< prev index next >

src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp

Print this page

  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/assembler.hpp"
  28 #include "asm/assembler.inline.hpp"
  29 #include "ci/ciEnv.hpp"

  30 #include "code/compiledIC.hpp"
  31 #include "compiler/compileTask.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "compiler/oopMap.hpp"
  34 #include "gc/shared/barrierSet.hpp"
  35 #include "gc/shared/barrierSetAssembler.hpp"
  36 #include "gc/shared/cardTableBarrierSet.hpp"
  37 #include "gc/shared/cardTable.hpp"
  38 #include "gc/shared/collectedHeap.hpp"
  39 #include "gc/shared/tlab_globals.hpp"
  40 #include "interpreter/bytecodeHistogram.hpp"
  41 #include "interpreter/interpreter.hpp"
  42 #include "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 

1154 }
1155 
1156 void MacroAssembler::post_call_nop() {
1157   if (!Continuations::enabled()) {
1158     return;
1159   }
1160   InstructionMark im(this);
1161   relocate(post_call_nop_Relocation::spec());
1162   InlineSkippedInstructionsCounter skipCounter(this);
1163   nop();
1164   movk(zr, 0);
1165   movk(zr, 0);
1166 }
1167 
1168 // these are no-ops overridden by InterpreterMacroAssembler
1169 
1170 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1171 
1172 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1173 




































1174 // Look up the method for a megamorphic invokeinterface call.
1175 // The target method is determined by <intf_klass, itable_index>.
1176 // The receiver klass is in recv_klass.
1177 // On success, the result will be in method_result, and execution falls through.
1178 // On failure, execution transfers to the given label.
1179 void MacroAssembler::lookup_interface_method(Register recv_klass,
1180                                              Register intf_klass,
1181                                              RegisterOrConstant itable_index,
1182                                              Register method_result,
1183                                              Register scan_temp,
1184                                              Label& L_no_such_interface,
1185                          bool return_method) {
1186   assert_different_registers(recv_klass, intf_klass, scan_temp);
1187   assert_different_registers(method_result, intf_klass, scan_temp);
1188   assert(recv_klass != method_result || !return_method,
1189      "recv_klass can be destroyed when method isn't needed");
1190   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1191          "caller must use same register for non-constant itable index as for method");
1192 
1193   // Compute start of first itableOffsetEntry (which is at the end of the vtable)

1842   ldarb(scratch, scratch);
1843   subs(zr, scratch, InstanceKlass::fully_initialized);
1844   br(Assembler::EQ, *L_fast_path);
1845 
1846   // Fast path check: current thread is initializer thread
1847   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
1848   cmp(rthread, scratch);
1849 
1850   if (L_slow_path == &L_fallthrough) {
1851     br(Assembler::EQ, *L_fast_path);
1852     bind(*L_slow_path);
1853   } else if (L_fast_path == &L_fallthrough) {
1854     br(Assembler::NE, *L_slow_path);
1855     bind(*L_fast_path);
1856   } else {
1857     Unimplemented();
1858   }
1859 }
1860 
1861 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
1862   if (!VerifyOops) return;




1863 
1864   // Pass register number to verify_oop_subroutine
1865   const char* b = nullptr;
1866   {
1867     ResourceMark rm;
1868     stringStream ss;
1869     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
1870     b = code_string(ss.as_string());
1871   }
1872   BLOCK_COMMENT("verify_oop {");
1873 
1874   strip_return_address(); // This might happen within a stack frame.
1875   protect_return_address();
1876   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
1877   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
1878 
1879   mov(r0, reg);
1880   movptr(rscratch1, (uintptr_t)(address)b);
1881 
1882   // call indirectly to solve generation ordering problem
1883   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
1884   ldr(rscratch2, Address(rscratch2));
1885   blr(rscratch2);
1886 
1887   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
1888   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
1889   authenticate_return_address();
1890 
1891   BLOCK_COMMENT("} verify_oop");
1892 }
1893 
1894 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
1895   if (!VerifyOops) return;




1896 
1897   const char* b = nullptr;
1898   {
1899     ResourceMark rm;
1900     stringStream ss;
1901     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
1902     b = code_string(ss.as_string());
1903   }
1904   BLOCK_COMMENT("verify_oop_addr {");
1905 
1906   strip_return_address(); // This might happen within a stack frame.
1907   protect_return_address();
1908   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
1909   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
1910 
1911   // addr may contain sp so we will have to adjust it based on the
1912   // pushes that we just did.
1913   if (addr.uses(sp)) {
1914     lea(r0, addr);
1915     ldr(r0, Address(r0, 4 * wordSize));

1973   call_VM_leaf_base(entry_point, 1);
1974 }
1975 
1976 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
1977   assert_different_registers(arg_1, c_rarg0);
1978   pass_arg0(this, arg_0);
1979   pass_arg1(this, arg_1);
1980   call_VM_leaf_base(entry_point, 2);
1981 }
1982 
1983 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
1984                                   Register arg_1, Register arg_2) {
1985   assert_different_registers(arg_1, c_rarg0);
1986   assert_different_registers(arg_2, c_rarg0, c_rarg1);
1987   pass_arg0(this, arg_0);
1988   pass_arg1(this, arg_1);
1989   pass_arg2(this, arg_2);
1990   call_VM_leaf_base(entry_point, 3);
1991 }
1992 




1993 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
1994   pass_arg0(this, arg_0);
1995   MacroAssembler::call_VM_leaf_base(entry_point, 1);
1996 }
1997 
1998 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
1999 
2000   assert_different_registers(arg_0, c_rarg1);
2001   pass_arg1(this, arg_1);
2002   pass_arg0(this, arg_0);
2003   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2004 }
2005 
2006 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2007   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2008   assert_different_registers(arg_1, c_rarg2);
2009   pass_arg2(this, arg_2);
2010   pass_arg1(this, arg_1);
2011   pass_arg0(this, arg_0);
2012   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2018   assert_different_registers(arg_2, c_rarg3);
2019   pass_arg3(this, arg_3);
2020   pass_arg2(this, arg_2);
2021   pass_arg1(this, arg_1);
2022   pass_arg0(this, arg_0);
2023   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2024 }
2025 
2026 void MacroAssembler::null_check(Register reg, int offset) {
2027   if (needs_explicit_null_check(offset)) {
2028     // provoke OS null exception if reg is null by
2029     // accessing M[reg] w/o changing any registers
2030     // NOTE: this is plenty to provoke a segv
2031     ldr(zr, Address(reg));
2032   } else {
2033     // nothing to do, (later) access of M[reg + offset]
2034     // will provoke OS null exception if reg is null
2035   }
2036 }
2037 









































































































2038 // MacroAssembler protected routines needed to implement
2039 // public methods
2040 
2041 void MacroAssembler::mov(Register r, Address dest) {
2042   code_section()->relocate(pc(), dest.rspec());
2043   uint64_t imm64 = (uint64_t)dest.target();
2044   movptr(r, imm64);
2045 }
2046 
2047 // Move a constant pointer into r.  In AArch64 mode the virtual
2048 // address space is 48 bits in size, so we only need three
2049 // instructions to create a patchable instruction sequence that can
2050 // reach anywhere.
2051 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2052 #ifndef PRODUCT
2053   {
2054     char buffer[64];
2055     snprintf(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2056     block_comment(buffer);
2057   }

4822   adrp(rscratch1, src2, offset);
4823   ldr(rscratch1, Address(rscratch1, offset));
4824   cmp(src1, rscratch1);
4825 }
4826 
4827 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
4828   cmp(obj1, obj2);
4829 }
4830 
4831 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
4832   load_method_holder(rresult, rmethod);
4833   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
4834 }
4835 
4836 void MacroAssembler::load_method_holder(Register holder, Register method) {
4837   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
4838   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
4839   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
4840 }
4841 








4842 void MacroAssembler::load_klass(Register dst, Register src) {
4843   if (UseCompressedClassPointers) {
4844     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
4845     decode_klass_not_null(dst);
4846   } else {
4847     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
4848   }
4849 }
4850 
4851 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
4852   if (RestoreMXCSROnJNICalls) {
4853     Label OK;
4854     get_fpcr(tmp1);
4855     mov(tmp2, tmp1);
4856     // Set FPCR to the state we need. We do want Round to Nearest. We
4857     // don't want non-IEEE rounding modes or floating-point traps.
4858     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
4859     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
4860     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
4861     eor(tmp2, tmp1, tmp2);

4897 
4898 void MacroAssembler::cmp_klass(Register oop, Register trial_klass, Register tmp) {
4899   if (UseCompressedClassPointers) {
4900     ldrw(tmp, Address(oop, oopDesc::klass_offset_in_bytes()));
4901     if (CompressedKlassPointers::base() == nullptr) {
4902       cmp(trial_klass, tmp, LSL, CompressedKlassPointers::shift());
4903       return;
4904     } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
4905                && CompressedKlassPointers::shift() == 0) {
4906       // Only the bottom 32 bits matter
4907       cmpw(trial_klass, tmp);
4908       return;
4909     }
4910     decode_klass_not_null(tmp);
4911   } else {
4912     ldr(tmp, Address(oop, oopDesc::klass_offset_in_bytes()));
4913   }
4914   cmp(trial_klass, tmp);
4915 }
4916 





4917 void MacroAssembler::store_klass(Register dst, Register src) {
4918   // FIXME: Should this be a store release?  concurrent gcs assumes
4919   // klass length is valid if klass field is not null.
4920   if (UseCompressedClassPointers) {
4921     encode_klass_not_null(src);
4922     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
4923   } else {
4924     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
4925   }
4926 }
4927 
4928 void MacroAssembler::store_klass_gap(Register dst, Register src) {
4929   if (UseCompressedClassPointers) {
4930     // Store to klass gap in destination
4931     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
4932   }
4933 }
4934 
4935 // Algorithm must match CompressedOops::encode.
4936 void MacroAssembler::encode_heap_oop(Register d, Register s) {

5223   if (as_raw) {
5224     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5225   } else {
5226     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5227   }
5228 }
5229 
5230 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5231                                      Address dst, Register val,
5232                                      Register tmp1, Register tmp2, Register tmp3) {
5233   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5234   decorators = AccessInternal::decorator_fixup(decorators, type);
5235   bool as_raw = (decorators & AS_RAW) != 0;
5236   if (as_raw) {
5237     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5238   } else {
5239     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5240   }
5241 }
5242 














































5243 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5244                                    Register tmp2, DecoratorSet decorators) {
5245   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5246 }
5247 
5248 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5249                                             Register tmp2, DecoratorSet decorators) {
5250   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5251 }
5252 
5253 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5254                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5255   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5256 }
5257 
5258 // Used for storing nulls.
5259 void MacroAssembler::store_heap_oop_null(Address dst) {
5260   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5261 }
5262 

5299     oop_index = oop_recorder()->allocate_metadata_index(obj);
5300   } else {
5301     oop_index = oop_recorder()->find_index(obj);
5302   }
5303   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5304   mov(dst, Address((address)obj, rspec));
5305 }
5306 
5307 Address MacroAssembler::constant_oop_address(jobject obj) {
5308 #ifdef ASSERT
5309   {
5310     ThreadInVMfromUnknown tiv;
5311     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5312     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5313   }
5314 #endif
5315   int oop_index = oop_recorder()->find_index(obj);
5316   return Address((address)obj, oop_Relocation::spec(oop_index));
5317 }
5318 
































































































5319 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5320 void MacroAssembler::tlab_allocate(Register obj,
5321                                    Register var_size_in_bytes,
5322                                    int con_size_in_bytes,
5323                                    Register t1,
5324                                    Register t2,
5325                                    Label& slow_case) {
5326   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5327   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5328 }
5329 
5330 void MacroAssembler::verify_tlab() {
5331 #ifdef ASSERT
5332   if (UseTLAB && VerifyOops) {
5333     Label next, ok;
5334 
5335     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5336 
5337     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5338     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5339     cmp(rscratch2, rscratch1);
5340     br(Assembler::HS, next);
5341     STOP("assert(top >= start)");
5342     should_not_reach_here();
5343 
5344     bind(next);
5345     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5346     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5347     cmp(rscratch2, rscratch1);
5348     br(Assembler::HS, ok);
5349     STOP("assert(top <= end)");
5350     should_not_reach_here();
5351 
5352     bind(ok);
5353     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5354   }
5355 #endif
5356 }
5357 




















5358 // Writes to stack successive pages until offset reached to check for
5359 // stack overflow + shadow pages.  This clobbers tmp.
5360 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5361   assert_different_registers(tmp, size, rscratch1);
5362   mov(tmp, sp);
5363   // Bang stack for total size given plus shadow page size.
5364   // Bang one page at a time because large size can bang beyond yellow and
5365   // red zones.
5366   Label loop;
5367   mov(rscratch1, (int)os::vm_page_size());
5368   bind(loop);
5369   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5370   subsw(size, size, rscratch1);
5371   str(size, Address(tmp));
5372   br(Assembler::GT, loop);
5373 
5374   // Bang down shadow pages too.
5375   // At this point, (tmp-0) is the last address touched, so don't
5376   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5377   // was post-decremented.)  Skip this address by starting at i=1, and

5463 }
5464 
5465 void MacroAssembler::remove_frame(int framesize) {
5466   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5467   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5468   if (framesize < ((1 << 9) + 2 * wordSize)) {
5469     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5470     add(sp, sp, framesize);
5471   } else {
5472     if (framesize < ((1 << 12) + 2 * wordSize))
5473       add(sp, sp, framesize - 2 * wordSize);
5474     else {
5475       mov(rscratch1, framesize - 2 * wordSize);
5476       add(sp, sp, rscratch1);
5477     }
5478     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5479   }
5480   authenticate_return_address();
5481 }
5482 



















































5483 
5484 // This method counts leading positive bytes (highest bit not set) in provided byte array
5485 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5486     // Simple and most common case of aligned small array which is not at the
5487     // end of memory page is placed here. All other cases are in stub.
5488     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5489     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5490     assert_different_registers(ary1, len, result);
5491 
5492     mov(result, len);
5493     cmpw(len, 0);
5494     br(LE, DONE);
5495     cmpw(len, 4 * wordSize);
5496     br(GE, STUB_LONG); // size > 32 then go to stub
5497 
5498     int shift = 64 - exact_log2(os::vm_page_size());
5499     lsl(rscratch1, ary1, shift);
5500     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5501     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5502     br(CS, STUB); // at the end of page then go to stub

6376 // On other systems, the helper is a usual C function.
6377 //
6378 void MacroAssembler::get_thread(Register dst) {
6379   RegSet saved_regs =
6380     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6381     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6382 
6383   protect_return_address();
6384   push(saved_regs, sp);
6385 
6386   mov(lr, CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper));
6387   blr(lr);
6388   if (dst != c_rarg0) {
6389     mov(dst, c_rarg0);
6390   }
6391 
6392   pop(saved_regs, sp);
6393   authenticate_return_address();
6394 }
6395 



























































































































































































































































































































































































































































6396 void MacroAssembler::cache_wb(Address line) {
6397   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6398   assert(line.index() == noreg, "index should be noreg");
6399   assert(line.offset() == 0, "offset should be 0");
6400   // would like to assert this
6401   // assert(line._ext.shift == 0, "shift should be zero");
6402   if (VM_Version::supports_dcpop()) {
6403     // writeback using clear virtual address to point of persistence
6404     dc(Assembler::CVAP, line.base());
6405   } else {
6406     // no need to generate anything as Unsafe.writebackMemory should
6407     // never invoke this stub
6408   }
6409 }
6410 
6411 void MacroAssembler::cache_wbsync(bool is_pre) {
6412   // we only need a barrier post sync
6413   if (!is_pre) {
6414     membar(Assembler::AnyAny);
6415   }

6774   }
6775 
6776   // Check if the lock-stack is full.
6777   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
6778   cmpw(top, (unsigned)LockStack::end_offset());
6779   br(Assembler::GE, slow);
6780 
6781   // Check for recursion.
6782   subw(t, top, oopSize);
6783   ldr(t, Address(rthread, t));
6784   cmp(obj, t);
6785   br(Assembler::EQ, push);
6786 
6787   // Check header for monitor (0b10).
6788   tst(mark, markWord::monitor_value);
6789   br(Assembler::NE, slow);
6790 
6791   // Try to lock. Transition lock bits 0b01 => 0b00
6792   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
6793   orr(mark, mark, markWord::unlocked_value);




6794   eor(t, mark, markWord::unlocked_value);
6795   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
6796           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
6797   br(Assembler::NE, slow);
6798 
6799   bind(push);
6800   // After successful lock, push object on lock-stack.
6801   str(obj, Address(rthread, top));
6802   addw(top, top, oopSize);
6803   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
6804 }
6805 
6806 // Implements lightweight-unlocking.
6807 //
6808 // - obj: the object to be unlocked
6809 // - t1, t2, t3: temporary registers
6810 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
6811 void MacroAssembler::lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
6812   assert(LockingMode == LM_LIGHTWEIGHT, "only used with new lightweight locking");
6813   // cmpxchg clobbers rscratch1.

  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 "precompiled.hpp"
  27 #include "asm/assembler.hpp"
  28 #include "asm/assembler.inline.hpp"
  29 #include "ci/ciEnv.hpp"
  30 #include "ci/ciInlineKlass.hpp"
  31 #include "code/compiledIC.hpp"
  32 #include "compiler/compileTask.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "compiler/oopMap.hpp"
  35 #include "gc/shared/barrierSet.hpp"
  36 #include "gc/shared/barrierSetAssembler.hpp"
  37 #include "gc/shared/cardTableBarrierSet.hpp"
  38 #include "gc/shared/cardTable.hpp"
  39 #include "gc/shared/collectedHeap.hpp"
  40 #include "gc/shared/tlab_globals.hpp"
  41 #include "interpreter/bytecodeHistogram.hpp"
  42 #include "interpreter/interpreter.hpp"
  43 #include "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/icache.hpp"
  54 #include "runtime/interfaceSupport.inline.hpp"
  55 #include "runtime/javaThread.hpp"
  56 #include "runtime/jniHandles.inline.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/signature_cc.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "utilities/globalDefinitions.hpp"
  61 #include "utilities/powerOfTwo.hpp"
  62 #include "vmreg_aarch64.inline.hpp"
  63 #ifdef COMPILER1
  64 #include "c1/c1_LIRAssembler.hpp"
  65 #endif
  66 #ifdef COMPILER2
  67 #include "oops/oop.hpp"
  68 #include "opto/compile.hpp"
  69 #include "opto/node.hpp"
  70 #include "opto/output.hpp"
  71 #endif
  72 
  73 #include <sys/types.h>
  74 
  75 #ifdef PRODUCT
  76 #define BLOCK_COMMENT(str) /* nothing */
  77 #else
  78 #define BLOCK_COMMENT(str) block_comment(str)
  79 #endif
  80 #define STOP(str) stop(str);
  81 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  82 

1158 }
1159 
1160 void MacroAssembler::post_call_nop() {
1161   if (!Continuations::enabled()) {
1162     return;
1163   }
1164   InstructionMark im(this);
1165   relocate(post_call_nop_Relocation::spec());
1166   InlineSkippedInstructionsCounter skipCounter(this);
1167   nop();
1168   movk(zr, 0);
1169   movk(zr, 0);
1170 }
1171 
1172 // these are no-ops overridden by InterpreterMacroAssembler
1173 
1174 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1175 
1176 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1177 
1178 void MacroAssembler::get_default_value_oop(Register inline_klass, Register temp_reg, Register obj) {
1179   assert_different_registers(inline_klass, temp_reg, obj, rscratch2);
1180 #ifdef ASSERT
1181   {
1182     Label done_check;
1183     test_klass_is_inline_type(inline_klass, temp_reg, done_check);
1184     stop("get_default_value_oop from non inline type klass");
1185     bind(done_check);
1186   }
1187 #endif
1188   Register offset = temp_reg;
1189   // Getting the offset of the pre-allocated default value
1190   ldr(offset, Address(inline_klass, in_bytes(InstanceKlass::adr_inlineklass_fixed_block_offset())));
1191   load_sized_value(offset, Address(offset, in_bytes(InlineKlass::default_value_offset_offset())), sizeof(int), true /*is_signed*/);
1192 
1193   // Getting the mirror
1194   ldr(obj, Address(inline_klass, in_bytes(Klass::java_mirror_offset())));
1195   resolve_oop_handle(obj, inline_klass, rscratch2);
1196 
1197   // Getting the pre-allocated default value from the mirror
1198   Address field(obj, offset);
1199   load_heap_oop(obj, field, inline_klass, rscratch2);
1200 }
1201 
1202 void MacroAssembler::get_empty_inline_type_oop(Register inline_klass, Register temp_reg, Register obj) {
1203 #ifdef ASSERT
1204   {
1205     Label done_check;
1206     test_klass_is_empty_inline_type(inline_klass, temp_reg, done_check);
1207     stop("get_empty_value from non-empty inline klass");
1208     bind(done_check);
1209   }
1210 #endif
1211   get_default_value_oop(inline_klass, temp_reg, obj);
1212 }
1213 
1214 // Look up the method for a megamorphic invokeinterface call.
1215 // The target method is determined by <intf_klass, itable_index>.
1216 // The receiver klass is in recv_klass.
1217 // On success, the result will be in method_result, and execution falls through.
1218 // On failure, execution transfers to the given label.
1219 void MacroAssembler::lookup_interface_method(Register recv_klass,
1220                                              Register intf_klass,
1221                                              RegisterOrConstant itable_index,
1222                                              Register method_result,
1223                                              Register scan_temp,
1224                                              Label& L_no_such_interface,
1225                          bool return_method) {
1226   assert_different_registers(recv_klass, intf_klass, scan_temp);
1227   assert_different_registers(method_result, intf_klass, scan_temp);
1228   assert(recv_klass != method_result || !return_method,
1229      "recv_klass can be destroyed when method isn't needed");
1230   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1231          "caller must use same register for non-constant itable index as for method");
1232 
1233   // Compute start of first itableOffsetEntry (which is at the end of the vtable)

1882   ldarb(scratch, scratch);
1883   subs(zr, scratch, InstanceKlass::fully_initialized);
1884   br(Assembler::EQ, *L_fast_path);
1885 
1886   // Fast path check: current thread is initializer thread
1887   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
1888   cmp(rthread, scratch);
1889 
1890   if (L_slow_path == &L_fallthrough) {
1891     br(Assembler::EQ, *L_fast_path);
1892     bind(*L_slow_path);
1893   } else if (L_fast_path == &L_fallthrough) {
1894     br(Assembler::NE, *L_slow_path);
1895     bind(*L_fast_path);
1896   } else {
1897     Unimplemented();
1898   }
1899 }
1900 
1901 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
1902   if (!VerifyOops || VerifyAdapterSharing) {
1903     // Below address of the code string confuses VerifyAdapterSharing
1904     // because it may differ between otherwise equivalent adapters.
1905     return;
1906   }
1907 
1908   // Pass register number to verify_oop_subroutine
1909   const char* b = nullptr;
1910   {
1911     ResourceMark rm;
1912     stringStream ss;
1913     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
1914     b = code_string(ss.as_string());
1915   }
1916   BLOCK_COMMENT("verify_oop {");
1917 
1918   strip_return_address(); // This might happen within a stack frame.
1919   protect_return_address();
1920   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
1921   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
1922 
1923   mov(r0, reg);
1924   movptr(rscratch1, (uintptr_t)(address)b);
1925 
1926   // call indirectly to solve generation ordering problem
1927   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
1928   ldr(rscratch2, Address(rscratch2));
1929   blr(rscratch2);
1930 
1931   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
1932   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
1933   authenticate_return_address();
1934 
1935   BLOCK_COMMENT("} verify_oop");
1936 }
1937 
1938 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
1939   if (!VerifyOops || VerifyAdapterSharing) {
1940     // Below address of the code string confuses VerifyAdapterSharing
1941     // because it may differ between otherwise equivalent adapters.
1942     return;
1943   }
1944 
1945   const char* b = nullptr;
1946   {
1947     ResourceMark rm;
1948     stringStream ss;
1949     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
1950     b = code_string(ss.as_string());
1951   }
1952   BLOCK_COMMENT("verify_oop_addr {");
1953 
1954   strip_return_address(); // This might happen within a stack frame.
1955   protect_return_address();
1956   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
1957   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
1958 
1959   // addr may contain sp so we will have to adjust it based on the
1960   // pushes that we just did.
1961   if (addr.uses(sp)) {
1962     lea(r0, addr);
1963     ldr(r0, Address(r0, 4 * wordSize));

2021   call_VM_leaf_base(entry_point, 1);
2022 }
2023 
2024 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2025   assert_different_registers(arg_1, c_rarg0);
2026   pass_arg0(this, arg_0);
2027   pass_arg1(this, arg_1);
2028   call_VM_leaf_base(entry_point, 2);
2029 }
2030 
2031 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2032                                   Register arg_1, Register arg_2) {
2033   assert_different_registers(arg_1, c_rarg0);
2034   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2035   pass_arg0(this, arg_0);
2036   pass_arg1(this, arg_1);
2037   pass_arg2(this, arg_2);
2038   call_VM_leaf_base(entry_point, 3);
2039 }
2040 
2041 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2042   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2043 }
2044 
2045 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2046   pass_arg0(this, arg_0);
2047   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2048 }
2049 
2050 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2051 
2052   assert_different_registers(arg_0, c_rarg1);
2053   pass_arg1(this, arg_1);
2054   pass_arg0(this, arg_0);
2055   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2056 }
2057 
2058 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2059   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2060   assert_different_registers(arg_1, c_rarg2);
2061   pass_arg2(this, arg_2);
2062   pass_arg1(this, arg_1);
2063   pass_arg0(this, arg_0);
2064   MacroAssembler::call_VM_leaf_base(entry_point, 3);

2070   assert_different_registers(arg_2, c_rarg3);
2071   pass_arg3(this, arg_3);
2072   pass_arg2(this, arg_2);
2073   pass_arg1(this, arg_1);
2074   pass_arg0(this, arg_0);
2075   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2076 }
2077 
2078 void MacroAssembler::null_check(Register reg, int offset) {
2079   if (needs_explicit_null_check(offset)) {
2080     // provoke OS null exception if reg is null by
2081     // accessing M[reg] w/o changing any registers
2082     // NOTE: this is plenty to provoke a segv
2083     ldr(zr, Address(reg));
2084   } else {
2085     // nothing to do, (later) access of M[reg + offset]
2086     // will provoke OS null exception if reg is null
2087   }
2088 }
2089 
2090 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2091   assert_different_registers(markword, rscratch2);
2092   andr(markword, markword, markWord::inline_type_mask_in_place);
2093   mov(rscratch2, markWord::inline_type_pattern);
2094   cmp(markword, rscratch2);
2095   br(Assembler::EQ, is_inline_type);
2096 }
2097 
2098 void MacroAssembler::test_klass_is_inline_type(Register klass, Register temp_reg, Label& is_inline_type) {
2099   ldrw(temp_reg, Address(klass, Klass::access_flags_offset()));
2100   andr(temp_reg, temp_reg, JVM_ACC_IDENTITY);
2101   cbz(temp_reg, is_inline_type);
2102 }
2103 
2104 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type) {
2105   assert_different_registers(tmp, rscratch1);
2106   cbz(object, not_inline_type);
2107   const int is_inline_type_mask = markWord::inline_type_pattern;
2108   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2109   mov(rscratch1, is_inline_type_mask);
2110   andr(tmp, tmp, rscratch1);
2111   cmp(tmp, rscratch1);
2112   br(Assembler::NE, not_inline_type);
2113 }
2114 
2115 void MacroAssembler::test_klass_is_empty_inline_type(Register klass, Register temp_reg, Label& is_empty_inline_type) {
2116 #ifdef ASSERT
2117   {
2118     Label done_check;
2119     test_klass_is_inline_type(klass, temp_reg, done_check);
2120     stop("test_klass_is_empty_inline_type with non inline type klass");
2121     bind(done_check);
2122   }
2123 #endif
2124   ldrw(temp_reg, Address(klass, InstanceKlass::misc_flags_offset()));
2125   andr(temp_reg, temp_reg, InstanceKlassFlags::is_empty_inline_type_value());
2126   cbnz(temp_reg, is_empty_inline_type);
2127 }
2128 
2129 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2130   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2131   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2132 }
2133 
2134 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2135   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2136   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2137 }
2138 
2139 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2140   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2141   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2142 }
2143 
2144 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
2145   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2146   tbnz(flags, ResolvedFieldEntry::has_null_marker_shift, has_null_marker);
2147 }
2148 
2149 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2150   Label test_mark_word;
2151   // load mark word
2152   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2153   // check displaced
2154   tst(temp_reg, markWord::unlocked_value);
2155   br(Assembler::NE, test_mark_word);
2156   // slow path use klass prototype
2157   load_prototype_header(temp_reg, oop);
2158 
2159   bind(test_mark_word);
2160   andr(temp_reg, temp_reg, test_bit);
2161   if (jmp_set) {
2162     cbnz(temp_reg, jmp_label);
2163   } else {
2164     cbz(temp_reg, jmp_label);
2165   }
2166 }
2167 
2168 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2169   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2170 }
2171 
2172 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2173                                                   Label&is_non_flat_array) {
2174   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2175 }
2176 
2177 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2178   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2179 }
2180 
2181 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2182   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2183 }
2184 
2185 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2186   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2187   br(Assembler::NE, is_flat_array);
2188 }
2189 
2190 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
2191   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2192   br(Assembler::EQ, is_non_flat_array);
2193 }
2194 
2195 // MacroAssembler protected routines needed to implement
2196 // public methods
2197 
2198 void MacroAssembler::mov(Register r, Address dest) {
2199   code_section()->relocate(pc(), dest.rspec());
2200   uint64_t imm64 = (uint64_t)dest.target();
2201   movptr(r, imm64);
2202 }
2203 
2204 // Move a constant pointer into r.  In AArch64 mode the virtual
2205 // address space is 48 bits in size, so we only need three
2206 // instructions to create a patchable instruction sequence that can
2207 // reach anywhere.
2208 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2209 #ifndef PRODUCT
2210   {
2211     char buffer[64];
2212     snprintf(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2213     block_comment(buffer);
2214   }

4979   adrp(rscratch1, src2, offset);
4980   ldr(rscratch1, Address(rscratch1, offset));
4981   cmp(src1, rscratch1);
4982 }
4983 
4984 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
4985   cmp(obj1, obj2);
4986 }
4987 
4988 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
4989   load_method_holder(rresult, rmethod);
4990   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
4991 }
4992 
4993 void MacroAssembler::load_method_holder(Register holder, Register method) {
4994   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
4995   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
4996   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
4997 }
4998 
4999 void MacroAssembler::load_metadata(Register dst, Register src) {
5000   if (UseCompressedClassPointers) {
5001     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5002   } else {
5003     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5004   }
5005 }
5006 
5007 void MacroAssembler::load_klass(Register dst, Register src) {
5008   if (UseCompressedClassPointers) {
5009     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5010     decode_klass_not_null(dst);
5011   } else {
5012     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5013   }
5014 }
5015 
5016 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5017   if (RestoreMXCSROnJNICalls) {
5018     Label OK;
5019     get_fpcr(tmp1);
5020     mov(tmp2, tmp1);
5021     // Set FPCR to the state we need. We do want Round to Nearest. We
5022     // don't want non-IEEE rounding modes or floating-point traps.
5023     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5024     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5025     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5026     eor(tmp2, tmp1, tmp2);

5062 
5063 void MacroAssembler::cmp_klass(Register oop, Register trial_klass, Register tmp) {
5064   if (UseCompressedClassPointers) {
5065     ldrw(tmp, Address(oop, oopDesc::klass_offset_in_bytes()));
5066     if (CompressedKlassPointers::base() == nullptr) {
5067       cmp(trial_klass, tmp, LSL, CompressedKlassPointers::shift());
5068       return;
5069     } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5070                && CompressedKlassPointers::shift() == 0) {
5071       // Only the bottom 32 bits matter
5072       cmpw(trial_klass, tmp);
5073       return;
5074     }
5075     decode_klass_not_null(tmp);
5076   } else {
5077     ldr(tmp, Address(oop, oopDesc::klass_offset_in_bytes()));
5078   }
5079   cmp(trial_klass, tmp);
5080 }
5081 
5082 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5083   load_klass(dst, src);
5084   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5085 }
5086 
5087 void MacroAssembler::store_klass(Register dst, Register src) {
5088   // FIXME: Should this be a store release?  concurrent gcs assumes
5089   // klass length is valid if klass field is not null.
5090   if (UseCompressedClassPointers) {
5091     encode_klass_not_null(src);
5092     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5093   } else {
5094     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5095   }
5096 }
5097 
5098 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5099   if (UseCompressedClassPointers) {
5100     // Store to klass gap in destination
5101     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5102   }
5103 }
5104 
5105 // Algorithm must match CompressedOops::encode.
5106 void MacroAssembler::encode_heap_oop(Register d, Register s) {

5393   if (as_raw) {
5394     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5395   } else {
5396     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5397   }
5398 }
5399 
5400 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5401                                      Address dst, Register val,
5402                                      Register tmp1, Register tmp2, Register tmp3) {
5403   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5404   decorators = AccessInternal::decorator_fixup(decorators, type);
5405   bool as_raw = (decorators & AS_RAW) != 0;
5406   if (as_raw) {
5407     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5408   } else {
5409     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5410   }
5411 }
5412 
5413 void MacroAssembler::access_value_copy(DecoratorSet decorators, Register src, Register dst,
5414                                        Register inline_klass) {
5415   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5416   bs->value_copy(this, decorators, src, dst, inline_klass);
5417 }
5418 
5419 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5420                                      Register inline_layout_info) {
5421   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5422   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5423 }
5424 
5425 void MacroAssembler::first_field_offset(Register inline_klass, Register offset) {
5426   ldr(offset, Address(inline_klass, InstanceKlass::adr_inlineklass_fixed_block_offset()));
5427   ldrw(offset, Address(offset, InlineKlass::first_field_offset_offset()));
5428 }
5429 
5430 void MacroAssembler::data_for_oop(Register oop, Register data, Register inline_klass) {
5431   // ((address) (void*) o) + vk->first_field_offset();
5432   Register offset = (data == oop) ? rscratch1 : data;
5433   first_field_offset(inline_klass, offset);
5434   if (data == oop) {
5435     add(data, data, offset);
5436   } else {
5437     lea(data, Address(oop, offset));
5438   }
5439 }
5440 
5441 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
5442                                                 Register index, Register data) {
5443   assert_different_registers(array, array_klass, index);
5444   assert_different_registers(rscratch1, array, index);
5445 
5446   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
5447   ldrw(rscratch1, Address(array_klass, Klass::layout_helper_offset()));
5448 
5449   // Klass::layout_helper_log2_element_size(lh)
5450   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
5451   lsr(rscratch1, rscratch1, Klass::_lh_log2_element_size_shift);
5452   andr(rscratch1, rscratch1, Klass::_lh_log2_element_size_mask);
5453   lslv(index, index, rscratch1);
5454 
5455   add(data, array, index);
5456   add(data, data, arrayOopDesc::base_offset_in_bytes(T_PRIMITIVE_OBJECT));
5457 }
5458 
5459 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5460                                    Register tmp2, DecoratorSet decorators) {
5461   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5462 }
5463 
5464 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5465                                             Register tmp2, DecoratorSet decorators) {
5466   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5467 }
5468 
5469 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5470                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5471   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5472 }
5473 
5474 // Used for storing nulls.
5475 void MacroAssembler::store_heap_oop_null(Address dst) {
5476   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5477 }
5478 

5515     oop_index = oop_recorder()->allocate_metadata_index(obj);
5516   } else {
5517     oop_index = oop_recorder()->find_index(obj);
5518   }
5519   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5520   mov(dst, Address((address)obj, rspec));
5521 }
5522 
5523 Address MacroAssembler::constant_oop_address(jobject obj) {
5524 #ifdef ASSERT
5525   {
5526     ThreadInVMfromUnknown tiv;
5527     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5528     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5529   }
5530 #endif
5531   int oop_index = oop_recorder()->find_index(obj);
5532   return Address((address)obj, oop_Relocation::spec(oop_index));
5533 }
5534 
5535 // Object / value buffer allocation...
5536 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
5537                                        Register t1, Register t2,
5538                                        bool clear_fields, Label& alloc_failed)
5539 {
5540   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
5541   Register layout_size = t1;
5542   assert(new_obj == r0, "needs to be r0");
5543   assert_different_registers(klass, new_obj, t1, t2);
5544 
5545   // get instance_size in InstanceKlass (scaled to a count of bytes)
5546   ldrw(layout_size, Address(klass, Klass::layout_helper_offset()));
5547   // test to see if it is malformed in some way
5548   tst(layout_size, Klass::_lh_instance_slow_path_bit);
5549   br(Assembler::NE, slow_case_no_pop);
5550 
5551   // Allocate the instance:
5552   //  If TLAB is enabled:
5553   //    Try to allocate in the TLAB.
5554   //    If fails, go to the slow path.
5555   //    Initialize the allocation.
5556   //    Exit.
5557   //
5558   //  Go to slow path.
5559 
5560   if (UseTLAB) {
5561     push(klass);
5562     tlab_allocate(new_obj, layout_size, 0, klass, t2, slow_case);
5563     if (ZeroTLAB || (!clear_fields)) {
5564       // the fields have been already cleared
5565       b(initialize_header);
5566     } else {
5567       // initialize both the header and fields
5568       b(initialize_object);
5569     }
5570 
5571     if (clear_fields) {
5572       // The object is initialized before the header.  If the object size is
5573       // zero, go directly to the header initialization.
5574       bind(initialize_object);
5575       subs(layout_size, layout_size, sizeof(oopDesc));
5576       br(Assembler::EQ, initialize_header);
5577 
5578       // Initialize topmost object field, divide size by 8, check if odd and
5579       // test if zero.
5580 
5581   #ifdef ASSERT
5582       // make sure instance_size was multiple of 8
5583       Label L;
5584       tst(layout_size, 7);
5585       br(Assembler::EQ, L);
5586       stop("object size is not multiple of 8 - adjust this code");
5587       bind(L);
5588       // must be > 0, no extra check needed here
5589   #endif
5590 
5591       lsr(layout_size, layout_size, LogBytesPerLong);
5592 
5593       // initialize remaining object fields: instance_size was a multiple of 8
5594       {
5595         Label loop;
5596         Register base = t2;
5597 
5598         bind(loop);
5599         add(rscratch1, new_obj, layout_size, Assembler::LSL, LogBytesPerLong);
5600         str(zr, Address(rscratch1, sizeof(oopDesc) - 1*oopSize));
5601         subs(layout_size, layout_size, 1);
5602         br(Assembler::NE, loop);
5603       }
5604     } // clear_fields
5605 
5606     // initialize object header only.
5607     bind(initialize_header);
5608     pop(klass);
5609     Register mark_word = t2;
5610     ldr(mark_word, Address(klass, Klass::prototype_header_offset()));
5611     str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes ()));
5612     store_klass_gap(new_obj, zr);  // zero klass gap for compressed oops
5613     mov(t2, klass);         // preserve klass
5614     store_klass(new_obj, t2);  // src klass reg is potentially compressed
5615 
5616     // TODO: Valhalla removed SharedRuntime::dtrace_object_alloc from here ?
5617 
5618     b(done);
5619   }
5620 
5621   if (UseTLAB) {
5622     bind(slow_case);
5623     pop(klass);
5624   }
5625   bind(slow_case_no_pop);
5626   b(alloc_failed);
5627 
5628   bind(done);
5629 }
5630 
5631 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5632 void MacroAssembler::tlab_allocate(Register obj,
5633                                    Register var_size_in_bytes,
5634                                    int con_size_in_bytes,
5635                                    Register t1,
5636                                    Register t2,
5637                                    Label& slow_case) {
5638   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5639   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5640 }
5641 
5642 void MacroAssembler::verify_tlab() {
5643 #ifdef ASSERT
5644   if (UseTLAB && VerifyOops) {
5645     Label next, ok;
5646 
5647     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5648 
5649     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5650     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5651     cmp(rscratch2, rscratch1);
5652     br(Assembler::HS, next);
5653     STOP("assert(top >= start)");
5654     should_not_reach_here();
5655 
5656     bind(next);
5657     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5658     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5659     cmp(rscratch2, rscratch1);
5660     br(Assembler::HS, ok);
5661     STOP("assert(top <= end)");
5662     should_not_reach_here();
5663 
5664     bind(ok);
5665     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5666   }
5667 #endif
5668 }
5669 
5670 void MacroAssembler::get_inline_type_field_klass(Register holder_klass, Register index, Register inline_klass) {
5671   inline_layout_info(holder_klass, index, inline_klass);
5672   ldr(inline_klass, Address(inline_klass, InlineLayoutInfo::klass_offset()));
5673 }
5674 
5675 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5676   assert_different_registers(holder_klass, index, layout_info);
5677   InlineLayoutInfo array[2];
5678   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5679   if (is_power_of_2(size)) {
5680     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5681   } else {
5682     mov(layout_info, size);
5683     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5684   }
5685   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5686   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5687   lea(layout_info, Address(layout_info, index));
5688 }
5689 
5690 // Writes to stack successive pages until offset reached to check for
5691 // stack overflow + shadow pages.  This clobbers tmp.
5692 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5693   assert_different_registers(tmp, size, rscratch1);
5694   mov(tmp, sp);
5695   // Bang stack for total size given plus shadow page size.
5696   // Bang one page at a time because large size can bang beyond yellow and
5697   // red zones.
5698   Label loop;
5699   mov(rscratch1, (int)os::vm_page_size());
5700   bind(loop);
5701   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5702   subsw(size, size, rscratch1);
5703   str(size, Address(tmp));
5704   br(Assembler::GT, loop);
5705 
5706   // Bang down shadow pages too.
5707   // At this point, (tmp-0) is the last address touched, so don't
5708   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5709   // was post-decremented.)  Skip this address by starting at i=1, and

5795 }
5796 
5797 void MacroAssembler::remove_frame(int framesize) {
5798   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5799   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5800   if (framesize < ((1 << 9) + 2 * wordSize)) {
5801     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5802     add(sp, sp, framesize);
5803   } else {
5804     if (framesize < ((1 << 12) + 2 * wordSize))
5805       add(sp, sp, framesize - 2 * wordSize);
5806     else {
5807       mov(rscratch1, framesize - 2 * wordSize);
5808       add(sp, sp, rscratch1);
5809     }
5810     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5811   }
5812   authenticate_return_address();
5813 }
5814 
5815 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
5816   if (needs_stack_repair) {
5817     // Remove the extension of the caller's frame used for inline type unpacking
5818     //
5819     // Right now the stack looks like this:
5820     //
5821     // | Arguments from caller     |
5822     // |---------------------------|  <-- caller's SP
5823     // | Saved LR #1               |
5824     // | Saved FP #1               |
5825     // |---------------------------|
5826     // | Extension space for       |
5827     // |   inline arg (un)packing  |
5828     // |---------------------------|  <-- start of this method's frame
5829     // | Saved LR #2               |
5830     // | Saved FP #2               |
5831     // |---------------------------|  <-- FP
5832     // | sp_inc                    |
5833     // | method locals             |
5834     // |---------------------------|  <-- SP
5835     //
5836     // There are two copies of FP and LR on the stack. They will be identical
5837     // unless the caller has been deoptimized, in which case LR #1 will be patched
5838     // to point at the deopt blob, and LR #2 will still point into the old method.
5839     //
5840     // The sp_inc stack slot holds the total size of the frame including the
5841     // extension space minus two words for the saved FP and LR.
5842 
5843     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
5844 
5845     ldr(rscratch1, Address(sp, sp_inc_offset));
5846     add(sp, sp, rscratch1);
5847     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5848   } else {
5849     remove_frame(initial_framesize);
5850   }
5851 }
5852 
5853 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
5854   int real_frame_size = frame_size + sp_inc;
5855   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
5856   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
5857   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
5858 
5859   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
5860 
5861   // Subtract two words for the saved FP and LR as these will be popped
5862   // separately. See remove_frame above.
5863   mov(rscratch1, real_frame_size - 2*wordSize);
5864   str(rscratch1, Address(sp, sp_inc_offset));
5865 }
5866 
5867 // This method counts leading positive bytes (highest bit not set) in provided byte array
5868 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5869     // Simple and most common case of aligned small array which is not at the
5870     // end of memory page is placed here. All other cases are in stub.
5871     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5872     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5873     assert_different_registers(ary1, len, result);
5874 
5875     mov(result, len);
5876     cmpw(len, 0);
5877     br(LE, DONE);
5878     cmpw(len, 4 * wordSize);
5879     br(GE, STUB_LONG); // size > 32 then go to stub
5880 
5881     int shift = 64 - exact_log2(os::vm_page_size());
5882     lsl(rscratch1, ary1, shift);
5883     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5884     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5885     br(CS, STUB); // at the end of page then go to stub

6759 // On other systems, the helper is a usual C function.
6760 //
6761 void MacroAssembler::get_thread(Register dst) {
6762   RegSet saved_regs =
6763     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6764     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6765 
6766   protect_return_address();
6767   push(saved_regs, sp);
6768 
6769   mov(lr, CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper));
6770   blr(lr);
6771   if (dst != c_rarg0) {
6772     mov(dst, c_rarg0);
6773   }
6774 
6775   pop(saved_regs, sp);
6776   authenticate_return_address();
6777 }
6778 
6779 #ifdef COMPILER2
6780 // C2 compiled method's prolog code
6781 // Moved here from aarch64.ad to support Valhalla code belows
6782 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
6783   if (C->clinit_barrier_on_entry()) {
6784     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
6785 
6786     Label L_skip_barrier;
6787 
6788     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
6789     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
6790     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
6791     bind(L_skip_barrier);
6792   }
6793 
6794   if (C->max_vector_size() > 0) {
6795     reinitialize_ptrue();
6796   }
6797 
6798   int bangsize = C->output()->bang_size_in_bytes();
6799   if (C->output()->need_stack_bang(bangsize))
6800     generate_stack_overflow_check(bangsize);
6801 
6802   // n.b. frame size includes space for return pc and rfp
6803   const long framesize = C->output()->frame_size_in_bytes();
6804   build_frame(framesize);
6805 
6806   if (C->needs_stack_repair()) {
6807     save_stack_increment(sp_inc, framesize);
6808   }
6809 
6810   if (VerifyStackAtCalls) {
6811     Unimplemented();
6812   }
6813 }
6814 #endif // COMPILER2
6815 
6816 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
6817   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
6818   // An inline type might be returned. If fields are in registers we
6819   // need to allocate an inline type instance and initialize it with
6820   // the value of the fields.
6821   Label skip;
6822   // We only need a new buffered inline type if a new one is not returned
6823   tbz(r0, 0, skip);
6824   int call_offset = -1;
6825 
6826   // Be careful not to clobber r1-7 which hold returned fields
6827   // Also do not use callee-saved registers as these may be live in the interpreter
6828   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
6829 
6830   // The following code is similar to allocate_instance but has some slight differences,
6831   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
6832   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
6833   Label slow_case;
6834   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
6835   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
6836 
6837   if (vk != nullptr) {
6838     // Called from C1, where the return type is statically known.
6839     movptr(klass, (intptr_t)vk->get_InlineKlass());
6840     jint lh = vk->layout_helper();
6841     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
6842     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
6843       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
6844     } else {
6845       b(slow_case);
6846     }
6847   } else {
6848     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
6849     andr(klass, r0, -2);
6850     if (UseTLAB) {
6851       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
6852       tst(tmp2, Klass::_lh_instance_slow_path_bit);
6853       br(Assembler::NE, slow_case);
6854       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
6855     } else {
6856       b(slow_case);
6857     }
6858   }
6859   if (UseTLAB) {
6860     // 2. Initialize buffered inline instance header
6861     Register buffer_obj = r0;
6862     mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
6863     str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
6864     store_klass_gap(buffer_obj, zr);
6865     if (vk == nullptr) {
6866       // store_klass corrupts klass, so save it for later use (interpreter case only).
6867       mov(tmp1, klass);
6868     }
6869     store_klass(buffer_obj, klass);
6870     // 3. Initialize its fields with an inline class specific handler
6871     if (vk != nullptr) {
6872       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
6873     } else {
6874       // tmp1 holds klass preserved above
6875       ldr(tmp1, Address(tmp1, InstanceKlass::adr_inlineklass_fixed_block_offset()));
6876       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
6877       blr(tmp1);
6878     }
6879 
6880     membar(Assembler::StoreStore);
6881     b(skip);
6882   } else {
6883     // Must have already branched to slow_case above.
6884     DEBUG_ONLY(should_not_reach_here());
6885   }
6886   bind(slow_case);
6887   // We failed to allocate a new inline type, fall back to a runtime
6888   // call. Some oop field may be live in some registers but we can't
6889   // tell. That runtime call will take care of preserving them
6890   // across a GC if there's one.
6891   mov(r0, r0_preserved);
6892 
6893   if (from_interpreter) {
6894     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
6895   } else {
6896     far_call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
6897     call_offset = offset();
6898   }
6899   membar(Assembler::StoreStore);
6900 
6901   bind(skip);
6902   return call_offset;
6903 }
6904 
6905 // Move a value between registers/stack slots and update the reg_state
6906 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
6907   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
6908   if (reg_state[to->value()] == reg_written) {
6909     return true; // Already written
6910   }
6911 
6912   if (from != to && bt != T_VOID) {
6913     if (reg_state[to->value()] == reg_readonly) {
6914       return false; // Not yet writable
6915     }
6916     if (from->is_reg()) {
6917       if (to->is_reg()) {
6918         if (from->is_Register() && to->is_Register()) {
6919           mov(to->as_Register(), from->as_Register());
6920         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
6921           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
6922         } else {
6923           ShouldNotReachHere();
6924         }
6925       } else {
6926         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
6927         Address to_addr = Address(sp, st_off);
6928         if (from->is_FloatRegister()) {
6929           if (bt == T_DOUBLE) {
6930              strd(from->as_FloatRegister(), to_addr);
6931           } else {
6932              assert(bt == T_FLOAT, "must be float");
6933              strs(from->as_FloatRegister(), to_addr);
6934           }
6935         } else {
6936           str(from->as_Register(), to_addr);
6937         }
6938       }
6939     } else {
6940       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
6941       if (to->is_reg()) {
6942         if (to->is_FloatRegister()) {
6943           if (bt == T_DOUBLE) {
6944             ldrd(to->as_FloatRegister(), from_addr);
6945           } else {
6946             assert(bt == T_FLOAT, "must be float");
6947             ldrs(to->as_FloatRegister(), from_addr);
6948           }
6949         } else {
6950           ldr(to->as_Register(), from_addr);
6951         }
6952       } else {
6953         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
6954         ldr(rscratch1, from_addr);
6955         str(rscratch1, Address(sp, st_off));
6956       }
6957     }
6958   }
6959 
6960   // Update register states
6961   reg_state[from->value()] = reg_writable;
6962   reg_state[to->value()] = reg_written;
6963   return true;
6964 }
6965 
6966 // Calculate the extra stack space required for packing or unpacking inline
6967 // args and adjust the stack pointer
6968 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
6969   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
6970   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
6971   assert(sp_inc > 0, "sanity");
6972 
6973   // Save a copy of the FP and LR here for deoptimization patching and frame walking
6974   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6975 
6976   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
6977   if (sp_inc < (1 << 9)) {
6978     sub(sp, sp, sp_inc);   // Fits in an immediate
6979   } else {
6980     mov(rscratch1, sp_inc);
6981     sub(sp, sp, rscratch1);
6982   }
6983 
6984   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
6985 }
6986 
6987 // Read all fields from an inline type oop and store the values in registers/stack slots
6988 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
6989                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
6990                                           RegState reg_state[]) {
6991   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
6992   assert(from->is_valid(), "source must be valid");
6993   bool progress = false;
6994 #ifdef ASSERT
6995   const int start_offset = offset();
6996 #endif
6997 
6998   Label L_null, L_notNull;
6999   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7000   Register tmp1 = r10;
7001   Register tmp2 = r11;
7002   Register fromReg = noreg;
7003   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, -1);
7004   bool done = true;
7005   bool mark_done = true;
7006   VMReg toReg;
7007   BasicType bt;
7008   // Check if argument requires a null check
7009   bool null_check = false;
7010   VMReg nullCheckReg;
7011   while (stream.next(nullCheckReg, bt)) {
7012     if (sig->at(stream.sig_index())._offset == -1) {
7013       null_check = true;
7014       break;
7015     }
7016   }
7017   stream.reset(sig_index, to_index);
7018   while (stream.next(toReg, bt)) {
7019     assert(toReg->is_valid(), "destination must be valid");
7020     int idx = (int)toReg->value();
7021     if (reg_state[idx] == reg_readonly) {
7022       if (idx != from->value()) {
7023         mark_done = false;
7024       }
7025       done = false;
7026       continue;
7027     } else if (reg_state[idx] == reg_written) {
7028       continue;
7029     }
7030     assert(reg_state[idx] == reg_writable, "must be writable");
7031     reg_state[idx] = reg_written;
7032     progress = true;
7033 
7034     if (fromReg == noreg) {
7035       if (from->is_reg()) {
7036         fromReg = from->as_Register();
7037       } else {
7038         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7039         ldr(tmp1, Address(sp, st_off));
7040         fromReg = tmp1;
7041       }
7042       if (null_check) {
7043         // Nullable inline type argument, emit null check
7044         cbz(fromReg, L_null);
7045       }
7046     }
7047     int off = sig->at(stream.sig_index())._offset;
7048     if (off == -1) {
7049       assert(null_check, "Missing null check at");
7050       if (toReg->is_stack()) {
7051         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7052         mov(tmp2, 1);
7053         str(tmp2, Address(sp, st_off));
7054       } else {
7055         mov(toReg->as_Register(), 1);
7056       }
7057       continue;
7058     }
7059     assert(off > 0, "offset in object should be positive");
7060     Address fromAddr = Address(fromReg, off);
7061     if (!toReg->is_FloatRegister()) {
7062       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7063       if (is_reference_type(bt)) {
7064         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7065       } else {
7066         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7067         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7068       }
7069       if (toReg->is_stack()) {
7070         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7071         str(dst, Address(sp, st_off));
7072       }
7073     } else if (bt == T_DOUBLE) {
7074       ldrd(toReg->as_FloatRegister(), fromAddr);
7075     } else {
7076       assert(bt == T_FLOAT, "must be float");
7077       ldrs(toReg->as_FloatRegister(), fromAddr);
7078     }
7079   }
7080   if (progress && null_check) {
7081     if (done) {
7082       b(L_notNull);
7083       bind(L_null);
7084       // Set IsInit field to zero to signal that the argument is null.
7085       // Also set all oop fields to zero to make the GC happy.
7086       stream.reset(sig_index, to_index);
7087       while (stream.next(toReg, bt)) {
7088         if (sig->at(stream.sig_index())._offset == -1 ||
7089             bt == T_OBJECT || bt == T_ARRAY) {
7090           if (toReg->is_stack()) {
7091             int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7092             str(zr, Address(sp, st_off));
7093           } else {
7094             mov(toReg->as_Register(), zr);
7095           }
7096         }
7097       }
7098       bind(L_notNull);
7099     } else {
7100       bind(L_null);
7101     }
7102   }
7103 
7104   sig_index = stream.sig_index();
7105   to_index = stream.regs_index();
7106 
7107   if (mark_done && reg_state[from->value()] != reg_written) {
7108     // This is okay because no one else will write to that slot
7109     reg_state[from->value()] = reg_writable;
7110   }
7111   from_index--;
7112   assert(progress || (start_offset == offset()), "should not emit code");
7113   return done;
7114 }
7115 
7116 // Pack fields back into an inline type oop
7117 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7118                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7119                                         RegState reg_state[], Register val_array) {
7120   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7121   assert(to->is_valid(), "destination must be valid");
7122 
7123   if (reg_state[to->value()] == reg_written) {
7124     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7125     return true; // Already written
7126   }
7127 
7128   // The GC barrier expanded by store_heap_oop below may call into the
7129   // runtime so use callee-saved registers for any values that need to be
7130   // preserved. The GC barrier assembler should take care of saving the
7131   // Java argument registers.
7132   // TODO 8284443 Isn't it an issue if below code uses r14 as tmp when it contains a spilled value?
7133   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7134   Register val_obj_tmp = r21;
7135   Register from_reg_tmp = r22;
7136   Register tmp1 = r14;
7137   Register tmp2 = r13;
7138   Register tmp3 = r12;
7139   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7140 
7141   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7142 
7143   if (reg_state[to->value()] == reg_readonly) {
7144     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7145       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7146       return false; // Not yet writable
7147     }
7148     val_obj = val_obj_tmp;
7149   }
7150 
7151   int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7152   load_heap_oop(val_obj, Address(val_array, index), tmp1, tmp2);
7153 
7154   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7155   VMReg fromReg;
7156   BasicType bt;
7157   Label L_null;
7158   while (stream.next(fromReg, bt)) {
7159     assert(fromReg->is_valid(), "source must be valid");
7160     reg_state[fromReg->value()] = reg_writable;
7161 
7162     int off = sig->at(stream.sig_index())._offset;
7163     if (off == -1) {
7164       // Nullable inline type argument, emit null check
7165       Label L_notNull;
7166       if (fromReg->is_stack()) {
7167         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7168         ldrb(tmp2, Address(sp, ld_off));
7169         cbnz(tmp2, L_notNull);
7170       } else {
7171         cbnz(fromReg->as_Register(), L_notNull);
7172       }
7173       mov(val_obj, 0);
7174       b(L_null);
7175       bind(L_notNull);
7176       continue;
7177     }
7178 
7179     assert(off > 0, "offset in object should be positive");
7180     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7181 
7182     // Pack the scalarized field into the value object.
7183     Address dst(val_obj, off);
7184 
7185     if (!fromReg->is_FloatRegister()) {
7186       Register src;
7187       if (fromReg->is_stack()) {
7188         src = from_reg_tmp;
7189         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7190         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7191       } else {
7192         src = fromReg->as_Register();
7193       }
7194       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7195       if (is_reference_type(bt)) {
7196         store_heap_oop(dst, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7197       } else {
7198         store_sized_value(dst, src, size_in_bytes);
7199       }
7200     } else if (bt == T_DOUBLE) {
7201       strd(fromReg->as_FloatRegister(), dst);
7202     } else {
7203       assert(bt == T_FLOAT, "must be float");
7204       strs(fromReg->as_FloatRegister(), dst);
7205     }
7206   }
7207   bind(L_null);
7208   sig_index = stream.sig_index();
7209   from_index = stream.regs_index();
7210 
7211   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7212   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7213   assert(success, "to register must be writeable");
7214 
7215   return true;
7216 }
7217 
7218 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7219   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7220 }
7221 
7222 void MacroAssembler::cache_wb(Address line) {
7223   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7224   assert(line.index() == noreg, "index should be noreg");
7225   assert(line.offset() == 0, "offset should be 0");
7226   // would like to assert this
7227   // assert(line._ext.shift == 0, "shift should be zero");
7228   if (VM_Version::supports_dcpop()) {
7229     // writeback using clear virtual address to point of persistence
7230     dc(Assembler::CVAP, line.base());
7231   } else {
7232     // no need to generate anything as Unsafe.writebackMemory should
7233     // never invoke this stub
7234   }
7235 }
7236 
7237 void MacroAssembler::cache_wbsync(bool is_pre) {
7238   // we only need a barrier post sync
7239   if (!is_pre) {
7240     membar(Assembler::AnyAny);
7241   }

7600   }
7601 
7602   // Check if the lock-stack is full.
7603   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7604   cmpw(top, (unsigned)LockStack::end_offset());
7605   br(Assembler::GE, slow);
7606 
7607   // Check for recursion.
7608   subw(t, top, oopSize);
7609   ldr(t, Address(rthread, t));
7610   cmp(obj, t);
7611   br(Assembler::EQ, push);
7612 
7613   // Check header for monitor (0b10).
7614   tst(mark, markWord::monitor_value);
7615   br(Assembler::NE, slow);
7616 
7617   // Try to lock. Transition lock bits 0b01 => 0b00
7618   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7619   orr(mark, mark, markWord::unlocked_value);
7620   if (EnableValhalla) {
7621     // Mask inline_type bit such that we go to the slow path if object is an inline type
7622     andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
7623   }
7624   eor(t, mark, markWord::unlocked_value);
7625   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7626           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7627   br(Assembler::NE, slow);
7628 
7629   bind(push);
7630   // After successful lock, push object on lock-stack.
7631   str(obj, Address(rthread, top));
7632   addw(top, top, oopSize);
7633   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7634 }
7635 
7636 // Implements lightweight-unlocking.
7637 //
7638 // - obj: the object to be unlocked
7639 // - t1, t2, t3: temporary registers
7640 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7641 void MacroAssembler::lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7642   assert(LockingMode == LM_LIGHTWEIGHT, "only used with new lightweight locking");
7643   // cmpxchg clobbers rscratch1.
< prev index next >