< 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   }

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








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

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





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

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






















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

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















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

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

5799 void MacroAssembler::build_frame(int framesize) {





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





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





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


















































































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

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




























































































































































































































































































































































































































































































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

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



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

   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "ci/ciInlineKlass.hpp"
  30 #include "code/compiledIC.hpp"
  31 #include "compiler/compileTask.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "compiler/oopMap.hpp"
  34 #include "gc/shared/barrierSet.hpp"
  35 #include "gc/shared/barrierSetAssembler.hpp"
  36 #include "gc/shared/cardTableBarrierSet.hpp"
  37 #include "gc/shared/cardTable.hpp"
  38 #include "gc/shared/collectedHeap.hpp"
  39 #include "gc/shared/tlab_globals.hpp"
  40 #include "interpreter/bytecodeHistogram.hpp"
  41 #include "interpreter/interpreter.hpp"
  42 #include "interpreter/interpreterRuntime.hpp"
  43 #include "jvm.h"
  44 #include "memory/resourceArea.hpp"
  45 #include "memory/universe.hpp"
  46 #include "nativeInst_aarch64.hpp"
  47 #include "oops/accessDecorators.hpp"
  48 #include "oops/compressedKlass.inline.hpp"
  49 #include "oops/compressedOops.inline.hpp"
  50 #include "oops/klass.inline.hpp"
  51 #include "oops/resolvedFieldEntry.hpp"
  52 #include "runtime/arguments.hpp"
  53 #include "runtime/continuation.hpp"
  54 #include "runtime/globals.hpp"
  55 #include "runtime/icache.hpp"
  56 #include "runtime/interfaceSupport.inline.hpp"
  57 #include "runtime/javaThread.hpp"
  58 #include "runtime/jniHandles.inline.hpp"
  59 #include "runtime/sharedRuntime.hpp"
  60 #include "runtime/signature_cc.hpp"
  61 #include "runtime/stubRoutines.hpp"
  62 #include "utilities/globalDefinitions.hpp"
  63 #include "utilities/integerCast.hpp"
  64 #include "utilities/powerOfTwo.hpp"
  65 #include "vmreg_aarch64.inline.hpp"
  66 #ifdef COMPILER1
  67 #include "c1/c1_LIRAssembler.hpp"
  68 #endif
  69 #ifdef COMPILER2
  70 #include "oops/oop.hpp"
  71 #include "opto/compile.hpp"
  72 #include "opto/node.hpp"
  73 #include "opto/output.hpp"
  74 #endif
  75 
  76 #include <sys/types.h>
  77 
  78 #ifdef PRODUCT
  79 #define BLOCK_COMMENT(str) /* nothing */
  80 #else
  81 #define BLOCK_COMMENT(str) block_comment(str)
  82 #endif
  83 #define STOP(str) stop(str);
  84 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  85 

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_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2413   Label test_mark_word;
2414   // load mark word
2415   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2416   // check displaced
2417   tst(temp_reg, markWord::unlocked_value);
2418   br(Assembler::NE, test_mark_word);
2419   // slow path use klass prototype
2420   load_prototype_header(temp_reg, oop);
2421 
2422   bind(test_mark_word);
2423   andr(temp_reg, temp_reg, test_bit);
2424   if (jmp_set) {
2425     cbnz(temp_reg, jmp_label);
2426   } else {
2427     cbz(temp_reg, jmp_label);
2428   }
2429 }
2430 
2431 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2432   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2433 }
2434 
2435 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2436                                                   Label&is_non_flat_array) {
2437   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2438 }
2439 
2440 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2441   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2442 }
2443 
2444 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2445   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2446 }
2447 
2448 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2449   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2450   br(Assembler::NE, is_flat_array);
2451 }
2452 
2453 // MacroAssembler protected routines needed to implement
2454 // public methods
2455 
2456 void MacroAssembler::mov(Register r, Address dest) {
2457   code_section()->relocate(pc(), dest.rspec());
2458   uint64_t imm64 = (uint64_t)dest.target();
2459   movptr(r, imm64);
2460 }
2461 
2462 // Move a constant pointer into r.  In AArch64 mode the virtual
2463 // address space is 48 bits in size, so we only need three
2464 // instructions to create a patchable instruction sequence that can
2465 // reach anywhere.
2466 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2467 #ifndef PRODUCT
2468   {
2469     char buffer[64];
2470     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2471     block_comment(buffer);
2472   }

5167   adrp(rscratch1, src2, offset);
5168   ldr(rscratch1, Address(rscratch1, offset));
5169   cmp(src1, rscratch1);
5170 }
5171 
5172 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5173   cmp(obj1, obj2);
5174 }
5175 
5176 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5177   load_method_holder(rresult, rmethod);
5178   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5179 }
5180 
5181 void MacroAssembler::load_method_holder(Register holder, Register method) {
5182   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5183   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5184   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5185 }
5186 
5187 void MacroAssembler::load_metadata(Register dst, Register src) {
5188   if (UseCompactObjectHeaders) {
5189     load_narrow_klass_compact(dst, src);
5190   } else {
5191     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5192   }
5193 }
5194 
5195 // Loads the obj's Klass* into dst.
5196 // Preserves all registers (incl src, rscratch1 and rscratch2).
5197 // Input:
5198 // src - the oop we want to load the klass from.
5199 // dst - output narrow klass.
5200 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5201   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5202   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5203   lsr(dst, dst, markWord::klass_shift);
5204 }
5205 
5206 void MacroAssembler::load_klass(Register dst, Register src) {
5207   if (UseCompactObjectHeaders) {
5208     load_narrow_klass_compact(dst, src);
5209   } else {
5210     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5211   }
5212   decode_klass_not_null(dst);
5213 }
5214 

5266   } else {
5267     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5268   }
5269   if (CompressedKlassPointers::base() == nullptr) {
5270     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5271     return;
5272   } else if (!AOTCodeCache::is_on_for_dump() &&
5273              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5274              && CompressedKlassPointers::shift() == 0) {
5275     // Only the bottom 32 bits matter
5276     cmpw(klass, tmp);
5277     return;
5278   }
5279   decode_klass_not_null(tmp);
5280   cmp(klass, tmp);
5281 }
5282 
5283 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5284   if (UseCompactObjectHeaders) {
5285     load_narrow_klass_compact(tmp1, obj1);
5286     load_narrow_klass_compact(tmp2, obj2);
5287   } else {
5288     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5289     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5290   }
5291   cmpw(tmp1, tmp2);
5292 }
5293 
5294 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5295   load_klass(dst, src);
5296   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5297 }
5298 
5299 void MacroAssembler::store_klass(Register dst, Register src) {
5300   // FIXME: Should this be a store release?  concurrent gcs assumes
5301   // klass length is valid if klass field is not null.
5302   assert(!UseCompactObjectHeaders, "not with compact headers");
5303   encode_klass_not_null(src);
5304   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5305 }
5306 
5307 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5308   assert(!UseCompactObjectHeaders, "not with compact headers");
5309   // Store to klass gap in destination
5310   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5311 }
5312 
5313 // Algorithm must match CompressedOops::encode.
5314 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5315 #ifdef ASSERT
5316   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5317 #endif
5318   verify_oop_msg(s, "broken oop in encode_heap_oop");

5668   if (as_raw) {
5669     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5670   } else {
5671     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5672   }
5673 }
5674 
5675 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5676                                      Address dst, Register val,
5677                                      Register tmp1, Register tmp2, Register tmp3) {
5678   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5679   decorators = AccessInternal::decorator_fixup(decorators, type);
5680   bool as_raw = (decorators & AS_RAW) != 0;
5681   if (as_raw) {
5682     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5683   } else {
5684     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5685   }
5686 }
5687 
5688 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5689                                      Register inline_layout_info) {
5690   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5691   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5692 }
5693 
5694 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5695   ldr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
5696   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5697 }
5698 
5699 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5700   // ((address) (void*) o) + vk->payload_offset();
5701   Register offset = (data == oop) ? rscratch1 : data;
5702   payload_offset(inline_klass, offset);
5703   if (data == oop) {
5704     add(data, data, offset);
5705   } else {
5706     lea(data, Address(oop, offset));
5707   }
5708 }
5709 
5710 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5711                                    Register tmp2, DecoratorSet decorators) {
5712   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5713 }
5714 
5715 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5716                                             Register tmp2, DecoratorSet decorators) {
5717   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5718 }
5719 
5720 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5721                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5722   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5723 }
5724 
5725 // Used for storing nulls.
5726 void MacroAssembler::store_heap_oop_null(Address dst) {
5727   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5728 }
5729 

5804     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5805     cmp(rscratch2, rscratch1);
5806     br(Assembler::HS, next);
5807     STOP("assert(top >= start)");
5808     should_not_reach_here();
5809 
5810     bind(next);
5811     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5812     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5813     cmp(rscratch2, rscratch1);
5814     br(Assembler::HS, ok);
5815     STOP("assert(top <= end)");
5816     should_not_reach_here();
5817 
5818     bind(ok);
5819     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5820   }
5821 #endif
5822 }
5823 
5824 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5825   assert_different_registers(holder_klass, index, layout_info);
5826   InlineLayoutInfo array[2];
5827   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5828   if (is_power_of_2(size)) {
5829     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5830   } else {
5831     mov(layout_info, size);
5832     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5833   }
5834   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5835   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5836   lea(layout_info, Address(layout_info, index));
5837 }
5838 
5839 // Writes to stack successive pages until offset reached to check for
5840 // stack overflow + shadow pages.  This clobbers tmp.
5841 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5842   assert_different_registers(tmp, size, rscratch1);
5843   mov(tmp, sp);
5844   // Bang stack for total size given plus shadow page size.
5845   // Bang one page at a time because large size can bang beyond yellow and
5846   // red zones.
5847   Label loop;
5848   mov(rscratch1, (int)os::vm_page_size());
5849   bind(loop);
5850   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5851   subsw(size, size, rscratch1);
5852   str(size, Address(tmp));
5853   br(Assembler::GT, loop);
5854 
5855   // Bang down shadow pages too.
5856   // At this point, (tmp-0) is the last address touched, so don't
5857   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5858   // was post-decremented.)  Skip this address by starting at i=1, and

5925 
5926   // Strictly speaking the card table base isn't an address at all, and it might
5927   // even be negative. It is thus materialised as a constant.
5928   mov(reg, (uint64_t)ctbs->card_table_base_const());
5929 }
5930 
5931 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5932 #if INCLUDE_CDS
5933   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5934   if (AOTCodeCache::is_on_for_dump()) {
5935     // all aotrc field addresses should be registered in the AOTCodeCache address table
5936     lea(reg, ExternalAddress(a));
5937   } else {
5938     mov(reg, (uint64_t)a);
5939   }
5940 #else
5941   ShouldNotReachHere();
5942 #endif
5943 }
5944 
5945 #ifdef ASSERT
5946 void MacroAssembler::build_frame(int framesize) {
5947   build_frame(framesize, false);
5948 }
5949 #endif
5950 
5951 void MacroAssembler::build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills)) {
5952   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5953   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5954   protect_return_address();
5955   if (framesize < ((1 << 9) + 2 * wordSize)) {
5956     sub(sp, sp, framesize);
5957     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5958       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5959       stp(rscratch1, rscratch1, Address(sp, framesize - 2 * wordSize));
5960     } else {
5961       stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5962     }
5963     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5964   } else {
5965     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5966       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5967       stp(rscratch1, rscratch1, Address(pre(sp, -2 * wordSize)));
5968     } else {
5969       stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5970     }
5971     if (PreserveFramePointer) mov(rfp, sp);
5972     if (framesize < ((1 << 12) + 2 * wordSize))
5973       sub(sp, sp, framesize - 2 * wordSize);
5974     else {
5975       mov(rscratch1, framesize - 2 * wordSize);
5976       sub(sp, sp, rscratch1);
5977     }
5978   }
5979   verify_cross_modify_fence_not_required();
5980 }
5981 
5982 void MacroAssembler::remove_frame(int framesize) {
5983   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5984   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5985   if (framesize < ((1 << 9) + 2 * wordSize)) {
5986     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5987     add(sp, sp, framesize);
5988   } else {
5989     if (framesize < ((1 << 12) + 2 * wordSize))
5990       add(sp, sp, framesize - 2 * wordSize);
5991     else {
5992       mov(rscratch1, framesize - 2 * wordSize);
5993       add(sp, sp, rscratch1);
5994     }
5995     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5996   }
5997   authenticate_return_address();
5998 }
5999 
6000 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
6001   if (needs_stack_repair) {
6002     // The method has a scalarized entry point (where fields of value object arguments
6003     // are passed through registers and stack), and a non-scalarized entry point (where
6004     // value object arguments are given as oops). The non-scalarized entry point will
6005     // first load each field of value object arguments and store them in registers and on
6006     // the stack in a way compatible with the scalarized entry point. To do so, some extra
6007     // stack space might be reserved (if argument registers are not enough). On leaving the
6008     // method, this space must be freed.
6009     //
6010     // In case we used the non-scalarized entry point the stack looks like this:
6011     //
6012     // | Arguments from caller     |
6013     // |---------------------------|  <-- caller's SP
6014     // | Saved LR #1               |
6015     // | Saved FP #1               |
6016     // |---------------------------|
6017     // | Extension space for       |
6018     // |   inline arg (un)packing  |
6019     // |---------------------------|  <-- start of this method's frame
6020     // | Saved LR #2               |
6021     // | Saved FP #2               |
6022     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6023     // | sp_inc                    |
6024     // | method locals             |
6025     // |---------------------------|  <-- SP
6026     //
6027     // There are two copies of FP and LR on the stack. They will be identical at
6028     // first, but that can change.
6029     // If the caller has been deoptimized, LR #1 will be patched to point at the
6030     // deopt blob, and LR #2 will still point into the old method.
6031     // If the saved FP (x29) was not used as the frame pointer, but to store an
6032     // oop, the GC will be aware only of FP #1 as the spilled location of x29 and
6033     // will fix only this one. Overall, FP/LR #2 are not reliable and are simply
6034     // needed to add space between the extension space and the locals, as there
6035     // would be between the real arguments and the locals if we don't need to
6036     // do unpacking (from the scalarized entry point).
6037     //
6038     // When restoring, one must then load FP #1 into x29, and LR #1 into x30,
6039     // while keeping in mind that from the scalarized entry point, there will be
6040     // only one copy of each. Indeed, in the case we used the scalarized calling
6041     // convention, the stack looks like this:
6042     //
6043     // | Arguments from caller     |
6044     // |---------------------------|  <-- caller's SP / start of this method's frame
6045     // | Saved LR                  |
6046     // | Saved FP                  |
6047     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6048     // | sp_inc                    |
6049     // | method locals             |
6050     // |---------------------------|  <-- SP
6051     //
6052     // The sp_inc stack slot holds the total size of the frame including the
6053     // extension space minus two words for the saved FP and LR. That is how to
6054     // find FP/LR #1. This size is expressed in bytes. Be careful when using it
6055     // from C++ in pointer arithmetic; you might need to divide it by wordSize.
6056     //
6057     // One can find sp_inc since the start the method's frame is SP + initial_framesize.
6058 
6059     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
6060 
6061     ldr(rscratch1, Address(sp, sp_inc_offset));
6062     add(sp, sp, rscratch1);
6063     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6064   } else {
6065     remove_frame(initial_framesize);
6066   }
6067 }
6068 
6069 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
6070   int real_frame_size = frame_size + sp_inc;
6071   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
6072   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
6073   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
6074 
6075   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
6076 
6077   // Subtract two words for the saved FP and LR as these will be popped
6078   // separately. See remove_frame above.
6079   mov(rscratch1, real_frame_size - 2*wordSize);
6080   str(rscratch1, Address(sp, sp_inc_offset));
6081 }
6082 
6083 // This method counts leading positive bytes (highest bit not set) in provided byte array
6084 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6085     // Simple and most common case of aligned small array which is not at the
6086     // end of memory page is placed here. All other cases are in stub.
6087     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6088     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6089     assert_different_registers(ary1, len, result);
6090 
6091     mov(result, len);
6092     cmpw(len, 0);
6093     br(LE, DONE);
6094     cmpw(len, 4 * wordSize);
6095     br(GE, STUB_LONG); // size > 32 then go to stub
6096 
6097     int shift = 64 - exact_log2(os::vm_page_size());
6098     lsl(rscratch1, ary1, shift);
6099     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6100     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6101     br(CS, STUB); // at the end of page then go to stub

6985 // On other systems, the helper is a usual C function.
6986 //
6987 void MacroAssembler::get_thread(Register dst) {
6988   RegSet saved_regs =
6989     BSD_ONLY(RegSet::range(r0, r17) + lr - dst)
6990     NOT_BSD (RegSet::range(r0, r1)  + lr - dst);
6991 
6992   protect_return_address();
6993   push(saved_regs, sp);
6994 
6995   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6996   blr(lr);
6997   if (dst != c_rarg0) {
6998     mov(dst, c_rarg0);
6999   }
7000 
7001   pop(saved_regs, sp);
7002   authenticate_return_address();
7003 }
7004 
7005 #ifdef COMPILER2
7006 // C2 compiled method's prolog code
7007 // Moved here from aarch64.ad to support Valhalla code below
7008 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
7009   if (C->clinit_barrier_on_entry()) {
7010     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
7011 
7012     Label L_skip_barrier;
7013 
7014     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
7015     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
7016     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
7017     bind(L_skip_barrier);
7018   }
7019 
7020   if (C->max_vector_size() > 0) {
7021     reinitialize_ptrue();
7022   }
7023 
7024   int bangsize = C->output()->bang_size_in_bytes();
7025   if (C->output()->need_stack_bang(bangsize))
7026     generate_stack_overflow_check(bangsize);
7027 
7028   // n.b. frame size includes space for return pc and rfp
7029   const long framesize = C->output()->frame_size_in_bytes();
7030   build_frame(framesize DEBUG_ONLY(COMMA sp_inc != 0));
7031 
7032   if (C->needs_stack_repair()) {
7033     save_stack_increment(sp_inc, framesize);
7034   }
7035 
7036   if (VerifyStackAtCalls) {
7037     Unimplemented();
7038   }
7039 }
7040 #endif // COMPILER2
7041 
7042 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
7043   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
7044   // An inline type might be returned. If fields are in registers we
7045   // need to allocate an inline type instance and initialize it with
7046   // the value of the fields.
7047   Label skip;
7048   // We only need a new buffered inline type if a new one is not returned
7049   tbz(r0, 0, skip);
7050   int call_offset = -1;
7051 
7052   // Be careful not to clobber r1-7 which hold returned fields
7053   // Also do not use callee-saved registers as these may be live in the interpreter
7054   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
7055 
7056   // The following code is similar to the instance allocation code in TemplateTable::_new
7057   //  but has some slight differences,
7058   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
7059   // allocating is not necessary if vk != nullptr, etc.
7060   Label slow_case;
7061   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
7062   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
7063 
7064   if (vk != nullptr) {
7065     // Called from C1, where the return type is statically known.
7066     movptr(klass, (intptr_t)vk->get_InlineKlass());
7067     jint lh = vk->layout_helper();
7068     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
7069     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
7070       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
7071     } else {
7072       b(slow_case);
7073     }
7074   } else {
7075     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
7076     andr(klass, r0, -2);
7077     if (UseTLAB) {
7078       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
7079       tst(tmp2, Klass::_lh_instance_slow_path_bit);
7080       br(Assembler::NE, slow_case);
7081       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
7082     } else {
7083       b(slow_case);
7084     }
7085   }
7086   if (UseTLAB) {
7087     // 2. Initialize buffered inline instance header
7088     Register buffer_obj = r0;
7089     if (UseCompactObjectHeaders) {
7090       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7091       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7092     } else {
7093       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7094       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7095       store_klass_gap(buffer_obj, zr);
7096       if (vk == nullptr) {
7097         // store_klass corrupts klass, so save it for later use (interpreter case only).
7098         mov(tmp1, klass);
7099       }
7100       store_klass(buffer_obj, klass);
7101       klass = tmp1;
7102     }
7103     // 3. Initialize its fields with an inline class specific handler
7104     if (vk != nullptr) {
7105       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7106     } else {
7107       ldr(tmp1, Address(klass, InlineKlass::adr_members_offset()));
7108       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7109       blr(tmp1);
7110     }
7111 
7112     membar(Assembler::StoreStore);
7113     b(skip);
7114   } else {
7115     // Must have already branched to slow_case above.
7116     DEBUG_ONLY(should_not_reach_here());
7117   }
7118   bind(slow_case);
7119   // We failed to allocate a new inline type, fall back to a runtime
7120   // call. Some oop field may be live in some registers but we can't
7121   // tell. That runtime call will take care of preserving them
7122   // across a GC if there's one.
7123   mov(r0, r0_preserved);
7124 
7125   if (from_interpreter) {
7126     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
7127   } else {
7128     far_call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
7129     call_offset = offset();
7130   }
7131   membar(Assembler::StoreStore);
7132 
7133   bind(skip);
7134   return call_offset;
7135 }
7136 
7137 // Move a value between registers/stack slots and update the reg_state
7138 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7139   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7140   if (reg_state[to->value()] == reg_written) {
7141     return true; // Already written
7142   }
7143 
7144   if (from != to && bt != T_VOID) {
7145     if (reg_state[to->value()] == reg_readonly) {
7146       return false; // Not yet writable
7147     }
7148     if (from->is_reg()) {
7149       if (to->is_reg()) {
7150         if (from->is_Register() && to->is_Register()) {
7151           mov(to->as_Register(), from->as_Register());
7152         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7153           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7154         } else {
7155           ShouldNotReachHere();
7156         }
7157       } else {
7158         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7159         Address to_addr = Address(sp, st_off);
7160         if (from->is_FloatRegister()) {
7161           if (bt == T_DOUBLE) {
7162              strd(from->as_FloatRegister(), to_addr);
7163           } else {
7164              assert(bt == T_FLOAT, "must be float");
7165              strs(from->as_FloatRegister(), to_addr);
7166           }
7167         } else {
7168           str(from->as_Register(), to_addr);
7169         }
7170       }
7171     } else {
7172       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7173       if (to->is_reg()) {
7174         if (to->is_FloatRegister()) {
7175           if (bt == T_DOUBLE) {
7176             ldrd(to->as_FloatRegister(), from_addr);
7177           } else {
7178             assert(bt == T_FLOAT, "must be float");
7179             ldrs(to->as_FloatRegister(), from_addr);
7180           }
7181         } else {
7182           ldr(to->as_Register(), from_addr);
7183         }
7184       } else {
7185         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7186         ldr(rscratch1, from_addr);
7187         str(rscratch1, Address(sp, st_off));
7188       }
7189     }
7190   }
7191 
7192   // Update register states
7193   reg_state[from->value()] = reg_writable;
7194   reg_state[to->value()] = reg_written;
7195   return true;
7196 }
7197 
7198 // Calculate the extra stack space required for packing or unpacking inline
7199 // args and adjust the stack pointer
7200 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7201   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7202   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7203   assert(sp_inc > 0, "sanity");
7204 
7205   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7206   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7207 
7208   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7209   if (sp_inc < (1 << 9)) {
7210     sub(sp, sp, sp_inc);   // Fits in an immediate
7211   } else {
7212     mov(rscratch1, sp_inc);
7213     sub(sp, sp, rscratch1);
7214   }
7215 
7216   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7217 }
7218 
7219 // Read all fields from an inline type oop and store the values in registers/stack slots
7220 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7221                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7222                                           RegState reg_state[]) {
7223   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7224   assert(from->is_valid(), "source must be valid");
7225   bool progress = false;
7226 #ifdef ASSERT
7227   const int start_offset = offset();
7228 #endif
7229 
7230   Label L_null, L_notNull;
7231   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7232   Register tmp1 = r10;
7233   Register tmp2 = r11;
7234 
7235 #ifdef ASSERT
7236   RegSet clobbered_gp_regs = MacroAssembler::call_clobbered_gp_registers();
7237   assert(clobbered_gp_regs.contains(tmp1), "tmp1 must be saved explicitly if it's not a clobber");
7238   assert(clobbered_gp_regs.contains(tmp2), "tmp2 must be saved explicitly if it's not a clobber");
7239   assert(clobbered_gp_regs.contains(r14), "r14 must be saved explicitly if it's not a clobber");
7240 #endif
7241 
7242   Register fromReg = noreg;
7243   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
7244   bool done = true;
7245   bool mark_done = true;
7246   VMReg toReg;
7247   BasicType bt;
7248   // Check if argument requires a null check
7249   bool null_check = false;
7250   VMReg nullCheckReg;
7251   while (stream.next(nullCheckReg, bt)) {
7252     if (sig->at(stream.sig_index())._offset == -1) {
7253       null_check = true;
7254       break;
7255     }
7256   }
7257   stream.reset(sig_index, to_index);
7258   while (stream.next(toReg, bt)) {
7259     assert(toReg->is_valid(), "destination must be valid");
7260     int idx = (int)toReg->value();
7261     if (reg_state[idx] == reg_readonly) {
7262       if (idx != from->value()) {
7263         mark_done = false;
7264       }
7265       done = false;
7266       continue;
7267     } else if (reg_state[idx] == reg_written) {
7268       continue;
7269     }
7270     assert(reg_state[idx] == reg_writable, "must be writable");
7271     reg_state[idx] = reg_written;
7272     progress = true;
7273 
7274     if (fromReg == noreg) {
7275       if (from->is_reg()) {
7276         fromReg = from->as_Register();
7277       } else {
7278         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7279         ldr(tmp1, Address(sp, st_off));
7280         fromReg = tmp1;
7281       }
7282       if (null_check) {
7283         // Nullable inline type argument, emit null check
7284         cbz(fromReg, L_null);
7285       }
7286     }
7287     int off = sig->at(stream.sig_index())._offset;
7288     if (off == -1) {
7289       assert(null_check, "Missing null check at");
7290       if (toReg->is_stack()) {
7291         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7292         mov(tmp2, 1);
7293         str(tmp2, Address(sp, st_off));
7294       } else {
7295         mov(toReg->as_Register(), 1);
7296       }
7297       continue;
7298     }
7299     if (sig->at(stream.sig_index())._vt_oop) {
7300       if (toReg->is_stack()) {
7301         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7302         str(fromReg, Address(sp, st_off));
7303       } else {
7304         mov(toReg->as_Register(), fromReg);
7305       }
7306       continue;
7307     }
7308     assert(off > 0, "offset in object should be positive");
7309     Address fromAddr = Address(fromReg, off);
7310     if (!toReg->is_FloatRegister()) {
7311       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7312       if (is_reference_type(bt)) {
7313         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7314       } else {
7315         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7316         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7317       }
7318       if (toReg->is_stack()) {
7319         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7320         str(dst, Address(sp, st_off));
7321       }
7322     } else if (bt == T_DOUBLE) {
7323       ldrd(toReg->as_FloatRegister(), fromAddr);
7324     } else {
7325       assert(bt == T_FLOAT, "must be float");
7326       ldrs(toReg->as_FloatRegister(), fromAddr);
7327     }
7328   }
7329   if (progress && null_check) {
7330     if (done) {
7331       b(L_notNull);
7332       bind(L_null);
7333       // Set null marker to zero to signal that the argument is null.
7334       // Also set all fields to zero since the runtime requires a canonical
7335       // representation of a flat null.
7336       stream.reset(sig_index, to_index);
7337       while (stream.next(toReg, bt)) {
7338         if (toReg->is_stack()) {
7339           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7340           str(zr, Address(sp, st_off));
7341         } else if (toReg->is_FloatRegister()) {
7342           mov(toReg->as_FloatRegister(), T2S, 0);
7343         } else {
7344           mov(toReg->as_Register(), zr);
7345         }
7346       }
7347       bind(L_notNull);
7348     } else {
7349       bind(L_null);
7350     }
7351   }
7352 
7353   sig_index = stream.sig_index();
7354   to_index = stream.regs_index();
7355 
7356   if (mark_done && reg_state[from->value()] != reg_written) {
7357     // This is okay because no one else will write to that slot
7358     reg_state[from->value()] = reg_writable;
7359   }
7360   from_index--;
7361   assert(progress || (start_offset == offset()), "should not emit code");
7362   return done;
7363 }
7364 
7365 // Pack fields back into an inline type oop
7366 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7367                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7368                                         RegState reg_state[], Register val_array) {
7369   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7370   assert(to->is_valid(), "destination must be valid");
7371 
7372   if (reg_state[to->value()] == reg_written) {
7373     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7374     return true; // Already written
7375   }
7376 
7377   // The GC barrier expanded by store_heap_oop below may call into the
7378   // runtime so use callee-saved registers for any values that need to be
7379   // preserved. The GC barrier assembler should take care of saving the
7380   // Java argument registers.
7381   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7382   Register val_obj_tmp = r21;
7383   Register from_reg_tmp = r22;
7384   Register tmp1 = r14;
7385   Register tmp2 = r13;
7386   Register tmp3 = r12;
7387   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7388 
7389   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7390 
7391   if (reg_state[to->value()] == reg_readonly) {
7392     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7393       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7394       return false; // Not yet writable
7395     }
7396     val_obj = val_obj_tmp;
7397   }
7398 
7399   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7400   VMReg fromReg;
7401   BasicType bt;
7402   Label L_null;
7403   while (stream.next(fromReg, bt)) {
7404     assert(fromReg->is_valid(), "source must be valid");
7405     reg_state[fromReg->value()] = reg_writable;
7406 
7407     int off = sig->at(stream.sig_index())._offset;
7408     if (off == -1) {
7409       // Nullable inline type argument, emit null check
7410       Label L_notNull;
7411       if (fromReg->is_stack()) {
7412         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7413         ldrb(tmp2, Address(sp, ld_off));
7414         cbnz(tmp2, L_notNull);
7415       } else {
7416         cbnz(fromReg->as_Register(), L_notNull);
7417       }
7418       mov(val_obj, 0);
7419       b(L_null);
7420       bind(L_notNull);
7421       continue;
7422     }
7423     if (sig->at(stream.sig_index())._vt_oop) {
7424       if (fromReg->is_stack()) {
7425         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7426         ldr(val_obj, Address(sp, ld_off));
7427       } else {
7428         mov(val_obj, fromReg->as_Register());
7429       }
7430       cbnz(val_obj, L_null);
7431       // get the buffer from the just allocated pool of buffers
7432       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7433       load_heap_oop(val_obj, Address(val_array, index), rscratch1, rscratch2);
7434       continue;
7435     }
7436 
7437     assert(off > 0, "offset in object should be positive");
7438     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7439 
7440     // Pack the scalarized field into the value object.
7441     Address dst(val_obj, off);
7442     if (!fromReg->is_FloatRegister()) {
7443       Register src;
7444       if (fromReg->is_stack()) {
7445         src = from_reg_tmp;
7446         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7447         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7448       } else {
7449         src = fromReg->as_Register();
7450       }
7451       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7452       if (is_reference_type(bt)) {
7453         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
7454         mov(tmp3, val_obj);
7455         Address dst_with_tmp3(tmp3, off);
7456         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7457       } else {
7458         store_sized_value(dst, src, size_in_bytes);
7459       }
7460     } else if (bt == T_DOUBLE) {
7461       strd(fromReg->as_FloatRegister(), dst);
7462     } else {
7463       assert(bt == T_FLOAT, "must be float");
7464       strs(fromReg->as_FloatRegister(), dst);
7465     }
7466   }
7467   bind(L_null);
7468   sig_index = stream.sig_index();
7469   from_index = stream.regs_index();
7470 
7471   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7472   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7473   assert(success, "to register must be writeable");
7474   return true;
7475 }
7476 
7477 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7478   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7479 }
7480 
7481 void MacroAssembler::cache_wb(Address line) {
7482   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7483   assert(line.index() == noreg, "index should be noreg");
7484   assert(line.offset() == 0, "offset should be 0");
7485   // would like to assert this
7486   // assert(line._ext.shift == 0, "shift should be zero");
7487   if (VM_Version::supports_dcpop()) {
7488     // writeback using clear virtual address to point of persistence
7489     dc(Assembler::CVAP, line.base());
7490   } else {
7491     // no need to generate anything as Unsafe.writebackMemory should
7492     // never invoke this stub
7493   }
7494 }
7495 
7496 void MacroAssembler::cache_wbsync(bool is_pre) {
7497   // we only need a barrier post sync
7498   if (!is_pre) {
7499     membar(Assembler::AnyAny);
7500   }

7896   }
7897 
7898   // Check if the lock-stack is full.
7899   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7900   cmpw(top, (unsigned)LockStack::end_offset());
7901   br(Assembler::GE, slow);
7902 
7903   // Check for recursion.
7904   subw(t, top, oopSize);
7905   ldr(t, Address(rthread, t));
7906   cmp(obj, t);
7907   br(Assembler::EQ, push);
7908 
7909   // Check header for monitor (0b10).
7910   tst(mark, markWord::monitor_value);
7911   br(Assembler::NE, slow);
7912 
7913   // Try to lock. Transition lock bits 0b01 => 0b00
7914   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7915   orr(mark, mark, markWord::unlocked_value);
7916   // Mask inline_type bit such that we go to the slow path if object is an inline type
7917   andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
7918 
7919   eor(t, mark, markWord::unlocked_value);
7920   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7921           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7922   br(Assembler::NE, slow);
7923 
7924   bind(push);
7925   // After successful lock, push object on lock-stack.
7926   str(obj, Address(rthread, top));
7927   addw(top, top, oopSize);
7928   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7929 }
7930 
7931 // Implements fast-unlocking.
7932 //
7933 // - obj: the object to be unlocked
7934 // - t1, t2, t3: temporary registers
7935 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7936 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7937   // cmpxchg clobbers rscratch1.
7938   assert_different_registers(obj, t1, t2, t3, rscratch1);
< prev index next >