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 (UseMD5Intrinsics) { 288 warning("MD5 intrinsics are not available on this CPU"); 289 FLAG_SET_DEFAULT(UseMD5Intrinsics, false); 290 } 291 292 if (has_vshasig()) { 293 if (FLAG_IS_DEFAULT(UseSHA)) { 294 UseSHA = true; 295 } 296 } else if (UseSHA) { 297 if (!FLAG_IS_DEFAULT(UseSHA)) 298 warning("SHA instructions are not available on this CPU"); 299 FLAG_SET_DEFAULT(UseSHA, false); 300 } 301 302 if (UseSHA1Intrinsics) { 303 warning("Intrinsics for SHA-1 crypto hash functions not available on this CPU."); 304 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false); 305 } 306 307 if (UseSHA && has_vshasig()) { 308 if (FLAG_IS_DEFAULT(UseSHA256Intrinsics)) { 309 FLAG_SET_DEFAULT(UseSHA256Intrinsics, true); 310 } 311 } else if (UseSHA256Intrinsics) { 312 warning("Intrinsics for SHA-224 and SHA-256 crypto hash functions not available on this CPU."); 313 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false); 314 } 315 316 if (UseSHA && has_vshasig()) { 317 if (FLAG_IS_DEFAULT(UseSHA512Intrinsics)) { 318 FLAG_SET_DEFAULT(UseSHA512Intrinsics, true); 319 } 320 } else if (UseSHA512Intrinsics) { 321 warning("Intrinsics for SHA-384 and SHA-512 crypto hash functions not available on this CPU."); 322 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false); 323 } 324 325 if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics)) { 326 FLAG_SET_DEFAULT(UseSHA, false); 327 } 328 329 #ifdef COMPILER2 330 if (FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) { 331 UseSquareToLenIntrinsic = true; 332 } 333 if (FLAG_IS_DEFAULT(UseMulAddIntrinsic)) { 334 UseMulAddIntrinsic = true; 335 } 336 if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) { 337 UseMultiplyToLenIntrinsic = true; 338 } 339 if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) { 340 UseMontgomeryMultiplyIntrinsic = true; 341 } 342 if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) { 343 UseMontgomerySquareIntrinsic = true; 344 } 345 #endif 346 347 if (UseVectorizedMismatchIntrinsic) { 348 warning("UseVectorizedMismatchIntrinsic specified, but not available on this CPU."); 349 FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false); 350 } 351 352 353 // Adjust RTM (Restricted Transactional Memory) flags. 354 if (UseRTMLocking) { 355 // If CPU or OS do not support TM: 356 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 357 // setting during arguments processing. See use_biased_locking(). 358 // VM_Version_init() is executed after UseBiasedLocking is used 359 // in Thread::allocate(). 360 if (PowerArchitecturePPC64 < 8) { 361 vm_exit_during_initialization("RTM instructions are not available on this CPU."); 362 } 363 364 if (!has_tm()) { 365 vm_exit_during_initialization("RTM is not supported on this OS version."); 366 } 367 } 368 369 if (UseRTMLocking) { 370 #if INCLUDE_RTM_OPT 371 if (!FLAG_IS_CMDLINE(UseRTMLocking)) { 372 // RTM locking should be used only for applications with 373 // high lock contention. For now we do not use it by default. 374 vm_exit_during_initialization("UseRTMLocking flag should be only set on command line"); 375 } 376 #else 377 // Only C2 does RTM locking optimization. 378 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 379 // setting during arguments processing. See use_biased_locking(). 380 vm_exit_during_initialization("RTM locking optimization is not supported in this VM"); 381 #endif 382 } else { // !UseRTMLocking 383 if (UseRTMForStackLocks) { 384 if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) { 385 warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off"); 386 } 387 FLAG_SET_DEFAULT(UseRTMForStackLocks, false); 388 } 389 if (UseRTMDeopt) { 390 FLAG_SET_DEFAULT(UseRTMDeopt, false); 391 } 392 #ifdef COMPILER2 393 if (PrintPreciseRTMLockingStatistics) { 394 FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false); 395 } 396 #endif 397 } 398 399 // This machine allows unaligned memory accesses 400 if (FLAG_IS_DEFAULT(UseUnalignedAccesses)) { 401 FLAG_SET_DEFAULT(UseUnalignedAccesses, true); 402 } 403 404 check_virtualizations(); 405 } 406 407 void VM_Version::check_virtualizations() { 408 #if defined(_AIX) 409 int rc = 0; 410 perfstat_partition_total_t pinfo; 411 rc = perfstat_partition_total(NULL, &pinfo, sizeof(perfstat_partition_total_t), 1); 412 if (rc == 1) { 413 Abstract_VM_Version::_detected_virtualization = PowerVM; 414 } 415 #else 416 const char* info_file = "/proc/ppc64/lparcfg"; 417 // system_type=...qemu indicates PowerKVM 418 // e.g. system_type=IBM pSeries (emulated by qemu) 419 char line[500]; 420 FILE* fp = fopen(info_file, "r"); 421 if (fp == NULL) { 422 return; 423 } 424 const char* system_type="system_type="; // in case this line contains qemu, it is KVM 425 const char* num_lpars="NumLpars="; // in case of non-KVM : if this line is found it is PowerVM 426 bool num_lpars_found = false; 427 428 while (fgets(line, sizeof(line), fp) != NULL) { 429 if (strncmp(line, system_type, strlen(system_type)) == 0) { 430 if (strstr(line, "qemu") != 0) { 431 Abstract_VM_Version::_detected_virtualization = PowerKVM; 432 fclose(fp); 433 return; 434 } 435 } 436 if (strncmp(line, num_lpars, strlen(num_lpars)) == 0) { 437 num_lpars_found = true; 438 } 439 } 440 if (num_lpars_found) { 441 Abstract_VM_Version::_detected_virtualization = PowerVM; 442 } else { 443 Abstract_VM_Version::_detected_virtualization = PowerFullPartitionMode; 444 } 445 fclose(fp); 446 #endif 447 } 448 449 void VM_Version::print_platform_virtualization_info(outputStream* st) { 450 #if defined(_AIX) 451 // more info about perfstat API see 452 // https://www.ibm.com/support/knowledgecenter/en/ssw_aix_72/com.ibm.aix.prftools/idprftools_perfstat_glob_partition.htm 453 int rc = 0; 454 perfstat_partition_total_t pinfo; 455 memset(&pinfo, 0, sizeof(perfstat_partition_total_t)); 456 rc = perfstat_partition_total(NULL, &pinfo, sizeof(perfstat_partition_total_t), 1); 457 if (rc != 1) { 458 return; 459 } else { 460 st->print_cr("Virtualization type : PowerVM"); 461 } 462 // CPU information 463 perfstat_cpu_total_t cpuinfo; 464 memset(&cpuinfo, 0, sizeof(perfstat_cpu_total_t)); 465 rc = perfstat_cpu_total(NULL, &cpuinfo, sizeof(perfstat_cpu_total_t), 1); 466 if (rc != 1) { 467 return; 468 } 469 470 st->print_cr("Processor description : %s", cpuinfo.description); 471 st->print_cr("Processor speed : %llu Hz", cpuinfo.processorHZ); 472 473 st->print_cr("LPAR partition name : %s", pinfo.name); 474 st->print_cr("LPAR partition number : %u", pinfo.lpar_id); 475 st->print_cr("LPAR partition type : %s", pinfo.type.b.shared_enabled ? "shared" : "dedicated"); 476 st->print_cr("LPAR mode : %s", pinfo.type.b.donate_enabled ? "donating" : pinfo.type.b.capped ? "capped" : "uncapped"); 477 st->print_cr("LPAR partition group ID : %u", pinfo.group_id); 478 st->print_cr("LPAR shared pool ID : %u", pinfo.pool_id); 479 480 st->print_cr("AMS (active memory sharing) : %s", pinfo.type.b.ams_capable ? "capable" : "not capable"); 481 st->print_cr("AMS (active memory sharing) : %s", pinfo.type.b.ams_enabled ? "on" : "off"); 482 st->print_cr("AME (active memory expansion) : %s", pinfo.type.b.ame_enabled ? "on" : "off"); 483 484 if (pinfo.type.b.ame_enabled) { 485 st->print_cr("AME true memory in bytes : %llu", pinfo.true_memory); 486 st->print_cr("AME expanded memory in bytes : %llu", pinfo.expanded_memory); 487 } 488 489 st->print_cr("SMT : %s", pinfo.type.b.smt_capable ? "capable" : "not capable"); 490 st->print_cr("SMT : %s", pinfo.type.b.smt_enabled ? "on" : "off"); 491 int ocpus = pinfo.online_cpus > 0 ? pinfo.online_cpus : 1; 492 st->print_cr("LPAR threads : %d", cpuinfo.ncpus/ocpus); 493 st->print_cr("LPAR online virtual cpus : %d", pinfo.online_cpus); 494 st->print_cr("LPAR logical cpus : %d", cpuinfo.ncpus); 495 st->print_cr("LPAR maximum virtual cpus : %u", pinfo.max_cpus); 496 st->print_cr("LPAR minimum virtual cpus : %u", pinfo.min_cpus); 497 st->print_cr("LPAR entitled capacity : %4.2f", (double) (pinfo.entitled_proc_capacity/100.0)); 498 st->print_cr("LPAR online memory : %llu MB", pinfo.online_memory); 499 st->print_cr("LPAR maximum memory : %llu MB", pinfo.max_memory); 500 st->print_cr("LPAR minimum memory : %llu MB", pinfo.min_memory); 501 #else 502 const char* info_file = "/proc/ppc64/lparcfg"; 503 const char* kw[] = { "system_type=", // qemu indicates PowerKVM 504 "partition_entitled_capacity=", // entitled processor capacity percentage 505 "partition_max_entitled_capacity=", 506 "capacity_weight=", // partition CPU weight 507 "partition_active_processors=", 508 "partition_potential_processors=", 509 "entitled_proc_capacity_available=", 510 "capped=", // 0 - uncapped, 1 - vcpus capped at entitled processor capacity percentage 511 "shared_processor_mode=", // (non)dedicated partition 512 "system_potential_processors=", 513 "pool=", // CPU-pool number 514 "pool_capacity=", 515 "NumLpars=", // on non-KVM machines, NumLpars is not found for full partition mode machines 516 NULL }; 517 if (!print_matching_lines_from_file(info_file, st, kw)) { 518 st->print_cr(" <%s Not Available>", info_file); 519 } 520 #endif 521 } 522 523 bool VM_Version::use_biased_locking() { 524 #if INCLUDE_RTM_OPT 525 // RTM locking is most useful when there is high lock contention and 526 // low data contention. With high lock contention the lock is usually 527 // inflated and biased locking is not suitable for that case. 528 // RTM locking code requires that biased locking is off. 529 // Note: we can't switch off UseBiasedLocking in get_processor_features() 530 // because it is used by Thread::allocate() which is called before 531 // VM_Version::initialize(). 532 if (UseRTMLocking && UseBiasedLocking) { 533 if (FLAG_IS_DEFAULT(UseBiasedLocking)) { 534 FLAG_SET_DEFAULT(UseBiasedLocking, false); 535 } else { 536 warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." ); 537 UseBiasedLocking = false; 538 } 539 } 540 #endif 541 return UseBiasedLocking; 542 } 543 544 void VM_Version::print_features() { 545 tty->print_cr("Version: %s L1_data_cache_line_size=%d", features_string(), L1_data_cache_line_size()); 546 547 if (Verbose) { 548 if (ContendedPaddingWidth > 0) { 549 tty->cr(); 550 tty->print_cr("ContendedPaddingWidth " INTX_FORMAT, ContendedPaddingWidth); 551 } 552 } 553 } 554 555 #ifdef COMPILER2 556 // Determine section size on power6: If section size is 8 instructions, 557 // there should be a difference between the two testloops of ~15 %. If 558 // no difference is detected the section is assumed to be 32 instructions. 559 void VM_Version::determine_section_size() { 560 561 int unroll = 80; 562 563 const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord; 564 565 // Allocate space for the code. 566 ResourceMark rm; 567 CodeBuffer cb("detect_section_size", code_size, 0); 568 MacroAssembler* a = new MacroAssembler(&cb); 569 570 uint32_t *code = (uint32_t *)a->pc(); 571 // Emit code. 572 void (*test1)() = (void(*)())(void *)a->function_entry(); 573 574 Label l1; 575 576 a->li(R4, 1); 577 a->sldi(R4, R4, 28); 578 a->b(l1); 579 a->align(CodeEntryAlignment); 580 581 a->bind(l1); 582 583 for (int i = 0; i < unroll; i++) { 584 // Schleife 1 585 // ------- sector 0 ------------ 586 // ;; 0 587 a->nop(); // 1 588 a->fpnop0(); // 2 589 a->fpnop1(); // 3 590 a->addi(R4,R4, -1); // 4 591 592 // ;; 1 593 a->nop(); // 5 594 a->fmr(F6, F6); // 6 595 a->fmr(F7, F7); // 7 596 a->endgroup(); // 8 597 // ------- sector 8 ------------ 598 599 // ;; 2 600 a->nop(); // 9 601 a->nop(); // 10 602 a->fmr(F8, F8); // 11 603 a->fmr(F9, F9); // 12 604 605 // ;; 3 606 a->nop(); // 13 607 a->fmr(F10, F10); // 14 608 a->fmr(F11, F11); // 15 609 a->endgroup(); // 16 610 // -------- sector 16 ------------- 611 612 // ;; 4 613 a->nop(); // 17 614 a->nop(); // 18 615 a->fmr(F15, F15); // 19 616 a->fmr(F16, F16); // 20 617 618 // ;; 5 619 a->nop(); // 21 620 a->fmr(F17, F17); // 22 621 a->fmr(F18, F18); // 23 622 a->endgroup(); // 24 623 // ------- sector 24 ------------ 624 625 // ;; 6 626 a->nop(); // 25 627 a->nop(); // 26 628 a->fmr(F19, F19); // 27 629 a->fmr(F20, F20); // 28 630 631 // ;; 7 632 a->nop(); // 29 633 a->fmr(F21, F21); // 30 634 a->fmr(F22, F22); // 31 635 a->brnop0(); // 32 636 637 // ------- sector 32 ------------ 638 } 639 640 // ;; 8 641 a->cmpdi(CCR0, R4, unroll); // 33 642 a->bge(CCR0, l1); // 34 643 a->blr(); 644 645 // Emit code. 646 void (*test2)() = (void(*)())(void *)a->function_entry(); 647 // uint32_t *code = (uint32_t *)a->pc(); 648 649 Label l2; 650 651 a->li(R4, 1); 652 a->sldi(R4, R4, 28); 653 a->b(l2); 654 a->align(CodeEntryAlignment); 655 656 a->bind(l2); 657 658 for (int i = 0; i < unroll; i++) { 659 // Schleife 2 660 // ------- sector 0 ------------ 661 // ;; 0 662 a->brnop0(); // 1 663 a->nop(); // 2 664 //a->cmpdi(CCR0, R4, unroll); 665 a->fpnop0(); // 3 666 a->fpnop1(); // 4 667 a->addi(R4,R4, -1); // 5 668 669 // ;; 1 670 671 a->nop(); // 6 672 a->fmr(F6, F6); // 7 673 a->fmr(F7, F7); // 8 674 // ------- sector 8 --------------- 675 676 // ;; 2 677 a->endgroup(); // 9 678 679 // ;; 3 680 a->nop(); // 10 681 a->nop(); // 11 682 a->fmr(F8, F8); // 12 683 684 // ;; 4 685 a->fmr(F9, F9); // 13 686 a->nop(); // 14 687 a->fmr(F10, F10); // 15 688 689 // ;; 5 690 a->fmr(F11, F11); // 16 691 // -------- sector 16 ------------- 692 693 // ;; 6 694 a->endgroup(); // 17 695 696 // ;; 7 697 a->nop(); // 18 698 a->nop(); // 19 699 a->fmr(F15, F15); // 20 700 701 // ;; 8 702 a->fmr(F16, F16); // 21 703 a->nop(); // 22 704 a->fmr(F17, F17); // 23 705 706 // ;; 9 707 a->fmr(F18, F18); // 24 708 // -------- sector 24 ------------- 709 710 // ;; 10 711 a->endgroup(); // 25 712 713 // ;; 11 714 a->nop(); // 26 715 a->nop(); // 27 716 a->fmr(F19, F19); // 28 717 718 // ;; 12 719 a->fmr(F20, F20); // 29 720 a->nop(); // 30 721 a->fmr(F21, F21); // 31 722 723 // ;; 13 724 a->fmr(F22, F22); // 32 725 } 726 727 // -------- sector 32 ------------- 728 // ;; 14 729 a->cmpdi(CCR0, R4, unroll); // 33 730 a->bge(CCR0, l2); // 34 731 732 a->blr(); 733 uint32_t *code_end = (uint32_t *)a->pc(); 734 a->flush(); 735 736 cb.insts()->set_end((u_char*)code_end); 737 738 double loop1_seconds,loop2_seconds, rel_diff; 739 uint64_t start1, stop1; 740 741 start1 = os::current_thread_cpu_time(false); 742 (*test1)(); 743 stop1 = os::current_thread_cpu_time(false); 744 loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0); 745 746 747 start1 = os::current_thread_cpu_time(false); 748 (*test2)(); 749 stop1 = os::current_thread_cpu_time(false); 750 751 loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0); 752 753 rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100; 754 755 if (PrintAssembly || PrintStubCode) { 756 ttyLocker ttyl; 757 tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 758 // Use existing decode function. This enables the [MachCode] format which is needed to DecodeErrorFile. 759 Disassembler::decode(&cb, (u_char*)code, (u_char*)code_end, tty); 760 tty->print_cr("Time loop1 :%f", loop1_seconds); 761 tty->print_cr("Time loop2 :%f", loop2_seconds); 762 tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff); 763 764 if (rel_diff > 12.0) { 765 tty->print_cr("Section Size 8 Instructions"); 766 } else{ 767 tty->print_cr("Section Size 32 Instructions or Power5"); 768 } 769 } 770 771 #if 0 // TODO: PPC port 772 // Set sector size (if not set explicitly). 773 if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) { 774 if (rel_diff > 12.0) { 775 PdScheduling::power6SectorSize = 0x20; 776 } else { 777 PdScheduling::power6SectorSize = 0x80; 778 } 779 } else if (Power6SectorSize128PPC64) { 780 PdScheduling::power6SectorSize = 0x80; 781 } else { 782 PdScheduling::power6SectorSize = 0x20; 783 } 784 #endif 785 if (UsePower6SchedulerPPC64) Unimplemented(); 786 } 787 #endif // COMPILER2 788 789 void VM_Version::determine_features() { 790 #if defined(ABI_ELFv2) 791 // 1 InstWord per call for the blr instruction. 792 const int code_size = (num_features+1+2*1)*BytesPerInstWord; 793 #else 794 // 7 InstWords for each call (function descriptor + blr instruction). 795 const int code_size = (num_features+1+2*7)*BytesPerInstWord; 796 #endif 797 int features = 0; 798 799 // create test area 800 enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size). 801 char test_area[BUFFER_SIZE]; 802 char *mid_of_test_area = &test_area[BUFFER_SIZE>>1]; 803 804 // Allocate space for the code. 805 ResourceMark rm; 806 CodeBuffer cb("detect_cpu_features", code_size, 0); 807 MacroAssembler* a = new MacroAssembler(&cb); 808 809 // Must be set to true so we can generate the test code. 810 _features = VM_Version::all_features_m; 811 812 // Emit code. 813 void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry(); 814 uint32_t *code = (uint32_t *)a->pc(); 815 // Don't use R0 in ldarx. 816 // Keep R3_ARG1 unmodified, it contains &field (see below). 817 // Keep R4_ARG2 unmodified, it contains offset = 0 (see below). 818 a->fsqrt(F3, F4); // code[0] -> fsqrt_m 819 a->fsqrts(F3, F4); // code[1] -> fsqrts_m 820 a->isel(R7, R5, R6, 0); // code[2] -> isel_m 821 a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m 822 a->cmpb(R7, R5, R6); // code[4] -> cmpb 823 a->popcntb(R7, R5); // code[5] -> popcntb 824 a->popcntw(R7, R5); // code[6] -> popcntw 825 a->fcfids(F3, F4); // code[7] -> fcfids 826 a->vand(VR0, VR0, VR0); // code[8] -> vand 827 // arg0 of lqarx must be an even register, (arg1 + arg2) must be a multiple of 16 828 a->lqarx_unchecked(R6, R3_ARG1, R4_ARG2, 1); // code[9] -> lqarx_m 829 a->vcipher(VR0, VR1, VR2); // code[10] -> vcipher 830 a->vpmsumb(VR0, VR1, VR2); // code[11] -> vpmsumb 831 a->mfdscr(R0); // code[12] -> mfdscr 832 a->lxvd2x(VSR0, R3_ARG1); // code[13] -> vsx 833 a->ldbrx(R7, R3_ARG1, R4_ARG2); // code[14] -> ldbrx 834 a->stdbrx(R7, R3_ARG1, R4_ARG2); // code[15] -> stdbrx 835 a->vshasigmaw(VR0, VR1, 1, 0xF); // code[16] -> vshasig 836 // rtm is determined by OS 837 a->darn(R7); // code[17] -> darn 838 a->blr(); 839 840 // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it. 841 void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry(); 842 a->dcbz(R3_ARG1); // R3_ARG1 = addr 843 a->blr(); 844 845 uint32_t *code_end = (uint32_t *)a->pc(); 846 a->flush(); 847 _features = VM_Version::unknown_m; 848 849 // Print the detection code. 850 if (PrintAssembly) { 851 ttyLocker ttyl; 852 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 853 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 854 } 855 856 // Measure cache line size. 857 memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF. 858 (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle. 859 int count = 0; // count zeroed bytes 860 for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++; 861 guarantee(is_power_of_2(count), "cache line size needs to be a power of 2"); 862 _L1_data_cache_line_size = count; 863 864 // Execute code. Illegal instructions will be replaced by 0 in the signal handler. 865 VM_Version::_is_determine_features_test_running = true; 866 // We must align the first argument to 16 bytes because of the lqarx check. 867 (*test)(align_up((address)mid_of_test_area, 16), 0); 868 VM_Version::_is_determine_features_test_running = false; 869 870 // determine which instructions are legal. 871 int feature_cntr = 0; 872 if (code[feature_cntr++]) features |= fsqrt_m; 873 if (code[feature_cntr++]) features |= fsqrts_m; 874 if (code[feature_cntr++]) features |= isel_m; 875 if (code[feature_cntr++]) features |= lxarxeh_m; 876 if (code[feature_cntr++]) features |= cmpb_m; 877 if (code[feature_cntr++]) features |= popcntb_m; 878 if (code[feature_cntr++]) features |= popcntw_m; 879 if (code[feature_cntr++]) features |= fcfids_m; 880 if (code[feature_cntr++]) features |= vand_m; 881 if (code[feature_cntr++]) features |= lqarx_m; 882 if (code[feature_cntr++]) features |= vcipher_m; 883 if (code[feature_cntr++]) features |= vpmsumb_m; 884 if (code[feature_cntr++]) features |= mfdscr_m; 885 if (code[feature_cntr++]) features |= vsx_m; 886 if (code[feature_cntr++]) features |= ldbrx_m; 887 if (code[feature_cntr++]) features |= stdbrx_m; 888 if (code[feature_cntr++]) features |= vshasig_m; 889 // feature rtm_m is determined by OS 890 if (code[feature_cntr++]) features |= darn_m; 891 892 // Print the detection code. 893 if (PrintAssembly) { 894 ttyLocker ttyl; 895 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code)); 896 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 897 } 898 899 _features = features; 900 901 #ifdef AIX 902 // To enable it on AIX it's necessary POWER8 or above and at least AIX 7.2. 903 // Actually, this is supported since AIX 7.1.. Unfortunately, this first 904 // contained bugs, so that it can only be enabled after AIX 7.1.3.30. 905 // The Java property os.version, which is used in RTM tests to decide 906 // whether the feature is available, only knows major and minor versions. 907 // We don't want to change this property, as user code might depend on it. 908 // So the tests can not check on subversion 3.30, and we only enable RTM 909 // with AIX 7.2. 910 if (has_lqarx()) { // POWER8 or above 911 if (os::Aix::os_version() >= 0x07020000) { // At least AIX 7.2. 912 _features |= rtm_m; 913 } 914 } 915 #endif 916 #if defined(LINUX) && defined(VM_LITTLE_ENDIAN) 917 unsigned long auxv = getauxval(AT_HWCAP2); 918 919 if (auxv & PPC_FEATURE2_HTM_NOSC) { 920 if (auxv & PPC_FEATURE2_HAS_HTM) { 921 // TM on POWER8 and POWER9 in compat mode (VM) is supported by the JVM. 922 // TM on POWER9 DD2.1 NV (baremetal) is not supported by the JVM (TM on 923 // POWER9 DD2.1 NV has a few issues that need a couple of firmware 924 // and kernel workarounds, so there is a new mode only supported 925 // on non-virtualized P9 machines called HTM with no Suspend Mode). 926 // TM on POWER9 D2.2+ NV is not supported at all by Linux. 927 _features |= rtm_m; 928 } 929 } 930 #endif 931 } 932 933 // Power 8: Configure Data Stream Control Register. 934 void VM_Version::config_dscr() { 935 // 7 InstWords for each call (function descriptor + blr instruction). 936 const int code_size = (2+2*7)*BytesPerInstWord; 937 938 // Allocate space for the code. 939 ResourceMark rm; 940 CodeBuffer cb("config_dscr", code_size, 0); 941 MacroAssembler* a = new MacroAssembler(&cb); 942 943 // Emit code. 944 uint64_t (*get_dscr)() = (uint64_t(*)())(void *)a->function_entry(); 945 uint32_t *code = (uint32_t *)a->pc(); 946 a->mfdscr(R3); 947 a->blr(); 948 949 void (*set_dscr)(long) = (void(*)(long))(void *)a->function_entry(); 950 a->mtdscr(R3); 951 a->blr(); 952 953 uint32_t *code_end = (uint32_t *)a->pc(); 954 a->flush(); 955 956 // Print the detection code. 957 if (PrintAssembly) { 958 ttyLocker ttyl; 959 tty->print_cr("Decoding dscr configuration stub at " INTPTR_FORMAT " before execution:", p2i(code)); 960 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 961 } 962 963 // Apply the configuration if needed. 964 _dscr_val = (*get_dscr)(); 965 if (Verbose) { 966 tty->print_cr("dscr value was 0x%lx" , _dscr_val); 967 } 968 bool change_requested = false; 969 if (DSCR_PPC64 != (uintx)-1) { 970 _dscr_val = DSCR_PPC64; 971 change_requested = true; 972 } 973 if (DSCR_DPFD_PPC64 <= 7) { 974 uint64_t mask = 0x7; 975 if ((_dscr_val & mask) != DSCR_DPFD_PPC64) { 976 _dscr_val = (_dscr_val & ~mask) | (DSCR_DPFD_PPC64); 977 change_requested = true; 978 } 979 } 980 if (DSCR_URG_PPC64 <= 7) { 981 uint64_t mask = 0x7 << 6; 982 if ((_dscr_val & mask) != DSCR_DPFD_PPC64 << 6) { 983 _dscr_val = (_dscr_val & ~mask) | (DSCR_URG_PPC64 << 6); 984 change_requested = true; 985 } 986 } 987 if (change_requested) { 988 (*set_dscr)(_dscr_val); 989 if (Verbose) { 990 tty->print_cr("dscr was set to 0x%lx" , (*get_dscr)()); 991 } 992 } 993 } 994 995 static uint64_t saved_features = 0; 996 997 void VM_Version::allow_all() { 998 saved_features = _features; 999 _features = all_features_m; 1000 } 1001 1002 void VM_Version::revert() { 1003 _features = saved_features; 1004 }