< 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 

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   }

5070   adrp(rscratch1, src2, offset);
5071   ldr(rscratch1, Address(rscratch1, offset));
5072   cmp(src1, rscratch1);
5073 }
5074 
5075 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5076   cmp(obj1, obj2);
5077 }
5078 
5079 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5080   load_method_holder(rresult, rmethod);
5081   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5082 }
5083 
5084 void MacroAssembler::load_method_holder(Register holder, Register method) {
5085   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5086   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5087   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5088 }
5089 








5090 // Loads the obj's Klass* into dst.
5091 // Preserves all registers (incl src, rscratch1 and rscratch2).
5092 // Input:
5093 // src - the oop we want to load the klass from.
5094 // dst - output narrow klass.
5095 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5096   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5097   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5098   lsr(dst, dst, markWord::klass_shift);
5099 }
5100 
5101 void MacroAssembler::load_klass(Register dst, Register src) {
5102   if (UseCompactObjectHeaders) {
5103     load_narrow_klass_compact(dst, src);
5104   } else {
5105     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5106   }
5107   decode_klass_not_null(dst);
5108 }
5109 

5161   } else {
5162     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5163   }
5164   if (CompressedKlassPointers::base() == nullptr) {
5165     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5166     return;
5167   } else if (!AOTCodeCache::is_on_for_dump() &&
5168              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5169              && CompressedKlassPointers::shift() == 0) {
5170     // Only the bottom 32 bits matter
5171     cmpw(klass, tmp);
5172     return;
5173   }
5174   decode_klass_not_null(tmp);
5175   cmp(klass, tmp);
5176 }
5177 
5178 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5179   if (UseCompactObjectHeaders) {
5180     load_narrow_klass_compact(tmp1, obj1);
5181     load_narrow_klass_compact(tmp2,  obj2);
5182   } else {
5183     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5184     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5185   }
5186   cmpw(tmp1, tmp2);
5187 }
5188 





5189 void MacroAssembler::store_klass(Register dst, Register src) {
5190   // FIXME: Should this be a store release?  concurrent gcs assumes
5191   // klass length is valid if klass field is not null.
5192   assert(!UseCompactObjectHeaders, "not with compact headers");
5193   encode_klass_not_null(src);
5194   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5195 }
5196 
5197 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5198   assert(!UseCompactObjectHeaders, "not with compact headers");
5199   // Store to klass gap in destination
5200   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5201 }
5202 
5203 // Algorithm must match CompressedOops::encode.
5204 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5205 #ifdef ASSERT
5206   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5207 #endif
5208   verify_oop_msg(s, "broken oop in encode_heap_oop");

5558   if (as_raw) {
5559     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5560   } else {
5561     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5562   }
5563 }
5564 
5565 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5566                                      Address dst, Register val,
5567                                      Register tmp1, Register tmp2, Register tmp3) {
5568   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5569   decorators = AccessInternal::decorator_fixup(decorators, type);
5570   bool as_raw = (decorators & AS_RAW) != 0;
5571   if (as_raw) {
5572     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5573   } else {
5574     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5575   }
5576 }
5577 








































5578 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5579                                    Register tmp2, DecoratorSet decorators) {
5580   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5581 }
5582 
5583 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5584                                             Register tmp2, DecoratorSet decorators) {
5585   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5586 }
5587 
5588 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5589                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5590   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5591 }
5592 
5593 // Used for storing nulls.
5594 void MacroAssembler::store_heap_oop_null(Address dst) {
5595   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5596 }
5597 

5633     oop_index = oop_recorder()->allocate_metadata_index(obj);
5634   } else {
5635     oop_index = oop_recorder()->find_index(obj);
5636   }
5637   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5638   mov(dst, Address((address)obj, rspec));
5639 }
5640 
5641 Address MacroAssembler::constant_oop_address(jobject obj) {
5642 #ifdef ASSERT
5643   {
5644     ThreadInVMfromUnknown tiv;
5645     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5646     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5647   }
5648 #endif
5649   int oop_index = oop_recorder()->find_index(obj);
5650   return Address((address)obj, oop_Relocation::spec(oop_index));
5651 }
5652 






































































































5653 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5654 void MacroAssembler::tlab_allocate(Register obj,
5655                                    Register var_size_in_bytes,
5656                                    int con_size_in_bytes,
5657                                    Register t1,
5658                                    Register t2,
5659                                    Label& slow_case) {
5660   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5661   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5662 }
5663 
5664 void MacroAssembler::verify_tlab() {
5665 #ifdef ASSERT
5666   if (UseTLAB && VerifyOops) {
5667     Label next, ok;
5668 
5669     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5670 
5671     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5672     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5673     cmp(rscratch2, rscratch1);
5674     br(Assembler::HS, next);
5675     STOP("assert(top >= start)");
5676     should_not_reach_here();
5677 
5678     bind(next);
5679     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5680     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5681     cmp(rscratch2, rscratch1);
5682     br(Assembler::HS, ok);
5683     STOP("assert(top <= end)");
5684     should_not_reach_here();
5685 
5686     bind(ok);
5687     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5688   }
5689 #endif
5690 }
5691 















5692 // Writes to stack successive pages until offset reached to check for
5693 // stack overflow + shadow pages.  This clobbers tmp.
5694 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5695   assert_different_registers(tmp, size, rscratch1);
5696   mov(tmp, sp);
5697   // Bang stack for total size given plus shadow page size.
5698   // Bang one page at a time because large size can bang beyond yellow and
5699   // red zones.
5700   Label loop;
5701   mov(rscratch1, (int)os::vm_page_size());
5702   bind(loop);
5703   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5704   subsw(size, size, rscratch1);
5705   str(size, Address(tmp));
5706   br(Assembler::GT, loop);
5707 
5708   // Bang down shadow pages too.
5709   // At this point, (tmp-0) is the last address touched, so don't
5710   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5711   // was post-decremented.)  Skip this address by starting at i=1, and

5778 
5779   // Strictly speaking the card table base isn't an address at all, and it might
5780   // even be negative. It is thus materialised as a constant.
5781   mov(reg, (uint64_t)ctbs->card_table_base_const());
5782 }
5783 
5784 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5785 #if INCLUDE_CDS
5786   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5787   if (AOTCodeCache::is_on_for_dump()) {
5788     // all aotrc field addresses should be registered in the AOTCodeCache address table
5789     lea(reg, ExternalAddress(a));
5790   } else {
5791     mov(reg, (uint64_t)a);
5792   }
5793 #else
5794   ShouldNotReachHere();
5795 #endif
5796 }
5797 

5798 void MacroAssembler::build_frame(int framesize) {





5799   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5800   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5801   protect_return_address();
5802   if (framesize < ((1 << 9) + 2 * wordSize)) {
5803     sub(sp, sp, framesize);
5804     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));





5805     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5806   } else {
5807     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));





5808     if (PreserveFramePointer) mov(rfp, sp);
5809     if (framesize < ((1 << 12) + 2 * wordSize))
5810       sub(sp, sp, framesize - 2 * wordSize);
5811     else {
5812       mov(rscratch1, framesize - 2 * wordSize);
5813       sub(sp, sp, rscratch1);
5814     }
5815   }
5816   verify_cross_modify_fence_not_required();
5817 }
5818 
5819 void MacroAssembler::remove_frame(int framesize) {
5820   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5821   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5822   if (framesize < ((1 << 9) + 2 * wordSize)) {
5823     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5824     add(sp, sp, framesize);
5825   } else {
5826     if (framesize < ((1 << 12) + 2 * wordSize))
5827       add(sp, sp, framesize - 2 * wordSize);
5828     else {
5829       mov(rscratch1, framesize - 2 * wordSize);
5830       add(sp, sp, rscratch1);
5831     }
5832     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5833   }
5834   authenticate_return_address();
5835 }
5836 


















































































5837 
5838 // This method counts leading positive bytes (highest bit not set) in provided byte array
5839 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5840     // Simple and most common case of aligned small array which is not at the
5841     // end of memory page is placed here. All other cases are in stub.
5842     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5843     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5844     assert_different_registers(ary1, len, result);
5845 
5846     mov(result, len);
5847     cmpw(len, 0);
5848     br(LE, DONE);
5849     cmpw(len, 4 * wordSize);
5850     br(GE, STUB_LONG); // size > 32 then go to stub
5851 
5852     int shift = 64 - exact_log2(os::vm_page_size());
5853     lsl(rscratch1, ary1, shift);
5854     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5855     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5856     br(CS, STUB); // at the end of page then go to stub

6739 // On other systems, the helper is a usual C function.
6740 //
6741 void MacroAssembler::get_thread(Register dst) {
6742   RegSet saved_regs =
6743     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6744     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6745 
6746   protect_return_address();
6747   push(saved_regs, sp);
6748 
6749   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6750   blr(lr);
6751   if (dst != c_rarg0) {
6752     mov(dst, c_rarg0);
6753   }
6754 
6755   pop(saved_regs, sp);
6756   authenticate_return_address();
6757 }
6758 



























































































































































































































































































































































































































































































6759 void MacroAssembler::cache_wb(Address line) {
6760   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6761   assert(line.index() == noreg, "index should be noreg");
6762   assert(line.offset() == 0, "offset should be 0");
6763   // would like to assert this
6764   // assert(line._ext.shift == 0, "shift should be zero");
6765   if (VM_Version::supports_dcpop()) {
6766     // writeback using clear virtual address to point of persistence
6767     dc(Assembler::CVAP, line.base());
6768   } else {
6769     // no need to generate anything as Unsafe.writebackMemory should
6770     // never invoke this stub
6771   }
6772 }
6773 
6774 void MacroAssembler::cache_wbsync(bool is_pre) {
6775   // we only need a barrier post sync
6776   if (!is_pre) {
6777     membar(Assembler::AnyAny);
6778   }

7174   }
7175 
7176   // Check if the lock-stack is full.
7177   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7178   cmpw(top, (unsigned)LockStack::end_offset());
7179   br(Assembler::GE, slow);
7180 
7181   // Check for recursion.
7182   subw(t, top, oopSize);
7183   ldr(t, Address(rthread, t));
7184   cmp(obj, t);
7185   br(Assembler::EQ, push);
7186 
7187   // Check header for monitor (0b10).
7188   tst(mark, markWord::monitor_value);
7189   br(Assembler::NE, slow);
7190 
7191   // Try to lock. Transition lock bits 0b01 => 0b00
7192   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7193   orr(mark, mark, markWord::unlocked_value);



7194   eor(t, mark, markWord::unlocked_value);
7195   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7196           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7197   br(Assembler::NE, slow);
7198 
7199   bind(push);
7200   // After successful lock, push object on lock-stack.
7201   str(obj, Address(rthread, top));
7202   addw(top, top, oopSize);
7203   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7204 }
7205 
7206 // Implements fast-unlocking.
7207 //
7208 // - obj: the object to be unlocked
7209 // - t1, t2, t3: temporary registers
7210 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7211 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7212   // cmpxchg clobbers rscratch1.
7213   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 

2012   ldarb(scratch, scratch);
2013   cmp(scratch, InstanceKlass::fully_initialized);
2014   br(Assembler::EQ, *L_fast_path);
2015 
2016   // Fast path check: current thread is initializer thread
2017   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2018   cmp(rthread, scratch);
2019 
2020   if (L_slow_path == &L_fallthrough) {
2021     br(Assembler::EQ, *L_fast_path);
2022     bind(*L_slow_path);
2023   } else if (L_fast_path == &L_fallthrough) {
2024     br(Assembler::NE, *L_slow_path);
2025     bind(*L_fast_path);
2026   } else {
2027     Unimplemented();
2028   }
2029 }
2030 
2031 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2032   if (!VerifyOops || VerifyAdapterSharing) {
2033     // Below address of the code string confuses VerifyAdapterSharing
2034     // because it may differ between otherwise equivalent adapters.
2035     return;
2036   }
2037 
2038   // Pass register number to verify_oop_subroutine
2039   const char* b = nullptr;
2040   {
2041     ResourceMark rm;
2042     stringStream ss;
2043     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2044     b = code_string(ss.as_string());
2045   }
2046   BLOCK_COMMENT("verify_oop {");
2047 
2048   strip_return_address(); // This might happen within a stack frame.
2049   protect_return_address();
2050   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2051   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2052 
2053   mov(r0, reg);
2054   movptr(rscratch1, (uintptr_t)(address)b);
2055 
2056   // call indirectly to solve generation ordering problem
2057   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2058   ldr(rscratch2, Address(rscratch2));
2059   blr(rscratch2);
2060 
2061   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2062   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2063   authenticate_return_address();
2064 
2065   BLOCK_COMMENT("} verify_oop");
2066 }
2067 
2068 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2069   if (!VerifyOops || VerifyAdapterSharing) {
2070     // Below address of the code string confuses VerifyAdapterSharing
2071     // because it may differ between otherwise equivalent adapters.
2072     return;
2073   }
2074 
2075   const char* b = nullptr;
2076   {
2077     ResourceMark rm;
2078     stringStream ss;
2079     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2080     b = code_string(ss.as_string());
2081   }
2082   BLOCK_COMMENT("verify_oop_addr {");
2083 
2084   strip_return_address(); // This might happen within a stack frame.
2085   protect_return_address();
2086   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2087   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2088 
2089   // addr may contain sp so we will have to adjust it based on the
2090   // pushes that we just did.
2091   if (addr.uses(sp)) {
2092     lea(r0, addr);
2093     ldr(r0, Address(r0, 4 * wordSize));

2306   call_VM_leaf_base(entry_point, 1);
2307 }
2308 
2309 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2310   assert_different_registers(arg_1, c_rarg0);
2311   pass_arg0(this, arg_0);
2312   pass_arg1(this, arg_1);
2313   call_VM_leaf_base(entry_point, 2);
2314 }
2315 
2316 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2317                                   Register arg_1, Register arg_2) {
2318   assert_different_registers(arg_1, c_rarg0);
2319   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2320   pass_arg0(this, arg_0);
2321   pass_arg1(this, arg_1);
2322   pass_arg2(this, arg_2);
2323   call_VM_leaf_base(entry_point, 3);
2324 }
2325 
2326 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2327   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2328 }
2329 
2330 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2331   pass_arg0(this, arg_0);
2332   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2333 }
2334 
2335 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2336 
2337   assert_different_registers(arg_0, c_rarg1);
2338   pass_arg1(this, arg_1);
2339   pass_arg0(this, arg_0);
2340   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2341 }
2342 
2343 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2344   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2345   assert_different_registers(arg_1, c_rarg2);
2346   pass_arg2(this, arg_2);
2347   pass_arg1(this, arg_1);
2348   pass_arg0(this, arg_0);
2349   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2355   assert_different_registers(arg_2, c_rarg3);
2356   pass_arg3(this, arg_3);
2357   pass_arg2(this, arg_2);
2358   pass_arg1(this, arg_1);
2359   pass_arg0(this, arg_0);
2360   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2361 }
2362 
2363 void MacroAssembler::null_check(Register reg, int offset) {
2364   if (needs_explicit_null_check(offset)) {
2365     // provoke OS null exception if reg is null by
2366     // accessing M[reg] w/o changing any registers
2367     // NOTE: this is plenty to provoke a segv
2368     ldr(zr, Address(reg));
2369   } else {
2370     // nothing to do, (later) access of M[reg + offset]
2371     // will provoke OS null exception if reg is null
2372   }
2373 }
2374 
2375 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2376   assert_different_registers(markword, rscratch2);
2377   mov(rscratch2, markWord::inline_type_pattern_mask);
2378   andr(markword, markword, rscratch2);
2379   mov(rscratch2, markWord::inline_type_pattern);
2380   cmp(markword, rscratch2);
2381   br(Assembler::EQ, is_inline_type);
2382 }
2383 
2384 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
2385   assert_different_registers(tmp, rscratch1);
2386   if (can_be_null) {
2387     cbz(object, not_inline_type);
2388   }
2389   const int is_inline_type_mask = markWord::inline_type_pattern;
2390   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2391   mov(rscratch1, is_inline_type_mask);
2392   andr(tmp, tmp, rscratch1);
2393   cmp(tmp, rscratch1);
2394   br(Assembler::NE, not_inline_type);
2395 }
2396 
2397 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2398   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2399   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2400 }
2401 
2402 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2403   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2404   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2405 }
2406 
2407 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2408   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2409   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2410 }
2411 
2412 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
2413   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2414   tbnz(flags, ResolvedFieldEntry::has_null_marker_shift, has_null_marker);
2415 }
2416 
2417 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2418   Label test_mark_word;
2419   // load mark word
2420   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2421   // check displaced
2422   tst(temp_reg, markWord::unlocked_value);
2423   br(Assembler::NE, test_mark_word);
2424   // slow path use klass prototype
2425   load_prototype_header(temp_reg, oop);
2426 
2427   bind(test_mark_word);
2428   andr(temp_reg, temp_reg, test_bit);
2429   if (jmp_set) {
2430     cbnz(temp_reg, jmp_label);
2431   } else {
2432     cbz(temp_reg, jmp_label);
2433   }
2434 }
2435 
2436 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2437   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2438 }
2439 
2440 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2441                                                   Label&is_non_flat_array) {
2442   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2443 }
2444 
2445 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2446   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2447 }
2448 
2449 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2450   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2451 }
2452 
2453 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2454   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2455   br(Assembler::NE, is_flat_array);
2456 }
2457 
2458 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
2459   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2460   br(Assembler::EQ, is_non_flat_array);
2461 }
2462 
2463 // MacroAssembler protected routines needed to implement
2464 // public methods
2465 
2466 void MacroAssembler::mov(Register r, Address dest) {
2467   code_section()->relocate(pc(), dest.rspec());
2468   uint64_t imm64 = (uint64_t)dest.target();
2469   movptr(r, imm64);
2470 }
2471 
2472 // Move a constant pointer into r.  In AArch64 mode the virtual
2473 // address space is 48 bits in size, so we only need three
2474 // instructions to create a patchable instruction sequence that can
2475 // reach anywhere.
2476 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2477 #ifndef PRODUCT
2478   {
2479     char buffer[64];
2480     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2481     block_comment(buffer);
2482   }

5176   adrp(rscratch1, src2, offset);
5177   ldr(rscratch1, Address(rscratch1, offset));
5178   cmp(src1, rscratch1);
5179 }
5180 
5181 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5182   cmp(obj1, obj2);
5183 }
5184 
5185 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5186   load_method_holder(rresult, rmethod);
5187   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5188 }
5189 
5190 void MacroAssembler::load_method_holder(Register holder, Register method) {
5191   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5192   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5193   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5194 }
5195 
5196 void MacroAssembler::load_metadata(Register dst, Register src) {
5197   if (UseCompactObjectHeaders) {
5198     load_narrow_klass_compact(dst, src);
5199   } else {
5200     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5201   }
5202 }
5203 
5204 // Loads the obj's Klass* into dst.
5205 // Preserves all registers (incl src, rscratch1 and rscratch2).
5206 // Input:
5207 // src - the oop we want to load the klass from.
5208 // dst - output narrow klass.
5209 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5210   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5211   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5212   lsr(dst, dst, markWord::klass_shift);
5213 }
5214 
5215 void MacroAssembler::load_klass(Register dst, Register src) {
5216   if (UseCompactObjectHeaders) {
5217     load_narrow_klass_compact(dst, src);
5218   } else {
5219     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5220   }
5221   decode_klass_not_null(dst);
5222 }
5223 

5275   } else {
5276     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5277   }
5278   if (CompressedKlassPointers::base() == nullptr) {
5279     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5280     return;
5281   } else if (!AOTCodeCache::is_on_for_dump() &&
5282              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5283              && CompressedKlassPointers::shift() == 0) {
5284     // Only the bottom 32 bits matter
5285     cmpw(klass, tmp);
5286     return;
5287   }
5288   decode_klass_not_null(tmp);
5289   cmp(klass, tmp);
5290 }
5291 
5292 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5293   if (UseCompactObjectHeaders) {
5294     load_narrow_klass_compact(tmp1, obj1);
5295     load_narrow_klass_compact(tmp2, obj2);
5296   } else {
5297     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5298     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5299   }
5300   cmpw(tmp1, tmp2);
5301 }
5302 
5303 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5304   load_klass(dst, src);
5305   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5306 }
5307 
5308 void MacroAssembler::store_klass(Register dst, Register src) {
5309   // FIXME: Should this be a store release?  concurrent gcs assumes
5310   // klass length is valid if klass field is not null.
5311   assert(!UseCompactObjectHeaders, "not with compact headers");
5312   encode_klass_not_null(src);
5313   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5314 }
5315 
5316 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5317   assert(!UseCompactObjectHeaders, "not with compact headers");
5318   // Store to klass gap in destination
5319   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5320 }
5321 
5322 // Algorithm must match CompressedOops::encode.
5323 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5324 #ifdef ASSERT
5325   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5326 #endif
5327   verify_oop_msg(s, "broken oop in encode_heap_oop");

5677   if (as_raw) {
5678     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5679   } else {
5680     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5681   }
5682 }
5683 
5684 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5685                                      Address dst, Register val,
5686                                      Register tmp1, Register tmp2, Register tmp3) {
5687   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5688   decorators = AccessInternal::decorator_fixup(decorators, type);
5689   bool as_raw = (decorators & AS_RAW) != 0;
5690   if (as_raw) {
5691     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5692   } else {
5693     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5694   }
5695 }
5696 
5697 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5698                                      Register inline_layout_info) {
5699   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5700   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5701 }
5702 
5703 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5704   ldr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
5705   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5706 }
5707 
5708 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5709   // ((address) (void*) o) + vk->payload_offset();
5710   Register offset = (data == oop) ? rscratch1 : data;
5711   payload_offset(inline_klass, offset);
5712   if (data == oop) {
5713     add(data, data, offset);
5714   } else {
5715     lea(data, Address(oop, offset));
5716   }
5717 }
5718 
5719 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
5720                                                 Register index, Register data) {
5721   assert_different_registers(array, array_klass, index);
5722   assert_different_registers(rscratch1, array, index);
5723 
5724   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
5725   ldrw(rscratch1, Address(array_klass, Klass::layout_helper_offset()));
5726 
5727   // Klass::layout_helper_log2_element_size(lh)
5728   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
5729   lsr(rscratch1, rscratch1, Klass::_lh_log2_element_size_shift);
5730   andr(rscratch1, rscratch1, Klass::_lh_log2_element_size_mask);
5731   lslv(index, index, rscratch1);
5732 
5733   add(data, array, index);
5734   add(data, data, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT));
5735 }
5736 
5737 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5738                                    Register tmp2, DecoratorSet decorators) {
5739   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5740 }
5741 
5742 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5743                                             Register tmp2, DecoratorSet decorators) {
5744   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5745 }
5746 
5747 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5748                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5749   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5750 }
5751 
5752 // Used for storing nulls.
5753 void MacroAssembler::store_heap_oop_null(Address dst) {
5754   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5755 }
5756 

5792     oop_index = oop_recorder()->allocate_metadata_index(obj);
5793   } else {
5794     oop_index = oop_recorder()->find_index(obj);
5795   }
5796   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5797   mov(dst, Address((address)obj, rspec));
5798 }
5799 
5800 Address MacroAssembler::constant_oop_address(jobject obj) {
5801 #ifdef ASSERT
5802   {
5803     ThreadInVMfromUnknown tiv;
5804     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5805     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5806   }
5807 #endif
5808   int oop_index = oop_recorder()->find_index(obj);
5809   return Address((address)obj, oop_Relocation::spec(oop_index));
5810 }
5811 
5812 // Object / value buffer allocation...
5813 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
5814                                        Register t1, Register t2,
5815                                        bool clear_fields, Label& alloc_failed)
5816 {
5817   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
5818   Register layout_size = t1;
5819   assert(new_obj == r0, "needs to be r0");
5820   assert_different_registers(klass, new_obj, t1, t2);
5821 
5822   // get instance_size in InstanceKlass (scaled to a count of bytes)
5823   ldrw(layout_size, Address(klass, Klass::layout_helper_offset()));
5824   // test to see if it is malformed in some way
5825   tst(layout_size, Klass::_lh_instance_slow_path_bit);
5826   br(Assembler::NE, slow_case_no_pop);
5827 
5828   // Allocate the instance:
5829   //  If TLAB is enabled:
5830   //    Try to allocate in the TLAB.
5831   //    If fails, go to the slow path.
5832   //    Initialize the allocation.
5833   //    Exit.
5834   //
5835   //  Go to slow path.
5836 
5837   if (UseTLAB) {
5838     push(klass);
5839     tlab_allocate(new_obj, layout_size, 0, klass, t2, slow_case);
5840     if (ZeroTLAB || (!clear_fields)) {
5841       // the fields have been already cleared
5842       b(initialize_header);
5843     } else {
5844       // initialize both the header and fields
5845       b(initialize_object);
5846     }
5847 
5848     if (clear_fields) {
5849       // The object is initialized before the header.  If the object size is
5850       // zero, go directly to the header initialization.
5851       bind(initialize_object);
5852       int header_size = oopDesc::header_size() * HeapWordSize;
5853       assert(is_aligned(header_size, BytesPerLong), "oop header size must be 8-byte-aligned");
5854       subs(layout_size, layout_size, header_size);
5855       br(Assembler::EQ, initialize_header);
5856 
5857       // Initialize topmost object field, divide size by 8, check if odd and
5858       // test if zero.
5859 
5860   #ifdef ASSERT
5861       // make sure instance_size was multiple of 8
5862       Label L;
5863       tst(layout_size, 7);
5864       br(Assembler::EQ, L);
5865       stop("object size is not multiple of 8 - adjust this code");
5866       bind(L);
5867       // must be > 0, no extra check needed here
5868   #endif
5869 
5870       lsr(layout_size, layout_size, LogBytesPerLong);
5871 
5872       // initialize remaining object fields: instance_size was a multiple of 8
5873       {
5874         Label loop;
5875         Register base = t2;
5876 
5877         bind(loop);
5878         add(rscratch1, new_obj, layout_size, Assembler::LSL, LogBytesPerLong);
5879         str(zr, Address(rscratch1, header_size - 1*oopSize));
5880         subs(layout_size, layout_size, 1);
5881         br(Assembler::NE, loop);
5882       }
5883     } // clear_fields
5884 
5885     // initialize object header only.
5886     bind(initialize_header);
5887     pop(klass);
5888     Register mark_word = t2;
5889     if (UseCompactObjectHeaders || Arguments::is_valhalla_enabled()) {
5890       ldr(mark_word, Address(klass, Klass::prototype_header_offset()));
5891       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5892     } else {
5893       mov(mark_word, (intptr_t)markWord::prototype().value());
5894       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5895     }
5896     if (!UseCompactObjectHeaders) {
5897       store_klass_gap(new_obj, zr);  // zero klass gap for compressed oops
5898       mov(t2, klass);                // preserve klass
5899       store_klass(new_obj, t2);      // src klass reg is potentially compressed
5900     }
5901     b(done);
5902   }
5903 
5904   if (UseTLAB) {
5905     bind(slow_case);
5906     pop(klass);
5907   }
5908   bind(slow_case_no_pop);
5909   b(alloc_failed);
5910 
5911   bind(done);
5912 }
5913 
5914 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5915 void MacroAssembler::tlab_allocate(Register obj,
5916                                    Register var_size_in_bytes,
5917                                    int con_size_in_bytes,
5918                                    Register t1,
5919                                    Register t2,
5920                                    Label& slow_case) {
5921   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5922   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5923 }
5924 
5925 void MacroAssembler::verify_tlab() {
5926 #ifdef ASSERT
5927   if (UseTLAB && VerifyOops) {
5928     Label next, ok;
5929 
5930     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5931 
5932     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5933     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5934     cmp(rscratch2, rscratch1);
5935     br(Assembler::HS, next);
5936     STOP("assert(top >= start)");
5937     should_not_reach_here();
5938 
5939     bind(next);
5940     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5941     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5942     cmp(rscratch2, rscratch1);
5943     br(Assembler::HS, ok);
5944     STOP("assert(top <= end)");
5945     should_not_reach_here();
5946 
5947     bind(ok);
5948     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5949   }
5950 #endif
5951 }
5952 
5953 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5954   assert_different_registers(holder_klass, index, layout_info);
5955   InlineLayoutInfo array[2];
5956   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5957   if (is_power_of_2(size)) {
5958     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5959   } else {
5960     mov(layout_info, size);
5961     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5962   }
5963   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5964   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5965   lea(layout_info, Address(layout_info, index));
5966 }
5967 
5968 // Writes to stack successive pages until offset reached to check for
5969 // stack overflow + shadow pages.  This clobbers tmp.
5970 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5971   assert_different_registers(tmp, size, rscratch1);
5972   mov(tmp, sp);
5973   // Bang stack for total size given plus shadow page size.
5974   // Bang one page at a time because large size can bang beyond yellow and
5975   // red zones.
5976   Label loop;
5977   mov(rscratch1, (int)os::vm_page_size());
5978   bind(loop);
5979   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5980   subsw(size, size, rscratch1);
5981   str(size, Address(tmp));
5982   br(Assembler::GT, loop);
5983 
5984   // Bang down shadow pages too.
5985   // At this point, (tmp-0) is the last address touched, so don't
5986   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5987   // was post-decremented.)  Skip this address by starting at i=1, and

6054 
6055   // Strictly speaking the card table base isn't an address at all, and it might
6056   // even be negative. It is thus materialised as a constant.
6057   mov(reg, (uint64_t)ctbs->card_table_base_const());
6058 }
6059 
6060 void MacroAssembler::load_aotrc_address(Register reg, address a) {
6061 #if INCLUDE_CDS
6062   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
6063   if (AOTCodeCache::is_on_for_dump()) {
6064     // all aotrc field addresses should be registered in the AOTCodeCache address table
6065     lea(reg, ExternalAddress(a));
6066   } else {
6067     mov(reg, (uint64_t)a);
6068   }
6069 #else
6070   ShouldNotReachHere();
6071 #endif
6072 }
6073 
6074 #ifdef ASSERT
6075 void MacroAssembler::build_frame(int framesize) {
6076   build_frame(framesize, false);
6077 }
6078 #endif
6079 
6080 void MacroAssembler::build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills)) {
6081   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
6082   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
6083   protect_return_address();
6084   if (framesize < ((1 << 9) + 2 * wordSize)) {
6085     sub(sp, sp, framesize);
6086     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
6087       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
6088       stp(rscratch1, rscratch1, Address(sp, framesize - 2 * wordSize));
6089     } else {
6090       stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
6091     }
6092     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
6093   } else {
6094     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
6095       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
6096       stp(rscratch1, rscratch1, Address(pre(sp, -2 * wordSize)));
6097     } else {
6098       stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6099     }
6100     if (PreserveFramePointer) mov(rfp, sp);
6101     if (framesize < ((1 << 12) + 2 * wordSize))
6102       sub(sp, sp, framesize - 2 * wordSize);
6103     else {
6104       mov(rscratch1, framesize - 2 * wordSize);
6105       sub(sp, sp, rscratch1);
6106     }
6107   }
6108   verify_cross_modify_fence_not_required();
6109 }
6110 
6111 void MacroAssembler::remove_frame(int framesize) {
6112   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
6113   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
6114   if (framesize < ((1 << 9) + 2 * wordSize)) {
6115     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
6116     add(sp, sp, framesize);
6117   } else {
6118     if (framesize < ((1 << 12) + 2 * wordSize))
6119       add(sp, sp, framesize - 2 * wordSize);
6120     else {
6121       mov(rscratch1, framesize - 2 * wordSize);
6122       add(sp, sp, rscratch1);
6123     }
6124     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6125   }
6126   authenticate_return_address();
6127 }
6128 
6129 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
6130   if (needs_stack_repair) {
6131     // The method has a scalarized entry point (where fields of value object arguments
6132     // are passed through registers and stack), and a non-scalarized entry point (where
6133     // value object arguments are given as oops). The non-scalarized entry point will
6134     // first load each field of value object arguments and store them in registers and on
6135     // the stack in a way compatible with the scalarized entry point. To do so, some extra
6136     // stack space might be reserved (if argument registers are not enough). On leaving the
6137     // method, this space must be freed.
6138     //
6139     // In case we used the non-scalarized entry point the stack looks like this:
6140     //
6141     // | Arguments from caller     |
6142     // |---------------------------|  <-- caller's SP
6143     // | Saved LR #1               |
6144     // | Saved FP #1               |
6145     // |---------------------------|
6146     // | Extension space for       |
6147     // |   inline arg (un)packing  |
6148     // |---------------------------|  <-- start of this method's frame
6149     // | Saved LR #2               |
6150     // | Saved FP #2               |
6151     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6152     // | sp_inc                    |
6153     // | method locals             |
6154     // |---------------------------|  <-- SP
6155     //
6156     // There are two copies of FP and LR on the stack. They will be identical at
6157     // first, but that can change.
6158     // If the caller has been deoptimized, LR #1 will be patched to point at the
6159     // deopt blob, and LR #2 will still point into the old method.
6160     // If the saved FP (x29) was not used as the frame pointer, but to store an
6161     // oop, the GC will be aware only of FP #1 as the spilled location of x29 and
6162     // will fix only this one. Overall, FP/LR #2 are not reliable and are simply
6163     // needed to add space between the extension space and the locals, as there
6164     // would be between the real arguments and the locals if we don't need to
6165     // do unpacking (from the scalarized entry point).
6166     //
6167     // When restoring, one must then load FP #1 into x29, and LR #1 into x30,
6168     // while keeping in mind that from the scalarized entry point, there will be
6169     // only one copy of each. Indeed, in the case we used the scalarized calling
6170     // convention, the stack looks like this:
6171     //
6172     // | Arguments from caller     |
6173     // |---------------------------|  <-- caller's SP / start of this method's frame
6174     // | Saved LR                  |
6175     // | Saved FP                  |
6176     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6177     // | sp_inc                    |
6178     // | method locals             |
6179     // |---------------------------|  <-- SP
6180     //
6181     // The sp_inc stack slot holds the total size of the frame including the
6182     // extension space minus two words for the saved FP and LR. That is how to
6183     // find FP/LR #1. This size is expressed in bytes. Be careful when using it
6184     // from C++ in pointer arithmetic; you might need to divide it by wordSize.
6185     //
6186     // One can find sp_inc since the start the method's frame is SP + initial_framesize.
6187 
6188     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
6189 
6190     ldr(rscratch1, Address(sp, sp_inc_offset));
6191     add(sp, sp, rscratch1);
6192     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6193   } else {
6194     remove_frame(initial_framesize);
6195   }
6196 }
6197 
6198 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
6199   int real_frame_size = frame_size + sp_inc;
6200   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
6201   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
6202   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
6203 
6204   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
6205 
6206   // Subtract two words for the saved FP and LR as these will be popped
6207   // separately. See remove_frame above.
6208   mov(rscratch1, real_frame_size - 2*wordSize);
6209   str(rscratch1, Address(sp, sp_inc_offset));
6210 }
6211 
6212 // This method counts leading positive bytes (highest bit not set) in provided byte array
6213 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6214     // Simple and most common case of aligned small array which is not at the
6215     // end of memory page is placed here. All other cases are in stub.
6216     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6217     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6218     assert_different_registers(ary1, len, result);
6219 
6220     mov(result, len);
6221     cmpw(len, 0);
6222     br(LE, DONE);
6223     cmpw(len, 4 * wordSize);
6224     br(GE, STUB_LONG); // size > 32 then go to stub
6225 
6226     int shift = 64 - exact_log2(os::vm_page_size());
6227     lsl(rscratch1, ary1, shift);
6228     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6229     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6230     br(CS, STUB); // at the end of page then go to stub

7113 // On other systems, the helper is a usual C function.
7114 //
7115 void MacroAssembler::get_thread(Register dst) {
7116   RegSet saved_regs =
7117     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
7118     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
7119 
7120   protect_return_address();
7121   push(saved_regs, sp);
7122 
7123   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
7124   blr(lr);
7125   if (dst != c_rarg0) {
7126     mov(dst, c_rarg0);
7127   }
7128 
7129   pop(saved_regs, sp);
7130   authenticate_return_address();
7131 }
7132 
7133 #ifdef COMPILER2
7134 // C2 compiled method's prolog code
7135 // Moved here from aarch64.ad to support Valhalla code below
7136 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
7137   if (C->clinit_barrier_on_entry()) {
7138     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
7139 
7140     Label L_skip_barrier;
7141 
7142     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
7143     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
7144     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
7145     bind(L_skip_barrier);
7146   }
7147 
7148   if (C->max_vector_size() > 0) {
7149     reinitialize_ptrue();
7150   }
7151 
7152   int bangsize = C->output()->bang_size_in_bytes();
7153   if (C->output()->need_stack_bang(bangsize))
7154     generate_stack_overflow_check(bangsize);
7155 
7156   // n.b. frame size includes space for return pc and rfp
7157   const long framesize = C->output()->frame_size_in_bytes();
7158   build_frame(framesize DEBUG_ONLY(COMMA sp_inc != 0));
7159 
7160   if (C->needs_stack_repair()) {
7161     save_stack_increment(sp_inc, framesize);
7162   }
7163 
7164   if (VerifyStackAtCalls) {
7165     Unimplemented();
7166   }
7167 }
7168 #endif // COMPILER2
7169 
7170 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
7171   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
7172   // An inline type might be returned. If fields are in registers we
7173   // need to allocate an inline type instance and initialize it with
7174   // the value of the fields.
7175   Label skip;
7176   // We only need a new buffered inline type if a new one is not returned
7177   tbz(r0, 0, skip);
7178   int call_offset = -1;
7179 
7180   // Be careful not to clobber r1-7 which hold returned fields
7181   // Also do not use callee-saved registers as these may be live in the interpreter
7182   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
7183 
7184   // The following code is similar to allocate_instance but has some slight differences,
7185   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
7186   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
7187   Label slow_case;
7188   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
7189   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
7190 
7191   if (vk != nullptr) {
7192     // Called from C1, where the return type is statically known.
7193     movptr(klass, (intptr_t)vk->get_InlineKlass());
7194     jint lh = vk->layout_helper();
7195     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
7196     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
7197       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
7198     } else {
7199       b(slow_case);
7200     }
7201   } else {
7202     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
7203     andr(klass, r0, -2);
7204     if (UseTLAB) {
7205       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
7206       tst(tmp2, Klass::_lh_instance_slow_path_bit);
7207       br(Assembler::NE, slow_case);
7208       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
7209     } else {
7210       b(slow_case);
7211     }
7212   }
7213   if (UseTLAB) {
7214     // 2. Initialize buffered inline instance header
7215     Register buffer_obj = r0;
7216     if (UseCompactObjectHeaders) {
7217       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7218       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7219     } else {
7220       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7221       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7222       store_klass_gap(buffer_obj, zr);
7223       if (vk == nullptr) {
7224         // store_klass corrupts klass, so save it for later use (interpreter case only).
7225         mov(tmp1, klass);
7226       }
7227       store_klass(buffer_obj, klass);
7228       klass = tmp1;
7229     }
7230     // 3. Initialize its fields with an inline class specific handler
7231     if (vk != nullptr) {
7232       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7233     } else {
7234       ldr(tmp1, Address(klass, InlineKlass::adr_members_offset()));
7235       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7236       blr(tmp1);
7237     }
7238 
7239     membar(Assembler::StoreStore);
7240     b(skip);
7241   } else {
7242     // Must have already branched to slow_case above.
7243     DEBUG_ONLY(should_not_reach_here());
7244   }
7245   bind(slow_case);
7246   // We failed to allocate a new inline type, fall back to a runtime
7247   // call. Some oop field may be live in some registers but we can't
7248   // tell. That runtime call will take care of preserving them
7249   // across a GC if there's one.
7250   mov(r0, r0_preserved);
7251 
7252   if (from_interpreter) {
7253     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
7254   } else {
7255     far_call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
7256     call_offset = offset();
7257   }
7258   membar(Assembler::StoreStore);
7259 
7260   bind(skip);
7261   return call_offset;
7262 }
7263 
7264 // Move a value between registers/stack slots and update the reg_state
7265 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7266   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7267   if (reg_state[to->value()] == reg_written) {
7268     return true; // Already written
7269   }
7270 
7271   if (from != to && bt != T_VOID) {
7272     if (reg_state[to->value()] == reg_readonly) {
7273       return false; // Not yet writable
7274     }
7275     if (from->is_reg()) {
7276       if (to->is_reg()) {
7277         if (from->is_Register() && to->is_Register()) {
7278           mov(to->as_Register(), from->as_Register());
7279         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7280           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7281         } else {
7282           ShouldNotReachHere();
7283         }
7284       } else {
7285         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7286         Address to_addr = Address(sp, st_off);
7287         if (from->is_FloatRegister()) {
7288           if (bt == T_DOUBLE) {
7289              strd(from->as_FloatRegister(), to_addr);
7290           } else {
7291              assert(bt == T_FLOAT, "must be float");
7292              strs(from->as_FloatRegister(), to_addr);
7293           }
7294         } else {
7295           str(from->as_Register(), to_addr);
7296         }
7297       }
7298     } else {
7299       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7300       if (to->is_reg()) {
7301         if (to->is_FloatRegister()) {
7302           if (bt == T_DOUBLE) {
7303             ldrd(to->as_FloatRegister(), from_addr);
7304           } else {
7305             assert(bt == T_FLOAT, "must be float");
7306             ldrs(to->as_FloatRegister(), from_addr);
7307           }
7308         } else {
7309           ldr(to->as_Register(), from_addr);
7310         }
7311       } else {
7312         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7313         ldr(rscratch1, from_addr);
7314         str(rscratch1, Address(sp, st_off));
7315       }
7316     }
7317   }
7318 
7319   // Update register states
7320   reg_state[from->value()] = reg_writable;
7321   reg_state[to->value()] = reg_written;
7322   return true;
7323 }
7324 
7325 // Calculate the extra stack space required for packing or unpacking inline
7326 // args and adjust the stack pointer
7327 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7328   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7329   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7330   assert(sp_inc > 0, "sanity");
7331 
7332   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7333   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7334 
7335   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7336   if (sp_inc < (1 << 9)) {
7337     sub(sp, sp, sp_inc);   // Fits in an immediate
7338   } else {
7339     mov(rscratch1, sp_inc);
7340     sub(sp, sp, rscratch1);
7341   }
7342 
7343   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7344 }
7345 
7346 // Read all fields from an inline type oop and store the values in registers/stack slots
7347 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7348                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7349                                           RegState reg_state[]) {
7350   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7351   assert(from->is_valid(), "source must be valid");
7352   bool progress = false;
7353 #ifdef ASSERT
7354   const int start_offset = offset();
7355 #endif
7356 
7357   Label L_null, L_notNull;
7358   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7359   Register tmp1 = r10;
7360   Register tmp2 = r11;
7361 
7362 #ifndef ASSERT
7363   RegSet clobbered_gp_regs = MacroAssembler::call_clobbered_gp_registers();
7364   assert(clobbered_gp_regs.contains(tmp1), "tmp1 must be saved explicitly if it's not a clobber");
7365   assert(clobbered_gp_regs.contains(tmp2), "tmp2 must be saved explicitly if it's not a clobber");
7366   assert(clobbered_gp_regs.contains(r14), "r14 must be saved explicitly if it's not a clobber");
7367 #endif
7368 
7369   Register fromReg = noreg;
7370   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
7371   bool done = true;
7372   bool mark_done = true;
7373   VMReg toReg;
7374   BasicType bt;
7375   // Check if argument requires a null check
7376   bool null_check = false;
7377   VMReg nullCheckReg;
7378   while (stream.next(nullCheckReg, bt)) {
7379     if (sig->at(stream.sig_index())._offset == -1) {
7380       null_check = true;
7381       break;
7382     }
7383   }
7384   stream.reset(sig_index, to_index);
7385   while (stream.next(toReg, bt)) {
7386     assert(toReg->is_valid(), "destination must be valid");
7387     int idx = (int)toReg->value();
7388     if (reg_state[idx] == reg_readonly) {
7389       if (idx != from->value()) {
7390         mark_done = false;
7391       }
7392       done = false;
7393       continue;
7394     } else if (reg_state[idx] == reg_written) {
7395       continue;
7396     }
7397     assert(reg_state[idx] == reg_writable, "must be writable");
7398     reg_state[idx] = reg_written;
7399     progress = true;
7400 
7401     if (fromReg == noreg) {
7402       if (from->is_reg()) {
7403         fromReg = from->as_Register();
7404       } else {
7405         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7406         ldr(tmp1, Address(sp, st_off));
7407         fromReg = tmp1;
7408       }
7409       if (null_check) {
7410         // Nullable inline type argument, emit null check
7411         cbz(fromReg, L_null);
7412       }
7413     }
7414     int off = sig->at(stream.sig_index())._offset;
7415     if (off == -1) {
7416       assert(null_check, "Missing null check at");
7417       if (toReg->is_stack()) {
7418         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7419         mov(tmp2, 1);
7420         str(tmp2, Address(sp, st_off));
7421       } else {
7422         mov(toReg->as_Register(), 1);
7423       }
7424       continue;
7425     }
7426     if (sig->at(stream.sig_index())._vt_oop) {
7427       if (toReg->is_stack()) {
7428         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7429         str(fromReg, Address(sp, st_off));
7430       } else {
7431         mov(toReg->as_Register(), fromReg);
7432       }
7433       continue;
7434     }
7435     assert(off > 0, "offset in object should be positive");
7436     Address fromAddr = Address(fromReg, off);
7437     if (!toReg->is_FloatRegister()) {
7438       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7439       if (is_reference_type(bt)) {
7440         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7441       } else {
7442         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7443         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7444       }
7445       if (toReg->is_stack()) {
7446         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7447         str(dst, Address(sp, st_off));
7448       }
7449     } else if (bt == T_DOUBLE) {
7450       ldrd(toReg->as_FloatRegister(), fromAddr);
7451     } else {
7452       assert(bt == T_FLOAT, "must be float");
7453       ldrs(toReg->as_FloatRegister(), fromAddr);
7454     }
7455   }
7456   if (progress && null_check) {
7457     if (done) {
7458       b(L_notNull);
7459       bind(L_null);
7460       // Set null marker to zero to signal that the argument is null.
7461       // Also set all fields to zero since the runtime requires a canonical
7462       // representation of a flat null.
7463       stream.reset(sig_index, to_index);
7464       while (stream.next(toReg, bt)) {
7465         if (toReg->is_stack()) {
7466           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7467           str(zr, Address(sp, st_off));
7468         } else if (toReg->is_FloatRegister()) {
7469           mov(toReg->as_FloatRegister(), T2S, 0);
7470         } else {
7471           mov(toReg->as_Register(), zr);
7472         }
7473       }
7474       bind(L_notNull);
7475     } else {
7476       bind(L_null);
7477     }
7478   }
7479 
7480   sig_index = stream.sig_index();
7481   to_index = stream.regs_index();
7482 
7483   if (mark_done && reg_state[from->value()] != reg_written) {
7484     // This is okay because no one else will write to that slot
7485     reg_state[from->value()] = reg_writable;
7486   }
7487   from_index--;
7488   assert(progress || (start_offset == offset()), "should not emit code");
7489   return done;
7490 }
7491 
7492 // Pack fields back into an inline type oop
7493 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7494                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7495                                         RegState reg_state[], Register val_array) {
7496   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7497   assert(to->is_valid(), "destination must be valid");
7498 
7499   if (reg_state[to->value()] == reg_written) {
7500     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7501     return true; // Already written
7502   }
7503 
7504   // The GC barrier expanded by store_heap_oop below may call into the
7505   // runtime so use callee-saved registers for any values that need to be
7506   // preserved. The GC barrier assembler should take care of saving the
7507   // Java argument registers.
7508   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7509   Register val_obj_tmp = r21;
7510   Register from_reg_tmp = r22;
7511   Register tmp1 = r14;
7512   Register tmp2 = r13;
7513   Register tmp3 = r12;
7514   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7515 
7516   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7517 
7518   if (reg_state[to->value()] == reg_readonly) {
7519     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7520       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7521       return false; // Not yet writable
7522     }
7523     val_obj = val_obj_tmp;
7524   }
7525 
7526   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7527   VMReg fromReg;
7528   BasicType bt;
7529   Label L_null;
7530   while (stream.next(fromReg, bt)) {
7531     assert(fromReg->is_valid(), "source must be valid");
7532     reg_state[fromReg->value()] = reg_writable;
7533 
7534     int off = sig->at(stream.sig_index())._offset;
7535     if (off == -1) {
7536       // Nullable inline type argument, emit null check
7537       Label L_notNull;
7538       if (fromReg->is_stack()) {
7539         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7540         ldrb(tmp2, Address(sp, ld_off));
7541         cbnz(tmp2, L_notNull);
7542       } else {
7543         cbnz(fromReg->as_Register(), L_notNull);
7544       }
7545       mov(val_obj, 0);
7546       b(L_null);
7547       bind(L_notNull);
7548       continue;
7549     }
7550     if (sig->at(stream.sig_index())._vt_oop) {
7551       if (fromReg->is_stack()) {
7552         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7553         ldr(val_obj, Address(sp, ld_off));
7554       } else {
7555         mov(val_obj, fromReg->as_Register());
7556       }
7557       cbnz(val_obj, L_null);
7558       // get the buffer from the just allocated pool of buffers
7559       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7560       load_heap_oop(val_obj, Address(val_array, index), rscratch1, rscratch2);
7561       continue;
7562     }
7563 
7564     assert(off > 0, "offset in object should be positive");
7565     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7566 
7567     // Pack the scalarized field into the value object.
7568     Address dst(val_obj, off);
7569     if (!fromReg->is_FloatRegister()) {
7570       Register src;
7571       if (fromReg->is_stack()) {
7572         src = from_reg_tmp;
7573         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7574         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7575       } else {
7576         src = fromReg->as_Register();
7577       }
7578       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7579       if (is_reference_type(bt)) {
7580         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
7581         mov(tmp3, val_obj);
7582         Address dst_with_tmp3(tmp3, off);
7583         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7584       } else {
7585         store_sized_value(dst, src, size_in_bytes);
7586       }
7587     } else if (bt == T_DOUBLE) {
7588       strd(fromReg->as_FloatRegister(), dst);
7589     } else {
7590       assert(bt == T_FLOAT, "must be float");
7591       strs(fromReg->as_FloatRegister(), dst);
7592     }
7593   }
7594   bind(L_null);
7595   sig_index = stream.sig_index();
7596   from_index = stream.regs_index();
7597 
7598   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7599   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7600   assert(success, "to register must be writeable");
7601   return true;
7602 }
7603 
7604 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7605   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7606 }
7607 
7608 void MacroAssembler::cache_wb(Address line) {
7609   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7610   assert(line.index() == noreg, "index should be noreg");
7611   assert(line.offset() == 0, "offset should be 0");
7612   // would like to assert this
7613   // assert(line._ext.shift == 0, "shift should be zero");
7614   if (VM_Version::supports_dcpop()) {
7615     // writeback using clear virtual address to point of persistence
7616     dc(Assembler::CVAP, line.base());
7617   } else {
7618     // no need to generate anything as Unsafe.writebackMemory should
7619     // never invoke this stub
7620   }
7621 }
7622 
7623 void MacroAssembler::cache_wbsync(bool is_pre) {
7624   // we only need a barrier post sync
7625   if (!is_pre) {
7626     membar(Assembler::AnyAny);
7627   }

8023   }
8024 
8025   // Check if the lock-stack is full.
8026   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
8027   cmpw(top, (unsigned)LockStack::end_offset());
8028   br(Assembler::GE, slow);
8029 
8030   // Check for recursion.
8031   subw(t, top, oopSize);
8032   ldr(t, Address(rthread, t));
8033   cmp(obj, t);
8034   br(Assembler::EQ, push);
8035 
8036   // Check header for monitor (0b10).
8037   tst(mark, markWord::monitor_value);
8038   br(Assembler::NE, slow);
8039 
8040   // Try to lock. Transition lock bits 0b01 => 0b00
8041   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
8042   orr(mark, mark, markWord::unlocked_value);
8043   // Mask inline_type bit such that we go to the slow path if object is an inline type
8044   andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
8045 
8046   eor(t, mark, markWord::unlocked_value);
8047   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
8048           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
8049   br(Assembler::NE, slow);
8050 
8051   bind(push);
8052   // After successful lock, push object on lock-stack.
8053   str(obj, Address(rthread, top));
8054   addw(top, top, oopSize);
8055   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
8056 }
8057 
8058 // Implements fast-unlocking.
8059 //
8060 // - obj: the object to be unlocked
8061 // - t1, t2, t3: temporary registers
8062 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
8063 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
8064   // cmpxchg clobbers rscratch1.
8065   assert_different_registers(obj, t1, t2, t3, rscratch1);
< prev index next >