1 /* 2 * Copyright (c) 2003, 2011, 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/assembler.hpp" 27 #include "interpreter/bytecodeHistogram.hpp" 28 #include "interpreter/interpreter.hpp" 29 #include "interpreter/interpreterGenerator.hpp" 30 #include "interpreter/interpreterRuntime.hpp" 31 #include "interpreter/templateTable.hpp" 32 #include "oops/arrayOop.hpp" 33 #include "oops/methodDataOop.hpp" 34 #include "oops/methodOop.hpp" 35 #include "oops/oop.inline.hpp" 36 #include "prims/jvmtiExport.hpp" 37 #include "prims/jvmtiThreadState.hpp" 38 #include "runtime/arguments.hpp" 39 #include "runtime/deoptimization.hpp" 40 #include "runtime/frame.inline.hpp" 41 #include "runtime/sharedRuntime.hpp" 42 #include "runtime/stubRoutines.hpp" 43 #include "runtime/synchronizer.hpp" 44 #include "runtime/timer.hpp" 45 #include "runtime/vframeArray.hpp" 46 #include "utilities/debug.hpp" 47 48 #define __ _masm-> 49 50 #ifndef CC_INTERP 51 52 const int method_offset = frame::interpreter_frame_method_offset * wordSize; 53 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize; 54 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize; 55 56 //----------------------------------------------------------------------------- 57 58 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 59 address entry = __ pc(); 60 61 #ifdef ASSERT 62 { 63 Label L; 64 __ lea(rax, Address(rbp, 65 frame::interpreter_frame_monitor_block_top_offset * 66 wordSize)); 67 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack 68 // grows negative) 69 __ jcc(Assembler::aboveEqual, L); // check if frame is complete 70 __ stop ("interpreter frame not set up"); 71 __ bind(L); 72 } 73 #endif // ASSERT 74 // Restore bcp under the assumption that the current frame is still 75 // interpreted 76 __ restore_bcp(); 77 78 // expression stack must be empty before entering the VM if an 79 // exception happened 80 __ empty_expression_stack(); 81 // throw exception 82 __ call_VM(noreg, 83 CAST_FROM_FN_PTR(address, 84 InterpreterRuntime::throw_StackOverflowError)); 85 return entry; 86 } 87 88 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler( 89 const char* name) { 90 address entry = __ pc(); 91 // expression stack must be empty before entering the VM if an 92 // exception happened 93 __ empty_expression_stack(); 94 // setup parameters 95 // ??? convention: expect aberrant index in register ebx 96 __ lea(c_rarg1, ExternalAddress((address)name)); 97 __ call_VM(noreg, 98 CAST_FROM_FN_PTR(address, 99 InterpreterRuntime:: 100 throw_ArrayIndexOutOfBoundsException), 101 c_rarg1, rbx); 102 return entry; 103 } 104 105 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 106 address entry = __ pc(); 107 108 // object is at TOS 109 __ pop(c_rarg1); 110 111 // expression stack must be empty before entering the VM if an 112 // exception happened 113 __ empty_expression_stack(); 114 115 __ call_VM(noreg, 116 CAST_FROM_FN_PTR(address, 117 InterpreterRuntime:: 118 throw_ClassCastException), 119 c_rarg1); 120 return entry; 121 } 122 123 // Arguments are: required type at TOS+8, failing object (or NULL) at TOS+4. 124 address TemplateInterpreterGenerator::generate_WrongMethodType_handler() { 125 address entry = __ pc(); 126 127 __ pop(c_rarg2); // failing object is at TOS 128 __ pop(c_rarg1); // required type is at TOS+8 129 130 __ verify_oop(c_rarg1); 131 __ verify_oop(c_rarg2); 132 133 // Various method handle types use interpreter registers as temps. 134 __ restore_bcp(); 135 __ restore_locals(); 136 137 // Expression stack must be empty before entering the VM for an exception. 138 __ empty_expression_stack(); 139 140 __ call_VM(noreg, 141 CAST_FROM_FN_PTR(address, 142 InterpreterRuntime::throw_WrongMethodTypeException), 143 // pass required type, failing object (or NULL) 144 c_rarg1, c_rarg2); 145 return entry; 146 } 147 148 address TemplateInterpreterGenerator::generate_exception_handler_common( 149 const char* name, const char* message, bool pass_oop) { 150 assert(!pass_oop || message == NULL, "either oop or message but not both"); 151 address entry = __ pc(); 152 if (pass_oop) { 153 // object is at TOS 154 __ pop(c_rarg2); 155 } 156 // expression stack must be empty before entering the VM if an 157 // exception happened 158 __ empty_expression_stack(); 159 // setup parameters 160 __ lea(c_rarg1, ExternalAddress((address)name)); 161 if (pass_oop) { 162 __ call_VM(rax, CAST_FROM_FN_PTR(address, 163 InterpreterRuntime:: 164 create_klass_exception), 165 c_rarg1, c_rarg2); 166 } else { 167 // kind of lame ExternalAddress can't take NULL because 168 // external_word_Relocation will assert. 169 if (message != NULL) { 170 __ lea(c_rarg2, ExternalAddress((address)message)); 171 } else { 172 __ movptr(c_rarg2, NULL_WORD); 173 } 174 __ call_VM(rax, 175 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), 176 c_rarg1, c_rarg2); 177 } 178 // throw exception 179 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); 180 return entry; 181 } 182 183 184 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { 185 address entry = __ pc(); 186 // NULL last_sp until next java call 187 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 188 __ dispatch_next(state); 189 return entry; 190 } 191 192 193 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) { 194 address entry = __ pc(); 195 196 // Restore stack bottom in case i2c adjusted stack 197 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 198 // and NULL it as marker that esp is now tos until next java call 199 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 200 201 __ restore_bcp(); 202 __ restore_locals(); 203 204 Label L_got_cache, L_giant_index; 205 if (EnableInvokeDynamic) { 206 __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic); 207 __ jcc(Assembler::equal, L_giant_index); 208 } 209 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2)); 210 __ bind(L_got_cache); 211 __ movl(rbx, Address(rbx, rcx, 212 Address::times_ptr, 213 in_bytes(constantPoolCacheOopDesc::base_offset()) + 214 3 * wordSize)); 215 __ andl(rbx, 0xFF); 216 __ lea(rsp, Address(rsp, rbx, Address::times_8)); 217 __ dispatch_next(state, step); 218 219 // out of the main line of code... 220 if (EnableInvokeDynamic) { 221 __ bind(L_giant_index); 222 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4)); 223 __ jmp(L_got_cache); 224 } 225 226 return entry; 227 } 228 229 230 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, 231 int step) { 232 address entry = __ pc(); 233 // NULL last_sp until next java call 234 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 235 __ restore_bcp(); 236 __ restore_locals(); 237 // handle exceptions 238 { 239 Label L; 240 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); 241 __ jcc(Assembler::zero, L); 242 __ call_VM(noreg, 243 CAST_FROM_FN_PTR(address, 244 InterpreterRuntime::throw_pending_exception)); 245 __ should_not_reach_here(); 246 __ bind(L); 247 } 248 __ dispatch_next(state, step); 249 return entry; 250 } 251 252 int AbstractInterpreter::BasicType_as_index(BasicType type) { 253 int i = 0; 254 switch (type) { 255 case T_BOOLEAN: i = 0; break; 256 case T_CHAR : i = 1; break; 257 case T_BYTE : i = 2; break; 258 case T_SHORT : i = 3; break; 259 case T_INT : i = 4; break; 260 case T_LONG : i = 5; break; 261 case T_VOID : i = 6; break; 262 case T_FLOAT : i = 7; break; 263 case T_DOUBLE : i = 8; break; 264 case T_OBJECT : i = 9; break; 265 case T_ARRAY : i = 9; break; 266 default : ShouldNotReachHere(); 267 } 268 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, 269 "index out of bounds"); 270 return i; 271 } 272 273 274 address TemplateInterpreterGenerator::generate_result_handler_for( 275 BasicType type) { 276 address entry = __ pc(); 277 switch (type) { 278 case T_BOOLEAN: __ c2bool(rax); break; 279 case T_CHAR : __ movzwl(rax, rax); break; 280 case T_BYTE : __ sign_extend_byte(rax); break; 281 case T_SHORT : __ sign_extend_short(rax); break; 282 case T_INT : /* nothing to do */ break; 283 case T_LONG : /* nothing to do */ break; 284 case T_VOID : /* nothing to do */ break; 285 case T_FLOAT : /* nothing to do */ break; 286 case T_DOUBLE : /* nothing to do */ break; 287 case T_OBJECT : 288 // retrieve result from frame 289 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize)); 290 // and verify it 291 __ verify_oop(rax); 292 break; 293 default : ShouldNotReachHere(); 294 } 295 __ ret(0); // return from result handler 296 return entry; 297 } 298 299 address TemplateInterpreterGenerator::generate_safept_entry_for( 300 TosState state, 301 address runtime_entry) { 302 address entry = __ pc(); 303 __ push(state); 304 __ call_VM(noreg, runtime_entry); 305 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 306 return entry; 307 } 308 309 310 311 // Helpers for commoning out cases in the various type of method entries. 312 // 313 314 315 // increment invocation count & check for overflow 316 // 317 // Note: checking for negative value instead of overflow 318 // so we have a 'sticky' overflow test 319 // 320 // rbx: method 321 // ecx: invocation counter 322 // 323 void InterpreterGenerator::generate_counter_incr( 324 Label* overflow, 325 Label* profile_method, 326 Label* profile_method_continue) { 327 const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) + 328 in_bytes(InvocationCounter::counter_offset())); 329 // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not. 330 if (TieredCompilation) { 331 int increment = InvocationCounter::count_increment; 332 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift; 333 Label no_mdo, done; 334 if (ProfileInterpreter) { 335 // Are we profiling? 336 __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset())); 337 __ testptr(rax, rax); 338 __ jccb(Assembler::zero, no_mdo); 339 // Increment counter in the MDO 340 const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) + 341 in_bytes(InvocationCounter::counter_offset())); 342 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow); 343 __ jmpb(done); 344 } 345 __ bind(no_mdo); 346 // Increment counter in methodOop (we don't need to load it, it's in ecx). 347 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow); 348 __ bind(done); 349 } else { 350 const Address backedge_counter(rbx, 351 methodOopDesc::backedge_counter_offset() + 352 InvocationCounter::counter_offset()); 353 354 if (ProfileInterpreter) { // %%% Merge this into methodDataOop 355 __ incrementl(Address(rbx, 356 methodOopDesc::interpreter_invocation_counter_offset())); 357 } 358 // Update standard invocation counters 359 __ movl(rax, backedge_counter); // load backedge counter 360 361 __ incrementl(rcx, InvocationCounter::count_increment); 362 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits 363 364 __ movl(invocation_counter, rcx); // save invocation count 365 __ addl(rcx, rax); // add both counters 366 367 // profile_method is non-null only for interpreted method so 368 // profile_method != NULL == !native_call 369 370 if (ProfileInterpreter && profile_method != NULL) { 371 // Test to see if we should create a method data oop 372 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit)); 373 __ jcc(Assembler::less, *profile_method_continue); 374 375 // if no method data exists, go to profile_method 376 __ test_method_data_pointer(rax, *profile_method); 377 } 378 379 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit)); 380 __ jcc(Assembler::aboveEqual, *overflow); 381 } 382 } 383 384 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) { 385 386 // Asm interpreter on entry 387 // r14 - locals 388 // r13 - bcp 389 // rbx - method 390 // edx - cpool --- DOES NOT APPEAR TO BE TRUE 391 // rbp - interpreter frame 392 393 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] 394 // Everything as it was on entry 395 // rdx is not restored. Doesn't appear to really be set. 396 397 const Address size_of_parameters(rbx, 398 methodOopDesc::size_of_parameters_offset()); 399 400 // InterpreterRuntime::frequency_counter_overflow takes two 401 // arguments, the first (thread) is passed by call_VM, the second 402 // indicates if the counter overflow occurs at a backwards branch 403 // (NULL bcp). We pass zero for it. The call returns the address 404 // of the verified entry point for the method or NULL if the 405 // compilation did not complete (either went background or bailed 406 // out). 407 __ movl(c_rarg1, 0); 408 __ call_VM(noreg, 409 CAST_FROM_FN_PTR(address, 410 InterpreterRuntime::frequency_counter_overflow), 411 c_rarg1); 412 413 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop 414 // Preserve invariant that r13/r14 contain bcp/locals of sender frame 415 // and jump to the interpreted entry. 416 __ jmp(*do_continue, relocInfo::none); 417 } 418 419 // See if we've got enough room on the stack for locals plus overhead. 420 // The expression stack grows down incrementally, so the normal guard 421 // page mechanism will work for that. 422 // 423 // NOTE: Since the additional locals are also always pushed (wasn't 424 // obvious in generate_method_entry) so the guard should work for them 425 // too. 426 // 427 // Args: 428 // rdx: number of additional locals this frame needs (what we must check) 429 // rbx: methodOop 430 // 431 // Kills: 432 // rax 433 void InterpreterGenerator::generate_stack_overflow_check(void) { 434 435 // monitor entry size: see picture of stack set 436 // (generate_method_entry) and frame_amd64.hpp 437 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 438 439 // total overhead size: entry_size + (saved rbp through expr stack 440 // bottom). be sure to change this if you add/subtract anything 441 // to/from the overhead area 442 const int overhead_size = 443 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size; 444 445 const int page_size = os::vm_page_size(); 446 447 Label after_frame_check; 448 449 // see if the frame is greater than one page in size. If so, 450 // then we need to verify there is enough stack space remaining 451 // for the additional locals. 452 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize); 453 __ jcc(Assembler::belowEqual, after_frame_check); 454 455 // compute rsp as if this were going to be the last frame on 456 // the stack before the red zone 457 458 const Address stack_base(r15_thread, Thread::stack_base_offset()); 459 const Address stack_size(r15_thread, Thread::stack_size_offset()); 460 461 // locals + overhead, in bytes 462 __ mov(rax, rdx); 463 __ shlptr(rax, Interpreter::logStackElementSize); // 2 slots per parameter. 464 __ addptr(rax, overhead_size); 465 466 #ifdef ASSERT 467 Label stack_base_okay, stack_size_okay; 468 // verify that thread stack base is non-zero 469 __ cmpptr(stack_base, (int32_t)NULL_WORD); 470 __ jcc(Assembler::notEqual, stack_base_okay); 471 __ stop("stack base is zero"); 472 __ bind(stack_base_okay); 473 // verify that thread stack size is non-zero 474 __ cmpptr(stack_size, 0); 475 __ jcc(Assembler::notEqual, stack_size_okay); 476 __ stop("stack size is zero"); 477 __ bind(stack_size_okay); 478 #endif 479 480 // Add stack base to locals and subtract stack size 481 __ addptr(rax, stack_base); 482 __ subptr(rax, stack_size); 483 484 // Use the maximum number of pages we might bang. 485 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages : 486 (StackRedPages+StackYellowPages); 487 488 // add in the red and yellow zone sizes 489 __ addptr(rax, max_pages * page_size); 490 491 // check against the current stack bottom 492 __ cmpptr(rsp, rax); 493 __ jcc(Assembler::above, after_frame_check); 494 495 __ pop(rax); // get return address 496 __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry())); 497 498 // all done with frame size check 499 __ bind(after_frame_check); 500 } 501 502 // Allocate monitor and lock method (asm interpreter) 503 // 504 // Args: 505 // rbx: methodOop 506 // r14: locals 507 // 508 // Kills: 509 // rax 510 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs) 511 // rscratch1, rscratch2 (scratch regs) 512 void InterpreterGenerator::lock_method(void) { 513 // synchronize method 514 const Address access_flags(rbx, methodOopDesc::access_flags_offset()); 515 const Address monitor_block_top( 516 rbp, 517 frame::interpreter_frame_monitor_block_top_offset * wordSize); 518 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 519 520 #ifdef ASSERT 521 { 522 Label L; 523 __ movl(rax, access_flags); 524 __ testl(rax, JVM_ACC_SYNCHRONIZED); 525 __ jcc(Assembler::notZero, L); 526 __ stop("method doesn't need synchronization"); 527 __ bind(L); 528 } 529 #endif // ASSERT 530 531 // get synchronization object 532 { 533 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + 534 Klass::java_mirror_offset_in_bytes(); 535 Label done; 536 __ movl(rax, access_flags); 537 __ testl(rax, JVM_ACC_STATIC); 538 // get receiver (assume this is frequent case) 539 __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0))); 540 __ jcc(Assembler::zero, done); 541 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset())); 542 __ movptr(rax, Address(rax, 543 constantPoolOopDesc::pool_holder_offset_in_bytes())); 544 __ movptr(rax, Address(rax, mirror_offset)); 545 546 #ifdef ASSERT 547 { 548 Label L; 549 __ testptr(rax, rax); 550 __ jcc(Assembler::notZero, L); 551 __ stop("synchronization object is NULL"); 552 __ bind(L); 553 } 554 #endif // ASSERT 555 556 __ bind(done); 557 } 558 559 // add space for monitor & lock 560 __ subptr(rsp, entry_size); // add space for a monitor entry 561 __ movptr(monitor_block_top, rsp); // set new monitor block top 562 // store object 563 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); 564 __ movptr(c_rarg1, rsp); // object address 565 __ lock_object(c_rarg1); 566 } 567 568 // Generate a fixed interpreter frame. This is identical setup for 569 // interpreted methods and for native methods hence the shared code. 570 // 571 // Args: 572 // rax: return address 573 // rbx: methodOop 574 // r14: pointer to locals 575 // r13: sender sp 576 // rdx: cp cache 577 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { 578 // initialize fixed part of activation frame 579 __ push(rax); // save return address 580 __ enter(); // save old & set new rbp 581 __ push(r13); // set sender sp 582 __ push((int)NULL_WORD); // leave last_sp as null 583 __ movptr(r13, Address(rbx, methodOopDesc::const_offset())); // get constMethodOop 584 __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase 585 __ push(rbx); // save methodOop 586 if (ProfileInterpreter) { 587 Label method_data_continue; 588 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset()))); 589 __ testptr(rdx, rdx); 590 __ jcc(Assembler::zero, method_data_continue); 591 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset())); 592 __ bind(method_data_continue); 593 __ push(rdx); // set the mdp (method data pointer) 594 } else { 595 __ push(0); 596 } 597 598 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset())); 599 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes())); 600 __ push(rdx); // set constant pool cache 601 __ push(r14); // set locals pointer 602 if (native_call) { 603 __ push(0); // no bcp 604 } else { 605 __ push(r13); // set bcp 606 } 607 __ push(0); // reserve word for pointer to expression stack bottom 608 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom 609 } 610 611 // End of helpers 612 613 // Various method entries 614 //------------------------------------------------------------------------------------------------------------------------ 615 // 616 // 617 618 // Call an accessor method (assuming it is resolved, otherwise drop 619 // into vanilla (slow path) entry 620 address InterpreterGenerator::generate_accessor_entry(void) { 621 // rbx: methodOop 622 623 // r13: senderSP must preserver for slow path, set SP to it on fast path 624 625 address entry_point = __ pc(); 626 Label xreturn_path; 627 628 // do fastpath for resolved accessor methods 629 if (UseFastAccessorMethods) { 630 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites 631 // thereof; parameter size = 1 632 // Note: We can only use this code if the getfield has been resolved 633 // and if we don't have a null-pointer exception => check for 634 // these conditions first and use slow path if necessary. 635 Label slow_path; 636 // If we need a safepoint check, generate full interpreter entry. 637 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 638 SafepointSynchronize::_not_synchronized); 639 640 __ jcc(Assembler::notEqual, slow_path); 641 // rbx: method 642 __ movptr(rax, Address(rsp, wordSize)); 643 644 // check if local 0 != NULL and read field 645 __ testptr(rax, rax); 646 __ jcc(Assembler::zero, slow_path); 647 648 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset())); 649 // read first instruction word and extract bytecode @ 1 and index @ 2 650 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset())); 651 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset())); 652 // Shift codes right to get the index on the right. 653 // The bytecode fetched looks like <index><0xb4><0x2a> 654 __ shrl(rdx, 2 * BitsPerByte); 655 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); 656 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes())); 657 658 // rax: local 0 659 // rbx: method 660 // rdx: constant pool cache index 661 // rdi: constant pool cache 662 663 // check if getfield has been resolved and read constant pool cache entry 664 // check the validity of the cache entry by testing whether _indices field 665 // contains Bytecode::_getfield in b1 byte. 666 assert(in_words(ConstantPoolCacheEntry::size()) == 4, 667 "adjust shift below"); 668 __ movl(rcx, 669 Address(rdi, 670 rdx, 671 Address::times_8, 672 constantPoolCacheOopDesc::base_offset() + 673 ConstantPoolCacheEntry::indices_offset())); 674 __ shrl(rcx, 2 * BitsPerByte); 675 __ andl(rcx, 0xFF); 676 __ cmpl(rcx, Bytecodes::_getfield); 677 __ jcc(Assembler::notEqual, slow_path); 678 679 // Note: constant pool entry is not valid before bytecode is resolved 680 __ movptr(rcx, 681 Address(rdi, 682 rdx, 683 Address::times_8, 684 constantPoolCacheOopDesc::base_offset() + 685 ConstantPoolCacheEntry::f2_offset())); 686 // edx: flags 687 __ movl(rdx, 688 Address(rdi, 689 rdx, 690 Address::times_8, 691 constantPoolCacheOopDesc::base_offset() + 692 ConstantPoolCacheEntry::flags_offset())); 693 694 Label notObj, notInt, notByte, notShort; 695 const Address field_address(rax, rcx, Address::times_1); 696 697 // Need to differentiate between igetfield, agetfield, bgetfield etc. 698 // because they are different sizes. 699 // Use the type from the constant pool cache 700 __ shrl(rdx, ConstantPoolCacheEntry::tosBits); 701 // Make sure we don't need to mask edx for tosBits after the above shift 702 ConstantPoolCacheEntry::verify_tosBits(); 703 704 __ cmpl(rdx, atos); 705 __ jcc(Assembler::notEqual, notObj); 706 // atos 707 __ load_heap_oop(rax, field_address); 708 __ jmp(xreturn_path); 709 710 __ bind(notObj); 711 __ cmpl(rdx, itos); 712 __ jcc(Assembler::notEqual, notInt); 713 // itos 714 __ movl(rax, field_address); 715 __ jmp(xreturn_path); 716 717 __ bind(notInt); 718 __ cmpl(rdx, btos); 719 __ jcc(Assembler::notEqual, notByte); 720 // btos 721 __ load_signed_byte(rax, field_address); 722 __ jmp(xreturn_path); 723 724 __ bind(notByte); 725 __ cmpl(rdx, stos); 726 __ jcc(Assembler::notEqual, notShort); 727 // stos 728 __ load_signed_short(rax, field_address); 729 __ jmp(xreturn_path); 730 731 __ bind(notShort); 732 #ifdef ASSERT 733 Label okay; 734 __ cmpl(rdx, ctos); 735 __ jcc(Assembler::equal, okay); 736 __ stop("what type is this?"); 737 __ bind(okay); 738 #endif 739 // ctos 740 __ load_unsigned_short(rax, field_address); 741 742 __ bind(xreturn_path); 743 744 // _ireturn/_areturn 745 __ pop(rdi); 746 __ mov(rsp, r13); 747 __ jmp(rdi); 748 __ ret(0); 749 750 // generate a vanilla interpreter entry as the slow path 751 __ bind(slow_path); 752 (void) generate_normal_entry(false); 753 } else { 754 (void) generate_normal_entry(false); 755 } 756 757 return entry_point; 758 } 759 760 // Method entry for java.lang.ref.Reference.get. 761 address InterpreterGenerator::generate_Reference_get_entry(void) { 762 #ifndef SERIALGC 763 // Code: _aload_0, _getfield, _areturn 764 // parameter size = 1 765 // 766 // The code that gets generated by this routine is split into 2 parts: 767 // 1. The "intrinsified" code for G1 (or any SATB based GC), 768 // 2. The slow path - which is an expansion of the regular method entry. 769 // 770 // Notes:- 771 // * In the G1 code we do not check whether we need to block for 772 // a safepoint. If G1 is enabled then we must execute the specialized 773 // code for Reference.get (except when the Reference object is null) 774 // so that we can log the value in the referent field with an SATB 775 // update buffer. 776 // If the code for the getfield template is modified so that the 777 // G1 pre-barrier code is executed when the current method is 778 // Reference.get() then going through the normal method entry 779 // will be fine. 780 // * The G1 code can, however, check the receiver object (the instance 781 // of java.lang.Reference) and jump to the slow path if null. If the 782 // Reference object is null then we obviously cannot fetch the referent 783 // and so we don't need to call the G1 pre-barrier. Thus we can use the 784 // regular method entry code to generate the NPE. 785 // 786 // This code is based on generate_accessor_enty. 787 // 788 // rbx: methodOop 789 790 // r13: senderSP must preserve for slow path, set SP to it on fast path 791 792 address entry = __ pc(); 793 794 const int referent_offset = java_lang_ref_Reference::referent_offset; 795 guarantee(referent_offset > 0, "referent offset not initialized"); 796 797 if (UseG1GC) { 798 Label slow_path; 799 // rbx: method 800 801 // Check if local 0 != NULL 802 // If the receiver is null then it is OK to jump to the slow path. 803 __ movptr(rax, Address(rsp, wordSize)); 804 805 __ testptr(rax, rax); 806 __ jcc(Assembler::zero, slow_path); 807 808 // rax: local 0 809 // rbx: method (but can be used as scratch now) 810 // rdx: scratch 811 // rdi: scratch 812 813 // Generate the G1 pre-barrier code to log the value of 814 // the referent field in an SATB buffer. 815 816 // Load the value of the referent field. 817 const Address field_address(rax, referent_offset); 818 __ load_heap_oop(rax, field_address); 819 820 // Generate the G1 pre-barrier code to log the value of 821 // the referent field in an SATB buffer. 822 __ g1_write_barrier_pre(noreg /* obj */, 823 rax /* pre_val */, 824 r15_thread /* thread */, 825 rbx /* tmp */, 826 true /* tosca_live */, 827 true /* expand_call */); 828 829 // _areturn 830 __ pop(rdi); // get return address 831 __ mov(rsp, r13); // set sp to sender sp 832 __ jmp(rdi); 833 __ ret(0); 834 835 // generate a vanilla interpreter entry as the slow path 836 __ bind(slow_path); 837 (void) generate_normal_entry(false); 838 839 return entry; 840 } 841 #endif // SERIALGC 842 843 // If G1 is not enabled then attempt to go through the accessor entry point 844 // Reference.get is an accessor 845 return generate_accessor_entry(); 846 } 847 848 849 // Interpreter stub for calling a native method. (asm interpreter) 850 // This sets up a somewhat different looking stack for calling the 851 // native method than the typical interpreter frame setup. 852 address InterpreterGenerator::generate_native_entry(bool synchronized) { 853 // determine code generation flags 854 bool inc_counter = UseCompiler || CountCompiledCalls; 855 856 // rbx: methodOop 857 // r13: sender sp 858 859 address entry_point = __ pc(); 860 861 const Address size_of_parameters(rbx, methodOopDesc:: 862 size_of_parameters_offset()); 863 const Address invocation_counter(rbx, methodOopDesc:: 864 invocation_counter_offset() + 865 InvocationCounter::counter_offset()); 866 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); 867 868 // get parameter size (always needed) 869 __ load_unsigned_short(rcx, size_of_parameters); 870 871 // native calls don't need the stack size check since they have no 872 // expression stack and the arguments are already on the stack and 873 // we only add a handful of words to the stack 874 875 // rbx: methodOop 876 // rcx: size of parameters 877 // r13: sender sp 878 __ pop(rax); // get return address 879 880 // for natives the size of locals is zero 881 882 // compute beginning of parameters (r14) 883 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize)); 884 885 // add 2 zero-initialized slots for native calls 886 // initialize result_handler slot 887 __ push((int) NULL_WORD); 888 // slot for oop temp 889 // (static native method holder mirror/jni oop result) 890 __ push((int) NULL_WORD); 891 892 if (inc_counter) { 893 __ movl(rcx, invocation_counter); // (pre-)fetch invocation count 894 } 895 896 // initialize fixed part of activation frame 897 generate_fixed_frame(true); 898 899 // make sure method is native & not abstract 900 #ifdef ASSERT 901 __ movl(rax, access_flags); 902 { 903 Label L; 904 __ testl(rax, JVM_ACC_NATIVE); 905 __ jcc(Assembler::notZero, L); 906 __ stop("tried to execute non-native method as native"); 907 __ bind(L); 908 } 909 { 910 Label L; 911 __ testl(rax, JVM_ACC_ABSTRACT); 912 __ jcc(Assembler::zero, L); 913 __ stop("tried to execute abstract method in interpreter"); 914 __ bind(L); 915 } 916 #endif 917 918 // Since at this point in the method invocation the exception handler 919 // would try to exit the monitor of synchronized methods which hasn't 920 // been entered yet, we set the thread local variable 921 // _do_not_unlock_if_synchronized to true. The remove_activation will 922 // check this flag. 923 924 const Address do_not_unlock_if_synchronized(r15_thread, 925 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 926 __ movbool(do_not_unlock_if_synchronized, true); 927 928 // increment invocation count & check for overflow 929 Label invocation_counter_overflow; 930 if (inc_counter) { 931 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); 932 } 933 934 Label continue_after_compile; 935 __ bind(continue_after_compile); 936 937 bang_stack_shadow_pages(true); 938 939 // reset the _do_not_unlock_if_synchronized flag 940 __ movbool(do_not_unlock_if_synchronized, false); 941 942 // check for synchronized methods 943 // Must happen AFTER invocation_counter check and stack overflow check, 944 // so method is not locked if overflows. 945 if (synchronized) { 946 lock_method(); 947 } else { 948 // no synchronization necessary 949 #ifdef ASSERT 950 { 951 Label L; 952 __ movl(rax, access_flags); 953 __ testl(rax, JVM_ACC_SYNCHRONIZED); 954 __ jcc(Assembler::zero, L); 955 __ stop("method needs synchronization"); 956 __ bind(L); 957 } 958 #endif 959 } 960 961 // start execution 962 #ifdef ASSERT 963 { 964 Label L; 965 const Address monitor_block_top(rbp, 966 frame::interpreter_frame_monitor_block_top_offset * wordSize); 967 __ movptr(rax, monitor_block_top); 968 __ cmpptr(rax, rsp); 969 __ jcc(Assembler::equal, L); 970 __ stop("broken stack frame setup in interpreter"); 971 __ bind(L); 972 } 973 #endif 974 975 // jvmti support 976 __ notify_method_entry(); 977 978 // work registers 979 const Register method = rbx; 980 const Register t = r11; 981 982 // allocate space for parameters 983 __ get_method(method); 984 __ verify_oop(method); 985 __ load_unsigned_short(t, 986 Address(method, 987 methodOopDesc::size_of_parameters_offset())); 988 __ shll(t, Interpreter::logStackElementSize); 989 990 __ subptr(rsp, t); 991 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 992 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI) 993 994 // get signature handler 995 { 996 Label L; 997 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); 998 __ testptr(t, t); 999 __ jcc(Assembler::notZero, L); 1000 __ call_VM(noreg, 1001 CAST_FROM_FN_PTR(address, 1002 InterpreterRuntime::prepare_native_call), 1003 method); 1004 __ get_method(method); 1005 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); 1006 __ bind(L); 1007 } 1008 1009 // call signature handler 1010 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14, 1011 "adjust this code"); 1012 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp, 1013 "adjust this code"); 1014 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1, 1015 "adjust this code"); 1016 1017 // The generated handlers do not touch RBX (the method oop). 1018 // However, large signatures cannot be cached and are generated 1019 // each time here. The slow-path generator can do a GC on return, 1020 // so we must reload it after the call. 1021 __ call(t); 1022 __ get_method(method); // slow path can do a GC, reload RBX 1023 1024 1025 // result handler is in rax 1026 // set result handler 1027 __ movptr(Address(rbp, 1028 (frame::interpreter_frame_result_handler_offset) * wordSize), 1029 rax); 1030 1031 // pass mirror handle if static call 1032 { 1033 Label L; 1034 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + 1035 Klass::java_mirror_offset_in_bytes(); 1036 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); 1037 __ testl(t, JVM_ACC_STATIC); 1038 __ jcc(Assembler::zero, L); 1039 // get mirror 1040 __ movptr(t, Address(method, methodOopDesc::constants_offset())); 1041 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes())); 1042 __ movptr(t, Address(t, mirror_offset)); 1043 // copy mirror into activation frame 1044 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), 1045 t); 1046 // pass handle to mirror 1047 __ lea(c_rarg1, 1048 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize)); 1049 __ bind(L); 1050 } 1051 1052 // get native function entry point 1053 { 1054 Label L; 1055 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); 1056 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry()); 1057 __ movptr(rscratch2, unsatisfied.addr()); 1058 __ cmpptr(rax, rscratch2); 1059 __ jcc(Assembler::notEqual, L); 1060 __ call_VM(noreg, 1061 CAST_FROM_FN_PTR(address, 1062 InterpreterRuntime::prepare_native_call), 1063 method); 1064 __ get_method(method); 1065 __ verify_oop(method); 1066 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); 1067 __ bind(L); 1068 } 1069 1070 // pass JNIEnv 1071 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset())); 1072 1073 // It is enough that the pc() points into the right code 1074 // segment. It does not have to be the correct return pc. 1075 __ set_last_Java_frame(rsp, rbp, (address) __ pc()); 1076 1077 // change thread state 1078 #ifdef ASSERT 1079 { 1080 Label L; 1081 __ movl(t, Address(r15_thread, JavaThread::thread_state_offset())); 1082 __ cmpl(t, _thread_in_Java); 1083 __ jcc(Assembler::equal, L); 1084 __ stop("Wrong thread state in native stub"); 1085 __ bind(L); 1086 } 1087 #endif 1088 1089 // Change state to native 1090 1091 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), 1092 _thread_in_native); 1093 1094 // Call the native method. 1095 __ call(rax); 1096 // result potentially in rax or xmm0 1097 1098 // Depending on runtime options, either restore the MXCSR 1099 // register after returning from the JNI Call or verify that 1100 // it wasn't changed during -Xcheck:jni. 1101 if (RestoreMXCSROnJNICalls) { 1102 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std())); 1103 } 1104 else if (CheckJNICalls) { 1105 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry()))); 1106 } 1107 1108 // NOTE: The order of these pushes is known to frame::interpreter_frame_result 1109 // in order to extract the result of a method call. If the order of these 1110 // pushes change or anything else is added to the stack then the code in 1111 // interpreter_frame_result must also change. 1112 1113 __ push(dtos); 1114 __ push(ltos); 1115 1116 // change thread state 1117 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), 1118 _thread_in_native_trans); 1119 1120 if (os::is_MP()) { 1121 if (UseMembar) { 1122 // Force this write out before the read below 1123 __ membar(Assembler::Membar_mask_bits( 1124 Assembler::LoadLoad | Assembler::LoadStore | 1125 Assembler::StoreLoad | Assembler::StoreStore)); 1126 } else { 1127 // Write serialization page so VM thread can do a pseudo remote membar. 1128 // We use the current thread pointer to calculate a thread specific 1129 // offset to write to within the page. This minimizes bus traffic 1130 // due to cache line collision. 1131 __ serialize_memory(r15_thread, rscratch2); 1132 } 1133 } 1134 1135 // check for safepoint operation in progress and/or pending suspend requests 1136 { 1137 Label Continue; 1138 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 1139 SafepointSynchronize::_not_synchronized); 1140 1141 Label L; 1142 __ jcc(Assembler::notEqual, L); 1143 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0); 1144 __ jcc(Assembler::equal, Continue); 1145 __ bind(L); 1146 1147 // Don't use call_VM as it will see a possible pending exception 1148 // and forward it and never return here preventing us from 1149 // clearing _last_native_pc down below. Also can't use 1150 // call_VM_leaf either as it will check to see if r13 & r14 are 1151 // preserved and correspond to the bcp/locals pointers. So we do a 1152 // runtime call by hand. 1153 // 1154 __ mov(c_rarg0, r15_thread); 1155 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) 1156 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 1157 __ andptr(rsp, -16); // align stack as required by ABI 1158 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans))); 1159 __ mov(rsp, r12); // restore sp 1160 __ reinit_heapbase(); 1161 __ bind(Continue); 1162 } 1163 1164 // change thread state 1165 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java); 1166 1167 // reset_last_Java_frame 1168 __ reset_last_Java_frame(true, true); 1169 1170 // reset handle block 1171 __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset())); 1172 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD); 1173 1174 // If result is an oop unbox and store it in frame where gc will see it 1175 // and result handler will pick it up 1176 1177 { 1178 Label no_oop, store_result; 1179 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT))); 1180 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize)); 1181 __ jcc(Assembler::notEqual, no_oop); 1182 // retrieve result 1183 __ pop(ltos); 1184 __ testptr(rax, rax); 1185 __ jcc(Assembler::zero, store_result); 1186 __ movptr(rax, Address(rax, 0)); 1187 __ bind(store_result); 1188 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax); 1189 // keep stack depth as expected by pushing oop which will eventually be discarde 1190 __ push(ltos); 1191 __ bind(no_oop); 1192 } 1193 1194 1195 { 1196 Label no_reguard; 1197 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()), 1198 JavaThread::stack_guard_yellow_disabled); 1199 __ jcc(Assembler::notEqual, no_reguard); 1200 1201 __ pusha(); // XXX only save smashed registers 1202 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) 1203 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 1204 __ andptr(rsp, -16); // align stack as required by ABI 1205 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); 1206 __ mov(rsp, r12); // restore sp 1207 __ popa(); // XXX only restore smashed registers 1208 __ reinit_heapbase(); 1209 1210 __ bind(no_reguard); 1211 } 1212 1213 1214 // The method register is junk from after the thread_in_native transition 1215 // until here. Also can't call_VM until the bcp has been 1216 // restored. Need bcp for throwing exception below so get it now. 1217 __ get_method(method); 1218 __ verify_oop(method); 1219 1220 // restore r13 to have legal interpreter frame, i.e., bci == 0 <=> 1221 // r13 == code_base() 1222 __ movptr(r13, Address(method, methodOopDesc::const_offset())); // get constMethodOop 1223 __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase 1224 // handle exceptions (exception handling will handle unlocking!) 1225 { 1226 Label L; 1227 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); 1228 __ jcc(Assembler::zero, L); 1229 // Note: At some point we may want to unify this with the code 1230 // used in call_VM_base(); i.e., we should use the 1231 // StubRoutines::forward_exception code. For now this doesn't work 1232 // here because the rsp is not correctly set at this point. 1233 __ MacroAssembler::call_VM(noreg, 1234 CAST_FROM_FN_PTR(address, 1235 InterpreterRuntime::throw_pending_exception)); 1236 __ should_not_reach_here(); 1237 __ bind(L); 1238 } 1239 1240 // do unlocking if necessary 1241 { 1242 Label L; 1243 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); 1244 __ testl(t, JVM_ACC_SYNCHRONIZED); 1245 __ jcc(Assembler::zero, L); 1246 // the code below should be shared with interpreter macro 1247 // assembler implementation 1248 { 1249 Label unlock; 1250 // BasicObjectLock will be first in list, since this is a 1251 // synchronized method. However, need to check that the object 1252 // has not been unlocked by an explicit monitorexit bytecode. 1253 const Address monitor(rbp, 1254 (intptr_t)(frame::interpreter_frame_initial_sp_offset * 1255 wordSize - sizeof(BasicObjectLock))); 1256 1257 // monitor expect in c_rarg1 for slow unlock path 1258 __ lea(c_rarg1, monitor); // address of first monitor 1259 1260 __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 1261 __ testptr(t, t); 1262 __ jcc(Assembler::notZero, unlock); 1263 1264 // Entry already unlocked, need to throw exception 1265 __ MacroAssembler::call_VM(noreg, 1266 CAST_FROM_FN_PTR(address, 1267 InterpreterRuntime::throw_illegal_monitor_state_exception)); 1268 __ should_not_reach_here(); 1269 1270 __ bind(unlock); 1271 __ unlock_object(c_rarg1); 1272 } 1273 __ bind(L); 1274 } 1275 1276 // jvmti support 1277 // Note: This must happen _after_ handling/throwing any exceptions since 1278 // the exception handler code notifies the runtime of method exits 1279 // too. If this happens before, method entry/exit notifications are 1280 // not properly paired (was bug - gri 11/22/99). 1281 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); 1282 1283 // restore potential result in edx:eax, call result handler to 1284 // restore potential result in ST0 & handle result 1285 1286 __ pop(ltos); 1287 __ pop(dtos); 1288 1289 __ movptr(t, Address(rbp, 1290 (frame::interpreter_frame_result_handler_offset) * wordSize)); 1291 __ call(t); 1292 1293 // remove activation 1294 __ movptr(t, Address(rbp, 1295 frame::interpreter_frame_sender_sp_offset * 1296 wordSize)); // get sender sp 1297 __ leave(); // remove frame anchor 1298 __ pop(rdi); // get return address 1299 __ mov(rsp, t); // set sp to sender sp 1300 __ jmp(rdi); 1301 1302 if (inc_counter) { 1303 // Handle overflow of counter and compile method 1304 __ bind(invocation_counter_overflow); 1305 generate_counter_overflow(&continue_after_compile); 1306 } 1307 1308 return entry_point; 1309 } 1310 1311 // 1312 // Generic interpreted method entry to (asm) interpreter 1313 // 1314 address InterpreterGenerator::generate_normal_entry(bool synchronized) { 1315 // determine code generation flags 1316 bool inc_counter = UseCompiler || CountCompiledCalls; 1317 1318 // ebx: methodOop 1319 // r13: sender sp 1320 address entry_point = __ pc(); 1321 1322 const Address size_of_parameters(rbx, 1323 methodOopDesc::size_of_parameters_offset()); 1324 const Address size_of_locals(rbx, methodOopDesc::size_of_locals_offset()); 1325 const Address invocation_counter(rbx, 1326 methodOopDesc::invocation_counter_offset() + 1327 InvocationCounter::counter_offset()); 1328 const Address access_flags(rbx, methodOopDesc::access_flags_offset()); 1329 1330 // get parameter size (always needed) 1331 __ load_unsigned_short(rcx, size_of_parameters); 1332 1333 // rbx: methodOop 1334 // rcx: size of parameters 1335 // r13: sender_sp (could differ from sp+wordSize if we were called via c2i ) 1336 1337 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words 1338 __ subl(rdx, rcx); // rdx = no. of additional locals 1339 1340 // YYY 1341 // __ incrementl(rdx); 1342 // __ andl(rdx, -2); 1343 1344 // see if we've got enough room on the stack for locals plus overhead. 1345 generate_stack_overflow_check(); 1346 1347 // get return address 1348 __ pop(rax); 1349 1350 // compute beginning of parameters (r14) 1351 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize)); 1352 1353 // rdx - # of additional locals 1354 // allocate space for locals 1355 // explicitly initialize locals 1356 { 1357 Label exit, loop; 1358 __ testl(rdx, rdx); 1359 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0 1360 __ bind(loop); 1361 __ push((int) NULL_WORD); // initialize local variables 1362 __ decrementl(rdx); // until everything initialized 1363 __ jcc(Assembler::greater, loop); 1364 __ bind(exit); 1365 } 1366 1367 // (pre-)fetch invocation count 1368 if (inc_counter) { 1369 __ movl(rcx, invocation_counter); 1370 } 1371 // initialize fixed part of activation frame 1372 generate_fixed_frame(false); 1373 1374 // make sure method is not native & not abstract 1375 #ifdef ASSERT 1376 __ movl(rax, access_flags); 1377 { 1378 Label L; 1379 __ testl(rax, JVM_ACC_NATIVE); 1380 __ jcc(Assembler::zero, L); 1381 __ stop("tried to execute native method as non-native"); 1382 __ bind(L); 1383 } 1384 { 1385 Label L; 1386 __ testl(rax, JVM_ACC_ABSTRACT); 1387 __ jcc(Assembler::zero, L); 1388 __ stop("tried to execute abstract method in interpreter"); 1389 __ bind(L); 1390 } 1391 #endif 1392 1393 // Since at this point in the method invocation the exception 1394 // handler would try to exit the monitor of synchronized methods 1395 // which hasn't been entered yet, we set the thread local variable 1396 // _do_not_unlock_if_synchronized to true. The remove_activation 1397 // will check this flag. 1398 1399 const Address do_not_unlock_if_synchronized(r15_thread, 1400 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1401 __ movbool(do_not_unlock_if_synchronized, true); 1402 1403 // increment invocation count & check for overflow 1404 Label invocation_counter_overflow; 1405 Label profile_method; 1406 Label profile_method_continue; 1407 if (inc_counter) { 1408 generate_counter_incr(&invocation_counter_overflow, 1409 &profile_method, 1410 &profile_method_continue); 1411 if (ProfileInterpreter) { 1412 __ bind(profile_method_continue); 1413 } 1414 } 1415 1416 Label continue_after_compile; 1417 __ bind(continue_after_compile); 1418 1419 // check for synchronized interpreted methods 1420 bang_stack_shadow_pages(false); 1421 1422 // reset the _do_not_unlock_if_synchronized flag 1423 __ movbool(do_not_unlock_if_synchronized, false); 1424 1425 // check for synchronized methods 1426 // Must happen AFTER invocation_counter check and stack overflow check, 1427 // so method is not locked if overflows. 1428 if (synchronized) { 1429 // Allocate monitor and lock method 1430 lock_method(); 1431 } else { 1432 // no synchronization necessary 1433 #ifdef ASSERT 1434 { 1435 Label L; 1436 __ movl(rax, access_flags); 1437 __ testl(rax, JVM_ACC_SYNCHRONIZED); 1438 __ jcc(Assembler::zero, L); 1439 __ stop("method needs synchronization"); 1440 __ bind(L); 1441 } 1442 #endif 1443 } 1444 1445 // start execution 1446 #ifdef ASSERT 1447 { 1448 Label L; 1449 const Address monitor_block_top (rbp, 1450 frame::interpreter_frame_monitor_block_top_offset * wordSize); 1451 __ movptr(rax, monitor_block_top); 1452 __ cmpptr(rax, rsp); 1453 __ jcc(Assembler::equal, L); 1454 __ stop("broken stack frame setup in interpreter"); 1455 __ bind(L); 1456 } 1457 #endif 1458 1459 // jvmti support 1460 __ notify_method_entry(); 1461 1462 __ dispatch_next(vtos); 1463 1464 // invocation counter overflow 1465 if (inc_counter) { 1466 if (ProfileInterpreter) { 1467 // We have decided to profile this method in the interpreter 1468 __ bind(profile_method); 1469 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1470 __ set_method_data_pointer_for_bcp(); 1471 __ get_method(rbx); 1472 __ jmp(profile_method_continue); 1473 } 1474 // Handle overflow of counter and compile method 1475 __ bind(invocation_counter_overflow); 1476 generate_counter_overflow(&continue_after_compile); 1477 } 1478 1479 return entry_point; 1480 } 1481 1482 // Entry points 1483 // 1484 // Here we generate the various kind of entries into the interpreter. 1485 // The two main entry type are generic bytecode methods and native 1486 // call method. These both come in synchronized and non-synchronized 1487 // versions but the frame layout they create is very similar. The 1488 // other method entry types are really just special purpose entries 1489 // that are really entry and interpretation all in one. These are for 1490 // trivial methods like accessor, empty, or special math methods. 1491 // 1492 // When control flow reaches any of the entry types for the interpreter 1493 // the following holds -> 1494 // 1495 // Arguments: 1496 // 1497 // rbx: methodOop 1498 // 1499 // Stack layout immediately at entry 1500 // 1501 // [ return address ] <--- rsp 1502 // [ parameter n ] 1503 // ... 1504 // [ parameter 1 ] 1505 // [ expression stack ] (caller's java expression stack) 1506 1507 // Assuming that we don't go to one of the trivial specialized entries 1508 // the stack will look like below when we are ready to execute the 1509 // first bytecode (or call the native routine). The register usage 1510 // will be as the template based interpreter expects (see 1511 // interpreter_amd64.hpp). 1512 // 1513 // local variables follow incoming parameters immediately; i.e. 1514 // the return address is moved to the end of the locals). 1515 // 1516 // [ monitor entry ] <--- rsp 1517 // ... 1518 // [ monitor entry ] 1519 // [ expr. stack bottom ] 1520 // [ saved r13 ] 1521 // [ current r14 ] 1522 // [ methodOop ] 1523 // [ saved ebp ] <--- rbp 1524 // [ return address ] 1525 // [ local variable m ] 1526 // ... 1527 // [ local variable 1 ] 1528 // [ parameter n ] 1529 // ... 1530 // [ parameter 1 ] <--- r14 1531 1532 address AbstractInterpreterGenerator::generate_method_entry( 1533 AbstractInterpreter::MethodKind kind) { 1534 // determine code generation flags 1535 bool synchronized = false; 1536 address entry_point = NULL; 1537 1538 switch (kind) { 1539 case Interpreter::zerolocals : break; 1540 case Interpreter::zerolocals_synchronized: synchronized = true; break; 1541 case Interpreter::native : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break; 1542 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true); break; 1543 case Interpreter::empty : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry(); break; 1544 case Interpreter::accessor : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry(); break; 1545 case Interpreter::abstract : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry(); break; 1546 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*) this)->generate_method_handle_entry();break; 1547 1548 case Interpreter::java_lang_math_sin : // fall thru 1549 case Interpreter::java_lang_math_cos : // fall thru 1550 case Interpreter::java_lang_math_tan : // fall thru 1551 case Interpreter::java_lang_math_abs : // fall thru 1552 case Interpreter::java_lang_math_log : // fall thru 1553 case Interpreter::java_lang_math_log10 : // fall thru 1554 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind); break; 1555 case Interpreter::java_lang_ref_reference_get 1556 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break; 1557 default : ShouldNotReachHere(); break; 1558 } 1559 1560 if (entry_point) { 1561 return entry_point; 1562 } 1563 1564 return ((InterpreterGenerator*) this)-> 1565 generate_normal_entry(synchronized); 1566 } 1567 1568 // These should never be compiled since the interpreter will prefer 1569 // the compiled version to the intrinsic version. 1570 bool AbstractInterpreter::can_be_compiled(methodHandle m) { 1571 switch (method_kind(m)) { 1572 case Interpreter::java_lang_math_sin : // fall thru 1573 case Interpreter::java_lang_math_cos : // fall thru 1574 case Interpreter::java_lang_math_tan : // fall thru 1575 case Interpreter::java_lang_math_abs : // fall thru 1576 case Interpreter::java_lang_math_log : // fall thru 1577 case Interpreter::java_lang_math_log10 : // fall thru 1578 case Interpreter::java_lang_math_sqrt : 1579 return false; 1580 default: 1581 return true; 1582 } 1583 } 1584 1585 // How much stack a method activation needs in words. 1586 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) { 1587 const int entry_size = frame::interpreter_frame_monitor_size(); 1588 1589 // total overhead size: entry_size + (saved rbp thru expr stack 1590 // bottom). be sure to change this if you add/subtract anything 1591 // to/from the overhead area 1592 const int overhead_size = 1593 -(frame::interpreter_frame_initial_sp_offset) + entry_size; 1594 1595 const int stub_code = frame::entry_frame_after_call_words; 1596 const int extra_stack = methodOopDesc::extra_stack_entries(); 1597 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) * 1598 Interpreter::stackElementWords; 1599 return (overhead_size + method_stack + stub_code); 1600 } 1601 1602 int AbstractInterpreter::layout_activation(methodOop method, 1603 int tempcount, 1604 int popframe_extra_args, 1605 int moncount, 1606 int caller_actual_parameters, 1607 int callee_param_count, 1608 int callee_locals, 1609 frame* caller, 1610 frame* interpreter_frame, 1611 bool is_top_frame) { 1612 // Note: This calculation must exactly parallel the frame setup 1613 // in AbstractInterpreterGenerator::generate_method_entry. 1614 // If interpreter_frame!=NULL, set up the method, locals, and monitors. 1615 // The frame interpreter_frame, if not NULL, is guaranteed to be the 1616 // right size, as determined by a previous call to this method. 1617 // It is also guaranteed to be walkable even though it is in a skeletal state 1618 1619 // fixed size of an interpreter frame: 1620 int max_locals = method->max_locals() * Interpreter::stackElementWords; 1621 int extra_locals = (method->max_locals() - method->size_of_parameters()) * 1622 Interpreter::stackElementWords; 1623 1624 int overhead = frame::sender_sp_offset - 1625 frame::interpreter_frame_initial_sp_offset; 1626 // Our locals were accounted for by the caller (or last_frame_adjust 1627 // on the transistion) Since the callee parameters already account 1628 // for the callee's params we only need to account for the extra 1629 // locals. 1630 int size = overhead + 1631 (callee_locals - callee_param_count)*Interpreter::stackElementWords + 1632 moncount * frame::interpreter_frame_monitor_size() + 1633 tempcount* Interpreter::stackElementWords + popframe_extra_args; 1634 if (interpreter_frame != NULL) { 1635 #ifdef ASSERT 1636 if (!EnableInvokeDynamic) 1637 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences? 1638 // Probably, since deoptimization doesn't work yet. 1639 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable"); 1640 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)"); 1641 #endif 1642 1643 interpreter_frame->interpreter_frame_set_method(method); 1644 // NOTE the difference in using sender_sp and 1645 // interpreter_frame_sender_sp interpreter_frame_sender_sp is 1646 // the original sp of the caller (the unextended_sp) and 1647 // sender_sp is fp+16 XXX 1648 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1; 1649 1650 interpreter_frame->interpreter_frame_set_locals(locals); 1651 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin(); 1652 BasicObjectLock* monbot = montop - moncount; 1653 interpreter_frame->interpreter_frame_set_monitor_end(monbot); 1654 1655 // Set last_sp 1656 intptr_t* esp = (intptr_t*) monbot - 1657 tempcount*Interpreter::stackElementWords - 1658 popframe_extra_args; 1659 interpreter_frame->interpreter_frame_set_last_sp(esp); 1660 1661 // All frames but the initial (oldest) interpreter frame we fill in have 1662 // a value for sender_sp that allows walking the stack but isn't 1663 // truly correct. Correct the value here. 1664 if (extra_locals != 0 && 1665 interpreter_frame->sender_sp() == 1666 interpreter_frame->interpreter_frame_sender_sp()) { 1667 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + 1668 extra_locals); 1669 } 1670 *interpreter_frame->interpreter_frame_cache_addr() = 1671 method->constants()->cache(); 1672 } 1673 return size; 1674 } 1675 1676 //----------------------------------------------------------------------------- 1677 // Exceptions 1678 1679 void TemplateInterpreterGenerator::generate_throw_exception() { 1680 // Entry point in previous activation (i.e., if the caller was 1681 // interpreted) 1682 Interpreter::_rethrow_exception_entry = __ pc(); 1683 // Restore sp to interpreter_frame_last_sp even though we are going 1684 // to empty the expression stack for the exception processing. 1685 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 1686 // rax: exception 1687 // rdx: return address/pc that threw exception 1688 __ restore_bcp(); // r13 points to call/send 1689 __ restore_locals(); 1690 __ reinit_heapbase(); // restore r12 as heapbase. 1691 // Entry point for exceptions thrown within interpreter code 1692 Interpreter::_throw_exception_entry = __ pc(); 1693 // expression stack is undefined here 1694 // rax: exception 1695 // r13: exception bcp 1696 __ verify_oop(rax); 1697 __ mov(c_rarg1, rax); 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(rdx, 1704 CAST_FROM_FN_PTR(address, 1705 InterpreterRuntime::exception_handler_for_exception), 1706 c_rarg1); 1707 // rax: exception handler entry point 1708 // rdx: preserved exception oop 1709 // r13: bcp for exception handler 1710 __ push_ptr(rdx); // push exception which is now the only value on the stack 1711 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!) 1712 1713 // If the exception is not handled in the current frame the frame is 1714 // removed and the exception is rethrown (i.e. exception 1715 // continuation is _rethrow_exception). 1716 // 1717 // Note: At this point the bci is still the bxi for the instruction 1718 // which caused the exception and the expression stack is 1719 // empty. Thus, for any VM calls at this point, GC will find a legal 1720 // oop map (with empty expression stack). 1721 1722 // In current activation 1723 // tos: exception 1724 // esi: exception bcp 1725 1726 // 1727 // JVMTI PopFrame support 1728 // 1729 1730 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 1731 __ empty_expression_stack(); 1732 // Set the popframe_processing bit in pending_popframe_condition 1733 // indicating that we are currently handling popframe, so that 1734 // call_VMs that may happen later do not trigger new popframe 1735 // handling cycles. 1736 __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset())); 1737 __ orl(rdx, JavaThread::popframe_processing_bit); 1738 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx); 1739 1740 { 1741 // Check to see whether we are returning to a deoptimized frame. 1742 // (The PopFrame call ensures that the caller of the popped frame is 1743 // either interpreted or compiled and deoptimizes it if compiled.) 1744 // In this case, we can't call dispatch_next() after the frame is 1745 // popped, but instead must save the incoming arguments and restore 1746 // them after deoptimization has occurred. 1747 // 1748 // Note that we don't compare the return PC against the 1749 // deoptimization blob's unpack entry because of the presence of 1750 // adapter frames in C2. 1751 Label caller_not_deoptimized; 1752 __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize)); 1753 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1754 InterpreterRuntime::interpreter_contains), c_rarg1); 1755 __ testl(rax, rax); 1756 __ jcc(Assembler::notZero, caller_not_deoptimized); 1757 1758 // Compute size of arguments for saving when returning to 1759 // deoptimized caller 1760 __ get_method(rax); 1761 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc:: 1762 size_of_parameters_offset()))); 1763 __ shll(rax, Interpreter::logStackElementSize); 1764 __ restore_locals(); // XXX do we need this? 1765 __ subptr(r14, rax); 1766 __ addptr(r14, wordSize); 1767 // Save these arguments 1768 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1769 Deoptimization:: 1770 popframe_preserve_args), 1771 r15_thread, rax, r14); 1772 1773 __ remove_activation(vtos, rdx, 1774 /* throw_monitor_exception */ false, 1775 /* install_monitor_exception */ false, 1776 /* notify_jvmdi */ false); 1777 1778 // Inform deoptimization that it is responsible for restoring 1779 // these arguments 1780 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), 1781 JavaThread::popframe_force_deopt_reexecution_bit); 1782 1783 // Continue in deoptimization handler 1784 __ jmp(rdx); 1785 1786 __ bind(caller_not_deoptimized); 1787 } 1788 1789 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */ 1790 /* throw_monitor_exception */ false, 1791 /* install_monitor_exception */ false, 1792 /* notify_jvmdi */ false); 1793 1794 // Finish with popframe handling 1795 // A previous I2C followed by a deoptimization might have moved the 1796 // outgoing arguments further up the stack. PopFrame expects the 1797 // mutations to those outgoing arguments to be preserved and other 1798 // constraints basically require this frame to look exactly as 1799 // though it had previously invoked an interpreted activation with 1800 // no space between the top of the expression stack (current 1801 // last_sp) and the top of stack. Rather than force deopt to 1802 // maintain this kind of invariant all the time we call a small 1803 // fixup routine to move the mutated arguments onto the top of our 1804 // expression stack if necessary. 1805 __ mov(c_rarg1, rsp); 1806 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 1807 // PC must point into interpreter here 1808 __ set_last_Java_frame(noreg, rbp, __ pc()); 1809 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2); 1810 __ reset_last_Java_frame(true, true); 1811 // Restore the last_sp and null it out 1812 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 1813 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 1814 1815 __ restore_bcp(); // XXX do we need this? 1816 __ restore_locals(); // XXX do we need this? 1817 // The method data pointer was incremented already during 1818 // call profiling. We have to restore the mdp for the current bcp. 1819 if (ProfileInterpreter) { 1820 __ set_method_data_pointer_for_bcp(); 1821 } 1822 1823 // Clear the popframe condition flag 1824 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), 1825 JavaThread::popframe_inactive); 1826 1827 __ dispatch_next(vtos); 1828 // end of PopFrame support 1829 1830 Interpreter::_remove_activation_entry = __ pc(); 1831 1832 // preserve exception over this code sequence 1833 __ pop_ptr(rax); 1834 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax); 1835 // remove the activation (without doing throws on illegalMonitorExceptions) 1836 __ remove_activation(vtos, rdx, false, true, false); 1837 // restore exception 1838 __ movptr(rax, Address(r15_thread, JavaThread::vm_result_offset())); 1839 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD); 1840 __ verify_oop(rax); 1841 1842 // In between activations - previous activation type unknown yet 1843 // compute continuation point - the continuation point expects the 1844 // following registers set up: 1845 // 1846 // rax: exception 1847 // rdx: return address/pc that threw exception 1848 // rsp: expression stack of caller 1849 // rbp: ebp of caller 1850 __ push(rax); // save exception 1851 __ push(rdx); // save return address 1852 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1853 SharedRuntime::exception_handler_for_return_address), 1854 r15_thread, rdx); 1855 __ mov(rbx, rax); // save exception handler 1856 __ pop(rdx); // restore return address 1857 __ pop(rax); // restore exception 1858 // Note that an "issuing PC" is actually the next PC after the call 1859 __ jmp(rbx); // jump to exception 1860 // handler of caller 1861 } 1862 1863 1864 // 1865 // JVMTI ForceEarlyReturn support 1866 // 1867 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 1868 address entry = __ pc(); 1869 1870 __ restore_bcp(); 1871 __ restore_locals(); 1872 __ empty_expression_stack(); 1873 __ load_earlyret_value(state); 1874 1875 __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset())); 1876 Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset()); 1877 1878 // Clear the earlyret state 1879 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive); 1880 1881 __ remove_activation(state, rsi, 1882 false, /* throw_monitor_exception */ 1883 false, /* install_monitor_exception */ 1884 true); /* notify_jvmdi */ 1885 __ jmp(rsi); 1886 1887 return entry; 1888 } // end of ForceEarlyReturn support 1889 1890 1891 //----------------------------------------------------------------------------- 1892 // Helper for vtos entry point generation 1893 1894 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, 1895 address& bep, 1896 address& cep, 1897 address& sep, 1898 address& aep, 1899 address& iep, 1900 address& lep, 1901 address& fep, 1902 address& dep, 1903 address& vep) { 1904 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 1905 Label L; 1906 aep = __ pc(); __ push_ptr(); __ jmp(L); 1907 fep = __ pc(); __ push_f(); __ jmp(L); 1908 dep = __ pc(); __ push_d(); __ jmp(L); 1909 lep = __ pc(); __ push_l(); __ jmp(L); 1910 bep = cep = sep = 1911 iep = __ pc(); __ push_i(); 1912 vep = __ pc(); 1913 __ bind(L); 1914 generate_and_dispatch(t); 1915 } 1916 1917 1918 //----------------------------------------------------------------------------- 1919 // Generation of individual instructions 1920 1921 // helpers for generate_and_dispatch 1922 1923 1924 InterpreterGenerator::InterpreterGenerator(StubQueue* code) 1925 : TemplateInterpreterGenerator(code) { 1926 generate_all(); // down here so it can be "virtual" 1927 } 1928 1929 //----------------------------------------------------------------------------- 1930 1931 // Non-product code 1932 #ifndef PRODUCT 1933 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 1934 address entry = __ pc(); 1935 1936 __ push(state); 1937 __ push(c_rarg0); 1938 __ push(c_rarg1); 1939 __ push(c_rarg2); 1940 __ push(c_rarg3); 1941 __ mov(c_rarg2, rax); // Pass itos 1942 #ifdef _WIN64 1943 __ movflt(xmm3, xmm0); // Pass ftos 1944 #endif 1945 __ call_VM(noreg, 1946 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), 1947 c_rarg1, c_rarg2, c_rarg3); 1948 __ pop(c_rarg3); 1949 __ pop(c_rarg2); 1950 __ pop(c_rarg1); 1951 __ pop(c_rarg0); 1952 __ pop(state); 1953 __ ret(0); // return from result handler 1954 1955 return entry; 1956 } 1957 1958 void TemplateInterpreterGenerator::count_bytecode() { 1959 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value)); 1960 } 1961 1962 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 1963 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()])); 1964 } 1965 1966 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 1967 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index)); 1968 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes); 1969 __ orl(rbx, 1970 ((int) t->bytecode()) << 1971 BytecodePairHistogram::log2_number_of_codes); 1972 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx); 1973 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters)); 1974 __ incrementl(Address(rscratch1, rbx, Address::times_4)); 1975 } 1976 1977 1978 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 1979 // Call a little run-time stub to avoid blow-up for each bytecode. 1980 // The run-time runtime saves the right registers, depending on 1981 // the tosca in-state for the given template. 1982 1983 assert(Interpreter::trace_code(t->tos_in()) != NULL, 1984 "entry must have been generated"); 1985 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) 1986 __ andptr(rsp, -16); // align stack as required by ABI 1987 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in()))); 1988 __ mov(rsp, r12); // restore sp 1989 __ reinit_heapbase(); 1990 } 1991 1992 1993 void TemplateInterpreterGenerator::stop_interpreter_at() { 1994 Label L; 1995 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value), 1996 StopInterpreterAt); 1997 __ jcc(Assembler::notEqual, L); 1998 __ int3(); 1999 __ bind(L); 2000 } 2001 #endif // !PRODUCT 2002 #endif // ! CC_INTERP