1 /* 2 * Copyright (c) 2003, 2017, 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::cache_offset_in_bytes(), result); 458 ld(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result); 459 // JNIHandles::resolve(result) 460 ld(result, 0, result); 461 #ifdef ASSERT 462 Label index_ok; 463 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result); 464 sldi(R0, R0, LogBytesPerHeapOop); 465 cmpd(CCR0, tmp, R0); 466 blt(CCR0, index_ok); 467 stop("resolved reference index out of bounds", 0x09256); 468 bind(index_ok); 469 #endif 470 // Add in the index. 471 add(result, tmp, result); 472 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result, is_null); 473 } 474 475 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 476 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2. 477 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1, 478 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) { 479 // Profile the not-null value's klass. 480 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2); 481 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype); 482 profile_typecheck_failed(Rtmp1, Rtmp2); 483 } 484 485 // Separate these two to allow for delay slot in middle. 486 // These are used to do a test and full jump to exception-throwing code. 487 488 // Check that index is in range for array, then shift index by index_shift, 489 // and put arrayOop + shifted_index into res. 490 // Note: res is still shy of address by array offset into object. 491 492 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, 493 int index_shift, Register Rtmp, Register Rres) { 494 // Check that index is in range for array, then shift index by index_shift, 495 // and put arrayOop + shifted_index into res. 496 // Note: res is still shy of address by array offset into object. 497 // Kills: 498 // - Rindex 499 // Writes: 500 // - Rres: Address that corresponds to the array index if check was successful. 501 verify_oop(Rarray); 502 const Register Rlength = R0; 503 const Register RsxtIndex = Rtmp; 504 Label LisNull, LnotOOR; 505 506 // Array nullcheck 507 if (!ImplicitNullChecks) { 508 cmpdi(CCR0, Rarray, 0); 509 beq(CCR0, LisNull); 510 } else { 511 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex); 512 } 513 514 // Rindex might contain garbage in upper bits (remember that we don't sign extend 515 // during integer arithmetic operations). So kill them and put value into same register 516 // where ArrayIndexOutOfBounds would expect the index in. 517 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit 518 519 // Index check 520 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray); 521 cmplw(CCR0, Rindex, Rlength); 522 sldi(RsxtIndex, RsxtIndex, index_shift); 523 blt(CCR0, LnotOOR); 524 // Index should be in R17_tos, array should be in R4_ARG2. 525 mr_if_needed(R17_tos, Rindex); 526 mr_if_needed(R4_ARG2, Rarray); 527 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry); 528 mtctr(Rtmp); 529 bctr(); 530 531 if (!ImplicitNullChecks) { 532 bind(LisNull); 533 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry); 534 mtctr(Rtmp); 535 bctr(); 536 } 537 538 align(32, 16); 539 bind(LnotOOR); 540 541 // Calc address 542 add(Rres, RsxtIndex, Rarray); 543 } 544 545 void InterpreterMacroAssembler::index_check(Register array, Register index, 546 int index_shift, Register tmp, Register res) { 547 // pop array 548 pop_ptr(array); 549 550 // check array 551 index_check_without_pop(array, index, index_shift, tmp, res); 552 } 553 554 void InterpreterMacroAssembler::get_const(Register Rdst) { 555 ld(Rdst, in_bytes(Method::const_offset()), R19_method); 556 } 557 558 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 559 get_const(Rdst); 560 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 561 } 562 563 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 564 get_constant_pool(Rdst); 565 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 566 } 567 568 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 569 get_constant_pool(Rcpool); 570 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool); 571 } 572 573 // Unlock if synchronized method. 574 // 575 // Unlock the receiver if this is a synchronized method. 576 // Unlock any Java monitors from synchronized blocks. 577 // 578 // If there are locked Java monitors 579 // If throw_monitor_exception 580 // throws IllegalMonitorStateException 581 // Else if install_monitor_exception 582 // installs IllegalMonitorStateException 583 // Else 584 // no error processing 585 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 586 bool throw_monitor_exception, 587 bool install_monitor_exception) { 588 Label Lunlocked, Lno_unlock; 589 { 590 Register Rdo_not_unlock_flag = R11_scratch1; 591 Register Raccess_flags = R12_scratch2; 592 593 // Check if synchronized method or unlocking prevented by 594 // JavaThread::do_not_unlock_if_synchronized flag. 595 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 596 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method); 597 li(R0, 0); 598 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag 599 600 push(state); 601 602 // Skip if we don't have to unlock. 603 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0. 604 beq(CCR0, Lunlocked); 605 606 cmpwi(CCR0, Rdo_not_unlock_flag, 0); 607 bne(CCR0, Lno_unlock); 608 } 609 610 // Unlock 611 { 612 Register Rmonitor_base = R11_scratch1; 613 614 Label Lunlock; 615 // If it's still locked, everything is ok, unlock it. 616 ld(Rmonitor_base, 0, R1_SP); 617 addi(Rmonitor_base, Rmonitor_base, 618 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base 619 620 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base); 621 cmpdi(CCR0, R0, 0); 622 bne(CCR0, Lunlock); 623 624 // If it's already unlocked, throw exception. 625 if (throw_monitor_exception) { 626 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 627 should_not_reach_here(); 628 } else { 629 if (install_monitor_exception) { 630 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 631 b(Lunlocked); 632 } 633 } 634 635 bind(Lunlock); 636 unlock_object(Rmonitor_base); 637 } 638 639 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not. 640 bind(Lunlocked); 641 { 642 Label Lexception, Lrestart; 643 Register Rcurrent_obj_addr = R11_scratch1; 644 const int delta = frame::interpreter_frame_monitor_size_in_bytes(); 645 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords"); 646 647 bind(Lrestart); 648 // Set up search loop: Calc num of iterations. 649 { 650 Register Riterations = R12_scratch2; 651 Register Rmonitor_base = Rcurrent_obj_addr; 652 ld(Rmonitor_base, 0, R1_SP); 653 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base 654 655 subf_(Riterations, R26_monitor, Rmonitor_base); 656 ble(CCR0, Lno_unlock); 657 658 addi(Rcurrent_obj_addr, Rmonitor_base, 659 BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes()); 660 // Check if any monitor is on stack, bail out if not 661 srdi(Riterations, Riterations, exact_log2(delta)); 662 mtctr(Riterations); 663 } 664 665 // The search loop: Look for locked monitors. 666 { 667 const Register Rcurrent_obj = R0; 668 Label Lloop; 669 670 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 671 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 672 bind(Lloop); 673 674 // Check if current entry is used. 675 cmpdi(CCR0, Rcurrent_obj, 0); 676 bne(CCR0, Lexception); 677 // Preload next iteration's compare value. 678 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 679 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 680 bdnz(Lloop); 681 } 682 // Fell through: Everything's unlocked => finish. 683 b(Lno_unlock); 684 685 // An object is still locked => need to throw exception. 686 bind(Lexception); 687 if (throw_monitor_exception) { 688 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 689 should_not_reach_here(); 690 } else { 691 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 692 // Unlock does not block, so don't have to worry about the frame. 693 Register Rmonitor_addr = R11_scratch1; 694 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta); 695 unlock_object(Rmonitor_addr); 696 if (install_monitor_exception) { 697 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 698 } 699 b(Lrestart); 700 } 701 } 702 703 align(32, 12); 704 bind(Lno_unlock); 705 pop(state); 706 } 707 708 // Support function for remove_activation & Co. 709 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, 710 Register Rscratch1, Register Rscratch2) { 711 // Pop interpreter frame. 712 ld(Rscratch1, 0, R1_SP); // *SP 713 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp 714 ld(Rscratch2, 0, Rscratch1); // **SP 715 #ifdef ASSERT 716 { 717 Label Lok; 718 ld(R0, _ijava_state_neg(ijava_reserved), Rscratch1); 719 cmpdi(CCR0, R0, 0x5afe); 720 beq(CCR0, Lok); 721 stop("frame corrupted (remove activation)", 0x5afe); 722 bind(Lok); 723 } 724 #endif 725 if (return_pc!=noreg) { 726 ld(return_pc, _abi(lr), Rscratch1); // LR 727 } 728 729 // Merge top frames. 730 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP 731 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP 732 } 733 734 void InterpreterMacroAssembler::narrow(Register result) { 735 Register ret_type = R11_scratch1; 736 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method); 737 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1); 738 739 Label notBool, notByte, notChar, done; 740 741 // common case first 742 cmpwi(CCR0, ret_type, T_INT); 743 beq(CCR0, done); 744 745 cmpwi(CCR0, ret_type, T_BOOLEAN); 746 bne(CCR0, notBool); 747 andi(result, result, 0x1); 748 b(done); 749 750 bind(notBool); 751 cmpwi(CCR0, ret_type, T_BYTE); 752 bne(CCR0, notByte); 753 extsb(result, result); 754 b(done); 755 756 bind(notByte); 757 cmpwi(CCR0, ret_type, T_CHAR); 758 bne(CCR0, notChar); 759 andi(result, result, 0xffff); 760 b(done); 761 762 bind(notChar); 763 // cmpwi(CCR0, ret_type, T_SHORT); // all that's left 764 // bne(CCR0, done); 765 extsh(result, result); 766 767 // Nothing to do for T_INT 768 bind(done); 769 } 770 771 // Remove activation. 772 // 773 // Unlock the receiver if this is a synchronized method. 774 // Unlock any Java monitors from synchronized blocks. 775 // Remove the activation from the stack. 776 // 777 // If there are locked Java monitors 778 // If throw_monitor_exception 779 // throws IllegalMonitorStateException 780 // Else if install_monitor_exception 781 // installs IllegalMonitorStateException 782 // Else 783 // no error processing 784 void InterpreterMacroAssembler::remove_activation(TosState state, 785 bool throw_monitor_exception, 786 bool install_monitor_exception) { 787 BLOCK_COMMENT("remove_activation {"); 788 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 789 790 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 791 notify_method_exit(false, state, NotifyJVMTI, true); 792 793 BLOCK_COMMENT("reserved_stack_check:"); 794 if (StackReservedPages > 0) { 795 // Test if reserved zone needs to be enabled. 796 Label no_reserved_zone_enabling; 797 798 // Compare frame pointers. There is no good stack pointer, as with stack 799 // frame compression we can get different SPs when we do calls. A subsequent 800 // call could have a smaller SP, so that this compare succeeds for an 801 // inner call of the method annotated with ReservedStack. 802 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread); 803 ld_ptr(R11_scratch1, _abi(callers_sp), R1_SP); // Load frame pointer. 804 cmpld(CCR0, R11_scratch1, R0); 805 blt_predict_taken(CCR0, no_reserved_zone_enabling); 806 807 // Enable reserved zone again, throw stack overflow exception. 808 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread); 809 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError)); 810 811 should_not_reach_here(); 812 813 bind(no_reserved_zone_enabling); 814 } 815 816 verify_oop(R17_tos, state); 817 verify_thread(); 818 819 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 820 mtlr(R0); 821 BLOCK_COMMENT("} remove_activation"); 822 } 823 824 // Lock object 825 // 826 // Registers alive 827 // monitor - Address of the BasicObjectLock to be used for locking, 828 // which must be initialized with the object to lock. 829 // object - Address of the object to be locked. 830 // 831 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 832 if (UseHeavyMonitors) { 833 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 834 monitor, /*check_for_exceptions=*/true); 835 } else { 836 // template code: 837 // 838 // markOop displaced_header = obj->mark().set_unlocked(); 839 // monitor->lock()->set_displaced_header(displaced_header); 840 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 841 // // We stored the monitor address into the object's mark word. 842 // } else if (THREAD->is_lock_owned((address)displaced_header)) 843 // // Simple recursive case. 844 // monitor->lock()->set_displaced_header(NULL); 845 // } else { 846 // // Slow path. 847 // InterpreterRuntime::monitorenter(THREAD, monitor); 848 // } 849 850 const Register displaced_header = R7_ARG5; 851 const Register object_mark_addr = R8_ARG6; 852 const Register current_header = R9_ARG7; 853 const Register tmp = R10_ARG8; 854 855 Label done; 856 Label cas_failed, slow_case; 857 858 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp); 859 860 // markOop displaced_header = obj->mark().set_unlocked(); 861 862 // Load markOop from object into displaced_header. 863 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object); 864 865 if (UseBiasedLocking) { 866 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case); 867 } 868 869 // Set displaced_header to be (markOop of object | UNLOCK_VALUE). 870 ori(displaced_header, displaced_header, markOopDesc::unlocked_value); 871 872 // monitor->lock()->set_displaced_header(displaced_header); 873 874 // Initialize the box (Must happen before we update the object mark!). 875 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 876 BasicLock::displaced_header_offset_in_bytes(), monitor); 877 878 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 879 880 // Store stack address of the BasicObjectLock (this is monitor) into object. 881 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 882 883 // Must fence, otherwise, preceding store(s) may float below cmpxchg. 884 // CmpxchgX sets CCR0 to cmpX(current, displaced). 885 cmpxchgd(/*flag=*/CCR0, 886 /*current_value=*/current_header, 887 /*compare_value=*/displaced_header, /*exchange_value=*/monitor, 888 /*where=*/object_mark_addr, 889 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq, 890 MacroAssembler::cmpxchgx_hint_acquire_lock(), 891 noreg, 892 &cas_failed, 893 /*check without membar and ldarx first*/true); 894 895 // If the compare-and-exchange succeeded, then we found an unlocked 896 // object and we have now locked it. 897 b(done); 898 bind(cas_failed); 899 900 // } else if (THREAD->is_lock_owned((address)displaced_header)) 901 // // Simple recursive case. 902 // monitor->lock()->set_displaced_header(NULL); 903 904 // We did not see an unlocked object so try the fast recursive case. 905 906 // Check if owner is self by comparing the value in the markOop of object 907 // (current_header) with the stack pointer. 908 sub(current_header, current_header, R1_SP); 909 910 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 911 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place); 912 913 and_(R0/*==0?*/, current_header, tmp); 914 // If condition is true we are done and hence we can store 0 in the displaced 915 // header indicating it is a recursive lock. 916 bne(CCR0, slow_case); 917 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() + 918 BasicLock::displaced_header_offset_in_bytes(), monitor); 919 b(done); 920 921 // } else { 922 // // Slow path. 923 // InterpreterRuntime::monitorenter(THREAD, monitor); 924 925 // None of the above fast optimizations worked so we have to get into the 926 // slow case of monitor enter. 927 bind(slow_case); 928 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 929 monitor, /*check_for_exceptions=*/true); 930 // } 931 align(32, 12); 932 bind(done); 933 } 934 } 935 936 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 937 // 938 // Registers alive 939 // monitor - Address of the BasicObjectLock to be used for locking, 940 // which must be initialized with the object to lock. 941 // 942 // Throw IllegalMonitorException if object is not locked by current thread. 943 void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) { 944 if (UseHeavyMonitors) { 945 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 946 monitor, check_for_exceptions); 947 } else { 948 949 // template code: 950 // 951 // if ((displaced_header = monitor->displaced_header()) == NULL) { 952 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 953 // monitor->set_obj(NULL); 954 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 955 // // We swapped the unlocked mark in displaced_header into the object's mark word. 956 // monitor->set_obj(NULL); 957 // } else { 958 // // Slow path. 959 // InterpreterRuntime::monitorexit(THREAD, monitor); 960 // } 961 962 const Register object = R7_ARG5; 963 const Register displaced_header = R8_ARG6; 964 const Register object_mark_addr = R9_ARG7; 965 const Register current_header = R10_ARG8; 966 967 Label free_slot; 968 Label slow_case; 969 970 assert_different_registers(object, displaced_header, object_mark_addr, current_header); 971 972 if (UseBiasedLocking) { 973 // The object address from the monitor is in object. 974 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); 975 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0"); 976 biased_locking_exit(CCR0, object, displaced_header, free_slot); 977 } 978 979 // Test first if we are in the fast recursive case. 980 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 981 BasicLock::displaced_header_offset_in_bytes(), monitor); 982 983 // If the displaced header is zero, we have a recursive unlock. 984 cmpdi(CCR0, displaced_header, 0); 985 beq(CCR0, free_slot); // recursive unlock 986 987 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 988 // // We swapped the unlocked mark in displaced_header into the object's mark word. 989 // monitor->set_obj(NULL); 990 991 // If we still have a lightweight lock, unlock the object and be done. 992 993 // The object address from the monitor is in object. 994 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); } 995 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 996 997 // We have the displaced header in displaced_header. If the lock is still 998 // lightweight, it will contain the monitor address and we'll store the 999 // displaced header back into the object's mark word. 1000 // CmpxchgX sets CCR0 to cmpX(current, monitor). 1001 cmpxchgd(/*flag=*/CCR0, 1002 /*current_value=*/current_header, 1003 /*compare_value=*/monitor, /*exchange_value=*/displaced_header, 1004 /*where=*/object_mark_addr, 1005 MacroAssembler::MemBarRel, 1006 MacroAssembler::cmpxchgx_hint_release_lock(), 1007 noreg, 1008 &slow_case); 1009 b(free_slot); 1010 1011 // } else { 1012 // // Slow path. 1013 // InterpreterRuntime::monitorexit(THREAD, monitor); 1014 1015 // The lock has been converted into a heavy lock and hence 1016 // we need to get into the slow case. 1017 bind(slow_case); 1018 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1019 monitor, check_for_exceptions); 1020 // } 1021 1022 Label done; 1023 b(done); // Monitor register may be overwritten! Runtime has already freed the slot. 1024 1025 // Exchange worked, do monitor->set_obj(NULL); 1026 align(32, 12); 1027 bind(free_slot); 1028 li(R0, 0); 1029 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor); 1030 bind(done); 1031 } 1032 } 1033 1034 // Load compiled (i2c) or interpreter entry when calling from interpreted and 1035 // do the call. Centralized so that all interpreter calls will do the same actions. 1036 // If jvmti single stepping is on for a thread we must not call compiled code. 1037 // 1038 // Input: 1039 // - Rtarget_method: method to call 1040 // - Rret_addr: return address 1041 // - 2 scratch regs 1042 // 1043 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, 1044 Register Rscratch1, Register Rscratch2) { 1045 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr); 1046 // Assume we want to go compiled if available. 1047 const Register Rtarget_addr = Rscratch1; 1048 const Register Rinterp_only = Rscratch2; 1049 1050 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method); 1051 1052 if (JvmtiExport::can_post_interpreter_events()) { 1053 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 1054 1055 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 1056 // compiled code in threads for which the event is enabled. Check here for 1057 // interp_only_mode if these events CAN be enabled. 1058 Label done; 1059 verify_thread(); 1060 cmpwi(CCR0, Rinterp_only, 0); 1061 beq(CCR0, done); 1062 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method); 1063 align(32, 12); 1064 bind(done); 1065 } 1066 1067 #ifdef ASSERT 1068 { 1069 Label Lok; 1070 cmpdi(CCR0, Rtarget_addr, 0); 1071 bne(CCR0, Lok); 1072 stop("null entry point"); 1073 bind(Lok); 1074 } 1075 #endif // ASSERT 1076 1077 mr(R21_sender_SP, R1_SP); 1078 1079 // Calc a precise SP for the call. The SP value we calculated in 1080 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space 1081 // if esp is not max. Also, the i2c adapter extends the stack space without restoring 1082 // our pre-calced value, so repeating calls via i2c would result in stack overflow. 1083 // Since esp already points to an empty slot, we just have to sub 1 additional slot 1084 // to meet the abi scratch requirements. 1085 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in 1086 // the return entry of the interpreter. 1087 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size); 1088 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address 1089 resize_frame_absolute(Rscratch2, Rscratch2, R0); 1090 1091 mr_if_needed(R19_method, Rtarget_method); 1092 mtctr(Rtarget_addr); 1093 mtlr(Rret_addr); 1094 1095 save_interpreter_state(Rscratch2); 1096 #ifdef ASSERT 1097 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp 1098 cmpd(CCR0, R21_sender_SP, Rscratch1); 1099 asm_assert_eq("top_frame_sp incorrect", 0x951); 1100 #endif 1101 1102 bctr(); 1103 } 1104 1105 // Set the method data pointer for the current bcp. 1106 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1107 assert(ProfileInterpreter, "must be profiling interpreter"); 1108 Label get_continue; 1109 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method); 1110 test_method_data_pointer(get_continue); 1111 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp); 1112 1113 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1114 add(R28_mdx, R28_mdx, R3_RET); 1115 bind(get_continue); 1116 } 1117 1118 // Test ImethodDataPtr. If it is null, continue at the specified label. 1119 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1120 assert(ProfileInterpreter, "must be profiling interpreter"); 1121 cmpdi(CCR0, R28_mdx, 0); 1122 beq(CCR0, zero_continue); 1123 } 1124 1125 void InterpreterMacroAssembler::verify_method_data_pointer() { 1126 assert(ProfileInterpreter, "must be profiling interpreter"); 1127 #ifdef ASSERT 1128 Label verify_continue; 1129 test_method_data_pointer(verify_continue); 1130 1131 // If the mdp is valid, it will point to a DataLayout header which is 1132 // consistent with the bcp. The converse is highly probable also. 1133 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx); 1134 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method); 1135 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset())); 1136 add(R11_scratch1, R12_scratch2, R12_scratch2); 1137 cmpd(CCR0, R11_scratch1, R14_bcp); 1138 beq(CCR0, verify_continue); 1139 1140 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx); 1141 1142 bind(verify_continue); 1143 #endif 1144 } 1145 1146 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1147 Register method_counters, 1148 Register Rscratch, 1149 Label &profile_continue) { 1150 assert(ProfileInterpreter, "must be profiling interpreter"); 1151 // Control will flow to "profile_continue" if the counter is less than the 1152 // limit or if we call profile_method(). 1153 Label done; 1154 1155 // If no method data exists, and the counter is high enough, make one. 1156 lwz(Rscratch, in_bytes(MethodCounters::interpreter_profile_limit_offset()), method_counters); 1157 1158 cmpdi(CCR0, R28_mdx, 0); 1159 // Test to see if we should create a method data oop. 1160 cmpd(CCR1, Rscratch, invocation_count); 1161 bne(CCR0, done); 1162 bge(CCR1, profile_continue); 1163 1164 // Build it now. 1165 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1166 set_method_data_pointer_for_bcp(); 1167 b(profile_continue); 1168 1169 align(32, 12); 1170 bind(done); 1171 } 1172 1173 void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register method_counters, 1174 Register target_bcp, Register disp, Register Rtmp) { 1175 assert_different_registers(backedge_count, target_bcp, disp, Rtmp, R4_ARG2); 1176 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 1177 1178 Label did_not_overflow; 1179 Label overflow_with_error; 1180 1181 lwz(Rtmp, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()), method_counters); 1182 cmpw(CCR0, backedge_count, Rtmp); 1183 1184 blt(CCR0, did_not_overflow); 1185 1186 // When ProfileInterpreter is on, the backedge_count comes from the 1187 // methodDataOop, which value does not get reset on the call to 1188 // frequency_counter_overflow(). To avoid excessive calls to the overflow 1189 // routine while the method is being compiled, add a second test to make sure 1190 // the overflow function is called only once every overflow_frequency. 1191 if (ProfileInterpreter) { 1192 const int overflow_frequency = 1024; 1193 andi_(Rtmp, backedge_count, overflow_frequency-1); 1194 bne(CCR0, did_not_overflow); 1195 } 1196 1197 // Overflow in loop, pass branch bytecode. 1198 subf(R4_ARG2, disp, target_bcp); // Compute branch bytecode (previous bcp). 1199 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); 1200 1201 // Was an OSR adapter generated? 1202 cmpdi(CCR0, R3_RET, 0); 1203 beq(CCR0, overflow_with_error); 1204 1205 // Has the nmethod been invalidated already? 1206 lbz(Rtmp, nmethod::state_offset(), R3_RET); 1207 cmpwi(CCR0, Rtmp, nmethod::in_use); 1208 bne(CCR0, overflow_with_error); 1209 1210 // Migrate the interpreter frame off of the stack. 1211 // We can use all registers because we will not return to interpreter from this point. 1212 1213 // Save nmethod. 1214 const Register osr_nmethod = R31; 1215 mr(osr_nmethod, R3_RET); 1216 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1); 1217 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread); 1218 reset_last_Java_frame(); 1219 // OSR buffer is in ARG1 1220 1221 // Remove the interpreter frame. 1222 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1223 1224 // Jump to the osr code. 1225 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod); 1226 mtlr(R0); 1227 mtctr(R11_scratch1); 1228 bctr(); 1229 1230 align(32, 12); 1231 bind(overflow_with_error); 1232 bind(did_not_overflow); 1233 } 1234 1235 // Store a value at some constant offset from the method data pointer. 1236 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1237 assert(ProfileInterpreter, "must be profiling interpreter"); 1238 1239 std(value, constant, R28_mdx); 1240 } 1241 1242 // Increment the value at some constant offset from the method data pointer. 1243 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1244 Register counter_addr, 1245 Register Rbumped_count, 1246 bool decrement) { 1247 // Locate the counter at a fixed offset from the mdp: 1248 addi(counter_addr, R28_mdx, constant); 1249 increment_mdp_data_at(counter_addr, Rbumped_count, decrement); 1250 } 1251 1252 // Increment the value at some non-fixed (reg + constant) offset from 1253 // the method data pointer. 1254 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1255 int constant, 1256 Register scratch, 1257 Register Rbumped_count, 1258 bool decrement) { 1259 // Add the constant to reg to get the offset. 1260 add(scratch, R28_mdx, reg); 1261 // Then calculate the counter address. 1262 addi(scratch, scratch, constant); 1263 increment_mdp_data_at(scratch, Rbumped_count, decrement); 1264 } 1265 1266 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr, 1267 Register Rbumped_count, 1268 bool decrement) { 1269 assert(ProfileInterpreter, "must be profiling interpreter"); 1270 1271 // Load the counter. 1272 ld(Rbumped_count, 0, counter_addr); 1273 1274 if (decrement) { 1275 // Decrement the register. Set condition codes. 1276 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment); 1277 // Store the decremented counter, if it is still negative. 1278 std(Rbumped_count, 0, counter_addr); 1279 // Note: add/sub overflow check are not ported, since 64 bit 1280 // calculation should never overflow. 1281 } else { 1282 // Increment the register. Set carry flag. 1283 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment); 1284 // Store the incremented counter. 1285 std(Rbumped_count, 0, counter_addr); 1286 } 1287 } 1288 1289 // Set a flag value at the current method data pointer position. 1290 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1291 Register scratch) { 1292 assert(ProfileInterpreter, "must be profiling interpreter"); 1293 // Load the data header. 1294 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1295 // Set the flag. 1296 ori(scratch, scratch, flag_constant); 1297 // Store the modified header. 1298 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1299 } 1300 1301 // Test the location at some offset from the method data pointer. 1302 // If it is not equal to value, branch to the not_equal_continue Label. 1303 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1304 Register value, 1305 Label& not_equal_continue, 1306 Register test_out) { 1307 assert(ProfileInterpreter, "must be profiling interpreter"); 1308 1309 ld(test_out, offset, R28_mdx); 1310 cmpd(CCR0, value, test_out); 1311 bne(CCR0, not_equal_continue); 1312 } 1313 1314 // Update the method data pointer by the displacement located at some fixed 1315 // offset from the method data pointer. 1316 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1317 Register scratch) { 1318 assert(ProfileInterpreter, "must be profiling interpreter"); 1319 1320 ld(scratch, offset_of_disp, R28_mdx); 1321 add(R28_mdx, scratch, R28_mdx); 1322 } 1323 1324 // Update the method data pointer by the displacement located at the 1325 // offset (reg + offset_of_disp). 1326 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1327 int offset_of_disp, 1328 Register scratch) { 1329 assert(ProfileInterpreter, "must be profiling interpreter"); 1330 1331 add(scratch, reg, R28_mdx); 1332 ld(scratch, offset_of_disp, scratch); 1333 add(R28_mdx, scratch, R28_mdx); 1334 } 1335 1336 // Update the method data pointer by a simple constant displacement. 1337 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1338 assert(ProfileInterpreter, "must be profiling interpreter"); 1339 addi(R28_mdx, R28_mdx, constant); 1340 } 1341 1342 // Update the method data pointer for a _ret bytecode whose target 1343 // was not among our cached targets. 1344 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1345 Register return_bci) { 1346 assert(ProfileInterpreter, "must be profiling interpreter"); 1347 1348 push(state); 1349 assert(return_bci->is_nonvolatile(), "need to protect return_bci"); 1350 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1351 pop(state); 1352 } 1353 1354 // Increments the backedge counter. 1355 // Returns backedge counter + invocation counter in Rdst. 1356 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst, 1357 const Register Rtmp1, Register Rscratch) { 1358 assert(UseCompiler, "incrementing must be useful"); 1359 assert_different_registers(Rdst, Rtmp1); 1360 const Register invocation_counter = Rtmp1; 1361 const Register counter = Rdst; 1362 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size."); 1363 1364 // Load backedge counter. 1365 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1366 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1367 // Load invocation counter. 1368 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) + 1369 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1370 1371 // Add the delta to the backedge counter. 1372 addi(counter, counter, InvocationCounter::count_increment); 1373 1374 // Mask the invocation counter. 1375 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value); 1376 1377 // Store new counter value. 1378 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1379 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1380 // Return invocation counter + backedge counter. 1381 add(counter, counter, invocation_counter); 1382 } 1383 1384 // Count a taken branch in the bytecodes. 1385 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1386 if (ProfileInterpreter) { 1387 Label profile_continue; 1388 1389 // If no method data exists, go to profile_continue. 1390 test_method_data_pointer(profile_continue); 1391 1392 // We are taking a branch. Increment the taken count. 1393 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count); 1394 1395 // The method data pointer needs to be updated to reflect the new target. 1396 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1397 bind (profile_continue); 1398 } 1399 } 1400 1401 // Count a not-taken branch in the bytecodes. 1402 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) { 1403 if (ProfileInterpreter) { 1404 Label profile_continue; 1405 1406 // If no method data exists, go to profile_continue. 1407 test_method_data_pointer(profile_continue); 1408 1409 // We are taking a branch. Increment the not taken count. 1410 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2); 1411 1412 // The method data pointer needs to be updated to correspond to the 1413 // next bytecode. 1414 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1415 bind (profile_continue); 1416 } 1417 } 1418 1419 // Count a non-virtual call in the bytecodes. 1420 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) { 1421 if (ProfileInterpreter) { 1422 Label profile_continue; 1423 1424 // If no method data exists, go to profile_continue. 1425 test_method_data_pointer(profile_continue); 1426 1427 // We are making a call. Increment the count. 1428 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1429 1430 // The method data pointer needs to be updated to reflect the new target. 1431 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1432 bind (profile_continue); 1433 } 1434 } 1435 1436 // Count a final call in the bytecodes. 1437 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) { 1438 if (ProfileInterpreter) { 1439 Label profile_continue; 1440 1441 // If no method data exists, go to profile_continue. 1442 test_method_data_pointer(profile_continue); 1443 1444 // We are making a call. Increment the count. 1445 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1446 1447 // The method data pointer needs to be updated to reflect the new target. 1448 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1449 bind (profile_continue); 1450 } 1451 } 1452 1453 // Count a virtual call in the bytecodes. 1454 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver, 1455 Register Rscratch1, 1456 Register Rscratch2, 1457 bool receiver_can_be_null) { 1458 if (!ProfileInterpreter) { return; } 1459 Label profile_continue; 1460 1461 // If no method data exists, go to profile_continue. 1462 test_method_data_pointer(profile_continue); 1463 1464 Label skip_receiver_profile; 1465 if (receiver_can_be_null) { 1466 Label not_null; 1467 cmpdi(CCR0, Rreceiver, 0); 1468 bne(CCR0, not_null); 1469 // We are making a call. Increment the count for null receiver. 1470 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2); 1471 b(skip_receiver_profile); 1472 bind(not_null); 1473 } 1474 1475 // Record the receiver type. 1476 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true); 1477 bind(skip_receiver_profile); 1478 1479 // The method data pointer needs to be updated to reflect the new target. 1480 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1481 bind (profile_continue); 1482 } 1483 1484 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) { 1485 if (ProfileInterpreter) { 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 mdp_delta = in_bytes(BitData::bit_data_size()); 1492 if (TypeProfileCasts) { 1493 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1494 1495 // Record the object type. 1496 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false); 1497 } 1498 1499 // The method data pointer needs to be updated. 1500 update_mdp_by_constant(mdp_delta); 1501 1502 bind (profile_continue); 1503 } 1504 } 1505 1506 void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) { 1507 if (ProfileInterpreter && TypeProfileCasts) { 1508 Label profile_continue; 1509 1510 // If no method data exists, go to profile_continue. 1511 test_method_data_pointer(profile_continue); 1512 1513 int count_offset = in_bytes(CounterData::count_offset()); 1514 // Back up the address, since we have already bumped the mdp. 1515 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1516 1517 // *Decrement* the counter. We expect to see zero or small negatives. 1518 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true); 1519 1520 bind (profile_continue); 1521 } 1522 } 1523 1524 // Count a ret in the bytecodes. 1525 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, 1526 Register scratch1, Register scratch2) { 1527 if (ProfileInterpreter) { 1528 Label profile_continue; 1529 uint row; 1530 1531 // If no method data exists, go to profile_continue. 1532 test_method_data_pointer(profile_continue); 1533 1534 // Update the total ret count. 1535 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 ); 1536 1537 for (row = 0; row < RetData::row_limit(); row++) { 1538 Label next_test; 1539 1540 // See if return_bci is equal to bci[n]: 1541 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1); 1542 1543 // return_bci is equal to bci[n]. Increment the count. 1544 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2); 1545 1546 // The method data pointer needs to be updated to reflect the new target. 1547 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1); 1548 b(profile_continue); 1549 bind(next_test); 1550 } 1551 1552 update_mdp_for_ret(state, return_bci); 1553 1554 bind (profile_continue); 1555 } 1556 } 1557 1558 // Count the default case of a switch construct. 1559 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) { 1560 if (ProfileInterpreter) { 1561 Label profile_continue; 1562 1563 // If no method data exists, go to profile_continue. 1564 test_method_data_pointer(profile_continue); 1565 1566 // Update the default case count 1567 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1568 scratch1, scratch2); 1569 1570 // The method data pointer needs to be updated. 1571 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()), 1572 scratch1); 1573 1574 bind (profile_continue); 1575 } 1576 } 1577 1578 // Count the index'th case of a switch construct. 1579 void InterpreterMacroAssembler::profile_switch_case(Register index, 1580 Register scratch1, 1581 Register scratch2, 1582 Register scratch3) { 1583 if (ProfileInterpreter) { 1584 assert_different_registers(index, scratch1, scratch2, scratch3); 1585 Label profile_continue; 1586 1587 // If no method data exists, go to profile_continue. 1588 test_method_data_pointer(profile_continue); 1589 1590 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes(). 1591 li(scratch3, in_bytes(MultiBranchData::case_array_offset())); 1592 1593 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works"); 1594 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1595 add(scratch1, scratch1, scratch3); 1596 1597 // Update the case count. 1598 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3); 1599 1600 // The method data pointer needs to be updated. 1601 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2); 1602 1603 bind (profile_continue); 1604 } 1605 } 1606 1607 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) { 1608 if (ProfileInterpreter) { 1609 assert_different_registers(Rscratch1, Rscratch2); 1610 Label profile_continue; 1611 1612 // If no method data exists, go to profile_continue. 1613 test_method_data_pointer(profile_continue); 1614 1615 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1); 1616 1617 // The method data pointer needs to be updated. 1618 int mdp_delta = in_bytes(BitData::bit_data_size()); 1619 if (TypeProfileCasts) { 1620 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1621 } 1622 update_mdp_by_constant(mdp_delta); 1623 1624 bind (profile_continue); 1625 } 1626 } 1627 1628 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver, 1629 Register Rscratch1, Register Rscratch2, 1630 bool is_virtual_call) { 1631 assert(ProfileInterpreter, "must be profiling"); 1632 assert_different_registers(Rreceiver, Rscratch1, Rscratch2); 1633 1634 Label done; 1635 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call); 1636 bind (done); 1637 } 1638 1639 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1640 Register receiver, Register scratch1, Register scratch2, 1641 int start_row, Label& done, bool is_virtual_call) { 1642 if (TypeProfileWidth == 0) { 1643 if (is_virtual_call) { 1644 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1645 } 1646 return; 1647 } 1648 1649 int last_row = VirtualCallData::row_limit() - 1; 1650 assert(start_row <= last_row, "must be work left to do"); 1651 // Test this row for both the receiver and for null. 1652 // Take any of three different outcomes: 1653 // 1. found receiver => increment count and goto done 1654 // 2. found null => keep looking for case 1, maybe allocate this cell 1655 // 3. found something else => keep looking for cases 1 and 2 1656 // Case 3 is handled by a recursive call. 1657 for (int row = start_row; row <= last_row; row++) { 1658 Label next_test; 1659 bool test_for_null_also = (row == start_row); 1660 1661 // See if the receiver is receiver[n]. 1662 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1663 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1); 1664 // delayed()->tst(scratch); 1665 1666 // The receiver is receiver[n]. Increment count[n]. 1667 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1668 increment_mdp_data_at(count_offset, scratch1, scratch2); 1669 b(done); 1670 bind(next_test); 1671 1672 if (test_for_null_also) { 1673 Label found_null; 1674 // Failed the equality check on receiver[n]... Test for null. 1675 if (start_row == last_row) { 1676 // The only thing left to do is handle the null case. 1677 if (is_virtual_call) { 1678 // Scratch1 contains test_out from test_mdp_data_at. 1679 cmpdi(CCR0, scratch1, 0); 1680 beq(CCR0, found_null); 1681 // Receiver did not match any saved receiver and there is no empty row for it. 1682 // Increment total counter to indicate polymorphic case. 1683 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1684 b(done); 1685 bind(found_null); 1686 } else { 1687 cmpdi(CCR0, scratch1, 0); 1688 bne(CCR0, done); 1689 } 1690 break; 1691 } 1692 // Since null is rare, make it be the branch-taken case. 1693 cmpdi(CCR0, scratch1, 0); 1694 beq(CCR0, found_null); 1695 1696 // Put all the "Case 3" tests here. 1697 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call); 1698 1699 // Found a null. Keep searching for a matching receiver, 1700 // but remember that this is an empty (unused) slot. 1701 bind(found_null); 1702 } 1703 } 1704 1705 // In the fall-through case, we found no matching receiver, but we 1706 // observed the receiver[start_row] is NULL. 1707 1708 // Fill in the receiver field and increment the count. 1709 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1710 set_mdp_data_at(recvr_offset, receiver); 1711 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1712 li(scratch1, DataLayout::counter_increment); 1713 set_mdp_data_at(count_offset, scratch1); 1714 if (start_row > 0) { 1715 b(done); 1716 } 1717 } 1718 1719 // Argument and return type profilig. 1720 // kills: tmp, tmp2, R0, CR0, CR1 1721 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base, 1722 RegisterOrConstant mdo_addr_offs, 1723 Register tmp, Register tmp2) { 1724 Label do_nothing, do_update; 1725 1726 // tmp2 = obj is allowed 1727 assert_different_registers(obj, mdo_addr_base, tmp, R0); 1728 assert_different_registers(tmp2, mdo_addr_base, tmp, R0); 1729 const Register klass = tmp2; 1730 1731 verify_oop(obj); 1732 1733 ld(tmp, mdo_addr_offs, mdo_addr_base); 1734 1735 // Set null_seen if obj is 0. 1736 cmpdi(CCR0, obj, 0); 1737 ori(R0, tmp, TypeEntries::null_seen); 1738 beq(CCR0, do_update); 1739 1740 load_klass(klass, obj); 1741 1742 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask)); 1743 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask); 1744 cmpd(CCR1, R0, klass); 1745 // Klass seen before, nothing to do (regardless of unknown bit). 1746 //beq(CCR1, do_nothing); 1747 1748 andi_(R0, klass, TypeEntries::type_unknown); 1749 // Already unknown. Nothing to do anymore. 1750 //bne(CCR0, do_nothing); 1751 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne 1752 beq(CCR0, do_nothing); 1753 1754 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask)); 1755 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1756 beq(CCR0, do_update); // First time here. Set profile type. 1757 1758 // Different than before. Cannot keep accurate profile. 1759 ori(R0, tmp, TypeEntries::type_unknown); 1760 1761 bind(do_update); 1762 // update profile 1763 std(R0, mdo_addr_offs, mdo_addr_base); 1764 1765 align(32, 12); 1766 bind(do_nothing); 1767 } 1768 1769 void InterpreterMacroAssembler::profile_arguments_type(Register callee, 1770 Register tmp1, Register tmp2, 1771 bool is_virtual) { 1772 if (!ProfileInterpreter) { 1773 return; 1774 } 1775 1776 assert_different_registers(callee, tmp1, tmp2, R28_mdx); 1777 1778 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1779 Label profile_continue; 1780 1781 test_method_data_pointer(profile_continue); 1782 1783 int off_to_start = is_virtual ? 1784 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1785 1786 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx); 1787 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1788 bne(CCR0, profile_continue); 1789 1790 if (MethodData::profile_arguments()) { 1791 Label done; 1792 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1793 add(R28_mdx, off_to_args, R28_mdx); 1794 1795 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1796 if (i > 0 || MethodData::profile_return()) { 1797 // If return value type is profiled we may have no argument to profile. 1798 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1799 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count()); 1800 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count()); 1801 blt(CCR0, done); 1802 } 1803 ld(tmp1, in_bytes(Method::const_offset()), callee); 1804 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1); 1805 // Stack offset o (zero based) from the start of the argument 1806 // list, for n arguments translates into offset n - o - 1 from 1807 // the end of the argument list. But there's an extra slot at 1808 // the top of the stack. So the offset is n - o from Lesp. 1809 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx); 1810 subf(tmp1, tmp2, tmp1); 1811 1812 sldi(tmp1, tmp1, Interpreter::logStackElementSize); 1813 ldx(tmp1, tmp1, R15_esp); 1814 1815 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1); 1816 1817 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1818 addi(R28_mdx, R28_mdx, to_add); 1819 off_to_args += to_add; 1820 } 1821 1822 if (MethodData::profile_return()) { 1823 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1824 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1825 } 1826 1827 bind(done); 1828 1829 if (MethodData::profile_return()) { 1830 // We're right after the type profile for the last 1831 // argument. tmp1 is the number of cells left in the 1832 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1833 // if there's a return to profile. 1834 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), 1835 "can't move past ret type"); 1836 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size)); 1837 add(R28_mdx, tmp1, R28_mdx); 1838 } 1839 } else { 1840 assert(MethodData::profile_return(), "either profile call args or call ret"); 1841 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 1842 } 1843 1844 // Mdp points right after the end of the 1845 // CallTypeData/VirtualCallTypeData, right after the cells for the 1846 // return value type if there's one. 1847 align(32, 12); 1848 bind(profile_continue); 1849 } 1850 } 1851 1852 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 1853 assert_different_registers(ret, tmp1, tmp2); 1854 if (ProfileInterpreter && MethodData::profile_return()) { 1855 Label profile_continue; 1856 1857 test_method_data_pointer(profile_continue); 1858 1859 if (MethodData::profile_return_jsr292_only()) { 1860 // If we don't profile all invoke bytecodes we must make sure 1861 // it's a bytecode we indeed profile. We can't go back to the 1862 // begining of the ProfileData we intend to update to check its 1863 // type because we're right after it and we don't known its 1864 // length. 1865 lbz(tmp1, 0, R14_bcp); 1866 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method); 1867 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic); 1868 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle); 1869 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1870 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm); 1871 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1872 bne(CCR0, profile_continue); 1873 } 1874 1875 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2); 1876 1877 align(32, 12); 1878 bind(profile_continue); 1879 } 1880 } 1881 1882 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, 1883 Register tmp3, Register tmp4) { 1884 if (ProfileInterpreter && MethodData::profile_parameters()) { 1885 Label profile_continue, done; 1886 1887 test_method_data_pointer(profile_continue); 1888 1889 // Load the offset of the area within the MDO used for 1890 // parameters. If it's negative we're not profiling any parameters. 1891 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx); 1892 cmpwi(CCR0, tmp1, 0); 1893 blt(CCR0, profile_continue); 1894 1895 // Compute a pointer to the area for parameters from the offset 1896 // and move the pointer to the slot for the last 1897 // parameters. Collect profiling from last parameter down. 1898 // mdo start + parameters offset + array length - 1 1899 1900 // Pointer to the parameter area in the MDO. 1901 const Register mdp = tmp1; 1902 add(mdp, tmp1, R28_mdx); 1903 1904 // Offset of the current profile entry to update. 1905 const Register entry_offset = tmp2; 1906 // entry_offset = array len in number of cells 1907 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp); 1908 1909 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1910 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 1911 1912 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 1913 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size)); 1914 // entry_offset in bytes 1915 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1916 1917 Label loop; 1918 align(32, 12); 1919 bind(loop); 1920 1921 // Load offset on the stack from the slot for this parameter. 1922 ld(tmp3, entry_offset, mdp); 1923 sldi(tmp3, tmp3, Interpreter::logStackElementSize); 1924 neg(tmp3, tmp3); 1925 // Read the parameter from the local area. 1926 ldx(tmp3, tmp3, R18_locals); 1927 1928 // Make entry_offset now point to the type field for this parameter. 1929 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1930 assert(type_base > off_base, "unexpected"); 1931 addi(entry_offset, entry_offset, type_base - off_base); 1932 1933 // Profile the parameter. 1934 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3); 1935 1936 // Go to next parameter. 1937 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base); 1938 cmpdi(CCR0, entry_offset, off_base + delta); 1939 addi(entry_offset, entry_offset, -delta); 1940 bge(CCR0, loop); 1941 1942 align(32, 12); 1943 bind(profile_continue); 1944 } 1945 } 1946 1947 // Add a InterpMonitorElem to stack (see frame_sparc.hpp). 1948 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) { 1949 1950 // Very-local scratch registers. 1951 const Register esp = Rtemp1; 1952 const Register slot = Rtemp2; 1953 1954 // Extracted monitor_size. 1955 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes(); 1956 assert(Assembler::is_aligned((unsigned int)monitor_size, 1957 (unsigned int)frame::alignment_in_bytes), 1958 "size of a monitor must respect alignment of SP"); 1959 1960 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor 1961 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp 1962 1963 // Shuffle expression stack down. Recall that stack_base points 1964 // just above the new expression stack bottom. Old_tos and new_tos 1965 // are used to scan thru the old and new expression stacks. 1966 if (!stack_is_empty) { 1967 Label copy_slot, copy_slot_finished; 1968 const Register n_slots = slot; 1969 1970 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack). 1971 subf(n_slots, esp, R26_monitor); 1972 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy. 1973 assert(LogBytesPerWord == 3, "conflicts assembler instructions"); 1974 beq(CCR0, copy_slot_finished); // Nothing to copy. 1975 1976 mtctr(n_slots); 1977 1978 // loop 1979 bind(copy_slot); 1980 ld(slot, 0, esp); // Move expression stack down. 1981 std(slot, -monitor_size, esp); // distance = monitor_size 1982 addi(esp, esp, BytesPerWord); 1983 bdnz(copy_slot); 1984 1985 bind(copy_slot_finished); 1986 } 1987 1988 addi(R15_esp, R15_esp, -monitor_size); 1989 addi(R26_monitor, R26_monitor, -monitor_size); 1990 1991 // Restart interpreter 1992 } 1993 1994 // ============================================================================ 1995 // Java locals access 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 // Kills: 2000 // - Rdst_value 2001 // - Rdst_address 2002 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) { 2003 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2004 subf(Rdst_address, Rdst_address, R18_locals); 2005 lwz(Rdst_value, 0, Rdst_address); 2006 } 2007 2008 // Load a local variable at index in Rindex into register Rdst_value. 2009 // Also puts address of local into Rdst_address as a service. 2010 // Kills: 2011 // - Rdst_value 2012 // - Rdst_address 2013 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) { 2014 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2015 subf(Rdst_address, Rdst_address, R18_locals); 2016 ld(Rdst_value, -8, Rdst_address); 2017 } 2018 2019 // Load a local variable at index in Rindex into register Rdst_value. 2020 // Also puts address of local into Rdst_address as a service. 2021 // Input: 2022 // - Rindex: slot nr of local variable 2023 // Kills: 2024 // - Rdst_value 2025 // - Rdst_address 2026 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, 2027 Register Rdst_address, 2028 Register Rindex) { 2029 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2030 subf(Rdst_address, Rdst_address, R18_locals); 2031 ld(Rdst_value, 0, Rdst_address); 2032 } 2033 2034 // Load a local variable at index in Rindex into register Rdst_value. 2035 // Also puts address of local into Rdst_address as a service. 2036 // Kills: 2037 // - Rdst_value 2038 // - Rdst_address 2039 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, 2040 Register Rdst_address, 2041 Register Rindex) { 2042 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2043 subf(Rdst_address, Rdst_address, R18_locals); 2044 lfs(Rdst_value, 0, Rdst_address); 2045 } 2046 2047 // Load a local variable at index in Rindex into register Rdst_value. 2048 // Also puts address of local into Rdst_address as a service. 2049 // Kills: 2050 // - Rdst_value 2051 // - Rdst_address 2052 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, 2053 Register Rdst_address, 2054 Register Rindex) { 2055 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2056 subf(Rdst_address, Rdst_address, R18_locals); 2057 lfd(Rdst_value, -8, Rdst_address); 2058 } 2059 2060 // Store an int value at local variable slot Rindex. 2061 // Kills: 2062 // - Rindex 2063 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) { 2064 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2065 subf(Rindex, Rindex, R18_locals); 2066 stw(Rvalue, 0, Rindex); 2067 } 2068 2069 // Store a long value at local variable slot Rindex. 2070 // Kills: 2071 // - Rindex 2072 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) { 2073 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2074 subf(Rindex, Rindex, R18_locals); 2075 std(Rvalue, -8, Rindex); 2076 } 2077 2078 // Store an oop value at local variable slot Rindex. 2079 // Kills: 2080 // - Rindex 2081 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) { 2082 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2083 subf(Rindex, Rindex, R18_locals); 2084 std(Rvalue, 0, Rindex); 2085 } 2086 2087 // Store an int value at local variable slot Rindex. 2088 // Kills: 2089 // - Rindex 2090 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) { 2091 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2092 subf(Rindex, Rindex, R18_locals); 2093 stfs(Rvalue, 0, Rindex); 2094 } 2095 2096 // Store an int value at local variable slot Rindex. 2097 // Kills: 2098 // - Rindex 2099 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) { 2100 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2101 subf(Rindex, Rindex, R18_locals); 2102 stfd(Rvalue, -8, Rindex); 2103 } 2104 2105 // Read pending exception from thread and jump to interpreter. 2106 // Throw exception entry if one if pending. Fall through otherwise. 2107 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) { 2108 assert_different_registers(Rscratch1, Rscratch2, R3); 2109 Register Rexception = Rscratch1; 2110 Register Rtmp = Rscratch2; 2111 Label Ldone; 2112 // Get pending exception oop. 2113 ld(Rexception, thread_(pending_exception)); 2114 cmpdi(CCR0, Rexception, 0); 2115 beq(CCR0, Ldone); 2116 li(Rtmp, 0); 2117 mr_if_needed(R3, Rexception); 2118 std(Rtmp, thread_(pending_exception)); // Clear exception in thread 2119 if (Interpreter::rethrow_exception_entry() != NULL) { 2120 // Already got entry address. 2121 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry()); 2122 } else { 2123 // Dynamically load entry address. 2124 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true); 2125 ld(Rtmp, simm16_rest, Rtmp); 2126 } 2127 mtctr(Rtmp); 2128 save_interpreter_state(Rtmp); 2129 bctr(); 2130 2131 align(32, 12); 2132 bind(Ldone); 2133 } 2134 2135 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) { 2136 save_interpreter_state(R11_scratch1); 2137 2138 MacroAssembler::call_VM(oop_result, entry_point, false); 2139 2140 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2141 2142 check_and_handle_popframe(R11_scratch1); 2143 check_and_handle_earlyret(R11_scratch1); 2144 // Now check exceptions manually. 2145 if (check_exceptions) { 2146 check_and_forward_exception(R11_scratch1, R12_scratch2); 2147 } 2148 } 2149 2150 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2151 Register arg_1, bool check_exceptions) { 2152 // ARG1 is reserved for the thread. 2153 mr_if_needed(R4_ARG2, arg_1); 2154 call_VM(oop_result, entry_point, check_exceptions); 2155 } 2156 2157 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2158 Register arg_1, Register arg_2, 2159 bool check_exceptions) { 2160 // ARG1 is reserved for the thread. 2161 mr_if_needed(R4_ARG2, arg_1); 2162 assert(arg_2 != R4_ARG2, "smashed argument"); 2163 mr_if_needed(R5_ARG3, arg_2); 2164 call_VM(oop_result, entry_point, check_exceptions); 2165 } 2166 2167 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2168 Register arg_1, Register arg_2, Register arg_3, 2169 bool check_exceptions) { 2170 // ARG1 is reserved for the thread. 2171 mr_if_needed(R4_ARG2, arg_1); 2172 assert(arg_2 != R4_ARG2, "smashed argument"); 2173 mr_if_needed(R5_ARG3, arg_2); 2174 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument"); 2175 mr_if_needed(R6_ARG4, arg_3); 2176 call_VM(oop_result, entry_point, check_exceptions); 2177 } 2178 2179 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) { 2180 ld(scratch, 0, R1_SP); 2181 std(R15_esp, _ijava_state_neg(esp), scratch); 2182 std(R14_bcp, _ijava_state_neg(bcp), scratch); 2183 std(R26_monitor, _ijava_state_neg(monitors), scratch); 2184 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); } 2185 // Other entries should be unchanged. 2186 } 2187 2188 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) { 2189 ld(scratch, 0, R1_SP); 2190 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception). 2191 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code. 2192 if (!bcp_and_mdx_only) { 2193 // Following ones are Metadata. 2194 ld(R19_method, _ijava_state_neg(method), scratch); 2195 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch); 2196 // Following ones are stack addresses and don't require reload. 2197 ld(R15_esp, _ijava_state_neg(esp), scratch); 2198 ld(R18_locals, _ijava_state_neg(locals), scratch); 2199 ld(R26_monitor, _ijava_state_neg(monitors), scratch); 2200 } 2201 #ifdef ASSERT 2202 { 2203 Label Lok; 2204 subf(R0, R1_SP, scratch); 2205 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size); 2206 bge(CCR0, Lok); 2207 stop("frame too small (restore istate)", 0x5432); 2208 bind(Lok); 2209 } 2210 { 2211 Label Lok; 2212 ld(R0, _ijava_state_neg(ijava_reserved), scratch); 2213 cmpdi(CCR0, R0, 0x5afe); 2214 beq(CCR0, Lok); 2215 stop("frame corrupted (restore istate)", 0x5afe); 2216 bind(Lok); 2217 } 2218 #endif 2219 } 2220 2221 void InterpreterMacroAssembler::get_method_counters(Register method, 2222 Register Rcounters, 2223 Label& skip) { 2224 BLOCK_COMMENT("Load and ev. allocate counter object {"); 2225 Label has_counters; 2226 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2227 cmpdi(CCR0, Rcounters, 0); 2228 bne(CCR0, has_counters); 2229 call_VM(noreg, CAST_FROM_FN_PTR(address, 2230 InterpreterRuntime::build_method_counters), method, false); 2231 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2232 cmpdi(CCR0, Rcounters, 0); 2233 beq(CCR0, skip); // No MethodCounters, OutOfMemory. 2234 BLOCK_COMMENT("} Load and ev. allocate counter object"); 2235 2236 bind(has_counters); 2237 } 2238 2239 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, 2240 Register iv_be_count, 2241 Register Rtmp_r0) { 2242 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 2243 Register invocation_count = iv_be_count; 2244 Register backedge_count = Rtmp_r0; 2245 int delta = InvocationCounter::count_increment; 2246 2247 // Load each counter in a register. 2248 // ld(inv_counter, Rtmp); 2249 // ld(be_counter, Rtmp2); 2250 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + 2251 InvocationCounter::counter_offset()); 2252 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + 2253 InvocationCounter::counter_offset()); 2254 2255 BLOCK_COMMENT("Increment profiling counters {"); 2256 2257 // Load the backedge counter. 2258 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int 2259 // Mask the backedge counter. 2260 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value); 2261 2262 // Load the invocation counter. 2263 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int 2264 // Add the delta to the invocation counter and store the result. 2265 addi(invocation_count, invocation_count, delta); 2266 // Store value. 2267 stw(invocation_count, inv_counter_offset, Rcounters); 2268 2269 // Add invocation counter + backedge counter. 2270 add(iv_be_count, backedge_count, invocation_count); 2271 2272 // Note that this macro must leave the backedge_count + invocation_count in 2273 // register iv_be_count! 2274 BLOCK_COMMENT("} Increment profiling counters"); 2275 } 2276 2277 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2278 if (state == atos) { MacroAssembler::verify_oop(reg); } 2279 } 2280 2281 // Local helper function for the verify_oop_or_return_address macro. 2282 static bool verify_return_address(Method* m, int bci) { 2283 #ifndef PRODUCT 2284 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci; 2285 // Assume it is a valid return address if it is inside m and is preceded by a jsr. 2286 if (!m->contains(pc)) return false; 2287 address jsr_pc; 2288 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2289 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2290 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2291 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2292 #endif // PRODUCT 2293 return false; 2294 } 2295 2296 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2297 if (VerifyFPU) { 2298 unimplemented("verfiyFPU"); 2299 } 2300 } 2301 2302 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2303 if (!VerifyOops) return; 2304 2305 // The VM documentation for the astore[_wide] bytecode allows 2306 // the TOS to be not only an oop but also a return address. 2307 Label test; 2308 Label skip; 2309 // See if it is an address (in the current method): 2310 2311 const int log2_bytecode_size_limit = 16; 2312 srdi_(Rtmp, reg, log2_bytecode_size_limit); 2313 bne(CCR0, test); 2314 2315 address fd = CAST_FROM_FN_PTR(address, verify_return_address); 2316 const int nbytes_save = MacroAssembler::num_volatile_regs * 8; 2317 save_volatile_gprs(R1_SP, -nbytes_save); // except R0 2318 save_LR_CR(Rtmp); // Save in old frame. 2319 push_frame_reg_args(nbytes_save, Rtmp); 2320 2321 load_const_optimized(Rtmp, fd, R0); 2322 mr_if_needed(R4_ARG2, reg); 2323 mr(R3_ARG1, R19_method); 2324 call_c(Rtmp); // call C 2325 2326 pop_frame(); 2327 restore_LR_CR(Rtmp); 2328 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0 2329 b(skip); 2330 2331 // Perform a more elaborate out-of-line call. 2332 // Not an address; verify it: 2333 bind(test); 2334 verify_oop(reg); 2335 bind(skip); 2336 } 2337 2338 // Inline assembly for: 2339 // 2340 // if (thread is in interp_only_mode) { 2341 // InterpreterRuntime::post_method_entry(); 2342 // } 2343 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) || 2344 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) { 2345 // SharedRuntime::jvmpi_method_entry(method, receiver); 2346 // } 2347 void InterpreterMacroAssembler::notify_method_entry() { 2348 // JVMTI 2349 // Whenever JVMTI puts a thread in interp_only_mode, method 2350 // entry/exit events are sent for that thread to track stack 2351 // depth. If it is possible to enter interp_only_mode we add 2352 // the code to check if the event should be sent. 2353 if (JvmtiExport::can_post_interpreter_events()) { 2354 Label jvmti_post_done; 2355 2356 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2357 cmpwi(CCR0, R0, 0); 2358 beq(CCR0, jvmti_post_done); 2359 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), 2360 /*check_exceptions=*/true); 2361 2362 bind(jvmti_post_done); 2363 } 2364 } 2365 2366 // Inline assembly for: 2367 // 2368 // if (thread is in interp_only_mode) { 2369 // // save result 2370 // InterpreterRuntime::post_method_exit(); 2371 // // restore result 2372 // } 2373 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) { 2374 // // save result 2375 // SharedRuntime::jvmpi_method_exit(); 2376 // // restore result 2377 // } 2378 // 2379 // Native methods have their result stored in d_tmp and l_tmp. 2380 // Java methods have their result stored in the expression stack. 2381 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state, 2382 NotifyMethodExitMode mode, bool check_exceptions) { 2383 // JVMTI 2384 // Whenever JVMTI puts a thread in interp_only_mode, method 2385 // entry/exit events are sent for that thread to track stack 2386 // depth. If it is possible to enter interp_only_mode we add 2387 // the code to check if the event should be sent. 2388 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2389 Label jvmti_post_done; 2390 2391 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2392 cmpwi(CCR0, R0, 0); 2393 beq(CCR0, jvmti_post_done); 2394 if (!is_native_method) { push(state); } // Expose tos to GC. 2395 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), 2396 /*check_exceptions=*/check_exceptions); 2397 if (!is_native_method) { pop(state); } 2398 2399 align(32, 12); 2400 bind(jvmti_post_done); 2401 } 2402 2403 // Dtrace support not implemented. 2404 } 2405