1 /* 2 * Copyright 1997-2010 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, 20 * CA 95054 USA or visit www.sun.com if you need additional information or 21 * have any questions. 22 * 23 */ 24 25 #include "incls/_precompiled.incl" 26 #include "incls/_interp_masm_sparc.cpp.incl" 27 28 #ifndef CC_INTERP 29 #ifndef FAST_DISPATCH 30 #define FAST_DISPATCH 1 31 #endif 32 #undef FAST_DISPATCH 33 34 // Implementation of InterpreterMacroAssembler 35 36 // This file specializes the assember with interpreter-specific macros 37 38 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS); 39 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS); 40 41 #else // CC_INTERP 42 #ifndef STATE 43 #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name)) 44 #endif // STATE 45 46 #endif // CC_INTERP 47 48 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) { 49 // Note: this algorithm is also used by C1's OSR entry sequence. 50 // Any changes should also be applied to CodeEmitter::emit_osr_entry(). 51 assert_different_registers(args_size, locals_size); 52 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted. 53 if (TaggedStackInterpreter) sll(locals_size, 1, locals_size); 54 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words 55 // Use br/mov combination because it works on both V8 and V9 and is 56 // faster. 57 Label skip_move; 58 br(Assembler::negative, true, Assembler::pt, skip_move); 59 delayed()->mov(G0, delta); 60 bind(skip_move); 61 round_to(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned) 62 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes 63 } 64 65 #ifndef CC_INTERP 66 67 // Dispatch code executed in the prolog of a bytecode which does not do it's 68 // own dispatch. The dispatch address is computed and placed in IdispatchAddress 69 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 70 assert_not_delayed(); 71 #ifdef FAST_DISPATCH 72 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since 73 // they both use I2. 74 assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive"); 75 ldub(Lbcp, bcp_incr, Lbyte_code); // load next bytecode 76 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); 77 // add offset to correct dispatch table 78 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 79 ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr 80 #else 81 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 82 // dispatch table to use 83 AddressLiteral tbl(Interpreter::dispatch_table(state)); 84 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 85 set(tbl, G3_scratch); // compute addr of table 86 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr 87 #endif 88 } 89 90 91 // Dispatch code executed in the epilog of a bytecode which does not do it's 92 // own dispatch. The dispatch address in IdispatchAddress is used for the 93 // dispatch. 94 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 95 assert_not_delayed(); 96 verify_FPU(1, state); 97 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 98 jmp( IdispatchAddress, 0 ); 99 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); 100 else delayed()->nop(); 101 } 102 103 104 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { 105 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) 106 assert_not_delayed(); 107 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 108 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr); 109 } 110 111 112 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) { 113 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) 114 assert_not_delayed(); 115 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 116 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false); 117 } 118 119 120 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 121 // load current bytecode 122 assert_not_delayed(); 123 ldub( Lbcp, 0, Lbyte_code); // load next bytecode 124 dispatch_base(state, table); 125 } 126 127 128 void InterpreterMacroAssembler::call_VM_leaf_base( 129 Register java_thread, 130 address entry_point, 131 int number_of_arguments 132 ) { 133 if (!java_thread->is_valid()) 134 java_thread = L7_thread_cache; 135 // super call 136 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments); 137 } 138 139 140 void InterpreterMacroAssembler::call_VM_base( 141 Register oop_result, 142 Register java_thread, 143 Register last_java_sp, 144 address entry_point, 145 int number_of_arguments, 146 bool check_exception 147 ) { 148 if (!java_thread->is_valid()) 149 java_thread = L7_thread_cache; 150 // See class ThreadInVMfromInterpreter, which assumes that the interpreter 151 // takes responsibility for setting its own thread-state on call-out. 152 // However, ThreadInVMfromInterpreter resets the state to "in_Java". 153 154 //save_bcp(); // save bcp 155 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception); 156 //restore_bcp(); // restore bcp 157 //restore_locals(); // restore locals pointer 158 } 159 160 161 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 162 if (JvmtiExport::can_pop_frame()) { 163 Label L; 164 165 // Check the "pending popframe condition" flag in the current thread 166 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg); 167 168 // Initiate popframe handling only if it is not already being processed. If the flag 169 // has the popframe_processing bit set, it means that this code is called *during* popframe 170 // handling - we don't want to reenter. 171 btst(JavaThread::popframe_pending_bit, scratch_reg); 172 br(zero, false, pt, L); 173 delayed()->nop(); 174 btst(JavaThread::popframe_processing_bit, scratch_reg); 175 br(notZero, false, pt, L); 176 delayed()->nop(); 177 178 // Call Interpreter::remove_activation_preserving_args_entry() to get the 179 // address of the same-named entrypoint in the generated interpreter code. 180 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 181 182 // Jump to Interpreter::_remove_activation_preserving_args_entry 183 jmpl(O0, G0, G0); 184 delayed()->nop(); 185 bind(L); 186 } 187 } 188 189 190 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 191 Register thr_state = G4_scratch; 192 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); 193 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset()); 194 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset()); 195 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset()); 196 switch (state) { 197 case ltos: ld_long(val_addr, Otos_l); break; 198 case atos: ld_ptr(oop_addr, Otos_l); 199 st_ptr(G0, oop_addr); break; 200 case btos: // fall through 201 case ctos: // fall through 202 case stos: // fall through 203 case itos: ld(val_addr, Otos_l1); break; 204 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break; 205 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break; 206 case vtos: /* nothing to do */ break; 207 default : ShouldNotReachHere(); 208 } 209 // Clean up tos value in the jvmti thread state 210 or3(G0, ilgl, G3_scratch); 211 stw(G3_scratch, tos_addr); 212 st_long(G0, val_addr); 213 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 214 } 215 216 217 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 218 if (JvmtiExport::can_force_early_return()) { 219 Label L; 220 Register thr_state = G3_scratch; 221 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); 222 tst(thr_state); 223 br(zero, false, pt, L); // if (thread->jvmti_thread_state() == NULL) exit; 224 delayed()->nop(); 225 226 // Initiate earlyret handling only if it is not already being processed. 227 // If the flag has the earlyret_processing bit set, it means that this code 228 // is called *during* earlyret handling - we don't want to reenter. 229 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch); 230 cmp(G4_scratch, JvmtiThreadState::earlyret_pending); 231 br(Assembler::notEqual, false, pt, L); 232 delayed()->nop(); 233 234 // Call Interpreter::remove_activation_early_entry() to get the address of the 235 // same-named entrypoint in the generated interpreter code 236 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1); 237 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1); 238 239 // Jump to Interpreter::_remove_activation_early_entry 240 jmpl(O0, G0, G0); 241 delayed()->nop(); 242 bind(L); 243 } 244 } 245 246 247 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) { 248 mov(arg_1, O0); 249 mov(arg_2, O1); 250 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2); 251 } 252 #endif /* CC_INTERP */ 253 254 255 #ifndef CC_INTERP 256 257 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) { 258 assert_not_delayed(); 259 dispatch_Lbyte_code(state, table); 260 } 261 262 263 void InterpreterMacroAssembler::dispatch_normal(TosState state) { 264 dispatch_base(state, Interpreter::normal_table(state)); 265 } 266 267 268 void InterpreterMacroAssembler::dispatch_only(TosState state) { 269 dispatch_base(state, Interpreter::dispatch_table(state)); 270 } 271 272 273 // common code to dispatch and dispatch_only 274 // dispatch value in Lbyte_code and increment Lbcp 275 276 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) { 277 verify_FPU(1, state); 278 // %%%%% maybe implement +VerifyActivationFrameSize here 279 //verify_thread(); //too slow; we will just verify on method entry & exit 280 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 281 #ifdef FAST_DISPATCH 282 if (table == Interpreter::dispatch_table(state)) { 283 // use IdispatchTables 284 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); 285 // add offset to correct dispatch table 286 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 287 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // get entry addr 288 } else { 289 #endif 290 // dispatch table to use 291 AddressLiteral tbl(table); 292 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 293 set(tbl, G3_scratch); // compute addr of table 294 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr 295 #ifdef FAST_DISPATCH 296 } 297 #endif 298 jmp( G3_scratch, 0 ); 299 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); 300 else delayed()->nop(); 301 } 302 303 304 // Helpers for expression stack 305 306 // Longs and doubles are Category 2 computational types in the 307 // JVM specification (section 3.11.1) and take 2 expression stack or 308 // local slots. 309 // Aligning them on 32 bit with tagged stacks is hard because the code generated 310 // for the dup* bytecodes depends on what types are already on the stack. 311 // If the types are split into the two stack/local slots, that is much easier 312 // (and we can use 0 for non-reference tags). 313 314 // Known good alignment in _LP64 but unknown otherwise 315 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) { 316 assert_not_delayed(); 317 318 #ifdef _LP64 319 ldf(FloatRegisterImpl::D, r1, offset, d); 320 #else 321 ldf(FloatRegisterImpl::S, r1, offset, d); 322 ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize(), d->successor()); 323 #endif 324 } 325 326 // Known good alignment in _LP64 but unknown otherwise 327 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) { 328 assert_not_delayed(); 329 330 #ifdef _LP64 331 stf(FloatRegisterImpl::D, d, r1, offset); 332 // store something more useful here 333 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize());) 334 #else 335 stf(FloatRegisterImpl::S, d, r1, offset); 336 stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize()); 337 #endif 338 } 339 340 341 // Known good alignment in _LP64 but unknown otherwise 342 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) { 343 assert_not_delayed(); 344 #ifdef _LP64 345 ldx(r1, offset, rd); 346 #else 347 ld(r1, offset, rd); 348 ld(r1, offset + Interpreter::stackElementSize(), rd->successor()); 349 #endif 350 } 351 352 // Known good alignment in _LP64 but unknown otherwise 353 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) { 354 assert_not_delayed(); 355 356 #ifdef _LP64 357 stx(l, r1, offset); 358 // store something more useful here 359 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize());) 360 #else 361 st(l, r1, offset); 362 st(l->successor(), r1, offset + Interpreter::stackElementSize()); 363 #endif 364 } 365 366 #ifdef ASSERT 367 void InterpreterMacroAssembler::verify_stack_tag(frame::Tag t, 368 Register r, 369 Register scratch) { 370 if (TaggedStackInterpreter) { 371 Label ok, long_ok; 372 ld_ptr(Lesp, Interpreter::expr_tag_offset_in_bytes(0), r); 373 if (t == frame::TagCategory2) { 374 cmp(r, G0); 375 brx(Assembler::equal, false, Assembler::pt, long_ok); 376 delayed()->ld_ptr(Lesp, Interpreter::expr_tag_offset_in_bytes(1), r); 377 stop("stack long/double tag value bad"); 378 bind(long_ok); 379 cmp(r, G0); 380 } else if (t == frame::TagValue) { 381 cmp(r, G0); 382 } else { 383 assert_different_registers(r, scratch); 384 mov(t, scratch); 385 cmp(r, scratch); 386 } 387 brx(Assembler::equal, false, Assembler::pt, ok); 388 delayed()->nop(); 389 // Also compare if the stack value is zero, then the tag might 390 // not have been set coming from deopt. 391 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); 392 cmp(r, G0); 393 brx(Assembler::equal, false, Assembler::pt, ok); 394 delayed()->nop(); 395 stop("Stack tag value is bad"); 396 bind(ok); 397 } 398 } 399 #endif // ASSERT 400 401 void InterpreterMacroAssembler::pop_i(Register r) { 402 assert_not_delayed(); 403 // Uses destination register r for scratch 404 debug_only(verify_stack_tag(frame::TagValue, r)); 405 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r); 406 inc(Lesp, Interpreter::stackElementSize()); 407 debug_only(verify_esp(Lesp)); 408 } 409 410 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) { 411 assert_not_delayed(); 412 // Uses destination register r for scratch 413 debug_only(verify_stack_tag(frame::TagReference, r, scratch)); 414 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); 415 inc(Lesp, Interpreter::stackElementSize()); 416 debug_only(verify_esp(Lesp)); 417 } 418 419 void InterpreterMacroAssembler::pop_l(Register r) { 420 assert_not_delayed(); 421 // Uses destination register r for scratch 422 debug_only(verify_stack_tag(frame::TagCategory2, r)); 423 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r); 424 inc(Lesp, 2*Interpreter::stackElementSize()); 425 debug_only(verify_esp(Lesp)); 426 } 427 428 429 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) { 430 assert_not_delayed(); 431 debug_only(verify_stack_tag(frame::TagValue, scratch)); 432 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f); 433 inc(Lesp, Interpreter::stackElementSize()); 434 debug_only(verify_esp(Lesp)); 435 } 436 437 438 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) { 439 assert_not_delayed(); 440 debug_only(verify_stack_tag(frame::TagCategory2, scratch)); 441 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f); 442 inc(Lesp, 2*Interpreter::stackElementSize()); 443 debug_only(verify_esp(Lesp)); 444 } 445 446 447 // (Note use register first, then decrement so dec can be done during store stall) 448 void InterpreterMacroAssembler::tag_stack(Register r) { 449 if (TaggedStackInterpreter) { 450 st_ptr(r, Lesp, Interpreter::tag_offset_in_bytes()); 451 } 452 } 453 454 void InterpreterMacroAssembler::tag_stack(frame::Tag t, Register r) { 455 if (TaggedStackInterpreter) { 456 assert (frame::TagValue == 0, "TagValue must be zero"); 457 if (t == frame::TagValue) { 458 st_ptr(G0, Lesp, Interpreter::tag_offset_in_bytes()); 459 } else if (t == frame::TagCategory2) { 460 st_ptr(G0, Lesp, Interpreter::tag_offset_in_bytes()); 461 // Tag next slot down too 462 st_ptr(G0, Lesp, -Interpreter::stackElementSize() + Interpreter::tag_offset_in_bytes()); 463 } else { 464 assert_different_registers(r, O3); 465 mov(t, O3); 466 st_ptr(O3, Lesp, Interpreter::tag_offset_in_bytes()); 467 } 468 } 469 } 470 471 void InterpreterMacroAssembler::push_i(Register r) { 472 assert_not_delayed(); 473 debug_only(verify_esp(Lesp)); 474 tag_stack(frame::TagValue, r); 475 st( r, Lesp, Interpreter::value_offset_in_bytes()); 476 dec( Lesp, Interpreter::stackElementSize()); 477 } 478 479 void InterpreterMacroAssembler::push_ptr(Register r) { 480 assert_not_delayed(); 481 tag_stack(frame::TagReference, r); 482 st_ptr( r, Lesp, Interpreter::value_offset_in_bytes()); 483 dec( Lesp, Interpreter::stackElementSize()); 484 } 485 486 void InterpreterMacroAssembler::push_ptr(Register r, Register tag) { 487 assert_not_delayed(); 488 tag_stack(tag); 489 st_ptr(r, Lesp, Interpreter::value_offset_in_bytes()); 490 dec( Lesp, Interpreter::stackElementSize()); 491 } 492 493 // remember: our convention for longs in SPARC is: 494 // O0 (Otos_l1) has high-order part in first word, 495 // O1 (Otos_l2) has low-order part in second word 496 497 void InterpreterMacroAssembler::push_l(Register r) { 498 assert_not_delayed(); 499 debug_only(verify_esp(Lesp)); 500 tag_stack(frame::TagCategory2, r); 501 // Longs are in stored in memory-correct order, even if unaligned. 502 // and may be separated by stack tags. 503 int offset = -Interpreter::stackElementSize() + Interpreter::value_offset_in_bytes(); 504 store_unaligned_long(r, Lesp, offset); 505 dec(Lesp, 2 * Interpreter::stackElementSize()); 506 } 507 508 509 void InterpreterMacroAssembler::push_f(FloatRegister f) { 510 assert_not_delayed(); 511 debug_only(verify_esp(Lesp)); 512 tag_stack(frame::TagValue, Otos_i); 513 stf(FloatRegisterImpl::S, f, Lesp, Interpreter::value_offset_in_bytes()); 514 dec(Lesp, Interpreter::stackElementSize()); 515 } 516 517 518 void InterpreterMacroAssembler::push_d(FloatRegister d) { 519 assert_not_delayed(); 520 debug_only(verify_esp(Lesp)); 521 tag_stack(frame::TagCategory2, Otos_i); 522 // Longs are in stored in memory-correct order, even if unaligned. 523 // and may be separated by stack tags. 524 int offset = -Interpreter::stackElementSize() + Interpreter::value_offset_in_bytes(); 525 store_unaligned_double(d, Lesp, offset); 526 dec(Lesp, 2 * Interpreter::stackElementSize()); 527 } 528 529 530 void InterpreterMacroAssembler::push(TosState state) { 531 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 532 switch (state) { 533 case atos: push_ptr(); break; 534 case btos: push_i(); break; 535 case ctos: 536 case stos: push_i(); break; 537 case itos: push_i(); break; 538 case ltos: push_l(); break; 539 case ftos: push_f(); break; 540 case dtos: push_d(); break; 541 case vtos: /* nothing to do */ break; 542 default : ShouldNotReachHere(); 543 } 544 } 545 546 547 void InterpreterMacroAssembler::pop(TosState state) { 548 switch (state) { 549 case atos: pop_ptr(); break; 550 case btos: pop_i(); break; 551 case ctos: 552 case stos: pop_i(); break; 553 case itos: pop_i(); break; 554 case ltos: pop_l(); break; 555 case ftos: pop_f(); break; 556 case dtos: pop_d(); break; 557 case vtos: /* nothing to do */ break; 558 default : ShouldNotReachHere(); 559 } 560 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 561 } 562 563 564 // Tagged stack helpers for swap and dup 565 void InterpreterMacroAssembler::load_ptr_and_tag(int n, Register val, 566 Register tag) { 567 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val); 568 if (TaggedStackInterpreter) { 569 ld_ptr(Lesp, Interpreter::expr_tag_offset_in_bytes(n), tag); 570 } 571 } 572 void InterpreterMacroAssembler::store_ptr_and_tag(int n, Register val, 573 Register tag) { 574 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n)); 575 if (TaggedStackInterpreter) { 576 st_ptr(tag, Lesp, Interpreter::expr_tag_offset_in_bytes(n)); 577 } 578 } 579 580 581 void InterpreterMacroAssembler::load_receiver(Register param_count, 582 Register recv) { 583 584 sll(param_count, Interpreter::logStackElementSize(), param_count); 585 if (TaggedStackInterpreter) { 586 add(param_count, Interpreter::value_offset_in_bytes(), param_count); // get obj address 587 } 588 ld_ptr(Lesp, param_count, recv); // gets receiver Oop 589 } 590 591 void InterpreterMacroAssembler::empty_expression_stack() { 592 // Reset Lesp. 593 sub( Lmonitors, wordSize, Lesp ); 594 595 // Reset SP by subtracting more space from Lesp. 596 Label done; 597 verify_oop(Lmethod); 598 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!"); 599 600 // A native does not need to do this, since its callee does not change SP. 601 ld(Lmethod, methodOopDesc::access_flags_offset(), Gframe_size); // Load access flags. 602 btst(JVM_ACC_NATIVE, Gframe_size); 603 br(Assembler::notZero, false, Assembler::pt, done); 604 delayed()->nop(); 605 606 // Compute max expression stack+register save area 607 lduh(Lmethod, in_bytes(methodOopDesc::max_stack_offset()), Gframe_size); // Load max stack. 608 if (TaggedStackInterpreter) sll ( Gframe_size, 1, Gframe_size); // max_stack * 2 for TAGS 609 add( Gframe_size, frame::memory_parameter_word_sp_offset, Gframe_size ); 610 611 // 612 // now set up a stack frame with the size computed above 613 // 614 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below 615 sll( Gframe_size, LogBytesPerWord, Gframe_size ); 616 sub( Lesp, Gframe_size, Gframe_size ); 617 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary 618 debug_only(verify_sp(Gframe_size, G4_scratch)); 619 #ifdef _LP64 620 sub(Gframe_size, STACK_BIAS, Gframe_size ); 621 #endif 622 mov(Gframe_size, SP); 623 624 bind(done); 625 } 626 627 628 #ifdef ASSERT 629 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) { 630 Label Bad, OK; 631 632 // Saved SP must be aligned. 633 #ifdef _LP64 634 btst(2*BytesPerWord-1, Rsp); 635 #else 636 btst(LongAlignmentMask, Rsp); 637 #endif 638 br(Assembler::notZero, false, Assembler::pn, Bad); 639 delayed()->nop(); 640 641 // Saved SP, plus register window size, must not be above FP. 642 add(Rsp, frame::register_save_words * wordSize, Rtemp); 643 #ifdef _LP64 644 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP 645 #endif 646 cmp(Rtemp, FP); 647 brx(Assembler::greaterUnsigned, false, Assembler::pn, Bad); 648 delayed()->nop(); 649 650 // Saved SP must not be ridiculously below current SP. 651 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K); 652 set(maxstack, Rtemp); 653 sub(SP, Rtemp, Rtemp); 654 #ifdef _LP64 655 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp 656 #endif 657 cmp(Rsp, Rtemp); 658 brx(Assembler::lessUnsigned, false, Assembler::pn, Bad); 659 delayed()->nop(); 660 661 br(Assembler::always, false, Assembler::pn, OK); 662 delayed()->nop(); 663 664 bind(Bad); 665 stop("on return to interpreted call, restored SP is corrupted"); 666 667 bind(OK); 668 } 669 670 671 void InterpreterMacroAssembler::verify_esp(Register Resp) { 672 // about to read or write Resp[0] 673 // make sure it is not in the monitors or the register save area 674 Label OK1, OK2; 675 676 cmp(Resp, Lmonitors); 677 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1); 678 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp); 679 stop("too many pops: Lesp points into monitor area"); 680 bind(OK1); 681 #ifdef _LP64 682 sub(Resp, STACK_BIAS, Resp); 683 #endif 684 cmp(Resp, SP); 685 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2); 686 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp); 687 stop("too many pushes: Lesp points into register window"); 688 bind(OK2); 689 } 690 #endif // ASSERT 691 692 // Load compiled (i2c) or interpreter entry when calling from interpreted and 693 // do the call. Centralized so that all interpreter calls will do the same actions. 694 // If jvmti single stepping is on for a thread we must not call compiled code. 695 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) { 696 697 // Assume we want to go compiled if available 698 699 ld_ptr(G5_method, in_bytes(methodOopDesc::from_interpreted_offset()), target); 700 701 if (JvmtiExport::can_post_interpreter_events()) { 702 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 703 // compiled code in threads for which the event is enabled. Check here for 704 // interp_only_mode if these events CAN be enabled. 705 verify_thread(); 706 Label skip_compiled_code; 707 708 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 709 ld(interp_only, scratch); 710 tst(scratch); 711 br(Assembler::notZero, true, Assembler::pn, skip_compiled_code); 712 delayed()->ld_ptr(G5_method, in_bytes(methodOopDesc::interpreter_entry_offset()), target); 713 bind(skip_compiled_code); 714 } 715 716 // the i2c_adapters need methodOop in G5_method (right? %%%) 717 // do the call 718 #ifdef ASSERT 719 { 720 Label ok; 721 br_notnull(target, false, Assembler::pt, ok); 722 delayed()->nop(); 723 stop("null entry point"); 724 bind(ok); 725 } 726 #endif // ASSERT 727 728 // Adjust Rret first so Llast_SP can be same as Rret 729 add(Rret, -frame::pc_return_offset, O7); 730 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer 731 // Record SP so we can remove any stack space allocated by adapter transition 732 jmp(target, 0); 733 delayed()->mov(SP, Llast_SP); 734 } 735 736 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) { 737 assert_not_delayed(); 738 739 Label not_taken; 740 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken); 741 else br (cc, false, Assembler::pn, not_taken); 742 delayed()->nop(); 743 744 TemplateTable::branch(false,false); 745 746 bind(not_taken); 747 748 profile_not_taken_branch(G3_scratch); 749 } 750 751 752 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp( 753 int bcp_offset, 754 Register Rtmp, 755 Register Rdst, 756 signedOrNot is_signed, 757 setCCOrNot should_set_CC ) { 758 assert(Rtmp != Rdst, "need separate temp register"); 759 assert_not_delayed(); 760 switch (is_signed) { 761 default: ShouldNotReachHere(); 762 763 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte 764 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte 765 } 766 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte 767 sll( Rdst, BitsPerByte, Rdst); 768 switch (should_set_CC ) { 769 default: ShouldNotReachHere(); 770 771 case set_CC: orcc( Rdst, Rtmp, Rdst ); break; 772 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break; 773 } 774 } 775 776 777 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp( 778 int bcp_offset, 779 Register Rtmp, 780 Register Rdst, 781 setCCOrNot should_set_CC ) { 782 assert(Rtmp != Rdst, "need separate temp register"); 783 assert_not_delayed(); 784 add( Lbcp, bcp_offset, Rtmp); 785 andcc( Rtmp, 3, G0); 786 Label aligned; 787 switch (should_set_CC ) { 788 default: ShouldNotReachHere(); 789 790 case set_CC: break; 791 case dont_set_CC: break; 792 } 793 794 br(Assembler::zero, true, Assembler::pn, aligned); 795 #ifdef _LP64 796 delayed()->ldsw(Rtmp, 0, Rdst); 797 #else 798 delayed()->ld(Rtmp, 0, Rdst); 799 #endif 800 801 ldub(Lbcp, bcp_offset + 3, Rdst); 802 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst); 803 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst); 804 #ifdef _LP64 805 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); 806 #else 807 // Unsigned load is faster than signed on some implementations 808 ldub(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); 809 #endif 810 or3(Rtmp, Rdst, Rdst ); 811 812 bind(aligned); 813 if (should_set_CC == set_CC) tst(Rdst); 814 } 815 816 817 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register cache, Register tmp, 818 int bcp_offset, bool giant_index) { 819 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 820 if (!giant_index) { 821 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); 822 } else { 823 assert(EnableInvokeDynamic, "giant index used only for EnableInvokeDynamic"); 824 get_4_byte_integer_at_bcp(bcp_offset, cache, tmp); 825 assert(constantPoolCacheOopDesc::decode_secondary_index(~123) == 123, "else change next line"); 826 xor3(tmp, -1, tmp); // convert to plain index 827 } 828 } 829 830 831 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp, 832 int bcp_offset, bool giant_index) { 833 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 834 assert_different_registers(cache, tmp); 835 assert_not_delayed(); 836 get_cache_index_at_bcp(cache, tmp, bcp_offset, giant_index); 837 // convert from field index to ConstantPoolCacheEntry index and from 838 // word index to byte offset 839 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); 840 add(LcpoolCache, tmp, cache); 841 } 842 843 844 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp, 845 int bcp_offset, bool giant_index) { 846 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 847 assert_different_registers(cache, tmp); 848 assert_not_delayed(); 849 assert(!giant_index,"NYI"); 850 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); 851 // convert from field index to ConstantPoolCacheEntry index 852 // and from word index to byte offset 853 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); 854 // skip past the header 855 add(tmp, in_bytes(constantPoolCacheOopDesc::base_offset()), tmp); 856 // construct pointer to cache entry 857 add(LcpoolCache, tmp, cache); 858 } 859 860 861 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 862 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. 863 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 864 Register Rsuper_klass, 865 Register Rtmp1, 866 Register Rtmp2, 867 Register Rtmp3, 868 Label &ok_is_subtype ) { 869 Label not_subtype; 870 871 // Profile the not-null value's klass. 872 profile_typecheck(Rsub_klass, Rtmp1); 873 874 check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass, 875 Rtmp1, Rtmp2, 876 &ok_is_subtype, ¬_subtype, NULL); 877 878 check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass, 879 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg, 880 &ok_is_subtype, NULL); 881 882 bind(not_subtype); 883 profile_typecheck_failed(Rtmp1); 884 } 885 886 // Separate these two to allow for delay slot in middle 887 // These are used to do a test and full jump to exception-throwing code. 888 889 // %%%%% Could possibly reoptimize this by testing to see if could use 890 // a single conditional branch (i.e. if span is small enough. 891 // If you go that route, than get rid of the split and give up 892 // on the delay-slot hack. 893 894 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition, 895 Label& ok ) { 896 assert_not_delayed(); 897 br(ok_condition, true, pt, ok); 898 // DELAY SLOT 899 } 900 901 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition, 902 Label& ok ) { 903 assert_not_delayed(); 904 bp( ok_condition, true, Assembler::xcc, pt, ok); 905 // DELAY SLOT 906 } 907 908 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition, 909 Label& ok ) { 910 assert_not_delayed(); 911 brx(ok_condition, true, pt, ok); 912 // DELAY SLOT 913 } 914 915 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point, 916 Register Rscratch, 917 Label& ok ) { 918 assert(throw_entry_point != NULL, "entry point must be generated by now"); 919 AddressLiteral dest(throw_entry_point); 920 jump_to(dest, Rscratch); 921 delayed()->nop(); 922 bind(ok); 923 } 924 925 926 // And if you cannot use the delay slot, here is a shorthand: 927 928 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition, 929 address throw_entry_point, 930 Register Rscratch ) { 931 Label ok; 932 if (ok_condition != never) { 933 throw_if_not_1_icc( ok_condition, ok); 934 delayed()->nop(); 935 } 936 throw_if_not_2( throw_entry_point, Rscratch, ok); 937 } 938 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition, 939 address throw_entry_point, 940 Register Rscratch ) { 941 Label ok; 942 if (ok_condition != never) { 943 throw_if_not_1_xcc( ok_condition, ok); 944 delayed()->nop(); 945 } 946 throw_if_not_2( throw_entry_point, Rscratch, ok); 947 } 948 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition, 949 address throw_entry_point, 950 Register Rscratch ) { 951 Label ok; 952 if (ok_condition != never) { 953 throw_if_not_1_x( ok_condition, ok); 954 delayed()->nop(); 955 } 956 throw_if_not_2( throw_entry_point, Rscratch, ok); 957 } 958 959 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res 960 // Note: res is still shy of address by array offset into object. 961 962 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) { 963 assert_not_delayed(); 964 965 verify_oop(array); 966 #ifdef _LP64 967 // sign extend since tos (index) can be a 32bit value 968 sra(index, G0, index); 969 #endif // _LP64 970 971 // check array 972 Label ptr_ok; 973 tst(array); 974 throw_if_not_1_x( notZero, ptr_ok ); 975 delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index 976 throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok); 977 978 Label index_ok; 979 cmp(index, tmp); 980 throw_if_not_1_icc( lessUnsigned, index_ok ); 981 if (index_shift > 0) delayed()->sll(index, index_shift, index); 982 else delayed()->add(array, index, res); // addr - const offset in index 983 // convention: move aberrant index into G3_scratch for exception message 984 mov(index, G3_scratch); 985 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok); 986 987 // add offset if didn't do it in delay slot 988 if (index_shift > 0) add(array, index, res); // addr - const offset in index 989 } 990 991 992 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { 993 assert_not_delayed(); 994 995 // pop array 996 pop_ptr(array); 997 998 // check array 999 index_check_without_pop(array, index, index_shift, tmp, res); 1000 } 1001 1002 1003 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 1004 ld_ptr(Lmethod, in_bytes(methodOopDesc::constants_offset()), Rdst); 1005 } 1006 1007 1008 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 1009 get_constant_pool(Rdst); 1010 ld_ptr(Rdst, constantPoolOopDesc::cache_offset_in_bytes(), Rdst); 1011 } 1012 1013 1014 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 1015 get_constant_pool(Rcpool); 1016 ld_ptr(Rcpool, constantPoolOopDesc::tags_offset_in_bytes(), Rtags); 1017 } 1018 1019 1020 // unlock if synchronized method 1021 // 1022 // Unlock the receiver if this is a synchronized method. 1023 // Unlock any Java monitors from syncronized blocks. 1024 // 1025 // If there are locked Java monitors 1026 // If throw_monitor_exception 1027 // throws IllegalMonitorStateException 1028 // Else if install_monitor_exception 1029 // installs IllegalMonitorStateException 1030 // Else 1031 // no error processing 1032 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 1033 bool throw_monitor_exception, 1034 bool install_monitor_exception) { 1035 Label unlocked, unlock, no_unlock; 1036 1037 // get the value of _do_not_unlock_if_synchronized into G1_scratch 1038 const Address do_not_unlock_if_synchronized(G2_thread, 1039 JavaThread::do_not_unlock_if_synchronized_offset()); 1040 ldbool(do_not_unlock_if_synchronized, G1_scratch); 1041 stbool(G0, do_not_unlock_if_synchronized); // reset the flag 1042 1043 // check if synchronized method 1044 const Address access_flags(Lmethod, methodOopDesc::access_flags_offset()); 1045 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1046 push(state); // save tos 1047 ld(access_flags, G3_scratch); // Load access flags. 1048 btst(JVM_ACC_SYNCHRONIZED, G3_scratch); 1049 br(zero, false, pt, unlocked); 1050 delayed()->nop(); 1051 1052 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 1053 // is set. 1054 tstbool(G1_scratch); 1055 br(Assembler::notZero, false, pn, no_unlock); 1056 delayed()->nop(); 1057 1058 // BasicObjectLock will be first in list, since this is a synchronized method. However, need 1059 // to check that the object has not been unlocked by an explicit monitorexit bytecode. 1060 1061 //Intel: if (throw_monitor_exception) ... else ... 1062 // Entry already unlocked, need to throw exception 1063 //... 1064 1065 // pass top-most monitor elem 1066 add( top_most_monitor(), O1 ); 1067 1068 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch); 1069 br_notnull(G3_scratch, false, pt, unlock); 1070 delayed()->nop(); 1071 1072 if (throw_monitor_exception) { 1073 // Entry already unlocked need to throw an exception 1074 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1075 should_not_reach_here(); 1076 } else { 1077 // Monitor already unlocked during a stack unroll. 1078 // If requested, install an illegal_monitor_state_exception. 1079 // Continue with stack unrolling. 1080 if (install_monitor_exception) { 1081 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1082 } 1083 ba(false, unlocked); 1084 delayed()->nop(); 1085 } 1086 1087 bind(unlock); 1088 1089 unlock_object(O1); 1090 1091 bind(unlocked); 1092 1093 // I0, I1: Might contain return value 1094 1095 // Check that all monitors are unlocked 1096 { Label loop, exception, entry, restart; 1097 1098 Register Rmptr = O0; 1099 Register Rtemp = O1; 1100 Register Rlimit = Lmonitors; 1101 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 1102 assert( (delta & LongAlignmentMask) == 0, 1103 "sizeof BasicObjectLock must be even number of doublewords"); 1104 1105 #ifdef ASSERT 1106 add(top_most_monitor(), Rmptr, delta); 1107 { Label L; 1108 // ensure that Rmptr starts out above (or at) Rlimit 1109 cmp(Rmptr, Rlimit); 1110 brx(Assembler::greaterEqualUnsigned, false, pn, L); 1111 delayed()->nop(); 1112 stop("monitor stack has negative size"); 1113 bind(L); 1114 } 1115 #endif 1116 bind(restart); 1117 ba(false, entry); 1118 delayed()-> 1119 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry 1120 1121 // Entry is still locked, need to throw exception 1122 bind(exception); 1123 if (throw_monitor_exception) { 1124 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1125 should_not_reach_here(); 1126 } else { 1127 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 1128 // Unlock does not block, so don't have to worry about the frame 1129 unlock_object(Rmptr); 1130 if (install_monitor_exception) { 1131 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1132 } 1133 ba(false, restart); 1134 delayed()->nop(); 1135 } 1136 1137 bind(loop); 1138 cmp(Rtemp, G0); // check if current entry is used 1139 brx(Assembler::notEqual, false, pn, exception); 1140 delayed()-> 1141 dec(Rmptr, delta); // otherwise advance to next entry 1142 #ifdef ASSERT 1143 { Label L; 1144 // ensure that Rmptr has not somehow stepped below Rlimit 1145 cmp(Rmptr, Rlimit); 1146 brx(Assembler::greaterEqualUnsigned, false, pn, L); 1147 delayed()->nop(); 1148 stop("ran off the end of the monitor stack"); 1149 bind(L); 1150 } 1151 #endif 1152 bind(entry); 1153 cmp(Rmptr, Rlimit); // check if bottom reached 1154 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry 1155 delayed()-> 1156 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp); 1157 } 1158 1159 bind(no_unlock); 1160 pop(state); 1161 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1162 } 1163 1164 1165 // remove activation 1166 // 1167 // Unlock the receiver if this is a synchronized method. 1168 // Unlock any Java monitors from syncronized blocks. 1169 // Remove the activation from the stack. 1170 // 1171 // If there are locked Java monitors 1172 // If throw_monitor_exception 1173 // throws IllegalMonitorStateException 1174 // Else if install_monitor_exception 1175 // installs IllegalMonitorStateException 1176 // Else 1177 // no error processing 1178 void InterpreterMacroAssembler::remove_activation(TosState state, 1179 bool throw_monitor_exception, 1180 bool install_monitor_exception) { 1181 1182 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 1183 1184 // save result (push state before jvmti call and pop it afterwards) and notify jvmti 1185 notify_method_exit(false, state, NotifyJVMTI); 1186 1187 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1188 verify_oop(Lmethod); 1189 verify_thread(); 1190 1191 // return tos 1192 assert(Otos_l1 == Otos_i, "adjust code below"); 1193 switch (state) { 1194 #ifdef _LP64 1195 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0 1196 #else 1197 case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through // O1 -> I1 1198 #endif 1199 case btos: // fall through 1200 case ctos: 1201 case stos: // fall through 1202 case atos: // fall through 1203 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0 1204 case ftos: // fall through 1205 case dtos: // fall through 1206 case vtos: /* nothing to do */ break; 1207 default : ShouldNotReachHere(); 1208 } 1209 1210 #if defined(COMPILER2) && !defined(_LP64) 1211 if (state == ltos) { 1212 // C2 expects long results in G1 we can't tell if we're returning to interpreted 1213 // or compiled so just be safe use G1 and O0/O1 1214 1215 // Shift bits into high (msb) of G1 1216 sllx(Otos_l1->after_save(), 32, G1); 1217 // Zero extend low bits 1218 srl (Otos_l2->after_save(), 0, Otos_l2->after_save()); 1219 or3 (Otos_l2->after_save(), G1, G1); 1220 } 1221 #endif /* COMPILER2 */ 1222 1223 } 1224 #endif /* CC_INTERP */ 1225 1226 1227 // Lock object 1228 // 1229 // Argument - lock_reg points to the BasicObjectLock to be used for locking, 1230 // it must be initialized with the object to lock 1231 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) { 1232 if (UseHeavyMonitors) { 1233 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1234 } 1235 else { 1236 Register obj_reg = Object; 1237 Register mark_reg = G4_scratch; 1238 Register temp_reg = G1_scratch; 1239 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes()); 1240 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1241 Label done; 1242 1243 Label slow_case; 1244 1245 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg); 1246 1247 // load markOop from object into mark_reg 1248 ld_ptr(mark_addr, mark_reg); 1249 1250 if (UseBiasedLocking) { 1251 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case); 1252 } 1253 1254 // get the address of basicLock on stack that will be stored in the object 1255 // we need a temporary register here as we do not want to clobber lock_reg 1256 // (cas clobbers the destination register) 1257 mov(lock_reg, temp_reg); 1258 // set mark reg to be (markOop of object | UNLOCK_VALUE) 1259 or3(mark_reg, markOopDesc::unlocked_value, mark_reg); 1260 // initialize the box (Must happen before we update the object mark!) 1261 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1262 // compare and exchange object_addr, markOop | 1, stack address of basicLock 1263 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1264 casx_under_lock(mark_addr.base(), mark_reg, temp_reg, 1265 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); 1266 1267 // if the compare and exchange succeeded we are done (we saw an unlocked object) 1268 cmp(mark_reg, temp_reg); 1269 brx(Assembler::equal, true, Assembler::pt, done); 1270 delayed()->nop(); 1271 1272 // We did not see an unlocked object so try the fast recursive case 1273 1274 // Check if owner is self by comparing the value in the markOop of object 1275 // with the stack pointer 1276 sub(temp_reg, SP, temp_reg); 1277 #ifdef _LP64 1278 sub(temp_reg, STACK_BIAS, temp_reg); 1279 #endif 1280 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 1281 1282 // Composite "andcc" test: 1283 // (a) %sp -vs- markword proximity check, and, 1284 // (b) verify mark word LSBs == 0 (Stack-locked). 1285 // 1286 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size()) 1287 // Note that the page size used for %sp proximity testing is arbitrary and is 1288 // unrelated to the actual MMU page size. We use a 'logical' page size of 1289 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate 1290 // field of the andcc instruction. 1291 andcc (temp_reg, 0xFFFFF003, G0) ; 1292 1293 // if condition is true we are done and hence we can store 0 in the displaced 1294 // header indicating it is a recursive lock and be done 1295 brx(Assembler::zero, true, Assembler::pt, done); 1296 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1297 1298 // none of the above fast optimizations worked so we have to get into the 1299 // slow case of monitor enter 1300 bind(slow_case); 1301 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1302 1303 bind(done); 1304 } 1305 } 1306 1307 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 1308 // 1309 // Argument - lock_reg points to the BasicObjectLock for lock 1310 // Throw IllegalMonitorException if object is not locked by current thread 1311 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1312 if (UseHeavyMonitors) { 1313 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1314 } else { 1315 Register obj_reg = G3_scratch; 1316 Register mark_reg = G4_scratch; 1317 Register displaced_header_reg = G1_scratch; 1318 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes()); 1319 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1320 Label done; 1321 1322 if (UseBiasedLocking) { 1323 // load the object out of the BasicObjectLock 1324 ld_ptr(lockobj_addr, obj_reg); 1325 biased_locking_exit(mark_addr, mark_reg, done, true); 1326 st_ptr(G0, lockobj_addr); // free entry 1327 } 1328 1329 // Test first if we are in the fast recursive case 1330 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes()); 1331 ld_ptr(lock_addr, displaced_header_reg); 1332 br_null(displaced_header_reg, true, Assembler::pn, done); 1333 delayed()->st_ptr(G0, lockobj_addr); // free entry 1334 1335 // See if it is still a light weight lock, if so we just unlock 1336 // the object and we are done 1337 1338 if (!UseBiasedLocking) { 1339 // load the object out of the BasicObjectLock 1340 ld_ptr(lockobj_addr, obj_reg); 1341 } 1342 1343 // we have the displaced header in displaced_header_reg 1344 // we expect to see the stack address of the basicLock in case the 1345 // lock is still a light weight lock (lock_reg) 1346 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1347 casx_under_lock(mark_addr.base(), lock_reg, displaced_header_reg, 1348 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); 1349 cmp(lock_reg, displaced_header_reg); 1350 brx(Assembler::equal, true, Assembler::pn, done); 1351 delayed()->st_ptr(G0, lockobj_addr); // free entry 1352 1353 // The lock has been converted into a heavy lock and hence 1354 // we need to get into the slow case 1355 1356 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1357 1358 bind(done); 1359 } 1360 } 1361 1362 #ifndef CC_INTERP 1363 1364 // Get the method data pointer from the methodOop and set the 1365 // specified register to its value. 1366 1367 void InterpreterMacroAssembler::set_method_data_pointer_offset(Register Roff) { 1368 assert(ProfileInterpreter, "must be profiling interpreter"); 1369 Label get_continue; 1370 1371 ld_ptr(Lmethod, in_bytes(methodOopDesc::method_data_offset()), ImethodDataPtr); 1372 test_method_data_pointer(get_continue); 1373 add(ImethodDataPtr, in_bytes(methodDataOopDesc::data_offset()), ImethodDataPtr); 1374 if (Roff != noreg) 1375 // Roff contains a method data index ("mdi"). It defaults to zero. 1376 add(ImethodDataPtr, Roff, ImethodDataPtr); 1377 bind(get_continue); 1378 } 1379 1380 // Set the method data pointer for the current bcp. 1381 1382 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1383 assert(ProfileInterpreter, "must be profiling interpreter"); 1384 Label zero_continue; 1385 1386 // Test MDO to avoid the call if it is NULL. 1387 ld_ptr(Lmethod, methodOopDesc::method_data_offset(), ImethodDataPtr); 1388 test_method_data_pointer(zero_continue); 1389 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp); 1390 set_method_data_pointer_offset(O0); 1391 bind(zero_continue); 1392 } 1393 1394 // Test ImethodDataPtr. If it is null, continue at the specified label 1395 1396 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1397 assert(ProfileInterpreter, "must be profiling interpreter"); 1398 #ifdef _LP64 1399 bpr(Assembler::rc_z, false, Assembler::pn, ImethodDataPtr, zero_continue); 1400 #else 1401 tst(ImethodDataPtr); 1402 br(Assembler::zero, false, Assembler::pn, zero_continue); 1403 #endif 1404 delayed()->nop(); 1405 } 1406 1407 void InterpreterMacroAssembler::verify_method_data_pointer() { 1408 assert(ProfileInterpreter, "must be profiling interpreter"); 1409 #ifdef ASSERT 1410 Label verify_continue; 1411 test_method_data_pointer(verify_continue); 1412 1413 // If the mdp is valid, it will point to a DataLayout header which is 1414 // consistent with the bcp. The converse is highly probable also. 1415 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch); 1416 ld_ptr(Lmethod, methodOopDesc::const_offset(), O5); 1417 add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), G3_scratch); 1418 add(G3_scratch, O5, G3_scratch); 1419 cmp(Lbcp, G3_scratch); 1420 brx(Assembler::equal, false, Assembler::pt, verify_continue); 1421 1422 Register temp_reg = O5; 1423 delayed()->mov(ImethodDataPtr, temp_reg); 1424 // %%% should use call_VM_leaf here? 1425 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr); 1426 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1); 1427 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS); 1428 stf(FloatRegisterImpl::D, Ftos_d, d_save); 1429 mov(temp_reg->after_save(), O2); 1430 save_thread(L7_thread_cache); 1431 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none); 1432 delayed()->nop(); 1433 restore_thread(L7_thread_cache); 1434 ldf(FloatRegisterImpl::D, d_save, Ftos_d); 1435 restore(); 1436 bind(verify_continue); 1437 #endif // ASSERT 1438 } 1439 1440 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1441 Register cur_bcp, 1442 Register Rtmp, 1443 Label &profile_continue) { 1444 assert(ProfileInterpreter, "must be profiling interpreter"); 1445 // Control will flow to "profile_continue" if the counter is less than the 1446 // limit or if we call profile_method() 1447 1448 Label done; 1449 1450 // if no method data exists, and the counter is high enough, make one 1451 #ifdef _LP64 1452 bpr(Assembler::rc_nz, false, Assembler::pn, ImethodDataPtr, done); 1453 #else 1454 tst(ImethodDataPtr); 1455 br(Assembler::notZero, false, Assembler::pn, done); 1456 #endif 1457 1458 // Test to see if we should create a method data oop 1459 AddressLiteral profile_limit((address) &InvocationCounter::InterpreterProfileLimit); 1460 #ifdef _LP64 1461 delayed()->nop(); 1462 sethi(profile_limit, Rtmp); 1463 #else 1464 delayed()->sethi(profile_limit, Rtmp); 1465 #endif 1466 ld(Rtmp, profile_limit.low10(), Rtmp); 1467 cmp(invocation_count, Rtmp); 1468 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue); 1469 delayed()->nop(); 1470 1471 // Build it now. 1472 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), cur_bcp); 1473 set_method_data_pointer_offset(O0); 1474 ba(false, profile_continue); 1475 delayed()->nop(); 1476 bind(done); 1477 } 1478 1479 // Store a value at some constant offset from the method data pointer. 1480 1481 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1482 assert(ProfileInterpreter, "must be profiling interpreter"); 1483 st_ptr(value, ImethodDataPtr, constant); 1484 } 1485 1486 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter, 1487 Register bumped_count, 1488 bool decrement) { 1489 assert(ProfileInterpreter, "must be profiling interpreter"); 1490 1491 // Load the counter. 1492 ld_ptr(counter, bumped_count); 1493 1494 if (decrement) { 1495 // Decrement the register. Set condition codes. 1496 subcc(bumped_count, DataLayout::counter_increment, bumped_count); 1497 1498 // If the decrement causes the counter to overflow, stay negative 1499 Label L; 1500 brx(Assembler::negative, true, Assembler::pn, L); 1501 1502 // Store the decremented counter, if it is still negative. 1503 delayed()->st_ptr(bumped_count, counter); 1504 bind(L); 1505 } else { 1506 // Increment the register. Set carry flag. 1507 addcc(bumped_count, DataLayout::counter_increment, bumped_count); 1508 1509 // If the increment causes the counter to overflow, pull back by 1. 1510 assert(DataLayout::counter_increment == 1, "subc works"); 1511 subc(bumped_count, G0, bumped_count); 1512 1513 // Store the incremented counter. 1514 st_ptr(bumped_count, counter); 1515 } 1516 } 1517 1518 // Increment the value at some constant offset from the method data pointer. 1519 1520 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1521 Register bumped_count, 1522 bool decrement) { 1523 // Locate the counter at a fixed offset from the mdp: 1524 Address counter(ImethodDataPtr, constant); 1525 increment_mdp_data_at(counter, bumped_count, decrement); 1526 } 1527 1528 // Increment the value at some non-fixed (reg + constant) offset from 1529 // the method data pointer. 1530 1531 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1532 int constant, 1533 Register bumped_count, 1534 Register scratch2, 1535 bool decrement) { 1536 // Add the constant to reg to get the offset. 1537 add(ImethodDataPtr, reg, scratch2); 1538 Address counter(scratch2, constant); 1539 increment_mdp_data_at(counter, bumped_count, decrement); 1540 } 1541 1542 // Set a flag value at the current method data pointer position. 1543 // Updates a single byte of the header, to avoid races with other header bits. 1544 1545 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1546 Register scratch) { 1547 assert(ProfileInterpreter, "must be profiling interpreter"); 1548 // Load the data header 1549 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch); 1550 1551 // Set the flag 1552 or3(scratch, flag_constant, scratch); 1553 1554 // Store the modified header. 1555 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset())); 1556 } 1557 1558 // Test the location at some offset from the method data pointer. 1559 // If it is not equal to value, branch to the not_equal_continue Label. 1560 // Set condition codes to match the nullness of the loaded value. 1561 1562 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1563 Register value, 1564 Label& not_equal_continue, 1565 Register scratch) { 1566 assert(ProfileInterpreter, "must be profiling interpreter"); 1567 ld_ptr(ImethodDataPtr, offset, scratch); 1568 cmp(value, scratch); 1569 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue); 1570 delayed()->tst(scratch); 1571 } 1572 1573 // Update the method data pointer by the displacement located at some fixed 1574 // offset from the method data pointer. 1575 1576 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1577 Register scratch) { 1578 assert(ProfileInterpreter, "must be profiling interpreter"); 1579 ld_ptr(ImethodDataPtr, offset_of_disp, scratch); 1580 add(ImethodDataPtr, scratch, ImethodDataPtr); 1581 } 1582 1583 // Update the method data pointer by the displacement located at the 1584 // offset (reg + offset_of_disp). 1585 1586 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1587 int offset_of_disp, 1588 Register scratch) { 1589 assert(ProfileInterpreter, "must be profiling interpreter"); 1590 add(reg, offset_of_disp, scratch); 1591 ld_ptr(ImethodDataPtr, scratch, scratch); 1592 add(ImethodDataPtr, scratch, ImethodDataPtr); 1593 } 1594 1595 // Update the method data pointer by a simple constant displacement. 1596 1597 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1598 assert(ProfileInterpreter, "must be profiling interpreter"); 1599 add(ImethodDataPtr, constant, ImethodDataPtr); 1600 } 1601 1602 // Update the method data pointer for a _ret bytecode whose target 1603 // was not among our cached targets. 1604 1605 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1606 Register return_bci) { 1607 assert(ProfileInterpreter, "must be profiling interpreter"); 1608 push(state); 1609 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile 1610 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1611 ld_ptr(l_tmp, return_bci); 1612 pop(state); 1613 } 1614 1615 // Count a taken branch in the bytecodes. 1616 1617 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1618 if (ProfileInterpreter) { 1619 Label profile_continue; 1620 1621 // If no method data exists, go to profile_continue. 1622 test_method_data_pointer(profile_continue); 1623 1624 // We are taking a branch. Increment the taken count. 1625 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count); 1626 1627 // The method data pointer needs to be updated to reflect the new target. 1628 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1629 bind (profile_continue); 1630 } 1631 } 1632 1633 1634 // Count a not-taken branch in the bytecodes. 1635 1636 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) { 1637 if (ProfileInterpreter) { 1638 Label profile_continue; 1639 1640 // If no method data exists, go to profile_continue. 1641 test_method_data_pointer(profile_continue); 1642 1643 // We are taking a branch. Increment the not taken count. 1644 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch); 1645 1646 // The method data pointer needs to be updated to correspond to the 1647 // next bytecode. 1648 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1649 bind (profile_continue); 1650 } 1651 } 1652 1653 1654 // Count a non-virtual call in the bytecodes. 1655 1656 void InterpreterMacroAssembler::profile_call(Register scratch) { 1657 if (ProfileInterpreter) { 1658 Label profile_continue; 1659 1660 // If no method data exists, go to profile_continue. 1661 test_method_data_pointer(profile_continue); 1662 1663 // We are making a call. Increment the count. 1664 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1665 1666 // The method data pointer needs to be updated to reflect the new target. 1667 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1668 bind (profile_continue); 1669 } 1670 } 1671 1672 1673 // Count a final call in the bytecodes. 1674 1675 void InterpreterMacroAssembler::profile_final_call(Register scratch) { 1676 if (ProfileInterpreter) { 1677 Label profile_continue; 1678 1679 // If no method data exists, go to profile_continue. 1680 test_method_data_pointer(profile_continue); 1681 1682 // We are making a call. Increment the count. 1683 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1684 1685 // The method data pointer needs to be updated to reflect the new target. 1686 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1687 bind (profile_continue); 1688 } 1689 } 1690 1691 1692 // Count a virtual call in the bytecodes. 1693 1694 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1695 Register scratch, 1696 bool receiver_can_be_null) { 1697 if (ProfileInterpreter) { 1698 Label profile_continue; 1699 1700 // If no method data exists, go to profile_continue. 1701 test_method_data_pointer(profile_continue); 1702 1703 1704 Label skip_receiver_profile; 1705 if (receiver_can_be_null) { 1706 Label not_null; 1707 tst(receiver); 1708 brx(Assembler::notZero, false, Assembler::pt, not_null); 1709 delayed()->nop(); 1710 // We are making a call. Increment the count for null receiver. 1711 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1712 ba(false, skip_receiver_profile); 1713 delayed()->nop(); 1714 bind(not_null); 1715 } 1716 1717 // Record the receiver type. 1718 record_klass_in_profile(receiver, scratch, true); 1719 bind(skip_receiver_profile); 1720 1721 // The method data pointer needs to be updated to reflect the new target. 1722 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1723 bind (profile_continue); 1724 } 1725 } 1726 1727 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1728 Register receiver, Register scratch, 1729 int start_row, Label& done, bool is_virtual_call) { 1730 if (TypeProfileWidth == 0) { 1731 if (is_virtual_call) { 1732 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1733 } 1734 return; 1735 } 1736 1737 int last_row = VirtualCallData::row_limit() - 1; 1738 assert(start_row <= last_row, "must be work left to do"); 1739 // Test this row for both the receiver and for null. 1740 // Take any of three different outcomes: 1741 // 1. found receiver => increment count and goto done 1742 // 2. found null => keep looking for case 1, maybe allocate this cell 1743 // 3. found something else => keep looking for cases 1 and 2 1744 // Case 3 is handled by a recursive call. 1745 for (int row = start_row; row <= last_row; row++) { 1746 Label next_test; 1747 bool test_for_null_also = (row == start_row); 1748 1749 // See if the receiver is receiver[n]. 1750 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1751 test_mdp_data_at(recvr_offset, receiver, next_test, scratch); 1752 // delayed()->tst(scratch); 1753 1754 // The receiver is receiver[n]. Increment count[n]. 1755 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1756 increment_mdp_data_at(count_offset, scratch); 1757 ba(false, done); 1758 delayed()->nop(); 1759 bind(next_test); 1760 1761 if (test_for_null_also) { 1762 Label found_null; 1763 // Failed the equality check on receiver[n]... Test for null. 1764 if (start_row == last_row) { 1765 // The only thing left to do is handle the null case. 1766 if (is_virtual_call) { 1767 brx(Assembler::zero, false, Assembler::pn, found_null); 1768 delayed()->nop(); 1769 // Receiver did not match any saved receiver and there is no empty row for it. 1770 // Increment total counter to indicate polymorphic case. 1771 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1772 ba(false, done); 1773 delayed()->nop(); 1774 bind(found_null); 1775 } else { 1776 brx(Assembler::notZero, false, Assembler::pt, done); 1777 delayed()->nop(); 1778 } 1779 break; 1780 } 1781 // Since null is rare, make it be the branch-taken case. 1782 brx(Assembler::zero, false, Assembler::pn, found_null); 1783 delayed()->nop(); 1784 1785 // Put all the "Case 3" tests here. 1786 record_klass_in_profile_helper(receiver, scratch, start_row + 1, done, is_virtual_call); 1787 1788 // Found a null. Keep searching for a matching receiver, 1789 // but remember that this is an empty (unused) slot. 1790 bind(found_null); 1791 } 1792 } 1793 1794 // In the fall-through case, we found no matching receiver, but we 1795 // observed the receiver[start_row] is NULL. 1796 1797 // Fill in the receiver field and increment the count. 1798 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1799 set_mdp_data_at(recvr_offset, receiver); 1800 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1801 mov(DataLayout::counter_increment, scratch); 1802 set_mdp_data_at(count_offset, scratch); 1803 if (start_row > 0) { 1804 ba(false, done); 1805 delayed()->nop(); 1806 } 1807 } 1808 1809 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1810 Register scratch, bool is_virtual_call) { 1811 assert(ProfileInterpreter, "must be profiling"); 1812 Label done; 1813 1814 record_klass_in_profile_helper(receiver, scratch, 0, done, is_virtual_call); 1815 1816 bind (done); 1817 } 1818 1819 1820 // Count a ret in the bytecodes. 1821 1822 void InterpreterMacroAssembler::profile_ret(TosState state, 1823 Register return_bci, 1824 Register scratch) { 1825 if (ProfileInterpreter) { 1826 Label profile_continue; 1827 uint row; 1828 1829 // If no method data exists, go to profile_continue. 1830 test_method_data_pointer(profile_continue); 1831 1832 // Update the total ret count. 1833 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1834 1835 for (row = 0; row < RetData::row_limit(); row++) { 1836 Label next_test; 1837 1838 // See if return_bci is equal to bci[n]: 1839 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), 1840 return_bci, next_test, scratch); 1841 1842 // return_bci is equal to bci[n]. Increment the count. 1843 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch); 1844 1845 // The method data pointer needs to be updated to reflect the new target. 1846 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch); 1847 ba(false, profile_continue); 1848 delayed()->nop(); 1849 bind(next_test); 1850 } 1851 1852 update_mdp_for_ret(state, return_bci); 1853 1854 bind (profile_continue); 1855 } 1856 } 1857 1858 // Profile an unexpected null in the bytecodes. 1859 void InterpreterMacroAssembler::profile_null_seen(Register scratch) { 1860 if (ProfileInterpreter) { 1861 Label profile_continue; 1862 1863 // If no method data exists, go to profile_continue. 1864 test_method_data_pointer(profile_continue); 1865 1866 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch); 1867 1868 // The method data pointer needs to be updated. 1869 int mdp_delta = in_bytes(BitData::bit_data_size()); 1870 if (TypeProfileCasts) { 1871 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1872 } 1873 update_mdp_by_constant(mdp_delta); 1874 1875 bind (profile_continue); 1876 } 1877 } 1878 1879 void InterpreterMacroAssembler::profile_typecheck(Register klass, 1880 Register scratch) { 1881 if (ProfileInterpreter) { 1882 Label profile_continue; 1883 1884 // If no method data exists, go to profile_continue. 1885 test_method_data_pointer(profile_continue); 1886 1887 int mdp_delta = in_bytes(BitData::bit_data_size()); 1888 if (TypeProfileCasts) { 1889 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1890 1891 // Record the object type. 1892 record_klass_in_profile(klass, scratch, false); 1893 } 1894 1895 // The method data pointer needs to be updated. 1896 update_mdp_by_constant(mdp_delta); 1897 1898 bind (profile_continue); 1899 } 1900 } 1901 1902 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) { 1903 if (ProfileInterpreter && TypeProfileCasts) { 1904 Label profile_continue; 1905 1906 // If no method data exists, go to profile_continue. 1907 test_method_data_pointer(profile_continue); 1908 1909 int count_offset = in_bytes(CounterData::count_offset()); 1910 // Back up the address, since we have already bumped the mdp. 1911 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1912 1913 // *Decrement* the counter. We expect to see zero or small negatives. 1914 increment_mdp_data_at(count_offset, scratch, true); 1915 1916 bind (profile_continue); 1917 } 1918 } 1919 1920 // Count the default case of a switch construct. 1921 1922 void InterpreterMacroAssembler::profile_switch_default(Register scratch) { 1923 if (ProfileInterpreter) { 1924 Label profile_continue; 1925 1926 // If no method data exists, go to profile_continue. 1927 test_method_data_pointer(profile_continue); 1928 1929 // Update the default case count 1930 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1931 scratch); 1932 1933 // The method data pointer needs to be updated. 1934 update_mdp_by_offset( 1935 in_bytes(MultiBranchData::default_displacement_offset()), 1936 scratch); 1937 1938 bind (profile_continue); 1939 } 1940 } 1941 1942 // Count the index'th case of a switch construct. 1943 1944 void InterpreterMacroAssembler::profile_switch_case(Register index, 1945 Register scratch, 1946 Register scratch2, 1947 Register scratch3) { 1948 if (ProfileInterpreter) { 1949 Label profile_continue; 1950 1951 // If no method data exists, go to profile_continue. 1952 test_method_data_pointer(profile_continue); 1953 1954 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() 1955 set(in_bytes(MultiBranchData::per_case_size()), scratch); 1956 smul(index, scratch, scratch); 1957 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch); 1958 1959 // Update the case count 1960 increment_mdp_data_at(scratch, 1961 in_bytes(MultiBranchData::relative_count_offset()), 1962 scratch2, 1963 scratch3); 1964 1965 // The method data pointer needs to be updated. 1966 update_mdp_by_offset(scratch, 1967 in_bytes(MultiBranchData::relative_displacement_offset()), 1968 scratch2); 1969 1970 bind (profile_continue); 1971 } 1972 } 1973 1974 // add a InterpMonitorElem to stack (see frame_sparc.hpp) 1975 1976 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty, 1977 Register Rtemp, 1978 Register Rtemp2 ) { 1979 1980 Register Rlimit = Lmonitors; 1981 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 1982 assert( (delta & LongAlignmentMask) == 0, 1983 "sizeof BasicObjectLock must be even number of doublewords"); 1984 1985 sub( SP, delta, SP); 1986 sub( Lesp, delta, Lesp); 1987 sub( Lmonitors, delta, Lmonitors); 1988 1989 if (!stack_is_empty) { 1990 1991 // must copy stack contents down 1992 1993 Label start_copying, next; 1994 1995 // untested("monitor stack expansion"); 1996 compute_stack_base(Rtemp); 1997 ba( false, start_copying ); 1998 delayed()->cmp( Rtemp, Rlimit); // done? duplicated below 1999 2000 // note: must copy from low memory upwards 2001 // On entry to loop, 2002 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS) 2003 // Loop mutates Rtemp 2004 2005 bind( next); 2006 2007 st_ptr(Rtemp2, Rtemp, 0); 2008 inc(Rtemp, wordSize); 2009 cmp(Rtemp, Rlimit); // are we done? (duplicated above) 2010 2011 bind( start_copying ); 2012 2013 brx( notEqual, true, pn, next ); 2014 delayed()->ld_ptr( Rtemp, delta, Rtemp2 ); 2015 2016 // done copying stack 2017 } 2018 } 2019 2020 // Locals 2021 #ifdef ASSERT 2022 void InterpreterMacroAssembler::verify_local_tag(frame::Tag t, 2023 Register base, 2024 Register scratch, 2025 int n) { 2026 if (TaggedStackInterpreter) { 2027 Label ok, long_ok; 2028 // Use dst for scratch 2029 assert_different_registers(base, scratch); 2030 ld_ptr(base, Interpreter::local_tag_offset_in_bytes(n), scratch); 2031 if (t == frame::TagCategory2) { 2032 cmp(scratch, G0); 2033 brx(Assembler::equal, false, Assembler::pt, long_ok); 2034 delayed()->ld_ptr(base, Interpreter::local_tag_offset_in_bytes(n+1), scratch); 2035 stop("local long/double tag value bad"); 2036 bind(long_ok); 2037 // compare second half tag 2038 cmp(scratch, G0); 2039 } else if (t == frame::TagValue) { 2040 cmp(scratch, G0); 2041 } else { 2042 assert_different_registers(O3, base, scratch); 2043 mov(t, O3); 2044 cmp(scratch, O3); 2045 } 2046 brx(Assembler::equal, false, Assembler::pt, ok); 2047 delayed()->nop(); 2048 // Also compare if the local value is zero, then the tag might 2049 // not have been set coming from deopt. 2050 ld_ptr(base, Interpreter::local_offset_in_bytes(n), scratch); 2051 cmp(scratch, G0); 2052 brx(Assembler::equal, false, Assembler::pt, ok); 2053 delayed()->nop(); 2054 stop("Local tag value is bad"); 2055 bind(ok); 2056 } 2057 } 2058 #endif // ASSERT 2059 2060 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) { 2061 assert_not_delayed(); 2062 sll(index, Interpreter::logStackElementSize(), index); 2063 sub(Llocals, index, index); 2064 debug_only(verify_local_tag(frame::TagReference, index, dst)); 2065 ld_ptr(index, Interpreter::value_offset_in_bytes(), dst); 2066 // Note: index must hold the effective address--the iinc template uses it 2067 } 2068 2069 // Just like access_local_ptr but the tag is a returnAddress 2070 void InterpreterMacroAssembler::access_local_returnAddress(Register index, 2071 Register dst ) { 2072 assert_not_delayed(); 2073 sll(index, Interpreter::logStackElementSize(), index); 2074 sub(Llocals, index, index); 2075 debug_only(verify_local_tag(frame::TagValue, index, dst)); 2076 ld_ptr(index, Interpreter::value_offset_in_bytes(), dst); 2077 } 2078 2079 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) { 2080 assert_not_delayed(); 2081 sll(index, Interpreter::logStackElementSize(), index); 2082 sub(Llocals, index, index); 2083 debug_only(verify_local_tag(frame::TagValue, index, dst)); 2084 ld(index, Interpreter::value_offset_in_bytes(), dst); 2085 // Note: index must hold the effective address--the iinc template uses it 2086 } 2087 2088 2089 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) { 2090 assert_not_delayed(); 2091 sll(index, Interpreter::logStackElementSize(), index); 2092 sub(Llocals, index, index); 2093 debug_only(verify_local_tag(frame::TagCategory2, index, dst)); 2094 // First half stored at index n+1 (which grows down from Llocals[n]) 2095 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst); 2096 } 2097 2098 2099 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) { 2100 assert_not_delayed(); 2101 sll(index, Interpreter::logStackElementSize(), index); 2102 sub(Llocals, index, index); 2103 debug_only(verify_local_tag(frame::TagValue, index, G1_scratch)); 2104 ldf(FloatRegisterImpl::S, index, Interpreter::value_offset_in_bytes(), dst); 2105 } 2106 2107 2108 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) { 2109 assert_not_delayed(); 2110 sll(index, Interpreter::logStackElementSize(), index); 2111 sub(Llocals, index, index); 2112 debug_only(verify_local_tag(frame::TagCategory2, index, G1_scratch)); 2113 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst); 2114 } 2115 2116 2117 #ifdef ASSERT 2118 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) { 2119 Label L; 2120 2121 assert(Rindex != Rscratch, "Registers cannot be same"); 2122 assert(Rindex != Rscratch1, "Registers cannot be same"); 2123 assert(Rlimit != Rscratch, "Registers cannot be same"); 2124 assert(Rlimit != Rscratch1, "Registers cannot be same"); 2125 assert(Rscratch1 != Rscratch, "Registers cannot be same"); 2126 2127 // untested("reg area corruption"); 2128 add(Rindex, offset, Rscratch); 2129 add(Rlimit, 64 + STACK_BIAS, Rscratch1); 2130 cmp(Rscratch, Rscratch1); 2131 brx(Assembler::greaterEqualUnsigned, false, pn, L); 2132 delayed()->nop(); 2133 stop("regsave area is being clobbered"); 2134 bind(L); 2135 } 2136 #endif // ASSERT 2137 2138 void InterpreterMacroAssembler::tag_local(frame::Tag t, 2139 Register base, 2140 Register src, 2141 int n) { 2142 if (TaggedStackInterpreter) { 2143 // have to store zero because local slots can be reused (rats!) 2144 if (t == frame::TagValue) { 2145 st_ptr(G0, base, Interpreter::local_tag_offset_in_bytes(n)); 2146 } else if (t == frame::TagCategory2) { 2147 st_ptr(G0, base, Interpreter::local_tag_offset_in_bytes(n)); 2148 st_ptr(G0, base, Interpreter::local_tag_offset_in_bytes(n+1)); 2149 } else { 2150 // assert that we don't stomp the value in 'src' 2151 // O3 is arbitrary because it's not used. 2152 assert_different_registers(src, base, O3); 2153 mov( t, O3); 2154 st_ptr(O3, base, Interpreter::local_tag_offset_in_bytes(n)); 2155 } 2156 } 2157 } 2158 2159 2160 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) { 2161 assert_not_delayed(); 2162 sll(index, Interpreter::logStackElementSize(), index); 2163 sub(Llocals, index, index); 2164 debug_only(check_for_regarea_stomp(index, Interpreter::value_offset_in_bytes(), FP, G1_scratch, G4_scratch);) 2165 tag_local(frame::TagValue, index, src); 2166 st(src, index, Interpreter::value_offset_in_bytes()); 2167 } 2168 2169 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src, 2170 Register tag ) { 2171 assert_not_delayed(); 2172 sll(index, Interpreter::logStackElementSize(), index); 2173 sub(Llocals, index, index); 2174 #ifdef ASSERT 2175 check_for_regarea_stomp(index, Interpreter::value_offset_in_bytes(), FP, G1_scratch, G4_scratch); 2176 #endif 2177 st_ptr(src, index, Interpreter::value_offset_in_bytes()); 2178 // Store tag register directly 2179 if (TaggedStackInterpreter) { 2180 st_ptr(tag, index, Interpreter::tag_offset_in_bytes()); 2181 } 2182 } 2183 2184 2185 2186 void InterpreterMacroAssembler::store_local_ptr( int n, Register src, 2187 Register tag ) { 2188 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n)); 2189 if (TaggedStackInterpreter) { 2190 st_ptr(tag, Llocals, Interpreter::local_tag_offset_in_bytes(n)); 2191 } 2192 } 2193 2194 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) { 2195 assert_not_delayed(); 2196 sll(index, Interpreter::logStackElementSize(), index); 2197 sub(Llocals, index, index); 2198 #ifdef ASSERT 2199 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2200 #endif 2201 tag_local(frame::TagCategory2, index, src); 2202 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1 2203 } 2204 2205 2206 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) { 2207 assert_not_delayed(); 2208 sll(index, Interpreter::logStackElementSize(), index); 2209 sub(Llocals, index, index); 2210 #ifdef ASSERT 2211 check_for_regarea_stomp(index, Interpreter::value_offset_in_bytes(), FP, G1_scratch, G4_scratch); 2212 #endif 2213 tag_local(frame::TagValue, index, G1_scratch); 2214 stf(FloatRegisterImpl::S, src, index, Interpreter::value_offset_in_bytes()); 2215 } 2216 2217 2218 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) { 2219 assert_not_delayed(); 2220 sll(index, Interpreter::logStackElementSize(), index); 2221 sub(Llocals, index, index); 2222 #ifdef ASSERT 2223 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2224 #endif 2225 tag_local(frame::TagCategory2, index, G1_scratch); 2226 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1)); 2227 } 2228 2229 2230 int InterpreterMacroAssembler::top_most_monitor_byte_offset() { 2231 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 2232 int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong); 2233 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS; 2234 } 2235 2236 2237 Address InterpreterMacroAssembler::top_most_monitor() { 2238 return Address(FP, top_most_monitor_byte_offset()); 2239 } 2240 2241 2242 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) { 2243 add( Lesp, wordSize, Rdest ); 2244 } 2245 2246 #endif /* CC_INTERP */ 2247 2248 void InterpreterMacroAssembler::increment_invocation_counter( Register Rtmp, Register Rtmp2 ) { 2249 assert(UseCompiler, "incrementing must be useful"); 2250 #ifdef CC_INTERP 2251 Address inv_counter(G5_method, methodOopDesc::invocation_counter_offset() + 2252 InvocationCounter::counter_offset()); 2253 Address be_counter (G5_method, methodOopDesc::backedge_counter_offset() + 2254 InvocationCounter::counter_offset()); 2255 #else 2256 Address inv_counter(Lmethod, methodOopDesc::invocation_counter_offset() + 2257 InvocationCounter::counter_offset()); 2258 Address be_counter (Lmethod, methodOopDesc::backedge_counter_offset() + 2259 InvocationCounter::counter_offset()); 2260 #endif /* CC_INTERP */ 2261 int delta = InvocationCounter::count_increment; 2262 2263 // Load each counter in a register 2264 ld( inv_counter, Rtmp ); 2265 ld( be_counter, Rtmp2 ); 2266 2267 assert( is_simm13( delta ), " delta too large."); 2268 2269 // Add the delta to the invocation counter and store the result 2270 add( Rtmp, delta, Rtmp ); 2271 2272 // Mask the backedge counter 2273 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2274 2275 // Store value 2276 st( Rtmp, inv_counter); 2277 2278 // Add invocation counter + backedge counter 2279 add( Rtmp, Rtmp2, Rtmp); 2280 2281 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 2282 } 2283 2284 void InterpreterMacroAssembler::increment_backedge_counter( Register Rtmp, Register Rtmp2 ) { 2285 assert(UseCompiler, "incrementing must be useful"); 2286 #ifdef CC_INTERP 2287 Address be_counter (G5_method, methodOopDesc::backedge_counter_offset() + 2288 InvocationCounter::counter_offset()); 2289 Address inv_counter(G5_method, methodOopDesc::invocation_counter_offset() + 2290 InvocationCounter::counter_offset()); 2291 #else 2292 Address be_counter (Lmethod, methodOopDesc::backedge_counter_offset() + 2293 InvocationCounter::counter_offset()); 2294 Address inv_counter(Lmethod, methodOopDesc::invocation_counter_offset() + 2295 InvocationCounter::counter_offset()); 2296 #endif /* CC_INTERP */ 2297 int delta = InvocationCounter::count_increment; 2298 // Load each counter in a register 2299 ld( be_counter, Rtmp ); 2300 ld( inv_counter, Rtmp2 ); 2301 2302 // Add the delta to the backedge counter 2303 add( Rtmp, delta, Rtmp ); 2304 2305 // Mask the invocation counter, add to backedge counter 2306 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2307 2308 // and store the result to memory 2309 st( Rtmp, be_counter ); 2310 2311 // Add backedge + invocation counter 2312 add( Rtmp, Rtmp2, Rtmp ); 2313 2314 // Note that this macro must leave backedge_count + invocation_count in Rtmp! 2315 } 2316 2317 #ifndef CC_INTERP 2318 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count, 2319 Register branch_bcp, 2320 Register Rtmp ) { 2321 Label did_not_overflow; 2322 Label overflow_with_error; 2323 assert_different_registers(backedge_count, Rtmp, branch_bcp); 2324 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 2325 2326 AddressLiteral limit(&InvocationCounter::InterpreterBackwardBranchLimit); 2327 load_contents(limit, Rtmp); 2328 cmp(backedge_count, Rtmp); 2329 br(Assembler::lessUnsigned, false, Assembler::pt, did_not_overflow); 2330 delayed()->nop(); 2331 2332 // When ProfileInterpreter is on, the backedge_count comes from the 2333 // methodDataOop, which value does not get reset on the call to 2334 // frequency_counter_overflow(). To avoid excessive calls to the overflow 2335 // routine while the method is being compiled, add a second test to make sure 2336 // the overflow function is called only once every overflow_frequency. 2337 if (ProfileInterpreter) { 2338 const int overflow_frequency = 1024; 2339 andcc(backedge_count, overflow_frequency-1, Rtmp); 2340 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow); 2341 delayed()->nop(); 2342 } 2343 2344 // overflow in loop, pass branch bytecode 2345 set(6,Rtmp); 2346 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp); 2347 2348 // Was an OSR adapter generated? 2349 // O0 = osr nmethod 2350 tst(O0); 2351 brx(Assembler::zero, false, Assembler::pn, overflow_with_error); 2352 delayed()->nop(); 2353 2354 // Has the nmethod been invalidated already? 2355 ld(O0, nmethod::entry_bci_offset(), O2); 2356 cmp(O2, InvalidOSREntryBci); 2357 br(Assembler::equal, false, Assembler::pn, overflow_with_error); 2358 delayed()->nop(); 2359 2360 // migrate the interpreter frame off of the stack 2361 2362 mov(G2_thread, L7); 2363 // save nmethod 2364 mov(O0, L6); 2365 set_last_Java_frame(SP, noreg); 2366 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7); 2367 reset_last_Java_frame(); 2368 mov(L7, G2_thread); 2369 2370 // move OSR nmethod to I1 2371 mov(L6, I1); 2372 2373 // OSR buffer to I0 2374 mov(O0, I0); 2375 2376 // remove the interpreter frame 2377 restore(I5_savedSP, 0, SP); 2378 2379 // Jump to the osr code. 2380 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2); 2381 jmp(O2, G0); 2382 delayed()->nop(); 2383 2384 bind(overflow_with_error); 2385 2386 bind(did_not_overflow); 2387 } 2388 2389 2390 2391 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) { 2392 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); } 2393 } 2394 2395 2396 // local helper function for the verify_oop_or_return_address macro 2397 static bool verify_return_address(methodOopDesc* m, int bci) { 2398 #ifndef PRODUCT 2399 address pc = (address)(m->constMethod()) 2400 + in_bytes(constMethodOopDesc::codes_offset()) + bci; 2401 // assume it is a valid return address if it is inside m and is preceded by a jsr 2402 if (!m->contains(pc)) return false; 2403 address jsr_pc; 2404 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2405 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2406 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2407 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2408 #endif // PRODUCT 2409 return false; 2410 } 2411 2412 2413 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2414 if (!VerifyOops) return; 2415 // the VM documentation for the astore[_wide] bytecode allows 2416 // the TOS to be not only an oop but also a return address 2417 Label test; 2418 Label skip; 2419 // See if it is an address (in the current method): 2420 2421 mov(reg, Rtmp); 2422 const int log2_bytecode_size_limit = 16; 2423 srl(Rtmp, log2_bytecode_size_limit, Rtmp); 2424 br_notnull( Rtmp, false, pt, test ); 2425 delayed()->nop(); 2426 2427 // %%% should use call_VM_leaf here? 2428 save_frame_and_mov(0, Lmethod, O0, reg, O1); 2429 save_thread(L7_thread_cache); 2430 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none); 2431 delayed()->nop(); 2432 restore_thread(L7_thread_cache); 2433 br_notnull( O0, false, pt, skip ); 2434 delayed()->restore(); 2435 2436 // Perform a more elaborate out-of-line call 2437 // Not an address; verify it: 2438 bind(test); 2439 verify_oop(reg); 2440 bind(skip); 2441 } 2442 2443 2444 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2445 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); 2446 } 2447 #endif /* CC_INTERP */ 2448 2449 // Inline assembly for: 2450 // 2451 // if (thread is in interp_only_mode) { 2452 // InterpreterRuntime::post_method_entry(); 2453 // } 2454 // if (DTraceMethodProbes) { 2455 // SharedRuntime::dtrace_method_entry(method, receiver); 2456 // } 2457 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 2458 // SharedRuntime::rc_trace_method_entry(method, receiver); 2459 // } 2460 2461 void InterpreterMacroAssembler::notify_method_entry() { 2462 2463 // C++ interpreter only uses this for native methods. 2464 2465 // Whenever JVMTI puts a thread in interp_only_mode, method 2466 // entry/exit events are sent for that thread to track stack 2467 // depth. If it is possible to enter interp_only_mode we add 2468 // the code to check if the event should be sent. 2469 if (JvmtiExport::can_post_interpreter_events()) { 2470 Label L; 2471 Register temp_reg = O5; 2472 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2473 ld(interp_only, temp_reg); 2474 tst(temp_reg); 2475 br(zero, false, pt, L); 2476 delayed()->nop(); 2477 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); 2478 bind(L); 2479 } 2480 2481 { 2482 Register temp_reg = O5; 2483 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2484 call_VM_leaf(noreg, 2485 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 2486 G2_thread, Lmethod); 2487 } 2488 2489 // RedefineClasses() tracing support for obsolete method entry 2490 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 2491 call_VM_leaf(noreg, 2492 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 2493 G2_thread, Lmethod); 2494 } 2495 } 2496 2497 2498 // Inline assembly for: 2499 // 2500 // if (thread is in interp_only_mode) { 2501 // // save result 2502 // InterpreterRuntime::post_method_exit(); 2503 // // restore result 2504 // } 2505 // if (DTraceMethodProbes) { 2506 // SharedRuntime::dtrace_method_exit(thread, method); 2507 // } 2508 // 2509 // Native methods have their result stored in d_tmp and l_tmp 2510 // Java methods have their result stored in the expression stack 2511 2512 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, 2513 TosState state, 2514 NotifyMethodExitMode mode) { 2515 // C++ interpreter only uses this for native methods. 2516 2517 // Whenever JVMTI puts a thread in interp_only_mode, method 2518 // entry/exit events are sent for that thread to track stack 2519 // depth. If it is possible to enter interp_only_mode we add 2520 // the code to check if the event should be sent. 2521 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2522 Label L; 2523 Register temp_reg = O5; 2524 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2525 ld(interp_only, temp_reg); 2526 tst(temp_reg); 2527 br(zero, false, pt, L); 2528 delayed()->nop(); 2529 2530 // Note: frame::interpreter_frame_result has a dependency on how the 2531 // method result is saved across the call to post_method_exit. For 2532 // native methods it assumes the result registers are saved to 2533 // l_scratch and d_scratch. If this changes then the interpreter_frame_result 2534 // implementation will need to be updated too. 2535 2536 save_return_value(state, is_native_method); 2537 call_VM(noreg, 2538 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2539 restore_return_value(state, is_native_method); 2540 bind(L); 2541 } 2542 2543 { 2544 Register temp_reg = O5; 2545 // Dtrace notification 2546 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2547 save_return_value(state, is_native_method); 2548 call_VM_leaf( 2549 noreg, 2550 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2551 G2_thread, Lmethod); 2552 restore_return_value(state, is_native_method); 2553 } 2554 } 2555 2556 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) { 2557 #ifdef CC_INTERP 2558 // result potentially in O0/O1: save it across calls 2559 stf(FloatRegisterImpl::D, F0, STATE(_native_fresult)); 2560 #ifdef _LP64 2561 stx(O0, STATE(_native_lresult)); 2562 #else 2563 std(O0, STATE(_native_lresult)); 2564 #endif 2565 #else // CC_INTERP 2566 if (is_native_call) { 2567 stf(FloatRegisterImpl::D, F0, d_tmp); 2568 #ifdef _LP64 2569 stx(O0, l_tmp); 2570 #else 2571 std(O0, l_tmp); 2572 #endif 2573 } else { 2574 push(state); 2575 } 2576 #endif // CC_INTERP 2577 } 2578 2579 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) { 2580 #ifdef CC_INTERP 2581 ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0); 2582 #ifdef _LP64 2583 ldx(STATE(_native_lresult), O0); 2584 #else 2585 ldd(STATE(_native_lresult), O0); 2586 #endif 2587 #else // CC_INTERP 2588 if (is_native_call) { 2589 ldf(FloatRegisterImpl::D, d_tmp, F0); 2590 #ifdef _LP64 2591 ldx(l_tmp, O0); 2592 #else 2593 ldd(l_tmp, O0); 2594 #endif 2595 } else { 2596 pop(state); 2597 } 2598 #endif // CC_INTERP 2599 }