1 /* 2 * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "interp_masm_sparc.hpp" 27 #include "interpreter/interpreter.hpp" 28 #include "interpreter/interpreterRuntime.hpp" 29 #include "logging/log.hpp" 30 #include "oops/arrayOop.hpp" 31 #include "oops/markOop.hpp" 32 #include "oops/methodData.hpp" 33 #include "oops/method.hpp" 34 #include "oops/methodCounters.hpp" 35 #include "prims/jvmtiExport.hpp" 36 #include "prims/jvmtiThreadState.hpp" 37 #include "runtime/basicLock.hpp" 38 #include "runtime/biasedLocking.hpp" 39 #include "runtime/safepointMechanism.hpp" 40 #include "runtime/sharedRuntime.hpp" 41 #include "runtime/thread.inline.hpp" 42 #include "utilities/align.hpp" 43 44 // Implementation of InterpreterMacroAssembler 45 46 // This file specializes the assember with interpreter-specific macros 47 48 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS); 49 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS); 50 51 void InterpreterMacroAssembler::jump_to_entry(address entry) { 52 assert(entry, "Entry must have been generated by now"); 53 AddressLiteral al(entry); 54 jump_to(al, G3_scratch); 55 delayed()->nop(); 56 } 57 58 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) { 59 // Note: this algorithm is also used by C1's OSR entry sequence. 60 // Any changes should also be applied to CodeEmitter::emit_osr_entry(). 61 assert_different_registers(args_size, locals_size); 62 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted. 63 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words 64 // Use br/mov combination because it works on both V8 and V9 and is 65 // faster. 66 Label skip_move; 67 br(Assembler::negative, true, Assembler::pt, skip_move); 68 delayed()->mov(G0, delta); 69 bind(skip_move); 70 align_up(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned) 71 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes 72 } 73 74 // Dispatch code executed in the prolog of a bytecode which does not do it's 75 // own dispatch. The dispatch address is computed and placed in IdispatchAddress 76 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 77 assert_not_delayed(); 78 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 79 // dispatch table to use 80 AddressLiteral tbl(Interpreter::dispatch_table(state)); 81 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 82 set(tbl, G3_scratch); // compute addr of table 83 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr 84 } 85 86 87 // Dispatch code executed in the epilog of a bytecode which does not do it's 88 // own dispatch. The dispatch address in IdispatchAddress is used for the 89 // dispatch. 90 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 91 assert_not_delayed(); 92 verify_FPU(1, state); 93 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 94 jmp( IdispatchAddress, 0 ); 95 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); 96 else delayed()->nop(); 97 } 98 99 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) { 100 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) 101 assert_not_delayed(); 102 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 103 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, true, generate_poll); 104 } 105 106 107 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) { 108 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) 109 assert_not_delayed(); 110 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode 111 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false); 112 } 113 114 115 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 116 // load current bytecode 117 assert_not_delayed(); 118 ldub( Lbcp, 0, Lbyte_code); // load next bytecode 119 dispatch_base(state, table); 120 } 121 122 123 void InterpreterMacroAssembler::call_VM_leaf_base( 124 Register java_thread, 125 address entry_point, 126 int number_of_arguments 127 ) { 128 if (!java_thread->is_valid()) 129 java_thread = L7_thread_cache; 130 // super call 131 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments); 132 } 133 134 135 void InterpreterMacroAssembler::call_VM_base( 136 Register oop_result, 137 Register java_thread, 138 Register last_java_sp, 139 address entry_point, 140 int number_of_arguments, 141 bool check_exception 142 ) { 143 if (!java_thread->is_valid()) 144 java_thread = L7_thread_cache; 145 // See class ThreadInVMfromInterpreter, which assumes that the interpreter 146 // takes responsibility for setting its own thread-state on call-out. 147 // However, ThreadInVMfromInterpreter resets the state to "in_Java". 148 149 //save_bcp(); // save bcp 150 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception); 151 //restore_bcp(); // restore bcp 152 //restore_locals(); // restore locals pointer 153 } 154 155 156 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 157 if (JvmtiExport::can_pop_frame()) { 158 Label L; 159 160 // Check the "pending popframe condition" flag in the current thread 161 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg); 162 163 // Initiate popframe handling only if it is not already being processed. If the flag 164 // has the popframe_processing bit set, it means that this code is called *during* popframe 165 // handling - we don't want to reenter. 166 btst(JavaThread::popframe_pending_bit, scratch_reg); 167 br(zero, false, pt, L); 168 delayed()->nop(); 169 btst(JavaThread::popframe_processing_bit, scratch_reg); 170 br(notZero, false, pt, L); 171 delayed()->nop(); 172 173 // Call Interpreter::remove_activation_preserving_args_entry() to get the 174 // address of the same-named entrypoint in the generated interpreter code. 175 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 176 177 // Jump to Interpreter::_remove_activation_preserving_args_entry 178 jmpl(O0, G0, G0); 179 delayed()->nop(); 180 bind(L); 181 } 182 } 183 184 185 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 186 Register thr_state = G4_scratch; 187 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); 188 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset()); 189 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset()); 190 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset()); 191 switch (state) { 192 case ltos: ld_long(val_addr, Otos_l); break; 193 case atos: ld_ptr(oop_addr, Otos_l); 194 st_ptr(G0, oop_addr); break; 195 case btos: // fall through 196 case ztos: // fall through 197 case ctos: // fall through 198 case stos: // fall through 199 case itos: ld(val_addr, Otos_l1); break; 200 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break; 201 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break; 202 case vtos: /* nothing to do */ break; 203 default : ShouldNotReachHere(); 204 } 205 // Clean up tos value in the jvmti thread state 206 or3(G0, ilgl, G3_scratch); 207 stw(G3_scratch, tos_addr); 208 st_long(G0, val_addr); 209 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 210 } 211 212 213 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 214 if (JvmtiExport::can_force_early_return()) { 215 Label L; 216 Register thr_state = G3_scratch; 217 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); 218 br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit; 219 220 // Initiate earlyret handling only if it is not already being processed. 221 // If the flag has the earlyret_processing bit set, it means that this code 222 // is called *during* earlyret handling - we don't want to reenter. 223 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch); 224 cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L); 225 226 // Call Interpreter::remove_activation_early_entry() to get the address of the 227 // same-named entrypoint in the generated interpreter code 228 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1); 229 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1); 230 231 // Jump to Interpreter::_remove_activation_early_entry 232 jmpl(O0, G0, G0); 233 delayed()->nop(); 234 bind(L); 235 } 236 } 237 238 239 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) { 240 mov(arg_1, O0); 241 mov(arg_2, O1); 242 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2); 243 } 244 245 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) { 246 assert_not_delayed(); 247 dispatch_Lbyte_code(state, table); 248 } 249 250 251 void InterpreterMacroAssembler::dispatch_normal(TosState state) { 252 dispatch_base(state, Interpreter::normal_table(state)); 253 } 254 255 256 void InterpreterMacroAssembler::dispatch_only(TosState state) { 257 dispatch_base(state, Interpreter::dispatch_table(state)); 258 } 259 260 261 // common code to dispatch and dispatch_only 262 // dispatch value in Lbyte_code and increment Lbcp 263 264 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify, bool generate_poll) { 265 verify_FPU(1, state); 266 // %%%%% maybe implement +VerifyActivationFrameSize here 267 //verify_thread(); //too slow; we will just verify on method entry & exit 268 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 269 // dispatch table to use 270 AddressLiteral tbl(table); 271 Label dispatch; 272 273 if (SafepointMechanism::uses_thread_local_poll() && generate_poll) { 274 AddressLiteral sfpt_tbl(Interpreter::safept_table(state)); 275 Label no_safepoint; 276 277 if (tbl.value() != sfpt_tbl.value()) { 278 ldx(Address(G2_thread, Thread::polling_page_offset()), G3_scratch, 0); 279 // Armed page has poll_bit set, if poll bit is cleared just continue. 280 and3(G3_scratch, SafepointMechanism::poll_bit(), G3_scratch); 281 282 br_null_short(G3_scratch, Assembler::pt, no_safepoint); 283 set(sfpt_tbl, G3_scratch); 284 ba_short(dispatch); 285 } 286 bind(no_safepoint); 287 } 288 289 set(tbl, G3_scratch); // compute addr of table 290 bind(dispatch); 291 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize 292 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr 293 jmp( G3_scratch, 0 ); 294 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); 295 else delayed()->nop(); 296 } 297 298 299 // Helpers for expression stack 300 301 // Longs and doubles are Category 2 computational types in the 302 // JVM specification (section 3.11.1) and take 2 expression stack or 303 // local slots. 304 // Aligning them on 32 bit with tagged stacks is hard because the code generated 305 // for the dup* bytecodes depends on what types are already on the stack. 306 // If the types are split into the two stack/local slots, that is much easier 307 // (and we can use 0 for non-reference tags). 308 309 // Known good alignment in _LP64 but unknown otherwise 310 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) { 311 assert_not_delayed(); 312 313 ldf(FloatRegisterImpl::D, r1, offset, d); 314 } 315 316 // Known good alignment in _LP64 but unknown otherwise 317 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) { 318 assert_not_delayed(); 319 320 stf(FloatRegisterImpl::D, d, r1, offset); 321 // store something more useful here 322 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);) 323 } 324 325 326 // Known good alignment in _LP64 but unknown otherwise 327 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) { 328 assert_not_delayed(); 329 ldx(r1, offset, rd); 330 } 331 332 // Known good alignment in _LP64 but unknown otherwise 333 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) { 334 assert_not_delayed(); 335 336 stx(l, r1, offset); 337 // store something more useful here 338 stx(G0, r1, offset+Interpreter::stackElementSize); 339 } 340 341 void InterpreterMacroAssembler::pop_i(Register r) { 342 assert_not_delayed(); 343 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r); 344 inc(Lesp, Interpreter::stackElementSize); 345 debug_only(verify_esp(Lesp)); 346 } 347 348 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) { 349 assert_not_delayed(); 350 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); 351 inc(Lesp, Interpreter::stackElementSize); 352 debug_only(verify_esp(Lesp)); 353 } 354 355 void InterpreterMacroAssembler::pop_l(Register r) { 356 assert_not_delayed(); 357 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r); 358 inc(Lesp, 2*Interpreter::stackElementSize); 359 debug_only(verify_esp(Lesp)); 360 } 361 362 363 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) { 364 assert_not_delayed(); 365 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f); 366 inc(Lesp, Interpreter::stackElementSize); 367 debug_only(verify_esp(Lesp)); 368 } 369 370 371 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) { 372 assert_not_delayed(); 373 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f); 374 inc(Lesp, 2*Interpreter::stackElementSize); 375 debug_only(verify_esp(Lesp)); 376 } 377 378 379 void InterpreterMacroAssembler::push_i(Register r) { 380 assert_not_delayed(); 381 debug_only(verify_esp(Lesp)); 382 st(r, Lesp, 0); 383 dec(Lesp, Interpreter::stackElementSize); 384 } 385 386 void InterpreterMacroAssembler::push_ptr(Register r) { 387 assert_not_delayed(); 388 st_ptr(r, Lesp, 0); 389 dec(Lesp, Interpreter::stackElementSize); 390 } 391 392 // remember: our convention for longs in SPARC is: 393 // O0 (Otos_l1) has high-order part in first word, 394 // O1 (Otos_l2) has low-order part in second word 395 396 void InterpreterMacroAssembler::push_l(Register r) { 397 assert_not_delayed(); 398 debug_only(verify_esp(Lesp)); 399 // Longs are stored in memory-correct order, even if unaligned. 400 int offset = -Interpreter::stackElementSize; 401 store_unaligned_long(r, Lesp, offset); 402 dec(Lesp, 2 * Interpreter::stackElementSize); 403 } 404 405 406 void InterpreterMacroAssembler::push_f(FloatRegister f) { 407 assert_not_delayed(); 408 debug_only(verify_esp(Lesp)); 409 stf(FloatRegisterImpl::S, f, Lesp, 0); 410 dec(Lesp, Interpreter::stackElementSize); 411 } 412 413 414 void InterpreterMacroAssembler::push_d(FloatRegister d) { 415 assert_not_delayed(); 416 debug_only(verify_esp(Lesp)); 417 // Longs are stored in memory-correct order, even if unaligned. 418 int offset = -Interpreter::stackElementSize; 419 store_unaligned_double(d, Lesp, offset); 420 dec(Lesp, 2 * Interpreter::stackElementSize); 421 } 422 423 424 void InterpreterMacroAssembler::push(TosState state) { 425 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 426 switch (state) { 427 case atos: push_ptr(); break; 428 case btos: // fall through 429 case ztos: // fall through 430 case ctos: // fall through 431 case stos: // fall through 432 case itos: push_i(); break; 433 case ltos: push_l(); break; 434 case ftos: push_f(); break; 435 case dtos: push_d(); break; 436 case vtos: /* nothing to do */ break; 437 default : ShouldNotReachHere(); 438 } 439 } 440 441 442 void InterpreterMacroAssembler::pop(TosState state) { 443 switch (state) { 444 case atos: pop_ptr(); break; 445 case btos: // fall through 446 case ztos: // fall through 447 case ctos: // fall through 448 case stos: // fall through 449 case itos: pop_i(); break; 450 case ltos: pop_l(); break; 451 case ftos: pop_f(); break; 452 case dtos: pop_d(); break; 453 case vtos: /* nothing to do */ break; 454 default : ShouldNotReachHere(); 455 } 456 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 457 } 458 459 460 // Helpers for swap and dup 461 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 462 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val); 463 } 464 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 465 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n)); 466 } 467 468 469 void InterpreterMacroAssembler::load_receiver(Register param_count, 470 Register recv) { 471 sll(param_count, Interpreter::logStackElementSize, param_count); 472 ld_ptr(Lesp, param_count, recv); // gets receiver oop 473 } 474 475 void InterpreterMacroAssembler::empty_expression_stack() { 476 // Reset Lesp. 477 sub( Lmonitors, wordSize, Lesp ); 478 479 // Reset SP by subtracting more space from Lesp. 480 Label done; 481 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!"); 482 483 // A native does not need to do this, since its callee does not change SP. 484 ld(Lmethod, Method::access_flags_offset(), Gframe_size); // Load access flags. 485 btst(JVM_ACC_NATIVE, Gframe_size); 486 br(Assembler::notZero, false, Assembler::pt, done); 487 delayed()->nop(); 488 489 // Compute max expression stack+register save area 490 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Gframe_size); 491 lduh(Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size); // Load max stack. 492 add(Gframe_size, frame::memory_parameter_word_sp_offset+Method::extra_stack_entries(), Gframe_size ); 493 494 // 495 // now set up a stack frame with the size computed above 496 // 497 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below 498 sll( Gframe_size, LogBytesPerWord, Gframe_size ); 499 sub( Lesp, Gframe_size, Gframe_size ); 500 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary 501 debug_only(verify_sp(Gframe_size, G4_scratch)); 502 sub(Gframe_size, STACK_BIAS, Gframe_size ); 503 mov(Gframe_size, SP); 504 505 bind(done); 506 } 507 508 509 #ifdef ASSERT 510 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) { 511 Label Bad, OK; 512 513 // Saved SP must be aligned. 514 btst(2*BytesPerWord-1, Rsp); 515 br(Assembler::notZero, false, Assembler::pn, Bad); 516 delayed()->nop(); 517 518 // Saved SP, plus register window size, must not be above FP. 519 add(Rsp, frame::register_save_words * wordSize, Rtemp); 520 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP 521 cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad); 522 523 // Saved SP must not be ridiculously below current SP. 524 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K); 525 set(maxstack, Rtemp); 526 sub(SP, Rtemp, Rtemp); 527 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp 528 cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad); 529 530 ba_short(OK); 531 532 bind(Bad); 533 stop("on return to interpreted call, restored SP is corrupted"); 534 535 bind(OK); 536 } 537 538 539 void InterpreterMacroAssembler::verify_esp(Register Resp) { 540 // about to read or write Resp[0] 541 // make sure it is not in the monitors or the register save area 542 Label OK1, OK2; 543 544 cmp(Resp, Lmonitors); 545 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1); 546 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp); 547 stop("too many pops: Lesp points into monitor area"); 548 bind(OK1); 549 sub(Resp, STACK_BIAS, Resp); 550 cmp(Resp, SP); 551 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2); 552 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp); 553 stop("too many pushes: Lesp points into register window"); 554 bind(OK2); 555 } 556 #endif // ASSERT 557 558 // Load compiled (i2c) or interpreter entry when calling from interpreted and 559 // do the call. Centralized so that all interpreter calls will do the same actions. 560 // If jvmti single stepping is on for a thread we must not call compiled code. 561 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) { 562 563 // Assume we want to go compiled if available 564 565 ld_ptr(G5_method, in_bytes(Method::from_interpreted_offset()), target); 566 567 if (JvmtiExport::can_post_interpreter_events()) { 568 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 569 // compiled code in threads for which the event is enabled. Check here for 570 // interp_only_mode if these events CAN be enabled. 571 verify_thread(); 572 Label skip_compiled_code; 573 574 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 575 ld(interp_only, scratch); 576 cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn); 577 delayed()->ld_ptr(G5_method, in_bytes(Method::interpreter_entry_offset()), target); 578 bind(skip_compiled_code); 579 } 580 581 // the i2c_adapters need Method* in G5_method (right? %%%) 582 // do the call 583 #ifdef ASSERT 584 { 585 Label ok; 586 br_notnull_short(target, Assembler::pt, ok); 587 stop("null entry point"); 588 bind(ok); 589 } 590 #endif // ASSERT 591 592 // Adjust Rret first so Llast_SP can be same as Rret 593 add(Rret, -frame::pc_return_offset, O7); 594 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer 595 // Record SP so we can remove any stack space allocated by adapter transition 596 jmp(target, 0); 597 delayed()->mov(SP, Llast_SP); 598 } 599 600 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) { 601 assert_not_delayed(); 602 603 Label not_taken; 604 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken); 605 else br (cc, false, Assembler::pn, not_taken); 606 delayed()->nop(); 607 608 TemplateTable::branch(false,false); 609 610 bind(not_taken); 611 612 profile_not_taken_branch(G3_scratch); 613 } 614 615 616 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp( 617 int bcp_offset, 618 Register Rtmp, 619 Register Rdst, 620 signedOrNot is_signed, 621 setCCOrNot should_set_CC ) { 622 assert(Rtmp != Rdst, "need separate temp register"); 623 assert_not_delayed(); 624 switch (is_signed) { 625 default: ShouldNotReachHere(); 626 627 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte 628 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte 629 } 630 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte 631 sll( Rdst, BitsPerByte, Rdst); 632 switch (should_set_CC ) { 633 default: ShouldNotReachHere(); 634 635 case set_CC: orcc( Rdst, Rtmp, Rdst ); break; 636 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break; 637 } 638 } 639 640 641 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp( 642 int bcp_offset, 643 Register Rtmp, 644 Register Rdst, 645 setCCOrNot should_set_CC ) { 646 assert(Rtmp != Rdst, "need separate temp register"); 647 assert_not_delayed(); 648 add( Lbcp, bcp_offset, Rtmp); 649 andcc( Rtmp, 3, G0); 650 Label aligned; 651 switch (should_set_CC ) { 652 default: ShouldNotReachHere(); 653 654 case set_CC: break; 655 case dont_set_CC: break; 656 } 657 658 br(Assembler::zero, true, Assembler::pn, aligned); 659 delayed()->ldsw(Rtmp, 0, Rdst); 660 661 ldub(Lbcp, bcp_offset + 3, Rdst); 662 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst); 663 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst); 664 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); 665 or3(Rtmp, Rdst, Rdst ); 666 667 bind(aligned); 668 if (should_set_CC == set_CC) tst(Rdst); 669 } 670 671 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register temp, Register index, 672 int bcp_offset, size_t index_size) { 673 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 674 if (index_size == sizeof(u2)) { 675 get_2_byte_integer_at_bcp(bcp_offset, temp, index, Unsigned); 676 } else if (index_size == sizeof(u4)) { 677 get_4_byte_integer_at_bcp(bcp_offset, temp, index); 678 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 679 xor3(index, -1, index); // convert to plain index 680 } else if (index_size == sizeof(u1)) { 681 ldub(Lbcp, bcp_offset, index); 682 } else { 683 ShouldNotReachHere(); 684 } 685 } 686 687 688 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp, 689 int bcp_offset, size_t index_size) { 690 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 691 assert_different_registers(cache, tmp); 692 assert_not_delayed(); 693 get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size); 694 // convert from field index to ConstantPoolCacheEntry index and from 695 // word index to byte offset 696 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); 697 add(LcpoolCache, tmp, cache); 698 } 699 700 701 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, 702 Register temp, 703 Register bytecode, 704 int byte_no, 705 int bcp_offset, 706 size_t index_size) { 707 get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size); 708 ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode); 709 const int shift_count = (1 + byte_no) * BitsPerByte; 710 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) || 711 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift), 712 "correct shift count"); 713 srl(bytecode, shift_count, bytecode); 714 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask"); 715 and3(bytecode, ConstantPoolCacheEntry::bytecode_1_mask, bytecode); 716 } 717 718 719 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp, 720 int bcp_offset, size_t index_size) { 721 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 722 assert_different_registers(cache, tmp); 723 assert_not_delayed(); 724 if (index_size == sizeof(u2)) { 725 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); 726 } else { 727 ShouldNotReachHere(); // other sizes not supported here 728 } 729 // convert from field index to ConstantPoolCacheEntry index 730 // and from word index to byte offset 731 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); 732 // skip past the header 733 add(tmp, in_bytes(ConstantPoolCache::base_offset()), tmp); 734 // construct pointer to cache entry 735 add(LcpoolCache, tmp, cache); 736 } 737 738 739 // Load object from cpool->resolved_references(index) 740 void InterpreterMacroAssembler::load_resolved_reference_at_index( 741 Register result, Register index) { 742 assert_different_registers(result, index); 743 assert_not_delayed(); 744 // convert from field index to resolved_references() index and from 745 // word index to byte offset. Since this is a java object, it can be compressed 746 Register tmp = index; // reuse 747 sll(index, LogBytesPerHeapOop, tmp); 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); 753 // Add in the index 754 add(result, tmp, result); 755 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result); 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 // convention: move aberrant index into G3_scratch for exception message 898 mov(index, G3_scratch); 899 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok); 900 901 // add offset if didn't do it in delay slot 902 if (index_shift > 0) add(array, index, res); // addr - const offset in index 903 } 904 905 906 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { 907 assert_not_delayed(); 908 909 // pop array 910 pop_ptr(array); 911 912 // check array 913 index_check_without_pop(array, index, index_shift, tmp, res); 914 } 915 916 917 void InterpreterMacroAssembler::get_const(Register Rdst) { 918 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst); 919 } 920 921 922 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 923 get_const(Rdst); 924 ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 925 } 926 927 928 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 929 get_constant_pool(Rdst); 930 ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 931 } 932 933 934 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 935 get_constant_pool(Rcpool); 936 ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags); 937 } 938 939 940 // unlock if synchronized method 941 // 942 // Unlock the receiver if this is a synchronized method. 943 // Unlock any Java monitors from syncronized blocks. 944 // 945 // If there are locked Java monitors 946 // If throw_monitor_exception 947 // throws IllegalMonitorStateException 948 // Else if install_monitor_exception 949 // installs IllegalMonitorStateException 950 // Else 951 // no error processing 952 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 953 bool throw_monitor_exception, 954 bool install_monitor_exception) { 955 Label unlocked, unlock, no_unlock; 956 957 // get the value of _do_not_unlock_if_synchronized into G1_scratch 958 const Address do_not_unlock_if_synchronized(G2_thread, 959 JavaThread::do_not_unlock_if_synchronized_offset()); 960 ldbool(do_not_unlock_if_synchronized, G1_scratch); 961 stbool(G0, do_not_unlock_if_synchronized); // reset the flag 962 963 // check if synchronized method 964 const Address access_flags(Lmethod, Method::access_flags_offset()); 965 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 966 push(state); // save tos 967 ld(access_flags, G3_scratch); // Load access flags. 968 btst(JVM_ACC_SYNCHRONIZED, G3_scratch); 969 br(zero, false, pt, unlocked); 970 delayed()->nop(); 971 972 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 973 // is set. 974 cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock); 975 delayed()->nop(); 976 977 // BasicObjectLock will be first in list, since this is a synchronized method. However, need 978 // to check that the object has not been unlocked by an explicit monitorexit bytecode. 979 980 //Intel: if (throw_monitor_exception) ... else ... 981 // Entry already unlocked, need to throw exception 982 //... 983 984 // pass top-most monitor elem 985 add( top_most_monitor(), O1 ); 986 987 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch); 988 br_notnull_short(G3_scratch, pt, unlock); 989 990 if (throw_monitor_exception) { 991 // Entry already unlocked need to throw an exception 992 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 993 should_not_reach_here(); 994 } else { 995 // Monitor already unlocked during a stack unroll. 996 // If requested, install an illegal_monitor_state_exception. 997 // Continue with stack unrolling. 998 if (install_monitor_exception) { 999 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1000 } 1001 ba_short(unlocked); 1002 } 1003 1004 bind(unlock); 1005 1006 unlock_object(O1); 1007 1008 bind(unlocked); 1009 1010 // I0, I1: Might contain return value 1011 1012 // Check that all monitors are unlocked 1013 { Label loop, exception, entry, restart; 1014 1015 Register Rmptr = O0; 1016 Register Rtemp = O1; 1017 Register Rlimit = Lmonitors; 1018 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 1019 assert( (delta & LongAlignmentMask) == 0, 1020 "sizeof BasicObjectLock must be even number of doublewords"); 1021 1022 #ifdef ASSERT 1023 add(top_most_monitor(), Rmptr, delta); 1024 { Label L; 1025 // ensure that Rmptr starts out above (or at) Rlimit 1026 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); 1027 stop("monitor stack has negative size"); 1028 bind(L); 1029 } 1030 #endif 1031 bind(restart); 1032 ba(entry); 1033 delayed()-> 1034 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry 1035 1036 // Entry is still locked, need to throw exception 1037 bind(exception); 1038 if (throw_monitor_exception) { 1039 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1040 should_not_reach_here(); 1041 } else { 1042 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 1043 // Unlock does not block, so don't have to worry about the frame 1044 unlock_object(Rmptr); 1045 if (install_monitor_exception) { 1046 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 1047 } 1048 ba_short(restart); 1049 } 1050 1051 bind(loop); 1052 cmp(Rtemp, G0); // check if current entry is used 1053 brx(Assembler::notEqual, false, pn, exception); 1054 delayed()-> 1055 dec(Rmptr, delta); // otherwise advance to next entry 1056 #ifdef ASSERT 1057 { Label L; 1058 // ensure that Rmptr has not somehow stepped below Rlimit 1059 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); 1060 stop("ran off the end of the monitor stack"); 1061 bind(L); 1062 } 1063 #endif 1064 bind(entry); 1065 cmp(Rmptr, Rlimit); // check if bottom reached 1066 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry 1067 delayed()-> 1068 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp); 1069 } 1070 1071 bind(no_unlock); 1072 pop(state); 1073 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1074 } 1075 1076 void InterpreterMacroAssembler::narrow(Register result) { 1077 1078 ld_ptr(Address(Lmethod, Method::const_offset()), G3_scratch); 1079 ldub(G3_scratch, in_bytes(ConstMethod::result_type_offset()), G3_scratch); 1080 1081 Label notBool, notByte, notChar, done; 1082 1083 // common case first 1084 cmp(G3_scratch, T_INT); 1085 br(Assembler::equal, true, pn, done); 1086 delayed()->nop(); 1087 1088 cmp(G3_scratch, T_BOOLEAN); 1089 br(Assembler::notEqual, true, pn, notBool); 1090 delayed()->cmp(G3_scratch, T_BYTE); 1091 and3(result, 1, result); 1092 ba(done); 1093 delayed()->nop(); 1094 1095 bind(notBool); 1096 // cmp(G3_scratch, T_BYTE); 1097 br(Assembler::notEqual, true, pn, notByte); 1098 delayed()->cmp(G3_scratch, T_CHAR); 1099 sll(result, 24, result); 1100 sra(result, 24, result); 1101 ba(done); 1102 delayed()->nop(); 1103 1104 bind(notByte); 1105 // cmp(G3_scratch, T_CHAR); 1106 sll(result, 16, result); 1107 br(Assembler::notEqual, true, pn, done); 1108 delayed()->sra(result, 16, result); 1109 // sll(result, 16, result); 1110 srl(result, 16, result); 1111 1112 // bind(notChar); 1113 // must be short, instructions already executed in delay slot 1114 // sll(result, 16, result); 1115 // sra(result, 16, result); 1116 1117 bind(done); 1118 } 1119 1120 // remove activation 1121 // 1122 // Unlock the receiver if this is a synchronized method. 1123 // Unlock any Java monitors from syncronized blocks. 1124 // Remove the activation from the stack. 1125 // 1126 // If there are locked Java monitors 1127 // If throw_monitor_exception 1128 // throws IllegalMonitorStateException 1129 // Else if install_monitor_exception 1130 // installs IllegalMonitorStateException 1131 // Else 1132 // no error processing 1133 void InterpreterMacroAssembler::remove_activation(TosState state, 1134 bool throw_monitor_exception, 1135 bool install_monitor_exception) { 1136 1137 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 1138 1139 // save result (push state before jvmti call and pop it afterwards) and notify jvmti 1140 notify_method_exit(false, state, NotifyJVMTI); 1141 1142 if (StackReservedPages > 0) { 1143 // testing if Stack Reserved Area needs to be re-enabled 1144 Label no_reserved_zone_enabling; 1145 ld_ptr(G2_thread, JavaThread::reserved_stack_activation_offset(), G3_scratch); 1146 cmp_and_brx_short(SP, G3_scratch, Assembler::lessUnsigned, Assembler::pt, no_reserved_zone_enabling); 1147 1148 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), G2_thread); 1149 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError), G2_thread); 1150 should_not_reach_here(); 1151 1152 bind(no_reserved_zone_enabling); 1153 } 1154 1155 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); 1156 verify_thread(); 1157 1158 // return tos 1159 assert(Otos_l1 == Otos_i, "adjust code below"); 1160 switch (state) { 1161 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0 1162 case btos: // fall through 1163 case ztos: // fall through 1164 case ctos: 1165 case stos: // fall through 1166 case atos: // fall through 1167 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0 1168 case ftos: // fall through 1169 case dtos: // fall through 1170 case vtos: /* nothing to do */ break; 1171 default : ShouldNotReachHere(); 1172 } 1173 } 1174 1175 // Lock object 1176 // 1177 // Argument - lock_reg points to the BasicObjectLock to be used for locking, 1178 // it must be initialized with the object to lock 1179 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) { 1180 if (UseHeavyMonitors) { 1181 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1182 } 1183 else { 1184 Register obj_reg = Object; 1185 Register mark_reg = G4_scratch; 1186 Register temp_reg = G1_scratch; 1187 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes()); 1188 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1189 Label done; 1190 1191 Label slow_case; 1192 1193 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg); 1194 1195 // load markOop from object into mark_reg 1196 ld_ptr(mark_addr, mark_reg); 1197 1198 if (UseBiasedLocking) { 1199 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case); 1200 } 1201 1202 // get the address of basicLock on stack that will be stored in the object 1203 // we need a temporary register here as we do not want to clobber lock_reg 1204 // (cas clobbers the destination register) 1205 mov(lock_reg, temp_reg); 1206 // set mark reg to be (markOop of object | UNLOCK_VALUE) 1207 or3(mark_reg, markOopDesc::unlocked_value, mark_reg); 1208 // initialize the box (Must happen before we update the object mark!) 1209 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1210 // compare and exchange object_addr, markOop | 1, stack address of basicLock 1211 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1212 cas_ptr(mark_addr.base(), mark_reg, temp_reg); 1213 1214 // if the compare and exchange succeeded we are done (we saw an unlocked object) 1215 cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done); 1216 1217 // We did not see an unlocked object so try the fast recursive case 1218 1219 // Check if owner is self by comparing the value in the markOop of object 1220 // with the stack pointer 1221 sub(temp_reg, SP, temp_reg); 1222 sub(temp_reg, STACK_BIAS, temp_reg); 1223 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 1224 1225 // Composite "andcc" test: 1226 // (a) %sp -vs- markword proximity check, and, 1227 // (b) verify mark word LSBs == 0 (Stack-locked). 1228 // 1229 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size()) 1230 // Note that the page size used for %sp proximity testing is arbitrary and is 1231 // unrelated to the actual MMU page size. We use a 'logical' page size of 1232 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate 1233 // field of the andcc instruction. 1234 andcc (temp_reg, 0xFFFFF003, G0) ; 1235 1236 // if condition is true we are done and hence we can store 0 in the displaced 1237 // header indicating it is a recursive lock and be done 1238 brx(Assembler::zero, true, Assembler::pt, done); 1239 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes()); 1240 1241 // none of the above fast optimizations worked so we have to get into the 1242 // slow case of monitor enter 1243 bind(slow_case); 1244 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); 1245 1246 bind(done); 1247 } 1248 } 1249 1250 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 1251 // 1252 // Argument - lock_reg points to the BasicObjectLock for lock 1253 // Throw IllegalMonitorException if object is not locked by current thread 1254 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1255 if (UseHeavyMonitors) { 1256 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1257 } else { 1258 Register obj_reg = G3_scratch; 1259 Register mark_reg = G4_scratch; 1260 Register displaced_header_reg = G1_scratch; 1261 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes()); 1262 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); 1263 Label done; 1264 1265 if (UseBiasedLocking) { 1266 // load the object out of the BasicObjectLock 1267 ld_ptr(lockobj_addr, obj_reg); 1268 biased_locking_exit(mark_addr, mark_reg, done, true); 1269 st_ptr(G0, lockobj_addr); // free entry 1270 } 1271 1272 // Test first if we are in the fast recursive case 1273 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes()); 1274 ld_ptr(lock_addr, displaced_header_reg); 1275 br_null(displaced_header_reg, true, Assembler::pn, done); 1276 delayed()->st_ptr(G0, lockobj_addr); // free entry 1277 1278 // See if it is still a light weight lock, if so we just unlock 1279 // the object and we are done 1280 1281 if (!UseBiasedLocking) { 1282 // load the object out of the BasicObjectLock 1283 ld_ptr(lockobj_addr, obj_reg); 1284 } 1285 1286 // we have the displaced header in displaced_header_reg 1287 // we expect to see the stack address of the basicLock in case the 1288 // lock is still a light weight lock (lock_reg) 1289 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 1290 cas_ptr(mark_addr.base(), lock_reg, displaced_header_reg); 1291 cmp(lock_reg, displaced_header_reg); 1292 brx(Assembler::equal, true, Assembler::pn, done); 1293 delayed()->st_ptr(G0, lockobj_addr); // free entry 1294 1295 // The lock has been converted into a heavy lock and hence 1296 // we need to get into the slow case 1297 1298 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1299 1300 bind(done); 1301 } 1302 } 1303 1304 // Get the method data pointer from the Method* and set the 1305 // specified register to its value. 1306 1307 void InterpreterMacroAssembler::set_method_data_pointer() { 1308 assert(ProfileInterpreter, "must be profiling interpreter"); 1309 Label get_continue; 1310 1311 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); 1312 test_method_data_pointer(get_continue); 1313 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); 1314 bind(get_continue); 1315 } 1316 1317 // Set the method data pointer for the current bcp. 1318 1319 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1320 assert(ProfileInterpreter, "must be profiling interpreter"); 1321 Label zero_continue; 1322 1323 // Test MDO to avoid the call if it is NULL. 1324 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); 1325 test_method_data_pointer(zero_continue); 1326 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp); 1327 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); 1328 add(ImethodDataPtr, O0, ImethodDataPtr); 1329 bind(zero_continue); 1330 } 1331 1332 // Test ImethodDataPtr. If it is null, continue at the specified label 1333 1334 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1335 assert(ProfileInterpreter, "must be profiling interpreter"); 1336 br_null_short(ImethodDataPtr, Assembler::pn, zero_continue); 1337 } 1338 1339 void InterpreterMacroAssembler::verify_method_data_pointer() { 1340 assert(ProfileInterpreter, "must be profiling interpreter"); 1341 #ifdef ASSERT 1342 Label verify_continue; 1343 test_method_data_pointer(verify_continue); 1344 1345 // If the mdp is valid, it will point to a DataLayout header which is 1346 // consistent with the bcp. The converse is highly probable also. 1347 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch); 1348 ld_ptr(Lmethod, Method::const_offset(), O5); 1349 add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch); 1350 add(G3_scratch, O5, G3_scratch); 1351 cmp(Lbcp, G3_scratch); 1352 brx(Assembler::equal, false, Assembler::pt, verify_continue); 1353 1354 Register temp_reg = O5; 1355 delayed()->mov(ImethodDataPtr, temp_reg); 1356 // %%% should use call_VM_leaf here? 1357 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr); 1358 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1); 1359 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS); 1360 stf(FloatRegisterImpl::D, Ftos_d, d_save); 1361 mov(temp_reg->after_save(), O2); 1362 save_thread(L7_thread_cache); 1363 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none); 1364 delayed()->nop(); 1365 restore_thread(L7_thread_cache); 1366 ldf(FloatRegisterImpl::D, d_save, Ftos_d); 1367 restore(); 1368 bind(verify_continue); 1369 #endif // ASSERT 1370 } 1371 1372 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1373 Register method_counters, 1374 Register Rtmp, 1375 Label &profile_continue) { 1376 assert(ProfileInterpreter, "must be profiling interpreter"); 1377 // Control will flow to "profile_continue" if the counter is less than the 1378 // limit or if we call profile_method() 1379 1380 Label done; 1381 1382 // if no method data exists, and the counter is high enough, make one 1383 br_notnull_short(ImethodDataPtr, Assembler::pn, done); 1384 1385 // Test to see if we should create a method data oop 1386 Address profile_limit(method_counters, MethodCounters::interpreter_profile_limit_offset()); 1387 ld(profile_limit, Rtmp); 1388 cmp(invocation_count, Rtmp); 1389 // Use long branches because call_VM() code and following code generated by 1390 // test_backedge_count_for_osr() is large in debug VM. 1391 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue); 1392 delayed()->nop(); 1393 1394 // Build it now. 1395 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1396 set_method_data_pointer_for_bcp(); 1397 ba(profile_continue); 1398 delayed()->nop(); 1399 bind(done); 1400 } 1401 1402 // Store a value at some constant offset from the method data pointer. 1403 1404 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1405 assert(ProfileInterpreter, "must be profiling interpreter"); 1406 st_ptr(value, ImethodDataPtr, constant); 1407 } 1408 1409 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter, 1410 Register bumped_count, 1411 bool decrement) { 1412 assert(ProfileInterpreter, "must be profiling interpreter"); 1413 1414 // Load the counter. 1415 ld_ptr(counter, bumped_count); 1416 1417 if (decrement) { 1418 // Decrement the register. Set condition codes. 1419 subcc(bumped_count, DataLayout::counter_increment, bumped_count); 1420 1421 // If the decrement causes the counter to overflow, stay negative 1422 Label L; 1423 brx(Assembler::negative, true, Assembler::pn, L); 1424 1425 // Store the decremented counter, if it is still negative. 1426 delayed()->st_ptr(bumped_count, counter); 1427 bind(L); 1428 } else { 1429 // Increment the register. Set carry flag. 1430 addcc(bumped_count, DataLayout::counter_increment, bumped_count); 1431 1432 // If the increment causes the counter to overflow, pull back by 1. 1433 assert(DataLayout::counter_increment == 1, "subc works"); 1434 subc(bumped_count, G0, bumped_count); 1435 1436 // Store the incremented counter. 1437 st_ptr(bumped_count, counter); 1438 } 1439 } 1440 1441 // Increment the value at some constant offset from the method data pointer. 1442 1443 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1444 Register bumped_count, 1445 bool decrement) { 1446 // Locate the counter at a fixed offset from the mdp: 1447 Address counter(ImethodDataPtr, constant); 1448 increment_mdp_data_at(counter, bumped_count, decrement); 1449 } 1450 1451 // Increment the value at some non-fixed (reg + constant) offset from 1452 // the method data pointer. 1453 1454 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1455 int constant, 1456 Register bumped_count, 1457 Register scratch2, 1458 bool decrement) { 1459 // Add the constant to reg to get the offset. 1460 add(ImethodDataPtr, reg, scratch2); 1461 Address counter(scratch2, constant); 1462 increment_mdp_data_at(counter, bumped_count, decrement); 1463 } 1464 1465 // Set a flag value at the current method data pointer position. 1466 // Updates a single byte of the header, to avoid races with other header bits. 1467 1468 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1469 Register scratch) { 1470 assert(ProfileInterpreter, "must be profiling interpreter"); 1471 // Load the data header 1472 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch); 1473 1474 // Set the flag 1475 or3(scratch, flag_constant, scratch); 1476 1477 // Store the modified header. 1478 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset())); 1479 } 1480 1481 // Test the location at some offset from the method data pointer. 1482 // If it is not equal to value, branch to the not_equal_continue Label. 1483 // Set condition codes to match the nullness of the loaded value. 1484 1485 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1486 Register value, 1487 Label& not_equal_continue, 1488 Register scratch) { 1489 assert(ProfileInterpreter, "must be profiling interpreter"); 1490 ld_ptr(ImethodDataPtr, offset, scratch); 1491 cmp(value, scratch); 1492 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue); 1493 delayed()->tst(scratch); 1494 } 1495 1496 // Update the method data pointer by the displacement located at some fixed 1497 // offset from the method data pointer. 1498 1499 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1500 Register scratch) { 1501 assert(ProfileInterpreter, "must be profiling interpreter"); 1502 ld_ptr(ImethodDataPtr, offset_of_disp, scratch); 1503 add(ImethodDataPtr, scratch, ImethodDataPtr); 1504 } 1505 1506 // Update the method data pointer by the displacement located at the 1507 // offset (reg + offset_of_disp). 1508 1509 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1510 int offset_of_disp, 1511 Register scratch) { 1512 assert(ProfileInterpreter, "must be profiling interpreter"); 1513 add(reg, offset_of_disp, scratch); 1514 ld_ptr(ImethodDataPtr, scratch, scratch); 1515 add(ImethodDataPtr, scratch, ImethodDataPtr); 1516 } 1517 1518 // Update the method data pointer by a simple constant displacement. 1519 1520 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1521 assert(ProfileInterpreter, "must be profiling interpreter"); 1522 add(ImethodDataPtr, constant, ImethodDataPtr); 1523 } 1524 1525 // Update the method data pointer for a _ret bytecode whose target 1526 // was not among our cached targets. 1527 1528 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1529 Register return_bci) { 1530 assert(ProfileInterpreter, "must be profiling interpreter"); 1531 push(state); 1532 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile 1533 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1534 ld_ptr(l_tmp, return_bci); 1535 pop(state); 1536 } 1537 1538 // Count a taken branch in the bytecodes. 1539 1540 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1541 if (ProfileInterpreter) { 1542 Label profile_continue; 1543 1544 // If no method data exists, go to profile_continue. 1545 test_method_data_pointer(profile_continue); 1546 1547 // We are taking a branch. Increment the taken count. 1548 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count); 1549 1550 // The method data pointer needs to be updated to reflect the new target. 1551 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1552 bind (profile_continue); 1553 } 1554 } 1555 1556 1557 // Count a not-taken branch in the bytecodes. 1558 1559 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) { 1560 if (ProfileInterpreter) { 1561 Label profile_continue; 1562 1563 // If no method data exists, go to profile_continue. 1564 test_method_data_pointer(profile_continue); 1565 1566 // We are taking a branch. Increment the not taken count. 1567 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch); 1568 1569 // The method data pointer needs to be updated to correspond to the 1570 // next bytecode. 1571 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1572 bind (profile_continue); 1573 } 1574 } 1575 1576 1577 // Count a non-virtual call in the bytecodes. 1578 1579 void InterpreterMacroAssembler::profile_call(Register scratch) { 1580 if (ProfileInterpreter) { 1581 Label profile_continue; 1582 1583 // If no method data exists, go to profile_continue. 1584 test_method_data_pointer(profile_continue); 1585 1586 // We are making a call. Increment the count. 1587 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1588 1589 // The method data pointer needs to be updated to reflect the new target. 1590 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1591 bind (profile_continue); 1592 } 1593 } 1594 1595 1596 // Count a final call in the bytecodes. 1597 1598 void InterpreterMacroAssembler::profile_final_call(Register scratch) { 1599 if (ProfileInterpreter) { 1600 Label profile_continue; 1601 1602 // If no method data exists, go to profile_continue. 1603 test_method_data_pointer(profile_continue); 1604 1605 // We are making a call. Increment the count. 1606 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1607 1608 // The method data pointer needs to be updated to reflect the new target. 1609 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1610 bind (profile_continue); 1611 } 1612 } 1613 1614 1615 // Count a virtual call in the bytecodes. 1616 1617 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1618 Register scratch, 1619 bool receiver_can_be_null) { 1620 if (ProfileInterpreter) { 1621 Label profile_continue; 1622 1623 // If no method data exists, go to profile_continue. 1624 test_method_data_pointer(profile_continue); 1625 1626 1627 Label skip_receiver_profile; 1628 if (receiver_can_be_null) { 1629 Label not_null; 1630 br_notnull_short(receiver, Assembler::pt, not_null); 1631 // We are making a call. Increment the count for null receiver. 1632 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1633 ba_short(skip_receiver_profile); 1634 bind(not_null); 1635 } 1636 1637 // Record the receiver type. 1638 record_klass_in_profile(receiver, scratch, true); 1639 bind(skip_receiver_profile); 1640 1641 // The method data pointer needs to be updated to reflect the new target. 1642 #if INCLUDE_JVMCI 1643 if (MethodProfileWidth == 0) { 1644 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1645 } 1646 #else 1647 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1648 #endif 1649 bind(profile_continue); 1650 } 1651 } 1652 1653 #if INCLUDE_JVMCI 1654 void InterpreterMacroAssembler::profile_called_method(Register method, Register scratch) { 1655 assert_different_registers(method, scratch); 1656 if (ProfileInterpreter && MethodProfileWidth > 0) { 1657 Label profile_continue; 1658 1659 // If no method data exists, go to profile_continue. 1660 test_method_data_pointer(profile_continue); 1661 1662 Label done; 1663 record_item_in_profile_helper(method, scratch, 0, done, MethodProfileWidth, 1664 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset())); 1665 bind(done); 1666 1667 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1668 bind(profile_continue); 1669 } 1670 } 1671 #endif // INCLUDE_JVMCI 1672 1673 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register scratch, 1674 Label& done, bool is_virtual_call) { 1675 if (TypeProfileWidth == 0) { 1676 if (is_virtual_call) { 1677 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1678 } 1679 #if INCLUDE_JVMCI 1680 else if (EnableJVMCI) { 1681 increment_mdp_data_at(in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()), scratch); 1682 } 1683 #endif 1684 } else { 1685 int non_profiled_offset = -1; 1686 if (is_virtual_call) { 1687 non_profiled_offset = in_bytes(CounterData::count_offset()); 1688 } 1689 #if INCLUDE_JVMCI 1690 else if (EnableJVMCI) { 1691 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()); 1692 } 1693 #endif 1694 1695 record_item_in_profile_helper(receiver, scratch, 0, done, TypeProfileWidth, 1696 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset); 1697 } 1698 } 1699 1700 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, 1701 Register scratch, int start_row, Label& done, int total_rows, 1702 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, 1703 int non_profiled_offset) { 1704 int last_row = total_rows - 1; 1705 assert(start_row <= last_row, "must be work left to do"); 1706 // Test this row for both the item and for null. 1707 // Take any of three different outcomes: 1708 // 1. found item => increment count and goto done 1709 // 2. found null => keep looking for case 1, maybe allocate this cell 1710 // 3. found something else => keep looking for cases 1 and 2 1711 // Case 3 is handled by a recursive call. 1712 for (int row = start_row; row <= last_row; row++) { 1713 Label next_test; 1714 bool test_for_null_also = (row == start_row); 1715 1716 // See if the item is item[n]. 1717 int item_offset = in_bytes(item_offset_fn(row)); 1718 test_mdp_data_at(item_offset, item, next_test, scratch); 1719 // delayed()->tst(scratch); 1720 1721 // The receiver is item[n]. Increment count[n]. 1722 int count_offset = in_bytes(item_count_offset_fn(row)); 1723 increment_mdp_data_at(count_offset, scratch); 1724 ba_short(done); 1725 bind(next_test); 1726 1727 if (test_for_null_also) { 1728 Label found_null; 1729 // Failed the equality check on item[n]... Test for null. 1730 if (start_row == last_row) { 1731 // The only thing left to do is handle the null case. 1732 if (non_profiled_offset >= 0) { 1733 brx(Assembler::zero, false, Assembler::pn, found_null); 1734 delayed()->nop(); 1735 // Item did not match any saved item and there is no empty row for it. 1736 // Increment total counter to indicate polymorphic case. 1737 increment_mdp_data_at(non_profiled_offset, scratch); 1738 ba_short(done); 1739 bind(found_null); 1740 } else { 1741 brx(Assembler::notZero, false, Assembler::pt, done); 1742 delayed()->nop(); 1743 } 1744 break; 1745 } 1746 // Since null is rare, make it be the branch-taken case. 1747 brx(Assembler::zero, false, Assembler::pn, found_null); 1748 delayed()->nop(); 1749 1750 // Put all the "Case 3" tests here. 1751 record_item_in_profile_helper(item, scratch, start_row + 1, done, total_rows, 1752 item_offset_fn, item_count_offset_fn, non_profiled_offset); 1753 1754 // Found a null. Keep searching for a matching item, 1755 // but remember that this is an empty (unused) slot. 1756 bind(found_null); 1757 } 1758 } 1759 1760 // In the fall-through case, we found no matching item, but we 1761 // observed the item[start_row] is NULL. 1762 1763 // Fill in the item field and increment the count. 1764 int item_offset = in_bytes(item_offset_fn(start_row)); 1765 set_mdp_data_at(item_offset, item); 1766 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1767 mov(DataLayout::counter_increment, scratch); 1768 set_mdp_data_at(count_offset, scratch); 1769 if (start_row > 0) { 1770 ba_short(done); 1771 } 1772 } 1773 1774 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1775 Register scratch, bool is_virtual_call) { 1776 assert(ProfileInterpreter, "must be profiling"); 1777 Label done; 1778 1779 record_klass_in_profile_helper(receiver, scratch, done, is_virtual_call); 1780 1781 bind (done); 1782 } 1783 1784 1785 // Count a ret in the bytecodes. 1786 1787 void InterpreterMacroAssembler::profile_ret(TosState state, 1788 Register return_bci, 1789 Register scratch) { 1790 if (ProfileInterpreter) { 1791 Label profile_continue; 1792 uint row; 1793 1794 // If no method data exists, go to profile_continue. 1795 test_method_data_pointer(profile_continue); 1796 1797 // Update the total ret count. 1798 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); 1799 1800 for (row = 0; row < RetData::row_limit(); row++) { 1801 Label next_test; 1802 1803 // See if return_bci is equal to bci[n]: 1804 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), 1805 return_bci, next_test, scratch); 1806 1807 // return_bci is equal to bci[n]. Increment the count. 1808 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch); 1809 1810 // The method data pointer needs to be updated to reflect the new target. 1811 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch); 1812 ba_short(profile_continue); 1813 bind(next_test); 1814 } 1815 1816 update_mdp_for_ret(state, return_bci); 1817 1818 bind (profile_continue); 1819 } 1820 } 1821 1822 // Profile an unexpected null in the bytecodes. 1823 void InterpreterMacroAssembler::profile_null_seen(Register scratch) { 1824 if (ProfileInterpreter) { 1825 Label profile_continue; 1826 1827 // If no method data exists, go to profile_continue. 1828 test_method_data_pointer(profile_continue); 1829 1830 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch); 1831 1832 // The method data pointer needs to be updated. 1833 int mdp_delta = in_bytes(BitData::bit_data_size()); 1834 if (TypeProfileCasts) { 1835 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size()); 1836 } 1837 update_mdp_by_constant(mdp_delta); 1838 1839 bind (profile_continue); 1840 } 1841 } 1842 1843 void InterpreterMacroAssembler::profile_typecheck(Register klass, 1844 Register scratch) { 1845 if (ProfileInterpreter) { 1846 Label profile_continue; 1847 1848 // If no method data exists, go to profile_continue. 1849 test_method_data_pointer(profile_continue); 1850 1851 int mdp_delta = in_bytes(BitData::bit_data_size()); 1852 if (TypeProfileCasts) { 1853 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size()); 1854 1855 // Record the object type. 1856 record_klass_in_profile(klass, scratch, false); 1857 } 1858 1859 // The method data pointer needs to be updated. 1860 update_mdp_by_constant(mdp_delta); 1861 1862 bind (profile_continue); 1863 } 1864 } 1865 1866 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) { 1867 if (ProfileInterpreter && TypeProfileCasts) { 1868 Label profile_continue; 1869 1870 // If no method data exists, go to profile_continue. 1871 test_method_data_pointer(profile_continue); 1872 1873 int count_offset = in_bytes(CounterData::count_offset()); 1874 // Back up the address, since we have already bumped the mdp. 1875 count_offset -= in_bytes(ReceiverTypeData::receiver_type_data_size()); 1876 1877 // *Decrement* the counter. We expect to see zero or small negatives. 1878 increment_mdp_data_at(count_offset, scratch, true); 1879 1880 bind (profile_continue); 1881 } 1882 } 1883 1884 // Count the default case of a switch construct. 1885 1886 void InterpreterMacroAssembler::profile_switch_default(Register scratch) { 1887 if (ProfileInterpreter) { 1888 Label profile_continue; 1889 1890 // If no method data exists, go to profile_continue. 1891 test_method_data_pointer(profile_continue); 1892 1893 // Update the default case count 1894 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1895 scratch); 1896 1897 // The method data pointer needs to be updated. 1898 update_mdp_by_offset( 1899 in_bytes(MultiBranchData::default_displacement_offset()), 1900 scratch); 1901 1902 bind (profile_continue); 1903 } 1904 } 1905 1906 // Count the index'th case of a switch construct. 1907 1908 void InterpreterMacroAssembler::profile_switch_case(Register index, 1909 Register scratch, 1910 Register scratch2, 1911 Register scratch3) { 1912 if (ProfileInterpreter) { 1913 Label profile_continue; 1914 1915 // If no method data exists, go to profile_continue. 1916 test_method_data_pointer(profile_continue); 1917 1918 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() 1919 set(in_bytes(MultiBranchData::per_case_size()), scratch); 1920 smul(index, scratch, scratch); 1921 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch); 1922 1923 // Update the case count 1924 increment_mdp_data_at(scratch, 1925 in_bytes(MultiBranchData::relative_count_offset()), 1926 scratch2, 1927 scratch3); 1928 1929 // The method data pointer needs to be updated. 1930 update_mdp_by_offset(scratch, 1931 in_bytes(MultiBranchData::relative_displacement_offset()), 1932 scratch2); 1933 1934 bind (profile_continue); 1935 } 1936 } 1937 1938 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) { 1939 Label not_null, do_nothing, do_update; 1940 1941 assert_different_registers(obj, mdo_addr.base(), tmp); 1942 1943 verify_oop(obj); 1944 1945 ld_ptr(mdo_addr, tmp); 1946 1947 br_notnull_short(obj, pt, not_null); 1948 or3(tmp, TypeEntries::null_seen, tmp); 1949 ba_short(do_update); 1950 1951 bind(not_null); 1952 load_klass(obj, obj); 1953 1954 xor3(obj, tmp, obj); 1955 btst(TypeEntries::type_klass_mask, obj); 1956 // klass seen before, nothing to do. The unknown bit may have been 1957 // set already but no need to check. 1958 brx(zero, false, pt, do_nothing); 1959 delayed()-> 1960 1961 btst(TypeEntries::type_unknown, obj); 1962 // already unknown. Nothing to do anymore. 1963 brx(notZero, false, pt, do_nothing); 1964 delayed()-> 1965 1966 btst(TypeEntries::type_mask, tmp); 1967 brx(zero, true, pt, do_update); 1968 // first time here. Set profile type. 1969 delayed()->or3(tmp, obj, tmp); 1970 1971 // different than before. Cannot keep accurate profile. 1972 or3(tmp, TypeEntries::type_unknown, tmp); 1973 1974 bind(do_update); 1975 // update profile 1976 st_ptr(tmp, mdo_addr); 1977 1978 bind(do_nothing); 1979 } 1980 1981 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) { 1982 if (!ProfileInterpreter) { 1983 return; 1984 } 1985 1986 assert_different_registers(callee, tmp1, tmp2, ImethodDataPtr); 1987 1988 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1989 Label profile_continue; 1990 1991 test_method_data_pointer(profile_continue); 1992 1993 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1994 1995 ldub(ImethodDataPtr, in_bytes(DataLayout::tag_offset()) - off_to_start, tmp1); 1996 cmp_and_br_short(tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag, notEqual, pn, profile_continue); 1997 1998 if (MethodData::profile_arguments()) { 1999 Label done; 2000 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 2001 add(ImethodDataPtr, off_to_args, ImethodDataPtr); 2002 2003 for (int i = 0; i < TypeProfileArgsLimit; i++) { 2004 if (i > 0 || MethodData::profile_return()) { 2005 // If return value type is profiled we may have no argument to profile 2006 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1); 2007 sub(tmp1, i*TypeStackSlotEntries::per_arg_count(), tmp1); 2008 cmp_and_br_short(tmp1, TypeStackSlotEntries::per_arg_count(), less, pn, done); 2009 } 2010 ld_ptr(Address(callee, Method::const_offset()), tmp1); 2011 lduh(Address(tmp1, ConstMethod::size_of_parameters_offset()), tmp1); 2012 // stack offset o (zero based) from the start of the argument 2013 // list, for n arguments translates into offset n - o - 1 from 2014 // the end of the argument list. But there's an extra slot at 2015 // the stop of the stack. So the offset is n - o from Lesp. 2016 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, tmp2); 2017 sub(tmp1, tmp2, tmp1); 2018 2019 // Can't use MacroAssembler::argument_address() which needs Gargs to be set up 2020 sll(tmp1, Interpreter::logStackElementSize, tmp1); 2021 ld_ptr(Lesp, tmp1, tmp1); 2022 2023 Address mdo_arg_addr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); 2024 profile_obj_type(tmp1, mdo_arg_addr, tmp2); 2025 2026 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 2027 add(ImethodDataPtr, to_add, ImethodDataPtr); 2028 off_to_args += to_add; 2029 } 2030 2031 if (MethodData::profile_return()) { 2032 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1); 2033 sub(tmp1, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count(), tmp1); 2034 } 2035 2036 bind(done); 2037 2038 if (MethodData::profile_return()) { 2039 // We're right after the type profile for the last 2040 // argument. tmp1 is the number of cells left in the 2041 // CallTypeData/VirtualCallTypeData to reach its end. Non null 2042 // if there's a return to profile. 2043 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 2044 sll(tmp1, exact_log2(DataLayout::cell_size), tmp1); 2045 add(ImethodDataPtr, tmp1, ImethodDataPtr); 2046 } 2047 } else { 2048 assert(MethodData::profile_return(), "either profile call args or call ret"); 2049 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 2050 } 2051 2052 // mdp points right after the end of the 2053 // CallTypeData/VirtualCallTypeData, right after the cells for the 2054 // return value type if there's one. 2055 2056 bind(profile_continue); 2057 } 2058 } 2059 2060 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 2061 assert_different_registers(ret, tmp1, tmp2); 2062 if (ProfileInterpreter && MethodData::profile_return()) { 2063 Label profile_continue, done; 2064 2065 test_method_data_pointer(profile_continue); 2066 2067 if (MethodData::profile_return_jsr292_only()) { 2068 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); 2069 2070 // If we don't profile all invoke bytecodes we must make sure 2071 // it's a bytecode we indeed profile. We can't go back to the 2072 // begining of the ProfileData we intend to update to check its 2073 // type because we're right after it and we don't known its 2074 // length. 2075 Label do_profile; 2076 ldub(Lbcp, 0, tmp1); 2077 cmp_and_br_short(tmp1, Bytecodes::_invokedynamic, equal, pn, do_profile); 2078 cmp(tmp1, Bytecodes::_invokehandle); 2079 br(equal, false, pn, do_profile); 2080 delayed()->lduh(Lmethod, Method::intrinsic_id_offset_in_bytes(), tmp1); 2081 cmp_and_br_short(tmp1, vmIntrinsics::_compiledLambdaForm, notEqual, pt, profile_continue); 2082 2083 bind(do_profile); 2084 } 2085 2086 Address mdo_ret_addr(ImethodDataPtr, -in_bytes(ReturnTypeEntry::size())); 2087 mov(ret, tmp1); 2088 profile_obj_type(tmp1, mdo_ret_addr, tmp2); 2089 2090 bind(profile_continue); 2091 } 2092 } 2093 2094 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) { 2095 if (ProfileInterpreter && MethodData::profile_parameters()) { 2096 Label profile_continue, done; 2097 2098 test_method_data_pointer(profile_continue); 2099 2100 // Load the offset of the area within the MDO used for 2101 // parameters. If it's negative we're not profiling any parameters. 2102 lduw(ImethodDataPtr, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), tmp1); 2103 cmp_and_br_short(tmp1, 0, less, pn, profile_continue); 2104 2105 // Compute a pointer to the area for parameters from the offset 2106 // and move the pointer to the slot for the last 2107 // parameters. Collect profiling from last parameter down. 2108 // mdo start + parameters offset + array length - 1 2109 2110 // Pointer to the parameter area in the MDO 2111 Register mdp = tmp1; 2112 add(ImethodDataPtr, tmp1, mdp); 2113 2114 // offset of the current profile entry to update 2115 Register entry_offset = tmp2; 2116 // entry_offset = array len in number of cells 2117 ld_ptr(mdp, ArrayData::array_len_offset(), entry_offset); 2118 2119 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 2120 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 2121 2122 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 2123 sub(entry_offset, TypeStackSlotEntries::per_arg_count() - (off_base / DataLayout::cell_size), entry_offset); 2124 // entry_offset in bytes 2125 sll(entry_offset, exact_log2(DataLayout::cell_size), entry_offset); 2126 2127 Label loop; 2128 bind(loop); 2129 2130 // load offset on the stack from the slot for this parameter 2131 ld_ptr(mdp, entry_offset, tmp3); 2132 sll(tmp3,Interpreter::logStackElementSize, tmp3); 2133 neg(tmp3); 2134 // read the parameter from the local area 2135 ld_ptr(Llocals, tmp3, tmp3); 2136 2137 // make entry_offset now point to the type field for this parameter 2138 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 2139 assert(type_base > off_base, "unexpected"); 2140 add(entry_offset, type_base - off_base, entry_offset); 2141 2142 // profile the parameter 2143 Address arg_type(mdp, entry_offset); 2144 profile_obj_type(tmp3, arg_type, tmp4); 2145 2146 // go to next parameter 2147 sub(entry_offset, TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base), entry_offset); 2148 cmp_and_br_short(entry_offset, off_base, greaterEqual, pt, loop); 2149 2150 bind(profile_continue); 2151 } 2152 } 2153 2154 // add a InterpMonitorElem to stack (see frame_sparc.hpp) 2155 2156 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty, 2157 Register Rtemp, 2158 Register Rtemp2 ) { 2159 2160 Register Rlimit = Lmonitors; 2161 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 2162 assert( (delta & LongAlignmentMask) == 0, 2163 "sizeof BasicObjectLock must be even number of doublewords"); 2164 2165 sub( SP, delta, SP); 2166 sub( Lesp, delta, Lesp); 2167 sub( Lmonitors, delta, Lmonitors); 2168 2169 if (!stack_is_empty) { 2170 2171 // must copy stack contents down 2172 2173 Label start_copying, next; 2174 2175 // untested("monitor stack expansion"); 2176 compute_stack_base(Rtemp); 2177 ba(start_copying); 2178 delayed()->cmp(Rtemp, Rlimit); // done? duplicated below 2179 2180 // note: must copy from low memory upwards 2181 // On entry to loop, 2182 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS) 2183 // Loop mutates Rtemp 2184 2185 bind( next); 2186 2187 st_ptr(Rtemp2, Rtemp, 0); 2188 inc(Rtemp, wordSize); 2189 cmp(Rtemp, Rlimit); // are we done? (duplicated above) 2190 2191 bind( start_copying ); 2192 2193 brx( notEqual, true, pn, next ); 2194 delayed()->ld_ptr( Rtemp, delta, Rtemp2 ); 2195 2196 // done copying stack 2197 } 2198 } 2199 2200 // Locals 2201 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) { 2202 assert_not_delayed(); 2203 sll(index, Interpreter::logStackElementSize, index); 2204 sub(Llocals, index, index); 2205 ld_ptr(index, 0, dst); 2206 // Note: index must hold the effective address--the iinc template uses it 2207 } 2208 2209 // Just like access_local_ptr but the tag is a returnAddress 2210 void InterpreterMacroAssembler::access_local_returnAddress(Register index, 2211 Register dst ) { 2212 assert_not_delayed(); 2213 sll(index, Interpreter::logStackElementSize, index); 2214 sub(Llocals, index, index); 2215 ld_ptr(index, 0, dst); 2216 } 2217 2218 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) { 2219 assert_not_delayed(); 2220 sll(index, Interpreter::logStackElementSize, index); 2221 sub(Llocals, index, index); 2222 ld(index, 0, dst); 2223 // Note: index must hold the effective address--the iinc template uses it 2224 } 2225 2226 2227 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) { 2228 assert_not_delayed(); 2229 sll(index, Interpreter::logStackElementSize, index); 2230 sub(Llocals, index, index); 2231 // First half stored at index n+1 (which grows down from Llocals[n]) 2232 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst); 2233 } 2234 2235 2236 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) { 2237 assert_not_delayed(); 2238 sll(index, Interpreter::logStackElementSize, index); 2239 sub(Llocals, index, index); 2240 ldf(FloatRegisterImpl::S, index, 0, dst); 2241 } 2242 2243 2244 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) { 2245 assert_not_delayed(); 2246 sll(index, Interpreter::logStackElementSize, index); 2247 sub(Llocals, index, index); 2248 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst); 2249 } 2250 2251 2252 #ifdef ASSERT 2253 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) { 2254 Label L; 2255 2256 assert(Rindex != Rscratch, "Registers cannot be same"); 2257 assert(Rindex != Rscratch1, "Registers cannot be same"); 2258 assert(Rlimit != Rscratch, "Registers cannot be same"); 2259 assert(Rlimit != Rscratch1, "Registers cannot be same"); 2260 assert(Rscratch1 != Rscratch, "Registers cannot be same"); 2261 2262 // untested("reg area corruption"); 2263 add(Rindex, offset, Rscratch); 2264 add(Rlimit, 64 + STACK_BIAS, Rscratch1); 2265 cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L); 2266 stop("regsave area is being clobbered"); 2267 bind(L); 2268 } 2269 #endif // ASSERT 2270 2271 2272 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) { 2273 assert_not_delayed(); 2274 sll(index, Interpreter::logStackElementSize, index); 2275 sub(Llocals, index, index); 2276 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);) 2277 st(src, index, 0); 2278 } 2279 2280 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) { 2281 assert_not_delayed(); 2282 sll(index, Interpreter::logStackElementSize, index); 2283 sub(Llocals, index, index); 2284 #ifdef ASSERT 2285 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2286 #endif 2287 st_ptr(src, index, 0); 2288 } 2289 2290 2291 2292 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) { 2293 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n)); 2294 } 2295 2296 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) { 2297 assert_not_delayed(); 2298 sll(index, Interpreter::logStackElementSize, index); 2299 sub(Llocals, index, index); 2300 #ifdef ASSERT 2301 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2302 #endif 2303 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1 2304 } 2305 2306 2307 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) { 2308 assert_not_delayed(); 2309 sll(index, Interpreter::logStackElementSize, index); 2310 sub(Llocals, index, index); 2311 #ifdef ASSERT 2312 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); 2313 #endif 2314 stf(FloatRegisterImpl::S, src, index, 0); 2315 } 2316 2317 2318 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) { 2319 assert_not_delayed(); 2320 sll(index, Interpreter::logStackElementSize, index); 2321 sub(Llocals, index, index); 2322 #ifdef ASSERT 2323 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); 2324 #endif 2325 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1)); 2326 } 2327 2328 2329 int InterpreterMacroAssembler::top_most_monitor_byte_offset() { 2330 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; 2331 int rounded_vm_local_words = align_up((int)frame::interpreter_frame_vm_local_words, WordsPerLong); 2332 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS; 2333 } 2334 2335 2336 Address InterpreterMacroAssembler::top_most_monitor() { 2337 return Address(FP, top_most_monitor_byte_offset()); 2338 } 2339 2340 2341 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) { 2342 add( Lesp, wordSize, Rdest ); 2343 } 2344 2345 void InterpreterMacroAssembler::get_method_counters(Register method, 2346 Register Rcounters, 2347 Label& skip) { 2348 Label has_counters; 2349 Address method_counters(method, in_bytes(Method::method_counters_offset())); 2350 ld_ptr(method_counters, Rcounters); 2351 br_notnull_short(Rcounters, Assembler::pt, has_counters); 2352 call_VM(noreg, CAST_FROM_FN_PTR(address, 2353 InterpreterRuntime::build_method_counters), method); 2354 ld_ptr(method_counters, Rcounters); 2355 br_null(Rcounters, false, Assembler::pn, skip); // No MethodCounters, OutOfMemory 2356 delayed()->nop(); 2357 bind(has_counters); 2358 } 2359 2360 void InterpreterMacroAssembler::increment_invocation_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) { 2361 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 2362 assert_different_registers(Rcounters, Rtmp, Rtmp2); 2363 2364 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() + 2365 InvocationCounter::counter_offset()); 2366 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() + 2367 InvocationCounter::counter_offset()); 2368 int delta = InvocationCounter::count_increment; 2369 2370 // Load each counter in a register 2371 ld( inv_counter, Rtmp ); 2372 ld( be_counter, Rtmp2 ); 2373 2374 assert( is_simm13( delta ), " delta too large."); 2375 2376 // Add the delta to the invocation counter and store the result 2377 add( Rtmp, delta, Rtmp ); 2378 2379 // Mask the backedge counter 2380 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2381 2382 // Store value 2383 st( Rtmp, inv_counter); 2384 2385 // Add invocation counter + backedge counter 2386 add( Rtmp, Rtmp2, Rtmp); 2387 2388 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 2389 } 2390 2391 void InterpreterMacroAssembler::increment_backedge_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) { 2392 assert(UseCompiler, "incrementing must be useful"); 2393 assert_different_registers(Rcounters, Rtmp, Rtmp2); 2394 2395 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() + 2396 InvocationCounter::counter_offset()); 2397 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() + 2398 InvocationCounter::counter_offset()); 2399 2400 int delta = InvocationCounter::count_increment; 2401 // Load each counter in a register 2402 ld( be_counter, Rtmp ); 2403 ld( inv_counter, Rtmp2 ); 2404 2405 // Add the delta to the backedge counter 2406 add( Rtmp, delta, Rtmp ); 2407 2408 // Mask the invocation counter, add to backedge counter 2409 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); 2410 2411 // and store the result to memory 2412 st( Rtmp, be_counter ); 2413 2414 // Add backedge + invocation counter 2415 add( Rtmp, Rtmp2, Rtmp ); 2416 2417 // Note that this macro must leave backedge_count + invocation_count in Rtmp! 2418 } 2419 2420 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count, 2421 Register method_counters, 2422 Register branch_bcp, 2423 Register Rtmp ) { 2424 Label did_not_overflow; 2425 Label overflow_with_error; 2426 assert_different_registers(backedge_count, Rtmp, branch_bcp); 2427 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 2428 2429 Address limit(method_counters, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())); 2430 ld(limit, Rtmp); 2431 cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow); 2432 2433 // When ProfileInterpreter is on, the backedge_count comes from the 2434 // MethodData*, which value does not get reset on the call to 2435 // frequency_counter_overflow(). To avoid excessive calls to the overflow 2436 // routine while the method is being compiled, add a second test to make sure 2437 // the overflow function is called only once every overflow_frequency. 2438 if (ProfileInterpreter) { 2439 const int overflow_frequency = 1024; 2440 andcc(backedge_count, overflow_frequency-1, Rtmp); 2441 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow); 2442 delayed()->nop(); 2443 } 2444 2445 // overflow in loop, pass branch bytecode 2446 set(6,Rtmp); 2447 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp); 2448 2449 // Was an OSR adapter generated? 2450 // O0 = osr nmethod 2451 br_null_short(O0, Assembler::pn, overflow_with_error); 2452 2453 // Has the nmethod been invalidated already? 2454 ldub(O0, nmethod::state_offset(), O2); 2455 cmp_and_br_short(O2, nmethod::in_use, Assembler::notEqual, Assembler::pn, overflow_with_error); 2456 2457 // migrate the interpreter frame off of the stack 2458 2459 mov(G2_thread, L7); 2460 // save nmethod 2461 mov(O0, L6); 2462 set_last_Java_frame(SP, noreg); 2463 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7); 2464 reset_last_Java_frame(); 2465 mov(L7, G2_thread); 2466 2467 // move OSR nmethod to I1 2468 mov(L6, I1); 2469 2470 // OSR buffer to I0 2471 mov(O0, I0); 2472 2473 // remove the interpreter frame 2474 restore(I5_savedSP, 0, SP); 2475 2476 // Jump to the osr code. 2477 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2); 2478 jmp(O2, G0); 2479 delayed()->nop(); 2480 2481 bind(overflow_with_error); 2482 2483 bind(did_not_overflow); 2484 } 2485 2486 2487 2488 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) { 2489 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); } 2490 } 2491 2492 2493 // local helper function for the verify_oop_or_return_address macro 2494 static bool verify_return_address(Method* m, int bci) { 2495 #ifndef PRODUCT 2496 address pc = (address)(m->constMethod()) 2497 + in_bytes(ConstMethod::codes_offset()) + bci; 2498 // assume it is a valid return address if it is inside m and is preceded by a jsr 2499 if (!m->contains(pc)) return false; 2500 address jsr_pc; 2501 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2502 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2503 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2504 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2505 #endif // PRODUCT 2506 return false; 2507 } 2508 2509 2510 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2511 if (!VerifyOops) return; 2512 // the VM documentation for the astore[_wide] bytecode allows 2513 // the TOS to be not only an oop but also a return address 2514 Label test; 2515 Label skip; 2516 // See if it is an address (in the current method): 2517 2518 mov(reg, Rtmp); 2519 const int log2_bytecode_size_limit = 16; 2520 srl(Rtmp, log2_bytecode_size_limit, Rtmp); 2521 br_notnull_short( Rtmp, pt, test ); 2522 2523 // %%% should use call_VM_leaf here? 2524 save_frame_and_mov(0, Lmethod, O0, reg, O1); 2525 save_thread(L7_thread_cache); 2526 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none); 2527 delayed()->nop(); 2528 restore_thread(L7_thread_cache); 2529 br_notnull( O0, false, pt, skip ); 2530 delayed()->restore(); 2531 2532 // Perform a more elaborate out-of-line call 2533 // Not an address; verify it: 2534 bind(test); 2535 verify_oop(reg); 2536 bind(skip); 2537 } 2538 2539 2540 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2541 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); 2542 } 2543 2544 2545 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 2546 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 2547 int increment, Address mask_addr, 2548 Register scratch1, Register scratch2, 2549 Condition cond, Label *where) { 2550 ld(counter_addr, scratch1); 2551 add(scratch1, increment, scratch1); 2552 ld(mask_addr, scratch2); 2553 andcc(scratch1, scratch2, G0); 2554 br(cond, false, Assembler::pn, *where); 2555 delayed()->st(scratch1, counter_addr); 2556 } 2557 2558 // Inline assembly for: 2559 // 2560 // if (thread is in interp_only_mode) { 2561 // InterpreterRuntime::post_method_entry(); 2562 // } 2563 // if (DTraceMethodProbes) { 2564 // SharedRuntime::dtrace_method_entry(method, receiver); 2565 // } 2566 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { 2567 // SharedRuntime::rc_trace_method_entry(method, receiver); 2568 // } 2569 2570 void InterpreterMacroAssembler::notify_method_entry() { 2571 2572 // Whenever JVMTI puts a thread in interp_only_mode, method 2573 // entry/exit events are sent for that thread to track stack 2574 // depth. If it is possible to enter interp_only_mode we add 2575 // the code to check if the event should be sent. 2576 if (JvmtiExport::can_post_interpreter_events()) { 2577 Label L; 2578 Register temp_reg = O5; 2579 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2580 ld(interp_only, temp_reg); 2581 cmp_and_br_short(temp_reg, 0, equal, pt, L); 2582 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); 2583 bind(L); 2584 } 2585 2586 { 2587 Register temp_reg = O5; 2588 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2589 call_VM_leaf(noreg, 2590 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 2591 G2_thread, Lmethod); 2592 } 2593 2594 // RedefineClasses() tracing support for obsolete method entry 2595 if (log_is_enabled(Trace, redefine, class, obsolete)) { 2596 call_VM_leaf(noreg, 2597 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 2598 G2_thread, Lmethod); 2599 } 2600 } 2601 2602 2603 // Inline assembly for: 2604 // 2605 // if (thread is in interp_only_mode) { 2606 // // save result 2607 // InterpreterRuntime::post_method_exit(); 2608 // // restore result 2609 // } 2610 // if (DTraceMethodProbes) { 2611 // SharedRuntime::dtrace_method_exit(thread, method); 2612 // } 2613 // 2614 // Native methods have their result stored in d_tmp and l_tmp 2615 // Java methods have their result stored in the expression stack 2616 2617 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, 2618 TosState state, 2619 NotifyMethodExitMode mode) { 2620 2621 // Whenever JVMTI puts a thread in interp_only_mode, method 2622 // entry/exit events are sent for that thread to track stack 2623 // depth. If it is possible to enter interp_only_mode we add 2624 // the code to check if the event should be sent. 2625 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2626 Label L; 2627 Register temp_reg = O5; 2628 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); 2629 ld(interp_only, temp_reg); 2630 cmp_and_br_short(temp_reg, 0, equal, pt, L); 2631 2632 // Note: frame::interpreter_frame_result has a dependency on how the 2633 // method result is saved across the call to post_method_exit. For 2634 // native methods it assumes the result registers are saved to 2635 // l_scratch and d_scratch. If this changes then the interpreter_frame_result 2636 // implementation will need to be updated too. 2637 2638 save_return_value(state, is_native_method); 2639 call_VM(noreg, 2640 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2641 restore_return_value(state, is_native_method); 2642 bind(L); 2643 } 2644 2645 { 2646 Register temp_reg = O5; 2647 // Dtrace notification 2648 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); 2649 save_return_value(state, is_native_method); 2650 call_VM_leaf( 2651 noreg, 2652 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2653 G2_thread, Lmethod); 2654 restore_return_value(state, is_native_method); 2655 } 2656 } 2657 2658 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) { 2659 if (is_native_call) { 2660 stf(FloatRegisterImpl::D, F0, d_tmp); 2661 stx(O0, l_tmp); 2662 } else { 2663 push(state); 2664 } 2665 } 2666 2667 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) { 2668 if (is_native_call) { 2669 ldf(FloatRegisterImpl::D, d_tmp, F0); 2670 ldx(l_tmp, O0); 2671 } else { 2672 pop(state); 2673 } 2674 }