interp_masm_ppc_64.cpp revision 11374:3fb9a97eb099
1/* 2 * Copyright (c) 2003, 2015, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2012, 2016 SAP SE. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 27#include "precompiled.hpp" 28#include "asm/macroAssembler.inline.hpp" 29#include "interp_masm_ppc_64.hpp" 30#include "interpreter/interpreterRuntime.hpp" 31#include "prims/jvmtiThreadState.hpp" 32#include "runtime/sharedRuntime.hpp" 33 34#ifdef PRODUCT 35#define BLOCK_COMMENT(str) // nothing 36#else 37#define BLOCK_COMMENT(str) block_comment(str) 38#endif 39 40void InterpreterMacroAssembler::null_check_throw(Register a, int offset, Register temp_reg) { 41 address exception_entry = Interpreter::throw_NullPointerException_entry(); 42 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry); 43} 44 45void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) { 46 assert(entry, "Entry must have been generated by now"); 47 if (is_within_range_of_b(entry, pc())) { 48 b(entry); 49 } else { 50 load_const_optimized(Rscratch, entry, R0); 51 mtctr(Rscratch); 52 bctr(); 53 } 54} 55 56void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { 57 Register bytecode = R12_scratch2; 58 if (bcp_incr != 0) { 59 lbzu(bytecode, bcp_incr, R14_bcp); 60 } else { 61 lbz(bytecode, 0, R14_bcp); 62 } 63 64 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state)); 65} 66 67void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 68 // Load current bytecode. 69 Register bytecode = R12_scratch2; 70 lbz(bytecode, 0, R14_bcp); 71 dispatch_Lbyte_code(state, bytecode, table); 72} 73 74// Dispatch code executed in the prolog of a bytecode which does not do it's 75// own dispatch. The dispatch address is computed and placed in R24_dispatch_addr. 76void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 77 Register bytecode = R12_scratch2; 78 lbz(bytecode, bcp_incr, R14_bcp); 79 80 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state)); 81 82 sldi(bytecode, bytecode, LogBytesPerWord); 83 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode); 84} 85 86// Dispatch code executed in the epilog of a bytecode which does not do it's 87// own dispatch. The dispatch address in R24_dispatch_addr is used for the 88// dispatch. 89void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 90 if (bcp_incr) { addi(R14_bcp, R14_bcp, bcp_incr); } 91 mtctr(R24_dispatch_addr); 92 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable); 93} 94 95void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 96 assert(scratch_reg != R0, "can't use R0 as scratch_reg here"); 97 if (JvmtiExport::can_pop_frame()) { 98 Label L; 99 100 // Check the "pending popframe condition" flag in the current thread. 101 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 102 103 // Initiate popframe handling only if it is not already being 104 // processed. If the flag has the popframe_processing bit set, it 105 // means that this code is called *during* popframe handling - we 106 // don't want to reenter. 107 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit); 108 beq(CCR0, L); 109 110 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit); 111 bne(CCR0, L); 112 113 // Call the Interpreter::remove_activation_preserving_args_entry() 114 // func to get the address of the same-named entrypoint in the 115 // generated interpreter code. 116#if defined(ABI_ELFv2) 117 call_c(CAST_FROM_FN_PTR(address, 118 Interpreter::remove_activation_preserving_args_entry), 119 relocInfo::none); 120#else 121 call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, 122 Interpreter::remove_activation_preserving_args_entry), 123 relocInfo::none); 124#endif 125 126 // Jump to Interpreter::_remove_activation_preserving_args_entry. 127 mtctr(R3_RET); 128 bctr(); 129 130 align(32, 12); 131 bind(L); 132 } 133} 134 135void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 136 const Register Rthr_state_addr = scratch_reg; 137 if (JvmtiExport::can_force_early_return()) { 138 Label Lno_early_ret; 139 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 140 cmpdi(CCR0, Rthr_state_addr, 0); 141 beq(CCR0, Lno_early_ret); 142 143 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr); 144 cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending); 145 bne(CCR0, Lno_early_ret); 146 147 // Jump to Interpreter::_earlyret_entry. 148 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr); 149 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry)); 150 mtlr(R3_RET); 151 blr(); 152 153 align(32, 12); 154 bind(Lno_early_ret); 155 } 156} 157 158void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) { 159 const Register RjvmtiState = Rscratch1; 160 const Register Rscratch2 = R0; 161 162 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 163 li(Rscratch2, 0); 164 165 switch (state) { 166 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 167 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 168 break; 169 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 170 break; 171 case btos: // fall through 172 case ztos: // fall through 173 case ctos: // fall through 174 case stos: // fall through 175 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 176 break; 177 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 178 break; 179 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 180 break; 181 case vtos: break; 182 default : ShouldNotReachHere(); 183 } 184 185 // Clean up tos value in the jvmti thread state. 186 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 187 // Set tos state field to illegal value. 188 li(Rscratch2, ilgl); 189 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState); 190} 191 192// Common code to dispatch and dispatch_only. 193// Dispatch value in Lbyte_code and increment Lbcp. 194 195void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) { 196 address table_base = (address)Interpreter::dispatch_table((TosState)0); 197 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base; 198 if (is_simm16(table_offs)) { 199 addi(dst, R25_templateTableBase, (int)table_offs); 200 } else { 201 load_const_optimized(dst, table, R0); 202 } 203} 204 205void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode, 206 address* table, bool verify) { 207 if (verify) { 208 unimplemented("dispatch_Lbyte_code: verify"); // See Sparc Implementation to implement this 209 } 210 211 assert_different_registers(bytecode, R11_scratch1); 212 213 // Calc dispatch table address. 214 load_dispatch_table(R11_scratch1, table); 215 216 sldi(R12_scratch2, bytecode, LogBytesPerWord); 217 ldx(R11_scratch1, R11_scratch1, R12_scratch2); 218 219 // Jump off! 220 mtctr(R11_scratch1); 221 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable); 222} 223 224void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) { 225 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize); 226 ldx(Rrecv_dst, Rrecv_dst, R15_esp); 227} 228 229// helpers for expression stack 230 231void InterpreterMacroAssembler::pop_i(Register r) { 232 lwzu(r, Interpreter::stackElementSize, R15_esp); 233} 234 235void InterpreterMacroAssembler::pop_ptr(Register r) { 236 ldu(r, Interpreter::stackElementSize, R15_esp); 237} 238 239void InterpreterMacroAssembler::pop_l(Register r) { 240 ld(r, Interpreter::stackElementSize, R15_esp); 241 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 242} 243 244void InterpreterMacroAssembler::pop_f(FloatRegister f) { 245 lfsu(f, Interpreter::stackElementSize, R15_esp); 246} 247 248void InterpreterMacroAssembler::pop_d(FloatRegister f) { 249 lfd(f, Interpreter::stackElementSize, R15_esp); 250 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 251} 252 253void InterpreterMacroAssembler::push_i(Register r) { 254 stw(r, 0, R15_esp); 255 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 256} 257 258void InterpreterMacroAssembler::push_ptr(Register r) { 259 std(r, 0, R15_esp); 260 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 261} 262 263void InterpreterMacroAssembler::push_l(Register r) { 264 std(r, - Interpreter::stackElementSize, R15_esp); 265 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 266} 267 268void InterpreterMacroAssembler::push_f(FloatRegister f) { 269 stfs(f, 0, R15_esp); 270 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 271} 272 273void InterpreterMacroAssembler::push_d(FloatRegister f) { 274 stfd(f, - Interpreter::stackElementSize, R15_esp); 275 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 276} 277 278void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) { 279 std(first, 0, R15_esp); 280 std(second, -Interpreter::stackElementSize, R15_esp); 281 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 282} 283 284void InterpreterMacroAssembler::push_l_pop_d(Register l, FloatRegister d) { 285 std(l, 0, R15_esp); 286 lfd(d, 0, R15_esp); 287} 288 289void InterpreterMacroAssembler::push_d_pop_l(FloatRegister d, Register l) { 290 stfd(d, 0, R15_esp); 291 ld(l, 0, R15_esp); 292} 293 294void InterpreterMacroAssembler::push(TosState state) { 295 switch (state) { 296 case atos: push_ptr(); break; 297 case btos: 298 case ztos: 299 case ctos: 300 case stos: 301 case itos: push_i(); break; 302 case ltos: push_l(); break; 303 case ftos: push_f(); break; 304 case dtos: push_d(); break; 305 case vtos: /* nothing to do */ break; 306 default : ShouldNotReachHere(); 307 } 308} 309 310void InterpreterMacroAssembler::pop(TosState state) { 311 switch (state) { 312 case atos: pop_ptr(); break; 313 case btos: 314 case ztos: 315 case ctos: 316 case stos: 317 case itos: pop_i(); break; 318 case ltos: pop_l(); break; 319 case ftos: pop_f(); break; 320 case dtos: pop_d(); break; 321 case vtos: /* nothing to do */ break; 322 default : ShouldNotReachHere(); 323 } 324 verify_oop(R17_tos, state); 325} 326 327void InterpreterMacroAssembler::empty_expression_stack() { 328 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 329} 330 331void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset, 332 Register Rdst, 333 signedOrNot is_signed) { 334#if defined(VM_LITTLE_ENDIAN) 335 if (bcp_offset) { 336 load_const_optimized(Rdst, bcp_offset); 337 lhbrx(Rdst, R14_bcp, Rdst); 338 } else { 339 lhbrx(Rdst, R14_bcp); 340 } 341 if (is_signed == Signed) { 342 extsh(Rdst, Rdst); 343 } 344#else 345 // Read Java big endian format. 346 if (is_signed == Signed) { 347 lha(Rdst, bcp_offset, R14_bcp); 348 } else { 349 lhz(Rdst, bcp_offset, R14_bcp); 350 } 351#endif 352} 353 354void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset, 355 Register Rdst, 356 signedOrNot is_signed) { 357#if defined(VM_LITTLE_ENDIAN) 358 if (bcp_offset) { 359 load_const_optimized(Rdst, bcp_offset); 360 lwbrx(Rdst, R14_bcp, Rdst); 361 } else { 362 lwbrx(Rdst, R14_bcp); 363 } 364 if (is_signed == Signed) { 365 extsw(Rdst, Rdst); 366 } 367#else 368 // Read Java big endian format. 369 if (bcp_offset & 3) { // Offset unaligned? 370 load_const_optimized(Rdst, bcp_offset); 371 if (is_signed == Signed) { 372 lwax(Rdst, R14_bcp, Rdst); 373 } else { 374 lwzx(Rdst, R14_bcp, Rdst); 375 } 376 } else { 377 if (is_signed == Signed) { 378 lwa(Rdst, bcp_offset, R14_bcp); 379 } else { 380 lwz(Rdst, bcp_offset, R14_bcp); 381 } 382 } 383#endif 384} 385 386 387// Load the constant pool cache index from the bytecode stream. 388// 389// Kills / writes: 390// - Rdst, Rscratch 391void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, 392 size_t index_size) { 393 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 394 // Cache index is always in the native format, courtesy of Rewriter. 395 if (index_size == sizeof(u2)) { 396 lhz(Rdst, bcp_offset, R14_bcp); 397 } else if (index_size == sizeof(u4)) { 398 if (bcp_offset & 3) { 399 load_const_optimized(Rdst, bcp_offset); 400 lwax(Rdst, R14_bcp, Rdst); 401 } else { 402 lwa(Rdst, bcp_offset, R14_bcp); 403 } 404 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 405 nand(Rdst, Rdst, Rdst); // convert to plain index 406 } else if (index_size == sizeof(u1)) { 407 lbz(Rdst, bcp_offset, R14_bcp); 408 } else { 409 ShouldNotReachHere(); 410 } 411 // Rdst now contains cp cache index. 412} 413 414void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset, 415 size_t index_size) { 416 get_cache_index_at_bcp(cache, bcp_offset, index_size); 417 sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord)); 418 add(cache, R27_constPoolCache, cache); 419} 420 421// Load 4-byte signed or unsigned integer in Java format (that is, big-endian format) 422// from (Rsrc)+offset. 423void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset, 424 signedOrNot is_signed) { 425#if defined(VM_LITTLE_ENDIAN) 426 if (offset) { 427 load_const_optimized(Rdst, offset); 428 lwbrx(Rdst, Rdst, Rsrc); 429 } else { 430 lwbrx(Rdst, Rsrc); 431 } 432 if (is_signed == Signed) { 433 extsw(Rdst, Rdst); 434 } 435#else 436 if (is_signed == Signed) { 437 lwa(Rdst, offset, Rsrc); 438 } else { 439 lwz(Rdst, offset, Rsrc); 440 } 441#endif 442} 443 444// Load object from cpool->resolved_references(index). 445void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index, Label *is_null) { 446 assert_different_registers(result, index); 447 get_constant_pool(result); 448 449 // Convert from field index to resolved_references() index and from 450 // word index to byte offset. Since this is a java object, it can be compressed. 451 Register tmp = index; // reuse 452 sldi(tmp, index, LogBytesPerHeapOop); 453 // Load pointer for resolved_references[] objArray. 454 ld(result, ConstantPool::resolved_references_offset_in_bytes(), result); 455 // JNIHandles::resolve(result) 456 ld(result, 0, result); 457#ifdef ASSERT 458 Label index_ok; 459 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result); 460 sldi(R0, R0, LogBytesPerHeapOop); 461 cmpd(CCR0, tmp, R0); 462 blt(CCR0, index_ok); 463 stop("resolved reference index out of bounds", 0x09256); 464 bind(index_ok); 465#endif 466 // Add in the index. 467 add(result, tmp, result); 468 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result, is_null); 469} 470 471// Generate a subtype check: branch to ok_is_subtype if sub_klass is 472// a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2. 473void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1, 474 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) { 475 // Profile the not-null value's klass. 476 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2); 477 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype); 478 profile_typecheck_failed(Rtmp1, Rtmp2); 479} 480 481void InterpreterMacroAssembler::generate_stack_overflow_check_with_compare_and_throw(Register Rmem_frame_size, Register Rscratch1) { 482 Label done; 483 BLOCK_COMMENT("stack_overflow_check_with_compare_and_throw {"); 484 sub(Rmem_frame_size, R1_SP, Rmem_frame_size); 485 ld(Rscratch1, thread_(stack_overflow_limit)); 486 cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1); 487 bgt(CCR0/*is_stack_overflow*/, done); 488 489 // Load target address of the runtime stub. 490 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order"); 491 load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0); 492 mtctr(Rscratch1); 493 // Restore caller_sp. 494#ifdef ASSERT 495 ld(Rscratch1, 0, R1_SP); 496 ld(R0, 0, R21_sender_SP); 497 cmpd(CCR0, R0, Rscratch1); 498 asm_assert_eq("backlink", 0x547); 499#endif // ASSERT 500 mr(R1_SP, R21_sender_SP); 501 bctr(); 502 503 align(32, 12); 504 bind(done); 505 BLOCK_COMMENT("} stack_overflow_check_with_compare_and_throw"); 506} 507 508// Separate these two to allow for delay slot in middle. 509// These are used to do a test and full jump to exception-throwing code. 510 511// Check that index is in range for array, then shift index by index_shift, 512// and put arrayOop + shifted_index into res. 513// Note: res is still shy of address by array offset into object. 514 515void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, 516 int index_shift, Register Rtmp, Register Rres) { 517 // Check that index is in range for array, then shift index by index_shift, 518 // and put arrayOop + shifted_index into res. 519 // Note: res is still shy of address by array offset into object. 520 // Kills: 521 // - Rindex 522 // Writes: 523 // - Rres: Address that corresponds to the array index if check was successful. 524 verify_oop(Rarray); 525 const Register Rlength = R0; 526 const Register RsxtIndex = Rtmp; 527 Label LisNull, LnotOOR; 528 529 // Array nullcheck 530 if (!ImplicitNullChecks) { 531 cmpdi(CCR0, Rarray, 0); 532 beq(CCR0, LisNull); 533 } else { 534 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex); 535 } 536 537 // Rindex might contain garbage in upper bits (remember that we don't sign extend 538 // during integer arithmetic operations). So kill them and put value into same register 539 // where ArrayIndexOutOfBounds would expect the index in. 540 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit 541 542 // Index check 543 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray); 544 cmplw(CCR0, Rindex, Rlength); 545 sldi(RsxtIndex, RsxtIndex, index_shift); 546 blt(CCR0, LnotOOR); 547 // Index should be in R17_tos, array should be in R4_ARG2. 548 mr_if_needed(R17_tos, Rindex); 549 mr_if_needed(R4_ARG2, Rarray); 550 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry); 551 mtctr(Rtmp); 552 bctr(); 553 554 if (!ImplicitNullChecks) { 555 bind(LisNull); 556 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry); 557 mtctr(Rtmp); 558 bctr(); 559 } 560 561 align(32, 16); 562 bind(LnotOOR); 563 564 // Calc address 565 add(Rres, RsxtIndex, Rarray); 566} 567 568void InterpreterMacroAssembler::index_check(Register array, Register index, 569 int index_shift, Register tmp, Register res) { 570 // pop array 571 pop_ptr(array); 572 573 // check array 574 index_check_without_pop(array, index, index_shift, tmp, res); 575} 576 577void InterpreterMacroAssembler::get_const(Register Rdst) { 578 ld(Rdst, in_bytes(Method::const_offset()), R19_method); 579} 580 581void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 582 get_const(Rdst); 583 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 584} 585 586void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 587 get_constant_pool(Rdst); 588 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 589} 590 591void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 592 get_constant_pool(Rcpool); 593 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool); 594} 595 596// Unlock if synchronized method. 597// 598// Unlock the receiver if this is a synchronized method. 599// Unlock any Java monitors from synchronized blocks. 600// 601// If there are locked Java monitors 602// If throw_monitor_exception 603// throws IllegalMonitorStateException 604// Else if install_monitor_exception 605// installs IllegalMonitorStateException 606// Else 607// no error processing 608void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 609 bool throw_monitor_exception, 610 bool install_monitor_exception) { 611 Label Lunlocked, Lno_unlock; 612 { 613 Register Rdo_not_unlock_flag = R11_scratch1; 614 Register Raccess_flags = R12_scratch2; 615 616 // Check if synchronized method or unlocking prevented by 617 // JavaThread::do_not_unlock_if_synchronized flag. 618 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 619 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method); 620 li(R0, 0); 621 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag 622 623 push(state); 624 625 // Skip if we don't have to unlock. 626 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0. 627 beq(CCR0, Lunlocked); 628 629 cmpwi(CCR0, Rdo_not_unlock_flag, 0); 630 bne(CCR0, Lno_unlock); 631 } 632 633 // Unlock 634 { 635 Register Rmonitor_base = R11_scratch1; 636 637 Label Lunlock; 638 // If it's still locked, everything is ok, unlock it. 639 ld(Rmonitor_base, 0, R1_SP); 640 addi(Rmonitor_base, Rmonitor_base, 641 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base 642 643 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base); 644 cmpdi(CCR0, R0, 0); 645 bne(CCR0, Lunlock); 646 647 // If it's already unlocked, throw exception. 648 if (throw_monitor_exception) { 649 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 650 should_not_reach_here(); 651 } else { 652 if (install_monitor_exception) { 653 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 654 b(Lunlocked); 655 } 656 } 657 658 bind(Lunlock); 659 unlock_object(Rmonitor_base); 660 } 661 662 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not. 663 bind(Lunlocked); 664 { 665 Label Lexception, Lrestart; 666 Register Rcurrent_obj_addr = R11_scratch1; 667 const int delta = frame::interpreter_frame_monitor_size_in_bytes(); 668 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords"); 669 670 bind(Lrestart); 671 // Set up search loop: Calc num of iterations. 672 { 673 Register Riterations = R12_scratch2; 674 Register Rmonitor_base = Rcurrent_obj_addr; 675 ld(Rmonitor_base, 0, R1_SP); 676 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base 677 678 subf_(Riterations, R26_monitor, Rmonitor_base); 679 ble(CCR0, Lno_unlock); 680 681 addi(Rcurrent_obj_addr, Rmonitor_base, 682 BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes()); 683 // Check if any monitor is on stack, bail out if not 684 srdi(Riterations, Riterations, exact_log2(delta)); 685 mtctr(Riterations); 686 } 687 688 // The search loop: Look for locked monitors. 689 { 690 const Register Rcurrent_obj = R0; 691 Label Lloop; 692 693 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 694 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 695 bind(Lloop); 696 697 // Check if current entry is used. 698 cmpdi(CCR0, Rcurrent_obj, 0); 699 bne(CCR0, Lexception); 700 // Preload next iteration's compare value. 701 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 702 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 703 bdnz(Lloop); 704 } 705 // Fell through: Everything's unlocked => finish. 706 b(Lno_unlock); 707 708 // An object is still locked => need to throw exception. 709 bind(Lexception); 710 if (throw_monitor_exception) { 711 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 712 should_not_reach_here(); 713 } else { 714 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 715 // Unlock does not block, so don't have to worry about the frame. 716 Register Rmonitor_addr = R11_scratch1; 717 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta); 718 unlock_object(Rmonitor_addr); 719 if (install_monitor_exception) { 720 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 721 } 722 b(Lrestart); 723 } 724 } 725 726 align(32, 12); 727 bind(Lno_unlock); 728 pop(state); 729} 730 731// Support function for remove_activation & Co. 732void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, 733 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 757void 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 807void InterpreterMacroAssembler::remove_activation(TosState state, 808 bool throw_monitor_exception, 809 bool install_monitor_exception) { 810 BLOCK_COMMENT("remove_activation {"); 811 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 812 813 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 814 notify_method_exit(false, state, NotifyJVMTI, true); 815 816 BLOCK_COMMENT("reserved_stack_check:"); 817 if (StackReservedPages > 0) { 818 // Test if reserved zone needs to be enabled. 819 Label no_reserved_zone_enabling; 820 821 // Compare frame pointers. There is no good stack pointer, as with stack 822 // frame compression we can get different SPs when we do calls. A subsequent 823 // call could have a smaller SP, so that this compare succeeds for an 824 // inner call of the method annotated with ReservedStack. 825 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread); 826 ld_ptr(R11_scratch1, _abi(callers_sp), R1_SP); // Load frame pointer. 827 cmpld(CCR0, R11_scratch1, R0); 828 blt_predict_taken(CCR0, no_reserved_zone_enabling); 829 830 // Enable reserved zone again, throw stack overflow exception. 831 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread); 832 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError)); 833 834 should_not_reach_here(); 835 836 bind(no_reserved_zone_enabling); 837 } 838 839 verify_oop(R17_tos, state); 840 verify_thread(); 841 842 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 843 mtlr(R0); 844 BLOCK_COMMENT("} remove_activation"); 845} 846 847// Lock object 848// 849// Registers alive 850// monitor - Address of the BasicObjectLock to be used for locking, 851// which must be initialized with the object to lock. 852// object - Address of the object to be locked. 853// 854void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 855 if (UseHeavyMonitors) { 856 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 857 monitor, /*check_for_exceptions=*/true); 858 } else { 859 // template code: 860 // 861 // markOop displaced_header = obj->mark().set_unlocked(); 862 // monitor->lock()->set_displaced_header(displaced_header); 863 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 864 // // We stored the monitor address into the object's mark word. 865 // } else if (THREAD->is_lock_owned((address)displaced_header)) 866 // // Simple recursive case. 867 // monitor->lock()->set_displaced_header(NULL); 868 // } else { 869 // // Slow path. 870 // InterpreterRuntime::monitorenter(THREAD, monitor); 871 // } 872 873 const Register displaced_header = R7_ARG5; 874 const Register object_mark_addr = R8_ARG6; 875 const Register current_header = R9_ARG7; 876 const Register tmp = R10_ARG8; 877 878 Label done; 879 Label cas_failed, slow_case; 880 881 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp); 882 883 // markOop displaced_header = obj->mark().set_unlocked(); 884 885 // Load markOop from object into displaced_header. 886 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object); 887 888 if (UseBiasedLocking) { 889 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case); 890 } 891 892 // Set displaced_header to be (markOop of object | UNLOCK_VALUE). 893 ori(displaced_header, displaced_header, markOopDesc::unlocked_value); 894 895 // monitor->lock()->set_displaced_header(displaced_header); 896 897 // Initialize the box (Must happen before we update the object mark!). 898 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 899 BasicLock::displaced_header_offset_in_bytes(), monitor); 900 901 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 902 903 // Store stack address of the BasicObjectLock (this is monitor) into object. 904 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 905 906 // Must fence, otherwise, preceding store(s) may float below cmpxchg. 907 // CmpxchgX sets CCR0 to cmpX(current, displaced). 908 cmpxchgd(/*flag=*/CCR0, 909 /*current_value=*/current_header, 910 /*compare_value=*/displaced_header, /*exchange_value=*/monitor, 911 /*where=*/object_mark_addr, 912 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq, 913 MacroAssembler::cmpxchgx_hint_acquire_lock(), 914 noreg, 915 &cas_failed, 916 /*check without membar and ldarx first*/true); 917 918 // If the compare-and-exchange succeeded, then we found an unlocked 919 // object and we have now locked it. 920 b(done); 921 bind(cas_failed); 922 923 // } else if (THREAD->is_lock_owned((address)displaced_header)) 924 // // Simple recursive case. 925 // monitor->lock()->set_displaced_header(NULL); 926 927 // We did not see an unlocked object so try the fast recursive case. 928 929 // Check if owner is self by comparing the value in the markOop of object 930 // (current_header) with the stack pointer. 931 sub(current_header, current_header, R1_SP); 932 933 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 934 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place); 935 936 and_(R0/*==0?*/, current_header, tmp); 937 // If condition is true we are done and hence we can store 0 in the displaced 938 // header indicating it is a recursive lock. 939 bne(CCR0, slow_case); 940 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() + 941 BasicLock::displaced_header_offset_in_bytes(), monitor); 942 b(done); 943 944 // } else { 945 // // Slow path. 946 // InterpreterRuntime::monitorenter(THREAD, monitor); 947 948 // None of the above fast optimizations worked so we have to get into the 949 // slow case of monitor enter. 950 bind(slow_case); 951 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 952 monitor, /*check_for_exceptions=*/true); 953 // } 954 align(32, 12); 955 bind(done); 956 } 957} 958 959// Unlocks an object. Used in monitorexit bytecode and remove_activation. 960// 961// Registers alive 962// monitor - Address of the BasicObjectLock to be used for locking, 963// which must be initialized with the object to lock. 964// 965// Throw IllegalMonitorException if object is not locked by current thread. 966void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) { 967 if (UseHeavyMonitors) { 968 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 969 monitor, check_for_exceptions); 970 } else { 971 972 // template code: 973 // 974 // if ((displaced_header = monitor->displaced_header()) == NULL) { 975 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 976 // monitor->set_obj(NULL); 977 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 978 // // We swapped the unlocked mark in displaced_header into the object's mark word. 979 // monitor->set_obj(NULL); 980 // } else { 981 // // Slow path. 982 // InterpreterRuntime::monitorexit(THREAD, monitor); 983 // } 984 985 const Register object = R7_ARG5; 986 const Register displaced_header = R8_ARG6; 987 const Register object_mark_addr = R9_ARG7; 988 const Register current_header = R10_ARG8; 989 990 Label free_slot; 991 Label slow_case; 992 993 assert_different_registers(object, displaced_header, object_mark_addr, current_header); 994 995 if (UseBiasedLocking) { 996 // The object address from the monitor is in object. 997 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); 998 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0"); 999 biased_locking_exit(CCR0, object, displaced_header, free_slot); 1000 } 1001 1002 // Test first if we are in the fast recursive case. 1003 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 1004 BasicLock::displaced_header_offset_in_bytes(), monitor); 1005 1006 // If the displaced header is zero, we have a recursive unlock. 1007 cmpdi(CCR0, displaced_header, 0); 1008 beq(CCR0, free_slot); // recursive unlock 1009 1010 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 1011 // // We swapped the unlocked mark in displaced_header into the object's mark word. 1012 // monitor->set_obj(NULL); 1013 1014 // If we still have a lightweight lock, unlock the object and be done. 1015 1016 // The object address from the monitor is in object. 1017 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); } 1018 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 1019 1020 // We have the displaced header in displaced_header. If the lock is still 1021 // lightweight, it will contain the monitor address and we'll store the 1022 // displaced header back into the object's mark word. 1023 // CmpxchgX sets CCR0 to cmpX(current, monitor). 1024 cmpxchgd(/*flag=*/CCR0, 1025 /*current_value=*/current_header, 1026 /*compare_value=*/monitor, /*exchange_value=*/displaced_header, 1027 /*where=*/object_mark_addr, 1028 MacroAssembler::MemBarRel, 1029 MacroAssembler::cmpxchgx_hint_release_lock(), 1030 noreg, 1031 &slow_case); 1032 b(free_slot); 1033 1034 // } else { 1035 // // Slow path. 1036 // InterpreterRuntime::monitorexit(THREAD, monitor); 1037 1038 // The lock has been converted into a heavy lock and hence 1039 // we need to get into the slow case. 1040 bind(slow_case); 1041 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1042 monitor, check_for_exceptions); 1043 // } 1044 1045 Label done; 1046 b(done); // Monitor register may be overwritten! Runtime has already freed the slot. 1047 1048 // Exchange worked, do monitor->set_obj(NULL); 1049 align(32, 12); 1050 bind(free_slot); 1051 li(R0, 0); 1052 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor); 1053 bind(done); 1054 } 1055} 1056 1057// Load compiled (i2c) or interpreter entry when calling from interpreted and 1058// do the call. Centralized so that all interpreter calls will do the same actions. 1059// If jvmti single stepping is on for a thread we must not call compiled code. 1060// 1061// Input: 1062// - Rtarget_method: method to call 1063// - Rret_addr: return address 1064// - 2 scratch regs 1065// 1066void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, 1067 Register Rscratch1, Register Rscratch2) { 1068 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr); 1069 // Assume we want to go compiled if available. 1070 const Register Rtarget_addr = Rscratch1; 1071 const Register Rinterp_only = Rscratch2; 1072 1073 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method); 1074 1075 if (JvmtiExport::can_post_interpreter_events()) { 1076 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 1077 1078 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 1079 // compiled code in threads for which the event is enabled. Check here for 1080 // interp_only_mode if these events CAN be enabled. 1081 Label done; 1082 verify_thread(); 1083 cmpwi(CCR0, Rinterp_only, 0); 1084 beq(CCR0, done); 1085 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method); 1086 align(32, 12); 1087 bind(done); 1088 } 1089 1090#ifdef ASSERT 1091 { 1092 Label Lok; 1093 cmpdi(CCR0, Rtarget_addr, 0); 1094 bne(CCR0, Lok); 1095 stop("null entry point"); 1096 bind(Lok); 1097 } 1098#endif // ASSERT 1099 1100 mr(R21_sender_SP, R1_SP); 1101 1102 // Calc a precise SP for the call. The SP value we calculated in 1103 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space 1104 // if esp is not max. Also, the i2c adapter extends the stack space without restoring 1105 // our pre-calced value, so repeating calls via i2c would result in stack overflow. 1106 // Since esp already points to an empty slot, we just have to sub 1 additional slot 1107 // to meet the abi scratch requirements. 1108 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in 1109 // the return entry of the interpreter. 1110 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size); 1111 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address 1112 resize_frame_absolute(Rscratch2, Rscratch2, R0); 1113 1114 mr_if_needed(R19_method, Rtarget_method); 1115 mtctr(Rtarget_addr); 1116 mtlr(Rret_addr); 1117 1118 save_interpreter_state(Rscratch2); 1119#ifdef ASSERT 1120 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp 1121 cmpd(CCR0, R21_sender_SP, Rscratch1); 1122 asm_assert_eq("top_frame_sp incorrect", 0x951); 1123#endif 1124 1125 bctr(); 1126} 1127 1128// Set the method data pointer for the current bcp. 1129void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1130 assert(ProfileInterpreter, "must be profiling interpreter"); 1131 Label get_continue; 1132 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method); 1133 test_method_data_pointer(get_continue); 1134 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp); 1135 1136 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1137 add(R28_mdx, R28_mdx, R3_RET); 1138 bind(get_continue); 1139} 1140 1141// Test ImethodDataPtr. If it is null, continue at the specified label. 1142void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1143 assert(ProfileInterpreter, "must be profiling interpreter"); 1144 cmpdi(CCR0, R28_mdx, 0); 1145 beq(CCR0, zero_continue); 1146} 1147 1148void InterpreterMacroAssembler::verify_method_data_pointer() { 1149 assert(ProfileInterpreter, "must be profiling interpreter"); 1150#ifdef ASSERT 1151 Label verify_continue; 1152 test_method_data_pointer(verify_continue); 1153 1154 // If the mdp is valid, it will point to a DataLayout header which is 1155 // consistent with the bcp. The converse is highly probable also. 1156 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx); 1157 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method); 1158 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset())); 1159 add(R11_scratch1, R12_scratch2, R12_scratch2); 1160 cmpd(CCR0, R11_scratch1, R14_bcp); 1161 beq(CCR0, verify_continue); 1162 1163 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx); 1164 1165 bind(verify_continue); 1166#endif 1167} 1168 1169void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1170 Register method_counters, 1171 Register Rscratch, 1172 Label &profile_continue) { 1173 assert(ProfileInterpreter, "must be profiling interpreter"); 1174 // Control will flow to "profile_continue" if the counter is less than the 1175 // limit or if we call profile_method(). 1176 Label done; 1177 1178 // If no method data exists, and the counter is high enough, make one. 1179 lwz(Rscratch, in_bytes(MethodCounters::interpreter_profile_limit_offset()), method_counters); 1180 1181 cmpdi(CCR0, R28_mdx, 0); 1182 // Test to see if we should create a method data oop. 1183 cmpd(CCR1, Rscratch, invocation_count); 1184 bne(CCR0, done); 1185 bge(CCR1, profile_continue); 1186 1187 // Build it now. 1188 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1189 set_method_data_pointer_for_bcp(); 1190 b(profile_continue); 1191 1192 align(32, 12); 1193 bind(done); 1194} 1195 1196void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register method_counters, 1197 Register target_bcp, Register disp, Register Rtmp) { 1198 assert_different_registers(backedge_count, target_bcp, disp, Rtmp, R4_ARG2); 1199 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 1200 1201 Label did_not_overflow; 1202 Label overflow_with_error; 1203 1204 lwz(Rtmp, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()), method_counters); 1205 cmpw(CCR0, backedge_count, Rtmp); 1206 1207 blt(CCR0, did_not_overflow); 1208 1209 // When ProfileInterpreter is on, the backedge_count comes from the 1210 // methodDataOop, which value does not get reset on the call to 1211 // frequency_counter_overflow(). To avoid excessive calls to the overflow 1212 // routine while the method is being compiled, add a second test to make sure 1213 // the overflow function is called only once every overflow_frequency. 1214 if (ProfileInterpreter) { 1215 const int overflow_frequency = 1024; 1216 andi_(Rtmp, backedge_count, overflow_frequency-1); 1217 bne(CCR0, did_not_overflow); 1218 } 1219 1220 // Overflow in loop, pass branch bytecode. 1221 subf(R4_ARG2, disp, target_bcp); // Compute branch bytecode (previous bcp). 1222 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); 1223 1224 // Was an OSR adapter generated? 1225 cmpdi(CCR0, R3_RET, 0); 1226 beq(CCR0, overflow_with_error); 1227 1228 // Has the nmethod been invalidated already? 1229 lbz(Rtmp, nmethod::state_offset(), R3_RET); 1230 cmpwi(CCR0, Rtmp, nmethod::in_use); 1231 bne(CCR0, overflow_with_error); 1232 1233 // Migrate the interpreter frame off of the stack. 1234 // We can use all registers because we will not return to interpreter from this point. 1235 1236 // Save nmethod. 1237 const Register osr_nmethod = R31; 1238 mr(osr_nmethod, R3_RET); 1239 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1); 1240 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread); 1241 reset_last_Java_frame(); 1242 // OSR buffer is in ARG1 1243 1244 // Remove the interpreter frame. 1245 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1246 1247 // Jump to the osr code. 1248 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod); 1249 mtlr(R0); 1250 mtctr(R11_scratch1); 1251 bctr(); 1252 1253 align(32, 12); 1254 bind(overflow_with_error); 1255 bind(did_not_overflow); 1256} 1257 1258// Store a value at some constant offset from the method data pointer. 1259void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1260 assert(ProfileInterpreter, "must be profiling interpreter"); 1261 1262 std(value, constant, R28_mdx); 1263} 1264 1265// Increment the value at some constant offset from the method data pointer. 1266void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1267 Register counter_addr, 1268 Register Rbumped_count, 1269 bool decrement) { 1270 // Locate the counter at a fixed offset from the mdp: 1271 addi(counter_addr, R28_mdx, constant); 1272 increment_mdp_data_at(counter_addr, Rbumped_count, decrement); 1273} 1274 1275// Increment the value at some non-fixed (reg + constant) offset from 1276// the method data pointer. 1277void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1278 int constant, 1279 Register scratch, 1280 Register Rbumped_count, 1281 bool decrement) { 1282 // Add the constant to reg to get the offset. 1283 add(scratch, R28_mdx, reg); 1284 // Then calculate the counter address. 1285 addi(scratch, scratch, constant); 1286 increment_mdp_data_at(scratch, Rbumped_count, decrement); 1287} 1288 1289void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr, 1290 Register Rbumped_count, 1291 bool decrement) { 1292 assert(ProfileInterpreter, "must be profiling interpreter"); 1293 1294 // Load the counter. 1295 ld(Rbumped_count, 0, counter_addr); 1296 1297 if (decrement) { 1298 // Decrement the register. Set condition codes. 1299 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment); 1300 // Store the decremented counter, if it is still negative. 1301 std(Rbumped_count, 0, counter_addr); 1302 // Note: add/sub overflow check are not ported, since 64 bit 1303 // calculation should never overflow. 1304 } else { 1305 // Increment the register. Set carry flag. 1306 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment); 1307 // Store the incremented counter. 1308 std(Rbumped_count, 0, counter_addr); 1309 } 1310} 1311 1312// Set a flag value at the current method data pointer position. 1313void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1314 Register scratch) { 1315 assert(ProfileInterpreter, "must be profiling interpreter"); 1316 // Load the data header. 1317 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1318 // Set the flag. 1319 ori(scratch, scratch, flag_constant); 1320 // Store the modified header. 1321 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1322} 1323 1324// Test the location at some offset from the method data pointer. 1325// If it is not equal to value, branch to the not_equal_continue Label. 1326void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1327 Register value, 1328 Label& not_equal_continue, 1329 Register test_out) { 1330 assert(ProfileInterpreter, "must be profiling interpreter"); 1331 1332 ld(test_out, offset, R28_mdx); 1333 cmpd(CCR0, value, test_out); 1334 bne(CCR0, not_equal_continue); 1335} 1336 1337// Update the method data pointer by the displacement located at some fixed 1338// offset from the method data pointer. 1339void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1340 Register scratch) { 1341 assert(ProfileInterpreter, "must be profiling interpreter"); 1342 1343 ld(scratch, offset_of_disp, R28_mdx); 1344 add(R28_mdx, scratch, R28_mdx); 1345} 1346 1347// Update the method data pointer by the displacement located at the 1348// offset (reg + offset_of_disp). 1349void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1350 int offset_of_disp, 1351 Register scratch) { 1352 assert(ProfileInterpreter, "must be profiling interpreter"); 1353 1354 add(scratch, reg, R28_mdx); 1355 ld(scratch, offset_of_disp, scratch); 1356 add(R28_mdx, scratch, R28_mdx); 1357} 1358 1359// Update the method data pointer by a simple constant displacement. 1360void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1361 assert(ProfileInterpreter, "must be profiling interpreter"); 1362 addi(R28_mdx, R28_mdx, constant); 1363} 1364 1365// Update the method data pointer for a _ret bytecode whose target 1366// was not among our cached targets. 1367void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1368 Register return_bci) { 1369 assert(ProfileInterpreter, "must be profiling interpreter"); 1370 1371 push(state); 1372 assert(return_bci->is_nonvolatile(), "need to protect return_bci"); 1373 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1374 pop(state); 1375} 1376 1377// Increments the backedge counter. 1378// Returns backedge counter + invocation counter in Rdst. 1379void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst, 1380 const Register Rtmp1, Register Rscratch) { 1381 assert(UseCompiler, "incrementing must be useful"); 1382 assert_different_registers(Rdst, Rtmp1); 1383 const Register invocation_counter = Rtmp1; 1384 const Register counter = Rdst; 1385 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size."); 1386 1387 // Load backedge counter. 1388 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1389 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1390 // Load invocation counter. 1391 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) + 1392 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1393 1394 // Add the delta to the backedge counter. 1395 addi(counter, counter, InvocationCounter::count_increment); 1396 1397 // Mask the invocation counter. 1398 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value); 1399 1400 // Store new counter value. 1401 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1402 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1403 // Return invocation counter + backedge counter. 1404 add(counter, counter, invocation_counter); 1405} 1406 1407// Count a taken branch in the bytecodes. 1408void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1409 if (ProfileInterpreter) { 1410 Label profile_continue; 1411 1412 // If no method data exists, go to profile_continue. 1413 test_method_data_pointer(profile_continue); 1414 1415 // We are taking a branch. Increment the taken count. 1416 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count); 1417 1418 // The method data pointer needs to be updated to reflect the new target. 1419 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1420 bind (profile_continue); 1421 } 1422} 1423 1424// Count a not-taken branch in the bytecodes. 1425void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) { 1426 if (ProfileInterpreter) { 1427 Label profile_continue; 1428 1429 // If no method data exists, go to profile_continue. 1430 test_method_data_pointer(profile_continue); 1431 1432 // We are taking a branch. Increment the not taken count. 1433 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2); 1434 1435 // The method data pointer needs to be updated to correspond to the 1436 // next bytecode. 1437 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1438 bind (profile_continue); 1439 } 1440} 1441 1442// Count a non-virtual call in the bytecodes. 1443void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) { 1444 if (ProfileInterpreter) { 1445 Label profile_continue; 1446 1447 // If no method data exists, go to profile_continue. 1448 test_method_data_pointer(profile_continue); 1449 1450 // We are making a call. Increment the count. 1451 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1452 1453 // The method data pointer needs to be updated to reflect the new target. 1454 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1455 bind (profile_continue); 1456 } 1457} 1458 1459// Count a final call in the bytecodes. 1460void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) { 1461 if (ProfileInterpreter) { 1462 Label profile_continue; 1463 1464 // If no method data exists, go to profile_continue. 1465 test_method_data_pointer(profile_continue); 1466 1467 // We are making a call. Increment the count. 1468 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1469 1470 // The method data pointer needs to be updated to reflect the new target. 1471 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1472 bind (profile_continue); 1473 } 1474} 1475 1476// Count a virtual call in the bytecodes. 1477void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver, 1478 Register Rscratch1, 1479 Register Rscratch2, 1480 bool receiver_can_be_null) { 1481 if (!ProfileInterpreter) { return; } 1482 Label profile_continue; 1483 1484 // If no method data exists, go to profile_continue. 1485 test_method_data_pointer(profile_continue); 1486 1487 Label skip_receiver_profile; 1488 if (receiver_can_be_null) { 1489 Label not_null; 1490 cmpdi(CCR0, Rreceiver, 0); 1491 bne(CCR0, not_null); 1492 // We are making a call. Increment the count for null receiver. 1493 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2); 1494 b(skip_receiver_profile); 1495 bind(not_null); 1496 } 1497 1498 // Record the receiver type. 1499 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true); 1500 bind(skip_receiver_profile); 1501 1502 // The method data pointer needs to be updated to reflect the new target. 1503 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1504 bind (profile_continue); 1505} 1506 1507void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) { 1508 if (ProfileInterpreter) { 1509 Label profile_continue; 1510 1511 // If no method data exists, go to profile_continue. 1512 test_method_data_pointer(profile_continue); 1513 1514 int mdp_delta = in_bytes(BitData::bit_data_size()); 1515 if (TypeProfileCasts) { 1516 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1517 1518 // Record the object type. 1519 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false); 1520 } 1521 1522 // The method data pointer needs to be updated. 1523 update_mdp_by_constant(mdp_delta); 1524 1525 bind (profile_continue); 1526 } 1527} 1528 1529void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) { 1530 if (ProfileInterpreter && TypeProfileCasts) { 1531 Label profile_continue; 1532 1533 // If no method data exists, go to profile_continue. 1534 test_method_data_pointer(profile_continue); 1535 1536 int count_offset = in_bytes(CounterData::count_offset()); 1537 // Back up the address, since we have already bumped the mdp. 1538 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1539 1540 // *Decrement* the counter. We expect to see zero or small negatives. 1541 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true); 1542 1543 bind (profile_continue); 1544 } 1545} 1546 1547// Count a ret in the bytecodes. 1548void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, 1549 Register scratch1, Register scratch2) { 1550 if (ProfileInterpreter) { 1551 Label profile_continue; 1552 uint row; 1553 1554 // If no method data exists, go to profile_continue. 1555 test_method_data_pointer(profile_continue); 1556 1557 // Update the total ret count. 1558 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 ); 1559 1560 for (row = 0; row < RetData::row_limit(); row++) { 1561 Label next_test; 1562 1563 // See if return_bci is equal to bci[n]: 1564 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1); 1565 1566 // return_bci is equal to bci[n]. Increment the count. 1567 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2); 1568 1569 // The method data pointer needs to be updated to reflect the new target. 1570 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1); 1571 b(profile_continue); 1572 bind(next_test); 1573 } 1574 1575 update_mdp_for_ret(state, return_bci); 1576 1577 bind (profile_continue); 1578 } 1579} 1580 1581// Count the default case of a switch construct. 1582void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) { 1583 if (ProfileInterpreter) { 1584 Label profile_continue; 1585 1586 // If no method data exists, go to profile_continue. 1587 test_method_data_pointer(profile_continue); 1588 1589 // Update the default case count 1590 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1591 scratch1, scratch2); 1592 1593 // The method data pointer needs to be updated. 1594 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()), 1595 scratch1); 1596 1597 bind (profile_continue); 1598 } 1599} 1600 1601// Count the index'th case of a switch construct. 1602void InterpreterMacroAssembler::profile_switch_case(Register index, 1603 Register scratch1, 1604 Register scratch2, 1605 Register scratch3) { 1606 if (ProfileInterpreter) { 1607 assert_different_registers(index, scratch1, scratch2, scratch3); 1608 Label profile_continue; 1609 1610 // If no method data exists, go to profile_continue. 1611 test_method_data_pointer(profile_continue); 1612 1613 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes(). 1614 li(scratch3, in_bytes(MultiBranchData::case_array_offset())); 1615 1616 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works"); 1617 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1618 add(scratch1, scratch1, scratch3); 1619 1620 // Update the case count. 1621 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3); 1622 1623 // The method data pointer needs to be updated. 1624 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2); 1625 1626 bind (profile_continue); 1627 } 1628} 1629 1630void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) { 1631 if (ProfileInterpreter) { 1632 assert_different_registers(Rscratch1, Rscratch2); 1633 Label profile_continue; 1634 1635 // If no method data exists, go to profile_continue. 1636 test_method_data_pointer(profile_continue); 1637 1638 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1); 1639 1640 // The method data pointer needs to be updated. 1641 int mdp_delta = in_bytes(BitData::bit_data_size()); 1642 if (TypeProfileCasts) { 1643 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1644 } 1645 update_mdp_by_constant(mdp_delta); 1646 1647 bind (profile_continue); 1648 } 1649} 1650 1651void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver, 1652 Register Rscratch1, Register Rscratch2, 1653 bool is_virtual_call) { 1654 assert(ProfileInterpreter, "must be profiling"); 1655 assert_different_registers(Rreceiver, Rscratch1, Rscratch2); 1656 1657 Label done; 1658 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call); 1659 bind (done); 1660} 1661 1662void InterpreterMacroAssembler::record_klass_in_profile_helper( 1663 Register receiver, Register scratch1, Register scratch2, 1664 int start_row, Label& done, bool is_virtual_call) { 1665 if (TypeProfileWidth == 0) { 1666 if (is_virtual_call) { 1667 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1668 } 1669 return; 1670 } 1671 1672 int last_row = VirtualCallData::row_limit() - 1; 1673 assert(start_row <= last_row, "must be work left to do"); 1674 // Test this row for both the receiver and for null. 1675 // Take any of three different outcomes: 1676 // 1. found receiver => increment count and goto done 1677 // 2. found null => keep looking for case 1, maybe allocate this cell 1678 // 3. found something else => keep looking for cases 1 and 2 1679 // Case 3 is handled by a recursive call. 1680 for (int row = start_row; row <= last_row; row++) { 1681 Label next_test; 1682 bool test_for_null_also = (row == start_row); 1683 1684 // See if the receiver is receiver[n]. 1685 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1686 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1); 1687 // delayed()->tst(scratch); 1688 1689 // The receiver is receiver[n]. Increment count[n]. 1690 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1691 increment_mdp_data_at(count_offset, scratch1, scratch2); 1692 b(done); 1693 bind(next_test); 1694 1695 if (test_for_null_also) { 1696 Label found_null; 1697 // Failed the equality check on receiver[n]... Test for null. 1698 if (start_row == last_row) { 1699 // The only thing left to do is handle the null case. 1700 if (is_virtual_call) { 1701 // Scratch1 contains test_out from test_mdp_data_at. 1702 cmpdi(CCR0, scratch1, 0); 1703 beq(CCR0, found_null); 1704 // Receiver did not match any saved receiver and there is no empty row for it. 1705 // Increment total counter to indicate polymorphic case. 1706 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1707 b(done); 1708 bind(found_null); 1709 } else { 1710 cmpdi(CCR0, scratch1, 0); 1711 bne(CCR0, done); 1712 } 1713 break; 1714 } 1715 // Since null is rare, make it be the branch-taken case. 1716 cmpdi(CCR0, scratch1, 0); 1717 beq(CCR0, found_null); 1718 1719 // Put all the "Case 3" tests here. 1720 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call); 1721 1722 // Found a null. Keep searching for a matching receiver, 1723 // but remember that this is an empty (unused) slot. 1724 bind(found_null); 1725 } 1726 } 1727 1728 // In the fall-through case, we found no matching receiver, but we 1729 // observed the receiver[start_row] is NULL. 1730 1731 // Fill in the receiver field and increment the count. 1732 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1733 set_mdp_data_at(recvr_offset, receiver); 1734 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1735 li(scratch1, DataLayout::counter_increment); 1736 set_mdp_data_at(count_offset, scratch1); 1737 if (start_row > 0) { 1738 b(done); 1739 } 1740} 1741 1742// Argument and return type profilig. 1743// kills: tmp, tmp2, R0, CR0, CR1 1744void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base, 1745 RegisterOrConstant mdo_addr_offs, 1746 Register tmp, Register tmp2) { 1747 Label do_nothing, do_update; 1748 1749 // tmp2 = obj is allowed 1750 assert_different_registers(obj, mdo_addr_base, tmp, R0); 1751 assert_different_registers(tmp2, mdo_addr_base, tmp, R0); 1752 const Register klass = tmp2; 1753 1754 verify_oop(obj); 1755 1756 ld(tmp, mdo_addr_offs, mdo_addr_base); 1757 1758 // Set null_seen if obj is 0. 1759 cmpdi(CCR0, obj, 0); 1760 ori(R0, tmp, TypeEntries::null_seen); 1761 beq(CCR0, do_update); 1762 1763 load_klass(klass, obj); 1764 1765 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask)); 1766 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask); 1767 cmpd(CCR1, R0, klass); 1768 // Klass seen before, nothing to do (regardless of unknown bit). 1769 //beq(CCR1, do_nothing); 1770 1771 andi_(R0, klass, TypeEntries::type_unknown); 1772 // Already unknown. Nothing to do anymore. 1773 //bne(CCR0, do_nothing); 1774 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne 1775 beq(CCR0, do_nothing); 1776 1777 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask)); 1778 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1779 beq(CCR0, do_update); // First time here. Set profile type. 1780 1781 // Different than before. Cannot keep accurate profile. 1782 ori(R0, tmp, TypeEntries::type_unknown); 1783 1784 bind(do_update); 1785 // update profile 1786 std(R0, mdo_addr_offs, mdo_addr_base); 1787 1788 align(32, 12); 1789 bind(do_nothing); 1790} 1791 1792void InterpreterMacroAssembler::profile_arguments_type(Register callee, 1793 Register tmp1, Register tmp2, 1794 bool is_virtual) { 1795 if (!ProfileInterpreter) { 1796 return; 1797 } 1798 1799 assert_different_registers(callee, tmp1, tmp2, R28_mdx); 1800 1801 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1802 Label profile_continue; 1803 1804 test_method_data_pointer(profile_continue); 1805 1806 int off_to_start = is_virtual ? 1807 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1808 1809 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx); 1810 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1811 bne(CCR0, profile_continue); 1812 1813 if (MethodData::profile_arguments()) { 1814 Label done; 1815 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1816 add(R28_mdx, off_to_args, R28_mdx); 1817 1818 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1819 if (i > 0 || MethodData::profile_return()) { 1820 // If return value type is profiled we may have no argument to profile. 1821 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1822 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count()); 1823 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count()); 1824 blt(CCR0, done); 1825 } 1826 ld(tmp1, in_bytes(Method::const_offset()), callee); 1827 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1); 1828 // Stack offset o (zero based) from the start of the argument 1829 // list, for n arguments translates into offset n - o - 1 from 1830 // the end of the argument list. But there's an extra slot at 1831 // the top of the stack. So the offset is n - o from Lesp. 1832 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx); 1833 subf(tmp1, tmp2, tmp1); 1834 1835 sldi(tmp1, tmp1, Interpreter::logStackElementSize); 1836 ldx(tmp1, tmp1, R15_esp); 1837 1838 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1); 1839 1840 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1841 addi(R28_mdx, R28_mdx, to_add); 1842 off_to_args += to_add; 1843 } 1844 1845 if (MethodData::profile_return()) { 1846 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1847 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1848 } 1849 1850 bind(done); 1851 1852 if (MethodData::profile_return()) { 1853 // We're right after the type profile for the last 1854 // argument. tmp1 is the number of cells left in the 1855 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1856 // if there's a return to profile. 1857 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), 1858 "can't move past ret type"); 1859 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size)); 1860 add(R28_mdx, tmp1, R28_mdx); 1861 } 1862 } else { 1863 assert(MethodData::profile_return(), "either profile call args or call ret"); 1864 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 1865 } 1866 1867 // Mdp points right after the end of the 1868 // CallTypeData/VirtualCallTypeData, right after the cells for the 1869 // return value type if there's one. 1870 align(32, 12); 1871 bind(profile_continue); 1872 } 1873} 1874 1875void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 1876 assert_different_registers(ret, tmp1, tmp2); 1877 if (ProfileInterpreter && MethodData::profile_return()) { 1878 Label profile_continue; 1879 1880 test_method_data_pointer(profile_continue); 1881 1882 if (MethodData::profile_return_jsr292_only()) { 1883 // If we don't profile all invoke bytecodes we must make sure 1884 // it's a bytecode we indeed profile. We can't go back to the 1885 // begining of the ProfileData we intend to update to check its 1886 // type because we're right after it and we don't known its 1887 // length. 1888 lbz(tmp1, 0, R14_bcp); 1889 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method); 1890 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic); 1891 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle); 1892 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1893 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm); 1894 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1895 bne(CCR0, profile_continue); 1896 } 1897 1898 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2); 1899 1900 align(32, 12); 1901 bind(profile_continue); 1902 } 1903} 1904 1905void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, 1906 Register tmp3, Register tmp4) { 1907 if (ProfileInterpreter && MethodData::profile_parameters()) { 1908 Label profile_continue, done; 1909 1910 test_method_data_pointer(profile_continue); 1911 1912 // Load the offset of the area within the MDO used for 1913 // parameters. If it's negative we're not profiling any parameters. 1914 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx); 1915 cmpwi(CCR0, tmp1, 0); 1916 blt(CCR0, profile_continue); 1917 1918 // Compute a pointer to the area for parameters from the offset 1919 // and move the pointer to the slot for the last 1920 // parameters. Collect profiling from last parameter down. 1921 // mdo start + parameters offset + array length - 1 1922 1923 // Pointer to the parameter area in the MDO. 1924 const Register mdp = tmp1; 1925 add(mdp, tmp1, R28_mdx); 1926 1927 // Offset of the current profile entry to update. 1928 const Register entry_offset = tmp2; 1929 // entry_offset = array len in number of cells 1930 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp); 1931 1932 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1933 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 1934 1935 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 1936 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size)); 1937 // entry_offset in bytes 1938 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1939 1940 Label loop; 1941 align(32, 12); 1942 bind(loop); 1943 1944 // Load offset on the stack from the slot for this parameter. 1945 ld(tmp3, entry_offset, mdp); 1946 sldi(tmp3, tmp3, Interpreter::logStackElementSize); 1947 neg(tmp3, tmp3); 1948 // Read the parameter from the local area. 1949 ldx(tmp3, tmp3, R18_locals); 1950 1951 // Make entry_offset now point to the type field for this parameter. 1952 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1953 assert(type_base > off_base, "unexpected"); 1954 addi(entry_offset, entry_offset, type_base - off_base); 1955 1956 // Profile the parameter. 1957 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3); 1958 1959 // Go to next parameter. 1960 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base); 1961 cmpdi(CCR0, entry_offset, off_base + delta); 1962 addi(entry_offset, entry_offset, -delta); 1963 bge(CCR0, loop); 1964 1965 align(32, 12); 1966 bind(profile_continue); 1967 } 1968} 1969 1970// Add a InterpMonitorElem to stack (see frame_sparc.hpp). 1971void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) { 1972 1973 // Very-local scratch registers. 1974 const Register esp = Rtemp1; 1975 const Register slot = Rtemp2; 1976 1977 // Extracted monitor_size. 1978 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes(); 1979 assert(Assembler::is_aligned((unsigned int)monitor_size, 1980 (unsigned int)frame::alignment_in_bytes), 1981 "size of a monitor must respect alignment of SP"); 1982 1983 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor 1984 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp 1985 1986 // Shuffle expression stack down. Recall that stack_base points 1987 // just above the new expression stack bottom. Old_tos and new_tos 1988 // are used to scan thru the old and new expression stacks. 1989 if (!stack_is_empty) { 1990 Label copy_slot, copy_slot_finished; 1991 const Register n_slots = slot; 1992 1993 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack). 1994 subf(n_slots, esp, R26_monitor); 1995 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy. 1996 assert(LogBytesPerWord == 3, "conflicts assembler instructions"); 1997 beq(CCR0, copy_slot_finished); // Nothing to copy. 1998 1999 mtctr(n_slots); 2000 2001 // loop 2002 bind(copy_slot); 2003 ld(slot, 0, esp); // Move expression stack down. 2004 std(slot, -monitor_size, esp); // distance = monitor_size 2005 addi(esp, esp, BytesPerWord); 2006 bdnz(copy_slot); 2007 2008 bind(copy_slot_finished); 2009 } 2010 2011 addi(R15_esp, R15_esp, -monitor_size); 2012 addi(R26_monitor, R26_monitor, -monitor_size); 2013 2014 // Restart interpreter 2015} 2016 2017// ============================================================================ 2018// Java locals access 2019 2020// Load a local variable at index in Rindex into register Rdst_value. 2021// Also puts address of local into Rdst_address as a service. 2022// Kills: 2023// - Rdst_value 2024// - Rdst_address 2025void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) { 2026 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2027 subf(Rdst_address, Rdst_address, R18_locals); 2028 lwz(Rdst_value, 0, Rdst_address); 2029} 2030 2031// Load a local variable at index in Rindex into register Rdst_value. 2032// Also puts address of local into Rdst_address as a service. 2033// Kills: 2034// - Rdst_value 2035// - Rdst_address 2036void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) { 2037 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2038 subf(Rdst_address, Rdst_address, R18_locals); 2039 ld(Rdst_value, -8, Rdst_address); 2040} 2041 2042// Load a local variable at index in Rindex into register Rdst_value. 2043// Also puts address of local into Rdst_address as a service. 2044// Input: 2045// - Rindex: slot nr of local variable 2046// Kills: 2047// - Rdst_value 2048// - Rdst_address 2049void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, 2050 Register Rdst_address, 2051 Register Rindex) { 2052 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2053 subf(Rdst_address, Rdst_address, R18_locals); 2054 ld(Rdst_value, 0, Rdst_address); 2055} 2056 2057// Load a local variable at index in Rindex into register Rdst_value. 2058// Also puts address of local into Rdst_address as a service. 2059// Kills: 2060// - Rdst_value 2061// - Rdst_address 2062void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, 2063 Register Rdst_address, 2064 Register Rindex) { 2065 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2066 subf(Rdst_address, Rdst_address, R18_locals); 2067 lfs(Rdst_value, 0, Rdst_address); 2068} 2069 2070// Load a local variable at index in Rindex into register Rdst_value. 2071// Also puts address of local into Rdst_address as a service. 2072// Kills: 2073// - Rdst_value 2074// - Rdst_address 2075void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, 2076 Register Rdst_address, 2077 Register Rindex) { 2078 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2079 subf(Rdst_address, Rdst_address, R18_locals); 2080 lfd(Rdst_value, -8, Rdst_address); 2081} 2082 2083// Store an int value at local variable slot Rindex. 2084// Kills: 2085// - Rindex 2086void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) { 2087 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2088 subf(Rindex, Rindex, R18_locals); 2089 stw(Rvalue, 0, Rindex); 2090} 2091 2092// Store a long value at local variable slot Rindex. 2093// Kills: 2094// - Rindex 2095void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) { 2096 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2097 subf(Rindex, Rindex, R18_locals); 2098 std(Rvalue, -8, Rindex); 2099} 2100 2101// Store an oop value at local variable slot Rindex. 2102// Kills: 2103// - Rindex 2104void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) { 2105 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2106 subf(Rindex, Rindex, R18_locals); 2107 std(Rvalue, 0, Rindex); 2108} 2109 2110// Store an int value at local variable slot Rindex. 2111// Kills: 2112// - Rindex 2113void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) { 2114 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2115 subf(Rindex, Rindex, R18_locals); 2116 stfs(Rvalue, 0, Rindex); 2117} 2118 2119// Store an int value at local variable slot Rindex. 2120// Kills: 2121// - Rindex 2122void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) { 2123 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2124 subf(Rindex, Rindex, R18_locals); 2125 stfd(Rvalue, -8, Rindex); 2126} 2127 2128// Read pending exception from thread and jump to interpreter. 2129// Throw exception entry if one if pending. Fall through otherwise. 2130void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) { 2131 assert_different_registers(Rscratch1, Rscratch2, R3); 2132 Register Rexception = Rscratch1; 2133 Register Rtmp = Rscratch2; 2134 Label Ldone; 2135 // Get pending exception oop. 2136 ld(Rexception, thread_(pending_exception)); 2137 cmpdi(CCR0, Rexception, 0); 2138 beq(CCR0, Ldone); 2139 li(Rtmp, 0); 2140 mr_if_needed(R3, Rexception); 2141 std(Rtmp, thread_(pending_exception)); // Clear exception in thread 2142 if (Interpreter::rethrow_exception_entry() != NULL) { 2143 // Already got entry address. 2144 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry()); 2145 } else { 2146 // Dynamically load entry address. 2147 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true); 2148 ld(Rtmp, simm16_rest, Rtmp); 2149 } 2150 mtctr(Rtmp); 2151 save_interpreter_state(Rtmp); 2152 bctr(); 2153 2154 align(32, 12); 2155 bind(Ldone); 2156} 2157 2158void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) { 2159 save_interpreter_state(R11_scratch1); 2160 2161 MacroAssembler::call_VM(oop_result, entry_point, false); 2162 2163 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2164 2165 check_and_handle_popframe(R11_scratch1); 2166 check_and_handle_earlyret(R11_scratch1); 2167 // Now check exceptions manually. 2168 if (check_exceptions) { 2169 check_and_forward_exception(R11_scratch1, R12_scratch2); 2170 } 2171} 2172 2173void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2174 Register arg_1, bool check_exceptions) { 2175 // ARG1 is reserved for the thread. 2176 mr_if_needed(R4_ARG2, arg_1); 2177 call_VM(oop_result, entry_point, check_exceptions); 2178} 2179 2180void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2181 Register arg_1, Register arg_2, 2182 bool check_exceptions) { 2183 // ARG1 is reserved for the thread. 2184 mr_if_needed(R4_ARG2, arg_1); 2185 assert(arg_2 != R4_ARG2, "smashed argument"); 2186 mr_if_needed(R5_ARG3, arg_2); 2187 call_VM(oop_result, entry_point, check_exceptions); 2188} 2189 2190void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2191 Register arg_1, Register arg_2, Register arg_3, 2192 bool check_exceptions) { 2193 // ARG1 is reserved for the thread. 2194 mr_if_needed(R4_ARG2, arg_1); 2195 assert(arg_2 != R4_ARG2, "smashed argument"); 2196 mr_if_needed(R5_ARG3, arg_2); 2197 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument"); 2198 mr_if_needed(R6_ARG4, arg_3); 2199 call_VM(oop_result, entry_point, check_exceptions); 2200} 2201 2202void InterpreterMacroAssembler::save_interpreter_state(Register scratch) { 2203 ld(scratch, 0, R1_SP); 2204 std(R15_esp, _ijava_state_neg(esp), scratch); 2205 std(R14_bcp, _ijava_state_neg(bcp), scratch); 2206 std(R26_monitor, _ijava_state_neg(monitors), scratch); 2207 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); } 2208 // Other entries should be unchanged. 2209} 2210 2211void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) { 2212 ld(scratch, 0, R1_SP); 2213 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception). 2214 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code. 2215 if (!bcp_and_mdx_only) { 2216 // Following ones are Metadata. 2217 ld(R19_method, _ijava_state_neg(method), scratch); 2218 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch); 2219 // Following ones are stack addresses and don't require reload. 2220 ld(R15_esp, _ijava_state_neg(esp), scratch); 2221 ld(R18_locals, _ijava_state_neg(locals), scratch); 2222 ld(R26_monitor, _ijava_state_neg(monitors), scratch); 2223 } 2224#ifdef ASSERT 2225 { 2226 Label Lok; 2227 subf(R0, R1_SP, scratch); 2228 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size); 2229 bge(CCR0, Lok); 2230 stop("frame too small (restore istate)", 0x5432); 2231 bind(Lok); 2232 } 2233 { 2234 Label Lok; 2235 ld(R0, _ijava_state_neg(ijava_reserved), scratch); 2236 cmpdi(CCR0, R0, 0x5afe); 2237 beq(CCR0, Lok); 2238 stop("frame corrupted (restore istate)", 0x5afe); 2239 bind(Lok); 2240 } 2241#endif 2242} 2243 2244void InterpreterMacroAssembler::get_method_counters(Register method, 2245 Register Rcounters, 2246 Label& skip) { 2247 BLOCK_COMMENT("Load and ev. allocate counter object {"); 2248 Label has_counters; 2249 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2250 cmpdi(CCR0, Rcounters, 0); 2251 bne(CCR0, has_counters); 2252 call_VM(noreg, CAST_FROM_FN_PTR(address, 2253 InterpreterRuntime::build_method_counters), method, false); 2254 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2255 cmpdi(CCR0, Rcounters, 0); 2256 beq(CCR0, skip); // No MethodCounters, OutOfMemory. 2257 BLOCK_COMMENT("} Load and ev. allocate counter object"); 2258 2259 bind(has_counters); 2260} 2261 2262void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, 2263 Register iv_be_count, 2264 Register Rtmp_r0) { 2265 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 2266 Register invocation_count = iv_be_count; 2267 Register backedge_count = Rtmp_r0; 2268 int delta = InvocationCounter::count_increment; 2269 2270 // Load each counter in a register. 2271 // ld(inv_counter, Rtmp); 2272 // ld(be_counter, Rtmp2); 2273 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + 2274 InvocationCounter::counter_offset()); 2275 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + 2276 InvocationCounter::counter_offset()); 2277 2278 BLOCK_COMMENT("Increment profiling counters {"); 2279 2280 // Load the backedge counter. 2281 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int 2282 // Mask the backedge counter. 2283 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value); 2284 2285 // Load the invocation counter. 2286 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int 2287 // Add the delta to the invocation counter and store the result. 2288 addi(invocation_count, invocation_count, delta); 2289 // Store value. 2290 stw(invocation_count, inv_counter_offset, Rcounters); 2291 2292 // Add invocation counter + backedge counter. 2293 add(iv_be_count, backedge_count, invocation_count); 2294 2295 // Note that this macro must leave the backedge_count + invocation_count in 2296 // register iv_be_count! 2297 BLOCK_COMMENT("} Increment profiling counters"); 2298} 2299 2300void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2301 if (state == atos) { MacroAssembler::verify_oop(reg); } 2302} 2303 2304// Local helper function for the verify_oop_or_return_address macro. 2305static bool verify_return_address(Method* m, int bci) { 2306#ifndef PRODUCT 2307 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci; 2308 // Assume it is a valid return address if it is inside m and is preceded by a jsr. 2309 if (!m->contains(pc)) return false; 2310 address jsr_pc; 2311 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2312 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2313 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2314 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2315#endif // PRODUCT 2316 return false; 2317} 2318 2319void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2320 if (VerifyFPU) { 2321 unimplemented("verfiyFPU"); 2322 } 2323} 2324 2325void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2326 if (!VerifyOops) return; 2327 2328 // The VM documentation for the astore[_wide] bytecode allows 2329 // the TOS to be not only an oop but also a return address. 2330 Label test; 2331 Label skip; 2332 // See if it is an address (in the current method): 2333 2334 const int log2_bytecode_size_limit = 16; 2335 srdi_(Rtmp, reg, log2_bytecode_size_limit); 2336 bne(CCR0, test); 2337 2338 address fd = CAST_FROM_FN_PTR(address, verify_return_address); 2339 const int nbytes_save = MacroAssembler::num_volatile_regs * 8; 2340 save_volatile_gprs(R1_SP, -nbytes_save); // except R0 2341 save_LR_CR(Rtmp); // Save in old frame. 2342 push_frame_reg_args(nbytes_save, Rtmp); 2343 2344 load_const_optimized(Rtmp, fd, R0); 2345 mr_if_needed(R4_ARG2, reg); 2346 mr(R3_ARG1, R19_method); 2347 call_c(Rtmp); // call C 2348 2349 pop_frame(); 2350 restore_LR_CR(Rtmp); 2351 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0 2352 b(skip); 2353 2354 // Perform a more elaborate out-of-line call. 2355 // Not an address; verify it: 2356 bind(test); 2357 verify_oop(reg); 2358 bind(skip); 2359} 2360 2361// Inline assembly for: 2362// 2363// if (thread is in interp_only_mode) { 2364// InterpreterRuntime::post_method_entry(); 2365// } 2366// if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) || 2367// *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) { 2368// SharedRuntime::jvmpi_method_entry(method, receiver); 2369// } 2370void InterpreterMacroAssembler::notify_method_entry() { 2371 // JVMTI 2372 // Whenever JVMTI puts a thread in interp_only_mode, method 2373 // entry/exit events are sent for that thread to track stack 2374 // depth. If it is possible to enter interp_only_mode we add 2375 // the code to check if the event should be sent. 2376 if (JvmtiExport::can_post_interpreter_events()) { 2377 Label jvmti_post_done; 2378 2379 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2380 cmpwi(CCR0, R0, 0); 2381 beq(CCR0, jvmti_post_done); 2382 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), 2383 /*check_exceptions=*/true); 2384 2385 bind(jvmti_post_done); 2386 } 2387} 2388 2389// Inline assembly for: 2390// 2391// if (thread is in interp_only_mode) { 2392// // save result 2393// InterpreterRuntime::post_method_exit(); 2394// // restore result 2395// } 2396// if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) { 2397// // save result 2398// SharedRuntime::jvmpi_method_exit(); 2399// // restore result 2400// } 2401// 2402// Native methods have their result stored in d_tmp and l_tmp. 2403// Java methods have their result stored in the expression stack. 2404void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state, 2405 NotifyMethodExitMode mode, bool check_exceptions) { 2406 // JVMTI 2407 // Whenever JVMTI puts a thread in interp_only_mode, method 2408 // entry/exit events are sent for that thread to track stack 2409 // depth. If it is possible to enter interp_only_mode we add 2410 // the code to check if the event should be sent. 2411 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2412 Label jvmti_post_done; 2413 2414 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2415 cmpwi(CCR0, R0, 0); 2416 beq(CCR0, jvmti_post_done); 2417 if (!is_native_method) { push(state); } // Expose tos to GC. 2418 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), 2419 /*check_exceptions=*/check_exceptions); 2420 if (!is_native_method) { pop(state); } 2421 2422 align(32, 12); 2423 bind(jvmti_post_done); 2424 } 2425 2426 // Dtrace support not implemented. 2427} 2428 2429