1 /* 2 * Copyright (c) 1997, 2020, 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 #ifndef SHARE_RUNTIME_DEOPTIMIZATION_HPP 26 #define SHARE_RUNTIME_DEOPTIMIZATION_HPP 27 28 #include "memory/allocation.hpp" 29 #include "runtime/frame.hpp" 30 31 class ProfileData; 32 class vframeArray; 33 class MonitorInfo; 34 class MonitorValue; 35 class ObjectValue; 36 class AutoBoxObjectValue; 37 class ScopeValue; 38 class compiledVFrame; 39 40 template<class E> class GrowableArray; 41 42 class Deoptimization : AllStatic { 43 friend class VMStructs; 44 friend class EscapeBarrier; 45 46 public: 47 // What condition caused the deoptimization? 48 enum DeoptReason { 49 Reason_many = -1, // indicates presence of several reasons 50 Reason_none = 0, // indicates absence of a relevant deopt. 51 // Next 8 reasons are recorded per bytecode in DataLayout::trap_bits. 52 // This is more complicated for JVMCI as JVMCI may deoptimize to *some* bytecode before the 53 // bytecode that actually caused the deopt (with inlining, JVMCI may even deoptimize to a 54 // bytecode in another method): 55 // - bytecode y in method b() causes deopt 56 // - JVMCI deoptimizes to bytecode x in method a() 57 // -> the deopt reason will be recorded for method a() at bytecode x 58 Reason_null_check, // saw unexpected null or zero divisor (@bci) 59 Reason_null_assert, // saw unexpected non-null or non-zero (@bci) 60 Reason_range_check, // saw unexpected array index (@bci) 61 Reason_class_check, // saw unexpected object class (@bci) 62 Reason_array_check, // saw unexpected array class (aastore @bci) 63 Reason_intrinsic, // saw unexpected operand to intrinsic (@bci) 64 Reason_bimorphic, // saw unexpected object class in bimorphic inlining (@bci) 65 66 #if INCLUDE_JVMCI 67 Reason_unreached0 = Reason_null_assert, 68 Reason_type_checked_inlining = Reason_intrinsic, 69 Reason_optimized_type_check = Reason_bimorphic, 70 #endif 71 72 Reason_profile_predicate, // compiler generated predicate moved from frequent branch in a loop failed 73 74 // recorded per method 75 Reason_unloaded, // unloaded class or constant pool entry 76 Reason_uninitialized, // bad class state (uninitialized) 77 Reason_initialized, // class has been fully initialized 78 Reason_unreached, // code is not reached, compiler 79 Reason_unhandled, // arbitrary compiler limitation 80 Reason_constraint, // arbitrary runtime constraint violated 81 Reason_div0_check, // a null_check due to division by zero 82 Reason_age, // nmethod too old; tier threshold reached 83 Reason_predicate, // compiler generated predicate failed 84 Reason_loop_limit_check, // compiler generated loop limits check failed 85 Reason_speculate_class_check, // saw unexpected object class from type speculation 86 Reason_speculate_null_check, // saw unexpected null from type speculation 87 Reason_speculate_null_assert, // saw unexpected null from type speculation 88 Reason_rtm_state_change, // rtm state change detected 89 Reason_unstable_if, // a branch predicted always false was taken 90 Reason_unstable_fused_if, // fused two ifs that had each one untaken branch. One is now taken. 91 #if INCLUDE_JVMCI 92 Reason_aliasing, // optimistic assumption about aliasing failed 93 Reason_transfer_to_interpreter, // explicit transferToInterpreter() 94 Reason_not_compiled_exception_handler, 95 Reason_unresolved, 96 Reason_jsr_mismatch, 97 #endif 98 99 // Reason_tenured is counted separately, add normal counted Reasons above. 100 // Related to MethodData::_trap_hist_limit where Reason_tenured isn't included 101 Reason_tenured, // age of the code has reached the limit 102 Reason_LIMIT, 103 104 // Note: Keep this enum in sync. with _trap_reason_name. 105 Reason_RECORDED_LIMIT = Reason_profile_predicate // some are not recorded per bc 106 // Note: Reason_RECORDED_LIMIT should fit into 31 bits of 107 // DataLayout::trap_bits. This dependency is enforced indirectly 108 // via asserts, to avoid excessive direct header-to-header dependencies. 109 // See Deoptimization::trap_state_reason and class DataLayout. 110 }; 111 112 // What action must be taken by the runtime? 113 enum DeoptAction { 114 Action_none, // just interpret, do not invalidate nmethod 115 Action_maybe_recompile, // recompile the nmethod; need not invalidate 116 Action_reinterpret, // invalidate the nmethod, reset IC, maybe recompile 117 Action_make_not_entrant, // invalidate the nmethod, recompile (probably) 118 Action_make_not_compilable, // invalidate the nmethod and do not compile 119 Action_LIMIT 120 // Note: Keep this enum in sync. with _trap_action_name. 121 }; 122 123 enum { 124 _action_bits = 3, 125 _reason_bits = 5, 126 _debug_id_bits = 23, 127 _action_shift = 0, 128 _reason_shift = _action_shift+_action_bits, 129 _debug_id_shift = _reason_shift+_reason_bits, 130 BC_CASE_LIMIT = PRODUCT_ONLY(1) NOT_PRODUCT(4) // for _deoptimization_hist 131 }; 132 133 enum UnpackType { 134 Unpack_deopt = 0, // normal deoptimization, use pc computed in unpack_vframe_on_stack 135 Unpack_exception = 1, // exception is pending 136 Unpack_uncommon_trap = 2, // redo last byte code (C2 only) 137 Unpack_reexecute = 3, // reexecute bytecode (C1 only) 138 Unpack_none = 4, // not deoptimizing the frame, just reallocating/relocking for JVMTI 139 Unpack_LIMIT = 5 140 }; 141 142 #if INCLUDE_JVMCI 143 // Can reconstruct virtualized unsafe large accesses to byte arrays. 144 static const int _support_large_access_byte_array_virtualization = 1; 145 #endif 146 147 // Make all nmethods that are marked_for_deoptimization not_entrant and deoptimize any live 148 // activations using those nmethods. If an nmethod is passed as an argument then it is 149 // marked_for_deoptimization and made not_entrant. Otherwise a scan of the code cache is done to 150 // find all marked nmethods and they are made not_entrant. 151 static void deoptimize_all_marked(nmethod* nmethod_only = NULL); 152 153 private: 154 // Revoke biased locks at deopt. 155 static void revoke_from_deopt_handler(JavaThread* thread, frame fr, RegisterMap* map); 156 157 static void revoke_for_object_deoptimization(JavaThread* deoptee_thread, frame fr, 158 RegisterMap* map, JavaThread* thread); 159 160 public: 161 // Deoptimizes a frame lazily. Deopt happens on return to the frame. 162 static void deoptimize(JavaThread* thread, frame fr, DeoptReason reason = Reason_constraint); 163 164 #if INCLUDE_JVMCI 165 static address deoptimize_for_missing_exception_handler(CompiledMethod* cm); 166 static oop get_cached_box(AutoBoxObjectValue* bv, frame* fr, RegisterMap* reg_map, TRAPS); 167 #endif 168 169 private: 170 // Does the actual work for deoptimizing a single frame 171 static void deoptimize_single_frame(JavaThread* thread, frame fr, DeoptReason reason); 172 173 #if COMPILER2_OR_JVMCI 174 // Deoptimize objects, that is reallocate and relock them, just before they 175 // escape through JVMTI. The given vframes cover one physical frame. 176 static bool deoptimize_objects_internal(JavaThread* thread, GrowableArray<compiledVFrame*>* chunk, 177 bool& realloc_failures); 178 179 public: 180 181 // Support for restoring non-escaping objects 182 static bool realloc_objects(JavaThread* thread, frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, TRAPS); 183 static void reassign_type_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, typeArrayOop obj, BasicType type); 184 static void reassign_object_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, objArrayOop obj); 185 static void reassign_fields(frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, bool realloc_failures, bool skip_internal); 186 static bool relock_objects(JavaThread* thread, GrowableArray<MonitorInfo*>* monitors, 187 JavaThread* deoptee_thread, frame& fr, int exec_mode, bool realloc_failures); 188 static void pop_frames_failed_reallocs(JavaThread* thread, vframeArray* array); 189 NOT_PRODUCT(static void print_objects(GrowableArray<ScopeValue*>* objects, bool realloc_failures);) 190 #endif // COMPILER2_OR_JVMCI 191 192 public: 193 static vframeArray* create_vframeArray(JavaThread* thread, frame fr, RegisterMap *reg_map, GrowableArray<compiledVFrame*>* chunk, bool realloc_failures); 194 195 // Interface used for unpacking deoptimized frames 196 197 // UnrollBlock is returned by fetch_unroll_info() to the deoptimization handler (blob). 198 // This is only a CheapObj to ease debugging after a deopt failure 199 class UnrollBlock : public CHeapObj<mtCompiler> { 200 friend class VMStructs; 201 friend class JVMCIVMStructs; 202 private: 203 int _size_of_deoptimized_frame; // Size, in bytes, of current deoptimized frame 204 int _caller_adjustment; // Adjustment, in bytes, to caller's SP by initial interpreted frame 205 int _number_of_frames; // Number frames to unroll 206 int _total_frame_sizes; // Total of number*sizes frames 207 intptr_t* _frame_sizes; // Array of frame sizes, in bytes, for unrolling the stack 208 address* _frame_pcs; // Array of frame pc's, in bytes, for unrolling the stack 209 intptr_t* _register_block; // Block for storing callee-saved registers. 210 BasicType _return_type; // Tells if we have to restore double or long return value 211 intptr_t _initial_info; // Platform dependent data for the sender frame (was FP on x86) 212 int _caller_actual_parameters; // The number of actual arguments at the 213 // interpreted caller of the deoptimized frame 214 int _unpack_kind; // exec_mode that can be changed during fetch_unroll_info 215 216 // The following fields are used as temps during the unpacking phase 217 // (which is tight on registers, especially on x86). They really ought 218 // to be PD variables but that involves moving this class into its own 219 // file to use the pd include mechanism. Maybe in a later cleanup ... 220 intptr_t _counter_temp; // SHOULD BE PD VARIABLE (x86 frame count temp) 221 intptr_t _sender_sp_temp; // SHOULD BE PD VARIABLE (x86 sender_sp) 222 public: 223 // Constructor 224 UnrollBlock(int size_of_deoptimized_frame, 225 int caller_adjustment, 226 int caller_actual_parameters, 227 int number_of_frames, 228 intptr_t* frame_sizes, 229 address* frames_pcs, 230 BasicType return_type, 231 int unpack_kind); 232 ~UnrollBlock(); 233 234 // Returns where a register is located. 235 intptr_t* value_addr_at(int register_number) const; 236 237 // Accessors frame_sizes() const238 intptr_t* frame_sizes() const { return _frame_sizes; } number_of_frames() const239 int number_of_frames() const { return _number_of_frames; } frame_pcs() const240 address* frame_pcs() const { return _frame_pcs ; } unpack_kind() const241 int unpack_kind() const { return _unpack_kind; } 242 243 // Returns the total size of frames 244 int size_of_frames() const; 245 set_initial_info(intptr_t info)246 void set_initial_info(intptr_t info) { _initial_info = info; } 247 caller_actual_parameters() const248 int caller_actual_parameters() const { return _caller_actual_parameters; } 249 250 // Accessors used by the code generator for the unpack stub. size_of_deoptimized_frame_offset_in_bytes()251 static int size_of_deoptimized_frame_offset_in_bytes() { return offset_of(UnrollBlock, _size_of_deoptimized_frame); } caller_adjustment_offset_in_bytes()252 static int caller_adjustment_offset_in_bytes() { return offset_of(UnrollBlock, _caller_adjustment); } number_of_frames_offset_in_bytes()253 static int number_of_frames_offset_in_bytes() { return offset_of(UnrollBlock, _number_of_frames); } frame_sizes_offset_in_bytes()254 static int frame_sizes_offset_in_bytes() { return offset_of(UnrollBlock, _frame_sizes); } total_frame_sizes_offset_in_bytes()255 static int total_frame_sizes_offset_in_bytes() { return offset_of(UnrollBlock, _total_frame_sizes); } frame_pcs_offset_in_bytes()256 static int frame_pcs_offset_in_bytes() { return offset_of(UnrollBlock, _frame_pcs); } register_block_offset_in_bytes()257 static int register_block_offset_in_bytes() { return offset_of(UnrollBlock, _register_block); } return_type_offset_in_bytes()258 static int return_type_offset_in_bytes() { return offset_of(UnrollBlock, _return_type); } counter_temp_offset_in_bytes()259 static int counter_temp_offset_in_bytes() { return offset_of(UnrollBlock, _counter_temp); } initial_info_offset_in_bytes()260 static int initial_info_offset_in_bytes() { return offset_of(UnrollBlock, _initial_info); } unpack_kind_offset_in_bytes()261 static int unpack_kind_offset_in_bytes() { return offset_of(UnrollBlock, _unpack_kind); } sender_sp_temp_offset_in_bytes()262 static int sender_sp_temp_offset_in_bytes() { return offset_of(UnrollBlock, _sender_sp_temp); } 263 return_type() const264 BasicType return_type() const { return _return_type; } 265 void print(); 266 }; 267 268 //** Returns an UnrollBlock continuing information 269 // how to make room for the resulting interpreter frames. 270 // Called by assembly stub after execution has returned to 271 // deoptimized frame. 272 // @argument thread. Thread where stub_frame resides. 273 // @see OptoRuntime::deoptimization_fetch_unroll_info_C 274 static UnrollBlock* fetch_unroll_info(JavaThread* thread, int exec_mode); 275 276 //** Unpacks vframeArray onto execution stack 277 // Called by assembly stub after execution has returned to 278 // deoptimized frame and after the stack unrolling. 279 // @argument thread. Thread where stub_frame resides. 280 // @argument exec_mode. Determines how execution should be continued in top frame. 281 // 0 means continue after current byte code 282 // 1 means exception has happened, handle exception 283 // 2 means reexecute current bytecode (for uncommon traps). 284 // @see OptoRuntime::deoptimization_unpack_frames_C 285 // Return BasicType of call return type, if any 286 static BasicType unpack_frames(JavaThread* thread, int exec_mode); 287 288 // Cleans up deoptimization bits on thread after unpacking or in the 289 // case of an exception. 290 static void cleanup_deopt_info(JavaThread *thread, 291 vframeArray * array); 292 293 // Restores callee saved values from deoptimized frame into oldest interpreter frame 294 // so caller of the deoptimized frame will get back the values it expects. 295 static void unwind_callee_save_values(frame* f, vframeArray* vframe_array); 296 297 //** Performs an uncommon trap for compiled code. 298 // The top most compiler frame is converted into interpreter frames 299 static UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index, jint exec_mode); 300 // Helper routine that enters the VM and may block 301 static void uncommon_trap_inner(JavaThread* thread, jint unloaded_class_index); 302 303 //** Deoptimizes the frame identified by id. 304 // Only called from VMDeoptimizeFrame 305 // @argument thread. Thread where stub_frame resides. 306 // @argument id. id of frame that should be deoptimized. 307 static void deoptimize_frame_internal(JavaThread* thread, intptr_t* id, DeoptReason reason); 308 309 // if thread is not the current thread then execute 310 // VM_DeoptimizeFrame otherwise deoptimize directly. 311 static void deoptimize_frame(JavaThread* thread, intptr_t* id, DeoptReason reason); 312 static void deoptimize_frame(JavaThread* thread, intptr_t* id); 313 314 // Statistics 315 static void gather_statistics(DeoptReason reason, DeoptAction action, 316 Bytecodes::Code bc = Bytecodes::_illegal); 317 static void print_statistics(); 318 319 // How much room to adjust the last frame's SP by, to make space for 320 // the callee's interpreter frame (which expects locals to be next to 321 // incoming arguments) 322 static int last_frame_adjust(int callee_parameters, int callee_locals); 323 324 // trap_request codes trap_request_reason(int trap_request)325 static DeoptReason trap_request_reason(int trap_request) { 326 if (trap_request < 0) 327 return (DeoptReason) 328 ((~(trap_request) >> _reason_shift) & right_n_bits(_reason_bits)); 329 else 330 // standard reason for unloaded CP entry 331 return Reason_unloaded; 332 } trap_request_action(int trap_request)333 static DeoptAction trap_request_action(int trap_request) { 334 if (trap_request < 0) 335 return (DeoptAction) 336 ((~(trap_request) >> _action_shift) & right_n_bits(_action_bits)); 337 else 338 // standard action for unloaded CP entry 339 return _unloaded_action; 340 } trap_request_debug_id(int trap_request)341 static int trap_request_debug_id(int trap_request) { 342 if (trap_request < 0) { 343 return ((~(trap_request) >> _debug_id_shift) & right_n_bits(_debug_id_bits)); 344 } else { 345 // standard action for unloaded CP entry 346 return 0; 347 } 348 } trap_request_index(int trap_request)349 static int trap_request_index(int trap_request) { 350 if (trap_request < 0) 351 return -1; 352 else 353 return trap_request; 354 } make_trap_request(DeoptReason reason,DeoptAction action,int index=-1)355 static int make_trap_request(DeoptReason reason, DeoptAction action, 356 int index = -1) { 357 assert((1 << _reason_bits) >= Reason_LIMIT, "enough bits"); 358 assert((1 << _action_bits) >= Action_LIMIT, "enough bits"); 359 int trap_request; 360 if (index != -1) 361 trap_request = index; 362 else 363 trap_request = (~(((reason) << _reason_shift) 364 + ((action) << _action_shift))); 365 assert(reason == trap_request_reason(trap_request), "valid reason"); 366 assert(action == trap_request_action(trap_request), "valid action"); 367 assert(index == trap_request_index(trap_request), "valid index"); 368 return trap_request; 369 } 370 371 // The trap_state stored in a MDO is decoded here. 372 // It records two items of information. 373 // reason: If a deoptimization happened here, what its reason was, 374 // or if there were multiple deopts with differing reasons. 375 // recompiled: If a deoptimization here triggered a recompilation. 376 // Note that not all reasons are recorded per-bci. 377 static DeoptReason trap_state_reason(int trap_state); 378 static int trap_state_has_reason(int trap_state, int reason); 379 static int trap_state_add_reason(int trap_state, int reason); 380 static bool trap_state_is_recompiled(int trap_state); 381 static int trap_state_set_recompiled(int trap_state, bool z); 382 static const char* format_trap_state(char* buf, size_t buflen, 383 int trap_state); 384 reason_is_recorded_per_bytecode(DeoptReason reason)385 static bool reason_is_recorded_per_bytecode(DeoptReason reason) { 386 return reason > Reason_none && reason <= Reason_RECORDED_LIMIT; 387 } 388 reason_recorded_per_bytecode_if_any(DeoptReason reason)389 static DeoptReason reason_recorded_per_bytecode_if_any(DeoptReason reason) { 390 if (reason_is_recorded_per_bytecode(reason)) 391 return reason; 392 else if (reason == Reason_div0_check) // null check due to divide-by-zero? 393 return Reason_null_check; // recorded per BCI as a null check 394 else if (reason == Reason_speculate_class_check) 395 return Reason_class_check; 396 else if (reason == Reason_speculate_null_check) 397 return Reason_null_check; 398 else if (reason == Reason_speculate_null_assert) 399 return Reason_null_assert; 400 else if (reason == Reason_unstable_if) 401 return Reason_intrinsic; 402 else if (reason == Reason_unstable_fused_if) 403 return Reason_range_check; 404 else 405 return Reason_none; 406 } 407 reason_is_speculate(int reason)408 static bool reason_is_speculate(int reason) { 409 if (reason == Reason_speculate_class_check || 410 reason == Reason_speculate_null_check || 411 reason == Reason_speculate_null_assert) { 412 return true; 413 } 414 return false; 415 } 416 reason_null_check(bool speculative)417 static DeoptReason reason_null_check(bool speculative) { 418 return speculative ? Deoptimization::Reason_speculate_null_check : Deoptimization::Reason_null_check; 419 } 420 reason_class_check(bool speculative)421 static DeoptReason reason_class_check(bool speculative) { 422 return speculative ? Deoptimization::Reason_speculate_class_check : Deoptimization::Reason_class_check; 423 } 424 reason_null_assert(bool speculative)425 static DeoptReason reason_null_assert(bool speculative) { 426 return speculative ? Deoptimization::Reason_speculate_null_assert : Deoptimization::Reason_null_assert; 427 } 428 per_method_trap_limit(int reason)429 static uint per_method_trap_limit(int reason) { 430 return reason_is_speculate(reason) ? (uint)PerMethodSpecTrapLimit : (uint)PerMethodTrapLimit; 431 } 432 433 static const char* trap_reason_name(int reason); 434 static const char* trap_action_name(int action); 435 // Format like reason='foo' action='bar' index='123'. 436 // This is suitable both for XML and for tty output. 437 static const char* format_trap_request(char* buf, size_t buflen, 438 int trap_request); 439 440 static jint total_deoptimization_count(); 441 442 // JVMTI PopFrame support 443 444 // Preserves incoming arguments to the popped frame when it is 445 // returning to a deoptimized caller 446 static void popframe_preserve_args(JavaThread* thread, int bytes_to_save, void* start_address); 447 448 static MethodData* get_method_data(JavaThread* thread, const methodHandle& m, bool create_if_missing); 449 private: 450 // Update the mdo's count and per-BCI reason bits, returning previous state: 451 static ProfileData* query_update_method_data(MethodData* trap_mdo, 452 int trap_bci, 453 DeoptReason reason, 454 bool update_total_trap_count, 455 #if INCLUDE_JVMCI 456 bool is_osr, 457 #endif 458 Method* compiled_method, 459 //outputs: 460 uint& ret_this_trap_count, 461 bool& ret_maybe_prior_trap, 462 bool& ret_maybe_prior_recompile); 463 // class loading support for uncommon trap 464 static void load_class_by_index(const constantPoolHandle& constant_pool, int index, TRAPS); 465 466 static UnrollBlock* fetch_unroll_info_helper(JavaThread* thread, int exec_mode); 467 468 static DeoptAction _unloaded_action; // == Action_reinterpret; 469 static const char* _trap_reason_name[]; 470 static const char* _trap_action_name[]; 471 472 static juint _deoptimization_hist[Reason_LIMIT][1+Action_LIMIT][BC_CASE_LIMIT]; 473 // Note: Histogram array size is 1-2 Kb. 474 475 public: 476 static void update_method_data_from_interpreter(MethodData* trap_mdo, int trap_bci, int reason); 477 }; 478 479 480 class DeoptimizationMarker : StackObj { // for profiling 481 static bool _is_active; 482 public: DeoptimizationMarker()483 DeoptimizationMarker() { _is_active = true; } ~DeoptimizationMarker()484 ~DeoptimizationMarker() { _is_active = false; } is_active()485 static bool is_active() { return _is_active; } 486 }; 487 488 #endif // SHARE_RUNTIME_DEOPTIMIZATION_HPP 489