1 /* 2 * Copyright (c) 2003, 2015, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "asm/macroAssembler.hpp" 27 #include "interpreter/bytecodeHistogram.hpp" 28 #include "interpreter/interpreter.hpp" 29 #include "interpreter/interpreterGenerator.hpp" 30 #include "interpreter/interpreterRuntime.hpp" 31 #include "interpreter/interp_masm.hpp" 32 #include "interpreter/templateTable.hpp" 33 #include "oops/arrayOop.hpp" 34 #include "oops/methodData.hpp" 35 #include "oops/method.hpp" 36 #include "oops/oop.inline.hpp" 37 #include "prims/jvmtiExport.hpp" 38 #include "prims/jvmtiThreadState.hpp" 39 #include "runtime/arguments.hpp" 40 #include "runtime/deoptimization.hpp" 41 #include "runtime/frame.inline.hpp" 42 #include "runtime/sharedRuntime.hpp" 43 #include "runtime/stubRoutines.hpp" 44 #include "runtime/synchronizer.hpp" 45 #include "runtime/timer.hpp" 46 #include "runtime/vframeArray.hpp" 47 #include "utilities/debug.hpp" 48 #include "utilities/macros.hpp" 49 50 #define __ _masm-> 51 52 #ifndef CC_INTERP 53 54 const int method_offset = frame::interpreter_frame_method_offset * wordSize; 55 const int bcp_offset = frame::interpreter_frame_bcp_offset * wordSize; 56 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize; 57 58 //----------------------------------------------------------------------------- 59 60 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 61 address entry = __ pc(); 62 63 #ifdef ASSERT 64 { 65 Label L; 66 __ lea(rax, Address(rbp, 67 frame::interpreter_frame_monitor_block_top_offset * 68 wordSize)); 69 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack 70 // grows negative) 71 __ jcc(Assembler::aboveEqual, L); // check if frame is complete 72 __ stop ("interpreter frame not set up"); 73 __ bind(L); 74 } 75 #endif // ASSERT 76 // Restore bcp under the assumption that the current frame is still 77 // interpreted 78 __ restore_bcp(); 79 80 // expression stack must be empty before entering the VM if an 81 // exception happened 82 __ empty_expression_stack(); 83 // throw exception 84 __ call_VM(noreg, 85 CAST_FROM_FN_PTR(address, 86 InterpreterRuntime::throw_StackOverflowError)); 87 return entry; 88 } 89 90 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler( 91 const char* name) { 92 address entry = __ pc(); 93 // expression stack must be empty before entering the VM if an 94 // exception happened 95 __ empty_expression_stack(); 96 // setup parameters 97 // ??? convention: expect aberrant index in register ebx 98 __ lea(c_rarg1, ExternalAddress((address)name)); 99 __ call_VM(noreg, 100 CAST_FROM_FN_PTR(address, 101 InterpreterRuntime:: 102 throw_ArrayIndexOutOfBoundsException), 103 c_rarg1, rbx); 104 return entry; 105 } 106 107 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 108 address entry = __ pc(); 109 110 // object is at TOS 111 __ pop(c_rarg1); 112 113 // expression stack must be empty before entering the VM if an 114 // exception happened 115 __ empty_expression_stack(); 116 117 __ call_VM(noreg, 118 CAST_FROM_FN_PTR(address, 119 InterpreterRuntime:: 120 throw_ClassCastException), 121 c_rarg1); 122 return entry; 123 } 124 125 address TemplateInterpreterGenerator::generate_exception_handler_common( 126 const char* name, const char* message, bool pass_oop) { 127 assert(!pass_oop || message == NULL, "either oop or message but not both"); 128 address entry = __ pc(); 129 if (pass_oop) { 130 // object is at TOS 131 __ pop(c_rarg2); 132 } 133 // expression stack must be empty before entering the VM if an 134 // exception happened 135 __ empty_expression_stack(); 136 // setup parameters 137 __ lea(c_rarg1, ExternalAddress((address)name)); 138 if (pass_oop) { 139 __ call_VM(rax, CAST_FROM_FN_PTR(address, 140 InterpreterRuntime:: 141 create_klass_exception), 142 c_rarg1, c_rarg2); 143 } else { 144 // kind of lame ExternalAddress can't take NULL because 145 // external_word_Relocation will assert. 146 if (message != NULL) { 147 __ lea(c_rarg2, ExternalAddress((address)message)); 148 } else { 149 __ movptr(c_rarg2, NULL_WORD); 150 } 151 __ call_VM(rax, 152 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), 153 c_rarg1, c_rarg2); 154 } 155 // throw exception 156 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); 157 return entry; 158 } 159 160 161 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { 162 address entry = __ pc(); 163 // NULL last_sp until next java call 164 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 165 __ dispatch_next(state); 166 return entry; 167 } 168 169 170 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) { 171 address entry = __ pc(); 172 173 // Restore stack bottom in case i2c adjusted stack 174 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 175 // and NULL it as marker that esp is now tos until next java call 176 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 177 178 __ restore_bcp(); 179 __ restore_locals(); 180 181 if (state == atos) { 182 Register mdp = rbx; 183 Register tmp = rcx; 184 __ profile_return_type(mdp, rax, tmp); 185 } 186 187 const Register cache = rbx; 188 const Register index = rcx; 189 __ get_cache_and_index_at_bcp(cache, index, 1, index_size); 190 191 const Register flags = cache; 192 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset())); 193 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask); 194 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale())); 195 __ dispatch_next(state, step); 196 197 return entry; 198 } 199 200 201 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) { 202 address entry = __ pc(); 203 // NULL last_sp until next java call 204 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 205 __ restore_bcp(); 206 __ restore_locals(); 207 #if INCLUDE_JVMCI 208 // Check if we need to take lock at entry of synchronized method. 209 if (UseJVMCICompiler) { 210 Label L; 211 __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0); 212 __ jcc(Assembler::zero, L); 213 // Clear flag. 214 __ movb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0); 215 // Satisfy calling convention for lock_method(). 216 __ get_method(rbx); 217 // Take lock. 218 lock_method(); 219 __ bind(L); 220 } 221 #endif 222 // handle exceptions 223 { 224 Label L; 225 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); 226 __ jcc(Assembler::zero, L); 227 __ call_VM(noreg, 228 CAST_FROM_FN_PTR(address, 229 InterpreterRuntime::throw_pending_exception)); 230 __ should_not_reach_here(); 231 __ bind(L); 232 } 233 __ dispatch_next(state, step); 234 return entry; 235 } 236 237 int AbstractInterpreter::BasicType_as_index(BasicType type) { 238 int i = 0; 239 switch (type) { 240 case T_BOOLEAN: i = 0; break; 241 case T_CHAR : i = 1; break; 242 case T_BYTE : i = 2; break; 243 case T_SHORT : i = 3; break; 244 case T_INT : i = 4; break; 245 case T_LONG : i = 5; break; 246 case T_VOID : i = 6; break; 247 case T_FLOAT : i = 7; break; 248 case T_DOUBLE : i = 8; break; 249 case T_OBJECT : i = 9; break; 250 case T_ARRAY : i = 9; break; 251 default : ShouldNotReachHere(); 252 } 253 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, 254 "index out of bounds"); 255 return i; 256 } 257 258 259 address TemplateInterpreterGenerator::generate_result_handler_for( 260 BasicType type) { 261 address entry = __ pc(); 262 switch (type) { 263 case T_BOOLEAN: __ c2bool(rax); break; 264 case T_CHAR : __ movzwl(rax, rax); break; 265 case T_BYTE : __ sign_extend_byte(rax); break; 266 case T_SHORT : __ sign_extend_short(rax); break; 267 case T_INT : /* nothing to do */ break; 268 case T_LONG : /* nothing to do */ break; 269 case T_VOID : /* nothing to do */ break; 270 case T_FLOAT : /* nothing to do */ break; 271 case T_DOUBLE : /* nothing to do */ break; 272 case T_OBJECT : 273 // retrieve result from frame 274 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize)); 275 // and verify it 276 __ verify_oop(rax); 277 break; 278 default : ShouldNotReachHere(); 279 } 280 __ ret(0); // return from result handler 281 return entry; 282 } 283 284 address TemplateInterpreterGenerator::generate_safept_entry_for( 285 TosState state, 286 address runtime_entry) { 287 address entry = __ pc(); 288 __ push(state); 289 __ call_VM(noreg, runtime_entry); 290 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 291 return entry; 292 } 293 294 295 296 // Helpers for commoning out cases in the various type of method entries. 297 // 298 299 300 // increment invocation count & check for overflow 301 // 302 // Note: checking for negative value instead of overflow 303 // so we have a 'sticky' overflow test 304 // 305 // rbx: method 306 // ecx: invocation counter 307 // 308 void InterpreterGenerator::generate_counter_incr( 309 Label* overflow, 310 Label* profile_method, 311 Label* profile_method_continue) { 312 Label done; 313 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not. 314 if (TieredCompilation) { 315 int increment = InvocationCounter::count_increment; 316 Label no_mdo; 317 if (ProfileInterpreter) { 318 // Are we profiling? 319 __ movptr(rax, Address(rbx, Method::method_data_offset())); 320 __ testptr(rax, rax); 321 __ jccb(Assembler::zero, no_mdo); 322 // Increment counter in the MDO 323 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) + 324 in_bytes(InvocationCounter::counter_offset())); 325 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset())); 326 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow); 327 __ jmp(done); 328 } 329 __ bind(no_mdo); 330 // Increment counter in MethodCounters 331 const Address invocation_counter(rax, 332 MethodCounters::invocation_counter_offset() + 333 InvocationCounter::counter_offset()); 334 __ get_method_counters(rbx, rax, done); 335 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset())); 336 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, 337 false, Assembler::zero, overflow); 338 __ bind(done); 339 } else { // not TieredCompilation 340 const Address backedge_counter(rax, 341 MethodCounters::backedge_counter_offset() + 342 InvocationCounter::counter_offset()); 343 const Address invocation_counter(rax, 344 MethodCounters::invocation_counter_offset() + 345 InvocationCounter::counter_offset()); 346 347 __ get_method_counters(rbx, rax, done); 348 349 if (ProfileInterpreter) { 350 __ incrementl(Address(rax, 351 MethodCounters::interpreter_invocation_counter_offset())); 352 } 353 // Update standard invocation counters 354 __ movl(rcx, invocation_counter); 355 __ incrementl(rcx, InvocationCounter::count_increment); 356 __ movl(invocation_counter, rcx); // save invocation count 357 358 __ movl(rax, backedge_counter); // load backedge counter 359 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits 360 361 __ addl(rcx, rax); // add both counters 362 363 // profile_method is non-null only for interpreted method so 364 // profile_method != NULL == !native_call 365 366 if (ProfileInterpreter && profile_method != NULL) { 367 // Test to see if we should create a method data oop 368 __ movptr(rax, Address(rbx, Method::method_counters_offset())); 369 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset()))); 370 __ jcc(Assembler::less, *profile_method_continue); 371 372 // if no method data exists, go to profile_method 373 __ test_method_data_pointer(rax, *profile_method); 374 } 375 376 __ movptr(rax, Address(rbx, Method::method_counters_offset())); 377 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset()))); 378 __ jcc(Assembler::aboveEqual, *overflow); 379 __ bind(done); 380 } 381 } 382 383 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) { 384 385 // Asm interpreter on entry 386 // r14 - locals 387 // r13 - bcp 388 // rbx - method 389 // edx - cpool --- DOES NOT APPEAR TO BE TRUE 390 // rbp - interpreter frame 391 392 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] 393 // Everything as it was on entry 394 // rdx is not restored. Doesn't appear to really be set. 395 396 // InterpreterRuntime::frequency_counter_overflow takes two 397 // arguments, the first (thread) is passed by call_VM, the second 398 // indicates if the counter overflow occurs at a backwards branch 399 // (NULL bcp). We pass zero for it. The call returns the address 400 // of the verified entry point for the method or NULL if the 401 // compilation did not complete (either went background or bailed 402 // out). 403 __ movl(c_rarg1, 0); 404 __ call_VM(noreg, 405 CAST_FROM_FN_PTR(address, 406 InterpreterRuntime::frequency_counter_overflow), 407 c_rarg1); 408 409 __ movptr(rbx, Address(rbp, method_offset)); // restore Method* 410 // Preserve invariant that r13/r14 contain bcp/locals of sender frame 411 // and jump to the interpreted entry. 412 __ jmp(*do_continue, relocInfo::none); 413 } 414 415 // See if we've got enough room on the stack for locals plus overhead. 416 // The expression stack grows down incrementally, so the normal guard 417 // page mechanism will work for that. 418 // 419 // NOTE: Since the additional locals are also always pushed (wasn't 420 // obvious in generate_fixed_frame) so the guard should work for them 421 // too. 422 // 423 // Args: 424 // rdx: number of additional locals this frame needs (what we must check) 425 // rbx: Method* 426 // 427 // Kills: 428 // rax 429 void InterpreterGenerator::generate_stack_overflow_check(void) { 430 431 // monitor entry size: see picture of stack in frame_x86.hpp 432 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 433 434 // total overhead size: entry_size + (saved rbp through expr stack 435 // bottom). be sure to change this if you add/subtract anything 436 // to/from the overhead area 437 const int overhead_size = 438 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size; 439 440 const int page_size = os::vm_page_size(); 441 442 Label after_frame_check; 443 444 // see if the frame is greater than one page in size. If so, 445 // then we need to verify there is enough stack space remaining 446 // for the additional locals. 447 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize); 448 __ jcc(Assembler::belowEqual, after_frame_check); 449 450 // compute rsp as if this were going to be the last frame on 451 // the stack before the red zone 452 453 const Address stack_base(r15_thread, Thread::stack_base_offset()); 454 const Address stack_size(r15_thread, Thread::stack_size_offset()); 455 456 // locals + overhead, in bytes 457 __ mov(rax, rdx); 458 __ shlptr(rax, Interpreter::logStackElementSize); // 2 slots per parameter. 459 __ addptr(rax, overhead_size); 460 461 #ifdef ASSERT 462 Label stack_base_okay, stack_size_okay; 463 // verify that thread stack base is non-zero 464 __ cmpptr(stack_base, (int32_t)NULL_WORD); 465 __ jcc(Assembler::notEqual, stack_base_okay); 466 __ stop("stack base is zero"); 467 __ bind(stack_base_okay); 468 // verify that thread stack size is non-zero 469 __ cmpptr(stack_size, 0); 470 __ jcc(Assembler::notEqual, stack_size_okay); 471 __ stop("stack size is zero"); 472 __ bind(stack_size_okay); 473 #endif 474 475 // Add stack base to locals and subtract stack size 476 __ addptr(rax, stack_base); 477 __ subptr(rax, stack_size); 478 479 // Use the maximum number of pages we might bang. 480 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages : 481 (StackRedPages+StackYellowPages); 482 483 // add in the red and yellow zone sizes 484 __ addptr(rax, max_pages * page_size); 485 486 // check against the current stack bottom 487 __ cmpptr(rsp, rax); 488 __ jcc(Assembler::above, after_frame_check); 489 490 // Restore sender's sp as SP. This is necessary if the sender's 491 // frame is an extended compiled frame (see gen_c2i_adapter()) 492 // and safer anyway in case of JSR292 adaptations. 493 494 __ pop(rax); // return address must be moved if SP is changed 495 __ mov(rsp, r13); 496 __ push(rax); 497 498 // Note: the restored frame is not necessarily interpreted. 499 // Use the shared runtime version of the StackOverflowError. 500 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated"); 501 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry())); 502 503 // all done with frame size check 504 __ bind(after_frame_check); 505 } 506 507 // Allocate monitor and lock method (asm interpreter) 508 // 509 // Args: 510 // rbx: Method* 511 // r14: locals 512 // 513 // Kills: 514 // rax 515 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs) 516 // rscratch1, rscratch2 (scratch regs) 517 void TemplateInterpreterGenerator::lock_method() { 518 // synchronize method 519 const Address access_flags(rbx, Method::access_flags_offset()); 520 const Address monitor_block_top( 521 rbp, 522 frame::interpreter_frame_monitor_block_top_offset * wordSize); 523 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 524 525 #ifdef ASSERT 526 { 527 Label L; 528 __ movl(rax, access_flags); 529 __ testl(rax, JVM_ACC_SYNCHRONIZED); 530 __ jcc(Assembler::notZero, L); 531 __ stop("method doesn't need synchronization"); 532 __ bind(L); 533 } 534 #endif // ASSERT 535 536 // get synchronization object 537 { 538 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 539 Label done; 540 __ movl(rax, access_flags); 541 __ testl(rax, JVM_ACC_STATIC); 542 // get receiver (assume this is frequent case) 543 __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0))); 544 __ jcc(Assembler::zero, done); 545 __ movptr(rax, Address(rbx, Method::const_offset())); 546 __ movptr(rax, Address(rax, ConstMethod::constants_offset())); 547 __ movptr(rax, Address(rax, 548 ConstantPool::pool_holder_offset_in_bytes())); 549 __ movptr(rax, Address(rax, mirror_offset)); 550 551 #ifdef ASSERT 552 { 553 Label L; 554 __ testptr(rax, rax); 555 __ jcc(Assembler::notZero, L); 556 __ stop("synchronization object is NULL"); 557 __ bind(L); 558 } 559 #endif // ASSERT 560 561 __ bind(done); 562 } 563 564 // add space for monitor & lock 565 __ subptr(rsp, entry_size); // add space for a monitor entry 566 __ movptr(monitor_block_top, rsp); // set new monitor block top 567 // store object 568 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); 569 __ movptr(c_rarg1, rsp); // object address 570 __ lock_object(c_rarg1); 571 } 572 573 // Generate a fixed interpreter frame. This is identical setup for 574 // interpreted methods and for native methods hence the shared code. 575 // 576 // Args: 577 // rax: return address 578 // rbx: Method* 579 // r14: pointer to locals 580 // r13: sender sp 581 // rdx: cp cache 582 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { 583 // initialize fixed part of activation frame 584 __ push(rax); // save return address 585 __ enter(); // save old & set new rbp 586 __ push(r13); // set sender sp 587 __ push((int)NULL_WORD); // leave last_sp as null 588 __ movptr(r13, Address(rbx, Method::const_offset())); // get ConstMethod* 589 __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase 590 __ push(rbx); // save Method* 591 if (ProfileInterpreter) { 592 Label method_data_continue; 593 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset()))); 594 __ testptr(rdx, rdx); 595 __ jcc(Assembler::zero, method_data_continue); 596 __ addptr(rdx, in_bytes(MethodData::data_offset())); 597 __ bind(method_data_continue); 598 __ push(rdx); // set the mdp (method data pointer) 599 } else { 600 __ push(0); 601 } 602 603 __ movptr(rdx, Address(rbx, Method::const_offset())); 604 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset())); 605 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes())); 606 __ push(rdx); // set constant pool cache 607 __ push(r14); // set locals pointer 608 if (native_call) { 609 __ push(0); // no bcp 610 } else { 611 __ push(r13); // set bcp 612 } 613 __ push(0); // reserve word for pointer to expression stack bottom 614 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom 615 } 616 617 // End of helpers 618 619 // Method entry for java.lang.ref.Reference.get. 620 address InterpreterGenerator::generate_Reference_get_entry(void) { 621 #if INCLUDE_ALL_GCS 622 // Code: _aload_0, _getfield, _areturn 623 // parameter size = 1 624 // 625 // The code that gets generated by this routine is split into 2 parts: 626 // 1. The "intrinsified" code for G1 (or any SATB based GC), 627 // 2. The slow path - which is an expansion of the regular method entry. 628 // 629 // Notes:- 630 // * In the G1 code we do not check whether we need to block for 631 // a safepoint. If G1 is enabled then we must execute the specialized 632 // code for Reference.get (except when the Reference object is null) 633 // so that we can log the value in the referent field with an SATB 634 // update buffer. 635 // If the code for the getfield template is modified so that the 636 // G1 pre-barrier code is executed when the current method is 637 // Reference.get() then going through the normal method entry 638 // will be fine. 639 // * The G1 code can, however, check the receiver object (the instance 640 // of java.lang.Reference) and jump to the slow path if null. If the 641 // Reference object is null then we obviously cannot fetch the referent 642 // and so we don't need to call the G1 pre-barrier. Thus we can use the 643 // regular method entry code to generate the NPE. 644 // 645 // rbx: Method* 646 647 // r13: senderSP must preserve for slow path, set SP to it on fast path 648 649 address entry = __ pc(); 650 651 const int referent_offset = java_lang_ref_Reference::referent_offset; 652 guarantee(referent_offset > 0, "referent offset not initialized"); 653 654 if (UseG1GC) { 655 Label slow_path; 656 // rbx: method 657 658 // Check if local 0 != NULL 659 // If the receiver is null then it is OK to jump to the slow path. 660 __ movptr(rax, Address(rsp, wordSize)); 661 662 __ testptr(rax, rax); 663 __ jcc(Assembler::zero, slow_path); 664 665 // rax: local 0 666 // rbx: method (but can be used as scratch now) 667 // rdx: scratch 668 // rdi: scratch 669 670 // Generate the G1 pre-barrier code to log the value of 671 // the referent field in an SATB buffer. 672 673 // Load the value of the referent field. 674 const Address field_address(rax, referent_offset); 675 __ load_heap_oop(rax, field_address); 676 677 // Generate the G1 pre-barrier code to log the value of 678 // the referent field in an SATB buffer. 679 __ g1_write_barrier_pre(noreg /* obj */, 680 rax /* pre_val */, 681 r15_thread /* thread */, 682 rbx /* tmp */, 683 true /* tosca_live */, 684 true /* expand_call */); 685 686 // _areturn 687 __ pop(rdi); // get return address 688 __ mov(rsp, r13); // set sp to sender sp 689 __ jmp(rdi); 690 __ ret(0); 691 692 // generate a vanilla interpreter entry as the slow path 693 __ bind(slow_path); 694 (void) generate_normal_entry(false); 695 696 return entry; 697 } 698 #endif // INCLUDE_ALL_GCS 699 700 // If G1 is not enabled then attempt to go through the accessor entry point 701 // Reference.get is an accessor 702 return generate_jump_to_normal_entry(); 703 } 704 705 /** 706 * Method entry for static native methods: 707 * int java.util.zip.CRC32.update(int crc, int b) 708 */ 709 address InterpreterGenerator::generate_CRC32_update_entry() { 710 if (UseCRC32Intrinsics) { 711 address entry = __ pc(); 712 713 // rbx,: Method* 714 // r13: senderSP must preserved for slow path, set SP to it on fast path 715 // c_rarg0: scratch (rdi on non-Win64, rcx on Win64) 716 // c_rarg1: scratch (rsi on non-Win64, rdx on Win64) 717 718 Label slow_path; 719 // If we need a safepoint check, generate full interpreter entry. 720 ExternalAddress state(SafepointSynchronize::address_of_state()); 721 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 722 SafepointSynchronize::_not_synchronized); 723 __ jcc(Assembler::notEqual, slow_path); 724 725 // We don't generate local frame and don't align stack because 726 // we call stub code and there is no safepoint on this path. 727 728 // Load parameters 729 const Register crc = rax; // crc 730 const Register val = c_rarg0; // source java byte value 731 const Register tbl = c_rarg1; // scratch 732 733 // Arguments are reversed on java expression stack 734 __ movl(val, Address(rsp, wordSize)); // byte value 735 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC 736 737 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr())); 738 __ notl(crc); // ~crc 739 __ update_byte_crc32(crc, val, tbl); 740 __ notl(crc); // ~crc 741 // result in rax 742 743 // _areturn 744 __ pop(rdi); // get return address 745 __ mov(rsp, r13); // set sp to sender sp 746 __ jmp(rdi); 747 748 // generate a vanilla native entry as the slow path 749 __ bind(slow_path); 750 751 (void) generate_native_entry(false); 752 753 return entry; 754 } 755 return generate_native_entry(false); 756 } 757 758 /** 759 * Method entry for static native methods: 760 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len) 761 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 762 */ 763 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 764 if (UseCRC32Intrinsics) { 765 address entry = __ pc(); 766 767 // rbx,: Method* 768 // r13: senderSP must preserved for slow path, set SP to it on fast path 769 770 Label slow_path; 771 // If we need a safepoint check, generate full interpreter entry. 772 ExternalAddress state(SafepointSynchronize::address_of_state()); 773 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 774 SafepointSynchronize::_not_synchronized); 775 __ jcc(Assembler::notEqual, slow_path); 776 777 // We don't generate local frame and don't align stack because 778 // we call stub code and there is no safepoint on this path. 779 780 // Load parameters 781 const Register crc = c_rarg0; // crc 782 const Register buf = c_rarg1; // source java byte array address 783 const Register len = c_rarg2; // length 784 const Register off = len; // offset (never overlaps with 'len') 785 786 // Arguments are reversed on java expression stack 787 // Calculate address of start element 788 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { 789 __ movptr(buf, Address(rsp, 3*wordSize)); // long buf 790 __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset 791 __ addq(buf, off); // + offset 792 __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC 793 } else { 794 __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array 795 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size 796 __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset 797 __ addq(buf, off); // + offset 798 __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC 799 } 800 // Can now load 'len' since we're finished with 'off' 801 __ movl(len, Address(rsp, wordSize)); // Length 802 803 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len); 804 // result in rax 805 806 // _areturn 807 __ pop(rdi); // get return address 808 __ mov(rsp, r13); // set sp to sender sp 809 __ jmp(rdi); 810 811 // generate a vanilla native entry as the slow path 812 __ bind(slow_path); 813 814 (void) generate_native_entry(false); 815 816 return entry; 817 } 818 return generate_native_entry(false); 819 } 820 821 // Interpreter stub for calling a native method. (asm interpreter) 822 // This sets up a somewhat different looking stack for calling the 823 // native method than the typical interpreter frame setup. 824 address InterpreterGenerator::generate_native_entry(bool synchronized) { 825 // determine code generation flags 826 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 827 828 // rbx: Method* 829 // r13: sender sp 830 831 address entry_point = __ pc(); 832 833 const Address constMethod (rbx, Method::const_offset()); 834 const Address access_flags (rbx, Method::access_flags_offset()); 835 const Address size_of_parameters(rcx, ConstMethod:: 836 size_of_parameters_offset()); 837 838 839 // get parameter size (always needed) 840 __ movptr(rcx, constMethod); 841 __ load_unsigned_short(rcx, size_of_parameters); 842 843 // native calls don't need the stack size check since they have no 844 // expression stack and the arguments are already on the stack and 845 // we only add a handful of words to the stack 846 847 // rbx: Method* 848 // rcx: size of parameters 849 // r13: sender sp 850 __ pop(rax); // get return address 851 852 // for natives the size of locals is zero 853 854 // compute beginning of parameters (r14) 855 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize)); 856 857 // add 2 zero-initialized slots for native calls 858 // initialize result_handler slot 859 __ push((int) NULL_WORD); 860 // slot for oop temp 861 // (static native method holder mirror/jni oop result) 862 __ push((int) NULL_WORD); 863 864 // initialize fixed part of activation frame 865 generate_fixed_frame(true); 866 867 // make sure method is native & not abstract 868 #ifdef ASSERT 869 __ movl(rax, access_flags); 870 { 871 Label L; 872 __ testl(rax, JVM_ACC_NATIVE); 873 __ jcc(Assembler::notZero, L); 874 __ stop("tried to execute non-native method as native"); 875 __ bind(L); 876 } 877 { 878 Label L; 879 __ testl(rax, JVM_ACC_ABSTRACT); 880 __ jcc(Assembler::zero, L); 881 __ stop("tried to execute abstract method in interpreter"); 882 __ bind(L); 883 } 884 #endif 885 886 // Since at this point in the method invocation the exception handler 887 // would try to exit the monitor of synchronized methods which hasn't 888 // been entered yet, we set the thread local variable 889 // _do_not_unlock_if_synchronized to true. The remove_activation will 890 // check this flag. 891 892 const Address do_not_unlock_if_synchronized(r15_thread, 893 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 894 __ movbool(do_not_unlock_if_synchronized, true); 895 896 // increment invocation count & check for overflow 897 Label invocation_counter_overflow; 898 if (inc_counter) { 899 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); 900 } 901 902 Label continue_after_compile; 903 __ bind(continue_after_compile); 904 905 bang_stack_shadow_pages(true); 906 907 // reset the _do_not_unlock_if_synchronized flag 908 __ movbool(do_not_unlock_if_synchronized, false); 909 910 // check for synchronized methods 911 // Must happen AFTER invocation_counter check and stack overflow check, 912 // so method is not locked if overflows. 913 if (synchronized) { 914 lock_method(); 915 } else { 916 // no synchronization necessary 917 #ifdef ASSERT 918 { 919 Label L; 920 __ movl(rax, access_flags); 921 __ testl(rax, JVM_ACC_SYNCHRONIZED); 922 __ jcc(Assembler::zero, L); 923 __ stop("method needs synchronization"); 924 __ bind(L); 925 } 926 #endif 927 } 928 929 // start execution 930 #ifdef ASSERT 931 { 932 Label L; 933 const Address monitor_block_top(rbp, 934 frame::interpreter_frame_monitor_block_top_offset * wordSize); 935 __ movptr(rax, monitor_block_top); 936 __ cmpptr(rax, rsp); 937 __ jcc(Assembler::equal, L); 938 __ stop("broken stack frame setup in interpreter"); 939 __ bind(L); 940 } 941 #endif 942 943 // jvmti support 944 __ notify_method_entry(); 945 946 // work registers 947 const Register method = rbx; 948 const Register t = r11; 949 950 // allocate space for parameters 951 __ get_method(method); 952 __ movptr(t, Address(method, Method::const_offset())); 953 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset())); 954 __ shll(t, Interpreter::logStackElementSize); 955 956 __ subptr(rsp, t); 957 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 958 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI) 959 960 // get signature handler 961 { 962 Label L; 963 __ movptr(t, Address(method, Method::signature_handler_offset())); 964 __ testptr(t, t); 965 __ jcc(Assembler::notZero, L); 966 __ call_VM(noreg, 967 CAST_FROM_FN_PTR(address, 968 InterpreterRuntime::prepare_native_call), 969 method); 970 __ get_method(method); 971 __ movptr(t, Address(method, Method::signature_handler_offset())); 972 __ bind(L); 973 } 974 975 // call signature handler 976 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14, 977 "adjust this code"); 978 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp, 979 "adjust this code"); 980 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1, 981 "adjust this code"); 982 983 // The generated handlers do not touch RBX (the method oop). 984 // However, large signatures cannot be cached and are generated 985 // each time here. The slow-path generator can do a GC on return, 986 // so we must reload it after the call. 987 __ call(t); 988 __ get_method(method); // slow path can do a GC, reload RBX 989 990 991 // result handler is in rax 992 // set result handler 993 __ movptr(Address(rbp, 994 (frame::interpreter_frame_result_handler_offset) * wordSize), 995 rax); 996 997 // pass mirror handle if static call 998 { 999 Label L; 1000 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 1001 __ movl(t, Address(method, Method::access_flags_offset())); 1002 __ testl(t, JVM_ACC_STATIC); 1003 __ jcc(Assembler::zero, L); 1004 // get mirror 1005 __ movptr(t, Address(method, Method::const_offset())); 1006 __ movptr(t, Address(t, ConstMethod::constants_offset())); 1007 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes())); 1008 __ movptr(t, Address(t, mirror_offset)); 1009 // copy mirror into activation frame 1010 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), 1011 t); 1012 // pass handle to mirror 1013 __ lea(c_rarg1, 1014 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize)); 1015 __ bind(L); 1016 } 1017 1018 // get native function entry point 1019 { 1020 Label L; 1021 __ movptr(rax, Address(method, Method::native_function_offset())); 1022 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry()); 1023 __ movptr(rscratch2, unsatisfied.addr()); 1024 __ cmpptr(rax, rscratch2); 1025 __ jcc(Assembler::notEqual, L); 1026 __ call_VM(noreg, 1027 CAST_FROM_FN_PTR(address, 1028 InterpreterRuntime::prepare_native_call), 1029 method); 1030 __ get_method(method); 1031 __ movptr(rax, Address(method, Method::native_function_offset())); 1032 __ bind(L); 1033 } 1034 1035 // pass JNIEnv 1036 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset())); 1037 1038 // It is enough that the pc() points into the right code 1039 // segment. It does not have to be the correct return pc. 1040 __ set_last_Java_frame(rsp, rbp, (address) __ pc()); 1041 1042 // change thread state 1043 #ifdef ASSERT 1044 { 1045 Label L; 1046 __ movl(t, Address(r15_thread, JavaThread::thread_state_offset())); 1047 __ cmpl(t, _thread_in_Java); 1048 __ jcc(Assembler::equal, L); 1049 __ stop("Wrong thread state in native stub"); 1050 __ bind(L); 1051 } 1052 #endif 1053 1054 // Change state to native 1055 1056 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), 1057 _thread_in_native); 1058 1059 // Call the native method. 1060 __ call(rax); 1061 // result potentially in rax or xmm0 1062 1063 // Verify or restore cpu control state after JNI call 1064 __ restore_cpu_control_state_after_jni(); 1065 1066 // NOTE: The order of these pushes is known to frame::interpreter_frame_result 1067 // in order to extract the result of a method call. If the order of these 1068 // pushes change or anything else is added to the stack then the code in 1069 // interpreter_frame_result must also change. 1070 1071 __ push(dtos); 1072 __ push(ltos); 1073 1074 // change thread state 1075 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), 1076 _thread_in_native_trans); 1077 1078 if (os::is_MP()) { 1079 if (UseMembar) { 1080 // Force this write out before the read below 1081 __ membar(Assembler::Membar_mask_bits( 1082 Assembler::LoadLoad | Assembler::LoadStore | 1083 Assembler::StoreLoad | Assembler::StoreStore)); 1084 } else { 1085 // Write serialization page so VM thread can do a pseudo remote membar. 1086 // We use the current thread pointer to calculate a thread specific 1087 // offset to write to within the page. This minimizes bus traffic 1088 // due to cache line collision. 1089 __ serialize_memory(r15_thread, rscratch2); 1090 } 1091 } 1092 1093 // check for safepoint operation in progress and/or pending suspend requests 1094 { 1095 Label Continue; 1096 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 1097 SafepointSynchronize::_not_synchronized); 1098 1099 Label L; 1100 __ jcc(Assembler::notEqual, L); 1101 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0); 1102 __ jcc(Assembler::equal, Continue); 1103 __ bind(L); 1104 1105 // Don't use call_VM as it will see a possible pending exception 1106 // and forward it and never return here preventing us from 1107 // clearing _last_native_pc down below. Also can't use 1108 // call_VM_leaf either as it will check to see if r13 & r14 are 1109 // preserved and correspond to the bcp/locals pointers. So we do a 1110 // runtime call by hand. 1111 // 1112 __ mov(c_rarg0, r15_thread); 1113 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) 1114 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 1115 __ andptr(rsp, -16); // align stack as required by ABI 1116 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans))); 1117 __ mov(rsp, r12); // restore sp 1118 __ reinit_heapbase(); 1119 __ bind(Continue); 1120 } 1121 1122 // change thread state 1123 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java); 1124 1125 // reset_last_Java_frame 1126 __ reset_last_Java_frame(true, true); 1127 1128 // reset handle block 1129 __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset())); 1130 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD); 1131 1132 // If result is an oop unbox and store it in frame where gc will see it 1133 // and result handler will pick it up 1134 1135 { 1136 Label no_oop, store_result; 1137 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT))); 1138 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize)); 1139 __ jcc(Assembler::notEqual, no_oop); 1140 // retrieve result 1141 __ pop(ltos); 1142 __ testptr(rax, rax); 1143 __ jcc(Assembler::zero, store_result); 1144 __ movptr(rax, Address(rax, 0)); 1145 __ bind(store_result); 1146 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax); 1147 // keep stack depth as expected by pushing oop which will eventually be discarde 1148 __ push(ltos); 1149 __ bind(no_oop); 1150 } 1151 1152 1153 { 1154 Label no_reguard; 1155 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()), 1156 JavaThread::stack_guard_yellow_disabled); 1157 __ jcc(Assembler::notEqual, no_reguard); 1158 1159 __ pusha(); // XXX only save smashed registers 1160 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) 1161 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 1162 __ andptr(rsp, -16); // align stack as required by ABI 1163 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); 1164 __ mov(rsp, r12); // restore sp 1165 __ popa(); // XXX only restore smashed registers 1166 __ reinit_heapbase(); 1167 1168 __ bind(no_reguard); 1169 } 1170 1171 1172 // The method register is junk from after the thread_in_native transition 1173 // until here. Also can't call_VM until the bcp has been 1174 // restored. Need bcp for throwing exception below so get it now. 1175 __ get_method(method); 1176 1177 // restore r13 to have legal interpreter frame, i.e., bci == 0 <=> 1178 // r13 == code_base() 1179 __ movptr(r13, Address(method, Method::const_offset())); // get ConstMethod* 1180 __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase 1181 // handle exceptions (exception handling will handle unlocking!) 1182 { 1183 Label L; 1184 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); 1185 __ jcc(Assembler::zero, L); 1186 // Note: At some point we may want to unify this with the code 1187 // used in call_VM_base(); i.e., we should use the 1188 // StubRoutines::forward_exception code. For now this doesn't work 1189 // here because the rsp is not correctly set at this point. 1190 __ MacroAssembler::call_VM(noreg, 1191 CAST_FROM_FN_PTR(address, 1192 InterpreterRuntime::throw_pending_exception)); 1193 __ should_not_reach_here(); 1194 __ bind(L); 1195 } 1196 1197 // do unlocking if necessary 1198 { 1199 Label L; 1200 __ movl(t, Address(method, Method::access_flags_offset())); 1201 __ testl(t, JVM_ACC_SYNCHRONIZED); 1202 __ jcc(Assembler::zero, L); 1203 // the code below should be shared with interpreter macro 1204 // assembler implementation 1205 { 1206 Label unlock; 1207 // BasicObjectLock will be first in list, since this is a 1208 // synchronized method. However, need to check that the object 1209 // has not been unlocked by an explicit monitorexit bytecode. 1210 const Address monitor(rbp, 1211 (intptr_t)(frame::interpreter_frame_initial_sp_offset * 1212 wordSize - sizeof(BasicObjectLock))); 1213 1214 // monitor expect in c_rarg1 for slow unlock path 1215 __ lea(c_rarg1, monitor); // address of first monitor 1216 1217 __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 1218 __ testptr(t, t); 1219 __ jcc(Assembler::notZero, unlock); 1220 1221 // Entry already unlocked, need to throw exception 1222 __ MacroAssembler::call_VM(noreg, 1223 CAST_FROM_FN_PTR(address, 1224 InterpreterRuntime::throw_illegal_monitor_state_exception)); 1225 __ should_not_reach_here(); 1226 1227 __ bind(unlock); 1228 __ unlock_object(c_rarg1); 1229 } 1230 __ bind(L); 1231 } 1232 1233 // jvmti support 1234 // Note: This must happen _after_ handling/throwing any exceptions since 1235 // the exception handler code notifies the runtime of method exits 1236 // too. If this happens before, method entry/exit notifications are 1237 // not properly paired (was bug - gri 11/22/99). 1238 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); 1239 1240 // restore potential result in edx:eax, call result handler to 1241 // restore potential result in ST0 & handle result 1242 1243 __ pop(ltos); 1244 __ pop(dtos); 1245 1246 __ movptr(t, Address(rbp, 1247 (frame::interpreter_frame_result_handler_offset) * wordSize)); 1248 __ call(t); 1249 1250 // remove activation 1251 __ movptr(t, Address(rbp, 1252 frame::interpreter_frame_sender_sp_offset * 1253 wordSize)); // get sender sp 1254 __ leave(); // remove frame anchor 1255 __ pop(rdi); // get return address 1256 __ mov(rsp, t); // set sp to sender sp 1257 __ jmp(rdi); 1258 1259 if (inc_counter) { 1260 // Handle overflow of counter and compile method 1261 __ bind(invocation_counter_overflow); 1262 generate_counter_overflow(&continue_after_compile); 1263 } 1264 1265 return entry_point; 1266 } 1267 1268 // 1269 // Generic interpreted method entry to (asm) interpreter 1270 // 1271 address InterpreterGenerator::generate_normal_entry(bool synchronized) { 1272 // determine code generation flags 1273 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 1274 1275 // ebx: Method* 1276 // r13: sender sp 1277 address entry_point = __ pc(); 1278 1279 const Address constMethod(rbx, Method::const_offset()); 1280 const Address access_flags(rbx, Method::access_flags_offset()); 1281 const Address size_of_parameters(rdx, 1282 ConstMethod::size_of_parameters_offset()); 1283 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset()); 1284 1285 1286 // get parameter size (always needed) 1287 __ movptr(rdx, constMethod); 1288 __ load_unsigned_short(rcx, size_of_parameters); 1289 1290 // rbx: Method* 1291 // rcx: size of parameters 1292 // r13: sender_sp (could differ from sp+wordSize if we were called via c2i ) 1293 1294 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words 1295 __ subl(rdx, rcx); // rdx = no. of additional locals 1296 1297 // YYY 1298 // __ incrementl(rdx); 1299 // __ andl(rdx, -2); 1300 1301 // see if we've got enough room on the stack for locals plus overhead. 1302 generate_stack_overflow_check(); 1303 1304 // get return address 1305 __ pop(rax); 1306 1307 // compute beginning of parameters (r14) 1308 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize)); 1309 1310 // rdx - # of additional locals 1311 // allocate space for locals 1312 // explicitly initialize locals 1313 { 1314 Label exit, loop; 1315 __ testl(rdx, rdx); 1316 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0 1317 __ bind(loop); 1318 __ push((int) NULL_WORD); // initialize local variables 1319 __ decrementl(rdx); // until everything initialized 1320 __ jcc(Assembler::greater, loop); 1321 __ bind(exit); 1322 } 1323 1324 // initialize fixed part of activation frame 1325 generate_fixed_frame(false); 1326 1327 // make sure method is not native & not abstract 1328 #ifdef ASSERT 1329 __ movl(rax, access_flags); 1330 { 1331 Label L; 1332 __ testl(rax, JVM_ACC_NATIVE); 1333 __ jcc(Assembler::zero, L); 1334 __ stop("tried to execute native method as non-native"); 1335 __ bind(L); 1336 } 1337 { 1338 Label L; 1339 __ testl(rax, JVM_ACC_ABSTRACT); 1340 __ jcc(Assembler::zero, L); 1341 __ stop("tried to execute abstract method in interpreter"); 1342 __ bind(L); 1343 } 1344 #endif 1345 1346 // Since at this point in the method invocation the exception 1347 // handler would try to exit the monitor of synchronized methods 1348 // which hasn't been entered yet, we set the thread local variable 1349 // _do_not_unlock_if_synchronized to true. The remove_activation 1350 // will check this flag. 1351 1352 const Address do_not_unlock_if_synchronized(r15_thread, 1353 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1354 __ movbool(do_not_unlock_if_synchronized, true); 1355 1356 __ profile_parameters_type(rax, rcx, rdx); 1357 // increment invocation count & check for overflow 1358 Label invocation_counter_overflow; 1359 Label profile_method; 1360 Label profile_method_continue; 1361 if (inc_counter) { 1362 generate_counter_incr(&invocation_counter_overflow, 1363 &profile_method, 1364 &profile_method_continue); 1365 if (ProfileInterpreter) { 1366 __ bind(profile_method_continue); 1367 } 1368 } 1369 1370 Label continue_after_compile; 1371 __ bind(continue_after_compile); 1372 1373 // check for synchronized interpreted methods 1374 bang_stack_shadow_pages(false); 1375 1376 // reset the _do_not_unlock_if_synchronized flag 1377 __ movbool(do_not_unlock_if_synchronized, false); 1378 1379 // check for synchronized methods 1380 // Must happen AFTER invocation_counter check and stack overflow check, 1381 // so method is not locked if overflows. 1382 if (synchronized) { 1383 // Allocate monitor and lock method 1384 lock_method(); 1385 } else { 1386 // no synchronization necessary 1387 #ifdef ASSERT 1388 { 1389 Label L; 1390 __ movl(rax, access_flags); 1391 __ testl(rax, JVM_ACC_SYNCHRONIZED); 1392 __ jcc(Assembler::zero, L); 1393 __ stop("method needs synchronization"); 1394 __ bind(L); 1395 } 1396 #endif 1397 } 1398 1399 // start execution 1400 #ifdef ASSERT 1401 { 1402 Label L; 1403 const Address monitor_block_top (rbp, 1404 frame::interpreter_frame_monitor_block_top_offset * wordSize); 1405 __ movptr(rax, monitor_block_top); 1406 __ cmpptr(rax, rsp); 1407 __ jcc(Assembler::equal, L); 1408 __ stop("broken stack frame setup in interpreter"); 1409 __ bind(L); 1410 } 1411 #endif 1412 1413 // jvmti support 1414 __ notify_method_entry(); 1415 1416 __ dispatch_next(vtos); 1417 1418 // invocation counter overflow 1419 if (inc_counter) { 1420 if (ProfileInterpreter) { 1421 // We have decided to profile this method in the interpreter 1422 __ bind(profile_method); 1423 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1424 __ set_method_data_pointer_for_bcp(); 1425 __ get_method(rbx); 1426 __ jmp(profile_method_continue); 1427 } 1428 // Handle overflow of counter and compile method 1429 __ bind(invocation_counter_overflow); 1430 generate_counter_overflow(&continue_after_compile); 1431 } 1432 1433 return entry_point; 1434 } 1435 1436 1437 // These should never be compiled since the interpreter will prefer 1438 // the compiled version to the intrinsic version. 1439 bool AbstractInterpreter::can_be_compiled(methodHandle m) { 1440 switch (method_kind(m)) { 1441 case Interpreter::java_lang_math_sin : // fall thru 1442 case Interpreter::java_lang_math_cos : // fall thru 1443 case Interpreter::java_lang_math_tan : // fall thru 1444 case Interpreter::java_lang_math_abs : // fall thru 1445 case Interpreter::java_lang_math_log : // fall thru 1446 case Interpreter::java_lang_math_log10 : // fall thru 1447 case Interpreter::java_lang_math_sqrt : // fall thru 1448 case Interpreter::java_lang_math_pow : // fall thru 1449 case Interpreter::java_lang_math_exp : 1450 return false; 1451 default: 1452 return true; 1453 } 1454 } 1455 1456 // How much stack a method activation needs in words. 1457 int AbstractInterpreter::size_top_interpreter_activation(Method* method) { 1458 const int entry_size = frame::interpreter_frame_monitor_size(); 1459 1460 // total overhead size: entry_size + (saved rbp thru expr stack 1461 // bottom). be sure to change this if you add/subtract anything 1462 // to/from the overhead area 1463 const int overhead_size = 1464 -(frame::interpreter_frame_initial_sp_offset) + entry_size; 1465 1466 const int stub_code = frame::entry_frame_after_call_words; 1467 const int method_stack = (method->max_locals() + method->max_stack()) * 1468 Interpreter::stackElementWords; 1469 return (overhead_size + method_stack + stub_code); 1470 } 1471 1472 //----------------------------------------------------------------------------- 1473 // Exceptions 1474 1475 void TemplateInterpreterGenerator::generate_throw_exception() { 1476 // Entry point in previous activation (i.e., if the caller was 1477 // interpreted) 1478 Interpreter::_rethrow_exception_entry = __ pc(); 1479 // Restore sp to interpreter_frame_last_sp even though we are going 1480 // to empty the expression stack for the exception processing. 1481 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 1482 // rax: exception 1483 // rdx: return address/pc that threw exception 1484 __ restore_bcp(); // r13 points to call/send 1485 __ restore_locals(); 1486 __ reinit_heapbase(); // restore r12 as heapbase. 1487 // Entry point for exceptions thrown within interpreter code 1488 Interpreter::_throw_exception_entry = __ pc(); 1489 // expression stack is undefined here 1490 // rax: exception 1491 // r13: exception bcp 1492 __ verify_oop(rax); 1493 __ mov(c_rarg1, rax); 1494 1495 // expression stack must be empty before entering the VM in case of 1496 // an exception 1497 __ empty_expression_stack(); 1498 // find exception handler address and preserve exception oop 1499 __ call_VM(rdx, 1500 CAST_FROM_FN_PTR(address, 1501 InterpreterRuntime::exception_handler_for_exception), 1502 c_rarg1); 1503 // rax: exception handler entry point 1504 // rdx: preserved exception oop 1505 // r13: bcp for exception handler 1506 __ push_ptr(rdx); // push exception which is now the only value on the stack 1507 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!) 1508 1509 // If the exception is not handled in the current frame the frame is 1510 // removed and the exception is rethrown (i.e. exception 1511 // continuation is _rethrow_exception). 1512 // 1513 // Note: At this point the bci is still the bxi for the instruction 1514 // which caused the exception and the expression stack is 1515 // empty. Thus, for any VM calls at this point, GC will find a legal 1516 // oop map (with empty expression stack). 1517 1518 // In current activation 1519 // tos: exception 1520 // esi: exception bcp 1521 1522 // 1523 // JVMTI PopFrame support 1524 // 1525 1526 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 1527 __ empty_expression_stack(); 1528 // Set the popframe_processing bit in pending_popframe_condition 1529 // indicating that we are currently handling popframe, so that 1530 // call_VMs that may happen later do not trigger new popframe 1531 // handling cycles. 1532 __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset())); 1533 __ orl(rdx, JavaThread::popframe_processing_bit); 1534 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx); 1535 1536 { 1537 // Check to see whether we are returning to a deoptimized frame. 1538 // (The PopFrame call ensures that the caller of the popped frame is 1539 // either interpreted or compiled and deoptimizes it if compiled.) 1540 // In this case, we can't call dispatch_next() after the frame is 1541 // popped, but instead must save the incoming arguments and restore 1542 // them after deoptimization has occurred. 1543 // 1544 // Note that we don't compare the return PC against the 1545 // deoptimization blob's unpack entry because of the presence of 1546 // adapter frames in C2. 1547 Label caller_not_deoptimized; 1548 __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize)); 1549 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1550 InterpreterRuntime::interpreter_contains), c_rarg1); 1551 __ testl(rax, rax); 1552 __ jcc(Assembler::notZero, caller_not_deoptimized); 1553 1554 // Compute size of arguments for saving when returning to 1555 // deoptimized caller 1556 __ get_method(rax); 1557 __ movptr(rax, Address(rax, Method::const_offset())); 1558 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod:: 1559 size_of_parameters_offset()))); 1560 __ shll(rax, Interpreter::logStackElementSize); 1561 __ restore_locals(); // XXX do we need this? 1562 __ subptr(r14, rax); 1563 __ addptr(r14, wordSize); 1564 // Save these arguments 1565 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1566 Deoptimization:: 1567 popframe_preserve_args), 1568 r15_thread, rax, r14); 1569 1570 __ remove_activation(vtos, rdx, 1571 /* throw_monitor_exception */ false, 1572 /* install_monitor_exception */ false, 1573 /* notify_jvmdi */ false); 1574 1575 // Inform deoptimization that it is responsible for restoring 1576 // these arguments 1577 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), 1578 JavaThread::popframe_force_deopt_reexecution_bit); 1579 1580 // Continue in deoptimization handler 1581 __ jmp(rdx); 1582 1583 __ bind(caller_not_deoptimized); 1584 } 1585 1586 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */ 1587 /* throw_monitor_exception */ false, 1588 /* install_monitor_exception */ false, 1589 /* notify_jvmdi */ false); 1590 1591 // Finish with popframe handling 1592 // A previous I2C followed by a deoptimization might have moved the 1593 // outgoing arguments further up the stack. PopFrame expects the 1594 // mutations to those outgoing arguments to be preserved and other 1595 // constraints basically require this frame to look exactly as 1596 // though it had previously invoked an interpreted activation with 1597 // no space between the top of the expression stack (current 1598 // last_sp) and the top of stack. Rather than force deopt to 1599 // maintain this kind of invariant all the time we call a small 1600 // fixup routine to move the mutated arguments onto the top of our 1601 // expression stack if necessary. 1602 __ mov(c_rarg1, rsp); 1603 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 1604 // PC must point into interpreter here 1605 __ set_last_Java_frame(noreg, rbp, __ pc()); 1606 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2); 1607 __ reset_last_Java_frame(true, true); 1608 // Restore the last_sp and null it out 1609 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 1610 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 1611 1612 __ restore_bcp(); // XXX do we need this? 1613 __ restore_locals(); // XXX do we need this? 1614 // The method data pointer was incremented already during 1615 // call profiling. We have to restore the mdp for the current bcp. 1616 if (ProfileInterpreter) { 1617 __ set_method_data_pointer_for_bcp(); 1618 } 1619 1620 // Clear the popframe condition flag 1621 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), 1622 JavaThread::popframe_inactive); 1623 1624 #if INCLUDE_JVMTI 1625 { 1626 Label L_done; 1627 const Register local0 = r14; 1628 1629 __ cmpb(Address(r13, 0), Bytecodes::_invokestatic); 1630 __ jcc(Assembler::notEqual, L_done); 1631 1632 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. 1633 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL. 1634 1635 __ get_method(rdx); 1636 __ movptr(rax, Address(local0, 0)); 1637 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, r13); 1638 1639 __ testptr(rax, rax); 1640 __ jcc(Assembler::zero, L_done); 1641 1642 __ movptr(Address(rbx, 0), rax); 1643 __ bind(L_done); 1644 } 1645 #endif // INCLUDE_JVMTI 1646 1647 __ dispatch_next(vtos); 1648 // end of PopFrame support 1649 1650 Interpreter::_remove_activation_entry = __ pc(); 1651 1652 // preserve exception over this code sequence 1653 __ pop_ptr(rax); 1654 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax); 1655 // remove the activation (without doing throws on illegalMonitorExceptions) 1656 __ remove_activation(vtos, rdx, false, true, false); 1657 // restore exception 1658 __ get_vm_result(rax, r15_thread); 1659 1660 // In between activations - previous activation type unknown yet 1661 // compute continuation point - the continuation point expects the 1662 // following registers set up: 1663 // 1664 // rax: exception 1665 // rdx: return address/pc that threw exception 1666 // rsp: expression stack of caller 1667 // rbp: ebp of caller 1668 __ push(rax); // save exception 1669 __ push(rdx); // save return address 1670 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1671 SharedRuntime::exception_handler_for_return_address), 1672 r15_thread, rdx); 1673 __ mov(rbx, rax); // save exception handler 1674 __ pop(rdx); // restore return address 1675 __ pop(rax); // restore exception 1676 // Note that an "issuing PC" is actually the next PC after the call 1677 __ jmp(rbx); // jump to exception 1678 // handler of caller 1679 } 1680 1681 1682 // 1683 // JVMTI ForceEarlyReturn support 1684 // 1685 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 1686 address entry = __ pc(); 1687 1688 __ restore_bcp(); 1689 __ restore_locals(); 1690 __ empty_expression_stack(); 1691 __ load_earlyret_value(state); 1692 1693 __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset())); 1694 Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset()); 1695 1696 // Clear the earlyret state 1697 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive); 1698 1699 __ remove_activation(state, rsi, 1700 false, /* throw_monitor_exception */ 1701 false, /* install_monitor_exception */ 1702 true); /* notify_jvmdi */ 1703 __ jmp(rsi); 1704 1705 return entry; 1706 } // end of ForceEarlyReturn support 1707 1708 1709 //----------------------------------------------------------------------------- 1710 // Helper for vtos entry point generation 1711 1712 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, 1713 address& bep, 1714 address& cep, 1715 address& sep, 1716 address& aep, 1717 address& iep, 1718 address& lep, 1719 address& fep, 1720 address& dep, 1721 address& vep) { 1722 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 1723 Label L; 1724 aep = __ pc(); __ push_ptr(); __ jmp(L); 1725 fep = __ pc(); __ push_f(xmm0); __ jmp(L); 1726 dep = __ pc(); __ push_d(xmm0); __ jmp(L); 1727 lep = __ pc(); __ push_l(); __ jmp(L); 1728 bep = cep = sep = 1729 iep = __ pc(); __ push_i(); 1730 vep = __ pc(); 1731 __ bind(L); 1732 generate_and_dispatch(t); 1733 } 1734 1735 1736 //----------------------------------------------------------------------------- 1737 // Generation of individual instructions 1738 1739 // helpers for generate_and_dispatch 1740 1741 1742 InterpreterGenerator::InterpreterGenerator(StubQueue* code) 1743 : TemplateInterpreterGenerator(code) { 1744 generate_all(); // down here so it can be "virtual" 1745 } 1746 1747 //----------------------------------------------------------------------------- 1748 1749 // Non-product code 1750 #ifndef PRODUCT 1751 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 1752 address entry = __ pc(); 1753 1754 __ push(state); 1755 __ push(c_rarg0); 1756 __ push(c_rarg1); 1757 __ push(c_rarg2); 1758 __ push(c_rarg3); 1759 __ mov(c_rarg2, rax); // Pass itos 1760 #ifdef _WIN64 1761 __ movflt(xmm3, xmm0); // Pass ftos 1762 #endif 1763 __ call_VM(noreg, 1764 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), 1765 c_rarg1, c_rarg2, c_rarg3); 1766 __ pop(c_rarg3); 1767 __ pop(c_rarg2); 1768 __ pop(c_rarg1); 1769 __ pop(c_rarg0); 1770 __ pop(state); 1771 __ ret(0); // return from result handler 1772 1773 return entry; 1774 } 1775 1776 void TemplateInterpreterGenerator::count_bytecode() { 1777 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value)); 1778 } 1779 1780 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 1781 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()])); 1782 } 1783 1784 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 1785 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index)); 1786 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes); 1787 __ orl(rbx, 1788 ((int) t->bytecode()) << 1789 BytecodePairHistogram::log2_number_of_codes); 1790 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx); 1791 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters)); 1792 __ incrementl(Address(rscratch1, rbx, Address::times_4)); 1793 } 1794 1795 1796 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 1797 // Call a little run-time stub to avoid blow-up for each bytecode. 1798 // The run-time runtime saves the right registers, depending on 1799 // the tosca in-state for the given template. 1800 1801 assert(Interpreter::trace_code(t->tos_in()) != NULL, 1802 "entry must have been generated"); 1803 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) 1804 __ andptr(rsp, -16); // align stack as required by ABI 1805 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in()))); 1806 __ mov(rsp, r12); // restore sp 1807 __ reinit_heapbase(); 1808 } 1809 1810 1811 void TemplateInterpreterGenerator::stop_interpreter_at() { 1812 Label L; 1813 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value), 1814 StopInterpreterAt); 1815 __ jcc(Assembler::notEqual, L); 1816 __ int3(); 1817 __ bind(L); 1818 } 1819 #endif // !PRODUCT 1820 #endif // ! CC_INTERP