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