1 /* 2 * Copyright (c) 1997, 2012, 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.inline.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 7 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 // recorded per method 66 Reason_unloaded, // unloaded class or constant pool entry 67 Reason_uninitialized, // bad class state (uninitialized) 68 Reason_unreached, // code is not reached, compiler 69 Reason_unhandled, // arbitrary compiler limitation 70 Reason_constraint, // arbitrary runtime constraint violated 71 Reason_div0_check, // a null_check due to division by zero 72 Reason_age, // nmethod too old; tier threshold reached 73 Reason_predicate, // compiler generated predicate failed 74 Reason_loop_limit_check, // compiler generated loop limits check failed 75 Reason_speculate_class_check, // saw unexpected object class from type speculation 76 Reason_speculate_null_check, // saw unexpected null from type speculation 77 Reason_rtm_state_change, // rtm state change detected 78 Reason_unstable_if, // a branch predicted always false was taken 79 Reason_unstable_fused_if, // fused two ifs that had each one untaken branch. One is now taken. 80 #if INCLUDE_JVMCI 81 Reason_aliasing, // optimistic assumption about aliasing failed 82 Reason_transfer_to_interpreter, // explicit transferToInterpreter() 83 Reason_not_compiled_exception_handler, 84 Reason_unresolved, 85 Reason_jsr_mismatch, 86 #endif 87 88 // Reason_tenured is counted separately, add normal counted Reasons above. 89 // Related to MethodData::_trap_hist_limit where Reason_tenured isn't included 90 Reason_tenured, // age of the code has reached the limit 91 Reason_LIMIT, 92 93 // Note: Keep this enum in sync. with _trap_reason_name. 94 Reason_RECORDED_LIMIT = Reason_bimorphic // some are not recorded per bc 95 // Note: Reason_RECORDED_LIMIT should be < 8 to fit into 3 bits of 96 // DataLayout::trap_bits. This dependency is enforced indirectly 97 // via asserts, to avoid excessive direct header-to-header dependencies. 98 // See Deoptimization::trap_state_reason and class DataLayout. 99 }; 100 101 // What action must be taken by the runtime? 102 enum DeoptAction { 103 Action_none, // just interpret, do not invalidate nmethod 104 Action_maybe_recompile, // recompile the nmethod; need not invalidate 105 Action_reinterpret, // invalidate the nmethod, reset IC, maybe recompile 106 Action_make_not_entrant, // invalidate the nmethod, recompile (probably) 107 Action_make_not_compilable, // invalidate the nmethod and do not compile 108 Action_LIMIT 109 // Note: Keep this enum in sync. with _trap_action_name. 110 }; 111 112 enum { 113 _action_bits = 3, 114 _reason_bits = 5, 115 _debug_id_bits = 23, 116 _action_shift = 0, 117 _reason_shift = _action_shift+_action_bits, 118 _debug_id_shift = _reason_shift+_reason_bits, 119 BC_CASE_LIMIT = PRODUCT_ONLY(1) NOT_PRODUCT(4) // for _deoptimization_hist 120 }; 121 122 enum UnpackType { 123 Unpack_deopt = 0, // normal deoptimization, use pc computed in unpack_vframe_on_stack 124 Unpack_exception = 1, // exception is pending 125 Unpack_uncommon_trap = 2, // redo last byte code (C2 only) 126 Unpack_reexecute = 3 // reexecute bytecode (C1 only) 127 }; 128 129 // Checks all compiled methods. Invalid methods are deleted and 130 // corresponding activations are deoptimized. 131 static int deoptimize_dependents(); 132 133 // Deoptimizes a frame lazily. nmethod gets patched deopt happens on return to the frame 134 static void deoptimize(JavaThread* thread, frame fr, RegisterMap *reg_map); 135 static void deoptimize(JavaThread* thread, frame fr, RegisterMap *reg_map, DeoptReason reason); 136 137 private: 138 // Does the actual work for deoptimizing a single frame 139 static void deoptimize_single_frame(JavaThread* thread, frame fr, DeoptReason reason); 140 141 // Helper function to revoke biases of all monitors in frame if UseBiasedLocking 142 // is enabled 143 static void revoke_biases_of_monitors(JavaThread* thread, frame fr, RegisterMap* map); 144 // Helper function to revoke biases of all monitors in frames 145 // executing in a particular CodeBlob if UseBiasedLocking is enabled 146 static void revoke_biases_of_monitors(CodeBlob* cb); 147 148 #if defined(COMPILER2) || INCLUDE_JVMCI 149 JVMCI_ONLY(public:) 150 151 // Support for restoring non-escaping objects 152 static bool realloc_objects(JavaThread* thread, frame* fr, GrowableArray<ScopeValue*>* objects, TRAPS); 153 static void reassign_type_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, typeArrayOop obj, BasicType type); 154 static void reassign_object_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, objArrayOop obj); 155 static void reassign_fields(frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, bool realloc_failures, bool skip_internal); 156 static void relock_objects(GrowableArray<MonitorInfo*>* monitors, JavaThread* thread, bool realloc_failures); 157 static void pop_frames_failed_reallocs(JavaThread* thread, vframeArray* array); 158 NOT_PRODUCT(static void print_objects(GrowableArray<ScopeValue*>* objects, bool realloc_failures);) 159 #endif // COMPILER2 || INCLUDE_JVMCI 160 161 public: 162 static vframeArray* create_vframeArray(JavaThread* thread, frame fr, RegisterMap *reg_map, GrowableArray<compiledVFrame*>* chunk, bool realloc_failures); 163 164 // Interface used for unpacking deoptimized frames 165 166 // UnrollBlock is returned by fetch_unroll_info() to the deoptimization handler (blob). 167 // This is only a CheapObj to ease debugging after a deopt failure 168 class UnrollBlock : public CHeapObj<mtCompiler> { 169 friend class VMStructs; 170 private: 171 int _size_of_deoptimized_frame; // Size, in bytes, of current deoptimized frame 172 int _caller_adjustment; // Adjustment, in bytes, to caller's SP by initial interpreted frame 173 int _number_of_frames; // Number frames to unroll 174 int _total_frame_sizes; // Total of number*sizes frames 175 intptr_t* _frame_sizes; // Array of frame sizes, in bytes, for unrolling the stack 176 address* _frame_pcs; // Array of frame pc's, in bytes, for unrolling the stack 177 intptr_t* _register_block; // Block for storing callee-saved registers. 178 BasicType _return_type; // Tells if we have to restore double or long return value 179 intptr_t _initial_info; // Platform dependent data for the sender frame (was FP on x86) 180 int _caller_actual_parameters; // The number of actual arguments at the 181 // interpreted caller of the deoptimized frame 182 183 // The following fields are used as temps during the unpacking phase 184 // (which is tight on registers, especially on x86). They really ought 185 // to be PD variables but that involves moving this class into its own 186 // file to use the pd include mechanism. Maybe in a later cleanup ... 187 intptr_t _counter_temp; // SHOULD BE PD VARIABLE (x86 frame count temp) 188 intptr_t _unpack_kind; // SHOULD BE PD VARIABLE (x86 unpack kind) 189 intptr_t _sender_sp_temp; // SHOULD BE PD VARIABLE (x86 sender_sp) 190 public: 191 // Constructor 192 UnrollBlock(int size_of_deoptimized_frame, 193 int caller_adjustment, 194 int caller_actual_parameters, 195 int number_of_frames, 196 intptr_t* frame_sizes, 197 address* frames_pcs, 198 BasicType return_type); 199 ~UnrollBlock(); 200 201 // Returns where a register is located. 202 intptr_t* value_addr_at(int register_number) const; 203 204 // Accessors 205 intptr_t* frame_sizes() const { return _frame_sizes; } 206 int number_of_frames() const { return _number_of_frames; } 207 address* frame_pcs() const { return _frame_pcs ; } 208 209 // Returns the total size of frames 210 int size_of_frames() const; 211 212 void set_initial_info(intptr_t info) { _initial_info = info; } 213 214 int caller_actual_parameters() const { return _caller_actual_parameters; } 215 216 // Accessors used by the code generator for the unpack stub. 217 static int size_of_deoptimized_frame_offset_in_bytes() { return offset_of(UnrollBlock, _size_of_deoptimized_frame); } 218 static int caller_adjustment_offset_in_bytes() { return offset_of(UnrollBlock, _caller_adjustment); } 219 static int number_of_frames_offset_in_bytes() { return offset_of(UnrollBlock, _number_of_frames); } 220 static int frame_sizes_offset_in_bytes() { return offset_of(UnrollBlock, _frame_sizes); } 221 static int total_frame_sizes_offset_in_bytes() { return offset_of(UnrollBlock, _total_frame_sizes); } 222 static int frame_pcs_offset_in_bytes() { return offset_of(UnrollBlock, _frame_pcs); } 223 static int register_block_offset_in_bytes() { return offset_of(UnrollBlock, _register_block); } 224 static int return_type_offset_in_bytes() { return offset_of(UnrollBlock, _return_type); } 225 static int counter_temp_offset_in_bytes() { return offset_of(UnrollBlock, _counter_temp); } 226 static int initial_info_offset_in_bytes() { return offset_of(UnrollBlock, _initial_info); } 227 static int unpack_kind_offset_in_bytes() { return offset_of(UnrollBlock, _unpack_kind); } 228 static int sender_sp_temp_offset_in_bytes() { return offset_of(UnrollBlock, _sender_sp_temp); } 229 230 BasicType return_type() const { return _return_type; } 231 void print(); 232 }; 233 234 //** Returns an UnrollBlock continuing information 235 // how to make room for the resulting interpreter frames. 236 // Called by assembly stub after execution has returned to 237 // deoptimized frame. 238 // @argument thread. Thread where stub_frame resides. 239 // @see OptoRuntime::deoptimization_fetch_unroll_info_C 240 static UnrollBlock* fetch_unroll_info(JavaThread* thread); 241 242 //** Unpacks vframeArray onto execution stack 243 // Called by assembly stub after execution has returned to 244 // deoptimized frame and after the stack unrolling. 245 // @argument thread. Thread where stub_frame resides. 246 // @argument exec_mode. Determines how execution should be continued in top frame. 247 // 0 means continue after current byte code 248 // 1 means exception has happened, handle exception 249 // 2 means reexecute current bytecode (for uncommon traps). 250 // @see OptoRuntime::deoptimization_unpack_frames_C 251 // Return BasicType of call return type, if any 252 static BasicType unpack_frames(JavaThread* thread, int exec_mode); 253 254 // Cleans up deoptimization bits on thread after unpacking or in the 255 // case of an exception. 256 static void cleanup_deopt_info(JavaThread *thread, 257 vframeArray * array); 258 259 // Restores callee saved values from deoptimized frame into oldest interpreter frame 260 // so caller of the deoptimized frame will get back the values it expects. 261 static void unwind_callee_save_values(frame* f, vframeArray* vframe_array); 262 263 //** Performs an uncommon trap for compiled code. 264 // The top most compiler frame is converted into interpreter frames 265 static UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index); 266 // Helper routine that enters the VM and may block 267 static void uncommon_trap_inner(JavaThread* thread, jint unloaded_class_index); 268 269 //** Deoptimizes the frame identified by id. 270 // Only called from VMDeoptimizeFrame 271 // @argument thread. Thread where stub_frame resides. 272 // @argument id. id of frame that should be deoptimized. 273 static void deoptimize_frame_internal(JavaThread* thread, intptr_t* id, DeoptReason reason); 274 275 // if thread is not the current thread then execute 276 // VM_DeoptimizeFrame otherwise deoptimize directly. 277 static void deoptimize_frame(JavaThread* thread, intptr_t* id, DeoptReason reason); 278 static void deoptimize_frame(JavaThread* thread, intptr_t* id); 279 280 // Statistics 281 static void gather_statistics(DeoptReason reason, DeoptAction action, 282 Bytecodes::Code bc = Bytecodes::_illegal); 283 static void print_statistics(); 284 285 // How much room to adjust the last frame's SP by, to make space for 286 // the callee's interpreter frame (which expects locals to be next to 287 // incoming arguments) 288 static int last_frame_adjust(int callee_parameters, int callee_locals); 289 290 // trap_request codes 291 static DeoptReason trap_request_reason(int trap_request) { 292 if (trap_request < 0) 293 return (DeoptReason) 294 ((~(trap_request) >> _reason_shift) & right_n_bits(_reason_bits)); 295 else 296 // standard reason for unloaded CP entry 297 return Reason_unloaded; 298 } 299 static DeoptAction trap_request_action(int trap_request) { 300 if (trap_request < 0) 301 return (DeoptAction) 302 ((~(trap_request) >> _action_shift) & right_n_bits(_action_bits)); 303 else 304 // standard action for unloaded CP entry 305 return _unloaded_action; 306 } 307 static int trap_request_debug_id(int trap_request) { 308 if (trap_request < 0) { 309 return ((~(trap_request) >> _debug_id_shift) & right_n_bits(_debug_id_bits)); 310 } else { 311 // standard action for unloaded CP entry 312 return 0; 313 } 314 } 315 static int trap_request_index(int trap_request) { 316 if (trap_request < 0) 317 return -1; 318 else 319 return trap_request; 320 } 321 static int make_trap_request(DeoptReason reason, DeoptAction action, 322 int index = -1) { 323 assert((1 << _reason_bits) >= Reason_LIMIT, "enough bits"); 324 assert((1 << _action_bits) >= Action_LIMIT, "enough bits"); 325 int trap_request; 326 if (index != -1) 327 trap_request = index; 328 else 329 trap_request = (~(((reason) << _reason_shift) 330 + ((action) << _action_shift))); 331 assert(reason == trap_request_reason(trap_request), "valid reason"); 332 assert(action == trap_request_action(trap_request), "valid action"); 333 assert(index == trap_request_index(trap_request), "valid index"); 334 return trap_request; 335 } 336 337 // The trap_state stored in a MDO is decoded here. 338 // It records two items of information. 339 // reason: If a deoptimization happened here, what its reason was, 340 // or if there were multiple deopts with differing reasons. 341 // recompiled: If a deoptimization here triggered a recompilation. 342 // Note that not all reasons are recorded per-bci. 343 static DeoptReason trap_state_reason(int trap_state); 344 static int trap_state_has_reason(int trap_state, int reason); 345 static int trap_state_add_reason(int trap_state, int reason); 346 static bool trap_state_is_recompiled(int trap_state); 347 static int trap_state_set_recompiled(int trap_state, bool z); 348 static const char* format_trap_state(char* buf, size_t buflen, 349 int trap_state); 350 351 static bool reason_is_recorded_per_bytecode(DeoptReason reason) { 352 return reason > Reason_none && reason <= Reason_RECORDED_LIMIT; 353 } 354 355 static DeoptReason reason_recorded_per_bytecode_if_any(DeoptReason reason) { 356 if (reason_is_recorded_per_bytecode(reason)) 357 return reason; 358 else if (reason == Reason_div0_check) // null check due to divide-by-zero? 359 return Reason_null_check; // recorded per BCI as a null check 360 else if (reason == Reason_speculate_class_check) 361 return Reason_class_check; 362 else if (reason == Reason_speculate_null_check) 363 return Reason_null_check; 364 else if (reason == Reason_unstable_if) 365 return Reason_intrinsic; 366 else if (reason == Reason_unstable_fused_if) 367 return Reason_range_check; 368 else 369 return Reason_none; 370 } 371 372 static bool reason_is_speculate(int reason) { 373 if (reason == Reason_speculate_class_check || reason == Reason_speculate_null_check) { 374 return true; 375 } 376 return false; 377 } 378 379 static DeoptReason reason_null_check(bool speculative) { 380 return speculative ? Deoptimization::Reason_speculate_null_check : Deoptimization::Reason_null_check; 381 } 382 383 static DeoptReason reason_class_check(bool speculative) { 384 return speculative ? Deoptimization::Reason_speculate_class_check : Deoptimization::Reason_class_check; 385 } 386 387 static uint per_method_trap_limit(int reason) { 388 return reason_is_speculate(reason) ? (uint)PerMethodSpecTrapLimit : (uint)PerMethodTrapLimit; 389 } 390 391 static const char* trap_reason_name(int reason); 392 static const char* trap_action_name(int action); 393 // Format like reason='foo' action='bar' index='123'. 394 // This is suitable both for XML and for tty output. 395 static const char* format_trap_request(char* buf, size_t buflen, 396 int trap_request); 397 398 static jint total_deoptimization_count(); 399 static jint deoptimization_count(DeoptReason reason); 400 401 // JVMTI PopFrame support 402 403 // Preserves incoming arguments to the popped frame when it is 404 // returning to a deoptimized caller 405 static void popframe_preserve_args(JavaThread* thread, int bytes_to_save, void* start_address); 406 407 static MethodData* get_method_data(JavaThread* thread, methodHandle m, bool create_if_missing); 408 private: 409 // Update the mdo's count and per-BCI reason bits, returning previous state: 410 static ProfileData* query_update_method_data(MethodData* trap_mdo, 411 int trap_bci, 412 DeoptReason reason, 413 bool update_total_trap_count, 414 #if INCLUDE_JVMCI 415 bool is_osr, 416 #endif 417 Method* compiled_method, 418 //outputs: 419 uint& ret_this_trap_count, 420 bool& ret_maybe_prior_trap, 421 bool& ret_maybe_prior_recompile); 422 // class loading support for uncommon trap 423 static void load_class_by_index(constantPoolHandle constant_pool, int index, TRAPS); 424 static void load_class_by_index(constantPoolHandle constant_pool, int index); 425 426 static UnrollBlock* fetch_unroll_info_helper(JavaThread* thread); 427 428 static DeoptAction _unloaded_action; // == Action_reinterpret; 429 static const char* _trap_reason_name[]; 430 static const char* _trap_action_name[]; 431 432 static juint _deoptimization_hist[Reason_LIMIT][1+Action_LIMIT][BC_CASE_LIMIT]; 433 // Note: Histogram array size is 1-2 Kb. 434 435 public: 436 static void update_method_data_from_interpreter(MethodData* trap_mdo, int trap_bci, int reason); 437 }; 438 439 class DeoptimizationMarker : StackObj { // for profiling 440 static bool _is_active; 441 public: 442 DeoptimizationMarker() { _is_active = true; } 443 ~DeoptimizationMarker() { _is_active = false; } 444 static bool is_active() { return _is_active; } 445 }; 446 447 #endif // SHARE_VM_RUNTIME_DEOPTIMIZATION_HPP