1 /* 2 * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "asm/assembler.hpp" 27 #include "interpreter/bytecodeHistogram.hpp" 28 #include "interpreter/interpreter.hpp" 29 #include "interpreter/interpreterGenerator.hpp" 30 #include "interpreter/interpreterRuntime.hpp" 31 #include "interpreter/templateTable.hpp" 32 #include "oops/arrayOop.hpp" 33 #include "oops/methodDataOop.hpp" 34 #include "oops/methodOop.hpp" 35 #include "oops/oop.inline.hpp" 36 #include "prims/jvmtiExport.hpp" 37 #include "prims/jvmtiThreadState.hpp" 38 #include "runtime/arguments.hpp" 39 #include "runtime/deoptimization.hpp" 40 #include "runtime/frame.inline.hpp" 41 #include "runtime/sharedRuntime.hpp" 42 #include "runtime/stubRoutines.hpp" 43 #include "runtime/synchronizer.hpp" 44 #include "runtime/timer.hpp" 45 #include "runtime/vframeArray.hpp" 46 #include "utilities/debug.hpp" 47 48 #define __ _masm-> 49 50 51 #ifndef CC_INTERP 52 const int method_offset = frame::interpreter_frame_method_offset * wordSize; 53 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize; 54 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize; 55 56 //------------------------------------------------------------------------------------------------------------------------ 57 58 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 59 address entry = __ pc(); 60 61 // Note: There should be a minimal interpreter frame set up when stack 62 // overflow occurs since we check explicitly for it now. 63 // 64 #ifdef ASSERT 65 { Label L; 66 __ lea(rax, Address(rbp, 67 frame::interpreter_frame_monitor_block_top_offset * wordSize)); 68 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp, 69 // (stack 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 exception 80 // happened 81 __ empty_expression_stack(); 82 __ empty_FPU_stack(); 83 // throw exception 84 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError)); 85 return entry; 86 } 87 88 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) { 89 address entry = __ pc(); 90 // expression stack must be empty before entering the VM if an exception happened 91 __ empty_expression_stack(); 92 __ empty_FPU_stack(); 93 // setup parameters 94 // ??? convention: expect aberrant index in register rbx, 95 __ lea(rax, ExternalAddress((address)name)); 96 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx); 97 return entry; 98 } 99 100 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 101 address entry = __ pc(); 102 // object is at TOS 103 __ pop(rax); 104 // expression stack must be empty before entering the VM if an exception 105 // happened 106 __ empty_expression_stack(); 107 __ empty_FPU_stack(); 108 __ call_VM(noreg, 109 CAST_FROM_FN_PTR(address, 110 InterpreterRuntime::throw_ClassCastException), 111 rax); 112 return entry; 113 } 114 115 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) { 116 assert(!pass_oop || message == NULL, "either oop or message but not both"); 117 address entry = __ pc(); 118 if (pass_oop) { 119 // object is at TOS 120 __ pop(rbx); 121 } 122 // expression stack must be empty before entering the VM if an exception happened 123 __ empty_expression_stack(); 124 __ empty_FPU_stack(); 125 // setup parameters 126 __ lea(rax, ExternalAddress((address)name)); 127 if (pass_oop) { 128 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx); 129 } else { 130 if (message != NULL) { 131 __ lea(rbx, ExternalAddress((address)message)); 132 } else { 133 __ movptr(rbx, NULL_WORD); 134 } 135 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx); 136 } 137 // throw exception 138 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); 139 return entry; 140 } 141 142 143 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { 144 address entry = __ pc(); 145 // NULL last_sp until next java call 146 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 147 __ dispatch_next(state); 148 return entry; 149 } 150 151 152 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) { 153 TosState incoming_state = state; 154 address entry = __ pc(); 155 156 #ifdef COMPILER2 157 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases 158 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) { 159 for (int i = 1; i < 8; i++) { 160 __ ffree(i); 161 } 162 } else if (UseSSE < 2) { 163 __ empty_FPU_stack(); 164 } 165 #endif 166 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) { 167 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled"); 168 } else { 169 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled"); 170 } 171 172 // In SSE mode, interpreter returns FP results in xmm0 but they need 173 // to end up back on the FPU so it can operate on them. 174 if (incoming_state == ftos && UseSSE >= 1) { 175 __ subptr(rsp, wordSize); 176 __ movflt(Address(rsp, 0), xmm0); 177 __ fld_s(Address(rsp, 0)); 178 __ addptr(rsp, wordSize); 179 } else if (incoming_state == dtos && UseSSE >= 2) { 180 __ subptr(rsp, 2*wordSize); 181 __ movdbl(Address(rsp, 0), xmm0); 182 __ fld_d(Address(rsp, 0)); 183 __ addptr(rsp, 2*wordSize); 184 } 185 186 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter"); 187 188 // Restore stack bottom in case i2c adjusted stack 189 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 190 // and NULL it as marker that rsp is now tos until next java call 191 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 192 193 __ restore_bcp(); 194 __ restore_locals(); 195 196 Label L_got_cache, L_giant_index; 197 if (EnableInvokeDynamic) { 198 __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic); 199 __ jcc(Assembler::equal, L_giant_index); 200 } 201 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2)); 202 __ bind(L_got_cache); 203 __ movl(rbx, Address(rbx, rcx, 204 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + 205 ConstantPoolCacheEntry::flags_offset())); 206 __ andptr(rbx, 0xFF); 207 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); 208 __ dispatch_next(state, step); 209 210 // out of the main line of code... 211 if (EnableInvokeDynamic) { 212 __ bind(L_giant_index); 213 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4)); 214 __ jmp(L_got_cache); 215 } 216 217 return entry; 218 } 219 220 221 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) { 222 address entry = __ pc(); 223 224 // In SSE mode, FP results are in xmm0 225 if (state == ftos && UseSSE > 0) { 226 __ subptr(rsp, wordSize); 227 __ movflt(Address(rsp, 0), xmm0); 228 __ fld_s(Address(rsp, 0)); 229 __ addptr(rsp, wordSize); 230 } else if (state == dtos && UseSSE >= 2) { 231 __ subptr(rsp, 2*wordSize); 232 __ movdbl(Address(rsp, 0), xmm0); 233 __ fld_d(Address(rsp, 0)); 234 __ addptr(rsp, 2*wordSize); 235 } 236 237 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter"); 238 239 // The stack is not extended by deopt but we must NULL last_sp as this 240 // entry is like a "return". 241 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 242 __ restore_bcp(); 243 __ restore_locals(); 244 // handle exceptions 245 { Label L; 246 const Register thread = rcx; 247 __ get_thread(thread); 248 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 249 __ jcc(Assembler::zero, L); 250 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception)); 251 __ should_not_reach_here(); 252 __ bind(L); 253 } 254 __ dispatch_next(state, step); 255 return entry; 256 } 257 258 259 int AbstractInterpreter::BasicType_as_index(BasicType type) { 260 int i = 0; 261 switch (type) { 262 case T_BOOLEAN: i = 0; break; 263 case T_CHAR : i = 1; break; 264 case T_BYTE : i = 2; break; 265 case T_SHORT : i = 3; break; 266 case T_INT : // fall through 267 case T_LONG : // fall through 268 case T_VOID : i = 4; break; 269 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE 270 case T_DOUBLE : i = 6; break; 271 case T_OBJECT : // fall through 272 case T_ARRAY : i = 7; break; 273 default : ShouldNotReachHere(); 274 } 275 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds"); 276 return i; 277 } 278 279 280 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) { 281 address entry = __ pc(); 282 switch (type) { 283 case T_BOOLEAN: __ c2bool(rax); break; 284 case T_CHAR : __ andptr(rax, 0xFFFF); break; 285 case T_BYTE : __ sign_extend_byte (rax); break; 286 case T_SHORT : __ sign_extend_short(rax); break; 287 case T_INT : /* nothing to do */ break; 288 case T_DOUBLE : 289 case T_FLOAT : 290 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); 291 __ pop(t); // remove return address first 292 // Must return a result for interpreter or compiler. In SSE 293 // mode, results are returned in xmm0 and the FPU stack must 294 // be empty. 295 if (type == T_FLOAT && UseSSE >= 1) { 296 // Load ST0 297 __ fld_d(Address(rsp, 0)); 298 // Store as float and empty fpu stack 299 __ fstp_s(Address(rsp, 0)); 300 // and reload 301 __ movflt(xmm0, Address(rsp, 0)); 302 } else if (type == T_DOUBLE && UseSSE >= 2 ) { 303 __ movdbl(xmm0, Address(rsp, 0)); 304 } else { 305 // restore ST0 306 __ fld_d(Address(rsp, 0)); 307 } 308 // and pop the temp 309 __ addptr(rsp, 2 * wordSize); 310 __ push(t); // restore return address 311 } 312 break; 313 case T_OBJECT : 314 // retrieve result from frame 315 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize)); 316 // and verify it 317 __ verify_oop(rax); 318 break; 319 default : ShouldNotReachHere(); 320 } 321 __ ret(0); // return from result handler 322 return entry; 323 } 324 325 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) { 326 address entry = __ pc(); 327 __ push(state); 328 __ call_VM(noreg, runtime_entry); 329 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 330 return entry; 331 } 332 333 334 // Helpers for commoning out cases in the various type of method entries. 335 // 336 337 // increment invocation count & check for overflow 338 // 339 // Note: checking for negative value instead of overflow 340 // so we have a 'sticky' overflow test 341 // 342 // rbx,: method 343 // rcx: invocation counter 344 // 345 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { 346 const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) + 347 in_bytes(InvocationCounter::counter_offset())); 348 // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not. 349 if (TieredCompilation) { 350 int increment = InvocationCounter::count_increment; 351 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift; 352 Label no_mdo, done; 353 if (ProfileInterpreter) { 354 // Are we profiling? 355 __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset())); 356 __ testptr(rax, rax); 357 __ jccb(Assembler::zero, no_mdo); 358 // Increment counter in the MDO 359 const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) + 360 in_bytes(InvocationCounter::counter_offset())); 361 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow); 362 __ jmpb(done); 363 } 364 __ bind(no_mdo); 365 // Increment counter in methodOop (we don't need to load it, it's in rcx). 366 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow); 367 __ bind(done); 368 } else { 369 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() + 370 InvocationCounter::counter_offset()); 371 372 if (ProfileInterpreter) { // %%% Merge this into methodDataOop 373 __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset())); 374 } 375 // Update standard invocation counters 376 __ movl(rax, backedge_counter); // load backedge counter 377 378 __ incrementl(rcx, InvocationCounter::count_increment); 379 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits 380 381 __ movl(invocation_counter, rcx); // save invocation count 382 __ addl(rcx, rax); // add both counters 383 384 // profile_method is non-null only for interpreted method so 385 // profile_method != NULL == !native_call 386 // BytecodeInterpreter only calls for native so code is elided. 387 388 if (ProfileInterpreter && profile_method != NULL) { 389 // Test to see if we should create a method data oop 390 __ cmp32(rcx, 391 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit)); 392 __ jcc(Assembler::less, *profile_method_continue); 393 394 // if no method data exists, go to profile_method 395 __ test_method_data_pointer(rax, *profile_method); 396 } 397 398 __ cmp32(rcx, 399 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit)); 400 __ jcc(Assembler::aboveEqual, *overflow); 401 } 402 } 403 404 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) { 405 406 // Asm interpreter on entry 407 // rdi - locals 408 // rsi - bcp 409 // rbx, - method 410 // rdx - cpool 411 // rbp, - interpreter frame 412 413 // C++ interpreter on entry 414 // rsi - new interpreter state pointer 415 // rbp - interpreter frame pointer 416 // rbx - method 417 418 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] 419 // rbx, - method 420 // rcx - rcvr (assuming there is one) 421 // top of stack return address of interpreter caller 422 // rsp - sender_sp 423 424 // C++ interpreter only 425 // rsi - previous interpreter state pointer 426 427 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); 428 429 // InterpreterRuntime::frequency_counter_overflow takes one argument 430 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp). 431 // The call returns the address of the verified entry point for the method or NULL 432 // if the compilation did not complete (either went background or bailed out). 433 __ movptr(rax, (intptr_t)false); 434 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax); 435 436 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop 437 438 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame 439 // and jump to the interpreted entry. 440 __ jmp(*do_continue, relocInfo::none); 441 442 } 443 444 void InterpreterGenerator::generate_stack_overflow_check(void) { 445 // see if we've got enough room on the stack for locals plus overhead. 446 // the expression stack grows down incrementally, so the normal guard 447 // page mechanism will work for that. 448 // 449 // Registers live on entry: 450 // 451 // Asm interpreter 452 // rdx: number of additional locals this frame needs (what we must check) 453 // rbx,: methodOop 454 455 // destroyed on exit 456 // rax, 457 458 // NOTE: since the additional locals are also always pushed (wasn't obvious in 459 // generate_method_entry) so the guard should work for them too. 460 // 461 462 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp 463 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 464 465 // total overhead size: entry_size + (saved rbp, thru expr stack bottom). 466 // be sure to change this if you add/subtract anything to/from the overhead area 467 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size; 468 469 const int page_size = os::vm_page_size(); 470 471 Label after_frame_check; 472 473 // see if the frame is greater than one page in size. If so, 474 // then we need to verify there is enough stack space remaining 475 // for the additional locals. 476 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize); 477 __ jcc(Assembler::belowEqual, after_frame_check); 478 479 // compute rsp as if this were going to be the last frame on 480 // the stack before the red zone 481 482 Label after_frame_check_pop; 483 484 __ push(rsi); 485 486 const Register thread = rsi; 487 488 __ get_thread(thread); 489 490 const Address stack_base(thread, Thread::stack_base_offset()); 491 const Address stack_size(thread, Thread::stack_size_offset()); 492 493 // locals + overhead, in bytes 494 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size)); 495 496 #ifdef ASSERT 497 Label stack_base_okay, stack_size_okay; 498 // verify that thread stack base is non-zero 499 __ cmpptr(stack_base, (int32_t)NULL_WORD); 500 __ jcc(Assembler::notEqual, stack_base_okay); 501 __ stop("stack base is zero"); 502 __ bind(stack_base_okay); 503 // verify that thread stack size is non-zero 504 __ cmpptr(stack_size, 0); 505 __ jcc(Assembler::notEqual, stack_size_okay); 506 __ stop("stack size is zero"); 507 __ bind(stack_size_okay); 508 #endif 509 510 // Add stack base to locals and subtract stack size 511 __ addptr(rax, stack_base); 512 __ subptr(rax, stack_size); 513 514 // Use the maximum number of pages we might bang. 515 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages : 516 (StackRedPages+StackYellowPages); 517 __ addptr(rax, max_pages * page_size); 518 519 // check against the current stack bottom 520 __ cmpptr(rsp, rax); 521 __ jcc(Assembler::above, after_frame_check_pop); 522 523 __ pop(rsi); // get saved bcp / (c++ prev state ). 524 525 // Restore sender's sp as SP. This is necessary if the sender's 526 // frame is an extended compiled frame (see gen_c2i_adapter()) 527 // and safer anyway in case of JSR292 adaptations. 528 529 __ pop(rax); // return address must be moved if SP is changed 530 __ mov(rsp, rsi); 531 __ push(rax); 532 533 // Note: the restored frame is not necessarily interpreted. 534 // Use the shared runtime version of the StackOverflowError. 535 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated"); 536 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry())); 537 // all done with frame size check 538 __ bind(after_frame_check_pop); 539 __ pop(rsi); 540 541 __ bind(after_frame_check); 542 } 543 544 // Allocate monitor and lock method (asm interpreter) 545 // rbx, - methodOop 546 // 547 void InterpreterGenerator::lock_method(void) { 548 // synchronize method 549 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); 550 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 551 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 552 553 #ifdef ASSERT 554 { Label L; 555 __ movl(rax, access_flags); 556 __ testl(rax, JVM_ACC_SYNCHRONIZED); 557 __ jcc(Assembler::notZero, L); 558 __ stop("method doesn't need synchronization"); 559 __ bind(L); 560 } 561 #endif // ASSERT 562 // get synchronization object 563 { Label done; 564 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 565 __ movl(rax, access_flags); 566 __ testl(rax, JVM_ACC_STATIC); 567 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case) 568 __ jcc(Assembler::zero, done); 569 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset())); 570 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes())); 571 __ movptr(rax, Address(rax, mirror_offset)); 572 __ bind(done); 573 } 574 // add space for monitor & lock 575 __ subptr(rsp, entry_size); // add space for a monitor entry 576 __ movptr(monitor_block_top, rsp); // set new monitor block top 577 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object 578 __ mov(rdx, rsp); // object address 579 __ lock_object(rdx); 580 } 581 582 // 583 // Generate a fixed interpreter frame. This is identical setup for interpreted methods 584 // and for native methods hence the shared code. 585 586 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { 587 // initialize fixed part of activation frame 588 __ push(rax); // save return address 589 __ enter(); // save old & set new rbp, 590 591 592 __ push(rsi); // set sender sp 593 __ push((int32_t)NULL_WORD); // leave last_sp as null 594 __ movptr(rsi, Address(rbx,methodOopDesc::const_offset())); // get constMethodOop 595 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase 596 __ push(rbx); // save methodOop 597 if (ProfileInterpreter) { 598 Label method_data_continue; 599 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset()))); 600 __ testptr(rdx, rdx); 601 __ jcc(Assembler::zero, method_data_continue); 602 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset())); 603 __ bind(method_data_continue); 604 __ push(rdx); // set the mdp (method data pointer) 605 } else { 606 __ push(0); 607 } 608 609 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset())); 610 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes())); 611 __ push(rdx); // set constant pool cache 612 __ push(rdi); // set locals pointer 613 if (native_call) { 614 __ push(0); // no bcp 615 } else { 616 __ push(rsi); // set bcp 617 } 618 __ push(0); // reserve word for pointer to expression stack bottom 619 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom 620 } 621 622 // End of helpers 623 624 // 625 // Various method entries 626 //------------------------------------------------------------------------------------------------------------------------ 627 // 628 // 629 630 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry 631 632 address InterpreterGenerator::generate_accessor_entry(void) { 633 634 // rbx,: methodOop 635 // rcx: receiver (preserve for slow entry into asm interpreter) 636 637 // rsi: senderSP must preserved for slow path, set SP to it on fast path 638 639 address entry_point = __ pc(); 640 Label xreturn_path; 641 642 // do fastpath for resolved accessor methods 643 if (UseFastAccessorMethods) { 644 Label slow_path; 645 // If we need a safepoint check, generate full interpreter entry. 646 ExternalAddress state(SafepointSynchronize::address_of_state()); 647 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 648 SafepointSynchronize::_not_synchronized); 649 650 __ jcc(Assembler::notEqual, slow_path); 651 // ASM/C++ Interpreter 652 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1 653 // Note: We can only use this code if the getfield has been resolved 654 // and if we don't have a null-pointer exception => check for 655 // these conditions first and use slow path if necessary. 656 // rbx,: method 657 // rcx: receiver 658 __ movptr(rax, Address(rsp, wordSize)); 659 660 // check if local 0 != NULL and read field 661 __ testptr(rax, rax); 662 __ jcc(Assembler::zero, slow_path); 663 664 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset())); 665 // read first instruction word and extract bytecode @ 1 and index @ 2 666 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset())); 667 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset())); 668 // Shift codes right to get the index on the right. 669 // The bytecode fetched looks like <index><0xb4><0x2a> 670 __ shrl(rdx, 2*BitsPerByte); 671 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); 672 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes())); 673 674 // rax,: local 0 675 // rbx,: method 676 // rcx: receiver - do not destroy since it is needed for slow path! 677 // rcx: scratch 678 // rdx: constant pool cache index 679 // rdi: constant pool cache 680 // rsi: sender sp 681 682 // check if getfield has been resolved and read constant pool cache entry 683 // check the validity of the cache entry by testing whether _indices field 684 // contains Bytecode::_getfield in b1 byte. 685 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below"); 686 __ movl(rcx, 687 Address(rdi, 688 rdx, 689 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset())); 690 __ shrl(rcx, 2*BitsPerByte); 691 __ andl(rcx, 0xFF); 692 __ cmpl(rcx, Bytecodes::_getfield); 693 __ jcc(Assembler::notEqual, slow_path); 694 695 // Note: constant pool entry is not valid before bytecode is resolved 696 __ movptr(rcx, 697 Address(rdi, 698 rdx, 699 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())); 700 __ movl(rdx, 701 Address(rdi, 702 rdx, 703 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset())); 704 705 Label notByte, notShort, notChar; 706 const Address field_address (rax, rcx, Address::times_1); 707 708 // Need to differentiate between igetfield, agetfield, bgetfield etc. 709 // because they are different sizes. 710 // Use the type from the constant pool cache 711 __ shrl(rdx, ConstantPoolCacheEntry::tosBits); 712 // Make sure we don't need to mask rdx for tosBits after the above shift 713 ConstantPoolCacheEntry::verify_tosBits(); 714 __ cmpl(rdx, btos); 715 __ jcc(Assembler::notEqual, notByte); 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 __ load_signed_short(rax, field_address); 723 __ jmp(xreturn_path); 724 725 __ bind(notShort); 726 __ cmpl(rdx, ctos); 727 __ jcc(Assembler::notEqual, notChar); 728 __ load_unsigned_short(rax, field_address); 729 __ jmp(xreturn_path); 730 731 __ bind(notChar); 732 #ifdef ASSERT 733 Label okay; 734 __ cmpl(rdx, atos); 735 __ jcc(Assembler::equal, okay); 736 __ cmpl(rdx, itos); 737 __ jcc(Assembler::equal, okay); 738 __ stop("what type is this?"); 739 __ bind(okay); 740 #endif // ASSERT 741 // All the rest are a 32 bit wordsize 742 // This is ok for now. Since fast accessors should be going away 743 __ movptr(rax, field_address); 744 745 __ bind(xreturn_path); 746 747 // _ireturn/_areturn 748 __ pop(rdi); // get return address 749 __ mov(rsp, rsi); // set sp to sender sp 750 __ jmp(rdi); 751 752 // generate a vanilla interpreter entry as the slow path 753 __ bind(slow_path); 754 755 (void) generate_normal_entry(false); 756 return entry_point; 757 } 758 return NULL; 759 760 } 761 762 // Method entry for java.lang.ref.Reference.get. 763 address InterpreterGenerator::generate_Reference_get_entry(void) { 764 #ifndef SERIALGC 765 // Code: _aload_0, _getfield, _areturn 766 // parameter size = 1 767 // 768 // The code that gets generated by this routine is split into 2 parts: 769 // 1. The "intrinsified" code for G1 (or any SATB based GC), 770 // 2. The slow path - which is an expansion of the regular method entry. 771 // 772 // Notes:- 773 // * In the G1 code we do not check whether we need to block for 774 // a safepoint. If G1 is enabled then we must execute the specialized 775 // code for Reference.get (except when the Reference object is null) 776 // so that we can log the value in the referent field with an SATB 777 // update buffer. 778 // If the code for the getfield template is modified so that the 779 // G1 pre-barrier code is executed when the current method is 780 // Reference.get() then going through the normal method entry 781 // will be fine. 782 // * The G1 code below can, however, check the receiver object (the instance 783 // of java.lang.Reference) and jump to the slow path if null. If the 784 // Reference object is null then we obviously cannot fetch the referent 785 // and so we don't need to call the G1 pre-barrier. Thus we can use the 786 // regular method entry code to generate the NPE. 787 // 788 // This code is based on generate_accessor_enty. 789 790 // rbx,: methodOop 791 // rcx: receiver (preserve for slow entry into asm interpreter) 792 793 // rsi: senderSP must preserved for slow path, set SP to it on fast path 794 795 address entry = __ pc(); 796 797 const int referent_offset = java_lang_ref_Reference::referent_offset; 798 guarantee(referent_offset > 0, "referent offset not initialized"); 799 800 if (UseG1GC) { 801 Label slow_path; 802 803 // Check if local 0 != NULL 804 // If the receiver is null then it is OK to jump to the slow path. 805 __ movptr(rax, Address(rsp, wordSize)); 806 __ testptr(rax, rax); 807 __ jcc(Assembler::zero, slow_path); 808 809 // rax: local 0 (must be preserved across the G1 barrier call) 810 // 811 // rbx: method (at this point it's scratch) 812 // rcx: receiver (at this point it's scratch) 813 // rdx: scratch 814 // rdi: scratch 815 // 816 // rsi: sender sp 817 818 // Preserve the sender sp in case the pre-barrier 819 // calls the runtime 820 __ push(rsi); 821 822 // Load the value of the referent field. 823 const Address field_address(rax, referent_offset); 824 __ movptr(rax, field_address); 825 826 // Generate the G1 pre-barrier code to log the value of 827 // the referent field in an SATB buffer. 828 __ get_thread(rcx); 829 __ g1_write_barrier_pre(noreg /* obj */, 830 rax /* pre_val */, 831 rcx /* thread */, 832 rbx /* tmp */, 833 true /* tosca_save */, 834 true /* expand_call */); 835 836 // _areturn 837 __ pop(rsi); // get sender sp 838 __ pop(rdi); // get return address 839 __ mov(rsp, rsi); // set sp to sender sp 840 __ jmp(rdi); 841 842 __ bind(slow_path); 843 (void) generate_normal_entry(false); 844 845 return entry; 846 } 847 #endif // SERIALGC 848 849 // If G1 is not enabled then attempt to go through the accessor entry point 850 // Reference.get is an accessor 851 return generate_accessor_entry(); 852 } 853 854 // 855 // Interpreter stub for calling a native method. (asm interpreter) 856 // This sets up a somewhat different looking stack for calling the native method 857 // than the typical interpreter frame setup. 858 // 859 860 address InterpreterGenerator::generate_native_entry(bool synchronized) { 861 // determine code generation flags 862 bool inc_counter = UseCompiler || CountCompiledCalls; 863 864 // rbx,: methodOop 865 // rsi: sender sp 866 // rsi: previous interpreter state (C++ interpreter) must preserve 867 address entry_point = __ pc(); 868 869 870 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); 871 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset()); 872 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); 873 874 // get parameter size (always needed) 875 __ load_unsigned_short(rcx, size_of_parameters); 876 877 // native calls don't need the stack size check since they have no expression stack 878 // and the arguments are already on the stack and we only add a handful of words 879 // to the stack 880 881 // rbx,: methodOop 882 // rcx: size of parameters 883 // rsi: sender sp 884 885 __ pop(rax); // get return address 886 // for natives the size of locals is zero 887 888 // compute beginning of parameters (rdi) 889 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize)); 890 891 892 // add 2 zero-initialized slots for native calls 893 // NULL result handler 894 __ push((int32_t)NULL_WORD); 895 // NULL oop temp (mirror or jni oop result) 896 __ push((int32_t)NULL_WORD); 897 898 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count 899 // initialize fixed part of activation frame 900 901 generate_fixed_frame(true); 902 903 // make sure method is native & not abstract 904 #ifdef ASSERT 905 __ movl(rax, access_flags); 906 { 907 Label L; 908 __ testl(rax, JVM_ACC_NATIVE); 909 __ jcc(Assembler::notZero, L); 910 __ stop("tried to execute non-native method as native"); 911 __ bind(L); 912 } 913 { Label L; 914 __ testl(rax, JVM_ACC_ABSTRACT); 915 __ jcc(Assembler::zero, L); 916 __ stop("tried to execute abstract method in interpreter"); 917 __ bind(L); 918 } 919 #endif 920 921 // Since at this point in the method invocation the exception handler 922 // would try to exit the monitor of synchronized methods which hasn't 923 // been entered yet, we set the thread local variable 924 // _do_not_unlock_if_synchronized to true. The remove_activation will 925 // check this flag. 926 927 __ get_thread(rax); 928 const Address do_not_unlock_if_synchronized(rax, 929 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 930 __ movbool(do_not_unlock_if_synchronized, true); 931 932 // increment invocation count & check for overflow 933 Label invocation_counter_overflow; 934 if (inc_counter) { 935 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); 936 } 937 938 Label continue_after_compile; 939 __ bind(continue_after_compile); 940 941 bang_stack_shadow_pages(true); 942 943 // reset the _do_not_unlock_if_synchronized flag 944 __ get_thread(rax); 945 __ movbool(do_not_unlock_if_synchronized, false); 946 947 // check for synchronized methods 948 // Must happen AFTER invocation_counter check and stack overflow check, 949 // so method is not locked if overflows. 950 // 951 if (synchronized) { 952 lock_method(); 953 } else { 954 // no synchronization necessary 955 #ifdef ASSERT 956 { Label L; 957 __ movl(rax, access_flags); 958 __ testl(rax, JVM_ACC_SYNCHRONIZED); 959 __ jcc(Assembler::zero, L); 960 __ stop("method needs synchronization"); 961 __ bind(L); 962 } 963 #endif 964 } 965 966 // start execution 967 #ifdef ASSERT 968 { Label L; 969 const Address monitor_block_top (rbp, 970 frame::interpreter_frame_monitor_block_top_offset * wordSize); 971 __ movptr(rax, monitor_block_top); 972 __ cmpptr(rax, rsp); 973 __ jcc(Assembler::equal, L); 974 __ stop("broken stack frame setup in interpreter"); 975 __ bind(L); 976 } 977 #endif 978 979 // jvmti/dtrace support 980 __ notify_method_entry(); 981 982 // work registers 983 const Register method = rbx; 984 const Register thread = rdi; 985 const Register t = rcx; 986 987 // allocate space for parameters 988 __ get_method(method); 989 __ verify_oop(method); 990 __ load_unsigned_short(t, Address(method, methodOopDesc::size_of_parameters_offset())); 991 __ shlptr(t, Interpreter::logStackElementSize); 992 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror 993 __ subptr(rsp, t); 994 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics 995 996 // get signature handler 997 { Label L; 998 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); 999 __ testptr(t, t); 1000 __ jcc(Assembler::notZero, L); 1001 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method); 1002 __ get_method(method); 1003 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); 1004 __ bind(L); 1005 } 1006 1007 // call signature handler 1008 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code"); 1009 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code"); 1010 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code"); 1011 // The generated handlers do not touch RBX (the method oop). 1012 // However, large signatures cannot be cached and are generated 1013 // each time here. The slow-path generator will blow RBX 1014 // sometime, so we must reload it after the call. 1015 __ call(t); 1016 __ get_method(method); // slow path call blows RBX on DevStudio 5.0 1017 1018 // result handler is in rax, 1019 // set result handler 1020 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax); 1021 1022 // pass mirror handle if static call 1023 { Label L; 1024 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 1025 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); 1026 __ testl(t, JVM_ACC_STATIC); 1027 __ jcc(Assembler::zero, L); 1028 // get mirror 1029 __ movptr(t, Address(method, methodOopDesc:: constants_offset())); 1030 __ movptr(t, Address(t, constantPoolOopDesc::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), t); 1034 // pass handle to mirror 1035 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize)); 1036 __ movptr(Address(rsp, wordSize), t); 1037 __ bind(L); 1038 } 1039 1040 // get native function entry point 1041 { Label L; 1042 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); 1043 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry()); 1044 __ cmpptr(rax, unsatisfied.addr()); 1045 __ jcc(Assembler::notEqual, L); 1046 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method); 1047 __ get_method(method); 1048 __ verify_oop(method); 1049 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); 1050 __ bind(L); 1051 } 1052 1053 // pass JNIEnv 1054 __ get_thread(thread); 1055 __ lea(t, Address(thread, JavaThread::jni_environment_offset())); 1056 __ movptr(Address(rsp, 0), t); 1057 1058 // set_last_Java_frame_before_call 1059 // It is enough that the pc() 1060 // points into the right code segment. It does not have to be the correct return pc. 1061 __ set_last_Java_frame(thread, noreg, rbp, __ pc()); 1062 1063 // change thread state 1064 #ifdef ASSERT 1065 { Label L; 1066 __ movl(t, Address(thread, JavaThread::thread_state_offset())); 1067 __ cmpl(t, _thread_in_Java); 1068 __ jcc(Assembler::equal, L); 1069 __ stop("Wrong thread state in native stub"); 1070 __ bind(L); 1071 } 1072 #endif 1073 1074 // Change state to native 1075 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native); 1076 __ call(rax); 1077 1078 // result potentially in rdx:rax or ST0 1079 1080 // Either restore the MXCSR register after returning from the JNI Call 1081 // or verify that it wasn't changed. 1082 if (VM_Version::supports_sse()) { 1083 if (RestoreMXCSROnJNICalls) { 1084 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std())); 1085 } 1086 else if (CheckJNICalls ) { 1087 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry())); 1088 } 1089 } 1090 1091 // Either restore the x87 floating pointer control word after returning 1092 // from the JNI call or verify that it wasn't changed. 1093 if (CheckJNICalls) { 1094 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry())); 1095 } 1096 1097 // save potential result in ST(0) & rdx:rax 1098 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 - 1099 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers) 1100 // It is safe to do this push because state is _thread_in_native and return address will be found 1101 // via _last_native_pc and not via _last_jave_sp 1102 1103 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result. 1104 // If the order changes or anything else is added to the stack the code in 1105 // interpreter_frame_result will have to be changed. 1106 1107 { Label L; 1108 Label push_double; 1109 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT)); 1110 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE)); 1111 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize), 1112 float_handler.addr()); 1113 __ jcc(Assembler::equal, push_double); 1114 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize), 1115 double_handler.addr()); 1116 __ jcc(Assembler::notEqual, L); 1117 __ bind(push_double); 1118 __ push(dtos); 1119 __ bind(L); 1120 } 1121 __ push(ltos); 1122 1123 // change thread state 1124 __ get_thread(thread); 1125 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans); 1126 if(os::is_MP()) { 1127 if (UseMembar) { 1128 // Force this write out before the read below 1129 __ membar(Assembler::Membar_mask_bits( 1130 Assembler::LoadLoad | Assembler::LoadStore | 1131 Assembler::StoreLoad | Assembler::StoreStore)); 1132 } else { 1133 // Write serialization page so VM thread can do a pseudo remote membar. 1134 // We use the current thread pointer to calculate a thread specific 1135 // offset to write to within the page. This minimizes bus traffic 1136 // due to cache line collision. 1137 __ serialize_memory(thread, rcx); 1138 } 1139 } 1140 1141 if (AlwaysRestoreFPU) { 1142 // Make sure the control word is correct. 1143 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); 1144 } 1145 1146 // check for safepoint operation in progress and/or pending suspend requests 1147 { Label Continue; 1148 1149 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 1150 SafepointSynchronize::_not_synchronized); 1151 1152 Label L; 1153 __ jcc(Assembler::notEqual, L); 1154 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0); 1155 __ jcc(Assembler::equal, Continue); 1156 __ bind(L); 1157 1158 // Don't use call_VM as it will see a possible pending exception and forward it 1159 // and never return here preventing us from clearing _last_native_pc down below. 1160 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are 1161 // preserved and correspond to the bcp/locals pointers. So we do a runtime call 1162 // by hand. 1163 // 1164 __ push(thread); 1165 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, 1166 JavaThread::check_special_condition_for_native_trans))); 1167 __ increment(rsp, wordSize); 1168 __ get_thread(thread); 1169 1170 __ bind(Continue); 1171 } 1172 1173 // change thread state 1174 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java); 1175 1176 __ reset_last_Java_frame(thread, true, true); 1177 1178 // reset handle block 1179 __ movptr(t, Address(thread, JavaThread::active_handles_offset())); 1180 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD); 1181 1182 // If result was an oop then unbox and save it in the frame 1183 { Label L; 1184 Label no_oop, store_result; 1185 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT)); 1186 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), 1187 handler.addr()); 1188 __ jcc(Assembler::notEqual, no_oop); 1189 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD); 1190 __ pop(ltos); 1191 __ testptr(rax, rax); 1192 __ jcc(Assembler::zero, store_result); 1193 // unbox 1194 __ movptr(rax, Address(rax, 0)); 1195 __ bind(store_result); 1196 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax); 1197 // keep stack depth as expected by pushing oop which will eventually be discarded 1198 __ push(ltos); 1199 __ bind(no_oop); 1200 } 1201 1202 { 1203 Label no_reguard; 1204 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled); 1205 __ jcc(Assembler::notEqual, no_reguard); 1206 1207 __ pusha(); 1208 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); 1209 __ popa(); 1210 1211 __ bind(no_reguard); 1212 } 1213 1214 // restore rsi to have legal interpreter frame, 1215 // i.e., bci == 0 <=> rsi == code_base() 1216 // Can't call_VM until bcp is within reasonable. 1217 __ get_method(method); // method is junk from thread_in_native to now. 1218 __ verify_oop(method); 1219 __ movptr(rsi, Address(method,methodOopDesc::const_offset())); // get constMethodOop 1220 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase 1221 1222 // handle exceptions (exception handling will handle unlocking!) 1223 { Label L; 1224 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 1225 __ jcc(Assembler::zero, L); 1226 // Note: At some point we may want to unify this with the code used in call_VM_base(); 1227 // i.e., we should use the StubRoutines::forward_exception code. For now this 1228 // doesn't work here because the rsp is not correctly set at this point. 1229 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception)); 1230 __ should_not_reach_here(); 1231 __ bind(L); 1232 } 1233 1234 // do unlocking if necessary 1235 { Label L; 1236 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); 1237 __ testl(t, JVM_ACC_SYNCHRONIZED); 1238 __ jcc(Assembler::zero, L); 1239 // the code below should be shared with interpreter macro assembler implementation 1240 { Label unlock; 1241 // BasicObjectLock will be first in list, since this is a synchronized method. However, need 1242 // to check that the object has not been unlocked by an explicit monitorexit bytecode. 1243 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock)); 1244 1245 __ lea(rdx, monitor); // address of first monitor 1246 1247 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); 1248 __ testptr(t, t); 1249 __ jcc(Assembler::notZero, unlock); 1250 1251 // Entry already unlocked, need to throw exception 1252 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1253 __ should_not_reach_here(); 1254 1255 __ bind(unlock); 1256 __ unlock_object(rdx); 1257 } 1258 __ bind(L); 1259 } 1260 1261 // jvmti/dtrace support 1262 // Note: This must happen _after_ handling/throwing any exceptions since 1263 // the exception handler code notifies the runtime of method exits 1264 // too. If this happens before, method entry/exit notifications are 1265 // not properly paired (was bug - gri 11/22/99). 1266 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); 1267 1268 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result 1269 __ pop(ltos); 1270 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize)); 1271 __ call(t); 1272 1273 // remove activation 1274 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp 1275 __ leave(); // remove frame anchor 1276 __ pop(rdi); // get return address 1277 __ mov(rsp, t); // set sp to sender sp 1278 __ jmp(rdi); 1279 1280 if (inc_counter) { 1281 // Handle overflow of counter and compile method 1282 __ bind(invocation_counter_overflow); 1283 generate_counter_overflow(&continue_after_compile); 1284 } 1285 1286 return entry_point; 1287 } 1288 1289 // 1290 // Generic interpreted method entry to (asm) interpreter 1291 // 1292 address InterpreterGenerator::generate_normal_entry(bool synchronized) { 1293 // determine code generation flags 1294 bool inc_counter = UseCompiler || CountCompiledCalls; 1295 1296 // rbx,: methodOop 1297 // rsi: sender sp 1298 address entry_point = __ pc(); 1299 1300 1301 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); 1302 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset()); 1303 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset()); 1304 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); 1305 1306 // get parameter size (always needed) 1307 __ load_unsigned_short(rcx, size_of_parameters); 1308 1309 // rbx,: methodOop 1310 // rcx: size of parameters 1311 1312 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i ) 1313 1314 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words 1315 __ subl(rdx, rcx); // rdx = no. of additional locals 1316 1317 // see if we've got enough room on the stack for locals plus overhead. 1318 generate_stack_overflow_check(); 1319 1320 // get return address 1321 __ pop(rax); 1322 1323 // compute beginning of parameters (rdi) 1324 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize)); 1325 1326 // rdx - # of additional locals 1327 // allocate space for locals 1328 // explicitly initialize locals 1329 { 1330 Label exit, loop; 1331 __ testl(rdx, rdx); 1332 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0 1333 __ bind(loop); 1334 __ push((int32_t)NULL_WORD); // initialize local variables 1335 __ decrement(rdx); // until everything initialized 1336 __ jcc(Assembler::greater, loop); 1337 __ bind(exit); 1338 } 1339 1340 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count 1341 // initialize fixed part of activation frame 1342 generate_fixed_frame(false); 1343 1344 // make sure method is not native & not abstract 1345 #ifdef ASSERT 1346 __ movl(rax, access_flags); 1347 { 1348 Label L; 1349 __ testl(rax, JVM_ACC_NATIVE); 1350 __ jcc(Assembler::zero, L); 1351 __ stop("tried to execute native method as non-native"); 1352 __ bind(L); 1353 } 1354 { Label L; 1355 __ testl(rax, JVM_ACC_ABSTRACT); 1356 __ jcc(Assembler::zero, L); 1357 __ stop("tried to execute abstract method in interpreter"); 1358 __ bind(L); 1359 } 1360 #endif 1361 1362 // Since at this point in the method invocation the exception handler 1363 // would try to exit the monitor of synchronized methods which hasn't 1364 // been entered yet, we set the thread local variable 1365 // _do_not_unlock_if_synchronized to true. The remove_activation will 1366 // check this flag. 1367 1368 __ get_thread(rax); 1369 const Address do_not_unlock_if_synchronized(rax, 1370 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1371 __ movbool(do_not_unlock_if_synchronized, true); 1372 1373 // increment invocation count & check for overflow 1374 Label invocation_counter_overflow; 1375 Label profile_method; 1376 Label profile_method_continue; 1377 if (inc_counter) { 1378 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue); 1379 if (ProfileInterpreter) { 1380 __ bind(profile_method_continue); 1381 } 1382 } 1383 Label continue_after_compile; 1384 __ bind(continue_after_compile); 1385 1386 bang_stack_shadow_pages(false); 1387 1388 // reset the _do_not_unlock_if_synchronized flag 1389 __ get_thread(rax); 1390 __ movbool(do_not_unlock_if_synchronized, false); 1391 1392 // check for synchronized methods 1393 // Must happen AFTER invocation_counter check and stack overflow check, 1394 // so method is not locked if overflows. 1395 // 1396 if (synchronized) { 1397 // Allocate monitor and lock method 1398 lock_method(); 1399 } else { 1400 // no synchronization necessary 1401 #ifdef ASSERT 1402 { Label L; 1403 __ movl(rax, access_flags); 1404 __ testl(rax, JVM_ACC_SYNCHRONIZED); 1405 __ jcc(Assembler::zero, L); 1406 __ stop("method needs synchronization"); 1407 __ bind(L); 1408 } 1409 #endif 1410 } 1411 1412 // start execution 1413 #ifdef ASSERT 1414 { Label L; 1415 const Address monitor_block_top (rbp, 1416 frame::interpreter_frame_monitor_block_top_offset * wordSize); 1417 __ movptr(rax, monitor_block_top); 1418 __ cmpptr(rax, rsp); 1419 __ jcc(Assembler::equal, L); 1420 __ stop("broken stack frame setup in interpreter"); 1421 __ bind(L); 1422 } 1423 #endif 1424 1425 // jvmti support 1426 __ notify_method_entry(); 1427 1428 __ dispatch_next(vtos); 1429 1430 // invocation counter overflow 1431 if (inc_counter) { 1432 if (ProfileInterpreter) { 1433 // We have decided to profile this method in the interpreter 1434 __ bind(profile_method); 1435 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1436 __ set_method_data_pointer_for_bcp(); 1437 __ get_method(rbx); 1438 __ jmp(profile_method_continue); 1439 } 1440 // Handle overflow of counter and compile method 1441 __ bind(invocation_counter_overflow); 1442 generate_counter_overflow(&continue_after_compile); 1443 } 1444 1445 return entry_point; 1446 } 1447 1448 //------------------------------------------------------------------------------------------------------------------------ 1449 // Entry points 1450 // 1451 // Here we generate the various kind of entries into the interpreter. 1452 // The two main entry type are generic bytecode methods and native call method. 1453 // These both come in synchronized and non-synchronized versions but the 1454 // frame layout they create is very similar. The other method entry 1455 // types are really just special purpose entries that are really entry 1456 // and interpretation all in one. These are for trivial methods like 1457 // accessor, empty, or special math methods. 1458 // 1459 // When control flow reaches any of the entry types for the interpreter 1460 // the following holds -> 1461 // 1462 // Arguments: 1463 // 1464 // rbx,: methodOop 1465 // rcx: receiver 1466 // 1467 // 1468 // Stack layout immediately at entry 1469 // 1470 // [ return address ] <--- rsp 1471 // [ parameter n ] 1472 // ... 1473 // [ parameter 1 ] 1474 // [ expression stack ] (caller's java expression stack) 1475 1476 // Assuming that we don't go to one of the trivial specialized 1477 // entries the stack will look like below when we are ready to execute 1478 // the first bytecode (or call the native routine). The register usage 1479 // will be as the template based interpreter expects (see interpreter_x86.hpp). 1480 // 1481 // local variables follow incoming parameters immediately; i.e. 1482 // the return address is moved to the end of the locals). 1483 // 1484 // [ monitor entry ] <--- rsp 1485 // ... 1486 // [ monitor entry ] 1487 // [ expr. stack bottom ] 1488 // [ saved rsi ] 1489 // [ current rdi ] 1490 // [ methodOop ] 1491 // [ saved rbp, ] <--- rbp, 1492 // [ return address ] 1493 // [ local variable m ] 1494 // ... 1495 // [ local variable 1 ] 1496 // [ parameter n ] 1497 // ... 1498 // [ parameter 1 ] <--- rdi 1499 1500 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) { 1501 // determine code generation flags 1502 bool synchronized = false; 1503 address entry_point = NULL; 1504 1505 switch (kind) { 1506 case Interpreter::zerolocals : break; 1507 case Interpreter::zerolocals_synchronized: synchronized = true; break; 1508 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break; 1509 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break; 1510 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break; 1511 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break; 1512 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break; 1513 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break; 1514 1515 case Interpreter::java_lang_math_sin : // fall thru 1516 case Interpreter::java_lang_math_cos : // fall thru 1517 case Interpreter::java_lang_math_tan : // fall thru 1518 case Interpreter::java_lang_math_abs : // fall thru 1519 case Interpreter::java_lang_math_log : // fall thru 1520 case Interpreter::java_lang_math_log10 : // fall thru 1521 case Interpreter::java_lang_math_sqrt : // fall thru 1522 case Interpreter::java_lang_math_pow : // fall thru 1523 case Interpreter::java_lang_math_exp : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break; 1524 case Interpreter::java_lang_ref_reference_get 1525 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break; 1526 default : ShouldNotReachHere(); break; 1527 } 1528 1529 if (entry_point) return entry_point; 1530 1531 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized); 1532 1533 } 1534 1535 // These should never be compiled since the interpreter will prefer 1536 // the compiled version to the intrinsic version. 1537 bool AbstractInterpreter::can_be_compiled(methodHandle m) { 1538 switch (method_kind(m)) { 1539 case Interpreter::java_lang_math_sin : // fall thru 1540 case Interpreter::java_lang_math_cos : // fall thru 1541 case Interpreter::java_lang_math_tan : // fall thru 1542 case Interpreter::java_lang_math_abs : // fall thru 1543 case Interpreter::java_lang_math_log : // fall thru 1544 case Interpreter::java_lang_math_log10 : // fall thru 1545 case Interpreter::java_lang_math_sqrt : // fall thru 1546 case Interpreter::java_lang_math_pow : // fall thru 1547 case Interpreter::java_lang_math_exp : 1548 return false; 1549 default: 1550 return true; 1551 } 1552 } 1553 1554 // How much stack a method activation needs in words. 1555 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) { 1556 1557 const int stub_code = 4; // see generate_call_stub 1558 // Save space for one monitor to get into the interpreted method in case 1559 // the method is synchronized 1560 int monitor_size = method->is_synchronized() ? 1561 1*frame::interpreter_frame_monitor_size() : 0; 1562 1563 // total overhead size: entry_size + (saved rbp, thru expr stack bottom). 1564 // be sure to change this if you add/subtract anything to/from the overhead area 1565 const int overhead_size = -frame::interpreter_frame_initial_sp_offset; 1566 1567 const int extra_stack = methodOopDesc::extra_stack_entries(); 1568 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) * 1569 Interpreter::stackElementWords; 1570 return overhead_size + method_stack + stub_code; 1571 } 1572 1573 // asm based interpreter deoptimization helpers 1574 1575 int AbstractInterpreter::layout_activation(methodOop method, 1576 int tempcount, 1577 int popframe_extra_args, 1578 int moncount, 1579 int caller_actual_parameters, 1580 int callee_param_count, 1581 int callee_locals, 1582 frame* caller, 1583 frame* interpreter_frame, 1584 bool is_top_frame) { 1585 // Note: This calculation must exactly parallel the frame setup 1586 // in AbstractInterpreterGenerator::generate_method_entry. 1587 // If interpreter_frame!=NULL, set up the method, locals, and monitors. 1588 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size, 1589 // as determined by a previous call to this method. 1590 // It is also guaranteed to be walkable even though it is in a skeletal state 1591 // NOTE: return size is in words not bytes 1592 1593 // fixed size of an interpreter frame: 1594 int max_locals = method->max_locals() * Interpreter::stackElementWords; 1595 int extra_locals = (method->max_locals() - method->size_of_parameters()) * 1596 Interpreter::stackElementWords; 1597 1598 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset; 1599 1600 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion) 1601 // Since the callee parameters already account for the callee's params we only need to account for 1602 // the extra locals. 1603 1604 1605 int size = overhead + 1606 ((callee_locals - callee_param_count)*Interpreter::stackElementWords) + 1607 (moncount*frame::interpreter_frame_monitor_size()) + 1608 tempcount*Interpreter::stackElementWords + popframe_extra_args; 1609 1610 if (interpreter_frame != NULL) { 1611 #ifdef ASSERT 1612 if (!EnableInvokeDynamic) 1613 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences? 1614 // Probably, since deoptimization doesn't work yet. 1615 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable"); 1616 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)"); 1617 #endif 1618 1619 interpreter_frame->interpreter_frame_set_method(method); 1620 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp 1621 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp) 1622 // and sender_sp is fp+8 1623 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1; 1624 1625 #ifdef ASSERT 1626 if (caller->is_interpreted_frame()) { 1627 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement"); 1628 } 1629 #endif 1630 1631 interpreter_frame->interpreter_frame_set_locals(locals); 1632 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin(); 1633 BasicObjectLock* monbot = montop - moncount; 1634 interpreter_frame->interpreter_frame_set_monitor_end(monbot); 1635 1636 // Set last_sp 1637 intptr_t* rsp = (intptr_t*) monbot - 1638 tempcount*Interpreter::stackElementWords - 1639 popframe_extra_args; 1640 interpreter_frame->interpreter_frame_set_last_sp(rsp); 1641 1642 // All frames but the initial (oldest) interpreter frame we fill in have a 1643 // value for sender_sp that allows walking the stack but isn't 1644 // truly correct. Correct the value here. 1645 1646 if (extra_locals != 0 && 1647 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) { 1648 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals); 1649 } 1650 *interpreter_frame->interpreter_frame_cache_addr() = 1651 method->constants()->cache(); 1652 } 1653 return size; 1654 } 1655 1656 1657 //------------------------------------------------------------------------------------------------------------------------ 1658 // Exceptions 1659 1660 void TemplateInterpreterGenerator::generate_throw_exception() { 1661 // Entry point in previous activation (i.e., if the caller was interpreted) 1662 Interpreter::_rethrow_exception_entry = __ pc(); 1663 const Register thread = rcx; 1664 1665 // Restore sp to interpreter_frame_last_sp even though we are going 1666 // to empty the expression stack for the exception processing. 1667 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 1668 // rax,: exception 1669 // rdx: return address/pc that threw exception 1670 __ restore_bcp(); // rsi points to call/send 1671 __ restore_locals(); 1672 1673 // Entry point for exceptions thrown within interpreter code 1674 Interpreter::_throw_exception_entry = __ pc(); 1675 // expression stack is undefined here 1676 // rax,: exception 1677 // rsi: exception bcp 1678 __ verify_oop(rax); 1679 1680 // expression stack must be empty before entering the VM in case of an exception 1681 __ empty_expression_stack(); 1682 __ empty_FPU_stack(); 1683 // find exception handler address and preserve exception oop 1684 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax); 1685 // rax,: exception handler entry point 1686 // rdx: preserved exception oop 1687 // rsi: bcp for exception handler 1688 __ push_ptr(rdx); // push exception which is now the only value on the stack 1689 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!) 1690 1691 // If the exception is not handled in the current frame the frame is removed and 1692 // the exception is rethrown (i.e. exception continuation is _rethrow_exception). 1693 // 1694 // Note: At this point the bci is still the bxi for the instruction which caused 1695 // the exception and the expression stack is empty. Thus, for any VM calls 1696 // at this point, GC will find a legal oop map (with empty expression stack). 1697 1698 // In current activation 1699 // tos: exception 1700 // rsi: exception bcp 1701 1702 // 1703 // JVMTI PopFrame support 1704 // 1705 1706 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 1707 __ empty_expression_stack(); 1708 __ empty_FPU_stack(); 1709 // Set the popframe_processing bit in pending_popframe_condition indicating that we are 1710 // currently handling popframe, so that call_VMs that may happen later do not trigger new 1711 // popframe handling cycles. 1712 __ get_thread(thread); 1713 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset())); 1714 __ orl(rdx, JavaThread::popframe_processing_bit); 1715 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx); 1716 1717 { 1718 // Check to see whether we are returning to a deoptimized frame. 1719 // (The PopFrame call ensures that the caller of the popped frame is 1720 // either interpreted or compiled and deoptimizes it if compiled.) 1721 // In this case, we can't call dispatch_next() after the frame is 1722 // popped, but instead must save the incoming arguments and restore 1723 // them after deoptimization has occurred. 1724 // 1725 // Note that we don't compare the return PC against the 1726 // deoptimization blob's unpack entry because of the presence of 1727 // adapter frames in C2. 1728 Label caller_not_deoptimized; 1729 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize)); 1730 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx); 1731 __ testl(rax, rax); 1732 __ jcc(Assembler::notZero, caller_not_deoptimized); 1733 1734 // Compute size of arguments for saving when returning to deoptimized caller 1735 __ get_method(rax); 1736 __ verify_oop(rax); 1737 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::size_of_parameters_offset()))); 1738 __ shlptr(rax, Interpreter::logStackElementSize); 1739 __ restore_locals(); 1740 __ subptr(rdi, rax); 1741 __ addptr(rdi, wordSize); 1742 // Save these arguments 1743 __ get_thread(thread); 1744 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi); 1745 1746 __ remove_activation(vtos, rdx, 1747 /* throw_monitor_exception */ false, 1748 /* install_monitor_exception */ false, 1749 /* notify_jvmdi */ false); 1750 1751 // Inform deoptimization that it is responsible for restoring these arguments 1752 __ get_thread(thread); 1753 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit); 1754 1755 // Continue in deoptimization handler 1756 __ jmp(rdx); 1757 1758 __ bind(caller_not_deoptimized); 1759 } 1760 1761 __ remove_activation(vtos, rdx, 1762 /* throw_monitor_exception */ false, 1763 /* install_monitor_exception */ false, 1764 /* notify_jvmdi */ false); 1765 1766 // Finish with popframe handling 1767 // A previous I2C followed by a deoptimization might have moved the 1768 // outgoing arguments further up the stack. PopFrame expects the 1769 // mutations to those outgoing arguments to be preserved and other 1770 // constraints basically require this frame to look exactly as 1771 // though it had previously invoked an interpreted activation with 1772 // no space between the top of the expression stack (current 1773 // last_sp) and the top of stack. Rather than force deopt to 1774 // maintain this kind of invariant all the time we call a small 1775 // fixup routine to move the mutated arguments onto the top of our 1776 // expression stack if necessary. 1777 __ mov(rax, rsp); 1778 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 1779 __ get_thread(thread); 1780 // PC must point into interpreter here 1781 __ set_last_Java_frame(thread, noreg, rbp, __ pc()); 1782 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx); 1783 __ get_thread(thread); 1784 __ reset_last_Java_frame(thread, true, true); 1785 // Restore the last_sp and null it out 1786 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); 1787 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 1788 1789 __ restore_bcp(); 1790 __ restore_locals(); 1791 // The method data pointer was incremented already during 1792 // call profiling. We have to restore the mdp for the current bcp. 1793 if (ProfileInterpreter) { 1794 __ set_method_data_pointer_for_bcp(); 1795 } 1796 1797 // Clear the popframe condition flag 1798 __ get_thread(thread); 1799 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive); 1800 1801 __ dispatch_next(vtos); 1802 // end of PopFrame support 1803 1804 Interpreter::_remove_activation_entry = __ pc(); 1805 1806 // preserve exception over this code sequence 1807 __ pop_ptr(rax); 1808 __ get_thread(thread); 1809 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax); 1810 // remove the activation (without doing throws on illegalMonitorExceptions) 1811 __ remove_activation(vtos, rdx, false, true, false); 1812 // restore exception 1813 __ get_thread(thread); 1814 __ movptr(rax, Address(thread, JavaThread::vm_result_offset())); 1815 __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD); 1816 __ verify_oop(rax); 1817 1818 // Inbetween activations - previous activation type unknown yet 1819 // compute continuation point - the continuation point expects 1820 // the following registers set up: 1821 // 1822 // rax: exception 1823 // rdx: return address/pc that threw exception 1824 // rsp: expression stack of caller 1825 // rbp: rbp, of caller 1826 __ push(rax); // save exception 1827 __ push(rdx); // save return address 1828 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx); 1829 __ mov(rbx, rax); // save exception handler 1830 __ pop(rdx); // restore return address 1831 __ pop(rax); // restore exception 1832 // Note that an "issuing PC" is actually the next PC after the call 1833 __ jmp(rbx); // jump to exception handler of caller 1834 } 1835 1836 1837 // 1838 // JVMTI ForceEarlyReturn support 1839 // 1840 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 1841 address entry = __ pc(); 1842 const Register thread = rcx; 1843 1844 __ restore_bcp(); 1845 __ restore_locals(); 1846 __ empty_expression_stack(); 1847 __ empty_FPU_stack(); 1848 __ load_earlyret_value(state); 1849 1850 __ get_thread(thread); 1851 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset())); 1852 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset()); 1853 1854 // Clear the earlyret state 1855 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive); 1856 1857 __ remove_activation(state, rsi, 1858 false, /* throw_monitor_exception */ 1859 false, /* install_monitor_exception */ 1860 true); /* notify_jvmdi */ 1861 __ jmp(rsi); 1862 return entry; 1863 } // end of ForceEarlyReturn support 1864 1865 1866 //------------------------------------------------------------------------------------------------------------------------ 1867 // Helper for vtos entry point generation 1868 1869 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) { 1870 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 1871 Label L; 1872 fep = __ pc(); __ push(ftos); __ jmp(L); 1873 dep = __ pc(); __ push(dtos); __ jmp(L); 1874 lep = __ pc(); __ push(ltos); __ jmp(L); 1875 aep = __ pc(); __ push(atos); __ jmp(L); 1876 bep = cep = sep = // fall through 1877 iep = __ pc(); __ push(itos); // fall through 1878 vep = __ pc(); __ bind(L); // fall through 1879 generate_and_dispatch(t); 1880 } 1881 1882 //------------------------------------------------------------------------------------------------------------------------ 1883 // Generation of individual instructions 1884 1885 // helpers for generate_and_dispatch 1886 1887 1888 1889 InterpreterGenerator::InterpreterGenerator(StubQueue* code) 1890 : TemplateInterpreterGenerator(code) { 1891 generate_all(); // down here so it can be "virtual" 1892 } 1893 1894 //------------------------------------------------------------------------------------------------------------------------ 1895 1896 // Non-product code 1897 #ifndef PRODUCT 1898 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 1899 address entry = __ pc(); 1900 1901 // prepare expression stack 1902 __ pop(rcx); // pop return address so expression stack is 'pure' 1903 __ push(state); // save tosca 1904 1905 // pass tosca registers as arguments & call tracer 1906 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx); 1907 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state) 1908 __ pop(state); // restore tosca 1909 1910 // return 1911 __ jmp(rcx); 1912 1913 return entry; 1914 } 1915 1916 1917 void TemplateInterpreterGenerator::count_bytecode() { 1918 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value)); 1919 } 1920 1921 1922 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 1923 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()])); 1924 } 1925 1926 1927 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 1928 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx); 1929 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes); 1930 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); 1931 ExternalAddress table((address) BytecodePairHistogram::_counters); 1932 Address index(noreg, rbx, Address::times_4); 1933 __ incrementl(ArrayAddress(table, index)); 1934 } 1935 1936 1937 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 1938 // Call a little run-time stub to avoid blow-up for each bytecode. 1939 // The run-time runtime saves the right registers, depending on 1940 // the tosca in-state for the given template. 1941 assert(Interpreter::trace_code(t->tos_in()) != NULL, 1942 "entry must have been generated"); 1943 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in()))); 1944 } 1945 1946 1947 void TemplateInterpreterGenerator::stop_interpreter_at() { 1948 Label L; 1949 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value), 1950 StopInterpreterAt); 1951 __ jcc(Assembler::notEqual, L); 1952 __ int3(); 1953 __ bind(L); 1954 } 1955 #endif // !PRODUCT 1956 #endif // CC_INTERP