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