1 /* 2 * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "interp_masm_sparc.hpp" 27 #include "interpreter/interpreter.hpp" 28 #include "interpreter/interpreterRuntime.hpp" 29 #include "logging/log.hpp" 30 #include "oops/arrayOop.hpp" 31 #include "oops/markOop.hpp" 32 #include "oops/methodData.hpp" 33 #include "oops/method.hpp" 34 #include "oops/methodCounters.hpp" 35 #include "prims/jvmtiExport.hpp" 36 #include "prims/jvmtiThreadState.hpp" 37 #include "runtime/basicLock.hpp" 38 #include "runtime/biasedLocking.hpp" 39 #include "runtime/sharedRuntime.hpp" 40 #include "runtime/thread.inline.hpp" 41 42 #ifndef FAST_DISPATCH 43 #define FAST_DISPATCH 1 44 #endif 45 #undef FAST_DISPATCH 46 47 // Implementation of InterpreterMacroAssembler 48 49 // This file specializes the assember with interpreter-specific macros 50 51 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS); 52 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS); 53 54 void InterpreterMacroAssembler::jump_to_entry(address entry) { 55 assert(entry, "Entry must have been generated by now"); 56 AddressLiteral al(entry); 57 jump_to(al, G3_scratch); 58 delayed()->nop(); 59 } 60 61 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) { 62 // Note: this algorithm is also used by C1's OSR entry sequence. 63 // Any changes should also be applied to CodeEmitter::emit_osr_entry(). 64 assert_different_registers(args_size, locals_size); 65 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted. 66 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words 67 // Use br/mov combination because it works on both V8 and V9 and is 68 // faster. 69 Label skip_move; 70 br(Assembler::negative, true, Assembler::pt, skip_move); 71 delayed()->mov(G0, delta); 72 bind(skip_move); 73 round_to(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned) 74 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes 75 } 76 77 // Dispatch code executed in the prolog of a bytecode which does not do it's 78 // own dispatch. The dispatch address is computed and placed in IdispatchAddress 79 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 80 assert_not_delayed(); 81 #ifdef FAST_DISPATCH 82 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since 83 // they both use I2. 84 assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive"); 85 ldub(Lbcp, bcp_incr, Lbyte_code); // load next bytecode 86 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); 87 // add offset to correct dispatch table 88 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 89 ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr 90 #else 91 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 92 // dispatch table to use 93 AddressLiteral tbl(Interpreter::dispatch_table(state)); 94 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 95 set(tbl, G3_scratch); // compute addr of table 96 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr 97 #endif 98 } 99 100 101 // Dispatch code executed in the epilog of a bytecode which does not do it's 102 // own dispatch. The dispatch address in IdispatchAddress is used for the 103 // dispatch. 104 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 105 assert_not_delayed(); 106 verify_FPU(1, state); 107 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 108 jmp( IdispatchAddress, 0 ); 109 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); 110 else delayed()->nop(); 111 } 112 113 114 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { 115 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) 116 assert_not_delayed(); 117 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 118 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr); 119 } 120 121 122 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) { 123 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) 124 assert_not_delayed(); 125 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 126 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false); 127 } 128 129 130 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 131 // load current bytecode 132 assert_not_delayed(); 133 ldub( Lbcp, 0, Lbyte_code); // load next bytecode 134 dispatch_base(state, table); 135 } 136 137 138 void InterpreterMacroAssembler::call_VM_leaf_base( 139 Register java_thread, 140 address entry_point, 141 int number_of_arguments 142 ) { 143 if (!java_thread->is_valid()) 144 java_thread = L7_thread_cache; 145 // super call 146 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments); 147 } 148 149 150 void InterpreterMacroAssembler::call_VM_base( 151 Register oop_result, 152 Register java_thread, 153 Register last_java_sp, 154 address entry_point, 155 int number_of_arguments, 156 bool check_exception 157 ) { 158 if (!java_thread->is_valid()) 159 java_thread = L7_thread_cache; 160 // See class ThreadInVMfromInterpreter, which assumes that the interpreter 161 // takes responsibility for setting its own thread-state on call-out. 162 // However, ThreadInVMfromInterpreter resets the state to "in_Java". 163 164 //save_bcp(); // save bcp 165 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception); 166 //restore_bcp(); // restore bcp 167 //restore_locals(); // restore locals pointer 168 } 169 170 171 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 172 if (JvmtiExport::can_pop_frame()) { 173 Label L; 174 175 // Check the "pending popframe condition" flag in the current thread 176 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg); 177 178 // Initiate popframe handling only if it is not already being processed. If the flag 179 // has the popframe_processing bit set, it means that this code is called *during* popframe 180 // handling - we don't want to reenter. 181 btst(JavaThread::popframe_pending_bit, scratch_reg); 182 br(zero, false, pt, L); 183 delayed()->nop(); 184 btst(JavaThread::popframe_processing_bit, scratch_reg); 185 br(notZero, false, pt, L); 186 delayed()->nop(); 187 188 // Call Interpreter::remove_activation_preserving_args_entry() to get the 189 // address of the same-named entrypoint in the generated interpreter code. 190 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 191 192 // Jump to Interpreter::_remove_activation_preserving_args_entry 193 jmpl(O0, G0, G0); 194 delayed()->nop(); 195 bind(L); 196 } 197 } 198 199 200 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 201 Register thr_state = G4_scratch; 202 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); 203 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset()); 204 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset()); 205 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset()); 206 switch (state) { 207 case ltos: ld_long(val_addr, Otos_l); break; 208 case atos: ld_ptr(oop_addr, Otos_l); 209 st_ptr(G0, oop_addr); break; 210 case btos: // fall through 211 case ztos: // fall through 212 case ctos: // fall through 213 case stos: // fall through 214 case itos: ld(val_addr, Otos_l1); break; 215 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break; 216 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break; 217 case vtos: /* nothing to do */ break; 218 default : ShouldNotReachHere(); 219 } 220 // Clean up tos value in the jvmti thread state 221 or3(G0, ilgl, G3_scratch); 222 stw(G3_scratch, tos_addr); 223 st_long(G0, val_addr); 224 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 225 } 226 227 228 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 229 if (JvmtiExport::can_force_early_return()) { 230 Label L; 231 Register thr_state = G3_scratch; 232 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); 233 br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit; 234 235 // Initiate earlyret handling only if it is not already being processed. 236 // If the flag has the earlyret_processing bit set, it means that this code 237 // is called *during* earlyret handling - we don't want to reenter. 238 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch); 239 cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L); 240 241 // Call Interpreter::remove_activation_early_entry() to get the address of the 242 // same-named entrypoint in the generated interpreter code 243 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1); 244 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1); 245 246 // Jump to Interpreter::_remove_activation_early_entry 247 jmpl(O0, G0, G0); 248 delayed()->nop(); 249 bind(L); 250 } 251 } 252 253 254 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) { 255 mov(arg_1, O0); 256 mov(arg_2, O1); 257 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2); 258 } 259 260 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) { 261 assert_not_delayed(); 262 dispatch_Lbyte_code(state, table); 263 } 264 265 266 void InterpreterMacroAssembler::dispatch_normal(TosState state) { 267 dispatch_base(state, Interpreter::normal_table(state)); 268 } 269 270 271 void InterpreterMacroAssembler::dispatch_only(TosState state) { 272 dispatch_base(state, Interpreter::dispatch_table(state)); 273 } 274 275 276 // common code to dispatch and dispatch_only 277 // dispatch value in Lbyte_code and increment Lbcp 278 279 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) { 280 verify_FPU(1, state); 281 // %%%%% maybe implement +VerifyActivationFrameSize here 282 //verify_thread(); //too slow; we will just verify on method entry & exit 283 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 284 #ifdef FAST_DISPATCH 285 if (table == Interpreter::dispatch_table(state)) { 286 // use IdispatchTables 287 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); 288 // add offset to correct dispatch table 289 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 290 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // get entry addr 291 } else { 292 #endif 293 // dispatch table to use 294 AddressLiteral tbl(table); 295 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 296 set(tbl, G3_scratch); // compute addr of table 297 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr 298 #ifdef FAST_DISPATCH 299 } 300 #endif 301 jmp( G3_scratch, 0 ); 302 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); 303 else delayed()->nop(); 304 } 305 306 307 // Helpers for expression stack 308 309 // Longs and doubles are Category 2 computational types in the 310 // JVM specification (section 3.11.1) and take 2 expression stack or 311 // local slots. 312 // Aligning them on 32 bit with tagged stacks is hard because the code generated 313 // for the dup* bytecodes depends on what types are already on the stack. 314 // If the types are split into the two stack/local slots, that is much easier 315 // (and we can use 0 for non-reference tags). 316 317 // Known good alignment in _LP64 but unknown otherwise 318 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) { 319 assert_not_delayed(); 320 321 ldf(FloatRegisterImpl::D, r1, offset, d); 322 } 323 324 // Known good alignment in _LP64 but unknown otherwise 325 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) { 326 assert_not_delayed(); 327 328 stf(FloatRegisterImpl::D, d, r1, offset); 329 // store something more useful here 330 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);) 331 } 332 333 334 // Known good alignment in _LP64 but unknown otherwise 335 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) { 336 assert_not_delayed(); 337 ldx(r1, offset, rd); 338 } 339 340 // Known good alignment in _LP64 but unknown otherwise 341 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) { 342 assert_not_delayed(); 343 344 stx(l, r1, offset); 345 // store something more useful here 346 stx(G0, r1, offset+Interpreter::stackElementSize); 347 } 348 349 void InterpreterMacroAssembler::pop_i(Register r) { 350 assert_not_delayed(); 351 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r); 352 inc(Lesp, Interpreter::stackElementSize); 353 debug_only(verify_esp(Lesp)); 354 } 355 356 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) { 357 assert_not_delayed(); 358 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); 359 inc(Lesp, Interpreter::stackElementSize); 360 debug_only(verify_esp(Lesp)); 361 } 362 363 void InterpreterMacroAssembler::pop_l(Register r) { 364 assert_not_delayed(); 365 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r); 366 inc(Lesp, 2*Interpreter::stackElementSize); 367 debug_only(verify_esp(Lesp)); 368 } 369 370 371 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) { 372 assert_not_delayed(); 373 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f); 374 inc(Lesp, Interpreter::stackElementSize); 375 debug_only(verify_esp(Lesp)); 376 } 377 378 379 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) { 380 assert_not_delayed(); 381 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f); 382 inc(Lesp, 2*Interpreter::stackElementSize); 383 debug_only(verify_esp(Lesp)); 384 } 385 386 387 void InterpreterMacroAssembler::push_i(Register r) { 388 assert_not_delayed(); 389 debug_only(verify_esp(Lesp)); 390 st(r, Lesp, 0); 391 dec(Lesp, Interpreter::stackElementSize); 392 } 393 394 void InterpreterMacroAssembler::push_ptr(Register r) { 395 assert_not_delayed(); 396 st_ptr(r, Lesp, 0); 397 dec(Lesp, Interpreter::stackElementSize); 398 } 399 400 // remember: our convention for longs in SPARC is: 401 // O0 (Otos_l1) has high-order part in first word, 402 // O1 (Otos_l2) has low-order part in second word 403 404 void InterpreterMacroAssembler::push_l(Register r) { 405 assert_not_delayed(); 406 debug_only(verify_esp(Lesp)); 407 // Longs are stored in memory-correct order, even if unaligned. 408 int offset = -Interpreter::stackElementSize; 409 store_unaligned_long(r, Lesp, offset); 410 dec(Lesp, 2 * Interpreter::stackElementSize); 411 } 412 413 414 void InterpreterMacroAssembler::push_f(FloatRegister f) { 415 assert_not_delayed(); 416 debug_only(verify_esp(Lesp)); 417 stf(FloatRegisterImpl::S, f, Lesp, 0); 418 dec(Lesp, Interpreter::stackElementSize); 419 } 420 421 422 void InterpreterMacroAssembler::push_d(FloatRegister d) { 423 assert_not_delayed(); 424 debug_only(verify_esp(Lesp)); 425 // Longs are stored in memory-correct order, even if unaligned. 426 int offset = -Interpreter::stackElementSize; 427 store_unaligned_double(d, Lesp, offset); 428 dec(Lesp, 2 * Interpreter::stackElementSize); 429 } 430 431 432 void InterpreterMacroAssembler::push(TosState state) { 433 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 434 switch (state) { 435 case atos: push_ptr(); break; 436 case btos: // fall through 437 case ztos: // fall through 438 case ctos: // fall through 439 case stos: // fall through 440 case itos: push_i(); break; 441 case ltos: push_l(); break; 442 case ftos: push_f(); break; 443 case dtos: push_d(); break; 444 case vtos: /* nothing to do */ break; 445 default : ShouldNotReachHere(); 446 } 447 } 448 449 450 void InterpreterMacroAssembler::pop(TosState state) { 451 switch (state) { 452 case atos: pop_ptr(); break; 453 case btos: // fall through 454 case ztos: // fall through 455 case ctos: // fall through 456 case stos: // fall through 457 case itos: pop_i(); break; 458 case ltos: pop_l(); break; 459 case ftos: pop_f(); break; 460 case dtos: pop_d(); break; 461 case vtos: /* nothing to do */ break; 462 default : ShouldNotReachHere(); 463 } 464 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 465 } 466 467 468 // Helpers for swap and dup 469 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 470 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val); 471 } 472 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 473 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n)); 474 } 475 476 477 void InterpreterMacroAssembler::load_receiver(Register param_count, 478 Register recv) { 479 sll(param_count, Interpreter::logStackElementSize, param_count); 480 ld_ptr(Lesp, param_count, recv); // gets receiver oop 481 } 482 483 void InterpreterMacroAssembler::empty_expression_stack() { 484 // Reset Lesp. 485 sub( Lmonitors, wordSize, Lesp ); 486 487 // Reset SP by subtracting more space from Lesp. 488 Label done; 489 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!"); 490 491 // A native does not need to do this, since its callee does not change SP. 492 ld(Lmethod, Method::access_flags_offset(), Gframe_size); // Load access flags. 493 btst(JVM_ACC_NATIVE, Gframe_size); 494 br(Assembler::notZero, false, Assembler::pt, done); 495 delayed()->nop(); 496 497 // Compute max expression stack+register save area 498 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Gframe_size); 499 lduh(Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size); // Load max stack. 500 add(Gframe_size, frame::memory_parameter_word_sp_offset+Method::extra_stack_entries(), Gframe_size ); 501 502 // 503 // now set up a stack frame with the size computed above 504 // 505 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below 506 sll( Gframe_size, LogBytesPerWord, Gframe_size ); 507 sub( Lesp, Gframe_size, Gframe_size ); 508 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary 509 debug_only(verify_sp(Gframe_size, G4_scratch)); 510 sub(Gframe_size, STACK_BIAS, Gframe_size ); 511 mov(Gframe_size, SP); 512 513 bind(done); 514 } 515 516 517 #ifdef ASSERT 518 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) { 519 Label Bad, OK; 520 521 // Saved SP must be aligned. 522 btst(2*BytesPerWord-1, Rsp); 523 br(Assembler::notZero, false, Assembler::pn, Bad); 524 delayed()->nop(); 525 526 // Saved SP, plus register window size, must not be above FP. 527 add(Rsp, frame::register_save_words * wordSize, Rtemp); 528 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP 529 cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad); 530 531 // Saved SP must not be ridiculously below current SP. 532 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K); 533 set(maxstack, Rtemp); 534 sub(SP, Rtemp, Rtemp); 535 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp 536 cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad); 537 538 ba_short(OK); 539 540 bind(Bad); 541 stop("on return to interpreted call, restored SP is corrupted"); 542 543 bind(OK); 544 } 545 546 547 void InterpreterMacroAssembler::verify_esp(Register Resp) { 548 // about to read or write Resp[0] 549 // make sure it is not in the monitors or the register save area 550 Label OK1, OK2; 551 552 cmp(Resp, Lmonitors); 553 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1); 554 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp); 555 stop("too many pops: Lesp points into monitor area"); 556 bind(OK1); 557 sub(Resp, STACK_BIAS, Resp); 558 cmp(Resp, SP); 559 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2); 560 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp); 561 stop("too many pushes: Lesp points into register window"); 562 bind(OK2); 563 } 564 #endif // ASSERT 565 566 // Load compiled (i2c) or interpreter entry when calling from interpreted and 567 // do the call. Centralized so that all interpreter calls will do the same actions. 568 // If jvmti single stepping is on for a thread we must not call compiled code. 569 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) { 570 571 // Assume we want to go compiled if available 572 573 ld_ptr(G5_method, in_bytes(Method::from_interpreted_offset()), target); 574 575 if (JvmtiExport::can_post_interpreter_events()) { 576 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 577 // compiled code in threads for which the event is enabled. Check here for 578 // interp_only_mode if these events CAN be enabled. 579 verify_thread(); 580 Label skip_compiled_code; 581 582 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 583 ld(interp_only, scratch); 584 cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn); 585 delayed()->ld_ptr(G5_method, in_bytes(Method::interpreter_entry_offset()), target); 586 bind(skip_compiled_code); 587 } 588 589 // the i2c_adapters need Method* in G5_method (right? %%%) 590 // do the call 591 #ifdef ASSERT 592 { 593 Label ok; 594 br_notnull_short(target, Assembler::pt, ok); 595 stop("null entry point"); 596 bind(ok); 597 } 598 #endif // ASSERT 599 600 // Adjust Rret first so Llast_SP can be same as Rret 601 add(Rret, -frame::pc_return_offset, O7); 602 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer 603 // Record SP so we can remove any stack space allocated by adapter transition 604 jmp(target, 0); 605 delayed()->mov(SP, Llast_SP); 606 } 607 608 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) { 609 assert_not_delayed(); 610 611 Label not_taken; 612 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken); 613 else br (cc, false, Assembler::pn, not_taken); 614 delayed()->nop(); 615 616 TemplateTable::branch(false,false); 617 618 bind(not_taken); 619 620 profile_not_taken_branch(G3_scratch); 621 } 622 623 624 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp( 625 int bcp_offset, 626 Register Rtmp, 627 Register Rdst, 628 signedOrNot is_signed, 629 setCCOrNot should_set_CC ) { 630 assert(Rtmp != Rdst, "need separate temp register"); 631 assert_not_delayed(); 632 switch (is_signed) { 633 default: ShouldNotReachHere(); 634 635 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte 636 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte 637 } 638 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte 639 sll( Rdst, BitsPerByte, Rdst); 640 switch (should_set_CC ) { 641 default: ShouldNotReachHere(); 642 643 case set_CC: orcc( Rdst, Rtmp, Rdst ); break; 644 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break; 645 } 646 } 647 648 649 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp( 650 int bcp_offset, 651 Register Rtmp, 652 Register Rdst, 653 setCCOrNot should_set_CC ) { 654 assert(Rtmp != Rdst, "need separate temp register"); 655 assert_not_delayed(); 656 add( Lbcp, bcp_offset, Rtmp); 657 andcc( Rtmp, 3, G0); 658 Label aligned; 659 switch (should_set_CC ) { 660 default: ShouldNotReachHere(); 661 662 case set_CC: break; 663 case dont_set_CC: break; 664 } 665 666 br(Assembler::zero, true, Assembler::pn, aligned); 667 delayed()->ldsw(Rtmp, 0, Rdst); 668 669 ldub(Lbcp, bcp_offset + 3, Rdst); 670 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst); 671 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst); 672 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); 673 or3(Rtmp, Rdst, Rdst ); 674 675 bind(aligned); 676 if (should_set_CC == set_CC) tst(Rdst); 677 } 678 679 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register temp, Register index, 680 int bcp_offset, size_t index_size) { 681 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 682 if (index_size == sizeof(u2)) { 683 get_2_byte_integer_at_bcp(bcp_offset, temp, index, Unsigned); 684 } else if (index_size == sizeof(u4)) { 685 get_4_byte_integer_at_bcp(bcp_offset, temp, index); 686 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 687 xor3(index, -1, index); // convert to plain index 688 } else if (index_size == sizeof(u1)) { 689 ldub(Lbcp, bcp_offset, index); 690 } else { 691 ShouldNotReachHere(); 692 } 693 } 694 695 696 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp, 697 int bcp_offset, size_t index_size) { 698 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 699 assert_different_registers(cache, tmp); 700 assert_not_delayed(); 701 get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size); 702 // convert from field index to ConstantPoolCacheEntry index and from 703 // word index to byte offset 704 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); 705 add(LcpoolCache, tmp, cache); 706 } 707 708 709 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, 710 Register temp, 711 Register bytecode, 712 int byte_no, 713 int bcp_offset, 714 size_t index_size) { 715 get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size); 716 ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode); 717 const int shift_count = (1 + byte_no) * BitsPerByte; 718 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) || 719 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift), 720 "correct shift count"); 721 srl(bytecode, shift_count, bytecode); 722 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask"); 723 and3(bytecode, ConstantPoolCacheEntry::bytecode_1_mask, bytecode); 724 } 725 726 727 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp, 728 int bcp_offset, size_t index_size) { 729 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 730 assert_different_registers(cache, tmp); 731 assert_not_delayed(); 732 if (index_size == sizeof(u2)) { 733 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); 734 } else { 735 ShouldNotReachHere(); // other sizes not supported here 736 } 737 // convert from field index to ConstantPoolCacheEntry index 738 // and from word index to byte offset 739 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); 740 // skip past the header 741 add(tmp, in_bytes(ConstantPoolCache::base_offset()), tmp); 742 // construct pointer to cache entry 743 add(LcpoolCache, tmp, cache); 744 } 745 746 747 // Load object from cpool->resolved_references(index) 748 void InterpreterMacroAssembler::load_resolved_reference_at_index( 749 Register result, Register index) { 750 assert_different_registers(result, index); 751 assert_not_delayed(); 752 // convert from field index to resolved_references() index and from 753 // word index to byte offset. Since this is a java object, it can be compressed 754 Register tmp = index; // reuse 755 sll(index, LogBytesPerHeapOop, tmp); 756 get_constant_pool(result); 757 // load pointer for resolved_references[] objArray 758 ld_ptr(result, ConstantPool::cache_offset_in_bytes(), result); 759 ld_ptr(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result); 760 // JNIHandles::resolve(result) 761 ld_ptr(result, 0, result); 762 // Add in the index 763 add(result, tmp, result); 764 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result); 765 } 766 767 768 // load cpool->resolved_klass_at(index) 769 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool, 770 Register Roffset, Register Rklass) { 771 // int value = *this_cp->int_at_addr(which); 772 // int resolved_klass_index = extract_low_short_from_int(value); 773 // 774 // Because SPARC is big-endian, the low_short is at (cpool->int_at_addr(which) + 2 bytes) 775 add(Roffset, Rcpool, Roffset); 776 lduh(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index 777 778 Register Rresolved_klasses = Rklass; 779 ld_ptr(Rcpool, ConstantPool::resolved_klasses_offset_in_bytes(), Rresolved_klasses); 780 sll(Roffset, LogBytesPerWord, Roffset); 781 add(Roffset, Array<Klass*>::base_offset_in_bytes(), Roffset); 782 ld_ptr(Rresolved_klasses, Roffset, Rklass); 783 } 784 785 786 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 787 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. 788 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 789 Register Rsuper_klass, 790 Register Rtmp1, 791 Register Rtmp2, 792 Register Rtmp3, 793 Label &ok_is_subtype ) { 794 Label not_subtype; 795 796 // Profile the not-null value's klass. 797 profile_typecheck(Rsub_klass, Rtmp1); 798 799 check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass, 800 Rtmp1, Rtmp2, 801 &ok_is_subtype, ¬_subtype, NULL); 802 803 check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass, 804 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg, 805 &ok_is_subtype, NULL); 806 807 bind(not_subtype); 808 profile_typecheck_failed(Rtmp1); 809 } 810 811 // Separate these two to allow for delay slot in middle 812 // These are used to do a test and full jump to exception-throwing code. 813 814 // %%%%% Could possibly reoptimize this by testing to see if could use 815 // a single conditional branch (i.e. if span is small enough. 816 // If you go that route, than get rid of the split and give up 817 // on the delay-slot hack. 818 819 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition, 820 Label& ok ) { 821 assert_not_delayed(); 822 br(ok_condition, true, pt, ok); 823 // DELAY SLOT 824 } 825 826 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition, 827 Label& ok ) { 828 assert_not_delayed(); 829 bp( ok_condition, true, Assembler::xcc, pt, ok); 830 // DELAY SLOT 831 } 832 833 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition, 834 Label& ok ) { 835 assert_not_delayed(); 836 brx(ok_condition, true, pt, ok); 837 // DELAY SLOT 838 } 839 840 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point, 841 Register Rscratch, 842 Label& ok ) { 843 assert(throw_entry_point != NULL, "entry point must be generated by now"); 844 AddressLiteral dest(throw_entry_point); 845 jump_to(dest, Rscratch); 846 delayed()->nop(); 847 bind(ok); 848 } 849 850 851 // And if you cannot use the delay slot, here is a shorthand: 852 853 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition, 854 address throw_entry_point, 855 Register Rscratch ) { 856 Label ok; 857 if (ok_condition != never) { 858 throw_if_not_1_icc( ok_condition, ok); 859 delayed()->nop(); 860 } 861 throw_if_not_2( throw_entry_point, Rscratch, ok); 862 } 863 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition, 864 address throw_entry_point, 865 Register Rscratch ) { 866 Label ok; 867 if (ok_condition != never) { 868 throw_if_not_1_xcc( ok_condition, ok); 869 delayed()->nop(); 870 } 871 throw_if_not_2( throw_entry_point, Rscratch, ok); 872 } 873 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition, 874 address throw_entry_point, 875 Register Rscratch ) { 876 Label ok; 877 if (ok_condition != never) { 878 throw_if_not_1_x( ok_condition, ok); 879 delayed()->nop(); 880 } 881 throw_if_not_2( throw_entry_point, Rscratch, ok); 882 } 883 884 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res 885 // Note: res is still shy of address by array offset into object. 886 887 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) { 888 assert_not_delayed(); 889 890 verify_oop(array); 891 // sign extend since tos (index) can be a 32bit value 892 sra(index, G0, index); 893 894 // check array 895 Label ptr_ok; 896 tst(array); 897 throw_if_not_1_x( notZero, ptr_ok ); 898 delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index 899 throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok); 900 901 Label index_ok; 902 cmp(index, tmp); 903 throw_if_not_1_icc( lessUnsigned, index_ok ); 904 if (index_shift > 0) delayed()->sll(index, index_shift, index); 905 else delayed()->add(array, index, res); // addr - const offset in index 906 // convention: move aberrant index into G3_scratch for exception message 907 mov(index, G3_scratch); 908 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok); 909 910 // add offset if didn't do it in delay slot 911 if (index_shift > 0) add(array, index, res); // addr - const offset in index 912 } 913 914 915 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { 916 assert_not_delayed(); 917 918 // pop array 919 pop_ptr(array); 920 921 // check array 922 index_check_without_pop(array, index, index_shift, tmp, res); 923 } 924 925 926 void InterpreterMacroAssembler::get_const(Register Rdst) { 927 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst); 928 } 929 930 931 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 932 get_const(Rdst); 933 ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 934 } 935 936 937 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 938 get_constant_pool(Rdst); 939 ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 940 } 941 942 943 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 944 get_constant_pool(Rcpool); 945 ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags); 946 } 947 948 949 // unlock if synchronized method 950 // 951 // Unlock the receiver if this is a synchronized method. 952 // Unlock any Java monitors from syncronized blocks. 953 // 954 // If there are locked Java monitors 955 // If throw_monitor_exception 956 // throws IllegalMonitorStateException 957 // Else if install_monitor_exception 958 // installs IllegalMonitorStateException 959 // Else 960 // no error processing 961 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 962 bool throw_monitor_exception, 963 bool install_monitor_exception) { 964 Label unlocked, unlock, no_unlock; 965 966 // get the value of _do_not_unlock_if_synchronized into G1_scratch 967 const Address do_not_unlock_if_synchronized(G2_thread, 968 JavaThread::do_not_unlock_if_synchronized_offset()); 969 ldbool(do_not_unlock_if_synchronized, G1_scratch); 970 stbool(G0, do_not_unlock_if_synchronized); // reset the flag 971 972 // check if synchronized method 973 const Address access_flags(Lmethod, Method::access_flags_offset()); 974 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 975 push(state); // save tos 976 ld(access_flags, G3_scratch); // Load access flags. 977 btst(JVM_ACC_SYNCHRONIZED, G3_scratch); 978 br(zero, false, pt, unlocked); 979 delayed()->nop(); 980 981 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 982 // is set. 983 cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock); 984 delayed()->nop(); 985 986 // BasicObjectLock will be first in list, since this is a synchronized method. However, need 987 // to check that the object has not been unlocked by an explicit monitorexit bytecode. 988 989 //Intel: if (throw_monitor_exception) ... else ... 990 // Entry already unlocked, need to throw exception 991 //... 992 993 // pass top-most monitor elem 994 add( top_most_monitor(), O1 ); 995 996 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch); 997 br_notnull_short(G3_scratch, pt, unlock); 998 999 if (throw_monitor_exception) { 1000 // Entry already unlocked need to throw an exception 1001 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1002 should_not_reach_here(); 1003 } else { 1004 // Monitor already unlocked during a stack unroll. 1005 // If requested, install an illegal_monitor_state_exception. 1006 // Continue with stack unrolling. 1007 if (install_monitor_exception) { 1008 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1009 } 1010 ba_short(unlocked); 1011 } 1012 1013 bind(unlock); 1014 1015 unlock_object(O1); 1016 1017 bind(unlocked); 1018 1019 // I0, I1: Might contain return value 1020 1021 // Check that all monitors are unlocked 1022 { Label loop, exception, entry, restart; 1023 1024 Register Rmptr = O0; 1025 Register Rtemp = O1; 1026 Register Rlimit = Lmonitors; 1027 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 1028 assert( (delta & LongAlignmentMask) == 0, 1029 "sizeof BasicObjectLock must be even number of doublewords"); 1030 1031 #ifdef ASSERT 1032 add(top_most_monitor(), Rmptr, delta); 1033 { Label L; 1034 // ensure that Rmptr starts out above (or at) Rlimit 1035 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); 1036 stop("monitor stack has negative size"); 1037 bind(L); 1038 } 1039 #endif 1040 bind(restart); 1041 ba(entry); 1042 delayed()-> 1043 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry 1044 1045 // Entry is still locked, need to throw exception 1046 bind(exception); 1047 if (throw_monitor_exception) { 1048 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1049 should_not_reach_here(); 1050 } else { 1051 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 1052 // Unlock does not block, so don't have to worry about the frame 1053 unlock_object(Rmptr); 1054 if (install_monitor_exception) { 1055 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1056 } 1057 ba_short(restart); 1058 } 1059 1060 bind(loop); 1061 cmp(Rtemp, G0); // check if current entry is used 1062 brx(Assembler::notEqual, false, pn, exception); 1063 delayed()-> 1064 dec(Rmptr, delta); // otherwise advance to next entry 1065 #ifdef ASSERT 1066 { Label L; 1067 // ensure that Rmptr has not somehow stepped below Rlimit 1068 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); 1069 stop("ran off the end of the monitor stack"); 1070 bind(L); 1071 } 1072 #endif 1073 bind(entry); 1074 cmp(Rmptr, Rlimit); // check if bottom reached 1075 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry 1076 delayed()-> 1077 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp); 1078 } 1079 1080 bind(no_unlock); 1081 pop(state); 1082 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1083 } 1084 1085 void InterpreterMacroAssembler::narrow(Register result) { 1086 1087 ld_ptr(Address(Lmethod, Method::const_offset()), G3_scratch); 1088 ldub(G3_scratch, in_bytes(ConstMethod::result_type_offset()), G3_scratch); 1089 1090 Label notBool, notByte, notChar, done; 1091 1092 // common case first 1093 cmp(G3_scratch, T_INT); 1094 br(Assembler::equal, true, pn, done); 1095 delayed()->nop(); 1096 1097 cmp(G3_scratch, T_BOOLEAN); 1098 br(Assembler::notEqual, true, pn, notBool); 1099 delayed()->cmp(G3_scratch, T_BYTE); 1100 and3(result, 1, result); 1101 ba(done); 1102 delayed()->nop(); 1103 1104 bind(notBool); 1105 // cmp(G3_scratch, T_BYTE); 1106 br(Assembler::notEqual, true, pn, notByte); 1107 delayed()->cmp(G3_scratch, T_CHAR); 1108 sll(result, 24, result); 1109 sra(result, 24, result); 1110 ba(done); 1111 delayed()->nop(); 1112 1113 bind(notByte); 1114 // cmp(G3_scratch, T_CHAR); 1115 sll(result, 16, result); 1116 br(Assembler::notEqual, true, pn, done); 1117 delayed()->sra(result, 16, result); 1118 // sll(result, 16, result); 1119 srl(result, 16, result); 1120 1121 // bind(notChar); 1122 // must be short, instructions already executed in delay slot 1123 // sll(result, 16, result); 1124 // sra(result, 16, result); 1125 1126 bind(done); 1127 } 1128 1129 // remove activation 1130 // 1131 // Unlock the receiver if this is a synchronized method. 1132 // Unlock any Java monitors from syncronized blocks. 1133 // Remove the activation from the stack. 1134 // 1135 // If there are locked Java monitors 1136 // If throw_monitor_exception 1137 // throws IllegalMonitorStateException 1138 // Else if install_monitor_exception 1139 // installs IllegalMonitorStateException 1140 // Else 1141 // no error processing 1142 void InterpreterMacroAssembler::remove_activation(TosState state, 1143 bool throw_monitor_exception, 1144 bool install_monitor_exception) { 1145 1146 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 1147 1148 // save result (push state before jvmti call and pop it afterwards) and notify jvmti 1149 notify_method_exit(false, state, NotifyJVMTI); 1150 1151 if (StackReservedPages > 0) { 1152 // testing if Stack Reserved Area needs to be re-enabled 1153 Label no_reserved_zone_enabling; 1154 ld_ptr(G2_thread, JavaThread::reserved_stack_activation_offset(), G3_scratch); 1155 cmp_and_brx_short(SP, G3_scratch, Assembler::lessUnsigned, Assembler::pt, no_reserved_zone_enabling); 1156 1157 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), G2_thread); 1158 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError), G2_thread); 1159 should_not_reach_here(); 1160 1161 bind(no_reserved_zone_enabling); 1162 } 1163 1164 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1165 verify_thread(); 1166 1167 // return tos 1168 assert(Otos_l1 == Otos_i, "adjust code below"); 1169 switch (state) { 1170 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0 1171 case btos: // fall through 1172 case ztos: // fall through 1173 case ctos: 1174 case stos: // fall through 1175 case atos: // fall through 1176 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0 1177 case ftos: // fall through 1178 case dtos: // fall through 1179 case vtos: /* nothing to do */ break; 1180 default : ShouldNotReachHere(); 1181 } 1182 } 1183 1184 // Lock object 1185 // 1186 // Argument - lock_reg points to the BasicObjectLock to be used for locking, 1187 // it must be initialized with the object to lock 1188 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) { 1189 if (UseHeavyMonitors) { 1190 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1191 } 1192 else { 1193 Register obj_reg = Object; 1194 Register mark_reg = G4_scratch; 1195 Register temp_reg = G1_scratch; 1196 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes()); 1197 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1198 Label done; 1199 1200 Label slow_case; 1201 1202 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg); 1203 1204 // load markOop from object into mark_reg 1205 ld_ptr(mark_addr, mark_reg); 1206 1207 if (UseBiasedLocking) { 1208 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case); 1209 } 1210 1211 // get the address of basicLock on stack that will be stored in the object 1212 // we need a temporary register here as we do not want to clobber lock_reg 1213 // (cas clobbers the destination register) 1214 mov(lock_reg, temp_reg); 1215 // set mark reg to be (markOop of object | UNLOCK_VALUE) 1216 or3(mark_reg, markOopDesc::unlocked_value, mark_reg); 1217 // initialize the box (Must happen before we update the object mark!) 1218 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1219 // compare and exchange object_addr, markOop | 1, stack address of basicLock 1220 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1221 cas_ptr(mark_addr.base(), mark_reg, temp_reg); 1222 1223 // if the compare and exchange succeeded we are done (we saw an unlocked object) 1224 cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done); 1225 1226 // We did not see an unlocked object so try the fast recursive case 1227 1228 // Check if owner is self by comparing the value in the markOop of object 1229 // with the stack pointer 1230 sub(temp_reg, SP, temp_reg); 1231 sub(temp_reg, STACK_BIAS, temp_reg); 1232 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 1233 1234 // Composite "andcc" test: 1235 // (a) %sp -vs- markword proximity check, and, 1236 // (b) verify mark word LSBs == 0 (Stack-locked). 1237 // 1238 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size()) 1239 // Note that the page size used for %sp proximity testing is arbitrary and is 1240 // unrelated to the actual MMU page size. We use a 'logical' page size of 1241 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate 1242 // field of the andcc instruction. 1243 andcc (temp_reg, 0xFFFFF003, G0) ; 1244 1245 // if condition is true we are done and hence we can store 0 in the displaced 1246 // header indicating it is a recursive lock and be done 1247 brx(Assembler::zero, true, Assembler::pt, done); 1248 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1249 1250 // none of the above fast optimizations worked so we have to get into the 1251 // slow case of monitor enter 1252 bind(slow_case); 1253 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1254 1255 bind(done); 1256 } 1257 } 1258 1259 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 1260 // 1261 // Argument - lock_reg points to the BasicObjectLock for lock 1262 // Throw IllegalMonitorException if object is not locked by current thread 1263 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1264 if (UseHeavyMonitors) { 1265 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1266 } else { 1267 Register obj_reg = G3_scratch; 1268 Register mark_reg = G4_scratch; 1269 Register displaced_header_reg = G1_scratch; 1270 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes()); 1271 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1272 Label done; 1273 1274 if (UseBiasedLocking) { 1275 // load the object out of the BasicObjectLock 1276 ld_ptr(lockobj_addr, obj_reg); 1277 biased_locking_exit(mark_addr, mark_reg, done, true); 1278 st_ptr(G0, lockobj_addr); // free entry 1279 } 1280 1281 // Test first if we are in the fast recursive case 1282 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes()); 1283 ld_ptr(lock_addr, displaced_header_reg); 1284 br_null(displaced_header_reg, true, Assembler::pn, done); 1285 delayed()->st_ptr(G0, lockobj_addr); // free entry 1286 1287 // See if it is still a light weight lock, if so we just unlock 1288 // the object and we are done 1289 1290 if (!UseBiasedLocking) { 1291 // load the object out of the BasicObjectLock 1292 ld_ptr(lockobj_addr, obj_reg); 1293 } 1294 1295 // we have the displaced header in displaced_header_reg 1296 // we expect to see the stack address of the basicLock in case the 1297 // lock is still a light weight lock (lock_reg) 1298 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1299 cas_ptr(mark_addr.base(), lock_reg, displaced_header_reg); 1300 cmp(lock_reg, displaced_header_reg); 1301 brx(Assembler::equal, true, Assembler::pn, done); 1302 delayed()->st_ptr(G0, lockobj_addr); // free entry 1303 1304 // The lock has been converted into a heavy lock and hence 1305 // we need to get into the slow case 1306 1307 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1308 1309 bind(done); 1310 } 1311 } 1312 1313 // Get the method data pointer from the Method* and set the 1314 // specified register to its value. 1315 1316 void InterpreterMacroAssembler::set_method_data_pointer() { 1317 assert(ProfileInterpreter, "must be profiling interpreter"); 1318 Label get_continue; 1319 1320 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); 1321 test_method_data_pointer(get_continue); 1322 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); 1323 bind(get_continue); 1324 } 1325 1326 // Set the method data pointer for the current bcp. 1327 1328 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1329 assert(ProfileInterpreter, "must be profiling interpreter"); 1330 Label zero_continue; 1331 1332 // Test MDO to avoid the call if it is NULL. 1333 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); 1334 test_method_data_pointer(zero_continue); 1335 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp); 1336 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); 1337 add(ImethodDataPtr, O0, ImethodDataPtr); 1338 bind(zero_continue); 1339 } 1340 1341 // Test ImethodDataPtr. If it is null, continue at the specified label 1342 1343 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1344 assert(ProfileInterpreter, "must be profiling interpreter"); 1345 br_null_short(ImethodDataPtr, Assembler::pn, zero_continue); 1346 } 1347 1348 void InterpreterMacroAssembler::verify_method_data_pointer() { 1349 assert(ProfileInterpreter, "must be profiling interpreter"); 1350 #ifdef ASSERT 1351 Label verify_continue; 1352 test_method_data_pointer(verify_continue); 1353 1354 // If the mdp is valid, it will point to a DataLayout header which is 1355 // consistent with the bcp. The converse is highly probable also. 1356 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch); 1357 ld_ptr(Lmethod, Method::const_offset(), O5); 1358 add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch); 1359 add(G3_scratch, O5, G3_scratch); 1360 cmp(Lbcp, G3_scratch); 1361 brx(Assembler::equal, false, Assembler::pt, verify_continue); 1362 1363 Register temp_reg = O5; 1364 delayed()->mov(ImethodDataPtr, temp_reg); 1365 // %%% should use call_VM_leaf here? 1366 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr); 1367 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1); 1368 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS); 1369 stf(FloatRegisterImpl::D, Ftos_d, d_save); 1370 mov(temp_reg->after_save(), O2); 1371 save_thread(L7_thread_cache); 1372 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none); 1373 delayed()->nop(); 1374 restore_thread(L7_thread_cache); 1375 ldf(FloatRegisterImpl::D, d_save, Ftos_d); 1376 restore(); 1377 bind(verify_continue); 1378 #endif // ASSERT 1379 } 1380 1381 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1382 Register method_counters, 1383 Register Rtmp, 1384 Label &profile_continue) { 1385 assert(ProfileInterpreter, "must be profiling interpreter"); 1386 // Control will flow to "profile_continue" if the counter is less than the 1387 // limit or if we call profile_method() 1388 1389 Label done; 1390 1391 // if no method data exists, and the counter is high enough, make one 1392 br_notnull_short(ImethodDataPtr, Assembler::pn, done); 1393 1394 // Test to see if we should create a method data oop 1395 Address profile_limit(method_counters, MethodCounters::interpreter_profile_limit_offset()); 1396 ld(profile_limit, Rtmp); 1397 cmp(invocation_count, Rtmp); 1398 // Use long branches because call_VM() code and following code generated by 1399 // test_backedge_count_for_osr() is large in debug VM. 1400 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue); 1401 delayed()->nop(); 1402 1403 // Build it now. 1404 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1405 set_method_data_pointer_for_bcp(); 1406 ba(profile_continue); 1407 delayed()->nop(); 1408 bind(done); 1409 } 1410 1411 // Store a value at some constant offset from the method data pointer. 1412 1413 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1414 assert(ProfileInterpreter, "must be profiling interpreter"); 1415 st_ptr(value, ImethodDataPtr, constant); 1416 } 1417 1418 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter, 1419 Register bumped_count, 1420 bool decrement) { 1421 assert(ProfileInterpreter, "must be profiling interpreter"); 1422 1423 // Load the counter. 1424 ld_ptr(counter, bumped_count); 1425 1426 if (decrement) { 1427 // Decrement the register. Set condition codes. 1428 subcc(bumped_count, DataLayout::counter_increment, bumped_count); 1429 1430 // If the decrement causes the counter to overflow, stay negative 1431 Label L; 1432 brx(Assembler::negative, true, Assembler::pn, L); 1433 1434 // Store the decremented counter, if it is still negative. 1435 delayed()->st_ptr(bumped_count, counter); 1436 bind(L); 1437 } else { 1438 // Increment the register. Set carry flag. 1439 addcc(bumped_count, DataLayout::counter_increment, bumped_count); 1440 1441 // If the increment causes the counter to overflow, pull back by 1. 1442 assert(DataLayout::counter_increment == 1, "subc works"); 1443 subc(bumped_count, G0, bumped_count); 1444 1445 // Store the incremented counter. 1446 st_ptr(bumped_count, counter); 1447 } 1448 } 1449 1450 // Increment the value at some constant offset from the method data pointer. 1451 1452 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1453 Register bumped_count, 1454 bool decrement) { 1455 // Locate the counter at a fixed offset from the mdp: 1456 Address counter(ImethodDataPtr, constant); 1457 increment_mdp_data_at(counter, bumped_count, decrement); 1458 } 1459 1460 // Increment the value at some non-fixed (reg + constant) offset from 1461 // the method data pointer. 1462 1463 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1464 int constant, 1465 Register bumped_count, 1466 Register scratch2, 1467 bool decrement) { 1468 // Add the constant to reg to get the offset. 1469 add(ImethodDataPtr, reg, scratch2); 1470 Address counter(scratch2, constant); 1471 increment_mdp_data_at(counter, bumped_count, decrement); 1472 } 1473 1474 // Set a flag value at the current method data pointer position. 1475 // Updates a single byte of the header, to avoid races with other header bits. 1476 1477 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1478 Register scratch) { 1479 assert(ProfileInterpreter, "must be profiling interpreter"); 1480 // Load the data header 1481 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch); 1482 1483 // Set the flag 1484 or3(scratch, flag_constant, scratch); 1485 1486 // Store the modified header. 1487 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset())); 1488 } 1489 1490 // Test the location at some offset from the method data pointer. 1491 // If it is not equal to value, branch to the not_equal_continue Label. 1492 // Set condition codes to match the nullness of the loaded value. 1493 1494 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1495 Register value, 1496 Label& not_equal_continue, 1497 Register scratch) { 1498 assert(ProfileInterpreter, "must be profiling interpreter"); 1499 ld_ptr(ImethodDataPtr, offset, scratch); 1500 cmp(value, scratch); 1501 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue); 1502 delayed()->tst(scratch); 1503 } 1504 1505 // Update the method data pointer by the displacement located at some fixed 1506 // offset from the method data pointer. 1507 1508 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1509 Register scratch) { 1510 assert(ProfileInterpreter, "must be profiling interpreter"); 1511 ld_ptr(ImethodDataPtr, offset_of_disp, scratch); 1512 add(ImethodDataPtr, scratch, ImethodDataPtr); 1513 } 1514 1515 // Update the method data pointer by the displacement located at the 1516 // offset (reg + offset_of_disp). 1517 1518 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1519 int offset_of_disp, 1520 Register scratch) { 1521 assert(ProfileInterpreter, "must be profiling interpreter"); 1522 add(reg, offset_of_disp, scratch); 1523 ld_ptr(ImethodDataPtr, scratch, scratch); 1524 add(ImethodDataPtr, scratch, ImethodDataPtr); 1525 } 1526 1527 // Update the method data pointer by a simple constant displacement. 1528 1529 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1530 assert(ProfileInterpreter, "must be profiling interpreter"); 1531 add(ImethodDataPtr, constant, ImethodDataPtr); 1532 } 1533 1534 // Update the method data pointer for a _ret bytecode whose target 1535 // was not among our cached targets. 1536 1537 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1538 Register return_bci) { 1539 assert(ProfileInterpreter, "must be profiling interpreter"); 1540 push(state); 1541 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile 1542 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1543 ld_ptr(l_tmp, return_bci); 1544 pop(state); 1545 } 1546 1547 // Count a taken branch in the bytecodes. 1548 1549 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1550 if (ProfileInterpreter) { 1551 Label profile_continue; 1552 1553 // If no method data exists, go to profile_continue. 1554 test_method_data_pointer(profile_continue); 1555 1556 // We are taking a branch. Increment the taken count. 1557 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count); 1558 1559 // The method data pointer needs to be updated to reflect the new target. 1560 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1561 bind (profile_continue); 1562 } 1563 } 1564 1565 1566 // Count a not-taken branch in the bytecodes. 1567 1568 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) { 1569 if (ProfileInterpreter) { 1570 Label profile_continue; 1571 1572 // If no method data exists, go to profile_continue. 1573 test_method_data_pointer(profile_continue); 1574 1575 // We are taking a branch. Increment the not taken count. 1576 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch); 1577 1578 // The method data pointer needs to be updated to correspond to the 1579 // next bytecode. 1580 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1581 bind (profile_continue); 1582 } 1583 } 1584 1585 1586 // Count a non-virtual call in the bytecodes. 1587 1588 void InterpreterMacroAssembler::profile_call(Register scratch) { 1589 if (ProfileInterpreter) { 1590 Label profile_continue; 1591 1592 // If no method data exists, go to profile_continue. 1593 test_method_data_pointer(profile_continue); 1594 1595 // We are making a call. Increment the count. 1596 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1597 1598 // The method data pointer needs to be updated to reflect the new target. 1599 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1600 bind (profile_continue); 1601 } 1602 } 1603 1604 1605 // Count a final call in the bytecodes. 1606 1607 void InterpreterMacroAssembler::profile_final_call(Register scratch) { 1608 if (ProfileInterpreter) { 1609 Label profile_continue; 1610 1611 // If no method data exists, go to profile_continue. 1612 test_method_data_pointer(profile_continue); 1613 1614 // We are making a call. Increment the count. 1615 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1616 1617 // The method data pointer needs to be updated to reflect the new target. 1618 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1619 bind (profile_continue); 1620 } 1621 } 1622 1623 1624 // Count a virtual call in the bytecodes. 1625 1626 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1627 Register scratch, 1628 bool receiver_can_be_null) { 1629 if (ProfileInterpreter) { 1630 Label profile_continue; 1631 1632 // If no method data exists, go to profile_continue. 1633 test_method_data_pointer(profile_continue); 1634 1635 1636 Label skip_receiver_profile; 1637 if (receiver_can_be_null) { 1638 Label not_null; 1639 br_notnull_short(receiver, Assembler::pt, not_null); 1640 // We are making a call. Increment the count for null receiver. 1641 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1642 ba_short(skip_receiver_profile); 1643 bind(not_null); 1644 } 1645 1646 // Record the receiver type. 1647 record_klass_in_profile(receiver, scratch, true); 1648 bind(skip_receiver_profile); 1649 1650 // The method data pointer needs to be updated to reflect the new target. 1651 #if INCLUDE_JVMCI 1652 if (MethodProfileWidth == 0) { 1653 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1654 } 1655 #else 1656 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1657 #endif 1658 bind(profile_continue); 1659 } 1660 } 1661 1662 #if INCLUDE_JVMCI 1663 void InterpreterMacroAssembler::profile_called_method(Register method, Register scratch) { 1664 assert_different_registers(method, scratch); 1665 if (ProfileInterpreter && MethodProfileWidth > 0) { 1666 Label profile_continue; 1667 1668 // If no method data exists, go to profile_continue. 1669 test_method_data_pointer(profile_continue); 1670 1671 Label done; 1672 record_item_in_profile_helper(method, scratch, 0, done, MethodProfileWidth, 1673 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset())); 1674 bind(done); 1675 1676 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1677 bind(profile_continue); 1678 } 1679 } 1680 #endif // INCLUDE_JVMCI 1681 1682 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register scratch, 1683 Label& done, bool is_virtual_call) { 1684 if (TypeProfileWidth == 0) { 1685 if (is_virtual_call) { 1686 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1687 } 1688 #if INCLUDE_JVMCI 1689 else if (EnableJVMCI) { 1690 increment_mdp_data_at(in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()), scratch); 1691 } 1692 #endif 1693 } else { 1694 int non_profiled_offset = -1; 1695 if (is_virtual_call) { 1696 non_profiled_offset = in_bytes(CounterData::count_offset()); 1697 } 1698 #if INCLUDE_JVMCI 1699 else if (EnableJVMCI) { 1700 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()); 1701 } 1702 #endif 1703 1704 record_item_in_profile_helper(receiver, scratch, 0, done, TypeProfileWidth, 1705 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset); 1706 } 1707 } 1708 1709 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, 1710 Register scratch, int start_row, Label& done, int total_rows, 1711 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, 1712 int non_profiled_offset) { 1713 int last_row = total_rows - 1; 1714 assert(start_row <= last_row, "must be work left to do"); 1715 // Test this row for both the item and for null. 1716 // Take any of three different outcomes: 1717 // 1. found item => increment count and goto done 1718 // 2. found null => keep looking for case 1, maybe allocate this cell 1719 // 3. found something else => keep looking for cases 1 and 2 1720 // Case 3 is handled by a recursive call. 1721 for (int row = start_row; row <= last_row; row++) { 1722 Label next_test; 1723 bool test_for_null_also = (row == start_row); 1724 1725 // See if the item is item[n]. 1726 int item_offset = in_bytes(item_offset_fn(row)); 1727 test_mdp_data_at(item_offset, item, next_test, scratch); 1728 // delayed()->tst(scratch); 1729 1730 // The receiver is item[n]. Increment count[n]. 1731 int count_offset = in_bytes(item_count_offset_fn(row)); 1732 increment_mdp_data_at(count_offset, scratch); 1733 ba_short(done); 1734 bind(next_test); 1735 1736 if (test_for_null_also) { 1737 Label found_null; 1738 // Failed the equality check on item[n]... Test for null. 1739 if (start_row == last_row) { 1740 // The only thing left to do is handle the null case. 1741 if (non_profiled_offset >= 0) { 1742 brx(Assembler::zero, false, Assembler::pn, found_null); 1743 delayed()->nop(); 1744 // Item did not match any saved item and there is no empty row for it. 1745 // Increment total counter to indicate polymorphic case. 1746 increment_mdp_data_at(non_profiled_offset, scratch); 1747 ba_short(done); 1748 bind(found_null); 1749 } else { 1750 brx(Assembler::notZero, false, Assembler::pt, done); 1751 delayed()->nop(); 1752 } 1753 break; 1754 } 1755 // Since null is rare, make it be the branch-taken case. 1756 brx(Assembler::zero, false, Assembler::pn, found_null); 1757 delayed()->nop(); 1758 1759 // Put all the "Case 3" tests here. 1760 record_item_in_profile_helper(item, scratch, start_row + 1, done, total_rows, 1761 item_offset_fn, item_count_offset_fn, non_profiled_offset); 1762 1763 // Found a null. Keep searching for a matching item, 1764 // but remember that this is an empty (unused) slot. 1765 bind(found_null); 1766 } 1767 } 1768 1769 // In the fall-through case, we found no matching item, but we 1770 // observed the item[start_row] is NULL. 1771 1772 // Fill in the item field and increment the count. 1773 int item_offset = in_bytes(item_offset_fn(start_row)); 1774 set_mdp_data_at(item_offset, item); 1775 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1776 mov(DataLayout::counter_increment, scratch); 1777 set_mdp_data_at(count_offset, scratch); 1778 if (start_row > 0) { 1779 ba_short(done); 1780 } 1781 } 1782 1783 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1784 Register scratch, bool is_virtual_call) { 1785 assert(ProfileInterpreter, "must be profiling"); 1786 Label done; 1787 1788 record_klass_in_profile_helper(receiver, scratch, done, is_virtual_call); 1789 1790 bind (done); 1791 } 1792 1793 1794 // Count a ret in the bytecodes. 1795 1796 void InterpreterMacroAssembler::profile_ret(TosState state, 1797 Register return_bci, 1798 Register scratch) { 1799 if (ProfileInterpreter) { 1800 Label profile_continue; 1801 uint row; 1802 1803 // If no method data exists, go to profile_continue. 1804 test_method_data_pointer(profile_continue); 1805 1806 // Update the total ret count. 1807 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1808 1809 for (row = 0; row < RetData::row_limit(); row++) { 1810 Label next_test; 1811 1812 // See if return_bci is equal to bci[n]: 1813 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), 1814 return_bci, next_test, scratch); 1815 1816 // return_bci is equal to bci[n]. Increment the count. 1817 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch); 1818 1819 // The method data pointer needs to be updated to reflect the new target. 1820 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch); 1821 ba_short(profile_continue); 1822 bind(next_test); 1823 } 1824 1825 update_mdp_for_ret(state, return_bci); 1826 1827 bind (profile_continue); 1828 } 1829 } 1830 1831 // Profile an unexpected null in the bytecodes. 1832 void InterpreterMacroAssembler::profile_null_seen(Register scratch) { 1833 if (ProfileInterpreter) { 1834 Label profile_continue; 1835 1836 // If no method data exists, go to profile_continue. 1837 test_method_data_pointer(profile_continue); 1838 1839 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch); 1840 1841 // The method data pointer needs to be updated. 1842 int mdp_delta = in_bytes(BitData::bit_data_size()); 1843 if (TypeProfileCasts) { 1844 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size()); 1845 } 1846 update_mdp_by_constant(mdp_delta); 1847 1848 bind (profile_continue); 1849 } 1850 } 1851 1852 void InterpreterMacroAssembler::profile_typecheck(Register klass, 1853 Register scratch) { 1854 if (ProfileInterpreter) { 1855 Label profile_continue; 1856 1857 // If no method data exists, go to profile_continue. 1858 test_method_data_pointer(profile_continue); 1859 1860 int mdp_delta = in_bytes(BitData::bit_data_size()); 1861 if (TypeProfileCasts) { 1862 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size()); 1863 1864 // Record the object type. 1865 record_klass_in_profile(klass, scratch, false); 1866 } 1867 1868 // The method data pointer needs to be updated. 1869 update_mdp_by_constant(mdp_delta); 1870 1871 bind (profile_continue); 1872 } 1873 } 1874 1875 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) { 1876 if (ProfileInterpreter && TypeProfileCasts) { 1877 Label profile_continue; 1878 1879 // If no method data exists, go to profile_continue. 1880 test_method_data_pointer(profile_continue); 1881 1882 int count_offset = in_bytes(CounterData::count_offset()); 1883 // Back up the address, since we have already bumped the mdp. 1884 count_offset -= in_bytes(ReceiverTypeData::receiver_type_data_size()); 1885 1886 // *Decrement* the counter. We expect to see zero or small negatives. 1887 increment_mdp_data_at(count_offset, scratch, true); 1888 1889 bind (profile_continue); 1890 } 1891 } 1892 1893 // Count the default case of a switch construct. 1894 1895 void InterpreterMacroAssembler::profile_switch_default(Register scratch) { 1896 if (ProfileInterpreter) { 1897 Label profile_continue; 1898 1899 // If no method data exists, go to profile_continue. 1900 test_method_data_pointer(profile_continue); 1901 1902 // Update the default case count 1903 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1904 scratch); 1905 1906 // The method data pointer needs to be updated. 1907 update_mdp_by_offset( 1908 in_bytes(MultiBranchData::default_displacement_offset()), 1909 scratch); 1910 1911 bind (profile_continue); 1912 } 1913 } 1914 1915 // Count the index'th case of a switch construct. 1916 1917 void InterpreterMacroAssembler::profile_switch_case(Register index, 1918 Register scratch, 1919 Register scratch2, 1920 Register scratch3) { 1921 if (ProfileInterpreter) { 1922 Label profile_continue; 1923 1924 // If no method data exists, go to profile_continue. 1925 test_method_data_pointer(profile_continue); 1926 1927 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() 1928 set(in_bytes(MultiBranchData::per_case_size()), scratch); 1929 smul(index, scratch, scratch); 1930 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch); 1931 1932 // Update the case count 1933 increment_mdp_data_at(scratch, 1934 in_bytes(MultiBranchData::relative_count_offset()), 1935 scratch2, 1936 scratch3); 1937 1938 // The method data pointer needs to be updated. 1939 update_mdp_by_offset(scratch, 1940 in_bytes(MultiBranchData::relative_displacement_offset()), 1941 scratch2); 1942 1943 bind (profile_continue); 1944 } 1945 } 1946 1947 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) { 1948 Label not_null, do_nothing, do_update; 1949 1950 assert_different_registers(obj, mdo_addr.base(), tmp); 1951 1952 verify_oop(obj); 1953 1954 ld_ptr(mdo_addr, tmp); 1955 1956 br_notnull_short(obj, pt, not_null); 1957 or3(tmp, TypeEntries::null_seen, tmp); 1958 ba_short(do_update); 1959 1960 bind(not_null); 1961 load_klass(obj, obj); 1962 1963 xor3(obj, tmp, obj); 1964 btst(TypeEntries::type_klass_mask, obj); 1965 // klass seen before, nothing to do. The unknown bit may have been 1966 // set already but no need to check. 1967 brx(zero, false, pt, do_nothing); 1968 delayed()-> 1969 1970 btst(TypeEntries::type_unknown, obj); 1971 // already unknown. Nothing to do anymore. 1972 brx(notZero, false, pt, do_nothing); 1973 delayed()-> 1974 1975 btst(TypeEntries::type_mask, tmp); 1976 brx(zero, true, pt, do_update); 1977 // first time here. Set profile type. 1978 delayed()->or3(tmp, obj, tmp); 1979 1980 // different than before. Cannot keep accurate profile. 1981 or3(tmp, TypeEntries::type_unknown, tmp); 1982 1983 bind(do_update); 1984 // update profile 1985 st_ptr(tmp, mdo_addr); 1986 1987 bind(do_nothing); 1988 } 1989 1990 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) { 1991 if (!ProfileInterpreter) { 1992 return; 1993 } 1994 1995 assert_different_registers(callee, tmp1, tmp2, ImethodDataPtr); 1996 1997 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1998 Label profile_continue; 1999 2000 test_method_data_pointer(profile_continue); 2001 2002 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 2003 2004 ldub(ImethodDataPtr, in_bytes(DataLayout::tag_offset()) - off_to_start, tmp1); 2005 cmp_and_br_short(tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag, notEqual, pn, profile_continue); 2006 2007 if (MethodData::profile_arguments()) { 2008 Label done; 2009 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 2010 add(ImethodDataPtr, off_to_args, ImethodDataPtr); 2011 2012 for (int i = 0; i < TypeProfileArgsLimit; i++) { 2013 if (i > 0 || MethodData::profile_return()) { 2014 // If return value type is profiled we may have no argument to profile 2015 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1); 2016 sub(tmp1, i*TypeStackSlotEntries::per_arg_count(), tmp1); 2017 cmp_and_br_short(tmp1, TypeStackSlotEntries::per_arg_count(), less, pn, done); 2018 } 2019 ld_ptr(Address(callee, Method::const_offset()), tmp1); 2020 lduh(Address(tmp1, ConstMethod::size_of_parameters_offset()), tmp1); 2021 // stack offset o (zero based) from the start of the argument 2022 // list, for n arguments translates into offset n - o - 1 from 2023 // the end of the argument list. But there's an extra slot at 2024 // the stop of the stack. So the offset is n - o from Lesp. 2025 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, tmp2); 2026 sub(tmp1, tmp2, tmp1); 2027 2028 // Can't use MacroAssembler::argument_address() which needs Gargs to be set up 2029 sll(tmp1, Interpreter::logStackElementSize, tmp1); 2030 ld_ptr(Lesp, tmp1, tmp1); 2031 2032 Address mdo_arg_addr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); 2033 profile_obj_type(tmp1, mdo_arg_addr, tmp2); 2034 2035 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 2036 add(ImethodDataPtr, to_add, ImethodDataPtr); 2037 off_to_args += to_add; 2038 } 2039 2040 if (MethodData::profile_return()) { 2041 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1); 2042 sub(tmp1, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count(), tmp1); 2043 } 2044 2045 bind(done); 2046 2047 if (MethodData::profile_return()) { 2048 // We're right after the type profile for the last 2049 // argument. tmp1 is the number of cells left in the 2050 // CallTypeData/VirtualCallTypeData to reach its end. Non null 2051 // if there's a return to profile. 2052 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 2053 sll(tmp1, exact_log2(DataLayout::cell_size), tmp1); 2054 add(ImethodDataPtr, tmp1, ImethodDataPtr); 2055 } 2056 } else { 2057 assert(MethodData::profile_return(), "either profile call args or call ret"); 2058 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 2059 } 2060 2061 // mdp points right after the end of the 2062 // CallTypeData/VirtualCallTypeData, right after the cells for the 2063 // return value type if there's one. 2064 2065 bind(profile_continue); 2066 } 2067 } 2068 2069 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 2070 assert_different_registers(ret, tmp1, tmp2); 2071 if (ProfileInterpreter && MethodData::profile_return()) { 2072 Label profile_continue, done; 2073 2074 test_method_data_pointer(profile_continue); 2075 2076 if (MethodData::profile_return_jsr292_only()) { 2077 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); 2078 2079 // If we don't profile all invoke bytecodes we must make sure 2080 // it's a bytecode we indeed profile. We can't go back to the 2081 // begining of the ProfileData we intend to update to check its 2082 // type because we're right after it and we don't known its 2083 // length. 2084 Label do_profile; 2085 ldub(Lbcp, 0, tmp1); 2086 cmp_and_br_short(tmp1, Bytecodes::_invokedynamic, equal, pn, do_profile); 2087 cmp(tmp1, Bytecodes::_invokehandle); 2088 br(equal, false, pn, do_profile); 2089 delayed()->lduh(Lmethod, Method::intrinsic_id_offset_in_bytes(), tmp1); 2090 cmp_and_br_short(tmp1, vmIntrinsics::_compiledLambdaForm, notEqual, pt, profile_continue); 2091 2092 bind(do_profile); 2093 } 2094 2095 Address mdo_ret_addr(ImethodDataPtr, -in_bytes(ReturnTypeEntry::size())); 2096 mov(ret, tmp1); 2097 profile_obj_type(tmp1, mdo_ret_addr, tmp2); 2098 2099 bind(profile_continue); 2100 } 2101 } 2102 2103 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) { 2104 if (ProfileInterpreter && MethodData::profile_parameters()) { 2105 Label profile_continue, done; 2106 2107 test_method_data_pointer(profile_continue); 2108 2109 // Load the offset of the area within the MDO used for 2110 // parameters. If it's negative we're not profiling any parameters. 2111 lduw(ImethodDataPtr, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), tmp1); 2112 cmp_and_br_short(tmp1, 0, less, pn, profile_continue); 2113 2114 // Compute a pointer to the area for parameters from the offset 2115 // and move the pointer to the slot for the last 2116 // parameters. Collect profiling from last parameter down. 2117 // mdo start + parameters offset + array length - 1 2118 2119 // Pointer to the parameter area in the MDO 2120 Register mdp = tmp1; 2121 add(ImethodDataPtr, tmp1, mdp); 2122 2123 // offset of the current profile entry to update 2124 Register entry_offset = tmp2; 2125 // entry_offset = array len in number of cells 2126 ld_ptr(mdp, ArrayData::array_len_offset(), entry_offset); 2127 2128 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 2129 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 2130 2131 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 2132 sub(entry_offset, TypeStackSlotEntries::per_arg_count() - (off_base / DataLayout::cell_size), entry_offset); 2133 // entry_offset in bytes 2134 sll(entry_offset, exact_log2(DataLayout::cell_size), entry_offset); 2135 2136 Label loop; 2137 bind(loop); 2138 2139 // load offset on the stack from the slot for this parameter 2140 ld_ptr(mdp, entry_offset, tmp3); 2141 sll(tmp3,Interpreter::logStackElementSize, tmp3); 2142 neg(tmp3); 2143 // read the parameter from the local area 2144 ld_ptr(Llocals, tmp3, tmp3); 2145 2146 // make entry_offset now point to the type field for this parameter 2147 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 2148 assert(type_base > off_base, "unexpected"); 2149 add(entry_offset, type_base - off_base, entry_offset); 2150 2151 // profile the parameter 2152 Address arg_type(mdp, entry_offset); 2153 profile_obj_type(tmp3, arg_type, tmp4); 2154 2155 // go to next parameter 2156 sub(entry_offset, TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base), entry_offset); 2157 cmp_and_br_short(entry_offset, off_base, greaterEqual, pt, loop); 2158 2159 bind(profile_continue); 2160 } 2161 } 2162 2163 // add a InterpMonitorElem to stack (see frame_sparc.hpp) 2164 2165 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty, 2166 Register Rtemp, 2167 Register Rtemp2 ) { 2168 2169 Register Rlimit = Lmonitors; 2170 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 2171 assert( (delta & LongAlignmentMask) == 0, 2172 "sizeof BasicObjectLock must be even number of doublewords"); 2173 2174 sub( SP, delta, SP); 2175 sub( Lesp, delta, Lesp); 2176 sub( Lmonitors, delta, Lmonitors); 2177 2178 if (!stack_is_empty) { 2179 2180 // must copy stack contents down 2181 2182 Label start_copying, next; 2183 2184 // untested("monitor stack expansion"); 2185 compute_stack_base(Rtemp); 2186 ba(start_copying); 2187 delayed()->cmp(Rtemp, Rlimit); // done? duplicated below 2188 2189 // note: must copy from low memory upwards 2190 // On entry to loop, 2191 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS) 2192 // Loop mutates Rtemp 2193 2194 bind( next); 2195 2196 st_ptr(Rtemp2, Rtemp, 0); 2197 inc(Rtemp, wordSize); 2198 cmp(Rtemp, Rlimit); // are we done? (duplicated above) 2199 2200 bind( start_copying ); 2201 2202 brx( notEqual, true, pn, next ); 2203 delayed()->ld_ptr( Rtemp, delta, Rtemp2 ); 2204 2205 // done copying stack 2206 } 2207 } 2208 2209 // Locals 2210 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) { 2211 assert_not_delayed(); 2212 sll(index, Interpreter::logStackElementSize, index); 2213 sub(Llocals, index, index); 2214 ld_ptr(index, 0, dst); 2215 // Note: index must hold the effective address--the iinc template uses it 2216 } 2217 2218 // Just like access_local_ptr but the tag is a returnAddress 2219 void InterpreterMacroAssembler::access_local_returnAddress(Register index, 2220 Register dst ) { 2221 assert_not_delayed(); 2222 sll(index, Interpreter::logStackElementSize, index); 2223 sub(Llocals, index, index); 2224 ld_ptr(index, 0, dst); 2225 } 2226 2227 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) { 2228 assert_not_delayed(); 2229 sll(index, Interpreter::logStackElementSize, index); 2230 sub(Llocals, index, index); 2231 ld(index, 0, dst); 2232 // Note: index must hold the effective address--the iinc template uses it 2233 } 2234 2235 2236 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) { 2237 assert_not_delayed(); 2238 sll(index, Interpreter::logStackElementSize, index); 2239 sub(Llocals, index, index); 2240 // First half stored at index n+1 (which grows down from Llocals[n]) 2241 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst); 2242 } 2243 2244 2245 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) { 2246 assert_not_delayed(); 2247 sll(index, Interpreter::logStackElementSize, index); 2248 sub(Llocals, index, index); 2249 ldf(FloatRegisterImpl::S, index, 0, dst); 2250 } 2251 2252 2253 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) { 2254 assert_not_delayed(); 2255 sll(index, Interpreter::logStackElementSize, index); 2256 sub(Llocals, index, index); 2257 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst); 2258 } 2259 2260 2261 #ifdef ASSERT 2262 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) { 2263 Label L; 2264 2265 assert(Rindex != Rscratch, "Registers cannot be same"); 2266 assert(Rindex != Rscratch1, "Registers cannot be same"); 2267 assert(Rlimit != Rscratch, "Registers cannot be same"); 2268 assert(Rlimit != Rscratch1, "Registers cannot be same"); 2269 assert(Rscratch1 != Rscratch, "Registers cannot be same"); 2270 2271 // untested("reg area corruption"); 2272 add(Rindex, offset, Rscratch); 2273 add(Rlimit, 64 + STACK_BIAS, Rscratch1); 2274 cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L); 2275 stop("regsave area is being clobbered"); 2276 bind(L); 2277 } 2278 #endif // ASSERT 2279 2280 2281 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) { 2282 assert_not_delayed(); 2283 sll(index, Interpreter::logStackElementSize, index); 2284 sub(Llocals, index, index); 2285 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);) 2286 st(src, index, 0); 2287 } 2288 2289 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) { 2290 assert_not_delayed(); 2291 sll(index, Interpreter::logStackElementSize, index); 2292 sub(Llocals, index, index); 2293 #ifdef ASSERT 2294 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2295 #endif 2296 st_ptr(src, index, 0); 2297 } 2298 2299 2300 2301 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) { 2302 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n)); 2303 } 2304 2305 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) { 2306 assert_not_delayed(); 2307 sll(index, Interpreter::logStackElementSize, index); 2308 sub(Llocals, index, index); 2309 #ifdef ASSERT 2310 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2311 #endif 2312 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1 2313 } 2314 2315 2316 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) { 2317 assert_not_delayed(); 2318 sll(index, Interpreter::logStackElementSize, index); 2319 sub(Llocals, index, index); 2320 #ifdef ASSERT 2321 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2322 #endif 2323 stf(FloatRegisterImpl::S, src, index, 0); 2324 } 2325 2326 2327 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) { 2328 assert_not_delayed(); 2329 sll(index, Interpreter::logStackElementSize, index); 2330 sub(Llocals, index, index); 2331 #ifdef ASSERT 2332 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2333 #endif 2334 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1)); 2335 } 2336 2337 2338 int InterpreterMacroAssembler::top_most_monitor_byte_offset() { 2339 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 2340 int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong); 2341 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS; 2342 } 2343 2344 2345 Address InterpreterMacroAssembler::top_most_monitor() { 2346 return Address(FP, top_most_monitor_byte_offset()); 2347 } 2348 2349 2350 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) { 2351 add( Lesp, wordSize, Rdest ); 2352 } 2353 2354 void InterpreterMacroAssembler::get_method_counters(Register method, 2355 Register Rcounters, 2356 Label& skip) { 2357 Label has_counters; 2358 Address method_counters(method, in_bytes(Method::method_counters_offset())); 2359 ld_ptr(method_counters, Rcounters); 2360 br_notnull_short(Rcounters, Assembler::pt, has_counters); 2361 call_VM(noreg, CAST_FROM_FN_PTR(address, 2362 InterpreterRuntime::build_method_counters), method); 2363 ld_ptr(method_counters, Rcounters); 2364 br_null(Rcounters, false, Assembler::pn, skip); // No MethodCounters, OutOfMemory 2365 delayed()->nop(); 2366 bind(has_counters); 2367 } 2368 2369 void InterpreterMacroAssembler::increment_invocation_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) { 2370 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 2371 assert_different_registers(Rcounters, Rtmp, Rtmp2); 2372 2373 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() + 2374 InvocationCounter::counter_offset()); 2375 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() + 2376 InvocationCounter::counter_offset()); 2377 int delta = InvocationCounter::count_increment; 2378 2379 // Load each counter in a register 2380 ld( inv_counter, Rtmp ); 2381 ld( be_counter, Rtmp2 ); 2382 2383 assert( is_simm13( delta ), " delta too large."); 2384 2385 // Add the delta to the invocation counter and store the result 2386 add( Rtmp, delta, Rtmp ); 2387 2388 // Mask the backedge counter 2389 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2390 2391 // Store value 2392 st( Rtmp, inv_counter); 2393 2394 // Add invocation counter + backedge counter 2395 add( Rtmp, Rtmp2, Rtmp); 2396 2397 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 2398 } 2399 2400 void InterpreterMacroAssembler::increment_backedge_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) { 2401 assert(UseCompiler, "incrementing must be useful"); 2402 assert_different_registers(Rcounters, Rtmp, Rtmp2); 2403 2404 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() + 2405 InvocationCounter::counter_offset()); 2406 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() + 2407 InvocationCounter::counter_offset()); 2408 2409 int delta = InvocationCounter::count_increment; 2410 // Load each counter in a register 2411 ld( be_counter, Rtmp ); 2412 ld( inv_counter, Rtmp2 ); 2413 2414 // Add the delta to the backedge counter 2415 add( Rtmp, delta, Rtmp ); 2416 2417 // Mask the invocation counter, add to backedge counter 2418 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2419 2420 // and store the result to memory 2421 st( Rtmp, be_counter ); 2422 2423 // Add backedge + invocation counter 2424 add( Rtmp, Rtmp2, Rtmp ); 2425 2426 // Note that this macro must leave backedge_count + invocation_count in Rtmp! 2427 } 2428 2429 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count, 2430 Register method_counters, 2431 Register branch_bcp, 2432 Register Rtmp ) { 2433 Label did_not_overflow; 2434 Label overflow_with_error; 2435 assert_different_registers(backedge_count, Rtmp, branch_bcp); 2436 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 2437 2438 Address limit(method_counters, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())); 2439 ld(limit, Rtmp); 2440 cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow); 2441 2442 // When ProfileInterpreter is on, the backedge_count comes from the 2443 // MethodData*, which value does not get reset on the call to 2444 // frequency_counter_overflow(). To avoid excessive calls to the overflow 2445 // routine while the method is being compiled, add a second test to make sure 2446 // the overflow function is called only once every overflow_frequency. 2447 if (ProfileInterpreter) { 2448 const int overflow_frequency = 1024; 2449 andcc(backedge_count, overflow_frequency-1, Rtmp); 2450 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow); 2451 delayed()->nop(); 2452 } 2453 2454 // overflow in loop, pass branch bytecode 2455 set(6,Rtmp); 2456 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp); 2457 2458 // Was an OSR adapter generated? 2459 // O0 = osr nmethod 2460 br_null_short(O0, Assembler::pn, overflow_with_error); 2461 2462 // Has the nmethod been invalidated already? 2463 ldub(O0, nmethod::state_offset(), O2); 2464 cmp_and_br_short(O2, nmethod::in_use, Assembler::notEqual, Assembler::pn, overflow_with_error); 2465 2466 // migrate the interpreter frame off of the stack 2467 2468 mov(G2_thread, L7); 2469 // save nmethod 2470 mov(O0, L6); 2471 set_last_Java_frame(SP, noreg); 2472 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7); 2473 reset_last_Java_frame(); 2474 mov(L7, G2_thread); 2475 2476 // move OSR nmethod to I1 2477 mov(L6, I1); 2478 2479 // OSR buffer to I0 2480 mov(O0, I0); 2481 2482 // remove the interpreter frame 2483 restore(I5_savedSP, 0, SP); 2484 2485 // Jump to the osr code. 2486 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2); 2487 jmp(O2, G0); 2488 delayed()->nop(); 2489 2490 bind(overflow_with_error); 2491 2492 bind(did_not_overflow); 2493 } 2494 2495 2496 2497 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) { 2498 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); } 2499 } 2500 2501 2502 // local helper function for the verify_oop_or_return_address macro 2503 static bool verify_return_address(Method* m, int bci) { 2504 #ifndef PRODUCT 2505 address pc = (address)(m->constMethod()) 2506 + in_bytes(ConstMethod::codes_offset()) + bci; 2507 // assume it is a valid return address if it is inside m and is preceded by a jsr 2508 if (!m->contains(pc)) return false; 2509 address jsr_pc; 2510 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2511 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2512 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2513 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2514 #endif // PRODUCT 2515 return false; 2516 } 2517 2518 2519 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2520 if (!VerifyOops) return; 2521 // the VM documentation for the astore[_wide] bytecode allows 2522 // the TOS to be not only an oop but also a return address 2523 Label test; 2524 Label skip; 2525 // See if it is an address (in the current method): 2526 2527 mov(reg, Rtmp); 2528 const int log2_bytecode_size_limit = 16; 2529 srl(Rtmp, log2_bytecode_size_limit, Rtmp); 2530 br_notnull_short( Rtmp, pt, test ); 2531 2532 // %%% should use call_VM_leaf here? 2533 save_frame_and_mov(0, Lmethod, O0, reg, O1); 2534 save_thread(L7_thread_cache); 2535 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none); 2536 delayed()->nop(); 2537 restore_thread(L7_thread_cache); 2538 br_notnull( O0, false, pt, skip ); 2539 delayed()->restore(); 2540 2541 // Perform a more elaborate out-of-line call 2542 // Not an address; verify it: 2543 bind(test); 2544 verify_oop(reg); 2545 bind(skip); 2546 } 2547 2548 2549 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2550 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); 2551 } 2552 2553 2554 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 2555 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 2556 int increment, Address mask_addr, 2557 Register scratch1, Register scratch2, 2558 Condition cond, Label *where) { 2559 ld(counter_addr, scratch1); 2560 add(scratch1, increment, scratch1); 2561 ld(mask_addr, scratch2); 2562 andcc(scratch1, scratch2, G0); 2563 br(cond, false, Assembler::pn, *where); 2564 delayed()->st(scratch1, counter_addr); 2565 } 2566 2567 // Inline assembly for: 2568 // 2569 // if (thread is in interp_only_mode) { 2570 // InterpreterRuntime::post_method_entry(); 2571 // } 2572 // if (DTraceMethodProbes) { 2573 // SharedRuntime::dtrace_method_entry(method, receiver); 2574 // } 2575 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 2576 // SharedRuntime::rc_trace_method_entry(method, receiver); 2577 // } 2578 2579 void InterpreterMacroAssembler::notify_method_entry() { 2580 2581 // Whenever JVMTI puts a thread in interp_only_mode, method 2582 // entry/exit events are sent for that thread to track stack 2583 // depth. If it is possible to enter interp_only_mode we add 2584 // the code to check if the event should be sent. 2585 if (JvmtiExport::can_post_interpreter_events()) { 2586 Label L; 2587 Register temp_reg = O5; 2588 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2589 ld(interp_only, temp_reg); 2590 cmp_and_br_short(temp_reg, 0, equal, pt, L); 2591 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); 2592 bind(L); 2593 } 2594 2595 { 2596 Register temp_reg = O5; 2597 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2598 call_VM_leaf(noreg, 2599 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 2600 G2_thread, Lmethod); 2601 } 2602 2603 // RedefineClasses() tracing support for obsolete method entry 2604 if (log_is_enabled(Trace, redefine, class, obsolete)) { 2605 call_VM_leaf(noreg, 2606 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 2607 G2_thread, Lmethod); 2608 } 2609 } 2610 2611 2612 // Inline assembly for: 2613 // 2614 // if (thread is in interp_only_mode) { 2615 // // save result 2616 // InterpreterRuntime::post_method_exit(); 2617 // // restore result 2618 // } 2619 // if (DTraceMethodProbes) { 2620 // SharedRuntime::dtrace_method_exit(thread, method); 2621 // } 2622 // 2623 // Native methods have their result stored in d_tmp and l_tmp 2624 // Java methods have their result stored in the expression stack 2625 2626 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, 2627 TosState state, 2628 NotifyMethodExitMode mode) { 2629 2630 // Whenever JVMTI puts a thread in interp_only_mode, method 2631 // entry/exit events are sent for that thread to track stack 2632 // depth. If it is possible to enter interp_only_mode we add 2633 // the code to check if the event should be sent. 2634 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2635 Label L; 2636 Register temp_reg = O5; 2637 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2638 ld(interp_only, temp_reg); 2639 cmp_and_br_short(temp_reg, 0, equal, pt, L); 2640 2641 // Note: frame::interpreter_frame_result has a dependency on how the 2642 // method result is saved across the call to post_method_exit. For 2643 // native methods it assumes the result registers are saved to 2644 // l_scratch and d_scratch. If this changes then the interpreter_frame_result 2645 // implementation will need to be updated too. 2646 2647 save_return_value(state, is_native_method); 2648 call_VM(noreg, 2649 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2650 restore_return_value(state, is_native_method); 2651 bind(L); 2652 } 2653 2654 { 2655 Register temp_reg = O5; 2656 // Dtrace notification 2657 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2658 save_return_value(state, is_native_method); 2659 call_VM_leaf( 2660 noreg, 2661 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2662 G2_thread, Lmethod); 2663 restore_return_value(state, is_native_method); 2664 } 2665 } 2666 2667 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) { 2668 if (is_native_call) { 2669 stf(FloatRegisterImpl::D, F0, d_tmp); 2670 stx(O0, l_tmp); 2671 } else { 2672 push(state); 2673 } 2674 } 2675 2676 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) { 2677 if (is_native_call) { 2678 ldf(FloatRegisterImpl::D, d_tmp, F0); 2679 ldx(l_tmp, O0); 2680 } else { 2681 pop(state); 2682 } 2683 }