interp_masm_ppc_64.cpp revision 9751:4a24de859a87
1/* 2 * Copyright (c) 2003, 2015, Oracle and/or its affiliates. All rights reserved. 3 * Copyright 2012, 2015 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#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#ifdef CC_INTERP 42 address exception_entry = StubRoutines::throw_NullPointerException_at_call_entry(); 43#else 44 address exception_entry = Interpreter::throw_NullPointerException_entry(); 45#endif 46 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry); 47} 48 49void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) { 50 assert(entry, "Entry must have been generated by now"); 51 if (is_within_range_of_b(entry, pc())) { 52 b(entry); 53 } else { 54 load_const_optimized(Rscratch, entry, R0); 55 mtctr(Rscratch); 56 bctr(); 57 } 58} 59 60#ifndef CC_INTERP 61 62void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { 63 Register bytecode = R12_scratch2; 64 if (bcp_incr != 0) { 65 lbzu(bytecode, bcp_incr, R14_bcp); 66 } else { 67 lbz(bytecode, 0, R14_bcp); 68 } 69 70 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state)); 71} 72 73void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 74 // Load current bytecode. 75 Register bytecode = R12_scratch2; 76 lbz(bytecode, 0, R14_bcp); 77 dispatch_Lbyte_code(state, bytecode, table); 78} 79 80// Dispatch code executed in the prolog of a bytecode which does not do it's 81// own dispatch. The dispatch address is computed and placed in R24_dispatch_addr. 82void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 83 Register bytecode = R12_scratch2; 84 lbz(bytecode, bcp_incr, R14_bcp); 85 86 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state)); 87 88 sldi(bytecode, bytecode, LogBytesPerWord); 89 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode); 90} 91 92// Dispatch code executed in the epilog of a bytecode which does not do it's 93// own dispatch. The dispatch address in R24_dispatch_addr is used for the 94// dispatch. 95void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 96 if (bcp_incr) { addi(R14_bcp, R14_bcp, bcp_incr); } 97 mtctr(R24_dispatch_addr); 98 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable); 99} 100 101void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 102 assert(scratch_reg != R0, "can't use R0 as scratch_reg here"); 103 if (JvmtiExport::can_pop_frame()) { 104 Label L; 105 106 // Check the "pending popframe condition" flag in the current thread. 107 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 108 109 // Initiate popframe handling only if it is not already being 110 // processed. If the flag has the popframe_processing bit set, it 111 // means that this code is called *during* popframe handling - we 112 // don't want to reenter. 113 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit); 114 beq(CCR0, L); 115 116 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit); 117 bne(CCR0, L); 118 119 // Call the Interpreter::remove_activation_preserving_args_entry() 120 // func to get the address of the same-named entrypoint in the 121 // generated interpreter code. 122#if defined(ABI_ELFv2) 123 call_c(CAST_FROM_FN_PTR(address, 124 Interpreter::remove_activation_preserving_args_entry), 125 relocInfo::none); 126#else 127 call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, 128 Interpreter::remove_activation_preserving_args_entry), 129 relocInfo::none); 130#endif 131 132 // Jump to Interpreter::_remove_activation_preserving_args_entry. 133 mtctr(R3_RET); 134 bctr(); 135 136 align(32, 12); 137 bind(L); 138 } 139} 140 141void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 142 const Register Rthr_state_addr = scratch_reg; 143 if (JvmtiExport::can_force_early_return()) { 144 Label Lno_early_ret; 145 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 146 cmpdi(CCR0, Rthr_state_addr, 0); 147 beq(CCR0, Lno_early_ret); 148 149 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr); 150 cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending); 151 bne(CCR0, Lno_early_ret); 152 153 // Jump to Interpreter::_earlyret_entry. 154 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr); 155 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry)); 156 mtlr(R3_RET); 157 blr(); 158 159 align(32, 12); 160 bind(Lno_early_ret); 161 } 162} 163 164void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) { 165 const Register RjvmtiState = Rscratch1; 166 const Register Rscratch2 = R0; 167 168 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 169 li(Rscratch2, 0); 170 171 switch (state) { 172 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 173 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 174 break; 175 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 176 break; 177 case btos: // 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 200void 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 210void 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 assert_different_registers(bytecode, R11_scratch1); 216 217 // Calc dispatch table address. 218 load_dispatch_table(R11_scratch1, table); 219 220 sldi(R12_scratch2, bytecode, LogBytesPerWord); 221 ldx(R11_scratch1, R11_scratch1, R12_scratch2); 222 223 // Jump off! 224 mtctr(R11_scratch1); 225 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable); 226} 227 228void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) { 229 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize); 230 ldx(Rrecv_dst, Rrecv_dst, R15_esp); 231} 232 233// helpers for expression stack 234 235void InterpreterMacroAssembler::pop_i(Register r) { 236 lwzu(r, Interpreter::stackElementSize, R15_esp); 237} 238 239void InterpreterMacroAssembler::pop_ptr(Register r) { 240 ldu(r, Interpreter::stackElementSize, R15_esp); 241} 242 243void InterpreterMacroAssembler::pop_l(Register r) { 244 ld(r, Interpreter::stackElementSize, R15_esp); 245 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 246} 247 248void InterpreterMacroAssembler::pop_f(FloatRegister f) { 249 lfsu(f, Interpreter::stackElementSize, R15_esp); 250} 251 252void InterpreterMacroAssembler::pop_d(FloatRegister f) { 253 lfd(f, Interpreter::stackElementSize, R15_esp); 254 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 255} 256 257void InterpreterMacroAssembler::push_i(Register r) { 258 stw(r, 0, R15_esp); 259 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 260} 261 262void InterpreterMacroAssembler::push_ptr(Register r) { 263 std(r, 0, R15_esp); 264 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 265} 266 267void InterpreterMacroAssembler::push_l(Register r) { 268 std(r, - Interpreter::stackElementSize, R15_esp); 269 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 270} 271 272void InterpreterMacroAssembler::push_f(FloatRegister f) { 273 stfs(f, 0, R15_esp); 274 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 275} 276 277void InterpreterMacroAssembler::push_d(FloatRegister f) { 278 stfd(f, - Interpreter::stackElementSize, R15_esp); 279 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 280} 281 282void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) { 283 std(first, 0, R15_esp); 284 std(second, -Interpreter::stackElementSize, R15_esp); 285 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 286} 287 288void InterpreterMacroAssembler::push_l_pop_d(Register l, FloatRegister d) { 289 std(l, 0, R15_esp); 290 lfd(d, 0, R15_esp); 291} 292 293void InterpreterMacroAssembler::push_d_pop_l(FloatRegister d, Register l) { 294 stfd(d, 0, R15_esp); 295 ld(l, 0, R15_esp); 296} 297 298void InterpreterMacroAssembler::push(TosState state) { 299 switch (state) { 300 case atos: push_ptr(); break; 301 case btos: 302 case ctos: 303 case stos: 304 case itos: push_i(); break; 305 case ltos: push_l(); break; 306 case ftos: push_f(); break; 307 case dtos: push_d(); break; 308 case vtos: /* nothing to do */ break; 309 default : ShouldNotReachHere(); 310 } 311} 312 313void InterpreterMacroAssembler::pop(TosState state) { 314 switch (state) { 315 case atos: pop_ptr(); break; 316 case btos: 317 case ctos: 318 case stos: 319 case itos: pop_i(); break; 320 case ltos: pop_l(); break; 321 case ftos: pop_f(); break; 322 case dtos: pop_d(); break; 323 case vtos: /* nothing to do */ break; 324 default : ShouldNotReachHere(); 325 } 326 verify_oop(R17_tos, state); 327} 328 329void InterpreterMacroAssembler::empty_expression_stack() { 330 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 331} 332 333void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset, 334 Register Rdst, 335 signedOrNot is_signed) { 336#if defined(VM_LITTLE_ENDIAN) 337 if (bcp_offset) { 338 load_const_optimized(Rdst, bcp_offset); 339 lhbrx(Rdst, R14_bcp, Rdst); 340 } else { 341 lhbrx(Rdst, R14_bcp); 342 } 343 if (is_signed == Signed) { 344 extsh(Rdst, Rdst); 345 } 346#else 347 // Read Java big endian format. 348 if (is_signed == Signed) { 349 lha(Rdst, bcp_offset, R14_bcp); 350 } else { 351 lhz(Rdst, bcp_offset, R14_bcp); 352 } 353#endif 354} 355 356void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset, 357 Register Rdst, 358 signedOrNot is_signed) { 359#if defined(VM_LITTLE_ENDIAN) 360 if (bcp_offset) { 361 load_const_optimized(Rdst, bcp_offset); 362 lwbrx(Rdst, R14_bcp, Rdst); 363 } else { 364 lwbrx(Rdst, R14_bcp); 365 } 366 if (is_signed == Signed) { 367 extsw(Rdst, Rdst); 368 } 369#else 370 // Read Java big endian format. 371 if (bcp_offset & 3) { // Offset unaligned? 372 load_const_optimized(Rdst, bcp_offset); 373 if (is_signed == Signed) { 374 lwax(Rdst, R14_bcp, Rdst); 375 } else { 376 lwzx(Rdst, R14_bcp, Rdst); 377 } 378 } else { 379 if (is_signed == Signed) { 380 lwa(Rdst, bcp_offset, R14_bcp); 381 } else { 382 lwz(Rdst, bcp_offset, R14_bcp); 383 } 384 } 385#endif 386} 387 388 389// Load the constant pool cache index from the bytecode stream. 390// 391// Kills / writes: 392// - Rdst, Rscratch 393void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, size_t index_size) { 394 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 395 // Cache index is always in the native format, courtesy of Rewriter. 396 if (index_size == sizeof(u2)) { 397 lhz(Rdst, bcp_offset, R14_bcp); 398 } else if (index_size == sizeof(u4)) { 399 if (bcp_offset & 3) { 400 load_const_optimized(Rdst, bcp_offset); 401 lwax(Rdst, R14_bcp, Rdst); 402 } else { 403 lwa(Rdst, bcp_offset, R14_bcp); 404 } 405 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 406 nand(Rdst, Rdst, Rdst); // convert to plain index 407 } else if (index_size == sizeof(u1)) { 408 lbz(Rdst, bcp_offset, R14_bcp); 409 } else { 410 ShouldNotReachHere(); 411 } 412 // Rdst now contains cp cache index. 413} 414 415void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset, 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 sub(Rmem_frame_size, R1_SP, Rmem_frame_size); 484 ld(Rscratch1, thread_(stack_overflow_limit)); 485 cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1); 486 bgt(CCR0/*is_stack_overflow*/, done); 487 488 // Load target address of the runtime stub. 489 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order"); 490 load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0); 491 mtctr(Rscratch1); 492 // Restore caller_sp. 493#ifdef ASSERT 494 ld(Rscratch1, 0, R1_SP); 495 ld(R0, 0, R21_sender_SP); 496 cmpd(CCR0, R0, Rscratch1); 497 asm_assert_eq("backlink", 0x547); 498#endif // ASSERT 499 mr(R1_SP, R21_sender_SP); 500 bctr(); 501 502 align(32, 12); 503 bind(done); 504} 505 506// Separate these two to allow for delay slot in middle. 507// These are used to do a test and full jump to exception-throwing code. 508 509// Check that index is in range for array, then shift index by index_shift, 510// and put arrayOop + shifted_index into res. 511// Note: res is still shy of address by array offset into object. 512 513void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, int index_shift, Register Rtmp, Register Rres) { 514 // Check that index is in range for array, then shift index by index_shift, 515 // and put arrayOop + shifted_index into res. 516 // Note: res is still shy of address by array offset into object. 517 // Kills: 518 // - Rindex 519 // Writes: 520 // - Rres: Address that corresponds to the array index if check was successful. 521 verify_oop(Rarray); 522 const Register Rlength = R0; 523 const Register RsxtIndex = Rtmp; 524 Label LisNull, LnotOOR; 525 526 // Array nullcheck 527 if (!ImplicitNullChecks) { 528 cmpdi(CCR0, Rarray, 0); 529 beq(CCR0, LisNull); 530 } else { 531 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex); 532 } 533 534 // Rindex might contain garbage in upper bits (remember that we don't sign extend 535 // during integer arithmetic operations). So kill them and put value into same register 536 // where ArrayIndexOutOfBounds would expect the index in. 537 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit 538 539 // Index check 540 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray); 541 cmplw(CCR0, Rindex, Rlength); 542 sldi(RsxtIndex, RsxtIndex, index_shift); 543 blt(CCR0, LnotOOR); 544 // Index should be in R17_tos, array should be in R4_ARG2. 545 mr_if_needed(R17_tos, Rindex); 546 mr_if_needed(R4_ARG2, Rarray); 547 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry); 548 mtctr(Rtmp); 549 bctr(); 550 551 if (!ImplicitNullChecks) { 552 bind(LisNull); 553 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry); 554 mtctr(Rtmp); 555 bctr(); 556 } 557 558 align(32, 16); 559 bind(LnotOOR); 560 561 // Calc address 562 add(Rres, RsxtIndex, Rarray); 563} 564 565void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { 566 // pop array 567 pop_ptr(array); 568 569 // check array 570 index_check_without_pop(array, index, index_shift, tmp, res); 571} 572 573void InterpreterMacroAssembler::get_const(Register Rdst) { 574 ld(Rdst, in_bytes(Method::const_offset()), R19_method); 575} 576 577void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 578 get_const(Rdst); 579 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 580} 581 582void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 583 get_constant_pool(Rdst); 584 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 585} 586 587void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 588 get_constant_pool(Rcpool); 589 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool); 590} 591 592// Unlock if synchronized method. 593// 594// Unlock the receiver if this is a synchronized method. 595// Unlock any Java monitors from synchronized blocks. 596// 597// If there are locked Java monitors 598// If throw_monitor_exception 599// throws IllegalMonitorStateException 600// Else if install_monitor_exception 601// installs IllegalMonitorStateException 602// Else 603// no error processing 604void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 605 bool throw_monitor_exception, 606 bool install_monitor_exception) { 607 Label Lunlocked, Lno_unlock; 608 { 609 Register Rdo_not_unlock_flag = R11_scratch1; 610 Register Raccess_flags = R12_scratch2; 611 612 // Check if synchronized method or unlocking prevented by 613 // JavaThread::do_not_unlock_if_synchronized flag. 614 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 615 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method); 616 li(R0, 0); 617 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag 618 619 push(state); 620 621 // Skip if we don't have to unlock. 622 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0. 623 beq(CCR0, Lunlocked); 624 625 cmpwi(CCR0, Rdo_not_unlock_flag, 0); 626 bne(CCR0, Lno_unlock); 627 } 628 629 // Unlock 630 { 631 Register Rmonitor_base = R11_scratch1; 632 633 Label Lunlock; 634 // If it's still locked, everything is ok, unlock it. 635 ld(Rmonitor_base, 0, R1_SP); 636 addi(Rmonitor_base, Rmonitor_base, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base 637 638 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base); 639 cmpdi(CCR0, R0, 0); 640 bne(CCR0, Lunlock); 641 642 // If it's already unlocked, throw exception. 643 if (throw_monitor_exception) { 644 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 645 should_not_reach_here(); 646 } else { 647 if (install_monitor_exception) { 648 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 649 b(Lunlocked); 650 } 651 } 652 653 bind(Lunlock); 654 unlock_object(Rmonitor_base); 655 } 656 657 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not. 658 bind(Lunlocked); 659 { 660 Label Lexception, Lrestart; 661 Register Rcurrent_obj_addr = R11_scratch1; 662 const int delta = frame::interpreter_frame_monitor_size_in_bytes(); 663 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords"); 664 665 bind(Lrestart); 666 // Set up search loop: Calc num of iterations. 667 { 668 Register Riterations = R12_scratch2; 669 Register Rmonitor_base = Rcurrent_obj_addr; 670 ld(Rmonitor_base, 0, R1_SP); 671 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base 672 673 subf_(Riterations, R26_monitor, Rmonitor_base); 674 ble(CCR0, Lno_unlock); 675 676 addi(Rcurrent_obj_addr, Rmonitor_base, BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes()); 677 // Check if any monitor is on stack, bail out if not 678 srdi(Riterations, Riterations, exact_log2(delta)); 679 mtctr(Riterations); 680 } 681 682 // The search loop: Look for locked monitors. 683 { 684 const Register Rcurrent_obj = R0; 685 Label Lloop; 686 687 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 688 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 689 bind(Lloop); 690 691 // Check if current entry is used. 692 cmpdi(CCR0, Rcurrent_obj, 0); 693 bne(CCR0, Lexception); 694 // Preload next iteration's compare value. 695 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 696 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 697 bdnz(Lloop); 698 } 699 // Fell through: Everything's unlocked => finish. 700 b(Lno_unlock); 701 702 // An object is still locked => need to throw exception. 703 bind(Lexception); 704 if (throw_monitor_exception) { 705 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 706 should_not_reach_here(); 707 } else { 708 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 709 // Unlock does not block, so don't have to worry about the frame. 710 Register Rmonitor_addr = R11_scratch1; 711 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta); 712 unlock_object(Rmonitor_addr); 713 if (install_monitor_exception) { 714 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 715 } 716 b(Lrestart); 717 } 718 } 719 720 align(32, 12); 721 bind(Lno_unlock); 722 pop(state); 723} 724 725// Support function for remove_activation & Co. 726void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, Register Rscratch1, Register Rscratch2) { 727 // Pop interpreter frame. 728 ld(Rscratch1, 0, R1_SP); // *SP 729 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp 730 ld(Rscratch2, 0, Rscratch1); // **SP 731#ifdef ASSERT 732 { 733 Label Lok; 734 ld(R0, _ijava_state_neg(ijava_reserved), Rscratch1); 735 cmpdi(CCR0, R0, 0x5afe); 736 beq(CCR0, Lok); 737 stop("frame corrupted (remove activation)", 0x5afe); 738 bind(Lok); 739 } 740#endif 741 if (return_pc!=noreg) { 742 ld(return_pc, _abi(lr), Rscratch1); // LR 743 } 744 745 // Merge top frames. 746 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP 747 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP 748} 749 750// Remove activation. 751// 752// Unlock the receiver if this is a synchronized method. 753// Unlock any Java monitors from synchronized blocks. 754// Remove the activation from the stack. 755// 756// If there are locked Java monitors 757// If throw_monitor_exception 758// throws IllegalMonitorStateException 759// Else if install_monitor_exception 760// installs IllegalMonitorStateException 761// Else 762// no error processing 763void InterpreterMacroAssembler::remove_activation(TosState state, 764 bool throw_monitor_exception, 765 bool install_monitor_exception) { 766 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 767 768 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 769 notify_method_exit(false, state, NotifyJVMTI, true); 770 771 verify_oop(R17_tos, state); 772 verify_thread(); 773 774 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 775 mtlr(R0); 776} 777 778#endif // !CC_INTERP 779 780// Lock object 781// 782// Registers alive 783// monitor - Address of the BasicObjectLock to be used for locking, 784// which must be initialized with the object to lock. 785// object - Address of the object to be locked. 786// 787void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 788 if (UseHeavyMonitors) { 789 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 790 monitor, /*check_for_exceptions=*/true CC_INTERP_ONLY(&& false)); 791 } else { 792 // template code: 793 // 794 // markOop displaced_header = obj->mark().set_unlocked(); 795 // monitor->lock()->set_displaced_header(displaced_header); 796 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 797 // // We stored the monitor address into the object's mark word. 798 // } else if (THREAD->is_lock_owned((address)displaced_header)) 799 // // Simple recursive case. 800 // monitor->lock()->set_displaced_header(NULL); 801 // } else { 802 // // Slow path. 803 // InterpreterRuntime::monitorenter(THREAD, monitor); 804 // } 805 806 const Register displaced_header = R7_ARG5; 807 const Register object_mark_addr = R8_ARG6; 808 const Register current_header = R9_ARG7; 809 const Register tmp = R10_ARG8; 810 811 Label done; 812 Label cas_failed, slow_case; 813 814 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp); 815 816 // markOop displaced_header = obj->mark().set_unlocked(); 817 818 // Load markOop from object into displaced_header. 819 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object); 820 821 if (UseBiasedLocking) { 822 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case); 823 } 824 825 // Set displaced_header to be (markOop of object | UNLOCK_VALUE). 826 ori(displaced_header, displaced_header, markOopDesc::unlocked_value); 827 828 // monitor->lock()->set_displaced_header(displaced_header); 829 830 // Initialize the box (Must happen before we update the object mark!). 831 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 832 BasicLock::displaced_header_offset_in_bytes(), monitor); 833 834 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 835 836 // Store stack address of the BasicObjectLock (this is monitor) into object. 837 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 838 839 // Must fence, otherwise, preceding store(s) may float below cmpxchg. 840 // CmpxchgX sets CCR0 to cmpX(current, displaced). 841 cmpxchgd(/*flag=*/CCR0, 842 /*current_value=*/current_header, 843 /*compare_value=*/displaced_header, /*exchange_value=*/monitor, 844 /*where=*/object_mark_addr, 845 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq, 846 MacroAssembler::cmpxchgx_hint_acquire_lock(), 847 noreg, 848 &cas_failed, 849 /*check without membar and ldarx first*/true); 850 851 // If the compare-and-exchange succeeded, then we found an unlocked 852 // object and we have now locked it. 853 b(done); 854 bind(cas_failed); 855 856 // } else if (THREAD->is_lock_owned((address)displaced_header)) 857 // // Simple recursive case. 858 // monitor->lock()->set_displaced_header(NULL); 859 860 // We did not see an unlocked object so try the fast recursive case. 861 862 // Check if owner is self by comparing the value in the markOop of object 863 // (current_header) with the stack pointer. 864 sub(current_header, current_header, R1_SP); 865 866 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 867 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place); 868 869 and_(R0/*==0?*/, current_header, tmp); 870 // If condition is true we are done and hence we can store 0 in the displaced 871 // header indicating it is a recursive lock. 872 bne(CCR0, slow_case); 873 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() + 874 BasicLock::displaced_header_offset_in_bytes(), monitor); 875 b(done); 876 877 // } else { 878 // // Slow path. 879 // InterpreterRuntime::monitorenter(THREAD, monitor); 880 881 // None of the above fast optimizations worked so we have to get into the 882 // slow case of monitor enter. 883 bind(slow_case); 884 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 885 monitor, /*check_for_exceptions=*/true CC_INTERP_ONLY(&& false)); 886 // } 887 align(32, 12); 888 bind(done); 889 } 890} 891 892// Unlocks an object. Used in monitorexit bytecode and remove_activation. 893// 894// Registers alive 895// monitor - Address of the BasicObjectLock to be used for locking, 896// which must be initialized with the object to lock. 897// 898// Throw IllegalMonitorException if object is not locked by current thread. 899void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) { 900 if (UseHeavyMonitors) { 901 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 902 monitor, check_for_exceptions CC_INTERP_ONLY(&& false)); 903 } else { 904 905 // template code: 906 // 907 // if ((displaced_header = monitor->displaced_header()) == NULL) { 908 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 909 // monitor->set_obj(NULL); 910 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 911 // // We swapped the unlocked mark in displaced_header into the object's mark word. 912 // monitor->set_obj(NULL); 913 // } else { 914 // // Slow path. 915 // InterpreterRuntime::monitorexit(THREAD, monitor); 916 // } 917 918 const Register object = R7_ARG5; 919 const Register displaced_header = R8_ARG6; 920 const Register object_mark_addr = R9_ARG7; 921 const Register current_header = R10_ARG8; 922 923 Label free_slot; 924 Label slow_case; 925 926 assert_different_registers(object, displaced_header, object_mark_addr, current_header); 927 928 if (UseBiasedLocking) { 929 // The object address from the monitor is in object. 930 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); 931 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0"); 932 biased_locking_exit(CCR0, object, displaced_header, free_slot); 933 } 934 935 // Test first if we are in the fast recursive case. 936 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 937 BasicLock::displaced_header_offset_in_bytes(), monitor); 938 939 // If the displaced header is zero, we have a recursive unlock. 940 cmpdi(CCR0, displaced_header, 0); 941 beq(CCR0, free_slot); // recursive unlock 942 943 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 944 // // We swapped the unlocked mark in displaced_header into the object's mark word. 945 // monitor->set_obj(NULL); 946 947 // If we still have a lightweight lock, unlock the object and be done. 948 949 // The object address from the monitor is in object. 950 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); } 951 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 952 953 // We have the displaced header in displaced_header. If the lock is still 954 // lightweight, it will contain the monitor address and we'll store the 955 // displaced header back into the object's mark word. 956 // CmpxchgX sets CCR0 to cmpX(current, monitor). 957 cmpxchgd(/*flag=*/CCR0, 958 /*current_value=*/current_header, 959 /*compare_value=*/monitor, /*exchange_value=*/displaced_header, 960 /*where=*/object_mark_addr, 961 MacroAssembler::MemBarRel, 962 MacroAssembler::cmpxchgx_hint_release_lock(), 963 noreg, 964 &slow_case); 965 b(free_slot); 966 967 // } else { 968 // // Slow path. 969 // InterpreterRuntime::monitorexit(THREAD, monitor); 970 971 // The lock has been converted into a heavy lock and hence 972 // we need to get into the slow case. 973 bind(slow_case); 974 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 975 monitor, check_for_exceptions CC_INTERP_ONLY(&& false)); 976 // } 977 978 Label done; 979 b(done); // Monitor register may be overwritten! Runtime has already freed the slot. 980 981 // Exchange worked, do monitor->set_obj(NULL); 982 align(32, 12); 983 bind(free_slot); 984 li(R0, 0); 985 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor); 986 bind(done); 987 } 988} 989 990#ifndef CC_INTERP 991 992// Load compiled (i2c) or interpreter entry when calling from interpreted and 993// do the call. Centralized so that all interpreter calls will do the same actions. 994// If jvmti single stepping is on for a thread we must not call compiled code. 995// 996// Input: 997// - Rtarget_method: method to call 998// - Rret_addr: return address 999// - 2 scratch regs 1000// 1001void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, Register Rscratch1, Register Rscratch2) { 1002 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr); 1003 // Assume we want to go compiled if available. 1004 const Register Rtarget_addr = Rscratch1; 1005 const Register Rinterp_only = Rscratch2; 1006 1007 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method); 1008 1009 if (JvmtiExport::can_post_interpreter_events()) { 1010 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 1011 1012 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 1013 // compiled code in threads for which the event is enabled. Check here for 1014 // interp_only_mode if these events CAN be enabled. 1015 Label done; 1016 verify_thread(); 1017 cmpwi(CCR0, Rinterp_only, 0); 1018 beq(CCR0, done); 1019 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method); 1020 align(32, 12); 1021 bind(done); 1022 } 1023 1024#ifdef ASSERT 1025 { 1026 Label Lok; 1027 cmpdi(CCR0, Rtarget_addr, 0); 1028 bne(CCR0, Lok); 1029 stop("null entry point"); 1030 bind(Lok); 1031 } 1032#endif // ASSERT 1033 1034 mr(R21_sender_SP, R1_SP); 1035 1036 // Calc a precise SP for the call. The SP value we calculated in 1037 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space 1038 // if esp is not max. Also, the i2c adapter extends the stack space without restoring 1039 // our pre-calced value, so repeating calls via i2c would result in stack overflow. 1040 // Since esp already points to an empty slot, we just have to sub 1 additional slot 1041 // to meet the abi scratch requirements. 1042 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in 1043 // the return entry of the interpreter. 1044 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size); 1045 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address 1046 resize_frame_absolute(Rscratch2, Rscratch2, R0); 1047 1048 mr_if_needed(R19_method, Rtarget_method); 1049 mtctr(Rtarget_addr); 1050 mtlr(Rret_addr); 1051 1052 save_interpreter_state(Rscratch2); 1053#ifdef ASSERT 1054 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp 1055 cmpd(CCR0, R21_sender_SP, Rscratch1); 1056 asm_assert_eq("top_frame_sp incorrect", 0x951); 1057#endif 1058 1059 bctr(); 1060} 1061 1062// Set the method data pointer for the current bcp. 1063void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1064 assert(ProfileInterpreter, "must be profiling interpreter"); 1065 Label get_continue; 1066 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method); 1067 test_method_data_pointer(get_continue); 1068 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp); 1069 1070 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1071 add(R28_mdx, R28_mdx, R3_RET); 1072 bind(get_continue); 1073} 1074 1075// Test ImethodDataPtr. If it is null, continue at the specified label. 1076void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1077 assert(ProfileInterpreter, "must be profiling interpreter"); 1078 cmpdi(CCR0, R28_mdx, 0); 1079 beq(CCR0, zero_continue); 1080} 1081 1082void InterpreterMacroAssembler::verify_method_data_pointer() { 1083 assert(ProfileInterpreter, "must be profiling interpreter"); 1084#ifdef ASSERT 1085 Label verify_continue; 1086 test_method_data_pointer(verify_continue); 1087 1088 // If the mdp is valid, it will point to a DataLayout header which is 1089 // consistent with the bcp. The converse is highly probable also. 1090 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx); 1091 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method); 1092 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset())); 1093 add(R11_scratch1, R12_scratch2, R12_scratch2); 1094 cmpd(CCR0, R11_scratch1, R14_bcp); 1095 beq(CCR0, verify_continue); 1096 1097 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx); 1098 1099 bind(verify_continue); 1100#endif 1101} 1102 1103void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1104 Register method_counters, 1105 Register Rscratch, 1106 Label &profile_continue) { 1107 assert(ProfileInterpreter, "must be profiling interpreter"); 1108 // Control will flow to "profile_continue" if the counter is less than the 1109 // limit or if we call profile_method(). 1110 Label done; 1111 1112 // If no method data exists, and the counter is high enough, make one. 1113 lwz(Rscratch, in_bytes(MethodCounters::interpreter_profile_limit_offset()), method_counters); 1114 1115 cmpdi(CCR0, R28_mdx, 0); 1116 // Test to see if we should create a method data oop. 1117 cmpd(CCR1, Rscratch, invocation_count); 1118 bne(CCR0, done); 1119 bge(CCR1, profile_continue); 1120 1121 // Build it now. 1122 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1123 set_method_data_pointer_for_bcp(); 1124 b(profile_continue); 1125 1126 align(32, 12); 1127 bind(done); 1128} 1129 1130void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register method_counters, 1131 Register target_bcp, Register disp, Register Rtmp) { 1132 assert_different_registers(backedge_count, target_bcp, disp, Rtmp, R4_ARG2); 1133 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 1134 1135 Label did_not_overflow; 1136 Label overflow_with_error; 1137 1138 lwz(Rtmp, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()), method_counters); 1139 cmpw(CCR0, backedge_count, Rtmp); 1140 1141 blt(CCR0, did_not_overflow); 1142 1143 // When ProfileInterpreter is on, the backedge_count comes from the 1144 // methodDataOop, which value does not get reset on the call to 1145 // frequency_counter_overflow(). To avoid excessive calls to the overflow 1146 // routine while the method is being compiled, add a second test to make sure 1147 // the overflow function is called only once every overflow_frequency. 1148 if (ProfileInterpreter) { 1149 const int overflow_frequency = 1024; 1150 andi_(Rtmp, backedge_count, overflow_frequency-1); 1151 bne(CCR0, did_not_overflow); 1152 } 1153 1154 // Overflow in loop, pass branch bytecode. 1155 subf(R4_ARG2, disp, target_bcp); // Compute branch bytecode (previous bcp). 1156 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); 1157 1158 // Was an OSR adapter generated? 1159 cmpdi(CCR0, R3_RET, 0); 1160 beq(CCR0, overflow_with_error); 1161 1162 // Has the nmethod been invalidated already? 1163 lbz(Rtmp, nmethod::state_offset(), R3_RET); 1164 cmpwi(CCR0, Rtmp, nmethod::in_use); 1165 bne(CCR0, overflow_with_error); 1166 1167 // Migrate the interpreter frame off of the stack. 1168 // We can use all registers because we will not return to interpreter from this point. 1169 1170 // Save nmethod. 1171 const Register osr_nmethod = R31; 1172 mr(osr_nmethod, R3_RET); 1173 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1); 1174 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread); 1175 reset_last_Java_frame(); 1176 // OSR buffer is in ARG1 1177 1178 // Remove the interpreter frame. 1179 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1180 1181 // Jump to the osr code. 1182 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod); 1183 mtlr(R0); 1184 mtctr(R11_scratch1); 1185 bctr(); 1186 1187 align(32, 12); 1188 bind(overflow_with_error); 1189 bind(did_not_overflow); 1190} 1191 1192// Store a value at some constant offset from the method data pointer. 1193void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1194 assert(ProfileInterpreter, "must be profiling interpreter"); 1195 1196 std(value, constant, R28_mdx); 1197} 1198 1199// Increment the value at some constant offset from the method data pointer. 1200void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1201 Register counter_addr, 1202 Register Rbumped_count, 1203 bool decrement) { 1204 // Locate the counter at a fixed offset from the mdp: 1205 addi(counter_addr, R28_mdx, constant); 1206 increment_mdp_data_at(counter_addr, Rbumped_count, decrement); 1207} 1208 1209// Increment the value at some non-fixed (reg + constant) offset from 1210// the method data pointer. 1211void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1212 int constant, 1213 Register scratch, 1214 Register Rbumped_count, 1215 bool decrement) { 1216 // Add the constant to reg to get the offset. 1217 add(scratch, R28_mdx, reg); 1218 // Then calculate the counter address. 1219 addi(scratch, scratch, constant); 1220 increment_mdp_data_at(scratch, Rbumped_count, decrement); 1221} 1222 1223void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr, 1224 Register Rbumped_count, 1225 bool decrement) { 1226 assert(ProfileInterpreter, "must be profiling interpreter"); 1227 1228 // Load the counter. 1229 ld(Rbumped_count, 0, counter_addr); 1230 1231 if (decrement) { 1232 // Decrement the register. Set condition codes. 1233 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment); 1234 // Store the decremented counter, if it is still negative. 1235 std(Rbumped_count, 0, counter_addr); 1236 // Note: add/sub overflow check are not ported, since 64 bit 1237 // calculation should never overflow. 1238 } else { 1239 // Increment the register. Set carry flag. 1240 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment); 1241 // Store the incremented counter. 1242 std(Rbumped_count, 0, counter_addr); 1243 } 1244} 1245 1246// Set a flag value at the current method data pointer position. 1247void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1248 Register scratch) { 1249 assert(ProfileInterpreter, "must be profiling interpreter"); 1250 // Load the data header. 1251 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1252 // Set the flag. 1253 ori(scratch, scratch, flag_constant); 1254 // Store the modified header. 1255 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1256} 1257 1258// Test the location at some offset from the method data pointer. 1259// If it is not equal to value, branch to the not_equal_continue Label. 1260void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1261 Register value, 1262 Label& not_equal_continue, 1263 Register test_out) { 1264 assert(ProfileInterpreter, "must be profiling interpreter"); 1265 1266 ld(test_out, offset, R28_mdx); 1267 cmpd(CCR0, value, test_out); 1268 bne(CCR0, not_equal_continue); 1269} 1270 1271// Update the method data pointer by the displacement located at some fixed 1272// offset from the method data pointer. 1273void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1274 Register scratch) { 1275 assert(ProfileInterpreter, "must be profiling interpreter"); 1276 1277 ld(scratch, offset_of_disp, R28_mdx); 1278 add(R28_mdx, scratch, R28_mdx); 1279} 1280 1281// Update the method data pointer by the displacement located at the 1282// offset (reg + offset_of_disp). 1283void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1284 int offset_of_disp, 1285 Register scratch) { 1286 assert(ProfileInterpreter, "must be profiling interpreter"); 1287 1288 add(scratch, reg, R28_mdx); 1289 ld(scratch, offset_of_disp, scratch); 1290 add(R28_mdx, scratch, R28_mdx); 1291} 1292 1293// Update the method data pointer by a simple constant displacement. 1294void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1295 assert(ProfileInterpreter, "must be profiling interpreter"); 1296 addi(R28_mdx, R28_mdx, constant); 1297} 1298 1299// Update the method data pointer for a _ret bytecode whose target 1300// was not among our cached targets. 1301void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1302 Register return_bci) { 1303 assert(ProfileInterpreter, "must be profiling interpreter"); 1304 1305 push(state); 1306 assert(return_bci->is_nonvolatile(), "need to protect return_bci"); 1307 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1308 pop(state); 1309} 1310 1311// Increments the backedge counter. 1312// Returns backedge counter + invocation counter in Rdst. 1313void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst, 1314 const Register Rtmp1, Register Rscratch) { 1315 assert(UseCompiler, "incrementing must be useful"); 1316 assert_different_registers(Rdst, Rtmp1); 1317 const Register invocation_counter = Rtmp1; 1318 const Register counter = Rdst; 1319 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size."); 1320 1321 // Load backedge counter. 1322 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1323 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1324 // Load invocation counter. 1325 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) + 1326 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1327 1328 // Add the delta to the backedge counter. 1329 addi(counter, counter, InvocationCounter::count_increment); 1330 1331 // Mask the invocation counter. 1332 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value); 1333 1334 // Store new counter value. 1335 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1336 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1337 // Return invocation counter + backedge counter. 1338 add(counter, counter, invocation_counter); 1339} 1340 1341// Count a taken branch in the bytecodes. 1342void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1343 if (ProfileInterpreter) { 1344 Label profile_continue; 1345 1346 // If no method data exists, go to profile_continue. 1347 test_method_data_pointer(profile_continue); 1348 1349 // We are taking a branch. Increment the taken count. 1350 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count); 1351 1352 // The method data pointer needs to be updated to reflect the new target. 1353 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1354 bind (profile_continue); 1355 } 1356} 1357 1358// Count a not-taken branch in the bytecodes. 1359void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) { 1360 if (ProfileInterpreter) { 1361 Label profile_continue; 1362 1363 // If no method data exists, go to profile_continue. 1364 test_method_data_pointer(profile_continue); 1365 1366 // We are taking a branch. Increment the not taken count. 1367 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2); 1368 1369 // The method data pointer needs to be updated to correspond to the 1370 // next bytecode. 1371 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1372 bind (profile_continue); 1373 } 1374} 1375 1376// Count a non-virtual call in the bytecodes. 1377void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) { 1378 if (ProfileInterpreter) { 1379 Label profile_continue; 1380 1381 // If no method data exists, go to profile_continue. 1382 test_method_data_pointer(profile_continue); 1383 1384 // We are making a call. Increment the count. 1385 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1386 1387 // The method data pointer needs to be updated to reflect the new target. 1388 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1389 bind (profile_continue); 1390 } 1391} 1392 1393// Count a final call in the bytecodes. 1394void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) { 1395 if (ProfileInterpreter) { 1396 Label profile_continue; 1397 1398 // If no method data exists, go to profile_continue. 1399 test_method_data_pointer(profile_continue); 1400 1401 // We are making a call. Increment the count. 1402 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1403 1404 // The method data pointer needs to be updated to reflect the new target. 1405 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1406 bind (profile_continue); 1407 } 1408} 1409 1410// Count a virtual call in the bytecodes. 1411void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver, 1412 Register Rscratch1, 1413 Register Rscratch2, 1414 bool receiver_can_be_null) { 1415 if (!ProfileInterpreter) { return; } 1416 Label profile_continue; 1417 1418 // If no method data exists, go to profile_continue. 1419 test_method_data_pointer(profile_continue); 1420 1421 Label skip_receiver_profile; 1422 if (receiver_can_be_null) { 1423 Label not_null; 1424 cmpdi(CCR0, Rreceiver, 0); 1425 bne(CCR0, not_null); 1426 // We are making a call. Increment the count for null receiver. 1427 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2); 1428 b(skip_receiver_profile); 1429 bind(not_null); 1430 } 1431 1432 // Record the receiver type. 1433 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true); 1434 bind(skip_receiver_profile); 1435 1436 // The method data pointer needs to be updated to reflect the new target. 1437 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1438 bind (profile_continue); 1439} 1440 1441void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) { 1442 if (ProfileInterpreter) { 1443 Label profile_continue; 1444 1445 // If no method data exists, go to profile_continue. 1446 test_method_data_pointer(profile_continue); 1447 1448 int mdp_delta = in_bytes(BitData::bit_data_size()); 1449 if (TypeProfileCasts) { 1450 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1451 1452 // Record the object type. 1453 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false); 1454 } 1455 1456 // The method data pointer needs to be updated. 1457 update_mdp_by_constant(mdp_delta); 1458 1459 bind (profile_continue); 1460 } 1461} 1462 1463void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) { 1464 if (ProfileInterpreter && TypeProfileCasts) { 1465 Label profile_continue; 1466 1467 // If no method data exists, go to profile_continue. 1468 test_method_data_pointer(profile_continue); 1469 1470 int count_offset = in_bytes(CounterData::count_offset()); 1471 // Back up the address, since we have already bumped the mdp. 1472 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1473 1474 // *Decrement* the counter. We expect to see zero or small negatives. 1475 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true); 1476 1477 bind (profile_continue); 1478 } 1479} 1480 1481// Count a ret in the bytecodes. 1482void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, Register scratch1, Register scratch2) { 1483 if (ProfileInterpreter) { 1484 Label profile_continue; 1485 uint row; 1486 1487 // If no method data exists, go to profile_continue. 1488 test_method_data_pointer(profile_continue); 1489 1490 // Update the total ret count. 1491 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 ); 1492 1493 for (row = 0; row < RetData::row_limit(); row++) { 1494 Label next_test; 1495 1496 // See if return_bci is equal to bci[n]: 1497 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1); 1498 1499 // return_bci is equal to bci[n]. Increment the count. 1500 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2); 1501 1502 // The method data pointer needs to be updated to reflect the new target. 1503 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1); 1504 b(profile_continue); 1505 bind(next_test); 1506 } 1507 1508 update_mdp_for_ret(state, return_bci); 1509 1510 bind (profile_continue); 1511 } 1512} 1513 1514// Count the default case of a switch construct. 1515void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) { 1516 if (ProfileInterpreter) { 1517 Label profile_continue; 1518 1519 // If no method data exists, go to profile_continue. 1520 test_method_data_pointer(profile_continue); 1521 1522 // Update the default case count 1523 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1524 scratch1, scratch2); 1525 1526 // The method data pointer needs to be updated. 1527 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()), 1528 scratch1); 1529 1530 bind (profile_continue); 1531 } 1532} 1533 1534// Count the index'th case of a switch construct. 1535void InterpreterMacroAssembler::profile_switch_case(Register index, 1536 Register scratch1, 1537 Register scratch2, 1538 Register scratch3) { 1539 if (ProfileInterpreter) { 1540 assert_different_registers(index, scratch1, scratch2, scratch3); 1541 Label profile_continue; 1542 1543 // If no method data exists, go to profile_continue. 1544 test_method_data_pointer(profile_continue); 1545 1546 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes(). 1547 li(scratch3, in_bytes(MultiBranchData::case_array_offset())); 1548 1549 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works"); 1550 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1551 add(scratch1, scratch1, scratch3); 1552 1553 // Update the case count. 1554 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3); 1555 1556 // The method data pointer needs to be updated. 1557 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2); 1558 1559 bind (profile_continue); 1560 } 1561} 1562 1563void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) { 1564 if (ProfileInterpreter) { 1565 assert_different_registers(Rscratch1, Rscratch2); 1566 Label profile_continue; 1567 1568 // If no method data exists, go to profile_continue. 1569 test_method_data_pointer(profile_continue); 1570 1571 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1); 1572 1573 // The method data pointer needs to be updated. 1574 int mdp_delta = in_bytes(BitData::bit_data_size()); 1575 if (TypeProfileCasts) { 1576 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1577 } 1578 update_mdp_by_constant(mdp_delta); 1579 1580 bind (profile_continue); 1581 } 1582} 1583 1584void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver, 1585 Register Rscratch1, Register Rscratch2, 1586 bool is_virtual_call) { 1587 assert(ProfileInterpreter, "must be profiling"); 1588 assert_different_registers(Rreceiver, Rscratch1, Rscratch2); 1589 1590 Label done; 1591 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call); 1592 bind (done); 1593} 1594 1595void InterpreterMacroAssembler::record_klass_in_profile_helper( 1596 Register receiver, Register scratch1, Register scratch2, 1597 int start_row, Label& done, bool is_virtual_call) { 1598 if (TypeProfileWidth == 0) { 1599 if (is_virtual_call) { 1600 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1601 } 1602 return; 1603 } 1604 1605 int last_row = VirtualCallData::row_limit() - 1; 1606 assert(start_row <= last_row, "must be work left to do"); 1607 // Test this row for both the receiver and for null. 1608 // Take any of three different outcomes: 1609 // 1. found receiver => increment count and goto done 1610 // 2. found null => keep looking for case 1, maybe allocate this cell 1611 // 3. found something else => keep looking for cases 1 and 2 1612 // Case 3 is handled by a recursive call. 1613 for (int row = start_row; row <= last_row; row++) { 1614 Label next_test; 1615 bool test_for_null_also = (row == start_row); 1616 1617 // See if the receiver is receiver[n]. 1618 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1619 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1); 1620 // delayed()->tst(scratch); 1621 1622 // The receiver is receiver[n]. Increment count[n]. 1623 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1624 increment_mdp_data_at(count_offset, scratch1, scratch2); 1625 b(done); 1626 bind(next_test); 1627 1628 if (test_for_null_also) { 1629 Label found_null; 1630 // Failed the equality check on receiver[n]... Test for null. 1631 if (start_row == last_row) { 1632 // The only thing left to do is handle the null case. 1633 if (is_virtual_call) { 1634 // Scratch1 contains test_out from test_mdp_data_at. 1635 cmpdi(CCR0, scratch1, 0); 1636 beq(CCR0, found_null); 1637 // Receiver did not match any saved receiver and there is no empty row for it. 1638 // Increment total counter to indicate polymorphic case. 1639 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1640 b(done); 1641 bind(found_null); 1642 } else { 1643 cmpdi(CCR0, scratch1, 0); 1644 bne(CCR0, done); 1645 } 1646 break; 1647 } 1648 // Since null is rare, make it be the branch-taken case. 1649 cmpdi(CCR0, scratch1, 0); 1650 beq(CCR0, found_null); 1651 1652 // Put all the "Case 3" tests here. 1653 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call); 1654 1655 // Found a null. Keep searching for a matching receiver, 1656 // but remember that this is an empty (unused) slot. 1657 bind(found_null); 1658 } 1659 } 1660 1661 // In the fall-through case, we found no matching receiver, but we 1662 // observed the receiver[start_row] is NULL. 1663 1664 // Fill in the receiver field and increment the count. 1665 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1666 set_mdp_data_at(recvr_offset, receiver); 1667 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1668 li(scratch1, DataLayout::counter_increment); 1669 set_mdp_data_at(count_offset, scratch1); 1670 if (start_row > 0) { 1671 b(done); 1672 } 1673} 1674 1675// Argument and return type profilig. 1676// kills: tmp, tmp2, R0, CR0, CR1 1677void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base, 1678 RegisterOrConstant mdo_addr_offs, Register tmp, Register tmp2) { 1679 Label do_nothing, do_update; 1680 1681 // tmp2 = obj is allowed 1682 assert_different_registers(obj, mdo_addr_base, tmp, R0); 1683 assert_different_registers(tmp2, mdo_addr_base, tmp, R0); 1684 const Register klass = tmp2; 1685 1686 verify_oop(obj); 1687 1688 ld(tmp, mdo_addr_offs, mdo_addr_base); 1689 1690 // Set null_seen if obj is 0. 1691 cmpdi(CCR0, obj, 0); 1692 ori(R0, tmp, TypeEntries::null_seen); 1693 beq(CCR0, do_update); 1694 1695 load_klass(klass, obj); 1696 1697 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask)); 1698 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask); 1699 cmpd(CCR1, R0, klass); 1700 // Klass seen before, nothing to do (regardless of unknown bit). 1701 //beq(CCR1, do_nothing); 1702 1703 andi_(R0, klass, TypeEntries::type_unknown); 1704 // Already unknown. Nothing to do anymore. 1705 //bne(CCR0, do_nothing); 1706 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne 1707 beq(CCR0, do_nothing); 1708 1709 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask)); 1710 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1711 beq(CCR0, do_update); // First time here. Set profile type. 1712 1713 // Different than before. Cannot keep accurate profile. 1714 ori(R0, tmp, TypeEntries::type_unknown); 1715 1716 bind(do_update); 1717 // update profile 1718 std(R0, mdo_addr_offs, mdo_addr_base); 1719 1720 align(32, 12); 1721 bind(do_nothing); 1722} 1723 1724void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) { 1725 if (!ProfileInterpreter) { 1726 return; 1727 } 1728 1729 assert_different_registers(callee, tmp1, tmp2, R28_mdx); 1730 1731 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1732 Label profile_continue; 1733 1734 test_method_data_pointer(profile_continue); 1735 1736 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1737 1738 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx); 1739 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1740 bne(CCR0, profile_continue); 1741 1742 if (MethodData::profile_arguments()) { 1743 Label done; 1744 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1745 add(R28_mdx, off_to_args, R28_mdx); 1746 1747 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1748 if (i > 0 || MethodData::profile_return()) { 1749 // If return value type is profiled we may have no argument to profile. 1750 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1751 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count()); 1752 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count()); 1753 blt(CCR0, done); 1754 } 1755 ld(tmp1, in_bytes(Method::const_offset()), callee); 1756 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1); 1757 // Stack offset o (zero based) from the start of the argument 1758 // list, for n arguments translates into offset n - o - 1 from 1759 // the end of the argument list. But there's an extra slot at 1760 // the top of the stack. So the offset is n - o from Lesp. 1761 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx); 1762 subf(tmp1, tmp2, tmp1); 1763 1764 sldi(tmp1, tmp1, Interpreter::logStackElementSize); 1765 ldx(tmp1, tmp1, R15_esp); 1766 1767 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1); 1768 1769 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1770 addi(R28_mdx, R28_mdx, to_add); 1771 off_to_args += to_add; 1772 } 1773 1774 if (MethodData::profile_return()) { 1775 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1776 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1777 } 1778 1779 bind(done); 1780 1781 if (MethodData::profile_return()) { 1782 // We're right after the type profile for the last 1783 // argument. tmp1 is the number of cells left in the 1784 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1785 // if there's a return to profile. 1786 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 1787 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size)); 1788 add(R28_mdx, tmp1, R28_mdx); 1789 } 1790 } else { 1791 assert(MethodData::profile_return(), "either profile call args or call ret"); 1792 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 1793 } 1794 1795 // Mdp points right after the end of the 1796 // CallTypeData/VirtualCallTypeData, right after the cells for the 1797 // return value type if there's one. 1798 align(32, 12); 1799 bind(profile_continue); 1800 } 1801} 1802 1803void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 1804 assert_different_registers(ret, tmp1, tmp2); 1805 if (ProfileInterpreter && MethodData::profile_return()) { 1806 Label profile_continue; 1807 1808 test_method_data_pointer(profile_continue); 1809 1810 if (MethodData::profile_return_jsr292_only()) { 1811 // If we don't profile all invoke bytecodes we must make sure 1812 // it's a bytecode we indeed profile. We can't go back to the 1813 // begining of the ProfileData we intend to update to check its 1814 // type because we're right after it and we don't known its 1815 // length. 1816 lbz(tmp1, 0, R14_bcp); 1817 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method); 1818 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic); 1819 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle); 1820 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1821 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm); 1822 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1823 bne(CCR0, profile_continue); 1824 } 1825 1826 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2); 1827 1828 align(32, 12); 1829 bind(profile_continue); 1830 } 1831} 1832 1833void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) { 1834 if (ProfileInterpreter && MethodData::profile_parameters()) { 1835 Label profile_continue, done; 1836 1837 test_method_data_pointer(profile_continue); 1838 1839 // Load the offset of the area within the MDO used for 1840 // parameters. If it's negative we're not profiling any parameters. 1841 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx); 1842 cmpwi(CCR0, tmp1, 0); 1843 blt(CCR0, profile_continue); 1844 1845 // Compute a pointer to the area for parameters from the offset 1846 // and move the pointer to the slot for the last 1847 // parameters. Collect profiling from last parameter down. 1848 // mdo start + parameters offset + array length - 1 1849 1850 // Pointer to the parameter area in the MDO. 1851 const Register mdp = tmp1; 1852 add(mdp, tmp1, R28_mdx); 1853 1854 // Offset of the current profile entry to update. 1855 const Register entry_offset = tmp2; 1856 // entry_offset = array len in number of cells 1857 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp); 1858 1859 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1860 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 1861 1862 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 1863 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size)); 1864 // entry_offset in bytes 1865 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1866 1867 Label loop; 1868 align(32, 12); 1869 bind(loop); 1870 1871 // Load offset on the stack from the slot for this parameter. 1872 ld(tmp3, entry_offset, mdp); 1873 sldi(tmp3, tmp3, Interpreter::logStackElementSize); 1874 neg(tmp3, tmp3); 1875 // Read the parameter from the local area. 1876 ldx(tmp3, tmp3, R18_locals); 1877 1878 // Make entry_offset now point to the type field for this parameter. 1879 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1880 assert(type_base > off_base, "unexpected"); 1881 addi(entry_offset, entry_offset, type_base - off_base); 1882 1883 // Profile the parameter. 1884 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3); 1885 1886 // Go to next parameter. 1887 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base); 1888 cmpdi(CCR0, entry_offset, off_base + delta); 1889 addi(entry_offset, entry_offset, -delta); 1890 bge(CCR0, loop); 1891 1892 align(32, 12); 1893 bind(profile_continue); 1894 } 1895} 1896 1897// Add a InterpMonitorElem to stack (see frame_sparc.hpp). 1898void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) { 1899 1900 // Very-local scratch registers. 1901 const Register esp = Rtemp1; 1902 const Register slot = Rtemp2; 1903 1904 // Extracted monitor_size. 1905 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes(); 1906 assert(Assembler::is_aligned((unsigned int)monitor_size, 1907 (unsigned int)frame::alignment_in_bytes), 1908 "size of a monitor must respect alignment of SP"); 1909 1910 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor 1911 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp 1912 1913 // Shuffle expression stack down. Recall that stack_base points 1914 // just above the new expression stack bottom. Old_tos and new_tos 1915 // are used to scan thru the old and new expression stacks. 1916 if (!stack_is_empty) { 1917 Label copy_slot, copy_slot_finished; 1918 const Register n_slots = slot; 1919 1920 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack). 1921 subf(n_slots, esp, R26_monitor); 1922 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy. 1923 assert(LogBytesPerWord == 3, "conflicts assembler instructions"); 1924 beq(CCR0, copy_slot_finished); // Nothing to copy. 1925 1926 mtctr(n_slots); 1927 1928 // loop 1929 bind(copy_slot); 1930 ld(slot, 0, esp); // Move expression stack down. 1931 std(slot, -monitor_size, esp); // distance = monitor_size 1932 addi(esp, esp, BytesPerWord); 1933 bdnz(copy_slot); 1934 1935 bind(copy_slot_finished); 1936 } 1937 1938 addi(R15_esp, R15_esp, -monitor_size); 1939 addi(R26_monitor, R26_monitor, -monitor_size); 1940 1941 // Restart interpreter 1942} 1943 1944// ============================================================================ 1945// Java locals access 1946 1947// Load a local variable at index in Rindex into register Rdst_value. 1948// Also puts address of local into Rdst_address as a service. 1949// Kills: 1950// - Rdst_value 1951// - Rdst_address 1952void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) { 1953 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 1954 subf(Rdst_address, Rdst_address, R18_locals); 1955 lwz(Rdst_value, 0, Rdst_address); 1956} 1957 1958// Load a local variable at index in Rindex into register Rdst_value. 1959// Also puts address of local into Rdst_address as a service. 1960// Kills: 1961// - Rdst_value 1962// - Rdst_address 1963void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) { 1964 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 1965 subf(Rdst_address, Rdst_address, R18_locals); 1966 ld(Rdst_value, -8, Rdst_address); 1967} 1968 1969// Load a local variable at index in Rindex into register Rdst_value. 1970// Also puts address of local into Rdst_address as a service. 1971// Input: 1972// - Rindex: slot nr of local variable 1973// Kills: 1974// - Rdst_value 1975// - Rdst_address 1976void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, Register Rdst_address, Register Rindex) { 1977 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 1978 subf(Rdst_address, Rdst_address, R18_locals); 1979 ld(Rdst_value, 0, Rdst_address); 1980} 1981 1982// Load a local variable at index in Rindex into register Rdst_value. 1983// Also puts address of local into Rdst_address as a service. 1984// Kills: 1985// - Rdst_value 1986// - Rdst_address 1987void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) { 1988 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 1989 subf(Rdst_address, Rdst_address, R18_locals); 1990 lfs(Rdst_value, 0, Rdst_address); 1991} 1992 1993// Load a local variable at index in Rindex into register Rdst_value. 1994// Also puts address of local into Rdst_address as a service. 1995// Kills: 1996// - Rdst_value 1997// - Rdst_address 1998void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) { 1999 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2000 subf(Rdst_address, Rdst_address, R18_locals); 2001 lfd(Rdst_value, -8, Rdst_address); 2002} 2003 2004// Store an int value at local variable slot Rindex. 2005// Kills: 2006// - Rindex 2007void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) { 2008 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2009 subf(Rindex, Rindex, R18_locals); 2010 stw(Rvalue, 0, Rindex); 2011} 2012 2013// Store a long value at local variable slot Rindex. 2014// Kills: 2015// - Rindex 2016void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) { 2017 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2018 subf(Rindex, Rindex, R18_locals); 2019 std(Rvalue, -8, Rindex); 2020} 2021 2022// Store an oop value at local variable slot Rindex. 2023// Kills: 2024// - Rindex 2025void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) { 2026 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2027 subf(Rindex, Rindex, R18_locals); 2028 std(Rvalue, 0, Rindex); 2029} 2030 2031// Store an int value at local variable slot Rindex. 2032// Kills: 2033// - Rindex 2034void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) { 2035 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2036 subf(Rindex, Rindex, R18_locals); 2037 stfs(Rvalue, 0, Rindex); 2038} 2039 2040// Store an int value at local variable slot Rindex. 2041// Kills: 2042// - Rindex 2043void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) { 2044 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2045 subf(Rindex, Rindex, R18_locals); 2046 stfd(Rvalue, -8, Rindex); 2047} 2048 2049// Read pending exception from thread and jump to interpreter. 2050// Throw exception entry if one if pending. Fall through otherwise. 2051void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) { 2052 assert_different_registers(Rscratch1, Rscratch2, R3); 2053 Register Rexception = Rscratch1; 2054 Register Rtmp = Rscratch2; 2055 Label Ldone; 2056 // Get pending exception oop. 2057 ld(Rexception, thread_(pending_exception)); 2058 cmpdi(CCR0, Rexception, 0); 2059 beq(CCR0, Ldone); 2060 li(Rtmp, 0); 2061 mr_if_needed(R3, Rexception); 2062 std(Rtmp, thread_(pending_exception)); // Clear exception in thread 2063 if (Interpreter::rethrow_exception_entry() != NULL) { 2064 // Already got entry address. 2065 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry()); 2066 } else { 2067 // Dynamically load entry address. 2068 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true); 2069 ld(Rtmp, simm16_rest, Rtmp); 2070 } 2071 mtctr(Rtmp); 2072 save_interpreter_state(Rtmp); 2073 bctr(); 2074 2075 align(32, 12); 2076 bind(Ldone); 2077} 2078 2079void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) { 2080 save_interpreter_state(R11_scratch1); 2081 2082 MacroAssembler::call_VM(oop_result, entry_point, false); 2083 2084 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2085 2086 check_and_handle_popframe(R11_scratch1); 2087 check_and_handle_earlyret(R11_scratch1); 2088 // Now check exceptions manually. 2089 if (check_exceptions) { 2090 check_and_forward_exception(R11_scratch1, R12_scratch2); 2091 } 2092} 2093 2094void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions) { 2095 // ARG1 is reserved for the thread. 2096 mr_if_needed(R4_ARG2, arg_1); 2097 call_VM(oop_result, entry_point, check_exceptions); 2098} 2099 2100void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) { 2101 // ARG1 is reserved for the thread. 2102 mr_if_needed(R4_ARG2, arg_1); 2103 assert(arg_2 != R4_ARG2, "smashed argument"); 2104 mr_if_needed(R5_ARG3, arg_2); 2105 call_VM(oop_result, entry_point, check_exceptions); 2106} 2107 2108void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) { 2109 // ARG1 is reserved for the thread. 2110 mr_if_needed(R4_ARG2, arg_1); 2111 assert(arg_2 != R4_ARG2, "smashed argument"); 2112 mr_if_needed(R5_ARG3, arg_2); 2113 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument"); 2114 mr_if_needed(R6_ARG4, arg_3); 2115 call_VM(oop_result, entry_point, check_exceptions); 2116} 2117 2118void InterpreterMacroAssembler::save_interpreter_state(Register scratch) { 2119 ld(scratch, 0, R1_SP); 2120 std(R15_esp, _ijava_state_neg(esp), scratch); 2121 std(R14_bcp, _ijava_state_neg(bcp), scratch); 2122 std(R26_monitor, _ijava_state_neg(monitors), scratch); 2123 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); } 2124 // Other entries should be unchanged. 2125} 2126 2127void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) { 2128 ld(scratch, 0, R1_SP); 2129 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception). 2130 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code. 2131 if (!bcp_and_mdx_only) { 2132 // Following ones are Metadata. 2133 ld(R19_method, _ijava_state_neg(method), scratch); 2134 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch); 2135 // Following ones are stack addresses and don't require reload. 2136 ld(R15_esp, _ijava_state_neg(esp), scratch); 2137 ld(R18_locals, _ijava_state_neg(locals), scratch); 2138 ld(R26_monitor, _ijava_state_neg(monitors), scratch); 2139 } 2140#ifdef ASSERT 2141 { 2142 Label Lok; 2143 subf(R0, R1_SP, scratch); 2144 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size); 2145 bge(CCR0, Lok); 2146 stop("frame too small (restore istate)", 0x5432); 2147 bind(Lok); 2148 } 2149 { 2150 Label Lok; 2151 ld(R0, _ijava_state_neg(ijava_reserved), scratch); 2152 cmpdi(CCR0, R0, 0x5afe); 2153 beq(CCR0, Lok); 2154 stop("frame corrupted (restore istate)", 0x5afe); 2155 bind(Lok); 2156 } 2157#endif 2158} 2159 2160#endif // !CC_INTERP 2161 2162void InterpreterMacroAssembler::get_method_counters(Register method, 2163 Register Rcounters, 2164 Label& skip) { 2165 BLOCK_COMMENT("Load and ev. allocate counter object {"); 2166 Label has_counters; 2167 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2168 cmpdi(CCR0, Rcounters, 0); 2169 bne(CCR0, has_counters); 2170 call_VM(noreg, CAST_FROM_FN_PTR(address, 2171 InterpreterRuntime::build_method_counters), method, false); 2172 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2173 cmpdi(CCR0, Rcounters, 0); 2174 beq(CCR0, skip); // No MethodCounters, OutOfMemory. 2175 BLOCK_COMMENT("} Load and ev. allocate counter object"); 2176 2177 bind(has_counters); 2178} 2179 2180void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register iv_be_count, Register Rtmp_r0) { 2181 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 2182 Register invocation_count = iv_be_count; 2183 Register backedge_count = Rtmp_r0; 2184 int delta = InvocationCounter::count_increment; 2185 2186 // Load each counter in a register. 2187 // ld(inv_counter, Rtmp); 2188 // ld(be_counter, Rtmp2); 2189 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + 2190 InvocationCounter::counter_offset()); 2191 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + 2192 InvocationCounter::counter_offset()); 2193 2194 BLOCK_COMMENT("Increment profiling counters {"); 2195 2196 // Load the backedge counter. 2197 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int 2198 // Mask the backedge counter. 2199 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value); 2200 2201 // Load the invocation counter. 2202 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int 2203 // Add the delta to the invocation counter and store the result. 2204 addi(invocation_count, invocation_count, delta); 2205 // Store value. 2206 stw(invocation_count, inv_counter_offset, Rcounters); 2207 2208 // Add invocation counter + backedge counter. 2209 add(iv_be_count, backedge_count, invocation_count); 2210 2211 // Note that this macro must leave the backedge_count + invocation_count in 2212 // register iv_be_count! 2213 BLOCK_COMMENT("} Increment profiling counters"); 2214} 2215 2216void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2217 if (state == atos) { MacroAssembler::verify_oop(reg); } 2218} 2219 2220#ifndef CC_INTERP 2221// Local helper function for the verify_oop_or_return_address macro. 2222static bool verify_return_address(Method* m, int bci) { 2223#ifndef PRODUCT 2224 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci; 2225 // Assume it is a valid return address if it is inside m and is preceded by a jsr. 2226 if (!m->contains(pc)) return false; 2227 address jsr_pc; 2228 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2229 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2230 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2231 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2232#endif // PRODUCT 2233 return false; 2234} 2235 2236void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2237 if (VerifyFPU) { 2238 unimplemented("verfiyFPU"); 2239 } 2240} 2241 2242void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2243 if (!VerifyOops) return; 2244 2245 // The VM documentation for the astore[_wide] bytecode allows 2246 // the TOS to be not only an oop but also a return address. 2247 Label test; 2248 Label skip; 2249 // See if it is an address (in the current method): 2250 2251 const int log2_bytecode_size_limit = 16; 2252 srdi_(Rtmp, reg, log2_bytecode_size_limit); 2253 bne(CCR0, test); 2254 2255 address fd = CAST_FROM_FN_PTR(address, verify_return_address); 2256 const int nbytes_save = MacroAssembler::num_volatile_regs * 8; 2257 save_volatile_gprs(R1_SP, -nbytes_save); // except R0 2258 save_LR_CR(Rtmp); // Save in old frame. 2259 push_frame_reg_args(nbytes_save, Rtmp); 2260 2261 load_const_optimized(Rtmp, fd, R0); 2262 mr_if_needed(R4_ARG2, reg); 2263 mr(R3_ARG1, R19_method); 2264 call_c(Rtmp); // call C 2265 2266 pop_frame(); 2267 restore_LR_CR(Rtmp); 2268 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0 2269 b(skip); 2270 2271 // Perform a more elaborate out-of-line call. 2272 // Not an address; verify it: 2273 bind(test); 2274 verify_oop(reg); 2275 bind(skip); 2276} 2277#endif // !CC_INTERP 2278 2279// Inline assembly for: 2280// 2281// if (thread is in interp_only_mode) { 2282// InterpreterRuntime::post_method_entry(); 2283// } 2284// if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) || 2285// *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) { 2286// SharedRuntime::jvmpi_method_entry(method, receiver); 2287// } 2288void InterpreterMacroAssembler::notify_method_entry() { 2289 // JVMTI 2290 // Whenever JVMTI puts a thread in interp_only_mode, method 2291 // entry/exit events are sent for that thread to track stack 2292 // depth. If it is possible to enter interp_only_mode we add 2293 // the code to check if the event should be sent. 2294 if (JvmtiExport::can_post_interpreter_events()) { 2295 Label jvmti_post_done; 2296 2297 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2298 cmpwi(CCR0, R0, 0); 2299 beq(CCR0, jvmti_post_done); 2300 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), 2301 /*check_exceptions=*/true CC_INTERP_ONLY(&& false)); 2302 2303 bind(jvmti_post_done); 2304 } 2305} 2306 2307// Inline assembly for: 2308// 2309// if (thread is in interp_only_mode) { 2310// // save result 2311// InterpreterRuntime::post_method_exit(); 2312// // restore result 2313// } 2314// if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) { 2315// // save result 2316// SharedRuntime::jvmpi_method_exit(); 2317// // restore result 2318// } 2319// 2320// Native methods have their result stored in d_tmp and l_tmp. 2321// Java methods have their result stored in the expression stack. 2322void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state, 2323 NotifyMethodExitMode mode, bool check_exceptions) { 2324 // JVMTI 2325 // Whenever JVMTI puts a thread in interp_only_mode, method 2326 // entry/exit events are sent for that thread to track stack 2327 // depth. If it is possible to enter interp_only_mode we add 2328 // the code to check if the event should be sent. 2329 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2330 Label jvmti_post_done; 2331 2332 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2333 cmpwi(CCR0, R0, 0); 2334 beq(CCR0, jvmti_post_done); 2335 CC_INTERP_ONLY(assert(is_native_method && !check_exceptions, "must not push state")); 2336 if (!is_native_method) push(state); // Expose tos to GC. 2337 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), 2338 /*check_exceptions=*/check_exceptions); 2339 if (!is_native_method) pop(state); 2340 2341 align(32, 12); 2342 bind(jvmti_post_done); 2343 } 2344 2345 // Dtrace support not implemented. 2346} 2347 2348#ifdef CC_INTERP 2349// Convert the current TOP_IJAVA_FRAME into a PARENT_IJAVA_FRAME 2350// (using parent_frame_resize) and push a new interpreter 2351// TOP_IJAVA_FRAME (using frame_size). 2352void InterpreterMacroAssembler::push_interpreter_frame(Register top_frame_size, Register parent_frame_resize, 2353 Register tmp1, Register tmp2, Register tmp3, 2354 Register tmp4, Register pc) { 2355 assert_different_registers(top_frame_size, parent_frame_resize, tmp1, tmp2, tmp3, tmp4); 2356 ld(tmp1, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2357 mr(tmp2/*top_frame_sp*/, R1_SP); 2358 // Move initial_caller_sp. 2359 ld(tmp4, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2360 neg(parent_frame_resize, parent_frame_resize); 2361 resize_frame(parent_frame_resize/*-parent_frame_resize*/, tmp3); 2362 2363 // Set LR in new parent frame. 2364 std(tmp1, _abi(lr), R1_SP); 2365 // Set top_frame_sp info for new parent frame. 2366 std(tmp2, _parent_ijava_frame_abi(top_frame_sp), R1_SP); 2367 std(tmp4, _parent_ijava_frame_abi(initial_caller_sp), R1_SP); 2368 2369 // Push new TOP_IJAVA_FRAME. 2370 push_frame(top_frame_size, tmp2); 2371 2372 get_PC_trash_LR(tmp3); 2373 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2374 // Used for non-initial callers by unextended_sp(). 2375 std(R1_SP, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2376} 2377 2378// Pop the topmost TOP_IJAVA_FRAME and convert the previous 2379// PARENT_IJAVA_FRAME back into a TOP_IJAVA_FRAME. 2380void InterpreterMacroAssembler::pop_interpreter_frame(Register tmp1, Register tmp2, Register tmp3, Register tmp4) { 2381 assert_different_registers(tmp1, tmp2, tmp3, tmp4); 2382 2383 ld(tmp1/*caller's sp*/, _abi(callers_sp), R1_SP); 2384 ld(tmp3, _abi(lr), tmp1); 2385 2386 ld(tmp4, _parent_ijava_frame_abi(initial_caller_sp), tmp1); 2387 2388 ld(tmp2/*caller's caller's sp*/, _abi(callers_sp), tmp1); 2389 // Merge top frame. 2390 std(tmp2, _abi(callers_sp), R1_SP); 2391 2392 ld(tmp2, _parent_ijava_frame_abi(top_frame_sp), tmp1); 2393 2394 // Update C stack pointer to caller's top_abi. 2395 resize_frame_absolute(tmp2/*addr*/, tmp1/*tmp*/, tmp2/*tmp*/); 2396 2397 // Update LR in top_frame. 2398 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2399 2400 std(tmp4, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2401 2402 // Store the top-frame stack-pointer for c2i adapters. 2403 std(R1_SP, _top_ijava_frame_abi(top_frame_sp), R1_SP); 2404} 2405 2406// Turn state's interpreter frame into the current TOP_IJAVA_FRAME. 2407void InterpreterMacroAssembler::pop_interpreter_frame_to_state(Register state, Register tmp1, Register tmp2, Register tmp3) { 2408 assert_different_registers(R14_state, R15_prev_state, tmp1, tmp2, tmp3); 2409 2410 if (state == R14_state) { 2411 ld(tmp1/*state's fp*/, state_(_last_Java_fp)); 2412 ld(tmp2/*state's sp*/, state_(_last_Java_sp)); 2413 } else if (state == R15_prev_state) { 2414 ld(tmp1/*state's fp*/, prev_state_(_last_Java_fp)); 2415 ld(tmp2/*state's sp*/, prev_state_(_last_Java_sp)); 2416 } else { 2417 ShouldNotReachHere(); 2418 } 2419 2420 // Merge top frames. 2421 std(tmp1, _abi(callers_sp), R1_SP); 2422 2423 // Tmp2 is new SP. 2424 // Tmp1 is parent's SP. 2425 resize_frame_absolute(tmp2/*addr*/, tmp1/*tmp*/, tmp2/*tmp*/); 2426 2427 // Update LR in top_frame. 2428 // Must be interpreter frame. 2429 get_PC_trash_LR(tmp3); 2430 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 2431 // Used for non-initial callers by unextended_sp(). 2432 std(R1_SP, _top_ijava_frame_abi(initial_caller_sp), R1_SP); 2433} 2434 2435// Set SP to initial caller's sp, but before fix the back chain. 2436void InterpreterMacroAssembler::resize_frame_to_initial_caller(Register tmp1, Register tmp2) { 2437 ld(tmp1, _parent_ijava_frame_abi(initial_caller_sp), R1_SP); 2438 ld(tmp2, _parent_ijava_frame_abi(callers_sp), R1_SP); 2439 std(tmp2, _parent_ijava_frame_abi(callers_sp), tmp1); // Fix back chain ... 2440 mr(R1_SP, tmp1); // ... and resize to initial caller. 2441} 2442 2443// Pop the current interpreter state (without popping the correspoding 2444// frame) and restore R14_state and R15_prev_state accordingly. 2445// Use prev_state_may_be_0 to indicate whether prev_state may be 0 2446// in order to generate an extra check before retrieving prev_state_(_prev_link). 2447void InterpreterMacroAssembler::pop_interpreter_state(bool prev_state_may_be_0) 2448{ 2449 // Move prev_state to state and restore prev_state from state_(_prev_link). 2450 Label prev_state_is_0; 2451 mr(R14_state, R15_prev_state); 2452 2453 // Don't retrieve /*state==*/prev_state_(_prev_link) 2454 // if /*state==*/prev_state is 0. 2455 if (prev_state_may_be_0) { 2456 cmpdi(CCR0, R15_prev_state, 0); 2457 beq(CCR0, prev_state_is_0); 2458 } 2459 2460 ld(R15_prev_state, /*state==*/prev_state_(_prev_link)); 2461 bind(prev_state_is_0); 2462} 2463 2464void InterpreterMacroAssembler::restore_prev_state() { 2465 // _prev_link is private, but cInterpreter is a friend. 2466 ld(R15_prev_state, state_(_prev_link)); 2467} 2468#endif // CC_INTERP 2469