< 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 

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) return;




2032 
2033   // Pass register number to verify_oop_subroutine
2034   const char* b = nullptr;
2035   {
2036     ResourceMark rm;
2037     stringStream ss;
2038     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2039     b = code_string(ss.as_string());
2040   }
2041   BLOCK_COMMENT("verify_oop {");
2042 
2043   strip_return_address(); // This might happen within a stack frame.
2044   protect_return_address();
2045   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2046   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2047 
2048   mov(r0, reg);
2049   movptr(rscratch1, (uintptr_t)(address)b);
2050 
2051   // call indirectly to solve generation ordering problem
2052   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2053   ldr(rscratch2, Address(rscratch2));
2054   blr(rscratch2);
2055 
2056   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2057   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2058   authenticate_return_address();
2059 
2060   BLOCK_COMMENT("} verify_oop");
2061 }
2062 
2063 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2064   if (!VerifyOops) return;




2065 
2066   const char* b = nullptr;
2067   {
2068     ResourceMark rm;
2069     stringStream ss;
2070     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2071     b = code_string(ss.as_string());
2072   }
2073   BLOCK_COMMENT("verify_oop_addr {");
2074 
2075   strip_return_address(); // This might happen within a stack frame.
2076   protect_return_address();
2077   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2078   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2079 
2080   // addr may contain sp so we will have to adjust it based on the
2081   // pushes that we just did.
2082   if (addr.uses(sp)) {
2083     lea(r0, addr);
2084     ldr(r0, Address(r0, 4 * wordSize));

2142   call_VM_leaf_base(entry_point, 1);
2143 }
2144 
2145 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2146   assert_different_registers(arg_1, c_rarg0);
2147   pass_arg0(this, arg_0);
2148   pass_arg1(this, arg_1);
2149   call_VM_leaf_base(entry_point, 2);
2150 }
2151 
2152 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2153                                   Register arg_1, Register arg_2) {
2154   assert_different_registers(arg_1, c_rarg0);
2155   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2156   pass_arg0(this, arg_0);
2157   pass_arg1(this, arg_1);
2158   pass_arg2(this, arg_2);
2159   call_VM_leaf_base(entry_point, 3);
2160 }
2161 




2162 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2163   pass_arg0(this, arg_0);
2164   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2165 }
2166 
2167 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2168 
2169   assert_different_registers(arg_0, c_rarg1);
2170   pass_arg1(this, arg_1);
2171   pass_arg0(this, arg_0);
2172   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2173 }
2174 
2175 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2176   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2177   assert_different_registers(arg_1, c_rarg2);
2178   pass_arg2(this, arg_2);
2179   pass_arg1(this, arg_1);
2180   pass_arg0(this, arg_0);
2181   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2187   assert_different_registers(arg_2, c_rarg3);
2188   pass_arg3(this, arg_3);
2189   pass_arg2(this, arg_2);
2190   pass_arg1(this, arg_1);
2191   pass_arg0(this, arg_0);
2192   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2193 }
2194 
2195 void MacroAssembler::null_check(Register reg, int offset) {
2196   if (needs_explicit_null_check(offset)) {
2197     // provoke OS null exception if reg is null by
2198     // accessing M[reg] w/o changing any registers
2199     // NOTE: this is plenty to provoke a segv
2200     ldr(zr, Address(reg));
2201   } else {
2202     // nothing to do, (later) access of M[reg + offset]
2203     // will provoke OS null exception if reg is null
2204   }
2205 }
2206 
























































































2207 // MacroAssembler protected routines needed to implement
2208 // public methods
2209 
2210 void MacroAssembler::mov(Register r, Address dest) {
2211   code_section()->relocate(pc(), dest.rspec());
2212   uint64_t imm64 = (uint64_t)dest.target();
2213   movptr(r, imm64);
2214 }
2215 
2216 // Move a constant pointer into r.  In AArch64 mode the virtual
2217 // address space is 48 bits in size, so we only need three
2218 // instructions to create a patchable instruction sequence that can
2219 // reach anywhere.
2220 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2221 #ifndef PRODUCT
2222   {
2223     char buffer[64];
2224     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2225     block_comment(buffer);
2226   }

4887   adrp(rscratch1, src2, offset);
4888   ldr(rscratch1, Address(rscratch1, offset));
4889   cmp(src1, rscratch1);
4890 }
4891 
4892 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
4893   cmp(obj1, obj2);
4894 }
4895 
4896 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
4897   load_method_holder(rresult, rmethod);
4898   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
4899 }
4900 
4901 void MacroAssembler::load_method_holder(Register holder, Register method) {
4902   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
4903   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
4904   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
4905 }
4906 










4907 // Loads the obj's Klass* into dst.
4908 // Preserves all registers (incl src, rscratch1 and rscratch2).
4909 // Input:
4910 // src - the oop we want to load the klass from.
4911 // dst - output narrow klass.
4912 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
4913   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
4914   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
4915   lsr(dst, dst, markWord::klass_shift);
4916 }
4917 
4918 void MacroAssembler::load_klass(Register dst, Register src) {
4919   if (UseCompactObjectHeaders) {
4920     load_narrow_klass_compact(dst, src);
4921     decode_klass_not_null(dst);
4922   } else if (UseCompressedClassPointers) {
4923     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
4924     decode_klass_not_null(dst);
4925   } else {
4926     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));

4997   }
4998   cmp(klass, tmp);
4999 }
5000 
5001 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5002   if (UseCompactObjectHeaders) {
5003     load_narrow_klass_compact(tmp1, obj1);
5004     load_narrow_klass_compact(tmp2,  obj2);
5005     cmpw(tmp1, tmp2);
5006   } else if (UseCompressedClassPointers) {
5007     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5008     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5009     cmpw(tmp1, tmp2);
5010   } else {
5011     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5012     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5013     cmp(tmp1, tmp2);
5014   }
5015 }
5016 





5017 void MacroAssembler::store_klass(Register dst, Register src) {
5018   // FIXME: Should this be a store release?  concurrent gcs assumes
5019   // klass length is valid if klass field is not null.
5020   assert(!UseCompactObjectHeaders, "not with compact headers");
5021   if (UseCompressedClassPointers) {
5022     encode_klass_not_null(src);
5023     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5024   } else {
5025     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5026   }
5027 }
5028 
5029 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5030   assert(!UseCompactObjectHeaders, "not with compact headers");
5031   if (UseCompressedClassPointers) {
5032     // Store to klass gap in destination
5033     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5034   }
5035 }
5036 

5398   if (as_raw) {
5399     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5400   } else {
5401     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5402   }
5403 }
5404 
5405 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5406                                      Address dst, Register val,
5407                                      Register tmp1, Register tmp2, Register tmp3) {
5408   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5409   decorators = AccessInternal::decorator_fixup(decorators, type);
5410   bool as_raw = (decorators & AS_RAW) != 0;
5411   if (as_raw) {
5412     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5413   } else {
5414     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5415   }
5416 }
5417 








































5418 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5419                                    Register tmp2, DecoratorSet decorators) {
5420   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5421 }
5422 
5423 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5424                                             Register tmp2, DecoratorSet decorators) {
5425   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5426 }
5427 
5428 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5429                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5430   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5431 }
5432 
5433 // Used for storing nulls.
5434 void MacroAssembler::store_heap_oop_null(Address dst) {
5435   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5436 }
5437 

5473     oop_index = oop_recorder()->allocate_metadata_index(obj);
5474   } else {
5475     oop_index = oop_recorder()->find_index(obj);
5476   }
5477   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5478   mov(dst, Address((address)obj, rspec));
5479 }
5480 
5481 Address MacroAssembler::constant_oop_address(jobject obj) {
5482 #ifdef ASSERT
5483   {
5484     ThreadInVMfromUnknown tiv;
5485     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5486     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5487   }
5488 #endif
5489   int oop_index = oop_recorder()->find_index(obj);
5490   return Address((address)obj, oop_Relocation::spec(oop_index));
5491 }
5492 






































































































5493 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5494 void MacroAssembler::tlab_allocate(Register obj,
5495                                    Register var_size_in_bytes,
5496                                    int con_size_in_bytes,
5497                                    Register t1,
5498                                    Register t2,
5499                                    Label& slow_case) {
5500   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5501   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5502 }
5503 
5504 void MacroAssembler::verify_tlab() {
5505 #ifdef ASSERT
5506   if (UseTLAB && VerifyOops) {
5507     Label next, ok;
5508 
5509     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5510 
5511     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5512     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5513     cmp(rscratch2, rscratch1);
5514     br(Assembler::HS, next);
5515     STOP("assert(top >= start)");
5516     should_not_reach_here();
5517 
5518     bind(next);
5519     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5520     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5521     cmp(rscratch2, rscratch1);
5522     br(Assembler::HS, ok);
5523     STOP("assert(top <= end)");
5524     should_not_reach_here();
5525 
5526     bind(ok);
5527     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5528   }
5529 #endif
5530 }
5531 




















5532 // Writes to stack successive pages until offset reached to check for
5533 // stack overflow + shadow pages.  This clobbers tmp.
5534 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5535   assert_different_registers(tmp, size, rscratch1);
5536   mov(tmp, sp);
5537   // Bang stack for total size given plus shadow page size.
5538   // Bang one page at a time because large size can bang beyond yellow and
5539   // red zones.
5540   Label loop;
5541   mov(rscratch1, (int)os::vm_page_size());
5542   bind(loop);
5543   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5544   subsw(size, size, rscratch1);
5545   str(size, Address(tmp));
5546   br(Assembler::GT, loop);
5547 
5548   // Bang down shadow pages too.
5549   // At this point, (tmp-0) is the last address touched, so don't
5550   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5551   // was post-decremented.)  Skip this address by starting at i=1, and

5637 }
5638 
5639 void MacroAssembler::remove_frame(int framesize) {
5640   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5641   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5642   if (framesize < ((1 << 9) + 2 * wordSize)) {
5643     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5644     add(sp, sp, framesize);
5645   } else {
5646     if (framesize < ((1 << 12) + 2 * wordSize))
5647       add(sp, sp, framesize - 2 * wordSize);
5648     else {
5649       mov(rscratch1, framesize - 2 * wordSize);
5650       add(sp, sp, rscratch1);
5651     }
5652     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5653   }
5654   authenticate_return_address();
5655 }
5656 


































































5657 
5658 // This method counts leading positive bytes (highest bit not set) in provided byte array
5659 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5660     // Simple and most common case of aligned small array which is not at the
5661     // end of memory page is placed here. All other cases are in stub.
5662     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5663     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5664     assert_different_registers(ary1, len, result);
5665 
5666     mov(result, len);
5667     cmpw(len, 0);
5668     br(LE, DONE);
5669     cmpw(len, 4 * wordSize);
5670     br(GE, STUB_LONG); // size > 32 then go to stub
5671 
5672     int shift = 64 - exact_log2(os::vm_page_size());
5673     lsl(rscratch1, ary1, shift);
5674     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5675     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5676     br(CS, STUB); // at the end of page then go to stub

6554 // On other systems, the helper is a usual C function.
6555 //
6556 void MacroAssembler::get_thread(Register dst) {
6557   RegSet saved_regs =
6558     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6559     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6560 
6561   protect_return_address();
6562   push(saved_regs, sp);
6563 
6564   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6565   blr(lr);
6566   if (dst != c_rarg0) {
6567     mov(dst, c_rarg0);
6568   }
6569 
6570   pop(saved_regs, sp);
6571   authenticate_return_address();
6572 }
6573 
































































































































































































































































































































































































































































6574 void MacroAssembler::cache_wb(Address line) {
6575   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6576   assert(line.index() == noreg, "index should be noreg");
6577   assert(line.offset() == 0, "offset should be 0");
6578   // would like to assert this
6579   // assert(line._ext.shift == 0, "shift should be zero");
6580   if (VM_Version::supports_dcpop()) {
6581     // writeback using clear virtual address to point of persistence
6582     dc(Assembler::CVAP, line.base());
6583   } else {
6584     // no need to generate anything as Unsafe.writebackMemory should
6585     // never invoke this stub
6586   }
6587 }
6588 
6589 void MacroAssembler::cache_wbsync(bool is_pre) {
6590   // we only need a barrier post sync
6591   if (!is_pre) {
6592     membar(Assembler::AnyAny);
6593   }

6964   }
6965 
6966   // Check if the lock-stack is full.
6967   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
6968   cmpw(top, (unsigned)LockStack::end_offset());
6969   br(Assembler::GE, slow);
6970 
6971   // Check for recursion.
6972   subw(t, top, oopSize);
6973   ldr(t, Address(rthread, t));
6974   cmp(obj, t);
6975   br(Assembler::EQ, push);
6976 
6977   // Check header for monitor (0b10).
6978   tst(mark, markWord::monitor_value);
6979   br(Assembler::NE, slow);
6980 
6981   // Try to lock. Transition lock bits 0b01 => 0b00
6982   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
6983   orr(mark, mark, markWord::unlocked_value);



6984   eor(t, mark, markWord::unlocked_value);
6985   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
6986           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
6987   br(Assembler::NE, slow);
6988 
6989   bind(push);
6990   // After successful lock, push object on lock-stack.
6991   str(obj, Address(rthread, top));
6992   addw(top, top, oopSize);
6993   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
6994 }
6995 
6996 // Implements lightweight-unlocking.
6997 //
6998 // - obj: the object to be unlocked
6999 // - t1, t2, t3: temporary registers
7000 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7001 void MacroAssembler::lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7002   // cmpxchg clobbers rscratch1.
7003   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/continuation.hpp"
  53 #include "runtime/globals.hpp"
  54 #include "runtime/icache.hpp"
  55 #include "runtime/interfaceSupport.inline.hpp"
  56 #include "runtime/javaThread.hpp"
  57 #include "runtime/jniHandles.inline.hpp"
  58 #include "runtime/sharedRuntime.hpp"
  59 #include "runtime/signature_cc.hpp"
  60 #include "runtime/stubRoutines.hpp"
  61 #include "utilities/globalDefinitions.hpp"
  62 #include "utilities/powerOfTwo.hpp"
  63 #include "vmreg_aarch64.inline.hpp"
  64 #ifdef COMPILER1
  65 #include "c1/c1_LIRAssembler.hpp"
  66 #endif
  67 #ifdef COMPILER2
  68 #include "oops/oop.hpp"
  69 #include "opto/compile.hpp"
  70 #include "opto/node.hpp"
  71 #include "opto/output.hpp"
  72 #endif
  73 
  74 #include <sys/types.h>
  75 
  76 #ifdef PRODUCT
  77 #define BLOCK_COMMENT(str) /* nothing */
  78 #else
  79 #define BLOCK_COMMENT(str) block_comment(str)
  80 #endif
  81 #define STOP(str) stop(str);
  82 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  83 

2016   ldarb(scratch, scratch);
2017   cmp(scratch, InstanceKlass::fully_initialized);
2018   br(Assembler::EQ, *L_fast_path);
2019 
2020   // Fast path check: current thread is initializer thread
2021   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2022   cmp(rthread, scratch);
2023 
2024   if (L_slow_path == &L_fallthrough) {
2025     br(Assembler::EQ, *L_fast_path);
2026     bind(*L_slow_path);
2027   } else if (L_fast_path == &L_fallthrough) {
2028     br(Assembler::NE, *L_slow_path);
2029     bind(*L_fast_path);
2030   } else {
2031     Unimplemented();
2032   }
2033 }
2034 
2035 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2036   if (!VerifyOops || VerifyAdapterSharing) {
2037     // Below address of the code string confuses VerifyAdapterSharing
2038     // because it may differ between otherwise equivalent adapters.
2039     return;
2040   }
2041 
2042   // Pass register number to verify_oop_subroutine
2043   const char* b = nullptr;
2044   {
2045     ResourceMark rm;
2046     stringStream ss;
2047     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2048     b = code_string(ss.as_string());
2049   }
2050   BLOCK_COMMENT("verify_oop {");
2051 
2052   strip_return_address(); // This might happen within a stack frame.
2053   protect_return_address();
2054   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2055   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2056 
2057   mov(r0, reg);
2058   movptr(rscratch1, (uintptr_t)(address)b);
2059 
2060   // call indirectly to solve generation ordering problem
2061   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2062   ldr(rscratch2, Address(rscratch2));
2063   blr(rscratch2);
2064 
2065   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2066   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2067   authenticate_return_address();
2068 
2069   BLOCK_COMMENT("} verify_oop");
2070 }
2071 
2072 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2073   if (!VerifyOops || VerifyAdapterSharing) {
2074     // Below address of the code string confuses VerifyAdapterSharing
2075     // because it may differ between otherwise equivalent adapters.
2076     return;
2077   }
2078 
2079   const char* b = nullptr;
2080   {
2081     ResourceMark rm;
2082     stringStream ss;
2083     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2084     b = code_string(ss.as_string());
2085   }
2086   BLOCK_COMMENT("verify_oop_addr {");
2087 
2088   strip_return_address(); // This might happen within a stack frame.
2089   protect_return_address();
2090   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2091   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2092 
2093   // addr may contain sp so we will have to adjust it based on the
2094   // pushes that we just did.
2095   if (addr.uses(sp)) {
2096     lea(r0, addr);
2097     ldr(r0, Address(r0, 4 * wordSize));

2155   call_VM_leaf_base(entry_point, 1);
2156 }
2157 
2158 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2159   assert_different_registers(arg_1, c_rarg0);
2160   pass_arg0(this, arg_0);
2161   pass_arg1(this, arg_1);
2162   call_VM_leaf_base(entry_point, 2);
2163 }
2164 
2165 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2166                                   Register arg_1, Register arg_2) {
2167   assert_different_registers(arg_1, c_rarg0);
2168   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2169   pass_arg0(this, arg_0);
2170   pass_arg1(this, arg_1);
2171   pass_arg2(this, arg_2);
2172   call_VM_leaf_base(entry_point, 3);
2173 }
2174 
2175 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2176   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2177 }
2178 
2179 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2180   pass_arg0(this, arg_0);
2181   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2182 }
2183 
2184 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2185 
2186   assert_different_registers(arg_0, c_rarg1);
2187   pass_arg1(this, arg_1);
2188   pass_arg0(this, arg_0);
2189   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2190 }
2191 
2192 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2193   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2194   assert_different_registers(arg_1, c_rarg2);
2195   pass_arg2(this, arg_2);
2196   pass_arg1(this, arg_1);
2197   pass_arg0(this, arg_0);
2198   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2204   assert_different_registers(arg_2, c_rarg3);
2205   pass_arg3(this, arg_3);
2206   pass_arg2(this, arg_2);
2207   pass_arg1(this, arg_1);
2208   pass_arg0(this, arg_0);
2209   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2210 }
2211 
2212 void MacroAssembler::null_check(Register reg, int offset) {
2213   if (needs_explicit_null_check(offset)) {
2214     // provoke OS null exception if reg is null by
2215     // accessing M[reg] w/o changing any registers
2216     // NOTE: this is plenty to provoke a segv
2217     ldr(zr, Address(reg));
2218   } else {
2219     // nothing to do, (later) access of M[reg + offset]
2220     // will provoke OS null exception if reg is null
2221   }
2222 }
2223 
2224 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2225   assert_different_registers(markword, rscratch2);
2226   mov(rscratch2, markWord::inline_type_mask_in_place);
2227   andr(markword, markword, rscratch2);
2228   mov(rscratch2, markWord::inline_type_pattern);
2229   cmp(markword, rscratch2);
2230   br(Assembler::EQ, is_inline_type);
2231 }
2232 
2233 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
2234   assert_different_registers(tmp, rscratch1);
2235   if (can_be_null) {
2236     cbz(object, not_inline_type);
2237   }
2238   const int is_inline_type_mask = markWord::inline_type_pattern;
2239   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2240   mov(rscratch1, is_inline_type_mask);
2241   andr(tmp, tmp, rscratch1);
2242   cmp(tmp, rscratch1);
2243   br(Assembler::NE, not_inline_type);
2244 }
2245 
2246 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2247   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2248   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2249 }
2250 
2251 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2252   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2253   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2254 }
2255 
2256 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2257   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2258   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2259 }
2260 
2261 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
2262   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2263   tbnz(flags, ResolvedFieldEntry::has_null_marker_shift, has_null_marker);
2264 }
2265 
2266 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2267   Label test_mark_word;
2268   // load mark word
2269   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2270   // check displaced
2271   tst(temp_reg, markWord::unlocked_value);
2272   br(Assembler::NE, test_mark_word);
2273   // slow path use klass prototype
2274   load_prototype_header(temp_reg, oop);
2275 
2276   bind(test_mark_word);
2277   andr(temp_reg, temp_reg, test_bit);
2278   if (jmp_set) {
2279     cbnz(temp_reg, jmp_label);
2280   } else {
2281     cbz(temp_reg, jmp_label);
2282   }
2283 }
2284 
2285 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2286   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2287 }
2288 
2289 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2290                                                   Label&is_non_flat_array) {
2291   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2292 }
2293 
2294 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2295   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2296 }
2297 
2298 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2299   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2300 }
2301 
2302 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2303   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2304   br(Assembler::NE, is_flat_array);
2305 }
2306 
2307 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
2308   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2309   br(Assembler::EQ, is_non_flat_array);
2310 }
2311 
2312 // MacroAssembler protected routines needed to implement
2313 // public methods
2314 
2315 void MacroAssembler::mov(Register r, Address dest) {
2316   code_section()->relocate(pc(), dest.rspec());
2317   uint64_t imm64 = (uint64_t)dest.target();
2318   movptr(r, imm64);
2319 }
2320 
2321 // Move a constant pointer into r.  In AArch64 mode the virtual
2322 // address space is 48 bits in size, so we only need three
2323 // instructions to create a patchable instruction sequence that can
2324 // reach anywhere.
2325 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2326 #ifndef PRODUCT
2327   {
2328     char buffer[64];
2329     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2330     block_comment(buffer);
2331   }

4992   adrp(rscratch1, src2, offset);
4993   ldr(rscratch1, Address(rscratch1, offset));
4994   cmp(src1, rscratch1);
4995 }
4996 
4997 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
4998   cmp(obj1, obj2);
4999 }
5000 
5001 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5002   load_method_holder(rresult, rmethod);
5003   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5004 }
5005 
5006 void MacroAssembler::load_method_holder(Register holder, Register method) {
5007   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5008   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5009   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5010 }
5011 
5012 void MacroAssembler::load_metadata(Register dst, Register src) {
5013   if (UseCompactObjectHeaders) {
5014     load_narrow_klass_compact(dst, src);
5015   } else if (UseCompressedClassPointers) {
5016     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5017   } else {
5018     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5019   }
5020 }
5021 
5022 // Loads the obj's Klass* into dst.
5023 // Preserves all registers (incl src, rscratch1 and rscratch2).
5024 // Input:
5025 // src - the oop we want to load the klass from.
5026 // dst - output narrow klass.
5027 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5028   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5029   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5030   lsr(dst, dst, markWord::klass_shift);
5031 }
5032 
5033 void MacroAssembler::load_klass(Register dst, Register src) {
5034   if (UseCompactObjectHeaders) {
5035     load_narrow_klass_compact(dst, src);
5036     decode_klass_not_null(dst);
5037   } else if (UseCompressedClassPointers) {
5038     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5039     decode_klass_not_null(dst);
5040   } else {
5041     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));

5112   }
5113   cmp(klass, tmp);
5114 }
5115 
5116 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5117   if (UseCompactObjectHeaders) {
5118     load_narrow_klass_compact(tmp1, obj1);
5119     load_narrow_klass_compact(tmp2,  obj2);
5120     cmpw(tmp1, tmp2);
5121   } else if (UseCompressedClassPointers) {
5122     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5123     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5124     cmpw(tmp1, tmp2);
5125   } else {
5126     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5127     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5128     cmp(tmp1, tmp2);
5129   }
5130 }
5131 
5132 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5133   load_klass(dst, src);
5134   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5135 }
5136 
5137 void MacroAssembler::store_klass(Register dst, Register src) {
5138   // FIXME: Should this be a store release?  concurrent gcs assumes
5139   // klass length is valid if klass field is not null.
5140   assert(!UseCompactObjectHeaders, "not with compact headers");
5141   if (UseCompressedClassPointers) {
5142     encode_klass_not_null(src);
5143     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5144   } else {
5145     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5146   }
5147 }
5148 
5149 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5150   assert(!UseCompactObjectHeaders, "not with compact headers");
5151   if (UseCompressedClassPointers) {
5152     // Store to klass gap in destination
5153     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5154   }
5155 }
5156 

5518   if (as_raw) {
5519     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5520   } else {
5521     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5522   }
5523 }
5524 
5525 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5526                                      Address dst, Register val,
5527                                      Register tmp1, Register tmp2, Register tmp3) {
5528   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5529   decorators = AccessInternal::decorator_fixup(decorators, type);
5530   bool as_raw = (decorators & AS_RAW) != 0;
5531   if (as_raw) {
5532     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5533   } else {
5534     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5535   }
5536 }
5537 
5538 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5539                                      Register inline_layout_info) {
5540   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5541   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5542 }
5543 
5544 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5545   ldr(offset, Address(inline_klass, InstanceKlass::adr_inlineklass_fixed_block_offset()));
5546   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5547 }
5548 
5549 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5550   // ((address) (void*) o) + vk->payload_offset();
5551   Register offset = (data == oop) ? rscratch1 : data;
5552   payload_offset(inline_klass, offset);
5553   if (data == oop) {
5554     add(data, data, offset);
5555   } else {
5556     lea(data, Address(oop, offset));
5557   }
5558 }
5559 
5560 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
5561                                                 Register index, Register data) {
5562   assert_different_registers(array, array_klass, index);
5563   assert_different_registers(rscratch1, array, index);
5564 
5565   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
5566   ldrw(rscratch1, Address(array_klass, Klass::layout_helper_offset()));
5567 
5568   // Klass::layout_helper_log2_element_size(lh)
5569   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
5570   lsr(rscratch1, rscratch1, Klass::_lh_log2_element_size_shift);
5571   andr(rscratch1, rscratch1, Klass::_lh_log2_element_size_mask);
5572   lslv(index, index, rscratch1);
5573 
5574   add(data, array, index);
5575   add(data, data, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT));
5576 }
5577 
5578 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5579                                    Register tmp2, DecoratorSet decorators) {
5580   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5581 }
5582 
5583 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5584                                             Register tmp2, DecoratorSet decorators) {
5585   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5586 }
5587 
5588 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5589                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5590   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5591 }
5592 
5593 // Used for storing nulls.
5594 void MacroAssembler::store_heap_oop_null(Address dst) {
5595   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5596 }
5597 

5633     oop_index = oop_recorder()->allocate_metadata_index(obj);
5634   } else {
5635     oop_index = oop_recorder()->find_index(obj);
5636   }
5637   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5638   mov(dst, Address((address)obj, rspec));
5639 }
5640 
5641 Address MacroAssembler::constant_oop_address(jobject obj) {
5642 #ifdef ASSERT
5643   {
5644     ThreadInVMfromUnknown tiv;
5645     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5646     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5647   }
5648 #endif
5649   int oop_index = oop_recorder()->find_index(obj);
5650   return Address((address)obj, oop_Relocation::spec(oop_index));
5651 }
5652 
5653 // Object / value buffer allocation...
5654 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
5655                                        Register t1, Register t2,
5656                                        bool clear_fields, Label& alloc_failed)
5657 {
5658   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
5659   Register layout_size = t1;
5660   assert(new_obj == r0, "needs to be r0");
5661   assert_different_registers(klass, new_obj, t1, t2);
5662 
5663   // get instance_size in InstanceKlass (scaled to a count of bytes)
5664   ldrw(layout_size, Address(klass, Klass::layout_helper_offset()));
5665   // test to see if it is malformed in some way
5666   tst(layout_size, Klass::_lh_instance_slow_path_bit);
5667   br(Assembler::NE, slow_case_no_pop);
5668 
5669   // Allocate the instance:
5670   //  If TLAB is enabled:
5671   //    Try to allocate in the TLAB.
5672   //    If fails, go to the slow path.
5673   //    Initialize the allocation.
5674   //    Exit.
5675   //
5676   //  Go to slow path.
5677 
5678   if (UseTLAB) {
5679     push(klass);
5680     tlab_allocate(new_obj, layout_size, 0, klass, t2, slow_case);
5681     if (ZeroTLAB || (!clear_fields)) {
5682       // the fields have been already cleared
5683       b(initialize_header);
5684     } else {
5685       // initialize both the header and fields
5686       b(initialize_object);
5687     }
5688 
5689     if (clear_fields) {
5690       // The object is initialized before the header.  If the object size is
5691       // zero, go directly to the header initialization.
5692       bind(initialize_object);
5693       int header_size = oopDesc::header_size() * HeapWordSize;
5694       assert(is_aligned(header_size, BytesPerLong), "oop header size must be 8-byte-aligned");
5695       subs(layout_size, layout_size, header_size);
5696       br(Assembler::EQ, initialize_header);
5697 
5698       // Initialize topmost object field, divide size by 8, check if odd and
5699       // test if zero.
5700 
5701   #ifdef ASSERT
5702       // make sure instance_size was multiple of 8
5703       Label L;
5704       tst(layout_size, 7);
5705       br(Assembler::EQ, L);
5706       stop("object size is not multiple of 8 - adjust this code");
5707       bind(L);
5708       // must be > 0, no extra check needed here
5709   #endif
5710 
5711       lsr(layout_size, layout_size, LogBytesPerLong);
5712 
5713       // initialize remaining object fields: instance_size was a multiple of 8
5714       {
5715         Label loop;
5716         Register base = t2;
5717 
5718         bind(loop);
5719         add(rscratch1, new_obj, layout_size, Assembler::LSL, LogBytesPerLong);
5720         str(zr, Address(rscratch1, header_size - 1*oopSize));
5721         subs(layout_size, layout_size, 1);
5722         br(Assembler::NE, loop);
5723       }
5724     } // clear_fields
5725 
5726     // initialize object header only.
5727     bind(initialize_header);
5728     pop(klass);
5729     Register mark_word = t2;
5730     if (UseCompactObjectHeaders || EnableValhalla) {
5731       ldr(mark_word, Address(klass, Klass::prototype_header_offset()));
5732       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5733     } else {
5734       mov(mark_word, (intptr_t)markWord::prototype().value());
5735       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5736     }
5737     if (!UseCompactObjectHeaders) {
5738       store_klass_gap(new_obj, zr);  // zero klass gap for compressed oops
5739       mov(t2, klass);                // preserve klass
5740       store_klass(new_obj, t2);      // src klass reg is potentially compressed
5741     }
5742     b(done);
5743   }
5744 
5745   if (UseTLAB) {
5746     bind(slow_case);
5747     pop(klass);
5748   }
5749   bind(slow_case_no_pop);
5750   b(alloc_failed);
5751 
5752   bind(done);
5753 }
5754 
5755 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5756 void MacroAssembler::tlab_allocate(Register obj,
5757                                    Register var_size_in_bytes,
5758                                    int con_size_in_bytes,
5759                                    Register t1,
5760                                    Register t2,
5761                                    Label& slow_case) {
5762   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5763   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5764 }
5765 
5766 void MacroAssembler::verify_tlab() {
5767 #ifdef ASSERT
5768   if (UseTLAB && VerifyOops) {
5769     Label next, ok;
5770 
5771     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5772 
5773     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5774     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5775     cmp(rscratch2, rscratch1);
5776     br(Assembler::HS, next);
5777     STOP("assert(top >= start)");
5778     should_not_reach_here();
5779 
5780     bind(next);
5781     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5782     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5783     cmp(rscratch2, rscratch1);
5784     br(Assembler::HS, ok);
5785     STOP("assert(top <= end)");
5786     should_not_reach_here();
5787 
5788     bind(ok);
5789     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5790   }
5791 #endif
5792 }
5793 
5794 void MacroAssembler::get_inline_type_field_klass(Register holder_klass, Register index, Register inline_klass) {
5795   inline_layout_info(holder_klass, index, inline_klass);
5796   ldr(inline_klass, Address(inline_klass, InlineLayoutInfo::klass_offset()));
5797 }
5798 
5799 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5800   assert_different_registers(holder_klass, index, layout_info);
5801   InlineLayoutInfo array[2];
5802   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5803   if (is_power_of_2(size)) {
5804     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5805   } else {
5806     mov(layout_info, size);
5807     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5808   }
5809   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5810   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5811   lea(layout_info, Address(layout_info, index));
5812 }
5813 
5814 // Writes to stack successive pages until offset reached to check for
5815 // stack overflow + shadow pages.  This clobbers tmp.
5816 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5817   assert_different_registers(tmp, size, rscratch1);
5818   mov(tmp, sp);
5819   // Bang stack for total size given plus shadow page size.
5820   // Bang one page at a time because large size can bang beyond yellow and
5821   // red zones.
5822   Label loop;
5823   mov(rscratch1, (int)os::vm_page_size());
5824   bind(loop);
5825   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5826   subsw(size, size, rscratch1);
5827   str(size, Address(tmp));
5828   br(Assembler::GT, loop);
5829 
5830   // Bang down shadow pages too.
5831   // At this point, (tmp-0) is the last address touched, so don't
5832   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5833   // was post-decremented.)  Skip this address by starting at i=1, and

5919 }
5920 
5921 void MacroAssembler::remove_frame(int framesize) {
5922   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5923   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5924   if (framesize < ((1 << 9) + 2 * wordSize)) {
5925     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5926     add(sp, sp, framesize);
5927   } else {
5928     if (framesize < ((1 << 12) + 2 * wordSize))
5929       add(sp, sp, framesize - 2 * wordSize);
5930     else {
5931       mov(rscratch1, framesize - 2 * wordSize);
5932       add(sp, sp, rscratch1);
5933     }
5934     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5935   }
5936   authenticate_return_address();
5937 }
5938 
5939 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
5940   if (needs_stack_repair) {
5941     // Remove the extension of the caller's frame used for inline type unpacking
5942     //
5943     // Right now the stack looks like this:
5944     //
5945     // | Arguments from caller     |
5946     // |---------------------------|  <-- caller's SP
5947     // | Saved LR #1               |
5948     // | Saved FP #1               |
5949     // |---------------------------|
5950     // | Extension space for       |
5951     // |   inline arg (un)packing  |
5952     // |---------------------------|  <-- start of this method's frame
5953     // | Saved LR #2               |
5954     // | Saved FP #2               |
5955     // |---------------------------|  <-- FP
5956     // | sp_inc                    |
5957     // | method locals             |
5958     // |---------------------------|  <-- SP
5959     //
5960     // There are two copies of FP and LR on the stack. They will be identical at
5961     // first, but that can change.
5962     // If the caller has been deoptimized, LR #1 will be patched to point at the
5963     // deopt blob, and LR #2 will still point into the old method.
5964     // If the saved FP (x29) was not used as the frame pointer, but to store an
5965     // oop, the GC will be aware only of FP #1 as the spilled location of x29 and
5966     // will fix only this one. Overall, FP/LR #2 are not reliable and are simply
5967     // needed to add space between the extension space and the locals, as there
5968     // would be between the real arguments and the locals if we don't need to
5969     // do unpacking.
5970     //
5971     // When restoring, one must then load FP #1 into x29, and LR #1 into x30,
5972     // while keeping in mind that from the scalarized entry point, there will be
5973     // only one copy of each.
5974     //
5975     // The sp_inc stack slot holds the total size of the frame including the
5976     // extension space minus two words for the saved FP and LR. That is how to
5977     // find FP/LR #1. This size is expressed in bytes. Be careful when using it
5978     // from C++ in pointer arithmetic; you might need to divide it by wordSize.
5979     //
5980     // TODO 8371993 store fake values instead of LR/FP#2
5981 
5982     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
5983 
5984     ldr(rscratch1, Address(sp, sp_inc_offset));
5985     add(sp, sp, rscratch1);
5986     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5987   } else {
5988     remove_frame(initial_framesize);
5989   }
5990 }
5991 
5992 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
5993   int real_frame_size = frame_size + sp_inc;
5994   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
5995   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
5996   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
5997 
5998   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
5999 
6000   // Subtract two words for the saved FP and LR as these will be popped
6001   // separately. See remove_frame above.
6002   mov(rscratch1, real_frame_size - 2*wordSize);
6003   str(rscratch1, Address(sp, sp_inc_offset));
6004 }
6005 
6006 // This method counts leading positive bytes (highest bit not set) in provided byte array
6007 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6008     // Simple and most common case of aligned small array which is not at the
6009     // end of memory page is placed here. All other cases are in stub.
6010     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6011     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6012     assert_different_registers(ary1, len, result);
6013 
6014     mov(result, len);
6015     cmpw(len, 0);
6016     br(LE, DONE);
6017     cmpw(len, 4 * wordSize);
6018     br(GE, STUB_LONG); // size > 32 then go to stub
6019 
6020     int shift = 64 - exact_log2(os::vm_page_size());
6021     lsl(rscratch1, ary1, shift);
6022     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6023     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6024     br(CS, STUB); // at the end of page then go to stub

6902 // On other systems, the helper is a usual C function.
6903 //
6904 void MacroAssembler::get_thread(Register dst) {
6905   RegSet saved_regs =
6906     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6907     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6908 
6909   protect_return_address();
6910   push(saved_regs, sp);
6911 
6912   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6913   blr(lr);
6914   if (dst != c_rarg0) {
6915     mov(dst, c_rarg0);
6916   }
6917 
6918   pop(saved_regs, sp);
6919   authenticate_return_address();
6920 }
6921 
6922 #ifdef COMPILER2
6923 // C2 compiled method's prolog code
6924 // Moved here from aarch64.ad to support Valhalla code belows
6925 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
6926   if (C->clinit_barrier_on_entry()) {
6927     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
6928 
6929     Label L_skip_barrier;
6930 
6931     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
6932     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
6933     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
6934     bind(L_skip_barrier);
6935   }
6936 
6937   if (C->max_vector_size() > 0) {
6938     reinitialize_ptrue();
6939   }
6940 
6941   int bangsize = C->output()->bang_size_in_bytes();
6942   if (C->output()->need_stack_bang(bangsize))
6943     generate_stack_overflow_check(bangsize);
6944 
6945   // n.b. frame size includes space for return pc and rfp
6946   const long framesize = C->output()->frame_size_in_bytes();
6947   build_frame(framesize);
6948 
6949   if (C->needs_stack_repair()) {
6950     save_stack_increment(sp_inc, framesize);
6951   }
6952 
6953   if (VerifyStackAtCalls) {
6954     Unimplemented();
6955   }
6956 }
6957 #endif // COMPILER2
6958 
6959 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
6960   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
6961   // An inline type might be returned. If fields are in registers we
6962   // need to allocate an inline type instance and initialize it with
6963   // the value of the fields.
6964   Label skip;
6965   // We only need a new buffered inline type if a new one is not returned
6966   tbz(r0, 0, skip);
6967   int call_offset = -1;
6968 
6969   // Be careful not to clobber r1-7 which hold returned fields
6970   // Also do not use callee-saved registers as these may be live in the interpreter
6971   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
6972 
6973   // The following code is similar to allocate_instance but has some slight differences,
6974   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
6975   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
6976   Label slow_case;
6977   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
6978   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
6979 
6980   if (vk != nullptr) {
6981     // Called from C1, where the return type is statically known.
6982     movptr(klass, (intptr_t)vk->get_InlineKlass());
6983     jint lh = vk->layout_helper();
6984     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
6985     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
6986       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
6987     } else {
6988       b(slow_case);
6989     }
6990   } else {
6991     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
6992     andr(klass, r0, -2);
6993     if (UseTLAB) {
6994       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
6995       tst(tmp2, Klass::_lh_instance_slow_path_bit);
6996       br(Assembler::NE, slow_case);
6997       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
6998     } else {
6999       b(slow_case);
7000     }
7001   }
7002   if (UseTLAB) {
7003     // 2. Initialize buffered inline instance header
7004     Register buffer_obj = r0;
7005     if (UseCompactObjectHeaders) {
7006       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7007       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7008     } else {
7009       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7010       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7011       store_klass_gap(buffer_obj, zr);
7012       if (vk == nullptr) {
7013         // store_klass corrupts klass, so save it for later use (interpreter case only).
7014         mov(tmp1, klass);
7015       }
7016       store_klass(buffer_obj, klass);
7017       klass = tmp1;
7018     }
7019     // 3. Initialize its fields with an inline class specific handler
7020     if (vk != nullptr) {
7021       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7022     } else {
7023       ldr(tmp1, Address(klass, InstanceKlass::adr_inlineklass_fixed_block_offset()));
7024       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7025       blr(tmp1);
7026     }
7027 
7028     membar(Assembler::StoreStore);
7029     b(skip);
7030   } else {
7031     // Must have already branched to slow_case above.
7032     DEBUG_ONLY(should_not_reach_here());
7033   }
7034   bind(slow_case);
7035   // We failed to allocate a new inline type, fall back to a runtime
7036   // call. Some oop field may be live in some registers but we can't
7037   // tell. That runtime call will take care of preserving them
7038   // across a GC if there's one.
7039   mov(r0, r0_preserved);
7040 
7041   if (from_interpreter) {
7042     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
7043   } else {
7044     far_call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
7045     call_offset = offset();
7046   }
7047   membar(Assembler::StoreStore);
7048 
7049   bind(skip);
7050   return call_offset;
7051 }
7052 
7053 // Move a value between registers/stack slots and update the reg_state
7054 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7055   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7056   if (reg_state[to->value()] == reg_written) {
7057     return true; // Already written
7058   }
7059 
7060   if (from != to && bt != T_VOID) {
7061     if (reg_state[to->value()] == reg_readonly) {
7062       return false; // Not yet writable
7063     }
7064     if (from->is_reg()) {
7065       if (to->is_reg()) {
7066         if (from->is_Register() && to->is_Register()) {
7067           mov(to->as_Register(), from->as_Register());
7068         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7069           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7070         } else {
7071           ShouldNotReachHere();
7072         }
7073       } else {
7074         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7075         Address to_addr = Address(sp, st_off);
7076         if (from->is_FloatRegister()) {
7077           if (bt == T_DOUBLE) {
7078              strd(from->as_FloatRegister(), to_addr);
7079           } else {
7080              assert(bt == T_FLOAT, "must be float");
7081              strs(from->as_FloatRegister(), to_addr);
7082           }
7083         } else {
7084           str(from->as_Register(), to_addr);
7085         }
7086       }
7087     } else {
7088       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7089       if (to->is_reg()) {
7090         if (to->is_FloatRegister()) {
7091           if (bt == T_DOUBLE) {
7092             ldrd(to->as_FloatRegister(), from_addr);
7093           } else {
7094             assert(bt == T_FLOAT, "must be float");
7095             ldrs(to->as_FloatRegister(), from_addr);
7096           }
7097         } else {
7098           ldr(to->as_Register(), from_addr);
7099         }
7100       } else {
7101         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7102         ldr(rscratch1, from_addr);
7103         str(rscratch1, Address(sp, st_off));
7104       }
7105     }
7106   }
7107 
7108   // Update register states
7109   reg_state[from->value()] = reg_writable;
7110   reg_state[to->value()] = reg_written;
7111   return true;
7112 }
7113 
7114 // Calculate the extra stack space required for packing or unpacking inline
7115 // args and adjust the stack pointer
7116 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7117   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7118   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7119   assert(sp_inc > 0, "sanity");
7120 
7121   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7122   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7123 
7124   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7125   if (sp_inc < (1 << 9)) {
7126     sub(sp, sp, sp_inc);   // Fits in an immediate
7127   } else {
7128     mov(rscratch1, sp_inc);
7129     sub(sp, sp, rscratch1);
7130   }
7131 
7132   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7133 }
7134 
7135 // Read all fields from an inline type oop and store the values in registers/stack slots
7136 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7137                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7138                                           RegState reg_state[]) {
7139   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7140   assert(from->is_valid(), "source must be valid");
7141   bool progress = false;
7142 #ifdef ASSERT
7143   const int start_offset = offset();
7144 #endif
7145 
7146   Label L_null, L_notNull;
7147   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7148   // TODO 8366717 We need to make sure that r14 (and potentially other long-life regs) are kept live in slowpath runtime calls in GC barriers
7149   Register tmp1 = r10;
7150   Register tmp2 = r11;
7151   Register fromReg = noreg;
7152   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, -1);
7153   bool done = true;
7154   bool mark_done = true;
7155   VMReg toReg;
7156   BasicType bt;
7157   // Check if argument requires a null check
7158   bool null_check = false;
7159   VMReg nullCheckReg;
7160   while (stream.next(nullCheckReg, bt)) {
7161     if (sig->at(stream.sig_index())._offset == -1) {
7162       null_check = true;
7163       break;
7164     }
7165   }
7166   stream.reset(sig_index, to_index);
7167   while (stream.next(toReg, bt)) {
7168     assert(toReg->is_valid(), "destination must be valid");
7169     int idx = (int)toReg->value();
7170     if (reg_state[idx] == reg_readonly) {
7171       if (idx != from->value()) {
7172         mark_done = false;
7173       }
7174       done = false;
7175       continue;
7176     } else if (reg_state[idx] == reg_written) {
7177       continue;
7178     }
7179     assert(reg_state[idx] == reg_writable, "must be writable");
7180     reg_state[idx] = reg_written;
7181     progress = true;
7182 
7183     if (fromReg == noreg) {
7184       if (from->is_reg()) {
7185         fromReg = from->as_Register();
7186       } else {
7187         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7188         ldr(tmp1, Address(sp, st_off));
7189         fromReg = tmp1;
7190       }
7191       if (null_check) {
7192         // Nullable inline type argument, emit null check
7193         cbz(fromReg, L_null);
7194       }
7195     }
7196     int off = sig->at(stream.sig_index())._offset;
7197     if (off == -1) {
7198       assert(null_check, "Missing null check at");
7199       if (toReg->is_stack()) {
7200         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7201         mov(tmp2, 1);
7202         str(tmp2, Address(sp, st_off));
7203       } else {
7204         mov(toReg->as_Register(), 1);
7205       }
7206       continue;
7207     }
7208     assert(off > 0, "offset in object should be positive");
7209     Address fromAddr = Address(fromReg, off);
7210     if (!toReg->is_FloatRegister()) {
7211       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7212       if (is_reference_type(bt)) {
7213         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7214       } else {
7215         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7216         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7217       }
7218       if (toReg->is_stack()) {
7219         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7220         str(dst, Address(sp, st_off));
7221       }
7222     } else if (bt == T_DOUBLE) {
7223       ldrd(toReg->as_FloatRegister(), fromAddr);
7224     } else {
7225       assert(bt == T_FLOAT, "must be float");
7226       ldrs(toReg->as_FloatRegister(), fromAddr);
7227     }
7228   }
7229   if (progress && null_check) {
7230     if (done) {
7231       b(L_notNull);
7232       bind(L_null);
7233       // Set null marker to zero to signal that the argument is null.
7234       // Also set all oop fields to zero to make the GC happy.
7235       stream.reset(sig_index, to_index);
7236       while (stream.next(toReg, bt)) {
7237         if (sig->at(stream.sig_index())._offset == -1 ||
7238             bt == T_OBJECT || bt == T_ARRAY) {
7239           if (toReg->is_stack()) {
7240             int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7241             str(zr, Address(sp, st_off));
7242           } else {
7243             mov(toReg->as_Register(), zr);
7244           }
7245         }
7246       }
7247       bind(L_notNull);
7248     } else {
7249       bind(L_null);
7250     }
7251   }
7252 
7253   // TODO 8366717 This is probably okay but looks fishy because stream is reset in the "Set null marker to zero" case just above. Same on x64.
7254   sig_index = stream.sig_index();
7255   to_index = stream.regs_index();
7256 
7257   if (mark_done && reg_state[from->value()] != reg_written) {
7258     // This is okay because no one else will write to that slot
7259     reg_state[from->value()] = reg_writable;
7260   }
7261   from_index--;
7262   assert(progress || (start_offset == offset()), "should not emit code");
7263   return done;
7264 }
7265 
7266 // Pack fields back into an inline type oop
7267 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7268                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7269                                         RegState reg_state[], Register val_array) {
7270   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7271   assert(to->is_valid(), "destination must be valid");
7272 
7273   if (reg_state[to->value()] == reg_written) {
7274     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7275     return true; // Already written
7276   }
7277 
7278   // The GC barrier expanded by store_heap_oop below may call into the
7279   // runtime so use callee-saved registers for any values that need to be
7280   // preserved. The GC barrier assembler should take care of saving the
7281   // Java argument registers.
7282   // TODO 8284443 Isn't it an issue if below code uses r14 as tmp when it contains a spilled value?
7283   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7284   Register val_obj_tmp = r21;
7285   Register from_reg_tmp = r22;
7286   Register tmp1 = r14;
7287   Register tmp2 = r13;
7288   Register tmp3 = r12;
7289   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7290 
7291   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7292 
7293   if (reg_state[to->value()] == reg_readonly) {
7294     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7295       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7296       return false; // Not yet writable
7297     }
7298     val_obj = val_obj_tmp;
7299   }
7300 
7301   int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7302   load_heap_oop(val_obj, Address(val_array, index), tmp1, tmp2);
7303 
7304   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7305   VMReg fromReg;
7306   BasicType bt;
7307   Label L_null;
7308   while (stream.next(fromReg, bt)) {
7309     assert(fromReg->is_valid(), "source must be valid");
7310     reg_state[fromReg->value()] = reg_writable;
7311 
7312     int off = sig->at(stream.sig_index())._offset;
7313     if (off == -1) {
7314       // Nullable inline type argument, emit null check
7315       Label L_notNull;
7316       if (fromReg->is_stack()) {
7317         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7318         ldrb(tmp2, Address(sp, ld_off));
7319         cbnz(tmp2, L_notNull);
7320       } else {
7321         cbnz(fromReg->as_Register(), L_notNull);
7322       }
7323       mov(val_obj, 0);
7324       b(L_null);
7325       bind(L_notNull);
7326       continue;
7327     }
7328 
7329     assert(off > 0, "offset in object should be positive");
7330     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7331 
7332     // Pack the scalarized field into the value object.
7333     Address dst(val_obj, off);
7334     if (!fromReg->is_FloatRegister()) {
7335       Register src;
7336       if (fromReg->is_stack()) {
7337         src = from_reg_tmp;
7338         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7339         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7340       } else {
7341         src = fromReg->as_Register();
7342       }
7343       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7344       if (is_reference_type(bt)) {
7345         store_heap_oop(dst, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7346       } else {
7347         store_sized_value(dst, src, size_in_bytes);
7348       }
7349     } else if (bt == T_DOUBLE) {
7350       strd(fromReg->as_FloatRegister(), dst);
7351     } else {
7352       assert(bt == T_FLOAT, "must be float");
7353       strs(fromReg->as_FloatRegister(), dst);
7354     }
7355   }
7356   bind(L_null);
7357   sig_index = stream.sig_index();
7358   from_index = stream.regs_index();
7359 
7360   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7361   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7362   assert(success, "to register must be writeable");
7363   return true;
7364 }
7365 
7366 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7367   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7368 }
7369 
7370 void MacroAssembler::cache_wb(Address line) {
7371   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7372   assert(line.index() == noreg, "index should be noreg");
7373   assert(line.offset() == 0, "offset should be 0");
7374   // would like to assert this
7375   // assert(line._ext.shift == 0, "shift should be zero");
7376   if (VM_Version::supports_dcpop()) {
7377     // writeback using clear virtual address to point of persistence
7378     dc(Assembler::CVAP, line.base());
7379   } else {
7380     // no need to generate anything as Unsafe.writebackMemory should
7381     // never invoke this stub
7382   }
7383 }
7384 
7385 void MacroAssembler::cache_wbsync(bool is_pre) {
7386   // we only need a barrier post sync
7387   if (!is_pre) {
7388     membar(Assembler::AnyAny);
7389   }

7760   }
7761 
7762   // Check if the lock-stack is full.
7763   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7764   cmpw(top, (unsigned)LockStack::end_offset());
7765   br(Assembler::GE, slow);
7766 
7767   // Check for recursion.
7768   subw(t, top, oopSize);
7769   ldr(t, Address(rthread, t));
7770   cmp(obj, t);
7771   br(Assembler::EQ, push);
7772 
7773   // Check header for monitor (0b10).
7774   tst(mark, markWord::monitor_value);
7775   br(Assembler::NE, slow);
7776 
7777   // Try to lock. Transition lock bits 0b01 => 0b00
7778   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7779   orr(mark, mark, markWord::unlocked_value);
7780   // Mask inline_type bit such that we go to the slow path if object is an inline type
7781   andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
7782 
7783   eor(t, mark, markWord::unlocked_value);
7784   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7785           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7786   br(Assembler::NE, slow);
7787 
7788   bind(push);
7789   // After successful lock, push object on lock-stack.
7790   str(obj, Address(rthread, top));
7791   addw(top, top, oopSize);
7792   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7793 }
7794 
7795 // Implements lightweight-unlocking.
7796 //
7797 // - obj: the object to be unlocked
7798 // - t1, t2, t3: temporary registers
7799 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7800 void MacroAssembler::lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7801   // cmpxchg clobbers rscratch1.
7802   assert_different_registers(obj, t1, t2, t3, rscratch1);
< prev index next >