interp_masm_ppc_64.cpp revision 11486:3950d1713ffa
1/* 2 * Copyright (c) 2003, 2016, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2012, 2016 SAP SE. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 27#include "precompiled.hpp" 28#include "asm/macroAssembler.inline.hpp" 29#include "interp_masm_ppc_64.hpp" 30#include "interpreter/interpreterRuntime.hpp" 31#include "prims/jvmtiThreadState.hpp" 32#include "runtime/sharedRuntime.hpp" 33 34#ifdef PRODUCT 35#define BLOCK_COMMENT(str) // nothing 36#else 37#define BLOCK_COMMENT(str) block_comment(str) 38#endif 39 40void InterpreterMacroAssembler::null_check_throw(Register a, int offset, Register temp_reg) { 41 address exception_entry = Interpreter::throw_NullPointerException_entry(); 42 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry); 43} 44 45void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) { 46 assert(entry, "Entry must have been generated by now"); 47 if (is_within_range_of_b(entry, pc())) { 48 b(entry); 49 } else { 50 load_const_optimized(Rscratch, entry, R0); 51 mtctr(Rscratch); 52 bctr(); 53 } 54} 55 56void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { 57 Register bytecode = R12_scratch2; 58 if (bcp_incr != 0) { 59 lbzu(bytecode, bcp_incr, R14_bcp); 60 } else { 61 lbz(bytecode, 0, R14_bcp); 62 } 63 64 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state)); 65} 66 67void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 68 // Load current bytecode. 69 Register bytecode = R12_scratch2; 70 lbz(bytecode, 0, R14_bcp); 71 dispatch_Lbyte_code(state, bytecode, table); 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 R24_dispatch_addr. 76void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 77 Register bytecode = R12_scratch2; 78 lbz(bytecode, bcp_incr, R14_bcp); 79 80 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state)); 81 82 sldi(bytecode, bytecode, LogBytesPerWord); 83 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode); 84} 85 86// Dispatch code executed in the epilog of a bytecode which does not do it's 87// own dispatch. The dispatch address in R24_dispatch_addr is used for the 88// dispatch. 89void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 90 if (bcp_incr) { addi(R14_bcp, R14_bcp, bcp_incr); } 91 mtctr(R24_dispatch_addr); 92 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable); 93} 94 95void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 96 assert(scratch_reg != R0, "can't use R0 as scratch_reg here"); 97 if (JvmtiExport::can_pop_frame()) { 98 Label L; 99 100 // Check the "pending popframe condition" flag in the current thread. 101 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 102 103 // Initiate popframe handling only if it is not already being 104 // processed. If the flag has the popframe_processing bit set, it 105 // means that this code is called *during* popframe handling - we 106 // don't want to reenter. 107 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit); 108 beq(CCR0, L); 109 110 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit); 111 bne(CCR0, L); 112 113 // Call the Interpreter::remove_activation_preserving_args_entry() 114 // func to get the address of the same-named entrypoint in the 115 // generated interpreter code. 116#if defined(ABI_ELFv2) 117 call_c(CAST_FROM_FN_PTR(address, 118 Interpreter::remove_activation_preserving_args_entry), 119 relocInfo::none); 120#else 121 call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, 122 Interpreter::remove_activation_preserving_args_entry), 123 relocInfo::none); 124#endif 125 126 // Jump to Interpreter::_remove_activation_preserving_args_entry. 127 mtctr(R3_RET); 128 bctr(); 129 130 align(32, 12); 131 bind(L); 132 } 133} 134 135void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 136 const Register Rthr_state_addr = scratch_reg; 137 if (JvmtiExport::can_force_early_return()) { 138 Label Lno_early_ret; 139 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 140 cmpdi(CCR0, Rthr_state_addr, 0); 141 beq(CCR0, Lno_early_ret); 142 143 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr); 144 cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending); 145 bne(CCR0, Lno_early_ret); 146 147 // Jump to Interpreter::_earlyret_entry. 148 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr); 149 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry)); 150 mtlr(R3_RET); 151 blr(); 152 153 align(32, 12); 154 bind(Lno_early_ret); 155 } 156} 157 158void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) { 159 const Register RjvmtiState = Rscratch1; 160 const Register Rscratch2 = R0; 161 162 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 163 li(Rscratch2, 0); 164 165 switch (state) { 166 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 167 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 168 break; 169 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 170 break; 171 case btos: // fall through 172 case ztos: // fall through 173 case ctos: // fall through 174 case stos: // fall through 175 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 176 break; 177 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 178 break; 179 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 180 break; 181 case vtos: break; 182 default : ShouldNotReachHere(); 183 } 184 185 // Clean up tos value in the jvmti thread state. 186 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 187 // Set tos state field to illegal value. 188 li(Rscratch2, ilgl); 189 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState); 190} 191 192// Common code to dispatch and dispatch_only. 193// Dispatch value in Lbyte_code and increment Lbcp. 194 195void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) { 196 address table_base = (address)Interpreter::dispatch_table((TosState)0); 197 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base; 198 if (is_simm16(table_offs)) { 199 addi(dst, R25_templateTableBase, (int)table_offs); 200 } else { 201 load_const_optimized(dst, table, R0); 202 } 203} 204 205void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode, 206 address* table, bool verify) { 207 if (verify) { 208 unimplemented("dispatch_Lbyte_code: verify"); // See Sparc Implementation to implement this 209 } 210 211 assert_different_registers(bytecode, R11_scratch1); 212 213 // Calc dispatch table address. 214 load_dispatch_table(R11_scratch1, table); 215 216 sldi(R12_scratch2, bytecode, LogBytesPerWord); 217 ldx(R11_scratch1, R11_scratch1, R12_scratch2); 218 219 // Jump off! 220 mtctr(R11_scratch1); 221 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable); 222} 223 224void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) { 225 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize); 226 ldx(Rrecv_dst, Rrecv_dst, R15_esp); 227} 228 229// helpers for expression stack 230 231void InterpreterMacroAssembler::pop_i(Register r) { 232 lwzu(r, Interpreter::stackElementSize, R15_esp); 233} 234 235void InterpreterMacroAssembler::pop_ptr(Register r) { 236 ldu(r, Interpreter::stackElementSize, R15_esp); 237} 238 239void InterpreterMacroAssembler::pop_l(Register r) { 240 ld(r, Interpreter::stackElementSize, R15_esp); 241 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 242} 243 244void InterpreterMacroAssembler::pop_f(FloatRegister f) { 245 lfsu(f, Interpreter::stackElementSize, R15_esp); 246} 247 248void InterpreterMacroAssembler::pop_d(FloatRegister f) { 249 lfd(f, Interpreter::stackElementSize, R15_esp); 250 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 251} 252 253void InterpreterMacroAssembler::push_i(Register r) { 254 stw(r, 0, R15_esp); 255 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 256} 257 258void InterpreterMacroAssembler::push_ptr(Register r) { 259 std(r, 0, R15_esp); 260 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 261} 262 263void InterpreterMacroAssembler::push_l(Register r) { 264 std(r, - Interpreter::stackElementSize, R15_esp); 265 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 266} 267 268void InterpreterMacroAssembler::push_f(FloatRegister f) { 269 stfs(f, 0, R15_esp); 270 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 271} 272 273void InterpreterMacroAssembler::push_d(FloatRegister f) { 274 stfd(f, - Interpreter::stackElementSize, R15_esp); 275 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 276} 277 278void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) { 279 std(first, 0, R15_esp); 280 std(second, -Interpreter::stackElementSize, R15_esp); 281 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 282} 283 284void InterpreterMacroAssembler::push_l_pop_d(Register l, FloatRegister d) { 285 std(l, 0, R15_esp); 286 lfd(d, 0, R15_esp); 287} 288 289void InterpreterMacroAssembler::push_d_pop_l(FloatRegister d, Register l) { 290 stfd(d, 0, R15_esp); 291 ld(l, 0, R15_esp); 292} 293 294void InterpreterMacroAssembler::push(TosState state) { 295 switch (state) { 296 case atos: push_ptr(); break; 297 case btos: 298 case ztos: 299 case ctos: 300 case stos: 301 case itos: push_i(); break; 302 case ltos: push_l(); break; 303 case ftos: push_f(); break; 304 case dtos: push_d(); break; 305 case vtos: /* nothing to do */ break; 306 default : ShouldNotReachHere(); 307 } 308} 309 310void InterpreterMacroAssembler::pop(TosState state) { 311 switch (state) { 312 case atos: pop_ptr(); break; 313 case btos: 314 case ztos: 315 case ctos: 316 case stos: 317 case itos: pop_i(); break; 318 case ltos: pop_l(); break; 319 case ftos: pop_f(); break; 320 case dtos: pop_d(); break; 321 case vtos: /* nothing to do */ break; 322 default : ShouldNotReachHere(); 323 } 324 verify_oop(R17_tos, state); 325} 326 327void InterpreterMacroAssembler::empty_expression_stack() { 328 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 329} 330 331void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset, 332 Register Rdst, 333 signedOrNot is_signed) { 334#if defined(VM_LITTLE_ENDIAN) 335 if (bcp_offset) { 336 load_const_optimized(Rdst, bcp_offset); 337 lhbrx(Rdst, R14_bcp, Rdst); 338 } else { 339 lhbrx(Rdst, R14_bcp); 340 } 341 if (is_signed == Signed) { 342 extsh(Rdst, Rdst); 343 } 344#else 345 // Read Java big endian format. 346 if (is_signed == Signed) { 347 lha(Rdst, bcp_offset, R14_bcp); 348 } else { 349 lhz(Rdst, bcp_offset, R14_bcp); 350 } 351#endif 352} 353 354void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset, 355 Register Rdst, 356 signedOrNot is_signed) { 357#if defined(VM_LITTLE_ENDIAN) 358 if (bcp_offset) { 359 load_const_optimized(Rdst, bcp_offset); 360 lwbrx(Rdst, R14_bcp, Rdst); 361 } else { 362 lwbrx(Rdst, R14_bcp); 363 } 364 if (is_signed == Signed) { 365 extsw(Rdst, Rdst); 366 } 367#else 368 // Read Java big endian format. 369 if (bcp_offset & 3) { // Offset unaligned? 370 load_const_optimized(Rdst, bcp_offset); 371 if (is_signed == Signed) { 372 lwax(Rdst, R14_bcp, Rdst); 373 } else { 374 lwzx(Rdst, R14_bcp, Rdst); 375 } 376 } else { 377 if (is_signed == Signed) { 378 lwa(Rdst, bcp_offset, R14_bcp); 379 } else { 380 lwz(Rdst, bcp_offset, R14_bcp); 381 } 382 } 383#endif 384} 385 386 387// Load the constant pool cache index from the bytecode stream. 388// 389// Kills / writes: 390// - Rdst, Rscratch 391void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, 392 size_t index_size) { 393 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 394 // Cache index is always in the native format, courtesy of Rewriter. 395 if (index_size == sizeof(u2)) { 396 lhz(Rdst, bcp_offset, R14_bcp); 397 } else if (index_size == sizeof(u4)) { 398 if (bcp_offset & 3) { 399 load_const_optimized(Rdst, bcp_offset); 400 lwax(Rdst, R14_bcp, Rdst); 401 } else { 402 lwa(Rdst, bcp_offset, R14_bcp); 403 } 404 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 405 nand(Rdst, Rdst, Rdst); // convert to plain index 406 } else if (index_size == sizeof(u1)) { 407 lbz(Rdst, bcp_offset, R14_bcp); 408 } else { 409 ShouldNotReachHere(); 410 } 411 // Rdst now contains cp cache index. 412} 413 414void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset, 415 size_t index_size) { 416 get_cache_index_at_bcp(cache, bcp_offset, index_size); 417 sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord)); 418 add(cache, R27_constPoolCache, cache); 419} 420 421// Load 4-byte signed or unsigned integer in Java format (that is, big-endian format) 422// from (Rsrc)+offset. 423void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset, 424 signedOrNot is_signed) { 425#if defined(VM_LITTLE_ENDIAN) 426 if (offset) { 427 load_const_optimized(Rdst, offset); 428 lwbrx(Rdst, Rdst, Rsrc); 429 } else { 430 lwbrx(Rdst, Rsrc); 431 } 432 if (is_signed == Signed) { 433 extsw(Rdst, Rdst); 434 } 435#else 436 if (is_signed == Signed) { 437 lwa(Rdst, offset, Rsrc); 438 } else { 439 lwz(Rdst, offset, Rsrc); 440 } 441#endif 442} 443 444// Load object from cpool->resolved_references(index). 445void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index, Label *is_null) { 446 assert_different_registers(result, index); 447 get_constant_pool(result); 448 449 // Convert from field index to resolved_references() index and from 450 // word index to byte offset. Since this is a java object, it can be compressed. 451 Register tmp = index; // reuse 452 sldi(tmp, index, LogBytesPerHeapOop); 453 // Load pointer for resolved_references[] objArray. 454 ld(result, ConstantPool::resolved_references_offset_in_bytes(), result); 455 // JNIHandles::resolve(result) 456 ld(result, 0, result); 457#ifdef ASSERT 458 Label index_ok; 459 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result); 460 sldi(R0, R0, LogBytesPerHeapOop); 461 cmpd(CCR0, tmp, R0); 462 blt(CCR0, index_ok); 463 stop("resolved reference index out of bounds", 0x09256); 464 bind(index_ok); 465#endif 466 // Add in the index. 467 add(result, tmp, result); 468 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result, is_null); 469} 470 471// Generate a subtype check: branch to ok_is_subtype if sub_klass is 472// a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2. 473void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1, 474 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) { 475 // Profile the not-null value's klass. 476 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2); 477 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype); 478 profile_typecheck_failed(Rtmp1, Rtmp2); 479} 480 481// Separate these two to allow for delay slot in middle. 482// These are used to do a test and full jump to exception-throwing code. 483 484// Check that index is in range for array, then shift index by index_shift, 485// and put arrayOop + shifted_index into res. 486// Note: res is still shy of address by array offset into object. 487 488void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, 489 int index_shift, Register Rtmp, Register Rres) { 490 // Check that index is in range for array, then shift index by index_shift, 491 // and put arrayOop + shifted_index into res. 492 // Note: res is still shy of address by array offset into object. 493 // Kills: 494 // - Rindex 495 // Writes: 496 // - Rres: Address that corresponds to the array index if check was successful. 497 verify_oop(Rarray); 498 const Register Rlength = R0; 499 const Register RsxtIndex = Rtmp; 500 Label LisNull, LnotOOR; 501 502 // Array nullcheck 503 if (!ImplicitNullChecks) { 504 cmpdi(CCR0, Rarray, 0); 505 beq(CCR0, LisNull); 506 } else { 507 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex); 508 } 509 510 // Rindex might contain garbage in upper bits (remember that we don't sign extend 511 // during integer arithmetic operations). So kill them and put value into same register 512 // where ArrayIndexOutOfBounds would expect the index in. 513 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit 514 515 // Index check 516 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray); 517 cmplw(CCR0, Rindex, Rlength); 518 sldi(RsxtIndex, RsxtIndex, index_shift); 519 blt(CCR0, LnotOOR); 520 // Index should be in R17_tos, array should be in R4_ARG2. 521 mr_if_needed(R17_tos, Rindex); 522 mr_if_needed(R4_ARG2, Rarray); 523 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry); 524 mtctr(Rtmp); 525 bctr(); 526 527 if (!ImplicitNullChecks) { 528 bind(LisNull); 529 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry); 530 mtctr(Rtmp); 531 bctr(); 532 } 533 534 align(32, 16); 535 bind(LnotOOR); 536 537 // Calc address 538 add(Rres, RsxtIndex, Rarray); 539} 540 541void InterpreterMacroAssembler::index_check(Register array, Register index, 542 int index_shift, Register tmp, Register res) { 543 // pop array 544 pop_ptr(array); 545 546 // check array 547 index_check_without_pop(array, index, index_shift, tmp, res); 548} 549 550void InterpreterMacroAssembler::get_const(Register Rdst) { 551 ld(Rdst, in_bytes(Method::const_offset()), R19_method); 552} 553 554void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 555 get_const(Rdst); 556 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 557} 558 559void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 560 get_constant_pool(Rdst); 561 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 562} 563 564void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 565 get_constant_pool(Rcpool); 566 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool); 567} 568 569// Unlock if synchronized method. 570// 571// Unlock the receiver if this is a synchronized method. 572// Unlock any Java monitors from synchronized blocks. 573// 574// If there are locked Java monitors 575// If throw_monitor_exception 576// throws IllegalMonitorStateException 577// Else if install_monitor_exception 578// installs IllegalMonitorStateException 579// Else 580// no error processing 581void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 582 bool throw_monitor_exception, 583 bool install_monitor_exception) { 584 Label Lunlocked, Lno_unlock; 585 { 586 Register Rdo_not_unlock_flag = R11_scratch1; 587 Register Raccess_flags = R12_scratch2; 588 589 // Check if synchronized method or unlocking prevented by 590 // JavaThread::do_not_unlock_if_synchronized flag. 591 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 592 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method); 593 li(R0, 0); 594 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag 595 596 push(state); 597 598 // Skip if we don't have to unlock. 599 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0. 600 beq(CCR0, Lunlocked); 601 602 cmpwi(CCR0, Rdo_not_unlock_flag, 0); 603 bne(CCR0, Lno_unlock); 604 } 605 606 // Unlock 607 { 608 Register Rmonitor_base = R11_scratch1; 609 610 Label Lunlock; 611 // If it's still locked, everything is ok, unlock it. 612 ld(Rmonitor_base, 0, R1_SP); 613 addi(Rmonitor_base, Rmonitor_base, 614 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base 615 616 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base); 617 cmpdi(CCR0, R0, 0); 618 bne(CCR0, Lunlock); 619 620 // If it's already unlocked, throw exception. 621 if (throw_monitor_exception) { 622 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 623 should_not_reach_here(); 624 } else { 625 if (install_monitor_exception) { 626 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 627 b(Lunlocked); 628 } 629 } 630 631 bind(Lunlock); 632 unlock_object(Rmonitor_base); 633 } 634 635 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not. 636 bind(Lunlocked); 637 { 638 Label Lexception, Lrestart; 639 Register Rcurrent_obj_addr = R11_scratch1; 640 const int delta = frame::interpreter_frame_monitor_size_in_bytes(); 641 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords"); 642 643 bind(Lrestart); 644 // Set up search loop: Calc num of iterations. 645 { 646 Register Riterations = R12_scratch2; 647 Register Rmonitor_base = Rcurrent_obj_addr; 648 ld(Rmonitor_base, 0, R1_SP); 649 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base 650 651 subf_(Riterations, R26_monitor, Rmonitor_base); 652 ble(CCR0, Lno_unlock); 653 654 addi(Rcurrent_obj_addr, Rmonitor_base, 655 BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes()); 656 // Check if any monitor is on stack, bail out if not 657 srdi(Riterations, Riterations, exact_log2(delta)); 658 mtctr(Riterations); 659 } 660 661 // The search loop: Look for locked monitors. 662 { 663 const Register Rcurrent_obj = R0; 664 Label Lloop; 665 666 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 667 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 668 bind(Lloop); 669 670 // Check if current entry is used. 671 cmpdi(CCR0, Rcurrent_obj, 0); 672 bne(CCR0, Lexception); 673 // Preload next iteration's compare value. 674 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 675 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 676 bdnz(Lloop); 677 } 678 // Fell through: Everything's unlocked => finish. 679 b(Lno_unlock); 680 681 // An object is still locked => need to throw exception. 682 bind(Lexception); 683 if (throw_monitor_exception) { 684 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 685 should_not_reach_here(); 686 } else { 687 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 688 // Unlock does not block, so don't have to worry about the frame. 689 Register Rmonitor_addr = R11_scratch1; 690 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta); 691 unlock_object(Rmonitor_addr); 692 if (install_monitor_exception) { 693 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 694 } 695 b(Lrestart); 696 } 697 } 698 699 align(32, 12); 700 bind(Lno_unlock); 701 pop(state); 702} 703 704// Support function for remove_activation & Co. 705void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, 706 Register Rscratch1, Register Rscratch2) { 707 // Pop interpreter frame. 708 ld(Rscratch1, 0, R1_SP); // *SP 709 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp 710 ld(Rscratch2, 0, Rscratch1); // **SP 711#ifdef ASSERT 712 { 713 Label Lok; 714 ld(R0, _ijava_state_neg(ijava_reserved), Rscratch1); 715 cmpdi(CCR0, R0, 0x5afe); 716 beq(CCR0, Lok); 717 stop("frame corrupted (remove activation)", 0x5afe); 718 bind(Lok); 719 } 720#endif 721 if (return_pc!=noreg) { 722 ld(return_pc, _abi(lr), Rscratch1); // LR 723 } 724 725 // Merge top frames. 726 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP 727 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP 728} 729 730void InterpreterMacroAssembler::narrow(Register result) { 731 Register ret_type = R11_scratch1; 732 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method); 733 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1); 734 735 Label notBool, notByte, notChar, done; 736 737 // common case first 738 cmpwi(CCR0, ret_type, T_INT); 739 beq(CCR0, done); 740 741 cmpwi(CCR0, ret_type, T_BOOLEAN); 742 bne(CCR0, notBool); 743 andi(result, result, 0x1); 744 b(done); 745 746 bind(notBool); 747 cmpwi(CCR0, ret_type, T_BYTE); 748 bne(CCR0, notByte); 749 extsb(result, result); 750 b(done); 751 752 bind(notByte); 753 cmpwi(CCR0, ret_type, T_CHAR); 754 bne(CCR0, notChar); 755 andi(result, result, 0xffff); 756 b(done); 757 758 bind(notChar); 759 // cmpwi(CCR0, ret_type, T_SHORT); // all that's left 760 // bne(CCR0, done); 761 extsh(result, result); 762 763 // Nothing to do for T_INT 764 bind(done); 765} 766 767// Remove activation. 768// 769// Unlock the receiver if this is a synchronized method. 770// Unlock any Java monitors from synchronized blocks. 771// Remove the activation from the stack. 772// 773// If there are locked Java monitors 774// If throw_monitor_exception 775// throws IllegalMonitorStateException 776// Else if install_monitor_exception 777// installs IllegalMonitorStateException 778// Else 779// no error processing 780void InterpreterMacroAssembler::remove_activation(TosState state, 781 bool throw_monitor_exception, 782 bool install_monitor_exception) { 783 BLOCK_COMMENT("remove_activation {"); 784 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 785 786 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 787 notify_method_exit(false, state, NotifyJVMTI, true); 788 789 BLOCK_COMMENT("reserved_stack_check:"); 790 if (StackReservedPages > 0) { 791 // Test if reserved zone needs to be enabled. 792 Label no_reserved_zone_enabling; 793 794 // Compare frame pointers. There is no good stack pointer, as with stack 795 // frame compression we can get different SPs when we do calls. A subsequent 796 // call could have a smaller SP, so that this compare succeeds for an 797 // inner call of the method annotated with ReservedStack. 798 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread); 799 ld_ptr(R11_scratch1, _abi(callers_sp), R1_SP); // Load frame pointer. 800 cmpld(CCR0, R11_scratch1, R0); 801 blt_predict_taken(CCR0, no_reserved_zone_enabling); 802 803 // Enable reserved zone again, throw stack overflow exception. 804 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread); 805 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError)); 806 807 should_not_reach_here(); 808 809 bind(no_reserved_zone_enabling); 810 } 811 812 verify_oop(R17_tos, state); 813 verify_thread(); 814 815 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 816 mtlr(R0); 817 BLOCK_COMMENT("} remove_activation"); 818} 819 820// Lock object 821// 822// Registers alive 823// monitor - Address of the BasicObjectLock to be used for locking, 824// which must be initialized with the object to lock. 825// object - Address of the object to be locked. 826// 827void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 828 if (UseHeavyMonitors) { 829 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 830 monitor, /*check_for_exceptions=*/true); 831 } else { 832 // template code: 833 // 834 // markOop displaced_header = obj->mark().set_unlocked(); 835 // monitor->lock()->set_displaced_header(displaced_header); 836 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 837 // // We stored the monitor address into the object's mark word. 838 // } else if (THREAD->is_lock_owned((address)displaced_header)) 839 // // Simple recursive case. 840 // monitor->lock()->set_displaced_header(NULL); 841 // } else { 842 // // Slow path. 843 // InterpreterRuntime::monitorenter(THREAD, monitor); 844 // } 845 846 const Register displaced_header = R7_ARG5; 847 const Register object_mark_addr = R8_ARG6; 848 const Register current_header = R9_ARG7; 849 const Register tmp = R10_ARG8; 850 851 Label done; 852 Label cas_failed, slow_case; 853 854 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp); 855 856 // markOop displaced_header = obj->mark().set_unlocked(); 857 858 // Load markOop from object into displaced_header. 859 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object); 860 861 if (UseBiasedLocking) { 862 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case); 863 } 864 865 // Set displaced_header to be (markOop of object | UNLOCK_VALUE). 866 ori(displaced_header, displaced_header, markOopDesc::unlocked_value); 867 868 // monitor->lock()->set_displaced_header(displaced_header); 869 870 // Initialize the box (Must happen before we update the object mark!). 871 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 872 BasicLock::displaced_header_offset_in_bytes(), monitor); 873 874 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 875 876 // Store stack address of the BasicObjectLock (this is monitor) into object. 877 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 878 879 // Must fence, otherwise, preceding store(s) may float below cmpxchg. 880 // CmpxchgX sets CCR0 to cmpX(current, displaced). 881 cmpxchgd(/*flag=*/CCR0, 882 /*current_value=*/current_header, 883 /*compare_value=*/displaced_header, /*exchange_value=*/monitor, 884 /*where=*/object_mark_addr, 885 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq, 886 MacroAssembler::cmpxchgx_hint_acquire_lock(), 887 noreg, 888 &cas_failed, 889 /*check without membar and ldarx first*/true); 890 891 // If the compare-and-exchange succeeded, then we found an unlocked 892 // object and we have now locked it. 893 b(done); 894 bind(cas_failed); 895 896 // } else if (THREAD->is_lock_owned((address)displaced_header)) 897 // // Simple recursive case. 898 // monitor->lock()->set_displaced_header(NULL); 899 900 // We did not see an unlocked object so try the fast recursive case. 901 902 // Check if owner is self by comparing the value in the markOop of object 903 // (current_header) with the stack pointer. 904 sub(current_header, current_header, R1_SP); 905 906 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 907 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place); 908 909 and_(R0/*==0?*/, current_header, tmp); 910 // If condition is true we are done and hence we can store 0 in the displaced 911 // header indicating it is a recursive lock. 912 bne(CCR0, slow_case); 913 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() + 914 BasicLock::displaced_header_offset_in_bytes(), monitor); 915 b(done); 916 917 // } else { 918 // // Slow path. 919 // InterpreterRuntime::monitorenter(THREAD, monitor); 920 921 // None of the above fast optimizations worked so we have to get into the 922 // slow case of monitor enter. 923 bind(slow_case); 924 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 925 monitor, /*check_for_exceptions=*/true); 926 // } 927 align(32, 12); 928 bind(done); 929 } 930} 931 932// Unlocks an object. Used in monitorexit bytecode and remove_activation. 933// 934// Registers alive 935// monitor - Address of the BasicObjectLock to be used for locking, 936// which must be initialized with the object to lock. 937// 938// Throw IllegalMonitorException if object is not locked by current thread. 939void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) { 940 if (UseHeavyMonitors) { 941 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 942 monitor, check_for_exceptions); 943 } else { 944 945 // template code: 946 // 947 // if ((displaced_header = monitor->displaced_header()) == NULL) { 948 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 949 // monitor->set_obj(NULL); 950 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 951 // // We swapped the unlocked mark in displaced_header into the object's mark word. 952 // monitor->set_obj(NULL); 953 // } else { 954 // // Slow path. 955 // InterpreterRuntime::monitorexit(THREAD, monitor); 956 // } 957 958 const Register object = R7_ARG5; 959 const Register displaced_header = R8_ARG6; 960 const Register object_mark_addr = R9_ARG7; 961 const Register current_header = R10_ARG8; 962 963 Label free_slot; 964 Label slow_case; 965 966 assert_different_registers(object, displaced_header, object_mark_addr, current_header); 967 968 if (UseBiasedLocking) { 969 // The object address from the monitor is in object. 970 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); 971 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0"); 972 biased_locking_exit(CCR0, object, displaced_header, free_slot); 973 } 974 975 // Test first if we are in the fast recursive case. 976 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 977 BasicLock::displaced_header_offset_in_bytes(), monitor); 978 979 // If the displaced header is zero, we have a recursive unlock. 980 cmpdi(CCR0, displaced_header, 0); 981 beq(CCR0, free_slot); // recursive unlock 982 983 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 984 // // We swapped the unlocked mark in displaced_header into the object's mark word. 985 // monitor->set_obj(NULL); 986 987 // If we still have a lightweight lock, unlock the object and be done. 988 989 // The object address from the monitor is in object. 990 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); } 991 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 992 993 // We have the displaced header in displaced_header. If the lock is still 994 // lightweight, it will contain the monitor address and we'll store the 995 // displaced header back into the object's mark word. 996 // CmpxchgX sets CCR0 to cmpX(current, monitor). 997 cmpxchgd(/*flag=*/CCR0, 998 /*current_value=*/current_header, 999 /*compare_value=*/monitor, /*exchange_value=*/displaced_header, 1000 /*where=*/object_mark_addr, 1001 MacroAssembler::MemBarRel, 1002 MacroAssembler::cmpxchgx_hint_release_lock(), 1003 noreg, 1004 &slow_case); 1005 b(free_slot); 1006 1007 // } else { 1008 // // Slow path. 1009 // InterpreterRuntime::monitorexit(THREAD, monitor); 1010 1011 // The lock has been converted into a heavy lock and hence 1012 // we need to get into the slow case. 1013 bind(slow_case); 1014 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1015 monitor, check_for_exceptions); 1016 // } 1017 1018 Label done; 1019 b(done); // Monitor register may be overwritten! Runtime has already freed the slot. 1020 1021 // Exchange worked, do monitor->set_obj(NULL); 1022 align(32, 12); 1023 bind(free_slot); 1024 li(R0, 0); 1025 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor); 1026 bind(done); 1027 } 1028} 1029 1030// Load compiled (i2c) or interpreter entry when calling from interpreted and 1031// do the call. Centralized so that all interpreter calls will do the same actions. 1032// If jvmti single stepping is on for a thread we must not call compiled code. 1033// 1034// Input: 1035// - Rtarget_method: method to call 1036// - Rret_addr: return address 1037// - 2 scratch regs 1038// 1039void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, 1040 Register Rscratch1, Register Rscratch2) { 1041 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr); 1042 // Assume we want to go compiled if available. 1043 const Register Rtarget_addr = Rscratch1; 1044 const Register Rinterp_only = Rscratch2; 1045 1046 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method); 1047 1048 if (JvmtiExport::can_post_interpreter_events()) { 1049 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 1050 1051 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 1052 // compiled code in threads for which the event is enabled. Check here for 1053 // interp_only_mode if these events CAN be enabled. 1054 Label done; 1055 verify_thread(); 1056 cmpwi(CCR0, Rinterp_only, 0); 1057 beq(CCR0, done); 1058 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method); 1059 align(32, 12); 1060 bind(done); 1061 } 1062 1063#ifdef ASSERT 1064 { 1065 Label Lok; 1066 cmpdi(CCR0, Rtarget_addr, 0); 1067 bne(CCR0, Lok); 1068 stop("null entry point"); 1069 bind(Lok); 1070 } 1071#endif // ASSERT 1072 1073 mr(R21_sender_SP, R1_SP); 1074 1075 // Calc a precise SP for the call. The SP value we calculated in 1076 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space 1077 // if esp is not max. Also, the i2c adapter extends the stack space without restoring 1078 // our pre-calced value, so repeating calls via i2c would result in stack overflow. 1079 // Since esp already points to an empty slot, we just have to sub 1 additional slot 1080 // to meet the abi scratch requirements. 1081 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in 1082 // the return entry of the interpreter. 1083 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size); 1084 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address 1085 resize_frame_absolute(Rscratch2, Rscratch2, R0); 1086 1087 mr_if_needed(R19_method, Rtarget_method); 1088 mtctr(Rtarget_addr); 1089 mtlr(Rret_addr); 1090 1091 save_interpreter_state(Rscratch2); 1092#ifdef ASSERT 1093 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp 1094 cmpd(CCR0, R21_sender_SP, Rscratch1); 1095 asm_assert_eq("top_frame_sp incorrect", 0x951); 1096#endif 1097 1098 bctr(); 1099} 1100 1101// Set the method data pointer for the current bcp. 1102void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1103 assert(ProfileInterpreter, "must be profiling interpreter"); 1104 Label get_continue; 1105 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method); 1106 test_method_data_pointer(get_continue); 1107 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp); 1108 1109 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1110 add(R28_mdx, R28_mdx, R3_RET); 1111 bind(get_continue); 1112} 1113 1114// Test ImethodDataPtr. If it is null, continue at the specified label. 1115void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1116 assert(ProfileInterpreter, "must be profiling interpreter"); 1117 cmpdi(CCR0, R28_mdx, 0); 1118 beq(CCR0, zero_continue); 1119} 1120 1121void InterpreterMacroAssembler::verify_method_data_pointer() { 1122 assert(ProfileInterpreter, "must be profiling interpreter"); 1123#ifdef ASSERT 1124 Label verify_continue; 1125 test_method_data_pointer(verify_continue); 1126 1127 // If the mdp is valid, it will point to a DataLayout header which is 1128 // consistent with the bcp. The converse is highly probable also. 1129 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx); 1130 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method); 1131 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset())); 1132 add(R11_scratch1, R12_scratch2, R12_scratch2); 1133 cmpd(CCR0, R11_scratch1, R14_bcp); 1134 beq(CCR0, verify_continue); 1135 1136 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx); 1137 1138 bind(verify_continue); 1139#endif 1140} 1141 1142void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1143 Register method_counters, 1144 Register Rscratch, 1145 Label &profile_continue) { 1146 assert(ProfileInterpreter, "must be profiling interpreter"); 1147 // Control will flow to "profile_continue" if the counter is less than the 1148 // limit or if we call profile_method(). 1149 Label done; 1150 1151 // If no method data exists, and the counter is high enough, make one. 1152 lwz(Rscratch, in_bytes(MethodCounters::interpreter_profile_limit_offset()), method_counters); 1153 1154 cmpdi(CCR0, R28_mdx, 0); 1155 // Test to see if we should create a method data oop. 1156 cmpd(CCR1, Rscratch, invocation_count); 1157 bne(CCR0, done); 1158 bge(CCR1, profile_continue); 1159 1160 // Build it now. 1161 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1162 set_method_data_pointer_for_bcp(); 1163 b(profile_continue); 1164 1165 align(32, 12); 1166 bind(done); 1167} 1168 1169void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register method_counters, 1170 Register target_bcp, Register disp, Register Rtmp) { 1171 assert_different_registers(backedge_count, target_bcp, disp, Rtmp, R4_ARG2); 1172 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 1173 1174 Label did_not_overflow; 1175 Label overflow_with_error; 1176 1177 lwz(Rtmp, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()), method_counters); 1178 cmpw(CCR0, backedge_count, Rtmp); 1179 1180 blt(CCR0, did_not_overflow); 1181 1182 // When ProfileInterpreter is on, the backedge_count comes from the 1183 // methodDataOop, which value does not get reset on the call to 1184 // frequency_counter_overflow(). To avoid excessive calls to the overflow 1185 // routine while the method is being compiled, add a second test to make sure 1186 // the overflow function is called only once every overflow_frequency. 1187 if (ProfileInterpreter) { 1188 const int overflow_frequency = 1024; 1189 andi_(Rtmp, backedge_count, overflow_frequency-1); 1190 bne(CCR0, did_not_overflow); 1191 } 1192 1193 // Overflow in loop, pass branch bytecode. 1194 subf(R4_ARG2, disp, target_bcp); // Compute branch bytecode (previous bcp). 1195 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); 1196 1197 // Was an OSR adapter generated? 1198 cmpdi(CCR0, R3_RET, 0); 1199 beq(CCR0, overflow_with_error); 1200 1201 // Has the nmethod been invalidated already? 1202 lbz(Rtmp, nmethod::state_offset(), R3_RET); 1203 cmpwi(CCR0, Rtmp, nmethod::in_use); 1204 bne(CCR0, overflow_with_error); 1205 1206 // Migrate the interpreter frame off of the stack. 1207 // We can use all registers because we will not return to interpreter from this point. 1208 1209 // Save nmethod. 1210 const Register osr_nmethod = R31; 1211 mr(osr_nmethod, R3_RET); 1212 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1); 1213 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread); 1214 reset_last_Java_frame(); 1215 // OSR buffer is in ARG1 1216 1217 // Remove the interpreter frame. 1218 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1219 1220 // Jump to the osr code. 1221 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod); 1222 mtlr(R0); 1223 mtctr(R11_scratch1); 1224 bctr(); 1225 1226 align(32, 12); 1227 bind(overflow_with_error); 1228 bind(did_not_overflow); 1229} 1230 1231// Store a value at some constant offset from the method data pointer. 1232void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1233 assert(ProfileInterpreter, "must be profiling interpreter"); 1234 1235 std(value, constant, R28_mdx); 1236} 1237 1238// Increment the value at some constant offset from the method data pointer. 1239void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1240 Register counter_addr, 1241 Register Rbumped_count, 1242 bool decrement) { 1243 // Locate the counter at a fixed offset from the mdp: 1244 addi(counter_addr, R28_mdx, constant); 1245 increment_mdp_data_at(counter_addr, Rbumped_count, decrement); 1246} 1247 1248// Increment the value at some non-fixed (reg + constant) offset from 1249// the method data pointer. 1250void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1251 int constant, 1252 Register scratch, 1253 Register Rbumped_count, 1254 bool decrement) { 1255 // Add the constant to reg to get the offset. 1256 add(scratch, R28_mdx, reg); 1257 // Then calculate the counter address. 1258 addi(scratch, scratch, constant); 1259 increment_mdp_data_at(scratch, Rbumped_count, decrement); 1260} 1261 1262void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr, 1263 Register Rbumped_count, 1264 bool decrement) { 1265 assert(ProfileInterpreter, "must be profiling interpreter"); 1266 1267 // Load the counter. 1268 ld(Rbumped_count, 0, counter_addr); 1269 1270 if (decrement) { 1271 // Decrement the register. Set condition codes. 1272 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment); 1273 // Store the decremented counter, if it is still negative. 1274 std(Rbumped_count, 0, counter_addr); 1275 // Note: add/sub overflow check are not ported, since 64 bit 1276 // calculation should never overflow. 1277 } else { 1278 // Increment the register. Set carry flag. 1279 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment); 1280 // Store the incremented counter. 1281 std(Rbumped_count, 0, counter_addr); 1282 } 1283} 1284 1285// Set a flag value at the current method data pointer position. 1286void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1287 Register scratch) { 1288 assert(ProfileInterpreter, "must be profiling interpreter"); 1289 // Load the data header. 1290 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1291 // Set the flag. 1292 ori(scratch, scratch, flag_constant); 1293 // Store the modified header. 1294 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1295} 1296 1297// Test the location at some offset from the method data pointer. 1298// If it is not equal to value, branch to the not_equal_continue Label. 1299void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1300 Register value, 1301 Label& not_equal_continue, 1302 Register test_out) { 1303 assert(ProfileInterpreter, "must be profiling interpreter"); 1304 1305 ld(test_out, offset, R28_mdx); 1306 cmpd(CCR0, value, test_out); 1307 bne(CCR0, not_equal_continue); 1308} 1309 1310// Update the method data pointer by the displacement located at some fixed 1311// offset from the method data pointer. 1312void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1313 Register scratch) { 1314 assert(ProfileInterpreter, "must be profiling interpreter"); 1315 1316 ld(scratch, offset_of_disp, R28_mdx); 1317 add(R28_mdx, scratch, R28_mdx); 1318} 1319 1320// Update the method data pointer by the displacement located at the 1321// offset (reg + offset_of_disp). 1322void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1323 int offset_of_disp, 1324 Register scratch) { 1325 assert(ProfileInterpreter, "must be profiling interpreter"); 1326 1327 add(scratch, reg, R28_mdx); 1328 ld(scratch, offset_of_disp, scratch); 1329 add(R28_mdx, scratch, R28_mdx); 1330} 1331 1332// Update the method data pointer by a simple constant displacement. 1333void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1334 assert(ProfileInterpreter, "must be profiling interpreter"); 1335 addi(R28_mdx, R28_mdx, constant); 1336} 1337 1338// Update the method data pointer for a _ret bytecode whose target 1339// was not among our cached targets. 1340void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1341 Register return_bci) { 1342 assert(ProfileInterpreter, "must be profiling interpreter"); 1343 1344 push(state); 1345 assert(return_bci->is_nonvolatile(), "need to protect return_bci"); 1346 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1347 pop(state); 1348} 1349 1350// Increments the backedge counter. 1351// Returns backedge counter + invocation counter in Rdst. 1352void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst, 1353 const Register Rtmp1, Register Rscratch) { 1354 assert(UseCompiler, "incrementing must be useful"); 1355 assert_different_registers(Rdst, Rtmp1); 1356 const Register invocation_counter = Rtmp1; 1357 const Register counter = Rdst; 1358 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size."); 1359 1360 // Load backedge counter. 1361 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1362 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1363 // Load invocation counter. 1364 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) + 1365 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1366 1367 // Add the delta to the backedge counter. 1368 addi(counter, counter, InvocationCounter::count_increment); 1369 1370 // Mask the invocation counter. 1371 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value); 1372 1373 // Store new counter value. 1374 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1375 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1376 // Return invocation counter + backedge counter. 1377 add(counter, counter, invocation_counter); 1378} 1379 1380// Count a taken branch in the bytecodes. 1381void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1382 if (ProfileInterpreter) { 1383 Label profile_continue; 1384 1385 // If no method data exists, go to profile_continue. 1386 test_method_data_pointer(profile_continue); 1387 1388 // We are taking a branch. Increment the taken count. 1389 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count); 1390 1391 // The method data pointer needs to be updated to reflect the new target. 1392 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1393 bind (profile_continue); 1394 } 1395} 1396 1397// Count a not-taken branch in the bytecodes. 1398void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) { 1399 if (ProfileInterpreter) { 1400 Label profile_continue; 1401 1402 // If no method data exists, go to profile_continue. 1403 test_method_data_pointer(profile_continue); 1404 1405 // We are taking a branch. Increment the not taken count. 1406 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2); 1407 1408 // The method data pointer needs to be updated to correspond to the 1409 // next bytecode. 1410 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1411 bind (profile_continue); 1412 } 1413} 1414 1415// Count a non-virtual call in the bytecodes. 1416void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) { 1417 if (ProfileInterpreter) { 1418 Label profile_continue; 1419 1420 // If no method data exists, go to profile_continue. 1421 test_method_data_pointer(profile_continue); 1422 1423 // We are making a call. Increment the count. 1424 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1425 1426 // The method data pointer needs to be updated to reflect the new target. 1427 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1428 bind (profile_continue); 1429 } 1430} 1431 1432// Count a final call in the bytecodes. 1433void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) { 1434 if (ProfileInterpreter) { 1435 Label profile_continue; 1436 1437 // If no method data exists, go to profile_continue. 1438 test_method_data_pointer(profile_continue); 1439 1440 // We are making a call. Increment the count. 1441 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1442 1443 // The method data pointer needs to be updated to reflect the new target. 1444 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1445 bind (profile_continue); 1446 } 1447} 1448 1449// Count a virtual call in the bytecodes. 1450void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver, 1451 Register Rscratch1, 1452 Register Rscratch2, 1453 bool receiver_can_be_null) { 1454 if (!ProfileInterpreter) { return; } 1455 Label profile_continue; 1456 1457 // If no method data exists, go to profile_continue. 1458 test_method_data_pointer(profile_continue); 1459 1460 Label skip_receiver_profile; 1461 if (receiver_can_be_null) { 1462 Label not_null; 1463 cmpdi(CCR0, Rreceiver, 0); 1464 bne(CCR0, not_null); 1465 // We are making a call. Increment the count for null receiver. 1466 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2); 1467 b(skip_receiver_profile); 1468 bind(not_null); 1469 } 1470 1471 // Record the receiver type. 1472 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true); 1473 bind(skip_receiver_profile); 1474 1475 // The method data pointer needs to be updated to reflect the new target. 1476 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1477 bind (profile_continue); 1478} 1479 1480void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) { 1481 if (ProfileInterpreter) { 1482 Label profile_continue; 1483 1484 // If no method data exists, go to profile_continue. 1485 test_method_data_pointer(profile_continue); 1486 1487 int mdp_delta = in_bytes(BitData::bit_data_size()); 1488 if (TypeProfileCasts) { 1489 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1490 1491 // Record the object type. 1492 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false); 1493 } 1494 1495 // The method data pointer needs to be updated. 1496 update_mdp_by_constant(mdp_delta); 1497 1498 bind (profile_continue); 1499 } 1500} 1501 1502void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) { 1503 if (ProfileInterpreter && TypeProfileCasts) { 1504 Label profile_continue; 1505 1506 // If no method data exists, go to profile_continue. 1507 test_method_data_pointer(profile_continue); 1508 1509 int count_offset = in_bytes(CounterData::count_offset()); 1510 // Back up the address, since we have already bumped the mdp. 1511 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1512 1513 // *Decrement* the counter. We expect to see zero or small negatives. 1514 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true); 1515 1516 bind (profile_continue); 1517 } 1518} 1519 1520// Count a ret in the bytecodes. 1521void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, 1522 Register scratch1, Register scratch2) { 1523 if (ProfileInterpreter) { 1524 Label profile_continue; 1525 uint row; 1526 1527 // If no method data exists, go to profile_continue. 1528 test_method_data_pointer(profile_continue); 1529 1530 // Update the total ret count. 1531 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 ); 1532 1533 for (row = 0; row < RetData::row_limit(); row++) { 1534 Label next_test; 1535 1536 // See if return_bci is equal to bci[n]: 1537 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1); 1538 1539 // return_bci is equal to bci[n]. Increment the count. 1540 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2); 1541 1542 // The method data pointer needs to be updated to reflect the new target. 1543 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1); 1544 b(profile_continue); 1545 bind(next_test); 1546 } 1547 1548 update_mdp_for_ret(state, return_bci); 1549 1550 bind (profile_continue); 1551 } 1552} 1553 1554// Count the default case of a switch construct. 1555void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) { 1556 if (ProfileInterpreter) { 1557 Label profile_continue; 1558 1559 // If no method data exists, go to profile_continue. 1560 test_method_data_pointer(profile_continue); 1561 1562 // Update the default case count 1563 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1564 scratch1, scratch2); 1565 1566 // The method data pointer needs to be updated. 1567 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()), 1568 scratch1); 1569 1570 bind (profile_continue); 1571 } 1572} 1573 1574// Count the index'th case of a switch construct. 1575void InterpreterMacroAssembler::profile_switch_case(Register index, 1576 Register scratch1, 1577 Register scratch2, 1578 Register scratch3) { 1579 if (ProfileInterpreter) { 1580 assert_different_registers(index, scratch1, scratch2, scratch3); 1581 Label profile_continue; 1582 1583 // If no method data exists, go to profile_continue. 1584 test_method_data_pointer(profile_continue); 1585 1586 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes(). 1587 li(scratch3, in_bytes(MultiBranchData::case_array_offset())); 1588 1589 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works"); 1590 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1591 add(scratch1, scratch1, scratch3); 1592 1593 // Update the case count. 1594 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3); 1595 1596 // The method data pointer needs to be updated. 1597 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2); 1598 1599 bind (profile_continue); 1600 } 1601} 1602 1603void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) { 1604 if (ProfileInterpreter) { 1605 assert_different_registers(Rscratch1, Rscratch2); 1606 Label profile_continue; 1607 1608 // If no method data exists, go to profile_continue. 1609 test_method_data_pointer(profile_continue); 1610 1611 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1); 1612 1613 // The method data pointer needs to be updated. 1614 int mdp_delta = in_bytes(BitData::bit_data_size()); 1615 if (TypeProfileCasts) { 1616 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1617 } 1618 update_mdp_by_constant(mdp_delta); 1619 1620 bind (profile_continue); 1621 } 1622} 1623 1624void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver, 1625 Register Rscratch1, Register Rscratch2, 1626 bool is_virtual_call) { 1627 assert(ProfileInterpreter, "must be profiling"); 1628 assert_different_registers(Rreceiver, Rscratch1, Rscratch2); 1629 1630 Label done; 1631 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call); 1632 bind (done); 1633} 1634 1635void InterpreterMacroAssembler::record_klass_in_profile_helper( 1636 Register receiver, Register scratch1, Register scratch2, 1637 int start_row, Label& done, bool is_virtual_call) { 1638 if (TypeProfileWidth == 0) { 1639 if (is_virtual_call) { 1640 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1641 } 1642 return; 1643 } 1644 1645 int last_row = VirtualCallData::row_limit() - 1; 1646 assert(start_row <= last_row, "must be work left to do"); 1647 // Test this row for both the receiver and for null. 1648 // Take any of three different outcomes: 1649 // 1. found receiver => increment count and goto done 1650 // 2. found null => keep looking for case 1, maybe allocate this cell 1651 // 3. found something else => keep looking for cases 1 and 2 1652 // Case 3 is handled by a recursive call. 1653 for (int row = start_row; row <= last_row; row++) { 1654 Label next_test; 1655 bool test_for_null_also = (row == start_row); 1656 1657 // See if the receiver is receiver[n]. 1658 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1659 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1); 1660 // delayed()->tst(scratch); 1661 1662 // The receiver is receiver[n]. Increment count[n]. 1663 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1664 increment_mdp_data_at(count_offset, scratch1, scratch2); 1665 b(done); 1666 bind(next_test); 1667 1668 if (test_for_null_also) { 1669 Label found_null; 1670 // Failed the equality check on receiver[n]... Test for null. 1671 if (start_row == last_row) { 1672 // The only thing left to do is handle the null case. 1673 if (is_virtual_call) { 1674 // Scratch1 contains test_out from test_mdp_data_at. 1675 cmpdi(CCR0, scratch1, 0); 1676 beq(CCR0, found_null); 1677 // Receiver did not match any saved receiver and there is no empty row for it. 1678 // Increment total counter to indicate polymorphic case. 1679 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1680 b(done); 1681 bind(found_null); 1682 } else { 1683 cmpdi(CCR0, scratch1, 0); 1684 bne(CCR0, done); 1685 } 1686 break; 1687 } 1688 // Since null is rare, make it be the branch-taken case. 1689 cmpdi(CCR0, scratch1, 0); 1690 beq(CCR0, found_null); 1691 1692 // Put all the "Case 3" tests here. 1693 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call); 1694 1695 // Found a null. Keep searching for a matching receiver, 1696 // but remember that this is an empty (unused) slot. 1697 bind(found_null); 1698 } 1699 } 1700 1701 // In the fall-through case, we found no matching receiver, but we 1702 // observed the receiver[start_row] is NULL. 1703 1704 // Fill in the receiver field and increment the count. 1705 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1706 set_mdp_data_at(recvr_offset, receiver); 1707 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1708 li(scratch1, DataLayout::counter_increment); 1709 set_mdp_data_at(count_offset, scratch1); 1710 if (start_row > 0) { 1711 b(done); 1712 } 1713} 1714 1715// Argument and return type profilig. 1716// kills: tmp, tmp2, R0, CR0, CR1 1717void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base, 1718 RegisterOrConstant mdo_addr_offs, 1719 Register tmp, Register tmp2) { 1720 Label do_nothing, do_update; 1721 1722 // tmp2 = obj is allowed 1723 assert_different_registers(obj, mdo_addr_base, tmp, R0); 1724 assert_different_registers(tmp2, mdo_addr_base, tmp, R0); 1725 const Register klass = tmp2; 1726 1727 verify_oop(obj); 1728 1729 ld(tmp, mdo_addr_offs, mdo_addr_base); 1730 1731 // Set null_seen if obj is 0. 1732 cmpdi(CCR0, obj, 0); 1733 ori(R0, tmp, TypeEntries::null_seen); 1734 beq(CCR0, do_update); 1735 1736 load_klass(klass, obj); 1737 1738 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask)); 1739 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask); 1740 cmpd(CCR1, R0, klass); 1741 // Klass seen before, nothing to do (regardless of unknown bit). 1742 //beq(CCR1, do_nothing); 1743 1744 andi_(R0, klass, TypeEntries::type_unknown); 1745 // Already unknown. Nothing to do anymore. 1746 //bne(CCR0, do_nothing); 1747 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne 1748 beq(CCR0, do_nothing); 1749 1750 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask)); 1751 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1752 beq(CCR0, do_update); // First time here. Set profile type. 1753 1754 // Different than before. Cannot keep accurate profile. 1755 ori(R0, tmp, TypeEntries::type_unknown); 1756 1757 bind(do_update); 1758 // update profile 1759 std(R0, mdo_addr_offs, mdo_addr_base); 1760 1761 align(32, 12); 1762 bind(do_nothing); 1763} 1764 1765void InterpreterMacroAssembler::profile_arguments_type(Register callee, 1766 Register tmp1, Register tmp2, 1767 bool is_virtual) { 1768 if (!ProfileInterpreter) { 1769 return; 1770 } 1771 1772 assert_different_registers(callee, tmp1, tmp2, R28_mdx); 1773 1774 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1775 Label profile_continue; 1776 1777 test_method_data_pointer(profile_continue); 1778 1779 int off_to_start = is_virtual ? 1780 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1781 1782 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx); 1783 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1784 bne(CCR0, profile_continue); 1785 1786 if (MethodData::profile_arguments()) { 1787 Label done; 1788 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1789 add(R28_mdx, off_to_args, R28_mdx); 1790 1791 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1792 if (i > 0 || MethodData::profile_return()) { 1793 // If return value type is profiled we may have no argument to profile. 1794 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1795 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count()); 1796 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count()); 1797 blt(CCR0, done); 1798 } 1799 ld(tmp1, in_bytes(Method::const_offset()), callee); 1800 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1); 1801 // Stack offset o (zero based) from the start of the argument 1802 // list, for n arguments translates into offset n - o - 1 from 1803 // the end of the argument list. But there's an extra slot at 1804 // the top of the stack. So the offset is n - o from Lesp. 1805 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx); 1806 subf(tmp1, tmp2, tmp1); 1807 1808 sldi(tmp1, tmp1, Interpreter::logStackElementSize); 1809 ldx(tmp1, tmp1, R15_esp); 1810 1811 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1); 1812 1813 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1814 addi(R28_mdx, R28_mdx, to_add); 1815 off_to_args += to_add; 1816 } 1817 1818 if (MethodData::profile_return()) { 1819 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1820 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1821 } 1822 1823 bind(done); 1824 1825 if (MethodData::profile_return()) { 1826 // We're right after the type profile for the last 1827 // argument. tmp1 is the number of cells left in the 1828 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1829 // if there's a return to profile. 1830 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), 1831 "can't move past ret type"); 1832 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size)); 1833 add(R28_mdx, tmp1, R28_mdx); 1834 } 1835 } else { 1836 assert(MethodData::profile_return(), "either profile call args or call ret"); 1837 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 1838 } 1839 1840 // Mdp points right after the end of the 1841 // CallTypeData/VirtualCallTypeData, right after the cells for the 1842 // return value type if there's one. 1843 align(32, 12); 1844 bind(profile_continue); 1845 } 1846} 1847 1848void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 1849 assert_different_registers(ret, tmp1, tmp2); 1850 if (ProfileInterpreter && MethodData::profile_return()) { 1851 Label profile_continue; 1852 1853 test_method_data_pointer(profile_continue); 1854 1855 if (MethodData::profile_return_jsr292_only()) { 1856 // If we don't profile all invoke bytecodes we must make sure 1857 // it's a bytecode we indeed profile. We can't go back to the 1858 // begining of the ProfileData we intend to update to check its 1859 // type because we're right after it and we don't known its 1860 // length. 1861 lbz(tmp1, 0, R14_bcp); 1862 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method); 1863 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic); 1864 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle); 1865 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1866 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm); 1867 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1868 bne(CCR0, profile_continue); 1869 } 1870 1871 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2); 1872 1873 align(32, 12); 1874 bind(profile_continue); 1875 } 1876} 1877 1878void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, 1879 Register tmp3, Register tmp4) { 1880 if (ProfileInterpreter && MethodData::profile_parameters()) { 1881 Label profile_continue, done; 1882 1883 test_method_data_pointer(profile_continue); 1884 1885 // Load the offset of the area within the MDO used for 1886 // parameters. If it's negative we're not profiling any parameters. 1887 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx); 1888 cmpwi(CCR0, tmp1, 0); 1889 blt(CCR0, profile_continue); 1890 1891 // Compute a pointer to the area for parameters from the offset 1892 // and move the pointer to the slot for the last 1893 // parameters. Collect profiling from last parameter down. 1894 // mdo start + parameters offset + array length - 1 1895 1896 // Pointer to the parameter area in the MDO. 1897 const Register mdp = tmp1; 1898 add(mdp, tmp1, R28_mdx); 1899 1900 // Offset of the current profile entry to update. 1901 const Register entry_offset = tmp2; 1902 // entry_offset = array len in number of cells 1903 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp); 1904 1905 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1906 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 1907 1908 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 1909 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size)); 1910 // entry_offset in bytes 1911 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1912 1913 Label loop; 1914 align(32, 12); 1915 bind(loop); 1916 1917 // Load offset on the stack from the slot for this parameter. 1918 ld(tmp3, entry_offset, mdp); 1919 sldi(tmp3, tmp3, Interpreter::logStackElementSize); 1920 neg(tmp3, tmp3); 1921 // Read the parameter from the local area. 1922 ldx(tmp3, tmp3, R18_locals); 1923 1924 // Make entry_offset now point to the type field for this parameter. 1925 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1926 assert(type_base > off_base, "unexpected"); 1927 addi(entry_offset, entry_offset, type_base - off_base); 1928 1929 // Profile the parameter. 1930 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3); 1931 1932 // Go to next parameter. 1933 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base); 1934 cmpdi(CCR0, entry_offset, off_base + delta); 1935 addi(entry_offset, entry_offset, -delta); 1936 bge(CCR0, loop); 1937 1938 align(32, 12); 1939 bind(profile_continue); 1940 } 1941} 1942 1943// Add a InterpMonitorElem to stack (see frame_sparc.hpp). 1944void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) { 1945 1946 // Very-local scratch registers. 1947 const Register esp = Rtemp1; 1948 const Register slot = Rtemp2; 1949 1950 // Extracted monitor_size. 1951 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes(); 1952 assert(Assembler::is_aligned((unsigned int)monitor_size, 1953 (unsigned int)frame::alignment_in_bytes), 1954 "size of a monitor must respect alignment of SP"); 1955 1956 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor 1957 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp 1958 1959 // Shuffle expression stack down. Recall that stack_base points 1960 // just above the new expression stack bottom. Old_tos and new_tos 1961 // are used to scan thru the old and new expression stacks. 1962 if (!stack_is_empty) { 1963 Label copy_slot, copy_slot_finished; 1964 const Register n_slots = slot; 1965 1966 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack). 1967 subf(n_slots, esp, R26_monitor); 1968 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy. 1969 assert(LogBytesPerWord == 3, "conflicts assembler instructions"); 1970 beq(CCR0, copy_slot_finished); // Nothing to copy. 1971 1972 mtctr(n_slots); 1973 1974 // loop 1975 bind(copy_slot); 1976 ld(slot, 0, esp); // Move expression stack down. 1977 std(slot, -monitor_size, esp); // distance = monitor_size 1978 addi(esp, esp, BytesPerWord); 1979 bdnz(copy_slot); 1980 1981 bind(copy_slot_finished); 1982 } 1983 1984 addi(R15_esp, R15_esp, -monitor_size); 1985 addi(R26_monitor, R26_monitor, -monitor_size); 1986 1987 // Restart interpreter 1988} 1989 1990// ============================================================================ 1991// Java locals access 1992 1993// Load a local variable at index in Rindex into register Rdst_value. 1994// Also puts address of local into Rdst_address as a service. 1995// Kills: 1996// - Rdst_value 1997// - Rdst_address 1998void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) { 1999 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2000 subf(Rdst_address, Rdst_address, R18_locals); 2001 lwz(Rdst_value, 0, Rdst_address); 2002} 2003 2004// Load a local variable at index in Rindex into register Rdst_value. 2005// Also puts address of local into Rdst_address as a service. 2006// Kills: 2007// - Rdst_value 2008// - Rdst_address 2009void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) { 2010 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2011 subf(Rdst_address, Rdst_address, R18_locals); 2012 ld(Rdst_value, -8, Rdst_address); 2013} 2014 2015// Load a local variable at index in Rindex into register Rdst_value. 2016// Also puts address of local into Rdst_address as a service. 2017// Input: 2018// - Rindex: slot nr of local variable 2019// Kills: 2020// - Rdst_value 2021// - Rdst_address 2022void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, 2023 Register Rdst_address, 2024 Register Rindex) { 2025 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2026 subf(Rdst_address, Rdst_address, R18_locals); 2027 ld(Rdst_value, 0, Rdst_address); 2028} 2029 2030// Load a local variable at index in Rindex into register Rdst_value. 2031// Also puts address of local into Rdst_address as a service. 2032// Kills: 2033// - Rdst_value 2034// - Rdst_address 2035void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, 2036 Register Rdst_address, 2037 Register Rindex) { 2038 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2039 subf(Rdst_address, Rdst_address, R18_locals); 2040 lfs(Rdst_value, 0, Rdst_address); 2041} 2042 2043// Load a local variable at index in Rindex into register Rdst_value. 2044// Also puts address of local into Rdst_address as a service. 2045// Kills: 2046// - Rdst_value 2047// - Rdst_address 2048void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, 2049 Register Rdst_address, 2050 Register Rindex) { 2051 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2052 subf(Rdst_address, Rdst_address, R18_locals); 2053 lfd(Rdst_value, -8, Rdst_address); 2054} 2055 2056// Store an int value at local variable slot Rindex. 2057// Kills: 2058// - Rindex 2059void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) { 2060 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2061 subf(Rindex, Rindex, R18_locals); 2062 stw(Rvalue, 0, Rindex); 2063} 2064 2065// Store a long value at local variable slot Rindex. 2066// Kills: 2067// - Rindex 2068void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) { 2069 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2070 subf(Rindex, Rindex, R18_locals); 2071 std(Rvalue, -8, Rindex); 2072} 2073 2074// Store an oop value at local variable slot Rindex. 2075// Kills: 2076// - Rindex 2077void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) { 2078 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2079 subf(Rindex, Rindex, R18_locals); 2080 std(Rvalue, 0, Rindex); 2081} 2082 2083// Store an int value at local variable slot Rindex. 2084// Kills: 2085// - Rindex 2086void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) { 2087 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2088 subf(Rindex, Rindex, R18_locals); 2089 stfs(Rvalue, 0, Rindex); 2090} 2091 2092// Store an int value at local variable slot Rindex. 2093// Kills: 2094// - Rindex 2095void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) { 2096 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2097 subf(Rindex, Rindex, R18_locals); 2098 stfd(Rvalue, -8, Rindex); 2099} 2100 2101// Read pending exception from thread and jump to interpreter. 2102// Throw exception entry if one if pending. Fall through otherwise. 2103void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) { 2104 assert_different_registers(Rscratch1, Rscratch2, R3); 2105 Register Rexception = Rscratch1; 2106 Register Rtmp = Rscratch2; 2107 Label Ldone; 2108 // Get pending exception oop. 2109 ld(Rexception, thread_(pending_exception)); 2110 cmpdi(CCR0, Rexception, 0); 2111 beq(CCR0, Ldone); 2112 li(Rtmp, 0); 2113 mr_if_needed(R3, Rexception); 2114 std(Rtmp, thread_(pending_exception)); // Clear exception in thread 2115 if (Interpreter::rethrow_exception_entry() != NULL) { 2116 // Already got entry address. 2117 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry()); 2118 } else { 2119 // Dynamically load entry address. 2120 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true); 2121 ld(Rtmp, simm16_rest, Rtmp); 2122 } 2123 mtctr(Rtmp); 2124 save_interpreter_state(Rtmp); 2125 bctr(); 2126 2127 align(32, 12); 2128 bind(Ldone); 2129} 2130 2131void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) { 2132 save_interpreter_state(R11_scratch1); 2133 2134 MacroAssembler::call_VM(oop_result, entry_point, false); 2135 2136 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2137 2138 check_and_handle_popframe(R11_scratch1); 2139 check_and_handle_earlyret(R11_scratch1); 2140 // Now check exceptions manually. 2141 if (check_exceptions) { 2142 check_and_forward_exception(R11_scratch1, R12_scratch2); 2143 } 2144} 2145 2146void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2147 Register arg_1, bool check_exceptions) { 2148 // ARG1 is reserved for the thread. 2149 mr_if_needed(R4_ARG2, arg_1); 2150 call_VM(oop_result, entry_point, check_exceptions); 2151} 2152 2153void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2154 Register arg_1, Register arg_2, 2155 bool check_exceptions) { 2156 // ARG1 is reserved for the thread. 2157 mr_if_needed(R4_ARG2, arg_1); 2158 assert(arg_2 != R4_ARG2, "smashed argument"); 2159 mr_if_needed(R5_ARG3, arg_2); 2160 call_VM(oop_result, entry_point, check_exceptions); 2161} 2162 2163void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2164 Register arg_1, Register arg_2, Register arg_3, 2165 bool check_exceptions) { 2166 // ARG1 is reserved for the thread. 2167 mr_if_needed(R4_ARG2, arg_1); 2168 assert(arg_2 != R4_ARG2, "smashed argument"); 2169 mr_if_needed(R5_ARG3, arg_2); 2170 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument"); 2171 mr_if_needed(R6_ARG4, arg_3); 2172 call_VM(oop_result, entry_point, check_exceptions); 2173} 2174 2175void InterpreterMacroAssembler::save_interpreter_state(Register scratch) { 2176 ld(scratch, 0, R1_SP); 2177 std(R15_esp, _ijava_state_neg(esp), scratch); 2178 std(R14_bcp, _ijava_state_neg(bcp), scratch); 2179 std(R26_monitor, _ijava_state_neg(monitors), scratch); 2180 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); } 2181 // Other entries should be unchanged. 2182} 2183 2184void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) { 2185 ld(scratch, 0, R1_SP); 2186 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception). 2187 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code. 2188 if (!bcp_and_mdx_only) { 2189 // Following ones are Metadata. 2190 ld(R19_method, _ijava_state_neg(method), scratch); 2191 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch); 2192 // Following ones are stack addresses and don't require reload. 2193 ld(R15_esp, _ijava_state_neg(esp), scratch); 2194 ld(R18_locals, _ijava_state_neg(locals), scratch); 2195 ld(R26_monitor, _ijava_state_neg(monitors), scratch); 2196 } 2197#ifdef ASSERT 2198 { 2199 Label Lok; 2200 subf(R0, R1_SP, scratch); 2201 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size); 2202 bge(CCR0, Lok); 2203 stop("frame too small (restore istate)", 0x5432); 2204 bind(Lok); 2205 } 2206 { 2207 Label Lok; 2208 ld(R0, _ijava_state_neg(ijava_reserved), scratch); 2209 cmpdi(CCR0, R0, 0x5afe); 2210 beq(CCR0, Lok); 2211 stop("frame corrupted (restore istate)", 0x5afe); 2212 bind(Lok); 2213 } 2214#endif 2215} 2216 2217void InterpreterMacroAssembler::get_method_counters(Register method, 2218 Register Rcounters, 2219 Label& skip) { 2220 BLOCK_COMMENT("Load and ev. allocate counter object {"); 2221 Label has_counters; 2222 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2223 cmpdi(CCR0, Rcounters, 0); 2224 bne(CCR0, has_counters); 2225 call_VM(noreg, CAST_FROM_FN_PTR(address, 2226 InterpreterRuntime::build_method_counters), method, false); 2227 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2228 cmpdi(CCR0, Rcounters, 0); 2229 beq(CCR0, skip); // No MethodCounters, OutOfMemory. 2230 BLOCK_COMMENT("} Load and ev. allocate counter object"); 2231 2232 bind(has_counters); 2233} 2234 2235void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, 2236 Register iv_be_count, 2237 Register Rtmp_r0) { 2238 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 2239 Register invocation_count = iv_be_count; 2240 Register backedge_count = Rtmp_r0; 2241 int delta = InvocationCounter::count_increment; 2242 2243 // Load each counter in a register. 2244 // ld(inv_counter, Rtmp); 2245 // ld(be_counter, Rtmp2); 2246 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + 2247 InvocationCounter::counter_offset()); 2248 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + 2249 InvocationCounter::counter_offset()); 2250 2251 BLOCK_COMMENT("Increment profiling counters {"); 2252 2253 // Load the backedge counter. 2254 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int 2255 // Mask the backedge counter. 2256 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value); 2257 2258 // Load the invocation counter. 2259 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int 2260 // Add the delta to the invocation counter and store the result. 2261 addi(invocation_count, invocation_count, delta); 2262 // Store value. 2263 stw(invocation_count, inv_counter_offset, Rcounters); 2264 2265 // Add invocation counter + backedge counter. 2266 add(iv_be_count, backedge_count, invocation_count); 2267 2268 // Note that this macro must leave the backedge_count + invocation_count in 2269 // register iv_be_count! 2270 BLOCK_COMMENT("} Increment profiling counters"); 2271} 2272 2273void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2274 if (state == atos) { MacroAssembler::verify_oop(reg); } 2275} 2276 2277// Local helper function for the verify_oop_or_return_address macro. 2278static bool verify_return_address(Method* m, int bci) { 2279#ifndef PRODUCT 2280 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci; 2281 // Assume it is a valid return address if it is inside m and is preceded by a jsr. 2282 if (!m->contains(pc)) return false; 2283 address jsr_pc; 2284 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2285 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2286 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2287 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2288#endif // PRODUCT 2289 return false; 2290} 2291 2292void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2293 if (VerifyFPU) { 2294 unimplemented("verfiyFPU"); 2295 } 2296} 2297 2298void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2299 if (!VerifyOops) return; 2300 2301 // The VM documentation for the astore[_wide] bytecode allows 2302 // the TOS to be not only an oop but also a return address. 2303 Label test; 2304 Label skip; 2305 // See if it is an address (in the current method): 2306 2307 const int log2_bytecode_size_limit = 16; 2308 srdi_(Rtmp, reg, log2_bytecode_size_limit); 2309 bne(CCR0, test); 2310 2311 address fd = CAST_FROM_FN_PTR(address, verify_return_address); 2312 const int nbytes_save = MacroAssembler::num_volatile_regs * 8; 2313 save_volatile_gprs(R1_SP, -nbytes_save); // except R0 2314 save_LR_CR(Rtmp); // Save in old frame. 2315 push_frame_reg_args(nbytes_save, Rtmp); 2316 2317 load_const_optimized(Rtmp, fd, R0); 2318 mr_if_needed(R4_ARG2, reg); 2319 mr(R3_ARG1, R19_method); 2320 call_c(Rtmp); // call C 2321 2322 pop_frame(); 2323 restore_LR_CR(Rtmp); 2324 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0 2325 b(skip); 2326 2327 // Perform a more elaborate out-of-line call. 2328 // Not an address; verify it: 2329 bind(test); 2330 verify_oop(reg); 2331 bind(skip); 2332} 2333 2334// Inline assembly for: 2335// 2336// if (thread is in interp_only_mode) { 2337// InterpreterRuntime::post_method_entry(); 2338// } 2339// if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) || 2340// *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) { 2341// SharedRuntime::jvmpi_method_entry(method, receiver); 2342// } 2343void InterpreterMacroAssembler::notify_method_entry() { 2344 // JVMTI 2345 // Whenever JVMTI puts a thread in interp_only_mode, method 2346 // entry/exit events are sent for that thread to track stack 2347 // depth. If it is possible to enter interp_only_mode we add 2348 // the code to check if the event should be sent. 2349 if (JvmtiExport::can_post_interpreter_events()) { 2350 Label jvmti_post_done; 2351 2352 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2353 cmpwi(CCR0, R0, 0); 2354 beq(CCR0, jvmti_post_done); 2355 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), 2356 /*check_exceptions=*/true); 2357 2358 bind(jvmti_post_done); 2359 } 2360} 2361 2362// Inline assembly for: 2363// 2364// if (thread is in interp_only_mode) { 2365// // save result 2366// InterpreterRuntime::post_method_exit(); 2367// // restore result 2368// } 2369// if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) { 2370// // save result 2371// SharedRuntime::jvmpi_method_exit(); 2372// // restore result 2373// } 2374// 2375// Native methods have their result stored in d_tmp and l_tmp. 2376// Java methods have their result stored in the expression stack. 2377void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state, 2378 NotifyMethodExitMode mode, bool check_exceptions) { 2379 // JVMTI 2380 // Whenever JVMTI puts a thread in interp_only_mode, method 2381 // entry/exit events are sent for that thread to track stack 2382 // depth. If it is possible to enter interp_only_mode we add 2383 // the code to check if the event should be sent. 2384 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2385 Label jvmti_post_done; 2386 2387 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2388 cmpwi(CCR0, R0, 0); 2389 beq(CCR0, jvmti_post_done); 2390 if (!is_native_method) { push(state); } // Expose tos to GC. 2391 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), 2392 /*check_exceptions=*/check_exceptions); 2393 if (!is_native_method) { pop(state); } 2394 2395 align(32, 12); 2396 bind(jvmti_post_done); 2397 } 2398 2399 // Dtrace support not implemented. 2400} 2401 2402