1 /* 2 * Copyright (c) 2016, 2024, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2016, 2024 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 // Major contributions by AHa, AS, JL, ML. 27 28 #include "precompiled.hpp" 29 #include "asm/macroAssembler.inline.hpp" 30 #include "gc/shared/barrierSet.hpp" 31 #include "gc/shared/barrierSetAssembler.hpp" 32 #include "interp_masm_s390.hpp" 33 #include "interpreter/interpreter.hpp" 34 #include "interpreter/interpreterRuntime.hpp" 35 #include "oops/arrayOop.hpp" 36 #include "oops/markWord.hpp" 37 #include "oops/methodCounters.hpp" 38 #include "oops/methodData.hpp" 39 #include "oops/resolvedFieldEntry.hpp" 40 #include "oops/resolvedIndyEntry.hpp" 41 #include "oops/resolvedMethodEntry.hpp" 42 #include "prims/jvmtiExport.hpp" 43 #include "prims/jvmtiThreadState.hpp" 44 #include "runtime/basicLock.hpp" 45 #include "runtime/frame.inline.hpp" 46 #include "runtime/javaThread.hpp" 47 #include "runtime/safepointMechanism.hpp" 48 #include "runtime/sharedRuntime.hpp" 49 #include "utilities/macros.hpp" 50 #include "utilities/powerOfTwo.hpp" 51 52 // Implementation of InterpreterMacroAssembler. 53 // This file specializes the assembler with interpreter-specific macros. 54 55 #ifdef PRODUCT 56 #define BLOCK_COMMENT(str) 57 #define BIND(label) bind(label); 58 #else 59 #define BLOCK_COMMENT(str) block_comment(str) 60 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") 61 #endif 62 63 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) { 64 assert(entry != nullptr, "Entry must have been generated by now"); 65 assert(Rscratch != Z_R0, "Can't use R0 for addressing"); 66 branch_optimized(Assembler::bcondAlways, entry); 67 } 68 69 void InterpreterMacroAssembler::empty_expression_stack(void) { 70 get_monitors(Z_R1_scratch); 71 add2reg(Z_esp, -Interpreter::stackElementSize, Z_R1_scratch); 72 } 73 74 // Dispatch code executed in the prolog of a bytecode which does not do it's 75 // own dispatch. 76 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 77 // On z/Architecture we are short on registers, therefore we do not preload the 78 // dispatch address of the next bytecode. 79 } 80 81 // Dispatch code executed in the epilog of a bytecode which does not do it's 82 // own dispatch. 83 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { 84 dispatch_next(state, step); 85 } 86 87 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) { 88 z_llgc(Z_bytecode, bcp_incr, Z_R0, Z_bcp); // Load next bytecode. 89 add2reg(Z_bcp, bcp_incr); // Advance bcp. Add2reg produces optimal code. 90 dispatch_base(state, Interpreter::dispatch_table(state), generate_poll); 91 } 92 93 // Common code to dispatch and dispatch_only. 94 // Dispatch value in Lbyte_code and increment Lbcp. 95 96 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table, bool generate_poll) { 97 verify_FPU(1, state); 98 99 #ifdef ASSERT 100 address reentry = nullptr; 101 { Label OK; 102 // Check if the frame pointer in Z_fp is correct. 103 z_cg(Z_fp, 0, Z_SP); 104 z_bre(OK); 105 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp: " FILE_AND_LINE); 106 bind(OK); 107 } 108 { Label OK; 109 // check if the locals pointer in Z_locals is correct 110 z_cg(Z_locals, _z_ijava_state_neg(locals), Z_fp); 111 z_bre(OK); 112 reentry = stop_chain_static(reentry, "invalid locals pointer Z_locals: " FILE_AND_LINE); 113 bind(OK); 114 } 115 #endif 116 117 // TODO: Maybe implement +VerifyActivationFrameSize here. 118 verify_oop(Z_tos, state); 119 120 // Dispatch table to use. 121 load_absolute_address(Z_tmp_1, (address)table); // Z_tmp_1 = table; 122 123 if (generate_poll) { 124 address *sfpt_tbl = Interpreter::safept_table(state); 125 if (table != sfpt_tbl) { 126 Label dispatch; 127 const Address poll_byte_addr(Z_thread, in_bytes(JavaThread::polling_word_offset()) + 7 /* Big Endian */); 128 // Armed page has poll_bit set, if poll bit is cleared just continue. 129 z_tm(poll_byte_addr, SafepointMechanism::poll_bit()); 130 z_braz(dispatch); 131 load_absolute_address(Z_tmp_1, (address)sfpt_tbl); // Z_tmp_1 = table; 132 bind(dispatch); 133 } 134 } 135 136 // 0 <= Z_bytecode < 256 => Use a 32 bit shift, because it is shorter than sllg. 137 // Z_bytecode must have been loaded zero-extended for this approach to be correct. 138 z_sll(Z_bytecode, LogBytesPerWord, Z_R0); // Multiply by wordSize. 139 z_lg(Z_tmp_1, 0, Z_bytecode, Z_tmp_1); // Get entry addr. 140 141 z_br(Z_tmp_1); 142 } 143 144 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) { 145 dispatch_base(state, Interpreter::dispatch_table(state), generate_poll); 146 } 147 148 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) { 149 dispatch_base(state, Interpreter::normal_table(state)); 150 } 151 152 void InterpreterMacroAssembler::dispatch_via(TosState state, address *table) { 153 // Load current bytecode. 154 z_llgc(Z_bytecode, Address(Z_bcp, (intptr_t)0)); 155 dispatch_base(state, table); 156 } 157 158 // The following call_VM*_base() methods overload and mask the respective 159 // declarations/definitions in class MacroAssembler. They are meant as a "detour" 160 // to perform additional, template interpreter specific tasks before actually 161 // calling their MacroAssembler counterparts. 162 163 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point) { 164 bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated. 165 // interpreter specific 166 // Note: No need to save/restore bcp (Z_R13) pointer since these are callee 167 // saved registers and no blocking/ GC can happen in leaf calls. 168 169 // super call 170 MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation); 171 } 172 173 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, bool allow_relocation) { 174 // interpreter specific 175 // Note: No need to save/restore bcp (Z_R13) pointer since these are callee 176 // saved registers and no blocking/ GC can happen in leaf calls. 177 178 // super call 179 MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation); 180 } 181 182 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp, 183 address entry_point, bool check_exceptions) { 184 bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated. 185 // interpreter specific 186 187 save_bcp(); 188 save_esp(); 189 // super call 190 MacroAssembler::call_VM_base(oop_result, last_java_sp, 191 entry_point, allow_relocation, check_exceptions); 192 restore_bcp(); 193 } 194 195 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp, 196 address entry_point, bool allow_relocation, 197 bool check_exceptions) { 198 // interpreter specific 199 200 save_bcp(); 201 save_esp(); 202 // super call 203 MacroAssembler::call_VM_base(oop_result, last_java_sp, 204 entry_point, allow_relocation, check_exceptions); 205 restore_bcp(); 206 } 207 208 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { 209 if (JvmtiExport::can_pop_frame()) { 210 BLOCK_COMMENT("check_and_handle_popframe {"); 211 Label L; 212 // Initiate popframe handling only if it is not already being 213 // processed. If the flag has the popframe_processing bit set, it 214 // means that this code is called *during* popframe handling - we 215 // don't want to reenter. 216 // TODO: Check if all four state combinations could be visible. 217 // If (processing and !pending) is an invisible/impossible state, 218 // there is optimization potential by testing both bits at once. 219 // Then, All_Zeroes and All_Ones means skip, Mixed means doit. 220 testbit(Address(Z_thread, JavaThread::popframe_condition_offset()), 221 exact_log2(JavaThread::popframe_pending_bit)); 222 z_bfalse(L); 223 testbit(Address(Z_thread, JavaThread::popframe_condition_offset()), 224 exact_log2(JavaThread::popframe_processing_bit)); 225 z_btrue(L); 226 227 // Call Interpreter::remove_activation_preserving_args_entry() to get the 228 // address of the same-named entrypoint in the generated interpreter code. 229 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 230 // The above call should (as its only effect) return the contents of the field 231 // _remove_activation_preserving_args_entry in Z_RET. 232 // We just jump there to have the work done. 233 z_br(Z_RET); 234 // There is no way for control to fall thru here. 235 236 bind(L); 237 BLOCK_COMMENT("} check_and_handle_popframe"); 238 } 239 } 240 241 242 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 243 Register RjvmtiState = Z_R1_scratch; 244 int tos_off = in_bytes(JvmtiThreadState::earlyret_tos_offset()); 245 int oop_off = in_bytes(JvmtiThreadState::earlyret_oop_offset()); 246 int val_off = in_bytes(JvmtiThreadState::earlyret_value_offset()); 247 int state_off = in_bytes(JavaThread::jvmti_thread_state_offset()); 248 249 z_lg(RjvmtiState, state_off, Z_thread); 250 251 switch (state) { 252 case atos: z_lg(Z_tos, oop_off, RjvmtiState); 253 store_const(Address(RjvmtiState, oop_off), 0L, 8, 8, Z_R0_scratch); 254 break; 255 case ltos: z_lg(Z_tos, val_off, RjvmtiState); break; 256 case btos: // fall through 257 case ztos: // fall through 258 case ctos: // fall through 259 case stos: // fall through 260 case itos: z_llgf(Z_tos, val_off, RjvmtiState); break; 261 case ftos: z_le(Z_ftos, val_off, RjvmtiState); break; 262 case dtos: z_ld(Z_ftos, val_off, RjvmtiState); break; 263 case vtos: /* nothing to do */ break; 264 default : ShouldNotReachHere(); 265 } 266 267 // Clean up tos value in the jvmti thread state. 268 store_const(Address(RjvmtiState, val_off), 0L, 8, 8, Z_R0_scratch); 269 // Set tos state field to illegal value. 270 store_const(Address(RjvmtiState, tos_off), ilgl, 4, 1, Z_R0_scratch); 271 } 272 273 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 274 if (JvmtiExport::can_force_early_return()) { 275 BLOCK_COMMENT("check_and_handle_earlyret {"); 276 Label L; 277 // arg regs are save, because we are just behind the call in call_VM_base 278 Register jvmti_thread_state = Z_ARG2; 279 Register tmp = Z_ARG3; 280 load_and_test_long(jvmti_thread_state, Address(Z_thread, JavaThread::jvmti_thread_state_offset())); 281 z_bre(L); // if (thread->jvmti_thread_state() == nullptr) exit; 282 283 // Initiate earlyret handling only if it is not already being processed. 284 // If the flag has the earlyret_processing bit set, it means that this code 285 // is called *during* earlyret handling - we don't want to reenter. 286 287 assert((JvmtiThreadState::earlyret_pending != 0) && (JvmtiThreadState::earlyret_inactive == 0), 288 "must fix this check, when changing the values of the earlyret enum"); 289 assert(JvmtiThreadState::earlyret_pending == 1, "must fix this check, when changing the values of the earlyret enum"); 290 291 load_and_test_int(tmp, Address(jvmti_thread_state, JvmtiThreadState::earlyret_state_offset())); 292 z_brz(L); // if (thread->jvmti_thread_state()->_earlyret_state != JvmtiThreadState::earlyret_pending) exit; 293 294 // Call Interpreter::remove_activation_early_entry() to get the address of the 295 // same-named entrypoint in the generated interpreter code. 296 assert(sizeof(TosState) == 4, "unexpected size"); 297 z_l(Z_ARG1, Address(jvmti_thread_state, JvmtiThreadState::earlyret_tos_offset())); 298 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Z_ARG1); 299 // The above call should (as its only effect) return the contents of the field 300 // _remove_activation_preserving_args_entry in Z_RET. 301 // We just jump there to have the work done. 302 z_br(Z_RET); 303 // There is no way for control to fall thru here. 304 305 bind(L); 306 BLOCK_COMMENT("} check_and_handle_earlyret"); 307 } 308 } 309 310 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2) { 311 lgr_if_needed(Z_ARG1, arg_1); 312 assert(arg_2 != Z_ARG1, "smashed argument"); 313 lgr_if_needed(Z_ARG2, arg_2); 314 MacroAssembler::call_VM_leaf_base(entry_point, true); 315 } 316 317 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, int bcp_offset, size_t index_size) { 318 Address param(Z_bcp, bcp_offset); 319 320 BLOCK_COMMENT("get_cache_index_at_bcp {"); 321 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 322 if (index_size == sizeof(u2)) { 323 load_sized_value(index, param, 2, false /*signed*/); 324 } else if (index_size == sizeof(u4)) { 325 326 load_sized_value(index, param, 4, false); 327 328 // Check if the secondary index definition is still ~x, otherwise 329 // we have to change the following assembler code to calculate the 330 // plain index. 331 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 332 not_(index); // Convert to plain index. 333 } else if (index_size == sizeof(u1)) { 334 z_llgc(index, param); 335 } else { 336 ShouldNotReachHere(); 337 } 338 BLOCK_COMMENT("}"); 339 } 340 341 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) { 342 // Get index out of bytecode pointer. 343 get_cache_index_at_bcp(index, 1, sizeof(u4)); 344 345 // Get the address of the ResolvedIndyEntry array 346 get_constant_pool_cache(cache); 347 z_lg(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset()))); 348 349 // Scale the index to form a byte offset into the ResolvedIndyEntry array 350 size_t entry_size = sizeof(ResolvedIndyEntry); 351 if (is_power_of_2(entry_size)) { 352 z_sllg(index, index, exact_log2(entry_size)); 353 } else { 354 z_mghi(index, entry_size); 355 } 356 357 // Calculate the final field address. 358 z_la(cache, Array<ResolvedIndyEntry>::base_offset_in_bytes(), index, cache); 359 } 360 361 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) { 362 // Get field index out of bytecode pointer. 363 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2)); 364 365 // Get the address of the ResolvedFieldEntry array. 366 get_constant_pool_cache(cache); 367 z_lg(cache, Address(cache, in_bytes(ConstantPoolCache::field_entries_offset()))); 368 369 // Scale the index to form a byte offset into the ResolvedFieldEntry array 370 size_t entry_size = sizeof(ResolvedFieldEntry); 371 if (is_power_of_2(entry_size)) { 372 z_sllg(index, index, exact_log2(entry_size)); 373 } else { 374 z_mghi(index, entry_size); 375 } 376 377 // Calculate the final field address. 378 z_la(cache, Array<ResolvedFieldEntry>::base_offset_in_bytes(), index, cache); 379 } 380 381 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) { 382 // Get field index out of bytecode pointer. 383 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2)); 384 385 // Get the address of the ResolvedMethodEntry array. 386 get_constant_pool_cache(cache); 387 z_lg(cache, Address(cache, in_bytes(ConstantPoolCache::method_entries_offset()))); 388 389 // Scale the index to form a byte offset into the ResolvedMethodEntry array 390 size_t entry_size = sizeof(ResolvedMethodEntry); 391 if (is_power_of_2(entry_size)) { 392 z_sllg(index, index, exact_log2(entry_size)); 393 } else { 394 z_mghi(index, entry_size); 395 } 396 397 // Calculate the final field address. 398 z_la(cache, Array<ResolvedMethodEntry>::base_offset_in_bytes(), index, cache); 399 } 400 401 // Load object from cpool->resolved_references(index). 402 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) { 403 assert_different_registers(result, index); 404 get_constant_pool(result); 405 406 // Convert 407 // - from field index to resolved_references() index and 408 // - from word index to byte offset. 409 // Since this is a java object, it is potentially compressed. 410 Register tmp = index; // reuse 411 z_sllg(index, index, LogBytesPerHeapOop); // Offset into resolved references array. 412 // Load pointer for resolved_references[] objArray. 413 z_lg(result, in_bytes(ConstantPool::cache_offset()), result); 414 z_lg(result, in_bytes(ConstantPoolCache::resolved_references_offset()), result); 415 resolve_oop_handle(result); // Load resolved references array itself. 416 #ifdef ASSERT 417 NearLabel index_ok; 418 z_lgf(Z_R0, Address(result, arrayOopDesc::length_offset_in_bytes())); 419 z_sllg(Z_R0, Z_R0, LogBytesPerHeapOop); 420 compare64_and_branch(tmp, Z_R0, Assembler::bcondLow, index_ok); 421 stop("resolved reference index out of bounds", 0x09256); 422 bind(index_ok); 423 #endif 424 z_agr(result, index); // Address of indexed array element. 425 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp, noreg); 426 } 427 428 // load cpool->resolved_klass_at(index) 429 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register cpool, Register offset, Register iklass) { 430 // int value = *(Rcpool->int_at_addr(which)); 431 // int resolved_klass_index = extract_low_short_from_int(value); 432 z_llgh(offset, Address(cpool, offset, sizeof(ConstantPool) + 2)); // offset = resolved_klass_index (s390 is big-endian) 433 z_sllg(offset, offset, LogBytesPerWord); // Convert 'index' to 'offset' 434 z_lg(iklass, Address(cpool, ConstantPool::resolved_klasses_offset())); // iklass = cpool->_resolved_klasses 435 z_lg(iklass, Address(iklass, offset, Array<Klass*>::base_offset_in_bytes())); 436 } 437 438 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 439 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. 440 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 441 Register Rsuper_klass, 442 Register Rtmp1, 443 Register Rtmp2, 444 Label &ok_is_subtype) { 445 // Profile the not-null value's klass. 446 profile_typecheck(Rtmp1, Rsub_klass, Rtmp2); 447 448 // Do the check. 449 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype); 450 } 451 452 // Pop topmost element from stack. It just disappears. 453 // Useful if consumed previously by access via stackTop(). 454 void InterpreterMacroAssembler::popx(int len) { 455 add2reg(Z_esp, len*Interpreter::stackElementSize); 456 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 457 } 458 459 // Get Address object of stack top. No checks. No pop. 460 // Purpose: - Provide address of stack operand to exploit reg-mem operations. 461 // - Avoid RISC-like mem2reg - reg-reg-op sequence. 462 Address InterpreterMacroAssembler::stackTop() { 463 return Address(Z_esp, Interpreter::expr_offset_in_bytes(0)); 464 } 465 466 void InterpreterMacroAssembler::pop_i(Register r) { 467 z_l(r, Interpreter::expr_offset_in_bytes(0), Z_esp); 468 add2reg(Z_esp, Interpreter::stackElementSize); 469 assert_different_registers(r, Z_R1_scratch); 470 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 471 } 472 473 void InterpreterMacroAssembler::pop_ptr(Register r) { 474 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp); 475 add2reg(Z_esp, Interpreter::stackElementSize); 476 assert_different_registers(r, Z_R1_scratch); 477 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 478 } 479 480 void InterpreterMacroAssembler::pop_l(Register r) { 481 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp); 482 add2reg(Z_esp, 2*Interpreter::stackElementSize); 483 assert_different_registers(r, Z_R1_scratch); 484 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 485 } 486 487 void InterpreterMacroAssembler::pop_f(FloatRegister f) { 488 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), false); 489 add2reg(Z_esp, Interpreter::stackElementSize); 490 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 491 } 492 493 void InterpreterMacroAssembler::pop_d(FloatRegister f) { 494 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), true); 495 add2reg(Z_esp, 2*Interpreter::stackElementSize); 496 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 497 } 498 499 void InterpreterMacroAssembler::push_i(Register r) { 500 assert_different_registers(r, Z_R1_scratch); 501 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 502 z_st(r, Address(Z_esp)); 503 add2reg(Z_esp, -Interpreter::stackElementSize); 504 } 505 506 void InterpreterMacroAssembler::push_ptr(Register r) { 507 z_stg(r, Address(Z_esp)); 508 add2reg(Z_esp, -Interpreter::stackElementSize); 509 } 510 511 void InterpreterMacroAssembler::push_l(Register r) { 512 assert_different_registers(r, Z_R1_scratch); 513 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 514 int offset = -Interpreter::stackElementSize; 515 z_stg(r, Address(Z_esp, offset)); 516 clear_mem(Address(Z_esp), Interpreter::stackElementSize); 517 add2reg(Z_esp, 2 * offset); 518 } 519 520 void InterpreterMacroAssembler::push_f(FloatRegister f) { 521 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 522 freg2mem_opt(f, Address(Z_esp), false); 523 add2reg(Z_esp, -Interpreter::stackElementSize); 524 } 525 526 void InterpreterMacroAssembler::push_d(FloatRegister d) { 527 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 528 int offset = -Interpreter::stackElementSize; 529 freg2mem_opt(d, Address(Z_esp, offset)); 530 add2reg(Z_esp, 2 * offset); 531 } 532 533 void InterpreterMacroAssembler::push(TosState state) { 534 verify_oop(Z_tos, state); 535 switch (state) { 536 case atos: push_ptr(); break; 537 case btos: push_i(); break; 538 case ztos: 539 case ctos: 540 case stos: push_i(); break; 541 case itos: push_i(); break; 542 case ltos: push_l(); break; 543 case ftos: push_f(); break; 544 case dtos: push_d(); break; 545 case vtos: /* nothing to do */ break; 546 default : ShouldNotReachHere(); 547 } 548 } 549 550 void InterpreterMacroAssembler::pop(TosState state) { 551 switch (state) { 552 case atos: pop_ptr(Z_tos); break; 553 case btos: pop_i(Z_tos); break; 554 case ztos: 555 case ctos: 556 case stos: pop_i(Z_tos); break; 557 case itos: pop_i(Z_tos); break; 558 case ltos: pop_l(Z_tos); break; 559 case ftos: pop_f(Z_ftos); break; 560 case dtos: pop_d(Z_ftos); break; 561 case vtos: /* nothing to do */ break; 562 default : ShouldNotReachHere(); 563 } 564 verify_oop(Z_tos, state); 565 } 566 567 // Helpers for swap and dup. 568 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 569 z_lg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n))); 570 } 571 572 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 573 z_stg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n))); 574 } 575 576 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted(Register method) { 577 // Satisfy interpreter calling convention (see generate_normal_entry()). 578 z_lgr(Z_R10, Z_SP); // Set sender sp (aka initial caller sp, aka unextended sp). 579 // Record top_frame_sp, because the callee might modify it, if it's compiled. 580 z_stg(Z_SP, _z_ijava_state_neg(top_frame_sp), Z_fp); 581 save_bcp(); 582 save_esp(); 583 z_lgr(Z_method, method); // Set Z_method (kills Z_fp!). 584 } 585 586 // Jump to from_interpreted entry of a call unless single stepping is possible 587 // in this thread in which case we must call the i2i entry. 588 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { 589 assert_different_registers(method, Z_R10 /*used for initial_caller_sp*/, temp); 590 prepare_to_jump_from_interpreted(method); 591 592 if (JvmtiExport::can_post_interpreter_events()) { 593 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 594 // compiled code in threads for which the event is enabled. Check here for 595 // interp_only_mode if these events CAN be enabled. 596 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset())); 597 MacroAssembler::load_and_test_int(Z_R0_scratch, Address(Z_thread, JavaThread::interp_only_mode_offset())); 598 z_bcr(bcondEqual, Z_R1_scratch); // Run compiled code if zero. 599 // Run interpreted. 600 z_lg(Z_R1_scratch, Address(method, Method::interpreter_entry_offset())); 601 z_br(Z_R1_scratch); 602 } else { 603 // Run compiled code. 604 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset())); 605 z_br(Z_R1_scratch); 606 } 607 } 608 609 #ifdef ASSERT 610 void InterpreterMacroAssembler::verify_esp(Register Resp, Register Rtemp) { 611 // About to read or write Resp[0]. 612 // Make sure it is not in the monitors or the TOP_IJAVA_FRAME_ABI. 613 address reentry = nullptr; 614 615 { 616 // Check if the frame pointer in Z_fp is correct. 617 NearLabel OK; 618 z_cg(Z_fp, 0, Z_SP); 619 z_bre(OK); 620 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp"); 621 bind(OK); 622 } 623 { 624 // Resp must not point into or below the operand stack, 625 // i.e. IJAVA_STATE.monitors > Resp. 626 NearLabel OK; 627 Register Rmonitors = Rtemp; 628 z_lg(Rmonitors, _z_ijava_state_neg(monitors), Z_fp); 629 compareU64_and_branch(Rmonitors, Resp, bcondHigh, OK); 630 reentry = stop_chain_static(reentry, "too many pops: Z_esp points into monitor area"); 631 bind(OK); 632 } 633 { 634 // Resp may point to the last word of TOP_IJAVA_FRAME_ABI, but not below 635 // i.e. !(Z_SP + frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize > Resp). 636 NearLabel OK; 637 Register Rabi_bottom = Rtemp; 638 add2reg(Rabi_bottom, frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize, Z_SP); 639 compareU64_and_branch(Rabi_bottom, Resp, bcondNotHigh, OK); 640 reentry = stop_chain_static(reentry, "too many pushes: Z_esp points into TOP_IJAVA_FRAME_ABI"); 641 bind(OK); 642 } 643 } 644 645 void InterpreterMacroAssembler::asm_assert_ijava_state_magic(Register tmp) { 646 Label magic_ok; 647 load_const_optimized(tmp, frame::z_istate_magic_number); 648 z_cg(tmp, Address(Z_fp, _z_ijava_state_neg(magic))); 649 z_bre(magic_ok); 650 stop_static("error: wrong magic number in ijava_state access"); 651 bind(magic_ok); 652 } 653 #endif // ASSERT 654 655 void InterpreterMacroAssembler::save_bcp() { 656 z_stg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))); 657 asm_assert_ijava_state_magic(Z_bcp); 658 NOT_PRODUCT(z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)))); 659 } 660 661 void InterpreterMacroAssembler::restore_bcp() { 662 asm_assert_ijava_state_magic(Z_bcp); 663 z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))); 664 } 665 666 void InterpreterMacroAssembler::save_esp() { 667 z_stg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp))); 668 } 669 670 void InterpreterMacroAssembler::restore_esp() { 671 asm_assert_ijava_state_magic(Z_esp); 672 z_lg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp))); 673 } 674 675 void InterpreterMacroAssembler::get_monitors(Register reg) { 676 asm_assert_ijava_state_magic(reg); 677 mem2reg_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors))); 678 } 679 680 void InterpreterMacroAssembler::save_monitors(Register reg) { 681 reg2mem_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors))); 682 } 683 684 void InterpreterMacroAssembler::get_mdp(Register mdp) { 685 z_lg(mdp, _z_ijava_state_neg(mdx), Z_fp); 686 } 687 688 void InterpreterMacroAssembler::save_mdp(Register mdp) { 689 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp); 690 } 691 692 // Values that are only read (besides initialization). 693 void InterpreterMacroAssembler::restore_locals() { 694 asm_assert_ijava_state_magic(Z_locals); 695 z_lg(Z_locals, Address(Z_fp, _z_ijava_state_neg(locals))); 696 } 697 698 void InterpreterMacroAssembler::get_method(Register reg) { 699 asm_assert_ijava_state_magic(reg); 700 z_lg(reg, Address(Z_fp, _z_ijava_state_neg(method))); 701 } 702 703 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(Register Rdst, int bcp_offset, 704 signedOrNot is_signed) { 705 // Rdst is an 8-byte return value!!! 706 707 // Unaligned loads incur only a small penalty on z/Architecture. The penalty 708 // is a few (2..3) ticks, even when the load crosses a cache line 709 // boundary. In case of a cache miss, the stall could, of course, be 710 // much longer. 711 712 switch (is_signed) { 713 case Signed: 714 z_lgh(Rdst, bcp_offset, Z_R0, Z_bcp); 715 break; 716 case Unsigned: 717 z_llgh(Rdst, bcp_offset, Z_R0, Z_bcp); 718 break; 719 default: 720 ShouldNotReachHere(); 721 } 722 } 723 724 725 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(Register Rdst, int bcp_offset, 726 setCCOrNot set_cc) { 727 // Rdst is an 8-byte return value!!! 728 729 // Unaligned loads incur only a small penalty on z/Architecture. The penalty 730 // is a few (2..3) ticks, even when the load crosses a cache line 731 // boundary. In case of a cache miss, the stall could, of course, be 732 // much longer. 733 734 // Both variants implement a sign-extending int2long load. 735 if (set_cc == set_CC) { 736 load_and_test_int2long(Rdst, Address(Z_bcp, (intptr_t)bcp_offset)); 737 } else { 738 mem2reg_signed_opt( Rdst, Address(Z_bcp, (intptr_t)bcp_offset)); 739 } 740 } 741 742 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 743 get_method(Rdst); 744 mem2reg_opt(Rdst, Address(Rdst, Method::const_offset())); 745 mem2reg_opt(Rdst, Address(Rdst, ConstMethod::constants_offset())); 746 } 747 748 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { 749 get_constant_pool(Rdst); 750 mem2reg_opt(Rdst, Address(Rdst, ConstantPool::cache_offset())); 751 } 752 753 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 754 get_constant_pool(Rcpool); 755 mem2reg_opt(Rtags, Address(Rcpool, ConstantPool::tags_offset())); 756 } 757 758 // Unlock if synchronized method. 759 // 760 // Unlock the receiver if this is a synchronized method. 761 // Unlock any Java monitors from synchronized blocks. 762 // 763 // If there are locked Java monitors 764 // If throw_monitor_exception 765 // throws IllegalMonitorStateException 766 // Else if install_monitor_exception 767 // installs IllegalMonitorStateException 768 // Else 769 // no error processing 770 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 771 bool throw_monitor_exception, 772 bool install_monitor_exception) { 773 NearLabel unlocked, unlock, no_unlock; 774 775 { 776 Register R_method = Z_ARG2; 777 Register R_do_not_unlock_if_synchronized = Z_ARG3; 778 779 // Get the value of _do_not_unlock_if_synchronized into G1_scratch. 780 const Address do_not_unlock_if_synchronized(Z_thread, 781 JavaThread::do_not_unlock_if_synchronized_offset()); 782 load_sized_value(R_do_not_unlock_if_synchronized, do_not_unlock_if_synchronized, 1, false /*unsigned*/); 783 z_mvi(do_not_unlock_if_synchronized, false); // Reset the flag. 784 785 // Check if synchronized method. 786 get_method(R_method); 787 verify_oop(Z_tos, state); 788 push(state); // Save tos/result. 789 testbit(method2_(R_method, access_flags), JVM_ACC_SYNCHRONIZED_BIT); 790 z_bfalse(unlocked); 791 792 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 793 // is set. 794 compareU64_and_branch(R_do_not_unlock_if_synchronized, (intptr_t)0L, bcondNotEqual, no_unlock); 795 } 796 797 // unlock monitor 798 799 // BasicObjectLock will be first in list, since this is a 800 // synchronized method. However, need to check that the object has 801 // not been unlocked by an explicit monitorexit bytecode. 802 const Address monitor(Z_fp, -(frame::z_ijava_state_size + (int) sizeof(BasicObjectLock))); 803 // We use Z_ARG2 so that if we go slow path it will be the correct 804 // register for unlock_object to pass to VM directly. 805 load_address(Z_ARG2, monitor); // Address of first monitor. 806 z_lg(Z_ARG3, Address(Z_ARG2, BasicObjectLock::obj_offset())); 807 compareU64_and_branch(Z_ARG3, (intptr_t)0L, bcondNotEqual, unlock); 808 809 if (throw_monitor_exception) { 810 // Entry already unlocked need to throw an exception. 811 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 812 should_not_reach_here(); 813 } else { 814 // Monitor already unlocked during a stack unroll. 815 // If requested, install an illegal_monitor_state_exception. 816 // Continue with stack unrolling. 817 if (install_monitor_exception) { 818 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 819 } 820 z_bru(unlocked); 821 } 822 823 bind(unlock); 824 825 unlock_object(Z_ARG2); 826 827 bind(unlocked); 828 829 // I0, I1: Might contain return value 830 831 // Check that all monitors are unlocked. 832 { 833 NearLabel loop, exception, entry, restart; 834 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes(); 835 // We use Z_ARG2 so that if we go slow path it will be the correct 836 // register for unlock_object to pass to VM directly. 837 Register R_current_monitor = Z_ARG2; 838 Register R_monitor_block_bot = Z_ARG1; 839 const Address monitor_block_top(Z_fp, _z_ijava_state_neg(monitors)); 840 const Address monitor_block_bot(Z_fp, -frame::z_ijava_state_size); 841 842 bind(restart); 843 // Starting with top-most entry. 844 z_lg(R_current_monitor, monitor_block_top); 845 // Points to word before bottom of monitor block. 846 load_address(R_monitor_block_bot, monitor_block_bot); 847 z_bru(entry); 848 849 // Entry already locked, need to throw exception. 850 bind(exception); 851 852 if (throw_monitor_exception) { 853 // Throw exception. 854 MacroAssembler::call_VM(noreg, 855 CAST_FROM_FN_PTR(address, InterpreterRuntime:: 856 throw_illegal_monitor_state_exception)); 857 should_not_reach_here(); 858 } else { 859 // Stack unrolling. Unlock object and install illegal_monitor_exception. 860 // Unlock does not block, so don't have to worry about the frame. 861 // We don't have to preserve c_rarg1 since we are going to throw an exception. 862 unlock_object(R_current_monitor); 863 if (install_monitor_exception) { 864 call_VM(noreg, CAST_FROM_FN_PTR(address, 865 InterpreterRuntime:: 866 new_illegal_monitor_state_exception)); 867 } 868 z_bru(restart); 869 } 870 871 bind(loop); 872 // Check if current entry is used. 873 load_and_test_long(Z_R0_scratch, Address(R_current_monitor, BasicObjectLock::obj_offset())); 874 z_brne(exception); 875 876 add2reg(R_current_monitor, entry_size); // Otherwise advance to next entry. 877 bind(entry); 878 compareU64_and_branch(R_current_monitor, R_monitor_block_bot, bcondNotEqual, loop); 879 } 880 881 bind(no_unlock); 882 pop(state); 883 verify_oop(Z_tos, state); 884 } 885 886 void InterpreterMacroAssembler::narrow(Register result, Register ret_type) { 887 get_method(ret_type); 888 z_lg(ret_type, Address(ret_type, in_bytes(Method::const_offset()))); 889 z_lb(ret_type, Address(ret_type, in_bytes(ConstMethod::result_type_offset()))); 890 891 Label notBool, notByte, notChar, done; 892 893 // common case first 894 compareU32_and_branch(ret_type, T_INT, bcondEqual, done); 895 896 compareU32_and_branch(ret_type, T_BOOLEAN, bcondNotEqual, notBool); 897 z_nilf(result, 0x1); 898 z_bru(done); 899 900 bind(notBool); 901 compareU32_and_branch(ret_type, T_BYTE, bcondNotEqual, notByte); 902 z_lbr(result, result); 903 z_bru(done); 904 905 bind(notByte); 906 compareU32_and_branch(ret_type, T_CHAR, bcondNotEqual, notChar); 907 z_nilf(result, 0xffff); 908 z_bru(done); 909 910 bind(notChar); 911 // compareU32_and_branch(ret_type, T_SHORT, bcondNotEqual, notShort); 912 z_lhr(result, result); 913 914 // Nothing to do for T_INT 915 bind(done); 916 } 917 918 // remove activation 919 // 920 // Unlock the receiver if this is a synchronized method. 921 // Unlock any Java monitors from synchronized blocks. 922 // Remove the activation from the stack. 923 // 924 // If there are locked Java monitors 925 // If throw_monitor_exception 926 // throws IllegalMonitorStateException 927 // Else if install_monitor_exception 928 // installs IllegalMonitorStateException 929 // Else 930 // no error processing 931 void InterpreterMacroAssembler::remove_activation(TosState state, 932 Register return_pc, 933 bool throw_monitor_exception, 934 bool install_monitor_exception, 935 bool notify_jvmti) { 936 BLOCK_COMMENT("remove_activation {"); 937 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 938 939 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 940 notify_method_exit(false, state, notify_jvmti ? NotifyJVMTI : SkipNotifyJVMTI); 941 942 if (StackReservedPages > 0) { 943 BLOCK_COMMENT("reserved_stack_check:"); 944 // Test if reserved zone needs to be enabled. 945 Label no_reserved_zone_enabling; 946 947 // check if already enabled - if so no re-enabling needed 948 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size"); 949 z_ly(Z_R0, Address(Z_thread, JavaThread::stack_guard_state_offset())); 950 compare32_and_branch(Z_R0, StackOverflow::stack_guard_enabled, bcondEqual, no_reserved_zone_enabling); 951 952 // Compare frame pointers. There is no good stack pointer, as with stack 953 // frame compression we can get different SPs when we do calls. A subsequent 954 // call could have a smaller SP, so that this compare succeeds for an 955 // inner call of the method annotated with ReservedStack. 956 z_lg(Z_R0, Address(Z_SP, (intptr_t)_z_abi(callers_sp))); 957 z_clg(Z_R0, Address(Z_thread, JavaThread::reserved_stack_activation_offset())); // Compare with frame pointer in memory. 958 z_brl(no_reserved_zone_enabling); 959 960 // Enable reserved zone again, throw stack overflow exception. 961 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), Z_thread); 962 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError)); 963 964 should_not_reach_here(); 965 966 bind(no_reserved_zone_enabling); 967 } 968 969 verify_oop(Z_tos, state); 970 971 pop_interpreter_frame(return_pc, Z_ARG2, Z_ARG3); 972 BLOCK_COMMENT("} remove_activation"); 973 } 974 975 // lock object 976 // 977 // Registers alive 978 // monitor (Z_R10) - Address of the BasicObjectLock to be used for locking, 979 // which must be initialized with the object to lock. 980 // object (Z_R11, Z_R2) - Address of the object to be locked. 981 // templateTable (monitorenter) is using Z_R2 for object 982 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 983 984 if (LockingMode == LM_MONITOR) { 985 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor); 986 return; 987 } 988 989 // template code: (for LM_LEGACY) 990 // 991 // markWord displaced_header = obj->mark().set_unlocked(); 992 // monitor->lock()->set_displaced_header(displaced_header); 993 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) { 994 // // We stored the monitor address into the object's mark word. 995 // } else if (THREAD->is_lock_owned((address)displaced_header)) 996 // // Simple recursive case. 997 // monitor->lock()->set_displaced_header(nullptr); 998 // } else { 999 // // Slow path. 1000 // InterpreterRuntime::monitorenter(THREAD, monitor); 1001 // } 1002 1003 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 1004 1005 const Register header = Z_ARG5; 1006 const Register object_mark_addr = Z_ARG4; 1007 const Register current_header = Z_ARG5; 1008 const Register tmp = Z_R1_scratch; 1009 1010 NearLabel done, slow_case; 1011 1012 // markWord header = obj->mark().set_unlocked(); 1013 1014 // Load markWord from object into header. 1015 z_lg(header, hdr_offset, object); 1016 1017 if (DiagnoseSyncOnValueBasedClasses != 0) { 1018 load_klass(tmp, object); 1019 testbit(Address(tmp, Klass::access_flags_offset()), exact_log2(JVM_ACC_IS_VALUE_BASED_CLASS)); 1020 z_btrue(slow_case); 1021 } 1022 1023 if (LockingMode == LM_LIGHTWEIGHT) { 1024 lightweight_lock(object, /* mark word */ header, tmp, slow_case); 1025 } else if (LockingMode == LM_LEGACY) { 1026 1027 // Set header to be (markWord of object | UNLOCK_VALUE). 1028 // This will not change anything if it was unlocked before. 1029 z_oill(header, markWord::unlocked_value); 1030 1031 // monitor->lock()->set_displaced_header(displaced_header); 1032 const int lock_offset = in_bytes(BasicObjectLock::lock_offset()); 1033 const int mark_offset = lock_offset + BasicLock::displaced_header_offset_in_bytes(); 1034 1035 // Initialize the box (Must happen before we update the object mark!). 1036 z_stg(header, mark_offset, monitor); 1037 1038 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) { 1039 1040 // not necessary, use offset in instruction directly. 1041 // add2reg(object_mark_addr, hdr_offset, object); 1042 1043 // Store stack address of the BasicObjectLock (this is monitor) into object. 1044 z_csg(header, monitor, hdr_offset, object); 1045 assert(current_header == header, 1046 "must be same register"); // Identified two registers from z/Architecture. 1047 1048 z_bre(done); 1049 1050 // } else if (THREAD->is_lock_owned((address)displaced_header)) 1051 // // Simple recursive case. 1052 // monitor->lock()->set_displaced_header(nullptr); 1053 1054 // We did not see an unlocked object so try the fast recursive case. 1055 1056 // Check if owner is self by comparing the value in the markWord of object 1057 // (current_header) with the stack pointer. 1058 z_sgr(current_header, Z_SP); 1059 1060 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 1061 1062 // The prior sequence "LGR, NGR, LTGR" can be done better 1063 // (Z_R1 is temp and not used after here). 1064 load_const_optimized(Z_R0, (~(os::vm_page_size() - 1) | markWord::lock_mask_in_place)); 1065 z_ngr(Z_R0, current_header); // AND sets CC (result eq/ne 0) 1066 1067 // If condition is true we are done and hence we can store 0 in the displaced 1068 // header indicating it is a recursive lock and be done. 1069 z_brne(slow_case); 1070 z_release(); // Member unnecessary on zarch AND because the above csg does a sync before and after. 1071 z_stg(Z_R0/*==0!*/, mark_offset, monitor); 1072 } 1073 z_bru(done); 1074 // } else { 1075 // // Slow path. 1076 // InterpreterRuntime::monitorenter(THREAD, monitor); 1077 1078 // None of the above fast optimizations worked so we have to get into the 1079 // slow case of monitor enter. 1080 bind(slow_case); 1081 if (LockingMode == LM_LIGHTWEIGHT) { 1082 // for lightweight locking we need to use monitorenter_obj, see interpreterRuntime.cpp 1083 call_VM(noreg, 1084 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj), 1085 object); 1086 } else { 1087 call_VM(noreg, 1088 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 1089 monitor); 1090 } 1091 // } 1092 1093 bind(done); 1094 } 1095 1096 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 1097 // 1098 // Registers alive 1099 // monitor - address of the BasicObjectLock to be used for locking, 1100 // which must be initialized with the object to lock. 1101 // 1102 // Throw IllegalMonitorException if object is not locked by current thread. 1103 void InterpreterMacroAssembler::unlock_object(Register monitor, Register object) { 1104 1105 if (LockingMode == LM_MONITOR) { 1106 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor); 1107 return; 1108 } 1109 1110 // else { 1111 // template code: (for LM_LEGACY): 1112 // 1113 // if ((displaced_header = monitor->displaced_header()) == nullptr) { 1114 // // Recursive unlock. Mark the monitor unlocked by setting the object field to null. 1115 // monitor->set_obj(nullptr); 1116 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) { 1117 // // We swapped the unlocked mark in displaced_header into the object's mark word. 1118 // monitor->set_obj(nullptr); 1119 // } else { 1120 // // Slow path. 1121 // InterpreterRuntime::monitorexit(monitor); 1122 // } 1123 1124 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 1125 1126 const Register header = Z_ARG4; 1127 const Register current_header = Z_R1_scratch; 1128 Address obj_entry(monitor, BasicObjectLock::obj_offset()); 1129 Label done, slow_case; 1130 1131 if (object == noreg) { 1132 // In the template interpreter, we must assure that the object 1133 // entry in the monitor is cleared on all paths. Thus we move 1134 // loading up to here, and clear the entry afterwards. 1135 object = Z_ARG3; // Use Z_ARG3 if caller didn't pass object. 1136 z_lg(object, obj_entry); 1137 } 1138 1139 assert_different_registers(monitor, object, header, current_header); 1140 1141 // if ((displaced_header = monitor->displaced_header()) == nullptr) { 1142 // // Recursive unlock. Mark the monitor unlocked by setting the object field to null. 1143 // monitor->set_obj(nullptr); 1144 1145 // monitor->lock()->set_displaced_header(displaced_header); 1146 const int lock_offset = in_bytes(BasicObjectLock::lock_offset()); 1147 const int mark_offset = lock_offset + BasicLock::displaced_header_offset_in_bytes(); 1148 1149 clear_mem(obj_entry, sizeof(oop)); 1150 if (LockingMode != LM_LIGHTWEIGHT) { 1151 // Test first if we are in the fast recursive case. 1152 MacroAssembler::load_and_test_long(header, Address(monitor, mark_offset)); 1153 z_bre(done); // header == 0 -> goto done 1154 } 1155 1156 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) { 1157 // // We swapped the unlocked mark in displaced_header into the object's mark word. 1158 // monitor->set_obj(nullptr); 1159 1160 // If we still have a lightweight lock, unlock the object and be done. 1161 if (LockingMode == LM_LIGHTWEIGHT) { 1162 // Check for non-symmetric locking. This is allowed by the spec and the interpreter 1163 // must handle it. 1164 1165 Register tmp = current_header; 1166 1167 // First check for lock-stack underflow. 1168 z_lgf(tmp, Address(Z_thread, JavaThread::lock_stack_top_offset())); 1169 compareU32_and_branch(tmp, (unsigned)LockStack::start_offset(), Assembler::bcondNotHigh, slow_case); 1170 1171 // Then check if the top of the lock-stack matches the unlocked object. 1172 z_aghi(tmp, -oopSize); 1173 z_lg(tmp, Address(Z_thread, tmp)); 1174 compare64_and_branch(tmp, object, Assembler::bcondNotEqual, slow_case); 1175 1176 z_lg(header, Address(object, hdr_offset)); 1177 z_lgr(tmp, header); 1178 z_nill(tmp, markWord::monitor_value); 1179 z_brne(slow_case); 1180 1181 lightweight_unlock(object, header, tmp, slow_case); 1182 1183 z_bru(done); 1184 } else { 1185 // The markword is expected to be at offset 0. 1186 // This is not required on s390, at least not here. 1187 assert(hdr_offset == 0, "unlock_object: review code below"); 1188 1189 // We have the displaced header in header. If the lock is still 1190 // lightweight, it will contain the monitor address and we'll store the 1191 // displaced header back into the object's mark word. 1192 z_lgr(current_header, monitor); 1193 z_csg(current_header, header, hdr_offset, object); 1194 z_bre(done); 1195 } 1196 1197 // } else { 1198 // // Slow path. 1199 // InterpreterRuntime::monitorexit(monitor); 1200 1201 // The lock has been converted into a heavy lock and hence 1202 // we need to get into the slow case. 1203 bind(slow_case); 1204 z_stg(object, obj_entry); // Restore object entry, has been cleared above. 1205 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor); 1206 1207 // } 1208 1209 bind(done); 1210 } 1211 1212 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) { 1213 assert(ProfileInterpreter, "must be profiling interpreter"); 1214 load_and_test_long(mdp, Address(Z_fp, _z_ijava_state_neg(mdx))); 1215 z_brz(zero_continue); 1216 } 1217 1218 // Set the method data pointer for the current bcp. 1219 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1220 assert(ProfileInterpreter, "must be profiling interpreter"); 1221 Label set_mdp; 1222 Register mdp = Z_ARG4; 1223 Register method = Z_ARG5; 1224 1225 get_method(method); 1226 // Test MDO to avoid the call if it is null. 1227 load_and_test_long(mdp, method2_(method, method_data)); 1228 z_brz(set_mdp); 1229 1230 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), method, Z_bcp); 1231 // Z_RET: mdi 1232 // Mdo is guaranteed to be non-zero here, we checked for it before the call. 1233 assert(method->is_nonvolatile(), "choose nonvolatile reg or reload from frame"); 1234 z_lg(mdp, method2_(method, method_data)); // Must reload, mdp is volatile reg. 1235 add2reg_with_index(mdp, in_bytes(MethodData::data_offset()), Z_RET, mdp); 1236 1237 bind(set_mdp); 1238 save_mdp(mdp); 1239 } 1240 1241 void InterpreterMacroAssembler::verify_method_data_pointer() { 1242 assert(ProfileInterpreter, "must be profiling interpreter"); 1243 #ifdef ASSERT 1244 NearLabel verify_continue; 1245 Register bcp_expected = Z_ARG3; 1246 Register mdp = Z_ARG4; 1247 Register method = Z_ARG5; 1248 1249 test_method_data_pointer(mdp, verify_continue); // If mdp is zero, continue 1250 get_method(method); 1251 1252 // If the mdp is valid, it will point to a DataLayout header which is 1253 // consistent with the bcp. The converse is highly probable also. 1254 load_sized_value(bcp_expected, Address(mdp, DataLayout::bci_offset()), 2, false /*signed*/); 1255 z_ag(bcp_expected, Address(method, Method::const_offset())); 1256 load_address(bcp_expected, Address(bcp_expected, ConstMethod::codes_offset())); 1257 compareU64_and_branch(bcp_expected, Z_bcp, bcondEqual, verify_continue); 1258 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), method, Z_bcp, mdp); 1259 bind(verify_continue); 1260 #endif // ASSERT 1261 } 1262 1263 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) { 1264 assert(ProfileInterpreter, "must be profiling interpreter"); 1265 z_stg(value, constant, mdp_in); 1266 } 1267 1268 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1269 int constant, 1270 Register tmp, 1271 bool decrement) { 1272 assert_different_registers(mdp_in, tmp); 1273 // counter address 1274 Address data(mdp_in, constant); 1275 const int delta = decrement ? -DataLayout::counter_increment : DataLayout::counter_increment; 1276 add2mem_64(Address(mdp_in, constant), delta, tmp); 1277 } 1278 1279 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, 1280 int flag_byte_constant) { 1281 assert(ProfileInterpreter, "must be profiling interpreter"); 1282 // Set the flag. 1283 z_oi(Address(mdp_in, DataLayout::flags_offset()), flag_byte_constant); 1284 } 1285 1286 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, 1287 int offset, 1288 Register value, 1289 Register test_value_out, 1290 Label& not_equal_continue) { 1291 assert(ProfileInterpreter, "must be profiling interpreter"); 1292 if (test_value_out == noreg) { 1293 z_cg(value, Address(mdp_in, offset)); 1294 z_brne(not_equal_continue); 1295 } else { 1296 // Put the test value into a register, so caller can use it: 1297 z_lg(test_value_out, Address(mdp_in, offset)); 1298 compareU64_and_branch(test_value_out, value, bcondNotEqual, not_equal_continue); 1299 } 1300 } 1301 1302 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) { 1303 update_mdp_by_offset(mdp_in, noreg, offset_of_disp); 1304 } 1305 1306 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1307 Register dataidx, 1308 int offset_of_disp) { 1309 assert(ProfileInterpreter, "must be profiling interpreter"); 1310 Address disp_address(mdp_in, dataidx, offset_of_disp); 1311 Assembler::z_ag(mdp_in, disp_address); 1312 save_mdp(mdp_in); 1313 } 1314 1315 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) { 1316 assert(ProfileInterpreter, "must be profiling interpreter"); 1317 add2reg(mdp_in, constant); 1318 save_mdp(mdp_in); 1319 } 1320 1321 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { 1322 assert(ProfileInterpreter, "must be profiling interpreter"); 1323 assert(return_bci->is_nonvolatile(), "choose nonvolatile reg or save/restore"); 1324 call_VM(noreg, 1325 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), 1326 return_bci); 1327 } 1328 1329 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) { 1330 if (ProfileInterpreter) { 1331 Label profile_continue; 1332 1333 // If no method data exists, go to profile_continue. 1334 // Otherwise, assign to mdp. 1335 test_method_data_pointer(mdp, profile_continue); 1336 1337 // We are taking a branch. Increment the taken count. 1338 // We inline increment_mdp_data_at to return bumped_count in a register 1339 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); 1340 Address data(mdp, JumpData::taken_offset()); 1341 z_lg(bumped_count, data); 1342 // 64-bit overflow is very unlikely. Saturation to 32-bit values is 1343 // performed when reading the counts. 1344 add2reg(bumped_count, DataLayout::counter_increment); 1345 z_stg(bumped_count, data); // Store back out 1346 1347 // The method data pointer needs to be updated to reflect the new target. 1348 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); 1349 bind(profile_continue); 1350 } 1351 } 1352 1353 // Kills Z_R1_scratch. 1354 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { 1355 if (ProfileInterpreter) { 1356 Label profile_continue; 1357 1358 // If no method data exists, go to profile_continue. 1359 test_method_data_pointer(mdp, profile_continue); 1360 1361 // We are taking a branch. Increment the not taken count. 1362 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()), Z_R1_scratch); 1363 1364 // The method data pointer needs to be updated to correspond to 1365 // the next bytecode. 1366 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); 1367 bind(profile_continue); 1368 } 1369 } 1370 1371 // Kills: Z_R1_scratch. 1372 void InterpreterMacroAssembler::profile_call(Register mdp) { 1373 if (ProfileInterpreter) { 1374 Label profile_continue; 1375 1376 // If no method data exists, go to profile_continue. 1377 test_method_data_pointer(mdp, profile_continue); 1378 1379 // We are making a call. Increment the count. 1380 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1381 1382 // The method data pointer needs to be updated to reflect the new target. 1383 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); 1384 bind(profile_continue); 1385 } 1386 } 1387 1388 void InterpreterMacroAssembler::profile_final_call(Register mdp) { 1389 if (ProfileInterpreter) { 1390 Label profile_continue; 1391 1392 // If no method data exists, go to profile_continue. 1393 test_method_data_pointer(mdp, profile_continue); 1394 1395 // We are making a call. Increment the count. 1396 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1397 1398 // The method data pointer needs to be updated to reflect the new target. 1399 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1400 bind(profile_continue); 1401 } 1402 } 1403 1404 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1405 Register mdp, 1406 Register reg2, 1407 bool receiver_can_be_null) { 1408 if (ProfileInterpreter) { 1409 NearLabel profile_continue; 1410 1411 // If no method data exists, go to profile_continue. 1412 test_method_data_pointer(mdp, profile_continue); 1413 1414 NearLabel skip_receiver_profile; 1415 if (receiver_can_be_null) { 1416 NearLabel not_null; 1417 compareU64_and_branch(receiver, (intptr_t)0L, bcondNotEqual, not_null); 1418 // We are making a call. Increment the count for null receiver. 1419 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1420 z_bru(skip_receiver_profile); 1421 bind(not_null); 1422 } 1423 1424 // Record the receiver type. 1425 record_klass_in_profile(receiver, mdp, reg2); 1426 bind(skip_receiver_profile); 1427 1428 // The method data pointer needs to be updated to reflect the new target. 1429 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1430 bind(profile_continue); 1431 } 1432 } 1433 1434 // This routine creates a state machine for updating the multi-row 1435 // type profile at a virtual call site (or other type-sensitive bytecode). 1436 // The machine visits each row (of receiver/count) until the receiver type 1437 // is found, or until it runs out of rows. At the same time, it remembers 1438 // the location of the first empty row. (An empty row records null for its 1439 // receiver, and can be allocated for a newly-observed receiver type.) 1440 // Because there are two degrees of freedom in the state, a simple linear 1441 // search will not work; it must be a decision tree. Hence this helper 1442 // function is recursive, to generate the required tree structured code. 1443 // It's the interpreter, so we are trading off code space for speed. 1444 // See below for example code. 1445 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1446 Register receiver, Register mdp, 1447 Register reg2, int start_row, 1448 Label& done) { 1449 if (TypeProfileWidth == 0) { 1450 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1451 return; 1452 } 1453 1454 int last_row = VirtualCallData::row_limit() - 1; 1455 assert(start_row <= last_row, "must be work left to do"); 1456 // Test this row for both the receiver and for null. 1457 // Take any of three different outcomes: 1458 // 1. found receiver => increment count and goto done 1459 // 2. found null => keep looking for case 1, maybe allocate this cell 1460 // 3. found something else => keep looking for cases 1 and 2 1461 // Case 3 is handled by a recursive call. 1462 for (int row = start_row; row <= last_row; row++) { 1463 NearLabel next_test; 1464 bool test_for_null_also = (row == start_row); 1465 1466 // See if the receiver is receiver[n]. 1467 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1468 test_mdp_data_at(mdp, recvr_offset, receiver, 1469 (test_for_null_also ? reg2 : noreg), 1470 next_test); 1471 // (Reg2 now contains the receiver from the CallData.) 1472 1473 // The receiver is receiver[n]. Increment count[n]. 1474 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1475 increment_mdp_data_at(mdp, count_offset); 1476 z_bru(done); 1477 bind(next_test); 1478 1479 if (test_for_null_also) { 1480 Label found_null; 1481 // Failed the equality check on receiver[n]... Test for null. 1482 z_ltgr(reg2, reg2); 1483 if (start_row == last_row) { 1484 // The only thing left to do is handle the null case. 1485 z_brz(found_null); 1486 // Receiver did not match any saved receiver and there is no empty row for it. 1487 // Increment total counter to indicate polymorphic case. 1488 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1489 z_bru(done); 1490 bind(found_null); 1491 break; 1492 } 1493 // Since null is rare, make it be the branch-taken case. 1494 z_brz(found_null); 1495 1496 // Put all the "Case 3" tests here. 1497 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done); 1498 1499 // Found a null. Keep searching for a matching receiver, 1500 // but remember that this is an empty (unused) slot. 1501 bind(found_null); 1502 } 1503 } 1504 1505 // In the fall-through case, we found no matching receiver, but we 1506 // observed the receiver[start_row] is null. 1507 1508 // Fill in the receiver field and increment the count. 1509 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1510 set_mdp_data_at(mdp, recvr_offset, receiver); 1511 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1512 load_const_optimized(reg2, DataLayout::counter_increment); 1513 set_mdp_data_at(mdp, count_offset, reg2); 1514 if (start_row > 0) { 1515 z_bru(done); 1516 } 1517 } 1518 1519 // Example state machine code for three profile rows: 1520 // // main copy of decision tree, rooted at row[1] 1521 // if (row[0].rec == rec) { row[0].incr(); goto done; } 1522 // if (row[0].rec != nullptr) { 1523 // // inner copy of decision tree, rooted at row[1] 1524 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1525 // if (row[1].rec != nullptr) { 1526 // // degenerate decision tree, rooted at row[2] 1527 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1528 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow 1529 // row[2].init(rec); goto done; 1530 // } else { 1531 // // remember row[1] is empty 1532 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1533 // row[1].init(rec); goto done; 1534 // } 1535 // } else { 1536 // // remember row[0] is empty 1537 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1538 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1539 // row[0].init(rec); goto done; 1540 // } 1541 // done: 1542 1543 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1544 Register mdp, Register reg2) { 1545 assert(ProfileInterpreter, "must be profiling"); 1546 Label done; 1547 1548 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done); 1549 1550 bind (done); 1551 } 1552 1553 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) { 1554 if (ProfileInterpreter) { 1555 NearLabel profile_continue; 1556 uint row; 1557 1558 // If no method data exists, go to profile_continue. 1559 test_method_data_pointer(mdp, profile_continue); 1560 1561 // Update the total ret count. 1562 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1563 1564 for (row = 0; row < RetData::row_limit(); row++) { 1565 NearLabel next_test; 1566 1567 // See if return_bci is equal to bci[n]: 1568 test_mdp_data_at(mdp, 1569 in_bytes(RetData::bci_offset(row)), 1570 return_bci, noreg, 1571 next_test); 1572 1573 // Return_bci is equal to bci[n]. Increment the count. 1574 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); 1575 1576 // The method data pointer needs to be updated to reflect the new target. 1577 update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row))); 1578 z_bru(profile_continue); 1579 bind(next_test); 1580 } 1581 1582 update_mdp_for_ret(return_bci); 1583 1584 bind(profile_continue); 1585 } 1586 } 1587 1588 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { 1589 if (ProfileInterpreter) { 1590 Label profile_continue; 1591 1592 // If no method data exists, go to profile_continue. 1593 test_method_data_pointer(mdp, profile_continue); 1594 1595 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1596 1597 // The method data pointer needs to be updated. 1598 int mdp_delta = in_bytes(BitData::bit_data_size()); 1599 if (TypeProfileCasts) { 1600 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1601 } 1602 update_mdp_by_constant(mdp, mdp_delta); 1603 1604 bind(profile_continue); 1605 } 1606 } 1607 1608 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { 1609 if (ProfileInterpreter) { 1610 Label profile_continue; 1611 1612 // If no method data exists, go to profile_continue. 1613 test_method_data_pointer(mdp, profile_continue); 1614 1615 // The method data pointer needs to be updated. 1616 int mdp_delta = in_bytes(BitData::bit_data_size()); 1617 if (TypeProfileCasts) { 1618 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1619 1620 // Record the object type. 1621 record_klass_in_profile(klass, mdp, reg2); 1622 } 1623 update_mdp_by_constant(mdp, mdp_delta); 1624 1625 bind(profile_continue); 1626 } 1627 } 1628 1629 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { 1630 if (ProfileInterpreter) { 1631 Label profile_continue; 1632 1633 // If no method data exists, go to profile_continue. 1634 test_method_data_pointer(mdp, profile_continue); 1635 1636 // Update the default case count. 1637 increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset())); 1638 1639 // The method data pointer needs to be updated. 1640 update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset())); 1641 1642 bind(profile_continue); 1643 } 1644 } 1645 1646 // Kills: index, scratch1, scratch2. 1647 void InterpreterMacroAssembler::profile_switch_case(Register index, 1648 Register mdp, 1649 Register scratch1, 1650 Register scratch2) { 1651 if (ProfileInterpreter) { 1652 Label profile_continue; 1653 assert_different_registers(index, mdp, scratch1, scratch2); 1654 1655 // If no method data exists, go to profile_continue. 1656 test_method_data_pointer(mdp, profile_continue); 1657 1658 // Build the base (index * per_case_size_in_bytes()) + 1659 // case_array_offset_in_bytes(). 1660 z_sllg(index, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1661 add2reg(index, in_bytes(MultiBranchData::case_array_offset())); 1662 1663 // Add the calculated base to the mdp -> address of the case' data. 1664 Address case_data_addr(mdp, index); 1665 Register case_data = scratch1; 1666 load_address(case_data, case_data_addr); 1667 1668 // Update the case count. 1669 increment_mdp_data_at(case_data, 1670 in_bytes(MultiBranchData::relative_count_offset()), 1671 scratch2); 1672 1673 // The method data pointer needs to be updated. 1674 update_mdp_by_offset(mdp, 1675 index, 1676 in_bytes(MultiBranchData::relative_displacement_offset())); 1677 1678 bind(profile_continue); 1679 } 1680 } 1681 1682 // kills: R0, R1, flags, loads klass from obj (if not null) 1683 void InterpreterMacroAssembler::profile_obj_type(Register obj, Address mdo_addr, Register klass, bool cmp_done) { 1684 NearLabel null_seen, init_klass, do_nothing, do_update; 1685 1686 // Klass = obj is allowed. 1687 const Register tmp = Z_R1; 1688 assert_different_registers(obj, mdo_addr.base(), tmp, Z_R0); 1689 assert_different_registers(klass, mdo_addr.base(), tmp, Z_R0); 1690 1691 z_lg(tmp, mdo_addr); 1692 if (cmp_done) { 1693 z_brz(null_seen); 1694 } else { 1695 compareU64_and_branch(obj, (intptr_t)0, Assembler::bcondEqual, null_seen); 1696 } 1697 1698 MacroAssembler::verify_oop(obj, FILE_AND_LINE); 1699 load_klass(klass, obj); 1700 1701 // Klass seen before, nothing to do (regardless of unknown bit). 1702 z_lgr(Z_R0, tmp); 1703 assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction"); 1704 z_nill(Z_R0, TypeEntries::type_klass_mask & 0xFFFF); 1705 compareU64_and_branch(Z_R0, klass, Assembler::bcondEqual, do_nothing); 1706 1707 // Already unknown. Nothing to do anymore. 1708 z_tmll(tmp, TypeEntries::type_unknown); 1709 z_brc(Assembler::bcondAllOne, do_nothing); 1710 1711 z_lgr(Z_R0, tmp); 1712 assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction"); 1713 z_nill(Z_R0, TypeEntries::type_mask & 0xFFFF); 1714 compareU64_and_branch(Z_R0, (intptr_t)0, Assembler::bcondEqual, init_klass); 1715 1716 // Different than before. Cannot keep accurate profile. 1717 z_oill(tmp, TypeEntries::type_unknown); 1718 z_bru(do_update); 1719 1720 bind(init_klass); 1721 // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1722 z_ogr(tmp, klass); 1723 z_bru(do_update); 1724 1725 bind(null_seen); 1726 // Set null_seen if obj is 0. 1727 z_oill(tmp, TypeEntries::null_seen); 1728 // fallthru: z_bru(do_update); 1729 1730 bind(do_update); 1731 z_stg(tmp, mdo_addr); 1732 1733 bind(do_nothing); 1734 } 1735 1736 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { 1737 if (!ProfileInterpreter) { 1738 return; 1739 } 1740 1741 assert_different_registers(mdp, callee, tmp); 1742 1743 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1744 Label profile_continue; 1745 1746 test_method_data_pointer(mdp, profile_continue); 1747 1748 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1749 1750 z_cliy(in_bytes(DataLayout::tag_offset()) - off_to_start, mdp, 1751 is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1752 z_brne(profile_continue); 1753 1754 if (MethodData::profile_arguments()) { 1755 NearLabel done; 1756 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1757 add2reg(mdp, off_to_args); 1758 1759 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1760 if (i > 0 || MethodData::profile_return()) { 1761 // If return value type is profiled we may have no argument to profile. 1762 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp); 1763 add2reg(tmp, -i*TypeStackSlotEntries::per_arg_count()); 1764 compare64_and_branch(tmp, TypeStackSlotEntries::per_arg_count(), Assembler::bcondLow, done); 1765 } 1766 z_lg(tmp, Address(callee, Method::const_offset())); 1767 z_lgh(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); 1768 // Stack offset o (zero based) from the start of the argument 1769 // list. For n arguments translates into offset n - o - 1 from 1770 // the end of the argument list. But there is an extra slot at 1771 // the top of the stack. So the offset is n - o from Lesp. 1772 z_sg(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args)); 1773 z_sllg(tmp, tmp, Interpreter::logStackElementSize); 1774 Address stack_slot_addr(tmp, Z_esp); 1775 z_ltg(tmp, stack_slot_addr); 1776 1777 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); 1778 profile_obj_type(tmp, mdo_arg_addr, tmp, /*ltg did compare to 0*/ true); 1779 1780 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1781 add2reg(mdp, to_add); 1782 off_to_args += to_add; 1783 } 1784 1785 if (MethodData::profile_return()) { 1786 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp); 1787 add2reg(tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1788 } 1789 1790 bind(done); 1791 1792 if (MethodData::profile_return()) { 1793 // We're right after the type profile for the last 1794 // argument. Tmp is the number of cells left in the 1795 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1796 // if there's a return to profile. 1797 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 1798 z_sllg(tmp, tmp, exact_log2(DataLayout::cell_size)); 1799 z_agr(mdp, tmp); 1800 } 1801 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp); 1802 } else { 1803 assert(MethodData::profile_return(), "either profile call args or call ret"); 1804 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); 1805 } 1806 1807 // Mdp points right after the end of the 1808 // CallTypeData/VirtualCallTypeData, right after the cells for the 1809 // return value type if there's one. 1810 bind(profile_continue); 1811 } 1812 } 1813 1814 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { 1815 assert_different_registers(mdp, ret, tmp); 1816 if (ProfileInterpreter && MethodData::profile_return()) { 1817 Label profile_continue; 1818 1819 test_method_data_pointer(mdp, profile_continue); 1820 1821 if (MethodData::profile_return_jsr292_only()) { 1822 // If we don't profile all invoke bytecodes we must make sure 1823 // it's a bytecode we indeed profile. We can't go back to the 1824 // beginning of the ProfileData we intend to update to check its 1825 // type because we're right after it and we don't known its 1826 // length. 1827 NearLabel do_profile; 1828 Address bc(Z_bcp); 1829 z_lb(tmp, bc); 1830 compare32_and_branch(tmp, Bytecodes::_invokedynamic, Assembler::bcondEqual, do_profile); 1831 compare32_and_branch(tmp, Bytecodes::_invokehandle, Assembler::bcondEqual, do_profile); 1832 get_method(tmp); 1833 // Supplement to 8139891: _intrinsic_id exceeded 1-byte size limit. 1834 if (Method::intrinsic_id_size_in_bytes() == 1) { 1835 z_cli(in_bytes(Method::intrinsic_id_offset()), tmp, static_cast<int>(vmIntrinsics::_compiledLambdaForm)); 1836 } else { 1837 assert(Method::intrinsic_id_size_in_bytes() == 2, "size error: check Method::_intrinsic_id"); 1838 z_lh(tmp, in_bytes(Method::intrinsic_id_offset()), Z_R0, tmp); 1839 z_chi(tmp, static_cast<int>(vmIntrinsics::_compiledLambdaForm)); 1840 } 1841 z_brne(profile_continue); 1842 1843 bind(do_profile); 1844 } 1845 1846 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size())); 1847 profile_obj_type(ret, mdo_ret_addr, tmp); 1848 1849 bind(profile_continue); 1850 } 1851 } 1852 1853 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) { 1854 if (ProfileInterpreter && MethodData::profile_parameters()) { 1855 Label profile_continue, done; 1856 1857 test_method_data_pointer(mdp, profile_continue); 1858 1859 // Load the offset of the area within the MDO used for 1860 // parameters. If it's negative we're not profiling any parameters. 1861 Address parm_di_addr(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())); 1862 load_and_test_int2long(tmp1, parm_di_addr); 1863 z_brl(profile_continue); 1864 1865 // Compute a pointer to the area for parameters from the offset 1866 // and move the pointer to the slot for the last 1867 // parameters. Collect profiling from last parameter down. 1868 // mdo start + parameters offset + array length - 1 1869 1870 // Pointer to the parameter area in the MDO. 1871 z_agr(mdp, tmp1); 1872 1873 // Offset of the current profile entry to update. 1874 const Register entry_offset = tmp1; 1875 // entry_offset = array len in number of cells. 1876 z_lg(entry_offset, Address(mdp, ArrayData::array_len_offset())); 1877 // entry_offset (number of cells) = array len - size of 1 entry 1878 add2reg(entry_offset, -TypeStackSlotEntries::per_arg_count()); 1879 // entry_offset in bytes 1880 z_sllg(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1881 1882 Label loop; 1883 bind(loop); 1884 1885 Address arg_off(mdp, entry_offset, ParametersTypeData::stack_slot_offset(0)); 1886 Address arg_type(mdp, entry_offset, ParametersTypeData::type_offset(0)); 1887 1888 // Load offset on the stack from the slot for this parameter. 1889 z_lg(tmp2, arg_off); 1890 z_sllg(tmp2, tmp2, Interpreter::logStackElementSize); 1891 z_lcgr(tmp2); // Negate. 1892 1893 // Profile the parameter. 1894 z_ltg(tmp2, Address(Z_locals, tmp2)); 1895 profile_obj_type(tmp2, arg_type, tmp2, /*ltg did compare to 0*/ true); 1896 1897 // Go to next parameter. 1898 z_aghi(entry_offset, -TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size); 1899 z_brnl(loop); 1900 1901 bind(profile_continue); 1902 } 1903 } 1904 1905 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 1906 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 1907 int increment, 1908 Address mask, 1909 Register scratch, 1910 bool preloaded, 1911 branch_condition cond, 1912 Label *where) { 1913 assert_different_registers(counter_addr.base(), scratch); 1914 if (preloaded) { 1915 add2reg(scratch, increment); 1916 reg2mem_opt(scratch, counter_addr, false); 1917 } else { 1918 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment) && counter_addr.is_RSYform()) { 1919 z_alsi(counter_addr.disp20(), counter_addr.base(), increment); 1920 mem2reg_signed_opt(scratch, counter_addr); 1921 } else { 1922 mem2reg_signed_opt(scratch, counter_addr); 1923 add2reg(scratch, increment); 1924 reg2mem_opt(scratch, counter_addr, false); 1925 } 1926 } 1927 z_n(scratch, mask); 1928 if (where) { z_brc(cond, *where); } 1929 } 1930 1931 // Get MethodCounters object for given method. Lazily allocated if necessary. 1932 // method - Ptr to Method object. 1933 // Rcounters - Ptr to MethodCounters object associated with Method object. 1934 // skip - Exit point if MethodCounters object can't be created (OOM condition). 1935 void InterpreterMacroAssembler::get_method_counters(Register Rmethod, 1936 Register Rcounters, 1937 Label& skip) { 1938 assert_different_registers(Rmethod, Rcounters); 1939 1940 BLOCK_COMMENT("get MethodCounters object {"); 1941 1942 Label has_counters; 1943 load_and_test_long(Rcounters, Address(Rmethod, Method::method_counters_offset())); 1944 z_brnz(has_counters); 1945 1946 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), Rmethod); 1947 z_ltgr(Rcounters, Z_RET); // Runtime call returns MethodCounters object. 1948 z_brz(skip); // No MethodCounters, out of memory. 1949 1950 bind(has_counters); 1951 1952 BLOCK_COMMENT("} get MethodCounters object"); 1953 } 1954 1955 // Increment invocation counter in MethodCounters object. 1956 // Return (invocation_counter+backedge_counter) as "result" in RctrSum. 1957 // Counter values are all unsigned. 1958 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register RctrSum) { 1959 assert(UseCompiler, "incrementing must be useful"); 1960 assert_different_registers(Rcounters, RctrSum); 1961 1962 int increment = InvocationCounter::count_increment; 1963 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()); 1964 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset()); 1965 1966 BLOCK_COMMENT("Increment invocation counter {"); 1967 1968 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) { 1969 // Increment the invocation counter in place, 1970 // then add the incremented value to the backedge counter. 1971 z_l(RctrSum, be_counter_offset, Rcounters); 1972 z_alsi(inv_counter_offset, Rcounters, increment); // Atomic increment @no extra cost! 1973 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits. 1974 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters); 1975 } else { 1976 // This path is optimized for low register consumption 1977 // at the cost of somewhat higher operand delays. 1978 // It does not need an extra temp register. 1979 1980 // Update the invocation counter. 1981 z_l(RctrSum, inv_counter_offset, Rcounters); 1982 if (RctrSum == Z_R0) { 1983 z_ahi(RctrSum, increment); 1984 } else { 1985 add2reg(RctrSum, increment); 1986 } 1987 z_st(RctrSum, inv_counter_offset, Rcounters); 1988 1989 // Mask off the state bits. 1990 z_nilf(RctrSum, InvocationCounter::count_mask_value); 1991 1992 // Add the backedge counter to the updated invocation counter to 1993 // form the result. 1994 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters); 1995 } 1996 1997 BLOCK_COMMENT("} Increment invocation counter"); 1998 1999 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 2000 } 2001 2002 2003 // increment backedge counter in MethodCounters object. 2004 // return (invocation_counter+backedge_counter) as "result" in RctrSum 2005 // counter values are all unsigned! 2006 void InterpreterMacroAssembler::increment_backedge_counter(Register Rcounters, Register RctrSum) { 2007 assert(UseCompiler, "incrementing must be useful"); 2008 assert_different_registers(Rcounters, RctrSum); 2009 2010 int increment = InvocationCounter::count_increment; 2011 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()); 2012 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset()); 2013 2014 BLOCK_COMMENT("Increment backedge counter {"); 2015 2016 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) { 2017 // Increment the invocation counter in place, 2018 // then add the incremented value to the backedge counter. 2019 z_l(RctrSum, inv_counter_offset, Rcounters); 2020 z_alsi(be_counter_offset, Rcounters, increment); // Atomic increment @no extra cost! 2021 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits. 2022 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters); 2023 } else { 2024 // This path is optimized for low register consumption 2025 // at the cost of somewhat higher operand delays. 2026 // It does not need an extra temp register. 2027 2028 // Update the invocation counter. 2029 z_l(RctrSum, be_counter_offset, Rcounters); 2030 if (RctrSum == Z_R0) { 2031 z_ahi(RctrSum, increment); 2032 } else { 2033 add2reg(RctrSum, increment); 2034 } 2035 z_st(RctrSum, be_counter_offset, Rcounters); 2036 2037 // Mask off the state bits. 2038 z_nilf(RctrSum, InvocationCounter::count_mask_value); 2039 2040 // Add the backedge counter to the updated invocation counter to 2041 // form the result. 2042 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters); 2043 } 2044 2045 BLOCK_COMMENT("} Increment backedge counter"); 2046 2047 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 2048 } 2049 2050 // Add an InterpMonitorElem to stack (see frame_s390.hpp). 2051 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, 2052 Register Rtemp1, 2053 Register Rtemp2, 2054 Register Rtemp3) { 2055 2056 const Register Rcurr_slot = Rtemp1; 2057 const Register Rlimit = Rtemp2; 2058 const jint delta = -frame::interpreter_frame_monitor_size_in_bytes(); 2059 2060 assert((delta & LongAlignmentMask) == 0, 2061 "sizeof BasicObjectLock must be even number of doublewords"); 2062 assert(2 * wordSize == -delta, "this works only as long as delta == -2*wordSize"); 2063 assert(Rcurr_slot != Z_R0, "Register must be usable as base register"); 2064 assert_different_registers(Rlimit, Rcurr_slot, Rtemp3); 2065 2066 get_monitors(Rlimit); 2067 2068 // Adjust stack pointer for additional monitor entry. 2069 resize_frame(RegisterOrConstant((intptr_t) delta), Z_fp, false); 2070 2071 if (!stack_is_empty) { 2072 // Must copy stack contents down. 2073 NearLabel next, done; 2074 2075 // Rtemp := addr(Tos), Z_esp is pointing below it! 2076 add2reg(Rcurr_slot, wordSize, Z_esp); 2077 2078 // Nothing to do, if already at monitor area. 2079 compareU64_and_branch(Rcurr_slot, Rlimit, bcondNotLow, done); 2080 2081 bind(next); 2082 2083 // Move one stack slot. 2084 mem2reg_opt(Rtemp3, Address(Rcurr_slot)); 2085 reg2mem_opt(Rtemp3, Address(Rcurr_slot, delta)); 2086 add2reg(Rcurr_slot, wordSize); 2087 compareU64_and_branch(Rcurr_slot, Rlimit, bcondLow, next); // Are we done? 2088 2089 bind(done); 2090 // Done copying stack. 2091 } 2092 2093 // Adjust expression stack and monitor pointers. 2094 add2reg(Z_esp, delta); 2095 add2reg(Rlimit, delta); 2096 save_monitors(Rlimit); 2097 } 2098 2099 // Note: Index holds the offset in bytes afterwards. 2100 // You can use this to store a new value (with Llocals as the base). 2101 void InterpreterMacroAssembler::access_local_int(Register index, Register dst) { 2102 z_sllg(index, index, LogBytesPerWord); 2103 mem2reg_opt(dst, Address(Z_locals, index), false); 2104 } 2105 2106 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2107 if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); } 2108 } 2109 2110 // Inline assembly for: 2111 // 2112 // if (thread is in interp_only_mode) { 2113 // InterpreterRuntime::post_method_entry(); 2114 // } 2115 2116 void InterpreterMacroAssembler::notify_method_entry() { 2117 2118 // JVMTI 2119 // Whenever JVMTI puts a thread in interp_only_mode, method 2120 // entry/exit events are sent for that thread to track stack 2121 // depth. If it is possible to enter interp_only_mode we add 2122 // the code to check if the event should be sent. 2123 if (JvmtiExport::can_post_interpreter_events()) { 2124 Label jvmti_post_done; 2125 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset())); 2126 z_bre(jvmti_post_done); 2127 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); 2128 bind(jvmti_post_done); 2129 } 2130 } 2131 2132 // Inline assembly for: 2133 // 2134 // if (thread is in interp_only_mode) { 2135 // if (!native_method) save result 2136 // InterpreterRuntime::post_method_exit(); 2137 // if (!native_method) restore result 2138 // } 2139 // if (DTraceMethodProbes) { 2140 // SharedRuntime::dtrace_method_exit(thread, method); 2141 // } 2142 // 2143 // For native methods their result is stored in z_ijava_state.lresult 2144 // and z_ijava_state.fresult before coming here. 2145 // Java methods have their result stored in the expression stack. 2146 // 2147 // Notice the dependency to frame::interpreter_frame_result(). 2148 void InterpreterMacroAssembler::notify_method_exit(bool native_method, 2149 TosState state, 2150 NotifyMethodExitMode mode) { 2151 // JVMTI 2152 // Whenever JVMTI puts a thread in interp_only_mode, method 2153 // entry/exit events are sent for that thread to track stack 2154 // depth. If it is possible to enter interp_only_mode we add 2155 // the code to check if the event should be sent. 2156 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2157 Label jvmti_post_done; 2158 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset())); 2159 z_bre(jvmti_post_done); 2160 if (!native_method) push(state); // see frame::interpreter_frame_result() 2161 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2162 if (!native_method) pop(state); 2163 bind(jvmti_post_done); 2164 } 2165 2166 #if 0 2167 // Dtrace currently not supported on z/Architecture. 2168 { 2169 SkipIfEqual skip(this, &DTraceMethodProbes, false); 2170 push(state); 2171 get_method(c_rarg1); 2172 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2173 r15_thread, c_rarg1); 2174 pop(state); 2175 } 2176 #endif 2177 } 2178 2179 void InterpreterMacroAssembler::skip_if_jvmti_mode(Label &Lskip, Register Rscratch) { 2180 if (!JvmtiExport::can_post_interpreter_events()) { 2181 return; 2182 } 2183 2184 load_and_test_int(Rscratch, Address(Z_thread, JavaThread::interp_only_mode_offset())); 2185 z_brnz(Lskip); 2186 2187 } 2188 2189 // Pop the topmost TOP_IJAVA_FRAME and set it's sender_sp as new Z_SP. 2190 // The return pc is loaded into the register return_pc. 2191 // 2192 // Registers updated: 2193 // return_pc - The return pc of the calling frame. 2194 // tmp1, tmp2 - scratch 2195 void InterpreterMacroAssembler::pop_interpreter_frame(Register return_pc, Register tmp1, Register tmp2) { 2196 // F0 Z_SP -> caller_sp (F1's) 2197 // ... 2198 // sender_sp (F1's) 2199 // ... 2200 // F1 Z_fp -> caller_sp (F2's) 2201 // return_pc (Continuation after return from F0.) 2202 // ... 2203 // F2 caller_sp 2204 2205 // Remove F0's activation. Restoring Z_SP to sender_sp reverts modifications 2206 // (a) by a c2i adapter and (b) by generate_fixed_frame(). 2207 // In case (a) the new top frame F1 is an unextended compiled frame. 2208 // In case (b) F1 is converted from PARENT_IJAVA_FRAME to TOP_IJAVA_FRAME. 2209 2210 // Case (b) seems to be redundant when returning to a interpreted caller, 2211 // because then the caller's top_frame_sp is installed as sp (see 2212 // TemplateInterpreterGenerator::generate_return_entry_for ()). But 2213 // pop_interpreter_frame() is also used in exception handling and there the 2214 // frame type of the caller is unknown, therefore top_frame_sp cannot be used, 2215 // so it is important that sender_sp is the caller's sp as TOP_IJAVA_FRAME. 2216 2217 Register R_f1_sender_sp = tmp1; 2218 Register R_f2_sp = tmp2; 2219 2220 // First check for the interpreter frame's magic. 2221 asm_assert_ijava_state_magic(R_f2_sp/*tmp*/); 2222 z_lg(R_f2_sp, _z_parent_ijava_frame_abi(callers_sp), Z_fp); 2223 z_lg(R_f1_sender_sp, _z_ijava_state_neg(sender_sp), Z_fp); 2224 if (return_pc->is_valid()) 2225 z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp); 2226 // Pop F0 by resizing to R_f1_sender_sp and using R_f2_sp as fp. 2227 resize_frame_absolute(R_f1_sender_sp, R_f2_sp, false/*load fp*/); 2228 2229 #ifdef ASSERT 2230 // The return_pc in the new top frame is dead... at least that's my 2231 // current understanding; to assert this I overwrite it. 2232 load_const_optimized(Z_ARG3, 0xb00b1); 2233 z_stg(Z_ARG3, _z_parent_ijava_frame_abi(return_pc), Z_SP); 2234 #endif 2235 } 2236 2237 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2238 if (VerifyFPU) { 2239 unimplemented("verifyFPU"); 2240 } 2241 }