1 /* 2 * Copyright (c) 2003, 2017, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2012, 2017 SAP SE. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 27 #include "precompiled.hpp" 28 #include "asm/macroAssembler.inline.hpp" 29 #include "interp_masm_ppc.hpp" 30 #include "interpreter/interpreterRuntime.hpp" 31 #include "prims/jvmtiThreadState.hpp" 32 #include "runtime/sharedRuntime.hpp" 33 34 #ifdef PRODUCT 35 #define BLOCK_COMMENT(str) // nothing 36 #else 37 #define BLOCK_COMMENT(str) block_comment(str) 38 #endif 39 40 void InterpreterMacroAssembler::null_check_throw(Register a, int offset, Register temp_reg) { 41 address exception_entry = Interpreter::throw_NullPointerException_entry(); 42 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry); 43 } 44 45 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) { 46 assert(entry, "Entry must have been generated by now"); 47 if (is_within_range_of_b(entry, pc())) { 48 b(entry); 49 } else { 50 load_const_optimized(Rscratch, entry, R0); 51 mtctr(Rscratch); 52 bctr(); 53 } 54 } 55 56 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { 57 Register bytecode = R12_scratch2; 58 if (bcp_incr != 0) { 59 lbzu(bytecode, bcp_incr, R14_bcp); 60 } else { 61 lbz(bytecode, 0, R14_bcp); 62 } 63 64 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state)); 65 } 66 67 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 68 // Load current bytecode. 69 Register bytecode = R12_scratch2; 70 lbz(bytecode, 0, R14_bcp); 71 dispatch_Lbyte_code(state, bytecode, table); 72 } 73 74 // Dispatch code executed in the prolog of a bytecode which does not do it's 75 // own dispatch. The dispatch address is computed and placed in R24_dispatch_addr. 76 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 77 Register bytecode = R12_scratch2; 78 lbz(bytecode, bcp_incr, R14_bcp); 79 80 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state)); 81 82 sldi(bytecode, bytecode, LogBytesPerWord); 83 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode); 84 } 85 86 // Dispatch code executed in the epilog of a bytecode which does not do it's 87 // own dispatch. The dispatch address in R24_dispatch_addr is used for the 88 // dispatch. 89 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { 90 if (bcp_incr) { addi(R14_bcp, R14_bcp, bcp_incr); } 91 mtctr(R24_dispatch_addr); 92 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable); 93 } 94 95 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 96 assert(scratch_reg != R0, "can't use R0 as scratch_reg here"); 97 if (JvmtiExport::can_pop_frame()) { 98 Label L; 99 100 // Check the "pending popframe condition" flag in the current thread. 101 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 102 103 // Initiate popframe handling only if it is not already being 104 // processed. If the flag has the popframe_processing bit set, it 105 // means that this code is called *during* popframe handling - we 106 // don't want to reenter. 107 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit); 108 beq(CCR0, L); 109 110 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit); 111 bne(CCR0, L); 112 113 // Call the Interpreter::remove_activation_preserving_args_entry() 114 // func to get the address of the same-named entrypoint in the 115 // generated interpreter code. 116 #if defined(ABI_ELFv2) 117 call_c(CAST_FROM_FN_PTR(address, 118 Interpreter::remove_activation_preserving_args_entry), 119 relocInfo::none); 120 #else 121 call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, 122 Interpreter::remove_activation_preserving_args_entry), 123 relocInfo::none); 124 #endif 125 126 // Jump to Interpreter::_remove_activation_preserving_args_entry. 127 mtctr(R3_RET); 128 bctr(); 129 130 align(32, 12); 131 bind(L); 132 } 133 } 134 135 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 136 const Register Rthr_state_addr = scratch_reg; 137 if (JvmtiExport::can_force_early_return()) { 138 Label Lno_early_ret; 139 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 140 cmpdi(CCR0, Rthr_state_addr, 0); 141 beq(CCR0, Lno_early_ret); 142 143 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr); 144 cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending); 145 bne(CCR0, Lno_early_ret); 146 147 // Jump to Interpreter::_earlyret_entry. 148 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr); 149 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry)); 150 mtlr(R3_RET); 151 blr(); 152 153 align(32, 12); 154 bind(Lno_early_ret); 155 } 156 } 157 158 void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) { 159 const Register RjvmtiState = Rscratch1; 160 const Register Rscratch2 = R0; 161 162 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 163 li(Rscratch2, 0); 164 165 switch (state) { 166 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 167 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); 168 break; 169 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 170 break; 171 case btos: // fall through 172 case ztos: // fall through 173 case ctos: // fall through 174 case stos: // fall through 175 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 176 break; 177 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 178 break; 179 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 180 break; 181 case vtos: break; 182 default : ShouldNotReachHere(); 183 } 184 185 // Clean up tos value in the jvmti thread state. 186 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); 187 // Set tos state field to illegal value. 188 li(Rscratch2, ilgl); 189 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState); 190 } 191 192 // Common code to dispatch and dispatch_only. 193 // Dispatch value in Lbyte_code and increment Lbcp. 194 195 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) { 196 address table_base = (address)Interpreter::dispatch_table((TosState)0); 197 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base; 198 if (is_simm16(table_offs)) { 199 addi(dst, R25_templateTableBase, (int)table_offs); 200 } else { 201 load_const_optimized(dst, table, R0); 202 } 203 } 204 205 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode, 206 address* table, bool verify) { 207 if (verify) { 208 unimplemented("dispatch_Lbyte_code: verify"); // See Sparc Implementation to implement this 209 } 210 211 assert_different_registers(bytecode, R11_scratch1); 212 213 // Calc dispatch table address. 214 load_dispatch_table(R11_scratch1, table); 215 216 sldi(R12_scratch2, bytecode, LogBytesPerWord); 217 ldx(R11_scratch1, R11_scratch1, R12_scratch2); 218 219 // Jump off! 220 mtctr(R11_scratch1); 221 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable); 222 } 223 224 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) { 225 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize); 226 ldx(Rrecv_dst, Rrecv_dst, R15_esp); 227 } 228 229 // helpers for expression stack 230 231 void InterpreterMacroAssembler::pop_i(Register r) { 232 lwzu(r, Interpreter::stackElementSize, R15_esp); 233 } 234 235 void InterpreterMacroAssembler::pop_ptr(Register r) { 236 ldu(r, Interpreter::stackElementSize, R15_esp); 237 } 238 239 void InterpreterMacroAssembler::pop_l(Register r) { 240 ld(r, Interpreter::stackElementSize, R15_esp); 241 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 242 } 243 244 void InterpreterMacroAssembler::pop_f(FloatRegister f) { 245 lfsu(f, Interpreter::stackElementSize, R15_esp); 246 } 247 248 void InterpreterMacroAssembler::pop_d(FloatRegister f) { 249 lfd(f, Interpreter::stackElementSize, R15_esp); 250 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); 251 } 252 253 void InterpreterMacroAssembler::push_i(Register r) { 254 stw(r, 0, R15_esp); 255 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 256 } 257 258 void InterpreterMacroAssembler::push_ptr(Register r) { 259 std(r, 0, R15_esp); 260 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 261 } 262 263 void InterpreterMacroAssembler::push_l(Register r) { 264 // Clear unused slot. 265 load_const_optimized(R0, 0L); 266 std(R0, 0, R15_esp); 267 std(r, - Interpreter::stackElementSize, R15_esp); 268 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 269 } 270 271 void InterpreterMacroAssembler::push_f(FloatRegister f) { 272 stfs(f, 0, R15_esp); 273 addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); 274 } 275 276 void InterpreterMacroAssembler::push_d(FloatRegister f) { 277 stfd(f, - Interpreter::stackElementSize, R15_esp); 278 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 279 } 280 281 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) { 282 std(first, 0, R15_esp); 283 std(second, -Interpreter::stackElementSize, R15_esp); 284 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); 285 } 286 287 void InterpreterMacroAssembler::move_l_to_d(Register l, FloatRegister d) { 288 if (VM_Version::has_mtfprd()) { 289 mtfprd(d, l); 290 } else { 291 std(l, 0, R15_esp); 292 lfd(d, 0, R15_esp); 293 } 294 } 295 296 void InterpreterMacroAssembler::move_d_to_l(FloatRegister d, Register l) { 297 if (VM_Version::has_mtfprd()) { 298 mffprd(l, d); 299 } else { 300 stfd(d, 0, R15_esp); 301 ld(l, 0, R15_esp); 302 } 303 } 304 305 void InterpreterMacroAssembler::push(TosState state) { 306 switch (state) { 307 case atos: push_ptr(); break; 308 case btos: 309 case ztos: 310 case ctos: 311 case stos: 312 case itos: push_i(); break; 313 case ltos: push_l(); break; 314 case ftos: push_f(); break; 315 case dtos: push_d(); break; 316 case vtos: /* nothing to do */ break; 317 default : ShouldNotReachHere(); 318 } 319 } 320 321 void InterpreterMacroAssembler::pop(TosState state) { 322 switch (state) { 323 case atos: pop_ptr(); break; 324 case btos: 325 case ztos: 326 case ctos: 327 case stos: 328 case itos: pop_i(); break; 329 case ltos: pop_l(); break; 330 case ftos: pop_f(); break; 331 case dtos: pop_d(); break; 332 case vtos: /* nothing to do */ break; 333 default : ShouldNotReachHere(); 334 } 335 verify_oop(R17_tos, state); 336 } 337 338 void InterpreterMacroAssembler::empty_expression_stack() { 339 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 340 } 341 342 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset, 343 Register Rdst, 344 signedOrNot is_signed) { 345 #if defined(VM_LITTLE_ENDIAN) 346 if (bcp_offset) { 347 load_const_optimized(Rdst, bcp_offset); 348 lhbrx(Rdst, R14_bcp, Rdst); 349 } else { 350 lhbrx(Rdst, R14_bcp); 351 } 352 if (is_signed == Signed) { 353 extsh(Rdst, Rdst); 354 } 355 #else 356 // Read Java big endian format. 357 if (is_signed == Signed) { 358 lha(Rdst, bcp_offset, R14_bcp); 359 } else { 360 lhz(Rdst, bcp_offset, R14_bcp); 361 } 362 #endif 363 } 364 365 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset, 366 Register Rdst, 367 signedOrNot is_signed) { 368 #if defined(VM_LITTLE_ENDIAN) 369 if (bcp_offset) { 370 load_const_optimized(Rdst, bcp_offset); 371 lwbrx(Rdst, R14_bcp, Rdst); 372 } else { 373 lwbrx(Rdst, R14_bcp); 374 } 375 if (is_signed == Signed) { 376 extsw(Rdst, Rdst); 377 } 378 #else 379 // Read Java big endian format. 380 if (bcp_offset & 3) { // Offset unaligned? 381 load_const_optimized(Rdst, bcp_offset); 382 if (is_signed == Signed) { 383 lwax(Rdst, R14_bcp, Rdst); 384 } else { 385 lwzx(Rdst, R14_bcp, Rdst); 386 } 387 } else { 388 if (is_signed == Signed) { 389 lwa(Rdst, bcp_offset, R14_bcp); 390 } else { 391 lwz(Rdst, bcp_offset, R14_bcp); 392 } 393 } 394 #endif 395 } 396 397 398 // Load the constant pool cache index from the bytecode stream. 399 // 400 // Kills / writes: 401 // - Rdst, Rscratch 402 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, 403 size_t index_size) { 404 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 405 // Cache index is always in the native format, courtesy of Rewriter. 406 if (index_size == sizeof(u2)) { 407 lhz(Rdst, bcp_offset, R14_bcp); 408 } else if (index_size == sizeof(u4)) { 409 if (bcp_offset & 3) { 410 load_const_optimized(Rdst, bcp_offset); 411 lwax(Rdst, R14_bcp, Rdst); 412 } else { 413 lwa(Rdst, bcp_offset, R14_bcp); 414 } 415 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 416 nand(Rdst, Rdst, Rdst); // convert to plain index 417 } else if (index_size == sizeof(u1)) { 418 lbz(Rdst, bcp_offset, R14_bcp); 419 } else { 420 ShouldNotReachHere(); 421 } 422 // Rdst now contains cp cache index. 423 } 424 425 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset, 426 size_t index_size) { 427 get_cache_index_at_bcp(cache, bcp_offset, index_size); 428 sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord)); 429 add(cache, R27_constPoolCache, cache); 430 } 431 432 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format) 433 // from (Rsrc)+offset. 434 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset, 435 signedOrNot is_signed) { 436 #if defined(VM_LITTLE_ENDIAN) 437 if (offset) { 438 load_const_optimized(Rdst, offset); 439 lwbrx(Rdst, Rdst, Rsrc); 440 } else { 441 lwbrx(Rdst, Rsrc); 442 } 443 if (is_signed == Signed) { 444 extsw(Rdst, Rdst); 445 } 446 #else 447 if (is_signed == Signed) { 448 lwa(Rdst, offset, Rsrc); 449 } else { 450 lwz(Rdst, offset, Rsrc); 451 } 452 #endif 453 } 454 455 // Load object from cpool->resolved_references(index). 456 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index, Label *is_null) { 457 assert_different_registers(result, index); 458 get_constant_pool(result); 459 460 // Convert from field index to resolved_references() index and from 461 // word index to byte offset. Since this is a java object, it can be compressed. 462 Register tmp = index; // reuse 463 sldi(tmp, index, LogBytesPerHeapOop); 464 // Load pointer for resolved_references[] objArray. 465 ld(result, ConstantPool::cache_offset_in_bytes(), result); 466 ld(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result); 467 resolve_oop_handle(result); 468 #ifdef ASSERT 469 Label index_ok; 470 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result); 471 sldi(R0, R0, LogBytesPerHeapOop); 472 cmpd(CCR0, tmp, R0); 473 blt(CCR0, index_ok); 474 stop("resolved reference index out of bounds", 0x09256); 475 bind(index_ok); 476 #endif 477 // Add in the index. 478 add(result, tmp, result); 479 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result, is_null); 480 } 481 482 // load cpool->resolved_klass_at(index) 483 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool, Register Roffset, Register Rklass) { 484 // int value = *(Rcpool->int_at_addr(which)); 485 // int resolved_klass_index = extract_low_short_from_int(value); 486 add(Roffset, Rcpool, Roffset); 487 #if defined(VM_LITTLE_ENDIAN) 488 lhz(Roffset, sizeof(ConstantPool), Roffset); // Roffset = resolved_klass_index 489 #else 490 lhz(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index 491 #endif 492 493 ld(Rklass, ConstantPool::resolved_klasses_offset_in_bytes(), Rcpool); // Rklass = Rcpool->_resolved_klasses 494 495 sldi(Roffset, Roffset, LogBytesPerWord); 496 addi(Roffset, Roffset, Array<Klass*>::base_offset_in_bytes()); 497 isync(); // Order load of instance Klass wrt. tags. 498 ldx(Rklass, Rklass, Roffset); 499 } 500 501 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 502 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2. 503 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1, 504 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) { 505 // Profile the not-null value's klass. 506 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2); 507 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype); 508 profile_typecheck_failed(Rtmp1, Rtmp2); 509 } 510 511 // Separate these two to allow for delay slot in middle. 512 // These are used to do a test and full jump to exception-throwing code. 513 514 // 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. 517 518 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, 519 int index_shift, Register Rtmp, Register Rres) { 520 // Check that index is in range for array, then shift index by index_shift, 521 // and put arrayOop + shifted_index into res. 522 // Note: res is still shy of address by array offset into object. 523 // Kills: 524 // - Rindex 525 // Writes: 526 // - Rres: Address that corresponds to the array index if check was successful. 527 verify_oop(Rarray); 528 const Register Rlength = R0; 529 const Register RsxtIndex = Rtmp; 530 Label LisNull, LnotOOR; 531 532 // Array nullcheck 533 if (!ImplicitNullChecks) { 534 cmpdi(CCR0, Rarray, 0); 535 beq(CCR0, LisNull); 536 } else { 537 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex); 538 } 539 540 // Rindex might contain garbage in upper bits (remember that we don't sign extend 541 // during integer arithmetic operations). So kill them and put value into same register 542 // where ArrayIndexOutOfBounds would expect the index in. 543 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit 544 545 // Index check 546 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray); 547 cmplw(CCR0, Rindex, Rlength); 548 sldi(RsxtIndex, RsxtIndex, index_shift); 549 blt(CCR0, LnotOOR); 550 // Index should be in R17_tos, array should be in R4_ARG2. 551 mr_if_needed(R17_tos, Rindex); 552 mr_if_needed(R4_ARG2, Rarray); 553 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry); 554 mtctr(Rtmp); 555 bctr(); 556 557 if (!ImplicitNullChecks) { 558 bind(LisNull); 559 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry); 560 mtctr(Rtmp); 561 bctr(); 562 } 563 564 align(32, 16); 565 bind(LnotOOR); 566 567 // Calc address 568 add(Rres, RsxtIndex, Rarray); 569 } 570 571 void InterpreterMacroAssembler::index_check(Register array, Register index, 572 int index_shift, Register tmp, Register res) { 573 // pop array 574 pop_ptr(array); 575 576 // check array 577 index_check_without_pop(array, index, index_shift, tmp, res); 578 } 579 580 void InterpreterMacroAssembler::get_const(Register Rdst) { 581 ld(Rdst, in_bytes(Method::const_offset()), R19_method); 582 } 583 584 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 585 get_const(Rdst); 586 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); 587 } 588 589 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 590 get_constant_pool(Rdst); 591 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); 592 } 593 594 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 595 get_constant_pool(Rcpool); 596 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool); 597 } 598 599 // Unlock if synchronized method. 600 // 601 // Unlock the receiver if this is a synchronized method. 602 // Unlock any Java monitors from synchronized blocks. 603 // 604 // If there are locked Java monitors 605 // If throw_monitor_exception 606 // throws IllegalMonitorStateException 607 // Else if install_monitor_exception 608 // installs IllegalMonitorStateException 609 // Else 610 // no error processing 611 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 612 bool throw_monitor_exception, 613 bool install_monitor_exception) { 614 Label Lunlocked, Lno_unlock; 615 { 616 Register Rdo_not_unlock_flag = R11_scratch1; 617 Register Raccess_flags = R12_scratch2; 618 619 // Check if synchronized method or unlocking prevented by 620 // JavaThread::do_not_unlock_if_synchronized flag. 621 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 622 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method); 623 li(R0, 0); 624 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag 625 626 push(state); 627 628 // Skip if we don't have to unlock. 629 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0. 630 beq(CCR0, Lunlocked); 631 632 cmpwi(CCR0, Rdo_not_unlock_flag, 0); 633 bne(CCR0, Lno_unlock); 634 } 635 636 // Unlock 637 { 638 Register Rmonitor_base = R11_scratch1; 639 640 Label Lunlock; 641 // If it's still locked, everything is ok, unlock it. 642 ld(Rmonitor_base, 0, R1_SP); 643 addi(Rmonitor_base, Rmonitor_base, 644 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base 645 646 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base); 647 cmpdi(CCR0, R0, 0); 648 bne(CCR0, Lunlock); 649 650 // If it's already unlocked, throw exception. 651 if (throw_monitor_exception) { 652 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 653 should_not_reach_here(); 654 } else { 655 if (install_monitor_exception) { 656 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 657 b(Lunlocked); 658 } 659 } 660 661 bind(Lunlock); 662 unlock_object(Rmonitor_base); 663 } 664 665 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not. 666 bind(Lunlocked); 667 { 668 Label Lexception, Lrestart; 669 Register Rcurrent_obj_addr = R11_scratch1; 670 const int delta = frame::interpreter_frame_monitor_size_in_bytes(); 671 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords"); 672 673 bind(Lrestart); 674 // Set up search loop: Calc num of iterations. 675 { 676 Register Riterations = R12_scratch2; 677 Register Rmonitor_base = Rcurrent_obj_addr; 678 ld(Rmonitor_base, 0, R1_SP); 679 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base 680 681 subf_(Riterations, R26_monitor, Rmonitor_base); 682 ble(CCR0, Lno_unlock); 683 684 addi(Rcurrent_obj_addr, Rmonitor_base, 685 BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes()); 686 // Check if any monitor is on stack, bail out if not 687 srdi(Riterations, Riterations, exact_log2(delta)); 688 mtctr(Riterations); 689 } 690 691 // The search loop: Look for locked monitors. 692 { 693 const Register Rcurrent_obj = R0; 694 Label Lloop; 695 696 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 697 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 698 bind(Lloop); 699 700 // Check if current entry is used. 701 cmpdi(CCR0, Rcurrent_obj, 0); 702 bne(CCR0, Lexception); 703 // Preload next iteration's compare value. 704 ld(Rcurrent_obj, 0, Rcurrent_obj_addr); 705 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); 706 bdnz(Lloop); 707 } 708 // Fell through: Everything's unlocked => finish. 709 b(Lno_unlock); 710 711 // An object is still locked => need to throw exception. 712 bind(Lexception); 713 if (throw_monitor_exception) { 714 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 715 should_not_reach_here(); 716 } else { 717 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. 718 // Unlock does not block, so don't have to worry about the frame. 719 Register Rmonitor_addr = R11_scratch1; 720 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta); 721 unlock_object(Rmonitor_addr); 722 if (install_monitor_exception) { 723 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 724 } 725 b(Lrestart); 726 } 727 } 728 729 align(32, 12); 730 bind(Lno_unlock); 731 pop(state); 732 } 733 734 // Support function for remove_activation & Co. 735 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, 736 Register Rscratch1, Register Rscratch2) { 737 // Pop interpreter frame. 738 ld(Rscratch1, 0, R1_SP); // *SP 739 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp 740 ld(Rscratch2, 0, Rscratch1); // **SP 741 #ifdef ASSERT 742 { 743 Label Lok; 744 ld(R0, _ijava_state_neg(ijava_reserved), Rscratch1); 745 cmpdi(CCR0, R0, 0x5afe); 746 beq(CCR0, Lok); 747 stop("frame corrupted (remove activation)", 0x5afe); 748 bind(Lok); 749 } 750 #endif 751 if (return_pc!=noreg) { 752 ld(return_pc, _abi(lr), Rscratch1); // LR 753 } 754 755 // Merge top frames. 756 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP 757 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP 758 } 759 760 void InterpreterMacroAssembler::narrow(Register result) { 761 Register ret_type = R11_scratch1; 762 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method); 763 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1); 764 765 Label notBool, notByte, notChar, done; 766 767 // common case first 768 cmpwi(CCR0, ret_type, T_INT); 769 beq(CCR0, done); 770 771 cmpwi(CCR0, ret_type, T_BOOLEAN); 772 bne(CCR0, notBool); 773 andi(result, result, 0x1); 774 b(done); 775 776 bind(notBool); 777 cmpwi(CCR0, ret_type, T_BYTE); 778 bne(CCR0, notByte); 779 extsb(result, result); 780 b(done); 781 782 bind(notByte); 783 cmpwi(CCR0, ret_type, T_CHAR); 784 bne(CCR0, notChar); 785 andi(result, result, 0xffff); 786 b(done); 787 788 bind(notChar); 789 // cmpwi(CCR0, ret_type, T_SHORT); // all that's left 790 // bne(CCR0, done); 791 extsh(result, result); 792 793 // Nothing to do for T_INT 794 bind(done); 795 } 796 797 // Remove activation. 798 // 799 // Unlock the receiver if this is a synchronized method. 800 // Unlock any Java monitors from synchronized blocks. 801 // Remove the activation from the stack. 802 // 803 // If there are locked Java monitors 804 // If throw_monitor_exception 805 // throws IllegalMonitorStateException 806 // Else if install_monitor_exception 807 // installs IllegalMonitorStateException 808 // Else 809 // no error processing 810 void InterpreterMacroAssembler::remove_activation(TosState state, 811 bool throw_monitor_exception, 812 bool install_monitor_exception) { 813 BLOCK_COMMENT("remove_activation {"); 814 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 815 816 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 817 notify_method_exit(false, state, NotifyJVMTI, true); 818 819 BLOCK_COMMENT("reserved_stack_check:"); 820 if (StackReservedPages > 0) { 821 // Test if reserved zone needs to be enabled. 822 Label no_reserved_zone_enabling; 823 824 // Compare frame pointers. There is no good stack pointer, as with stack 825 // frame compression we can get different SPs when we do calls. A subsequent 826 // call could have a smaller SP, so that this compare succeeds for an 827 // inner call of the method annotated with ReservedStack. 828 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread); 829 ld_ptr(R11_scratch1, _abi(callers_sp), R1_SP); // Load frame pointer. 830 cmpld(CCR0, R11_scratch1, R0); 831 blt_predict_taken(CCR0, no_reserved_zone_enabling); 832 833 // Enable reserved zone again, throw stack overflow exception. 834 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread); 835 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError)); 836 837 should_not_reach_here(); 838 839 bind(no_reserved_zone_enabling); 840 } 841 842 verify_oop(R17_tos, state); 843 verify_thread(); 844 845 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 846 mtlr(R0); 847 BLOCK_COMMENT("} remove_activation"); 848 } 849 850 // Lock object 851 // 852 // Registers alive 853 // monitor - Address of the BasicObjectLock to be used for locking, 854 // which must be initialized with the object to lock. 855 // object - Address of the object to be locked. 856 // 857 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 858 if (UseHeavyMonitors) { 859 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 860 monitor, /*check_for_exceptions=*/true); 861 } else { 862 // template code: 863 // 864 // markOop displaced_header = obj->mark().set_unlocked(); 865 // monitor->lock()->set_displaced_header(displaced_header); 866 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 867 // // We stored the monitor address into the object's mark word. 868 // } else if (THREAD->is_lock_owned((address)displaced_header)) 869 // // Simple recursive case. 870 // monitor->lock()->set_displaced_header(NULL); 871 // } else { 872 // // Slow path. 873 // InterpreterRuntime::monitorenter(THREAD, monitor); 874 // } 875 876 const Register displaced_header = R7_ARG5; 877 const Register object_mark_addr = R8_ARG6; 878 const Register current_header = R9_ARG7; 879 const Register tmp = R10_ARG8; 880 881 Label done; 882 Label cas_failed, slow_case; 883 884 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp); 885 886 // markOop displaced_header = obj->mark().set_unlocked(); 887 888 // Load markOop from object into displaced_header. 889 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object); 890 891 if (UseBiasedLocking) { 892 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case); 893 } 894 895 // Set displaced_header to be (markOop of object | UNLOCK_VALUE). 896 ori(displaced_header, displaced_header, markOopDesc::unlocked_value); 897 898 // monitor->lock()->set_displaced_header(displaced_header); 899 900 // Initialize the box (Must happen before we update the object mark!). 901 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 902 BasicLock::displaced_header_offset_in_bytes(), monitor); 903 904 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 905 906 // Store stack address of the BasicObjectLock (this is monitor) into object. 907 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 908 909 // Must fence, otherwise, preceding store(s) may float below cmpxchg. 910 // CmpxchgX sets CCR0 to cmpX(current, displaced). 911 cmpxchgd(/*flag=*/CCR0, 912 /*current_value=*/current_header, 913 /*compare_value=*/displaced_header, /*exchange_value=*/monitor, 914 /*where=*/object_mark_addr, 915 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq, 916 MacroAssembler::cmpxchgx_hint_acquire_lock(), 917 noreg, 918 &cas_failed, 919 /*check without membar and ldarx first*/true); 920 921 // If the compare-and-exchange succeeded, then we found an unlocked 922 // object and we have now locked it. 923 b(done); 924 bind(cas_failed); 925 926 // } else if (THREAD->is_lock_owned((address)displaced_header)) 927 // // Simple recursive case. 928 // monitor->lock()->set_displaced_header(NULL); 929 930 // We did not see an unlocked object so try the fast recursive case. 931 932 // Check if owner is self by comparing the value in the markOop of object 933 // (current_header) with the stack pointer. 934 sub(current_header, current_header, R1_SP); 935 936 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 937 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place); 938 939 and_(R0/*==0?*/, current_header, tmp); 940 // If condition is true we are done and hence we can store 0 in the displaced 941 // header indicating it is a recursive lock. 942 bne(CCR0, slow_case); 943 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() + 944 BasicLock::displaced_header_offset_in_bytes(), monitor); 945 b(done); 946 947 // } else { 948 // // Slow path. 949 // InterpreterRuntime::monitorenter(THREAD, monitor); 950 951 // None of the above fast optimizations worked so we have to get into the 952 // slow case of monitor enter. 953 bind(slow_case); 954 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 955 monitor, /*check_for_exceptions=*/true); 956 // } 957 align(32, 12); 958 bind(done); 959 } 960 } 961 962 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 963 // 964 // Registers alive 965 // monitor - Address of the BasicObjectLock to be used for locking, 966 // which must be initialized with the object to lock. 967 // 968 // Throw IllegalMonitorException if object is not locked by current thread. 969 void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) { 970 if (UseHeavyMonitors) { 971 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 972 monitor, check_for_exceptions); 973 } else { 974 975 // template code: 976 // 977 // if ((displaced_header = monitor->displaced_header()) == NULL) { 978 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 979 // monitor->set_obj(NULL); 980 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 981 // // We swapped the unlocked mark in displaced_header into the object's mark word. 982 // monitor->set_obj(NULL); 983 // } else { 984 // // Slow path. 985 // InterpreterRuntime::monitorexit(THREAD, monitor); 986 // } 987 988 const Register object = R7_ARG5; 989 const Register displaced_header = R8_ARG6; 990 const Register object_mark_addr = R9_ARG7; 991 const Register current_header = R10_ARG8; 992 993 Label free_slot; 994 Label slow_case; 995 996 assert_different_registers(object, displaced_header, object_mark_addr, current_header); 997 998 if (UseBiasedLocking) { 999 // The object address from the monitor is in object. 1000 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); 1001 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0"); 1002 biased_locking_exit(CCR0, object, displaced_header, free_slot); 1003 } 1004 1005 // Test first if we are in the fast recursive case. 1006 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 1007 BasicLock::displaced_header_offset_in_bytes(), monitor); 1008 1009 // If the displaced header is zero, we have a recursive unlock. 1010 cmpdi(CCR0, displaced_header, 0); 1011 beq(CCR0, free_slot); // recursive unlock 1012 1013 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 1014 // // We swapped the unlocked mark in displaced_header into the object's mark word. 1015 // monitor->set_obj(NULL); 1016 1017 // If we still have a lightweight lock, unlock the object and be done. 1018 1019 // The object address from the monitor is in object. 1020 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); } 1021 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); 1022 1023 // We have the displaced header in displaced_header. If the lock is still 1024 // lightweight, it will contain the monitor address and we'll store the 1025 // displaced header back into the object's mark word. 1026 // CmpxchgX sets CCR0 to cmpX(current, monitor). 1027 cmpxchgd(/*flag=*/CCR0, 1028 /*current_value=*/current_header, 1029 /*compare_value=*/monitor, /*exchange_value=*/displaced_header, 1030 /*where=*/object_mark_addr, 1031 MacroAssembler::MemBarRel, 1032 MacroAssembler::cmpxchgx_hint_release_lock(), 1033 noreg, 1034 &slow_case); 1035 b(free_slot); 1036 1037 // } else { 1038 // // Slow path. 1039 // InterpreterRuntime::monitorexit(THREAD, monitor); 1040 1041 // The lock has been converted into a heavy lock and hence 1042 // we need to get into the slow case. 1043 bind(slow_case); 1044 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1045 monitor, check_for_exceptions); 1046 // } 1047 1048 Label done; 1049 b(done); // Monitor register may be overwritten! Runtime has already freed the slot. 1050 1051 // Exchange worked, do monitor->set_obj(NULL); 1052 align(32, 12); 1053 bind(free_slot); 1054 li(R0, 0); 1055 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor); 1056 bind(done); 1057 } 1058 } 1059 1060 // Load compiled (i2c) or interpreter entry when calling from interpreted and 1061 // do the call. Centralized so that all interpreter calls will do the same actions. 1062 // If jvmti single stepping is on for a thread we must not call compiled code. 1063 // 1064 // Input: 1065 // - Rtarget_method: method to call 1066 // - Rret_addr: return address 1067 // - 2 scratch regs 1068 // 1069 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, 1070 Register Rscratch1, Register Rscratch2) { 1071 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr); 1072 // Assume we want to go compiled if available. 1073 const Register Rtarget_addr = Rscratch1; 1074 const Register Rinterp_only = Rscratch2; 1075 1076 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method); 1077 1078 if (JvmtiExport::can_post_interpreter_events()) { 1079 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 1080 1081 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 1082 // compiled code in threads for which the event is enabled. Check here for 1083 // interp_only_mode if these events CAN be enabled. 1084 Label done; 1085 verify_thread(); 1086 cmpwi(CCR0, Rinterp_only, 0); 1087 beq(CCR0, done); 1088 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method); 1089 align(32, 12); 1090 bind(done); 1091 } 1092 1093 #ifdef ASSERT 1094 { 1095 Label Lok; 1096 cmpdi(CCR0, Rtarget_addr, 0); 1097 bne(CCR0, Lok); 1098 stop("null entry point"); 1099 bind(Lok); 1100 } 1101 #endif // ASSERT 1102 1103 mr(R21_sender_SP, R1_SP); 1104 1105 // Calc a precise SP for the call. The SP value we calculated in 1106 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space 1107 // if esp is not max. Also, the i2c adapter extends the stack space without restoring 1108 // our pre-calced value, so repeating calls via i2c would result in stack overflow. 1109 // Since esp already points to an empty slot, we just have to sub 1 additional slot 1110 // to meet the abi scratch requirements. 1111 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in 1112 // the return entry of the interpreter. 1113 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size); 1114 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address 1115 resize_frame_absolute(Rscratch2, Rscratch2, R0); 1116 1117 mr_if_needed(R19_method, Rtarget_method); 1118 mtctr(Rtarget_addr); 1119 mtlr(Rret_addr); 1120 1121 save_interpreter_state(Rscratch2); 1122 #ifdef ASSERT 1123 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp 1124 cmpd(CCR0, R21_sender_SP, Rscratch1); 1125 asm_assert_eq("top_frame_sp incorrect", 0x951); 1126 #endif 1127 1128 bctr(); 1129 } 1130 1131 // Set the method data pointer for the current bcp. 1132 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1133 assert(ProfileInterpreter, "must be profiling interpreter"); 1134 Label get_continue; 1135 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method); 1136 test_method_data_pointer(get_continue); 1137 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp); 1138 1139 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 1140 add(R28_mdx, R28_mdx, R3_RET); 1141 bind(get_continue); 1142 } 1143 1144 // Test ImethodDataPtr. If it is null, continue at the specified label. 1145 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { 1146 assert(ProfileInterpreter, "must be profiling interpreter"); 1147 cmpdi(CCR0, R28_mdx, 0); 1148 beq(CCR0, zero_continue); 1149 } 1150 1151 void InterpreterMacroAssembler::verify_method_data_pointer() { 1152 assert(ProfileInterpreter, "must be profiling interpreter"); 1153 #ifdef ASSERT 1154 Label verify_continue; 1155 test_method_data_pointer(verify_continue); 1156 1157 // If the mdp is valid, it will point to a DataLayout header which is 1158 // consistent with the bcp. The converse is highly probable also. 1159 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx); 1160 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method); 1161 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset())); 1162 add(R11_scratch1, R12_scratch2, R12_scratch2); 1163 cmpd(CCR0, R11_scratch1, R14_bcp); 1164 beq(CCR0, verify_continue); 1165 1166 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx); 1167 1168 bind(verify_continue); 1169 #endif 1170 } 1171 1172 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, 1173 Register method_counters, 1174 Register Rscratch, 1175 Label &profile_continue) { 1176 assert(ProfileInterpreter, "must be profiling interpreter"); 1177 // Control will flow to "profile_continue" if the counter is less than the 1178 // limit or if we call profile_method(). 1179 Label done; 1180 1181 // If no method data exists, and the counter is high enough, make one. 1182 lwz(Rscratch, in_bytes(MethodCounters::interpreter_profile_limit_offset()), method_counters); 1183 1184 cmpdi(CCR0, R28_mdx, 0); 1185 // Test to see if we should create a method data oop. 1186 cmpd(CCR1, Rscratch, invocation_count); 1187 bne(CCR0, done); 1188 bge(CCR1, profile_continue); 1189 1190 // Build it now. 1191 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1192 set_method_data_pointer_for_bcp(); 1193 b(profile_continue); 1194 1195 align(32, 12); 1196 bind(done); 1197 } 1198 1199 void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register method_counters, 1200 Register target_bcp, Register disp, Register Rtmp) { 1201 assert_different_registers(backedge_count, target_bcp, disp, Rtmp, R4_ARG2); 1202 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); 1203 1204 Label did_not_overflow; 1205 Label overflow_with_error; 1206 1207 lwz(Rtmp, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()), method_counters); 1208 cmpw(CCR0, backedge_count, Rtmp); 1209 1210 blt(CCR0, did_not_overflow); 1211 1212 // When ProfileInterpreter is on, the backedge_count comes from the 1213 // methodDataOop, which value does not get reset on the call to 1214 // frequency_counter_overflow(). To avoid excessive calls to the overflow 1215 // routine while the method is being compiled, add a second test to make sure 1216 // the overflow function is called only once every overflow_frequency. 1217 if (ProfileInterpreter) { 1218 const int overflow_frequency = 1024; 1219 andi_(Rtmp, backedge_count, overflow_frequency-1); 1220 bne(CCR0, did_not_overflow); 1221 } 1222 1223 // Overflow in loop, pass branch bytecode. 1224 subf(R4_ARG2, disp, target_bcp); // Compute branch bytecode (previous bcp). 1225 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); 1226 1227 // Was an OSR adapter generated? 1228 cmpdi(CCR0, R3_RET, 0); 1229 beq(CCR0, overflow_with_error); 1230 1231 // Has the nmethod been invalidated already? 1232 lbz(Rtmp, nmethod::state_offset(), R3_RET); 1233 cmpwi(CCR0, Rtmp, nmethod::in_use); 1234 bne(CCR0, overflow_with_error); 1235 1236 // Migrate the interpreter frame off of the stack. 1237 // We can use all registers because we will not return to interpreter from this point. 1238 1239 // Save nmethod. 1240 const Register osr_nmethod = R31; 1241 mr(osr_nmethod, R3_RET); 1242 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1); 1243 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread); 1244 reset_last_Java_frame(); 1245 // OSR buffer is in ARG1 1246 1247 // Remove the interpreter frame. 1248 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1249 1250 // Jump to the osr code. 1251 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod); 1252 mtlr(R0); 1253 mtctr(R11_scratch1); 1254 bctr(); 1255 1256 align(32, 12); 1257 bind(overflow_with_error); 1258 bind(did_not_overflow); 1259 } 1260 1261 // Store a value at some constant offset from the method data pointer. 1262 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { 1263 assert(ProfileInterpreter, "must be profiling interpreter"); 1264 1265 std(value, constant, R28_mdx); 1266 } 1267 1268 // Increment the value at some constant offset from the method data pointer. 1269 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, 1270 Register counter_addr, 1271 Register Rbumped_count, 1272 bool decrement) { 1273 // Locate the counter at a fixed offset from the mdp: 1274 addi(counter_addr, R28_mdx, constant); 1275 increment_mdp_data_at(counter_addr, Rbumped_count, decrement); 1276 } 1277 1278 // Increment the value at some non-fixed (reg + constant) offset from 1279 // the method data pointer. 1280 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, 1281 int constant, 1282 Register scratch, 1283 Register Rbumped_count, 1284 bool decrement) { 1285 // Add the constant to reg to get the offset. 1286 add(scratch, R28_mdx, reg); 1287 // Then calculate the counter address. 1288 addi(scratch, scratch, constant); 1289 increment_mdp_data_at(scratch, Rbumped_count, decrement); 1290 } 1291 1292 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr, 1293 Register Rbumped_count, 1294 bool decrement) { 1295 assert(ProfileInterpreter, "must be profiling interpreter"); 1296 1297 // Load the counter. 1298 ld(Rbumped_count, 0, counter_addr); 1299 1300 if (decrement) { 1301 // Decrement the register. Set condition codes. 1302 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment); 1303 // Store the decremented counter, if it is still negative. 1304 std(Rbumped_count, 0, counter_addr); 1305 // Note: add/sub overflow check are not ported, since 64 bit 1306 // calculation should never overflow. 1307 } else { 1308 // Increment the register. Set carry flag. 1309 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment); 1310 // Store the incremented counter. 1311 std(Rbumped_count, 0, counter_addr); 1312 } 1313 } 1314 1315 // Set a flag value at the current method data pointer position. 1316 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, 1317 Register scratch) { 1318 assert(ProfileInterpreter, "must be profiling interpreter"); 1319 // Load the data header. 1320 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1321 // Set the flag. 1322 ori(scratch, scratch, flag_constant); 1323 // Store the modified header. 1324 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); 1325 } 1326 1327 // Test the location at some offset from the method data pointer. 1328 // If it is not equal to value, branch to the not_equal_continue Label. 1329 void InterpreterMacroAssembler::test_mdp_data_at(int offset, 1330 Register value, 1331 Label& not_equal_continue, 1332 Register test_out) { 1333 assert(ProfileInterpreter, "must be profiling interpreter"); 1334 1335 ld(test_out, offset, R28_mdx); 1336 cmpd(CCR0, value, test_out); 1337 bne(CCR0, not_equal_continue); 1338 } 1339 1340 // Update the method data pointer by the displacement located at some fixed 1341 // offset from the method data pointer. 1342 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, 1343 Register scratch) { 1344 assert(ProfileInterpreter, "must be profiling interpreter"); 1345 1346 ld(scratch, offset_of_disp, R28_mdx); 1347 add(R28_mdx, scratch, R28_mdx); 1348 } 1349 1350 // Update the method data pointer by the displacement located at the 1351 // offset (reg + offset_of_disp). 1352 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, 1353 int offset_of_disp, 1354 Register scratch) { 1355 assert(ProfileInterpreter, "must be profiling interpreter"); 1356 1357 add(scratch, reg, R28_mdx); 1358 ld(scratch, offset_of_disp, scratch); 1359 add(R28_mdx, scratch, R28_mdx); 1360 } 1361 1362 // Update the method data pointer by a simple constant displacement. 1363 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { 1364 assert(ProfileInterpreter, "must be profiling interpreter"); 1365 addi(R28_mdx, R28_mdx, constant); 1366 } 1367 1368 // Update the method data pointer for a _ret bytecode whose target 1369 // was not among our cached targets. 1370 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, 1371 Register return_bci) { 1372 assert(ProfileInterpreter, "must be profiling interpreter"); 1373 1374 push(state); 1375 assert(return_bci->is_nonvolatile(), "need to protect return_bci"); 1376 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); 1377 pop(state); 1378 } 1379 1380 // Increments the backedge counter. 1381 // Returns backedge counter + invocation counter in Rdst. 1382 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst, 1383 const Register Rtmp1, Register Rscratch) { 1384 assert(UseCompiler, "incrementing must be useful"); 1385 assert_different_registers(Rdst, Rtmp1); 1386 const Register invocation_counter = Rtmp1; 1387 const Register counter = Rdst; 1388 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size."); 1389 1390 // Load backedge counter. 1391 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1392 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1393 // Load invocation counter. 1394 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) + 1395 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1396 1397 // Add the delta to the backedge counter. 1398 addi(counter, counter, InvocationCounter::count_increment); 1399 1400 // Mask the invocation counter. 1401 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value); 1402 1403 // Store new counter value. 1404 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) + 1405 in_bytes(InvocationCounter::counter_offset()), Rcounters); 1406 // Return invocation counter + backedge counter. 1407 add(counter, counter, invocation_counter); 1408 } 1409 1410 // Count a taken branch in the bytecodes. 1411 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { 1412 if (ProfileInterpreter) { 1413 Label profile_continue; 1414 1415 // If no method data exists, go to profile_continue. 1416 test_method_data_pointer(profile_continue); 1417 1418 // We are taking a branch. Increment the taken count. 1419 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count); 1420 1421 // The method data pointer needs to be updated to reflect the new target. 1422 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); 1423 bind (profile_continue); 1424 } 1425 } 1426 1427 // Count a not-taken branch in the bytecodes. 1428 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) { 1429 if (ProfileInterpreter) { 1430 Label profile_continue; 1431 1432 // If no method data exists, go to profile_continue. 1433 test_method_data_pointer(profile_continue); 1434 1435 // We are taking a branch. Increment the not taken count. 1436 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2); 1437 1438 // The method data pointer needs to be updated to correspond to the 1439 // next bytecode. 1440 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); 1441 bind (profile_continue); 1442 } 1443 } 1444 1445 // Count a non-virtual call in the bytecodes. 1446 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) { 1447 if (ProfileInterpreter) { 1448 Label profile_continue; 1449 1450 // If no method data exists, go to profile_continue. 1451 test_method_data_pointer(profile_continue); 1452 1453 // We are making a call. Increment the count. 1454 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1455 1456 // The method data pointer needs to be updated to reflect the new target. 1457 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); 1458 bind (profile_continue); 1459 } 1460 } 1461 1462 // Count a final call in the bytecodes. 1463 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) { 1464 if (ProfileInterpreter) { 1465 Label profile_continue; 1466 1467 // If no method data exists, go to profile_continue. 1468 test_method_data_pointer(profile_continue); 1469 1470 // We are making a call. Increment the count. 1471 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1472 1473 // The method data pointer needs to be updated to reflect the new target. 1474 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1475 bind (profile_continue); 1476 } 1477 } 1478 1479 // Count a virtual call in the bytecodes. 1480 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver, 1481 Register Rscratch1, 1482 Register Rscratch2, 1483 bool receiver_can_be_null) { 1484 if (!ProfileInterpreter) { return; } 1485 Label profile_continue; 1486 1487 // If no method data exists, go to profile_continue. 1488 test_method_data_pointer(profile_continue); 1489 1490 Label skip_receiver_profile; 1491 if (receiver_can_be_null) { 1492 Label not_null; 1493 cmpdi(CCR0, Rreceiver, 0); 1494 bne(CCR0, not_null); 1495 // We are making a call. Increment the count for null receiver. 1496 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2); 1497 b(skip_receiver_profile); 1498 bind(not_null); 1499 } 1500 1501 // Record the receiver type. 1502 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true); 1503 bind(skip_receiver_profile); 1504 1505 // The method data pointer needs to be updated to reflect the new target. 1506 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); 1507 bind (profile_continue); 1508 } 1509 1510 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) { 1511 if (ProfileInterpreter) { 1512 Label profile_continue; 1513 1514 // If no method data exists, go to profile_continue. 1515 test_method_data_pointer(profile_continue); 1516 1517 int mdp_delta = in_bytes(BitData::bit_data_size()); 1518 if (TypeProfileCasts) { 1519 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1520 1521 // Record the object type. 1522 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false); 1523 } 1524 1525 // The method data pointer needs to be updated. 1526 update_mdp_by_constant(mdp_delta); 1527 1528 bind (profile_continue); 1529 } 1530 } 1531 1532 void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) { 1533 if (ProfileInterpreter && TypeProfileCasts) { 1534 Label profile_continue; 1535 1536 // If no method data exists, go to profile_continue. 1537 test_method_data_pointer(profile_continue); 1538 1539 int count_offset = in_bytes(CounterData::count_offset()); 1540 // Back up the address, since we have already bumped the mdp. 1541 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1542 1543 // *Decrement* the counter. We expect to see zero or small negatives. 1544 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true); 1545 1546 bind (profile_continue); 1547 } 1548 } 1549 1550 // Count a ret in the bytecodes. 1551 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, 1552 Register scratch1, Register scratch2) { 1553 if (ProfileInterpreter) { 1554 Label profile_continue; 1555 uint row; 1556 1557 // If no method data exists, go to profile_continue. 1558 test_method_data_pointer(profile_continue); 1559 1560 // Update the total ret count. 1561 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 ); 1562 1563 for (row = 0; row < RetData::row_limit(); row++) { 1564 Label next_test; 1565 1566 // See if return_bci is equal to bci[n]: 1567 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1); 1568 1569 // return_bci is equal to bci[n]. Increment the count. 1570 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2); 1571 1572 // The method data pointer needs to be updated to reflect the new target. 1573 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1); 1574 b(profile_continue); 1575 bind(next_test); 1576 } 1577 1578 update_mdp_for_ret(state, return_bci); 1579 1580 bind (profile_continue); 1581 } 1582 } 1583 1584 // Count the default case of a switch construct. 1585 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) { 1586 if (ProfileInterpreter) { 1587 Label profile_continue; 1588 1589 // If no method data exists, go to profile_continue. 1590 test_method_data_pointer(profile_continue); 1591 1592 // Update the default case count 1593 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), 1594 scratch1, scratch2); 1595 1596 // The method data pointer needs to be updated. 1597 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()), 1598 scratch1); 1599 1600 bind (profile_continue); 1601 } 1602 } 1603 1604 // Count the index'th case of a switch construct. 1605 void InterpreterMacroAssembler::profile_switch_case(Register index, 1606 Register scratch1, 1607 Register scratch2, 1608 Register scratch3) { 1609 if (ProfileInterpreter) { 1610 assert_different_registers(index, scratch1, scratch2, scratch3); 1611 Label profile_continue; 1612 1613 // If no method data exists, go to profile_continue. 1614 test_method_data_pointer(profile_continue); 1615 1616 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes(). 1617 li(scratch3, in_bytes(MultiBranchData::case_array_offset())); 1618 1619 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works"); 1620 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1621 add(scratch1, scratch1, scratch3); 1622 1623 // Update the case count. 1624 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3); 1625 1626 // The method data pointer needs to be updated. 1627 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2); 1628 1629 bind (profile_continue); 1630 } 1631 } 1632 1633 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) { 1634 if (ProfileInterpreter) { 1635 assert_different_registers(Rscratch1, Rscratch2); 1636 Label profile_continue; 1637 1638 // If no method data exists, go to profile_continue. 1639 test_method_data_pointer(profile_continue); 1640 1641 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1); 1642 1643 // The method data pointer needs to be updated. 1644 int mdp_delta = in_bytes(BitData::bit_data_size()); 1645 if (TypeProfileCasts) { 1646 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1647 } 1648 update_mdp_by_constant(mdp_delta); 1649 1650 bind (profile_continue); 1651 } 1652 } 1653 1654 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver, 1655 Register Rscratch1, Register Rscratch2, 1656 bool is_virtual_call) { 1657 assert(ProfileInterpreter, "must be profiling"); 1658 assert_different_registers(Rreceiver, Rscratch1, Rscratch2); 1659 1660 Label done; 1661 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call); 1662 bind (done); 1663 } 1664 1665 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1666 Register receiver, Register scratch1, Register scratch2, 1667 int start_row, Label& done, bool is_virtual_call) { 1668 if (TypeProfileWidth == 0) { 1669 if (is_virtual_call) { 1670 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1671 } 1672 return; 1673 } 1674 1675 int last_row = VirtualCallData::row_limit() - 1; 1676 assert(start_row <= last_row, "must be work left to do"); 1677 // Test this row for both the receiver and for null. 1678 // Take any of three different outcomes: 1679 // 1. found receiver => increment count and goto done 1680 // 2. found null => keep looking for case 1, maybe allocate this cell 1681 // 3. found something else => keep looking for cases 1 and 2 1682 // Case 3 is handled by a recursive call. 1683 for (int row = start_row; row <= last_row; row++) { 1684 Label next_test; 1685 bool test_for_null_also = (row == start_row); 1686 1687 // See if the receiver is receiver[n]. 1688 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1689 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1); 1690 // delayed()->tst(scratch); 1691 1692 // The receiver is receiver[n]. Increment count[n]. 1693 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1694 increment_mdp_data_at(count_offset, scratch1, scratch2); 1695 b(done); 1696 bind(next_test); 1697 1698 if (test_for_null_also) { 1699 Label found_null; 1700 // Failed the equality check on receiver[n]... Test for null. 1701 if (start_row == last_row) { 1702 // The only thing left to do is handle the null case. 1703 if (is_virtual_call) { 1704 // Scratch1 contains test_out from test_mdp_data_at. 1705 cmpdi(CCR0, scratch1, 0); 1706 beq(CCR0, found_null); 1707 // Receiver did not match any saved receiver and there is no empty row for it. 1708 // Increment total counter to indicate polymorphic case. 1709 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); 1710 b(done); 1711 bind(found_null); 1712 } else { 1713 cmpdi(CCR0, scratch1, 0); 1714 bne(CCR0, done); 1715 } 1716 break; 1717 } 1718 // Since null is rare, make it be the branch-taken case. 1719 cmpdi(CCR0, scratch1, 0); 1720 beq(CCR0, found_null); 1721 1722 // Put all the "Case 3" tests here. 1723 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call); 1724 1725 // Found a null. Keep searching for a matching receiver, 1726 // but remember that this is an empty (unused) slot. 1727 bind(found_null); 1728 } 1729 } 1730 1731 // In the fall-through case, we found no matching receiver, but we 1732 // observed the receiver[start_row] is NULL. 1733 1734 // Fill in the receiver field and increment the count. 1735 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1736 set_mdp_data_at(recvr_offset, receiver); 1737 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1738 li(scratch1, DataLayout::counter_increment); 1739 set_mdp_data_at(count_offset, scratch1); 1740 if (start_row > 0) { 1741 b(done); 1742 } 1743 } 1744 1745 // Argument and return type profilig. 1746 // kills: tmp, tmp2, R0, CR0, CR1 1747 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base, 1748 RegisterOrConstant mdo_addr_offs, 1749 Register tmp, Register tmp2) { 1750 Label do_nothing, do_update; 1751 1752 // tmp2 = obj is allowed 1753 assert_different_registers(obj, mdo_addr_base, tmp, R0); 1754 assert_different_registers(tmp2, mdo_addr_base, tmp, R0); 1755 const Register klass = tmp2; 1756 1757 verify_oop(obj); 1758 1759 ld(tmp, mdo_addr_offs, mdo_addr_base); 1760 1761 // Set null_seen if obj is 0. 1762 cmpdi(CCR0, obj, 0); 1763 ori(R0, tmp, TypeEntries::null_seen); 1764 beq(CCR0, do_update); 1765 1766 load_klass(klass, obj); 1767 1768 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask)); 1769 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask); 1770 cmpd(CCR1, R0, klass); 1771 // Klass seen before, nothing to do (regardless of unknown bit). 1772 //beq(CCR1, do_nothing); 1773 1774 andi_(R0, klass, TypeEntries::type_unknown); 1775 // Already unknown. Nothing to do anymore. 1776 //bne(CCR0, do_nothing); 1777 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne 1778 beq(CCR0, do_nothing); 1779 1780 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask)); 1781 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1782 beq(CCR0, do_update); // First time here. Set profile type. 1783 1784 // Different than before. Cannot keep accurate profile. 1785 ori(R0, tmp, TypeEntries::type_unknown); 1786 1787 bind(do_update); 1788 // update profile 1789 std(R0, mdo_addr_offs, mdo_addr_base); 1790 1791 align(32, 12); 1792 bind(do_nothing); 1793 } 1794 1795 void InterpreterMacroAssembler::profile_arguments_type(Register callee, 1796 Register tmp1, Register tmp2, 1797 bool is_virtual) { 1798 if (!ProfileInterpreter) { 1799 return; 1800 } 1801 1802 assert_different_registers(callee, tmp1, tmp2, R28_mdx); 1803 1804 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1805 Label profile_continue; 1806 1807 test_method_data_pointer(profile_continue); 1808 1809 int off_to_start = is_virtual ? 1810 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1811 1812 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx); 1813 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1814 bne(CCR0, profile_continue); 1815 1816 if (MethodData::profile_arguments()) { 1817 Label done; 1818 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1819 add(R28_mdx, off_to_args, R28_mdx); 1820 1821 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1822 if (i > 0 || MethodData::profile_return()) { 1823 // If return value type is profiled we may have no argument to profile. 1824 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1825 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count()); 1826 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count()); 1827 blt(CCR0, done); 1828 } 1829 ld(tmp1, in_bytes(Method::const_offset()), callee); 1830 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1); 1831 // Stack offset o (zero based) from the start of the argument 1832 // list, for n arguments translates into offset n - o - 1 from 1833 // the end of the argument list. But there's an extra slot at 1834 // the top of the stack. So the offset is n - o from Lesp. 1835 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx); 1836 subf(tmp1, tmp2, tmp1); 1837 1838 sldi(tmp1, tmp1, Interpreter::logStackElementSize); 1839 ldx(tmp1, tmp1, R15_esp); 1840 1841 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1); 1842 1843 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1844 addi(R28_mdx, R28_mdx, to_add); 1845 off_to_args += to_add; 1846 } 1847 1848 if (MethodData::profile_return()) { 1849 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx); 1850 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1851 } 1852 1853 bind(done); 1854 1855 if (MethodData::profile_return()) { 1856 // We're right after the type profile for the last 1857 // argument. tmp1 is the number of cells left in the 1858 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1859 // if there's a return to profile. 1860 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), 1861 "can't move past ret type"); 1862 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size)); 1863 add(R28_mdx, tmp1, R28_mdx); 1864 } 1865 } else { 1866 assert(MethodData::profile_return(), "either profile call args or call ret"); 1867 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size())); 1868 } 1869 1870 // Mdp points right after the end of the 1871 // CallTypeData/VirtualCallTypeData, right after the cells for the 1872 // return value type if there's one. 1873 align(32, 12); 1874 bind(profile_continue); 1875 } 1876 } 1877 1878 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) { 1879 assert_different_registers(ret, tmp1, tmp2); 1880 if (ProfileInterpreter && MethodData::profile_return()) { 1881 Label profile_continue; 1882 1883 test_method_data_pointer(profile_continue); 1884 1885 if (MethodData::profile_return_jsr292_only()) { 1886 // If we don't profile all invoke bytecodes we must make sure 1887 // it's a bytecode we indeed profile. We can't go back to the 1888 // begining of the ProfileData we intend to update to check its 1889 // type because we're right after it and we don't known its 1890 // length. 1891 lbz(tmp1, 0, R14_bcp); 1892 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method); 1893 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic); 1894 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle); 1895 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1896 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm); 1897 cror(CCR0, Assembler::equal, CCR1, Assembler::equal); 1898 bne(CCR0, profile_continue); 1899 } 1900 1901 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2); 1902 1903 align(32, 12); 1904 bind(profile_continue); 1905 } 1906 } 1907 1908 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, 1909 Register tmp3, Register tmp4) { 1910 if (ProfileInterpreter && MethodData::profile_parameters()) { 1911 Label profile_continue, done; 1912 1913 test_method_data_pointer(profile_continue); 1914 1915 // Load the offset of the area within the MDO used for 1916 // parameters. If it's negative we're not profiling any parameters. 1917 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx); 1918 cmpwi(CCR0, tmp1, 0); 1919 blt(CCR0, profile_continue); 1920 1921 // Compute a pointer to the area for parameters from the offset 1922 // and move the pointer to the slot for the last 1923 // parameters. Collect profiling from last parameter down. 1924 // mdo start + parameters offset + array length - 1 1925 1926 // Pointer to the parameter area in the MDO. 1927 const Register mdp = tmp1; 1928 add(mdp, tmp1, R28_mdx); 1929 1930 // Offset of the current profile entry to update. 1931 const Register entry_offset = tmp2; 1932 // entry_offset = array len in number of cells 1933 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp); 1934 1935 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1936 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells"); 1937 1938 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field 1939 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size)); 1940 // entry_offset in bytes 1941 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1942 1943 Label loop; 1944 align(32, 12); 1945 bind(loop); 1946 1947 // Load offset on the stack from the slot for this parameter. 1948 ld(tmp3, entry_offset, mdp); 1949 sldi(tmp3, tmp3, Interpreter::logStackElementSize); 1950 neg(tmp3, tmp3); 1951 // Read the parameter from the local area. 1952 ldx(tmp3, tmp3, R18_locals); 1953 1954 // Make entry_offset now point to the type field for this parameter. 1955 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1956 assert(type_base > off_base, "unexpected"); 1957 addi(entry_offset, entry_offset, type_base - off_base); 1958 1959 // Profile the parameter. 1960 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3); 1961 1962 // Go to next parameter. 1963 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base); 1964 cmpdi(CCR0, entry_offset, off_base + delta); 1965 addi(entry_offset, entry_offset, -delta); 1966 bge(CCR0, loop); 1967 1968 align(32, 12); 1969 bind(profile_continue); 1970 } 1971 } 1972 1973 // Add a InterpMonitorElem to stack (see frame_sparc.hpp). 1974 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) { 1975 1976 // Very-local scratch registers. 1977 const Register esp = Rtemp1; 1978 const Register slot = Rtemp2; 1979 1980 // Extracted monitor_size. 1981 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes(); 1982 assert(Assembler::is_aligned((unsigned int)monitor_size, 1983 (unsigned int)frame::alignment_in_bytes), 1984 "size of a monitor must respect alignment of SP"); 1985 1986 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor 1987 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp 1988 1989 // Shuffle expression stack down. Recall that stack_base points 1990 // just above the new expression stack bottom. Old_tos and new_tos 1991 // are used to scan thru the old and new expression stacks. 1992 if (!stack_is_empty) { 1993 Label copy_slot, copy_slot_finished; 1994 const Register n_slots = slot; 1995 1996 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack). 1997 subf(n_slots, esp, R26_monitor); 1998 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy. 1999 assert(LogBytesPerWord == 3, "conflicts assembler instructions"); 2000 beq(CCR0, copy_slot_finished); // Nothing to copy. 2001 2002 mtctr(n_slots); 2003 2004 // loop 2005 bind(copy_slot); 2006 ld(slot, 0, esp); // Move expression stack down. 2007 std(slot, -monitor_size, esp); // distance = monitor_size 2008 addi(esp, esp, BytesPerWord); 2009 bdnz(copy_slot); 2010 2011 bind(copy_slot_finished); 2012 } 2013 2014 addi(R15_esp, R15_esp, -monitor_size); 2015 addi(R26_monitor, R26_monitor, -monitor_size); 2016 2017 // Restart interpreter 2018 } 2019 2020 // ============================================================================ 2021 // Java locals access 2022 2023 // Load a local variable at index in Rindex into register Rdst_value. 2024 // Also puts address of local into Rdst_address as a service. 2025 // Kills: 2026 // - Rdst_value 2027 // - Rdst_address 2028 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) { 2029 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2030 subf(Rdst_address, Rdst_address, R18_locals); 2031 lwz(Rdst_value, 0, Rdst_address); 2032 } 2033 2034 // Load a local variable at index in Rindex into register Rdst_value. 2035 // Also puts address of local into Rdst_address as a service. 2036 // Kills: 2037 // - Rdst_value 2038 // - Rdst_address 2039 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) { 2040 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2041 subf(Rdst_address, Rdst_address, R18_locals); 2042 ld(Rdst_value, -8, Rdst_address); 2043 } 2044 2045 // Load a local variable at index in Rindex into register Rdst_value. 2046 // Also puts address of local into Rdst_address as a service. 2047 // Input: 2048 // - Rindex: slot nr of local variable 2049 // Kills: 2050 // - Rdst_value 2051 // - Rdst_address 2052 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, 2053 Register Rdst_address, 2054 Register Rindex) { 2055 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2056 subf(Rdst_address, Rdst_address, R18_locals); 2057 ld(Rdst_value, 0, Rdst_address); 2058 } 2059 2060 // Load a local variable at index in Rindex into register Rdst_value. 2061 // Also puts address of local into Rdst_address as a service. 2062 // Kills: 2063 // - Rdst_value 2064 // - Rdst_address 2065 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, 2066 Register Rdst_address, 2067 Register Rindex) { 2068 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2069 subf(Rdst_address, Rdst_address, R18_locals); 2070 lfs(Rdst_value, 0, Rdst_address); 2071 } 2072 2073 // Load a local variable at index in Rindex into register Rdst_value. 2074 // Also puts address of local into Rdst_address as a service. 2075 // Kills: 2076 // - Rdst_value 2077 // - Rdst_address 2078 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, 2079 Register Rdst_address, 2080 Register Rindex) { 2081 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); 2082 subf(Rdst_address, Rdst_address, R18_locals); 2083 lfd(Rdst_value, -8, Rdst_address); 2084 } 2085 2086 // Store an int value at local variable slot Rindex. 2087 // Kills: 2088 // - Rindex 2089 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) { 2090 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2091 subf(Rindex, Rindex, R18_locals); 2092 stw(Rvalue, 0, Rindex); 2093 } 2094 2095 // Store a long value at local variable slot Rindex. 2096 // Kills: 2097 // - Rindex 2098 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) { 2099 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2100 subf(Rindex, Rindex, R18_locals); 2101 std(Rvalue, -8, Rindex); 2102 } 2103 2104 // Store an oop value at local variable slot Rindex. 2105 // Kills: 2106 // - Rindex 2107 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) { 2108 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2109 subf(Rindex, Rindex, R18_locals); 2110 std(Rvalue, 0, Rindex); 2111 } 2112 2113 // Store an int value at local variable slot Rindex. 2114 // Kills: 2115 // - Rindex 2116 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) { 2117 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2118 subf(Rindex, Rindex, R18_locals); 2119 stfs(Rvalue, 0, Rindex); 2120 } 2121 2122 // Store an int value at local variable slot Rindex. 2123 // Kills: 2124 // - Rindex 2125 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) { 2126 sldi(Rindex, Rindex, Interpreter::logStackElementSize); 2127 subf(Rindex, Rindex, R18_locals); 2128 stfd(Rvalue, -8, Rindex); 2129 } 2130 2131 // Read pending exception from thread and jump to interpreter. 2132 // Throw exception entry if one if pending. Fall through otherwise. 2133 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) { 2134 assert_different_registers(Rscratch1, Rscratch2, R3); 2135 Register Rexception = Rscratch1; 2136 Register Rtmp = Rscratch2; 2137 Label Ldone; 2138 // Get pending exception oop. 2139 ld(Rexception, thread_(pending_exception)); 2140 cmpdi(CCR0, Rexception, 0); 2141 beq(CCR0, Ldone); 2142 li(Rtmp, 0); 2143 mr_if_needed(R3, Rexception); 2144 std(Rtmp, thread_(pending_exception)); // Clear exception in thread 2145 if (Interpreter::rethrow_exception_entry() != NULL) { 2146 // Already got entry address. 2147 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry()); 2148 } else { 2149 // Dynamically load entry address. 2150 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true); 2151 ld(Rtmp, simm16_rest, Rtmp); 2152 } 2153 mtctr(Rtmp); 2154 save_interpreter_state(Rtmp); 2155 bctr(); 2156 2157 align(32, 12); 2158 bind(Ldone); 2159 } 2160 2161 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) { 2162 save_interpreter_state(R11_scratch1); 2163 2164 MacroAssembler::call_VM(oop_result, entry_point, false); 2165 2166 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); 2167 2168 check_and_handle_popframe(R11_scratch1); 2169 check_and_handle_earlyret(R11_scratch1); 2170 // Now check exceptions manually. 2171 if (check_exceptions) { 2172 check_and_forward_exception(R11_scratch1, R12_scratch2); 2173 } 2174 } 2175 2176 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2177 Register arg_1, bool check_exceptions) { 2178 // ARG1 is reserved for the thread. 2179 mr_if_needed(R4_ARG2, arg_1); 2180 call_VM(oop_result, entry_point, check_exceptions); 2181 } 2182 2183 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2184 Register arg_1, Register arg_2, 2185 bool check_exceptions) { 2186 // ARG1 is reserved for the thread. 2187 mr_if_needed(R4_ARG2, arg_1); 2188 assert(arg_2 != R4_ARG2, "smashed argument"); 2189 mr_if_needed(R5_ARG3, arg_2); 2190 call_VM(oop_result, entry_point, check_exceptions); 2191 } 2192 2193 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, 2194 Register arg_1, Register arg_2, Register arg_3, 2195 bool check_exceptions) { 2196 // ARG1 is reserved for the thread. 2197 mr_if_needed(R4_ARG2, arg_1); 2198 assert(arg_2 != R4_ARG2, "smashed argument"); 2199 mr_if_needed(R5_ARG3, arg_2); 2200 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument"); 2201 mr_if_needed(R6_ARG4, arg_3); 2202 call_VM(oop_result, entry_point, check_exceptions); 2203 } 2204 2205 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) { 2206 ld(scratch, 0, R1_SP); 2207 std(R15_esp, _ijava_state_neg(esp), scratch); 2208 std(R14_bcp, _ijava_state_neg(bcp), scratch); 2209 std(R26_monitor, _ijava_state_neg(monitors), scratch); 2210 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); } 2211 // Other entries should be unchanged. 2212 } 2213 2214 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) { 2215 ld(scratch, 0, R1_SP); 2216 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception). 2217 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code. 2218 if (!bcp_and_mdx_only) { 2219 // Following ones are Metadata. 2220 ld(R19_method, _ijava_state_neg(method), scratch); 2221 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch); 2222 // Following ones are stack addresses and don't require reload. 2223 ld(R15_esp, _ijava_state_neg(esp), scratch); 2224 ld(R18_locals, _ijava_state_neg(locals), scratch); 2225 ld(R26_monitor, _ijava_state_neg(monitors), scratch); 2226 } 2227 #ifdef ASSERT 2228 { 2229 Label Lok; 2230 subf(R0, R1_SP, scratch); 2231 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size); 2232 bge(CCR0, Lok); 2233 stop("frame too small (restore istate)", 0x5432); 2234 bind(Lok); 2235 } 2236 { 2237 Label Lok; 2238 ld(R0, _ijava_state_neg(ijava_reserved), scratch); 2239 cmpdi(CCR0, R0, 0x5afe); 2240 beq(CCR0, Lok); 2241 stop("frame corrupted (restore istate)", 0x5afe); 2242 bind(Lok); 2243 } 2244 #endif 2245 } 2246 2247 void InterpreterMacroAssembler::get_method_counters(Register method, 2248 Register Rcounters, 2249 Label& skip) { 2250 BLOCK_COMMENT("Load and ev. allocate counter object {"); 2251 Label has_counters; 2252 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2253 cmpdi(CCR0, Rcounters, 0); 2254 bne(CCR0, has_counters); 2255 call_VM(noreg, CAST_FROM_FN_PTR(address, 2256 InterpreterRuntime::build_method_counters), method, false); 2257 ld(Rcounters, in_bytes(Method::method_counters_offset()), method); 2258 cmpdi(CCR0, Rcounters, 0); 2259 beq(CCR0, skip); // No MethodCounters, OutOfMemory. 2260 BLOCK_COMMENT("} Load and ev. allocate counter object"); 2261 2262 bind(has_counters); 2263 } 2264 2265 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, 2266 Register iv_be_count, 2267 Register Rtmp_r0) { 2268 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 2269 Register invocation_count = iv_be_count; 2270 Register backedge_count = Rtmp_r0; 2271 int delta = InvocationCounter::count_increment; 2272 2273 // Load each counter in a register. 2274 // ld(inv_counter, Rtmp); 2275 // ld(be_counter, Rtmp2); 2276 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + 2277 InvocationCounter::counter_offset()); 2278 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + 2279 InvocationCounter::counter_offset()); 2280 2281 BLOCK_COMMENT("Increment profiling counters {"); 2282 2283 // Load the backedge counter. 2284 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int 2285 // Mask the backedge counter. 2286 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value); 2287 2288 // Load the invocation counter. 2289 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int 2290 // Add the delta to the invocation counter and store the result. 2291 addi(invocation_count, invocation_count, delta); 2292 // Store value. 2293 stw(invocation_count, inv_counter_offset, Rcounters); 2294 2295 // Add invocation counter + backedge counter. 2296 add(iv_be_count, backedge_count, invocation_count); 2297 2298 // Note that this macro must leave the backedge_count + invocation_count in 2299 // register iv_be_count! 2300 BLOCK_COMMENT("} Increment profiling counters"); 2301 } 2302 2303 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2304 if (state == atos) { MacroAssembler::verify_oop(reg); } 2305 } 2306 2307 // Local helper function for the verify_oop_or_return_address macro. 2308 static bool verify_return_address(Method* m, int bci) { 2309 #ifndef PRODUCT 2310 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci; 2311 // Assume it is a valid return address if it is inside m and is preceded by a jsr. 2312 if (!m->contains(pc)) return false; 2313 address jsr_pc; 2314 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); 2315 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; 2316 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); 2317 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; 2318 #endif // PRODUCT 2319 return false; 2320 } 2321 2322 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2323 if (VerifyFPU) { 2324 unimplemented("verfiyFPU"); 2325 } 2326 } 2327 2328 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { 2329 if (!VerifyOops) return; 2330 2331 // The VM documentation for the astore[_wide] bytecode allows 2332 // the TOS to be not only an oop but also a return address. 2333 Label test; 2334 Label skip; 2335 // See if it is an address (in the current method): 2336 2337 const int log2_bytecode_size_limit = 16; 2338 srdi_(Rtmp, reg, log2_bytecode_size_limit); 2339 bne(CCR0, test); 2340 2341 address fd = CAST_FROM_FN_PTR(address, verify_return_address); 2342 const int nbytes_save = MacroAssembler::num_volatile_regs * 8; 2343 save_volatile_gprs(R1_SP, -nbytes_save); // except R0 2344 save_LR_CR(Rtmp); // Save in old frame. 2345 push_frame_reg_args(nbytes_save, Rtmp); 2346 2347 load_const_optimized(Rtmp, fd, R0); 2348 mr_if_needed(R4_ARG2, reg); 2349 mr(R3_ARG1, R19_method); 2350 call_c(Rtmp); // call C 2351 2352 pop_frame(); 2353 restore_LR_CR(Rtmp); 2354 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0 2355 b(skip); 2356 2357 // Perform a more elaborate out-of-line call. 2358 // Not an address; verify it: 2359 bind(test); 2360 verify_oop(reg); 2361 bind(skip); 2362 } 2363 2364 // Inline assembly for: 2365 // 2366 // if (thread is in interp_only_mode) { 2367 // InterpreterRuntime::post_method_entry(); 2368 // } 2369 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) || 2370 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) { 2371 // SharedRuntime::jvmpi_method_entry(method, receiver); 2372 // } 2373 void InterpreterMacroAssembler::notify_method_entry() { 2374 // JVMTI 2375 // Whenever JVMTI puts a thread in interp_only_mode, method 2376 // entry/exit events are sent for that thread to track stack 2377 // depth. If it is possible to enter interp_only_mode we add 2378 // the code to check if the event should be sent. 2379 if (JvmtiExport::can_post_interpreter_events()) { 2380 Label jvmti_post_done; 2381 2382 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2383 cmpwi(CCR0, R0, 0); 2384 beq(CCR0, jvmti_post_done); 2385 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), 2386 /*check_exceptions=*/true); 2387 2388 bind(jvmti_post_done); 2389 } 2390 } 2391 2392 // Inline assembly for: 2393 // 2394 // if (thread is in interp_only_mode) { 2395 // // save result 2396 // InterpreterRuntime::post_method_exit(); 2397 // // restore result 2398 // } 2399 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) { 2400 // // save result 2401 // SharedRuntime::jvmpi_method_exit(); 2402 // // restore result 2403 // } 2404 // 2405 // Native methods have their result stored in d_tmp and l_tmp. 2406 // Java methods have their result stored in the expression stack. 2407 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state, 2408 NotifyMethodExitMode mode, bool check_exceptions) { 2409 // JVMTI 2410 // Whenever JVMTI puts a thread in interp_only_mode, method 2411 // entry/exit events are sent for that thread to track stack 2412 // depth. If it is possible to enter interp_only_mode we add 2413 // the code to check if the event should be sent. 2414 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2415 Label jvmti_post_done; 2416 2417 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); 2418 cmpwi(CCR0, R0, 0); 2419 beq(CCR0, jvmti_post_done); 2420 if (!is_native_method) { push(state); } // Expose tos to GC. 2421 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), 2422 /*check_exceptions=*/check_exceptions); 2423 if (!is_native_method) { pop(state); } 2424 2425 align(32, 12); 2426 bind(jvmti_post_done); 2427 } 2428 2429 // Dtrace support not implemented. 2430 } 2431