< 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 

1991   ldarb(scratch, scratch);
1992   cmp(scratch, InstanceKlass::fully_initialized);
1993   br(Assembler::EQ, *L_fast_path);
1994 
1995   // Fast path check: current thread is initializer thread
1996   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
1997   cmp(rthread, scratch);
1998 
1999   if (L_slow_path == &L_fallthrough) {
2000     br(Assembler::EQ, *L_fast_path);
2001     bind(*L_slow_path);
2002   } else if (L_fast_path == &L_fallthrough) {
2003     br(Assembler::NE, *L_slow_path);
2004     bind(*L_fast_path);
2005   } else {
2006     Unimplemented();
2007   }
2008 }
2009 
2010 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2011   if (!VerifyOops) return;




2012 
2013   // Pass register number to verify_oop_subroutine
2014   const char* b = nullptr;
2015   {
2016     ResourceMark rm;
2017     stringStream ss;
2018     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2019     b = code_string(ss.as_string());
2020   }
2021   BLOCK_COMMENT("verify_oop {");
2022 
2023   strip_return_address(); // This might happen within a stack frame.
2024   protect_return_address();
2025   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2026   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2027 
2028   mov(r0, reg);
2029   movptr(rscratch1, (uintptr_t)(address)b);
2030 
2031   // call indirectly to solve generation ordering problem
2032   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2033   ldr(rscratch2, Address(rscratch2));
2034   blr(rscratch2);
2035 
2036   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2037   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2038   authenticate_return_address();
2039 
2040   BLOCK_COMMENT("} verify_oop");
2041 }
2042 
2043 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2044   if (!VerifyOops) return;




2045 
2046   const char* b = nullptr;
2047   {
2048     ResourceMark rm;
2049     stringStream ss;
2050     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2051     b = code_string(ss.as_string());
2052   }
2053   BLOCK_COMMENT("verify_oop_addr {");
2054 
2055   strip_return_address(); // This might happen within a stack frame.
2056   protect_return_address();
2057   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2058   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2059 
2060   // addr may contain sp so we will have to adjust it based on the
2061   // pushes that we just did.
2062   if (addr.uses(sp)) {
2063     lea(r0, addr);
2064     ldr(r0, Address(r0, 4 * wordSize));

2274   call_VM_leaf_base(entry_point, 1);
2275 }
2276 
2277 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2278   assert_different_registers(arg_1, c_rarg0);
2279   pass_arg0(this, arg_0);
2280   pass_arg1(this, arg_1);
2281   call_VM_leaf_base(entry_point, 2);
2282 }
2283 
2284 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2285                                   Register arg_1, Register arg_2) {
2286   assert_different_registers(arg_1, c_rarg0);
2287   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2288   pass_arg0(this, arg_0);
2289   pass_arg1(this, arg_1);
2290   pass_arg2(this, arg_2);
2291   call_VM_leaf_base(entry_point, 3);
2292 }
2293 




2294 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2295   pass_arg0(this, arg_0);
2296   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2297 }
2298 
2299 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2300 
2301   assert_different_registers(arg_0, c_rarg1);
2302   pass_arg1(this, arg_1);
2303   pass_arg0(this, arg_0);
2304   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2305 }
2306 
2307 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2308   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2309   assert_different_registers(arg_1, c_rarg2);
2310   pass_arg2(this, arg_2);
2311   pass_arg1(this, arg_1);
2312   pass_arg0(this, arg_0);
2313   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2319   assert_different_registers(arg_2, c_rarg3);
2320   pass_arg3(this, arg_3);
2321   pass_arg2(this, arg_2);
2322   pass_arg1(this, arg_1);
2323   pass_arg0(this, arg_0);
2324   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2325 }
2326 
2327 void MacroAssembler::null_check(Register reg, int offset) {
2328   if (needs_explicit_null_check(offset)) {
2329     // provoke OS null exception if reg is null by
2330     // accessing M[reg] w/o changing any registers
2331     // NOTE: this is plenty to provoke a segv
2332     ldr(zr, Address(reg));
2333   } else {
2334     // nothing to do, (later) access of M[reg + offset]
2335     // will provoke OS null exception if reg is null
2336   }
2337 }
2338 














































































2339 // MacroAssembler protected routines needed to implement
2340 // public methods
2341 
2342 void MacroAssembler::mov(Register r, Address dest) {
2343   code_section()->relocate(pc(), dest.rspec());
2344   uint64_t imm64 = (uint64_t)dest.target();
2345   movptr(r, imm64);
2346 }
2347 
2348 // Move a constant pointer into r.  In AArch64 mode the virtual
2349 // address space is 48 bits in size, so we only need three
2350 // instructions to create a patchable instruction sequence that can
2351 // reach anywhere.
2352 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2353 #ifndef PRODUCT
2354   {
2355     char buffer[64];
2356     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2357     block_comment(buffer);
2358   }

5057   adrp(rscratch1, src2, offset);
5058   ldr(rscratch1, Address(rscratch1, offset));
5059   cmp(src1, rscratch1);
5060 }
5061 
5062 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5063   cmp(obj1, obj2);
5064 }
5065 
5066 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5067   load_method_holder(rresult, rmethod);
5068   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5069 }
5070 
5071 void MacroAssembler::load_method_holder(Register holder, Register method) {
5072   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5073   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5074   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5075 }
5076 








5077 // Loads the obj's Klass* into dst.
5078 // Preserves all registers (incl src, rscratch1 and rscratch2).
5079 // Input:
5080 // src - the oop we want to load the klass from.
5081 // dst - output narrow klass.
5082 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5083   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5084   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5085   lsr(dst, dst, markWord::klass_shift);
5086 }
5087 
5088 void MacroAssembler::load_klass(Register dst, Register src) {
5089   if (UseCompactObjectHeaders) {
5090     load_narrow_klass_compact(dst, src);
5091   } else {
5092     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5093   }
5094   decode_klass_not_null(dst);
5095 }
5096 

5148   } else {
5149     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5150   }
5151   if (CompressedKlassPointers::base() == nullptr) {
5152     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5153     return;
5154   } else if (!AOTCodeCache::is_on_for_dump() &&
5155              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5156              && CompressedKlassPointers::shift() == 0) {
5157     // Only the bottom 32 bits matter
5158     cmpw(klass, tmp);
5159     return;
5160   }
5161   decode_klass_not_null(tmp);
5162   cmp(klass, tmp);
5163 }
5164 
5165 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5166   if (UseCompactObjectHeaders) {
5167     load_narrow_klass_compact(tmp1, obj1);
5168     load_narrow_klass_compact(tmp2,  obj2);
5169   } else {
5170     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5171     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5172   }
5173   cmpw(tmp1, tmp2);
5174 }
5175 





5176 void MacroAssembler::store_klass(Register dst, Register src) {
5177   // FIXME: Should this be a store release?  concurrent gcs assumes
5178   // klass length is valid if klass field is not null.
5179   assert(!UseCompactObjectHeaders, "not with compact headers");
5180   encode_klass_not_null(src);
5181   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5182 }
5183 
5184 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5185   assert(!UseCompactObjectHeaders, "not with compact headers");
5186   // Store to klass gap in destination
5187   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5188 }
5189 
5190 // Algorithm must match CompressedOops::encode.
5191 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5192 #ifdef ASSERT
5193   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5194 #endif
5195   verify_oop_msg(s, "broken oop in encode_heap_oop");

5545   if (as_raw) {
5546     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5547   } else {
5548     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5549   }
5550 }
5551 
5552 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5553                                      Address dst, Register val,
5554                                      Register tmp1, Register tmp2, Register tmp3) {
5555   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5556   decorators = AccessInternal::decorator_fixup(decorators, type);
5557   bool as_raw = (decorators & AS_RAW) != 0;
5558   if (as_raw) {
5559     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5560   } else {
5561     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5562   }
5563 }
5564 






















5565 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5566                                    Register tmp2, DecoratorSet decorators) {
5567   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5568 }
5569 
5570 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5571                                             Register tmp2, DecoratorSet decorators) {
5572   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5573 }
5574 
5575 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5576                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5577   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5578 }
5579 
5580 // Used for storing nulls.
5581 void MacroAssembler::store_heap_oop_null(Address dst) {
5582   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5583 }
5584 

5659     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5660     cmp(rscratch2, rscratch1);
5661     br(Assembler::HS, next);
5662     STOP("assert(top >= start)");
5663     should_not_reach_here();
5664 
5665     bind(next);
5666     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5667     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5668     cmp(rscratch2, rscratch1);
5669     br(Assembler::HS, ok);
5670     STOP("assert(top <= end)");
5671     should_not_reach_here();
5672 
5673     bind(ok);
5674     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5675   }
5676 #endif
5677 }
5678 















5679 // Writes to stack successive pages until offset reached to check for
5680 // stack overflow + shadow pages.  This clobbers tmp.
5681 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5682   assert_different_registers(tmp, size, rscratch1);
5683   mov(tmp, sp);
5684   // Bang stack for total size given plus shadow page size.
5685   // Bang one page at a time because large size can bang beyond yellow and
5686   // red zones.
5687   Label loop;
5688   mov(rscratch1, (int)os::vm_page_size());
5689   bind(loop);
5690   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5691   subsw(size, size, rscratch1);
5692   str(size, Address(tmp));
5693   br(Assembler::GT, loop);
5694 
5695   // Bang down shadow pages too.
5696   // At this point, (tmp-0) is the last address touched, so don't
5697   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5698   // was post-decremented.)  Skip this address by starting at i=1, and

5765 
5766   // Strictly speaking the card table base isn't an address at all, and it might
5767   // even be negative. It is thus materialised as a constant.
5768   mov(reg, (uint64_t)ctbs->card_table_base_const());
5769 }
5770 
5771 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5772 #if INCLUDE_CDS
5773   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5774   if (AOTCodeCache::is_on_for_dump()) {
5775     // all aotrc field addresses should be registered in the AOTCodeCache address table
5776     lea(reg, ExternalAddress(a));
5777   } else {
5778     mov(reg, (uint64_t)a);
5779   }
5780 #else
5781   ShouldNotReachHere();
5782 #endif
5783 }
5784 

5785 void MacroAssembler::build_frame(int framesize) {





5786   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5787   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5788   protect_return_address();
5789   if (framesize < ((1 << 9) + 2 * wordSize)) {
5790     sub(sp, sp, framesize);
5791     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));





5792     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5793   } else {
5794     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));





5795     if (PreserveFramePointer) mov(rfp, sp);
5796     if (framesize < ((1 << 12) + 2 * wordSize))
5797       sub(sp, sp, framesize - 2 * wordSize);
5798     else {
5799       mov(rscratch1, framesize - 2 * wordSize);
5800       sub(sp, sp, rscratch1);
5801     }
5802   }
5803   verify_cross_modify_fence_not_required();
5804 }
5805 
5806 void MacroAssembler::remove_frame(int framesize) {
5807   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5808   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5809   if (framesize < ((1 << 9) + 2 * wordSize)) {
5810     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5811     add(sp, sp, framesize);
5812   } else {
5813     if (framesize < ((1 << 12) + 2 * wordSize))
5814       add(sp, sp, framesize - 2 * wordSize);
5815     else {
5816       mov(rscratch1, framesize - 2 * wordSize);
5817       add(sp, sp, rscratch1);
5818     }
5819     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5820   }
5821   authenticate_return_address();
5822 }
5823 


















































































5824 
5825 // This method counts leading positive bytes (highest bit not set) in provided byte array
5826 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5827     // Simple and most common case of aligned small array which is not at the
5828     // end of memory page is placed here. All other cases are in stub.
5829     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5830     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5831     assert_different_registers(ary1, len, result);
5832 
5833     mov(result, len);
5834     cmpw(len, 0);
5835     br(LE, DONE);
5836     cmpw(len, 4 * wordSize);
5837     br(GE, STUB_LONG); // size > 32 then go to stub
5838 
5839     int shift = 64 - exact_log2(os::vm_page_size());
5840     lsl(rscratch1, ary1, shift);
5841     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5842     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5843     br(CS, STUB); // at the end of page then go to stub

6727 // On other systems, the helper is a usual C function.
6728 //
6729 void MacroAssembler::get_thread(Register dst) {
6730   RegSet saved_regs =
6731     BSD_ONLY(RegSet::range(r0, r17) + lr - dst)
6732     NOT_BSD (RegSet::range(r0, r1)  + lr - dst);
6733 
6734   protect_return_address();
6735   push(saved_regs, sp);
6736 
6737   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6738   blr(lr);
6739   if (dst != c_rarg0) {
6740     mov(dst, c_rarg0);
6741   }
6742 
6743   pop(saved_regs, sp);
6744   authenticate_return_address();
6745 }
6746 




























































































































































































































































































































































































































































































6747 void MacroAssembler::cache_wb(Address line) {
6748   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6749   assert(line.index() == noreg, "index should be noreg");
6750   assert(line.offset() == 0, "offset should be 0");
6751   // would like to assert this
6752   // assert(line._ext.shift == 0, "shift should be zero");
6753   if (VM_Version::supports_dcpop()) {
6754     // writeback using clear virtual address to point of persistence
6755     dc(Assembler::CVAP, line.base());
6756   } else {
6757     // no need to generate anything as Unsafe.writebackMemory should
6758     // never invoke this stub
6759   }
6760 }
6761 
6762 void MacroAssembler::cache_wbsync(bool is_pre) {
6763   // we only need a barrier post sync
6764   if (!is_pre) {
6765     membar(Assembler::AnyAny);
6766   }

7162   }
7163 
7164   // Check if the lock-stack is full.
7165   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7166   cmpw(top, (unsigned)LockStack::end_offset());
7167   br(Assembler::GE, slow);
7168 
7169   // Check for recursion.
7170   subw(t, top, oopSize);
7171   ldr(t, Address(rthread, t));
7172   cmp(obj, t);
7173   br(Assembler::EQ, push);
7174 
7175   // Check header for monitor (0b10).
7176   tst(mark, markWord::monitor_value);
7177   br(Assembler::NE, slow);
7178 
7179   // Try to lock. Transition lock bits 0b01 => 0b00
7180   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7181   orr(mark, mark, markWord::unlocked_value);



7182   eor(t, mark, markWord::unlocked_value);
7183   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7184           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7185   br(Assembler::NE, slow);
7186 
7187   bind(push);
7188   // After successful lock, push object on lock-stack.
7189   str(obj, Address(rthread, top));
7190   addw(top, top, oopSize);
7191   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7192 }
7193 
7194 // Implements fast-unlocking.
7195 //
7196 // - obj: the object to be unlocked
7197 // - t1, t2, t3: temporary registers
7198 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7199 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7200   // cmpxchg clobbers rscratch1.
7201   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 

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

2288   call_VM_leaf_base(entry_point, 1);
2289 }
2290 
2291 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2292   assert_different_registers(arg_1, c_rarg0);
2293   pass_arg0(this, arg_0);
2294   pass_arg1(this, arg_1);
2295   call_VM_leaf_base(entry_point, 2);
2296 }
2297 
2298 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2299                                   Register arg_1, Register arg_2) {
2300   assert_different_registers(arg_1, c_rarg0);
2301   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2302   pass_arg0(this, arg_0);
2303   pass_arg1(this, arg_1);
2304   pass_arg2(this, arg_2);
2305   call_VM_leaf_base(entry_point, 3);
2306 }
2307 
2308 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2309   MacroAssembler::call_VM_leaf_base(entry_point, 1);
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 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2358   assert_different_registers(markword, rscratch2);
2359   mov(rscratch2, markWord::inline_type_pattern_mask);
2360   andr(markword, markword, rscratch2);
2361   mov(rscratch2, markWord::inline_type_pattern);
2362   cmp(markword, rscratch2);
2363   br(Assembler::EQ, is_inline_type);
2364 }
2365 
2366 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
2367   assert_different_registers(tmp, rscratch1);
2368   if (can_be_null) {
2369     cbz(object, not_inline_type);
2370   }
2371   const int is_inline_type_mask = markWord::inline_type_pattern;
2372   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2373   mov(rscratch1, is_inline_type_mask);
2374   andr(tmp, tmp, rscratch1);
2375   cmp(tmp, rscratch1);
2376   br(Assembler::NE, not_inline_type);
2377 }
2378 
2379 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2380   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2381   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2382 }
2383 
2384 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2385   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2386   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2387 }
2388 
2389 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2390   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2391   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2392 }
2393 
2394 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2395   Label test_mark_word;
2396   // load mark word
2397   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2398   // check displaced
2399   tst(temp_reg, markWord::unlocked_value);
2400   br(Assembler::NE, test_mark_word);
2401   // slow path use klass prototype
2402   load_prototype_header(temp_reg, oop);
2403 
2404   bind(test_mark_word);
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   }

5153   adrp(rscratch1, src2, offset);
5154   ldr(rscratch1, Address(rscratch1, offset));
5155   cmp(src1, rscratch1);
5156 }
5157 
5158 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5159   cmp(obj1, obj2);
5160 }
5161 
5162 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5163   load_method_holder(rresult, rmethod);
5164   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5165 }
5166 
5167 void MacroAssembler::load_method_holder(Register holder, Register method) {
5168   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5169   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5170   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5171 }
5172 
5173 void MacroAssembler::load_metadata(Register dst, Register src) {
5174   if (UseCompactObjectHeaders) {
5175     load_narrow_klass_compact(dst, src);
5176   } else {
5177     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5178   }
5179 }
5180 
5181 // Loads the obj's Klass* into dst.
5182 // Preserves all registers (incl src, rscratch1 and rscratch2).
5183 // Input:
5184 // src - the oop we want to load the klass from.
5185 // dst - output narrow klass.
5186 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5187   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5188   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5189   lsr(dst, dst, markWord::klass_shift);
5190 }
5191 
5192 void MacroAssembler::load_klass(Register dst, Register src) {
5193   if (UseCompactObjectHeaders) {
5194     load_narrow_klass_compact(dst, src);
5195   } else {
5196     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5197   }
5198   decode_klass_not_null(dst);
5199 }
5200 

5252   } else {
5253     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5254   }
5255   if (CompressedKlassPointers::base() == nullptr) {
5256     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5257     return;
5258   } else if (!AOTCodeCache::is_on_for_dump() &&
5259              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5260              && CompressedKlassPointers::shift() == 0) {
5261     // Only the bottom 32 bits matter
5262     cmpw(klass, tmp);
5263     return;
5264   }
5265   decode_klass_not_null(tmp);
5266   cmp(klass, tmp);
5267 }
5268 
5269 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5270   if (UseCompactObjectHeaders) {
5271     load_narrow_klass_compact(tmp1, obj1);
5272     load_narrow_klass_compact(tmp2, obj2);
5273   } else {
5274     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5275     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5276   }
5277   cmpw(tmp1, tmp2);
5278 }
5279 
5280 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5281   load_klass(dst, src);
5282   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5283 }
5284 
5285 void MacroAssembler::store_klass(Register dst, Register src) {
5286   // FIXME: Should this be a store release?  concurrent gcs assumes
5287   // klass length is valid if klass field is not null.
5288   assert(!UseCompactObjectHeaders, "not with compact headers");
5289   encode_klass_not_null(src);
5290   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5291 }
5292 
5293 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5294   assert(!UseCompactObjectHeaders, "not with compact headers");
5295   // Store to klass gap in destination
5296   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5297 }
5298 
5299 // Algorithm must match CompressedOops::encode.
5300 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5301 #ifdef ASSERT
5302   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5303 #endif
5304   verify_oop_msg(s, "broken oop in encode_heap_oop");

5654   if (as_raw) {
5655     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5656   } else {
5657     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5658   }
5659 }
5660 
5661 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5662                                      Address dst, Register val,
5663                                      Register tmp1, Register tmp2, Register tmp3) {
5664   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5665   decorators = AccessInternal::decorator_fixup(decorators, type);
5666   bool as_raw = (decorators & AS_RAW) != 0;
5667   if (as_raw) {
5668     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5669   } else {
5670     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5671   }
5672 }
5673 
5674 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5675                                      Register inline_layout_info) {
5676   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5677   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5678 }
5679 
5680 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5681   ldr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
5682   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5683 }
5684 
5685 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5686   // ((address) (void*) o) + vk->payload_offset();
5687   Register offset = (data == oop) ? rscratch1 : data;
5688   payload_offset(inline_klass, offset);
5689   if (data == oop) {
5690     add(data, data, offset);
5691   } else {
5692     lea(data, Address(oop, offset));
5693   }
5694 }
5695 
5696 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5697                                    Register tmp2, DecoratorSet decorators) {
5698   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5699 }
5700 
5701 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5702                                             Register tmp2, DecoratorSet decorators) {
5703   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5704 }
5705 
5706 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5707                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5708   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5709 }
5710 
5711 // Used for storing nulls.
5712 void MacroAssembler::store_heap_oop_null(Address dst) {
5713   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5714 }
5715 

5790     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5791     cmp(rscratch2, rscratch1);
5792     br(Assembler::HS, next);
5793     STOP("assert(top >= start)");
5794     should_not_reach_here();
5795 
5796     bind(next);
5797     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5798     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5799     cmp(rscratch2, rscratch1);
5800     br(Assembler::HS, ok);
5801     STOP("assert(top <= end)");
5802     should_not_reach_here();
5803 
5804     bind(ok);
5805     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5806   }
5807 #endif
5808 }
5809 
5810 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5811   assert_different_registers(holder_klass, index, layout_info);
5812   InlineLayoutInfo array[2];
5813   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5814   if (is_power_of_2(size)) {
5815     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5816   } else {
5817     mov(layout_info, size);
5818     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5819   }
5820   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5821   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5822   lea(layout_info, Address(layout_info, index));
5823 }
5824 
5825 // Writes to stack successive pages until offset reached to check for
5826 // stack overflow + shadow pages.  This clobbers tmp.
5827 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5828   assert_different_registers(tmp, size, rscratch1);
5829   mov(tmp, sp);
5830   // Bang stack for total size given plus shadow page size.
5831   // Bang one page at a time because large size can bang beyond yellow and
5832   // red zones.
5833   Label loop;
5834   mov(rscratch1, (int)os::vm_page_size());
5835   bind(loop);
5836   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5837   subsw(size, size, rscratch1);
5838   str(size, Address(tmp));
5839   br(Assembler::GT, loop);
5840 
5841   // Bang down shadow pages too.
5842   // At this point, (tmp-0) is the last address touched, so don't
5843   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5844   // was post-decremented.)  Skip this address by starting at i=1, and

5911 
5912   // Strictly speaking the card table base isn't an address at all, and it might
5913   // even be negative. It is thus materialised as a constant.
5914   mov(reg, (uint64_t)ctbs->card_table_base_const());
5915 }
5916 
5917 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5918 #if INCLUDE_CDS
5919   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5920   if (AOTCodeCache::is_on_for_dump()) {
5921     // all aotrc field addresses should be registered in the AOTCodeCache address table
5922     lea(reg, ExternalAddress(a));
5923   } else {
5924     mov(reg, (uint64_t)a);
5925   }
5926 #else
5927   ShouldNotReachHere();
5928 #endif
5929 }
5930 
5931 #ifdef ASSERT
5932 void MacroAssembler::build_frame(int framesize) {
5933   build_frame(framesize, false);
5934 }
5935 #endif
5936 
5937 void MacroAssembler::build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills)) {
5938   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5939   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5940   protect_return_address();
5941   if (framesize < ((1 << 9) + 2 * wordSize)) {
5942     sub(sp, sp, framesize);
5943     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5944       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5945       stp(rscratch1, rscratch1, Address(sp, framesize - 2 * wordSize));
5946     } else {
5947       stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5948     }
5949     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5950   } else {
5951     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5952       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5953       stp(rscratch1, rscratch1, Address(pre(sp, -2 * wordSize)));
5954     } else {
5955       stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5956     }
5957     if (PreserveFramePointer) mov(rfp, sp);
5958     if (framesize < ((1 << 12) + 2 * wordSize))
5959       sub(sp, sp, framesize - 2 * wordSize);
5960     else {
5961       mov(rscratch1, framesize - 2 * wordSize);
5962       sub(sp, sp, rscratch1);
5963     }
5964   }
5965   verify_cross_modify_fence_not_required();
5966 }
5967 
5968 void MacroAssembler::remove_frame(int framesize) {
5969   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5970   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5971   if (framesize < ((1 << 9) + 2 * wordSize)) {
5972     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5973     add(sp, sp, framesize);
5974   } else {
5975     if (framesize < ((1 << 12) + 2 * wordSize))
5976       add(sp, sp, framesize - 2 * wordSize);
5977     else {
5978       mov(rscratch1, framesize - 2 * wordSize);
5979       add(sp, sp, rscratch1);
5980     }
5981     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5982   }
5983   authenticate_return_address();
5984 }
5985 
5986 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
5987   if (needs_stack_repair) {
5988     // The method has a scalarized entry point (where fields of value object arguments
5989     // are passed through registers and stack), and a non-scalarized entry point (where
5990     // value object arguments are given as oops). The non-scalarized entry point will
5991     // first load each field of value object arguments and store them in registers and on
5992     // the stack in a way compatible with the scalarized entry point. To do so, some extra
5993     // stack space might be reserved (if argument registers are not enough). On leaving the
5994     // method, this space must be freed.
5995     //
5996     // In case we used the non-scalarized entry point the stack looks like this:
5997     //
5998     // | Arguments from caller     |
5999     // |---------------------------|  <-- caller's SP
6000     // | Saved LR #1               |
6001     // | Saved FP #1               |
6002     // |---------------------------|
6003     // | Extension space for       |
6004     // |   inline arg (un)packing  |
6005     // |---------------------------|  <-- start of this method's frame
6006     // | Saved LR #2               |
6007     // | Saved FP #2               |
6008     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6009     // | sp_inc                    |
6010     // | method locals             |
6011     // |---------------------------|  <-- SP
6012     //
6013     // There are two copies of FP and LR on the stack. They will be identical at
6014     // first, but that can change.
6015     // If the caller has been deoptimized, LR #1 will be patched to point at the
6016     // deopt blob, and LR #2 will still point into the old method.
6017     // If the saved FP (x29) was not used as the frame pointer, but to store an
6018     // oop, the GC will be aware only of FP #1 as the spilled location of x29 and
6019     // will fix only this one. Overall, FP/LR #2 are not reliable and are simply
6020     // needed to add space between the extension space and the locals, as there
6021     // would be between the real arguments and the locals if we don't need to
6022     // do unpacking (from the scalarized entry point).
6023     //
6024     // When restoring, one must then load FP #1 into x29, and LR #1 into x30,
6025     // while keeping in mind that from the scalarized entry point, there will be
6026     // only one copy of each. Indeed, in the case we used the scalarized calling
6027     // convention, the stack looks like this:
6028     //
6029     // | Arguments from caller     |
6030     // |---------------------------|  <-- caller's SP / start of this method's frame
6031     // | Saved LR                  |
6032     // | Saved FP                  |
6033     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6034     // | sp_inc                    |
6035     // | method locals             |
6036     // |---------------------------|  <-- SP
6037     //
6038     // The sp_inc stack slot holds the total size of the frame including the
6039     // extension space minus two words for the saved FP and LR. That is how to
6040     // find FP/LR #1. This size is expressed in bytes. Be careful when using it
6041     // from C++ in pointer arithmetic; you might need to divide it by wordSize.
6042     //
6043     // One can find sp_inc since the start the method's frame is SP + initial_framesize.
6044 
6045     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
6046 
6047     ldr(rscratch1, Address(sp, sp_inc_offset));
6048     add(sp, sp, rscratch1);
6049     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6050   } else {
6051     remove_frame(initial_framesize);
6052   }
6053 }
6054 
6055 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
6056   int real_frame_size = frame_size + sp_inc;
6057   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
6058   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
6059   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
6060 
6061   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
6062 
6063   // Subtract two words for the saved FP and LR as these will be popped
6064   // separately. See remove_frame above.
6065   mov(rscratch1, real_frame_size - 2*wordSize);
6066   str(rscratch1, Address(sp, sp_inc_offset));
6067 }
6068 
6069 // This method counts leading positive bytes (highest bit not set) in provided byte array
6070 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6071     // Simple and most common case of aligned small array which is not at the
6072     // end of memory page is placed here. All other cases are in stub.
6073     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6074     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6075     assert_different_registers(ary1, len, result);
6076 
6077     mov(result, len);
6078     cmpw(len, 0);
6079     br(LE, DONE);
6080     cmpw(len, 4 * wordSize);
6081     br(GE, STUB_LONG); // size > 32 then go to stub
6082 
6083     int shift = 64 - exact_log2(os::vm_page_size());
6084     lsl(rscratch1, ary1, shift);
6085     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6086     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6087     br(CS, STUB); // at the end of page then go to stub

6971 // On other systems, the helper is a usual C function.
6972 //
6973 void MacroAssembler::get_thread(Register dst) {
6974   RegSet saved_regs =
6975     BSD_ONLY(RegSet::range(r0, r17) + lr - dst)
6976     NOT_BSD (RegSet::range(r0, r1)  + lr - dst);
6977 
6978   protect_return_address();
6979   push(saved_regs, sp);
6980 
6981   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6982   blr(lr);
6983   if (dst != c_rarg0) {
6984     mov(dst, c_rarg0);
6985   }
6986 
6987   pop(saved_regs, sp);
6988   authenticate_return_address();
6989 }
6990 
6991 #ifdef COMPILER2
6992 // C2 compiled method's prolog code
6993 // Moved here from aarch64.ad to support Valhalla code below
6994 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
6995   if (C->clinit_barrier_on_entry()) {
6996     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
6997 
6998     Label L_skip_barrier;
6999 
7000     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
7001     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
7002     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
7003     bind(L_skip_barrier);
7004   }
7005 
7006   if (C->max_vector_size() > 0) {
7007     reinitialize_ptrue();
7008   }
7009 
7010   int bangsize = C->output()->bang_size_in_bytes();
7011   if (C->output()->need_stack_bang(bangsize))
7012     generate_stack_overflow_check(bangsize);
7013 
7014   // n.b. frame size includes space for return pc and rfp
7015   const long framesize = C->output()->frame_size_in_bytes();
7016   build_frame(framesize DEBUG_ONLY(COMMA sp_inc != 0));
7017 
7018   if (C->needs_stack_repair()) {
7019     save_stack_increment(sp_inc, framesize);
7020   }
7021 
7022   if (VerifyStackAtCalls) {
7023     Unimplemented();
7024   }
7025 }
7026 #endif // COMPILER2
7027 
7028 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
7029   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
7030   // An inline type might be returned. If fields are in registers we
7031   // need to allocate an inline type instance and initialize it with
7032   // the value of the fields.
7033   Label skip;
7034   // We only need a new buffered inline type if a new one is not returned
7035   tbz(r0, 0, skip);
7036   int call_offset = -1;
7037 
7038   // Be careful not to clobber r1-7 which hold returned fields
7039   // Also do not use callee-saved registers as these may be live in the interpreter
7040   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
7041 
7042   // The following code is similar to the instance allocation code in TemplateTable::_new
7043   //  but has some slight differences,
7044   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
7045   // allocating is not necessary if vk != nullptr, etc.
7046   Label slow_case;
7047   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
7048   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
7049 
7050   if (vk != nullptr) {
7051     // Called from C1, where the return type is statically known.
7052     movptr(klass, (intptr_t)vk->get_InlineKlass());
7053     jint lh = vk->layout_helper();
7054     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
7055     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
7056       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
7057     } else {
7058       b(slow_case);
7059     }
7060   } else {
7061     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
7062     andr(klass, r0, -2);
7063     if (UseTLAB) {
7064       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
7065       tst(tmp2, Klass::_lh_instance_slow_path_bit);
7066       br(Assembler::NE, slow_case);
7067       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
7068     } else {
7069       b(slow_case);
7070     }
7071   }
7072   if (UseTLAB) {
7073     // 2. Initialize buffered inline instance header
7074     Register buffer_obj = r0;
7075     if (UseCompactObjectHeaders) {
7076       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7077       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7078     } else {
7079       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7080       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7081       store_klass_gap(buffer_obj, zr);
7082       if (vk == nullptr) {
7083         // store_klass corrupts klass, so save it for later use (interpreter case only).
7084         mov(tmp1, klass);
7085       }
7086       store_klass(buffer_obj, klass);
7087       klass = tmp1;
7088     }
7089     // 3. Initialize its fields with an inline class specific handler
7090     if (vk != nullptr) {
7091       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7092     } else {
7093       ldr(tmp1, Address(klass, InlineKlass::adr_members_offset()));
7094       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7095       blr(tmp1);
7096     }
7097 
7098     membar(Assembler::StoreStore);
7099     b(skip);
7100   } else {
7101     // Must have already branched to slow_case above.
7102     DEBUG_ONLY(should_not_reach_here());
7103   }
7104   bind(slow_case);
7105   // We failed to allocate a new inline type, fall back to a runtime
7106   // call. Some oop field may be live in some registers but we can't
7107   // tell. That runtime call will take care of preserving them
7108   // across a GC if there's one.
7109   mov(r0, r0_preserved);
7110 
7111   if (from_interpreter) {
7112     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
7113   } else {
7114     far_call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
7115     call_offset = offset();
7116   }
7117   membar(Assembler::StoreStore);
7118 
7119   bind(skip);
7120   return call_offset;
7121 }
7122 
7123 // Move a value between registers/stack slots and update the reg_state
7124 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7125   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7126   if (reg_state[to->value()] == reg_written) {
7127     return true; // Already written
7128   }
7129 
7130   if (from != to && bt != T_VOID) {
7131     if (reg_state[to->value()] == reg_readonly) {
7132       return false; // Not yet writable
7133     }
7134     if (from->is_reg()) {
7135       if (to->is_reg()) {
7136         if (from->is_Register() && to->is_Register()) {
7137           mov(to->as_Register(), from->as_Register());
7138         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7139           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7140         } else {
7141           ShouldNotReachHere();
7142         }
7143       } else {
7144         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7145         Address to_addr = Address(sp, st_off);
7146         if (from->is_FloatRegister()) {
7147           if (bt == T_DOUBLE) {
7148              strd(from->as_FloatRegister(), to_addr);
7149           } else {
7150              assert(bt == T_FLOAT, "must be float");
7151              strs(from->as_FloatRegister(), to_addr);
7152           }
7153         } else {
7154           str(from->as_Register(), to_addr);
7155         }
7156       }
7157     } else {
7158       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7159       if (to->is_reg()) {
7160         if (to->is_FloatRegister()) {
7161           if (bt == T_DOUBLE) {
7162             ldrd(to->as_FloatRegister(), from_addr);
7163           } else {
7164             assert(bt == T_FLOAT, "must be float");
7165             ldrs(to->as_FloatRegister(), from_addr);
7166           }
7167         } else {
7168           ldr(to->as_Register(), from_addr);
7169         }
7170       } else {
7171         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7172         ldr(rscratch1, from_addr);
7173         str(rscratch1, Address(sp, st_off));
7174       }
7175     }
7176   }
7177 
7178   // Update register states
7179   reg_state[from->value()] = reg_writable;
7180   reg_state[to->value()] = reg_written;
7181   return true;
7182 }
7183 
7184 // Calculate the extra stack space required for packing or unpacking inline
7185 // args and adjust the stack pointer
7186 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7187   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7188   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7189   assert(sp_inc > 0, "sanity");
7190 
7191   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7192   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7193 
7194   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7195   if (sp_inc < (1 << 9)) {
7196     sub(sp, sp, sp_inc);   // Fits in an immediate
7197   } else {
7198     mov(rscratch1, sp_inc);
7199     sub(sp, sp, rscratch1);
7200   }
7201 
7202   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7203 }
7204 
7205 // Read all fields from an inline type oop and store the values in registers/stack slots
7206 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7207                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7208                                           RegState reg_state[]) {
7209   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7210   assert(from->is_valid(), "source must be valid");
7211   bool progress = false;
7212 #ifdef ASSERT
7213   const int start_offset = offset();
7214 #endif
7215 
7216   Label L_null, L_notNull;
7217   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7218   Register tmp1 = r10;
7219   Register tmp2 = r11;
7220 
7221 #ifdef ASSERT
7222   RegSet clobbered_gp_regs = MacroAssembler::call_clobbered_gp_registers();
7223   assert(clobbered_gp_regs.contains(tmp1), "tmp1 must be saved explicitly if it's not a clobber");
7224   assert(clobbered_gp_regs.contains(tmp2), "tmp2 must be saved explicitly if it's not a clobber");
7225   assert(clobbered_gp_regs.contains(r14), "r14 must be saved explicitly if it's not a clobber");
7226 #endif
7227 
7228   Register fromReg = noreg;
7229   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
7230   bool done = true;
7231   bool mark_done = true;
7232   VMReg toReg;
7233   BasicType bt;
7234   // Check if argument requires a null check
7235   bool null_check = false;
7236   VMReg nullCheckReg;
7237   while (stream.next(nullCheckReg, bt)) {
7238     if (sig->at(stream.sig_index())._offset == -1) {
7239       null_check = true;
7240       break;
7241     }
7242   }
7243   stream.reset(sig_index, to_index);
7244   while (stream.next(toReg, bt)) {
7245     assert(toReg->is_valid(), "destination must be valid");
7246     int idx = (int)toReg->value();
7247     if (reg_state[idx] == reg_readonly) {
7248       if (idx != from->value()) {
7249         mark_done = false;
7250       }
7251       done = false;
7252       continue;
7253     } else if (reg_state[idx] == reg_written) {
7254       continue;
7255     }
7256     assert(reg_state[idx] == reg_writable, "must be writable");
7257     reg_state[idx] = reg_written;
7258     progress = true;
7259 
7260     if (fromReg == noreg) {
7261       if (from->is_reg()) {
7262         fromReg = from->as_Register();
7263       } else {
7264         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7265         ldr(tmp1, Address(sp, st_off));
7266         fromReg = tmp1;
7267       }
7268       if (null_check) {
7269         // Nullable inline type argument, emit null check
7270         cbz(fromReg, L_null);
7271       }
7272     }
7273     int off = sig->at(stream.sig_index())._offset;
7274     if (off == -1) {
7275       assert(null_check, "Missing null check at");
7276       if (toReg->is_stack()) {
7277         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7278         mov(tmp2, 1);
7279         str(tmp2, Address(sp, st_off));
7280       } else {
7281         mov(toReg->as_Register(), 1);
7282       }
7283       continue;
7284     }
7285     if (sig->at(stream.sig_index())._vt_oop) {
7286       if (toReg->is_stack()) {
7287         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7288         str(fromReg, Address(sp, st_off));
7289       } else {
7290         mov(toReg->as_Register(), fromReg);
7291       }
7292       continue;
7293     }
7294     assert(off > 0, "offset in object should be positive");
7295     Address fromAddr = Address(fromReg, off);
7296     if (!toReg->is_FloatRegister()) {
7297       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7298       if (is_reference_type(bt)) {
7299         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7300       } else {
7301         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7302         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7303       }
7304       if (toReg->is_stack()) {
7305         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7306         str(dst, Address(sp, st_off));
7307       }
7308     } else if (bt == T_DOUBLE) {
7309       ldrd(toReg->as_FloatRegister(), fromAddr);
7310     } else {
7311       assert(bt == T_FLOAT, "must be float");
7312       ldrs(toReg->as_FloatRegister(), fromAddr);
7313     }
7314   }
7315   if (progress && null_check) {
7316     if (done) {
7317       b(L_notNull);
7318       bind(L_null);
7319       // Set null marker to zero to signal that the argument is null.
7320       // Also set all fields to zero since the runtime requires a canonical
7321       // representation of a flat null.
7322       stream.reset(sig_index, to_index);
7323       while (stream.next(toReg, bt)) {
7324         if (toReg->is_stack()) {
7325           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7326           str(zr, Address(sp, st_off));
7327         } else if (toReg->is_FloatRegister()) {
7328           mov(toReg->as_FloatRegister(), T2S, 0);
7329         } else {
7330           mov(toReg->as_Register(), zr);
7331         }
7332       }
7333       bind(L_notNull);
7334     } else {
7335       bind(L_null);
7336     }
7337   }
7338 
7339   sig_index = stream.sig_index();
7340   to_index = stream.regs_index();
7341 
7342   if (mark_done && reg_state[from->value()] != reg_written) {
7343     // This is okay because no one else will write to that slot
7344     reg_state[from->value()] = reg_writable;
7345   }
7346   from_index--;
7347   assert(progress || (start_offset == offset()), "should not emit code");
7348   return done;
7349 }
7350 
7351 // Pack fields back into an inline type oop
7352 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7353                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7354                                         RegState reg_state[], Register val_array) {
7355   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7356   assert(to->is_valid(), "destination must be valid");
7357 
7358   if (reg_state[to->value()] == reg_written) {
7359     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7360     return true; // Already written
7361   }
7362 
7363   // The GC barrier expanded by store_heap_oop below may call into the
7364   // runtime so use callee-saved registers for any values that need to be
7365   // preserved. The GC barrier assembler should take care of saving the
7366   // Java argument registers.
7367   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7368   Register val_obj_tmp = r21;
7369   Register from_reg_tmp = r22;
7370   Register tmp1 = r14;
7371   Register tmp2 = r13;
7372   Register tmp3 = r12;
7373   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7374 
7375   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7376 
7377   if (reg_state[to->value()] == reg_readonly) {
7378     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7379       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7380       return false; // Not yet writable
7381     }
7382     val_obj = val_obj_tmp;
7383   }
7384 
7385   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7386   VMReg fromReg;
7387   BasicType bt;
7388   Label L_null;
7389   while (stream.next(fromReg, bt)) {
7390     assert(fromReg->is_valid(), "source must be valid");
7391     reg_state[fromReg->value()] = reg_writable;
7392 
7393     int off = sig->at(stream.sig_index())._offset;
7394     if (off == -1) {
7395       // Nullable inline type argument, emit null check
7396       Label L_notNull;
7397       if (fromReg->is_stack()) {
7398         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7399         ldrb(tmp2, Address(sp, ld_off));
7400         cbnz(tmp2, L_notNull);
7401       } else {
7402         cbnz(fromReg->as_Register(), L_notNull);
7403       }
7404       mov(val_obj, 0);
7405       b(L_null);
7406       bind(L_notNull);
7407       continue;
7408     }
7409     if (sig->at(stream.sig_index())._vt_oop) {
7410       if (fromReg->is_stack()) {
7411         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7412         ldr(val_obj, Address(sp, ld_off));
7413       } else {
7414         mov(val_obj, fromReg->as_Register());
7415       }
7416       cbnz(val_obj, L_null);
7417       // get the buffer from the just allocated pool of buffers
7418       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7419       load_heap_oop(val_obj, Address(val_array, index), rscratch1, rscratch2);
7420       continue;
7421     }
7422 
7423     assert(off > 0, "offset in object should be positive");
7424     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7425 
7426     // Pack the scalarized field into the value object.
7427     Address dst(val_obj, off);
7428     if (!fromReg->is_FloatRegister()) {
7429       Register src;
7430       if (fromReg->is_stack()) {
7431         src = from_reg_tmp;
7432         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7433         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7434       } else {
7435         src = fromReg->as_Register();
7436       }
7437       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7438       if (is_reference_type(bt)) {
7439         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
7440         mov(tmp3, val_obj);
7441         Address dst_with_tmp3(tmp3, off);
7442         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7443       } else {
7444         store_sized_value(dst, src, size_in_bytes);
7445       }
7446     } else if (bt == T_DOUBLE) {
7447       strd(fromReg->as_FloatRegister(), dst);
7448     } else {
7449       assert(bt == T_FLOAT, "must be float");
7450       strs(fromReg->as_FloatRegister(), dst);
7451     }
7452   }
7453   bind(L_null);
7454   sig_index = stream.sig_index();
7455   from_index = stream.regs_index();
7456 
7457   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7458   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7459   assert(success, "to register must be writable");
7460   return true;
7461 }
7462 
7463 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7464   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7465 }
7466 
7467 void MacroAssembler::cache_wb(Address line) {
7468   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7469   assert(line.index() == noreg, "index should be noreg");
7470   assert(line.offset() == 0, "offset should be 0");
7471   // would like to assert this
7472   // assert(line._ext.shift == 0, "shift should be zero");
7473   if (VM_Version::supports_dcpop()) {
7474     // writeback using clear virtual address to point of persistence
7475     dc(Assembler::CVAP, line.base());
7476   } else {
7477     // no need to generate anything as Unsafe.writebackMemory should
7478     // never invoke this stub
7479   }
7480 }
7481 
7482 void MacroAssembler::cache_wbsync(bool is_pre) {
7483   // we only need a barrier post sync
7484   if (!is_pre) {
7485     membar(Assembler::AnyAny);
7486   }

7882   }
7883 
7884   // Check if the lock-stack is full.
7885   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7886   cmpw(top, (unsigned)LockStack::end_offset());
7887   br(Assembler::GE, slow);
7888 
7889   // Check for recursion.
7890   subw(t, top, oopSize);
7891   ldr(t, Address(rthread, t));
7892   cmp(obj, t);
7893   br(Assembler::EQ, push);
7894 
7895   // Check header for monitor (0b10).
7896   tst(mark, markWord::monitor_value);
7897   br(Assembler::NE, slow);
7898 
7899   // Try to lock. Transition lock bits 0b01 => 0b00
7900   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7901   orr(mark, mark, markWord::unlocked_value);
7902   // Mask inline_type bit such that we go to the slow path if object is an inline type
7903   andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
7904 
7905   eor(t, mark, markWord::unlocked_value);
7906   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7907           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7908   br(Assembler::NE, slow);
7909 
7910   bind(push);
7911   // After successful lock, push object on lock-stack.
7912   str(obj, Address(rthread, top));
7913   addw(top, top, oopSize);
7914   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7915 }
7916 
7917 // Implements fast-unlocking.
7918 //
7919 // - obj: the object to be unlocked
7920 // - t1, t2, t3: temporary registers
7921 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7922 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7923   // cmpxchg clobbers rscratch1.
7924   assert_different_registers(obj, t1, t2, t3, rscratch1);
< prev index next >