1 /* 2 * Copyright (c) 1997, 2020, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2012, 2020 SAP SE. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 #include "precompiled.hpp" 27 #include "jvm.h" 28 #include "asm/assembler.inline.hpp" 29 #include "asm/macroAssembler.inline.hpp" 30 #include "compiler/disassembler.hpp" 31 #include "memory/resourceArea.hpp" 32 #include "runtime/java.hpp" 33 #include "runtime/os.hpp" 34 #include "runtime/stubCodeGenerator.hpp" 35 #include "runtime/vm_version.hpp" 36 #include "utilities/align.hpp" 37 #include "utilities/defaultStream.hpp" 38 #include "utilities/globalDefinitions.hpp" 39 #include "utilities/powerOfTwo.hpp" 40 41 #include <sys/sysinfo.h> 42 #if defined(_AIX) 43 #include <libperfstat.h> 44 #endif 45 46 #if defined(LINUX) && defined(VM_LITTLE_ENDIAN) 47 #include <sys/auxv.h> 48 49 #ifndef PPC_FEATURE2_HTM_NOSC 50 #define PPC_FEATURE2_HTM_NOSC (1 << 24) 51 #endif 52 #endif 53 54 bool VM_Version::_is_determine_features_test_running = false; 55 uint64_t VM_Version::_dscr_val = 0; 56 57 #define MSG(flag) \ 58 if (flag && !FLAG_IS_DEFAULT(flag)) \ 59 jio_fprintf(defaultStream::error_stream(), \ 60 "warning: -XX:+" #flag " requires -XX:+UseSIGTRAP\n" \ 61 " -XX:+" #flag " will be disabled!\n"); 62 63 void VM_Version::initialize() { 64 65 // Test which instructions are supported and measure cache line size. 66 determine_features(); 67 68 // If PowerArchitecturePPC64 hasn't been specified explicitly determine from features. 69 if (FLAG_IS_DEFAULT(PowerArchitecturePPC64)) { 70 if (VM_Version::has_darn()) { 71 FLAG_SET_ERGO(PowerArchitecturePPC64, 9); 72 } else if (VM_Version::has_lqarx()) { 73 FLAG_SET_ERGO(PowerArchitecturePPC64, 8); 74 } else if (VM_Version::has_popcntw()) { 75 FLAG_SET_ERGO(PowerArchitecturePPC64, 7); 76 } else if (VM_Version::has_cmpb()) { 77 FLAG_SET_ERGO(PowerArchitecturePPC64, 6); 78 } else if (VM_Version::has_popcntb()) { 79 FLAG_SET_ERGO(PowerArchitecturePPC64, 5); 80 } else { 81 FLAG_SET_ERGO(PowerArchitecturePPC64, 0); 82 } 83 } 84 85 bool PowerArchitecturePPC64_ok = false; 86 switch (PowerArchitecturePPC64) { 87 case 9: if (!VM_Version::has_darn() ) break; 88 case 8: if (!VM_Version::has_lqarx() ) break; 89 case 7: if (!VM_Version::has_popcntw()) break; 90 case 6: if (!VM_Version::has_cmpb() ) break; 91 case 5: if (!VM_Version::has_popcntb()) break; 92 case 0: PowerArchitecturePPC64_ok = true; break; 93 default: break; 94 } 95 guarantee(PowerArchitecturePPC64_ok, "PowerArchitecturePPC64 cannot be set to " 96 UINTX_FORMAT " on this machine", PowerArchitecturePPC64); 97 98 // Power 8: Configure Data Stream Control Register. 99 if (PowerArchitecturePPC64 >= 8 && has_mfdscr()) { 100 config_dscr(); 101 } 102 103 if (!UseSIGTRAP) { 104 MSG(TrapBasedICMissChecks); 105 MSG(TrapBasedNullChecks); 106 FLAG_SET_ERGO(TrapBasedNullChecks, false); 107 FLAG_SET_ERGO(TrapBasedICMissChecks, false); 108 } 109 110 #ifdef COMPILER2 111 if (!UseSIGTRAP) { 112 MSG(TrapBasedRangeChecks); 113 FLAG_SET_ERGO(TrapBasedRangeChecks, false); 114 } 115 116 // On Power6 test for section size. 117 if (PowerArchitecturePPC64 == 6) { 118 determine_section_size(); 119 // TODO: PPC port } else { 120 // TODO: PPC port PdScheduling::power6SectorSize = 0x20; 121 } 122 123 if (PowerArchitecturePPC64 >= 8) { 124 if (FLAG_IS_DEFAULT(SuperwordUseVSX)) { 125 FLAG_SET_ERGO(SuperwordUseVSX, true); 126 } 127 } else { 128 if (SuperwordUseVSX) { 129 warning("SuperwordUseVSX specified, but needs at least Power8."); 130 FLAG_SET_DEFAULT(SuperwordUseVSX, false); 131 } 132 } 133 MaxVectorSize = SuperwordUseVSX ? 16 : 8; 134 135 if (PowerArchitecturePPC64 >= 9) { 136 if (FLAG_IS_DEFAULT(UseCountTrailingZerosInstructionsPPC64)) { 137 FLAG_SET_ERGO(UseCountTrailingZerosInstructionsPPC64, true); 138 } 139 if (FLAG_IS_DEFAULT(UseCharacterCompareIntrinsics)) { 140 FLAG_SET_ERGO(UseCharacterCompareIntrinsics, true); 141 } 142 if (FLAG_IS_DEFAULT(UseVectorByteReverseInstructionsPPC64)) { 143 FLAG_SET_ERGO(UseVectorByteReverseInstructionsPPC64, true); 144 } 145 } else { 146 if (UseCountTrailingZerosInstructionsPPC64) { 147 warning("UseCountTrailingZerosInstructionsPPC64 specified, but needs at least Power9."); 148 FLAG_SET_DEFAULT(UseCountTrailingZerosInstructionsPPC64, false); 149 } 150 if (UseCharacterCompareIntrinsics) { 151 warning("UseCharacterCompareIntrinsics specified, but needs at least Power9."); 152 FLAG_SET_DEFAULT(UseCharacterCompareIntrinsics, false); 153 } 154 if (UseVectorByteReverseInstructionsPPC64) { 155 warning("UseVectorByteReverseInstructionsPPC64 specified, but needs at least Power9."); 156 FLAG_SET_DEFAULT(UseVectorByteReverseInstructionsPPC64, false); 157 } 158 } 159 #endif 160 161 // Create and print feature-string. 162 char buf[(num_features+1) * 16]; // Max 16 chars per feature. 163 jio_snprintf(buf, sizeof(buf), 164 "ppc64%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s", 165 (has_fsqrt() ? " fsqrt" : ""), 166 (has_isel() ? " isel" : ""), 167 (has_lxarxeh() ? " lxarxeh" : ""), 168 (has_cmpb() ? " cmpb" : ""), 169 (has_popcntb() ? " popcntb" : ""), 170 (has_popcntw() ? " popcntw" : ""), 171 (has_fcfids() ? " fcfids" : ""), 172 (has_vand() ? " vand" : ""), 173 (has_lqarx() ? " lqarx" : ""), 174 (has_vcipher() ? " aes" : ""), 175 (has_vpmsumb() ? " vpmsumb" : ""), 176 (has_mfdscr() ? " mfdscr" : ""), 177 (has_vsx() ? " vsx" : ""), 178 (has_ldbrx() ? " ldbrx" : ""), 179 (has_stdbrx() ? " stdbrx" : ""), 180 (has_vshasig() ? " sha" : ""), 181 (has_tm() ? " rtm" : ""), 182 (has_darn() ? " darn" : "") 183 // Make sure number of %s matches num_features! 184 ); 185 _features_string = os::strdup(buf); 186 if (Verbose) { 187 print_features(); 188 } 189 190 // PPC64 supports 8-byte compare-exchange operations (see Atomic::cmpxchg) 191 // and 'atomic long memory ops' (see Unsafe_GetLongVolatile). 192 _supports_cx8 = true; 193 194 // Used by C1. 195 _supports_atomic_getset4 = true; 196 _supports_atomic_getadd4 = true; 197 _supports_atomic_getset8 = true; 198 _supports_atomic_getadd8 = true; 199 200 intx cache_line_size = L1_data_cache_line_size(); 201 202 if (PowerArchitecturePPC64 >= 9) { 203 if (os::supports_map_sync() == true) { 204 _data_cache_line_flush_size = cache_line_size; 205 } 206 } 207 208 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1; 209 210 if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size; 211 // PPC processors have an automatic prefetch engine. 212 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 1; 213 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3 * cache_line_size; 214 215 assert(AllocatePrefetchLines > 0, "invalid value"); 216 if (AllocatePrefetchLines < 1) { // Set valid value in product VM. 217 AllocatePrefetchLines = 1; // Conservative value. 218 } 219 220 if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) { 221 AllocatePrefetchStyle = 1; // Fall back if inappropriate. 222 } 223 224 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive"); 225 226 if (FLAG_IS_DEFAULT(ContendedPaddingWidth) && (cache_line_size > ContendedPaddingWidth)) { 227 ContendedPaddingWidth = cache_line_size; 228 } 229 230 // If running on Power8 or newer hardware, the implementation uses the available vector instructions. 231 // In all other cases, the implementation uses only generally available instructions. 232 if (!UseCRC32Intrinsics) { 233 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) { 234 FLAG_SET_DEFAULT(UseCRC32Intrinsics, true); 235 } 236 } 237 238 // Implementation does not use any of the vector instructions available with Power8. 239 // Their exploitation is still pending (aka "work in progress"). 240 if (!UseCRC32CIntrinsics) { 241 if (FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) { 242 FLAG_SET_DEFAULT(UseCRC32CIntrinsics, true); 243 } 244 } 245 246 // TODO: Provide implementation. 247 if (UseAdler32Intrinsics) { 248 warning("Adler32Intrinsics not available on this CPU."); 249 FLAG_SET_DEFAULT(UseAdler32Intrinsics, false); 250 } 251 252 // The AES intrinsic stubs require AES instruction support. 253 if (has_vcipher()) { 254 if (FLAG_IS_DEFAULT(UseAES)) { 255 UseAES = true; 256 } 257 } else if (UseAES) { 258 if (!FLAG_IS_DEFAULT(UseAES)) 259 warning("AES instructions are not available on this CPU"); 260 FLAG_SET_DEFAULT(UseAES, false); 261 } 262 263 if (UseAES && has_vcipher()) { 264 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) { 265 UseAESIntrinsics = true; 266 } 267 } else if (UseAESIntrinsics) { 268 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) 269 warning("AES intrinsics are not available on this CPU"); 270 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 271 } 272 273 if (UseAESCTRIntrinsics) { 274 warning("AES/CTR intrinsics are not available on this CPU"); 275 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false); 276 } 277 278 if (UseGHASHIntrinsics) { 279 warning("GHASH intrinsics are not available on this CPU"); 280 FLAG_SET_DEFAULT(UseGHASHIntrinsics, false); 281 } 282 283 if (FLAG_IS_DEFAULT(UseFMA)) { 284 FLAG_SET_DEFAULT(UseFMA, true); 285 } 286 287 if (has_vshasig()) { 288 if (FLAG_IS_DEFAULT(UseSHA)) { 289 UseSHA = true; 290 } 291 } else if (UseSHA) { 292 if (!FLAG_IS_DEFAULT(UseSHA)) 293 warning("SHA instructions are not available on this CPU"); 294 FLAG_SET_DEFAULT(UseSHA, false); 295 } 296 297 if (UseSHA1Intrinsics) { 298 warning("Intrinsics for SHA-1 crypto hash functions not available on this CPU."); 299 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false); 300 } 301 302 if (UseSHA && has_vshasig()) { 303 if (FLAG_IS_DEFAULT(UseSHA256Intrinsics)) { 304 FLAG_SET_DEFAULT(UseSHA256Intrinsics, true); 305 } 306 } else if (UseSHA256Intrinsics) { 307 warning("Intrinsics for SHA-224 and SHA-256 crypto hash functions not available on this CPU."); 308 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false); 309 } 310 311 if (UseSHA && has_vshasig()) { 312 if (FLAG_IS_DEFAULT(UseSHA512Intrinsics)) { 313 FLAG_SET_DEFAULT(UseSHA512Intrinsics, true); 314 } 315 } else if (UseSHA512Intrinsics) { 316 warning("Intrinsics for SHA-384 and SHA-512 crypto hash functions not available on this CPU."); 317 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false); 318 } 319 320 if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics)) { 321 FLAG_SET_DEFAULT(UseSHA, false); 322 } 323 324 #ifdef COMPILER2 325 if (FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) { 326 UseSquareToLenIntrinsic = true; 327 } 328 if (FLAG_IS_DEFAULT(UseMulAddIntrinsic)) { 329 UseMulAddIntrinsic = true; 330 } 331 if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) { 332 UseMultiplyToLenIntrinsic = true; 333 } 334 if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) { 335 UseMontgomeryMultiplyIntrinsic = true; 336 } 337 if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) { 338 UseMontgomerySquareIntrinsic = true; 339 } 340 #endif 341 342 if (UseVectorizedMismatchIntrinsic) { 343 warning("UseVectorizedMismatchIntrinsic specified, but not available on this CPU."); 344 FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false); 345 } 346 347 348 // Adjust RTM (Restricted Transactional Memory) flags. 349 if (UseRTMLocking) { 350 // If CPU or OS do not support TM: 351 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 352 // setting during arguments processing. See use_biased_locking(). 353 // VM_Version_init() is executed after UseBiasedLocking is used 354 // in Thread::allocate(). 355 if (PowerArchitecturePPC64 < 8) { 356 vm_exit_during_initialization("RTM instructions are not available on this CPU."); 357 } 358 359 if (!has_tm()) { 360 vm_exit_during_initialization("RTM is not supported on this OS version."); 361 } 362 } 363 364 if (UseRTMLocking) { 365 #if INCLUDE_RTM_OPT 366 if (!FLAG_IS_CMDLINE(UseRTMLocking)) { 367 // RTM locking should be used only for applications with 368 // high lock contention. For now we do not use it by default. 369 vm_exit_during_initialization("UseRTMLocking flag should be only set on command line"); 370 } 371 #else 372 // Only C2 does RTM locking optimization. 373 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 374 // setting during arguments processing. See use_biased_locking(). 375 vm_exit_during_initialization("RTM locking optimization is not supported in this VM"); 376 #endif 377 } else { // !UseRTMLocking 378 if (UseRTMForStackLocks) { 379 if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) { 380 warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off"); 381 } 382 FLAG_SET_DEFAULT(UseRTMForStackLocks, false); 383 } 384 if (UseRTMDeopt) { 385 FLAG_SET_DEFAULT(UseRTMDeopt, false); 386 } 387 #ifdef COMPILER2 388 if (PrintPreciseRTMLockingStatistics) { 389 FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false); 390 } 391 #endif 392 } 393 394 // This machine allows unaligned memory accesses 395 if (FLAG_IS_DEFAULT(UseUnalignedAccesses)) { 396 FLAG_SET_DEFAULT(UseUnalignedAccesses, true); 397 } 398 399 check_virtualizations(); 400 } 401 402 void VM_Version::check_virtualizations() { 403 #if defined(_AIX) 404 int rc = 0; 405 perfstat_partition_total_t pinfo; 406 rc = perfstat_partition_total(NULL, &pinfo, sizeof(perfstat_partition_total_t), 1); 407 if (rc == 1) { 408 Abstract_VM_Version::_detected_virtualization = PowerVM; 409 } 410 #else 411 const char* info_file = "/proc/ppc64/lparcfg"; 412 // system_type=...qemu indicates PowerKVM 413 // e.g. system_type=IBM pSeries (emulated by qemu) 414 char line[500]; 415 FILE* fp = fopen(info_file, "r"); 416 if (fp == NULL) { 417 return; 418 } 419 const char* system_type="system_type="; // in case this line contains qemu, it is KVM 420 const char* num_lpars="NumLpars="; // in case of non-KVM : if this line is found it is PowerVM 421 bool num_lpars_found = false; 422 423 while (fgets(line, sizeof(line), fp) != NULL) { 424 if (strncmp(line, system_type, strlen(system_type)) == 0) { 425 if (strstr(line, "qemu") != 0) { 426 Abstract_VM_Version::_detected_virtualization = PowerKVM; 427 fclose(fp); 428 return; 429 } 430 } 431 if (strncmp(line, num_lpars, strlen(num_lpars)) == 0) { 432 num_lpars_found = true; 433 } 434 } 435 if (num_lpars_found) { 436 Abstract_VM_Version::_detected_virtualization = PowerVM; 437 } else { 438 Abstract_VM_Version::_detected_virtualization = PowerFullPartitionMode; 439 } 440 fclose(fp); 441 #endif 442 } 443 444 void VM_Version::print_platform_virtualization_info(outputStream* st) { 445 #if defined(_AIX) 446 // more info about perfstat API see 447 // https://www.ibm.com/support/knowledgecenter/en/ssw_aix_72/com.ibm.aix.prftools/idprftools_perfstat_glob_partition.htm 448 int rc = 0; 449 perfstat_partition_total_t pinfo; 450 memset(&pinfo, 0, sizeof(perfstat_partition_total_t)); 451 rc = perfstat_partition_total(NULL, &pinfo, sizeof(perfstat_partition_total_t), 1); 452 if (rc != 1) { 453 return; 454 } else { 455 st->print_cr("Virtualization type : PowerVM"); 456 } 457 // CPU information 458 perfstat_cpu_total_t cpuinfo; 459 memset(&cpuinfo, 0, sizeof(perfstat_cpu_total_t)); 460 rc = perfstat_cpu_total(NULL, &cpuinfo, sizeof(perfstat_cpu_total_t), 1); 461 if (rc != 1) { 462 return; 463 } 464 465 st->print_cr("Processor description : %s", cpuinfo.description); 466 st->print_cr("Processor speed : %llu Hz", cpuinfo.processorHZ); 467 468 st->print_cr("LPAR partition name : %s", pinfo.name); 469 st->print_cr("LPAR partition number : %u", pinfo.lpar_id); 470 st->print_cr("LPAR partition type : %s", pinfo.type.b.shared_enabled ? "shared" : "dedicated"); 471 st->print_cr("LPAR mode : %s", pinfo.type.b.donate_enabled ? "donating" : pinfo.type.b.capped ? "capped" : "uncapped"); 472 st->print_cr("LPAR partition group ID : %u", pinfo.group_id); 473 st->print_cr("LPAR shared pool ID : %u", pinfo.pool_id); 474 475 st->print_cr("AMS (active memory sharing) : %s", pinfo.type.b.ams_capable ? "capable" : "not capable"); 476 st->print_cr("AMS (active memory sharing) : %s", pinfo.type.b.ams_enabled ? "on" : "off"); 477 st->print_cr("AME (active memory expansion) : %s", pinfo.type.b.ame_enabled ? "on" : "off"); 478 479 if (pinfo.type.b.ame_enabled) { 480 st->print_cr("AME true memory in bytes : %llu", pinfo.true_memory); 481 st->print_cr("AME expanded memory in bytes : %llu", pinfo.expanded_memory); 482 } 483 484 st->print_cr("SMT : %s", pinfo.type.b.smt_capable ? "capable" : "not capable"); 485 st->print_cr("SMT : %s", pinfo.type.b.smt_enabled ? "on" : "off"); 486 int ocpus = pinfo.online_cpus > 0 ? pinfo.online_cpus : 1; 487 st->print_cr("LPAR threads : %d", cpuinfo.ncpus/ocpus); 488 st->print_cr("LPAR online virtual cpus : %d", pinfo.online_cpus); 489 st->print_cr("LPAR logical cpus : %d", cpuinfo.ncpus); 490 st->print_cr("LPAR maximum virtual cpus : %u", pinfo.max_cpus); 491 st->print_cr("LPAR minimum virtual cpus : %u", pinfo.min_cpus); 492 st->print_cr("LPAR entitled capacity : %4.2f", (double) (pinfo.entitled_proc_capacity/100.0)); 493 st->print_cr("LPAR online memory : %llu MB", pinfo.online_memory); 494 st->print_cr("LPAR maximum memory : %llu MB", pinfo.max_memory); 495 st->print_cr("LPAR minimum memory : %llu MB", pinfo.min_memory); 496 #else 497 const char* info_file = "/proc/ppc64/lparcfg"; 498 const char* kw[] = { "system_type=", // qemu indicates PowerKVM 499 "partition_entitled_capacity=", // entitled processor capacity percentage 500 "partition_max_entitled_capacity=", 501 "capacity_weight=", // partition CPU weight 502 "partition_active_processors=", 503 "partition_potential_processors=", 504 "entitled_proc_capacity_available=", 505 "capped=", // 0 - uncapped, 1 - vcpus capped at entitled processor capacity percentage 506 "shared_processor_mode=", // (non)dedicated partition 507 "system_potential_processors=", 508 "pool=", // CPU-pool number 509 "pool_capacity=", 510 "NumLpars=", // on non-KVM machines, NumLpars is not found for full partition mode machines 511 NULL }; 512 if (!print_matching_lines_from_file(info_file, st, kw)) { 513 st->print_cr(" <%s Not Available>", info_file); 514 } 515 #endif 516 } 517 518 bool VM_Version::use_biased_locking() { 519 #if INCLUDE_RTM_OPT 520 // RTM locking is most useful when there is high lock contention and 521 // low data contention. With high lock contention the lock is usually 522 // inflated and biased locking is not suitable for that case. 523 // RTM locking code requires that biased locking is off. 524 // Note: we can't switch off UseBiasedLocking in get_processor_features() 525 // because it is used by Thread::allocate() which is called before 526 // VM_Version::initialize(). 527 if (UseRTMLocking && UseBiasedLocking) { 528 if (FLAG_IS_DEFAULT(UseBiasedLocking)) { 529 FLAG_SET_DEFAULT(UseBiasedLocking, false); 530 } else { 531 warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." ); 532 UseBiasedLocking = false; 533 } 534 } 535 #endif 536 return UseBiasedLocking; 537 } 538 539 void VM_Version::print_features() { 540 tty->print_cr("Version: %s L1_data_cache_line_size=%d", features_string(), L1_data_cache_line_size()); 541 542 if (Verbose) { 543 if (ContendedPaddingWidth > 0) { 544 tty->cr(); 545 tty->print_cr("ContendedPaddingWidth " INTX_FORMAT, ContendedPaddingWidth); 546 } 547 } 548 } 549 550 #ifdef COMPILER2 551 // Determine section size on power6: If section size is 8 instructions, 552 // there should be a difference between the two testloops of ~15 %. If 553 // no difference is detected the section is assumed to be 32 instructions. 554 void VM_Version::determine_section_size() { 555 556 int unroll = 80; 557 558 const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord; 559 560 // Allocate space for the code. 561 ResourceMark rm; 562 CodeBuffer cb("detect_section_size", code_size, 0); 563 MacroAssembler* a = new MacroAssembler(&cb); 564 565 uint32_t *code = (uint32_t *)a->pc(); 566 // Emit code. 567 void (*test1)() = (void(*)())(void *)a->function_entry(); 568 569 Label l1; 570 571 a->li(R4, 1); 572 a->sldi(R4, R4, 28); 573 a->b(l1); 574 a->align(CodeEntryAlignment); 575 576 a->bind(l1); 577 578 for (int i = 0; i < unroll; i++) { 579 // Schleife 1 580 // ------- sector 0 ------------ 581 // ;; 0 582 a->nop(); // 1 583 a->fpnop0(); // 2 584 a->fpnop1(); // 3 585 a->addi(R4,R4, -1); // 4 586 587 // ;; 1 588 a->nop(); // 5 589 a->fmr(F6, F6); // 6 590 a->fmr(F7, F7); // 7 591 a->endgroup(); // 8 592 // ------- sector 8 ------------ 593 594 // ;; 2 595 a->nop(); // 9 596 a->nop(); // 10 597 a->fmr(F8, F8); // 11 598 a->fmr(F9, F9); // 12 599 600 // ;; 3 601 a->nop(); // 13 602 a->fmr(F10, F10); // 14 603 a->fmr(F11, F11); // 15 604 a->endgroup(); // 16 605 // -------- sector 16 ------------- 606 607 // ;; 4 608 a->nop(); // 17 609 a->nop(); // 18 610 a->fmr(F15, F15); // 19 611 a->fmr(F16, F16); // 20 612 613 // ;; 5 614 a->nop(); // 21 615 a->fmr(F17, F17); // 22 616 a->fmr(F18, F18); // 23 617 a->endgroup(); // 24 618 // ------- sector 24 ------------ 619 620 // ;; 6 621 a->nop(); // 25 622 a->nop(); // 26 623 a->fmr(F19, F19); // 27 624 a->fmr(F20, F20); // 28 625 626 // ;; 7 627 a->nop(); // 29 628 a->fmr(F21, F21); // 30 629 a->fmr(F22, F22); // 31 630 a->brnop0(); // 32 631 632 // ------- sector 32 ------------ 633 } 634 635 // ;; 8 636 a->cmpdi(CCR0, R4, unroll); // 33 637 a->bge(CCR0, l1); // 34 638 a->blr(); 639 640 // Emit code. 641 void (*test2)() = (void(*)())(void *)a->function_entry(); 642 // uint32_t *code = (uint32_t *)a->pc(); 643 644 Label l2; 645 646 a->li(R4, 1); 647 a->sldi(R4, R4, 28); 648 a->b(l2); 649 a->align(CodeEntryAlignment); 650 651 a->bind(l2); 652 653 for (int i = 0; i < unroll; i++) { 654 // Schleife 2 655 // ------- sector 0 ------------ 656 // ;; 0 657 a->brnop0(); // 1 658 a->nop(); // 2 659 //a->cmpdi(CCR0, R4, unroll); 660 a->fpnop0(); // 3 661 a->fpnop1(); // 4 662 a->addi(R4,R4, -1); // 5 663 664 // ;; 1 665 666 a->nop(); // 6 667 a->fmr(F6, F6); // 7 668 a->fmr(F7, F7); // 8 669 // ------- sector 8 --------------- 670 671 // ;; 2 672 a->endgroup(); // 9 673 674 // ;; 3 675 a->nop(); // 10 676 a->nop(); // 11 677 a->fmr(F8, F8); // 12 678 679 // ;; 4 680 a->fmr(F9, F9); // 13 681 a->nop(); // 14 682 a->fmr(F10, F10); // 15 683 684 // ;; 5 685 a->fmr(F11, F11); // 16 686 // -------- sector 16 ------------- 687 688 // ;; 6 689 a->endgroup(); // 17 690 691 // ;; 7 692 a->nop(); // 18 693 a->nop(); // 19 694 a->fmr(F15, F15); // 20 695 696 // ;; 8 697 a->fmr(F16, F16); // 21 698 a->nop(); // 22 699 a->fmr(F17, F17); // 23 700 701 // ;; 9 702 a->fmr(F18, F18); // 24 703 // -------- sector 24 ------------- 704 705 // ;; 10 706 a->endgroup(); // 25 707 708 // ;; 11 709 a->nop(); // 26 710 a->nop(); // 27 711 a->fmr(F19, F19); // 28 712 713 // ;; 12 714 a->fmr(F20, F20); // 29 715 a->nop(); // 30 716 a->fmr(F21, F21); // 31 717 718 // ;; 13 719 a->fmr(F22, F22); // 32 720 } 721 722 // -------- sector 32 ------------- 723 // ;; 14 724 a->cmpdi(CCR0, R4, unroll); // 33 725 a->bge(CCR0, l2); // 34 726 727 a->blr(); 728 uint32_t *code_end = (uint32_t *)a->pc(); 729 a->flush(); 730 731 cb.insts()->set_end((u_char*)code_end); 732 733 double loop1_seconds,loop2_seconds, rel_diff; 734 uint64_t start1, stop1; 735 736 start1 = os::current_thread_cpu_time(false); 737 (*test1)(); 738 stop1 = os::current_thread_cpu_time(false); 739 loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0); 740 741 742 start1 = os::current_thread_cpu_time(false); 743 (*test2)(); 744 stop1 = os::current_thread_cpu_time(false); 745 746 loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0); 747 748 rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100; 749 750 if (PrintAssembly || PrintStubCode) { 751 ttyLocker ttyl; 752 tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 753 // Use existing decode function. This enables the [MachCode] format which is needed to DecodeErrorFile. 754 Disassembler::decode(&cb, (u_char*)code, (u_char*)code_end, tty); 755 tty->print_cr("Time loop1 :%f", loop1_seconds); 756 tty->print_cr("Time loop2 :%f", loop2_seconds); 757 tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff); 758 759 if (rel_diff > 12.0) { 760 tty->print_cr("Section Size 8 Instructions"); 761 } else{ 762 tty->print_cr("Section Size 32 Instructions or Power5"); 763 } 764 } 765 766 #if 0 // TODO: PPC port 767 // Set sector size (if not set explicitly). 768 if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) { 769 if (rel_diff > 12.0) { 770 PdScheduling::power6SectorSize = 0x20; 771 } else { 772 PdScheduling::power6SectorSize = 0x80; 773 } 774 } else if (Power6SectorSize128PPC64) { 775 PdScheduling::power6SectorSize = 0x80; 776 } else { 777 PdScheduling::power6SectorSize = 0x20; 778 } 779 #endif 780 if (UsePower6SchedulerPPC64) Unimplemented(); 781 } 782 #endif // COMPILER2 783 784 void VM_Version::determine_features() { 785 #if defined(ABI_ELFv2) 786 // 1 InstWord per call for the blr instruction. 787 const int code_size = (num_features+1+2*1)*BytesPerInstWord; 788 #else 789 // 7 InstWords for each call (function descriptor + blr instruction). 790 const int code_size = (num_features+1+2*7)*BytesPerInstWord; 791 #endif 792 int features = 0; 793 794 // create test area 795 enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size). 796 char test_area[BUFFER_SIZE]; 797 char *mid_of_test_area = &test_area[BUFFER_SIZE>>1]; 798 799 // Allocate space for the code. 800 ResourceMark rm; 801 CodeBuffer cb("detect_cpu_features", code_size, 0); 802 MacroAssembler* a = new MacroAssembler(&cb); 803 804 // Must be set to true so we can generate the test code. 805 _features = VM_Version::all_features_m; 806 807 // Emit code. 808 void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry(); 809 uint32_t *code = (uint32_t *)a->pc(); 810 // Don't use R0 in ldarx. 811 // Keep R3_ARG1 unmodified, it contains &field (see below). 812 // Keep R4_ARG2 unmodified, it contains offset = 0 (see below). 813 a->fsqrt(F3, F4); // code[0] -> fsqrt_m 814 a->fsqrts(F3, F4); // code[1] -> fsqrts_m 815 a->isel(R7, R5, R6, 0); // code[2] -> isel_m 816 a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m 817 a->cmpb(R7, R5, R6); // code[4] -> cmpb 818 a->popcntb(R7, R5); // code[5] -> popcntb 819 a->popcntw(R7, R5); // code[6] -> popcntw 820 a->fcfids(F3, F4); // code[7] -> fcfids 821 a->vand(VR0, VR0, VR0); // code[8] -> vand 822 // arg0 of lqarx must be an even register, (arg1 + arg2) must be a multiple of 16 823 a->lqarx_unchecked(R6, R3_ARG1, R4_ARG2, 1); // code[9] -> lqarx_m 824 a->vcipher(VR0, VR1, VR2); // code[10] -> vcipher 825 a->vpmsumb(VR0, VR1, VR2); // code[11] -> vpmsumb 826 a->mfdscr(R0); // code[12] -> mfdscr 827 a->lxvd2x(VSR0, R3_ARG1); // code[13] -> vsx 828 a->ldbrx(R7, R3_ARG1, R4_ARG2); // code[14] -> ldbrx 829 a->stdbrx(R7, R3_ARG1, R4_ARG2); // code[15] -> stdbrx 830 a->vshasigmaw(VR0, VR1, 1, 0xF); // code[16] -> vshasig 831 // rtm is determined by OS 832 a->darn(R7); // code[17] -> darn 833 a->blr(); 834 835 // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it. 836 void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry(); 837 a->dcbz(R3_ARG1); // R3_ARG1 = addr 838 a->blr(); 839 840 uint32_t *code_end = (uint32_t *)a->pc(); 841 a->flush(); 842 _features = VM_Version::unknown_m; 843 844 // Print the detection code. 845 if (PrintAssembly) { 846 ttyLocker ttyl; 847 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 848 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 849 } 850 851 // Measure cache line size. 852 memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF. 853 (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle. 854 int count = 0; // count zeroed bytes 855 for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++; 856 guarantee(is_power_of_2(count), "cache line size needs to be a power of 2"); 857 _L1_data_cache_line_size = count; 858 859 // Execute code. Illegal instructions will be replaced by 0 in the signal handler. 860 VM_Version::_is_determine_features_test_running = true; 861 // We must align the first argument to 16 bytes because of the lqarx check. 862 (*test)(align_up((address)mid_of_test_area, 16), 0); 863 VM_Version::_is_determine_features_test_running = false; 864 865 // determine which instructions are legal. 866 int feature_cntr = 0; 867 if (code[feature_cntr++]) features |= fsqrt_m; 868 if (code[feature_cntr++]) features |= fsqrts_m; 869 if (code[feature_cntr++]) features |= isel_m; 870 if (code[feature_cntr++]) features |= lxarxeh_m; 871 if (code[feature_cntr++]) features |= cmpb_m; 872 if (code[feature_cntr++]) features |= popcntb_m; 873 if (code[feature_cntr++]) features |= popcntw_m; 874 if (code[feature_cntr++]) features |= fcfids_m; 875 if (code[feature_cntr++]) features |= vand_m; 876 if (code[feature_cntr++]) features |= lqarx_m; 877 if (code[feature_cntr++]) features |= vcipher_m; 878 if (code[feature_cntr++]) features |= vpmsumb_m; 879 if (code[feature_cntr++]) features |= mfdscr_m; 880 if (code[feature_cntr++]) features |= vsx_m; 881 if (code[feature_cntr++]) features |= ldbrx_m; 882 if (code[feature_cntr++]) features |= stdbrx_m; 883 if (code[feature_cntr++]) features |= vshasig_m; 884 // feature rtm_m is determined by OS 885 if (code[feature_cntr++]) features |= darn_m; 886 887 // Print the detection code. 888 if (PrintAssembly) { 889 ttyLocker ttyl; 890 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code)); 891 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 892 } 893 894 _features = features; 895 896 #ifdef AIX 897 // To enable it on AIX it's necessary POWER8 or above and at least AIX 7.2. 898 // Actually, this is supported since AIX 7.1.. Unfortunately, this first 899 // contained bugs, so that it can only be enabled after AIX 7.1.3.30. 900 // The Java property os.version, which is used in RTM tests to decide 901 // whether the feature is available, only knows major and minor versions. 902 // We don't want to change this property, as user code might depend on it. 903 // So the tests can not check on subversion 3.30, and we only enable RTM 904 // with AIX 7.2. 905 if (has_lqarx()) { // POWER8 or above 906 if (os::Aix::os_version() >= 0x07020000) { // At least AIX 7.2. 907 _features |= rtm_m; 908 } 909 } 910 #endif 911 #if defined(LINUX) && defined(VM_LITTLE_ENDIAN) 912 unsigned long auxv = getauxval(AT_HWCAP2); 913 914 if (auxv & PPC_FEATURE2_HTM_NOSC) { 915 if (auxv & PPC_FEATURE2_HAS_HTM) { 916 // TM on POWER8 and POWER9 in compat mode (VM) is supported by the JVM. 917 // TM on POWER9 DD2.1 NV (baremetal) is not supported by the JVM (TM on 918 // POWER9 DD2.1 NV has a few issues that need a couple of firmware 919 // and kernel workarounds, so there is a new mode only supported 920 // on non-virtualized P9 machines called HTM with no Suspend Mode). 921 // TM on POWER9 D2.2+ NV is not supported at all by Linux. 922 _features |= rtm_m; 923 } 924 } 925 #endif 926 } 927 928 // Power 8: Configure Data Stream Control Register. 929 void VM_Version::config_dscr() { 930 // 7 InstWords for each call (function descriptor + blr instruction). 931 const int code_size = (2+2*7)*BytesPerInstWord; 932 933 // Allocate space for the code. 934 ResourceMark rm; 935 CodeBuffer cb("config_dscr", code_size, 0); 936 MacroAssembler* a = new MacroAssembler(&cb); 937 938 // Emit code. 939 uint64_t (*get_dscr)() = (uint64_t(*)())(void *)a->function_entry(); 940 uint32_t *code = (uint32_t *)a->pc(); 941 a->mfdscr(R3); 942 a->blr(); 943 944 void (*set_dscr)(long) = (void(*)(long))(void *)a->function_entry(); 945 a->mtdscr(R3); 946 a->blr(); 947 948 uint32_t *code_end = (uint32_t *)a->pc(); 949 a->flush(); 950 951 // Print the detection code. 952 if (PrintAssembly) { 953 ttyLocker ttyl; 954 tty->print_cr("Decoding dscr configuration stub at " INTPTR_FORMAT " before execution:", p2i(code)); 955 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 956 } 957 958 // Apply the configuration if needed. 959 _dscr_val = (*get_dscr)(); 960 if (Verbose) { 961 tty->print_cr("dscr value was 0x%lx" , _dscr_val); 962 } 963 bool change_requested = false; 964 if (DSCR_PPC64 != (uintx)-1) { 965 _dscr_val = DSCR_PPC64; 966 change_requested = true; 967 } 968 if (DSCR_DPFD_PPC64 <= 7) { 969 uint64_t mask = 0x7; 970 if ((_dscr_val & mask) != DSCR_DPFD_PPC64) { 971 _dscr_val = (_dscr_val & ~mask) | (DSCR_DPFD_PPC64); 972 change_requested = true; 973 } 974 } 975 if (DSCR_URG_PPC64 <= 7) { 976 uint64_t mask = 0x7 << 6; 977 if ((_dscr_val & mask) != DSCR_DPFD_PPC64 << 6) { 978 _dscr_val = (_dscr_val & ~mask) | (DSCR_URG_PPC64 << 6); 979 change_requested = true; 980 } 981 } 982 if (change_requested) { 983 (*set_dscr)(_dscr_val); 984 if (Verbose) { 985 tty->print_cr("dscr was set to 0x%lx" , (*get_dscr)()); 986 } 987 } 988 } 989 990 static uint64_t saved_features = 0; 991 992 void VM_Version::allow_all() { 993 saved_features = _features; 994 _features = all_features_m; 995 } 996 997 void VM_Version::revert() { 998 _features = saved_features; 999 }