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, Register tmp) { 743 assert_different_registers(result, index, tmp); 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 sll(index, LogBytesPerHeapOop, index); 748 get_constant_pool(result); 749 // load pointer for resolved_references[] objArray 750 ld_ptr(result, ConstantPool::cache_offset_in_bytes(), result); 751 ld_ptr(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result); 752 resolve_oop_handle(result, tmp); 753 // Add in the index 754 add(result, index, result); 755 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result, tmp); 756 // The resulting oop is null if the reference is not yet resolved. 757 // It is Universe::the_null_sentinel() if the reference resolved to NULL via condy. 758 } 759 760 761 // load cpool->resolved_klass_at(index) 762 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool, 763 Register Roffset, Register Rklass) { 764 // int value = *this_cp->int_at_addr(which); 765 // int resolved_klass_index = extract_low_short_from_int(value); 766 // 767 // Because SPARC is big-endian, the low_short is at (cpool->int_at_addr(which) + 2 bytes) 768 add(Roffset, Rcpool, Roffset); 769 lduh(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index 770 771 Register Rresolved_klasses = Rklass; 772 ld_ptr(Rcpool, ConstantPool::resolved_klasses_offset_in_bytes(), Rresolved_klasses); 773 sll(Roffset, LogBytesPerWord, Roffset); 774 add(Roffset, Array<Klass*>::base_offset_in_bytes(), Roffset); 775 ld_ptr(Rresolved_klasses, Roffset, Rklass); 776 } 777 778 779 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 780 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. 781 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 782 Register Rsuper_klass, 783 Register Rtmp1, 784 Register Rtmp2, 785 Register Rtmp3, 786 Label &ok_is_subtype ) { 787 Label not_subtype; 788 789 // Profile the not-null value's klass. 790 profile_typecheck(Rsub_klass, Rtmp1); 791 792 check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass, 793 Rtmp1, Rtmp2, 794 &ok_is_subtype, ¬_subtype, NULL); 795 796 check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass, 797 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg, 798 &ok_is_subtype, NULL); 799 800 bind(not_subtype); 801 profile_typecheck_failed(Rtmp1); 802 } 803 804 // Separate these two to allow for delay slot in middle 805 // These are used to do a test and full jump to exception-throwing code. 806 807 // %%%%% Could possibly reoptimize this by testing to see if could use 808 // a single conditional branch (i.e. if span is small enough. 809 // If you go that route, than get rid of the split and give up 810 // on the delay-slot hack. 811 812 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition, 813 Label& ok ) { 814 assert_not_delayed(); 815 br(ok_condition, true, pt, ok); 816 // DELAY SLOT 817 } 818 819 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition, 820 Label& ok ) { 821 assert_not_delayed(); 822 bp( ok_condition, true, Assembler::xcc, pt, ok); 823 // DELAY SLOT 824 } 825 826 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition, 827 Label& ok ) { 828 assert_not_delayed(); 829 brx(ok_condition, true, pt, ok); 830 // DELAY SLOT 831 } 832 833 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point, 834 Register Rscratch, 835 Label& ok ) { 836 assert(throw_entry_point != NULL, "entry point must be generated by now"); 837 AddressLiteral dest(throw_entry_point); 838 jump_to(dest, Rscratch); 839 delayed()->nop(); 840 bind(ok); 841 } 842 843 844 // And if you cannot use the delay slot, here is a shorthand: 845 846 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition, 847 address throw_entry_point, 848 Register Rscratch ) { 849 Label ok; 850 if (ok_condition != never) { 851 throw_if_not_1_icc( ok_condition, ok); 852 delayed()->nop(); 853 } 854 throw_if_not_2( throw_entry_point, Rscratch, ok); 855 } 856 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition, 857 address throw_entry_point, 858 Register Rscratch ) { 859 Label ok; 860 if (ok_condition != never) { 861 throw_if_not_1_xcc( ok_condition, ok); 862 delayed()->nop(); 863 } 864 throw_if_not_2( throw_entry_point, Rscratch, ok); 865 } 866 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition, 867 address throw_entry_point, 868 Register Rscratch ) { 869 Label ok; 870 if (ok_condition != never) { 871 throw_if_not_1_x( ok_condition, ok); 872 delayed()->nop(); 873 } 874 throw_if_not_2( throw_entry_point, Rscratch, ok); 875 } 876 877 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res 878 // Note: res is still shy of address by array offset into object. 879 880 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) { 881 assert_not_delayed(); 882 883 verify_oop(array); 884 // Sign extend since tos (index) can be a 32bit value. 885 sra(index, G0, index); 886 887 // Check array. 888 Label ptr_ok; 889 tst(array); 890 throw_if_not_1_x(notZero, ptr_ok); 891 delayed()->ld(array, arrayOopDesc::length_offset_in_bytes(), tmp); // Check index. 892 throw_if_not_2(Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok); 893 894 Label index_ok; 895 cmp(index, tmp); 896 throw_if_not_1_icc(lessUnsigned, index_ok); 897 if (index_shift > 0) { 898 delayed()->sll(index, index_shift, index); 899 } else { 900 delayed()->add(array, index, res); // addr - const offset in index 901 } 902 // Pass the array to create more detailed exceptions. 903 // Convention: move aberrant index into Otos_i for exception message. 904 mov(index, Otos_i); 905 mov(array, G3_scratch); 906 throw_if_not_2(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok); 907 908 // add offset if didn't do it in delay slot 909 if (index_shift > 0) { add(array, index, res); } // addr - const offset in index 910 } 911 912 913 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { 914 assert_not_delayed(); 915 916 // pop array 917 pop_ptr(array); 918 919 // check array 920 index_check_without_pop(array, index, index_shift, tmp, res); 921 } 922 923 924 void InterpreterMacroAssembler::get_const(Register Rdst) { 925 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst); 926 } 927 928 929 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 930 get_const(Rdst); 931 ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 932 } 933 934 935 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 936 get_constant_pool(Rdst); 937 ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 938 } 939 940 941 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 942 get_constant_pool(Rcpool); 943 ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags); 944 } 945 946 947 // unlock if synchronized method 948 // 949 // Unlock the receiver if this is a synchronized method. 950 // Unlock any Java monitors from syncronized blocks. 951 // 952 // If there are locked Java monitors 953 // If throw_monitor_exception 954 // throws IllegalMonitorStateException 955 // Else if install_monitor_exception 956 // installs IllegalMonitorStateException 957 // Else 958 // no error processing 959 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 960 bool throw_monitor_exception, 961 bool install_monitor_exception) { 962 Label unlocked, unlock, no_unlock; 963 964 // get the value of _do_not_unlock_if_synchronized into G1_scratch 965 const Address do_not_unlock_if_synchronized(G2_thread, 966 JavaThread::do_not_unlock_if_synchronized_offset()); 967 ldbool(do_not_unlock_if_synchronized, G1_scratch); 968 stbool(G0, do_not_unlock_if_synchronized); // reset the flag 969 970 // check if synchronized method 971 const Address access_flags(Lmethod, Method::access_flags_offset()); 972 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 973 push(state); // save tos 974 ld(access_flags, G3_scratch); // Load access flags. 975 btst(JVM_ACC_SYNCHRONIZED, G3_scratch); 976 br(zero, false, pt, unlocked); 977 delayed()->nop(); 978 979 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 980 // is set. 981 cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock); 982 delayed()->nop(); 983 984 // BasicObjectLock will be first in list, since this is a synchronized method. However, need 985 // to check that the object has not been unlocked by an explicit monitorexit bytecode. 986 987 //Intel: if (throw_monitor_exception) ... else ... 988 // Entry already unlocked, need to throw exception 989 //... 990 991 // pass top-most monitor elem 992 add( top_most_monitor(), O1 ); 993 994 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch); 995 br_notnull_short(G3_scratch, pt, unlock); 996 997 if (throw_monitor_exception) { 998 // Entry already unlocked need to throw an exception 999 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1000 should_not_reach_here(); 1001 } else { 1002 // Monitor already unlocked during a stack unroll. 1003 // If requested, install an illegal_monitor_state_exception. 1004 // Continue with stack unrolling. 1005 if (install_monitor_exception) { 1006 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1007 } 1008 ba_short(unlocked); 1009 } 1010 1011 bind(unlock); 1012 1013 unlock_object(O1); 1014 1015 bind(unlocked); 1016 1017 // I0, I1: Might contain return value 1018 1019 // Check that all monitors are unlocked 1020 { Label loop, exception, entry, restart; 1021 1022 Register Rmptr = O0; 1023 Register Rtemp = O1; 1024 Register Rlimit = Lmonitors; 1025 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 1026 assert( (delta & LongAlignmentMask) == 0, 1027 "sizeof BasicObjectLock must be even number of doublewords"); 1028 1029 #ifdef ASSERT 1030 add(top_most_monitor(), Rmptr, delta); 1031 { Label L; 1032 // ensure that Rmptr starts out above (or at) Rlimit 1033 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); 1034 stop("monitor stack has negative size"); 1035 bind(L); 1036 } 1037 #endif 1038 bind(restart); 1039 ba(entry); 1040 delayed()-> 1041 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry 1042 1043 // Entry is still locked, need to throw exception 1044 bind(exception); 1045 if (throw_monitor_exception) { 1046 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1047 should_not_reach_here(); 1048 } else { 1049 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 1050 // Unlock does not block, so don't have to worry about the frame 1051 unlock_object(Rmptr); 1052 if (install_monitor_exception) { 1053 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1054 } 1055 ba_short(restart); 1056 } 1057 1058 bind(loop); 1059 cmp(Rtemp, G0); // check if current entry is used 1060 brx(Assembler::notEqual, false, pn, exception); 1061 delayed()-> 1062 dec(Rmptr, delta); // otherwise advance to next entry 1063 #ifdef ASSERT 1064 { Label L; 1065 // ensure that Rmptr has not somehow stepped below Rlimit 1066 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); 1067 stop("ran off the end of the monitor stack"); 1068 bind(L); 1069 } 1070 #endif 1071 bind(entry); 1072 cmp(Rmptr, Rlimit); // check if bottom reached 1073 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry 1074 delayed()-> 1075 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp); 1076 } 1077 1078 bind(no_unlock); 1079 pop(state); 1080 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1081 } 1082 1083 void InterpreterMacroAssembler::narrow(Register result) { 1084 1085 ld_ptr(Address(Lmethod, Method::const_offset()), G3_scratch); 1086 ldub(G3_scratch, in_bytes(ConstMethod::result_type_offset()), G3_scratch); 1087 1088 Label notBool, notByte, notChar, done; 1089 1090 // common case first 1091 cmp(G3_scratch, T_INT); 1092 br(Assembler::equal, true, pn, done); 1093 delayed()->nop(); 1094 1095 cmp(G3_scratch, T_BOOLEAN); 1096 br(Assembler::notEqual, true, pn, notBool); 1097 delayed()->cmp(G3_scratch, T_BYTE); 1098 and3(result, 1, result); 1099 ba(done); 1100 delayed()->nop(); 1101 1102 bind(notBool); 1103 // cmp(G3_scratch, T_BYTE); 1104 br(Assembler::notEqual, true, pn, notByte); 1105 delayed()->cmp(G3_scratch, T_CHAR); 1106 sll(result, 24, result); 1107 sra(result, 24, result); 1108 ba(done); 1109 delayed()->nop(); 1110 1111 bind(notByte); 1112 // cmp(G3_scratch, T_CHAR); 1113 sll(result, 16, result); 1114 br(Assembler::notEqual, true, pn, done); 1115 delayed()->sra(result, 16, result); 1116 // sll(result, 16, result); 1117 srl(result, 16, result); 1118 1119 // bind(notChar); 1120 // must be short, instructions already executed in delay slot 1121 // sll(result, 16, result); 1122 // sra(result, 16, result); 1123 1124 bind(done); 1125 } 1126 1127 // remove activation 1128 // 1129 // Unlock the receiver if this is a synchronized method. 1130 // Unlock any Java monitors from syncronized blocks. 1131 // Remove the activation from the stack. 1132 // 1133 // If there are locked Java monitors 1134 // If throw_monitor_exception 1135 // throws IllegalMonitorStateException 1136 // Else if install_monitor_exception 1137 // installs IllegalMonitorStateException 1138 // Else 1139 // no error processing 1140 void InterpreterMacroAssembler::remove_activation(TosState state, 1141 bool throw_monitor_exception, 1142 bool install_monitor_exception) { 1143 1144 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 1145 1146 // save result (push state before jvmti call and pop it afterwards) and notify jvmti 1147 notify_method_exit(false, state, NotifyJVMTI); 1148 1149 if (StackReservedPages > 0) { 1150 // testing if Stack Reserved Area needs to be re-enabled 1151 Label no_reserved_zone_enabling; 1152 ld_ptr(G2_thread, JavaThread::reserved_stack_activation_offset(), G3_scratch); 1153 cmp_and_brx_short(SP, G3_scratch, Assembler::lessUnsigned, Assembler::pt, no_reserved_zone_enabling); 1154 1155 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), G2_thread); 1156 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError), G2_thread); 1157 should_not_reach_here(); 1158 1159 bind(no_reserved_zone_enabling); 1160 } 1161 1162 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1163 verify_thread(); 1164 1165 // return tos 1166 assert(Otos_l1 == Otos_i, "adjust code below"); 1167 switch (state) { 1168 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0 1169 case btos: // fall through 1170 case ztos: // fall through 1171 case ctos: 1172 case stos: // fall through 1173 case atos: // fall through 1174 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0 1175 case ftos: // fall through 1176 case dtos: // fall through 1177 case vtos: /* nothing to do */ break; 1178 default : ShouldNotReachHere(); 1179 } 1180 } 1181 1182 // Lock object 1183 // 1184 // Argument - lock_reg points to the BasicObjectLock to be used for locking, 1185 // it must be initialized with the object to lock 1186 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) { 1187 if (UseHeavyMonitors) { 1188 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1189 } 1190 else { 1191 Register obj_reg = Object; 1192 Register mark_reg = G4_scratch; 1193 Register temp_reg = G1_scratch; 1194 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes()); 1195 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1196 Label done; 1197 1198 Label slow_case; 1199 1200 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg); 1201 1202 // load markOop from object into mark_reg 1203 ld_ptr(mark_addr, mark_reg); 1204 1205 if (UseBiasedLocking) { 1206 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case); 1207 } 1208 1209 // get the address of basicLock on stack that will be stored in the object 1210 // we need a temporary register here as we do not want to clobber lock_reg 1211 // (cas clobbers the destination register) 1212 mov(lock_reg, temp_reg); 1213 // set mark reg to be (markOop of object | UNLOCK_VALUE) 1214 or3(mark_reg, markOopDesc::unlocked_value, mark_reg); 1215 // initialize the box (Must happen before we update the object mark!) 1216 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1217 // compare and exchange object_addr, markOop | 1, stack address of basicLock 1218 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1219 cas_ptr(mark_addr.base(), mark_reg, temp_reg); 1220 1221 // if the compare and exchange succeeded we are done (we saw an unlocked object) 1222 cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done); 1223 1224 // We did not see an unlocked object so try the fast recursive case 1225 1226 // Check if owner is self by comparing the value in the markOop of object 1227 // with the stack pointer 1228 sub(temp_reg, SP, temp_reg); 1229 sub(temp_reg, STACK_BIAS, temp_reg); 1230 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 1231 1232 // Composite "andcc" test: 1233 // (a) %sp -vs- markword proximity check, and, 1234 // (b) verify mark word LSBs == 0 (Stack-locked). 1235 // 1236 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size()) 1237 // Note that the page size used for %sp proximity testing is arbitrary and is 1238 // unrelated to the actual MMU page size. We use a 'logical' page size of 1239 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate 1240 // field of the andcc instruction. 1241 andcc (temp_reg, 0xFFFFF003, G0) ; 1242 1243 // if condition is true we are done and hence we can store 0 in the displaced 1244 // header indicating it is a recursive lock and be done 1245 brx(Assembler::zero, true, Assembler::pt, done); 1246 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1247 1248 // none of the above fast optimizations worked so we have to get into the 1249 // slow case of monitor enter 1250 bind(slow_case); 1251 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1252 1253 bind(done); 1254 } 1255 } 1256 1257 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 1258 // 1259 // Argument - lock_reg points to the BasicObjectLock for lock 1260 // Throw IllegalMonitorException if object is not locked by current thread 1261 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1262 if (UseHeavyMonitors) { 1263 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1264 } else { 1265 Register obj_reg = G3_scratch; 1266 Register mark_reg = G4_scratch; 1267 Register displaced_header_reg = G1_scratch; 1268 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes()); 1269 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1270 Label done; 1271 1272 if (UseBiasedLocking) { 1273 // load the object out of the BasicObjectLock 1274 ld_ptr(lockobj_addr, obj_reg); 1275 biased_locking_exit(mark_addr, mark_reg, done, true); 1276 st_ptr(G0, lockobj_addr); // free entry 1277 } 1278 1279 // Test first if we are in the fast recursive case 1280 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes()); 1281 ld_ptr(lock_addr, displaced_header_reg); 1282 br_null(displaced_header_reg, true, Assembler::pn, done); 1283 delayed()->st_ptr(G0, lockobj_addr); // free entry 1284 1285 // See if it is still a light weight lock, if so we just unlock 1286 // the object and we are done 1287 1288 if (!UseBiasedLocking) { 1289 // load the object out of the BasicObjectLock 1290 ld_ptr(lockobj_addr, obj_reg); 1291 } 1292 1293 // we have the displaced header in displaced_header_reg 1294 // we expect to see the stack address of the basicLock in case the 1295 // lock is still a light weight lock (lock_reg) 1296 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1297 cas_ptr(mark_addr.base(), lock_reg, displaced_header_reg); 1298 cmp(lock_reg, displaced_header_reg); 1299 brx(Assembler::equal, true, Assembler::pn, done); 1300 delayed()->st_ptr(G0, lockobj_addr); // free entry 1301 1302 // The lock has been converted into a heavy lock and hence 1303 // we need to get into the slow case 1304 1305 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1306 1307 bind(done); 1308 } 1309 } 1310 1311 // Get the method data pointer from the Method* and set the 1312 // specified register to its value. 1313 1314 void InterpreterMacroAssembler::set_method_data_pointer() { 1315 assert(ProfileInterpreter, "must be profiling interpreter"); 1316 Label get_continue; 1317 1318 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); 1319 test_method_data_pointer(get_continue); 1320 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); 1321 bind(get_continue); 1322 } 1323 1324 // Set the method data pointer for the current bcp. 1325 1326 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1327 assert(ProfileInterpreter, "must be profiling interpreter"); 1328 Label zero_continue; 1329 1330 // Test MDO to avoid the call if it is NULL. 1331 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); 1332 test_method_data_pointer(zero_continue); 1333 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp); 1334 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); 1335 add(ImethodDataPtr, O0, ImethodDataPtr); 1336 bind(zero_continue); 1337 } 1338 1339 // Test ImethodDataPtr. If it is null, continue at the specified label 1340 1341 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1342 assert(ProfileInterpreter, "must be profiling interpreter"); 1343 br_null_short(ImethodDataPtr, Assembler::pn, zero_continue); 1344 } 1345 1346 void InterpreterMacroAssembler::verify_method_data_pointer() { 1347 assert(ProfileInterpreter, "must be profiling interpreter"); 1348 #ifdef ASSERT 1349 Label verify_continue; 1350 test_method_data_pointer(verify_continue); 1351 1352 // If the mdp is valid, it will point to a DataLayout header which is 1353 // consistent with the bcp. The converse is highly probable also. 1354 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch); 1355 ld_ptr(Lmethod, Method::const_offset(), O5); 1356 add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch); 1357 add(G3_scratch, O5, G3_scratch); 1358 cmp(Lbcp, G3_scratch); 1359 brx(Assembler::equal, false, Assembler::pt, verify_continue); 1360 1361 Register temp_reg = O5; 1362 delayed()->mov(ImethodDataPtr, temp_reg); 1363 // %%% should use call_VM_leaf here? 1364 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr); 1365 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1); 1366 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS); 1367 stf(FloatRegisterImpl::D, Ftos_d, d_save); 1368 mov(temp_reg->after_save(), O2); 1369 save_thread(L7_thread_cache); 1370 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none); 1371 delayed()->nop(); 1372 restore_thread(L7_thread_cache); 1373 ldf(FloatRegisterImpl::D, d_save, Ftos_d); 1374 restore(); 1375 bind(verify_continue); 1376 #endif // ASSERT 1377 } 1378 1379 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1380 Register method_counters, 1381 Register Rtmp, 1382 Label &profile_continue) { 1383 assert(ProfileInterpreter, "must be profiling interpreter"); 1384 // Control will flow to "profile_continue" if the counter is less than the 1385 // limit or if we call profile_method() 1386 1387 Label done; 1388 1389 // if no method data exists, and the counter is high enough, make one 1390 br_notnull_short(ImethodDataPtr, Assembler::pn, done); 1391 1392 // Test to see if we should create a method data oop 1393 Address profile_limit(method_counters, MethodCounters::interpreter_profile_limit_offset()); 1394 ld(profile_limit, Rtmp); 1395 cmp(invocation_count, Rtmp); 1396 // Use long branches because call_VM() code and following code generated by 1397 // test_backedge_count_for_osr() is large in debug VM. 1398 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue); 1399 delayed()->nop(); 1400 1401 // Build it now. 1402 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1403 set_method_data_pointer_for_bcp(); 1404 ba(profile_continue); 1405 delayed()->nop(); 1406 bind(done); 1407 } 1408 1409 // Store a value at some constant offset from the method data pointer. 1410 1411 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1412 assert(ProfileInterpreter, "must be profiling interpreter"); 1413 st_ptr(value, ImethodDataPtr, constant); 1414 } 1415 1416 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter, 1417 Register bumped_count, 1418 bool decrement) { 1419 assert(ProfileInterpreter, "must be profiling interpreter"); 1420 1421 // Load the counter. 1422 ld_ptr(counter, bumped_count); 1423 1424 if (decrement) { 1425 // Decrement the register. Set condition codes. 1426 subcc(bumped_count, DataLayout::counter_increment, bumped_count); 1427 1428 // If the decrement causes the counter to overflow, stay negative 1429 Label L; 1430 brx(Assembler::negative, true, Assembler::pn, L); 1431 1432 // Store the decremented counter, if it is still negative. 1433 delayed()->st_ptr(bumped_count, counter); 1434 bind(L); 1435 } else { 1436 // Increment the register. Set carry flag. 1437 addcc(bumped_count, DataLayout::counter_increment, bumped_count); 1438 1439 // If the increment causes the counter to overflow, pull back by 1. 1440 assert(DataLayout::counter_increment == 1, "subc works"); 1441 subc(bumped_count, G0, bumped_count); 1442 1443 // Store the incremented counter. 1444 st_ptr(bumped_count, counter); 1445 } 1446 } 1447 1448 // Increment the value at some constant offset from the method data pointer. 1449 1450 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1451 Register bumped_count, 1452 bool decrement) { 1453 // Locate the counter at a fixed offset from the mdp: 1454 Address counter(ImethodDataPtr, constant); 1455 increment_mdp_data_at(counter, bumped_count, decrement); 1456 } 1457 1458 // Increment the value at some non-fixed (reg + constant) offset from 1459 // the method data pointer. 1460 1461 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1462 int constant, 1463 Register bumped_count, 1464 Register scratch2, 1465 bool decrement) { 1466 // Add the constant to reg to get the offset. 1467 add(ImethodDataPtr, reg, scratch2); 1468 Address counter(scratch2, constant); 1469 increment_mdp_data_at(counter, bumped_count, decrement); 1470 } 1471 1472 // Set a flag value at the current method data pointer position. 1473 // Updates a single byte of the header, to avoid races with other header bits. 1474 1475 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1476 Register scratch) { 1477 assert(ProfileInterpreter, "must be profiling interpreter"); 1478 // Load the data header 1479 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch); 1480 1481 // Set the flag 1482 or3(scratch, flag_constant, scratch); 1483 1484 // Store the modified header. 1485 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset())); 1486 } 1487 1488 // Test the location at some offset from the method data pointer. 1489 // If it is not equal to value, branch to the not_equal_continue Label. 1490 // Set condition codes to match the nullness of the loaded value. 1491 1492 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1493 Register value, 1494 Label& not_equal_continue, 1495 Register scratch) { 1496 assert(ProfileInterpreter, "must be profiling interpreter"); 1497 ld_ptr(ImethodDataPtr, offset, scratch); 1498 cmp(value, scratch); 1499 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue); 1500 delayed()->tst(scratch); 1501 } 1502 1503 // Update the method data pointer by the displacement located at some fixed 1504 // offset from the method data pointer. 1505 1506 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1507 Register scratch) { 1508 assert(ProfileInterpreter, "must be profiling interpreter"); 1509 ld_ptr(ImethodDataPtr, offset_of_disp, scratch); 1510 add(ImethodDataPtr, scratch, ImethodDataPtr); 1511 } 1512 1513 // Update the method data pointer by the displacement located at the 1514 // offset (reg + offset_of_disp). 1515 1516 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1517 int offset_of_disp, 1518 Register scratch) { 1519 assert(ProfileInterpreter, "must be profiling interpreter"); 1520 add(reg, offset_of_disp, scratch); 1521 ld_ptr(ImethodDataPtr, scratch, scratch); 1522 add(ImethodDataPtr, scratch, ImethodDataPtr); 1523 } 1524 1525 // Update the method data pointer by a simple constant displacement. 1526 1527 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1528 assert(ProfileInterpreter, "must be profiling interpreter"); 1529 add(ImethodDataPtr, constant, ImethodDataPtr); 1530 } 1531 1532 // Update the method data pointer for a _ret bytecode whose target 1533 // was not among our cached targets. 1534 1535 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1536 Register return_bci) { 1537 assert(ProfileInterpreter, "must be profiling interpreter"); 1538 push(state); 1539 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile 1540 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1541 ld_ptr(l_tmp, return_bci); 1542 pop(state); 1543 } 1544 1545 // Count a taken branch in the bytecodes. 1546 1547 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1548 if (ProfileInterpreter) { 1549 Label profile_continue; 1550 1551 // If no method data exists, go to profile_continue. 1552 test_method_data_pointer(profile_continue); 1553 1554 // We are taking a branch. Increment the taken count. 1555 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count); 1556 1557 // The method data pointer needs to be updated to reflect the new target. 1558 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1559 bind (profile_continue); 1560 } 1561 } 1562 1563 1564 // Count a not-taken branch in the bytecodes. 1565 1566 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) { 1567 if (ProfileInterpreter) { 1568 Label profile_continue; 1569 1570 // If no method data exists, go to profile_continue. 1571 test_method_data_pointer(profile_continue); 1572 1573 // We are taking a branch. Increment the not taken count. 1574 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch); 1575 1576 // The method data pointer needs to be updated to correspond to the 1577 // next bytecode. 1578 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1579 bind (profile_continue); 1580 } 1581 } 1582 1583 1584 // Count a non-virtual call in the bytecodes. 1585 1586 void InterpreterMacroAssembler::profile_call(Register scratch) { 1587 if (ProfileInterpreter) { 1588 Label profile_continue; 1589 1590 // If no method data exists, go to profile_continue. 1591 test_method_data_pointer(profile_continue); 1592 1593 // We are making a call. Increment the count. 1594 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1595 1596 // The method data pointer needs to be updated to reflect the new target. 1597 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1598 bind (profile_continue); 1599 } 1600 } 1601 1602 1603 // Count a final call in the bytecodes. 1604 1605 void InterpreterMacroAssembler::profile_final_call(Register scratch) { 1606 if (ProfileInterpreter) { 1607 Label profile_continue; 1608 1609 // If no method data exists, go to profile_continue. 1610 test_method_data_pointer(profile_continue); 1611 1612 // We are making a call. Increment the count. 1613 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1614 1615 // The method data pointer needs to be updated to reflect the new target. 1616 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1617 bind (profile_continue); 1618 } 1619 } 1620 1621 1622 // Count a virtual call in the bytecodes. 1623 1624 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1625 Register scratch, 1626 bool receiver_can_be_null) { 1627 if (ProfileInterpreter) { 1628 Label profile_continue; 1629 1630 // If no method data exists, go to profile_continue. 1631 test_method_data_pointer(profile_continue); 1632 1633 1634 Label skip_receiver_profile; 1635 if (receiver_can_be_null) { 1636 Label not_null; 1637 br_notnull_short(receiver, Assembler::pt, not_null); 1638 // We are making a call. Increment the count for null receiver. 1639 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1640 ba_short(skip_receiver_profile); 1641 bind(not_null); 1642 } 1643 1644 // Record the receiver type. 1645 record_klass_in_profile(receiver, scratch, true); 1646 bind(skip_receiver_profile); 1647 1648 // The method data pointer needs to be updated to reflect the new target. 1649 #if INCLUDE_JVMCI 1650 if (MethodProfileWidth == 0) { 1651 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1652 } 1653 #else 1654 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1655 #endif 1656 bind(profile_continue); 1657 } 1658 } 1659 1660 #if INCLUDE_JVMCI 1661 void InterpreterMacroAssembler::profile_called_method(Register method, Register scratch) { 1662 assert_different_registers(method, scratch); 1663 if (ProfileInterpreter && MethodProfileWidth > 0) { 1664 Label profile_continue; 1665 1666 // If no method data exists, go to profile_continue. 1667 test_method_data_pointer(profile_continue); 1668 1669 Label done; 1670 record_item_in_profile_helper(method, scratch, 0, done, MethodProfileWidth, 1671 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset())); 1672 bind(done); 1673 1674 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1675 bind(profile_continue); 1676 } 1677 } 1678 #endif // INCLUDE_JVMCI 1679 1680 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register scratch, 1681 Label& done, bool is_virtual_call) { 1682 if (TypeProfileWidth == 0) { 1683 if (is_virtual_call) { 1684 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1685 } 1686 #if INCLUDE_JVMCI 1687 else if (EnableJVMCI) { 1688 increment_mdp_data_at(in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()), scratch); 1689 } 1690 #endif 1691 } else { 1692 int non_profiled_offset = -1; 1693 if (is_virtual_call) { 1694 non_profiled_offset = in_bytes(CounterData::count_offset()); 1695 } 1696 #if INCLUDE_JVMCI 1697 else if (EnableJVMCI) { 1698 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()); 1699 } 1700 #endif 1701 1702 record_item_in_profile_helper(receiver, scratch, 0, done, TypeProfileWidth, 1703 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset); 1704 } 1705 } 1706 1707 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, 1708 Register scratch, int start_row, Label& done, int total_rows, 1709 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, 1710 int non_profiled_offset) { 1711 int last_row = total_rows - 1; 1712 assert(start_row <= last_row, "must be work left to do"); 1713 // Test this row for both the item and for null. 1714 // Take any of three different outcomes: 1715 // 1. found item => increment count and goto done 1716 // 2. found null => keep looking for case 1, maybe allocate this cell 1717 // 3. found something else => keep looking for cases 1 and 2 1718 // Case 3 is handled by a recursive call. 1719 for (int row = start_row; row <= last_row; row++) { 1720 Label next_test; 1721 bool test_for_null_also = (row == start_row); 1722 1723 // See if the item is item[n]. 1724 int item_offset = in_bytes(item_offset_fn(row)); 1725 test_mdp_data_at(item_offset, item, next_test, scratch); 1726 // delayed()->tst(scratch); 1727 1728 // The receiver is item[n]. Increment count[n]. 1729 int count_offset = in_bytes(item_count_offset_fn(row)); 1730 increment_mdp_data_at(count_offset, scratch); 1731 ba_short(done); 1732 bind(next_test); 1733 1734 if (test_for_null_also) { 1735 Label found_null; 1736 // Failed the equality check on item[n]... Test for null. 1737 if (start_row == last_row) { 1738 // The only thing left to do is handle the null case. 1739 if (non_profiled_offset >= 0) { 1740 brx(Assembler::zero, false, Assembler::pn, found_null); 1741 delayed()->nop(); 1742 // Item did not match any saved item and there is no empty row for it. 1743 // Increment total counter to indicate polymorphic case. 1744 increment_mdp_data_at(non_profiled_offset, scratch); 1745 ba_short(done); 1746 bind(found_null); 1747 } else { 1748 brx(Assembler::notZero, false, Assembler::pt, done); 1749 delayed()->nop(); 1750 } 1751 break; 1752 } 1753 // Since null is rare, make it be the branch-taken case. 1754 brx(Assembler::zero, false, Assembler::pn, found_null); 1755 delayed()->nop(); 1756 1757 // Put all the "Case 3" tests here. 1758 record_item_in_profile_helper(item, scratch, start_row + 1, done, total_rows, 1759 item_offset_fn, item_count_offset_fn, non_profiled_offset); 1760 1761 // Found a null. Keep searching for a matching item, 1762 // but remember that this is an empty (unused) slot. 1763 bind(found_null); 1764 } 1765 } 1766 1767 // In the fall-through case, we found no matching item, but we 1768 // observed the item[start_row] is NULL. 1769 1770 // Fill in the item field and increment the count. 1771 int item_offset = in_bytes(item_offset_fn(start_row)); 1772 set_mdp_data_at(item_offset, item); 1773 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1774 mov(DataLayout::counter_increment, scratch); 1775 set_mdp_data_at(count_offset, scratch); 1776 if (start_row > 0) { 1777 ba_short(done); 1778 } 1779 } 1780 1781 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1782 Register scratch, bool is_virtual_call) { 1783 assert(ProfileInterpreter, "must be profiling"); 1784 Label done; 1785 1786 record_klass_in_profile_helper(receiver, scratch, done, is_virtual_call); 1787 1788 bind (done); 1789 } 1790 1791 1792 // Count a ret in the bytecodes. 1793 1794 void InterpreterMacroAssembler::profile_ret(TosState state, 1795 Register return_bci, 1796 Register scratch) { 1797 if (ProfileInterpreter) { 1798 Label profile_continue; 1799 uint row; 1800 1801 // If no method data exists, go to profile_continue. 1802 test_method_data_pointer(profile_continue); 1803 1804 // Update the total ret count. 1805 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1806 1807 for (row = 0; row < RetData::row_limit(); row++) { 1808 Label next_test; 1809 1810 // See if return_bci is equal to bci[n]: 1811 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), 1812 return_bci, next_test, scratch); 1813 1814 // return_bci is equal to bci[n]. Increment the count. 1815 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch); 1816 1817 // The method data pointer needs to be updated to reflect the new target. 1818 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch); 1819 ba_short(profile_continue); 1820 bind(next_test); 1821 } 1822 1823 update_mdp_for_ret(state, return_bci); 1824 1825 bind (profile_continue); 1826 } 1827 } 1828 1829 // Profile an unexpected null in the bytecodes. 1830 void InterpreterMacroAssembler::profile_null_seen(Register scratch) { 1831 if (ProfileInterpreter) { 1832 Label profile_continue; 1833 1834 // If no method data exists, go to profile_continue. 1835 test_method_data_pointer(profile_continue); 1836 1837 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch); 1838 1839 // The method data pointer needs to be updated. 1840 int mdp_delta = in_bytes(BitData::bit_data_size()); 1841 if (TypeProfileCasts) { 1842 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size()); 1843 } 1844 update_mdp_by_constant(mdp_delta); 1845 1846 bind (profile_continue); 1847 } 1848 } 1849 1850 void InterpreterMacroAssembler::profile_typecheck(Register klass, 1851 Register scratch) { 1852 if (ProfileInterpreter) { 1853 Label profile_continue; 1854 1855 // If no method data exists, go to profile_continue. 1856 test_method_data_pointer(profile_continue); 1857 1858 int mdp_delta = in_bytes(BitData::bit_data_size()); 1859 if (TypeProfileCasts) { 1860 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size()); 1861 1862 // Record the object type. 1863 record_klass_in_profile(klass, scratch, false); 1864 } 1865 1866 // The method data pointer needs to be updated. 1867 update_mdp_by_constant(mdp_delta); 1868 1869 bind (profile_continue); 1870 } 1871 } 1872 1873 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) { 1874 if (ProfileInterpreter && TypeProfileCasts) { 1875 Label profile_continue; 1876 1877 // If no method data exists, go to profile_continue. 1878 test_method_data_pointer(profile_continue); 1879 1880 int count_offset = in_bytes(CounterData::count_offset()); 1881 // Back up the address, since we have already bumped the mdp. 1882 count_offset -= in_bytes(ReceiverTypeData::receiver_type_data_size()); 1883 1884 // *Decrement* the counter. We expect to see zero or small negatives. 1885 increment_mdp_data_at(count_offset, scratch, true); 1886 1887 bind (profile_continue); 1888 } 1889 } 1890 1891 // Count the default case of a switch construct. 1892 1893 void InterpreterMacroAssembler::profile_switch_default(Register scratch) { 1894 if (ProfileInterpreter) { 1895 Label profile_continue; 1896 1897 // If no method data exists, go to profile_continue. 1898 test_method_data_pointer(profile_continue); 1899 1900 // Update the default case count 1901 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1902 scratch); 1903 1904 // The method data pointer needs to be updated. 1905 update_mdp_by_offset( 1906 in_bytes(MultiBranchData::default_displacement_offset()), 1907 scratch); 1908 1909 bind (profile_continue); 1910 } 1911 } 1912 1913 // Count the index'th case of a switch construct. 1914 1915 void InterpreterMacroAssembler::profile_switch_case(Register index, 1916 Register scratch, 1917 Register scratch2, 1918 Register scratch3) { 1919 if (ProfileInterpreter) { 1920 Label profile_continue; 1921 1922 // If no method data exists, go to profile_continue. 1923 test_method_data_pointer(profile_continue); 1924 1925 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() 1926 set(in_bytes(MultiBranchData::per_case_size()), scratch); 1927 smul(index, scratch, scratch); 1928 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch); 1929 1930 // Update the case count 1931 increment_mdp_data_at(scratch, 1932 in_bytes(MultiBranchData::relative_count_offset()), 1933 scratch2, 1934 scratch3); 1935 1936 // The method data pointer needs to be updated. 1937 update_mdp_by_offset(scratch, 1938 in_bytes(MultiBranchData::relative_displacement_offset()), 1939 scratch2); 1940 1941 bind (profile_continue); 1942 } 1943 } 1944 1945 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) { 1946 Label not_null, do_nothing, do_update; 1947 1948 assert_different_registers(obj, mdo_addr.base(), tmp); 1949 1950 verify_oop(obj); 1951 1952 ld_ptr(mdo_addr, tmp); 1953 1954 br_notnull_short(obj, pt, not_null); 1955 or3(tmp, TypeEntries::null_seen, tmp); 1956 ba_short(do_update); 1957 1958 bind(not_null); 1959 load_klass(obj, obj); 1960 1961 xor3(obj, tmp, obj); 1962 btst(TypeEntries::type_klass_mask, obj); 1963 // klass seen before, nothing to do. The unknown bit may have been 1964 // set already but no need to check. 1965 brx(zero, false, pt, do_nothing); 1966 delayed()-> 1967 1968 btst(TypeEntries::type_unknown, obj); 1969 // already unknown. Nothing to do anymore. 1970 brx(notZero, false, pt, do_nothing); 1971 delayed()-> 1972 1973 btst(TypeEntries::type_mask, tmp); 1974 brx(zero, true, pt, do_update); 1975 // first time here. Set profile type. 1976 delayed()->or3(tmp, obj, tmp); 1977 1978 // different than before. Cannot keep accurate profile. 1979 or3(tmp, TypeEntries::type_unknown, tmp); 1980 1981 bind(do_update); 1982 // update profile 1983 st_ptr(tmp, mdo_addr); 1984 1985 bind(do_nothing); 1986 } 1987 1988 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) { 1989 if (!ProfileInterpreter) { 1990 return; 1991 } 1992 1993 assert_different_registers(callee, tmp1, tmp2, ImethodDataPtr); 1994 1995 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1996 Label profile_continue; 1997 1998 test_method_data_pointer(profile_continue); 1999 2000 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 2001 2002 ldub(ImethodDataPtr, in_bytes(DataLayout::tag_offset()) - off_to_start, tmp1); 2003 cmp_and_br_short(tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag, notEqual, pn, profile_continue); 2004 2005 if (MethodData::profile_arguments()) { 2006 Label done; 2007 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 2008 add(ImethodDataPtr, off_to_args, ImethodDataPtr); 2009 2010 for (int i = 0; i < TypeProfileArgsLimit; i++) { 2011 if (i > 0 || MethodData::profile_return()) { 2012 // If return value type is profiled we may have no argument to profile 2013 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1); 2014 sub(tmp1, i*TypeStackSlotEntries::per_arg_count(), tmp1); 2015 cmp_and_br_short(tmp1, TypeStackSlotEntries::per_arg_count(), less, pn, done); 2016 } 2017 ld_ptr(Address(callee, Method::const_offset()), tmp1); 2018 lduh(Address(tmp1, ConstMethod::size_of_parameters_offset()), tmp1); 2019 // stack offset o (zero based) from the start of the argument 2020 // list, for n arguments translates into offset n - o - 1 from 2021 // the end of the argument list. But there's an extra slot at 2022 // the stop of the stack. So the offset is n - o from Lesp. 2023 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, tmp2); 2024 sub(tmp1, tmp2, tmp1); 2025 2026 // Can't use MacroAssembler::argument_address() which needs Gargs to be set up 2027 sll(tmp1, Interpreter::logStackElementSize, tmp1); 2028 ld_ptr(Lesp, tmp1, tmp1); 2029 2030 Address mdo_arg_addr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); 2031 profile_obj_type(tmp1, mdo_arg_addr, tmp2); 2032 2033 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 2034 add(ImethodDataPtr, to_add, ImethodDataPtr); 2035 off_to_args += to_add; 2036 } 2037 2038 if (MethodData::profile_return()) { 2039 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1); 2040 sub(tmp1, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count(), tmp1); 2041 } 2042 2043 bind(done); 2044 2045 if (MethodData::profile_return()) { 2046 // We're right after the type profile for the last 2047 // argument. tmp1 is the number of cells left in the 2048 // CallTypeData/VirtualCallTypeData to reach its end. Non null 2049 // if there's a return to profile. 2050 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 2051 sll(tmp1, exact_log2(DataLayout::cell_size), tmp1); 2052 add(ImethodDataPtr, tmp1, ImethodDataPtr); 2053 } 2054 } else { 2055 assert(MethodData::profile_return(), "either profile call args or call ret"); 2056 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 2057 } 2058 2059 // mdp points right after the end of the 2060 // CallTypeData/VirtualCallTypeData, right after the cells for the 2061 // return value type if there's one. 2062 2063 bind(profile_continue); 2064 } 2065 } 2066 2067 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 2068 assert_different_registers(ret, tmp1, tmp2); 2069 if (ProfileInterpreter && MethodData::profile_return()) { 2070 Label profile_continue, done; 2071 2072 test_method_data_pointer(profile_continue); 2073 2074 if (MethodData::profile_return_jsr292_only()) { 2075 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); 2076 2077 // If we don't profile all invoke bytecodes we must make sure 2078 // it's a bytecode we indeed profile. We can't go back to the 2079 // begining of the ProfileData we intend to update to check its 2080 // type because we're right after it and we don't known its 2081 // length. 2082 Label do_profile; 2083 ldub(Lbcp, 0, tmp1); 2084 cmp_and_br_short(tmp1, Bytecodes::_invokedynamic, equal, pn, do_profile); 2085 cmp(tmp1, Bytecodes::_invokehandle); 2086 br(equal, false, pn, do_profile); 2087 delayed()->lduh(Lmethod, Method::intrinsic_id_offset_in_bytes(), tmp1); 2088 cmp_and_br_short(tmp1, vmIntrinsics::_compiledLambdaForm, notEqual, pt, profile_continue); 2089 2090 bind(do_profile); 2091 } 2092 2093 Address mdo_ret_addr(ImethodDataPtr, -in_bytes(ReturnTypeEntry::size())); 2094 mov(ret, tmp1); 2095 profile_obj_type(tmp1, mdo_ret_addr, tmp2); 2096 2097 bind(profile_continue); 2098 } 2099 } 2100 2101 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) { 2102 if (ProfileInterpreter && MethodData::profile_parameters()) { 2103 Label profile_continue, done; 2104 2105 test_method_data_pointer(profile_continue); 2106 2107 // Load the offset of the area within the MDO used for 2108 // parameters. If it's negative we're not profiling any parameters. 2109 lduw(ImethodDataPtr, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), tmp1); 2110 cmp_and_br_short(tmp1, 0, less, pn, profile_continue); 2111 2112 // Compute a pointer to the area for parameters from the offset 2113 // and move the pointer to the slot for the last 2114 // parameters. Collect profiling from last parameter down. 2115 // mdo start + parameters offset + array length - 1 2116 2117 // Pointer to the parameter area in the MDO 2118 Register mdp = tmp1; 2119 add(ImethodDataPtr, tmp1, mdp); 2120 2121 // offset of the current profile entry to update 2122 Register entry_offset = tmp2; 2123 // entry_offset = array len in number of cells 2124 ld_ptr(mdp, ArrayData::array_len_offset(), entry_offset); 2125 2126 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 2127 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 2128 2129 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 2130 sub(entry_offset, TypeStackSlotEntries::per_arg_count() - (off_base / DataLayout::cell_size), entry_offset); 2131 // entry_offset in bytes 2132 sll(entry_offset, exact_log2(DataLayout::cell_size), entry_offset); 2133 2134 Label loop; 2135 bind(loop); 2136 2137 // load offset on the stack from the slot for this parameter 2138 ld_ptr(mdp, entry_offset, tmp3); 2139 sll(tmp3,Interpreter::logStackElementSize, tmp3); 2140 neg(tmp3); 2141 // read the parameter from the local area 2142 ld_ptr(Llocals, tmp3, tmp3); 2143 2144 // make entry_offset now point to the type field for this parameter 2145 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 2146 assert(type_base > off_base, "unexpected"); 2147 add(entry_offset, type_base - off_base, entry_offset); 2148 2149 // profile the parameter 2150 Address arg_type(mdp, entry_offset); 2151 profile_obj_type(tmp3, arg_type, tmp4); 2152 2153 // go to next parameter 2154 sub(entry_offset, TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base), entry_offset); 2155 cmp_and_br_short(entry_offset, off_base, greaterEqual, pt, loop); 2156 2157 bind(profile_continue); 2158 } 2159 } 2160 2161 // add a InterpMonitorElem to stack (see frame_sparc.hpp) 2162 2163 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty, 2164 Register Rtemp, 2165 Register Rtemp2 ) { 2166 2167 Register Rlimit = Lmonitors; 2168 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 2169 assert( (delta & LongAlignmentMask) == 0, 2170 "sizeof BasicObjectLock must be even number of doublewords"); 2171 2172 sub( SP, delta, SP); 2173 sub( Lesp, delta, Lesp); 2174 sub( Lmonitors, delta, Lmonitors); 2175 2176 if (!stack_is_empty) { 2177 2178 // must copy stack contents down 2179 2180 Label start_copying, next; 2181 2182 // untested("monitor stack expansion"); 2183 compute_stack_base(Rtemp); 2184 ba(start_copying); 2185 delayed()->cmp(Rtemp, Rlimit); // done? duplicated below 2186 2187 // note: must copy from low memory upwards 2188 // On entry to loop, 2189 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS) 2190 // Loop mutates Rtemp 2191 2192 bind( next); 2193 2194 st_ptr(Rtemp2, Rtemp, 0); 2195 inc(Rtemp, wordSize); 2196 cmp(Rtemp, Rlimit); // are we done? (duplicated above) 2197 2198 bind( start_copying ); 2199 2200 brx( notEqual, true, pn, next ); 2201 delayed()->ld_ptr( Rtemp, delta, Rtemp2 ); 2202 2203 // done copying stack 2204 } 2205 } 2206 2207 // Locals 2208 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) { 2209 assert_not_delayed(); 2210 sll(index, Interpreter::logStackElementSize, index); 2211 sub(Llocals, index, index); 2212 ld_ptr(index, 0, dst); 2213 // Note: index must hold the effective address--the iinc template uses it 2214 } 2215 2216 // Just like access_local_ptr but the tag is a returnAddress 2217 void InterpreterMacroAssembler::access_local_returnAddress(Register index, 2218 Register dst ) { 2219 assert_not_delayed(); 2220 sll(index, Interpreter::logStackElementSize, index); 2221 sub(Llocals, index, index); 2222 ld_ptr(index, 0, dst); 2223 } 2224 2225 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) { 2226 assert_not_delayed(); 2227 sll(index, Interpreter::logStackElementSize, index); 2228 sub(Llocals, index, index); 2229 ld(index, 0, dst); 2230 // Note: index must hold the effective address--the iinc template uses it 2231 } 2232 2233 2234 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) { 2235 assert_not_delayed(); 2236 sll(index, Interpreter::logStackElementSize, index); 2237 sub(Llocals, index, index); 2238 // First half stored at index n+1 (which grows down from Llocals[n]) 2239 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst); 2240 } 2241 2242 2243 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) { 2244 assert_not_delayed(); 2245 sll(index, Interpreter::logStackElementSize, index); 2246 sub(Llocals, index, index); 2247 ldf(FloatRegisterImpl::S, index, 0, dst); 2248 } 2249 2250 2251 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) { 2252 assert_not_delayed(); 2253 sll(index, Interpreter::logStackElementSize, index); 2254 sub(Llocals, index, index); 2255 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst); 2256 } 2257 2258 2259 #ifdef ASSERT 2260 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) { 2261 Label L; 2262 2263 assert(Rindex != Rscratch, "Registers cannot be same"); 2264 assert(Rindex != Rscratch1, "Registers cannot be same"); 2265 assert(Rlimit != Rscratch, "Registers cannot be same"); 2266 assert(Rlimit != Rscratch1, "Registers cannot be same"); 2267 assert(Rscratch1 != Rscratch, "Registers cannot be same"); 2268 2269 // untested("reg area corruption"); 2270 add(Rindex, offset, Rscratch); 2271 add(Rlimit, 64 + STACK_BIAS, Rscratch1); 2272 cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L); 2273 stop("regsave area is being clobbered"); 2274 bind(L); 2275 } 2276 #endif // ASSERT 2277 2278 2279 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) { 2280 assert_not_delayed(); 2281 sll(index, Interpreter::logStackElementSize, index); 2282 sub(Llocals, index, index); 2283 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);) 2284 st(src, index, 0); 2285 } 2286 2287 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) { 2288 assert_not_delayed(); 2289 sll(index, Interpreter::logStackElementSize, index); 2290 sub(Llocals, index, index); 2291 #ifdef ASSERT 2292 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2293 #endif 2294 st_ptr(src, index, 0); 2295 } 2296 2297 2298 2299 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) { 2300 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n)); 2301 } 2302 2303 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) { 2304 assert_not_delayed(); 2305 sll(index, Interpreter::logStackElementSize, index); 2306 sub(Llocals, index, index); 2307 #ifdef ASSERT 2308 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2309 #endif 2310 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1 2311 } 2312 2313 2314 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) { 2315 assert_not_delayed(); 2316 sll(index, Interpreter::logStackElementSize, index); 2317 sub(Llocals, index, index); 2318 #ifdef ASSERT 2319 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2320 #endif 2321 stf(FloatRegisterImpl::S, src, index, 0); 2322 } 2323 2324 2325 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) { 2326 assert_not_delayed(); 2327 sll(index, Interpreter::logStackElementSize, index); 2328 sub(Llocals, index, index); 2329 #ifdef ASSERT 2330 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2331 #endif 2332 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1)); 2333 } 2334 2335 2336 int InterpreterMacroAssembler::top_most_monitor_byte_offset() { 2337 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 2338 int rounded_vm_local_words = align_up((int)frame::interpreter_frame_vm_local_words, WordsPerLong); 2339 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS; 2340 } 2341 2342 2343 Address InterpreterMacroAssembler::top_most_monitor() { 2344 return Address(FP, top_most_monitor_byte_offset()); 2345 } 2346 2347 2348 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) { 2349 add( Lesp, wordSize, Rdest ); 2350 } 2351 2352 void InterpreterMacroAssembler::get_method_counters(Register method, 2353 Register Rcounters, 2354 Label& skip) { 2355 Label has_counters; 2356 Address method_counters(method, in_bytes(Method::method_counters_offset())); 2357 ld_ptr(method_counters, Rcounters); 2358 br_notnull_short(Rcounters, Assembler::pt, has_counters); 2359 call_VM(noreg, CAST_FROM_FN_PTR(address, 2360 InterpreterRuntime::build_method_counters), method); 2361 ld_ptr(method_counters, Rcounters); 2362 br_null(Rcounters, false, Assembler::pn, skip); // No MethodCounters, OutOfMemory 2363 delayed()->nop(); 2364 bind(has_counters); 2365 } 2366 2367 void InterpreterMacroAssembler::increment_invocation_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) { 2368 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 2369 assert_different_registers(Rcounters, Rtmp, Rtmp2); 2370 2371 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() + 2372 InvocationCounter::counter_offset()); 2373 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() + 2374 InvocationCounter::counter_offset()); 2375 int delta = InvocationCounter::count_increment; 2376 2377 // Load each counter in a register 2378 ld( inv_counter, Rtmp ); 2379 ld( be_counter, Rtmp2 ); 2380 2381 assert( is_simm13( delta ), " delta too large."); 2382 2383 // Add the delta to the invocation counter and store the result 2384 add( Rtmp, delta, Rtmp ); 2385 2386 // Mask the backedge counter 2387 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2388 2389 // Store value 2390 st( Rtmp, inv_counter); 2391 2392 // Add invocation counter + backedge counter 2393 add( Rtmp, Rtmp2, Rtmp); 2394 2395 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 2396 } 2397 2398 void InterpreterMacroAssembler::increment_backedge_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) { 2399 assert(UseCompiler, "incrementing must be useful"); 2400 assert_different_registers(Rcounters, Rtmp, Rtmp2); 2401 2402 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() + 2403 InvocationCounter::counter_offset()); 2404 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() + 2405 InvocationCounter::counter_offset()); 2406 2407 int delta = InvocationCounter::count_increment; 2408 // Load each counter in a register 2409 ld( be_counter, Rtmp ); 2410 ld( inv_counter, Rtmp2 ); 2411 2412 // Add the delta to the backedge counter 2413 add( Rtmp, delta, Rtmp ); 2414 2415 // Mask the invocation counter, add to backedge counter 2416 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2417 2418 // and store the result to memory 2419 st( Rtmp, be_counter ); 2420 2421 // Add backedge + invocation counter 2422 add( Rtmp, Rtmp2, Rtmp ); 2423 2424 // Note that this macro must leave backedge_count + invocation_count in Rtmp! 2425 } 2426 2427 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count, 2428 Register method_counters, 2429 Register branch_bcp, 2430 Register Rtmp ) { 2431 Label did_not_overflow; 2432 Label overflow_with_error; 2433 assert_different_registers(backedge_count, Rtmp, branch_bcp); 2434 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 2435 2436 Address limit(method_counters, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())); 2437 ld(limit, Rtmp); 2438 cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow); 2439 2440 // When ProfileInterpreter is on, the backedge_count comes from the 2441 // MethodData*, which value does not get reset on the call to 2442 // frequency_counter_overflow(). To avoid excessive calls to the overflow 2443 // routine while the method is being compiled, add a second test to make sure 2444 // the overflow function is called only once every overflow_frequency. 2445 if (ProfileInterpreter) { 2446 const int overflow_frequency = 1024; 2447 andcc(backedge_count, overflow_frequency-1, Rtmp); 2448 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow); 2449 delayed()->nop(); 2450 } 2451 2452 // overflow in loop, pass branch bytecode 2453 set(6,Rtmp); 2454 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp); 2455 2456 // Was an OSR adapter generated? 2457 // O0 = osr nmethod 2458 br_null_short(O0, Assembler::pn, overflow_with_error); 2459 2460 // Has the nmethod been invalidated already? 2461 ldub(O0, nmethod::state_offset(), O2); 2462 cmp_and_br_short(O2, nmethod::in_use, Assembler::notEqual, Assembler::pn, overflow_with_error); 2463 2464 // migrate the interpreter frame off of the stack 2465 2466 mov(G2_thread, L7); 2467 // save nmethod 2468 mov(O0, L6); 2469 set_last_Java_frame(SP, noreg); 2470 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7); 2471 reset_last_Java_frame(); 2472 mov(L7, G2_thread); 2473 2474 // move OSR nmethod to I1 2475 mov(L6, I1); 2476 2477 // OSR buffer to I0 2478 mov(O0, I0); 2479 2480 // remove the interpreter frame 2481 restore(I5_savedSP, 0, SP); 2482 2483 // Jump to the osr code. 2484 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2); 2485 jmp(O2, G0); 2486 delayed()->nop(); 2487 2488 bind(overflow_with_error); 2489 2490 bind(did_not_overflow); 2491 } 2492 2493 2494 2495 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) { 2496 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); } 2497 } 2498 2499 2500 // local helper function for the verify_oop_or_return_address macro 2501 static bool verify_return_address(Method* m, int bci) { 2502 #ifndef PRODUCT 2503 address pc = (address)(m->constMethod()) 2504 + in_bytes(ConstMethod::codes_offset()) + bci; 2505 // assume it is a valid return address if it is inside m and is preceded by a jsr 2506 if (!m->contains(pc)) return false; 2507 address jsr_pc; 2508 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2509 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2510 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2511 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2512 #endif // PRODUCT 2513 return false; 2514 } 2515 2516 2517 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2518 if (!VerifyOops) return; 2519 // the VM documentation for the astore[_wide] bytecode allows 2520 // the TOS to be not only an oop but also a return address 2521 Label test; 2522 Label skip; 2523 // See if it is an address (in the current method): 2524 2525 mov(reg, Rtmp); 2526 const int log2_bytecode_size_limit = 16; 2527 srl(Rtmp, log2_bytecode_size_limit, Rtmp); 2528 br_notnull_short( Rtmp, pt, test ); 2529 2530 // %%% should use call_VM_leaf here? 2531 save_frame_and_mov(0, Lmethod, O0, reg, O1); 2532 save_thread(L7_thread_cache); 2533 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none); 2534 delayed()->nop(); 2535 restore_thread(L7_thread_cache); 2536 br_notnull( O0, false, pt, skip ); 2537 delayed()->restore(); 2538 2539 // Perform a more elaborate out-of-line call 2540 // Not an address; verify it: 2541 bind(test); 2542 verify_oop(reg); 2543 bind(skip); 2544 } 2545 2546 2547 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2548 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); 2549 } 2550 2551 2552 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 2553 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 2554 int increment, Address mask_addr, 2555 Register scratch1, Register scratch2, 2556 Condition cond, Label *where) { 2557 ld(counter_addr, scratch1); 2558 add(scratch1, increment, scratch1); 2559 ld(mask_addr, scratch2); 2560 andcc(scratch1, scratch2, G0); 2561 br(cond, false, Assembler::pn, *where); 2562 delayed()->st(scratch1, counter_addr); 2563 } 2564 2565 // Inline assembly for: 2566 // 2567 // if (thread is in interp_only_mode) { 2568 // InterpreterRuntime::post_method_entry(); 2569 // } 2570 // if (DTraceMethodProbes) { 2571 // SharedRuntime::dtrace_method_entry(method, receiver); 2572 // } 2573 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 2574 // SharedRuntime::rc_trace_method_entry(method, receiver); 2575 // } 2576 2577 void InterpreterMacroAssembler::notify_method_entry() { 2578 2579 // Whenever JVMTI puts a thread in interp_only_mode, method 2580 // entry/exit events are sent for that thread to track stack 2581 // depth. If it is possible to enter interp_only_mode we add 2582 // the code to check if the event should be sent. 2583 if (JvmtiExport::can_post_interpreter_events()) { 2584 Label L; 2585 Register temp_reg = O5; 2586 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2587 ld(interp_only, temp_reg); 2588 cmp_and_br_short(temp_reg, 0, equal, pt, L); 2589 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); 2590 bind(L); 2591 } 2592 2593 { 2594 Register temp_reg = O5; 2595 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2596 call_VM_leaf(noreg, 2597 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 2598 G2_thread, Lmethod); 2599 } 2600 2601 // RedefineClasses() tracing support for obsolete method entry 2602 if (log_is_enabled(Trace, redefine, class, obsolete)) { 2603 call_VM_leaf(noreg, 2604 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 2605 G2_thread, Lmethod); 2606 } 2607 } 2608 2609 2610 // Inline assembly for: 2611 // 2612 // if (thread is in interp_only_mode) { 2613 // // save result 2614 // InterpreterRuntime::post_method_exit(); 2615 // // restore result 2616 // } 2617 // if (DTraceMethodProbes) { 2618 // SharedRuntime::dtrace_method_exit(thread, method); 2619 // } 2620 // 2621 // Native methods have their result stored in d_tmp and l_tmp 2622 // Java methods have their result stored in the expression stack 2623 2624 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, 2625 TosState state, 2626 NotifyMethodExitMode mode) { 2627 2628 // Whenever JVMTI puts a thread in interp_only_mode, method 2629 // entry/exit events are sent for that thread to track stack 2630 // depth. If it is possible to enter interp_only_mode we add 2631 // the code to check if the event should be sent. 2632 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2633 Label L; 2634 Register temp_reg = O5; 2635 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2636 ld(interp_only, temp_reg); 2637 cmp_and_br_short(temp_reg, 0, equal, pt, L); 2638 2639 // Note: frame::interpreter_frame_result has a dependency on how the 2640 // method result is saved across the call to post_method_exit. For 2641 // native methods it assumes the result registers are saved to 2642 // l_scratch and d_scratch. If this changes then the interpreter_frame_result 2643 // implementation will need to be updated too. 2644 2645 save_return_value(state, is_native_method); 2646 call_VM(noreg, 2647 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2648 restore_return_value(state, is_native_method); 2649 bind(L); 2650 } 2651 2652 { 2653 Register temp_reg = O5; 2654 // Dtrace notification 2655 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2656 save_return_value(state, is_native_method); 2657 call_VM_leaf( 2658 noreg, 2659 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2660 G2_thread, Lmethod); 2661 restore_return_value(state, is_native_method); 2662 } 2663 } 2664 2665 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) { 2666 if (is_native_call) { 2667 stf(FloatRegisterImpl::D, F0, d_tmp); 2668 stx(O0, l_tmp); 2669 } else { 2670 push(state); 2671 } 2672 } 2673 2674 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) { 2675 if (is_native_call) { 2676 ldf(FloatRegisterImpl::D, d_tmp, F0); 2677 ldx(l_tmp, O0); 2678 } else { 2679 pop(state); 2680 } 2681 }