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