1 /* 2 * Copyright (c) 1997, 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 52 #ifndef CC_INTERP 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 // Note: There should be a minimal interpreter frame set up when stack 63 // overflow occurs since we check explicitly for it now. 64 // 65 #ifdef ASSERT 66 { Label L; 67 __ lea(rax, Address(rbp, 68 frame::interpreter_frame_monitor_block_top_offset * wordSize)); 69 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp, 70 // (stack grows negative) 71 __ jcc(Assembler::aboveEqual, L); // check if frame is complete 72 __ stop ("interpreter frame not set up"); 73 __ bind(L); 74 } 75 #endif // ASSERT 76 // Restore bcp under the assumption that the current frame is still 77 // interpreted 78 __ restore_bcp(); 79 80 // expression stack must be empty before entering the VM if an exception 81 // happened 82 __ empty_expression_stack(); 83 __ empty_FPU_stack(); 84 // throw exception 85 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError)); 86 return entry; 87 } 88 89 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) { 90 address entry = __ pc(); 91 // expression stack must be empty before entering the VM if an exception happened 92 __ empty_expression_stack(); 93 __ empty_FPU_stack(); 94 // setup parameters 95 // ??? convention: expect aberrant index in register rbx, 96 __ lea(rax, ExternalAddress((address)name)); 97 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx); 98 return entry; 99 } 100 101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 102 address entry = __ pc(); 103 // object is at TOS 104 __ pop(rax); 105 // expression stack must be empty before entering the VM if an exception 106 // happened 107 __ empty_expression_stack(); 108 __ empty_FPU_stack(); 109 __ call_VM(noreg, 110 CAST_FROM_FN_PTR(address, 111 InterpreterRuntime::throw_ClassCastException), 112 rax); 113 return entry; 114 } 115 116 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) { 117 assert(!pass_oop || message == NULL, "either oop or message but not both"); 118 address entry = __ pc(); 119 if (pass_oop) { 120 // object is at TOS 121 __ pop(rbx); 122 } 123 // expression stack must be empty before entering the VM if an exception happened 124 __ empty_expression_stack(); 125 __ empty_FPU_stack(); 126 // setup parameters 127 __ lea(rax, ExternalAddress((address)name)); 128 if (pass_oop) { 129 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx); 130 } else { 131 if (message != NULL) { 132 __ lea(rbx, ExternalAddress((address)message)); 133 } else { 134 __ movptr(rbx, NULL_WORD); 135 } 136 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx); 137 } 138 // throw exception 139 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); 140 return entry; 141 } 142 143 144 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { 145 address entry = __ pc(); 146 // NULL last_sp until next java call 147 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 148 __ dispatch_next(state); 149 return entry; 150 } 151 152 153 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) { 154 TosState incoming_state = state; 155 address entry = __ pc(); 156 157 #ifdef COMPILER2 158 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases 159 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) { 160 for (int i = 1; i < 8; i++) { 161 __ ffree(i); 162 } 163 } else if (UseSSE < 2) { 164 __ empty_FPU_stack(); 165 } 166 #endif 167 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) { 168 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled"); 169 } else { 170 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled"); 171 } 172 173 // In SSE mode, interpreter returns FP results in xmm0 but they need 174 // to end up back on the FPU so it can operate on them. 175 if (incoming_state == ftos && UseSSE >= 1) { 176 __ subptr(rsp, wordSize); 177 __ movflt(Address(rsp, 0), xmm0); 178 __ fld_s(Address(rsp, 0)); 179 __ addptr(rsp, wordSize); 180 } else if (incoming_state == dtos && UseSSE >= 2) { 181 __ subptr(rsp, 2*wordSize); 182 __ movdbl(Address(rsp, 0), xmm0); 183 __ fld_d(Address(rsp, 0)); 184 __ addptr(rsp, 2*wordSize); 185 } 186 187 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter"); 188 189 // Restore stack bottom in case i2c adjusted stack 190 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 191 // and NULL it as marker that rsp is now tos until next java call 192 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 193 194 if (incoming_state == atos) { 195 Register mdp = rsi; 196 Register tmp = rdi; 197 __ profile_return_type(mdp, rax, tmp); 198 } 199 __ restore_bcp(); 200 __ restore_locals(); 201 202 Label L_got_cache, L_giant_index; 203 if (EnableInvokeDynamic) { 204 __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic); 205 __ jcc(Assembler::equal, L_giant_index); 206 } 207 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2)); 208 __ bind(L_got_cache); 209 __ movl(rbx, Address(rbx, rcx, 210 Address::times_ptr, ConstantPoolCache::base_offset() + 211 ConstantPoolCacheEntry::flags_offset())); 212 __ andptr(rbx, 0xFF); 213 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); 214 __ dispatch_next(state, step); 215 216 // out of the main line of code... 217 if (EnableInvokeDynamic) { 218 __ bind(L_giant_index); 219 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4)); 220 __ jmp(L_got_cache); 221 } 222 223 return entry; 224 } 225 226 227 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) { 228 address entry = __ pc(); 229 230 // In SSE mode, FP results are in xmm0 231 if (state == ftos && UseSSE > 0) { 232 __ subptr(rsp, wordSize); 233 __ movflt(Address(rsp, 0), xmm0); 234 __ fld_s(Address(rsp, 0)); 235 __ addptr(rsp, wordSize); 236 } else if (state == dtos && UseSSE >= 2) { 237 __ subptr(rsp, 2*wordSize); 238 __ movdbl(Address(rsp, 0), xmm0); 239 __ fld_d(Address(rsp, 0)); 240 __ addptr(rsp, 2*wordSize); 241 } 242 243 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter"); 244 245 // The stack is not extended by deopt but we must NULL last_sp as this 246 // entry is like a "return". 247 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 248 __ restore_bcp(); 249 __ restore_locals(); 250 // handle exceptions 251 { Label L; 252 const Register thread = rcx; 253 __ get_thread(thread); 254 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 255 __ jcc(Assembler::zero, L); 256 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception)); 257 __ should_not_reach_here(); 258 __ bind(L); 259 } 260 __ dispatch_next(state, step); 261 return entry; 262 } 263 264 265 int AbstractInterpreter::BasicType_as_index(BasicType type) { 266 int i = 0; 267 switch (type) { 268 case T_BOOLEAN: i = 0; break; 269 case T_CHAR : i = 1; break; 270 case T_BYTE : i = 2; break; 271 case T_SHORT : i = 3; break; 272 case T_INT : // fall through 273 case T_LONG : // fall through 274 case T_VOID : i = 4; break; 275 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE 276 case T_DOUBLE : i = 6; break; 277 case T_OBJECT : // fall through 278 case T_ARRAY : i = 7; break; 279 default : ShouldNotReachHere(); 280 } 281 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds"); 282 return i; 283 } 284 285 286 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) { 287 address entry = __ pc(); 288 switch (type) { 289 case T_BOOLEAN: __ c2bool(rax); break; 290 case T_CHAR : __ andptr(rax, 0xFFFF); break; 291 case T_BYTE : __ sign_extend_byte (rax); break; 292 case T_SHORT : __ sign_extend_short(rax); break; 293 case T_INT : /* nothing to do */ break; 294 case T_DOUBLE : 295 case T_FLOAT : 296 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); 297 __ pop(t); // remove return address first 298 // Must return a result for interpreter or compiler. In SSE 299 // mode, results are returned in xmm0 and the FPU stack must 300 // be empty. 301 if (type == T_FLOAT && UseSSE >= 1) { 302 // Load ST0 303 __ fld_d(Address(rsp, 0)); 304 // Store as float and empty fpu stack 305 __ fstp_s(Address(rsp, 0)); 306 // and reload 307 __ movflt(xmm0, Address(rsp, 0)); 308 } else if (type == T_DOUBLE && UseSSE >= 2 ) { 309 __ movdbl(xmm0, Address(rsp, 0)); 310 } else { 311 // restore ST0 312 __ fld_d(Address(rsp, 0)); 313 } 314 // and pop the temp 315 __ addptr(rsp, 2 * wordSize); 316 __ push(t); // restore return address 317 } 318 break; 319 case T_OBJECT : 320 // retrieve result from frame 321 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize)); 322 // and verify it 323 __ verify_oop(rax); 324 break; 325 default : ShouldNotReachHere(); 326 } 327 __ ret(0); // return from result handler 328 return entry; 329 } 330 331 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) { 332 address entry = __ pc(); 333 __ push(state); 334 __ call_VM(noreg, runtime_entry); 335 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 336 return entry; 337 } 338 339 340 // Helpers for commoning out cases in the various type of method entries. 341 // 342 343 // increment invocation count & check for overflow 344 // 345 // Note: checking for negative value instead of overflow 346 // so we have a 'sticky' overflow test 347 // 348 // rbx,: method 349 // rcx: invocation counter 350 // 351 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { 352 Label done; 353 // Note: In tiered we increment either counters in MethodCounters* or in MDO 354 // depending if we're profiling or not. 355 if (TieredCompilation) { 356 int increment = InvocationCounter::count_increment; 357 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift; 358 Label no_mdo; 359 if (ProfileInterpreter) { 360 // Are we profiling? 361 __ movptr(rax, Address(rbx, Method::method_data_offset())); 362 __ testptr(rax, rax); 363 __ jccb(Assembler::zero, no_mdo); 364 // Increment counter in the MDO 365 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) + 366 in_bytes(InvocationCounter::counter_offset())); 367 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow); 368 __ jmp(done); 369 } 370 __ bind(no_mdo); 371 // Increment counter in MethodCounters 372 const Address invocation_counter(rax, 373 MethodCounters::invocation_counter_offset() + 374 InvocationCounter::counter_offset()); 375 376 __ get_method_counters(rbx, rax, done); 377 __ increment_mask_and_jump(invocation_counter, increment, mask, 378 rcx, false, Assembler::zero, overflow); 379 __ bind(done); 380 } else { 381 const Address backedge_counter (rax, 382 MethodCounters::backedge_counter_offset() + 383 InvocationCounter::counter_offset()); 384 const Address invocation_counter(rax, 385 MethodCounters::invocation_counter_offset() + 386 InvocationCounter::counter_offset()); 387 388 __ get_method_counters(rbx, rax, done); 389 390 if (ProfileInterpreter) { 391 __ incrementl(Address(rax, 392 MethodCounters::interpreter_invocation_counter_offset())); 393 } 394 395 // Update standard invocation counters 396 __ movl(rcx, invocation_counter); 397 __ incrementl(rcx, InvocationCounter::count_increment); 398 __ movl(invocation_counter, rcx); // save invocation count 399 400 __ movl(rax, backedge_counter); // load backedge counter 401 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits 402 403 __ addl(rcx, rax); // add both counters 404 405 // profile_method is non-null only for interpreted method so 406 // profile_method != NULL == !native_call 407 // BytecodeInterpreter only calls for native so code is elided. 408 409 if (ProfileInterpreter && profile_method != NULL) { 410 // Test to see if we should create a method data oop 411 __ cmp32(rcx, 412 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit)); 413 __ jcc(Assembler::less, *profile_method_continue); 414 415 // if no method data exists, go to profile_method 416 __ test_method_data_pointer(rax, *profile_method); 417 } 418 419 __ cmp32(rcx, 420 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit)); 421 __ jcc(Assembler::aboveEqual, *overflow); 422 __ bind(done); 423 } 424 } 425 426 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) { 427 428 // Asm interpreter on entry 429 // rdi - locals 430 // rsi - bcp 431 // rbx, - method 432 // rdx - cpool 433 // rbp, - interpreter frame 434 435 // C++ interpreter on entry 436 // rsi - new interpreter state pointer 437 // rbp - interpreter frame pointer 438 // rbx - method 439 440 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] 441 // rbx, - method 442 // rcx - rcvr (assuming there is one) 443 // top of stack return address of interpreter caller 444 // rsp - sender_sp 445 446 // C++ interpreter only 447 // rsi - previous interpreter state pointer 448 449 // InterpreterRuntime::frequency_counter_overflow takes one argument 450 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp). 451 // The call returns the address of the verified entry point for the method or NULL 452 // if the compilation did not complete (either went background or bailed out). 453 __ movptr(rax, (intptr_t)false); 454 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax); 455 456 __ movptr(rbx, Address(rbp, method_offset)); // restore Method* 457 458 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame 459 // and jump to the interpreted entry. 460 __ jmp(*do_continue, relocInfo::none); 461 462 } 463 464 void InterpreterGenerator::generate_stack_overflow_check(void) { 465 // see if we've got enough room on the stack for locals plus overhead. 466 // the expression stack grows down incrementally, so the normal guard 467 // page mechanism will work for that. 468 // 469 // Registers live on entry: 470 // 471 // Asm interpreter 472 // rdx: number of additional locals this frame needs (what we must check) 473 // rbx,: Method* 474 475 // destroyed on exit 476 // rax, 477 478 // NOTE: since the additional locals are also always pushed (wasn't obvious in 479 // generate_method_entry) so the guard should work for them too. 480 // 481 482 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp 483 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 484 485 // total overhead size: entry_size + (saved rbp, thru expr stack bottom). 486 // be sure to change this if you add/subtract anything to/from the overhead area 487 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size; 488 489 const int page_size = os::vm_page_size(); 490 491 Label after_frame_check; 492 493 // see if the frame is greater than one page in size. If so, 494 // then we need to verify there is enough stack space remaining 495 // for the additional locals. 496 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize); 497 __ jcc(Assembler::belowEqual, after_frame_check); 498 499 // compute rsp as if this were going to be the last frame on 500 // the stack before the red zone 501 502 Label after_frame_check_pop; 503 504 __ push(rsi); 505 506 const Register thread = rsi; 507 508 __ get_thread(thread); 509 510 const Address stack_base(thread, Thread::stack_base_offset()); 511 const Address stack_size(thread, Thread::stack_size_offset()); 512 513 // locals + overhead, in bytes 514 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size)); 515 516 #ifdef ASSERT 517 Label stack_base_okay, stack_size_okay; 518 // verify that thread stack base is non-zero 519 __ cmpptr(stack_base, (int32_t)NULL_WORD); 520 __ jcc(Assembler::notEqual, stack_base_okay); 521 __ stop("stack base is zero"); 522 __ bind(stack_base_okay); 523 // verify that thread stack size is non-zero 524 __ cmpptr(stack_size, 0); 525 __ jcc(Assembler::notEqual, stack_size_okay); 526 __ stop("stack size is zero"); 527 __ bind(stack_size_okay); 528 #endif 529 530 // Add stack base to locals and subtract stack size 531 __ addptr(rax, stack_base); 532 __ subptr(rax, stack_size); 533 534 // Use the maximum number of pages we might bang. 535 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages : 536 (StackRedPages+StackYellowPages); 537 __ addptr(rax, max_pages * page_size); 538 539 // check against the current stack bottom 540 __ cmpptr(rsp, rax); 541 __ jcc(Assembler::above, after_frame_check_pop); 542 543 __ pop(rsi); // get saved bcp / (c++ prev state ). 544 545 // Restore sender's sp as SP. This is necessary if the sender's 546 // frame is an extended compiled frame (see gen_c2i_adapter()) 547 // and safer anyway in case of JSR292 adaptations. 548 549 __ pop(rax); // return address must be moved if SP is changed 550 __ mov(rsp, rsi); 551 __ push(rax); 552 553 // Note: the restored frame is not necessarily interpreted. 554 // Use the shared runtime version of the StackOverflowError. 555 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated"); 556 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry())); 557 // all done with frame size check 558 __ bind(after_frame_check_pop); 559 __ pop(rsi); 560 561 __ bind(after_frame_check); 562 } 563 564 // Allocate monitor and lock method (asm interpreter) 565 // rbx, - Method* 566 // 567 void InterpreterGenerator::lock_method(void) { 568 // synchronize method 569 const Address access_flags (rbx, Method::access_flags_offset()); 570 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 571 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 572 573 #ifdef ASSERT 574 { Label L; 575 __ movl(rax, access_flags); 576 __ testl(rax, JVM_ACC_SYNCHRONIZED); 577 __ jcc(Assembler::notZero, L); 578 __ stop("method doesn't need synchronization"); 579 __ bind(L); 580 } 581 #endif // ASSERT 582 // get synchronization object 583 { Label done; 584 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 585 __ movl(rax, access_flags); 586 __ testl(rax, JVM_ACC_STATIC); 587 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case) 588 __ jcc(Assembler::zero, done); 589 __ movptr(rax, Address(rbx, Method::const_offset())); 590 __ movptr(rax, Address(rax, ConstMethod::constants_offset())); 591 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes())); 592 __ movptr(rax, Address(rax, mirror_offset)); 593 __ bind(done); 594 } 595 // add space for monitor & lock 596 __ subptr(rsp, entry_size); // add space for a monitor entry 597 __ movptr(monitor_block_top, rsp); // set new monitor block top 598 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object 599 __ mov(rdx, rsp); // object address 600 __ lock_object(rdx); 601 } 602 603 // 604 // Generate a fixed interpreter frame. This is identical setup for interpreted methods 605 // and for native methods hence the shared code. 606 607 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { 608 // initialize fixed part of activation frame 609 __ push(rax); // save return address 610 __ enter(); // save old & set new rbp, 611 612 613 __ push(rsi); // set sender sp 614 __ push((int32_t)NULL_WORD); // leave last_sp as null 615 __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod* 616 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase 617 __ push(rbx); // save Method* 618 if (ProfileInterpreter) { 619 Label method_data_continue; 620 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset()))); 621 __ testptr(rdx, rdx); 622 __ jcc(Assembler::zero, method_data_continue); 623 __ addptr(rdx, in_bytes(MethodData::data_offset())); 624 __ bind(method_data_continue); 625 __ push(rdx); // set the mdp (method data pointer) 626 } else { 627 __ push(0); 628 } 629 630 __ movptr(rdx, Address(rbx, Method::const_offset())); 631 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset())); 632 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes())); 633 __ push(rdx); // set constant pool cache 634 __ push(rdi); // set locals pointer 635 if (native_call) { 636 __ push(0); // no bcp 637 } else { 638 __ push(rsi); // set bcp 639 } 640 __ push(0); // reserve word for pointer to expression stack bottom 641 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom 642 } 643 644 // End of helpers 645 646 // 647 // Various method entries 648 //------------------------------------------------------------------------------------------------------------------------ 649 // 650 // 651 652 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry 653 654 address InterpreterGenerator::generate_accessor_entry(void) { 655 656 // rbx,: Method* 657 // rcx: receiver (preserve for slow entry into asm interpreter) 658 659 // rsi: senderSP must preserved for slow path, set SP to it on fast path 660 661 address entry_point = __ pc(); 662 Label xreturn_path; 663 664 // do fastpath for resolved accessor methods 665 if (UseFastAccessorMethods) { 666 Label slow_path; 667 // If we need a safepoint check, generate full interpreter entry. 668 ExternalAddress state(SafepointSynchronize::address_of_state()); 669 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 670 SafepointSynchronize::_not_synchronized); 671 672 __ jcc(Assembler::notEqual, slow_path); 673 // ASM/C++ Interpreter 674 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1 675 // Note: We can only use this code if the getfield has been resolved 676 // and if we don't have a null-pointer exception => check for 677 // these conditions first and use slow path if necessary. 678 // rbx,: method 679 // rcx: receiver 680 __ movptr(rax, Address(rsp, wordSize)); 681 682 // check if local 0 != NULL and read field 683 __ testptr(rax, rax); 684 __ jcc(Assembler::zero, slow_path); 685 686 // read first instruction word and extract bytecode @ 1 and index @ 2 687 __ movptr(rdx, Address(rbx, Method::const_offset())); 688 __ movptr(rdi, Address(rdx, ConstMethod::constants_offset())); 689 __ movl(rdx, Address(rdx, ConstMethod::codes_offset())); 690 // Shift codes right to get the index on the right. 691 // The bytecode fetched looks like <index><0xb4><0x2a> 692 __ shrl(rdx, 2*BitsPerByte); 693 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); 694 __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes())); 695 696 // rax,: local 0 697 // rbx,: method 698 // rcx: receiver - do not destroy since it is needed for slow path! 699 // rcx: scratch 700 // rdx: constant pool cache index 701 // rdi: constant pool cache 702 // rsi: sender sp 703 704 // check if getfield has been resolved and read constant pool cache entry 705 // check the validity of the cache entry by testing whether _indices field 706 // contains Bytecode::_getfield in b1 byte. 707 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below"); 708 __ movl(rcx, 709 Address(rdi, 710 rdx, 711 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset())); 712 __ shrl(rcx, 2*BitsPerByte); 713 __ andl(rcx, 0xFF); 714 __ cmpl(rcx, Bytecodes::_getfield); 715 __ jcc(Assembler::notEqual, slow_path); 716 717 // Note: constant pool entry is not valid before bytecode is resolved 718 __ movptr(rcx, 719 Address(rdi, 720 rdx, 721 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset())); 722 __ movl(rdx, 723 Address(rdi, 724 rdx, 725 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset())); 726 727 Label notByte, notShort, notChar; 728 const Address field_address (rax, rcx, Address::times_1); 729 730 // Need to differentiate between igetfield, agetfield, bgetfield etc. 731 // because they are different sizes. 732 // Use the type from the constant pool cache 733 __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift); 734 // Make sure we don't need to mask rdx after the above shift 735 ConstantPoolCacheEntry::verify_tos_state_shift(); 736 __ cmpl(rdx, btos); 737 __ jcc(Assembler::notEqual, notByte); 738 __ load_signed_byte(rax, field_address); 739 __ jmp(xreturn_path); 740 741 __ bind(notByte); 742 __ cmpl(rdx, stos); 743 __ jcc(Assembler::notEqual, notShort); 744 __ load_signed_short(rax, field_address); 745 __ jmp(xreturn_path); 746 747 __ bind(notShort); 748 __ cmpl(rdx, ctos); 749 __ jcc(Assembler::notEqual, notChar); 750 __ load_unsigned_short(rax, field_address); 751 __ jmp(xreturn_path); 752 753 __ bind(notChar); 754 #ifdef ASSERT 755 Label okay; 756 __ cmpl(rdx, atos); 757 __ jcc(Assembler::equal, okay); 758 __ cmpl(rdx, itos); 759 __ jcc(Assembler::equal, okay); 760 __ stop("what type is this?"); 761 __ bind(okay); 762 #endif // ASSERT 763 // All the rest are a 32 bit wordsize 764 // This is ok for now. Since fast accessors should be going away 765 __ movptr(rax, field_address); 766 767 __ bind(xreturn_path); 768 769 // _ireturn/_areturn 770 __ pop(rdi); // get return address 771 __ mov(rsp, rsi); // set sp to sender sp 772 __ jmp(rdi); 773 774 // generate a vanilla interpreter entry as the slow path 775 __ bind(slow_path); 776 777 (void) generate_normal_entry(false); 778 return entry_point; 779 } 780 return NULL; 781 782 } 783 784 // Method entry for java.lang.ref.Reference.get. 785 address InterpreterGenerator::generate_Reference_get_entry(void) { 786 #if INCLUDE_ALL_GCS 787 // Code: _aload_0, _getfield, _areturn 788 // parameter size = 1 789 // 790 // The code that gets generated by this routine is split into 2 parts: 791 // 1. The "intrinsified" code for G1 (or any SATB based GC), 792 // 2. The slow path - which is an expansion of the regular method entry. 793 // 794 // Notes:- 795 // * In the G1 code we do not check whether we need to block for 796 // a safepoint. If G1 is enabled then we must execute the specialized 797 // code for Reference.get (except when the Reference object is null) 798 // so that we can log the value in the referent field with an SATB 799 // update buffer. 800 // If the code for the getfield template is modified so that the 801 // G1 pre-barrier code is executed when the current method is 802 // Reference.get() then going through the normal method entry 803 // will be fine. 804 // * The G1 code below can, however, check the receiver object (the instance 805 // of java.lang.Reference) and jump to the slow path if null. If the 806 // Reference object is null then we obviously cannot fetch the referent 807 // and so we don't need to call the G1 pre-barrier. Thus we can use the 808 // regular method entry code to generate the NPE. 809 // 810 // This code is based on generate_accessor_enty. 811 812 // rbx,: Method* 813 // rcx: receiver (preserve for slow entry into asm interpreter) 814 815 // rsi: senderSP must preserved for slow path, set SP to it on fast path 816 817 address entry = __ pc(); 818 819 const int referent_offset = java_lang_ref_Reference::referent_offset; 820 guarantee(referent_offset > 0, "referent offset not initialized"); 821 822 if (UseG1GC) { 823 Label slow_path; 824 825 // Check if local 0 != NULL 826 // If the receiver is null then it is OK to jump to the slow path. 827 __ movptr(rax, Address(rsp, wordSize)); 828 __ testptr(rax, rax); 829 __ jcc(Assembler::zero, slow_path); 830 831 // rax: local 0 (must be preserved across the G1 barrier call) 832 // 833 // rbx: method (at this point it's scratch) 834 // rcx: receiver (at this point it's scratch) 835 // rdx: scratch 836 // rdi: scratch 837 // 838 // rsi: sender sp 839 840 // Preserve the sender sp in case the pre-barrier 841 // calls the runtime 842 __ push(rsi); 843 844 // Load the value of the referent field. 845 const Address field_address(rax, referent_offset); 846 __ movptr(rax, field_address); 847 848 // Generate the G1 pre-barrier code to log the value of 849 // the referent field in an SATB buffer. 850 __ get_thread(rcx); 851 __ g1_write_barrier_pre(noreg /* obj */, 852 rax /* pre_val */, 853 rcx /* thread */, 854 rbx /* tmp */, 855 true /* tosca_save */, 856 true /* expand_call */); 857 858 // _areturn 859 __ pop(rsi); // get sender sp 860 __ pop(rdi); // get return address 861 __ mov(rsp, rsi); // set sp to sender sp 862 __ jmp(rdi); 863 864 __ bind(slow_path); 865 (void) generate_normal_entry(false); 866 867 return entry; 868 } 869 #endif // INCLUDE_ALL_GCS 870 871 // If G1 is not enabled then attempt to go through the accessor entry point 872 // Reference.get is an accessor 873 return generate_accessor_entry(); 874 } 875 876 /** 877 * Method entry for static native methods: 878 * int java.util.zip.CRC32.update(int crc, int b) 879 */ 880 address InterpreterGenerator::generate_CRC32_update_entry() { 881 if (UseCRC32Intrinsics) { 882 address entry = __ pc(); 883 884 // rbx,: Method* 885 // rsi: senderSP must preserved for slow path, set SP to it on fast path 886 // rdx: scratch 887 // rdi: scratch 888 889 Label slow_path; 890 // If we need a safepoint check, generate full interpreter entry. 891 ExternalAddress state(SafepointSynchronize::address_of_state()); 892 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 893 SafepointSynchronize::_not_synchronized); 894 __ jcc(Assembler::notEqual, slow_path); 895 896 // We don't generate local frame and don't align stack because 897 // we call stub code and there is no safepoint on this path. 898 899 // Load parameters 900 const Register crc = rax; // crc 901 const Register val = rdx; // source java byte value 902 const Register tbl = rdi; // scratch 903 904 // Arguments are reversed on java expression stack 905 __ movl(val, Address(rsp, wordSize)); // byte value 906 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC 907 908 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr())); 909 __ notl(crc); // ~crc 910 __ update_byte_crc32(crc, val, tbl); 911 __ notl(crc); // ~crc 912 // result in rax 913 914 // _areturn 915 __ pop(rdi); // get return address 916 __ mov(rsp, rsi); // set sp to sender sp 917 __ jmp(rdi); 918 919 // generate a vanilla native entry as the slow path 920 __ bind(slow_path); 921 922 (void) generate_native_entry(false); 923 924 return entry; 925 } 926 return generate_native_entry(false); 927 } 928 929 /** 930 * Method entry for static native methods: 931 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len) 932 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) 933 */ 934 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { 935 if (UseCRC32Intrinsics) { 936 address entry = __ pc(); 937 938 // rbx,: Method* 939 // rsi: senderSP must preserved for slow path, set SP to it on fast path 940 // rdx: scratch 941 // rdi: scratch 942 943 Label slow_path; 944 // If we need a safepoint check, generate full interpreter entry. 945 ExternalAddress state(SafepointSynchronize::address_of_state()); 946 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 947 SafepointSynchronize::_not_synchronized); 948 __ jcc(Assembler::notEqual, slow_path); 949 950 // We don't generate local frame and don't align stack because 951 // we call stub code and there is no safepoint on this path. 952 953 // Load parameters 954 const Register crc = rax; // crc 955 const Register buf = rdx; // source java byte array address 956 const Register len = rdi; // length 957 958 // Arguments are reversed on java expression stack 959 __ movl(len, Address(rsp, wordSize)); // Length 960 // Calculate address of start element 961 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { 962 __ movptr(buf, Address(rsp, 3*wordSize)); // long buf 963 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset 964 __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC 965 } else { 966 __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array 967 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size 968 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset 969 __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC 970 } 971 972 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len); 973 // result in rax 974 975 // _areturn 976 __ pop(rdi); // get return address 977 __ mov(rsp, rsi); // set sp to sender sp 978 __ jmp(rdi); 979 980 // generate a vanilla native entry as the slow path 981 __ bind(slow_path); 982 983 (void) generate_native_entry(false); 984 985 return entry; 986 } 987 return generate_native_entry(false); 988 } 989 990 // 991 // Interpreter stub for calling a native method. (asm interpreter) 992 // This sets up a somewhat different looking stack for calling the native method 993 // than the typical interpreter frame setup. 994 // 995 996 address InterpreterGenerator::generate_native_entry(bool synchronized) { 997 // determine code generation flags 998 bool inc_counter = UseCompiler || CountCompiledCalls; 999 1000 // rbx,: Method* 1001 // rsi: sender sp 1002 // rsi: previous interpreter state (C++ interpreter) must preserve 1003 address entry_point = __ pc(); 1004 1005 const Address constMethod (rbx, Method::const_offset()); 1006 const Address access_flags (rbx, Method::access_flags_offset()); 1007 const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset()); 1008 1009 // get parameter size (always needed) 1010 __ movptr(rcx, constMethod); 1011 __ load_unsigned_short(rcx, size_of_parameters); 1012 1013 // native calls don't need the stack size check since they have no expression stack 1014 // and the arguments are already on the stack and we only add a handful of words 1015 // to the stack 1016 1017 // rbx,: Method* 1018 // rcx: size of parameters 1019 // rsi: sender sp 1020 1021 __ pop(rax); // get return address 1022 // for natives the size of locals is zero 1023 1024 // compute beginning of parameters (rdi) 1025 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize)); 1026 1027 1028 // add 2 zero-initialized slots for native calls 1029 // NULL result handler 1030 __ push((int32_t)NULL_WORD); 1031 // NULL oop temp (mirror or jni oop result) 1032 __ push((int32_t)NULL_WORD); 1033 1034 // initialize fixed part of activation frame 1035 generate_fixed_frame(true); 1036 1037 // make sure method is native & not abstract 1038 #ifdef ASSERT 1039 __ movl(rax, access_flags); 1040 { 1041 Label L; 1042 __ testl(rax, JVM_ACC_NATIVE); 1043 __ jcc(Assembler::notZero, L); 1044 __ stop("tried to execute non-native method as native"); 1045 __ bind(L); 1046 } 1047 { Label L; 1048 __ testl(rax, JVM_ACC_ABSTRACT); 1049 __ jcc(Assembler::zero, L); 1050 __ stop("tried to execute abstract method in interpreter"); 1051 __ bind(L); 1052 } 1053 #endif 1054 1055 // Since at this point in the method invocation the exception handler 1056 // would try to exit the monitor of synchronized methods which hasn't 1057 // been entered yet, we set the thread local variable 1058 // _do_not_unlock_if_synchronized to true. The remove_activation will 1059 // check this flag. 1060 1061 __ get_thread(rax); 1062 const Address do_not_unlock_if_synchronized(rax, 1063 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1064 __ movbool(do_not_unlock_if_synchronized, true); 1065 1066 // increment invocation count & check for overflow 1067 Label invocation_counter_overflow; 1068 if (inc_counter) { 1069 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); 1070 } 1071 1072 Label continue_after_compile; 1073 __ bind(continue_after_compile); 1074 1075 bang_stack_shadow_pages(true); 1076 1077 // reset the _do_not_unlock_if_synchronized flag 1078 __ get_thread(rax); 1079 __ movbool(do_not_unlock_if_synchronized, false); 1080 1081 // check for synchronized methods 1082 // Must happen AFTER invocation_counter check and stack overflow check, 1083 // so method is not locked if overflows. 1084 // 1085 if (synchronized) { 1086 lock_method(); 1087 } else { 1088 // no synchronization necessary 1089 #ifdef ASSERT 1090 { Label L; 1091 __ movl(rax, access_flags); 1092 __ testl(rax, JVM_ACC_SYNCHRONIZED); 1093 __ jcc(Assembler::zero, L); 1094 __ stop("method needs synchronization"); 1095 __ bind(L); 1096 } 1097 #endif 1098 } 1099 1100 // start execution 1101 #ifdef ASSERT 1102 { Label L; 1103 const Address monitor_block_top (rbp, 1104 frame::interpreter_frame_monitor_block_top_offset * wordSize); 1105 __ movptr(rax, monitor_block_top); 1106 __ cmpptr(rax, rsp); 1107 __ jcc(Assembler::equal, L); 1108 __ stop("broken stack frame setup in interpreter"); 1109 __ bind(L); 1110 } 1111 #endif 1112 1113 // jvmti/dtrace support 1114 __ notify_method_entry(); 1115 1116 // work registers 1117 const Register method = rbx; 1118 const Register thread = rdi; 1119 const Register t = rcx; 1120 1121 // allocate space for parameters 1122 __ get_method(method); 1123 __ movptr(t, Address(method, Method::const_offset())); 1124 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset())); 1125 1126 __ shlptr(t, Interpreter::logStackElementSize); 1127 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror 1128 __ subptr(rsp, t); 1129 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics 1130 1131 // get signature handler 1132 { Label L; 1133 __ movptr(t, Address(method, Method::signature_handler_offset())); 1134 __ testptr(t, t); 1135 __ jcc(Assembler::notZero, L); 1136 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method); 1137 __ get_method(method); 1138 __ movptr(t, Address(method, Method::signature_handler_offset())); 1139 __ bind(L); 1140 } 1141 1142 // call signature handler 1143 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code"); 1144 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code"); 1145 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code"); 1146 // The generated handlers do not touch RBX (the method oop). 1147 // However, large signatures cannot be cached and are generated 1148 // each time here. The slow-path generator will blow RBX 1149 // sometime, so we must reload it after the call. 1150 __ call(t); 1151 __ get_method(method); // slow path call blows RBX on DevStudio 5.0 1152 1153 // result handler is in rax, 1154 // set result handler 1155 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax); 1156 1157 // pass mirror handle if static call 1158 { Label L; 1159 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 1160 __ movl(t, Address(method, Method::access_flags_offset())); 1161 __ testl(t, JVM_ACC_STATIC); 1162 __ jcc(Assembler::zero, L); 1163 // get mirror 1164 __ movptr(t, Address(method, Method:: const_offset())); 1165 __ movptr(t, Address(t, ConstMethod::constants_offset())); 1166 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes())); 1167 __ movptr(t, Address(t, mirror_offset)); 1168 // copy mirror into activation frame 1169 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t); 1170 // pass handle to mirror 1171 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize)); 1172 __ movptr(Address(rsp, wordSize), t); 1173 __ bind(L); 1174 } 1175 1176 // get native function entry point 1177 { Label L; 1178 __ movptr(rax, Address(method, Method::native_function_offset())); 1179 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry()); 1180 __ cmpptr(rax, unsatisfied.addr()); 1181 __ jcc(Assembler::notEqual, L); 1182 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method); 1183 __ get_method(method); 1184 __ movptr(rax, Address(method, Method::native_function_offset())); 1185 __ bind(L); 1186 } 1187 1188 // pass JNIEnv 1189 __ get_thread(thread); 1190 __ lea(t, Address(thread, JavaThread::jni_environment_offset())); 1191 __ movptr(Address(rsp, 0), t); 1192 1193 // set_last_Java_frame_before_call 1194 // It is enough that the pc() 1195 // points into the right code segment. It does not have to be the correct return pc. 1196 __ set_last_Java_frame(thread, noreg, rbp, __ pc()); 1197 1198 // change thread state 1199 #ifdef ASSERT 1200 { Label L; 1201 __ movl(t, Address(thread, JavaThread::thread_state_offset())); 1202 __ cmpl(t, _thread_in_Java); 1203 __ jcc(Assembler::equal, L); 1204 __ stop("Wrong thread state in native stub"); 1205 __ bind(L); 1206 } 1207 #endif 1208 1209 // Change state to native 1210 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native); 1211 __ call(rax); 1212 1213 // result potentially in rdx:rax or ST0 1214 1215 // Verify or restore cpu control state after JNI call 1216 __ restore_cpu_control_state_after_jni(); 1217 1218 // save potential result in ST(0) & rdx:rax 1219 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 - 1220 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers) 1221 // It is safe to do this push because state is _thread_in_native and return address will be found 1222 // via _last_native_pc and not via _last_jave_sp 1223 1224 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result. 1225 // If the order changes or anything else is added to the stack the code in 1226 // interpreter_frame_result will have to be changed. 1227 1228 { Label L; 1229 Label push_double; 1230 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT)); 1231 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE)); 1232 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize), 1233 float_handler.addr()); 1234 __ jcc(Assembler::equal, push_double); 1235 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize), 1236 double_handler.addr()); 1237 __ jcc(Assembler::notEqual, L); 1238 __ bind(push_double); 1239 __ push(dtos); 1240 __ bind(L); 1241 } 1242 __ push(ltos); 1243 1244 // change thread state 1245 __ get_thread(thread); 1246 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans); 1247 if(os::is_MP()) { 1248 if (UseMembar) { 1249 // Force this write out before the read below 1250 __ membar(Assembler::Membar_mask_bits( 1251 Assembler::LoadLoad | Assembler::LoadStore | 1252 Assembler::StoreLoad | Assembler::StoreStore)); 1253 } else { 1254 // Write serialization page so VM thread can do a pseudo remote membar. 1255 // We use the current thread pointer to calculate a thread specific 1256 // offset to write to within the page. This minimizes bus traffic 1257 // due to cache line collision. 1258 __ serialize_memory(thread, rcx); 1259 } 1260 } 1261 1262 if (AlwaysRestoreFPU) { 1263 // Make sure the control word is correct. 1264 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); 1265 } 1266 1267 // check for safepoint operation in progress and/or pending suspend requests 1268 { Label Continue; 1269 1270 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 1271 SafepointSynchronize::_not_synchronized); 1272 1273 Label L; 1274 __ jcc(Assembler::notEqual, L); 1275 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0); 1276 __ jcc(Assembler::equal, Continue); 1277 __ bind(L); 1278 1279 // Don't use call_VM as it will see a possible pending exception and forward it 1280 // and never return here preventing us from clearing _last_native_pc down below. 1281 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are 1282 // preserved and correspond to the bcp/locals pointers. So we do a runtime call 1283 // by hand. 1284 // 1285 __ push(thread); 1286 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, 1287 JavaThread::check_special_condition_for_native_trans))); 1288 __ increment(rsp, wordSize); 1289 __ get_thread(thread); 1290 1291 __ bind(Continue); 1292 } 1293 1294 // change thread state 1295 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java); 1296 1297 __ reset_last_Java_frame(thread, true, true); 1298 1299 // reset handle block 1300 __ movptr(t, Address(thread, JavaThread::active_handles_offset())); 1301 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD); 1302 1303 // If result was an oop then unbox and save it in the frame 1304 { Label L; 1305 Label no_oop, store_result; 1306 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT)); 1307 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), 1308 handler.addr()); 1309 __ jcc(Assembler::notEqual, no_oop); 1310 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD); 1311 __ pop(ltos); 1312 __ testptr(rax, rax); 1313 __ jcc(Assembler::zero, store_result); 1314 // unbox 1315 __ movptr(rax, Address(rax, 0)); 1316 __ bind(store_result); 1317 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax); 1318 // keep stack depth as expected by pushing oop which will eventually be discarded 1319 __ push(ltos); 1320 __ bind(no_oop); 1321 } 1322 1323 { 1324 Label no_reguard; 1325 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled); 1326 __ jcc(Assembler::notEqual, no_reguard); 1327 1328 __ pusha(); 1329 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); 1330 __ popa(); 1331 1332 __ bind(no_reguard); 1333 } 1334 1335 // restore rsi to have legal interpreter frame, 1336 // i.e., bci == 0 <=> rsi == code_base() 1337 // Can't call_VM until bcp is within reasonable. 1338 __ get_method(method); // method is junk from thread_in_native to now. 1339 __ movptr(rsi, Address(method,Method::const_offset())); // get ConstMethod* 1340 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase 1341 1342 // handle exceptions (exception handling will handle unlocking!) 1343 { Label L; 1344 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 1345 __ jcc(Assembler::zero, L); 1346 // Note: At some point we may want to unify this with the code used in call_VM_base(); 1347 // i.e., we should use the StubRoutines::forward_exception code. For now this 1348 // doesn't work here because the rsp is not correctly set at this point. 1349 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception)); 1350 __ should_not_reach_here(); 1351 __ bind(L); 1352 } 1353 1354 // do unlocking if necessary 1355 { Label L; 1356 __ movl(t, Address(method, Method::access_flags_offset())); 1357 __ testl(t, JVM_ACC_SYNCHRONIZED); 1358 __ jcc(Assembler::zero, L); 1359 // the code below should be shared with interpreter macro assembler implementation 1360 { Label unlock; 1361 // BasicObjectLock will be first in list, since this is a synchronized method. However, need 1362 // to check that the object has not been unlocked by an explicit monitorexit bytecode. 1363 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock)); 1364 1365 __ lea(rdx, monitor); // address of first monitor 1366 1367 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); 1368 __ testptr(t, t); 1369 __ jcc(Assembler::notZero, unlock); 1370 1371 // Entry already unlocked, need to throw exception 1372 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1373 __ should_not_reach_here(); 1374 1375 __ bind(unlock); 1376 __ unlock_object(rdx); 1377 } 1378 __ bind(L); 1379 } 1380 1381 // jvmti/dtrace support 1382 // Note: This must happen _after_ handling/throwing any exceptions since 1383 // the exception handler code notifies the runtime of method exits 1384 // too. If this happens before, method entry/exit notifications are 1385 // not properly paired (was bug - gri 11/22/99). 1386 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); 1387 1388 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result 1389 __ pop(ltos); 1390 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize)); 1391 __ call(t); 1392 1393 // remove activation 1394 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp 1395 __ leave(); // remove frame anchor 1396 __ pop(rdi); // get return address 1397 __ mov(rsp, t); // set sp to sender sp 1398 __ jmp(rdi); 1399 1400 if (inc_counter) { 1401 // Handle overflow of counter and compile method 1402 __ bind(invocation_counter_overflow); 1403 generate_counter_overflow(&continue_after_compile); 1404 } 1405 1406 return entry_point; 1407 } 1408 1409 // 1410 // Generic interpreted method entry to (asm) interpreter 1411 // 1412 address InterpreterGenerator::generate_normal_entry(bool synchronized) { 1413 // determine code generation flags 1414 bool inc_counter = UseCompiler || CountCompiledCalls; 1415 1416 // rbx,: Method* 1417 // rsi: sender sp 1418 address entry_point = __ pc(); 1419 1420 const Address constMethod (rbx, Method::const_offset()); 1421 const Address access_flags (rbx, Method::access_flags_offset()); 1422 const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset()); 1423 const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset()); 1424 1425 // get parameter size (always needed) 1426 __ movptr(rdx, constMethod); 1427 __ load_unsigned_short(rcx, size_of_parameters); 1428 1429 // rbx,: Method* 1430 // rcx: size of parameters 1431 1432 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i ) 1433 1434 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words 1435 __ subl(rdx, rcx); // rdx = no. of additional locals 1436 1437 // see if we've got enough room on the stack for locals plus overhead. 1438 generate_stack_overflow_check(); 1439 1440 // get return address 1441 __ pop(rax); 1442 1443 // compute beginning of parameters (rdi) 1444 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize)); 1445 1446 // rdx - # of additional locals 1447 // allocate space for locals 1448 // explicitly initialize locals 1449 { 1450 Label exit, loop; 1451 __ testl(rdx, rdx); 1452 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0 1453 __ bind(loop); 1454 __ push((int32_t)NULL_WORD); // initialize local variables 1455 __ decrement(rdx); // until everything initialized 1456 __ jcc(Assembler::greater, loop); 1457 __ bind(exit); 1458 } 1459 1460 // initialize fixed part of activation frame 1461 generate_fixed_frame(false); 1462 1463 // make sure method is not native & not abstract 1464 #ifdef ASSERT 1465 __ movl(rax, access_flags); 1466 { 1467 Label L; 1468 __ testl(rax, JVM_ACC_NATIVE); 1469 __ jcc(Assembler::zero, L); 1470 __ stop("tried to execute native method as non-native"); 1471 __ bind(L); 1472 } 1473 { Label L; 1474 __ testl(rax, JVM_ACC_ABSTRACT); 1475 __ jcc(Assembler::zero, L); 1476 __ stop("tried to execute abstract method in interpreter"); 1477 __ bind(L); 1478 } 1479 #endif 1480 1481 // Since at this point in the method invocation the exception handler 1482 // would try to exit the monitor of synchronized methods which hasn't 1483 // been entered yet, we set the thread local variable 1484 // _do_not_unlock_if_synchronized to true. The remove_activation will 1485 // check this flag. 1486 1487 __ get_thread(rax); 1488 const Address do_not_unlock_if_synchronized(rax, 1489 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1490 __ movbool(do_not_unlock_if_synchronized, true); 1491 1492 // increment invocation count & check for overflow 1493 Label invocation_counter_overflow; 1494 Label profile_method; 1495 Label profile_method_continue; 1496 if (inc_counter) { 1497 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue); 1498 if (ProfileInterpreter) { 1499 __ bind(profile_method_continue); 1500 } 1501 } 1502 Label continue_after_compile; 1503 __ bind(continue_after_compile); 1504 1505 bang_stack_shadow_pages(false); 1506 1507 // reset the _do_not_unlock_if_synchronized flag 1508 __ get_thread(rax); 1509 __ movbool(do_not_unlock_if_synchronized, false); 1510 1511 // check for synchronized methods 1512 // Must happen AFTER invocation_counter check and stack overflow check, 1513 // so method is not locked if overflows. 1514 // 1515 if (synchronized) { 1516 // Allocate monitor and lock method 1517 lock_method(); 1518 } else { 1519 // no synchronization necessary 1520 #ifdef ASSERT 1521 { Label L; 1522 __ movl(rax, access_flags); 1523 __ testl(rax, JVM_ACC_SYNCHRONIZED); 1524 __ jcc(Assembler::zero, L); 1525 __ stop("method needs synchronization"); 1526 __ bind(L); 1527 } 1528 #endif 1529 } 1530 1531 // start execution 1532 #ifdef ASSERT 1533 { Label L; 1534 const Address monitor_block_top (rbp, 1535 frame::interpreter_frame_monitor_block_top_offset * wordSize); 1536 __ movptr(rax, monitor_block_top); 1537 __ cmpptr(rax, rsp); 1538 __ jcc(Assembler::equal, L); 1539 __ stop("broken stack frame setup in interpreter"); 1540 __ bind(L); 1541 } 1542 #endif 1543 1544 // jvmti support 1545 __ notify_method_entry(); 1546 1547 __ dispatch_next(vtos); 1548 1549 // invocation counter overflow 1550 if (inc_counter) { 1551 if (ProfileInterpreter) { 1552 // We have decided to profile this method in the interpreter 1553 __ bind(profile_method); 1554 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1555 __ set_method_data_pointer_for_bcp(); 1556 __ get_method(rbx); 1557 __ jmp(profile_method_continue); 1558 } 1559 // Handle overflow of counter and compile method 1560 __ bind(invocation_counter_overflow); 1561 generate_counter_overflow(&continue_after_compile); 1562 } 1563 1564 return entry_point; 1565 } 1566 1567 //------------------------------------------------------------------------------------------------------------------------ 1568 // Entry points 1569 // 1570 // Here we generate the various kind of entries into the interpreter. 1571 // The two main entry type are generic bytecode methods and native call method. 1572 // These both come in synchronized and non-synchronized versions but the 1573 // frame layout they create is very similar. The other method entry 1574 // types are really just special purpose entries that are really entry 1575 // and interpretation all in one. These are for trivial methods like 1576 // accessor, empty, or special math methods. 1577 // 1578 // When control flow reaches any of the entry types for the interpreter 1579 // the following holds -> 1580 // 1581 // Arguments: 1582 // 1583 // rbx,: Method* 1584 // rcx: receiver 1585 // 1586 // 1587 // Stack layout immediately at entry 1588 // 1589 // [ return address ] <--- rsp 1590 // [ parameter n ] 1591 // ... 1592 // [ parameter 1 ] 1593 // [ expression stack ] (caller's java expression stack) 1594 1595 // Assuming that we don't go to one of the trivial specialized 1596 // entries the stack will look like below when we are ready to execute 1597 // the first bytecode (or call the native routine). The register usage 1598 // will be as the template based interpreter expects (see interpreter_x86.hpp). 1599 // 1600 // local variables follow incoming parameters immediately; i.e. 1601 // the return address is moved to the end of the locals). 1602 // 1603 // [ monitor entry ] <--- rsp 1604 // ... 1605 // [ monitor entry ] 1606 // [ expr. stack bottom ] 1607 // [ saved rsi ] 1608 // [ current rdi ] 1609 // [ Method* ] 1610 // [ saved rbp, ] <--- rbp, 1611 // [ return address ] 1612 // [ local variable m ] 1613 // ... 1614 // [ local variable 1 ] 1615 // [ parameter n ] 1616 // ... 1617 // [ parameter 1 ] <--- rdi 1618 1619 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) { 1620 // determine code generation flags 1621 bool synchronized = false; 1622 address entry_point = NULL; 1623 InterpreterGenerator* ig_this = (InterpreterGenerator*)this; 1624 1625 switch (kind) { 1626 case Interpreter::zerolocals : break; 1627 case Interpreter::zerolocals_synchronized: synchronized = true; break; 1628 case Interpreter::native : entry_point = ig_this->generate_native_entry(false); break; 1629 case Interpreter::native_synchronized : entry_point = ig_this->generate_native_entry(true); break; 1630 case Interpreter::empty : entry_point = ig_this->generate_empty_entry(); break; 1631 case Interpreter::accessor : entry_point = ig_this->generate_accessor_entry(); break; 1632 case Interpreter::abstract : entry_point = ig_this->generate_abstract_entry(); break; 1633 1634 case Interpreter::java_lang_math_sin : // fall thru 1635 case Interpreter::java_lang_math_cos : // fall thru 1636 case Interpreter::java_lang_math_tan : // fall thru 1637 case Interpreter::java_lang_math_abs : // fall thru 1638 case Interpreter::java_lang_math_log : // fall thru 1639 case Interpreter::java_lang_math_log10 : // fall thru 1640 case Interpreter::java_lang_math_sqrt : // fall thru 1641 case Interpreter::java_lang_math_pow : // fall thru 1642 case Interpreter::java_lang_math_exp : entry_point = ig_this->generate_math_entry(kind); break; 1643 case Interpreter::java_lang_ref_reference_get 1644 : entry_point = ig_this->generate_Reference_get_entry(); break; 1645 case Interpreter::java_util_zip_CRC32_update 1646 : entry_point = ig_this->generate_CRC32_update_entry(); break; 1647 case Interpreter::java_util_zip_CRC32_updateBytes 1648 : // fall thru 1649 case Interpreter::java_util_zip_CRC32_updateByteBuffer 1650 : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break; 1651 default: 1652 fatal(err_msg("unexpected method kind: %d", kind)); 1653 break; 1654 } 1655 1656 if (entry_point) return entry_point; 1657 1658 return ig_this->generate_normal_entry(synchronized); 1659 1660 } 1661 1662 // These should never be compiled since the interpreter will prefer 1663 // the compiled version to the intrinsic version. 1664 bool AbstractInterpreter::can_be_compiled(methodHandle m) { 1665 switch (method_kind(m)) { 1666 case Interpreter::java_lang_math_sin : // fall thru 1667 case Interpreter::java_lang_math_cos : // fall thru 1668 case Interpreter::java_lang_math_tan : // fall thru 1669 case Interpreter::java_lang_math_abs : // fall thru 1670 case Interpreter::java_lang_math_log : // fall thru 1671 case Interpreter::java_lang_math_log10 : // fall thru 1672 case Interpreter::java_lang_math_sqrt : // fall thru 1673 case Interpreter::java_lang_math_pow : // fall thru 1674 case Interpreter::java_lang_math_exp : 1675 return false; 1676 default: 1677 return true; 1678 } 1679 } 1680 1681 // How much stack a method activation needs in words. 1682 int AbstractInterpreter::size_top_interpreter_activation(Method* method) { 1683 1684 const int stub_code = 4; // see generate_call_stub 1685 // Save space for one monitor to get into the interpreted method in case 1686 // the method is synchronized 1687 int monitor_size = method->is_synchronized() ? 1688 1*frame::interpreter_frame_monitor_size() : 0; 1689 1690 // total overhead size: entry_size + (saved rbp, thru expr stack bottom). 1691 // be sure to change this if you add/subtract anything to/from the overhead area 1692 const int overhead_size = -frame::interpreter_frame_initial_sp_offset; 1693 1694 const int method_stack = (method->max_locals() + method->max_stack()) * 1695 Interpreter::stackElementWords; 1696 return overhead_size + method_stack + stub_code; 1697 } 1698 1699 // asm based interpreter deoptimization helpers 1700 1701 int AbstractInterpreter::layout_activation(Method* method, 1702 int tempcount, 1703 int popframe_extra_args, 1704 int moncount, 1705 int caller_actual_parameters, 1706 int callee_param_count, 1707 int callee_locals, 1708 frame* caller, 1709 frame* interpreter_frame, 1710 bool is_top_frame, 1711 bool is_bottom_frame) { 1712 // Note: This calculation must exactly parallel the frame setup 1713 // in AbstractInterpreterGenerator::generate_method_entry. 1714 // If interpreter_frame!=NULL, set up the method, locals, and monitors. 1715 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size, 1716 // as determined by a previous call to this method. 1717 // It is also guaranteed to be walkable even though it is in a skeletal state 1718 // NOTE: return size is in words not bytes 1719 1720 // fixed size of an interpreter frame: 1721 int max_locals = method->max_locals() * Interpreter::stackElementWords; 1722 int extra_locals = (method->max_locals() - method->size_of_parameters()) * 1723 Interpreter::stackElementWords; 1724 1725 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset; 1726 1727 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion) 1728 // Since the callee parameters already account for the callee's params we only need to account for 1729 // the extra locals. 1730 1731 1732 int size = overhead + 1733 ((callee_locals - callee_param_count)*Interpreter::stackElementWords) + 1734 (moncount*frame::interpreter_frame_monitor_size()) + 1735 tempcount*Interpreter::stackElementWords + popframe_extra_args; 1736 1737 if (interpreter_frame != NULL) { 1738 #ifdef ASSERT 1739 if (!EnableInvokeDynamic) 1740 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences? 1741 // Probably, since deoptimization doesn't work yet. 1742 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable"); 1743 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)"); 1744 #endif 1745 1746 interpreter_frame->interpreter_frame_set_method(method); 1747 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp 1748 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp) 1749 // and sender_sp is fp+8 1750 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1; 1751 1752 #ifdef ASSERT 1753 if (caller->is_interpreted_frame()) { 1754 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement"); 1755 } 1756 #endif 1757 1758 interpreter_frame->interpreter_frame_set_locals(locals); 1759 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin(); 1760 BasicObjectLock* monbot = montop - moncount; 1761 interpreter_frame->interpreter_frame_set_monitor_end(monbot); 1762 1763 // Set last_sp 1764 intptr_t* rsp = (intptr_t*) monbot - 1765 tempcount*Interpreter::stackElementWords - 1766 popframe_extra_args; 1767 interpreter_frame->interpreter_frame_set_last_sp(rsp); 1768 1769 // All frames but the initial (oldest) interpreter frame we fill in have a 1770 // value for sender_sp that allows walking the stack but isn't 1771 // truly correct. Correct the value here. 1772 1773 if (extra_locals != 0 && 1774 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) { 1775 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals); 1776 } 1777 *interpreter_frame->interpreter_frame_cache_addr() = 1778 method->constants()->cache(); 1779 } 1780 return size; 1781 } 1782 1783 1784 //------------------------------------------------------------------------------------------------------------------------ 1785 // Exceptions 1786 1787 void TemplateInterpreterGenerator::generate_throw_exception() { 1788 // Entry point in previous activation (i.e., if the caller was interpreted) 1789 Interpreter::_rethrow_exception_entry = __ pc(); 1790 const Register thread = rcx; 1791 1792 // Restore sp to interpreter_frame_last_sp even though we are going 1793 // to empty the expression stack for the exception processing. 1794 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 1795 // rax,: exception 1796 // rdx: return address/pc that threw exception 1797 __ restore_bcp(); // rsi points to call/send 1798 __ restore_locals(); 1799 1800 // Entry point for exceptions thrown within interpreter code 1801 Interpreter::_throw_exception_entry = __ pc(); 1802 // expression stack is undefined here 1803 // rax,: exception 1804 // rsi: exception bcp 1805 __ verify_oop(rax); 1806 1807 // expression stack must be empty before entering the VM in case of an exception 1808 __ empty_expression_stack(); 1809 __ empty_FPU_stack(); 1810 // find exception handler address and preserve exception oop 1811 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax); 1812 // rax,: exception handler entry point 1813 // rdx: preserved exception oop 1814 // rsi: bcp for exception handler 1815 __ push_ptr(rdx); // push exception which is now the only value on the stack 1816 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!) 1817 1818 // If the exception is not handled in the current frame the frame is removed and 1819 // the exception is rethrown (i.e. exception continuation is _rethrow_exception). 1820 // 1821 // Note: At this point the bci is still the bxi for the instruction which caused 1822 // the exception and the expression stack is empty. Thus, for any VM calls 1823 // at this point, GC will find a legal oop map (with empty expression stack). 1824 1825 // In current activation 1826 // tos: exception 1827 // rsi: exception bcp 1828 1829 // 1830 // JVMTI PopFrame support 1831 // 1832 1833 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 1834 __ empty_expression_stack(); 1835 __ empty_FPU_stack(); 1836 // Set the popframe_processing bit in pending_popframe_condition indicating that we are 1837 // currently handling popframe, so that call_VMs that may happen later do not trigger new 1838 // popframe handling cycles. 1839 __ get_thread(thread); 1840 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset())); 1841 __ orl(rdx, JavaThread::popframe_processing_bit); 1842 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx); 1843 1844 { 1845 // Check to see whether we are returning to a deoptimized frame. 1846 // (The PopFrame call ensures that the caller of the popped frame is 1847 // either interpreted or compiled and deoptimizes it if compiled.) 1848 // In this case, we can't call dispatch_next() after the frame is 1849 // popped, but instead must save the incoming arguments and restore 1850 // them after deoptimization has occurred. 1851 // 1852 // Note that we don't compare the return PC against the 1853 // deoptimization blob's unpack entry because of the presence of 1854 // adapter frames in C2. 1855 Label caller_not_deoptimized; 1856 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize)); 1857 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx); 1858 __ testl(rax, rax); 1859 __ jcc(Assembler::notZero, caller_not_deoptimized); 1860 1861 // Compute size of arguments for saving when returning to deoptimized caller 1862 __ get_method(rax); 1863 __ movptr(rax, Address(rax, Method::const_offset())); 1864 __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset())); 1865 __ shlptr(rax, Interpreter::logStackElementSize); 1866 __ restore_locals(); 1867 __ subptr(rdi, rax); 1868 __ addptr(rdi, wordSize); 1869 // Save these arguments 1870 __ get_thread(thread); 1871 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi); 1872 1873 __ remove_activation(vtos, rdx, 1874 /* throw_monitor_exception */ false, 1875 /* install_monitor_exception */ false, 1876 /* notify_jvmdi */ false); 1877 1878 // Inform deoptimization that it is responsible for restoring these arguments 1879 __ get_thread(thread); 1880 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit); 1881 1882 // Continue in deoptimization handler 1883 __ jmp(rdx); 1884 1885 __ bind(caller_not_deoptimized); 1886 } 1887 1888 __ remove_activation(vtos, rdx, 1889 /* throw_monitor_exception */ false, 1890 /* install_monitor_exception */ false, 1891 /* notify_jvmdi */ false); 1892 1893 // Finish with popframe handling 1894 // A previous I2C followed by a deoptimization might have moved the 1895 // outgoing arguments further up the stack. PopFrame expects the 1896 // mutations to those outgoing arguments to be preserved and other 1897 // constraints basically require this frame to look exactly as 1898 // though it had previously invoked an interpreted activation with 1899 // no space between the top of the expression stack (current 1900 // last_sp) and the top of stack. Rather than force deopt to 1901 // maintain this kind of invariant all the time we call a small 1902 // fixup routine to move the mutated arguments onto the top of our 1903 // expression stack if necessary. 1904 __ mov(rax, rsp); 1905 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 1906 __ get_thread(thread); 1907 // PC must point into interpreter here 1908 __ set_last_Java_frame(thread, noreg, rbp, __ pc()); 1909 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx); 1910 __ get_thread(thread); 1911 __ reset_last_Java_frame(thread, true, true); 1912 // Restore the last_sp and null it out 1913 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 1914 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 1915 1916 __ restore_bcp(); 1917 __ restore_locals(); 1918 // The method data pointer was incremented already during 1919 // call profiling. We have to restore the mdp for the current bcp. 1920 if (ProfileInterpreter) { 1921 __ set_method_data_pointer_for_bcp(); 1922 } 1923 1924 // Clear the popframe condition flag 1925 __ get_thread(thread); 1926 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive); 1927 1928 #if INCLUDE_JVMTI 1929 if (EnableInvokeDynamic) { 1930 Label L_done; 1931 const Register local0 = rdi; 1932 1933 __ cmpb(Address(rsi, 0), Bytecodes::_invokestatic); 1934 __ jcc(Assembler::notEqual, L_done); 1935 1936 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. 1937 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL. 1938 1939 __ get_method(rdx); 1940 __ movptr(rax, Address(local0, 0)); 1941 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rsi); 1942 1943 __ testptr(rax, rax); 1944 __ jcc(Assembler::zero, L_done); 1945 1946 __ movptr(Address(rbx, 0), rax); 1947 __ bind(L_done); 1948 } 1949 #endif // INCLUDE_JVMTI 1950 1951 __ dispatch_next(vtos); 1952 // end of PopFrame support 1953 1954 Interpreter::_remove_activation_entry = __ pc(); 1955 1956 // preserve exception over this code sequence 1957 __ pop_ptr(rax); 1958 __ get_thread(thread); 1959 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax); 1960 // remove the activation (without doing throws on illegalMonitorExceptions) 1961 __ remove_activation(vtos, rdx, false, true, false); 1962 // restore exception 1963 __ get_thread(thread); 1964 __ get_vm_result(rax, thread); 1965 1966 // Inbetween activations - previous activation type unknown yet 1967 // compute continuation point - the continuation point expects 1968 // the following registers set up: 1969 // 1970 // rax: exception 1971 // rdx: return address/pc that threw exception 1972 // rsp: expression stack of caller 1973 // rbp: rbp, of caller 1974 __ push(rax); // save exception 1975 __ push(rdx); // save return address 1976 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx); 1977 __ mov(rbx, rax); // save exception handler 1978 __ pop(rdx); // restore return address 1979 __ pop(rax); // restore exception 1980 // Note that an "issuing PC" is actually the next PC after the call 1981 __ jmp(rbx); // jump to exception handler of caller 1982 } 1983 1984 1985 // 1986 // JVMTI ForceEarlyReturn support 1987 // 1988 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 1989 address entry = __ pc(); 1990 const Register thread = rcx; 1991 1992 __ restore_bcp(); 1993 __ restore_locals(); 1994 __ empty_expression_stack(); 1995 __ empty_FPU_stack(); 1996 __ load_earlyret_value(state); 1997 1998 __ get_thread(thread); 1999 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset())); 2000 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset()); 2001 2002 // Clear the earlyret state 2003 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive); 2004 2005 __ remove_activation(state, rsi, 2006 false, /* throw_monitor_exception */ 2007 false, /* install_monitor_exception */ 2008 true); /* notify_jvmdi */ 2009 __ jmp(rsi); 2010 return entry; 2011 } // end of ForceEarlyReturn support 2012 2013 2014 //------------------------------------------------------------------------------------------------------------------------ 2015 // Helper for vtos entry point generation 2016 2017 void TemplateInterpreterGenerator::set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) { 2018 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 2019 Label L; 2020 fep = __ pc(); __ push(ftos); __ jmp(L); 2021 dep = __ pc(); __ push(dtos); __ jmp(L); 2022 lep = __ pc(); __ push(ltos); __ jmp(L); 2023 aep = __ pc(); __ push(atos); __ jmp(L); 2024 bep = cep = sep = // fall through 2025 iep = __ pc(); __ push(itos); // fall through 2026 vep = __ pc(); __ bind(L); // fall through 2027 generate_and_dispatch(t); 2028 } 2029 2030 //------------------------------------------------------------------------------------------------------------------------ 2031 // Generation of individual instructions 2032 2033 // helpers for generate_and_dispatch 2034 2035 2036 2037 InterpreterGenerator::InterpreterGenerator(StubQueue* code) 2038 : TemplateInterpreterGenerator(code) { 2039 generate_all(); // down here so it can be "virtual" 2040 } 2041 2042 //------------------------------------------------------------------------------------------------------------------------ 2043 2044 // Non-product code 2045 #ifndef PRODUCT 2046 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 2047 address entry = __ pc(); 2048 2049 // prepare expression stack 2050 __ pop(rcx); // pop return address so expression stack is 'pure' 2051 __ push(state); // save tosca 2052 2053 // pass tosca registers as arguments & call tracer 2054 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx); 2055 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state) 2056 __ pop(state); // restore tosca 2057 2058 // return 2059 __ jmp(rcx); 2060 2061 return entry; 2062 } 2063 2064 2065 void TemplateInterpreterGenerator::count_bytecode() { 2066 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value)); 2067 } 2068 2069 2070 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 2071 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()])); 2072 } 2073 2074 2075 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 2076 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx); 2077 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes); 2078 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); 2079 ExternalAddress table((address) BytecodePairHistogram::_counters); 2080 Address index(noreg, rbx, Address::times_4); 2081 __ incrementl(ArrayAddress(table, index)); 2082 } 2083 2084 2085 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 2086 // Call a little run-time stub to avoid blow-up for each bytecode. 2087 // The run-time runtime saves the right registers, depending on 2088 // the tosca in-state for the given template. 2089 assert(Interpreter::trace_code(t->tos_in()) != NULL, 2090 "entry must have been generated"); 2091 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in()))); 2092 } 2093 2094 2095 void TemplateInterpreterGenerator::stop_interpreter_at() { 2096 Label L; 2097 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value), 2098 StopInterpreterAt); 2099 __ jcc(Assembler::notEqual, L); 2100 __ int3(); 2101 __ bind(L); 2102 } 2103 #endif // !PRODUCT 2104 #endif // CC_INTERP