1 /* 2 * Copyright (c) 2003, 2013, 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/templateTable.hpp" 32 #include "oops/arrayOop.hpp" 33 #include "oops/methodData.hpp" 34 #include "oops/method.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 #include "utilities/macros.hpp" 48 49 #define __ _masm-> 50 51 #ifndef CC_INTERP 52 53 const int method_offset = frame::interpreter_frame_method_offset * wordSize; 54 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize; 55 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize; 56 57 //----------------------------------------------------------------------------- 58 59 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 60 address entry = __ pc(); 61 62 #ifdef ASSERT 63 { 64 Label L; 65 __ lea(rax, Address(rbp, 66 frame::interpreter_frame_monitor_block_top_offset * 67 wordSize)); 68 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack 69 // grows negative) 70 __ jcc(Assembler::aboveEqual, L); // check if frame is complete 71 __ stop ("interpreter frame not set up"); 72 __ bind(L); 73 } 74 #endif // ASSERT 75 // Restore bcp under the assumption that the current frame is still 76 // interpreted 77 __ restore_bcp(); 78 79 // expression stack must be empty before entering the VM if an 80 // exception happened 81 __ empty_expression_stack(); 82 // throw exception 83 __ call_VM(noreg, 84 CAST_FROM_FN_PTR(address, 85 InterpreterRuntime::throw_StackOverflowError)); 86 return entry; 87 } 88 89 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler( 90 const char* name) { 91 address entry = __ pc(); 92 // expression stack must be empty before entering the VM if an 93 // exception happened 94 __ empty_expression_stack(); 95 // setup parameters 96 // ??? convention: expect aberrant index in register ebx 97 __ lea(c_rarg1, ExternalAddress((address)name)); 98 __ call_VM(noreg, 99 CAST_FROM_FN_PTR(address, 100 InterpreterRuntime:: 101 throw_ArrayIndexOutOfBoundsException), 102 c_rarg1, rbx); 103 return entry; 104 } 105 106 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 107 address entry = __ pc(); 108 109 // object is at TOS 110 __ pop(c_rarg1); 111 112 // expression stack must be empty before entering the VM if an 113 // exception happened 114 __ empty_expression_stack(); 115 116 __ call_VM(noreg, 117 CAST_FROM_FN_PTR(address, 118 InterpreterRuntime:: 119 throw_ClassCastException), 120 c_rarg1); 121 return entry; 122 } 123 124 address TemplateInterpreterGenerator::generate_exception_handler_common( 125 const char* name, const char* message, bool pass_oop) { 126 assert(!pass_oop || message == NULL, "either oop or message but not both"); 127 address entry = __ pc(); 128 if (pass_oop) { 129 // object is at TOS 130 __ pop(c_rarg2); 131 } 132 // expression stack must be empty before entering the VM if an 133 // exception happened 134 __ empty_expression_stack(); 135 // setup parameters 136 __ lea(c_rarg1, ExternalAddress((address)name)); 137 if (pass_oop) { 138 __ call_VM(rax, CAST_FROM_FN_PTR(address, 139 InterpreterRuntime:: 140 create_klass_exception), 141 c_rarg1, c_rarg2); 142 } else { 143 // kind of lame ExternalAddress can't take NULL because 144 // external_word_Relocation will assert. 145 if (message != NULL) { 146 __ lea(c_rarg2, ExternalAddress((address)message)); 147 } else { 148 __ movptr(c_rarg2, NULL_WORD); 149 } 150 __ call_VM(rax, 151 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), 152 c_rarg1, c_rarg2); 153 } 154 // throw exception 155 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); 156 return entry; 157 } 158 159 160 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { 161 address entry = __ pc(); 162 // NULL last_sp until next java call 163 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 164 __ dispatch_next(state); 165 return entry; 166 } 167 168 169 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) { 170 address entry = __ pc(); 171 172 // Restore stack bottom in case i2c adjusted stack 173 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 174 // and NULL it as marker that esp is now tos until next java call 175 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 176 177 __ restore_bcp(); 178 __ restore_locals(); 179 180 Label L_got_cache, L_giant_index; 181 if (EnableInvokeDynamic) { 182 __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic); 183 __ jcc(Assembler::equal, L_giant_index); 184 } 185 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2)); 186 __ bind(L_got_cache); 187 __ movl(rbx, Address(rbx, rcx, 188 Address::times_ptr, 189 in_bytes(ConstantPoolCache::base_offset()) + 190 3 * wordSize)); 191 __ andl(rbx, 0xFF); 192 __ lea(rsp, Address(rsp, rbx, Address::times_8)); 193 __ dispatch_next(state, step); 194 195 // out of the main line of code... 196 if (EnableInvokeDynamic) { 197 __ bind(L_giant_index); 198 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4)); 199 __ jmp(L_got_cache); 200 } 201 202 return entry; 203 } 204 205 206 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, 207 int step) { 208 address entry = __ pc(); 209 // NULL last_sp until next java call 210 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 211 __ restore_bcp(); 212 __ restore_locals(); 213 // handle exceptions 214 { 215 Label L; 216 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); 217 __ jcc(Assembler::zero, L); 218 __ call_VM(noreg, 219 CAST_FROM_FN_PTR(address, 220 InterpreterRuntime::throw_pending_exception)); 221 __ should_not_reach_here(); 222 __ bind(L); 223 } 224 __ dispatch_next(state, step); 225 return entry; 226 } 227 228 int AbstractInterpreter::BasicType_as_index(BasicType type) { 229 int i = 0; 230 switch (type) { 231 case T_BOOLEAN: i = 0; break; 232 case T_CHAR : i = 1; break; 233 case T_BYTE : i = 2; break; 234 case T_SHORT : i = 3; break; 235 case T_INT : i = 4; break; 236 case T_LONG : i = 5; break; 237 case T_VOID : i = 6; break; 238 case T_FLOAT : i = 7; break; 239 case T_DOUBLE : i = 8; break; 240 case T_OBJECT : i = 9; break; 241 case T_ARRAY : i = 9; break; 242 default : ShouldNotReachHere(); 243 } 244 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, 245 "index out of bounds"); 246 return i; 247 } 248 249 250 address TemplateInterpreterGenerator::generate_result_handler_for( 251 BasicType type) { 252 address entry = __ pc(); 253 switch (type) { 254 case T_BOOLEAN: __ c2bool(rax); break; 255 case T_CHAR : __ movzwl(rax, rax); break; 256 case T_BYTE : __ sign_extend_byte(rax); break; 257 case T_SHORT : __ sign_extend_short(rax); break; 258 case T_INT : /* nothing to do */ break; 259 case T_LONG : /* nothing to do */ break; 260 case T_VOID : /* nothing to do */ break; 261 case T_FLOAT : /* nothing to do */ break; 262 case T_DOUBLE : /* nothing to do */ break; 263 case T_OBJECT : 264 // retrieve result from frame 265 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize)); 266 // and verify it 267 __ verify_oop(rax); 268 break; 269 default : ShouldNotReachHere(); 270 } 271 __ ret(0); // return from result handler 272 return entry; 273 } 274 275 address TemplateInterpreterGenerator::generate_safept_entry_for( 276 TosState state, 277 address runtime_entry) { 278 address entry = __ pc(); 279 __ push(state); 280 __ call_VM(noreg, runtime_entry); 281 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 282 return entry; 283 } 284 285 286 287 // Helpers for commoning out cases in the various type of method entries. 288 // 289 290 291 // increment invocation count & check for overflow 292 // 293 // Note: checking for negative value instead of overflow 294 // so we have a 'sticky' overflow test 295 // 296 // rbx: method 297 // ecx: invocation counter 298 // 299 void InterpreterGenerator::generate_counter_incr( 300 Label* overflow, 301 Label* profile_method, 302 Label* profile_method_continue) { 303 Label done; 304 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not. 305 if (TieredCompilation) { 306 int increment = InvocationCounter::count_increment; 307 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift; 308 Label no_mdo; 309 if (ProfileInterpreter) { 310 // Are we profiling? 311 __ movptr(rax, Address(rbx, Method::method_data_offset())); 312 __ testptr(rax, rax); 313 __ jccb(Assembler::zero, no_mdo); 314 // Increment counter in the MDO 315 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) + 316 in_bytes(InvocationCounter::counter_offset())); 317 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow); 318 __ jmp(done); 319 } 320 __ bind(no_mdo); 321 // Increment counter in MethodCounters 322 const Address invocation_counter(rax, 323 MethodCounters::invocation_counter_offset() + 324 InvocationCounter::counter_offset()); 325 __ get_method_counters(rbx, rax, done); 326 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, 327 false, Assembler::zero, overflow); 328 __ bind(done); 329 } else { 330 const Address backedge_counter(rax, 331 MethodCounters::backedge_counter_offset() + 332 InvocationCounter::counter_offset()); 333 const Address invocation_counter(rax, 334 MethodCounters::invocation_counter_offset() + 335 InvocationCounter::counter_offset()); 336 337 __ get_method_counters(rbx, rax, done); 338 339 if (ProfileInterpreter) { 340 __ incrementl(Address(rax, 341 MethodCounters::interpreter_invocation_counter_offset())); 342 } 343 // Update standard invocation counters 344 __ movl(rcx, invocation_counter); 345 __ incrementl(rcx, InvocationCounter::count_increment); 346 __ movl(invocation_counter, rcx); // save invocation count 347 348 __ movl(rax, backedge_counter); // load backedge counter 349 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits 350 351 __ addl(rcx, rax); // add both counters 352 353 // profile_method is non-null only for interpreted method so 354 // profile_method != NULL == !native_call 355 356 if (ProfileInterpreter && profile_method != NULL) { 357 // Test to see if we should create a method data oop 358 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit)); 359 __ jcc(Assembler::less, *profile_method_continue); 360 361 // if no method data exists, go to profile_method 362 __ test_method_data_pointer(rax, *profile_method); 363 } 364 365 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit)); 366 __ jcc(Assembler::aboveEqual, *overflow); 367 __ bind(done); 368 } 369 } 370 371 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) { 372 373 // Asm interpreter on entry 374 // r14 - locals 375 // r13 - bcp 376 // rbx - method 377 // edx - cpool --- DOES NOT APPEAR TO BE TRUE 378 // rbp - interpreter frame 379 380 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] 381 // Everything as it was on entry 382 // rdx is not restored. Doesn't appear to really be set. 383 384 // InterpreterRuntime::frequency_counter_overflow takes two 385 // arguments, the first (thread) is passed by call_VM, the second 386 // indicates if the counter overflow occurs at a backwards branch 387 // (NULL bcp). We pass zero for it. The call returns the address 388 // of the verified entry point for the method or NULL if the 389 // compilation did not complete (either went background or bailed 390 // out). 391 __ movl(c_rarg1, 0); 392 __ call_VM(noreg, 393 CAST_FROM_FN_PTR(address, 394 InterpreterRuntime::frequency_counter_overflow), 395 c_rarg1); 396 397 __ movptr(rbx, Address(rbp, method_offset)); // restore Method* 398 // Preserve invariant that r13/r14 contain bcp/locals of sender frame 399 // and jump to the interpreted entry. 400 __ jmp(*do_continue, relocInfo::none); 401 } 402 403 // See if we've got enough room on the stack for locals plus overhead. 404 // The expression stack grows down incrementally, so the normal guard 405 // page mechanism will work for that. 406 // 407 // NOTE: Since the additional locals are also always pushed (wasn't 408 // obvious in generate_method_entry) so the guard should work for them 409 // too. 410 // 411 // Args: 412 // rdx: number of additional locals this frame needs (what we must check) 413 // rbx: Method* 414 // 415 // Kills: 416 // rax 417 void InterpreterGenerator::generate_stack_overflow_check(void) { 418 419 // monitor entry size: see picture of stack set 420 // (generate_method_entry) and frame_amd64.hpp 421 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 422 423 // total overhead size: entry_size + (saved rbp through expr stack 424 // bottom). be sure to change this if you add/subtract anything 425 // to/from the overhead area 426 const int overhead_size = 427 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size; 428 429 const int page_size = os::vm_page_size(); 430 431 Label after_frame_check; 432 433 // see if the frame is greater than one page in size. If so, 434 // then we need to verify there is enough stack space remaining 435 // for the additional locals. 436 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize); 437 __ jcc(Assembler::belowEqual, after_frame_check); 438 439 // compute rsp as if this were going to be the last frame on 440 // the stack before the red zone 441 442 const Address stack_base(r15_thread, Thread::stack_base_offset()); 443 const Address stack_size(r15_thread, Thread::stack_size_offset()); 444 445 // locals + overhead, in bytes 446 __ mov(rax, rdx); 447 __ shlptr(rax, Interpreter::logStackElementSize); // 2 slots per parameter. 448 __ addptr(rax, overhead_size); 449 450 #ifdef ASSERT 451 Label stack_base_okay, stack_size_okay; 452 // verify that thread stack base is non-zero 453 __ cmpptr(stack_base, (int32_t)NULL_WORD); 454 __ jcc(Assembler::notEqual, stack_base_okay); 455 __ stop("stack base is zero"); 456 __ bind(stack_base_okay); 457 // verify that thread stack size is non-zero 458 __ cmpptr(stack_size, 0); 459 __ jcc(Assembler::notEqual, stack_size_okay); 460 __ stop("stack size is zero"); 461 __ bind(stack_size_okay); 462 #endif 463 464 // Add stack base to locals and subtract stack size 465 __ addptr(rax, stack_base); 466 __ subptr(rax, stack_size); 467 468 // Use the maximum number of pages we might bang. 469 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages : 470 (StackRedPages+StackYellowPages); 471 472 // add in the red and yellow zone sizes 473 __ addptr(rax, max_pages * page_size); 474 475 // check against the current stack bottom 476 __ cmpptr(rsp, rax); 477 __ jcc(Assembler::above, after_frame_check); 478 479 // Restore sender's sp as SP. This is necessary if the sender's 480 // frame is an extended compiled frame (see gen_c2i_adapter()) 481 // and safer anyway in case of JSR292 adaptations. 482 483 __ pop(rax); // return address must be moved if SP is changed 484 __ mov(rsp, r13); 485 __ push(rax); 486 487 // Note: the restored frame is not necessarily interpreted. 488 // Use the shared runtime version of the StackOverflowError. 489 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated"); 490 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry())); 491 492 // all done with frame size check 493 __ bind(after_frame_check); 494 } 495 496 // Allocate monitor and lock method (asm interpreter) 497 // 498 // Args: 499 // rbx: Method* 500 // r14: locals 501 // 502 // Kills: 503 // rax 504 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs) 505 // rscratch1, rscratch2 (scratch regs) 506 void InterpreterGenerator::lock_method(void) { 507 // synchronize method 508 const Address access_flags(rbx, Method::access_flags_offset()); 509 const Address monitor_block_top( 510 rbp, 511 frame::interpreter_frame_monitor_block_top_offset * wordSize); 512 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 513 514 #ifdef ASSERT 515 { 516 Label L; 517 __ movl(rax, access_flags); 518 __ testl(rax, JVM_ACC_SYNCHRONIZED); 519 __ jcc(Assembler::notZero, L); 520 __ stop("method doesn't need synchronization"); 521 __ bind(L); 522 } 523 #endif // ASSERT 524 525 // get synchronization object 526 { 527 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 528 Label done; 529 __ movl(rax, access_flags); 530 __ testl(rax, JVM_ACC_STATIC); 531 // get receiver (assume this is frequent case) 532 __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0))); 533 __ jcc(Assembler::zero, done); 534 __ movptr(rax, Address(rbx, Method::const_offset())); 535 __ movptr(rax, Address(rax, ConstMethod::constants_offset())); 536 __ movptr(rax, Address(rax, 537 ConstantPool::pool_holder_offset_in_bytes())); 538 __ movptr(rax, Address(rax, mirror_offset)); 539 540 #ifdef ASSERT 541 { 542 Label L; 543 __ testptr(rax, rax); 544 __ jcc(Assembler::notZero, L); 545 __ stop("synchronization object is NULL"); 546 __ bind(L); 547 } 548 #endif // ASSERT 549 550 __ bind(done); 551 } 552 553 // add space for monitor & lock 554 __ subptr(rsp, entry_size); // add space for a monitor entry 555 __ movptr(monitor_block_top, rsp); // set new monitor block top 556 // store object 557 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); 558 __ movptr(c_rarg1, rsp); // object address 559 __ lock_object(c_rarg1); 560 } 561 562 // Generate a fixed interpreter frame. This is identical setup for 563 // interpreted methods and for native methods hence the shared code. 564 // 565 // Args: 566 // rax: return address 567 // rbx: Method* 568 // r14: pointer to locals 569 // r13: sender sp 570 // rdx: cp cache 571 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { 572 // initialize fixed part of activation frame 573 __ push(rax); // save return address 574 __ enter(); // save old & set new rbp 575 __ push(r13); // set sender sp 576 __ push((int)NULL_WORD); // leave last_sp as null 577 __ movptr(r13, Address(rbx, Method::const_offset())); // get ConstMethod* 578 __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase 579 __ push(rbx); // save Method* 580 if (ProfileInterpreter) { 581 Label method_data_continue; 582 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset()))); 583 __ testptr(rdx, rdx); 584 __ jcc(Assembler::zero, method_data_continue); 585 __ addptr(rdx, in_bytes(MethodData::data_offset())); 586 __ bind(method_data_continue); 587 __ push(rdx); // set the mdp (method data pointer) 588 } else { 589 __ push(0); 590 } 591 592 __ movptr(rdx, Address(rbx, Method::const_offset())); 593 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset())); 594 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes())); 595 __ push(rdx); // set constant pool cache 596 __ push(r14); // set locals pointer 597 if (native_call) { 598 __ push(0); // no bcp 599 } else { 600 __ push(r13); // set bcp 601 } 602 __ push(0); // reserve word for pointer to expression stack bottom 603 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom 604 } 605 606 // End of helpers 607 608 // Various method entries 609 //------------------------------------------------------------------------------------------------------------------------ 610 // 611 // 612 613 // Call an accessor method (assuming it is resolved, otherwise drop 614 // into vanilla (slow path) entry 615 address InterpreterGenerator::generate_accessor_entry(void) { 616 // rbx: Method* 617 618 // r13: senderSP must preserver for slow path, set SP to it on fast path 619 620 address entry_point = __ pc(); 621 Label xreturn_path; 622 623 // do fastpath for resolved accessor methods 624 if (UseFastAccessorMethods) { 625 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites 626 // thereof; parameter size = 1 627 // Note: We can only use this code if the getfield has been resolved 628 // and if we don't have a null-pointer exception => check for 629 // these conditions first and use slow path if necessary. 630 Label slow_path; 631 // If we need a safepoint check, generate full interpreter entry. 632 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 633 SafepointSynchronize::_not_synchronized); 634 635 __ jcc(Assembler::notEqual, slow_path); 636 // rbx: method 637 __ movptr(rax, Address(rsp, wordSize)); 638 639 // check if local 0 != NULL and read field 640 __ testptr(rax, rax); 641 __ jcc(Assembler::zero, slow_path); 642 643 // read first instruction word and extract bytecode @ 1 and index @ 2 644 __ movptr(rdx, Address(rbx, Method::const_offset())); 645 __ movptr(rdi, Address(rdx, ConstMethod::constants_offset())); 646 __ movl(rdx, Address(rdx, ConstMethod::codes_offset())); 647 // Shift codes right to get the index on the right. 648 // The bytecode fetched looks like <index><0xb4><0x2a> 649 __ shrl(rdx, 2 * BitsPerByte); 650 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); 651 __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes())); 652 653 // rax: local 0 654 // rbx: method 655 // rdx: constant pool cache index 656 // rdi: constant pool cache 657 658 // check if getfield has been resolved and read constant pool cache entry 659 // check the validity of the cache entry by testing whether _indices field 660 // contains Bytecode::_getfield in b1 byte. 661 assert(in_words(ConstantPoolCacheEntry::size()) == 4, 662 "adjust shift below"); 663 __ movl(rcx, 664 Address(rdi, 665 rdx, 666 Address::times_8, 667 ConstantPoolCache::base_offset() + 668 ConstantPoolCacheEntry::indices_offset())); 669 __ shrl(rcx, 2 * BitsPerByte); 670 __ andl(rcx, 0xFF); 671 __ cmpl(rcx, Bytecodes::_getfield); 672 __ jcc(Assembler::notEqual, slow_path); 673 674 // Note: constant pool entry is not valid before bytecode is resolved 675 __ movptr(rcx, 676 Address(rdi, 677 rdx, 678 Address::times_8, 679 ConstantPoolCache::base_offset() + 680 ConstantPoolCacheEntry::f2_offset())); 681 // edx: flags 682 __ movl(rdx, 683 Address(rdi, 684 rdx, 685 Address::times_8, 686 ConstantPoolCache::base_offset() + 687 ConstantPoolCacheEntry::flags_offset())); 688 689 Label notObj, notInt, notByte, notShort; 690 const Address field_address(rax, rcx, Address::times_1); 691 692 // Need to differentiate between igetfield, agetfield, bgetfield etc. 693 // because they are different sizes. 694 // Use the type from the constant pool cache 695 __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift); 696 // Make sure we don't need to mask edx after the above shift 697 ConstantPoolCacheEntry::verify_tos_state_shift(); 698 699 __ cmpl(rdx, atos); 700 __ jcc(Assembler::notEqual, notObj); 701 // atos 702 __ load_heap_oop(rax, field_address); 703 __ jmp(xreturn_path); 704 705 __ bind(notObj); 706 __ cmpl(rdx, itos); 707 __ jcc(Assembler::notEqual, notInt); 708 // itos 709 __ movl(rax, field_address); 710 __ jmp(xreturn_path); 711 712 __ bind(notInt); 713 __ cmpl(rdx, btos); 714 __ jcc(Assembler::notEqual, notByte); 715 // btos 716 __ load_signed_byte(rax, field_address); 717 __ jmp(xreturn_path); 718 719 __ bind(notByte); 720 __ cmpl(rdx, stos); 721 __ jcc(Assembler::notEqual, notShort); 722 // stos 723 __ load_signed_short(rax, field_address); 724 __ jmp(xreturn_path); 725 726 __ bind(notShort); 727 #ifdef ASSERT 728 Label okay; 729 __ cmpl(rdx, ctos); 730 __ jcc(Assembler::equal, okay); 731 __ stop("what type is this?"); 732 __ bind(okay); 733 #endif 734 // ctos 735 __ load_unsigned_short(rax, field_address); 736 737 __ bind(xreturn_path); 738 739 // _ireturn/_areturn 740 __ pop(rdi); 741 __ mov(rsp, r13); 742 __ jmp(rdi); 743 __ ret(0); 744 745 // generate a vanilla interpreter entry as the slow path 746 __ bind(slow_path); 747 (void) generate_normal_entry(false); 748 } else { 749 (void) generate_normal_entry(false); 750 } 751 752 return entry_point; 753 } 754 755 // Method entry for java.lang.ref.Reference.get. 756 address InterpreterGenerator::generate_Reference_get_entry(void) { 757 #if INCLUDE_ALL_GCS 758 // Code: _aload_0, _getfield, _areturn 759 // parameter size = 1 760 // 761 // The code that gets generated by this routine is split into 2 parts: 762 // 1. The "intrinsified" code for G1 (or any SATB based GC), 763 // 2. The slow path - which is an expansion of the regular method entry. 764 // 765 // Notes:- 766 // * In the G1 code we do not check whether we need to block for 767 // a safepoint. If G1 is enabled then we must execute the specialized 768 // code for Reference.get (except when the Reference object is null) 769 // so that we can log the value in the referent field with an SATB 770 // update buffer. 771 // If the code for the getfield template is modified so that the 772 // G1 pre-barrier code is executed when the current method is 773 // Reference.get() then going through the normal method entry 774 // will be fine. 775 // * The G1 code can, however, check the receiver object (the instance 776 // of java.lang.Reference) and jump to the slow path if null. If the 777 // Reference object is null then we obviously cannot fetch the referent 778 // and so we don't need to call the G1 pre-barrier. Thus we can use the 779 // regular method entry code to generate the NPE. 780 // 781 // This code is based on generate_accessor_enty. 782 // 783 // rbx: Method* 784 785 // r13: senderSP must preserve for slow path, set SP to it on fast path 786 787 address entry = __ pc(); 788 789 const int referent_offset = java_lang_ref_Reference::referent_offset; 790 guarantee(referent_offset > 0, "referent offset not initialized"); 791 792 if (UseG1GC) { 793 Label slow_path; 794 // rbx: method 795 796 // Check if local 0 != NULL 797 // If the receiver is null then it is OK to jump to the slow path. 798 __ movptr(rax, Address(rsp, wordSize)); 799 800 __ testptr(rax, rax); 801 __ jcc(Assembler::zero, slow_path); 802 803 // rax: local 0 804 // rbx: method (but can be used as scratch now) 805 // rdx: scratch 806 // rdi: scratch 807 808 // Generate the G1 pre-barrier code to log the value of 809 // the referent field in an SATB buffer. 810 811 // Load the value of the referent field. 812 const Address field_address(rax, referent_offset); 813 __ load_heap_oop(rax, field_address); 814 815 // Generate the G1 pre-barrier code to log the value of 816 // the referent field in an SATB buffer. 817 __ g1_write_barrier_pre(noreg /* obj */, 818 rax /* pre_val */, 819 r15_thread /* thread */, 820 rbx /* tmp */, 821 true /* tosca_live */, 822 true /* expand_call */); 823 824 // _areturn 825 __ pop(rdi); // get return address 826 __ mov(rsp, r13); // set sp to sender sp 827 __ jmp(rdi); 828 __ ret(0); 829 830 // generate a vanilla interpreter entry as the slow path 831 __ bind(slow_path); 832 (void) generate_normal_entry(false); 833 834 return entry; 835 } 836 #endif // INCLUDE_ALL_GCS 837 838 // If G1 is not enabled then attempt to go through the accessor entry point 839 // Reference.get is an accessor 840 return generate_accessor_entry(); 841 } 842 843 844 // Interpreter stub for calling a native method. (asm interpreter) 845 // This sets up a somewhat different looking stack for calling the 846 // native method than the typical interpreter frame setup. 847 address InterpreterGenerator::generate_native_entry(bool synchronized) { 848 // determine code generation flags 849 bool inc_counter = UseCompiler || CountCompiledCalls; 850 851 // rbx: Method* 852 // r13: sender sp 853 854 address entry_point = __ pc(); 855 856 const Address constMethod (rbx, Method::const_offset()); 857 const Address access_flags (rbx, Method::access_flags_offset()); 858 const Address size_of_parameters(rcx, ConstMethod:: 859 size_of_parameters_offset()); 860 861 862 // get parameter size (always needed) 863 __ movptr(rcx, constMethod); 864 __ load_unsigned_short(rcx, size_of_parameters); 865 866 // native calls don't need the stack size check since they have no 867 // expression stack and the arguments are already on the stack and 868 // we only add a handful of words to the stack 869 870 // rbx: Method* 871 // rcx: size of parameters 872 // r13: sender sp 873 __ pop(rax); // get return address 874 875 // for natives the size of locals is zero 876 877 // compute beginning of parameters (r14) 878 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize)); 879 880 // add 2 zero-initialized slots for native calls 881 // initialize result_handler slot 882 __ push((int) NULL_WORD); 883 // slot for oop temp 884 // (static native method holder mirror/jni oop result) 885 __ push((int) NULL_WORD); 886 887 // initialize fixed part of activation frame 888 generate_fixed_frame(true); 889 890 // make sure method is native & not abstract 891 #ifdef ASSERT 892 __ movl(rax, access_flags); 893 { 894 Label L; 895 __ testl(rax, JVM_ACC_NATIVE); 896 __ jcc(Assembler::notZero, L); 897 __ stop("tried to execute non-native method as native"); 898 __ bind(L); 899 } 900 { 901 Label L; 902 __ testl(rax, JVM_ACC_ABSTRACT); 903 __ jcc(Assembler::zero, L); 904 __ stop("tried to execute abstract method in interpreter"); 905 __ bind(L); 906 } 907 #endif 908 909 // Since at this point in the method invocation the exception handler 910 // would try to exit the monitor of synchronized methods which hasn't 911 // been entered yet, we set the thread local variable 912 // _do_not_unlock_if_synchronized to true. The remove_activation will 913 // check this flag. 914 915 const Address do_not_unlock_if_synchronized(r15_thread, 916 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 917 __ movbool(do_not_unlock_if_synchronized, true); 918 919 // increment invocation count & check for overflow 920 Label invocation_counter_overflow; 921 if (inc_counter) { 922 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); 923 } 924 925 Label continue_after_compile; 926 __ bind(continue_after_compile); 927 928 bang_stack_shadow_pages(true); 929 930 // reset the _do_not_unlock_if_synchronized flag 931 __ movbool(do_not_unlock_if_synchronized, false); 932 933 // check for synchronized methods 934 // Must happen AFTER invocation_counter check and stack overflow check, 935 // so method is not locked if overflows. 936 if (synchronized) { 937 lock_method(); 938 } else { 939 // no synchronization necessary 940 #ifdef ASSERT 941 { 942 Label L; 943 __ movl(rax, access_flags); 944 __ testl(rax, JVM_ACC_SYNCHRONIZED); 945 __ jcc(Assembler::zero, L); 946 __ stop("method needs synchronization"); 947 __ bind(L); 948 } 949 #endif 950 } 951 952 // start execution 953 #ifdef ASSERT 954 { 955 Label L; 956 const Address monitor_block_top(rbp, 957 frame::interpreter_frame_monitor_block_top_offset * wordSize); 958 __ movptr(rax, monitor_block_top); 959 __ cmpptr(rax, rsp); 960 __ jcc(Assembler::equal, L); 961 __ stop("broken stack frame setup in interpreter"); 962 __ bind(L); 963 } 964 #endif 965 966 // jvmti support 967 __ notify_method_entry(); 968 969 // work registers 970 const Register method = rbx; 971 const Register t = r11; 972 973 // allocate space for parameters 974 __ get_method(method); 975 __ movptr(t, Address(method, Method::const_offset())); 976 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset())); 977 __ shll(t, Interpreter::logStackElementSize); 978 979 __ subptr(rsp, t); 980 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 981 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI) 982 983 // get signature handler 984 { 985 Label L; 986 __ movptr(t, Address(method, Method::signature_handler_offset())); 987 __ testptr(t, t); 988 __ jcc(Assembler::notZero, L); 989 __ call_VM(noreg, 990 CAST_FROM_FN_PTR(address, 991 InterpreterRuntime::prepare_native_call), 992 method); 993 __ get_method(method); 994 __ movptr(t, Address(method, Method::signature_handler_offset())); 995 __ bind(L); 996 } 997 998 // call signature handler 999 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14, 1000 "adjust this code"); 1001 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp, 1002 "adjust this code"); 1003 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1, 1004 "adjust this code"); 1005 1006 // The generated handlers do not touch RBX (the method oop). 1007 // However, large signatures cannot be cached and are generated 1008 // each time here. The slow-path generator can do a GC on return, 1009 // so we must reload it after the call. 1010 __ call(t); 1011 __ get_method(method); // slow path can do a GC, reload RBX 1012 1013 1014 // result handler is in rax 1015 // set result handler 1016 __ movptr(Address(rbp, 1017 (frame::interpreter_frame_result_handler_offset) * wordSize), 1018 rax); 1019 1020 // pass mirror handle if static call 1021 { 1022 Label L; 1023 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 1024 __ movl(t, Address(method, Method::access_flags_offset())); 1025 __ testl(t, JVM_ACC_STATIC); 1026 __ jcc(Assembler::zero, L); 1027 // get mirror 1028 __ movptr(t, Address(method, Method::const_offset())); 1029 __ movptr(t, Address(t, ConstMethod::constants_offset())); 1030 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes())); 1031 __ movptr(t, Address(t, mirror_offset)); 1032 // copy mirror into activation frame 1033 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), 1034 t); 1035 // pass handle to mirror 1036 __ lea(c_rarg1, 1037 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize)); 1038 __ bind(L); 1039 } 1040 1041 // get native function entry point 1042 { 1043 Label L; 1044 __ movptr(rax, Address(method, Method::native_function_offset())); 1045 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry()); 1046 __ movptr(rscratch2, unsatisfied.addr()); 1047 __ cmpptr(rax, rscratch2); 1048 __ jcc(Assembler::notEqual, L); 1049 __ call_VM(noreg, 1050 CAST_FROM_FN_PTR(address, 1051 InterpreterRuntime::prepare_native_call), 1052 method); 1053 __ get_method(method); 1054 __ movptr(rax, Address(method, Method::native_function_offset())); 1055 __ bind(L); 1056 } 1057 1058 // pass JNIEnv 1059 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset())); 1060 1061 // It is enough that the pc() points into the right code 1062 // segment. It does not have to be the correct return pc. 1063 __ set_last_Java_frame(rsp, rbp, (address) __ pc()); 1064 1065 // change thread state 1066 #ifdef ASSERT 1067 { 1068 Label L; 1069 __ movl(t, Address(r15_thread, JavaThread::thread_state_offset())); 1070 __ cmpl(t, _thread_in_Java); 1071 __ jcc(Assembler::equal, L); 1072 __ stop("Wrong thread state in native stub"); 1073 __ bind(L); 1074 } 1075 #endif 1076 1077 // Change state to native 1078 1079 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), 1080 _thread_in_native); 1081 1082 // Call the native method. 1083 __ call(rax); 1084 // result potentially in rax or xmm0 1085 1086 // Verify or restore cpu control state after JNI call 1087 __ restore_cpu_control_state_after_jni(); 1088 1089 // NOTE: The order of these pushes is known to frame::interpreter_frame_result 1090 // in order to extract the result of a method call. If the order of these 1091 // pushes change or anything else is added to the stack then the code in 1092 // interpreter_frame_result must also change. 1093 1094 __ push(dtos); 1095 __ push(ltos); 1096 1097 // change thread state 1098 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), 1099 _thread_in_native_trans); 1100 1101 if (os::is_MP()) { 1102 if (UseMembar) { 1103 // Force this write out before the read below 1104 __ membar(Assembler::Membar_mask_bits( 1105 Assembler::LoadLoad | Assembler::LoadStore | 1106 Assembler::StoreLoad | Assembler::StoreStore)); 1107 } else { 1108 // Write serialization page so VM thread can do a pseudo remote membar. 1109 // We use the current thread pointer to calculate a thread specific 1110 // offset to write to within the page. This minimizes bus traffic 1111 // due to cache line collision. 1112 __ serialize_memory(r15_thread, rscratch2); 1113 } 1114 } 1115 1116 // check for safepoint operation in progress and/or pending suspend requests 1117 { 1118 Label Continue; 1119 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 1120 SafepointSynchronize::_not_synchronized); 1121 1122 Label L; 1123 __ jcc(Assembler::notEqual, L); 1124 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0); 1125 __ jcc(Assembler::equal, Continue); 1126 __ bind(L); 1127 1128 // Don't use call_VM as it will see a possible pending exception 1129 // and forward it and never return here preventing us from 1130 // clearing _last_native_pc down below. Also can't use 1131 // call_VM_leaf either as it will check to see if r13 & r14 are 1132 // preserved and correspond to the bcp/locals pointers. So we do a 1133 // runtime call by hand. 1134 // 1135 __ mov(c_rarg0, r15_thread); 1136 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) 1137 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 1138 __ andptr(rsp, -16); // align stack as required by ABI 1139 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans))); 1140 __ mov(rsp, r12); // restore sp 1141 __ reinit_heapbase(); 1142 __ bind(Continue); 1143 } 1144 1145 // change thread state 1146 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java); 1147 1148 // reset_last_Java_frame 1149 __ reset_last_Java_frame(true, true); 1150 1151 // reset handle block 1152 __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset())); 1153 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD); 1154 1155 // If result is an oop unbox and store it in frame where gc will see it 1156 // and result handler will pick it up 1157 1158 { 1159 Label no_oop, store_result; 1160 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT))); 1161 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize)); 1162 __ jcc(Assembler::notEqual, no_oop); 1163 // retrieve result 1164 __ pop(ltos); 1165 __ testptr(rax, rax); 1166 __ jcc(Assembler::zero, store_result); 1167 __ movptr(rax, Address(rax, 0)); 1168 __ bind(store_result); 1169 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax); 1170 // keep stack depth as expected by pushing oop which will eventually be discarde 1171 __ push(ltos); 1172 __ bind(no_oop); 1173 } 1174 1175 1176 { 1177 Label no_reguard; 1178 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()), 1179 JavaThread::stack_guard_yellow_disabled); 1180 __ jcc(Assembler::notEqual, no_reguard); 1181 1182 __ pusha(); // XXX only save smashed registers 1183 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) 1184 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 1185 __ andptr(rsp, -16); // align stack as required by ABI 1186 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); 1187 __ mov(rsp, r12); // restore sp 1188 __ popa(); // XXX only restore smashed registers 1189 __ reinit_heapbase(); 1190 1191 __ bind(no_reguard); 1192 } 1193 1194 1195 // The method register is junk from after the thread_in_native transition 1196 // until here. Also can't call_VM until the bcp has been 1197 // restored. Need bcp for throwing exception below so get it now. 1198 __ get_method(method); 1199 1200 // restore r13 to have legal interpreter frame, i.e., bci == 0 <=> 1201 // r13 == code_base() 1202 __ movptr(r13, Address(method, Method::const_offset())); // get ConstMethod* 1203 __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase 1204 // handle exceptions (exception handling will handle unlocking!) 1205 { 1206 Label L; 1207 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); 1208 __ jcc(Assembler::zero, L); 1209 // Note: At some point we may want to unify this with the code 1210 // used in call_VM_base(); i.e., we should use the 1211 // StubRoutines::forward_exception code. For now this doesn't work 1212 // here because the rsp is not correctly set at this point. 1213 __ MacroAssembler::call_VM(noreg, 1214 CAST_FROM_FN_PTR(address, 1215 InterpreterRuntime::throw_pending_exception)); 1216 __ should_not_reach_here(); 1217 __ bind(L); 1218 } 1219 1220 // do unlocking if necessary 1221 { 1222 Label L; 1223 __ movl(t, Address(method, Method::access_flags_offset())); 1224 __ testl(t, JVM_ACC_SYNCHRONIZED); 1225 __ jcc(Assembler::zero, L); 1226 // the code below should be shared with interpreter macro 1227 // assembler implementation 1228 { 1229 Label unlock; 1230 // BasicObjectLock will be first in list, since this is a 1231 // synchronized method. However, need to check that the object 1232 // has not been unlocked by an explicit monitorexit bytecode. 1233 const Address monitor(rbp, 1234 (intptr_t)(frame::interpreter_frame_initial_sp_offset * 1235 wordSize - sizeof(BasicObjectLock))); 1236 1237 // monitor expect in c_rarg1 for slow unlock path 1238 __ lea(c_rarg1, monitor); // address of first monitor 1239 1240 __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 1241 __ testptr(t, t); 1242 __ jcc(Assembler::notZero, unlock); 1243 1244 // Entry already unlocked, need to throw exception 1245 __ MacroAssembler::call_VM(noreg, 1246 CAST_FROM_FN_PTR(address, 1247 InterpreterRuntime::throw_illegal_monitor_state_exception)); 1248 __ should_not_reach_here(); 1249 1250 __ bind(unlock); 1251 __ unlock_object(c_rarg1); 1252 } 1253 __ bind(L); 1254 } 1255 1256 // jvmti support 1257 // Note: This must happen _after_ handling/throwing any exceptions since 1258 // the exception handler code notifies the runtime of method exits 1259 // too. If this happens before, method entry/exit notifications are 1260 // not properly paired (was bug - gri 11/22/99). 1261 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); 1262 1263 // restore potential result in edx:eax, call result handler to 1264 // restore potential result in ST0 & handle result 1265 1266 __ pop(ltos); 1267 __ pop(dtos); 1268 1269 __ movptr(t, Address(rbp, 1270 (frame::interpreter_frame_result_handler_offset) * wordSize)); 1271 __ call(t); 1272 1273 // remove activation 1274 __ movptr(t, Address(rbp, 1275 frame::interpreter_frame_sender_sp_offset * 1276 wordSize)); // get sender sp 1277 __ leave(); // remove frame anchor 1278 __ pop(rdi); // get return address 1279 __ mov(rsp, t); // set sp to sender sp 1280 __ jmp(rdi); 1281 1282 if (inc_counter) { 1283 // Handle overflow of counter and compile method 1284 __ bind(invocation_counter_overflow); 1285 generate_counter_overflow(&continue_after_compile); 1286 } 1287 1288 return entry_point; 1289 } 1290 1291 // 1292 // Generic interpreted method entry to (asm) interpreter 1293 // 1294 address InterpreterGenerator::generate_normal_entry(bool synchronized) { 1295 // determine code generation flags 1296 bool inc_counter = UseCompiler || CountCompiledCalls; 1297 1298 // ebx: Method* 1299 // r13: sender sp 1300 address entry_point = __ pc(); 1301 1302 const Address constMethod(rbx, Method::const_offset()); 1303 const Address access_flags(rbx, Method::access_flags_offset()); 1304 const Address size_of_parameters(rdx, 1305 ConstMethod::size_of_parameters_offset()); 1306 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset()); 1307 1308 1309 // get parameter size (always needed) 1310 __ movptr(rdx, constMethod); 1311 __ load_unsigned_short(rcx, size_of_parameters); 1312 1313 // rbx: Method* 1314 // rcx: size of parameters 1315 // r13: sender_sp (could differ from sp+wordSize if we were called via c2i ) 1316 1317 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words 1318 __ subl(rdx, rcx); // rdx = no. of additional locals 1319 1320 // YYY 1321 // __ incrementl(rdx); 1322 // __ andl(rdx, -2); 1323 1324 // see if we've got enough room on the stack for locals plus overhead. 1325 generate_stack_overflow_check(); 1326 1327 // get return address 1328 __ pop(rax); 1329 1330 // compute beginning of parameters (r14) 1331 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize)); 1332 1333 // rdx - # of additional locals 1334 // allocate space for locals 1335 // explicitly initialize locals 1336 { 1337 Label exit, loop; 1338 __ testl(rdx, rdx); 1339 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0 1340 __ bind(loop); 1341 __ push((int) NULL_WORD); // initialize local variables 1342 __ decrementl(rdx); // until everything initialized 1343 __ jcc(Assembler::greater, loop); 1344 __ bind(exit); 1345 } 1346 1347 // initialize fixed part of activation frame 1348 generate_fixed_frame(false); 1349 1350 // make sure method is not native & not abstract 1351 #ifdef ASSERT 1352 __ movl(rax, access_flags); 1353 { 1354 Label L; 1355 __ testl(rax, JVM_ACC_NATIVE); 1356 __ jcc(Assembler::zero, L); 1357 __ stop("tried to execute native method as non-native"); 1358 __ bind(L); 1359 } 1360 { 1361 Label L; 1362 __ testl(rax, JVM_ACC_ABSTRACT); 1363 __ jcc(Assembler::zero, L); 1364 __ stop("tried to execute abstract method in interpreter"); 1365 __ bind(L); 1366 } 1367 #endif 1368 1369 // Since at this point in the method invocation the exception 1370 // handler would try to exit the monitor of synchronized methods 1371 // which hasn't been entered yet, we set the thread local variable 1372 // _do_not_unlock_if_synchronized to true. The remove_activation 1373 // will check this flag. 1374 1375 const Address do_not_unlock_if_synchronized(r15_thread, 1376 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1377 __ movbool(do_not_unlock_if_synchronized, true); 1378 1379 // increment invocation count & check for overflow 1380 Label invocation_counter_overflow; 1381 Label profile_method; 1382 Label profile_method_continue; 1383 if (inc_counter) { 1384 generate_counter_incr(&invocation_counter_overflow, 1385 &profile_method, 1386 &profile_method_continue); 1387 if (ProfileInterpreter) { 1388 __ bind(profile_method_continue); 1389 } 1390 } 1391 1392 Label continue_after_compile; 1393 __ bind(continue_after_compile); 1394 1395 // check for synchronized interpreted methods 1396 bang_stack_shadow_pages(false); 1397 1398 // reset the _do_not_unlock_if_synchronized flag 1399 __ movbool(do_not_unlock_if_synchronized, false); 1400 1401 // check for synchronized methods 1402 // Must happen AFTER invocation_counter check and stack overflow check, 1403 // so method is not locked if overflows. 1404 if (synchronized) { 1405 // Allocate monitor and lock method 1406 lock_method(); 1407 } else { 1408 // no synchronization necessary 1409 #ifdef ASSERT 1410 { 1411 Label L; 1412 __ movl(rax, access_flags); 1413 __ testl(rax, JVM_ACC_SYNCHRONIZED); 1414 __ jcc(Assembler::zero, L); 1415 __ stop("method needs synchronization"); 1416 __ bind(L); 1417 } 1418 #endif 1419 } 1420 1421 // start execution 1422 #ifdef ASSERT 1423 { 1424 Label L; 1425 const Address monitor_block_top (rbp, 1426 frame::interpreter_frame_monitor_block_top_offset * wordSize); 1427 __ movptr(rax, monitor_block_top); 1428 __ cmpptr(rax, rsp); 1429 __ jcc(Assembler::equal, L); 1430 __ stop("broken stack frame setup in interpreter"); 1431 __ bind(L); 1432 } 1433 #endif 1434 1435 // jvmti support 1436 __ notify_method_entry(); 1437 1438 __ dispatch_next(vtos); 1439 1440 // invocation counter overflow 1441 if (inc_counter) { 1442 if (ProfileInterpreter) { 1443 // We have decided to profile this method in the interpreter 1444 __ bind(profile_method); 1445 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1446 __ set_method_data_pointer_for_bcp(); 1447 __ get_method(rbx); 1448 __ jmp(profile_method_continue); 1449 } 1450 // Handle overflow of counter and compile method 1451 __ bind(invocation_counter_overflow); 1452 generate_counter_overflow(&continue_after_compile); 1453 } 1454 1455 return entry_point; 1456 } 1457 1458 // Entry points 1459 // 1460 // Here we generate the various kind of entries into the interpreter. 1461 // The two main entry type are generic bytecode methods and native 1462 // call method. These both come in synchronized and non-synchronized 1463 // versions but the frame layout they create is very similar. The 1464 // other method entry types are really just special purpose entries 1465 // that are really entry and interpretation all in one. These are for 1466 // trivial methods like accessor, empty, or special math methods. 1467 // 1468 // When control flow reaches any of the entry types for the interpreter 1469 // the following holds -> 1470 // 1471 // Arguments: 1472 // 1473 // rbx: Method* 1474 // 1475 // Stack layout immediately at entry 1476 // 1477 // [ return address ] <--- rsp 1478 // [ parameter n ] 1479 // ... 1480 // [ parameter 1 ] 1481 // [ expression stack ] (caller's java expression stack) 1482 1483 // Assuming that we don't go to one of the trivial specialized entries 1484 // the stack will look like below when we are ready to execute the 1485 // first bytecode (or call the native routine). The register usage 1486 // will be as the template based interpreter expects (see 1487 // interpreter_amd64.hpp). 1488 // 1489 // local variables follow incoming parameters immediately; i.e. 1490 // the return address is moved to the end of the locals). 1491 // 1492 // [ monitor entry ] <--- rsp 1493 // ... 1494 // [ monitor entry ] 1495 // [ expr. stack bottom ] 1496 // [ saved r13 ] 1497 // [ current r14 ] 1498 // [ Method* ] 1499 // [ saved ebp ] <--- rbp 1500 // [ return address ] 1501 // [ local variable m ] 1502 // ... 1503 // [ local variable 1 ] 1504 // [ parameter n ] 1505 // ... 1506 // [ parameter 1 ] <--- r14 1507 1508 address AbstractInterpreterGenerator::generate_method_entry( 1509 AbstractInterpreter::MethodKind kind) { 1510 // determine code generation flags 1511 bool synchronized = false; 1512 address entry_point = NULL; 1513 1514 switch (kind) { 1515 case Interpreter::zerolocals : break; 1516 case Interpreter::zerolocals_synchronized: synchronized = true; break; 1517 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break; 1518 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break; 1519 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break; 1520 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break; 1521 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break; 1522 1523 case Interpreter::java_lang_math_sin : // fall thru 1524 case Interpreter::java_lang_math_cos : // fall thru 1525 case Interpreter::java_lang_math_tan : // fall thru 1526 case Interpreter::java_lang_math_abs : // fall thru 1527 case Interpreter::java_lang_math_log : // fall thru 1528 case Interpreter::java_lang_math_log10 : // fall thru 1529 case Interpreter::java_lang_math_sqrt : // fall thru 1530 case Interpreter::java_lang_math_pow : // fall thru 1531 case Interpreter::java_lang_math_exp : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break; 1532 case Interpreter::java_lang_ref_reference_get 1533 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break; 1534 default: 1535 fatal(err_msg("unexpected method kind: %d", kind)); 1536 break; 1537 } 1538 1539 if (entry_point) { 1540 return entry_point; 1541 } 1542 1543 return ((InterpreterGenerator*) this)-> 1544 generate_normal_entry(synchronized); 1545 } 1546 1547 // These should never be compiled since the interpreter will prefer 1548 // the compiled version to the intrinsic version. 1549 bool AbstractInterpreter::can_be_compiled(methodHandle m) { 1550 switch (method_kind(m)) { 1551 case Interpreter::java_lang_math_sin : // fall thru 1552 case Interpreter::java_lang_math_cos : // fall thru 1553 case Interpreter::java_lang_math_tan : // fall thru 1554 case Interpreter::java_lang_math_abs : // fall thru 1555 case Interpreter::java_lang_math_log : // fall thru 1556 case Interpreter::java_lang_math_log10 : // fall thru 1557 case Interpreter::java_lang_math_sqrt : // fall thru 1558 case Interpreter::java_lang_math_pow : // fall thru 1559 case Interpreter::java_lang_math_exp : 1560 return false; 1561 default: 1562 return true; 1563 } 1564 } 1565 1566 // How much stack a method activation needs in words. 1567 int AbstractInterpreter::size_top_interpreter_activation(Method* method) { 1568 const int entry_size = frame::interpreter_frame_monitor_size(); 1569 1570 // total overhead size: entry_size + (saved rbp thru expr stack 1571 // bottom). be sure to change this if you add/subtract anything 1572 // to/from the overhead area 1573 const int overhead_size = 1574 -(frame::interpreter_frame_initial_sp_offset) + entry_size; 1575 1576 const int stub_code = frame::entry_frame_after_call_words; 1577 const int method_stack = (method->max_locals() + method->max_stack()) * 1578 Interpreter::stackElementWords; 1579 return (overhead_size + method_stack + stub_code); 1580 } 1581 1582 int AbstractInterpreter::layout_activation(Method* method, 1583 int tempcount, 1584 int popframe_extra_args, 1585 int moncount, 1586 int caller_actual_parameters, 1587 int callee_param_count, 1588 int callee_locals, 1589 frame* caller, 1590 frame* interpreter_frame, 1591 bool is_top_frame, 1592 bool is_bottom_frame) { 1593 // Note: This calculation must exactly parallel the frame setup 1594 // in AbstractInterpreterGenerator::generate_method_entry. 1595 // If interpreter_frame!=NULL, set up the method, locals, and monitors. 1596 // The frame interpreter_frame, if not NULL, is guaranteed to be the 1597 // right size, as determined by a previous call to this method. 1598 // It is also guaranteed to be walkable even though it is in a skeletal state 1599 1600 // fixed size of an interpreter frame: 1601 int max_locals = method->max_locals() * Interpreter::stackElementWords; 1602 int extra_locals = (method->max_locals() - method->size_of_parameters()) * 1603 Interpreter::stackElementWords; 1604 1605 int overhead = frame::sender_sp_offset - 1606 frame::interpreter_frame_initial_sp_offset; 1607 // Our locals were accounted for by the caller (or last_frame_adjust 1608 // on the transistion) Since the callee parameters already account 1609 // for the callee's params we only need to account for the extra 1610 // locals. 1611 int size = overhead + 1612 (callee_locals - callee_param_count)*Interpreter::stackElementWords + 1613 moncount * frame::interpreter_frame_monitor_size() + 1614 tempcount* Interpreter::stackElementWords + popframe_extra_args; 1615 if (interpreter_frame != NULL) { 1616 #ifdef ASSERT 1617 if (!EnableInvokeDynamic) 1618 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences? 1619 // Probably, since deoptimization doesn't work yet. 1620 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable"); 1621 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)"); 1622 #endif 1623 1624 interpreter_frame->interpreter_frame_set_method(method); 1625 // NOTE the difference in using sender_sp and 1626 // interpreter_frame_sender_sp interpreter_frame_sender_sp is 1627 // the original sp of the caller (the unextended_sp) and 1628 // sender_sp is fp+16 XXX 1629 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1; 1630 1631 #ifdef ASSERT 1632 if (caller->is_interpreted_frame()) { 1633 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement"); 1634 } 1635 #endif 1636 1637 interpreter_frame->interpreter_frame_set_locals(locals); 1638 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin(); 1639 BasicObjectLock* monbot = montop - moncount; 1640 interpreter_frame->interpreter_frame_set_monitor_end(monbot); 1641 1642 // Set last_sp 1643 intptr_t* esp = (intptr_t*) monbot - 1644 tempcount*Interpreter::stackElementWords - 1645 popframe_extra_args; 1646 interpreter_frame->interpreter_frame_set_last_sp(esp); 1647 1648 // All frames but the initial (oldest) interpreter frame we fill in have 1649 // a value for sender_sp that allows walking the stack but isn't 1650 // truly correct. Correct the value here. 1651 if (extra_locals != 0 && 1652 interpreter_frame->sender_sp() == 1653 interpreter_frame->interpreter_frame_sender_sp()) { 1654 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + 1655 extra_locals); 1656 } 1657 *interpreter_frame->interpreter_frame_cache_addr() = 1658 method->constants()->cache(); 1659 } 1660 return size; 1661 } 1662 1663 //----------------------------------------------------------------------------- 1664 // Exceptions 1665 1666 void TemplateInterpreterGenerator::generate_throw_exception() { 1667 // Entry point in previous activation (i.e., if the caller was 1668 // interpreted) 1669 Interpreter::_rethrow_exception_entry = __ pc(); 1670 // Restore sp to interpreter_frame_last_sp even though we are going 1671 // to empty the expression stack for the exception processing. 1672 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 1673 // rax: exception 1674 // rdx: return address/pc that threw exception 1675 __ restore_bcp(); // r13 points to call/send 1676 __ restore_locals(); 1677 __ reinit_heapbase(); // restore r12 as heapbase. 1678 // Entry point for exceptions thrown within interpreter code 1679 Interpreter::_throw_exception_entry = __ pc(); 1680 // expression stack is undefined here 1681 // rax: exception 1682 // r13: exception bcp 1683 __ verify_oop(rax); 1684 __ mov(c_rarg1, rax); 1685 1686 // expression stack must be empty before entering the VM in case of 1687 // an exception 1688 __ empty_expression_stack(); 1689 // find exception handler address and preserve exception oop 1690 __ call_VM(rdx, 1691 CAST_FROM_FN_PTR(address, 1692 InterpreterRuntime::exception_handler_for_exception), 1693 c_rarg1); 1694 // rax: exception handler entry point 1695 // rdx: preserved exception oop 1696 // r13: bcp for exception handler 1697 __ push_ptr(rdx); // push exception which is now the only value on the stack 1698 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!) 1699 1700 // If the exception is not handled in the current frame the frame is 1701 // removed and the exception is rethrown (i.e. exception 1702 // continuation is _rethrow_exception). 1703 // 1704 // Note: At this point the bci is still the bxi for the instruction 1705 // which caused the exception and the expression stack is 1706 // empty. Thus, for any VM calls at this point, GC will find a legal 1707 // oop map (with empty expression stack). 1708 1709 // In current activation 1710 // tos: exception 1711 // esi: exception bcp 1712 1713 // 1714 // JVMTI PopFrame support 1715 // 1716 1717 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 1718 __ empty_expression_stack(); 1719 // Set the popframe_processing bit in pending_popframe_condition 1720 // indicating that we are currently handling popframe, so that 1721 // call_VMs that may happen later do not trigger new popframe 1722 // handling cycles. 1723 __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset())); 1724 __ orl(rdx, JavaThread::popframe_processing_bit); 1725 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx); 1726 1727 { 1728 // Check to see whether we are returning to a deoptimized frame. 1729 // (The PopFrame call ensures that the caller of the popped frame is 1730 // either interpreted or compiled and deoptimizes it if compiled.) 1731 // In this case, we can't call dispatch_next() after the frame is 1732 // popped, but instead must save the incoming arguments and restore 1733 // them after deoptimization has occurred. 1734 // 1735 // Note that we don't compare the return PC against the 1736 // deoptimization blob's unpack entry because of the presence of 1737 // adapter frames in C2. 1738 Label caller_not_deoptimized; 1739 __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize)); 1740 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1741 InterpreterRuntime::interpreter_contains), c_rarg1); 1742 __ testl(rax, rax); 1743 __ jcc(Assembler::notZero, caller_not_deoptimized); 1744 1745 // Compute size of arguments for saving when returning to 1746 // deoptimized caller 1747 __ get_method(rax); 1748 __ movptr(rax, Address(rax, Method::const_offset())); 1749 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod:: 1750 size_of_parameters_offset()))); 1751 __ shll(rax, Interpreter::logStackElementSize); 1752 __ restore_locals(); // XXX do we need this? 1753 __ subptr(r14, rax); 1754 __ addptr(r14, wordSize); 1755 // Save these arguments 1756 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1757 Deoptimization:: 1758 popframe_preserve_args), 1759 r15_thread, rax, r14); 1760 1761 __ remove_activation(vtos, rdx, 1762 /* throw_monitor_exception */ false, 1763 /* install_monitor_exception */ false, 1764 /* notify_jvmdi */ false); 1765 1766 // Inform deoptimization that it is responsible for restoring 1767 // these arguments 1768 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), 1769 JavaThread::popframe_force_deopt_reexecution_bit); 1770 1771 // Continue in deoptimization handler 1772 __ jmp(rdx); 1773 1774 __ bind(caller_not_deoptimized); 1775 } 1776 1777 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */ 1778 /* throw_monitor_exception */ false, 1779 /* install_monitor_exception */ false, 1780 /* notify_jvmdi */ false); 1781 1782 // Finish with popframe handling 1783 // A previous I2C followed by a deoptimization might have moved the 1784 // outgoing arguments further up the stack. PopFrame expects the 1785 // mutations to those outgoing arguments to be preserved and other 1786 // constraints basically require this frame to look exactly as 1787 // though it had previously invoked an interpreted activation with 1788 // no space between the top of the expression stack (current 1789 // last_sp) and the top of stack. Rather than force deopt to 1790 // maintain this kind of invariant all the time we call a small 1791 // fixup routine to move the mutated arguments onto the top of our 1792 // expression stack if necessary. 1793 __ mov(c_rarg1, rsp); 1794 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 1795 // PC must point into interpreter here 1796 __ set_last_Java_frame(noreg, rbp, __ pc()); 1797 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2); 1798 __ reset_last_Java_frame(true, true); 1799 // Restore the last_sp and null it out 1800 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 1801 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 1802 1803 __ restore_bcp(); // XXX do we need this? 1804 __ restore_locals(); // XXX do we need this? 1805 // The method data pointer was incremented already during 1806 // call profiling. We have to restore the mdp for the current bcp. 1807 if (ProfileInterpreter) { 1808 __ set_method_data_pointer_for_bcp(); 1809 } 1810 1811 // Clear the popframe condition flag 1812 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), 1813 JavaThread::popframe_inactive); 1814 1815 __ dispatch_next(vtos); 1816 // end of PopFrame support 1817 1818 Interpreter::_remove_activation_entry = __ pc(); 1819 1820 // preserve exception over this code sequence 1821 __ pop_ptr(rax); 1822 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax); 1823 // remove the activation (without doing throws on illegalMonitorExceptions) 1824 __ remove_activation(vtos, rdx, false, true, false); 1825 // restore exception 1826 __ get_vm_result(rax, r15_thread); 1827 1828 // In between activations - previous activation type unknown yet 1829 // compute continuation point - the continuation point expects the 1830 // following registers set up: 1831 // 1832 // rax: exception 1833 // rdx: return address/pc that threw exception 1834 // rsp: expression stack of caller 1835 // rbp: ebp of caller 1836 __ push(rax); // save exception 1837 __ push(rdx); // save return address 1838 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, 1839 SharedRuntime::exception_handler_for_return_address), 1840 r15_thread, rdx); 1841 __ mov(rbx, rax); // save exception handler 1842 __ pop(rdx); // restore return address 1843 __ pop(rax); // restore exception 1844 // Note that an "issuing PC" is actually the next PC after the call 1845 __ jmp(rbx); // jump to exception 1846 // handler of caller 1847 } 1848 1849 1850 // 1851 // JVMTI ForceEarlyReturn support 1852 // 1853 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 1854 address entry = __ pc(); 1855 1856 __ restore_bcp(); 1857 __ restore_locals(); 1858 __ empty_expression_stack(); 1859 __ load_earlyret_value(state); 1860 1861 __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset())); 1862 Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset()); 1863 1864 // Clear the earlyret state 1865 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive); 1866 1867 __ remove_activation(state, rsi, 1868 false, /* throw_monitor_exception */ 1869 false, /* install_monitor_exception */ 1870 true); /* notify_jvmdi */ 1871 __ jmp(rsi); 1872 1873 return entry; 1874 } // end of ForceEarlyReturn support 1875 1876 1877 //----------------------------------------------------------------------------- 1878 // Helper for vtos entry point generation 1879 1880 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, 1881 address& bep, 1882 address& cep, 1883 address& sep, 1884 address& aep, 1885 address& iep, 1886 address& lep, 1887 address& fep, 1888 address& dep, 1889 address& vep) { 1890 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 1891 Label L; 1892 aep = __ pc(); __ push_ptr(); __ jmp(L); 1893 fep = __ pc(); __ push_f(); __ jmp(L); 1894 dep = __ pc(); __ push_d(); __ jmp(L); 1895 lep = __ pc(); __ push_l(); __ jmp(L); 1896 bep = cep = sep = 1897 iep = __ pc(); __ push_i(); 1898 vep = __ pc(); 1899 __ bind(L); 1900 generate_and_dispatch(t); 1901 } 1902 1903 1904 //----------------------------------------------------------------------------- 1905 // Generation of individual instructions 1906 1907 // helpers for generate_and_dispatch 1908 1909 1910 InterpreterGenerator::InterpreterGenerator(StubQueue* code) 1911 : TemplateInterpreterGenerator(code) { 1912 generate_all(); // down here so it can be "virtual" 1913 } 1914 1915 //----------------------------------------------------------------------------- 1916 1917 // Non-product code 1918 #ifndef PRODUCT 1919 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 1920 address entry = __ pc(); 1921 1922 __ push(state); 1923 __ push(c_rarg0); 1924 __ push(c_rarg1); 1925 __ push(c_rarg2); 1926 __ push(c_rarg3); 1927 __ mov(c_rarg2, rax); // Pass itos 1928 #ifdef _WIN64 1929 __ movflt(xmm3, xmm0); // Pass ftos 1930 #endif 1931 __ call_VM(noreg, 1932 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), 1933 c_rarg1, c_rarg2, c_rarg3); 1934 __ pop(c_rarg3); 1935 __ pop(c_rarg2); 1936 __ pop(c_rarg1); 1937 __ pop(c_rarg0); 1938 __ pop(state); 1939 __ ret(0); // return from result handler 1940 1941 return entry; 1942 } 1943 1944 void TemplateInterpreterGenerator::count_bytecode() { 1945 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value)); 1946 } 1947 1948 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 1949 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()])); 1950 } 1951 1952 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 1953 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index)); 1954 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes); 1955 __ orl(rbx, 1956 ((int) t->bytecode()) << 1957 BytecodePairHistogram::log2_number_of_codes); 1958 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx); 1959 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters)); 1960 __ incrementl(Address(rscratch1, rbx, Address::times_4)); 1961 } 1962 1963 1964 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 1965 // Call a little run-time stub to avoid blow-up for each bytecode. 1966 // The run-time runtime saves the right registers, depending on 1967 // the tosca in-state for the given template. 1968 1969 assert(Interpreter::trace_code(t->tos_in()) != NULL, 1970 "entry must have been generated"); 1971 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM) 1972 __ andptr(rsp, -16); // align stack as required by ABI 1973 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in()))); 1974 __ mov(rsp, r12); // restore sp 1975 __ reinit_heapbase(); 1976 } 1977 1978 1979 void TemplateInterpreterGenerator::stop_interpreter_at() { 1980 Label L; 1981 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value), 1982 StopInterpreterAt); 1983 __ jcc(Assembler::notEqual, L); 1984 __ int3(); 1985 __ bind(L); 1986 } 1987 #endif // !PRODUCT 1988 #endif // ! CC_INTERP