1 /* 2 * Copyright (c) 1997, 2010, 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 "interp_masm_sparc.hpp" 27 #include "interpreter/interpreter.hpp" 28 #include "interpreter/interpreterRuntime.hpp" 29 #include "oops/arrayOop.hpp" 30 #include "oops/markOop.hpp" 31 #include "oops/methodDataOop.hpp" 32 #include "oops/methodOop.hpp" 33 #include "prims/jvmtiExport.hpp" 34 #include "prims/jvmtiRedefineClassesTrace.hpp" 35 #include "prims/jvmtiThreadState.hpp" 36 #include "runtime/basicLock.hpp" 37 #include "runtime/biasedLocking.hpp" 38 #include "runtime/sharedRuntime.hpp" 39 #ifdef TARGET_OS_FAMILY_linux 40 # include "thread_linux.inline.hpp" 41 #endif 42 #ifdef TARGET_OS_FAMILY_solaris 43 # include "thread_solaris.inline.hpp" 44 #endif 45 46 #ifndef CC_INTERP 47 #ifndef FAST_DISPATCH 48 #define FAST_DISPATCH 1 49 #endif 50 #undef FAST_DISPATCH 51 52 // Implementation of InterpreterMacroAssembler 53 54 // This file specializes the assember with interpreter-specific macros 55 56 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS); 57 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS); 58 59 #else // CC_INTERP 60 #ifndef STATE 61 #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name)) 62 #endif // STATE 63 64 #endif // CC_INTERP 65 66 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) { 67 // Note: this algorithm is also used by C1's OSR entry sequence. 68 // Any changes should also be applied to CodeEmitter::emit_osr_entry(). 69 assert_different_registers(args_size, locals_size); 70 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted. 71 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words 72 // Use br/mov combination because it works on both V8 and V9 and is 73 // faster. 74 Label skip_move; 75 br(Assembler::negative, true, Assembler::pt, skip_move); 76 delayed()->mov(G0, delta); 77 bind(skip_move); 78 round_to(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned) 79 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes 80 } 81 82 #ifndef CC_INTERP 83 84 // Dispatch code executed in the prolog of a bytecode which does not do it's 85 // own dispatch. The dispatch address is computed and placed in IdispatchAddress 86 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 87 assert_not_delayed(); 88 #ifdef FAST_DISPATCH 89 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since 90 // they both use I2. 91 assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive"); 92 ldub(Lbcp, bcp_incr, Lbyte_code); // load next bytecode 93 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); 94 // add offset to correct dispatch table 95 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 96 ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr 97 #else 98 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 99 // dispatch table to use 100 AddressLiteral tbl(Interpreter::dispatch_table(state)); 101 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 102 set(tbl, G3_scratch); // compute addr of table 103 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr 104 #endif 105 } 106 107 108 // Dispatch code executed in the epilog of a bytecode which does not do it's 109 // own dispatch. The dispatch address in IdispatchAddress is used for the 110 // dispatch. 111 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 112 assert_not_delayed(); 113 verify_FPU(1, state); 114 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 115 jmp( IdispatchAddress, 0 ); 116 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); 117 else delayed()->nop(); 118 } 119 120 121 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { 122 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) 123 assert_not_delayed(); 124 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 125 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr); 126 } 127 128 129 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) { 130 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) 131 assert_not_delayed(); 132 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 133 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false); 134 } 135 136 137 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 138 // load current bytecode 139 assert_not_delayed(); 140 ldub( Lbcp, 0, Lbyte_code); // load next bytecode 141 dispatch_base(state, table); 142 } 143 144 145 void InterpreterMacroAssembler::call_VM_leaf_base( 146 Register java_thread, 147 address entry_point, 148 int number_of_arguments 149 ) { 150 if (!java_thread->is_valid()) 151 java_thread = L7_thread_cache; 152 // super call 153 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments); 154 } 155 156 157 void InterpreterMacroAssembler::call_VM_base( 158 Register oop_result, 159 Register java_thread, 160 Register last_java_sp, 161 address entry_point, 162 int number_of_arguments, 163 bool check_exception 164 ) { 165 if (!java_thread->is_valid()) 166 java_thread = L7_thread_cache; 167 // See class ThreadInVMfromInterpreter, which assumes that the interpreter 168 // takes responsibility for setting its own thread-state on call-out. 169 // However, ThreadInVMfromInterpreter resets the state to "in_Java". 170 171 //save_bcp(); // save bcp 172 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception); 173 //restore_bcp(); // restore bcp 174 //restore_locals(); // restore locals pointer 175 } 176 177 178 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 179 if (JvmtiExport::can_pop_frame()) { 180 Label L; 181 182 // Check the "pending popframe condition" flag in the current thread 183 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg); 184 185 // Initiate popframe handling only if it is not already being processed. If the flag 186 // has the popframe_processing bit set, it means that this code is called *during* popframe 187 // handling - we don't want to reenter. 188 btst(JavaThread::popframe_pending_bit, scratch_reg); 189 br(zero, false, pt, L); 190 delayed()->nop(); 191 btst(JavaThread::popframe_processing_bit, scratch_reg); 192 br(notZero, false, pt, L); 193 delayed()->nop(); 194 195 // Call Interpreter::remove_activation_preserving_args_entry() to get the 196 // address of the same-named entrypoint in the generated interpreter code. 197 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 198 199 // Jump to Interpreter::_remove_activation_preserving_args_entry 200 jmpl(O0, G0, G0); 201 delayed()->nop(); 202 bind(L); 203 } 204 } 205 206 207 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 208 Register thr_state = G4_scratch; 209 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); 210 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset()); 211 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset()); 212 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset()); 213 switch (state) { 214 case ltos: ld_long(val_addr, Otos_l); break; 215 case atos: ld_ptr(oop_addr, Otos_l); 216 st_ptr(G0, oop_addr); break; 217 case btos: // fall through 218 case ctos: // fall through 219 case stos: // fall through 220 case itos: ld(val_addr, Otos_l1); break; 221 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break; 222 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break; 223 case vtos: /* nothing to do */ break; 224 default : ShouldNotReachHere(); 225 } 226 // Clean up tos value in the jvmti thread state 227 or3(G0, ilgl, G3_scratch); 228 stw(G3_scratch, tos_addr); 229 st_long(G0, val_addr); 230 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 231 } 232 233 234 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 235 if (JvmtiExport::can_force_early_return()) { 236 Label L; 237 Register thr_state = G3_scratch; 238 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); 239 tst(thr_state); 240 br(zero, false, pt, L); // if (thread->jvmti_thread_state() == NULL) exit; 241 delayed()->nop(); 242 243 // Initiate earlyret handling only if it is not already being processed. 244 // If the flag has the earlyret_processing bit set, it means that this code 245 // is called *during* earlyret handling - we don't want to reenter. 246 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch); 247 cmp(G4_scratch, JvmtiThreadState::earlyret_pending); 248 br(Assembler::notEqual, false, pt, L); 249 delayed()->nop(); 250 251 // Call Interpreter::remove_activation_early_entry() to get the address of the 252 // same-named entrypoint in the generated interpreter code 253 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1); 254 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1); 255 256 // Jump to Interpreter::_remove_activation_early_entry 257 jmpl(O0, G0, G0); 258 delayed()->nop(); 259 bind(L); 260 } 261 } 262 263 264 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) { 265 mov(arg_1, O0); 266 mov(arg_2, O1); 267 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2); 268 } 269 #endif /* CC_INTERP */ 270 271 272 #ifndef CC_INTERP 273 274 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) { 275 assert_not_delayed(); 276 dispatch_Lbyte_code(state, table); 277 } 278 279 280 void InterpreterMacroAssembler::dispatch_normal(TosState state) { 281 dispatch_base(state, Interpreter::normal_table(state)); 282 } 283 284 285 void InterpreterMacroAssembler::dispatch_only(TosState state) { 286 dispatch_base(state, Interpreter::dispatch_table(state)); 287 } 288 289 290 // common code to dispatch and dispatch_only 291 // dispatch value in Lbyte_code and increment Lbcp 292 293 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) { 294 verify_FPU(1, state); 295 // %%%%% maybe implement +VerifyActivationFrameSize here 296 //verify_thread(); //too slow; we will just verify on method entry & exit 297 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 298 #ifdef FAST_DISPATCH 299 if (table == Interpreter::dispatch_table(state)) { 300 // use IdispatchTables 301 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); 302 // add offset to correct dispatch table 303 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 304 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // get entry addr 305 } else { 306 #endif 307 // dispatch table to use 308 AddressLiteral tbl(table); 309 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 310 set(tbl, G3_scratch); // compute addr of table 311 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr 312 #ifdef FAST_DISPATCH 313 } 314 #endif 315 jmp( G3_scratch, 0 ); 316 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); 317 else delayed()->nop(); 318 } 319 320 321 // Helpers for expression stack 322 323 // Longs and doubles are Category 2 computational types in the 324 // JVM specification (section 3.11.1) and take 2 expression stack or 325 // local slots. 326 // Aligning them on 32 bit with tagged stacks is hard because the code generated 327 // for the dup* bytecodes depends on what types are already on the stack. 328 // If the types are split into the two stack/local slots, that is much easier 329 // (and we can use 0 for non-reference tags). 330 331 // Known good alignment in _LP64 but unknown otherwise 332 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) { 333 assert_not_delayed(); 334 335 #ifdef _LP64 336 ldf(FloatRegisterImpl::D, r1, offset, d); 337 #else 338 ldf(FloatRegisterImpl::S, r1, offset, d); 339 ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize, d->successor()); 340 #endif 341 } 342 343 // Known good alignment in _LP64 but unknown otherwise 344 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) { 345 assert_not_delayed(); 346 347 #ifdef _LP64 348 stf(FloatRegisterImpl::D, d, r1, offset); 349 // store something more useful here 350 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);) 351 #else 352 stf(FloatRegisterImpl::S, d, r1, offset); 353 stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize); 354 #endif 355 } 356 357 358 // Known good alignment in _LP64 but unknown otherwise 359 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) { 360 assert_not_delayed(); 361 #ifdef _LP64 362 ldx(r1, offset, rd); 363 #else 364 ld(r1, offset, rd); 365 ld(r1, offset + Interpreter::stackElementSize, rd->successor()); 366 #endif 367 } 368 369 // Known good alignment in _LP64 but unknown otherwise 370 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) { 371 assert_not_delayed(); 372 373 #ifdef _LP64 374 stx(l, r1, offset); 375 // store something more useful here 376 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);) 377 #else 378 st(l, r1, offset); 379 st(l->successor(), r1, offset + Interpreter::stackElementSize); 380 #endif 381 } 382 383 void InterpreterMacroAssembler::pop_i(Register r) { 384 assert_not_delayed(); 385 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r); 386 inc(Lesp, Interpreter::stackElementSize); 387 debug_only(verify_esp(Lesp)); 388 } 389 390 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) { 391 assert_not_delayed(); 392 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); 393 inc(Lesp, Interpreter::stackElementSize); 394 debug_only(verify_esp(Lesp)); 395 } 396 397 void InterpreterMacroAssembler::pop_l(Register r) { 398 assert_not_delayed(); 399 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r); 400 inc(Lesp, 2*Interpreter::stackElementSize); 401 debug_only(verify_esp(Lesp)); 402 } 403 404 405 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) { 406 assert_not_delayed(); 407 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f); 408 inc(Lesp, Interpreter::stackElementSize); 409 debug_only(verify_esp(Lesp)); 410 } 411 412 413 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) { 414 assert_not_delayed(); 415 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f); 416 inc(Lesp, 2*Interpreter::stackElementSize); 417 debug_only(verify_esp(Lesp)); 418 } 419 420 421 void InterpreterMacroAssembler::push_i(Register r) { 422 assert_not_delayed(); 423 debug_only(verify_esp(Lesp)); 424 st(r, Lesp, 0); 425 dec(Lesp, Interpreter::stackElementSize); 426 } 427 428 void InterpreterMacroAssembler::push_ptr(Register r) { 429 assert_not_delayed(); 430 st_ptr(r, Lesp, 0); 431 dec(Lesp, Interpreter::stackElementSize); 432 } 433 434 // remember: our convention for longs in SPARC is: 435 // O0 (Otos_l1) has high-order part in first word, 436 // O1 (Otos_l2) has low-order part in second word 437 438 void InterpreterMacroAssembler::push_l(Register r) { 439 assert_not_delayed(); 440 debug_only(verify_esp(Lesp)); 441 // Longs are stored in memory-correct order, even if unaligned. 442 int offset = -Interpreter::stackElementSize; 443 store_unaligned_long(r, Lesp, offset); 444 dec(Lesp, 2 * Interpreter::stackElementSize); 445 } 446 447 448 void InterpreterMacroAssembler::push_f(FloatRegister f) { 449 assert_not_delayed(); 450 debug_only(verify_esp(Lesp)); 451 stf(FloatRegisterImpl::S, f, Lesp, 0); 452 dec(Lesp, Interpreter::stackElementSize); 453 } 454 455 456 void InterpreterMacroAssembler::push_d(FloatRegister d) { 457 assert_not_delayed(); 458 debug_only(verify_esp(Lesp)); 459 // Longs are stored in memory-correct order, even if unaligned. 460 int offset = -Interpreter::stackElementSize; 461 store_unaligned_double(d, Lesp, offset); 462 dec(Lesp, 2 * Interpreter::stackElementSize); 463 } 464 465 466 void InterpreterMacroAssembler::push(TosState state) { 467 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 468 switch (state) { 469 case atos: push_ptr(); break; 470 case btos: push_i(); break; 471 case ctos: 472 case stos: push_i(); break; 473 case itos: push_i(); break; 474 case ltos: push_l(); break; 475 case ftos: push_f(); break; 476 case dtos: push_d(); break; 477 case vtos: /* nothing to do */ break; 478 default : ShouldNotReachHere(); 479 } 480 } 481 482 483 void InterpreterMacroAssembler::pop(TosState state) { 484 switch (state) { 485 case atos: pop_ptr(); break; 486 case btos: pop_i(); break; 487 case ctos: 488 case stos: pop_i(); break; 489 case itos: pop_i(); break; 490 case ltos: pop_l(); break; 491 case ftos: pop_f(); break; 492 case dtos: pop_d(); break; 493 case vtos: /* nothing to do */ break; 494 default : ShouldNotReachHere(); 495 } 496 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 497 } 498 499 500 // Helpers for swap and dup 501 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 502 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val); 503 } 504 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 505 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n)); 506 } 507 508 509 void InterpreterMacroAssembler::load_receiver(Register param_count, 510 Register recv) { 511 sll(param_count, Interpreter::logStackElementSize, param_count); 512 ld_ptr(Lesp, param_count, recv); // gets receiver Oop 513 } 514 515 void InterpreterMacroAssembler::empty_expression_stack() { 516 // Reset Lesp. 517 sub( Lmonitors, wordSize, Lesp ); 518 519 // Reset SP by subtracting more space from Lesp. 520 Label done; 521 verify_oop(Lmethod); 522 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!"); 523 524 // A native does not need to do this, since its callee does not change SP. 525 ld(Lmethod, methodOopDesc::access_flags_offset(), Gframe_size); // Load access flags. 526 btst(JVM_ACC_NATIVE, Gframe_size); 527 br(Assembler::notZero, false, Assembler::pt, done); 528 delayed()->nop(); 529 530 // Compute max expression stack+register save area 531 lduh(Lmethod, in_bytes(methodOopDesc::max_stack_offset()), Gframe_size); // Load max stack. 532 add( Gframe_size, frame::memory_parameter_word_sp_offset, Gframe_size ); 533 534 // 535 // now set up a stack frame with the size computed above 536 // 537 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below 538 sll( Gframe_size, LogBytesPerWord, Gframe_size ); 539 sub( Lesp, Gframe_size, Gframe_size ); 540 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary 541 debug_only(verify_sp(Gframe_size, G4_scratch)); 542 #ifdef _LP64 543 sub(Gframe_size, STACK_BIAS, Gframe_size ); 544 #endif 545 mov(Gframe_size, SP); 546 547 bind(done); 548 } 549 550 551 #ifdef ASSERT 552 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) { 553 Label Bad, OK; 554 555 // Saved SP must be aligned. 556 #ifdef _LP64 557 btst(2*BytesPerWord-1, Rsp); 558 #else 559 btst(LongAlignmentMask, Rsp); 560 #endif 561 br(Assembler::notZero, false, Assembler::pn, Bad); 562 delayed()->nop(); 563 564 // Saved SP, plus register window size, must not be above FP. 565 add(Rsp, frame::register_save_words * wordSize, Rtemp); 566 #ifdef _LP64 567 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP 568 #endif 569 cmp(Rtemp, FP); 570 brx(Assembler::greaterUnsigned, false, Assembler::pn, Bad); 571 delayed()->nop(); 572 573 // Saved SP must not be ridiculously below current SP. 574 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K); 575 set(maxstack, Rtemp); 576 sub(SP, Rtemp, Rtemp); 577 #ifdef _LP64 578 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp 579 #endif 580 cmp(Rsp, Rtemp); 581 brx(Assembler::lessUnsigned, false, Assembler::pn, Bad); 582 delayed()->nop(); 583 584 br(Assembler::always, false, Assembler::pn, OK); 585 delayed()->nop(); 586 587 bind(Bad); 588 stop("on return to interpreted call, restored SP is corrupted"); 589 590 bind(OK); 591 } 592 593 594 void InterpreterMacroAssembler::verify_esp(Register Resp) { 595 // about to read or write Resp[0] 596 // make sure it is not in the monitors or the register save area 597 Label OK1, OK2; 598 599 cmp(Resp, Lmonitors); 600 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1); 601 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp); 602 stop("too many pops: Lesp points into monitor area"); 603 bind(OK1); 604 #ifdef _LP64 605 sub(Resp, STACK_BIAS, Resp); 606 #endif 607 cmp(Resp, SP); 608 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2); 609 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp); 610 stop("too many pushes: Lesp points into register window"); 611 bind(OK2); 612 } 613 #endif // ASSERT 614 615 // Load compiled (i2c) or interpreter entry when calling from interpreted and 616 // do the call. Centralized so that all interpreter calls will do the same actions. 617 // If jvmti single stepping is on for a thread we must not call compiled code. 618 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) { 619 620 // Assume we want to go compiled if available 621 622 ld_ptr(G5_method, in_bytes(methodOopDesc::from_interpreted_offset()), target); 623 624 if (JvmtiExport::can_post_interpreter_events()) { 625 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 626 // compiled code in threads for which the event is enabled. Check here for 627 // interp_only_mode if these events CAN be enabled. 628 verify_thread(); 629 Label skip_compiled_code; 630 631 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 632 ld(interp_only, scratch); 633 tst(scratch); 634 br(Assembler::notZero, true, Assembler::pn, skip_compiled_code); 635 delayed()->ld_ptr(G5_method, in_bytes(methodOopDesc::interpreter_entry_offset()), target); 636 bind(skip_compiled_code); 637 } 638 639 // the i2c_adapters need methodOop in G5_method (right? %%%) 640 // do the call 641 #ifdef ASSERT 642 { 643 Label ok; 644 br_notnull(target, false, Assembler::pt, ok); 645 delayed()->nop(); 646 stop("null entry point"); 647 bind(ok); 648 } 649 #endif // ASSERT 650 651 // Adjust Rret first so Llast_SP can be same as Rret 652 add(Rret, -frame::pc_return_offset, O7); 653 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer 654 // Record SP so we can remove any stack space allocated by adapter transition 655 jmp(target, 0); 656 delayed()->mov(SP, Llast_SP); 657 } 658 659 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) { 660 assert_not_delayed(); 661 662 Label not_taken; 663 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken); 664 else br (cc, false, Assembler::pn, not_taken); 665 delayed()->nop(); 666 667 TemplateTable::branch(false,false); 668 669 bind(not_taken); 670 671 profile_not_taken_branch(G3_scratch); 672 } 673 674 675 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp( 676 int bcp_offset, 677 Register Rtmp, 678 Register Rdst, 679 signedOrNot is_signed, 680 setCCOrNot should_set_CC ) { 681 assert(Rtmp != Rdst, "need separate temp register"); 682 assert_not_delayed(); 683 switch (is_signed) { 684 default: ShouldNotReachHere(); 685 686 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte 687 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte 688 } 689 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte 690 sll( Rdst, BitsPerByte, Rdst); 691 switch (should_set_CC ) { 692 default: ShouldNotReachHere(); 693 694 case set_CC: orcc( Rdst, Rtmp, Rdst ); break; 695 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break; 696 } 697 } 698 699 700 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp( 701 int bcp_offset, 702 Register Rtmp, 703 Register Rdst, 704 setCCOrNot should_set_CC ) { 705 assert(Rtmp != Rdst, "need separate temp register"); 706 assert_not_delayed(); 707 add( Lbcp, bcp_offset, Rtmp); 708 andcc( Rtmp, 3, G0); 709 Label aligned; 710 switch (should_set_CC ) { 711 default: ShouldNotReachHere(); 712 713 case set_CC: break; 714 case dont_set_CC: break; 715 } 716 717 br(Assembler::zero, true, Assembler::pn, aligned); 718 #ifdef _LP64 719 delayed()->ldsw(Rtmp, 0, Rdst); 720 #else 721 delayed()->ld(Rtmp, 0, Rdst); 722 #endif 723 724 ldub(Lbcp, bcp_offset + 3, Rdst); 725 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst); 726 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst); 727 #ifdef _LP64 728 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); 729 #else 730 // Unsigned load is faster than signed on some implementations 731 ldub(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); 732 #endif 733 or3(Rtmp, Rdst, Rdst ); 734 735 bind(aligned); 736 if (should_set_CC == set_CC) tst(Rdst); 737 } 738 739 740 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register cache, Register tmp, 741 int bcp_offset, size_t index_size) { 742 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 743 if (index_size == sizeof(u2)) { 744 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); 745 } else if (index_size == sizeof(u4)) { 746 assert(EnableInvokeDynamic, "giant index used only for EnableInvokeDynamic"); 747 get_4_byte_integer_at_bcp(bcp_offset, cache, tmp); 748 assert(constantPoolCacheOopDesc::decode_secondary_index(~123) == 123, "else change next line"); 749 xor3(tmp, -1, tmp); // convert to plain index 750 } else if (index_size == sizeof(u1)) { 751 assert(EnableMethodHandles, "tiny index used only for EnableMethodHandles"); 752 ldub(Lbcp, bcp_offset, tmp); 753 } else { 754 ShouldNotReachHere(); 755 } 756 } 757 758 759 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp, 760 int bcp_offset, size_t index_size) { 761 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 762 assert_different_registers(cache, tmp); 763 assert_not_delayed(); 764 get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size); 765 // convert from field index to ConstantPoolCacheEntry index and from 766 // word index to byte offset 767 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); 768 add(LcpoolCache, tmp, cache); 769 } 770 771 772 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp, 773 int bcp_offset, size_t index_size) { 774 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 775 assert_different_registers(cache, tmp); 776 assert_not_delayed(); 777 if (index_size == sizeof(u2)) { 778 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); 779 } else { 780 ShouldNotReachHere(); // other sizes not supported here 781 } 782 // convert from field index to ConstantPoolCacheEntry index 783 // and from word index to byte offset 784 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); 785 // skip past the header 786 add(tmp, in_bytes(constantPoolCacheOopDesc::base_offset()), tmp); 787 // construct pointer to cache entry 788 add(LcpoolCache, tmp, cache); 789 } 790 791 792 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 793 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. 794 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 795 Register Rsuper_klass, 796 Register Rtmp1, 797 Register Rtmp2, 798 Register Rtmp3, 799 Label &ok_is_subtype ) { 800 Label not_subtype; 801 802 // Profile the not-null value's klass. 803 profile_typecheck(Rsub_klass, Rtmp1); 804 805 check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass, 806 Rtmp1, Rtmp2, 807 &ok_is_subtype, ¬_subtype, NULL); 808 809 check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass, 810 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg, 811 &ok_is_subtype, NULL); 812 813 bind(not_subtype); 814 profile_typecheck_failed(Rtmp1); 815 } 816 817 // Separate these two to allow for delay slot in middle 818 // These are used to do a test and full jump to exception-throwing code. 819 820 // %%%%% Could possibly reoptimize this by testing to see if could use 821 // a single conditional branch (i.e. if span is small enough. 822 // If you go that route, than get rid of the split and give up 823 // on the delay-slot hack. 824 825 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition, 826 Label& ok ) { 827 assert_not_delayed(); 828 br(ok_condition, true, pt, ok); 829 // DELAY SLOT 830 } 831 832 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition, 833 Label& ok ) { 834 assert_not_delayed(); 835 bp( ok_condition, true, Assembler::xcc, pt, ok); 836 // DELAY SLOT 837 } 838 839 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition, 840 Label& ok ) { 841 assert_not_delayed(); 842 brx(ok_condition, true, pt, ok); 843 // DELAY SLOT 844 } 845 846 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point, 847 Register Rscratch, 848 Label& ok ) { 849 assert(throw_entry_point != NULL, "entry point must be generated by now"); 850 AddressLiteral dest(throw_entry_point); 851 jump_to(dest, Rscratch); 852 delayed()->nop(); 853 bind(ok); 854 } 855 856 857 // And if you cannot use the delay slot, here is a shorthand: 858 859 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition, 860 address throw_entry_point, 861 Register Rscratch ) { 862 Label ok; 863 if (ok_condition != never) { 864 throw_if_not_1_icc( ok_condition, ok); 865 delayed()->nop(); 866 } 867 throw_if_not_2( throw_entry_point, Rscratch, ok); 868 } 869 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition, 870 address throw_entry_point, 871 Register Rscratch ) { 872 Label ok; 873 if (ok_condition != never) { 874 throw_if_not_1_xcc( ok_condition, ok); 875 delayed()->nop(); 876 } 877 throw_if_not_2( throw_entry_point, Rscratch, ok); 878 } 879 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition, 880 address throw_entry_point, 881 Register Rscratch ) { 882 Label ok; 883 if (ok_condition != never) { 884 throw_if_not_1_x( ok_condition, ok); 885 delayed()->nop(); 886 } 887 throw_if_not_2( throw_entry_point, Rscratch, ok); 888 } 889 890 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res 891 // Note: res is still shy of address by array offset into object. 892 893 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) { 894 assert_not_delayed(); 895 896 verify_oop(array); 897 #ifdef _LP64 898 // sign extend since tos (index) can be a 32bit value 899 sra(index, G0, index); 900 #endif // _LP64 901 902 // check array 903 Label ptr_ok; 904 tst(array); 905 throw_if_not_1_x( notZero, ptr_ok ); 906 delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index 907 throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok); 908 909 Label index_ok; 910 cmp(index, tmp); 911 throw_if_not_1_icc( lessUnsigned, index_ok ); 912 if (index_shift > 0) delayed()->sll(index, index_shift, index); 913 else delayed()->add(array, index, res); // addr - const offset in index 914 // convention: move aberrant index into G3_scratch for exception message 915 mov(index, G3_scratch); 916 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok); 917 918 // add offset if didn't do it in delay slot 919 if (index_shift > 0) add(array, index, res); // addr - const offset in index 920 } 921 922 923 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { 924 assert_not_delayed(); 925 926 // pop array 927 pop_ptr(array); 928 929 // check array 930 index_check_without_pop(array, index, index_shift, tmp, res); 931 } 932 933 934 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 935 ld_ptr(Lmethod, in_bytes(methodOopDesc::constants_offset()), Rdst); 936 } 937 938 939 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 940 get_constant_pool(Rdst); 941 ld_ptr(Rdst, constantPoolOopDesc::cache_offset_in_bytes(), Rdst); 942 } 943 944 945 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 946 get_constant_pool(Rcpool); 947 ld_ptr(Rcpool, constantPoolOopDesc::tags_offset_in_bytes(), Rtags); 948 } 949 950 951 // unlock if synchronized method 952 // 953 // Unlock the receiver if this is a synchronized method. 954 // Unlock any Java monitors from syncronized blocks. 955 // 956 // If there are locked Java monitors 957 // If throw_monitor_exception 958 // throws IllegalMonitorStateException 959 // Else if install_monitor_exception 960 // installs IllegalMonitorStateException 961 // Else 962 // no error processing 963 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 964 bool throw_monitor_exception, 965 bool install_monitor_exception) { 966 Label unlocked, unlock, no_unlock; 967 968 // get the value of _do_not_unlock_if_synchronized into G1_scratch 969 const Address do_not_unlock_if_synchronized(G2_thread, 970 JavaThread::do_not_unlock_if_synchronized_offset()); 971 ldbool(do_not_unlock_if_synchronized, G1_scratch); 972 stbool(G0, do_not_unlock_if_synchronized); // reset the flag 973 974 // check if synchronized method 975 const Address access_flags(Lmethod, methodOopDesc::access_flags_offset()); 976 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 977 push(state); // save tos 978 ld(access_flags, G3_scratch); // Load access flags. 979 btst(JVM_ACC_SYNCHRONIZED, G3_scratch); 980 br(zero, false, pt, unlocked); 981 delayed()->nop(); 982 983 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 984 // is set. 985 tstbool(G1_scratch); 986 br(Assembler::notZero, false, pn, no_unlock); 987 delayed()->nop(); 988 989 // BasicObjectLock will be first in list, since this is a synchronized method. However, need 990 // to check that the object has not been unlocked by an explicit monitorexit bytecode. 991 992 //Intel: if (throw_monitor_exception) ... else ... 993 // Entry already unlocked, need to throw exception 994 //... 995 996 // pass top-most monitor elem 997 add( top_most_monitor(), O1 ); 998 999 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch); 1000 br_notnull(G3_scratch, false, pt, unlock); 1001 delayed()->nop(); 1002 1003 if (throw_monitor_exception) { 1004 // Entry already unlocked need to throw an exception 1005 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1006 should_not_reach_here(); 1007 } else { 1008 // Monitor already unlocked during a stack unroll. 1009 // If requested, install an illegal_monitor_state_exception. 1010 // Continue with stack unrolling. 1011 if (install_monitor_exception) { 1012 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1013 } 1014 ba(false, unlocked); 1015 delayed()->nop(); 1016 } 1017 1018 bind(unlock); 1019 1020 unlock_object(O1); 1021 1022 bind(unlocked); 1023 1024 // I0, I1: Might contain return value 1025 1026 // Check that all monitors are unlocked 1027 { Label loop, exception, entry, restart; 1028 1029 Register Rmptr = O0; 1030 Register Rtemp = O1; 1031 Register Rlimit = Lmonitors; 1032 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 1033 assert( (delta & LongAlignmentMask) == 0, 1034 "sizeof BasicObjectLock must be even number of doublewords"); 1035 1036 #ifdef ASSERT 1037 add(top_most_monitor(), Rmptr, delta); 1038 { Label L; 1039 // ensure that Rmptr starts out above (or at) Rlimit 1040 cmp(Rmptr, Rlimit); 1041 brx(Assembler::greaterEqualUnsigned, false, pn, L); 1042 delayed()->nop(); 1043 stop("monitor stack has negative size"); 1044 bind(L); 1045 } 1046 #endif 1047 bind(restart); 1048 ba(false, entry); 1049 delayed()-> 1050 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry 1051 1052 // Entry is still locked, need to throw exception 1053 bind(exception); 1054 if (throw_monitor_exception) { 1055 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1056 should_not_reach_here(); 1057 } else { 1058 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 1059 // Unlock does not block, so don't have to worry about the frame 1060 unlock_object(Rmptr); 1061 if (install_monitor_exception) { 1062 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1063 } 1064 ba(false, restart); 1065 delayed()->nop(); 1066 } 1067 1068 bind(loop); 1069 cmp(Rtemp, G0); // check if current entry is used 1070 brx(Assembler::notEqual, false, pn, exception); 1071 delayed()-> 1072 dec(Rmptr, delta); // otherwise advance to next entry 1073 #ifdef ASSERT 1074 { Label L; 1075 // ensure that Rmptr has not somehow stepped below Rlimit 1076 cmp(Rmptr, Rlimit); 1077 brx(Assembler::greaterEqualUnsigned, false, pn, L); 1078 delayed()->nop(); 1079 stop("ran off the end of the monitor stack"); 1080 bind(L); 1081 } 1082 #endif 1083 bind(entry); 1084 cmp(Rmptr, Rlimit); // check if bottom reached 1085 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry 1086 delayed()-> 1087 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp); 1088 } 1089 1090 bind(no_unlock); 1091 pop(state); 1092 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1093 } 1094 1095 1096 // remove activation 1097 // 1098 // Unlock the receiver if this is a synchronized method. 1099 // Unlock any Java monitors from syncronized blocks. 1100 // Remove the activation from the stack. 1101 // 1102 // If there are locked Java monitors 1103 // If throw_monitor_exception 1104 // throws IllegalMonitorStateException 1105 // Else if install_monitor_exception 1106 // installs IllegalMonitorStateException 1107 // Else 1108 // no error processing 1109 void InterpreterMacroAssembler::remove_activation(TosState state, 1110 bool throw_monitor_exception, 1111 bool install_monitor_exception) { 1112 1113 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 1114 1115 // save result (push state before jvmti call and pop it afterwards) and notify jvmti 1116 notify_method_exit(false, state, NotifyJVMTI); 1117 1118 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1119 verify_oop(Lmethod); 1120 verify_thread(); 1121 1122 // return tos 1123 assert(Otos_l1 == Otos_i, "adjust code below"); 1124 switch (state) { 1125 #ifdef _LP64 1126 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0 1127 #else 1128 case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through // O1 -> I1 1129 #endif 1130 case btos: // fall through 1131 case ctos: 1132 case stos: // fall through 1133 case atos: // fall through 1134 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0 1135 case ftos: // fall through 1136 case dtos: // fall through 1137 case vtos: /* nothing to do */ break; 1138 default : ShouldNotReachHere(); 1139 } 1140 1141 #if defined(COMPILER2) && !defined(_LP64) 1142 if (state == ltos) { 1143 // C2 expects long results in G1 we can't tell if we're returning to interpreted 1144 // or compiled so just be safe use G1 and O0/O1 1145 1146 // Shift bits into high (msb) of G1 1147 sllx(Otos_l1->after_save(), 32, G1); 1148 // Zero extend low bits 1149 srl (Otos_l2->after_save(), 0, Otos_l2->after_save()); 1150 or3 (Otos_l2->after_save(), G1, G1); 1151 } 1152 #endif /* COMPILER2 */ 1153 1154 } 1155 #endif /* CC_INTERP */ 1156 1157 1158 // Lock object 1159 // 1160 // Argument - lock_reg points to the BasicObjectLock to be used for locking, 1161 // it must be initialized with the object to lock 1162 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) { 1163 if (UseHeavyMonitors) { 1164 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1165 } 1166 else { 1167 Register obj_reg = Object; 1168 Register mark_reg = G4_scratch; 1169 Register temp_reg = G1_scratch; 1170 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes()); 1171 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1172 Label done; 1173 1174 Label slow_case; 1175 1176 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg); 1177 1178 // load markOop from object into mark_reg 1179 ld_ptr(mark_addr, mark_reg); 1180 1181 if (UseBiasedLocking) { 1182 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case); 1183 } 1184 1185 // get the address of basicLock on stack that will be stored in the object 1186 // we need a temporary register here as we do not want to clobber lock_reg 1187 // (cas clobbers the destination register) 1188 mov(lock_reg, temp_reg); 1189 // set mark reg to be (markOop of object | UNLOCK_VALUE) 1190 or3(mark_reg, markOopDesc::unlocked_value, mark_reg); 1191 // initialize the box (Must happen before we update the object mark!) 1192 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1193 // compare and exchange object_addr, markOop | 1, stack address of basicLock 1194 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1195 casx_under_lock(mark_addr.base(), mark_reg, temp_reg, 1196 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); 1197 1198 // if the compare and exchange succeeded we are done (we saw an unlocked object) 1199 cmp(mark_reg, temp_reg); 1200 brx(Assembler::equal, true, Assembler::pt, done); 1201 delayed()->nop(); 1202 1203 // We did not see an unlocked object so try the fast recursive case 1204 1205 // Check if owner is self by comparing the value in the markOop of object 1206 // with the stack pointer 1207 sub(temp_reg, SP, temp_reg); 1208 #ifdef _LP64 1209 sub(temp_reg, STACK_BIAS, temp_reg); 1210 #endif 1211 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 1212 1213 // Composite "andcc" test: 1214 // (a) %sp -vs- markword proximity check, and, 1215 // (b) verify mark word LSBs == 0 (Stack-locked). 1216 // 1217 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size()) 1218 // Note that the page size used for %sp proximity testing is arbitrary and is 1219 // unrelated to the actual MMU page size. We use a 'logical' page size of 1220 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate 1221 // field of the andcc instruction. 1222 andcc (temp_reg, 0xFFFFF003, G0) ; 1223 1224 // if condition is true we are done and hence we can store 0 in the displaced 1225 // header indicating it is a recursive lock and be done 1226 brx(Assembler::zero, true, Assembler::pt, done); 1227 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1228 1229 // none of the above fast optimizations worked so we have to get into the 1230 // slow case of monitor enter 1231 bind(slow_case); 1232 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1233 1234 bind(done); 1235 } 1236 } 1237 1238 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 1239 // 1240 // Argument - lock_reg points to the BasicObjectLock for lock 1241 // Throw IllegalMonitorException if object is not locked by current thread 1242 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1243 if (UseHeavyMonitors) { 1244 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1245 } else { 1246 Register obj_reg = G3_scratch; 1247 Register mark_reg = G4_scratch; 1248 Register displaced_header_reg = G1_scratch; 1249 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes()); 1250 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1251 Label done; 1252 1253 if (UseBiasedLocking) { 1254 // load the object out of the BasicObjectLock 1255 ld_ptr(lockobj_addr, obj_reg); 1256 biased_locking_exit(mark_addr, mark_reg, done, true); 1257 st_ptr(G0, lockobj_addr); // free entry 1258 } 1259 1260 // Test first if we are in the fast recursive case 1261 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes()); 1262 ld_ptr(lock_addr, displaced_header_reg); 1263 br_null(displaced_header_reg, true, Assembler::pn, done); 1264 delayed()->st_ptr(G0, lockobj_addr); // free entry 1265 1266 // See if it is still a light weight lock, if so we just unlock 1267 // the object and we are done 1268 1269 if (!UseBiasedLocking) { 1270 // load the object out of the BasicObjectLock 1271 ld_ptr(lockobj_addr, obj_reg); 1272 } 1273 1274 // we have the displaced header in displaced_header_reg 1275 // we expect to see the stack address of the basicLock in case the 1276 // lock is still a light weight lock (lock_reg) 1277 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1278 casx_under_lock(mark_addr.base(), lock_reg, displaced_header_reg, 1279 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); 1280 cmp(lock_reg, displaced_header_reg); 1281 brx(Assembler::equal, true, Assembler::pn, done); 1282 delayed()->st_ptr(G0, lockobj_addr); // free entry 1283 1284 // The lock has been converted into a heavy lock and hence 1285 // we need to get into the slow case 1286 1287 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1288 1289 bind(done); 1290 } 1291 } 1292 1293 #ifndef CC_INTERP 1294 1295 // Get the method data pointer from the methodOop and set the 1296 // specified register to its value. 1297 1298 void InterpreterMacroAssembler::set_method_data_pointer_offset(Register Roff) { 1299 assert(ProfileInterpreter, "must be profiling interpreter"); 1300 Label get_continue; 1301 1302 ld_ptr(Lmethod, in_bytes(methodOopDesc::method_data_offset()), ImethodDataPtr); 1303 test_method_data_pointer(get_continue); 1304 add(ImethodDataPtr, in_bytes(methodDataOopDesc::data_offset()), ImethodDataPtr); 1305 if (Roff != noreg) 1306 // Roff contains a method data index ("mdi"). It defaults to zero. 1307 add(ImethodDataPtr, Roff, ImethodDataPtr); 1308 bind(get_continue); 1309 } 1310 1311 // Set the method data pointer for the current bcp. 1312 1313 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1314 assert(ProfileInterpreter, "must be profiling interpreter"); 1315 Label zero_continue; 1316 1317 // Test MDO to avoid the call if it is NULL. 1318 ld_ptr(Lmethod, methodOopDesc::method_data_offset(), ImethodDataPtr); 1319 test_method_data_pointer(zero_continue); 1320 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp); 1321 set_method_data_pointer_offset(O0); 1322 bind(zero_continue); 1323 } 1324 1325 // Test ImethodDataPtr. If it is null, continue at the specified label 1326 1327 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1328 assert(ProfileInterpreter, "must be profiling interpreter"); 1329 #ifdef _LP64 1330 bpr(Assembler::rc_z, false, Assembler::pn, ImethodDataPtr, zero_continue); 1331 #else 1332 tst(ImethodDataPtr); 1333 br(Assembler::zero, false, Assembler::pn, zero_continue); 1334 #endif 1335 delayed()->nop(); 1336 } 1337 1338 void InterpreterMacroAssembler::verify_method_data_pointer() { 1339 assert(ProfileInterpreter, "must be profiling interpreter"); 1340 #ifdef ASSERT 1341 Label verify_continue; 1342 test_method_data_pointer(verify_continue); 1343 1344 // If the mdp is valid, it will point to a DataLayout header which is 1345 // consistent with the bcp. The converse is highly probable also. 1346 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch); 1347 ld_ptr(Lmethod, methodOopDesc::const_offset(), O5); 1348 add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), G3_scratch); 1349 add(G3_scratch, O5, G3_scratch); 1350 cmp(Lbcp, G3_scratch); 1351 brx(Assembler::equal, false, Assembler::pt, verify_continue); 1352 1353 Register temp_reg = O5; 1354 delayed()->mov(ImethodDataPtr, temp_reg); 1355 // %%% should use call_VM_leaf here? 1356 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr); 1357 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1); 1358 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS); 1359 stf(FloatRegisterImpl::D, Ftos_d, d_save); 1360 mov(temp_reg->after_save(), O2); 1361 save_thread(L7_thread_cache); 1362 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none); 1363 delayed()->nop(); 1364 restore_thread(L7_thread_cache); 1365 ldf(FloatRegisterImpl::D, d_save, Ftos_d); 1366 restore(); 1367 bind(verify_continue); 1368 #endif // ASSERT 1369 } 1370 1371 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1372 Register cur_bcp, 1373 Register Rtmp, 1374 Label &profile_continue) { 1375 assert(ProfileInterpreter, "must be profiling interpreter"); 1376 // Control will flow to "profile_continue" if the counter is less than the 1377 // limit or if we call profile_method() 1378 1379 Label done; 1380 1381 // if no method data exists, and the counter is high enough, make one 1382 #ifdef _LP64 1383 bpr(Assembler::rc_nz, false, Assembler::pn, ImethodDataPtr, done); 1384 #else 1385 tst(ImethodDataPtr); 1386 br(Assembler::notZero, false, Assembler::pn, done); 1387 #endif 1388 1389 // Test to see if we should create a method data oop 1390 AddressLiteral profile_limit((address) &InvocationCounter::InterpreterProfileLimit); 1391 #ifdef _LP64 1392 delayed()->nop(); 1393 sethi(profile_limit, Rtmp); 1394 #else 1395 delayed()->sethi(profile_limit, Rtmp); 1396 #endif 1397 ld(Rtmp, profile_limit.low10(), Rtmp); 1398 cmp(invocation_count, Rtmp); 1399 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue); 1400 delayed()->nop(); 1401 1402 // Build it now. 1403 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), cur_bcp); 1404 set_method_data_pointer_offset(O0); 1405 ba(false, profile_continue); 1406 delayed()->nop(); 1407 bind(done); 1408 } 1409 1410 // Store a value at some constant offset from the method data pointer. 1411 1412 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1413 assert(ProfileInterpreter, "must be profiling interpreter"); 1414 st_ptr(value, ImethodDataPtr, constant); 1415 } 1416 1417 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter, 1418 Register bumped_count, 1419 bool decrement) { 1420 assert(ProfileInterpreter, "must be profiling interpreter"); 1421 1422 // Load the counter. 1423 ld_ptr(counter, bumped_count); 1424 1425 if (decrement) { 1426 // Decrement the register. Set condition codes. 1427 subcc(bumped_count, DataLayout::counter_increment, bumped_count); 1428 1429 // If the decrement causes the counter to overflow, stay negative 1430 Label L; 1431 brx(Assembler::negative, true, Assembler::pn, L); 1432 1433 // Store the decremented counter, if it is still negative. 1434 delayed()->st_ptr(bumped_count, counter); 1435 bind(L); 1436 } else { 1437 // Increment the register. Set carry flag. 1438 addcc(bumped_count, DataLayout::counter_increment, bumped_count); 1439 1440 // If the increment causes the counter to overflow, pull back by 1. 1441 assert(DataLayout::counter_increment == 1, "subc works"); 1442 subc(bumped_count, G0, bumped_count); 1443 1444 // Store the incremented counter. 1445 st_ptr(bumped_count, counter); 1446 } 1447 } 1448 1449 // Increment the value at some constant offset from the method data pointer. 1450 1451 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1452 Register bumped_count, 1453 bool decrement) { 1454 // Locate the counter at a fixed offset from the mdp: 1455 Address counter(ImethodDataPtr, constant); 1456 increment_mdp_data_at(counter, bumped_count, decrement); 1457 } 1458 1459 // Increment the value at some non-fixed (reg + constant) offset from 1460 // the method data pointer. 1461 1462 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1463 int constant, 1464 Register bumped_count, 1465 Register scratch2, 1466 bool decrement) { 1467 // Add the constant to reg to get the offset. 1468 add(ImethodDataPtr, reg, scratch2); 1469 Address counter(scratch2, constant); 1470 increment_mdp_data_at(counter, bumped_count, decrement); 1471 } 1472 1473 // Set a flag value at the current method data pointer position. 1474 // Updates a single byte of the header, to avoid races with other header bits. 1475 1476 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1477 Register scratch) { 1478 assert(ProfileInterpreter, "must be profiling interpreter"); 1479 // Load the data header 1480 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch); 1481 1482 // Set the flag 1483 or3(scratch, flag_constant, scratch); 1484 1485 // Store the modified header. 1486 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset())); 1487 } 1488 1489 // Test the location at some offset from the method data pointer. 1490 // If it is not equal to value, branch to the not_equal_continue Label. 1491 // Set condition codes to match the nullness of the loaded value. 1492 1493 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1494 Register value, 1495 Label& not_equal_continue, 1496 Register scratch) { 1497 assert(ProfileInterpreter, "must be profiling interpreter"); 1498 ld_ptr(ImethodDataPtr, offset, scratch); 1499 cmp(value, scratch); 1500 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue); 1501 delayed()->tst(scratch); 1502 } 1503 1504 // Update the method data pointer by the displacement located at some fixed 1505 // offset from the method data pointer. 1506 1507 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1508 Register scratch) { 1509 assert(ProfileInterpreter, "must be profiling interpreter"); 1510 ld_ptr(ImethodDataPtr, offset_of_disp, scratch); 1511 add(ImethodDataPtr, scratch, ImethodDataPtr); 1512 } 1513 1514 // Update the method data pointer by the displacement located at the 1515 // offset (reg + offset_of_disp). 1516 1517 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1518 int offset_of_disp, 1519 Register scratch) { 1520 assert(ProfileInterpreter, "must be profiling interpreter"); 1521 add(reg, offset_of_disp, scratch); 1522 ld_ptr(ImethodDataPtr, scratch, scratch); 1523 add(ImethodDataPtr, scratch, ImethodDataPtr); 1524 } 1525 1526 // Update the method data pointer by a simple constant displacement. 1527 1528 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1529 assert(ProfileInterpreter, "must be profiling interpreter"); 1530 add(ImethodDataPtr, constant, ImethodDataPtr); 1531 } 1532 1533 // Update the method data pointer for a _ret bytecode whose target 1534 // was not among our cached targets. 1535 1536 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1537 Register return_bci) { 1538 assert(ProfileInterpreter, "must be profiling interpreter"); 1539 push(state); 1540 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile 1541 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1542 ld_ptr(l_tmp, return_bci); 1543 pop(state); 1544 } 1545 1546 // Count a taken branch in the bytecodes. 1547 1548 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1549 if (ProfileInterpreter) { 1550 Label profile_continue; 1551 1552 // If no method data exists, go to profile_continue. 1553 test_method_data_pointer(profile_continue); 1554 1555 // We are taking a branch. Increment the taken count. 1556 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count); 1557 1558 // The method data pointer needs to be updated to reflect the new target. 1559 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1560 bind (profile_continue); 1561 } 1562 } 1563 1564 1565 // Count a not-taken branch in the bytecodes. 1566 1567 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) { 1568 if (ProfileInterpreter) { 1569 Label profile_continue; 1570 1571 // If no method data exists, go to profile_continue. 1572 test_method_data_pointer(profile_continue); 1573 1574 // We are taking a branch. Increment the not taken count. 1575 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch); 1576 1577 // The method data pointer needs to be updated to correspond to the 1578 // next bytecode. 1579 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1580 bind (profile_continue); 1581 } 1582 } 1583 1584 1585 // Count a non-virtual call in the bytecodes. 1586 1587 void InterpreterMacroAssembler::profile_call(Register scratch) { 1588 if (ProfileInterpreter) { 1589 Label profile_continue; 1590 1591 // If no method data exists, go to profile_continue. 1592 test_method_data_pointer(profile_continue); 1593 1594 // We are making a call. Increment the count. 1595 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1596 1597 // The method data pointer needs to be updated to reflect the new target. 1598 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1599 bind (profile_continue); 1600 } 1601 } 1602 1603 1604 // Count a final call in the bytecodes. 1605 1606 void InterpreterMacroAssembler::profile_final_call(Register scratch) { 1607 if (ProfileInterpreter) { 1608 Label profile_continue; 1609 1610 // If no method data exists, go to profile_continue. 1611 test_method_data_pointer(profile_continue); 1612 1613 // We are making a call. Increment the count. 1614 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1615 1616 // The method data pointer needs to be updated to reflect the new target. 1617 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1618 bind (profile_continue); 1619 } 1620 } 1621 1622 1623 // Count a virtual call in the bytecodes. 1624 1625 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1626 Register scratch, 1627 bool receiver_can_be_null) { 1628 if (ProfileInterpreter) { 1629 Label profile_continue; 1630 1631 // If no method data exists, go to profile_continue. 1632 test_method_data_pointer(profile_continue); 1633 1634 1635 Label skip_receiver_profile; 1636 if (receiver_can_be_null) { 1637 Label not_null; 1638 tst(receiver); 1639 brx(Assembler::notZero, false, Assembler::pt, not_null); 1640 delayed()->nop(); 1641 // We are making a call. Increment the count for null receiver. 1642 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1643 ba(false, skip_receiver_profile); 1644 delayed()->nop(); 1645 bind(not_null); 1646 } 1647 1648 // Record the receiver type. 1649 record_klass_in_profile(receiver, scratch, true); 1650 bind(skip_receiver_profile); 1651 1652 // The method data pointer needs to be updated to reflect the new target. 1653 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1654 bind (profile_continue); 1655 } 1656 } 1657 1658 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1659 Register receiver, Register scratch, 1660 int start_row, Label& done, bool is_virtual_call) { 1661 if (TypeProfileWidth == 0) { 1662 if (is_virtual_call) { 1663 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1664 } 1665 return; 1666 } 1667 1668 int last_row = VirtualCallData::row_limit() - 1; 1669 assert(start_row <= last_row, "must be work left to do"); 1670 // Test this row for both the receiver and for null. 1671 // Take any of three different outcomes: 1672 // 1. found receiver => increment count and goto done 1673 // 2. found null => keep looking for case 1, maybe allocate this cell 1674 // 3. found something else => keep looking for cases 1 and 2 1675 // Case 3 is handled by a recursive call. 1676 for (int row = start_row; row <= last_row; row++) { 1677 Label next_test; 1678 bool test_for_null_also = (row == start_row); 1679 1680 // See if the receiver is receiver[n]. 1681 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1682 test_mdp_data_at(recvr_offset, receiver, next_test, scratch); 1683 // delayed()->tst(scratch); 1684 1685 // The receiver is receiver[n]. Increment count[n]. 1686 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1687 increment_mdp_data_at(count_offset, scratch); 1688 ba(false, done); 1689 delayed()->nop(); 1690 bind(next_test); 1691 1692 if (test_for_null_also) { 1693 Label found_null; 1694 // Failed the equality check on receiver[n]... Test for null. 1695 if (start_row == last_row) { 1696 // The only thing left to do is handle the null case. 1697 if (is_virtual_call) { 1698 brx(Assembler::zero, false, Assembler::pn, found_null); 1699 delayed()->nop(); 1700 // Receiver did not match any saved receiver and there is no empty row for it. 1701 // Increment total counter to indicate polymorphic case. 1702 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1703 ba(false, done); 1704 delayed()->nop(); 1705 bind(found_null); 1706 } else { 1707 brx(Assembler::notZero, false, Assembler::pt, done); 1708 delayed()->nop(); 1709 } 1710 break; 1711 } 1712 // Since null is rare, make it be the branch-taken case. 1713 brx(Assembler::zero, false, Assembler::pn, found_null); 1714 delayed()->nop(); 1715 1716 // Put all the "Case 3" tests here. 1717 record_klass_in_profile_helper(receiver, scratch, start_row + 1, done, is_virtual_call); 1718 1719 // Found a null. Keep searching for a matching receiver, 1720 // but remember that this is an empty (unused) slot. 1721 bind(found_null); 1722 } 1723 } 1724 1725 // In the fall-through case, we found no matching receiver, but we 1726 // observed the receiver[start_row] is NULL. 1727 1728 // Fill in the receiver field and increment the count. 1729 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1730 set_mdp_data_at(recvr_offset, receiver); 1731 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1732 mov(DataLayout::counter_increment, scratch); 1733 set_mdp_data_at(count_offset, scratch); 1734 if (start_row > 0) { 1735 ba(false, done); 1736 delayed()->nop(); 1737 } 1738 } 1739 1740 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1741 Register scratch, bool is_virtual_call) { 1742 assert(ProfileInterpreter, "must be profiling"); 1743 Label done; 1744 1745 record_klass_in_profile_helper(receiver, scratch, 0, done, is_virtual_call); 1746 1747 bind (done); 1748 } 1749 1750 1751 // Count a ret in the bytecodes. 1752 1753 void InterpreterMacroAssembler::profile_ret(TosState state, 1754 Register return_bci, 1755 Register scratch) { 1756 if (ProfileInterpreter) { 1757 Label profile_continue; 1758 uint row; 1759 1760 // If no method data exists, go to profile_continue. 1761 test_method_data_pointer(profile_continue); 1762 1763 // Update the total ret count. 1764 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1765 1766 for (row = 0; row < RetData::row_limit(); row++) { 1767 Label next_test; 1768 1769 // See if return_bci is equal to bci[n]: 1770 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), 1771 return_bci, next_test, scratch); 1772 1773 // return_bci is equal to bci[n]. Increment the count. 1774 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch); 1775 1776 // The method data pointer needs to be updated to reflect the new target. 1777 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch); 1778 ba(false, profile_continue); 1779 delayed()->nop(); 1780 bind(next_test); 1781 } 1782 1783 update_mdp_for_ret(state, return_bci); 1784 1785 bind (profile_continue); 1786 } 1787 } 1788 1789 // Profile an unexpected null in the bytecodes. 1790 void InterpreterMacroAssembler::profile_null_seen(Register scratch) { 1791 if (ProfileInterpreter) { 1792 Label profile_continue; 1793 1794 // If no method data exists, go to profile_continue. 1795 test_method_data_pointer(profile_continue); 1796 1797 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch); 1798 1799 // The method data pointer needs to be updated. 1800 int mdp_delta = in_bytes(BitData::bit_data_size()); 1801 if (TypeProfileCasts) { 1802 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1803 } 1804 update_mdp_by_constant(mdp_delta); 1805 1806 bind (profile_continue); 1807 } 1808 } 1809 1810 void InterpreterMacroAssembler::profile_typecheck(Register klass, 1811 Register scratch) { 1812 if (ProfileInterpreter) { 1813 Label profile_continue; 1814 1815 // If no method data exists, go to profile_continue. 1816 test_method_data_pointer(profile_continue); 1817 1818 int mdp_delta = in_bytes(BitData::bit_data_size()); 1819 if (TypeProfileCasts) { 1820 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1821 1822 // Record the object type. 1823 record_klass_in_profile(klass, scratch, false); 1824 } 1825 1826 // The method data pointer needs to be updated. 1827 update_mdp_by_constant(mdp_delta); 1828 1829 bind (profile_continue); 1830 } 1831 } 1832 1833 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) { 1834 if (ProfileInterpreter && TypeProfileCasts) { 1835 Label profile_continue; 1836 1837 // If no method data exists, go to profile_continue. 1838 test_method_data_pointer(profile_continue); 1839 1840 int count_offset = in_bytes(CounterData::count_offset()); 1841 // Back up the address, since we have already bumped the mdp. 1842 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1843 1844 // *Decrement* the counter. We expect to see zero or small negatives. 1845 increment_mdp_data_at(count_offset, scratch, true); 1846 1847 bind (profile_continue); 1848 } 1849 } 1850 1851 // Count the default case of a switch construct. 1852 1853 void InterpreterMacroAssembler::profile_switch_default(Register scratch) { 1854 if (ProfileInterpreter) { 1855 Label profile_continue; 1856 1857 // If no method data exists, go to profile_continue. 1858 test_method_data_pointer(profile_continue); 1859 1860 // Update the default case count 1861 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1862 scratch); 1863 1864 // The method data pointer needs to be updated. 1865 update_mdp_by_offset( 1866 in_bytes(MultiBranchData::default_displacement_offset()), 1867 scratch); 1868 1869 bind (profile_continue); 1870 } 1871 } 1872 1873 // Count the index'th case of a switch construct. 1874 1875 void InterpreterMacroAssembler::profile_switch_case(Register index, 1876 Register scratch, 1877 Register scratch2, 1878 Register scratch3) { 1879 if (ProfileInterpreter) { 1880 Label profile_continue; 1881 1882 // If no method data exists, go to profile_continue. 1883 test_method_data_pointer(profile_continue); 1884 1885 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() 1886 set(in_bytes(MultiBranchData::per_case_size()), scratch); 1887 smul(index, scratch, scratch); 1888 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch); 1889 1890 // Update the case count 1891 increment_mdp_data_at(scratch, 1892 in_bytes(MultiBranchData::relative_count_offset()), 1893 scratch2, 1894 scratch3); 1895 1896 // The method data pointer needs to be updated. 1897 update_mdp_by_offset(scratch, 1898 in_bytes(MultiBranchData::relative_displacement_offset()), 1899 scratch2); 1900 1901 bind (profile_continue); 1902 } 1903 } 1904 1905 // add a InterpMonitorElem to stack (see frame_sparc.hpp) 1906 1907 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty, 1908 Register Rtemp, 1909 Register Rtemp2 ) { 1910 1911 Register Rlimit = Lmonitors; 1912 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 1913 assert( (delta & LongAlignmentMask) == 0, 1914 "sizeof BasicObjectLock must be even number of doublewords"); 1915 1916 sub( SP, delta, SP); 1917 sub( Lesp, delta, Lesp); 1918 sub( Lmonitors, delta, Lmonitors); 1919 1920 if (!stack_is_empty) { 1921 1922 // must copy stack contents down 1923 1924 Label start_copying, next; 1925 1926 // untested("monitor stack expansion"); 1927 compute_stack_base(Rtemp); 1928 ba( false, start_copying ); 1929 delayed()->cmp( Rtemp, Rlimit); // done? duplicated below 1930 1931 // note: must copy from low memory upwards 1932 // On entry to loop, 1933 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS) 1934 // Loop mutates Rtemp 1935 1936 bind( next); 1937 1938 st_ptr(Rtemp2, Rtemp, 0); 1939 inc(Rtemp, wordSize); 1940 cmp(Rtemp, Rlimit); // are we done? (duplicated above) 1941 1942 bind( start_copying ); 1943 1944 brx( notEqual, true, pn, next ); 1945 delayed()->ld_ptr( Rtemp, delta, Rtemp2 ); 1946 1947 // done copying stack 1948 } 1949 } 1950 1951 // Locals 1952 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) { 1953 assert_not_delayed(); 1954 sll(index, Interpreter::logStackElementSize, index); 1955 sub(Llocals, index, index); 1956 ld_ptr(index, 0, dst); 1957 // Note: index must hold the effective address--the iinc template uses it 1958 } 1959 1960 // Just like access_local_ptr but the tag is a returnAddress 1961 void InterpreterMacroAssembler::access_local_returnAddress(Register index, 1962 Register dst ) { 1963 assert_not_delayed(); 1964 sll(index, Interpreter::logStackElementSize, index); 1965 sub(Llocals, index, index); 1966 ld_ptr(index, 0, dst); 1967 } 1968 1969 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) { 1970 assert_not_delayed(); 1971 sll(index, Interpreter::logStackElementSize, index); 1972 sub(Llocals, index, index); 1973 ld(index, 0, dst); 1974 // Note: index must hold the effective address--the iinc template uses it 1975 } 1976 1977 1978 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) { 1979 assert_not_delayed(); 1980 sll(index, Interpreter::logStackElementSize, index); 1981 sub(Llocals, index, index); 1982 // First half stored at index n+1 (which grows down from Llocals[n]) 1983 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst); 1984 } 1985 1986 1987 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) { 1988 assert_not_delayed(); 1989 sll(index, Interpreter::logStackElementSize, index); 1990 sub(Llocals, index, index); 1991 ldf(FloatRegisterImpl::S, index, 0, dst); 1992 } 1993 1994 1995 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) { 1996 assert_not_delayed(); 1997 sll(index, Interpreter::logStackElementSize, index); 1998 sub(Llocals, index, index); 1999 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst); 2000 } 2001 2002 2003 #ifdef ASSERT 2004 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) { 2005 Label L; 2006 2007 assert(Rindex != Rscratch, "Registers cannot be same"); 2008 assert(Rindex != Rscratch1, "Registers cannot be same"); 2009 assert(Rlimit != Rscratch, "Registers cannot be same"); 2010 assert(Rlimit != Rscratch1, "Registers cannot be same"); 2011 assert(Rscratch1 != Rscratch, "Registers cannot be same"); 2012 2013 // untested("reg area corruption"); 2014 add(Rindex, offset, Rscratch); 2015 add(Rlimit, 64 + STACK_BIAS, Rscratch1); 2016 cmp(Rscratch, Rscratch1); 2017 brx(Assembler::greaterEqualUnsigned, false, pn, L); 2018 delayed()->nop(); 2019 stop("regsave area is being clobbered"); 2020 bind(L); 2021 } 2022 #endif // ASSERT 2023 2024 2025 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) { 2026 assert_not_delayed(); 2027 sll(index, Interpreter::logStackElementSize, index); 2028 sub(Llocals, index, index); 2029 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);) 2030 st(src, index, 0); 2031 } 2032 2033 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) { 2034 assert_not_delayed(); 2035 sll(index, Interpreter::logStackElementSize, index); 2036 sub(Llocals, index, index); 2037 #ifdef ASSERT 2038 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2039 #endif 2040 st_ptr(src, index, 0); 2041 } 2042 2043 2044 2045 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) { 2046 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n)); 2047 } 2048 2049 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) { 2050 assert_not_delayed(); 2051 sll(index, Interpreter::logStackElementSize, index); 2052 sub(Llocals, index, index); 2053 #ifdef ASSERT 2054 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2055 #endif 2056 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1 2057 } 2058 2059 2060 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) { 2061 assert_not_delayed(); 2062 sll(index, Interpreter::logStackElementSize, index); 2063 sub(Llocals, index, index); 2064 #ifdef ASSERT 2065 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2066 #endif 2067 stf(FloatRegisterImpl::S, src, index, 0); 2068 } 2069 2070 2071 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) { 2072 assert_not_delayed(); 2073 sll(index, Interpreter::logStackElementSize, index); 2074 sub(Llocals, index, index); 2075 #ifdef ASSERT 2076 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2077 #endif 2078 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1)); 2079 } 2080 2081 2082 int InterpreterMacroAssembler::top_most_monitor_byte_offset() { 2083 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 2084 int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong); 2085 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS; 2086 } 2087 2088 2089 Address InterpreterMacroAssembler::top_most_monitor() { 2090 return Address(FP, top_most_monitor_byte_offset()); 2091 } 2092 2093 2094 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) { 2095 add( Lesp, wordSize, Rdest ); 2096 } 2097 2098 #endif /* CC_INTERP */ 2099 2100 void InterpreterMacroAssembler::increment_invocation_counter( Register Rtmp, Register Rtmp2 ) { 2101 assert(UseCompiler, "incrementing must be useful"); 2102 #ifdef CC_INTERP 2103 Address inv_counter(G5_method, methodOopDesc::invocation_counter_offset() + 2104 InvocationCounter::counter_offset()); 2105 Address be_counter (G5_method, methodOopDesc::backedge_counter_offset() + 2106 InvocationCounter::counter_offset()); 2107 #else 2108 Address inv_counter(Lmethod, methodOopDesc::invocation_counter_offset() + 2109 InvocationCounter::counter_offset()); 2110 Address be_counter (Lmethod, methodOopDesc::backedge_counter_offset() + 2111 InvocationCounter::counter_offset()); 2112 #endif /* CC_INTERP */ 2113 int delta = InvocationCounter::count_increment; 2114 2115 // Load each counter in a register 2116 ld( inv_counter, Rtmp ); 2117 ld( be_counter, Rtmp2 ); 2118 2119 assert( is_simm13( delta ), " delta too large."); 2120 2121 // Add the delta to the invocation counter and store the result 2122 add( Rtmp, delta, Rtmp ); 2123 2124 // Mask the backedge counter 2125 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2126 2127 // Store value 2128 st( Rtmp, inv_counter); 2129 2130 // Add invocation counter + backedge counter 2131 add( Rtmp, Rtmp2, Rtmp); 2132 2133 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 2134 } 2135 2136 void InterpreterMacroAssembler::increment_backedge_counter( Register Rtmp, Register Rtmp2 ) { 2137 assert(UseCompiler, "incrementing must be useful"); 2138 #ifdef CC_INTERP 2139 Address be_counter (G5_method, methodOopDesc::backedge_counter_offset() + 2140 InvocationCounter::counter_offset()); 2141 Address inv_counter(G5_method, methodOopDesc::invocation_counter_offset() + 2142 InvocationCounter::counter_offset()); 2143 #else 2144 Address be_counter (Lmethod, methodOopDesc::backedge_counter_offset() + 2145 InvocationCounter::counter_offset()); 2146 Address inv_counter(Lmethod, methodOopDesc::invocation_counter_offset() + 2147 InvocationCounter::counter_offset()); 2148 #endif /* CC_INTERP */ 2149 int delta = InvocationCounter::count_increment; 2150 // Load each counter in a register 2151 ld( be_counter, Rtmp ); 2152 ld( inv_counter, Rtmp2 ); 2153 2154 // Add the delta to the backedge counter 2155 add( Rtmp, delta, Rtmp ); 2156 2157 // Mask the invocation counter, add to backedge counter 2158 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2159 2160 // and store the result to memory 2161 st( Rtmp, be_counter ); 2162 2163 // Add backedge + invocation counter 2164 add( Rtmp, Rtmp2, Rtmp ); 2165 2166 // Note that this macro must leave backedge_count + invocation_count in Rtmp! 2167 } 2168 2169 #ifndef CC_INTERP 2170 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count, 2171 Register branch_bcp, 2172 Register Rtmp ) { 2173 Label did_not_overflow; 2174 Label overflow_with_error; 2175 assert_different_registers(backedge_count, Rtmp, branch_bcp); 2176 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 2177 2178 AddressLiteral limit(&InvocationCounter::InterpreterBackwardBranchLimit); 2179 load_contents(limit, Rtmp); 2180 cmp(backedge_count, Rtmp); 2181 br(Assembler::lessUnsigned, false, Assembler::pt, did_not_overflow); 2182 delayed()->nop(); 2183 2184 // When ProfileInterpreter is on, the backedge_count comes from the 2185 // methodDataOop, which value does not get reset on the call to 2186 // frequency_counter_overflow(). To avoid excessive calls to the overflow 2187 // routine while the method is being compiled, add a second test to make sure 2188 // the overflow function is called only once every overflow_frequency. 2189 if (ProfileInterpreter) { 2190 const int overflow_frequency = 1024; 2191 andcc(backedge_count, overflow_frequency-1, Rtmp); 2192 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow); 2193 delayed()->nop(); 2194 } 2195 2196 // overflow in loop, pass branch bytecode 2197 set(6,Rtmp); 2198 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp); 2199 2200 // Was an OSR adapter generated? 2201 // O0 = osr nmethod 2202 tst(O0); 2203 brx(Assembler::zero, false, Assembler::pn, overflow_with_error); 2204 delayed()->nop(); 2205 2206 // Has the nmethod been invalidated already? 2207 ld(O0, nmethod::entry_bci_offset(), O2); 2208 cmp(O2, InvalidOSREntryBci); 2209 br(Assembler::equal, false, Assembler::pn, overflow_with_error); 2210 delayed()->nop(); 2211 2212 // migrate the interpreter frame off of the stack 2213 2214 mov(G2_thread, L7); 2215 // save nmethod 2216 mov(O0, L6); 2217 set_last_Java_frame(SP, noreg); 2218 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7); 2219 reset_last_Java_frame(); 2220 mov(L7, G2_thread); 2221 2222 // move OSR nmethod to I1 2223 mov(L6, I1); 2224 2225 // OSR buffer to I0 2226 mov(O0, I0); 2227 2228 // remove the interpreter frame 2229 restore(I5_savedSP, 0, SP); 2230 2231 // Jump to the osr code. 2232 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2); 2233 jmp(O2, G0); 2234 delayed()->nop(); 2235 2236 bind(overflow_with_error); 2237 2238 bind(did_not_overflow); 2239 } 2240 2241 2242 2243 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) { 2244 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); } 2245 } 2246 2247 2248 // local helper function for the verify_oop_or_return_address macro 2249 static bool verify_return_address(methodOopDesc* m, int bci) { 2250 #ifndef PRODUCT 2251 address pc = (address)(m->constMethod()) 2252 + in_bytes(constMethodOopDesc::codes_offset()) + bci; 2253 // assume it is a valid return address if it is inside m and is preceded by a jsr 2254 if (!m->contains(pc)) return false; 2255 address jsr_pc; 2256 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2257 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2258 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2259 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2260 #endif // PRODUCT 2261 return false; 2262 } 2263 2264 2265 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2266 if (!VerifyOops) return; 2267 // the VM documentation for the astore[_wide] bytecode allows 2268 // the TOS to be not only an oop but also a return address 2269 Label test; 2270 Label skip; 2271 // See if it is an address (in the current method): 2272 2273 mov(reg, Rtmp); 2274 const int log2_bytecode_size_limit = 16; 2275 srl(Rtmp, log2_bytecode_size_limit, Rtmp); 2276 br_notnull( Rtmp, false, pt, test ); 2277 delayed()->nop(); 2278 2279 // %%% should use call_VM_leaf here? 2280 save_frame_and_mov(0, Lmethod, O0, reg, O1); 2281 save_thread(L7_thread_cache); 2282 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none); 2283 delayed()->nop(); 2284 restore_thread(L7_thread_cache); 2285 br_notnull( O0, false, pt, skip ); 2286 delayed()->restore(); 2287 2288 // Perform a more elaborate out-of-line call 2289 // Not an address; verify it: 2290 bind(test); 2291 verify_oop(reg); 2292 bind(skip); 2293 } 2294 2295 2296 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2297 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); 2298 } 2299 #endif /* CC_INTERP */ 2300 2301 // Inline assembly for: 2302 // 2303 // if (thread is in interp_only_mode) { 2304 // InterpreterRuntime::post_method_entry(); 2305 // } 2306 // if (DTraceMethodProbes) { 2307 // SharedRuntime::dtrace_method_entry(method, receiver); 2308 // } 2309 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 2310 // SharedRuntime::rc_trace_method_entry(method, receiver); 2311 // } 2312 2313 void InterpreterMacroAssembler::notify_method_entry() { 2314 2315 // C++ interpreter only uses this for native methods. 2316 2317 // Whenever JVMTI puts a thread in interp_only_mode, method 2318 // entry/exit events are sent for that thread to track stack 2319 // depth. If it is possible to enter interp_only_mode we add 2320 // the code to check if the event should be sent. 2321 if (JvmtiExport::can_post_interpreter_events()) { 2322 Label L; 2323 Register temp_reg = O5; 2324 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2325 ld(interp_only, temp_reg); 2326 tst(temp_reg); 2327 br(zero, false, pt, L); 2328 delayed()->nop(); 2329 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); 2330 bind(L); 2331 } 2332 2333 { 2334 Register temp_reg = O5; 2335 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2336 call_VM_leaf(noreg, 2337 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 2338 G2_thread, Lmethod); 2339 } 2340 2341 // RedefineClasses() tracing support for obsolete method entry 2342 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 2343 call_VM_leaf(noreg, 2344 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 2345 G2_thread, Lmethod); 2346 } 2347 } 2348 2349 2350 // Inline assembly for: 2351 // 2352 // if (thread is in interp_only_mode) { 2353 // // save result 2354 // InterpreterRuntime::post_method_exit(); 2355 // // restore result 2356 // } 2357 // if (DTraceMethodProbes) { 2358 // SharedRuntime::dtrace_method_exit(thread, method); 2359 // } 2360 // 2361 // Native methods have their result stored in d_tmp and l_tmp 2362 // Java methods have their result stored in the expression stack 2363 2364 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, 2365 TosState state, 2366 NotifyMethodExitMode mode) { 2367 // C++ interpreter only uses this for native methods. 2368 2369 // Whenever JVMTI puts a thread in interp_only_mode, method 2370 // entry/exit events are sent for that thread to track stack 2371 // depth. If it is possible to enter interp_only_mode we add 2372 // the code to check if the event should be sent. 2373 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2374 Label L; 2375 Register temp_reg = O5; 2376 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2377 ld(interp_only, temp_reg); 2378 tst(temp_reg); 2379 br(zero, false, pt, L); 2380 delayed()->nop(); 2381 2382 // Note: frame::interpreter_frame_result has a dependency on how the 2383 // method result is saved across the call to post_method_exit. For 2384 // native methods it assumes the result registers are saved to 2385 // l_scratch and d_scratch. If this changes then the interpreter_frame_result 2386 // implementation will need to be updated too. 2387 2388 save_return_value(state, is_native_method); 2389 call_VM(noreg, 2390 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2391 restore_return_value(state, is_native_method); 2392 bind(L); 2393 } 2394 2395 { 2396 Register temp_reg = O5; 2397 // Dtrace notification 2398 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2399 save_return_value(state, is_native_method); 2400 call_VM_leaf( 2401 noreg, 2402 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2403 G2_thread, Lmethod); 2404 restore_return_value(state, is_native_method); 2405 } 2406 } 2407 2408 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) { 2409 #ifdef CC_INTERP 2410 // result potentially in O0/O1: save it across calls 2411 stf(FloatRegisterImpl::D, F0, STATE(_native_fresult)); 2412 #ifdef _LP64 2413 stx(O0, STATE(_native_lresult)); 2414 #else 2415 std(O0, STATE(_native_lresult)); 2416 #endif 2417 #else // CC_INTERP 2418 if (is_native_call) { 2419 stf(FloatRegisterImpl::D, F0, d_tmp); 2420 #ifdef _LP64 2421 stx(O0, l_tmp); 2422 #else 2423 std(O0, l_tmp); 2424 #endif 2425 } else { 2426 push(state); 2427 } 2428 #endif // CC_INTERP 2429 } 2430 2431 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) { 2432 #ifdef CC_INTERP 2433 ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0); 2434 #ifdef _LP64 2435 ldx(STATE(_native_lresult), O0); 2436 #else 2437 ldd(STATE(_native_lresult), O0); 2438 #endif 2439 #else // CC_INTERP 2440 if (is_native_call) { 2441 ldf(FloatRegisterImpl::D, d_tmp, F0); 2442 #ifdef _LP64 2443 ldx(l_tmp, O0); 2444 #else 2445 ldd(l_tmp, O0); 2446 #endif 2447 } else { 2448 pop(state); 2449 } 2450 #endif // CC_INTERP 2451 } 2452 2453 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 2454 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 2455 int increment, int mask, 2456 Register scratch1, Register scratch2, 2457 Condition cond, Label *where) { 2458 ld(counter_addr, scratch1); 2459 add(scratch1, increment, scratch1); 2460 if (is_simm13(mask)) { 2461 andcc(scratch1, mask, G0); 2462 } else { 2463 set(mask, scratch2); 2464 andcc(scratch1, scratch2, G0); 2465 } 2466 br(cond, false, Assembler::pn, *where); 2467 delayed()->st(scratch1, counter_addr); 2468 }