< prev index next >

src/hotspot/cpu/x86/sharedRuntime_x86_64.cpp

Print this page

  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #ifndef _WINDOWS
  27 #include "alloca.h"
  28 #endif
  29 #include "asm/macroAssembler.hpp"
  30 #include "asm/macroAssembler.inline.hpp"

  31 #include "code/compiledIC.hpp"
  32 #include "code/debugInfoRec.hpp"
  33 #include "code/icBuffer.hpp"
  34 #include "code/nativeInst.hpp"
  35 #include "code/vtableStubs.hpp"
  36 #include "compiler/oopMap.hpp"
  37 #include "gc/shared/collectedHeap.hpp"
  38 #include "gc/shared/gcLocker.hpp"
  39 #include "gc/shared/barrierSet.hpp"
  40 #include "gc/shared/barrierSetAssembler.hpp"
  41 #include "interpreter/interpreter.hpp"
  42 #include "logging/log.hpp"
  43 #include "memory/resourceArea.hpp"
  44 #include "memory/universe.hpp"
  45 #include "oops/compiledICHolder.hpp"
  46 #include "oops/klass.inline.hpp"
  47 #include "oops/method.inline.hpp"
  48 #include "prims/methodHandles.hpp"
  49 #include "runtime/continuation.hpp"
  50 #include "runtime/continuationEntry.inline.hpp"

 507     case T_SHORT:
 508     case T_INT:
 509       if (int_args < Argument::n_int_register_parameters_j) {
 510         regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
 511       } else {
 512         regs[i].set1(VMRegImpl::stack2reg(stk_args));
 513         stk_args += 2;
 514       }
 515       break;
 516     case T_VOID:
 517       // halves of T_LONG or T_DOUBLE
 518       assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
 519       regs[i].set_bad();
 520       break;
 521     case T_LONG:
 522       assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
 523       // fall through
 524     case T_OBJECT:
 525     case T_ARRAY:
 526     case T_ADDRESS:

 527       if (int_args < Argument::n_int_register_parameters_j) {
 528         regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
 529       } else {
 530         regs[i].set2(VMRegImpl::stack2reg(stk_args));
 531         stk_args += 2;
 532       }
 533       break;
 534     case T_FLOAT:
 535       if (fp_args < Argument::n_float_register_parameters_j) {
 536         regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
 537       } else {
 538         regs[i].set1(VMRegImpl::stack2reg(stk_args));
 539         stk_args += 2;
 540       }
 541       break;
 542     case T_DOUBLE:
 543       assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
 544       if (fp_args < Argument::n_float_register_parameters_j) {
 545         regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
 546       } else {
 547         regs[i].set2(VMRegImpl::stack2reg(stk_args));
 548         stk_args += 2;
 549       }
 550       break;
 551     default:
 552       ShouldNotReachHere();
 553       break;
 554     }
 555   }
 556 
 557   return align_up(stk_args, 2);
 558 }
 559 


















































































 560 // Patch the callers callsite with entry to compiled code if it exists.
 561 static void patch_callers_callsite(MacroAssembler *masm) {
 562   Label L;
 563   __ cmpptr(Address(rbx, in_bytes(Method::code_offset())), NULL_WORD);
 564   __ jcc(Assembler::equal, L);
 565 
 566   // Save the current stack pointer
 567   __ mov(r13, rsp);
 568   // Schedule the branch target address early.
 569   // Call into the VM to patch the caller, then jump to compiled callee
 570   // rax isn't live so capture return address while we easily can
 571   __ movptr(rax, Address(rsp, 0));
 572 
 573   // align stack so push_CPU_state doesn't fault
 574   __ andptr(rsp, -(StackAlignmentInBytes));
 575   __ push_CPU_state();
 576   __ vzeroupper();
 577   // VM needs caller's callsite
 578   // VM needs target method
 579   // This needs to be a long call since we will relocate this adapter to

 582   // Allocate argument register save area
 583   if (frame::arg_reg_save_area_bytes != 0) {
 584     __ subptr(rsp, frame::arg_reg_save_area_bytes);
 585   }
 586   __ mov(c_rarg0, rbx);
 587   __ mov(c_rarg1, rax);
 588   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)));
 589 
 590   // De-allocate argument register save area
 591   if (frame::arg_reg_save_area_bytes != 0) {
 592     __ addptr(rsp, frame::arg_reg_save_area_bytes);
 593   }
 594 
 595   __ vzeroupper();
 596   __ pop_CPU_state();
 597   // restore sp
 598   __ mov(rsp, r13);
 599   __ bind(L);
 600 }
 601 










































































































 602 
 603 static void gen_c2i_adapter(MacroAssembler *masm,
 604                             int total_args_passed,
 605                             int comp_args_on_stack,
 606                             const BasicType *sig_bt,
 607                             const VMRegPair *regs,
 608                             Label& skip_fixup) {































 609   // Before we get into the guts of the C2I adapter, see if we should be here
 610   // at all.  We've come from compiled code and are attempting to jump to the
 611   // interpreter, which means the caller made a static call to get here
 612   // (vcalls always get a compiled target if there is one).  Check for a
 613   // compiled target.  If there is one, we need to patch the caller's call.
 614   patch_callers_callsite(masm);
 615 
 616   __ bind(skip_fixup);
 617 










































 618   // Since all args are passed on the stack, total_args_passed *
 619   // Interpreter::stackElementSize is the space we need.
 620 
 621   assert(total_args_passed >= 0, "total_args_passed is %d", total_args_passed);
 622 
 623   int extraspace = (total_args_passed * Interpreter::stackElementSize);
 624 
 625   // stack is aligned, keep it that way
 626   // This is not currently needed or enforced by the interpreter, but
 627   // we might as well conform to the ABI.
 628   extraspace = align_up(extraspace, 2*wordSize);
 629 
 630   // set senderSP value
 631   __ lea(r13, Address(rsp, wordSize));
 632 
 633 #ifdef ASSERT
 634   __ check_stack_alignment(r13, "sender stack not aligned");
 635 #endif
 636   if (extraspace > 0) {
 637     // Pop the return address
 638     __ pop(rax);
 639 
 640     __ subptr(rsp, extraspace);
 641 
 642     // Push the return address
 643     __ push(rax);
 644 
 645     // Account for the return address location since we store it first rather
 646     // than hold it in a register across all the shuffling
 647     extraspace += wordSize;
 648   }
 649 
 650 #ifdef ASSERT
 651   __ check_stack_alignment(rsp, "callee stack not aligned", wordSize, rax);
 652 #endif
 653 
 654   // Now write the args into the outgoing interpreter space
 655   for (int i = 0; i < total_args_passed; i++) {
 656     if (sig_bt[i] == T_VOID) {
 657       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
 658       continue;
 659     }
 660 
 661     // offset to start parameters
 662     int st_off   = (total_args_passed - i) * Interpreter::stackElementSize;
 663     int next_off = st_off - Interpreter::stackElementSize;
 664 
 665     // Say 4 args:
 666     // i   st_off
 667     // 0   32 T_LONG
 668     // 1   24 T_VOID
 669     // 2   16 T_OBJECT
 670     // 3    8 T_BOOL
 671     // -    0 return address
 672     //
 673     // However to make thing extra confusing. Because we can fit a long/double in
 674     // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
 675     // leaves one slot empty and only stores to a single slot. In this case the
 676     // slot that is occupied is the T_VOID slot. See I said it was confusing.
 677 
 678     VMReg r_1 = regs[i].first();
 679     VMReg r_2 = regs[i].second();
 680     if (!r_1->is_valid()) {
 681       assert(!r_2->is_valid(), "");
 682       continue;
 683     }
 684     if (r_1->is_stack()) {
 685       // memory to memory use rax
 686       int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace;
 687       if (!r_2->is_valid()) {
 688         // sign extend??
 689         __ movl(rax, Address(rsp, ld_off));
 690         __ movptr(Address(rsp, st_off), rax);
 691 
 692       } else {
 693 
 694         __ movq(rax, Address(rsp, ld_off));
 695 
 696         // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
 697         // T_DOUBLE and T_LONG use two slots in the interpreter
 698         if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
 699           // ld_off == LSW, ld_off+wordSize == MSW
 700           // st_off == MSW, next_off == LSW
 701           __ movq(Address(rsp, next_off), rax);
 702 #ifdef ASSERT
 703           // Overwrite the unused slot with known junk
 704           __ mov64(rax, CONST64(0xdeadffffdeadaaaa));
 705           __ movptr(Address(rsp, st_off), rax);
 706 #endif /* ASSERT */
 707         } else {
 708           __ movq(Address(rsp, st_off), rax);
 709         }
 710       }
 711     } else if (r_1->is_Register()) {
 712       Register r = r_1->as_Register();
 713       if (!r_2->is_valid()) {
 714         // must be only an int (or less ) so move only 32bits to slot
 715         // why not sign extend??
 716         __ movl(Address(rsp, st_off), r);
 717       } else {
 718         // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
 719         // T_DOUBLE and T_LONG use two slots in the interpreter
 720         if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
 721           // long/double in gpr
 722 #ifdef ASSERT
 723           // Overwrite the unused slot with known junk
 724           __ mov64(rax, CONST64(0xdeadffffdeadaaab));
 725           __ movptr(Address(rsp, st_off), rax);
 726 #endif /* ASSERT */
 727           __ movq(Address(rsp, next_off), r);

























 728         } else {
 729           __ movptr(Address(rsp, st_off), r);





















 730         }
 731       }
 732     } else {
 733       assert(r_1->is_XMMRegister(), "");
 734       if (!r_2->is_valid()) {
 735         // only a float use just part of the slot
 736         __ movflt(Address(rsp, st_off), r_1->as_XMMRegister());
 737       } else {
 738 #ifdef ASSERT
 739         // Overwrite the unused slot with known junk
 740         __ mov64(rax, CONST64(0xdeadffffdeadaaac));
 741         __ movptr(Address(rsp, st_off), rax);
 742 #endif /* ASSERT */
 743         __ movdbl(Address(rsp, next_off), r_1->as_XMMRegister());
 744       }
 745     }
 746   }
 747 
 748   // Schedule the branch target address early.
 749   __ movptr(rcx, Address(rbx, in_bytes(Method::interpreter_entry_offset())));
 750   __ jmp(rcx);
 751 }
 752 
 753 static void range_check(MacroAssembler* masm, Register pc_reg, Register temp_reg,
 754                         address code_start, address code_end,
 755                         Label& L_ok) {
 756   Label L_fail;
 757   __ lea(temp_reg, ExternalAddress(code_start));
 758   __ cmpptr(pc_reg, temp_reg);
 759   __ jcc(Assembler::belowEqual, L_fail);
 760   __ lea(temp_reg, ExternalAddress(code_end));
 761   __ cmpptr(pc_reg, temp_reg);
 762   __ jcc(Assembler::below, L_ok);
 763   __ bind(L_fail);
 764 }
 765 
 766 void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,
 767                                     int total_args_passed,
 768                                     int comp_args_on_stack,
 769                                     const BasicType *sig_bt,
 770                                     const VMRegPair *regs) {
 771 
 772   // Note: r13 contains the senderSP on entry. We must preserve it since
 773   // we may do a i2c -> c2i transition if we lose a race where compiled
 774   // code goes non-entrant while we get args ready.
 775   // In addition we use r13 to locate all the interpreter args as
 776   // we must align the stack to 16 bytes on an i2c entry else we
 777   // lose alignment we expect in all compiled code and register
 778   // save code can segv when fxsave instructions find improperly
 779   // aligned stack pointer.
 780 
 781   // Adapters can be frameless because they do not require the caller
 782   // to perform additional cleanup work, such as correcting the stack pointer.
 783   // An i2c adapter is frameless because the *caller* frame, which is interpreted,
 784   // routinely repairs its own stack pointer (from interpreter_frame_last_sp),
 785   // even if a callee has modified the stack pointer.
 786   // A c2i adapter is frameless because the *callee* frame, which is interpreted,
 787   // routinely repairs its caller's stack pointer (from sender_sp, which is set
 788   // up via the senderSP register).
 789   // In other words, if *either* the caller or callee is interpreted, we can

 840   // Convert 4-byte c2 stack slots to words.
 841   int comp_words_on_stack = align_up(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord;
 842 
 843   if (comp_args_on_stack) {
 844     __ subptr(rsp, comp_words_on_stack * wordSize);
 845   }
 846 
 847   // Ensure compiled code always sees stack at proper alignment
 848   __ andptr(rsp, -16);
 849 
 850   // push the return address and misalign the stack that youngest frame always sees
 851   // as far as the placement of the call instruction
 852   __ push(rax);
 853 
 854   // Put saved SP in another register
 855   const Register saved_sp = rax;
 856   __ movptr(saved_sp, r11);
 857 
 858   // Will jump to the compiled code just as if compiled code was doing it.
 859   // Pre-load the register-jump target early, to schedule it better.
 860   __ movptr(r11, Address(rbx, in_bytes(Method::from_compiled_offset())));
 861 
 862 #if INCLUDE_JVMCI
 863   if (EnableJVMCI) {
 864     // check if this call should be routed towards a specific entry point
 865     __ cmpptr(Address(r15_thread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())), 0);
 866     Label no_alternative_target;
 867     __ jcc(Assembler::equal, no_alternative_target);
 868     __ movptr(r11, Address(r15_thread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
 869     __ movptr(Address(r15_thread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())), 0);
 870     __ bind(no_alternative_target);
 871   }
 872 #endif // INCLUDE_JVMCI
 873 


 874   // Now generate the shuffle code.  Pick up all register args and move the
 875   // rest through the floating point stack top.
 876   for (int i = 0; i < total_args_passed; i++) {
 877     if (sig_bt[i] == T_VOID) {


 878       // Longs and doubles are passed in native word order, but misaligned
 879       // in the 32-bit build.
 880       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");

 881       continue;
 882     }
 883 
 884     // Pick up 0, 1 or 2 words from SP+offset.
 885 
 886     assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
 887             "scrambled load targets?");
 888     // Load in argument order going down.
 889     int ld_off = (total_args_passed - i)*Interpreter::stackElementSize;
 890     // Point to interpreter value (vs. tag)
 891     int next_off = ld_off - Interpreter::stackElementSize;
 892     //
 893     //
 894     //
 895     VMReg r_1 = regs[i].first();
 896     VMReg r_2 = regs[i].second();
 897     if (!r_1->is_valid()) {
 898       assert(!r_2->is_valid(), "");
 899       continue;
 900     }

 902       // Convert stack slot to an SP offset (+ wordSize to account for return address )
 903       int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize;
 904 
 905       // We can use r13 as a temp here because compiled code doesn't need r13 as an input
 906       // and if we end up going thru a c2i because of a miss a reasonable value of r13
 907       // will be generated.
 908       if (!r_2->is_valid()) {
 909         // sign extend???
 910         __ movl(r13, Address(saved_sp, ld_off));
 911         __ movptr(Address(rsp, st_off), r13);
 912       } else {
 913         //
 914         // We are using two optoregs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
 915         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
 916         // So we must adjust where to pick up the data to match the interpreter.
 917         //
 918         // Interpreter local[n] == MSW, local[n+1] == LSW however locals
 919         // are accessed as negative so LSW is at LOW address
 920 
 921         // ld_off is MSW so get LSW
 922         const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
 923                            next_off : ld_off;
 924         __ movq(r13, Address(saved_sp, offset));
 925         // st_off is LSW (i.e. reg.first())
 926         __ movq(Address(rsp, st_off), r13);
 927       }
 928     } else if (r_1->is_Register()) {  // Register argument
 929       Register r = r_1->as_Register();
 930       assert(r != rax, "must be different");
 931       if (r_2->is_valid()) {
 932         //
 933         // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
 934         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
 935         // So we must adjust where to pick up the data to match the interpreter.
 936 
 937         const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
 938                            next_off : ld_off;
 939 
 940         // this can be a misaligned move
 941         __ movq(r, Address(saved_sp, offset));
 942       } else {
 943         // sign extend and use a full word?
 944         __ movl(r, Address(saved_sp, ld_off));
 945       }
 946     } else {
 947       if (!r_2->is_valid()) {
 948         __ movflt(r_1->as_XMMRegister(), Address(saved_sp, ld_off));
 949       } else {
 950         __ movdbl(r_1->as_XMMRegister(), Address(saved_sp, next_off));
 951       }
 952     }
 953   }
 954 
 955   __ push_cont_fastpath(); // Set JavaThread::_cont_fastpath to the sp of the oldest interpreted frame we know about
 956 
 957   // 6243940 We might end up in handle_wrong_method if
 958   // the callee is deoptimized as we race thru here. If that
 959   // happens we don't want to take a safepoint because the
 960   // caller frame will look interpreted and arguments are now
 961   // "compiled" so it is much better to make this transition
 962   // invisible to the stack walking code. Unfortunately if
 963   // we try and find the callee by normal means a safepoint
 964   // is possible. So we stash the desired callee in the thread
 965   // and the vm will find there should this case occur.
 966 
 967   __ movptr(Address(r15_thread, JavaThread::callee_target_offset()), rbx);
 968 
 969   // put Method* where a c2i would expect should we end up there
 970   // only needed because eof c2 resolve stubs return Method* as a result in
 971   // rax
 972   __ mov(rax, rbx);
 973   __ jmp(r11);
 974 }
 975 






















 976 // ---------------------------------------------------------------
 977 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
 978                                                             int total_args_passed,
 979                                                             int comp_args_on_stack,
 980                                                             const BasicType *sig_bt,
 981                                                             const VMRegPair *regs,
 982                                                             AdapterFingerPrint* fingerprint) {






 983   address i2c_entry = __ pc();
 984 
 985   gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
 986 
 987   // -------------------------------------------------------------------------
 988   // Generate a C2I adapter.  On entry we know rbx holds the Method* during calls
 989   // to the interpreter.  The args start out packed in the compiled layout.  They
 990   // need to be unpacked into the interpreter layout.  This will almost always
 991   // require some stack space.  We grow the current (compiled) stack, then repack
 992   // the args.  We  finally end in a jump to the generic interpreter entry point.
 993   // On exit from the interpreter, the interpreter will restore our SP (lest the
 994   // compiled code, which relies solely on SP and not RBP, get sick).
 995 
 996   address c2i_unverified_entry = __ pc();

 997   Label skip_fixup;
 998   Label ok;
 999 
1000   Register holder = rax;
1001   Register receiver = j_rarg0;
1002   Register temp = rbx;
1003 
1004   {
1005     __ load_klass(temp, receiver, rscratch1);
1006     __ cmpptr(temp, Address(holder, CompiledICHolder::holder_klass_offset()));
1007     __ movptr(rbx, Address(holder, CompiledICHolder::holder_metadata_offset()));
1008     __ jcc(Assembler::equal, ok);
1009     __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1010 
1011     __ bind(ok);
1012     // Method might have been compiled since the call site was patched to
1013     // interpreted if that is the case treat it as a miss so we can get
1014     // the call site corrected.
1015     __ cmpptr(Address(rbx, in_bytes(Method::code_offset())), NULL_WORD);
1016     __ jcc(Assembler::equal, skip_fixup);
1017     __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1018   }
1019 
1020   address c2i_entry = __ pc();
1021 
1022   // Class initialization barrier for static methods
1023   address c2i_no_clinit_check_entry = nullptr;
1024   if (VM_Version::supports_fast_class_init_checks()) {
1025     Label L_skip_barrier;
1026     Register method = rbx;
1027 
1028     { // Bypass the barrier for non-static methods
1029       Register flags = rscratch1;
1030       __ movl(flags, Address(method, Method::access_flags_offset()));
1031       __ testl(flags, JVM_ACC_STATIC);
1032       __ jcc(Assembler::zero, L_skip_barrier); // non-static
1033     }
1034 
1035     Register klass = rscratch1;
1036     __ load_method_holder(klass, method);
1037     __ clinit_barrier(klass, r15_thread, &L_skip_barrier /*L_fast_path*/);
1038 
1039     __ jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub())); // slow path
1040 
1041     __ bind(L_skip_barrier);
1042     c2i_no_clinit_check_entry = __ pc();













1043   }
1044 
1045   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
1046   bs->c2i_entry_barrier(masm);
1047 
1048   gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);





1049 
1050   return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry, c2i_no_clinit_check_entry);
1051 }
1052 
1053 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
1054                                          VMRegPair *regs,
1055                                          VMRegPair *regs2,
1056                                          int total_args_passed) {
1057   assert(regs2 == nullptr, "not needed on x86");
1058 // We return the amount of VMRegImpl stack slots we need to reserve for all
1059 // the arguments NOT counting out_preserve_stack_slots.
1060 
1061 // NOTE: These arrays will have to change when c1 is ported
1062 #ifdef _WIN64
1063     static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
1064       c_rarg0, c_rarg1, c_rarg2, c_rarg3
1065     };
1066     static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
1067       c_farg0, c_farg1, c_farg2, c_farg3
1068     };
1069 #else
1070     static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {

1088       case T_BYTE:
1089       case T_SHORT:
1090       case T_INT:
1091         if (int_args < Argument::n_int_register_parameters_c) {
1092           regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
1093 #ifdef _WIN64
1094           fp_args++;
1095           // Allocate slots for callee to stuff register args the stack.
1096           stk_args += 2;
1097 #endif
1098         } else {
1099           regs[i].set1(VMRegImpl::stack2reg(stk_args));
1100           stk_args += 2;
1101         }
1102         break;
1103       case T_LONG:
1104         assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
1105         // fall through
1106       case T_OBJECT:
1107       case T_ARRAY:

1108       case T_ADDRESS:
1109       case T_METADATA:
1110         if (int_args < Argument::n_int_register_parameters_c) {
1111           regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
1112 #ifdef _WIN64
1113           fp_args++;
1114           stk_args += 2;
1115 #endif
1116         } else {
1117           regs[i].set2(VMRegImpl::stack2reg(stk_args));
1118           stk_args += 2;
1119         }
1120         break;
1121       case T_FLOAT:
1122         if (fp_args < Argument::n_float_register_parameters_c) {
1123           regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
1124 #ifdef _WIN64
1125           int_args++;
1126           // Allocate slots for callee to stuff register args the stack.
1127           stk_args += 2;

2014 
2015   int temploc = -1;
2016   for (int ai = 0; ai < arg_order.length(); ai += 2) {
2017     int i = arg_order.at(ai);
2018     int c_arg = arg_order.at(ai + 1);
2019     __ block_comment(err_msg("move %d -> %d", i, c_arg));
2020 #ifdef ASSERT
2021     if (in_regs[i].first()->is_Register()) {
2022       assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!");
2023     } else if (in_regs[i].first()->is_XMMRegister()) {
2024       assert(!freg_destroyed[in_regs[i].first()->as_XMMRegister()->encoding()], "destroyed reg!");
2025     }
2026     if (out_regs[c_arg].first()->is_Register()) {
2027       reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
2028     } else if (out_regs[c_arg].first()->is_XMMRegister()) {
2029       freg_destroyed[out_regs[c_arg].first()->as_XMMRegister()->encoding()] = true;
2030     }
2031 #endif /* ASSERT */
2032     switch (in_sig_bt[i]) {
2033       case T_ARRAY:

2034       case T_OBJECT:
2035         __ object_move(map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
2036                     ((i == 0) && (!is_static)),
2037                     &receiver_offset);
2038         break;
2039       case T_VOID:
2040         break;
2041 
2042       case T_FLOAT:
2043         __ float_move(in_regs[i], out_regs[c_arg]);
2044           break;
2045 
2046       case T_DOUBLE:
2047         assert( i + 1 < total_in_args &&
2048                 in_sig_bt[i + 1] == T_VOID &&
2049                 out_sig_bt[c_arg+1] == T_VOID, "bad arg list");
2050         __ double_move(in_regs[i], out_regs[c_arg]);
2051         break;
2052 
2053       case T_LONG :

2140     const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes();
2141 
2142     // Get the handle (the 2nd argument)
2143     __ mov(oop_handle_reg, c_rarg1);
2144 
2145     // Get address of the box
2146 
2147     __ lea(lock_reg, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size));
2148 
2149     // Load the oop from the handle
2150     __ movptr(obj_reg, Address(oop_handle_reg, 0));
2151 
2152     if (LockingMode == LM_MONITOR) {
2153       __ jmp(slow_path_lock);
2154     } else if (LockingMode == LM_LEGACY) {
2155       // Load immediate 1 into swap_reg %rax
2156       __ movl(swap_reg, 1);
2157 
2158       // Load (object->mark() | 1) into swap_reg %rax
2159       __ orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));




2160 
2161       // Save (object->mark() | 1) into BasicLock's displaced header
2162       __ movptr(Address(lock_reg, mark_word_offset), swap_reg);
2163 
2164       // src -> dest iff dest == rax else rax <- dest
2165       __ lock();
2166       __ cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
2167       __ jcc(Assembler::equal, count_mon);
2168 
2169       // Hmm should this move to the slow path code area???
2170 
2171       // Test if the oopMark is an obvious stack pointer, i.e.,
2172       //  1) (mark & 3) == 0, and
2173       //  2) rsp <= mark < mark + os::pagesize()
2174       // These 3 tests can be done by evaluating the following
2175       // expression: ((mark - rsp) & (3 - os::vm_page_size())),
2176       // assuming both stack pointer and pagesize have their
2177       // least significant 2 bits clear.
2178       // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
2179 

2204   // Now set thread in native
2205   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native);
2206 
2207   __ call(RuntimeAddress(native_func));
2208 
2209   // Verify or restore cpu control state after JNI call
2210   __ restore_cpu_control_state_after_jni(rscratch1);
2211 
2212   // Unpack native results.
2213   switch (ret_type) {
2214   case T_BOOLEAN: __ c2bool(rax);            break;
2215   case T_CHAR   : __ movzwl(rax, rax);      break;
2216   case T_BYTE   : __ sign_extend_byte (rax); break;
2217   case T_SHORT  : __ sign_extend_short(rax); break;
2218   case T_INT    : /* nothing to do */        break;
2219   case T_DOUBLE :
2220   case T_FLOAT  :
2221     // Result is in xmm0 we'll save as needed
2222     break;
2223   case T_ARRAY:                 // Really a handle

2224   case T_OBJECT:                // Really a handle
2225       break; // can't de-handlize until after safepoint check
2226   case T_VOID: break;
2227   case T_LONG: break;
2228   default       : ShouldNotReachHere();
2229   }
2230 
2231   Label after_transition;
2232 
2233   // Switch thread to "native transition" state before reading the synchronization state.
2234   // This additional state is necessary because reading and testing the synchronization
2235   // state is not atomic w.r.t. GC, as this scenario demonstrates:
2236   //     Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
2237   //     VM thread changes sync state to synchronizing and suspends threads for GC.
2238   //     Thread A is resumed to finish this native method, but doesn't block here since it
2239   //     didn't see any synchronization is progress, and escapes.
2240   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
2241 
2242   // Force this write out before the read below
2243   if (!UseSystemMemoryBarrier) {

3701   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), NULL_WORD);
3702 #endif
3703   // Clear the exception oop so GC no longer processes it as a root.
3704   __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), NULL_WORD);
3705 
3706   // rax: exception oop
3707   // r8:  exception handler
3708   // rdx: exception pc
3709   // Jump to handler
3710 
3711   __ jmp(r8);
3712 
3713   // Make sure all code is generated
3714   masm->flush();
3715 
3716   // Set exception blob
3717   _exception_blob =  ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
3718 }
3719 #endif // COMPILER2
3720 















































































































  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #ifndef _WINDOWS
  27 #include "alloca.h"
  28 #endif
  29 #include "asm/macroAssembler.hpp"
  30 #include "asm/macroAssembler.inline.hpp"
  31 #include "classfile/symbolTable.hpp"
  32 #include "code/compiledIC.hpp"
  33 #include "code/debugInfoRec.hpp"
  34 #include "code/icBuffer.hpp"
  35 #include "code/nativeInst.hpp"
  36 #include "code/vtableStubs.hpp"
  37 #include "compiler/oopMap.hpp"
  38 #include "gc/shared/collectedHeap.hpp"
  39 #include "gc/shared/gcLocker.hpp"
  40 #include "gc/shared/barrierSet.hpp"
  41 #include "gc/shared/barrierSetAssembler.hpp"
  42 #include "interpreter/interpreter.hpp"
  43 #include "logging/log.hpp"
  44 #include "memory/resourceArea.hpp"
  45 #include "memory/universe.hpp"
  46 #include "oops/compiledICHolder.hpp"
  47 #include "oops/klass.inline.hpp"
  48 #include "oops/method.inline.hpp"
  49 #include "prims/methodHandles.hpp"
  50 #include "runtime/continuation.hpp"
  51 #include "runtime/continuationEntry.inline.hpp"

 508     case T_SHORT:
 509     case T_INT:
 510       if (int_args < Argument::n_int_register_parameters_j) {
 511         regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
 512       } else {
 513         regs[i].set1(VMRegImpl::stack2reg(stk_args));
 514         stk_args += 2;
 515       }
 516       break;
 517     case T_VOID:
 518       // halves of T_LONG or T_DOUBLE
 519       assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
 520       regs[i].set_bad();
 521       break;
 522     case T_LONG:
 523       assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
 524       // fall through
 525     case T_OBJECT:
 526     case T_ARRAY:
 527     case T_ADDRESS:
 528     case T_PRIMITIVE_OBJECT:
 529       if (int_args < Argument::n_int_register_parameters_j) {
 530         regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
 531       } else {
 532         regs[i].set2(VMRegImpl::stack2reg(stk_args));
 533         stk_args += 2;
 534       }
 535       break;
 536     case T_FLOAT:
 537       if (fp_args < Argument::n_float_register_parameters_j) {
 538         regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
 539       } else {
 540         regs[i].set1(VMRegImpl::stack2reg(stk_args));
 541         stk_args += 2;
 542       }
 543       break;
 544     case T_DOUBLE:
 545       assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
 546       if (fp_args < Argument::n_float_register_parameters_j) {
 547         regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
 548       } else {
 549         regs[i].set2(VMRegImpl::stack2reg(stk_args));
 550         stk_args += 2;
 551       }
 552       break;
 553     default:
 554       ShouldNotReachHere();
 555       break;
 556     }
 557   }
 558 
 559   return align_up(stk_args, 2);
 560 }
 561 
 562 // Same as java_calling_convention() but for multiple return
 563 // values. There's no way to store them on the stack so if we don't
 564 // have enough registers, multiple values can't be returned.
 565 const uint SharedRuntime::java_return_convention_max_int = Argument::n_int_register_parameters_j+1;
 566 const uint SharedRuntime::java_return_convention_max_float = Argument::n_float_register_parameters_j;
 567 int SharedRuntime::java_return_convention(const BasicType *sig_bt,
 568                                           VMRegPair *regs,
 569                                           int total_args_passed) {
 570   // Create the mapping between argument positions and
 571   // registers.
 572   static const Register INT_ArgReg[java_return_convention_max_int] = {
 573     rax, j_rarg5, j_rarg4, j_rarg3, j_rarg2, j_rarg1, j_rarg0
 574   };
 575   static const XMMRegister FP_ArgReg[java_return_convention_max_float] = {
 576     j_farg0, j_farg1, j_farg2, j_farg3,
 577     j_farg4, j_farg5, j_farg6, j_farg7
 578   };
 579 
 580 
 581   uint int_args = 0;
 582   uint fp_args = 0;
 583 
 584   for (int i = 0; i < total_args_passed; i++) {
 585     switch (sig_bt[i]) {
 586     case T_BOOLEAN:
 587     case T_CHAR:
 588     case T_BYTE:
 589     case T_SHORT:
 590     case T_INT:
 591       if (int_args < Argument::n_int_register_parameters_j+1) {
 592         regs[i].set1(INT_ArgReg[int_args]->as_VMReg());
 593         int_args++;
 594       } else {
 595         return -1;
 596       }
 597       break;
 598     case T_VOID:
 599       // halves of T_LONG or T_DOUBLE
 600       assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
 601       regs[i].set_bad();
 602       break;
 603     case T_LONG:
 604       assert(sig_bt[i + 1] == T_VOID, "expecting half");
 605       // fall through
 606     case T_OBJECT:
 607     case T_PRIMITIVE_OBJECT:
 608     case T_ARRAY:
 609     case T_ADDRESS:
 610     case T_METADATA:
 611       if (int_args < Argument::n_int_register_parameters_j+1) {
 612         regs[i].set2(INT_ArgReg[int_args]->as_VMReg());
 613         int_args++;
 614       } else {
 615         return -1;
 616       }
 617       break;
 618     case T_FLOAT:
 619       if (fp_args < Argument::n_float_register_parameters_j) {
 620         regs[i].set1(FP_ArgReg[fp_args]->as_VMReg());
 621         fp_args++;
 622       } else {
 623         return -1;
 624       }
 625       break;
 626     case T_DOUBLE:
 627       assert(sig_bt[i + 1] == T_VOID, "expecting half");
 628       if (fp_args < Argument::n_float_register_parameters_j) {
 629         regs[i].set2(FP_ArgReg[fp_args]->as_VMReg());
 630         fp_args++;
 631       } else {
 632         return -1;
 633       }
 634       break;
 635     default:
 636       ShouldNotReachHere();
 637       break;
 638     }
 639   }
 640 
 641   return int_args + fp_args;
 642 }
 643 
 644 // Patch the callers callsite with entry to compiled code if it exists.
 645 static void patch_callers_callsite(MacroAssembler *masm) {
 646   Label L;
 647   __ cmpptr(Address(rbx, in_bytes(Method::code_offset())), NULL_WORD);
 648   __ jcc(Assembler::equal, L);
 649 
 650   // Save the current stack pointer
 651   __ mov(r13, rsp);
 652   // Schedule the branch target address early.
 653   // Call into the VM to patch the caller, then jump to compiled callee
 654   // rax isn't live so capture return address while we easily can
 655   __ movptr(rax, Address(rsp, 0));
 656 
 657   // align stack so push_CPU_state doesn't fault
 658   __ andptr(rsp, -(StackAlignmentInBytes));
 659   __ push_CPU_state();
 660   __ vzeroupper();
 661   // VM needs caller's callsite
 662   // VM needs target method
 663   // This needs to be a long call since we will relocate this adapter to

 666   // Allocate argument register save area
 667   if (frame::arg_reg_save_area_bytes != 0) {
 668     __ subptr(rsp, frame::arg_reg_save_area_bytes);
 669   }
 670   __ mov(c_rarg0, rbx);
 671   __ mov(c_rarg1, rax);
 672   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)));
 673 
 674   // De-allocate argument register save area
 675   if (frame::arg_reg_save_area_bytes != 0) {
 676     __ addptr(rsp, frame::arg_reg_save_area_bytes);
 677   }
 678 
 679   __ vzeroupper();
 680   __ pop_CPU_state();
 681   // restore sp
 682   __ mov(rsp, r13);
 683   __ bind(L);
 684 }
 685 
 686 // For each inline type argument, sig includes the list of fields of
 687 // the inline type. This utility function computes the number of
 688 // arguments for the call if inline types are passed by reference (the
 689 // calling convention the interpreter expects).
 690 static int compute_total_args_passed_int(const GrowableArray<SigEntry>* sig_extended) {
 691   int total_args_passed = 0;
 692   if (InlineTypePassFieldsAsArgs) {
 693     for (int i = 0; i < sig_extended->length(); i++) {
 694       BasicType bt = sig_extended->at(i)._bt;
 695       if (bt == T_PRIMITIVE_OBJECT) {
 696         // In sig_extended, an inline type argument starts with:
 697         // T_PRIMITIVE_OBJECT, followed by the types of the fields of the
 698         // inline type and T_VOID to mark the end of the value
 699         // type. Inline types are flattened so, for instance, in the
 700         // case of an inline type with an int field and an inline type
 701         // field that itself has 2 fields, an int and a long:
 702         // T_PRIMITIVE_OBJECT T_INT T_PRIMITIVE_OBJECT T_INT T_LONG T_VOID (second
 703         // slot for the T_LONG) T_VOID (inner T_PRIMITIVE_OBJECT) T_VOID
 704         // (outer T_PRIMITIVE_OBJECT)
 705         total_args_passed++;
 706         int vt = 1;
 707         do {
 708           i++;
 709           BasicType bt = sig_extended->at(i)._bt;
 710           BasicType prev_bt = sig_extended->at(i-1)._bt;
 711           if (bt == T_PRIMITIVE_OBJECT) {
 712             vt++;
 713           } else if (bt == T_VOID &&
 714                      prev_bt != T_LONG &&
 715                      prev_bt != T_DOUBLE) {
 716             vt--;
 717           }
 718         } while (vt != 0);
 719       } else {
 720         total_args_passed++;
 721       }
 722     }
 723   } else {
 724     total_args_passed = sig_extended->length();
 725   }
 726   return total_args_passed;
 727 }
 728 
 729 
 730 static void gen_c2i_adapter_helper(MacroAssembler* masm,
 731                                    BasicType bt,
 732                                    BasicType prev_bt,
 733                                    size_t size_in_bytes,
 734                                    const VMRegPair& reg_pair,
 735                                    const Address& to,
 736                                    int extraspace,
 737                                    bool is_oop) {
 738   assert(bt != T_PRIMITIVE_OBJECT || !InlineTypePassFieldsAsArgs, "no inline type here");
 739   if (bt == T_VOID) {
 740     assert(prev_bt == T_LONG || prev_bt == T_DOUBLE, "missing half");
 741     return;
 742   }
 743 
 744   // Say 4 args:
 745   // i   st_off
 746   // 0   32 T_LONG
 747   // 1   24 T_VOID
 748   // 2   16 T_OBJECT
 749   // 3    8 T_BOOL
 750   // -    0 return address
 751   //
 752   // However to make thing extra confusing. Because we can fit a long/double in
 753   // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
 754   // leaves one slot empty and only stores to a single slot. In this case the
 755   // slot that is occupied is the T_VOID slot. See I said it was confusing.
 756 
 757   bool wide = (size_in_bytes == wordSize);
 758   VMReg r_1 = reg_pair.first();
 759   VMReg r_2 = reg_pair.second();
 760   assert(r_2->is_valid() == wide, "invalid size");
 761   if (!r_1->is_valid()) {
 762     assert(!r_2->is_valid(), "must be invalid");
 763     return;
 764   }
 765 
 766   if (!r_1->is_XMMRegister()) {
 767     Register val = rax;
 768     if (r_1->is_stack()) {
 769       int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace;
 770       __ load_sized_value(val, Address(rsp, ld_off), size_in_bytes, /* is_signed */ false);
 771     } else {
 772       val = r_1->as_Register();
 773     }
 774     assert_different_registers(to.base(), val, rscratch1);
 775     if (is_oop) {
 776       __ push(r13);
 777       __ push(rbx);
 778       __ store_heap_oop(to, val, rscratch1, r13, rbx, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
 779       __ pop(rbx);
 780       __ pop(r13);
 781     } else {
 782       __ store_sized_value(to, val, size_in_bytes);
 783     }
 784   } else {
 785     if (wide) {
 786       __ movdbl(to, r_1->as_XMMRegister());
 787     } else {
 788       __ movflt(to, r_1->as_XMMRegister());
 789     }
 790   }
 791 }
 792 
 793 static void gen_c2i_adapter(MacroAssembler *masm,
 794                             const GrowableArray<SigEntry>* sig_extended,


 795                             const VMRegPair *regs,
 796                             bool requires_clinit_barrier,
 797                             address& c2i_no_clinit_check_entry,
 798                             Label& skip_fixup,
 799                             address start,
 800                             OopMapSet* oop_maps,
 801                             int& frame_complete,
 802                             int& frame_size_in_words,
 803                             bool alloc_inline_receiver) {
 804   if (requires_clinit_barrier && VM_Version::supports_fast_class_init_checks()) {
 805     Label L_skip_barrier;
 806     Register method = rbx;
 807 
 808     { // Bypass the barrier for non-static methods
 809       Register flags = rscratch1;
 810       __ movl(flags, Address(method, Method::access_flags_offset()));
 811       __ testl(flags, JVM_ACC_STATIC);
 812       __ jcc(Assembler::zero, L_skip_barrier); // non-static
 813     }
 814 
 815     Register klass = rscratch1;
 816     __ load_method_holder(klass, method);
 817     __ clinit_barrier(klass, r15_thread, &L_skip_barrier /*L_fast_path*/);
 818 
 819     __ jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub())); // slow path
 820 
 821     __ bind(L_skip_barrier);
 822     c2i_no_clinit_check_entry = __ pc();
 823   }
 824 
 825   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 826   bs->c2i_entry_barrier(masm);
 827 
 828   // Before we get into the guts of the C2I adapter, see if we should be here
 829   // at all.  We've come from compiled code and are attempting to jump to the
 830   // interpreter, which means the caller made a static call to get here
 831   // (vcalls always get a compiled target if there is one).  Check for a
 832   // compiled target.  If there is one, we need to patch the caller's call.
 833   patch_callers_callsite(masm);
 834 
 835   __ bind(skip_fixup);
 836 
 837   if (InlineTypePassFieldsAsArgs) {
 838     // Is there an inline type argument?
 839     bool has_inline_argument = false;
 840     for (int i = 0; i < sig_extended->length() && !has_inline_argument; i++) {
 841       has_inline_argument = (sig_extended->at(i)._bt == T_PRIMITIVE_OBJECT);
 842     }
 843     if (has_inline_argument) {
 844       // There is at least an inline type argument: we're coming from
 845       // compiled code so we have no buffers to back the inline types.
 846       // Allocate the buffers here with a runtime call.
 847       OopMap* map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words, /*save_vectors*/ false);
 848 
 849       frame_complete = __ offset();
 850 
 851       __ set_last_Java_frame(noreg, noreg, nullptr, rscratch1);
 852 
 853       __ mov(c_rarg0, r15_thread);
 854       __ mov(c_rarg1, rbx);
 855       __ mov64(c_rarg2, (int64_t)alloc_inline_receiver);
 856       __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::allocate_inline_types)));
 857 
 858       oop_maps->add_gc_map((int)(__ pc() - start), map);
 859       __ reset_last_Java_frame(false);
 860 
 861       RegisterSaver::restore_live_registers(masm);
 862 
 863       Label no_exception;
 864       __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 865       __ jcc(Assembler::equal, no_exception);
 866 
 867       __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), NULL_WORD);
 868       __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
 869       __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 870 
 871       __ bind(no_exception);
 872 
 873       // We get an array of objects from the runtime call
 874       __ get_vm_result(rscratch2, r15_thread); // Use rscratch2 (r11) as temporary because rscratch1 (r10) is trashed by movptr()
 875       __ get_vm_result_2(rbx, r15_thread); // TODO: required to keep the callee Method live?
 876     }
 877   }
 878 
 879   // Since all args are passed on the stack, total_args_passed *
 880   // Interpreter::stackElementSize is the space we need.
 881   int total_args_passed = compute_total_args_passed_int(sig_extended);
 882   assert(total_args_passed >= 0, "total_args_passed is %d", total_args_passed);
 883 
 884   int extraspace = (total_args_passed * Interpreter::stackElementSize);
 885 
 886   // stack is aligned, keep it that way
 887   // This is not currently needed or enforced by the interpreter, but
 888   // we might as well conform to the ABI.
 889   extraspace = align_up(extraspace, 2*wordSize);
 890 
 891   // set senderSP value
 892   __ lea(r13, Address(rsp, wordSize));
 893 
 894 #ifdef ASSERT
 895   __ check_stack_alignment(r13, "sender stack not aligned");
 896 #endif
 897   if (extraspace > 0) {
 898     // Pop the return address
 899     __ pop(rax);
 900 
 901     __ subptr(rsp, extraspace);
 902 
 903     // Push the return address
 904     __ push(rax);
 905 
 906     // Account for the return address location since we store it first rather
 907     // than hold it in a register across all the shuffling
 908     extraspace += wordSize;
 909   }
 910 
 911 #ifdef ASSERT
 912   __ check_stack_alignment(rsp, "callee stack not aligned", wordSize, rax);
 913 #endif
 914 
 915   // Now write the args into the outgoing interpreter space









 916 
 917   // next_arg_comp is the next argument from the compiler point of
 918   // view (inline type fields are passed in registers/on the stack). In
 919   // sig_extended, an inline type argument starts with: T_PRIMITIVE_OBJECT,
 920   // followed by the types of the fields of the inline type and T_VOID
 921   // to mark the end of the inline type. ignored counts the number of
 922   // T_PRIMITIVE_OBJECT/T_VOID. next_vt_arg is the next inline type argument:
 923   // used to get the buffer for that argument from the pool of buffers
 924   // we allocated above and want to pass to the
 925   // interpreter. next_arg_int is the next argument from the
 926   // interpreter point of view (inline types are passed by reference).
 927   for (int next_arg_comp = 0, ignored = 0, next_vt_arg = 0, next_arg_int = 0;
 928        next_arg_comp < sig_extended->length(); next_arg_comp++) {
 929     assert(ignored <= next_arg_comp, "shouldn't skip over more slots than there are arguments");
 930     assert(next_arg_int <= total_args_passed, "more arguments for the interpreter than expected?");
 931     BasicType bt = sig_extended->at(next_arg_comp)._bt;
 932     int st_off = (total_args_passed - next_arg_int) * Interpreter::stackElementSize;
 933     if (!InlineTypePassFieldsAsArgs || bt != T_PRIMITIVE_OBJECT) {
 934       int next_off = st_off - Interpreter::stackElementSize;
 935       const int offset = (bt == T_LONG || bt == T_DOUBLE) ? next_off : st_off;
 936       const VMRegPair reg_pair = regs[next_arg_comp-ignored];
 937       size_t size_in_bytes = reg_pair.second()->is_valid() ? 8 : 4;
 938       gen_c2i_adapter_helper(masm, bt, next_arg_comp > 0 ? sig_extended->at(next_arg_comp-1)._bt : T_ILLEGAL,
 939                              size_in_bytes, reg_pair, Address(rsp, offset), extraspace, false);
 940       next_arg_int++;













 941 #ifdef ASSERT
 942       if (bt == T_LONG || bt == T_DOUBLE) {
 943         // Overwrite the unused slot with known junk
 944         __ mov64(rax, CONST64(0xdeadffffdeadaaaa));
 945         __ movptr(Address(rsp, st_off), rax);



 946       }















 947 #endif /* ASSERT */
 948     } else {
 949       ignored++;
 950       // get the buffer from the just allocated pool of buffers
 951       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + next_vt_arg * type2aelembytes(T_PRIMITIVE_OBJECT);
 952       __ load_heap_oop(r14, Address(rscratch2, index));
 953       next_vt_arg++; next_arg_int++;
 954       int vt = 1;
 955       // write fields we get from compiled code in registers/stack
 956       // slots to the buffer: we know we are done with that inline type
 957       // argument when we hit the T_VOID that acts as an end of inline
 958       // type delimiter for this inline type. Inline types are flattened
 959       // so we might encounter embedded inline types. Each entry in
 960       // sig_extended contains a field offset in the buffer.
 961       Label L_null;
 962       do {
 963         next_arg_comp++;
 964         BasicType bt = sig_extended->at(next_arg_comp)._bt;
 965         BasicType prev_bt = sig_extended->at(next_arg_comp-1)._bt;
 966         if (bt == T_PRIMITIVE_OBJECT) {
 967           vt++;
 968           ignored++;
 969         } else if (bt == T_VOID &&
 970                    prev_bt != T_LONG &&
 971                    prev_bt != T_DOUBLE) {
 972           vt--;
 973           ignored++;
 974         } else {
 975           int off = sig_extended->at(next_arg_comp)._offset;
 976           if (off == -1) {
 977             // Nullable inline type argument, emit null check
 978             VMReg reg = regs[next_arg_comp-ignored].first();
 979             Label L_notNull;
 980             if (reg->is_stack()) {
 981               int ld_off = reg->reg2stack() * VMRegImpl::stack_slot_size + extraspace;
 982               __ testb(Address(rsp, ld_off), 1);
 983             } else {
 984               __ testb(reg->as_Register(), 1);
 985             }
 986             __ jcc(Assembler::notZero, L_notNull);
 987             __ movptr(Address(rsp, st_off), 0);
 988             __ jmp(L_null);
 989             __ bind(L_notNull);
 990             continue;
 991           }
 992           assert(off > 0, "offset in object should be positive");
 993           size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
 994           bool is_oop = is_reference_type(bt);
 995           gen_c2i_adapter_helper(masm, bt, next_arg_comp > 0 ? sig_extended->at(next_arg_comp-1)._bt : T_ILLEGAL,
 996                                  size_in_bytes, regs[next_arg_comp-ignored], Address(r14, off), extraspace, is_oop);
 997         }
 998       } while (vt != 0);
 999       // pass the buffer to the interpreter
1000       __ movptr(Address(rsp, st_off), r14);
1001       __ bind(L_null);










1002     }
1003   }
1004 
1005   // Schedule the branch target address early.
1006   __ movptr(rcx, Address(rbx, in_bytes(Method::interpreter_entry_offset())));
1007   __ jmp(rcx);
1008 }
1009 
1010 static void range_check(MacroAssembler* masm, Register pc_reg, Register temp_reg,
1011                         address code_start, address code_end,
1012                         Label& L_ok) {
1013   Label L_fail;
1014   __ lea(temp_reg, ExternalAddress(code_start));
1015   __ cmpptr(pc_reg, temp_reg);
1016   __ jcc(Assembler::belowEqual, L_fail);
1017   __ lea(temp_reg, ExternalAddress(code_end));
1018   __ cmpptr(pc_reg, temp_reg);
1019   __ jcc(Assembler::below, L_ok);
1020   __ bind(L_fail);
1021 }
1022 
1023 void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,

1024                                     int comp_args_on_stack,
1025                                     const GrowableArray<SigEntry>* sig,
1026                                     const VMRegPair *regs) {
1027 
1028   // Note: r13 contains the senderSP on entry. We must preserve it since
1029   // we may do a i2c -> c2i transition if we lose a race where compiled
1030   // code goes non-entrant while we get args ready.
1031   // In addition we use r13 to locate all the interpreter args as
1032   // we must align the stack to 16 bytes on an i2c entry else we
1033   // lose alignment we expect in all compiled code and register
1034   // save code can segv when fxsave instructions find improperly
1035   // aligned stack pointer.
1036 
1037   // Adapters can be frameless because they do not require the caller
1038   // to perform additional cleanup work, such as correcting the stack pointer.
1039   // An i2c adapter is frameless because the *caller* frame, which is interpreted,
1040   // routinely repairs its own stack pointer (from interpreter_frame_last_sp),
1041   // even if a callee has modified the stack pointer.
1042   // A c2i adapter is frameless because the *callee* frame, which is interpreted,
1043   // routinely repairs its caller's stack pointer (from sender_sp, which is set
1044   // up via the senderSP register).
1045   // In other words, if *either* the caller or callee is interpreted, we can

1096   // Convert 4-byte c2 stack slots to words.
1097   int comp_words_on_stack = align_up(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord;
1098 
1099   if (comp_args_on_stack) {
1100     __ subptr(rsp, comp_words_on_stack * wordSize);
1101   }
1102 
1103   // Ensure compiled code always sees stack at proper alignment
1104   __ andptr(rsp, -16);
1105 
1106   // push the return address and misalign the stack that youngest frame always sees
1107   // as far as the placement of the call instruction
1108   __ push(rax);
1109 
1110   // Put saved SP in another register
1111   const Register saved_sp = rax;
1112   __ movptr(saved_sp, r11);
1113 
1114   // Will jump to the compiled code just as if compiled code was doing it.
1115   // Pre-load the register-jump target early, to schedule it better.
1116   __ movptr(r11, Address(rbx, in_bytes(Method::from_compiled_inline_offset())));
1117 
1118 #if INCLUDE_JVMCI
1119   if (EnableJVMCI) {
1120     // check if this call should be routed towards a specific entry point
1121     __ cmpptr(Address(r15_thread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())), 0);
1122     Label no_alternative_target;
1123     __ jcc(Assembler::equal, no_alternative_target);
1124     __ movptr(r11, Address(r15_thread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
1125     __ movptr(Address(r15_thread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())), 0);
1126     __ bind(no_alternative_target);
1127   }
1128 #endif // INCLUDE_JVMCI
1129 
1130   int total_args_passed = sig->length();
1131 
1132   // Now generate the shuffle code.  Pick up all register args and move the
1133   // rest through the floating point stack top.
1134   for (int i = 0; i < total_args_passed; i++) {
1135     BasicType bt = sig->at(i)._bt;
1136     assert(bt != T_PRIMITIVE_OBJECT, "i2c adapter doesn't unpack inline type args");
1137     if (bt == T_VOID) {
1138       // Longs and doubles are passed in native word order, but misaligned
1139       // in the 32-bit build.
1140       BasicType prev_bt = (i > 0) ? sig->at(i-1)._bt : T_ILLEGAL;
1141       assert(i > 0 && (prev_bt == T_LONG || prev_bt == T_DOUBLE), "missing half");
1142       continue;
1143     }
1144 
1145     // Pick up 0, 1 or 2 words from SP+offset.
1146 
1147     assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
1148             "scrambled load targets?");
1149     // Load in argument order going down.
1150     int ld_off = (total_args_passed - i)*Interpreter::stackElementSize;
1151     // Point to interpreter value (vs. tag)
1152     int next_off = ld_off - Interpreter::stackElementSize;
1153     //
1154     //
1155     //
1156     VMReg r_1 = regs[i].first();
1157     VMReg r_2 = regs[i].second();
1158     if (!r_1->is_valid()) {
1159       assert(!r_2->is_valid(), "");
1160       continue;
1161     }

1163       // Convert stack slot to an SP offset (+ wordSize to account for return address )
1164       int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize;
1165 
1166       // We can use r13 as a temp here because compiled code doesn't need r13 as an input
1167       // and if we end up going thru a c2i because of a miss a reasonable value of r13
1168       // will be generated.
1169       if (!r_2->is_valid()) {
1170         // sign extend???
1171         __ movl(r13, Address(saved_sp, ld_off));
1172         __ movptr(Address(rsp, st_off), r13);
1173       } else {
1174         //
1175         // We are using two optoregs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
1176         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
1177         // So we must adjust where to pick up the data to match the interpreter.
1178         //
1179         // Interpreter local[n] == MSW, local[n+1] == LSW however locals
1180         // are accessed as negative so LSW is at LOW address
1181 
1182         // ld_off is MSW so get LSW
1183         const int offset = (bt==T_LONG||bt==T_DOUBLE)?
1184                            next_off : ld_off;
1185         __ movq(r13, Address(saved_sp, offset));
1186         // st_off is LSW (i.e. reg.first())
1187         __ movq(Address(rsp, st_off), r13);
1188       }
1189     } else if (r_1->is_Register()) {  // Register argument
1190       Register r = r_1->as_Register();
1191       assert(r != rax, "must be different");
1192       if (r_2->is_valid()) {
1193         //
1194         // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
1195         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
1196         // So we must adjust where to pick up the data to match the interpreter.
1197 
1198         const int offset = (bt==T_LONG||bt==T_DOUBLE)?
1199                            next_off : ld_off;
1200 
1201         // this can be a misaligned move
1202         __ movq(r, Address(saved_sp, offset));
1203       } else {
1204         // sign extend and use a full word?
1205         __ movl(r, Address(saved_sp, ld_off));
1206       }
1207     } else {
1208       if (!r_2->is_valid()) {
1209         __ movflt(r_1->as_XMMRegister(), Address(saved_sp, ld_off));
1210       } else {
1211         __ movdbl(r_1->as_XMMRegister(), Address(saved_sp, next_off));
1212       }
1213     }
1214   }
1215 
1216   __ push_cont_fastpath(); // Set JavaThread::_cont_fastpath to the sp of the oldest interpreted frame we know about
1217 
1218   // 6243940 We might end up in handle_wrong_method if
1219   // the callee is deoptimized as we race thru here. If that
1220   // happens we don't want to take a safepoint because the
1221   // caller frame will look interpreted and arguments are now
1222   // "compiled" so it is much better to make this transition
1223   // invisible to the stack walking code. Unfortunately if
1224   // we try and find the callee by normal means a safepoint
1225   // is possible. So we stash the desired callee in the thread
1226   // and the vm will find there should this case occur.
1227 
1228   __ movptr(Address(r15_thread, JavaThread::callee_target_offset()), rbx);
1229 
1230   // put Method* where a c2i would expect should we end up there
1231   // only needed because of c2 resolve stubs return Method* as a result in
1232   // rax
1233   __ mov(rax, rbx);
1234   __ jmp(r11);
1235 }
1236 
1237 static void gen_inline_cache_check(MacroAssembler *masm, Label& skip_fixup) {
1238   Label ok;
1239 
1240   Register holder = rax;
1241   Register receiver = j_rarg0;
1242   Register temp = rbx;
1243 
1244   __ load_klass(temp, receiver, rscratch1);
1245   __ cmpptr(temp, Address(holder, CompiledICHolder::holder_klass_offset()));
1246   __ movptr(rbx, Address(holder, CompiledICHolder::holder_metadata_offset()));
1247   __ jcc(Assembler::equal, ok);
1248   __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1249 
1250   __ bind(ok);
1251   // Method might have been compiled since the call site was patched to
1252   // interpreted if that is the case treat it as a miss so we can get
1253   // the call site corrected.
1254   __ cmpptr(Address(rbx, in_bytes(Method::code_offset())), NULL_WORD);
1255   __ jcc(Assembler::equal, skip_fixup);
1256   __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1257 }
1258 
1259 // ---------------------------------------------------------------
1260 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler* masm,

1261                                                             int comp_args_on_stack,
1262                                                             const GrowableArray<SigEntry>* sig,
1263                                                             const VMRegPair* regs,
1264                                                             const GrowableArray<SigEntry>* sig_cc,
1265                                                             const VMRegPair* regs_cc,
1266                                                             const GrowableArray<SigEntry>* sig_cc_ro,
1267                                                             const VMRegPair* regs_cc_ro,
1268                                                             AdapterFingerPrint* fingerprint,
1269                                                             AdapterBlob*& new_adapter,
1270                                                             bool allocate_code_blob) {
1271   address i2c_entry = __ pc();
1272   gen_i2c_adapter(masm, comp_args_on_stack, sig, regs);

1273 
1274   // -------------------------------------------------------------------------
1275   // Generate a C2I adapter.  On entry we know rbx holds the Method* during calls
1276   // to the interpreter.  The args start out packed in the compiled layout.  They
1277   // need to be unpacked into the interpreter layout.  This will almost always
1278   // require some stack space.  We grow the current (compiled) stack, then repack
1279   // the args.  We  finally end in a jump to the generic interpreter entry point.
1280   // On exit from the interpreter, the interpreter will restore our SP (lest the
1281   // compiled code, which relies solely on SP and not RBP, get sick).
1282 
1283   address c2i_unverified_entry        = __ pc();
1284   address c2i_unverified_inline_entry = __ pc();
1285   Label skip_fixup;





1286 
1287   gen_inline_cache_check(masm, skip_fixup);





1288 
1289   OopMapSet* oop_maps = new OopMapSet();
1290   int frame_complete = CodeOffsets::frame_never_safe;
1291   int frame_size_in_words = 0;







1292 
1293   // Scalarized c2i adapter with non-scalarized receiver (i.e., don't pack receiver)
1294   address c2i_no_clinit_check_entry = nullptr;
1295   address c2i_inline_ro_entry = __ pc();
1296   if (regs_cc != regs_cc_ro) {
1297     // No class init barrier needed because method is guaranteed to be non-static
1298     gen_c2i_adapter(masm, sig_cc_ro, regs_cc_ro, /* requires_clinit_barrier = */ false, c2i_no_clinit_check_entry,
1299                     skip_fixup, i2c_entry, oop_maps, frame_complete, frame_size_in_words, /* alloc_inline_receiver = */ false);
1300     skip_fixup.reset();
1301   }









1302 
1303   // Scalarized c2i adapter
1304   address c2i_entry        = __ pc();
1305   address c2i_inline_entry = __ pc();
1306   gen_c2i_adapter(masm, sig_cc, regs_cc, /* requires_clinit_barrier = */ true, c2i_no_clinit_check_entry,
1307                   skip_fixup, i2c_entry, oop_maps, frame_complete, frame_size_in_words, /* alloc_inline_receiver = */ true);
1308 
1309   // Non-scalarized c2i adapter
1310   if (regs != regs_cc) {
1311     c2i_unverified_inline_entry = __ pc();
1312     Label inline_entry_skip_fixup;
1313     gen_inline_cache_check(masm, inline_entry_skip_fixup);
1314 
1315     c2i_inline_entry = __ pc();
1316     gen_c2i_adapter(masm, sig, regs, /* requires_clinit_barrier = */ true, c2i_no_clinit_check_entry,
1317                     inline_entry_skip_fixup, i2c_entry, oop_maps, frame_complete, frame_size_in_words, /* alloc_inline_receiver = */ false);
1318   }
1319 


1320 
1321   // The c2i adapters might safepoint and trigger a GC. The caller must make sure that
1322   // the GC knows about the location of oop argument locations passed to the c2i adapter.
1323   if (allocate_code_blob) {
1324     bool caller_must_gc_arguments = (regs != regs_cc);
1325     new_adapter = AdapterBlob::create(masm->code(), frame_complete, frame_size_in_words, oop_maps, caller_must_gc_arguments);
1326   }
1327 
1328   return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_inline_entry, c2i_inline_ro_entry, c2i_unverified_entry, c2i_unverified_inline_entry, c2i_no_clinit_check_entry);
1329 }
1330 
1331 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
1332                                          VMRegPair *regs,
1333                                          VMRegPair *regs2,
1334                                          int total_args_passed) {
1335   assert(regs2 == nullptr, "not needed on x86");
1336 // We return the amount of VMRegImpl stack slots we need to reserve for all
1337 // the arguments NOT counting out_preserve_stack_slots.
1338 
1339 // NOTE: These arrays will have to change when c1 is ported
1340 #ifdef _WIN64
1341     static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
1342       c_rarg0, c_rarg1, c_rarg2, c_rarg3
1343     };
1344     static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
1345       c_farg0, c_farg1, c_farg2, c_farg3
1346     };
1347 #else
1348     static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {

1366       case T_BYTE:
1367       case T_SHORT:
1368       case T_INT:
1369         if (int_args < Argument::n_int_register_parameters_c) {
1370           regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
1371 #ifdef _WIN64
1372           fp_args++;
1373           // Allocate slots for callee to stuff register args the stack.
1374           stk_args += 2;
1375 #endif
1376         } else {
1377           regs[i].set1(VMRegImpl::stack2reg(stk_args));
1378           stk_args += 2;
1379         }
1380         break;
1381       case T_LONG:
1382         assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
1383         // fall through
1384       case T_OBJECT:
1385       case T_ARRAY:
1386       case T_PRIMITIVE_OBJECT:
1387       case T_ADDRESS:
1388       case T_METADATA:
1389         if (int_args < Argument::n_int_register_parameters_c) {
1390           regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
1391 #ifdef _WIN64
1392           fp_args++;
1393           stk_args += 2;
1394 #endif
1395         } else {
1396           regs[i].set2(VMRegImpl::stack2reg(stk_args));
1397           stk_args += 2;
1398         }
1399         break;
1400       case T_FLOAT:
1401         if (fp_args < Argument::n_float_register_parameters_c) {
1402           regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
1403 #ifdef _WIN64
1404           int_args++;
1405           // Allocate slots for callee to stuff register args the stack.
1406           stk_args += 2;

2293 
2294   int temploc = -1;
2295   for (int ai = 0; ai < arg_order.length(); ai += 2) {
2296     int i = arg_order.at(ai);
2297     int c_arg = arg_order.at(ai + 1);
2298     __ block_comment(err_msg("move %d -> %d", i, c_arg));
2299 #ifdef ASSERT
2300     if (in_regs[i].first()->is_Register()) {
2301       assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!");
2302     } else if (in_regs[i].first()->is_XMMRegister()) {
2303       assert(!freg_destroyed[in_regs[i].first()->as_XMMRegister()->encoding()], "destroyed reg!");
2304     }
2305     if (out_regs[c_arg].first()->is_Register()) {
2306       reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
2307     } else if (out_regs[c_arg].first()->is_XMMRegister()) {
2308       freg_destroyed[out_regs[c_arg].first()->as_XMMRegister()->encoding()] = true;
2309     }
2310 #endif /* ASSERT */
2311     switch (in_sig_bt[i]) {
2312       case T_ARRAY:
2313       case T_PRIMITIVE_OBJECT:
2314       case T_OBJECT:
2315         __ object_move(map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
2316                     ((i == 0) && (!is_static)),
2317                     &receiver_offset);
2318         break;
2319       case T_VOID:
2320         break;
2321 
2322       case T_FLOAT:
2323         __ float_move(in_regs[i], out_regs[c_arg]);
2324           break;
2325 
2326       case T_DOUBLE:
2327         assert( i + 1 < total_in_args &&
2328                 in_sig_bt[i + 1] == T_VOID &&
2329                 out_sig_bt[c_arg+1] == T_VOID, "bad arg list");
2330         __ double_move(in_regs[i], out_regs[c_arg]);
2331         break;
2332 
2333       case T_LONG :

2420     const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes();
2421 
2422     // Get the handle (the 2nd argument)
2423     __ mov(oop_handle_reg, c_rarg1);
2424 
2425     // Get address of the box
2426 
2427     __ lea(lock_reg, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size));
2428 
2429     // Load the oop from the handle
2430     __ movptr(obj_reg, Address(oop_handle_reg, 0));
2431 
2432     if (LockingMode == LM_MONITOR) {
2433       __ jmp(slow_path_lock);
2434     } else if (LockingMode == LM_LEGACY) {
2435       // Load immediate 1 into swap_reg %rax
2436       __ movl(swap_reg, 1);
2437 
2438       // Load (object->mark() | 1) into swap_reg %rax
2439       __ orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
2440       if (EnableValhalla) {
2441         // Mask inline_type bit such that we go to the slow path if object is an inline type
2442         __ andptr(swap_reg, ~((int) markWord::inline_type_bit_in_place));
2443       }
2444 
2445       // Save (object->mark() | 1) into BasicLock's displaced header
2446       __ movptr(Address(lock_reg, mark_word_offset), swap_reg);
2447 
2448       // src -> dest iff dest == rax else rax <- dest
2449       __ lock();
2450       __ cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
2451       __ jcc(Assembler::equal, count_mon);
2452 
2453       // Hmm should this move to the slow path code area???
2454 
2455       // Test if the oopMark is an obvious stack pointer, i.e.,
2456       //  1) (mark & 3) == 0, and
2457       //  2) rsp <= mark < mark + os::pagesize()
2458       // These 3 tests can be done by evaluating the following
2459       // expression: ((mark - rsp) & (3 - os::vm_page_size())),
2460       // assuming both stack pointer and pagesize have their
2461       // least significant 2 bits clear.
2462       // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
2463 

2488   // Now set thread in native
2489   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native);
2490 
2491   __ call(RuntimeAddress(native_func));
2492 
2493   // Verify or restore cpu control state after JNI call
2494   __ restore_cpu_control_state_after_jni(rscratch1);
2495 
2496   // Unpack native results.
2497   switch (ret_type) {
2498   case T_BOOLEAN: __ c2bool(rax);            break;
2499   case T_CHAR   : __ movzwl(rax, rax);      break;
2500   case T_BYTE   : __ sign_extend_byte (rax); break;
2501   case T_SHORT  : __ sign_extend_short(rax); break;
2502   case T_INT    : /* nothing to do */        break;
2503   case T_DOUBLE :
2504   case T_FLOAT  :
2505     // Result is in xmm0 we'll save as needed
2506     break;
2507   case T_ARRAY:                 // Really a handle
2508   case T_PRIMITIVE_OBJECT:           // Really a handle
2509   case T_OBJECT:                // Really a handle
2510       break; // can't de-handlize until after safepoint check
2511   case T_VOID: break;
2512   case T_LONG: break;
2513   default       : ShouldNotReachHere();
2514   }
2515 
2516   Label after_transition;
2517 
2518   // Switch thread to "native transition" state before reading the synchronization state.
2519   // This additional state is necessary because reading and testing the synchronization
2520   // state is not atomic w.r.t. GC, as this scenario demonstrates:
2521   //     Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
2522   //     VM thread changes sync state to synchronizing and suspends threads for GC.
2523   //     Thread A is resumed to finish this native method, but doesn't block here since it
2524   //     didn't see any synchronization is progress, and escapes.
2525   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
2526 
2527   // Force this write out before the read below
2528   if (!UseSystemMemoryBarrier) {

3986   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), NULL_WORD);
3987 #endif
3988   // Clear the exception oop so GC no longer processes it as a root.
3989   __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), NULL_WORD);
3990 
3991   // rax: exception oop
3992   // r8:  exception handler
3993   // rdx: exception pc
3994   // Jump to handler
3995 
3996   __ jmp(r8);
3997 
3998   // Make sure all code is generated
3999   masm->flush();
4000 
4001   // Set exception blob
4002   _exception_blob =  ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
4003 }
4004 #endif // COMPILER2
4005 
4006 BufferedInlineTypeBlob* SharedRuntime::generate_buffered_inline_type_adapter(const InlineKlass* vk) {
4007   BufferBlob* buf = BufferBlob::create("inline types pack/unpack", 16 * K);
4008   CodeBuffer buffer(buf);
4009   short buffer_locs[20];
4010   buffer.insts()->initialize_shared_locs((relocInfo*)buffer_locs,
4011                                          sizeof(buffer_locs)/sizeof(relocInfo));
4012 
4013   MacroAssembler* masm = new MacroAssembler(&buffer);
4014 
4015   const Array<SigEntry>* sig_vk = vk->extended_sig();
4016   const Array<VMRegPair>* regs = vk->return_regs();
4017 
4018   int pack_fields_jobject_off = __ offset();
4019   // Resolve pre-allocated buffer from JNI handle.
4020   // We cannot do this in generate_call_stub() because it requires GC code to be initialized.
4021   __ movptr(rax, Address(r13, 0));
4022   __ resolve_jobject(rax /* value */,
4023                      r15_thread /* thread */,
4024                      r12 /* tmp */);
4025   __ movptr(Address(r13, 0), rax);
4026 
4027   int pack_fields_off = __ offset();
4028 
4029   int j = 1;
4030   for (int i = 0; i < sig_vk->length(); i++) {
4031     BasicType bt = sig_vk->at(i)._bt;
4032     if (bt == T_PRIMITIVE_OBJECT) {
4033       continue;
4034     }
4035     if (bt == T_VOID) {
4036       if (sig_vk->at(i-1)._bt == T_LONG ||
4037           sig_vk->at(i-1)._bt == T_DOUBLE) {
4038         j++;
4039       }
4040       continue;
4041     }
4042     int off = sig_vk->at(i)._offset;
4043     assert(off > 0, "offset in object should be positive");
4044     VMRegPair pair = regs->at(j);
4045     VMReg r_1 = pair.first();
4046     VMReg r_2 = pair.second();
4047     Address to(rax, off);
4048     if (bt == T_FLOAT) {
4049       __ movflt(to, r_1->as_XMMRegister());
4050     } else if (bt == T_DOUBLE) {
4051       __ movdbl(to, r_1->as_XMMRegister());
4052     } else {
4053       Register val = r_1->as_Register();
4054       assert_different_registers(to.base(), val, r14, r13, rbx, rscratch1);
4055       if (is_reference_type(bt)) {
4056         __ store_heap_oop(to, val, r14, r13, rbx, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
4057       } else {
4058         __ store_sized_value(to, r_1->as_Register(), type2aelembytes(bt));
4059       }
4060     }
4061     j++;
4062   }
4063   assert(j == regs->length(), "missed a field?");
4064 
4065   __ ret(0);
4066 
4067   int unpack_fields_off = __ offset();
4068 
4069   Label skip;
4070   __ testptr(rax, rax);
4071   __ jcc(Assembler::zero, skip);
4072 
4073   j = 1;
4074   for (int i = 0; i < sig_vk->length(); i++) {
4075     BasicType bt = sig_vk->at(i)._bt;
4076     if (bt == T_PRIMITIVE_OBJECT) {
4077       continue;
4078     }
4079     if (bt == T_VOID) {
4080       if (sig_vk->at(i-1)._bt == T_LONG ||
4081           sig_vk->at(i-1)._bt == T_DOUBLE) {
4082         j++;
4083       }
4084       continue;
4085     }
4086     int off = sig_vk->at(i)._offset;
4087     assert(off > 0, "offset in object should be positive");
4088     VMRegPair pair = regs->at(j);
4089     VMReg r_1 = pair.first();
4090     VMReg r_2 = pair.second();
4091     Address from(rax, off);
4092     if (bt == T_FLOAT) {
4093       __ movflt(r_1->as_XMMRegister(), from);
4094     } else if (bt == T_DOUBLE) {
4095       __ movdbl(r_1->as_XMMRegister(), from);
4096     } else if (bt == T_OBJECT || bt == T_ARRAY) {
4097       assert_different_registers(rax, r_1->as_Register());
4098       __ load_heap_oop(r_1->as_Register(), from);
4099     } else {
4100       assert(is_java_primitive(bt), "unexpected basic type");
4101       assert_different_registers(rax, r_1->as_Register());
4102       size_t size_in_bytes = type2aelembytes(bt);
4103       __ load_sized_value(r_1->as_Register(), from, size_in_bytes, bt != T_CHAR && bt != T_BOOLEAN);
4104     }
4105     j++;
4106   }
4107   assert(j == regs->length(), "missed a field?");
4108 
4109   __ bind(skip);
4110   __ ret(0);
4111 
4112   __ flush();
4113 
4114   return BufferedInlineTypeBlob::create(&buffer, pack_fields_off, pack_fields_jobject_off, unpack_fields_off);
4115 }
< prev index next >