1 /* 2 * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2016 SAP SE. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 // Major contributions by AHa, AS, JL, ML. 27 28 #include "precompiled.hpp" 29 #include "asm/macroAssembler.inline.hpp" 30 #include "interp_masm_s390.hpp" 31 #include "interpreter/interpreter.hpp" 32 #include "interpreter/interpreterRuntime.hpp" 33 #include "oops/arrayOop.hpp" 34 #include "oops/markOop.hpp" 35 #include "prims/jvmtiExport.hpp" 36 #include "prims/jvmtiThreadState.hpp" 37 #include "runtime/basicLock.hpp" 38 #include "runtime/biasedLocking.hpp" 39 #include "runtime/sharedRuntime.hpp" 40 #include "runtime/thread.inline.hpp" 41 42 // Implementation of InterpreterMacroAssembler. 43 // This file specializes the assember with interpreter-specific macros. 44 45 #ifdef PRODUCT 46 #define BLOCK_COMMENT(str) 47 #define BIND(label) bind(label); 48 #else 49 #define BLOCK_COMMENT(str) block_comment(str) 50 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") 51 #endif 52 53 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) { 54 assert(entry != NULL, "Entry must have been generated by now"); 55 assert(Rscratch != Z_R0, "Can't use R0 for addressing"); 56 branch_optimized(Assembler::bcondAlways, entry); 57 } 58 59 void InterpreterMacroAssembler::empty_expression_stack(void) { 60 get_monitors(Z_R1_scratch); 61 add2reg(Z_esp, -Interpreter::stackElementSize, Z_R1_scratch); 62 } 63 64 // Dispatch code executed in the prolog of a bytecode which does not do it's 65 // own dispatch. 66 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { 67 // On z/Architecture we are short on registers, therefore we do not preload the 68 // dispatch address of the next bytecode. 69 } 70 71 // Dispatch code executed in the epilog of a bytecode which does not do it's 72 // own dispatch. 73 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { 74 dispatch_next(state, step); 75 } 76 77 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { 78 z_llgc(Z_bytecode, bcp_incr, Z_R0, Z_bcp); // Load next bytecode. 79 add2reg(Z_bcp, bcp_incr); // Advance bcp. Add2reg produces optimal code. 80 dispatch_base(state, Interpreter::dispatch_table(state)); 81 } 82 83 // Common code to dispatch and dispatch_only. 84 // Dispatch value in Lbyte_code and increment Lbcp. 85 86 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) { 87 verify_FPU(1, state); 88 89 #ifdef ASSERT 90 address reentry = NULL; 91 { Label OK; 92 // Check if the frame pointer in Z_fp is correct. 93 z_cg(Z_fp, 0, Z_SP); 94 z_bre(OK); 95 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp: " FILE_AND_LINE); 96 bind(OK); 97 } 98 { Label OK; 99 // check if the locals pointer in Z_locals is correct 100 z_cg(Z_locals, _z_ijava_state_neg(locals), Z_fp); 101 z_bre(OK); 102 reentry = stop_chain_static(reentry, "invalid locals pointer Z_locals: " FILE_AND_LINE); 103 bind(OK); 104 } 105 #endif 106 107 // TODO: Maybe implement +VerifyActivationFrameSize here. 108 // verify_thread(); // Too slow. We will just verify on method entry & exit. 109 verify_oop(Z_tos, state); 110 #ifdef FAST_DISPATCH 111 if (table == Interpreter::dispatch_table(state)) { 112 // Use IdispatchTables. 113 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); 114 // Add offset to correct dispatch table. 115 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // Multiply by wordSize. 116 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // Get entry addr. 117 } else 118 #endif 119 { 120 // Dispatch table to use. 121 load_absolute_address(Z_tmp_1, (address) table); // Z_tmp_1 = table; 122 123 // 0 <= Z_bytecode < 256 => Use a 32 bit shift, because it is shorter than sllg. 124 // Z_bytecode must have been loaded zero-extended for this approach to be correct. 125 z_sll(Z_bytecode, LogBytesPerWord, Z_R0); // Multiply by wordSize. 126 z_lg(Z_tmp_1, 0, Z_bytecode, Z_tmp_1); // Get entry addr. 127 } 128 z_br(Z_tmp_1); 129 } 130 131 void InterpreterMacroAssembler::dispatch_only(TosState state) { 132 dispatch_base(state, Interpreter::dispatch_table(state)); 133 } 134 135 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) { 136 dispatch_base(state, Interpreter::normal_table(state)); 137 } 138 139 void InterpreterMacroAssembler::dispatch_via(TosState state, address *table) { 140 // Load current bytecode. 141 z_llgc(Z_bytecode, Address(Z_bcp, (intptr_t)0)); 142 dispatch_base(state, table); 143 } 144 145 // The following call_VM*_base() methods overload and mask the respective 146 // declarations/definitions in class MacroAssembler. They are meant as a "detour" 147 // to perform additional, template interpreter specific tasks before actually 148 // calling their MacroAssembler counterparts. 149 150 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point) { 151 bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated. 152 // interpreter specific 153 // Note: No need to save/restore bcp (Z_R13) pointer since these are callee 154 // saved registers and no blocking/ GC can happen in leaf calls. 155 156 // super call 157 MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation); 158 } 159 160 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, bool allow_relocation) { 161 // interpreter specific 162 // Note: No need to save/restore bcp (Z_R13) pointer since these are callee 163 // saved registers and no blocking/ GC can happen in leaf calls. 164 165 // super call 166 MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation); 167 } 168 169 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp, 170 address entry_point, bool check_exceptions) { 171 bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated. 172 // interpreter specific 173 174 save_bcp(); 175 save_esp(); 176 // super call 177 MacroAssembler::call_VM_base(oop_result, last_java_sp, 178 entry_point, allow_relocation, check_exceptions); 179 restore_bcp(); 180 } 181 182 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp, 183 address entry_point, bool allow_relocation, 184 bool check_exceptions) { 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::check_and_handle_popframe(Register scratch_reg) { 196 if (JvmtiExport::can_pop_frame()) { 197 BLOCK_COMMENT("check_and_handle_popframe {"); 198 Label L; 199 // Initiate popframe handling only if it is not already being 200 // processed. If the flag has the popframe_processing bit set, it 201 // means that this code is called *during* popframe handling - we 202 // don't want to reenter. 203 // TODO: Check if all four state combinations could be visible. 204 // If (processing and !pending) is an invisible/impossible state, 205 // there is optimization potential by testing both bits at once. 206 // Then, All_Zeroes and All_Ones means skip, Mixed means doit. 207 testbit(Address(Z_thread, JavaThread::popframe_condition_offset()), 208 exact_log2(JavaThread::popframe_pending_bit)); 209 z_bfalse(L); 210 testbit(Address(Z_thread, JavaThread::popframe_condition_offset()), 211 exact_log2(JavaThread::popframe_processing_bit)); 212 z_btrue(L); 213 214 // Call Interpreter::remove_activation_preserving_args_entry() to get the 215 // address of the same-named entrypoint in the generated interpreter code. 216 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 217 // The above call should (as its only effect) return the contents of the field 218 // _remove_activation_preserving_args_entry in Z_RET. 219 // We just jump there to have the work done. 220 z_br(Z_RET); 221 // There is no way for control to fall thru here. 222 223 bind(L); 224 BLOCK_COMMENT("} check_and_handle_popframe"); 225 } 226 } 227 228 229 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 230 Register RjvmtiState = Z_R1_scratch; 231 int tos_off = in_bytes(JvmtiThreadState::earlyret_tos_offset()); 232 int oop_off = in_bytes(JvmtiThreadState::earlyret_oop_offset()); 233 int val_off = in_bytes(JvmtiThreadState::earlyret_value_offset()); 234 int state_off = in_bytes(JavaThread::jvmti_thread_state_offset()); 235 236 z_lg(RjvmtiState, state_off, Z_thread); 237 238 switch (state) { 239 case atos: z_lg(Z_tos, oop_off, RjvmtiState); 240 store_const(Address(RjvmtiState, oop_off), 0L, 8, 8, Z_R0_scratch); 241 break; 242 case ltos: z_lg(Z_tos, val_off, RjvmtiState); break; 243 case btos: // fall through 244 case ztos: // fall through 245 case ctos: // fall through 246 case stos: // fall through 247 case itos: z_llgf(Z_tos, val_off, RjvmtiState); break; 248 case ftos: z_le(Z_ftos, val_off, RjvmtiState); break; 249 case dtos: z_ld(Z_ftos, val_off, RjvmtiState); break; 250 case vtos: /* nothing to do */ break; 251 default : ShouldNotReachHere(); 252 } 253 254 // Clean up tos value in the jvmti thread state. 255 store_const(Address(RjvmtiState, val_off), 0L, 8, 8, Z_R0_scratch); 256 // Set tos state field to illegal value. 257 store_const(Address(RjvmtiState, tos_off), ilgl, 4, 1, Z_R0_scratch); 258 } 259 260 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { 261 if (JvmtiExport::can_force_early_return()) { 262 BLOCK_COMMENT("check_and_handle_earlyret {"); 263 Label L; 264 // arg regs are save, because we are just behind the call in call_VM_base 265 Register jvmti_thread_state = Z_ARG2; 266 Register tmp = Z_ARG3; 267 load_and_test_long(jvmti_thread_state, Address(Z_thread, JavaThread::jvmti_thread_state_offset())); 268 z_bre(L); // if (thread->jvmti_thread_state() == NULL) exit; 269 270 // Initiate earlyret handling only if it is not already being processed. 271 // If the flag has the earlyret_processing bit set, it means that this code 272 // is called *during* earlyret handling - we don't want to reenter. 273 274 assert((JvmtiThreadState::earlyret_pending != 0) && (JvmtiThreadState::earlyret_inactive == 0), 275 "must fix this check, when changing the values of the earlyret enum"); 276 assert(JvmtiThreadState::earlyret_pending == 1, "must fix this check, when changing the values of the earlyret enum"); 277 278 load_and_test_int(tmp, Address(jvmti_thread_state, JvmtiThreadState::earlyret_state_offset())); 279 z_brz(L); // if (thread->jvmti_thread_state()->_earlyret_state != JvmtiThreadState::earlyret_pending) exit; 280 281 // Call Interpreter::remove_activation_early_entry() to get the address of the 282 // same-named entrypoint in the generated interpreter code. 283 assert(sizeof(TosState) == 4, "unexpected size"); 284 z_l(Z_ARG1, Address(jvmti_thread_state, JvmtiThreadState::earlyret_tos_offset())); 285 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Z_ARG1); 286 // The above call should (as its only effect) return the contents of the field 287 // _remove_activation_preserving_args_entry in Z_RET. 288 // We just jump there to have the work done. 289 z_br(Z_RET); 290 // There is no way for control to fall thru here. 291 292 bind(L); 293 BLOCK_COMMENT("} check_and_handle_earlyret"); 294 } 295 } 296 297 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2) { 298 lgr_if_needed(Z_ARG1, arg_1); 299 assert(arg_2 != Z_ARG1, "smashed argument"); 300 lgr_if_needed(Z_ARG2, arg_2); 301 MacroAssembler::call_VM_leaf_base(entry_point, true); 302 } 303 304 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, int bcp_offset, size_t index_size) { 305 Address param(Z_bcp, bcp_offset); 306 307 BLOCK_COMMENT("get_cache_index_at_bcp {"); 308 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 309 if (index_size == sizeof(u2)) { 310 load_sized_value(index, param, 2, false /*signed*/); 311 } else if (index_size == sizeof(u4)) { 312 313 load_sized_value(index, param, 4, false); 314 315 // Check if the secondary index definition is still ~x, otherwise 316 // we have to change the following assembler code to calculate the 317 // plain index. 318 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 319 not_(index); // Convert to plain index. 320 } else if (index_size == sizeof(u1)) { 321 z_llgc(index, param); 322 } else { 323 ShouldNotReachHere(); 324 } 325 BLOCK_COMMENT("}"); 326 } 327 328 329 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register cpe_offset, 330 int bcp_offset, size_t index_size) { 331 BLOCK_COMMENT("get_cache_and_index_at_bcp {"); 332 assert_different_registers(cache, cpe_offset); 333 get_cache_index_at_bcp(cpe_offset, bcp_offset, index_size); 334 z_lg(cache, Address(Z_fp, _z_ijava_state_neg(cpoolCache))); 335 // Convert from field index to ConstantPoolCache offset in bytes. 336 z_sllg(cpe_offset, cpe_offset, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord)); 337 BLOCK_COMMENT("}"); 338 } 339 340 // Kills Z_R0_scratch. 341 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, 342 Register cpe_offset, 343 Register bytecode, 344 int byte_no, 345 int bcp_offset, 346 size_t index_size) { 347 BLOCK_COMMENT("get_cache_and_index_and_bytecode_at_bcp {"); 348 get_cache_and_index_at_bcp(cache, cpe_offset, bcp_offset, index_size); 349 350 // We want to load (from CP cache) the bytecode that corresponds to the passed-in byte_no. 351 // It is located at (cache + cpe_offset + base_offset + indices_offset + (8-1) (last byte in DW) - (byte_no+1). 352 // Instead of loading, shifting and masking a DW, we just load that one byte of interest with z_llgc (unsigned). 353 const int base_ix_off = in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()); 354 const int off_in_DW = (8-1) - (1+byte_no); 355 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask"); 356 assert(ConstantPoolCacheEntry::bytecode_1_mask == 0xff, ""); 357 load_sized_value(bytecode, Address(cache, cpe_offset, base_ix_off+off_in_DW), 1, false /*signed*/); 358 359 BLOCK_COMMENT("}"); 360 } 361 362 // Load object from cpool->resolved_references(index). 363 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) { 364 assert_different_registers(result, index); 365 get_constant_pool(result); 366 367 // Convert 368 // - from field index to resolved_references() index and 369 // - from word index to byte offset. 370 // Since this is a java object, it is potentially compressed. 371 Register tmp = index; // reuse 372 z_sllg(index, index, LogBytesPerHeapOop); // Offset into resolved references array. 373 // Load pointer for resolved_references[] objArray. 374 z_lg(result, ConstantPool::resolved_references_offset_in_bytes(), result); 375 // JNIHandles::resolve(result) 376 z_lg(result, 0, result); // Load resolved references array itself. 377 #ifdef ASSERT 378 NearLabel index_ok; 379 z_lgf(Z_R0, Address(result, arrayOopDesc::length_offset_in_bytes())); 380 z_sllg(Z_R0, Z_R0, LogBytesPerHeapOop); 381 compare64_and_branch(tmp, Z_R0, Assembler::bcondLow, index_ok); 382 stop("resolved reference index out of bounds", 0x09256); 383 bind(index_ok); 384 #endif 385 z_agr(result, index); // Address of indexed array element. 386 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result); 387 } 388 389 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, 390 Register tmp, 391 int bcp_offset, 392 size_t index_size) { 393 BLOCK_COMMENT("get_cache_entry_pointer_at_bcp {"); 394 get_cache_and_index_at_bcp(cache, tmp, bcp_offset, index_size); 395 add2reg_with_index(cache, in_bytes(ConstantPoolCache::base_offset()), tmp, cache); 396 BLOCK_COMMENT("}"); 397 } 398 399 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 400 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. 401 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 402 Register Rsuper_klass, 403 Register Rtmp1, 404 Register Rtmp2, 405 Label &ok_is_subtype) { 406 // Profile the not-null value's klass. 407 profile_typecheck(Rtmp1, Rsub_klass, Rtmp2); 408 409 // Do the check. 410 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype); 411 412 // Profile the failure of the check. 413 profile_typecheck_failed(Rtmp1, Rtmp2); 414 } 415 416 // Pop topmost element from stack. It just disappears. 417 // Useful if consumed previously by access via stackTop(). 418 void InterpreterMacroAssembler::popx(int len) { 419 add2reg(Z_esp, len*Interpreter::stackElementSize); 420 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 421 } 422 423 // Get Address object of stack top. No checks. No pop. 424 // Purpose: - Provide address of stack operand to exploit reg-mem operations. 425 // - Avoid RISC-like mem2reg - reg-reg-op sequence. 426 Address InterpreterMacroAssembler::stackTop() { 427 return Address(Z_esp, Interpreter::expr_offset_in_bytes(0)); 428 } 429 430 void InterpreterMacroAssembler::pop_i(Register r) { 431 z_l(r, Interpreter::expr_offset_in_bytes(0), Z_esp); 432 add2reg(Z_esp, Interpreter::stackElementSize); 433 assert_different_registers(r, Z_R1_scratch); 434 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 435 } 436 437 void InterpreterMacroAssembler::pop_ptr(Register r) { 438 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp); 439 add2reg(Z_esp, Interpreter::stackElementSize); 440 assert_different_registers(r, Z_R1_scratch); 441 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 442 } 443 444 void InterpreterMacroAssembler::pop_l(Register r) { 445 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp); 446 add2reg(Z_esp, 2*Interpreter::stackElementSize); 447 assert_different_registers(r, Z_R1_scratch); 448 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 449 } 450 451 void InterpreterMacroAssembler::pop_f(FloatRegister f) { 452 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), false); 453 add2reg(Z_esp, Interpreter::stackElementSize); 454 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 455 } 456 457 void InterpreterMacroAssembler::pop_d(FloatRegister f) { 458 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), true); 459 add2reg(Z_esp, 2*Interpreter::stackElementSize); 460 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 461 } 462 463 void InterpreterMacroAssembler::push_i(Register r) { 464 assert_different_registers(r, Z_R1_scratch); 465 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 466 z_st(r, Address(Z_esp)); 467 add2reg(Z_esp, -Interpreter::stackElementSize); 468 } 469 470 void InterpreterMacroAssembler::push_ptr(Register r) { 471 z_stg(r, Address(Z_esp)); 472 add2reg(Z_esp, -Interpreter::stackElementSize); 473 } 474 475 void InterpreterMacroAssembler::push_l(Register r) { 476 assert_different_registers(r, Z_R1_scratch); 477 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 478 int offset = -Interpreter::stackElementSize; 479 z_stg(r, Address(Z_esp, offset)); 480 clear_mem(Address(Z_esp), Interpreter::stackElementSize); 481 add2reg(Z_esp, 2 * offset); 482 } 483 484 void InterpreterMacroAssembler::push_f(FloatRegister f) { 485 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 486 freg2mem_opt(f, Address(Z_esp), false); 487 add2reg(Z_esp, -Interpreter::stackElementSize); 488 } 489 490 void InterpreterMacroAssembler::push_d(FloatRegister d) { 491 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 492 int offset = -Interpreter::stackElementSize; 493 freg2mem_opt(d, Address(Z_esp, offset)); 494 add2reg(Z_esp, 2 * offset); 495 } 496 497 void InterpreterMacroAssembler::push(TosState state) { 498 verify_oop(Z_tos, state); 499 switch (state) { 500 case atos: push_ptr(); break; 501 case btos: push_i(); break; 502 case ztos: 503 case ctos: 504 case stos: push_i(); break; 505 case itos: push_i(); break; 506 case ltos: push_l(); break; 507 case ftos: push_f(); break; 508 case dtos: push_d(); break; 509 case vtos: /* nothing to do */ break; 510 default : ShouldNotReachHere(); 511 } 512 } 513 514 void InterpreterMacroAssembler::pop(TosState state) { 515 switch (state) { 516 case atos: pop_ptr(Z_tos); break; 517 case btos: pop_i(Z_tos); break; 518 case ztos: 519 case ctos: 520 case stos: pop_i(Z_tos); break; 521 case itos: pop_i(Z_tos); break; 522 case ltos: pop_l(Z_tos); break; 523 case ftos: pop_f(Z_ftos); break; 524 case dtos: pop_d(Z_ftos); break; 525 case vtos: /* nothing to do */ break; 526 default : ShouldNotReachHere(); 527 } 528 verify_oop(Z_tos, state); 529 } 530 531 // Helpers for swap and dup. 532 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 533 z_lg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n))); 534 } 535 536 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 537 z_stg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n))); 538 } 539 540 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted(Register method) { 541 // Satisfy interpreter calling convention (see generate_normal_entry()). 542 z_lgr(Z_R10, Z_SP); // Set sender sp (aka initial caller sp, aka unextended sp). 543 // Record top_frame_sp, because the callee might modify it, if it's compiled. 544 z_stg(Z_SP, _z_ijava_state_neg(top_frame_sp), Z_fp); 545 save_bcp(); 546 save_esp(); 547 z_lgr(Z_method, method); // Set Z_method (kills Z_fp!). 548 } 549 550 // Jump to from_interpreted entry of a call unless single stepping is possible 551 // in this thread in which case we must call the i2i entry. 552 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { 553 assert_different_registers(method, Z_R10 /*used for initial_caller_sp*/, temp); 554 prepare_to_jump_from_interpreted(method); 555 556 if (JvmtiExport::can_post_interpreter_events()) { 557 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 558 // compiled code in threads for which the event is enabled. Check here for 559 // interp_only_mode if these events CAN be enabled. 560 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset())); 561 MacroAssembler::load_and_test_int(Z_R0_scratch, Address(Z_thread, JavaThread::interp_only_mode_offset())); 562 z_bcr(bcondEqual, Z_R1_scratch); // Run compiled code if zero. 563 // Run interpreted. 564 z_lg(Z_R1_scratch, Address(method, Method::interpreter_entry_offset())); 565 z_br(Z_R1_scratch); 566 } else { 567 // Run compiled code. 568 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset())); 569 z_br(Z_R1_scratch); 570 } 571 } 572 573 #ifdef ASSERT 574 void InterpreterMacroAssembler::verify_esp(Register Resp, Register Rtemp) { 575 // About to read or write Resp[0]. 576 // Make sure it is not in the monitors or the TOP_IJAVA_FRAME_ABI. 577 address reentry = NULL; 578 579 { 580 // Check if the frame pointer in Z_fp is correct. 581 NearLabel OK; 582 z_cg(Z_fp, 0, Z_SP); 583 z_bre(OK); 584 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp"); 585 bind(OK); 586 } 587 { 588 // Resp must not point into or below the operand stack, 589 // i.e. IJAVA_STATE.monitors > Resp. 590 NearLabel OK; 591 Register Rmonitors = Rtemp; 592 z_lg(Rmonitors, _z_ijava_state_neg(monitors), Z_fp); 593 compareU64_and_branch(Rmonitors, Resp, bcondHigh, OK); 594 reentry = stop_chain_static(reentry, "too many pops: Z_esp points into monitor area"); 595 bind(OK); 596 } 597 { 598 // Resp may point to the last word of TOP_IJAVA_FRAME_ABI, but not below 599 // i.e. !(Z_SP + frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize > Resp). 600 NearLabel OK; 601 Register Rabi_bottom = Rtemp; 602 add2reg(Rabi_bottom, frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize, Z_SP); 603 compareU64_and_branch(Rabi_bottom, Resp, bcondNotHigh, OK); 604 reentry = stop_chain_static(reentry, "too many pushes: Z_esp points into TOP_IJAVA_FRAME_ABI"); 605 bind(OK); 606 } 607 } 608 609 void InterpreterMacroAssembler::asm_assert_ijava_state_magic(Register tmp) { 610 Label magic_ok; 611 load_const_optimized(tmp, frame::z_istate_magic_number); 612 z_cg(tmp, Address(Z_fp, _z_ijava_state_neg(magic))); 613 z_bre(magic_ok); 614 stop_static("error: wrong magic number in ijava_state access"); 615 bind(magic_ok); 616 } 617 #endif // ASSERT 618 619 void InterpreterMacroAssembler::save_bcp() { 620 z_stg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))); 621 asm_assert_ijava_state_magic(Z_bcp); 622 NOT_PRODUCT(z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)))); 623 } 624 625 void InterpreterMacroAssembler::restore_bcp() { 626 asm_assert_ijava_state_magic(Z_bcp); 627 z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))); 628 } 629 630 void InterpreterMacroAssembler::save_esp() { 631 z_stg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp))); 632 } 633 634 void InterpreterMacroAssembler::restore_esp() { 635 asm_assert_ijava_state_magic(Z_esp); 636 z_lg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp))); 637 } 638 639 void InterpreterMacroAssembler::get_monitors(Register reg) { 640 asm_assert_ijava_state_magic(reg); 641 mem2reg_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors))); 642 } 643 644 void InterpreterMacroAssembler::save_monitors(Register reg) { 645 reg2mem_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors))); 646 } 647 648 void InterpreterMacroAssembler::get_mdp(Register mdp) { 649 z_lg(mdp, _z_ijava_state_neg(mdx), Z_fp); 650 } 651 652 void InterpreterMacroAssembler::save_mdp(Register mdp) { 653 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp); 654 } 655 656 // Values that are only read (besides initialization). 657 void InterpreterMacroAssembler::restore_locals() { 658 asm_assert_ijava_state_magic(Z_locals); 659 z_lg(Z_locals, Address(Z_fp, _z_ijava_state_neg(locals))); 660 } 661 662 void InterpreterMacroAssembler::get_method(Register reg) { 663 asm_assert_ijava_state_magic(reg); 664 z_lg(reg, Address(Z_fp, _z_ijava_state_neg(method))); 665 } 666 667 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(Register Rdst, int bcp_offset, 668 signedOrNot is_signed) { 669 // Rdst is an 8-byte return value!!! 670 671 // Unaligned loads incur only a small penalty on z/Architecture. The penalty 672 // is a few (2..3) ticks, even when the load crosses a cache line 673 // boundary. In case of a cache miss, the stall could, of course, be 674 // much longer. 675 676 switch (is_signed) { 677 case Signed: 678 z_lgh(Rdst, bcp_offset, Z_R0, Z_bcp); 679 break; 680 case Unsigned: 681 z_llgh(Rdst, bcp_offset, Z_R0, Z_bcp); 682 break; 683 default: 684 ShouldNotReachHere(); 685 } 686 } 687 688 689 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(Register Rdst, int bcp_offset, 690 setCCOrNot set_cc) { 691 // Rdst is an 8-byte return value!!! 692 693 // Unaligned loads incur only a small penalty on z/Architecture. The penalty 694 // is a few (2..3) ticks, even when the load crosses a cache line 695 // boundary. In case of a cache miss, the stall could, of course, be 696 // much longer. 697 698 // Both variants implement a sign-extending int2long load. 699 if (set_cc == set_CC) { 700 load_and_test_int2long(Rdst, Address(Z_bcp, (intptr_t)bcp_offset)); 701 } else { 702 mem2reg_signed_opt( Rdst, Address(Z_bcp, (intptr_t)bcp_offset)); 703 } 704 } 705 706 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 707 get_method(Rdst); 708 mem2reg_opt(Rdst, Address(Rdst, Method::const_offset())); 709 mem2reg_opt(Rdst, Address(Rdst, ConstMethod::constants_offset())); 710 } 711 712 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 713 get_constant_pool(Rcpool); 714 mem2reg_opt(Rtags, Address(Rcpool, ConstantPool::tags_offset_in_bytes())); 715 } 716 717 // Unlock if synchronized method. 718 // 719 // Unlock the receiver if this is a synchronized method. 720 // Unlock any Java monitors from syncronized blocks. 721 // 722 // If there are locked Java monitors 723 // If throw_monitor_exception 724 // throws IllegalMonitorStateException 725 // Else if install_monitor_exception 726 // installs IllegalMonitorStateException 727 // Else 728 // no error processing 729 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 730 bool throw_monitor_exception, 731 bool install_monitor_exception) { 732 NearLabel unlocked, unlock, no_unlock; 733 734 { 735 Register R_method = Z_ARG2; 736 Register R_do_not_unlock_if_synchronized = Z_ARG3; 737 738 // Get the value of _do_not_unlock_if_synchronized into G1_scratch. 739 const Address do_not_unlock_if_synchronized(Z_thread, 740 JavaThread::do_not_unlock_if_synchronized_offset()); 741 load_sized_value(R_do_not_unlock_if_synchronized, do_not_unlock_if_synchronized, 1, false /*unsigned*/); 742 z_mvi(do_not_unlock_if_synchronized, false); // Reset the flag. 743 744 // Check if synchronized method. 745 get_method(R_method); 746 verify_oop(Z_tos, state); 747 push(state); // Save tos/result. 748 testbit(method2_(R_method, access_flags), JVM_ACC_SYNCHRONIZED_BIT); 749 z_bfalse(unlocked); 750 751 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 752 // is set. 753 compareU64_and_branch(R_do_not_unlock_if_synchronized, (intptr_t)0L, bcondNotEqual, no_unlock); 754 } 755 756 // unlock monitor 757 758 // BasicObjectLock will be first in list, since this is a 759 // synchronized method. However, need to check that the object has 760 // not been unlocked by an explicit monitorexit bytecode. 761 const Address monitor(Z_fp, -(frame::z_ijava_state_size + (int) sizeof(BasicObjectLock))); 762 // We use Z_ARG2 so that if we go slow path it will be the correct 763 // register for unlock_object to pass to VM directly. 764 load_address(Z_ARG2, monitor); // Address of first monitor. 765 z_lg(Z_ARG3, Address(Z_ARG2, BasicObjectLock::obj_offset_in_bytes())); 766 compareU64_and_branch(Z_ARG3, (intptr_t)0L, bcondNotEqual, unlock); 767 768 if (throw_monitor_exception) { 769 // Entry already unlocked need to throw an exception. 770 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 771 should_not_reach_here(); 772 } else { 773 // Monitor already unlocked during a stack unroll. 774 // If requested, install an illegal_monitor_state_exception. 775 // Continue with stack unrolling. 776 if (install_monitor_exception) { 777 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 778 } 779 z_bru(unlocked); 780 } 781 782 bind(unlock); 783 784 unlock_object(Z_ARG2); 785 786 bind(unlocked); 787 788 // I0, I1: Might contain return value 789 790 // Check that all monitors are unlocked. 791 { 792 NearLabel loop, exception, entry, restart; 793 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 794 // We use Z_ARG2 so that if we go slow path it will be the correct 795 // register for unlock_object to pass to VM directly. 796 Register R_current_monitor = Z_ARG2; 797 Register R_monitor_block_bot = Z_ARG1; 798 const Address monitor_block_top(Z_fp, _z_ijava_state_neg(monitors)); 799 const Address monitor_block_bot(Z_fp, -frame::z_ijava_state_size); 800 801 bind(restart); 802 // Starting with top-most entry. 803 z_lg(R_current_monitor, monitor_block_top); 804 // Points to word before bottom of monitor block. 805 load_address(R_monitor_block_bot, monitor_block_bot); 806 z_bru(entry); 807 808 // Entry already locked, need to throw exception. 809 bind(exception); 810 811 if (throw_monitor_exception) { 812 // Throw exception. 813 MacroAssembler::call_VM(noreg, 814 CAST_FROM_FN_PTR(address, InterpreterRuntime:: 815 throw_illegal_monitor_state_exception)); 816 should_not_reach_here(); 817 } else { 818 // Stack unrolling. Unlock object and install illegal_monitor_exception. 819 // Unlock does not block, so don't have to worry about the frame. 820 // We don't have to preserve c_rarg1 since we are going to throw an exception. 821 unlock_object(R_current_monitor); 822 if (install_monitor_exception) { 823 call_VM(noreg, CAST_FROM_FN_PTR(address, 824 InterpreterRuntime:: 825 new_illegal_monitor_state_exception)); 826 } 827 z_bru(restart); 828 } 829 830 bind(loop); 831 // Check if current entry is used. 832 load_and_test_long(Z_R0_scratch, Address(R_current_monitor, BasicObjectLock::obj_offset_in_bytes())); 833 z_brne(exception); 834 835 add2reg(R_current_monitor, entry_size); // Otherwise advance to next entry. 836 bind(entry); 837 compareU64_and_branch(R_current_monitor, R_monitor_block_bot, bcondNotEqual, loop); 838 } 839 840 bind(no_unlock); 841 pop(state); 842 verify_oop(Z_tos, state); 843 } 844 845 // remove activation 846 // 847 // Unlock the receiver if this is a synchronized method. 848 // Unlock any Java monitors from syncronized blocks. 849 // Remove the activation from the stack. 850 // 851 // If there are locked Java monitors 852 // If throw_monitor_exception 853 // throws IllegalMonitorStateException 854 // Else if install_monitor_exception 855 // installs IllegalMonitorStateException 856 // Else 857 // no error processing 858 void InterpreterMacroAssembler::remove_activation(TosState state, 859 Register return_pc, 860 bool throw_monitor_exception, 861 bool install_monitor_exception, 862 bool notify_jvmti) { 863 BLOCK_COMMENT("remove_activation {"); 864 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 865 866 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 867 notify_method_exit(false, state, notify_jvmti ? NotifyJVMTI : SkipNotifyJVMTI); 868 869 if (StackReservedPages > 0) { 870 BLOCK_COMMENT("reserved_stack_check:"); 871 // Test if reserved zone needs to be enabled. 872 Label no_reserved_zone_enabling; 873 874 // Compare frame pointers. There is no good stack pointer, as with stack 875 // frame compression we can get different SPs when we do calls. A subsequent 876 // call could have a smaller SP, so that this compare succeeds for an 877 // inner call of the method annotated with ReservedStack. 878 z_lg(Z_R0, Address(Z_SP, (intptr_t)_z_abi(callers_sp))); 879 z_clg(Z_R0, Address(Z_thread, JavaThread::reserved_stack_activation_offset())); // Compare with frame pointer in memory. 880 z_brl(no_reserved_zone_enabling); 881 882 // Enable reserved zone again, throw stack overflow exception. 883 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), Z_thread); 884 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError)); 885 886 should_not_reach_here(); 887 888 bind(no_reserved_zone_enabling); 889 } 890 891 verify_oop(Z_tos, state); 892 verify_thread(); 893 894 pop_interpreter_frame(return_pc, Z_ARG2, Z_ARG3); 895 BLOCK_COMMENT("} remove_activation"); 896 } 897 898 // lock object 899 // 900 // Registers alive 901 // monitor - Address of the BasicObjectLock to be used for locking, 902 // which must be initialized with the object to lock. 903 // object - Address of the object to be locked. 904 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 905 906 if (UseHeavyMonitors) { 907 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 908 monitor, /*check_for_exceptions=*/false); 909 return; 910 } 911 912 // template code: 913 // 914 // markOop displaced_header = obj->mark().set_unlocked(); 915 // monitor->lock()->set_displaced_header(displaced_header); 916 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 917 // // We stored the monitor address into the object's mark word. 918 // } else if (THREAD->is_lock_owned((address)displaced_header)) 919 // // Simple recursive case. 920 // monitor->lock()->set_displaced_header(NULL); 921 // } else { 922 // // Slow path. 923 // InterpreterRuntime::monitorenter(THREAD, monitor); 924 // } 925 926 const Register displaced_header = Z_ARG5; 927 const Register object_mark_addr = Z_ARG4; 928 const Register current_header = Z_ARG5; 929 930 NearLabel done; 931 NearLabel slow_case; 932 933 // markOop displaced_header = obj->mark().set_unlocked(); 934 935 // Load markOop from object into displaced_header. 936 z_lg(displaced_header, oopDesc::mark_offset_in_bytes(), object); 937 938 if (UseBiasedLocking) { 939 biased_locking_enter(object, displaced_header, Z_R1, Z_R0, done, &slow_case); 940 } 941 942 // Set displaced_header to be (markOop of object | UNLOCK_VALUE). 943 z_oill(displaced_header, markOopDesc::unlocked_value); 944 945 // monitor->lock()->set_displaced_header(displaced_header); 946 947 // Initialize the box (Must happen before we update the object mark!). 948 z_stg(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 949 BasicLock::displaced_header_offset_in_bytes(), monitor); 950 951 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 952 953 // Store stack address of the BasicObjectLock (this is monitor) into object. 954 add2reg(object_mark_addr, oopDesc::mark_offset_in_bytes(), object); 955 956 z_csg(displaced_header, monitor, 0, object_mark_addr); 957 assert(current_header==displaced_header, "must be same register"); // Identified two registers from z/Architecture. 958 959 z_bre(done); 960 961 // } else if (THREAD->is_lock_owned((address)displaced_header)) 962 // // Simple recursive case. 963 // monitor->lock()->set_displaced_header(NULL); 964 965 // We did not see an unlocked object so try the fast recursive case. 966 967 // Check if owner is self by comparing the value in the markOop of object 968 // (current_header) with the stack pointer. 969 z_sgr(current_header, Z_SP); 970 971 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 972 973 // The prior sequence "LGR, NGR, LTGR" can be done better 974 // (Z_R1 is temp and not used after here). 975 load_const_optimized(Z_R0, (~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place)); 976 z_ngr(Z_R0, current_header); // AND sets CC (result eq/ne 0) 977 978 // If condition is true we are done and hence we can store 0 in the displaced 979 // header indicating it is a recursive lock and be done. 980 z_brne(slow_case); 981 z_release(); // Membar unnecessary on zarch AND because the above csg does a sync before and after. 982 z_stg(Z_R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() + 983 BasicLock::displaced_header_offset_in_bytes(), monitor); 984 z_bru(done); 985 986 // } else { 987 // // Slow path. 988 // InterpreterRuntime::monitorenter(THREAD, monitor); 989 990 // None of the above fast optimizations worked so we have to get into the 991 // slow case of monitor enter. 992 bind(slow_case); 993 994 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 995 monitor, /*check_for_exceptions=*/false); 996 997 // } 998 999 bind(done); 1000 } 1001 1002 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 1003 // 1004 // Registers alive 1005 // monitor - address of the BasicObjectLock to be used for locking, 1006 // which must be initialized with the object to lock. 1007 // 1008 // Throw IllegalMonitorException if object is not locked by current thread. 1009 void InterpreterMacroAssembler::unlock_object(Register monitor, Register object) { 1010 1011 if (UseHeavyMonitors) { 1012 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1013 monitor, /*check_for_exceptions=*/ true); 1014 return; 1015 } 1016 1017 // else { 1018 // template code: 1019 // 1020 // if ((displaced_header = monitor->displaced_header()) == NULL) { 1021 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 1022 // monitor->set_obj(NULL); 1023 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 1024 // // We swapped the unlocked mark in displaced_header into the object's mark word. 1025 // monitor->set_obj(NULL); 1026 // } else { 1027 // // Slow path. 1028 // InterpreterRuntime::monitorexit(THREAD, monitor); 1029 // } 1030 1031 const Register displaced_header = Z_ARG4; 1032 const Register current_header = Z_R1; 1033 Address obj_entry(monitor, BasicObjectLock::obj_offset_in_bytes()); 1034 Label done; 1035 1036 if (object == noreg) { 1037 // In the template interpreter, we must assure that the object 1038 // entry in the monitor is cleared on all paths. Thus we move 1039 // loading up to here, and clear the entry afterwards. 1040 object = Z_ARG3; // Use Z_ARG3 if caller didn't pass object. 1041 z_lg(object, obj_entry); 1042 } 1043 1044 assert_different_registers(monitor, object, displaced_header, current_header); 1045 1046 // if ((displaced_header = monitor->displaced_header()) == NULL) { 1047 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 1048 // monitor->set_obj(NULL); 1049 1050 clear_mem(obj_entry, sizeof(oop)); 1051 1052 if (UseBiasedLocking) { 1053 // The object address from the monitor is in object. 1054 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0"); 1055 biased_locking_exit(object, displaced_header, done); 1056 } 1057 1058 // Test first if we are in the fast recursive case. 1059 MacroAssembler::load_and_test_long(displaced_header, 1060 Address(monitor, BasicObjectLock::lock_offset_in_bytes() + 1061 BasicLock::displaced_header_offset_in_bytes())); 1062 z_bre(done); // displaced_header == 0 -> goto done 1063 1064 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 1065 // // We swapped the unlocked mark in displaced_header into the object's mark word. 1066 // monitor->set_obj(NULL); 1067 1068 // If we still have a lightweight lock, unlock the object and be done. 1069 1070 // The markword is expected to be at offset 0. 1071 assert(oopDesc::mark_offset_in_bytes() == 0, "unlock_object: review code below"); 1072 1073 // We have the displaced header in displaced_header. If the lock is still 1074 // lightweight, it will contain the monitor address and we'll store the 1075 // displaced header back into the object's mark word. 1076 z_lgr(current_header, monitor); 1077 z_csg(current_header, displaced_header, 0, object); 1078 z_bre(done); 1079 1080 // } else { 1081 // // Slow path. 1082 // InterpreterRuntime::monitorexit(THREAD, monitor); 1083 1084 // The lock has been converted into a heavy lock and hence 1085 // we need to get into the slow case. 1086 z_stg(object, obj_entry); // Restore object entry, has been cleared above. 1087 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1088 monitor, /*check_for_exceptions=*/false); 1089 1090 // } 1091 1092 bind(done); 1093 } 1094 1095 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) { 1096 assert(ProfileInterpreter, "must be profiling interpreter"); 1097 load_and_test_long(mdp, Address(Z_fp, _z_ijava_state_neg(mdx))); 1098 z_brz(zero_continue); 1099 } 1100 1101 // Set the method data pointer for the current bcp. 1102 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1103 assert(ProfileInterpreter, "must be profiling interpreter"); 1104 Label set_mdp; 1105 Register mdp = Z_ARG4; 1106 Register method = Z_ARG5; 1107 1108 get_method(method); 1109 // Test MDO to avoid the call if it is NULL. 1110 load_and_test_long(mdp, method2_(method, method_data)); 1111 z_brz(set_mdp); 1112 1113 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), method, Z_bcp); 1114 // Z_RET: mdi 1115 // Mdo is guaranteed to be non-zero here, we checked for it before the call. 1116 assert(method->is_nonvolatile(), "choose nonvolatile reg or reload from frame"); 1117 z_lg(mdp, method2_(method, method_data)); // Must reload, mdp is volatile reg. 1118 add2reg_with_index(mdp, in_bytes(MethodData::data_offset()), Z_RET, mdp); 1119 1120 bind(set_mdp); 1121 save_mdp(mdp); 1122 } 1123 1124 void InterpreterMacroAssembler::verify_method_data_pointer() { 1125 assert(ProfileInterpreter, "must be profiling interpreter"); 1126 #ifdef ASSERT 1127 NearLabel verify_continue; 1128 Register bcp_expected = Z_ARG3; 1129 Register mdp = Z_ARG4; 1130 Register method = Z_ARG5; 1131 1132 test_method_data_pointer(mdp, verify_continue); // If mdp is zero, continue 1133 get_method(method); 1134 1135 // If the mdp is valid, it will point to a DataLayout header which is 1136 // consistent with the bcp. The converse is highly probable also. 1137 load_sized_value(bcp_expected, Address(mdp, DataLayout::bci_offset()), 2, false /*signed*/); 1138 z_ag(bcp_expected, Address(method, Method::const_offset())); 1139 load_address(bcp_expected, Address(bcp_expected, ConstMethod::codes_offset())); 1140 compareU64_and_branch(bcp_expected, Z_bcp, bcondEqual, verify_continue); 1141 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), method, Z_bcp, mdp); 1142 bind(verify_continue); 1143 #endif // ASSERT 1144 } 1145 1146 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) { 1147 assert(ProfileInterpreter, "must be profiling interpreter"); 1148 z_stg(value, constant, mdp_in); 1149 } 1150 1151 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1152 int constant, 1153 Register tmp, 1154 bool decrement) { 1155 assert_different_registers(mdp_in, tmp); 1156 // counter address 1157 Address data(mdp_in, constant); 1158 const int delta = decrement ? -DataLayout::counter_increment : DataLayout::counter_increment; 1159 add2mem_64(Address(mdp_in, constant), delta, tmp); 1160 } 1161 1162 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, 1163 int flag_byte_constant) { 1164 assert(ProfileInterpreter, "must be profiling interpreter"); 1165 // Set the flag. 1166 z_oi(Address(mdp_in, DataLayout::flags_offset()), flag_byte_constant); 1167 } 1168 1169 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, 1170 int offset, 1171 Register value, 1172 Register test_value_out, 1173 Label& not_equal_continue) { 1174 assert(ProfileInterpreter, "must be profiling interpreter"); 1175 if (test_value_out == noreg) { 1176 z_cg(value, Address(mdp_in, offset)); 1177 z_brne(not_equal_continue); 1178 } else { 1179 // Put the test value into a register, so caller can use it: 1180 z_lg(test_value_out, Address(mdp_in, offset)); 1181 compareU64_and_branch(test_value_out, value, bcondNotEqual, not_equal_continue); 1182 } 1183 } 1184 1185 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) { 1186 update_mdp_by_offset(mdp_in, noreg, offset_of_disp); 1187 } 1188 1189 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1190 Register dataidx, 1191 int offset_of_disp) { 1192 assert(ProfileInterpreter, "must be profiling interpreter"); 1193 Address disp_address(mdp_in, dataidx, offset_of_disp); 1194 Assembler::z_ag(mdp_in, disp_address); 1195 save_mdp(mdp_in); 1196 } 1197 1198 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) { 1199 assert(ProfileInterpreter, "must be profiling interpreter"); 1200 add2reg(mdp_in, constant); 1201 save_mdp(mdp_in); 1202 } 1203 1204 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { 1205 assert(ProfileInterpreter, "must be profiling interpreter"); 1206 assert(return_bci->is_nonvolatile(), "choose nonvolatile reg or save/restore"); 1207 call_VM(noreg, 1208 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), 1209 return_bci); 1210 } 1211 1212 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) { 1213 if (ProfileInterpreter) { 1214 Label profile_continue; 1215 1216 // If no method data exists, go to profile_continue. 1217 // Otherwise, assign to mdp. 1218 test_method_data_pointer(mdp, profile_continue); 1219 1220 // We are taking a branch. Increment the taken count. 1221 // We inline increment_mdp_data_at to return bumped_count in a register 1222 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); 1223 Address data(mdp, JumpData::taken_offset()); 1224 z_lg(bumped_count, data); 1225 // 64-bit overflow is very unlikely. Saturation to 32-bit values is 1226 // performed when reading the counts. 1227 add2reg(bumped_count, DataLayout::counter_increment); 1228 z_stg(bumped_count, data); // Store back out 1229 1230 // The method data pointer needs to be updated to reflect the new target. 1231 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); 1232 bind(profile_continue); 1233 } 1234 } 1235 1236 // Kills Z_R1_scratch. 1237 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { 1238 if (ProfileInterpreter) { 1239 Label profile_continue; 1240 1241 // If no method data exists, go to profile_continue. 1242 test_method_data_pointer(mdp, profile_continue); 1243 1244 // We are taking a branch. Increment the not taken count. 1245 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()), Z_R1_scratch); 1246 1247 // The method data pointer needs to be updated to correspond to 1248 // the next bytecode. 1249 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); 1250 bind(profile_continue); 1251 } 1252 } 1253 1254 // Kills: Z_R1_scratch. 1255 void InterpreterMacroAssembler::profile_call(Register mdp) { 1256 if (ProfileInterpreter) { 1257 Label profile_continue; 1258 1259 // If no method data exists, go to profile_continue. 1260 test_method_data_pointer(mdp, profile_continue); 1261 1262 // We are making a call. Increment the count. 1263 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1264 1265 // The method data pointer needs to be updated to reflect the new target. 1266 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); 1267 bind(profile_continue); 1268 } 1269 } 1270 1271 void InterpreterMacroAssembler::profile_final_call(Register mdp) { 1272 if (ProfileInterpreter) { 1273 Label profile_continue; 1274 1275 // If no method data exists, go to profile_continue. 1276 test_method_data_pointer(mdp, profile_continue); 1277 1278 // We are making a call. Increment the count. 1279 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1280 1281 // The method data pointer needs to be updated to reflect the new target. 1282 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1283 bind(profile_continue); 1284 } 1285 } 1286 1287 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1288 Register mdp, 1289 Register reg2, 1290 bool receiver_can_be_null) { 1291 if (ProfileInterpreter) { 1292 NearLabel profile_continue; 1293 1294 // If no method data exists, go to profile_continue. 1295 test_method_data_pointer(mdp, profile_continue); 1296 1297 NearLabel skip_receiver_profile; 1298 if (receiver_can_be_null) { 1299 NearLabel not_null; 1300 compareU64_and_branch(receiver, (intptr_t)0L, bcondNotEqual, not_null); 1301 // We are making a call. Increment the count for null receiver. 1302 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1303 z_bru(skip_receiver_profile); 1304 bind(not_null); 1305 } 1306 1307 // Record the receiver type. 1308 record_klass_in_profile(receiver, mdp, reg2, true); 1309 bind(skip_receiver_profile); 1310 1311 // The method data pointer needs to be updated to reflect the new target. 1312 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1313 bind(profile_continue); 1314 } 1315 } 1316 1317 // This routine creates a state machine for updating the multi-row 1318 // type profile at a virtual call site (or other type-sensitive bytecode). 1319 // The machine visits each row (of receiver/count) until the receiver type 1320 // is found, or until it runs out of rows. At the same time, it remembers 1321 // the location of the first empty row. (An empty row records null for its 1322 // receiver, and can be allocated for a newly-observed receiver type.) 1323 // Because there are two degrees of freedom in the state, a simple linear 1324 // search will not work; it must be a decision tree. Hence this helper 1325 // function is recursive, to generate the required tree structured code. 1326 // It's the interpreter, so we are trading off code space for speed. 1327 // See below for example code. 1328 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1329 Register receiver, Register mdp, 1330 Register reg2, int start_row, 1331 Label& done, bool is_virtual_call) { 1332 if (TypeProfileWidth == 0) { 1333 if (is_virtual_call) { 1334 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1335 } 1336 return; 1337 } 1338 1339 int last_row = VirtualCallData::row_limit() - 1; 1340 assert(start_row <= last_row, "must be work left to do"); 1341 // Test this row for both the receiver and for null. 1342 // Take any of three different outcomes: 1343 // 1. found receiver => increment count and goto done 1344 // 2. found null => keep looking for case 1, maybe allocate this cell 1345 // 3. found something else => keep looking for cases 1 and 2 1346 // Case 3 is handled by a recursive call. 1347 for (int row = start_row; row <= last_row; row++) { 1348 NearLabel next_test; 1349 bool test_for_null_also = (row == start_row); 1350 1351 // See if the receiver is receiver[n]. 1352 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1353 test_mdp_data_at(mdp, recvr_offset, receiver, 1354 (test_for_null_also ? reg2 : noreg), 1355 next_test); 1356 // (Reg2 now contains the receiver from the CallData.) 1357 1358 // The receiver is receiver[n]. Increment count[n]. 1359 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1360 increment_mdp_data_at(mdp, count_offset); 1361 z_bru(done); 1362 bind(next_test); 1363 1364 if (test_for_null_also) { 1365 Label found_null; 1366 // Failed the equality check on receiver[n]... Test for null. 1367 z_ltgr(reg2, reg2); 1368 if (start_row == last_row) { 1369 // The only thing left to do is handle the null case. 1370 if (is_virtual_call) { 1371 z_brz(found_null); 1372 // Receiver did not match any saved receiver and there is no empty row for it. 1373 // Increment total counter to indicate polymorphic case. 1374 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1375 z_bru(done); 1376 bind(found_null); 1377 } else { 1378 z_brnz(done); 1379 } 1380 break; 1381 } 1382 // Since null is rare, make it be the branch-taken case. 1383 z_brz(found_null); 1384 1385 // Put all the "Case 3" tests here. 1386 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call); 1387 1388 // Found a null. Keep searching for a matching receiver, 1389 // but remember that this is an empty (unused) slot. 1390 bind(found_null); 1391 } 1392 } 1393 1394 // In the fall-through case, we found no matching receiver, but we 1395 // observed the receiver[start_row] is NULL. 1396 1397 // Fill in the receiver field and increment the count. 1398 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1399 set_mdp_data_at(mdp, recvr_offset, receiver); 1400 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1401 load_const_optimized(reg2, DataLayout::counter_increment); 1402 set_mdp_data_at(mdp, count_offset, reg2); 1403 if (start_row > 0) { 1404 z_bru(done); 1405 } 1406 } 1407 1408 // Example state machine code for three profile rows: 1409 // // main copy of decision tree, rooted at row[1] 1410 // if (row[0].rec == rec) { row[0].incr(); goto done; } 1411 // if (row[0].rec != NULL) { 1412 // // inner copy of decision tree, rooted at row[1] 1413 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1414 // if (row[1].rec != NULL) { 1415 // // degenerate decision tree, rooted at row[2] 1416 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1417 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow 1418 // row[2].init(rec); goto done; 1419 // } else { 1420 // // remember row[1] is empty 1421 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1422 // row[1].init(rec); goto done; 1423 // } 1424 // } else { 1425 // // remember row[0] is empty 1426 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1427 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1428 // row[0].init(rec); goto done; 1429 // } 1430 // done: 1431 1432 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1433 Register mdp, Register reg2, 1434 bool is_virtual_call) { 1435 assert(ProfileInterpreter, "must be profiling"); 1436 Label done; 1437 1438 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); 1439 1440 bind (done); 1441 } 1442 1443 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) { 1444 if (ProfileInterpreter) { 1445 NearLabel profile_continue; 1446 uint row; 1447 1448 // If no method data exists, go to profile_continue. 1449 test_method_data_pointer(mdp, profile_continue); 1450 1451 // Update the total ret count. 1452 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1453 1454 for (row = 0; row < RetData::row_limit(); row++) { 1455 NearLabel next_test; 1456 1457 // See if return_bci is equal to bci[n]: 1458 test_mdp_data_at(mdp, 1459 in_bytes(RetData::bci_offset(row)), 1460 return_bci, noreg, 1461 next_test); 1462 1463 // Return_bci is equal to bci[n]. Increment the count. 1464 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); 1465 1466 // The method data pointer needs to be updated to reflect the new target. 1467 update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row))); 1468 z_bru(profile_continue); 1469 bind(next_test); 1470 } 1471 1472 update_mdp_for_ret(return_bci); 1473 1474 bind(profile_continue); 1475 } 1476 } 1477 1478 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { 1479 if (ProfileInterpreter) { 1480 Label profile_continue; 1481 1482 // If no method data exists, go to profile_continue. 1483 test_method_data_pointer(mdp, profile_continue); 1484 1485 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1486 1487 // The method data pointer needs to be updated. 1488 int mdp_delta = in_bytes(BitData::bit_data_size()); 1489 if (TypeProfileCasts) { 1490 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1491 } 1492 update_mdp_by_constant(mdp, mdp_delta); 1493 1494 bind(profile_continue); 1495 } 1496 } 1497 1498 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp, Register tmp) { 1499 if (ProfileInterpreter && TypeProfileCasts) { 1500 Label profile_continue; 1501 1502 // If no method data exists, go to profile_continue. 1503 test_method_data_pointer(mdp, profile_continue); 1504 1505 int count_offset = in_bytes(CounterData::count_offset()); 1506 // Back up the address, since we have already bumped the mdp. 1507 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1508 1509 // *Decrement* the counter. We expect to see zero or small negatives. 1510 increment_mdp_data_at(mdp, count_offset, tmp, true); 1511 1512 bind (profile_continue); 1513 } 1514 } 1515 1516 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { 1517 if (ProfileInterpreter) { 1518 Label profile_continue; 1519 1520 // If no method data exists, go to profile_continue. 1521 test_method_data_pointer(mdp, profile_continue); 1522 1523 // The method data pointer needs to be updated. 1524 int mdp_delta = in_bytes(BitData::bit_data_size()); 1525 if (TypeProfileCasts) { 1526 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1527 1528 // Record the object type. 1529 record_klass_in_profile(klass, mdp, reg2, false); 1530 } 1531 update_mdp_by_constant(mdp, mdp_delta); 1532 1533 bind(profile_continue); 1534 } 1535 } 1536 1537 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { 1538 if (ProfileInterpreter) { 1539 Label profile_continue; 1540 1541 // If no method data exists, go to profile_continue. 1542 test_method_data_pointer(mdp, profile_continue); 1543 1544 // Update the default case count. 1545 increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset())); 1546 1547 // The method data pointer needs to be updated. 1548 update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset())); 1549 1550 bind(profile_continue); 1551 } 1552 } 1553 1554 // Kills: index, scratch1, scratch2. 1555 void InterpreterMacroAssembler::profile_switch_case(Register index, 1556 Register mdp, 1557 Register scratch1, 1558 Register scratch2) { 1559 if (ProfileInterpreter) { 1560 Label profile_continue; 1561 assert_different_registers(index, mdp, scratch1, scratch2); 1562 1563 // If no method data exists, go to profile_continue. 1564 test_method_data_pointer(mdp, profile_continue); 1565 1566 // Build the base (index * per_case_size_in_bytes()) + 1567 // case_array_offset_in_bytes(). 1568 z_sllg(index, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1569 add2reg(index, in_bytes(MultiBranchData::case_array_offset())); 1570 1571 // Add the calculated base to the mdp -> address of the case' data. 1572 Address case_data_addr(mdp, index); 1573 Register case_data = scratch1; 1574 load_address(case_data, case_data_addr); 1575 1576 // Update the case count. 1577 increment_mdp_data_at(case_data, 1578 in_bytes(MultiBranchData::relative_count_offset()), 1579 scratch2); 1580 1581 // The method data pointer needs to be updated. 1582 update_mdp_by_offset(mdp, 1583 index, 1584 in_bytes(MultiBranchData::relative_displacement_offset())); 1585 1586 bind(profile_continue); 1587 } 1588 } 1589 1590 // kills: R0, R1, flags, loads klass from obj (if not null) 1591 void InterpreterMacroAssembler::profile_obj_type(Register obj, Address mdo_addr, Register klass, bool cmp_done) { 1592 NearLabel null_seen, init_klass, do_nothing, do_update; 1593 1594 // Klass = obj is allowed. 1595 const Register tmp = Z_R1; 1596 assert_different_registers(obj, mdo_addr.base(), tmp, Z_R0); 1597 assert_different_registers(klass, mdo_addr.base(), tmp, Z_R0); 1598 1599 z_lg(tmp, mdo_addr); 1600 if (cmp_done) { 1601 z_brz(null_seen); 1602 } else { 1603 compareU64_and_branch(obj, (intptr_t)0, Assembler::bcondEqual, null_seen); 1604 } 1605 1606 verify_oop(obj); 1607 load_klass(klass, obj); 1608 1609 // Klass seen before, nothing to do (regardless of unknown bit). 1610 z_lgr(Z_R0, tmp); 1611 assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction"); 1612 z_nill(Z_R0, TypeEntries::type_klass_mask & 0xFFFF); 1613 compareU64_and_branch(Z_R0, klass, Assembler::bcondEqual, do_nothing); 1614 1615 // Already unknown. Nothing to do anymore. 1616 z_tmll(tmp, TypeEntries::type_unknown); 1617 z_brc(Assembler::bcondAllOne, do_nothing); 1618 1619 z_lgr(Z_R0, tmp); 1620 assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction"); 1621 z_nill(Z_R0, TypeEntries::type_mask & 0xFFFF); 1622 compareU64_and_branch(Z_R0, (intptr_t)0, Assembler::bcondEqual, init_klass); 1623 1624 // Different than before. Cannot keep accurate profile. 1625 z_oill(tmp, TypeEntries::type_unknown); 1626 z_bru(do_update); 1627 1628 bind(init_klass); 1629 // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1630 z_ogr(tmp, klass); 1631 z_bru(do_update); 1632 1633 bind(null_seen); 1634 // Set null_seen if obj is 0. 1635 z_oill(tmp, TypeEntries::null_seen); 1636 // fallthru: z_bru(do_update); 1637 1638 bind(do_update); 1639 z_stg(tmp, mdo_addr); 1640 1641 bind(do_nothing); 1642 } 1643 1644 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { 1645 if (!ProfileInterpreter) { 1646 return; 1647 } 1648 1649 assert_different_registers(mdp, callee, tmp); 1650 1651 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1652 Label profile_continue; 1653 1654 test_method_data_pointer(mdp, profile_continue); 1655 1656 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1657 1658 z_cliy(in_bytes(DataLayout::tag_offset()) - off_to_start, mdp, 1659 is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1660 z_brne(profile_continue); 1661 1662 if (MethodData::profile_arguments()) { 1663 NearLabel done; 1664 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1665 add2reg(mdp, off_to_args); 1666 1667 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1668 if (i > 0 || MethodData::profile_return()) { 1669 // If return value type is profiled we may have no argument to profile. 1670 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp); 1671 add2reg(tmp, -i*TypeStackSlotEntries::per_arg_count()); 1672 compare64_and_branch(tmp, TypeStackSlotEntries::per_arg_count(), Assembler::bcondLow, done); 1673 } 1674 z_lg(tmp, Address(callee, Method::const_offset())); 1675 z_lgh(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); 1676 // Stack offset o (zero based) from the start of the argument 1677 // list. For n arguments translates into offset n - o - 1 from 1678 // the end of the argument list. But there is an extra slot at 1679 // the top of the stack. So the offset is n - o from Lesp. 1680 z_sg(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args)); 1681 z_sllg(tmp, tmp, Interpreter::logStackElementSize); 1682 Address stack_slot_addr(tmp, Z_esp); 1683 z_ltg(tmp, stack_slot_addr); 1684 1685 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); 1686 profile_obj_type(tmp, mdo_arg_addr, tmp, /*ltg did compare to 0*/ true); 1687 1688 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1689 add2reg(mdp, to_add); 1690 off_to_args += to_add; 1691 } 1692 1693 if (MethodData::profile_return()) { 1694 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp); 1695 add2reg(tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1696 } 1697 1698 bind(done); 1699 1700 if (MethodData::profile_return()) { 1701 // We're right after the type profile for the last 1702 // argument. Tmp is the number of cells left in the 1703 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1704 // if there's a return to profile. 1705 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 1706 z_sllg(tmp, tmp, exact_log2(DataLayout::cell_size)); 1707 z_agr(mdp, tmp); 1708 } 1709 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp); 1710 } else { 1711 assert(MethodData::profile_return(), "either profile call args or call ret"); 1712 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); 1713 } 1714 1715 // Mdp points right after the end of the 1716 // CallTypeData/VirtualCallTypeData, right after the cells for the 1717 // return value type if there's one. 1718 bind(profile_continue); 1719 } 1720 } 1721 1722 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { 1723 assert_different_registers(mdp, ret, tmp); 1724 if (ProfileInterpreter && MethodData::profile_return()) { 1725 Label profile_continue; 1726 1727 test_method_data_pointer(mdp, profile_continue); 1728 1729 if (MethodData::profile_return_jsr292_only()) { 1730 // If we don't profile all invoke bytecodes we must make sure 1731 // it's a bytecode we indeed profile. We can't go back to the 1732 // beginning of the ProfileData we intend to update to check its 1733 // type because we're right after it and we don't known its 1734 // length. 1735 NearLabel do_profile; 1736 Address bc(Z_bcp); 1737 z_lb(tmp, bc); 1738 compare32_and_branch(tmp, Bytecodes::_invokedynamic, Assembler::bcondEqual, do_profile); 1739 compare32_and_branch(tmp, Bytecodes::_invokehandle, Assembler::bcondEqual, do_profile); 1740 get_method(tmp); 1741 // Supplement to 8139891: _intrinsic_id exceeded 1-byte size limit. 1742 if (Method::intrinsic_id_size_in_bytes() == 1) { 1743 z_cli(Method::intrinsic_id_offset_in_bytes(), tmp, vmIntrinsics::_compiledLambdaForm); 1744 } else { 1745 assert(Method::intrinsic_id_size_in_bytes() == 2, "size error: check Method::_intrinsic_id"); 1746 z_lh(tmp, Method::intrinsic_id_offset_in_bytes(), Z_R0, tmp); 1747 z_chi(tmp, vmIntrinsics::_compiledLambdaForm); 1748 } 1749 z_brne(profile_continue); 1750 1751 bind(do_profile); 1752 } 1753 1754 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size())); 1755 profile_obj_type(ret, mdo_ret_addr, tmp); 1756 1757 bind(profile_continue); 1758 } 1759 } 1760 1761 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) { 1762 if (ProfileInterpreter && MethodData::profile_parameters()) { 1763 Label profile_continue, done; 1764 1765 test_method_data_pointer(mdp, profile_continue); 1766 1767 // Load the offset of the area within the MDO used for 1768 // parameters. If it's negative we're not profiling any parameters. 1769 Address parm_di_addr(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())); 1770 load_and_test_int2long(tmp1, parm_di_addr); 1771 z_brl(profile_continue); 1772 1773 // Compute a pointer to the area for parameters from the offset 1774 // and move the pointer to the slot for the last 1775 // parameters. Collect profiling from last parameter down. 1776 // mdo start + parameters offset + array length - 1 1777 1778 // Pointer to the parameter area in the MDO. 1779 z_agr(mdp, tmp1); 1780 1781 // Offset of the current profile entry to update. 1782 const Register entry_offset = tmp1; 1783 // entry_offset = array len in number of cells. 1784 z_lg(entry_offset, Address(mdp, ArrayData::array_len_offset())); 1785 // entry_offset (number of cells) = array len - size of 1 entry 1786 add2reg(entry_offset, -TypeStackSlotEntries::per_arg_count()); 1787 // entry_offset in bytes 1788 z_sllg(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1789 1790 Label loop; 1791 bind(loop); 1792 1793 Address arg_off(mdp, entry_offset, ParametersTypeData::stack_slot_offset(0)); 1794 Address arg_type(mdp, entry_offset, ParametersTypeData::type_offset(0)); 1795 1796 // Load offset on the stack from the slot for this parameter. 1797 z_lg(tmp2, arg_off); 1798 z_sllg(tmp2, tmp2, Interpreter::logStackElementSize); 1799 z_lcgr(tmp2); // Negate. 1800 1801 // Profile the parameter. 1802 z_ltg(tmp2, Address(Z_locals, tmp2)); 1803 profile_obj_type(tmp2, arg_type, tmp2, /*ltg did compare to 0*/ true); 1804 1805 // Go to next parameter. 1806 z_aghi(entry_offset, -TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size); 1807 z_brnl(loop); 1808 1809 bind(profile_continue); 1810 } 1811 } 1812 1813 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 1814 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 1815 int increment, 1816 Address mask, 1817 Register scratch, 1818 bool preloaded, 1819 branch_condition cond, 1820 Label *where) { 1821 assert_different_registers(counter_addr.base(), scratch); 1822 if (preloaded) { 1823 add2reg(scratch, increment); 1824 reg2mem_opt(scratch, counter_addr, false); 1825 } else { 1826 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment) && counter_addr.is_RSYform()) { 1827 z_alsi(counter_addr.disp20(), counter_addr.base(), increment); 1828 mem2reg_signed_opt(scratch, counter_addr); 1829 } else { 1830 mem2reg_signed_opt(scratch, counter_addr); 1831 add2reg(scratch, increment); 1832 reg2mem_opt(scratch, counter_addr, false); 1833 } 1834 } 1835 z_n(scratch, mask); 1836 if (where) { z_brc(cond, *where); } 1837 } 1838 1839 // Get MethodCounters object for given method. Lazily allocated if necessary. 1840 // method - Ptr to Method object. 1841 // Rcounters - Ptr to MethodCounters object associated with Method object. 1842 // skip - Exit point if MethodCounters object can't be created (OOM condition). 1843 void InterpreterMacroAssembler::get_method_counters(Register Rmethod, 1844 Register Rcounters, 1845 Label& skip) { 1846 assert_different_registers(Rmethod, Rcounters); 1847 1848 BLOCK_COMMENT("get MethodCounters object {"); 1849 1850 Label has_counters; 1851 load_and_test_long(Rcounters, Address(Rmethod, Method::method_counters_offset())); 1852 z_brnz(has_counters); 1853 1854 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), Rmethod, false); 1855 z_ltgr(Rcounters, Z_RET); // Runtime call returns MethodCounters object. 1856 z_brz(skip); // No MethodCounters, out of memory. 1857 1858 bind(has_counters); 1859 1860 BLOCK_COMMENT("} get MethodCounters object"); 1861 } 1862 1863 // Increment invocation counter in MethodCounters object. 1864 // Return (invocation_counter+backedge_counter) as "result" in RctrSum. 1865 // Counter values are all unsigned. 1866 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register RctrSum) { 1867 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 1868 assert_different_registers(Rcounters, RctrSum); 1869 1870 int increment = InvocationCounter::count_increment; 1871 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()); 1872 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset()); 1873 1874 BLOCK_COMMENT("Increment invocation counter {"); 1875 1876 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) { 1877 // Increment the invocation counter in place, 1878 // then add the incremented value to the backedge counter. 1879 z_l(RctrSum, be_counter_offset, Rcounters); 1880 z_alsi(inv_counter_offset, Rcounters, increment); // Atomic increment @no extra cost! 1881 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits. 1882 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters); 1883 } else { 1884 // This path is optimized for low register consumption 1885 // at the cost of somewhat higher operand delays. 1886 // It does not need an extra temp register. 1887 1888 // Update the invocation counter. 1889 z_l(RctrSum, inv_counter_offset, Rcounters); 1890 if (RctrSum == Z_R0) { 1891 z_ahi(RctrSum, increment); 1892 } else { 1893 add2reg(RctrSum, increment); 1894 } 1895 z_st(RctrSum, inv_counter_offset, Rcounters); 1896 1897 // Mask off the state bits. 1898 z_nilf(RctrSum, InvocationCounter::count_mask_value); 1899 1900 // Add the backedge counter to the updated invocation counter to 1901 // form the result. 1902 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters); 1903 } 1904 1905 BLOCK_COMMENT("} Increment invocation counter"); 1906 1907 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 1908 } 1909 1910 1911 // increment backedge counter in MethodCounters object. 1912 // return (invocation_counter+backedge_counter) as "result" in RctrSum 1913 // counter values are all unsigned! 1914 void InterpreterMacroAssembler::increment_backedge_counter(Register Rcounters, Register RctrSum) { 1915 assert(UseCompiler, "incrementing must be useful"); 1916 assert_different_registers(Rcounters, RctrSum); 1917 1918 int increment = InvocationCounter::count_increment; 1919 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()); 1920 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset()); 1921 1922 BLOCK_COMMENT("Increment backedge counter {"); 1923 1924 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) { 1925 // Increment the invocation counter in place, 1926 // then add the incremented value to the backedge counter. 1927 z_l(RctrSum, inv_counter_offset, Rcounters); 1928 z_alsi(be_counter_offset, Rcounters, increment); // Atomic increment @no extra cost! 1929 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits. 1930 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters); 1931 } else { 1932 // This path is optimized for low register consumption 1933 // at the cost of somewhat higher operand delays. 1934 // It does not need an extra temp register. 1935 1936 // Update the invocation counter. 1937 z_l(RctrSum, be_counter_offset, Rcounters); 1938 if (RctrSum == Z_R0) { 1939 z_ahi(RctrSum, increment); 1940 } else { 1941 add2reg(RctrSum, increment); 1942 } 1943 z_st(RctrSum, be_counter_offset, Rcounters); 1944 1945 // Mask off the state bits. 1946 z_nilf(RctrSum, InvocationCounter::count_mask_value); 1947 1948 // Add the backedge counter to the updated invocation counter to 1949 // form the result. 1950 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters); 1951 } 1952 1953 BLOCK_COMMENT("} Increment backedge counter"); 1954 1955 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 1956 } 1957 1958 // Add an InterpMonitorElem to stack (see frame_s390.hpp). 1959 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, 1960 Register Rtemp1, 1961 Register Rtemp2, 1962 Register Rtemp3) { 1963 1964 const Register Rcurr_slot = Rtemp1; 1965 const Register Rlimit = Rtemp2; 1966 const jint delta = -frame::interpreter_frame_monitor_size() * wordSize; 1967 1968 assert((delta & LongAlignmentMask) == 0, 1969 "sizeof BasicObjectLock must be even number of doublewords"); 1970 assert(2 * wordSize == -delta, "this works only as long as delta == -2*wordSize"); 1971 assert(Rcurr_slot != Z_R0, "Register must be usable as base register"); 1972 assert_different_registers(Rlimit, Rcurr_slot, Rtemp3); 1973 1974 get_monitors(Rlimit); 1975 1976 // Adjust stack pointer for additional monitor entry. 1977 resize_frame(RegisterOrConstant((intptr_t) delta), Z_fp, false); 1978 1979 if (!stack_is_empty) { 1980 // Must copy stack contents down. 1981 NearLabel next, done; 1982 1983 // Rtemp := addr(Tos), Z_esp is pointing below it! 1984 add2reg(Rcurr_slot, wordSize, Z_esp); 1985 1986 // Nothing to do, if already at monitor area. 1987 compareU64_and_branch(Rcurr_slot, Rlimit, bcondNotLow, done); 1988 1989 bind(next); 1990 1991 // Move one stack slot. 1992 mem2reg_opt(Rtemp3, Address(Rcurr_slot)); 1993 reg2mem_opt(Rtemp3, Address(Rcurr_slot, delta)); 1994 add2reg(Rcurr_slot, wordSize); 1995 compareU64_and_branch(Rcurr_slot, Rlimit, bcondLow, next); // Are we done? 1996 1997 bind(done); 1998 // Done copying stack. 1999 } 2000 2001 // Adjust expression stack and monitor pointers. 2002 add2reg(Z_esp, delta); 2003 add2reg(Rlimit, delta); 2004 save_monitors(Rlimit); 2005 } 2006 2007 // Note: Index holds the offset in bytes afterwards. 2008 // You can use this to store a new value (with Llocals as the base). 2009 void InterpreterMacroAssembler::access_local_int(Register index, Register dst) { 2010 z_sllg(index, index, LogBytesPerWord); 2011 mem2reg_opt(dst, Address(Z_locals, index), false); 2012 } 2013 2014 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2015 if (state == atos) { MacroAssembler::verify_oop(reg); } 2016 } 2017 2018 // Inline assembly for: 2019 // 2020 // if (thread is in interp_only_mode) { 2021 // InterpreterRuntime::post_method_entry(); 2022 // } 2023 2024 void InterpreterMacroAssembler::notify_method_entry() { 2025 2026 // JVMTI 2027 // Whenever JVMTI puts a thread in interp_only_mode, method 2028 // entry/exit events are sent for that thread to track stack 2029 // depth. If it is possible to enter interp_only_mode we add 2030 // the code to check if the event should be sent. 2031 if (JvmtiExport::can_post_interpreter_events()) { 2032 Label jvmti_post_done; 2033 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset())); 2034 z_bre(jvmti_post_done); 2035 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), /*check_exceptions=*/false); 2036 bind(jvmti_post_done); 2037 } 2038 } 2039 2040 // Inline assembly for: 2041 // 2042 // if (thread is in interp_only_mode) { 2043 // if (!native_method) save result 2044 // InterpreterRuntime::post_method_exit(); 2045 // if (!native_method) restore result 2046 // } 2047 // if (DTraceMethodProbes) { 2048 // SharedRuntime::dtrace_method_exit(thread, method); 2049 // } 2050 // 2051 // For native methods their result is stored in z_ijava_state.lresult 2052 // and z_ijava_state.fresult before coming here. 2053 // Java methods have their result stored in the expression stack. 2054 // 2055 // Notice the dependency to frame::interpreter_frame_result(). 2056 void InterpreterMacroAssembler::notify_method_exit(bool native_method, 2057 TosState state, 2058 NotifyMethodExitMode mode) { 2059 // JVMTI 2060 // Whenever JVMTI puts a thread in interp_only_mode, method 2061 // entry/exit events are sent for that thread to track stack 2062 // depth. If it is possible to enter interp_only_mode we add 2063 // the code to check if the event should be sent. 2064 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2065 Label jvmti_post_done; 2066 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset())); 2067 z_bre(jvmti_post_done); 2068 if (!native_method) push(state); // see frame::interpreter_frame_result() 2069 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), /*check_exceptions=*/false); 2070 if (!native_method) pop(state); 2071 bind(jvmti_post_done); 2072 } 2073 2074 #if 0 2075 // Dtrace currently not supported on z/Architecture. 2076 { 2077 SkipIfEqual skip(this, &DTraceMethodProbes, false); 2078 push(state); 2079 get_method(c_rarg1); 2080 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2081 r15_thread, c_rarg1); 2082 pop(state); 2083 } 2084 #endif 2085 } 2086 2087 void InterpreterMacroAssembler::skip_if_jvmti_mode(Label &Lskip, Register Rscratch) { 2088 if (!JvmtiExport::can_post_interpreter_events()) { 2089 return; 2090 } 2091 2092 load_and_test_int(Rscratch, Address(Z_thread, JavaThread::interp_only_mode_offset())); 2093 z_brnz(Lskip); 2094 2095 } 2096 2097 // Pop the topmost TOP_IJAVA_FRAME and set it's sender_sp as new Z_SP. 2098 // The return pc is loaded into the register return_pc. 2099 // 2100 // Registers updated: 2101 // return_pc - The return pc of the calling frame. 2102 // tmp1, tmp2 - scratch 2103 void InterpreterMacroAssembler::pop_interpreter_frame(Register return_pc, Register tmp1, Register tmp2) { 2104 // F0 Z_SP -> caller_sp (F1's) 2105 // ... 2106 // sender_sp (F1's) 2107 // ... 2108 // F1 Z_fp -> caller_sp (F2's) 2109 // return_pc (Continuation after return from F0.) 2110 // ... 2111 // F2 caller_sp 2112 2113 // Remove F0's activation. Restoring Z_SP to sender_sp reverts modifications 2114 // (a) by a c2i adapter and (b) by generate_fixed_frame(). 2115 // In case (a) the new top frame F1 is an unextended compiled frame. 2116 // In case (b) F1 is converted from PARENT_IJAVA_FRAME to TOP_IJAVA_FRAME. 2117 2118 // Case (b) seems to be redundant when returning to a interpreted caller, 2119 // because then the caller's top_frame_sp is installed as sp (see 2120 // TemplateInterpreterGenerator::generate_return_entry_for ()). But 2121 // pop_interpreter_frame() is also used in exception handling and there the 2122 // frame type of the caller is unknown, therefore top_frame_sp cannot be used, 2123 // so it is important that sender_sp is the caller's sp as TOP_IJAVA_FRAME. 2124 2125 Register R_f1_sender_sp = tmp1; 2126 Register R_f2_sp = tmp2; 2127 2128 // Tirst check the for the interpreter frame's magic. 2129 asm_assert_ijava_state_magic(R_f2_sp/*tmp*/); 2130 z_lg(R_f2_sp, _z_parent_ijava_frame_abi(callers_sp), Z_fp); 2131 z_lg(R_f1_sender_sp, _z_ijava_state_neg(sender_sp), Z_fp); 2132 if (return_pc->is_valid()) 2133 z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp); 2134 // Pop F0 by resizing to R_f1_sender_sp and using R_f2_sp as fp. 2135 resize_frame_absolute(R_f1_sender_sp, R_f2_sp, false/*load fp*/); 2136 2137 #ifdef ASSERT 2138 // The return_pc in the new top frame is dead... at least that's my 2139 // current understanding; to assert this I overwrite it. 2140 load_const_optimized(Z_ARG3, 0xb00b1); 2141 z_stg(Z_ARG3, _z_parent_ijava_frame_abi(return_pc), Z_SP); 2142 #endif 2143 } 2144 2145 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2146 if (VerifyFPU) { 2147 unimplemented("verfiyFPU"); 2148 } 2149 } 2150