| 1 | /* |
| 2 | * Copyright (c) 1997, 2018, 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 "classfile/vmSymbols.hpp" |
| 27 | #include "code/vmreg.inline.hpp" |
| 28 | #include "interpreter/bytecode.hpp" |
| 29 | #include "interpreter/interpreter.hpp" |
| 30 | #include "memory/allocation.inline.hpp" |
| 31 | #include "memory/resourceArea.hpp" |
| 32 | #include "oops/methodData.hpp" |
| 33 | #include "oops/oop.inline.hpp" |
| 34 | #include "prims/jvmtiThreadState.hpp" |
| 35 | #include "runtime/frame.inline.hpp" |
| 36 | #include "runtime/handles.inline.hpp" |
| 37 | #include "runtime/monitorChunk.hpp" |
| 38 | #include "runtime/sharedRuntime.hpp" |
| 39 | #include "runtime/vframe.hpp" |
| 40 | #include "runtime/vframeArray.hpp" |
| 41 | #include "runtime/vframe_hp.hpp" |
| 42 | #include "utilities/copy.hpp" |
| 43 | #include "utilities/events.hpp" |
| 44 | #ifdef COMPILER2 |
| 45 | #include "opto/runtime.hpp" |
| 46 | #endif |
| 47 | |
| 48 | int vframeArrayElement:: bci(void) const { return (_bci == SynchronizationEntryBCI ? 0 : _bci); } |
| 49 | |
| 50 | void vframeArrayElement::free_monitors(JavaThread* jt) { |
| 51 | if (_monitors != NULL) { |
| 52 | MonitorChunk* chunk = _monitors; |
| 53 | _monitors = NULL; |
| 54 | jt->remove_monitor_chunk(chunk); |
| 55 | delete chunk; |
| 56 | } |
| 57 | } |
| 58 | |
| 59 | void vframeArrayElement::fill_in(compiledVFrame* vf, bool realloc_failures) { |
| 60 | |
| 61 | // Copy the information from the compiled vframe to the |
| 62 | // interpreter frame we will be creating to replace vf |
| 63 | |
| 64 | _method = vf->method(); |
| 65 | _bci = vf->raw_bci(); |
| 66 | _reexecute = vf->should_reexecute(); |
| 67 | #ifdef ASSERT |
| 68 | _removed_monitors = false; |
| 69 | #endif |
| 70 | |
| 71 | int index; |
| 72 | |
| 73 | // Get the monitors off-stack |
| 74 | |
| 75 | GrowableArray<MonitorInfo*>* list = vf->monitors(); |
| 76 | if (list->is_empty()) { |
| 77 | _monitors = NULL; |
| 78 | } else { |
| 79 | |
| 80 | // Allocate monitor chunk |
| 81 | _monitors = new MonitorChunk(list->length()); |
| 82 | vf->thread()->add_monitor_chunk(_monitors); |
| 83 | |
| 84 | // Migrate the BasicLocks from the stack to the monitor chunk |
| 85 | for (index = 0; index < list->length(); index++) { |
| 86 | MonitorInfo* monitor = list->at(index); |
| 87 | assert(!monitor->owner_is_scalar_replaced() || realloc_failures, "object should be reallocated already" ); |
| 88 | BasicObjectLock* dest = _monitors->at(index); |
| 89 | if (monitor->owner_is_scalar_replaced()) { |
| 90 | dest->set_obj(NULL); |
| 91 | } else { |
| 92 | assert(monitor->owner() == NULL || (!monitor->owner()->is_unlocked() && !monitor->owner()->has_bias_pattern()), "object must be null or locked, and unbiased" ); |
| 93 | dest->set_obj(monitor->owner()); |
| 94 | monitor->lock()->move_to(monitor->owner(), dest->lock()); |
| 95 | } |
| 96 | } |
| 97 | } |
| 98 | |
| 99 | // Convert the vframe locals and expressions to off stack |
| 100 | // values. Because we will not gc all oops can be converted to |
| 101 | // intptr_t (i.e. a stack slot) and we are fine. This is |
| 102 | // good since we are inside a HandleMark and the oops in our |
| 103 | // collection would go away between packing them here and |
| 104 | // unpacking them in unpack_on_stack. |
| 105 | |
| 106 | // First the locals go off-stack |
| 107 | |
| 108 | // FIXME this seems silly it creates a StackValueCollection |
| 109 | // in order to get the size to then copy them and |
| 110 | // convert the types to intptr_t size slots. Seems like it |
| 111 | // could do it in place... Still uses less memory than the |
| 112 | // old way though |
| 113 | |
| 114 | StackValueCollection *locs = vf->locals(); |
| 115 | _locals = new StackValueCollection(locs->size()); |
| 116 | for(index = 0; index < locs->size(); index++) { |
| 117 | StackValue* value = locs->at(index); |
| 118 | switch(value->type()) { |
| 119 | case T_OBJECT: |
| 120 | assert(!value->obj_is_scalar_replaced() || realloc_failures, "object should be reallocated already" ); |
| 121 | // preserve object type |
| 122 | _locals->add( new StackValue(cast_from_oop<intptr_t>((value->get_obj()())), T_OBJECT )); |
| 123 | break; |
| 124 | case T_CONFLICT: |
| 125 | // A dead local. Will be initialized to null/zero. |
| 126 | _locals->add( new StackValue()); |
| 127 | break; |
| 128 | case T_INT: |
| 129 | _locals->add( new StackValue(value->get_int())); |
| 130 | break; |
| 131 | default: |
| 132 | ShouldNotReachHere(); |
| 133 | } |
| 134 | } |
| 135 | |
| 136 | // Now the expressions off-stack |
| 137 | // Same silliness as above |
| 138 | |
| 139 | StackValueCollection *exprs = vf->expressions(); |
| 140 | _expressions = new StackValueCollection(exprs->size()); |
| 141 | for(index = 0; index < exprs->size(); index++) { |
| 142 | StackValue* value = exprs->at(index); |
| 143 | switch(value->type()) { |
| 144 | case T_OBJECT: |
| 145 | assert(!value->obj_is_scalar_replaced() || realloc_failures, "object should be reallocated already" ); |
| 146 | // preserve object type |
| 147 | _expressions->add( new StackValue(cast_from_oop<intptr_t>((value->get_obj()())), T_OBJECT )); |
| 148 | break; |
| 149 | case T_CONFLICT: |
| 150 | // A dead stack element. Will be initialized to null/zero. |
| 151 | // This can occur when the compiler emits a state in which stack |
| 152 | // elements are known to be dead (because of an imminent exception). |
| 153 | _expressions->add( new StackValue()); |
| 154 | break; |
| 155 | case T_INT: |
| 156 | _expressions->add( new StackValue(value->get_int())); |
| 157 | break; |
| 158 | default: |
| 159 | ShouldNotReachHere(); |
| 160 | } |
| 161 | } |
| 162 | } |
| 163 | |
| 164 | int unpack_counter = 0; |
| 165 | |
| 166 | void vframeArrayElement::unpack_on_stack(int caller_actual_parameters, |
| 167 | int callee_parameters, |
| 168 | int callee_locals, |
| 169 | frame* caller, |
| 170 | bool is_top_frame, |
| 171 | bool is_bottom_frame, |
| 172 | int exec_mode) { |
| 173 | JavaThread* thread = (JavaThread*) Thread::current(); |
| 174 | |
| 175 | bool realloc_failure_exception = thread->frames_to_pop_failed_realloc() > 0; |
| 176 | |
| 177 | // Look at bci and decide on bcp and continuation pc |
| 178 | address bcp; |
| 179 | // C++ interpreter doesn't need a pc since it will figure out what to do when it |
| 180 | // begins execution |
| 181 | address pc; |
| 182 | bool use_next_mdp = false; // true if we should use the mdp associated with the next bci |
| 183 | // rather than the one associated with bcp |
| 184 | if (raw_bci() == SynchronizationEntryBCI) { |
| 185 | // We are deoptimizing while hanging in prologue code for synchronized method |
| 186 | bcp = method()->bcp_from(0); // first byte code |
| 187 | pc = Interpreter::deopt_entry(vtos, 0); // step = 0 since we don't skip current bytecode |
| 188 | } else if (should_reexecute()) { //reexecute this bytecode |
| 189 | assert(is_top_frame, "reexecute allowed only for the top frame" ); |
| 190 | bcp = method()->bcp_from(bci()); |
| 191 | pc = Interpreter::deopt_reexecute_entry(method(), bcp); |
| 192 | } else { |
| 193 | bcp = method()->bcp_from(bci()); |
| 194 | pc = Interpreter::deopt_continue_after_entry(method(), bcp, callee_parameters, is_top_frame); |
| 195 | use_next_mdp = true; |
| 196 | } |
| 197 | assert(Bytecodes::is_defined(*bcp), "must be a valid bytecode" ); |
| 198 | |
| 199 | // Monitorenter and pending exceptions: |
| 200 | // |
| 201 | // For Compiler2, there should be no pending exception when deoptimizing at monitorenter |
| 202 | // because there is no safepoint at the null pointer check (it is either handled explicitly |
| 203 | // or prior to the monitorenter) and asynchronous exceptions are not made "pending" by the |
| 204 | // runtime interface for the slow case (see JRT_ENTRY_FOR_MONITORENTER). If an asynchronous |
| 205 | // exception was processed, the bytecode pointer would have to be extended one bytecode beyond |
| 206 | // the monitorenter to place it in the proper exception range. |
| 207 | // |
| 208 | // For Compiler1, deoptimization can occur while throwing a NullPointerException at monitorenter, |
| 209 | // in which case bcp should point to the monitorenter since it is within the exception's range. |
| 210 | // |
| 211 | // For realloc failure exception we just pop frames, skip the guarantee. |
| 212 | |
| 213 | assert(*bcp != Bytecodes::_monitorenter || is_top_frame, "a _monitorenter must be a top frame" ); |
| 214 | assert(thread->deopt_compiled_method() != NULL, "compiled method should be known" ); |
| 215 | guarantee(realloc_failure_exception || !(thread->deopt_compiled_method()->is_compiled_by_c2() && |
| 216 | *bcp == Bytecodes::_monitorenter && |
| 217 | exec_mode == Deoptimization::Unpack_exception), |
| 218 | "shouldn't get exception during monitorenter" ); |
| 219 | |
| 220 | int popframe_preserved_args_size_in_bytes = 0; |
| 221 | int popframe_preserved_args_size_in_words = 0; |
| 222 | if (is_top_frame) { |
| 223 | JvmtiThreadState *state = thread->jvmti_thread_state(); |
| 224 | if (JvmtiExport::can_pop_frame() && |
| 225 | (thread->has_pending_popframe() || thread->popframe_forcing_deopt_reexecution())) { |
| 226 | if (thread->has_pending_popframe()) { |
| 227 | // Pop top frame after deoptimization |
| 228 | #ifndef CC_INTERP |
| 229 | pc = Interpreter::remove_activation_preserving_args_entry(); |
| 230 | #else |
| 231 | // Do an uncommon trap type entry. c++ interpreter will know |
| 232 | // to pop frame and preserve the args |
| 233 | pc = Interpreter::deopt_entry(vtos, 0); |
| 234 | use_next_mdp = false; |
| 235 | #endif |
| 236 | } else { |
| 237 | // Reexecute invoke in top frame |
| 238 | pc = Interpreter::deopt_entry(vtos, 0); |
| 239 | use_next_mdp = false; |
| 240 | popframe_preserved_args_size_in_bytes = in_bytes(thread->popframe_preserved_args_size()); |
| 241 | // Note: the PopFrame-related extension of the expression stack size is done in |
| 242 | // Deoptimization::fetch_unroll_info_helper |
| 243 | popframe_preserved_args_size_in_words = in_words(thread->popframe_preserved_args_size_in_words()); |
| 244 | } |
| 245 | } else if (!realloc_failure_exception && JvmtiExport::can_force_early_return() && state != NULL && state->is_earlyret_pending()) { |
| 246 | // Force early return from top frame after deoptimization |
| 247 | #ifndef CC_INTERP |
| 248 | pc = Interpreter::remove_activation_early_entry(state->earlyret_tos()); |
| 249 | #endif |
| 250 | } else { |
| 251 | if (realloc_failure_exception && JvmtiExport::can_force_early_return() && state != NULL && state->is_earlyret_pending()) { |
| 252 | state->clr_earlyret_pending(); |
| 253 | state->set_earlyret_oop(NULL); |
| 254 | state->clr_earlyret_value(); |
| 255 | } |
| 256 | // Possibly override the previous pc computation of the top (youngest) frame |
| 257 | switch (exec_mode) { |
| 258 | case Deoptimization::Unpack_deopt: |
| 259 | // use what we've got |
| 260 | break; |
| 261 | case Deoptimization::Unpack_exception: |
| 262 | // exception is pending |
| 263 | pc = SharedRuntime::raw_exception_handler_for_return_address(thread, pc); |
| 264 | // [phh] We're going to end up in some handler or other, so it doesn't |
| 265 | // matter what mdp we point to. See exception_handler_for_exception() |
| 266 | // in interpreterRuntime.cpp. |
| 267 | break; |
| 268 | case Deoptimization::Unpack_uncommon_trap: |
| 269 | case Deoptimization::Unpack_reexecute: |
| 270 | // redo last byte code |
| 271 | pc = Interpreter::deopt_entry(vtos, 0); |
| 272 | use_next_mdp = false; |
| 273 | break; |
| 274 | default: |
| 275 | ShouldNotReachHere(); |
| 276 | } |
| 277 | } |
| 278 | } |
| 279 | |
| 280 | // Setup the interpreter frame |
| 281 | |
| 282 | assert(method() != NULL, "method must exist" ); |
| 283 | int temps = expressions()->size(); |
| 284 | |
| 285 | int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors(); |
| 286 | |
| 287 | Interpreter::layout_activation(method(), |
| 288 | temps + callee_parameters, |
| 289 | popframe_preserved_args_size_in_words, |
| 290 | locks, |
| 291 | caller_actual_parameters, |
| 292 | callee_parameters, |
| 293 | callee_locals, |
| 294 | caller, |
| 295 | iframe(), |
| 296 | is_top_frame, |
| 297 | is_bottom_frame); |
| 298 | |
| 299 | // Update the pc in the frame object and overwrite the temporary pc |
| 300 | // we placed in the skeletal frame now that we finally know the |
| 301 | // exact interpreter address we should use. |
| 302 | |
| 303 | _frame.patch_pc(thread, pc); |
| 304 | |
| 305 | assert (!method()->is_synchronized() || locks > 0 || _removed_monitors || raw_bci() == SynchronizationEntryBCI, "synchronized methods must have monitors" ); |
| 306 | |
| 307 | BasicObjectLock* top = iframe()->interpreter_frame_monitor_begin(); |
| 308 | for (int index = 0; index < locks; index++) { |
| 309 | top = iframe()->previous_monitor_in_interpreter_frame(top); |
| 310 | BasicObjectLock* src = _monitors->at(index); |
| 311 | top->set_obj(src->obj()); |
| 312 | src->lock()->move_to(src->obj(), top->lock()); |
| 313 | } |
| 314 | if (ProfileInterpreter) { |
| 315 | iframe()->interpreter_frame_set_mdp(0); // clear out the mdp. |
| 316 | } |
| 317 | iframe()->interpreter_frame_set_bcp(bcp); |
| 318 | if (ProfileInterpreter) { |
| 319 | MethodData* mdo = method()->method_data(); |
| 320 | if (mdo != NULL) { |
| 321 | int bci = iframe()->interpreter_frame_bci(); |
| 322 | if (use_next_mdp) ++bci; |
| 323 | address mdp = mdo->bci_to_dp(bci); |
| 324 | iframe()->interpreter_frame_set_mdp(mdp); |
| 325 | } |
| 326 | } |
| 327 | |
| 328 | if (PrintDeoptimizationDetails) { |
| 329 | tty->print_cr("Expressions size: %d" , expressions()->size()); |
| 330 | } |
| 331 | |
| 332 | // Unpack expression stack |
| 333 | // If this is an intermediate frame (i.e. not top frame) then this |
| 334 | // only unpacks the part of the expression stack not used by callee |
| 335 | // as parameters. The callee parameters are unpacked as part of the |
| 336 | // callee locals. |
| 337 | int i; |
| 338 | for(i = 0; i < expressions()->size(); i++) { |
| 339 | StackValue *value = expressions()->at(i); |
| 340 | intptr_t* addr = iframe()->interpreter_frame_expression_stack_at(i); |
| 341 | switch(value->type()) { |
| 342 | case T_INT: |
| 343 | *addr = value->get_int(); |
| 344 | #ifndef PRODUCT |
| 345 | if (PrintDeoptimizationDetails) { |
| 346 | tty->print_cr("Reconstructed expression %d (INT): %d" , i, (int)(*addr)); |
| 347 | } |
| 348 | #endif |
| 349 | break; |
| 350 | case T_OBJECT: |
| 351 | *addr = value->get_int(T_OBJECT); |
| 352 | #ifndef PRODUCT |
| 353 | if (PrintDeoptimizationDetails) { |
| 354 | tty->print("Reconstructed expression %d (OBJECT): " , i); |
| 355 | oop o = (oop)(address)(*addr); |
| 356 | if (o == NULL) { |
| 357 | tty->print_cr("NULL" ); |
| 358 | } else { |
| 359 | ResourceMark rm; |
| 360 | tty->print_raw_cr(o->klass()->name()->as_C_string()); |
| 361 | } |
| 362 | } |
| 363 | #endif |
| 364 | break; |
| 365 | case T_CONFLICT: |
| 366 | // A dead stack slot. Initialize to null in case it is an oop. |
| 367 | *addr = NULL_WORD; |
| 368 | break; |
| 369 | default: |
| 370 | ShouldNotReachHere(); |
| 371 | } |
| 372 | } |
| 373 | |
| 374 | |
| 375 | // Unpack the locals |
| 376 | for(i = 0; i < locals()->size(); i++) { |
| 377 | StackValue *value = locals()->at(i); |
| 378 | intptr_t* addr = iframe()->interpreter_frame_local_at(i); |
| 379 | switch(value->type()) { |
| 380 | case T_INT: |
| 381 | *addr = value->get_int(); |
| 382 | #ifndef PRODUCT |
| 383 | if (PrintDeoptimizationDetails) { |
| 384 | tty->print_cr("Reconstructed local %d (INT): %d" , i, (int)(*addr)); |
| 385 | } |
| 386 | #endif |
| 387 | break; |
| 388 | case T_OBJECT: |
| 389 | *addr = value->get_int(T_OBJECT); |
| 390 | #ifndef PRODUCT |
| 391 | if (PrintDeoptimizationDetails) { |
| 392 | tty->print("Reconstructed local %d (OBJECT): " , i); |
| 393 | oop o = (oop)(address)(*addr); |
| 394 | if (o == NULL) { |
| 395 | tty->print_cr("NULL" ); |
| 396 | } else { |
| 397 | ResourceMark rm; |
| 398 | tty->print_raw_cr(o->klass()->name()->as_C_string()); |
| 399 | } |
| 400 | } |
| 401 | #endif |
| 402 | break; |
| 403 | case T_CONFLICT: |
| 404 | // A dead location. If it is an oop then we need a NULL to prevent GC from following it |
| 405 | *addr = NULL_WORD; |
| 406 | break; |
| 407 | default: |
| 408 | ShouldNotReachHere(); |
| 409 | } |
| 410 | } |
| 411 | |
| 412 | if (is_top_frame && JvmtiExport::can_pop_frame() && thread->popframe_forcing_deopt_reexecution()) { |
| 413 | // An interpreted frame was popped but it returns to a deoptimized |
| 414 | // frame. The incoming arguments to the interpreted activation |
| 415 | // were preserved in thread-local storage by the |
| 416 | // remove_activation_preserving_args_entry in the interpreter; now |
| 417 | // we put them back into the just-unpacked interpreter frame. |
| 418 | // Note that this assumes that the locals arena grows toward lower |
| 419 | // addresses. |
| 420 | if (popframe_preserved_args_size_in_words != 0) { |
| 421 | void* saved_args = thread->popframe_preserved_args(); |
| 422 | assert(saved_args != NULL, "must have been saved by interpreter" ); |
| 423 | #ifdef ASSERT |
| 424 | assert(popframe_preserved_args_size_in_words <= |
| 425 | iframe()->interpreter_frame_expression_stack_size()*Interpreter::stackElementWords, |
| 426 | "expression stack size should have been extended" ); |
| 427 | #endif // ASSERT |
| 428 | int top_element = iframe()->interpreter_frame_expression_stack_size()-1; |
| 429 | intptr_t* base; |
| 430 | if (frame::interpreter_frame_expression_stack_direction() < 0) { |
| 431 | base = iframe()->interpreter_frame_expression_stack_at(top_element); |
| 432 | } else { |
| 433 | base = iframe()->interpreter_frame_expression_stack(); |
| 434 | } |
| 435 | Copy::conjoint_jbytes(saved_args, |
| 436 | base, |
| 437 | popframe_preserved_args_size_in_bytes); |
| 438 | thread->popframe_free_preserved_args(); |
| 439 | } |
| 440 | } |
| 441 | |
| 442 | #ifndef PRODUCT |
| 443 | if (PrintDeoptimizationDetails) { |
| 444 | ttyLocker ttyl; |
| 445 | tty->print_cr("[%d Interpreted Frame]" , ++unpack_counter); |
| 446 | iframe()->print_on(tty); |
| 447 | RegisterMap map(thread); |
| 448 | vframe* f = vframe::new_vframe(iframe(), &map, thread); |
| 449 | f->print(); |
| 450 | |
| 451 | tty->print_cr("locals size %d" , locals()->size()); |
| 452 | tty->print_cr("expression size %d" , expressions()->size()); |
| 453 | |
| 454 | method()->print_value(); |
| 455 | tty->cr(); |
| 456 | // method()->print_codes(); |
| 457 | } else if (TraceDeoptimization) { |
| 458 | tty->print(" " ); |
| 459 | method()->print_value(); |
| 460 | Bytecodes::Code code = Bytecodes::java_code_at(method(), bcp); |
| 461 | int bci = method()->bci_from(bcp); |
| 462 | tty->print(" - %s" , Bytecodes::name(code)); |
| 463 | tty->print(" @ bci %d " , bci); |
| 464 | tty->print_cr("sp = " PTR_FORMAT, p2i(iframe()->sp())); |
| 465 | } |
| 466 | #endif // PRODUCT |
| 467 | |
| 468 | // The expression stack and locals are in the resource area don't leave |
| 469 | // a dangling pointer in the vframeArray we leave around for debug |
| 470 | // purposes |
| 471 | |
| 472 | _locals = _expressions = NULL; |
| 473 | |
| 474 | } |
| 475 | |
| 476 | int vframeArrayElement::on_stack_size(int callee_parameters, |
| 477 | int callee_locals, |
| 478 | bool is_top_frame, |
| 479 | int ) const { |
| 480 | assert(method()->max_locals() == locals()->size(), "just checking" ); |
| 481 | int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors(); |
| 482 | int temps = expressions()->size(); |
| 483 | return Interpreter::size_activation(method()->max_stack(), |
| 484 | temps + callee_parameters, |
| 485 | popframe_extra_stack_expression_els, |
| 486 | locks, |
| 487 | callee_parameters, |
| 488 | callee_locals, |
| 489 | is_top_frame); |
| 490 | } |
| 491 | |
| 492 | |
| 493 | intptr_t* vframeArray::unextended_sp() const { |
| 494 | return _original.unextended_sp(); |
| 495 | } |
| 496 | |
| 497 | vframeArray* vframeArray::allocate(JavaThread* thread, int frame_size, GrowableArray<compiledVFrame*>* chunk, |
| 498 | RegisterMap *reg_map, frame sender, frame caller, frame self, |
| 499 | bool realloc_failures) { |
| 500 | |
| 501 | // Allocate the vframeArray |
| 502 | vframeArray * result = (vframeArray*) AllocateHeap(sizeof(vframeArray) + // fixed part |
| 503 | sizeof(vframeArrayElement) * (chunk->length() - 1), // variable part |
| 504 | mtCompiler); |
| 505 | result->_frames = chunk->length(); |
| 506 | result->_owner_thread = thread; |
| 507 | result->_sender = sender; |
| 508 | result->_caller = caller; |
| 509 | result->_original = self; |
| 510 | result->set_unroll_block(NULL); // initialize it |
| 511 | result->fill_in(thread, frame_size, chunk, reg_map, realloc_failures); |
| 512 | return result; |
| 513 | } |
| 514 | |
| 515 | void vframeArray::fill_in(JavaThread* thread, |
| 516 | int frame_size, |
| 517 | GrowableArray<compiledVFrame*>* chunk, |
| 518 | const RegisterMap *reg_map, |
| 519 | bool realloc_failures) { |
| 520 | // Set owner first, it is used when adding monitor chunks |
| 521 | |
| 522 | _frame_size = frame_size; |
| 523 | for(int i = 0; i < chunk->length(); i++) { |
| 524 | element(i)->fill_in(chunk->at(i), realloc_failures); |
| 525 | } |
| 526 | |
| 527 | // Copy registers for callee-saved registers |
| 528 | if (reg_map != NULL) { |
| 529 | for(int i = 0; i < RegisterMap::reg_count; i++) { |
| 530 | #ifdef AMD64 |
| 531 | // The register map has one entry for every int (32-bit value), so |
| 532 | // 64-bit physical registers have two entries in the map, one for |
| 533 | // each half. Ignore the high halves of 64-bit registers, just like |
| 534 | // frame::oopmapreg_to_location does. |
| 535 | // |
| 536 | // [phh] FIXME: this is a temporary hack! This code *should* work |
| 537 | // correctly w/o this hack, possibly by changing RegisterMap::pd_location |
| 538 | // in frame_amd64.cpp and the values of the phantom high half registers |
| 539 | // in amd64.ad. |
| 540 | // if (VMReg::Name(i) < SharedInfo::stack0 && is_even(i)) { |
| 541 | intptr_t* src = (intptr_t*) reg_map->location(VMRegImpl::as_VMReg(i)); |
| 542 | _callee_registers[i] = src != NULL ? *src : NULL_WORD; |
| 543 | // } else { |
| 544 | // jint* src = (jint*) reg_map->location(VMReg::Name(i)); |
| 545 | // _callee_registers[i] = src != NULL ? *src : NULL_WORD; |
| 546 | // } |
| 547 | #else |
| 548 | jint* src = (jint*) reg_map->location(VMRegImpl::as_VMReg(i)); |
| 549 | _callee_registers[i] = src != NULL ? *src : NULL_WORD; |
| 550 | #endif |
| 551 | if (src == NULL) { |
| 552 | set_location_valid(i, false); |
| 553 | } else { |
| 554 | set_location_valid(i, true); |
| 555 | jint* dst = (jint*) register_location(i); |
| 556 | *dst = *src; |
| 557 | } |
| 558 | } |
| 559 | } |
| 560 | } |
| 561 | |
| 562 | void vframeArray::unpack_to_stack(frame &unpack_frame, int exec_mode, int caller_actual_parameters) { |
| 563 | // stack picture |
| 564 | // unpack_frame |
| 565 | // [new interpreter frames ] (frames are skeletal but walkable) |
| 566 | // caller_frame |
| 567 | // |
| 568 | // This routine fills in the missing data for the skeletal interpreter frames |
| 569 | // in the above picture. |
| 570 | |
| 571 | // Find the skeletal interpreter frames to unpack into |
| 572 | JavaThread* THREAD = JavaThread::current(); |
| 573 | RegisterMap map(THREAD, false); |
| 574 | // Get the youngest frame we will unpack (last to be unpacked) |
| 575 | frame me = unpack_frame.sender(&map); |
| 576 | int index; |
| 577 | for (index = 0; index < frames(); index++ ) { |
| 578 | *element(index)->iframe() = me; |
| 579 | // Get the caller frame (possibly skeletal) |
| 580 | me = me.sender(&map); |
| 581 | } |
| 582 | |
| 583 | // Do the unpacking of interpreter frames; the frame at index 0 represents the top activation, so it has no callee |
| 584 | // Unpack the frames from the oldest (frames() -1) to the youngest (0) |
| 585 | frame* caller_frame = &me; |
| 586 | for (index = frames() - 1; index >= 0 ; index--) { |
| 587 | vframeArrayElement* elem = element(index); // caller |
| 588 | int callee_parameters, callee_locals; |
| 589 | if (index == 0) { |
| 590 | callee_parameters = callee_locals = 0; |
| 591 | } else { |
| 592 | methodHandle caller = elem->method(); |
| 593 | methodHandle callee = element(index - 1)->method(); |
| 594 | Bytecode_invoke inv(caller, elem->bci()); |
| 595 | // invokedynamic instructions don't have a class but obviously don't have a MemberName appendix. |
| 596 | // NOTE: Use machinery here that avoids resolving of any kind. |
| 597 | const bool has_member_arg = |
| 598 | !inv.is_invokedynamic() && MethodHandles::has_member_arg(inv.klass(), inv.name()); |
| 599 | callee_parameters = callee->size_of_parameters() + (has_member_arg ? 1 : 0); |
| 600 | callee_locals = callee->max_locals(); |
| 601 | } |
| 602 | elem->unpack_on_stack(caller_actual_parameters, |
| 603 | callee_parameters, |
| 604 | callee_locals, |
| 605 | caller_frame, |
| 606 | index == 0, |
| 607 | index == frames() - 1, |
| 608 | exec_mode); |
| 609 | if (index == frames() - 1) { |
| 610 | Deoptimization::unwind_callee_save_values(elem->iframe(), this); |
| 611 | } |
| 612 | caller_frame = elem->iframe(); |
| 613 | caller_actual_parameters = callee_parameters; |
| 614 | } |
| 615 | deallocate_monitor_chunks(); |
| 616 | } |
| 617 | |
| 618 | void vframeArray::deallocate_monitor_chunks() { |
| 619 | JavaThread* jt = JavaThread::current(); |
| 620 | for (int index = 0; index < frames(); index++ ) { |
| 621 | element(index)->free_monitors(jt); |
| 622 | } |
| 623 | } |
| 624 | |
| 625 | #ifndef PRODUCT |
| 626 | |
| 627 | bool vframeArray::structural_compare(JavaThread* thread, GrowableArray<compiledVFrame*>* chunk) { |
| 628 | if (owner_thread() != thread) return false; |
| 629 | int index = 0; |
| 630 | #if 0 // FIXME can't do this comparison |
| 631 | |
| 632 | // Compare only within vframe array. |
| 633 | for (deoptimizedVFrame* vf = deoptimizedVFrame::cast(vframe_at(first_index())); vf; vf = vf->deoptimized_sender_or_null()) { |
| 634 | if (index >= chunk->length() || !vf->structural_compare(chunk->at(index))) return false; |
| 635 | index++; |
| 636 | } |
| 637 | if (index != chunk->length()) return false; |
| 638 | #endif |
| 639 | |
| 640 | return true; |
| 641 | } |
| 642 | |
| 643 | #endif |
| 644 | |
| 645 | address vframeArray::register_location(int i) const { |
| 646 | assert(0 <= i && i < RegisterMap::reg_count, "index out of bounds" ); |
| 647 | return (address) & _callee_registers[i]; |
| 648 | } |
| 649 | |
| 650 | |
| 651 | #ifndef PRODUCT |
| 652 | |
| 653 | // Printing |
| 654 | |
| 655 | // Note: we cannot have print_on as const, as we allocate inside the method |
| 656 | void vframeArray::print_on_2(outputStream* st) { |
| 657 | st->print_cr(" - sp: " INTPTR_FORMAT, p2i(sp())); |
| 658 | st->print(" - thread: " ); |
| 659 | Thread::current()->print(); |
| 660 | st->print_cr(" - frame size: %d" , frame_size()); |
| 661 | for (int index = 0; index < frames() ; index++ ) { |
| 662 | element(index)->print(st); |
| 663 | } |
| 664 | } |
| 665 | |
| 666 | void vframeArrayElement::print(outputStream* st) { |
| 667 | st->print_cr(" - interpreter_frame -> sp: " INTPTR_FORMAT, p2i(iframe()->sp())); |
| 668 | } |
| 669 | |
| 670 | void vframeArray::print_value_on(outputStream* st) const { |
| 671 | st->print_cr("vframeArray [%d] " , frames()); |
| 672 | } |
| 673 | |
| 674 | |
| 675 | #endif |
| 676 | |