1 /*
   2  * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
   4  * Copyright (c) 2020, 2023, Huawei Technologies Co., Ltd. All rights reserved.
   5  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   6  *
   7  * This code is free software; you can redistribute it and/or modify it
   8  * under the terms of the GNU General Public License version 2 only, as
   9  * published by the Free Software Foundation.
  10  *
  11  * This code is distributed in the hope that it will be useful, but WITHOUT
  12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  14  * version 2 for more details (a copy is included in the LICENSE file that
  15  * accompanied this code).
  16  *
  17  * You should have received a copy of the GNU General Public License version
  18  * 2 along with this work; if not, write to the Free Software Foundation,
  19  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  20  *
  21  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  22  * or visit www.oracle.com if you need additional information or have any
  23  * questions.
  24  *
  25  */
  26 
  27 #include "precompiled.hpp"
  28 #include "asm/macroAssembler.hpp"
  29 #include "asm/macroAssembler.inline.hpp"
  30 #include "code/compiledIC.hpp"
  31 #include "code/debugInfoRec.hpp"
  32 #include "code/vtableStubs.hpp"
  33 #include "compiler/oopMap.hpp"
  34 #include "gc/shared/barrierSetAssembler.hpp"
  35 #include "interpreter/interp_masm.hpp"
  36 #include "interpreter/interpreter.hpp"
  37 #include "logging/log.hpp"
  38 #include "memory/resourceArea.hpp"
  39 #include "nativeInst_riscv.hpp"
  40 #include "oops/klass.inline.hpp"
  41 #include "oops/method.inline.hpp"
  42 #include "prims/methodHandles.hpp"
  43 #include "runtime/continuation.hpp"
  44 #include "runtime/continuationEntry.inline.hpp"
  45 #include "runtime/globals.hpp"
  46 #include "runtime/jniHandles.hpp"
  47 #include "runtime/safepointMechanism.hpp"
  48 #include "runtime/sharedRuntime.hpp"
  49 #include "runtime/signature.hpp"
  50 #include "runtime/stubRoutines.hpp"
  51 #include "runtime/vframeArray.hpp"
  52 #include "utilities/align.hpp"
  53 #include "utilities/formatBuffer.hpp"
  54 #include "vmreg_riscv.inline.hpp"
  55 #ifdef COMPILER1
  56 #include "c1/c1_Runtime1.hpp"
  57 #endif
  58 #ifdef COMPILER2
  59 #include "adfiles/ad_riscv.hpp"
  60 #include "opto/runtime.hpp"
  61 #endif
  62 #if INCLUDE_JVMCI
  63 #include "jvmci/jvmciJavaClasses.hpp"
  64 #endif
  65 
  66 #define __ masm->
  67 
  68 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
  69 
  70 class SimpleRuntimeFrame {
  71 public:
  72 
  73   // Most of the runtime stubs have this simple frame layout.
  74   // This class exists to make the layout shared in one place.
  75   // Offsets are for compiler stack slots, which are jints.
  76   enum layout {
  77     // The frame sender code expects that fp will be in the "natural" place and
  78     // will override any oopMap setting for it. We must therefore force the layout
  79     // so that it agrees with the frame sender code.
  80     // we don't expect any arg reg save area so riscv asserts that
  81     // frame::arg_reg_save_area_bytes == 0
  82     fp_off = 0, fp_off2,
  83     return_off, return_off2,
  84     framesize
  85   };
  86 };
  87 
  88 class RegisterSaver {
  89   const bool _save_vectors;
  90  public:
  91   RegisterSaver(bool save_vectors) : _save_vectors(UseRVV && save_vectors) {}
  92   ~RegisterSaver() {}
  93   OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words);
  94   void restore_live_registers(MacroAssembler* masm);
  95 
  96   // Offsets into the register save area
  97   // Used by deoptimization when it is managing result register
  98   // values on its own
  99   // gregs:28, float_register:32; except: x1(ra) & x2(sp) & gp(x3) & tp(x4)
 100   // |---v0---|<---SP
 101   // |---v1---|save vectors only in generate_handler_blob
 102   // |-- .. --|
 103   // |---v31--|-----
 104   // |---f0---|
 105   // |---f1---|
 106   // |   ..   |
 107   // |---f31--|
 108   // |---reserved slot for stack alignment---|
 109   // |---x5---|
 110   // |   x6   |
 111   // |---.. --|
 112   // |---x31--|
 113   // |---fp---|
 114   // |---ra---|
 115   int v0_offset_in_bytes(void) { return 0; }
 116   int f0_offset_in_bytes(void) {
 117     int f0_offset = 0;
 118 #ifdef COMPILER2
 119     if (_save_vectors) {
 120       f0_offset += Matcher::scalable_vector_reg_size(T_INT) * VectorRegister::number_of_registers *
 121                    BytesPerInt;
 122     }
 123 #endif
 124     return f0_offset;
 125   }
 126   int reserved_slot_offset_in_bytes(void) {
 127     return f0_offset_in_bytes() +
 128            FloatRegister::max_slots_per_register *
 129            FloatRegister::number_of_registers *
 130            BytesPerInt;
 131   }
 132 
 133   int reg_offset_in_bytes(Register r) {
 134     assert (r->encoding() > 4, "ra, sp, gp and tp not saved");
 135     return reserved_slot_offset_in_bytes() + (r->encoding() - 4 /* x1, x2, x3, x4 */) * wordSize;
 136   }
 137 
 138   int freg_offset_in_bytes(FloatRegister f) {
 139     return f0_offset_in_bytes() + f->encoding() * wordSize;
 140   }
 141 
 142   int ra_offset_in_bytes(void) {
 143     return reserved_slot_offset_in_bytes() +
 144            (Register::number_of_registers - 3) *
 145            Register::max_slots_per_register *
 146            BytesPerInt;
 147   }
 148 };
 149 
 150 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words) {
 151   int vector_size_in_bytes = 0;
 152   int vector_size_in_slots = 0;
 153 #ifdef COMPILER2
 154   if (_save_vectors) {
 155     vector_size_in_bytes += Matcher::scalable_vector_reg_size(T_BYTE);
 156     vector_size_in_slots += Matcher::scalable_vector_reg_size(T_INT);
 157   }
 158 #endif
 159 
 160   int frame_size_in_bytes = align_up(additional_frame_words * wordSize + ra_offset_in_bytes() + wordSize, 16);
 161   // OopMap frame size is in compiler stack slots (jint's) not bytes or words
 162   int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
 163   // The caller will allocate additional_frame_words
 164   int additional_frame_slots = additional_frame_words * wordSize / BytesPerInt;
 165   // CodeBlob frame size is in words.
 166   int frame_size_in_words = frame_size_in_bytes / wordSize;
 167   *total_frame_words = frame_size_in_words;
 168 
 169   // Save Integer, Float and Vector registers.
 170   __ enter();
 171   __ push_CPU_state(_save_vectors, vector_size_in_bytes);
 172 
 173   // Set an oopmap for the call site.  This oopmap will map all
 174   // oop-registers and debug-info registers as callee-saved.  This
 175   // will allow deoptimization at this safepoint to find all possible
 176   // debug-info recordings, as well as let GC find all oops.
 177 
 178   OopMapSet *oop_maps = new OopMapSet();
 179   OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
 180   assert_cond(oop_maps != nullptr && oop_map != nullptr);
 181 
 182   int sp_offset_in_slots = 0;
 183   int step_in_slots = 0;
 184   if (_save_vectors) {
 185     step_in_slots = vector_size_in_slots;
 186     for (int i = 0; i < VectorRegister::number_of_registers; i++, sp_offset_in_slots += step_in_slots) {
 187       VectorRegister r = as_VectorRegister(i);
 188       oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset_in_slots), r->as_VMReg());
 189     }
 190   }
 191 
 192   step_in_slots = FloatRegister::max_slots_per_register;
 193   for (int i = 0; i < FloatRegister::number_of_registers; i++, sp_offset_in_slots += step_in_slots) {
 194     FloatRegister r = as_FloatRegister(i);
 195     oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset_in_slots), r->as_VMReg());
 196   }
 197 
 198   step_in_slots = Register::max_slots_per_register;
 199   // skip the slot reserved for alignment, see MacroAssembler::push_reg;
 200   // also skip x5 ~ x6 on the stack because they are caller-saved registers.
 201   sp_offset_in_slots += Register::max_slots_per_register * 3;
 202   // besides, we ignore x0 ~ x4 because push_CPU_state won't push them on the stack.
 203   for (int i = 7; i < Register::number_of_registers; i++, sp_offset_in_slots += step_in_slots) {
 204     Register r = as_Register(i);
 205     if (r != xthread) {
 206       oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset_in_slots + additional_frame_slots), r->as_VMReg());
 207     }
 208   }
 209 
 210   return oop_map;
 211 }
 212 
 213 void RegisterSaver::restore_live_registers(MacroAssembler* masm) {
 214 #ifdef COMPILER2
 215   __ pop_CPU_state(_save_vectors, Matcher::scalable_vector_reg_size(T_BYTE));
 216 #else
 217 #if !INCLUDE_JVMCI
 218   assert(!_save_vectors, "vectors are generated only by C2 and JVMCI");
 219 #endif
 220   __ pop_CPU_state(_save_vectors);
 221 #endif
 222   __ leave();
 223 }
 224 
 225 // Is vector's size (in bytes) bigger than a size saved by default?
 226 // riscv does not ovlerlay the floating-point registers on vector registers like aarch64.
 227 bool SharedRuntime::is_wide_vector(int size) {
 228   return UseRVV;
 229 }
 230 
 231 // ---------------------------------------------------------------------------
 232 // Read the array of BasicTypes from a signature, and compute where the
 233 // arguments should go.  Values in the VMRegPair regs array refer to 4-byte
 234 // quantities.  Values less than VMRegImpl::stack0 are registers, those above
 235 // refer to 4-byte stack slots.  All stack slots are based off of the stack pointer
 236 // as framesizes are fixed.
 237 // VMRegImpl::stack0 refers to the first slot 0(sp).
 238 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher.
 239 // Register up to Register::number_of_registers) are the 64-bit
 240 // integer registers.
 241 
 242 // Note: the INPUTS in sig_bt are in units of Java argument words,
 243 // which are 64-bit.  The OUTPUTS are in 32-bit units.
 244 
 245 // The Java calling convention is a "shifted" version of the C ABI.
 246 // By skipping the first C ABI register we can call non-static jni
 247 // methods with small numbers of arguments without having to shuffle
 248 // the arguments at all. Since we control the java ABI we ought to at
 249 // least get some advantage out of it.
 250 
 251 int SharedRuntime::java_calling_convention(const BasicType *sig_bt,
 252                                            VMRegPair *regs,
 253                                            int total_args_passed) {
 254   // Create the mapping between argument positions and
 255   // registers.
 256   static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = {
 257     j_rarg0, j_rarg1, j_rarg2, j_rarg3,
 258     j_rarg4, j_rarg5, j_rarg6, j_rarg7
 259   };
 260   static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters_j] = {
 261     j_farg0, j_farg1, j_farg2, j_farg3,
 262     j_farg4, j_farg5, j_farg6, j_farg7
 263   };
 264 
 265   uint int_args = 0;
 266   uint fp_args = 0;
 267   uint stk_args = 0;
 268 
 269   for (int i = 0; i < total_args_passed; i++) {
 270     switch (sig_bt[i]) {
 271       case T_BOOLEAN: // fall through
 272       case T_CHAR:    // fall through
 273       case T_BYTE:    // fall through
 274       case T_SHORT:   // fall through
 275       case T_INT:
 276         if (int_args < Argument::n_int_register_parameters_j) {
 277           regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
 278         } else {
 279           stk_args = align_up(stk_args, 2);
 280           regs[i].set1(VMRegImpl::stack2reg(stk_args));
 281           stk_args += 1;
 282         }
 283         break;
 284       case T_VOID:
 285         // halves of T_LONG or T_DOUBLE
 286         assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
 287         regs[i].set_bad();
 288         break;
 289       case T_LONG:      // fall through
 290         assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
 291       case T_OBJECT:    // fall through
 292       case T_ARRAY:     // fall through
 293       case T_ADDRESS:
 294         if (int_args < Argument::n_int_register_parameters_j) {
 295           regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
 296         } else {
 297           stk_args = align_up(stk_args, 2);
 298           regs[i].set2(VMRegImpl::stack2reg(stk_args));
 299           stk_args += 2;
 300         }
 301         break;
 302       case T_FLOAT:
 303         if (fp_args < Argument::n_float_register_parameters_j) {
 304           regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
 305         } else {
 306           stk_args = align_up(stk_args, 2);
 307           regs[i].set1(VMRegImpl::stack2reg(stk_args));
 308           stk_args += 1;
 309         }
 310         break;
 311       case T_DOUBLE:
 312         assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
 313         if (fp_args < Argument::n_float_register_parameters_j) {
 314           regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
 315         } else {
 316           stk_args = align_up(stk_args, 2);
 317           regs[i].set2(VMRegImpl::stack2reg(stk_args));
 318           stk_args += 2;
 319         }
 320         break;
 321       default:
 322         ShouldNotReachHere();
 323     }
 324   }
 325 
 326   return stk_args;
 327 }
 328 
 329 // Patch the callers callsite with entry to compiled code if it exists.
 330 static void patch_callers_callsite(MacroAssembler *masm) {
 331   Label L;
 332   __ ld(t0, Address(xmethod, in_bytes(Method::code_offset())));
 333   __ beqz(t0, L);
 334 
 335   __ enter();
 336   __ push_CPU_state();
 337 
 338   // VM needs caller's callsite
 339   // VM needs target method
 340   // This needs to be a long call since we will relocate this adapter to
 341   // the codeBuffer and it may not reach
 342 
 343 #ifndef PRODUCT
 344   assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
 345 #endif
 346 
 347   __ mv(c_rarg0, xmethod);
 348   __ mv(c_rarg1, ra);
 349   __ rt_call(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite));
 350 
 351   __ pop_CPU_state();
 352   // restore sp
 353   __ leave();
 354   __ bind(L);
 355 }
 356 
 357 static void gen_c2i_adapter(MacroAssembler *masm,
 358                             int total_args_passed,
 359                             int comp_args_on_stack,
 360                             const BasicType *sig_bt,
 361                             const VMRegPair *regs,
 362                             Label& skip_fixup) {
 363   // Before we get into the guts of the C2I adapter, see if we should be here
 364   // at all.  We've come from compiled code and are attempting to jump to the
 365   // interpreter, which means the caller made a static call to get here
 366   // (vcalls always get a compiled target if there is one).  Check for a
 367   // compiled target.  If there is one, we need to patch the caller's call.
 368   patch_callers_callsite(masm);
 369 
 370   __ bind(skip_fixup);
 371 
 372   int words_pushed = 0;
 373 
 374   // Since all args are passed on the stack, total_args_passed *
 375   // Interpreter::stackElementSize is the space we need.
 376 
 377   int extraspace = total_args_passed * Interpreter::stackElementSize;
 378 
 379   __ mv(x19_sender_sp, sp);
 380 
 381   // stack is aligned, keep it that way
 382   extraspace = align_up(extraspace, 2 * wordSize);
 383 
 384   if (extraspace) {
 385     __ sub(sp, sp, extraspace);
 386   }
 387 
 388   // Now write the args into the outgoing interpreter space
 389   for (int i = 0; i < total_args_passed; i++) {
 390     if (sig_bt[i] == T_VOID) {
 391       assert(i > 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "missing half");
 392       continue;
 393     }
 394 
 395     // offset to start parameters
 396     int st_off   = (total_args_passed - i - 1) * Interpreter::stackElementSize;
 397     int next_off = st_off - Interpreter::stackElementSize;
 398 
 399     // Say 4 args:
 400     // i   st_off
 401     // 0   32 T_LONG
 402     // 1   24 T_VOID
 403     // 2   16 T_OBJECT
 404     // 3    8 T_BOOL
 405     // -    0 return address
 406     //
 407     // However to make thing extra confusing. Because we can fit a Java long/double in
 408     // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
 409     // leaves one slot empty and only stores to a single slot. In this case the
 410     // slot that is occupied is the T_VOID slot. See I said it was confusing.
 411 
 412     VMReg r_1 = regs[i].first();
 413     VMReg r_2 = regs[i].second();
 414     if (!r_1->is_valid()) {
 415       assert(!r_2->is_valid(), "");
 416       continue;
 417     }
 418     if (r_1->is_stack()) {
 419       // memory to memory use t0
 420       int ld_off = (r_1->reg2stack() * VMRegImpl::stack_slot_size
 421                     + extraspace
 422                     + words_pushed * wordSize);
 423       if (!r_2->is_valid()) {
 424         __ lwu(t0, Address(sp, ld_off));
 425         __ sd(t0, Address(sp, st_off), /*temp register*/esp);
 426       } else {
 427         __ ld(t0, Address(sp, ld_off), /*temp register*/esp);
 428 
 429         // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
 430         // T_DOUBLE and T_LONG use two slots in the interpreter
 431         if (sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
 432           // ld_off == LSW, ld_off+wordSize == MSW
 433           // st_off == MSW, next_off == LSW
 434           __ sd(t0, Address(sp, next_off), /*temp register*/esp);
 435 #ifdef ASSERT
 436           // Overwrite the unused slot with known junk
 437           __ mv(t0, 0xdeadffffdeadaaaaul);
 438           __ sd(t0, Address(sp, st_off), /*temp register*/esp);
 439 #endif /* ASSERT */
 440         } else {
 441           __ sd(t0, Address(sp, st_off), /*temp register*/esp);
 442         }
 443       }
 444     } else if (r_1->is_Register()) {
 445       Register r = r_1->as_Register();
 446       if (!r_2->is_valid()) {
 447         // must be only an int (or less ) so move only 32bits to slot
 448         __ sd(r, Address(sp, st_off));
 449       } else {
 450         // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
 451         // T_DOUBLE and T_LONG use two slots in the interpreter
 452         if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
 453           // long/double in gpr
 454 #ifdef ASSERT
 455           // Overwrite the unused slot with known junk
 456           __ mv(t0, 0xdeadffffdeadaaabul);
 457           __ sd(t0, Address(sp, st_off), /*temp register*/esp);
 458 #endif /* ASSERT */
 459           __ sd(r, Address(sp, next_off));
 460         } else {
 461           __ sd(r, Address(sp, st_off));
 462         }
 463       }
 464     } else {
 465       assert(r_1->is_FloatRegister(), "");
 466       if (!r_2->is_valid()) {
 467         // only a float use just part of the slot
 468         __ fsw(r_1->as_FloatRegister(), Address(sp, st_off));
 469       } else {
 470 #ifdef ASSERT
 471         // Overwrite the unused slot with known junk
 472         __ mv(t0, 0xdeadffffdeadaaacul);
 473         __ sd(t0, Address(sp, st_off), /*temp register*/esp);
 474 #endif /* ASSERT */
 475         __ fsd(r_1->as_FloatRegister(), Address(sp, next_off));
 476       }
 477     }
 478   }
 479 
 480   __ mv(esp, sp); // Interp expects args on caller's expression stack
 481 
 482   __ ld(t0, Address(xmethod, in_bytes(Method::interpreter_entry_offset())));
 483   __ jr(t0);
 484 }
 485 
 486 void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,
 487                                     int total_args_passed,
 488                                     int comp_args_on_stack,
 489                                     const BasicType *sig_bt,
 490                                     const VMRegPair *regs) {
 491   // Note: x19_sender_sp contains the senderSP on entry. We must
 492   // preserve it since we may do a i2c -> c2i transition if we lose a
 493   // race where compiled code goes non-entrant while we get args
 494   // ready.
 495 
 496   // Cut-out for having no stack args.
 497   int comp_words_on_stack = align_up(comp_args_on_stack * VMRegImpl::stack_slot_size, wordSize) >> LogBytesPerWord;
 498   if (comp_args_on_stack != 0) {
 499     __ sub(t0, sp, comp_words_on_stack * wordSize);
 500     __ andi(sp, t0, -16);
 501   }
 502 
 503   // Will jump to the compiled code just as if compiled code was doing it.
 504   // Pre-load the register-jump target early, to schedule it better.
 505   __ ld(t1, Address(xmethod, in_bytes(Method::from_compiled_offset())));
 506 
 507 #if INCLUDE_JVMCI
 508   if (EnableJVMCI) {
 509     // check if this call should be routed towards a specific entry point
 510     __ ld(t0, Address(xthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
 511     Label no_alternative_target;
 512     __ beqz(t0, no_alternative_target);
 513     __ mv(t1, t0);
 514     __ sd(zr, Address(xthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
 515     __ bind(no_alternative_target);
 516   }
 517 #endif // INCLUDE_JVMCI
 518 
 519   // Now generate the shuffle code.
 520   for (int i = 0; i < total_args_passed; i++) {
 521     if (sig_bt[i] == T_VOID) {
 522       assert(i > 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "missing half");
 523       continue;
 524     }
 525 
 526     // Pick up 0, 1 or 2 words from SP+offset.
 527 
 528     assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
 529            "scrambled load targets?");
 530     // Load in argument order going down.
 531     int ld_off = (total_args_passed - i - 1) * Interpreter::stackElementSize;
 532     // Point to interpreter value (vs. tag)
 533     int next_off = ld_off - Interpreter::stackElementSize;
 534 
 535     VMReg r_1 = regs[i].first();
 536     VMReg r_2 = regs[i].second();
 537     if (!r_1->is_valid()) {
 538       assert(!r_2->is_valid(), "");
 539       continue;
 540     }
 541     if (r_1->is_stack()) {
 542       // Convert stack slot to an SP offset (+ wordSize to account for return address )
 543       int st_off = regs[i].first()->reg2stack() * VMRegImpl::stack_slot_size;
 544       if (!r_2->is_valid()) {
 545         __ lw(t0, Address(esp, ld_off));
 546         __ sd(t0, Address(sp, st_off), /*temp register*/t2);
 547       } else {
 548         //
 549         // We are using two optoregs. This can be either T_OBJECT,
 550         // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
 551         // two slots but only uses one for thr T_LONG or T_DOUBLE case
 552         // So we must adjust where to pick up the data to match the
 553         // interpreter.
 554         //
 555         // Interpreter local[n] == MSW, local[n+1] == LSW however locals
 556         // are accessed as negative so LSW is at LOW address
 557 
 558         // ld_off is MSW so get LSW
 559         const int offset = (sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) ?
 560                            next_off : ld_off;
 561         __ ld(t0, Address(esp, offset));
 562         // st_off is LSW (i.e. reg.first())
 563         __ sd(t0, Address(sp, st_off), /*temp register*/t2);
 564       }
 565     } else if (r_1->is_Register()) {  // Register argument
 566       Register r = r_1->as_Register();
 567       if (r_2->is_valid()) {
 568         //
 569         // We are using two VMRegs. This can be either T_OBJECT,
 570         // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
 571         // two slots but only uses one for thr T_LONG or T_DOUBLE case
 572         // So we must adjust where to pick up the data to match the
 573         // interpreter.
 574 
 575         const int offset = (sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) ?
 576                            next_off : ld_off;
 577 
 578         // this can be a misaligned move
 579         __ ld(r, Address(esp, offset));
 580       } else {
 581         // sign extend and use a full word?
 582         __ lw(r, Address(esp, ld_off));
 583       }
 584     } else {
 585       if (!r_2->is_valid()) {
 586         __ flw(r_1->as_FloatRegister(), Address(esp, ld_off));
 587       } else {
 588         __ fld(r_1->as_FloatRegister(), Address(esp, next_off));
 589       }
 590     }
 591   }
 592 
 593   __ push_cont_fastpath(xthread); // Set JavaThread::_cont_fastpath to the sp of the oldest interpreted frame we know about
 594 
 595   // 6243940 We might end up in handle_wrong_method if
 596   // the callee is deoptimized as we race thru here. If that
 597   // happens we don't want to take a safepoint because the
 598   // caller frame will look interpreted and arguments are now
 599   // "compiled" so it is much better to make this transition
 600   // invisible to the stack walking code. Unfortunately if
 601   // we try and find the callee by normal means a safepoint
 602   // is possible. So we stash the desired callee in the thread
 603   // and the vm will find there should this case occur.
 604 
 605   __ sd(xmethod, Address(xthread, JavaThread::callee_target_offset()));
 606 
 607   __ jr(t1);
 608 }
 609 
 610 // ---------------------------------------------------------------
 611 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
 612                                                             int total_args_passed,
 613                                                             int comp_args_on_stack,
 614                                                             const BasicType *sig_bt,
 615                                                             const VMRegPair *regs,
 616                                                             AdapterFingerPrint* fingerprint) {
 617   address i2c_entry = __ pc();
 618   gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
 619 
 620   address c2i_unverified_entry = __ pc();
 621   Label skip_fixup;
 622 
 623   const Register receiver = j_rarg0;
 624   const Register data = t1;
 625   const Register tmp = t2;  // A call-clobbered register not used for arg passing
 626 
 627   // -------------------------------------------------------------------------
 628   // Generate a C2I adapter.  On entry we know xmethod holds the Method* during calls
 629   // to the interpreter.  The args start out packed in the compiled layout.  They
 630   // need to be unpacked into the interpreter layout.  This will almost always
 631   // require some stack space.  We grow the current (compiled) stack, then repack
 632   // the args.  We  finally end in a jump to the generic interpreter entry point.
 633   // On exit from the interpreter, the interpreter will restore our SP (lest the
 634   // compiled code, which relies solely on SP and not FP, get sick).
 635 
 636   {
 637     __ block_comment("c2i_unverified_entry {");
 638 
 639     __ ic_check();
 640     __ ld(xmethod, Address(data, CompiledICData::speculated_method_offset()));
 641 
 642     __ ld(t0, Address(xmethod, in_bytes(Method::code_offset())));
 643     __ beqz(t0, skip_fixup);
 644     __ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 645     __ block_comment("} c2i_unverified_entry");
 646   }
 647 
 648   address c2i_entry = __ pc();
 649 
 650   // Class initialization barrier for static methods
 651   address c2i_no_clinit_check_entry = nullptr;
 652   if (VM_Version::supports_fast_class_init_checks()) {
 653     Label L_skip_barrier;
 654 
 655     { // Bypass the barrier for non-static methods
 656       __ lwu(t0, Address(xmethod, Method::access_flags_offset()));
 657       __ test_bit(t1, t0, exact_log2(JVM_ACC_STATIC));
 658       __ beqz(t1, L_skip_barrier); // non-static
 659     }
 660 
 661     __ load_method_holder(t1, xmethod);
 662     __ clinit_barrier(t1, t0, &L_skip_barrier);
 663     __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
 664 
 665     __ bind(L_skip_barrier);
 666     c2i_no_clinit_check_entry = __ pc();
 667   }
 668 
 669   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 670   bs->c2i_entry_barrier(masm);
 671 
 672   gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
 673 
 674   return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry, c2i_no_clinit_check_entry);
 675 }
 676 
 677 int SharedRuntime::vector_calling_convention(VMRegPair *regs,
 678                                              uint num_bits,
 679                                              uint total_args_passed) {
 680   Unimplemented();
 681   return 0;
 682 }
 683 
 684 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
 685                                          VMRegPair *regs,
 686                                          int total_args_passed) {
 687 
 688   // We return the amount of VMRegImpl stack slots we need to reserve for all
 689   // the arguments NOT counting out_preserve_stack_slots.
 690 
 691   static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
 692     c_rarg0, c_rarg1, c_rarg2, c_rarg3,
 693     c_rarg4, c_rarg5,  c_rarg6,  c_rarg7
 694   };
 695   static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
 696     c_farg0, c_farg1, c_farg2, c_farg3,
 697     c_farg4, c_farg5, c_farg6, c_farg7
 698   };
 699 
 700   uint int_args = 0;
 701   uint fp_args = 0;
 702   uint stk_args = 0; // inc by 2 each time
 703 
 704   for (int i = 0; i < total_args_passed; i++) {
 705     switch (sig_bt[i]) {
 706       case T_BOOLEAN:  // fall through
 707       case T_CHAR:     // fall through
 708       case T_BYTE:     // fall through
 709       case T_SHORT:    // fall through
 710       case T_INT:
 711         if (int_args < Argument::n_int_register_parameters_c) {
 712           regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
 713         } else {
 714           regs[i].set1(VMRegImpl::stack2reg(stk_args));
 715           stk_args += 2;
 716         }
 717         break;
 718       case T_LONG:      // fall through
 719         assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
 720       case T_OBJECT:    // fall through
 721       case T_ARRAY:     // fall through
 722       case T_ADDRESS:   // fall through
 723       case T_METADATA:
 724         if (int_args < Argument::n_int_register_parameters_c) {
 725           regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
 726         } else {
 727           regs[i].set2(VMRegImpl::stack2reg(stk_args));
 728           stk_args += 2;
 729         }
 730         break;
 731       case T_FLOAT:
 732         if (fp_args < Argument::n_float_register_parameters_c) {
 733           regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
 734         } else if (int_args < Argument::n_int_register_parameters_c) {
 735           regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
 736         } else {
 737           regs[i].set1(VMRegImpl::stack2reg(stk_args));
 738           stk_args += 2;
 739         }
 740         break;
 741       case T_DOUBLE:
 742         assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
 743         if (fp_args < Argument::n_float_register_parameters_c) {
 744           regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
 745         } else if (int_args < Argument::n_int_register_parameters_c) {
 746           regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
 747         } else {
 748           regs[i].set2(VMRegImpl::stack2reg(stk_args));
 749           stk_args += 2;
 750         }
 751         break;
 752       case T_VOID: // Halves of longs and doubles
 753         assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
 754         regs[i].set_bad();
 755         break;
 756       default:
 757         ShouldNotReachHere();
 758     }
 759   }
 760 
 761   return stk_args;
 762 }
 763 
 764 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
 765   // We always ignore the frame_slots arg and just use the space just below frame pointer
 766   // which by this time is free to use
 767   switch (ret_type) {
 768     case T_FLOAT:
 769       __ fsw(f10, Address(fp, -3 * wordSize));
 770       break;
 771     case T_DOUBLE:
 772       __ fsd(f10, Address(fp, -3 * wordSize));
 773       break;
 774     case T_VOID:  break;
 775     default: {
 776       __ sd(x10, Address(fp, -3 * wordSize));
 777     }
 778   }
 779 }
 780 
 781 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
 782   // We always ignore the frame_slots arg and just use the space just below frame pointer
 783   // which by this time is free to use
 784   switch (ret_type) {
 785     case T_FLOAT:
 786       __ flw(f10, Address(fp, -3 * wordSize));
 787       break;
 788     case T_DOUBLE:
 789       __ fld(f10, Address(fp, -3 * wordSize));
 790       break;
 791     case T_VOID:  break;
 792     default: {
 793       __ ld(x10, Address(fp, -3 * wordSize));
 794     }
 795   }
 796 }
 797 
 798 static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
 799   RegSet x;
 800   for ( int i = first_arg ; i < arg_count ; i++ ) {
 801     if (args[i].first()->is_Register()) {
 802       x = x + args[i].first()->as_Register();
 803     } else if (args[i].first()->is_FloatRegister()) {
 804       __ addi(sp, sp, -2 * wordSize);
 805       __ fsd(args[i].first()->as_FloatRegister(), Address(sp, 0));
 806     }
 807   }
 808   __ push_reg(x, sp);
 809 }
 810 
 811 static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
 812   RegSet x;
 813   for ( int i = first_arg ; i < arg_count ; i++ ) {
 814     if (args[i].first()->is_Register()) {
 815       x = x + args[i].first()->as_Register();
 816     } else {
 817       ;
 818     }
 819   }
 820   __ pop_reg(x, sp);
 821   for ( int i = arg_count - 1 ; i >= first_arg ; i-- ) {
 822     if (args[i].first()->is_Register()) {
 823       ;
 824     } else if (args[i].first()->is_FloatRegister()) {
 825       __ fld(args[i].first()->as_FloatRegister(), Address(sp, 0));
 826       __ add(sp, sp, 2 * wordSize);
 827     }
 828   }
 829 }
 830 
 831 static void verify_oop_args(MacroAssembler* masm,
 832                             const methodHandle& method,
 833                             const BasicType* sig_bt,
 834                             const VMRegPair* regs) {
 835   const Register temp_reg = x9;  // not part of any compiled calling seq
 836   if (VerifyOops) {
 837     for (int i = 0; i < method->size_of_parameters(); i++) {
 838       if (sig_bt[i] == T_OBJECT ||
 839           sig_bt[i] == T_ARRAY) {
 840         VMReg r = regs[i].first();
 841         assert(r->is_valid(), "bad oop arg");
 842         if (r->is_stack()) {
 843           __ ld(temp_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
 844           __ verify_oop(temp_reg);
 845         } else {
 846           __ verify_oop(r->as_Register());
 847         }
 848       }
 849     }
 850   }
 851 }
 852 
 853 // on exit, sp points to the ContinuationEntry
 854 static OopMap* continuation_enter_setup(MacroAssembler* masm, int& stack_slots) {
 855   assert(ContinuationEntry::size() % VMRegImpl::stack_slot_size == 0, "");
 856   assert(in_bytes(ContinuationEntry::cont_offset())  % VMRegImpl::stack_slot_size == 0, "");
 857   assert(in_bytes(ContinuationEntry::chunk_offset()) % VMRegImpl::stack_slot_size == 0, "");
 858 
 859   stack_slots += (int)ContinuationEntry::size() / wordSize;
 860   __ sub(sp, sp, (int)ContinuationEntry::size()); // place Continuation metadata
 861 
 862   OopMap* map = new OopMap(((int)ContinuationEntry::size() + wordSize) / VMRegImpl::stack_slot_size, 0 /* arg_slots*/);
 863 
 864   __ ld(t0, Address(xthread, JavaThread::cont_entry_offset()));
 865   __ sd(t0, Address(sp, ContinuationEntry::parent_offset()));
 866   __ sd(sp, Address(xthread, JavaThread::cont_entry_offset()));
 867 
 868   return map;
 869 }
 870 
 871 // on entry c_rarg1 points to the continuation
 872 //          sp points to ContinuationEntry
 873 //          c_rarg3 -- isVirtualThread
 874 static void fill_continuation_entry(MacroAssembler* masm) {
 875 #ifdef ASSERT
 876   __ mv(t0, ContinuationEntry::cookie_value());
 877   __ sw(t0, Address(sp, ContinuationEntry::cookie_offset()));
 878 #endif
 879 
 880   __ sd(c_rarg1, Address(sp, ContinuationEntry::cont_offset()));
 881   __ sw(c_rarg3, Address(sp, ContinuationEntry::flags_offset()));
 882   __ sd(zr,      Address(sp, ContinuationEntry::chunk_offset()));
 883   __ sw(zr,      Address(sp, ContinuationEntry::argsize_offset()));
 884   __ sw(zr,      Address(sp, ContinuationEntry::pin_count_offset()));
 885 
 886   __ ld(t0, Address(xthread, JavaThread::cont_fastpath_offset()));
 887   __ sd(t0, Address(sp, ContinuationEntry::parent_cont_fastpath_offset()));
 888   __ ld(t0, Address(xthread, JavaThread::held_monitor_count_offset()));
 889   __ sd(t0, Address(sp, ContinuationEntry::parent_held_monitor_count_offset()));
 890 
 891   __ sd(zr, Address(xthread, JavaThread::cont_fastpath_offset()));
 892   __ sd(zr, Address(xthread, JavaThread::held_monitor_count_offset()));
 893 }
 894 
 895 // on entry, sp points to the ContinuationEntry
 896 // on exit, fp points to the spilled fp + 2 * wordSize in the entry frame
 897 static void continuation_enter_cleanup(MacroAssembler* masm) {
 898 #ifndef PRODUCT
 899   Label OK;
 900   __ ld(t0, Address(xthread, JavaThread::cont_entry_offset()));
 901   __ beq(sp, t0, OK);
 902   __ stop("incorrect sp");
 903   __ bind(OK);
 904 #endif
 905 
 906   __ ld(t0, Address(sp, ContinuationEntry::parent_cont_fastpath_offset()));
 907   __ sd(t0, Address(xthread, JavaThread::cont_fastpath_offset()));
 908   __ ld(t0, Address(sp, ContinuationEntry::parent_held_monitor_count_offset()));
 909   __ sd(t0, Address(xthread, JavaThread::held_monitor_count_offset()));
 910 
 911   __ ld(t0, Address(sp, ContinuationEntry::parent_offset()));
 912   __ sd(t0, Address(xthread, JavaThread::cont_entry_offset()));
 913   __ add(fp, sp, (int)ContinuationEntry::size() + 2 * wordSize /* 2 extra words to match up with leave() */);
 914 }
 915 
 916 // enterSpecial(Continuation c, boolean isContinue, boolean isVirtualThread)
 917 // On entry: c_rarg1 -- the continuation object
 918 //           c_rarg2 -- isContinue
 919 //           c_rarg3 -- isVirtualThread
 920 static void gen_continuation_enter(MacroAssembler* masm,
 921                                    const methodHandle& method,
 922                                    const BasicType* sig_bt,
 923                                    const VMRegPair* regs,
 924                                    int& exception_offset,
 925                                    OopMapSet*oop_maps,
 926                                    int& frame_complete,
 927                                    int& stack_slots,
 928                                    int& interpreted_entry_offset,
 929                                    int& compiled_entry_offset) {
 930   // verify_oop_args(masm, method, sig_bt, regs);
 931   Address resolve(SharedRuntime::get_resolve_static_call_stub(), relocInfo::static_call_type);
 932 
 933   address start = __ pc();
 934 
 935   Label call_thaw, exit;
 936 
 937   // i2i entry used at interp_only_mode only
 938   interpreted_entry_offset = __ pc() - start;
 939   {
 940 #ifdef ASSERT
 941     Label is_interp_only;
 942     __ lw(t0, Address(xthread, JavaThread::interp_only_mode_offset()));
 943     __ bnez(t0, is_interp_only);
 944     __ stop("enterSpecial interpreter entry called when not in interp_only_mode");
 945     __ bind(is_interp_only);
 946 #endif
 947 
 948     // Read interpreter arguments into registers (this is an ad-hoc i2c adapter)
 949     __ ld(c_rarg1, Address(esp, Interpreter::stackElementSize * 2));
 950     __ ld(c_rarg2, Address(esp, Interpreter::stackElementSize * 1));
 951     __ ld(c_rarg3, Address(esp, Interpreter::stackElementSize * 0));
 952     __ push_cont_fastpath(xthread);
 953 
 954     __ enter();
 955     stack_slots = 2; // will be adjusted in setup
 956     OopMap* map = continuation_enter_setup(masm, stack_slots);
 957     // The frame is complete here, but we only record it for the compiled entry, so the frame would appear unsafe,
 958     // but that's okay because at the very worst we'll miss an async sample, but we're in interp_only_mode anyway.
 959 
 960     fill_continuation_entry(masm);
 961 
 962     __ bnez(c_rarg2, call_thaw);
 963 
 964     // Make sure the call is patchable
 965     __ align(NativeInstruction::instruction_size);
 966 
 967     const address tr_call = __ trampoline_call(resolve);
 968     if (tr_call == nullptr) {
 969       fatal("CodeCache is full at gen_continuation_enter");
 970     }
 971 
 972     oop_maps->add_gc_map(__ pc() - start, map);
 973     __ post_call_nop();
 974 
 975     __ j(exit);
 976 
 977     CodeBuffer* cbuf = masm->code_section()->outer();
 978     address stub = CompiledDirectCall::emit_to_interp_stub(*cbuf, tr_call);
 979     if (stub == nullptr) {
 980       fatal("CodeCache is full at gen_continuation_enter");
 981     }
 982   }
 983 
 984   // compiled entry
 985   __ align(CodeEntryAlignment);
 986   compiled_entry_offset = __ pc() - start;
 987 
 988   __ enter();
 989   stack_slots = 2; // will be adjusted in setup
 990   OopMap* map = continuation_enter_setup(masm, stack_slots);
 991   frame_complete = __ pc() - start;
 992 
 993   fill_continuation_entry(masm);
 994 
 995   __ bnez(c_rarg2, call_thaw);
 996 
 997   // Make sure the call is patchable
 998   __ align(NativeInstruction::instruction_size);
 999 
1000   const address tr_call = __ trampoline_call(resolve);
1001   if (tr_call == nullptr) {
1002     fatal("CodeCache is full at gen_continuation_enter");
1003   }
1004 
1005   oop_maps->add_gc_map(__ pc() - start, map);
1006   __ post_call_nop();
1007 
1008   __ j(exit);
1009 
1010   __ bind(call_thaw);
1011 
1012   __ rt_call(CAST_FROM_FN_PTR(address, StubRoutines::cont_thaw()));
1013   oop_maps->add_gc_map(__ pc() - start, map->deep_copy());
1014   ContinuationEntry::_return_pc_offset = __ pc() - start;
1015   __ post_call_nop();
1016 
1017   __ bind(exit);
1018   continuation_enter_cleanup(masm);
1019   __ leave();
1020   __ ret();
1021 
1022   // exception handling
1023   exception_offset = __ pc() - start;
1024   {
1025     __ mv(x9, x10); // save return value contaning the exception oop in callee-saved x9
1026 
1027     continuation_enter_cleanup(masm);
1028 
1029     __ ld(c_rarg1, Address(fp, -1 * wordSize)); // return address
1030     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), xthread, c_rarg1);
1031 
1032     // see OptoRuntime::generate_exception_blob: x10 -- exception oop, x13 -- exception pc
1033 
1034     __ mv(x11, x10); // the exception handler
1035     __ mv(x10, x9); // restore return value contaning the exception oop
1036     __ verify_oop(x10);
1037 
1038     __ leave();
1039     __ mv(x13, ra);
1040     __ jr(x11); // the exception handler
1041   }
1042 
1043   CodeBuffer* cbuf = masm->code_section()->outer();
1044   address stub = CompiledDirectCall::emit_to_interp_stub(*cbuf, tr_call);
1045   if (stub == nullptr) {
1046     fatal("CodeCache is full at gen_continuation_enter");
1047   }
1048 }
1049 
1050 static void gen_continuation_yield(MacroAssembler* masm,
1051                                    const methodHandle& method,
1052                                    const BasicType* sig_bt,
1053                                    const VMRegPair* regs,
1054                                    OopMapSet* oop_maps,
1055                                    int& frame_complete,
1056                                    int& stack_slots,
1057                                    int& compiled_entry_offset) {
1058   enum layout {
1059     fp_off,
1060     fp_off2,
1061     return_off,
1062     return_off2,
1063     framesize // inclusive of return address
1064   };
1065   // assert(is_even(framesize/2), "sp not 16-byte aligned");
1066 
1067   stack_slots = framesize / VMRegImpl::slots_per_word;
1068   assert(stack_slots == 2, "recheck layout");
1069 
1070   address start = __ pc();
1071 
1072   compiled_entry_offset = __ pc() - start;
1073   __ enter();
1074 
1075   __ mv(c_rarg1, sp);
1076 
1077   frame_complete = __ pc() - start;
1078   address the_pc = __ pc();
1079 
1080   __ post_call_nop(); // this must be exactly after the pc value that is pushed into the frame info, we use this nop for fast CodeBlob lookup
1081 
1082   __ mv(c_rarg0, xthread);
1083   __ set_last_Java_frame(sp, fp, the_pc, t0);
1084   __ call_VM_leaf(Continuation::freeze_entry(), 2);
1085   __ reset_last_Java_frame(true);
1086 
1087   Label pinned;
1088 
1089   __ bnez(x10, pinned);
1090 
1091   // We've succeeded, set sp to the ContinuationEntry
1092   __ ld(sp, Address(xthread, JavaThread::cont_entry_offset()));
1093   continuation_enter_cleanup(masm);
1094 
1095   __ bind(pinned); // pinned -- return to caller
1096 
1097   // handle pending exception thrown by freeze
1098   __ ld(t0, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1099   Label ok;
1100   __ beqz(t0, ok);
1101   __ leave();
1102   __ la(t0, RuntimeAddress(StubRoutines::forward_exception_entry()));
1103   __ jr(t0);
1104   __ bind(ok);
1105 
1106   __ leave();
1107   __ ret();
1108 
1109   OopMap* map = new OopMap(framesize, 1);
1110   oop_maps->add_gc_map(the_pc - start, map);
1111 }
1112 
1113 static void gen_special_dispatch(MacroAssembler* masm,
1114                                  const methodHandle& method,
1115                                  const BasicType* sig_bt,
1116                                  const VMRegPair* regs) {
1117   verify_oop_args(masm, method, sig_bt, regs);
1118   vmIntrinsics::ID iid = method->intrinsic_id();
1119 
1120   // Now write the args into the outgoing interpreter space
1121   bool     has_receiver   = false;
1122   Register receiver_reg   = noreg;
1123   int      member_arg_pos = -1;
1124   Register member_reg     = noreg;
1125   int      ref_kind       = MethodHandles::signature_polymorphic_intrinsic_ref_kind(iid);
1126   if (ref_kind != 0) {
1127     member_arg_pos = method->size_of_parameters() - 1;  // trailing MemberName argument
1128     member_reg = x9;  // known to be free at this point
1129     has_receiver = MethodHandles::ref_kind_has_receiver(ref_kind);
1130   } else if (iid == vmIntrinsics::_invokeBasic) {
1131     has_receiver = true;
1132   } else if (iid == vmIntrinsics::_linkToNative) {
1133     member_arg_pos = method->size_of_parameters() - 1;  // trailing NativeEntryPoint argument
1134     member_reg = x9;  // known to be free at this point
1135   } else {
1136     fatal("unexpected intrinsic id %d", vmIntrinsics::as_int(iid));
1137   }
1138 
1139   if (member_reg != noreg) {
1140     // Load the member_arg into register, if necessary.
1141     SharedRuntime::check_member_name_argument_is_last_argument(method, sig_bt, regs);
1142     VMReg r = regs[member_arg_pos].first();
1143     if (r->is_stack()) {
1144       __ ld(member_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
1145     } else {
1146       // no data motion is needed
1147       member_reg = r->as_Register();
1148     }
1149   }
1150 
1151   if (has_receiver) {
1152     // Make sure the receiver is loaded into a register.
1153     assert(method->size_of_parameters() > 0, "oob");
1154     assert(sig_bt[0] == T_OBJECT, "receiver argument must be an object");
1155     VMReg r = regs[0].first();
1156     assert(r->is_valid(), "bad receiver arg");
1157     if (r->is_stack()) {
1158       // Porting note:  This assumes that compiled calling conventions always
1159       // pass the receiver oop in a register.  If this is not true on some
1160       // platform, pick a temp and load the receiver from stack.
1161       fatal("receiver always in a register");
1162       receiver_reg = x12;  // known to be free at this point
1163       __ ld(receiver_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
1164     } else {
1165       // no data motion is needed
1166       receiver_reg = r->as_Register();
1167     }
1168   }
1169 
1170   // Figure out which address we are really jumping to:
1171   MethodHandles::generate_method_handle_dispatch(masm, iid,
1172                                                  receiver_reg, member_reg, /*for_compiler_entry:*/ true);
1173 }
1174 
1175 // ---------------------------------------------------------------------------
1176 // Generate a native wrapper for a given method.  The method takes arguments
1177 // in the Java compiled code convention, marshals them to the native
1178 // convention (handlizes oops, etc), transitions to native, makes the call,
1179 // returns to java state (possibly blocking), unhandlizes any result and
1180 // returns.
1181 //
1182 // Critical native functions are a shorthand for the use of
1183 // GetPrimtiveArrayCritical and disallow the use of any other JNI
1184 // functions.  The wrapper is expected to unpack the arguments before
1185 // passing them to the callee and perform checks before and after the
1186 // native call to ensure that they GCLocker
1187 // lock_critical/unlock_critical semantics are followed.  Some other
1188 // parts of JNI setup are skipped like the tear down of the JNI handle
1189 // block and the check for pending exceptions it's impossible for them
1190 // to be thrown.
1191 //
1192 // They are roughly structured like this:
1193 //    if (GCLocker::needs_gc()) SharedRuntime::block_for_jni_critical()
1194 //    tranistion to thread_in_native
1195 //    unpack array arguments and call native entry point
1196 //    check for safepoint in progress
1197 //    check if any thread suspend flags are set
1198 //      call into JVM and possible unlock the JNI critical
1199 //      if a GC was suppressed while in the critical native.
1200 //    transition back to thread_in_Java
1201 //    return to caller
1202 //
1203 nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
1204                                                 const methodHandle& method,
1205                                                 int compile_id,
1206                                                 BasicType* in_sig_bt,
1207                                                 VMRegPair* in_regs,
1208                                                 BasicType ret_type) {
1209   if (method->is_continuation_native_intrinsic()) {
1210     int exception_offset = -1;
1211     OopMapSet* oop_maps = new OopMapSet();
1212     int frame_complete = -1;
1213     int stack_slots = -1;
1214     int interpreted_entry_offset = -1;
1215     int vep_offset = -1;
1216     if (method->is_continuation_enter_intrinsic()) {
1217       gen_continuation_enter(masm,
1218                              method,
1219                              in_sig_bt,
1220                              in_regs,
1221                              exception_offset,
1222                              oop_maps,
1223                              frame_complete,
1224                              stack_slots,
1225                              interpreted_entry_offset,
1226                              vep_offset);
1227     } else if (method->is_continuation_yield_intrinsic()) {
1228       gen_continuation_yield(masm,
1229                              method,
1230                              in_sig_bt,
1231                              in_regs,
1232                              oop_maps,
1233                              frame_complete,
1234                              stack_slots,
1235                              vep_offset);
1236     } else {
1237       guarantee(false, "Unknown Continuation native intrinsic");
1238     }
1239 
1240 #ifdef ASSERT
1241     if (method->is_continuation_enter_intrinsic()) {
1242       assert(interpreted_entry_offset != -1, "Must be set");
1243       assert(exception_offset != -1,         "Must be set");
1244     } else {
1245       assert(interpreted_entry_offset == -1, "Must be unset");
1246       assert(exception_offset == -1,         "Must be unset");
1247     }
1248     assert(frame_complete != -1,    "Must be set");
1249     assert(stack_slots != -1,       "Must be set");
1250     assert(vep_offset != -1,        "Must be set");
1251 #endif
1252 
1253     __ flush();
1254     nmethod* nm = nmethod::new_native_nmethod(method,
1255                                               compile_id,
1256                                               masm->code(),
1257                                               vep_offset,
1258                                               frame_complete,
1259                                               stack_slots,
1260                                               in_ByteSize(-1),
1261                                               in_ByteSize(-1),
1262                                               oop_maps,
1263                                               exception_offset);
1264     if (nm == nullptr) return nm;
1265     if (method->is_continuation_enter_intrinsic()) {
1266       ContinuationEntry::set_enter_code(nm, interpreted_entry_offset);
1267     } else if (method->is_continuation_yield_intrinsic()) {
1268       _cont_doYield_stub = nm;
1269     } else {
1270       guarantee(false, "Unknown Continuation native intrinsic");
1271     }
1272     return nm;
1273   }
1274 
1275   if (method->is_method_handle_intrinsic()) {
1276     vmIntrinsics::ID iid = method->intrinsic_id();
1277     intptr_t start = (intptr_t)__ pc();
1278     int vep_offset = ((intptr_t)__ pc()) - start;
1279 
1280     // First instruction must be a nop as it may need to be patched on deoptimisation
1281     {
1282       Assembler::IncompressibleRegion ir(masm);  // keep the nop as 4 bytes for patching.
1283       MacroAssembler::assert_alignment(__ pc());
1284       __ nop();  // 4 bytes
1285     }
1286     gen_special_dispatch(masm,
1287                          method,
1288                          in_sig_bt,
1289                          in_regs);
1290     int frame_complete = ((intptr_t)__ pc()) - start;  // not complete, period
1291     __ flush();
1292     int stack_slots = SharedRuntime::out_preserve_stack_slots();  // no out slots at all, actually
1293     return nmethod::new_native_nmethod(method,
1294                                        compile_id,
1295                                        masm->code(),
1296                                        vep_offset,
1297                                        frame_complete,
1298                                        stack_slots / VMRegImpl::slots_per_word,
1299                                        in_ByteSize(-1),
1300                                        in_ByteSize(-1),
1301                                        (OopMapSet*)nullptr);
1302   }
1303   address native_func = method->native_function();
1304   assert(native_func != nullptr, "must have function");
1305 
1306   // An OopMap for lock (and class if static)
1307   OopMapSet *oop_maps = new OopMapSet();
1308   assert_cond(oop_maps != nullptr);
1309   intptr_t start = (intptr_t)__ pc();
1310 
1311   // We have received a description of where all the java arg are located
1312   // on entry to the wrapper. We need to convert these args to where
1313   // the jni function will expect them. To figure out where they go
1314   // we convert the java signature to a C signature by inserting
1315   // the hidden arguments as arg[0] and possibly arg[1] (static method)
1316 
1317   const int total_in_args = method->size_of_parameters();
1318   int total_c_args = total_in_args + (method->is_static() ? 2 : 1);
1319 
1320   BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args);
1321   VMRegPair* out_regs   = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args);
1322   BasicType* in_elem_bt = nullptr;
1323 
1324   int argc = 0;
1325   out_sig_bt[argc++] = T_ADDRESS;
1326   if (method->is_static()) {
1327     out_sig_bt[argc++] = T_OBJECT;
1328   }
1329 
1330   for (int i = 0; i < total_in_args ; i++) {
1331     out_sig_bt[argc++] = in_sig_bt[i];
1332   }
1333 
1334   // Now figure out where the args must be stored and how much stack space
1335   // they require.
1336   int out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args);
1337 
1338   // Compute framesize for the wrapper.  We need to handlize all oops in
1339   // incoming registers
1340 
1341   // Calculate the total number of stack slots we will need.
1342 
1343   // First count the abi requirement plus all of the outgoing args
1344   int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
1345 
1346   // Now the space for the inbound oop handle area
1347   int total_save_slots = 8 * VMRegImpl::slots_per_word;  // 8 arguments passed in registers
1348 
1349   int oop_handle_offset = stack_slots;
1350   stack_slots += total_save_slots;
1351 
1352   // Now any space we need for handlizing a klass if static method
1353 
1354   int klass_slot_offset = 0;
1355   int klass_offset = -1;
1356   int lock_slot_offset = 0;
1357   bool is_static = false;
1358 
1359   if (method->is_static()) {
1360     klass_slot_offset = stack_slots;
1361     stack_slots += VMRegImpl::slots_per_word;
1362     klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size;
1363     is_static = true;
1364   }
1365 
1366   // Plus a lock if needed
1367 
1368   if (method->is_synchronized()) {
1369     lock_slot_offset = stack_slots;
1370     stack_slots += VMRegImpl::slots_per_word;
1371   }
1372 
1373   // Now a place (+2) to save return values or temp during shuffling
1374   // + 4 for return address (which we own) and saved fp
1375   stack_slots += 6;
1376 
1377   // Ok The space we have allocated will look like:
1378   //
1379   //
1380   // FP-> |                     |
1381   //      | 2 slots (ra)        |
1382   //      | 2 slots (fp)        |
1383   //      |---------------------|
1384   //      | 2 slots for moves   |
1385   //      |---------------------|
1386   //      | lock box (if sync)  |
1387   //      |---------------------| <- lock_slot_offset
1388   //      | klass (if static)   |
1389   //      |---------------------| <- klass_slot_offset
1390   //      | oopHandle area      |
1391   //      |---------------------| <- oop_handle_offset (8 java arg registers)
1392   //      | outbound memory     |
1393   //      | based arguments     |
1394   //      |                     |
1395   //      |---------------------|
1396   //      |                     |
1397   // SP-> | out_preserved_slots |
1398   //
1399   //
1400 
1401 
1402   // Now compute actual number of stack words we need rounding to make
1403   // stack properly aligned.
1404   stack_slots = align_up(stack_slots, StackAlignmentInSlots);
1405 
1406   int stack_size = stack_slots * VMRegImpl::stack_slot_size;
1407 
1408   // First thing make an ic check to see if we should even be here
1409 
1410   // We are free to use all registers as temps without saving them and
1411   // restoring them except fp. fp is the only callee save register
1412   // as far as the interpreter and the compiler(s) are concerned.
1413 
1414 
1415   const Register ic_reg = t1;
1416   const Register receiver = j_rarg0;
1417 
1418   __ verify_oop(receiver);
1419   assert_different_registers(receiver, t0, t1);
1420 
1421   __ ic_check();
1422 
1423   int vep_offset = ((intptr_t)__ pc()) - start;
1424 
1425   // If we have to make this method not-entrant we'll overwrite its
1426   // first instruction with a jump.
1427   {
1428     Assembler::IncompressibleRegion ir(masm);  // keep the nop as 4 bytes for patching.
1429     MacroAssembler::assert_alignment(__ pc());
1430     __ nop();  // 4 bytes
1431   }
1432 
1433   if (VM_Version::supports_fast_class_init_checks() && method->needs_clinit_barrier()) {
1434     Label L_skip_barrier;
1435     __ mov_metadata(t1, method->method_holder()); // InstanceKlass*
1436     __ clinit_barrier(t1, t0, &L_skip_barrier);
1437     __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
1438 
1439     __ bind(L_skip_barrier);
1440   }
1441 
1442   // Generate stack overflow check
1443   __ bang_stack_with_offset(checked_cast<int>(StackOverflow::stack_shadow_zone_size()));
1444 
1445   // Generate a new frame for the wrapper.
1446   __ enter();
1447   // -2 because return address is already present and so is saved fp
1448   __ sub(sp, sp, stack_size - 2 * wordSize);
1449 
1450   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
1451   assert_cond(bs != nullptr);
1452   bs->nmethod_entry_barrier(masm, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
1453 
1454   // Frame is now completed as far as size and linkage.
1455   int frame_complete = ((intptr_t)__ pc()) - start;
1456 
1457   // We use x18 as the oop handle for the receiver/klass
1458   // It is callee save so it survives the call to native
1459 
1460   const Register oop_handle_reg = x18;
1461 
1462   //
1463   // We immediately shuffle the arguments so that any vm call we have to
1464   // make from here on out (sync slow path, jvmti, etc.) we will have
1465   // captured the oops from our caller and have a valid oopMap for
1466   // them.
1467 
1468   // -----------------
1469   // The Grand Shuffle
1470 
1471   // The Java calling convention is either equal (linux) or denser (win64) than the
1472   // c calling convention. However the because of the jni_env argument the c calling
1473   // convention always has at least one more (and two for static) arguments than Java.
1474   // Therefore if we move the args from java -> c backwards then we will never have
1475   // a register->register conflict and we don't have to build a dependency graph
1476   // and figure out how to break any cycles.
1477   //
1478 
1479   // Record esp-based slot for receiver on stack for non-static methods
1480   int receiver_offset = -1;
1481 
1482   // This is a trick. We double the stack slots so we can claim
1483   // the oops in the caller's frame. Since we are sure to have
1484   // more args than the caller doubling is enough to make
1485   // sure we can capture all the incoming oop args from the
1486   // caller.
1487   //
1488   OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1489   assert_cond(map != nullptr);
1490 
1491   int float_args = 0;
1492   int int_args = 0;
1493 
1494 #ifdef ASSERT
1495   bool reg_destroyed[Register::number_of_registers];
1496   bool freg_destroyed[FloatRegister::number_of_registers];
1497   for ( int r = 0 ; r < Register::number_of_registers ; r++ ) {
1498     reg_destroyed[r] = false;
1499   }
1500   for ( int f = 0 ; f < FloatRegister::number_of_registers ; f++ ) {
1501     freg_destroyed[f] = false;
1502   }
1503 
1504 #endif /* ASSERT */
1505 
1506   // For JNI natives the incoming and outgoing registers are offset upwards.
1507   GrowableArray<int> arg_order(2 * total_in_args);
1508   VMRegPair tmp_vmreg;
1509   tmp_vmreg.set2(x9->as_VMReg());
1510 
1511   for (int i = total_in_args - 1, c_arg = total_c_args - 1; i >= 0; i--, c_arg--) {
1512     arg_order.push(i);
1513     arg_order.push(c_arg);
1514   }
1515 
1516   int temploc = -1;
1517   for (int ai = 0; ai < arg_order.length(); ai += 2) {
1518     int i = arg_order.at(ai);
1519     int c_arg = arg_order.at(ai + 1);
1520     __ block_comment(err_msg("mv %d -> %d", i, c_arg));
1521     assert(c_arg != -1 && i != -1, "wrong order");
1522 #ifdef ASSERT
1523     if (in_regs[i].first()->is_Register()) {
1524       assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!");
1525     } else if (in_regs[i].first()->is_FloatRegister()) {
1526       assert(!freg_destroyed[in_regs[i].first()->as_FloatRegister()->encoding()], "destroyed reg!");
1527     }
1528     if (out_regs[c_arg].first()->is_Register()) {
1529       reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
1530     } else if (out_regs[c_arg].first()->is_FloatRegister()) {
1531       freg_destroyed[out_regs[c_arg].first()->as_FloatRegister()->encoding()] = true;
1532     }
1533 #endif /* ASSERT */
1534     switch (in_sig_bt[i]) {
1535       case T_ARRAY:
1536       case T_OBJECT:
1537         __ object_move(map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
1538                        ((i == 0) && (!is_static)),
1539                        &receiver_offset);
1540         int_args++;
1541         break;
1542       case T_VOID:
1543         break;
1544 
1545       case T_FLOAT:
1546         __ float_move(in_regs[i], out_regs[c_arg]);
1547         float_args++;
1548         break;
1549 
1550       case T_DOUBLE:
1551         assert( i + 1 < total_in_args &&
1552                 in_sig_bt[i + 1] == T_VOID &&
1553                 out_sig_bt[c_arg + 1] == T_VOID, "bad arg list");
1554         __ double_move(in_regs[i], out_regs[c_arg]);
1555         float_args++;
1556         break;
1557 
1558       case T_LONG :
1559         __ long_move(in_regs[i], out_regs[c_arg]);
1560         int_args++;
1561         break;
1562 
1563       case T_ADDRESS:
1564         assert(false, "found T_ADDRESS in java args");
1565         break;
1566 
1567       default:
1568         __ move32_64(in_regs[i], out_regs[c_arg]);
1569         int_args++;
1570     }
1571   }
1572 
1573   // point c_arg at the first arg that is already loaded in case we
1574   // need to spill before we call out
1575   int c_arg = total_c_args - total_in_args;
1576 
1577   // Pre-load a static method's oop into c_rarg1.
1578   if (method->is_static()) {
1579 
1580     //  load oop into a register
1581     __ movoop(c_rarg1,
1582               JNIHandles::make_local(method->method_holder()->java_mirror()));
1583 
1584     // Now handlize the static class mirror it's known not-null.
1585     __ sd(c_rarg1, Address(sp, klass_offset));
1586     map->set_oop(VMRegImpl::stack2reg(klass_slot_offset));
1587 
1588     // Now get the handle
1589     __ la(c_rarg1, Address(sp, klass_offset));
1590     // and protect the arg if we must spill
1591     c_arg--;
1592   }
1593 
1594   // Change state to native (we save the return address in the thread, since it might not
1595   // be pushed on the stack when we do a stack traversal).
1596   // We use the same pc/oopMap repeatedly when we call out
1597 
1598   Label native_return;
1599   __ set_last_Java_frame(sp, noreg, native_return, t0);
1600 
1601   Label dtrace_method_entry, dtrace_method_entry_done;
1602   {
1603     ExternalAddress target((address)&DTraceMethodProbes);
1604     __ relocate(target.rspec(), [&] {
1605       int32_t offset;
1606       __ la(t0, target.target(), offset);
1607       __ lbu(t0, Address(t0, offset));
1608     });
1609     __ bnez(t0, dtrace_method_entry);
1610     __ bind(dtrace_method_entry_done);
1611   }
1612 
1613   // RedefineClasses() tracing support for obsolete method entry
1614   if (log_is_enabled(Trace, redefine, class, obsolete)) {
1615     // protect the args we've loaded
1616     save_args(masm, total_c_args, c_arg, out_regs);
1617     __ mov_metadata(c_rarg1, method());
1618     __ call_VM_leaf(
1619       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1620       xthread, c_rarg1);
1621     restore_args(masm, total_c_args, c_arg, out_regs);
1622   }
1623 
1624   // Lock a synchronized method
1625 
1626   // Register definitions used by locking and unlocking
1627 
1628   const Register swap_reg = x10;
1629   const Register obj_reg  = x9;  // Will contain the oop
1630   const Register lock_reg = x30;  // Address of compiler lock object (BasicLock)
1631   const Register old_hdr  = x30;  // value of old header at unlock time
1632   const Register lock_tmp = x31;  // Temporary used by lightweight_lock/unlock
1633   const Register tmp      = ra;
1634 
1635   Label slow_path_lock;
1636   Label lock_done;
1637 
1638   if (method->is_synchronized()) {
1639     Label count;
1640 
1641     const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes();
1642 
1643     // Get the handle (the 2nd argument)
1644     __ mv(oop_handle_reg, c_rarg1);
1645 
1646     // Get address of the box
1647 
1648     __ la(lock_reg, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1649 
1650     // Load the oop from the handle
1651     __ ld(obj_reg, Address(oop_handle_reg, 0));
1652 
1653     if (LockingMode == LM_MONITOR) {
1654       __ j(slow_path_lock);
1655     } else if (LockingMode == LM_LEGACY) {
1656       // Load (object->mark() | 1) into swap_reg % x10
1657       __ ld(t0, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1658       __ ori(swap_reg, t0, 1);
1659 
1660       // Save (object->mark() | 1) into BasicLock's displaced header
1661       __ sd(swap_reg, Address(lock_reg, mark_word_offset));
1662 
1663       // src -> dest if dest == x10 else x10 <- dest
1664       __ cmpxchg_obj_header(x10, lock_reg, obj_reg, lock_tmp, count, /*fallthrough*/nullptr);
1665 
1666       // Test if the oopMark is an obvious stack pointer, i.e.,
1667       //  1) (mark & 3) == 0, and
1668       //  2) sp <= mark < mark + os::pagesize()
1669       // These 3 tests can be done by evaluating the following
1670       // expression: ((mark - sp) & (3 - os::vm_page_size())),
1671       // assuming both stack pointer and pagesize have their
1672       // least significant 2 bits clear.
1673       // NOTE: the oopMark is in swap_reg % 10 as the result of cmpxchg
1674 
1675       __ sub(swap_reg, swap_reg, sp);
1676       __ andi(swap_reg, swap_reg, 3 - (int)os::vm_page_size());
1677 
1678       // Save the test result, for recursive case, the result is zero
1679       __ sd(swap_reg, Address(lock_reg, mark_word_offset));
1680       __ bnez(swap_reg, slow_path_lock);
1681     } else {
1682       assert(LockingMode == LM_LIGHTWEIGHT, "");
1683       __ ld(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1684       __ lightweight_lock(obj_reg, swap_reg, tmp, lock_tmp, slow_path_lock);
1685     }
1686 
1687     __ bind(count);
1688     __ increment(Address(xthread, JavaThread::held_monitor_count_offset()));
1689 
1690     // Slow path will re-enter here
1691     __ bind(lock_done);
1692   }
1693 
1694 
1695   // Finally just about ready to make the JNI call
1696 
1697   // get JNIEnv* which is first argument to native
1698   __ la(c_rarg0, Address(xthread, in_bytes(JavaThread::jni_environment_offset())));
1699 
1700   // Now set thread in native
1701   __ la(t1, Address(xthread, JavaThread::thread_state_offset()));
1702   __ mv(t0, _thread_in_native);
1703   __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1704   __ sw(t0, Address(t1));
1705 
1706   __ rt_call(native_func);
1707 
1708   __ bind(native_return);
1709 
1710   intptr_t return_pc = (intptr_t) __ pc();
1711   oop_maps->add_gc_map(return_pc - start, map);
1712 
1713   // Unpack native results.
1714   if (ret_type != T_OBJECT && ret_type != T_ARRAY) {
1715     __ cast_primitive_type(ret_type, x10);
1716   }
1717 
1718   Label safepoint_in_progress, safepoint_in_progress_done;
1719   Label after_transition;
1720 
1721   // Switch thread to "native transition" state before reading the synchronization state.
1722   // This additional state is necessary because reading and testing the synchronization
1723   // state is not atomic w.r.t. GC, as this scenario demonstrates:
1724   //     Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
1725   //     VM thread changes sync state to synchronizing and suspends threads for GC.
1726   //     Thread A is resumed to finish this native method, but doesn't block here since it
1727   //     didn't see any synchronization is progress, and escapes.
1728   __ mv(t0, _thread_in_native_trans);
1729 
1730   __ sw(t0, Address(xthread, JavaThread::thread_state_offset()));
1731 
1732   // Force this write out before the read below
1733   if (!UseSystemMemoryBarrier) {
1734     __ membar(MacroAssembler::AnyAny);
1735   }
1736 
1737   // check for safepoint operation in progress and/or pending suspend requests
1738   {
1739     // We need an acquire here to ensure that any subsequent load of the
1740     // global SafepointSynchronize::_state flag is ordered after this load
1741     // of the thread-local polling word. We don't want this poll to
1742     // return false (i.e. not safepointing) and a later poll of the global
1743     // SafepointSynchronize::_state spuriously to return true.
1744     // This is to avoid a race when we're in a native->Java transition
1745     // racing the code which wakes up from a safepoint.
1746 
1747     __ safepoint_poll(safepoint_in_progress, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1748     __ lwu(t0, Address(xthread, JavaThread::suspend_flags_offset()));
1749     __ bnez(t0, safepoint_in_progress);
1750     __ bind(safepoint_in_progress_done);
1751   }
1752 
1753   // change thread state
1754   __ la(t1, Address(xthread, JavaThread::thread_state_offset()));
1755   __ mv(t0, _thread_in_Java);
1756   __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1757   __ sw(t0, Address(t1));
1758   __ bind(after_transition);
1759 
1760   Label reguard;
1761   Label reguard_done;
1762   __ lbu(t0, Address(xthread, JavaThread::stack_guard_state_offset()));
1763   __ mv(t1, StackOverflow::stack_guard_yellow_reserved_disabled);
1764   __ beq(t0, t1, reguard);
1765   __ bind(reguard_done);
1766 
1767   // native result if any is live
1768 
1769   // Unlock
1770   Label unlock_done;
1771   Label slow_path_unlock;
1772   if (method->is_synchronized()) {
1773 
1774     // Get locked oop from the handle we passed to jni
1775     __ ld(obj_reg, Address(oop_handle_reg, 0));
1776 
1777     Label done, not_recursive;
1778 
1779     if (LockingMode == LM_LEGACY) {
1780       // Simple recursive lock?
1781       __ ld(t0, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1782       __ bnez(t0, not_recursive);
1783       __ decrement(Address(xthread, JavaThread::held_monitor_count_offset()));
1784       __ j(done);
1785     }
1786 
1787     __ bind(not_recursive);
1788 
1789     // Must save x10 if if it is live now because cmpxchg must use it
1790     if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1791       save_native_result(masm, ret_type, stack_slots);
1792     }
1793 
1794     if (LockingMode == LM_MONITOR) {
1795       __ j(slow_path_unlock);
1796     } else if (LockingMode == LM_LEGACY) {
1797       // get address of the stack lock
1798       __ la(x10, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1799       //  get old displaced header
1800       __ ld(old_hdr, Address(x10, 0));
1801 
1802       // Atomic swap old header if oop still contains the stack lock
1803       Label count;
1804       __ cmpxchg_obj_header(x10, old_hdr, obj_reg, lock_tmp, count, &slow_path_unlock);
1805       __ bind(count);
1806       __ decrement(Address(xthread, JavaThread::held_monitor_count_offset()));
1807     } else {
1808       assert(LockingMode == LM_LIGHTWEIGHT, "");
1809       __ ld(old_hdr, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1810       __ test_bit(t0, old_hdr, exact_log2(markWord::monitor_value));
1811       __ bnez(t0, slow_path_unlock);
1812       __ lightweight_unlock(obj_reg, old_hdr, swap_reg, lock_tmp, slow_path_unlock);
1813       __ decrement(Address(xthread, JavaThread::held_monitor_count_offset()));
1814     }
1815 
1816     // slow path re-enters here
1817     __ bind(unlock_done);
1818     if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1819       restore_native_result(masm, ret_type, stack_slots);
1820     }
1821 
1822     __ bind(done);
1823   }
1824 
1825   Label dtrace_method_exit, dtrace_method_exit_done;
1826   {
1827     ExternalAddress target((address)&DTraceMethodProbes);
1828     __ relocate(target.rspec(), [&] {
1829       int32_t offset;
1830       __ la(t0, target.target(), offset);
1831       __ lbu(t0, Address(t0, offset));
1832     });
1833     __ bnez(t0, dtrace_method_exit);
1834     __ bind(dtrace_method_exit_done);
1835   }
1836 
1837   __ reset_last_Java_frame(false);
1838 
1839   // Unbox oop result, e.g. JNIHandles::resolve result.
1840   if (is_reference_type(ret_type)) {
1841     __ resolve_jobject(x10, x11, x12);
1842   }
1843 
1844   if (CheckJNICalls) {
1845     // clear_pending_jni_exception_check
1846     __ sd(zr, Address(xthread, JavaThread::pending_jni_exception_check_fn_offset()));
1847   }
1848 
1849   // reset handle block
1850   __ ld(x12, Address(xthread, JavaThread::active_handles_offset()));
1851   __ sd(zr, Address(x12, JNIHandleBlock::top_offset()));
1852 
1853   __ leave();
1854 
1855   // Any exception pending?
1856   Label exception_pending;
1857   __ ld(t0, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1858   __ bnez(t0, exception_pending);
1859 
1860   // We're done
1861   __ ret();
1862 
1863   // Unexpected paths are out of line and go here
1864 
1865   // forward the exception
1866   __ bind(exception_pending);
1867 
1868   // and forward the exception
1869   __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1870 
1871   // Slow path locking & unlocking
1872   if (method->is_synchronized()) {
1873 
1874     __ block_comment("Slow path lock {");
1875     __ bind(slow_path_lock);
1876 
1877     // has last_Java_frame setup. No exceptions so do vanilla call not call_VM
1878     // args are (oop obj, BasicLock* lock, JavaThread* thread)
1879 
1880     // protect the args we've loaded
1881     save_args(masm, total_c_args, c_arg, out_regs);
1882 
1883     __ mv(c_rarg0, obj_reg);
1884     __ mv(c_rarg1, lock_reg);
1885     __ mv(c_rarg2, xthread);
1886 
1887     // Not a leaf but we have last_Java_frame setup as we want
1888     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), 3);
1889     restore_args(masm, total_c_args, c_arg, out_regs);
1890 
1891 #ifdef ASSERT
1892     { Label L;
1893       __ ld(t0, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1894       __ beqz(t0, L);
1895       __ stop("no pending exception allowed on exit from monitorenter");
1896       __ bind(L);
1897     }
1898 #endif
1899     __ j(lock_done);
1900 
1901     __ block_comment("} Slow path lock");
1902 
1903     __ block_comment("Slow path unlock {");
1904     __ bind(slow_path_unlock);
1905 
1906     if (ret_type == T_FLOAT || ret_type == T_DOUBLE) {
1907       save_native_result(masm, ret_type, stack_slots);
1908     }
1909 
1910     __ mv(c_rarg2, xthread);
1911     __ la(c_rarg1, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1912     __ mv(c_rarg0, obj_reg);
1913 
1914     // Save pending exception around call to VM (which contains an EXCEPTION_MARK)
1915     // NOTE that obj_reg == x9 currently
1916     __ ld(x9, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1917     __ sd(zr, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1918 
1919     __ rt_call(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C));
1920 
1921 #ifdef ASSERT
1922     {
1923       Label L;
1924       __ ld(t0, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1925       __ beqz(t0, L);
1926       __ stop("no pending exception allowed on exit complete_monitor_unlocking_C");
1927       __ bind(L);
1928     }
1929 #endif /* ASSERT */
1930 
1931     __ sd(x9, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1932 
1933     if (ret_type == T_FLOAT || ret_type == T_DOUBLE) {
1934       restore_native_result(masm, ret_type, stack_slots);
1935     }
1936     __ j(unlock_done);
1937 
1938     __ block_comment("} Slow path unlock");
1939 
1940   } // synchronized
1941 
1942   // SLOW PATH Reguard the stack if needed
1943 
1944   __ bind(reguard);
1945   save_native_result(masm, ret_type, stack_slots);
1946   __ rt_call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1947   restore_native_result(masm, ret_type, stack_slots);
1948   // and continue
1949   __ j(reguard_done);
1950 
1951   // SLOW PATH safepoint
1952   {
1953     __ block_comment("safepoint {");
1954     __ bind(safepoint_in_progress);
1955 
1956     // Don't use call_VM as it will see a possible pending exception and forward it
1957     // and never return here preventing us from clearing _last_native_pc down below.
1958     //
1959     save_native_result(masm, ret_type, stack_slots);
1960     __ mv(c_rarg0, xthread);
1961 #ifndef PRODUCT
1962     assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
1963 #endif
1964     __ rt_call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1965 
1966     // Restore any method result value
1967     restore_native_result(masm, ret_type, stack_slots);
1968 
1969     __ j(safepoint_in_progress_done);
1970     __ block_comment("} safepoint");
1971   }
1972 
1973   // SLOW PATH dtrace support
1974   {
1975     __ block_comment("dtrace entry {");
1976     __ bind(dtrace_method_entry);
1977 
1978     // We have all of the arguments setup at this point. We must not touch any register
1979     // argument registers at this point (what if we save/restore them there are no oop?
1980 
1981     save_args(masm, total_c_args, c_arg, out_regs);
1982     __ mov_metadata(c_rarg1, method());
1983     __ call_VM_leaf(
1984       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1985       xthread, c_rarg1);
1986     restore_args(masm, total_c_args, c_arg, out_regs);
1987     __ j(dtrace_method_entry_done);
1988     __ block_comment("} dtrace entry");
1989   }
1990 
1991   {
1992     __ block_comment("dtrace exit {");
1993     __ bind(dtrace_method_exit);
1994     save_native_result(masm, ret_type, stack_slots);
1995     __ mov_metadata(c_rarg1, method());
1996     __ call_VM_leaf(
1997          CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1998          xthread, c_rarg1);
1999     restore_native_result(masm, ret_type, stack_slots);
2000     __ j(dtrace_method_exit_done);
2001     __ block_comment("} dtrace exit");
2002   }
2003 
2004   __ flush();
2005 
2006   nmethod *nm = nmethod::new_native_nmethod(method,
2007                                             compile_id,
2008                                             masm->code(),
2009                                             vep_offset,
2010                                             frame_complete,
2011                                             stack_slots / VMRegImpl::slots_per_word,
2012                                             (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)),
2013                                             in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size),
2014                                             oop_maps);
2015   assert(nm != nullptr, "create native nmethod fail!");
2016   return nm;
2017 }
2018 
2019 // this function returns the adjust size (in number of words) to a c2i adapter
2020 // activation for use during deoptimization
2021 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals) {
2022   assert(callee_locals >= callee_parameters,
2023          "test and remove; got more parms than locals");
2024   if (callee_locals < callee_parameters) {
2025     return 0;                   // No adjustment for negative locals
2026   }
2027   int diff = (callee_locals - callee_parameters) * Interpreter::stackElementWords;
2028   // diff is counted in stack words
2029   return align_up(diff, 2);
2030 }
2031 
2032 //------------------------------generate_deopt_blob----------------------------
2033 void SharedRuntime::generate_deopt_blob() {
2034   // Allocate space for the code
2035   ResourceMark rm;
2036   // Setup code generation tools
2037   int pad = 0;
2038 #if INCLUDE_JVMCI
2039   if (EnableJVMCI) {
2040     pad += 512; // Increase the buffer size when compiling for JVMCI
2041   }
2042 #endif
2043   CodeBuffer buffer("deopt_blob", 2048 + pad, 1024);
2044   MacroAssembler* masm = new MacroAssembler(&buffer);
2045   int frame_size_in_words = -1;
2046   OopMap* map = nullptr;
2047   OopMapSet *oop_maps = new OopMapSet();
2048   assert_cond(masm != nullptr && oop_maps != nullptr);
2049   RegisterSaver reg_saver(COMPILER2_OR_JVMCI != 0);
2050 
2051   // -------------
2052   // This code enters when returning to a de-optimized nmethod.  A return
2053   // address has been pushed on the stack, and return values are in
2054   // registers.
2055   // If we are doing a normal deopt then we were called from the patched
2056   // nmethod from the point we returned to the nmethod. So the return
2057   // address on the stack is wrong by NativeCall::instruction_size
2058   // We will adjust the value so it looks like we have the original return
2059   // address on the stack (like when we eagerly deoptimized).
2060   // In the case of an exception pending when deoptimizing, we enter
2061   // with a return address on the stack that points after the call we patched
2062   // into the exception handler. We have the following register state from,
2063   // e.g., the forward exception stub (see stubGenerator_riscv.cpp).
2064   //    x10: exception oop
2065   //    x9: exception handler
2066   //    x13: throwing pc
2067   // So in this case we simply jam x13 into the useless return address and
2068   // the stack looks just like we want.
2069   //
2070   // At this point we need to de-opt.  We save the argument return
2071   // registers.  We call the first C routine, fetch_unroll_info().  This
2072   // routine captures the return values and returns a structure which
2073   // describes the current frame size and the sizes of all replacement frames.
2074   // The current frame is compiled code and may contain many inlined
2075   // functions, each with their own JVM state.  We pop the current frame, then
2076   // push all the new frames.  Then we call the C routine unpack_frames() to
2077   // populate these frames.  Finally unpack_frames() returns us the new target
2078   // address.  Notice that callee-save registers are BLOWN here; they have
2079   // already been captured in the vframeArray at the time the return PC was
2080   // patched.
2081   address start = __ pc();
2082   Label cont;
2083 
2084   // Prolog for non exception case!
2085 
2086   // Save everything in sight.
2087   map = reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2088 
2089   // Normal deoptimization.  Save exec mode for unpack_frames.
2090   __ mv(xcpool, Deoptimization::Unpack_deopt); // callee-saved
2091   __ j(cont);
2092 
2093   int reexecute_offset = __ pc() - start;
2094 #if INCLUDE_JVMCI && !defined(COMPILER1)
2095   if (EnableJVMCI && UseJVMCICompiler) {
2096     // JVMCI does not use this kind of deoptimization
2097     __ should_not_reach_here();
2098   }
2099 #endif
2100 
2101   // Reexecute case
2102   // return address is the pc describes what bci to do re-execute at
2103 
2104   // No need to update map as each call to save_live_registers will produce identical oopmap
2105   (void) reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2106 
2107   __ mv(xcpool, Deoptimization::Unpack_reexecute); // callee-saved
2108   __ j(cont);
2109 
2110 #if INCLUDE_JVMCI
2111   Label after_fetch_unroll_info_call;
2112   int implicit_exception_uncommon_trap_offset = 0;
2113   int uncommon_trap_offset = 0;
2114 
2115   if (EnableJVMCI) {
2116     implicit_exception_uncommon_trap_offset = __ pc() - start;
2117 
2118     __ ld(ra, Address(xthread, in_bytes(JavaThread::jvmci_implicit_exception_pc_offset())));
2119     __ sd(zr, Address(xthread, in_bytes(JavaThread::jvmci_implicit_exception_pc_offset())));
2120 
2121     uncommon_trap_offset = __ pc() - start;
2122 
2123     // Save everything in sight.
2124     reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2125     // fetch_unroll_info needs to call last_java_frame()
2126     Label retaddr;
2127     __ set_last_Java_frame(sp, noreg, retaddr, t0);
2128 
2129     __ lw(c_rarg1, Address(xthread, in_bytes(JavaThread::pending_deoptimization_offset())));
2130     __ mv(t0, -1);
2131     __ sw(t0, Address(xthread, in_bytes(JavaThread::pending_deoptimization_offset())));
2132 
2133     __ mv(xcpool, Deoptimization::Unpack_reexecute);
2134     __ mv(c_rarg0, xthread);
2135     __ orrw(c_rarg2, zr, xcpool); // exec mode
2136     __ rt_call(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap));
2137     __ bind(retaddr);
2138     oop_maps->add_gc_map( __ pc()-start, map->deep_copy());
2139 
2140     __ reset_last_Java_frame(false);
2141 
2142     __ j(after_fetch_unroll_info_call);
2143   } // EnableJVMCI
2144 #endif // INCLUDE_JVMCI
2145 
2146   int exception_offset = __ pc() - start;
2147 
2148   // Prolog for exception case
2149 
2150   // all registers are dead at this entry point, except for x10, and
2151   // x13 which contain the exception oop and exception pc
2152   // respectively.  Set them in TLS and fall thru to the
2153   // unpack_with_exception_in_tls entry point.
2154 
2155   __ sd(x13, Address(xthread, JavaThread::exception_pc_offset()));
2156   __ sd(x10, Address(xthread, JavaThread::exception_oop_offset()));
2157 
2158   int exception_in_tls_offset = __ pc() - start;
2159 
2160   // new implementation because exception oop is now passed in JavaThread
2161 
2162   // Prolog for exception case
2163   // All registers must be preserved because they might be used by LinearScan
2164   // Exceptiop oop and throwing PC are passed in JavaThread
2165   // tos: stack at point of call to method that threw the exception (i.e. only
2166   // args are on the stack, no return address)
2167 
2168   // The return address pushed by save_live_registers will be patched
2169   // later with the throwing pc. The correct value is not available
2170   // now because loading it from memory would destroy registers.
2171 
2172   // NB: The SP at this point must be the SP of the method that is
2173   // being deoptimized.  Deoptimization assumes that the frame created
2174   // here by save_live_registers is immediately below the method's SP.
2175   // This is a somewhat fragile mechanism.
2176 
2177   // Save everything in sight.
2178   map = reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2179 
2180   // Now it is safe to overwrite any register
2181 
2182   // Deopt during an exception.  Save exec mode for unpack_frames.
2183   __ mv(xcpool, Deoptimization::Unpack_exception); // callee-saved
2184 
2185   // load throwing pc from JavaThread and patch it as the return address
2186   // of the current frame. Then clear the field in JavaThread
2187 
2188   __ ld(x13, Address(xthread, JavaThread::exception_pc_offset()));
2189   __ sd(x13, Address(fp, frame::return_addr_offset * wordSize));
2190   __ sd(zr, Address(xthread, JavaThread::exception_pc_offset()));
2191 
2192 #ifdef ASSERT
2193   // verify that there is really an exception oop in JavaThread
2194   __ ld(x10, Address(xthread, JavaThread::exception_oop_offset()));
2195   __ verify_oop(x10);
2196 
2197   // verify that there is no pending exception
2198   Label no_pending_exception;
2199   __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
2200   __ beqz(t0, no_pending_exception);
2201   __ stop("must not have pending exception here");
2202   __ bind(no_pending_exception);
2203 #endif
2204 
2205   __ bind(cont);
2206 
2207   // Call C code.  Need thread and this frame, but NOT official VM entry
2208   // crud.  We cannot block on this call, no GC can happen.
2209   //
2210   // UnrollBlock* fetch_unroll_info(JavaThread* thread)
2211 
2212   // fetch_unroll_info needs to call last_java_frame().
2213 
2214   Label retaddr;
2215   __ set_last_Java_frame(sp, noreg, retaddr, t0);
2216 #ifdef ASSERT
2217   {
2218     Label L;
2219     __ ld(t0, Address(xthread,
2220                               JavaThread::last_Java_fp_offset()));
2221     __ beqz(t0, L);
2222     __ stop("SharedRuntime::generate_deopt_blob: last_Java_fp not cleared");
2223     __ bind(L);
2224   }
2225 #endif // ASSERT
2226   __ mv(c_rarg0, xthread);
2227   __ mv(c_rarg1, xcpool);
2228   __ rt_call(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info));
2229   __ bind(retaddr);
2230 
2231   // Need to have an oopmap that tells fetch_unroll_info where to
2232   // find any register it might need.
2233   oop_maps->add_gc_map(__ pc() - start, map);
2234 
2235   __ reset_last_Java_frame(false);
2236 
2237 #if INCLUDE_JVMCI
2238   if (EnableJVMCI) {
2239     __ bind(after_fetch_unroll_info_call);
2240   }
2241 #endif
2242 
2243   // Load UnrollBlock* into x15
2244   __ mv(x15, x10);
2245 
2246   __ lwu(xcpool, Address(x15, Deoptimization::UnrollBlock::unpack_kind_offset()));
2247   Label noException;
2248   __ mv(t0, Deoptimization::Unpack_exception);
2249   __ bne(xcpool, t0, noException); // Was exception pending?
2250   __ ld(x10, Address(xthread, JavaThread::exception_oop_offset()));
2251   __ ld(x13, Address(xthread, JavaThread::exception_pc_offset()));
2252   __ sd(zr, Address(xthread, JavaThread::exception_oop_offset()));
2253   __ sd(zr, Address(xthread, JavaThread::exception_pc_offset()));
2254 
2255   __ verify_oop(x10);
2256 
2257   // Overwrite the result registers with the exception results.
2258   __ sd(x10, Address(sp, reg_saver.reg_offset_in_bytes(x10)));
2259 
2260   __ bind(noException);
2261 
2262   // Only register save data is on the stack.
2263   // Now restore the result registers.  Everything else is either dead
2264   // or captured in the vframeArray.
2265 
2266   // Restore fp result register
2267   __ fld(f10, Address(sp, reg_saver.freg_offset_in_bytes(f10)));
2268   // Restore integer result register
2269   __ ld(x10, Address(sp, reg_saver.reg_offset_in_bytes(x10)));
2270 
2271   // Pop all of the register save area off the stack
2272   __ add(sp, sp, frame_size_in_words * wordSize);
2273 
2274   // All of the register save area has been popped of the stack. Only the
2275   // return address remains.
2276 
2277   // Pop all the frames we must move/replace.
2278   //
2279   // Frame picture (youngest to oldest)
2280   // 1: self-frame (no frame link)
2281   // 2: deopting frame  (no frame link)
2282   // 3: caller of deopting frame (could be compiled/interpreted).
2283   //
2284   // Note: by leaving the return address of self-frame on the stack
2285   // and using the size of frame 2 to adjust the stack
2286   // when we are done the return to frame 3 will still be on the stack.
2287 
2288   // Pop deoptimized frame
2289   __ lwu(x12, Address(x15, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset()));
2290   __ sub(x12, x12, 2 * wordSize);
2291   __ add(sp, sp, x12);
2292   __ ld(fp, Address(sp, 0));
2293   __ ld(ra, Address(sp, wordSize));
2294   __ addi(sp, sp, 2 * wordSize);
2295   // RA should now be the return address to the caller (3)
2296 
2297 #ifdef ASSERT
2298   // Compilers generate code that bang the stack by as much as the
2299   // interpreter would need. So this stack banging should never
2300   // trigger a fault. Verify that it does not on non product builds.
2301   __ lwu(x9, Address(x15, Deoptimization::UnrollBlock::total_frame_sizes_offset()));
2302   __ bang_stack_size(x9, x12);
2303 #endif
2304   // Load address of array of frame pcs into x12
2305   __ ld(x12, Address(x15, Deoptimization::UnrollBlock::frame_pcs_offset()));
2306 
2307   // Load address of array of frame sizes into x14
2308   __ ld(x14, Address(x15, Deoptimization::UnrollBlock::frame_sizes_offset()));
2309 
2310   // Load counter into x13
2311   __ lwu(x13, Address(x15, Deoptimization::UnrollBlock::number_of_frames_offset()));
2312 
2313   // Now adjust the caller's stack to make up for the extra locals
2314   // but record the original sp so that we can save it in the skeletal interpreter
2315   // frame and the stack walking of interpreter_sender will get the unextended sp
2316   // value and not the "real" sp value.
2317 
2318   const Register sender_sp = x16;
2319 
2320   __ mv(sender_sp, sp);
2321   __ lwu(x9, Address(x15,
2322                      Deoptimization::UnrollBlock::
2323                      caller_adjustment_offset()));
2324   __ sub(sp, sp, x9);
2325 
2326   // Push interpreter frames in a loop
2327   __ mv(t0, 0xDEADDEAD);               // Make a recognizable pattern
2328   __ mv(t1, t0);
2329   Label loop;
2330   __ bind(loop);
2331   __ ld(x9, Address(x14, 0));          // Load frame size
2332   __ addi(x14, x14, wordSize);
2333   __ sub(x9, x9, 2 * wordSize);        // We'll push pc and fp by hand
2334   __ ld(ra, Address(x12, 0));          // Load pc
2335   __ addi(x12, x12, wordSize);
2336   __ enter();                          // Save old & set new fp
2337   __ sub(sp, sp, x9);                  // Prolog
2338   // This value is corrected by layout_activation_impl
2339   __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
2340   __ sd(sender_sp, Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize)); // Make it walkable
2341   __ mv(sender_sp, sp);                // Pass sender_sp to next frame
2342   __ addi(x13, x13, -1);               // Decrement counter
2343   __ bnez(x13, loop);
2344 
2345     // Re-push self-frame
2346   __ ld(ra, Address(x12));
2347   __ enter();
2348 
2349   // Allocate a full sized register save area.  We subtract 2 because
2350   // enter() just pushed 2 words
2351   __ sub(sp, sp, (frame_size_in_words - 2) * wordSize);
2352 
2353   // Restore frame locals after moving the frame
2354   __ fsd(f10, Address(sp, reg_saver.freg_offset_in_bytes(f10)));
2355   __ sd(x10, Address(sp, reg_saver.reg_offset_in_bytes(x10)));
2356 
2357   // Call C code.  Need thread but NOT official VM entry
2358   // crud.  We cannot block on this call, no GC can happen.  Call should
2359   // restore return values to their stack-slots with the new SP.
2360   //
2361   // void Deoptimization::unpack_frames(JavaThread* thread, int exec_mode)
2362 
2363   // Use fp because the frames look interpreted now
2364   // Don't need the precise return PC here, just precise enough to point into this code blob.
2365   address the_pc = __ pc();
2366   __ set_last_Java_frame(sp, fp, the_pc, t0);
2367 
2368   __ mv(c_rarg0, xthread);
2369   __ mv(c_rarg1, xcpool); // second arg: exec_mode
2370   __ rt_call(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames));
2371 
2372   // Set an oopmap for the call site
2373   // Use the same PC we used for the last java frame
2374   oop_maps->add_gc_map(the_pc - start,
2375                        new OopMap(frame_size_in_words, 0));
2376 
2377   // Clear fp AND pc
2378   __ reset_last_Java_frame(true);
2379 
2380   // Collect return values
2381   __ fld(f10, Address(sp, reg_saver.freg_offset_in_bytes(f10)));
2382   __ ld(x10, Address(sp, reg_saver.reg_offset_in_bytes(x10)));
2383 
2384   // Pop self-frame.
2385   __ leave();                           // Epilog
2386 
2387   // Jump to interpreter
2388   __ ret();
2389 
2390   // Make sure all code is generated
2391   masm->flush();
2392 
2393   _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words);
2394   assert(_deopt_blob != nullptr, "create deoptimization blob fail!");
2395   _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset);
2396 #if INCLUDE_JVMCI
2397   if (EnableJVMCI) {
2398     _deopt_blob->set_uncommon_trap_offset(uncommon_trap_offset);
2399     _deopt_blob->set_implicit_exception_uncommon_trap_offset(implicit_exception_uncommon_trap_offset);
2400   }
2401 #endif
2402 }
2403 
2404 // Number of stack slots between incoming argument block and the start of
2405 // a new frame. The PROLOG must add this many slots to the stack. The
2406 // EPILOG must remove this many slots.
2407 // RISCV needs two words for RA (return address) and FP (frame pointer).
2408 uint SharedRuntime::in_preserve_stack_slots() {
2409   return 2 * VMRegImpl::slots_per_word;
2410 }
2411 
2412 uint SharedRuntime::out_preserve_stack_slots() {
2413   return 0;
2414 }
2415 
2416 VMReg SharedRuntime::thread_register() {
2417   Unimplemented();
2418   return nullptr;
2419 }
2420 
2421 #ifdef COMPILER2
2422 //------------------------------generate_uncommon_trap_blob--------------------
2423 void SharedRuntime::generate_uncommon_trap_blob() {
2424   // Allocate space for the code
2425   ResourceMark rm;
2426   // Setup code generation tools
2427   CodeBuffer buffer("uncommon_trap_blob", 2048, 1024);
2428   MacroAssembler* masm = new MacroAssembler(&buffer);
2429   assert_cond(masm != nullptr);
2430 
2431   assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
2432 
2433   address start = __ pc();
2434 
2435   // Push self-frame.  We get here with a return address in RA
2436   // and sp should be 16 byte aligned
2437   // push fp and retaddr by hand
2438   __ addi(sp, sp, -2 * wordSize);
2439   __ sd(ra, Address(sp, wordSize));
2440   __ sd(fp, Address(sp, 0));
2441   // we don't expect an arg reg save area
2442 #ifndef PRODUCT
2443   assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
2444 #endif
2445   // compiler left unloaded_class_index in j_rarg0 move to where the
2446   // runtime expects it.
2447   __ sign_extend(c_rarg1, j_rarg0, 32);
2448 
2449   // we need to set the past SP to the stack pointer of the stub frame
2450   // and the pc to the address where this runtime call will return
2451   // although actually any pc in this code blob will do).
2452   Label retaddr;
2453   __ set_last_Java_frame(sp, noreg, retaddr, t0);
2454 
2455   // Call C code.  Need thread but NOT official VM entry
2456   // crud.  We cannot block on this call, no GC can happen.  Call should
2457   // capture callee-saved registers as well as return values.
2458   //
2459   // UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index, jint exec_mode)
2460   //
2461   // n.b. 3 gp args, 0 fp args, integral return type
2462 
2463   __ mv(c_rarg0, xthread);
2464   __ mv(c_rarg2, Deoptimization::Unpack_uncommon_trap);
2465   __ rt_call(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap));
2466   __ bind(retaddr);
2467 
2468   // Set an oopmap for the call site
2469   OopMapSet* oop_maps = new OopMapSet();
2470   OopMap* map = new OopMap(SimpleRuntimeFrame::framesize, 0);
2471   assert_cond(oop_maps != nullptr && map != nullptr);
2472 
2473   // location of fp is known implicitly by the frame sender code
2474 
2475   oop_maps->add_gc_map(__ pc() - start, map);
2476 
2477   __ reset_last_Java_frame(false);
2478 
2479   // move UnrollBlock* into x14
2480   __ mv(x14, x10);
2481 
2482 #ifdef ASSERT
2483   { Label L;
2484     __ lwu(t0, Address(x14, Deoptimization::UnrollBlock::unpack_kind_offset()));
2485     __ mv(t1, Deoptimization::Unpack_uncommon_trap);
2486     __ beq(t0, t1, L);
2487     __ stop("SharedRuntime::generate_uncommon_trap_blob: expected Unpack_uncommon_trap");
2488     __ bind(L);
2489   }
2490 #endif
2491 
2492   // Pop all the frames we must move/replace.
2493   //
2494   // Frame picture (youngest to oldest)
2495   // 1: self-frame (no frame link)
2496   // 2: deopting frame  (no frame link)
2497   // 3: caller of deopting frame (could be compiled/interpreted).
2498 
2499   __ add(sp, sp, (SimpleRuntimeFrame::framesize) << LogBytesPerInt); // Epilog!
2500 
2501   // Pop deoptimized frame (int)
2502   __ lwu(x12, Address(x14,
2503                       Deoptimization::UnrollBlock::
2504                       size_of_deoptimized_frame_offset()));
2505   __ sub(x12, x12, 2 * wordSize);
2506   __ add(sp, sp, x12);
2507   __ ld(fp, Address(sp, 0));
2508   __ ld(ra, Address(sp, wordSize));
2509   __ addi(sp, sp, 2 * wordSize);
2510   // RA should now be the return address to the caller (3) frame
2511 
2512 #ifdef ASSERT
2513   // Compilers generate code that bang the stack by as much as the
2514   // interpreter would need. So this stack banging should never
2515   // trigger a fault. Verify that it does not on non product builds.
2516   __ lwu(x11, Address(x14,
2517                       Deoptimization::UnrollBlock::
2518                       total_frame_sizes_offset()));
2519   __ bang_stack_size(x11, x12);
2520 #endif
2521 
2522   // Load address of array of frame pcs into x12 (address*)
2523   __ ld(x12, Address(x14,
2524                      Deoptimization::UnrollBlock::frame_pcs_offset()));
2525 
2526   // Load address of array of frame sizes into x15 (intptr_t*)
2527   __ ld(x15, Address(x14,
2528                      Deoptimization::UnrollBlock::
2529                      frame_sizes_offset()));
2530 
2531   // Counter
2532   __ lwu(x13, Address(x14,
2533                       Deoptimization::UnrollBlock::
2534                       number_of_frames_offset())); // (int)
2535 
2536   // Now adjust the caller's stack to make up for the extra locals but
2537   // record the original sp so that we can save it in the skeletal
2538   // interpreter frame and the stack walking of interpreter_sender
2539   // will get the unextended sp value and not the "real" sp value.
2540 
2541   const Register sender_sp = t1; // temporary register
2542 
2543   __ lwu(x11, Address(x14,
2544                       Deoptimization::UnrollBlock::
2545                       caller_adjustment_offset())); // (int)
2546   __ mv(sender_sp, sp);
2547   __ sub(sp, sp, x11);
2548 
2549   // Push interpreter frames in a loop
2550   Label loop;
2551   __ bind(loop);
2552   __ ld(x11, Address(x15, 0));       // Load frame size
2553   __ sub(x11, x11, 2 * wordSize);    // We'll push pc and fp by hand
2554   __ ld(ra, Address(x12, 0));        // Save return address
2555   __ enter();                        // and old fp & set new fp
2556   __ sub(sp, sp, x11);               // Prolog
2557   __ sd(sender_sp, Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize)); // Make it walkable
2558   // This value is corrected by layout_activation_impl
2559   __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
2560   __ mv(sender_sp, sp);              // Pass sender_sp to next frame
2561   __ add(x15, x15, wordSize);        // Bump array pointer (sizes)
2562   __ add(x12, x12, wordSize);        // Bump array pointer (pcs)
2563   __ subw(x13, x13, 1);              // Decrement counter
2564   __ bgtz(x13, loop);
2565   __ ld(ra, Address(x12, 0));        // save final return address
2566   // Re-push self-frame
2567   __ enter();                        // & old fp & set new fp
2568 
2569   // Use fp because the frames look interpreted now
2570   // Save "the_pc" since it cannot easily be retrieved using the last_java_SP after we aligned SP.
2571   // Don't need the precise return PC here, just precise enough to point into this code blob.
2572   address the_pc = __ pc();
2573   __ set_last_Java_frame(sp, fp, the_pc, t0);
2574 
2575   // Call C code.  Need thread but NOT official VM entry
2576   // crud.  We cannot block on this call, no GC can happen.  Call should
2577   // restore return values to their stack-slots with the new SP.
2578   //
2579   // BasicType unpack_frames(JavaThread* thread, int exec_mode)
2580   //
2581 
2582   // n.b. 2 gp args, 0 fp args, integral return type
2583 
2584   // sp should already be aligned
2585   __ mv(c_rarg0, xthread);
2586   __ mv(c_rarg1, Deoptimization::Unpack_uncommon_trap);
2587   __ rt_call(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames));
2588 
2589   // Set an oopmap for the call site
2590   // Use the same PC we used for the last java frame
2591   oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
2592 
2593   // Clear fp AND pc
2594   __ reset_last_Java_frame(true);
2595 
2596   // Pop self-frame.
2597   __ leave();                 // Epilog
2598 
2599   // Jump to interpreter
2600   __ ret();
2601 
2602   // Make sure all code is generated
2603   masm->flush();
2604 
2605   _uncommon_trap_blob =  UncommonTrapBlob::create(&buffer, oop_maps,
2606                                                   SimpleRuntimeFrame::framesize >> 1);
2607 }
2608 #endif // COMPILER2
2609 
2610 //------------------------------generate_handler_blob------
2611 //
2612 // Generate a special Compile2Runtime blob that saves all registers,
2613 // and setup oopmap.
2614 //
2615 SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, int poll_type) {
2616   ResourceMark rm;
2617   OopMapSet *oop_maps = new OopMapSet();
2618   assert_cond(oop_maps != nullptr);
2619   OopMap* map = nullptr;
2620 
2621   // Allocate space for the code.  Setup code generation tools.
2622   CodeBuffer buffer("handler_blob", 2048, 1024);
2623   MacroAssembler* masm = new MacroAssembler(&buffer);
2624   assert_cond(masm != nullptr);
2625 
2626   address start   = __ pc();
2627   address call_pc = nullptr;
2628   int frame_size_in_words = -1;
2629   bool cause_return = (poll_type == POLL_AT_RETURN);
2630   RegisterSaver reg_saver(poll_type == POLL_AT_VECTOR_LOOP /* save_vectors */);
2631 
2632   // Save Integer and Float registers.
2633   map = reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2634 
2635   // The following is basically a call_VM.  However, we need the precise
2636   // address of the call in order to generate an oopmap. Hence, we do all the
2637   // work ourselves.
2638 
2639   Label retaddr;
2640   __ set_last_Java_frame(sp, noreg, retaddr, t0);
2641 
2642   // The return address must always be correct so that frame constructor never
2643   // sees an invalid pc.
2644 
2645   if (!cause_return) {
2646     // overwrite the return address pushed by save_live_registers
2647     // Additionally, x18 is a callee-saved register so we can look at
2648     // it later to determine if someone changed the return address for
2649     // us!
2650     __ ld(x18, Address(xthread, JavaThread::saved_exception_pc_offset()));
2651     __ sd(x18, Address(fp, frame::return_addr_offset * wordSize));
2652   }
2653 
2654   // Do the call
2655   __ mv(c_rarg0, xthread);
2656   __ rt_call(call_ptr);
2657   __ bind(retaddr);
2658 
2659   // Set an oopmap for the call site.  This oopmap will map all
2660   // oop-registers and debug-info registers as callee-saved.  This
2661   // will allow deoptimization at this safepoint to find all possible
2662   // debug-info recordings, as well as let GC find all oops.
2663 
2664   oop_maps->add_gc_map( __ pc() - start, map);
2665 
2666   Label noException;
2667 
2668   __ reset_last_Java_frame(false);
2669 
2670   __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2671 
2672   __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
2673   __ beqz(t0, noException);
2674 
2675   // Exception pending
2676 
2677   reg_saver.restore_live_registers(masm);
2678 
2679   __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2680 
2681   // No exception case
2682   __ bind(noException);
2683 
2684   Label no_adjust, bail;
2685   if (!cause_return) {
2686     // If our stashed return pc was modified by the runtime we avoid touching it
2687     __ ld(t0, Address(fp, frame::return_addr_offset * wordSize));
2688     __ bne(x18, t0, no_adjust);
2689 
2690 #ifdef ASSERT
2691     // Verify the correct encoding of the poll we're about to skip.
2692     // See NativeInstruction::is_lwu_to_zr()
2693     __ lwu(t0, Address(x18));
2694     __ andi(t1, t0, 0b0000011);
2695     __ mv(t2, 0b0000011);
2696     __ bne(t1, t2, bail); // 0-6:0b0000011
2697     __ srli(t1, t0, 7);
2698     __ andi(t1, t1, 0b00000);
2699     __ bnez(t1, bail);    // 7-11:0b00000
2700     __ srli(t1, t0, 12);
2701     __ andi(t1, t1, 0b110);
2702     __ mv(t2, 0b110);
2703     __ bne(t1, t2, bail); // 12-14:0b110
2704 #endif
2705     // Adjust return pc forward to step over the safepoint poll instruction
2706     __ add(x18, x18, NativeInstruction::instruction_size);
2707     __ sd(x18, Address(fp, frame::return_addr_offset * wordSize));
2708   }
2709 
2710   __ bind(no_adjust);
2711   // Normal exit, restore registers and exit.
2712 
2713   reg_saver.restore_live_registers(masm);
2714   __ ret();
2715 
2716 #ifdef ASSERT
2717   __ bind(bail);
2718   __ stop("Attempting to adjust pc to skip safepoint poll but the return point is not what we expected");
2719 #endif
2720 
2721   // Make sure all code is generated
2722   masm->flush();
2723 
2724   // Fill-out other meta info
2725   return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words);
2726 }
2727 
2728 //
2729 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss
2730 //
2731 // Generate a stub that calls into vm to find out the proper destination
2732 // of a java call. All the argument registers are live at this point
2733 // but since this is generic code we don't know what they are and the caller
2734 // must do any gc of the args.
2735 //
2736 RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) {
2737   assert(StubRoutines::forward_exception_entry() != nullptr, "must be generated before");
2738 
2739   // allocate space for the code
2740   ResourceMark rm;
2741 
2742   CodeBuffer buffer(name, 1000, 512);
2743   MacroAssembler* masm = new MacroAssembler(&buffer);
2744   assert_cond(masm != nullptr);
2745 
2746   int frame_size_in_words = -1;
2747   RegisterSaver reg_saver(false /* save_vectors */);
2748 
2749   OopMapSet *oop_maps = new OopMapSet();
2750   assert_cond(oop_maps != nullptr);
2751   OopMap* map = nullptr;
2752 
2753   int start = __ offset();
2754 
2755   map = reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2756 
2757   int frame_complete = __ offset();
2758 
2759   {
2760     Label retaddr;
2761     __ set_last_Java_frame(sp, noreg, retaddr, t0);
2762 
2763     __ mv(c_rarg0, xthread);
2764     __ rt_call(destination);
2765     __ bind(retaddr);
2766   }
2767 
2768   // Set an oopmap for the call site.
2769   // We need this not only for callee-saved registers, but also for volatile
2770   // registers that the compiler might be keeping live across a safepoint.
2771 
2772   oop_maps->add_gc_map( __ offset() - start, map);
2773 
2774   // x10 contains the address we are going to jump to assuming no exception got installed
2775 
2776   // clear last_Java_sp
2777   __ reset_last_Java_frame(false);
2778   // check for pending exceptions
2779   Label pending;
2780   __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
2781   __ bnez(t0, pending);
2782 
2783   // get the returned Method*
2784   __ get_vm_result_2(xmethod, xthread);
2785   __ sd(xmethod, Address(sp, reg_saver.reg_offset_in_bytes(xmethod)));
2786 
2787   // x10 is where we want to jump, overwrite t0 which is saved and temporary
2788   __ sd(x10, Address(sp, reg_saver.reg_offset_in_bytes(t0)));
2789   reg_saver.restore_live_registers(masm);
2790 
2791   // We are back to the original state on entry and ready to go.
2792 
2793   __ jr(t0);
2794 
2795   // Pending exception after the safepoint
2796 
2797   __ bind(pending);
2798 
2799   reg_saver.restore_live_registers(masm);
2800 
2801   // exception pending => remove activation and forward to exception handler
2802 
2803   __ sd(zr, Address(xthread, JavaThread::vm_result_offset()));
2804 
2805   __ ld(x10, Address(xthread, Thread::pending_exception_offset()));
2806   __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2807 
2808   // -------------
2809   // make sure all code is generated
2810   masm->flush();
2811 
2812   // return the  blob
2813   return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true);
2814 }
2815 
2816 #ifdef COMPILER2
2817 //------------------------------generate_exception_blob---------------------------
2818 // creates exception blob at the end
2819 // Using exception blob, this code is jumped from a compiled method.
2820 // (see emit_exception_handler in riscv.ad file)
2821 //
2822 // Given an exception pc at a call we call into the runtime for the
2823 // handler in this method. This handler might merely restore state
2824 // (i.e. callee save registers) unwind the frame and jump to the
2825 // exception handler for the nmethod if there is no Java level handler
2826 // for the nmethod.
2827 //
2828 // This code is entered with a jmp.
2829 //
2830 // Arguments:
2831 //   x10: exception oop
2832 //   x13: exception pc
2833 //
2834 // Results:
2835 //   x10: exception oop
2836 //   x13: exception pc in caller
2837 //   destination: exception handler of caller
2838 //
2839 // Note: the exception pc MUST be at a call (precise debug information)
2840 //       Registers x10, x13, x12, x14, x15, t0 are not callee saved.
2841 //
2842 
2843 void OptoRuntime::generate_exception_blob() {
2844   assert(!OptoRuntime::is_callee_saved_register(R13_num), "");
2845   assert(!OptoRuntime::is_callee_saved_register(R10_num), "");
2846   assert(!OptoRuntime::is_callee_saved_register(R12_num), "");
2847 
2848   assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
2849 
2850   // Allocate space for the code
2851   ResourceMark rm;
2852   // Setup code generation tools
2853   CodeBuffer buffer("exception_blob", 2048, 1024);
2854   MacroAssembler* masm = new MacroAssembler(&buffer);
2855   assert_cond(masm != nullptr);
2856 
2857   // TODO check various assumptions made here
2858   //
2859   // make sure we do so before running this
2860 
2861   address start = __ pc();
2862 
2863   // push fp and retaddr by hand
2864   // Exception pc is 'return address' for stack walker
2865   __ addi(sp, sp, -2 * wordSize);
2866   __ sd(ra, Address(sp, wordSize));
2867   __ sd(fp, Address(sp));
2868   // there are no callee save registers and we don't expect an
2869   // arg reg save area
2870 #ifndef PRODUCT
2871   assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
2872 #endif
2873   // Store exception in Thread object. We cannot pass any arguments to the
2874   // handle_exception call, since we do not want to make any assumption
2875   // about the size of the frame where the exception happened in.
2876   __ sd(x10, Address(xthread, JavaThread::exception_oop_offset()));
2877   __ sd(x13, Address(xthread, JavaThread::exception_pc_offset()));
2878 
2879   // This call does all the hard work.  It checks if an exception handler
2880   // exists in the method.
2881   // If so, it returns the handler address.
2882   // If not, it prepares for stack-unwinding, restoring the callee-save
2883   // registers of the frame being removed.
2884   //
2885   // address OptoRuntime::handle_exception_C(JavaThread* thread)
2886   //
2887   // n.b. 1 gp arg, 0 fp args, integral return type
2888 
2889   // the stack should always be aligned
2890   address the_pc = __ pc();
2891   __ set_last_Java_frame(sp, noreg, the_pc, t0);
2892   __ mv(c_rarg0, xthread);
2893   __ rt_call(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C));
2894 
2895   // handle_exception_C is a special VM call which does not require an explicit
2896   // instruction sync afterwards.
2897 
2898   // Set an oopmap for the call site.  This oopmap will only be used if we
2899   // are unwinding the stack.  Hence, all locations will be dead.
2900   // Callee-saved registers will be the same as the frame above (i.e.,
2901   // handle_exception_stub), since they were restored when we got the
2902   // exception.
2903 
2904   OopMapSet* oop_maps = new OopMapSet();
2905   assert_cond(oop_maps != nullptr);
2906 
2907   oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
2908 
2909   __ reset_last_Java_frame(false);
2910 
2911   // Restore callee-saved registers
2912 
2913   // fp is an implicitly saved callee saved register (i.e. the calling
2914   // convention will save restore it in prolog/epilog) Other than that
2915   // there are no callee save registers now that adapter frames are gone.
2916   // and we dont' expect an arg reg save area
2917   __ ld(fp, Address(sp));
2918   __ ld(x13, Address(sp, wordSize));
2919   __ addi(sp, sp , 2 * wordSize);
2920 
2921   // x10: exception handler
2922 
2923   // We have a handler in x10 (could be deopt blob).
2924   __ mv(t0, x10);
2925 
2926   // Get the exception oop
2927   __ ld(x10, Address(xthread, JavaThread::exception_oop_offset()));
2928   // Get the exception pc in case we are deoptimized
2929   __ ld(x14, Address(xthread, JavaThread::exception_pc_offset()));
2930 #ifdef ASSERT
2931   __ sd(zr, Address(xthread, JavaThread::exception_handler_pc_offset()));
2932   __ sd(zr, Address(xthread, JavaThread::exception_pc_offset()));
2933 #endif
2934   // Clear the exception oop so GC no longer processes it as a root.
2935   __ sd(zr, Address(xthread, JavaThread::exception_oop_offset()));
2936 
2937   // x10: exception oop
2938   // t0:  exception handler
2939   // x14: exception pc
2940   // Jump to handler
2941 
2942   __ jr(t0);
2943 
2944   // Make sure all code is generated
2945   masm->flush();
2946 
2947   // Set exception blob
2948   _exception_blob =  ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
2949 }
2950 #endif // COMPILER2