1 /* 2 * Copyright (c) 1997, 2018, 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/frame.inline.hpp" 40 #include "runtime/safepointMechanism.hpp" 41 #include "runtime/sharedRuntime.hpp" 42 #include "runtime/thread.inline.hpp" 43 #include "utilities/align.hpp" 44 45 // Implementation of InterpreterMacroAssembler 46 47 // This file specializes the assember with interpreter-specific macros 48 49 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS); 50 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS); 51 52 void InterpreterMacroAssembler::jump_to_entry(address entry) { 53 assert(entry, "Entry must have been generated by now"); 54 AddressLiteral al(entry); 55 jump_to(al, G3_scratch); 56 delayed()->nop(); 57 } 58 59 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) { 60 // Note: this algorithm is also used by C1's OSR entry sequence. 61 // Any changes should also be applied to CodeEmitter::emit_osr_entry(). 62 assert_different_registers(args_size, locals_size); 63 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted. 64 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words 65 // Use br/mov combination because it works on both V8 and V9 and is 66 // faster. 67 Label skip_move; 68 br(Assembler::negative, true, Assembler::pt, skip_move); 69 delayed()->mov(G0, delta); 70 bind(skip_move); 71 align_up(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned) 72 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes 73 } 74 75 // Dispatch code executed in the prolog of a bytecode which does not do it's 76 // own dispatch. The dispatch address is computed and placed in IdispatchAddress 77 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 78 assert_not_delayed(); 79 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 80 // dispatch table to use 81 AddressLiteral tbl(Interpreter::dispatch_table(state)); 82 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 83 set(tbl, G3_scratch); // compute addr of table 84 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr 85 } 86 87 88 // Dispatch code executed in the epilog of a bytecode which does not do it's 89 // own dispatch. The dispatch address in IdispatchAddress is used for the 90 // dispatch. 91 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 92 assert_not_delayed(); 93 verify_FPU(1, state); 94 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 95 jmp( IdispatchAddress, 0 ); 96 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); 97 else delayed()->nop(); 98 } 99 100 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) { 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, true, generate_poll); 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, bool generate_poll) { 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() && generate_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, tmp); 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 // Pass the array to create more detailed exceptions. 899 // convention: move aberrant index into G3_scratch for exception message 900 mov(index, Otos_i); 901 mov(array, G3_scratch); 902 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok); 903 904 // add offset if didn't do it in delay slot 905 if (index_shift > 0) add(array, index, res); // addr - const offset in index 906 } 907 908 909 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { 910 assert_not_delayed(); 911 912 // pop array 913 pop_ptr(array); 914 915 // check array 916 index_check_without_pop(array, index, index_shift, tmp, res); 917 } 918 919 920 void InterpreterMacroAssembler::get_const(Register Rdst) { 921 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst); 922 } 923 924 925 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 926 get_const(Rdst); 927 ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 928 } 929 930 931 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 932 get_constant_pool(Rdst); 933 ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 934 } 935 936 937 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 938 get_constant_pool(Rcpool); 939 ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags); 940 } 941 942 943 // unlock if synchronized method 944 // 945 // Unlock the receiver if this is a synchronized method. 946 // Unlock any Java monitors from syncronized blocks. 947 // 948 // If there are locked Java monitors 949 // If throw_monitor_exception 950 // throws IllegalMonitorStateException 951 // Else if install_monitor_exception 952 // installs IllegalMonitorStateException 953 // Else 954 // no error processing 955 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 956 bool throw_monitor_exception, 957 bool install_monitor_exception) { 958 Label unlocked, unlock, no_unlock; 959 960 // get the value of _do_not_unlock_if_synchronized into G1_scratch 961 const Address do_not_unlock_if_synchronized(G2_thread, 962 JavaThread::do_not_unlock_if_synchronized_offset()); 963 ldbool(do_not_unlock_if_synchronized, G1_scratch); 964 stbool(G0, do_not_unlock_if_synchronized); // reset the flag 965 966 // check if synchronized method 967 const Address access_flags(Lmethod, Method::access_flags_offset()); 968 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 969 push(state); // save tos 970 ld(access_flags, G3_scratch); // Load access flags. 971 btst(JVM_ACC_SYNCHRONIZED, G3_scratch); 972 br(zero, false, pt, unlocked); 973 delayed()->nop(); 974 975 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 976 // is set. 977 cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock); 978 delayed()->nop(); 979 980 // BasicObjectLock will be first in list, since this is a synchronized method. However, need 981 // to check that the object has not been unlocked by an explicit monitorexit bytecode. 982 983 //Intel: if (throw_monitor_exception) ... else ... 984 // Entry already unlocked, need to throw exception 985 //... 986 987 // pass top-most monitor elem 988 add( top_most_monitor(), O1 ); 989 990 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch); 991 br_notnull_short(G3_scratch, pt, unlock); 992 993 if (throw_monitor_exception) { 994 // Entry already unlocked need to throw an exception 995 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 996 should_not_reach_here(); 997 } else { 998 // Monitor already unlocked during a stack unroll. 999 // If requested, install an illegal_monitor_state_exception. 1000 // Continue with stack unrolling. 1001 if (install_monitor_exception) { 1002 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1003 } 1004 ba_short(unlocked); 1005 } 1006 1007 bind(unlock); 1008 1009 unlock_object(O1); 1010 1011 bind(unlocked); 1012 1013 // I0, I1: Might contain return value 1014 1015 // Check that all monitors are unlocked 1016 { Label loop, exception, entry, restart; 1017 1018 Register Rmptr = O0; 1019 Register Rtemp = O1; 1020 Register Rlimit = Lmonitors; 1021 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 1022 assert( (delta & LongAlignmentMask) == 0, 1023 "sizeof BasicObjectLock must be even number of doublewords"); 1024 1025 #ifdef ASSERT 1026 add(top_most_monitor(), Rmptr, delta); 1027 { Label L; 1028 // ensure that Rmptr starts out above (or at) Rlimit 1029 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); 1030 stop("monitor stack has negative size"); 1031 bind(L); 1032 } 1033 #endif 1034 bind(restart); 1035 ba(entry); 1036 delayed()-> 1037 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry 1038 1039 // Entry is still locked, need to throw exception 1040 bind(exception); 1041 if (throw_monitor_exception) { 1042 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1043 should_not_reach_here(); 1044 } else { 1045 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 1046 // Unlock does not block, so don't have to worry about the frame 1047 unlock_object(Rmptr); 1048 if (install_monitor_exception) { 1049 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1050 } 1051 ba_short(restart); 1052 } 1053 1054 bind(loop); 1055 cmp(Rtemp, G0); // check if current entry is used 1056 brx(Assembler::notEqual, false, pn, exception); 1057 delayed()-> 1058 dec(Rmptr, delta); // otherwise advance to next entry 1059 #ifdef ASSERT 1060 { Label L; 1061 // ensure that Rmptr has not somehow stepped below Rlimit 1062 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); 1063 stop("ran off the end of the monitor stack"); 1064 bind(L); 1065 } 1066 #endif 1067 bind(entry); 1068 cmp(Rmptr, Rlimit); // check if bottom reached 1069 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry 1070 delayed()-> 1071 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp); 1072 } 1073 1074 bind(no_unlock); 1075 pop(state); 1076 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1077 } 1078 1079 void InterpreterMacroAssembler::narrow(Register result) { 1080 1081 ld_ptr(Address(Lmethod, Method::const_offset()), G3_scratch); 1082 ldub(G3_scratch, in_bytes(ConstMethod::result_type_offset()), G3_scratch); 1083 1084 Label notBool, notByte, notChar, done; 1085 1086 // common case first 1087 cmp(G3_scratch, T_INT); 1088 br(Assembler::equal, true, pn, done); 1089 delayed()->nop(); 1090 1091 cmp(G3_scratch, T_BOOLEAN); 1092 br(Assembler::notEqual, true, pn, notBool); 1093 delayed()->cmp(G3_scratch, T_BYTE); 1094 and3(result, 1, result); 1095 ba(done); 1096 delayed()->nop(); 1097 1098 bind(notBool); 1099 // cmp(G3_scratch, T_BYTE); 1100 br(Assembler::notEqual, true, pn, notByte); 1101 delayed()->cmp(G3_scratch, T_CHAR); 1102 sll(result, 24, result); 1103 sra(result, 24, result); 1104 ba(done); 1105 delayed()->nop(); 1106 1107 bind(notByte); 1108 // cmp(G3_scratch, T_CHAR); 1109 sll(result, 16, result); 1110 br(Assembler::notEqual, true, pn, done); 1111 delayed()->sra(result, 16, result); 1112 // sll(result, 16, result); 1113 srl(result, 16, result); 1114 1115 // bind(notChar); 1116 // must be short, instructions already executed in delay slot 1117 // sll(result, 16, result); 1118 // sra(result, 16, result); 1119 1120 bind(done); 1121 } 1122 1123 // remove activation 1124 // 1125 // Unlock the receiver if this is a synchronized method. 1126 // Unlock any Java monitors from syncronized blocks. 1127 // Remove the activation from the stack. 1128 // 1129 // If there are locked Java monitors 1130 // If throw_monitor_exception 1131 // throws IllegalMonitorStateException 1132 // Else if install_monitor_exception 1133 // installs IllegalMonitorStateException 1134 // Else 1135 // no error processing 1136 void InterpreterMacroAssembler::remove_activation(TosState state, 1137 bool throw_monitor_exception, 1138 bool install_monitor_exception) { 1139 1140 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 1141 1142 // save result (push state before jvmti call and pop it afterwards) and notify jvmti 1143 notify_method_exit(false, state, NotifyJVMTI); 1144 1145 if (StackReservedPages > 0) { 1146 // testing if Stack Reserved Area needs to be re-enabled 1147 Label no_reserved_zone_enabling; 1148 ld_ptr(G2_thread, JavaThread::reserved_stack_activation_offset(), G3_scratch); 1149 cmp_and_brx_short(SP, G3_scratch, Assembler::lessUnsigned, Assembler::pt, no_reserved_zone_enabling); 1150 1151 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), G2_thread); 1152 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError), G2_thread); 1153 should_not_reach_here(); 1154 1155 bind(no_reserved_zone_enabling); 1156 } 1157 1158 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1159 verify_thread(); 1160 1161 // return tos 1162 assert(Otos_l1 == Otos_i, "adjust code below"); 1163 switch (state) { 1164 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0 1165 case btos: // fall through 1166 case ztos: // fall through 1167 case ctos: 1168 case stos: // fall through 1169 case atos: // fall through 1170 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0 1171 case ftos: // fall through 1172 case dtos: // fall through 1173 case vtos: /* nothing to do */ break; 1174 default : ShouldNotReachHere(); 1175 } 1176 } 1177 1178 // Lock object 1179 // 1180 // Argument - lock_reg points to the BasicObjectLock to be used for locking, 1181 // it must be initialized with the object to lock 1182 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) { 1183 if (UseHeavyMonitors) { 1184 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1185 } 1186 else { 1187 Register obj_reg = Object; 1188 Register mark_reg = G4_scratch; 1189 Register temp_reg = G1_scratch; 1190 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes()); 1191 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1192 Label done; 1193 1194 Label slow_case; 1195 1196 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg); 1197 1198 // load markOop from object into mark_reg 1199 ld_ptr(mark_addr, mark_reg); 1200 1201 if (UseBiasedLocking) { 1202 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case); 1203 } 1204 1205 // get the address of basicLock on stack that will be stored in the object 1206 // we need a temporary register here as we do not want to clobber lock_reg 1207 // (cas clobbers the destination register) 1208 mov(lock_reg, temp_reg); 1209 // set mark reg to be (markOop of object | UNLOCK_VALUE) 1210 or3(mark_reg, markOopDesc::unlocked_value, mark_reg); 1211 // initialize the box (Must happen before we update the object mark!) 1212 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1213 // compare and exchange object_addr, markOop | 1, stack address of basicLock 1214 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1215 cas_ptr(mark_addr.base(), mark_reg, temp_reg); 1216 1217 // if the compare and exchange succeeded we are done (we saw an unlocked object) 1218 cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done); 1219 1220 // We did not see an unlocked object so try the fast recursive case 1221 1222 // Check if owner is self by comparing the value in the markOop of object 1223 // with the stack pointer 1224 sub(temp_reg, SP, temp_reg); 1225 sub(temp_reg, STACK_BIAS, temp_reg); 1226 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 1227 1228 // Composite "andcc" test: 1229 // (a) %sp -vs- markword proximity check, and, 1230 // (b) verify mark word LSBs == 0 (Stack-locked). 1231 // 1232 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size()) 1233 // Note that the page size used for %sp proximity testing is arbitrary and is 1234 // unrelated to the actual MMU page size. We use a 'logical' page size of 1235 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate 1236 // field of the andcc instruction. 1237 andcc (temp_reg, 0xFFFFF003, G0) ; 1238 1239 // if condition is true we are done and hence we can store 0 in the displaced 1240 // header indicating it is a recursive lock and be done 1241 brx(Assembler::zero, true, Assembler::pt, done); 1242 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1243 1244 // none of the above fast optimizations worked so we have to get into the 1245 // slow case of monitor enter 1246 bind(slow_case); 1247 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1248 1249 bind(done); 1250 } 1251 } 1252 1253 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 1254 // 1255 // Argument - lock_reg points to the BasicObjectLock for lock 1256 // Throw IllegalMonitorException if object is not locked by current thread 1257 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1258 if (UseHeavyMonitors) { 1259 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1260 } else { 1261 Register obj_reg = G3_scratch; 1262 Register mark_reg = G4_scratch; 1263 Register displaced_header_reg = G1_scratch; 1264 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes()); 1265 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1266 Label done; 1267 1268 if (UseBiasedLocking) { 1269 // load the object out of the BasicObjectLock 1270 ld_ptr(lockobj_addr, obj_reg); 1271 biased_locking_exit(mark_addr, mark_reg, done, true); 1272 st_ptr(G0, lockobj_addr); // free entry 1273 } 1274 1275 // Test first if we are in the fast recursive case 1276 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes()); 1277 ld_ptr(lock_addr, displaced_header_reg); 1278 br_null(displaced_header_reg, true, Assembler::pn, done); 1279 delayed()->st_ptr(G0, lockobj_addr); // free entry 1280 1281 // See if it is still a light weight lock, if so we just unlock 1282 // the object and we are done 1283 1284 if (!UseBiasedLocking) { 1285 // load the object out of the BasicObjectLock 1286 ld_ptr(lockobj_addr, obj_reg); 1287 } 1288 1289 // we have the displaced header in displaced_header_reg 1290 // we expect to see the stack address of the basicLock in case the 1291 // lock is still a light weight lock (lock_reg) 1292 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1293 cas_ptr(mark_addr.base(), lock_reg, displaced_header_reg); 1294 cmp(lock_reg, displaced_header_reg); 1295 brx(Assembler::equal, true, Assembler::pn, done); 1296 delayed()->st_ptr(G0, lockobj_addr); // free entry 1297 1298 // The lock has been converted into a heavy lock and hence 1299 // we need to get into the slow case 1300 1301 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1302 1303 bind(done); 1304 } 1305 } 1306 1307 // Get the method data pointer from the Method* and set the 1308 // specified register to its value. 1309 1310 void InterpreterMacroAssembler::set_method_data_pointer() { 1311 assert(ProfileInterpreter, "must be profiling interpreter"); 1312 Label get_continue; 1313 1314 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); 1315 test_method_data_pointer(get_continue); 1316 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); 1317 bind(get_continue); 1318 } 1319 1320 // Set the method data pointer for the current bcp. 1321 1322 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1323 assert(ProfileInterpreter, "must be profiling interpreter"); 1324 Label zero_continue; 1325 1326 // Test MDO to avoid the call if it is NULL. 1327 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); 1328 test_method_data_pointer(zero_continue); 1329 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp); 1330 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); 1331 add(ImethodDataPtr, O0, ImethodDataPtr); 1332 bind(zero_continue); 1333 } 1334 1335 // Test ImethodDataPtr. If it is null, continue at the specified label 1336 1337 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1338 assert(ProfileInterpreter, "must be profiling interpreter"); 1339 br_null_short(ImethodDataPtr, Assembler::pn, zero_continue); 1340 } 1341 1342 void InterpreterMacroAssembler::verify_method_data_pointer() { 1343 assert(ProfileInterpreter, "must be profiling interpreter"); 1344 #ifdef ASSERT 1345 Label verify_continue; 1346 test_method_data_pointer(verify_continue); 1347 1348 // If the mdp is valid, it will point to a DataLayout header which is 1349 // consistent with the bcp. The converse is highly probable also. 1350 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch); 1351 ld_ptr(Lmethod, Method::const_offset(), O5); 1352 add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch); 1353 add(G3_scratch, O5, G3_scratch); 1354 cmp(Lbcp, G3_scratch); 1355 brx(Assembler::equal, false, Assembler::pt, verify_continue); 1356 1357 Register temp_reg = O5; 1358 delayed()->mov(ImethodDataPtr, temp_reg); 1359 // %%% should use call_VM_leaf here? 1360 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr); 1361 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1); 1362 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS); 1363 stf(FloatRegisterImpl::D, Ftos_d, d_save); 1364 mov(temp_reg->after_save(), O2); 1365 save_thread(L7_thread_cache); 1366 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none); 1367 delayed()->nop(); 1368 restore_thread(L7_thread_cache); 1369 ldf(FloatRegisterImpl::D, d_save, Ftos_d); 1370 restore(); 1371 bind(verify_continue); 1372 #endif // ASSERT 1373 } 1374 1375 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1376 Register method_counters, 1377 Register Rtmp, 1378 Label &profile_continue) { 1379 assert(ProfileInterpreter, "must be profiling interpreter"); 1380 // Control will flow to "profile_continue" if the counter is less than the 1381 // limit or if we call profile_method() 1382 1383 Label done; 1384 1385 // if no method data exists, and the counter is high enough, make one 1386 br_notnull_short(ImethodDataPtr, Assembler::pn, done); 1387 1388 // Test to see if we should create a method data oop 1389 Address profile_limit(method_counters, MethodCounters::interpreter_profile_limit_offset()); 1390 ld(profile_limit, Rtmp); 1391 cmp(invocation_count, Rtmp); 1392 // Use long branches because call_VM() code and following code generated by 1393 // test_backedge_count_for_osr() is large in debug VM. 1394 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue); 1395 delayed()->nop(); 1396 1397 // Build it now. 1398 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1399 set_method_data_pointer_for_bcp(); 1400 ba(profile_continue); 1401 delayed()->nop(); 1402 bind(done); 1403 } 1404 1405 // Store a value at some constant offset from the method data pointer. 1406 1407 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1408 assert(ProfileInterpreter, "must be profiling interpreter"); 1409 st_ptr(value, ImethodDataPtr, constant); 1410 } 1411 1412 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter, 1413 Register bumped_count, 1414 bool decrement) { 1415 assert(ProfileInterpreter, "must be profiling interpreter"); 1416 1417 // Load the counter. 1418 ld_ptr(counter, bumped_count); 1419 1420 if (decrement) { 1421 // Decrement the register. Set condition codes. 1422 subcc(bumped_count, DataLayout::counter_increment, bumped_count); 1423 1424 // If the decrement causes the counter to overflow, stay negative 1425 Label L; 1426 brx(Assembler::negative, true, Assembler::pn, L); 1427 1428 // Store the decremented counter, if it is still negative. 1429 delayed()->st_ptr(bumped_count, counter); 1430 bind(L); 1431 } else { 1432 // Increment the register. Set carry flag. 1433 addcc(bumped_count, DataLayout::counter_increment, bumped_count); 1434 1435 // If the increment causes the counter to overflow, pull back by 1. 1436 assert(DataLayout::counter_increment == 1, "subc works"); 1437 subc(bumped_count, G0, bumped_count); 1438 1439 // Store the incremented counter. 1440 st_ptr(bumped_count, counter); 1441 } 1442 } 1443 1444 // Increment the value at some constant offset from the method data pointer. 1445 1446 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1447 Register bumped_count, 1448 bool decrement) { 1449 // Locate the counter at a fixed offset from the mdp: 1450 Address counter(ImethodDataPtr, constant); 1451 increment_mdp_data_at(counter, bumped_count, decrement); 1452 } 1453 1454 // Increment the value at some non-fixed (reg + constant) offset from 1455 // the method data pointer. 1456 1457 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1458 int constant, 1459 Register bumped_count, 1460 Register scratch2, 1461 bool decrement) { 1462 // Add the constant to reg to get the offset. 1463 add(ImethodDataPtr, reg, scratch2); 1464 Address counter(scratch2, constant); 1465 increment_mdp_data_at(counter, bumped_count, decrement); 1466 } 1467 1468 // Set a flag value at the current method data pointer position. 1469 // Updates a single byte of the header, to avoid races with other header bits. 1470 1471 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1472 Register scratch) { 1473 assert(ProfileInterpreter, "must be profiling interpreter"); 1474 // Load the data header 1475 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch); 1476 1477 // Set the flag 1478 or3(scratch, flag_constant, scratch); 1479 1480 // Store the modified header. 1481 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset())); 1482 } 1483 1484 // Test the location at some offset from the method data pointer. 1485 // If it is not equal to value, branch to the not_equal_continue Label. 1486 // Set condition codes to match the nullness of the loaded value. 1487 1488 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1489 Register value, 1490 Label& not_equal_continue, 1491 Register scratch) { 1492 assert(ProfileInterpreter, "must be profiling interpreter"); 1493 ld_ptr(ImethodDataPtr, offset, scratch); 1494 cmp(value, scratch); 1495 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue); 1496 delayed()->tst(scratch); 1497 } 1498 1499 // Update the method data pointer by the displacement located at some fixed 1500 // offset from the method data pointer. 1501 1502 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1503 Register scratch) { 1504 assert(ProfileInterpreter, "must be profiling interpreter"); 1505 ld_ptr(ImethodDataPtr, offset_of_disp, scratch); 1506 add(ImethodDataPtr, scratch, ImethodDataPtr); 1507 } 1508 1509 // Update the method data pointer by the displacement located at the 1510 // offset (reg + offset_of_disp). 1511 1512 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1513 int offset_of_disp, 1514 Register scratch) { 1515 assert(ProfileInterpreter, "must be profiling interpreter"); 1516 add(reg, offset_of_disp, scratch); 1517 ld_ptr(ImethodDataPtr, scratch, scratch); 1518 add(ImethodDataPtr, scratch, ImethodDataPtr); 1519 } 1520 1521 // Update the method data pointer by a simple constant displacement. 1522 1523 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1524 assert(ProfileInterpreter, "must be profiling interpreter"); 1525 add(ImethodDataPtr, constant, ImethodDataPtr); 1526 } 1527 1528 // Update the method data pointer for a _ret bytecode whose target 1529 // was not among our cached targets. 1530 1531 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1532 Register return_bci) { 1533 assert(ProfileInterpreter, "must be profiling interpreter"); 1534 push(state); 1535 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile 1536 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1537 ld_ptr(l_tmp, return_bci); 1538 pop(state); 1539 } 1540 1541 // Count a taken branch in the bytecodes. 1542 1543 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1544 if (ProfileInterpreter) { 1545 Label profile_continue; 1546 1547 // If no method data exists, go to profile_continue. 1548 test_method_data_pointer(profile_continue); 1549 1550 // We are taking a branch. Increment the taken count. 1551 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count); 1552 1553 // The method data pointer needs to be updated to reflect the new target. 1554 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1555 bind (profile_continue); 1556 } 1557 } 1558 1559 1560 // Count a not-taken branch in the bytecodes. 1561 1562 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) { 1563 if (ProfileInterpreter) { 1564 Label profile_continue; 1565 1566 // If no method data exists, go to profile_continue. 1567 test_method_data_pointer(profile_continue); 1568 1569 // We are taking a branch. Increment the not taken count. 1570 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch); 1571 1572 // The method data pointer needs to be updated to correspond to the 1573 // next bytecode. 1574 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1575 bind (profile_continue); 1576 } 1577 } 1578 1579 1580 // Count a non-virtual call in the bytecodes. 1581 1582 void InterpreterMacroAssembler::profile_call(Register scratch) { 1583 if (ProfileInterpreter) { 1584 Label profile_continue; 1585 1586 // If no method data exists, go to profile_continue. 1587 test_method_data_pointer(profile_continue); 1588 1589 // We are making a call. Increment the count. 1590 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1591 1592 // The method data pointer needs to be updated to reflect the new target. 1593 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1594 bind (profile_continue); 1595 } 1596 } 1597 1598 1599 // Count a final call in the bytecodes. 1600 1601 void InterpreterMacroAssembler::profile_final_call(Register scratch) { 1602 if (ProfileInterpreter) { 1603 Label profile_continue; 1604 1605 // If no method data exists, go to profile_continue. 1606 test_method_data_pointer(profile_continue); 1607 1608 // We are making a call. Increment the count. 1609 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1610 1611 // The method data pointer needs to be updated to reflect the new target. 1612 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1613 bind (profile_continue); 1614 } 1615 } 1616 1617 1618 // Count a virtual call in the bytecodes. 1619 1620 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1621 Register scratch, 1622 bool receiver_can_be_null) { 1623 if (ProfileInterpreter) { 1624 Label profile_continue; 1625 1626 // If no method data exists, go to profile_continue. 1627 test_method_data_pointer(profile_continue); 1628 1629 1630 Label skip_receiver_profile; 1631 if (receiver_can_be_null) { 1632 Label not_null; 1633 br_notnull_short(receiver, Assembler::pt, not_null); 1634 // We are making a call. Increment the count for null receiver. 1635 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1636 ba_short(skip_receiver_profile); 1637 bind(not_null); 1638 } 1639 1640 // Record the receiver type. 1641 record_klass_in_profile(receiver, scratch, true); 1642 bind(skip_receiver_profile); 1643 1644 // The method data pointer needs to be updated to reflect the new target. 1645 #if INCLUDE_JVMCI 1646 if (MethodProfileWidth == 0) { 1647 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1648 } 1649 #else 1650 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1651 #endif 1652 bind(profile_continue); 1653 } 1654 } 1655 1656 #if INCLUDE_JVMCI 1657 void InterpreterMacroAssembler::profile_called_method(Register method, Register scratch) { 1658 assert_different_registers(method, scratch); 1659 if (ProfileInterpreter && MethodProfileWidth > 0) { 1660 Label profile_continue; 1661 1662 // If no method data exists, go to profile_continue. 1663 test_method_data_pointer(profile_continue); 1664 1665 Label done; 1666 record_item_in_profile_helper(method, scratch, 0, done, MethodProfileWidth, 1667 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset())); 1668 bind(done); 1669 1670 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1671 bind(profile_continue); 1672 } 1673 } 1674 #endif // INCLUDE_JVMCI 1675 1676 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register scratch, 1677 Label& done, bool is_virtual_call) { 1678 if (TypeProfileWidth == 0) { 1679 if (is_virtual_call) { 1680 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1681 } 1682 #if INCLUDE_JVMCI 1683 else if (EnableJVMCI) { 1684 increment_mdp_data_at(in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()), scratch); 1685 } 1686 #endif 1687 } else { 1688 int non_profiled_offset = -1; 1689 if (is_virtual_call) { 1690 non_profiled_offset = in_bytes(CounterData::count_offset()); 1691 } 1692 #if INCLUDE_JVMCI 1693 else if (EnableJVMCI) { 1694 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()); 1695 } 1696 #endif 1697 1698 record_item_in_profile_helper(receiver, scratch, 0, done, TypeProfileWidth, 1699 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset); 1700 } 1701 } 1702 1703 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, 1704 Register scratch, int start_row, Label& done, int total_rows, 1705 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, 1706 int non_profiled_offset) { 1707 int last_row = total_rows - 1; 1708 assert(start_row <= last_row, "must be work left to do"); 1709 // Test this row for both the item and for null. 1710 // Take any of three different outcomes: 1711 // 1. found item => increment count and goto done 1712 // 2. found null => keep looking for case 1, maybe allocate this cell 1713 // 3. found something else => keep looking for cases 1 and 2 1714 // Case 3 is handled by a recursive call. 1715 for (int row = start_row; row <= last_row; row++) { 1716 Label next_test; 1717 bool test_for_null_also = (row == start_row); 1718 1719 // See if the item is item[n]. 1720 int item_offset = in_bytes(item_offset_fn(row)); 1721 test_mdp_data_at(item_offset, item, next_test, scratch); 1722 // delayed()->tst(scratch); 1723 1724 // The receiver is item[n]. Increment count[n]. 1725 int count_offset = in_bytes(item_count_offset_fn(row)); 1726 increment_mdp_data_at(count_offset, scratch); 1727 ba_short(done); 1728 bind(next_test); 1729 1730 if (test_for_null_also) { 1731 Label found_null; 1732 // Failed the equality check on item[n]... Test for null. 1733 if (start_row == last_row) { 1734 // The only thing left to do is handle the null case. 1735 if (non_profiled_offset >= 0) { 1736 brx(Assembler::zero, false, Assembler::pn, found_null); 1737 delayed()->nop(); 1738 // Item did not match any saved item and there is no empty row for it. 1739 // Increment total counter to indicate polymorphic case. 1740 increment_mdp_data_at(non_profiled_offset, scratch); 1741 ba_short(done); 1742 bind(found_null); 1743 } else { 1744 brx(Assembler::notZero, false, Assembler::pt, done); 1745 delayed()->nop(); 1746 } 1747 break; 1748 } 1749 // Since null is rare, make it be the branch-taken case. 1750 brx(Assembler::zero, false, Assembler::pn, found_null); 1751 delayed()->nop(); 1752 1753 // Put all the "Case 3" tests here. 1754 record_item_in_profile_helper(item, scratch, start_row + 1, done, total_rows, 1755 item_offset_fn, item_count_offset_fn, non_profiled_offset); 1756 1757 // Found a null. Keep searching for a matching item, 1758 // but remember that this is an empty (unused) slot. 1759 bind(found_null); 1760 } 1761 } 1762 1763 // In the fall-through case, we found no matching item, but we 1764 // observed the item[start_row] is NULL. 1765 1766 // Fill in the item field and increment the count. 1767 int item_offset = in_bytes(item_offset_fn(start_row)); 1768 set_mdp_data_at(item_offset, item); 1769 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1770 mov(DataLayout::counter_increment, scratch); 1771 set_mdp_data_at(count_offset, scratch); 1772 if (start_row > 0) { 1773 ba_short(done); 1774 } 1775 } 1776 1777 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1778 Register scratch, bool is_virtual_call) { 1779 assert(ProfileInterpreter, "must be profiling"); 1780 Label done; 1781 1782 record_klass_in_profile_helper(receiver, scratch, done, is_virtual_call); 1783 1784 bind (done); 1785 } 1786 1787 1788 // Count a ret in the bytecodes. 1789 1790 void InterpreterMacroAssembler::profile_ret(TosState state, 1791 Register return_bci, 1792 Register scratch) { 1793 if (ProfileInterpreter) { 1794 Label profile_continue; 1795 uint row; 1796 1797 // If no method data exists, go to profile_continue. 1798 test_method_data_pointer(profile_continue); 1799 1800 // Update the total ret count. 1801 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1802 1803 for (row = 0; row < RetData::row_limit(); row++) { 1804 Label next_test; 1805 1806 // See if return_bci is equal to bci[n]: 1807 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), 1808 return_bci, next_test, scratch); 1809 1810 // return_bci is equal to bci[n]. Increment the count. 1811 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch); 1812 1813 // The method data pointer needs to be updated to reflect the new target. 1814 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch); 1815 ba_short(profile_continue); 1816 bind(next_test); 1817 } 1818 1819 update_mdp_for_ret(state, return_bci); 1820 1821 bind (profile_continue); 1822 } 1823 } 1824 1825 // Profile an unexpected null in the bytecodes. 1826 void InterpreterMacroAssembler::profile_null_seen(Register scratch) { 1827 if (ProfileInterpreter) { 1828 Label profile_continue; 1829 1830 // If no method data exists, go to profile_continue. 1831 test_method_data_pointer(profile_continue); 1832 1833 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch); 1834 1835 // The method data pointer needs to be updated. 1836 int mdp_delta = in_bytes(BitData::bit_data_size()); 1837 if (TypeProfileCasts) { 1838 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size()); 1839 } 1840 update_mdp_by_constant(mdp_delta); 1841 1842 bind (profile_continue); 1843 } 1844 } 1845 1846 void InterpreterMacroAssembler::profile_typecheck(Register klass, 1847 Register scratch) { 1848 if (ProfileInterpreter) { 1849 Label profile_continue; 1850 1851 // If no method data exists, go to profile_continue. 1852 test_method_data_pointer(profile_continue); 1853 1854 int mdp_delta = in_bytes(BitData::bit_data_size()); 1855 if (TypeProfileCasts) { 1856 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size()); 1857 1858 // Record the object type. 1859 record_klass_in_profile(klass, scratch, false); 1860 } 1861 1862 // The method data pointer needs to be updated. 1863 update_mdp_by_constant(mdp_delta); 1864 1865 bind (profile_continue); 1866 } 1867 } 1868 1869 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) { 1870 if (ProfileInterpreter && TypeProfileCasts) { 1871 Label profile_continue; 1872 1873 // If no method data exists, go to profile_continue. 1874 test_method_data_pointer(profile_continue); 1875 1876 int count_offset = in_bytes(CounterData::count_offset()); 1877 // Back up the address, since we have already bumped the mdp. 1878 count_offset -= in_bytes(ReceiverTypeData::receiver_type_data_size()); 1879 1880 // *Decrement* the counter. We expect to see zero or small negatives. 1881 increment_mdp_data_at(count_offset, scratch, true); 1882 1883 bind (profile_continue); 1884 } 1885 } 1886 1887 // Count the default case of a switch construct. 1888 1889 void InterpreterMacroAssembler::profile_switch_default(Register scratch) { 1890 if (ProfileInterpreter) { 1891 Label profile_continue; 1892 1893 // If no method data exists, go to profile_continue. 1894 test_method_data_pointer(profile_continue); 1895 1896 // Update the default case count 1897 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1898 scratch); 1899 1900 // The method data pointer needs to be updated. 1901 update_mdp_by_offset( 1902 in_bytes(MultiBranchData::default_displacement_offset()), 1903 scratch); 1904 1905 bind (profile_continue); 1906 } 1907 } 1908 1909 // Count the index'th case of a switch construct. 1910 1911 void InterpreterMacroAssembler::profile_switch_case(Register index, 1912 Register scratch, 1913 Register scratch2, 1914 Register scratch3) { 1915 if (ProfileInterpreter) { 1916 Label profile_continue; 1917 1918 // If no method data exists, go to profile_continue. 1919 test_method_data_pointer(profile_continue); 1920 1921 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() 1922 set(in_bytes(MultiBranchData::per_case_size()), scratch); 1923 smul(index, scratch, scratch); 1924 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch); 1925 1926 // Update the case count 1927 increment_mdp_data_at(scratch, 1928 in_bytes(MultiBranchData::relative_count_offset()), 1929 scratch2, 1930 scratch3); 1931 1932 // The method data pointer needs to be updated. 1933 update_mdp_by_offset(scratch, 1934 in_bytes(MultiBranchData::relative_displacement_offset()), 1935 scratch2); 1936 1937 bind (profile_continue); 1938 } 1939 } 1940 1941 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) { 1942 Label not_null, do_nothing, do_update; 1943 1944 assert_different_registers(obj, mdo_addr.base(), tmp); 1945 1946 verify_oop(obj); 1947 1948 ld_ptr(mdo_addr, tmp); 1949 1950 br_notnull_short(obj, pt, not_null); 1951 or3(tmp, TypeEntries::null_seen, tmp); 1952 ba_short(do_update); 1953 1954 bind(not_null); 1955 load_klass(obj, obj); 1956 1957 xor3(obj, tmp, obj); 1958 btst(TypeEntries::type_klass_mask, obj); 1959 // klass seen before, nothing to do. The unknown bit may have been 1960 // set already but no need to check. 1961 brx(zero, false, pt, do_nothing); 1962 delayed()-> 1963 1964 btst(TypeEntries::type_unknown, obj); 1965 // already unknown. Nothing to do anymore. 1966 brx(notZero, false, pt, do_nothing); 1967 delayed()-> 1968 1969 btst(TypeEntries::type_mask, tmp); 1970 brx(zero, true, pt, do_update); 1971 // first time here. Set profile type. 1972 delayed()->or3(tmp, obj, tmp); 1973 1974 // different than before. Cannot keep accurate profile. 1975 or3(tmp, TypeEntries::type_unknown, tmp); 1976 1977 bind(do_update); 1978 // update profile 1979 st_ptr(tmp, mdo_addr); 1980 1981 bind(do_nothing); 1982 } 1983 1984 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) { 1985 if (!ProfileInterpreter) { 1986 return; 1987 } 1988 1989 assert_different_registers(callee, tmp1, tmp2, ImethodDataPtr); 1990 1991 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1992 Label profile_continue; 1993 1994 test_method_data_pointer(profile_continue); 1995 1996 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1997 1998 ldub(ImethodDataPtr, in_bytes(DataLayout::tag_offset()) - off_to_start, tmp1); 1999 cmp_and_br_short(tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag, notEqual, pn, profile_continue); 2000 2001 if (MethodData::profile_arguments()) { 2002 Label done; 2003 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 2004 add(ImethodDataPtr, off_to_args, ImethodDataPtr); 2005 2006 for (int i = 0; i < TypeProfileArgsLimit; i++) { 2007 if (i > 0 || MethodData::profile_return()) { 2008 // If return value type is profiled we may have no argument to profile 2009 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1); 2010 sub(tmp1, i*TypeStackSlotEntries::per_arg_count(), tmp1); 2011 cmp_and_br_short(tmp1, TypeStackSlotEntries::per_arg_count(), less, pn, done); 2012 } 2013 ld_ptr(Address(callee, Method::const_offset()), tmp1); 2014 lduh(Address(tmp1, ConstMethod::size_of_parameters_offset()), tmp1); 2015 // stack offset o (zero based) from the start of the argument 2016 // list, for n arguments translates into offset n - o - 1 from 2017 // the end of the argument list. But there's an extra slot at 2018 // the stop of the stack. So the offset is n - o from Lesp. 2019 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, tmp2); 2020 sub(tmp1, tmp2, tmp1); 2021 2022 // Can't use MacroAssembler::argument_address() which needs Gargs to be set up 2023 sll(tmp1, Interpreter::logStackElementSize, tmp1); 2024 ld_ptr(Lesp, tmp1, tmp1); 2025 2026 Address mdo_arg_addr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); 2027 profile_obj_type(tmp1, mdo_arg_addr, tmp2); 2028 2029 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 2030 add(ImethodDataPtr, to_add, ImethodDataPtr); 2031 off_to_args += to_add; 2032 } 2033 2034 if (MethodData::profile_return()) { 2035 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1); 2036 sub(tmp1, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count(), tmp1); 2037 } 2038 2039 bind(done); 2040 2041 if (MethodData::profile_return()) { 2042 // We're right after the type profile for the last 2043 // argument. tmp1 is the number of cells left in the 2044 // CallTypeData/VirtualCallTypeData to reach its end. Non null 2045 // if there's a return to profile. 2046 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 2047 sll(tmp1, exact_log2(DataLayout::cell_size), tmp1); 2048 add(ImethodDataPtr, tmp1, ImethodDataPtr); 2049 } 2050 } else { 2051 assert(MethodData::profile_return(), "either profile call args or call ret"); 2052 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 2053 } 2054 2055 // mdp points right after the end of the 2056 // CallTypeData/VirtualCallTypeData, right after the cells for the 2057 // return value type if there's one. 2058 2059 bind(profile_continue); 2060 } 2061 } 2062 2063 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 2064 assert_different_registers(ret, tmp1, tmp2); 2065 if (ProfileInterpreter && MethodData::profile_return()) { 2066 Label profile_continue, done; 2067 2068 test_method_data_pointer(profile_continue); 2069 2070 if (MethodData::profile_return_jsr292_only()) { 2071 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); 2072 2073 // If we don't profile all invoke bytecodes we must make sure 2074 // it's a bytecode we indeed profile. We can't go back to the 2075 // begining of the ProfileData we intend to update to check its 2076 // type because we're right after it and we don't known its 2077 // length. 2078 Label do_profile; 2079 ldub(Lbcp, 0, tmp1); 2080 cmp_and_br_short(tmp1, Bytecodes::_invokedynamic, equal, pn, do_profile); 2081 cmp(tmp1, Bytecodes::_invokehandle); 2082 br(equal, false, pn, do_profile); 2083 delayed()->lduh(Lmethod, Method::intrinsic_id_offset_in_bytes(), tmp1); 2084 cmp_and_br_short(tmp1, vmIntrinsics::_compiledLambdaForm, notEqual, pt, profile_continue); 2085 2086 bind(do_profile); 2087 } 2088 2089 Address mdo_ret_addr(ImethodDataPtr, -in_bytes(ReturnTypeEntry::size())); 2090 mov(ret, tmp1); 2091 profile_obj_type(tmp1, mdo_ret_addr, tmp2); 2092 2093 bind(profile_continue); 2094 } 2095 } 2096 2097 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) { 2098 if (ProfileInterpreter && MethodData::profile_parameters()) { 2099 Label profile_continue, done; 2100 2101 test_method_data_pointer(profile_continue); 2102 2103 // Load the offset of the area within the MDO used for 2104 // parameters. If it's negative we're not profiling any parameters. 2105 lduw(ImethodDataPtr, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), tmp1); 2106 cmp_and_br_short(tmp1, 0, less, pn, profile_continue); 2107 2108 // Compute a pointer to the area for parameters from the offset 2109 // and move the pointer to the slot for the last 2110 // parameters. Collect profiling from last parameter down. 2111 // mdo start + parameters offset + array length - 1 2112 2113 // Pointer to the parameter area in the MDO 2114 Register mdp = tmp1; 2115 add(ImethodDataPtr, tmp1, mdp); 2116 2117 // offset of the current profile entry to update 2118 Register entry_offset = tmp2; 2119 // entry_offset = array len in number of cells 2120 ld_ptr(mdp, ArrayData::array_len_offset(), entry_offset); 2121 2122 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 2123 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 2124 2125 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 2126 sub(entry_offset, TypeStackSlotEntries::per_arg_count() - (off_base / DataLayout::cell_size), entry_offset); 2127 // entry_offset in bytes 2128 sll(entry_offset, exact_log2(DataLayout::cell_size), entry_offset); 2129 2130 Label loop; 2131 bind(loop); 2132 2133 // load offset on the stack from the slot for this parameter 2134 ld_ptr(mdp, entry_offset, tmp3); 2135 sll(tmp3,Interpreter::logStackElementSize, tmp3); 2136 neg(tmp3); 2137 // read the parameter from the local area 2138 ld_ptr(Llocals, tmp3, tmp3); 2139 2140 // make entry_offset now point to the type field for this parameter 2141 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 2142 assert(type_base > off_base, "unexpected"); 2143 add(entry_offset, type_base - off_base, entry_offset); 2144 2145 // profile the parameter 2146 Address arg_type(mdp, entry_offset); 2147 profile_obj_type(tmp3, arg_type, tmp4); 2148 2149 // go to next parameter 2150 sub(entry_offset, TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base), entry_offset); 2151 cmp_and_br_short(entry_offset, off_base, greaterEqual, pt, loop); 2152 2153 bind(profile_continue); 2154 } 2155 } 2156 2157 // add a InterpMonitorElem to stack (see frame_sparc.hpp) 2158 2159 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty, 2160 Register Rtemp, 2161 Register Rtemp2 ) { 2162 2163 Register Rlimit = Lmonitors; 2164 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 2165 assert( (delta & LongAlignmentMask) == 0, 2166 "sizeof BasicObjectLock must be even number of doublewords"); 2167 2168 sub( SP, delta, SP); 2169 sub( Lesp, delta, Lesp); 2170 sub( Lmonitors, delta, Lmonitors); 2171 2172 if (!stack_is_empty) { 2173 2174 // must copy stack contents down 2175 2176 Label start_copying, next; 2177 2178 // untested("monitor stack expansion"); 2179 compute_stack_base(Rtemp); 2180 ba(start_copying); 2181 delayed()->cmp(Rtemp, Rlimit); // done? duplicated below 2182 2183 // note: must copy from low memory upwards 2184 // On entry to loop, 2185 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS) 2186 // Loop mutates Rtemp 2187 2188 bind( next); 2189 2190 st_ptr(Rtemp2, Rtemp, 0); 2191 inc(Rtemp, wordSize); 2192 cmp(Rtemp, Rlimit); // are we done? (duplicated above) 2193 2194 bind( start_copying ); 2195 2196 brx( notEqual, true, pn, next ); 2197 delayed()->ld_ptr( Rtemp, delta, Rtemp2 ); 2198 2199 // done copying stack 2200 } 2201 } 2202 2203 // Locals 2204 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) { 2205 assert_not_delayed(); 2206 sll(index, Interpreter::logStackElementSize, index); 2207 sub(Llocals, index, index); 2208 ld_ptr(index, 0, dst); 2209 // Note: index must hold the effective address--the iinc template uses it 2210 } 2211 2212 // Just like access_local_ptr but the tag is a returnAddress 2213 void InterpreterMacroAssembler::access_local_returnAddress(Register index, 2214 Register dst ) { 2215 assert_not_delayed(); 2216 sll(index, Interpreter::logStackElementSize, index); 2217 sub(Llocals, index, index); 2218 ld_ptr(index, 0, dst); 2219 } 2220 2221 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) { 2222 assert_not_delayed(); 2223 sll(index, Interpreter::logStackElementSize, index); 2224 sub(Llocals, index, index); 2225 ld(index, 0, dst); 2226 // Note: index must hold the effective address--the iinc template uses it 2227 } 2228 2229 2230 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) { 2231 assert_not_delayed(); 2232 sll(index, Interpreter::logStackElementSize, index); 2233 sub(Llocals, index, index); 2234 // First half stored at index n+1 (which grows down from Llocals[n]) 2235 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst); 2236 } 2237 2238 2239 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) { 2240 assert_not_delayed(); 2241 sll(index, Interpreter::logStackElementSize, index); 2242 sub(Llocals, index, index); 2243 ldf(FloatRegisterImpl::S, index, 0, dst); 2244 } 2245 2246 2247 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) { 2248 assert_not_delayed(); 2249 sll(index, Interpreter::logStackElementSize, index); 2250 sub(Llocals, index, index); 2251 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst); 2252 } 2253 2254 2255 #ifdef ASSERT 2256 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) { 2257 Label L; 2258 2259 assert(Rindex != Rscratch, "Registers cannot be same"); 2260 assert(Rindex != Rscratch1, "Registers cannot be same"); 2261 assert(Rlimit != Rscratch, "Registers cannot be same"); 2262 assert(Rlimit != Rscratch1, "Registers cannot be same"); 2263 assert(Rscratch1 != Rscratch, "Registers cannot be same"); 2264 2265 // untested("reg area corruption"); 2266 add(Rindex, offset, Rscratch); 2267 add(Rlimit, 64 + STACK_BIAS, Rscratch1); 2268 cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L); 2269 stop("regsave area is being clobbered"); 2270 bind(L); 2271 } 2272 #endif // ASSERT 2273 2274 2275 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) { 2276 assert_not_delayed(); 2277 sll(index, Interpreter::logStackElementSize, index); 2278 sub(Llocals, index, index); 2279 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);) 2280 st(src, index, 0); 2281 } 2282 2283 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) { 2284 assert_not_delayed(); 2285 sll(index, Interpreter::logStackElementSize, index); 2286 sub(Llocals, index, index); 2287 #ifdef ASSERT 2288 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2289 #endif 2290 st_ptr(src, index, 0); 2291 } 2292 2293 2294 2295 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) { 2296 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n)); 2297 } 2298 2299 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) { 2300 assert_not_delayed(); 2301 sll(index, Interpreter::logStackElementSize, index); 2302 sub(Llocals, index, index); 2303 #ifdef ASSERT 2304 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2305 #endif 2306 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1 2307 } 2308 2309 2310 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) { 2311 assert_not_delayed(); 2312 sll(index, Interpreter::logStackElementSize, index); 2313 sub(Llocals, index, index); 2314 #ifdef ASSERT 2315 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2316 #endif 2317 stf(FloatRegisterImpl::S, src, index, 0); 2318 } 2319 2320 2321 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) { 2322 assert_not_delayed(); 2323 sll(index, Interpreter::logStackElementSize, index); 2324 sub(Llocals, index, index); 2325 #ifdef ASSERT 2326 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2327 #endif 2328 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1)); 2329 } 2330 2331 2332 int InterpreterMacroAssembler::top_most_monitor_byte_offset() { 2333 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 2334 int rounded_vm_local_words = align_up((int)frame::interpreter_frame_vm_local_words, WordsPerLong); 2335 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS; 2336 } 2337 2338 2339 Address InterpreterMacroAssembler::top_most_monitor() { 2340 return Address(FP, top_most_monitor_byte_offset()); 2341 } 2342 2343 2344 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) { 2345 add( Lesp, wordSize, Rdest ); 2346 } 2347 2348 void InterpreterMacroAssembler::get_method_counters(Register method, 2349 Register Rcounters, 2350 Label& skip) { 2351 Label has_counters; 2352 Address method_counters(method, in_bytes(Method::method_counters_offset())); 2353 ld_ptr(method_counters, Rcounters); 2354 br_notnull_short(Rcounters, Assembler::pt, has_counters); 2355 call_VM(noreg, CAST_FROM_FN_PTR(address, 2356 InterpreterRuntime::build_method_counters), method); 2357 ld_ptr(method_counters, Rcounters); 2358 br_null(Rcounters, false, Assembler::pn, skip); // No MethodCounters, OutOfMemory 2359 delayed()->nop(); 2360 bind(has_counters); 2361 } 2362 2363 void InterpreterMacroAssembler::increment_invocation_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) { 2364 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 2365 assert_different_registers(Rcounters, Rtmp, Rtmp2); 2366 2367 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() + 2368 InvocationCounter::counter_offset()); 2369 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() + 2370 InvocationCounter::counter_offset()); 2371 int delta = InvocationCounter::count_increment; 2372 2373 // Load each counter in a register 2374 ld( inv_counter, Rtmp ); 2375 ld( be_counter, Rtmp2 ); 2376 2377 assert( is_simm13( delta ), " delta too large."); 2378 2379 // Add the delta to the invocation counter and store the result 2380 add( Rtmp, delta, Rtmp ); 2381 2382 // Mask the backedge counter 2383 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2384 2385 // Store value 2386 st( Rtmp, inv_counter); 2387 2388 // Add invocation counter + backedge counter 2389 add( Rtmp, Rtmp2, Rtmp); 2390 2391 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 2392 } 2393 2394 void InterpreterMacroAssembler::increment_backedge_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) { 2395 assert(UseCompiler, "incrementing must be useful"); 2396 assert_different_registers(Rcounters, Rtmp, Rtmp2); 2397 2398 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() + 2399 InvocationCounter::counter_offset()); 2400 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() + 2401 InvocationCounter::counter_offset()); 2402 2403 int delta = InvocationCounter::count_increment; 2404 // Load each counter in a register 2405 ld( be_counter, Rtmp ); 2406 ld( inv_counter, Rtmp2 ); 2407 2408 // Add the delta to the backedge counter 2409 add( Rtmp, delta, Rtmp ); 2410 2411 // Mask the invocation counter, add to backedge counter 2412 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2413 2414 // and store the result to memory 2415 st( Rtmp, be_counter ); 2416 2417 // Add backedge + invocation counter 2418 add( Rtmp, Rtmp2, Rtmp ); 2419 2420 // Note that this macro must leave backedge_count + invocation_count in Rtmp! 2421 } 2422 2423 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count, 2424 Register method_counters, 2425 Register branch_bcp, 2426 Register Rtmp ) { 2427 Label did_not_overflow; 2428 Label overflow_with_error; 2429 assert_different_registers(backedge_count, Rtmp, branch_bcp); 2430 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 2431 2432 Address limit(method_counters, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())); 2433 ld(limit, Rtmp); 2434 cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow); 2435 2436 // When ProfileInterpreter is on, the backedge_count comes from the 2437 // MethodData*, which value does not get reset on the call to 2438 // frequency_counter_overflow(). To avoid excessive calls to the overflow 2439 // routine while the method is being compiled, add a second test to make sure 2440 // the overflow function is called only once every overflow_frequency. 2441 if (ProfileInterpreter) { 2442 const int overflow_frequency = 1024; 2443 andcc(backedge_count, overflow_frequency-1, Rtmp); 2444 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow); 2445 delayed()->nop(); 2446 } 2447 2448 // overflow in loop, pass branch bytecode 2449 set(6,Rtmp); 2450 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp); 2451 2452 // Was an OSR adapter generated? 2453 // O0 = osr nmethod 2454 br_null_short(O0, Assembler::pn, overflow_with_error); 2455 2456 // Has the nmethod been invalidated already? 2457 ldub(O0, nmethod::state_offset(), O2); 2458 cmp_and_br_short(O2, nmethod::in_use, Assembler::notEqual, Assembler::pn, overflow_with_error); 2459 2460 // migrate the interpreter frame off of the stack 2461 2462 mov(G2_thread, L7); 2463 // save nmethod 2464 mov(O0, L6); 2465 set_last_Java_frame(SP, noreg); 2466 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7); 2467 reset_last_Java_frame(); 2468 mov(L7, G2_thread); 2469 2470 // move OSR nmethod to I1 2471 mov(L6, I1); 2472 2473 // OSR buffer to I0 2474 mov(O0, I0); 2475 2476 // remove the interpreter frame 2477 restore(I5_savedSP, 0, SP); 2478 2479 // Jump to the osr code. 2480 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2); 2481 jmp(O2, G0); 2482 delayed()->nop(); 2483 2484 bind(overflow_with_error); 2485 2486 bind(did_not_overflow); 2487 } 2488 2489 2490 2491 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) { 2492 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); } 2493 } 2494 2495 2496 // local helper function for the verify_oop_or_return_address macro 2497 static bool verify_return_address(Method* m, int bci) { 2498 #ifndef PRODUCT 2499 address pc = (address)(m->constMethod()) 2500 + in_bytes(ConstMethod::codes_offset()) + bci; 2501 // assume it is a valid return address if it is inside m and is preceded by a jsr 2502 if (!m->contains(pc)) return false; 2503 address jsr_pc; 2504 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2505 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2506 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2507 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2508 #endif // PRODUCT 2509 return false; 2510 } 2511 2512 2513 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2514 if (!VerifyOops) return; 2515 // the VM documentation for the astore[_wide] bytecode allows 2516 // the TOS to be not only an oop but also a return address 2517 Label test; 2518 Label skip; 2519 // See if it is an address (in the current method): 2520 2521 mov(reg, Rtmp); 2522 const int log2_bytecode_size_limit = 16; 2523 srl(Rtmp, log2_bytecode_size_limit, Rtmp); 2524 br_notnull_short( Rtmp, pt, test ); 2525 2526 // %%% should use call_VM_leaf here? 2527 save_frame_and_mov(0, Lmethod, O0, reg, O1); 2528 save_thread(L7_thread_cache); 2529 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none); 2530 delayed()->nop(); 2531 restore_thread(L7_thread_cache); 2532 br_notnull( O0, false, pt, skip ); 2533 delayed()->restore(); 2534 2535 // Perform a more elaborate out-of-line call 2536 // Not an address; verify it: 2537 bind(test); 2538 verify_oop(reg); 2539 bind(skip); 2540 } 2541 2542 2543 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2544 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); 2545 } 2546 2547 2548 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 2549 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 2550 int increment, Address mask_addr, 2551 Register scratch1, Register scratch2, 2552 Condition cond, Label *where) { 2553 ld(counter_addr, scratch1); 2554 add(scratch1, increment, scratch1); 2555 ld(mask_addr, scratch2); 2556 andcc(scratch1, scratch2, G0); 2557 br(cond, false, Assembler::pn, *where); 2558 delayed()->st(scratch1, counter_addr); 2559 } 2560 2561 // Inline assembly for: 2562 // 2563 // if (thread is in interp_only_mode) { 2564 // InterpreterRuntime::post_method_entry(); 2565 // } 2566 // if (DTraceMethodProbes) { 2567 // SharedRuntime::dtrace_method_entry(method, receiver); 2568 // } 2569 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 2570 // SharedRuntime::rc_trace_method_entry(method, receiver); 2571 // } 2572 2573 void InterpreterMacroAssembler::notify_method_entry() { 2574 2575 // Whenever JVMTI puts a thread in interp_only_mode, method 2576 // entry/exit events are sent for that thread to track stack 2577 // depth. If it is possible to enter interp_only_mode we add 2578 // the code to check if the event should be sent. 2579 if (JvmtiExport::can_post_interpreter_events()) { 2580 Label L; 2581 Register temp_reg = O5; 2582 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2583 ld(interp_only, temp_reg); 2584 cmp_and_br_short(temp_reg, 0, equal, pt, L); 2585 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); 2586 bind(L); 2587 } 2588 2589 { 2590 Register temp_reg = O5; 2591 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2592 call_VM_leaf(noreg, 2593 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 2594 G2_thread, Lmethod); 2595 } 2596 2597 // RedefineClasses() tracing support for obsolete method entry 2598 if (log_is_enabled(Trace, redefine, class, obsolete)) { 2599 call_VM_leaf(noreg, 2600 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 2601 G2_thread, Lmethod); 2602 } 2603 } 2604 2605 2606 // Inline assembly for: 2607 // 2608 // if (thread is in interp_only_mode) { 2609 // // save result 2610 // InterpreterRuntime::post_method_exit(); 2611 // // restore result 2612 // } 2613 // if (DTraceMethodProbes) { 2614 // SharedRuntime::dtrace_method_exit(thread, method); 2615 // } 2616 // 2617 // Native methods have their result stored in d_tmp and l_tmp 2618 // Java methods have their result stored in the expression stack 2619 2620 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, 2621 TosState state, 2622 NotifyMethodExitMode mode) { 2623 2624 // Whenever JVMTI puts a thread in interp_only_mode, method 2625 // entry/exit events are sent for that thread to track stack 2626 // depth. If it is possible to enter interp_only_mode we add 2627 // the code to check if the event should be sent. 2628 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2629 Label L; 2630 Register temp_reg = O5; 2631 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2632 ld(interp_only, temp_reg); 2633 cmp_and_br_short(temp_reg, 0, equal, pt, L); 2634 2635 // Note: frame::interpreter_frame_result has a dependency on how the 2636 // method result is saved across the call to post_method_exit. For 2637 // native methods it assumes the result registers are saved to 2638 // l_scratch and d_scratch. If this changes then the interpreter_frame_result 2639 // implementation will need to be updated too. 2640 2641 save_return_value(state, is_native_method); 2642 call_VM(noreg, 2643 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2644 restore_return_value(state, is_native_method); 2645 bind(L); 2646 } 2647 2648 { 2649 Register temp_reg = O5; 2650 // Dtrace notification 2651 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2652 save_return_value(state, is_native_method); 2653 call_VM_leaf( 2654 noreg, 2655 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2656 G2_thread, Lmethod); 2657 restore_return_value(state, is_native_method); 2658 } 2659 } 2660 2661 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) { 2662 if (is_native_call) { 2663 stf(FloatRegisterImpl::D, F0, d_tmp); 2664 stx(O0, l_tmp); 2665 } else { 2666 push(state); 2667 } 2668 } 2669 2670 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) { 2671 if (is_native_call) { 2672 ldf(FloatRegisterImpl::D, d_tmp, F0); 2673 ldx(l_tmp, O0); 2674 } else { 2675 pop(state); 2676 } 2677 }