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