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 ztos: // fall through 178 case ctos: // fall through 179 case stos: // fall through 180 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 181 break; 182 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 183 break; 184 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 185 break; 186 case vtos: break; 187 default : ShouldNotReachHere(); 188 } 189 190 // Clean up tos value in the jvmti thread state. 191 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 192 // Set tos state field to illegal value. 193 li(Rscratch2, ilgl); 194 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState); 195 } 196 197 // Common code to dispatch and dispatch_only. 198 // Dispatch value in Lbyte_code and increment Lbcp. 199 200 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) { 201 address table_base = (address)Interpreter::dispatch_table((TosState)0); 202 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base; 203 if (is_simm16(table_offs)) { 204 addi(dst, R25_templateTableBase, (int)table_offs); 205 } else { 206 load_const_optimized(dst, table, R0); 207 } 208 } 209 210 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode, address* table, bool verify) { 211 if (verify) { 212 unimplemented("dispatch_Lbyte_code: verify"); // See Sparc Implementation to implement this 213 } 214 215 #ifdef FAST_DISPATCH 216 unimplemented("dispatch_Lbyte_code FAST_DISPATCH"); 217 #else 218 assert_different_registers(bytecode, R11_scratch1); 219 220 // Calc dispatch table address. 221 load_dispatch_table(R11_scratch1, table); 222 223 sldi(R12_scratch2, bytecode, LogBytesPerWord); 224 ldx(R11_scratch1, R11_scratch1, R12_scratch2); 225 226 // Jump off! 227 mtctr(R11_scratch1); 228 bctr(); 229 #endif 230 } 231 232 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) { 233 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize); 234 ldx(Rrecv_dst, Rrecv_dst, R15_esp); 235 } 236 237 // helpers for expression stack 238 239 void InterpreterMacroAssembler::pop_i(Register r) { 240 lwzu(r, Interpreter::stackElementSize, R15_esp); 241 } 242 243 void InterpreterMacroAssembler::pop_ptr(Register r) { 244 ldu(r, Interpreter::stackElementSize, R15_esp); 245 } 246 247 void InterpreterMacroAssembler::pop_l(Register r) { 248 ld(r, Interpreter::stackElementSize, R15_esp); 249 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 250 } 251 252 void InterpreterMacroAssembler::pop_f(FloatRegister f) { 253 lfsu(f, Interpreter::stackElementSize, R15_esp); 254 } 255 256 void InterpreterMacroAssembler::pop_d(FloatRegister f) { 257 lfd(f, Interpreter::stackElementSize, R15_esp); 258 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 259 } 260 261 void InterpreterMacroAssembler::push_i(Register r) { 262 stw(r, 0, R15_esp); 263 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 264 } 265 266 void InterpreterMacroAssembler::push_ptr(Register r) { 267 std(r, 0, R15_esp); 268 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 269 } 270 271 void InterpreterMacroAssembler::push_l(Register r) { 272 std(r, - Interpreter::stackElementSize, R15_esp); 273 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 274 } 275 276 void InterpreterMacroAssembler::push_f(FloatRegister f) { 277 stfs(f, 0, R15_esp); 278 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 279 } 280 281 void InterpreterMacroAssembler::push_d(FloatRegister f) { 282 stfd(f, - Interpreter::stackElementSize, R15_esp); 283 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 284 } 285 286 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) { 287 std(first, 0, R15_esp); 288 std(second, -Interpreter::stackElementSize, R15_esp); 289 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 290 } 291 292 void InterpreterMacroAssembler::push_l_pop_d(Register l, FloatRegister d) { 293 std(l, 0, R15_esp); 294 lfd(d, 0, R15_esp); 295 } 296 297 void InterpreterMacroAssembler::push_d_pop_l(FloatRegister d, Register l) { 298 stfd(d, 0, R15_esp); 299 ld(l, 0, R15_esp); 300 } 301 302 void InterpreterMacroAssembler::push(TosState state) { 303 switch (state) { 304 case atos: push_ptr(); break; 305 case btos: 306 case ztos: 307 case ctos: 308 case stos: 309 case itos: push_i(); break; 310 case ltos: push_l(); break; 311 case ftos: push_f(); break; 312 case dtos: push_d(); break; 313 case vtos: /* nothing to do */ break; 314 default : ShouldNotReachHere(); 315 } 316 } 317 318 void InterpreterMacroAssembler::pop(TosState state) { 319 switch (state) { 320 case atos: pop_ptr(); break; 321 case btos: 322 case ztos: 323 case ctos: 324 case stos: 325 case itos: pop_i(); break; 326 case ltos: pop_l(); break; 327 case ftos: pop_f(); break; 328 case dtos: pop_d(); break; 329 case vtos: /* nothing to do */ break; 330 default : ShouldNotReachHere(); 331 } 332 verify_oop(R17_tos, state); 333 } 334 335 void InterpreterMacroAssembler::empty_expression_stack() { 336 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 337 } 338 339 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset, 340 Register Rdst, 341 signedOrNot is_signed) { 342 #if defined(VM_LITTLE_ENDIAN) 343 if (bcp_offset) { 344 load_const_optimized(Rdst, bcp_offset); 345 lhbrx(Rdst, R14_bcp, Rdst); 346 } else { 347 lhbrx(Rdst, R14_bcp); 348 } 349 if (is_signed == Signed) { 350 extsh(Rdst, Rdst); 351 } 352 #else 353 // Read Java big endian format. 354 if (is_signed == Signed) { 355 lha(Rdst, bcp_offset, R14_bcp); 356 } else { 357 lhz(Rdst, bcp_offset, R14_bcp); 358 } 359 #endif 360 } 361 362 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset, 363 Register Rdst, 364 signedOrNot is_signed) { 365 #if defined(VM_LITTLE_ENDIAN) 366 if (bcp_offset) { 367 load_const_optimized(Rdst, bcp_offset); 368 lwbrx(Rdst, R14_bcp, Rdst); 369 } else { 370 lwbrx(Rdst, R14_bcp); 371 } 372 if (is_signed == Signed) { 373 extsw(Rdst, Rdst); 374 } 375 #else 376 // Read Java big endian format. 377 if (bcp_offset & 3) { // Offset unaligned? 378 load_const_optimized(Rdst, bcp_offset); 379 if (is_signed == Signed) { 380 lwax(Rdst, R14_bcp, Rdst); 381 } else { 382 lwzx(Rdst, R14_bcp, Rdst); 383 } 384 } else { 385 if (is_signed == Signed) { 386 lwa(Rdst, bcp_offset, R14_bcp); 387 } else { 388 lwz(Rdst, bcp_offset, R14_bcp); 389 } 390 } 391 #endif 392 } 393 394 395 // Load the constant pool cache index from the bytecode stream. 396 // 397 // Kills / writes: 398 // - Rdst, Rscratch 399 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, size_t index_size) { 400 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 401 // Cache index is always in the native format, courtesy of Rewriter. 402 if (index_size == sizeof(u2)) { 403 lhz(Rdst, bcp_offset, R14_bcp); 404 } else if (index_size == sizeof(u4)) { 405 assert(EnableInvokeDynamic, "giant index used only for JSR 292"); 406 if (bcp_offset & 3) { 407 load_const_optimized(Rdst, bcp_offset); 408 lwax(Rdst, R14_bcp, Rdst); 409 } else { 410 lwa(Rdst, bcp_offset, R14_bcp); 411 } 412 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 413 nand(Rdst, Rdst, Rdst); // convert to plain index 414 } else if (index_size == sizeof(u1)) { 415 lbz(Rdst, bcp_offset, R14_bcp); 416 } else { 417 ShouldNotReachHere(); 418 } 419 // Rdst now contains cp cache index. 420 } 421 422 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset, size_t index_size) { 423 get_cache_index_at_bcp(cache, bcp_offset, index_size); 424 sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord)); 425 add(cache, R27_constPoolCache, cache); 426 } 427 428 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format) 429 // from (Rsrc)+offset. 430 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset, 431 signedOrNot is_signed) { 432 #if defined(VM_LITTLE_ENDIAN) 433 if (offset) { 434 load_const_optimized(Rdst, offset); 435 lwbrx(Rdst, Rdst, Rsrc); 436 } else { 437 lwbrx(Rdst, Rsrc); 438 } 439 if (is_signed == Signed) { 440 extsw(Rdst, Rdst); 441 } 442 #else 443 if (is_signed == Signed) { 444 lwa(Rdst, offset, Rsrc); 445 } else { 446 lwz(Rdst, offset, Rsrc); 447 } 448 #endif 449 } 450 451 // Load object from cpool->resolved_references(index). 452 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) { 453 assert_different_registers(result, index); 454 get_constant_pool(result); 455 456 // Convert from field index to resolved_references() index and from 457 // word index to byte offset. Since this is a java object, it can be compressed. 458 Register tmp = index; // reuse 459 sldi(tmp, index, LogBytesPerHeapOop); 460 // Load pointer for resolved_references[] objArray. 461 ld(result, ConstantPool::resolved_references_offset_in_bytes(), result); 462 // JNIHandles::resolve(result) 463 ld(result, 0, result); 464 #ifdef ASSERT 465 Label index_ok; 466 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result); 467 sldi(R0, R0, LogBytesPerHeapOop); 468 cmpd(CCR0, tmp, R0); 469 blt(CCR0, index_ok); 470 stop("resolved reference index out of bounds", 0x09256); 471 bind(index_ok); 472 #endif 473 // Add in the index. 474 add(result, tmp, result); 475 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result); 476 } 477 478 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 479 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2. 480 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1, 481 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) { 482 // Profile the not-null value's klass. 483 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2); 484 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype); 485 profile_typecheck_failed(Rtmp1, Rtmp2); 486 } 487 488 void InterpreterMacroAssembler::generate_stack_overflow_check_with_compare_and_throw(Register Rmem_frame_size, Register Rscratch1) { 489 Label done; 490 sub(Rmem_frame_size, R1_SP, Rmem_frame_size); 491 ld(Rscratch1, thread_(stack_overflow_limit)); 492 cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1); 493 bgt(CCR0/*is_stack_overflow*/, done); 494 495 // Load target address of the runtime stub. 496 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order"); 497 load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0); 498 mtctr(Rscratch1); 499 // Restore caller_sp. 500 #ifdef ASSERT 501 ld(Rscratch1, 0, R1_SP); 502 ld(R0, 0, R21_sender_SP); 503 cmpd(CCR0, R0, Rscratch1); 504 asm_assert_eq("backlink", 0x547); 505 #endif // ASSERT 506 mr(R1_SP, R21_sender_SP); 507 bctr(); 508 509 align(32, 12); 510 bind(done); 511 } 512 513 // Separate these two to allow for delay slot in middle. 514 // These are used to do a test and full jump to exception-throwing code. 515 516 // Check that index is in range for array, then shift index by index_shift, 517 // and put arrayOop + shifted_index into res. 518 // Note: res is still shy of address by array offset into object. 519 520 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, int index_shift, Register Rtmp, Register Rres) { 521 // Check that index is in range for array, then shift index by index_shift, 522 // and put arrayOop + shifted_index into res. 523 // Note: res is still shy of address by array offset into object. 524 // Kills: 525 // - Rindex 526 // Writes: 527 // - Rres: Address that corresponds to the array index if check was successful. 528 verify_oop(Rarray); 529 const Register Rlength = R0; 530 const Register RsxtIndex = Rtmp; 531 Label LisNull, LnotOOR; 532 533 // Array nullcheck 534 if (!ImplicitNullChecks) { 535 cmpdi(CCR0, Rarray, 0); 536 beq(CCR0, LisNull); 537 } else { 538 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex); 539 } 540 541 // Rindex might contain garbage in upper bits (remember that we don't sign extend 542 // during integer arithmetic operations). So kill them and put value into same register 543 // where ArrayIndexOutOfBounds would expect the index in. 544 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit 545 546 // Index check 547 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray); 548 cmplw(CCR0, Rindex, Rlength); 549 sldi(RsxtIndex, RsxtIndex, index_shift); 550 blt(CCR0, LnotOOR); 551 // Index should be in R17_tos, array should be in R4_ARG2. 552 mr(R17_tos, Rindex); 553 mr(R4_ARG2, Rarray); 554 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry); 555 mtctr(Rtmp); 556 bctr(); 557 558 if (!ImplicitNullChecks) { 559 bind(LisNull); 560 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry); 561 mtctr(Rtmp); 562 bctr(); 563 } 564 565 align(32, 16); 566 bind(LnotOOR); 567 568 // Calc address 569 add(Rres, RsxtIndex, Rarray); 570 } 571 572 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { 573 // pop array 574 pop_ptr(array); 575 576 // check array 577 index_check_without_pop(array, index, index_shift, tmp, res); 578 } 579 580 void InterpreterMacroAssembler::get_const(Register Rdst) { 581 ld(Rdst, in_bytes(Method::const_offset()), R19_method); 582 } 583 584 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 585 get_const(Rdst); 586 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 587 } 588 589 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 590 get_constant_pool(Rdst); 591 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 592 } 593 594 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 595 get_constant_pool(Rcpool); 596 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool); 597 } 598 599 // Unlock if synchronized method. 600 // 601 // Unlock the receiver if this is a synchronized method. 602 // Unlock any Java monitors from synchronized blocks. 603 // 604 // If there are locked Java monitors 605 // If throw_monitor_exception 606 // throws IllegalMonitorStateException 607 // Else if install_monitor_exception 608 // installs IllegalMonitorStateException 609 // Else 610 // no error processing 611 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 612 bool throw_monitor_exception, 613 bool install_monitor_exception) { 614 Label Lunlocked, Lno_unlock; 615 { 616 Register Rdo_not_unlock_flag = R11_scratch1; 617 Register Raccess_flags = R12_scratch2; 618 619 // Check if synchronized method or unlocking prevented by 620 // JavaThread::do_not_unlock_if_synchronized flag. 621 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 622 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method); 623 li(R0, 0); 624 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag 625 626 push(state); 627 628 // Skip if we don't have to unlock. 629 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0. 630 beq(CCR0, Lunlocked); 631 632 cmpwi(CCR0, Rdo_not_unlock_flag, 0); 633 bne(CCR0, Lno_unlock); 634 } 635 636 // Unlock 637 { 638 Register Rmonitor_base = R11_scratch1; 639 640 Label Lunlock; 641 // If it's still locked, everything is ok, unlock it. 642 ld(Rmonitor_base, 0, R1_SP); 643 addi(Rmonitor_base, Rmonitor_base, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base 644 645 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base); 646 cmpdi(CCR0, R0, 0); 647 bne(CCR0, Lunlock); 648 649 // If it's already unlocked, throw exception. 650 if (throw_monitor_exception) { 651 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 652 should_not_reach_here(); 653 } else { 654 if (install_monitor_exception) { 655 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 656 b(Lunlocked); 657 } 658 } 659 660 bind(Lunlock); 661 unlock_object(Rmonitor_base); 662 } 663 664 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not. 665 bind(Lunlocked); 666 { 667 Label Lexception, Lrestart; 668 Register Rcurrent_obj_addr = R11_scratch1; 669 const int delta = frame::interpreter_frame_monitor_size_in_bytes(); 670 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords"); 671 672 bind(Lrestart); 673 // Set up search loop: Calc num of iterations. 674 { 675 Register Riterations = R12_scratch2; 676 Register Rmonitor_base = Rcurrent_obj_addr; 677 ld(Rmonitor_base, 0, R1_SP); 678 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base 679 680 subf_(Riterations, R26_monitor, Rmonitor_base); 681 ble(CCR0, Lno_unlock); 682 683 addi(Rcurrent_obj_addr, Rmonitor_base, BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes()); 684 // Check if any monitor is on stack, bail out if not 685 srdi(Riterations, Riterations, exact_log2(delta)); 686 mtctr(Riterations); 687 } 688 689 // The search loop: Look for locked monitors. 690 { 691 const Register Rcurrent_obj = R0; 692 Label Lloop; 693 694 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 695 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 696 bind(Lloop); 697 698 // Check if current entry is used. 699 cmpdi(CCR0, Rcurrent_obj, 0); 700 bne(CCR0, Lexception); 701 // Preload next iteration's compare value. 702 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 703 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 704 bdnz(Lloop); 705 } 706 // Fell through: Everything's unlocked => finish. 707 b(Lno_unlock); 708 709 // An object is still locked => need to throw exception. 710 bind(Lexception); 711 if (throw_monitor_exception) { 712 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 713 should_not_reach_here(); 714 } else { 715 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 716 // Unlock does not block, so don't have to worry about the frame. 717 Register Rmonitor_addr = R11_scratch1; 718 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta); 719 unlock_object(Rmonitor_addr); 720 if (install_monitor_exception) { 721 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 722 } 723 b(Lrestart); 724 } 725 } 726 727 align(32, 12); 728 bind(Lno_unlock); 729 pop(state); 730 } 731 732 // Support function for remove_activation & Co. 733 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, Register Rscratch1, Register Rscratch2) { 734 // Pop interpreter frame. 735 ld(Rscratch1, 0, R1_SP); // *SP 736 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp 737 ld(Rscratch2, 0, Rscratch1); // **SP 738 #ifdef ASSERT 739 { 740 Label Lok; 741 ld(R0, _ijava_state_neg(ijava_reserved), Rscratch1); 742 cmpdi(CCR0, R0, 0x5afe); 743 beq(CCR0, Lok); 744 stop("frame corrupted (remove activation)", 0x5afe); 745 bind(Lok); 746 } 747 #endif 748 if (return_pc!=noreg) { 749 ld(return_pc, _abi(lr), Rscratch1); // LR 750 } 751 752 // Merge top frames. 753 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP 754 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP 755 } 756 757 void InterpreterMacroAssembler::narrow(Register result) { 758 Register ret_type = R11_scratch1; 759 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method); 760 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1); 761 762 Label notBool, notByte, notChar, done; 763 764 // common case first 765 cmpwi(CCR0, ret_type, T_INT); 766 beq(CCR0, done); 767 768 cmpwi(CCR0, ret_type, T_BOOLEAN); 769 bne(CCR0, notBool); 770 andi(result, result, 0x1); 771 b(done); 772 773 bind(notBool); 774 cmpwi(CCR0, ret_type, T_BYTE); 775 bne(CCR0, notByte); 776 extsb(result, result); 777 b(done); 778 779 bind(notByte); 780 cmpwi(CCR0, ret_type, T_CHAR); 781 bne(CCR0, notChar); 782 andi(result, result, 0xffff); 783 b(done); 784 785 bind(notChar); 786 // cmpwi(CCR0, ret_type, T_SHORT); // all that's left 787 // bne(CCR0, done); 788 extsh(result, result); 789 790 // Nothing to do for T_INT 791 bind(done); 792 } 793 794 // Remove activation. 795 // 796 // Unlock the receiver if this is a synchronized method. 797 // Unlock any Java monitors from synchronized blocks. 798 // Remove the activation from the stack. 799 // 800 // If there are locked Java monitors 801 // If throw_monitor_exception 802 // throws IllegalMonitorStateException 803 // Else if install_monitor_exception 804 // installs IllegalMonitorStateException 805 // Else 806 // no error processing 807 void InterpreterMacroAssembler::remove_activation(TosState state, 808 bool throw_monitor_exception, 809 bool install_monitor_exception) { 810 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 811 812 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 813 notify_method_exit(false, state, NotifyJVMTI, true); 814 815 verify_oop(R17_tos, state); 816 verify_thread(); 817 818 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 819 mtlr(R0); 820 } 821 822 #endif // !CC_INTERP 823 824 // Lock object 825 // 826 // Registers alive 827 // monitor - Address of the BasicObjectLock to be used for locking, 828 // which must be initialized with the object to lock. 829 // object - Address of the object to be locked. 830 // 831 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 832 if (UseHeavyMonitors) { 833 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 834 monitor, /*check_for_exceptions=*/true CC_INTERP_ONLY(&& false)); 835 } else { 836 // template code: 837 // 838 // markOop displaced_header = obj->mark().set_unlocked(); 839 // monitor->lock()->set_displaced_header(displaced_header); 840 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 841 // // We stored the monitor address into the object's mark word. 842 // } else if (THREAD->is_lock_owned((address)displaced_header)) 843 // // Simple recursive case. 844 // monitor->lock()->set_displaced_header(NULL); 845 // } else { 846 // // Slow path. 847 // InterpreterRuntime::monitorenter(THREAD, monitor); 848 // } 849 850 const Register displaced_header = R7_ARG5; 851 const Register object_mark_addr = R8_ARG6; 852 const Register current_header = R9_ARG7; 853 const Register tmp = R10_ARG8; 854 855 Label done; 856 Label cas_failed, slow_case; 857 858 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp); 859 860 // markOop displaced_header = obj->mark().set_unlocked(); 861 862 // Load markOop from object into displaced_header. 863 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object); 864 865 if (UseBiasedLocking) { 866 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case); 867 } 868 869 // Set displaced_header to be (markOop of object | UNLOCK_VALUE). 870 ori(displaced_header, displaced_header, markOopDesc::unlocked_value); 871 872 // monitor->lock()->set_displaced_header(displaced_header); 873 874 // Initialize the box (Must happen before we update the object mark!). 875 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 876 BasicLock::displaced_header_offset_in_bytes(), monitor); 877 878 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 879 880 // Store stack address of the BasicObjectLock (this is monitor) into object. 881 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 882 883 // Must fence, otherwise, preceding store(s) may float below cmpxchg. 884 // CmpxchgX sets CCR0 to cmpX(current, displaced). 885 fence(); // TODO: replace by MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq ? 886 cmpxchgd(/*flag=*/CCR0, 887 /*current_value=*/current_header, 888 /*compare_value=*/displaced_header, /*exchange_value=*/monitor, 889 /*where=*/object_mark_addr, 890 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq, 891 MacroAssembler::cmpxchgx_hint_acquire_lock(), 892 noreg, 893 &cas_failed); 894 895 // If the compare-and-exchange succeeded, then we found an unlocked 896 // object and we have now locked it. 897 b(done); 898 bind(cas_failed); 899 900 // } else if (THREAD->is_lock_owned((address)displaced_header)) 901 // // Simple recursive case. 902 // monitor->lock()->set_displaced_header(NULL); 903 904 // We did not see an unlocked object so try the fast recursive case. 905 906 // Check if owner is self by comparing the value in the markOop of object 907 // (current_header) with the stack pointer. 908 sub(current_header, current_header, R1_SP); 909 910 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 911 load_const_optimized(tmp, 912 (address) (~(os::vm_page_size()-1) | 913 markOopDesc::lock_mask_in_place)); 914 915 and_(R0/*==0?*/, current_header, tmp); 916 // If condition is true we are done and hence we can store 0 in the displaced 917 // header indicating it is a recursive lock. 918 bne(CCR0, slow_case); 919 release(); 920 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() + 921 BasicLock::displaced_header_offset_in_bytes(), monitor); 922 b(done); 923 924 // } else { 925 // // Slow path. 926 // InterpreterRuntime::monitorenter(THREAD, monitor); 927 928 // None of the above fast optimizations worked so we have to get into the 929 // slow case of monitor enter. 930 bind(slow_case); 931 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 932 monitor, /*check_for_exceptions=*/true CC_INTERP_ONLY(&& false)); 933 // } 934 align(32, 12); 935 bind(done); 936 } 937 } 938 939 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 940 // 941 // Registers alive 942 // monitor - Address of the BasicObjectLock to be used for locking, 943 // which must be initialized with the object to lock. 944 // 945 // Throw IllegalMonitorException if object is not locked by current thread. 946 void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) { 947 if (UseHeavyMonitors) { 948 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 949 monitor, check_for_exceptions CC_INTERP_ONLY(&& false)); 950 } else { 951 952 // template code: 953 // 954 // if ((displaced_header = monitor->displaced_header()) == NULL) { 955 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 956 // monitor->set_obj(NULL); 957 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 958 // // We swapped the unlocked mark in displaced_header into the object's mark word. 959 // monitor->set_obj(NULL); 960 // } else { 961 // // Slow path. 962 // InterpreterRuntime::monitorexit(THREAD, monitor); 963 // } 964 965 const Register object = R7_ARG5; 966 const Register displaced_header = R8_ARG6; 967 const Register object_mark_addr = R9_ARG7; 968 const Register current_header = R10_ARG8; 969 970 Label free_slot; 971 Label slow_case; 972 973 assert_different_registers(object, displaced_header, object_mark_addr, current_header); 974 975 if (UseBiasedLocking) { 976 // The object address from the monitor is in object. 977 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); 978 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0"); 979 biased_locking_exit(CCR0, object, displaced_header, free_slot); 980 } 981 982 // Test first if we are in the fast recursive case. 983 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 984 BasicLock::displaced_header_offset_in_bytes(), monitor); 985 986 // If the displaced header is zero, we have a recursive unlock. 987 cmpdi(CCR0, displaced_header, 0); 988 beq(CCR0, free_slot); // recursive unlock 989 990 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 991 // // We swapped the unlocked mark in displaced_header into the object's mark word. 992 // monitor->set_obj(NULL); 993 994 // If we still have a lightweight lock, unlock the object and be done. 995 996 // The object address from the monitor is in object. 997 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); } 998 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 999 1000 // We have the displaced header in displaced_header. If the lock is still 1001 // lightweight, it will contain the monitor address and we'll store the 1002 // displaced header back into the object's mark word. 1003 // CmpxchgX sets CCR0 to cmpX(current, monitor). 1004 cmpxchgd(/*flag=*/CCR0, 1005 /*current_value=*/current_header, 1006 /*compare_value=*/monitor, /*exchange_value=*/displaced_header, 1007 /*where=*/object_mark_addr, 1008 MacroAssembler::MemBarRel, 1009 MacroAssembler::cmpxchgx_hint_release_lock(), 1010 noreg, 1011 &slow_case); 1012 b(free_slot); 1013 1014 // } else { 1015 // // Slow path. 1016 // InterpreterRuntime::monitorexit(THREAD, monitor); 1017 1018 // The lock has been converted into a heavy lock and hence 1019 // we need to get into the slow case. 1020 bind(slow_case); 1021 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1022 monitor, check_for_exceptions CC_INTERP_ONLY(&& false)); 1023 // } 1024 1025 Label done; 1026 b(done); // Monitor register may be overwritten! Runtime has already freed the slot. 1027 1028 // Exchange worked, do monitor->set_obj(NULL); 1029 align(32, 12); 1030 bind(free_slot); 1031 li(R0, 0); 1032 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor); 1033 bind(done); 1034 } 1035 } 1036 1037 #ifndef CC_INTERP 1038 1039 // Load compiled (i2c) or interpreter entry when calling from interpreted and 1040 // do the call. Centralized so that all interpreter calls will do the same actions. 1041 // If jvmti single stepping is on for a thread we must not call compiled code. 1042 // 1043 // Input: 1044 // - Rtarget_method: method to call 1045 // - Rret_addr: return address 1046 // - 2 scratch regs 1047 // 1048 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, Register Rscratch1, Register Rscratch2) { 1049 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr); 1050 // Assume we want to go compiled if available. 1051 const Register Rtarget_addr = Rscratch1; 1052 const Register Rinterp_only = Rscratch2; 1053 1054 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method); 1055 1056 if (JvmtiExport::can_post_interpreter_events()) { 1057 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 1058 1059 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 1060 // compiled code in threads for which the event is enabled. Check here for 1061 // interp_only_mode if these events CAN be enabled. 1062 Label done; 1063 verify_thread(); 1064 cmpwi(CCR0, Rinterp_only, 0); 1065 beq(CCR0, done); 1066 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method); 1067 align(32, 12); 1068 bind(done); 1069 } 1070 1071 #ifdef ASSERT 1072 { 1073 Label Lok; 1074 cmpdi(CCR0, Rtarget_addr, 0); 1075 bne(CCR0, Lok); 1076 stop("null entry point"); 1077 bind(Lok); 1078 } 1079 #endif // ASSERT 1080 1081 mr(R21_sender_SP, R1_SP); 1082 1083 // Calc a precise SP for the call. The SP value we calculated in 1084 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space 1085 // if esp is not max. Also, the i2c adapter extends the stack space without restoring 1086 // our pre-calced value, so repeating calls via i2c would result in stack overflow. 1087 // Since esp already points to an empty slot, we just have to sub 1 additional slot 1088 // to meet the abi scratch requirements. 1089 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in 1090 // the return entry of the interpreter. 1091 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size); 1092 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address 1093 resize_frame_absolute(Rscratch2, Rscratch2, R0); 1094 1095 mr_if_needed(R19_method, Rtarget_method); 1096 mtctr(Rtarget_addr); 1097 mtlr(Rret_addr); 1098 1099 save_interpreter_state(Rscratch2); 1100 #ifdef ASSERT 1101 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp 1102 cmpd(CCR0, R21_sender_SP, Rscratch1); 1103 asm_assert_eq("top_frame_sp incorrect", 0x951); 1104 #endif 1105 1106 bctr(); 1107 } 1108 1109 // Set the method data pointer for the current bcp. 1110 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1111 assert(ProfileInterpreter, "must be profiling interpreter"); 1112 Label get_continue; 1113 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method); 1114 test_method_data_pointer(get_continue); 1115 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp); 1116 1117 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1118 add(R28_mdx, R28_mdx, R3_RET); 1119 bind(get_continue); 1120 } 1121 1122 // Test ImethodDataPtr. If it is null, continue at the specified label. 1123 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1124 assert(ProfileInterpreter, "must be profiling interpreter"); 1125 cmpdi(CCR0, R28_mdx, 0); 1126 beq(CCR0, zero_continue); 1127 } 1128 1129 void InterpreterMacroAssembler::verify_method_data_pointer() { 1130 assert(ProfileInterpreter, "must be profiling interpreter"); 1131 #ifdef ASSERT 1132 Label verify_continue; 1133 test_method_data_pointer(verify_continue); 1134 1135 // If the mdp is valid, it will point to a DataLayout header which is 1136 // consistent with the bcp. The converse is highly probable also. 1137 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx); 1138 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method); 1139 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset())); 1140 add(R11_scratch1, R12_scratch2, R12_scratch2); 1141 cmpd(CCR0, R11_scratch1, R14_bcp); 1142 beq(CCR0, verify_continue); 1143 1144 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx); 1145 1146 bind(verify_continue); 1147 #endif 1148 } 1149 1150 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1151 Register Rscratch, 1152 Label &profile_continue) { 1153 assert(ProfileInterpreter, "must be profiling interpreter"); 1154 // Control will flow to "profile_continue" if the counter is less than the 1155 // limit or if we call profile_method(). 1156 Label done; 1157 1158 // If no method data exists, and the counter is high enough, make one. 1159 int ipl_offs = load_const_optimized(Rscratch, &InvocationCounter::InterpreterProfileLimit, R0, true); 1160 lwz(Rscratch, ipl_offs, Rscratch); 1161 1162 cmpdi(CCR0, R28_mdx, 0); 1163 // Test to see if we should create a method data oop. 1164 cmpd(CCR1, Rscratch /* InterpreterProfileLimit */, invocation_count); 1165 bne(CCR0, done); 1166 bge(CCR1, profile_continue); 1167 1168 // Build it now. 1169 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1170 set_method_data_pointer_for_bcp(); 1171 b(profile_continue); 1172 1173 align(32, 12); 1174 bind(done); 1175 } 1176 1177 void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register branch_bcp, Register Rtmp) { 1178 assert_different_registers(backedge_count, Rtmp, branch_bcp); 1179 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 1180 1181 Label did_not_overflow; 1182 Label overflow_with_error; 1183 1184 int ibbl_offs = load_const_optimized(Rtmp, &InvocationCounter::InterpreterBackwardBranchLimit, R0, true); 1185 lwz(Rtmp, ibbl_offs, Rtmp); 1186 cmpw(CCR0, backedge_count, Rtmp); 1187 1188 blt(CCR0, did_not_overflow); 1189 1190 // When ProfileInterpreter is on, the backedge_count comes from the 1191 // methodDataOop, which value does not get reset on the call to 1192 // frequency_counter_overflow(). To avoid excessive calls to the overflow 1193 // routine while the method is being compiled, add a second test to make sure 1194 // the overflow function is called only once every overflow_frequency. 1195 if (ProfileInterpreter) { 1196 const int overflow_frequency = 1024; 1197 li(Rtmp, overflow_frequency-1); 1198 andr(Rtmp, Rtmp, backedge_count); 1199 cmpwi(CCR0, Rtmp, 0); 1200 bne(CCR0, did_not_overflow); 1201 } 1202 1203 // Overflow in loop, pass branch bytecode. 1204 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, true); 1205 1206 // Was an OSR adapter generated? 1207 // O0 = osr nmethod 1208 cmpdi(CCR0, R3_RET, 0); 1209 beq(CCR0, overflow_with_error); 1210 1211 // Has the nmethod been invalidated already? 1212 lwz(Rtmp, nmethod::entry_bci_offset(), R3_RET); 1213 cmpwi(CCR0, Rtmp, InvalidOSREntryBci); 1214 beq(CCR0, overflow_with_error); 1215 1216 // Migrate the interpreter frame off of the stack. 1217 // We can use all registers because we will not return to interpreter from this point. 1218 1219 // Save nmethod. 1220 const Register osr_nmethod = R31; 1221 mr(osr_nmethod, R3_RET); 1222 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1); 1223 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread); 1224 reset_last_Java_frame(); 1225 // OSR buffer is in ARG1 1226 1227 // Remove the interpreter frame. 1228 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1229 1230 // Jump to the osr code. 1231 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod); 1232 mtlr(R0); 1233 mtctr(R11_scratch1); 1234 bctr(); 1235 1236 align(32, 12); 1237 bind(overflow_with_error); 1238 bind(did_not_overflow); 1239 } 1240 1241 // Store a value at some constant offset from the method data pointer. 1242 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1243 assert(ProfileInterpreter, "must be profiling interpreter"); 1244 1245 std(value, constant, R28_mdx); 1246 } 1247 1248 // Increment the value at some constant offset from the method data pointer. 1249 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1250 Register counter_addr, 1251 Register Rbumped_count, 1252 bool decrement) { 1253 // Locate the counter at a fixed offset from the mdp: 1254 addi(counter_addr, R28_mdx, constant); 1255 increment_mdp_data_at(counter_addr, Rbumped_count, decrement); 1256 } 1257 1258 // Increment the value at some non-fixed (reg + constant) offset from 1259 // the method data pointer. 1260 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1261 int constant, 1262 Register scratch, 1263 Register Rbumped_count, 1264 bool decrement) { 1265 // Add the constant to reg to get the offset. 1266 add(scratch, R28_mdx, reg); 1267 // Then calculate the counter address. 1268 addi(scratch, scratch, constant); 1269 increment_mdp_data_at(scratch, Rbumped_count, decrement); 1270 } 1271 1272 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr, 1273 Register Rbumped_count, 1274 bool decrement) { 1275 assert(ProfileInterpreter, "must be profiling interpreter"); 1276 1277 // Load the counter. 1278 ld(Rbumped_count, 0, counter_addr); 1279 1280 if (decrement) { 1281 // Decrement the register. Set condition codes. 1282 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment); 1283 // Store the decremented counter, if it is still negative. 1284 std(Rbumped_count, 0, counter_addr); 1285 // Note: add/sub overflow check are not ported, since 64 bit 1286 // calculation should never overflow. 1287 } else { 1288 // Increment the register. Set carry flag. 1289 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment); 1290 // Store the incremented counter. 1291 std(Rbumped_count, 0, counter_addr); 1292 } 1293 } 1294 1295 // Set a flag value at the current method data pointer position. 1296 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1297 Register scratch) { 1298 assert(ProfileInterpreter, "must be profiling interpreter"); 1299 // Load the data header. 1300 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1301 // Set the flag. 1302 ori(scratch, scratch, flag_constant); 1303 // Store the modified header. 1304 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1305 } 1306 1307 // Test the location at some offset from the method data pointer. 1308 // If it is not equal to value, branch to the not_equal_continue Label. 1309 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1310 Register value, 1311 Label& not_equal_continue, 1312 Register test_out) { 1313 assert(ProfileInterpreter, "must be profiling interpreter"); 1314 1315 ld(test_out, offset, R28_mdx); 1316 cmpd(CCR0, value, test_out); 1317 bne(CCR0, not_equal_continue); 1318 } 1319 1320 // Update the method data pointer by the displacement located at some fixed 1321 // offset from the method data pointer. 1322 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1323 Register scratch) { 1324 assert(ProfileInterpreter, "must be profiling interpreter"); 1325 1326 ld(scratch, offset_of_disp, R28_mdx); 1327 add(R28_mdx, scratch, R28_mdx); 1328 } 1329 1330 // Update the method data pointer by the displacement located at the 1331 // offset (reg + offset_of_disp). 1332 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1333 int offset_of_disp, 1334 Register scratch) { 1335 assert(ProfileInterpreter, "must be profiling interpreter"); 1336 1337 add(scratch, reg, R28_mdx); 1338 ld(scratch, offset_of_disp, scratch); 1339 add(R28_mdx, scratch, R28_mdx); 1340 } 1341 1342 // Update the method data pointer by a simple constant displacement. 1343 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1344 assert(ProfileInterpreter, "must be profiling interpreter"); 1345 addi(R28_mdx, R28_mdx, constant); 1346 } 1347 1348 // Update the method data pointer for a _ret bytecode whose target 1349 // was not among our cached targets. 1350 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1351 Register return_bci) { 1352 assert(ProfileInterpreter, "must be profiling interpreter"); 1353 1354 push(state); 1355 assert(return_bci->is_nonvolatile(), "need to protect return_bci"); 1356 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1357 pop(state); 1358 } 1359 1360 // Increments the backedge counter. 1361 // Returns backedge counter + invocation counter in Rdst. 1362 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst, 1363 const Register Rtmp1, Register Rscratch) { 1364 assert(UseCompiler, "incrementing must be useful"); 1365 assert_different_registers(Rdst, Rtmp1); 1366 const Register invocation_counter = Rtmp1; 1367 const Register counter = Rdst; 1368 // TODO ppc port assert(4 == InvocationCounter::sz_counter(), "unexpected field size."); 1369 1370 // Load backedge counter. 1371 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1372 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1373 // Load invocation counter. 1374 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) + 1375 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1376 1377 // Add the delta to the backedge counter. 1378 addi(counter, counter, InvocationCounter::count_increment); 1379 1380 // Mask the invocation counter. 1381 li(Rscratch, InvocationCounter::count_mask_value); 1382 andr(invocation_counter, invocation_counter, Rscratch); 1383 1384 // Store new counter value. 1385 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1386 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1387 // Return invocation counter + backedge counter. 1388 add(counter, counter, invocation_counter); 1389 } 1390 1391 // Count a taken branch in the bytecodes. 1392 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1393 if (ProfileInterpreter) { 1394 Label profile_continue; 1395 1396 // If no method data exists, go to profile_continue. 1397 test_method_data_pointer(profile_continue); 1398 1399 // We are taking a branch. Increment the taken count. 1400 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count); 1401 1402 // The method data pointer needs to be updated to reflect the new target. 1403 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1404 bind (profile_continue); 1405 } 1406 } 1407 1408 // Count a not-taken branch in the bytecodes. 1409 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) { 1410 if (ProfileInterpreter) { 1411 Label profile_continue; 1412 1413 // If no method data exists, go to profile_continue. 1414 test_method_data_pointer(profile_continue); 1415 1416 // We are taking a branch. Increment the not taken count. 1417 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2); 1418 1419 // The method data pointer needs to be updated to correspond to the 1420 // next bytecode. 1421 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1422 bind (profile_continue); 1423 } 1424 } 1425 1426 // Count a non-virtual call in the bytecodes. 1427 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) { 1428 if (ProfileInterpreter) { 1429 Label profile_continue; 1430 1431 // If no method data exists, go to profile_continue. 1432 test_method_data_pointer(profile_continue); 1433 1434 // We are making a call. Increment the count. 1435 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1436 1437 // The method data pointer needs to be updated to reflect the new target. 1438 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1439 bind (profile_continue); 1440 } 1441 } 1442 1443 // Count a final call in the bytecodes. 1444 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) { 1445 if (ProfileInterpreter) { 1446 Label profile_continue; 1447 1448 // If no method data exists, go to profile_continue. 1449 test_method_data_pointer(profile_continue); 1450 1451 // We are making a call. Increment the count. 1452 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1453 1454 // The method data pointer needs to be updated to reflect the new target. 1455 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1456 bind (profile_continue); 1457 } 1458 } 1459 1460 // Count a virtual call in the bytecodes. 1461 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver, 1462 Register Rscratch1, 1463 Register Rscratch2, 1464 bool receiver_can_be_null) { 1465 if (!ProfileInterpreter) { return; } 1466 Label profile_continue; 1467 1468 // If no method data exists, go to profile_continue. 1469 test_method_data_pointer(profile_continue); 1470 1471 Label skip_receiver_profile; 1472 if (receiver_can_be_null) { 1473 Label not_null; 1474 cmpdi(CCR0, Rreceiver, 0); 1475 bne(CCR0, not_null); 1476 // We are making a call. Increment the count for null receiver. 1477 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2); 1478 b(skip_receiver_profile); 1479 bind(not_null); 1480 } 1481 1482 // Record the receiver type. 1483 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true); 1484 bind(skip_receiver_profile); 1485 1486 // The method data pointer needs to be updated to reflect the new target. 1487 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1488 bind (profile_continue); 1489 } 1490 1491 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) { 1492 if (ProfileInterpreter) { 1493 Label profile_continue; 1494 1495 // If no method data exists, go to profile_continue. 1496 test_method_data_pointer(profile_continue); 1497 1498 int mdp_delta = in_bytes(BitData::bit_data_size()); 1499 if (TypeProfileCasts) { 1500 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1501 1502 // Record the object type. 1503 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false); 1504 } 1505 1506 // The method data pointer needs to be updated. 1507 update_mdp_by_constant(mdp_delta); 1508 1509 bind (profile_continue); 1510 } 1511 } 1512 1513 void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) { 1514 if (ProfileInterpreter && TypeProfileCasts) { 1515 Label profile_continue; 1516 1517 // If no method data exists, go to profile_continue. 1518 test_method_data_pointer(profile_continue); 1519 1520 int count_offset = in_bytes(CounterData::count_offset()); 1521 // Back up the address, since we have already bumped the mdp. 1522 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1523 1524 // *Decrement* the counter. We expect to see zero or small negatives. 1525 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true); 1526 1527 bind (profile_continue); 1528 } 1529 } 1530 1531 // Count a ret in the bytecodes. 1532 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, Register scratch1, Register scratch2) { 1533 if (ProfileInterpreter) { 1534 Label profile_continue; 1535 uint row; 1536 1537 // If no method data exists, go to profile_continue. 1538 test_method_data_pointer(profile_continue); 1539 1540 // Update the total ret count. 1541 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 ); 1542 1543 for (row = 0; row < RetData::row_limit(); row++) { 1544 Label next_test; 1545 1546 // See if return_bci is equal to bci[n]: 1547 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1); 1548 1549 // return_bci is equal to bci[n]. Increment the count. 1550 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2); 1551 1552 // The method data pointer needs to be updated to reflect the new target. 1553 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1); 1554 b(profile_continue); 1555 bind(next_test); 1556 } 1557 1558 update_mdp_for_ret(state, return_bci); 1559 1560 bind (profile_continue); 1561 } 1562 } 1563 1564 // Count the default case of a switch construct. 1565 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) { 1566 if (ProfileInterpreter) { 1567 Label profile_continue; 1568 1569 // If no method data exists, go to profile_continue. 1570 test_method_data_pointer(profile_continue); 1571 1572 // Update the default case count 1573 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1574 scratch1, scratch2); 1575 1576 // The method data pointer needs to be updated. 1577 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()), 1578 scratch1); 1579 1580 bind (profile_continue); 1581 } 1582 } 1583 1584 // Count the index'th case of a switch construct. 1585 void InterpreterMacroAssembler::profile_switch_case(Register index, 1586 Register scratch1, 1587 Register scratch2, 1588 Register scratch3) { 1589 if (ProfileInterpreter) { 1590 assert_different_registers(index, scratch1, scratch2, scratch3); 1591 Label profile_continue; 1592 1593 // If no method data exists, go to profile_continue. 1594 test_method_data_pointer(profile_continue); 1595 1596 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes(). 1597 li(scratch3, in_bytes(MultiBranchData::case_array_offset())); 1598 1599 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works"); 1600 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1601 add(scratch1, scratch1, scratch3); 1602 1603 // Update the case count. 1604 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3); 1605 1606 // The method data pointer needs to be updated. 1607 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2); 1608 1609 bind (profile_continue); 1610 } 1611 } 1612 1613 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) { 1614 if (ProfileInterpreter) { 1615 assert_different_registers(Rscratch1, Rscratch2); 1616 Label profile_continue; 1617 1618 // If no method data exists, go to profile_continue. 1619 test_method_data_pointer(profile_continue); 1620 1621 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1); 1622 1623 // The method data pointer needs to be updated. 1624 int mdp_delta = in_bytes(BitData::bit_data_size()); 1625 if (TypeProfileCasts) { 1626 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1627 } 1628 update_mdp_by_constant(mdp_delta); 1629 1630 bind (profile_continue); 1631 } 1632 } 1633 1634 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver, 1635 Register Rscratch1, Register Rscratch2, 1636 bool is_virtual_call) { 1637 assert(ProfileInterpreter, "must be profiling"); 1638 assert_different_registers(Rreceiver, Rscratch1, Rscratch2); 1639 1640 Label done; 1641 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call); 1642 bind (done); 1643 } 1644 1645 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1646 Register receiver, Register scratch1, Register scratch2, 1647 int start_row, Label& done, bool is_virtual_call) { 1648 if (TypeProfileWidth == 0) { 1649 if (is_virtual_call) { 1650 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1651 } 1652 return; 1653 } 1654 1655 int last_row = VirtualCallData::row_limit() - 1; 1656 assert(start_row <= last_row, "must be work left to do"); 1657 // Test this row for both the receiver and for null. 1658 // Take any of three different outcomes: 1659 // 1. found receiver => increment count and goto done 1660 // 2. found null => keep looking for case 1, maybe allocate this cell 1661 // 3. found something else => keep looking for cases 1 and 2 1662 // Case 3 is handled by a recursive call. 1663 for (int row = start_row; row <= last_row; row++) { 1664 Label next_test; 1665 bool test_for_null_also = (row == start_row); 1666 1667 // See if the receiver is receiver[n]. 1668 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1669 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1); 1670 // delayed()->tst(scratch); 1671 1672 // The receiver is receiver[n]. Increment count[n]. 1673 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1674 increment_mdp_data_at(count_offset, scratch1, scratch2); 1675 b(done); 1676 bind(next_test); 1677 1678 if (test_for_null_also) { 1679 Label found_null; 1680 // Failed the equality check on receiver[n]... Test for null. 1681 if (start_row == last_row) { 1682 // The only thing left to do is handle the null case. 1683 if (is_virtual_call) { 1684 // Scratch1 contains test_out from test_mdp_data_at. 1685 cmpdi(CCR0, scratch1, 0); 1686 beq(CCR0, found_null); 1687 // Receiver did not match any saved receiver and there is no empty row for it. 1688 // Increment total counter to indicate polymorphic case. 1689 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1690 b(done); 1691 bind(found_null); 1692 } else { 1693 cmpdi(CCR0, scratch1, 0); 1694 bne(CCR0, done); 1695 } 1696 break; 1697 } 1698 // Since null is rare, make it be the branch-taken case. 1699 cmpdi(CCR0, scratch1, 0); 1700 beq(CCR0, found_null); 1701 1702 // Put all the "Case 3" tests here. 1703 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call); 1704 1705 // Found a null. Keep searching for a matching receiver, 1706 // but remember that this is an empty (unused) slot. 1707 bind(found_null); 1708 } 1709 } 1710 1711 // In the fall-through case, we found no matching receiver, but we 1712 // observed the receiver[start_row] is NULL. 1713 1714 // Fill in the receiver field and increment the count. 1715 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1716 set_mdp_data_at(recvr_offset, receiver); 1717 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1718 li(scratch1, DataLayout::counter_increment); 1719 set_mdp_data_at(count_offset, scratch1); 1720 if (start_row > 0) { 1721 b(done); 1722 } 1723 } 1724 1725 // Argument and return type profilig. 1726 // kills: tmp, tmp2, R0, CR0, CR1 1727 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base, 1728 RegisterOrConstant mdo_addr_offs, Register tmp, Register tmp2) { 1729 Label do_nothing, do_update; 1730 1731 // tmp2 = obj is allowed 1732 assert_different_registers(obj, mdo_addr_base, tmp, R0); 1733 assert_different_registers(tmp2, mdo_addr_base, tmp, R0); 1734 const Register klass = tmp2; 1735 1736 verify_oop(obj); 1737 1738 ld(tmp, mdo_addr_offs, mdo_addr_base); 1739 1740 // Set null_seen if obj is 0. 1741 cmpdi(CCR0, obj, 0); 1742 ori(R0, tmp, TypeEntries::null_seen); 1743 beq(CCR0, do_update); 1744 1745 load_klass(klass, obj); 1746 1747 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask)); 1748 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask); 1749 cmpd(CCR1, R0, klass); 1750 // Klass seen before, nothing to do (regardless of unknown bit). 1751 //beq(CCR1, do_nothing); 1752 1753 andi_(R0, klass, TypeEntries::type_unknown); 1754 // Already unknown. Nothing to do anymore. 1755 //bne(CCR0, do_nothing); 1756 crorc(/*CCR0 eq*/2, /*CCR1 eq*/4+2, /*CCR0 eq*/2); // cr0 eq = cr1 eq or cr0 ne 1757 beq(CCR0, do_nothing); 1758 1759 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask)); 1760 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1761 beq(CCR0, do_update); // First time here. Set profile type. 1762 1763 // Different than before. Cannot keep accurate profile. 1764 ori(R0, tmp, TypeEntries::type_unknown); 1765 1766 bind(do_update); 1767 // update profile 1768 std(R0, mdo_addr_offs, mdo_addr_base); 1769 1770 align(32, 12); 1771 bind(do_nothing); 1772 } 1773 1774 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) { 1775 if (!ProfileInterpreter) { 1776 return; 1777 } 1778 1779 assert_different_registers(callee, tmp1, tmp2, R28_mdx); 1780 1781 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1782 Label profile_continue; 1783 1784 test_method_data_pointer(profile_continue); 1785 1786 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1787 1788 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx); 1789 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1790 bne(CCR0, profile_continue); 1791 1792 if (MethodData::profile_arguments()) { 1793 Label done; 1794 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1795 add(R28_mdx, off_to_args, R28_mdx); 1796 1797 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1798 if (i > 0 || MethodData::profile_return()) { 1799 // If return value type is profiled we may have no argument to profile. 1800 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1801 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count()); 1802 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count()); 1803 blt(CCR0, done); 1804 } 1805 ld(tmp1, in_bytes(Method::const_offset()), callee); 1806 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1); 1807 // Stack offset o (zero based) from the start of the argument 1808 // list, for n arguments translates into offset n - o - 1 from 1809 // the end of the argument list. But there's an extra slot at 1810 // the top of the stack. So the offset is n - o from Lesp. 1811 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx); 1812 subf(tmp1, tmp2, tmp1); 1813 1814 sldi(tmp1, tmp1, Interpreter::logStackElementSize); 1815 ldx(tmp1, tmp1, R15_esp); 1816 1817 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1); 1818 1819 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1820 addi(R28_mdx, R28_mdx, to_add); 1821 off_to_args += to_add; 1822 } 1823 1824 if (MethodData::profile_return()) { 1825 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1826 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1827 } 1828 1829 bind(done); 1830 1831 if (MethodData::profile_return()) { 1832 // We're right after the type profile for the last 1833 // argument. tmp1 is the number of cells left in the 1834 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1835 // if there's a return to profile. 1836 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 1837 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size)); 1838 add(R28_mdx, tmp1, R28_mdx); 1839 } 1840 } else { 1841 assert(MethodData::profile_return(), "either profile call args or call ret"); 1842 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 1843 } 1844 1845 // Mdp points right after the end of the 1846 // CallTypeData/VirtualCallTypeData, right after the cells for the 1847 // return value type if there's one. 1848 align(32, 12); 1849 bind(profile_continue); 1850 } 1851 } 1852 1853 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 1854 assert_different_registers(ret, tmp1, tmp2); 1855 if (ProfileInterpreter && MethodData::profile_return()) { 1856 Label profile_continue; 1857 1858 test_method_data_pointer(profile_continue); 1859 1860 if (MethodData::profile_return_jsr292_only()) { 1861 // If we don't profile all invoke bytecodes we must make sure 1862 // it's a bytecode we indeed profile. We can't go back to the 1863 // begining of the ProfileData we intend to update to check its 1864 // type because we're right after it and we don't known its 1865 // length. 1866 lbz(tmp1, 0, R14_bcp); 1867 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method); 1868 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic); 1869 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle); 1870 cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2); 1871 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm); 1872 cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2); 1873 bne(CCR0, profile_continue); 1874 } 1875 1876 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2); 1877 1878 align(32, 12); 1879 bind(profile_continue); 1880 } 1881 } 1882 1883 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) { 1884 if (ProfileInterpreter && MethodData::profile_parameters()) { 1885 Label profile_continue, done; 1886 1887 test_method_data_pointer(profile_continue); 1888 1889 // Load the offset of the area within the MDO used for 1890 // parameters. If it's negative we're not profiling any parameters. 1891 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx); 1892 cmpwi(CCR0, tmp1, 0); 1893 blt(CCR0, profile_continue); 1894 1895 // Compute a pointer to the area for parameters from the offset 1896 // and move the pointer to the slot for the last 1897 // parameters. Collect profiling from last parameter down. 1898 // mdo start + parameters offset + array length - 1 1899 1900 // Pointer to the parameter area in the MDO. 1901 const Register mdp = tmp1; 1902 add(mdp, tmp1, R28_mdx); 1903 1904 // Pffset of the current profile entry to update. 1905 const Register entry_offset = tmp2; 1906 // entry_offset = array len in number of cells 1907 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp); 1908 1909 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1910 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 1911 1912 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 1913 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size)); 1914 // entry_offset in bytes 1915 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1916 1917 Label loop; 1918 align(32, 12); 1919 bind(loop); 1920 1921 // Load offset on the stack from the slot for this parameter. 1922 ld(tmp3, entry_offset, mdp); 1923 sldi(tmp3, tmp3, Interpreter::logStackElementSize); 1924 neg(tmp3, tmp3); 1925 // Read the parameter from the local area. 1926 ldx(tmp3, tmp3, R18_locals); 1927 1928 // Make entry_offset now point to the type field for this parameter. 1929 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1930 assert(type_base > off_base, "unexpected"); 1931 addi(entry_offset, entry_offset, type_base - off_base); 1932 1933 // Profile the parameter. 1934 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3); 1935 1936 // Go to next parameter. 1937 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base); 1938 cmpdi(CCR0, entry_offset, off_base + delta); 1939 addi(entry_offset, entry_offset, -delta); 1940 bge(CCR0, loop); 1941 1942 align(32, 12); 1943 bind(profile_continue); 1944 } 1945 } 1946 1947 // Add a InterpMonitorElem to stack (see frame_sparc.hpp). 1948 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) { 1949 1950 // Very-local scratch registers. 1951 const Register esp = Rtemp1; 1952 const Register slot = Rtemp2; 1953 1954 // Extracted monitor_size. 1955 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes(); 1956 assert(Assembler::is_aligned((unsigned int)monitor_size, 1957 (unsigned int)frame::alignment_in_bytes), 1958 "size of a monitor must respect alignment of SP"); 1959 1960 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor 1961 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp 1962 1963 // Shuffle expression stack down. Recall that stack_base points 1964 // just above the new expression stack bottom. Old_tos and new_tos 1965 // are used to scan thru the old and new expression stacks. 1966 if (!stack_is_empty) { 1967 Label copy_slot, copy_slot_finished; 1968 const Register n_slots = slot; 1969 1970 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack). 1971 subf(n_slots, esp, R26_monitor); 1972 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy. 1973 assert(LogBytesPerWord == 3, "conflicts assembler instructions"); 1974 beq(CCR0, copy_slot_finished); // Nothing to copy. 1975 1976 mtctr(n_slots); 1977 1978 // loop 1979 bind(copy_slot); 1980 ld(slot, 0, esp); // Move expression stack down. 1981 std(slot, -monitor_size, esp); // distance = monitor_size 1982 addi(esp, esp, BytesPerWord); 1983 bdnz(copy_slot); 1984 1985 bind(copy_slot_finished); 1986 } 1987 1988 addi(R15_esp, R15_esp, -monitor_size); 1989 addi(R26_monitor, R26_monitor, -monitor_size); 1990 1991 // Restart interpreter 1992 } 1993 1994 // ============================================================================ 1995 // Java locals access 1996 1997 // Load a local variable at index in Rindex into register Rdst_value. 1998 // Also puts address of local into Rdst_address as a service. 1999 // Kills: 2000 // - Rdst_value 2001 // - Rdst_address 2002 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) { 2003 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2004 subf(Rdst_address, Rdst_address, R18_locals); 2005 lwz(Rdst_value, 0, Rdst_address); 2006 } 2007 2008 // Load a local variable at index in Rindex into register Rdst_value. 2009 // Also puts address of local into Rdst_address as a service. 2010 // Kills: 2011 // - Rdst_value 2012 // - Rdst_address 2013 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) { 2014 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2015 subf(Rdst_address, Rdst_address, R18_locals); 2016 ld(Rdst_value, -8, Rdst_address); 2017 } 2018 2019 // Load a local variable at index in Rindex into register Rdst_value. 2020 // Also puts address of local into Rdst_address as a service. 2021 // Input: 2022 // - Rindex: slot nr of local variable 2023 // Kills: 2024 // - Rdst_value 2025 // - Rdst_address 2026 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, Register Rdst_address, Register Rindex) { 2027 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2028 subf(Rdst_address, Rdst_address, R18_locals); 2029 ld(Rdst_value, 0, Rdst_address); 2030 } 2031 2032 // Load a local variable at index in Rindex into register Rdst_value. 2033 // Also puts address of local into Rdst_address as a service. 2034 // Kills: 2035 // - Rdst_value 2036 // - Rdst_address 2037 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) { 2038 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2039 subf(Rdst_address, Rdst_address, R18_locals); 2040 lfs(Rdst_value, 0, Rdst_address); 2041 } 2042 2043 // Load a local variable at index in Rindex into register Rdst_value. 2044 // Also puts address of local into Rdst_address as a service. 2045 // Kills: 2046 // - Rdst_value 2047 // - Rdst_address 2048 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) { 2049 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2050 subf(Rdst_address, Rdst_address, R18_locals); 2051 lfd(Rdst_value, -8, Rdst_address); 2052 } 2053 2054 // Store an int value at local variable slot Rindex. 2055 // Kills: 2056 // - Rindex 2057 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) { 2058 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2059 subf(Rindex, Rindex, R18_locals); 2060 stw(Rvalue, 0, Rindex); 2061 } 2062 2063 // Store a long value at local variable slot Rindex. 2064 // Kills: 2065 // - Rindex 2066 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) { 2067 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2068 subf(Rindex, Rindex, R18_locals); 2069 std(Rvalue, -8, Rindex); 2070 } 2071 2072 // Store an oop value at local variable slot Rindex. 2073 // Kills: 2074 // - Rindex 2075 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) { 2076 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2077 subf(Rindex, Rindex, R18_locals); 2078 std(Rvalue, 0, Rindex); 2079 } 2080 2081 // Store an int value at local variable slot Rindex. 2082 // Kills: 2083 // - Rindex 2084 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) { 2085 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2086 subf(Rindex, Rindex, R18_locals); 2087 stfs(Rvalue, 0, Rindex); 2088 } 2089 2090 // Store an int value at local variable slot Rindex. 2091 // Kills: 2092 // - Rindex 2093 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) { 2094 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2095 subf(Rindex, Rindex, R18_locals); 2096 stfd(Rvalue, -8, Rindex); 2097 } 2098 2099 // Read pending exception from thread and jump to interpreter. 2100 // Throw exception entry if one if pending. Fall through otherwise. 2101 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) { 2102 assert_different_registers(Rscratch1, Rscratch2, R3); 2103 Register Rexception = Rscratch1; 2104 Register Rtmp = Rscratch2; 2105 Label Ldone; 2106 // Get pending exception oop. 2107 ld(Rexception, thread_(pending_exception)); 2108 cmpdi(CCR0, Rexception, 0); 2109 beq(CCR0, Ldone); 2110 li(Rtmp, 0); 2111 mr_if_needed(R3, Rexception); 2112 std(Rtmp, thread_(pending_exception)); // Clear exception in thread 2113 if (Interpreter::rethrow_exception_entry() != NULL) { 2114 // Already got entry address. 2115 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry()); 2116 } else { 2117 // Dynamically load entry address. 2118 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true); 2119 ld(Rtmp, simm16_rest, Rtmp); 2120 } 2121 mtctr(Rtmp); 2122 save_interpreter_state(Rtmp); 2123 bctr(); 2124 2125 align(32, 12); 2126 bind(Ldone); 2127 } 2128 2129 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) { 2130 save_interpreter_state(R11_scratch1); 2131 2132 MacroAssembler::call_VM(oop_result, entry_point, false); 2133 2134 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2135 2136 check_and_handle_popframe(R11_scratch1); 2137 check_and_handle_earlyret(R11_scratch1); 2138 // Now check exceptions manually. 2139 if (check_exceptions) { 2140 check_and_forward_exception(R11_scratch1, R12_scratch2); 2141 } 2142 } 2143 2144 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions) { 2145 // ARG1 is reserved for the thread. 2146 mr_if_needed(R4_ARG2, arg_1); 2147 call_VM(oop_result, entry_point, check_exceptions); 2148 } 2149 2150 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) { 2151 // ARG1 is reserved for the thread. 2152 mr_if_needed(R4_ARG2, arg_1); 2153 assert(arg_2 != R4_ARG2, "smashed argument"); 2154 mr_if_needed(R5_ARG3, arg_2); 2155 call_VM(oop_result, entry_point, check_exceptions); 2156 } 2157 2158 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) { 2159 // ARG1 is reserved for the thread. 2160 mr_if_needed(R4_ARG2, arg_1); 2161 assert(arg_2 != R4_ARG2, "smashed argument"); 2162 mr_if_needed(R5_ARG3, arg_2); 2163 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument"); 2164 mr_if_needed(R6_ARG4, arg_3); 2165 call_VM(oop_result, entry_point, check_exceptions); 2166 } 2167 2168 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) { 2169 ld(scratch, 0, R1_SP); 2170 std(R15_esp, _ijava_state_neg(esp), scratch); 2171 std(R14_bcp, _ijava_state_neg(bcp), scratch); 2172 std(R26_monitor, _ijava_state_neg(monitors), scratch); 2173 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); } 2174 // Other entries should be unchanged. 2175 } 2176 2177 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) { 2178 ld(scratch, 0, R1_SP); 2179 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception). 2180 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code. 2181 if (!bcp_and_mdx_only) { 2182 // Following ones are Metadata. 2183 ld(R19_method, _ijava_state_neg(method), scratch); 2184 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch); 2185 // Following ones are stack addresses and don't require reload. 2186 ld(R15_esp, _ijava_state_neg(esp), scratch); 2187 ld(R18_locals, _ijava_state_neg(locals), scratch); 2188 ld(R26_monitor, _ijava_state_neg(monitors), scratch); 2189 } 2190 #ifdef ASSERT 2191 { 2192 Label Lok; 2193 subf(R0, R1_SP, scratch); 2194 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size); 2195 bge(CCR0, Lok); 2196 stop("frame too small (restore istate)", 0x5432); 2197 bind(Lok); 2198 } 2199 { 2200 Label Lok; 2201 ld(R0, _ijava_state_neg(ijava_reserved), scratch); 2202 cmpdi(CCR0, R0, 0x5afe); 2203 beq(CCR0, Lok); 2204 stop("frame corrupted (restore istate)", 0x5afe); 2205 bind(Lok); 2206 } 2207 #endif 2208 } 2209 2210 #endif // !CC_INTERP 2211 2212 void InterpreterMacroAssembler::get_method_counters(Register method, 2213 Register Rcounters, 2214 Label& skip) { 2215 BLOCK_COMMENT("Load and ev. allocate counter object {"); 2216 Label has_counters; 2217 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2218 cmpdi(CCR0, Rcounters, 0); 2219 bne(CCR0, has_counters); 2220 call_VM(noreg, CAST_FROM_FN_PTR(address, 2221 InterpreterRuntime::build_method_counters), method, false); 2222 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2223 cmpdi(CCR0, Rcounters, 0); 2224 beq(CCR0, skip); // No MethodCounters, OutOfMemory. 2225 BLOCK_COMMENT("} Load and ev. allocate counter object"); 2226 2227 bind(has_counters); 2228 } 2229 2230 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register iv_be_count, Register Rtmp_r0) { 2231 assert(UseCompiler, "incrementing must be useful"); 2232 Register invocation_count = iv_be_count; 2233 Register backedge_count = Rtmp_r0; 2234 int delta = InvocationCounter::count_increment; 2235 2236 // Load each counter in a register. 2237 // ld(inv_counter, Rtmp); 2238 // ld(be_counter, Rtmp2); 2239 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + 2240 InvocationCounter::counter_offset()); 2241 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + 2242 InvocationCounter::counter_offset()); 2243 2244 BLOCK_COMMENT("Increment profiling counters {"); 2245 2246 // Load the backedge counter. 2247 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int 2248 // Mask the backedge counter. 2249 Register tmp = invocation_count; 2250 li(tmp, InvocationCounter::count_mask_value); 2251 andr(backedge_count, tmp, backedge_count); // Cannot use andi, need sign extension of count_mask_value. 2252 2253 // Load the invocation counter. 2254 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int 2255 // Add the delta to the invocation counter and store the result. 2256 addi(invocation_count, invocation_count, delta); 2257 // Store value. 2258 stw(invocation_count, inv_counter_offset, Rcounters); 2259 2260 // Add invocation counter + backedge counter. 2261 add(iv_be_count, backedge_count, invocation_count); 2262 2263 // Note that this macro must leave the backedge_count + invocation_count in 2264 // register iv_be_count! 2265 BLOCK_COMMENT("} Increment profiling counters"); 2266 } 2267 2268 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2269 if (state == atos) { MacroAssembler::verify_oop(reg); } 2270 } 2271 2272 #ifndef CC_INTERP 2273 // Local helper function for the verify_oop_or_return_address macro. 2274 static bool verify_return_address(Method* m, int bci) { 2275 #ifndef PRODUCT 2276 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci; 2277 // Assume it is a valid return address if it is inside m and is preceded by a jsr. 2278 if (!m->contains(pc)) return false; 2279 address jsr_pc; 2280 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2281 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2282 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2283 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2284 #endif // PRODUCT 2285 return false; 2286 } 2287 2288 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2289 if (VerifyFPU) { 2290 unimplemented("verfiyFPU"); 2291 } 2292 } 2293 2294 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2295 if (!VerifyOops) return; 2296 2297 // The VM documentation for the astore[_wide] bytecode allows 2298 // the TOS to be not only an oop but also a return address. 2299 Label test; 2300 Label skip; 2301 // See if it is an address (in the current method): 2302 2303 const int log2_bytecode_size_limit = 16; 2304 srdi_(Rtmp, reg, log2_bytecode_size_limit); 2305 bne(CCR0, test); 2306 2307 address fd = CAST_FROM_FN_PTR(address, verify_return_address); 2308 const int nbytes_save = 11*8; // volatile gprs except R0 2309 save_volatile_gprs(R1_SP, -nbytes_save); // except R0 2310 save_LR_CR(Rtmp); // Save in old frame. 2311 push_frame_reg_args(nbytes_save, Rtmp); 2312 2313 load_const_optimized(Rtmp, fd, R0); 2314 mr_if_needed(R4_ARG2, reg); 2315 mr(R3_ARG1, R19_method); 2316 call_c(Rtmp); // call C 2317 2318 pop_frame(); 2319 restore_LR_CR(Rtmp); 2320 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0 2321 b(skip); 2322 2323 // Perform a more elaborate out-of-line call. 2324 // Not an address; verify it: 2325 bind(test); 2326 verify_oop(reg); 2327 bind(skip); 2328 } 2329 #endif // !CC_INTERP 2330 2331 // Inline assembly for: 2332 // 2333 // if (thread is in interp_only_mode) { 2334 // InterpreterRuntime::post_method_entry(); 2335 // } 2336 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) || 2337 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) { 2338 // SharedRuntime::jvmpi_method_entry(method, receiver); 2339 // } 2340 void InterpreterMacroAssembler::notify_method_entry() { 2341 // JVMTI 2342 // Whenever JVMTI puts a thread in interp_only_mode, method 2343 // entry/exit events are sent for that thread to track stack 2344 // depth. If it is possible to enter interp_only_mode we add 2345 // the code to check if the event should be sent. 2346 if (JvmtiExport::can_post_interpreter_events()) { 2347 Label jvmti_post_done; 2348 2349 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2350 cmpwi(CCR0, R0, 0); 2351 beq(CCR0, jvmti_post_done); 2352 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), 2353 /*check_exceptions=*/true CC_INTERP_ONLY(&& false)); 2354 2355 bind(jvmti_post_done); 2356 } 2357 } 2358 2359 // Inline assembly for: 2360 // 2361 // if (thread is in interp_only_mode) { 2362 // // save result 2363 // InterpreterRuntime::post_method_exit(); 2364 // // restore result 2365 // } 2366 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) { 2367 // // save result 2368 // SharedRuntime::jvmpi_method_exit(); 2369 // // restore result 2370 // } 2371 // 2372 // Native methods have their result stored in d_tmp and l_tmp. 2373 // Java methods have their result stored in the expression stack. 2374 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state, 2375 NotifyMethodExitMode mode, bool check_exceptions) { 2376 // JVMTI 2377 // Whenever JVMTI puts a thread in interp_only_mode, method 2378 // entry/exit events are sent for that thread to track stack 2379 // depth. If it is possible to enter interp_only_mode we add 2380 // the code to check if the event should be sent. 2381 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2382 Label jvmti_post_done; 2383 2384 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2385 cmpwi(CCR0, R0, 0); 2386 beq(CCR0, jvmti_post_done); 2387 CC_INTERP_ONLY(assert(is_native_method && !check_exceptions, "must not push state")); 2388 if (!is_native_method) push(state); // Expose tos to GC. 2389 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), 2390 /*check_exceptions=*/check_exceptions); 2391 if (!is_native_method) pop(state); 2392 2393 align(32, 12); 2394 bind(jvmti_post_done); 2395 } 2396 2397 // Dtrace support not implemented. 2398 } 2399 2400 #ifdef CC_INTERP 2401 // Convert the current TOP_IJAVA_FRAME into a PARENT_IJAVA_FRAME 2402 // (using parent_frame_resize) and push a new interpreter 2403 // TOP_IJAVA_FRAME (using frame_size). 2404 void InterpreterMacroAssembler::push_interpreter_frame(Register top_frame_size, Register parent_frame_resize, 2405 Register tmp1, Register tmp2, Register tmp3, 2406 Register tmp4, Register pc) { 2407 assert_different_registers(top_frame_size, parent_frame_resize, tmp1, tmp2, tmp3, tmp4); 2408 ld(tmp1, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2409 mr(tmp2/*top_frame_sp*/, R1_SP); 2410 // Move initial_caller_sp. 2411 ld(tmp4, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2412 neg(parent_frame_resize, parent_frame_resize); 2413 resize_frame(parent_frame_resize/*-parent_frame_resize*/, tmp3); 2414 2415 // Set LR in new parent frame. 2416 std(tmp1, _abi(lr), R1_SP); 2417 // Set top_frame_sp info for new parent frame. 2418 std(tmp2, _parent_ijava_frame_abi(top_frame_sp), R1_SP); 2419 std(tmp4, _parent_ijava_frame_abi(initial_caller_sp), R1_SP); 2420 2421 // Push new TOP_IJAVA_FRAME. 2422 push_frame(top_frame_size, tmp2); 2423 2424 get_PC_trash_LR(tmp3); 2425 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2426 // Used for non-initial callers by unextended_sp(). 2427 std(R1_SP, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2428 } 2429 2430 // Pop the topmost TOP_IJAVA_FRAME and convert the previous 2431 // PARENT_IJAVA_FRAME back into a TOP_IJAVA_FRAME. 2432 void InterpreterMacroAssembler::pop_interpreter_frame(Register tmp1, Register tmp2, Register tmp3, Register tmp4) { 2433 assert_different_registers(tmp1, tmp2, tmp3, tmp4); 2434 2435 ld(tmp1/*caller's sp*/, _abi(callers_sp), R1_SP); 2436 ld(tmp3, _abi(lr), tmp1); 2437 2438 ld(tmp4, _parent_ijava_frame_abi(initial_caller_sp), tmp1); 2439 2440 ld(tmp2/*caller's caller's sp*/, _abi(callers_sp), tmp1); 2441 // Merge top frame. 2442 std(tmp2, _abi(callers_sp), R1_SP); 2443 2444 ld(tmp2, _parent_ijava_frame_abi(top_frame_sp), tmp1); 2445 2446 // Update C stack pointer to caller's top_abi. 2447 resize_frame_absolute(tmp2/*addr*/, tmp1/*tmp*/, tmp2/*tmp*/); 2448 2449 // Update LR in top_frame. 2450 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2451 2452 std(tmp4, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2453 2454 // Store the top-frame stack-pointer for c2i adapters. 2455 std(R1_SP, _top_ijava_frame_abi(top_frame_sp), R1_SP); 2456 } 2457 2458 // Turn state's interpreter frame into the current TOP_IJAVA_FRAME. 2459 void InterpreterMacroAssembler::pop_interpreter_frame_to_state(Register state, Register tmp1, Register tmp2, Register tmp3) { 2460 assert_different_registers(R14_state, R15_prev_state, tmp1, tmp2, tmp3); 2461 2462 if (state == R14_state) { 2463 ld(tmp1/*state's fp*/, state_(_last_Java_fp)); 2464 ld(tmp2/*state's sp*/, state_(_last_Java_sp)); 2465 } else if (state == R15_prev_state) { 2466 ld(tmp1/*state's fp*/, prev_state_(_last_Java_fp)); 2467 ld(tmp2/*state's sp*/, prev_state_(_last_Java_sp)); 2468 } else { 2469 ShouldNotReachHere(); 2470 } 2471 2472 // Merge top frames. 2473 std(tmp1, _abi(callers_sp), R1_SP); 2474 2475 // Tmp2 is new SP. 2476 // Tmp1 is parent's SP. 2477 resize_frame_absolute(tmp2/*addr*/, tmp1/*tmp*/, tmp2/*tmp*/); 2478 2479 // Update LR in top_frame. 2480 // Must be interpreter frame. 2481 get_PC_trash_LR(tmp3); 2482 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2483 // Used for non-initial callers by unextended_sp(). 2484 std(R1_SP, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2485 } 2486 2487 // Set SP to initial caller's sp, but before fix the back chain. 2488 void InterpreterMacroAssembler::resize_frame_to_initial_caller(Register tmp1, Register tmp2) { 2489 ld(tmp1, _parent_ijava_frame_abi(initial_caller_sp), R1_SP); 2490 ld(tmp2, _parent_ijava_frame_abi(callers_sp), R1_SP); 2491 std(tmp2, _parent_ijava_frame_abi(callers_sp), tmp1); // Fix back chain ... 2492 mr(R1_SP, tmp1); // ... and resize to initial caller. 2493 } 2494 2495 // Pop the current interpreter state (without popping the correspoding 2496 // frame) and restore R14_state and R15_prev_state accordingly. 2497 // Use prev_state_may_be_0 to indicate whether prev_state may be 0 2498 // in order to generate an extra check before retrieving prev_state_(_prev_link). 2499 void InterpreterMacroAssembler::pop_interpreter_state(bool prev_state_may_be_0) 2500 { 2501 // Move prev_state to state and restore prev_state from state_(_prev_link). 2502 Label prev_state_is_0; 2503 mr(R14_state, R15_prev_state); 2504 2505 // Don't retrieve /*state==*/prev_state_(_prev_link) 2506 // if /*state==*/prev_state is 0. 2507 if (prev_state_may_be_0) { 2508 cmpdi(CCR0, R15_prev_state, 0); 2509 beq(CCR0, prev_state_is_0); 2510 } 2511 2512 ld(R15_prev_state, /*state==*/prev_state_(_prev_link)); 2513 bind(prev_state_is_0); 2514 } 2515 2516 void InterpreterMacroAssembler::restore_prev_state() { 2517 // _prev_link is private, but cInterpreter is a friend. 2518 ld(R15_prev_state, state_(_prev_link)); 2519 } 2520 #endif // CC_INTERP