1 /* 2 * Copyright (c) 2007, 2012, 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/cppInterpreter.hpp" 29 #include "interpreter/interpreter.hpp" 30 #include "interpreter/interpreterGenerator.hpp" 31 #include "interpreter/interpreterRuntime.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/interfaceSupport.hpp" 42 #include "runtime/sharedRuntime.hpp" 43 #include "runtime/stubRoutines.hpp" 44 #include "runtime/synchronizer.hpp" 45 #include "runtime/timer.hpp" 46 #include "runtime/vframeArray.hpp" 47 #include "utilities/debug.hpp" 48 #ifdef SHARK 49 #include "shark/shark_globals.hpp" 50 #endif 51 52 #ifdef CC_INTERP 53 54 // Routine exists to make tracebacks look decent in debugger 55 // while we are recursed in the frame manager/c++ interpreter. 56 // We could use an address in the frame manager but having 57 // frames look natural in the debugger is a plus. 58 extern "C" void RecursiveInterpreterActivation(interpreterState istate ) 59 { 60 // 61 ShouldNotReachHere(); 62 } 63 64 65 #define __ _masm-> 66 #define STATE(field_name) (Address(state, byte_offset_of(BytecodeInterpreter, field_name))) 67 68 Label fast_accessor_slow_entry_path; // fast accessor methods need to be able to jmp to unsynchronized 69 // c++ interpreter entry point this holds that entry point label. 70 71 // default registers for state and sender_sp 72 // state and sender_sp are the same on 32bit because we have no choice. 73 // state could be rsi on 64bit but it is an arg reg and not callee save 74 // so r13 is better choice. 75 76 const Register state = NOT_LP64(rsi) LP64_ONLY(r13); 77 const Register sender_sp_on_entry = NOT_LP64(rsi) LP64_ONLY(r13); 78 79 // NEEDED for JVMTI? 80 // address AbstractInterpreter::_remove_activation_preserving_args_entry; 81 82 static address unctrap_frame_manager_entry = NULL; 83 84 static address deopt_frame_manager_return_atos = NULL; 85 static address deopt_frame_manager_return_btos = NULL; 86 static address deopt_frame_manager_return_itos = NULL; 87 static address deopt_frame_manager_return_ltos = NULL; 88 static address deopt_frame_manager_return_ftos = NULL; 89 static address deopt_frame_manager_return_dtos = NULL; 90 static address deopt_frame_manager_return_vtos = NULL; 91 92 int AbstractInterpreter::BasicType_as_index(BasicType type) { 93 int i = 0; 94 switch (type) { 95 case T_BOOLEAN: i = 0; break; 96 case T_CHAR : i = 1; break; 97 case T_BYTE : i = 2; break; 98 case T_SHORT : i = 3; break; 99 case T_INT : i = 4; break; 100 case T_VOID : i = 5; break; 101 case T_FLOAT : i = 8; break; 102 case T_LONG : i = 9; break; 103 case T_DOUBLE : i = 6; break; 104 case T_OBJECT : // fall through 105 case T_ARRAY : i = 7; break; 106 default : ShouldNotReachHere(); 107 } 108 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds"); 109 return i; 110 } 111 112 // Is this pc anywhere within code owned by the interpreter? 113 // This only works for pc that might possibly be exposed to frame 114 // walkers. It clearly misses all of the actual c++ interpreter 115 // implementation 116 bool CppInterpreter::contains(address pc) { 117 return (_code->contains(pc) || 118 pc == CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation)); 119 } 120 121 122 address CppInterpreterGenerator::generate_result_handler_for(BasicType type) { 123 address entry = __ pc(); 124 switch (type) { 125 case T_BOOLEAN: __ c2bool(rax); break; 126 case T_CHAR : __ andl(rax, 0xFFFF); break; 127 case T_BYTE : __ sign_extend_byte (rax); break; 128 case T_SHORT : __ sign_extend_short(rax); break; 129 case T_VOID : // fall thru 130 case T_LONG : // fall thru 131 case T_INT : /* nothing to do */ break; 132 133 case T_DOUBLE : 134 case T_FLOAT : 135 { 136 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); 137 __ pop(t); // remove return address first 138 // Must return a result for interpreter or compiler. In SSE 139 // mode, results are returned in xmm0 and the FPU stack must 140 // be empty. 141 if (type == T_FLOAT && UseSSE >= 1) { 142 #ifndef _LP64 143 // Load ST0 144 __ fld_d(Address(rsp, 0)); 145 // Store as float and empty fpu stack 146 __ fstp_s(Address(rsp, 0)); 147 #endif // !_LP64 148 // and reload 149 __ movflt(xmm0, Address(rsp, 0)); 150 } else if (type == T_DOUBLE && UseSSE >= 2 ) { 151 __ movdbl(xmm0, Address(rsp, 0)); 152 } else { 153 // restore ST0 154 __ fld_d(Address(rsp, 0)); 155 } 156 // and pop the temp 157 __ addptr(rsp, 2 * wordSize); 158 __ push(t); // restore return address 159 } 160 break; 161 case T_OBJECT : 162 // retrieve result from frame 163 __ movptr(rax, STATE(_oop_temp)); 164 // and verify it 165 __ verify_oop(rax); 166 break; 167 default : ShouldNotReachHere(); 168 } 169 __ ret(0); // return from result handler 170 return entry; 171 } 172 173 // tosca based result to c++ interpreter stack based result. 174 // Result goes to top of native stack. 175 176 #undef EXTEND // SHOULD NOT BE NEEDED 177 address CppInterpreterGenerator::generate_tosca_to_stack_converter(BasicType type) { 178 // A result is in the tosca (abi result) from either a native method call or compiled 179 // code. Place this result on the java expression stack so C++ interpreter can use it. 180 address entry = __ pc(); 181 182 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); 183 __ pop(t); // remove return address first 184 switch (type) { 185 case T_VOID: 186 break; 187 case T_BOOLEAN: 188 #ifdef EXTEND 189 __ c2bool(rax); 190 #endif 191 __ push(rax); 192 break; 193 case T_CHAR : 194 #ifdef EXTEND 195 __ andl(rax, 0xFFFF); 196 #endif 197 __ push(rax); 198 break; 199 case T_BYTE : 200 #ifdef EXTEND 201 __ sign_extend_byte (rax); 202 #endif 203 __ push(rax); 204 break; 205 case T_SHORT : 206 #ifdef EXTEND 207 __ sign_extend_short(rax); 208 #endif 209 __ push(rax); 210 break; 211 case T_LONG : 212 __ push(rdx); // pushes useless junk on 64bit 213 __ push(rax); 214 break; 215 case T_INT : 216 __ push(rax); 217 break; 218 case T_FLOAT : 219 // Result is in ST(0)/xmm0 220 __ subptr(rsp, wordSize); 221 if ( UseSSE < 1) { 222 __ fstp_s(Address(rsp, 0)); 223 } else { 224 __ movflt(Address(rsp, 0), xmm0); 225 } 226 break; 227 case T_DOUBLE : 228 __ subptr(rsp, 2*wordSize); 229 if ( UseSSE < 2 ) { 230 __ fstp_d(Address(rsp, 0)); 231 } else { 232 __ movdbl(Address(rsp, 0), xmm0); 233 } 234 break; 235 case T_OBJECT : 236 __ verify_oop(rax); // verify it 237 __ push(rax); 238 break; 239 default : ShouldNotReachHere(); 240 } 241 __ jmp(t); // return from result handler 242 return entry; 243 } 244 245 address CppInterpreterGenerator::generate_stack_to_stack_converter(BasicType type) { 246 // A result is in the java expression stack of the interpreted method that has just 247 // returned. Place this result on the java expression stack of the caller. 248 // 249 // The current interpreter activation in rsi/r13 is for the method just returning its 250 // result. So we know that the result of this method is on the top of the current 251 // execution stack (which is pre-pushed) and will be return to the top of the caller 252 // stack. The top of the callers stack is the bottom of the locals of the current 253 // activation. 254 // Because of the way activation are managed by the frame manager the value of rsp is 255 // below both the stack top of the current activation and naturally the stack top 256 // of the calling activation. This enable this routine to leave the return address 257 // to the frame manager on the stack and do a vanilla return. 258 // 259 // On entry: rsi/r13 - interpreter state of activation returning a (potential) result 260 // On Return: rsi/r13 - unchanged 261 // rax - new stack top for caller activation (i.e. activation in _prev_link) 262 // 263 // Can destroy rdx, rcx. 264 // 265 266 address entry = __ pc(); 267 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); 268 switch (type) { 269 case T_VOID: 270 __ movptr(rax, STATE(_locals)); // pop parameters get new stack value 271 __ addptr(rax, wordSize); // account for prepush before we return 272 break; 273 case T_FLOAT : 274 case T_BOOLEAN: 275 case T_CHAR : 276 case T_BYTE : 277 case T_SHORT : 278 case T_INT : 279 // 1 word result 280 __ movptr(rdx, STATE(_stack)); 281 __ movptr(rax, STATE(_locals)); // address for result 282 __ movl(rdx, Address(rdx, wordSize)); // get result 283 __ movptr(Address(rax, 0), rdx); // and store it 284 break; 285 case T_LONG : 286 case T_DOUBLE : 287 // return top two words on current expression stack to caller's expression stack 288 // The caller's expression stack is adjacent to the current frame manager's intepretState 289 // except we allocated one extra word for this intepretState so we won't overwrite it 290 // when we return a two word result. 291 292 __ movptr(rax, STATE(_locals)); // address for result 293 __ movptr(rcx, STATE(_stack)); 294 __ subptr(rax, wordSize); // need addition word besides locals[0] 295 __ movptr(rdx, Address(rcx, 2*wordSize)); // get result word (junk in 64bit) 296 __ movptr(Address(rax, wordSize), rdx); // and store it 297 __ movptr(rdx, Address(rcx, wordSize)); // get result word 298 __ movptr(Address(rax, 0), rdx); // and store it 299 break; 300 case T_OBJECT : 301 __ movptr(rdx, STATE(_stack)); 302 __ movptr(rax, STATE(_locals)); // address for result 303 __ movptr(rdx, Address(rdx, wordSize)); // get result 304 __ verify_oop(rdx); // verify it 305 __ movptr(Address(rax, 0), rdx); // and store it 306 break; 307 default : ShouldNotReachHere(); 308 } 309 __ ret(0); 310 return entry; 311 } 312 313 address CppInterpreterGenerator::generate_stack_to_native_abi_converter(BasicType type) { 314 // A result is in the java expression stack of the interpreted method that has just 315 // returned. Place this result in the native abi that the caller expects. 316 // 317 // Similar to generate_stack_to_stack_converter above. Called at a similar time from the 318 // frame manager execept in this situation the caller is native code (c1/c2/call_stub) 319 // and so rather than return result onto caller's java expression stack we return the 320 // result in the expected location based on the native abi. 321 // On entry: rsi/r13 - interpreter state of activation returning a (potential) result 322 // On Return: rsi/r13 - unchanged 323 // Other registers changed [rax/rdx/ST(0) as needed for the result returned] 324 325 address entry = __ pc(); 326 switch (type) { 327 case T_VOID: 328 break; 329 case T_BOOLEAN: 330 case T_CHAR : 331 case T_BYTE : 332 case T_SHORT : 333 case T_INT : 334 __ movptr(rdx, STATE(_stack)); // get top of stack 335 __ movl(rax, Address(rdx, wordSize)); // get result word 1 336 break; 337 case T_LONG : 338 __ movptr(rdx, STATE(_stack)); // get top of stack 339 __ movptr(rax, Address(rdx, wordSize)); // get result low word 340 NOT_LP64(__ movl(rdx, Address(rdx, 2*wordSize));) // get result high word 341 break; 342 case T_FLOAT : 343 __ movptr(rdx, STATE(_stack)); // get top of stack 344 if ( UseSSE >= 1) { 345 __ movflt(xmm0, Address(rdx, wordSize)); 346 } else { 347 __ fld_s(Address(rdx, wordSize)); // pushd float result 348 } 349 break; 350 case T_DOUBLE : 351 __ movptr(rdx, STATE(_stack)); // get top of stack 352 if ( UseSSE > 1) { 353 __ movdbl(xmm0, Address(rdx, wordSize)); 354 } else { 355 __ fld_d(Address(rdx, wordSize)); // push double result 356 } 357 break; 358 case T_OBJECT : 359 __ movptr(rdx, STATE(_stack)); // get top of stack 360 __ movptr(rax, Address(rdx, wordSize)); // get result word 1 361 __ verify_oop(rax); // verify it 362 break; 363 default : ShouldNotReachHere(); 364 } 365 __ ret(0); 366 return entry; 367 } 368 369 address CppInterpreter::return_entry(TosState state, int length) { 370 // make it look good in the debugger 371 return CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation); 372 } 373 374 address CppInterpreter::deopt_entry(TosState state, int length) { 375 address ret = NULL; 376 if (length != 0) { 377 switch (state) { 378 case atos: ret = deopt_frame_manager_return_atos; break; 379 case btos: ret = deopt_frame_manager_return_btos; break; 380 case ctos: 381 case stos: 382 case itos: ret = deopt_frame_manager_return_itos; break; 383 case ltos: ret = deopt_frame_manager_return_ltos; break; 384 case ftos: ret = deopt_frame_manager_return_ftos; break; 385 case dtos: ret = deopt_frame_manager_return_dtos; break; 386 case vtos: ret = deopt_frame_manager_return_vtos; break; 387 } 388 } else { 389 ret = unctrap_frame_manager_entry; // re-execute the bytecode ( e.g. uncommon trap) 390 } 391 assert(ret != NULL, "Not initialized"); 392 return ret; 393 } 394 395 // C++ Interpreter 396 void CppInterpreterGenerator::generate_compute_interpreter_state(const Register state, 397 const Register locals, 398 const Register sender_sp, 399 bool native) { 400 401 // On entry the "locals" argument points to locals[0] (or where it would be in case no locals in 402 // a static method). "state" contains any previous frame manager state which we must save a link 403 // to in the newly generated state object. On return "state" is a pointer to the newly allocated 404 // state object. We must allocate and initialize a new interpretState object and the method 405 // expression stack. Because the returned result (if any) of the method will be placed on the caller's 406 // expression stack and this will overlap with locals[0] (and locals[1] if double/long) we must 407 // be sure to leave space on the caller's stack so that this result will not overwrite values when 408 // locals[0] and locals[1] do not exist (and in fact are return address and saved rbp). So when 409 // we are non-native we in essence ensure that locals[0-1] exist. We play an extra trick in 410 // non-product builds and initialize this last local with the previous interpreterState as 411 // this makes things look real nice in the debugger. 412 413 // State on entry 414 // Assumes locals == &locals[0] 415 // Assumes state == any previous frame manager state (assuming call path from c++ interpreter) 416 // Assumes rax = return address 417 // rcx == senders_sp 418 // rbx == method 419 // Modifies rcx, rdx, rax 420 // Returns: 421 // state == address of new interpreterState 422 // rsp == bottom of method's expression stack. 423 424 const Address const_offset (rbx, methodOopDesc::const_offset()); 425 426 427 // On entry sp is the sender's sp. This includes the space for the arguments 428 // that the sender pushed. If the sender pushed no args (a static) and the 429 // caller returns a long then we need two words on the sender's stack which 430 // are not present (although when we return a restore full size stack the 431 // space will be present). If we didn't allocate two words here then when 432 // we "push" the result of the caller's stack we would overwrite the return 433 // address and the saved rbp. Not good. So simply allocate 2 words now 434 // just to be safe. This is the "static long no_params() method" issue. 435 // See Lo.java for a testcase. 436 // We don't need this for native calls because they return result in 437 // register and the stack is expanded in the caller before we store 438 // the results on the stack. 439 440 if (!native) { 441 #ifdef PRODUCT 442 __ subptr(rsp, 2*wordSize); 443 #else /* PRODUCT */ 444 __ push((int32_t)NULL_WORD); 445 __ push(state); // make it look like a real argument 446 #endif /* PRODUCT */ 447 } 448 449 // Now that we are assure of space for stack result, setup typical linkage 450 451 __ push(rax); 452 __ enter(); 453 454 __ mov(rax, state); // save current state 455 456 __ lea(rsp, Address(rsp, -(int)sizeof(BytecodeInterpreter))); 457 __ mov(state, rsp); 458 459 // rsi/r13 == state/locals rax == prevstate 460 461 // initialize the "shadow" frame so that use since C++ interpreter not directly 462 // recursive. Simpler to recurse but we can't trim expression stack as we call 463 // new methods. 464 __ movptr(STATE(_locals), locals); // state->_locals = locals() 465 __ movptr(STATE(_self_link), state); // point to self 466 __ movptr(STATE(_prev_link), rax); // state->_link = state on entry (NULL or previous state) 467 __ movptr(STATE(_sender_sp), sender_sp); // state->_sender_sp = sender_sp 468 #ifdef _LP64 469 __ movptr(STATE(_thread), r15_thread); // state->_bcp = codes() 470 #else 471 __ get_thread(rax); // get vm's javathread* 472 __ movptr(STATE(_thread), rax); // state->_bcp = codes() 473 #endif // _LP64 474 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset())); // get constantMethodOop 475 __ lea(rdx, Address(rdx, constMethodOopDesc::codes_offset())); // get code base 476 if (native) { 477 __ movptr(STATE(_bcp), (int32_t)NULL_WORD); // state->_bcp = NULL 478 } else { 479 __ movptr(STATE(_bcp), rdx); // state->_bcp = codes() 480 } 481 __ xorptr(rdx, rdx); 482 __ movptr(STATE(_oop_temp), rdx); // state->_oop_temp = NULL (only really needed for native) 483 __ movptr(STATE(_mdx), rdx); // state->_mdx = NULL 484 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset())); 485 __ movptr(rdx, Address(rdx, constMethodOopDesc::constants_offset())); 486 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes())); 487 __ movptr(STATE(_constants), rdx); // state->_constants = constants() 488 489 __ movptr(STATE(_method), rbx); // state->_method = method() 490 __ movl(STATE(_msg), (int32_t) BytecodeInterpreter::method_entry); // state->_msg = initial method entry 491 __ movptr(STATE(_result._to_call._callee), (int32_t) NULL_WORD); // state->_result._to_call._callee_callee = NULL 492 493 494 __ movptr(STATE(_monitor_base), rsp); // set monitor block bottom (grows down) this would point to entry [0] 495 // entries run from -1..x where &monitor[x] == 496 497 { 498 // Must not attempt to lock method until we enter interpreter as gc won't be able to find the 499 // initial frame. However we allocate a free monitor so we don't have to shuffle the expression stack 500 // immediately. 501 502 // synchronize method 503 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); 504 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 505 Label not_synced; 506 507 __ movl(rax, access_flags); 508 __ testl(rax, JVM_ACC_SYNCHRONIZED); 509 __ jcc(Assembler::zero, not_synced); 510 511 // Allocate initial monitor and pre initialize it 512 // get synchronization object 513 514 Label done; 515 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 516 __ movl(rax, access_flags); 517 __ testl(rax, JVM_ACC_STATIC); 518 __ movptr(rax, Address(locals, 0)); // get receiver (assume this is frequent case) 519 __ jcc(Assembler::zero, done); 520 __ movptr(rax, Address(rbx, methodOopDesc::const_offset())); 521 __ movptr(rax, Address(rax, constMethodOopDesc::constants_offset())); 522 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes())); 523 __ movptr(rax, Address(rax, mirror_offset)); 524 __ bind(done); 525 // add space for monitor & lock 526 __ subptr(rsp, entry_size); // add space for a monitor entry 527 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object 528 __ bind(not_synced); 529 } 530 531 __ movptr(STATE(_stack_base), rsp); // set expression stack base ( == &monitors[-count]) 532 if (native) { 533 __ movptr(STATE(_stack), rsp); // set current expression stack tos 534 __ movptr(STATE(_stack_limit), rsp); 535 } else { 536 __ subptr(rsp, wordSize); // pre-push stack 537 __ movptr(STATE(_stack), rsp); // set current expression stack tos 538 539 // compute full expression stack limit 540 541 const Address size_of_stack (rbx, methodOopDesc::max_stack_offset()); 542 const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_words(); 543 __ load_unsigned_short(rdx, size_of_stack); // get size of expression stack in words 544 __ negptr(rdx); // so we can subtract in next step 545 // Allocate expression stack 546 __ lea(rsp, Address(rsp, rdx, Address::times_ptr, -extra_stack)); 547 __ movptr(STATE(_stack_limit), rsp); 548 } 549 550 #ifdef _LP64 551 // Make sure stack is properly aligned and sized for the abi 552 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 553 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI) 554 #endif // _LP64 555 556 557 558 } 559 560 // Helpers for commoning out cases in the various type of method entries. 561 // 562 563 // increment invocation count & check for overflow 564 // 565 // Note: checking for negative value instead of overflow 566 // so we have a 'sticky' overflow test 567 // 568 // rbx,: method 569 // rcx: invocation counter 570 // 571 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { 572 573 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset()); 574 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset()); 575 576 if (ProfileInterpreter) { // %%% Merge this into methodDataOop 577 __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset())); 578 } 579 // Update standard invocation counters 580 __ movl(rax, backedge_counter); // load backedge counter 581 582 __ increment(rcx, InvocationCounter::count_increment); 583 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits 584 585 __ movl(invocation_counter, rcx); // save invocation count 586 __ addl(rcx, rax); // add both counters 587 588 // profile_method is non-null only for interpreted method so 589 // profile_method != NULL == !native_call 590 // BytecodeInterpreter only calls for native so code is elided. 591 592 __ cmp32(rcx, 593 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit)); 594 __ jcc(Assembler::aboveEqual, *overflow); 595 596 } 597 598 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) { 599 600 // C++ interpreter on entry 601 // rsi/r13 - new interpreter state pointer 602 // rbp - interpreter frame pointer 603 // rbx - method 604 605 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] 606 // rbx, - method 607 // rcx - rcvr (assuming there is one) 608 // top of stack return address of interpreter caller 609 // rsp - sender_sp 610 611 // C++ interpreter only 612 // rsi/r13 - previous interpreter state pointer 613 614 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); 615 616 // InterpreterRuntime::frequency_counter_overflow takes one argument 617 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp). 618 // The call returns the address of the verified entry point for the method or NULL 619 // if the compilation did not complete (either went background or bailed out). 620 __ movptr(rax, (int32_t)false); 621 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax); 622 623 // for c++ interpreter can rsi really be munged? 624 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); // restore state 625 __ movptr(rbx, Address(state, byte_offset_of(BytecodeInterpreter, _method))); // restore method 626 __ movptr(rdi, Address(state, byte_offset_of(BytecodeInterpreter, _locals))); // get locals pointer 627 628 __ jmp(*do_continue, relocInfo::none); 629 630 } 631 632 void InterpreterGenerator::generate_stack_overflow_check(void) { 633 // see if we've got enough room on the stack for locals plus overhead. 634 // the expression stack grows down incrementally, so the normal guard 635 // page mechanism will work for that. 636 // 637 // Registers live on entry: 638 // 639 // Asm interpreter 640 // rdx: number of additional locals this frame needs (what we must check) 641 // rbx,: methodOop 642 643 // C++ Interpreter 644 // rsi/r13: previous interpreter frame state object 645 // rdi: &locals[0] 646 // rcx: # of locals 647 // rdx: number of additional locals this frame needs (what we must check) 648 // rbx: methodOop 649 650 // destroyed on exit 651 // rax, 652 653 // NOTE: since the additional locals are also always pushed (wasn't obvious in 654 // generate_method_entry) so the guard should work for them too. 655 // 656 657 // monitor entry size: see picture of stack set (generate_method_entry) and frame_i486.hpp 658 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 659 660 // total overhead size: entry_size + (saved rbp, thru expr stack bottom). 661 // be sure to change this if you add/subtract anything to/from the overhead area 662 const int overhead_size = (int)sizeof(BytecodeInterpreter); 663 664 const int page_size = os::vm_page_size(); 665 666 Label after_frame_check; 667 668 // compute rsp as if this were going to be the last frame on 669 // the stack before the red zone 670 671 Label after_frame_check_pop; 672 673 // save rsi == caller's bytecode ptr (c++ previous interp. state) 674 // QQQ problem here?? rsi overload???? 675 __ push(state); 676 677 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rsi); 678 679 NOT_LP64(__ get_thread(thread)); 680 681 const Address stack_base(thread, Thread::stack_base_offset()); 682 const Address stack_size(thread, Thread::stack_size_offset()); 683 684 // locals + overhead, in bytes 685 const Address size_of_stack (rbx, methodOopDesc::max_stack_offset()); 686 // Always give one monitor to allow us to start interp if sync method. 687 // Any additional monitors need a check when moving the expression stack 688 const int one_monitor = frame::interpreter_frame_monitor_size() * wordSize; 689 const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries(); 690 __ load_unsigned_short(rax, size_of_stack); // get size of expression stack in words 691 __ lea(rax, Address(noreg, rax, Interpreter::stackElementScale(), extra_stack + one_monitor)); 692 __ lea(rax, Address(rax, rdx, Interpreter::stackElementScale(), overhead_size)); 693 694 #ifdef ASSERT 695 Label stack_base_okay, stack_size_okay; 696 // verify that thread stack base is non-zero 697 __ cmpptr(stack_base, (int32_t)0); 698 __ jcc(Assembler::notEqual, stack_base_okay); 699 __ stop("stack base is zero"); 700 __ bind(stack_base_okay); 701 // verify that thread stack size is non-zero 702 __ cmpptr(stack_size, (int32_t)0); 703 __ jcc(Assembler::notEqual, stack_size_okay); 704 __ stop("stack size is zero"); 705 __ bind(stack_size_okay); 706 #endif 707 708 // Add stack base to locals and subtract stack size 709 __ addptr(rax, stack_base); 710 __ subptr(rax, stack_size); 711 712 // We should have a magic number here for the size of the c++ interpreter frame. 713 // We can't actually tell this ahead of time. The debug version size is around 3k 714 // product is 1k and fastdebug is 4k 715 const int slop = 6 * K; 716 717 // Use the maximum number of pages we might bang. 718 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages : 719 (StackRedPages+StackYellowPages); 720 // Only need this if we are stack banging which is temporary while 721 // we're debugging. 722 __ addptr(rax, slop + 2*max_pages * page_size); 723 724 // check against the current stack bottom 725 __ cmpptr(rsp, rax); 726 __ jcc(Assembler::above, after_frame_check_pop); 727 728 __ pop(state); // get c++ prev state. 729 730 // throw exception return address becomes throwing pc 731 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError)); 732 733 // all done with frame size check 734 __ bind(after_frame_check_pop); 735 __ pop(state); 736 737 __ bind(after_frame_check); 738 } 739 740 // Find preallocated monitor and lock method (C++ interpreter) 741 // rbx - methodOop 742 // 743 void InterpreterGenerator::lock_method(void) { 744 // assumes state == rsi/r13 == pointer to current interpreterState 745 // minimally destroys rax, rdx|c_rarg1, rdi 746 // 747 // synchronize method 748 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 749 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); 750 751 const Register monitor = NOT_LP64(rdx) LP64_ONLY(c_rarg1); 752 753 // find initial monitor i.e. monitors[-1] 754 __ movptr(monitor, STATE(_monitor_base)); // get monitor bottom limit 755 __ subptr(monitor, entry_size); // point to initial monitor 756 757 #ifdef ASSERT 758 { Label L; 759 __ movl(rax, access_flags); 760 __ testl(rax, JVM_ACC_SYNCHRONIZED); 761 __ jcc(Assembler::notZero, L); 762 __ stop("method doesn't need synchronization"); 763 __ bind(L); 764 } 765 #endif // ASSERT 766 // get synchronization object 767 { Label done; 768 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 769 __ movl(rax, access_flags); 770 __ movptr(rdi, STATE(_locals)); // prepare to get receiver (assume common case) 771 __ testl(rax, JVM_ACC_STATIC); 772 __ movptr(rax, Address(rdi, 0)); // get receiver (assume this is frequent case) 773 __ jcc(Assembler::zero, done); 774 __ movptr(rax, Address(rbx, methodOopDesc::const_offset())); 775 __ movptr(rax, Address(rax, constMethodOopDesc::constants_offset())); 776 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes())); 777 __ movptr(rax, Address(rax, mirror_offset)); 778 __ bind(done); 779 } 780 #ifdef ASSERT 781 { Label L; 782 __ cmpptr(rax, Address(monitor, BasicObjectLock::obj_offset_in_bytes())); // correct object? 783 __ jcc(Assembler::equal, L); 784 __ stop("wrong synchronization lobject"); 785 __ bind(L); 786 } 787 #endif // ASSERT 788 // can destroy rax, rdx|c_rarg1, rcx, and (via call_VM) rdi! 789 __ lock_object(monitor); 790 } 791 792 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry 793 794 address InterpreterGenerator::generate_accessor_entry(void) { 795 796 // rbx: methodOop 797 798 // rsi/r13: senderSP must preserved for slow path, set SP to it on fast path 799 800 Label xreturn_path; 801 802 // do fastpath for resolved accessor methods 803 if (UseFastAccessorMethods) { 804 805 address entry_point = __ pc(); 806 807 Label slow_path; 808 // If we need a safepoint check, generate full interpreter entry. 809 ExternalAddress state(SafepointSynchronize::address_of_state()); 810 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 811 SafepointSynchronize::_not_synchronized); 812 813 __ jcc(Assembler::notEqual, slow_path); 814 // ASM/C++ Interpreter 815 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1 816 // Note: We can only use this code if the getfield has been resolved 817 // and if we don't have a null-pointer exception => check for 818 // these conditions first and use slow path if necessary. 819 // rbx,: method 820 // rcx: receiver 821 __ movptr(rax, Address(rsp, wordSize)); 822 823 // check if local 0 != NULL and read field 824 __ testptr(rax, rax); 825 __ jcc(Assembler::zero, slow_path); 826 827 // read first instruction word and extract bytecode @ 1 and index @ 2 828 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset())); 829 __ movptr(rdi, Address(rdx, constMethodOopDesc::constants_offset())); 830 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset())); 831 // Shift codes right to get the index on the right. 832 // The bytecode fetched looks like <index><0xb4><0x2a> 833 __ shrl(rdx, 2*BitsPerByte); 834 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); 835 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes())); 836 837 // rax,: local 0 838 // rbx,: method 839 // rcx: receiver - do not destroy since it is needed for slow path! 840 // rcx: scratch 841 // rdx: constant pool cache index 842 // rdi: constant pool cache 843 // rsi/r13: sender sp 844 845 // check if getfield has been resolved and read constant pool cache entry 846 // check the validity of the cache entry by testing whether _indices field 847 // contains Bytecode::_getfield in b1 byte. 848 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below"); 849 __ movl(rcx, 850 Address(rdi, 851 rdx, 852 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset())); 853 __ shrl(rcx, 2*BitsPerByte); 854 __ andl(rcx, 0xFF); 855 __ cmpl(rcx, Bytecodes::_getfield); 856 __ jcc(Assembler::notEqual, slow_path); 857 858 // Note: constant pool entry is not valid before bytecode is resolved 859 __ movptr(rcx, 860 Address(rdi, 861 rdx, 862 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())); 863 __ movl(rdx, 864 Address(rdi, 865 rdx, 866 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset())); 867 868 Label notByte, notShort, notChar; 869 const Address field_address (rax, rcx, Address::times_1); 870 871 // Need to differentiate between igetfield, agetfield, bgetfield etc. 872 // because they are different sizes. 873 // Use the type from the constant pool cache 874 __ shrl(rdx, ConstantPoolCacheEntry::tosBits); 875 // Make sure we don't need to mask rdx for tosBits after the above shift 876 ConstantPoolCacheEntry::verify_tosBits(); 877 #ifdef _LP64 878 Label notObj; 879 __ cmpl(rdx, atos); 880 __ jcc(Assembler::notEqual, notObj); 881 // atos 882 __ movptr(rax, field_address); 883 __ jmp(xreturn_path); 884 885 __ bind(notObj); 886 #endif // _LP64 887 __ cmpl(rdx, btos); 888 __ jcc(Assembler::notEqual, notByte); 889 __ load_signed_byte(rax, field_address); 890 __ jmp(xreturn_path); 891 892 __ bind(notByte); 893 __ cmpl(rdx, stos); 894 __ jcc(Assembler::notEqual, notShort); 895 __ load_signed_short(rax, field_address); 896 __ jmp(xreturn_path); 897 898 __ bind(notShort); 899 __ cmpl(rdx, ctos); 900 __ jcc(Assembler::notEqual, notChar); 901 __ load_unsigned_short(rax, field_address); 902 __ jmp(xreturn_path); 903 904 __ bind(notChar); 905 #ifdef ASSERT 906 Label okay; 907 #ifndef _LP64 908 __ cmpl(rdx, atos); 909 __ jcc(Assembler::equal, okay); 910 #endif // _LP64 911 __ cmpl(rdx, itos); 912 __ jcc(Assembler::equal, okay); 913 __ stop("what type is this?"); 914 __ bind(okay); 915 #endif // ASSERT 916 // All the rest are a 32 bit wordsize 917 __ movl(rax, field_address); 918 919 __ bind(xreturn_path); 920 921 // _ireturn/_areturn 922 __ pop(rdi); // get return address 923 __ mov(rsp, sender_sp_on_entry); // set sp to sender sp 924 __ jmp(rdi); 925 926 // generate a vanilla interpreter entry as the slow path 927 __ bind(slow_path); 928 // We will enter c++ interpreter looking like it was 929 // called by the call_stub this will cause it to return 930 // a tosca result to the invoker which might have been 931 // the c++ interpreter itself. 932 933 __ jmp(fast_accessor_slow_entry_path); 934 return entry_point; 935 936 } else { 937 return NULL; 938 } 939 940 } 941 942 address InterpreterGenerator::generate_Reference_get_entry(void) { 943 #ifndef SERIALGC 944 if (UseG1GC) { 945 // We need to generate have a routine that generates code to: 946 // * load the value in the referent field 947 // * passes that value to the pre-barrier. 948 // 949 // In the case of G1 this will record the value of the 950 // referent in an SATB buffer if marking is active. 951 // This will cause concurrent marking to mark the referent 952 // field as live. 953 Unimplemented(); 954 } 955 #endif // SERIALGC 956 957 // If G1 is not enabled then attempt to go through the accessor entry point 958 // Reference.get is an accessor 959 return generate_accessor_entry(); 960 } 961 962 // 963 // C++ Interpreter stub for calling a native method. 964 // This sets up a somewhat different looking stack for calling the native method 965 // than the typical interpreter frame setup but still has the pointer to 966 // an interpreter state. 967 // 968 969 address InterpreterGenerator::generate_native_entry(bool synchronized) { 970 // determine code generation flags 971 bool inc_counter = UseCompiler || CountCompiledCalls; 972 973 // rbx: methodOop 974 // rcx: receiver (unused) 975 // rsi/r13: previous interpreter state (if called from C++ interpreter) must preserve 976 // in any case. If called via c1/c2/call_stub rsi/r13 is junk (to use) but harmless 977 // to save/restore. 978 address entry_point = __ pc(); 979 980 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); 981 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset()); 982 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset()); 983 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); 984 985 // rsi/r13 == state/locals rdi == prevstate 986 const Register locals = rdi; 987 988 // get parameter size (always needed) 989 __ load_unsigned_short(rcx, size_of_parameters); 990 991 // rbx: methodOop 992 // rcx: size of parameters 993 __ pop(rax); // get return address 994 // for natives the size of locals is zero 995 996 // compute beginning of parameters /locals 997 __ lea(locals, Address(rsp, rcx, Address::times_ptr, -wordSize)); 998 999 // initialize fixed part of activation frame 1000 1001 // Assumes rax = return address 1002 1003 // allocate and initialize new interpreterState and method expression stack 1004 // IN(locals) -> locals 1005 // IN(state) -> previous frame manager state (NULL from stub/c1/c2) 1006 // destroys rax, rcx, rdx 1007 // OUT (state) -> new interpreterState 1008 // OUT(rsp) -> bottom of methods expression stack 1009 1010 // save sender_sp 1011 __ mov(rcx, sender_sp_on_entry); 1012 // start with NULL previous state 1013 __ movptr(state, (int32_t)NULL_WORD); 1014 generate_compute_interpreter_state(state, locals, rcx, true); 1015 1016 #ifdef ASSERT 1017 { Label L; 1018 __ movptr(rax, STATE(_stack_base)); 1019 #ifdef _LP64 1020 // duplicate the alignment rsp got after setting stack_base 1021 __ subptr(rax, frame::arg_reg_save_area_bytes); // windows 1022 __ andptr(rax, -16); // must be 16 byte boundary (see amd64 ABI) 1023 #endif // _LP64 1024 __ cmpptr(rax, rsp); 1025 __ jcc(Assembler::equal, L); 1026 __ stop("broken stack frame setup in interpreter"); 1027 __ bind(L); 1028 } 1029 #endif 1030 1031 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count 1032 1033 const Register unlock_thread = LP64_ONLY(r15_thread) NOT_LP64(rax); 1034 NOT_LP64(__ movptr(unlock_thread, STATE(_thread));) // get thread 1035 // Since at this point in the method invocation the exception handler 1036 // would try to exit the monitor of synchronized methods which hasn't 1037 // been entered yet, we set the thread local variable 1038 // _do_not_unlock_if_synchronized to true. The remove_activation will 1039 // check this flag. 1040 1041 const Address do_not_unlock_if_synchronized(unlock_thread, 1042 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1043 __ movbool(do_not_unlock_if_synchronized, true); 1044 1045 // make sure method is native & not abstract 1046 #ifdef ASSERT 1047 __ movl(rax, access_flags); 1048 { 1049 Label L; 1050 __ testl(rax, JVM_ACC_NATIVE); 1051 __ jcc(Assembler::notZero, L); 1052 __ stop("tried to execute non-native method as native"); 1053 __ bind(L); 1054 } 1055 { Label L; 1056 __ testl(rax, JVM_ACC_ABSTRACT); 1057 __ jcc(Assembler::zero, L); 1058 __ stop("tried to execute abstract method in interpreter"); 1059 __ bind(L); 1060 } 1061 #endif 1062 1063 1064 // increment invocation count & check for overflow 1065 Label invocation_counter_overflow; 1066 if (inc_counter) { 1067 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); 1068 } 1069 1070 Label continue_after_compile; 1071 1072 __ bind(continue_after_compile); 1073 1074 bang_stack_shadow_pages(true); 1075 1076 // reset the _do_not_unlock_if_synchronized flag 1077 NOT_LP64(__ movl(rax, STATE(_thread));) // get thread 1078 __ movbool(do_not_unlock_if_synchronized, false); 1079 1080 1081 // check for synchronized native methods 1082 // 1083 // Note: This must happen *after* invocation counter check, since 1084 // when overflow happens, the method should not be locked. 1085 if (synchronized) { 1086 // potentially kills rax, rcx, rdx, rdi 1087 lock_method(); 1088 } else { 1089 // no synchronization necessary 1090 #ifdef ASSERT 1091 { Label L; 1092 __ movl(rax, access_flags); 1093 __ testl(rax, JVM_ACC_SYNCHRONIZED); 1094 __ jcc(Assembler::zero, L); 1095 __ stop("method needs synchronization"); 1096 __ bind(L); 1097 } 1098 #endif 1099 } 1100 1101 // start execution 1102 1103 // jvmti support 1104 __ notify_method_entry(); 1105 1106 // work registers 1107 const Register method = rbx; 1108 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rdi); 1109 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); // rcx|rscratch1 1110 1111 // allocate space for parameters 1112 __ movptr(method, STATE(_method)); 1113 __ verify_oop(method); 1114 __ load_unsigned_short(t, Address(method, methodOopDesc::size_of_parameters_offset())); 1115 __ shll(t, 2); 1116 #ifdef _LP64 1117 __ subptr(rsp, t); 1118 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows 1119 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI) 1120 #else 1121 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror 1122 __ subptr(rsp, t); 1123 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics 1124 #endif // _LP64 1125 1126 // get signature handler 1127 Label pending_exception_present; 1128 1129 { Label L; 1130 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); 1131 __ testptr(t, t); 1132 __ jcc(Assembler::notZero, L); 1133 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method, false); 1134 __ movptr(method, STATE(_method)); 1135 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 1136 __ jcc(Assembler::notEqual, pending_exception_present); 1137 __ verify_oop(method); 1138 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); 1139 __ bind(L); 1140 } 1141 #ifdef ASSERT 1142 { 1143 Label L; 1144 __ push(t); 1145 __ get_thread(t); // get vm's javathread* 1146 __ cmpptr(t, STATE(_thread)); 1147 __ jcc(Assembler::equal, L); 1148 __ int3(); 1149 __ bind(L); 1150 __ pop(t); 1151 } 1152 #endif // 1153 1154 const Register from_ptr = InterpreterRuntime::SignatureHandlerGenerator::from(); 1155 // call signature handler 1156 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code"); 1157 1158 // The generated handlers do not touch RBX (the method oop). 1159 // However, large signatures cannot be cached and are generated 1160 // each time here. The slow-path generator will blow RBX 1161 // sometime, so we must reload it after the call. 1162 __ movptr(from_ptr, STATE(_locals)); // get the from pointer 1163 __ call(t); 1164 __ movptr(method, STATE(_method)); 1165 __ verify_oop(method); 1166 1167 // result handler is in rax 1168 // set result handler 1169 __ movptr(STATE(_result_handler), rax); 1170 1171 1172 // get native function entry point 1173 { Label L; 1174 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); 1175 __ testptr(rax, rax); 1176 __ jcc(Assembler::notZero, L); 1177 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method); 1178 __ movptr(method, STATE(_method)); 1179 __ verify_oop(method); 1180 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); 1181 __ bind(L); 1182 } 1183 1184 // pass mirror handle if static call 1185 { Label L; 1186 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 1187 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); 1188 __ testl(t, JVM_ACC_STATIC); 1189 __ jcc(Assembler::zero, L); 1190 // get mirror 1191 __ movptr(t, Address(method, methodOopDesc:: const_offset())); 1192 __ movptr(t, Address(t, constMethodOopDesc::constants_offset())); 1193 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes())); 1194 __ movptr(t, Address(t, mirror_offset)); 1195 // copy mirror into activation object 1196 __ movptr(STATE(_oop_temp), t); 1197 // pass handle to mirror 1198 #ifdef _LP64 1199 __ lea(c_rarg1, STATE(_oop_temp)); 1200 #else 1201 __ lea(t, STATE(_oop_temp)); 1202 __ movptr(Address(rsp, wordSize), t); 1203 #endif // _LP64 1204 __ bind(L); 1205 } 1206 #ifdef ASSERT 1207 { 1208 Label L; 1209 __ push(t); 1210 __ get_thread(t); // get vm's javathread* 1211 __ cmpptr(t, STATE(_thread)); 1212 __ jcc(Assembler::equal, L); 1213 __ int3(); 1214 __ bind(L); 1215 __ pop(t); 1216 } 1217 #endif // 1218 1219 // pass JNIEnv 1220 #ifdef _LP64 1221 __ lea(c_rarg0, Address(thread, JavaThread::jni_environment_offset())); 1222 #else 1223 __ movptr(thread, STATE(_thread)); // get thread 1224 __ lea(t, Address(thread, JavaThread::jni_environment_offset())); 1225 1226 __ movptr(Address(rsp, 0), t); 1227 #endif // _LP64 1228 1229 #ifdef ASSERT 1230 { 1231 Label L; 1232 __ push(t); 1233 __ get_thread(t); // get vm's javathread* 1234 __ cmpptr(t, STATE(_thread)); 1235 __ jcc(Assembler::equal, L); 1236 __ int3(); 1237 __ bind(L); 1238 __ pop(t); 1239 } 1240 #endif // 1241 1242 #ifdef ASSERT 1243 { Label L; 1244 __ movl(t, Address(thread, JavaThread::thread_state_offset())); 1245 __ cmpl(t, _thread_in_Java); 1246 __ jcc(Assembler::equal, L); 1247 __ stop("Wrong thread state in native stub"); 1248 __ bind(L); 1249 } 1250 #endif 1251 1252 // Change state to native (we save the return address in the thread, since it might not 1253 // be pushed on the stack when we do a a stack traversal). It is enough that the pc() 1254 // points into the right code segment. It does not have to be the correct return pc. 1255 1256 __ set_last_Java_frame(thread, noreg, rbp, __ pc()); 1257 1258 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native); 1259 1260 __ call(rax); 1261 1262 // result potentially in rdx:rax or ST0 1263 __ movptr(method, STATE(_method)); 1264 NOT_LP64(__ movptr(thread, STATE(_thread));) // get thread 1265 1266 // The potential result is in ST(0) & rdx:rax 1267 // With C++ interpreter we leave any possible result in ST(0) until we are in result handler and then 1268 // we do the appropriate stuff for returning the result. rdx:rax must always be saved because just about 1269 // anything we do here will destroy it, st(0) is only saved if we re-enter the vm where it would 1270 // be destroyed. 1271 // It is safe to do these pushes because state is _thread_in_native and return address will be found 1272 // via _last_native_pc and not via _last_jave_sp 1273 1274 // Must save the value of ST(0)/xmm0 since it could be destroyed before we get to result handler 1275 { Label Lpush, Lskip; 1276 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT)); 1277 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE)); 1278 __ cmpptr(STATE(_result_handler), float_handler.addr()); 1279 __ jcc(Assembler::equal, Lpush); 1280 __ cmpptr(STATE(_result_handler), double_handler.addr()); 1281 __ jcc(Assembler::notEqual, Lskip); 1282 __ bind(Lpush); 1283 __ subptr(rsp, 2*wordSize); 1284 if ( UseSSE < 2 ) { 1285 __ fstp_d(Address(rsp, 0)); 1286 } else { 1287 __ movdbl(Address(rsp, 0), xmm0); 1288 } 1289 __ bind(Lskip); 1290 } 1291 1292 // save rax:rdx for potential use by result handler. 1293 __ push(rax); 1294 #ifndef _LP64 1295 __ push(rdx); 1296 #endif // _LP64 1297 1298 // Either restore the MXCSR register after returning from the JNI Call 1299 // or verify that it wasn't changed. 1300 if (VM_Version::supports_sse()) { 1301 if (RestoreMXCSROnJNICalls) { 1302 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std())); 1303 } 1304 else if (CheckJNICalls ) { 1305 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry())); 1306 } 1307 } 1308 1309 #ifndef _LP64 1310 // Either restore the x87 floating pointer control word after returning 1311 // from the JNI call or verify that it wasn't changed. 1312 if (CheckJNICalls) { 1313 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry())); 1314 } 1315 #endif // _LP64 1316 1317 1318 // change thread state 1319 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans); 1320 if(os::is_MP()) { 1321 // Write serialization page so VM thread can do a pseudo remote membar. 1322 // We use the current thread pointer to calculate a thread specific 1323 // offset to write to within the page. This minimizes bus traffic 1324 // due to cache line collision. 1325 __ serialize_memory(thread, rcx); 1326 } 1327 1328 // check for safepoint operation in progress and/or pending suspend requests 1329 { Label Continue; 1330 1331 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), 1332 SafepointSynchronize::_not_synchronized); 1333 1334 // threads running native code and they are expected to self-suspend 1335 // when leaving the _thread_in_native state. We need to check for 1336 // pending suspend requests here. 1337 Label L; 1338 __ jcc(Assembler::notEqual, L); 1339 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0); 1340 __ jcc(Assembler::equal, Continue); 1341 __ bind(L); 1342 1343 // Don't use call_VM as it will see a possible pending exception and forward it 1344 // and never return here preventing us from clearing _last_native_pc down below. 1345 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are 1346 // preserved and correspond to the bcp/locals pointers. 1347 // 1348 1349 ((MacroAssembler*)_masm)->call_VM_leaf(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), 1350 thread); 1351 __ increment(rsp, wordSize); 1352 1353 __ movptr(method, STATE(_method)); 1354 __ verify_oop(method); 1355 __ movptr(thread, STATE(_thread)); // get thread 1356 1357 __ bind(Continue); 1358 } 1359 1360 // change thread state 1361 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java); 1362 1363 __ reset_last_Java_frame(thread, true, true); 1364 1365 // reset handle block 1366 __ movptr(t, Address(thread, JavaThread::active_handles_offset())); 1367 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD); 1368 1369 // If result was an oop then unbox and save it in the frame 1370 { Label L; 1371 Label no_oop, store_result; 1372 ExternalAddress oop_handler(AbstractInterpreter::result_handler(T_OBJECT)); 1373 __ cmpptr(STATE(_result_handler), oop_handler.addr()); 1374 __ jcc(Assembler::notEqual, no_oop); 1375 #ifndef _LP64 1376 __ pop(rdx); 1377 #endif // _LP64 1378 __ pop(rax); 1379 __ testptr(rax, rax); 1380 __ jcc(Assembler::zero, store_result); 1381 // unbox 1382 __ movptr(rax, Address(rax, 0)); 1383 __ bind(store_result); 1384 __ movptr(STATE(_oop_temp), rax); 1385 // keep stack depth as expected by pushing oop which will eventually be discarded 1386 __ push(rax); 1387 #ifndef _LP64 1388 __ push(rdx); 1389 #endif // _LP64 1390 __ bind(no_oop); 1391 } 1392 1393 { 1394 Label no_reguard; 1395 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled); 1396 __ jcc(Assembler::notEqual, no_reguard); 1397 1398 __ pusha(); 1399 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); 1400 __ popa(); 1401 1402 __ bind(no_reguard); 1403 } 1404 1405 1406 // QQQ Seems like for native methods we simply return and the caller will see the pending 1407 // exception and do the right thing. Certainly the interpreter will, don't know about 1408 // compiled methods. 1409 // Seems that the answer to above is no this is wrong. The old code would see the exception 1410 // and forward it before doing the unlocking and notifying jvmdi that method has exited. 1411 // This seems wrong need to investigate the spec. 1412 1413 // handle exceptions (exception handling will handle unlocking!) 1414 { Label L; 1415 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 1416 __ jcc(Assembler::zero, L); 1417 __ bind(pending_exception_present); 1418 1419 // There are potential results on the stack (rax/rdx, ST(0)) we ignore these and simply 1420 // return and let caller deal with exception. This skips the unlocking here which 1421 // seems wrong but seems to be what asm interpreter did. Can't find this in the spec. 1422 // Note: must preverve method in rbx 1423 // 1424 1425 // remove activation 1426 1427 __ movptr(t, STATE(_sender_sp)); 1428 __ leave(); // remove frame anchor 1429 __ pop(rdi); // get return address 1430 __ movptr(state, STATE(_prev_link)); // get previous state for return 1431 __ mov(rsp, t); // set sp to sender sp 1432 __ push(rdi); // push throwing pc 1433 // The skips unlocking!! This seems to be what asm interpreter does but seems 1434 // very wrong. Not clear if this violates the spec. 1435 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 1436 __ bind(L); 1437 } 1438 1439 // do unlocking if necessary 1440 { Label L; 1441 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); 1442 __ testl(t, JVM_ACC_SYNCHRONIZED); 1443 __ jcc(Assembler::zero, L); 1444 // the code below should be shared with interpreter macro assembler implementation 1445 { Label unlock; 1446 const Register monitor = NOT_LP64(rdx) LP64_ONLY(c_rarg1); 1447 // BasicObjectLock will be first in list, since this is a synchronized method. However, need 1448 // to check that the object has not been unlocked by an explicit monitorexit bytecode. 1449 __ movptr(monitor, STATE(_monitor_base)); 1450 __ subptr(monitor, frame::interpreter_frame_monitor_size() * wordSize); // address of initial monitor 1451 1452 __ movptr(t, Address(monitor, BasicObjectLock::obj_offset_in_bytes())); 1453 __ testptr(t, t); 1454 __ jcc(Assembler::notZero, unlock); 1455 1456 // Entry already unlocked, need to throw exception 1457 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1458 __ should_not_reach_here(); 1459 1460 __ bind(unlock); 1461 __ unlock_object(monitor); 1462 // unlock can blow rbx so restore it for path that needs it below 1463 __ movptr(method, STATE(_method)); 1464 } 1465 __ bind(L); 1466 } 1467 1468 // jvmti support 1469 // Note: This must happen _after_ handling/throwing any exceptions since 1470 // the exception handler code notifies the runtime of method exits 1471 // too. If this happens before, method entry/exit notifications are 1472 // not properly paired (was bug - gri 11/22/99). 1473 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); 1474 1475 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result 1476 #ifndef _LP64 1477 __ pop(rdx); 1478 #endif // _LP64 1479 __ pop(rax); 1480 __ movptr(t, STATE(_result_handler)); // get result handler 1481 __ call(t); // call result handler to convert to tosca form 1482 1483 // remove activation 1484 1485 __ movptr(t, STATE(_sender_sp)); 1486 1487 __ leave(); // remove frame anchor 1488 __ pop(rdi); // get return address 1489 __ movptr(state, STATE(_prev_link)); // get previous state for return (if c++ interpreter was caller) 1490 __ mov(rsp, t); // set sp to sender sp 1491 __ jmp(rdi); 1492 1493 // invocation counter overflow 1494 if (inc_counter) { 1495 // Handle overflow of counter and compile method 1496 __ bind(invocation_counter_overflow); 1497 generate_counter_overflow(&continue_after_compile); 1498 } 1499 1500 return entry_point; 1501 } 1502 1503 // Generate entries that will put a result type index into rcx 1504 void CppInterpreterGenerator::generate_deopt_handling() { 1505 1506 Label return_from_deopt_common; 1507 1508 // Generate entries that will put a result type index into rcx 1509 // deopt needs to jump to here to enter the interpreter (return a result) 1510 deopt_frame_manager_return_atos = __ pc(); 1511 1512 // rax is live here 1513 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_OBJECT)); // Result stub address array index 1514 __ jmp(return_from_deopt_common); 1515 1516 1517 // deopt needs to jump to here to enter the interpreter (return a result) 1518 deopt_frame_manager_return_btos = __ pc(); 1519 1520 // rax is live here 1521 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_BOOLEAN)); // Result stub address array index 1522 __ jmp(return_from_deopt_common); 1523 1524 // deopt needs to jump to here to enter the interpreter (return a result) 1525 deopt_frame_manager_return_itos = __ pc(); 1526 1527 // rax is live here 1528 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_INT)); // Result stub address array index 1529 __ jmp(return_from_deopt_common); 1530 1531 // deopt needs to jump to here to enter the interpreter (return a result) 1532 1533 deopt_frame_manager_return_ltos = __ pc(); 1534 // rax,rdx are live here 1535 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_LONG)); // Result stub address array index 1536 __ jmp(return_from_deopt_common); 1537 1538 // deopt needs to jump to here to enter the interpreter (return a result) 1539 1540 deopt_frame_manager_return_ftos = __ pc(); 1541 // st(0) is live here 1542 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index 1543 __ jmp(return_from_deopt_common); 1544 1545 // deopt needs to jump to here to enter the interpreter (return a result) 1546 deopt_frame_manager_return_dtos = __ pc(); 1547 1548 // st(0) is live here 1549 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index 1550 __ jmp(return_from_deopt_common); 1551 1552 // deopt needs to jump to here to enter the interpreter (return a result) 1553 deopt_frame_manager_return_vtos = __ pc(); 1554 1555 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_VOID)); 1556 1557 // Deopt return common 1558 // an index is present in rcx that lets us move any possible result being 1559 // return to the interpreter's stack 1560 // 1561 // Because we have a full sized interpreter frame on the youngest 1562 // activation the stack is pushed too deep to share the tosca to 1563 // stack converters directly. We shrink the stack to the desired 1564 // amount and then push result and then re-extend the stack. 1565 // We could have the code in size_activation layout a short 1566 // frame for the top activation but that would look different 1567 // than say sparc (which needs a full size activation because 1568 // the windows are in the way. Really it could be short? QQQ 1569 // 1570 __ bind(return_from_deopt_common); 1571 1572 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); 1573 1574 // setup rsp so we can push the "result" as needed. 1575 __ movptr(rsp, STATE(_stack)); // trim stack (is prepushed) 1576 __ addptr(rsp, wordSize); // undo prepush 1577 1578 ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack); 1579 // Address index(noreg, rcx, Address::times_ptr); 1580 __ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_ptr))); 1581 // __ movl(rcx, Address(noreg, rcx, Address::times_ptr, int(AbstractInterpreter::_tosca_to_stack))); 1582 __ call(rcx); // call result converter 1583 1584 __ movl(STATE(_msg), (int)BytecodeInterpreter::deopt_resume); 1585 __ lea(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present) 1586 __ movptr(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed, 1587 // result if any on stack already ) 1588 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth 1589 } 1590 1591 // Generate the code to handle a more_monitors message from the c++ interpreter 1592 void CppInterpreterGenerator::generate_more_monitors() { 1593 1594 1595 Label entry, loop; 1596 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 1597 // 1. compute new pointers // rsp: old expression stack top 1598 __ movptr(rdx, STATE(_stack_base)); // rdx: old expression stack bottom 1599 __ subptr(rsp, entry_size); // move expression stack top limit 1600 __ subptr(STATE(_stack), entry_size); // update interpreter stack top 1601 __ subptr(STATE(_stack_limit), entry_size); // inform interpreter 1602 __ subptr(rdx, entry_size); // move expression stack bottom 1603 __ movptr(STATE(_stack_base), rdx); // inform interpreter 1604 __ movptr(rcx, STATE(_stack)); // set start value for copy loop 1605 __ jmp(entry); 1606 // 2. move expression stack contents 1607 __ bind(loop); 1608 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location 1609 __ movptr(Address(rcx, 0), rbx); // and store it at new location 1610 __ addptr(rcx, wordSize); // advance to next word 1611 __ bind(entry); 1612 __ cmpptr(rcx, rdx); // check if bottom reached 1613 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word 1614 // now zero the slot so we can find it. 1615 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD); 1616 __ movl(STATE(_msg), (int)BytecodeInterpreter::got_monitors); 1617 } 1618 1619 1620 // Initial entry to C++ interpreter from the call_stub. 1621 // This entry point is called the frame manager since it handles the generation 1622 // of interpreter activation frames via requests directly from the vm (via call_stub) 1623 // and via requests from the interpreter. The requests from the call_stub happen 1624 // directly thru the entry point. Requests from the interpreter happen via returning 1625 // from the interpreter and examining the message the interpreter has returned to 1626 // the frame manager. The frame manager can take the following requests: 1627 1628 // NO_REQUEST - error, should never happen. 1629 // MORE_MONITORS - need a new monitor. Shuffle the expression stack on down and 1630 // allocate a new monitor. 1631 // CALL_METHOD - setup a new activation to call a new method. Very similar to what 1632 // happens during entry during the entry via the call stub. 1633 // RETURN_FROM_METHOD - remove an activation. Return to interpreter or call stub. 1634 // 1635 // Arguments: 1636 // 1637 // rbx: methodOop 1638 // rcx: receiver - unused (retrieved from stack as needed) 1639 // rsi/r13: previous frame manager state (NULL from the call_stub/c1/c2) 1640 // 1641 // 1642 // Stack layout at entry 1643 // 1644 // [ return address ] <--- rsp 1645 // [ parameter n ] 1646 // ... 1647 // [ parameter 1 ] 1648 // [ expression stack ] 1649 // 1650 // 1651 // We are free to blow any registers we like because the call_stub which brought us here 1652 // initially has preserved the callee save registers already. 1653 // 1654 // 1655 1656 static address interpreter_frame_manager = NULL; 1657 1658 address InterpreterGenerator::generate_normal_entry(bool synchronized) { 1659 1660 // rbx: methodOop 1661 // rsi/r13: sender sp 1662 1663 // Because we redispatch "recursive" interpreter entries thru this same entry point 1664 // the "input" register usage is a little strange and not what you expect coming 1665 // from the call_stub. From the call stub rsi/rdi (current/previous) interpreter 1666 // state are NULL but on "recursive" dispatches they are what you'd expect. 1667 // rsi: current interpreter state (C++ interpreter) must preserve (null from call_stub/c1/c2) 1668 1669 1670 // A single frame manager is plenty as we don't specialize for synchronized. We could and 1671 // the code is pretty much ready. Would need to change the test below and for good measure 1672 // modify generate_interpreter_state to only do the (pre) sync stuff stuff for synchronized 1673 // routines. Not clear this is worth it yet. 1674 1675 if (interpreter_frame_manager) return interpreter_frame_manager; 1676 1677 address entry_point = __ pc(); 1678 1679 // Fast accessor methods share this entry point. 1680 // This works because frame manager is in the same codelet 1681 if (UseFastAccessorMethods && !synchronized) __ bind(fast_accessor_slow_entry_path); 1682 1683 Label dispatch_entry_2; 1684 __ movptr(rcx, sender_sp_on_entry); 1685 __ movptr(state, (int32_t)NULL_WORD); // no current activation 1686 1687 __ jmp(dispatch_entry_2); 1688 1689 const Register locals = rdi; 1690 1691 Label re_dispatch; 1692 1693 __ bind(re_dispatch); 1694 1695 // save sender sp (doesn't include return address 1696 __ lea(rcx, Address(rsp, wordSize)); 1697 1698 __ bind(dispatch_entry_2); 1699 1700 // save sender sp 1701 __ push(rcx); 1702 1703 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); 1704 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset()); 1705 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); 1706 1707 // const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 1708 // const Address monitor_block_bot (rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 1709 // const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock)); 1710 1711 // get parameter size (always needed) 1712 __ load_unsigned_short(rcx, size_of_parameters); 1713 1714 // rbx: methodOop 1715 // rcx: size of parameters 1716 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words 1717 1718 __ subptr(rdx, rcx); // rdx = no. of additional locals 1719 1720 // see if we've got enough room on the stack for locals plus overhead. 1721 generate_stack_overflow_check(); // C++ 1722 1723 // c++ interpreter does not use stack banging or any implicit exceptions 1724 // leave for now to verify that check is proper. 1725 bang_stack_shadow_pages(false); 1726 1727 1728 1729 // compute beginning of parameters (rdi) 1730 __ lea(locals, Address(rsp, rcx, Address::times_ptr, wordSize)); 1731 1732 // save sender's sp 1733 // __ movl(rcx, rsp); 1734 1735 // get sender's sp 1736 __ pop(rcx); 1737 1738 // get return address 1739 __ pop(rax); 1740 1741 // rdx - # of additional locals 1742 // allocate space for locals 1743 // explicitly initialize locals 1744 { 1745 Label exit, loop; 1746 __ testl(rdx, rdx); // (32bit ok) 1747 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0 1748 __ bind(loop); 1749 __ push((int32_t)NULL_WORD); // initialize local variables 1750 __ decrement(rdx); // until everything initialized 1751 __ jcc(Assembler::greater, loop); 1752 __ bind(exit); 1753 } 1754 1755 1756 // Assumes rax = return address 1757 1758 // allocate and initialize new interpreterState and method expression stack 1759 // IN(locals) -> locals 1760 // IN(state) -> any current interpreter activation 1761 // destroys rax, rcx, rdx, rdi 1762 // OUT (state) -> new interpreterState 1763 // OUT(rsp) -> bottom of methods expression stack 1764 1765 generate_compute_interpreter_state(state, locals, rcx, false); 1766 1767 // Call interpreter 1768 1769 Label call_interpreter; 1770 __ bind(call_interpreter); 1771 1772 // c++ interpreter does not use stack banging or any implicit exceptions 1773 // leave for now to verify that check is proper. 1774 bang_stack_shadow_pages(false); 1775 1776 1777 // Call interpreter enter here if message is 1778 // set and we know stack size is valid 1779 1780 Label call_interpreter_2; 1781 1782 __ bind(call_interpreter_2); 1783 1784 { 1785 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread); 1786 1787 #ifdef _LP64 1788 __ mov(c_rarg0, state); 1789 #else 1790 __ push(state); // push arg to interpreter 1791 __ movptr(thread, STATE(_thread)); 1792 #endif // _LP64 1793 1794 // We can setup the frame anchor with everything we want at this point 1795 // as we are thread_in_Java and no safepoints can occur until we go to 1796 // vm mode. We do have to clear flags on return from vm but that is it 1797 // 1798 __ movptr(Address(thread, JavaThread::last_Java_fp_offset()), rbp); 1799 __ movptr(Address(thread, JavaThread::last_Java_sp_offset()), rsp); 1800 1801 // Call the interpreter 1802 1803 RuntimeAddress normal(CAST_FROM_FN_PTR(address, BytecodeInterpreter::run)); 1804 RuntimeAddress checking(CAST_FROM_FN_PTR(address, BytecodeInterpreter::runWithChecks)); 1805 1806 __ call(JvmtiExport::can_post_interpreter_events() ? checking : normal); 1807 NOT_LP64(__ pop(rax);) // discard parameter to run 1808 // 1809 // state is preserved since it is callee saved 1810 // 1811 1812 // reset_last_Java_frame 1813 1814 NOT_LP64(__ movl(thread, STATE(_thread));) 1815 __ reset_last_Java_frame(thread, true, true); 1816 } 1817 1818 // examine msg from interpreter to determine next action 1819 1820 __ movl(rdx, STATE(_msg)); // Get new message 1821 1822 Label call_method; 1823 Label return_from_interpreted_method; 1824 Label throw_exception; 1825 Label bad_msg; 1826 Label do_OSR; 1827 1828 __ cmpl(rdx, (int32_t)BytecodeInterpreter::call_method); 1829 __ jcc(Assembler::equal, call_method); 1830 __ cmpl(rdx, (int32_t)BytecodeInterpreter::return_from_method); 1831 __ jcc(Assembler::equal, return_from_interpreted_method); 1832 __ cmpl(rdx, (int32_t)BytecodeInterpreter::do_osr); 1833 __ jcc(Assembler::equal, do_OSR); 1834 __ cmpl(rdx, (int32_t)BytecodeInterpreter::throwing_exception); 1835 __ jcc(Assembler::equal, throw_exception); 1836 __ cmpl(rdx, (int32_t)BytecodeInterpreter::more_monitors); 1837 __ jcc(Assembler::notEqual, bad_msg); 1838 1839 // Allocate more monitor space, shuffle expression stack.... 1840 1841 generate_more_monitors(); 1842 1843 __ jmp(call_interpreter); 1844 1845 // uncommon trap needs to jump to here to enter the interpreter (re-execute current bytecode) 1846 unctrap_frame_manager_entry = __ pc(); 1847 // 1848 // Load the registers we need. 1849 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); 1850 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth 1851 __ jmp(call_interpreter_2); 1852 1853 1854 1855 //============================================================================= 1856 // Returning from a compiled method into a deopted method. The bytecode at the 1857 // bcp has completed. The result of the bytecode is in the native abi (the tosca 1858 // for the template based interpreter). Any stack space that was used by the 1859 // bytecode that has completed has been removed (e.g. parameters for an invoke) 1860 // so all that we have to do is place any pending result on the expression stack 1861 // and resume execution on the next bytecode. 1862 1863 1864 generate_deopt_handling(); 1865 __ jmp(call_interpreter); 1866 1867 1868 // Current frame has caught an exception we need to dispatch to the 1869 // handler. We can get here because a native interpreter frame caught 1870 // an exception in which case there is no handler and we must rethrow 1871 // If it is a vanilla interpreted frame the we simply drop into the 1872 // interpreter and let it do the lookup. 1873 1874 Interpreter::_rethrow_exception_entry = __ pc(); 1875 // rax: exception 1876 // rdx: return address/pc that threw exception 1877 1878 Label return_with_exception; 1879 Label unwind_and_forward; 1880 1881 // restore state pointer. 1882 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); 1883 1884 __ movptr(rbx, STATE(_method)); // get method 1885 #ifdef _LP64 1886 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax); 1887 #else 1888 __ movl(rcx, STATE(_thread)); // get thread 1889 1890 // Store exception with interpreter will expect it 1891 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax); 1892 #endif // _LP64 1893 1894 // is current frame vanilla or native? 1895 1896 __ movl(rdx, access_flags); 1897 __ testl(rdx, JVM_ACC_NATIVE); 1898 __ jcc(Assembler::zero, return_with_exception); // vanilla interpreted frame, handle directly 1899 1900 // We drop thru to unwind a native interpreted frame with a pending exception 1901 // We jump here for the initial interpreter frame with exception pending 1902 // We unwind the current acivation and forward it to our caller. 1903 1904 __ bind(unwind_and_forward); 1905 1906 // unwind rbp, return stack to unextended value and re-push return address 1907 1908 __ movptr(rcx, STATE(_sender_sp)); 1909 __ leave(); 1910 __ pop(rdx); 1911 __ mov(rsp, rcx); 1912 __ push(rdx); 1913 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 1914 1915 // Return point from a call which returns a result in the native abi 1916 // (c1/c2/jni-native). This result must be processed onto the java 1917 // expression stack. 1918 // 1919 // A pending exception may be present in which case there is no result present 1920 1921 Label resume_interpreter; 1922 Label do_float; 1923 Label do_double; 1924 Label done_conv; 1925 1926 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases 1927 if (UseSSE < 2) { 1928 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); 1929 __ movptr(rbx, STATE(_result._to_call._callee)); // get method just executed 1930 __ movl(rcx, Address(rbx, methodOopDesc::result_index_offset())); 1931 __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index 1932 __ jcc(Assembler::equal, do_float); 1933 __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index 1934 __ jcc(Assembler::equal, do_double); 1935 #if !defined(_LP64) || defined(COMPILER1) || !defined(COMPILER2) 1936 __ empty_FPU_stack(); 1937 #endif // COMPILER2 1938 __ jmp(done_conv); 1939 1940 __ bind(do_float); 1941 #ifdef COMPILER2 1942 for (int i = 1; i < 8; i++) { 1943 __ ffree(i); 1944 } 1945 #endif // COMPILER2 1946 __ jmp(done_conv); 1947 __ bind(do_double); 1948 #ifdef COMPILER2 1949 for (int i = 1; i < 8; i++) { 1950 __ ffree(i); 1951 } 1952 #endif // COMPILER2 1953 __ jmp(done_conv); 1954 } else { 1955 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled"); 1956 __ jmp(done_conv); 1957 } 1958 1959 // Return point to interpreter from compiled/native method 1960 InternalAddress return_from_native_method(__ pc()); 1961 1962 __ bind(done_conv); 1963 1964 1965 // Result if any is in tosca. The java expression stack is in the state that the 1966 // calling convention left it (i.e. params may or may not be present) 1967 // Copy the result from tosca and place it on java expression stack. 1968 1969 // Restore rsi/r13 as compiled code may not preserve it 1970 1971 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); 1972 1973 // restore stack to what we had when we left (in case i2c extended it) 1974 1975 __ movptr(rsp, STATE(_stack)); 1976 __ lea(rsp, Address(rsp, wordSize)); 1977 1978 // If there is a pending exception then we don't really have a result to process 1979 1980 #ifdef _LP64 1981 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 1982 #else 1983 __ movptr(rcx, STATE(_thread)); // get thread 1984 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 1985 #endif // _LP64 1986 __ jcc(Assembler::notZero, return_with_exception); 1987 1988 // get method just executed 1989 __ movptr(rbx, STATE(_result._to_call._callee)); 1990 1991 // callee left args on top of expression stack, remove them 1992 __ load_unsigned_short(rcx, Address(rbx, methodOopDesc::size_of_parameters_offset())); 1993 __ lea(rsp, Address(rsp, rcx, Address::times_ptr)); 1994 1995 __ movl(rcx, Address(rbx, methodOopDesc::result_index_offset())); 1996 ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack); 1997 // Address index(noreg, rax, Address::times_ptr); 1998 __ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_ptr))); 1999 // __ movl(rcx, Address(noreg, rcx, Address::times_ptr, int(AbstractInterpreter::_tosca_to_stack))); 2000 __ call(rcx); // call result converter 2001 __ jmp(resume_interpreter); 2002 2003 // An exception is being caught on return to a vanilla interpreter frame. 2004 // Empty the stack and resume interpreter 2005 2006 __ bind(return_with_exception); 2007 2008 // Exception present, empty stack 2009 __ movptr(rsp, STATE(_stack_base)); 2010 __ jmp(resume_interpreter); 2011 2012 // Return from interpreted method we return result appropriate to the caller (i.e. "recursive" 2013 // interpreter call, or native) and unwind this interpreter activation. 2014 // All monitors should be unlocked. 2015 2016 __ bind(return_from_interpreted_method); 2017 2018 Label return_to_initial_caller; 2019 2020 __ movptr(rbx, STATE(_method)); // get method just executed 2021 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from "recursive" interpreter call? 2022 __ movl(rax, Address(rbx, methodOopDesc::result_index_offset())); // get result type index 2023 __ jcc(Assembler::equal, return_to_initial_caller); // back to native code (call_stub/c1/c2) 2024 2025 // Copy result to callers java stack 2026 ExternalAddress stack_to_stack((address)CppInterpreter::_stack_to_stack); 2027 // Address index(noreg, rax, Address::times_ptr); 2028 2029 __ movptr(rax, ArrayAddress(stack_to_stack, Address(noreg, rax, Address::times_ptr))); 2030 // __ movl(rax, Address(noreg, rax, Address::times_ptr, int(AbstractInterpreter::_stack_to_stack))); 2031 __ call(rax); // call result converter 2032 2033 Label unwind_recursive_activation; 2034 __ bind(unwind_recursive_activation); 2035 2036 // returning to interpreter method from "recursive" interpreter call 2037 // result converter left rax pointing to top of the java stack for method we are returning 2038 // to. Now all we must do is unwind the state from the completed call 2039 2040 __ movptr(state, STATE(_prev_link)); // unwind state 2041 __ leave(); // pop the frame 2042 __ mov(rsp, rax); // unwind stack to remove args 2043 2044 // Resume the interpreter. The current frame contains the current interpreter 2045 // state object. 2046 // 2047 2048 __ bind(resume_interpreter); 2049 2050 // state == interpreterState object for method we are resuming 2051 2052 __ movl(STATE(_msg), (int)BytecodeInterpreter::method_resume); 2053 __ lea(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present) 2054 __ movptr(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed, 2055 // result if any on stack already ) 2056 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth 2057 __ jmp(call_interpreter_2); // No need to bang 2058 2059 // interpreter returning to native code (call_stub/c1/c2) 2060 // convert result and unwind initial activation 2061 // rax - result index 2062 2063 __ bind(return_to_initial_caller); 2064 ExternalAddress stack_to_native((address)CppInterpreter::_stack_to_native_abi); 2065 // Address index(noreg, rax, Address::times_ptr); 2066 2067 __ movptr(rax, ArrayAddress(stack_to_native, Address(noreg, rax, Address::times_ptr))); 2068 __ call(rax); // call result converter 2069 2070 Label unwind_initial_activation; 2071 __ bind(unwind_initial_activation); 2072 2073 // RETURN TO CALL_STUB/C1/C2 code (result if any in rax/rdx ST(0)) 2074 2075 /* Current stack picture 2076 2077 [ incoming parameters ] 2078 [ extra locals ] 2079 [ return address to CALL_STUB/C1/C2] 2080 fp -> [ CALL_STUB/C1/C2 fp ] 2081 BytecodeInterpreter object 2082 expression stack 2083 sp -> 2084 2085 */ 2086 2087 // return restoring the stack to the original sender_sp value 2088 2089 __ movptr(rcx, STATE(_sender_sp)); 2090 __ leave(); 2091 __ pop(rdi); // get return address 2092 // set stack to sender's sp 2093 __ mov(rsp, rcx); 2094 __ jmp(rdi); // return to call_stub 2095 2096 // OSR request, adjust return address to make current frame into adapter frame 2097 // and enter OSR nmethod 2098 2099 __ bind(do_OSR); 2100 2101 Label remove_initial_frame; 2102 2103 // We are going to pop this frame. Is there another interpreter frame underneath 2104 // it or is it callstub/compiled? 2105 2106 // Move buffer to the expected parameter location 2107 __ movptr(rcx, STATE(_result._osr._osr_buf)); 2108 2109 __ movptr(rax, STATE(_result._osr._osr_entry)); 2110 2111 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from "recursive" interpreter call? 2112 __ jcc(Assembler::equal, remove_initial_frame); // back to native code (call_stub/c1/c2) 2113 2114 __ movptr(sender_sp_on_entry, STATE(_sender_sp)); // get sender's sp in expected register 2115 __ leave(); // pop the frame 2116 __ mov(rsp, sender_sp_on_entry); // trim any stack expansion 2117 2118 2119 // We know we are calling compiled so push specialized return 2120 // method uses specialized entry, push a return so we look like call stub setup 2121 // this path will handle fact that result is returned in registers and not 2122 // on the java stack. 2123 2124 __ pushptr(return_from_native_method.addr()); 2125 2126 __ jmp(rax); 2127 2128 __ bind(remove_initial_frame); 2129 2130 __ movptr(rdx, STATE(_sender_sp)); 2131 __ leave(); 2132 // get real return 2133 __ pop(rsi); 2134 // set stack to sender's sp 2135 __ mov(rsp, rdx); 2136 // repush real return 2137 __ push(rsi); 2138 // Enter OSR nmethod 2139 __ jmp(rax); 2140 2141 2142 2143 2144 // Call a new method. All we do is (temporarily) trim the expression stack 2145 // push a return address to bring us back to here and leap to the new entry. 2146 2147 __ bind(call_method); 2148 2149 // stack points to next free location and not top element on expression stack 2150 // method expects sp to be pointing to topmost element 2151 2152 __ movptr(rsp, STATE(_stack)); // pop args to c++ interpreter, set sp to java stack top 2153 __ lea(rsp, Address(rsp, wordSize)); 2154 2155 __ movptr(rbx, STATE(_result._to_call._callee)); // get method to execute 2156 2157 // don't need a return address if reinvoking interpreter 2158 2159 // Make it look like call_stub calling conventions 2160 2161 // Get (potential) receiver 2162 __ load_unsigned_short(rcx, size_of_parameters); // get size of parameters in words 2163 2164 ExternalAddress recursive(CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation)); 2165 __ pushptr(recursive.addr()); // make it look good in the debugger 2166 2167 InternalAddress entry(entry_point); 2168 __ cmpptr(STATE(_result._to_call._callee_entry_point), entry.addr()); // returning to interpreter? 2169 __ jcc(Assembler::equal, re_dispatch); // yes 2170 2171 __ pop(rax); // pop dummy address 2172 2173 2174 // get specialized entry 2175 __ movptr(rax, STATE(_result._to_call._callee_entry_point)); 2176 // set sender SP 2177 __ mov(sender_sp_on_entry, rsp); 2178 2179 // method uses specialized entry, push a return so we look like call stub setup 2180 // this path will handle fact that result is returned in registers and not 2181 // on the java stack. 2182 2183 __ pushptr(return_from_native_method.addr()); 2184 2185 __ jmp(rax); 2186 2187 __ bind(bad_msg); 2188 __ stop("Bad message from interpreter"); 2189 2190 // Interpreted method "returned" with an exception pass it on... 2191 // Pass result, unwind activation and continue/return to interpreter/call_stub 2192 // We handle result (if any) differently based on return to interpreter or call_stub 2193 2194 Label unwind_initial_with_pending_exception; 2195 2196 __ bind(throw_exception); 2197 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from recursive interpreter call? 2198 __ jcc(Assembler::equal, unwind_initial_with_pending_exception); // no, back to native code (call_stub/c1/c2) 2199 __ movptr(rax, STATE(_locals)); // pop parameters get new stack value 2200 __ addptr(rax, wordSize); // account for prepush before we return 2201 __ jmp(unwind_recursive_activation); 2202 2203 __ bind(unwind_initial_with_pending_exception); 2204 2205 // We will unwind the current (initial) interpreter frame and forward 2206 // the exception to the caller. We must put the exception in the 2207 // expected register and clear pending exception and then forward. 2208 2209 __ jmp(unwind_and_forward); 2210 2211 interpreter_frame_manager = entry_point; 2212 return entry_point; 2213 } 2214 2215 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) { 2216 // determine code generation flags 2217 bool synchronized = false; 2218 address entry_point = NULL; 2219 2220 switch (kind) { 2221 case Interpreter::zerolocals : break; 2222 case Interpreter::zerolocals_synchronized: synchronized = true; break; 2223 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break; 2224 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break; 2225 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break; 2226 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break; 2227 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break; 2228 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break; 2229 2230 case Interpreter::java_lang_math_sin : // fall thru 2231 case Interpreter::java_lang_math_cos : // fall thru 2232 case Interpreter::java_lang_math_tan : // fall thru 2233 case Interpreter::java_lang_math_abs : // fall thru 2234 case Interpreter::java_lang_math_log : // fall thru 2235 case Interpreter::java_lang_math_log10 : // fall thru 2236 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break; 2237 case Interpreter::java_lang_ref_reference_get 2238 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break; 2239 default : ShouldNotReachHere(); break; 2240 } 2241 2242 if (entry_point) return entry_point; 2243 2244 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized); 2245 2246 } 2247 2248 InterpreterGenerator::InterpreterGenerator(StubQueue* code) 2249 : CppInterpreterGenerator(code) { 2250 generate_all(); // down here so it can be "virtual" 2251 } 2252 2253 // Deoptimization helpers for C++ interpreter 2254 2255 // How much stack a method activation needs in words. 2256 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) { 2257 2258 const int stub_code = 4; // see generate_call_stub 2259 // Save space for one monitor to get into the interpreted method in case 2260 // the method is synchronized 2261 int monitor_size = method->is_synchronized() ? 2262 1*frame::interpreter_frame_monitor_size() : 0; 2263 2264 // total static overhead size. Account for interpreter state object, return 2265 // address, saved rbp and 2 words for a "static long no_params() method" issue. 2266 2267 const int overhead_size = sizeof(BytecodeInterpreter)/wordSize + 2268 ( frame::sender_sp_offset - frame::link_offset) + 2; 2269 2270 const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries(); 2271 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) * 2272 Interpreter::stackElementWords(); 2273 return overhead_size + method_stack + stub_code; 2274 } 2275 2276 // returns the activation size. 2277 static int size_activation_helper(int extra_locals_size, int monitor_size) { 2278 return (extra_locals_size + // the addition space for locals 2279 2*BytesPerWord + // return address and saved rbp 2280 2*BytesPerWord + // "static long no_params() method" issue 2281 sizeof(BytecodeInterpreter) + // interpreterState 2282 monitor_size); // monitors 2283 } 2284 2285 void BytecodeInterpreter::layout_interpreterState(interpreterState to_fill, 2286 frame* caller, 2287 frame* current, 2288 methodOop method, 2289 intptr_t* locals, 2290 intptr_t* stack, 2291 intptr_t* stack_base, 2292 intptr_t* monitor_base, 2293 intptr_t* frame_bottom, 2294 bool is_top_frame 2295 ) 2296 { 2297 // What about any vtable? 2298 // 2299 to_fill->_thread = JavaThread::current(); 2300 // This gets filled in later but make it something recognizable for now 2301 to_fill->_bcp = method->code_base(); 2302 to_fill->_locals = locals; 2303 to_fill->_constants = method->constants()->cache(); 2304 to_fill->_method = method; 2305 to_fill->_mdx = NULL; 2306 to_fill->_stack = stack; 2307 if (is_top_frame && JavaThread::current()->popframe_forcing_deopt_reexecution() ) { 2308 to_fill->_msg = deopt_resume2; 2309 } else { 2310 to_fill->_msg = method_resume; 2311 } 2312 to_fill->_result._to_call._bcp_advance = 0; 2313 to_fill->_result._to_call._callee_entry_point = NULL; // doesn't matter to anyone 2314 to_fill->_result._to_call._callee = NULL; // doesn't matter to anyone 2315 to_fill->_prev_link = NULL; 2316 2317 to_fill->_sender_sp = caller->unextended_sp(); 2318 2319 if (caller->is_interpreted_frame()) { 2320 interpreterState prev = caller->get_interpreterState(); 2321 to_fill->_prev_link = prev; 2322 // *current->register_addr(GR_Iprev_state) = (intptr_t) prev; 2323 // Make the prev callee look proper 2324 prev->_result._to_call._callee = method; 2325 if (*prev->_bcp == Bytecodes::_invokeinterface) { 2326 prev->_result._to_call._bcp_advance = 5; 2327 } else { 2328 prev->_result._to_call._bcp_advance = 3; 2329 } 2330 } 2331 to_fill->_oop_temp = NULL; 2332 to_fill->_stack_base = stack_base; 2333 // Need +1 here because stack_base points to the word just above the first expr stack entry 2334 // and stack_limit is supposed to point to the word just below the last expr stack entry. 2335 // See generate_compute_interpreter_state. 2336 int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries(); 2337 to_fill->_stack_limit = stack_base - (method->max_stack() + extra_stack + 1); 2338 to_fill->_monitor_base = (BasicObjectLock*) monitor_base; 2339 2340 to_fill->_self_link = to_fill; 2341 assert(stack >= to_fill->_stack_limit && stack < to_fill->_stack_base, 2342 "Stack top out of range"); 2343 } 2344 2345 int AbstractInterpreter::layout_activation(methodOop method, 2346 int tempcount, // 2347 int popframe_extra_args, 2348 int moncount, 2349 int caller_actual_parameters, 2350 int callee_param_count, 2351 int callee_locals, 2352 frame* caller, 2353 frame* interpreter_frame, 2354 bool is_top_frame) { 2355 2356 assert(popframe_extra_args == 0, "FIX ME"); 2357 // NOTE this code must exactly mimic what InterpreterGenerator::generate_compute_interpreter_state() 2358 // does as far as allocating an interpreter frame. 2359 // If interpreter_frame!=NULL, set up the method, locals, and monitors. 2360 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size, 2361 // as determined by a previous call to this method. 2362 // It is also guaranteed to be walkable even though it is in a skeletal state 2363 // NOTE: return size is in words not bytes 2364 // NOTE: tempcount is the current size of the java expression stack. For top most 2365 // frames we will allocate a full sized expression stack and not the curback 2366 // version that non-top frames have. 2367 2368 // Calculate the amount our frame will be adjust by the callee. For top frame 2369 // this is zero. 2370 2371 // NOTE: ia64 seems to do this wrong (or at least backwards) in that it 2372 // calculates the extra locals based on itself. Not what the callee does 2373 // to it. So it ignores last_frame_adjust value. Seems suspicious as far 2374 // as getting sender_sp correct. 2375 2376 int extra_locals_size = (callee_locals - callee_param_count) * BytesPerWord; 2377 int monitor_size = sizeof(BasicObjectLock) * moncount; 2378 2379 // First calculate the frame size without any java expression stack 2380 int short_frame_size = size_activation_helper(extra_locals_size, 2381 monitor_size); 2382 2383 // Now with full size expression stack 2384 int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries(); 2385 int full_frame_size = short_frame_size + (method->max_stack() + extra_stack) * BytesPerWord; 2386 2387 // and now with only live portion of the expression stack 2388 short_frame_size = short_frame_size + tempcount * BytesPerWord; 2389 2390 // the size the activation is right now. Only top frame is full size 2391 int frame_size = (is_top_frame ? full_frame_size : short_frame_size); 2392 2393 if (interpreter_frame != NULL) { 2394 #ifdef ASSERT 2395 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable"); 2396 #endif 2397 2398 // MUCHO HACK 2399 2400 intptr_t* frame_bottom = (intptr_t*) ((intptr_t)interpreter_frame->sp() - (full_frame_size - frame_size)); 2401 2402 /* Now fillin the interpreterState object */ 2403 2404 // The state object is the first thing on the frame and easily located 2405 2406 interpreterState cur_state = (interpreterState) ((intptr_t)interpreter_frame->fp() - sizeof(BytecodeInterpreter)); 2407 2408 2409 // Find the locals pointer. This is rather simple on x86 because there is no 2410 // confusing rounding at the callee to account for. We can trivially locate 2411 // our locals based on the current fp(). 2412 // Note: the + 2 is for handling the "static long no_params() method" issue. 2413 // (too bad I don't really remember that issue well...) 2414 2415 intptr_t* locals; 2416 // If the caller is interpreted we need to make sure that locals points to the first 2417 // argument that the caller passed and not in an area where the stack might have been extended. 2418 // because the stack to stack to converter needs a proper locals value in order to remove the 2419 // arguments from the caller and place the result in the proper location. Hmm maybe it'd be 2420 // simpler if we simply stored the result in the BytecodeInterpreter object and let the c++ code 2421 // adjust the stack?? HMMM QQQ 2422 // 2423 if (caller->is_interpreted_frame()) { 2424 // locals must agree with the caller because it will be used to set the 2425 // caller's tos when we return. 2426 interpreterState prev = caller->get_interpreterState(); 2427 // stack() is prepushed. 2428 locals = prev->stack() + method->size_of_parameters(); 2429 // locals = caller->unextended_sp() + (method->size_of_parameters() - 1); 2430 if (locals != interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2) { 2431 // os::breakpoint(); 2432 } 2433 } else { 2434 // this is where a c2i would have placed locals (except for the +2) 2435 locals = interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2; 2436 } 2437 2438 intptr_t* monitor_base = (intptr_t*) cur_state; 2439 intptr_t* stack_base = (intptr_t*) ((intptr_t) monitor_base - monitor_size); 2440 /* +1 because stack is always prepushed */ 2441 intptr_t* stack = (intptr_t*) ((intptr_t) stack_base - (tempcount + 1) * BytesPerWord); 2442 2443 2444 BytecodeInterpreter::layout_interpreterState(cur_state, 2445 caller, 2446 interpreter_frame, 2447 method, 2448 locals, 2449 stack, 2450 stack_base, 2451 monitor_base, 2452 frame_bottom, 2453 is_top_frame); 2454 2455 // BytecodeInterpreter::pd_layout_interpreterState(cur_state, interpreter_return_address, interpreter_frame->fp()); 2456 } 2457 return frame_size/BytesPerWord; 2458 } 2459 2460 #endif // CC_INTERP (all)