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 #ifndef SHARE_VM_RUNTIME_DEOPTIMIZATION_HPP 26 #define SHARE_VM_RUNTIME_DEOPTIMIZATION_HPP 27 28 #include "memory/allocation.hpp" 29 #include "runtime/frame.hpp" 30 31 class ProfileData; 32 class vframeArray; 33 class MonitorValue; 34 class ObjectValue; 35 36 class Deoptimization : AllStatic { 37 friend class VMStructs; 38 39 public: 40 // What condition caused the deoptimization? 41 enum DeoptReason { 42 Reason_many = -1, // indicates presence of several reasons 43 Reason_none = 0, // indicates absence of a relevant deopt. 44 // Next 8 reasons are recorded per bytecode in DataLayout::trap_bits. 45 // This is more complicated for JVMCI as JVMCI may deoptimize to *some* bytecode before the 46 // bytecode that actually caused the deopt (with inlining, JVMCI may even deoptimize to a 47 // bytecode in another method): 48 // - bytecode y in method b() causes deopt 49 // - JVMCI deoptimizes to bytecode x in method a() 50 // -> the deopt reason will be recorded for method a() at bytecode x 51 Reason_null_check, // saw unexpected null or zero divisor (@bci) 52 Reason_null_assert, // saw unexpected non-null or non-zero (@bci) 53 Reason_range_check, // saw unexpected array index (@bci) 54 Reason_class_check, // saw unexpected object class (@bci) 55 Reason_array_check, // saw unexpected array class (aastore @bci) 56 Reason_intrinsic, // saw unexpected operand to intrinsic (@bci) 57 Reason_bimorphic, // saw unexpected object class in bimorphic inlining (@bci) 58 59 #if INCLUDE_JVMCI 60 Reason_unreached0 = Reason_null_assert, 61 Reason_type_checked_inlining = Reason_intrinsic, 62 Reason_optimized_type_check = Reason_bimorphic, 63 #endif 64 65 Reason_profile_predicate, // compiler generated predicate moved from frequent branch in a loop failed 66 67 // recorded per method 68 Reason_unloaded, // unloaded class or constant pool entry 69 Reason_uninitialized, // bad class state (uninitialized) 70 Reason_unreached, // code is not reached, compiler 71 Reason_unhandled, // arbitrary compiler limitation 72 Reason_constraint, // arbitrary runtime constraint violated 73 Reason_div0_check, // a null_check due to division by zero 74 Reason_age, // nmethod too old; tier threshold reached 75 Reason_predicate, // compiler generated predicate failed 76 Reason_loop_limit_check, // compiler generated loop limits check failed 77 Reason_speculate_class_check, // saw unexpected object class from type speculation 78 Reason_speculate_null_check, // saw unexpected null from type speculation 79 Reason_speculate_null_assert, // saw unexpected null from type speculation 80 Reason_rtm_state_change, // rtm state change detected 81 Reason_unstable_if, // a branch predicted always false was taken 82 Reason_unstable_fused_if, // fused two ifs that had each one untaken branch. One is now taken. 83 #if INCLUDE_JVMCI 84 Reason_aliasing, // optimistic assumption about aliasing failed 85 Reason_transfer_to_interpreter, // explicit transferToInterpreter() 86 Reason_not_compiled_exception_handler, 87 Reason_unresolved, 88 Reason_jsr_mismatch, 89 #endif 90 91 // Reason_tenured is counted separately, add normal counted Reasons above. 92 // Related to MethodData::_trap_hist_limit where Reason_tenured isn't included 93 Reason_tenured, // age of the code has reached the limit 94 Reason_LIMIT, 95 96 // Note: Keep this enum in sync. with _trap_reason_name. 97 Reason_RECORDED_LIMIT = Reason_profile_predicate // some are not recorded per bc 98 // Note: Reason_RECORDED_LIMIT should fit into 31 bits of 99 // DataLayout::trap_bits. This dependency is enforced indirectly 100 // via asserts, to avoid excessive direct header-to-header dependencies. 101 // See Deoptimization::trap_state_reason and class DataLayout. 102 }; 103 104 // What action must be taken by the runtime? 105 enum DeoptAction { 106 Action_none, // just interpret, do not invalidate nmethod 107 Action_maybe_recompile, // recompile the nmethod; need not invalidate 108 Action_reinterpret, // invalidate the nmethod, reset IC, maybe recompile 109 Action_make_not_entrant, // invalidate the nmethod, recompile (probably) 110 Action_make_not_compilable, // invalidate the nmethod and do not compile 111 Action_LIMIT 112 // Note: Keep this enum in sync. with _trap_action_name. 113 }; 114 115 enum { 116 _action_bits = 3, 117 _reason_bits = 5, 118 _debug_id_bits = 23, 119 _action_shift = 0, 120 _reason_shift = _action_shift+_action_bits, 121 _debug_id_shift = _reason_shift+_reason_bits, 122 BC_CASE_LIMIT = PRODUCT_ONLY(1) NOT_PRODUCT(4) // for _deoptimization_hist 123 }; 124 125 enum UnpackType { 126 Unpack_deopt = 0, // normal deoptimization, use pc computed in unpack_vframe_on_stack 127 Unpack_exception = 1, // exception is pending 128 Unpack_uncommon_trap = 2, // redo last byte code (C2 only) 129 Unpack_reexecute = 3, // reexecute bytecode (C1 only) 130 Unpack_LIMIT = 4 131 }; 132 133 // Checks all compiled methods. Invalid methods are deleted and 134 // corresponding activations are deoptimized. 135 static int deoptimize_dependents(); 136 137 // Deoptimizes a frame lazily. nmethod gets patched deopt happens on return to the frame 138 static void deoptimize(JavaThread* thread, frame fr, RegisterMap *reg_map); 139 static void deoptimize(JavaThread* thread, frame fr, RegisterMap *reg_map, DeoptReason reason); 140 141 #if INCLUDE_JVMCI 142 static address deoptimize_for_missing_exception_handler(CompiledMethod* cm); 143 #endif 144 145 private: 146 // Does the actual work for deoptimizing a single frame 147 static void deoptimize_single_frame(JavaThread* thread, frame fr, DeoptReason reason); 148 149 // Helper function to revoke biases of all monitors in frame if UseBiasedLocking 150 // is enabled 151 static void revoke_biases_of_monitors(JavaThread* thread, frame fr, RegisterMap* map); 152 // Helper function to revoke biases of all monitors in frames 153 // executing in a particular CodeBlob if UseBiasedLocking is enabled 154 static void revoke_biases_of_monitors(CodeBlob* cb); 155 156 #if COMPILER2_OR_JVMCI 157 JVMCI_ONLY(public:) 158 159 // Support for restoring non-escaping objects 160 static bool realloc_objects(JavaThread* thread, frame* fr, GrowableArray<ScopeValue*>* objects, TRAPS); 161 static void reassign_type_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, typeArrayOop obj, BasicType type); 162 static void reassign_object_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, objArrayOop obj); 163 static void reassign_fields(frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, bool realloc_failures, bool skip_internal); 164 static void relock_objects(GrowableArray<MonitorInfo*>* monitors, JavaThread* thread, bool realloc_failures); 165 static void pop_frames_failed_reallocs(JavaThread* thread, vframeArray* array); 166 NOT_PRODUCT(static void print_objects(GrowableArray<ScopeValue*>* objects, bool realloc_failures);) 167 #endif // COMPILER2_OR_JVMCI 168 169 public: 170 static vframeArray* create_vframeArray(JavaThread* thread, frame fr, RegisterMap *reg_map, GrowableArray<compiledVFrame*>* chunk, bool realloc_failures); 171 172 // Interface used for unpacking deoptimized frames 173 174 // UnrollBlock is returned by fetch_unroll_info() to the deoptimization handler (blob). 175 // This is only a CheapObj to ease debugging after a deopt failure 176 class UnrollBlock : public CHeapObj<mtCompiler> { 177 friend class VMStructs; 178 friend class JVMCIVMStructs; 179 private: 180 int _size_of_deoptimized_frame; // Size, in bytes, of current deoptimized frame 181 int _caller_adjustment; // Adjustment, in bytes, to caller's SP by initial interpreted frame 182 int _number_of_frames; // Number frames to unroll 183 int _total_frame_sizes; // Total of number*sizes frames 184 intptr_t* _frame_sizes; // Array of frame sizes, in bytes, for unrolling the stack 185 address* _frame_pcs; // Array of frame pc's, in bytes, for unrolling the stack 186 intptr_t* _register_block; // Block for storing callee-saved registers. 187 BasicType _return_type; // Tells if we have to restore double or long return value 188 intptr_t _initial_info; // Platform dependent data for the sender frame (was FP on x86) 189 int _caller_actual_parameters; // The number of actual arguments at the 190 // interpreted caller of the deoptimized frame 191 int _unpack_kind; // exec_mode that can be changed during fetch_unroll_info 192 193 // The following fields are used as temps during the unpacking phase 194 // (which is tight on registers, especially on x86). They really ought 195 // to be PD variables but that involves moving this class into its own 196 // file to use the pd include mechanism. Maybe in a later cleanup ... 197 intptr_t _counter_temp; // SHOULD BE PD VARIABLE (x86 frame count temp) 198 intptr_t _sender_sp_temp; // SHOULD BE PD VARIABLE (x86 sender_sp) 199 public: 200 // Constructor 201 UnrollBlock(int size_of_deoptimized_frame, 202 int caller_adjustment, 203 int caller_actual_parameters, 204 int number_of_frames, 205 intptr_t* frame_sizes, 206 address* frames_pcs, 207 BasicType return_type, 208 int unpack_kind); 209 ~UnrollBlock(); 210 211 // Returns where a register is located. 212 intptr_t* value_addr_at(int register_number) const; 213 214 // Accessors frame_sizes() const215 intptr_t* frame_sizes() const { return _frame_sizes; } number_of_frames() const216 int number_of_frames() const { return _number_of_frames; } frame_pcs() const217 address* frame_pcs() const { return _frame_pcs ; } unpack_kind() const218 int unpack_kind() const { return _unpack_kind; } 219 220 // Returns the total size of frames 221 int size_of_frames() const; 222 set_initial_info(intptr_t info)223 void set_initial_info(intptr_t info) { _initial_info = info; } 224 caller_actual_parameters() const225 int caller_actual_parameters() const { return _caller_actual_parameters; } 226 227 // Accessors used by the code generator for the unpack stub. size_of_deoptimized_frame_offset_in_bytes()228 static int size_of_deoptimized_frame_offset_in_bytes() { return offset_of(UnrollBlock, _size_of_deoptimized_frame); } caller_adjustment_offset_in_bytes()229 static int caller_adjustment_offset_in_bytes() { return offset_of(UnrollBlock, _caller_adjustment); } number_of_frames_offset_in_bytes()230 static int number_of_frames_offset_in_bytes() { return offset_of(UnrollBlock, _number_of_frames); } frame_sizes_offset_in_bytes()231 static int frame_sizes_offset_in_bytes() { return offset_of(UnrollBlock, _frame_sizes); } total_frame_sizes_offset_in_bytes()232 static int total_frame_sizes_offset_in_bytes() { return offset_of(UnrollBlock, _total_frame_sizes); } frame_pcs_offset_in_bytes()233 static int frame_pcs_offset_in_bytes() { return offset_of(UnrollBlock, _frame_pcs); } register_block_offset_in_bytes()234 static int register_block_offset_in_bytes() { return offset_of(UnrollBlock, _register_block); } return_type_offset_in_bytes()235 static int return_type_offset_in_bytes() { return offset_of(UnrollBlock, _return_type); } counter_temp_offset_in_bytes()236 static int counter_temp_offset_in_bytes() { return offset_of(UnrollBlock, _counter_temp); } initial_info_offset_in_bytes()237 static int initial_info_offset_in_bytes() { return offset_of(UnrollBlock, _initial_info); } unpack_kind_offset_in_bytes()238 static int unpack_kind_offset_in_bytes() { return offset_of(UnrollBlock, _unpack_kind); } sender_sp_temp_offset_in_bytes()239 static int sender_sp_temp_offset_in_bytes() { return offset_of(UnrollBlock, _sender_sp_temp); } 240 return_type() const241 BasicType return_type() const { return _return_type; } 242 void print(); 243 }; 244 245 //** Returns an UnrollBlock continuing information 246 // how to make room for the resulting interpreter frames. 247 // Called by assembly stub after execution has returned to 248 // deoptimized frame. 249 // @argument thread. Thread where stub_frame resides. 250 // @see OptoRuntime::deoptimization_fetch_unroll_info_C 251 static UnrollBlock* fetch_unroll_info(JavaThread* thread, int exec_mode); 252 253 //** Unpacks vframeArray onto execution stack 254 // Called by assembly stub after execution has returned to 255 // deoptimized frame and after the stack unrolling. 256 // @argument thread. Thread where stub_frame resides. 257 // @argument exec_mode. Determines how execution should be continued in top frame. 258 // 0 means continue after current byte code 259 // 1 means exception has happened, handle exception 260 // 2 means reexecute current bytecode (for uncommon traps). 261 // @see OptoRuntime::deoptimization_unpack_frames_C 262 // Return BasicType of call return type, if any 263 static BasicType unpack_frames(JavaThread* thread, int exec_mode); 264 265 // Cleans up deoptimization bits on thread after unpacking or in the 266 // case of an exception. 267 static void cleanup_deopt_info(JavaThread *thread, 268 vframeArray * array); 269 270 // Restores callee saved values from deoptimized frame into oldest interpreter frame 271 // so caller of the deoptimized frame will get back the values it expects. 272 static void unwind_callee_save_values(frame* f, vframeArray* vframe_array); 273 274 //** Performs an uncommon trap for compiled code. 275 // The top most compiler frame is converted into interpreter frames 276 static UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index, jint exec_mode); 277 // Helper routine that enters the VM and may block 278 static void uncommon_trap_inner(JavaThread* thread, jint unloaded_class_index); 279 280 //** Deoptimizes the frame identified by id. 281 // Only called from VMDeoptimizeFrame 282 // @argument thread. Thread where stub_frame resides. 283 // @argument id. id of frame that should be deoptimized. 284 static void deoptimize_frame_internal(JavaThread* thread, intptr_t* id, DeoptReason reason); 285 286 // if thread is not the current thread then execute 287 // VM_DeoptimizeFrame otherwise deoptimize directly. 288 static void deoptimize_frame(JavaThread* thread, intptr_t* id, DeoptReason reason); 289 static void deoptimize_frame(JavaThread* thread, intptr_t* id); 290 291 // Statistics 292 static void gather_statistics(DeoptReason reason, DeoptAction action, 293 Bytecodes::Code bc = Bytecodes::_illegal); 294 static void print_statistics(); 295 296 // How much room to adjust the last frame's SP by, to make space for 297 // the callee's interpreter frame (which expects locals to be next to 298 // incoming arguments) 299 static int last_frame_adjust(int callee_parameters, int callee_locals); 300 301 // trap_request codes trap_request_reason(int trap_request)302 static DeoptReason trap_request_reason(int trap_request) { 303 if (trap_request < 0) 304 return (DeoptReason) 305 ((~(trap_request) >> _reason_shift) & right_n_bits(_reason_bits)); 306 else 307 // standard reason for unloaded CP entry 308 return Reason_unloaded; 309 } trap_request_action(int trap_request)310 static DeoptAction trap_request_action(int trap_request) { 311 if (trap_request < 0) 312 return (DeoptAction) 313 ((~(trap_request) >> _action_shift) & right_n_bits(_action_bits)); 314 else 315 // standard action for unloaded CP entry 316 return _unloaded_action; 317 } trap_request_debug_id(int trap_request)318 static int trap_request_debug_id(int trap_request) { 319 if (trap_request < 0) { 320 return ((~(trap_request) >> _debug_id_shift) & right_n_bits(_debug_id_bits)); 321 } else { 322 // standard action for unloaded CP entry 323 return 0; 324 } 325 } trap_request_index(int trap_request)326 static int trap_request_index(int trap_request) { 327 if (trap_request < 0) 328 return -1; 329 else 330 return trap_request; 331 } make_trap_request(DeoptReason reason,DeoptAction action,int index=-1)332 static int make_trap_request(DeoptReason reason, DeoptAction action, 333 int index = -1) { 334 assert((1 << _reason_bits) >= Reason_LIMIT, "enough bits"); 335 assert((1 << _action_bits) >= Action_LIMIT, "enough bits"); 336 int trap_request; 337 if (index != -1) 338 trap_request = index; 339 else 340 trap_request = (~(((reason) << _reason_shift) 341 + ((action) << _action_shift))); 342 assert(reason == trap_request_reason(trap_request), "valid reason"); 343 assert(action == trap_request_action(trap_request), "valid action"); 344 assert(index == trap_request_index(trap_request), "valid index"); 345 return trap_request; 346 } 347 348 // The trap_state stored in a MDO is decoded here. 349 // It records two items of information. 350 // reason: If a deoptimization happened here, what its reason was, 351 // or if there were multiple deopts with differing reasons. 352 // recompiled: If a deoptimization here triggered a recompilation. 353 // Note that not all reasons are recorded per-bci. 354 static DeoptReason trap_state_reason(int trap_state); 355 static int trap_state_has_reason(int trap_state, int reason); 356 static int trap_state_add_reason(int trap_state, int reason); 357 static bool trap_state_is_recompiled(int trap_state); 358 static int trap_state_set_recompiled(int trap_state, bool z); 359 static const char* format_trap_state(char* buf, size_t buflen, 360 int trap_state); 361 reason_is_recorded_per_bytecode(DeoptReason reason)362 static bool reason_is_recorded_per_bytecode(DeoptReason reason) { 363 return reason > Reason_none && reason <= Reason_RECORDED_LIMIT; 364 } 365 reason_recorded_per_bytecode_if_any(DeoptReason reason)366 static DeoptReason reason_recorded_per_bytecode_if_any(DeoptReason reason) { 367 if (reason_is_recorded_per_bytecode(reason)) 368 return reason; 369 else if (reason == Reason_div0_check) // null check due to divide-by-zero? 370 return Reason_null_check; // recorded per BCI as a null check 371 else if (reason == Reason_speculate_class_check) 372 return Reason_class_check; 373 else if (reason == Reason_speculate_null_check) 374 return Reason_null_check; 375 else if (reason == Reason_speculate_null_assert) 376 return Reason_null_assert; 377 else if (reason == Reason_unstable_if) 378 return Reason_intrinsic; 379 else if (reason == Reason_unstable_fused_if) 380 return Reason_range_check; 381 else 382 return Reason_none; 383 } 384 reason_is_speculate(int reason)385 static bool reason_is_speculate(int reason) { 386 if (reason == Reason_speculate_class_check || 387 reason == Reason_speculate_null_check || 388 reason == Reason_speculate_null_assert) { 389 return true; 390 } 391 return false; 392 } 393 reason_null_check(bool speculative)394 static DeoptReason reason_null_check(bool speculative) { 395 return speculative ? Deoptimization::Reason_speculate_null_check : Deoptimization::Reason_null_check; 396 } 397 reason_class_check(bool speculative)398 static DeoptReason reason_class_check(bool speculative) { 399 return speculative ? Deoptimization::Reason_speculate_class_check : Deoptimization::Reason_class_check; 400 } 401 reason_null_assert(bool speculative)402 static DeoptReason reason_null_assert(bool speculative) { 403 return speculative ? Deoptimization::Reason_speculate_null_assert : Deoptimization::Reason_null_assert; 404 } 405 per_method_trap_limit(int reason)406 static uint per_method_trap_limit(int reason) { 407 return reason_is_speculate(reason) ? (uint)PerMethodSpecTrapLimit : (uint)PerMethodTrapLimit; 408 } 409 410 static const char* trap_reason_name(int reason); 411 static const char* trap_action_name(int action); 412 // Format like reason='foo' action='bar' index='123'. 413 // This is suitable both for XML and for tty output. 414 static const char* format_trap_request(char* buf, size_t buflen, 415 int trap_request); 416 417 static jint total_deoptimization_count(); 418 static jint deoptimization_count(DeoptReason reason); 419 420 // JVMTI PopFrame support 421 422 // Preserves incoming arguments to the popped frame when it is 423 // returning to a deoptimized caller 424 static void popframe_preserve_args(JavaThread* thread, int bytes_to_save, void* start_address); 425 426 static MethodData* get_method_data(JavaThread* thread, const methodHandle& m, bool create_if_missing); 427 private: 428 // Update the mdo's count and per-BCI reason bits, returning previous state: 429 static ProfileData* query_update_method_data(MethodData* trap_mdo, 430 int trap_bci, 431 DeoptReason reason, 432 bool update_total_trap_count, 433 #if INCLUDE_JVMCI 434 bool is_osr, 435 #endif 436 Method* compiled_method, 437 //outputs: 438 uint& ret_this_trap_count, 439 bool& ret_maybe_prior_trap, 440 bool& ret_maybe_prior_recompile); 441 // class loading support for uncommon trap 442 static void load_class_by_index(const constantPoolHandle& constant_pool, int index, TRAPS); 443 static void load_class_by_index(const constantPoolHandle& constant_pool, int index); 444 445 static UnrollBlock* fetch_unroll_info_helper(JavaThread* thread, int exec_mode); 446 447 static DeoptAction _unloaded_action; // == Action_reinterpret; 448 static const char* _trap_reason_name[]; 449 static const char* _trap_action_name[]; 450 451 static juint _deoptimization_hist[Reason_LIMIT][1+Action_LIMIT][BC_CASE_LIMIT]; 452 // Note: Histogram array size is 1-2 Kb. 453 454 public: 455 static void update_method_data_from_interpreter(MethodData* trap_mdo, int trap_bci, int reason); 456 }; 457 458 class DeoptimizationMarker : StackObj { // for profiling 459 static bool _is_active; 460 public: DeoptimizationMarker()461 DeoptimizationMarker() { _is_active = true; } ~DeoptimizationMarker()462 ~DeoptimizationMarker() { _is_active = false; } is_active()463 static bool is_active() { return _is_active; } 464 }; 465 466 #endif // SHARE_VM_RUNTIME_DEOPTIMIZATION_HPP 467