 old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4); 495 tcg_patch32(src_rw, old); 496 return true; 497 } 498 break; 499 default: 500 g_assert_not_reached(); 501 } 502 return false; 503} 504 505static int is_const_p16(uint64_t val) 506{ 507 for (int i = 0; i < 4; ++i) { 508 uint64_t mask = 0xffffull << (i * 16); 509 if ((val & ~mask) == 0) { 510 return i; 511 } 512 } 513 return -1; 514} 515 516static int is_const_p32(uint64_t val) 517{ 518 if ((val & 0xffffffff00000000ull) == 0) { 519 return 0; 520 } 521 if ((val & 0x00000000ffffffffull) == 0) { 522 return 1; 523 } 524 return -1; 525} 526 527/* 528 * Accept bit patterns like these: 529 * 0....01....1 530 * 1....10....0 531 * 1..10..01..1 532 * 0..01..10..0 533 * Copied from gcc sources. 534 */ 535static bool risbg_mask(uint64_t c) 536{ 537 uint64_t lsb; 538 /* We don't change the number of transitions by inverting, 539 so make sure we start with the LSB zero. */ 540 if (c & 1) { 541 c = ~c; 542 } 543 /* Reject all zeros or all ones. */ 544 if (c == 0) { 545 return false; 546 } 547 /* Find the first transition. */ 548 lsb = c & -c; 549 /* Invert to look for a second transition. */ 550 c = ~c; 551 /* Erase the first transition. */ 552 c &= -lsb; 553 /* Find the second transition, if any. */ 554 lsb = c & -c; 555 /* Match if all the bits are 1's, or if c is zero. */ 556 return c == -lsb; 557} 558 559/* Test if a constant matches the constraint. */ 560static bool tcg_target_const_match(int64_t val, int ct, 561 TCGType type, TCGCond cond, int vece) 562{ 563 uint64_t uval = val; 564 565 if (ct & TCG_CT_CONST) { 566 return true; 567 } 568 if (type == TCG_TYPE_I32) { 569 uval = (uint32_t)val; 570 val = (int32_t)val; 571 } 572 573 if (ct & TCG_CT_CONST_CMP) { 574 if (is_tst_cond(cond)) { 575 if (is_const_p16(uval) >= 0) { 576 return true; /* TMxx */ 577 } 578 if (risbg_mask(uval)) { 579 return true; /* RISBG */ 580 } 581 return false; 582 } 583 584 if (type == TCG_TYPE_I32) { 585 return true; 586 } 587 588 switch (cond) { 589 case TCG_COND_EQ: 590 case TCG_COND_NE: 591 ct |= TCG_CT_CONST_S32 | TCG_CT_CONST_U32; /* CGFI or CLGFI */ 592 break; 593 case TCG_COND_LT: 594 case TCG_COND_GE: 595 case TCG_COND_LE: 596 case TCG_COND_GT: 597 ct |= TCG_CT_CONST_S32; /* CGFI */ 598 break; 599 case TCG_COND_LTU: 600 case TCG_COND_GEU: 601 case TCG_COND_LEU: 602 case TCG_COND_GTU: 603 ct |= TCG_CT_CONST_U32; /* CLGFI */ 604 break; 605 case TCG_COND_TSTNE: 606 case TCG_COND_TSTEQ: 607 /* checked above, fallthru */ 608 default: 609 g_assert_not_reached(); 610 } 611 } 612 613 if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) { 614 return true; 615 } 616 if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) { 617 return true; 618 } 619 if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) { 620 return true; 621 } 622 if ((ct & TCG_CT_CONST_ZERO) && val == 0) { 623 return true; 624 } 625 if ((ct & TCG_CT_CONST_M1) && val == -1) { 626 return true; 627 } 628 629 if (ct & TCG_CT_CONST_INV) { 630 val = ~val; 631 } 632 if ((ct & TCG_CT_CONST_P32) && is_const_p32(val) >= 0) { 633 return true; 634 } 635 if ((ct & TCG_CT_CONST_INVRISBG) && risbg_mask(~val)) { 636 return true; 637 } 638 return false; 639} 640 641/* Emit instructions according to the given instruction format. */ 642 643static void tcg_out_insn_RR(TCGContext *s, S390Opcode op, TCGReg r1, TCGReg r2) 644{ 645 tcg_out16(s, (op << 8) | (r1 << 4) | r2); 646} 647 648static void tcg_out_insn_RRE(TCGContext *s, S390Opcode op, 649 TCGReg r1, TCGReg r2) 650{ 651 tcg_out32(s, (op << 16) | (r1 << 4) | r2); 652} 653 654/* RRF-a without the m4 field */ 655static void tcg_out_insn_RRFa(TCGContext *s, S390Opcode op, 656 TCGReg r1, TCGReg r2, TCGReg r3) 657{ 658 tcg_out32(s, (op << 16) | (r3 << 12) | (r1 << 4) | r2); 659} 660 661/* RRF-a with the m4 field */ 662static void tcg_out_insn_RRFam(TCGContext *s, S390Opcode op, 663 TCGReg r1, TCGReg r2, TCGReg r3, int m4) 664{ 665 tcg_out32(s, (op << 16) | (r3 << 12) | (m4 << 8) | (r1 << 4) | r2); 666} 667 668static void tcg_out_insn_RRFc(TCGContext *s, S390Opcode op, 669 TCGReg r1, TCGReg r2, int m3) 670{ 671 tcg_out32(s, (op << 16) | (m3 << 12) | (r1 << 4) | r2); 672} 673 674static void tcg_out_insn_RI(TCGContext *s, S390Opcode op, TCGReg r1, int i2) 675{ 676 tcg_out32(s, (op << 16) | (r1 << 20) | (i2 & 0xffff)); 677} 678 679static void tcg_out_insn_RIEg(TCGContext *s, S390Opcode op, TCGReg r1, 680 int i2, int m3) 681{ 682 tcg_out16(s, (op & 0xff00) | (r1 << 4) | m3); 683 tcg_out32(s, (i2 << 16) | (op & 0xff)); 684} 685 686static void tcg_out_insn_RIL(TCGContext *s, S390Opcode op, TCGReg r1, int i2) 687{ 688 tcg_out16(s, op | (r1 << 4)); 689 tcg_out32(s, i2); 690} 691 692static void tcg_out_insn_RS(TCGContext *s, S390Opcode op, TCGReg r1, 693 TCGReg b2, TCGReg r3, int disp) 694{ 695 tcg_out32(s, (op << 24) | (r1 << 20) | (r3 << 16) | (b2 << 12) 696 | (disp & 0xfff)); 697} 698 699static void tcg_out_insn_RSY(TCGContext *s, S390Opcode op, TCGReg r1, 700 TCGReg b2, TCGReg r3, int disp) 701{ 702 tcg_out16(s, (op & 0xff00) | (r1 << 4) | r3); 703 tcg_out32(s, (op & 0xff) | (b2 << 28) 704 | ((disp & 0xfff) << 16) | ((disp & 0xff000) >> 4)); 705} 706 707#define tcg_out_insn_RX tcg_out_insn_RS 708#define tcg_out_insn_RXY tcg_out_insn_RSY 709 710static int RXB(TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4) 711{ 712 /* 713 * Shift bit 4 of each regno to its corresponding bit of RXB. 714 * RXB itself begins at bit 8 of the instruction so 8 - 4 = 4 715 * is the left-shift of the 4th operand. 716 */ 717 return ((v1 & 0x10) << (4 + 3)) 718 | ((v2 & 0x10) << (4 + 2)) 719 | ((v3 & 0x10) << (4 + 1)) 720 | ((v4 & 0x10) << (4 + 0)); 721} 722 723static void tcg_out_insn_VRIa(TCGContext *s, S390Opcode op, 724 TCGReg v1, uint16_t i2, int m3) 725{ 726 tcg_debug_assert(is_vector_reg(v1)); 727 tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4)); 728 tcg_out16(s, i2); 729 tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12)); 730} 731 732static void tcg_out_insn_VRIb(TCGContext *s, S390Opcode op, 733 TCGReg v1, uint8_t i2, uint8_t i3, int m4) 734{ 735 tcg_debug_assert(is_vector_reg(v1)); 736 tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4)); 737 tcg_out16(s, (i2 << 8) | (i3 & 0xff)); 738 tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12)); 739} 740 741static void tcg_out_insn_VRIc(TCGContext *s, S390Opcode op, 742 TCGReg v1, uint16_t i2, TCGReg v3, int m4) 743{ 744 tcg_debug_assert(is_vector_reg(v1)); 745 tcg_debug_assert(is_vector_reg(v3)); 746 tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf)); 747 tcg_out16(s, i2); 748 tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12)); 749} 750 751static void tcg_out_insn_VRRa(TCGContext *s, S390Opcode op, 752 TCGReg v1, TCGReg v2, int m3) 753{ 754 tcg_debug_assert(is_vector_reg(v1)); 755 tcg_debug_assert(is_vector_reg(v2)); 756 tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf)); 757 tcg_out32(s, (op & 0x00ff) | RXB(v1, v2, 0, 0) | (m3 << 12)); 758} 759 760static void tcg_out_insn_VRRc(TCGContext *s, S390Opcode op, 761 TCGReg v1, TCGReg v2, TCGReg v3, int m4) 762{ 763 tcg_debug_assert(is_vector_reg(v1)); 764 tcg_debug_assert(is_vector_reg(v2)); 765 tcg_debug_assert(is_vector_reg(v3)); 766 tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf)); 767 tcg_out16(s, v3 << 12); 768 tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, 0) | (m4 << 12)); 769} 770 771static void tcg_out_insn_VRRe(TCGContext *s, S390Opcode op, 772 TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4) 773{ 774 tcg_debug_assert(is_vector_reg(v1)); 775 tcg_debug_assert(is_vector_reg(v2)); 776 tcg_debug_assert(is_vector_reg(v3)); 777 tcg_debug_assert(is_vector_reg(v4)); 778 tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf)); 779 tcg_out16(s, v3 << 12); 780 tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, v4) | (v4 << 12)); 781} 782 783static void tcg_out_insn_VRRf(TCGContext *s, S390Opcode op, 784 TCGReg v1, TCGReg r2, TCGReg r3) 785{ 786 tcg_debug_assert(is_vector_reg(v1)); 787 tcg_debug_assert(is_general_reg(r2)); 788 tcg_debug_assert(is_general_reg(r3)); 789 tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r2); 790 tcg_out16(s, r3 << 12); 791 tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0)); 792} 793 794static void tcg_out_insn_VRSa(TCGContext *s, S390Opcode op, TCGReg v1, 795 intptr_t d2, TCGReg b2, TCGReg v3, int m4) 796{ 797 tcg_debug_assert(is_vector_reg(v1)); 798 tcg_debug_assert(d2 >= 0 && d2 <= 0xfff); 799 tcg_debug_assert(is_general_reg(b2)); 800 tcg_debug_assert(is_vector_reg(v3)); 801 tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf)); 802 tcg_out16(s, b2 << 12 | d2); 803 tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12)); 804} 805 806static void tcg_out_insn_VRSb(TCGContext *s, S390Opcode op, TCGReg v1, 807 intptr_t d2, TCGReg b2, TCGReg r3, int m4) 808{ 809 tcg_debug_assert(is_vector_reg(v1)); 810 tcg_debug_assert(d2 >= 0 && d2 <= 0xfff); 811 tcg_debug_assert(is_general_reg(b2)); 812 tcg_debug_assert(is_general_reg(r3)); 813 tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r3); 814 tcg_out16(s, b2 << 12 | d2); 815 tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12)); 816} 817 818static void tcg_out_insn_VRSc(TCGContext *s, S390Opcode op, TCGReg r1, 819 intptr_t d2, TCGReg b2, TCGReg v3, int m4) 820{ 821 tcg_debug_assert(is_general_reg(r1)); 822 tcg_debug_assert(d2 >= 0 && d2 <= 0xfff); 823 tcg_debug_assert(is_general_reg(b2)); 824 tcg_debug_assert(is_vector_reg(v3)); 825 tcg_out16(s, (op & 0xff00) | (r1 << 4) | (v3 & 0xf)); 826 tcg_out16(s, b2 << 12 | d2); 827 tcg_out16(s, (op & 0x00ff) | RXB(0, v3, 0, 0) | (m4 << 12)); 828} 829 830static void tcg_out_insn_VRX(TCGContext *s, S390Opcode op, TCGReg v1, 831 TCGReg b2, TCGReg x2, intptr_t d2, int m3) 832{ 833 tcg_debug_assert(is_vector_reg(v1)); 834 tcg_debug_assert(d2 >= 0 && d2 <= 0xfff); 835 tcg_debug_assert(is_general_reg(x2)); 836 tcg_debug_assert(is_general_reg(b2)); 837 tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | x2); 838 tcg_out16(s, (b2 << 12) | d2); 839 tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12)); 840} 841 842/* Emit an opcode with "type-checking" of the format. */ 843#define tcg_out_insn(S, FMT, OP, ...) \ 844 glue(tcg_out_insn_,FMT)(S, glue(glue(FMT,_),OP), ## __VA_ARGS__) 845 846 847/* emit 64-bit shifts */ 848static void tcg_out_sh64(TCGContext* s, S390Opcode op, TCGReg dest, 849 TCGReg src, TCGReg sh_reg, int sh_imm) 850{ 851 tcg_out_insn_RSY(s, op, dest, sh_reg, src, sh_imm); 852} 853 854/* emit 32-bit shifts */ 855static void tcg_out_sh32(TCGContext* s, S390Opcode op, TCGReg dest, 856 TCGReg sh_reg, int sh_imm) 857{ 858 tcg_out_insn_RS(s, op, dest, sh_reg, 0, sh_imm); 859} 860 861static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src) 862{ 863 if (src == dst) { 864 return true; 865 } 866 switch (type) { 867 case TCG_TYPE_I32: 868 if (likely(is_general_reg(dst) && is_general_reg(src))) { 869 tcg_out_insn(s, RR, LR, dst, src); 870 break; 871 } 872 /* fallthru */ 873 874 case TCG_TYPE_I64: 875 if (likely(is_general_reg(dst))) { 876 if (likely(is_general_reg(src))) { 877 tcg_out_insn(s, RRE, LGR, dst, src); 878 } else { 879 tcg_out_insn(s, VRSc, VLGV, dst, 0, 0, src, 3); 880 } 881 break; 882 } else if (is_general_reg(src)) { 883 tcg_out_insn(s, VRSb, VLVG, dst, 0, 0, src, 3); 884 break; 885 } 886 /* fallthru */ 887 888 case TCG_TYPE_V64: 889 case TCG_TYPE_V128: 890 tcg_out_insn(s, VRRa, VLR, dst, src, 0); 891 break; 892 893 default: 894 g_assert_not_reached(); 895 } 896 return true; 897} 898 899static const S390Opcode li_insns[4] = { 900 RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH 901}; 902static const S390Opcode oi_insns[4] = { 903 RI_OILL, RI_OILH, RI_OIHL, RI_OIHH 904}; 905static const S390Opcode lif_insns[2] = { 906 RIL_LLILF, RIL_LLIHF, 907}; 908static const S390Opcode tm_insns[4] = { 909 RI_TMLL, RI_TMLH, RI_TMHL, RI_TMHH 910}; 911 912/* load a register with an immediate value */ 913static void tcg_out_movi(TCGContext *s, TCGType type, 914 TCGReg ret, tcg_target_long sval) 915{ 916 tcg_target_ulong uval = sval; 917 ptrdiff_t pc_off; 918 int i; 919 920 if (type == TCG_TYPE_I32) { 921 uval = (uint32_t)sval; 922 sval = (int32_t)sval; 923 } 924 925 /* Try all 32-bit insns that can load it in one go. */ 926 if (sval >= -0x8000 && sval < 0x8000) { 927 tcg_out_insn(s, RI, LGHI, ret, sval); 928 return; 929 } 930 931 i = is_const_p16(uval); 932 if (i >= 0) { 933 tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16)); 934 return; 935 } 936 937 /* Try all 48-bit insns that can load it in one go. */ 938 if (sval == (int32_t)sval) { 939 tcg_out_insn(s, RIL, LGFI, ret, sval); 940 return; 941 } 942 943 i = is_const_p32(uval); 944 if (i >= 0) { 945 tcg_out_insn_RIL(s, lif_insns[i], ret, uval >> (i * 32)); 946 return; 947 } 948 949 /* Try for PC-relative address load. For odd addresses, add one. */ 950 pc_off = tcg_pcrel_diff(s, (void *)sval) >> 1; 951 if (pc_off == (int32_t)pc_off) { 952 tcg_out_insn(s, RIL, LARL, ret, pc_off); 953 if (sval & 1) { 954 tcg_out_insn(s, RI, AGHI, ret, 1); 955 } 956 return; 957 } 958 959 /* Otherwise, load it by parts. */ 960 i = is_const_p16((uint32_t)uval); 961 if (i >= 0) { 962 tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16)); 963 } else { 964 tcg_out_insn(s, RIL, LLILF, ret, uval); 965 } 966 uval >>= 32; 967 i = is_const_p16(uval); 968 if (i >= 0) { 969 tcg_out_insn_RI(s, oi_insns[i + 2], ret, uval >> (i * 16)); 970 } else { 971 tcg_out_insn(s, RIL, OIHF, ret, uval); 972 } 973} 974 975/* Emit a load/store type instruction. Inputs are: 976 DATA: The register to be loaded or stored. 977 BASE+OFS: The effective address. 978 OPC_RX: If the operation has an RX format opcode (e.g. STC), otherwise 0. 979 OPC_RXY: The RXY format opcode for the operation (e.g. STCY). */ 980 981static void tcg_out_mem(TCGContext *s, S390Opcode opc_rx, S390Opcode opc_rxy, 982 TCGReg data, TCGReg base, TCGReg index, 983 tcg_target_long ofs) 984{ 985 if (ofs < -0x80000 || ofs >= 0x80000) { 986 /* Combine the low 20 bits of the offset with the actual load insn; 987 the high 44 bits must come from an immediate load. */ 988 tcg_target_long low = ((ofs & 0xfffff) ^ 0x80000) - 0x80000; 989 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - low); 990 ofs = low; 991 992 /* If we were already given an index register, add it in. */ 993 if (index != TCG_REG_NONE) { 994 tcg_out_insn(s, RRE, AGR, TCG_TMP0, index); 995 } 996 index = TCG_TMP0; 997 } 998 999 if (opc_rx && ofs >= 0 && ofs < 0x1000) { 1000 tcg_out_insn_RX(s, opc_rx, data, base, index, ofs); 1001 } else { 1002 tcg_out_insn_RXY(s, opc_rxy, data, base, index, ofs); 1003 } 1004} 1005 1006static void tcg_out_vrx_mem(TCGContext *s, S390Opcode opc_vrx, 1007 TCGReg data, TCGReg base, TCGReg index, 1008 tcg_target_long ofs, int m3) 1009{ 1010 if (ofs < 0 || ofs >= 0x1000) { 1011 if (ofs >= -0x80000 && ofs < 0x80000) { 1012 tcg_out_insn(s, RXY, LAY, TCG_TMP0, base, index, ofs); 1013 base = TCG_TMP0; 1014 index = TCG_REG_NONE; 1015 ofs = 0; 1016 } else { 1017 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs); 1018 if (index != TCG_REG_NONE) { 1019 tcg_out_insn(s, RRE, AGR, TCG_TMP0, index); 1020 } 1021 index = TCG_TMP0; 1022 ofs = 0; 1023 } 1024 } 1025 tcg_out_insn_VRX(s, opc_vrx, data, base, index, ofs, m3); 1026} 1027 1028/* load data without address translation or endianness conversion */ 1029static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg data, 1030 TCGReg base, intptr_t ofs) 1031{ 1032 switch (type) { 1033 case TCG_TYPE_I32: 1034 if (likely(is_general_reg(data))) { 1035 tcg_out_mem(s, RX_L, RXY_LY, data, base, TCG_REG_NONE, ofs); 1036 break; 1037 } 1038 tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_32); 1039 break; 1040 1041 case TCG_TYPE_I64: 1042 if (likely(is_general_reg(data))) { 1043 tcg_out_mem(s, 0, RXY_LG, data, base, TCG_REG_NONE, ofs); 1044 break; 1045 } 1046 /* fallthru */ 1047 1048 case TCG_TYPE_V64: 1049 tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_64); 1050 break; 1051 1052 case TCG_TYPE_V128: 1053 /* Hint quadword aligned. */ 1054 tcg_out_vrx_mem(s, VRX_VL, data, base, TCG_REG_NONE, ofs, 4); 1055 break; 1056 1057 default: 1058 g_assert_not_reached(); 1059 } 1060} 1061 1062static void tcg_out_st(TCGContext *s, TCGType type, TCGReg data, 1063 TCGReg base, intptr_t ofs) 1064{ 1065 switch (type) { 1066 case TCG_TYPE_I32: 1067 if (likely(is_general_reg(data))) { 1068 tcg_out_mem(s, RX_ST, RXY_STY, data, base, TCG_REG_NONE, ofs); 1069 } else { 1070 tcg_out_vrx_mem(s, VRX_VSTEF, data, base, TCG_REG_NONE, ofs, 1); 1071 } 1072 break; 1073 1074 case TCG_TYPE_I64: 1075 if (likely(is_general_reg(data))) { 1076 tcg_out_mem(s, 0, RXY_STG, data, base, TCG_REG_NONE, ofs); 1077 break; 1078 } 1079 /* fallthru */ 1080 1081 case TCG_TYPE_V64: 1082 tcg_out_vrx_mem(s, VRX_VSTEG, data, base, TCG_REG_NONE, ofs, 0); 1083 break; 1084 1085 case TCG_TYPE_V128: 1086 /* Hint quadword aligned. */ 1087 tcg_out_vrx_mem(s, VRX_VST, data, base, TCG_REG_NONE, ofs, 4); 1088 break; 1089 1090 default: 1091 g_assert_not_reached(); 1092 } 1093} 1094 1095static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val, 1096 TCGReg base, intptr_t ofs) 1097{ 1098 return false; 1099} 1100 1101static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2) 1102{ 1103 return false; 1104} 1105 1106static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs, 1107 tcg_target_long imm) 1108{ 1109 /* This function is only used for passing structs by reference. */ 1110 tcg_out_mem(s, RX_LA, RXY_LAY, rd, rs, TCG_REG_NONE, imm); 1111} 1112 1113static inline void tcg_out_risbg(TCGContext *s, TCGReg dest, TCGReg src, 1114 int msb, int lsb, int ofs, int z) 1115{ 1116 /* Format RIE-f */ 1117 tcg_out16(s, (RIEf_RISBG & 0xff00) | (dest << 4) | src); 1118 tcg_out16(s, (msb << 8) | (z << 7) | lsb); 1119 tcg_out16(s, (ofs << 8) | (RIEf_RISBG & 0xff)); 1120} 1121 1122static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src) 1123{ 1124 tcg_out_insn(s, RRE, LGBR, dest, src); 1125} 1126 1127static void tcg_out_ext8u(TCGContext *s, TCGReg dest, TCGReg src) 1128{ 1129 tcg_out_insn(s, RRE, LLGCR, dest, src); 1130} 1131 1132static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src) 1133{ 1134 tcg_out_insn(s, RRE, LGHR, dest, src); 1135} 1136 1137static void tcg_out_ext16u(TCGContext *s, TCGReg dest, TCGReg src) 1138{ 1139 tcg_out_insn(s, RRE, LLGHR, dest, src); 1140} 1141 1142static void tcg_out_ext32s(TCGContext *s, TCGReg dest, TCGReg src) 1143{ 1144 tcg_out_insn(s, RRE, LGFR, dest, src); 1145} 1146 1147static void tcg_out_ext32u(TCGContext *s, TCGReg dest, TCGReg src) 1148{ 1149 tcg_out_insn(s, RRE, LLGFR, dest, src); 1150} 1151 1152static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dest, TCGReg src) 1153{ 1154 tcg_out_ext32s(s, dest, src); 1155} 1156 1157static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dest, TCGReg src) 1158{ 1159 tcg_out_ext32u(s, dest, src); 1160} 1161 1162static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg dest, TCGReg src) 1163{ 1164 tcg_out_mov(s, TCG_TYPE_I32, dest, src); 1165} 1166 1167static void tgen_andi_risbg(TCGContext *s, TCGReg out, TCGReg in, uint64_t val) 1168{ 1169 int msb, lsb; 1170 if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) { 1171 /* Achieve wraparound by swapping msb and lsb. */ 1172 msb = 64 - ctz64(~val); 1173 lsb = clz64(~val) - 1; 1174 } else { 1175 msb = clz64(val); 1176 lsb = 63 - ctz64(val); 1177 } 1178 tcg_out_risbg(s, out, in, msb, lsb, 0, 1); 1179} 1180 1181static void tgen_andi(TCGContext *s, TCGType type, TCGReg dest, uint64_t val) 1182{ 1183 static const S390Opcode ni_insns[4] = { 1184 RI_NILL, RI_NILH, RI_NIHL, RI_NIHH 1185 }; 1186 static const S390Opcode nif_insns[2] = { 1187 RIL_NILF, RIL_NIHF 1188 }; 1189 uint64_t valid = (type == TCG_TYPE_I32 ? 0xffffffffull : -1ull); 1190 int i; 1191 1192 /* Look for the zero-extensions. */ 1193 if ((val & valid) == 0xffffffff) { 1194 tcg_out_ext32u(s, dest, dest); 1195 return; 1196 } 1197 if ((val & valid) == 0xff) { 1198 tcg_out_ext8u(s, dest, dest); 1199 return; 1200 } 1201 if ((val & valid) == 0xffff) { 1202 tcg_out_ext16u(s, dest, dest); 1203 return; 1204 } 1205 1206 i = is_const_p16(~val & valid); 1207 if (i >= 0) { 1208 tcg_out_insn_RI(s, ni_insns[i], dest, val >> (i * 16)); 1209 return; 1210 } 1211 1212 i = is_const_p32(~val & valid); 1213 tcg_debug_assert(i == 0 || type != TCG_TYPE_I32); 1214 if (i >= 0) { 1215 tcg_out_insn_RIL(s, nif_insns[i], dest, val >> (i * 32)); 1216 return; 1217 } 1218 1219 if (risbg_mask(val)) { 1220 tgen_andi_risbg(s, dest, dest, val); 1221 return; 1222 } 1223 1224 g_assert_not_reached(); 1225} 1226 1227static void tgen_ori(TCGContext *s, TCGReg dest, uint64_t val) 1228{ 1229 static const S390Opcode oif_insns[2] = { 1230 RIL_OILF, RIL_OIHF 1231 }; 1232 1233 int i; 1234 1235 i = is_const_p16(val); 1236 if (i >= 0) { 1237 tcg_out_insn_RI(s, oi_insns[i], dest, val >> (i * 16)); 1238 return; 1239 } 1240 1241 i = is_const_p32(val); 1242 if (i >= 0) { 1243 tcg_out_insn_RIL(s, oif_insns[i], dest, val >> (i * 32)); 1244 return; 1245 } 1246 1247 g_assert_not_reached(); 1248} 1249 1250static void tgen_xori(TCGContext *s, TCGReg dest, uint64_t val) 1251{ 1252 switch (is_const_p32(val)) { 1253 case 0: 1254 tcg_out_insn(s, RIL, XILF, dest, val); 1255 break; 1256 case 1: 1257 tcg_out_insn(s, RIL, XIHF, dest, val >> 32); 1258 break; 1259 default: 1260 g_assert_not_reached(); 1261 } 1262} 1263 1264static int tgen_cmp2(TCGContext *s, TCGType type, TCGCond c, TCGReg r1, 1265 TCGArg c2, bool c2const, bool need_carry, int *inv_cc) 1266{ 1267 bool is_unsigned = is_unsigned_cond(c); 1268 TCGCond inv_c = tcg_invert_cond(c); 1269 S390Opcode op; 1270 1271 if (is_tst_cond(c)) { 1272 tcg_debug_assert(!need_carry); 1273 1274 if (!c2const) { 1275 if (type == TCG_TYPE_I32) { 1276 tcg_out_insn(s, RRFa, NRK, TCG_REG_R0, r1, c2); 1277 } else { 1278 tcg_out_insn(s, RRFa, NGRK, TCG_REG_R0, r1, c2); 1279 } 1280 goto exit; 1281 } 1282 1283 if (type == TCG_TYPE_I32) { 1284 c2 = (uint32_t)c2; 1285 } 1286 1287 int i = is_const_p16(c2); 1288 if (i >= 0) { 1289 tcg_out_insn_RI(s, tm_insns[i], r1, c2 >> (i * 16)); 1290 *inv_cc = c == TCG_COND_TSTEQ ? S390_TM_NE : S390_TM_EQ; 1291 return *inv_cc ^ 15; 1292 } 1293 1294 if (risbg_mask(c2)) { 1295 tgen_andi_risbg(s, TCG_REG_R0, r1, c2); 1296 goto exit; 1297 } 1298 g_assert_not_reached(); 1299 } 1300 1301 if (c2const) { 1302 if (c2 == 0) { 1303 if (!(is_unsigned && need_carry)) { 1304 if (type == TCG_TYPE_I32) { 1305 tcg_out_insn(s, RR, LTR, r1, r1); 1306 } else { 1307 tcg_out_insn(s, RRE, LTGR, r1, r1); 1308 } 1309 *inv_cc = tcg_cond_to_ltr_cond[inv_c]; 1310 return tcg_cond_to_ltr_cond[c]; 1311 } 1312 } 1313 1314 if (!is_unsigned && c2 == (int16_t)c2) { 1315 op = (type == TCG_TYPE_I32 ? RI_CHI : RI_CGHI); 1316 tcg_out_insn_RI(s, op, r1, c2); 1317 goto exit; 1318 } 1319 1320 if (type == TCG_TYPE_I32) { 1321 op = (is_unsigned ? RIL_CLFI : RIL_CFI); 1322 tcg_out_insn_RIL(s, op, r1, c2); 1323 goto exit; 1324 } 1325 1326 /* Should match TCG_CT_CONST_CMP. */ 1327 switch (c) { 1328 case TCG_COND_LT: 1329 case TCG_COND_GE: 1330 case TCG_COND_LE: 1331 case TCG_COND_GT: 1332 tcg_debug_assert(c2 == (int32_t)c2); 1333 op = RIL_CGFI; 1334 break; 1335 case TCG_COND_EQ: 1336 case TCG_COND_NE: 1337 if (c2 == (int32_t)c2) { 1338 op = RIL_CGFI; 1339 break; 1340 } 1341 /* fall through */ 1342 case TCG_COND_LTU: 1343 case TCG_COND_GEU: 1344 case TCG_COND_LEU: 1345 case TCG_COND_GTU: 1346 tcg_debug_assert(c2 == (uint32_t)c2); 1347 op = RIL_CLGFI; 1348 break; 1349 default: 1350 g_assert_not_reached(); 1351 } 1352 tcg_out_insn_RIL(s, op, r1, c2); 1353 } else if (type == TCG_TYPE_I32) { 1354 op = (is_unsigned ? RR_CLR : RR_CR); 1355 tcg_out_insn_RR(s, op, r1, c2); 1356 } else { 1357 op = (is_unsigned ? RRE_CLGR : RRE_CGR); 1358 tcg_out_insn_RRE(s, op, r1, c2); 1359 } 1360 1361 exit: 1362 *inv_cc = tcg_cond_to_s390_cond[inv_c]; 1363 return tcg_cond_to_s390_cond[c]; 1364} 1365 1366static int tgen_cmp(TCGContext *s, TCGType type, TCGCond c, TCGReg r1, 1367 TCGArg c2, bool c2const, bool need_carry) 1368{ 1369 int inv_cc; 1370 return tgen_cmp2(s, type, c, r1, c2, c2const, need_carry, &inv_cc); 1371} 1372 1373static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond, 1374 TCGReg dest, TCGReg c1, TCGArg c2, 1375 bool c2const, bool neg) 1376{ 1377 int cc; 1378 1379 /* With LOC2, we can always emit the minimum 3 insns. */ 1380 if (HAVE_FACILITY(LOAD_ON_COND2)) { 1381 /* Emit: d = 0, d = (cc ? 1 : d). */ 1382 cc = tgen_cmp(s, type, cond, c1, c2, c2const, false); 1383 tcg_out_movi(s, TCG_TYPE_I64, dest, 0); 1384 tcg_out_insn(s, RIEg, LOCGHI, dest, neg ? -1 : 1, cc); 1385 return; 1386 } 1387 1388 switch (cond) { 1389 case TCG_COND_GEU: 1390 case TCG_COND_LTU: 1391 case TCG_COND_LT: 1392 case TCG_COND_GE: 1393 /* Swap operands so that we can use LEU/GTU/GT/LE. */ 1394 if (!c2const) { 1395 TCGReg t = c1; 1396 c1 = c2; 1397 c2 = t; 1398 cond = tcg_swap_cond(cond); 1399 } 1400 break; 1401 default: 1402 break; 1403 } 1404 1405 switch (cond) { 1406 case TCG_COND_NE: 1407 /* X != 0 is X > 0. */ 1408 if (c2const && c2 == 0) { 1409 cond = TCG_COND_GTU; 1410 } else { 1411 break; 1412 } 1413 /* fallthru */ 1414 1415 case TCG_COND_GTU: 1416 case TCG_COND_GT: 1417 /* 1418 * The result of a compare has CC=2 for GT and CC=3 unused. 1419 * ADD LOGICAL WITH CARRY considers (CC & 2) the carry bit. 1420 */ 1421 tgen_cmp(s, type, cond, c1, c2, c2const, true); 1422 tcg_out_movi(s, type, dest, 0); 1423 tcg_out_insn(s, RRE, ALCGR, dest, dest); 1424 if (neg) { 1425 if (type == TCG_TYPE_I32) { 1426 tcg_out_insn(s, RR, LCR, dest, dest); 1427 } else { 1428 tcg_out_insn(s, RRE, LCGR, dest, dest); 1429 } 1430 } 1431 return; 1432 1433 case TCG_COND_EQ: 1434 /* X == 0 is X <= 0. */ 1435 if (c2const && c2 == 0) { 1436 cond = TCG_COND_LEU; 1437 } else { 1438 break; 1439 } 1440 /* fallthru */ 1441 1442 case TCG_COND_LEU: 1443 case TCG_COND_LE: 1444 /* 1445 * As above, but we're looking for borrow, or !carry. 1446 * The second insn computes d - d - borrow, or -1 for true 1447 * and 0 for false. So we must mask to 1 bit afterward. 1448 */ 1449 tgen_cmp(s, type, cond, c1, c2, c2const, true); 1450 tcg_out_insn(s, RRE, SLBGR, dest, dest); 1451 if (!neg) { 1452 tgen_andi(s, type, dest, 1); 1453 } 1454 return; 1455 1456 default: 1457 g_assert_not_reached(); 1458 } 1459 1460 cc = tgen_cmp(s, type, cond, c1, c2, c2const, false); 1461 /* Emit: d = 0, t = 1, d = (cc ? t : d). */ 1462 tcg_out_movi(s, TCG_TYPE_I64, dest, 0); 1463 tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, neg ? -1 : 1); 1464 tcg_out_insn(s, RRFc, LOCGR, dest, TCG_TMP0, cc); 1465} 1466 1467static void tgen_movcond_int(TCGContext *s, TCGType type, TCGReg dest, 1468 TCGArg v3, int v3const, TCGReg v4, 1469 int cc, int inv_cc) 1470{ 1471 TCGReg src; 1472 1473 if (v3const) { 1474 if (dest == v4) { 1475 if (HAVE_FACILITY(LOAD_ON_COND2)) { 1476 /* Emit: if (cc) dest = v3. */ 1477 tcg_out_insn(s, RIEg, LOCGHI, dest, v3, cc); 1478 return; 1479 } 1480 tcg_out_insn(s, RI, LGHI, TCG_TMP0, v3); 1481 src = TCG_TMP0; 1482 } else { 1483 /* LGR+LOCGHI is larger than LGHI+LOCGR. */ 1484 tcg_out_insn(s, RI, LGHI, dest, v3); 1485 cc = inv_cc; 1486 src = v4; 1487 } 1488 } else { 1489 if (HAVE_FACILITY(MISC_INSN_EXT3)) { 1490 /* Emit: dest = cc ? v3 : v4. */ 1491 tcg_out_insn(s, RRFam, SELGR, dest, v3, v4, cc); 1492 return; 1493 } 1494 if (dest == v4) { 1495 src = v3; 1496 } else { 1497 tcg_out_mov(s, type, dest, v3); 1498 cc = inv_cc; 1499 src = v4; 1500 } 1501 } 1502 1503 /* Emit: if (cc) dest = src. */ 1504 tcg_out_insn(s, RRFc, LOCGR, dest, src, cc); 1505} 1506 1507static void tgen_movcond(TCGContext *s, TCGType type, TCGCond c, TCGReg dest, 1508 TCGReg c1, TCGArg c2, int c2const, 1509 TCGArg v3, int v3const, TCGReg v4) 1510{ 1511 int cc, inv_cc; 1512 1513 cc = tgen_cmp2(s, type, c, c1, c2, c2const, false, &inv_cc); 1514 tgen_movcond_int(s, type, dest, v3, v3const, v4, cc, inv_cc); 1515} 1516 1517static void tgen_deposit(TCGContext *s, TCGReg dest, TCGReg src, 1518 int ofs, int len, int z) 1519{ 1520 int lsb = (63 - ofs); 1521 int msb = lsb - (len - 1); 1522 tcg_out_risbg(s, dest, src, msb, lsb, ofs, z); 1523} 1524 1525static void tgen_extract(TCGContext *s, TCGReg dest, TCGReg src, 1526 int ofs, int len) 1527{ 1528 if (ofs == 0) { 1529 switch (len) { 1530 case 8: 1531 tcg_out_ext8u(s, dest, src); 1532 return; 1533 case 16: 1534 tcg_out_ext16u(s, dest, src); 1535 return; 1536 case 32: 1537 tcg_out_ext32u(s, dest, src); 1538 return; 1539 } 1540 } 1541 tcg_out_risbg(s, dest, src, 64 - len, 63, 64 - ofs, 1); 1542} 1543 1544static void tgen_sextract(TCGContext *s, TCGReg dest, TCGReg src, 1545 int ofs, int len) 1546{ 1547 if (ofs == 0) { 1548 switch (len) { 1549 case 8: 1550 tcg_out_ext8s(s, TCG_TYPE_REG, dest, src); 1551 return; 1552 case 16: 1553 tcg_out_ext16s(s, TCG_TYPE_REG, dest, src); 1554 return; 1555 case 32: 1556 tcg_out_ext32s(s, dest, src); 1557 return; 1558 } 1559 } 1560 g_assert_not_reached(); 1561} 1562 1563static void tgen_gotoi(TCGContext *s, int cc, const tcg_insn_unit *dest) 1564{ 1565 ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1; 1566 if (off == (int16_t)off) { 1567 tcg_out_insn(s, RI, BRC, cc, off); 1568 } else if (off == (int32_t)off) { 1569 tcg_out_insn(s, RIL, BRCL, cc, off); 1570 } else { 1571 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest); 1572 tcg_out_insn(s, RR, BCR, cc, TCG_TMP0); 1573 } 1574} 1575 1576static void tgen_branch(TCGContext *s, int cc, TCGLabel *l) 1577{ 1578 if (l->has_value) { 1579 tgen_gotoi(s, cc, l->u.value_ptr); 1580 } else { 1581 tcg_out16(s, RI_BRC | (cc << 4)); 1582 tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, l, 2); 1583 s->code_ptr += 1; 1584 } 1585} 1586 1587static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc, 1588 TCGReg r1, TCGReg r2, TCGLabel *l) 1589{ 1590 tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2); 1591 /* Format RIE-b */ 1592 tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2); 1593 tcg_out16(s, 0); 1594 tcg_out16(s, cc << 12 | (opc & 0xff)); 1595} 1596 1597static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc, 1598 TCGReg r1, int i2, TCGLabel *l) 1599{ 1600 tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2); 1601 /* Format RIE-c */ 1602 tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc); 1603 tcg_out16(s, 0); 1604 tcg_out16(s, (i2 << 8) | (opc & 0xff)); 1605} 1606 1607static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c, 1608 TCGReg r1, TCGArg c2, int c2const, TCGLabel *l) 1609{ 1610 int cc; 1611 1612 if (!is_tst_cond(c)) { 1613 bool is_unsigned = is_unsigned_cond(c); 1614 bool in_range; 1615 S390Opcode opc; 1616 1617 cc = tcg_cond_to_s390_cond[c]; 1618 1619 if (!c2const) { 1620 opc = (type == TCG_TYPE_I32 1621 ? (is_unsigned ? RIEb_CLRJ : RIEb_CRJ) 1622 : (is_unsigned ? RIEb_CLGRJ : RIEb_CGRJ)); 1623 tgen_compare_branch(s, opc, cc, r1, c2, l); 1624 return; 1625 } 1626 1627 /* 1628 * COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field. 1629 * If the immediate we've been given does not fit that range, we'll 1630 * fall back to separate compare and branch instructions using the 1631 * larger comparison range afforded by COMPARE IMMEDIATE. 1632 */ 1633 if (type == TCG_TYPE_I32) { 1634 if (is_unsigned) { 1635 opc = RIEc_CLIJ; 1636 in_range = (uint32_t)c2 == (uint8_t)c2; 1637 } else { 1638 opc = RIEc_CIJ; 1639 in_range = (int32_t)c2 == (int8_t)c2; 1640 } 1641 } else { 1642 if (is_unsigned) { 1643 opc = RIEc_CLGIJ; 1644 in_range = (uint64_t)c2 == (uint8_t)c2; 1645 } else { 1646 opc = RIEc_CGIJ; 1647 in_range = (int64_t)c2 == (int8_t)c2; 1648 } 1649 } 1650 if (in_range) { 1651 tgen_compare_imm_branch(s, opc, cc, r1, c2, l); 1652 return; 1653 } 1654 } 1655 1656 cc = tgen_cmp(s, type, c, r1, c2, c2const, false); 1657 tgen_branch(s, cc, l); 1658} 1659 1660static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *dest) 1661{ 1662 ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1; 1663 if (off == (int32_t)off) { 1664 tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off); 1665 } else { 1666 tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest); 1667 tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0); 1668 } 1669} 1670 1671static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest, 1672 const TCGHelperInfo *info) 1673{ 1674 tcg_out_call_int(s, dest); 1675} 1676 1677typedef struct { 1678 TCGReg base; 1679 TCGReg index; 1680 int disp; 1681 TCGAtomAlign aa; 1682} HostAddress; 1683 1684bool tcg_target_has_memory_bswap(MemOp memop) 1685{ 1686 TCGAtomAlign aa; 1687 1688 if ((memop & MO_SIZE) <= MO_64) { 1689 return true; 1690 } 1691 1692 /* 1693 * Reject 16-byte memop with 16-byte atomicity, 1694 * but do allow a pair of 64-bit operations. 1695 */ 1696 aa = atom_and_align_for_opc(tcg_ctx, memop, MO_ATOM_IFALIGN, true); 1697 return aa.atom <= MO_64; 1698} 1699 1700static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg data, 1701 HostAddress h) 1702{ 1703 switch (opc & (MO_SSIZE | MO_BSWAP)) { 1704 case MO_UB: 1705 tcg_out_insn(s, RXY, LLGC, data, h.base, h.index, h.disp); 1706 break; 1707 case MO_SB: 1708 tcg_out_insn(s, RXY, LGB, data, h.base, h.index, h.disp); 1709 break; 1710 1711 case MO_UW | MO_BSWAP: 1712 /* swapped unsigned halfword load with upper bits zeroed */ 1713 tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp); 1714 tcg_out_ext16u(s, data, data); 1715 break; 1716 case MO_UW: 1717 tcg_out_insn(s, RXY, LLGH, data, h.base, h.index, h.disp); 1718 break; 1719 1720 case MO_SW | MO_BSWAP: 1721 /* swapped sign-extended halfword load */ 1722 tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp); 1723 tcg_out_ext16s(s, TCG_TYPE_REG, data, data); 1724 break; 1725 case MO_SW: 1726 tcg_out_insn(s, RXY, LGH, data, h.base, h.index, h.disp); 1727 break; 1728 1729 case MO_UL | MO_BSWAP: 1730 /* swapped unsigned int load with upper bits zeroed */ 1731 tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp); 1732 tcg_out_ext32u(s, data, data); 1733 break; 1734 case MO_UL: 1735 tcg_out_insn(s, RXY, LLGF, data, h.base, h.index, h.disp); 1736 break; 1737 1738 case MO_SL | MO_BSWAP: 1739 /* swapped sign-extended int load */ 1740 tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp); 1741 tcg_out_ext32s(s, data, data); 1742 break; 1743 case MO_SL: 1744 tcg_out_insn(s, RXY, LGF, data, h.base, h.index, h.disp); 1745 break; 1746 1747 case MO_UQ | MO_BSWAP: 1748 tcg_out_insn(s, RXY, LRVG, data, h.base, h.index, h.disp); 1749 break; 1750 case MO_UQ: 1751 tcg_out_insn(s, RXY, LG, data, h.base, h.index, h.disp); 1752 break; 1753 1754 default: 1755 g_assert_not_reached(); 1756 } 1757} 1758 1759static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg data, 1760 HostAddress h) 1761{ 1762 switch (opc & (MO_SIZE | MO_BSWAP)) { 1763 case MO_UB: 1764 if (h.disp >= 0 && h.disp < 0x1000) { 1765 tcg_out_insn(s, RX, STC, data, h.base, h.index, h.disp); 1766 } else { 1767 tcg_out_insn(s, RXY, STCY, data, h.base, h.index, h.disp); 1768 } 1769 break; 1770 1771 case MO_UW | MO_BSWAP: 1772 tcg_out_insn(s, RXY, STRVH, data, h.base, h.index, h.disp); 1773 break; 1774 case MO_UW: 1775 if (h.disp >= 0 && h.disp < 0x1000) { 1776 tcg_out_insn(s, RX, STH, data, h.base, h.index, h.disp); 1777 } else { 1778 tcg_out_insn(s, RXY, STHY, data, h.base, h.index, h.disp); 1779 } 1780 break; 1781 1782 case MO_UL | MO_BSWAP: 1783 tcg_out_insn(s, RXY, STRV, data, h.base, h.index, h.disp); 1784 break; 1785 case MO_UL: 1786 if (h.disp >= 0 && h.disp < 0x1000) { 1787 tcg_out_insn(s, RX, ST, data, h.base, h.index, h.disp); 1788 } else { 1789 tcg_out_insn(s, RXY, STY, data, h.base, h.index, h.disp); 1790 } 1791 break; 1792 1793 case MO_UQ | MO_BSWAP: 1794 tcg_out_insn(s, RXY, STRVG, data, h.base, h.index, h.disp); 1795 break; 1796 case MO_UQ: 1797 tcg_out_insn(s, RXY, STG, data, h.base, h.index, h.disp); 1798 break; 1799 1800 default: 1801 g_assert_not_reached(); 1802 } 1803} 1804 1805static const TCGLdstHelperParam ldst_helper_param = { 1806 .ntmp = 1, .tmp = { TCG_TMP0 } 1807}; 1808 1809static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb) 1810{ 1811 MemOp opc = get_memop(lb->oi); 1812 1813 if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL, 1814 (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) { 1815 return false; 1816 } 1817 1818 tcg_out_ld_helper_args(s, lb, &ldst_helper_param); 1819 tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]); 1820 tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param); 1821 1822 tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr); 1823 return true; 1824} 1825 1826static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb) 1827{ 1828 MemOp opc = get_memop(lb->oi); 1829 1830 if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL, 1831 (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) { 1832 return false; 1833 } 1834 1835 tcg_out_st_helper_args(s, lb, &ldst_helper_param); 1836 tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE]); 1837 1838 tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr); 1839 return true; 1840} 1841 1842/* We're expecting to use a 20-bit negative offset on the tlb memory ops. */ 1843#define MIN_TLB_MASK_TABLE_OFS -(1 << 19) 1844 1845/* 1846 * For system-mode, perform the TLB load and compare. 1847 * For user-mode, perform any required alignment tests. 1848 * In both cases, return a TCGLabelQemuLdst structure if the slow path 1849 * is required and fill in @h with the host address for the fast path. 1850 */ 1851static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h, 1852 TCGReg addr_reg, MemOpIdx oi, 1853 bool is_ld) 1854{ 1855 TCGType addr_type = s->addr_type; 1856 TCGLabelQemuLdst *ldst = NULL; 1857 MemOp opc = get_memop(oi); 1858 MemOp s_bits = opc & MO_SIZE; 1859 unsigned a_mask; 1860 1861 h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, s_bits == MO_128); 1862 a_mask = (1 << h->aa.align) - 1; 1863 1864 if (tcg_use_softmmu) { 1865 unsigned s_mask = (1 << s_bits) - 1; 1866 int mem_index = get_mmuidx(oi); 1867 int fast_off = tlb_mask_table_ofs(s, mem_index); 1868 int mask_off = fast_off + offsetof(CPUTLBDescFast, mask); 1869 int table_off = fast_off + offsetof(CPUTLBDescFast, table); 1870 int ofs, a_off; 1871 uint64_t tlb_mask; 1872 1873 ldst = new_ldst_label(s); 1874 ldst->is_ld = is_ld; 1875 ldst->oi = oi; 1876 ldst->addr_reg = addr_reg; 1877 1878 tcg_out_sh64(s, RSY_SRLG, TCG_TMP0, addr_reg, TCG_REG_NONE, 1879 s->page_bits - CPU_TLB_ENTRY_BITS); 1880 1881 tcg_out_insn(s, RXY, NG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, mask_off); 1882 tcg_out_insn(s, RXY, AG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, table_off); 1883 1884 /* 1885 * For aligned accesses, we check the first byte and include the 1886 * alignment bits within the address. For unaligned access, we 1887 * check that we don't cross pages using the address of the last 1888 * byte of the access. 1889 */ 1890 a_off = (a_mask >= s_mask ? 0 : s_mask - a_mask); 1891 tlb_mask = (uint64_t)s->page_mask | a_mask; 1892 if (a_off == 0) { 1893 tgen_andi_risbg(s, TCG_REG_R0, addr_reg, tlb_mask); 1894 } else { 1895 tcg_out_insn(s, RX, LA, TCG_REG_R0, addr_reg, TCG_REG_NONE, a_off); 1896 tgen_andi(s, addr_type, TCG_REG_R0, tlb_mask); 1897 } 1898 1899 if (is_ld) { 1900 ofs = offsetof(CPUTLBEntry, addr_read); 1901 } else { 1902 ofs = offsetof(CPUTLBEntry, addr_write); 1903 } 1904 if (addr_type == TCG_TYPE_I32) { 1905 ofs += HOST_BIG_ENDIAN * 4; 1906 tcg_out_insn(s, RX, C, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs); 1907 } else { 1908 tcg_out_insn(s, RXY, CG, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs); 1909 } 1910 1911 tcg_out16(s, RI_BRC | (S390_CC_NE << 4)); 1912 ldst->label_ptr[0] = s->code_ptr++; 1913 1914 h->index = TCG_TMP0; 1915 tcg_out_insn(s, RXY, LG, h->index, TCG_TMP0, TCG_REG_NONE, 1916 offsetof(CPUTLBEntry, addend)); 1917 1918 if (addr_type == TCG_TYPE_I32) { 1919 tcg_out_insn(s, RRE, ALGFR, h->index, addr_reg); 1920 h->base = TCG_REG_NONE; 1921 } else { 1922 h->base = addr_reg; 1923 } 1924 h->disp = 0; 1925 } else { 1926 if (a_mask) { 1927 ldst = new_ldst_label(s); 1928 ldst->is_ld = is_ld; 1929 ldst->oi = oi; 1930 ldst->addr_reg = addr_reg; 1931 1932 tcg_debug_assert(a_mask <= 0xffff); 1933 tcg_out_insn(s, RI, TMLL, addr_reg, a_mask); 1934 1935 tcg_out16(s, RI_BRC | (S390_TM_NE << 4)); 1936 ldst->label_ptr[0] = s->code_ptr++; 1937 } 1938 1939 h->base = addr_reg; 1940 if (addr_type == TCG_TYPE_I32) { 1941 tcg_out_ext32u(s, TCG_TMP0, addr_reg); 1942 h->base = TCG_TMP0; 1943 } 1944 if (guest_base < 0x80000) { 1945 h->index = TCG_REG_NONE; 1946 h->disp = guest_base; 1947 } else { 1948 h->index = TCG_GUEST_BASE_REG; 1949 h->disp = 0; 1950 } 1951 } 1952 1953 return ldst; 1954} 1955 1956static void tcg_out_qemu_ld(TCGContext* s, TCGReg data_reg, TCGReg addr_reg, 1957 MemOpIdx oi, TCGType data_type) 1958{ 1959 TCGLabelQemuLdst *ldst; 1960 HostAddress h; 1961 1962 ldst = prepare_host_addr(s, &h, addr_reg, oi, true); 1963 tcg_out_qemu_ld_direct(s, get_memop(oi), data_reg, h); 1964 1965 if (ldst) { 1966 ldst->type = data_type; 1967 ldst->datalo_reg = data_reg; 1968 ldst->raddr = tcg_splitwx_to_rx(s->code_ptr); 1969 } 1970} 1971 1972static void tcg_out_qemu_st(TCGContext* s, TCGReg data_reg, TCGReg addr_reg, 1973 MemOpIdx oi, TCGType data_type) 1974{ 1975 TCGLabelQemuLdst *ldst; 1976 HostAddress h; 1977 1978 ldst = prepare_host_addr(s, &h, addr_reg, oi, false); 1979 tcg_out_qemu_st_direct(s, get_memop(oi), data_reg, h); 1980 1981 if (ldst) { 1982 ldst->type = data_type; 1983 ldst->datalo_reg = data_reg; 1984 ldst->raddr = tcg_splitwx_to_rx(s->code_ptr); 1985 } 1986} 1987 1988static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi, 1989 TCGReg addr_reg, MemOpIdx oi, bool is_ld) 1990{ 1991 TCGLabel *l1 = NULL, *l2 = NULL; 1992 TCGLabelQemuLdst *ldst; 1993 HostAddress h; 1994 bool need_bswap; 1995 bool use_pair; 1996 S390Opcode insn; 1997 1998 ldst = prepare_host_addr(s, &h, addr_reg, oi, is_ld); 1999 2000 use_pair = h.aa.atom < MO_128; 2001 need_bswap = get_memop(oi) & MO_BSWAP; 2002 2003 if (!use_pair) { 2004 /* 2005 * Atomicity requires we use LPQ. If we've already checked for 2006 * 16-byte alignment, that's all we need. If we arrive with 2007 * lesser alignment, we have determined that less than 16-byte 2008 * alignment can be satisfied with two 8-byte loads. 2009 */ 2010 if (h.aa.align < MO_128) { 2011 use_pair = true; 2012 l1 = gen_new_label(); 2013 l2 = gen_new_label(); 2014 2015 tcg_out_insn(s, RI, TMLL, addr_reg, 15); 2016 tgen_branch(s, S390_TM_NE, l1); 2017 } 2018 2019 tcg_debug_assert(!need_bswap); 2020 tcg_debug_assert(datalo & 1); 2021 tcg_debug_assert(datahi == datalo - 1); 2022 insn = is_ld ? RXY_LPQ : RXY_STPQ; 2023 tcg_out_insn_RXY(s, insn, datahi, h.base, h.index, h.disp); 2024 2025 if (use_pair) { 2026 tgen_branch(s, S390_CC_ALWAYS, l2); 2027 tcg_out_label(s, l1); 2028 } 2029 } 2030 if (use_pair) { 2031 TCGReg d1, d2; 2032 2033 if (need_bswap) { 2034 d1 = datalo, d2 = datahi; 2035 insn = is_ld ? RXY_LRVG : RXY_STRVG; 2036 } else { 2037 d1 = datahi, d2 = datalo; 2038 insn = is_ld ? RXY_LG : RXY_STG; 2039 } 2040 2041 if (h.base == d1 || h.index == d1) { 2042 tcg_out_insn(s, RXY, LAY, TCG_TMP0, h.base, h.index, h.disp); 2043 h.base = TCG_TMP0; 2044 h.index = TCG_REG_NONE; 2045 h.disp = 0; 2046 } 2047 tcg_out_insn_RXY(s, insn, d1, h.base, h.index, h.disp); 2048 tcg_out_insn_RXY(s, insn, d2, h.base, h.index, h.disp + 8); 2049 } 2050 if (l2) { 2051 tcg_out_label(s, l2); 2052 } 2053 2054 if (ldst) { 2055 ldst->type = TCG_TYPE_I128; 2056 ldst->datalo_reg = datalo; 2057 ldst->datahi_reg = datahi; 2058 ldst->raddr = tcg_splitwx_to_rx(s->code_ptr); 2059 } 2060} 2061 2062static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0) 2063{ 2064 /* Reuse the zeroing that exists for goto_ptr. */ 2065 if (a0 == 0) { 2066 tgen_gotoi(s, S390_CC_ALWAYS, tcg_code_gen_epilogue); 2067 } else { 2068 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, a0); 2069 tgen_gotoi(s, S390_CC_ALWAYS, tb_ret_addr); 2070 } 2071} 2072 2073static void tcg_out_goto_tb(TCGContext *s, int which) 2074{ 2075 /* 2076 * Branch displacement must be aligned for atomic patching; 2077 * see if we need to add extra nop before branch 2078 */ 2079 if (!QEMU_PTR_IS_ALIGNED(s->code_ptr + 1, 4)) { 2080 tcg_out16(s, NOP); 2081 } 2082 tcg_out16(s, RIL_BRCL | (S390_CC_ALWAYS << 4)); 2083 set_jmp_insn_offset(s, which); 2084 s->code_ptr += 2; 2085 set_jmp_reset_offset(s, which); 2086} 2087 2088void tb_target_set_jmp_target(const TranslationBlock *tb, int n, 2089 uintptr_t jmp_rx, uintptr_t jmp_rw) 2090{ 2091 if (!HAVE_FACILITY(GEN_INST_EXT)) { 2092 return; 2093 } 2094 /* patch the branch destination */ 2095 uintptr_t addr = tb->jmp_target_addr[n]; 2096 intptr_t disp = addr - (jmp_rx - 2); 2097 qatomic_set((int32_t *)jmp_rw, disp / 2); 2098 /* no need to flush icache explicitly */ 2099} 2100 2101 2102static void tgen_add(TCGContext *s, TCGType type, 2103 TCGReg a0, TCGReg a1, TCGReg a2) 2104{ 2105 if (a0 != a1) { 2106 tcg_out_insn(s, RX, LA, a0, a1, a2, 0); 2107 } else if (type == TCG_TYPE_I32) { 2108 tcg_out_insn(s, RR, AR, a0, a2); 2109 } else { 2110 tcg_out_insn(s, RRE, AGR, a0, a2); 2111 } 2112} 2113 2114static void tgen_addi(TCGContext *s, TCGType type, 2115 TCGReg a0, TCGReg a1, tcg_target_long a2) 2116{ 2117 if (a0 == a1) { 2118 if (type == TCG_TYPE_I32) { 2119 if (a2 == (int16_t)a2) { 2120 tcg_out_insn(s, RI, AHI, a0, a2); 2121 } else { 2122 tcg_out_insn(s, RIL, AFI, a0, a2); 2123 } 2124 return; 2125 } 2126 if (a2 == (int16_t)a2) { 2127 tcg_out_insn(s, RI, AGHI, a0, a2); 2128 return; 2129 } 2130 if (a2 == (int32_t)a2) { 2131 tcg_out_insn(s, RIL, AGFI, a0, a2); 2132 return; 2133 } 2134 if (a2 == (uint32_t)a2) { 2135 tcg_out_insn(s, RIL, ALGFI, a0, a2); 2136 return; 2137 } 2138 if (-a2 == (uint32_t)-a2) { 2139 tcg_out_insn(s, RIL, SLGFI, a0, -a2); 2140 return; 2141 } 2142 } 2143 tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2); 2144} 2145 2146static const TCGOutOpBinary outop_add = { 2147 .base.static_constraint = C_O1_I2(r, r, ri), 2148 .out_rrr = tgen_add, 2149 .out_rri = tgen_addi, 2150}; 2151 2152static void tgen_and(TCGContext *s, TCGType type, 2153 TCGReg a0, TCGReg a1, TCGReg a2) 2154{ 2155 if (type != TCG_TYPE_I32) { 2156 tcg_out_insn(s, RRFa, NGRK, a0, a1, a2); 2157 } else if (a0 == a1) { 2158 tcg_out_insn(s, RR, NR, a0, a2); 2159 } else { 2160 tcg_out_insn(s, RRFa, NRK, a0, a1, a2); 2161 } 2162} 2163 2164static void tgen_andi_3(TCGContext *s, TCGType type, 2165 TCGReg a0, TCGReg a1, tcg_target_long a2) 2166{ 2167 tcg_out_mov(s, type, a0, a1); 2168 tgen_andi(s, type, a0, a2); 2169} 2170 2171static const TCGOutOpBinary outop_and = { 2172 .base.static_constraint = C_O1_I2(r, r, rNKR), 2173 .out_rrr = tgen_and, 2174 .out_rri = tgen_andi_3, 2175}; 2176 2177static void tgen_andc(TCGContext *s, TCGType type, 2178 TCGReg a0, TCGReg a1, TCGReg a2) 2179{ 2180 if (type == TCG_TYPE_I32) { 2181 tcg_out_insn(s, RRFa, NCRK, a0, a1, a2); 2182 } else { 2183 tcg_out_insn(s, RRFa, NCGRK, a0, a1, a2); 2184 } 2185} 2186 2187static TCGConstraintSetIndex cset_misc3_rrr(TCGType type, unsigned flags) 2188{ 2189 return HAVE_FACILITY(MISC_INSN_EXT3) ? C_O1_I2(r, r, r) : C_NotImplemented; 2190} 2191 2192static const TCGOutOpBinary outop_andc = { 2193 .base.static_constraint = C_Dynamic, 2194 .base.dynamic_constraint = cset_misc3_rrr, 2195 .out_rrr = tgen_andc, 2196}; 2197 2198static void tgen_clz_int(TCGContext *s, TCGReg dest, TCGReg a1, 2199 TCGArg a2, int a2const) 2200{ 2201 /* 2202 * Since this sets both R and R+1, we have no choice but to store the 2203 * result into R0, allowing R1 == TCG_TMP0 to be clobbered as well. 2204 */ 2205 QEMU_BUILD_BUG_ON(TCG_TMP0 != TCG_REG_R1); 2206 tcg_out_insn(s, RRE, FLOGR, TCG_REG_R0, a1); 2207 2208 if (a2const && a2 == 64) { 2209 tcg_out_mov(s, TCG_TYPE_I64, dest, TCG_REG_R0); 2210 return; 2211 } 2212 2213 /* 2214 * Conditions from FLOGR are: 2215 * 2 -> one bit found 2216 * 8 -> no one bit found 2217 */ 2218 tgen_movcond_int(s, TCG_TYPE_I64, dest, a2, a2const, TCG_REG_R0, 8, 2); 2219} 2220 2221static void tgen_clz(TCGContext *s, TCGType type, 2222 TCGReg a0, TCGReg a1, TCGReg a2) 2223{ 2224 tgen_clz_int(s, a0, a1, a2, false); 2225} 2226 2227static void tgen_clzi(TCGContext *s, TCGType type, 2228 TCGReg a0, TCGReg a1, tcg_target_long a2) 2229{ 2230 tgen_clz_int(s, a0, a1, a2, true); 2231} 2232 2233static TCGConstraintSetIndex cset_clz(TCGType type, unsigned flags) 2234{ 2235 return type == TCG_TYPE_I64 ? C_O1_I2(r, r, rI) : C_NotImplemented; 2236} 2237 2238static const TCGOutOpBinary outop_clz = { 2239 .base.static_constraint = C_Dynamic, 2240 .base.dynamic_constraint = cset_clz, 2241 .out_rrr = tgen_clz, 2242 .out_rri = tgen_clzi, 2243}; 2244 2245static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg dest, TCGReg src) 2246{ 2247 /* With MIE3, and bit 0 of m4 set, we get the complete result. */ 2248 if (HAVE_FACILITY(MISC_INSN_EXT3)) { 2249 if (type == TCG_TYPE_I32) { 2250 tcg_out_ext32u(s, dest, src); 2251 src = dest; 2252 } 2253 tcg_out_insn(s, RRFc, POPCNT, dest, src, 8); 2254 return; 2255 } 2256 2257 /* Without MIE3, each byte gets the count of bits for the byte. */ 2258 tcg_out_insn(s, RRFc, POPCNT, dest, src, 0); 2259 2260 /* Multiply to sum each byte at the top of the word. */ 2261 if (type == TCG_TYPE_I32) { 2262 tcg_out_insn(s, RIL, MSFI, dest, 0x01010101); 2263 tcg_out_sh32(s, RS_SRL, dest, TCG_REG_NONE, 24); 2264 } else { 2265 tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 0x0101010101010101ull); 2266 tcg_out_insn(s, RRE, MSGR, dest, TCG_TMP0); 2267 tcg_out_sh64(s, RSY_SRLG, dest, dest, TCG_REG_NONE, 56); 2268 } 2269} 2270 2271static const TCGOutOpUnary outop_ctpop = { 2272 .base.static_constraint = C_O1_I1(r, r), 2273 .out_rr = tgen_ctpop, 2274}; 2275 2276static const TCGOutOpBinary outop_ctz = { 2277 .base.static_constraint = C_NotImplemented, 2278}; 2279 2280static const TCGOutOpBinary outop_divs = { 2281 .base.static_constraint = C_NotImplemented, 2282}; 2283 2284static void tgen_divs2(TCGContext *s, TCGType type, 2285 TCGReg a0, TCGReg a1, TCGReg a4) 2286{ 2287 tcg_debug_assert((a1 & 1) == 0); 2288 tcg_debug_assert(a0 == a1 + 1); 2289 if (type == TCG_TYPE_I32) { 2290 tcg_out_insn(s, RR, DR, a1, a4); 2291 } else { 2292 /* 2293 * TODO: Move the sign-extend of the numerator from a2 into a3 2294 * into the tcg backend, instead of in early expansion. It is 2295 * required for 32-bit DR, but not 64-bit DSGR. 2296 */ 2297 tcg_out_insn(s, RRE, DSGR, a1, a4); 2298 } 2299} 2300 2301static const TCGOutOpDivRem outop_divs2 = { 2302 .base.static_constraint = C_O2_I3(o, m, 0, 1, r), 2303 .out_rr01r = tgen_divs2, 2304}; 2305 2306static const TCGOutOpBinary outop_divu = { 2307 .base.static_constraint = C_NotImplemented, 2308}; 2309 2310static void tgen_divu2(TCGContext *s, TCGType type, 2311 TCGReg a0, TCGReg a1, TCGReg a4) 2312{ 2313 tcg_debug_assert((a1 & 1) == 0); 2314 tcg_debug_assert(a0 == a1 + 1); 2315 if (type == TCG_TYPE_I32) { 2316 tcg_out_insn(s, RRE, DLR, a1, a4); 2317 } else { 2318 tcg_out_insn(s, RRE, DLGR, a1, a4); 2319 } 2320} 2321 2322static const TCGOutOpDivRem outop_divu2 = { 2323 .base.static_constraint = C_O2_I3(o, m, 0, 1, r), 2324 .out_rr01r = tgen_divu2, 2325}; 2326 2327static void tgen_eqv(TCGContext *s, TCGType type, 2328 TCGReg a0, TCGReg a1, TCGReg a2) 2329{ 2330 if (type == TCG_TYPE_I32) { 2331 tcg_out_insn(s, RRFa, NXRK, a0, a1, a2); 2332 } else { 2333 tcg_out_insn(s, RRFa, NXGRK, a0, a1, a2); 2334 } 2335} 2336 2337static const TCGOutOpBinary outop_eqv = { 2338 .base.static_constraint = C_Dynamic, 2339 .base.dynamic_constraint = cset_misc3_rrr, 2340 .out_rrr = tgen_eqv, 2341}; 2342 2343static void tgen_mul(TCGContext *s, TCGType type, 2344 TCGReg a0, TCGReg a1, TCGReg a2) 2345{ 2346 if (type == TCG_TYPE_I32) { 2347 if (a0 == a1) { 2348 tcg_out_insn(s, RRE, MSR, a0, a2); 2349 } else { 2350 tcg_out_insn(s, RRFa, MSRKC, a0, a1, a2); 2351 } 2352 } else { 2353 if (a0 == a1) { 2354 tcg_out_insn(s, RRE, MSGR, a0, a2); 2355 } else { 2356 tcg_out_insn(s, RRFa, MSGRKC, a0, a1, a2); 2357 } 2358 } 2359} 2360 2361static void tgen_muli(TCGContext *s, TCGType type, 2362 TCGReg a0, TCGReg a1, tcg_target_long a2) 2363{ 2364 tcg_out_mov(s, type, a0, a1); 2365 if (type == TCG_TYPE_I32) { 2366 if (a2 == (int16_t)a2) { 2367 tcg_out_insn(s, RI, MHI, a0, a2); 2368 } else { 2369 tcg_out_insn(s, RIL, MSFI, a0, a2); 2370 } 2371 } else { 2372 if (a2 == (int16_t)a2) { 2373 tcg_out_insn(s, RI, MGHI, a0, a2); 2374 } else { 2375 tcg_out_insn(s, RIL, MSGFI, a0, a2); 2376 } 2377 } 2378} 2379 2380static TCGConstraintSetIndex cset_mul(TCGType type, unsigned flags) 2381{ 2382 return (HAVE_FACILITY(MISC_INSN_EXT2) 2383 ? C_O1_I2(r, r, rJ) 2384 : C_O1_I2(r, 0, rJ)); 2385} 2386 2387static const TCGOutOpBinary outop_mul = { 2388 .base.static_constraint = C_Dynamic, 2389 .base.dynamic_constraint = cset_mul, 2390 .out_rrr = tgen_mul, 2391 .out_rri = tgen_muli, 2392}; 2393 2394static const TCGOutOpBinary outop_mulsh = { 2395 .base.static_constraint = C_NotImplemented, 2396}; 2397 2398static const TCGOutOpBinary outop_muluh = { 2399 .base.static_constraint = C_NotImplemented, 2400}; 2401 2402static void tgen_nand(TCGContext *s, TCGType type, 2403 TCGReg a0, TCGReg a1, TCGReg a2) 2404{ 2405 if (type == TCG_TYPE_I32) { 2406 tcg_out_insn(s, RRFa, NNRK, a0, a1, a2); 2407 } else { 2408 tcg_out_insn(s, RRFa, NNGRK, a0, a1, a2); 2409 } 2410} 2411 2412static const TCGOutOpBinary outop_nand = { 2413 .base.static_constraint = C_Dynamic, 2414 .base.dynamic_constraint = cset_misc3_rrr, 2415 .out_rrr = tgen_nand, 2416}; 2417 2418static void tgen_nor(TCGContext *s, TCGType type, 2419 TCGReg a0, TCGReg a1, TCGReg a2) 2420{ 2421 if (type == TCG_TYPE_I32) { 2422 tcg_out_insn(s, RRFa, NORK, a0, a1, a2); 2423 } else { 2424 tcg_out_insn(s, RRFa, NOGRK, a0, a1, a2); 2425 } 2426} 2427 2428static const TCGOutOpBinary outop_nor = { 2429 .base.static_constraint = C_Dynamic, 2430 .base.dynamic_constraint = cset_misc3_rrr, 2431 .out_rrr = tgen_nor, 2432}; 2433 2434static void tgen_or(TCGContext *s, TCGType type, 2435 TCGReg a0, TCGReg a1, TCGReg a2) 2436{ 2437 if (type != TCG_TYPE_I32) { 2438 tcg_out_insn(s, RRFa, OGRK, a0, a1, a2); 2439 } else if (a0 == a1) { 2440 tcg_out_insn(s, RR, OR, a0, a2); 2441 } else { 2442 tcg_out_insn(s, RRFa, ORK, a0, a1, a2); 2443 } 2444} 2445 2446static void tgen_ori_3(TCGContext *s, TCGType type, 2447 TCGReg a0, TCGReg a1, tcg_target_long a2) 2448{ 2449 tcg_out_mov(s, type, a0, a1); 2450 tgen_ori(s, a0, type == TCG_TYPE_I32 ? (uint32_t)a2 : a2); 2451} 2452 2453static const TCGOutOpBinary outop_or = { 2454 .base.static_constraint = C_O1_I2(r, r, rK), 2455 .out_rrr = tgen_or, 2456 .out_rri = tgen_ori_3, 2457}; 2458 2459static void tgen_orc(TCGContext *s, TCGType type, 2460 TCGReg a0, TCGReg a1, TCGReg a2) 2461{ 2462 if (type == TCG_TYPE_I32) { 2463 tcg_out_insn(s, RRFa, OCRK, a0, a1, a2); 2464 } else { 2465 tcg_out_insn(s, RRFa, OCGRK, a0, a1, a2); 2466 } 2467} 2468 2469static const TCGOutOpBinary outop_orc = { 2470 .base.static_constraint = C_Dynamic, 2471 .base.dynamic_constraint = cset_misc3_rrr, 2472 .out_rrr = tgen_orc, 2473}; 2474 2475static const TCGOutOpBinary outop_rems = { 2476 .base.static_constraint = C_NotImplemented, 2477}; 2478 2479static const TCGOutOpBinary outop_remu = { 2480 .base.static_constraint = C_NotImplemented, 2481}; 2482 2483static void tgen_rotl_int(TCGContext *s, TCGType type, TCGReg dst, 2484 TCGReg src, TCGReg v, tcg_target_long i) 2485{ 2486 S390Opcode insn = type == TCG_TYPE_I32 ? RSY_RLL : RSY_RLLG; 2487 tcg_out_sh64(s, insn, dst, src, v, i); 2488} 2489 2490static void tgen_rotl(TCGContext *s, TCGType type, 2491 TCGReg a0, TCGReg a1, TCGReg a2) 2492{ 2493 tgen_rotl_int(s, type, a0, a1, a2, 0); 2494} 2495 2496static void tgen_rotli(TCGContext *s, TCGType type, 2497 TCGReg a0, TCGReg a1, tcg_target_long a2) 2498{ 2499 tgen_rotl_int(s, type, a0, a1, TCG_REG_NONE, a2); 2500} 2501 2502static const TCGOutOpBinary outop_rotl = { 2503 .base.static_constraint = C_O1_I2(r, r, ri), 2504 .out_rrr = tgen_rotl, 2505 .out_rri = tgen_rotli, 2506}; 2507 2508static const TCGOutOpBinary outop_rotr = { 2509 .base.static_constraint = C_NotImplemented, 2510}; 2511 2512static void tgen_sar_int(TCGContext *s, TCGType type, TCGReg dst, 2513 TCGReg src, TCGReg v, tcg_target_long i) 2514{ 2515 if (type != TCG_TYPE_I32) { 2516 tcg_out_sh64(s, RSY_SRAG, dst, src, v, i); 2517 } else if (dst == src) { 2518 tcg_out_sh32(s, RS_SRA, dst, v, i); 2519 } else { 2520 tcg_out_sh64(s, RSY_SRAK, dst, src, v, i); 2521 } 2522} 2523 2524static void tgen_sar(TCGContext *s, TCGType type, 2525 TCGReg a0, TCGReg a1, TCGReg a2) 2526{ 2527 tgen_sar_int(s, type, a0, a1, a2, 0); 2528} 2529 2530static void tgen_sari(TCGContext *s, TCGType type, 2531 TCGReg a0, TCGReg a1, tcg_target_long a2) 2532{ 2533 tgen_sar_int(s, type, a0, a1, TCG_REG_NONE, a2); 2534} 2535 2536static const TCGOutOpBinary outop_sar = { 2537 .base.static_constraint = C_O1_I2(r, r, ri), 2538 .out_rrr = tgen_sar, 2539 .out_rri = tgen_sari, 2540}; 2541 2542static void tgen_shl_int(TCGContext *s, TCGType type, TCGReg dst, 2543 TCGReg src, TCGReg v, tcg_target_long i) 2544{ 2545 if (type != TCG_TYPE_I32) { 2546 tcg_out_sh64(s, RSY_SLLG, dst, src, v, i); 2547 } else if (dst == src) { 2548 tcg_out_sh32(s, RS_SLL, dst, v, i); 2549 } else { 2550 tcg_out_sh64(s, RSY_SLLK, dst, src, v, i); 2551 } 2552} 2553 2554static void tgen_shl(TCGContext *s, TCGType type, 2555 TCGReg a0, TCGReg a1, TCGReg a2) 2556{ 2557 tgen_shl_int(s, type, a0, a1, a2, 0); 2558} 2559 2560static void tgen_shli(TCGContext *s, TCGType type, 2561 TCGReg a0, TCGReg a1, tcg_target_long a2) 2562{ 2563 tgen_shl_int(s, type, a0, a1, TCG_REG_NONE, a2); 2564} 2565 2566static const TCGOutOpBinary outop_shl = { 2567 .base.static_constraint = C_O1_I2(r, r, ri), 2568 .out_rrr = tgen_shl, 2569 .out_rri = tgen_shli, 2570}; 2571 2572static void tgen_shr_int(TCGContext *s, TCGType type, TCGReg dst, 2573 TCGReg src, TCGReg v, tcg_target_long i) 2574{ 2575 if (type != TCG_TYPE_I32) { 2576 tcg_out_sh64(s, RSY_SRLG, dst, src, v, i); 2577 } else if (dst == src) { 2578 tcg_out_sh32(s, RS_SRL, dst, v, i); 2579 } else { 2580 tcg_out_sh64(s, RSY_SRLK, dst, src, v, i); 2581 } 2582} 2583 2584static void tgen_shr(TCGContext *s, TCGType type, 2585 TCGReg a0, TCGReg a1, TCGReg a2) 2586{ 2587 tgen_shr_int(s, type, a0, a1, a2, 0); 2588} 2589 2590static void tgen_shri(TCGContext *s, TCGType type, 2591 TCGReg a0, TCGReg a1, tcg_target_long a2) 2592{ 2593 tgen_shr_int(s, type, a0, a1, TCG_REG_NONE, a2); 2594} 2595 2596static const TCGOutOpBinary outop_shr = { 2597 .base.static_constraint = C_O1_I2(r, r, ri), 2598 .out_rrr = tgen_shr, 2599 .out_rri = tgen_shri, 2600}; 2601 2602static void tgen_sub(TCGContext *s, TCGType type, 2603 TCGReg a0, TCGReg a1, TCGReg a2) 2604{ 2605 if (type != TCG_TYPE_I32) { 2606 tcg_out_insn(s, RRFa, SGRK, a0, a1, a2); 2607 } else if (a0 == a1) { 2608 tcg_out_insn(s, RR, SR, a0, a2); 2609 } else { 2610 tcg_out_insn(s, RRFa, SRK, a0, a1, a2); 2611 } 2612} 2613 2614static const TCGOutOpSubtract outop_sub = { 2615 .base.static_constraint = C_O1_I2(r, r, r), 2616 .out_rrr = tgen_sub, 2617}; 2618 2619static void tgen_xor(TCGContext *s, TCGType type, 2620 TCGReg a0, TCGReg a1, TCGReg a2) 2621{ 2622 if (type != TCG_TYPE_I32) { 2623 tcg_out_insn(s, RRFa, XGRK, a0, a1, a2); 2624 } else if (a0 == a1) { 2625 tcg_out_insn(s, RR, XR, a0, a2); 2626 } else { 2627 tcg_out_insn(s, RRFa, XRK, a0, a1, a2); 2628 } 2629} 2630 2631static void tgen_xori_3(TCGContext *s, TCGType type, 2632 TCGReg a0, TCGReg a1, tcg_target_long a2) 2633{ 2634 tcg_out_mov(s, type, a0, a1); 2635 tgen_xori(s, a0, type == TCG_TYPE_I32 ? (uint32_t)a2 : a2); 2636} 2637 2638static const TCGOutOpBinary outop_xor = { 2639 .base.static_constraint = C_O1_I2(r, r, rK), 2640 .out_rrr = tgen_xor, 2641 .out_rri = tgen_xori_3, 2642}; 2643 2644static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1) 2645{ 2646 if (type == TCG_TYPE_I32) { 2647 tcg_out_insn(s, RR, LCR, a0, a1); 2648 } else { 2649 tcg_out_insn(s, RRE, LCGR, a0, a1); 2650 } 2651} 2652 2653static const TCGOutOpUnary outop_neg = { 2654 .base.static_constraint = C_O1_I1(r, r), 2655 .out_rr = tgen_neg, 2656}; 2657 2658static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1) 2659{ 2660 tgen_nor(s, type, a0, a1, a1); 2661} 2662 2663static TCGConstraintSetIndex cset_not(TCGType type, unsigned flags) 2664{ 2665 return HAVE_FACILITY(MISC_INSN_EXT3) ? C_O1_I1(r, r) : C_NotImplemented; 2666} 2667 2668static const TCGOutOpUnary outop_not = { 2669 .base.static_constraint = C_Dynamic, 2670 .base.dynamic_constraint = cset_not, 2671 .out_rr = tgen_not, 2672}; 2673 2674 2675# define OP_32_64(x) \ 2676 case glue(glue(INDEX_op_,x),_i32): \ 2677 case glue(glue(INDEX_op_,x),_i64) 2678 2679static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type, 2680 const TCGArg args[TCG_MAX_OP_ARGS], 2681 const int const_args[TCG_MAX_OP_ARGS]) 2682{ 2683 TCGArg a0, a1, a2; 2684 2685 switch (opc) { 2686 case INDEX_op_goto_ptr: 2687 a0 = args[0]; 2688 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, a0); 2689 break; 2690 2691 OP_32_64(ld8u): 2692 /* ??? LLC (RXY format) is only present with the extended-immediate 2693 facility, whereas LLGC is always present. */ 2694 tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]); 2695 break; 2696 2697 OP_32_64(ld8s): 2698 /* ??? LB is no smaller than LGB, so no point to using it. */ 2699 tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]); 2700 break; 2701 2702 OP_32_64(ld16u): 2703 /* ??? LLH (RXY format) is only present with the extended-immediate 2704 facility, whereas LLGH is always present. */ 2705 tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]); 2706 break; 2707 2708 case INDEX_op_ld16s_i32: 2709 tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]); 2710 break; 2711 2712 case INDEX_op_ld_i32: 2713 tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]); 2714 break; 2715 2716 OP_32_64(st8): 2717 tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1], 2718 TCG_REG_NONE, args[2]); 2719 break; 2720 2721 OP_32_64(st16): 2722 tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1], 2723 TCG_REG_NONE, args[2]); 2724 break; 2725 2726 case INDEX_op_st_i32: 2727 tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]); 2728 break; 2729 2730 case INDEX_op_bswap16_i32: 2731 a0 = args[0], a1 = args[1], a2 = args[2]; 2732 tcg_out_insn(s, RRE, LRVR, a0, a1); 2733 if (a2 & TCG_BSWAP_OS) { 2734 tcg_out_sh32(s, RS_SRA, a0, TCG_REG_NONE, 16); 2735 } else { 2736 tcg_out_sh32(s, RS_SRL, a0, TCG_REG_NONE, 16); 2737 } 2738 break; 2739 case INDEX_op_bswap16_i64: 2740 a0 = args[0], a1 = args[1], a2 = args[2]; 2741 tcg_out_insn(s, RRE, LRVGR, a0, a1); 2742 if (a2 & TCG_BSWAP_OS) { 2743 tcg_out_sh64(s, RSY_SRAG, a0, a0, TCG_REG_NONE, 48); 2744 } else { 2745 tcg_out_sh64(s, RSY_SRLG, a0, a0, TCG_REG_NONE, 48); 2746 } 2747 break; 2748 2749 case INDEX_op_bswap32_i32: 2750 tcg_out_insn(s, RRE, LRVR, args[0], args[1]); 2751 break; 2752 case INDEX_op_bswap32_i64: 2753 a0 = args[0], a1 = args[1], a2 = args[2]; 2754 tcg_out_insn(s, RRE, LRVR, a0, a1); 2755 if (a2 & TCG_BSWAP_OS) { 2756 tcg_out_ext32s(s, a0, a0); 2757 } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) { 2758 tcg_out_ext32u(s, a0, a0); 2759 } 2760 break; 2761 2762 case INDEX_op_add2_i32: 2763 if (const_args[4]) { 2764 tcg_out_insn(s, RIL, ALFI, args[0], args[4]); 2765 } else { 2766 tcg_out_insn(s, RR, ALR, args[0], args[4]); 2767 } 2768 tcg_out_insn(s, RRE, ALCR, args[1], args[5]); 2769 break; 2770 case INDEX_op_sub2_i32: 2771 if (const_args[4]) { 2772 tcg_out_insn(s, RIL, SLFI, args[0], args[4]); 2773 } else { 2774 tcg_out_insn(s, RR, SLR, args[0], args[4]); 2775 } 2776 tcg_out_insn(s, RRE, SLBR, args[1], args[5]); 2777 break; 2778 2779 case INDEX_op_br: 2780 tgen_branch(s, S390_CC_ALWAYS, arg_label(args[0])); 2781 break; 2782 2783 case INDEX_op_brcond_i32: 2784 tgen_brcond(s, TCG_TYPE_I32, args[2], args[0], 2785 args[1], const_args[1], arg_label(args[3])); 2786 break; 2787 case INDEX_op_setcond_i32: 2788 tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], 2789 args[2], const_args[2], false); 2790 break; 2791 case INDEX_op_negsetcond_i32: 2792 tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1], 2793 args[2], const_args[2], true); 2794 break; 2795 case INDEX_op_movcond_i32: 2796 tgen_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1], 2797 args[2], const_args[2], args[3], const_args[3], args[4]); 2798 break; 2799 2800 case INDEX_op_qemu_ld_i32: 2801 tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I32); 2802 break; 2803 case INDEX_op_qemu_ld_i64: 2804 tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I64); 2805 break; 2806 case INDEX_op_qemu_st_i32: 2807 tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I32); 2808 break; 2809 case INDEX_op_qemu_st_i64: 2810 tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I64); 2811 break; 2812 case INDEX_op_qemu_ld_i128: 2813 tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], true); 2814 break; 2815 case INDEX_op_qemu_st_i128: 2816 tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], false); 2817 break; 2818 2819 case INDEX_op_ld16s_i64: 2820 tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]); 2821 break; 2822 case INDEX_op_ld32u_i64: 2823 tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]); 2824 break; 2825 case INDEX_op_ld32s_i64: 2826 tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]); 2827 break; 2828 case INDEX_op_ld_i64: 2829 tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]); 2830 break; 2831 2832 case INDEX_op_st32_i64: 2833 tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]); 2834 break; 2835 case INDEX_op_st_i64: 2836 tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]); 2837 break; 2838 2839 case INDEX_op_bswap64_i64: 2840 tcg_out_insn(s, RRE, LRVGR, args[0], args[1]); 2841 break; 2842 2843 case INDEX_op_mulu2_i64: 2844 tcg_debug_assert(args[0] == args[2]); 2845 tcg_debug_assert((args[1] & 1) == 0); 2846 tcg_debug_assert(args[0] == args[1] + 1); 2847 tcg_out_insn(s, RRE, MLGR, args[1], args[3]); 2848 break; 2849 case INDEX_op_muls2_i64: 2850 tcg_debug_assert((args[1] & 1) == 0); 2851 tcg_debug_assert(args[0] == args[1] + 1); 2852 tcg_out_insn(s, RRFa, MGRK, args[1], args[2], args[3]); 2853 break; 2854 2855 case INDEX_op_add2_i64: 2856 if (const_args[4]) { 2857 if ((int64_t)args[4] >= 0) { 2858 tcg_out_insn(s, RIL, ALGFI, args[0], args[4]); 2859 } else { 2860 tcg_out_insn(s, RIL, SLGFI, args[0], -args[4]); 2861 } 2862 } else { 2863 tcg_out_insn(s, RRE, ALGR, args[0], args[4]); 2864 } 2865 tcg_out_insn(s, RRE, ALCGR, args[1], args[5]); 2866 break; 2867 case INDEX_op_sub2_i64: 2868 if (const_args[4]) { 2869 if ((int64_t)args[4] >= 0) { 2870 tcg_out_insn(s, RIL, SLGFI, args[0], args[4]); 2871 } else { 2872 tcg_out_insn(s, RIL, ALGFI, args[0], -args[4]); 2873 } 2874 } else { 2875 tcg_out_insn(s, RRE, SLGR, args[0], args[4]); 2876 } 2877 tcg_out_insn(s, RRE, SLBGR, args[1], args[5]); 2878 break; 2879 2880 case INDEX_op_brcond_i64: 2881 tgen_brcond(s, TCG_TYPE_I64, args[2], args[0], 2882 args[1], const_args[1], arg_label(args[3])); 2883 break; 2884 case INDEX_op_setcond_i64: 2885 tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], 2886 args[2], const_args[2], false); 2887 break; 2888 case INDEX_op_negsetcond_i64: 2889 tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1], 2890 args[2], const_args[2], true); 2891 break; 2892 case INDEX_op_movcond_i64: 2893 tgen_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1], 2894 args[2], const_args[2], args[3], const_args[3], args[4]); 2895 break; 2896 2897 OP_32_64(deposit): 2898 a0 = args[0], a1 = args[1], a2 = args[2]; 2899 if (const_args[1]) { 2900 tgen_deposit(s, a0, a2, args[3], args[4], 1); 2901 } else { 2902 /* Since we can't support "0Z" as a constraint, we allow a1 in 2903 any register. Fix things up as if a matching constraint. */ 2904 if (a0 != a1) { 2905 if (a0 == a2) { 2906 tcg_out_mov(s, type, TCG_TMP0, a2); 2907 a2 = TCG_TMP0; 2908 } 2909 tcg_out_mov(s, type, a0, a1); 2910 } 2911 tgen_deposit(s, a0, a2, args[3], args[4], 0); 2912 } 2913 break; 2914 2915 OP_32_64(extract): 2916 tgen_extract(s, args[0], args[1], args[2], args[3]); 2917 break; 2918 OP_32_64(sextract): 2919 tgen_sextract(s, args[0], args[1], args[2], args[3]); 2920 break; 2921 2922 case INDEX_op_mb: 2923 /* The host memory model is quite strong, we simply need to 2924 serialize the instruction stream. */ 2925 if (args[0] & TCG_MO_ST_LD) { 2926 /* fast-bcr-serialization facility (45) is present */ 2927 tcg_out_insn(s, RR, BCR, 14, 0); 2928 } 2929 break; 2930 2931 case INDEX_op_call: /* Always emitted via tcg_out_call. */ 2932 case INDEX_op_exit_tb: /* Always emitted via tcg_out_exit_tb. */ 2933 case INDEX_op_goto_tb: /* Always emitted via tcg_out_goto_tb. */ 2934 case INDEX_op_ext_i32_i64: /* Always emitted via tcg_reg_alloc_op. */ 2935 case INDEX_op_extu_i32_i64: 2936 case INDEX_op_extrl_i64_i32: 2937 default: 2938 g_assert_not_reached(); 2939 } 2940} 2941 2942static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece, 2943 TCGReg dst, TCGReg src) 2944{ 2945 if (is_general_reg(src)) { 2946 /* Replicate general register into two MO_64. */ 2947 tcg_out_insn(s, VRRf, VLVGP, dst, src, src); 2948 if (vece == MO_64) { 2949 return true; 2950 } 2951 src = dst; 2952 } 2953 2954 /* 2955 * Recall that the "standard" integer, within a vector, is the 2956 * rightmost element of the leftmost doubleword, a-la VLLEZ. 2957 */ 2958 tcg_out_insn(s, VRIc, VREP, dst, (8 >> vece) - 1, src, vece); 2959 return true; 2960} 2961 2962static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece, 2963 TCGReg dst, TCGReg base, intptr_t offset) 2964{ 2965 tcg_out_vrx_mem(s, VRX_VLREP, dst, base, TCG_REG_NONE, offset, vece); 2966 return true; 2967} 2968 2969static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece, 2970 TCGReg dst, int64_t val) 2971{ 2972 int i, mask, msb, lsb; 2973 2974 /* Look for int16_t elements. */ 2975 if (vece <= MO_16 || 2976 (vece == MO_32 ? (int32_t)val : val) == (int16_t)val) { 2977 tcg_out_insn(s, VRIa, VREPI, dst, val, vece); 2978 return; 2979 } 2980 2981 /* Look for bit masks. */ 2982 if (vece == MO_32) { 2983 if (risbg_mask((int32_t)val)) { 2984 /* Handle wraparound by swapping msb and lsb. */ 2985 if ((val & 0x80000001u) == 0x80000001u) { 2986 msb = 32 - ctz32(~val); 2987 lsb = clz32(~val) - 1; 2988 } else { 2989 msb = clz32(val); 2990 lsb = 31 - ctz32(val); 2991 } 2992 tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_32); 2993 return; 2994 } 2995 } else { 2996 if (risbg_mask(val)) { 2997 /* Handle wraparound by swapping msb and lsb. */ 2998 if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) { 2999 /* Handle wraparound by swapping msb and lsb. */ 3000 msb = 64 - ctz64(~val); 3001 lsb = clz64(~val) - 1; 3002 } else { 3003 msb = clz64(val); 3004 lsb = 63 - ctz64(val); 3005 } 3006 tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_64); 3007 return; 3008 } 3009 } 3010 3011 /* Look for all bytes 0x00 or 0xff. */ 3012 for (i = mask = 0; i < 8; i++) { 3013 uint8_t byte = val >> (i * 8); 3014 if (byte == 0xff) { 3015 mask |= 1 << i; 3016 } else if (byte != 0) { 3017 break; 3018 } 3019 } 3020 if (i == 8) { 3021 tcg_out_insn(s, VRIa, VGBM, dst, mask * 0x0101, 0); 3022 return; 3023 } 3024 3025 /* Otherwise, stuff it in the constant pool. */ 3026 tcg_out_insn(s, RIL, LARL, TCG_TMP0, 0); 3027 new_pool_label(s, val, R_390_PC32DBL, s->code_ptr - 2, 2); 3028 tcg_out_insn(s, VRX, VLREP, dst, TCG_TMP0, TCG_REG_NONE, 0, MO_64); 3029} 3030 3031static bool tcg_out_cmp_vec_noinv(TCGContext *s, unsigned vece, TCGReg a0, 3032 TCGReg a1, TCGReg a2, TCGCond cond) 3033{ 3034 bool need_swap = false, need_inv = false; 3035 3036 switch (cond) { 3037 case TCG_COND_EQ: 3038 case TCG_COND_GT: 3039 case TCG_COND_GTU: 3040 break; 3041 case TCG_COND_NE: 3042 case TCG_COND_LE: 3043 case TCG_COND_LEU: 3044 need_inv = true; 3045 break; 3046 case TCG_COND_LT: 3047 case TCG_COND_LTU: 3048 need_swap = true; 3049 break; 3050 case TCG_COND_GE: 3051 case TCG_COND_GEU: 3052 need_swap = need_inv = true; 3053 break; 3054 default: 3055 g_assert_not_reached(); 3056 } 3057 3058 if (need_inv) { 3059 cond = tcg_invert_cond(cond); 3060 } 3061 if (need_swap) { 3062 TCGReg swap = a1; 3063 a1 = a2; 3064 a2 = swap; 3065 cond = tcg_swap_cond(cond); 3066 } 3067 3068 switch (cond) { 3069 case TCG_COND_EQ: 3070 tcg_out_insn(s, VRRc, VCEQ, a0, a1, a2, vece); 3071 break; 3072 case TCG_COND_GT: 3073 tcg_out_insn(s, VRRc, VCH, a0, a1, a2, vece); 3074 break; 3075 case TCG_COND_GTU: 3076 tcg_out_insn(s, VRRc, VCHL, a0, a1, a2, vece); 3077 break; 3078 default: 3079 g_assert_not_reached(); 3080 } 3081 return need_inv; 3082} 3083 3084static void tcg_out_cmp_vec(TCGContext *s, unsigned vece, TCGReg a0, 3085 TCGReg a1, TCGReg a2, TCGCond cond) 3086{ 3087 if (tcg_out_cmp_vec_noinv(s, vece, a0, a1, a2, cond)) { 3088 tcg_out_insn(s, VRRc, VNO, a0, a0, a0, 0); 3089 } 3090} 3091 3092static void tcg_out_cmpsel_vec(TCGContext *s, unsigned vece, TCGReg a0, 3093 TCGReg c1, TCGReg c2, TCGArg v3, 3094 int const_v3, TCGReg v4, TCGCond cond) 3095{ 3096 bool inv = tcg_out_cmp_vec_noinv(s, vece, TCG_VEC_TMP0, c1, c2, cond); 3097 3098 if (!const_v3) { 3099 if (inv) { 3100 tcg_out_insn(s, VRRe, VSEL, a0, v4, v3, TCG_VEC_TMP0); 3101 } else { 3102 tcg_out_insn(s, VRRe, VSEL, a0, v3, v4, TCG_VEC_TMP0); 3103 } 3104 } else if (v3) { 3105 if (inv) { 3106 tcg_out_insn(s, VRRc, VOC, a0, v4, TCG_VEC_TMP0, 0); 3107 } else { 3108 tcg_out_insn(s, VRRc, VO, a0, v4, TCG_VEC_TMP0, 0); 3109 } 3110 } else { 3111 if (inv) { 3112 tcg_out_insn(s, VRRc, VN, a0, v4, TCG_VEC_TMP0, 0); 3113 } else { 3114 tcg_out_insn(s, VRRc, VNC, a0, v4, TCG_VEC_TMP0, 0); 3115 } 3116 } 3117} 3118 3119static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc, 3120 unsigned vecl, unsigned vece, 3121 const TCGArg args[TCG_MAX_OP_ARGS], 3122 const int const_args[TCG_MAX_OP_ARGS]) 3123{ 3124 TCGType type = vecl + TCG_TYPE_V64; 3125 TCGArg a0 = args[0], a1 = args[1], a2 = args[2]; 3126 3127 switch (opc) { 3128 case INDEX_op_ld_vec: 3129 tcg_out_ld(s, type, a0, a1, a2); 3130 break; 3131 case INDEX_op_st_vec: 3132 tcg_out_st(s, type, a0, a1, a2); 3133 break; 3134 case INDEX_op_dupm_vec: 3135 tcg_out_dupm_vec(s, type, vece, a0, a1, a2); 3136 break; 3137 3138 case INDEX_op_abs_vec: 3139 tcg_out_insn(s, VRRa, VLP, a0, a1, vece); 3140 break; 3141 case INDEX_op_neg_vec: 3142 tcg_out_insn(s, VRRa, VLC, a0, a1, vece); 3143 break; 3144 case INDEX_op_not_vec: 3145 tcg_out_insn(s, VRRc, VNO, a0, a1, a1, 0); 3146 break; 3147 3148 case INDEX_op_add_vec: 3149 tcg_out_insn(s, VRRc, VA, a0, a1, a2, vece); 3150 break; 3151 case INDEX_op_sub_vec: 3152 tcg_out_insn(s, VRRc, VS, a0, a1, a2, vece); 3153 break; 3154 case INDEX_op_and_vec: 3155 tcg_out_insn(s, VRRc, VN, a0, a1, a2, 0); 3156 break; 3157 case INDEX_op_andc_vec: 3158 tcg_out_insn(s, VRRc, VNC, a0, a1, a2, 0); 3159 break; 3160 case INDEX_op_mul_vec: 3161 tcg_out_insn(s, VRRc, VML, a0, a1, a2, vece); 3162 break; 3163 case INDEX_op_or_vec: 3164 tcg_out_insn(s, VRRc, VO, a0, a1, a2, 0); 3165 break; 3166 case INDEX_op_orc_vec: 3167 tcg_out_insn(s, VRRc, VOC, a0, a1, a2, 0); 3168 break; 3169 case INDEX_op_xor_vec: 3170 tcg_out_insn(s, VRRc, VX, a0, a1, a2, 0); 3171 break; 3172 case INDEX_op_nand_vec: 3173 tcg_out_insn(s, VRRc, VNN, a0, a1, a2, 0); 3174 break; 3175 case INDEX_op_nor_vec: 3176 tcg_out_insn(s, VRRc, VNO, a0, a1, a2, 0); 3177 break; 3178 case INDEX_op_eqv_vec: 3179 tcg_out_insn(s, VRRc, VNX, a0, a1, a2, 0); 3180 break; 3181 3182 case INDEX_op_shli_vec: 3183 tcg_out_insn(s, VRSa, VESL, a0, a2, TCG_REG_NONE, a1, vece); 3184 break; 3185 case INDEX_op_shri_vec: 3186 tcg_out_insn(s, VRSa, VESRL, a0, a2, TCG_REG_NONE, a1, vece); 3187 break; 3188 case INDEX_op_sari_vec: 3189 tcg_out_insn(s, VRSa, VESRA, a0, a2, TCG_REG_NONE, a1, vece); 3190 break; 3191 case INDEX_op_rotli_vec: 3192 tcg_out_insn(s, VRSa, VERLL, a0, a2, TCG_REG_NONE, a1, vece); 3193 break; 3194 case INDEX_op_shls_vec: 3195 tcg_out_insn(s, VRSa, VESL, a0, 0, a2, a1, vece); 3196 break; 3197 case INDEX_op_shrs_vec: 3198 tcg_out_insn(s, VRSa, VESRL, a0, 0, a2, a1, vece); 3199 break; 3200 case INDEX_op_sars_vec: 3201 tcg_out_insn(s, VRSa, VESRA, a0, 0, a2, a1, vece); 3202 break; 3203 case INDEX_op_rotls_vec: 3204 tcg_out_insn(s, VRSa, VERLL, a0, 0, a2, a1, vece); 3205 break; 3206 case INDEX_op_shlv_vec: 3207 tcg_out_insn(s, VRRc, VESLV, a0, a1, a2, vece); 3208 break; 3209 case INDEX_op_shrv_vec: 3210 tcg_out_insn(s, VRRc, VESRLV, a0, a1, a2, vece); 3211 break; 3212 case INDEX_op_sarv_vec: 3213 tcg_out_insn(s, VRRc, VESRAV, a0, a1, a2, vece); 3214 break; 3215 case INDEX_op_rotlv_vec: 3216 tcg_out_insn(s, VRRc, VERLLV, a0, a1, a2, vece); 3217 break; 3218 3219 case INDEX_op_smin_vec: 3220 tcg_out_insn(s, VRRc, VMN, a0, a1, a2, vece); 3221 break; 3222 case INDEX_op_smax_vec: 3223 tcg_out_insn(s, VRRc, VMX, a0, a1, a2, vece); 3224 break; 3225 case INDEX_op_umin_vec: 3226 tcg_out_insn(s, VRRc, VMNL, a0, a1, a2, vece); 3227 break; 3228 case INDEX_op_umax_vec: 3229 tcg_out_insn(s, VRRc, VMXL, a0, a1, a2, vece); 3230 break; 3231 3232 case INDEX_op_bitsel_vec: 3233 tcg_out_insn(s, VRRe, VSEL, a0, a2, args[3], a1); 3234 break; 3235 3236 case INDEX_op_cmp_vec: 3237 tcg_out_cmp_vec(s, vece, a0, a1, a2, args[3]); 3238 break; 3239 case INDEX_op_cmpsel_vec: 3240 tcg_out_cmpsel_vec(s, vece, a0, a1, a2, args[3], const_args[3], 3241 args[4], args[5]); 3242 break; 3243 3244 case INDEX_op_s390_vuph_vec: 3245 tcg_out_insn(s, VRRa, VUPH, a0, a1, vece); 3246 break; 3247 case INDEX_op_s390_vupl_vec: 3248 tcg_out_insn(s, VRRa, VUPL, a0, a1, vece); 3249 break; 3250 case INDEX_op_s390_vpks_vec: 3251 tcg_out_insn(s, VRRc, VPKS, a0, a1, a2, vece); 3252 break; 3253 3254 case INDEX_op_mov_vec: /* Always emitted via tcg_out_mov. */ 3255 case INDEX_op_dup_vec: /* Always emitted via tcg_out_dup_vec. */ 3256 default: 3257 g_assert_not_reached(); 3258 } 3259} 3260 3261int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece) 3262{ 3263 switch (opc) { 3264 case INDEX_op_abs_vec: 3265 case INDEX_op_add_vec: 3266 case INDEX_op_and_vec: 3267 case INDEX_op_andc_vec: 3268 case INDEX_op_bitsel_vec: 3269 case INDEX_op_eqv_vec: 3270 case INDEX_op_nand_vec: 3271 case INDEX_op_neg_vec: 3272 case INDEX_op_nor_vec: 3273 case INDEX_op_not_vec: 3274 case INDEX_op_or_vec: 3275 case INDEX_op_orc_vec: 3276 case INDEX_op_rotli_vec: 3277 case INDEX_op_rotls_vec: 3278 case INDEX_op_rotlv_vec: 3279 case INDEX_op_sari_vec: 3280 case INDEX_op_sars_vec: 3281 case INDEX_op_sarv_vec: 3282 case INDEX_op_shli_vec: 3283 case INDEX_op_shls_vec: 3284 case INDEX_op_shlv_vec: 3285 case INDEX_op_shri_vec: 3286 case INDEX_op_shrs_vec: 3287 case INDEX_op_shrv_vec: 3288 case INDEX_op_smax_vec: 3289 case INDEX_op_smin_vec: 3290 case INDEX_op_sub_vec: 3291 case INDEX_op_umax_vec: 3292 case INDEX_op_umin_vec: 3293 case INDEX_op_xor_vec: 3294 case INDEX_op_cmp_vec: 3295 case INDEX_op_cmpsel_vec: 3296 return 1; 3297 case INDEX_op_rotrv_vec: 3298 return -1; 3299 case INDEX_op_mul_vec: 3300 return vece < MO_64; 3301 case INDEX_op_ssadd_vec: 3302 case INDEX_op_sssub_vec: 3303 return vece < MO_64 ? -1 : 0; 3304 default: 3305 return 0; 3306 } 3307} 3308 3309static void expand_vec_sat(TCGType type, unsigned vece, TCGv_vec v0, 3310 TCGv_vec v1, TCGv_vec v2, TCGOpcode add_sub_opc) 3311{ 3312 TCGv_vec h1 = tcg_temp_new_vec(type); 3313 TCGv_vec h2 = tcg_temp_new_vec(type); 3314 TCGv_vec l1 = tcg_temp_new_vec(type); 3315 TCGv_vec l2 = tcg_temp_new_vec(type); 3316 3317 tcg_debug_assert (vece < MO_64); 3318 3319 /* Unpack with sign-extension. */ 3320 vec_gen_2(INDEX_op_s390_vuph_vec, type, vece, 3321 tcgv_vec_arg(h1), tcgv_vec_arg(v1)); 3322 vec_gen_2(INDEX_op_s390_vuph_vec, type, vece, 3323 tcgv_vec_arg(h2), tcgv_vec_arg(v2)); 3324 3325 vec_gen_2(INDEX_op_s390_vupl_vec, type, vece, 3326 tcgv_vec_arg(l1), tcgv_vec_arg(v1)); 3327 vec_gen_2(INDEX_op_s390_vupl_vec, type, vece, 3328 tcgv_vec_arg(l2), tcgv_vec_arg(v2)); 3329 3330 /* Arithmetic on a wider element size. */ 3331 vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(h1), 3332 tcgv_vec_arg(h1), tcgv_vec_arg(h2)); 3333 vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(l1), 3334 tcgv_vec_arg(l1), tcgv_vec_arg(l2)); 3335 3336 /* Pack with saturation. */ 3337 vec_gen_3(INDEX_op_s390_vpks_vec, type, vece + 1, 3338 tcgv_vec_arg(v0), tcgv_vec_arg(h1), tcgv_vec_arg(l1)); 3339 3340 tcg_temp_free_vec(h1); 3341 tcg_temp_free_vec(h2); 3342 tcg_temp_free_vec(l1); 3343 tcg_temp_free_vec(l2); 3344} 3345 3346void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece, 3347 TCGArg a0, ...) 3348{ 3349 va_list va; 3350 TCGv_vec v0, v1, v2, t0; 3351 3352 va_start(va, a0); 3353 v0 = temp_tcgv_vec(arg_temp(a0)); 3354 v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg))); 3355 v2 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg))); 3356 3357 switch (opc) { 3358 case INDEX_op_rotrv_vec: 3359 t0 = tcg_temp_new_vec(type); 3360 tcg_gen_neg_vec(vece, t0, v2); 3361 tcg_gen_rotlv_vec(vece, v0, v1, t0); 3362 tcg_temp_free_vec(t0); 3363 break; 3364 3365 case INDEX_op_ssadd_vec: 3366 expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_add_vec); 3367 break; 3368 case INDEX_op_sssub_vec: 3369 expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_sub_vec); 3370 break; 3371 3372 default: 3373 g_assert_not_reached(); 3374 } 3375 va_end(va); 3376} 3377 3378static TCGConstraintSetIndex 3379tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags) 3380{ 3381 switch (op) { 3382 case INDEX_op_goto_ptr: 3383 return C_O0_I1(r); 3384 3385 case INDEX_op_ld8u_i32: 3386 case INDEX_op_ld8u_i64: 3387 case INDEX_op_ld8s_i32: 3388 case INDEX_op_ld8s_i64: 3389 case INDEX_op_ld16u_i32: 3390 case INDEX_op_ld16u_i64: 3391 case INDEX_op_ld16s_i32: 3392 case INDEX_op_ld16s_i64: 3393 case INDEX_op_ld_i32: 3394 case INDEX_op_ld32u_i64: 3395 case INDEX_op_ld32s_i64: 3396 case INDEX_op_ld_i64: 3397 return C_O1_I1(r, r); 3398 3399 case INDEX_op_st8_i32: 3400 case INDEX_op_st8_i64: 3401 case INDEX_op_st16_i32: 3402 case INDEX_op_st16_i64: 3403 case INDEX_op_st_i32: 3404 case INDEX_op_st32_i64: 3405 case INDEX_op_st_i64: 3406 return C_O0_I2(r, r); 3407 3408 case INDEX_op_setcond_i32: 3409 case INDEX_op_negsetcond_i32: 3410 case INDEX_op_setcond_i64: 3411 case INDEX_op_negsetcond_i64: 3412 return C_O1_I2(r, r, rC); 3413 3414 case INDEX_op_brcond_i32: 3415 return C_O0_I2(r, ri); 3416 case INDEX_op_brcond_i64: 3417 return C_O0_I2(r, rC); 3418 3419 case INDEX_op_bswap16_i32: 3420 case INDEX_op_bswap16_i64: 3421 case INDEX_op_bswap32_i32: 3422 case INDEX_op_bswap32_i64: 3423 case INDEX_op_bswap64_i64: 3424 case INDEX_op_ext_i32_i64: 3425 case INDEX_op_extu_i32_i64: 3426 case INDEX_op_extract_i32: 3427 case INDEX_op_extract_i64: 3428 case INDEX_op_sextract_i32: 3429 case INDEX_op_sextract_i64: 3430 return C_O1_I1(r, r); 3431 3432 case INDEX_op_qemu_ld_i32: 3433 case INDEX_op_qemu_ld_i64: 3434 return C_O1_I1(r, r); 3435 case INDEX_op_qemu_st_i64: 3436 case INDEX_op_qemu_st_i32: 3437 return C_O0_I2(r, r); 3438 case INDEX_op_qemu_ld_i128: 3439 return C_O2_I1(o, m, r); 3440 case INDEX_op_qemu_st_i128: 3441 return C_O0_I3(o, m, r); 3442 3443 case INDEX_op_deposit_i32: 3444 case INDEX_op_deposit_i64: 3445 return C_O1_I2(r, rZ, r); 3446 3447 case INDEX_op_movcond_i32: 3448 return C_O1_I4(r, r, ri, rI, r); 3449 case INDEX_op_movcond_i64: 3450 return C_O1_I4(r, r, rC, rI, r); 3451 3452 case INDEX_op_mulu2_i64: 3453 return C_O2_I2(o, m, 0, r); 3454 case INDEX_op_muls2_i64: 3455 return C_O2_I2(o, m, r, r); 3456 3457 case INDEX_op_add2_i32: 3458 case INDEX_op_sub2_i32: 3459 return C_N1_O1_I4(r, r, 0, 1, ri, r); 3460 3461 case INDEX_op_add2_i64: 3462 case INDEX_op_sub2_i64: 3463 return C_N1_O1_I4(r, r, 0, 1, rJU, r); 3464 3465 case INDEX_op_st_vec: 3466 return C_O0_I2(v, r); 3467 case INDEX_op_ld_vec: 3468 case INDEX_op_dupm_vec: 3469 return C_O1_I1(v, r); 3470 case INDEX_op_dup_vec: 3471 return C_O1_I1(v, vr); 3472 case INDEX_op_abs_vec: 3473 case INDEX_op_neg_vec: 3474 case INDEX_op_not_vec: 3475 case INDEX_op_rotli_vec: 3476 case INDEX_op_sari_vec: 3477 case INDEX_op_shli_vec: 3478 case INDEX_op_shri_vec: 3479 case INDEX_op_s390_vuph_vec: 3480 case INDEX_op_s390_vupl_vec: 3481 return C_O1_I1(v, v); 3482 case INDEX_op_add_vec: 3483 case INDEX_op_sub_vec: 3484 case INDEX_op_and_vec: 3485 case INDEX_op_andc_vec: 3486 case INDEX_op_or_vec: 3487 case INDEX_op_orc_vec: 3488 case INDEX_op_xor_vec: 3489 case INDEX_op_nand_vec: 3490 case INDEX_op_nor_vec: 3491 case INDEX_op_eqv_vec: 3492 case INDEX_op_cmp_vec: 3493 case INDEX_op_mul_vec: 3494 case INDEX_op_rotlv_vec: 3495 case INDEX_op_rotrv_vec: 3496 case INDEX_op_shlv_vec: 3497 case INDEX_op_shrv_vec: 3498 case INDEX_op_sarv_vec: 3499 case INDEX_op_smax_vec: 3500 case INDEX_op_smin_vec: 3501 case INDEX_op_umax_vec: 3502 case INDEX_op_umin_vec: 3503 case INDEX_op_s390_vpks_vec: 3504 return C_O1_I2(v, v, v); 3505 case INDEX_op_rotls_vec: 3506 case INDEX_op_shls_vec: 3507 case INDEX_op_shrs_vec: 3508 case INDEX_op_sars_vec: 3509 return C_O1_I2(v, v, r); 3510 case INDEX_op_bitsel_vec: 3511 return C_O1_I3(v, v, v, v); 3512 case INDEX_op_cmpsel_vec: 3513 return (TCG_TARGET_HAS_orc_vec 3514 ? C_O1_I4(v, v, v, vZM, v) 3515 : C_O1_I4(v, v, v, vZ, v)); 3516 3517 default: 3518 return C_NotImplemented; 3519 } 3520} 3521 3522/* 3523 * Mainline glibc added HWCAP_S390_VX before it was kernel abi. 3524 * Some distros have fixed this up locally, others have not. 3525 */ 3526#ifndef HWCAP_S390_VXRS 3527#define HWCAP_S390_VXRS 2048 3528#endif 3529 3530static void query_s390_facilities(void) 3531{ 3532 unsigned long hwcap = qemu_getauxval(AT_HWCAP); 3533 const char *which; 3534 3535 /* Is STORE FACILITY LIST EXTENDED available? Honestly, I believe this 3536 is present on all 64-bit systems, but let's check for it anyway. */ 3537 if (hwcap & HWCAP_S390_STFLE) { 3538 register int r0 __asm__("0") = ARRAY_SIZE(s390_facilities) - 1; 3539 register void *r1 __asm__("1") = s390_facilities; 3540 3541 /* stfle 0(%r1) */ 3542 asm volatile(".word 0xb2b0,0x1000" 3543 : "=r"(r0) : "r"(r0), "r"(r1) : "memory", "cc"); 3544 } 3545 3546 /* 3547 * Use of vector registers requires os support beyond the facility bit. 3548 * If the kernel does not advertise support, disable the facility bits. 3549 * There is nothing else we currently care about in the 3rd word, so 3550 * disable VECTOR with one store. 3551 */ 3552 if (!(hwcap & HWCAP_S390_VXRS)) { 3553 s390_facilities[2] = 0; 3554 } 3555 3556 /* 3557 * Minimum supported cpu revision is z196. 3558 * Check for all required facilities. 3559 * ZARCH_ACTIVE is done via preprocessor check for 64-bit. 3560 */ 3561 if (!HAVE_FACILITY(LONG_DISP)) { 3562 which = "long-displacement"; 3563 goto fail; 3564 } 3565 if (!HAVE_FACILITY(EXT_IMM)) { 3566 which = "extended-immediate"; 3567 goto fail; 3568 } 3569 if (!HAVE_FACILITY(GEN_INST_EXT)) { 3570 which = "general-instructions-extension"; 3571 goto fail; 3572 } 3573 /* 3574 * Facility 45 is a big bin that contains: distinct-operands, 3575 * fast-BCR-serialization, high-word, population-count, 3576 * interlocked-access-1, and load/store-on-condition-1 3577 */ 3578 if (!HAVE_FACILITY(45)) { 3579 which = "45"; 3580 goto fail; 3581 } 3582 return; 3583 3584 fail: 3585 error_report("%s: missing required facility %s", __func__, which); 3586 exit(EXIT_FAILURE); 3587} 3588 3589static void tcg_target_init(TCGContext *s) 3590{ 3591 query_s390_facilities(); 3592 3593 tcg_target_available_regs[TCG_TYPE_I32] = 0xffff; 3594 tcg_target_available_regs[TCG_TYPE_I64] = 0xffff; 3595 if (HAVE_FACILITY(VECTOR)) { 3596 tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull; 3597 tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull; 3598 } 3599 3600 tcg_target_call_clobber_regs = 0; 3601 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0); 3602 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1); 3603 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2); 3604 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3); 3605 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4); 3606 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5); 3607 /* The r6 register is technically call-saved, but it's also a parameter 3608 register, so it can get killed by setup for the qemu_st helper. */ 3609 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6); 3610 /* The return register can be considered call-clobbered. */ 3611 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14); 3612 3613 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0); 3614 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1); 3615 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2); 3616 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3); 3617 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4); 3618 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5); 3619 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6); 3620 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7); 3621 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16); 3622 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17); 3623 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18); 3624 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19); 3625 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V20); 3626 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V21); 3627 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V22); 3628 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V23); 3629 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V24); 3630 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V25); 3631 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V26); 3632 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V27); 3633 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V28); 3634 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V29); 3635 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V30); 3636 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V31); 3637 3638 s->reserved_regs = 0; 3639 tcg_regset_set_reg(s->reserved_regs, TCG_TMP0); 3640 tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP0); 3641 /* XXX many insns can't be used with R0, so we better avoid it for now */ 3642 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0); 3643 tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK); 3644} 3645 3646#define FRAME_SIZE ((int)(TCG_TARGET_CALL_STACK_OFFSET \ 3647 + TCG_STATIC_CALL_ARGS_SIZE \ 3648 + CPU_TEMP_BUF_NLONGS * sizeof(long))) 3649 3650static void tcg_target_qemu_prologue(TCGContext *s) 3651{ 3652 /* stmg %r6,%r15,48(%r15) (save registers) */ 3653 tcg_out_insn(s, RXY, STMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 48); 3654 3655 /* aghi %r15,-frame_size */ 3656 tcg_out_insn(s, RI, AGHI, TCG_REG_R15, -FRAME_SIZE); 3657 3658 tcg_set_frame(s, TCG_REG_CALL_STACK, 3659 TCG_STATIC_CALL_ARGS_SIZE + TCG_TARGET_CALL_STACK_OFFSET, 3660 CPU_TEMP_BUF_NLONGS * sizeof(long)); 3661 3662 if (!tcg_use_softmmu && guest_base >= 0x80000) { 3663 tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base); 3664 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG); 3665 } 3666 3667 tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]); 3668 3669 /* br %r3 (go to TB) */ 3670 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, tcg_target_call_iarg_regs[1]); 3671 3672 /* 3673 * Return path for goto_ptr. Set return value to 0, a-la exit_tb, 3674 * and fall through to the rest of the epilogue. 3675 */ 3676 tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr); 3677 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, 0); 3678 3679 /* TB epilogue */ 3680 tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr); 3681 3682 /* lmg %r6,%r15,fs+48(%r15) (restore registers) */ 3683 tcg_out_insn(s, RXY, LMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 3684 FRAME_SIZE + 48); 3685 3686 /* br %r14 (return) */ 3687 tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14); 3688} 3689 3690static void tcg_out_tb_start(TCGContext *s) 3691{ 3692 /* nothing to do */ 3693} 3694 3695static void tcg_out_nop_fill(tcg_insn_unit *p, int count) 3696{ 3697 memset(p, 0x07, count * sizeof(tcg_insn_unit)); 3698} 3699 3700typedef struct { 3701 DebugFrameHeader h; 3702 uint8_t fde_def_cfa[4]; 3703 uint8_t fde_reg_ofs[18]; 3704} DebugFrame; 3705 3706/* We're expecting a 2 byte uleb128 encoded value. */ 3707QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14)); 3708 3709#define ELF_HOST_MACHINE EM_S390 3710 3711static const DebugFrame debug_frame = { 3712 .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */ 3713 .h.cie.id = -1, 3714 .h.cie.version = 1, 3715 .h.cie.code_align = 1, 3716 .h.cie.data_align = 8, /* sleb128 8 */ 3717 .h.cie.return_column = TCG_REG_R14, 3718 3719 /* Total FDE size does not include the "len" member. */ 3720 .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset), 3721 3722 .fde_def_cfa = { 3723 12, TCG_REG_CALL_STACK, /* DW_CFA_def_cfa %r15, ... */ 3724 (FRAME_SIZE & 0x7f) | 0x80, /* ... uleb128 FRAME_SIZE */ 3725 (FRAME_SIZE >> 7) 3726 }, 3727 .fde_reg_ofs = { 3728 0x86, 6, /* DW_CFA_offset, %r6, 48 */ 3729 0x87, 7, /* DW_CFA_offset, %r7, 56 */ 3730 0x88, 8, /* DW_CFA_offset, %r8, 64 */ 3731 0x89, 9, /* DW_CFA_offset, %r92, 72 */ 3732 0x8a, 10, /* DW_CFA_offset, %r10, 80 */ 3733 0x8b, 11, /* DW_CFA_offset, %r11, 88 */ 3734 0x8c, 12, /* DW_CFA_offset, %r12, 96 */ 3735 0x8d, 13, /* DW_CFA_offset, %r13, 104 */ 3736 0x8e, 14, /* DW_CFA_offset, %r14, 112 */ 3737 } 3738}; 3739 3740void tcg_register_jit(const void *buf, size_t buf_size) 3741{ 3742 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame)); 3743} 3744