1 /* 2 * Copyright (c) 2003, 2022, 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/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/methodData.hpp" 41 #include "oops/method.hpp" 42 #include "oops/oop.inline.hpp" 43 #include "oops/inlineKlass.hpp" 44 #include "prims/jvmtiExport.hpp" 45 #include "prims/jvmtiThreadState.hpp" 46 #include "runtime/arguments.hpp" 47 #include "runtime/deoptimization.hpp" 48 #include "runtime/frame.inline.hpp" 49 #include "runtime/jniHandles.hpp" 50 #include "runtime/sharedRuntime.hpp" 51 #include "runtime/stubRoutines.hpp" 52 #include "runtime/synchronizer.hpp" 53 #include "runtime/timer.hpp" 54 #include "runtime/vframeArray.hpp" 55 #include "utilities/debug.hpp" 56 #include "utilities/powerOfTwo.hpp" 57 #include <sys/types.h> 58 59 #ifndef PRODUCT 60 #include "oops/method.hpp" 61 #endif // !PRODUCT 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 __ _masm-> 71 72 //----------------------------------------------------------------------------- 73 74 extern "C" void entry(CodeBuffer*); 75 76 //----------------------------------------------------------------------------- 77 78 address TemplateInterpreterGenerator::generate_slow_signature_handler() { 79 address entry = __ pc(); 80 81 __ andr(esp, esp, -16); 82 __ mov(c_rarg3, esp); 83 // rmethod 84 // rlocals 85 // c_rarg3: first stack arg - wordSize 86 87 // adjust sp 88 __ sub(sp, c_rarg3, 18 * wordSize); 89 __ str(lr, Address(__ pre(sp, -2 * wordSize))); 90 __ call_VM(noreg, 91 CAST_FROM_FN_PTR(address, 92 InterpreterRuntime::slow_signature_handler), 93 rmethod, rlocals, c_rarg3); 94 95 // r0: result handler 96 97 // Stack layout: 98 // rsp: return address <- sp 99 // 1 garbage 100 // 8 integer args (if static first is unused) 101 // 1 float/double identifiers 102 // 8 double args 103 // stack args <- esp 104 // garbage 105 // expression stack bottom 106 // bcp (NULL) 107 // ... 108 109 // Restore LR 110 __ ldr(lr, Address(__ post(sp, 2 * wordSize))); 111 112 // Do FP first so we can use c_rarg3 as temp 113 __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers 114 115 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) { 116 const FloatRegister r = as_FloatRegister(i); 117 118 Label d, done; 119 120 __ tbnz(c_rarg3, i, d); 121 __ ldrs(r, Address(sp, (10 + i) * wordSize)); 122 __ b(done); 123 __ bind(d); 124 __ ldrd(r, Address(sp, (10 + i) * wordSize)); 125 __ bind(done); 126 } 127 128 // c_rarg0 contains the result from the call of 129 // InterpreterRuntime::slow_signature_handler so we don't touch it 130 // here. It will be loaded with the JNIEnv* later. 131 __ ldr(c_rarg1, Address(sp, 1 * wordSize)); 132 for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) { 133 Register rm = as_Register(i), rn = as_Register(i+1); 134 __ ldp(rm, rn, Address(sp, i * wordSize)); 135 } 136 137 __ add(sp, sp, 18 * wordSize); 138 __ ret(lr); 139 140 return entry; 141 } 142 143 144 // 145 // Various method entries 146 // 147 148 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) { 149 // rmethod: Method* 150 // r13: sender sp 151 // esp: args 152 153 if (!InlineIntrinsics) return NULL; // Generate a vanilla entry 154 155 // These don't need a safepoint check because they aren't virtually 156 // callable. We won't enter these intrinsics from compiled code. 157 // If in the future we added an intrinsic which was virtually callable 158 // we'd have to worry about how to safepoint so that this code is used. 159 160 // mathematical functions inlined by compiler 161 // (interpreter must provide identical implementation 162 // in order to avoid monotonicity bugs when switching 163 // from interpreter to compiler in the middle of some 164 // computation) 165 // 166 // stack: 167 // [ arg ] <-- esp 168 // [ arg ] 169 // retaddr in lr 170 171 address entry_point = NULL; 172 Register continuation = lr; 173 switch (kind) { 174 case Interpreter::java_lang_math_abs: 175 entry_point = __ pc(); 176 __ ldrd(v0, Address(esp)); 177 __ fabsd(v0, v0); 178 __ mov(sp, r13); // Restore caller's SP 179 break; 180 case Interpreter::java_lang_math_sqrt: 181 entry_point = __ pc(); 182 __ ldrd(v0, Address(esp)); 183 __ fsqrtd(v0, v0); 184 __ mov(sp, r13); 185 break; 186 case Interpreter::java_lang_math_sin : 187 case Interpreter::java_lang_math_cos : 188 case Interpreter::java_lang_math_tan : 189 case Interpreter::java_lang_math_log : 190 case Interpreter::java_lang_math_log10 : 191 case Interpreter::java_lang_math_exp : 192 entry_point = __ pc(); 193 __ ldrd(v0, Address(esp)); 194 __ mov(sp, r13); 195 __ mov(r19, lr); 196 continuation = r19; // The first callee-saved register 197 generate_transcendental_entry(kind, 1); 198 break; 199 case Interpreter::java_lang_math_pow : 200 entry_point = __ pc(); 201 __ mov(r19, lr); 202 continuation = r19; 203 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize)); 204 __ ldrd(v1, Address(esp)); 205 __ mov(sp, r13); 206 generate_transcendental_entry(kind, 2); 207 break; 208 case Interpreter::java_lang_math_fmaD : 209 if (UseFMA) { 210 entry_point = __ pc(); 211 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize)); 212 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize)); 213 __ ldrd(v2, Address(esp)); 214 __ fmaddd(v0, v0, v1, v2); 215 __ mov(sp, r13); // Restore caller's SP 216 } 217 break; 218 case Interpreter::java_lang_math_fmaF : 219 if (UseFMA) { 220 entry_point = __ pc(); 221 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize)); 222 __ ldrs(v1, Address(esp, Interpreter::stackElementSize)); 223 __ ldrs(v2, Address(esp)); 224 __ fmadds(v0, v0, v1, v2); 225 __ mov(sp, r13); // Restore caller's SP 226 } 227 break; 228 default: 229 ; 230 } 231 if (entry_point) { 232 __ br(continuation); 233 } 234 235 return entry_point; 236 } 237 238 // double trigonometrics and transcendentals 239 // static jdouble dsin(jdouble x); 240 // static jdouble dcos(jdouble x); 241 // static jdouble dtan(jdouble x); 242 // static jdouble dlog(jdouble x); 243 // static jdouble dlog10(jdouble x); 244 // static jdouble dexp(jdouble x); 245 // static jdouble dpow(jdouble x, jdouble y); 246 247 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) { 248 address fn; 249 switch (kind) { 250 case Interpreter::java_lang_math_sin : 251 if (StubRoutines::dsin() == NULL) { 252 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); 253 } else { 254 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin()); 255 } 256 break; 257 case Interpreter::java_lang_math_cos : 258 if (StubRoutines::dcos() == NULL) { 259 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); 260 } else { 261 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos()); 262 } 263 break; 264 case Interpreter::java_lang_math_tan : 265 if (StubRoutines::dtan() == NULL) { 266 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); 267 } else { 268 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan()); 269 } 270 break; 271 case Interpreter::java_lang_math_log : 272 if (StubRoutines::dlog() == NULL) { 273 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); 274 } else { 275 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog()); 276 } 277 break; 278 case Interpreter::java_lang_math_log10 : 279 if (StubRoutines::dlog10() == NULL) { 280 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); 281 } else { 282 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10()); 283 } 284 break; 285 case Interpreter::java_lang_math_exp : 286 if (StubRoutines::dexp() == NULL) { 287 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); 288 } else { 289 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp()); 290 } 291 break; 292 case Interpreter::java_lang_math_pow : 293 if (StubRoutines::dpow() == NULL) { 294 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); 295 } else { 296 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow()); 297 } 298 break; 299 default: 300 ShouldNotReachHere(); 301 fn = NULL; // unreachable 302 } 303 __ mov(rscratch1, fn); 304 __ blr(rscratch1); 305 } 306 307 // Abstract method entry 308 // Attempt to execute abstract method. Throw exception 309 address TemplateInterpreterGenerator::generate_abstract_entry(void) { 310 // rmethod: Method* 311 // r13: sender SP 312 313 address entry_point = __ pc(); 314 315 // abstract method entry 316 317 // pop return address, reset last_sp to NULL 318 __ empty_expression_stack(); 319 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed) 320 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 321 322 // throw exception 323 __ call_VM(noreg, CAST_FROM_FN_PTR(address, 324 InterpreterRuntime::throw_AbstractMethodErrorWithMethod), 325 rmethod); 326 // the call_VM checks for exception, so we should never return here. 327 __ should_not_reach_here(); 328 329 return entry_point; 330 } 331 332 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 333 address entry = __ pc(); 334 335 #ifdef ASSERT 336 { 337 Label L; 338 __ ldr(rscratch1, Address(rfp, 339 frame::interpreter_frame_monitor_block_top_offset * 340 wordSize)); 341 __ mov(rscratch2, sp); 342 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack 343 // grows negative) 344 __ br(Assembler::HS, L); // check if frame is complete 345 __ stop ("interpreter frame not set up"); 346 __ bind(L); 347 } 348 #endif // ASSERT 349 // Restore bcp under the assumption that the current frame is still 350 // interpreted 351 __ restore_bcp(); 352 353 // expression stack must be empty before entering the VM if an 354 // exception happened 355 __ empty_expression_stack(); 356 // throw exception 357 __ call_VM(noreg, 358 CAST_FROM_FN_PTR(address, 359 InterpreterRuntime::throw_StackOverflowError)); 360 return entry; 361 } 362 363 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() { 364 address entry = __ pc(); 365 // expression stack must be empty before entering the VM if an 366 // exception happened 367 __ empty_expression_stack(); 368 // setup parameters 369 370 // ??? convention: expect aberrant index in register r1 371 __ movw(c_rarg2, r1); 372 // ??? convention: expect array in register r3 373 __ mov(c_rarg1, r3); 374 __ call_VM(noreg, 375 CAST_FROM_FN_PTR(address, 376 InterpreterRuntime:: 377 throw_ArrayIndexOutOfBoundsException), 378 c_rarg1, c_rarg2); 379 return entry; 380 } 381 382 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 383 address entry = __ pc(); 384 385 // object is at TOS 386 __ pop(c_rarg1); 387 388 // expression stack must be empty before entering the VM if an 389 // exception happened 390 __ empty_expression_stack(); 391 392 __ call_VM(noreg, 393 CAST_FROM_FN_PTR(address, 394 InterpreterRuntime:: 395 throw_ClassCastException), 396 c_rarg1); 397 return entry; 398 } 399 400 address TemplateInterpreterGenerator::generate_exception_handler_common( 401 const char* name, const char* message, bool pass_oop) { 402 assert(!pass_oop || message == NULL, "either oop or message but not both"); 403 address entry = __ pc(); 404 if (pass_oop) { 405 // object is at TOS 406 __ pop(c_rarg2); 407 } 408 // expression stack must be empty before entering the VM if an 409 // exception happened 410 __ empty_expression_stack(); 411 // setup parameters 412 __ lea(c_rarg1, Address((address)name)); 413 if (pass_oop) { 414 __ call_VM(r0, CAST_FROM_FN_PTR(address, 415 InterpreterRuntime:: 416 create_klass_exception), 417 c_rarg1, c_rarg2); 418 } else { 419 // kind of lame ExternalAddress can't take NULL because 420 // external_word_Relocation will assert. 421 if (message != NULL) { 422 __ lea(c_rarg2, Address((address)message)); 423 } else { 424 __ mov(c_rarg2, NULL_WORD); 425 } 426 __ call_VM(r0, 427 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), 428 c_rarg1, c_rarg2); 429 } 430 // throw exception 431 __ b(address(Interpreter::throw_exception_entry())); 432 return entry; 433 } 434 435 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) { 436 address entry = __ pc(); 437 438 // Restore stack bottom in case i2c adjusted stack 439 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 440 // and NULL it as marker that esp is now tos until next java call 441 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 442 443 if (state == atos && InlineTypeReturnedAsFields) { 444 __ store_inline_type_fields_to_buf(NULL, true); 445 } 446 447 __ restore_bcp(); 448 __ restore_locals(); 449 __ restore_constant_pool_cache(); 450 __ get_method(rmethod); 451 452 if (state == atos) { 453 Register obj = r0; 454 Register mdp = r1; 455 Register tmp = r2; 456 __ profile_return_type(mdp, obj, tmp); 457 } 458 459 // Pop N words from the stack 460 __ get_cache_and_index_at_bcp(r1, r2, 1, index_size); 461 __ ldr(r1, Address(r1, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset())); 462 __ andr(r1, r1, ConstantPoolCacheEntry::parameter_size_mask); 463 464 __ add(esp, esp, r1, Assembler::LSL, 3); 465 466 // Restore machine SP 467 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); 468 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset())); 469 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2); 470 __ ldr(rscratch2, 471 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize)); 472 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3); 473 __ andr(sp, rscratch1, -16); 474 475 __ check_and_handle_popframe(rthread); 476 __ check_and_handle_earlyret(rthread); 477 478 __ get_dispatch(); 479 __ dispatch_next(state, step); 480 481 return entry; 482 } 483 484 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, 485 int step, 486 address continuation) { 487 address entry = __ pc(); 488 __ restore_bcp(); 489 __ restore_locals(); 490 __ restore_constant_pool_cache(); 491 __ get_method(rmethod); 492 __ get_dispatch(); 493 494 // Calculate stack limit 495 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); 496 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset())); 497 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2); 498 __ ldr(rscratch2, 499 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize)); 500 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3); 501 __ andr(sp, rscratch1, -16); 502 503 // Restore expression stack pointer 504 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 505 // NULL last_sp until next java call 506 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 507 508 #if INCLUDE_JVMCI 509 // Check if we need to take lock at entry of synchronized method. This can 510 // only occur on method entry so emit it only for vtos with step 0. 511 if (EnableJVMCI && state == vtos && step == 0) { 512 Label L; 513 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset())); 514 __ cbz(rscratch1, L); 515 // Clear flag. 516 __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset())); 517 // Take lock. 518 lock_method(); 519 __ bind(L); 520 } else { 521 #ifdef ASSERT 522 if (EnableJVMCI) { 523 Label L; 524 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset())); 525 __ cbz(rscratch1, L); 526 __ stop("unexpected pending monitor in deopt entry"); 527 __ bind(L); 528 } 529 #endif 530 } 531 #endif 532 // handle exceptions 533 { 534 Label L; 535 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset())); 536 __ cbz(rscratch1, L); 537 __ call_VM(noreg, 538 CAST_FROM_FN_PTR(address, 539 InterpreterRuntime::throw_pending_exception)); 540 __ should_not_reach_here(); 541 __ bind(L); 542 } 543 544 if (continuation == NULL) { 545 __ dispatch_next(state, step); 546 } else { 547 __ jump_to_entry(continuation); 548 } 549 return entry; 550 } 551 552 address TemplateInterpreterGenerator::generate_result_handler_for( 553 BasicType type) { 554 address entry = __ pc(); 555 switch (type) { 556 case T_BOOLEAN: __ c2bool(r0); break; 557 case T_CHAR : __ uxth(r0, r0); break; 558 case T_BYTE : __ sxtb(r0, r0); break; 559 case T_SHORT : __ sxth(r0, r0); break; 560 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this 561 case T_LONG : /* nothing to do */ break; 562 case T_VOID : /* nothing to do */ break; 563 case T_FLOAT : /* nothing to do */ break; 564 case T_DOUBLE : /* nothing to do */ break; 565 case T_PRIMITIVE_OBJECT: // fall through (value types are handled with oops) 566 case T_OBJECT : 567 // retrieve result from frame 568 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize)); 569 // and verify it 570 __ verify_oop(r0); 571 break; 572 default : ShouldNotReachHere(); 573 } 574 __ ret(lr); // return from result handler 575 return entry; 576 } 577 578 address TemplateInterpreterGenerator::generate_safept_entry_for( 579 TosState state, 580 address runtime_entry) { 581 address entry = __ pc(); 582 __ push(state); 583 __ call_VM(noreg, runtime_entry); 584 __ membar(Assembler::AnyAny); 585 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 586 return entry; 587 } 588 589 // Helpers for commoning out cases in the various type of method entries. 590 // 591 592 593 // increment invocation count & check for overflow 594 // 595 // Note: checking for negative value instead of overflow 596 // so we have a 'sticky' overflow test 597 // 598 // rmethod: method 599 // 600 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) { 601 Label done; 602 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not. 603 int increment = InvocationCounter::count_increment; 604 Label no_mdo; 605 if (ProfileInterpreter) { 606 // Are we profiling? 607 __ ldr(r0, Address(rmethod, Method::method_data_offset())); 608 __ cbz(r0, no_mdo); 609 // Increment counter in the MDO 610 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) + 611 in_bytes(InvocationCounter::counter_offset())); 612 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset())); 613 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow); 614 __ b(done); 615 } 616 __ bind(no_mdo); 617 // Increment counter in MethodCounters 618 const Address invocation_counter(rscratch2, 619 MethodCounters::invocation_counter_offset() + 620 InvocationCounter::counter_offset()); 621 __ get_method_counters(rmethod, rscratch2, done); 622 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset())); 623 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow); 624 __ bind(done); 625 } 626 627 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) { 628 629 // Asm interpreter on entry 630 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] 631 // Everything as it was on entry 632 633 // InterpreterRuntime::frequency_counter_overflow takes two 634 // arguments, the first (thread) is passed by call_VM, the second 635 // indicates if the counter overflow occurs at a backwards branch 636 // (NULL bcp). We pass zero for it. The call returns the address 637 // of the verified entry point for the method or NULL if the 638 // compilation did not complete (either went background or bailed 639 // out). 640 __ mov(c_rarg1, 0); 641 __ call_VM(noreg, 642 CAST_FROM_FN_PTR(address, 643 InterpreterRuntime::frequency_counter_overflow), 644 c_rarg1); 645 646 __ b(do_continue); 647 } 648 649 // See if we've got enough room on the stack for locals plus overhead 650 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError 651 // without going through the signal handler, i.e., reserved and yellow zones 652 // will not be made usable. The shadow zone must suffice to handle the 653 // overflow. 654 // The expression stack grows down incrementally, so the normal guard 655 // page mechanism will work for that. 656 // 657 // NOTE: Since the additional locals are also always pushed (wasn't 658 // obvious in generate_method_entry) so the guard should work for them 659 // too. 660 // 661 // Args: 662 // r3: number of additional locals this frame needs (what we must check) 663 // rmethod: Method* 664 // 665 // Kills: 666 // r0 667 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) { 668 669 // monitor entry size: see picture of stack set 670 // (generate_method_entry) and frame_amd64.hpp 671 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 672 673 // total overhead size: entry_size + (saved rbp through expr stack 674 // bottom). be sure to change this if you add/subtract anything 675 // to/from the overhead area 676 const int overhead_size = 677 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size; 678 679 const int page_size = os::vm_page_size(); 680 681 Label after_frame_check; 682 683 // see if the frame is greater than one page in size. If so, 684 // then we need to verify there is enough stack space remaining 685 // for the additional locals. 686 // 687 // Note that we use SUBS rather than CMP here because the immediate 688 // field of this instruction may overflow. SUBS can cope with this 689 // because it is a macro that will expand to some number of MOV 690 // instructions and a register operation. 691 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize); 692 __ br(Assembler::LS, after_frame_check); 693 694 // compute rsp as if this were going to be the last frame on 695 // the stack before the red zone 696 697 // locals + overhead, in bytes 698 __ mov(r0, overhead_size); 699 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter. 700 701 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset()); 702 __ ldr(rscratch1, stack_limit); 703 704 #ifdef ASSERT 705 Label limit_okay; 706 // Verify that thread stack limit is non-zero. 707 __ cbnz(rscratch1, limit_okay); 708 __ stop("stack overflow limit is zero"); 709 __ bind(limit_okay); 710 #endif 711 712 // Add stack limit to locals. 713 __ add(r0, r0, rscratch1); 714 715 // Check against the current stack bottom. 716 __ cmp(sp, r0); 717 __ br(Assembler::HI, after_frame_check); 718 719 // Remove the incoming args, peeling the machine SP back to where it 720 // was in the caller. This is not strictly necessary, but unless we 721 // do so the stack frame may have a garbage FP; this ensures a 722 // correct call stack that we can always unwind. The ANDR should be 723 // unnecessary because the sender SP in r13 is always aligned, but 724 // it doesn't hurt. 725 __ andr(sp, r13, -16); 726 727 // Note: the restored frame is not necessarily interpreted. 728 // Use the shared runtime version of the StackOverflowError. 729 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated"); 730 __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry())); 731 732 // all done with frame size check 733 __ bind(after_frame_check); 734 } 735 736 // Allocate monitor and lock method (asm interpreter) 737 // 738 // Args: 739 // rmethod: Method* 740 // rlocals: locals 741 // 742 // Kills: 743 // r0 744 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs) 745 // rscratch1, rscratch2 (scratch regs) 746 void TemplateInterpreterGenerator::lock_method() { 747 // synchronize method 748 const Address access_flags(rmethod, Method::access_flags_offset()); 749 const Address monitor_block_top( 750 rfp, 751 frame::interpreter_frame_monitor_block_top_offset * wordSize); 752 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 753 754 #ifdef ASSERT 755 { 756 Label L; 757 __ ldrw(r0, access_flags); 758 __ tst(r0, JVM_ACC_SYNCHRONIZED); 759 __ br(Assembler::NE, L); 760 __ stop("method doesn't need synchronization"); 761 __ bind(L); 762 } 763 #endif // ASSERT 764 765 // get synchronization object 766 { 767 Label done; 768 __ ldrw(r0, access_flags); 769 __ tst(r0, JVM_ACC_STATIC); 770 // get receiver (assume this is frequent case) 771 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0))); 772 __ br(Assembler::EQ, done); 773 __ load_mirror(r0, rmethod); 774 775 #ifdef ASSERT 776 { 777 Label L; 778 __ cbnz(r0, L); 779 __ stop("synchronization object is NULL"); 780 __ bind(L); 781 } 782 #endif // ASSERT 783 784 __ bind(done); 785 } 786 787 // add space for monitor & lock 788 __ sub(sp, sp, entry_size); // add space for a monitor entry 789 __ sub(esp, esp, entry_size); 790 __ mov(rscratch1, esp); 791 __ str(rscratch1, monitor_block_top); // set new monitor block top 792 // store object 793 __ str(r0, Address(esp, BasicObjectLock::obj_offset_in_bytes())); 794 __ mov(c_rarg1, esp); // object address 795 __ lock_object(c_rarg1); 796 } 797 798 // Generate a fixed interpreter frame. This is identical setup for 799 // interpreted methods and for native methods hence the shared code. 800 // 801 // Args: 802 // lr: return address 803 // rmethod: Method* 804 // rlocals: pointer to locals 805 // rcpool: cp cache 806 // stack_pointer: previous sp 807 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { 808 // initialize fixed part of activation frame 809 if (native_call) { 810 __ sub(esp, sp, 14 * wordSize); 811 __ mov(rbcp, zr); 812 __ stp(esp, zr, Address(__ pre(sp, -14 * wordSize))); 813 // add 2 zero-initialized slots for native calls 814 __ stp(zr, zr, Address(sp, 12 * wordSize)); 815 } else { 816 __ sub(esp, sp, 12 * wordSize); 817 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod 818 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase 819 __ stp(esp, rbcp, Address(__ pre(sp, -12 * wordSize))); 820 } 821 822 if (ProfileInterpreter) { 823 Label method_data_continue; 824 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset())); 825 __ cbz(rscratch1, method_data_continue); 826 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset()))); 827 __ bind(method_data_continue); 828 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer) 829 } else { 830 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp) 831 } 832 833 // Get mirror and store it in the frame as GC root for this Method* 834 __ load_mirror(r10, rmethod); 835 __ stp(r10, zr, Address(sp, 4 * wordSize)); 836 837 __ ldr(rcpool, Address(rmethod, Method::const_offset())); 838 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset())); 839 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset_in_bytes())); 840 __ stp(rlocals, rcpool, Address(sp, 2 * wordSize)); 841 842 __ protect_return_address(); 843 __ stp(rfp, lr, Address(sp, 10 * wordSize)); 844 __ lea(rfp, Address(sp, 10 * wordSize)); 845 846 // set sender sp 847 // leave last_sp as null 848 __ stp(zr, r13, Address(sp, 8 * wordSize)); 849 850 // Move SP out of the way 851 if (! native_call) { 852 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); 853 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset())); 854 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2); 855 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3); 856 __ andr(sp, rscratch1, -16); 857 } 858 } 859 860 // End of helpers 861 862 // Various method entries 863 //------------------------------------------------------------------------------------------------------------------------ 864 // 865 // 866 867 // Method entry for java.lang.ref.Reference.get. 868 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) { 869 // Code: _aload_0, _getfield, _areturn 870 // parameter size = 1 871 // 872 // The code that gets generated by this routine is split into 2 parts: 873 // 1. The "intrinsified" code for G1 (or any SATB based GC), 874 // 2. The slow path - which is an expansion of the regular method entry. 875 // 876 // Notes:- 877 // * In the G1 code we do not check whether we need to block for 878 // a safepoint. If G1 is enabled then we must execute the specialized 879 // code for Reference.get (except when the Reference object is null) 880 // so that we can log the value in the referent field with an SATB 881 // update buffer. 882 // If the code for the getfield template is modified so that the 883 // G1 pre-barrier code is executed when the current method is 884 // Reference.get() then going through the normal method entry 885 // will be fine. 886 // * The G1 code can, however, check the receiver object (the instance 887 // of java.lang.Reference) and jump to the slow path if null. If the 888 // Reference object is null then we obviously cannot fetch the referent 889 // and so we don't need to call the G1 pre-barrier. Thus we can use the 890 // regular method entry code to generate the NPE. 891 // 892 // This code is based on generate_accessor_entry. 893 // 894 // rmethod: Method* 895 // r13: senderSP must preserve for slow path, set SP to it on fast path 896 897 // LR is live. It must be saved around calls. 898 899 address entry = __ pc(); 900 901 const int referent_offset = java_lang_ref_Reference::referent_offset(); 902 903 Label slow_path; 904 const Register local_0 = c_rarg0; 905 // Check if local 0 != NULL 906 // If the receiver is null then it is OK to jump to the slow path. 907 __ ldr(local_0, Address(esp, 0)); 908 __ cbz(local_0, slow_path); 909 910 __ mov(r19, r13); // Move senderSP to a callee-saved register 911 912 // Load the value of the referent field. 913 const Address field_address(local_0, referent_offset); 914 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler(); 915 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch2, /*tmp2*/ rscratch1); 916 917 // areturn 918 __ andr(sp, r19, -16); // done with stack 919 __ ret(lr); 920 921 // generate a vanilla interpreter entry as the slow path 922 __ bind(slow_path); 923 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals)); 924 return entry; 925 926 } 927 928 /** 929 * Method entry for static native methods: 930 * int java.util.zip.CRC32.update(int crc, int b) 931 */ 932 address TemplateInterpreterGenerator::generate_CRC32_update_entry() { 933 if (UseCRC32Intrinsics) { 934 address entry = __ pc(); 935 936 // rmethod: Method* 937 // r13: senderSP must preserved for slow path 938 // esp: args 939 940 Label slow_path; 941 // If we need a safepoint check, generate full interpreter entry. 942 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */); 943 944 // We don't generate local frame and don't align stack because 945 // we call stub code and there is no safepoint on this path. 946 947 // Load parameters 948 const Register crc = c_rarg0; // crc 949 const Register val = c_rarg1; // source java byte value 950 const Register tbl = c_rarg2; // scratch 951 952 // Arguments are reversed on java expression stack 953 __ ldrw(val, Address(esp, 0)); // byte value 954 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC 955 956 uint64_t offset; 957 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset); 958 __ add(tbl, tbl, offset); 959 960 __ mvnw(crc, crc); // ~crc 961 __ update_byte_crc32(crc, val, tbl); 962 __ mvnw(crc, crc); // ~crc 963 964 // result in c_rarg0 965 966 __ andr(sp, r13, -16); 967 __ ret(lr); 968 969 // generate a vanilla native entry as the slow path 970 __ bind(slow_path); 971 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native)); 972 return entry; 973 } 974 return NULL; 975 } 976 977 /** 978 * Method entry for static native methods: 979 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len) 980 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 981 */ 982 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 983 if (UseCRC32Intrinsics) { 984 address entry = __ pc(); 985 986 // rmethod,: Method* 987 // r13: senderSP must preserved for slow path 988 989 Label slow_path; 990 // If we need a safepoint check, generate full interpreter entry. 991 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */); 992 993 // We don't generate local frame and don't align stack because 994 // we call stub code and there is no safepoint on this path. 995 996 // Load parameters 997 const Register crc = c_rarg0; // crc 998 const Register buf = c_rarg1; // source java byte array address 999 const Register len = c_rarg2; // length 1000 const Register off = len; // offset (never overlaps with 'len') 1001 1002 // Arguments are reversed on java expression stack 1003 // Calculate address of start element 1004 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { 1005 __ ldr(buf, Address(esp, 2*wordSize)); // long buf 1006 __ ldrw(off, Address(esp, wordSize)); // offset 1007 __ add(buf, buf, off); // + offset 1008 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC 1009 } else { 1010 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array 1011 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size 1012 __ ldrw(off, Address(esp, wordSize)); // offset 1013 __ add(buf, buf, off); // + offset 1014 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC 1015 } 1016 // Can now load 'len' since we're finished with 'off' 1017 __ ldrw(len, Address(esp, 0x0)); // Length 1018 1019 __ andr(sp, r13, -16); // Restore the caller's SP 1020 1021 // We are frameless so we can just jump to the stub. 1022 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32())); 1023 1024 // generate a vanilla native entry as the slow path 1025 __ bind(slow_path); 1026 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native)); 1027 return entry; 1028 } 1029 return NULL; 1030 } 1031 1032 /** 1033 * Method entry for intrinsic-candidate (non-native) methods: 1034 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end) 1035 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end) 1036 * Unlike CRC32, CRC32C does not have any methods marked as native 1037 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses 1038 */ 1039 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 1040 if (UseCRC32CIntrinsics) { 1041 address entry = __ pc(); 1042 1043 // Prepare jump to stub using parameters from the stack 1044 const Register crc = c_rarg0; // initial crc 1045 const Register buf = c_rarg1; // source java byte array address 1046 const Register len = c_rarg2; // len argument to the kernel 1047 1048 const Register end = len; // index of last element to process 1049 const Register off = crc; // offset 1050 1051 __ ldrw(end, Address(esp)); // int end 1052 __ ldrw(off, Address(esp, wordSize)); // int offset 1053 __ sub(len, end, off); 1054 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf 1055 __ add(buf, buf, off); // + offset 1056 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { 1057 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc 1058 } else { 1059 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size 1060 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc 1061 } 1062 1063 __ andr(sp, r13, -16); // Restore the caller's SP 1064 1065 // Jump to the stub. 1066 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C())); 1067 1068 return entry; 1069 } 1070 return NULL; 1071 } 1072 1073 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) { 1074 // Bang each page in the shadow zone. We can't assume it's been done for 1075 // an interpreter frame with greater than a page of locals, so each page 1076 // needs to be checked. Only true for non-native. 1077 const int n_shadow_pages = (int)(StackOverflow::stack_shadow_zone_size() / os::vm_page_size()); 1078 const int start_page = native_call ? n_shadow_pages : 1; 1079 const int page_size = os::vm_page_size(); 1080 for (int pages = start_page; pages <= n_shadow_pages ; pages++) { 1081 __ sub(rscratch2, sp, pages*page_size); 1082 __ str(zr, Address(rscratch2)); 1083 } 1084 } 1085 1086 1087 // Interpreter stub for calling a native method. (asm interpreter) 1088 // This sets up a somewhat different looking stack for calling the 1089 // native method than the typical interpreter frame setup. 1090 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) { 1091 // determine code generation flags 1092 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 1093 1094 // r1: Method* 1095 // rscratch1: sender sp 1096 1097 address entry_point = __ pc(); 1098 1099 const Address constMethod (rmethod, Method::const_offset()); 1100 const Address access_flags (rmethod, Method::access_flags_offset()); 1101 const Address size_of_parameters(r2, ConstMethod:: 1102 size_of_parameters_offset()); 1103 1104 // get parameter size (always needed) 1105 __ ldr(r2, constMethod); 1106 __ load_unsigned_short(r2, size_of_parameters); 1107 1108 // Native calls don't need the stack size check since they have no 1109 // expression stack and the arguments are already on the stack and 1110 // we only add a handful of words to the stack. 1111 1112 // rmethod: Method* 1113 // r2: size of parameters 1114 // rscratch1: sender sp 1115 1116 // for natives the size of locals is zero 1117 1118 // compute beginning of parameters (rlocals) 1119 __ add(rlocals, esp, r2, ext::uxtx, 3); 1120 __ add(rlocals, rlocals, -wordSize); 1121 1122 // Pull SP back to minimum size: this avoids holes in the stack 1123 __ andr(sp, esp, -16); 1124 1125 // initialize fixed part of activation frame 1126 generate_fixed_frame(true); 1127 1128 // make sure method is native & not abstract 1129 #ifdef ASSERT 1130 __ ldrw(r0, access_flags); 1131 { 1132 Label L; 1133 __ tst(r0, JVM_ACC_NATIVE); 1134 __ br(Assembler::NE, L); 1135 __ stop("tried to execute non-native method as native"); 1136 __ bind(L); 1137 } 1138 { 1139 Label L; 1140 __ tst(r0, JVM_ACC_ABSTRACT); 1141 __ br(Assembler::EQ, L); 1142 __ stop("tried to execute abstract method in interpreter"); 1143 __ bind(L); 1144 } 1145 #endif 1146 1147 // Since at this point in the method invocation the exception 1148 // handler would try to exit the monitor of synchronized methods 1149 // which hasn't been entered yet, we set the thread local variable 1150 // _do_not_unlock_if_synchronized to true. The remove_activation 1151 // will check this flag. 1152 1153 const Address do_not_unlock_if_synchronized(rthread, 1154 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1155 __ mov(rscratch2, true); 1156 __ strb(rscratch2, do_not_unlock_if_synchronized); 1157 1158 // increment invocation count & check for overflow 1159 Label invocation_counter_overflow; 1160 if (inc_counter) { 1161 generate_counter_incr(&invocation_counter_overflow); 1162 } 1163 1164 Label continue_after_compile; 1165 __ bind(continue_after_compile); 1166 1167 bang_stack_shadow_pages(true); 1168 1169 // reset the _do_not_unlock_if_synchronized flag 1170 __ strb(zr, do_not_unlock_if_synchronized); 1171 1172 // check for synchronized methods 1173 // Must happen AFTER invocation_counter check and stack overflow check, 1174 // so method is not locked if overflows. 1175 if (synchronized) { 1176 lock_method(); 1177 } else { 1178 // no synchronization necessary 1179 #ifdef ASSERT 1180 { 1181 Label L; 1182 __ ldrw(r0, access_flags); 1183 __ tst(r0, JVM_ACC_SYNCHRONIZED); 1184 __ br(Assembler::EQ, L); 1185 __ stop("method needs synchronization"); 1186 __ bind(L); 1187 } 1188 #endif 1189 } 1190 1191 // start execution 1192 #ifdef ASSERT 1193 { 1194 Label L; 1195 const Address monitor_block_top(rfp, 1196 frame::interpreter_frame_monitor_block_top_offset * wordSize); 1197 __ ldr(rscratch1, monitor_block_top); 1198 __ cmp(esp, rscratch1); 1199 __ br(Assembler::EQ, L); 1200 __ stop("broken stack frame setup in interpreter"); 1201 __ bind(L); 1202 } 1203 #endif 1204 1205 // jvmti support 1206 __ notify_method_entry(); 1207 1208 // work registers 1209 const Register t = r17; 1210 const Register result_handler = r19; 1211 1212 // allocate space for parameters 1213 __ ldr(t, Address(rmethod, Method::const_offset())); 1214 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset())); 1215 1216 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize); 1217 __ andr(sp, rscratch1, -16); 1218 __ mov(esp, rscratch1); 1219 1220 // get signature handler 1221 { 1222 Label L; 1223 __ ldr(t, Address(rmethod, Method::signature_handler_offset())); 1224 __ cbnz(t, L); 1225 __ call_VM(noreg, 1226 CAST_FROM_FN_PTR(address, 1227 InterpreterRuntime::prepare_native_call), 1228 rmethod); 1229 __ ldr(t, Address(rmethod, Method::signature_handler_offset())); 1230 __ bind(L); 1231 } 1232 1233 // call signature handler 1234 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals, 1235 "adjust this code"); 1236 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp, 1237 "adjust this code"); 1238 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1, 1239 "adjust this code"); 1240 1241 // The generated handlers do not touch rmethod (the method). 1242 // However, large signatures cannot be cached and are generated 1243 // each time here. The slow-path generator can do a GC on return, 1244 // so we must reload it after the call. 1245 __ blr(t); 1246 __ get_method(rmethod); // slow path can do a GC, reload rmethod 1247 1248 1249 // result handler is in r0 1250 // set result handler 1251 __ mov(result_handler, r0); 1252 // pass mirror handle if static call 1253 { 1254 Label L; 1255 __ ldrw(t, Address(rmethod, Method::access_flags_offset())); 1256 __ tbz(t, exact_log2(JVM_ACC_STATIC), L); 1257 // get mirror 1258 __ load_mirror(t, rmethod); 1259 // copy mirror into activation frame 1260 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize)); 1261 // pass handle to mirror 1262 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize); 1263 __ bind(L); 1264 } 1265 1266 // get native function entry point in r10 1267 { 1268 Label L; 1269 __ ldr(r10, Address(rmethod, Method::native_function_offset())); 1270 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry()); 1271 __ mov(rscratch2, unsatisfied); 1272 __ ldr(rscratch2, rscratch2); 1273 __ cmp(r10, rscratch2); 1274 __ br(Assembler::NE, L); 1275 __ call_VM(noreg, 1276 CAST_FROM_FN_PTR(address, 1277 InterpreterRuntime::prepare_native_call), 1278 rmethod); 1279 __ get_method(rmethod); 1280 __ ldr(r10, Address(rmethod, Method::native_function_offset())); 1281 __ bind(L); 1282 } 1283 1284 // pass JNIEnv 1285 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset())); 1286 1287 // Set the last Java PC in the frame anchor to be the return address from 1288 // the call to the native method: this will allow the debugger to 1289 // generate an accurate stack trace. 1290 Label native_return; 1291 __ set_last_Java_frame(esp, rfp, native_return, rscratch1); 1292 1293 // change thread state 1294 #ifdef ASSERT 1295 { 1296 Label L; 1297 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset())); 1298 __ cmp(t, (u1)_thread_in_Java); 1299 __ br(Assembler::EQ, L); 1300 __ stop("Wrong thread state in native stub"); 1301 __ bind(L); 1302 } 1303 #endif 1304 1305 // Change state to native 1306 __ mov(rscratch1, _thread_in_native); 1307 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset())); 1308 __ stlrw(rscratch1, rscratch2); 1309 1310 // Call the native method. 1311 __ blr(r10); 1312 __ bind(native_return); 1313 __ get_method(rmethod); 1314 // result potentially in r0 or v0 1315 1316 // make room for the pushes we're about to do 1317 __ sub(rscratch1, esp, 4 * wordSize); 1318 __ andr(sp, rscratch1, -16); 1319 1320 // NOTE: The order of these pushes is known to frame::interpreter_frame_result 1321 // in order to extract the result of a method call. If the order of these 1322 // pushes change or anything else is added to the stack then the code in 1323 // interpreter_frame_result must also change. 1324 __ push(dtos); 1325 __ push(ltos); 1326 1327 __ verify_sve_vector_length(); 1328 1329 // change thread state 1330 __ mov(rscratch1, _thread_in_native_trans); 1331 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset())); 1332 __ stlrw(rscratch1, rscratch2); 1333 1334 // Force this write out before the read below 1335 __ dmb(Assembler::ISH); 1336 1337 // check for safepoint operation in progress and/or pending suspend requests 1338 { 1339 Label L, Continue; 1340 1341 // We need an acquire here to ensure that any subsequent load of the 1342 // global SafepointSynchronize::_state flag is ordered after this load 1343 // of the thread-local polling word. We don't want this poll to 1344 // return false (i.e. not safepointing) and a later poll of the global 1345 // SafepointSynchronize::_state spuriously to return true. 1346 // 1347 // This is to avoid a race when we're in a native->Java transition 1348 // racing the code which wakes up from a safepoint. 1349 __ safepoint_poll(L, true /* at_return */, true /* acquire */, false /* in_nmethod */); 1350 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset())); 1351 __ cbz(rscratch2, Continue); 1352 __ bind(L); 1353 1354 // Don't use call_VM as it will see a possible pending exception 1355 // and forward it and never return here preventing us from 1356 // clearing _last_native_pc down below. So we do a runtime call by 1357 // hand. 1358 // 1359 __ mov(c_rarg0, rthread); 1360 __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)); 1361 __ blr(rscratch2); 1362 __ get_method(rmethod); 1363 __ reinit_heapbase(); 1364 __ bind(Continue); 1365 } 1366 1367 // change thread state 1368 __ mov(rscratch1, _thread_in_Java); 1369 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset())); 1370 __ stlrw(rscratch1, rscratch2); 1371 1372 // reset_last_Java_frame 1373 __ reset_last_Java_frame(true); 1374 1375 if (CheckJNICalls) { 1376 // clear_pending_jni_exception_check 1377 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset())); 1378 } 1379 1380 // reset handle block 1381 __ ldr(t, Address(rthread, JavaThread::active_handles_offset())); 1382 __ str(zr, Address(t, JNIHandleBlock::top_offset_in_bytes())); 1383 1384 // If result is an oop unbox and store it in frame where gc will see it 1385 // and result handler will pick it up 1386 1387 { 1388 Label no_oop; 1389 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT))); 1390 __ cmp(t, result_handler); 1391 __ br(Assembler::NE, no_oop); 1392 // Unbox oop result, e.g. JNIHandles::resolve result. 1393 __ pop(ltos); 1394 __ resolve_jobject(r0, rthread, t); 1395 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize)); 1396 // keep stack depth as expected by pushing oop which will eventually be discarded 1397 __ push(ltos); 1398 __ bind(no_oop); 1399 } 1400 1401 { 1402 Label no_reguard; 1403 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset()))); 1404 __ ldrw(rscratch1, Address(rscratch1)); 1405 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled); 1406 __ br(Assembler::NE, no_reguard); 1407 1408 __ push_call_clobbered_registers(); 1409 __ mov(c_rarg0, rthread); 1410 __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)); 1411 __ blr(rscratch2); 1412 __ pop_call_clobbered_registers(); 1413 1414 __ bind(no_reguard); 1415 } 1416 1417 // The method register is junk from after the thread_in_native transition 1418 // until here. Also can't call_VM until the bcp has been 1419 // restored. Need bcp for throwing exception below so get it now. 1420 __ get_method(rmethod); 1421 1422 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=> 1423 // rbcp == code_base() 1424 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod* 1425 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase 1426 // handle exceptions (exception handling will handle unlocking!) 1427 { 1428 Label L; 1429 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset())); 1430 __ cbz(rscratch1, L); 1431 // Note: At some point we may want to unify this with the code 1432 // used in call_VM_base(); i.e., we should use the 1433 // StubRoutines::forward_exception code. For now this doesn't work 1434 // here because the rsp is not correctly set at this point. 1435 __ MacroAssembler::call_VM(noreg, 1436 CAST_FROM_FN_PTR(address, 1437 InterpreterRuntime::throw_pending_exception)); 1438 __ should_not_reach_here(); 1439 __ bind(L); 1440 } 1441 1442 // do unlocking if necessary 1443 { 1444 Label L; 1445 __ ldrw(t, Address(rmethod, Method::access_flags_offset())); 1446 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L); 1447 // the code below should be shared with interpreter macro 1448 // assembler implementation 1449 { 1450 Label unlock; 1451 // BasicObjectLock will be first in list, since this is a 1452 // synchronized method. However, need to check that the object 1453 // has not been unlocked by an explicit monitorexit bytecode. 1454 1455 // monitor expect in c_rarg1 for slow unlock path 1456 __ lea (c_rarg1, Address(rfp, // address of first monitor 1457 (intptr_t)(frame::interpreter_frame_initial_sp_offset * 1458 wordSize - sizeof(BasicObjectLock)))); 1459 1460 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 1461 __ cbnz(t, unlock); 1462 1463 // Entry already unlocked, need to throw exception 1464 __ MacroAssembler::call_VM(noreg, 1465 CAST_FROM_FN_PTR(address, 1466 InterpreterRuntime::throw_illegal_monitor_state_exception)); 1467 __ should_not_reach_here(); 1468 1469 __ bind(unlock); 1470 __ unlock_object(c_rarg1); 1471 } 1472 __ bind(L); 1473 } 1474 1475 // jvmti support 1476 // Note: This must happen _after_ handling/throwing any exceptions since 1477 // the exception handler code notifies the runtime of method exits 1478 // too. If this happens before, method entry/exit notifications are 1479 // not properly paired (was bug - gri 11/22/99). 1480 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); 1481 1482 // restore potential result in r0:d0, call result handler to 1483 // restore potential result in ST0 & handle result 1484 1485 __ pop(ltos); 1486 __ pop(dtos); 1487 1488 __ blr(result_handler); 1489 1490 // remove activation 1491 __ ldr(esp, Address(rfp, 1492 frame::interpreter_frame_sender_sp_offset * 1493 wordSize)); // get sender sp 1494 // remove frame anchor 1495 __ leave(); 1496 1497 // resture sender sp 1498 __ mov(sp, esp); 1499 1500 __ ret(lr); 1501 1502 if (inc_counter) { 1503 // Handle overflow of counter and compile method 1504 __ bind(invocation_counter_overflow); 1505 generate_counter_overflow(continue_after_compile); 1506 } 1507 1508 return entry_point; 1509 } 1510 1511 // 1512 // Generic interpreted method entry to (asm) interpreter 1513 // 1514 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) { 1515 // determine code generation flags 1516 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 1517 1518 // rscratch1: sender sp 1519 address entry_point = __ pc(); 1520 1521 const Address constMethod(rmethod, Method::const_offset()); 1522 const Address access_flags(rmethod, Method::access_flags_offset()); 1523 const Address size_of_parameters(r3, 1524 ConstMethod::size_of_parameters_offset()); 1525 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset()); 1526 1527 // get parameter size (always needed) 1528 // need to load the const method first 1529 __ ldr(r3, constMethod); 1530 __ load_unsigned_short(r2, size_of_parameters); 1531 1532 // r2: size of parameters 1533 1534 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words 1535 __ sub(r3, r3, r2); // r3 = no. of additional locals 1536 1537 // see if we've got enough room on the stack for locals plus overhead. 1538 generate_stack_overflow_check(); 1539 1540 // compute beginning of parameters (rlocals) 1541 __ add(rlocals, esp, r2, ext::uxtx, 3); 1542 __ sub(rlocals, rlocals, wordSize); 1543 1544 __ mov(rscratch1, esp); 1545 1546 // r3 - # of additional locals 1547 // allocate space for locals 1548 // explicitly initialize locals 1549 // Initializing memory allocated for locals in the same direction as 1550 // the stack grows to ensure page initialization order according 1551 // to windows-aarch64 stack page growth requirement (see 1552 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack) 1553 { 1554 Label exit, loop; 1555 __ ands(zr, r3, r3); 1556 __ br(Assembler::LE, exit); // do nothing if r3 <= 0 1557 __ bind(loop); 1558 __ str(zr, Address(__ pre(rscratch1, -wordSize))); 1559 __ sub(r3, r3, 1); // until everything initialized 1560 __ cbnz(r3, loop); 1561 __ bind(exit); 1562 } 1563 1564 // Padding between locals and fixed part of activation frame to ensure 1565 // SP is always 16-byte aligned. 1566 __ andr(sp, rscratch1, -16); 1567 1568 // And the base dispatch table 1569 __ get_dispatch(); 1570 1571 // initialize fixed part of activation frame 1572 generate_fixed_frame(false); 1573 1574 // make sure method is not native & not abstract 1575 #ifdef ASSERT 1576 __ ldrw(r0, access_flags); 1577 { 1578 Label L; 1579 __ tst(r0, JVM_ACC_NATIVE); 1580 __ br(Assembler::EQ, L); 1581 __ stop("tried to execute native method as non-native"); 1582 __ bind(L); 1583 } 1584 { 1585 Label L; 1586 __ tst(r0, JVM_ACC_ABSTRACT); 1587 __ br(Assembler::EQ, L); 1588 __ stop("tried to execute abstract method in interpreter"); 1589 __ bind(L); 1590 } 1591 #endif 1592 1593 // Since at this point in the method invocation the exception 1594 // handler would try to exit the monitor of synchronized methods 1595 // which hasn't been entered yet, we set the thread local variable 1596 // _do_not_unlock_if_synchronized to true. The remove_activation 1597 // will check this flag. 1598 1599 const Address do_not_unlock_if_synchronized(rthread, 1600 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1601 __ mov(rscratch2, true); 1602 __ strb(rscratch2, do_not_unlock_if_synchronized); 1603 1604 Register mdp = r3; 1605 __ profile_parameters_type(mdp, r1, r2); 1606 1607 // increment invocation count & check for overflow 1608 Label invocation_counter_overflow; 1609 if (inc_counter) { 1610 generate_counter_incr(&invocation_counter_overflow); 1611 } 1612 1613 Label continue_after_compile; 1614 __ bind(continue_after_compile); 1615 1616 bang_stack_shadow_pages(false); 1617 1618 // reset the _do_not_unlock_if_synchronized flag 1619 __ strb(zr, do_not_unlock_if_synchronized); 1620 1621 // check for synchronized methods 1622 // Must happen AFTER invocation_counter check and stack overflow check, 1623 // so method is not locked if overflows. 1624 if (synchronized) { 1625 // Allocate monitor and lock method 1626 lock_method(); 1627 } else { 1628 // no synchronization necessary 1629 #ifdef ASSERT 1630 { 1631 Label L; 1632 __ ldrw(r0, access_flags); 1633 __ tst(r0, JVM_ACC_SYNCHRONIZED); 1634 __ br(Assembler::EQ, L); 1635 __ stop("method needs synchronization"); 1636 __ bind(L); 1637 } 1638 #endif 1639 } 1640 1641 // start execution 1642 #ifdef ASSERT 1643 { 1644 Label L; 1645 const Address monitor_block_top (rfp, 1646 frame::interpreter_frame_monitor_block_top_offset * wordSize); 1647 __ ldr(rscratch1, monitor_block_top); 1648 __ cmp(esp, rscratch1); 1649 __ br(Assembler::EQ, L); 1650 __ stop("broken stack frame setup in interpreter"); 1651 __ bind(L); 1652 } 1653 #endif 1654 1655 // jvmti support 1656 __ notify_method_entry(); 1657 1658 __ dispatch_next(vtos); 1659 1660 // invocation counter overflow 1661 if (inc_counter) { 1662 // Handle overflow of counter and compile method 1663 __ bind(invocation_counter_overflow); 1664 generate_counter_overflow(continue_after_compile); 1665 } 1666 1667 return entry_point; 1668 } 1669 1670 //----------------------------------------------------------------------------- 1671 // Exceptions 1672 1673 void TemplateInterpreterGenerator::generate_throw_exception() { 1674 // Entry point in previous activation (i.e., if the caller was 1675 // interpreted) 1676 Interpreter::_rethrow_exception_entry = __ pc(); 1677 // Restore sp to interpreter_frame_last_sp even though we are going 1678 // to empty the expression stack for the exception processing. 1679 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1680 // r0: exception 1681 // r3: return address/pc that threw exception 1682 __ restore_bcp(); // rbcp points to call/send 1683 __ restore_locals(); 1684 __ restore_constant_pool_cache(); 1685 __ reinit_heapbase(); // restore rheapbase as heapbase. 1686 __ get_dispatch(); 1687 1688 // Entry point for exceptions thrown within interpreter code 1689 Interpreter::_throw_exception_entry = __ pc(); 1690 // If we came here via a NullPointerException on the receiver of a 1691 // method, rmethod may be corrupt. 1692 __ get_method(rmethod); 1693 // expression stack is undefined here 1694 // r0: exception 1695 // rbcp: exception bcp 1696 __ verify_oop(r0); 1697 __ mov(c_rarg1, r0); 1698 1699 // expression stack must be empty before entering the VM in case of 1700 // an exception 1701 __ empty_expression_stack(); 1702 // find exception handler address and preserve exception oop 1703 __ call_VM(r3, 1704 CAST_FROM_FN_PTR(address, 1705 InterpreterRuntime::exception_handler_for_exception), 1706 c_rarg1); 1707 1708 // Calculate stack limit 1709 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); 1710 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset())); 1711 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4); 1712 __ ldr(rscratch2, 1713 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize)); 1714 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3); 1715 __ andr(sp, rscratch1, -16); 1716 1717 // r0: exception handler entry point 1718 // r3: preserved exception oop 1719 // rbcp: bcp for exception handler 1720 __ push_ptr(r3); // push exception which is now the only value on the stack 1721 __ br(r0); // jump to exception handler (may be _remove_activation_entry!) 1722 1723 // If the exception is not handled in the current frame the frame is 1724 // removed and the exception is rethrown (i.e. exception 1725 // continuation is _rethrow_exception). 1726 // 1727 // Note: At this point the bci is still the bxi for the instruction 1728 // which caused the exception and the expression stack is 1729 // empty. Thus, for any VM calls at this point, GC will find a legal 1730 // oop map (with empty expression stack). 1731 1732 // 1733 // JVMTI PopFrame support 1734 // 1735 1736 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 1737 __ empty_expression_stack(); 1738 // Set the popframe_processing bit in pending_popframe_condition 1739 // indicating that we are currently handling popframe, so that 1740 // call_VMs that may happen later do not trigger new popframe 1741 // handling cycles. 1742 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset())); 1743 __ orr(r3, r3, JavaThread::popframe_processing_bit); 1744 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset())); 1745 1746 { 1747 // Check to see whether we are returning to a deoptimized frame. 1748 // (The PopFrame call ensures that the caller of the popped frame is 1749 // either interpreted or compiled and deoptimizes it if compiled.) 1750 // In this case, we can't call dispatch_next() after the frame is 1751 // popped, but instead must save the incoming arguments and restore 1752 // them after deoptimization has occurred. 1753 // 1754 // Note that we don't compare the return PC against the 1755 // deoptimization blob's unpack entry because of the presence of 1756 // adapter frames in C2. 1757 Label caller_not_deoptimized; 1758 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize)); 1759 // This is a return address, so requires authenticating for PAC. 1760 __ authenticate_return_address(c_rarg1, rscratch1); 1761 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1762 InterpreterRuntime::interpreter_contains), c_rarg1); 1763 __ cbnz(r0, caller_not_deoptimized); 1764 1765 // Compute size of arguments for saving when returning to 1766 // deoptimized caller 1767 __ get_method(r0); 1768 __ ldr(r0, Address(r0, Method::const_offset())); 1769 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod:: 1770 size_of_parameters_offset()))); 1771 __ lsl(r0, r0, Interpreter::logStackElementSize); 1772 __ restore_locals(); // XXX do we need this? 1773 __ sub(rlocals, rlocals, r0); 1774 __ add(rlocals, rlocals, wordSize); 1775 // Save these arguments 1776 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1777 Deoptimization:: 1778 popframe_preserve_args), 1779 rthread, r0, rlocals); 1780 1781 __ remove_activation(vtos, 1782 /* throw_monitor_exception */ false, 1783 /* install_monitor_exception */ false, 1784 /* notify_jvmdi */ false); 1785 1786 // Inform deoptimization that it is responsible for restoring 1787 // these arguments 1788 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit); 1789 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset())); 1790 1791 // Continue in deoptimization handler 1792 __ ret(lr); 1793 1794 __ bind(caller_not_deoptimized); 1795 } 1796 1797 __ remove_activation(vtos, 1798 /* throw_monitor_exception */ false, 1799 /* install_monitor_exception */ false, 1800 /* notify_jvmdi */ false); 1801 1802 // Restore the last_sp and null it out 1803 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1804 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1805 1806 __ restore_bcp(); 1807 __ restore_locals(); 1808 __ restore_constant_pool_cache(); 1809 __ get_method(rmethod); 1810 __ get_dispatch(); 1811 1812 // The method data pointer was incremented already during 1813 // call profiling. We have to restore the mdp for the current bcp. 1814 if (ProfileInterpreter) { 1815 __ set_method_data_pointer_for_bcp(); 1816 } 1817 1818 // Clear the popframe condition flag 1819 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset())); 1820 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive"); 1821 1822 #if INCLUDE_JVMTI 1823 { 1824 Label L_done; 1825 1826 __ ldrb(rscratch1, Address(rbcp, 0)); 1827 __ cmpw(rscratch1, Bytecodes::_invokestatic); 1828 __ br(Assembler::NE, L_done); 1829 1830 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. 1831 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL. 1832 1833 __ ldr(c_rarg0, Address(rlocals, 0)); 1834 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp); 1835 1836 __ cbz(r0, L_done); 1837 1838 __ str(r0, Address(esp, 0)); 1839 __ bind(L_done); 1840 } 1841 #endif // INCLUDE_JVMTI 1842 1843 // Restore machine SP 1844 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); 1845 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset())); 1846 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4); 1847 __ ldr(rscratch2, 1848 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize)); 1849 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3); 1850 __ andr(sp, rscratch1, -16); 1851 1852 __ dispatch_next(vtos); 1853 // end of PopFrame support 1854 1855 Interpreter::_remove_activation_entry = __ pc(); 1856 1857 // preserve exception over this code sequence 1858 __ pop_ptr(r0); 1859 __ str(r0, Address(rthread, JavaThread::vm_result_offset())); 1860 // remove the activation (without doing throws on illegalMonitorExceptions) 1861 __ remove_activation(vtos, false, true, false); 1862 // restore exception 1863 __ get_vm_result(r0, rthread); 1864 1865 // In between activations - previous activation type unknown yet 1866 // compute continuation point - the continuation point expects the 1867 // following registers set up: 1868 // 1869 // r0: exception 1870 // lr: return address/pc that threw exception 1871 // esp: expression stack of caller 1872 // rfp: fp of caller 1873 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address 1874 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1875 SharedRuntime::exception_handler_for_return_address), 1876 rthread, lr); 1877 __ mov(r1, r0); // save exception handler 1878 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address 1879 // We might be returning to a deopt handler that expects r3 to 1880 // contain the exception pc 1881 __ mov(r3, lr); 1882 // Note that an "issuing PC" is actually the next PC after the call 1883 __ br(r1); // jump to exception 1884 // handler of caller 1885 } 1886 1887 1888 // 1889 // JVMTI ForceEarlyReturn support 1890 // 1891 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 1892 address entry = __ pc(); 1893 1894 __ restore_bcp(); 1895 __ restore_locals(); 1896 __ empty_expression_stack(); 1897 __ load_earlyret_value(state); 1898 1899 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset())); 1900 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset()); 1901 1902 // Clear the earlyret state 1903 assert(JvmtiThreadState::earlyret_inactive == 0, "should be"); 1904 __ str(zr, cond_addr); 1905 1906 __ remove_activation(state, 1907 false, /* throw_monitor_exception */ 1908 false, /* install_monitor_exception */ 1909 true); /* notify_jvmdi */ 1910 __ ret(lr); 1911 1912 return entry; 1913 } // end of ForceEarlyReturn support 1914 1915 1916 1917 //----------------------------------------------------------------------------- 1918 // Helper for vtos entry point generation 1919 1920 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, 1921 address& bep, 1922 address& cep, 1923 address& sep, 1924 address& aep, 1925 address& iep, 1926 address& lep, 1927 address& fep, 1928 address& dep, 1929 address& vep) { 1930 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 1931 Label L; 1932 aep = __ pc(); __ push_ptr(); __ b(L); 1933 fep = __ pc(); __ push_f(); __ b(L); 1934 dep = __ pc(); __ push_d(); __ b(L); 1935 lep = __ pc(); __ push_l(); __ b(L); 1936 bep = cep = sep = 1937 iep = __ pc(); __ push_i(); 1938 vep = __ pc(); 1939 __ bind(L); 1940 generate_and_dispatch(t); 1941 } 1942 1943 //----------------------------------------------------------------------------- 1944 1945 // Non-product code 1946 #ifndef PRODUCT 1947 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 1948 address entry = __ pc(); 1949 1950 __ protect_return_address(); 1951 __ push(lr); 1952 __ push(state); 1953 __ push(RegSet::range(r0, r15), sp); 1954 __ mov(c_rarg2, r0); // Pass itos 1955 __ call_VM(noreg, 1956 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), 1957 c_rarg1, c_rarg2, c_rarg3); 1958 __ pop(RegSet::range(r0, r15), sp); 1959 __ pop(state); 1960 __ pop(lr); 1961 __ authenticate_return_address(); 1962 __ ret(lr); // return from result handler 1963 1964 return entry; 1965 } 1966 1967 void TemplateInterpreterGenerator::count_bytecode() { 1968 Register rscratch3 = r0; 1969 __ push(rscratch1); 1970 __ push(rscratch2); 1971 __ push(rscratch3); 1972 __ mov(rscratch3, (address) &BytecodeCounter::_counter_value); 1973 __ atomic_add(noreg, 1, rscratch3); 1974 __ pop(rscratch3); 1975 __ pop(rscratch2); 1976 __ pop(rscratch1); 1977 } 1978 1979 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { ; } 1980 1981 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { ; } 1982 1983 1984 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 1985 // Call a little run-time stub to avoid blow-up for each bytecode. 1986 // The run-time runtime saves the right registers, depending on 1987 // the tosca in-state for the given template. 1988 1989 assert(Interpreter::trace_code(t->tos_in()) != NULL, 1990 "entry must have been generated"); 1991 __ bl(Interpreter::trace_code(t->tos_in())); 1992 __ reinit_heapbase(); 1993 } 1994 1995 1996 void TemplateInterpreterGenerator::stop_interpreter_at() { 1997 Label L; 1998 __ push(rscratch1); 1999 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value); 2000 __ ldr(rscratch1, Address(rscratch1)); 2001 __ mov(rscratch2, StopInterpreterAt); 2002 __ cmpw(rscratch1, rscratch2); 2003 __ br(Assembler::NE, L); 2004 __ brk(0); 2005 __ bind(L); 2006 __ pop(rscratch1); 2007 } 2008 2009 #endif // !PRODUCT