< 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/powerOfTwo.hpp"

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

2005   ldarb(scratch, scratch);
2006   cmp(scratch, InstanceKlass::fully_initialized);
2007   br(Assembler::EQ, *L_fast_path);
2008 
2009   // Fast path check: current thread is initializer thread
2010   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2011   cmp(rthread, scratch);
2012 
2013   if (L_slow_path == &L_fallthrough) {
2014     br(Assembler::EQ, *L_fast_path);
2015     bind(*L_slow_path);
2016   } else if (L_fast_path == &L_fallthrough) {
2017     br(Assembler::NE, *L_slow_path);
2018     bind(*L_fast_path);
2019   } else {
2020     Unimplemented();
2021   }
2022 }
2023 
2024 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2025   if (!VerifyOops) return;




2026 
2027   // Pass register number to verify_oop_subroutine
2028   const char* b = nullptr;
2029   {
2030     ResourceMark rm;
2031     stringStream ss;
2032     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2033     b = code_string(ss.as_string());
2034   }
2035   BLOCK_COMMENT("verify_oop {");
2036 
2037   strip_return_address(); // This might happen within a stack frame.
2038   protect_return_address();
2039   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2040   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2041 
2042   mov(r0, reg);
2043   movptr(rscratch1, (uintptr_t)(address)b);
2044 
2045   // call indirectly to solve generation ordering problem
2046   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2047   ldr(rscratch2, Address(rscratch2));
2048   blr(rscratch2);
2049 
2050   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2051   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2052   authenticate_return_address();
2053 
2054   BLOCK_COMMENT("} verify_oop");
2055 }
2056 
2057 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2058   if (!VerifyOops) return;




2059 
2060   const char* b = nullptr;
2061   {
2062     ResourceMark rm;
2063     stringStream ss;
2064     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2065     b = code_string(ss.as_string());
2066   }
2067   BLOCK_COMMENT("verify_oop_addr {");
2068 
2069   strip_return_address(); // This might happen within a stack frame.
2070   protect_return_address();
2071   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2072   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2073 
2074   // addr may contain sp so we will have to adjust it based on the
2075   // pushes that we just did.
2076   if (addr.uses(sp)) {
2077     lea(r0, addr);
2078     ldr(r0, Address(r0, 4 * wordSize));

2291   call_VM_leaf_base(entry_point, 1);
2292 }
2293 
2294 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2295   assert_different_registers(arg_1, c_rarg0);
2296   pass_arg0(this, arg_0);
2297   pass_arg1(this, arg_1);
2298   call_VM_leaf_base(entry_point, 2);
2299 }
2300 
2301 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2302                                   Register arg_1, Register arg_2) {
2303   assert_different_registers(arg_1, c_rarg0);
2304   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2305   pass_arg0(this, arg_0);
2306   pass_arg1(this, arg_1);
2307   pass_arg2(this, arg_2);
2308   call_VM_leaf_base(entry_point, 3);
2309 }
2310 




2311 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2312   pass_arg0(this, arg_0);
2313   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2314 }
2315 
2316 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2317 
2318   assert_different_registers(arg_0, c_rarg1);
2319   pass_arg1(this, arg_1);
2320   pass_arg0(this, arg_0);
2321   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2322 }
2323 
2324 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2325   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2326   assert_different_registers(arg_1, c_rarg2);
2327   pass_arg2(this, arg_2);
2328   pass_arg1(this, arg_1);
2329   pass_arg0(this, arg_0);
2330   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2336   assert_different_registers(arg_2, c_rarg3);
2337   pass_arg3(this, arg_3);
2338   pass_arg2(this, arg_2);
2339   pass_arg1(this, arg_1);
2340   pass_arg0(this, arg_0);
2341   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2342 }
2343 
2344 void MacroAssembler::null_check(Register reg, int offset) {
2345   if (needs_explicit_null_check(offset)) {
2346     // provoke OS null exception if reg is null by
2347     // accessing M[reg] w/o changing any registers
2348     // NOTE: this is plenty to provoke a segv
2349     ldr(zr, Address(reg));
2350   } else {
2351     // nothing to do, (later) access of M[reg + offset]
2352     // will provoke OS null exception if reg is null
2353   }
2354 }
2355 
























































































2356 // MacroAssembler protected routines needed to implement
2357 // public methods
2358 
2359 void MacroAssembler::mov(Register r, Address dest) {
2360   code_section()->relocate(pc(), dest.rspec());
2361   uint64_t imm64 = (uint64_t)dest.target();
2362   movptr(r, imm64);
2363 }
2364 
2365 // Move a constant pointer into r.  In AArch64 mode the virtual
2366 // address space is 48 bits in size, so we only need three
2367 // instructions to create a patchable instruction sequence that can
2368 // reach anywhere.
2369 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2370 #ifndef PRODUCT
2371   {
2372     char buffer[64];
2373     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2374     block_comment(buffer);
2375   }

5034   adrp(rscratch1, src2, offset);
5035   ldr(rscratch1, Address(rscratch1, offset));
5036   cmp(src1, rscratch1);
5037 }
5038 
5039 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5040   cmp(obj1, obj2);
5041 }
5042 
5043 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5044   load_method_holder(rresult, rmethod);
5045   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5046 }
5047 
5048 void MacroAssembler::load_method_holder(Register holder, Register method) {
5049   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5050   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5051   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5052 }
5053 








5054 // Loads the obj's Klass* into dst.
5055 // Preserves all registers (incl src, rscratch1 and rscratch2).
5056 // Input:
5057 // src - the oop we want to load the klass from.
5058 // dst - output narrow klass.
5059 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5060   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5061   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5062   lsr(dst, dst, markWord::klass_shift);
5063 }
5064 
5065 void MacroAssembler::load_klass(Register dst, Register src) {
5066   if (UseCompactObjectHeaders) {
5067     load_narrow_klass_compact(dst, src);
5068   } else {
5069     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5070   }
5071   decode_klass_not_null(dst);
5072 }
5073 

5125   } else {
5126     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5127   }
5128   if (CompressedKlassPointers::base() == nullptr) {
5129     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5130     return;
5131   } else if (!AOTCodeCache::is_on_for_dump() &&
5132              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5133              && CompressedKlassPointers::shift() == 0) {
5134     // Only the bottom 32 bits matter
5135     cmpw(klass, tmp);
5136     return;
5137   }
5138   decode_klass_not_null(tmp);
5139   cmp(klass, tmp);
5140 }
5141 
5142 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5143   if (UseCompactObjectHeaders) {
5144     load_narrow_klass_compact(tmp1, obj1);
5145     load_narrow_klass_compact(tmp2,  obj2);
5146   } else {
5147     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5148     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5149   }
5150   cmpw(tmp1, tmp2);
5151 }
5152 





5153 void MacroAssembler::store_klass(Register dst, Register src) {
5154   // FIXME: Should this be a store release?  concurrent gcs assumes
5155   // klass length is valid if klass field is not null.
5156   assert(!UseCompactObjectHeaders, "not with compact headers");
5157   encode_klass_not_null(src);
5158   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5159 }
5160 
5161 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5162   assert(!UseCompactObjectHeaders, "not with compact headers");
5163   // Store to klass gap in destination
5164   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5165 }
5166 
5167 // Algorithm must match CompressedOops::encode.
5168 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5169 #ifdef ASSERT
5170   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5171 #endif
5172   verify_oop_msg(s, "broken oop in encode_heap_oop");

5522   if (as_raw) {
5523     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5524   } else {
5525     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5526   }
5527 }
5528 
5529 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5530                                      Address dst, Register val,
5531                                      Register tmp1, Register tmp2, Register tmp3) {
5532   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5533   decorators = AccessInternal::decorator_fixup(decorators, type);
5534   bool as_raw = (decorators & AS_RAW) != 0;
5535   if (as_raw) {
5536     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5537   } else {
5538     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5539   }
5540 }
5541 








































5542 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5543                                    Register tmp2, DecoratorSet decorators) {
5544   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5545 }
5546 
5547 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5548                                             Register tmp2, DecoratorSet decorators) {
5549   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5550 }
5551 
5552 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5553                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5554   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5555 }
5556 
5557 // Used for storing nulls.
5558 void MacroAssembler::store_heap_oop_null(Address dst) {
5559   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5560 }
5561 

5597     oop_index = oop_recorder()->allocate_metadata_index(obj);
5598   } else {
5599     oop_index = oop_recorder()->find_index(obj);
5600   }
5601   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5602   mov(dst, Address((address)obj, rspec));
5603 }
5604 
5605 Address MacroAssembler::constant_oop_address(jobject obj) {
5606 #ifdef ASSERT
5607   {
5608     ThreadInVMfromUnknown tiv;
5609     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5610     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5611   }
5612 #endif
5613   int oop_index = oop_recorder()->find_index(obj);
5614   return Address((address)obj, oop_Relocation::spec(oop_index));
5615 }
5616 






































































































5617 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5618 void MacroAssembler::tlab_allocate(Register obj,
5619                                    Register var_size_in_bytes,
5620                                    int con_size_in_bytes,
5621                                    Register t1,
5622                                    Register t2,
5623                                    Label& slow_case) {
5624   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5625   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5626 }
5627 
5628 void MacroAssembler::verify_tlab() {
5629 #ifdef ASSERT
5630   if (UseTLAB && VerifyOops) {
5631     Label next, ok;
5632 
5633     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5634 
5635     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5636     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5637     cmp(rscratch2, rscratch1);
5638     br(Assembler::HS, next);
5639     STOP("assert(top >= start)");
5640     should_not_reach_here();
5641 
5642     bind(next);
5643     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5644     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5645     cmp(rscratch2, rscratch1);
5646     br(Assembler::HS, ok);
5647     STOP("assert(top <= end)");
5648     should_not_reach_here();
5649 
5650     bind(ok);
5651     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5652   }
5653 #endif
5654 }
5655 















5656 // Writes to stack successive pages until offset reached to check for
5657 // stack overflow + shadow pages.  This clobbers tmp.
5658 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5659   assert_different_registers(tmp, size, rscratch1);
5660   mov(tmp, sp);
5661   // Bang stack for total size given plus shadow page size.
5662   // Bang one page at a time because large size can bang beyond yellow and
5663   // red zones.
5664   Label loop;
5665   mov(rscratch1, (int)os::vm_page_size());
5666   bind(loop);
5667   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5668   subsw(size, size, rscratch1);
5669   str(size, Address(tmp));
5670   br(Assembler::GT, loop);
5671 
5672   // Bang down shadow pages too.
5673   // At this point, (tmp-0) is the last address touched, so don't
5674   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5675   // was post-decremented.)  Skip this address by starting at i=1, and

5742 
5743   // Strictly speaking the card table base isn't an address at all, and it might
5744   // even be negative. It is thus materialised as a constant.
5745   mov(reg, (uint64_t)ctbs->card_table_base_const());
5746 }
5747 
5748 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5749 #if INCLUDE_CDS
5750   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5751   if (AOTCodeCache::is_on_for_dump()) {
5752     // all aotrc field addresses should be registered in the AOTCodeCache address table
5753     lea(reg, ExternalAddress(a));
5754   } else {
5755     mov(reg, (uint64_t)a);
5756   }
5757 #else
5758   ShouldNotReachHere();
5759 #endif
5760 }
5761 

5762 void MacroAssembler::build_frame(int framesize) {





5763   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5764   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5765   protect_return_address();
5766   if (framesize < ((1 << 9) + 2 * wordSize)) {
5767     sub(sp, sp, framesize);
5768     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));





5769     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5770   } else {
5771     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));





5772     if (PreserveFramePointer) mov(rfp, sp);
5773     if (framesize < ((1 << 12) + 2 * wordSize))
5774       sub(sp, sp, framesize - 2 * wordSize);
5775     else {
5776       mov(rscratch1, framesize - 2 * wordSize);
5777       sub(sp, sp, rscratch1);
5778     }
5779   }
5780   verify_cross_modify_fence_not_required();
5781 }
5782 
5783 void MacroAssembler::remove_frame(int framesize) {
5784   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5785   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5786   if (framesize < ((1 << 9) + 2 * wordSize)) {
5787     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5788     add(sp, sp, framesize);
5789   } else {
5790     if (framesize < ((1 << 12) + 2 * wordSize))
5791       add(sp, sp, framesize - 2 * wordSize);
5792     else {
5793       mov(rscratch1, framesize - 2 * wordSize);
5794       add(sp, sp, rscratch1);
5795     }
5796     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5797   }
5798   authenticate_return_address();
5799 }
5800 


















































































5801 
5802 // This method counts leading positive bytes (highest bit not set) in provided byte array
5803 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5804     // Simple and most common case of aligned small array which is not at the
5805     // end of memory page is placed here. All other cases are in stub.
5806     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5807     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5808     assert_different_registers(ary1, len, result);
5809 
5810     mov(result, len);
5811     cmpw(len, 0);
5812     br(LE, DONE);
5813     cmpw(len, 4 * wordSize);
5814     br(GE, STUB_LONG); // size > 32 then go to stub
5815 
5816     int shift = 64 - exact_log2(os::vm_page_size());
5817     lsl(rscratch1, ary1, shift);
5818     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5819     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5820     br(CS, STUB); // at the end of page then go to stub

6703 // On other systems, the helper is a usual C function.
6704 //
6705 void MacroAssembler::get_thread(Register dst) {
6706   RegSet saved_regs =
6707     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6708     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6709 
6710   protect_return_address();
6711   push(saved_regs, sp);
6712 
6713   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6714   blr(lr);
6715   if (dst != c_rarg0) {
6716     mov(dst, c_rarg0);
6717   }
6718 
6719   pop(saved_regs, sp);
6720   authenticate_return_address();
6721 }
6722 




























































































































































































































































































































































































































































































6723 void MacroAssembler::cache_wb(Address line) {
6724   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6725   assert(line.index() == noreg, "index should be noreg");
6726   assert(line.offset() == 0, "offset should be 0");
6727   // would like to assert this
6728   // assert(line._ext.shift == 0, "shift should be zero");
6729   if (VM_Version::supports_dcpop()) {
6730     // writeback using clear virtual address to point of persistence
6731     dc(Assembler::CVAP, line.base());
6732   } else {
6733     // no need to generate anything as Unsafe.writebackMemory should
6734     // never invoke this stub
6735   }
6736 }
6737 
6738 void MacroAssembler::cache_wbsync(bool is_pre) {
6739   // we only need a barrier post sync
6740   if (!is_pre) {
6741     membar(Assembler::AnyAny);
6742   }

7138   }
7139 
7140   // Check if the lock-stack is full.
7141   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7142   cmpw(top, (unsigned)LockStack::end_offset());
7143   br(Assembler::GE, slow);
7144 
7145   // Check for recursion.
7146   subw(t, top, oopSize);
7147   ldr(t, Address(rthread, t));
7148   cmp(obj, t);
7149   br(Assembler::EQ, push);
7150 
7151   // Check header for monitor (0b10).
7152   tst(mark, markWord::monitor_value);
7153   br(Assembler::NE, slow);
7154 
7155   // Try to lock. Transition lock bits 0b01 => 0b00
7156   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7157   orr(mark, mark, markWord::unlocked_value);



7158   eor(t, mark, markWord::unlocked_value);
7159   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7160           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7161   br(Assembler::NE, slow);
7162 
7163   bind(push);
7164   // After successful lock, push object on lock-stack.
7165   str(obj, Address(rthread, top));
7166   addw(top, top, oopSize);
7167   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7168 }
7169 
7170 // Implements fast-unlocking.
7171 //
7172 // - obj: the object to be unlocked
7173 // - t1, t2, t3: temporary registers
7174 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7175 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7176   // cmpxchg clobbers rscratch1.
7177   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/powerOfTwo.hpp"
  64 #include "vmreg_aarch64.inline.hpp"
  65 #ifdef COMPILER1
  66 #include "c1/c1_LIRAssembler.hpp"
  67 #endif
  68 #ifdef COMPILER2
  69 #include "oops/oop.hpp"
  70 #include "opto/compile.hpp"
  71 #include "opto/node.hpp"
  72 #include "opto/output.hpp"
  73 #endif
  74 
  75 #include <sys/types.h>
  76 
  77 #ifdef PRODUCT
  78 #define BLOCK_COMMENT(str) /* nothing */
  79 #else
  80 #define BLOCK_COMMENT(str) block_comment(str)
  81 #endif
  82 #define STOP(str) stop(str);
  83 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  84 

2011   ldarb(scratch, scratch);
2012   cmp(scratch, InstanceKlass::fully_initialized);
2013   br(Assembler::EQ, *L_fast_path);
2014 
2015   // Fast path check: current thread is initializer thread
2016   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2017   cmp(rthread, scratch);
2018 
2019   if (L_slow_path == &L_fallthrough) {
2020     br(Assembler::EQ, *L_fast_path);
2021     bind(*L_slow_path);
2022   } else if (L_fast_path == &L_fallthrough) {
2023     br(Assembler::NE, *L_slow_path);
2024     bind(*L_fast_path);
2025   } else {
2026     Unimplemented();
2027   }
2028 }
2029 
2030 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2031   if (!VerifyOops || VerifyAdapterSharing) {
2032     // Below address of the code string confuses VerifyAdapterSharing
2033     // because it may differ between otherwise equivalent adapters.
2034     return;
2035   }
2036 
2037   // Pass register number to verify_oop_subroutine
2038   const char* b = nullptr;
2039   {
2040     ResourceMark rm;
2041     stringStream ss;
2042     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2043     b = code_string(ss.as_string());
2044   }
2045   BLOCK_COMMENT("verify_oop {");
2046 
2047   strip_return_address(); // This might happen within a stack frame.
2048   protect_return_address();
2049   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2050   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2051 
2052   mov(r0, reg);
2053   movptr(rscratch1, (uintptr_t)(address)b);
2054 
2055   // call indirectly to solve generation ordering problem
2056   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2057   ldr(rscratch2, Address(rscratch2));
2058   blr(rscratch2);
2059 
2060   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2061   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2062   authenticate_return_address();
2063 
2064   BLOCK_COMMENT("} verify_oop");
2065 }
2066 
2067 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2068   if (!VerifyOops || VerifyAdapterSharing) {
2069     // Below address of the code string confuses VerifyAdapterSharing
2070     // because it may differ between otherwise equivalent adapters.
2071     return;
2072   }
2073 
2074   const char* b = nullptr;
2075   {
2076     ResourceMark rm;
2077     stringStream ss;
2078     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2079     b = code_string(ss.as_string());
2080   }
2081   BLOCK_COMMENT("verify_oop_addr {");
2082 
2083   strip_return_address(); // This might happen within a stack frame.
2084   protect_return_address();
2085   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2086   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2087 
2088   // addr may contain sp so we will have to adjust it based on the
2089   // pushes that we just did.
2090   if (addr.uses(sp)) {
2091     lea(r0, addr);
2092     ldr(r0, Address(r0, 4 * wordSize));

2305   call_VM_leaf_base(entry_point, 1);
2306 }
2307 
2308 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2309   assert_different_registers(arg_1, c_rarg0);
2310   pass_arg0(this, arg_0);
2311   pass_arg1(this, arg_1);
2312   call_VM_leaf_base(entry_point, 2);
2313 }
2314 
2315 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2316                                   Register arg_1, Register arg_2) {
2317   assert_different_registers(arg_1, c_rarg0);
2318   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2319   pass_arg0(this, arg_0);
2320   pass_arg1(this, arg_1);
2321   pass_arg2(this, arg_2);
2322   call_VM_leaf_base(entry_point, 3);
2323 }
2324 
2325 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2326   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2327 }
2328 
2329 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2330   pass_arg0(this, arg_0);
2331   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2332 }
2333 
2334 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2335 
2336   assert_different_registers(arg_0, c_rarg1);
2337   pass_arg1(this, arg_1);
2338   pass_arg0(this, arg_0);
2339   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2340 }
2341 
2342 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2343   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2344   assert_different_registers(arg_1, c_rarg2);
2345   pass_arg2(this, arg_2);
2346   pass_arg1(this, arg_1);
2347   pass_arg0(this, arg_0);
2348   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2354   assert_different_registers(arg_2, c_rarg3);
2355   pass_arg3(this, arg_3);
2356   pass_arg2(this, arg_2);
2357   pass_arg1(this, arg_1);
2358   pass_arg0(this, arg_0);
2359   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2360 }
2361 
2362 void MacroAssembler::null_check(Register reg, int offset) {
2363   if (needs_explicit_null_check(offset)) {
2364     // provoke OS null exception if reg is null by
2365     // accessing M[reg] w/o changing any registers
2366     // NOTE: this is plenty to provoke a segv
2367     ldr(zr, Address(reg));
2368   } else {
2369     // nothing to do, (later) access of M[reg + offset]
2370     // will provoke OS null exception if reg is null
2371   }
2372 }
2373 
2374 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2375   assert_different_registers(markword, rscratch2);
2376   mov(rscratch2, markWord::inline_type_pattern_mask);
2377   andr(markword, markword, rscratch2);
2378   mov(rscratch2, markWord::inline_type_pattern);
2379   cmp(markword, rscratch2);
2380   br(Assembler::EQ, is_inline_type);
2381 }
2382 
2383 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
2384   assert_different_registers(tmp, rscratch1);
2385   if (can_be_null) {
2386     cbz(object, not_inline_type);
2387   }
2388   const int is_inline_type_mask = markWord::inline_type_pattern;
2389   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2390   mov(rscratch1, is_inline_type_mask);
2391   andr(tmp, tmp, rscratch1);
2392   cmp(tmp, rscratch1);
2393   br(Assembler::NE, not_inline_type);
2394 }
2395 
2396 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2397   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2398   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2399 }
2400 
2401 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2402   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2403   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2404 }
2405 
2406 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2407   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2408   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2409 }
2410 
2411 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
2412   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2413   tbnz(flags, ResolvedFieldEntry::has_null_marker_shift, has_null_marker);
2414 }
2415 
2416 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2417   Label test_mark_word;
2418   // load mark word
2419   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2420   // check displaced
2421   tst(temp_reg, markWord::unlocked_value);
2422   br(Assembler::NE, test_mark_word);
2423   // slow path use klass prototype
2424   load_prototype_header(temp_reg, oop);
2425 
2426   bind(test_mark_word);
2427   andr(temp_reg, temp_reg, test_bit);
2428   if (jmp_set) {
2429     cbnz(temp_reg, jmp_label);
2430   } else {
2431     cbz(temp_reg, jmp_label);
2432   }
2433 }
2434 
2435 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2436   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2437 }
2438 
2439 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2440                                                   Label&is_non_flat_array) {
2441   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2442 }
2443 
2444 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2445   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2446 }
2447 
2448 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2449   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2450 }
2451 
2452 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2453   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2454   br(Assembler::NE, is_flat_array);
2455 }
2456 
2457 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
2458   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2459   br(Assembler::EQ, is_non_flat_array);
2460 }
2461 
2462 // MacroAssembler protected routines needed to implement
2463 // public methods
2464 
2465 void MacroAssembler::mov(Register r, Address dest) {
2466   code_section()->relocate(pc(), dest.rspec());
2467   uint64_t imm64 = (uint64_t)dest.target();
2468   movptr(r, imm64);
2469 }
2470 
2471 // Move a constant pointer into r.  In AArch64 mode the virtual
2472 // address space is 48 bits in size, so we only need three
2473 // instructions to create a patchable instruction sequence that can
2474 // reach anywhere.
2475 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2476 #ifndef PRODUCT
2477   {
2478     char buffer[64];
2479     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2480     block_comment(buffer);
2481   }

5140   adrp(rscratch1, src2, offset);
5141   ldr(rscratch1, Address(rscratch1, offset));
5142   cmp(src1, rscratch1);
5143 }
5144 
5145 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5146   cmp(obj1, obj2);
5147 }
5148 
5149 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5150   load_method_holder(rresult, rmethod);
5151   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5152 }
5153 
5154 void MacroAssembler::load_method_holder(Register holder, Register method) {
5155   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5156   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5157   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5158 }
5159 
5160 void MacroAssembler::load_metadata(Register dst, Register src) {
5161   if (UseCompactObjectHeaders) {
5162     load_narrow_klass_compact(dst, src);
5163   } else {
5164     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5165   }
5166 }
5167 
5168 // Loads the obj's Klass* into dst.
5169 // Preserves all registers (incl src, rscratch1 and rscratch2).
5170 // Input:
5171 // src - the oop we want to load the klass from.
5172 // dst - output narrow klass.
5173 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5174   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5175   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5176   lsr(dst, dst, markWord::klass_shift);
5177 }
5178 
5179 void MacroAssembler::load_klass(Register dst, Register src) {
5180   if (UseCompactObjectHeaders) {
5181     load_narrow_klass_compact(dst, src);
5182   } else {
5183     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5184   }
5185   decode_klass_not_null(dst);
5186 }
5187 

5239   } else {
5240     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5241   }
5242   if (CompressedKlassPointers::base() == nullptr) {
5243     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5244     return;
5245   } else if (!AOTCodeCache::is_on_for_dump() &&
5246              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5247              && CompressedKlassPointers::shift() == 0) {
5248     // Only the bottom 32 bits matter
5249     cmpw(klass, tmp);
5250     return;
5251   }
5252   decode_klass_not_null(tmp);
5253   cmp(klass, tmp);
5254 }
5255 
5256 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5257   if (UseCompactObjectHeaders) {
5258     load_narrow_klass_compact(tmp1, obj1);
5259     load_narrow_klass_compact(tmp2, obj2);
5260   } else {
5261     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5262     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5263   }
5264   cmpw(tmp1, tmp2);
5265 }
5266 
5267 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5268   load_klass(dst, src);
5269   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5270 }
5271 
5272 void MacroAssembler::store_klass(Register dst, Register src) {
5273   // FIXME: Should this be a store release?  concurrent gcs assumes
5274   // klass length is valid if klass field is not null.
5275   assert(!UseCompactObjectHeaders, "not with compact headers");
5276   encode_klass_not_null(src);
5277   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5278 }
5279 
5280 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5281   assert(!UseCompactObjectHeaders, "not with compact headers");
5282   // Store to klass gap in destination
5283   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5284 }
5285 
5286 // Algorithm must match CompressedOops::encode.
5287 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5288 #ifdef ASSERT
5289   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5290 #endif
5291   verify_oop_msg(s, "broken oop in encode_heap_oop");

5641   if (as_raw) {
5642     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5643   } else {
5644     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5645   }
5646 }
5647 
5648 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5649                                      Address dst, Register val,
5650                                      Register tmp1, Register tmp2, Register tmp3) {
5651   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5652   decorators = AccessInternal::decorator_fixup(decorators, type);
5653   bool as_raw = (decorators & AS_RAW) != 0;
5654   if (as_raw) {
5655     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5656   } else {
5657     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5658   }
5659 }
5660 
5661 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5662                                      Register inline_layout_info) {
5663   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5664   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5665 }
5666 
5667 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5668   ldr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
5669   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5670 }
5671 
5672 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5673   // ((address) (void*) o) + vk->payload_offset();
5674   Register offset = (data == oop) ? rscratch1 : data;
5675   payload_offset(inline_klass, offset);
5676   if (data == oop) {
5677     add(data, data, offset);
5678   } else {
5679     lea(data, Address(oop, offset));
5680   }
5681 }
5682 
5683 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
5684                                                 Register index, Register data) {
5685   assert_different_registers(array, array_klass, index);
5686   assert_different_registers(rscratch1, array, index);
5687 
5688   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
5689   ldrw(rscratch1, Address(array_klass, Klass::layout_helper_offset()));
5690 
5691   // Klass::layout_helper_log2_element_size(lh)
5692   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
5693   lsr(rscratch1, rscratch1, Klass::_lh_log2_element_size_shift);
5694   andr(rscratch1, rscratch1, Klass::_lh_log2_element_size_mask);
5695   lslv(index, index, rscratch1);
5696 
5697   add(data, array, index);
5698   add(data, data, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT));
5699 }
5700 
5701 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5702                                    Register tmp2, DecoratorSet decorators) {
5703   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5704 }
5705 
5706 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5707                                             Register tmp2, DecoratorSet decorators) {
5708   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5709 }
5710 
5711 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5712                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5713   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5714 }
5715 
5716 // Used for storing nulls.
5717 void MacroAssembler::store_heap_oop_null(Address dst) {
5718   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5719 }
5720 

5756     oop_index = oop_recorder()->allocate_metadata_index(obj);
5757   } else {
5758     oop_index = oop_recorder()->find_index(obj);
5759   }
5760   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5761   mov(dst, Address((address)obj, rspec));
5762 }
5763 
5764 Address MacroAssembler::constant_oop_address(jobject obj) {
5765 #ifdef ASSERT
5766   {
5767     ThreadInVMfromUnknown tiv;
5768     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5769     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5770   }
5771 #endif
5772   int oop_index = oop_recorder()->find_index(obj);
5773   return Address((address)obj, oop_Relocation::spec(oop_index));
5774 }
5775 
5776 // Object / value buffer allocation...
5777 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
5778                                        Register t1, Register t2,
5779                                        bool clear_fields, Label& alloc_failed)
5780 {
5781   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
5782   Register layout_size = t1;
5783   assert(new_obj == r0, "needs to be r0");
5784   assert_different_registers(klass, new_obj, t1, t2);
5785 
5786   // get instance_size in InstanceKlass (scaled to a count of bytes)
5787   ldrw(layout_size, Address(klass, Klass::layout_helper_offset()));
5788   // test to see if it is malformed in some way
5789   tst(layout_size, Klass::_lh_instance_slow_path_bit);
5790   br(Assembler::NE, slow_case_no_pop);
5791 
5792   // Allocate the instance:
5793   //  If TLAB is enabled:
5794   //    Try to allocate in the TLAB.
5795   //    If fails, go to the slow path.
5796   //    Initialize the allocation.
5797   //    Exit.
5798   //
5799   //  Go to slow path.
5800 
5801   if (UseTLAB) {
5802     push(klass);
5803     tlab_allocate(new_obj, layout_size, 0, klass, t2, slow_case);
5804     if (ZeroTLAB || (!clear_fields)) {
5805       // the fields have been already cleared
5806       b(initialize_header);
5807     } else {
5808       // initialize both the header and fields
5809       b(initialize_object);
5810     }
5811 
5812     if (clear_fields) {
5813       // The object is initialized before the header.  If the object size is
5814       // zero, go directly to the header initialization.
5815       bind(initialize_object);
5816       int header_size = oopDesc::header_size() * HeapWordSize;
5817       assert(is_aligned(header_size, BytesPerLong), "oop header size must be 8-byte-aligned");
5818       subs(layout_size, layout_size, header_size);
5819       br(Assembler::EQ, initialize_header);
5820 
5821       // Initialize topmost object field, divide size by 8, check if odd and
5822       // test if zero.
5823 
5824   #ifdef ASSERT
5825       // make sure instance_size was multiple of 8
5826       Label L;
5827       tst(layout_size, 7);
5828       br(Assembler::EQ, L);
5829       stop("object size is not multiple of 8 - adjust this code");
5830       bind(L);
5831       // must be > 0, no extra check needed here
5832   #endif
5833 
5834       lsr(layout_size, layout_size, LogBytesPerLong);
5835 
5836       // initialize remaining object fields: instance_size was a multiple of 8
5837       {
5838         Label loop;
5839         Register base = t2;
5840 
5841         bind(loop);
5842         add(rscratch1, new_obj, layout_size, Assembler::LSL, LogBytesPerLong);
5843         str(zr, Address(rscratch1, header_size - 1*oopSize));
5844         subs(layout_size, layout_size, 1);
5845         br(Assembler::NE, loop);
5846       }
5847     } // clear_fields
5848 
5849     // initialize object header only.
5850     bind(initialize_header);
5851     pop(klass);
5852     Register mark_word = t2;
5853     if (UseCompactObjectHeaders || Arguments::is_valhalla_enabled()) {
5854       ldr(mark_word, Address(klass, Klass::prototype_header_offset()));
5855       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5856     } else {
5857       mov(mark_word, (intptr_t)markWord::prototype().value());
5858       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5859     }
5860     if (!UseCompactObjectHeaders) {
5861       store_klass_gap(new_obj, zr);  // zero klass gap for compressed oops
5862       mov(t2, klass);                // preserve klass
5863       store_klass(new_obj, t2);      // src klass reg is potentially compressed
5864     }
5865     b(done);
5866   }
5867 
5868   if (UseTLAB) {
5869     bind(slow_case);
5870     pop(klass);
5871   }
5872   bind(slow_case_no_pop);
5873   b(alloc_failed);
5874 
5875   bind(done);
5876 }
5877 
5878 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5879 void MacroAssembler::tlab_allocate(Register obj,
5880                                    Register var_size_in_bytes,
5881                                    int con_size_in_bytes,
5882                                    Register t1,
5883                                    Register t2,
5884                                    Label& slow_case) {
5885   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5886   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5887 }
5888 
5889 void MacroAssembler::verify_tlab() {
5890 #ifdef ASSERT
5891   if (UseTLAB && VerifyOops) {
5892     Label next, ok;
5893 
5894     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5895 
5896     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5897     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5898     cmp(rscratch2, rscratch1);
5899     br(Assembler::HS, next);
5900     STOP("assert(top >= start)");
5901     should_not_reach_here();
5902 
5903     bind(next);
5904     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5905     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5906     cmp(rscratch2, rscratch1);
5907     br(Assembler::HS, ok);
5908     STOP("assert(top <= end)");
5909     should_not_reach_here();
5910 
5911     bind(ok);
5912     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5913   }
5914 #endif
5915 }
5916 
5917 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5918   assert_different_registers(holder_klass, index, layout_info);
5919   InlineLayoutInfo array[2];
5920   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5921   if (is_power_of_2(size)) {
5922     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5923   } else {
5924     mov(layout_info, size);
5925     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5926   }
5927   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5928   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5929   lea(layout_info, Address(layout_info, index));
5930 }
5931 
5932 // Writes to stack successive pages until offset reached to check for
5933 // stack overflow + shadow pages.  This clobbers tmp.
5934 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5935   assert_different_registers(tmp, size, rscratch1);
5936   mov(tmp, sp);
5937   // Bang stack for total size given plus shadow page size.
5938   // Bang one page at a time because large size can bang beyond yellow and
5939   // red zones.
5940   Label loop;
5941   mov(rscratch1, (int)os::vm_page_size());
5942   bind(loop);
5943   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5944   subsw(size, size, rscratch1);
5945   str(size, Address(tmp));
5946   br(Assembler::GT, loop);
5947 
5948   // Bang down shadow pages too.
5949   // At this point, (tmp-0) is the last address touched, so don't
5950   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5951   // was post-decremented.)  Skip this address by starting at i=1, and

6018 
6019   // Strictly speaking the card table base isn't an address at all, and it might
6020   // even be negative. It is thus materialised as a constant.
6021   mov(reg, (uint64_t)ctbs->card_table_base_const());
6022 }
6023 
6024 void MacroAssembler::load_aotrc_address(Register reg, address a) {
6025 #if INCLUDE_CDS
6026   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
6027   if (AOTCodeCache::is_on_for_dump()) {
6028     // all aotrc field addresses should be registered in the AOTCodeCache address table
6029     lea(reg, ExternalAddress(a));
6030   } else {
6031     mov(reg, (uint64_t)a);
6032   }
6033 #else
6034   ShouldNotReachHere();
6035 #endif
6036 }
6037 
6038 #ifdef ASSERT
6039 void MacroAssembler::build_frame(int framesize) {
6040   build_frame(framesize, false);
6041 }
6042 #endif
6043 
6044 void MacroAssembler::build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills)) {
6045   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
6046   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
6047   protect_return_address();
6048   if (framesize < ((1 << 9) + 2 * wordSize)) {
6049     sub(sp, sp, framesize);
6050     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
6051       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
6052       stp(rscratch1, rscratch1, Address(sp, framesize - 2 * wordSize));
6053     } else {
6054       stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
6055     }
6056     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
6057   } else {
6058     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
6059       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
6060       stp(rscratch1, rscratch1, Address(pre(sp, -2 * wordSize)));
6061     } else {
6062       stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6063     }
6064     if (PreserveFramePointer) mov(rfp, sp);
6065     if (framesize < ((1 << 12) + 2 * wordSize))
6066       sub(sp, sp, framesize - 2 * wordSize);
6067     else {
6068       mov(rscratch1, framesize - 2 * wordSize);
6069       sub(sp, sp, rscratch1);
6070     }
6071   }
6072   verify_cross_modify_fence_not_required();
6073 }
6074 
6075 void MacroAssembler::remove_frame(int framesize) {
6076   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
6077   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
6078   if (framesize < ((1 << 9) + 2 * wordSize)) {
6079     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
6080     add(sp, sp, framesize);
6081   } else {
6082     if (framesize < ((1 << 12) + 2 * wordSize))
6083       add(sp, sp, framesize - 2 * wordSize);
6084     else {
6085       mov(rscratch1, framesize - 2 * wordSize);
6086       add(sp, sp, rscratch1);
6087     }
6088     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6089   }
6090   authenticate_return_address();
6091 }
6092 
6093 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
6094   if (needs_stack_repair) {
6095     // The method has a scalarized entry point (where fields of value object arguments
6096     // are passed through registers and stack), and a non-scalarized entry point (where
6097     // value object arguments are given as oops). The non-scalarized entry point will
6098     // first load each field of value object arguments and store them in registers and on
6099     // the stack in a way compatible with the scalarized entry point. To do so, some extra
6100     // stack space might be reserved (if argument registers are not enough). On leaving the
6101     // method, this space must be freed.
6102     //
6103     // In case we used the non-scalarized entry point the stack looks like this:
6104     //
6105     // | Arguments from caller     |
6106     // |---------------------------|  <-- caller's SP
6107     // | Saved LR #1               |
6108     // | Saved FP #1               |
6109     // |---------------------------|
6110     // | Extension space for       |
6111     // |   inline arg (un)packing  |
6112     // |---------------------------|  <-- start of this method's frame
6113     // | Saved LR #2               |
6114     // | Saved FP #2               |
6115     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6116     // | sp_inc                    |
6117     // | method locals             |
6118     // |---------------------------|  <-- SP
6119     //
6120     // There are two copies of FP and LR on the stack. They will be identical at
6121     // first, but that can change.
6122     // If the caller has been deoptimized, LR #1 will be patched to point at the
6123     // deopt blob, and LR #2 will still point into the old method.
6124     // If the saved FP (x29) was not used as the frame pointer, but to store an
6125     // oop, the GC will be aware only of FP #1 as the spilled location of x29 and
6126     // will fix only this one. Overall, FP/LR #2 are not reliable and are simply
6127     // needed to add space between the extension space and the locals, as there
6128     // would be between the real arguments and the locals if we don't need to
6129     // do unpacking (from the scalarized entry point).
6130     //
6131     // When restoring, one must then load FP #1 into x29, and LR #1 into x30,
6132     // while keeping in mind that from the scalarized entry point, there will be
6133     // only one copy of each. Indeed, in the case we used the scalarized calling
6134     // convention, the stack looks like this:
6135     //
6136     // | Arguments from caller     |
6137     // |---------------------------|  <-- caller's SP / start of this method's frame
6138     // | Saved LR                  |
6139     // | Saved FP                  |
6140     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6141     // | sp_inc                    |
6142     // | method locals             |
6143     // |---------------------------|  <-- SP
6144     //
6145     // The sp_inc stack slot holds the total size of the frame including the
6146     // extension space minus two words for the saved FP and LR. That is how to
6147     // find FP/LR #1. This size is expressed in bytes. Be careful when using it
6148     // from C++ in pointer arithmetic; you might need to divide it by wordSize.
6149     //
6150     // One can find sp_inc since the start the method's frame is SP + initial_framesize.
6151 
6152     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
6153 
6154     ldr(rscratch1, Address(sp, sp_inc_offset));
6155     add(sp, sp, rscratch1);
6156     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6157   } else {
6158     remove_frame(initial_framesize);
6159   }
6160 }
6161 
6162 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
6163   int real_frame_size = frame_size + sp_inc;
6164   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
6165   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
6166   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
6167 
6168   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
6169 
6170   // Subtract two words for the saved FP and LR as these will be popped
6171   // separately. See remove_frame above.
6172   mov(rscratch1, real_frame_size - 2*wordSize);
6173   str(rscratch1, Address(sp, sp_inc_offset));
6174 }
6175 
6176 // This method counts leading positive bytes (highest bit not set) in provided byte array
6177 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6178     // Simple and most common case of aligned small array which is not at the
6179     // end of memory page is placed here. All other cases are in stub.
6180     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6181     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6182     assert_different_registers(ary1, len, result);
6183 
6184     mov(result, len);
6185     cmpw(len, 0);
6186     br(LE, DONE);
6187     cmpw(len, 4 * wordSize);
6188     br(GE, STUB_LONG); // size > 32 then go to stub
6189 
6190     int shift = 64 - exact_log2(os::vm_page_size());
6191     lsl(rscratch1, ary1, shift);
6192     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6193     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6194     br(CS, STUB); // at the end of page then go to stub

7077 // On other systems, the helper is a usual C function.
7078 //
7079 void MacroAssembler::get_thread(Register dst) {
7080   RegSet saved_regs =
7081     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
7082     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
7083 
7084   protect_return_address();
7085   push(saved_regs, sp);
7086 
7087   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
7088   blr(lr);
7089   if (dst != c_rarg0) {
7090     mov(dst, c_rarg0);
7091   }
7092 
7093   pop(saved_regs, sp);
7094   authenticate_return_address();
7095 }
7096 
7097 #ifdef COMPILER2
7098 // C2 compiled method's prolog code
7099 // Moved here from aarch64.ad to support Valhalla code below
7100 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
7101   if (C->clinit_barrier_on_entry()) {
7102     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
7103 
7104     Label L_skip_barrier;
7105 
7106     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
7107     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
7108     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
7109     bind(L_skip_barrier);
7110   }
7111 
7112   if (C->max_vector_size() > 0) {
7113     reinitialize_ptrue();
7114   }
7115 
7116   int bangsize = C->output()->bang_size_in_bytes();
7117   if (C->output()->need_stack_bang(bangsize))
7118     generate_stack_overflow_check(bangsize);
7119 
7120   // n.b. frame size includes space for return pc and rfp
7121   const long framesize = C->output()->frame_size_in_bytes();
7122   build_frame(framesize DEBUG_ONLY(COMMA sp_inc != 0));
7123 
7124   if (C->needs_stack_repair()) {
7125     save_stack_increment(sp_inc, framesize);
7126   }
7127 
7128   if (VerifyStackAtCalls) {
7129     Unimplemented();
7130   }
7131 }
7132 #endif // COMPILER2
7133 
7134 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
7135   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
7136   // An inline type might be returned. If fields are in registers we
7137   // need to allocate an inline type instance and initialize it with
7138   // the value of the fields.
7139   Label skip;
7140   // We only need a new buffered inline type if a new one is not returned
7141   tbz(r0, 0, skip);
7142   int call_offset = -1;
7143 
7144   // Be careful not to clobber r1-7 which hold returned fields
7145   // Also do not use callee-saved registers as these may be live in the interpreter
7146   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
7147 
7148   // The following code is similar to allocate_instance but has some slight differences,
7149   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
7150   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
7151   Label slow_case;
7152   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
7153   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
7154 
7155   if (vk != nullptr) {
7156     // Called from C1, where the return type is statically known.
7157     movptr(klass, (intptr_t)vk->get_InlineKlass());
7158     jint lh = vk->layout_helper();
7159     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
7160     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
7161       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
7162     } else {
7163       b(slow_case);
7164     }
7165   } else {
7166     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
7167     andr(klass, r0, -2);
7168     if (UseTLAB) {
7169       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
7170       tst(tmp2, Klass::_lh_instance_slow_path_bit);
7171       br(Assembler::NE, slow_case);
7172       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
7173     } else {
7174       b(slow_case);
7175     }
7176   }
7177   if (UseTLAB) {
7178     // 2. Initialize buffered inline instance header
7179     Register buffer_obj = r0;
7180     if (UseCompactObjectHeaders) {
7181       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7182       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7183     } else {
7184       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7185       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7186       store_klass_gap(buffer_obj, zr);
7187       if (vk == nullptr) {
7188         // store_klass corrupts klass, so save it for later use (interpreter case only).
7189         mov(tmp1, klass);
7190       }
7191       store_klass(buffer_obj, klass);
7192       klass = tmp1;
7193     }
7194     // 3. Initialize its fields with an inline class specific handler
7195     if (vk != nullptr) {
7196       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7197     } else {
7198       ldr(tmp1, Address(klass, InlineKlass::adr_members_offset()));
7199       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7200       blr(tmp1);
7201     }
7202 
7203     membar(Assembler::StoreStore);
7204     b(skip);
7205   } else {
7206     // Must have already branched to slow_case above.
7207     DEBUG_ONLY(should_not_reach_here());
7208   }
7209   bind(slow_case);
7210   // We failed to allocate a new inline type, fall back to a runtime
7211   // call. Some oop field may be live in some registers but we can't
7212   // tell. That runtime call will take care of preserving them
7213   // across a GC if there's one.
7214   mov(r0, r0_preserved);
7215 
7216   if (from_interpreter) {
7217     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
7218   } else {
7219     far_call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
7220     call_offset = offset();
7221   }
7222   membar(Assembler::StoreStore);
7223 
7224   bind(skip);
7225   return call_offset;
7226 }
7227 
7228 // Move a value between registers/stack slots and update the reg_state
7229 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7230   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7231   if (reg_state[to->value()] == reg_written) {
7232     return true; // Already written
7233   }
7234 
7235   if (from != to && bt != T_VOID) {
7236     if (reg_state[to->value()] == reg_readonly) {
7237       return false; // Not yet writable
7238     }
7239     if (from->is_reg()) {
7240       if (to->is_reg()) {
7241         if (from->is_Register() && to->is_Register()) {
7242           mov(to->as_Register(), from->as_Register());
7243         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7244           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7245         } else {
7246           ShouldNotReachHere();
7247         }
7248       } else {
7249         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7250         Address to_addr = Address(sp, st_off);
7251         if (from->is_FloatRegister()) {
7252           if (bt == T_DOUBLE) {
7253              strd(from->as_FloatRegister(), to_addr);
7254           } else {
7255              assert(bt == T_FLOAT, "must be float");
7256              strs(from->as_FloatRegister(), to_addr);
7257           }
7258         } else {
7259           str(from->as_Register(), to_addr);
7260         }
7261       }
7262     } else {
7263       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7264       if (to->is_reg()) {
7265         if (to->is_FloatRegister()) {
7266           if (bt == T_DOUBLE) {
7267             ldrd(to->as_FloatRegister(), from_addr);
7268           } else {
7269             assert(bt == T_FLOAT, "must be float");
7270             ldrs(to->as_FloatRegister(), from_addr);
7271           }
7272         } else {
7273           ldr(to->as_Register(), from_addr);
7274         }
7275       } else {
7276         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7277         ldr(rscratch1, from_addr);
7278         str(rscratch1, Address(sp, st_off));
7279       }
7280     }
7281   }
7282 
7283   // Update register states
7284   reg_state[from->value()] = reg_writable;
7285   reg_state[to->value()] = reg_written;
7286   return true;
7287 }
7288 
7289 // Calculate the extra stack space required for packing or unpacking inline
7290 // args and adjust the stack pointer
7291 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7292   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7293   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7294   assert(sp_inc > 0, "sanity");
7295 
7296   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7297   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7298 
7299   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7300   if (sp_inc < (1 << 9)) {
7301     sub(sp, sp, sp_inc);   // Fits in an immediate
7302   } else {
7303     mov(rscratch1, sp_inc);
7304     sub(sp, sp, rscratch1);
7305   }
7306 
7307   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7308 }
7309 
7310 // Read all fields from an inline type oop and store the values in registers/stack slots
7311 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7312                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7313                                           RegState reg_state[]) {
7314   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7315   assert(from->is_valid(), "source must be valid");
7316   bool progress = false;
7317 #ifdef ASSERT
7318   const int start_offset = offset();
7319 #endif
7320 
7321   Label L_null, L_notNull;
7322   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7323   Register tmp1 = r10;
7324   Register tmp2 = r11;
7325 
7326 #ifndef ASSERT
7327   RegSet clobbered_gp_regs = MacroAssembler::call_clobbered_gp_registers();
7328   assert(clobbered_gp_regs.contains(tmp1), "tmp1 must be saved explicitly if it's not a clobber");
7329   assert(clobbered_gp_regs.contains(tmp2), "tmp2 must be saved explicitly if it's not a clobber");
7330   assert(clobbered_gp_regs.contains(r14), "r14 must be saved explicitly if it's not a clobber");
7331 #endif
7332 
7333   Register fromReg = noreg;
7334   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
7335   bool done = true;
7336   bool mark_done = true;
7337   VMReg toReg;
7338   BasicType bt;
7339   // Check if argument requires a null check
7340   bool null_check = false;
7341   VMReg nullCheckReg;
7342   while (stream.next(nullCheckReg, bt)) {
7343     if (sig->at(stream.sig_index())._offset == -1) {
7344       null_check = true;
7345       break;
7346     }
7347   }
7348   stream.reset(sig_index, to_index);
7349   while (stream.next(toReg, bt)) {
7350     assert(toReg->is_valid(), "destination must be valid");
7351     int idx = (int)toReg->value();
7352     if (reg_state[idx] == reg_readonly) {
7353       if (idx != from->value()) {
7354         mark_done = false;
7355       }
7356       done = false;
7357       continue;
7358     } else if (reg_state[idx] == reg_written) {
7359       continue;
7360     }
7361     assert(reg_state[idx] == reg_writable, "must be writable");
7362     reg_state[idx] = reg_written;
7363     progress = true;
7364 
7365     if (fromReg == noreg) {
7366       if (from->is_reg()) {
7367         fromReg = from->as_Register();
7368       } else {
7369         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7370         ldr(tmp1, Address(sp, st_off));
7371         fromReg = tmp1;
7372       }
7373       if (null_check) {
7374         // Nullable inline type argument, emit null check
7375         cbz(fromReg, L_null);
7376       }
7377     }
7378     int off = sig->at(stream.sig_index())._offset;
7379     if (off == -1) {
7380       assert(null_check, "Missing null check at");
7381       if (toReg->is_stack()) {
7382         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7383         mov(tmp2, 1);
7384         str(tmp2, Address(sp, st_off));
7385       } else {
7386         mov(toReg->as_Register(), 1);
7387       }
7388       continue;
7389     }
7390     if (sig->at(stream.sig_index())._vt_oop) {
7391       if (toReg->is_stack()) {
7392         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7393         str(fromReg, Address(sp, st_off));
7394       } else {
7395         mov(toReg->as_Register(), fromReg);
7396       }
7397       continue;
7398     }
7399     assert(off > 0, "offset in object should be positive");
7400     Address fromAddr = Address(fromReg, off);
7401     if (!toReg->is_FloatRegister()) {
7402       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7403       if (is_reference_type(bt)) {
7404         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7405       } else {
7406         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7407         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7408       }
7409       if (toReg->is_stack()) {
7410         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7411         str(dst, Address(sp, st_off));
7412       }
7413     } else if (bt == T_DOUBLE) {
7414       ldrd(toReg->as_FloatRegister(), fromAddr);
7415     } else {
7416       assert(bt == T_FLOAT, "must be float");
7417       ldrs(toReg->as_FloatRegister(), fromAddr);
7418     }
7419   }
7420   if (progress && null_check) {
7421     if (done) {
7422       b(L_notNull);
7423       bind(L_null);
7424       // Set null marker to zero to signal that the argument is null.
7425       // Also set all fields to zero since the runtime requires a canonical
7426       // representation of a flat null.
7427       stream.reset(sig_index, to_index);
7428       while (stream.next(toReg, bt)) {
7429         if (toReg->is_stack()) {
7430           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7431           str(zr, Address(sp, st_off));
7432         } else if (toReg->is_FloatRegister()) {
7433           mov(toReg->as_FloatRegister(), T2S, 0);
7434         } else {
7435           mov(toReg->as_Register(), zr);
7436         }
7437       }
7438       bind(L_notNull);
7439     } else {
7440       bind(L_null);
7441     }
7442   }
7443 
7444   sig_index = stream.sig_index();
7445   to_index = stream.regs_index();
7446 
7447   if (mark_done && reg_state[from->value()] != reg_written) {
7448     // This is okay because no one else will write to that slot
7449     reg_state[from->value()] = reg_writable;
7450   }
7451   from_index--;
7452   assert(progress || (start_offset == offset()), "should not emit code");
7453   return done;
7454 }
7455 
7456 // Pack fields back into an inline type oop
7457 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7458                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7459                                         RegState reg_state[], Register val_array) {
7460   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7461   assert(to->is_valid(), "destination must be valid");
7462 
7463   if (reg_state[to->value()] == reg_written) {
7464     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7465     return true; // Already written
7466   }
7467 
7468   // The GC barrier expanded by store_heap_oop below may call into the
7469   // runtime so use callee-saved registers for any values that need to be
7470   // preserved. The GC barrier assembler should take care of saving the
7471   // Java argument registers.
7472   // TODO 8284443 Isn't it an issue if below code uses r14 as tmp when it contains a spilled value?
7473   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7474   Register val_obj_tmp = r21;
7475   Register from_reg_tmp = r22;
7476   Register tmp1 = r14;
7477   Register tmp2 = r13;
7478   Register tmp3 = r12;
7479   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7480 
7481   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7482 
7483   if (reg_state[to->value()] == reg_readonly) {
7484     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7485       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7486       return false; // Not yet writable
7487     }
7488     val_obj = val_obj_tmp;
7489   }
7490 
7491   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7492   VMReg fromReg;
7493   BasicType bt;
7494   Label L_null;
7495   while (stream.next(fromReg, bt)) {
7496     assert(fromReg->is_valid(), "source must be valid");
7497     reg_state[fromReg->value()] = reg_writable;
7498 
7499     int off = sig->at(stream.sig_index())._offset;
7500     if (off == -1) {
7501       // Nullable inline type argument, emit null check
7502       Label L_notNull;
7503       if (fromReg->is_stack()) {
7504         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7505         ldrb(tmp2, Address(sp, ld_off));
7506         cbnz(tmp2, L_notNull);
7507       } else {
7508         cbnz(fromReg->as_Register(), L_notNull);
7509       }
7510       mov(val_obj, 0);
7511       b(L_null);
7512       bind(L_notNull);
7513       continue;
7514     }
7515     if (sig->at(stream.sig_index())._vt_oop) {
7516       if (fromReg->is_stack()) {
7517         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7518         ldr(val_obj, Address(sp, ld_off));
7519       } else {
7520         mov(val_obj, fromReg->as_Register());
7521       }
7522       cbnz(val_obj, L_null);
7523       // get the buffer from the just allocated pool of buffers
7524       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7525       load_heap_oop(val_obj, Address(val_array, index), rscratch1, rscratch2);
7526       continue;
7527     }
7528 
7529     assert(off > 0, "offset in object should be positive");
7530     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7531 
7532     // Pack the scalarized field into the value object.
7533     Address dst(val_obj, off);
7534     if (!fromReg->is_FloatRegister()) {
7535       Register src;
7536       if (fromReg->is_stack()) {
7537         src = from_reg_tmp;
7538         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7539         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7540       } else {
7541         src = fromReg->as_Register();
7542       }
7543       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7544       if (is_reference_type(bt)) {
7545         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
7546         mov(tmp3, val_obj);
7547         Address dst_with_tmp3(tmp3, off);
7548         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7549       } else {
7550         store_sized_value(dst, src, size_in_bytes);
7551       }
7552     } else if (bt == T_DOUBLE) {
7553       strd(fromReg->as_FloatRegister(), dst);
7554     } else {
7555       assert(bt == T_FLOAT, "must be float");
7556       strs(fromReg->as_FloatRegister(), dst);
7557     }
7558   }
7559   bind(L_null);
7560   sig_index = stream.sig_index();
7561   from_index = stream.regs_index();
7562 
7563   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7564   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7565   assert(success, "to register must be writeable");
7566   return true;
7567 }
7568 
7569 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7570   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7571 }
7572 
7573 void MacroAssembler::cache_wb(Address line) {
7574   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7575   assert(line.index() == noreg, "index should be noreg");
7576   assert(line.offset() == 0, "offset should be 0");
7577   // would like to assert this
7578   // assert(line._ext.shift == 0, "shift should be zero");
7579   if (VM_Version::supports_dcpop()) {
7580     // writeback using clear virtual address to point of persistence
7581     dc(Assembler::CVAP, line.base());
7582   } else {
7583     // no need to generate anything as Unsafe.writebackMemory should
7584     // never invoke this stub
7585   }
7586 }
7587 
7588 void MacroAssembler::cache_wbsync(bool is_pre) {
7589   // we only need a barrier post sync
7590   if (!is_pre) {
7591     membar(Assembler::AnyAny);
7592   }

7988   }
7989 
7990   // Check if the lock-stack is full.
7991   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7992   cmpw(top, (unsigned)LockStack::end_offset());
7993   br(Assembler::GE, slow);
7994 
7995   // Check for recursion.
7996   subw(t, top, oopSize);
7997   ldr(t, Address(rthread, t));
7998   cmp(obj, t);
7999   br(Assembler::EQ, push);
8000 
8001   // Check header for monitor (0b10).
8002   tst(mark, markWord::monitor_value);
8003   br(Assembler::NE, slow);
8004 
8005   // Try to lock. Transition lock bits 0b01 => 0b00
8006   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
8007   orr(mark, mark, markWord::unlocked_value);
8008   // Mask inline_type bit such that we go to the slow path if object is an inline type
8009   andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
8010 
8011   eor(t, mark, markWord::unlocked_value);
8012   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
8013           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
8014   br(Assembler::NE, slow);
8015 
8016   bind(push);
8017   // After successful lock, push object on lock-stack.
8018   str(obj, Address(rthread, top));
8019   addw(top, top, oopSize);
8020   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
8021 }
8022 
8023 // Implements fast-unlocking.
8024 //
8025 // - obj: the object to be unlocked
8026 // - t1, t2, t3: temporary registers
8027 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
8028 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
8029   // cmpxchg clobbers rscratch1.
8030   assert_different_registers(obj, t1, t2, t3, rscratch1);
< prev index next >