1 /* 2 * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 #include "asm/macroAssembler.inline.hpp" 27 #include "classfile/javaClasses.hpp" 28 #include "compiler/disassembler.hpp" 29 #include "compiler/compiler_globals.hpp" 30 #include "gc/shared/barrierSetAssembler.hpp" 31 #include "interpreter/bytecodeHistogram.hpp" 32 #include "interpreter/interpreter.hpp" 33 #include "interpreter/interpreterRuntime.hpp" 34 #include "interpreter/interp_masm.hpp" 35 #include "interpreter/templateInterpreterGenerator.hpp" 36 #include "interpreter/templateTable.hpp" 37 #include "interpreter/bytecodeTracer.hpp" 38 #include "memory/resourceArea.hpp" 39 #include "oops/arrayOop.hpp" 40 #include "oops/method.hpp" 41 #include "oops/methodCounters.hpp" 42 #include "oops/methodData.hpp" 43 #include "oops/oop.inline.hpp" 44 #include "oops/resolvedIndyEntry.hpp" 45 #include "oops/resolvedMethodEntry.hpp" 46 #include "prims/jvmtiExport.hpp" 47 #include "prims/jvmtiThreadState.hpp" 48 #include "runtime/arguments.hpp" 49 #include "runtime/deoptimization.hpp" 50 #include "runtime/frame.inline.hpp" 51 #include "runtime/globals.hpp" 52 #include "runtime/jniHandles.hpp" 53 #include "runtime/sharedRuntime.hpp" 54 #include "runtime/stubRoutines.hpp" 55 #include "runtime/synchronizer.hpp" 56 #include "runtime/timer.hpp" 57 #include "runtime/vframeArray.hpp" 58 #include "utilities/checkedCast.hpp" 59 #include "utilities/debug.hpp" 60 #include "utilities/powerOfTwo.hpp" 61 #include <sys/types.h> 62 63 // Size of interpreter code. Increase if too small. Interpreter will 64 // fail with a guarantee ("not enough space for interpreter generation"); 65 // if too small. 66 // Run with +PrintInterpreter to get the VM to print out the size. 67 // Max size with JVMTI 68 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024; 69 70 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)-> 71 72 address TemplateInterpreterGenerator::generate_slow_signature_handler() { 73 address entry = __ pc(); 74 75 __ andr(esp, esp, -16); 76 __ mov(c_rarg3, esp); 77 // rmethod 78 // rlocals 79 // c_rarg3: first stack arg - wordSize 80 81 // adjust sp 82 __ sub(sp, c_rarg3, 18 * wordSize); 83 __ str(lr, Address(__ pre(sp, -2 * wordSize))); 84 __ call_VM(noreg, 85 CAST_FROM_FN_PTR(address, 86 InterpreterRuntime::slow_signature_handler), 87 rmethod, rlocals, c_rarg3); 88 89 // r0: result handler 90 91 // Stack layout: 92 // rsp: return address <- sp 93 // 1 garbage 94 // 8 integer args (if static first is unused) 95 // 1 float/double identifiers 96 // 8 double args 97 // stack args <- esp 98 // garbage 99 // expression stack bottom 100 // bcp (null) 101 // ... 102 103 // Restore LR 104 __ ldr(lr, Address(__ post(sp, 2 * wordSize))); 105 106 // Do FP first so we can use c_rarg3 as temp 107 __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers 108 109 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) { 110 const FloatRegister r = as_FloatRegister(i); 111 112 Label d, done; 113 114 __ tbnz(c_rarg3, i, d); 115 __ ldrs(r, Address(sp, (10 + i) * wordSize)); 116 __ b(done); 117 __ bind(d); 118 __ ldrd(r, Address(sp, (10 + i) * wordSize)); 119 __ bind(done); 120 } 121 122 // c_rarg0 contains the result from the call of 123 // InterpreterRuntime::slow_signature_handler so we don't touch it 124 // here. It will be loaded with the JNIEnv* later. 125 __ ldr(c_rarg1, Address(sp, 1 * wordSize)); 126 for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) { 127 Register rm = as_Register(i), rn = as_Register(i+1); 128 __ ldp(rm, rn, Address(sp, i * wordSize)); 129 } 130 131 __ add(sp, sp, 18 * wordSize); 132 __ ret(lr); 133 134 return entry; 135 } 136 137 138 // 139 // Various method entries 140 // 141 142 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) { 143 // rmethod: Method* 144 // r19_sender_sp: sender sp 145 // esp: args 146 147 // These don't need a safepoint check because they aren't virtually 148 // callable. We won't enter these intrinsics from compiled code. 149 // If in the future we added an intrinsic which was virtually callable 150 // we'd have to worry about how to safepoint so that this code is used. 151 152 // mathematical functions inlined by compiler 153 // (interpreter must provide identical implementation 154 // in order to avoid monotonicity bugs when switching 155 // from interpreter to compiler in the middle of some 156 // computation) 157 // 158 // stack: 159 // [ arg ] <-- esp 160 // [ arg ] 161 // retaddr in lr 162 163 address entry_point = nullptr; 164 Register continuation = lr; 165 switch (kind) { 166 case Interpreter::java_lang_math_abs: 167 entry_point = __ pc(); 168 __ ldrd(v0, Address(esp)); 169 __ fabsd(v0, v0); 170 __ mov(sp, r19_sender_sp); // Restore caller's SP 171 break; 172 case Interpreter::java_lang_math_sqrt: 173 entry_point = __ pc(); 174 __ ldrd(v0, Address(esp)); 175 __ fsqrtd(v0, v0); 176 __ mov(sp, r19_sender_sp); 177 break; 178 case Interpreter::java_lang_math_sin : 179 case Interpreter::java_lang_math_cos : 180 case Interpreter::java_lang_math_tan : 181 case Interpreter::java_lang_math_log : 182 case Interpreter::java_lang_math_log10 : 183 case Interpreter::java_lang_math_exp : 184 entry_point = __ pc(); 185 __ ldrd(v0, Address(esp)); 186 __ mov(sp, r19_sender_sp); 187 __ mov(r23, lr); 188 continuation = r23; // The first free callee-saved register 189 generate_transcendental_entry(kind, 1); 190 break; 191 case Interpreter::java_lang_math_pow : 192 entry_point = __ pc(); 193 __ mov(r23, lr); 194 continuation = r23; 195 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize)); 196 __ ldrd(v1, Address(esp)); 197 __ mov(sp, r19_sender_sp); 198 generate_transcendental_entry(kind, 2); 199 break; 200 case Interpreter::java_lang_math_fmaD : 201 if (UseFMA) { 202 entry_point = __ pc(); 203 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize)); 204 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize)); 205 __ ldrd(v2, Address(esp)); 206 __ fmaddd(v0, v0, v1, v2); 207 __ mov(sp, r19_sender_sp); // Restore caller's SP 208 } 209 break; 210 case Interpreter::java_lang_math_fmaF : 211 if (UseFMA) { 212 entry_point = __ pc(); 213 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize)); 214 __ ldrs(v1, Address(esp, Interpreter::stackElementSize)); 215 __ ldrs(v2, Address(esp)); 216 __ fmadds(v0, v0, v1, v2); 217 __ mov(sp, r19_sender_sp); // Restore caller's SP 218 } 219 break; 220 default: 221 ; 222 } 223 if (entry_point) { 224 __ br(continuation); 225 } 226 227 return entry_point; 228 } 229 230 // double trigonometrics and transcendentals 231 // static jdouble dsin(jdouble x); 232 // static jdouble dcos(jdouble x); 233 // static jdouble dtan(jdouble x); 234 // static jdouble dlog(jdouble x); 235 // static jdouble dlog10(jdouble x); 236 // static jdouble dexp(jdouble x); 237 // static jdouble dpow(jdouble x, jdouble y); 238 239 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) { 240 address fn; 241 switch (kind) { 242 case Interpreter::java_lang_math_sin : 243 if (StubRoutines::dsin() == nullptr) { 244 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); 245 } else { 246 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin()); 247 } 248 break; 249 case Interpreter::java_lang_math_cos : 250 if (StubRoutines::dcos() == nullptr) { 251 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); 252 } else { 253 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos()); 254 } 255 break; 256 case Interpreter::java_lang_math_tan : 257 if (StubRoutines::dtan() == nullptr) { 258 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); 259 } else { 260 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan()); 261 } 262 break; 263 case Interpreter::java_lang_math_log : 264 if (StubRoutines::dlog() == nullptr) { 265 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); 266 } else { 267 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog()); 268 } 269 break; 270 case Interpreter::java_lang_math_log10 : 271 if (StubRoutines::dlog10() == nullptr) { 272 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); 273 } else { 274 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10()); 275 } 276 break; 277 case Interpreter::java_lang_math_exp : 278 if (StubRoutines::dexp() == nullptr) { 279 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); 280 } else { 281 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp()); 282 } 283 break; 284 case Interpreter::java_lang_math_pow : 285 if (StubRoutines::dpow() == nullptr) { 286 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); 287 } else { 288 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow()); 289 } 290 break; 291 default: 292 ShouldNotReachHere(); 293 fn = nullptr; // unreachable 294 } 295 __ mov(rscratch1, fn); 296 __ blr(rscratch1); 297 } 298 299 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() { 300 assert(VM_Version::supports_float16(), "this intrinsic is not supported"); 301 // r19_sender_sp: sender sp 302 // stack: 303 // [ arg ] <-- esp 304 // [ arg ] 305 // retaddr in lr 306 // result in v0 307 308 address entry_point = __ pc(); 309 __ ldrw(c_rarg0, Address(esp)); 310 __ flt16_to_flt(v0, c_rarg0, v1); 311 __ mov(sp, r19_sender_sp); // Restore caller's SP 312 __ br(lr); 313 return entry_point; 314 } 315 316 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() { 317 assert(VM_Version::supports_float16(), "this intrinsic is not supported"); 318 // r19_sender_sp: sender sp 319 // stack: 320 // [ arg ] <-- esp 321 // [ arg ] 322 // retaddr in lr 323 // result in c_rarg0 324 325 address entry_point = __ pc(); 326 __ ldrs(v0, Address(esp)); 327 __ flt_to_flt16(c_rarg0, v0, v1); 328 __ mov(sp, r19_sender_sp); // Restore caller's SP 329 __ br(lr); 330 return entry_point; 331 } 332 333 // Abstract method entry 334 // Attempt to execute abstract method. Throw exception 335 address TemplateInterpreterGenerator::generate_abstract_entry(void) { 336 // rmethod: Method* 337 // r19_sender_sp: sender SP 338 339 address entry_point = __ pc(); 340 341 // abstract method entry 342 343 // pop return address, reset last_sp to null 344 __ empty_expression_stack(); 345 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed) 346 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 347 348 // throw exception 349 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 350 InterpreterRuntime::throw_AbstractMethodErrorWithMethod), 351 rmethod); 352 // the call_VM checks for exception, so we should never return here. 353 __ should_not_reach_here(); 354 355 return entry_point; 356 } 357 358 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 359 address entry = __ pc(); 360 361 #ifdef ASSERT 362 { 363 Label L; 364 __ ldr(rscratch1, Address(rfp, 365 frame::interpreter_frame_monitor_block_top_offset * 366 wordSize)); 367 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize))); 368 __ mov(rscratch2, sp); 369 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack 370 // grows negative) 371 __ br(Assembler::HS, L); // check if frame is complete 372 __ stop ("interpreter frame not set up"); 373 __ bind(L); 374 } 375 #endif // ASSERT 376 // Restore bcp under the assumption that the current frame is still 377 // interpreted 378 __ restore_bcp(); 379 380 // expression stack must be empty before entering the VM if an 381 // exception happened 382 __ empty_expression_stack(); 383 // throw exception 384 __ call_VM(noreg, 385 CAST_FROM_FN_PTR(address, 386 InterpreterRuntime::throw_StackOverflowError)); 387 return entry; 388 } 389 390 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() { 391 address entry = __ pc(); 392 // expression stack must be empty before entering the VM if an 393 // exception happened 394 __ empty_expression_stack(); 395 // setup parameters 396 397 // ??? convention: expect aberrant index in register r1 398 __ movw(c_rarg2, r1); 399 // ??? convention: expect array in register r3 400 __ mov(c_rarg1, r3); 401 __ call_VM(noreg, 402 CAST_FROM_FN_PTR(address, 403 InterpreterRuntime:: 404 throw_ArrayIndexOutOfBoundsException), 405 c_rarg1, c_rarg2); 406 return entry; 407 } 408 409 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 410 address entry = __ pc(); 411 412 // object is at TOS 413 __ pop(c_rarg1); 414 415 // expression stack must be empty before entering the VM if an 416 // exception happened 417 __ empty_expression_stack(); 418 419 __ call_VM(noreg, 420 CAST_FROM_FN_PTR(address, 421 InterpreterRuntime:: 422 throw_ClassCastException), 423 c_rarg1); 424 return entry; 425 } 426 427 address TemplateInterpreterGenerator::generate_exception_handler_common( 428 const char* name, const char* message, bool pass_oop) { 429 assert(!pass_oop || message == nullptr, "either oop or message but not both"); 430 address entry = __ pc(); 431 if (pass_oop) { 432 // object is at TOS 433 __ pop(c_rarg2); 434 } 435 // expression stack must be empty before entering the VM if an 436 // exception happened 437 __ empty_expression_stack(); 438 // setup parameters 439 __ lea(c_rarg1, Address((address)name)); 440 if (pass_oop) { 441 __ call_VM(r0, CAST_FROM_FN_PTR(address, 442 InterpreterRuntime:: 443 create_klass_exception), 444 c_rarg1, c_rarg2); 445 } else { 446 // kind of lame ExternalAddress can't take null because 447 // external_word_Relocation will assert. 448 if (message != nullptr) { 449 __ lea(c_rarg2, Address((address)message)); 450 } else { 451 __ mov(c_rarg2, NULL_WORD); 452 } 453 __ call_VM(r0, 454 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), 455 c_rarg1, c_rarg2); 456 } 457 // throw exception 458 __ b(address(Interpreter::throw_exception_entry())); 459 return entry; 460 } 461 462 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) { 463 address entry = __ pc(); 464 465 // Restore stack bottom in case i2c adjusted stack 466 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 467 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize))); 468 // and null it as marker that esp is now tos until next java call 469 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 470 __ restore_bcp(); 471 __ restore_locals(); 472 __ restore_constant_pool_cache(); 473 __ get_method(rmethod); 474 475 if (state == atos) { 476 Register obj = r0; 477 Register mdp = r1; 478 Register tmp = r2; 479 __ profile_return_type(mdp, obj, tmp); 480 } 481 482 const Register cache = r1; 483 const Register index = r2; 484 485 if (index_size == sizeof(u4)) { 486 __ load_resolved_indy_entry(cache, index); 487 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset()))); 488 __ add(esp, esp, cache, Assembler::LSL, 3); 489 } else { 490 // Pop N words from the stack 491 assert(index_size == sizeof(u2), "Can only be u2"); 492 __ load_method_entry(cache, index); 493 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset()))); 494 __ add(esp, esp, cache, Assembler::LSL, 3); 495 } 496 497 // Restore machine SP 498 __ restore_sp_after_call(); 499 500 __ check_and_handle_popframe(rthread); 501 __ check_and_handle_earlyret(rthread); 502 503 __ get_dispatch(); 504 __ dispatch_next(state, step); 505 506 return entry; 507 } 508 509 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, 510 int step, 511 address continuation) { 512 address entry = __ pc(); 513 __ restore_bcp(); 514 __ restore_locals(); 515 __ restore_constant_pool_cache(); 516 __ get_method(rmethod); 517 __ get_dispatch(); 518 519 __ restore_sp_after_call(); // Restore SP to extended SP 520 521 // Restore expression stack pointer 522 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 523 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize))); 524 // null last_sp until next java call 525 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 526 527 #if INCLUDE_JVMCI 528 // Check if we need to take lock at entry of synchronized method. This can 529 // only occur on method entry so emit it only for vtos with step 0. 530 if (EnableJVMCI && state == vtos && step == 0) { 531 Label L; 532 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset())); 533 __ cbz(rscratch1, L); 534 // Clear flag. 535 __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset())); 536 // Take lock. 537 lock_method(); 538 __ bind(L); 539 } else { 540 #ifdef ASSERT 541 if (EnableJVMCI) { 542 Label L; 543 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset())); 544 __ cbz(rscratch1, L); 545 __ stop("unexpected pending monitor in deopt entry"); 546 __ bind(L); 547 } 548 #endif 549 } 550 #endif 551 // handle exceptions 552 { 553 Label L; 554 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset())); 555 __ cbz(rscratch1, L); 556 __ call_VM(noreg, 557 CAST_FROM_FN_PTR(address, 558 InterpreterRuntime::throw_pending_exception)); 559 __ should_not_reach_here(); 560 __ bind(L); 561 } 562 563 if (continuation == nullptr) { 564 __ dispatch_next(state, step); 565 } else { 566 __ jump_to_entry(continuation); 567 } 568 return entry; 569 } 570 571 address TemplateInterpreterGenerator::generate_result_handler_for( 572 BasicType type) { 573 address entry = __ pc(); 574 switch (type) { 575 case T_BOOLEAN: __ c2bool(r0); break; 576 case T_CHAR : __ uxth(r0, r0); break; 577 case T_BYTE : __ sxtb(r0, r0); break; 578 case T_SHORT : __ sxth(r0, r0); break; 579 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this 580 case T_LONG : /* nothing to do */ break; 581 case T_VOID : /* nothing to do */ break; 582 case T_FLOAT : /* nothing to do */ break; 583 case T_DOUBLE : /* nothing to do */ break; 584 case T_OBJECT : 585 // retrieve result from frame 586 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize)); 587 // and verify it 588 __ verify_oop(r0); 589 break; 590 default : ShouldNotReachHere(); 591 } 592 __ ret(lr); // return from result handler 593 return entry; 594 } 595 596 address TemplateInterpreterGenerator::generate_safept_entry_for( 597 TosState state, 598 address runtime_entry) { 599 address entry = __ pc(); 600 __ push(state); 601 __ push_cont_fastpath(rthread); 602 __ call_VM(noreg, runtime_entry); 603 __ pop_cont_fastpath(rthread); 604 __ membar(Assembler::AnyAny); 605 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 606 return entry; 607 } 608 609 address TemplateInterpreterGenerator::generate_cont_resume_interpreter_adapter() { 610 if (!Continuations::enabled()) return nullptr; 611 address start = __ pc(); 612 613 __ restore_bcp(); 614 __ restore_locals(); 615 616 // Restore constant pool cache 617 __ ldr(rcpool, Address(rfp, frame::interpreter_frame_cache_offset * wordSize)); 618 619 // Restore Java expression stack pointer 620 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 621 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize))); 622 // and null it as marker that esp is now tos until next java call 623 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 624 625 // Restore machine SP 626 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize)); 627 __ lea(sp, Address(rfp, rscratch1, Address::lsl(LogBytesPerWord))); 628 629 // Restore method 630 __ ldr(rmethod, Address(rfp, frame::interpreter_frame_method_offset * wordSize)); 631 632 // Restore dispatch 633 uint64_t offset; 634 __ adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset); 635 __ add(rdispatch, rdispatch, offset); 636 637 __ ret(lr); 638 639 return start; 640 } 641 642 643 // Helpers for commoning out cases in the various type of method entries. 644 // 645 646 647 // increment invocation count & check for overflow 648 // 649 // Note: checking for negative value instead of overflow 650 // so we have a 'sticky' overflow test 651 // 652 // rmethod: method 653 // 654 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) { 655 Label done; 656 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not. 657 int increment = InvocationCounter::count_increment; 658 Label no_mdo; 659 if (ProfileInterpreter) { 660 // Are we profiling? 661 __ ldr(r0, Address(rmethod, Method::method_data_offset())); 662 __ cbz(r0, no_mdo); 663 // Increment counter in the MDO 664 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) + 665 in_bytes(InvocationCounter::counter_offset())); 666 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset())); 667 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow); 668 __ b(done); 669 } 670 __ bind(no_mdo); 671 // Increment counter in MethodCounters 672 const Address invocation_counter(rscratch2, 673 MethodCounters::invocation_counter_offset() + 674 InvocationCounter::counter_offset()); 675 __ get_method_counters(rmethod, rscratch2, done); 676 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset())); 677 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow); 678 __ bind(done); 679 } 680 681 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) { 682 683 // Asm interpreter on entry 684 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] 685 // Everything as it was on entry 686 687 // InterpreterRuntime::frequency_counter_overflow takes two 688 // arguments, the first (thread) is passed by call_VM, the second 689 // indicates if the counter overflow occurs at a backwards branch 690 // (null bcp). We pass zero for it. The call returns the address 691 // of the verified entry point for the method or null if the 692 // compilation did not complete (either went background or bailed 693 // out). 694 __ mov(c_rarg1, 0); 695 __ call_VM(noreg, 696 CAST_FROM_FN_PTR(address, 697 InterpreterRuntime::frequency_counter_overflow), 698 c_rarg1); 699 700 __ b(do_continue); 701 } 702 703 // See if we've got enough room on the stack for locals plus overhead 704 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError 705 // without going through the signal handler, i.e., reserved and yellow zones 706 // will not be made usable. The shadow zone must suffice to handle the 707 // overflow. 708 // The expression stack grows down incrementally, so the normal guard 709 // page mechanism will work for that. 710 // 711 // NOTE: Since the additional locals are also always pushed (wasn't 712 // obvious in generate_method_entry) so the guard should work for them 713 // too. 714 // 715 // Args: 716 // r3: number of additional locals this frame needs (what we must check) 717 // rmethod: Method* 718 // 719 // Kills: 720 // r0 721 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) { 722 723 // monitor entry size: see picture of stack set 724 // (generate_method_entry) and frame_amd64.hpp 725 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes(); 726 727 // total overhead size: entry_size + (saved rbp through expr stack 728 // bottom). be sure to change this if you add/subtract anything 729 // to/from the overhead area 730 const int overhead_size = 731 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size; 732 733 const size_t page_size = os::vm_page_size(); 734 735 Label after_frame_check; 736 737 // see if the frame is greater than one page in size. If so, 738 // then we need to verify there is enough stack space remaining 739 // for the additional locals. 740 // 741 // Note that we use SUBS rather than CMP here because the immediate 742 // field of this instruction may overflow. SUBS can cope with this 743 // because it is a macro that will expand to some number of MOV 744 // instructions and a register operation. 745 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize); 746 __ br(Assembler::LS, after_frame_check); 747 748 // compute rsp as if this were going to be the last frame on 749 // the stack before the red zone 750 751 // locals + overhead, in bytes 752 __ mov(r0, overhead_size); 753 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter. 754 755 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset()); 756 __ ldr(rscratch1, stack_limit); 757 758 #ifdef ASSERT 759 Label limit_okay; 760 // Verify that thread stack limit is non-zero. 761 __ cbnz(rscratch1, limit_okay); 762 __ stop("stack overflow limit is zero"); 763 __ bind(limit_okay); 764 #endif 765 766 // Add stack limit to locals. 767 __ add(r0, r0, rscratch1); 768 769 // Check against the current stack bottom. 770 __ cmp(sp, r0); 771 __ br(Assembler::HI, after_frame_check); 772 773 // Remove the incoming args, peeling the machine SP back to where it 774 // was in the caller. This is not strictly necessary, but unless we 775 // do so the stack frame may have a garbage FP; this ensures a 776 // correct call stack that we can always unwind. The ANDR should be 777 // unnecessary because the sender SP in r19 is always aligned, but 778 // it doesn't hurt. 779 __ andr(sp, r19_sender_sp, -16); 780 781 // Note: the restored frame is not necessarily interpreted. 782 // Use the shared runtime version of the StackOverflowError. 783 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "stub not yet generated"); 784 __ far_jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry())); 785 786 // all done with frame size check 787 __ bind(after_frame_check); 788 } 789 790 // Allocate monitor and lock method (asm interpreter) 791 // 792 // Args: 793 // rmethod: Method* 794 // rlocals: locals 795 // 796 // Kills: 797 // r0 798 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs) 799 // rscratch1, rscratch2 (scratch regs) 800 void TemplateInterpreterGenerator::lock_method() { 801 // synchronize method 802 const Address access_flags(rmethod, Method::access_flags_offset()); 803 const Address monitor_block_top( 804 rfp, 805 frame::interpreter_frame_monitor_block_top_offset * wordSize); 806 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes(); 807 808 #ifdef ASSERT 809 { 810 Label L; 811 __ ldrh(r0, access_flags); 812 __ tst(r0, JVM_ACC_SYNCHRONIZED); 813 __ br(Assembler::NE, L); 814 __ stop("method doesn't need synchronization"); 815 __ bind(L); 816 } 817 #endif // ASSERT 818 819 // get synchronization object 820 { 821 Label done; 822 __ ldrh(r0, access_flags); 823 __ tst(r0, JVM_ACC_STATIC); 824 // get receiver (assume this is frequent case) 825 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0))); 826 __ br(Assembler::EQ, done); 827 __ load_mirror(r0, rmethod, r5, rscratch2); 828 829 #ifdef ASSERT 830 { 831 Label L; 832 __ cbnz(r0, L); 833 __ stop("synchronization object is null"); 834 __ bind(L); 835 } 836 #endif // ASSERT 837 838 __ bind(done); 839 } 840 841 // add space for monitor & lock 842 __ check_extended_sp(); 843 __ sub(sp, sp, entry_size); // add space for a monitor entry 844 __ sub(esp, esp, entry_size); 845 __ sub(rscratch1, sp, rfp); 846 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize); 847 __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize)); 848 __ sub(rscratch1, esp, rfp); 849 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize); 850 __ str(rscratch1, monitor_block_top); // set new monitor block top 851 852 // store object 853 __ str(r0, Address(esp, BasicObjectLock::obj_offset())); 854 __ mov(c_rarg1, esp); // object address 855 __ lock_object(c_rarg1); 856 } 857 858 // Generate a fixed interpreter frame. This is identical setup for 859 // interpreted methods and for native methods hence the shared code. 860 // 861 // Args: 862 // lr: return address 863 // rmethod: Method* 864 // rlocals: pointer to locals 865 // rcpool: cp cache 866 // stack_pointer: previous sp 867 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { 868 // Save ConstMethod* in r5_const_method for later use to avoid loading multiple times 869 Register r5_const_method = r5; 870 __ ldr(r5_const_method, Address(rmethod, Method::const_offset())); 871 872 // initialize fixed part of activation frame 873 if (native_call) { 874 __ sub(esp, sp, 14 * wordSize); 875 __ mov(rbcp, zr); 876 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset); 877 __ stp(rscratch1, zr, Address(__ pre(sp, -14 * wordSize))); 878 // add 2 zero-initialized slots for native calls 879 __ stp(zr, zr, Address(sp, 12 * wordSize)); 880 } else { 881 __ sub(esp, sp, 12 * wordSize); 882 __ add(rbcp, r5_const_method, in_bytes(ConstMethod::codes_offset())); // get codebase 883 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset); 884 __ stp(rscratch1, rbcp, Address(__ pre(sp, -12 * wordSize))); 885 } 886 887 if (ProfileInterpreter) { 888 Label method_data_continue; 889 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset())); 890 __ cbz(rscratch1, method_data_continue); 891 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset()))); 892 __ bind(method_data_continue); 893 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer) 894 } else { 895 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp) 896 } 897 898 __ protect_return_address(); 899 __ stp(rfp, lr, Address(sp, 10 * wordSize)); 900 __ lea(rfp, Address(sp, 10 * wordSize)); 901 902 // Save ConstantPool* in r11_constants for later use to avoid loading multiple times 903 Register r11_constants = r11; 904 __ ldr(r11_constants, Address(r5_const_method, ConstMethod::constants_offset())); 905 __ ldr(rcpool, Address(r11_constants, ConstantPool::cache_offset())); 906 __ sub(rscratch1, rlocals, rfp); 907 __ lsr(rscratch1, rscratch1, Interpreter::logStackElementSize); // rscratch1 = rlocals - fp(); 908 // Store relativized rlocals, see frame::interpreter_frame_locals(). 909 __ stp(rscratch1, rcpool, Address(sp, 2 * wordSize)); 910 911 // set sender sp 912 // leave last_sp as null 913 __ stp(zr, r19_sender_sp, Address(sp, 8 * wordSize)); 914 915 // Get mirror. Resolve ConstantPool* -> InstanceKlass* -> Java mirror. 916 __ ldr(r10, Address(r11_constants, ConstantPool::pool_holder_offset())); 917 __ ldr(r10, Address(r10, in_bytes(Klass::java_mirror_offset()))); 918 __ resolve_oop_handle(r10, rscratch1, rscratch2); 919 if (! native_call) { 920 __ ldrh(rscratch1, Address(r5_const_method, ConstMethod::max_stack_offset())); 921 __ add(rscratch1, rscratch1, MAX2(3, Method::extra_stack_entries())); 922 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3); 923 __ andr(rscratch1, rscratch1, -16); 924 __ sub(rscratch2, rscratch1, rfp); 925 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize); 926 // Store extended SP and mirror 927 __ stp(r10, rscratch2, Address(sp, 4 * wordSize)); 928 // Move SP out of the way 929 __ mov(sp, rscratch1); 930 } else { 931 // Make sure there is room for the exception oop pushed in case method throws 932 // an exception (see TemplateInterpreterGenerator::generate_throw_exception()) 933 __ sub(rscratch1, sp, 2 * wordSize); 934 __ sub(rscratch2, rscratch1, rfp); 935 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize); 936 __ stp(r10, rscratch2, Address(sp, 4 * wordSize)); 937 __ mov(sp, rscratch1); 938 } 939 } 940 941 // End of helpers 942 943 // Various method entries 944 //------------------------------------------------------------------------------------------------------------------------ 945 // 946 // 947 948 // Method entry for java.lang.ref.Reference.get. 949 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) { 950 // Code: _aload_0, _getfield, _areturn 951 // parameter size = 1 952 // 953 // The code that gets generated by this routine is split into 2 parts: 954 // 1. The "intrinsified" code for G1 (or any SATB based GC), 955 // 2. The slow path - which is an expansion of the regular method entry. 956 // 957 // Notes:- 958 // * In the G1 code we do not check whether we need to block for 959 // a safepoint. If G1 is enabled then we must execute the specialized 960 // code for Reference.get (except when the Reference object is null) 961 // so that we can log the value in the referent field with an SATB 962 // update buffer. 963 // If the code for the getfield template is modified so that the 964 // G1 pre-barrier code is executed when the current method is 965 // Reference.get() then going through the normal method entry 966 // will be fine. 967 // * The G1 code can, however, check the receiver object (the instance 968 // of java.lang.Reference) and jump to the slow path if null. If the 969 // Reference object is null then we obviously cannot fetch the referent 970 // and so we don't need to call the G1 pre-barrier. Thus we can use the 971 // regular method entry code to generate the NPE. 972 // 973 // This code is based on generate_accessor_entry. 974 // 975 // rmethod: Method* 976 // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path 977 978 // LR is live. It must be saved around calls. 979 980 address entry = __ pc(); 981 982 const int referent_offset = java_lang_ref_Reference::referent_offset(); 983 984 Label slow_path; 985 const Register local_0 = c_rarg0; 986 // Check if local 0 != null 987 // If the receiver is null then it is OK to jump to the slow path. 988 __ ldr(local_0, Address(esp, 0)); 989 __ cbz(local_0, slow_path); 990 991 // Load the value of the referent field. 992 const Address field_address(local_0, referent_offset); 993 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler(); 994 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2); 995 996 // areturn 997 __ andr(sp, r19_sender_sp, -16); // done with stack 998 __ ret(lr); 999 1000 // generate a vanilla interpreter entry as the slow path 1001 __ bind(slow_path); 1002 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals)); 1003 return entry; 1004 1005 } 1006 1007 /** 1008 * Method entry for static native methods: 1009 * int java.util.zip.CRC32.update(int crc, int b) 1010 */ 1011 address TemplateInterpreterGenerator::generate_CRC32_update_entry() { 1012 assert(UseCRC32Intrinsics, "this intrinsic is not supported"); 1013 address entry = __ pc(); 1014 1015 // rmethod: Method* 1016 // r19_sender_sp: senderSP must preserved for slow path 1017 // esp: args 1018 1019 Label slow_path; 1020 // If we need a safepoint check, generate full interpreter entry. 1021 __ safepoint_poll(slow_path, false /* at_return */, false /* in_nmethod */); 1022 1023 // We don't generate local frame and don't align stack because 1024 // we call stub code and there is no safepoint on this path. 1025 1026 // Load parameters 1027 const Register crc = c_rarg0; // crc 1028 const Register val = c_rarg1; // source java byte value 1029 const Register tbl = c_rarg2; // scratch 1030 1031 // Arguments are reversed on java expression stack 1032 __ ldrw(val, Address(esp, 0)); // byte value 1033 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC 1034 1035 uint64_t offset; 1036 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset); 1037 __ add(tbl, tbl, offset); 1038 1039 __ mvnw(crc, crc); // ~crc 1040 __ update_byte_crc32(crc, val, tbl); 1041 __ mvnw(crc, crc); // ~crc 1042 1043 // result in c_rarg0 1044 1045 __ andr(sp, r19_sender_sp, -16); 1046 __ ret(lr); 1047 1048 // generate a vanilla native entry as the slow path 1049 __ bind(slow_path); 1050 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native)); 1051 return entry; 1052 } 1053 1054 /** 1055 * Method entry for static native methods: 1056 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len) 1057 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 1058 */ 1059 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 1060 assert(UseCRC32Intrinsics, "this intrinsic is not supported"); 1061 address entry = __ pc(); 1062 1063 // rmethod,: Method* 1064 // r19_sender_sp: senderSP must preserved for slow path 1065 1066 Label slow_path; 1067 // If we need a safepoint check, generate full interpreter entry. 1068 __ safepoint_poll(slow_path, false /* at_return */, false /* in_nmethod */); 1069 1070 // We don't generate local frame and don't align stack because 1071 // we call stub code and there is no safepoint on this path. 1072 1073 // Load parameters 1074 const Register crc = c_rarg0; // crc 1075 const Register buf = c_rarg1; // source java byte array address 1076 const Register len = c_rarg2; // length 1077 const Register off = len; // offset (never overlaps with 'len') 1078 1079 // Arguments are reversed on java expression stack 1080 // Calculate address of start element 1081 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { 1082 __ ldr(buf, Address(esp, 2*wordSize)); // long buf 1083 __ ldrw(off, Address(esp, wordSize)); // offset 1084 __ add(buf, buf, off); // + offset 1085 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC 1086 } else { 1087 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array 1088 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size 1089 __ ldrw(off, Address(esp, wordSize)); // offset 1090 __ add(buf, buf, off); // + offset 1091 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC 1092 } 1093 // Can now load 'len' since we're finished with 'off' 1094 __ ldrw(len, Address(esp, 0x0)); // Length 1095 1096 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP 1097 1098 // We are frameless so we can just jump to the stub. 1099 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32())); 1100 1101 // generate a vanilla native entry as the slow path 1102 __ bind(slow_path); 1103 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native)); 1104 return entry; 1105 } 1106 1107 /** 1108 * Method entry for intrinsic-candidate (non-native) methods: 1109 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end) 1110 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end) 1111 * Unlike CRC32, CRC32C does not have any methods marked as native 1112 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses 1113 */ 1114 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 1115 assert(UseCRC32CIntrinsics, "this intrinsic is not supported"); 1116 address entry = __ pc(); 1117 1118 // Prepare jump to stub using parameters from the stack 1119 const Register crc = c_rarg0; // initial crc 1120 const Register buf = c_rarg1; // source java byte array address 1121 const Register len = c_rarg2; // len argument to the kernel 1122 1123 const Register end = len; // index of last element to process 1124 const Register off = crc; // offset 1125 1126 __ ldrw(end, Address(esp)); // int end 1127 __ ldrw(off, Address(esp, wordSize)); // int offset 1128 __ sub(len, end, off); 1129 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf 1130 __ add(buf, buf, off); // + offset 1131 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { 1132 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc 1133 } else { 1134 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size 1135 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc 1136 } 1137 1138 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP 1139 1140 // Jump to the stub. 1141 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C())); 1142 1143 return entry; 1144 } 1145 1146 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) { 1147 // See more discussion in stackOverflow.hpp. 1148 1149 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size()); 1150 const int page_size = (int)os::vm_page_size(); 1151 const int n_shadow_pages = shadow_zone_size / page_size; 1152 1153 #ifdef ASSERT 1154 Label L_good_limit; 1155 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit())); 1156 __ cbnz(rscratch1, L_good_limit); 1157 __ stop("shadow zone safe limit is not initialized"); 1158 __ bind(L_good_limit); 1159 1160 Label L_good_watermark; 1161 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark())); 1162 __ cbnz(rscratch1, L_good_watermark); 1163 __ stop("shadow zone growth watermark is not initialized"); 1164 __ bind(L_good_watermark); 1165 #endif 1166 1167 Label L_done; 1168 1169 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark())); 1170 __ cmp(sp, rscratch1); 1171 __ br(Assembler::HI, L_done); 1172 1173 for (int p = 1; p <= n_shadow_pages; p++) { 1174 __ sub(rscratch2, sp, p*page_size); 1175 __ str(zr, Address(rscratch2)); 1176 } 1177 1178 // Record the new watermark, but only if the update is above the safe limit. 1179 // Otherwise, the next time around the check above would pass the safe limit. 1180 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit())); 1181 __ cmp(sp, rscratch1); 1182 __ br(Assembler::LS, L_done); 1183 __ mov(rscratch1, sp); 1184 __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark())); 1185 1186 __ bind(L_done); 1187 } 1188 1189 // Interpreter stub for calling a native method. (asm interpreter) 1190 // This sets up a somewhat different looking stack for calling the 1191 // native method than the typical interpreter frame setup. 1192 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) { 1193 // determine code generation flags 1194 bool inc_counter = UseCompiler || CountCompiledCalls; 1195 1196 // r1: Method* 1197 // rscratch1: sender sp 1198 1199 address entry_point = __ pc(); 1200 1201 const Address constMethod (rmethod, Method::const_offset()); 1202 const Address access_flags (rmethod, Method::access_flags_offset()); 1203 const Address size_of_parameters(r2, ConstMethod:: 1204 size_of_parameters_offset()); 1205 1206 // get parameter size (always needed) 1207 __ ldr(r2, constMethod); 1208 __ load_unsigned_short(r2, size_of_parameters); 1209 1210 // Native calls don't need the stack size check since they have no 1211 // expression stack and the arguments are already on the stack and 1212 // we only add a handful of words to the stack. 1213 1214 // rmethod: Method* 1215 // r2: size of parameters 1216 // rscratch1: sender sp 1217 1218 // for natives the size of locals is zero 1219 1220 // compute beginning of parameters (rlocals) 1221 __ add(rlocals, esp, r2, ext::uxtx, 3); 1222 __ add(rlocals, rlocals, -wordSize); 1223 1224 // Pull SP back to minimum size: this avoids holes in the stack 1225 __ andr(sp, esp, -16); 1226 1227 // initialize fixed part of activation frame 1228 generate_fixed_frame(true); 1229 1230 // make sure method is native & not abstract 1231 #ifdef ASSERT 1232 __ ldrh(r0, access_flags); 1233 { 1234 Label L; 1235 __ tst(r0, JVM_ACC_NATIVE); 1236 __ br(Assembler::NE, L); 1237 __ stop("tried to execute non-native method as native"); 1238 __ bind(L); 1239 } 1240 { 1241 Label L; 1242 __ tst(r0, JVM_ACC_ABSTRACT); 1243 __ br(Assembler::EQ, L); 1244 __ stop("tried to execute abstract method in interpreter"); 1245 __ bind(L); 1246 } 1247 #endif 1248 1249 // Since at this point in the method invocation the exception 1250 // handler would try to exit the monitor of synchronized methods 1251 // which hasn't been entered yet, we set the thread local variable 1252 // _do_not_unlock_if_synchronized to true. The remove_activation 1253 // will check this flag. 1254 1255 const Address do_not_unlock_if_synchronized(rthread, 1256 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1257 __ mov(rscratch2, true); 1258 __ strb(rscratch2, do_not_unlock_if_synchronized); 1259 1260 // increment invocation count & check for overflow 1261 Label invocation_counter_overflow; 1262 if (inc_counter) { 1263 generate_counter_incr(&invocation_counter_overflow); 1264 } 1265 1266 Label continue_after_compile; 1267 __ bind(continue_after_compile); 1268 1269 bang_stack_shadow_pages(true); 1270 1271 // reset the _do_not_unlock_if_synchronized flag 1272 __ strb(zr, do_not_unlock_if_synchronized); 1273 1274 // check for synchronized methods 1275 // Must happen AFTER invocation_counter check and stack overflow check, 1276 // so method is not locked if overflows. 1277 if (synchronized) { 1278 lock_method(); 1279 } else { 1280 // no synchronization necessary 1281 #ifdef ASSERT 1282 { 1283 Label L; 1284 __ ldrh(r0, access_flags); 1285 __ tst(r0, JVM_ACC_SYNCHRONIZED); 1286 __ br(Assembler::EQ, L); 1287 __ stop("method needs synchronization"); 1288 __ bind(L); 1289 } 1290 #endif 1291 } 1292 1293 // start execution 1294 #ifdef ASSERT 1295 { 1296 Label L; 1297 const Address monitor_block_top(rfp, 1298 frame::interpreter_frame_monitor_block_top_offset * wordSize); 1299 __ ldr(rscratch1, monitor_block_top); 1300 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize))); 1301 __ cmp(esp, rscratch1); 1302 __ br(Assembler::EQ, L); 1303 __ stop("broken stack frame setup in interpreter 1"); 1304 __ bind(L); 1305 } 1306 #endif 1307 1308 // jvmti support 1309 __ notify_method_entry(); 1310 1311 // work registers 1312 const Register t = r17; 1313 const Register result_handler = r19; 1314 1315 // allocate space for parameters 1316 __ ldr(t, Address(rmethod, Method::const_offset())); 1317 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset())); 1318 1319 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize); 1320 __ andr(sp, rscratch1, -16); 1321 __ mov(esp, rscratch1); 1322 1323 // get signature handler 1324 { 1325 Label L; 1326 __ ldr(t, Address(rmethod, Method::signature_handler_offset())); 1327 __ cbnz(t, L); 1328 __ call_VM(noreg, 1329 CAST_FROM_FN_PTR(address, 1330 InterpreterRuntime::prepare_native_call), 1331 rmethod); 1332 __ ldr(t, Address(rmethod, Method::signature_handler_offset())); 1333 __ bind(L); 1334 } 1335 1336 // call signature handler 1337 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals, 1338 "adjust this code"); 1339 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp, 1340 "adjust this code"); 1341 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1, 1342 "adjust this code"); 1343 1344 // The generated handlers do not touch rmethod (the method). 1345 // However, large signatures cannot be cached and are generated 1346 // each time here. The slow-path generator can do a GC on return, 1347 // so we must reload it after the call. 1348 __ blr(t); 1349 __ get_method(rmethod); // slow path can do a GC, reload rmethod 1350 1351 1352 // result handler is in r0 1353 // set result handler 1354 __ mov(result_handler, r0); 1355 // Save it in the frame in case of preemption; we cannot rely on callee saved registers. 1356 __ str(r0, Address(rfp, frame::interpreter_frame_result_handler_offset * wordSize)); 1357 1358 // pass mirror handle if static call 1359 { 1360 Label L; 1361 __ ldrh(t, Address(rmethod, Method::access_flags_offset())); 1362 __ tbz(t, exact_log2(JVM_ACC_STATIC), L); 1363 // get mirror 1364 __ load_mirror(t, rmethod, r10, rscratch2); 1365 // copy mirror into activation frame 1366 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize)); 1367 // pass handle to mirror 1368 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize); 1369 __ bind(L); 1370 } 1371 1372 // get native function entry point in r10 1373 { 1374 Label L; 1375 __ ldr(r10, Address(rmethod, Method::native_function_offset())); 1376 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry()); 1377 __ lea(rscratch2, unsatisfied); 1378 __ ldr(rscratch2, rscratch2); 1379 __ cmp(r10, rscratch2); 1380 __ br(Assembler::NE, L); 1381 __ call_VM(noreg, 1382 CAST_FROM_FN_PTR(address, 1383 InterpreterRuntime::prepare_native_call), 1384 rmethod); 1385 __ get_method(rmethod); 1386 __ ldr(r10, Address(rmethod, Method::native_function_offset())); 1387 __ bind(L); 1388 } 1389 1390 // pass JNIEnv 1391 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset())); 1392 1393 // It is enough that the pc() points into the right code 1394 // segment. It does not have to be the correct return pc. 1395 // For convenience we use the pc we want to resume to in 1396 // case of preemption on Object.wait. 1397 Label native_return; 1398 __ set_last_Java_frame(esp, rfp, native_return, rscratch1); 1399 1400 // change thread state 1401 #ifdef ASSERT 1402 { 1403 Label L; 1404 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset())); 1405 __ cmp(t, (u1)_thread_in_Java); 1406 __ br(Assembler::EQ, L); 1407 __ stop("Wrong thread state in native stub"); 1408 __ bind(L); 1409 } 1410 #endif 1411 1412 // Change state to native 1413 __ mov(rscratch1, _thread_in_native); 1414 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset())); 1415 __ stlrw(rscratch1, rscratch2); 1416 1417 __ push_cont_fastpath(); 1418 1419 // Call the native method. 1420 __ blr(r10); 1421 1422 __ pop_cont_fastpath(); 1423 1424 __ get_method(rmethod); 1425 // result potentially in r0 or v0 1426 1427 // Restore cpu control state after JNI call 1428 __ restore_cpu_control_state_after_jni(rscratch1, rscratch2); 1429 1430 // make room for the pushes we're about to do 1431 __ sub(rscratch1, esp, 4 * wordSize); 1432 __ andr(sp, rscratch1, -16); 1433 1434 // NOTE: The order of these pushes is known to frame::interpreter_frame_result 1435 // in order to extract the result of a method call. If the order of these 1436 // pushes change or anything else is added to the stack then the code in 1437 // interpreter_frame_result must also change. 1438 __ push(dtos); 1439 __ push(ltos); 1440 1441 __ verify_sve_vector_length(); 1442 1443 // change thread state 1444 __ mov(rscratch1, _thread_in_native_trans); 1445 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset())); 1446 __ stlrw(rscratch1, rscratch2); 1447 1448 // Force this write out before the read below 1449 if (!UseSystemMemoryBarrier) { 1450 __ dmb(Assembler::ISH); 1451 } 1452 1453 // check for safepoint operation in progress and/or pending suspend requests 1454 { 1455 Label L, Continue; 1456 1457 // No need for acquire as Java threads always disarm themselves. 1458 __ safepoint_poll(L, true /* at_return */, false /* in_nmethod */); 1459 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset())); 1460 __ cbz(rscratch2, Continue); 1461 __ bind(L); 1462 1463 // Don't use call_VM as it will see a possible pending exception 1464 // and forward it and never return here preventing us from 1465 // clearing _last_native_pc down below. So we do a runtime call by 1466 // hand. 1467 // 1468 __ mov(c_rarg0, rthread); 1469 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans))); 1470 __ blr(rscratch2); 1471 __ get_method(rmethod); 1472 __ reinit_heapbase(); 1473 __ bind(Continue); 1474 } 1475 1476 // change thread state 1477 __ mov(rscratch1, _thread_in_Java); 1478 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset())); 1479 __ stlrw(rscratch1, rscratch2); 1480 1481 // Check preemption for Object.wait() 1482 Label not_preempted; 1483 __ ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset())); 1484 __ cbz(rscratch1, not_preempted); 1485 __ str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset())); 1486 __ br(rscratch1); 1487 __ bind(native_return); 1488 __ restore_after_resume(true /* is_native */); 1489 // reload result_handler 1490 __ ldr(result_handler, Address(rfp, frame::interpreter_frame_result_handler_offset*wordSize)); 1491 __ bind(not_preempted); 1492 1493 // reset_last_Java_frame 1494 __ reset_last_Java_frame(true); 1495 1496 if (CheckJNICalls) { 1497 // clear_pending_jni_exception_check 1498 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset())); 1499 } 1500 1501 // reset handle block 1502 __ ldr(t, Address(rthread, JavaThread::active_handles_offset())); 1503 __ str(zr, Address(t, JNIHandleBlock::top_offset())); 1504 1505 // If result is an oop unbox and store it in frame where gc will see it 1506 // and result handler will pick it up 1507 1508 { 1509 Label no_oop; 1510 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT))); 1511 __ cmp(t, result_handler); 1512 __ br(Assembler::NE, no_oop); 1513 // Unbox oop result, e.g. JNIHandles::resolve result. 1514 __ pop(ltos); 1515 __ resolve_jobject(r0, t, rscratch2); 1516 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize)); 1517 // keep stack depth as expected by pushing oop which will eventually be discarded 1518 __ push(ltos); 1519 __ bind(no_oop); 1520 } 1521 1522 { 1523 Label no_reguard; 1524 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset()))); 1525 __ ldrw(rscratch1, Address(rscratch1)); 1526 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled); 1527 __ br(Assembler::NE, no_reguard); 1528 1529 __ push_call_clobbered_registers(); 1530 __ mov(c_rarg0, rthread); 1531 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); 1532 __ blr(rscratch2); 1533 __ pop_call_clobbered_registers(); 1534 1535 __ bind(no_reguard); 1536 } 1537 1538 // The method register is junk from after the thread_in_native transition 1539 // until here. Also can't call_VM until the bcp has been 1540 // restored. Need bcp for throwing exception below so get it now. 1541 __ get_method(rmethod); 1542 1543 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=> 1544 // rbcp == code_base() 1545 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod* 1546 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase 1547 // handle exceptions (exception handling will handle unlocking!) 1548 { 1549 Label L; 1550 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset())); 1551 __ cbz(rscratch1, L); 1552 // Note: At some point we may want to unify this with the code 1553 // used in call_VM_base(); i.e., we should use the 1554 // StubRoutines::forward_exception code. For now this doesn't work 1555 // here because the rsp is not correctly set at this point. 1556 __ MacroAssembler::call_VM(noreg, 1557 CAST_FROM_FN_PTR(address, 1558 InterpreterRuntime::throw_pending_exception)); 1559 __ should_not_reach_here(); 1560 __ bind(L); 1561 } 1562 1563 // do unlocking if necessary 1564 { 1565 Label L; 1566 __ ldrh(t, Address(rmethod, Method::access_flags_offset())); 1567 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L); 1568 // the code below should be shared with interpreter macro 1569 // assembler implementation 1570 { 1571 Label unlock; 1572 // BasicObjectLock will be first in list, since this is a 1573 // synchronized method. However, need to check that the object 1574 // has not been unlocked by an explicit monitorexit bytecode. 1575 1576 // monitor expect in c_rarg1 for slow unlock path 1577 __ lea (c_rarg1, Address(rfp, // address of first monitor 1578 (intptr_t)(frame::interpreter_frame_initial_sp_offset * 1579 wordSize - sizeof(BasicObjectLock)))); 1580 1581 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset())); 1582 __ cbnz(t, unlock); 1583 1584 // Entry already unlocked, need to throw exception 1585 __ MacroAssembler::call_VM(noreg, 1586 CAST_FROM_FN_PTR(address, 1587 InterpreterRuntime::throw_illegal_monitor_state_exception)); 1588 __ should_not_reach_here(); 1589 1590 __ bind(unlock); 1591 __ unlock_object(c_rarg1); 1592 } 1593 __ bind(L); 1594 } 1595 1596 #if INCLUDE_JFR 1597 __ enter_jfr_critical_section(); 1598 1599 // This poll test is to uphold the invariant that a JFR sampled frame 1600 // must not return to its caller without a prior safepoint poll check. 1601 // The earlier poll check in this routine is insufficient for this purpose 1602 // because the thread has transitioned back to Java. 1603 1604 Label slow_path; 1605 Label fast_path; 1606 __ safepoint_poll(slow_path, true /* at_return */, false /* in_nmethod */); 1607 __ br(Assembler::AL, fast_path); 1608 __ bind(slow_path); 1609 __ push(dtos); 1610 __ push(ltos); 1611 __ set_last_Java_frame(esp, rfp, __ pc(), rscratch1); 1612 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread); 1613 __ reset_last_Java_frame(true); 1614 __ pop(ltos); 1615 __ pop(dtos); 1616 __ bind(fast_path); 1617 1618 #endif // INCLUDE_JFR 1619 1620 // jvmti support 1621 // Note: This must happen _after_ handling/throwing any exceptions since 1622 // the exception handler code notifies the runtime of method exits 1623 // too. If this happens before, method entry/exit notifications are 1624 // not properly paired (was bug - gri 11/22/99). 1625 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); 1626 1627 // restore potential result in r0:d0, call result handler to 1628 // restore potential result in ST0 & handle result 1629 1630 __ pop(ltos); 1631 __ pop(dtos); 1632 1633 __ blr(result_handler); 1634 1635 // remove activation 1636 __ ldr(esp, Address(rfp, 1637 frame::interpreter_frame_sender_sp_offset * 1638 wordSize)); // get sender sp 1639 // remove frame anchor 1640 __ leave(); 1641 1642 JFR_ONLY(__ leave_jfr_critical_section();) 1643 1644 // restore sender sp 1645 __ mov(sp, esp); 1646 1647 __ ret(lr); 1648 1649 if (inc_counter) { 1650 // Handle overflow of counter and compile method 1651 __ bind(invocation_counter_overflow); 1652 generate_counter_overflow(continue_after_compile); 1653 } 1654 1655 return entry_point; 1656 } 1657 1658 // 1659 // Generic interpreted method entry to (asm) interpreter 1660 // 1661 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) { 1662 // determine code generation flags 1663 bool inc_counter = UseCompiler || CountCompiledCalls; 1664 1665 // rscratch1: sender sp 1666 address entry_point = __ pc(); 1667 1668 const Address constMethod(rmethod, Method::const_offset()); 1669 const Address access_flags(rmethod, Method::access_flags_offset()); 1670 const Address size_of_parameters(r3, 1671 ConstMethod::size_of_parameters_offset()); 1672 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset()); 1673 1674 // get parameter size (always needed) 1675 // need to load the const method first 1676 __ ldr(r3, constMethod); 1677 __ load_unsigned_short(r2, size_of_parameters); 1678 1679 // r2: size of parameters 1680 1681 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words 1682 __ sub(r3, r3, r2); // r3 = no. of additional locals 1683 1684 // see if we've got enough room on the stack for locals plus overhead. 1685 generate_stack_overflow_check(); 1686 1687 // compute beginning of parameters (rlocals) 1688 __ add(rlocals, esp, r2, ext::uxtx, 3); 1689 __ sub(rlocals, rlocals, wordSize); 1690 1691 __ mov(rscratch1, esp); 1692 1693 // r3 - # of additional locals 1694 // allocate space for locals 1695 // explicitly initialize locals 1696 // Initializing memory allocated for locals in the same direction as 1697 // the stack grows to ensure page initialization order according 1698 // to windows-aarch64 stack page growth requirement (see 1699 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack) 1700 { 1701 Label exit, loop; 1702 __ ands(zr, r3, r3); 1703 __ br(Assembler::LE, exit); // do nothing if r3 <= 0 1704 __ bind(loop); 1705 __ str(zr, Address(__ pre(rscratch1, -wordSize))); 1706 __ sub(r3, r3, 1); // until everything initialized 1707 __ cbnz(r3, loop); 1708 __ bind(exit); 1709 } 1710 1711 // Padding between locals and fixed part of activation frame to ensure 1712 // SP is always 16-byte aligned. 1713 __ andr(sp, rscratch1, -16); 1714 1715 // And the base dispatch table 1716 __ get_dispatch(); 1717 1718 // initialize fixed part of activation frame 1719 generate_fixed_frame(false); 1720 1721 // make sure method is not native & not abstract 1722 #ifdef ASSERT 1723 __ ldrh(r0, access_flags); 1724 { 1725 Label L; 1726 __ tst(r0, JVM_ACC_NATIVE); 1727 __ br(Assembler::EQ, L); 1728 __ stop("tried to execute native method as non-native"); 1729 __ bind(L); 1730 } 1731 { 1732 Label L; 1733 __ tst(r0, JVM_ACC_ABSTRACT); 1734 __ br(Assembler::EQ, L); 1735 __ stop("tried to execute abstract method in interpreter"); 1736 __ bind(L); 1737 } 1738 #endif 1739 1740 // Since at this point in the method invocation the exception 1741 // handler would try to exit the monitor of synchronized methods 1742 // which hasn't been entered yet, we set the thread local variable 1743 // _do_not_unlock_if_synchronized to true. The remove_activation 1744 // will check this flag. 1745 1746 const Address do_not_unlock_if_synchronized(rthread, 1747 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1748 __ mov(rscratch2, true); 1749 __ strb(rscratch2, do_not_unlock_if_synchronized); 1750 1751 Register mdp = r3; 1752 __ profile_parameters_type(mdp, r1, r2); 1753 1754 // increment invocation count & check for overflow 1755 Label invocation_counter_overflow; 1756 if (inc_counter) { 1757 generate_counter_incr(&invocation_counter_overflow); 1758 } 1759 1760 Label continue_after_compile; 1761 __ bind(continue_after_compile); 1762 1763 bang_stack_shadow_pages(false); 1764 1765 // reset the _do_not_unlock_if_synchronized flag 1766 __ strb(zr, do_not_unlock_if_synchronized); 1767 1768 // check for synchronized methods 1769 // Must happen AFTER invocation_counter check and stack overflow check, 1770 // so method is not locked if overflows. 1771 if (synchronized) { 1772 // Allocate monitor and lock method 1773 lock_method(); 1774 } else { 1775 // no synchronization necessary 1776 #ifdef ASSERT 1777 { 1778 Label L; 1779 __ ldrh(r0, access_flags); 1780 __ tst(r0, JVM_ACC_SYNCHRONIZED); 1781 __ br(Assembler::EQ, L); 1782 __ stop("method needs synchronization"); 1783 __ bind(L); 1784 } 1785 #endif 1786 } 1787 1788 // start execution 1789 #ifdef ASSERT 1790 { 1791 Label L; 1792 const Address monitor_block_top (rfp, 1793 frame::interpreter_frame_monitor_block_top_offset * wordSize); 1794 __ ldr(rscratch1, monitor_block_top); 1795 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize))); 1796 __ cmp(esp, rscratch1); 1797 __ br(Assembler::EQ, L); 1798 __ stop("broken stack frame setup in interpreter 2"); 1799 __ bind(L); 1800 } 1801 #endif 1802 1803 // jvmti support 1804 __ notify_method_entry(); 1805 1806 __ dispatch_next(vtos); 1807 1808 // invocation counter overflow 1809 if (inc_counter) { 1810 // Handle overflow of counter and compile method 1811 __ bind(invocation_counter_overflow); 1812 generate_counter_overflow(continue_after_compile); 1813 } 1814 1815 return entry_point; 1816 } 1817 1818 // Method entry for java.lang.Thread.currentThread 1819 address TemplateInterpreterGenerator::generate_currentThread() { 1820 address entry_point = __ pc(); 1821 1822 __ ldr(r0, Address(rthread, JavaThread::vthread_offset())); 1823 __ resolve_oop_handle(r0, rscratch1, rscratch2); 1824 __ ret(lr); 1825 1826 return entry_point; 1827 } 1828 1829 //----------------------------------------------------------------------------- 1830 // Exceptions 1831 1832 void TemplateInterpreterGenerator::generate_throw_exception() { 1833 // Entry point in previous activation (i.e., if the caller was 1834 // interpreted) 1835 Interpreter::_rethrow_exception_entry = __ pc(); 1836 // Restore sp to interpreter_frame_last_sp even though we are going 1837 // to empty the expression stack for the exception processing. 1838 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1839 // r0: exception 1840 // r3: return address/pc that threw exception 1841 __ restore_bcp(); // rbcp points to call/send 1842 __ restore_locals(); 1843 __ restore_constant_pool_cache(); 1844 __ reinit_heapbase(); // restore rheapbase as heapbase. 1845 __ get_dispatch(); 1846 1847 // Entry point for exceptions thrown within interpreter code 1848 Interpreter::_throw_exception_entry = __ pc(); 1849 // If we came here via a NullPointerException on the receiver of a 1850 // method, rmethod may be corrupt. 1851 __ get_method(rmethod); 1852 // expression stack is undefined here 1853 // r0: exception 1854 // rbcp: exception bcp 1855 __ verify_oop(r0); 1856 __ mov(c_rarg1, r0); 1857 1858 // expression stack must be empty before entering the VM in case of 1859 // an exception 1860 __ empty_expression_stack(); 1861 // find exception handler address and preserve exception oop 1862 __ call_VM(r3, 1863 CAST_FROM_FN_PTR(address, 1864 InterpreterRuntime::exception_handler_for_exception), 1865 c_rarg1); 1866 1867 // Restore machine SP 1868 __ restore_sp_after_call(); 1869 1870 // r0: exception handler entry point 1871 // r3: preserved exception oop 1872 // rbcp: bcp for exception handler 1873 __ push_ptr(r3); // push exception which is now the only value on the stack 1874 __ br(r0); // jump to exception handler (may be _remove_activation_entry!) 1875 1876 // If the exception is not handled in the current frame the frame is 1877 // removed and the exception is rethrown (i.e. exception 1878 // continuation is _rethrow_exception). 1879 // 1880 // Note: At this point the bci is still the bxi for the instruction 1881 // which caused the exception and the expression stack is 1882 // empty. Thus, for any VM calls at this point, GC will find a legal 1883 // oop map (with empty expression stack). 1884 1885 // 1886 // JVMTI PopFrame support 1887 // 1888 1889 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 1890 __ empty_expression_stack(); 1891 __ restore_bcp(); // We could have returned from deoptimizing this frame, so restore rbcp. 1892 // Set the popframe_processing bit in pending_popframe_condition 1893 // indicating that we are currently handling popframe, so that 1894 // call_VMs that may happen later do not trigger new popframe 1895 // handling cycles. 1896 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset())); 1897 __ orr(r3, r3, JavaThread::popframe_processing_bit); 1898 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset())); 1899 1900 { 1901 // Check to see whether we are returning to a deoptimized frame. 1902 // (The PopFrame call ensures that the caller of the popped frame is 1903 // either interpreted or compiled and deoptimizes it if compiled.) 1904 // In this case, we can't call dispatch_next() after the frame is 1905 // popped, but instead must save the incoming arguments and restore 1906 // them after deoptimization has occurred. 1907 // 1908 // Note that we don't compare the return PC against the 1909 // deoptimization blob's unpack entry because of the presence of 1910 // adapter frames in C2. 1911 Label caller_not_deoptimized; 1912 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize)); 1913 // This is a return address, so requires authenticating for PAC. 1914 __ authenticate_return_address(c_rarg1); 1915 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1916 InterpreterRuntime::interpreter_contains), c_rarg1); 1917 __ cbnz(r0, caller_not_deoptimized); 1918 1919 // Compute size of arguments for saving when returning to 1920 // deoptimized caller 1921 __ get_method(r0); 1922 __ ldr(r0, Address(r0, Method::const_offset())); 1923 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod:: 1924 size_of_parameters_offset()))); 1925 __ lsl(r0, r0, Interpreter::logStackElementSize); 1926 __ restore_locals(); // XXX do we need this? 1927 __ sub(rlocals, rlocals, r0); 1928 __ add(rlocals, rlocals, wordSize); 1929 // Save these arguments 1930 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1931 Deoptimization:: 1932 popframe_preserve_args), 1933 rthread, r0, rlocals); 1934 1935 __ remove_activation(vtos, 1936 /* throw_monitor_exception */ false, 1937 /* install_monitor_exception */ false, 1938 /* notify_jvmdi */ false); 1939 1940 // Inform deoptimization that it is responsible for restoring 1941 // these arguments 1942 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit); 1943 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset())); 1944 1945 // Continue in deoptimization handler 1946 __ ret(lr); 1947 1948 __ bind(caller_not_deoptimized); 1949 } 1950 1951 __ remove_activation(vtos, 1952 /* throw_monitor_exception */ false, 1953 /* install_monitor_exception */ false, 1954 /* notify_jvmdi */ false); 1955 1956 // Restore the last_sp and null it out 1957 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1958 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize))); 1959 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1960 1961 __ restore_bcp(); 1962 __ restore_locals(); 1963 __ restore_constant_pool_cache(); 1964 __ get_method(rmethod); 1965 __ get_dispatch(); 1966 1967 // The method data pointer was incremented already during 1968 // call profiling. We have to restore the mdp for the current bcp. 1969 if (ProfileInterpreter) { 1970 __ set_method_data_pointer_for_bcp(); 1971 } 1972 1973 // Clear the popframe condition flag 1974 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset())); 1975 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive"); 1976 1977 #if INCLUDE_JVMTI 1978 { 1979 Label L_done; 1980 1981 __ ldrb(rscratch1, Address(rbcp, 0)); 1982 __ cmpw(rscratch1, Bytecodes::_invokestatic); 1983 __ br(Assembler::NE, L_done); 1984 1985 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. 1986 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null. 1987 1988 __ ldr(c_rarg0, Address(rlocals, 0)); 1989 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp); 1990 1991 __ cbz(r0, L_done); 1992 1993 __ str(r0, Address(esp, 0)); 1994 __ bind(L_done); 1995 } 1996 #endif // INCLUDE_JVMTI 1997 1998 // Restore machine SP 1999 __ restore_sp_after_call(); 2000 2001 __ dispatch_next(vtos); 2002 // end of PopFrame support 2003 2004 Interpreter::_remove_activation_entry = __ pc(); 2005 2006 // preserve exception over this code sequence 2007 __ pop_ptr(r0); 2008 __ str(r0, Address(rthread, JavaThread::vm_result_oop_offset())); 2009 // remove the activation (without doing throws on illegalMonitorExceptions) 2010 __ remove_activation(vtos, false, true, false); 2011 // restore exception 2012 __ get_vm_result_oop(r0, rthread); 2013 2014 // In between activations - previous activation type unknown yet 2015 // compute continuation point - the continuation point expects the 2016 // following registers set up: 2017 // 2018 // r0: exception 2019 // lr: return address/pc that threw exception 2020 // esp: expression stack of caller 2021 // rfp: fp of caller 2022 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address 2023 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 2024 SharedRuntime::exception_handler_for_return_address), 2025 rthread, lr); 2026 __ mov(r1, r0); // save exception handler 2027 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address 2028 // We might be returning to a deopt handler that expects r3 to 2029 // contain the exception pc 2030 __ mov(r3, lr); 2031 // Note that an "issuing PC" is actually the next PC after the call 2032 __ br(r1); // jump to exception 2033 // handler of caller 2034 } 2035 2036 2037 // 2038 // JVMTI ForceEarlyReturn support 2039 // 2040 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 2041 address entry = __ pc(); 2042 2043 __ restore_bcp(); 2044 __ restore_locals(); 2045 __ empty_expression_stack(); 2046 __ load_earlyret_value(state); 2047 2048 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset())); 2049 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset()); 2050 2051 // Clear the earlyret state 2052 assert(JvmtiThreadState::earlyret_inactive == 0, "should be"); 2053 __ str(zr, cond_addr); 2054 2055 __ remove_activation(state, 2056 false, /* throw_monitor_exception */ 2057 false, /* install_monitor_exception */ 2058 true); /* notify_jvmdi */ 2059 __ ret(lr); 2060 2061 return entry; 2062 } // end of ForceEarlyReturn support 2063 2064 2065 2066 //----------------------------------------------------------------------------- 2067 // Helper for vtos entry point generation 2068 2069 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, 2070 address& bep, 2071 address& cep, 2072 address& sep, 2073 address& aep, 2074 address& iep, 2075 address& lep, 2076 address& fep, 2077 address& dep, 2078 address& vep) { 2079 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 2080 Label L; 2081 aep = __ pc(); // atos entry point 2082 __ push_ptr(); 2083 __ b(L); 2084 fep = __ pc(); // ftos entry point 2085 __ push_f(); 2086 __ b(L); 2087 dep = __ pc(); // dtos entry point 2088 __ push_d(); 2089 __ b(L); 2090 lep = __ pc(); // ltos entry point 2091 __ push_l(); 2092 __ b(L); 2093 bep = cep = sep = iep = __ pc(); // [bcsi]tos entry point 2094 __ push_i(); 2095 vep = __ pc(); // vtos entry point 2096 __ bind(L); 2097 generate_and_dispatch(t); 2098 } 2099 2100 //----------------------------------------------------------------------------- 2101 2102 // Non-product code 2103 #ifndef PRODUCT 2104 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 2105 address entry = __ pc(); 2106 2107 __ protect_return_address(); 2108 __ push(lr); 2109 __ push(state); 2110 __ push(RegSet::range(r0, r15), sp); 2111 __ mov(c_rarg2, r0); // Pass itos 2112 __ call_VM(noreg, 2113 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), 2114 c_rarg1, c_rarg2, c_rarg3); 2115 __ pop(RegSet::range(r0, r15), sp); 2116 __ pop(state); 2117 __ pop(lr); 2118 __ authenticate_return_address(); 2119 __ ret(lr); // return from result handler 2120 2121 return entry; 2122 } 2123 2124 void TemplateInterpreterGenerator::count_bytecode() { 2125 __ mov(r10, (address) &BytecodeCounter::_counter_value); 2126 __ atomic_add(noreg, 1, r10); 2127 } 2128 2129 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 2130 __ mov(r10, (address) &BytecodeHistogram::_counters[t->bytecode()]); 2131 __ atomic_addw(noreg, 1, r10); 2132 } 2133 2134 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 2135 // Calculate new index for counter: 2136 // _index = (_index >> log2_number_of_codes) | 2137 // (bytecode << log2_number_of_codes); 2138 Register index_addr = rscratch1; 2139 Register index = rscratch2; 2140 __ mov(index_addr, (address) &BytecodePairHistogram::_index); 2141 __ ldrw(index, index_addr); 2142 __ mov(r10, 2143 ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); 2144 __ orrw(index, r10, index, Assembler::LSR, 2145 BytecodePairHistogram::log2_number_of_codes); 2146 __ strw(index, index_addr); 2147 2148 // Bump bucket contents: 2149 // _counters[_index] ++; 2150 Register counter_addr = rscratch1; 2151 __ mov(r10, (address) &BytecodePairHistogram::_counters); 2152 __ lea(counter_addr, Address(r10, index, Address::lsl(LogBytesPerInt))); 2153 __ atomic_addw(noreg, 1, counter_addr); 2154 } 2155 2156 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 2157 // Call a little run-time stub to avoid blow-up for each bytecode. 2158 // The run-time runtime saves the right registers, depending on 2159 // the tosca in-state for the given template. 2160 2161 assert(Interpreter::trace_code(t->tos_in()) != nullptr, 2162 "entry must have been generated"); 2163 __ bl(RuntimeAddress(Interpreter::trace_code(t->tos_in()))); 2164 __ reinit_heapbase(); 2165 } 2166 2167 2168 void TemplateInterpreterGenerator::stop_interpreter_at() { 2169 Label L; 2170 __ push(rscratch1); 2171 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value); 2172 __ ldr(rscratch1, Address(rscratch1)); 2173 __ mov(rscratch2, StopInterpreterAt); 2174 __ cmp(rscratch1, rscratch2); 2175 __ br(Assembler::NE, L); 2176 __ brk(0); 2177 __ bind(L); 2178 __ pop(rscratch1); 2179 } 2180 2181 #endif // !PRODUCT