1 /*
   2  * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright 2012, 2018 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 uint64_t VM_Version::_dscr_val = 0;
  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_lqarx()) {
  64       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 8);
  65     } else if (VM_Version::has_popcntw()) {
  66       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7);
  67     } else if (VM_Version::has_cmpb()) {
  68       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 6);
  69     } else if (VM_Version::has_popcntb()) {
  70       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 5);
  71     } else {
  72       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 0);
  73     }
  74   }
  75   guarantee(PowerArchitecturePPC64 == 0 || PowerArchitecturePPC64 == 5 ||
  76             PowerArchitecturePPC64 == 6 || PowerArchitecturePPC64 == 7 ||
  77             PowerArchitecturePPC64 == 8,
  78             "PowerArchitecturePPC64 should be 0, 5, 6, 7, or 8");
  79 
  80   // Power 8: Configure Data Stream Control Register.
  81   if (PowerArchitecturePPC64 >= 8) {
  82     config_dscr();
  83   }
  84 
  85   if (!UseSIGTRAP) {
  86     MSG(TrapBasedICMissChecks);
  87     MSG(TrapBasedNotEntrantChecks);
  88     MSG(TrapBasedNullChecks);
  89     FLAG_SET_ERGO(bool, TrapBasedNotEntrantChecks, false);
  90     FLAG_SET_ERGO(bool, TrapBasedNullChecks,       false);
  91     FLAG_SET_ERGO(bool, TrapBasedICMissChecks,     false);
  92   }
  93 
  94 #ifdef COMPILER2
  95   if (!UseSIGTRAP) {
  96     MSG(TrapBasedRangeChecks);
  97     FLAG_SET_ERGO(bool, TrapBasedRangeChecks, false);
  98   }
  99 
 100   // On Power6 test for section size.
 101   if (PowerArchitecturePPC64 == 6) {
 102     determine_section_size();
 103   // TODO: PPC port } else {
 104   // TODO: PPC port PdScheduling::power6SectorSize = 0x20;
 105   }
 106 
 107   MaxVectorSize = 8;
 108 #endif
 109 
 110   // Create and print feature-string.
 111   char buf[(num_features+1) * 16]; // Max 16 chars per feature.
 112   jio_snprintf(buf, sizeof(buf),
 113                "ppc64%s%s%s%s%s%s%s%s%s%s%s%s%s",
 114                (has_fsqrt()   ? " fsqrt"   : ""),
 115                (has_isel()    ? " isel"    : ""),
 116                (has_lxarxeh() ? " lxarxeh" : ""),
 117                (has_cmpb()    ? " cmpb"    : ""),
 118                //(has_mftgpr()? " mftgpr"  : ""),
 119                (has_popcntb() ? " popcntb" : ""),
 120                (has_popcntw() ? " popcntw" : ""),
 121                (has_fcfids()  ? " fcfids"  : ""),
 122                (has_vand()    ? " vand"    : ""),
 123                (has_lqarx()   ? " lqarx"   : ""),
 124                (has_vcipher() ? " aes"     : ""),
 125                (has_vpmsumb() ? " vpmsumb" : ""),
 126                (has_mfdscr()  ? " mfdscr"  : ""),
 127                (has_vsx()     ? " vsx"     : "")
 128                // Make sure number of %s matches num_features!
 129               );
 130   _features_str = strdup(buf);
 131   if (Verbose) {
 132     print_features();
 133   }
 134 
 135   // PPC64 supports 8-byte compare-exchange operations (see
 136   // Atomic::cmpxchg and StubGenerator::generate_atomic_cmpxchg_ptr)
 137   // and 'atomic long memory ops' (see Unsafe_GetLongVolatile).
 138   _supports_cx8 = true;
 139 
 140   UseSSE = 0; // Only on x86 and x64
 141 
 142   intx cache_line_size = _measured_cache_line_size;
 143 
 144   if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1;
 145 
 146   if (AllocatePrefetchStyle == 4) {
 147     AllocatePrefetchStepSize = cache_line_size; // Need exact value.
 148     if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default.
 149     if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined?
 150   } else {
 151     if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size;
 152     if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value.
 153     if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined?
 154   }
 155 
 156   assert(AllocatePrefetchLines > 0, "invalid value");
 157   if (AllocatePrefetchLines < 1) { // Set valid value in product VM.
 158     AllocatePrefetchLines = 1; // Conservative value.
 159   }
 160 
 161   if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) {
 162     AllocatePrefetchStyle = 1; // Fall back if inappropriate.
 163   }
 164 
 165   assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive");
 166 
 167   // Implementation does not use any of the vector instructions
 168   // available with Power8. Their exploitation is still pending.
 169   if (!UseCRC32Intrinsics) {
 170     if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) {
 171       FLAG_SET_DEFAULT(UseCRC32Intrinsics, true);
 172     }
 173   }
 174 
 175   // The AES intrinsic stubs require AES instruction support.
 176 #if defined(VM_LITTLE_ENDIAN)
 177   if (has_vcipher()) {
 178     if (FLAG_IS_DEFAULT(UseAES)) {
 179       UseAES = true;
 180     }
 181   } else if (UseAES) {
 182     if (!FLAG_IS_DEFAULT(UseAES))
 183       warning("AES instructions are not available on this CPU");
 184     FLAG_SET_DEFAULT(UseAES, false);
 185   }
 186 
 187   if (UseAES && has_vcipher()) {
 188     if (FLAG_IS_DEFAULT(UseAESIntrinsics)) {
 189       UseAESIntrinsics = true;
 190     }
 191   } else if (UseAESIntrinsics) {
 192     if (!FLAG_IS_DEFAULT(UseAESIntrinsics))
 193       warning("AES intrinsics are not available on this CPU");
 194     FLAG_SET_DEFAULT(UseAESIntrinsics, false);
 195   }
 196 
 197 #else
 198   if (UseAES) {
 199     warning("AES instructions are not available on this CPU");
 200     FLAG_SET_DEFAULT(UseAES, false);
 201   }
 202   if (UseAESIntrinsics) {
 203     if (!FLAG_IS_DEFAULT(UseAESIntrinsics))
 204       warning("AES intrinsics are not available on this CPU");
 205     FLAG_SET_DEFAULT(UseAESIntrinsics, false);
 206   }
 207 #endif
 208 
 209   if (UseSHA) {
 210     warning("SHA instructions are not available on this CPU");
 211     FLAG_SET_DEFAULT(UseSHA, false);
 212   }
 213   if (UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics) {
 214     warning("SHA intrinsics are not available on this CPU");
 215     FLAG_SET_DEFAULT(UseSHA1Intrinsics, false);
 216     FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
 217     FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
 218   }
 219 
 220   if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) {
 221     UseMontgomeryMultiplyIntrinsic = true;
 222   }
 223   if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) {
 224     UseMontgomerySquareIntrinsic = true;
 225   }
 226 }
 227 
 228 void VM_Version::print_features() {
 229   tty->print_cr("Version: %s cache_line_size = %d", cpu_features(), (int) get_cache_line_size());
 230 }
 231 
 232 #ifdef COMPILER2
 233 // Determine section size on power6: If section size is 8 instructions,
 234 // there should be a difference between the two testloops of ~15 %. If
 235 // no difference is detected the section is assumed to be 32 instructions.
 236 void VM_Version::determine_section_size() {
 237 
 238   int unroll = 80;
 239 
 240   const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord;
 241 
 242   // Allocate space for the code.
 243   ResourceMark rm;
 244   CodeBuffer cb("detect_section_size", code_size, 0);
 245   MacroAssembler* a = new MacroAssembler(&cb);
 246 
 247   uint32_t *code = (uint32_t *)a->pc();
 248   // Emit code.
 249   void (*test1)() = (void(*)())(void *)a->function_entry();
 250 
 251   Label l1;
 252 
 253   a->li(R4, 1);
 254   a->sldi(R4, R4, 28);
 255   a->b(l1);
 256   a->align(CodeEntryAlignment);
 257 
 258   a->bind(l1);
 259 
 260   for (int i = 0; i < unroll; i++) {
 261     // Schleife 1
 262     // ------- sector 0 ------------
 263     // ;; 0
 264     a->nop();                   // 1
 265     a->fpnop0();                // 2
 266     a->fpnop1();                // 3
 267     a->addi(R4,R4, -1); // 4
 268 
 269     // ;;  1
 270     a->nop();                   // 5
 271     a->fmr(F6, F6);             // 6
 272     a->fmr(F7, F7);             // 7
 273     a->endgroup();              // 8
 274     // ------- sector 8 ------------
 275 
 276     // ;;  2
 277     a->nop();                   // 9
 278     a->nop();                   // 10
 279     a->fmr(F8, F8);             // 11
 280     a->fmr(F9, F9);             // 12
 281 
 282     // ;;  3
 283     a->nop();                   // 13
 284     a->fmr(F10, F10);           // 14
 285     a->fmr(F11, F11);           // 15
 286     a->endgroup();              // 16
 287     // -------- sector 16 -------------
 288 
 289     // ;;  4
 290     a->nop();                   // 17
 291     a->nop();                   // 18
 292     a->fmr(F15, F15);           // 19
 293     a->fmr(F16, F16);           // 20
 294 
 295     // ;;  5
 296     a->nop();                   // 21
 297     a->fmr(F17, F17);           // 22
 298     a->fmr(F18, F18);           // 23
 299     a->endgroup();              // 24
 300     // ------- sector 24  ------------
 301 
 302     // ;;  6
 303     a->nop();                   // 25
 304     a->nop();                   // 26
 305     a->fmr(F19, F19);           // 27
 306     a->fmr(F20, F20);           // 28
 307 
 308     // ;;  7
 309     a->nop();                   // 29
 310     a->fmr(F21, F21);           // 30
 311     a->fmr(F22, F22);           // 31
 312     a->brnop0();                // 32
 313 
 314     // ------- sector 32 ------------
 315   }
 316 
 317   // ;; 8
 318   a->cmpdi(CCR0, R4, unroll);   // 33
 319   a->bge(CCR0, l1);             // 34
 320   a->blr();
 321 
 322   // Emit code.
 323   void (*test2)() = (void(*)())(void *)a->function_entry();
 324   // uint32_t *code = (uint32_t *)a->pc();
 325 
 326   Label l2;
 327 
 328   a->li(R4, 1);
 329   a->sldi(R4, R4, 28);
 330   a->b(l2);
 331   a->align(CodeEntryAlignment);
 332 
 333   a->bind(l2);
 334 
 335   for (int i = 0; i < unroll; i++) {
 336     // Schleife 2
 337     // ------- sector 0 ------------
 338     // ;; 0
 339     a->brnop0();                  // 1
 340     a->nop();                     // 2
 341     //a->cmpdi(CCR0, R4, unroll);
 342     a->fpnop0();                  // 3
 343     a->fpnop1();                  // 4
 344     a->addi(R4,R4, -1);           // 5
 345 
 346     // ;; 1
 347 
 348     a->nop();                     // 6
 349     a->fmr(F6, F6);               // 7
 350     a->fmr(F7, F7);               // 8
 351     // ------- sector 8 ---------------
 352 
 353     // ;; 2
 354     a->endgroup();                // 9
 355 
 356     // ;; 3
 357     a->nop();                     // 10
 358     a->nop();                     // 11
 359     a->fmr(F8, F8);               // 12
 360 
 361     // ;; 4
 362     a->fmr(F9, F9);               // 13
 363     a->nop();                     // 14
 364     a->fmr(F10, F10);             // 15
 365 
 366     // ;; 5
 367     a->fmr(F11, F11);             // 16
 368     // -------- sector 16 -------------
 369 
 370     // ;; 6
 371     a->endgroup();                // 17
 372 
 373     // ;; 7
 374     a->nop();                     // 18
 375     a->nop();                     // 19
 376     a->fmr(F15, F15);             // 20
 377 
 378     // ;; 8
 379     a->fmr(F16, F16);             // 21
 380     a->nop();                     // 22
 381     a->fmr(F17, F17);             // 23
 382 
 383     // ;; 9
 384     a->fmr(F18, F18);             // 24
 385     // -------- sector 24 -------------
 386 
 387     // ;; 10
 388     a->endgroup();                // 25
 389 
 390     // ;; 11
 391     a->nop();                     // 26
 392     a->nop();                     // 27
 393     a->fmr(F19, F19);             // 28
 394 
 395     // ;; 12
 396     a->fmr(F20, F20);             // 29
 397     a->nop();                     // 30
 398     a->fmr(F21, F21);             // 31
 399 
 400     // ;; 13
 401     a->fmr(F22, F22);             // 32
 402   }
 403 
 404   // -------- sector 32 -------------
 405   // ;; 14
 406   a->cmpdi(CCR0, R4, unroll); // 33
 407   a->bge(CCR0, l2);           // 34
 408 
 409   a->blr();
 410   uint32_t *code_end = (uint32_t *)a->pc();
 411   a->flush();
 412 
 413   double loop1_seconds,loop2_seconds, rel_diff;
 414   uint64_t start1, stop1;
 415 
 416   start1 = os::current_thread_cpu_time(false);
 417   (*test1)();
 418   stop1 = os::current_thread_cpu_time(false);
 419   loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0);
 420 
 421 
 422   start1 = os::current_thread_cpu_time(false);
 423   (*test2)();
 424   stop1 = os::current_thread_cpu_time(false);
 425 
 426   loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0);
 427 
 428   rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100;
 429 
 430   if (PrintAssembly) {
 431     ttyLocker ttyl;
 432     tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", p2i(code));
 433     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 434     tty->print_cr("Time loop1 :%f", loop1_seconds);
 435     tty->print_cr("Time loop2 :%f", loop2_seconds);
 436     tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff);
 437 
 438     if (rel_diff > 12.0) {
 439       tty->print_cr("Section Size 8 Instructions");
 440     } else{
 441       tty->print_cr("Section Size 32 Instructions or Power5");
 442     }
 443   }
 444 
 445 #if 0 // TODO: PPC port
 446   // Set sector size (if not set explicitly).
 447   if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) {
 448     if (rel_diff > 12.0) {
 449       PdScheduling::power6SectorSize = 0x20;
 450     } else {
 451       PdScheduling::power6SectorSize = 0x80;
 452     }
 453   } else if (Power6SectorSize128PPC64) {
 454     PdScheduling::power6SectorSize = 0x80;
 455   } else {
 456     PdScheduling::power6SectorSize = 0x20;
 457   }
 458 #endif
 459   if (UsePower6SchedulerPPC64) Unimplemented();
 460 }
 461 #endif // COMPILER2
 462 
 463 void VM_Version::determine_features() {
 464 #if defined(ABI_ELFv2)
 465   const int code_size = (num_features+1+2*7)*BytesPerInstWord; // TODO(asmundak): calculation is incorrect.
 466 #else
 467   // 7 InstWords for each call (function descriptor + blr instruction).
 468   const int code_size = (num_features+1+2*7)*BytesPerInstWord;
 469 #endif
 470   int features = 0;
 471 
 472   // create test area
 473   enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size).
 474   char test_area[BUFFER_SIZE];
 475   char *mid_of_test_area = &test_area[BUFFER_SIZE>>1];
 476 
 477   // Allocate space for the code.
 478   ResourceMark rm;
 479   CodeBuffer cb("detect_cpu_features", code_size, 0);
 480   MacroAssembler* a = new MacroAssembler(&cb);
 481 
 482   // Must be set to true so we can generate the test code.
 483   _features = VM_Version::all_features_m;
 484 
 485   // Emit code.
 486   void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry();
 487   uint32_t *code = (uint32_t *)a->pc();
 488   // Don't use R0 in ldarx.
 489   // Keep R3_ARG1 unmodified, it contains &field (see below).
 490   // Keep R4_ARG2 unmodified, it contains offset = 0 (see below).
 491   a->fsqrt(F3, F4);                            // code[0] -> fsqrt_m
 492   a->fsqrts(F3, F4);                           // code[1] -> fsqrts_m
 493   a->isel(R7, R5, R6, 0);                      // code[2] -> isel_m
 494   a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m
 495   a->cmpb(R7, R5, R6);                         // code[4] -> bcmp
 496   //a->mftgpr(R7, F3);                         // code[5] -> mftgpr
 497   a->popcntb(R7, R5);                          // code[6] -> popcntb
 498   a->popcntw(R7, R5);                          // code[7] -> popcntw
 499   a->fcfids(F3, F4);                           // code[8] -> fcfids
 500   a->vand(VR0, VR0, VR0);                      // code[9] -> vand
 501   a->lqarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[10] -> lqarx_m
 502   a->vcipher(VR0, VR1, VR2);                   // code[11] -> vcipher
 503   a->vpmsumb(VR0, VR1, VR2);                   // code[12] -> vpmsumb
 504   a->mfdscr(R0);                               // code[13] -> mfdscr
 505   a->lxvd2x(VSR0, 0, R3_ARG1);                 // code[14] -> vsx
 506   a->blr();
 507 
 508   // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it.
 509   void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry();
 510   a->dcbz(R3_ARG1); // R3_ARG1 = addr
 511   a->blr();
 512 
 513   uint32_t *code_end = (uint32_t *)a->pc();
 514   a->flush();
 515   _features = VM_Version::unknown_m;
 516 
 517   // Print the detection code.
 518   if (PrintAssembly) {
 519     ttyLocker ttyl;
 520     tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", p2i(code));
 521     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 522   }
 523 
 524   // Measure cache line size.
 525   memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF.
 526   (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle.
 527   int count = 0; // count zeroed bytes
 528   for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++;
 529   guarantee(is_power_of_2(count), "cache line size needs to be a power of 2");
 530   _measured_cache_line_size = count;
 531 
 532   // Execute code. Illegal instructions will be replaced by 0 in the signal handler.
 533   VM_Version::_is_determine_features_test_running = true;
 534   (*test)((address)mid_of_test_area, (uint64_t)0);
 535   VM_Version::_is_determine_features_test_running = false;
 536 
 537   // determine which instructions are legal.
 538   int feature_cntr = 0;
 539   if (code[feature_cntr++]) features |= fsqrt_m;
 540   if (code[feature_cntr++]) features |= fsqrts_m;
 541   if (code[feature_cntr++]) features |= isel_m;
 542   if (code[feature_cntr++]) features |= lxarxeh_m;
 543   if (code[feature_cntr++]) features |= cmpb_m;
 544   //if(code[feature_cntr++])features |= mftgpr_m;
 545   if (code[feature_cntr++]) features |= popcntb_m;
 546   if (code[feature_cntr++]) features |= popcntw_m;
 547   if (code[feature_cntr++]) features |= fcfids_m;
 548   if (code[feature_cntr++]) features |= vand_m;
 549   if (code[feature_cntr++]) features |= lqarx_m;
 550   if (code[feature_cntr++]) features |= vcipher_m;
 551   if (code[feature_cntr++]) features |= vpmsumb_m;
 552   if (code[feature_cntr++]) features |= mfdscr_m;
 553   if (code[feature_cntr++]) features |= vsx_m;
 554 
 555   // Print the detection code.
 556   if (PrintAssembly) {
 557     ttyLocker ttyl;
 558     tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", p2i(code));
 559     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 560   }
 561 
 562   _features = features;
 563 }
 564 
 565 // Power 8: Configure Data Stream Control Register.
 566 void VM_Version::config_dscr() {
 567   assert(has_lqarx(), "Only execute on Power 8 or later!");
 568 
 569   // 7 InstWords for each call (function descriptor + blr instruction).
 570   const int code_size = (2+2*7)*BytesPerInstWord;
 571 
 572   // Allocate space for the code.
 573   ResourceMark rm;
 574   CodeBuffer cb("config_dscr", code_size, 0);
 575   MacroAssembler* a = new MacroAssembler(&cb);
 576 
 577   // Emit code.
 578   uint64_t (*get_dscr)() = (uint64_t(*)())(void *)a->function_entry();
 579   uint32_t *code = (uint32_t *)a->pc();
 580   a->mfdscr(R3);
 581   a->blr();
 582 
 583   void (*set_dscr)(long) = (void(*)(long))(void *)a->function_entry();
 584   a->mtdscr(R3);
 585   a->blr();
 586 
 587   uint32_t *code_end = (uint32_t *)a->pc();
 588   a->flush();
 589 
 590   // Print the detection code.
 591   if (PrintAssembly) {
 592     ttyLocker ttyl;
 593     tty->print_cr("Decoding dscr configuration stub at " INTPTR_FORMAT " before execution:", code);
 594     Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
 595   }
 596 
 597   // Apply the configuration if needed.
 598   _dscr_val = (*get_dscr)();
 599   if (Verbose) {
 600     tty->print_cr("dscr value was 0x%lx" , _dscr_val);
 601   }
 602   bool change_requested = false;
 603   if (DSCR_PPC64 != (uintx)-1) {
 604     _dscr_val = DSCR_PPC64;
 605     change_requested = true;
 606   }
 607   if (DSCR_DPFD_PPC64 <= 7) {
 608     uint64_t mask = 0x7;
 609     if ((_dscr_val & mask) != DSCR_DPFD_PPC64) {
 610       _dscr_val = (_dscr_val & ~mask) | (DSCR_DPFD_PPC64);
 611       change_requested = true;
 612     }
 613   }
 614   if (DSCR_URG_PPC64 <= 7) {
 615     uint64_t mask = 0x7 << 6;
 616     if ((_dscr_val & mask) != DSCR_DPFD_PPC64 << 6) {
 617       _dscr_val = (_dscr_val & ~mask) | (DSCR_URG_PPC64 << 6);
 618       change_requested = true;
 619     }
 620   }
 621   if (change_requested) {
 622     (*set_dscr)(_dscr_val);
 623     if (Verbose) {
 624       tty->print_cr("dscr was set to 0x%lx" , (*get_dscr)());
 625     }
 626   }
 627 }
 628 
 629 static int saved_features = 0;
 630 
 631 void VM_Version::allow_all() {
 632   saved_features = _features;
 633   _features      = all_features_m;
 634 }
 635 
 636 void VM_Version::revert() {
 637   _features = saved_features;
 638 }