< prev index next >

src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp

Print this page

   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 "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"

  29 #include "code/compiledIC.hpp"
  30 #include "compiler/compileTask.hpp"
  31 #include "compiler/disassembler.hpp"
  32 #include "compiler/oopMap.hpp"
  33 #include "gc/shared/barrierSet.hpp"
  34 #include "gc/shared/barrierSetAssembler.hpp"
  35 #include "gc/shared/cardTableBarrierSet.hpp"
  36 #include "gc/shared/cardTable.hpp"
  37 #include "gc/shared/collectedHeap.hpp"
  38 #include "gc/shared/tlab_globals.hpp"
  39 #include "interpreter/bytecodeHistogram.hpp"
  40 #include "interpreter/interpreter.hpp"
  41 #include "interpreter/interpreterRuntime.hpp"
  42 #include "jvm.h"
  43 #include "memory/resourceArea.hpp"
  44 #include "memory/universe.hpp"
  45 #include "nativeInst_aarch64.hpp"
  46 #include "oops/accessDecorators.hpp"
  47 #include "oops/compressedKlass.inline.hpp"
  48 #include "oops/compressedOops.inline.hpp"
  49 #include "oops/klass.inline.hpp"


  50 #include "runtime/continuation.hpp"

  51 #include "runtime/icache.hpp"
  52 #include "runtime/interfaceSupport.inline.hpp"
  53 #include "runtime/javaThread.hpp"
  54 #include "runtime/jniHandles.inline.hpp"
  55 #include "runtime/sharedRuntime.hpp"

  56 #include "runtime/stubRoutines.hpp"
  57 #include "utilities/globalDefinitions.hpp"
  58 #include "utilities/integerCast.hpp"
  59 #include "utilities/powerOfTwo.hpp"

  60 #ifdef COMPILER1
  61 #include "c1/c1_LIRAssembler.hpp"
  62 #endif
  63 #ifdef COMPILER2
  64 #include "oops/oop.hpp"
  65 #include "opto/compile.hpp"
  66 #include "opto/node.hpp"
  67 #include "opto/output.hpp"
  68 #endif
  69 
  70 #include <sys/types.h>
  71 
  72 #ifdef PRODUCT
  73 #define BLOCK_COMMENT(str) /* nothing */
  74 #else
  75 #define BLOCK_COMMENT(str) block_comment(str)
  76 #endif
  77 #define STOP(str) stop(str);
  78 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  79 

 488   // narrow OOPs by setting the upper 16 bits in the first
 489   // instruction.
 490   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 491     // Move narrow OOP
 492     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 493     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 494     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 495     instructions = 2;
 496   } else {
 497     // Move wide OOP
 498     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 499     uintptr_t dest = (uintptr_t)o;
 500     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 501     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 502     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 503     instructions = 3;
 504   }
 505   return instructions * NativeInstruction::instruction_size;
 506 }
 507 
 508 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 509   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 510   // We encode narrow ones by setting the upper 16 bits in the first
 511   // instruction.
 512   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 513   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 514          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 515 
 516   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 517 
 518   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 519   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 520   return 2 * NativeInstruction::instruction_size;
 521 }
 522 
 523 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 524   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 525   if (at_return) {
 526     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 527     // we may safely use the sp instead to perform the stack watermark check.
 528     cmp(in_nmethod ? sp : rfp, tmp);
 529     br(Assembler::HI, slow_path);
 530   } else {
 531     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 532   }
 533 }
 534 
 535 void MacroAssembler::rt_call(address dest, Register tmp) {
 536   CodeBlob *cb = CodeCache::find_blob(dest);
 537   if (cb) {
 538     far_call(RuntimeAddress(dest));
 539   } else {
 540     lea(tmp, RuntimeAddress(dest));
 541     blr(tmp);
 542   }

2006   ldarb(scratch, scratch);
2007   cmp(scratch, InstanceKlass::fully_initialized);
2008   br(Assembler::EQ, *L_fast_path);
2009 
2010   // Fast path check: current thread is initializer thread
2011   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2012   cmp(rthread, scratch);
2013 
2014   if (L_slow_path == &L_fallthrough) {
2015     br(Assembler::EQ, *L_fast_path);
2016     bind(*L_slow_path);
2017   } else if (L_fast_path == &L_fallthrough) {
2018     br(Assembler::NE, *L_slow_path);
2019     bind(*L_fast_path);
2020   } else {
2021     Unimplemented();
2022   }
2023 }
2024 
2025 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2026   if (!VerifyOops) return;




2027 
2028   // Pass register number to verify_oop_subroutine
2029   const char* b = nullptr;
2030   {
2031     ResourceMark rm;
2032     stringStream ss;
2033     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2034     b = code_string(ss.as_string());
2035   }
2036   BLOCK_COMMENT("verify_oop {");
2037 
2038   strip_return_address(); // This might happen within a stack frame.
2039   protect_return_address();
2040   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2041   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2042 
2043   mov(r0, reg);
2044   movptr(rscratch1, (uintptr_t)(address)b);
2045 
2046   // call indirectly to solve generation ordering problem
2047   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2048   ldr(rscratch2, Address(rscratch2));
2049   blr(rscratch2);
2050 
2051   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2052   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2053   authenticate_return_address();
2054 
2055   BLOCK_COMMENT("} verify_oop");
2056 }
2057 
2058 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2059   if (!VerifyOops) return;




2060 
2061   const char* b = nullptr;
2062   {
2063     ResourceMark rm;
2064     stringStream ss;
2065     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2066     b = code_string(ss.as_string());
2067   }
2068   BLOCK_COMMENT("verify_oop_addr {");
2069 
2070   strip_return_address(); // This might happen within a stack frame.
2071   protect_return_address();
2072   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2073   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2074 
2075   // addr may contain sp so we will have to adjust it based on the
2076   // pushes that we just did.
2077   if (addr.uses(sp)) {
2078     lea(r0, addr);
2079     ldr(r0, Address(r0, 4 * wordSize));

2292   call_VM_leaf_base(entry_point, 1);
2293 }
2294 
2295 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2296   assert_different_registers(arg_1, c_rarg0);
2297   pass_arg0(this, arg_0);
2298   pass_arg1(this, arg_1);
2299   call_VM_leaf_base(entry_point, 2);
2300 }
2301 
2302 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2303                                   Register arg_1, Register arg_2) {
2304   assert_different_registers(arg_1, c_rarg0);
2305   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2306   pass_arg0(this, arg_0);
2307   pass_arg1(this, arg_1);
2308   pass_arg2(this, arg_2);
2309   call_VM_leaf_base(entry_point, 3);
2310 }
2311 




2312 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2313   pass_arg0(this, arg_0);
2314   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2315 }
2316 
2317 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2318 
2319   assert_different_registers(arg_0, c_rarg1);
2320   pass_arg1(this, arg_1);
2321   pass_arg0(this, arg_0);
2322   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2323 }
2324 
2325 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2326   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2327   assert_different_registers(arg_1, c_rarg2);
2328   pass_arg2(this, arg_2);
2329   pass_arg1(this, arg_1);
2330   pass_arg0(this, arg_0);
2331   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2337   assert_different_registers(arg_2, c_rarg3);
2338   pass_arg3(this, arg_3);
2339   pass_arg2(this, arg_2);
2340   pass_arg1(this, arg_1);
2341   pass_arg0(this, arg_0);
2342   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2343 }
2344 
2345 void MacroAssembler::null_check(Register reg, int offset) {
2346   if (needs_explicit_null_check(offset)) {
2347     // provoke OS null exception if reg is null by
2348     // accessing M[reg] w/o changing any registers
2349     // NOTE: this is plenty to provoke a segv
2350     ldr(zr, Address(reg));
2351   } else {
2352     // nothing to do, (later) access of M[reg + offset]
2353     // will provoke OS null exception if reg is null
2354   }
2355 }
2356 














































































2357 // MacroAssembler protected routines needed to implement
2358 // public methods
2359 
2360 void MacroAssembler::mov(Register r, Address dest) {
2361   code_section()->relocate(pc(), dest.rspec());
2362   uint64_t imm64 = (uint64_t)dest.target();
2363   movptr(r, imm64);
2364 }
2365 
2366 // Move a constant pointer into r.  In AArch64 mode the virtual
2367 // address space is 48 bits in size, so we only need three
2368 // instructions to create a patchable instruction sequence that can
2369 // reach anywhere.
2370 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2371 #ifndef PRODUCT
2372   {
2373     char buffer[64];
2374     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2375     block_comment(buffer);
2376   }

5071   adrp(rscratch1, src2, offset);
5072   ldr(rscratch1, Address(rscratch1, offset));
5073   cmp(src1, rscratch1);
5074 }
5075 
5076 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5077   cmp(obj1, obj2);
5078 }
5079 
5080 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5081   load_method_holder(rresult, rmethod);
5082   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5083 }
5084 
5085 void MacroAssembler::load_method_holder(Register holder, Register method) {
5086   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5087   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5088   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5089 }
5090 








5091 // Loads the obj's Klass* into dst.
5092 // Preserves all registers (incl src, rscratch1 and rscratch2).
5093 // Input:
5094 // src - the oop we want to load the klass from.
5095 // dst - output narrow klass.
5096 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5097   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5098   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5099   lsr(dst, dst, markWord::klass_shift);
5100 }
5101 
5102 void MacroAssembler::load_klass(Register dst, Register src) {
5103   if (UseCompactObjectHeaders) {
5104     load_narrow_klass_compact(dst, src);
5105   } else {
5106     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5107   }
5108   decode_klass_not_null(dst);
5109 }
5110 

5162   } else {
5163     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5164   }
5165   if (CompressedKlassPointers::base() == nullptr) {
5166     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5167     return;
5168   } else if (!AOTCodeCache::is_on_for_dump() &&
5169              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5170              && CompressedKlassPointers::shift() == 0) {
5171     // Only the bottom 32 bits matter
5172     cmpw(klass, tmp);
5173     return;
5174   }
5175   decode_klass_not_null(tmp);
5176   cmp(klass, tmp);
5177 }
5178 
5179 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5180   if (UseCompactObjectHeaders) {
5181     load_narrow_klass_compact(tmp1, obj1);
5182     load_narrow_klass_compact(tmp2,  obj2);
5183   } else {
5184     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5185     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5186   }
5187   cmpw(tmp1, tmp2);
5188 }
5189 





5190 void MacroAssembler::store_klass(Register dst, Register src) {
5191   // FIXME: Should this be a store release?  concurrent gcs assumes
5192   // klass length is valid if klass field is not null.
5193   assert(!UseCompactObjectHeaders, "not with compact headers");
5194   encode_klass_not_null(src);
5195   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5196 }
5197 
5198 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5199   assert(!UseCompactObjectHeaders, "not with compact headers");
5200   // Store to klass gap in destination
5201   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5202 }
5203 
5204 // Algorithm must match CompressedOops::encode.
5205 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5206 #ifdef ASSERT
5207   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5208 #endif
5209   verify_oop_msg(s, "broken oop in encode_heap_oop");

5559   if (as_raw) {
5560     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5561   } else {
5562     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5563   }
5564 }
5565 
5566 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5567                                      Address dst, Register val,
5568                                      Register tmp1, Register tmp2, Register tmp3) {
5569   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5570   decorators = AccessInternal::decorator_fixup(decorators, type);
5571   bool as_raw = (decorators & AS_RAW) != 0;
5572   if (as_raw) {
5573     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5574   } else {
5575     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5576   }
5577 }
5578 






















5579 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5580                                    Register tmp2, DecoratorSet decorators) {
5581   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5582 }
5583 
5584 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5585                                             Register tmp2, DecoratorSet decorators) {
5586   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5587 }
5588 
5589 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5590                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5591   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5592 }
5593 
5594 // Used for storing nulls.
5595 void MacroAssembler::store_heap_oop_null(Address dst) {
5596   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5597 }
5598 

5673     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5674     cmp(rscratch2, rscratch1);
5675     br(Assembler::HS, next);
5676     STOP("assert(top >= start)");
5677     should_not_reach_here();
5678 
5679     bind(next);
5680     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5681     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5682     cmp(rscratch2, rscratch1);
5683     br(Assembler::HS, ok);
5684     STOP("assert(top <= end)");
5685     should_not_reach_here();
5686 
5687     bind(ok);
5688     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5689   }
5690 #endif
5691 }
5692 















5693 // Writes to stack successive pages until offset reached to check for
5694 // stack overflow + shadow pages.  This clobbers tmp.
5695 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5696   assert_different_registers(tmp, size, rscratch1);
5697   mov(tmp, sp);
5698   // Bang stack for total size given plus shadow page size.
5699   // Bang one page at a time because large size can bang beyond yellow and
5700   // red zones.
5701   Label loop;
5702   mov(rscratch1, (int)os::vm_page_size());
5703   bind(loop);
5704   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5705   subsw(size, size, rscratch1);
5706   str(size, Address(tmp));
5707   br(Assembler::GT, loop);
5708 
5709   // Bang down shadow pages too.
5710   // At this point, (tmp-0) is the last address touched, so don't
5711   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5712   // was post-decremented.)  Skip this address by starting at i=1, and

5779 
5780   // Strictly speaking the card table base isn't an address at all, and it might
5781   // even be negative. It is thus materialised as a constant.
5782   mov(reg, (uint64_t)ctbs->card_table_base_const());
5783 }
5784 
5785 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5786 #if INCLUDE_CDS
5787   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5788   if (AOTCodeCache::is_on_for_dump()) {
5789     // all aotrc field addresses should be registered in the AOTCodeCache address table
5790     lea(reg, ExternalAddress(a));
5791   } else {
5792     mov(reg, (uint64_t)a);
5793   }
5794 #else
5795   ShouldNotReachHere();
5796 #endif
5797 }
5798 

5799 void MacroAssembler::build_frame(int framesize) {





5800   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5801   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5802   protect_return_address();
5803   if (framesize < ((1 << 9) + 2 * wordSize)) {
5804     sub(sp, sp, framesize);
5805     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));





5806     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5807   } else {
5808     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));





5809     if (PreserveFramePointer) mov(rfp, sp);
5810     if (framesize < ((1 << 12) + 2 * wordSize))
5811       sub(sp, sp, framesize - 2 * wordSize);
5812     else {
5813       mov(rscratch1, framesize - 2 * wordSize);
5814       sub(sp, sp, rscratch1);
5815     }
5816   }
5817   verify_cross_modify_fence_not_required();
5818 }
5819 
5820 void MacroAssembler::remove_frame(int framesize) {
5821   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5822   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5823   if (framesize < ((1 << 9) + 2 * wordSize)) {
5824     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5825     add(sp, sp, framesize);
5826   } else {
5827     if (framesize < ((1 << 12) + 2 * wordSize))
5828       add(sp, sp, framesize - 2 * wordSize);
5829     else {
5830       mov(rscratch1, framesize - 2 * wordSize);
5831       add(sp, sp, rscratch1);
5832     }
5833     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5834   }
5835   authenticate_return_address();
5836 }
5837 


















































































5838 
5839 // This method counts leading positive bytes (highest bit not set) in provided byte array
5840 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5841     // Simple and most common case of aligned small array which is not at the
5842     // end of memory page is placed here. All other cases are in stub.
5843     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5844     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5845     assert_different_registers(ary1, len, result);
5846 
5847     mov(result, len);
5848     cmpw(len, 0);
5849     br(LE, DONE);
5850     cmpw(len, 4 * wordSize);
5851     br(GE, STUB_LONG); // size > 32 then go to stub
5852 
5853     int shift = 64 - exact_log2(os::vm_page_size());
5854     lsl(rscratch1, ary1, shift);
5855     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5856     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5857     br(CS, STUB); // at the end of page then go to stub

6741 // On other systems, the helper is a usual C function.
6742 //
6743 void MacroAssembler::get_thread(Register dst) {
6744   RegSet saved_regs =
6745     BSD_ONLY(RegSet::range(r0, r17) + lr - dst)
6746     NOT_BSD (RegSet::range(r0, r1)  + lr - dst);
6747 
6748   protect_return_address();
6749   push(saved_regs, sp);
6750 
6751   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6752   blr(lr);
6753   if (dst != c_rarg0) {
6754     mov(dst, c_rarg0);
6755   }
6756 
6757   pop(saved_regs, sp);
6758   authenticate_return_address();
6759 }
6760 




























































































































































































































































































































































































































































































6761 void MacroAssembler::cache_wb(Address line) {
6762   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6763   assert(line.index() == noreg, "index should be noreg");
6764   assert(line.offset() == 0, "offset should be 0");
6765   // would like to assert this
6766   // assert(line._ext.shift == 0, "shift should be zero");
6767   if (VM_Version::supports_dcpop()) {
6768     // writeback using clear virtual address to point of persistence
6769     dc(Assembler::CVAP, line.base());
6770   } else {
6771     // no need to generate anything as Unsafe.writebackMemory should
6772     // never invoke this stub
6773   }
6774 }
6775 
6776 void MacroAssembler::cache_wbsync(bool is_pre) {
6777   // we only need a barrier post sync
6778   if (!is_pre) {
6779     membar(Assembler::AnyAny);
6780   }

7176   }
7177 
7178   // Check if the lock-stack is full.
7179   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7180   cmpw(top, (unsigned)LockStack::end_offset());
7181   br(Assembler::GE, slow);
7182 
7183   // Check for recursion.
7184   subw(t, top, oopSize);
7185   ldr(t, Address(rthread, t));
7186   cmp(obj, t);
7187   br(Assembler::EQ, push);
7188 
7189   // Check header for monitor (0b10).
7190   tst(mark, markWord::monitor_value);
7191   br(Assembler::NE, slow);
7192 
7193   // Try to lock. Transition lock bits 0b01 => 0b00
7194   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7195   orr(mark, mark, markWord::unlocked_value);



7196   eor(t, mark, markWord::unlocked_value);
7197   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7198           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7199   br(Assembler::NE, slow);
7200 
7201   bind(push);
7202   // After successful lock, push object on lock-stack.
7203   str(obj, Address(rthread, top));
7204   addw(top, top, oopSize);
7205   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7206 }
7207 
7208 // Implements fast-unlocking.
7209 //
7210 // - obj: the object to be unlocked
7211 // - t1, t2, t3: temporary registers
7212 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7213 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7214   // cmpxchg clobbers rscratch1.
7215   assert_different_registers(obj, t1, t2, t3, rscratch1);

   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 "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "ci/ciInlineKlass.hpp"
  30 #include "code/compiledIC.hpp"
  31 #include "compiler/compileTask.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "compiler/oopMap.hpp"
  34 #include "gc/shared/barrierSet.hpp"
  35 #include "gc/shared/barrierSetAssembler.hpp"
  36 #include "gc/shared/cardTableBarrierSet.hpp"
  37 #include "gc/shared/cardTable.hpp"
  38 #include "gc/shared/collectedHeap.hpp"
  39 #include "gc/shared/tlab_globals.hpp"
  40 #include "interpreter/bytecodeHistogram.hpp"
  41 #include "interpreter/interpreter.hpp"
  42 #include "interpreter/interpreterRuntime.hpp"
  43 #include "jvm.h"
  44 #include "memory/resourceArea.hpp"
  45 #include "memory/universe.hpp"
  46 #include "nativeInst_aarch64.hpp"
  47 #include "oops/accessDecorators.hpp"
  48 #include "oops/compressedKlass.inline.hpp"
  49 #include "oops/compressedOops.inline.hpp"
  50 #include "oops/klass.inline.hpp"
  51 #include "oops/resolvedFieldEntry.hpp"
  52 #include "runtime/arguments.hpp"
  53 #include "runtime/continuation.hpp"
  54 #include "runtime/globals.hpp"
  55 #include "runtime/icache.hpp"
  56 #include "runtime/interfaceSupport.inline.hpp"
  57 #include "runtime/javaThread.hpp"
  58 #include "runtime/jniHandles.inline.hpp"
  59 #include "runtime/sharedRuntime.hpp"
  60 #include "runtime/signature_cc.hpp"
  61 #include "runtime/stubRoutines.hpp"
  62 #include "utilities/globalDefinitions.hpp"
  63 #include "utilities/integerCast.hpp"
  64 #include "utilities/powerOfTwo.hpp"
  65 #include "vmreg_aarch64.inline.hpp"
  66 #ifdef COMPILER1
  67 #include "c1/c1_LIRAssembler.hpp"
  68 #endif
  69 #ifdef COMPILER2
  70 #include "oops/oop.hpp"
  71 #include "opto/compile.hpp"
  72 #include "opto/node.hpp"
  73 #include "opto/output.hpp"
  74 #endif
  75 
  76 #include <sys/types.h>
  77 
  78 #ifdef PRODUCT
  79 #define BLOCK_COMMENT(str) /* nothing */
  80 #else
  81 #define BLOCK_COMMENT(str) block_comment(str)
  82 #endif
  83 #define STOP(str) stop(str);
  84 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  85 

 494   // narrow OOPs by setting the upper 16 bits in the first
 495   // instruction.
 496   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 497     // Move narrow OOP
 498     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 499     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 500     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 501     instructions = 2;
 502   } else {
 503     // Move wide OOP
 504     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 505     uintptr_t dest = (uintptr_t)o;
 506     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 507     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 508     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 509     instructions = 3;
 510   }
 511   return instructions * NativeInstruction::instruction_size;
 512 }
 513 















 514 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 515   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 516   if (at_return) {
 517     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 518     // we may safely use the sp instead to perform the stack watermark check.
 519     cmp(in_nmethod ? sp : rfp, tmp);
 520     br(Assembler::HI, slow_path);
 521   } else {
 522     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 523   }
 524 }
 525 
 526 void MacroAssembler::rt_call(address dest, Register tmp) {
 527   CodeBlob *cb = CodeCache::find_blob(dest);
 528   if (cb) {
 529     far_call(RuntimeAddress(dest));
 530   } else {
 531     lea(tmp, RuntimeAddress(dest));
 532     blr(tmp);
 533   }

1997   ldarb(scratch, scratch);
1998   cmp(scratch, InstanceKlass::fully_initialized);
1999   br(Assembler::EQ, *L_fast_path);
2000 
2001   // Fast path check: current thread is initializer thread
2002   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2003   cmp(rthread, scratch);
2004 
2005   if (L_slow_path == &L_fallthrough) {
2006     br(Assembler::EQ, *L_fast_path);
2007     bind(*L_slow_path);
2008   } else if (L_fast_path == &L_fallthrough) {
2009     br(Assembler::NE, *L_slow_path);
2010     bind(*L_fast_path);
2011   } else {
2012     Unimplemented();
2013   }
2014 }
2015 
2016 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2017   if (!VerifyOops || VerifyAdapterSharing) {
2018     // Below address of the code string confuses VerifyAdapterSharing
2019     // because it may differ between otherwise equivalent adapters.
2020     return;
2021   }
2022 
2023   // Pass register number to verify_oop_subroutine
2024   const char* b = nullptr;
2025   {
2026     ResourceMark rm;
2027     stringStream ss;
2028     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2029     b = code_string(ss.as_string());
2030   }
2031   BLOCK_COMMENT("verify_oop {");
2032 
2033   strip_return_address(); // This might happen within a stack frame.
2034   protect_return_address();
2035   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2036   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2037 
2038   mov(r0, reg);
2039   movptr(rscratch1, (uintptr_t)(address)b);
2040 
2041   // call indirectly to solve generation ordering problem
2042   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2043   ldr(rscratch2, Address(rscratch2));
2044   blr(rscratch2);
2045 
2046   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2047   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2048   authenticate_return_address();
2049 
2050   BLOCK_COMMENT("} verify_oop");
2051 }
2052 
2053 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2054   if (!VerifyOops || VerifyAdapterSharing) {
2055     // Below address of the code string confuses VerifyAdapterSharing
2056     // because it may differ between otherwise equivalent adapters.
2057     return;
2058   }
2059 
2060   const char* b = nullptr;
2061   {
2062     ResourceMark rm;
2063     stringStream ss;
2064     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2065     b = code_string(ss.as_string());
2066   }
2067   BLOCK_COMMENT("verify_oop_addr {");
2068 
2069   strip_return_address(); // This might happen within a stack frame.
2070   protect_return_address();
2071   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2072   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2073 
2074   // addr may contain sp so we will have to adjust it based on the
2075   // pushes that we just did.
2076   if (addr.uses(sp)) {
2077     lea(r0, addr);
2078     ldr(r0, Address(r0, 4 * wordSize));

2291   call_VM_leaf_base(entry_point, 1);
2292 }
2293 
2294 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2295   assert_different_registers(arg_1, c_rarg0);
2296   pass_arg0(this, arg_0);
2297   pass_arg1(this, arg_1);
2298   call_VM_leaf_base(entry_point, 2);
2299 }
2300 
2301 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2302                                   Register arg_1, Register arg_2) {
2303   assert_different_registers(arg_1, c_rarg0);
2304   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2305   pass_arg0(this, arg_0);
2306   pass_arg1(this, arg_1);
2307   pass_arg2(this, arg_2);
2308   call_VM_leaf_base(entry_point, 3);
2309 }
2310 
2311 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2312   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2313 }
2314 
2315 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2316   pass_arg0(this, arg_0);
2317   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2318 }
2319 
2320 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2321 
2322   assert_different_registers(arg_0, c_rarg1);
2323   pass_arg1(this, arg_1);
2324   pass_arg0(this, arg_0);
2325   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2326 }
2327 
2328 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2329   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2330   assert_different_registers(arg_1, c_rarg2);
2331   pass_arg2(this, arg_2);
2332   pass_arg1(this, arg_1);
2333   pass_arg0(this, arg_0);
2334   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2340   assert_different_registers(arg_2, c_rarg3);
2341   pass_arg3(this, arg_3);
2342   pass_arg2(this, arg_2);
2343   pass_arg1(this, arg_1);
2344   pass_arg0(this, arg_0);
2345   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2346 }
2347 
2348 void MacroAssembler::null_check(Register reg, int offset) {
2349   if (needs_explicit_null_check(offset)) {
2350     // provoke OS null exception if reg is null by
2351     // accessing M[reg] w/o changing any registers
2352     // NOTE: this is plenty to provoke a segv
2353     ldr(zr, Address(reg));
2354   } else {
2355     // nothing to do, (later) access of M[reg + offset]
2356     // will provoke OS null exception if reg is null
2357   }
2358 }
2359 
2360 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2361   assert_different_registers(markword, rscratch2);
2362   mov(rscratch2, markWord::inline_type_pattern_mask);
2363   andr(markword, markword, rscratch2);
2364   mov(rscratch2, markWord::inline_type_pattern);
2365   cmp(markword, rscratch2);
2366   br(Assembler::EQ, is_inline_type);
2367 }
2368 
2369 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
2370   assert_different_registers(tmp, rscratch1);
2371   if (can_be_null) {
2372     cbz(object, not_inline_type);
2373   }
2374   const int is_inline_type_mask = markWord::inline_type_pattern;
2375   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2376   mov(rscratch1, is_inline_type_mask);
2377   andr(tmp, tmp, rscratch1);
2378   cmp(tmp, rscratch1);
2379   br(Assembler::NE, not_inline_type);
2380 }
2381 
2382 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2383   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2384   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2385 }
2386 
2387 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2388   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2389   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2390 }
2391 
2392 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2393   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2394   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2395 }
2396 
2397 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2398   Label test_mark_word;
2399   // load mark word
2400   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2401   // check displaced
2402   tst(temp_reg, markWord::unlocked_value);
2403   br(Assembler::NE, test_mark_word);
2404   // slow path use klass prototype
2405   load_prototype_header(temp_reg, oop);
2406 
2407   bind(test_mark_word);
2408   andr(temp_reg, temp_reg, test_bit);
2409   if (jmp_set) {
2410     cbnz(temp_reg, jmp_label);
2411   } else {
2412     cbz(temp_reg, jmp_label);
2413   }
2414 }
2415 
2416 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2417   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2418 }
2419 
2420 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2421                                                   Label&is_non_flat_array) {
2422   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2423 }
2424 
2425 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2426   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2427 }
2428 
2429 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2430   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2431 }
2432 
2433 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2434   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2435   br(Assembler::NE, is_flat_array);
2436 }
2437 
2438 // MacroAssembler protected routines needed to implement
2439 // public methods
2440 
2441 void MacroAssembler::mov(Register r, Address dest) {
2442   code_section()->relocate(pc(), dest.rspec());
2443   uint64_t imm64 = (uint64_t)dest.target();
2444   movptr(r, imm64);
2445 }
2446 
2447 // Move a constant pointer into r.  In AArch64 mode the virtual
2448 // address space is 48 bits in size, so we only need three
2449 // instructions to create a patchable instruction sequence that can
2450 // reach anywhere.
2451 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2452 #ifndef PRODUCT
2453   {
2454     char buffer[64];
2455     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2456     block_comment(buffer);
2457   }

5152   adrp(rscratch1, src2, offset);
5153   ldr(rscratch1, Address(rscratch1, offset));
5154   cmp(src1, rscratch1);
5155 }
5156 
5157 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5158   cmp(obj1, obj2);
5159 }
5160 
5161 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5162   load_method_holder(rresult, rmethod);
5163   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5164 }
5165 
5166 void MacroAssembler::load_method_holder(Register holder, Register method) {
5167   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5168   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5169   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5170 }
5171 
5172 void MacroAssembler::load_metadata(Register dst, Register src) {
5173   if (UseCompactObjectHeaders) {
5174     load_narrow_klass_compact(dst, src);
5175   } else {
5176     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5177   }
5178 }
5179 
5180 // Loads the obj's Klass* into dst.
5181 // Preserves all registers (incl src, rscratch1 and rscratch2).
5182 // Input:
5183 // src - the oop we want to load the klass from.
5184 // dst - output narrow klass.
5185 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5186   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5187   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5188   lsr(dst, dst, markWord::klass_shift);
5189 }
5190 
5191 void MacroAssembler::load_klass(Register dst, Register src) {
5192   if (UseCompactObjectHeaders) {
5193     load_narrow_klass_compact(dst, src);
5194   } else {
5195     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5196   }
5197   decode_klass_not_null(dst);
5198 }
5199 

5251   } else {
5252     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5253   }
5254   if (CompressedKlassPointers::base() == nullptr) {
5255     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5256     return;
5257   } else if (!AOTCodeCache::is_on_for_dump() &&
5258              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5259              && CompressedKlassPointers::shift() == 0) {
5260     // Only the bottom 32 bits matter
5261     cmpw(klass, tmp);
5262     return;
5263   }
5264   decode_klass_not_null(tmp);
5265   cmp(klass, tmp);
5266 }
5267 
5268 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5269   if (UseCompactObjectHeaders) {
5270     load_narrow_klass_compact(tmp1, obj1);
5271     load_narrow_klass_compact(tmp2, obj2);
5272   } else {
5273     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5274     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5275   }
5276   cmpw(tmp1, tmp2);
5277 }
5278 
5279 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5280   load_klass(dst, src);
5281   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5282 }
5283 
5284 void MacroAssembler::store_klass(Register dst, Register src) {
5285   // FIXME: Should this be a store release?  concurrent gcs assumes
5286   // klass length is valid if klass field is not null.
5287   assert(!UseCompactObjectHeaders, "not with compact headers");
5288   encode_klass_not_null(src);
5289   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5290 }
5291 
5292 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5293   assert(!UseCompactObjectHeaders, "not with compact headers");
5294   // Store to klass gap in destination
5295   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5296 }
5297 
5298 // Algorithm must match CompressedOops::encode.
5299 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5300 #ifdef ASSERT
5301   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5302 #endif
5303   verify_oop_msg(s, "broken oop in encode_heap_oop");

5653   if (as_raw) {
5654     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5655   } else {
5656     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5657   }
5658 }
5659 
5660 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5661                                      Address dst, Register val,
5662                                      Register tmp1, Register tmp2, Register tmp3) {
5663   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5664   decorators = AccessInternal::decorator_fixup(decorators, type);
5665   bool as_raw = (decorators & AS_RAW) != 0;
5666   if (as_raw) {
5667     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5668   } else {
5669     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5670   }
5671 }
5672 
5673 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5674                                      Register inline_layout_info) {
5675   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5676   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5677 }
5678 
5679 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5680   ldr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
5681   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5682 }
5683 
5684 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5685   // ((address) (void*) o) + vk->payload_offset();
5686   Register offset = (data == oop) ? rscratch1 : data;
5687   payload_offset(inline_klass, offset);
5688   if (data == oop) {
5689     add(data, data, offset);
5690   } else {
5691     lea(data, Address(oop, offset));
5692   }
5693 }
5694 
5695 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5696                                    Register tmp2, DecoratorSet decorators) {
5697   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5698 }
5699 
5700 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5701                                             Register tmp2, DecoratorSet decorators) {
5702   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5703 }
5704 
5705 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5706                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5707   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5708 }
5709 
5710 // Used for storing nulls.
5711 void MacroAssembler::store_heap_oop_null(Address dst) {
5712   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5713 }
5714 

5789     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5790     cmp(rscratch2, rscratch1);
5791     br(Assembler::HS, next);
5792     STOP("assert(top >= start)");
5793     should_not_reach_here();
5794 
5795     bind(next);
5796     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5797     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5798     cmp(rscratch2, rscratch1);
5799     br(Assembler::HS, ok);
5800     STOP("assert(top <= end)");
5801     should_not_reach_here();
5802 
5803     bind(ok);
5804     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5805   }
5806 #endif
5807 }
5808 
5809 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5810   assert_different_registers(holder_klass, index, layout_info);
5811   InlineLayoutInfo array[2];
5812   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5813   if (is_power_of_2(size)) {
5814     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5815   } else {
5816     mov(layout_info, size);
5817     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5818   }
5819   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5820   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5821   lea(layout_info, Address(layout_info, index));
5822 }
5823 
5824 // Writes to stack successive pages until offset reached to check for
5825 // stack overflow + shadow pages.  This clobbers tmp.
5826 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5827   assert_different_registers(tmp, size, rscratch1);
5828   mov(tmp, sp);
5829   // Bang stack for total size given plus shadow page size.
5830   // Bang one page at a time because large size can bang beyond yellow and
5831   // red zones.
5832   Label loop;
5833   mov(rscratch1, (int)os::vm_page_size());
5834   bind(loop);
5835   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5836   subsw(size, size, rscratch1);
5837   str(size, Address(tmp));
5838   br(Assembler::GT, loop);
5839 
5840   // Bang down shadow pages too.
5841   // At this point, (tmp-0) is the last address touched, so don't
5842   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5843   // was post-decremented.)  Skip this address by starting at i=1, and

5910 
5911   // Strictly speaking the card table base isn't an address at all, and it might
5912   // even be negative. It is thus materialised as a constant.
5913   mov(reg, (uint64_t)ctbs->card_table_base_const());
5914 }
5915 
5916 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5917 #if INCLUDE_CDS
5918   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5919   if (AOTCodeCache::is_on_for_dump()) {
5920     // all aotrc field addresses should be registered in the AOTCodeCache address table
5921     lea(reg, ExternalAddress(a));
5922   } else {
5923     mov(reg, (uint64_t)a);
5924   }
5925 #else
5926   ShouldNotReachHere();
5927 #endif
5928 }
5929 
5930 #ifdef ASSERT
5931 void MacroAssembler::build_frame(int framesize) {
5932   build_frame(framesize, false);
5933 }
5934 #endif
5935 
5936 void MacroAssembler::build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills)) {
5937   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5938   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5939   protect_return_address();
5940   if (framesize < ((1 << 9) + 2 * wordSize)) {
5941     sub(sp, sp, framesize);
5942     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5943       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5944       stp(rscratch1, rscratch1, Address(sp, framesize - 2 * wordSize));
5945     } else {
5946       stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5947     }
5948     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5949   } else {
5950     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5951       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5952       stp(rscratch1, rscratch1, Address(pre(sp, -2 * wordSize)));
5953     } else {
5954       stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5955     }
5956     if (PreserveFramePointer) mov(rfp, sp);
5957     if (framesize < ((1 << 12) + 2 * wordSize))
5958       sub(sp, sp, framesize - 2 * wordSize);
5959     else {
5960       mov(rscratch1, framesize - 2 * wordSize);
5961       sub(sp, sp, rscratch1);
5962     }
5963   }
5964   verify_cross_modify_fence_not_required();
5965 }
5966 
5967 void MacroAssembler::remove_frame(int framesize) {
5968   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5969   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5970   if (framesize < ((1 << 9) + 2 * wordSize)) {
5971     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5972     add(sp, sp, framesize);
5973   } else {
5974     if (framesize < ((1 << 12) + 2 * wordSize))
5975       add(sp, sp, framesize - 2 * wordSize);
5976     else {
5977       mov(rscratch1, framesize - 2 * wordSize);
5978       add(sp, sp, rscratch1);
5979     }
5980     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5981   }
5982   authenticate_return_address();
5983 }
5984 
5985 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
5986   if (needs_stack_repair) {
5987     // The method has a scalarized entry point (where fields of value object arguments
5988     // are passed through registers and stack), and a non-scalarized entry point (where
5989     // value object arguments are given as oops). The non-scalarized entry point will
5990     // first load each field of value object arguments and store them in registers and on
5991     // the stack in a way compatible with the scalarized entry point. To do so, some extra
5992     // stack space might be reserved (if argument registers are not enough). On leaving the
5993     // method, this space must be freed.
5994     //
5995     // In case we used the non-scalarized entry point the stack looks like this:
5996     //
5997     // | Arguments from caller     |
5998     // |---------------------------|  <-- caller's SP
5999     // | Saved LR #1               |
6000     // | Saved FP #1               |
6001     // |---------------------------|
6002     // | Extension space for       |
6003     // |   inline arg (un)packing  |
6004     // |---------------------------|  <-- start of this method's frame
6005     // | Saved LR #2               |
6006     // | Saved FP #2               |
6007     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6008     // | sp_inc                    |
6009     // | method locals             |
6010     // |---------------------------|  <-- SP
6011     //
6012     // There are two copies of FP and LR on the stack. They will be identical at
6013     // first, but that can change.
6014     // If the caller has been deoptimized, LR #1 will be patched to point at the
6015     // deopt blob, and LR #2 will still point into the old method.
6016     // If the saved FP (x29) was not used as the frame pointer, but to store an
6017     // oop, the GC will be aware only of FP #1 as the spilled location of x29 and
6018     // will fix only this one. Overall, FP/LR #2 are not reliable and are simply
6019     // needed to add space between the extension space and the locals, as there
6020     // would be between the real arguments and the locals if we don't need to
6021     // do unpacking (from the scalarized entry point).
6022     //
6023     // When restoring, one must then load FP #1 into x29, and LR #1 into x30,
6024     // while keeping in mind that from the scalarized entry point, there will be
6025     // only one copy of each. Indeed, in the case we used the scalarized calling
6026     // convention, the stack looks like this:
6027     //
6028     // | Arguments from caller     |
6029     // |---------------------------|  <-- caller's SP / start of this method's frame
6030     // | Saved LR                  |
6031     // | Saved FP                  |
6032     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6033     // | sp_inc                    |
6034     // | method locals             |
6035     // |---------------------------|  <-- SP
6036     //
6037     // The sp_inc stack slot holds the total size of the frame including the
6038     // extension space minus two words for the saved FP and LR. That is how to
6039     // find FP/LR #1. This size is expressed in bytes. Be careful when using it
6040     // from C++ in pointer arithmetic; you might need to divide it by wordSize.
6041     //
6042     // One can find sp_inc since the start the method's frame is SP + initial_framesize.
6043 
6044     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
6045 
6046     ldr(rscratch1, Address(sp, sp_inc_offset));
6047     add(sp, sp, rscratch1);
6048     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6049   } else {
6050     remove_frame(initial_framesize);
6051   }
6052 }
6053 
6054 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
6055   int real_frame_size = frame_size + sp_inc;
6056   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
6057   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
6058   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
6059 
6060   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
6061 
6062   // Subtract two words for the saved FP and LR as these will be popped
6063   // separately. See remove_frame above.
6064   mov(rscratch1, real_frame_size - 2*wordSize);
6065   str(rscratch1, Address(sp, sp_inc_offset));
6066 }
6067 
6068 // This method counts leading positive bytes (highest bit not set) in provided byte array
6069 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6070     // Simple and most common case of aligned small array which is not at the
6071     // end of memory page is placed here. All other cases are in stub.
6072     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6073     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6074     assert_different_registers(ary1, len, result);
6075 
6076     mov(result, len);
6077     cmpw(len, 0);
6078     br(LE, DONE);
6079     cmpw(len, 4 * wordSize);
6080     br(GE, STUB_LONG); // size > 32 then go to stub
6081 
6082     int shift = 64 - exact_log2(os::vm_page_size());
6083     lsl(rscratch1, ary1, shift);
6084     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6085     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6086     br(CS, STUB); // at the end of page then go to stub

6970 // On other systems, the helper is a usual C function.
6971 //
6972 void MacroAssembler::get_thread(Register dst) {
6973   RegSet saved_regs =
6974     BSD_ONLY(RegSet::range(r0, r17) + lr - dst)
6975     NOT_BSD (RegSet::range(r0, r1)  + lr - dst);
6976 
6977   protect_return_address();
6978   push(saved_regs, sp);
6979 
6980   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6981   blr(lr);
6982   if (dst != c_rarg0) {
6983     mov(dst, c_rarg0);
6984   }
6985 
6986   pop(saved_regs, sp);
6987   authenticate_return_address();
6988 }
6989 
6990 #ifdef COMPILER2
6991 // C2 compiled method's prolog code
6992 // Moved here from aarch64.ad to support Valhalla code below
6993 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
6994   if (C->clinit_barrier_on_entry()) {
6995     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
6996 
6997     Label L_skip_barrier;
6998 
6999     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
7000     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
7001     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
7002     bind(L_skip_barrier);
7003   }
7004 
7005   if (C->max_vector_size() > 0) {
7006     reinitialize_ptrue();
7007   }
7008 
7009   int bangsize = C->output()->bang_size_in_bytes();
7010   if (C->output()->need_stack_bang(bangsize))
7011     generate_stack_overflow_check(bangsize);
7012 
7013   // n.b. frame size includes space for return pc and rfp
7014   const long framesize = C->output()->frame_size_in_bytes();
7015   build_frame(framesize DEBUG_ONLY(COMMA sp_inc != 0));
7016 
7017   if (C->needs_stack_repair()) {
7018     save_stack_increment(sp_inc, framesize);
7019   }
7020 
7021   if (VerifyStackAtCalls) {
7022     Unimplemented();
7023   }
7024 }
7025 #endif // COMPILER2
7026 
7027 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
7028   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
7029   // An inline type might be returned. If fields are in registers we
7030   // need to allocate an inline type instance and initialize it with
7031   // the value of the fields.
7032   Label skip;
7033   // We only need a new buffered inline type if a new one is not returned
7034   tbz(r0, 0, skip);
7035   int call_offset = -1;
7036 
7037   // Be careful not to clobber r1-7 which hold returned fields
7038   // Also do not use callee-saved registers as these may be live in the interpreter
7039   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
7040 
7041   // The following code is similar to the instance allocation code in TemplateTable::_new
7042   //  but has some slight differences,
7043   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
7044   // allocating is not necessary if vk != nullptr, etc.
7045   Label slow_case;
7046   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
7047   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
7048 
7049   if (vk != nullptr) {
7050     // Called from C1, where the return type is statically known.
7051     movptr(klass, (intptr_t)vk->get_InlineKlass());
7052     jint lh = vk->layout_helper();
7053     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
7054     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
7055       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
7056     } else {
7057       b(slow_case);
7058     }
7059   } else {
7060     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
7061     andr(klass, r0, -2);
7062     if (UseTLAB) {
7063       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
7064       tst(tmp2, Klass::_lh_instance_slow_path_bit);
7065       br(Assembler::NE, slow_case);
7066       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
7067     } else {
7068       b(slow_case);
7069     }
7070   }
7071   if (UseTLAB) {
7072     // 2. Initialize buffered inline instance header
7073     Register buffer_obj = r0;
7074     if (UseCompactObjectHeaders) {
7075       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7076       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7077     } else {
7078       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7079       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7080       store_klass_gap(buffer_obj, zr);
7081       if (vk == nullptr) {
7082         // store_klass corrupts klass, so save it for later use (interpreter case only).
7083         mov(tmp1, klass);
7084       }
7085       store_klass(buffer_obj, klass);
7086       klass = tmp1;
7087     }
7088     // 3. Initialize its fields with an inline class specific handler
7089     if (vk != nullptr) {
7090       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7091     } else {
7092       ldr(tmp1, Address(klass, InlineKlass::adr_members_offset()));
7093       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7094       blr(tmp1);
7095     }
7096 
7097     membar(Assembler::StoreStore);
7098     b(skip);
7099   } else {
7100     // Must have already branched to slow_case above.
7101     DEBUG_ONLY(should_not_reach_here());
7102   }
7103   bind(slow_case);
7104   // We failed to allocate a new inline type, fall back to a runtime
7105   // call. Some oop field may be live in some registers but we can't
7106   // tell. That runtime call will take care of preserving them
7107   // across a GC if there's one.
7108   mov(r0, r0_preserved);
7109 
7110   if (from_interpreter) {
7111     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
7112   } else {
7113     far_call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
7114     call_offset = offset();
7115   }
7116   membar(Assembler::StoreStore);
7117 
7118   bind(skip);
7119   return call_offset;
7120 }
7121 
7122 // Move a value between registers/stack slots and update the reg_state
7123 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7124   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7125   if (reg_state[to->value()] == reg_written) {
7126     return true; // Already written
7127   }
7128 
7129   if (from != to && bt != T_VOID) {
7130     if (reg_state[to->value()] == reg_readonly) {
7131       return false; // Not yet writable
7132     }
7133     if (from->is_reg()) {
7134       if (to->is_reg()) {
7135         if (from->is_Register() && to->is_Register()) {
7136           mov(to->as_Register(), from->as_Register());
7137         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7138           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7139         } else {
7140           ShouldNotReachHere();
7141         }
7142       } else {
7143         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7144         Address to_addr = Address(sp, st_off);
7145         if (from->is_FloatRegister()) {
7146           if (bt == T_DOUBLE) {
7147              strd(from->as_FloatRegister(), to_addr);
7148           } else {
7149              assert(bt == T_FLOAT, "must be float");
7150              strs(from->as_FloatRegister(), to_addr);
7151           }
7152         } else {
7153           str(from->as_Register(), to_addr);
7154         }
7155       }
7156     } else {
7157       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7158       if (to->is_reg()) {
7159         if (to->is_FloatRegister()) {
7160           if (bt == T_DOUBLE) {
7161             ldrd(to->as_FloatRegister(), from_addr);
7162           } else {
7163             assert(bt == T_FLOAT, "must be float");
7164             ldrs(to->as_FloatRegister(), from_addr);
7165           }
7166         } else {
7167           ldr(to->as_Register(), from_addr);
7168         }
7169       } else {
7170         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7171         ldr(rscratch1, from_addr);
7172         str(rscratch1, Address(sp, st_off));
7173       }
7174     }
7175   }
7176 
7177   // Update register states
7178   reg_state[from->value()] = reg_writable;
7179   reg_state[to->value()] = reg_written;
7180   return true;
7181 }
7182 
7183 // Calculate the extra stack space required for packing or unpacking inline
7184 // args and adjust the stack pointer
7185 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7186   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7187   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7188   assert(sp_inc > 0, "sanity");
7189 
7190   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7191   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7192 
7193   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7194   if (sp_inc < (1 << 9)) {
7195     sub(sp, sp, sp_inc);   // Fits in an immediate
7196   } else {
7197     mov(rscratch1, sp_inc);
7198     sub(sp, sp, rscratch1);
7199   }
7200 
7201   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7202 }
7203 
7204 // Read all fields from an inline type oop and store the values in registers/stack slots
7205 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7206                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7207                                           RegState reg_state[]) {
7208   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7209   assert(from->is_valid(), "source must be valid");
7210   bool progress = false;
7211 #ifdef ASSERT
7212   const int start_offset = offset();
7213 #endif
7214 
7215   Label L_null, L_notNull;
7216   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7217   Register tmp1 = r10;
7218   Register tmp2 = r11;
7219 
7220 #ifdef ASSERT
7221   RegSet clobbered_gp_regs = MacroAssembler::call_clobbered_gp_registers();
7222   assert(clobbered_gp_regs.contains(tmp1), "tmp1 must be saved explicitly if it's not a clobber");
7223   assert(clobbered_gp_regs.contains(tmp2), "tmp2 must be saved explicitly if it's not a clobber");
7224   assert(clobbered_gp_regs.contains(r14), "r14 must be saved explicitly if it's not a clobber");
7225 #endif
7226 
7227   Register fromReg = noreg;
7228   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
7229   bool done = true;
7230   bool mark_done = true;
7231   VMReg toReg;
7232   BasicType bt;
7233   // Check if argument requires a null check
7234   bool null_check = false;
7235   VMReg nullCheckReg;
7236   while (stream.next(nullCheckReg, bt)) {
7237     if (sig->at(stream.sig_index())._offset == -1) {
7238       null_check = true;
7239       break;
7240     }
7241   }
7242   stream.reset(sig_index, to_index);
7243   while (stream.next(toReg, bt)) {
7244     assert(toReg->is_valid(), "destination must be valid");
7245     int idx = (int)toReg->value();
7246     if (reg_state[idx] == reg_readonly) {
7247       if (idx != from->value()) {
7248         mark_done = false;
7249       }
7250       done = false;
7251       continue;
7252     } else if (reg_state[idx] == reg_written) {
7253       continue;
7254     }
7255     assert(reg_state[idx] == reg_writable, "must be writable");
7256     reg_state[idx] = reg_written;
7257     progress = true;
7258 
7259     if (fromReg == noreg) {
7260       if (from->is_reg()) {
7261         fromReg = from->as_Register();
7262       } else {
7263         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7264         ldr(tmp1, Address(sp, st_off));
7265         fromReg = tmp1;
7266       }
7267       if (null_check) {
7268         // Nullable inline type argument, emit null check
7269         cbz(fromReg, L_null);
7270       }
7271     }
7272     int off = sig->at(stream.sig_index())._offset;
7273     if (off == -1) {
7274       assert(null_check, "Missing null check at");
7275       if (toReg->is_stack()) {
7276         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7277         mov(tmp2, 1);
7278         str(tmp2, Address(sp, st_off));
7279       } else {
7280         mov(toReg->as_Register(), 1);
7281       }
7282       continue;
7283     }
7284     if (sig->at(stream.sig_index())._vt_oop) {
7285       if (toReg->is_stack()) {
7286         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7287         str(fromReg, Address(sp, st_off));
7288       } else {
7289         mov(toReg->as_Register(), fromReg);
7290       }
7291       continue;
7292     }
7293     assert(off > 0, "offset in object should be positive");
7294     Address fromAddr = Address(fromReg, off);
7295     if (!toReg->is_FloatRegister()) {
7296       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7297       if (is_reference_type(bt)) {
7298         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7299       } else {
7300         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7301         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7302       }
7303       if (toReg->is_stack()) {
7304         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7305         str(dst, Address(sp, st_off));
7306       }
7307     } else if (bt == T_DOUBLE) {
7308       ldrd(toReg->as_FloatRegister(), fromAddr);
7309     } else {
7310       assert(bt == T_FLOAT, "must be float");
7311       ldrs(toReg->as_FloatRegister(), fromAddr);
7312     }
7313   }
7314   if (progress && null_check) {
7315     if (done) {
7316       b(L_notNull);
7317       bind(L_null);
7318       // Set null marker to zero to signal that the argument is null.
7319       // Also set all fields to zero since the runtime requires a canonical
7320       // representation of a flat null.
7321       stream.reset(sig_index, to_index);
7322       while (stream.next(toReg, bt)) {
7323         if (toReg->is_stack()) {
7324           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7325           str(zr, Address(sp, st_off));
7326         } else if (toReg->is_FloatRegister()) {
7327           mov(toReg->as_FloatRegister(), T2S, 0);
7328         } else {
7329           mov(toReg->as_Register(), zr);
7330         }
7331       }
7332       bind(L_notNull);
7333     } else {
7334       bind(L_null);
7335     }
7336   }
7337 
7338   sig_index = stream.sig_index();
7339   to_index = stream.regs_index();
7340 
7341   if (mark_done && reg_state[from->value()] != reg_written) {
7342     // This is okay because no one else will write to that slot
7343     reg_state[from->value()] = reg_writable;
7344   }
7345   from_index--;
7346   assert(progress || (start_offset == offset()), "should not emit code");
7347   return done;
7348 }
7349 
7350 // Pack fields back into an inline type oop
7351 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7352                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7353                                         RegState reg_state[], Register val_array) {
7354   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7355   assert(to->is_valid(), "destination must be valid");
7356 
7357   if (reg_state[to->value()] == reg_written) {
7358     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7359     return true; // Already written
7360   }
7361 
7362   // The GC barrier expanded by store_heap_oop below may call into the
7363   // runtime so use callee-saved registers for any values that need to be
7364   // preserved. The GC barrier assembler should take care of saving the
7365   // Java argument registers.
7366   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7367   Register val_obj_tmp = r21;
7368   Register from_reg_tmp = r22;
7369   Register tmp1 = r14;
7370   Register tmp2 = r13;
7371   Register tmp3 = r12;
7372   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7373 
7374   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7375 
7376   if (reg_state[to->value()] == reg_readonly) {
7377     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7378       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7379       return false; // Not yet writable
7380     }
7381     val_obj = val_obj_tmp;
7382   }
7383 
7384   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7385   VMReg fromReg;
7386   BasicType bt;
7387   Label L_null;
7388   while (stream.next(fromReg, bt)) {
7389     assert(fromReg->is_valid(), "source must be valid");
7390     reg_state[fromReg->value()] = reg_writable;
7391 
7392     int off = sig->at(stream.sig_index())._offset;
7393     if (off == -1) {
7394       // Nullable inline type argument, emit null check
7395       Label L_notNull;
7396       if (fromReg->is_stack()) {
7397         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7398         ldrb(tmp2, Address(sp, ld_off));
7399         cbnz(tmp2, L_notNull);
7400       } else {
7401         cbnz(fromReg->as_Register(), L_notNull);
7402       }
7403       mov(val_obj, 0);
7404       b(L_null);
7405       bind(L_notNull);
7406       continue;
7407     }
7408     if (sig->at(stream.sig_index())._vt_oop) {
7409       if (fromReg->is_stack()) {
7410         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7411         ldr(val_obj, Address(sp, ld_off));
7412       } else {
7413         mov(val_obj, fromReg->as_Register());
7414       }
7415       cbnz(val_obj, L_null);
7416       // get the buffer from the just allocated pool of buffers
7417       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7418       load_heap_oop(val_obj, Address(val_array, index), rscratch1, rscratch2);
7419       continue;
7420     }
7421 
7422     assert(off > 0, "offset in object should be positive");
7423     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7424 
7425     // Pack the scalarized field into the value object.
7426     Address dst(val_obj, off);
7427     if (!fromReg->is_FloatRegister()) {
7428       Register src;
7429       if (fromReg->is_stack()) {
7430         src = from_reg_tmp;
7431         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7432         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7433       } else {
7434         src = fromReg->as_Register();
7435       }
7436       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7437       if (is_reference_type(bt)) {
7438         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
7439         mov(tmp3, val_obj);
7440         Address dst_with_tmp3(tmp3, off);
7441         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7442       } else {
7443         store_sized_value(dst, src, size_in_bytes);
7444       }
7445     } else if (bt == T_DOUBLE) {
7446       strd(fromReg->as_FloatRegister(), dst);
7447     } else {
7448       assert(bt == T_FLOAT, "must be float");
7449       strs(fromReg->as_FloatRegister(), dst);
7450     }
7451   }
7452   bind(L_null);
7453   sig_index = stream.sig_index();
7454   from_index = stream.regs_index();
7455 
7456   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7457   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7458   assert(success, "to register must be writeable");
7459   return true;
7460 }
7461 
7462 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7463   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7464 }
7465 
7466 void MacroAssembler::cache_wb(Address line) {
7467   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7468   assert(line.index() == noreg, "index should be noreg");
7469   assert(line.offset() == 0, "offset should be 0");
7470   // would like to assert this
7471   // assert(line._ext.shift == 0, "shift should be zero");
7472   if (VM_Version::supports_dcpop()) {
7473     // writeback using clear virtual address to point of persistence
7474     dc(Assembler::CVAP, line.base());
7475   } else {
7476     // no need to generate anything as Unsafe.writebackMemory should
7477     // never invoke this stub
7478   }
7479 }
7480 
7481 void MacroAssembler::cache_wbsync(bool is_pre) {
7482   // we only need a barrier post sync
7483   if (!is_pre) {
7484     membar(Assembler::AnyAny);
7485   }

7881   }
7882 
7883   // Check if the lock-stack is full.
7884   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7885   cmpw(top, (unsigned)LockStack::end_offset());
7886   br(Assembler::GE, slow);
7887 
7888   // Check for recursion.
7889   subw(t, top, oopSize);
7890   ldr(t, Address(rthread, t));
7891   cmp(obj, t);
7892   br(Assembler::EQ, push);
7893 
7894   // Check header for monitor (0b10).
7895   tst(mark, markWord::monitor_value);
7896   br(Assembler::NE, slow);
7897 
7898   // Try to lock. Transition lock bits 0b01 => 0b00
7899   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7900   orr(mark, mark, markWord::unlocked_value);
7901   // Mask inline_type bit such that we go to the slow path if object is an inline type
7902   andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
7903 
7904   eor(t, mark, markWord::unlocked_value);
7905   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7906           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7907   br(Assembler::NE, slow);
7908 
7909   bind(push);
7910   // After successful lock, push object on lock-stack.
7911   str(obj, Address(rthread, top));
7912   addw(top, top, oopSize);
7913   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7914 }
7915 
7916 // Implements fast-unlocking.
7917 //
7918 // - obj: the object to be unlocked
7919 // - t1, t2, t3: temporary registers
7920 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7921 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7922   // cmpxchg clobbers rscratch1.
7923   assert_different_registers(obj, t1, t2, t3, rscratch1);
< prev index next >