1 /* 2 * Copyright (c) 2003, 2021, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2012, 2021 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 "gc/shared/barrierSet.hpp" 30 #include "gc/shared/barrierSetAssembler.hpp" 31 #include "interp_masm_ppc.hpp" 32 #include "interpreter/interpreterRuntime.hpp" 33 #include "oops/methodData.hpp" 34 #include "prims/jvmtiExport.hpp" 35 #include "prims/jvmtiThreadState.hpp" 36 #include "runtime/frame.inline.hpp" 37 #include "runtime/safepointMechanism.hpp" 38 #include "runtime/sharedRuntime.hpp" 39 #include "runtime/vm_version.hpp" 40 #include "utilities/powerOfTwo.hpp" 41 42 // Implementation of InterpreterMacroAssembler. 43 44 // This file specializes the assembler with interpreter-specific macros. 45 46 #ifdef PRODUCT 47 #define BLOCK_COMMENT(str) // nothing 48 #else 49 #define BLOCK_COMMENT(str) block_comment(str) 50 #endif 51 52 void InterpreterMacroAssembler::null_check_throw(Register a, int offset, Register temp_reg) { 53 address exception_entry = Interpreter::throw_NullPointerException_entry(); 54 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry); 55 } 56 57 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) { 58 assert(entry, "Entry must have been generated by now"); 59 if (is_within_range_of_b(entry, pc())) { 60 b(entry); 61 } else { 62 load_const_optimized(Rscratch, entry, R0); 63 mtctr(Rscratch); 64 bctr(); 65 } 66 } 67 68 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) { 69 Register bytecode = R12_scratch2; 70 if (bcp_incr != 0) { 71 lbzu(bytecode, bcp_incr, R14_bcp); 72 } else { 73 lbz(bytecode, 0, R14_bcp); 74 } 75 76 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state), generate_poll); 77 } 78 79 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 80 // Load current bytecode. 81 Register bytecode = R12_scratch2; 82 lbz(bytecode, 0, R14_bcp); 83 dispatch_Lbyte_code(state, bytecode, table); 84 } 85 86 // Dispatch code executed in the prolog of a bytecode which does not do it's 87 // own dispatch. The dispatch address is computed and placed in R24_dispatch_addr. 88 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 89 Register bytecode = R12_scratch2; 90 lbz(bytecode, bcp_incr, R14_bcp); 91 92 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state)); 93 94 sldi(bytecode, bytecode, LogBytesPerWord); 95 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode); 96 } 97 98 // Dispatch code executed in the epilog of a bytecode which does not do it's 99 // own dispatch. The dispatch address in R24_dispatch_addr is used for the 100 // dispatch. 101 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 102 if (bcp_incr) { addi(R14_bcp, R14_bcp, bcp_incr); } 103 mtctr(R24_dispatch_addr); 104 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable); 105 } 106 107 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 108 assert(scratch_reg != R0, "can't use R0 as scratch_reg here"); 109 if (JvmtiExport::can_pop_frame()) { 110 Label L; 111 112 // Check the "pending popframe condition" flag in the current thread. 113 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 114 115 // Initiate popframe handling only if it is not already being 116 // processed. If the flag has the popframe_processing bit set, it 117 // means that this code is called *during* popframe handling - we 118 // don't want to reenter. 119 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit); 120 beq(CCR0, L); 121 122 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit); 123 bne(CCR0, L); 124 125 // Call the Interpreter::remove_activation_preserving_args_entry() 126 // func to get the address of the same-named entrypoint in the 127 // generated interpreter code. 128 #if defined(ABI_ELFv2) 129 call_c(CAST_FROM_FN_PTR(address, 130 Interpreter::remove_activation_preserving_args_entry), 131 relocInfo::none); 132 #else 133 call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, 134 Interpreter::remove_activation_preserving_args_entry), 135 relocInfo::none); 136 #endif 137 138 // Jump to Interpreter::_remove_activation_preserving_args_entry. 139 mtctr(R3_RET); 140 bctr(); 141 142 align(32, 12); 143 bind(L); 144 } 145 } 146 147 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 148 const Register Rthr_state_addr = scratch_reg; 149 if (JvmtiExport::can_force_early_return()) { 150 Label Lno_early_ret; 151 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 152 cmpdi(CCR0, Rthr_state_addr, 0); 153 beq(CCR0, Lno_early_ret); 154 155 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr); 156 cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending); 157 bne(CCR0, Lno_early_ret); 158 159 // Jump to Interpreter::_earlyret_entry. 160 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr); 161 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry)); 162 mtlr(R3_RET); 163 blr(); 164 165 align(32, 12); 166 bind(Lno_early_ret); 167 } 168 } 169 170 void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) { 171 const Register RjvmtiState = Rscratch1; 172 const Register Rscratch2 = R0; 173 174 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 175 li(Rscratch2, 0); 176 177 switch (state) { 178 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 179 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 180 break; 181 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 182 break; 183 case btos: // fall through 184 case ztos: // fall through 185 case ctos: // fall through 186 case stos: // fall through 187 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 188 break; 189 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 190 break; 191 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 192 break; 193 case vtos: break; 194 default : ShouldNotReachHere(); 195 } 196 197 // Clean up tos value in the jvmti thread state. 198 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 199 // Set tos state field to illegal value. 200 li(Rscratch2, ilgl); 201 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState); 202 } 203 204 // Common code to dispatch and dispatch_only. 205 // Dispatch value in Lbyte_code and increment Lbcp. 206 207 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) { 208 address table_base = (address)Interpreter::dispatch_table((TosState)0); 209 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base; 210 if (is_simm16(table_offs)) { 211 addi(dst, R25_templateTableBase, (int)table_offs); 212 } else { 213 load_const_optimized(dst, table, R0); 214 } 215 } 216 217 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode, 218 address* table, bool generate_poll) { 219 assert_different_registers(bytecode, R11_scratch1); 220 221 // Calc dispatch table address. 222 load_dispatch_table(R11_scratch1, table); 223 224 if (generate_poll) { 225 address *sfpt_tbl = Interpreter::safept_table(state); 226 if (table != sfpt_tbl) { 227 Label dispatch; 228 ld(R0, in_bytes(JavaThread::polling_word_offset()), R16_thread); 229 // Armed page has poll_bit set, if poll bit is cleared just continue. 230 andi_(R0, R0, SafepointMechanism::poll_bit()); 231 beq(CCR0, dispatch); 232 load_dispatch_table(R11_scratch1, sfpt_tbl); 233 align(32, 16); 234 bind(dispatch); 235 } 236 } 237 238 sldi(R12_scratch2, bytecode, LogBytesPerWord); 239 ldx(R11_scratch1, R11_scratch1, R12_scratch2); 240 241 // Jump off! 242 mtctr(R11_scratch1); 243 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable); 244 } 245 246 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) { 247 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize); 248 ldx(Rrecv_dst, Rrecv_dst, R15_esp); 249 } 250 251 // helpers for expression stack 252 253 void InterpreterMacroAssembler::pop_i(Register r) { 254 lwzu(r, Interpreter::stackElementSize, R15_esp); 255 } 256 257 void InterpreterMacroAssembler::pop_ptr(Register r) { 258 ldu(r, Interpreter::stackElementSize, R15_esp); 259 } 260 261 void InterpreterMacroAssembler::pop_l(Register r) { 262 ld(r, Interpreter::stackElementSize, R15_esp); 263 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 264 } 265 266 void InterpreterMacroAssembler::pop_f(FloatRegister f) { 267 lfsu(f, Interpreter::stackElementSize, R15_esp); 268 } 269 270 void InterpreterMacroAssembler::pop_d(FloatRegister f) { 271 lfd(f, Interpreter::stackElementSize, R15_esp); 272 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 273 } 274 275 void InterpreterMacroAssembler::push_i(Register r) { 276 stw(r, 0, R15_esp); 277 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 278 } 279 280 void InterpreterMacroAssembler::push_ptr(Register r) { 281 std(r, 0, R15_esp); 282 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 283 } 284 285 void InterpreterMacroAssembler::push_l(Register r) { 286 // Clear unused slot. 287 load_const_optimized(R0, 0L); 288 std(R0, 0, R15_esp); 289 std(r, - Interpreter::stackElementSize, R15_esp); 290 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 291 } 292 293 void InterpreterMacroAssembler::push_f(FloatRegister f) { 294 stfs(f, 0, R15_esp); 295 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 296 } 297 298 void InterpreterMacroAssembler::push_d(FloatRegister f) { 299 stfd(f, - Interpreter::stackElementSize, R15_esp); 300 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 301 } 302 303 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) { 304 std(first, 0, R15_esp); 305 std(second, -Interpreter::stackElementSize, R15_esp); 306 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 307 } 308 309 void InterpreterMacroAssembler::move_l_to_d(Register l, FloatRegister d) { 310 if (VM_Version::has_mtfprd()) { 311 mtfprd(d, l); 312 } else { 313 std(l, 0, R15_esp); 314 lfd(d, 0, R15_esp); 315 } 316 } 317 318 void InterpreterMacroAssembler::move_d_to_l(FloatRegister d, Register l) { 319 if (VM_Version::has_mtfprd()) { 320 mffprd(l, d); 321 } else { 322 stfd(d, 0, R15_esp); 323 ld(l, 0, R15_esp); 324 } 325 } 326 327 void InterpreterMacroAssembler::push(TosState state) { 328 switch (state) { 329 case atos: push_ptr(); break; 330 case btos: 331 case ztos: 332 case ctos: 333 case stos: 334 case itos: push_i(); break; 335 case ltos: push_l(); break; 336 case ftos: push_f(); break; 337 case dtos: push_d(); break; 338 case vtos: /* nothing to do */ break; 339 default : ShouldNotReachHere(); 340 } 341 } 342 343 void InterpreterMacroAssembler::pop(TosState state) { 344 switch (state) { 345 case atos: pop_ptr(); break; 346 case btos: 347 case ztos: 348 case ctos: 349 case stos: 350 case itos: pop_i(); break; 351 case ltos: pop_l(); break; 352 case ftos: pop_f(); break; 353 case dtos: pop_d(); break; 354 case vtos: /* nothing to do */ break; 355 default : ShouldNotReachHere(); 356 } 357 verify_oop(R17_tos, state); 358 } 359 360 void InterpreterMacroAssembler::empty_expression_stack() { 361 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 362 } 363 364 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset, 365 Register Rdst, 366 signedOrNot is_signed) { 367 #if defined(VM_LITTLE_ENDIAN) 368 if (bcp_offset) { 369 load_const_optimized(Rdst, bcp_offset); 370 lhbrx(Rdst, R14_bcp, Rdst); 371 } else { 372 lhbrx(Rdst, R14_bcp); 373 } 374 if (is_signed == Signed) { 375 extsh(Rdst, Rdst); 376 } 377 #else 378 // Read Java big endian format. 379 if (is_signed == Signed) { 380 lha(Rdst, bcp_offset, R14_bcp); 381 } else { 382 lhz(Rdst, bcp_offset, R14_bcp); 383 } 384 #endif 385 } 386 387 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset, 388 Register Rdst, 389 signedOrNot is_signed) { 390 #if defined(VM_LITTLE_ENDIAN) 391 if (bcp_offset) { 392 load_const_optimized(Rdst, bcp_offset); 393 lwbrx(Rdst, R14_bcp, Rdst); 394 } else { 395 lwbrx(Rdst, R14_bcp); 396 } 397 if (is_signed == Signed) { 398 extsw(Rdst, Rdst); 399 } 400 #else 401 // Read Java big endian format. 402 if (bcp_offset & 3) { // Offset unaligned? 403 load_const_optimized(Rdst, bcp_offset); 404 if (is_signed == Signed) { 405 lwax(Rdst, R14_bcp, Rdst); 406 } else { 407 lwzx(Rdst, R14_bcp, Rdst); 408 } 409 } else { 410 if (is_signed == Signed) { 411 lwa(Rdst, bcp_offset, R14_bcp); 412 } else { 413 lwz(Rdst, bcp_offset, R14_bcp); 414 } 415 } 416 #endif 417 } 418 419 420 // Load the constant pool cache index from the bytecode stream. 421 // 422 // Kills / writes: 423 // - Rdst, Rscratch 424 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, 425 size_t index_size) { 426 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 427 // Cache index is always in the native format, courtesy of Rewriter. 428 if (index_size == sizeof(u2)) { 429 lhz(Rdst, bcp_offset, R14_bcp); 430 } else if (index_size == sizeof(u4)) { 431 if (bcp_offset & 3) { 432 load_const_optimized(Rdst, bcp_offset); 433 lwax(Rdst, R14_bcp, Rdst); 434 } else { 435 lwa(Rdst, bcp_offset, R14_bcp); 436 } 437 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 438 nand(Rdst, Rdst, Rdst); // convert to plain index 439 } else if (index_size == sizeof(u1)) { 440 lbz(Rdst, bcp_offset, R14_bcp); 441 } else { 442 ShouldNotReachHere(); 443 } 444 // Rdst now contains cp cache index. 445 } 446 447 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset, 448 size_t index_size) { 449 get_cache_index_at_bcp(cache, bcp_offset, index_size); 450 sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord)); 451 add(cache, R27_constPoolCache, cache); 452 } 453 454 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format) 455 // from (Rsrc)+offset. 456 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset, 457 signedOrNot is_signed) { 458 #if defined(VM_LITTLE_ENDIAN) 459 if (offset) { 460 load_const_optimized(Rdst, offset); 461 lwbrx(Rdst, Rdst, Rsrc); 462 } else { 463 lwbrx(Rdst, Rsrc); 464 } 465 if (is_signed == Signed) { 466 extsw(Rdst, Rdst); 467 } 468 #else 469 if (is_signed == Signed) { 470 lwa(Rdst, offset, Rsrc); 471 } else { 472 lwz(Rdst, offset, Rsrc); 473 } 474 #endif 475 } 476 477 // Load object from cpool->resolved_references(index). 478 // Kills: 479 // - index 480 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index, 481 Register tmp1, Register tmp2, 482 Label *L_handle_null) { 483 assert_different_registers(result, index, tmp1, tmp2); 484 assert(index->is_nonvolatile(), "needs to survive C-call in resolve_oop_handle"); 485 get_constant_pool(result); 486 487 // Convert from field index to resolved_references() index and from 488 // word index to byte offset. Since this is a java object, it can be compressed. 489 sldi(index, index, LogBytesPerHeapOop); 490 // Load pointer for resolved_references[] objArray. 491 ld(result, ConstantPool::cache_offset_in_bytes(), result); 492 ld(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result); 493 resolve_oop_handle(result, tmp1, tmp2, MacroAssembler::PRESERVATION_NONE); 494 #ifdef ASSERT 495 Label index_ok; 496 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result); 497 sldi(R0, R0, LogBytesPerHeapOop); 498 cmpd(CCR0, index, R0); 499 blt(CCR0, index_ok); 500 stop("resolved reference index out of bounds"); 501 bind(index_ok); 502 #endif 503 // Add in the index. 504 add(result, index, result); 505 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result, 506 tmp1, tmp2, 507 MacroAssembler::PRESERVATION_NONE, 508 0, L_handle_null); 509 } 510 511 // load cpool->resolved_klass_at(index) 512 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool, Register Roffset, Register Rklass) { 513 // int value = *(Rcpool->int_at_addr(which)); 514 // int resolved_klass_index = extract_low_short_from_int(value); 515 add(Roffset, Rcpool, Roffset); 516 #if defined(VM_LITTLE_ENDIAN) 517 lhz(Roffset, sizeof(ConstantPool), Roffset); // Roffset = resolved_klass_index 518 #else 519 lhz(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index 520 #endif 521 522 ld(Rklass, ConstantPool::resolved_klasses_offset_in_bytes(), Rcpool); // Rklass = Rcpool->_resolved_klasses 523 524 sldi(Roffset, Roffset, LogBytesPerWord); 525 addi(Roffset, Roffset, Array<Klass*>::base_offset_in_bytes()); 526 isync(); // Order load of instance Klass wrt. tags. 527 ldx(Rklass, Rklass, Roffset); 528 } 529 530 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no, 531 Register cache, 532 Register method) { 533 const int method_offset = in_bytes( 534 ConstantPoolCache::base_offset() + 535 ((byte_no == TemplateTable::f2_byte) 536 ? ConstantPoolCacheEntry::f2_offset() 537 : ConstantPoolCacheEntry::f1_offset())); 538 539 ld(method, method_offset, cache); // get f1 Method* 540 } 541 542 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 543 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2. 544 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1, 545 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) { 546 // Profile the not-null value's klass. 547 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2); 548 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype); 549 profile_typecheck_failed(Rtmp1, Rtmp2); 550 } 551 552 // Separate these two to allow for delay slot in middle. 553 // These are used to do a test and full jump to exception-throwing code. 554 555 // Check that index is in range for array, then shift index by index_shift, 556 // and put arrayOop + shifted_index into res. 557 // Note: res is still shy of address by array offset into object. 558 559 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, 560 int index_shift, Register Rtmp, Register Rres) { 561 // Check that index is in range for array, then shift index by index_shift, 562 // and put arrayOop + shifted_index into res. 563 // Note: res is still shy of address by array offset into object. 564 // Kills: 565 // - Rindex 566 // Writes: 567 // - Rres: Address that corresponds to the array index if check was successful. 568 verify_oop(Rarray); 569 const Register Rlength = R0; 570 const Register RsxtIndex = Rtmp; 571 Label LisNull, LnotOOR; 572 573 // Array nullcheck 574 if (!ImplicitNullChecks) { 575 cmpdi(CCR0, Rarray, 0); 576 beq(CCR0, LisNull); 577 } else { 578 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex); 579 } 580 581 // Rindex might contain garbage in upper bits (remember that we don't sign extend 582 // during integer arithmetic operations). So kill them and put value into same register 583 // where ArrayIndexOutOfBounds would expect the index in. 584 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit 585 586 // Index check 587 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray); 588 cmplw(CCR0, Rindex, Rlength); 589 sldi(RsxtIndex, RsxtIndex, index_shift); 590 blt(CCR0, LnotOOR); 591 // Index should be in R17_tos, array should be in R4_ARG2. 592 mr_if_needed(R17_tos, Rindex); 593 mr_if_needed(R4_ARG2, Rarray); 594 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry); 595 mtctr(Rtmp); 596 bctr(); 597 598 if (!ImplicitNullChecks) { 599 bind(LisNull); 600 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry); 601 mtctr(Rtmp); 602 bctr(); 603 } 604 605 align(32, 16); 606 bind(LnotOOR); 607 608 // Calc address 609 add(Rres, RsxtIndex, Rarray); 610 } 611 612 void InterpreterMacroAssembler::index_check(Register array, Register index, 613 int index_shift, Register tmp, Register res) { 614 // pop array 615 pop_ptr(array); 616 617 // check array 618 index_check_without_pop(array, index, index_shift, tmp, res); 619 } 620 621 void InterpreterMacroAssembler::get_const(Register Rdst) { 622 ld(Rdst, in_bytes(Method::const_offset()), R19_method); 623 } 624 625 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 626 get_const(Rdst); 627 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 628 } 629 630 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 631 get_constant_pool(Rdst); 632 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 633 } 634 635 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 636 get_constant_pool(Rcpool); 637 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool); 638 } 639 640 // Unlock if synchronized method. 641 // 642 // Unlock the receiver if this is a synchronized method. 643 // Unlock any Java monitors from synchronized blocks. 644 // 645 // If there are locked Java monitors 646 // If throw_monitor_exception 647 // throws IllegalMonitorStateException 648 // Else if install_monitor_exception 649 // installs IllegalMonitorStateException 650 // Else 651 // no error processing 652 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 653 bool throw_monitor_exception, 654 bool install_monitor_exception) { 655 Label Lunlocked, Lno_unlock; 656 { 657 Register Rdo_not_unlock_flag = R11_scratch1; 658 Register Raccess_flags = R12_scratch2; 659 660 // Check if synchronized method or unlocking prevented by 661 // JavaThread::do_not_unlock_if_synchronized flag. 662 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 663 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method); 664 li(R0, 0); 665 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag 666 667 push(state); 668 669 // Skip if we don't have to unlock. 670 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0. 671 beq(CCR0, Lunlocked); 672 673 cmpwi(CCR0, Rdo_not_unlock_flag, 0); 674 bne(CCR0, Lno_unlock); 675 } 676 677 // Unlock 678 { 679 Register Rmonitor_base = R11_scratch1; 680 681 Label Lunlock; 682 // If it's still locked, everything is ok, unlock it. 683 ld(Rmonitor_base, 0, R1_SP); 684 addi(Rmonitor_base, Rmonitor_base, 685 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base 686 687 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base); 688 cmpdi(CCR0, R0, 0); 689 bne(CCR0, Lunlock); 690 691 // If it's already unlocked, throw exception. 692 if (throw_monitor_exception) { 693 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 694 should_not_reach_here(); 695 } else { 696 if (install_monitor_exception) { 697 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 698 b(Lunlocked); 699 } 700 } 701 702 bind(Lunlock); 703 unlock_object(Rmonitor_base); 704 } 705 706 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not. 707 bind(Lunlocked); 708 { 709 Label Lexception, Lrestart; 710 Register Rcurrent_obj_addr = R11_scratch1; 711 const int delta = frame::interpreter_frame_monitor_size_in_bytes(); 712 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords"); 713 714 bind(Lrestart); 715 // Set up search loop: Calc num of iterations. 716 { 717 Register Riterations = R12_scratch2; 718 Register Rmonitor_base = Rcurrent_obj_addr; 719 ld(Rmonitor_base, 0, R1_SP); 720 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base 721 722 subf_(Riterations, R26_monitor, Rmonitor_base); 723 ble(CCR0, Lno_unlock); 724 725 addi(Rcurrent_obj_addr, Rmonitor_base, 726 BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes()); 727 // Check if any monitor is on stack, bail out if not 728 srdi(Riterations, Riterations, exact_log2(delta)); 729 mtctr(Riterations); 730 } 731 732 // The search loop: Look for locked monitors. 733 { 734 const Register Rcurrent_obj = R0; 735 Label Lloop; 736 737 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 738 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 739 bind(Lloop); 740 741 // Check if current entry is used. 742 cmpdi(CCR0, Rcurrent_obj, 0); 743 bne(CCR0, Lexception); 744 // Preload next iteration's compare value. 745 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 746 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 747 bdnz(Lloop); 748 } 749 // Fell through: Everything's unlocked => finish. 750 b(Lno_unlock); 751 752 // An object is still locked => need to throw exception. 753 bind(Lexception); 754 if (throw_monitor_exception) { 755 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 756 should_not_reach_here(); 757 } else { 758 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 759 // Unlock does not block, so don't have to worry about the frame. 760 Register Rmonitor_addr = R11_scratch1; 761 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta); 762 unlock_object(Rmonitor_addr); 763 if (install_monitor_exception) { 764 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 765 } 766 b(Lrestart); 767 } 768 } 769 770 align(32, 12); 771 bind(Lno_unlock); 772 pop(state); 773 } 774 775 // Support function for remove_activation & Co. 776 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, 777 Register Rscratch1, Register Rscratch2) { 778 // Pop interpreter frame. 779 ld(Rscratch1, 0, R1_SP); // *SP 780 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp 781 ld(Rscratch2, 0, Rscratch1); // **SP 782 if (return_pc!=noreg) { 783 ld(return_pc, _abi0(lr), Rscratch1); // LR 784 } 785 786 // Merge top frames. 787 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP 788 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP 789 } 790 791 void InterpreterMacroAssembler::narrow(Register result) { 792 Register ret_type = R11_scratch1; 793 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method); 794 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1); 795 796 Label notBool, notByte, notChar, done; 797 798 // common case first 799 cmpwi(CCR0, ret_type, T_INT); 800 beq(CCR0, done); 801 802 cmpwi(CCR0, ret_type, T_BOOLEAN); 803 bne(CCR0, notBool); 804 andi(result, result, 0x1); 805 b(done); 806 807 bind(notBool); 808 cmpwi(CCR0, ret_type, T_BYTE); 809 bne(CCR0, notByte); 810 extsb(result, result); 811 b(done); 812 813 bind(notByte); 814 cmpwi(CCR0, ret_type, T_CHAR); 815 bne(CCR0, notChar); 816 andi(result, result, 0xffff); 817 b(done); 818 819 bind(notChar); 820 // cmpwi(CCR0, ret_type, T_SHORT); // all that's left 821 // bne(CCR0, done); 822 extsh(result, result); 823 824 // Nothing to do for T_INT 825 bind(done); 826 } 827 828 // Remove activation. 829 // 830 // Apply stack watermark barrier. 831 // Unlock the receiver if this is a synchronized method. 832 // Unlock any Java monitors from synchronized blocks. 833 // Remove the activation from the stack. 834 // 835 // If there are locked Java monitors 836 // If throw_monitor_exception 837 // throws IllegalMonitorStateException 838 // Else if install_monitor_exception 839 // installs IllegalMonitorStateException 840 // Else 841 // no error processing 842 void InterpreterMacroAssembler::remove_activation(TosState state, 843 bool throw_monitor_exception, 844 bool install_monitor_exception) { 845 BLOCK_COMMENT("remove_activation {"); 846 847 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily, 848 // that would normally not be safe to use. Such bad returns into unsafe territory of 849 // the stack, will call InterpreterRuntime::at_unwind. 850 Label slow_path; 851 Label fast_path; 852 safepoint_poll(slow_path, R11_scratch1, true /* at_return */, false /* in_nmethod */); 853 b(fast_path); 854 bind(slow_path); 855 push(state); 856 set_last_Java_frame(R1_SP, noreg); 857 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), R16_thread); 858 reset_last_Java_frame(); 859 pop(state); 860 align(32); 861 bind(fast_path); 862 863 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 864 865 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 866 notify_method_exit(false, state, NotifyJVMTI, true); 867 868 BLOCK_COMMENT("reserved_stack_check:"); 869 if (StackReservedPages > 0) { 870 // Test if reserved zone needs to be enabled. 871 Label no_reserved_zone_enabling; 872 873 // Compare frame pointers. There is no good stack pointer, as with stack 874 // frame compression we can get different SPs when we do calls. A subsequent 875 // call could have a smaller SP, so that this compare succeeds for an 876 // inner call of the method annotated with ReservedStack. 877 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread); 878 ld_ptr(R11_scratch1, _abi0(callers_sp), R1_SP); // Load frame pointer. 879 cmpld(CCR0, R11_scratch1, R0); 880 blt_predict_taken(CCR0, no_reserved_zone_enabling); 881 882 // Enable reserved zone again, throw stack overflow exception. 883 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread); 884 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError)); 885 886 should_not_reach_here(); 887 888 bind(no_reserved_zone_enabling); 889 } 890 891 verify_oop(R17_tos, state); 892 verify_thread(); 893 894 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 895 mtlr(R0); 896 BLOCK_COMMENT("} remove_activation"); 897 } 898 899 // Lock object 900 // 901 // Registers alive 902 // monitor - Address of the BasicObjectLock to be used for locking, 903 // which must be initialized with the object to lock. 904 // object - Address of the object to be locked. 905 // 906 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 907 if (UseHeavyMonitors) { 908 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor); 909 } else { 910 // template code: 911 // 912 // markWord displaced_header = obj->mark().set_unlocked(); 913 // monitor->lock()->set_displaced_header(displaced_header); 914 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) { 915 // // We stored the monitor address into the object's mark word. 916 // } else if (THREAD->is_lock_owned((address)displaced_header)) 917 // // Simple recursive case. 918 // monitor->lock()->set_displaced_header(NULL); 919 // } else { 920 // // Slow path. 921 // InterpreterRuntime::monitorenter(THREAD, monitor); 922 // } 923 924 const Register displaced_header = R7_ARG5; 925 const Register object_mark_addr = R8_ARG6; 926 const Register current_header = R9_ARG7; 927 const Register tmp = R10_ARG8; 928 929 Label done; 930 Label cas_failed, slow_case; 931 932 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp); 933 934 // markWord displaced_header = obj->mark().set_unlocked(); 935 936 // Load markWord from object into displaced_header. 937 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object); 938 939 if (DiagnoseSyncOnValueBasedClasses != 0) { 940 load_klass(tmp, object); 941 lwz(tmp, in_bytes(Klass::access_flags_offset()), tmp); 942 testbitdi(CCR0, R0, tmp, exact_log2(JVM_ACC_IS_VALUE_BASED_CLASS)); 943 bne(CCR0, slow_case); 944 } 945 946 // Set displaced_header to be (markWord of object | UNLOCK_VALUE). 947 ori(displaced_header, displaced_header, markWord::unlocked_value); 948 949 // monitor->lock()->set_displaced_header(displaced_header); 950 951 // Initialize the box (Must happen before we update the object mark!). 952 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 953 BasicLock::displaced_header_offset_in_bytes(), monitor); 954 955 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) { 956 957 // Store stack address of the BasicObjectLock (this is monitor) into object. 958 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 959 960 // Must fence, otherwise, preceding store(s) may float below cmpxchg. 961 // CmpxchgX sets CCR0 to cmpX(current, displaced). 962 cmpxchgd(/*flag=*/CCR0, 963 /*current_value=*/current_header, 964 /*compare_value=*/displaced_header, /*exchange_value=*/monitor, 965 /*where=*/object_mark_addr, 966 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq, 967 MacroAssembler::cmpxchgx_hint_acquire_lock(), 968 noreg, 969 &cas_failed, 970 /*check without membar and ldarx first*/true); 971 972 // If the compare-and-exchange succeeded, then we found an unlocked 973 // object and we have now locked it. 974 b(done); 975 bind(cas_failed); 976 977 // } else if (THREAD->is_lock_owned((address)displaced_header)) 978 // // Simple recursive case. 979 // monitor->lock()->set_displaced_header(NULL); 980 981 // We did not see an unlocked object so try the fast recursive case. 982 983 // Check if owner is self by comparing the value in the markWord of object 984 // (current_header) with the stack pointer. 985 sub(current_header, current_header, R1_SP); 986 987 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 988 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markWord::lock_mask_in_place); 989 990 and_(R0/*==0?*/, current_header, tmp); 991 // If condition is true we are done and hence we can store 0 in the displaced 992 // header indicating it is a recursive lock. 993 bne(CCR0, slow_case); 994 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() + 995 BasicLock::displaced_header_offset_in_bytes(), monitor); 996 b(done); 997 998 // } else { 999 // // Slow path. 1000 // InterpreterRuntime::monitorenter(THREAD, monitor); 1001 1002 // None of the above fast optimizations worked so we have to get into the 1003 // slow case of monitor enter. 1004 bind(slow_case); 1005 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor); 1006 // } 1007 align(32, 12); 1008 bind(done); 1009 } 1010 } 1011 1012 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 1013 // 1014 // Registers alive 1015 // monitor - Address of the BasicObjectLock to be used for locking, 1016 // which must be initialized with the object to lock. 1017 // 1018 // Throw IllegalMonitorException if object is not locked by current thread. 1019 void InterpreterMacroAssembler::unlock_object(Register monitor) { 1020 if (UseHeavyMonitors) { 1021 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor); 1022 } else { 1023 1024 // template code: 1025 // 1026 // if ((displaced_header = monitor->displaced_header()) == NULL) { 1027 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 1028 // monitor->set_obj(NULL); 1029 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) { 1030 // // We swapped the unlocked mark in displaced_header into the object's mark word. 1031 // monitor->set_obj(NULL); 1032 // } else { 1033 // // Slow path. 1034 // InterpreterRuntime::monitorexit(monitor); 1035 // } 1036 1037 const Register object = R7_ARG5; 1038 const Register displaced_header = R8_ARG6; 1039 const Register object_mark_addr = R9_ARG7; 1040 const Register current_header = R10_ARG8; 1041 1042 Label free_slot; 1043 Label slow_case; 1044 1045 assert_different_registers(object, displaced_header, object_mark_addr, current_header); 1046 1047 // Test first if we are in the fast recursive case. 1048 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 1049 BasicLock::displaced_header_offset_in_bytes(), monitor); 1050 1051 // If the displaced header is zero, we have a recursive unlock. 1052 cmpdi(CCR0, displaced_header, 0); 1053 beq(CCR0, free_slot); // recursive unlock 1054 1055 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) { 1056 // // We swapped the unlocked mark in displaced_header into the object's mark word. 1057 // monitor->set_obj(NULL); 1058 1059 // If we still have a lightweight lock, unlock the object and be done. 1060 1061 // The object address from the monitor is in object. 1062 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); 1063 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 1064 1065 // We have the displaced header in displaced_header. If the lock is still 1066 // lightweight, it will contain the monitor address and we'll store the 1067 // displaced header back into the object's mark word. 1068 // CmpxchgX sets CCR0 to cmpX(current, monitor). 1069 cmpxchgd(/*flag=*/CCR0, 1070 /*current_value=*/current_header, 1071 /*compare_value=*/monitor, /*exchange_value=*/displaced_header, 1072 /*where=*/object_mark_addr, 1073 MacroAssembler::MemBarRel, 1074 MacroAssembler::cmpxchgx_hint_release_lock(), 1075 noreg, 1076 &slow_case); 1077 b(free_slot); 1078 1079 // } else { 1080 // // Slow path. 1081 // InterpreterRuntime::monitorexit(monitor); 1082 1083 // The lock has been converted into a heavy lock and hence 1084 // we need to get into the slow case. 1085 bind(slow_case); 1086 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor); 1087 // } 1088 1089 Label done; 1090 b(done); // Monitor register may be overwritten! Runtime has already freed the slot. 1091 1092 // Exchange worked, do monitor->set_obj(NULL); 1093 align(32, 12); 1094 bind(free_slot); 1095 li(R0, 0); 1096 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor); 1097 bind(done); 1098 } 1099 } 1100 1101 // Load compiled (i2c) or interpreter entry when calling from interpreted and 1102 // do the call. Centralized so that all interpreter calls will do the same actions. 1103 // If jvmti single stepping is on for a thread we must not call compiled code. 1104 // 1105 // Input: 1106 // - Rtarget_method: method to call 1107 // - Rret_addr: return address 1108 // - 2 scratch regs 1109 // 1110 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, 1111 Register Rscratch1, Register Rscratch2) { 1112 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr); 1113 // Assume we want to go compiled if available. 1114 const Register Rtarget_addr = Rscratch1; 1115 const Register Rinterp_only = Rscratch2; 1116 1117 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method); 1118 1119 if (JvmtiExport::can_post_interpreter_events()) { 1120 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 1121 1122 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 1123 // compiled code in threads for which the event is enabled. Check here for 1124 // interp_only_mode if these events CAN be enabled. 1125 Label done; 1126 verify_thread(); 1127 cmpwi(CCR0, Rinterp_only, 0); 1128 beq(CCR0, done); 1129 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method); 1130 align(32, 12); 1131 bind(done); 1132 } 1133 1134 #ifdef ASSERT 1135 { 1136 Label Lok; 1137 cmpdi(CCR0, Rtarget_addr, 0); 1138 bne(CCR0, Lok); 1139 stop("null entry point"); 1140 bind(Lok); 1141 } 1142 #endif // ASSERT 1143 1144 mr(R21_sender_SP, R1_SP); 1145 1146 // Calc a precise SP for the call. The SP value we calculated in 1147 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space 1148 // if esp is not max. Also, the i2c adapter extends the stack space without restoring 1149 // our pre-calced value, so repeating calls via i2c would result in stack overflow. 1150 // Since esp already points to an empty slot, we just have to sub 1 additional slot 1151 // to meet the abi scratch requirements. 1152 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in 1153 // the return entry of the interpreter. 1154 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size); 1155 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address 1156 resize_frame_absolute(Rscratch2, Rscratch2, R0); 1157 1158 mr_if_needed(R19_method, Rtarget_method); 1159 mtctr(Rtarget_addr); 1160 mtlr(Rret_addr); 1161 1162 save_interpreter_state(Rscratch2); 1163 #ifdef ASSERT 1164 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp 1165 cmpd(CCR0, R21_sender_SP, Rscratch1); 1166 asm_assert_eq("top_frame_sp incorrect"); 1167 #endif 1168 1169 bctr(); 1170 } 1171 1172 // Set the method data pointer for the current bcp. 1173 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1174 assert(ProfileInterpreter, "must be profiling interpreter"); 1175 Label get_continue; 1176 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method); 1177 test_method_data_pointer(get_continue); 1178 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp); 1179 1180 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1181 add(R28_mdx, R28_mdx, R3_RET); 1182 bind(get_continue); 1183 } 1184 1185 // Test ImethodDataPtr. If it is null, continue at the specified label. 1186 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1187 assert(ProfileInterpreter, "must be profiling interpreter"); 1188 cmpdi(CCR0, R28_mdx, 0); 1189 beq(CCR0, zero_continue); 1190 } 1191 1192 void InterpreterMacroAssembler::verify_method_data_pointer() { 1193 assert(ProfileInterpreter, "must be profiling interpreter"); 1194 #ifdef ASSERT 1195 Label verify_continue; 1196 test_method_data_pointer(verify_continue); 1197 1198 // If the mdp is valid, it will point to a DataLayout header which is 1199 // consistent with the bcp. The converse is highly probable also. 1200 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx); 1201 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method); 1202 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset())); 1203 add(R11_scratch1, R12_scratch2, R12_scratch2); 1204 cmpd(CCR0, R11_scratch1, R14_bcp); 1205 beq(CCR0, verify_continue); 1206 1207 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx); 1208 1209 bind(verify_continue); 1210 #endif 1211 } 1212 1213 // Store a value at some constant offset from the method data pointer. 1214 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1215 assert(ProfileInterpreter, "must be profiling interpreter"); 1216 1217 std(value, constant, R28_mdx); 1218 } 1219 1220 // Increment the value at some constant offset from the method data pointer. 1221 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1222 Register counter_addr, 1223 Register Rbumped_count, 1224 bool decrement) { 1225 // Locate the counter at a fixed offset from the mdp: 1226 addi(counter_addr, R28_mdx, constant); 1227 increment_mdp_data_at(counter_addr, Rbumped_count, decrement); 1228 } 1229 1230 // Increment the value at some non-fixed (reg + constant) offset from 1231 // the method data pointer. 1232 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1233 int constant, 1234 Register scratch, 1235 Register Rbumped_count, 1236 bool decrement) { 1237 // Add the constant to reg to get the offset. 1238 add(scratch, R28_mdx, reg); 1239 // Then calculate the counter address. 1240 addi(scratch, scratch, constant); 1241 increment_mdp_data_at(scratch, Rbumped_count, decrement); 1242 } 1243 1244 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr, 1245 Register Rbumped_count, 1246 bool decrement) { 1247 assert(ProfileInterpreter, "must be profiling interpreter"); 1248 1249 // Load the counter. 1250 ld(Rbumped_count, 0, counter_addr); 1251 1252 if (decrement) { 1253 // Decrement the register. Set condition codes. 1254 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment); 1255 // Store the decremented counter, if it is still negative. 1256 std(Rbumped_count, 0, counter_addr); 1257 // Note: add/sub overflow check are not ported, since 64 bit 1258 // calculation should never overflow. 1259 } else { 1260 // Increment the register. Set carry flag. 1261 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment); 1262 // Store the incremented counter. 1263 std(Rbumped_count, 0, counter_addr); 1264 } 1265 } 1266 1267 // Set a flag value at the current method data pointer position. 1268 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1269 Register scratch) { 1270 assert(ProfileInterpreter, "must be profiling interpreter"); 1271 // Load the data header. 1272 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1273 // Set the flag. 1274 ori(scratch, scratch, flag_constant); 1275 // Store the modified header. 1276 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1277 } 1278 1279 // Test the location at some offset from the method data pointer. 1280 // If it is not equal to value, branch to the not_equal_continue Label. 1281 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1282 Register value, 1283 Label& not_equal_continue, 1284 Register test_out) { 1285 assert(ProfileInterpreter, "must be profiling interpreter"); 1286 1287 ld(test_out, offset, R28_mdx); 1288 cmpd(CCR0, value, test_out); 1289 bne(CCR0, not_equal_continue); 1290 } 1291 1292 // Update the method data pointer by the displacement located at some fixed 1293 // offset from the method data pointer. 1294 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1295 Register scratch) { 1296 assert(ProfileInterpreter, "must be profiling interpreter"); 1297 1298 ld(scratch, offset_of_disp, R28_mdx); 1299 add(R28_mdx, scratch, R28_mdx); 1300 } 1301 1302 // Update the method data pointer by the displacement located at the 1303 // offset (reg + offset_of_disp). 1304 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1305 int offset_of_disp, 1306 Register scratch) { 1307 assert(ProfileInterpreter, "must be profiling interpreter"); 1308 1309 add(scratch, reg, R28_mdx); 1310 ld(scratch, offset_of_disp, scratch); 1311 add(R28_mdx, scratch, R28_mdx); 1312 } 1313 1314 // Update the method data pointer by a simple constant displacement. 1315 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1316 assert(ProfileInterpreter, "must be profiling interpreter"); 1317 addi(R28_mdx, R28_mdx, constant); 1318 } 1319 1320 // Update the method data pointer for a _ret bytecode whose target 1321 // was not among our cached targets. 1322 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1323 Register return_bci) { 1324 assert(ProfileInterpreter, "must be profiling interpreter"); 1325 1326 push(state); 1327 assert(return_bci->is_nonvolatile(), "need to protect return_bci"); 1328 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1329 pop(state); 1330 } 1331 1332 // Increments the backedge counter. 1333 // Returns backedge counter + invocation counter in Rdst. 1334 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst, 1335 const Register Rtmp1, Register Rscratch) { 1336 assert(UseCompiler, "incrementing must be useful"); 1337 assert_different_registers(Rdst, Rtmp1); 1338 const Register invocation_counter = Rtmp1; 1339 const Register counter = Rdst; 1340 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size."); 1341 1342 // Load backedge counter. 1343 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1344 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1345 // Load invocation counter. 1346 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) + 1347 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1348 1349 // Add the delta to the backedge counter. 1350 addi(counter, counter, InvocationCounter::count_increment); 1351 1352 // Mask the invocation counter. 1353 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value); 1354 1355 // Store new counter value. 1356 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1357 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1358 // Return invocation counter + backedge counter. 1359 add(counter, counter, invocation_counter); 1360 } 1361 1362 // Count a taken branch in the bytecodes. 1363 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1364 if (ProfileInterpreter) { 1365 Label profile_continue; 1366 1367 // If no method data exists, go to profile_continue. 1368 test_method_data_pointer(profile_continue); 1369 1370 // We are taking a branch. Increment the taken count. 1371 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count); 1372 1373 // The method data pointer needs to be updated to reflect the new target. 1374 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1375 bind (profile_continue); 1376 } 1377 } 1378 1379 // Count a not-taken branch in the bytecodes. 1380 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) { 1381 if (ProfileInterpreter) { 1382 Label profile_continue; 1383 1384 // If no method data exists, go to profile_continue. 1385 test_method_data_pointer(profile_continue); 1386 1387 // We are taking a branch. Increment the not taken count. 1388 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2); 1389 1390 // The method data pointer needs to be updated to correspond to the 1391 // next bytecode. 1392 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1393 bind (profile_continue); 1394 } 1395 } 1396 1397 // Count a non-virtual call in the bytecodes. 1398 void InterpreterMacroAssembler::profile_call(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 making a call. Increment the count. 1406 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1407 1408 // The method data pointer needs to be updated to reflect the new target. 1409 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1410 bind (profile_continue); 1411 } 1412 } 1413 1414 // Count a final call in the bytecodes. 1415 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) { 1416 if (ProfileInterpreter) { 1417 Label profile_continue; 1418 1419 // If no method data exists, go to profile_continue. 1420 test_method_data_pointer(profile_continue); 1421 1422 // We are making a call. Increment the count. 1423 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1424 1425 // The method data pointer needs to be updated to reflect the new target. 1426 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1427 bind (profile_continue); 1428 } 1429 } 1430 1431 // Count a virtual call in the bytecodes. 1432 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver, 1433 Register Rscratch1, 1434 Register Rscratch2, 1435 bool receiver_can_be_null) { 1436 if (!ProfileInterpreter) { return; } 1437 Label profile_continue; 1438 1439 // If no method data exists, go to profile_continue. 1440 test_method_data_pointer(profile_continue); 1441 1442 Label skip_receiver_profile; 1443 if (receiver_can_be_null) { 1444 Label not_null; 1445 cmpdi(CCR0, Rreceiver, 0); 1446 bne(CCR0, not_null); 1447 // We are making a call. Increment the count for null receiver. 1448 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2); 1449 b(skip_receiver_profile); 1450 bind(not_null); 1451 } 1452 1453 // Record the receiver type. 1454 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true); 1455 bind(skip_receiver_profile); 1456 1457 // The method data pointer needs to be updated to reflect the new target. 1458 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1459 bind (profile_continue); 1460 } 1461 1462 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) { 1463 if (ProfileInterpreter) { 1464 Label profile_continue; 1465 1466 // If no method data exists, go to profile_continue. 1467 test_method_data_pointer(profile_continue); 1468 1469 int mdp_delta = in_bytes(BitData::bit_data_size()); 1470 if (TypeProfileCasts) { 1471 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1472 1473 // Record the object type. 1474 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false); 1475 } 1476 1477 // The method data pointer needs to be updated. 1478 update_mdp_by_constant(mdp_delta); 1479 1480 bind (profile_continue); 1481 } 1482 } 1483 1484 void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) { 1485 if (ProfileInterpreter && TypeProfileCasts) { 1486 Label profile_continue; 1487 1488 // If no method data exists, go to profile_continue. 1489 test_method_data_pointer(profile_continue); 1490 1491 int count_offset = in_bytes(CounterData::count_offset()); 1492 // Back up the address, since we have already bumped the mdp. 1493 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1494 1495 // *Decrement* the counter. We expect to see zero or small negatives. 1496 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true); 1497 1498 bind (profile_continue); 1499 } 1500 } 1501 1502 // Count a ret in the bytecodes. 1503 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, 1504 Register scratch1, Register scratch2) { 1505 if (ProfileInterpreter) { 1506 Label profile_continue; 1507 uint row; 1508 1509 // If no method data exists, go to profile_continue. 1510 test_method_data_pointer(profile_continue); 1511 1512 // Update the total ret count. 1513 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 ); 1514 1515 for (row = 0; row < RetData::row_limit(); row++) { 1516 Label next_test; 1517 1518 // See if return_bci is equal to bci[n]: 1519 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1); 1520 1521 // return_bci is equal to bci[n]. Increment the count. 1522 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2); 1523 1524 // The method data pointer needs to be updated to reflect the new target. 1525 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1); 1526 b(profile_continue); 1527 bind(next_test); 1528 } 1529 1530 update_mdp_for_ret(state, return_bci); 1531 1532 bind (profile_continue); 1533 } 1534 } 1535 1536 // Count the default case of a switch construct. 1537 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) { 1538 if (ProfileInterpreter) { 1539 Label profile_continue; 1540 1541 // If no method data exists, go to profile_continue. 1542 test_method_data_pointer(profile_continue); 1543 1544 // Update the default case count 1545 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1546 scratch1, scratch2); 1547 1548 // The method data pointer needs to be updated. 1549 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()), 1550 scratch1); 1551 1552 bind (profile_continue); 1553 } 1554 } 1555 1556 // Count the index'th case of a switch construct. 1557 void InterpreterMacroAssembler::profile_switch_case(Register index, 1558 Register scratch1, 1559 Register scratch2, 1560 Register scratch3) { 1561 if (ProfileInterpreter) { 1562 assert_different_registers(index, scratch1, scratch2, scratch3); 1563 Label profile_continue; 1564 1565 // If no method data exists, go to profile_continue. 1566 test_method_data_pointer(profile_continue); 1567 1568 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes(). 1569 li(scratch3, in_bytes(MultiBranchData::case_array_offset())); 1570 1571 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works"); 1572 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1573 add(scratch1, scratch1, scratch3); 1574 1575 // Update the case count. 1576 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3); 1577 1578 // The method data pointer needs to be updated. 1579 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2); 1580 1581 bind (profile_continue); 1582 } 1583 } 1584 1585 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) { 1586 if (ProfileInterpreter) { 1587 assert_different_registers(Rscratch1, Rscratch2); 1588 Label profile_continue; 1589 1590 // If no method data exists, go to profile_continue. 1591 test_method_data_pointer(profile_continue); 1592 1593 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1); 1594 1595 // The method data pointer needs to be updated. 1596 int mdp_delta = in_bytes(BitData::bit_data_size()); 1597 if (TypeProfileCasts) { 1598 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1599 } 1600 update_mdp_by_constant(mdp_delta); 1601 1602 bind (profile_continue); 1603 } 1604 } 1605 1606 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver, 1607 Register Rscratch1, Register Rscratch2, 1608 bool is_virtual_call) { 1609 assert(ProfileInterpreter, "must be profiling"); 1610 assert_different_registers(Rreceiver, Rscratch1, Rscratch2); 1611 1612 Label done; 1613 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call); 1614 bind (done); 1615 } 1616 1617 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1618 Register receiver, Register scratch1, Register scratch2, 1619 int start_row, Label& done, bool is_virtual_call) { 1620 if (TypeProfileWidth == 0) { 1621 if (is_virtual_call) { 1622 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1623 } 1624 return; 1625 } 1626 1627 int last_row = VirtualCallData::row_limit() - 1; 1628 assert(start_row <= last_row, "must be work left to do"); 1629 // Test this row for both the receiver and for null. 1630 // Take any of three different outcomes: 1631 // 1. found receiver => increment count and goto done 1632 // 2. found null => keep looking for case 1, maybe allocate this cell 1633 // 3. found something else => keep looking for cases 1 and 2 1634 // Case 3 is handled by a recursive call. 1635 for (int row = start_row; row <= last_row; row++) { 1636 Label next_test; 1637 bool test_for_null_also = (row == start_row); 1638 1639 // See if the receiver is receiver[n]. 1640 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1641 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1); 1642 // delayed()->tst(scratch); 1643 1644 // The receiver is receiver[n]. Increment count[n]. 1645 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1646 increment_mdp_data_at(count_offset, scratch1, scratch2); 1647 b(done); 1648 bind(next_test); 1649 1650 if (test_for_null_also) { 1651 Label found_null; 1652 // Failed the equality check on receiver[n]... Test for null. 1653 if (start_row == last_row) { 1654 // The only thing left to do is handle the null case. 1655 if (is_virtual_call) { 1656 // Scratch1 contains test_out from test_mdp_data_at. 1657 cmpdi(CCR0, scratch1, 0); 1658 beq(CCR0, found_null); 1659 // Receiver did not match any saved receiver and there is no empty row for it. 1660 // Increment total counter to indicate polymorphic case. 1661 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1662 b(done); 1663 bind(found_null); 1664 } else { 1665 cmpdi(CCR0, scratch1, 0); 1666 bne(CCR0, done); 1667 } 1668 break; 1669 } 1670 // Since null is rare, make it be the branch-taken case. 1671 cmpdi(CCR0, scratch1, 0); 1672 beq(CCR0, found_null); 1673 1674 // Put all the "Case 3" tests here. 1675 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call); 1676 1677 // Found a null. Keep searching for a matching receiver, 1678 // but remember that this is an empty (unused) slot. 1679 bind(found_null); 1680 } 1681 } 1682 1683 // In the fall-through case, we found no matching receiver, but we 1684 // observed the receiver[start_row] is NULL. 1685 1686 // Fill in the receiver field and increment the count. 1687 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1688 set_mdp_data_at(recvr_offset, receiver); 1689 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1690 li(scratch1, DataLayout::counter_increment); 1691 set_mdp_data_at(count_offset, scratch1); 1692 if (start_row > 0) { 1693 b(done); 1694 } 1695 } 1696 1697 // Argument and return type profilig. 1698 // kills: tmp, tmp2, R0, CR0, CR1 1699 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base, 1700 RegisterOrConstant mdo_addr_offs, 1701 Register tmp, Register tmp2) { 1702 Label do_nothing, do_update; 1703 1704 // tmp2 = obj is allowed 1705 assert_different_registers(obj, mdo_addr_base, tmp, R0); 1706 assert_different_registers(tmp2, mdo_addr_base, tmp, R0); 1707 const Register klass = tmp2; 1708 1709 verify_oop(obj); 1710 1711 ld(tmp, mdo_addr_offs, mdo_addr_base); 1712 1713 // Set null_seen if obj is 0. 1714 cmpdi(CCR0, obj, 0); 1715 ori(R0, tmp, TypeEntries::null_seen); 1716 beq(CCR0, do_update); 1717 1718 load_klass(klass, obj); 1719 1720 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask)); 1721 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask); 1722 cmpd(CCR1, R0, klass); 1723 // Klass seen before, nothing to do (regardless of unknown bit). 1724 //beq(CCR1, do_nothing); 1725 1726 andi_(R0, klass, TypeEntries::type_unknown); 1727 // Already unknown. Nothing to do anymore. 1728 //bne(CCR0, do_nothing); 1729 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne 1730 beq(CCR0, do_nothing); 1731 1732 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask)); 1733 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1734 beq(CCR0, do_update); // First time here. Set profile type. 1735 1736 // Different than before. Cannot keep accurate profile. 1737 ori(R0, tmp, TypeEntries::type_unknown); 1738 1739 bind(do_update); 1740 // update profile 1741 std(R0, mdo_addr_offs, mdo_addr_base); 1742 1743 align(32, 12); 1744 bind(do_nothing); 1745 } 1746 1747 void InterpreterMacroAssembler::profile_arguments_type(Register callee, 1748 Register tmp1, Register tmp2, 1749 bool is_virtual) { 1750 if (!ProfileInterpreter) { 1751 return; 1752 } 1753 1754 assert_different_registers(callee, tmp1, tmp2, R28_mdx); 1755 1756 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1757 Label profile_continue; 1758 1759 test_method_data_pointer(profile_continue); 1760 1761 int off_to_start = is_virtual ? 1762 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1763 1764 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx); 1765 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1766 bne(CCR0, profile_continue); 1767 1768 if (MethodData::profile_arguments()) { 1769 Label done; 1770 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1771 add(R28_mdx, off_to_args, R28_mdx); 1772 1773 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1774 if (i > 0 || MethodData::profile_return()) { 1775 // If return value type is profiled we may have no argument to profile. 1776 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1777 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count()); 1778 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count()); 1779 blt(CCR0, done); 1780 } 1781 ld(tmp1, in_bytes(Method::const_offset()), callee); 1782 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1); 1783 // Stack offset o (zero based) from the start of the argument 1784 // list, for n arguments translates into offset n - o - 1 from 1785 // the end of the argument list. But there's an extra slot at 1786 // the top of the stack. So the offset is n - o from Lesp. 1787 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx); 1788 subf(tmp1, tmp2, tmp1); 1789 1790 sldi(tmp1, tmp1, Interpreter::logStackElementSize); 1791 ldx(tmp1, tmp1, R15_esp); 1792 1793 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1); 1794 1795 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1796 addi(R28_mdx, R28_mdx, to_add); 1797 off_to_args += to_add; 1798 } 1799 1800 if (MethodData::profile_return()) { 1801 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1802 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1803 } 1804 1805 bind(done); 1806 1807 if (MethodData::profile_return()) { 1808 // We're right after the type profile for the last 1809 // argument. tmp1 is the number of cells left in the 1810 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1811 // if there's a return to profile. 1812 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), 1813 "can't move past ret type"); 1814 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size)); 1815 add(R28_mdx, tmp1, R28_mdx); 1816 } 1817 } else { 1818 assert(MethodData::profile_return(), "either profile call args or call ret"); 1819 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 1820 } 1821 1822 // Mdp points right after the end of the 1823 // CallTypeData/VirtualCallTypeData, right after the cells for the 1824 // return value type if there's one. 1825 align(32, 12); 1826 bind(profile_continue); 1827 } 1828 } 1829 1830 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 1831 assert_different_registers(ret, tmp1, tmp2); 1832 if (ProfileInterpreter && MethodData::profile_return()) { 1833 Label profile_continue; 1834 1835 test_method_data_pointer(profile_continue); 1836 1837 if (MethodData::profile_return_jsr292_only()) { 1838 // If we don't profile all invoke bytecodes we must make sure 1839 // it's a bytecode we indeed profile. We can't go back to the 1840 // begining of the ProfileData we intend to update to check its 1841 // type because we're right after it and we don't known its 1842 // length. 1843 lbz(tmp1, 0, R14_bcp); 1844 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method); 1845 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic); 1846 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle); 1847 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1848 cmpwi(CCR1, tmp2, static_cast<int>(vmIntrinsics::_compiledLambdaForm)); 1849 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1850 bne(CCR0, profile_continue); 1851 } 1852 1853 profile_obj_type(ret, R28_mdx, -in_bytes(SingleTypeEntry::size()), tmp1, tmp2); 1854 1855 align(32, 12); 1856 bind(profile_continue); 1857 } 1858 } 1859 1860 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, 1861 Register tmp3, Register tmp4) { 1862 if (ProfileInterpreter && MethodData::profile_parameters()) { 1863 Label profile_continue, done; 1864 1865 test_method_data_pointer(profile_continue); 1866 1867 // Load the offset of the area within the MDO used for 1868 // parameters. If it's negative we're not profiling any parameters. 1869 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx); 1870 cmpwi(CCR0, tmp1, 0); 1871 blt(CCR0, profile_continue); 1872 1873 // Compute a pointer to the area for parameters from the offset 1874 // and move the pointer to the slot for the last 1875 // parameters. Collect profiling from last parameter down. 1876 // mdo start + parameters offset + array length - 1 1877 1878 // Pointer to the parameter area in the MDO. 1879 const Register mdp = tmp1; 1880 add(mdp, tmp1, R28_mdx); 1881 1882 // Offset of the current profile entry to update. 1883 const Register entry_offset = tmp2; 1884 // entry_offset = array len in number of cells 1885 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp); 1886 1887 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1888 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 1889 1890 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 1891 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size)); 1892 // entry_offset in bytes 1893 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1894 1895 Label loop; 1896 align(32, 12); 1897 bind(loop); 1898 1899 // Load offset on the stack from the slot for this parameter. 1900 ld(tmp3, entry_offset, mdp); 1901 sldi(tmp3, tmp3, Interpreter::logStackElementSize); 1902 neg(tmp3, tmp3); 1903 // Read the parameter from the local area. 1904 ldx(tmp3, tmp3, R18_locals); 1905 1906 // Make entry_offset now point to the type field for this parameter. 1907 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1908 assert(type_base > off_base, "unexpected"); 1909 addi(entry_offset, entry_offset, type_base - off_base); 1910 1911 // Profile the parameter. 1912 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3); 1913 1914 // Go to next parameter. 1915 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base); 1916 cmpdi(CCR0, entry_offset, off_base + delta); 1917 addi(entry_offset, entry_offset, -delta); 1918 bge(CCR0, loop); 1919 1920 align(32, 12); 1921 bind(profile_continue); 1922 } 1923 } 1924 1925 // Add a InterpMonitorElem to stack (see frame_sparc.hpp). 1926 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) { 1927 1928 // Very-local scratch registers. 1929 const Register esp = Rtemp1; 1930 const Register slot = Rtemp2; 1931 1932 // Extracted monitor_size. 1933 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes(); 1934 assert(Assembler::is_aligned((unsigned int)monitor_size, 1935 (unsigned int)frame::alignment_in_bytes), 1936 "size of a monitor must respect alignment of SP"); 1937 1938 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor 1939 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp 1940 1941 // Shuffle expression stack down. Recall that stack_base points 1942 // just above the new expression stack bottom. Old_tos and new_tos 1943 // are used to scan thru the old and new expression stacks. 1944 if (!stack_is_empty) { 1945 Label copy_slot, copy_slot_finished; 1946 const Register n_slots = slot; 1947 1948 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack). 1949 subf(n_slots, esp, R26_monitor); 1950 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy. 1951 assert(LogBytesPerWord == 3, "conflicts assembler instructions"); 1952 beq(CCR0, copy_slot_finished); // Nothing to copy. 1953 1954 mtctr(n_slots); 1955 1956 // loop 1957 bind(copy_slot); 1958 ld(slot, 0, esp); // Move expression stack down. 1959 std(slot, -monitor_size, esp); // distance = monitor_size 1960 addi(esp, esp, BytesPerWord); 1961 bdnz(copy_slot); 1962 1963 bind(copy_slot_finished); 1964 } 1965 1966 addi(R15_esp, R15_esp, -monitor_size); 1967 addi(R26_monitor, R26_monitor, -monitor_size); 1968 1969 // Restart interpreter 1970 } 1971 1972 // ============================================================================ 1973 // Java locals access 1974 1975 // Load a local variable at index in Rindex into register Rdst_value. 1976 // Also puts address of local into Rdst_address as a service. 1977 // Kills: 1978 // - Rdst_value 1979 // - Rdst_address 1980 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) { 1981 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 1982 subf(Rdst_address, Rdst_address, R18_locals); 1983 lwz(Rdst_value, 0, Rdst_address); 1984 } 1985 1986 // Load a local variable at index in Rindex into register Rdst_value. 1987 // Also puts address of local into Rdst_address as a service. 1988 // Kills: 1989 // - Rdst_value 1990 // - Rdst_address 1991 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) { 1992 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 1993 subf(Rdst_address, Rdst_address, R18_locals); 1994 ld(Rdst_value, -8, Rdst_address); 1995 } 1996 1997 // Load a local variable at index in Rindex into register Rdst_value. 1998 // Also puts address of local into Rdst_address as a service. 1999 // Input: 2000 // - Rindex: slot nr of local variable 2001 // Kills: 2002 // - Rdst_value 2003 // - Rdst_address 2004 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, 2005 Register Rdst_address, 2006 Register Rindex) { 2007 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2008 subf(Rdst_address, Rdst_address, R18_locals); 2009 ld(Rdst_value, 0, Rdst_address); 2010 } 2011 2012 // Load a local variable at index in Rindex into register Rdst_value. 2013 // Also puts address of local into Rdst_address as a service. 2014 // Kills: 2015 // - Rdst_value 2016 // - Rdst_address 2017 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, 2018 Register Rdst_address, 2019 Register Rindex) { 2020 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2021 subf(Rdst_address, Rdst_address, R18_locals); 2022 lfs(Rdst_value, 0, Rdst_address); 2023 } 2024 2025 // Load a local variable at index in Rindex into register Rdst_value. 2026 // Also puts address of local into Rdst_address as a service. 2027 // Kills: 2028 // - Rdst_value 2029 // - Rdst_address 2030 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, 2031 Register Rdst_address, 2032 Register Rindex) { 2033 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2034 subf(Rdst_address, Rdst_address, R18_locals); 2035 lfd(Rdst_value, -8, Rdst_address); 2036 } 2037 2038 // Store an int value at local variable slot Rindex. 2039 // Kills: 2040 // - Rindex 2041 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) { 2042 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2043 subf(Rindex, Rindex, R18_locals); 2044 stw(Rvalue, 0, Rindex); 2045 } 2046 2047 // Store a long value at local variable slot Rindex. 2048 // Kills: 2049 // - Rindex 2050 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) { 2051 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2052 subf(Rindex, Rindex, R18_locals); 2053 std(Rvalue, -8, Rindex); 2054 } 2055 2056 // Store an oop value at local variable slot Rindex. 2057 // Kills: 2058 // - Rindex 2059 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) { 2060 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2061 subf(Rindex, Rindex, R18_locals); 2062 std(Rvalue, 0, Rindex); 2063 } 2064 2065 // Store an int value at local variable slot Rindex. 2066 // Kills: 2067 // - Rindex 2068 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) { 2069 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2070 subf(Rindex, Rindex, R18_locals); 2071 stfs(Rvalue, 0, Rindex); 2072 } 2073 2074 // Store an int value at local variable slot Rindex. 2075 // Kills: 2076 // - Rindex 2077 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) { 2078 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2079 subf(Rindex, Rindex, R18_locals); 2080 stfd(Rvalue, -8, Rindex); 2081 } 2082 2083 // Read pending exception from thread and jump to interpreter. 2084 // Throw exception entry if one if pending. Fall through otherwise. 2085 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) { 2086 assert_different_registers(Rscratch1, Rscratch2, R3); 2087 Register Rexception = Rscratch1; 2088 Register Rtmp = Rscratch2; 2089 Label Ldone; 2090 // Get pending exception oop. 2091 ld(Rexception, thread_(pending_exception)); 2092 cmpdi(CCR0, Rexception, 0); 2093 beq(CCR0, Ldone); 2094 li(Rtmp, 0); 2095 mr_if_needed(R3, Rexception); 2096 std(Rtmp, thread_(pending_exception)); // Clear exception in thread 2097 if (Interpreter::rethrow_exception_entry() != NULL) { 2098 // Already got entry address. 2099 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry()); 2100 } else { 2101 // Dynamically load entry address. 2102 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true); 2103 ld(Rtmp, simm16_rest, Rtmp); 2104 } 2105 mtctr(Rtmp); 2106 save_interpreter_state(Rtmp); 2107 bctr(); 2108 2109 align(32, 12); 2110 bind(Ldone); 2111 } 2112 2113 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) { 2114 save_interpreter_state(R11_scratch1); 2115 2116 MacroAssembler::call_VM(oop_result, entry_point, false); 2117 2118 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2119 2120 check_and_handle_popframe(R11_scratch1); 2121 check_and_handle_earlyret(R11_scratch1); 2122 // Now check exceptions manually. 2123 if (check_exceptions) { 2124 check_and_forward_exception(R11_scratch1, R12_scratch2); 2125 } 2126 } 2127 2128 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2129 Register arg_1, bool check_exceptions) { 2130 // ARG1 is reserved for the thread. 2131 mr_if_needed(R4_ARG2, arg_1); 2132 call_VM(oop_result, entry_point, check_exceptions); 2133 } 2134 2135 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2136 Register arg_1, Register arg_2, 2137 bool check_exceptions) { 2138 // ARG1 is reserved for the thread. 2139 mr_if_needed(R4_ARG2, arg_1); 2140 assert(arg_2 != R4_ARG2, "smashed argument"); 2141 mr_if_needed(R5_ARG3, arg_2); 2142 call_VM(oop_result, entry_point, check_exceptions); 2143 } 2144 2145 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2146 Register arg_1, Register arg_2, Register arg_3, 2147 bool check_exceptions) { 2148 // ARG1 is reserved for the thread. 2149 mr_if_needed(R4_ARG2, arg_1); 2150 assert(arg_2 != R4_ARG2, "smashed argument"); 2151 mr_if_needed(R5_ARG3, arg_2); 2152 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument"); 2153 mr_if_needed(R6_ARG4, arg_3); 2154 call_VM(oop_result, entry_point, check_exceptions); 2155 } 2156 2157 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) { 2158 ld(scratch, 0, R1_SP); 2159 std(R15_esp, _ijava_state_neg(esp), scratch); 2160 std(R14_bcp, _ijava_state_neg(bcp), scratch); 2161 std(R26_monitor, _ijava_state_neg(monitors), scratch); 2162 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); } 2163 // Other entries should be unchanged. 2164 } 2165 2166 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) { 2167 ld(scratch, 0, R1_SP); 2168 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception). 2169 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code. 2170 if (!bcp_and_mdx_only) { 2171 // Following ones are Metadata. 2172 ld(R19_method, _ijava_state_neg(method), scratch); 2173 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch); 2174 // Following ones are stack addresses and don't require reload. 2175 ld(R15_esp, _ijava_state_neg(esp), scratch); 2176 ld(R18_locals, _ijava_state_neg(locals), scratch); 2177 ld(R26_monitor, _ijava_state_neg(monitors), scratch); 2178 } 2179 #ifdef ASSERT 2180 { 2181 Label Lok; 2182 subf(R0, R1_SP, scratch); 2183 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size); 2184 bge(CCR0, Lok); 2185 stop("frame too small (restore istate)"); 2186 bind(Lok); 2187 } 2188 #endif 2189 } 2190 2191 void InterpreterMacroAssembler::get_method_counters(Register method, 2192 Register Rcounters, 2193 Label& skip) { 2194 BLOCK_COMMENT("Load and ev. allocate counter object {"); 2195 Label has_counters; 2196 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2197 cmpdi(CCR0, Rcounters, 0); 2198 bne(CCR0, has_counters); 2199 call_VM(noreg, CAST_FROM_FN_PTR(address, 2200 InterpreterRuntime::build_method_counters), method); 2201 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2202 cmpdi(CCR0, Rcounters, 0); 2203 beq(CCR0, skip); // No MethodCounters, OutOfMemory. 2204 BLOCK_COMMENT("} Load and ev. allocate counter object"); 2205 2206 bind(has_counters); 2207 } 2208 2209 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, 2210 Register iv_be_count, 2211 Register Rtmp_r0) { 2212 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 2213 Register invocation_count = iv_be_count; 2214 Register backedge_count = Rtmp_r0; 2215 int delta = InvocationCounter::count_increment; 2216 2217 // Load each counter in a register. 2218 // ld(inv_counter, Rtmp); 2219 // ld(be_counter, Rtmp2); 2220 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + 2221 InvocationCounter::counter_offset()); 2222 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + 2223 InvocationCounter::counter_offset()); 2224 2225 BLOCK_COMMENT("Increment profiling counters {"); 2226 2227 // Load the backedge counter. 2228 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int 2229 // Mask the backedge counter. 2230 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value); 2231 2232 // Load the invocation counter. 2233 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int 2234 // Add the delta to the invocation counter and store the result. 2235 addi(invocation_count, invocation_count, delta); 2236 // Store value. 2237 stw(invocation_count, inv_counter_offset, Rcounters); 2238 2239 // Add invocation counter + backedge counter. 2240 add(iv_be_count, backedge_count, invocation_count); 2241 2242 // Note that this macro must leave the backedge_count + invocation_count in 2243 // register iv_be_count! 2244 BLOCK_COMMENT("} Increment profiling counters"); 2245 } 2246 2247 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2248 if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); } 2249 } 2250 2251 // Local helper function for the verify_oop_or_return_address macro. 2252 static bool verify_return_address(Method* m, int bci) { 2253 #ifndef PRODUCT 2254 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci; 2255 // Assume it is a valid return address if it is inside m and is preceded by a jsr. 2256 if (!m->contains(pc)) return false; 2257 address jsr_pc; 2258 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2259 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2260 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2261 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2262 #endif // PRODUCT 2263 return false; 2264 } 2265 2266 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2267 if (VerifyFPU) { 2268 unimplemented("verfiyFPU"); 2269 } 2270 } 2271 2272 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2273 if (!VerifyOops) return; 2274 2275 // The VM documentation for the astore[_wide] bytecode allows 2276 // the TOS to be not only an oop but also a return address. 2277 Label test; 2278 Label skip; 2279 // See if it is an address (in the current method): 2280 2281 const int log2_bytecode_size_limit = 16; 2282 srdi_(Rtmp, reg, log2_bytecode_size_limit); 2283 bne(CCR0, test); 2284 2285 address fd = CAST_FROM_FN_PTR(address, verify_return_address); 2286 const int nbytes_save = MacroAssembler::num_volatile_regs * 8; 2287 save_volatile_gprs(R1_SP, -nbytes_save); // except R0 2288 save_LR_CR(Rtmp); // Save in old frame. 2289 push_frame_reg_args(nbytes_save, Rtmp); 2290 2291 load_const_optimized(Rtmp, fd, R0); 2292 mr_if_needed(R4_ARG2, reg); 2293 mr(R3_ARG1, R19_method); 2294 call_c(Rtmp); // call C 2295 2296 pop_frame(); 2297 restore_LR_CR(Rtmp); 2298 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0 2299 b(skip); 2300 2301 // Perform a more elaborate out-of-line call. 2302 // Not an address; verify it: 2303 bind(test); 2304 verify_oop(reg); 2305 bind(skip); 2306 } 2307 2308 // Inline assembly for: 2309 // 2310 // if (thread is in interp_only_mode) { 2311 // InterpreterRuntime::post_method_entry(); 2312 // } 2313 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) || 2314 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) { 2315 // SharedRuntime::jvmpi_method_entry(method, receiver); 2316 // } 2317 void InterpreterMacroAssembler::notify_method_entry() { 2318 // JVMTI 2319 // Whenever JVMTI puts a thread in interp_only_mode, method 2320 // entry/exit events are sent for that thread to track stack 2321 // depth. If it is possible to enter interp_only_mode we add 2322 // the code to check if the event should be sent. 2323 if (JvmtiExport::can_post_interpreter_events()) { 2324 Label jvmti_post_done; 2325 2326 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2327 cmpwi(CCR0, R0, 0); 2328 beq(CCR0, jvmti_post_done); 2329 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); 2330 2331 bind(jvmti_post_done); 2332 } 2333 } 2334 2335 // Inline assembly for: 2336 // 2337 // if (thread is in interp_only_mode) { 2338 // // save result 2339 // InterpreterRuntime::post_method_exit(); 2340 // // restore result 2341 // } 2342 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) { 2343 // // save result 2344 // SharedRuntime::jvmpi_method_exit(); 2345 // // restore result 2346 // } 2347 // 2348 // Native methods have their result stored in d_tmp and l_tmp. 2349 // Java methods have their result stored in the expression stack. 2350 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state, 2351 NotifyMethodExitMode mode, bool check_exceptions) { 2352 // JVMTI 2353 // Whenever JVMTI puts a thread in interp_only_mode, method 2354 // entry/exit events are sent for that thread to track stack 2355 // depth. If it is possible to enter interp_only_mode we add 2356 // the code to check if the event should be sent. 2357 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2358 Label jvmti_post_done; 2359 2360 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2361 cmpwi(CCR0, R0, 0); 2362 beq(CCR0, jvmti_post_done); 2363 if (!is_native_method) { push(state); } // Expose tos to GC. 2364 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), check_exceptions); 2365 if (!is_native_method) { pop(state); } 2366 2367 align(32, 12); 2368 bind(jvmti_post_done); 2369 } 2370 2371 // Dtrace support not implemented. 2372 }