1 /* 2 * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved. 3 * Copyright 2013, 2014 SAP AG. 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 #include "precompiled.hpp" 27 #ifndef CC_INTERP 28 #include "asm/macroAssembler.inline.hpp" 29 #include "interpreter/bytecodeHistogram.hpp" 30 #include "interpreter/interpreter.hpp" 31 #include "interpreter/interpreterGenerator.hpp" 32 #include "interpreter/interpreterRuntime.hpp" 33 #include "interpreter/interp_masm.hpp" 34 #include "interpreter/templateTable.hpp" 35 #include "oops/arrayOop.hpp" 36 #include "oops/methodData.hpp" 37 #include "oops/method.hpp" 38 #include "oops/oop.inline.hpp" 39 #include "prims/jvmtiExport.hpp" 40 #include "prims/jvmtiThreadState.hpp" 41 #include "runtime/arguments.hpp" 42 #include "runtime/deoptimization.hpp" 43 #include "runtime/frame.inline.hpp" 44 #include "runtime/sharedRuntime.hpp" 45 #include "runtime/stubRoutines.hpp" 46 #include "runtime/synchronizer.hpp" 47 #include "runtime/timer.hpp" 48 #include "runtime/vframeArray.hpp" 49 #include "utilities/debug.hpp" 50 #include "utilities/macros.hpp" 51 52 #undef __ 53 #define __ _masm-> 54 55 #ifdef PRODUCT 56 #define BLOCK_COMMENT(str) /* nothing */ 57 #else 58 #define BLOCK_COMMENT(str) __ block_comment(str) 59 #endif 60 61 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") 62 63 //----------------------------------------------------------------------------- 64 65 // Actually we should never reach here since we do stack overflow checks before pushing any frame. 66 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { 67 address entry = __ pc(); 68 __ unimplemented("generate_StackOverflowError_handler"); 69 return entry; 70 } 71 72 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) { 73 address entry = __ pc(); 74 __ empty_expression_stack(); 75 __ load_const_optimized(R4_ARG2, (address) name); 76 // Index is in R17_tos. 77 __ mr(R5_ARG3, R17_tos); 78 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException)); 79 return entry; 80 } 81 82 #if 0 83 // Call special ClassCastException constructor taking object to cast 84 // and target class as arguments. 85 address TemplateInterpreterGenerator::generate_ClassCastException_verbose_handler() { 86 address entry = __ pc(); 87 88 // Expression stack must be empty before entering the VM if an 89 // exception happened. 90 __ empty_expression_stack(); 91 92 // Thread will be loaded to R3_ARG1. 93 // Target class oop is in register R5_ARG3 by convention! 94 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose, R17_tos, R5_ARG3)); 95 // Above call must not return here since exception pending. 96 DEBUG_ONLY(__ should_not_reach_here();) 97 return entry; 98 } 99 #endif 100 101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { 102 address entry = __ pc(); 103 // Expression stack must be empty before entering the VM if an 104 // exception happened. 105 __ empty_expression_stack(); 106 107 // Load exception object. 108 // Thread will be loaded to R3_ARG1. 109 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos); 110 #ifdef ASSERT 111 // Above call must not return here since exception pending. 112 __ should_not_reach_here(); 113 #endif 114 return entry; 115 } 116 117 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) { 118 address entry = __ pc(); 119 //__ untested("generate_exception_handler_common"); 120 Register Rexception = R17_tos; 121 122 // Expression stack must be empty before entering the VM if an exception happened. 123 __ empty_expression_stack(); 124 125 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1); 126 if (pass_oop) { 127 __ mr(R5_ARG3, Rexception); 128 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false); 129 } else { 130 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1); 131 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false); 132 } 133 134 // Throw exception. 135 __ mr(R3_ARG1, Rexception); 136 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2); 137 __ mtctr(R11_scratch1); 138 __ bctr(); 139 140 return entry; 141 } 142 143 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { 144 address entry = __ pc(); 145 __ unimplemented("generate_continuation_for"); 146 return entry; 147 } 148 149 // This entry is returned to when a call returns to the interpreter. 150 // When we arrive here, we expect that the callee stack frame is already popped. 151 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) { 152 address entry = __ pc(); 153 154 // Move the value out of the return register back to the TOS cache of current frame. 155 switch (state) { 156 case ltos: 157 case btos: 158 case ctos: 159 case stos: 160 case atos: 161 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache 162 case ftos: 163 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET 164 case vtos: break; // Nothing to do, this was a void return. 165 default : ShouldNotReachHere(); 166 } 167 168 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp. 169 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 170 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 171 172 // Compiled code destroys templateTableBase, reload. 173 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2); 174 175 const Register cache = R11_scratch1; 176 const Register size = R12_scratch2; 177 __ get_cache_and_index_at_bcp(cache, 1, index_size); 178 179 // Get least significant byte of 64 bit value: 180 #if defined(VM_LITTLE_ENDIAN) 181 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache); 182 #else 183 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache); 184 #endif 185 __ sldi(size, size, Interpreter::logStackElementSize); 186 __ add(R15_esp, R15_esp, size); 187 __ dispatch_next(state, step); 188 return entry; 189 } 190 191 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) { 192 address entry = __ pc(); 193 // If state != vtos, we're returning from a native method, which put it's result 194 // into the result register. So move the value out of the return register back 195 // to the TOS cache of current frame. 196 197 switch (state) { 198 case ltos: 199 case btos: 200 case ctos: 201 case stos: 202 case atos: 203 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache 204 case ftos: 205 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET 206 case vtos: break; // Nothing to do, this was a void return. 207 default : ShouldNotReachHere(); 208 } 209 210 // Load LcpoolCache @@@ should be already set! 211 __ get_constant_pool_cache(R27_constPoolCache); 212 213 // Handle a pending exception, fall through if none. 214 __ check_and_forward_exception(R11_scratch1, R12_scratch2); 215 216 // Start executing bytecodes. 217 __ dispatch_next(state, step); 218 219 return entry; 220 } 221 222 // A result handler converts the native result into java format. 223 // Use the shared code between c++ and template interpreter. 224 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) { 225 return AbstractInterpreterGenerator::generate_result_handler_for(type); 226 } 227 228 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) { 229 address entry = __ pc(); 230 231 __ push(state); 232 __ call_VM(noreg, runtime_entry); 233 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); 234 235 return entry; 236 } 237 238 // Helpers for commoning out cases in the various type of method entries. 239 240 // Increment invocation count & check for overflow. 241 // 242 // Note: checking for negative value instead of overflow 243 // so we have a 'sticky' overflow test. 244 // 245 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { 246 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not. 247 Register Rscratch1 = R11_scratch1; 248 Register Rscratch2 = R12_scratch2; 249 Register R3_counters = R3_ARG1; 250 Label done; 251 252 if (TieredCompilation) { 253 const int increment = InvocationCounter::count_increment; 254 const int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift; 255 Label no_mdo; 256 if (ProfileInterpreter) { 257 const Register Rmdo = Rscratch1; 258 // If no method data exists, go to profile_continue. 259 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method); 260 __ cmpdi(CCR0, Rmdo, 0); 261 __ beq(CCR0, no_mdo); 262 263 // Increment backedge counter in the MDO. 264 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); 265 __ lwz(Rscratch2, mdo_bc_offs, Rmdo); 266 __ addi(Rscratch2, Rscratch2, increment); 267 __ stw(Rscratch2, mdo_bc_offs, Rmdo); 268 __ load_const_optimized(Rscratch1, mask, R0); 269 __ and_(Rscratch1, Rscratch2, Rscratch1); 270 __ bne(CCR0, done); 271 __ b(*overflow); 272 } 273 274 // Increment counter in MethodCounters*. 275 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); 276 __ bind(no_mdo); 277 __ get_method_counters(R19_method, R3_counters, done); 278 __ lwz(Rscratch2, mo_bc_offs, R3_counters); 279 __ addi(Rscratch2, Rscratch2, increment); 280 __ stw(Rscratch2, mo_bc_offs, R3_counters); 281 __ load_const_optimized(Rscratch1, mask, R0); 282 __ and_(Rscratch1, Rscratch2, Rscratch1); 283 __ beq(CCR0, *overflow); 284 285 __ bind(done); 286 287 } else { 288 289 // Update standard invocation counters. 290 Register Rsum_ivc_bec = R4_ARG2; 291 __ get_method_counters(R19_method, R3_counters, done); 292 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2); 293 // Increment interpreter invocation counter. 294 if (ProfileInterpreter) { // %%% Merge this into methodDataOop. 295 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters); 296 __ addi(R12_scratch2, R12_scratch2, 1); 297 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters); 298 } 299 // Check if we must create a method data obj. 300 if (ProfileInterpreter && profile_method != NULL) { 301 const Register profile_limit = Rscratch1; 302 int pl_offs = __ load_const_optimized(profile_limit, &InvocationCounter::InterpreterProfileLimit, R0, true); 303 __ lwz(profile_limit, pl_offs, profile_limit); 304 // Test to see if we should create a method data oop. 305 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit); 306 __ blt(CCR0, *profile_method_continue); 307 // If no method data exists, go to profile_method. 308 __ test_method_data_pointer(*profile_method); 309 } 310 // Finally check for counter overflow. 311 if (overflow) { 312 const Register invocation_limit = Rscratch1; 313 int il_offs = __ load_const_optimized(invocation_limit, &InvocationCounter::InterpreterInvocationLimit, R0, true); 314 __ lwz(invocation_limit, il_offs, invocation_limit); 315 assert(4 == sizeof(InvocationCounter::InterpreterInvocationLimit), "unexpected field size"); 316 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit); 317 __ bge(CCR0, *overflow); 318 } 319 320 __ bind(done); 321 } 322 } 323 324 // Generate code to initiate compilation on invocation counter overflow. 325 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) { 326 // Generate code to initiate compilation on the counter overflow. 327 328 // InterpreterRuntime::frequency_counter_overflow takes one arguments, 329 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp) 330 // We pass zero in. 331 // The call returns the address of the verified entry point for the method or NULL 332 // if the compilation did not complete (either went background or bailed out). 333 // 334 // Unlike the C++ interpreter above: Check exceptions! 335 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed 336 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur. 337 338 __ li(R4_ARG2, 0); 339 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); 340 341 // Returns verified_entry_point or NULL. 342 // We ignore it in any case. 343 __ b(continue_entry); 344 } 345 346 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) { 347 assert_different_registers(Rmem_frame_size, Rscratch1); 348 __ generate_stack_overflow_check_with_compare_and_throw(Rmem_frame_size, Rscratch1); 349 } 350 351 void TemplateInterpreterGenerator::unlock_method(bool check_exceptions) { 352 __ unlock_object(R26_monitor, check_exceptions); 353 } 354 355 // Lock the current method, interpreter register window must be set up! 356 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) { 357 const Register Robj_to_lock = Rscratch2; 358 359 { 360 if (!flags_preloaded) { 361 __ lwz(Rflags, method_(access_flags)); 362 } 363 364 #ifdef ASSERT 365 // Check if methods needs synchronization. 366 { 367 Label Lok; 368 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT); 369 __ btrue(CCR0,Lok); 370 __ stop("method doesn't need synchronization"); 371 __ bind(Lok); 372 } 373 #endif // ASSERT 374 } 375 376 // Get synchronization object to Rscratch2. 377 { 378 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 379 Label Lstatic; 380 Label Ldone; 381 382 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT); 383 __ btrue(CCR0, Lstatic); 384 385 // Non-static case: load receiver obj from stack and we're done. 386 __ ld(Robj_to_lock, R18_locals); 387 __ b(Ldone); 388 389 __ bind(Lstatic); // Static case: Lock the java mirror 390 __ ld(Robj_to_lock, in_bytes(Method::const_offset()), R19_method); 391 __ ld(Robj_to_lock, in_bytes(ConstMethod::constants_offset()), Robj_to_lock); 392 __ ld(Robj_to_lock, ConstantPool::pool_holder_offset_in_bytes(), Robj_to_lock); 393 __ ld(Robj_to_lock, mirror_offset, Robj_to_lock); 394 395 __ bind(Ldone); 396 __ verify_oop(Robj_to_lock); 397 } 398 399 // Got the oop to lock => execute! 400 __ add_monitor_to_stack(true, Rscratch1, R0); 401 402 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor); 403 __ lock_object(R26_monitor, Robj_to_lock); 404 } 405 406 // Generate a fixed interpreter frame for pure interpreter 407 // and I2N native transition frames. 408 // 409 // Before (stack grows downwards): 410 // 411 // | ... | 412 // |------------- | 413 // | java arg0 | 414 // | ... | 415 // | java argn | 416 // | | <- R15_esp 417 // | | 418 // |--------------| 419 // | abi_112 | 420 // | | <- R1_SP 421 // |==============| 422 // 423 // 424 // After: 425 // 426 // | ... | 427 // | java arg0 |<- R18_locals 428 // | ... | 429 // | java argn | 430 // |--------------| 431 // | | 432 // | java locals | 433 // | | 434 // |--------------| 435 // | abi_48 | 436 // |==============| 437 // | | 438 // | istate | 439 // | | 440 // |--------------| 441 // | monitor |<- R26_monitor 442 // |--------------| 443 // | |<- R15_esp 444 // | expression | 445 // | stack | 446 // | | 447 // |--------------| 448 // | | 449 // | abi_112 |<- R1_SP 450 // |==============| 451 // 452 // The top most frame needs an abi space of 112 bytes. This space is needed, 453 // since we call to c. The c function may spill their arguments to the caller 454 // frame. When we call to java, we don't need these spill slots. In order to save 455 // space on the stack, we resize the caller. However, java local reside in 456 // the caller frame and the frame has to be increased. The frame_size for the 457 // current frame was calculated based on max_stack as size for the expression 458 // stack. At the call, just a part of the expression stack might be used. 459 // We don't want to waste this space and cut the frame back accordingly. 460 // The resulting amount for resizing is calculated as follows: 461 // resize = (number_of_locals - number_of_arguments) * slot_size 462 // + (R1_SP - R15_esp) + 48 463 // 464 // The size for the callee frame is calculated: 465 // framesize = 112 + max_stack + monitor + state_size 466 // 467 // maxstack: Max number of slots on the expression stack, loaded from the method. 468 // monitor: We statically reserve room for one monitor object. 469 // state_size: We save the current state of the interpreter to this area. 470 // 471 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) { 472 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes. 473 top_frame_size = R7_ARG5, 474 Rconst_method = R8_ARG6; 475 476 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size); 477 478 __ ld(Rconst_method, method_(const)); 479 __ lhz(Rsize_of_parameters /* number of params */, 480 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method); 481 if (native_call) { 482 // If we're calling a native method, we reserve space for the worst-case signature 483 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2). 484 // We add two slots to the parameter_count, one for the jni 485 // environment and one for a possible native mirror. 486 Label skip_native_calculate_max_stack; 487 __ addi(top_frame_size, Rsize_of_parameters, 2); 488 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters); 489 __ bge(CCR0, skip_native_calculate_max_stack); 490 __ li(top_frame_size, Argument::n_register_parameters); 491 __ bind(skip_native_calculate_max_stack); 492 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); 493 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize); 494 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! 495 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters. 496 } else { 497 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method); 498 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); 499 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize); 500 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method); 501 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0 502 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! 503 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize); 504 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1); 505 } 506 507 // Compute top frame size. 508 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size); 509 510 // Cut back area between esp and max_stack. 511 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize); 512 513 __ round_to(top_frame_size, frame::alignment_in_bytes); 514 __ round_to(parent_frame_resize, frame::alignment_in_bytes); 515 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size. 516 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48. 517 518 { 519 // -------------------------------------------------------------------------- 520 // Stack overflow check 521 522 Label cont; 523 __ add(R11_scratch1, parent_frame_resize, top_frame_size); 524 generate_stack_overflow_check(R11_scratch1, R12_scratch2); 525 } 526 527 // Set up interpreter state registers. 528 529 __ add(R18_locals, R15_esp, Rsize_of_parameters); 530 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method); 531 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache); 532 533 // Set method data pointer. 534 if (ProfileInterpreter) { 535 Label zero_continue; 536 __ ld(R28_mdx, method_(method_data)); 537 __ cmpdi(CCR0, R28_mdx, 0); 538 __ beq(CCR0, zero_continue); 539 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); 540 __ bind(zero_continue); 541 } 542 543 if (native_call) { 544 __ li(R14_bcp, 0); // Must initialize. 545 } else { 546 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method); 547 } 548 549 // Resize parent frame. 550 __ mflr(R12_scratch2); 551 __ neg(parent_frame_resize, parent_frame_resize); 552 __ resize_frame(parent_frame_resize, R11_scratch1); 553 __ std(R12_scratch2, _abi(lr), R1_SP); 554 555 __ addi(R26_monitor, R1_SP, - frame::ijava_state_size); 556 __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); 557 558 // Store values. 559 // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls 560 // in InterpreterMacroAssembler::call_from_interpreter. 561 __ std(R19_method, _ijava_state_neg(method), R1_SP); 562 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP); 563 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP); 564 __ std(R18_locals, _ijava_state_neg(locals), R1_SP); 565 566 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only 567 // be found in the frame after save_interpreter_state is done. This is always true 568 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong, 569 // because e.g. frame::interpreter_frame_bcp() will not access the correct value 570 // (Enhanced Stack Trace). 571 // The signal handler does not save the interpreter state into the frame. 572 __ li(R0, 0); 573 #ifdef ASSERT 574 // Fill remaining slots with constants. 575 __ load_const_optimized(R11_scratch1, 0x5afe); 576 __ load_const_optimized(R12_scratch2, 0xdead); 577 #endif 578 // We have to initialize some frame slots for native calls (accessed by GC). 579 if (native_call) { 580 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP); 581 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP); 582 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); } 583 } 584 #ifdef ASSERT 585 else { 586 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP); 587 __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP); 588 __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP); 589 } 590 __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP); 591 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP); 592 __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP); 593 __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP); 594 #endif 595 __ subf(R12_scratch2, top_frame_size, R1_SP); 596 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); 597 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP); 598 599 // Push top frame. 600 __ push_frame(top_frame_size, R11_scratch1); 601 } 602 603 // End of helpers 604 605 606 // Support abs and sqrt like in compiler. 607 // For others we can use a normal (native) entry. 608 609 inline bool math_entry_available(AbstractInterpreter::MethodKind kind) { 610 // Provide math entry with debugging on demand. 611 // Note: Debugging changes which code will get executed: 612 // Debugging or disabled InlineIntrinsics: java method will get interpreted and performs a native call. 613 // Not debugging and enabled InlineIntrinics: processor instruction will get used. 614 // Result might differ slightly due to rounding etc. 615 if (!InlineIntrinsics && (!FLAG_IS_ERGO(InlineIntrinsics))) return false; // Generate a vanilla entry. 616 617 return ((kind==Interpreter::java_lang_math_sqrt && VM_Version::has_fsqrt()) || 618 (kind==Interpreter::java_lang_math_abs)); 619 } 620 621 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) { 622 if (!math_entry_available(kind)) { 623 NOT_PRODUCT(__ should_not_reach_here();) 624 return Interpreter::entry_for_kind(Interpreter::zerolocals); 625 } 626 627 Label Lslow_path; 628 const Register Rjvmti_mode = R11_scratch1; 629 address entry = __ pc(); 630 631 // Provide math entry with debugging on demand. 632 __ lwz(Rjvmti_mode, thread_(interp_only_mode)); 633 __ cmpwi(CCR0, Rjvmti_mode, 0); 634 __ bne(CCR0, Lslow_path); // jvmti_mode!=0 635 636 __ lfd(F1_RET, Interpreter::stackElementSize, R15_esp); 637 638 // Pop c2i arguments (if any) off when we return. 639 #ifdef ASSERT 640 __ ld(R9_ARG7, 0, R1_SP); 641 __ ld(R10_ARG8, 0, R21_sender_SP); 642 __ cmpd(CCR0, R9_ARG7, R10_ARG8); 643 __ asm_assert_eq("backlink", 0x545); 644 #endif // ASSERT 645 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. 646 647 if (kind == Interpreter::java_lang_math_sqrt) { 648 __ fsqrt(F1_RET, F1_RET); 649 } else if (kind == Interpreter::java_lang_math_abs) { 650 __ fabs(F1_RET, F1_RET); 651 } else { 652 ShouldNotReachHere(); 653 } 654 655 // And we're done. 656 __ blr(); 657 658 // Provide slow path for JVMTI case. 659 __ bind(Lslow_path); 660 __ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R12_scratch2); 661 __ flush(); 662 663 return entry; 664 } 665 666 // Interpreter stub for calling a native method. (asm interpreter) 667 // This sets up a somewhat different looking stack for calling the 668 // native method than the typical interpreter frame setup. 669 // 670 // On entry: 671 // R19_method - method 672 // R16_thread - JavaThread* 673 // R15_esp - intptr_t* sender tos 674 // 675 // abstract stack (grows up) 676 // [ IJava (caller of JNI callee) ] <-- ASP 677 // ... 678 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) { 679 680 address entry = __ pc(); 681 682 const bool inc_counter = UseCompiler || CountCompiledCalls; 683 684 // ----------------------------------------------------------------------------- 685 // Allocate a new frame that represents the native callee (i2n frame). 686 // This is not a full-blown interpreter frame, but in particular, the 687 // following registers are valid after this: 688 // - R19_method 689 // - R18_local (points to start of argumuments to native function) 690 // 691 // abstract stack (grows up) 692 // [ IJava (caller of JNI callee) ] <-- ASP 693 // ... 694 695 const Register signature_handler_fd = R11_scratch1; 696 const Register pending_exception = R0; 697 const Register result_handler_addr = R31; 698 const Register native_method_fd = R11_scratch1; 699 const Register access_flags = R22_tmp2; 700 const Register active_handles = R11_scratch1; // R26_monitor saved to state. 701 const Register sync_state = R12_scratch2; 702 const Register sync_state_addr = sync_state; // Address is dead after use. 703 const Register suspend_flags = R11_scratch1; 704 705 //============================================================================= 706 // Allocate new frame and initialize interpreter state. 707 708 Label exception_return; 709 Label exception_return_sync_check; 710 Label stack_overflow_return; 711 712 // Generate new interpreter state and jump to stack_overflow_return in case of 713 // a stack overflow. 714 //generate_compute_interpreter_state(stack_overflow_return); 715 716 Register size_of_parameters = R22_tmp2; 717 718 generate_fixed_frame(true, size_of_parameters, noreg /* unused */); 719 720 //============================================================================= 721 // Increment invocation counter. On overflow, entry to JNI method 722 // will be compiled. 723 Label invocation_counter_overflow, continue_after_compile; 724 if (inc_counter) { 725 if (synchronized) { 726 // Since at this point in the method invocation the exception handler 727 // would try to exit the monitor of synchronized methods which hasn't 728 // been entered yet, we set the thread local variable 729 // _do_not_unlock_if_synchronized to true. If any exception was thrown by 730 // runtime, exception handling i.e. unlock_if_synchronized_method will 731 // check this thread local flag. 732 // This flag has two effects, one is to force an unwind in the topmost 733 // interpreter frame and not perform an unlock while doing so. 734 __ li(R0, 1); 735 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 736 } 737 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); 738 739 __ BIND(continue_after_compile); 740 // Reset the _do_not_unlock_if_synchronized flag. 741 if (synchronized) { 742 __ li(R0, 0); 743 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 744 } 745 } 746 747 // access_flags = method->access_flags(); 748 // Load access flags. 749 assert(access_flags->is_nonvolatile(), 750 "access_flags must be in a non-volatile register"); 751 // Type check. 752 assert(4 == sizeof(AccessFlags), "unexpected field size"); 753 __ lwz(access_flags, method_(access_flags)); 754 755 // We don't want to reload R19_method and access_flags after calls 756 // to some helper functions. 757 assert(R19_method->is_nonvolatile(), 758 "R19_method must be a non-volatile register"); 759 760 // Check for synchronized methods. Must happen AFTER invocation counter 761 // check, so method is not locked if counter overflows. 762 763 if (synchronized) { 764 lock_method(access_flags, R11_scratch1, R12_scratch2, true); 765 766 // Update monitor in state. 767 __ ld(R11_scratch1, 0, R1_SP); 768 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1); 769 } 770 771 // jvmti/jvmpi support 772 __ notify_method_entry(); 773 774 //============================================================================= 775 // Get and call the signature handler. 776 777 __ ld(signature_handler_fd, method_(signature_handler)); 778 Label call_signature_handler; 779 780 __ cmpdi(CCR0, signature_handler_fd, 0); 781 __ bne(CCR0, call_signature_handler); 782 783 // Method has never been called. Either generate a specialized 784 // handler or point to the slow one. 785 // 786 // Pass parameter 'false' to avoid exception check in call_VM. 787 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false); 788 789 // Check for an exception while looking up the target method. If we 790 // incurred one, bail. 791 __ ld(pending_exception, thread_(pending_exception)); 792 __ cmpdi(CCR0, pending_exception, 0); 793 __ bne(CCR0, exception_return_sync_check); // Has pending exception. 794 795 // Reload signature handler, it may have been created/assigned in the meanwhile. 796 __ ld(signature_handler_fd, method_(signature_handler)); 797 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below). 798 799 __ BIND(call_signature_handler); 800 801 // Before we call the signature handler we push a new frame to 802 // protect the interpreter frame volatile registers when we return 803 // from jni but before we can get back to Java. 804 805 // First set the frame anchor while the SP/FP registers are 806 // convenient and the slow signature handler can use this same frame 807 // anchor. 808 809 // We have a TOP_IJAVA_FRAME here, which belongs to us. 810 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/); 811 812 // Now the interpreter frame (and its call chain) have been 813 // invalidated and flushed. We are now protected against eager 814 // being enabled in native code. Even if it goes eager the 815 // registers will be reloaded as clean and we will invalidate after 816 // the call so no spurious flush should be possible. 817 818 // Call signature handler and pass locals address. 819 // 820 // Our signature handlers copy required arguments to the C stack 821 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13. 822 __ mr(R3_ARG1, R18_locals); 823 #if !defined(ABI_ELFv2) 824 __ ld(signature_handler_fd, 0, signature_handler_fd); 825 #endif 826 827 __ call_stub(signature_handler_fd); 828 829 // Remove the register parameter varargs slots we allocated in 830 // compute_interpreter_state. SP+16 ends up pointing to the ABI 831 // outgoing argument area. 832 // 833 // Not needed on PPC64. 834 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord); 835 836 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register"); 837 // Save across call to native method. 838 __ mr(result_handler_addr, R3_RET); 839 840 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror. 841 842 // Set up fixed parameters and call the native method. 843 // If the method is static, get mirror into R4_ARG2. 844 { 845 Label method_is_not_static; 846 // Access_flags is non-volatile and still, no need to restore it. 847 848 // Restore access flags. 849 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT); 850 __ bfalse(CCR0, method_is_not_static); 851 852 // constants = method->constants(); 853 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method); 854 __ ld(R11_scratch1, in_bytes(ConstMethod::constants_offset()), R11_scratch1); 855 // pool_holder = method->constants()->pool_holder(); 856 __ ld(R11_scratch1/*pool_holder*/, ConstantPool::pool_holder_offset_in_bytes(), 857 R11_scratch1/*constants*/); 858 859 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 860 861 // mirror = pool_holder->klass_part()->java_mirror(); 862 __ ld(R0/*mirror*/, mirror_offset, R11_scratch1/*pool_holder*/); 863 // state->_native_mirror = mirror; 864 865 __ ld(R11_scratch1, 0, R1_SP); 866 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); 867 // R4_ARG2 = &state->_oop_temp; 868 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp)); 869 __ BIND(method_is_not_static); 870 } 871 872 // At this point, arguments have been copied off the stack into 873 // their JNI positions. Oops are boxed in-place on the stack, with 874 // handles copied to arguments. The result handler address is in a 875 // register. 876 877 // Pass JNIEnv address as first parameter. 878 __ addir(R3_ARG1, thread_(jni_environment)); 879 880 // Load the native_method entry before we change the thread state. 881 __ ld(native_method_fd, method_(native_function)); 882 883 //============================================================================= 884 // Transition from _thread_in_Java to _thread_in_native. As soon as 885 // we make this change the safepoint code needs to be certain that 886 // the last Java frame we established is good. The pc in that frame 887 // just needs to be near here not an actual return address. 888 889 // We use release_store_fence to update values like the thread state, where 890 // we don't want the current thread to continue until all our prior memory 891 // accesses (including the new thread state) are visible to other threads. 892 __ li(R0, _thread_in_native); 893 __ release(); 894 895 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size"); 896 __ stw(R0, thread_(thread_state)); 897 898 if (UseMembar) { 899 __ fence(); 900 } 901 902 //============================================================================= 903 // Call the native method. Argument registers must not have been 904 // overwritten since "__ call_stub(signature_handler);" (except for 905 // ARG1 and ARG2 for static methods). 906 __ call_c(native_method_fd); 907 908 __ li(R0, 0); 909 __ ld(R11_scratch1, 0, R1_SP); 910 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1); 911 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); 912 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset 913 914 // Note: C++ interpreter needs the following here: 915 // The frame_manager_lr field, which we use for setting the last 916 // java frame, gets overwritten by the signature handler. Restore 917 // it now. 918 //__ get_PC_trash_LR(R11_scratch1); 919 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP); 920 921 // Because of GC R19_method may no longer be valid. 922 923 // Block, if necessary, before resuming in _thread_in_Java state. 924 // In order for GC to work, don't clear the last_Java_sp until after 925 // blocking. 926 927 //============================================================================= 928 // Switch thread to "native transition" state before reading the 929 // synchronization state. This additional state is necessary 930 // because reading and testing the synchronization state is not 931 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A, 932 // in _thread_in_native state, loads _not_synchronized and is 933 // preempted. VM thread changes sync state to synchronizing and 934 // suspends threads for GC. Thread A is resumed to finish this 935 // native method, but doesn't block here since it didn't see any 936 // synchronization in progress, and escapes. 937 938 // We use release_store_fence to update values like the thread state, where 939 // we don't want the current thread to continue until all our prior memory 940 // accesses (including the new thread state) are visible to other threads. 941 __ li(R0/*thread_state*/, _thread_in_native_trans); 942 __ release(); 943 __ stw(R0/*thread_state*/, thread_(thread_state)); 944 if (UseMembar) { 945 __ fence(); 946 } 947 // Write serialization page so that the VM thread can do a pseudo remote 948 // membar. We use the current thread pointer to calculate a thread 949 // specific offset to write to within the page. This minimizes bus 950 // traffic due to cache line collision. 951 else { 952 __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2); 953 } 954 955 // Now before we return to java we must look for a current safepoint 956 // (a new safepoint can not start since we entered native_trans). 957 // We must check here because a current safepoint could be modifying 958 // the callers registers right this moment. 959 960 // Acquire isn't strictly necessary here because of the fence, but 961 // sync_state is declared to be volatile, so we do it anyway 962 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path). 963 int sync_state_offs = __ load_const_optimized(sync_state_addr, SafepointSynchronize::address_of_state(), /*temp*/R0, true); 964 965 // TODO PPC port assert(4 == SafepointSynchronize::sz_state(), "unexpected field size"); 966 __ lwz(sync_state, sync_state_offs, sync_state_addr); 967 968 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size"); 969 __ lwz(suspend_flags, thread_(suspend_flags)); 970 971 Label sync_check_done; 972 Label do_safepoint; 973 // No synchronization in progress nor yet synchronized. 974 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized); 975 // Not suspended. 976 __ cmpwi(CCR1, suspend_flags, 0); 977 978 __ bne(CCR0, do_safepoint); 979 __ beq(CCR1, sync_check_done); 980 __ bind(do_safepoint); 981 __ isync(); 982 // Block. We do the call directly and leave the current 983 // last_Java_frame setup undisturbed. We must save any possible 984 // native result across the call. No oop is present. 985 986 __ mr(R3_ARG1, R16_thread); 987 #if defined(ABI_ELFv2) 988 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), 989 relocInfo::none); 990 #else 991 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans), 992 relocInfo::none); 993 #endif 994 995 __ bind(sync_check_done); 996 997 //============================================================================= 998 // <<<<<< Back in Interpreter Frame >>>>> 999 1000 // We are in thread_in_native_trans here and back in the normal 1001 // interpreter frame. We don't have to do anything special about 1002 // safepoints and we can switch to Java mode anytime we are ready. 1003 1004 // Note: frame::interpreter_frame_result has a dependency on how the 1005 // method result is saved across the call to post_method_exit. For 1006 // native methods it assumes that the non-FPU/non-void result is 1007 // saved in _native_lresult and a FPU result in _native_fresult. If 1008 // this changes then the interpreter_frame_result implementation 1009 // will need to be updated too. 1010 1011 // On PPC64, we have stored the result directly after the native call. 1012 1013 //============================================================================= 1014 // Back in Java 1015 1016 // We use release_store_fence to update values like the thread state, where 1017 // we don't want the current thread to continue until all our prior memory 1018 // accesses (including the new thread state) are visible to other threads. 1019 __ li(R0/*thread_state*/, _thread_in_Java); 1020 __ release(); 1021 __ stw(R0/*thread_state*/, thread_(thread_state)); 1022 if (UseMembar) { 1023 __ fence(); 1024 } 1025 1026 __ reset_last_Java_frame(); 1027 1028 // Jvmdi/jvmpi support. Whether we've got an exception pending or 1029 // not, and whether unlocking throws an exception or not, we notify 1030 // on native method exit. If we do have an exception, we'll end up 1031 // in the caller's context to handle it, so if we don't do the 1032 // notify here, we'll drop it on the floor. 1033 __ notify_method_exit(true/*native method*/, 1034 ilgl /*illegal state (not used for native methods)*/, 1035 InterpreterMacroAssembler::NotifyJVMTI, 1036 false /*check_exceptions*/); 1037 1038 //============================================================================= 1039 // Handle exceptions 1040 1041 if (synchronized) { 1042 // Don't check for exceptions since we're still in the i2n frame. Do that 1043 // manually afterwards. 1044 unlock_method(false); 1045 } 1046 1047 // Reset active handles after returning from native. 1048 // thread->active_handles()->clear(); 1049 __ ld(active_handles, thread_(active_handles)); 1050 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size"); 1051 __ li(R0, 0); 1052 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles); 1053 1054 Label exception_return_sync_check_already_unlocked; 1055 __ ld(R0/*pending_exception*/, thread_(pending_exception)); 1056 __ cmpdi(CCR0, R0/*pending_exception*/, 0); 1057 __ bne(CCR0, exception_return_sync_check_already_unlocked); 1058 1059 //----------------------------------------------------------------------------- 1060 // No exception pending. 1061 1062 // Move native method result back into proper registers and return. 1063 // Invoke result handler (may unbox/promote). 1064 __ ld(R11_scratch1, 0, R1_SP); 1065 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1); 1066 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); 1067 __ call_stub(result_handler_addr); 1068 1069 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); 1070 1071 // Must use the return pc which was loaded from the caller's frame 1072 // as the VM uses return-pc-patching for deoptimization. 1073 __ mtlr(R0); 1074 __ blr(); 1075 1076 //----------------------------------------------------------------------------- 1077 // An exception is pending. We call into the runtime only if the 1078 // caller was not interpreted. If it was interpreted the 1079 // interpreter will do the correct thing. If it isn't interpreted 1080 // (call stub/compiled code) we will change our return and continue. 1081 1082 __ BIND(exception_return_sync_check); 1083 1084 if (synchronized) { 1085 // Don't check for exceptions since we're still in the i2n frame. Do that 1086 // manually afterwards. 1087 unlock_method(false); 1088 } 1089 __ BIND(exception_return_sync_check_already_unlocked); 1090 1091 const Register return_pc = R31; 1092 1093 __ ld(return_pc, 0, R1_SP); 1094 __ ld(return_pc, _abi(lr), return_pc); 1095 1096 // Get the address of the exception handler. 1097 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), 1098 R16_thread, 1099 return_pc /* return pc */); 1100 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2); 1101 1102 // Load the PC of the the exception handler into LR. 1103 __ mtlr(R3_RET); 1104 1105 // Load exception into R3_ARG1 and clear pending exception in thread. 1106 __ ld(R3_ARG1/*exception*/, thread_(pending_exception)); 1107 __ li(R4_ARG2, 0); 1108 __ std(R4_ARG2, thread_(pending_exception)); 1109 1110 // Load the original return pc into R4_ARG2. 1111 __ mr(R4_ARG2/*issuing_pc*/, return_pc); 1112 1113 // Return to exception handler. 1114 __ blr(); 1115 1116 //============================================================================= 1117 // Counter overflow. 1118 1119 if (inc_counter) { 1120 // Handle invocation counter overflow. 1121 __ bind(invocation_counter_overflow); 1122 1123 generate_counter_overflow(continue_after_compile); 1124 } 1125 1126 return entry; 1127 } 1128 1129 // Generic interpreted method entry to (asm) interpreter. 1130 // 1131 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) { 1132 bool inc_counter = UseCompiler || CountCompiledCalls; 1133 address entry = __ pc(); 1134 // Generate the code to allocate the interpreter stack frame. 1135 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame. 1136 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame. 1137 1138 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals); 1139 1140 #ifdef FAST_DISPATCH 1141 __ unimplemented("Fast dispatch in generate_normal_entry"); 1142 #if 0 1143 __ set((intptr_t)Interpreter::dispatch_table(), IdispatchTables); 1144 // Set bytecode dispatch table base. 1145 #endif 1146 #endif 1147 1148 // -------------------------------------------------------------------------- 1149 // Zero out non-parameter locals. 1150 // Note: *Always* zero out non-parameter locals as Sparc does. It's not 1151 // worth to ask the flag, just do it. 1152 Register Rslot_addr = R6_ARG4, 1153 Rnum = R7_ARG5; 1154 Label Lno_locals, Lzero_loop; 1155 1156 // Set up the zeroing loop. 1157 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals); 1158 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals); 1159 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize); 1160 __ beq(CCR0, Lno_locals); 1161 __ li(R0, 0); 1162 __ mtctr(Rnum); 1163 1164 // The zero locals loop. 1165 __ bind(Lzero_loop); 1166 __ std(R0, 0, Rslot_addr); 1167 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize); 1168 __ bdnz(Lzero_loop); 1169 1170 __ bind(Lno_locals); 1171 1172 // -------------------------------------------------------------------------- 1173 // Counter increment and overflow check. 1174 Label invocation_counter_overflow, 1175 profile_method, 1176 profile_method_continue; 1177 if (inc_counter || ProfileInterpreter) { 1178 1179 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1; 1180 if (synchronized) { 1181 // Since at this point in the method invocation the exception handler 1182 // would try to exit the monitor of synchronized methods which hasn't 1183 // been entered yet, we set the thread local variable 1184 // _do_not_unlock_if_synchronized to true. If any exception was thrown by 1185 // runtime, exception handling i.e. unlock_if_synchronized_method will 1186 // check this thread local flag. 1187 // This flag has two effects, one is to force an unwind in the topmost 1188 // interpreter frame and not perform an unlock while doing so. 1189 __ li(R0, 1); 1190 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1191 } 1192 // Increment invocation counter and check for overflow. 1193 if (inc_counter) { 1194 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue); 1195 } 1196 1197 __ bind(profile_method_continue); 1198 1199 // Reset the _do_not_unlock_if_synchronized flag. 1200 if (synchronized) { 1201 __ li(R0, 0); 1202 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); 1203 } 1204 } 1205 1206 // -------------------------------------------------------------------------- 1207 // Locking of synchronized methods. Must happen AFTER invocation_counter 1208 // check and stack overflow check, so method is not locked if overflows. 1209 if (synchronized) { 1210 lock_method(R3_ARG1, R4_ARG2, R5_ARG3); 1211 } 1212 #ifdef ASSERT 1213 else { 1214 Label Lok; 1215 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method); 1216 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED); 1217 __ asm_assert_eq("method needs synchronization", 0x8521); 1218 __ bind(Lok); 1219 } 1220 #endif // ASSERT 1221 1222 __ verify_thread(); 1223 1224 // -------------------------------------------------------------------------- 1225 // JVMTI support 1226 __ notify_method_entry(); 1227 1228 // -------------------------------------------------------------------------- 1229 // Start executing instructions. 1230 __ dispatch_next(vtos); 1231 1232 // -------------------------------------------------------------------------- 1233 // Out of line counter overflow and MDO creation code. 1234 if (ProfileInterpreter) { 1235 // We have decided to profile this method in the interpreter. 1236 __ bind(profile_method); 1237 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1238 __ set_method_data_pointer_for_bcp(); 1239 __ b(profile_method_continue); 1240 } 1241 1242 if (inc_counter) { 1243 // Handle invocation counter overflow. 1244 __ bind(invocation_counter_overflow); 1245 generate_counter_overflow(profile_method_continue); 1246 } 1247 return entry; 1248 } 1249 1250 // These should never be compiled since the interpreter will prefer 1251 // the compiled version to the intrinsic version. 1252 bool AbstractInterpreter::can_be_compiled(methodHandle m) { 1253 return !math_entry_available(method_kind(m)); 1254 } 1255 1256 // How much stack a method activation needs in stack slots. 1257 // We must calc this exactly like in generate_fixed_frame. 1258 // Note: This returns the conservative size assuming maximum alignment. 1259 int AbstractInterpreter::size_top_interpreter_activation(Method* method) { 1260 const int max_alignment_size = 2; 1261 const int abi_scratch = frame::abi_reg_args_size; 1262 return method->max_locals() + method->max_stack() + 1263 frame::interpreter_frame_monitor_size() + max_alignment_size + abi_scratch; 1264 } 1265 1266 // Returns number of stackElementWords needed for the interpreter frame with the 1267 // given sections. 1268 // This overestimates the stack by one slot in case of alignments. 1269 int AbstractInterpreter::size_activation(int max_stack, 1270 int temps, 1271 int extra_args, 1272 int monitors, 1273 int callee_params, 1274 int callee_locals, 1275 bool is_top_frame) { 1276 // Note: This calculation must exactly parallel the frame setup 1277 // in InterpreterGenerator::generate_fixed_frame. 1278 assert(Interpreter::stackElementWords == 1, "sanity"); 1279 const int max_alignment_space = StackAlignmentInBytes / Interpreter::stackElementSize; 1280 const int abi_scratch = is_top_frame ? (frame::abi_reg_args_size / Interpreter::stackElementSize) : 1281 (frame::abi_minframe_size / Interpreter::stackElementSize); 1282 const int size = 1283 max_stack + 1284 (callee_locals - callee_params) + 1285 monitors * frame::interpreter_frame_monitor_size() + 1286 max_alignment_space + 1287 abi_scratch + 1288 frame::ijava_state_size / Interpreter::stackElementSize; 1289 1290 // Fixed size of an interpreter frame, align to 16-byte. 1291 return (size & -2); 1292 } 1293 1294 // Fills a sceletal interpreter frame generated during deoptimizations. 1295 // 1296 // Parameters: 1297 // 1298 // interpreter_frame != NULL: 1299 // set up the method, locals, and monitors. 1300 // The frame interpreter_frame, if not NULL, is guaranteed to be the 1301 // right size, as determined by a previous call to this method. 1302 // It is also guaranteed to be walkable even though it is in a skeletal state 1303 // 1304 // is_top_frame == true: 1305 // We're processing the *oldest* interpreter frame! 1306 // 1307 // pop_frame_extra_args: 1308 // If this is != 0 we are returning to a deoptimized frame by popping 1309 // off the callee frame. We want to re-execute the call that called the 1310 // callee interpreted, but since the return to the interpreter would pop 1311 // the arguments off advance the esp by dummy popframe_extra_args slots. 1312 // Popping off those will establish the stack layout as it was before the call. 1313 // 1314 void AbstractInterpreter::layout_activation(Method* method, 1315 int tempcount, 1316 int popframe_extra_args, 1317 int moncount, 1318 int caller_actual_parameters, 1319 int callee_param_count, 1320 int callee_locals_count, 1321 frame* caller, 1322 frame* interpreter_frame, 1323 bool is_top_frame, 1324 bool is_bottom_frame) { 1325 1326 const int abi_scratch = is_top_frame ? (frame::abi_reg_args_size / Interpreter::stackElementSize) : 1327 (frame::abi_minframe_size / Interpreter::stackElementSize); 1328 1329 intptr_t* locals_base = (caller->is_interpreted_frame()) ? 1330 caller->interpreter_frame_esp() + caller_actual_parameters : 1331 caller->sp() + method->max_locals() - 1 + (frame::abi_minframe_size / Interpreter::stackElementSize) ; 1332 1333 intptr_t* monitor_base = caller->sp() - frame::ijava_state_size / Interpreter::stackElementSize ; 1334 intptr_t* monitor = monitor_base - (moncount * frame::interpreter_frame_monitor_size()); 1335 intptr_t* esp_base = monitor - 1; 1336 intptr_t* esp = esp_base - tempcount - popframe_extra_args; 1337 intptr_t* sp = (intptr_t *) (((intptr_t) (esp_base - callee_locals_count + callee_param_count - method->max_stack()- abi_scratch)) & -StackAlignmentInBytes); 1338 intptr_t* sender_sp = caller->sp() + (frame::abi_minframe_size - frame::abi_reg_args_size) / Interpreter::stackElementSize; 1339 intptr_t* top_frame_sp = is_top_frame ? sp : sp + (frame::abi_minframe_size - frame::abi_reg_args_size) / Interpreter::stackElementSize; 1340 1341 interpreter_frame->interpreter_frame_set_method(method); 1342 interpreter_frame->interpreter_frame_set_locals(locals_base); 1343 interpreter_frame->interpreter_frame_set_cpcache(method->constants()->cache()); 1344 interpreter_frame->interpreter_frame_set_esp(esp); 1345 interpreter_frame->interpreter_frame_set_monitor_end((BasicObjectLock *)monitor); 1346 interpreter_frame->interpreter_frame_set_top_frame_sp(top_frame_sp); 1347 if (!is_bottom_frame) { 1348 interpreter_frame->interpreter_frame_set_sender_sp(sender_sp); 1349 } 1350 } 1351 1352 // ============================================================================= 1353 // Exceptions 1354 1355 void TemplateInterpreterGenerator::generate_throw_exception() { 1356 Register Rexception = R17_tos, 1357 Rcontinuation = R3_RET; 1358 1359 // -------------------------------------------------------------------------- 1360 // Entry point if an method returns with a pending exception (rethrow). 1361 Interpreter::_rethrow_exception_entry = __ pc(); 1362 { 1363 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp. 1364 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 1365 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 1366 1367 // Compiled code destroys templateTableBase, reload. 1368 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1); 1369 } 1370 1371 // Entry point if a interpreted method throws an exception (throw). 1372 Interpreter::_throw_exception_entry = __ pc(); 1373 { 1374 __ mr(Rexception, R3_RET); 1375 1376 __ verify_thread(); 1377 __ verify_oop(Rexception); 1378 1379 // Expression stack must be empty before entering the VM in case of an exception. 1380 __ empty_expression_stack(); 1381 // Find exception handler address and preserve exception oop. 1382 // Call C routine to find handler and jump to it. 1383 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception); 1384 __ mtctr(Rcontinuation); 1385 // Push exception for exception handler bytecodes. 1386 __ push_ptr(Rexception); 1387 1388 // Jump to exception handler (may be remove activation entry!). 1389 __ bctr(); 1390 } 1391 1392 // If the exception is not handled in the current frame the frame is 1393 // removed and the exception is rethrown (i.e. exception 1394 // continuation is _rethrow_exception). 1395 // 1396 // Note: At this point the bci is still the bxi for the instruction 1397 // which caused the exception and the expression stack is 1398 // empty. Thus, for any VM calls at this point, GC will find a legal 1399 // oop map (with empty expression stack). 1400 1401 // In current activation 1402 // tos: exception 1403 // bcp: exception bcp 1404 1405 // -------------------------------------------------------------------------- 1406 // JVMTI PopFrame support 1407 1408 Interpreter::_remove_activation_preserving_args_entry = __ pc(); 1409 { 1410 // Set the popframe_processing bit in popframe_condition indicating that we are 1411 // currently handling popframe, so that call_VMs that may happen later do not 1412 // trigger new popframe handling cycles. 1413 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 1414 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit); 1415 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 1416 1417 // Empty the expression stack, as in normal exception handling. 1418 __ empty_expression_stack(); 1419 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false); 1420 1421 // Check to see whether we are returning to a deoptimized frame. 1422 // (The PopFrame call ensures that the caller of the popped frame is 1423 // either interpreted or compiled and deoptimizes it if compiled.) 1424 // Note that we don't compare the return PC against the 1425 // deoptimization blob's unpack entry because of the presence of 1426 // adapter frames in C2. 1427 Label Lcaller_not_deoptimized; 1428 Register return_pc = R3_ARG1; 1429 __ ld(return_pc, 0, R1_SP); 1430 __ ld(return_pc, _abi(lr), return_pc); 1431 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc); 1432 __ cmpdi(CCR0, R3_RET, 0); 1433 __ bne(CCR0, Lcaller_not_deoptimized); 1434 1435 // The deoptimized case. 1436 // In this case, we can't call dispatch_next() after the frame is 1437 // popped, but instead must save the incoming arguments and restore 1438 // them after deoptimization has occurred. 1439 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method); 1440 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2); 1441 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize); 1442 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize); 1443 __ subf(R5_ARG3, R4_ARG2, R5_ARG3); 1444 // Save these arguments. 1445 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3); 1446 1447 // Inform deoptimization that it is responsible for restoring these arguments. 1448 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit); 1449 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 1450 1451 // Return from the current method into the deoptimization blob. Will eventually 1452 // end up in the deopt interpeter entry, deoptimization prepared everything that 1453 // we will reexecute the call that called us. 1454 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2); 1455 __ mtlr(return_pc); 1456 __ blr(); 1457 1458 // The non-deoptimized case. 1459 __ bind(Lcaller_not_deoptimized); 1460 1461 // Clear the popframe condition flag. 1462 __ li(R0, 0); 1463 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); 1464 1465 // Get out of the current method and re-execute the call that called us. 1466 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); 1467 __ restore_interpreter_state(R11_scratch1); 1468 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); 1469 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); 1470 if (ProfileInterpreter) { 1471 __ set_method_data_pointer_for_bcp(); 1472 } 1473 #if INCLUDE_JVMTI 1474 Label L_done; 1475 1476 __ lbz(R11_scratch1, 0, R14_bcp); 1477 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic); 1478 __ bne(CCR0, L_done); 1479 1480 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. 1481 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL. 1482 __ ld(R4_ARG2, 0, R18_locals); 1483 __ call_VM(R11_scratch1, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), 1484 R4_ARG2, R19_method, R14_bcp); 1485 1486 __ cmpdi(CCR0, R11_scratch1, 0); 1487 __ beq(CCR0, L_done); 1488 1489 __ std(R11_scratch1, wordSize, R15_esp); 1490 __ bind(L_done); 1491 #endif // INCLUDE_JVMTI 1492 __ dispatch_next(vtos); 1493 } 1494 // end of JVMTI PopFrame support 1495 1496 // -------------------------------------------------------------------------- 1497 // Remove activation exception entry. 1498 // This is jumped to if an interpreted method can't handle an exception itself 1499 // (we come from the throw/rethrow exception entry above). We're going to call 1500 // into the VM to find the exception handler in the caller, pop the current 1501 // frame and return the handler we calculated. 1502 Interpreter::_remove_activation_entry = __ pc(); 1503 { 1504 __ pop_ptr(Rexception); 1505 __ verify_thread(); 1506 __ verify_oop(Rexception); 1507 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread); 1508 1509 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true); 1510 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false); 1511 1512 __ get_vm_result(Rexception); 1513 1514 // We are done with this activation frame; find out where to go next. 1515 // The continuation point will be an exception handler, which expects 1516 // the following registers set up: 1517 // 1518 // RET: exception oop 1519 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled. 1520 1521 Register return_pc = R31; // Needs to survive the runtime call. 1522 __ ld(return_pc, 0, R1_SP); 1523 __ ld(return_pc, _abi(lr), return_pc); 1524 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc); 1525 1526 // Remove the current activation. 1527 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); 1528 1529 __ mr(R4_ARG2, return_pc); 1530 __ mtlr(R3_RET); 1531 __ mr(R3_RET, Rexception); 1532 __ blr(); 1533 } 1534 } 1535 1536 // JVMTI ForceEarlyReturn support. 1537 // Returns "in the middle" of a method with a "fake" return value. 1538 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { 1539 1540 Register Rscratch1 = R11_scratch1, 1541 Rscratch2 = R12_scratch2; 1542 1543 address entry = __ pc(); 1544 __ empty_expression_stack(); 1545 1546 __ load_earlyret_value(state, Rscratch1); 1547 1548 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); 1549 // Clear the earlyret state. 1550 __ li(R0, 0); 1551 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1); 1552 1553 __ remove_activation(state, false, false); 1554 // Copied from TemplateTable::_return. 1555 // Restoration of lr done by remove_activation. 1556 switch (state) { 1557 case ltos: 1558 case btos: 1559 case ctos: 1560 case stos: 1561 case atos: 1562 case itos: __ mr(R3_RET, R17_tos); break; 1563 case ftos: 1564 case dtos: __ fmr(F1_RET, F15_ftos); break; 1565 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need 1566 // to get visible before the reference to the object gets stored anywhere. 1567 __ membar(Assembler::StoreStore); break; 1568 default : ShouldNotReachHere(); 1569 } 1570 __ blr(); 1571 1572 return entry; 1573 } // end of ForceEarlyReturn support 1574 1575 //----------------------------------------------------------------------------- 1576 // Helper for vtos entry point generation 1577 1578 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, 1579 address& bep, 1580 address& cep, 1581 address& sep, 1582 address& aep, 1583 address& iep, 1584 address& lep, 1585 address& fep, 1586 address& dep, 1587 address& vep) { 1588 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); 1589 Label L; 1590 1591 aep = __ pc(); __ push_ptr(); __ b(L); 1592 fep = __ pc(); __ push_f(); __ b(L); 1593 dep = __ pc(); __ push_d(); __ b(L); 1594 lep = __ pc(); __ push_l(); __ b(L); 1595 __ align(32, 12, 24); // align L 1596 bep = cep = sep = 1597 iep = __ pc(); __ push_i(); 1598 vep = __ pc(); 1599 __ bind(L); 1600 generate_and_dispatch(t); 1601 } 1602 1603 //----------------------------------------------------------------------------- 1604 // Generation of individual instructions 1605 1606 // helpers for generate_and_dispatch 1607 1608 InterpreterGenerator::InterpreterGenerator(StubQueue* code) 1609 : TemplateInterpreterGenerator(code) { 1610 generate_all(); // Down here so it can be "virtual". 1611 } 1612 1613 //----------------------------------------------------------------------------- 1614 1615 // Non-product code 1616 #ifndef PRODUCT 1617 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { 1618 //__ flush_bundle(); 1619 address entry = __ pc(); 1620 1621 const char *bname = NULL; 1622 uint tsize = 0; 1623 switch(state) { 1624 case ftos: 1625 bname = "trace_code_ftos {"; 1626 tsize = 2; 1627 break; 1628 case btos: 1629 bname = "trace_code_btos {"; 1630 tsize = 2; 1631 break; 1632 case ctos: 1633 bname = "trace_code_ctos {"; 1634 tsize = 2; 1635 break; 1636 case stos: 1637 bname = "trace_code_stos {"; 1638 tsize = 2; 1639 break; 1640 case itos: 1641 bname = "trace_code_itos {"; 1642 tsize = 2; 1643 break; 1644 case ltos: 1645 bname = "trace_code_ltos {"; 1646 tsize = 3; 1647 break; 1648 case atos: 1649 bname = "trace_code_atos {"; 1650 tsize = 2; 1651 break; 1652 case vtos: 1653 // Note: In case of vtos, the topmost of stack value could be a int or doubl 1654 // In case of a double (2 slots) we won't see the 2nd stack value. 1655 // Maybe we simply should print the topmost 3 stack slots to cope with the problem. 1656 bname = "trace_code_vtos {"; 1657 tsize = 2; 1658 1659 break; 1660 case dtos: 1661 bname = "trace_code_dtos {"; 1662 tsize = 3; 1663 break; 1664 default: 1665 ShouldNotReachHere(); 1666 } 1667 BLOCK_COMMENT(bname); 1668 1669 // Support short-cut for TraceBytecodesAt. 1670 // Don't call into the VM if we don't want to trace to speed up things. 1671 Label Lskip_vm_call; 1672 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { 1673 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true); 1674 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); 1675 __ ld(R11_scratch1, offs1, R11_scratch1); 1676 __ lwa(R12_scratch2, offs2, R12_scratch2); 1677 __ cmpd(CCR0, R12_scratch2, R11_scratch1); 1678 __ blt(CCR0, Lskip_vm_call); 1679 } 1680 1681 __ push(state); 1682 // Load 2 topmost expression stack values. 1683 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp); 1684 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp); 1685 __ mflr(R31); 1686 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false); 1687 __ mtlr(R31); 1688 __ pop(state); 1689 1690 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { 1691 __ bind(Lskip_vm_call); 1692 } 1693 __ blr(); 1694 BLOCK_COMMENT("} trace_code"); 1695 return entry; 1696 } 1697 1698 void TemplateInterpreterGenerator::count_bytecode() { 1699 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true); 1700 __ lwz(R12_scratch2, offs, R11_scratch1); 1701 __ addi(R12_scratch2, R12_scratch2, 1); 1702 __ stw(R12_scratch2, offs, R11_scratch1); 1703 } 1704 1705 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { 1706 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true); 1707 __ lwz(R12_scratch2, offs, R11_scratch1); 1708 __ addi(R12_scratch2, R12_scratch2, 1); 1709 __ stw(R12_scratch2, offs, R11_scratch1); 1710 } 1711 1712 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { 1713 const Register addr = R11_scratch1, 1714 tmp = R12_scratch2; 1715 // Get index, shift out old bytecode, bring in new bytecode, and store it. 1716 // _index = (_index >> log2_number_of_codes) | 1717 // (bytecode << log2_number_of_codes); 1718 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true); 1719 __ lwz(tmp, offs1, addr); 1720 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes); 1721 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); 1722 __ stw(tmp, offs1, addr); 1723 1724 // Bump bucket contents. 1725 // _counters[_index] ++; 1726 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true); 1727 __ sldi(tmp, tmp, LogBytesPerInt); 1728 __ add(addr, tmp, addr); 1729 __ lwz(tmp, offs2, addr); 1730 __ addi(tmp, tmp, 1); 1731 __ stw(tmp, offs2, addr); 1732 } 1733 1734 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { 1735 // Call a little run-time stub to avoid blow-up for each bytecode. 1736 // The run-time runtime saves the right registers, depending on 1737 // the tosca in-state for the given template. 1738 1739 assert(Interpreter::trace_code(t->tos_in()) != NULL, 1740 "entry must have been generated"); 1741 1742 // Note: we destroy LR here. 1743 __ bl(Interpreter::trace_code(t->tos_in())); 1744 } 1745 1746 void TemplateInterpreterGenerator::stop_interpreter_at() { 1747 Label L; 1748 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true); 1749 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); 1750 __ ld(R11_scratch1, offs1, R11_scratch1); 1751 __ lwa(R12_scratch2, offs2, R12_scratch2); 1752 __ cmpd(CCR0, R12_scratch2, R11_scratch1); 1753 __ bne(CCR0, L); 1754 __ illtrap(); 1755 __ bind(L); 1756 } 1757 1758 #endif // !PRODUCT 1759 #endif // !CC_INTERP