1 /* 2 * Copyright (c) 1997, 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 "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/methodData.hpp" 32 #include "oops/method.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 #include "runtime/thread.inline.hpp" 40 41 #ifndef CC_INTERP 42 #ifndef FAST_DISPATCH 43 #define FAST_DISPATCH 1 44 #endif 45 #undef FAST_DISPATCH 46 47 // Implementation of InterpreterMacroAssembler 48 49 // This file specializes the assember with interpreter-specific macros 50 51 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS); 52 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS); 53 54 #else // CC_INTERP 55 #ifndef STATE 56 #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name)) 57 #endif // STATE 58 59 #endif // CC_INTERP 60 61 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) { 62 // Note: this algorithm is also used by C1's OSR entry sequence. 63 // Any changes should also be applied to CodeEmitter::emit_osr_entry(). 64 assert_different_registers(args_size, locals_size); 65 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted. 66 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words 67 // Use br/mov combination because it works on both V8 and V9 and is 68 // faster. 69 Label skip_move; 70 br(Assembler::negative, true, Assembler::pt, skip_move); 71 delayed()->mov(G0, delta); 72 bind(skip_move); 73 round_to(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned) 74 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes 75 } 76 77 #ifndef CC_INTERP 78 79 // Dispatch code executed in the prolog of a bytecode which does not do it's 80 // own dispatch. The dispatch address is computed and placed in IdispatchAddress 81 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 82 assert_not_delayed(); 83 #ifdef FAST_DISPATCH 84 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since 85 // they both use I2. 86 assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive"); 87 ldub(Lbcp, bcp_incr, Lbyte_code); // load next bytecode 88 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); 89 // add offset to correct dispatch table 90 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 91 ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr 92 #else 93 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 94 // dispatch table to use 95 AddressLiteral tbl(Interpreter::dispatch_table(state)); 96 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 97 set(tbl, G3_scratch); // compute addr of table 98 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr 99 #endif 100 } 101 102 103 // Dispatch code executed in the epilog of a bytecode which does not do it's 104 // own dispatch. The dispatch address in IdispatchAddress is used for the 105 // dispatch. 106 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 107 assert_not_delayed(); 108 verify_FPU(1, state); 109 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 110 jmp( IdispatchAddress, 0 ); 111 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); 112 else delayed()->nop(); 113 } 114 115 116 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { 117 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) 118 assert_not_delayed(); 119 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 120 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr); 121 } 122 123 124 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) { 125 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) 126 assert_not_delayed(); 127 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 128 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false); 129 } 130 131 132 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 133 // load current bytecode 134 assert_not_delayed(); 135 ldub( Lbcp, 0, Lbyte_code); // load next bytecode 136 dispatch_base(state, table); 137 } 138 139 140 void InterpreterMacroAssembler::call_VM_leaf_base( 141 Register java_thread, 142 address entry_point, 143 int number_of_arguments 144 ) { 145 if (!java_thread->is_valid()) 146 java_thread = L7_thread_cache; 147 // super call 148 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments); 149 } 150 151 152 void InterpreterMacroAssembler::call_VM_base( 153 Register oop_result, 154 Register java_thread, 155 Register last_java_sp, 156 address entry_point, 157 int number_of_arguments, 158 bool check_exception 159 ) { 160 if (!java_thread->is_valid()) 161 java_thread = L7_thread_cache; 162 // See class ThreadInVMfromInterpreter, which assumes that the interpreter 163 // takes responsibility for setting its own thread-state on call-out. 164 // However, ThreadInVMfromInterpreter resets the state to "in_Java". 165 166 //save_bcp(); // save bcp 167 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception); 168 //restore_bcp(); // restore bcp 169 //restore_locals(); // restore locals pointer 170 } 171 172 173 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 174 if (JvmtiExport::can_pop_frame()) { 175 Label L; 176 177 // Check the "pending popframe condition" flag in the current thread 178 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg); 179 180 // Initiate popframe handling only if it is not already being processed. If the flag 181 // has the popframe_processing bit set, it means that this code is called *during* popframe 182 // handling - we don't want to reenter. 183 btst(JavaThread::popframe_pending_bit, scratch_reg); 184 br(zero, false, pt, L); 185 delayed()->nop(); 186 btst(JavaThread::popframe_processing_bit, scratch_reg); 187 br(notZero, false, pt, L); 188 delayed()->nop(); 189 190 // Call Interpreter::remove_activation_preserving_args_entry() to get the 191 // address of the same-named entrypoint in the generated interpreter code. 192 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 193 194 // Jump to Interpreter::_remove_activation_preserving_args_entry 195 jmpl(O0, G0, G0); 196 delayed()->nop(); 197 bind(L); 198 } 199 } 200 201 202 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 203 Register thr_state = G4_scratch; 204 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); 205 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset()); 206 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset()); 207 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset()); 208 switch (state) { 209 case ltos: ld_long(val_addr, Otos_l); break; 210 case atos: ld_ptr(oop_addr, Otos_l); 211 st_ptr(G0, oop_addr); break; 212 case btos: // fall through 213 case ctos: // fall through 214 case stos: // fall through 215 case itos: ld(val_addr, Otos_l1); break; 216 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break; 217 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break; 218 case vtos: /* nothing to do */ break; 219 default : ShouldNotReachHere(); 220 } 221 // Clean up tos value in the jvmti thread state 222 or3(G0, ilgl, G3_scratch); 223 stw(G3_scratch, tos_addr); 224 st_long(G0, val_addr); 225 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 226 } 227 228 229 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 230 if (JvmtiExport::can_force_early_return()) { 231 Label L; 232 Register thr_state = G3_scratch; 233 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); 234 br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit; 235 236 // Initiate earlyret handling only if it is not already being processed. 237 // If the flag has the earlyret_processing bit set, it means that this code 238 // is called *during* earlyret handling - we don't want to reenter. 239 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch); 240 cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L); 241 242 // Call Interpreter::remove_activation_early_entry() to get the address of the 243 // same-named entrypoint in the generated interpreter code 244 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1); 245 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1); 246 247 // Jump to Interpreter::_remove_activation_early_entry 248 jmpl(O0, G0, G0); 249 delayed()->nop(); 250 bind(L); 251 } 252 } 253 254 255 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) { 256 mov(arg_1, O0); 257 mov(arg_2, O1); 258 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2); 259 } 260 #endif /* CC_INTERP */ 261 262 263 #ifndef CC_INTERP 264 265 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) { 266 assert_not_delayed(); 267 dispatch_Lbyte_code(state, table); 268 } 269 270 271 void InterpreterMacroAssembler::dispatch_normal(TosState state) { 272 dispatch_base(state, Interpreter::normal_table(state)); 273 } 274 275 276 void InterpreterMacroAssembler::dispatch_only(TosState state) { 277 dispatch_base(state, Interpreter::dispatch_table(state)); 278 } 279 280 281 // common code to dispatch and dispatch_only 282 // dispatch value in Lbyte_code and increment Lbcp 283 284 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) { 285 verify_FPU(1, state); 286 // %%%%% maybe implement +VerifyActivationFrameSize here 287 //verify_thread(); //too slow; we will just verify on method entry & exit 288 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 289 #ifdef FAST_DISPATCH 290 if (table == Interpreter::dispatch_table(state)) { 291 // use IdispatchTables 292 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); 293 // add offset to correct dispatch table 294 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 295 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // get entry addr 296 } else { 297 #endif 298 // dispatch table to use 299 AddressLiteral tbl(table); 300 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 301 set(tbl, G3_scratch); // compute addr of table 302 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr 303 #ifdef FAST_DISPATCH 304 } 305 #endif 306 jmp( G3_scratch, 0 ); 307 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); 308 else delayed()->nop(); 309 } 310 311 312 // Helpers for expression stack 313 314 // Longs and doubles are Category 2 computational types in the 315 // JVM specification (section 3.11.1) and take 2 expression stack or 316 // local slots. 317 // Aligning them on 32 bit with tagged stacks is hard because the code generated 318 // for the dup* bytecodes depends on what types are already on the stack. 319 // If the types are split into the two stack/local slots, that is much easier 320 // (and we can use 0 for non-reference tags). 321 322 // Known good alignment in _LP64 but unknown otherwise 323 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) { 324 assert_not_delayed(); 325 326 #ifdef _LP64 327 ldf(FloatRegisterImpl::D, r1, offset, d); 328 #else 329 ldf(FloatRegisterImpl::S, r1, offset, d); 330 ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize, d->successor()); 331 #endif 332 } 333 334 // Known good alignment in _LP64 but unknown otherwise 335 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) { 336 assert_not_delayed(); 337 338 #ifdef _LP64 339 stf(FloatRegisterImpl::D, d, r1, offset); 340 // store something more useful here 341 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);) 342 #else 343 stf(FloatRegisterImpl::S, d, r1, offset); 344 stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize); 345 #endif 346 } 347 348 349 // Known good alignment in _LP64 but unknown otherwise 350 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) { 351 assert_not_delayed(); 352 #ifdef _LP64 353 ldx(r1, offset, rd); 354 #else 355 ld(r1, offset, rd); 356 ld(r1, offset + Interpreter::stackElementSize, rd->successor()); 357 #endif 358 } 359 360 // Known good alignment in _LP64 but unknown otherwise 361 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) { 362 assert_not_delayed(); 363 364 #ifdef _LP64 365 stx(l, r1, offset); 366 // store something more useful here 367 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);) 368 #else 369 st(l, r1, offset); 370 st(l->successor(), r1, offset + Interpreter::stackElementSize); 371 #endif 372 } 373 374 void InterpreterMacroAssembler::pop_i(Register r) { 375 assert_not_delayed(); 376 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r); 377 inc(Lesp, Interpreter::stackElementSize); 378 debug_only(verify_esp(Lesp)); 379 } 380 381 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) { 382 assert_not_delayed(); 383 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); 384 inc(Lesp, Interpreter::stackElementSize); 385 debug_only(verify_esp(Lesp)); 386 } 387 388 void InterpreterMacroAssembler::pop_l(Register r) { 389 assert_not_delayed(); 390 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r); 391 inc(Lesp, 2*Interpreter::stackElementSize); 392 debug_only(verify_esp(Lesp)); 393 } 394 395 396 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) { 397 assert_not_delayed(); 398 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f); 399 inc(Lesp, Interpreter::stackElementSize); 400 debug_only(verify_esp(Lesp)); 401 } 402 403 404 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) { 405 assert_not_delayed(); 406 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f); 407 inc(Lesp, 2*Interpreter::stackElementSize); 408 debug_only(verify_esp(Lesp)); 409 } 410 411 412 void InterpreterMacroAssembler::push_i(Register r) { 413 assert_not_delayed(); 414 debug_only(verify_esp(Lesp)); 415 st(r, Lesp, 0); 416 dec(Lesp, Interpreter::stackElementSize); 417 } 418 419 void InterpreterMacroAssembler::push_ptr(Register r) { 420 assert_not_delayed(); 421 st_ptr(r, Lesp, 0); 422 dec(Lesp, Interpreter::stackElementSize); 423 } 424 425 // remember: our convention for longs in SPARC is: 426 // O0 (Otos_l1) has high-order part in first word, 427 // O1 (Otos_l2) has low-order part in second word 428 429 void InterpreterMacroAssembler::push_l(Register r) { 430 assert_not_delayed(); 431 debug_only(verify_esp(Lesp)); 432 // Longs are stored in memory-correct order, even if unaligned. 433 int offset = -Interpreter::stackElementSize; 434 store_unaligned_long(r, Lesp, offset); 435 dec(Lesp, 2 * Interpreter::stackElementSize); 436 } 437 438 439 void InterpreterMacroAssembler::push_f(FloatRegister f) { 440 assert_not_delayed(); 441 debug_only(verify_esp(Lesp)); 442 stf(FloatRegisterImpl::S, f, Lesp, 0); 443 dec(Lesp, Interpreter::stackElementSize); 444 } 445 446 447 void InterpreterMacroAssembler::push_d(FloatRegister d) { 448 assert_not_delayed(); 449 debug_only(verify_esp(Lesp)); 450 // Longs are stored in memory-correct order, even if unaligned. 451 int offset = -Interpreter::stackElementSize; 452 store_unaligned_double(d, Lesp, offset); 453 dec(Lesp, 2 * Interpreter::stackElementSize); 454 } 455 456 457 void InterpreterMacroAssembler::push(TosState state) { 458 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 459 switch (state) { 460 case atos: push_ptr(); break; 461 case btos: push_i(); break; 462 case ctos: 463 case stos: push_i(); break; 464 case itos: push_i(); break; 465 case ltos: push_l(); break; 466 case ftos: push_f(); break; 467 case dtos: push_d(); break; 468 case vtos: /* nothing to do */ break; 469 default : ShouldNotReachHere(); 470 } 471 } 472 473 474 void InterpreterMacroAssembler::pop(TosState state) { 475 switch (state) { 476 case atos: pop_ptr(); break; 477 case btos: pop_i(); break; 478 case ctos: 479 case stos: pop_i(); break; 480 case itos: pop_i(); break; 481 case ltos: pop_l(); break; 482 case ftos: pop_f(); break; 483 case dtos: pop_d(); break; 484 case vtos: /* nothing to do */ break; 485 default : ShouldNotReachHere(); 486 } 487 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 488 } 489 490 491 // Helpers for swap and dup 492 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 493 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val); 494 } 495 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 496 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n)); 497 } 498 499 500 void InterpreterMacroAssembler::load_receiver(Register param_count, 501 Register recv) { 502 sll(param_count, Interpreter::logStackElementSize, param_count); 503 ld_ptr(Lesp, param_count, recv); // gets receiver oop 504 } 505 506 void InterpreterMacroAssembler::empty_expression_stack() { 507 // Reset Lesp. 508 sub( Lmonitors, wordSize, Lesp ); 509 510 // Reset SP by subtracting more space from Lesp. 511 Label done; 512 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!"); 513 514 // A native does not need to do this, since its callee does not change SP. 515 ld(Lmethod, Method::access_flags_offset(), Gframe_size); // Load access flags. 516 btst(JVM_ACC_NATIVE, Gframe_size); 517 br(Assembler::notZero, false, Assembler::pt, done); 518 delayed()->nop(); 519 520 // Compute max expression stack+register save area 521 lduh(Lmethod, in_bytes(Method::max_stack_offset()), Gframe_size); // Load max stack. 522 add( Gframe_size, frame::memory_parameter_word_sp_offset, Gframe_size ); 523 524 // 525 // now set up a stack frame with the size computed above 526 // 527 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below 528 sll( Gframe_size, LogBytesPerWord, Gframe_size ); 529 sub( Lesp, Gframe_size, Gframe_size ); 530 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary 531 debug_only(verify_sp(Gframe_size, G4_scratch)); 532 #ifdef _LP64 533 sub(Gframe_size, STACK_BIAS, Gframe_size ); 534 #endif 535 mov(Gframe_size, SP); 536 537 bind(done); 538 } 539 540 541 #ifdef ASSERT 542 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) { 543 Label Bad, OK; 544 545 // Saved SP must be aligned. 546 #ifdef _LP64 547 btst(2*BytesPerWord-1, Rsp); 548 #else 549 btst(LongAlignmentMask, Rsp); 550 #endif 551 br(Assembler::notZero, false, Assembler::pn, Bad); 552 delayed()->nop(); 553 554 // Saved SP, plus register window size, must not be above FP. 555 add(Rsp, frame::register_save_words * wordSize, Rtemp); 556 #ifdef _LP64 557 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP 558 #endif 559 cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad); 560 561 // Saved SP must not be ridiculously below current SP. 562 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K); 563 set(maxstack, Rtemp); 564 sub(SP, Rtemp, Rtemp); 565 #ifdef _LP64 566 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp 567 #endif 568 cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad); 569 570 ba_short(OK); 571 572 bind(Bad); 573 stop("on return to interpreted call, restored SP is corrupted"); 574 575 bind(OK); 576 } 577 578 579 void InterpreterMacroAssembler::verify_esp(Register Resp) { 580 // about to read or write Resp[0] 581 // make sure it is not in the monitors or the register save area 582 Label OK1, OK2; 583 584 cmp(Resp, Lmonitors); 585 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1); 586 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp); 587 stop("too many pops: Lesp points into monitor area"); 588 bind(OK1); 589 #ifdef _LP64 590 sub(Resp, STACK_BIAS, Resp); 591 #endif 592 cmp(Resp, SP); 593 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2); 594 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp); 595 stop("too many pushes: Lesp points into register window"); 596 bind(OK2); 597 } 598 #endif // ASSERT 599 600 // Load compiled (i2c) or interpreter entry when calling from interpreted and 601 // do the call. Centralized so that all interpreter calls will do the same actions. 602 // If jvmti single stepping is on for a thread we must not call compiled code. 603 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) { 604 605 // Assume we want to go compiled if available 606 607 ld_ptr(G5_method, in_bytes(Method::from_interpreted_offset()), target); 608 609 if (JvmtiExport::can_post_interpreter_events()) { 610 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 611 // compiled code in threads for which the event is enabled. Check here for 612 // interp_only_mode if these events CAN be enabled. 613 verify_thread(); 614 Label skip_compiled_code; 615 616 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 617 ld(interp_only, scratch); 618 cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn); 619 delayed()->ld_ptr(G5_method, in_bytes(Method::interpreter_entry_offset()), target); 620 bind(skip_compiled_code); 621 } 622 623 // the i2c_adapters need Method* in G5_method (right? %%%) 624 // do the call 625 #ifdef ASSERT 626 { 627 Label ok; 628 br_notnull_short(target, Assembler::pt, ok); 629 stop("null entry point"); 630 bind(ok); 631 } 632 #endif // ASSERT 633 634 // Adjust Rret first so Llast_SP can be same as Rret 635 add(Rret, -frame::pc_return_offset, O7); 636 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer 637 // Record SP so we can remove any stack space allocated by adapter transition 638 jmp(target, 0); 639 delayed()->mov(SP, Llast_SP); 640 } 641 642 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) { 643 assert_not_delayed(); 644 645 Label not_taken; 646 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken); 647 else br (cc, false, Assembler::pn, not_taken); 648 delayed()->nop(); 649 650 TemplateTable::branch(false,false); 651 652 bind(not_taken); 653 654 profile_not_taken_branch(G3_scratch); 655 } 656 657 658 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp( 659 int bcp_offset, 660 Register Rtmp, 661 Register Rdst, 662 signedOrNot is_signed, 663 setCCOrNot should_set_CC ) { 664 assert(Rtmp != Rdst, "need separate temp register"); 665 assert_not_delayed(); 666 switch (is_signed) { 667 default: ShouldNotReachHere(); 668 669 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte 670 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte 671 } 672 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte 673 sll( Rdst, BitsPerByte, Rdst); 674 switch (should_set_CC ) { 675 default: ShouldNotReachHere(); 676 677 case set_CC: orcc( Rdst, Rtmp, Rdst ); break; 678 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break; 679 } 680 } 681 682 683 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp( 684 int bcp_offset, 685 Register Rtmp, 686 Register Rdst, 687 setCCOrNot should_set_CC ) { 688 assert(Rtmp != Rdst, "need separate temp register"); 689 assert_not_delayed(); 690 add( Lbcp, bcp_offset, Rtmp); 691 andcc( Rtmp, 3, G0); 692 Label aligned; 693 switch (should_set_CC ) { 694 default: ShouldNotReachHere(); 695 696 case set_CC: break; 697 case dont_set_CC: break; 698 } 699 700 br(Assembler::zero, true, Assembler::pn, aligned); 701 #ifdef _LP64 702 delayed()->ldsw(Rtmp, 0, Rdst); 703 #else 704 delayed()->ld(Rtmp, 0, Rdst); 705 #endif 706 707 ldub(Lbcp, bcp_offset + 3, Rdst); 708 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst); 709 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst); 710 #ifdef _LP64 711 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); 712 #else 713 // Unsigned load is faster than signed on some implementations 714 ldub(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); 715 #endif 716 or3(Rtmp, Rdst, Rdst ); 717 718 bind(aligned); 719 if (should_set_CC == set_CC) tst(Rdst); 720 } 721 722 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register temp, Register index, 723 int bcp_offset, size_t index_size) { 724 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 725 if (index_size == sizeof(u2)) { 726 get_2_byte_integer_at_bcp(bcp_offset, temp, index, Unsigned); 727 } else if (index_size == sizeof(u4)) { 728 assert(EnableInvokeDynamic, "giant index used only for JSR 292"); 729 get_4_byte_integer_at_bcp(bcp_offset, temp, index); 730 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 731 xor3(index, -1, index); // convert to plain index 732 } else if (index_size == sizeof(u1)) { 733 ldub(Lbcp, bcp_offset, index); 734 } else { 735 ShouldNotReachHere(); 736 } 737 } 738 739 740 void InterpreterMacroAssembler::get_cache_and_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 assert_different_registers(cache, tmp); 744 assert_not_delayed(); 745 get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size); 746 // convert from field index to ConstantPoolCacheEntry index and from 747 // word index to byte offset 748 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); 749 add(LcpoolCache, tmp, cache); 750 } 751 752 753 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, 754 Register temp, 755 Register bytecode, 756 int byte_no, 757 int bcp_offset, 758 size_t index_size) { 759 get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size); 760 ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode); 761 const int shift_count = (1 + byte_no) * BitsPerByte; 762 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) || 763 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift), 764 "correct shift count"); 765 srl(bytecode, shift_count, bytecode); 766 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask"); 767 and3(bytecode, ConstantPoolCacheEntry::bytecode_1_mask, bytecode); 768 } 769 770 771 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp, 772 int bcp_offset, size_t index_size) { 773 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 774 assert_different_registers(cache, tmp); 775 assert_not_delayed(); 776 if (index_size == sizeof(u2)) { 777 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); 778 } else { 779 ShouldNotReachHere(); // other sizes not supported here 780 } 781 // convert from field index to ConstantPoolCacheEntry index 782 // and from word index to byte offset 783 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); 784 // skip past the header 785 add(tmp, in_bytes(ConstantPoolCache::base_offset()), tmp); 786 // construct pointer to cache entry 787 add(LcpoolCache, tmp, cache); 788 } 789 790 791 // Load object from cpool->resolved_references(index) 792 void InterpreterMacroAssembler::load_resolved_reference_at_index( 793 Register result, Register index) { 794 assert_different_registers(result, index); 795 assert_not_delayed(); 796 // convert from field index to resolved_references() index and from 797 // word index to byte offset. Since this is a java object, it can be compressed 798 Register tmp = index; // reuse 799 sll(index, LogBytesPerHeapOop, tmp); 800 get_constant_pool(result); 801 // load pointer for resolved_references[] objArray 802 ld_ptr(result, ConstantPool::resolved_references_offset_in_bytes(), result); 803 // JNIHandles::resolve(result) 804 ld_ptr(result, 0, result); 805 // Add in the index 806 add(result, tmp, result); 807 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result); 808 } 809 810 811 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 812 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. 813 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 814 Register Rsuper_klass, 815 Register Rtmp1, 816 Register Rtmp2, 817 Register Rtmp3, 818 Label &ok_is_subtype ) { 819 Label not_subtype; 820 821 // Profile the not-null value's klass. 822 profile_typecheck(Rsub_klass, Rtmp1); 823 824 check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass, 825 Rtmp1, Rtmp2, 826 &ok_is_subtype, ¬_subtype, NULL); 827 828 check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass, 829 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg, 830 &ok_is_subtype, NULL); 831 832 bind(not_subtype); 833 profile_typecheck_failed(Rtmp1); 834 } 835 836 // Separate these two to allow for delay slot in middle 837 // These are used to do a test and full jump to exception-throwing code. 838 839 // %%%%% Could possibly reoptimize this by testing to see if could use 840 // a single conditional branch (i.e. if span is small enough. 841 // If you go that route, than get rid of the split and give up 842 // on the delay-slot hack. 843 844 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition, 845 Label& ok ) { 846 assert_not_delayed(); 847 br(ok_condition, true, pt, ok); 848 // DELAY SLOT 849 } 850 851 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition, 852 Label& ok ) { 853 assert_not_delayed(); 854 bp( ok_condition, true, Assembler::xcc, pt, ok); 855 // DELAY SLOT 856 } 857 858 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition, 859 Label& ok ) { 860 assert_not_delayed(); 861 brx(ok_condition, true, pt, ok); 862 // DELAY SLOT 863 } 864 865 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point, 866 Register Rscratch, 867 Label& ok ) { 868 assert(throw_entry_point != NULL, "entry point must be generated by now"); 869 AddressLiteral dest(throw_entry_point); 870 jump_to(dest, Rscratch); 871 delayed()->nop(); 872 bind(ok); 873 } 874 875 876 // And if you cannot use the delay slot, here is a shorthand: 877 878 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition, 879 address throw_entry_point, 880 Register Rscratch ) { 881 Label ok; 882 if (ok_condition != never) { 883 throw_if_not_1_icc( ok_condition, ok); 884 delayed()->nop(); 885 } 886 throw_if_not_2( throw_entry_point, Rscratch, ok); 887 } 888 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition, 889 address throw_entry_point, 890 Register Rscratch ) { 891 Label ok; 892 if (ok_condition != never) { 893 throw_if_not_1_xcc( ok_condition, ok); 894 delayed()->nop(); 895 } 896 throw_if_not_2( throw_entry_point, Rscratch, ok); 897 } 898 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition, 899 address throw_entry_point, 900 Register Rscratch ) { 901 Label ok; 902 if (ok_condition != never) { 903 throw_if_not_1_x( ok_condition, ok); 904 delayed()->nop(); 905 } 906 throw_if_not_2( throw_entry_point, Rscratch, ok); 907 } 908 909 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res 910 // Note: res is still shy of address by array offset into object. 911 912 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) { 913 assert_not_delayed(); 914 915 verify_oop(array); 916 #ifdef _LP64 917 // sign extend since tos (index) can be a 32bit value 918 sra(index, G0, index); 919 #endif // _LP64 920 921 // check array 922 Label ptr_ok; 923 tst(array); 924 throw_if_not_1_x( notZero, ptr_ok ); 925 delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index 926 throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok); 927 928 Label index_ok; 929 cmp(index, tmp); 930 throw_if_not_1_icc( lessUnsigned, index_ok ); 931 if (index_shift > 0) delayed()->sll(index, index_shift, index); 932 else delayed()->add(array, index, res); // addr - const offset in index 933 // convention: move aberrant index into G3_scratch for exception message 934 mov(index, G3_scratch); 935 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok); 936 937 // add offset if didn't do it in delay slot 938 if (index_shift > 0) add(array, index, res); // addr - const offset in index 939 } 940 941 942 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { 943 assert_not_delayed(); 944 945 // pop array 946 pop_ptr(array); 947 948 // check array 949 index_check_without_pop(array, index, index_shift, tmp, res); 950 } 951 952 953 void InterpreterMacroAssembler::get_const(Register Rdst) { 954 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst); 955 } 956 957 958 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 959 get_const(Rdst); 960 ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 961 } 962 963 964 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 965 get_constant_pool(Rdst); 966 ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 967 } 968 969 970 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 971 get_constant_pool(Rcpool); 972 ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags); 973 } 974 975 976 // unlock if synchronized method 977 // 978 // Unlock the receiver if this is a synchronized method. 979 // Unlock any Java monitors from syncronized blocks. 980 // 981 // If there are locked Java monitors 982 // If throw_monitor_exception 983 // throws IllegalMonitorStateException 984 // Else if install_monitor_exception 985 // installs IllegalMonitorStateException 986 // Else 987 // no error processing 988 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 989 bool throw_monitor_exception, 990 bool install_monitor_exception) { 991 Label unlocked, unlock, no_unlock; 992 993 // get the value of _do_not_unlock_if_synchronized into G1_scratch 994 const Address do_not_unlock_if_synchronized(G2_thread, 995 JavaThread::do_not_unlock_if_synchronized_offset()); 996 ldbool(do_not_unlock_if_synchronized, G1_scratch); 997 stbool(G0, do_not_unlock_if_synchronized); // reset the flag 998 999 // check if synchronized method 1000 const Address access_flags(Lmethod, Method::access_flags_offset()); 1001 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1002 push(state); // save tos 1003 ld(access_flags, G3_scratch); // Load access flags. 1004 btst(JVM_ACC_SYNCHRONIZED, G3_scratch); 1005 br(zero, false, pt, unlocked); 1006 delayed()->nop(); 1007 1008 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 1009 // is set. 1010 cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock); 1011 delayed()->nop(); 1012 1013 // BasicObjectLock will be first in list, since this is a synchronized method. However, need 1014 // to check that the object has not been unlocked by an explicit monitorexit bytecode. 1015 1016 //Intel: if (throw_monitor_exception) ... else ... 1017 // Entry already unlocked, need to throw exception 1018 //... 1019 1020 // pass top-most monitor elem 1021 add( top_most_monitor(), O1 ); 1022 1023 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch); 1024 br_notnull_short(G3_scratch, pt, unlock); 1025 1026 if (throw_monitor_exception) { 1027 // Entry already unlocked need to throw an exception 1028 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1029 should_not_reach_here(); 1030 } else { 1031 // Monitor already unlocked during a stack unroll. 1032 // If requested, install an illegal_monitor_state_exception. 1033 // Continue with stack unrolling. 1034 if (install_monitor_exception) { 1035 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1036 } 1037 ba_short(unlocked); 1038 } 1039 1040 bind(unlock); 1041 1042 unlock_object(O1); 1043 1044 bind(unlocked); 1045 1046 // I0, I1: Might contain return value 1047 1048 // Check that all monitors are unlocked 1049 { Label loop, exception, entry, restart; 1050 1051 Register Rmptr = O0; 1052 Register Rtemp = O1; 1053 Register Rlimit = Lmonitors; 1054 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 1055 assert( (delta & LongAlignmentMask) == 0, 1056 "sizeof BasicObjectLock must be even number of doublewords"); 1057 1058 #ifdef ASSERT 1059 add(top_most_monitor(), Rmptr, delta); 1060 { Label L; 1061 // ensure that Rmptr starts out above (or at) Rlimit 1062 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); 1063 stop("monitor stack has negative size"); 1064 bind(L); 1065 } 1066 #endif 1067 bind(restart); 1068 ba(entry); 1069 delayed()-> 1070 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry 1071 1072 // Entry is still locked, need to throw exception 1073 bind(exception); 1074 if (throw_monitor_exception) { 1075 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1076 should_not_reach_here(); 1077 } else { 1078 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 1079 // Unlock does not block, so don't have to worry about the frame 1080 unlock_object(Rmptr); 1081 if (install_monitor_exception) { 1082 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1083 } 1084 ba_short(restart); 1085 } 1086 1087 bind(loop); 1088 cmp(Rtemp, G0); // check if current entry is used 1089 brx(Assembler::notEqual, false, pn, exception); 1090 delayed()-> 1091 dec(Rmptr, delta); // otherwise advance to next entry 1092 #ifdef ASSERT 1093 { Label L; 1094 // ensure that Rmptr has not somehow stepped below Rlimit 1095 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); 1096 stop("ran off the end of the monitor stack"); 1097 bind(L); 1098 } 1099 #endif 1100 bind(entry); 1101 cmp(Rmptr, Rlimit); // check if bottom reached 1102 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry 1103 delayed()-> 1104 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp); 1105 } 1106 1107 bind(no_unlock); 1108 pop(state); 1109 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1110 } 1111 1112 1113 // remove activation 1114 // 1115 // Unlock the receiver if this is a synchronized method. 1116 // Unlock any Java monitors from syncronized blocks. 1117 // Remove the activation from the stack. 1118 // 1119 // If there are locked Java monitors 1120 // If throw_monitor_exception 1121 // throws IllegalMonitorStateException 1122 // Else if install_monitor_exception 1123 // installs IllegalMonitorStateException 1124 // Else 1125 // no error processing 1126 void InterpreterMacroAssembler::remove_activation(TosState state, 1127 bool throw_monitor_exception, 1128 bool install_monitor_exception) { 1129 1130 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 1131 1132 // save result (push state before jvmti call and pop it afterwards) and notify jvmti 1133 notify_method_exit(false, state, NotifyJVMTI); 1134 1135 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1136 verify_thread(); 1137 1138 // return tos 1139 assert(Otos_l1 == Otos_i, "adjust code below"); 1140 switch (state) { 1141 #ifdef _LP64 1142 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0 1143 #else 1144 case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through // O1 -> I1 1145 #endif 1146 case btos: // fall through 1147 case ctos: 1148 case stos: // fall through 1149 case atos: // fall through 1150 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0 1151 case ftos: // fall through 1152 case dtos: // fall through 1153 case vtos: /* nothing to do */ break; 1154 default : ShouldNotReachHere(); 1155 } 1156 1157 #if defined(COMPILER2) && !defined(_LP64) 1158 if (state == ltos) { 1159 // C2 expects long results in G1 we can't tell if we're returning to interpreted 1160 // or compiled so just be safe use G1 and O0/O1 1161 1162 // Shift bits into high (msb) of G1 1163 sllx(Otos_l1->after_save(), 32, G1); 1164 // Zero extend low bits 1165 srl (Otos_l2->after_save(), 0, Otos_l2->after_save()); 1166 or3 (Otos_l2->after_save(), G1, G1); 1167 } 1168 #endif /* COMPILER2 */ 1169 1170 } 1171 #endif /* CC_INTERP */ 1172 1173 1174 // Lock object 1175 // 1176 // Argument - lock_reg points to the BasicObjectLock to be used for locking, 1177 // it must be initialized with the object to lock 1178 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) { 1179 if (UseHeavyMonitors) { 1180 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1181 } 1182 else { 1183 Register obj_reg = Object; 1184 Register mark_reg = G4_scratch; 1185 Register temp_reg = G1_scratch; 1186 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes()); 1187 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1188 Label done; 1189 1190 Label slow_case; 1191 1192 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg); 1193 1194 // load markOop from object into mark_reg 1195 ld_ptr(mark_addr, mark_reg); 1196 1197 if (UseBiasedLocking) { 1198 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case); 1199 } 1200 1201 // get the address of basicLock on stack that will be stored in the object 1202 // we need a temporary register here as we do not want to clobber lock_reg 1203 // (cas clobbers the destination register) 1204 mov(lock_reg, temp_reg); 1205 // set mark reg to be (markOop of object | UNLOCK_VALUE) 1206 or3(mark_reg, markOopDesc::unlocked_value, mark_reg); 1207 // initialize the box (Must happen before we update the object mark!) 1208 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1209 // compare and exchange object_addr, markOop | 1, stack address of basicLock 1210 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1211 casx_under_lock(mark_addr.base(), mark_reg, temp_reg, 1212 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); 1213 1214 // if the compare and exchange succeeded we are done (we saw an unlocked object) 1215 cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done); 1216 1217 // We did not see an unlocked object so try the fast recursive case 1218 1219 // Check if owner is self by comparing the value in the markOop of object 1220 // with the stack pointer 1221 sub(temp_reg, SP, temp_reg); 1222 #ifdef _LP64 1223 sub(temp_reg, STACK_BIAS, temp_reg); 1224 #endif 1225 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 1226 1227 // Composite "andcc" test: 1228 // (a) %sp -vs- markword proximity check, and, 1229 // (b) verify mark word LSBs == 0 (Stack-locked). 1230 // 1231 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size()) 1232 // Note that the page size used for %sp proximity testing is arbitrary and is 1233 // unrelated to the actual MMU page size. We use a 'logical' page size of 1234 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate 1235 // field of the andcc instruction. 1236 andcc (temp_reg, 0xFFFFF003, G0) ; 1237 1238 // if condition is true we are done and hence we can store 0 in the displaced 1239 // header indicating it is a recursive lock and be done 1240 brx(Assembler::zero, true, Assembler::pt, done); 1241 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1242 1243 // none of the above fast optimizations worked so we have to get into the 1244 // slow case of monitor enter 1245 bind(slow_case); 1246 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1247 1248 bind(done); 1249 } 1250 } 1251 1252 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 1253 // 1254 // Argument - lock_reg points to the BasicObjectLock for lock 1255 // Throw IllegalMonitorException if object is not locked by current thread 1256 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1257 if (UseHeavyMonitors) { 1258 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1259 } else { 1260 Register obj_reg = G3_scratch; 1261 Register mark_reg = G4_scratch; 1262 Register displaced_header_reg = G1_scratch; 1263 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes()); 1264 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1265 Label done; 1266 1267 if (UseBiasedLocking) { 1268 // load the object out of the BasicObjectLock 1269 ld_ptr(lockobj_addr, obj_reg); 1270 biased_locking_exit(mark_addr, mark_reg, done, true); 1271 st_ptr(G0, lockobj_addr); // free entry 1272 } 1273 1274 // Test first if we are in the fast recursive case 1275 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes()); 1276 ld_ptr(lock_addr, displaced_header_reg); 1277 br_null(displaced_header_reg, true, Assembler::pn, done); 1278 delayed()->st_ptr(G0, lockobj_addr); // free entry 1279 1280 // See if it is still a light weight lock, if so we just unlock 1281 // the object and we are done 1282 1283 if (!UseBiasedLocking) { 1284 // load the object out of the BasicObjectLock 1285 ld_ptr(lockobj_addr, obj_reg); 1286 } 1287 1288 // we have the displaced header in displaced_header_reg 1289 // we expect to see the stack address of the basicLock in case the 1290 // lock is still a light weight lock (lock_reg) 1291 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1292 casx_under_lock(mark_addr.base(), lock_reg, displaced_header_reg, 1293 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); 1294 cmp(lock_reg, displaced_header_reg); 1295 brx(Assembler::equal, true, Assembler::pn, done); 1296 delayed()->st_ptr(G0, lockobj_addr); // free entry 1297 1298 // The lock has been converted into a heavy lock and hence 1299 // we need to get into the slow case 1300 1301 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1302 1303 bind(done); 1304 } 1305 } 1306 1307 #ifndef CC_INTERP 1308 1309 // Get the method data pointer from the Method* and set the 1310 // specified register to its value. 1311 1312 void InterpreterMacroAssembler::set_method_data_pointer() { 1313 assert(ProfileInterpreter, "must be profiling interpreter"); 1314 Label get_continue; 1315 1316 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); 1317 test_method_data_pointer(get_continue); 1318 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); 1319 bind(get_continue); 1320 } 1321 1322 // Set the method data pointer for the current bcp. 1323 1324 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1325 assert(ProfileInterpreter, "must be profiling interpreter"); 1326 Label zero_continue; 1327 1328 // Test MDO to avoid the call if it is NULL. 1329 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); 1330 test_method_data_pointer(zero_continue); 1331 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp); 1332 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); 1333 add(ImethodDataPtr, O0, ImethodDataPtr); 1334 bind(zero_continue); 1335 } 1336 1337 // Test ImethodDataPtr. If it is null, continue at the specified label 1338 1339 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1340 assert(ProfileInterpreter, "must be profiling interpreter"); 1341 br_null_short(ImethodDataPtr, Assembler::pn, zero_continue); 1342 } 1343 1344 void InterpreterMacroAssembler::verify_method_data_pointer() { 1345 assert(ProfileInterpreter, "must be profiling interpreter"); 1346 #ifdef ASSERT 1347 Label verify_continue; 1348 test_method_data_pointer(verify_continue); 1349 1350 // If the mdp is valid, it will point to a DataLayout header which is 1351 // consistent with the bcp. The converse is highly probable also. 1352 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch); 1353 ld_ptr(Lmethod, Method::const_offset(), O5); 1354 add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch); 1355 add(G3_scratch, O5, G3_scratch); 1356 cmp(Lbcp, G3_scratch); 1357 brx(Assembler::equal, false, Assembler::pt, verify_continue); 1358 1359 Register temp_reg = O5; 1360 delayed()->mov(ImethodDataPtr, temp_reg); 1361 // %%% should use call_VM_leaf here? 1362 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr); 1363 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1); 1364 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS); 1365 stf(FloatRegisterImpl::D, Ftos_d, d_save); 1366 mov(temp_reg->after_save(), O2); 1367 save_thread(L7_thread_cache); 1368 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none); 1369 delayed()->nop(); 1370 restore_thread(L7_thread_cache); 1371 ldf(FloatRegisterImpl::D, d_save, Ftos_d); 1372 restore(); 1373 bind(verify_continue); 1374 #endif // ASSERT 1375 } 1376 1377 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1378 Register Rtmp, 1379 Label &profile_continue) { 1380 assert(ProfileInterpreter, "must be profiling interpreter"); 1381 // Control will flow to "profile_continue" if the counter is less than the 1382 // limit or if we call profile_method() 1383 1384 Label done; 1385 1386 // if no method data exists, and the counter is high enough, make one 1387 br_notnull_short(ImethodDataPtr, Assembler::pn, done); 1388 1389 // Test to see if we should create a method data oop 1390 AddressLiteral profile_limit((address) &InvocationCounter::InterpreterProfileLimit); 1391 sethi(profile_limit, Rtmp); 1392 ld(Rtmp, profile_limit.low10(), Rtmp); 1393 cmp(invocation_count, Rtmp); 1394 // Use long branches because call_VM() code and following code generated by 1395 // test_backedge_count_for_osr() is large in debug VM. 1396 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue); 1397 delayed()->nop(); 1398 1399 // Build it now. 1400 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1401 set_method_data_pointer_for_bcp(); 1402 ba(profile_continue); 1403 delayed()->nop(); 1404 bind(done); 1405 } 1406 1407 // Store a value at some constant offset from the method data pointer. 1408 1409 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1410 assert(ProfileInterpreter, "must be profiling interpreter"); 1411 st_ptr(value, ImethodDataPtr, constant); 1412 } 1413 1414 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter, 1415 Register bumped_count, 1416 bool decrement) { 1417 assert(ProfileInterpreter, "must be profiling interpreter"); 1418 1419 // Load the counter. 1420 ld_ptr(counter, bumped_count); 1421 1422 if (decrement) { 1423 // Decrement the register. Set condition codes. 1424 subcc(bumped_count, DataLayout::counter_increment, bumped_count); 1425 1426 // If the decrement causes the counter to overflow, stay negative 1427 Label L; 1428 brx(Assembler::negative, true, Assembler::pn, L); 1429 1430 // Store the decremented counter, if it is still negative. 1431 delayed()->st_ptr(bumped_count, counter); 1432 bind(L); 1433 } else { 1434 // Increment the register. Set carry flag. 1435 addcc(bumped_count, DataLayout::counter_increment, bumped_count); 1436 1437 // If the increment causes the counter to overflow, pull back by 1. 1438 assert(DataLayout::counter_increment == 1, "subc works"); 1439 subc(bumped_count, G0, bumped_count); 1440 1441 // Store the incremented counter. 1442 st_ptr(bumped_count, counter); 1443 } 1444 } 1445 1446 // Increment the value at some constant offset from the method data pointer. 1447 1448 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1449 Register bumped_count, 1450 bool decrement) { 1451 // Locate the counter at a fixed offset from the mdp: 1452 Address counter(ImethodDataPtr, constant); 1453 increment_mdp_data_at(counter, bumped_count, decrement); 1454 } 1455 1456 // Increment the value at some non-fixed (reg + constant) offset from 1457 // the method data pointer. 1458 1459 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1460 int constant, 1461 Register bumped_count, 1462 Register scratch2, 1463 bool decrement) { 1464 // Add the constant to reg to get the offset. 1465 add(ImethodDataPtr, reg, scratch2); 1466 Address counter(scratch2, constant); 1467 increment_mdp_data_at(counter, bumped_count, decrement); 1468 } 1469 1470 // Set a flag value at the current method data pointer position. 1471 // Updates a single byte of the header, to avoid races with other header bits. 1472 1473 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1474 Register scratch) { 1475 assert(ProfileInterpreter, "must be profiling interpreter"); 1476 // Load the data header 1477 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch); 1478 1479 // Set the flag 1480 or3(scratch, flag_constant, scratch); 1481 1482 // Store the modified header. 1483 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset())); 1484 } 1485 1486 // Test the location at some offset from the method data pointer. 1487 // If it is not equal to value, branch to the not_equal_continue Label. 1488 // Set condition codes to match the nullness of the loaded value. 1489 1490 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1491 Register value, 1492 Label& not_equal_continue, 1493 Register scratch) { 1494 assert(ProfileInterpreter, "must be profiling interpreter"); 1495 ld_ptr(ImethodDataPtr, offset, scratch); 1496 cmp(value, scratch); 1497 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue); 1498 delayed()->tst(scratch); 1499 } 1500 1501 // Update the method data pointer by the displacement located at some fixed 1502 // offset from the method data pointer. 1503 1504 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1505 Register scratch) { 1506 assert(ProfileInterpreter, "must be profiling interpreter"); 1507 ld_ptr(ImethodDataPtr, offset_of_disp, scratch); 1508 add(ImethodDataPtr, scratch, ImethodDataPtr); 1509 } 1510 1511 // Update the method data pointer by the displacement located at the 1512 // offset (reg + offset_of_disp). 1513 1514 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1515 int offset_of_disp, 1516 Register scratch) { 1517 assert(ProfileInterpreter, "must be profiling interpreter"); 1518 add(reg, offset_of_disp, scratch); 1519 ld_ptr(ImethodDataPtr, scratch, scratch); 1520 add(ImethodDataPtr, scratch, ImethodDataPtr); 1521 } 1522 1523 // Update the method data pointer by a simple constant displacement. 1524 1525 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1526 assert(ProfileInterpreter, "must be profiling interpreter"); 1527 add(ImethodDataPtr, constant, ImethodDataPtr); 1528 } 1529 1530 // Update the method data pointer for a _ret bytecode whose target 1531 // was not among our cached targets. 1532 1533 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1534 Register return_bci) { 1535 assert(ProfileInterpreter, "must be profiling interpreter"); 1536 push(state); 1537 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile 1538 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1539 ld_ptr(l_tmp, return_bci); 1540 pop(state); 1541 } 1542 1543 // Count a taken branch in the bytecodes. 1544 1545 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1546 if (ProfileInterpreter) { 1547 Label profile_continue; 1548 1549 // If no method data exists, go to profile_continue. 1550 test_method_data_pointer(profile_continue); 1551 1552 // We are taking a branch. Increment the taken count. 1553 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count); 1554 1555 // The method data pointer needs to be updated to reflect the new target. 1556 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1557 bind (profile_continue); 1558 } 1559 } 1560 1561 1562 // Count a not-taken branch in the bytecodes. 1563 1564 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) { 1565 if (ProfileInterpreter) { 1566 Label profile_continue; 1567 1568 // If no method data exists, go to profile_continue. 1569 test_method_data_pointer(profile_continue); 1570 1571 // We are taking a branch. Increment the not taken count. 1572 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch); 1573 1574 // The method data pointer needs to be updated to correspond to the 1575 // next bytecode. 1576 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1577 bind (profile_continue); 1578 } 1579 } 1580 1581 1582 // Count a non-virtual call in the bytecodes. 1583 1584 void InterpreterMacroAssembler::profile_call(Register scratch) { 1585 if (ProfileInterpreter) { 1586 Label profile_continue; 1587 1588 // If no method data exists, go to profile_continue. 1589 test_method_data_pointer(profile_continue); 1590 1591 // We are making a call. Increment the count. 1592 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1593 1594 // The method data pointer needs to be updated to reflect the new target. 1595 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1596 bind (profile_continue); 1597 } 1598 } 1599 1600 1601 // Count a final call in the bytecodes. 1602 1603 void InterpreterMacroAssembler::profile_final_call(Register scratch) { 1604 if (ProfileInterpreter) { 1605 Label profile_continue; 1606 1607 // If no method data exists, go to profile_continue. 1608 test_method_data_pointer(profile_continue); 1609 1610 // We are making a call. Increment the count. 1611 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1612 1613 // The method data pointer needs to be updated to reflect the new target. 1614 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1615 bind (profile_continue); 1616 } 1617 } 1618 1619 1620 // Count a virtual call in the bytecodes. 1621 1622 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1623 Register scratch, 1624 bool receiver_can_be_null) { 1625 if (ProfileInterpreter) { 1626 Label profile_continue; 1627 1628 // If no method data exists, go to profile_continue. 1629 test_method_data_pointer(profile_continue); 1630 1631 1632 Label skip_receiver_profile; 1633 if (receiver_can_be_null) { 1634 Label not_null; 1635 br_notnull_short(receiver, Assembler::pt, not_null); 1636 // We are making a call. Increment the count for null receiver. 1637 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1638 ba_short(skip_receiver_profile); 1639 bind(not_null); 1640 } 1641 1642 // Record the receiver type. 1643 record_klass_in_profile(receiver, scratch, true); 1644 bind(skip_receiver_profile); 1645 1646 // The method data pointer needs to be updated to reflect the new target. 1647 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1648 bind (profile_continue); 1649 } 1650 } 1651 1652 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1653 Register receiver, Register scratch, 1654 int start_row, Label& done, bool is_virtual_call) { 1655 if (TypeProfileWidth == 0) { 1656 if (is_virtual_call) { 1657 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1658 } 1659 return; 1660 } 1661 1662 int last_row = VirtualCallData::row_limit() - 1; 1663 assert(start_row <= last_row, "must be work left to do"); 1664 // Test this row for both the receiver and for null. 1665 // Take any of three different outcomes: 1666 // 1. found receiver => increment count and goto done 1667 // 2. found null => keep looking for case 1, maybe allocate this cell 1668 // 3. found something else => keep looking for cases 1 and 2 1669 // Case 3 is handled by a recursive call. 1670 for (int row = start_row; row <= last_row; row++) { 1671 Label next_test; 1672 bool test_for_null_also = (row == start_row); 1673 1674 // See if the receiver is receiver[n]. 1675 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1676 test_mdp_data_at(recvr_offset, receiver, next_test, scratch); 1677 // delayed()->tst(scratch); 1678 1679 // The receiver is receiver[n]. Increment count[n]. 1680 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1681 increment_mdp_data_at(count_offset, scratch); 1682 ba_short(done); 1683 bind(next_test); 1684 1685 if (test_for_null_also) { 1686 Label found_null; 1687 // Failed the equality check on receiver[n]... Test for null. 1688 if (start_row == last_row) { 1689 // The only thing left to do is handle the null case. 1690 if (is_virtual_call) { 1691 brx(Assembler::zero, false, Assembler::pn, found_null); 1692 delayed()->nop(); 1693 // Receiver did not match any saved receiver and there is no empty row for it. 1694 // Increment total counter to indicate polymorphic case. 1695 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1696 ba_short(done); 1697 bind(found_null); 1698 } else { 1699 brx(Assembler::notZero, false, Assembler::pt, done); 1700 delayed()->nop(); 1701 } 1702 break; 1703 } 1704 // Since null is rare, make it be the branch-taken case. 1705 brx(Assembler::zero, false, Assembler::pn, found_null); 1706 delayed()->nop(); 1707 1708 // Put all the "Case 3" tests here. 1709 record_klass_in_profile_helper(receiver, scratch, start_row + 1, done, is_virtual_call); 1710 1711 // Found a null. Keep searching for a matching receiver, 1712 // but remember that this is an empty (unused) slot. 1713 bind(found_null); 1714 } 1715 } 1716 1717 // In the fall-through case, we found no matching receiver, but we 1718 // observed the receiver[start_row] is NULL. 1719 1720 // Fill in the receiver field and increment the count. 1721 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1722 set_mdp_data_at(recvr_offset, receiver); 1723 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1724 mov(DataLayout::counter_increment, scratch); 1725 set_mdp_data_at(count_offset, scratch); 1726 if (start_row > 0) { 1727 ba_short(done); 1728 } 1729 } 1730 1731 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1732 Register scratch, bool is_virtual_call) { 1733 assert(ProfileInterpreter, "must be profiling"); 1734 Label done; 1735 1736 record_klass_in_profile_helper(receiver, scratch, 0, done, is_virtual_call); 1737 1738 bind (done); 1739 } 1740 1741 1742 // Count a ret in the bytecodes. 1743 1744 void InterpreterMacroAssembler::profile_ret(TosState state, 1745 Register return_bci, 1746 Register scratch) { 1747 if (ProfileInterpreter) { 1748 Label profile_continue; 1749 uint row; 1750 1751 // If no method data exists, go to profile_continue. 1752 test_method_data_pointer(profile_continue); 1753 1754 // Update the total ret count. 1755 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1756 1757 for (row = 0; row < RetData::row_limit(); row++) { 1758 Label next_test; 1759 1760 // See if return_bci is equal to bci[n]: 1761 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), 1762 return_bci, next_test, scratch); 1763 1764 // return_bci is equal to bci[n]. Increment the count. 1765 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch); 1766 1767 // The method data pointer needs to be updated to reflect the new target. 1768 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch); 1769 ba_short(profile_continue); 1770 bind(next_test); 1771 } 1772 1773 update_mdp_for_ret(state, return_bci); 1774 1775 bind (profile_continue); 1776 } 1777 } 1778 1779 // Profile an unexpected null in the bytecodes. 1780 void InterpreterMacroAssembler::profile_null_seen(Register scratch) { 1781 if (ProfileInterpreter) { 1782 Label profile_continue; 1783 1784 // If no method data exists, go to profile_continue. 1785 test_method_data_pointer(profile_continue); 1786 1787 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch); 1788 1789 // The method data pointer needs to be updated. 1790 int mdp_delta = in_bytes(BitData::bit_data_size()); 1791 if (TypeProfileCasts) { 1792 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1793 } 1794 update_mdp_by_constant(mdp_delta); 1795 1796 bind (profile_continue); 1797 } 1798 } 1799 1800 void InterpreterMacroAssembler::profile_typecheck(Register klass, 1801 Register scratch) { 1802 if (ProfileInterpreter) { 1803 Label profile_continue; 1804 1805 // If no method data exists, go to profile_continue. 1806 test_method_data_pointer(profile_continue); 1807 1808 int mdp_delta = in_bytes(BitData::bit_data_size()); 1809 if (TypeProfileCasts) { 1810 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1811 1812 // Record the object type. 1813 record_klass_in_profile(klass, scratch, false); 1814 } 1815 1816 // The method data pointer needs to be updated. 1817 update_mdp_by_constant(mdp_delta); 1818 1819 bind (profile_continue); 1820 } 1821 } 1822 1823 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) { 1824 if (ProfileInterpreter && TypeProfileCasts) { 1825 Label profile_continue; 1826 1827 // If no method data exists, go to profile_continue. 1828 test_method_data_pointer(profile_continue); 1829 1830 int count_offset = in_bytes(CounterData::count_offset()); 1831 // Back up the address, since we have already bumped the mdp. 1832 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1833 1834 // *Decrement* the counter. We expect to see zero or small negatives. 1835 increment_mdp_data_at(count_offset, scratch, true); 1836 1837 bind (profile_continue); 1838 } 1839 } 1840 1841 // Count the default case of a switch construct. 1842 1843 void InterpreterMacroAssembler::profile_switch_default(Register scratch) { 1844 if (ProfileInterpreter) { 1845 Label profile_continue; 1846 1847 // If no method data exists, go to profile_continue. 1848 test_method_data_pointer(profile_continue); 1849 1850 // Update the default case count 1851 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1852 scratch); 1853 1854 // The method data pointer needs to be updated. 1855 update_mdp_by_offset( 1856 in_bytes(MultiBranchData::default_displacement_offset()), 1857 scratch); 1858 1859 bind (profile_continue); 1860 } 1861 } 1862 1863 // Count the index'th case of a switch construct. 1864 1865 void InterpreterMacroAssembler::profile_switch_case(Register index, 1866 Register scratch, 1867 Register scratch2, 1868 Register scratch3) { 1869 if (ProfileInterpreter) { 1870 Label profile_continue; 1871 1872 // If no method data exists, go to profile_continue. 1873 test_method_data_pointer(profile_continue); 1874 1875 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() 1876 set(in_bytes(MultiBranchData::per_case_size()), scratch); 1877 smul(index, scratch, scratch); 1878 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch); 1879 1880 // Update the case count 1881 increment_mdp_data_at(scratch, 1882 in_bytes(MultiBranchData::relative_count_offset()), 1883 scratch2, 1884 scratch3); 1885 1886 // The method data pointer needs to be updated. 1887 update_mdp_by_offset(scratch, 1888 in_bytes(MultiBranchData::relative_displacement_offset()), 1889 scratch2); 1890 1891 bind (profile_continue); 1892 } 1893 } 1894 1895 // add a InterpMonitorElem to stack (see frame_sparc.hpp) 1896 1897 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty, 1898 Register Rtemp, 1899 Register Rtemp2 ) { 1900 1901 Register Rlimit = Lmonitors; 1902 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 1903 assert( (delta & LongAlignmentMask) == 0, 1904 "sizeof BasicObjectLock must be even number of doublewords"); 1905 1906 sub( SP, delta, SP); 1907 sub( Lesp, delta, Lesp); 1908 sub( Lmonitors, delta, Lmonitors); 1909 1910 if (!stack_is_empty) { 1911 1912 // must copy stack contents down 1913 1914 Label start_copying, next; 1915 1916 // untested("monitor stack expansion"); 1917 compute_stack_base(Rtemp); 1918 ba(start_copying); 1919 delayed()->cmp(Rtemp, Rlimit); // done? duplicated below 1920 1921 // note: must copy from low memory upwards 1922 // On entry to loop, 1923 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS) 1924 // Loop mutates Rtemp 1925 1926 bind( next); 1927 1928 st_ptr(Rtemp2, Rtemp, 0); 1929 inc(Rtemp, wordSize); 1930 cmp(Rtemp, Rlimit); // are we done? (duplicated above) 1931 1932 bind( start_copying ); 1933 1934 brx( notEqual, true, pn, next ); 1935 delayed()->ld_ptr( Rtemp, delta, Rtemp2 ); 1936 1937 // done copying stack 1938 } 1939 } 1940 1941 // Locals 1942 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) { 1943 assert_not_delayed(); 1944 sll(index, Interpreter::logStackElementSize, index); 1945 sub(Llocals, index, index); 1946 ld_ptr(index, 0, dst); 1947 // Note: index must hold the effective address--the iinc template uses it 1948 } 1949 1950 // Just like access_local_ptr but the tag is a returnAddress 1951 void InterpreterMacroAssembler::access_local_returnAddress(Register index, 1952 Register dst ) { 1953 assert_not_delayed(); 1954 sll(index, Interpreter::logStackElementSize, index); 1955 sub(Llocals, index, index); 1956 ld_ptr(index, 0, dst); 1957 } 1958 1959 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) { 1960 assert_not_delayed(); 1961 sll(index, Interpreter::logStackElementSize, index); 1962 sub(Llocals, index, index); 1963 ld(index, 0, dst); 1964 // Note: index must hold the effective address--the iinc template uses it 1965 } 1966 1967 1968 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) { 1969 assert_not_delayed(); 1970 sll(index, Interpreter::logStackElementSize, index); 1971 sub(Llocals, index, index); 1972 // First half stored at index n+1 (which grows down from Llocals[n]) 1973 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst); 1974 } 1975 1976 1977 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) { 1978 assert_not_delayed(); 1979 sll(index, Interpreter::logStackElementSize, index); 1980 sub(Llocals, index, index); 1981 ldf(FloatRegisterImpl::S, index, 0, dst); 1982 } 1983 1984 1985 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) { 1986 assert_not_delayed(); 1987 sll(index, Interpreter::logStackElementSize, index); 1988 sub(Llocals, index, index); 1989 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst); 1990 } 1991 1992 1993 #ifdef ASSERT 1994 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) { 1995 Label L; 1996 1997 assert(Rindex != Rscratch, "Registers cannot be same"); 1998 assert(Rindex != Rscratch1, "Registers cannot be same"); 1999 assert(Rlimit != Rscratch, "Registers cannot be same"); 2000 assert(Rlimit != Rscratch1, "Registers cannot be same"); 2001 assert(Rscratch1 != Rscratch, "Registers cannot be same"); 2002 2003 // untested("reg area corruption"); 2004 add(Rindex, offset, Rscratch); 2005 add(Rlimit, 64 + STACK_BIAS, Rscratch1); 2006 cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L); 2007 stop("regsave area is being clobbered"); 2008 bind(L); 2009 } 2010 #endif // ASSERT 2011 2012 2013 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) { 2014 assert_not_delayed(); 2015 sll(index, Interpreter::logStackElementSize, index); 2016 sub(Llocals, index, index); 2017 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);) 2018 st(src, index, 0); 2019 } 2020 2021 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) { 2022 assert_not_delayed(); 2023 sll(index, Interpreter::logStackElementSize, index); 2024 sub(Llocals, index, index); 2025 #ifdef ASSERT 2026 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2027 #endif 2028 st_ptr(src, index, 0); 2029 } 2030 2031 2032 2033 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) { 2034 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n)); 2035 } 2036 2037 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) { 2038 assert_not_delayed(); 2039 sll(index, Interpreter::logStackElementSize, index); 2040 sub(Llocals, index, index); 2041 #ifdef ASSERT 2042 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2043 #endif 2044 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1 2045 } 2046 2047 2048 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) { 2049 assert_not_delayed(); 2050 sll(index, Interpreter::logStackElementSize, index); 2051 sub(Llocals, index, index); 2052 #ifdef ASSERT 2053 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2054 #endif 2055 stf(FloatRegisterImpl::S, src, index, 0); 2056 } 2057 2058 2059 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) { 2060 assert_not_delayed(); 2061 sll(index, Interpreter::logStackElementSize, index); 2062 sub(Llocals, index, index); 2063 #ifdef ASSERT 2064 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2065 #endif 2066 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1)); 2067 } 2068 2069 2070 int InterpreterMacroAssembler::top_most_monitor_byte_offset() { 2071 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 2072 int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong); 2073 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS; 2074 } 2075 2076 2077 Address InterpreterMacroAssembler::top_most_monitor() { 2078 return Address(FP, top_most_monitor_byte_offset()); 2079 } 2080 2081 2082 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) { 2083 add( Lesp, wordSize, Rdest ); 2084 } 2085 2086 #endif /* CC_INTERP */ 2087 2088 void InterpreterMacroAssembler::increment_invocation_counter( Register Rtmp, Register Rtmp2 ) { 2089 assert(UseCompiler, "incrementing must be useful"); 2090 #ifdef CC_INTERP 2091 Address inv_counter(G5_method, Method::invocation_counter_offset() + 2092 InvocationCounter::counter_offset()); 2093 Address be_counter (G5_method, Method::backedge_counter_offset() + 2094 InvocationCounter::counter_offset()); 2095 #else 2096 Address inv_counter(Lmethod, Method::invocation_counter_offset() + 2097 InvocationCounter::counter_offset()); 2098 Address be_counter (Lmethod, Method::backedge_counter_offset() + 2099 InvocationCounter::counter_offset()); 2100 #endif /* CC_INTERP */ 2101 int delta = InvocationCounter::count_increment; 2102 2103 // Load each counter in a register 2104 ld( inv_counter, Rtmp ); 2105 ld( be_counter, Rtmp2 ); 2106 2107 assert( is_simm13( delta ), " delta too large."); 2108 2109 // Add the delta to the invocation counter and store the result 2110 add( Rtmp, delta, Rtmp ); 2111 2112 // Mask the backedge counter 2113 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2114 2115 // Store value 2116 st( Rtmp, inv_counter); 2117 2118 // Add invocation counter + backedge counter 2119 add( Rtmp, Rtmp2, Rtmp); 2120 2121 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 2122 } 2123 2124 void InterpreterMacroAssembler::increment_backedge_counter( Register Rtmp, Register Rtmp2 ) { 2125 assert(UseCompiler, "incrementing must be useful"); 2126 #ifdef CC_INTERP 2127 Address be_counter (G5_method, Method::backedge_counter_offset() + 2128 InvocationCounter::counter_offset()); 2129 Address inv_counter(G5_method, Method::invocation_counter_offset() + 2130 InvocationCounter::counter_offset()); 2131 #else 2132 Address be_counter (Lmethod, Method::backedge_counter_offset() + 2133 InvocationCounter::counter_offset()); 2134 Address inv_counter(Lmethod, Method::invocation_counter_offset() + 2135 InvocationCounter::counter_offset()); 2136 #endif /* CC_INTERP */ 2137 int delta = InvocationCounter::count_increment; 2138 // Load each counter in a register 2139 ld( be_counter, Rtmp ); 2140 ld( inv_counter, Rtmp2 ); 2141 2142 // Add the delta to the backedge counter 2143 add( Rtmp, delta, Rtmp ); 2144 2145 // Mask the invocation counter, add to backedge counter 2146 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2147 2148 // and store the result to memory 2149 st( Rtmp, be_counter ); 2150 2151 // Add backedge + invocation counter 2152 add( Rtmp, Rtmp2, Rtmp ); 2153 2154 // Note that this macro must leave backedge_count + invocation_count in Rtmp! 2155 } 2156 2157 #ifndef CC_INTERP 2158 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count, 2159 Register branch_bcp, 2160 Register Rtmp ) { 2161 Label did_not_overflow; 2162 Label overflow_with_error; 2163 assert_different_registers(backedge_count, Rtmp, branch_bcp); 2164 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 2165 2166 AddressLiteral limit(&InvocationCounter::InterpreterBackwardBranchLimit); 2167 load_contents(limit, Rtmp); 2168 cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow); 2169 2170 // When ProfileInterpreter is on, the backedge_count comes from the 2171 // MethodData*, which value does not get reset on the call to 2172 // frequency_counter_overflow(). To avoid excessive calls to the overflow 2173 // routine while the method is being compiled, add a second test to make sure 2174 // the overflow function is called only once every overflow_frequency. 2175 if (ProfileInterpreter) { 2176 const int overflow_frequency = 1024; 2177 andcc(backedge_count, overflow_frequency-1, Rtmp); 2178 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow); 2179 delayed()->nop(); 2180 } 2181 2182 // overflow in loop, pass branch bytecode 2183 set(6,Rtmp); 2184 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp); 2185 2186 // Was an OSR adapter generated? 2187 // O0 = osr nmethod 2188 br_null_short(O0, Assembler::pn, overflow_with_error); 2189 2190 // Has the nmethod been invalidated already? 2191 ld(O0, nmethod::entry_bci_offset(), O2); 2192 cmp_and_br_short(O2, InvalidOSREntryBci, Assembler::equal, Assembler::pn, overflow_with_error); 2193 2194 // migrate the interpreter frame off of the stack 2195 2196 mov(G2_thread, L7); 2197 // save nmethod 2198 mov(O0, L6); 2199 set_last_Java_frame(SP, noreg); 2200 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7); 2201 reset_last_Java_frame(); 2202 mov(L7, G2_thread); 2203 2204 // move OSR nmethod to I1 2205 mov(L6, I1); 2206 2207 // OSR buffer to I0 2208 mov(O0, I0); 2209 2210 // remove the interpreter frame 2211 restore(I5_savedSP, 0, SP); 2212 2213 // Jump to the osr code. 2214 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2); 2215 jmp(O2, G0); 2216 delayed()->nop(); 2217 2218 bind(overflow_with_error); 2219 2220 bind(did_not_overflow); 2221 } 2222 2223 2224 2225 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) { 2226 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); } 2227 } 2228 2229 2230 // local helper function for the verify_oop_or_return_address macro 2231 static bool verify_return_address(Method* m, int bci) { 2232 #ifndef PRODUCT 2233 address pc = (address)(m->constMethod()) 2234 + in_bytes(ConstMethod::codes_offset()) + bci; 2235 // assume it is a valid return address if it is inside m and is preceded by a jsr 2236 if (!m->contains(pc)) return false; 2237 address jsr_pc; 2238 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2239 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2240 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2241 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2242 #endif // PRODUCT 2243 return false; 2244 } 2245 2246 2247 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2248 if (!VerifyOops) return; 2249 // the VM documentation for the astore[_wide] bytecode allows 2250 // the TOS to be not only an oop but also a return address 2251 Label test; 2252 Label skip; 2253 // See if it is an address (in the current method): 2254 2255 mov(reg, Rtmp); 2256 const int log2_bytecode_size_limit = 16; 2257 srl(Rtmp, log2_bytecode_size_limit, Rtmp); 2258 br_notnull_short( Rtmp, pt, test ); 2259 2260 // %%% should use call_VM_leaf here? 2261 save_frame_and_mov(0, Lmethod, O0, reg, O1); 2262 save_thread(L7_thread_cache); 2263 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none); 2264 delayed()->nop(); 2265 restore_thread(L7_thread_cache); 2266 br_notnull( O0, false, pt, skip ); 2267 delayed()->restore(); 2268 2269 // Perform a more elaborate out-of-line call 2270 // Not an address; verify it: 2271 bind(test); 2272 verify_oop(reg); 2273 bind(skip); 2274 } 2275 2276 2277 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2278 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); 2279 } 2280 #endif /* CC_INTERP */ 2281 2282 // Inline assembly for: 2283 // 2284 // if (thread is in interp_only_mode) { 2285 // InterpreterRuntime::post_method_entry(); 2286 // } 2287 // if (DTraceMethodProbes) { 2288 // SharedRuntime::dtrace_method_entry(method, receiver); 2289 // } 2290 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 2291 // SharedRuntime::rc_trace_method_entry(method, receiver); 2292 // } 2293 2294 void InterpreterMacroAssembler::notify_method_entry() { 2295 2296 // C++ interpreter only uses this for native methods. 2297 2298 // Whenever JVMTI puts a thread in interp_only_mode, method 2299 // entry/exit events are sent for that thread to track stack 2300 // depth. If it is possible to enter interp_only_mode we add 2301 // the code to check if the event should be sent. 2302 if (JvmtiExport::can_post_interpreter_events()) { 2303 Label L; 2304 Register temp_reg = O5; 2305 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2306 ld(interp_only, temp_reg); 2307 cmp_and_br_short(temp_reg, 0, equal, pt, L); 2308 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); 2309 bind(L); 2310 } 2311 2312 { 2313 Register temp_reg = O5; 2314 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2315 call_VM_leaf(noreg, 2316 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 2317 G2_thread, Lmethod); 2318 } 2319 2320 // RedefineClasses() tracing support for obsolete method entry 2321 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 2322 call_VM_leaf(noreg, 2323 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 2324 G2_thread, Lmethod); 2325 } 2326 } 2327 2328 2329 // Inline assembly for: 2330 // 2331 // if (thread is in interp_only_mode) { 2332 // // save result 2333 // InterpreterRuntime::post_method_exit(); 2334 // // restore result 2335 // } 2336 // if (DTraceMethodProbes) { 2337 // SharedRuntime::dtrace_method_exit(thread, method); 2338 // } 2339 // 2340 // Native methods have their result stored in d_tmp and l_tmp 2341 // Java methods have their result stored in the expression stack 2342 2343 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, 2344 TosState state, 2345 NotifyMethodExitMode mode) { 2346 // C++ interpreter only uses this for native methods. 2347 2348 // Whenever JVMTI puts a thread in interp_only_mode, method 2349 // entry/exit events are sent for that thread to track stack 2350 // depth. If it is possible to enter interp_only_mode we add 2351 // the code to check if the event should be sent. 2352 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2353 Label L; 2354 Register temp_reg = O5; 2355 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2356 ld(interp_only, temp_reg); 2357 cmp_and_br_short(temp_reg, 0, equal, pt, L); 2358 2359 // Note: frame::interpreter_frame_result has a dependency on how the 2360 // method result is saved across the call to post_method_exit. For 2361 // native methods it assumes the result registers are saved to 2362 // l_scratch and d_scratch. If this changes then the interpreter_frame_result 2363 // implementation will need to be updated too. 2364 2365 save_return_value(state, is_native_method); 2366 call_VM(noreg, 2367 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2368 restore_return_value(state, is_native_method); 2369 bind(L); 2370 } 2371 2372 { 2373 Register temp_reg = O5; 2374 // Dtrace notification 2375 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2376 save_return_value(state, is_native_method); 2377 call_VM_leaf( 2378 noreg, 2379 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2380 G2_thread, Lmethod); 2381 restore_return_value(state, is_native_method); 2382 } 2383 } 2384 2385 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) { 2386 #ifdef CC_INTERP 2387 // result potentially in O0/O1: save it across calls 2388 stf(FloatRegisterImpl::D, F0, STATE(_native_fresult)); 2389 #ifdef _LP64 2390 stx(O0, STATE(_native_lresult)); 2391 #else 2392 std(O0, STATE(_native_lresult)); 2393 #endif 2394 #else // CC_INTERP 2395 if (is_native_call) { 2396 stf(FloatRegisterImpl::D, F0, d_tmp); 2397 #ifdef _LP64 2398 stx(O0, l_tmp); 2399 #else 2400 std(O0, l_tmp); 2401 #endif 2402 } else { 2403 push(state); 2404 } 2405 #endif // CC_INTERP 2406 } 2407 2408 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) { 2409 #ifdef CC_INTERP 2410 ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0); 2411 #ifdef _LP64 2412 ldx(STATE(_native_lresult), O0); 2413 #else 2414 ldd(STATE(_native_lresult), O0); 2415 #endif 2416 #else // CC_INTERP 2417 if (is_native_call) { 2418 ldf(FloatRegisterImpl::D, d_tmp, F0); 2419 #ifdef _LP64 2420 ldx(l_tmp, O0); 2421 #else 2422 ldd(l_tmp, O0); 2423 #endif 2424 } else { 2425 pop(state); 2426 } 2427 #endif // CC_INTERP 2428 } 2429 2430 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 2431 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 2432 int increment, int mask, 2433 Register scratch1, Register scratch2, 2434 Condition cond, Label *where) { 2435 ld(counter_addr, scratch1); 2436 add(scratch1, increment, scratch1); 2437 if (is_simm13(mask)) { 2438 andcc(scratch1, mask, G0); 2439 } else { 2440 set(mask, scratch2); 2441 andcc(scratch1, scratch2, G0); 2442 } 2443 br(cond, false, Assembler::pn, *where); 2444 delayed()->st(scratch1, counter_addr); 2445 }