1 /* 2 * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #ifndef CPU_X86_VM_VM_VERSION_X86_HPP 26 #define CPU_X86_VM_VM_VERSION_X86_HPP 27 28 #include "runtime/globals_extension.hpp" 29 #include "runtime/vm_version.hpp" 30 31 class VM_Version : public Abstract_VM_Version { 32 friend class VMStructs; 33 friend class JVMCIVMStructs; 34 35 public: 36 // cpuid result register layouts. These are all unions of a uint32_t 37 // (in case anyone wants access to the register as a whole) and a bitfield. 38 39 union StdCpuid1Eax { 40 uint32_t value; 41 struct { 42 uint32_t stepping : 4, 43 model : 4, 44 family : 4, 45 proc_type : 2, 46 : 2, 47 ext_model : 4, 48 ext_family : 8, 49 : 4; 50 } bits; 51 }; 52 53 union StdCpuid1Ebx { // example, unused 54 uint32_t value; 55 struct { 56 uint32_t brand_id : 8, 57 clflush_size : 8, 58 threads_per_cpu : 8, 59 apic_id : 8; 60 } bits; 61 }; 62 63 union StdCpuid1Ecx { 64 uint32_t value; 65 struct { 66 uint32_t sse3 : 1, 67 clmul : 1, 68 : 1, 69 monitor : 1, 70 : 1, 71 vmx : 1, 72 : 1, 73 est : 1, 74 : 1, 75 ssse3 : 1, 76 cid : 1, 77 : 1, 78 fma : 1, 79 cmpxchg16: 1, 80 : 4, 81 dca : 1, 82 sse4_1 : 1, 83 sse4_2 : 1, 84 : 2, 85 popcnt : 1, 86 : 1, 87 aes : 1, 88 : 1, 89 osxsave : 1, 90 avx : 1, 91 : 3; 92 } bits; 93 }; 94 95 union StdCpuid1Edx { 96 uint32_t value; 97 struct { 98 uint32_t : 4, 99 tsc : 1, 100 : 3, 101 cmpxchg8 : 1, 102 : 6, 103 cmov : 1, 104 : 3, 105 clflush : 1, 106 : 3, 107 mmx : 1, 108 fxsr : 1, 109 sse : 1, 110 sse2 : 1, 111 : 1, 112 ht : 1, 113 : 3; 114 } bits; 115 }; 116 117 union DcpCpuid4Eax { 118 uint32_t value; 119 struct { 120 uint32_t cache_type : 5, 121 : 21, 122 cores_per_cpu : 6; 123 } bits; 124 }; 125 126 union DcpCpuid4Ebx { 127 uint32_t value; 128 struct { 129 uint32_t L1_line_size : 12, 130 partitions : 10, 131 associativity : 10; 132 } bits; 133 }; 134 135 union TplCpuidBEbx { 136 uint32_t value; 137 struct { 138 uint32_t logical_cpus : 16, 139 : 16; 140 } bits; 141 }; 142 143 union ExtCpuid1Ecx { 144 uint32_t value; 145 struct { 146 uint32_t LahfSahf : 1, 147 CmpLegacy : 1, 148 : 3, 149 lzcnt_intel : 1, 150 lzcnt : 1, 151 sse4a : 1, 152 misalignsse : 1, 153 prefetchw : 1, 154 : 22; 155 } bits; 156 }; 157 158 union ExtCpuid1Edx { 159 uint32_t value; 160 struct { 161 uint32_t : 22, 162 mmx_amd : 1, 163 mmx : 1, 164 fxsr : 1, 165 : 4, 166 long_mode : 1, 167 tdnow2 : 1, 168 tdnow : 1; 169 } bits; 170 }; 171 172 union ExtCpuid5Ex { 173 uint32_t value; 174 struct { 175 uint32_t L1_line_size : 8, 176 L1_tag_lines : 8, 177 L1_assoc : 8, 178 L1_size : 8; 179 } bits; 180 }; 181 182 union ExtCpuid7Edx { 183 uint32_t value; 184 struct { 185 uint32_t : 8, 186 tsc_invariance : 1, 187 : 23; 188 } bits; 189 }; 190 191 union ExtCpuid8Ecx { 192 uint32_t value; 193 struct { 194 uint32_t cores_per_cpu : 8, 195 : 24; 196 } bits; 197 }; 198 199 union SefCpuid7Eax { 200 uint32_t value; 201 }; 202 203 union SefCpuid7Ebx { 204 uint32_t value; 205 struct { 206 uint32_t fsgsbase : 1, 207 : 2, 208 bmi1 : 1, 209 : 1, 210 avx2 : 1, 211 : 2, 212 bmi2 : 1, 213 erms : 1, 214 : 1, 215 rtm : 1, 216 : 4, 217 avx512f : 1, 218 avx512dq : 1, 219 : 1, 220 adx : 1, 221 : 6, 222 avx512pf : 1, 223 avx512er : 1, 224 avx512cd : 1, 225 sha : 1, 226 avx512bw : 1, 227 avx512vl : 1; 228 } bits; 229 }; 230 231 union SefCpuid7Ecx { 232 uint32_t value; 233 struct { 234 uint32_t prefetchwt1 : 1, 235 avx512_vbmi : 1, 236 umip : 1, 237 pku : 1, 238 ospke : 1, 239 : 1, 240 avx512_vbmi2 : 1, 241 : 1, 242 gfni : 1, 243 vaes : 1, 244 vpclmulqdq : 1, 245 avx512_vnni : 1, 246 avx512_bitalg : 1, 247 : 1, 248 avx512_vpopcntdq : 1, 249 : 17; 250 } bits; 251 }; 252 253 union SefCpuid7Edx { 254 uint32_t value; 255 struct { 256 uint32_t : 2, 257 avx512_4vnniw : 1, 258 avx512_4fmaps : 1, 259 : 28; 260 } bits; 261 }; 262 263 union ExtCpuid1EEbx { 264 uint32_t value; 265 struct { 266 uint32_t : 8, 267 threads_per_core : 8, 268 : 16; 269 } bits; 270 }; 271 272 union XemXcr0Eax { 273 uint32_t value; 274 struct { 275 uint32_t x87 : 1, 276 sse : 1, 277 ymm : 1, 278 bndregs : 1, 279 bndcsr : 1, 280 opmask : 1, 281 zmm512 : 1, 282 zmm32 : 1, 283 : 24; 284 } bits; 285 }; 286 287 protected: 288 static int _cpu; 289 static int _model; 290 static int _stepping; 291 292 static address _cpuinfo_segv_addr; // address of instruction which causes SEGV 293 static address _cpuinfo_cont_addr; // address of instruction after the one which causes SEGV 294 295 enum Feature_Flag { 296 CPU_CX8 = (1 << 0), // next bits are from cpuid 1 (EDX) 297 CPU_CMOV = (1 << 1), 298 CPU_FXSR = (1 << 2), 299 CPU_HT = (1 << 3), 300 CPU_MMX = (1 << 4), 301 CPU_3DNOW_PREFETCH = (1 << 5), // Processor supports 3dnow prefetch and prefetchw instructions 302 // may not necessarily support other 3dnow instructions 303 CPU_SSE = (1 << 6), 304 CPU_SSE2 = (1 << 7), 305 CPU_SSE3 = (1 << 8), // SSE3 comes from cpuid 1 (ECX) 306 CPU_SSSE3 = (1 << 9), 307 CPU_SSE4A = (1 << 10), 308 CPU_SSE4_1 = (1 << 11), 309 CPU_SSE4_2 = (1 << 12), 310 CPU_POPCNT = (1 << 13), 311 CPU_LZCNT = (1 << 14), 312 CPU_TSC = (1 << 15), 313 CPU_TSCINV = (1 << 16), 314 CPU_AVX = (1 << 17), 315 CPU_AVX2 = (1 << 18), 316 CPU_AES = (1 << 19), 317 CPU_ERMS = (1 << 20), // enhanced 'rep movsb/stosb' instructions 318 CPU_CLMUL = (1 << 21), // carryless multiply for CRC 319 CPU_BMI1 = (1 << 22), 320 CPU_BMI2 = (1 << 23), 321 CPU_RTM = (1 << 24), // Restricted Transactional Memory instructions 322 CPU_ADX = (1 << 25), 323 CPU_AVX512F = (1 << 26), // AVX 512bit foundation instructions 324 CPU_AVX512DQ = (1 << 27), 325 CPU_AVX512PF = (1 << 28), 326 CPU_AVX512ER = (1 << 29), 327 CPU_AVX512CD = (1 << 30) 328 // Keeping sign bit 31 unassigned. 329 }; 330 331 #define CPU_AVX512BW ((uint64_t)UCONST64(0x100000000)) // enums are limited to 31 bit 332 #define CPU_AVX512VL ((uint64_t)UCONST64(0x200000000)) // EVEX instructions with smaller vector length 333 #define CPU_SHA ((uint64_t)UCONST64(0x400000000)) // SHA instructions 334 #define CPU_FMA ((uint64_t)UCONST64(0x800000000)) // FMA instructions 335 #define CPU_VZEROUPPER ((uint64_t)UCONST64(0x1000000000)) // Vzeroupper instruction 336 #define CPU_AVX512_VPOPCNTDQ ((uint64_t)UCONST64(0x2000000000)) // Vector popcount 337 #define CPU_VPCLMULQDQ ((uint64_t)UCONST64(0x4000000000)) //Vector carryless multiplication 338 #define CPU_VAES ((uint64_t)UCONST64(0x8000000000)) // Vector AES instructions 339 340 enum Extended_Family { 341 // AMD 342 CPU_FAMILY_AMD_11H = 0x11, 343 // ZX 344 CPU_FAMILY_ZX_CORE_F6 = 6, 345 CPU_FAMILY_ZX_CORE_F7 = 7, 346 // Intel 347 CPU_FAMILY_INTEL_CORE = 6, 348 CPU_MODEL_NEHALEM = 0x1e, 349 CPU_MODEL_NEHALEM_EP = 0x1a, 350 CPU_MODEL_NEHALEM_EX = 0x2e, 351 CPU_MODEL_WESTMERE = 0x25, 352 CPU_MODEL_WESTMERE_EP = 0x2c, 353 CPU_MODEL_WESTMERE_EX = 0x2f, 354 CPU_MODEL_SANDYBRIDGE = 0x2a, 355 CPU_MODEL_SANDYBRIDGE_EP = 0x2d, 356 CPU_MODEL_IVYBRIDGE_EP = 0x3a, 357 CPU_MODEL_HASWELL_E3 = 0x3c, 358 CPU_MODEL_HASWELL_E7 = 0x3f, 359 CPU_MODEL_BROADWELL = 0x3d, 360 CPU_MODEL_SKYLAKE = CPU_MODEL_HASWELL_E3 361 }; 362 363 // cpuid information block. All info derived from executing cpuid with 364 // various function numbers is stored here. Intel and AMD info is 365 // merged in this block: accessor methods disentangle it. 366 // 367 // The info block is laid out in subblocks of 4 dwords corresponding to 368 // eax, ebx, ecx and edx, whether or not they contain anything useful. 369 struct CpuidInfo { 370 // cpuid function 0 371 uint32_t std_max_function; 372 uint32_t std_vendor_name_0; 373 uint32_t std_vendor_name_1; 374 uint32_t std_vendor_name_2; 375 376 // cpuid function 1 377 StdCpuid1Eax std_cpuid1_eax; 378 StdCpuid1Ebx std_cpuid1_ebx; 379 StdCpuid1Ecx std_cpuid1_ecx; 380 StdCpuid1Edx std_cpuid1_edx; 381 382 // cpuid function 4 (deterministic cache parameters) 383 DcpCpuid4Eax dcp_cpuid4_eax; 384 DcpCpuid4Ebx dcp_cpuid4_ebx; 385 uint32_t dcp_cpuid4_ecx; // unused currently 386 uint32_t dcp_cpuid4_edx; // unused currently 387 388 // cpuid function 7 (structured extended features) 389 SefCpuid7Eax sef_cpuid7_eax; 390 SefCpuid7Ebx sef_cpuid7_ebx; 391 SefCpuid7Ecx sef_cpuid7_ecx; 392 SefCpuid7Edx sef_cpuid7_edx; 393 394 // cpuid function 0xB (processor topology) 395 // ecx = 0 396 uint32_t tpl_cpuidB0_eax; 397 TplCpuidBEbx tpl_cpuidB0_ebx; 398 uint32_t tpl_cpuidB0_ecx; // unused currently 399 uint32_t tpl_cpuidB0_edx; // unused currently 400 401 // ecx = 1 402 uint32_t tpl_cpuidB1_eax; 403 TplCpuidBEbx tpl_cpuidB1_ebx; 404 uint32_t tpl_cpuidB1_ecx; // unused currently 405 uint32_t tpl_cpuidB1_edx; // unused currently 406 407 // ecx = 2 408 uint32_t tpl_cpuidB2_eax; 409 TplCpuidBEbx tpl_cpuidB2_ebx; 410 uint32_t tpl_cpuidB2_ecx; // unused currently 411 uint32_t tpl_cpuidB2_edx; // unused currently 412 413 // cpuid function 0x80000000 // example, unused 414 uint32_t ext_max_function; 415 uint32_t ext_vendor_name_0; 416 uint32_t ext_vendor_name_1; 417 uint32_t ext_vendor_name_2; 418 419 // cpuid function 0x80000001 420 uint32_t ext_cpuid1_eax; // reserved 421 uint32_t ext_cpuid1_ebx; // reserved 422 ExtCpuid1Ecx ext_cpuid1_ecx; 423 ExtCpuid1Edx ext_cpuid1_edx; 424 425 // cpuid functions 0x80000002 thru 0x80000004: example, unused 426 uint32_t proc_name_0, proc_name_1, proc_name_2, proc_name_3; 427 uint32_t proc_name_4, proc_name_5, proc_name_6, proc_name_7; 428 uint32_t proc_name_8, proc_name_9, proc_name_10,proc_name_11; 429 430 // cpuid function 0x80000005 // AMD L1, Intel reserved 431 uint32_t ext_cpuid5_eax; // unused currently 432 uint32_t ext_cpuid5_ebx; // reserved 433 ExtCpuid5Ex ext_cpuid5_ecx; // L1 data cache info (AMD) 434 ExtCpuid5Ex ext_cpuid5_edx; // L1 instruction cache info (AMD) 435 436 // cpuid function 0x80000007 437 uint32_t ext_cpuid7_eax; // reserved 438 uint32_t ext_cpuid7_ebx; // reserved 439 uint32_t ext_cpuid7_ecx; // reserved 440 ExtCpuid7Edx ext_cpuid7_edx; // tscinv 441 442 // cpuid function 0x80000008 443 uint32_t ext_cpuid8_eax; // unused currently 444 uint32_t ext_cpuid8_ebx; // reserved 445 ExtCpuid8Ecx ext_cpuid8_ecx; 446 uint32_t ext_cpuid8_edx; // reserved 447 448 // cpuid function 0x8000001E // AMD 17h 449 uint32_t ext_cpuid1E_eax; 450 ExtCpuid1EEbx ext_cpuid1E_ebx; // threads per core (AMD17h) 451 uint32_t ext_cpuid1E_ecx; 452 uint32_t ext_cpuid1E_edx; // unused currently 453 454 // extended control register XCR0 (the XFEATURE_ENABLED_MASK register) 455 XemXcr0Eax xem_xcr0_eax; 456 uint32_t xem_xcr0_edx; // reserved 457 458 // Space to save ymm registers after signal handle 459 int ymm_save[8*4]; // Save ymm0, ymm7, ymm8, ymm15 460 461 // Space to save zmm registers after signal handle 462 int zmm_save[16*4]; // Save zmm0, zmm7, zmm8, zmm31 463 }; 464 465 // The actual cpuid info block 466 static CpuidInfo _cpuid_info; 467 468 // Extractors and predicates 469 static uint32_t extended_cpu_family() { 470 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.family; 471 result += _cpuid_info.std_cpuid1_eax.bits.ext_family; 472 return result; 473 } 474 475 static uint32_t extended_cpu_model() { 476 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.model; 477 result |= _cpuid_info.std_cpuid1_eax.bits.ext_model << 4; 478 return result; 479 } 480 481 static uint32_t cpu_stepping() { 482 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.stepping; 483 return result; 484 } 485 486 static uint logical_processor_count() { 487 uint result = threads_per_core(); 488 return result; 489 } 490 491 static uint64_t feature_flags() { 492 uint64_t result = 0; 493 if (_cpuid_info.std_cpuid1_edx.bits.cmpxchg8 != 0) 494 result |= CPU_CX8; 495 if (_cpuid_info.std_cpuid1_edx.bits.cmov != 0) 496 result |= CPU_CMOV; 497 if (_cpuid_info.std_cpuid1_edx.bits.fxsr != 0 || (is_amd() && 498 _cpuid_info.ext_cpuid1_edx.bits.fxsr != 0)) 499 result |= CPU_FXSR; 500 // HT flag is set for multi-core processors also. 501 if (threads_per_core() > 1) 502 result |= CPU_HT; 503 if (_cpuid_info.std_cpuid1_edx.bits.mmx != 0 || (is_amd() && 504 _cpuid_info.ext_cpuid1_edx.bits.mmx != 0)) 505 result |= CPU_MMX; 506 if (_cpuid_info.std_cpuid1_edx.bits.sse != 0) 507 result |= CPU_SSE; 508 if (_cpuid_info.std_cpuid1_edx.bits.sse2 != 0) 509 result |= CPU_SSE2; 510 if (_cpuid_info.std_cpuid1_ecx.bits.sse3 != 0) 511 result |= CPU_SSE3; 512 if (_cpuid_info.std_cpuid1_ecx.bits.ssse3 != 0) 513 result |= CPU_SSSE3; 514 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_1 != 0) 515 result |= CPU_SSE4_1; 516 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_2 != 0) 517 result |= CPU_SSE4_2; 518 if (_cpuid_info.std_cpuid1_ecx.bits.popcnt != 0) 519 result |= CPU_POPCNT; 520 if (_cpuid_info.std_cpuid1_ecx.bits.avx != 0 && 521 _cpuid_info.std_cpuid1_ecx.bits.osxsave != 0 && 522 _cpuid_info.xem_xcr0_eax.bits.sse != 0 && 523 _cpuid_info.xem_xcr0_eax.bits.ymm != 0) { 524 result |= CPU_AVX; 525 result |= CPU_VZEROUPPER; 526 if (_cpuid_info.sef_cpuid7_ebx.bits.avx2 != 0) 527 result |= CPU_AVX2; 528 if (_cpuid_info.sef_cpuid7_ebx.bits.avx512f != 0 && 529 _cpuid_info.xem_xcr0_eax.bits.opmask != 0 && 530 _cpuid_info.xem_xcr0_eax.bits.zmm512 != 0 && 531 _cpuid_info.xem_xcr0_eax.bits.zmm32 != 0) { 532 result |= CPU_AVX512F; 533 if (_cpuid_info.sef_cpuid7_ebx.bits.avx512cd != 0) 534 result |= CPU_AVX512CD; 535 if (_cpuid_info.sef_cpuid7_ebx.bits.avx512dq != 0) 536 result |= CPU_AVX512DQ; 537 if (_cpuid_info.sef_cpuid7_ebx.bits.avx512pf != 0) 538 result |= CPU_AVX512PF; 539 if (_cpuid_info.sef_cpuid7_ebx.bits.avx512er != 0) 540 result |= CPU_AVX512ER; 541 if (_cpuid_info.sef_cpuid7_ebx.bits.avx512bw != 0) 542 result |= CPU_AVX512BW; 543 if (_cpuid_info.sef_cpuid7_ebx.bits.avx512vl != 0) 544 result |= CPU_AVX512VL; 545 if (_cpuid_info.sef_cpuid7_ecx.bits.avx512_vpopcntdq != 0) 546 result |= CPU_AVX512_VPOPCNTDQ; 547 if (_cpuid_info.sef_cpuid7_ecx.bits.vpclmulqdq != 0) 548 result |= CPU_VPCLMULQDQ; 549 if (_cpuid_info.sef_cpuid7_ecx.bits.vaes != 0) 550 result |= CPU_VAES; 551 } 552 } 553 if(_cpuid_info.sef_cpuid7_ebx.bits.bmi1 != 0) 554 result |= CPU_BMI1; 555 if (_cpuid_info.std_cpuid1_edx.bits.tsc != 0) 556 result |= CPU_TSC; 557 if (_cpuid_info.ext_cpuid7_edx.bits.tsc_invariance != 0) 558 result |= CPU_TSCINV; 559 if (_cpuid_info.std_cpuid1_ecx.bits.aes != 0) 560 result |= CPU_AES; 561 if (_cpuid_info.sef_cpuid7_ebx.bits.erms != 0) 562 result |= CPU_ERMS; 563 if (_cpuid_info.std_cpuid1_ecx.bits.clmul != 0) 564 result |= CPU_CLMUL; 565 if (_cpuid_info.sef_cpuid7_ebx.bits.rtm != 0) 566 result |= CPU_RTM; 567 if(_cpuid_info.sef_cpuid7_ebx.bits.adx != 0) 568 result |= CPU_ADX; 569 if(_cpuid_info.sef_cpuid7_ebx.bits.bmi2 != 0) 570 result |= CPU_BMI2; 571 if (_cpuid_info.sef_cpuid7_ebx.bits.sha != 0) 572 result |= CPU_SHA; 573 if (_cpuid_info.std_cpuid1_ecx.bits.fma != 0) 574 result |= CPU_FMA; 575 576 // AMD features. 577 if (is_amd()) { 578 if ((_cpuid_info.ext_cpuid1_edx.bits.tdnow != 0) || 579 (_cpuid_info.ext_cpuid1_ecx.bits.prefetchw != 0)) 580 result |= CPU_3DNOW_PREFETCH; 581 if (_cpuid_info.ext_cpuid1_ecx.bits.lzcnt != 0) 582 result |= CPU_LZCNT; 583 if (_cpuid_info.ext_cpuid1_ecx.bits.sse4a != 0) 584 result |= CPU_SSE4A; 585 } 586 // Intel features. 587 if(is_intel()) { 588 if(_cpuid_info.ext_cpuid1_ecx.bits.lzcnt_intel != 0) 589 result |= CPU_LZCNT; 590 // for Intel, ecx.bits.misalignsse bit (bit 8) indicates support for prefetchw 591 if (_cpuid_info.ext_cpuid1_ecx.bits.misalignsse != 0) { 592 result |= CPU_3DNOW_PREFETCH; 593 } 594 } 595 596 // ZX features. 597 if (is_zx()) { 598 if (_cpuid_info.ext_cpuid1_ecx.bits.lzcnt_intel != 0) 599 result |= CPU_LZCNT; 600 // for ZX, ecx.bits.misalignsse bit (bit 8) indicates support for prefetchw 601 if (_cpuid_info.ext_cpuid1_ecx.bits.misalignsse != 0) { 602 result |= CPU_3DNOW_PREFETCH; 603 } 604 } 605 606 return result; 607 } 608 609 static bool os_supports_avx_vectors() { 610 bool retVal = false; 611 if (supports_evex()) { 612 // Verify that OS save/restore all bits of EVEX registers 613 // during signal processing. 614 int nreg = 2 LP64_ONLY(+2); 615 retVal = true; 616 for (int i = 0; i < 16 * nreg; i++) { // 64 bytes per zmm register 617 if (_cpuid_info.zmm_save[i] != ymm_test_value()) { 618 retVal = false; 619 break; 620 } 621 } 622 } else if (supports_avx()) { 623 // Verify that OS save/restore all bits of AVX registers 624 // during signal processing. 625 int nreg = 2 LP64_ONLY(+2); 626 retVal = true; 627 for (int i = 0; i < 8 * nreg; i++) { // 32 bytes per ymm register 628 if (_cpuid_info.ymm_save[i] != ymm_test_value()) { 629 retVal = false; 630 break; 631 } 632 } 633 // zmm_save will be set on a EVEX enabled machine even if we choose AVX code gen 634 if (retVal == false) { 635 // Verify that OS save/restore all bits of EVEX registers 636 // during signal processing. 637 int nreg = 2 LP64_ONLY(+2); 638 retVal = true; 639 for (int i = 0; i < 16 * nreg; i++) { // 64 bytes per zmm register 640 if (_cpuid_info.zmm_save[i] != ymm_test_value()) { 641 retVal = false; 642 break; 643 } 644 } 645 } 646 } 647 return retVal; 648 } 649 650 static void get_processor_features(); 651 652 public: 653 // Offsets for cpuid asm stub 654 static ByteSize std_cpuid0_offset() { return byte_offset_of(CpuidInfo, std_max_function); } 655 static ByteSize std_cpuid1_offset() { return byte_offset_of(CpuidInfo, std_cpuid1_eax); } 656 static ByteSize dcp_cpuid4_offset() { return byte_offset_of(CpuidInfo, dcp_cpuid4_eax); } 657 static ByteSize sef_cpuid7_offset() { return byte_offset_of(CpuidInfo, sef_cpuid7_eax); } 658 static ByteSize ext_cpuid1_offset() { return byte_offset_of(CpuidInfo, ext_cpuid1_eax); } 659 static ByteSize ext_cpuid5_offset() { return byte_offset_of(CpuidInfo, ext_cpuid5_eax); } 660 static ByteSize ext_cpuid7_offset() { return byte_offset_of(CpuidInfo, ext_cpuid7_eax); } 661 static ByteSize ext_cpuid8_offset() { return byte_offset_of(CpuidInfo, ext_cpuid8_eax); } 662 static ByteSize ext_cpuid1E_offset() { return byte_offset_of(CpuidInfo, ext_cpuid1E_eax); } 663 static ByteSize tpl_cpuidB0_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB0_eax); } 664 static ByteSize tpl_cpuidB1_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB1_eax); } 665 static ByteSize tpl_cpuidB2_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB2_eax); } 666 static ByteSize xem_xcr0_offset() { return byte_offset_of(CpuidInfo, xem_xcr0_eax); } 667 static ByteSize ymm_save_offset() { return byte_offset_of(CpuidInfo, ymm_save); } 668 static ByteSize zmm_save_offset() { return byte_offset_of(CpuidInfo, zmm_save); } 669 670 // The value used to check ymm register after signal handle 671 static int ymm_test_value() { return 0xCAFEBABE; } 672 673 static void get_cpu_info_wrapper(); 674 static void set_cpuinfo_segv_addr(address pc) { _cpuinfo_segv_addr = pc; } 675 static bool is_cpuinfo_segv_addr(address pc) { return _cpuinfo_segv_addr == pc; } 676 static void set_cpuinfo_cont_addr(address pc) { _cpuinfo_cont_addr = pc; } 677 static address cpuinfo_cont_addr() { return _cpuinfo_cont_addr; } 678 679 static void clean_cpuFeatures() { _features = 0; } 680 static void set_avx_cpuFeatures() { _features = (CPU_SSE | CPU_SSE2 | CPU_AVX | CPU_VZEROUPPER ); } 681 static void set_evex_cpuFeatures() { _features = (CPU_AVX512F | CPU_SSE | CPU_SSE2 | CPU_VZEROUPPER ); } 682 683 684 // Initialization 685 static void initialize(); 686 687 // Override Abstract_VM_Version implementation 688 static bool use_biased_locking(); 689 690 // Asserts 691 static void assert_is_initialized() { 692 assert(_cpuid_info.std_cpuid1_eax.bits.family != 0, "VM_Version not initialized"); 693 } 694 695 // 696 // Processor family: 697 // 3 - 386 698 // 4 - 486 699 // 5 - Pentium 700 // 6 - PentiumPro, Pentium II, Celeron, Xeon, Pentium III, Athlon, 701 // Pentium M, Core Solo, Core Duo, Core2 Duo 702 // family 6 model: 9, 13, 14, 15 703 // 0x0f - Pentium 4, Opteron 704 // 705 // Note: The cpu family should be used to select between 706 // instruction sequences which are valid on all Intel 707 // processors. Use the feature test functions below to 708 // determine whether a particular instruction is supported. 709 // 710 static int cpu_family() { return _cpu;} 711 static bool is_P6() { return cpu_family() >= 6; } 712 static bool is_amd() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x68747541; } // 'htuA' 713 static bool is_intel() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x756e6547; } // 'uneG' 714 static bool is_zx() { assert_is_initialized(); return (_cpuid_info.std_vendor_name_0 == 0x746e6543) || (_cpuid_info.std_vendor_name_0 == 0x68532020); } // 'tneC'||'hS ' 715 static bool is_atom_family() { return ((cpu_family() == 0x06) && ((extended_cpu_model() == 0x36) || (extended_cpu_model() == 0x37) || (extended_cpu_model() == 0x4D))); } //Silvermont and Centerton 716 static bool is_knights_family() { return ((cpu_family() == 0x06) && ((extended_cpu_model() == 0x57) || (extended_cpu_model() == 0x85))); } // Xeon Phi 3200/5200/7200 and Future Xeon Phi 717 718 static bool supports_processor_topology() { 719 return (_cpuid_info.std_max_function >= 0xB) && 720 // eax[4:0] | ebx[0:15] == 0 indicates invalid topology level. 721 // Some cpus have max cpuid >= 0xB but do not support processor topology. 722 (((_cpuid_info.tpl_cpuidB0_eax & 0x1f) | _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus) != 0); 723 } 724 725 static uint cores_per_cpu() { 726 uint result = 1; 727 if (is_intel()) { 728 bool supports_topology = supports_processor_topology(); 729 if (supports_topology) { 730 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus / 731 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 732 } 733 if (!supports_topology || result == 0) { 734 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1); 735 } 736 } else if (is_amd()) { 737 result = (_cpuid_info.ext_cpuid8_ecx.bits.cores_per_cpu + 1); 738 } else if (is_zx()) { 739 bool supports_topology = supports_processor_topology(); 740 if (supports_topology) { 741 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus / 742 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 743 } 744 if (!supports_topology || result == 0) { 745 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1); 746 } 747 } 748 return result; 749 } 750 751 static uint threads_per_core() { 752 uint result = 1; 753 if (is_intel() && supports_processor_topology()) { 754 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 755 } else if (is_zx() && supports_processor_topology()) { 756 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 757 } else if (_cpuid_info.std_cpuid1_edx.bits.ht != 0) { 758 if (cpu_family() >= 0x17) { 759 result = _cpuid_info.ext_cpuid1E_ebx.bits.threads_per_core + 1; 760 } else { 761 result = _cpuid_info.std_cpuid1_ebx.bits.threads_per_cpu / 762 cores_per_cpu(); 763 } 764 } 765 return (result == 0 ? 1 : result); 766 } 767 768 static intx L1_line_size() { 769 intx result = 0; 770 if (is_intel()) { 771 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1); 772 } else if (is_amd()) { 773 result = _cpuid_info.ext_cpuid5_ecx.bits.L1_line_size; 774 } else if (is_zx()) { 775 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1); 776 } 777 if (result < 32) // not defined ? 778 result = 32; // 32 bytes by default on x86 and other x64 779 return result; 780 } 781 782 static intx prefetch_data_size() { 783 return L1_line_size(); 784 } 785 786 // 787 // Feature identification 788 // 789 static bool supports_cpuid() { return _features != 0; } 790 static bool supports_cmpxchg8() { return (_features & CPU_CX8) != 0; } 791 static bool supports_cmov() { return (_features & CPU_CMOV) != 0; } 792 static bool supports_fxsr() { return (_features & CPU_FXSR) != 0; } 793 static bool supports_ht() { return (_features & CPU_HT) != 0; } 794 static bool supports_mmx() { return (_features & CPU_MMX) != 0; } 795 static bool supports_sse() { return (_features & CPU_SSE) != 0; } 796 static bool supports_sse2() { return (_features & CPU_SSE2) != 0; } 797 static bool supports_sse3() { return (_features & CPU_SSE3) != 0; } 798 static bool supports_ssse3() { return (_features & CPU_SSSE3)!= 0; } 799 static bool supports_sse4_1() { return (_features & CPU_SSE4_1) != 0; } 800 static bool supports_sse4_2() { return (_features & CPU_SSE4_2) != 0; } 801 static bool supports_popcnt() { return (_features & CPU_POPCNT) != 0; } 802 static bool supports_avx() { return (_features & CPU_AVX) != 0; } 803 static bool supports_avx2() { return (_features & CPU_AVX2) != 0; } 804 static bool supports_tsc() { return (_features & CPU_TSC) != 0; } 805 static bool supports_aes() { return (_features & CPU_AES) != 0; } 806 static bool supports_erms() { return (_features & CPU_ERMS) != 0; } 807 static bool supports_clmul() { return (_features & CPU_CLMUL) != 0; } 808 static bool supports_rtm() { return (_features & CPU_RTM) != 0; } 809 static bool supports_bmi1() { return (_features & CPU_BMI1) != 0; } 810 static bool supports_bmi2() { return (_features & CPU_BMI2) != 0; } 811 static bool supports_adx() { return (_features & CPU_ADX) != 0; } 812 static bool supports_evex() { return (_features & CPU_AVX512F) != 0; } 813 static bool supports_avx512dq() { return (_features & CPU_AVX512DQ) != 0; } 814 static bool supports_avx512pf() { return (_features & CPU_AVX512PF) != 0; } 815 static bool supports_avx512er() { return (_features & CPU_AVX512ER) != 0; } 816 static bool supports_avx512cd() { return (_features & CPU_AVX512CD) != 0; } 817 static bool supports_avx512bw() { return (_features & CPU_AVX512BW) != 0; } 818 static bool supports_avx512vl() { return (_features & CPU_AVX512VL) != 0; } 819 static bool supports_avx512vlbw() { return (supports_avx512bw() && supports_avx512vl()); } 820 static bool supports_avx512novl() { return (supports_evex() && !supports_avx512vl()); } 821 static bool supports_avx512nobw() { return (supports_evex() && !supports_avx512bw()); } 822 static bool supports_avx256only() { return (supports_avx2() && !supports_evex()); } 823 static bool supports_avxonly() { return ((supports_avx2() || supports_avx()) && !supports_evex()); } 824 static bool supports_sha() { return (_features & CPU_SHA) != 0; } 825 static bool supports_fma() { return (_features & CPU_FMA) != 0 && supports_avx(); } 826 static bool supports_vzeroupper() { return (_features & CPU_VZEROUPPER) != 0; } 827 static bool supports_vpopcntdq() { return (_features & CPU_AVX512_VPOPCNTDQ) != 0; } 828 static bool supports_vpclmulqdq() { return (_features & CPU_VPCLMULQDQ) != 0; } 829 static bool supports_vaes() { return (_features & CPU_VAES) != 0; } 830 831 // Intel features 832 static bool is_intel_family_core() { return is_intel() && 833 extended_cpu_family() == CPU_FAMILY_INTEL_CORE; } 834 835 static bool is_intel_tsc_synched_at_init() { 836 if (is_intel_family_core()) { 837 uint32_t ext_model = extended_cpu_model(); 838 if (ext_model == CPU_MODEL_NEHALEM_EP || 839 ext_model == CPU_MODEL_WESTMERE_EP || 840 ext_model == CPU_MODEL_SANDYBRIDGE_EP || 841 ext_model == CPU_MODEL_IVYBRIDGE_EP) { 842 // <= 2-socket invariant tsc support. EX versions are usually used 843 // in > 2-socket systems and likely don't synchronize tscs at 844 // initialization. 845 // Code that uses tsc values must be prepared for them to arbitrarily 846 // jump forward or backward. 847 return true; 848 } 849 } 850 return false; 851 } 852 853 // AMD features 854 static bool supports_3dnow_prefetch() { return (_features & CPU_3DNOW_PREFETCH) != 0; } 855 static bool supports_mmx_ext() { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.mmx_amd != 0; } 856 static bool supports_lzcnt() { return (_features & CPU_LZCNT) != 0; } 857 static bool supports_sse4a() { return (_features & CPU_SSE4A) != 0; } 858 859 static bool is_amd_Barcelona() { return is_amd() && 860 extended_cpu_family() == CPU_FAMILY_AMD_11H; } 861 862 // Intel and AMD newer cores support fast timestamps well 863 static bool supports_tscinv_bit() { 864 return (_features & CPU_TSCINV) != 0; 865 } 866 static bool supports_tscinv() { 867 return supports_tscinv_bit() && 868 ( (is_amd() && !is_amd_Barcelona()) || 869 is_intel_tsc_synched_at_init() ); 870 } 871 872 // Intel Core and newer cpus have fast IDIV instruction (excluding Atom). 873 static bool has_fast_idiv() { return is_intel() && cpu_family() == 6 && 874 supports_sse3() && _model != 0x1C; } 875 876 static bool supports_compare_and_exchange() { return true; } 877 878 static intx allocate_prefetch_distance(bool use_watermark_prefetch) { 879 // Hardware prefetching (distance/size in bytes): 880 // Pentium 3 - 64 / 32 881 // Pentium 4 - 256 / 128 882 // Athlon - 64 / 32 ???? 883 // Opteron - 128 / 64 only when 2 sequential cache lines accessed 884 // Core - 128 / 64 885 // 886 // Software prefetching (distance in bytes / instruction with best score): 887 // Pentium 3 - 128 / prefetchnta 888 // Pentium 4 - 512 / prefetchnta 889 // Athlon - 128 / prefetchnta 890 // Opteron - 256 / prefetchnta 891 // Core - 256 / prefetchnta 892 // It will be used only when AllocatePrefetchStyle > 0 893 894 if (is_amd()) { // AMD 895 if (supports_sse2()) { 896 return 256; // Opteron 897 } else { 898 return 128; // Athlon 899 } 900 } else { // Intel 901 if (supports_sse3() && cpu_family() == 6) { 902 if (supports_sse4_2() && supports_ht()) { // Nehalem based cpus 903 return 192; 904 } else if (use_watermark_prefetch) { // watermark prefetching on Core 905 #ifdef _LP64 906 return 384; 907 #else 908 return 320; 909 #endif 910 } 911 } 912 if (supports_sse2()) { 913 if (cpu_family() == 6) { 914 return 256; // Pentium M, Core, Core2 915 } else { 916 return 512; // Pentium 4 917 } 918 } else { 919 return 128; // Pentium 3 (and all other old CPUs) 920 } 921 } 922 } 923 924 // SSE2 and later processors implement a 'pause' instruction 925 // that can be used for efficient implementation of 926 // the intrinsic for java.lang.Thread.onSpinWait() 927 static bool supports_on_spin_wait() { return supports_sse2(); } 928 }; 929 930 #endif // CPU_X86_VM_VM_VERSION_X86_HPP