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.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 40 void 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 45 void 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 56 void 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 67 void 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. 76 void 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. 89 void 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 95 void 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 } 134 135 void 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 } 157 158 void 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. 194 195 void 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 } 204 205 void 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 } 223 224 void 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 230 231 void InterpreterMacroAssembler::pop_i(Register r) { 232 lwzu(r, Interpreter::stackElementSize, R15_esp); 233 } 234 235 void InterpreterMacroAssembler::pop_ptr(Register r) { 236 ldu(r, Interpreter::stackElementSize, R15_esp); 237 } 238 239 void InterpreterMacroAssembler::pop_l(Register r) { 240 ld(r, Interpreter::stackElementSize, R15_esp); 241 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 242 } 243 244 void InterpreterMacroAssembler::pop_f(FloatRegister f) { 245 lfsu(f, Interpreter::stackElementSize, R15_esp); 246 } 247 248 void InterpreterMacroAssembler::pop_d(FloatRegister f) { 249 lfd(f, Interpreter::stackElementSize, R15_esp); 250 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 251 } 252 253 void InterpreterMacroAssembler::push_i(Register r) { 254 stw(r, 0, R15_esp); 255 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 256 } 257 258 void InterpreterMacroAssembler::push_ptr(Register r) { 259 std(r, 0, R15_esp); 260 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 261 } 262 263 void InterpreterMacroAssembler::push_l(Register r) { 264 // Clear unused slot. 265 load_const_optimized(R0, 0L); 266 std(R0, 0, R15_esp); 267 std(r, - Interpreter::stackElementSize, R15_esp); 268 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 269 } 270 271 void InterpreterMacroAssembler::push_f(FloatRegister f) { 272 stfs(f, 0, R15_esp); 273 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 274 } 275 276 void InterpreterMacroAssembler::push_d(FloatRegister f) { 277 stfd(f, - Interpreter::stackElementSize, R15_esp); 278 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 279 } 280 281 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) { 282 std(first, 0, R15_esp); 283 std(second, -Interpreter::stackElementSize, R15_esp); 284 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 285 } 286 287 void InterpreterMacroAssembler::push_l_pop_d(Register l, FloatRegister d) { 288 std(l, 0, R15_esp); 289 lfd(d, 0, R15_esp); 290 } 291 292 void InterpreterMacroAssembler::push_d_pop_l(FloatRegister d, Register l) { 293 stfd(d, 0, R15_esp); 294 ld(l, 0, R15_esp); 295 } 296 297 void InterpreterMacroAssembler::push(TosState state) { 298 switch (state) { 299 case atos: push_ptr(); break; 300 case btos: 301 case ztos: 302 case ctos: 303 case stos: 304 case itos: push_i(); break; 305 case ltos: push_l(); break; 306 case ftos: push_f(); break; 307 case dtos: push_d(); break; 308 case vtos: /* nothing to do */ break; 309 default : ShouldNotReachHere(); 310 } 311 } 312 313 void InterpreterMacroAssembler::pop(TosState state) { 314 switch (state) { 315 case atos: pop_ptr(); break; 316 case btos: 317 case ztos: 318 case ctos: 319 case stos: 320 case itos: pop_i(); break; 321 case ltos: pop_l(); break; 322 case ftos: pop_f(); break; 323 case dtos: pop_d(); break; 324 case vtos: /* nothing to do */ break; 325 default : ShouldNotReachHere(); 326 } 327 verify_oop(R17_tos, state); 328 } 329 330 void InterpreterMacroAssembler::empty_expression_stack() { 331 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 332 } 333 334 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset, 335 Register Rdst, 336 signedOrNot is_signed) { 337 #if defined(VM_LITTLE_ENDIAN) 338 if (bcp_offset) { 339 load_const_optimized(Rdst, bcp_offset); 340 lhbrx(Rdst, R14_bcp, Rdst); 341 } else { 342 lhbrx(Rdst, R14_bcp); 343 } 344 if (is_signed == Signed) { 345 extsh(Rdst, Rdst); 346 } 347 #else 348 // Read Java big endian format. 349 if (is_signed == Signed) { 350 lha(Rdst, bcp_offset, R14_bcp); 351 } else { 352 lhz(Rdst, bcp_offset, R14_bcp); 353 } 354 #endif 355 } 356 357 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset, 358 Register Rdst, 359 signedOrNot is_signed) { 360 #if defined(VM_LITTLE_ENDIAN) 361 if (bcp_offset) { 362 load_const_optimized(Rdst, bcp_offset); 363 lwbrx(Rdst, R14_bcp, Rdst); 364 } else { 365 lwbrx(Rdst, R14_bcp); 366 } 367 if (is_signed == Signed) { 368 extsw(Rdst, Rdst); 369 } 370 #else 371 // Read Java big endian format. 372 if (bcp_offset & 3) { // Offset unaligned? 373 load_const_optimized(Rdst, bcp_offset); 374 if (is_signed == Signed) { 375 lwax(Rdst, R14_bcp, Rdst); 376 } else { 377 lwzx(Rdst, R14_bcp, Rdst); 378 } 379 } else { 380 if (is_signed == Signed) { 381 lwa(Rdst, bcp_offset, R14_bcp); 382 } else { 383 lwz(Rdst, bcp_offset, R14_bcp); 384 } 385 } 386 #endif 387 } 388 389 390 // Load the constant pool cache index from the bytecode stream. 391 // 392 // Kills / writes: 393 // - Rdst, Rscratch 394 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, 395 size_t index_size) { 396 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 397 // Cache index is always in the native format, courtesy of Rewriter. 398 if (index_size == sizeof(u2)) { 399 lhz(Rdst, bcp_offset, R14_bcp); 400 } else if (index_size == sizeof(u4)) { 401 if (bcp_offset & 3) { 402 load_const_optimized(Rdst, bcp_offset); 403 lwax(Rdst, R14_bcp, Rdst); 404 } else { 405 lwa(Rdst, bcp_offset, R14_bcp); 406 } 407 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 408 nand(Rdst, Rdst, Rdst); // convert to plain index 409 } else if (index_size == sizeof(u1)) { 410 lbz(Rdst, bcp_offset, R14_bcp); 411 } else { 412 ShouldNotReachHere(); 413 } 414 // Rdst now contains cp cache index. 415 } 416 417 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset, 418 size_t index_size) { 419 get_cache_index_at_bcp(cache, bcp_offset, index_size); 420 sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord)); 421 add(cache, R27_constPoolCache, cache); 422 } 423 424 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format) 425 // from (Rsrc)+offset. 426 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset, 427 signedOrNot is_signed) { 428 #if defined(VM_LITTLE_ENDIAN) 429 if (offset) { 430 load_const_optimized(Rdst, offset); 431 lwbrx(Rdst, Rdst, Rsrc); 432 } else { 433 lwbrx(Rdst, Rsrc); 434 } 435 if (is_signed == Signed) { 436 extsw(Rdst, Rdst); 437 } 438 #else 439 if (is_signed == Signed) { 440 lwa(Rdst, offset, Rsrc); 441 } else { 442 lwz(Rdst, offset, Rsrc); 443 } 444 #endif 445 } 446 447 // Load object from cpool->resolved_references(index). 448 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index, Label *is_null) { 449 assert_different_registers(result, index); 450 get_constant_pool(result); 451 452 // Convert from field index to resolved_references() index and from 453 // word index to byte offset. Since this is a java object, it can be compressed. 454 Register tmp = index; // reuse 455 sldi(tmp, index, LogBytesPerHeapOop); 456 // Load pointer for resolved_references[] objArray. 457 ld(result, ConstantPool::resolved_references_offset_in_bytes(), result); 458 // JNIHandles::resolve(result) 459 ld(result, 0, result); 460 #ifdef ASSERT 461 Label index_ok; 462 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result); 463 sldi(R0, R0, LogBytesPerHeapOop); 464 cmpd(CCR0, tmp, R0); 465 blt(CCR0, index_ok); 466 stop("resolved reference index out of bounds", 0x09256); 467 bind(index_ok); 468 #endif 469 // Add in the index. 470 add(result, tmp, result); 471 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result, is_null); 472 } 473 474 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 475 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2. 476 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1, 477 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) { 478 // Profile the not-null value's klass. 479 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2); 480 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype); 481 profile_typecheck_failed(Rtmp1, Rtmp2); 482 } 483 484 // Separate these two to allow for delay slot in middle. 485 // These are used to do a test and full jump to exception-throwing code. 486 487 // Check that index is in range for array, then shift index by index_shift, 488 // and put arrayOop + shifted_index into res. 489 // Note: res is still shy of address by array offset into object. 490 491 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, 492 int index_shift, Register Rtmp, Register Rres) { 493 // Check that index is in range for array, then shift index by index_shift, 494 // and put arrayOop + shifted_index into res. 495 // Note: res is still shy of address by array offset into object. 496 // Kills: 497 // - Rindex 498 // Writes: 499 // - Rres: Address that corresponds to the array index if check was successful. 500 verify_oop(Rarray); 501 const Register Rlength = R0; 502 const Register RsxtIndex = Rtmp; 503 Label LisNull, LnotOOR; 504 505 // Array nullcheck 506 if (!ImplicitNullChecks) { 507 cmpdi(CCR0, Rarray, 0); 508 beq(CCR0, LisNull); 509 } else { 510 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex); 511 } 512 513 // Rindex might contain garbage in upper bits (remember that we don't sign extend 514 // during integer arithmetic operations). So kill them and put value into same register 515 // where ArrayIndexOutOfBounds would expect the index in. 516 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit 517 518 // Index check 519 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray); 520 cmplw(CCR0, Rindex, Rlength); 521 sldi(RsxtIndex, RsxtIndex, index_shift); 522 blt(CCR0, LnotOOR); 523 // Index should be in R17_tos, array should be in R4_ARG2. 524 mr_if_needed(R17_tos, Rindex); 525 mr_if_needed(R4_ARG2, Rarray); 526 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry); 527 mtctr(Rtmp); 528 bctr(); 529 530 if (!ImplicitNullChecks) { 531 bind(LisNull); 532 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry); 533 mtctr(Rtmp); 534 bctr(); 535 } 536 537 align(32, 16); 538 bind(LnotOOR); 539 540 // Calc address 541 add(Rres, RsxtIndex, Rarray); 542 } 543 544 void InterpreterMacroAssembler::index_check(Register array, Register index, 545 int index_shift, Register tmp, Register res) { 546 // pop array 547 pop_ptr(array); 548 549 // check array 550 index_check_without_pop(array, index, index_shift, tmp, res); 551 } 552 553 void InterpreterMacroAssembler::get_const(Register Rdst) { 554 ld(Rdst, in_bytes(Method::const_offset()), R19_method); 555 } 556 557 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 558 get_const(Rdst); 559 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 560 } 561 562 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 563 get_constant_pool(Rdst); 564 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 565 } 566 567 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 568 get_constant_pool(Rcpool); 569 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool); 570 } 571 572 // Unlock if synchronized method. 573 // 574 // Unlock the receiver if this is a synchronized method. 575 // Unlock any Java monitors from synchronized blocks. 576 // 577 // If there are locked Java monitors 578 // If throw_monitor_exception 579 // throws IllegalMonitorStateException 580 // Else if install_monitor_exception 581 // installs IllegalMonitorStateException 582 // Else 583 // no error processing 584 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 585 bool throw_monitor_exception, 586 bool install_monitor_exception) { 587 Label Lunlocked, Lno_unlock; 588 { 589 Register Rdo_not_unlock_flag = R11_scratch1; 590 Register Raccess_flags = R12_scratch2; 591 592 // Check if synchronized method or unlocking prevented by 593 // JavaThread::do_not_unlock_if_synchronized flag. 594 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 595 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method); 596 li(R0, 0); 597 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag 598 599 push(state); 600 601 // Skip if we don't have to unlock. 602 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0. 603 beq(CCR0, Lunlocked); 604 605 cmpwi(CCR0, Rdo_not_unlock_flag, 0); 606 bne(CCR0, Lno_unlock); 607 } 608 609 // Unlock 610 { 611 Register Rmonitor_base = R11_scratch1; 612 613 Label Lunlock; 614 // If it's still locked, everything is ok, unlock it. 615 ld(Rmonitor_base, 0, R1_SP); 616 addi(Rmonitor_base, Rmonitor_base, 617 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base 618 619 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base); 620 cmpdi(CCR0, R0, 0); 621 bne(CCR0, Lunlock); 622 623 // If it's already unlocked, throw exception. 624 if (throw_monitor_exception) { 625 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 626 should_not_reach_here(); 627 } else { 628 if (install_monitor_exception) { 629 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 630 b(Lunlocked); 631 } 632 } 633 634 bind(Lunlock); 635 unlock_object(Rmonitor_base); 636 } 637 638 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not. 639 bind(Lunlocked); 640 { 641 Label Lexception, Lrestart; 642 Register Rcurrent_obj_addr = R11_scratch1; 643 const int delta = frame::interpreter_frame_monitor_size_in_bytes(); 644 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords"); 645 646 bind(Lrestart); 647 // Set up search loop: Calc num of iterations. 648 { 649 Register Riterations = R12_scratch2; 650 Register Rmonitor_base = Rcurrent_obj_addr; 651 ld(Rmonitor_base, 0, R1_SP); 652 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base 653 654 subf_(Riterations, R26_monitor, Rmonitor_base); 655 ble(CCR0, Lno_unlock); 656 657 addi(Rcurrent_obj_addr, Rmonitor_base, 658 BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes()); 659 // Check if any monitor is on stack, bail out if not 660 srdi(Riterations, Riterations, exact_log2(delta)); 661 mtctr(Riterations); 662 } 663 664 // The search loop: Look for locked monitors. 665 { 666 const Register Rcurrent_obj = R0; 667 Label Lloop; 668 669 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 670 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 671 bind(Lloop); 672 673 // Check if current entry is used. 674 cmpdi(CCR0, Rcurrent_obj, 0); 675 bne(CCR0, Lexception); 676 // Preload next iteration's compare value. 677 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 678 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 679 bdnz(Lloop); 680 } 681 // Fell through: Everything's unlocked => finish. 682 b(Lno_unlock); 683 684 // An object is still locked => need to throw exception. 685 bind(Lexception); 686 if (throw_monitor_exception) { 687 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 688 should_not_reach_here(); 689 } else { 690 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 691 // Unlock does not block, so don't have to worry about the frame. 692 Register Rmonitor_addr = R11_scratch1; 693 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta); 694 unlock_object(Rmonitor_addr); 695 if (install_monitor_exception) { 696 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 697 } 698 b(Lrestart); 699 } 700 } 701 702 align(32, 12); 703 bind(Lno_unlock); 704 pop(state); 705 } 706 707 // Support function for remove_activation & Co. 708 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, 709 Register Rscratch1, Register Rscratch2) { 710 // Pop interpreter frame. 711 ld(Rscratch1, 0, R1_SP); // *SP 712 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp 713 ld(Rscratch2, 0, Rscratch1); // **SP 714 #ifdef ASSERT 715 { 716 Label Lok; 717 ld(R0, _ijava_state_neg(ijava_reserved), Rscratch1); 718 cmpdi(CCR0, R0, 0x5afe); 719 beq(CCR0, Lok); 720 stop("frame corrupted (remove activation)", 0x5afe); 721 bind(Lok); 722 } 723 #endif 724 if (return_pc!=noreg) { 725 ld(return_pc, _abi(lr), Rscratch1); // LR 726 } 727 728 // Merge top frames. 729 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP 730 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP 731 } 732 733 void InterpreterMacroAssembler::narrow(Register result) { 734 Register ret_type = R11_scratch1; 735 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method); 736 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1); 737 738 Label notBool, notByte, notChar, done; 739 740 // common case first 741 cmpwi(CCR0, ret_type, T_INT); 742 beq(CCR0, done); 743 744 cmpwi(CCR0, ret_type, T_BOOLEAN); 745 bne(CCR0, notBool); 746 andi(result, result, 0x1); 747 b(done); 748 749 bind(notBool); 750 cmpwi(CCR0, ret_type, T_BYTE); 751 bne(CCR0, notByte); 752 extsb(result, result); 753 b(done); 754 755 bind(notByte); 756 cmpwi(CCR0, ret_type, T_CHAR); 757 bne(CCR0, notChar); 758 andi(result, result, 0xffff); 759 b(done); 760 761 bind(notChar); 762 // cmpwi(CCR0, ret_type, T_SHORT); // all that's left 763 // bne(CCR0, done); 764 extsh(result, result); 765 766 // Nothing to do for T_INT 767 bind(done); 768 } 769 770 // Remove activation. 771 // 772 // Unlock the receiver if this is a synchronized method. 773 // Unlock any Java monitors from synchronized blocks. 774 // Remove the activation from the stack. 775 // 776 // If there are locked Java monitors 777 // If throw_monitor_exception 778 // throws IllegalMonitorStateException 779 // Else if install_monitor_exception 780 // installs IllegalMonitorStateException 781 // Else 782 // no error processing 783 void InterpreterMacroAssembler::remove_activation(TosState state, 784 bool throw_monitor_exception, 785 bool install_monitor_exception) { 786 BLOCK_COMMENT("remove_activation {"); 787 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 788 789 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 790 notify_method_exit(false, state, NotifyJVMTI, true); 791 792 BLOCK_COMMENT("reserved_stack_check:"); 793 if (StackReservedPages > 0) { 794 // Test if reserved zone needs to be enabled. 795 Label no_reserved_zone_enabling; 796 797 // Compare frame pointers. There is no good stack pointer, as with stack 798 // frame compression we can get different SPs when we do calls. A subsequent 799 // call could have a smaller SP, so that this compare succeeds for an 800 // inner call of the method annotated with ReservedStack. 801 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread); 802 ld_ptr(R11_scratch1, _abi(callers_sp), R1_SP); // Load frame pointer. 803 cmpld(CCR0, R11_scratch1, R0); 804 blt_predict_taken(CCR0, no_reserved_zone_enabling); 805 806 // Enable reserved zone again, throw stack overflow exception. 807 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread); 808 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError)); 809 810 should_not_reach_here(); 811 812 bind(no_reserved_zone_enabling); 813 } 814 815 verify_oop(R17_tos, state); 816 verify_thread(); 817 818 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 819 mtlr(R0); 820 BLOCK_COMMENT("} remove_activation"); 821 } 822 823 // Lock object 824 // 825 // Registers alive 826 // monitor - Address of the BasicObjectLock to be used for locking, 827 // which must be initialized with the object to lock. 828 // object - Address of the object to be locked. 829 // 830 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 831 if (UseHeavyMonitors) { 832 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 833 monitor, /*check_for_exceptions=*/true); 834 } else { 835 // template code: 836 // 837 // markOop displaced_header = obj->mark().set_unlocked(); 838 // monitor->lock()->set_displaced_header(displaced_header); 839 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 840 // // We stored the monitor address into the object's mark word. 841 // } else if (THREAD->is_lock_owned((address)displaced_header)) 842 // // Simple recursive case. 843 // monitor->lock()->set_displaced_header(NULL); 844 // } else { 845 // // Slow path. 846 // InterpreterRuntime::monitorenter(THREAD, monitor); 847 // } 848 849 const Register displaced_header = R7_ARG5; 850 const Register object_mark_addr = R8_ARG6; 851 const Register current_header = R9_ARG7; 852 const Register tmp = R10_ARG8; 853 854 Label done; 855 Label cas_failed, slow_case; 856 857 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp); 858 859 // markOop displaced_header = obj->mark().set_unlocked(); 860 861 // Load markOop from object into displaced_header. 862 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object); 863 864 if (UseBiasedLocking) { 865 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case); 866 } 867 868 // Set displaced_header to be (markOop of object | UNLOCK_VALUE). 869 ori(displaced_header, displaced_header, markOopDesc::unlocked_value); 870 871 // monitor->lock()->set_displaced_header(displaced_header); 872 873 // Initialize the box (Must happen before we update the object mark!). 874 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 875 BasicLock::displaced_header_offset_in_bytes(), monitor); 876 877 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 878 879 // Store stack address of the BasicObjectLock (this is monitor) into object. 880 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 881 882 // Must fence, otherwise, preceding store(s) may float below cmpxchg. 883 // CmpxchgX sets CCR0 to cmpX(current, displaced). 884 cmpxchgd(/*flag=*/CCR0, 885 /*current_value=*/current_header, 886 /*compare_value=*/displaced_header, /*exchange_value=*/monitor, 887 /*where=*/object_mark_addr, 888 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq, 889 MacroAssembler::cmpxchgx_hint_acquire_lock(), 890 noreg, 891 &cas_failed, 892 /*check without membar and ldarx first*/true); 893 894 // If the compare-and-exchange succeeded, then we found an unlocked 895 // object and we have now locked it. 896 b(done); 897 bind(cas_failed); 898 899 // } else if (THREAD->is_lock_owned((address)displaced_header)) 900 // // Simple recursive case. 901 // monitor->lock()->set_displaced_header(NULL); 902 903 // We did not see an unlocked object so try the fast recursive case. 904 905 // Check if owner is self by comparing the value in the markOop of object 906 // (current_header) with the stack pointer. 907 sub(current_header, current_header, R1_SP); 908 909 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 910 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place); 911 912 and_(R0/*==0?*/, current_header, tmp); 913 // If condition is true we are done and hence we can store 0 in the displaced 914 // header indicating it is a recursive lock. 915 bne(CCR0, slow_case); 916 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() + 917 BasicLock::displaced_header_offset_in_bytes(), monitor); 918 b(done); 919 920 // } else { 921 // // Slow path. 922 // InterpreterRuntime::monitorenter(THREAD, monitor); 923 924 // None of the above fast optimizations worked so we have to get into the 925 // slow case of monitor enter. 926 bind(slow_case); 927 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 928 monitor, /*check_for_exceptions=*/true); 929 // } 930 align(32, 12); 931 bind(done); 932 } 933 } 934 935 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 936 // 937 // Registers alive 938 // monitor - Address of the BasicObjectLock to be used for locking, 939 // which must be initialized with the object to lock. 940 // 941 // Throw IllegalMonitorException if object is not locked by current thread. 942 void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) { 943 if (UseHeavyMonitors) { 944 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 945 monitor, check_for_exceptions); 946 } else { 947 948 // template code: 949 // 950 // if ((displaced_header = monitor->displaced_header()) == NULL) { 951 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 952 // monitor->set_obj(NULL); 953 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 954 // // We swapped the unlocked mark in displaced_header into the object's mark word. 955 // monitor->set_obj(NULL); 956 // } else { 957 // // Slow path. 958 // InterpreterRuntime::monitorexit(THREAD, monitor); 959 // } 960 961 const Register object = R7_ARG5; 962 const Register displaced_header = R8_ARG6; 963 const Register object_mark_addr = R9_ARG7; 964 const Register current_header = R10_ARG8; 965 966 Label free_slot; 967 Label slow_case; 968 969 assert_different_registers(object, displaced_header, object_mark_addr, current_header); 970 971 if (UseBiasedLocking) { 972 // The object address from the monitor is in object. 973 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); 974 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0"); 975 biased_locking_exit(CCR0, object, displaced_header, free_slot); 976 } 977 978 // Test first if we are in the fast recursive case. 979 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 980 BasicLock::displaced_header_offset_in_bytes(), monitor); 981 982 // If the displaced header is zero, we have a recursive unlock. 983 cmpdi(CCR0, displaced_header, 0); 984 beq(CCR0, free_slot); // recursive unlock 985 986 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 987 // // We swapped the unlocked mark in displaced_header into the object's mark word. 988 // monitor->set_obj(NULL); 989 990 // If we still have a lightweight lock, unlock the object and be done. 991 992 // The object address from the monitor is in object. 993 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); } 994 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 995 996 // We have the displaced header in displaced_header. If the lock is still 997 // lightweight, it will contain the monitor address and we'll store the 998 // displaced header back into the object's mark word. 999 // CmpxchgX sets CCR0 to cmpX(current, monitor). 1000 cmpxchgd(/*flag=*/CCR0, 1001 /*current_value=*/current_header, 1002 /*compare_value=*/monitor, /*exchange_value=*/displaced_header, 1003 /*where=*/object_mark_addr, 1004 MacroAssembler::MemBarRel, 1005 MacroAssembler::cmpxchgx_hint_release_lock(), 1006 noreg, 1007 &slow_case); 1008 b(free_slot); 1009 1010 // } else { 1011 // // Slow path. 1012 // InterpreterRuntime::monitorexit(THREAD, monitor); 1013 1014 // The lock has been converted into a heavy lock and hence 1015 // we need to get into the slow case. 1016 bind(slow_case); 1017 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1018 monitor, check_for_exceptions); 1019 // } 1020 1021 Label done; 1022 b(done); // Monitor register may be overwritten! Runtime has already freed the slot. 1023 1024 // Exchange worked, do monitor->set_obj(NULL); 1025 align(32, 12); 1026 bind(free_slot); 1027 li(R0, 0); 1028 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor); 1029 bind(done); 1030 } 1031 } 1032 1033 // Load compiled (i2c) or interpreter entry when calling from interpreted and 1034 // do the call. Centralized so that all interpreter calls will do the same actions. 1035 // If jvmti single stepping is on for a thread we must not call compiled code. 1036 // 1037 // Input: 1038 // - Rtarget_method: method to call 1039 // - Rret_addr: return address 1040 // - 2 scratch regs 1041 // 1042 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, 1043 Register Rscratch1, Register Rscratch2) { 1044 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr); 1045 // Assume we want to go compiled if available. 1046 const Register Rtarget_addr = Rscratch1; 1047 const Register Rinterp_only = Rscratch2; 1048 1049 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method); 1050 1051 if (JvmtiExport::can_post_interpreter_events()) { 1052 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 1053 1054 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 1055 // compiled code in threads for which the event is enabled. Check here for 1056 // interp_only_mode if these events CAN be enabled. 1057 Label done; 1058 verify_thread(); 1059 cmpwi(CCR0, Rinterp_only, 0); 1060 beq(CCR0, done); 1061 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method); 1062 align(32, 12); 1063 bind(done); 1064 } 1065 1066 #ifdef ASSERT 1067 { 1068 Label Lok; 1069 cmpdi(CCR0, Rtarget_addr, 0); 1070 bne(CCR0, Lok); 1071 stop("null entry point"); 1072 bind(Lok); 1073 } 1074 #endif // ASSERT 1075 1076 mr(R21_sender_SP, R1_SP); 1077 1078 // Calc a precise SP for the call. The SP value we calculated in 1079 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space 1080 // if esp is not max. Also, the i2c adapter extends the stack space without restoring 1081 // our pre-calced value, so repeating calls via i2c would result in stack overflow. 1082 // Since esp already points to an empty slot, we just have to sub 1 additional slot 1083 // to meet the abi scratch requirements. 1084 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in 1085 // the return entry of the interpreter. 1086 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size); 1087 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address 1088 resize_frame_absolute(Rscratch2, Rscratch2, R0); 1089 1090 mr_if_needed(R19_method, Rtarget_method); 1091 mtctr(Rtarget_addr); 1092 mtlr(Rret_addr); 1093 1094 save_interpreter_state(Rscratch2); 1095 #ifdef ASSERT 1096 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp 1097 cmpd(CCR0, R21_sender_SP, Rscratch1); 1098 asm_assert_eq("top_frame_sp incorrect", 0x951); 1099 #endif 1100 1101 bctr(); 1102 } 1103 1104 // Set the method data pointer for the current bcp. 1105 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1106 assert(ProfileInterpreter, "must be profiling interpreter"); 1107 Label get_continue; 1108 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method); 1109 test_method_data_pointer(get_continue); 1110 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp); 1111 1112 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1113 add(R28_mdx, R28_mdx, R3_RET); 1114 bind(get_continue); 1115 } 1116 1117 // Test ImethodDataPtr. If it is null, continue at the specified label. 1118 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1119 assert(ProfileInterpreter, "must be profiling interpreter"); 1120 cmpdi(CCR0, R28_mdx, 0); 1121 beq(CCR0, zero_continue); 1122 } 1123 1124 void InterpreterMacroAssembler::verify_method_data_pointer() { 1125 assert(ProfileInterpreter, "must be profiling interpreter"); 1126 #ifdef ASSERT 1127 Label verify_continue; 1128 test_method_data_pointer(verify_continue); 1129 1130 // If the mdp is valid, it will point to a DataLayout header which is 1131 // consistent with the bcp. The converse is highly probable also. 1132 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx); 1133 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method); 1134 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset())); 1135 add(R11_scratch1, R12_scratch2, R12_scratch2); 1136 cmpd(CCR0, R11_scratch1, R14_bcp); 1137 beq(CCR0, verify_continue); 1138 1139 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx); 1140 1141 bind(verify_continue); 1142 #endif 1143 } 1144 1145 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1146 Register method_counters, 1147 Register Rscratch, 1148 Label &profile_continue) { 1149 assert(ProfileInterpreter, "must be profiling interpreter"); 1150 // Control will flow to "profile_continue" if the counter is less than the 1151 // limit or if we call profile_method(). 1152 Label done; 1153 1154 // If no method data exists, and the counter is high enough, make one. 1155 lwz(Rscratch, in_bytes(MethodCounters::interpreter_profile_limit_offset()), method_counters); 1156 1157 cmpdi(CCR0, R28_mdx, 0); 1158 // Test to see if we should create a method data oop. 1159 cmpd(CCR1, Rscratch, invocation_count); 1160 bne(CCR0, done); 1161 bge(CCR1, profile_continue); 1162 1163 // Build it now. 1164 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1165 set_method_data_pointer_for_bcp(); 1166 b(profile_continue); 1167 1168 align(32, 12); 1169 bind(done); 1170 } 1171 1172 void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register method_counters, 1173 Register target_bcp, Register disp, Register Rtmp) { 1174 assert_different_registers(backedge_count, target_bcp, disp, Rtmp, R4_ARG2); 1175 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 1176 1177 Label did_not_overflow; 1178 Label overflow_with_error; 1179 1180 lwz(Rtmp, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()), method_counters); 1181 cmpw(CCR0, backedge_count, Rtmp); 1182 1183 blt(CCR0, did_not_overflow); 1184 1185 // When ProfileInterpreter is on, the backedge_count comes from the 1186 // methodDataOop, which value does not get reset on the call to 1187 // frequency_counter_overflow(). To avoid excessive calls to the overflow 1188 // routine while the method is being compiled, add a second test to make sure 1189 // the overflow function is called only once every overflow_frequency. 1190 if (ProfileInterpreter) { 1191 const int overflow_frequency = 1024; 1192 andi_(Rtmp, backedge_count, overflow_frequency-1); 1193 bne(CCR0, did_not_overflow); 1194 } 1195 1196 // Overflow in loop, pass branch bytecode. 1197 subf(R4_ARG2, disp, target_bcp); // Compute branch bytecode (previous bcp). 1198 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); 1199 1200 // Was an OSR adapter generated? 1201 cmpdi(CCR0, R3_RET, 0); 1202 beq(CCR0, overflow_with_error); 1203 1204 // Has the nmethod been invalidated already? 1205 lbz(Rtmp, nmethod::state_offset(), R3_RET); 1206 cmpwi(CCR0, Rtmp, nmethod::in_use); 1207 bne(CCR0, overflow_with_error); 1208 1209 // Migrate the interpreter frame off of the stack. 1210 // We can use all registers because we will not return to interpreter from this point. 1211 1212 // Save nmethod. 1213 const Register osr_nmethod = R31; 1214 mr(osr_nmethod, R3_RET); 1215 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1); 1216 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread); 1217 reset_last_Java_frame(); 1218 // OSR buffer is in ARG1 1219 1220 // Remove the interpreter frame. 1221 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1222 1223 // Jump to the osr code. 1224 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod); 1225 mtlr(R0); 1226 mtctr(R11_scratch1); 1227 bctr(); 1228 1229 align(32, 12); 1230 bind(overflow_with_error); 1231 bind(did_not_overflow); 1232 } 1233 1234 // Store a value at some constant offset from the method data pointer. 1235 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1236 assert(ProfileInterpreter, "must be profiling interpreter"); 1237 1238 std(value, constant, R28_mdx); 1239 } 1240 1241 // Increment the value at some constant offset from the method data pointer. 1242 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1243 Register counter_addr, 1244 Register Rbumped_count, 1245 bool decrement) { 1246 // Locate the counter at a fixed offset from the mdp: 1247 addi(counter_addr, R28_mdx, constant); 1248 increment_mdp_data_at(counter_addr, Rbumped_count, decrement); 1249 } 1250 1251 // Increment the value at some non-fixed (reg + constant) offset from 1252 // the method data pointer. 1253 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1254 int constant, 1255 Register scratch, 1256 Register Rbumped_count, 1257 bool decrement) { 1258 // Add the constant to reg to get the offset. 1259 add(scratch, R28_mdx, reg); 1260 // Then calculate the counter address. 1261 addi(scratch, scratch, constant); 1262 increment_mdp_data_at(scratch, Rbumped_count, decrement); 1263 } 1264 1265 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr, 1266 Register Rbumped_count, 1267 bool decrement) { 1268 assert(ProfileInterpreter, "must be profiling interpreter"); 1269 1270 // Load the counter. 1271 ld(Rbumped_count, 0, counter_addr); 1272 1273 if (decrement) { 1274 // Decrement the register. Set condition codes. 1275 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment); 1276 // Store the decremented counter, if it is still negative. 1277 std(Rbumped_count, 0, counter_addr); 1278 // Note: add/sub overflow check are not ported, since 64 bit 1279 // calculation should never overflow. 1280 } else { 1281 // Increment the register. Set carry flag. 1282 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment); 1283 // Store the incremented counter. 1284 std(Rbumped_count, 0, counter_addr); 1285 } 1286 } 1287 1288 // Set a flag value at the current method data pointer position. 1289 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1290 Register scratch) { 1291 assert(ProfileInterpreter, "must be profiling interpreter"); 1292 // Load the data header. 1293 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1294 // Set the flag. 1295 ori(scratch, scratch, flag_constant); 1296 // Store the modified header. 1297 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1298 } 1299 1300 // Test the location at some offset from the method data pointer. 1301 // If it is not equal to value, branch to the not_equal_continue Label. 1302 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1303 Register value, 1304 Label& not_equal_continue, 1305 Register test_out) { 1306 assert(ProfileInterpreter, "must be profiling interpreter"); 1307 1308 ld(test_out, offset, R28_mdx); 1309 cmpd(CCR0, value, test_out); 1310 bne(CCR0, not_equal_continue); 1311 } 1312 1313 // Update the method data pointer by the displacement located at some fixed 1314 // offset from the method data pointer. 1315 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1316 Register scratch) { 1317 assert(ProfileInterpreter, "must be profiling interpreter"); 1318 1319 ld(scratch, offset_of_disp, R28_mdx); 1320 add(R28_mdx, scratch, R28_mdx); 1321 } 1322 1323 // Update the method data pointer by the displacement located at the 1324 // offset (reg + offset_of_disp). 1325 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1326 int offset_of_disp, 1327 Register scratch) { 1328 assert(ProfileInterpreter, "must be profiling interpreter"); 1329 1330 add(scratch, reg, R28_mdx); 1331 ld(scratch, offset_of_disp, scratch); 1332 add(R28_mdx, scratch, R28_mdx); 1333 } 1334 1335 // Update the method data pointer by a simple constant displacement. 1336 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1337 assert(ProfileInterpreter, "must be profiling interpreter"); 1338 addi(R28_mdx, R28_mdx, constant); 1339 } 1340 1341 // Update the method data pointer for a _ret bytecode whose target 1342 // was not among our cached targets. 1343 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1344 Register return_bci) { 1345 assert(ProfileInterpreter, "must be profiling interpreter"); 1346 1347 push(state); 1348 assert(return_bci->is_nonvolatile(), "need to protect return_bci"); 1349 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1350 pop(state); 1351 } 1352 1353 // Increments the backedge counter. 1354 // Returns backedge counter + invocation counter in Rdst. 1355 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst, 1356 const Register Rtmp1, Register Rscratch) { 1357 assert(UseCompiler, "incrementing must be useful"); 1358 assert_different_registers(Rdst, Rtmp1); 1359 const Register invocation_counter = Rtmp1; 1360 const Register counter = Rdst; 1361 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size."); 1362 1363 // Load backedge counter. 1364 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1365 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1366 // Load invocation counter. 1367 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) + 1368 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1369 1370 // Add the delta to the backedge counter. 1371 addi(counter, counter, InvocationCounter::count_increment); 1372 1373 // Mask the invocation counter. 1374 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value); 1375 1376 // Store new counter value. 1377 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1378 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1379 // Return invocation counter + backedge counter. 1380 add(counter, counter, invocation_counter); 1381 } 1382 1383 // Count a taken branch in the bytecodes. 1384 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1385 if (ProfileInterpreter) { 1386 Label profile_continue; 1387 1388 // If no method data exists, go to profile_continue. 1389 test_method_data_pointer(profile_continue); 1390 1391 // We are taking a branch. Increment the taken count. 1392 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count); 1393 1394 // The method data pointer needs to be updated to reflect the new target. 1395 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1396 bind (profile_continue); 1397 } 1398 } 1399 1400 // Count a not-taken branch in the bytecodes. 1401 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) { 1402 if (ProfileInterpreter) { 1403 Label profile_continue; 1404 1405 // If no method data exists, go to profile_continue. 1406 test_method_data_pointer(profile_continue); 1407 1408 // We are taking a branch. Increment the not taken count. 1409 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2); 1410 1411 // The method data pointer needs to be updated to correspond to the 1412 // next bytecode. 1413 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1414 bind (profile_continue); 1415 } 1416 } 1417 1418 // Count a non-virtual call in the bytecodes. 1419 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) { 1420 if (ProfileInterpreter) { 1421 Label profile_continue; 1422 1423 // If no method data exists, go to profile_continue. 1424 test_method_data_pointer(profile_continue); 1425 1426 // We are making a call. Increment the count. 1427 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1428 1429 // The method data pointer needs to be updated to reflect the new target. 1430 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1431 bind (profile_continue); 1432 } 1433 } 1434 1435 // Count a final call in the bytecodes. 1436 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) { 1437 if (ProfileInterpreter) { 1438 Label profile_continue; 1439 1440 // If no method data exists, go to profile_continue. 1441 test_method_data_pointer(profile_continue); 1442 1443 // We are making a call. Increment the count. 1444 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1445 1446 // The method data pointer needs to be updated to reflect the new target. 1447 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1448 bind (profile_continue); 1449 } 1450 } 1451 1452 // Count a virtual call in the bytecodes. 1453 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver, 1454 Register Rscratch1, 1455 Register Rscratch2, 1456 bool receiver_can_be_null) { 1457 if (!ProfileInterpreter) { return; } 1458 Label profile_continue; 1459 1460 // If no method data exists, go to profile_continue. 1461 test_method_data_pointer(profile_continue); 1462 1463 Label skip_receiver_profile; 1464 if (receiver_can_be_null) { 1465 Label not_null; 1466 cmpdi(CCR0, Rreceiver, 0); 1467 bne(CCR0, not_null); 1468 // We are making a call. Increment the count for null receiver. 1469 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2); 1470 b(skip_receiver_profile); 1471 bind(not_null); 1472 } 1473 1474 // Record the receiver type. 1475 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true); 1476 bind(skip_receiver_profile); 1477 1478 // The method data pointer needs to be updated to reflect the new target. 1479 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1480 bind (profile_continue); 1481 } 1482 1483 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) { 1484 if (ProfileInterpreter) { 1485 Label profile_continue; 1486 1487 // If no method data exists, go to profile_continue. 1488 test_method_data_pointer(profile_continue); 1489 1490 int mdp_delta = in_bytes(BitData::bit_data_size()); 1491 if (TypeProfileCasts) { 1492 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1493 1494 // Record the object type. 1495 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false); 1496 } 1497 1498 // The method data pointer needs to be updated. 1499 update_mdp_by_constant(mdp_delta); 1500 1501 bind (profile_continue); 1502 } 1503 } 1504 1505 void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) { 1506 if (ProfileInterpreter && TypeProfileCasts) { 1507 Label profile_continue; 1508 1509 // If no method data exists, go to profile_continue. 1510 test_method_data_pointer(profile_continue); 1511 1512 int count_offset = in_bytes(CounterData::count_offset()); 1513 // Back up the address, since we have already bumped the mdp. 1514 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1515 1516 // *Decrement* the counter. We expect to see zero or small negatives. 1517 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true); 1518 1519 bind (profile_continue); 1520 } 1521 } 1522 1523 // Count a ret in the bytecodes. 1524 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, 1525 Register scratch1, Register scratch2) { 1526 if (ProfileInterpreter) { 1527 Label profile_continue; 1528 uint row; 1529 1530 // If no method data exists, go to profile_continue. 1531 test_method_data_pointer(profile_continue); 1532 1533 // Update the total ret count. 1534 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 ); 1535 1536 for (row = 0; row < RetData::row_limit(); row++) { 1537 Label next_test; 1538 1539 // See if return_bci is equal to bci[n]: 1540 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1); 1541 1542 // return_bci is equal to bci[n]. Increment the count. 1543 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2); 1544 1545 // The method data pointer needs to be updated to reflect the new target. 1546 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1); 1547 b(profile_continue); 1548 bind(next_test); 1549 } 1550 1551 update_mdp_for_ret(state, return_bci); 1552 1553 bind (profile_continue); 1554 } 1555 } 1556 1557 // Count the default case of a switch construct. 1558 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) { 1559 if (ProfileInterpreter) { 1560 Label profile_continue; 1561 1562 // If no method data exists, go to profile_continue. 1563 test_method_data_pointer(profile_continue); 1564 1565 // Update the default case count 1566 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1567 scratch1, scratch2); 1568 1569 // The method data pointer needs to be updated. 1570 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()), 1571 scratch1); 1572 1573 bind (profile_continue); 1574 } 1575 } 1576 1577 // Count the index'th case of a switch construct. 1578 void InterpreterMacroAssembler::profile_switch_case(Register index, 1579 Register scratch1, 1580 Register scratch2, 1581 Register scratch3) { 1582 if (ProfileInterpreter) { 1583 assert_different_registers(index, scratch1, scratch2, scratch3); 1584 Label profile_continue; 1585 1586 // If no method data exists, go to profile_continue. 1587 test_method_data_pointer(profile_continue); 1588 1589 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes(). 1590 li(scratch3, in_bytes(MultiBranchData::case_array_offset())); 1591 1592 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works"); 1593 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1594 add(scratch1, scratch1, scratch3); 1595 1596 // Update the case count. 1597 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3); 1598 1599 // The method data pointer needs to be updated. 1600 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2); 1601 1602 bind (profile_continue); 1603 } 1604 } 1605 1606 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) { 1607 if (ProfileInterpreter) { 1608 assert_different_registers(Rscratch1, Rscratch2); 1609 Label profile_continue; 1610 1611 // If no method data exists, go to profile_continue. 1612 test_method_data_pointer(profile_continue); 1613 1614 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1); 1615 1616 // The method data pointer needs to be updated. 1617 int mdp_delta = in_bytes(BitData::bit_data_size()); 1618 if (TypeProfileCasts) { 1619 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1620 } 1621 update_mdp_by_constant(mdp_delta); 1622 1623 bind (profile_continue); 1624 } 1625 } 1626 1627 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver, 1628 Register Rscratch1, Register Rscratch2, 1629 bool is_virtual_call) { 1630 assert(ProfileInterpreter, "must be profiling"); 1631 assert_different_registers(Rreceiver, Rscratch1, Rscratch2); 1632 1633 Label done; 1634 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call); 1635 bind (done); 1636 } 1637 1638 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1639 Register receiver, Register scratch1, Register scratch2, 1640 int start_row, Label& done, bool is_virtual_call) { 1641 if (TypeProfileWidth == 0) { 1642 if (is_virtual_call) { 1643 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1644 } 1645 return; 1646 } 1647 1648 int last_row = VirtualCallData::row_limit() - 1; 1649 assert(start_row <= last_row, "must be work left to do"); 1650 // Test this row for both the receiver and for null. 1651 // Take any of three different outcomes: 1652 // 1. found receiver => increment count and goto done 1653 // 2. found null => keep looking for case 1, maybe allocate this cell 1654 // 3. found something else => keep looking for cases 1 and 2 1655 // Case 3 is handled by a recursive call. 1656 for (int row = start_row; row <= last_row; row++) { 1657 Label next_test; 1658 bool test_for_null_also = (row == start_row); 1659 1660 // See if the receiver is receiver[n]. 1661 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1662 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1); 1663 // delayed()->tst(scratch); 1664 1665 // The receiver is receiver[n]. Increment count[n]. 1666 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1667 increment_mdp_data_at(count_offset, scratch1, scratch2); 1668 b(done); 1669 bind(next_test); 1670 1671 if (test_for_null_also) { 1672 Label found_null; 1673 // Failed the equality check on receiver[n]... Test for null. 1674 if (start_row == last_row) { 1675 // The only thing left to do is handle the null case. 1676 if (is_virtual_call) { 1677 // Scratch1 contains test_out from test_mdp_data_at. 1678 cmpdi(CCR0, scratch1, 0); 1679 beq(CCR0, found_null); 1680 // Receiver did not match any saved receiver and there is no empty row for it. 1681 // Increment total counter to indicate polymorphic case. 1682 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1683 b(done); 1684 bind(found_null); 1685 } else { 1686 cmpdi(CCR0, scratch1, 0); 1687 bne(CCR0, done); 1688 } 1689 break; 1690 } 1691 // Since null is rare, make it be the branch-taken case. 1692 cmpdi(CCR0, scratch1, 0); 1693 beq(CCR0, found_null); 1694 1695 // Put all the "Case 3" tests here. 1696 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call); 1697 1698 // Found a null. Keep searching for a matching receiver, 1699 // but remember that this is an empty (unused) slot. 1700 bind(found_null); 1701 } 1702 } 1703 1704 // In the fall-through case, we found no matching receiver, but we 1705 // observed the receiver[start_row] is NULL. 1706 1707 // Fill in the receiver field and increment the count. 1708 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1709 set_mdp_data_at(recvr_offset, receiver); 1710 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1711 li(scratch1, DataLayout::counter_increment); 1712 set_mdp_data_at(count_offset, scratch1); 1713 if (start_row > 0) { 1714 b(done); 1715 } 1716 } 1717 1718 // Argument and return type profilig. 1719 // kills: tmp, tmp2, R0, CR0, CR1 1720 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base, 1721 RegisterOrConstant mdo_addr_offs, 1722 Register tmp, Register tmp2) { 1723 Label do_nothing, do_update; 1724 1725 // tmp2 = obj is allowed 1726 assert_different_registers(obj, mdo_addr_base, tmp, R0); 1727 assert_different_registers(tmp2, mdo_addr_base, tmp, R0); 1728 const Register klass = tmp2; 1729 1730 verify_oop(obj); 1731 1732 ld(tmp, mdo_addr_offs, mdo_addr_base); 1733 1734 // Set null_seen if obj is 0. 1735 cmpdi(CCR0, obj, 0); 1736 ori(R0, tmp, TypeEntries::null_seen); 1737 beq(CCR0, do_update); 1738 1739 load_klass(klass, obj); 1740 1741 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask)); 1742 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask); 1743 cmpd(CCR1, R0, klass); 1744 // Klass seen before, nothing to do (regardless of unknown bit). 1745 //beq(CCR1, do_nothing); 1746 1747 andi_(R0, klass, TypeEntries::type_unknown); 1748 // Already unknown. Nothing to do anymore. 1749 //bne(CCR0, do_nothing); 1750 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne 1751 beq(CCR0, do_nothing); 1752 1753 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask)); 1754 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1755 beq(CCR0, do_update); // First time here. Set profile type. 1756 1757 // Different than before. Cannot keep accurate profile. 1758 ori(R0, tmp, TypeEntries::type_unknown); 1759 1760 bind(do_update); 1761 // update profile 1762 std(R0, mdo_addr_offs, mdo_addr_base); 1763 1764 align(32, 12); 1765 bind(do_nothing); 1766 } 1767 1768 void InterpreterMacroAssembler::profile_arguments_type(Register callee, 1769 Register tmp1, Register tmp2, 1770 bool is_virtual) { 1771 if (!ProfileInterpreter) { 1772 return; 1773 } 1774 1775 assert_different_registers(callee, tmp1, tmp2, R28_mdx); 1776 1777 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1778 Label profile_continue; 1779 1780 test_method_data_pointer(profile_continue); 1781 1782 int off_to_start = is_virtual ? 1783 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1784 1785 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx); 1786 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1787 bne(CCR0, profile_continue); 1788 1789 if (MethodData::profile_arguments()) { 1790 Label done; 1791 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1792 add(R28_mdx, off_to_args, R28_mdx); 1793 1794 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1795 if (i > 0 || MethodData::profile_return()) { 1796 // If return value type is profiled we may have no argument to profile. 1797 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1798 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count()); 1799 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count()); 1800 blt(CCR0, done); 1801 } 1802 ld(tmp1, in_bytes(Method::const_offset()), callee); 1803 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1); 1804 // Stack offset o (zero based) from the start of the argument 1805 // list, for n arguments translates into offset n - o - 1 from 1806 // the end of the argument list. But there's an extra slot at 1807 // the top of the stack. So the offset is n - o from Lesp. 1808 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx); 1809 subf(tmp1, tmp2, tmp1); 1810 1811 sldi(tmp1, tmp1, Interpreter::logStackElementSize); 1812 ldx(tmp1, tmp1, R15_esp); 1813 1814 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1); 1815 1816 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1817 addi(R28_mdx, R28_mdx, to_add); 1818 off_to_args += to_add; 1819 } 1820 1821 if (MethodData::profile_return()) { 1822 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1823 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1824 } 1825 1826 bind(done); 1827 1828 if (MethodData::profile_return()) { 1829 // We're right after the type profile for the last 1830 // argument. tmp1 is the number of cells left in the 1831 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1832 // if there's a return to profile. 1833 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), 1834 "can't move past ret type"); 1835 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size)); 1836 add(R28_mdx, tmp1, R28_mdx); 1837 } 1838 } else { 1839 assert(MethodData::profile_return(), "either profile call args or call ret"); 1840 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 1841 } 1842 1843 // Mdp points right after the end of the 1844 // CallTypeData/VirtualCallTypeData, right after the cells for the 1845 // return value type if there's one. 1846 align(32, 12); 1847 bind(profile_continue); 1848 } 1849 } 1850 1851 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 1852 assert_different_registers(ret, tmp1, tmp2); 1853 if (ProfileInterpreter && MethodData::profile_return()) { 1854 Label profile_continue; 1855 1856 test_method_data_pointer(profile_continue); 1857 1858 if (MethodData::profile_return_jsr292_only()) { 1859 // If we don't profile all invoke bytecodes we must make sure 1860 // it's a bytecode we indeed profile. We can't go back to the 1861 // begining of the ProfileData we intend to update to check its 1862 // type because we're right after it and we don't known its 1863 // length. 1864 lbz(tmp1, 0, R14_bcp); 1865 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method); 1866 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic); 1867 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle); 1868 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1869 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm); 1870 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1871 bne(CCR0, profile_continue); 1872 } 1873 1874 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2); 1875 1876 align(32, 12); 1877 bind(profile_continue); 1878 } 1879 } 1880 1881 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, 1882 Register tmp3, Register tmp4) { 1883 if (ProfileInterpreter && MethodData::profile_parameters()) { 1884 Label profile_continue, done; 1885 1886 test_method_data_pointer(profile_continue); 1887 1888 // Load the offset of the area within the MDO used for 1889 // parameters. If it's negative we're not profiling any parameters. 1890 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx); 1891 cmpwi(CCR0, tmp1, 0); 1892 blt(CCR0, profile_continue); 1893 1894 // Compute a pointer to the area for parameters from the offset 1895 // and move the pointer to the slot for the last 1896 // parameters. Collect profiling from last parameter down. 1897 // mdo start + parameters offset + array length - 1 1898 1899 // Pointer to the parameter area in the MDO. 1900 const Register mdp = tmp1; 1901 add(mdp, tmp1, R28_mdx); 1902 1903 // Offset of the current profile entry to update. 1904 const Register entry_offset = tmp2; 1905 // entry_offset = array len in number of cells 1906 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp); 1907 1908 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1909 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 1910 1911 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 1912 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size)); 1913 // entry_offset in bytes 1914 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1915 1916 Label loop; 1917 align(32, 12); 1918 bind(loop); 1919 1920 // Load offset on the stack from the slot for this parameter. 1921 ld(tmp3, entry_offset, mdp); 1922 sldi(tmp3, tmp3, Interpreter::logStackElementSize); 1923 neg(tmp3, tmp3); 1924 // Read the parameter from the local area. 1925 ldx(tmp3, tmp3, R18_locals); 1926 1927 // Make entry_offset now point to the type field for this parameter. 1928 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1929 assert(type_base > off_base, "unexpected"); 1930 addi(entry_offset, entry_offset, type_base - off_base); 1931 1932 // Profile the parameter. 1933 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3); 1934 1935 // Go to next parameter. 1936 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base); 1937 cmpdi(CCR0, entry_offset, off_base + delta); 1938 addi(entry_offset, entry_offset, -delta); 1939 bge(CCR0, loop); 1940 1941 align(32, 12); 1942 bind(profile_continue); 1943 } 1944 } 1945 1946 // Add a InterpMonitorElem to stack (see frame_sparc.hpp). 1947 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) { 1948 1949 // Very-local scratch registers. 1950 const Register esp = Rtemp1; 1951 const Register slot = Rtemp2; 1952 1953 // Extracted monitor_size. 1954 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes(); 1955 assert(Assembler::is_aligned((unsigned int)monitor_size, 1956 (unsigned int)frame::alignment_in_bytes), 1957 "size of a monitor must respect alignment of SP"); 1958 1959 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor 1960 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp 1961 1962 // Shuffle expression stack down. Recall that stack_base points 1963 // just above the new expression stack bottom. Old_tos and new_tos 1964 // are used to scan thru the old and new expression stacks. 1965 if (!stack_is_empty) { 1966 Label copy_slot, copy_slot_finished; 1967 const Register n_slots = slot; 1968 1969 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack). 1970 subf(n_slots, esp, R26_monitor); 1971 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy. 1972 assert(LogBytesPerWord == 3, "conflicts assembler instructions"); 1973 beq(CCR0, copy_slot_finished); // Nothing to copy. 1974 1975 mtctr(n_slots); 1976 1977 // loop 1978 bind(copy_slot); 1979 ld(slot, 0, esp); // Move expression stack down. 1980 std(slot, -monitor_size, esp); // distance = monitor_size 1981 addi(esp, esp, BytesPerWord); 1982 bdnz(copy_slot); 1983 1984 bind(copy_slot_finished); 1985 } 1986 1987 addi(R15_esp, R15_esp, -monitor_size); 1988 addi(R26_monitor, R26_monitor, -monitor_size); 1989 1990 // Restart interpreter 1991 } 1992 1993 // ============================================================================ 1994 // Java locals access 1995 1996 // Load a local variable at index in Rindex into register Rdst_value. 1997 // Also puts address of local into Rdst_address as a service. 1998 // Kills: 1999 // - Rdst_value 2000 // - Rdst_address 2001 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) { 2002 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2003 subf(Rdst_address, Rdst_address, R18_locals); 2004 lwz(Rdst_value, 0, Rdst_address); 2005 } 2006 2007 // Load a local variable at index in Rindex into register Rdst_value. 2008 // Also puts address of local into Rdst_address as a service. 2009 // Kills: 2010 // - Rdst_value 2011 // - Rdst_address 2012 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) { 2013 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2014 subf(Rdst_address, Rdst_address, R18_locals); 2015 ld(Rdst_value, -8, Rdst_address); 2016 } 2017 2018 // Load a local variable at index in Rindex into register Rdst_value. 2019 // Also puts address of local into Rdst_address as a service. 2020 // Input: 2021 // - Rindex: slot nr of local variable 2022 // Kills: 2023 // - Rdst_value 2024 // - Rdst_address 2025 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, 2026 Register Rdst_address, 2027 Register Rindex) { 2028 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2029 subf(Rdst_address, Rdst_address, R18_locals); 2030 ld(Rdst_value, 0, Rdst_address); 2031 } 2032 2033 // Load a local variable at index in Rindex into register Rdst_value. 2034 // Also puts address of local into Rdst_address as a service. 2035 // Kills: 2036 // - Rdst_value 2037 // - Rdst_address 2038 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, 2039 Register Rdst_address, 2040 Register Rindex) { 2041 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2042 subf(Rdst_address, Rdst_address, R18_locals); 2043 lfs(Rdst_value, 0, Rdst_address); 2044 } 2045 2046 // Load a local variable at index in Rindex into register Rdst_value. 2047 // Also puts address of local into Rdst_address as a service. 2048 // Kills: 2049 // - Rdst_value 2050 // - Rdst_address 2051 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, 2052 Register Rdst_address, 2053 Register Rindex) { 2054 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2055 subf(Rdst_address, Rdst_address, R18_locals); 2056 lfd(Rdst_value, -8, Rdst_address); 2057 } 2058 2059 // Store an int value at local variable slot Rindex. 2060 // Kills: 2061 // - Rindex 2062 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) { 2063 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2064 subf(Rindex, Rindex, R18_locals); 2065 stw(Rvalue, 0, Rindex); 2066 } 2067 2068 // Store a long value at local variable slot Rindex. 2069 // Kills: 2070 // - Rindex 2071 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) { 2072 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2073 subf(Rindex, Rindex, R18_locals); 2074 std(Rvalue, -8, Rindex); 2075 } 2076 2077 // Store an oop value at local variable slot Rindex. 2078 // Kills: 2079 // - Rindex 2080 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) { 2081 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2082 subf(Rindex, Rindex, R18_locals); 2083 std(Rvalue, 0, Rindex); 2084 } 2085 2086 // Store an int value at local variable slot Rindex. 2087 // Kills: 2088 // - Rindex 2089 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) { 2090 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2091 subf(Rindex, Rindex, R18_locals); 2092 stfs(Rvalue, 0, Rindex); 2093 } 2094 2095 // Store an int value at local variable slot Rindex. 2096 // Kills: 2097 // - Rindex 2098 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) { 2099 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2100 subf(Rindex, Rindex, R18_locals); 2101 stfd(Rvalue, -8, Rindex); 2102 } 2103 2104 // Read pending exception from thread and jump to interpreter. 2105 // Throw exception entry if one if pending. Fall through otherwise. 2106 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) { 2107 assert_different_registers(Rscratch1, Rscratch2, R3); 2108 Register Rexception = Rscratch1; 2109 Register Rtmp = Rscratch2; 2110 Label Ldone; 2111 // Get pending exception oop. 2112 ld(Rexception, thread_(pending_exception)); 2113 cmpdi(CCR0, Rexception, 0); 2114 beq(CCR0, Ldone); 2115 li(Rtmp, 0); 2116 mr_if_needed(R3, Rexception); 2117 std(Rtmp, thread_(pending_exception)); // Clear exception in thread 2118 if (Interpreter::rethrow_exception_entry() != NULL) { 2119 // Already got entry address. 2120 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry()); 2121 } else { 2122 // Dynamically load entry address. 2123 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true); 2124 ld(Rtmp, simm16_rest, Rtmp); 2125 } 2126 mtctr(Rtmp); 2127 save_interpreter_state(Rtmp); 2128 bctr(); 2129 2130 align(32, 12); 2131 bind(Ldone); 2132 } 2133 2134 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) { 2135 save_interpreter_state(R11_scratch1); 2136 2137 MacroAssembler::call_VM(oop_result, entry_point, false); 2138 2139 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2140 2141 check_and_handle_popframe(R11_scratch1); 2142 check_and_handle_earlyret(R11_scratch1); 2143 // Now check exceptions manually. 2144 if (check_exceptions) { 2145 check_and_forward_exception(R11_scratch1, R12_scratch2); 2146 } 2147 } 2148 2149 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2150 Register arg_1, bool check_exceptions) { 2151 // ARG1 is reserved for the thread. 2152 mr_if_needed(R4_ARG2, arg_1); 2153 call_VM(oop_result, entry_point, check_exceptions); 2154 } 2155 2156 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2157 Register arg_1, Register arg_2, 2158 bool check_exceptions) { 2159 // ARG1 is reserved for the thread. 2160 mr_if_needed(R4_ARG2, arg_1); 2161 assert(arg_2 != R4_ARG2, "smashed argument"); 2162 mr_if_needed(R5_ARG3, arg_2); 2163 call_VM(oop_result, entry_point, check_exceptions); 2164 } 2165 2166 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2167 Register arg_1, Register arg_2, Register arg_3, 2168 bool check_exceptions) { 2169 // ARG1 is reserved for the thread. 2170 mr_if_needed(R4_ARG2, arg_1); 2171 assert(arg_2 != R4_ARG2, "smashed argument"); 2172 mr_if_needed(R5_ARG3, arg_2); 2173 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument"); 2174 mr_if_needed(R6_ARG4, arg_3); 2175 call_VM(oop_result, entry_point, check_exceptions); 2176 } 2177 2178 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) { 2179 ld(scratch, 0, R1_SP); 2180 std(R15_esp, _ijava_state_neg(esp), scratch); 2181 std(R14_bcp, _ijava_state_neg(bcp), scratch); 2182 std(R26_monitor, _ijava_state_neg(monitors), scratch); 2183 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); } 2184 // Other entries should be unchanged. 2185 } 2186 2187 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) { 2188 ld(scratch, 0, R1_SP); 2189 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception). 2190 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code. 2191 if (!bcp_and_mdx_only) { 2192 // Following ones are Metadata. 2193 ld(R19_method, _ijava_state_neg(method), scratch); 2194 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch); 2195 // Following ones are stack addresses and don't require reload. 2196 ld(R15_esp, _ijava_state_neg(esp), scratch); 2197 ld(R18_locals, _ijava_state_neg(locals), scratch); 2198 ld(R26_monitor, _ijava_state_neg(monitors), scratch); 2199 } 2200 #ifdef ASSERT 2201 { 2202 Label Lok; 2203 subf(R0, R1_SP, scratch); 2204 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size); 2205 bge(CCR0, Lok); 2206 stop("frame too small (restore istate)", 0x5432); 2207 bind(Lok); 2208 } 2209 { 2210 Label Lok; 2211 ld(R0, _ijava_state_neg(ijava_reserved), scratch); 2212 cmpdi(CCR0, R0, 0x5afe); 2213 beq(CCR0, Lok); 2214 stop("frame corrupted (restore istate)", 0x5afe); 2215 bind(Lok); 2216 } 2217 #endif 2218 } 2219 2220 void InterpreterMacroAssembler::get_method_counters(Register method, 2221 Register Rcounters, 2222 Label& skip) { 2223 BLOCK_COMMENT("Load and ev. allocate counter object {"); 2224 Label has_counters; 2225 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2226 cmpdi(CCR0, Rcounters, 0); 2227 bne(CCR0, has_counters); 2228 call_VM(noreg, CAST_FROM_FN_PTR(address, 2229 InterpreterRuntime::build_method_counters), method, false); 2230 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2231 cmpdi(CCR0, Rcounters, 0); 2232 beq(CCR0, skip); // No MethodCounters, OutOfMemory. 2233 BLOCK_COMMENT("} Load and ev. allocate counter object"); 2234 2235 bind(has_counters); 2236 } 2237 2238 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, 2239 Register iv_be_count, 2240 Register Rtmp_r0) { 2241 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 2242 Register invocation_count = iv_be_count; 2243 Register backedge_count = Rtmp_r0; 2244 int delta = InvocationCounter::count_increment; 2245 2246 // Load each counter in a register. 2247 // ld(inv_counter, Rtmp); 2248 // ld(be_counter, Rtmp2); 2249 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + 2250 InvocationCounter::counter_offset()); 2251 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + 2252 InvocationCounter::counter_offset()); 2253 2254 BLOCK_COMMENT("Increment profiling counters {"); 2255 2256 // Load the backedge counter. 2257 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int 2258 // Mask the backedge counter. 2259 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value); 2260 2261 // Load the invocation counter. 2262 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int 2263 // Add the delta to the invocation counter and store the result. 2264 addi(invocation_count, invocation_count, delta); 2265 // Store value. 2266 stw(invocation_count, inv_counter_offset, Rcounters); 2267 2268 // Add invocation counter + backedge counter. 2269 add(iv_be_count, backedge_count, invocation_count); 2270 2271 // Note that this macro must leave the backedge_count + invocation_count in 2272 // register iv_be_count! 2273 BLOCK_COMMENT("} Increment profiling counters"); 2274 } 2275 2276 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2277 if (state == atos) { MacroAssembler::verify_oop(reg); } 2278 } 2279 2280 // Local helper function for the verify_oop_or_return_address macro. 2281 static bool verify_return_address(Method* m, int bci) { 2282 #ifndef PRODUCT 2283 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci; 2284 // Assume it is a valid return address if it is inside m and is preceded by a jsr. 2285 if (!m->contains(pc)) return false; 2286 address jsr_pc; 2287 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2288 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2289 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2290 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2291 #endif // PRODUCT 2292 return false; 2293 } 2294 2295 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2296 if (VerifyFPU) { 2297 unimplemented("verfiyFPU"); 2298 } 2299 } 2300 2301 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2302 if (!VerifyOops) return; 2303 2304 // The VM documentation for the astore[_wide] bytecode allows 2305 // the TOS to be not only an oop but also a return address. 2306 Label test; 2307 Label skip; 2308 // See if it is an address (in the current method): 2309 2310 const int log2_bytecode_size_limit = 16; 2311 srdi_(Rtmp, reg, log2_bytecode_size_limit); 2312 bne(CCR0, test); 2313 2314 address fd = CAST_FROM_FN_PTR(address, verify_return_address); 2315 const int nbytes_save = MacroAssembler::num_volatile_regs * 8; 2316 save_volatile_gprs(R1_SP, -nbytes_save); // except R0 2317 save_LR_CR(Rtmp); // Save in old frame. 2318 push_frame_reg_args(nbytes_save, Rtmp); 2319 2320 load_const_optimized(Rtmp, fd, R0); 2321 mr_if_needed(R4_ARG2, reg); 2322 mr(R3_ARG1, R19_method); 2323 call_c(Rtmp); // call C 2324 2325 pop_frame(); 2326 restore_LR_CR(Rtmp); 2327 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0 2328 b(skip); 2329 2330 // Perform a more elaborate out-of-line call. 2331 // Not an address; verify it: 2332 bind(test); 2333 verify_oop(reg); 2334 bind(skip); 2335 } 2336 2337 // Inline assembly for: 2338 // 2339 // if (thread is in interp_only_mode) { 2340 // InterpreterRuntime::post_method_entry(); 2341 // } 2342 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) || 2343 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) { 2344 // SharedRuntime::jvmpi_method_entry(method, receiver); 2345 // } 2346 void InterpreterMacroAssembler::notify_method_entry() { 2347 // JVMTI 2348 // Whenever JVMTI puts a thread in interp_only_mode, method 2349 // entry/exit events are sent for that thread to track stack 2350 // depth. If it is possible to enter interp_only_mode we add 2351 // the code to check if the event should be sent. 2352 if (JvmtiExport::can_post_interpreter_events()) { 2353 Label jvmti_post_done; 2354 2355 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2356 cmpwi(CCR0, R0, 0); 2357 beq(CCR0, jvmti_post_done); 2358 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), 2359 /*check_exceptions=*/true); 2360 2361 bind(jvmti_post_done); 2362 } 2363 } 2364 2365 // Inline assembly for: 2366 // 2367 // if (thread is in interp_only_mode) { 2368 // // save result 2369 // InterpreterRuntime::post_method_exit(); 2370 // // restore result 2371 // } 2372 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) { 2373 // // save result 2374 // SharedRuntime::jvmpi_method_exit(); 2375 // // restore result 2376 // } 2377 // 2378 // Native methods have their result stored in d_tmp and l_tmp. 2379 // Java methods have their result stored in the expression stack. 2380 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state, 2381 NotifyMethodExitMode mode, bool check_exceptions) { 2382 // JVMTI 2383 // Whenever JVMTI puts a thread in interp_only_mode, method 2384 // entry/exit events are sent for that thread to track stack 2385 // depth. If it is possible to enter interp_only_mode we add 2386 // the code to check if the event should be sent. 2387 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2388 Label jvmti_post_done; 2389 2390 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2391 cmpwi(CCR0, R0, 0); 2392 beq(CCR0, jvmti_post_done); 2393 if (!is_native_method) { push(state); } // Expose tos to GC. 2394 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), 2395 /*check_exceptions=*/check_exceptions); 2396 if (!is_native_method) { pop(state); } 2397 2398 align(32, 12); 2399 bind(jvmti_post_done); 2400 } 2401 2402 // Dtrace support not implemented. 2403 } 2404