1 /* 2 * Copyright (c) 2016, 2017, 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::cache_offset_in_bytes(), result); 375 z_lg(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result); 376 // JNIHandles::resolve(result) 377 z_lg(result, 0, result); // Load resolved references array itself. 378 #ifdef ASSERT 379 NearLabel index_ok; 380 z_lgf(Z_R0, Address(result, arrayOopDesc::length_offset_in_bytes())); 381 z_sllg(Z_R0, Z_R0, LogBytesPerHeapOop); 382 compare64_and_branch(tmp, Z_R0, Assembler::bcondLow, index_ok); 383 stop("resolved reference index out of bounds", 0x09256); 384 bind(index_ok); 385 #endif 386 z_agr(result, index); // Address of indexed array element. 387 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result); 388 } 389 390 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, 391 Register tmp, 392 int bcp_offset, 393 size_t index_size) { 394 BLOCK_COMMENT("get_cache_entry_pointer_at_bcp {"); 395 get_cache_and_index_at_bcp(cache, tmp, bcp_offset, index_size); 396 add2reg_with_index(cache, in_bytes(ConstantPoolCache::base_offset()), tmp, cache); 397 BLOCK_COMMENT("}"); 398 } 399 400 // Generate a subtype check: branch to ok_is_subtype if sub_klass is 401 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. 402 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 403 Register Rsuper_klass, 404 Register Rtmp1, 405 Register Rtmp2, 406 Label &ok_is_subtype) { 407 // Profile the not-null value's klass. 408 profile_typecheck(Rtmp1, Rsub_klass, Rtmp2); 409 410 // Do the check. 411 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype); 412 413 // Profile the failure of the check. 414 profile_typecheck_failed(Rtmp1, Rtmp2); 415 } 416 417 // Pop topmost element from stack. It just disappears. 418 // Useful if consumed previously by access via stackTop(). 419 void InterpreterMacroAssembler::popx(int len) { 420 add2reg(Z_esp, len*Interpreter::stackElementSize); 421 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 422 } 423 424 // Get Address object of stack top. No checks. No pop. 425 // Purpose: - Provide address of stack operand to exploit reg-mem operations. 426 // - Avoid RISC-like mem2reg - reg-reg-op sequence. 427 Address InterpreterMacroAssembler::stackTop() { 428 return Address(Z_esp, Interpreter::expr_offset_in_bytes(0)); 429 } 430 431 void InterpreterMacroAssembler::pop_i(Register r) { 432 z_l(r, Interpreter::expr_offset_in_bytes(0), Z_esp); 433 add2reg(Z_esp, Interpreter::stackElementSize); 434 assert_different_registers(r, Z_R1_scratch); 435 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 436 } 437 438 void InterpreterMacroAssembler::pop_ptr(Register r) { 439 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp); 440 add2reg(Z_esp, Interpreter::stackElementSize); 441 assert_different_registers(r, Z_R1_scratch); 442 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 443 } 444 445 void InterpreterMacroAssembler::pop_l(Register r) { 446 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp); 447 add2reg(Z_esp, 2*Interpreter::stackElementSize); 448 assert_different_registers(r, Z_R1_scratch); 449 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 450 } 451 452 void InterpreterMacroAssembler::pop_f(FloatRegister f) { 453 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), false); 454 add2reg(Z_esp, Interpreter::stackElementSize); 455 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 456 } 457 458 void InterpreterMacroAssembler::pop_d(FloatRegister f) { 459 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), true); 460 add2reg(Z_esp, 2*Interpreter::stackElementSize); 461 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 462 } 463 464 void InterpreterMacroAssembler::push_i(Register r) { 465 assert_different_registers(r, Z_R1_scratch); 466 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 467 z_st(r, Address(Z_esp)); 468 add2reg(Z_esp, -Interpreter::stackElementSize); 469 } 470 471 void InterpreterMacroAssembler::push_ptr(Register r) { 472 z_stg(r, Address(Z_esp)); 473 add2reg(Z_esp, -Interpreter::stackElementSize); 474 } 475 476 void InterpreterMacroAssembler::push_l(Register r) { 477 assert_different_registers(r, Z_R1_scratch); 478 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 479 int offset = -Interpreter::stackElementSize; 480 z_stg(r, Address(Z_esp, offset)); 481 clear_mem(Address(Z_esp), Interpreter::stackElementSize); 482 add2reg(Z_esp, 2 * offset); 483 } 484 485 void InterpreterMacroAssembler::push_f(FloatRegister f) { 486 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 487 freg2mem_opt(f, Address(Z_esp), false); 488 add2reg(Z_esp, -Interpreter::stackElementSize); 489 } 490 491 void InterpreterMacroAssembler::push_d(FloatRegister d) { 492 debug_only(verify_esp(Z_esp, Z_R1_scratch)); 493 int offset = -Interpreter::stackElementSize; 494 freg2mem_opt(d, Address(Z_esp, offset)); 495 add2reg(Z_esp, 2 * offset); 496 } 497 498 void InterpreterMacroAssembler::push(TosState state) { 499 verify_oop(Z_tos, state); 500 switch (state) { 501 case atos: push_ptr(); break; 502 case btos: push_i(); break; 503 case ztos: 504 case ctos: 505 case stos: push_i(); break; 506 case itos: push_i(); break; 507 case ltos: push_l(); break; 508 case ftos: push_f(); break; 509 case dtos: push_d(); break; 510 case vtos: /* nothing to do */ break; 511 default : ShouldNotReachHere(); 512 } 513 } 514 515 void InterpreterMacroAssembler::pop(TosState state) { 516 switch (state) { 517 case atos: pop_ptr(Z_tos); break; 518 case btos: pop_i(Z_tos); break; 519 case ztos: 520 case ctos: 521 case stos: pop_i(Z_tos); break; 522 case itos: pop_i(Z_tos); break; 523 case ltos: pop_l(Z_tos); break; 524 case ftos: pop_f(Z_ftos); break; 525 case dtos: pop_d(Z_ftos); break; 526 case vtos: /* nothing to do */ break; 527 default : ShouldNotReachHere(); 528 } 529 verify_oop(Z_tos, state); 530 } 531 532 // Helpers for swap and dup. 533 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 534 z_lg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n))); 535 } 536 537 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 538 z_stg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n))); 539 } 540 541 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted(Register method) { 542 // Satisfy interpreter calling convention (see generate_normal_entry()). 543 z_lgr(Z_R10, Z_SP); // Set sender sp (aka initial caller sp, aka unextended sp). 544 // Record top_frame_sp, because the callee might modify it, if it's compiled. 545 z_stg(Z_SP, _z_ijava_state_neg(top_frame_sp), Z_fp); 546 save_bcp(); 547 save_esp(); 548 z_lgr(Z_method, method); // Set Z_method (kills Z_fp!). 549 } 550 551 // Jump to from_interpreted entry of a call unless single stepping is possible 552 // in this thread in which case we must call the i2i entry. 553 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { 554 assert_different_registers(method, Z_R10 /*used for initial_caller_sp*/, temp); 555 prepare_to_jump_from_interpreted(method); 556 557 if (JvmtiExport::can_post_interpreter_events()) { 558 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 559 // compiled code in threads for which the event is enabled. Check here for 560 // interp_only_mode if these events CAN be enabled. 561 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset())); 562 MacroAssembler::load_and_test_int(Z_R0_scratch, Address(Z_thread, JavaThread::interp_only_mode_offset())); 563 z_bcr(bcondEqual, Z_R1_scratch); // Run compiled code if zero. 564 // Run interpreted. 565 z_lg(Z_R1_scratch, Address(method, Method::interpreter_entry_offset())); 566 z_br(Z_R1_scratch); 567 } else { 568 // Run compiled code. 569 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset())); 570 z_br(Z_R1_scratch); 571 } 572 } 573 574 #ifdef ASSERT 575 void InterpreterMacroAssembler::verify_esp(Register Resp, Register Rtemp) { 576 // About to read or write Resp[0]. 577 // Make sure it is not in the monitors or the TOP_IJAVA_FRAME_ABI. 578 address reentry = NULL; 579 580 { 581 // Check if the frame pointer in Z_fp is correct. 582 NearLabel OK; 583 z_cg(Z_fp, 0, Z_SP); 584 z_bre(OK); 585 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp"); 586 bind(OK); 587 } 588 { 589 // Resp must not point into or below the operand stack, 590 // i.e. IJAVA_STATE.monitors > Resp. 591 NearLabel OK; 592 Register Rmonitors = Rtemp; 593 z_lg(Rmonitors, _z_ijava_state_neg(monitors), Z_fp); 594 compareU64_and_branch(Rmonitors, Resp, bcondHigh, OK); 595 reentry = stop_chain_static(reentry, "too many pops: Z_esp points into monitor area"); 596 bind(OK); 597 } 598 { 599 // Resp may point to the last word of TOP_IJAVA_FRAME_ABI, but not below 600 // i.e. !(Z_SP + frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize > Resp). 601 NearLabel OK; 602 Register Rabi_bottom = Rtemp; 603 add2reg(Rabi_bottom, frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize, Z_SP); 604 compareU64_and_branch(Rabi_bottom, Resp, bcondNotHigh, OK); 605 reentry = stop_chain_static(reentry, "too many pushes: Z_esp points into TOP_IJAVA_FRAME_ABI"); 606 bind(OK); 607 } 608 } 609 610 void InterpreterMacroAssembler::asm_assert_ijava_state_magic(Register tmp) { 611 Label magic_ok; 612 load_const_optimized(tmp, frame::z_istate_magic_number); 613 z_cg(tmp, Address(Z_fp, _z_ijava_state_neg(magic))); 614 z_bre(magic_ok); 615 stop_static("error: wrong magic number in ijava_state access"); 616 bind(magic_ok); 617 } 618 #endif // ASSERT 619 620 void InterpreterMacroAssembler::save_bcp() { 621 z_stg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))); 622 asm_assert_ijava_state_magic(Z_bcp); 623 NOT_PRODUCT(z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)))); 624 } 625 626 void InterpreterMacroAssembler::restore_bcp() { 627 asm_assert_ijava_state_magic(Z_bcp); 628 z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))); 629 } 630 631 void InterpreterMacroAssembler::save_esp() { 632 z_stg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp))); 633 } 634 635 void InterpreterMacroAssembler::restore_esp() { 636 asm_assert_ijava_state_magic(Z_esp); 637 z_lg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp))); 638 } 639 640 void InterpreterMacroAssembler::get_monitors(Register reg) { 641 asm_assert_ijava_state_magic(reg); 642 mem2reg_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors))); 643 } 644 645 void InterpreterMacroAssembler::save_monitors(Register reg) { 646 reg2mem_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors))); 647 } 648 649 void InterpreterMacroAssembler::get_mdp(Register mdp) { 650 z_lg(mdp, _z_ijava_state_neg(mdx), Z_fp); 651 } 652 653 void InterpreterMacroAssembler::save_mdp(Register mdp) { 654 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp); 655 } 656 657 // Values that are only read (besides initialization). 658 void InterpreterMacroAssembler::restore_locals() { 659 asm_assert_ijava_state_magic(Z_locals); 660 z_lg(Z_locals, Address(Z_fp, _z_ijava_state_neg(locals))); 661 } 662 663 void InterpreterMacroAssembler::get_method(Register reg) { 664 asm_assert_ijava_state_magic(reg); 665 z_lg(reg, Address(Z_fp, _z_ijava_state_neg(method))); 666 } 667 668 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(Register Rdst, int bcp_offset, 669 signedOrNot is_signed) { 670 // Rdst is an 8-byte return value!!! 671 672 // Unaligned loads incur only a small penalty on z/Architecture. The penalty 673 // is a few (2..3) ticks, even when the load crosses a cache line 674 // boundary. In case of a cache miss, the stall could, of course, be 675 // much longer. 676 677 switch (is_signed) { 678 case Signed: 679 z_lgh(Rdst, bcp_offset, Z_R0, Z_bcp); 680 break; 681 case Unsigned: 682 z_llgh(Rdst, bcp_offset, Z_R0, Z_bcp); 683 break; 684 default: 685 ShouldNotReachHere(); 686 } 687 } 688 689 690 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(Register Rdst, int bcp_offset, 691 setCCOrNot set_cc) { 692 // Rdst is an 8-byte return value!!! 693 694 // Unaligned loads incur only a small penalty on z/Architecture. The penalty 695 // is a few (2..3) ticks, even when the load crosses a cache line 696 // boundary. In case of a cache miss, the stall could, of course, be 697 // much longer. 698 699 // Both variants implement a sign-extending int2long load. 700 if (set_cc == set_CC) { 701 load_and_test_int2long(Rdst, Address(Z_bcp, (intptr_t)bcp_offset)); 702 } else { 703 mem2reg_signed_opt( Rdst, Address(Z_bcp, (intptr_t)bcp_offset)); 704 } 705 } 706 707 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { 708 get_method(Rdst); 709 mem2reg_opt(Rdst, Address(Rdst, Method::const_offset())); 710 mem2reg_opt(Rdst, Address(Rdst, ConstMethod::constants_offset())); 711 } 712 713 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { 714 get_constant_pool(Rcpool); 715 mem2reg_opt(Rtags, Address(Rcpool, ConstantPool::tags_offset_in_bytes())); 716 } 717 718 // Unlock if synchronized method. 719 // 720 // Unlock the receiver if this is a synchronized method. 721 // Unlock any Java monitors from syncronized blocks. 722 // 723 // If there are locked Java monitors 724 // If throw_monitor_exception 725 // throws IllegalMonitorStateException 726 // Else if install_monitor_exception 727 // installs IllegalMonitorStateException 728 // Else 729 // no error processing 730 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, 731 bool throw_monitor_exception, 732 bool install_monitor_exception) { 733 NearLabel unlocked, unlock, no_unlock; 734 735 { 736 Register R_method = Z_ARG2; 737 Register R_do_not_unlock_if_synchronized = Z_ARG3; 738 739 // Get the value of _do_not_unlock_if_synchronized into G1_scratch. 740 const Address do_not_unlock_if_synchronized(Z_thread, 741 JavaThread::do_not_unlock_if_synchronized_offset()); 742 load_sized_value(R_do_not_unlock_if_synchronized, do_not_unlock_if_synchronized, 1, false /*unsigned*/); 743 z_mvi(do_not_unlock_if_synchronized, false); // Reset the flag. 744 745 // Check if synchronized method. 746 get_method(R_method); 747 verify_oop(Z_tos, state); 748 push(state); // Save tos/result. 749 testbit(method2_(R_method, access_flags), JVM_ACC_SYNCHRONIZED_BIT); 750 z_bfalse(unlocked); 751 752 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 753 // is set. 754 compareU64_and_branch(R_do_not_unlock_if_synchronized, (intptr_t)0L, bcondNotEqual, no_unlock); 755 } 756 757 // unlock monitor 758 759 // BasicObjectLock will be first in list, since this is a 760 // synchronized method. However, need to check that the object has 761 // not been unlocked by an explicit monitorexit bytecode. 762 const Address monitor(Z_fp, -(frame::z_ijava_state_size + (int) sizeof(BasicObjectLock))); 763 // We use Z_ARG2 so that if we go slow path it will be the correct 764 // register for unlock_object to pass to VM directly. 765 load_address(Z_ARG2, monitor); // Address of first monitor. 766 z_lg(Z_ARG3, Address(Z_ARG2, BasicObjectLock::obj_offset_in_bytes())); 767 compareU64_and_branch(Z_ARG3, (intptr_t)0L, bcondNotEqual, unlock); 768 769 if (throw_monitor_exception) { 770 // Entry already unlocked need to throw an exception. 771 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 772 should_not_reach_here(); 773 } else { 774 // Monitor already unlocked during a stack unroll. 775 // If requested, install an illegal_monitor_state_exception. 776 // Continue with stack unrolling. 777 if (install_monitor_exception) { 778 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); 779 } 780 z_bru(unlocked); 781 } 782 783 bind(unlock); 784 785 unlock_object(Z_ARG2); 786 787 bind(unlocked); 788 789 // I0, I1: Might contain return value 790 791 // Check that all monitors are unlocked. 792 { 793 NearLabel loop, exception, entry, restart; 794 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 795 // We use Z_ARG2 so that if we go slow path it will be the correct 796 // register for unlock_object to pass to VM directly. 797 Register R_current_monitor = Z_ARG2; 798 Register R_monitor_block_bot = Z_ARG1; 799 const Address monitor_block_top(Z_fp, _z_ijava_state_neg(monitors)); 800 const Address monitor_block_bot(Z_fp, -frame::z_ijava_state_size); 801 802 bind(restart); 803 // Starting with top-most entry. 804 z_lg(R_current_monitor, monitor_block_top); 805 // Points to word before bottom of monitor block. 806 load_address(R_monitor_block_bot, monitor_block_bot); 807 z_bru(entry); 808 809 // Entry already locked, need to throw exception. 810 bind(exception); 811 812 if (throw_monitor_exception) { 813 // Throw exception. 814 MacroAssembler::call_VM(noreg, 815 CAST_FROM_FN_PTR(address, InterpreterRuntime:: 816 throw_illegal_monitor_state_exception)); 817 should_not_reach_here(); 818 } else { 819 // Stack unrolling. Unlock object and install illegal_monitor_exception. 820 // Unlock does not block, so don't have to worry about the frame. 821 // We don't have to preserve c_rarg1 since we are going to throw an exception. 822 unlock_object(R_current_monitor); 823 if (install_monitor_exception) { 824 call_VM(noreg, CAST_FROM_FN_PTR(address, 825 InterpreterRuntime:: 826 new_illegal_monitor_state_exception)); 827 } 828 z_bru(restart); 829 } 830 831 bind(loop); 832 // Check if current entry is used. 833 load_and_test_long(Z_R0_scratch, Address(R_current_monitor, BasicObjectLock::obj_offset_in_bytes())); 834 z_brne(exception); 835 836 add2reg(R_current_monitor, entry_size); // Otherwise advance to next entry. 837 bind(entry); 838 compareU64_and_branch(R_current_monitor, R_monitor_block_bot, bcondNotEqual, loop); 839 } 840 841 bind(no_unlock); 842 pop(state); 843 verify_oop(Z_tos, state); 844 } 845 846 // remove activation 847 // 848 // Unlock the receiver if this is a synchronized method. 849 // Unlock any Java monitors from syncronized blocks. 850 // Remove the activation from the stack. 851 // 852 // If there are locked Java monitors 853 // If throw_monitor_exception 854 // throws IllegalMonitorStateException 855 // Else if install_monitor_exception 856 // installs IllegalMonitorStateException 857 // Else 858 // no error processing 859 void InterpreterMacroAssembler::remove_activation(TosState state, 860 Register return_pc, 861 bool throw_monitor_exception, 862 bool install_monitor_exception, 863 bool notify_jvmti) { 864 BLOCK_COMMENT("remove_activation {"); 865 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); 866 867 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. 868 notify_method_exit(false, state, notify_jvmti ? NotifyJVMTI : SkipNotifyJVMTI); 869 870 if (StackReservedPages > 0) { 871 BLOCK_COMMENT("reserved_stack_check:"); 872 // Test if reserved zone needs to be enabled. 873 Label no_reserved_zone_enabling; 874 875 // Compare frame pointers. There is no good stack pointer, as with stack 876 // frame compression we can get different SPs when we do calls. A subsequent 877 // call could have a smaller SP, so that this compare succeeds for an 878 // inner call of the method annotated with ReservedStack. 879 z_lg(Z_R0, Address(Z_SP, (intptr_t)_z_abi(callers_sp))); 880 z_clg(Z_R0, Address(Z_thread, JavaThread::reserved_stack_activation_offset())); // Compare with frame pointer in memory. 881 z_brl(no_reserved_zone_enabling); 882 883 // Enable reserved zone again, throw stack overflow exception. 884 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), Z_thread); 885 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError)); 886 887 should_not_reach_here(); 888 889 bind(no_reserved_zone_enabling); 890 } 891 892 verify_oop(Z_tos, state); 893 verify_thread(); 894 895 pop_interpreter_frame(return_pc, Z_ARG2, Z_ARG3); 896 BLOCK_COMMENT("} remove_activation"); 897 } 898 899 // lock object 900 // 901 // Registers alive 902 // monitor - Address of the BasicObjectLock to be used for locking, 903 // which must be initialized with the object to lock. 904 // object - Address of the object to be locked. 905 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) { 906 907 if (UseHeavyMonitors) { 908 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 909 monitor, /*check_for_exceptions=*/false); 910 return; 911 } 912 913 // template code: 914 // 915 // markOop displaced_header = obj->mark().set_unlocked(); 916 // monitor->lock()->set_displaced_header(displaced_header); 917 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 918 // // We stored the monitor address into the object's mark word. 919 // } else if (THREAD->is_lock_owned((address)displaced_header)) 920 // // Simple recursive case. 921 // monitor->lock()->set_displaced_header(NULL); 922 // } else { 923 // // Slow path. 924 // InterpreterRuntime::monitorenter(THREAD, monitor); 925 // } 926 927 const Register displaced_header = Z_ARG5; 928 const Register object_mark_addr = Z_ARG4; 929 const Register current_header = Z_ARG5; 930 931 NearLabel done; 932 NearLabel slow_case; 933 934 // markOop displaced_header = obj->mark().set_unlocked(); 935 936 // Load markOop from object into displaced_header. 937 z_lg(displaced_header, oopDesc::mark_offset_in_bytes(), object); 938 939 if (UseBiasedLocking) { 940 biased_locking_enter(object, displaced_header, Z_R1, Z_R0, done, &slow_case); 941 } 942 943 // Set displaced_header to be (markOop of object | UNLOCK_VALUE). 944 z_oill(displaced_header, markOopDesc::unlocked_value); 945 946 // monitor->lock()->set_displaced_header(displaced_header); 947 948 // Initialize the box (Must happen before we update the object mark!). 949 z_stg(displaced_header, BasicObjectLock::lock_offset_in_bytes() + 950 BasicLock::displaced_header_offset_in_bytes(), monitor); 951 952 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) { 953 954 // Store stack address of the BasicObjectLock (this is monitor) into object. 955 add2reg(object_mark_addr, oopDesc::mark_offset_in_bytes(), object); 956 957 z_csg(displaced_header, monitor, 0, object_mark_addr); 958 assert(current_header==displaced_header, "must be same register"); // Identified two registers from z/Architecture. 959 960 z_bre(done); 961 962 // } else if (THREAD->is_lock_owned((address)displaced_header)) 963 // // Simple recursive case. 964 // monitor->lock()->set_displaced_header(NULL); 965 966 // We did not see an unlocked object so try the fast recursive case. 967 968 // Check if owner is self by comparing the value in the markOop of object 969 // (current_header) with the stack pointer. 970 z_sgr(current_header, Z_SP); 971 972 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 973 974 // The prior sequence "LGR, NGR, LTGR" can be done better 975 // (Z_R1 is temp and not used after here). 976 load_const_optimized(Z_R0, (~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place)); 977 z_ngr(Z_R0, current_header); // AND sets CC (result eq/ne 0) 978 979 // If condition is true we are done and hence we can store 0 in the displaced 980 // header indicating it is a recursive lock and be done. 981 z_brne(slow_case); 982 z_release(); // Membar unnecessary on zarch AND because the above csg does a sync before and after. 983 z_stg(Z_R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() + 984 BasicLock::displaced_header_offset_in_bytes(), monitor); 985 z_bru(done); 986 987 // } else { 988 // // Slow path. 989 // InterpreterRuntime::monitorenter(THREAD, monitor); 990 991 // None of the above fast optimizations worked so we have to get into the 992 // slow case of monitor enter. 993 bind(slow_case); 994 995 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 996 monitor, /*check_for_exceptions=*/false); 997 998 // } 999 1000 bind(done); 1001 } 1002 1003 // Unlocks an object. Used in monitorexit bytecode and remove_activation. 1004 // 1005 // Registers alive 1006 // monitor - address of the BasicObjectLock to be used for locking, 1007 // which must be initialized with the object to lock. 1008 // 1009 // Throw IllegalMonitorException if object is not locked by current thread. 1010 void InterpreterMacroAssembler::unlock_object(Register monitor, Register object) { 1011 1012 if (UseHeavyMonitors) { 1013 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1014 monitor, /*check_for_exceptions=*/ true); 1015 return; 1016 } 1017 1018 // else { 1019 // template code: 1020 // 1021 // if ((displaced_header = monitor->displaced_header()) == NULL) { 1022 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 1023 // monitor->set_obj(NULL); 1024 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 1025 // // We swapped the unlocked mark in displaced_header into the object's mark word. 1026 // monitor->set_obj(NULL); 1027 // } else { 1028 // // Slow path. 1029 // InterpreterRuntime::monitorexit(THREAD, monitor); 1030 // } 1031 1032 const Register displaced_header = Z_ARG4; 1033 const Register current_header = Z_R1; 1034 Address obj_entry(monitor, BasicObjectLock::obj_offset_in_bytes()); 1035 Label done; 1036 1037 if (object == noreg) { 1038 // In the template interpreter, we must assure that the object 1039 // entry in the monitor is cleared on all paths. Thus we move 1040 // loading up to here, and clear the entry afterwards. 1041 object = Z_ARG3; // Use Z_ARG3 if caller didn't pass object. 1042 z_lg(object, obj_entry); 1043 } 1044 1045 assert_different_registers(monitor, object, displaced_header, current_header); 1046 1047 // if ((displaced_header = monitor->displaced_header()) == NULL) { 1048 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. 1049 // monitor->set_obj(NULL); 1050 1051 clear_mem(obj_entry, sizeof(oop)); 1052 1053 if (UseBiasedLocking) { 1054 // The object address from the monitor is in object. 1055 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0"); 1056 biased_locking_exit(object, displaced_header, done); 1057 } 1058 1059 // Test first if we are in the fast recursive case. 1060 MacroAssembler::load_and_test_long(displaced_header, 1061 Address(monitor, BasicObjectLock::lock_offset_in_bytes() + 1062 BasicLock::displaced_header_offset_in_bytes())); 1063 z_bre(done); // displaced_header == 0 -> goto done 1064 1065 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { 1066 // // We swapped the unlocked mark in displaced_header into the object's mark word. 1067 // monitor->set_obj(NULL); 1068 1069 // If we still have a lightweight lock, unlock the object and be done. 1070 1071 // The markword is expected to be at offset 0. 1072 assert(oopDesc::mark_offset_in_bytes() == 0, "unlock_object: review code below"); 1073 1074 // We have the displaced header in displaced_header. If the lock is still 1075 // lightweight, it will contain the monitor address and we'll store the 1076 // displaced header back into the object's mark word. 1077 z_lgr(current_header, monitor); 1078 z_csg(current_header, displaced_header, 0, object); 1079 z_bre(done); 1080 1081 // } else { 1082 // // Slow path. 1083 // InterpreterRuntime::monitorexit(THREAD, monitor); 1084 1085 // The lock has been converted into a heavy lock and hence 1086 // we need to get into the slow case. 1087 z_stg(object, obj_entry); // Restore object entry, has been cleared above. 1088 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1089 monitor, /*check_for_exceptions=*/false); 1090 1091 // } 1092 1093 bind(done); 1094 } 1095 1096 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) { 1097 assert(ProfileInterpreter, "must be profiling interpreter"); 1098 load_and_test_long(mdp, Address(Z_fp, _z_ijava_state_neg(mdx))); 1099 z_brz(zero_continue); 1100 } 1101 1102 // Set the method data pointer for the current bcp. 1103 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1104 assert(ProfileInterpreter, "must be profiling interpreter"); 1105 Label set_mdp; 1106 Register mdp = Z_ARG4; 1107 Register method = Z_ARG5; 1108 1109 get_method(method); 1110 // Test MDO to avoid the call if it is NULL. 1111 load_and_test_long(mdp, method2_(method, method_data)); 1112 z_brz(set_mdp); 1113 1114 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), method, Z_bcp); 1115 // Z_RET: mdi 1116 // Mdo is guaranteed to be non-zero here, we checked for it before the call. 1117 assert(method->is_nonvolatile(), "choose nonvolatile reg or reload from frame"); 1118 z_lg(mdp, method2_(method, method_data)); // Must reload, mdp is volatile reg. 1119 add2reg_with_index(mdp, in_bytes(MethodData::data_offset()), Z_RET, mdp); 1120 1121 bind(set_mdp); 1122 save_mdp(mdp); 1123 } 1124 1125 void InterpreterMacroAssembler::verify_method_data_pointer() { 1126 assert(ProfileInterpreter, "must be profiling interpreter"); 1127 #ifdef ASSERT 1128 NearLabel verify_continue; 1129 Register bcp_expected = Z_ARG3; 1130 Register mdp = Z_ARG4; 1131 Register method = Z_ARG5; 1132 1133 test_method_data_pointer(mdp, verify_continue); // If mdp is zero, continue 1134 get_method(method); 1135 1136 // If the mdp is valid, it will point to a DataLayout header which is 1137 // consistent with the bcp. The converse is highly probable also. 1138 load_sized_value(bcp_expected, Address(mdp, DataLayout::bci_offset()), 2, false /*signed*/); 1139 z_ag(bcp_expected, Address(method, Method::const_offset())); 1140 load_address(bcp_expected, Address(bcp_expected, ConstMethod::codes_offset())); 1141 compareU64_and_branch(bcp_expected, Z_bcp, bcondEqual, verify_continue); 1142 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), method, Z_bcp, mdp); 1143 bind(verify_continue); 1144 #endif // ASSERT 1145 } 1146 1147 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) { 1148 assert(ProfileInterpreter, "must be profiling interpreter"); 1149 z_stg(value, constant, mdp_in); 1150 } 1151 1152 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1153 int constant, 1154 Register tmp, 1155 bool decrement) { 1156 assert_different_registers(mdp_in, tmp); 1157 // counter address 1158 Address data(mdp_in, constant); 1159 const int delta = decrement ? -DataLayout::counter_increment : DataLayout::counter_increment; 1160 add2mem_64(Address(mdp_in, constant), delta, tmp); 1161 } 1162 1163 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, 1164 int flag_byte_constant) { 1165 assert(ProfileInterpreter, "must be profiling interpreter"); 1166 // Set the flag. 1167 z_oi(Address(mdp_in, DataLayout::flags_offset()), flag_byte_constant); 1168 } 1169 1170 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, 1171 int offset, 1172 Register value, 1173 Register test_value_out, 1174 Label& not_equal_continue) { 1175 assert(ProfileInterpreter, "must be profiling interpreter"); 1176 if (test_value_out == noreg) { 1177 z_cg(value, Address(mdp_in, offset)); 1178 z_brne(not_equal_continue); 1179 } else { 1180 // Put the test value into a register, so caller can use it: 1181 z_lg(test_value_out, Address(mdp_in, offset)); 1182 compareU64_and_branch(test_value_out, value, bcondNotEqual, not_equal_continue); 1183 } 1184 } 1185 1186 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) { 1187 update_mdp_by_offset(mdp_in, noreg, offset_of_disp); 1188 } 1189 1190 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1191 Register dataidx, 1192 int offset_of_disp) { 1193 assert(ProfileInterpreter, "must be profiling interpreter"); 1194 Address disp_address(mdp_in, dataidx, offset_of_disp); 1195 Assembler::z_ag(mdp_in, disp_address); 1196 save_mdp(mdp_in); 1197 } 1198 1199 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) { 1200 assert(ProfileInterpreter, "must be profiling interpreter"); 1201 add2reg(mdp_in, constant); 1202 save_mdp(mdp_in); 1203 } 1204 1205 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { 1206 assert(ProfileInterpreter, "must be profiling interpreter"); 1207 assert(return_bci->is_nonvolatile(), "choose nonvolatile reg or save/restore"); 1208 call_VM(noreg, 1209 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), 1210 return_bci); 1211 } 1212 1213 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) { 1214 if (ProfileInterpreter) { 1215 Label profile_continue; 1216 1217 // If no method data exists, go to profile_continue. 1218 // Otherwise, assign to mdp. 1219 test_method_data_pointer(mdp, profile_continue); 1220 1221 // We are taking a branch. Increment the taken count. 1222 // We inline increment_mdp_data_at to return bumped_count in a register 1223 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); 1224 Address data(mdp, JumpData::taken_offset()); 1225 z_lg(bumped_count, data); 1226 // 64-bit overflow is very unlikely. Saturation to 32-bit values is 1227 // performed when reading the counts. 1228 add2reg(bumped_count, DataLayout::counter_increment); 1229 z_stg(bumped_count, data); // Store back out 1230 1231 // The method data pointer needs to be updated to reflect the new target. 1232 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); 1233 bind(profile_continue); 1234 } 1235 } 1236 1237 // Kills Z_R1_scratch. 1238 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { 1239 if (ProfileInterpreter) { 1240 Label profile_continue; 1241 1242 // If no method data exists, go to profile_continue. 1243 test_method_data_pointer(mdp, profile_continue); 1244 1245 // We are taking a branch. Increment the not taken count. 1246 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()), Z_R1_scratch); 1247 1248 // The method data pointer needs to be updated to correspond to 1249 // the next bytecode. 1250 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); 1251 bind(profile_continue); 1252 } 1253 } 1254 1255 // Kills: Z_R1_scratch. 1256 void InterpreterMacroAssembler::profile_call(Register mdp) { 1257 if (ProfileInterpreter) { 1258 Label profile_continue; 1259 1260 // If no method data exists, go to profile_continue. 1261 test_method_data_pointer(mdp, profile_continue); 1262 1263 // We are making a call. Increment the count. 1264 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1265 1266 // The method data pointer needs to be updated to reflect the new target. 1267 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); 1268 bind(profile_continue); 1269 } 1270 } 1271 1272 void InterpreterMacroAssembler::profile_final_call(Register mdp) { 1273 if (ProfileInterpreter) { 1274 Label profile_continue; 1275 1276 // If no method data exists, go to profile_continue. 1277 test_method_data_pointer(mdp, profile_continue); 1278 1279 // We are making a call. Increment the count. 1280 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1281 1282 // The method data pointer needs to be updated to reflect the new target. 1283 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1284 bind(profile_continue); 1285 } 1286 } 1287 1288 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1289 Register mdp, 1290 Register reg2, 1291 bool receiver_can_be_null) { 1292 if (ProfileInterpreter) { 1293 NearLabel profile_continue; 1294 1295 // If no method data exists, go to profile_continue. 1296 test_method_data_pointer(mdp, profile_continue); 1297 1298 NearLabel skip_receiver_profile; 1299 if (receiver_can_be_null) { 1300 NearLabel not_null; 1301 compareU64_and_branch(receiver, (intptr_t)0L, bcondNotEqual, not_null); 1302 // We are making a call. Increment the count for null receiver. 1303 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1304 z_bru(skip_receiver_profile); 1305 bind(not_null); 1306 } 1307 1308 // Record the receiver type. 1309 record_klass_in_profile(receiver, mdp, reg2, true); 1310 bind(skip_receiver_profile); 1311 1312 // The method data pointer needs to be updated to reflect the new target. 1313 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1314 bind(profile_continue); 1315 } 1316 } 1317 1318 // This routine creates a state machine for updating the multi-row 1319 // type profile at a virtual call site (or other type-sensitive bytecode). 1320 // The machine visits each row (of receiver/count) until the receiver type 1321 // is found, or until it runs out of rows. At the same time, it remembers 1322 // the location of the first empty row. (An empty row records null for its 1323 // receiver, and can be allocated for a newly-observed receiver type.) 1324 // Because there are two degrees of freedom in the state, a simple linear 1325 // search will not work; it must be a decision tree. Hence this helper 1326 // function is recursive, to generate the required tree structured code. 1327 // It's the interpreter, so we are trading off code space for speed. 1328 // See below for example code. 1329 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1330 Register receiver, Register mdp, 1331 Register reg2, int start_row, 1332 Label& done, bool is_virtual_call) { 1333 if (TypeProfileWidth == 0) { 1334 if (is_virtual_call) { 1335 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1336 } 1337 return; 1338 } 1339 1340 int last_row = VirtualCallData::row_limit() - 1; 1341 assert(start_row <= last_row, "must be work left to do"); 1342 // Test this row for both the receiver and for null. 1343 // Take any of three different outcomes: 1344 // 1. found receiver => increment count and goto done 1345 // 2. found null => keep looking for case 1, maybe allocate this cell 1346 // 3. found something else => keep looking for cases 1 and 2 1347 // Case 3 is handled by a recursive call. 1348 for (int row = start_row; row <= last_row; row++) { 1349 NearLabel next_test; 1350 bool test_for_null_also = (row == start_row); 1351 1352 // See if the receiver is receiver[n]. 1353 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); 1354 test_mdp_data_at(mdp, recvr_offset, receiver, 1355 (test_for_null_also ? reg2 : noreg), 1356 next_test); 1357 // (Reg2 now contains the receiver from the CallData.) 1358 1359 // The receiver is receiver[n]. Increment count[n]. 1360 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); 1361 increment_mdp_data_at(mdp, count_offset); 1362 z_bru(done); 1363 bind(next_test); 1364 1365 if (test_for_null_also) { 1366 Label found_null; 1367 // Failed the equality check on receiver[n]... Test for null. 1368 z_ltgr(reg2, reg2); 1369 if (start_row == last_row) { 1370 // The only thing left to do is handle the null case. 1371 if (is_virtual_call) { 1372 z_brz(found_null); 1373 // Receiver did not match any saved receiver and there is no empty row for it. 1374 // Increment total counter to indicate polymorphic case. 1375 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1376 z_bru(done); 1377 bind(found_null); 1378 } else { 1379 z_brnz(done); 1380 } 1381 break; 1382 } 1383 // Since null is rare, make it be the branch-taken case. 1384 z_brz(found_null); 1385 1386 // Put all the "Case 3" tests here. 1387 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call); 1388 1389 // Found a null. Keep searching for a matching receiver, 1390 // but remember that this is an empty (unused) slot. 1391 bind(found_null); 1392 } 1393 } 1394 1395 // In the fall-through case, we found no matching receiver, but we 1396 // observed the receiver[start_row] is NULL. 1397 1398 // Fill in the receiver field and increment the count. 1399 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); 1400 set_mdp_data_at(mdp, recvr_offset, receiver); 1401 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); 1402 load_const_optimized(reg2, DataLayout::counter_increment); 1403 set_mdp_data_at(mdp, count_offset, reg2); 1404 if (start_row > 0) { 1405 z_bru(done); 1406 } 1407 } 1408 1409 // Example state machine code for three profile rows: 1410 // // main copy of decision tree, rooted at row[1] 1411 // if (row[0].rec == rec) { row[0].incr(); goto done; } 1412 // if (row[0].rec != NULL) { 1413 // // inner copy of decision tree, rooted at row[1] 1414 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1415 // if (row[1].rec != NULL) { 1416 // // degenerate decision tree, rooted at row[2] 1417 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1418 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow 1419 // row[2].init(rec); goto done; 1420 // } else { 1421 // // remember row[1] is empty 1422 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1423 // row[1].init(rec); goto done; 1424 // } 1425 // } else { 1426 // // remember row[0] is empty 1427 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1428 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1429 // row[0].init(rec); goto done; 1430 // } 1431 // done: 1432 1433 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1434 Register mdp, Register reg2, 1435 bool is_virtual_call) { 1436 assert(ProfileInterpreter, "must be profiling"); 1437 Label done; 1438 1439 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); 1440 1441 bind (done); 1442 } 1443 1444 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) { 1445 if (ProfileInterpreter) { 1446 NearLabel profile_continue; 1447 uint row; 1448 1449 // If no method data exists, go to profile_continue. 1450 test_method_data_pointer(mdp, profile_continue); 1451 1452 // Update the total ret count. 1453 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1454 1455 for (row = 0; row < RetData::row_limit(); row++) { 1456 NearLabel next_test; 1457 1458 // See if return_bci is equal to bci[n]: 1459 test_mdp_data_at(mdp, 1460 in_bytes(RetData::bci_offset(row)), 1461 return_bci, noreg, 1462 next_test); 1463 1464 // Return_bci is equal to bci[n]. Increment the count. 1465 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); 1466 1467 // The method data pointer needs to be updated to reflect the new target. 1468 update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row))); 1469 z_bru(profile_continue); 1470 bind(next_test); 1471 } 1472 1473 update_mdp_for_ret(return_bci); 1474 1475 bind(profile_continue); 1476 } 1477 } 1478 1479 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { 1480 if (ProfileInterpreter) { 1481 Label profile_continue; 1482 1483 // If no method data exists, go to profile_continue. 1484 test_method_data_pointer(mdp, profile_continue); 1485 1486 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1487 1488 // The method data pointer needs to be updated. 1489 int mdp_delta = in_bytes(BitData::bit_data_size()); 1490 if (TypeProfileCasts) { 1491 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1492 } 1493 update_mdp_by_constant(mdp, mdp_delta); 1494 1495 bind(profile_continue); 1496 } 1497 } 1498 1499 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp, Register tmp) { 1500 if (ProfileInterpreter && TypeProfileCasts) { 1501 Label profile_continue; 1502 1503 // If no method data exists, go to profile_continue. 1504 test_method_data_pointer(mdp, profile_continue); 1505 1506 int count_offset = in_bytes(CounterData::count_offset()); 1507 // Back up the address, since we have already bumped the mdp. 1508 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1509 1510 // *Decrement* the counter. We expect to see zero or small negatives. 1511 increment_mdp_data_at(mdp, count_offset, tmp, true); 1512 1513 bind (profile_continue); 1514 } 1515 } 1516 1517 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { 1518 if (ProfileInterpreter) { 1519 Label profile_continue; 1520 1521 // If no method data exists, go to profile_continue. 1522 test_method_data_pointer(mdp, profile_continue); 1523 1524 // The method data pointer needs to be updated. 1525 int mdp_delta = in_bytes(BitData::bit_data_size()); 1526 if (TypeProfileCasts) { 1527 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1528 1529 // Record the object type. 1530 record_klass_in_profile(klass, mdp, reg2, false); 1531 } 1532 update_mdp_by_constant(mdp, mdp_delta); 1533 1534 bind(profile_continue); 1535 } 1536 } 1537 1538 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { 1539 if (ProfileInterpreter) { 1540 Label profile_continue; 1541 1542 // If no method data exists, go to profile_continue. 1543 test_method_data_pointer(mdp, profile_continue); 1544 1545 // Update the default case count. 1546 increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset())); 1547 1548 // The method data pointer needs to be updated. 1549 update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset())); 1550 1551 bind(profile_continue); 1552 } 1553 } 1554 1555 // Kills: index, scratch1, scratch2. 1556 void InterpreterMacroAssembler::profile_switch_case(Register index, 1557 Register mdp, 1558 Register scratch1, 1559 Register scratch2) { 1560 if (ProfileInterpreter) { 1561 Label profile_continue; 1562 assert_different_registers(index, mdp, scratch1, scratch2); 1563 1564 // If no method data exists, go to profile_continue. 1565 test_method_data_pointer(mdp, profile_continue); 1566 1567 // Build the base (index * per_case_size_in_bytes()) + 1568 // case_array_offset_in_bytes(). 1569 z_sllg(index, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); 1570 add2reg(index, in_bytes(MultiBranchData::case_array_offset())); 1571 1572 // Add the calculated base to the mdp -> address of the case' data. 1573 Address case_data_addr(mdp, index); 1574 Register case_data = scratch1; 1575 load_address(case_data, case_data_addr); 1576 1577 // Update the case count. 1578 increment_mdp_data_at(case_data, 1579 in_bytes(MultiBranchData::relative_count_offset()), 1580 scratch2); 1581 1582 // The method data pointer needs to be updated. 1583 update_mdp_by_offset(mdp, 1584 index, 1585 in_bytes(MultiBranchData::relative_displacement_offset())); 1586 1587 bind(profile_continue); 1588 } 1589 } 1590 1591 // kills: R0, R1, flags, loads klass from obj (if not null) 1592 void InterpreterMacroAssembler::profile_obj_type(Register obj, Address mdo_addr, Register klass, bool cmp_done) { 1593 NearLabel null_seen, init_klass, do_nothing, do_update; 1594 1595 // Klass = obj is allowed. 1596 const Register tmp = Z_R1; 1597 assert_different_registers(obj, mdo_addr.base(), tmp, Z_R0); 1598 assert_different_registers(klass, mdo_addr.base(), tmp, Z_R0); 1599 1600 z_lg(tmp, mdo_addr); 1601 if (cmp_done) { 1602 z_brz(null_seen); 1603 } else { 1604 compareU64_and_branch(obj, (intptr_t)0, Assembler::bcondEqual, null_seen); 1605 } 1606 1607 verify_oop(obj); 1608 load_klass(klass, obj); 1609 1610 // Klass seen before, nothing to do (regardless of unknown bit). 1611 z_lgr(Z_R0, tmp); 1612 assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction"); 1613 z_nill(Z_R0, TypeEntries::type_klass_mask & 0xFFFF); 1614 compareU64_and_branch(Z_R0, klass, Assembler::bcondEqual, do_nothing); 1615 1616 // Already unknown. Nothing to do anymore. 1617 z_tmll(tmp, TypeEntries::type_unknown); 1618 z_brc(Assembler::bcondAllOne, do_nothing); 1619 1620 z_lgr(Z_R0, tmp); 1621 assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction"); 1622 z_nill(Z_R0, TypeEntries::type_mask & 0xFFFF); 1623 compareU64_and_branch(Z_R0, (intptr_t)0, Assembler::bcondEqual, init_klass); 1624 1625 // Different than before. Cannot keep accurate profile. 1626 z_oill(tmp, TypeEntries::type_unknown); 1627 z_bru(do_update); 1628 1629 bind(init_klass); 1630 // Combine klass and null_seen bit (only used if (tmp & type_mask)==0). 1631 z_ogr(tmp, klass); 1632 z_bru(do_update); 1633 1634 bind(null_seen); 1635 // Set null_seen if obj is 0. 1636 z_oill(tmp, TypeEntries::null_seen); 1637 // fallthru: z_bru(do_update); 1638 1639 bind(do_update); 1640 z_stg(tmp, mdo_addr); 1641 1642 bind(do_nothing); 1643 } 1644 1645 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { 1646 if (!ProfileInterpreter) { 1647 return; 1648 } 1649 1650 assert_different_registers(mdp, callee, tmp); 1651 1652 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1653 Label profile_continue; 1654 1655 test_method_data_pointer(mdp, profile_continue); 1656 1657 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1658 1659 z_cliy(in_bytes(DataLayout::tag_offset()) - off_to_start, mdp, 1660 is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 1661 z_brne(profile_continue); 1662 1663 if (MethodData::profile_arguments()) { 1664 NearLabel done; 1665 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1666 add2reg(mdp, off_to_args); 1667 1668 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1669 if (i > 0 || MethodData::profile_return()) { 1670 // If return value type is profiled we may have no argument to profile. 1671 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp); 1672 add2reg(tmp, -i*TypeStackSlotEntries::per_arg_count()); 1673 compare64_and_branch(tmp, TypeStackSlotEntries::per_arg_count(), Assembler::bcondLow, done); 1674 } 1675 z_lg(tmp, Address(callee, Method::const_offset())); 1676 z_lgh(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); 1677 // Stack offset o (zero based) from the start of the argument 1678 // list. For n arguments translates into offset n - o - 1 from 1679 // the end of the argument list. But there is an extra slot at 1680 // the top of the stack. So the offset is n - o from Lesp. 1681 z_sg(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args)); 1682 z_sllg(tmp, tmp, Interpreter::logStackElementSize); 1683 Address stack_slot_addr(tmp, Z_esp); 1684 z_ltg(tmp, stack_slot_addr); 1685 1686 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); 1687 profile_obj_type(tmp, mdo_arg_addr, tmp, /*ltg did compare to 0*/ true); 1688 1689 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1690 add2reg(mdp, to_add); 1691 off_to_args += to_add; 1692 } 1693 1694 if (MethodData::profile_return()) { 1695 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp); 1696 add2reg(tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1697 } 1698 1699 bind(done); 1700 1701 if (MethodData::profile_return()) { 1702 // We're right after the type profile for the last 1703 // argument. Tmp is the number of cells left in the 1704 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1705 // if there's a return to profile. 1706 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 1707 z_sllg(tmp, tmp, exact_log2(DataLayout::cell_size)); 1708 z_agr(mdp, tmp); 1709 } 1710 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp); 1711 } else { 1712 assert(MethodData::profile_return(), "either profile call args or call ret"); 1713 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); 1714 } 1715 1716 // Mdp points right after the end of the 1717 // CallTypeData/VirtualCallTypeData, right after the cells for the 1718 // return value type if there's one. 1719 bind(profile_continue); 1720 } 1721 } 1722 1723 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { 1724 assert_different_registers(mdp, ret, tmp); 1725 if (ProfileInterpreter && MethodData::profile_return()) { 1726 Label profile_continue; 1727 1728 test_method_data_pointer(mdp, profile_continue); 1729 1730 if (MethodData::profile_return_jsr292_only()) { 1731 // If we don't profile all invoke bytecodes we must make sure 1732 // it's a bytecode we indeed profile. We can't go back to the 1733 // beginning of the ProfileData we intend to update to check its 1734 // type because we're right after it and we don't known its 1735 // length. 1736 NearLabel do_profile; 1737 Address bc(Z_bcp); 1738 z_lb(tmp, bc); 1739 compare32_and_branch(tmp, Bytecodes::_invokedynamic, Assembler::bcondEqual, do_profile); 1740 compare32_and_branch(tmp, Bytecodes::_invokehandle, Assembler::bcondEqual, do_profile); 1741 get_method(tmp); 1742 // Supplement to 8139891: _intrinsic_id exceeded 1-byte size limit. 1743 if (Method::intrinsic_id_size_in_bytes() == 1) { 1744 z_cli(Method::intrinsic_id_offset_in_bytes(), tmp, vmIntrinsics::_compiledLambdaForm); 1745 } else { 1746 assert(Method::intrinsic_id_size_in_bytes() == 2, "size error: check Method::_intrinsic_id"); 1747 z_lh(tmp, Method::intrinsic_id_offset_in_bytes(), Z_R0, tmp); 1748 z_chi(tmp, vmIntrinsics::_compiledLambdaForm); 1749 } 1750 z_brne(profile_continue); 1751 1752 bind(do_profile); 1753 } 1754 1755 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size())); 1756 profile_obj_type(ret, mdo_ret_addr, tmp); 1757 1758 bind(profile_continue); 1759 } 1760 } 1761 1762 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) { 1763 if (ProfileInterpreter && MethodData::profile_parameters()) { 1764 Label profile_continue, done; 1765 1766 test_method_data_pointer(mdp, profile_continue); 1767 1768 // Load the offset of the area within the MDO used for 1769 // parameters. If it's negative we're not profiling any parameters. 1770 Address parm_di_addr(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())); 1771 load_and_test_int2long(tmp1, parm_di_addr); 1772 z_brl(profile_continue); 1773 1774 // Compute a pointer to the area for parameters from the offset 1775 // and move the pointer to the slot for the last 1776 // parameters. Collect profiling from last parameter down. 1777 // mdo start + parameters offset + array length - 1 1778 1779 // Pointer to the parameter area in the MDO. 1780 z_agr(mdp, tmp1); 1781 1782 // Offset of the current profile entry to update. 1783 const Register entry_offset = tmp1; 1784 // entry_offset = array len in number of cells. 1785 z_lg(entry_offset, Address(mdp, ArrayData::array_len_offset())); 1786 // entry_offset (number of cells) = array len - size of 1 entry 1787 add2reg(entry_offset, -TypeStackSlotEntries::per_arg_count()); 1788 // entry_offset in bytes 1789 z_sllg(entry_offset, entry_offset, exact_log2(DataLayout::cell_size)); 1790 1791 Label loop; 1792 bind(loop); 1793 1794 Address arg_off(mdp, entry_offset, ParametersTypeData::stack_slot_offset(0)); 1795 Address arg_type(mdp, entry_offset, ParametersTypeData::type_offset(0)); 1796 1797 // Load offset on the stack from the slot for this parameter. 1798 z_lg(tmp2, arg_off); 1799 z_sllg(tmp2, tmp2, Interpreter::logStackElementSize); 1800 z_lcgr(tmp2); // Negate. 1801 1802 // Profile the parameter. 1803 z_ltg(tmp2, Address(Z_locals, tmp2)); 1804 profile_obj_type(tmp2, arg_type, tmp2, /*ltg did compare to 0*/ true); 1805 1806 // Go to next parameter. 1807 z_aghi(entry_offset, -TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size); 1808 z_brnl(loop); 1809 1810 bind(profile_continue); 1811 } 1812 } 1813 1814 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 1815 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 1816 int increment, 1817 Address mask, 1818 Register scratch, 1819 bool preloaded, 1820 branch_condition cond, 1821 Label *where) { 1822 assert_different_registers(counter_addr.base(), scratch); 1823 if (preloaded) { 1824 add2reg(scratch, increment); 1825 reg2mem_opt(scratch, counter_addr, false); 1826 } else { 1827 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment) && counter_addr.is_RSYform()) { 1828 z_alsi(counter_addr.disp20(), counter_addr.base(), increment); 1829 mem2reg_signed_opt(scratch, counter_addr); 1830 } else { 1831 mem2reg_signed_opt(scratch, counter_addr); 1832 add2reg(scratch, increment); 1833 reg2mem_opt(scratch, counter_addr, false); 1834 } 1835 } 1836 z_n(scratch, mask); 1837 if (where) { z_brc(cond, *where); } 1838 } 1839 1840 // Get MethodCounters object for given method. Lazily allocated if necessary. 1841 // method - Ptr to Method object. 1842 // Rcounters - Ptr to MethodCounters object associated with Method object. 1843 // skip - Exit point if MethodCounters object can't be created (OOM condition). 1844 void InterpreterMacroAssembler::get_method_counters(Register Rmethod, 1845 Register Rcounters, 1846 Label& skip) { 1847 assert_different_registers(Rmethod, Rcounters); 1848 1849 BLOCK_COMMENT("get MethodCounters object {"); 1850 1851 Label has_counters; 1852 load_and_test_long(Rcounters, Address(Rmethod, Method::method_counters_offset())); 1853 z_brnz(has_counters); 1854 1855 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), Rmethod, false); 1856 z_ltgr(Rcounters, Z_RET); // Runtime call returns MethodCounters object. 1857 z_brz(skip); // No MethodCounters, out of memory. 1858 1859 bind(has_counters); 1860 1861 BLOCK_COMMENT("} get MethodCounters object"); 1862 } 1863 1864 // Increment invocation counter in MethodCounters object. 1865 // Return (invocation_counter+backedge_counter) as "result" in RctrSum. 1866 // Counter values are all unsigned. 1867 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register RctrSum) { 1868 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful"); 1869 assert_different_registers(Rcounters, RctrSum); 1870 1871 int increment = InvocationCounter::count_increment; 1872 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()); 1873 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset()); 1874 1875 BLOCK_COMMENT("Increment invocation counter {"); 1876 1877 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) { 1878 // Increment the invocation counter in place, 1879 // then add the incremented value to the backedge counter. 1880 z_l(RctrSum, be_counter_offset, Rcounters); 1881 z_alsi(inv_counter_offset, Rcounters, increment); // Atomic increment @no extra cost! 1882 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits. 1883 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters); 1884 } else { 1885 // This path is optimized for low register consumption 1886 // at the cost of somewhat higher operand delays. 1887 // It does not need an extra temp register. 1888 1889 // Update the invocation counter. 1890 z_l(RctrSum, inv_counter_offset, Rcounters); 1891 if (RctrSum == Z_R0) { 1892 z_ahi(RctrSum, increment); 1893 } else { 1894 add2reg(RctrSum, increment); 1895 } 1896 z_st(RctrSum, inv_counter_offset, Rcounters); 1897 1898 // Mask off the state bits. 1899 z_nilf(RctrSum, InvocationCounter::count_mask_value); 1900 1901 // Add the backedge counter to the updated invocation counter to 1902 // form the result. 1903 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters); 1904 } 1905 1906 BLOCK_COMMENT("} Increment invocation counter"); 1907 1908 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 1909 } 1910 1911 1912 // increment backedge counter in MethodCounters object. 1913 // return (invocation_counter+backedge_counter) as "result" in RctrSum 1914 // counter values are all unsigned! 1915 void InterpreterMacroAssembler::increment_backedge_counter(Register Rcounters, Register RctrSum) { 1916 assert(UseCompiler, "incrementing must be useful"); 1917 assert_different_registers(Rcounters, RctrSum); 1918 1919 int increment = InvocationCounter::count_increment; 1920 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()); 1921 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset()); 1922 1923 BLOCK_COMMENT("Increment backedge counter {"); 1924 1925 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) { 1926 // Increment the invocation counter in place, 1927 // then add the incremented value to the backedge counter. 1928 z_l(RctrSum, inv_counter_offset, Rcounters); 1929 z_alsi(be_counter_offset, Rcounters, increment); // Atomic increment @no extra cost! 1930 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits. 1931 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters); 1932 } else { 1933 // This path is optimized for low register consumption 1934 // at the cost of somewhat higher operand delays. 1935 // It does not need an extra temp register. 1936 1937 // Update the invocation counter. 1938 z_l(RctrSum, be_counter_offset, Rcounters); 1939 if (RctrSum == Z_R0) { 1940 z_ahi(RctrSum, increment); 1941 } else { 1942 add2reg(RctrSum, increment); 1943 } 1944 z_st(RctrSum, be_counter_offset, Rcounters); 1945 1946 // Mask off the state bits. 1947 z_nilf(RctrSum, InvocationCounter::count_mask_value); 1948 1949 // Add the backedge counter to the updated invocation counter to 1950 // form the result. 1951 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters); 1952 } 1953 1954 BLOCK_COMMENT("} Increment backedge counter"); 1955 1956 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! 1957 } 1958 1959 // Add an InterpMonitorElem to stack (see frame_s390.hpp). 1960 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, 1961 Register Rtemp1, 1962 Register Rtemp2, 1963 Register Rtemp3) { 1964 1965 const Register Rcurr_slot = Rtemp1; 1966 const Register Rlimit = Rtemp2; 1967 const jint delta = -frame::interpreter_frame_monitor_size() * wordSize; 1968 1969 assert((delta & LongAlignmentMask) == 0, 1970 "sizeof BasicObjectLock must be even number of doublewords"); 1971 assert(2 * wordSize == -delta, "this works only as long as delta == -2*wordSize"); 1972 assert(Rcurr_slot != Z_R0, "Register must be usable as base register"); 1973 assert_different_registers(Rlimit, Rcurr_slot, Rtemp3); 1974 1975 get_monitors(Rlimit); 1976 1977 // Adjust stack pointer for additional monitor entry. 1978 resize_frame(RegisterOrConstant((intptr_t) delta), Z_fp, false); 1979 1980 if (!stack_is_empty) { 1981 // Must copy stack contents down. 1982 NearLabel next, done; 1983 1984 // Rtemp := addr(Tos), Z_esp is pointing below it! 1985 add2reg(Rcurr_slot, wordSize, Z_esp); 1986 1987 // Nothing to do, if already at monitor area. 1988 compareU64_and_branch(Rcurr_slot, Rlimit, bcondNotLow, done); 1989 1990 bind(next); 1991 1992 // Move one stack slot. 1993 mem2reg_opt(Rtemp3, Address(Rcurr_slot)); 1994 reg2mem_opt(Rtemp3, Address(Rcurr_slot, delta)); 1995 add2reg(Rcurr_slot, wordSize); 1996 compareU64_and_branch(Rcurr_slot, Rlimit, bcondLow, next); // Are we done? 1997 1998 bind(done); 1999 // Done copying stack. 2000 } 2001 2002 // Adjust expression stack and monitor pointers. 2003 add2reg(Z_esp, delta); 2004 add2reg(Rlimit, delta); 2005 save_monitors(Rlimit); 2006 } 2007 2008 // Note: Index holds the offset in bytes afterwards. 2009 // You can use this to store a new value (with Llocals as the base). 2010 void InterpreterMacroAssembler::access_local_int(Register index, Register dst) { 2011 z_sllg(index, index, LogBytesPerWord); 2012 mem2reg_opt(dst, Address(Z_locals, index), false); 2013 } 2014 2015 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2016 if (state == atos) { MacroAssembler::verify_oop(reg); } 2017 } 2018 2019 // Inline assembly for: 2020 // 2021 // if (thread is in interp_only_mode) { 2022 // InterpreterRuntime::post_method_entry(); 2023 // } 2024 2025 void InterpreterMacroAssembler::notify_method_entry() { 2026 2027 // JVMTI 2028 // Whenever JVMTI puts a thread in interp_only_mode, method 2029 // entry/exit events are sent for that thread to track stack 2030 // depth. If it is possible to enter interp_only_mode we add 2031 // the code to check if the event should be sent. 2032 if (JvmtiExport::can_post_interpreter_events()) { 2033 Label jvmti_post_done; 2034 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset())); 2035 z_bre(jvmti_post_done); 2036 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), /*check_exceptions=*/false); 2037 bind(jvmti_post_done); 2038 } 2039 } 2040 2041 // Inline assembly for: 2042 // 2043 // if (thread is in interp_only_mode) { 2044 // if (!native_method) save result 2045 // InterpreterRuntime::post_method_exit(); 2046 // if (!native_method) restore result 2047 // } 2048 // if (DTraceMethodProbes) { 2049 // SharedRuntime::dtrace_method_exit(thread, method); 2050 // } 2051 // 2052 // For native methods their result is stored in z_ijava_state.lresult 2053 // and z_ijava_state.fresult before coming here. 2054 // Java methods have their result stored in the expression stack. 2055 // 2056 // Notice the dependency to frame::interpreter_frame_result(). 2057 void InterpreterMacroAssembler::notify_method_exit(bool native_method, 2058 TosState state, 2059 NotifyMethodExitMode mode) { 2060 // JVMTI 2061 // Whenever JVMTI puts a thread in interp_only_mode, method 2062 // entry/exit events are sent for that thread to track stack 2063 // depth. If it is possible to enter interp_only_mode we add 2064 // the code to check if the event should be sent. 2065 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2066 Label jvmti_post_done; 2067 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset())); 2068 z_bre(jvmti_post_done); 2069 if (!native_method) push(state); // see frame::interpreter_frame_result() 2070 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), /*check_exceptions=*/false); 2071 if (!native_method) pop(state); 2072 bind(jvmti_post_done); 2073 } 2074 2075 #if 0 2076 // Dtrace currently not supported on z/Architecture. 2077 { 2078 SkipIfEqual skip(this, &DTraceMethodProbes, false); 2079 push(state); 2080 get_method(c_rarg1); 2081 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2082 r15_thread, c_rarg1); 2083 pop(state); 2084 } 2085 #endif 2086 } 2087 2088 void InterpreterMacroAssembler::skip_if_jvmti_mode(Label &Lskip, Register Rscratch) { 2089 if (!JvmtiExport::can_post_interpreter_events()) { 2090 return; 2091 } 2092 2093 load_and_test_int(Rscratch, Address(Z_thread, JavaThread::interp_only_mode_offset())); 2094 z_brnz(Lskip); 2095 2096 } 2097 2098 // Pop the topmost TOP_IJAVA_FRAME and set it's sender_sp as new Z_SP. 2099 // The return pc is loaded into the register return_pc. 2100 // 2101 // Registers updated: 2102 // return_pc - The return pc of the calling frame. 2103 // tmp1, tmp2 - scratch 2104 void InterpreterMacroAssembler::pop_interpreter_frame(Register return_pc, Register tmp1, Register tmp2) { 2105 // F0 Z_SP -> caller_sp (F1's) 2106 // ... 2107 // sender_sp (F1's) 2108 // ... 2109 // F1 Z_fp -> caller_sp (F2's) 2110 // return_pc (Continuation after return from F0.) 2111 // ... 2112 // F2 caller_sp 2113 2114 // Remove F0's activation. Restoring Z_SP to sender_sp reverts modifications 2115 // (a) by a c2i adapter and (b) by generate_fixed_frame(). 2116 // In case (a) the new top frame F1 is an unextended compiled frame. 2117 // In case (b) F1 is converted from PARENT_IJAVA_FRAME to TOP_IJAVA_FRAME. 2118 2119 // Case (b) seems to be redundant when returning to a interpreted caller, 2120 // because then the caller's top_frame_sp is installed as sp (see 2121 // TemplateInterpreterGenerator::generate_return_entry_for ()). But 2122 // pop_interpreter_frame() is also used in exception handling and there the 2123 // frame type of the caller is unknown, therefore top_frame_sp cannot be used, 2124 // so it is important that sender_sp is the caller's sp as TOP_IJAVA_FRAME. 2125 2126 Register R_f1_sender_sp = tmp1; 2127 Register R_f2_sp = tmp2; 2128 2129 // Tirst check the for the interpreter frame's magic. 2130 asm_assert_ijava_state_magic(R_f2_sp/*tmp*/); 2131 z_lg(R_f2_sp, _z_parent_ijava_frame_abi(callers_sp), Z_fp); 2132 z_lg(R_f1_sender_sp, _z_ijava_state_neg(sender_sp), Z_fp); 2133 if (return_pc->is_valid()) 2134 z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp); 2135 // Pop F0 by resizing to R_f1_sender_sp and using R_f2_sp as fp. 2136 resize_frame_absolute(R_f1_sender_sp, R_f2_sp, false/*load fp*/); 2137 2138 #ifdef ASSERT 2139 // The return_pc in the new top frame is dead... at least that's my 2140 // current understanding; to assert this I overwrite it. 2141 load_const_optimized(Z_ARG3, 0xb00b1); 2142 z_stg(Z_ARG3, _z_parent_ijava_frame_abi(return_pc), Z_SP); 2143 #endif 2144 } 2145 2146 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2147 if (VerifyFPU) { 2148 unimplemented("verfiyFPU"); 2149 } 2150 } 2151