1 /* 2 * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved. 3 * Copyright 2012, 2014 SAP AG. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 27 #include "precompiled.hpp" 28 #include "asm/macroAssembler.inline.hpp" 29 #include "interp_masm_ppc_64.hpp" 30 #include "interpreter/interpreterRuntime.hpp" 31 #include "prims/jvmtiThreadState.hpp" 32 33 #ifdef PRODUCT 34 #define BLOCK_COMMENT(str) // nothing 35 #else 36 #define BLOCK_COMMENT(str) block_comment(str) 37 #endif 38 39 void InterpreterMacroAssembler::null_check_throw(Register a, int offset, Register temp_reg) { 40 #ifdef CC_INTERP 41 address exception_entry = StubRoutines::throw_NullPointerException_at_call_entry(); 42 #else 43 address exception_entry = Interpreter::throw_NullPointerException_entry(); 44 #endif 45 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry); 46 } 47 48 void InterpreterMacroAssembler::branch_to_entry(address entry, Register Rscratch) { 49 assert(entry, "Entry must have been generated by now"); 50 if (is_within_range_of_b(entry, pc())) { 51 b(entry); 52 } else { 53 load_const_optimized(Rscratch, entry, R0); 54 mtctr(Rscratch); 55 bctr(); 56 } 57 } 58 59 #ifndef CC_INTERP 60 61 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { 62 Register bytecode = R12_scratch2; 63 if (bcp_incr != 0) { 64 lbzu(bytecode, bcp_incr, R14_bcp); 65 } else { 66 lbz(bytecode, 0, R14_bcp); 67 } 68 69 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state)); 70 } 71 72 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 73 // Load current bytecode. 74 Register bytecode = R12_scratch2; 75 lbz(bytecode, 0, R14_bcp); 76 dispatch_Lbyte_code(state, bytecode, table); 77 } 78 79 // Dispatch code executed in the prolog of a bytecode which does not do it's 80 // own dispatch. The dispatch address is computed and placed in R24_dispatch_addr. 81 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 82 Register bytecode = R12_scratch2; 83 lbz(bytecode, bcp_incr, R14_bcp); 84 85 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state)); 86 87 sldi(bytecode, bytecode, LogBytesPerWord); 88 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode); 89 } 90 91 // Dispatch code executed in the epilog of a bytecode which does not do it's 92 // own dispatch. The dispatch address in R24_dispatch_addr is used for the 93 // dispatch. 94 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 95 mtctr(R24_dispatch_addr); 96 addi(R14_bcp, R14_bcp, bcp_incr); 97 bctr(); 98 } 99 100 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 101 assert(scratch_reg != R0, "can't use R0 as scratch_reg here"); 102 if (JvmtiExport::can_pop_frame()) { 103 Label L; 104 105 // Check the "pending popframe condition" flag in the current thread. 106 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 107 108 // Initiate popframe handling only if it is not already being 109 // processed. If the flag has the popframe_processing bit set, it 110 // means that this code is called *during* popframe handling - we 111 // don't want to reenter. 112 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit); 113 beq(CCR0, L); 114 115 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit); 116 bne(CCR0, L); 117 118 // Call the Interpreter::remove_activation_preserving_args_entry() 119 // func to get the address of the same-named entrypoint in the 120 // generated interpreter code. 121 #if defined(ABI_ELFv2) 122 call_c(CAST_FROM_FN_PTR(address, 123 Interpreter::remove_activation_preserving_args_entry), 124 relocInfo::none); 125 #else 126 call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, 127 Interpreter::remove_activation_preserving_args_entry), 128 relocInfo::none); 129 #endif 130 131 // Jump to Interpreter::_remove_activation_preserving_args_entry. 132 mtctr(R3_RET); 133 bctr(); 134 135 align(32, 12); 136 bind(L); 137 } 138 } 139 140 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 141 const Register Rthr_state_addr = scratch_reg; 142 if (JvmtiExport::can_force_early_return()) { 143 Label Lno_early_ret; 144 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 145 cmpdi(CCR0, Rthr_state_addr, 0); 146 beq(CCR0, Lno_early_ret); 147 148 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr); 149 cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending); 150 bne(CCR0, Lno_early_ret); 151 152 // Jump to Interpreter::_earlyret_entry. 153 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr); 154 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry)); 155 mtlr(R3_RET); 156 blr(); 157 158 align(32, 12); 159 bind(Lno_early_ret); 160 } 161 } 162 163 void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) { 164 const Register RjvmtiState = Rscratch1; 165 const Register Rscratch2 = R0; 166 167 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 168 li(Rscratch2, 0); 169 170 switch (state) { 171 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 172 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 173 break; 174 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 175 break; 176 case btos: // fall through 177 case ctos: // fall through 178 case stos: // fall through 179 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 180 break; 181 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 182 break; 183 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 184 break; 185 case vtos: break; 186 default : ShouldNotReachHere(); 187 } 188 189 // Clean up tos value in the jvmti thread state. 190 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 191 // Set tos state field to illegal value. 192 li(Rscratch2, ilgl); 193 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState); 194 } 195 196 // Common code to dispatch and dispatch_only. 197 // Dispatch value in Lbyte_code and increment Lbcp. 198 199 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) { 200 address table_base = (address)Interpreter::dispatch_table((TosState)0); 201 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base; 202 if (is_simm16(table_offs)) { 203 addi(dst, R25_templateTableBase, (int)table_offs); 204 } else { 205 load_const_optimized(dst, table, R0); 206 } 207 } 208 209 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode, address* table, bool verify) { 210 if (verify) { 211 unimplemented("dispatch_Lbyte_code: verify"); // See Sparc Implementation to implement this 212 } 213 214 #ifdef FAST_DISPATCH 215 unimplemented("dispatch_Lbyte_code FAST_DISPATCH"); 216 #else 217 assert_different_registers(bytecode, R11_scratch1); 218 219 // Calc dispatch table address. 220 load_dispatch_table(R11_scratch1, table); 221 222 sldi(R12_scratch2, bytecode, LogBytesPerWord); 223 ldx(R11_scratch1, R11_scratch1, R12_scratch2); 224 225 // Jump off! 226 mtctr(R11_scratch1); 227 bctr(); 228 #endif 229 } 230 231 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) { 232 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize); 233 ldx(Rrecv_dst, Rrecv_dst, R15_esp); 234 } 235 236 // helpers for expression stack 237 238 void InterpreterMacroAssembler::pop_i(Register r) { 239 lwzu(r, Interpreter::stackElementSize, R15_esp); 240 } 241 242 void InterpreterMacroAssembler::pop_ptr(Register r) { 243 ldu(r, Interpreter::stackElementSize, R15_esp); 244 } 245 246 void InterpreterMacroAssembler::pop_l(Register r) { 247 ld(r, Interpreter::stackElementSize, R15_esp); 248 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 249 } 250 251 void InterpreterMacroAssembler::pop_f(FloatRegister f) { 252 lfsu(f, Interpreter::stackElementSize, R15_esp); 253 } 254 255 void InterpreterMacroAssembler::pop_d(FloatRegister f) { 256 lfd(f, Interpreter::stackElementSize, R15_esp); 257 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 258 } 259 260 void InterpreterMacroAssembler::push_i(Register r) { 261 stw(r, 0, R15_esp); 262 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 263 } 264 265 void InterpreterMacroAssembler::push_ptr(Register r) { 266 std(r, 0, R15_esp); 267 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 268 } 269 270 void InterpreterMacroAssembler::push_l(Register r) { 271 std(r, - Interpreter::stackElementSize, R15_esp); 272 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 273 } 274 275 void InterpreterMacroAssembler::push_f(FloatRegister f) { 276 stfs(f, 0, R15_esp); 277 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 278 } 279 280 void InterpreterMacroAssembler::push_d(FloatRegister f) { 281 stfd(f, - Interpreter::stackElementSize, R15_esp); 282 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 283 } 284 285 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) { 286 std(first, 0, R15_esp); 287 std(second, -Interpreter::stackElementSize, R15_esp); 288 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 289 } 290 291 void InterpreterMacroAssembler::push_l_pop_d(Register l, FloatRegister d) { 292 std(l, 0, R15_esp); 293 lfd(d, 0, R15_esp); 294 } 295 296 void InterpreterMacroAssembler::push_d_pop_l(FloatRegister d, Register l) { 297 stfd(d, 0, R15_esp); 298 ld(l, 0, R15_esp); 299 } 300 301 void InterpreterMacroAssembler::push(TosState state) { 302 switch (state) { 303 case atos: push_ptr(); break; 304 case btos: 305 case ctos: 306 case stos: 307 case itos: push_i(); break; 308 case ltos: push_l(); break; 309 case ftos: push_f(); break; 310 case dtos: push_d(); break; 311 case vtos: /* nothing to do */ break; 312 default : ShouldNotReachHere(); 313 } 314 } 315 316 void InterpreterMacroAssembler::pop(TosState state) { 317 switch (state) { 318 case atos: pop_ptr(); break; 319 case btos: 320 case ctos: 321 case stos: 322 case itos: pop_i(); break; 323 case ltos: pop_l(); break; 324 case ftos: pop_f(); break; 325 case dtos: pop_d(); break; 326 case vtos: /* nothing to do */ break; 327 default : ShouldNotReachHere(); 328 } 329 verify_oop(R17_tos, state); 330 } 331 332 void InterpreterMacroAssembler::empty_expression_stack() { 333 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 334 } 335 336 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset, 337 Register Rdst, 338 signedOrNot is_signed) { 339 #if defined(VM_LITTLE_ENDIAN) 340 if (bcp_offset) { 341 load_const_optimized(Rdst, bcp_offset); 342 lhbrx(Rdst, R14_bcp, Rdst); 343 } else { 344 lhbrx(Rdst, R14_bcp); 345 } 346 if (is_signed == Signed) { 347 extsh(Rdst, Rdst); 348 } 349 #else 350 // Read Java big endian format. 351 if (is_signed == Signed) { 352 lha(Rdst, bcp_offset, R14_bcp); 353 } else { 354 lhz(Rdst, bcp_offset, R14_bcp); 355 } 356 #endif 357 } 358 359 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset, 360 Register Rdst, 361 signedOrNot is_signed) { 362 #if defined(VM_LITTLE_ENDIAN) 363 if (bcp_offset) { 364 load_const_optimized(Rdst, bcp_offset); 365 lwbrx(Rdst, R14_bcp, Rdst); 366 } else { 367 lwbrx(Rdst, R14_bcp); 368 } 369 if (is_signed == Signed) { 370 extsw(Rdst, Rdst); 371 } 372 #else 373 // Read Java big endian format. 374 if (bcp_offset & 3) { // Offset unaligned? 375 load_const_optimized(Rdst, bcp_offset); 376 if (is_signed == Signed) { 377 lwax(Rdst, R14_bcp, Rdst); 378 } else { 379 lwzx(Rdst, R14_bcp, Rdst); 380 } 381 } else { 382 if (is_signed == Signed) { 383 lwa(Rdst, bcp_offset, R14_bcp); 384 } else { 385 lwz(Rdst, bcp_offset, R14_bcp); 386 } 387 } 388 #endif 389 } 390 391 392 // Load the constant pool cache index from the bytecode stream. 393 // 394 // Kills / writes: 395 // - Rdst, Rscratch 396 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, size_t index_size) { 397 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 398 // Cache index is always in the native format, courtesy of Rewriter. 399 if (index_size == sizeof(u2)) { 400 lhz(Rdst, bcp_offset, R14_bcp); 401 } else if (index_size == sizeof(u4)) { 402 if (bcp_offset & 3) { 403 load_const_optimized(Rdst, bcp_offset); 404 lwax(Rdst, R14_bcp, Rdst); 405 } else { 406 lwa(Rdst, bcp_offset, R14_bcp); 407 } 408 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 409 nand(Rdst, Rdst, Rdst); // convert to plain index 410 } else if (index_size == sizeof(u1)) { 411 lbz(Rdst, bcp_offset, R14_bcp); 412 } else { 413 ShouldNotReachHere(); 414 } 415 // Rdst now contains cp cache index. 416 } 417 418 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset, 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) { 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); 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 void InterpreterMacroAssembler::generate_stack_overflow_check_with_compare_and_throw(Register Rmem_frame_size, Register Rscratch1) { 485 Label done; 486 sub(Rmem_frame_size, R1_SP, Rmem_frame_size); 487 ld(Rscratch1, thread_(stack_overflow_limit)); 488 cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1); 489 bgt(CCR0/*is_stack_overflow*/, done); 490 491 // Load target address of the runtime stub. 492 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order"); 493 load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0); 494 mtctr(Rscratch1); 495 // Restore caller_sp. 496 #ifdef ASSERT 497 ld(Rscratch1, 0, R1_SP); 498 ld(R0, 0, R21_sender_SP); 499 cmpd(CCR0, R0, Rscratch1); 500 asm_assert_eq("backlink", 0x547); 501 #endif // ASSERT 502 mr(R1_SP, R21_sender_SP); 503 bctr(); 504 505 align(32, 12); 506 bind(done); 507 } 508 509 // Separate these two to allow for delay slot in middle. 510 // These are used to do a test and full jump to exception-throwing code. 511 512 // Check that index is in range for array, then shift index by index_shift, 513 // and put arrayOop + shifted_index into res. 514 // Note: res is still shy of address by array offset into object. 515 516 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, int index_shift, Register Rtmp, Register Rres) { 517 // Check that index is in range for array, then shift index by index_shift, 518 // and put arrayOop + shifted_index into res. 519 // Note: res is still shy of address by array offset into object. 520 // Kills: 521 // - Rindex 522 // Writes: 523 // - Rres: Address that corresponds to the array index if check was successful. 524 verify_oop(Rarray); 525 const Register Rlength = R0; 526 const Register RsxtIndex = Rtmp; 527 Label LisNull, LnotOOR; 528 529 // Array nullcheck 530 if (!ImplicitNullChecks) { 531 cmpdi(CCR0, Rarray, 0); 532 beq(CCR0, LisNull); 533 } else { 534 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex); 535 } 536 537 // Rindex might contain garbage in upper bits (remember that we don't sign extend 538 // during integer arithmetic operations). So kill them and put value into same register 539 // where ArrayIndexOutOfBounds would expect the index in. 540 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit 541 542 // Index check 543 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray); 544 cmplw(CCR0, Rindex, Rlength); 545 sldi(RsxtIndex, RsxtIndex, index_shift); 546 blt(CCR0, LnotOOR); 547 // Index should be in R17_tos, array should be in R4_ARG2. 548 mr(R17_tos, Rindex); 549 mr(R4_ARG2, Rarray); 550 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry); 551 mtctr(Rtmp); 552 bctr(); 553 554 if (!ImplicitNullChecks) { 555 bind(LisNull); 556 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry); 557 mtctr(Rtmp); 558 bctr(); 559 } 560 561 align(32, 16); 562 bind(LnotOOR); 563 564 // Calc address 565 add(Rres, RsxtIndex, Rarray); 566 } 567 568 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { 569 // pop array 570 pop_ptr(array); 571 572 // check array 573 index_check_without_pop(array, index, index_shift, tmp, res); 574 } 575 576 void InterpreterMacroAssembler::get_const(Register Rdst) { 577 ld(Rdst, in_bytes(Method::const_offset()), R19_method); 578 } 579 580 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 581 get_const(Rdst); 582 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 583 } 584 585 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 586 get_constant_pool(Rdst); 587 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 588 } 589 590 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 591 get_constant_pool(Rcpool); 592 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool); 593 } 594 595 // Unlock if synchronized method. 596 // 597 // Unlock the receiver if this is a synchronized method. 598 // Unlock any Java monitors from synchronized blocks. 599 // 600 // If there are locked Java monitors 601 // If throw_monitor_exception 602 // throws IllegalMonitorStateException 603 // Else if install_monitor_exception 604 // installs IllegalMonitorStateException 605 // Else 606 // no error processing 607 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 608 bool throw_monitor_exception, 609 bool install_monitor_exception) { 610 Label Lunlocked, Lno_unlock; 611 { 612 Register Rdo_not_unlock_flag = R11_scratch1; 613 Register Raccess_flags = R12_scratch2; 614 615 // Check if synchronized method or unlocking prevented by 616 // JavaThread::do_not_unlock_if_synchronized flag. 617 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 618 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method); 619 li(R0, 0); 620 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag 621 622 push(state); 623 624 // Skip if we don't have to unlock. 625 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0. 626 beq(CCR0, Lunlocked); 627 628 cmpwi(CCR0, Rdo_not_unlock_flag, 0); 629 bne(CCR0, Lno_unlock); 630 } 631 632 // Unlock 633 { 634 Register Rmonitor_base = R11_scratch1; 635 636 Label Lunlock; 637 // If it's still locked, everything is ok, unlock it. 638 ld(Rmonitor_base, 0, R1_SP); 639 addi(Rmonitor_base, Rmonitor_base, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base 640 641 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base); 642 cmpdi(CCR0, R0, 0); 643 bne(CCR0, Lunlock); 644 645 // If it's already unlocked, throw exception. 646 if (throw_monitor_exception) { 647 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 648 should_not_reach_here(); 649 } else { 650 if (install_monitor_exception) { 651 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 652 b(Lunlocked); 653 } 654 } 655 656 bind(Lunlock); 657 unlock_object(Rmonitor_base); 658 } 659 660 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not. 661 bind(Lunlocked); 662 { 663 Label Lexception, Lrestart; 664 Register Rcurrent_obj_addr = R11_scratch1; 665 const int delta = frame::interpreter_frame_monitor_size_in_bytes(); 666 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords"); 667 668 bind(Lrestart); 669 // Set up search loop: Calc num of iterations. 670 { 671 Register Riterations = R12_scratch2; 672 Register Rmonitor_base = Rcurrent_obj_addr; 673 ld(Rmonitor_base, 0, R1_SP); 674 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base 675 676 subf_(Riterations, R26_monitor, Rmonitor_base); 677 ble(CCR0, Lno_unlock); 678 679 addi(Rcurrent_obj_addr, Rmonitor_base, BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes()); 680 // Check if any monitor is on stack, bail out if not 681 srdi(Riterations, Riterations, exact_log2(delta)); 682 mtctr(Riterations); 683 } 684 685 // The search loop: Look for locked monitors. 686 { 687 const Register Rcurrent_obj = R0; 688 Label Lloop; 689 690 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 691 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 692 bind(Lloop); 693 694 // Check if current entry is used. 695 cmpdi(CCR0, Rcurrent_obj, 0); 696 bne(CCR0, Lexception); 697 // Preload next iteration's compare value. 698 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 699 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 700 bdnz(Lloop); 701 } 702 // Fell through: Everything's unlocked => finish. 703 b(Lno_unlock); 704 705 // An object is still locked => need to throw exception. 706 bind(Lexception); 707 if (throw_monitor_exception) { 708 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 709 should_not_reach_here(); 710 } else { 711 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 712 // Unlock does not block, so don't have to worry about the frame. 713 Register Rmonitor_addr = R11_scratch1; 714 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta); 715 unlock_object(Rmonitor_addr); 716 if (install_monitor_exception) { 717 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 718 } 719 b(Lrestart); 720 } 721 } 722 723 align(32, 12); 724 bind(Lno_unlock); 725 pop(state); 726 } 727 728 // Support function for remove_activation & Co. 729 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, Register Rscratch1, Register Rscratch2) { 730 // Pop interpreter frame. 731 ld(Rscratch1, 0, R1_SP); // *SP 732 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp 733 ld(Rscratch2, 0, Rscratch1); // **SP 734 #ifdef ASSERT 735 { 736 Label Lok; 737 ld(R0, _ijava_state_neg(ijava_reserved), Rscratch1); 738 cmpdi(CCR0, R0, 0x5afe); 739 beq(CCR0, Lok); 740 stop("frame corrupted (remove activation)", 0x5afe); 741 bind(Lok); 742 } 743 #endif 744 if (return_pc!=noreg) { 745 ld(return_pc, _abi(lr), Rscratch1); // LR 746 } 747 748 // Merge top frames. 749 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP 750 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP 751 } 752 753 // Remove activation. 754 // 755 // Unlock the receiver if this is a synchronized method. 756 // Unlock any Java monitors from synchronized blocks. 757 // Remove the activation from the stack. 758 // 759 // If there are locked Java monitors 760 // If throw_monitor_exception 761 // throws IllegalMonitorStateException 762 // Else if install_monitor_exception 763 // installs IllegalMonitorStateException 764 // Else 765 // no error processing 766 void InterpreterMacroAssembler::remove_activation(TosState state, 767 bool throw_monitor_exception, 768 bool install_monitor_exception) { 769 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 770 771 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 772 notify_method_exit(false, state, NotifyJVMTI, true); 773 774 verify_oop(R17_tos, state); 775 verify_thread(); 776 777 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 778 mtlr(R0); 779 } 780 781 #endif // !CC_INTERP 782 783 // Lock object 784 // 785 // Registers alive 786 // monitor - Address of the BasicObjectLock to be used for locking, 787 // which must be initialized with the object to lock. 788 // object - Address of the object to be locked. 789 // 790 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 791 if (UseHeavyMonitors) { 792 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 793 monitor, /*check_for_exceptions=*/true CC_INTERP_ONLY(&& false)); 794 } else { 795 // template code: 796 // 797 // markOop displaced_header = obj->mark().set_unlocked(); 798 // monitor->lock()->set_displaced_header(displaced_header); 799 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 800 // // We stored the monitor address into the object's mark word. 801 // } else if (THREAD->is_lock_owned((address)displaced_header)) 802 // // Simple recursive case. 803 // monitor->lock()->set_displaced_header(NULL); 804 // } else { 805 // // Slow path. 806 // InterpreterRuntime::monitorenter(THREAD, monitor); 807 // } 808 809 const Register displaced_header = R7_ARG5; 810 const Register object_mark_addr = R8_ARG6; 811 const Register current_header = R9_ARG7; 812 const Register tmp = R10_ARG8; 813 814 Label done; 815 Label cas_failed, slow_case; 816 817 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp); 818 819 // markOop displaced_header = obj->mark().set_unlocked(); 820 821 // Load markOop from object into displaced_header. 822 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object); 823 824 if (UseBiasedLocking) { 825 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case); 826 } 827 828 // Set displaced_header to be (markOop of object | UNLOCK_VALUE). 829 ori(displaced_header, displaced_header, markOopDesc::unlocked_value); 830 831 // monitor->lock()->set_displaced_header(displaced_header); 832 833 // Initialize the box (Must happen before we update the object mark!). 834 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 835 BasicLock::displaced_header_offset_in_bytes(), monitor); 836 837 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 838 839 // Store stack address of the BasicObjectLock (this is monitor) into object. 840 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 841 842 // Must fence, otherwise, preceding store(s) may float below cmpxchg. 843 // CmpxchgX sets CCR0 to cmpX(current, displaced). 844 fence(); // TODO: replace by MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq ? 845 cmpxchgd(/*flag=*/CCR0, 846 /*current_value=*/current_header, 847 /*compare_value=*/displaced_header, /*exchange_value=*/monitor, 848 /*where=*/object_mark_addr, 849 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq, 850 MacroAssembler::cmpxchgx_hint_acquire_lock(), 851 noreg, 852 &cas_failed); 853 854 // If the compare-and-exchange succeeded, then we found an unlocked 855 // object and we have now locked it. 856 b(done); 857 bind(cas_failed); 858 859 // } else if (THREAD->is_lock_owned((address)displaced_header)) 860 // // Simple recursive case. 861 // monitor->lock()->set_displaced_header(NULL); 862 863 // We did not see an unlocked object so try the fast recursive case. 864 865 // Check if owner is self by comparing the value in the markOop of object 866 // (current_header) with the stack pointer. 867 sub(current_header, current_header, R1_SP); 868 869 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 870 load_const_optimized(tmp, 871 (address) (~(os::vm_page_size()-1) | 872 markOopDesc::lock_mask_in_place)); 873 874 and_(R0/*==0?*/, current_header, tmp); 875 // If condition is true we are done and hence we can store 0 in the displaced 876 // header indicating it is a recursive lock. 877 bne(CCR0, slow_case); 878 release(); 879 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() + 880 BasicLock::displaced_header_offset_in_bytes(), monitor); 881 b(done); 882 883 // } else { 884 // // Slow path. 885 // InterpreterRuntime::monitorenter(THREAD, monitor); 886 887 // None of the above fast optimizations worked so we have to get into the 888 // slow case of monitor enter. 889 bind(slow_case); 890 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 891 monitor, /*check_for_exceptions=*/true CC_INTERP_ONLY(&& false)); 892 // } 893 align(32, 12); 894 bind(done); 895 } 896 } 897 898 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 899 // 900 // Registers alive 901 // monitor - Address of the BasicObjectLock to be used for locking, 902 // which must be initialized with the object to lock. 903 // 904 // Throw IllegalMonitorException if object is not locked by current thread. 905 void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) { 906 if (UseHeavyMonitors) { 907 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 908 monitor, check_for_exceptions CC_INTERP_ONLY(&& false)); 909 } else { 910 911 // template code: 912 // 913 // if ((displaced_header = monitor->displaced_header()) == NULL) { 914 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 915 // monitor->set_obj(NULL); 916 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 917 // // We swapped the unlocked mark in displaced_header into the object's mark word. 918 // monitor->set_obj(NULL); 919 // } else { 920 // // Slow path. 921 // InterpreterRuntime::monitorexit(THREAD, monitor); 922 // } 923 924 const Register object = R7_ARG5; 925 const Register displaced_header = R8_ARG6; 926 const Register object_mark_addr = R9_ARG7; 927 const Register current_header = R10_ARG8; 928 929 Label free_slot; 930 Label slow_case; 931 932 assert_different_registers(object, displaced_header, object_mark_addr, current_header); 933 934 if (UseBiasedLocking) { 935 // The object address from the monitor is in object. 936 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); 937 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0"); 938 biased_locking_exit(CCR0, object, displaced_header, free_slot); 939 } 940 941 // Test first if we are in the fast recursive case. 942 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 943 BasicLock::displaced_header_offset_in_bytes(), monitor); 944 945 // If the displaced header is zero, we have a recursive unlock. 946 cmpdi(CCR0, displaced_header, 0); 947 beq(CCR0, free_slot); // recursive unlock 948 949 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 950 // // We swapped the unlocked mark in displaced_header into the object's mark word. 951 // monitor->set_obj(NULL); 952 953 // If we still have a lightweight lock, unlock the object and be done. 954 955 // The object address from the monitor is in object. 956 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); } 957 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 958 959 // We have the displaced header in displaced_header. If the lock is still 960 // lightweight, it will contain the monitor address and we'll store the 961 // displaced header back into the object's mark word. 962 // CmpxchgX sets CCR0 to cmpX(current, monitor). 963 cmpxchgd(/*flag=*/CCR0, 964 /*current_value=*/current_header, 965 /*compare_value=*/monitor, /*exchange_value=*/displaced_header, 966 /*where=*/object_mark_addr, 967 MacroAssembler::MemBarRel, 968 MacroAssembler::cmpxchgx_hint_release_lock(), 969 noreg, 970 &slow_case); 971 b(free_slot); 972 973 // } else { 974 // // Slow path. 975 // InterpreterRuntime::monitorexit(THREAD, monitor); 976 977 // The lock has been converted into a heavy lock and hence 978 // we need to get into the slow case. 979 bind(slow_case); 980 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 981 monitor, check_for_exceptions CC_INTERP_ONLY(&& false)); 982 // } 983 984 Label done; 985 b(done); // Monitor register may be overwritten! Runtime has already freed the slot. 986 987 // Exchange worked, do monitor->set_obj(NULL); 988 align(32, 12); 989 bind(free_slot); 990 li(R0, 0); 991 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor); 992 bind(done); 993 } 994 } 995 996 #ifndef CC_INTERP 997 998 // Load compiled (i2c) or interpreter entry when calling from interpreted and 999 // do the call. Centralized so that all interpreter calls will do the same actions. 1000 // If jvmti single stepping is on for a thread we must not call compiled code. 1001 // 1002 // Input: 1003 // - Rtarget_method: method to call 1004 // - Rret_addr: return address 1005 // - 2 scratch regs 1006 // 1007 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, Register Rscratch1, Register Rscratch2) { 1008 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr); 1009 // Assume we want to go compiled if available. 1010 const Register Rtarget_addr = Rscratch1; 1011 const Register Rinterp_only = Rscratch2; 1012 1013 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method); 1014 1015 if (JvmtiExport::can_post_interpreter_events()) { 1016 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 1017 1018 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 1019 // compiled code in threads for which the event is enabled. Check here for 1020 // interp_only_mode if these events CAN be enabled. 1021 Label done; 1022 verify_thread(); 1023 cmpwi(CCR0, Rinterp_only, 0); 1024 beq(CCR0, done); 1025 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method); 1026 align(32, 12); 1027 bind(done); 1028 } 1029 1030 #ifdef ASSERT 1031 { 1032 Label Lok; 1033 cmpdi(CCR0, Rtarget_addr, 0); 1034 bne(CCR0, Lok); 1035 stop("null entry point"); 1036 bind(Lok); 1037 } 1038 #endif // ASSERT 1039 1040 mr(R21_sender_SP, R1_SP); 1041 1042 // Calc a precise SP for the call. The SP value we calculated in 1043 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space 1044 // if esp is not max. Also, the i2c adapter extends the stack space without restoring 1045 // our pre-calced value, so repeating calls via i2c would result in stack overflow. 1046 // Since esp already points to an empty slot, we just have to sub 1 additional slot 1047 // to meet the abi scratch requirements. 1048 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in 1049 // the return entry of the interpreter. 1050 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size); 1051 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address 1052 resize_frame_absolute(Rscratch2, Rscratch2, R0); 1053 1054 mr_if_needed(R19_method, Rtarget_method); 1055 mtctr(Rtarget_addr); 1056 mtlr(Rret_addr); 1057 1058 save_interpreter_state(Rscratch2); 1059 #ifdef ASSERT 1060 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp 1061 cmpd(CCR0, R21_sender_SP, Rscratch1); 1062 asm_assert_eq("top_frame_sp incorrect", 0x951); 1063 #endif 1064 1065 bctr(); 1066 } 1067 1068 // Set the method data pointer for the current bcp. 1069 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1070 assert(ProfileInterpreter, "must be profiling interpreter"); 1071 Label get_continue; 1072 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method); 1073 test_method_data_pointer(get_continue); 1074 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp); 1075 1076 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1077 add(R28_mdx, R28_mdx, R3_RET); 1078 bind(get_continue); 1079 } 1080 1081 // Test ImethodDataPtr. If it is null, continue at the specified label. 1082 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1083 assert(ProfileInterpreter, "must be profiling interpreter"); 1084 cmpdi(CCR0, R28_mdx, 0); 1085 beq(CCR0, zero_continue); 1086 } 1087 1088 void InterpreterMacroAssembler::verify_method_data_pointer() { 1089 assert(ProfileInterpreter, "must be profiling interpreter"); 1090 #ifdef ASSERT 1091 Label verify_continue; 1092 test_method_data_pointer(verify_continue); 1093 1094 // If the mdp is valid, it will point to a DataLayout header which is 1095 // consistent with the bcp. The converse is highly probable also. 1096 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx); 1097 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method); 1098 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset())); 1099 add(R11_scratch1, R12_scratch2, R12_scratch2); 1100 cmpd(CCR0, R11_scratch1, R14_bcp); 1101 beq(CCR0, verify_continue); 1102 1103 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx); 1104 1105 bind(verify_continue); 1106 #endif 1107 } 1108 1109 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1110 Register Rscratch, 1111 Label &profile_continue) { 1112 assert(ProfileInterpreter, "must be profiling interpreter"); 1113 // Control will flow to "profile_continue" if the counter is less than the 1114 // limit or if we call profile_method(). 1115 Label done; 1116 1117 // If no method data exists, and the counter is high enough, make one. 1118 int ipl_offs = load_const_optimized(Rscratch, &InvocationCounter::InterpreterProfileLimit, R0, true); 1119 lwz(Rscratch, ipl_offs, Rscratch); 1120 1121 cmpdi(CCR0, R28_mdx, 0); 1122 // Test to see if we should create a method data oop. 1123 cmpd(CCR1, Rscratch /* InterpreterProfileLimit */, invocation_count); 1124 bne(CCR0, done); 1125 bge(CCR1, profile_continue); 1126 1127 // Build it now. 1128 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1129 set_method_data_pointer_for_bcp(); 1130 b(profile_continue); 1131 1132 align(32, 12); 1133 bind(done); 1134 } 1135 1136 void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register branch_bcp, Register Rtmp) { 1137 assert_different_registers(backedge_count, Rtmp, branch_bcp); 1138 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 1139 1140 Label did_not_overflow; 1141 Label overflow_with_error; 1142 1143 int ibbl_offs = load_const_optimized(Rtmp, &InvocationCounter::InterpreterBackwardBranchLimit, R0, true); 1144 lwz(Rtmp, ibbl_offs, Rtmp); 1145 cmpw(CCR0, backedge_count, Rtmp); 1146 1147 blt(CCR0, did_not_overflow); 1148 1149 // When ProfileInterpreter is on, the backedge_count comes from the 1150 // methodDataOop, which value does not get reset on the call to 1151 // frequency_counter_overflow(). To avoid excessive calls to the overflow 1152 // routine while the method is being compiled, add a second test to make sure 1153 // the overflow function is called only once every overflow_frequency. 1154 if (ProfileInterpreter) { 1155 const int overflow_frequency = 1024; 1156 li(Rtmp, overflow_frequency-1); 1157 andr(Rtmp, Rtmp, backedge_count); 1158 cmpwi(CCR0, Rtmp, 0); 1159 bne(CCR0, did_not_overflow); 1160 } 1161 1162 // Overflow in loop, pass branch bytecode. 1163 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, true); 1164 1165 // Was an OSR adapter generated? 1166 // O0 = osr nmethod 1167 cmpdi(CCR0, R3_RET, 0); 1168 beq(CCR0, overflow_with_error); 1169 1170 // Has the nmethod been invalidated already? 1171 lbz(Rtmp, nmethod::state_offset(), R3_RET); 1172 cmpwi(CCR0, Rtmp, nmethod::in_use); 1173 bne(CCR0, overflow_with_error); 1174 1175 // Migrate the interpreter frame off of the stack. 1176 // We can use all registers because we will not return to interpreter from this point. 1177 1178 // Save nmethod. 1179 const Register osr_nmethod = R31; 1180 mr(osr_nmethod, R3_RET); 1181 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1); 1182 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread); 1183 reset_last_Java_frame(); 1184 // OSR buffer is in ARG1 1185 1186 // Remove the interpreter frame. 1187 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1188 1189 // Jump to the osr code. 1190 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod); 1191 mtlr(R0); 1192 mtctr(R11_scratch1); 1193 bctr(); 1194 1195 align(32, 12); 1196 bind(overflow_with_error); 1197 bind(did_not_overflow); 1198 } 1199 1200 // Store a value at some constant offset from the method data pointer. 1201 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1202 assert(ProfileInterpreter, "must be profiling interpreter"); 1203 1204 std(value, constant, R28_mdx); 1205 } 1206 1207 // Increment the value at some constant offset from the method data pointer. 1208 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1209 Register counter_addr, 1210 Register Rbumped_count, 1211 bool decrement) { 1212 // Locate the counter at a fixed offset from the mdp: 1213 addi(counter_addr, R28_mdx, constant); 1214 increment_mdp_data_at(counter_addr, Rbumped_count, decrement); 1215 } 1216 1217 // Increment the value at some non-fixed (reg + constant) offset from 1218 // the method data pointer. 1219 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1220 int constant, 1221 Register scratch, 1222 Register Rbumped_count, 1223 bool decrement) { 1224 // Add the constant to reg to get the offset. 1225 add(scratch, R28_mdx, reg); 1226 // Then calculate the counter address. 1227 addi(scratch, scratch, constant); 1228 increment_mdp_data_at(scratch, Rbumped_count, decrement); 1229 } 1230 1231 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr, 1232 Register Rbumped_count, 1233 bool decrement) { 1234 assert(ProfileInterpreter, "must be profiling interpreter"); 1235 1236 // Load the counter. 1237 ld(Rbumped_count, 0, counter_addr); 1238 1239 if (decrement) { 1240 // Decrement the register. Set condition codes. 1241 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment); 1242 // Store the decremented counter, if it is still negative. 1243 std(Rbumped_count, 0, counter_addr); 1244 // Note: add/sub overflow check are not ported, since 64 bit 1245 // calculation should never overflow. 1246 } else { 1247 // Increment the register. Set carry flag. 1248 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment); 1249 // Store the incremented counter. 1250 std(Rbumped_count, 0, counter_addr); 1251 } 1252 } 1253 1254 // Set a flag value at the current method data pointer position. 1255 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1256 Register scratch) { 1257 assert(ProfileInterpreter, "must be profiling interpreter"); 1258 // Load the data header. 1259 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1260 // Set the flag. 1261 ori(scratch, scratch, flag_constant); 1262 // Store the modified header. 1263 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1264 } 1265 1266 // Test the location at some offset from the method data pointer. 1267 // If it is not equal to value, branch to the not_equal_continue Label. 1268 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1269 Register value, 1270 Label& not_equal_continue, 1271 Register test_out) { 1272 assert(ProfileInterpreter, "must be profiling interpreter"); 1273 1274 ld(test_out, offset, R28_mdx); 1275 cmpd(CCR0, value, test_out); 1276 bne(CCR0, not_equal_continue); 1277 } 1278 1279 // Update the method data pointer by the displacement located at some fixed 1280 // offset from the method data pointer. 1281 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1282 Register scratch) { 1283 assert(ProfileInterpreter, "must be profiling interpreter"); 1284 1285 ld(scratch, offset_of_disp, R28_mdx); 1286 add(R28_mdx, scratch, R28_mdx); 1287 } 1288 1289 // Update the method data pointer by the displacement located at the 1290 // offset (reg + offset_of_disp). 1291 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1292 int offset_of_disp, 1293 Register scratch) { 1294 assert(ProfileInterpreter, "must be profiling interpreter"); 1295 1296 add(scratch, reg, R28_mdx); 1297 ld(scratch, offset_of_disp, scratch); 1298 add(R28_mdx, scratch, R28_mdx); 1299 } 1300 1301 // Update the method data pointer by a simple constant displacement. 1302 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1303 assert(ProfileInterpreter, "must be profiling interpreter"); 1304 addi(R28_mdx, R28_mdx, constant); 1305 } 1306 1307 // Update the method data pointer for a _ret bytecode whose target 1308 // was not among our cached targets. 1309 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1310 Register return_bci) { 1311 assert(ProfileInterpreter, "must be profiling interpreter"); 1312 1313 push(state); 1314 assert(return_bci->is_nonvolatile(), "need to protect return_bci"); 1315 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1316 pop(state); 1317 } 1318 1319 // Increments the backedge counter. 1320 // Returns backedge counter + invocation counter in Rdst. 1321 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst, 1322 const Register Rtmp1, Register Rscratch) { 1323 assert(UseCompiler, "incrementing must be useful"); 1324 assert_different_registers(Rdst, Rtmp1); 1325 const Register invocation_counter = Rtmp1; 1326 const Register counter = Rdst; 1327 // TODO ppc port assert(4 == InvocationCounter::sz_counter(), "unexpected field size."); 1328 1329 // Load backedge counter. 1330 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1331 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1332 // Load invocation counter. 1333 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) + 1334 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1335 1336 // Add the delta to the backedge counter. 1337 addi(counter, counter, InvocationCounter::count_increment); 1338 1339 // Mask the invocation counter. 1340 li(Rscratch, InvocationCounter::count_mask_value); 1341 andr(invocation_counter, invocation_counter, Rscratch); 1342 1343 // Store new counter value. 1344 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1345 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1346 // Return invocation counter + backedge counter. 1347 add(counter, counter, invocation_counter); 1348 } 1349 1350 // Count a taken branch in the bytecodes. 1351 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1352 if (ProfileInterpreter) { 1353 Label profile_continue; 1354 1355 // If no method data exists, go to profile_continue. 1356 test_method_data_pointer(profile_continue); 1357 1358 // We are taking a branch. Increment the taken count. 1359 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count); 1360 1361 // The method data pointer needs to be updated to reflect the new target. 1362 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1363 bind (profile_continue); 1364 } 1365 } 1366 1367 // Count a not-taken branch in the bytecodes. 1368 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) { 1369 if (ProfileInterpreter) { 1370 Label profile_continue; 1371 1372 // If no method data exists, go to profile_continue. 1373 test_method_data_pointer(profile_continue); 1374 1375 // We are taking a branch. Increment the not taken count. 1376 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2); 1377 1378 // The method data pointer needs to be updated to correspond to the 1379 // next bytecode. 1380 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1381 bind (profile_continue); 1382 } 1383 } 1384 1385 // Count a non-virtual call in the bytecodes. 1386 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) { 1387 if (ProfileInterpreter) { 1388 Label profile_continue; 1389 1390 // If no method data exists, go to profile_continue. 1391 test_method_data_pointer(profile_continue); 1392 1393 // We are making a call. Increment the count. 1394 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1395 1396 // The method data pointer needs to be updated to reflect the new target. 1397 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1398 bind (profile_continue); 1399 } 1400 } 1401 1402 // Count a final call in the bytecodes. 1403 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) { 1404 if (ProfileInterpreter) { 1405 Label profile_continue; 1406 1407 // If no method data exists, go to profile_continue. 1408 test_method_data_pointer(profile_continue); 1409 1410 // We are making a call. Increment the count. 1411 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1412 1413 // The method data pointer needs to be updated to reflect the new target. 1414 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1415 bind (profile_continue); 1416 } 1417 } 1418 1419 // Count a virtual call in the bytecodes. 1420 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver, 1421 Register Rscratch1, 1422 Register Rscratch2, 1423 bool receiver_can_be_null) { 1424 if (!ProfileInterpreter) { return; } 1425 Label profile_continue; 1426 1427 // If no method data exists, go to profile_continue. 1428 test_method_data_pointer(profile_continue); 1429 1430 Label skip_receiver_profile; 1431 if (receiver_can_be_null) { 1432 Label not_null; 1433 cmpdi(CCR0, Rreceiver, 0); 1434 bne(CCR0, not_null); 1435 // We are making a call. Increment the count for null receiver. 1436 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2); 1437 b(skip_receiver_profile); 1438 bind(not_null); 1439 } 1440 1441 // Record the receiver type. 1442 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true); 1443 bind(skip_receiver_profile); 1444 1445 // The method data pointer needs to be updated to reflect the new target. 1446 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1447 bind (profile_continue); 1448 } 1449 1450 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) { 1451 if (ProfileInterpreter) { 1452 Label profile_continue; 1453 1454 // If no method data exists, go to profile_continue. 1455 test_method_data_pointer(profile_continue); 1456 1457 int mdp_delta = in_bytes(BitData::bit_data_size()); 1458 if (TypeProfileCasts) { 1459 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1460 1461 // Record the object type. 1462 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false); 1463 } 1464 1465 // The method data pointer needs to be updated. 1466 update_mdp_by_constant(mdp_delta); 1467 1468 bind (profile_continue); 1469 } 1470 } 1471 1472 void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) { 1473 if (ProfileInterpreter && TypeProfileCasts) { 1474 Label profile_continue; 1475 1476 // If no method data exists, go to profile_continue. 1477 test_method_data_pointer(profile_continue); 1478 1479 int count_offset = in_bytes(CounterData::count_offset()); 1480 // Back up the address, since we have already bumped the mdp. 1481 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1482 1483 // *Decrement* the counter. We expect to see zero or small negatives. 1484 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true); 1485 1486 bind (profile_continue); 1487 } 1488 } 1489 1490 // Count a ret in the bytecodes. 1491 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, Register scratch1, Register scratch2) { 1492 if (ProfileInterpreter) { 1493 Label profile_continue; 1494 uint row; 1495 1496 // If no method data exists, go to profile_continue. 1497 test_method_data_pointer(profile_continue); 1498 1499 // Update the total ret count. 1500 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 ); 1501 1502 for (row = 0; row < RetData::row_limit(); row++) { 1503 Label next_test; 1504 1505 // See if return_bci is equal to bci[n]: 1506 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1); 1507 1508 // return_bci is equal to bci[n]. Increment the count. 1509 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2); 1510 1511 // The method data pointer needs to be updated to reflect the new target. 1512 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1); 1513 b(profile_continue); 1514 bind(next_test); 1515 } 1516 1517 update_mdp_for_ret(state, return_bci); 1518 1519 bind (profile_continue); 1520 } 1521 } 1522 1523 // Count the default case of a switch construct. 1524 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) { 1525 if (ProfileInterpreter) { 1526 Label profile_continue; 1527 1528 // If no method data exists, go to profile_continue. 1529 test_method_data_pointer(profile_continue); 1530 1531 // Update the default case count 1532 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1533 scratch1, scratch2); 1534 1535 // The method data pointer needs to be updated. 1536 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()), 1537 scratch1); 1538 1539 bind (profile_continue); 1540 } 1541 } 1542 1543 // Count the index'th case of a switch construct. 1544 void InterpreterMacroAssembler::profile_switch_case(Register index, 1545 Register scratch1, 1546 Register scratch2, 1547 Register scratch3) { 1548 if (ProfileInterpreter) { 1549 assert_different_registers(index, scratch1, scratch2, scratch3); 1550 Label profile_continue; 1551 1552 // If no method data exists, go to profile_continue. 1553 test_method_data_pointer(profile_continue); 1554 1555 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes(). 1556 li(scratch3, in_bytes(MultiBranchData::case_array_offset())); 1557 1558 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works"); 1559 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1560 add(scratch1, scratch1, scratch3); 1561 1562 // Update the case count. 1563 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3); 1564 1565 // The method data pointer needs to be updated. 1566 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2); 1567 1568 bind (profile_continue); 1569 } 1570 } 1571 1572 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) { 1573 if (ProfileInterpreter) { 1574 assert_different_registers(Rscratch1, Rscratch2); 1575 Label profile_continue; 1576 1577 // If no method data exists, go to profile_continue. 1578 test_method_data_pointer(profile_continue); 1579 1580 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1); 1581 1582 // The method data pointer needs to be updated. 1583 int mdp_delta = in_bytes(BitData::bit_data_size()); 1584 if (TypeProfileCasts) { 1585 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1586 } 1587 update_mdp_by_constant(mdp_delta); 1588 1589 bind (profile_continue); 1590 } 1591 } 1592 1593 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver, 1594 Register Rscratch1, Register Rscratch2, 1595 bool is_virtual_call) { 1596 assert(ProfileInterpreter, "must be profiling"); 1597 assert_different_registers(Rreceiver, Rscratch1, Rscratch2); 1598 1599 Label done; 1600 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call); 1601 bind (done); 1602 } 1603 1604 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1605 Register receiver, Register scratch1, Register scratch2, 1606 int start_row, Label& done, bool is_virtual_call) { 1607 if (TypeProfileWidth == 0) { 1608 if (is_virtual_call) { 1609 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1610 } 1611 return; 1612 } 1613 1614 int last_row = VirtualCallData::row_limit() - 1; 1615 assert(start_row <= last_row, "must be work left to do"); 1616 // Test this row for both the receiver and for null. 1617 // Take any of three different outcomes: 1618 // 1. found receiver => increment count and goto done 1619 // 2. found null => keep looking for case 1, maybe allocate this cell 1620 // 3. found something else => keep looking for cases 1 and 2 1621 // Case 3 is handled by a recursive call. 1622 for (int row = start_row; row <= last_row; row++) { 1623 Label next_test; 1624 bool test_for_null_also = (row == start_row); 1625 1626 // See if the receiver is receiver[n]. 1627 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1628 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1); 1629 // delayed()->tst(scratch); 1630 1631 // The receiver is receiver[n]. Increment count[n]. 1632 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1633 increment_mdp_data_at(count_offset, scratch1, scratch2); 1634 b(done); 1635 bind(next_test); 1636 1637 if (test_for_null_also) { 1638 Label found_null; 1639 // Failed the equality check on receiver[n]... Test for null. 1640 if (start_row == last_row) { 1641 // The only thing left to do is handle the null case. 1642 if (is_virtual_call) { 1643 // Scratch1 contains test_out from test_mdp_data_at. 1644 cmpdi(CCR0, scratch1, 0); 1645 beq(CCR0, found_null); 1646 // Receiver did not match any saved receiver and there is no empty row for it. 1647 // Increment total counter to indicate polymorphic case. 1648 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1649 b(done); 1650 bind(found_null); 1651 } else { 1652 cmpdi(CCR0, scratch1, 0); 1653 bne(CCR0, done); 1654 } 1655 break; 1656 } 1657 // Since null is rare, make it be the branch-taken case. 1658 cmpdi(CCR0, scratch1, 0); 1659 beq(CCR0, found_null); 1660 1661 // Put all the "Case 3" tests here. 1662 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call); 1663 1664 // Found a null. Keep searching for a matching receiver, 1665 // but remember that this is an empty (unused) slot. 1666 bind(found_null); 1667 } 1668 } 1669 1670 // In the fall-through case, we found no matching receiver, but we 1671 // observed the receiver[start_row] is NULL. 1672 1673 // Fill in the receiver field and increment the count. 1674 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1675 set_mdp_data_at(recvr_offset, receiver); 1676 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1677 li(scratch1, DataLayout::counter_increment); 1678 set_mdp_data_at(count_offset, scratch1); 1679 if (start_row > 0) { 1680 b(done); 1681 } 1682 } 1683 1684 // Argument and return type profilig. 1685 // kills: tmp, tmp2, R0, CR0, CR1 1686 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base, 1687 RegisterOrConstant mdo_addr_offs, Register tmp, Register tmp2) { 1688 Label do_nothing, do_update; 1689 1690 // tmp2 = obj is allowed 1691 assert_different_registers(obj, mdo_addr_base, tmp, R0); 1692 assert_different_registers(tmp2, mdo_addr_base, tmp, R0); 1693 const Register klass = tmp2; 1694 1695 verify_oop(obj); 1696 1697 ld(tmp, mdo_addr_offs, mdo_addr_base); 1698 1699 // Set null_seen if obj is 0. 1700 cmpdi(CCR0, obj, 0); 1701 ori(R0, tmp, TypeEntries::null_seen); 1702 beq(CCR0, do_update); 1703 1704 load_klass(klass, obj); 1705 1706 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask)); 1707 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask); 1708 cmpd(CCR1, R0, klass); 1709 // Klass seen before, nothing to do (regardless of unknown bit). 1710 //beq(CCR1, do_nothing); 1711 1712 andi_(R0, klass, TypeEntries::type_unknown); 1713 // Already unknown. Nothing to do anymore. 1714 //bne(CCR0, do_nothing); 1715 crorc(/*CCR0 eq*/2, /*CCR1 eq*/4+2, /*CCR0 eq*/2); // cr0 eq = cr1 eq or cr0 ne 1716 beq(CCR0, do_nothing); 1717 1718 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask)); 1719 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1720 beq(CCR0, do_update); // First time here. Set profile type. 1721 1722 // Different than before. Cannot keep accurate profile. 1723 ori(R0, tmp, TypeEntries::type_unknown); 1724 1725 bind(do_update); 1726 // update profile 1727 std(R0, mdo_addr_offs, mdo_addr_base); 1728 1729 align(32, 12); 1730 bind(do_nothing); 1731 } 1732 1733 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) { 1734 if (!ProfileInterpreter) { 1735 return; 1736 } 1737 1738 assert_different_registers(callee, tmp1, tmp2, R28_mdx); 1739 1740 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1741 Label profile_continue; 1742 1743 test_method_data_pointer(profile_continue); 1744 1745 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1746 1747 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx); 1748 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1749 bne(CCR0, profile_continue); 1750 1751 if (MethodData::profile_arguments()) { 1752 Label done; 1753 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1754 add(R28_mdx, off_to_args, R28_mdx); 1755 1756 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1757 if (i > 0 || MethodData::profile_return()) { 1758 // If return value type is profiled we may have no argument to profile. 1759 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1760 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count()); 1761 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count()); 1762 blt(CCR0, done); 1763 } 1764 ld(tmp1, in_bytes(Method::const_offset()), callee); 1765 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1); 1766 // Stack offset o (zero based) from the start of the argument 1767 // list, for n arguments translates into offset n - o - 1 from 1768 // the end of the argument list. But there's an extra slot at 1769 // the top of the stack. So the offset is n - o from Lesp. 1770 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx); 1771 subf(tmp1, tmp2, tmp1); 1772 1773 sldi(tmp1, tmp1, Interpreter::logStackElementSize); 1774 ldx(tmp1, tmp1, R15_esp); 1775 1776 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1); 1777 1778 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1779 addi(R28_mdx, R28_mdx, to_add); 1780 off_to_args += to_add; 1781 } 1782 1783 if (MethodData::profile_return()) { 1784 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1785 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1786 } 1787 1788 bind(done); 1789 1790 if (MethodData::profile_return()) { 1791 // We're right after the type profile for the last 1792 // argument. tmp1 is the number of cells left in the 1793 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1794 // if there's a return to profile. 1795 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 1796 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size)); 1797 add(R28_mdx, tmp1, R28_mdx); 1798 } 1799 } else { 1800 assert(MethodData::profile_return(), "either profile call args or call ret"); 1801 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 1802 } 1803 1804 // Mdp points right after the end of the 1805 // CallTypeData/VirtualCallTypeData, right after the cells for the 1806 // return value type if there's one. 1807 align(32, 12); 1808 bind(profile_continue); 1809 } 1810 } 1811 1812 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 1813 assert_different_registers(ret, tmp1, tmp2); 1814 if (ProfileInterpreter && MethodData::profile_return()) { 1815 Label profile_continue; 1816 1817 test_method_data_pointer(profile_continue); 1818 1819 if (MethodData::profile_return_jsr292_only()) { 1820 // If we don't profile all invoke bytecodes we must make sure 1821 // it's a bytecode we indeed profile. We can't go back to the 1822 // begining of the ProfileData we intend to update to check its 1823 // type because we're right after it and we don't known its 1824 // length. 1825 lbz(tmp1, 0, R14_bcp); 1826 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method); 1827 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic); 1828 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle); 1829 cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2); 1830 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm); 1831 cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2); 1832 bne(CCR0, profile_continue); 1833 } 1834 1835 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2); 1836 1837 align(32, 12); 1838 bind(profile_continue); 1839 } 1840 } 1841 1842 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) { 1843 if (ProfileInterpreter && MethodData::profile_parameters()) { 1844 Label profile_continue, done; 1845 1846 test_method_data_pointer(profile_continue); 1847 1848 // Load the offset of the area within the MDO used for 1849 // parameters. If it's negative we're not profiling any parameters. 1850 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx); 1851 cmpwi(CCR0, tmp1, 0); 1852 blt(CCR0, profile_continue); 1853 1854 // Compute a pointer to the area for parameters from the offset 1855 // and move the pointer to the slot for the last 1856 // parameters. Collect profiling from last parameter down. 1857 // mdo start + parameters offset + array length - 1 1858 1859 // Pointer to the parameter area in the MDO. 1860 const Register mdp = tmp1; 1861 add(mdp, tmp1, R28_mdx); 1862 1863 // Pffset of the current profile entry to update. 1864 const Register entry_offset = tmp2; 1865 // entry_offset = array len in number of cells 1866 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp); 1867 1868 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1869 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 1870 1871 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 1872 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size)); 1873 // entry_offset in bytes 1874 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1875 1876 Label loop; 1877 align(32, 12); 1878 bind(loop); 1879 1880 // Load offset on the stack from the slot for this parameter. 1881 ld(tmp3, entry_offset, mdp); 1882 sldi(tmp3, tmp3, Interpreter::logStackElementSize); 1883 neg(tmp3, tmp3); 1884 // Read the parameter from the local area. 1885 ldx(tmp3, tmp3, R18_locals); 1886 1887 // Make entry_offset now point to the type field for this parameter. 1888 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1889 assert(type_base > off_base, "unexpected"); 1890 addi(entry_offset, entry_offset, type_base - off_base); 1891 1892 // Profile the parameter. 1893 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3); 1894 1895 // Go to next parameter. 1896 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base); 1897 cmpdi(CCR0, entry_offset, off_base + delta); 1898 addi(entry_offset, entry_offset, -delta); 1899 bge(CCR0, loop); 1900 1901 align(32, 12); 1902 bind(profile_continue); 1903 } 1904 } 1905 1906 // Add a InterpMonitorElem to stack (see frame_sparc.hpp). 1907 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) { 1908 1909 // Very-local scratch registers. 1910 const Register esp = Rtemp1; 1911 const Register slot = Rtemp2; 1912 1913 // Extracted monitor_size. 1914 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes(); 1915 assert(Assembler::is_aligned((unsigned int)monitor_size, 1916 (unsigned int)frame::alignment_in_bytes), 1917 "size of a monitor must respect alignment of SP"); 1918 1919 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor 1920 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp 1921 1922 // Shuffle expression stack down. Recall that stack_base points 1923 // just above the new expression stack bottom. Old_tos and new_tos 1924 // are used to scan thru the old and new expression stacks. 1925 if (!stack_is_empty) { 1926 Label copy_slot, copy_slot_finished; 1927 const Register n_slots = slot; 1928 1929 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack). 1930 subf(n_slots, esp, R26_monitor); 1931 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy. 1932 assert(LogBytesPerWord == 3, "conflicts assembler instructions"); 1933 beq(CCR0, copy_slot_finished); // Nothing to copy. 1934 1935 mtctr(n_slots); 1936 1937 // loop 1938 bind(copy_slot); 1939 ld(slot, 0, esp); // Move expression stack down. 1940 std(slot, -monitor_size, esp); // distance = monitor_size 1941 addi(esp, esp, BytesPerWord); 1942 bdnz(copy_slot); 1943 1944 bind(copy_slot_finished); 1945 } 1946 1947 addi(R15_esp, R15_esp, -monitor_size); 1948 addi(R26_monitor, R26_monitor, -monitor_size); 1949 1950 // Restart interpreter 1951 } 1952 1953 // ============================================================================ 1954 // Java locals access 1955 1956 // Load a local variable at index in Rindex into register Rdst_value. 1957 // Also puts address of local into Rdst_address as a service. 1958 // Kills: 1959 // - Rdst_value 1960 // - Rdst_address 1961 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) { 1962 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 1963 subf(Rdst_address, Rdst_address, R18_locals); 1964 lwz(Rdst_value, 0, Rdst_address); 1965 } 1966 1967 // Load a local variable at index in Rindex into register Rdst_value. 1968 // Also puts address of local into Rdst_address as a service. 1969 // Kills: 1970 // - Rdst_value 1971 // - Rdst_address 1972 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) { 1973 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 1974 subf(Rdst_address, Rdst_address, R18_locals); 1975 ld(Rdst_value, -8, Rdst_address); 1976 } 1977 1978 // Load a local variable at index in Rindex into register Rdst_value. 1979 // Also puts address of local into Rdst_address as a service. 1980 // Input: 1981 // - Rindex: slot nr of local variable 1982 // Kills: 1983 // - Rdst_value 1984 // - Rdst_address 1985 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, Register Rdst_address, Register Rindex) { 1986 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 1987 subf(Rdst_address, Rdst_address, R18_locals); 1988 ld(Rdst_value, 0, Rdst_address); 1989 } 1990 1991 // Load a local variable at index in Rindex into register Rdst_value. 1992 // Also puts address of local into Rdst_address as a service. 1993 // Kills: 1994 // - Rdst_value 1995 // - Rdst_address 1996 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) { 1997 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 1998 subf(Rdst_address, Rdst_address, R18_locals); 1999 lfs(Rdst_value, 0, Rdst_address); 2000 } 2001 2002 // Load a local variable at index in Rindex into register Rdst_value. 2003 // Also puts address of local into Rdst_address as a service. 2004 // Kills: 2005 // - Rdst_value 2006 // - Rdst_address 2007 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) { 2008 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2009 subf(Rdst_address, Rdst_address, R18_locals); 2010 lfd(Rdst_value, -8, Rdst_address); 2011 } 2012 2013 // Store an int value at local variable slot Rindex. 2014 // Kills: 2015 // - Rindex 2016 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) { 2017 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2018 subf(Rindex, Rindex, R18_locals); 2019 stw(Rvalue, 0, Rindex); 2020 } 2021 2022 // Store a long value at local variable slot Rindex. 2023 // Kills: 2024 // - Rindex 2025 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) { 2026 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2027 subf(Rindex, Rindex, R18_locals); 2028 std(Rvalue, -8, Rindex); 2029 } 2030 2031 // Store an oop value at local variable slot Rindex. 2032 // Kills: 2033 // - Rindex 2034 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) { 2035 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2036 subf(Rindex, Rindex, R18_locals); 2037 std(Rvalue, 0, Rindex); 2038 } 2039 2040 // Store an int value at local variable slot Rindex. 2041 // Kills: 2042 // - Rindex 2043 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) { 2044 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2045 subf(Rindex, Rindex, R18_locals); 2046 stfs(Rvalue, 0, Rindex); 2047 } 2048 2049 // Store an int value at local variable slot Rindex. 2050 // Kills: 2051 // - Rindex 2052 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) { 2053 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2054 subf(Rindex, Rindex, R18_locals); 2055 stfd(Rvalue, -8, Rindex); 2056 } 2057 2058 // Read pending exception from thread and jump to interpreter. 2059 // Throw exception entry if one if pending. Fall through otherwise. 2060 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) { 2061 assert_different_registers(Rscratch1, Rscratch2, R3); 2062 Register Rexception = Rscratch1; 2063 Register Rtmp = Rscratch2; 2064 Label Ldone; 2065 // Get pending exception oop. 2066 ld(Rexception, thread_(pending_exception)); 2067 cmpdi(CCR0, Rexception, 0); 2068 beq(CCR0, Ldone); 2069 li(Rtmp, 0); 2070 mr_if_needed(R3, Rexception); 2071 std(Rtmp, thread_(pending_exception)); // Clear exception in thread 2072 if (Interpreter::rethrow_exception_entry() != NULL) { 2073 // Already got entry address. 2074 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry()); 2075 } else { 2076 // Dynamically load entry address. 2077 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true); 2078 ld(Rtmp, simm16_rest, Rtmp); 2079 } 2080 mtctr(Rtmp); 2081 save_interpreter_state(Rtmp); 2082 bctr(); 2083 2084 align(32, 12); 2085 bind(Ldone); 2086 } 2087 2088 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) { 2089 save_interpreter_state(R11_scratch1); 2090 2091 MacroAssembler::call_VM(oop_result, entry_point, false); 2092 2093 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2094 2095 check_and_handle_popframe(R11_scratch1); 2096 check_and_handle_earlyret(R11_scratch1); 2097 // Now check exceptions manually. 2098 if (check_exceptions) { 2099 check_and_forward_exception(R11_scratch1, R12_scratch2); 2100 } 2101 } 2102 2103 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions) { 2104 // ARG1 is reserved for the thread. 2105 mr_if_needed(R4_ARG2, arg_1); 2106 call_VM(oop_result, entry_point, check_exceptions); 2107 } 2108 2109 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) { 2110 // ARG1 is reserved for the thread. 2111 mr_if_needed(R4_ARG2, arg_1); 2112 assert(arg_2 != R4_ARG2, "smashed argument"); 2113 mr_if_needed(R5_ARG3, arg_2); 2114 call_VM(oop_result, entry_point, check_exceptions); 2115 } 2116 2117 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) { 2118 // ARG1 is reserved for the thread. 2119 mr_if_needed(R4_ARG2, arg_1); 2120 assert(arg_2 != R4_ARG2, "smashed argument"); 2121 mr_if_needed(R5_ARG3, arg_2); 2122 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument"); 2123 mr_if_needed(R6_ARG4, arg_3); 2124 call_VM(oop_result, entry_point, check_exceptions); 2125 } 2126 2127 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) { 2128 ld(scratch, 0, R1_SP); 2129 std(R15_esp, _ijava_state_neg(esp), scratch); 2130 std(R14_bcp, _ijava_state_neg(bcp), scratch); 2131 std(R26_monitor, _ijava_state_neg(monitors), scratch); 2132 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); } 2133 // Other entries should be unchanged. 2134 } 2135 2136 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) { 2137 ld(scratch, 0, R1_SP); 2138 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception). 2139 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code. 2140 if (!bcp_and_mdx_only) { 2141 // Following ones are Metadata. 2142 ld(R19_method, _ijava_state_neg(method), scratch); 2143 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch); 2144 // Following ones are stack addresses and don't require reload. 2145 ld(R15_esp, _ijava_state_neg(esp), scratch); 2146 ld(R18_locals, _ijava_state_neg(locals), scratch); 2147 ld(R26_monitor, _ijava_state_neg(monitors), scratch); 2148 } 2149 #ifdef ASSERT 2150 { 2151 Label Lok; 2152 subf(R0, R1_SP, scratch); 2153 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size); 2154 bge(CCR0, Lok); 2155 stop("frame too small (restore istate)", 0x5432); 2156 bind(Lok); 2157 } 2158 { 2159 Label Lok; 2160 ld(R0, _ijava_state_neg(ijava_reserved), scratch); 2161 cmpdi(CCR0, R0, 0x5afe); 2162 beq(CCR0, Lok); 2163 stop("frame corrupted (restore istate)", 0x5afe); 2164 bind(Lok); 2165 } 2166 #endif 2167 } 2168 2169 #endif // !CC_INTERP 2170 2171 void InterpreterMacroAssembler::get_method_counters(Register method, 2172 Register Rcounters, 2173 Label& skip) { 2174 BLOCK_COMMENT("Load and ev. allocate counter object {"); 2175 Label has_counters; 2176 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2177 cmpdi(CCR0, Rcounters, 0); 2178 bne(CCR0, has_counters); 2179 call_VM(noreg, CAST_FROM_FN_PTR(address, 2180 InterpreterRuntime::build_method_counters), method, false); 2181 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2182 cmpdi(CCR0, Rcounters, 0); 2183 beq(CCR0, skip); // No MethodCounters, OutOfMemory. 2184 BLOCK_COMMENT("} Load and ev. allocate counter object"); 2185 2186 bind(has_counters); 2187 } 2188 2189 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register iv_be_count, Register Rtmp_r0) { 2190 assert(UseCompiler, "incrementing must be useful"); 2191 Register invocation_count = iv_be_count; 2192 Register backedge_count = Rtmp_r0; 2193 int delta = InvocationCounter::count_increment; 2194 2195 // Load each counter in a register. 2196 // ld(inv_counter, Rtmp); 2197 // ld(be_counter, Rtmp2); 2198 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + 2199 InvocationCounter::counter_offset()); 2200 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + 2201 InvocationCounter::counter_offset()); 2202 2203 BLOCK_COMMENT("Increment profiling counters {"); 2204 2205 // Load the backedge counter. 2206 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int 2207 // Mask the backedge counter. 2208 Register tmp = invocation_count; 2209 li(tmp, InvocationCounter::count_mask_value); 2210 andr(backedge_count, tmp, backedge_count); // Cannot use andi, need sign extension of count_mask_value. 2211 2212 // Load the invocation counter. 2213 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int 2214 // Add the delta to the invocation counter and store the result. 2215 addi(invocation_count, invocation_count, delta); 2216 // Store value. 2217 stw(invocation_count, inv_counter_offset, Rcounters); 2218 2219 // Add invocation counter + backedge counter. 2220 add(iv_be_count, backedge_count, invocation_count); 2221 2222 // Note that this macro must leave the backedge_count + invocation_count in 2223 // register iv_be_count! 2224 BLOCK_COMMENT("} Increment profiling counters"); 2225 } 2226 2227 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2228 if (state == atos) { MacroAssembler::verify_oop(reg); } 2229 } 2230 2231 #ifndef CC_INTERP 2232 // Local helper function for the verify_oop_or_return_address macro. 2233 static bool verify_return_address(Method* m, int bci) { 2234 #ifndef PRODUCT 2235 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci; 2236 // Assume it is a valid return address if it is inside m and is preceded by a jsr. 2237 if (!m->contains(pc)) return false; 2238 address jsr_pc; 2239 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2240 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2241 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2242 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2243 #endif // PRODUCT 2244 return false; 2245 } 2246 2247 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2248 if (VerifyFPU) { 2249 unimplemented("verfiyFPU"); 2250 } 2251 } 2252 2253 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2254 if (!VerifyOops) return; 2255 2256 // The VM documentation for the astore[_wide] bytecode allows 2257 // the TOS to be not only an oop but also a return address. 2258 Label test; 2259 Label skip; 2260 // See if it is an address (in the current method): 2261 2262 const int log2_bytecode_size_limit = 16; 2263 srdi_(Rtmp, reg, log2_bytecode_size_limit); 2264 bne(CCR0, test); 2265 2266 address fd = CAST_FROM_FN_PTR(address, verify_return_address); 2267 const int nbytes_save = 11*8; // volatile gprs except R0 2268 save_volatile_gprs(R1_SP, -nbytes_save); // except R0 2269 save_LR_CR(Rtmp); // Save in old frame. 2270 push_frame_reg_args(nbytes_save, Rtmp); 2271 2272 load_const_optimized(Rtmp, fd, R0); 2273 mr_if_needed(R4_ARG2, reg); 2274 mr(R3_ARG1, R19_method); 2275 call_c(Rtmp); // call C 2276 2277 pop_frame(); 2278 restore_LR_CR(Rtmp); 2279 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0 2280 b(skip); 2281 2282 // Perform a more elaborate out-of-line call. 2283 // Not an address; verify it: 2284 bind(test); 2285 verify_oop(reg); 2286 bind(skip); 2287 } 2288 #endif // !CC_INTERP 2289 2290 // Inline assembly for: 2291 // 2292 // if (thread is in interp_only_mode) { 2293 // InterpreterRuntime::post_method_entry(); 2294 // } 2295 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) || 2296 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) { 2297 // SharedRuntime::jvmpi_method_entry(method, receiver); 2298 // } 2299 void InterpreterMacroAssembler::notify_method_entry() { 2300 // JVMTI 2301 // Whenever JVMTI puts a thread in interp_only_mode, method 2302 // entry/exit events are sent for that thread to track stack 2303 // depth. If it is possible to enter interp_only_mode we add 2304 // the code to check if the event should be sent. 2305 if (JvmtiExport::can_post_interpreter_events()) { 2306 Label jvmti_post_done; 2307 2308 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2309 cmpwi(CCR0, R0, 0); 2310 beq(CCR0, jvmti_post_done); 2311 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), 2312 /*check_exceptions=*/true CC_INTERP_ONLY(&& false)); 2313 2314 bind(jvmti_post_done); 2315 } 2316 } 2317 2318 // Inline assembly for: 2319 // 2320 // if (thread is in interp_only_mode) { 2321 // // save result 2322 // InterpreterRuntime::post_method_exit(); 2323 // // restore result 2324 // } 2325 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) { 2326 // // save result 2327 // SharedRuntime::jvmpi_method_exit(); 2328 // // restore result 2329 // } 2330 // 2331 // Native methods have their result stored in d_tmp and l_tmp. 2332 // Java methods have their result stored in the expression stack. 2333 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state, 2334 NotifyMethodExitMode mode, bool check_exceptions) { 2335 // JVMTI 2336 // Whenever JVMTI puts a thread in interp_only_mode, method 2337 // entry/exit events are sent for that thread to track stack 2338 // depth. If it is possible to enter interp_only_mode we add 2339 // the code to check if the event should be sent. 2340 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2341 Label jvmti_post_done; 2342 2343 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2344 cmpwi(CCR0, R0, 0); 2345 beq(CCR0, jvmti_post_done); 2346 CC_INTERP_ONLY(assert(is_native_method && !check_exceptions, "must not push state")); 2347 if (!is_native_method) push(state); // Expose tos to GC. 2348 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), 2349 /*check_exceptions=*/check_exceptions); 2350 if (!is_native_method) pop(state); 2351 2352 align(32, 12); 2353 bind(jvmti_post_done); 2354 } 2355 2356 // Dtrace support not implemented. 2357 } 2358 2359 #ifdef CC_INTERP 2360 // Convert the current TOP_IJAVA_FRAME into a PARENT_IJAVA_FRAME 2361 // (using parent_frame_resize) and push a new interpreter 2362 // TOP_IJAVA_FRAME (using frame_size). 2363 void InterpreterMacroAssembler::push_interpreter_frame(Register top_frame_size, Register parent_frame_resize, 2364 Register tmp1, Register tmp2, Register tmp3, 2365 Register tmp4, Register pc) { 2366 assert_different_registers(top_frame_size, parent_frame_resize, tmp1, tmp2, tmp3, tmp4); 2367 ld(tmp1, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2368 mr(tmp2/*top_frame_sp*/, R1_SP); 2369 // Move initial_caller_sp. 2370 ld(tmp4, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2371 neg(parent_frame_resize, parent_frame_resize); 2372 resize_frame(parent_frame_resize/*-parent_frame_resize*/, tmp3); 2373 2374 // Set LR in new parent frame. 2375 std(tmp1, _abi(lr), R1_SP); 2376 // Set top_frame_sp info for new parent frame. 2377 std(tmp2, _parent_ijava_frame_abi(top_frame_sp), R1_SP); 2378 std(tmp4, _parent_ijava_frame_abi(initial_caller_sp), R1_SP); 2379 2380 // Push new TOP_IJAVA_FRAME. 2381 push_frame(top_frame_size, tmp2); 2382 2383 get_PC_trash_LR(tmp3); 2384 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2385 // Used for non-initial callers by unextended_sp(). 2386 std(R1_SP, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2387 } 2388 2389 // Pop the topmost TOP_IJAVA_FRAME and convert the previous 2390 // PARENT_IJAVA_FRAME back into a TOP_IJAVA_FRAME. 2391 void InterpreterMacroAssembler::pop_interpreter_frame(Register tmp1, Register tmp2, Register tmp3, Register tmp4) { 2392 assert_different_registers(tmp1, tmp2, tmp3, tmp4); 2393 2394 ld(tmp1/*caller's sp*/, _abi(callers_sp), R1_SP); 2395 ld(tmp3, _abi(lr), tmp1); 2396 2397 ld(tmp4, _parent_ijava_frame_abi(initial_caller_sp), tmp1); 2398 2399 ld(tmp2/*caller's caller's sp*/, _abi(callers_sp), tmp1); 2400 // Merge top frame. 2401 std(tmp2, _abi(callers_sp), R1_SP); 2402 2403 ld(tmp2, _parent_ijava_frame_abi(top_frame_sp), tmp1); 2404 2405 // Update C stack pointer to caller's top_abi. 2406 resize_frame_absolute(tmp2/*addr*/, tmp1/*tmp*/, tmp2/*tmp*/); 2407 2408 // Update LR in top_frame. 2409 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2410 2411 std(tmp4, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2412 2413 // Store the top-frame stack-pointer for c2i adapters. 2414 std(R1_SP, _top_ijava_frame_abi(top_frame_sp), R1_SP); 2415 } 2416 2417 // Turn state's interpreter frame into the current TOP_IJAVA_FRAME. 2418 void InterpreterMacroAssembler::pop_interpreter_frame_to_state(Register state, Register tmp1, Register tmp2, Register tmp3) { 2419 assert_different_registers(R14_state, R15_prev_state, tmp1, tmp2, tmp3); 2420 2421 if (state == R14_state) { 2422 ld(tmp1/*state's fp*/, state_(_last_Java_fp)); 2423 ld(tmp2/*state's sp*/, state_(_last_Java_sp)); 2424 } else if (state == R15_prev_state) { 2425 ld(tmp1/*state's fp*/, prev_state_(_last_Java_fp)); 2426 ld(tmp2/*state's sp*/, prev_state_(_last_Java_sp)); 2427 } else { 2428 ShouldNotReachHere(); 2429 } 2430 2431 // Merge top frames. 2432 std(tmp1, _abi(callers_sp), R1_SP); 2433 2434 // Tmp2 is new SP. 2435 // Tmp1 is parent's SP. 2436 resize_frame_absolute(tmp2/*addr*/, tmp1/*tmp*/, tmp2/*tmp*/); 2437 2438 // Update LR in top_frame. 2439 // Must be interpreter frame. 2440 get_PC_trash_LR(tmp3); 2441 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2442 // Used for non-initial callers by unextended_sp(). 2443 std(R1_SP, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2444 } 2445 2446 // Set SP to initial caller's sp, but before fix the back chain. 2447 void InterpreterMacroAssembler::resize_frame_to_initial_caller(Register tmp1, Register tmp2) { 2448 ld(tmp1, _parent_ijava_frame_abi(initial_caller_sp), R1_SP); 2449 ld(tmp2, _parent_ijava_frame_abi(callers_sp), R1_SP); 2450 std(tmp2, _parent_ijava_frame_abi(callers_sp), tmp1); // Fix back chain ... 2451 mr(R1_SP, tmp1); // ... and resize to initial caller. 2452 } 2453 2454 // Pop the current interpreter state (without popping the correspoding 2455 // frame) and restore R14_state and R15_prev_state accordingly. 2456 // Use prev_state_may_be_0 to indicate whether prev_state may be 0 2457 // in order to generate an extra check before retrieving prev_state_(_prev_link). 2458 void InterpreterMacroAssembler::pop_interpreter_state(bool prev_state_may_be_0) 2459 { 2460 // Move prev_state to state and restore prev_state from state_(_prev_link). 2461 Label prev_state_is_0; 2462 mr(R14_state, R15_prev_state); 2463 2464 // Don't retrieve /*state==*/prev_state_(_prev_link) 2465 // if /*state==*/prev_state is 0. 2466 if (prev_state_may_be_0) { 2467 cmpdi(CCR0, R15_prev_state, 0); 2468 beq(CCR0, prev_state_is_0); 2469 } 2470 2471 ld(R15_prev_state, /*state==*/prev_state_(_prev_link)); 2472 bind(prev_state_is_0); 2473 } 2474 2475 void InterpreterMacroAssembler::restore_prev_state() { 2476 // _prev_link is private, but cInterpreter is a friend. 2477 ld(R15_prev_state, state_(_prev_link)); 2478 } 2479 #endif // CC_INTERP