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