< prev index next >

src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp

Print this page

  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 "cds/archiveBuilder.hpp"
  29 #include "ci/ciEnv.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 "runtime/continuation.hpp"

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

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

  61 #ifdef COMPILER1
  62 #include "c1/c1_LIRAssembler.hpp"
  63 #endif
  64 #ifdef COMPILER2
  65 #include "oops/oop.hpp"
  66 #include "opto/compile.hpp"
  67 #include "opto/node.hpp"
  68 #include "opto/output.hpp"
  69 #endif
  70 
  71 #include <sys/types.h>
  72 
  73 #ifdef PRODUCT
  74 #define BLOCK_COMMENT(str) /* nothing */
  75 #else
  76 #define BLOCK_COMMENT(str) block_comment(str)
  77 #endif
  78 #define STOP(str) stop(str);
  79 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  80 

1990   ldarb(scratch, scratch);
1991   cmp(scratch, InstanceKlass::fully_initialized);
1992   br(Assembler::EQ, *L_fast_path);
1993 
1994   // Fast path check: current thread is initializer thread
1995   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
1996   cmp(rthread, scratch);
1997 
1998   if (L_slow_path == &L_fallthrough) {
1999     br(Assembler::EQ, *L_fast_path);
2000     bind(*L_slow_path);
2001   } else if (L_fast_path == &L_fallthrough) {
2002     br(Assembler::NE, *L_slow_path);
2003     bind(*L_fast_path);
2004   } else {
2005     Unimplemented();
2006   }
2007 }
2008 
2009 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2010   if (!VerifyOops) return;




2011 
2012   // Pass register number to verify_oop_subroutine
2013   const char* b = nullptr;
2014   {
2015     ResourceMark rm;
2016     stringStream ss;
2017     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2018     b = code_string(ss.as_string());
2019   }
2020   BLOCK_COMMENT("verify_oop {");
2021 
2022   strip_return_address(); // This might happen within a stack frame.
2023   protect_return_address();
2024   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2025   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2026 
2027   mov(r0, reg);
2028   movptr(rscratch1, (uintptr_t)(address)b);
2029 
2030   // call indirectly to solve generation ordering problem
2031   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2032   ldr(rscratch2, Address(rscratch2));
2033   blr(rscratch2);
2034 
2035   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2036   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2037   authenticate_return_address();
2038 
2039   BLOCK_COMMENT("} verify_oop");
2040 }
2041 
2042 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2043   if (!VerifyOops) return;




2044 
2045   const char* b = nullptr;
2046   {
2047     ResourceMark rm;
2048     stringStream ss;
2049     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2050     b = code_string(ss.as_string());
2051   }
2052   BLOCK_COMMENT("verify_oop_addr {");
2053 
2054   strip_return_address(); // This might happen within a stack frame.
2055   protect_return_address();
2056   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2057   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2058 
2059   // addr may contain sp so we will have to adjust it based on the
2060   // pushes that we just did.
2061   if (addr.uses(sp)) {
2062     lea(r0, addr);
2063     ldr(r0, Address(r0, 4 * wordSize));

2272   call_VM_leaf_base(entry_point, 1);
2273 }
2274 
2275 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2276   assert_different_registers(arg_1, c_rarg0);
2277   pass_arg0(this, arg_0);
2278   pass_arg1(this, arg_1);
2279   call_VM_leaf_base(entry_point, 2);
2280 }
2281 
2282 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2283                                   Register arg_1, Register arg_2) {
2284   assert_different_registers(arg_1, c_rarg0);
2285   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2286   pass_arg0(this, arg_0);
2287   pass_arg1(this, arg_1);
2288   pass_arg2(this, arg_2);
2289   call_VM_leaf_base(entry_point, 3);
2290 }
2291 




2292 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2293   pass_arg0(this, arg_0);
2294   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2295 }
2296 
2297 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2298 
2299   assert_different_registers(arg_0, c_rarg1);
2300   pass_arg1(this, arg_1);
2301   pass_arg0(this, arg_0);
2302   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2303 }
2304 
2305 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2306   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2307   assert_different_registers(arg_1, c_rarg2);
2308   pass_arg2(this, arg_2);
2309   pass_arg1(this, arg_1);
2310   pass_arg0(this, arg_0);
2311   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2317   assert_different_registers(arg_2, c_rarg3);
2318   pass_arg3(this, arg_3);
2319   pass_arg2(this, arg_2);
2320   pass_arg1(this, arg_1);
2321   pass_arg0(this, arg_0);
2322   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2323 }
2324 
2325 void MacroAssembler::null_check(Register reg, int offset) {
2326   if (needs_explicit_null_check(offset)) {
2327     // provoke OS null exception if reg is null by
2328     // accessing M[reg] w/o changing any registers
2329     // NOTE: this is plenty to provoke a segv
2330     ldr(zr, Address(reg));
2331   } else {
2332     // nothing to do, (later) access of M[reg + offset]
2333     // will provoke OS null exception if reg is null
2334   }
2335 }
2336 
















































































2337 // MacroAssembler protected routines needed to implement
2338 // public methods
2339 
2340 void MacroAssembler::mov(Register r, Address dest) {
2341   code_section()->relocate(pc(), dest.rspec());
2342   uint64_t imm64 = (uint64_t)dest.target();
2343   movptr(r, imm64);
2344 }
2345 
2346 // Move a constant pointer into r.  In AArch64 mode the virtual
2347 // address space is 48 bits in size, so we only need three
2348 // instructions to create a patchable instruction sequence that can
2349 // reach anywhere.
2350 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2351 #ifndef PRODUCT
2352   {
2353     char buffer[64];
2354     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2355     block_comment(buffer);
2356   }

5079   adrp(rscratch1, src2, offset);
5080   ldr(rscratch1, Address(rscratch1, offset));
5081   cmp(src1, rscratch1);
5082 }
5083 
5084 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5085   cmp(obj1, obj2);
5086 }
5087 
5088 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5089   load_method_holder(rresult, rmethod);
5090   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5091 }
5092 
5093 void MacroAssembler::load_method_holder(Register holder, Register method) {
5094   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5095   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5096   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5097 }
5098 








5099 // Loads the obj's narrow Klass from a compact object header (+COH) into dst.
5100 // Preserves all registers (incl src, rscratch1 and rscratch2).
5101 // Input:
5102 // src - the oop we want to load the klass from.
5103 // dst - output narrow klass.
5104 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5105   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5106   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5107   lsr(dst, dst, markWord::klass_shift);
5108 }
5109 
5110 // Loads the obj's narrow Klass from any header (compact or not) into dst.
5111 void MacroAssembler::load_narrow_klass(Register dst, Register src) {
5112   if (UseCompactObjectHeaders) {
5113     load_narrow_klass_compact(dst, src);
5114   } else {
5115     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5116   }
5117 }
5118 

5175   } else {
5176     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5177   }
5178   if (CompressedKlassPointers::base() == nullptr) {
5179     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5180     return;
5181   } else if (!AOTCodeCache::is_on_for_dump() &&
5182              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5183              && CompressedKlassPointers::shift() == 0) {
5184     // Only the bottom 32 bits matter
5185     cmpw(klass, tmp);
5186     return;
5187   }
5188   decode_klass_not_null(tmp, tmp, tmp2);
5189   cmp(klass, tmp);
5190 }
5191 
5192 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5193   if (UseCompactObjectHeaders) {
5194     load_narrow_klass_compact(tmp1, obj1);
5195     load_narrow_klass_compact(tmp2,  obj2);
5196   } else {
5197     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5198     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5199   }
5200   cmpw(tmp1, tmp2);
5201 }
5202 






5203 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
5204   // FIXME: Should this be a store release?  concurrent gcs assumes
5205   // klass length is valid if klass field is not null.
5206   assert(!UseCompactObjectHeaders, "not with compact headers");
5207   encode_klass_not_null(src, src, tmp);
5208   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5209 }
5210 
5211 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5212   assert(!UseCompactObjectHeaders, "not with compact headers");
5213   // Store to klass gap in destination
5214   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5215 }
5216 
5217 // Algorithm must match CompressedOops::encode.
5218 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5219 #ifdef ASSERT
5220   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5221 #endif
5222   verify_oop_msg(s, "broken oop in encode_heap_oop");

5554   if (as_raw) {
5555     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5556   } else {
5557     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5558   }
5559 }
5560 
5561 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5562                                      Address dst, Register val,
5563                                      Register tmp1, Register tmp2, Register tmp3) {
5564   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5565   decorators = AccessInternal::decorator_fixup(decorators, type);
5566   bool as_raw = (decorators & AS_RAW) != 0;
5567   if (as_raw) {
5568     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5569   } else {
5570     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5571   }
5572 }
5573 






















5574 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5575                                    Register tmp2, DecoratorSet decorators) {
5576   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5577 }
5578 
5579 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5580                                             Register tmp2, DecoratorSet decorators) {
5581   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5582 }
5583 
5584 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5585                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5586   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5587 }
5588 
5589 // Used for storing nulls.
5590 void MacroAssembler::store_heap_oop_null(Address dst) {
5591   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5592 }
5593 

5668     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5669     cmp(rscratch2, rscratch1);
5670     br(Assembler::HS, next);
5671     STOP("assert(top >= start)");
5672     should_not_reach_here();
5673 
5674     bind(next);
5675     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5676     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5677     cmp(rscratch2, rscratch1);
5678     br(Assembler::HS, ok);
5679     STOP("assert(top <= end)");
5680     should_not_reach_here();
5681 
5682     bind(ok);
5683     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5684   }
5685 #endif
5686 }
5687 















5688 // Writes to stack successive pages until offset reached to check for
5689 // stack overflow + shadow pages.  This clobbers tmp.
5690 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5691   assert_different_registers(tmp, size, rscratch1);
5692   mov(tmp, sp);
5693   // Bang stack for total size given plus shadow page size.
5694   // Bang one page at a time because large size can bang beyond yellow and
5695   // red zones.
5696   Label loop;
5697   mov(rscratch1, (int)os::vm_page_size());
5698   bind(loop);
5699   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5700   subsw(size, size, rscratch1);
5701   str(size, Address(tmp));
5702   br(Assembler::GT, loop);
5703 
5704   // Bang down shadow pages too.
5705   // At this point, (tmp-0) is the last address touched, so don't
5706   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5707   // was post-decremented.)  Skip this address by starting at i=1, and

5774 
5775   // Strictly speaking the card table base isn't an address at all, and it might
5776   // even be negative. It is thus materialised as a constant.
5777   mov(reg, (uint64_t)ctbs->card_table_base_const());
5778 }
5779 
5780 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5781 #if INCLUDE_CDS
5782   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5783   if (AOTCodeCache::is_on_for_dump()) {
5784     // all aotrc field addresses should be registered in the AOTCodeCache address table
5785     lea(reg, ExternalAddress(a));
5786   } else {
5787     mov(reg, (uint64_t)a);
5788   }
5789 #else
5790   ShouldNotReachHere();
5791 #endif
5792 }
5793 

5794 void MacroAssembler::build_frame(int framesize) {





5795   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5796   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5797   protect_return_address();
5798   if (framesize < ((1 << 9) + 2 * wordSize)) {
5799     sub(sp, sp, framesize);
5800     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));





5801     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5802   } else {
5803     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));





5804     if (PreserveFramePointer) mov(rfp, sp);
5805     if (framesize < ((1 << 12) + 2 * wordSize))
5806       sub(sp, sp, framesize - 2 * wordSize);
5807     else {
5808       mov(rscratch1, framesize - 2 * wordSize);
5809       sub(sp, sp, rscratch1);
5810     }
5811   }
5812   verify_cross_modify_fence_not_required();
5813 }
5814 
5815 void MacroAssembler::remove_frame(int framesize) {
5816   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5817   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5818   if (framesize < ((1 << 9) + 2 * wordSize)) {
5819     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5820     add(sp, sp, framesize);
5821   } else {
5822     if (framesize < ((1 << 12) + 2 * wordSize))
5823       add(sp, sp, framesize - 2 * wordSize);
5824     else {
5825       mov(rscratch1, framesize - 2 * wordSize);
5826       add(sp, sp, rscratch1);
5827     }
5828     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5829   }
5830   authenticate_return_address();
5831 }
5832 


















































































5833 
5834 // This method counts leading positive bytes (highest bit not set) in provided byte array
5835 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5836     // Simple and most common case of aligned small array which is not at the
5837     // end of memory page is placed here. All other cases are in stub.
5838     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5839     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5840     assert_different_registers(ary1, len, result);
5841 
5842     mov(result, len);
5843     cmpw(len, 0);
5844     br(LE, DONE);
5845     cmpw(len, 4 * wordSize);
5846     br(GE, STUB_LONG); // size > 32 then go to stub
5847 
5848     int shift = 64 - exact_log2(os::vm_page_size());
5849     lsl(rscratch1, ary1, shift);
5850     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5851     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5852     br(CS, STUB); // at the end of page then go to stub

6736 // On other systems, the helper is a usual C function.
6737 //
6738 void MacroAssembler::get_thread(Register dst) {
6739   RegSet saved_regs =
6740     BSD_ONLY(RegSet::range(r0, r17) + lr - dst)
6741     NOT_BSD (RegSet::range(r0, r1)  + lr - dst);
6742 
6743   protect_return_address();
6744   push(saved_regs, sp);
6745 
6746   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6747   blr(lr);
6748   if (dst != c_rarg0) {
6749     mov(dst, c_rarg0);
6750   }
6751 
6752   pop(saved_regs, sp);
6753   authenticate_return_address();
6754 }
6755 




























































































































































































































































































































































































































































































6756 void MacroAssembler::cache_wb(Address line) {
6757   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6758   assert(line.index() == noreg, "index should be noreg");
6759   assert(line.offset() == 0, "offset should be 0");
6760   // would like to assert this
6761   // assert(line._ext.shift == 0, "shift should be zero");
6762   if (VM_Version::supports_dcpop()) {
6763     // writeback using clear virtual address to point of persistence
6764     dc(Assembler::CVAP, line.base());
6765   } else {
6766     // no need to generate anything as Unsafe.writebackMemory should
6767     // never invoke this stub
6768   }
6769 }
6770 
6771 void MacroAssembler::cache_wbsync(bool is_pre) {
6772   // we only need a barrier post sync
6773   if (!is_pre) {
6774     membar(Assembler::AnyAny);
6775   }

7171   }
7172 
7173   // Check if the lock-stack is full.
7174   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7175   cmpw(top, (unsigned)LockStack::end_offset());
7176   br(Assembler::GE, slow);
7177 
7178   // Check for recursion.
7179   subw(t, top, oopSize);
7180   ldr(t, Address(rthread, t));
7181   cmp(obj, t);
7182   br(Assembler::EQ, push);
7183 
7184   // Check header for monitor (0b10).
7185   tst(mark, markWord::monitor_value);
7186   br(Assembler::NE, slow);
7187 
7188   // Try to lock. Transition lock bits 0b01 => 0b00
7189   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7190   orr(mark, mark, markWord::unlocked_value);



7191   eor(t, mark, markWord::unlocked_value);
7192   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword, memory_order_acquire);
7193   br(Assembler::NE, slow);
7194 
7195   bind(push);
7196   // After successful lock, push object on lock-stack.
7197   str(obj, Address(rthread, top));
7198   addw(top, top, oopSize);
7199   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7200 }
7201 
7202 // Implements fast-unlocking.
7203 //
7204 // - obj: the object to be unlocked
7205 // - t1, t2, t3: temporary registers
7206 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7207 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7208   // cmpxchg clobbers rscratch1.
7209   assert_different_registers(obj, t1, t2, t3, rscratch1);
7210 

  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 "cds/archiveBuilder.hpp"
  29 #include "ci/ciEnv.hpp"
  30 #include "ci/ciInlineKlass.hpp"
  31 #include "code/compiledIC.hpp"
  32 #include "compiler/compileTask.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "compiler/oopMap.hpp"
  35 #include "gc/shared/barrierSet.hpp"
  36 #include "gc/shared/barrierSetAssembler.hpp"
  37 #include "gc/shared/cardTableBarrierSet.hpp"
  38 #include "gc/shared/cardTable.hpp"
  39 #include "gc/shared/collectedHeap.hpp"
  40 #include "gc/shared/tlab_globals.hpp"
  41 #include "interpreter/bytecodeHistogram.hpp"
  42 #include "interpreter/interpreter.hpp"
  43 #include "interpreter/interpreterRuntime.hpp"
  44 #include "jvm.h"
  45 #include "memory/resourceArea.hpp"
  46 #include "memory/universe.hpp"
  47 #include "nativeInst_aarch64.hpp"
  48 #include "oops/accessDecorators.hpp"
  49 #include "oops/compressedKlass.inline.hpp"
  50 #include "oops/compressedOops.inline.hpp"
  51 #include "oops/klass.inline.hpp"
  52 #include "oops/resolvedFieldEntry.hpp"
  53 #include "runtime/arguments.hpp"
  54 #include "runtime/continuation.hpp"
  55 #include "runtime/globals.hpp"
  56 #include "runtime/icache.hpp"
  57 #include "runtime/interfaceSupport.inline.hpp"
  58 #include "runtime/javaThread.hpp"
  59 #include "runtime/jniHandles.inline.hpp"
  60 #include "runtime/sharedRuntime.hpp"
  61 #include "runtime/signature_cc.hpp"
  62 #include "runtime/stubRoutines.hpp"
  63 #include "utilities/globalDefinitions.hpp"
  64 #include "utilities/integerCast.hpp"
  65 #include "utilities/powerOfTwo.hpp"
  66 #include "vmreg_aarch64.inline.hpp"
  67 #ifdef COMPILER1
  68 #include "c1/c1_LIRAssembler.hpp"
  69 #endif
  70 #ifdef COMPILER2
  71 #include "oops/oop.hpp"
  72 #include "opto/compile.hpp"
  73 #include "opto/node.hpp"
  74 #include "opto/output.hpp"
  75 #endif
  76 
  77 #include <sys/types.h>
  78 
  79 #ifdef PRODUCT
  80 #define BLOCK_COMMENT(str) /* nothing */
  81 #else
  82 #define BLOCK_COMMENT(str) block_comment(str)
  83 #endif
  84 #define STOP(str) stop(str);
  85 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  86 

1996   ldarb(scratch, scratch);
1997   cmp(scratch, InstanceKlass::fully_initialized);
1998   br(Assembler::EQ, *L_fast_path);
1999 
2000   // Fast path check: current thread is initializer thread
2001   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2002   cmp(rthread, scratch);
2003 
2004   if (L_slow_path == &L_fallthrough) {
2005     br(Assembler::EQ, *L_fast_path);
2006     bind(*L_slow_path);
2007   } else if (L_fast_path == &L_fallthrough) {
2008     br(Assembler::NE, *L_slow_path);
2009     bind(*L_fast_path);
2010   } else {
2011     Unimplemented();
2012   }
2013 }
2014 
2015 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2016   if (!VerifyOops || VerifyAdapterSharing) {
2017     // Below address of the code string confuses VerifyAdapterSharing
2018     // because it may differ between otherwise equivalent adapters.
2019     return;
2020   }
2021 
2022   // Pass register number to verify_oop_subroutine
2023   const char* b = nullptr;
2024   {
2025     ResourceMark rm;
2026     stringStream ss;
2027     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2028     b = code_string(ss.as_string());
2029   }
2030   BLOCK_COMMENT("verify_oop {");
2031 
2032   strip_return_address(); // This might happen within a stack frame.
2033   protect_return_address();
2034   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2035   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2036 
2037   mov(r0, reg);
2038   movptr(rscratch1, (uintptr_t)(address)b);
2039 
2040   // call indirectly to solve generation ordering problem
2041   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2042   ldr(rscratch2, Address(rscratch2));
2043   blr(rscratch2);
2044 
2045   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2046   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2047   authenticate_return_address();
2048 
2049   BLOCK_COMMENT("} verify_oop");
2050 }
2051 
2052 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2053   if (!VerifyOops || VerifyAdapterSharing) {
2054     // Below address of the code string confuses VerifyAdapterSharing
2055     // because it may differ between otherwise equivalent adapters.
2056     return;
2057   }
2058 
2059   const char* b = nullptr;
2060   {
2061     ResourceMark rm;
2062     stringStream ss;
2063     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2064     b = code_string(ss.as_string());
2065   }
2066   BLOCK_COMMENT("verify_oop_addr {");
2067 
2068   strip_return_address(); // This might happen within a stack frame.
2069   protect_return_address();
2070   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2071   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2072 
2073   // addr may contain sp so we will have to adjust it based on the
2074   // pushes that we just did.
2075   if (addr.uses(sp)) {
2076     lea(r0, addr);
2077     ldr(r0, Address(r0, 4 * wordSize));

2286   call_VM_leaf_base(entry_point, 1);
2287 }
2288 
2289 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2290   assert_different_registers(arg_1, c_rarg0);
2291   pass_arg0(this, arg_0);
2292   pass_arg1(this, arg_1);
2293   call_VM_leaf_base(entry_point, 2);
2294 }
2295 
2296 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2297                                   Register arg_1, Register arg_2) {
2298   assert_different_registers(arg_1, c_rarg0);
2299   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2300   pass_arg0(this, arg_0);
2301   pass_arg1(this, arg_1);
2302   pass_arg2(this, arg_2);
2303   call_VM_leaf_base(entry_point, 3);
2304 }
2305 
2306 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2307   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2308 }
2309 
2310 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2311   pass_arg0(this, arg_0);
2312   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2313 }
2314 
2315 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2316 
2317   assert_different_registers(arg_0, c_rarg1);
2318   pass_arg1(this, arg_1);
2319   pass_arg0(this, arg_0);
2320   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2321 }
2322 
2323 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2324   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2325   assert_different_registers(arg_1, c_rarg2);
2326   pass_arg2(this, arg_2);
2327   pass_arg1(this, arg_1);
2328   pass_arg0(this, arg_0);
2329   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2335   assert_different_registers(arg_2, c_rarg3);
2336   pass_arg3(this, arg_3);
2337   pass_arg2(this, arg_2);
2338   pass_arg1(this, arg_1);
2339   pass_arg0(this, arg_0);
2340   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2341 }
2342 
2343 void MacroAssembler::null_check(Register reg, int offset) {
2344   if (needs_explicit_null_check(offset)) {
2345     // provoke OS null exception if reg is null by
2346     // accessing M[reg] w/o changing any registers
2347     // NOTE: this is plenty to provoke a segv
2348     ldr(zr, Address(reg));
2349   } else {
2350     // nothing to do, (later) access of M[reg + offset]
2351     // will provoke OS null exception if reg is null
2352   }
2353 }
2354 
2355 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2356   assert_different_registers(markword, rscratch2);
2357   mov(rscratch2, markWord::inline_type_pattern_mask);
2358   andr(markword, markword, rscratch2);
2359   mov(rscratch2, markWord::inline_type_pattern);
2360   cmp(markword, rscratch2);
2361   br(Assembler::EQ, is_inline_type);
2362 }
2363 
2364 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
2365   assert_different_registers(tmp, rscratch1);
2366   if (can_be_null) {
2367     cbz(object, not_inline_type);
2368   }
2369   const int is_inline_type_mask = markWord::inline_type_pattern;
2370   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2371   mov(rscratch1, is_inline_type_mask);
2372   andr(tmp, tmp, rscratch1);
2373   cmp(tmp, rscratch1);
2374   br(Assembler::NE, not_inline_type);
2375 }
2376 
2377 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2378   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2379   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2380 }
2381 
2382 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2383   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2384   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2385 }
2386 
2387 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2388   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2389   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2390 }
2391 
2392 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2393   // load mark word
2394   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2395   if (!UseObjectMonitorTable) {
2396     Label test_mark_word;
2397     // check displaced
2398     tst(temp_reg, markWord::unlocked_value);
2399     br(Assembler::NE, test_mark_word);
2400     // slow path use klass prototype
2401     load_prototype_header(temp_reg, oop);
2402 
2403     bind(test_mark_word);
2404   }
2405   andr(temp_reg, temp_reg, test_bit);
2406   if (jmp_set) {
2407     cbnz(temp_reg, jmp_label);
2408   } else {
2409     cbz(temp_reg, jmp_label);
2410   }
2411 }
2412 
2413 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2414   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2415 }
2416 
2417 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2418                                                   Label&is_non_flat_array) {
2419   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2420 }
2421 
2422 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2423   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2424 }
2425 
2426 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2427   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2428 }
2429 
2430 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2431   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2432   br(Assembler::NE, is_flat_array);
2433 }
2434 
2435 // MacroAssembler protected routines needed to implement
2436 // public methods
2437 
2438 void MacroAssembler::mov(Register r, Address dest) {
2439   code_section()->relocate(pc(), dest.rspec());
2440   uint64_t imm64 = (uint64_t)dest.target();
2441   movptr(r, imm64);
2442 }
2443 
2444 // Move a constant pointer into r.  In AArch64 mode the virtual
2445 // address space is 48 bits in size, so we only need three
2446 // instructions to create a patchable instruction sequence that can
2447 // reach anywhere.
2448 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2449 #ifndef PRODUCT
2450   {
2451     char buffer[64];
2452     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2453     block_comment(buffer);
2454   }

5177   adrp(rscratch1, src2, offset);
5178   ldr(rscratch1, Address(rscratch1, offset));
5179   cmp(src1, rscratch1);
5180 }
5181 
5182 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5183   cmp(obj1, obj2);
5184 }
5185 
5186 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5187   load_method_holder(rresult, rmethod);
5188   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5189 }
5190 
5191 void MacroAssembler::load_method_holder(Register holder, Register method) {
5192   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5193   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5194   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5195 }
5196 
5197 void MacroAssembler::load_metadata(Register dst, Register src) {
5198   if (UseCompactObjectHeaders) {
5199     load_narrow_klass_compact(dst, src);
5200   } else {
5201     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5202   }
5203 }
5204 
5205 // Loads the obj's narrow Klass from a compact object header (+COH) into dst.
5206 // Preserves all registers (incl src, rscratch1 and rscratch2).
5207 // Input:
5208 // src - the oop we want to load the klass from.
5209 // dst - output narrow klass.
5210 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5211   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5212   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5213   lsr(dst, dst, markWord::klass_shift);
5214 }
5215 
5216 // Loads the obj's narrow Klass from any header (compact or not) into dst.
5217 void MacroAssembler::load_narrow_klass(Register dst, Register src) {
5218   if (UseCompactObjectHeaders) {
5219     load_narrow_klass_compact(dst, src);
5220   } else {
5221     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5222   }
5223 }
5224 

5281   } else {
5282     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5283   }
5284   if (CompressedKlassPointers::base() == nullptr) {
5285     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5286     return;
5287   } else if (!AOTCodeCache::is_on_for_dump() &&
5288              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5289              && CompressedKlassPointers::shift() == 0) {
5290     // Only the bottom 32 bits matter
5291     cmpw(klass, tmp);
5292     return;
5293   }
5294   decode_klass_not_null(tmp, tmp, tmp2);
5295   cmp(klass, tmp);
5296 }
5297 
5298 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5299   if (UseCompactObjectHeaders) {
5300     load_narrow_klass_compact(tmp1, obj1);
5301     load_narrow_klass_compact(tmp2, obj2);
5302   } else {
5303     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5304     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5305   }
5306   cmpw(tmp1, tmp2);
5307 }
5308 
5309 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5310   Register tmp = (dst == rscratch1) ? rscratch2 : rscratch1;
5311   load_klass(dst, src, tmp);
5312   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5313 }
5314 
5315 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
5316   // FIXME: Should this be a store release?  concurrent gcs assumes
5317   // klass length is valid if klass field is not null.
5318   assert(!UseCompactObjectHeaders, "not with compact headers");
5319   encode_klass_not_null(src, src, tmp);
5320   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5321 }
5322 
5323 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5324   assert(!UseCompactObjectHeaders, "not with compact headers");
5325   // Store to klass gap in destination
5326   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5327 }
5328 
5329 // Algorithm must match CompressedOops::encode.
5330 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5331 #ifdef ASSERT
5332   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5333 #endif
5334   verify_oop_msg(s, "broken oop in encode_heap_oop");

5666   if (as_raw) {
5667     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5668   } else {
5669     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5670   }
5671 }
5672 
5673 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5674                                      Address dst, Register val,
5675                                      Register tmp1, Register tmp2, Register tmp3) {
5676   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5677   decorators = AccessInternal::decorator_fixup(decorators, type);
5678   bool as_raw = (decorators & AS_RAW) != 0;
5679   if (as_raw) {
5680     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5681   } else {
5682     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5683   }
5684 }
5685 
5686 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5687                                      Register inline_layout_info) {
5688   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5689   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5690 }
5691 
5692 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5693   ldr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
5694   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5695 }
5696 
5697 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5698   // ((address) (void*) o) + vk->payload_offset();
5699   Register offset = (data == oop) ? rscratch1 : data;
5700   payload_offset(inline_klass, offset);
5701   if (data == oop) {
5702     add(data, data, offset);
5703   } else {
5704     lea(data, Address(oop, offset));
5705   }
5706 }
5707 
5708 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5709                                    Register tmp2, DecoratorSet decorators) {
5710   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5711 }
5712 
5713 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5714                                             Register tmp2, DecoratorSet decorators) {
5715   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5716 }
5717 
5718 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5719                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5720   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5721 }
5722 
5723 // Used for storing nulls.
5724 void MacroAssembler::store_heap_oop_null(Address dst) {
5725   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5726 }
5727 

5802     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5803     cmp(rscratch2, rscratch1);
5804     br(Assembler::HS, next);
5805     STOP("assert(top >= start)");
5806     should_not_reach_here();
5807 
5808     bind(next);
5809     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5810     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5811     cmp(rscratch2, rscratch1);
5812     br(Assembler::HS, ok);
5813     STOP("assert(top <= end)");
5814     should_not_reach_here();
5815 
5816     bind(ok);
5817     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5818   }
5819 #endif
5820 }
5821 
5822 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5823   assert_different_registers(holder_klass, index, layout_info);
5824   InlineLayoutInfo array[2];
5825   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5826   if (is_power_of_2(size)) {
5827     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5828   } else {
5829     mov(layout_info, size);
5830     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5831   }
5832   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5833   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5834   lea(layout_info, Address(layout_info, index));
5835 }
5836 
5837 // Writes to stack successive pages until offset reached to check for
5838 // stack overflow + shadow pages.  This clobbers tmp.
5839 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5840   assert_different_registers(tmp, size, rscratch1);
5841   mov(tmp, sp);
5842   // Bang stack for total size given plus shadow page size.
5843   // Bang one page at a time because large size can bang beyond yellow and
5844   // red zones.
5845   Label loop;
5846   mov(rscratch1, (int)os::vm_page_size());
5847   bind(loop);
5848   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5849   subsw(size, size, rscratch1);
5850   str(size, Address(tmp));
5851   br(Assembler::GT, loop);
5852 
5853   // Bang down shadow pages too.
5854   // At this point, (tmp-0) is the last address touched, so don't
5855   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5856   // was post-decremented.)  Skip this address by starting at i=1, and

5923 
5924   // Strictly speaking the card table base isn't an address at all, and it might
5925   // even be negative. It is thus materialised as a constant.
5926   mov(reg, (uint64_t)ctbs->card_table_base_const());
5927 }
5928 
5929 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5930 #if INCLUDE_CDS
5931   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5932   if (AOTCodeCache::is_on_for_dump()) {
5933     // all aotrc field addresses should be registered in the AOTCodeCache address table
5934     lea(reg, ExternalAddress(a));
5935   } else {
5936     mov(reg, (uint64_t)a);
5937   }
5938 #else
5939   ShouldNotReachHere();
5940 #endif
5941 }
5942 
5943 #ifdef ASSERT
5944 void MacroAssembler::build_frame(int framesize) {
5945   build_frame(framesize, false);
5946 }
5947 #endif
5948 
5949 void MacroAssembler::build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills)) {
5950   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5951   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5952   protect_return_address();
5953   if (framesize < ((1 << 9) + 2 * wordSize)) {
5954     sub(sp, sp, framesize);
5955     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5956       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5957       stp(rscratch1, rscratch1, Address(sp, framesize - 2 * wordSize));
5958     } else {
5959       stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5960     }
5961     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5962   } else {
5963     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5964       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5965       stp(rscratch1, rscratch1, Address(pre(sp, -2 * wordSize)));
5966     } else {
5967       stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5968     }
5969     if (PreserveFramePointer) mov(rfp, sp);
5970     if (framesize < ((1 << 12) + 2 * wordSize))
5971       sub(sp, sp, framesize - 2 * wordSize);
5972     else {
5973       mov(rscratch1, framesize - 2 * wordSize);
5974       sub(sp, sp, rscratch1);
5975     }
5976   }
5977   verify_cross_modify_fence_not_required();
5978 }
5979 
5980 void MacroAssembler::remove_frame(int framesize) {
5981   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5982   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5983   if (framesize < ((1 << 9) + 2 * wordSize)) {
5984     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5985     add(sp, sp, framesize);
5986   } else {
5987     if (framesize < ((1 << 12) + 2 * wordSize))
5988       add(sp, sp, framesize - 2 * wordSize);
5989     else {
5990       mov(rscratch1, framesize - 2 * wordSize);
5991       add(sp, sp, rscratch1);
5992     }
5993     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5994   }
5995   authenticate_return_address();
5996 }
5997 
5998 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
5999   if (needs_stack_repair) {
6000     // The method has a scalarized entry point (where fields of value object arguments
6001     // are passed through registers and stack), and a non-scalarized entry point (where
6002     // value object arguments are given as oops). The non-scalarized entry point will
6003     // first load each field of value object arguments and store them in registers and on
6004     // the stack in a way compatible with the scalarized entry point. To do so, some extra
6005     // stack space might be reserved (if argument registers are not enough). On leaving the
6006     // method, this space must be freed.
6007     //
6008     // In case we used the non-scalarized entry point the stack looks like this:
6009     //
6010     // | Arguments from caller     |
6011     // |---------------------------|  <-- caller's SP
6012     // | Saved LR #1               |
6013     // | Saved FP #1               |
6014     // |---------------------------|
6015     // | Extension space for       |
6016     // |   inline arg (un)packing  |
6017     // |---------------------------|  <-- start of this method's frame
6018     // | Saved LR #2               |
6019     // | Saved FP #2               |
6020     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6021     // | sp_inc                    |
6022     // | method locals             |
6023     // |---------------------------|  <-- SP
6024     //
6025     // There are two copies of FP and LR on the stack. They will be identical at
6026     // first, but that can change.
6027     // If the caller has been deoptimized, LR #1 will be patched to point at the
6028     // deopt blob, and LR #2 will still point into the old method.
6029     // If the saved FP (x29) was not used as the frame pointer, but to store an
6030     // oop, the GC will be aware only of FP #1 as the spilled location of x29 and
6031     // will fix only this one. Overall, FP/LR #2 are not reliable and are simply
6032     // needed to add space between the extension space and the locals, as there
6033     // would be between the real arguments and the locals if we don't need to
6034     // do unpacking (from the scalarized entry point).
6035     //
6036     // When restoring, one must then load FP #1 into x29, and LR #1 into x30,
6037     // while keeping in mind that from the scalarized entry point, there will be
6038     // only one copy of each. Indeed, in the case we used the scalarized calling
6039     // convention, the stack looks like this:
6040     //
6041     // | Arguments from caller     |
6042     // |---------------------------|  <-- caller's SP / start of this method's frame
6043     // | Saved LR                  |
6044     // | Saved FP                  |
6045     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6046     // | sp_inc                    |
6047     // | method locals             |
6048     // |---------------------------|  <-- SP
6049     //
6050     // The sp_inc stack slot holds the total size of the frame including the
6051     // extension space minus two words for the saved FP and LR. That is how to
6052     // find FP/LR #1. This size is expressed in bytes. Be careful when using it
6053     // from C++ in pointer arithmetic; you might need to divide it by wordSize.
6054     //
6055     // One can find sp_inc since the start the method's frame is SP + initial_framesize.
6056 
6057     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
6058 
6059     ldr(rscratch1, Address(sp, sp_inc_offset));
6060     add(sp, sp, rscratch1);
6061     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6062   } else {
6063     remove_frame(initial_framesize);
6064   }
6065 }
6066 
6067 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
6068   int real_frame_size = frame_size + sp_inc;
6069   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
6070   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
6071   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
6072 
6073   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
6074 
6075   // Subtract two words for the saved FP and LR as these will be popped
6076   // separately. See remove_frame above.
6077   mov(rscratch1, real_frame_size - 2*wordSize);
6078   str(rscratch1, Address(sp, sp_inc_offset));
6079 }
6080 
6081 // This method counts leading positive bytes (highest bit not set) in provided byte array
6082 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6083     // Simple and most common case of aligned small array which is not at the
6084     // end of memory page is placed here. All other cases are in stub.
6085     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6086     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6087     assert_different_registers(ary1, len, result);
6088 
6089     mov(result, len);
6090     cmpw(len, 0);
6091     br(LE, DONE);
6092     cmpw(len, 4 * wordSize);
6093     br(GE, STUB_LONG); // size > 32 then go to stub
6094 
6095     int shift = 64 - exact_log2(os::vm_page_size());
6096     lsl(rscratch1, ary1, shift);
6097     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6098     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6099     br(CS, STUB); // at the end of page then go to stub

6983 // On other systems, the helper is a usual C function.
6984 //
6985 void MacroAssembler::get_thread(Register dst) {
6986   RegSet saved_regs =
6987     BSD_ONLY(RegSet::range(r0, r17) + lr - dst)
6988     NOT_BSD (RegSet::range(r0, r1)  + lr - dst);
6989 
6990   protect_return_address();
6991   push(saved_regs, sp);
6992 
6993   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6994   blr(lr);
6995   if (dst != c_rarg0) {
6996     mov(dst, c_rarg0);
6997   }
6998 
6999   pop(saved_regs, sp);
7000   authenticate_return_address();
7001 }
7002 
7003 #ifdef COMPILER2
7004 // C2 compiled method's prolog code
7005 // Moved here from aarch64.ad to support Valhalla code below
7006 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
7007   if (C->clinit_barrier_on_entry()) {
7008     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
7009 
7010     Label L_skip_barrier;
7011 
7012     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
7013     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
7014     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
7015     bind(L_skip_barrier);
7016   }
7017 
7018   if (C->max_vector_size() > 0) {
7019     reinitialize_ptrue();
7020   }
7021 
7022   int bangsize = C->output()->bang_size_in_bytes();
7023   if (C->output()->need_stack_bang(bangsize))
7024     generate_stack_overflow_check(bangsize);
7025 
7026   // n.b. frame size includes space for return pc and rfp
7027   const long framesize = C->output()->frame_size_in_bytes();
7028   build_frame(framesize DEBUG_ONLY(COMMA sp_inc != 0));
7029 
7030   if (C->needs_stack_repair()) {
7031     save_stack_increment(sp_inc, framesize);
7032   }
7033 
7034   if (VerifyStackAtCalls) {
7035     Unimplemented();
7036   }
7037 }
7038 #endif // COMPILER2
7039 
7040 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
7041   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
7042   // An inline type might be returned. If fields are in registers we
7043   // need to allocate an inline type instance and initialize it with
7044   // the value of the fields.
7045   Label skip;
7046   // We only need a new buffered inline type if a new one is not returned
7047   tbz(r0, 0, skip);
7048   int call_offset = -1;
7049 
7050   // Be careful not to clobber r1-7 which hold returned fields
7051   // Also do not use callee-saved registers as these may be live in the interpreter
7052   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
7053 
7054   // The following code is similar to the instance allocation code in TemplateTable::_new
7055   //  but has some slight differences,
7056   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
7057   // allocating is not necessary if vk != nullptr, etc.
7058   Label slow_case;
7059   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
7060   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
7061 
7062   if (vk != nullptr) {
7063     // Called from C1, where the return type is statically known.
7064     movptr(klass, (intptr_t)vk->get_InlineKlass());
7065     jint lh = vk->layout_helper();
7066     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
7067     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
7068       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
7069     } else {
7070       b(slow_case);
7071     }
7072   } else {
7073     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
7074     andr(klass, r0, -2);
7075     if (UseTLAB) {
7076       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
7077       tst(tmp2, Klass::_lh_instance_slow_path_bit);
7078       br(Assembler::NE, slow_case);
7079       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
7080     } else {
7081       b(slow_case);
7082     }
7083   }
7084   if (UseTLAB) {
7085     // 2. Initialize buffered inline instance header
7086     Register buffer_obj = r0;
7087     if (UseCompactObjectHeaders) {
7088       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7089       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7090     } else {
7091       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7092       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7093       store_klass_gap(buffer_obj, zr);
7094       if (vk == nullptr) {
7095         // store_klass corrupts klass, so save it for later use (interpreter case only).
7096         mov(tmp1, klass);
7097       }
7098       store_klass(buffer_obj, klass, rscratch1);
7099       klass = tmp1;
7100     }
7101     // 3. Initialize its fields with an inline class specific handler
7102     if (vk != nullptr) {
7103       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7104     } else {
7105       ldr(tmp1, Address(klass, InlineKlass::adr_members_offset()));
7106       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7107       blr(tmp1);
7108     }
7109 
7110     membar(Assembler::StoreStore);
7111     b(skip);
7112   } else {
7113     // Must have already branched to slow_case above.
7114     DEBUG_ONLY(should_not_reach_here());
7115   }
7116   bind(slow_case);
7117   // We failed to allocate a new inline type, fall back to a runtime
7118   // call. Some oop field may be live in some registers but we can't
7119   // tell. That runtime call will take care of preserving them
7120   // across a GC if there's one.
7121   mov(r0, r0_preserved);
7122 
7123   if (from_interpreter) {
7124     super_call_VM_leaf(SharedRuntime::store_inline_type_fields_to_buf_entry());
7125   } else {
7126     far_call(RuntimeAddress(SharedRuntime::store_inline_type_fields_to_buf_entry()));
7127     call_offset = offset();
7128   }
7129   membar(Assembler::StoreStore);
7130 
7131   bind(skip);
7132   return call_offset;
7133 }
7134 
7135 // Move a value between registers/stack slots and update the reg_state
7136 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7137   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7138   if (reg_state[to->value()] == reg_written) {
7139     return true; // Already written
7140   }
7141 
7142   if (from != to && bt != T_VOID) {
7143     if (reg_state[to->value()] == reg_readonly) {
7144       return false; // Not yet writable
7145     }
7146     if (from->is_reg()) {
7147       if (to->is_reg()) {
7148         if (from->is_Register() && to->is_Register()) {
7149           mov(to->as_Register(), from->as_Register());
7150         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7151           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7152         } else {
7153           ShouldNotReachHere();
7154         }
7155       } else {
7156         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7157         Address to_addr = Address(sp, st_off);
7158         if (from->is_FloatRegister()) {
7159           if (bt == T_DOUBLE) {
7160              strd(from->as_FloatRegister(), to_addr);
7161           } else {
7162              assert(bt == T_FLOAT, "must be float");
7163              strs(from->as_FloatRegister(), to_addr);
7164           }
7165         } else {
7166           str(from->as_Register(), to_addr);
7167         }
7168       }
7169     } else {
7170       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7171       if (to->is_reg()) {
7172         if (to->is_FloatRegister()) {
7173           if (bt == T_DOUBLE) {
7174             ldrd(to->as_FloatRegister(), from_addr);
7175           } else {
7176             assert(bt == T_FLOAT, "must be float");
7177             ldrs(to->as_FloatRegister(), from_addr);
7178           }
7179         } else {
7180           ldr(to->as_Register(), from_addr);
7181         }
7182       } else {
7183         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7184         ldr(rscratch1, from_addr);
7185         str(rscratch1, Address(sp, st_off));
7186       }
7187     }
7188   }
7189 
7190   // Update register states
7191   reg_state[from->value()] = reg_writable;
7192   reg_state[to->value()] = reg_written;
7193   return true;
7194 }
7195 
7196 // Calculate the extra stack space required for packing or unpacking inline
7197 // args and adjust the stack pointer
7198 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7199   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7200   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7201   assert(sp_inc > 0, "sanity");
7202 
7203   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7204   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7205 
7206   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7207   if (sp_inc < (1 << 9)) {
7208     sub(sp, sp, sp_inc);   // Fits in an immediate
7209   } else {
7210     mov(rscratch1, sp_inc);
7211     sub(sp, sp, rscratch1);
7212   }
7213 
7214   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7215 }
7216 
7217 // Read all fields from an inline type oop and store the values in registers/stack slots
7218 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7219                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7220                                           RegState reg_state[]) {
7221   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7222   assert(from->is_valid(), "source must be valid");
7223   bool progress = false;
7224 #ifdef ASSERT
7225   const int start_offset = offset();
7226 #endif
7227 
7228   Label L_null, L_notNull;
7229   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7230   Register tmp1 = r10;
7231   Register tmp2 = r11;
7232 
7233 #ifdef ASSERT
7234   RegSet clobbered_gp_regs = MacroAssembler::call_clobbered_gp_registers();
7235   assert(clobbered_gp_regs.contains(tmp1), "tmp1 must be saved explicitly if it's not a clobber");
7236   assert(clobbered_gp_regs.contains(tmp2), "tmp2 must be saved explicitly if it's not a clobber");
7237   assert(clobbered_gp_regs.contains(r14), "r14 must be saved explicitly if it's not a clobber");
7238 #endif
7239 
7240   Register fromReg = noreg;
7241   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
7242   bool done = true;
7243   bool mark_done = true;
7244   VMReg toReg;
7245   BasicType bt;
7246   // Check if argument requires a null check
7247   bool null_check = false;
7248   VMReg nullCheckReg;
7249   while (stream.next(nullCheckReg, bt)) {
7250     if (sig->at(stream.sig_index())._offset == -1) {
7251       null_check = true;
7252       break;
7253     }
7254   }
7255   stream.reset(sig_index, to_index);
7256   while (stream.next(toReg, bt)) {
7257     assert(toReg->is_valid(), "destination must be valid");
7258     int idx = (int)toReg->value();
7259     if (reg_state[idx] == reg_readonly) {
7260       if (idx != from->value()) {
7261         mark_done = false;
7262       }
7263       done = false;
7264       continue;
7265     } else if (reg_state[idx] == reg_written) {
7266       continue;
7267     }
7268     assert(reg_state[idx] == reg_writable, "must be writable");
7269     reg_state[idx] = reg_written;
7270     progress = true;
7271 
7272     if (fromReg == noreg) {
7273       if (from->is_reg()) {
7274         fromReg = from->as_Register();
7275       } else {
7276         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7277         ldr(tmp1, Address(sp, st_off));
7278         fromReg = tmp1;
7279       }
7280       if (null_check) {
7281         // Nullable inline type argument, emit null check
7282         cbz(fromReg, L_null);
7283       }
7284     }
7285     int off = sig->at(stream.sig_index())._offset;
7286     if (off == -1) {
7287       assert(null_check, "Missing null check at");
7288       if (toReg->is_stack()) {
7289         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7290         mov(tmp2, 1);
7291         str(tmp2, Address(sp, st_off));
7292       } else {
7293         mov(toReg->as_Register(), 1);
7294       }
7295       continue;
7296     }
7297     if (sig->at(stream.sig_index())._vt_oop) {
7298       if (toReg->is_stack()) {
7299         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7300         str(fromReg, Address(sp, st_off));
7301       } else {
7302         mov(toReg->as_Register(), fromReg);
7303       }
7304       continue;
7305     }
7306     assert(off > 0, "offset in object should be positive");
7307     Address fromAddr = Address(fromReg, off);
7308     if (!toReg->is_FloatRegister()) {
7309       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7310       if (is_reference_type(bt)) {
7311         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7312       } else {
7313         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7314         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7315       }
7316       if (toReg->is_stack()) {
7317         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7318         str(dst, Address(sp, st_off));
7319       }
7320     } else if (bt == T_DOUBLE) {
7321       ldrd(toReg->as_FloatRegister(), fromAddr);
7322     } else {
7323       assert(bt == T_FLOAT, "must be float");
7324       ldrs(toReg->as_FloatRegister(), fromAddr);
7325     }
7326   }
7327   if (progress && null_check) {
7328     if (done) {
7329       b(L_notNull);
7330       bind(L_null);
7331       // Set null marker to zero to signal that the argument is null.
7332       // Also set all fields to zero since the runtime requires a canonical
7333       // representation of a flat null.
7334       stream.reset(sig_index, to_index);
7335       while (stream.next(toReg, bt)) {
7336         if (toReg->is_stack()) {
7337           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7338           str(zr, Address(sp, st_off));
7339         } else if (toReg->is_FloatRegister()) {
7340           mov(toReg->as_FloatRegister(), T2S, 0);
7341         } else {
7342           mov(toReg->as_Register(), zr);
7343         }
7344       }
7345       bind(L_notNull);
7346     } else {
7347       bind(L_null);
7348     }
7349   }
7350 
7351   sig_index = stream.sig_index();
7352   to_index = stream.regs_index();
7353 
7354   if (mark_done && reg_state[from->value()] != reg_written) {
7355     // This is okay because no one else will write to that slot
7356     reg_state[from->value()] = reg_writable;
7357   }
7358   from_index--;
7359   assert(progress || (start_offset == offset()), "should not emit code");
7360   return done;
7361 }
7362 
7363 // Pack fields back into an inline type oop
7364 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7365                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7366                                         RegState reg_state[], Register val_array) {
7367   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7368   assert(to->is_valid(), "destination must be valid");
7369 
7370   if (reg_state[to->value()] == reg_written) {
7371     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7372     return true; // Already written
7373   }
7374 
7375   // The GC barrier expanded by store_heap_oop below may call into the
7376   // runtime so use callee-saved registers for any values that need to be
7377   // preserved. The GC barrier assembler should take care of saving the
7378   // Java argument registers.
7379   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7380   Register val_obj_tmp = r21;
7381   Register from_reg_tmp = r22;
7382   Register tmp1 = r14;
7383   Register tmp2 = r13;
7384   Register tmp3 = r12;
7385   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7386 
7387   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7388 
7389   if (reg_state[to->value()] == reg_readonly) {
7390     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7391       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7392       return false; // Not yet writable
7393     }
7394     val_obj = val_obj_tmp;
7395   }
7396 
7397   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7398   VMReg fromReg;
7399   BasicType bt;
7400   Label L_null;
7401   while (stream.next(fromReg, bt)) {
7402     assert(fromReg->is_valid(), "source must be valid");
7403     reg_state[fromReg->value()] = reg_writable;
7404 
7405     int off = sig->at(stream.sig_index())._offset;
7406     if (off == -1) {
7407       // Nullable inline type argument, emit null check
7408       Label L_notNull;
7409       if (fromReg->is_stack()) {
7410         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7411         ldrb(tmp2, Address(sp, ld_off));
7412         cbnz(tmp2, L_notNull);
7413       } else {
7414         cbnz(fromReg->as_Register(), L_notNull);
7415       }
7416       mov(val_obj, 0);
7417       b(L_null);
7418       bind(L_notNull);
7419       continue;
7420     }
7421     if (sig->at(stream.sig_index())._vt_oop) {
7422       if (fromReg->is_stack()) {
7423         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7424         ldr(val_obj, Address(sp, ld_off));
7425       } else {
7426         mov(val_obj, fromReg->as_Register());
7427       }
7428       cbnz(val_obj, L_null);
7429       // get the buffer from the just allocated pool of buffers
7430       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7431       load_heap_oop(val_obj, Address(val_array, index), rscratch1, rscratch2);
7432       continue;
7433     }
7434 
7435     assert(off > 0, "offset in object should be positive");
7436     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7437 
7438     // Pack the scalarized field into the value object.
7439     Address dst(val_obj, off);
7440     if (!fromReg->is_FloatRegister()) {
7441       Register src;
7442       if (fromReg->is_stack()) {
7443         src = from_reg_tmp;
7444         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7445         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7446       } else {
7447         src = fromReg->as_Register();
7448       }
7449       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7450       if (is_reference_type(bt)) {
7451         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
7452         mov(tmp3, val_obj);
7453         Address dst_with_tmp3(tmp3, off);
7454         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7455       } else {
7456         store_sized_value(dst, src, size_in_bytes);
7457       }
7458     } else if (bt == T_DOUBLE) {
7459       strd(fromReg->as_FloatRegister(), dst);
7460     } else {
7461       assert(bt == T_FLOAT, "must be float");
7462       strs(fromReg->as_FloatRegister(), dst);
7463     }
7464   }
7465   bind(L_null);
7466   sig_index = stream.sig_index();
7467   from_index = stream.regs_index();
7468 
7469   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7470   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7471   assert(success, "to register must be writable");
7472   return true;
7473 }
7474 
7475 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7476   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7477 }
7478 
7479 void MacroAssembler::cache_wb(Address line) {
7480   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7481   assert(line.index() == noreg, "index should be noreg");
7482   assert(line.offset() == 0, "offset should be 0");
7483   // would like to assert this
7484   // assert(line._ext.shift == 0, "shift should be zero");
7485   if (VM_Version::supports_dcpop()) {
7486     // writeback using clear virtual address to point of persistence
7487     dc(Assembler::CVAP, line.base());
7488   } else {
7489     // no need to generate anything as Unsafe.writebackMemory should
7490     // never invoke this stub
7491   }
7492 }
7493 
7494 void MacroAssembler::cache_wbsync(bool is_pre) {
7495   // we only need a barrier post sync
7496   if (!is_pre) {
7497     membar(Assembler::AnyAny);
7498   }

7894   }
7895 
7896   // Check if the lock-stack is full.
7897   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7898   cmpw(top, (unsigned)LockStack::end_offset());
7899   br(Assembler::GE, slow);
7900 
7901   // Check for recursion.
7902   subw(t, top, oopSize);
7903   ldr(t, Address(rthread, t));
7904   cmp(obj, t);
7905   br(Assembler::EQ, push);
7906 
7907   // Check header for monitor (0b10).
7908   tst(mark, markWord::monitor_value);
7909   br(Assembler::NE, slow);
7910 
7911   // Try to lock. Transition lock bits 0b01 => 0b00
7912   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7913   orr(mark, mark, markWord::unlocked_value);
7914   // Mask inline_type bit such that we go to the slow path if object is an inline type
7915   andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
7916 
7917   eor(t, mark, markWord::unlocked_value);
7918   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword, memory_order_acquire);
7919   br(Assembler::NE, slow);
7920 
7921   bind(push);
7922   // After successful lock, push object on lock-stack.
7923   str(obj, Address(rthread, top));
7924   addw(top, top, oopSize);
7925   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7926 }
7927 
7928 // Implements fast-unlocking.
7929 //
7930 // - obj: the object to be unlocked
7931 // - t1, t2, t3: temporary registers
7932 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7933 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7934   // cmpxchg clobbers rscratch1.
7935   assert_different_registers(obj, t1, t2, t3, rscratch1);
7936 
< prev index next >