1 /* 2 * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved. 3 * Copyright 2012, 2015 SAP AG. 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 "asm/assembler.inline.hpp" 28 #include "asm/macroAssembler.inline.hpp" 29 #include "compiler/disassembler.hpp" 30 #include "memory/resourceArea.hpp" 31 #include "runtime/java.hpp" 32 #include "runtime/os.hpp" 33 #include "runtime/stubCodeGenerator.hpp" 34 #include "utilities/defaultStream.hpp" 35 #include "utilities/globalDefinitions.hpp" 36 #include "vm_version_ppc.hpp" 37 38 # include <sys/sysinfo.h> 39 40 int VM_Version::_features = VM_Version::unknown_m; 41 int VM_Version::_measured_cache_line_size = 32; // pessimistic init value 42 const char* VM_Version::_features_str = ""; 43 bool VM_Version::_is_determine_features_test_running = false; 44 45 46 #define MSG(flag) \ 47 if (flag && !FLAG_IS_DEFAULT(flag)) \ 48 jio_fprintf(defaultStream::error_stream(), \ 49 "warning: -XX:+" #flag " requires -XX:+UseSIGTRAP\n" \ 50 " -XX:+" #flag " will be disabled!\n"); 51 52 void VM_Version::initialize() { 53 54 // Test which instructions are supported and measure cache line size. 55 determine_features(); 56 57 // If PowerArchitecturePPC64 hasn't been specified explicitly determine from features. 58 if (FLAG_IS_DEFAULT(PowerArchitecturePPC64)) { 59 if (VM_Version::has_lqarx()) { 60 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 8); 61 } else if (VM_Version::has_popcntw()) { 62 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7); 63 } else if (VM_Version::has_cmpb()) { 64 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 6); 65 } else if (VM_Version::has_popcntb()) { 66 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 5); 67 } else { 68 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 0); 69 } 70 } 71 guarantee(PowerArchitecturePPC64 == 0 || PowerArchitecturePPC64 == 5 || 72 PowerArchitecturePPC64 == 6 || PowerArchitecturePPC64 == 7 || 73 PowerArchitecturePPC64 == 8, 74 "PowerArchitecturePPC64 should be 0, 5, 6, 7, or 8"); 75 76 // Power 8: Configure Data Stream Control Register. 77 if (PowerArchitecturePPC64 >= 8) { 78 config_dscr(); 79 } 80 81 if (!UseSIGTRAP) { 82 MSG(TrapBasedICMissChecks); 83 MSG(TrapBasedNotEntrantChecks); 84 MSG(TrapBasedNullChecks); 85 FLAG_SET_ERGO(bool, TrapBasedNotEntrantChecks, false); 86 FLAG_SET_ERGO(bool, TrapBasedNullChecks, false); 87 FLAG_SET_ERGO(bool, TrapBasedICMissChecks, false); 88 } 89 90 #ifdef COMPILER2 91 if (!UseSIGTRAP) { 92 MSG(TrapBasedRangeChecks); 93 FLAG_SET_ERGO(bool, TrapBasedRangeChecks, false); 94 } 95 96 // On Power6 test for section size. 97 if (PowerArchitecturePPC64 == 6) { 98 determine_section_size(); 99 // TODO: PPC port } else { 100 // TODO: PPC port PdScheduling::power6SectorSize = 0x20; 101 } 102 103 MaxVectorSize = 8; 104 #endif 105 106 // Create and print feature-string. 107 char buf[(num_features+1) * 16]; // Max 16 chars per feature. 108 jio_snprintf(buf, sizeof(buf), 109 "ppc64%s%s%s%s%s%s%s%s%s%s%s%s", 110 (has_fsqrt() ? " fsqrt" : ""), 111 (has_isel() ? " isel" : ""), 112 (has_lxarxeh() ? " lxarxeh" : ""), 113 (has_cmpb() ? " cmpb" : ""), 114 //(has_mftgpr()? " mftgpr" : ""), 115 (has_popcntb() ? " popcntb" : ""), 116 (has_popcntw() ? " popcntw" : ""), 117 (has_fcfids() ? " fcfids" : ""), 118 (has_vand() ? " vand" : ""), 119 (has_lqarx() ? " lqarx" : ""), 120 (has_vcipher() ? " vcipher" : ""), 121 (has_vpmsumb() ? " vpmsumb" : ""), 122 (has_tcheck() ? " tcheck" : "") 123 // Make sure number of %s matches num_features! 124 ); 125 _features_str = os::strdup(buf); 126 if (Verbose) { 127 print_features(); 128 } 129 130 // PPC64 supports 8-byte compare-exchange operations (see 131 // Atomic::cmpxchg and StubGenerator::generate_atomic_cmpxchg_ptr) 132 // and 'atomic long memory ops' (see Unsafe_GetLongVolatile). 133 _supports_cx8 = true; 134 135 UseSSE = 0; // Only on x86 and x64 136 137 intx cache_line_size = _measured_cache_line_size; 138 139 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1; 140 141 if (AllocatePrefetchStyle == 4) { 142 AllocatePrefetchStepSize = cache_line_size; // Need exact value. 143 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default. 144 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined? 145 } else { 146 if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size; 147 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value. 148 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined? 149 } 150 151 assert(AllocatePrefetchLines > 0, "invalid value"); 152 if (AllocatePrefetchLines < 1) { // Set valid value in product VM. 153 AllocatePrefetchLines = 1; // Conservative value. 154 } 155 156 if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) { 157 AllocatePrefetchStyle = 1; // Fall back if inappropriate. 158 } 159 160 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive"); 161 162 // Implementation does not use any of the vector instructions 163 // available with Power8. Their exploitation is still pending. 164 if (!UseCRC32Intrinsics) { 165 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) { 166 FLAG_SET_DEFAULT(UseCRC32Intrinsics, true); 167 } 168 } 169 170 if (UseCRC32CIntrinsics) { 171 if (!FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) 172 warning("CRC32C intrinsics are not available on this CPU"); 173 FLAG_SET_DEFAULT(UseCRC32CIntrinsics, false); 174 } 175 176 // The AES intrinsic stubs require AES instruction support. 177 if (UseAES) { 178 warning("AES instructions are not available on this CPU"); 179 FLAG_SET_DEFAULT(UseAES, false); 180 } 181 if (UseAESIntrinsics) { 182 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) 183 warning("AES intrinsics are not available on this CPU"); 184 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 185 } 186 187 if (UseGHASHIntrinsics) { 188 warning("GHASH intrinsics are not available on this CPU"); 189 FLAG_SET_DEFAULT(UseGHASHIntrinsics, false); 190 } 191 192 if (UseSHA) { 193 warning("SHA instructions are not available on this CPU"); 194 FLAG_SET_DEFAULT(UseSHA, false); 195 } 196 if (UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics) { 197 warning("SHA intrinsics are not available on this CPU"); 198 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false); 199 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false); 200 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false); 201 } 202 203 if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) { 204 UseMultiplyToLenIntrinsic = true; 205 } 206 207 // Adjust RTM (Restricted Transactional Memory) flags. 208 if (!has_tcheck() && UseRTMLocking) { 209 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 210 // setting during arguments processing. See use_biased_locking(). 211 // VM_Version_init() is executed after UseBiasedLocking is used 212 // in Thread::allocate(). 213 vm_exit_during_initialization("RTM instructions are not available on this CPU"); 214 } 215 216 if (UseRTMLocking) { 217 #if INCLUDE_RTM_OPT 218 if (!UnlockExperimentalVMOptions) { 219 vm_exit_during_initialization("UseRTMLocking is only available as experimental option on this platform. " 220 "It must be enabled via -XX:+UnlockExperimentalVMOptions flag."); 221 } else { 222 warning("UseRTMLocking is only available as experimental option on this platform."); 223 } 224 if (!FLAG_IS_CMDLINE(UseRTMLocking)) { 225 // RTM locking should be used only for applications with 226 // high lock contention. For now we do not use it by default. 227 vm_exit_during_initialization("UseRTMLocking flag should be only set on command line"); 228 } 229 if (!is_power_of_2(RTMTotalCountIncrRate)) { 230 warning("RTMTotalCountIncrRate must be a power of 2, resetting it to 64"); 231 FLAG_SET_DEFAULT(RTMTotalCountIncrRate, 64); 232 } 233 if (RTMAbortRatio < 0 || RTMAbortRatio > 100) { 234 warning("RTMAbortRatio must be in the range 0 to 100, resetting it to 50"); 235 FLAG_SET_DEFAULT(RTMAbortRatio, 50); 236 } 237 guarantee(RTMSpinLoopCount > 0, "unsupported"); 238 #else 239 // Only C2 does RTM locking optimization. 240 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 241 // setting during arguments processing. See use_biased_locking(). 242 vm_exit_during_initialization("RTM locking optimization is not supported in this VM"); 243 #endif 244 } else { // !UseRTMLocking 245 if (UseRTMForStackLocks) { 246 if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) { 247 warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off"); 248 } 249 FLAG_SET_DEFAULT(UseRTMForStackLocks, false); 250 } 251 if (UseRTMDeopt) { 252 FLAG_SET_DEFAULT(UseRTMDeopt, false); 253 } 254 if (PrintPreciseRTMLockingStatistics) { 255 FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false); 256 } 257 } 258 259 // This machine does not allow unaligned memory accesses 260 if (UseUnalignedAccesses) { 261 if (!FLAG_IS_DEFAULT(UseUnalignedAccesses)) 262 warning("Unaligned memory access is not available on this CPU"); 263 FLAG_SET_DEFAULT(UseUnalignedAccesses, false); 264 } 265 } 266 267 bool VM_Version::use_biased_locking() { 268 #if INCLUDE_RTM_OPT 269 // RTM locking is most useful when there is high lock contention and 270 // low data contention. With high lock contention the lock is usually 271 // inflated and biased locking is not suitable for that case. 272 // RTM locking code requires that biased locking is off. 273 // Note: we can't switch off UseBiasedLocking in get_processor_features() 274 // because it is used by Thread::allocate() which is called before 275 // VM_Version::initialize(). 276 if (UseRTMLocking && UseBiasedLocking) { 277 if (FLAG_IS_DEFAULT(UseBiasedLocking)) { 278 FLAG_SET_DEFAULT(UseBiasedLocking, false); 279 } else { 280 warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." ); 281 UseBiasedLocking = false; 282 } 283 } 284 #endif 285 return UseBiasedLocking; 286 } 287 288 void VM_Version::print_features() { 289 tty->print_cr("Version: %s cache_line_size = %d", cpu_features(), (int) get_cache_line_size()); 290 } 291 292 #ifdef COMPILER2 293 // Determine section size on power6: If section size is 8 instructions, 294 // there should be a difference between the two testloops of ~15 %. If 295 // no difference is detected the section is assumed to be 32 instructions. 296 void VM_Version::determine_section_size() { 297 298 int unroll = 80; 299 300 const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord; 301 302 // Allocate space for the code. 303 ResourceMark rm; 304 CodeBuffer cb("detect_section_size", code_size, 0); 305 MacroAssembler* a = new MacroAssembler(&cb); 306 307 uint32_t *code = (uint32_t *)a->pc(); 308 // Emit code. 309 void (*test1)() = (void(*)())(void *)a->function_entry(); 310 311 Label l1; 312 313 a->li(R4, 1); 314 a->sldi(R4, R4, 28); 315 a->b(l1); 316 a->align(CodeEntryAlignment); 317 318 a->bind(l1); 319 320 for (int i = 0; i < unroll; i++) { 321 // Schleife 1 322 // ------- sector 0 ------------ 323 // ;; 0 324 a->nop(); // 1 325 a->fpnop0(); // 2 326 a->fpnop1(); // 3 327 a->addi(R4,R4, -1); // 4 328 329 // ;; 1 330 a->nop(); // 5 331 a->fmr(F6, F6); // 6 332 a->fmr(F7, F7); // 7 333 a->endgroup(); // 8 334 // ------- sector 8 ------------ 335 336 // ;; 2 337 a->nop(); // 9 338 a->nop(); // 10 339 a->fmr(F8, F8); // 11 340 a->fmr(F9, F9); // 12 341 342 // ;; 3 343 a->nop(); // 13 344 a->fmr(F10, F10); // 14 345 a->fmr(F11, F11); // 15 346 a->endgroup(); // 16 347 // -------- sector 16 ------------- 348 349 // ;; 4 350 a->nop(); // 17 351 a->nop(); // 18 352 a->fmr(F15, F15); // 19 353 a->fmr(F16, F16); // 20 354 355 // ;; 5 356 a->nop(); // 21 357 a->fmr(F17, F17); // 22 358 a->fmr(F18, F18); // 23 359 a->endgroup(); // 24 360 // ------- sector 24 ------------ 361 362 // ;; 6 363 a->nop(); // 25 364 a->nop(); // 26 365 a->fmr(F19, F19); // 27 366 a->fmr(F20, F20); // 28 367 368 // ;; 7 369 a->nop(); // 29 370 a->fmr(F21, F21); // 30 371 a->fmr(F22, F22); // 31 372 a->brnop0(); // 32 373 374 // ------- sector 32 ------------ 375 } 376 377 // ;; 8 378 a->cmpdi(CCR0, R4, unroll); // 33 379 a->bge(CCR0, l1); // 34 380 a->blr(); 381 382 // Emit code. 383 void (*test2)() = (void(*)())(void *)a->function_entry(); 384 // uint32_t *code = (uint32_t *)a->pc(); 385 386 Label l2; 387 388 a->li(R4, 1); 389 a->sldi(R4, R4, 28); 390 a->b(l2); 391 a->align(CodeEntryAlignment); 392 393 a->bind(l2); 394 395 for (int i = 0; i < unroll; i++) { 396 // Schleife 2 397 // ------- sector 0 ------------ 398 // ;; 0 399 a->brnop0(); // 1 400 a->nop(); // 2 401 //a->cmpdi(CCR0, R4, unroll); 402 a->fpnop0(); // 3 403 a->fpnop1(); // 4 404 a->addi(R4,R4, -1); // 5 405 406 // ;; 1 407 408 a->nop(); // 6 409 a->fmr(F6, F6); // 7 410 a->fmr(F7, F7); // 8 411 // ------- sector 8 --------------- 412 413 // ;; 2 414 a->endgroup(); // 9 415 416 // ;; 3 417 a->nop(); // 10 418 a->nop(); // 11 419 a->fmr(F8, F8); // 12 420 421 // ;; 4 422 a->fmr(F9, F9); // 13 423 a->nop(); // 14 424 a->fmr(F10, F10); // 15 425 426 // ;; 5 427 a->fmr(F11, F11); // 16 428 // -------- sector 16 ------------- 429 430 // ;; 6 431 a->endgroup(); // 17 432 433 // ;; 7 434 a->nop(); // 18 435 a->nop(); // 19 436 a->fmr(F15, F15); // 20 437 438 // ;; 8 439 a->fmr(F16, F16); // 21 440 a->nop(); // 22 441 a->fmr(F17, F17); // 23 442 443 // ;; 9 444 a->fmr(F18, F18); // 24 445 // -------- sector 24 ------------- 446 447 // ;; 10 448 a->endgroup(); // 25 449 450 // ;; 11 451 a->nop(); // 26 452 a->nop(); // 27 453 a->fmr(F19, F19); // 28 454 455 // ;; 12 456 a->fmr(F20, F20); // 29 457 a->nop(); // 30 458 a->fmr(F21, F21); // 31 459 460 // ;; 13 461 a->fmr(F22, F22); // 32 462 } 463 464 // -------- sector 32 ------------- 465 // ;; 14 466 a->cmpdi(CCR0, R4, unroll); // 33 467 a->bge(CCR0, l2); // 34 468 469 a->blr(); 470 uint32_t *code_end = (uint32_t *)a->pc(); 471 a->flush(); 472 473 double loop1_seconds,loop2_seconds, rel_diff; 474 uint64_t start1, stop1; 475 476 start1 = os::current_thread_cpu_time(false); 477 (*test1)(); 478 stop1 = os::current_thread_cpu_time(false); 479 loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0); 480 481 482 start1 = os::current_thread_cpu_time(false); 483 (*test2)(); 484 stop1 = os::current_thread_cpu_time(false); 485 486 loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0); 487 488 rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100; 489 490 if (PrintAssembly) { 491 ttyLocker ttyl; 492 tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 493 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 494 tty->print_cr("Time loop1 :%f", loop1_seconds); 495 tty->print_cr("Time loop2 :%f", loop2_seconds); 496 tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff); 497 498 if (rel_diff > 12.0) { 499 tty->print_cr("Section Size 8 Instructions"); 500 } else{ 501 tty->print_cr("Section Size 32 Instructions or Power5"); 502 } 503 } 504 505 #if 0 // TODO: PPC port 506 // Set sector size (if not set explicitly). 507 if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) { 508 if (rel_diff > 12.0) { 509 PdScheduling::power6SectorSize = 0x20; 510 } else { 511 PdScheduling::power6SectorSize = 0x80; 512 } 513 } else if (Power6SectorSize128PPC64) { 514 PdScheduling::power6SectorSize = 0x80; 515 } else { 516 PdScheduling::power6SectorSize = 0x20; 517 } 518 #endif 519 if (UsePower6SchedulerPPC64) Unimplemented(); 520 } 521 #endif // COMPILER2 522 523 void VM_Version::determine_features() { 524 #if defined(ABI_ELFv2) 525 // 1 InstWord per call for the blr instruction. 526 const int code_size = (num_features+1+2*1)*BytesPerInstWord; 527 #else 528 // 7 InstWords for each call (function descriptor + blr instruction). 529 const int code_size = (num_features+1+2*7)*BytesPerInstWord; 530 #endif 531 int features = 0; 532 533 // create test area 534 enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size). 535 char test_area[BUFFER_SIZE]; 536 char *mid_of_test_area = &test_area[BUFFER_SIZE>>1]; 537 538 // Allocate space for the code. 539 ResourceMark rm; 540 CodeBuffer cb("detect_cpu_features", code_size, 0); 541 MacroAssembler* a = new MacroAssembler(&cb); 542 543 // Must be set to true so we can generate the test code. 544 _features = VM_Version::all_features_m; 545 546 // Emit code. 547 void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry(); 548 uint32_t *code = (uint32_t *)a->pc(); 549 // Don't use R0 in ldarx. 550 // Keep R3_ARG1 unmodified, it contains &field (see below). 551 // Keep R4_ARG2 unmodified, it contains offset = 0 (see below). 552 a->fsqrt(F3, F4); // code[0] -> fsqrt_m 553 a->fsqrts(F3, F4); // code[1] -> fsqrts_m 554 a->isel(R7, R5, R6, 0); // code[2] -> isel_m 555 a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m 556 a->cmpb(R7, R5, R6); // code[4] -> cmpb 557 a->popcntb(R7, R5); // code[5] -> popcntb 558 a->popcntw(R7, R5); // code[6] -> popcntw 559 a->fcfids(F3, F4); // code[7] -> fcfids 560 a->vand(VR0, VR0, VR0); // code[8] -> vand 561 // arg0 of lqarx must be an even register, (arg1 + arg2) must be a multiple of 16 562 a->lqarx_unchecked(R6, R3_ARG1, R4_ARG2, 1); // code[9] -> lqarx_m 563 a->vcipher(VR0, VR1, VR2); // code[10] -> vcipher 564 a->vpmsumb(VR0, VR1, VR2); // code[11] -> vpmsumb 565 a->tcheck(0); // code[12] -> tcheck 566 a->blr(); 567 568 // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it. 569 void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry(); 570 a->dcbz(R3_ARG1); // R3_ARG1 = addr 571 a->blr(); 572 573 uint32_t *code_end = (uint32_t *)a->pc(); 574 a->flush(); 575 _features = VM_Version::unknown_m; 576 577 // Print the detection code. 578 if (PrintAssembly) { 579 ttyLocker ttyl; 580 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 581 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 582 } 583 584 // Measure cache line size. 585 memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF. 586 (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle. 587 int count = 0; // count zeroed bytes 588 for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++; 589 guarantee(is_power_of_2(count), "cache line size needs to be a power of 2"); 590 _measured_cache_line_size = count; 591 592 // Execute code. Illegal instructions will be replaced by 0 in the signal handler. 593 VM_Version::_is_determine_features_test_running = true; 594 // We must align the first argument to 16 bytes because of the lqarx check. 595 (*test)((address)align_size_up((intptr_t)mid_of_test_area, 16), (uint64_t)0); 596 VM_Version::_is_determine_features_test_running = false; 597 598 // determine which instructions are legal. 599 int feature_cntr = 0; 600 if (code[feature_cntr++]) features |= fsqrt_m; 601 if (code[feature_cntr++]) features |= fsqrts_m; 602 if (code[feature_cntr++]) features |= isel_m; 603 if (code[feature_cntr++]) features |= lxarxeh_m; 604 if (code[feature_cntr++]) features |= cmpb_m; 605 if (code[feature_cntr++]) features |= popcntb_m; 606 if (code[feature_cntr++]) features |= popcntw_m; 607 if (code[feature_cntr++]) features |= fcfids_m; 608 if (code[feature_cntr++]) features |= vand_m; 609 if (code[feature_cntr++]) features |= lqarx_m; 610 if (code[feature_cntr++]) features |= vcipher_m; 611 if (code[feature_cntr++]) features |= vpmsumb_m; 612 if (code[feature_cntr++]) features |= tcheck_m; 613 614 // Print the detection code. 615 if (PrintAssembly) { 616 ttyLocker ttyl; 617 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code)); 618 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 619 } 620 621 _features = features; 622 } 623 624 // Power 8: Configure Data Stream Control Register. 625 void VM_Version::config_dscr() { 626 assert(has_tcheck(), "Only execute on Power 8 or later!"); 627 628 // 7 InstWords for each call (function descriptor + blr instruction). 629 const int code_size = (2+2*7)*BytesPerInstWord; 630 631 // Allocate space for the code. 632 ResourceMark rm; 633 CodeBuffer cb("config_dscr", code_size, 0); 634 MacroAssembler* a = new MacroAssembler(&cb); 635 636 // Emit code. 637 uint64_t (*get_dscr)() = (uint64_t(*)())(void *)a->function_entry(); 638 uint32_t *code = (uint32_t *)a->pc(); 639 a->mfdscr(R3); 640 a->blr(); 641 642 void (*set_dscr)(long) = (void(*)(long))(void *)a->function_entry(); 643 a->mtdscr(R3); 644 a->blr(); 645 646 uint32_t *code_end = (uint32_t *)a->pc(); 647 a->flush(); 648 649 // Print the detection code. 650 if (PrintAssembly) { 651 ttyLocker ttyl; 652 tty->print_cr("Decoding dscr configuration stub at " INTPTR_FORMAT " before execution:", p2i(code)); 653 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 654 } 655 656 // Apply the configuration if needed. 657 uint64_t dscr_val = (*get_dscr)(); 658 if (Verbose) { 659 tty->print_cr("dscr value was 0x%lx" , dscr_val); 660 } 661 bool change_requested = false; 662 if (DSCR_PPC64 != (uintx)-1) { 663 dscr_val = DSCR_PPC64; 664 change_requested = true; 665 } 666 if (DSCR_DPFD_PPC64 <= 7) { 667 uint64_t mask = 0x7; 668 if ((dscr_val & mask) != DSCR_DPFD_PPC64) { 669 dscr_val = (dscr_val & ~mask) | (DSCR_DPFD_PPC64); 670 change_requested = true; 671 } 672 } 673 if (DSCR_URG_PPC64 <= 7) { 674 uint64_t mask = 0x7 << 6; 675 if ((dscr_val & mask) != DSCR_DPFD_PPC64 << 6) { 676 dscr_val = (dscr_val & ~mask) | (DSCR_URG_PPC64 << 6); 677 change_requested = true; 678 } 679 } 680 if (change_requested) { 681 (*set_dscr)(dscr_val); 682 if (Verbose) { 683 tty->print_cr("dscr was set to 0x%lx" , (*get_dscr)()); 684 } 685 } 686 } 687 688 static int saved_features = 0; 689 690 void VM_Version::allow_all() { 691 saved_features = _features; 692 _features = all_features_m; 693 } 694 695 void VM_Version::revert() { 696 _features = saved_features; 697 }