1 /* 2 * Copyright (c) 2008, 2017, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "asm/assembler.hpp" 27 #include "interpreter/bytecodeHistogram.hpp" 28 #include "interpreter/interp_masm.hpp" 29 #include "interpreter/interpreter.hpp" 30 #include "interpreter/interpreterRuntime.hpp" 31 #include "interpreter/templateInterpreterGenerator.hpp" 32 #include "interpreter/templateTable.hpp" 33 #include "oops/arrayOop.hpp" 34 #include "oops/methodData.hpp" 35 #include "oops/method.hpp" 36 #include "oops/oop.inline.hpp" 37 #include "prims/jvmtiExport.hpp" 38 #include "prims/jvmtiThreadState.hpp" 39 #include "prims/methodHandles.hpp" 40 #include "runtime/arguments.hpp" 41 #include "runtime/deoptimization.hpp" 42 #include "runtime/frame.inline.hpp" 43 #include "runtime/sharedRuntime.hpp" 44 #include "runtime/stubRoutines.hpp" 45 #include "runtime/synchronizer.hpp" 46 #include "runtime/timer.hpp" 47 #include "runtime/vframeArray.hpp" 48 #include "utilities/debug.hpp" 49 #include "utilities/macros.hpp" 50 51 // Size of interpreter code. Increase if too small. Interpreter will 52 // fail with a guarantee ("not enough space for interpreter generation"); 53 // if too small. 54 // Run with +PrintInterpreter to get the VM to print out the size. 55 // Max size with JVMTI 56 int TemplateInterpreter::InterpreterCodeSize = 180 * 1024; 57 58 #define __ _masm-> 59 60 //------------------------------------------------------------------------------------------------------------------------ 61 62 address TemplateInterpreterGenerator::generate_slow_signature_handler() { 63 address entry = __ pc(); 64 65 // callee-save register for saving LR, shared with generate_native_entry 66 const Register Rsaved_ret_addr = AARCH64_ONLY(R21) NOT_AARCH64(Rtmp_save0); 67 68 __ mov(Rsaved_ret_addr, LR); 69 70 __ mov(R1, Rmethod); 71 __ mov(R2, Rlocals); 72 __ mov(R3, SP); 73 74 #ifdef AARCH64 75 // expand expr. stack and extended SP to avoid cutting SP in call_VM 76 __ mov(Rstack_top, SP); 77 __ str(Rstack_top, Address(FP, frame::interpreter_frame_extended_sp_offset * wordSize)); 78 __ check_stack_top(); 79 80 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), R1, R2, R3, false); 81 82 __ ldp(ZR, c_rarg1, Address(SP, 2*wordSize, post_indexed)); 83 __ ldp(c_rarg2, c_rarg3, Address(SP, 2*wordSize, post_indexed)); 84 __ ldp(c_rarg4, c_rarg5, Address(SP, 2*wordSize, post_indexed)); 85 __ ldp(c_rarg6, c_rarg7, Address(SP, 2*wordSize, post_indexed)); 86 87 __ ldp_d(V0, V1, Address(SP, 2*wordSize, post_indexed)); 88 __ ldp_d(V2, V3, Address(SP, 2*wordSize, post_indexed)); 89 __ ldp_d(V4, V5, Address(SP, 2*wordSize, post_indexed)); 90 __ ldp_d(V6, V7, Address(SP, 2*wordSize, post_indexed)); 91 #else 92 93 // Safer to save R9 (when scratched) since callers may have been 94 // written assuming R9 survives. This is suboptimal but 95 // probably not important for this slow case call site. 96 // Note for R9 saving: slow_signature_handler may copy register 97 // arguments above the current SP (passed as R3). It is safe for 98 // call_VM to use push and pop to protect additional values on the 99 // stack if needed. 100 __ call_VM(CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), true /* save R9 if needed*/); 101 __ add(SP, SP, wordSize); // Skip R0 102 __ pop(RegisterSet(R1, R3)); // Load arguments passed in registers 103 #ifdef __ABI_HARD__ 104 // Few alternatives to an always-load-FP-registers approach: 105 // - parse method signature to detect FP arguments 106 // - keep a counter/flag on a stack indicationg number of FP arguments in the method. 107 // The later has been originally implemented and tested but a conditional path could 108 // eliminate any gain imposed by avoiding 8 double word loads. 109 __ fldmiad(SP, FloatRegisterSet(D0, 8), writeback); 110 #endif // __ABI_HARD__ 111 #endif // AARCH64 112 113 __ ret(Rsaved_ret_addr); 114 115 return entry; 116 } 117 118 119 // 120 // Various method entries (that c++ and asm interpreter agree upon) 121 //------------------------------------------------------------------------------------------------------------------------ 122 // 123 // 124 125 // Abstract method entry 126 // Attempt to execute abstract method. Throw exception 127 address TemplateInterpreterGenerator::generate_abstract_entry(void) { 128 address entry_point = __ pc(); 129 130 #ifdef AARCH64 131 __ restore_sp_after_call(Rtemp); 132 __ restore_stack_top(); 133 #endif 134 135 __ empty_expression_stack(); 136 137 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError)); 138 139 DEBUG_ONLY(STOP("generate_abstract_entry");) // Should not reach here 140 return entry_point; 141 } 142 143 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) { 144 if (!InlineIntrinsics) return NULL; // Generate a vanilla entry 145 146 // TODO: ARM 147 return NULL; 148 149 address entry_point = __ pc(); 150 STOP("generate_math_entry"); 151 return entry_point; 152 } 153 154 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 155 address entry = __ pc(); 156 157 // Note: There should be a minimal interpreter frame set up when stack 158 // overflow occurs since we check explicitly for it now. 159 // 160 #ifdef ASSERT 161 { Label L; 162 __ sub(Rtemp, FP, - frame::interpreter_frame_monitor_block_top_offset * wordSize); 163 __ cmp(SP, Rtemp); // Rtemp = maximal SP for current FP, 164 // (stack grows negative) 165 __ b(L, ls); // check if frame is complete 166 __ stop ("interpreter frame not set up"); 167 __ bind(L); 168 } 169 #endif // ASSERT 170 171 // Restore bcp under the assumption that the current frame is still 172 // interpreted 173 __ restore_bcp(); 174 175 // expression stack must be empty before entering the VM if an exception 176 // happened 177 __ empty_expression_stack(); 178 179 // throw exception 180 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError)); 181 182 __ should_not_reach_here(); 183 184 return entry; 185 } 186 187 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) { 188 address entry = __ pc(); 189 190 // index is in R4_ArrayIndexOutOfBounds_index 191 192 InlinedString Lname(name); 193 194 // expression stack must be empty before entering the VM if an exception happened 195 __ empty_expression_stack(); 196 197 // setup parameters 198 __ ldr_literal(R1, Lname); 199 __ mov(R2, R4_ArrayIndexOutOfBounds_index); 200 201 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R1, R2); 202 203 __ nop(); // to avoid filling CPU pipeline with invalid instructions 204 __ nop(); 205 __ should_not_reach_here(); 206 __ bind_literal(Lname); 207 208 return entry; 209 } 210 211 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 212 address entry = __ pc(); 213 214 // object is in R2_ClassCastException_obj 215 216 // expression stack must be empty before entering the VM if an exception 217 // happened 218 __ empty_expression_stack(); 219 220 __ mov(R1, R2_ClassCastException_obj); 221 __ call_VM(noreg, 222 CAST_FROM_FN_PTR(address, 223 InterpreterRuntime::throw_ClassCastException), 224 R1); 225 226 __ should_not_reach_here(); 227 228 return entry; 229 } 230 231 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) { 232 assert(!pass_oop || message == NULL, "either oop or message but not both"); 233 address entry = __ pc(); 234 235 InlinedString Lname(name); 236 InlinedString Lmessage(message); 237 238 if (pass_oop) { 239 // object is at TOS 240 __ pop_ptr(R2); 241 } 242 243 // expression stack must be empty before entering the VM if an exception happened 244 __ empty_expression_stack(); 245 246 // setup parameters 247 __ ldr_literal(R1, Lname); 248 249 if (pass_oop) { 250 __ call_VM(Rexception_obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), R1, R2); 251 } else { 252 if (message != NULL) { 253 __ ldr_literal(R2, Lmessage); 254 } else { 255 __ mov(R2, 0); 256 } 257 __ call_VM(Rexception_obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), R1, R2); 258 } 259 260 // throw exception 261 __ b(Interpreter::throw_exception_entry()); 262 263 __ nop(); // to avoid filling CPU pipeline with invalid instructions 264 __ nop(); 265 __ bind_literal(Lname); 266 if (!pass_oop && (message != NULL)) { 267 __ bind_literal(Lmessage); 268 } 269 270 return entry; 271 } 272 273 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) { 274 address entry = __ pc(); 275 276 __ interp_verify_oop(R0_tos, state, __FILE__, __LINE__); 277 278 #ifdef AARCH64 279 __ restore_sp_after_call(Rtemp); // Restore SP to extended SP 280 __ restore_stack_top(); 281 #else 282 // Restore stack bottom in case i2c adjusted stack 283 __ ldr(SP, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize)); 284 // and NULL it as marker that SP is now tos until next java call 285 __ mov(Rtemp, (int)NULL_WORD); 286 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize)); 287 #endif // AARCH64 288 289 __ restore_method(); 290 __ restore_bcp(); 291 __ restore_dispatch(); 292 __ restore_locals(); 293 294 const Register Rcache = R2_tmp; 295 const Register Rindex = R3_tmp; 296 __ get_cache_and_index_at_bcp(Rcache, Rindex, 1, index_size); 297 298 __ add(Rtemp, Rcache, AsmOperand(Rindex, lsl, LogBytesPerWord)); 299 __ ldrb(Rtemp, Address(Rtemp, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset())); 300 __ check_stack_top(); 301 __ add(Rstack_top, Rstack_top, AsmOperand(Rtemp, lsl, Interpreter::logStackElementSize)); 302 303 #ifndef AARCH64 304 __ convert_retval_to_tos(state); 305 #endif // !AARCH64 306 307 __ check_and_handle_popframe(); 308 __ check_and_handle_earlyret(); 309 310 __ dispatch_next(state, step); 311 312 return entry; 313 } 314 315 316 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) { 317 address entry = __ pc(); 318 319 __ interp_verify_oop(R0_tos, state, __FILE__, __LINE__); 320 321 #ifdef AARCH64 322 __ restore_sp_after_call(Rtemp); // Restore SP to extended SP 323 __ restore_stack_top(); 324 #else 325 // The stack is not extended by deopt but we must NULL last_sp as this 326 // entry is like a "return". 327 __ mov(Rtemp, 0); 328 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize)); 329 #endif // AARCH64 330 331 __ restore_method(); 332 __ restore_bcp(); 333 __ restore_dispatch(); 334 __ restore_locals(); 335 336 // handle exceptions 337 { Label L; 338 __ ldr(Rtemp, Address(Rthread, Thread::pending_exception_offset())); 339 __ cbz(Rtemp, L); 340 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception)); 341 __ should_not_reach_here(); 342 __ bind(L); 343 } 344 345 __ dispatch_next(state, step); 346 347 return entry; 348 } 349 350 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) { 351 #ifdef AARCH64 352 address entry = __ pc(); 353 switch (type) { 354 case T_BOOLEAN: 355 __ tst(R0, 0xff); 356 __ cset(R0, ne); 357 break; 358 case T_CHAR : __ zero_extend(R0, R0, 16); break; 359 case T_BYTE : __ sign_extend(R0, R0, 8); break; 360 case T_SHORT : __ sign_extend(R0, R0, 16); break; 361 case T_INT : // fall through 362 case T_LONG : // fall through 363 case T_VOID : // fall through 364 case T_FLOAT : // fall through 365 case T_DOUBLE : /* nothing to do */ break; 366 case T_OBJECT : 367 // retrieve result from frame 368 __ ldr(R0, Address(FP, frame::interpreter_frame_oop_temp_offset * wordSize)); 369 // and verify it 370 __ verify_oop(R0); 371 break; 372 default : ShouldNotReachHere(); 373 } 374 __ ret(); 375 return entry; 376 #else 377 // Result handlers are not used on 32-bit ARM 378 // since the returned value is already in appropriate format. 379 __ should_not_reach_here(); // to avoid empty code block 380 381 // The result handler non-zero indicates an object is returned and this is 382 // used in the native entry code. 383 return type == T_OBJECT ? (address)(-1) : NULL; 384 #endif // AARCH64 385 } 386 387 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) { 388 address entry = __ pc(); 389 __ push(state); 390 __ call_VM(noreg, runtime_entry); 391 392 // load current bytecode 393 __ ldrb(R3_bytecode, Address(Rbcp)); 394 __ dispatch_only_normal(vtos); 395 return entry; 396 } 397 398 399 // Helpers for commoning out cases in the various type of method entries. 400 // 401 402 // increment invocation count & check for overflow 403 // 404 // Note: checking for negative value instead of overflow 405 // so we have a 'sticky' overflow test 406 // 407 // In: Rmethod. 408 // 409 // Uses R0, R1, Rtemp. 410 // 411 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, 412 Label* profile_method, 413 Label* profile_method_continue) { 414 Label done; 415 const Register Rcounters = Rtemp; 416 const Address invocation_counter(Rcounters, 417 MethodCounters::invocation_counter_offset() + 418 InvocationCounter::counter_offset()); 419 420 // Note: In tiered we increment either counters in MethodCounters* or 421 // in MDO depending if we're profiling or not. 422 if (TieredCompilation) { 423 int increment = InvocationCounter::count_increment; 424 Label no_mdo; 425 if (ProfileInterpreter) { 426 // Are we profiling? 427 __ ldr(R1_tmp, Address(Rmethod, Method::method_data_offset())); 428 __ cbz(R1_tmp, no_mdo); 429 // Increment counter in the MDO 430 const Address mdo_invocation_counter(R1_tmp, 431 in_bytes(MethodData::invocation_counter_offset()) + 432 in_bytes(InvocationCounter::counter_offset())); 433 const Address mask(R1_tmp, in_bytes(MethodData::invoke_mask_offset())); 434 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, R0_tmp, Rtemp, eq, overflow); 435 __ b(done); 436 } 437 __ bind(no_mdo); 438 __ get_method_counters(Rmethod, Rcounters, done); 439 const Address mask(Rcounters, in_bytes(MethodCounters::invoke_mask_offset())); 440 __ increment_mask_and_jump(invocation_counter, increment, mask, R0_tmp, R1_tmp, eq, overflow); 441 __ bind(done); 442 } else { // not TieredCompilation 443 const Address backedge_counter(Rcounters, 444 MethodCounters::backedge_counter_offset() + 445 InvocationCounter::counter_offset()); 446 447 const Register Ricnt = R0_tmp; // invocation counter 448 const Register Rbcnt = R1_tmp; // backedge counter 449 450 __ get_method_counters(Rmethod, Rcounters, done); 451 452 if (ProfileInterpreter) { 453 const Register Riic = R1_tmp; 454 __ ldr_s32(Riic, Address(Rcounters, MethodCounters::interpreter_invocation_counter_offset())); 455 __ add(Riic, Riic, 1); 456 __ str_32(Riic, Address(Rcounters, MethodCounters::interpreter_invocation_counter_offset())); 457 } 458 459 // Update standard invocation counters 460 461 __ ldr_u32(Ricnt, invocation_counter); 462 __ ldr_u32(Rbcnt, backedge_counter); 463 464 __ add(Ricnt, Ricnt, InvocationCounter::count_increment); 465 466 #ifdef AARCH64 467 __ andr(Rbcnt, Rbcnt, (unsigned int)InvocationCounter::count_mask_value); // mask out the status bits 468 #else 469 __ bic(Rbcnt, Rbcnt, ~InvocationCounter::count_mask_value); // mask out the status bits 470 #endif // AARCH64 471 472 __ str_32(Ricnt, invocation_counter); // save invocation count 473 __ add(Ricnt, Ricnt, Rbcnt); // add both counters 474 475 // profile_method is non-null only for interpreted method so 476 // profile_method != NULL == !native_call 477 // BytecodeInterpreter only calls for native so code is elided. 478 479 if (ProfileInterpreter && profile_method != NULL) { 480 assert(profile_method_continue != NULL, "should be non-null"); 481 482 // Test to see if we should create a method data oop 483 // Reuse R1_tmp as we don't need backedge counters anymore. 484 Address profile_limit(Rcounters, in_bytes(MethodCounters::interpreter_profile_limit_offset())); 485 __ ldr_s32(R1_tmp, profile_limit); 486 __ cmp_32(Ricnt, R1_tmp); 487 __ b(*profile_method_continue, lt); 488 489 // if no method data exists, go to profile_method 490 __ test_method_data_pointer(R1_tmp, *profile_method); 491 } 492 493 Address invoke_limit(Rcounters, in_bytes(MethodCounters::interpreter_invocation_limit_offset())); 494 __ ldr_s32(R1_tmp, invoke_limit); 495 __ cmp_32(Ricnt, R1_tmp); 496 __ b(*overflow, hs); 497 __ bind(done); 498 } 499 } 500 501 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) { 502 // InterpreterRuntime::frequency_counter_overflow takes one argument 503 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp). 504 // The call returns the address of the verified entry point for the method or NULL 505 // if the compilation did not complete (either went background or bailed out). 506 __ mov(R1, (int)false); 507 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R1); 508 509 // jump to the interpreted entry. 510 __ b(do_continue); 511 } 512 513 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) { 514 // Check if we've got enough room on the stack for 515 // - overhead; 516 // - locals; 517 // - expression stack. 518 // 519 // Registers on entry: 520 // 521 // R3 = number of additional locals 522 // R11 = max expression stack slots (AArch64 only) 523 // Rthread 524 // Rmethod 525 // Registers used: R0, R1, R2, Rtemp. 526 527 const Register Radditional_locals = R3; 528 const Register RmaxStack = AARCH64_ONLY(R11) NOT_AARCH64(R2); 529 530 // monitor entry size 531 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 532 533 // total overhead size: entry_size + (saved registers, thru expr stack bottom). 534 // be sure to change this if you add/subtract anything to/from the overhead area 535 const int overhead_size = (frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset)*wordSize + entry_size; 536 537 // Pages reserved for VM runtime calls and subsequent Java calls. 538 const int reserved_pages = JavaThread::stack_shadow_zone_size(); 539 540 // Thread::stack_size() includes guard pages, and they should not be touched. 541 const int guard_pages = JavaThread::stack_guard_zone_size(); 542 543 __ ldr(R0, Address(Rthread, Thread::stack_base_offset())); 544 __ ldr(R1, Address(Rthread, Thread::stack_size_offset())); 545 #ifndef AARCH64 546 __ ldr(Rtemp, Address(Rmethod, Method::const_offset())); 547 __ ldrh(RmaxStack, Address(Rtemp, ConstMethod::max_stack_offset())); 548 #endif // !AARCH64 549 __ sub_slow(Rtemp, SP, overhead_size + reserved_pages + guard_pages + Method::extra_stack_words()); 550 551 // reserve space for additional locals 552 __ sub(Rtemp, Rtemp, AsmOperand(Radditional_locals, lsl, Interpreter::logStackElementSize)); 553 554 // stack size 555 __ sub(R0, R0, R1); 556 557 // reserve space for expression stack 558 __ sub(Rtemp, Rtemp, AsmOperand(RmaxStack, lsl, Interpreter::logStackElementSize)); 559 560 __ cmp(Rtemp, R0); 561 562 #ifdef AARCH64 563 Label L; 564 __ b(L, hi); 565 __ mov(SP, Rsender_sp); // restore SP 566 __ b(StubRoutines::throw_StackOverflowError_entry()); 567 __ bind(L); 568 #else 569 __ mov(SP, Rsender_sp, ls); // restore SP 570 __ b(StubRoutines::throw_StackOverflowError_entry(), ls); 571 #endif // AARCH64 572 } 573 574 575 // Allocate monitor and lock method (asm interpreter) 576 // 577 void TemplateInterpreterGenerator::lock_method() { 578 // synchronize method 579 580 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 581 assert ((entry_size % StackAlignmentInBytes) == 0, "should keep stack alignment"); 582 583 #ifdef ASSERT 584 { Label L; 585 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset())); 586 __ tbnz(Rtemp, JVM_ACC_SYNCHRONIZED_BIT, L); 587 __ stop("method doesn't need synchronization"); 588 __ bind(L); 589 } 590 #endif // ASSERT 591 592 // get synchronization object 593 { Label done; 594 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset())); 595 #ifdef AARCH64 596 __ ldr(R0, Address(Rlocals, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case) 597 __ tbz(Rtemp, JVM_ACC_STATIC_BIT, done); 598 #else 599 __ tst(Rtemp, JVM_ACC_STATIC); 600 __ ldr(R0, Address(Rlocals, Interpreter::local_offset_in_bytes(0)), eq); // get receiver (assume this is frequent case) 601 __ b(done, eq); 602 #endif // AARCH64 603 __ load_mirror(R0, Rmethod, Rtemp); 604 __ bind(done); 605 } 606 607 // add space for monitor & lock 608 609 #ifdef AARCH64 610 __ check_extended_sp(Rtemp); 611 __ sub(SP, SP, entry_size); // adjust extended SP 612 __ mov(Rtemp, SP); 613 __ str(Rtemp, Address(FP, frame::interpreter_frame_extended_sp_offset * wordSize)); 614 #endif // AARCH64 615 616 __ sub(Rstack_top, Rstack_top, entry_size); 617 __ check_stack_top_on_expansion(); 618 // add space for a monitor entry 619 __ str(Rstack_top, Address(FP, frame::interpreter_frame_monitor_block_top_offset * wordSize)); 620 // set new monitor block top 621 __ str(R0, Address(Rstack_top, BasicObjectLock::obj_offset_in_bytes())); 622 // store object 623 __ mov(R1, Rstack_top); // monitor entry address 624 __ lock_object(R1); 625 } 626 627 #ifdef AARCH64 628 629 // 630 // Generate a fixed interpreter frame. This is identical setup for interpreted methods 631 // and for native methods hence the shared code. 632 // 633 // On entry: 634 // R10 = ConstMethod 635 // R11 = max expr. stack (in slots), if !native_call 636 // 637 // On exit: 638 // Rbcp, Rstack_top are initialized, SP is extended 639 // 640 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { 641 // Incoming registers 642 const Register RconstMethod = R10; 643 const Register RmaxStack = R11; 644 // Temporary registers 645 const Register RextendedSP = R0; 646 const Register Rcache = R1; 647 const Register Rmdp = ProfileInterpreter ? R2 : ZR; 648 649 // Generates the following stack layout (stack grows up in this picture): 650 // 651 // [ expr. stack bottom ] 652 // [ saved Rbcp ] 653 // [ current Rlocals ] 654 // [ cache ] 655 // [ mdx ] 656 // [ mirror ] 657 // [ Method* ] 658 // [ extended SP ] 659 // [ expr. stack top ] 660 // [ sender_sp ] 661 // [ saved FP ] <--- FP 662 // [ saved LR ] 663 664 // initialize fixed part of activation frame 665 __ stp(FP, LR, Address(SP, -2*wordSize, pre_indexed)); 666 __ mov(FP, SP); // establish new FP 667 668 // setup Rbcp 669 if (native_call) { 670 __ mov(Rbcp, ZR); // bcp = 0 for native calls 671 } else { 672 __ add(Rbcp, RconstMethod, in_bytes(ConstMethod::codes_offset())); // get codebase 673 } 674 675 // Rstack_top & RextendedSP 676 __ sub(Rstack_top, SP, 10*wordSize); 677 if (native_call) { 678 __ sub(RextendedSP, Rstack_top, round_to(wordSize, StackAlignmentInBytes)); // reserve 1 slot for exception handling 679 } else { 680 __ sub(RextendedSP, Rstack_top, AsmOperand(RmaxStack, lsl, Interpreter::logStackElementSize)); 681 __ align_reg(RextendedSP, RextendedSP, StackAlignmentInBytes); 682 } 683 __ mov(SP, RextendedSP); 684 __ check_stack_top(); 685 686 // Load Rmdp 687 if (ProfileInterpreter) { 688 __ ldr(Rtemp, Address(Rmethod, Method::method_data_offset())); 689 __ tst(Rtemp, Rtemp); 690 __ add(Rtemp, Rtemp, in_bytes(MethodData::data_offset())); 691 __ csel(Rmdp, ZR, Rtemp, eq); 692 } 693 694 // Load Rcache 695 __ ldr(Rtemp, Address(RconstMethod, ConstMethod::constants_offset())); 696 __ ldr(Rcache, Address(Rtemp, ConstantPool::cache_offset_in_bytes())); 697 // Get mirror and store it in the frame as GC root for this Method* 698 __ load_mirror(Rtemp, Rmethod, Rtemp); 699 700 // Build fixed frame 701 __ stp(Rstack_top, Rbcp, Address(FP, -10*wordSize)); 702 __ stp(Rlocals, Rcache, Address(FP, -8*wordSize)); 703 __ stp(Rmdp, Rtemp, Address(FP, -6*wordSize)); 704 __ stp(Rmethod, RextendedSP, Address(FP, -4*wordSize)); 705 __ stp(ZR, Rsender_sp, Address(FP, -2*wordSize)); 706 assert(frame::interpreter_frame_initial_sp_offset == -10, "interpreter frame broken"); 707 assert(frame::interpreter_frame_stack_top_offset == -2, "stack top broken"); 708 } 709 710 #else // AARCH64 711 712 // 713 // Generate a fixed interpreter frame. This is identical setup for interpreted methods 714 // and for native methods hence the shared code. 715 716 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { 717 // Generates the following stack layout: 718 // 719 // [ expr. stack bottom ] 720 // [ saved Rbcp ] 721 // [ current Rlocals ] 722 // [ cache ] 723 // [ mdx ] 724 // [ Method* ] 725 // [ last_sp ] 726 // [ sender_sp ] 727 // [ saved FP ] <--- FP 728 // [ saved LR ] 729 730 // initialize fixed part of activation frame 731 __ push(LR); // save return address 732 __ push(FP); // save FP 733 __ mov(FP, SP); // establish new FP 734 735 __ push(Rsender_sp); 736 737 __ mov(R0, 0); 738 __ push(R0); // leave last_sp as null 739 740 // setup Rbcp 741 if (native_call) { 742 __ mov(Rbcp, 0); // bcp = 0 for native calls 743 } else { 744 __ ldr(Rtemp, Address(Rmethod, Method::const_offset())); // get ConstMethod* 745 __ add(Rbcp, Rtemp, ConstMethod::codes_offset()); // get codebase 746 } 747 748 __ push(Rmethod); // save Method* 749 // Get mirror and store it in the frame as GC root for this Method* 750 __ load_mirror(Rtemp, Rmethod, Rtemp); 751 __ push(Rtemp); 752 753 if (ProfileInterpreter) { 754 __ ldr(Rtemp, Address(Rmethod, Method::method_data_offset())); 755 __ tst(Rtemp, Rtemp); 756 __ add(Rtemp, Rtemp, in_bytes(MethodData::data_offset()), ne); 757 __ push(Rtemp); // set the mdp (method data pointer) 758 } else { 759 __ push(R0); 760 } 761 762 __ ldr(Rtemp, Address(Rmethod, Method::const_offset())); 763 __ ldr(Rtemp, Address(Rtemp, ConstMethod::constants_offset())); 764 __ ldr(Rtemp, Address(Rtemp, ConstantPool::cache_offset_in_bytes())); 765 __ push(Rtemp); // set constant pool cache 766 __ push(Rlocals); // set locals pointer 767 __ push(Rbcp); // set bcp 768 __ push(R0); // reserve word for pointer to expression stack bottom 769 __ str(SP, Address(SP, 0)); // set expression stack bottom 770 } 771 772 #endif // AARCH64 773 774 // End of helpers 775 776 //------------------------------------------------------------------------------------------------------------------------ 777 // Entry points 778 // 779 // Here we generate the various kind of entries into the interpreter. 780 // The two main entry type are generic bytecode methods and native call method. 781 // These both come in synchronized and non-synchronized versions but the 782 // frame layout they create is very similar. The other method entry 783 // types are really just special purpose entries that are really entry 784 // and interpretation all in one. These are for trivial methods like 785 // accessor, empty, or special math methods. 786 // 787 // When control flow reaches any of the entry types for the interpreter 788 // the following holds -> 789 // 790 // Arguments: 791 // 792 // Rmethod: Method* 793 // Rthread: thread 794 // Rsender_sp: sender sp 795 // Rparams (SP on 32-bit ARM): pointer to method parameters 796 // 797 // LR: return address 798 // 799 // Stack layout immediately at entry 800 // 801 // [ optional padding(*)] <--- SP (AArch64) 802 // [ parameter n ] <--- Rparams (SP on 32-bit ARM) 803 // ... 804 // [ parameter 1 ] 805 // [ expression stack ] (caller's java expression stack) 806 807 // Assuming that we don't go to one of the trivial specialized 808 // entries the stack will look like below when we are ready to execute 809 // the first bytecode (or call the native routine). The register usage 810 // will be as the template based interpreter expects. 811 // 812 // local variables follow incoming parameters immediately; i.e. 813 // the return address is saved at the end of the locals. 814 // 815 // [ reserved stack (*) ] <--- SP (AArch64) 816 // [ expr. stack ] <--- Rstack_top (SP on 32-bit ARM) 817 // [ monitor entry ] 818 // ... 819 // [ monitor entry ] 820 // [ expr. stack bottom ] 821 // [ saved Rbcp ] 822 // [ current Rlocals ] 823 // [ cache ] 824 // [ mdx ] 825 // [ mirror ] 826 // [ Method* ] 827 // 828 // 32-bit ARM: 829 // [ last_sp ] 830 // 831 // AArch64: 832 // [ extended SP (*) ] 833 // [ stack top (*) ] 834 // 835 // [ sender_sp ] 836 // [ saved FP ] <--- FP 837 // [ saved LR ] 838 // [ optional padding(*)] 839 // [ local variable m ] 840 // ... 841 // [ local variable 1 ] 842 // [ parameter n ] 843 // ... 844 // [ parameter 1 ] <--- Rlocals 845 // 846 // (*) - AArch64 only 847 // 848 849 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) { 850 #if INCLUDE_ALL_GCS 851 if (UseG1GC) { 852 // Code: _aload_0, _getfield, _areturn 853 // parameter size = 1 854 // 855 // The code that gets generated by this routine is split into 2 parts: 856 // 1. The "intrinsified" code for G1 (or any SATB based GC), 857 // 2. The slow path - which is an expansion of the regular method entry. 858 // 859 // Notes:- 860 // * In the G1 code we do not check whether we need to block for 861 // a safepoint. If G1 is enabled then we must execute the specialized 862 // code for Reference.get (except when the Reference object is null) 863 // so that we can log the value in the referent field with an SATB 864 // update buffer. 865 // If the code for the getfield template is modified so that the 866 // G1 pre-barrier code is executed when the current method is 867 // Reference.get() then going through the normal method entry 868 // will be fine. 869 // * The G1 code can, however, check the receiver object (the instance 870 // of java.lang.Reference) and jump to the slow path if null. If the 871 // Reference object is null then we obviously cannot fetch the referent 872 // and so we don't need to call the G1 pre-barrier. Thus we can use the 873 // regular method entry code to generate the NPE. 874 // 875 // This code is based on generate_accessor_enty. 876 // 877 // Rmethod: Method* 878 // Rthread: thread 879 // Rsender_sp: sender sp, must be preserved for slow path, set SP to it on fast path 880 // Rparams: parameters 881 882 address entry = __ pc(); 883 Label slow_path; 884 const Register Rthis = R0; 885 const Register Rret_addr = Rtmp_save1; 886 assert_different_registers(Rthis, Rret_addr, Rsender_sp); 887 888 const int referent_offset = java_lang_ref_Reference::referent_offset; 889 guarantee(referent_offset > 0, "referent offset not initialized"); 890 891 // Check if local 0 != NULL 892 // If the receiver is null then it is OK to jump to the slow path. 893 __ ldr(Rthis, Address(Rparams)); 894 __ cbz(Rthis, slow_path); 895 896 // Generate the G1 pre-barrier code to log the value of 897 // the referent field in an SATB buffer. 898 899 // Load the value of the referent field. 900 __ load_heap_oop(R0, Address(Rthis, referent_offset)); 901 902 // Preserve LR 903 __ mov(Rret_addr, LR); 904 905 __ g1_write_barrier_pre(noreg, // store_addr 906 noreg, // new_val 907 R0, // pre_val 908 Rtemp, // tmp1 909 R1_tmp); // tmp2 910 911 // _areturn 912 __ mov(SP, Rsender_sp); 913 __ ret(Rret_addr); 914 915 // generate a vanilla interpreter entry as the slow path 916 __ bind(slow_path); 917 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals)); 918 return entry; 919 } 920 #endif // INCLUDE_ALL_GCS 921 922 // If G1 is not enabled then attempt to go through the normal entry point 923 return NULL; 924 } 925 926 // Not supported 927 address TemplateInterpreterGenerator::generate_CRC32_update_entry() { return NULL; } 928 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { return NULL; } 929 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) { return NULL; } 930 931 // 932 // Interpreter stub for calling a native method. (asm interpreter) 933 // This sets up a somewhat different looking stack for calling the native method 934 // than the typical interpreter frame setup. 935 // 936 937 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) { 938 // determine code generation flags 939 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 940 941 // Incoming registers: 942 // 943 // Rmethod: Method* 944 // Rthread: thread 945 // Rsender_sp: sender sp 946 // Rparams: parameters 947 948 address entry_point = __ pc(); 949 950 // Register allocation 951 const Register Rsize_of_params = AARCH64_ONLY(R20) NOT_AARCH64(R6); 952 const Register Rsig_handler = AARCH64_ONLY(R21) NOT_AARCH64(Rtmp_save0 /* R4 */); 953 const Register Rnative_code = AARCH64_ONLY(R22) NOT_AARCH64(Rtmp_save1 /* R5 */); 954 const Register Rresult_handler = AARCH64_ONLY(Rsig_handler) NOT_AARCH64(R6); 955 956 #ifdef AARCH64 957 const Register RconstMethod = R10; // also used in generate_fixed_frame (should match) 958 const Register Rsaved_result = Rnative_code; 959 const FloatRegister Dsaved_result = V8; 960 #else 961 const Register Rsaved_result_lo = Rtmp_save0; // R4 962 const Register Rsaved_result_hi = Rtmp_save1; // R5 963 FloatRegister saved_result_fp; 964 #endif // AARCH64 965 966 967 #ifdef AARCH64 968 __ ldr(RconstMethod, Address(Rmethod, Method::const_offset())); 969 __ ldrh(Rsize_of_params, Address(RconstMethod, ConstMethod::size_of_parameters_offset())); 970 #else 971 __ ldr(Rsize_of_params, Address(Rmethod, Method::const_offset())); 972 __ ldrh(Rsize_of_params, Address(Rsize_of_params, ConstMethod::size_of_parameters_offset())); 973 #endif // AARCH64 974 975 // native calls don't need the stack size check since they have no expression stack 976 // and the arguments are already on the stack and we only add a handful of words 977 // to the stack 978 979 // compute beginning of parameters (Rlocals) 980 __ sub(Rlocals, Rparams, wordSize); 981 __ add(Rlocals, Rlocals, AsmOperand(Rsize_of_params, lsl, Interpreter::logStackElementSize)); 982 983 #ifdef AARCH64 984 int extra_stack_reserve = 2*wordSize; // extra space for oop_temp 985 if(__ can_post_interpreter_events()) { 986 // extra space for saved results 987 extra_stack_reserve += 2*wordSize; 988 } 989 // reserve extra stack space and nullify oop_temp slot 990 __ stp(ZR, ZR, Address(SP, -extra_stack_reserve, pre_indexed)); 991 #else 992 // reserve stack space for oop_temp 993 __ mov(R0, 0); 994 __ push(R0); 995 #endif // AARCH64 996 997 generate_fixed_frame(true); // Note: R9 is now saved in the frame 998 999 // make sure method is native & not abstract 1000 #ifdef ASSERT 1001 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset())); 1002 { 1003 Label L; 1004 __ tbnz(Rtemp, JVM_ACC_NATIVE_BIT, L); 1005 __ stop("tried to execute non-native method as native"); 1006 __ bind(L); 1007 } 1008 { Label L; 1009 __ tbz(Rtemp, JVM_ACC_ABSTRACT_BIT, L); 1010 __ stop("tried to execute abstract method in interpreter"); 1011 __ bind(L); 1012 } 1013 #endif 1014 1015 // increment invocation count & check for overflow 1016 Label invocation_counter_overflow; 1017 if (inc_counter) { 1018 if (synchronized) { 1019 // Avoid unlocking method's monitor in case of exception, as it has not 1020 // been locked yet. 1021 __ set_do_not_unlock_if_synchronized(true, Rtemp); 1022 } 1023 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); 1024 } 1025 1026 Label continue_after_compile; 1027 __ bind(continue_after_compile); 1028 1029 if (inc_counter && synchronized) { 1030 __ set_do_not_unlock_if_synchronized(false, Rtemp); 1031 } 1032 1033 // check for synchronized methods 1034 // Must happen AFTER invocation_counter check and stack overflow check, 1035 // so method is not locked if overflows. 1036 // 1037 if (synchronized) { 1038 lock_method(); 1039 } else { 1040 // no synchronization necessary 1041 #ifdef ASSERT 1042 { Label L; 1043 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset())); 1044 __ tbz(Rtemp, JVM_ACC_SYNCHRONIZED_BIT, L); 1045 __ stop("method needs synchronization"); 1046 __ bind(L); 1047 } 1048 #endif 1049 } 1050 1051 // start execution 1052 #ifdef ASSERT 1053 { Label L; 1054 __ ldr(Rtemp, Address(FP, frame::interpreter_frame_monitor_block_top_offset * wordSize)); 1055 __ cmp(Rtemp, Rstack_top); 1056 __ b(L, eq); 1057 __ stop("broken stack frame setup in interpreter"); 1058 __ bind(L); 1059 } 1060 #endif 1061 __ check_extended_sp(Rtemp); 1062 1063 // jvmti/dtrace support 1064 __ notify_method_entry(); 1065 #if R9_IS_SCRATCHED 1066 __ restore_method(); 1067 #endif 1068 1069 { 1070 Label L; 1071 __ ldr(Rsig_handler, Address(Rmethod, Method::signature_handler_offset())); 1072 __ cbnz(Rsig_handler, L); 1073 __ mov(R1, Rmethod); 1074 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R1, true); 1075 __ ldr(Rsig_handler, Address(Rmethod, Method::signature_handler_offset())); 1076 __ bind(L); 1077 } 1078 1079 { 1080 Label L; 1081 __ ldr(Rnative_code, Address(Rmethod, Method::native_function_offset())); 1082 __ cbnz(Rnative_code, L); 1083 __ mov(R1, Rmethod); 1084 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R1); 1085 __ ldr(Rnative_code, Address(Rmethod, Method::native_function_offset())); 1086 __ bind(L); 1087 } 1088 1089 // Allocate stack space for arguments 1090 1091 #ifdef AARCH64 1092 __ sub(Rtemp, SP, Rsize_of_params, ex_uxtw, LogBytesPerWord); 1093 __ align_reg(SP, Rtemp, StackAlignmentInBytes); 1094 1095 // Allocate more stack space to accomodate all arguments passed on GP and FP registers: 1096 // 8 * wordSize for GPRs 1097 // 8 * wordSize for FPRs 1098 int reg_arguments = round_to(8*wordSize + 8*wordSize, StackAlignmentInBytes); 1099 #else 1100 1101 // C functions need aligned stack 1102 __ bic(SP, SP, StackAlignmentInBytes - 1); 1103 // Multiply by BytesPerLong instead of BytesPerWord, because calling convention 1104 // may require empty slots due to long alignment, e.g. func(int, jlong, int, jlong) 1105 __ sub(SP, SP, AsmOperand(Rsize_of_params, lsl, LogBytesPerLong)); 1106 1107 #ifdef __ABI_HARD__ 1108 // Allocate more stack space to accomodate all GP as well as FP registers: 1109 // 4 * wordSize 1110 // 8 * BytesPerLong 1111 int reg_arguments = round_to((4*wordSize) + (8*BytesPerLong), StackAlignmentInBytes); 1112 #else 1113 // Reserve at least 4 words on the stack for loading 1114 // of parameters passed on registers (R0-R3). 1115 // See generate_slow_signature_handler(). 1116 // It is also used for JNIEnv & class additional parameters. 1117 int reg_arguments = 4 * wordSize; 1118 #endif // __ABI_HARD__ 1119 #endif // AARCH64 1120 1121 __ sub(SP, SP, reg_arguments); 1122 1123 1124 // Note: signature handler blows R4 (32-bit ARM) or R21 (AArch64) besides all scratch registers. 1125 // See AbstractInterpreterGenerator::generate_slow_signature_handler(). 1126 __ call(Rsig_handler); 1127 #if R9_IS_SCRATCHED 1128 __ restore_method(); 1129 #endif 1130 __ mov(Rresult_handler, R0); 1131 1132 // Pass JNIEnv and mirror for static methods 1133 { 1134 Label L; 1135 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset())); 1136 __ add(R0, Rthread, in_bytes(JavaThread::jni_environment_offset())); 1137 __ tbz(Rtemp, JVM_ACC_STATIC_BIT, L); 1138 __ load_mirror(Rtemp, Rmethod, Rtemp); 1139 __ add(R1, FP, frame::interpreter_frame_oop_temp_offset * wordSize); 1140 __ str(Rtemp, Address(R1, 0)); 1141 __ bind(L); 1142 } 1143 1144 __ set_last_Java_frame(SP, FP, true, Rtemp); 1145 1146 // Changing state to _thread_in_native must be the last thing to do 1147 // before the jump to native code. At this moment stack must be 1148 // safepoint-safe and completely prepared for stack walking. 1149 #ifdef ASSERT 1150 { 1151 Label L; 1152 __ ldr_u32(Rtemp, Address(Rthread, JavaThread::thread_state_offset())); 1153 __ cmp_32(Rtemp, _thread_in_Java); 1154 __ b(L, eq); 1155 __ stop("invalid thread state"); 1156 __ bind(L); 1157 } 1158 #endif 1159 1160 #ifdef AARCH64 1161 __ mov(Rtemp, _thread_in_native); 1162 __ add(Rtemp2, Rthread, in_bytes(JavaThread::thread_state_offset())); 1163 // STLR is used to force all preceding writes to be observed prior to thread state change 1164 __ stlr_w(Rtemp, Rtemp2); 1165 #else 1166 // Force all preceding writes to be observed prior to thread state change 1167 __ membar(MacroAssembler::StoreStore, Rtemp); 1168 1169 __ mov(Rtemp, _thread_in_native); 1170 __ str(Rtemp, Address(Rthread, JavaThread::thread_state_offset())); 1171 #endif // AARCH64 1172 1173 __ call(Rnative_code); 1174 #if R9_IS_SCRATCHED 1175 __ restore_method(); 1176 #endif 1177 1178 // Set FPSCR/FPCR to a known state 1179 if (AlwaysRestoreFPU) { 1180 __ restore_default_fp_mode(); 1181 } 1182 1183 // Do safepoint check 1184 __ mov(Rtemp, _thread_in_native_trans); 1185 __ str_32(Rtemp, Address(Rthread, JavaThread::thread_state_offset())); 1186 1187 // Force this write out before the read below 1188 __ membar(MacroAssembler::StoreLoad, Rtemp); 1189 1190 __ ldr_global_s32(Rtemp, SafepointSynchronize::address_of_state()); 1191 1192 // Protect the return value in the interleaved code: save it to callee-save registers. 1193 #ifdef AARCH64 1194 __ mov(Rsaved_result, R0); 1195 __ fmov_d(Dsaved_result, D0); 1196 #else 1197 __ mov(Rsaved_result_lo, R0); 1198 __ mov(Rsaved_result_hi, R1); 1199 #ifdef __ABI_HARD__ 1200 // preserve native FP result in a callee-saved register 1201 saved_result_fp = D8; 1202 __ fcpyd(saved_result_fp, D0); 1203 #else 1204 saved_result_fp = fnoreg; 1205 #endif // __ABI_HARD__ 1206 #endif // AARCH64 1207 1208 { 1209 __ ldr_u32(R3, Address(Rthread, JavaThread::suspend_flags_offset())); 1210 __ cmp(Rtemp, SafepointSynchronize::_not_synchronized); 1211 __ cond_cmp(R3, 0, eq); 1212 1213 #ifdef AARCH64 1214 Label L; 1215 __ b(L, eq); 1216 __ mov(R0, Rthread); 1217 __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), relocInfo::none); 1218 __ bind(L); 1219 #else 1220 __ mov(R0, Rthread, ne); 1221 __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), relocInfo::none, ne); 1222 #if R9_IS_SCRATCHED 1223 __ restore_method(); 1224 #endif 1225 #endif // AARCH64 1226 } 1227 1228 // Perform Native->Java thread transition 1229 __ mov(Rtemp, _thread_in_Java); 1230 __ str_32(Rtemp, Address(Rthread, JavaThread::thread_state_offset())); 1231 1232 // Zero handles and last_java_sp 1233 __ reset_last_Java_frame(Rtemp); 1234 __ ldr(R3, Address(Rthread, JavaThread::active_handles_offset())); 1235 __ str_32(__ zero_register(Rtemp), Address(R3, JNIHandleBlock::top_offset_in_bytes())); 1236 if (CheckJNICalls) { 1237 __ str(__ zero_register(Rtemp), Address(Rthread, JavaThread::pending_jni_exception_check_fn_offset())); 1238 } 1239 1240 // Unbox if the result is non-zero object 1241 #ifdef AARCH64 1242 { 1243 Label L, Lnull; 1244 __ mov_slow(Rtemp, AbstractInterpreter::result_handler(T_OBJECT)); 1245 __ cmp(Rresult_handler, Rtemp); 1246 __ b(L, ne); 1247 __ cbz(Rsaved_result, Lnull); 1248 __ ldr(Rsaved_result, Address(Rsaved_result)); 1249 __ bind(Lnull); 1250 // Store oop on the stack for GC 1251 __ str(Rsaved_result, Address(FP, frame::interpreter_frame_oop_temp_offset * wordSize)); 1252 __ bind(L); 1253 } 1254 #else 1255 __ tst(Rsaved_result_lo, Rresult_handler); 1256 __ ldr(Rsaved_result_lo, Address(Rsaved_result_lo), ne); 1257 1258 // Store oop on the stack for GC 1259 __ cmp(Rresult_handler, 0); 1260 __ str(Rsaved_result_lo, Address(FP, frame::interpreter_frame_oop_temp_offset * wordSize), ne); 1261 #endif // AARCH64 1262 1263 #ifdef AARCH64 1264 // Restore SP (drop native parameters area), to keep SP in sync with extended_sp in frame 1265 __ restore_sp_after_call(Rtemp); 1266 __ check_stack_top(); 1267 #endif // AARCH64 1268 1269 // reguard stack if StackOverflow exception happened while in native. 1270 { 1271 __ ldr_u32(Rtemp, Address(Rthread, JavaThread::stack_guard_state_offset())); 1272 __ cmp_32(Rtemp, JavaThread::stack_guard_yellow_reserved_disabled); 1273 #ifdef AARCH64 1274 Label L; 1275 __ b(L, ne); 1276 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), relocInfo::none); 1277 __ bind(L); 1278 #else 1279 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), relocInfo::none, eq); 1280 #if R9_IS_SCRATCHED 1281 __ restore_method(); 1282 #endif 1283 #endif // AARCH64 1284 } 1285 1286 // check pending exceptions 1287 { 1288 __ ldr(Rtemp, Address(Rthread, Thread::pending_exception_offset())); 1289 #ifdef AARCH64 1290 Label L; 1291 __ cbz(Rtemp, L); 1292 __ mov_pc_to(Rexception_pc); 1293 __ b(StubRoutines::forward_exception_entry()); 1294 __ bind(L); 1295 #else 1296 __ cmp(Rtemp, 0); 1297 __ mov(Rexception_pc, PC, ne); 1298 __ b(StubRoutines::forward_exception_entry(), ne); 1299 #endif // AARCH64 1300 } 1301 1302 if (synchronized) { 1303 // address of first monitor 1304 __ sub(R1, FP, - (frame::interpreter_frame_monitor_block_bottom_offset - frame::interpreter_frame_monitor_size()) * wordSize); 1305 __ unlock_object(R1); 1306 } 1307 1308 // jvmti/dtrace support 1309 // Note: This must happen _after_ handling/throwing any exceptions since 1310 // the exception handler code notifies the runtime of method exits 1311 // too. If this happens before, method entry/exit notifications are 1312 // not properly paired (was bug - gri 11/22/99). 1313 #ifdef AARCH64 1314 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI, true, Rsaved_result, noreg, Dsaved_result); 1315 #else 1316 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI, true, Rsaved_result_lo, Rsaved_result_hi, saved_result_fp); 1317 #endif // AARCH64 1318 1319 // Restore the result. Oop result is restored from the stack. 1320 #ifdef AARCH64 1321 __ mov(R0, Rsaved_result); 1322 __ fmov_d(D0, Dsaved_result); 1323 1324 __ blr(Rresult_handler); 1325 #else 1326 __ cmp(Rresult_handler, 0); 1327 __ ldr(R0, Address(FP, frame::interpreter_frame_oop_temp_offset * wordSize), ne); 1328 __ mov(R0, Rsaved_result_lo, eq); 1329 __ mov(R1, Rsaved_result_hi); 1330 1331 #ifdef __ABI_HARD__ 1332 // reload native FP result 1333 __ fcpyd(D0, D8); 1334 #endif // __ABI_HARD__ 1335 1336 #ifdef ASSERT 1337 if (VerifyOops) { 1338 Label L; 1339 __ cmp(Rresult_handler, 0); 1340 __ b(L, eq); 1341 __ verify_oop(R0); 1342 __ bind(L); 1343 } 1344 #endif // ASSERT 1345 #endif // AARCH64 1346 1347 // Restore FP/LR, sender_sp and return 1348 #ifdef AARCH64 1349 __ ldr(Rtemp, Address(FP, frame::interpreter_frame_sender_sp_offset * wordSize)); 1350 __ ldp(FP, LR, Address(FP)); 1351 __ mov(SP, Rtemp); 1352 #else 1353 __ mov(Rtemp, FP); 1354 __ ldmia(FP, RegisterSet(FP) | RegisterSet(LR)); 1355 __ ldr(SP, Address(Rtemp, frame::interpreter_frame_sender_sp_offset * wordSize)); 1356 #endif // AARCH64 1357 1358 __ ret(); 1359 1360 if (inc_counter) { 1361 // Handle overflow of counter and compile method 1362 __ bind(invocation_counter_overflow); 1363 generate_counter_overflow(continue_after_compile); 1364 } 1365 1366 return entry_point; 1367 } 1368 1369 // 1370 // Generic interpreted method entry to (asm) interpreter 1371 // 1372 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) { 1373 // determine code generation flags 1374 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; 1375 1376 // Rmethod: Method* 1377 // Rthread: thread 1378 // Rsender_sp: sender sp (could differ from SP if we were called via c2i) 1379 // Rparams: pointer to the last parameter in the stack 1380 1381 address entry_point = __ pc(); 1382 1383 const Register RconstMethod = AARCH64_ONLY(R10) NOT_AARCH64(R3); 1384 1385 #ifdef AARCH64 1386 const Register RmaxStack = R11; 1387 const Register RlocalsBase = R12; 1388 #endif // AARCH64 1389 1390 __ ldr(RconstMethod, Address(Rmethod, Method::const_offset())); 1391 1392 __ ldrh(R2, Address(RconstMethod, ConstMethod::size_of_parameters_offset())); 1393 __ ldrh(R3, Address(RconstMethod, ConstMethod::size_of_locals_offset())); 1394 1395 // setup Rlocals 1396 __ sub(Rlocals, Rparams, wordSize); 1397 __ add(Rlocals, Rlocals, AsmOperand(R2, lsl, Interpreter::logStackElementSize)); 1398 1399 __ sub(R3, R3, R2); // number of additional locals 1400 1401 #ifdef AARCH64 1402 // setup RmaxStack 1403 __ ldrh(RmaxStack, Address(RconstMethod, ConstMethod::max_stack_offset())); 1404 __ add(RmaxStack, RmaxStack, MAX2(1, Method::extra_stack_entries())); // reserve slots for exception handler and JSR292 appendix argument 1405 #endif // AARCH64 1406 1407 // see if we've got enough room on the stack for locals plus overhead. 1408 generate_stack_overflow_check(); 1409 1410 #ifdef AARCH64 1411 1412 // allocate space for locals 1413 { 1414 __ sub(RlocalsBase, Rparams, AsmOperand(R3, lsl, Interpreter::logStackElementSize)); 1415 __ align_reg(SP, RlocalsBase, StackAlignmentInBytes); 1416 } 1417 1418 // explicitly initialize locals 1419 { 1420 Label zero_loop, done; 1421 __ cbz(R3, done); 1422 1423 __ tbz(R3, 0, zero_loop); 1424 __ subs(R3, R3, 1); 1425 __ str(ZR, Address(RlocalsBase, wordSize, post_indexed)); 1426 __ b(done, eq); 1427 1428 __ bind(zero_loop); 1429 __ subs(R3, R3, 2); 1430 __ stp(ZR, ZR, Address(RlocalsBase, 2*wordSize, post_indexed)); 1431 __ b(zero_loop, ne); 1432 1433 __ bind(done); 1434 } 1435 1436 #else 1437 // allocate space for locals 1438 // explicitly initialize locals 1439 1440 // Loop is unrolled 4 times 1441 Label loop; 1442 __ mov(R0, 0); 1443 __ bind(loop); 1444 1445 // #1 1446 __ subs(R3, R3, 1); 1447 __ push(R0, ge); 1448 1449 // #2 1450 __ subs(R3, R3, 1, ge); 1451 __ push(R0, ge); 1452 1453 // #3 1454 __ subs(R3, R3, 1, ge); 1455 __ push(R0, ge); 1456 1457 // #4 1458 __ subs(R3, R3, 1, ge); 1459 __ push(R0, ge); 1460 1461 __ b(loop, gt); 1462 #endif // AARCH64 1463 1464 // initialize fixed part of activation frame 1465 generate_fixed_frame(false); 1466 1467 __ restore_dispatch(); 1468 1469 // make sure method is not native & not abstract 1470 #ifdef ASSERT 1471 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset())); 1472 { 1473 Label L; 1474 __ tbz(Rtemp, JVM_ACC_NATIVE_BIT, L); 1475 __ stop("tried to execute native method as non-native"); 1476 __ bind(L); 1477 } 1478 { Label L; 1479 __ tbz(Rtemp, JVM_ACC_ABSTRACT_BIT, L); 1480 __ stop("tried to execute abstract method in interpreter"); 1481 __ bind(L); 1482 } 1483 #endif 1484 1485 // increment invocation count & check for overflow 1486 Label invocation_counter_overflow; 1487 Label profile_method; 1488 Label profile_method_continue; 1489 if (inc_counter) { 1490 if (synchronized) { 1491 // Avoid unlocking method's monitor in case of exception, as it has not 1492 // been locked yet. 1493 __ set_do_not_unlock_if_synchronized(true, Rtemp); 1494 } 1495 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue); 1496 if (ProfileInterpreter) { 1497 __ bind(profile_method_continue); 1498 } 1499 } 1500 Label continue_after_compile; 1501 __ bind(continue_after_compile); 1502 1503 if (inc_counter && synchronized) { 1504 __ set_do_not_unlock_if_synchronized(false, Rtemp); 1505 } 1506 #if R9_IS_SCRATCHED 1507 __ restore_method(); 1508 #endif 1509 1510 // check for synchronized methods 1511 // Must happen AFTER invocation_counter check and stack overflow check, 1512 // so method is not locked if overflows. 1513 // 1514 if (synchronized) { 1515 // Allocate monitor and lock method 1516 lock_method(); 1517 } else { 1518 // no synchronization necessary 1519 #ifdef ASSERT 1520 { Label L; 1521 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset())); 1522 __ tbz(Rtemp, JVM_ACC_SYNCHRONIZED_BIT, L); 1523 __ stop("method needs synchronization"); 1524 __ bind(L); 1525 } 1526 #endif 1527 } 1528 1529 // start execution 1530 #ifdef ASSERT 1531 { Label L; 1532 __ ldr(Rtemp, Address(FP, frame::interpreter_frame_monitor_block_top_offset * wordSize)); 1533 __ cmp(Rtemp, Rstack_top); 1534 __ b(L, eq); 1535 __ stop("broken stack frame setup in interpreter"); 1536 __ bind(L); 1537 } 1538 #endif 1539 __ check_extended_sp(Rtemp); 1540 1541 // jvmti support 1542 __ notify_method_entry(); 1543 #if R9_IS_SCRATCHED 1544 __ restore_method(); 1545 #endif 1546 1547 __ dispatch_next(vtos); 1548 1549 // invocation counter overflow 1550 if (inc_counter) { 1551 if (ProfileInterpreter) { 1552 // We have decided to profile this method in the interpreter 1553 __ bind(profile_method); 1554 1555 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1556 __ set_method_data_pointer_for_bcp(); 1557 1558 __ b(profile_method_continue); 1559 } 1560 1561 // Handle overflow of counter and compile method 1562 __ bind(invocation_counter_overflow); 1563 generate_counter_overflow(continue_after_compile); 1564 } 1565 1566 return entry_point; 1567 } 1568 1569 //------------------------------------------------------------------------------------------------------------------------ 1570 // Exceptions 1571 1572 void TemplateInterpreterGenerator::generate_throw_exception() { 1573 // Entry point in previous activation (i.e., if the caller was interpreted) 1574 Interpreter::_rethrow_exception_entry = __ pc(); 1575 // Rexception_obj: exception 1576 1577 #ifndef AARCH64 1578 // Clear interpreter_frame_last_sp. 1579 __ mov(Rtemp, 0); 1580 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize)); 1581 #endif // !AARCH64 1582 1583 #if R9_IS_SCRATCHED 1584 __ restore_method(); 1585 #endif 1586 __ restore_bcp(); 1587 __ restore_dispatch(); 1588 __ restore_locals(); 1589 1590 #ifdef AARCH64 1591 __ restore_sp_after_call(Rtemp); 1592 #endif // AARCH64 1593 1594 // Entry point for exceptions thrown within interpreter code 1595 Interpreter::_throw_exception_entry = __ pc(); 1596 1597 // expression stack is undefined here 1598 // Rexception_obj: exception 1599 // Rbcp: exception bcp 1600 __ verify_oop(Rexception_obj); 1601 1602 // expression stack must be empty before entering the VM in case of an exception 1603 __ empty_expression_stack(); 1604 // find exception handler address and preserve exception oop 1605 __ mov(R1, Rexception_obj); 1606 __ call_VM(Rexception_obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), R1); 1607 // R0: exception handler entry point 1608 // Rexception_obj: preserved exception oop 1609 // Rbcp: bcp for exception handler 1610 __ push_ptr(Rexception_obj); // push exception which is now the only value on the stack 1611 __ jump(R0); // jump to exception handler (may be _remove_activation_entry!) 1612 1613 // If the exception is not handled in the current frame the frame is removed and 1614 // the exception is rethrown (i.e. exception continuation is _rethrow_exception). 1615 // 1616 // Note: At this point the bci is still the bxi for the instruction which caused 1617 // the exception and the expression stack is empty. Thus, for any VM calls 1618 // at this point, GC will find a legal oop map (with empty expression stack). 1619 1620 // In current activation 1621 // tos: exception 1622 // Rbcp: exception bcp 1623 1624 // 1625 // JVMTI PopFrame support 1626 // 1627 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 1628 1629 #ifdef AARCH64 1630 __ restore_sp_after_call(Rtemp); // restore SP to extended SP 1631 #endif // AARCH64 1632 1633 __ empty_expression_stack(); 1634 1635 // Set the popframe_processing bit in _popframe_condition indicating that we are 1636 // currently handling popframe, so that call_VMs that may happen later do not trigger new 1637 // popframe handling cycles. 1638 1639 __ ldr_s32(Rtemp, Address(Rthread, JavaThread::popframe_condition_offset())); 1640 __ orr(Rtemp, Rtemp, (unsigned)JavaThread::popframe_processing_bit); 1641 __ str_32(Rtemp, Address(Rthread, JavaThread::popframe_condition_offset())); 1642 1643 { 1644 // Check to see whether we are returning to a deoptimized frame. 1645 // (The PopFrame call ensures that the caller of the popped frame is 1646 // either interpreted or compiled and deoptimizes it if compiled.) 1647 // In this case, we can't call dispatch_next() after the frame is 1648 // popped, but instead must save the incoming arguments and restore 1649 // them after deoptimization has occurred. 1650 // 1651 // Note that we don't compare the return PC against the 1652 // deoptimization blob's unpack entry because of the presence of 1653 // adapter frames in C2. 1654 Label caller_not_deoptimized; 1655 __ ldr(R0, Address(FP, frame::return_addr_offset * wordSize)); 1656 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), R0); 1657 __ cbnz_32(R0, caller_not_deoptimized); 1658 #ifdef AARCH64 1659 __ NOT_TESTED(); 1660 #endif 1661 1662 // Compute size of arguments for saving when returning to deoptimized caller 1663 __ restore_method(); 1664 __ ldr(R0, Address(Rmethod, Method::const_offset())); 1665 __ ldrh(R0, Address(R0, ConstMethod::size_of_parameters_offset())); 1666 1667 __ logical_shift_left(R1, R0, Interpreter::logStackElementSize); 1668 // Save these arguments 1669 __ restore_locals(); 1670 __ sub(R2, Rlocals, R1); 1671 __ add(R2, R2, wordSize); 1672 __ mov(R0, Rthread); 1673 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R0, R1, R2); 1674 1675 __ remove_activation(vtos, LR, 1676 /* throw_monitor_exception */ false, 1677 /* install_monitor_exception */ false, 1678 /* notify_jvmdi */ false); 1679 1680 // Inform deoptimization that it is responsible for restoring these arguments 1681 __ mov(Rtemp, JavaThread::popframe_force_deopt_reexecution_bit); 1682 __ str_32(Rtemp, Address(Rthread, JavaThread::popframe_condition_offset())); 1683 1684 // Continue in deoptimization handler 1685 __ ret(); 1686 1687 __ bind(caller_not_deoptimized); 1688 } 1689 1690 __ remove_activation(vtos, R4, 1691 /* throw_monitor_exception */ false, 1692 /* install_monitor_exception */ false, 1693 /* notify_jvmdi */ false); 1694 1695 #ifndef AARCH64 1696 // Finish with popframe handling 1697 // A previous I2C followed by a deoptimization might have moved the 1698 // outgoing arguments further up the stack. PopFrame expects the 1699 // mutations to those outgoing arguments to be preserved and other 1700 // constraints basically require this frame to look exactly as 1701 // though it had previously invoked an interpreted activation with 1702 // no space between the top of the expression stack (current 1703 // last_sp) and the top of stack. Rather than force deopt to 1704 // maintain this kind of invariant all the time we call a small 1705 // fixup routine to move the mutated arguments onto the top of our 1706 // expression stack if necessary. 1707 __ mov(R1, SP); 1708 __ ldr(R2, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize)); 1709 // PC must point into interpreter here 1710 __ set_last_Java_frame(SP, FP, true, Rtemp); 1711 __ mov(R0, Rthread); 1712 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), R0, R1, R2); 1713 __ reset_last_Java_frame(Rtemp); 1714 #endif // !AARCH64 1715 1716 #ifdef AARCH64 1717 __ restore_sp_after_call(Rtemp); 1718 __ restore_stack_top(); 1719 #else 1720 // Restore the last_sp and null it out 1721 __ ldr(SP, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize)); 1722 __ mov(Rtemp, (int)NULL_WORD); 1723 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize)); 1724 #endif // AARCH64 1725 1726 __ restore_bcp(); 1727 __ restore_dispatch(); 1728 __ restore_locals(); 1729 __ restore_method(); 1730 1731 // The method data pointer was incremented already during 1732 // call profiling. We have to restore the mdp for the current bcp. 1733 if (ProfileInterpreter) { 1734 __ set_method_data_pointer_for_bcp(); 1735 } 1736 1737 // Clear the popframe condition flag 1738 assert(JavaThread::popframe_inactive == 0, "adjust this code"); 1739 __ str_32(__ zero_register(Rtemp), Address(Rthread, JavaThread::popframe_condition_offset())); 1740 1741 #if INCLUDE_JVMTI 1742 { 1743 Label L_done; 1744 1745 __ ldrb(Rtemp, Address(Rbcp, 0)); 1746 __ cmp(Rtemp, Bytecodes::_invokestatic); 1747 __ b(L_done, ne); 1748 1749 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. 1750 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL. 1751 1752 // get local0 1753 __ ldr(R1, Address(Rlocals, 0)); 1754 __ mov(R2, Rmethod); 1755 __ mov(R3, Rbcp); 1756 __ call_VM(R0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R1, R2, R3); 1757 1758 __ cbz(R0, L_done); 1759 1760 __ str(R0, Address(Rstack_top)); 1761 __ bind(L_done); 1762 } 1763 #endif // INCLUDE_JVMTI 1764 1765 __ dispatch_next(vtos); 1766 // end of PopFrame support 1767 1768 Interpreter::_remove_activation_entry = __ pc(); 1769 1770 // preserve exception over this code sequence 1771 __ pop_ptr(R0_tos); 1772 __ str(R0_tos, Address(Rthread, JavaThread::vm_result_offset())); 1773 // remove the activation (without doing throws on illegalMonitorExceptions) 1774 __ remove_activation(vtos, Rexception_pc, false, true, false); 1775 // restore exception 1776 __ get_vm_result(Rexception_obj, Rtemp); 1777 1778 // Inbetween activations - previous activation type unknown yet 1779 // compute continuation point - the continuation point expects 1780 // the following registers set up: 1781 // 1782 // Rexception_obj: exception 1783 // Rexception_pc: return address/pc that threw exception 1784 // SP: expression stack of caller 1785 // FP: frame pointer of caller 1786 __ mov(c_rarg0, Rthread); 1787 __ mov(c_rarg1, Rexception_pc); 1788 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), c_rarg0, c_rarg1); 1789 // Note that an "issuing PC" is actually the next PC after the call 1790 1791 __ jump(R0); // jump to exception handler of caller 1792 } 1793 1794 1795 // 1796 // JVMTI ForceEarlyReturn support 1797 // 1798 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 1799 address entry = __ pc(); 1800 1801 #ifdef AARCH64 1802 __ restore_sp_after_call(Rtemp); // restore SP to extended SP 1803 #endif // AARCH64 1804 1805 __ restore_bcp(); 1806 __ restore_dispatch(); 1807 __ restore_locals(); 1808 1809 __ empty_expression_stack(); 1810 1811 __ load_earlyret_value(state); 1812 1813 // Clear the earlyret state 1814 __ ldr(Rtemp, Address(Rthread, JavaThread::jvmti_thread_state_offset())); 1815 1816 assert(JvmtiThreadState::earlyret_inactive == 0, "adjust this code"); 1817 __ str_32(__ zero_register(R2), Address(Rtemp, JvmtiThreadState::earlyret_state_offset())); 1818 1819 __ remove_activation(state, LR, 1820 false, /* throw_monitor_exception */ 1821 false, /* install_monitor_exception */ 1822 true); /* notify_jvmdi */ 1823 1824 #ifndef AARCH64 1825 // According to interpreter calling conventions, result is returned in R0/R1, 1826 // so ftos (S0) and dtos (D0) are moved to R0/R1. 1827 // This conversion should be done after remove_activation, as it uses 1828 // push(state) & pop(state) to preserve return value. 1829 __ convert_tos_to_retval(state); 1830 #endif // !AARCH64 1831 __ ret(); 1832 1833 return entry; 1834 } // end of ForceEarlyReturn support 1835 1836 1837 //------------------------------------------------------------------------------------------------------------------------ 1838 // Helper for vtos entry point generation 1839 1840 void TemplateInterpreterGenerator::set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) { 1841 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 1842 Label L; 1843 1844 #ifdef __SOFTFP__ 1845 dep = __ pc(); // fall through 1846 #else 1847 fep = __ pc(); __ push(ftos); __ b(L); 1848 dep = __ pc(); __ push(dtos); __ b(L); 1849 #endif // __SOFTFP__ 1850 1851 lep = __ pc(); __ push(ltos); __ b(L); 1852 1853 if (AARCH64_ONLY(true) NOT_AARCH64(VerifyOops)) { // can't share atos entry with itos on AArch64 or if VerifyOops 1854 aep = __ pc(); __ push(atos); __ b(L); 1855 } else { 1856 aep = __ pc(); // fall through 1857 } 1858 1859 #ifdef __SOFTFP__ 1860 fep = __ pc(); // fall through 1861 #endif // __SOFTFP__ 1862 1863 bep = cep = sep = // fall through 1864 iep = __ pc(); __ push(itos); // fall through 1865 vep = __ pc(); __ bind(L); // fall through 1866 generate_and_dispatch(t); 1867 } 1868 1869 //------------------------------------------------------------------------------------------------------------------------ 1870 1871 // Non-product code 1872 #ifndef PRODUCT 1873 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 1874 address entry = __ pc(); 1875 1876 // prepare expression stack 1877 __ push(state); // save tosca 1878 1879 // pass tosca registers as arguments 1880 __ mov(R2, R0_tos); 1881 #ifdef AARCH64 1882 __ mov(R3, ZR); 1883 #else 1884 __ mov(R3, R1_tos_hi); 1885 #endif // AARCH64 1886 __ mov(R1, LR); // save return address 1887 1888 // call tracer 1889 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), R1, R2, R3); 1890 1891 __ mov(LR, R0); // restore return address 1892 __ pop(state); // restore tosca 1893 1894 // return 1895 __ ret(); 1896 1897 return entry; 1898 } 1899 1900 1901 void TemplateInterpreterGenerator::count_bytecode() { 1902 __ inc_global_counter((address) &BytecodeCounter::_counter_value, 0, Rtemp, R2_tmp, true); 1903 } 1904 1905 1906 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 1907 __ inc_global_counter((address)&BytecodeHistogram::_counters[0], sizeof(BytecodeHistogram::_counters[0]) * t->bytecode(), Rtemp, R2_tmp, true); 1908 } 1909 1910 1911 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 1912 const Register Rindex_addr = R2_tmp; 1913 Label Lcontinue; 1914 InlinedAddress Lcounters((address)BytecodePairHistogram::_counters); 1915 InlinedAddress Lindex((address)&BytecodePairHistogram::_index); 1916 const Register Rcounters_addr = R2_tmp; 1917 const Register Rindex = R4_tmp; 1918 1919 // calculate new index for counter: 1920 // index = (_index >> log2_number_of_codes) | (bytecode << log2_number_of_codes). 1921 // (_index >> log2_number_of_codes) is previous bytecode 1922 1923 __ ldr_literal(Rindex_addr, Lindex); 1924 __ ldr_s32(Rindex, Address(Rindex_addr)); 1925 __ mov_slow(Rtemp, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); 1926 __ orr(Rindex, Rtemp, AsmOperand(Rindex, lsr, BytecodePairHistogram::log2_number_of_codes)); 1927 __ str_32(Rindex, Address(Rindex_addr)); 1928 1929 // Rindex (R4) contains index of counter 1930 1931 __ ldr_literal(Rcounters_addr, Lcounters); 1932 __ ldr_s32(Rtemp, Address::indexed_32(Rcounters_addr, Rindex)); 1933 __ adds_32(Rtemp, Rtemp, 1); 1934 __ b(Lcontinue, mi); // avoid overflow 1935 __ str_32(Rtemp, Address::indexed_32(Rcounters_addr, Rindex)); 1936 1937 __ b(Lcontinue); 1938 1939 __ bind_literal(Lindex); 1940 __ bind_literal(Lcounters); 1941 1942 __ bind(Lcontinue); 1943 } 1944 1945 1946 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 1947 // Call a little run-time stub to avoid blow-up for each bytecode. 1948 // The run-time runtime saves the right registers, depending on 1949 // the tosca in-state for the given template. 1950 assert(Interpreter::trace_code(t->tos_in()) != NULL, 1951 "entry must have been generated"); 1952 address trace_entry = Interpreter::trace_code(t->tos_in()); 1953 __ call(trace_entry, relocInfo::none); 1954 } 1955 1956 1957 void TemplateInterpreterGenerator::stop_interpreter_at() { 1958 Label Lcontinue; 1959 const Register stop_at = R2_tmp; 1960 1961 __ ldr_global_s32(Rtemp, (address) &BytecodeCounter::_counter_value); 1962 __ mov_slow(stop_at, StopInterpreterAt); 1963 1964 // test bytecode counter 1965 __ cmp(Rtemp, stop_at); 1966 __ b(Lcontinue, ne); 1967 1968 __ trace_state("stop_interpreter_at"); 1969 __ breakpoint(); 1970 1971 __ bind(Lcontinue); 1972 } 1973 #endif // !PRODUCT