1 /* 2 * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. 3 * Copyright 2012, 2014 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/stubCodeGenerator.hpp" 33 #include "utilities/defaultStream.hpp" 34 #include "vm_version_ppc.hpp" 35 #ifdef TARGET_OS_FAMILY_aix 36 # include "os_aix.inline.hpp" 37 #endif 38 #ifdef TARGET_OS_FAMILY_linux 39 # include "os_linux.inline.hpp" 40 #endif 41 42 # include <sys/sysinfo.h> 43 44 int VM_Version::_features = VM_Version::unknown_m; 45 int VM_Version::_measured_cache_line_size = 128; // default value 46 const char* VM_Version::_features_str = ""; 47 bool VM_Version::_is_determine_features_test_running = false; 48 49 50 #define MSG(flag) \ 51 if (flag && !FLAG_IS_DEFAULT(flag)) \ 52 jio_fprintf(defaultStream::error_stream(), \ 53 "warning: -XX:+" #flag " requires -XX:+UseSIGTRAP\n" \ 54 " -XX:+" #flag " will be disabled!\n"); 55 56 void VM_Version::initialize() { 57 58 // Test which instructions are supported and measure cache line size. 59 determine_features(); 60 61 // If PowerArchitecturePPC64 hasn't been specified explicitly determine from features. 62 if (FLAG_IS_DEFAULT(PowerArchitecturePPC64)) { 63 if (VM_Version::has_popcntw()) { 64 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7); 65 } else if (VM_Version::has_cmpb()) { 66 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 6); 67 } else if (VM_Version::has_popcntb()) { 68 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 5); 69 } else { 70 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 0); 71 } 72 } 73 guarantee(PowerArchitecturePPC64 == 0 || PowerArchitecturePPC64 == 5 || 74 PowerArchitecturePPC64 == 6 || PowerArchitecturePPC64 == 7, 75 "PowerArchitecturePPC64 should be 0, 5, 6 or 7"); 76 77 if (!UseSIGTRAP) { 78 MSG(TrapBasedICMissChecks); 79 MSG(TrapBasedNotEntrantChecks); 80 MSG(TrapBasedNullChecks); 81 FLAG_SET_ERGO(bool, TrapBasedNotEntrantChecks, false); 82 FLAG_SET_ERGO(bool, TrapBasedNullChecks, false); 83 FLAG_SET_ERGO(bool, TrapBasedICMissChecks, false); 84 } 85 86 #ifdef COMPILER2 87 if (!UseSIGTRAP) { 88 MSG(TrapBasedRangeChecks); 89 FLAG_SET_ERGO(bool, TrapBasedRangeChecks, false); 90 } 91 92 // On Power6 test for section size. 93 if (PowerArchitecturePPC64 == 6) { 94 determine_section_size(); 95 // TODO: PPC port } else { 96 // TODO: PPC port PdScheduling::power6SectorSize = 0x20; 97 } 98 99 MaxVectorSize = 8; 100 #endif 101 102 // Create and print feature-string. 103 char buf[(num_features+1) * 16]; // Max 16 chars per feature. 104 jio_snprintf(buf, sizeof(buf), 105 "ppc64%s%s%s%s%s%s%s%s%s", 106 (has_fsqrt() ? " fsqrt" : ""), 107 (has_isel() ? " isel" : ""), 108 (has_lxarxeh() ? " lxarxeh" : ""), 109 (has_cmpb() ? " cmpb" : ""), 110 //(has_mftgpr()? " mftgpr" : ""), 111 (has_popcntb() ? " popcntb" : ""), 112 (has_popcntw() ? " popcntw" : ""), 113 (has_fcfids() ? " fcfids" : ""), 114 (has_vand() ? " vand" : ""), 115 (has_vcipher() ? " aes" : "") 116 // Make sure number of %s matches num_features! 117 ); 118 _features_str = strdup(buf); 119 NOT_PRODUCT(if (Verbose) print_features();); 120 121 // PPC64 supports 8-byte compare-exchange operations (see 122 // Atomic::cmpxchg and StubGenerator::generate_atomic_cmpxchg_ptr) 123 // and 'atomic long memory ops' (see Unsafe_GetLongVolatile). 124 _supports_cx8 = true; 125 126 UseSSE = 0; // Only on x86 and x64 127 128 intx cache_line_size = _measured_cache_line_size; 129 130 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1; 131 132 if (AllocatePrefetchStyle == 4) { 133 AllocatePrefetchStepSize = cache_line_size; // Need exact value. 134 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default. 135 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined? 136 } else { 137 if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size; 138 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value. 139 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined? 140 } 141 142 assert(AllocatePrefetchLines > 0, "invalid value"); 143 if (AllocatePrefetchLines < 1) { // Set valid value in product VM. 144 AllocatePrefetchLines = 1; // Conservative value. 145 } 146 147 if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) { 148 AllocatePrefetchStyle = 1; // Fall back if inappropriate. 149 } 150 151 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive"); 152 153 // Implementation does not use any of the vector instructions 154 // available with Power8. Their exploitation is still pending. 155 if (!UseCRC32Intrinsics) { 156 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) { 157 FLAG_SET_DEFAULT(UseCRC32Intrinsics, true); 158 } 159 } 160 161 // The AES intrinsic stubs require AES instruction support. 162 #if defined(VM_LITTLE_ENDIAN) 163 if (has_vcipher()) { 164 if (FLAG_IS_DEFAULT(UseAES)) { 165 UseAES = true; 166 } 167 } else if (UseAES) { 168 if (!FLAG_IS_DEFAULT(UseAES)) 169 warning("AES instructions are not available on this CPU"); 170 FLAG_SET_DEFAULT(UseAES, false); 171 } 172 173 if (UseAES && has_vcipher()) { 174 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) { 175 UseAESIntrinsics = true; 176 } 177 } else if (UseAESIntrinsics) { 178 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) 179 warning("AES intrinsics are not available on this CPU"); 180 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 181 } 182 183 #else 184 if (UseAES) { 185 warning("AES instructions are not available on this CPU"); 186 FLAG_SET_DEFAULT(UseAES, false); 187 } 188 if (UseAESIntrinsics) { 189 if (!FLAG_IS_DEFAULT(UseAESIntrinsics)) 190 warning("AES intrinsics are not available on this CPU"); 191 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 192 } 193 #endif 194 195 if (UseSHA) { 196 warning("SHA instructions are not available on this CPU"); 197 FLAG_SET_DEFAULT(UseSHA, false); 198 } 199 if (UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics) { 200 warning("SHA intrinsics are not available on this CPU"); 201 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false); 202 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false); 203 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false); 204 } 205 206 if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) { 207 UseMontgomeryMultiplyIntrinsic = true; 208 } 209 if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) { 210 UseMontgomerySquareIntrinsic = true; 211 } 212 } 213 214 void VM_Version::print_features() { 215 tty->print_cr("Version: %s cache_line_size = %d", cpu_features(), (int) get_cache_line_size()); 216 } 217 218 #ifdef COMPILER2 219 // Determine section size on power6: If section size is 8 instructions, 220 // there should be a difference between the two testloops of ~15 %. If 221 // no difference is detected the section is assumed to be 32 instructions. 222 void VM_Version::determine_section_size() { 223 224 int unroll = 80; 225 226 const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord; 227 228 // Allocate space for the code. 229 ResourceMark rm; 230 CodeBuffer cb("detect_section_size", code_size, 0); 231 MacroAssembler* a = new MacroAssembler(&cb); 232 233 uint32_t *code = (uint32_t *)a->pc(); 234 // Emit code. 235 void (*test1)() = (void(*)())(void *)a->function_entry(); 236 237 Label l1; 238 239 a->li(R4, 1); 240 a->sldi(R4, R4, 28); 241 a->b(l1); 242 a->align(CodeEntryAlignment); 243 244 a->bind(l1); 245 246 for (int i = 0; i < unroll; i++) { 247 // Schleife 1 248 // ------- sector 0 ------------ 249 // ;; 0 250 a->nop(); // 1 251 a->fpnop0(); // 2 252 a->fpnop1(); // 3 253 a->addi(R4,R4, -1); // 4 254 255 // ;; 1 256 a->nop(); // 5 257 a->fmr(F6, F6); // 6 258 a->fmr(F7, F7); // 7 259 a->endgroup(); // 8 260 // ------- sector 8 ------------ 261 262 // ;; 2 263 a->nop(); // 9 264 a->nop(); // 10 265 a->fmr(F8, F8); // 11 266 a->fmr(F9, F9); // 12 267 268 // ;; 3 269 a->nop(); // 13 270 a->fmr(F10, F10); // 14 271 a->fmr(F11, F11); // 15 272 a->endgroup(); // 16 273 // -------- sector 16 ------------- 274 275 // ;; 4 276 a->nop(); // 17 277 a->nop(); // 18 278 a->fmr(F15, F15); // 19 279 a->fmr(F16, F16); // 20 280 281 // ;; 5 282 a->nop(); // 21 283 a->fmr(F17, F17); // 22 284 a->fmr(F18, F18); // 23 285 a->endgroup(); // 24 286 // ------- sector 24 ------------ 287 288 // ;; 6 289 a->nop(); // 25 290 a->nop(); // 26 291 a->fmr(F19, F19); // 27 292 a->fmr(F20, F20); // 28 293 294 // ;; 7 295 a->nop(); // 29 296 a->fmr(F21, F21); // 30 297 a->fmr(F22, F22); // 31 298 a->brnop0(); // 32 299 300 // ------- sector 32 ------------ 301 } 302 303 // ;; 8 304 a->cmpdi(CCR0, R4, unroll); // 33 305 a->bge(CCR0, l1); // 34 306 a->blr(); 307 308 // Emit code. 309 void (*test2)() = (void(*)())(void *)a->function_entry(); 310 // uint32_t *code = (uint32_t *)a->pc(); 311 312 Label l2; 313 314 a->li(R4, 1); 315 a->sldi(R4, R4, 28); 316 a->b(l2); 317 a->align(CodeEntryAlignment); 318 319 a->bind(l2); 320 321 for (int i = 0; i < unroll; i++) { 322 // Schleife 2 323 // ------- sector 0 ------------ 324 // ;; 0 325 a->brnop0(); // 1 326 a->nop(); // 2 327 //a->cmpdi(CCR0, R4, unroll); 328 a->fpnop0(); // 3 329 a->fpnop1(); // 4 330 a->addi(R4,R4, -1); // 5 331 332 // ;; 1 333 334 a->nop(); // 6 335 a->fmr(F6, F6); // 7 336 a->fmr(F7, F7); // 8 337 // ------- sector 8 --------------- 338 339 // ;; 2 340 a->endgroup(); // 9 341 342 // ;; 3 343 a->nop(); // 10 344 a->nop(); // 11 345 a->fmr(F8, F8); // 12 346 347 // ;; 4 348 a->fmr(F9, F9); // 13 349 a->nop(); // 14 350 a->fmr(F10, F10); // 15 351 352 // ;; 5 353 a->fmr(F11, F11); // 16 354 // -------- sector 16 ------------- 355 356 // ;; 6 357 a->endgroup(); // 17 358 359 // ;; 7 360 a->nop(); // 18 361 a->nop(); // 19 362 a->fmr(F15, F15); // 20 363 364 // ;; 8 365 a->fmr(F16, F16); // 21 366 a->nop(); // 22 367 a->fmr(F17, F17); // 23 368 369 // ;; 9 370 a->fmr(F18, F18); // 24 371 // -------- sector 24 ------------- 372 373 // ;; 10 374 a->endgroup(); // 25 375 376 // ;; 11 377 a->nop(); // 26 378 a->nop(); // 27 379 a->fmr(F19, F19); // 28 380 381 // ;; 12 382 a->fmr(F20, F20); // 29 383 a->nop(); // 30 384 a->fmr(F21, F21); // 31 385 386 // ;; 13 387 a->fmr(F22, F22); // 32 388 } 389 390 // -------- sector 32 ------------- 391 // ;; 14 392 a->cmpdi(CCR0, R4, unroll); // 33 393 a->bge(CCR0, l2); // 34 394 395 a->blr(); 396 uint32_t *code_end = (uint32_t *)a->pc(); 397 a->flush(); 398 399 double loop1_seconds,loop2_seconds, rel_diff; 400 uint64_t start1, stop1; 401 402 start1 = os::current_thread_cpu_time(false); 403 (*test1)(); 404 stop1 = os::current_thread_cpu_time(false); 405 loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0); 406 407 408 start1 = os::current_thread_cpu_time(false); 409 (*test2)(); 410 stop1 = os::current_thread_cpu_time(false); 411 412 loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0); 413 414 rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100; 415 416 if (PrintAssembly) { 417 ttyLocker ttyl; 418 tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 419 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 420 tty->print_cr("Time loop1 :%f", loop1_seconds); 421 tty->print_cr("Time loop2 :%f", loop2_seconds); 422 tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff); 423 424 if (rel_diff > 12.0) { 425 tty->print_cr("Section Size 8 Instructions"); 426 } else{ 427 tty->print_cr("Section Size 32 Instructions or Power5"); 428 } 429 } 430 431 #if 0 // TODO: PPC port 432 // Set sector size (if not set explicitly). 433 if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) { 434 if (rel_diff > 12.0) { 435 PdScheduling::power6SectorSize = 0x20; 436 } else { 437 PdScheduling::power6SectorSize = 0x80; 438 } 439 } else if (Power6SectorSize128PPC64) { 440 PdScheduling::power6SectorSize = 0x80; 441 } else { 442 PdScheduling::power6SectorSize = 0x20; 443 } 444 #endif 445 if (UsePower6SchedulerPPC64) Unimplemented(); 446 } 447 #endif // COMPILER2 448 449 void VM_Version::determine_features() { 450 #if defined(ABI_ELFv2) 451 const int code_size = (num_features+1+2*7)*BytesPerInstWord; // TODO(asmundak): calculation is incorrect. 452 #else 453 // 7 InstWords for each call (function descriptor + blr instruction). 454 const int code_size = (num_features+1+2*7)*BytesPerInstWord; 455 #endif 456 int features = 0; 457 458 // create test area 459 enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size). 460 char test_area[BUFFER_SIZE]; 461 char *mid_of_test_area = &test_area[BUFFER_SIZE>>1]; 462 463 // Allocate space for the code. 464 ResourceMark rm; 465 CodeBuffer cb("detect_cpu_features", code_size, 0); 466 MacroAssembler* a = new MacroAssembler(&cb); 467 468 // Must be set to true so we can generate the test code. 469 _features = VM_Version::all_features_m; 470 471 // Emit code. 472 void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry(); 473 uint32_t *code = (uint32_t *)a->pc(); 474 // Don't use R0 in ldarx. 475 // Keep R3_ARG1 unmodified, it contains &field (see below). 476 // Keep R4_ARG2 unmodified, it contains offset = 0 (see below). 477 a->fsqrt(F3, F4); // code[0] -> fsqrt_m 478 a->fsqrts(F3, F4); // code[1] -> fsqrts_m 479 a->isel(R7, R5, R6, 0); // code[2] -> isel_m 480 a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m 481 a->cmpb(R7, R5, R6); // code[4] -> bcmp 482 //a->mftgpr(R7, F3); // code[5] -> mftgpr 483 a->popcntb(R7, R5); // code[6] -> popcntb 484 a->popcntw(R7, R5); // code[7] -> popcntw 485 a->fcfids(F3, F4); // code[8] -> fcfids 486 a->vand(VR0, VR0, VR0); // code[9] -> vand 487 a->vcipher(VR0, VR1, VR2); // code[10] -> vcipher 488 a->blr(); 489 490 // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it. 491 void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry(); 492 a->dcbz(R3_ARG1); // R3_ARG1 = addr 493 a->blr(); 494 495 uint32_t *code_end = (uint32_t *)a->pc(); 496 a->flush(); 497 _features = VM_Version::unknown_m; 498 499 // Print the detection code. 500 if (PrintAssembly) { 501 ttyLocker ttyl; 502 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code)); 503 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 504 } 505 506 // Measure cache line size. 507 memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF. 508 (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle. 509 int count = 0; // count zeroed bytes 510 for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++; 511 guarantee(is_power_of_2(count), "cache line size needs to be a power of 2"); 512 _measured_cache_line_size = count; 513 514 // Execute code. Illegal instructions will be replaced by 0 in the signal handler. 515 VM_Version::_is_determine_features_test_running = true; 516 (*test)((address)mid_of_test_area, (uint64_t)0); 517 VM_Version::_is_determine_features_test_running = false; 518 519 // determine which instructions are legal. 520 int feature_cntr = 0; 521 if (code[feature_cntr++]) features |= fsqrt_m; 522 if (code[feature_cntr++]) features |= fsqrts_m; 523 if (code[feature_cntr++]) features |= isel_m; 524 if (code[feature_cntr++]) features |= lxarxeh_m; 525 if (code[feature_cntr++]) features |= cmpb_m; 526 //if(code[feature_cntr++])features |= mftgpr_m; 527 if (code[feature_cntr++]) features |= popcntb_m; 528 if (code[feature_cntr++]) features |= popcntw_m; 529 if (code[feature_cntr++]) features |= fcfids_m; 530 if (code[feature_cntr++]) features |= vand_m; 531 if (code[feature_cntr++]) features |= vcipher_m; 532 533 // Print the detection code. 534 if (PrintAssembly) { 535 ttyLocker ttyl; 536 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code)); 537 Disassembler::decode((u_char*)code, (u_char*)code_end, tty); 538 } 539 540 _features = features; 541 } 542 543 544 static int saved_features = 0; 545 546 void VM_Version::allow_all() { 547 saved_features = _features; 548 _features = all_features_m; 549 } 550 551 void VM_Version::revert() { 552 _features = saved_features; 553 }