1 /*
   2  * Copyright (c) 1997, 2019, 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 ScopeValue;
  37 class compiledVFrame;
  38 
  39 template<class E> class GrowableArray;
  40 
  41 class Deoptimization : AllStatic {
  42   friend class VMStructs;
  43 
  44  public:
  45   // What condition caused the deoptimization?
  46   enum DeoptReason {
  47     Reason_many = -1,             // indicates presence of several reasons
  48     Reason_none = 0,              // indicates absence of a relevant deopt.
  49     // Next 8 reasons are recorded per bytecode in DataLayout::trap_bits.
  50     // This is more complicated for JVMCI as JVMCI may deoptimize to *some* bytecode before the
  51     // bytecode that actually caused the deopt (with inlining, JVMCI may even deoptimize to a
  52     // bytecode in another method):
  53     //  - bytecode y in method b() causes deopt
  54     //  - JVMCI deoptimizes to bytecode x in method a()
  55     // -> the deopt reason will be recorded for method a() at bytecode x
  56     Reason_null_check,            // saw unexpected null or zero divisor (@bci)
  57     Reason_null_assert,           // saw unexpected non-null or non-zero (@bci)
  58     Reason_range_check,           // saw unexpected array index (@bci)
  59     Reason_class_check,           // saw unexpected object class (@bci)
  60     Reason_array_check,           // saw unexpected array class (aastore @bci)
  61     Reason_intrinsic,             // saw unexpected operand to intrinsic (@bci)
  62     Reason_bimorphic,             // saw unexpected object class in bimorphic inlining (@bci)
  63 
  64 #if INCLUDE_JVMCI
  65     Reason_unreached0             = Reason_null_assert,
  66     Reason_type_checked_inlining  = Reason_intrinsic,
  67     Reason_optimized_type_check   = Reason_bimorphic,
  68 #endif
  69 
  70     Reason_profile_predicate,     // compiler generated predicate moved from frequent branch in a loop failed
  71 
  72     // recorded per method
  73     Reason_unloaded,              // unloaded class or constant pool entry
  74     Reason_uninitialized,         // bad class state (uninitialized)
  75     Reason_initialized,           // class has been fully initialized
  76     Reason_unreached,             // code is not reached, compiler
  77     Reason_unhandled,             // arbitrary compiler limitation
  78     Reason_constraint,            // arbitrary runtime constraint violated
  79     Reason_div0_check,            // a null_check due to division by zero
  80     Reason_age,                   // nmethod too old; tier threshold reached
  81     Reason_predicate,             // compiler generated predicate failed
  82     Reason_loop_limit_check,      // compiler generated loop limits check failed
  83     Reason_speculate_class_check, // saw unexpected object class from type speculation
  84     Reason_speculate_null_check,  // saw unexpected null from type speculation
  85     Reason_speculate_null_assert, // saw unexpected null from type speculation
  86     Reason_rtm_state_change,      // rtm state change detected
  87     Reason_unstable_if,           // a branch predicted always false was taken
  88     Reason_unstable_fused_if,     // fused two ifs that had each one untaken branch. One is now taken.
  89 #if INCLUDE_JVMCI
  90     Reason_aliasing,              // optimistic assumption about aliasing failed
  91     Reason_transfer_to_interpreter, // explicit transferToInterpreter()
  92     Reason_not_compiled_exception_handler,
  93     Reason_unresolved,
  94     Reason_jsr_mismatch,
  95 #endif
  96 
  97     // Reason_tenured is counted separately, add normal counted Reasons above.
  98     // Related to MethodData::_trap_hist_limit where Reason_tenured isn't included
  99     Reason_tenured,               // age of the code has reached the limit
 100     Reason_LIMIT,
 101 
 102     // Note:  Keep this enum in sync. with _trap_reason_name.
 103     Reason_RECORDED_LIMIT = Reason_profile_predicate  // some are not recorded per bc
 104     // Note:  Reason_RECORDED_LIMIT should fit into 31 bits of
 105     // DataLayout::trap_bits.  This dependency is enforced indirectly
 106     // via asserts, to avoid excessive direct header-to-header dependencies.
 107     // See Deoptimization::trap_state_reason and class DataLayout.
 108   };
 109 
 110   // What action must be taken by the runtime?
 111   enum DeoptAction {
 112     Action_none,                  // just interpret, do not invalidate nmethod
 113     Action_maybe_recompile,       // recompile the nmethod; need not invalidate
 114     Action_reinterpret,           // invalidate the nmethod, reset IC, maybe recompile
 115     Action_make_not_entrant,      // invalidate the nmethod, recompile (probably)
 116     Action_make_not_compilable,   // invalidate the nmethod and do not compile
 117     Action_LIMIT
 118     // Note:  Keep this enum in sync. with _trap_action_name.
 119   };
 120 
 121   enum {
 122     _action_bits = 3,
 123     _reason_bits = 5,
 124     _debug_id_bits = 23,
 125     _action_shift = 0,
 126     _reason_shift = _action_shift+_action_bits,
 127     _debug_id_shift = _reason_shift+_reason_bits,
 128     BC_CASE_LIMIT = PRODUCT_ONLY(1) NOT_PRODUCT(4) // for _deoptimization_hist
 129   };
 130 
 131   enum UnpackType {
 132     Unpack_deopt                = 0, // normal deoptimization, use pc computed in unpack_vframe_on_stack
 133     Unpack_exception            = 1, // exception is pending
 134     Unpack_uncommon_trap        = 2, // redo last byte code (C2 only)
 135     Unpack_reexecute            = 3, // reexecute bytecode (C1 only)
 136     Unpack_LIMIT                = 4
 137   };
 138 
 139   // Checks all compiled methods. Invalid methods are deleted and
 140   // corresponding activations are deoptimized.
 141   static int deoptimize_dependents();
 142 
 143   // Deoptimizes a frame lazily. nmethod gets patched deopt happens on return to the frame
 144   static void deoptimize(JavaThread* thread, frame fr, RegisterMap *reg_map);
 145   static void deoptimize(JavaThread* thread, frame fr, RegisterMap *reg_map, DeoptReason reason);
 146 
 147 #if INCLUDE_JVMCI
 148   static address deoptimize_for_missing_exception_handler(CompiledMethod* cm);
 149 #endif
 150 
 151   private:
 152   // Does the actual work for deoptimizing a single frame
 153   static void deoptimize_single_frame(JavaThread* thread, frame fr, DeoptReason reason);
 154 
 155   // Helper function to revoke biases of all monitors in frame if UseBiasedLocking
 156   // is enabled
 157   static void revoke_biases_of_monitors(JavaThread* thread, frame fr, RegisterMap* map);
 158 
 159 #if COMPILER2_OR_JVMCI
 160 JVMCI_ONLY(public:)
 161 
 162   // Support for restoring non-escaping objects
 163   static bool realloc_objects(JavaThread* thread, frame* fr, GrowableArray<ScopeValue*>* objects, TRAPS);
 164   static void reassign_type_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, typeArrayOop obj, BasicType type);
 165   static void reassign_object_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, objArrayOop obj);
 166   static void reassign_fields(frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, bool realloc_failures, bool skip_internal);
 167   static void relock_objects(GrowableArray<MonitorInfo*>* monitors, JavaThread* thread, bool realloc_failures);
 168   static void pop_frames_failed_reallocs(JavaThread* thread, vframeArray* array);
 169   NOT_PRODUCT(static void print_objects(GrowableArray<ScopeValue*>* objects, bool realloc_failures);)
 170 #endif // COMPILER2_OR_JVMCI
 171 
 172   public:
 173   static vframeArray* create_vframeArray(JavaThread* thread, frame fr, RegisterMap *reg_map, GrowableArray<compiledVFrame*>* chunk, bool realloc_failures);
 174 
 175   // Interface used for unpacking deoptimized frames
 176 
 177   // UnrollBlock is returned by fetch_unroll_info() to the deoptimization handler (blob).
 178   // This is only a CheapObj to ease debugging after a deopt failure
 179   class UnrollBlock : public CHeapObj<mtCompiler> {
 180     friend class VMStructs;
 181     friend class JVMCIVMStructs;
 182    private:
 183     int       _size_of_deoptimized_frame; // Size, in bytes, of current deoptimized frame
 184     int       _caller_adjustment;         // Adjustment, in bytes, to caller's SP by initial interpreted frame
 185     int       _number_of_frames;          // Number frames to unroll
 186     int       _total_frame_sizes;         // Total of number*sizes frames
 187     intptr_t* _frame_sizes;               // Array of frame sizes, in bytes, for unrolling the stack
 188     address*  _frame_pcs;                 // Array of frame pc's, in bytes, for unrolling the stack
 189     intptr_t* _register_block;            // Block for storing callee-saved registers.
 190     BasicType _return_type;               // Tells if we have to restore double or long return value
 191     intptr_t  _initial_info;              // Platform dependent data for the sender frame (was FP on x86)
 192     int       _caller_actual_parameters;  // The number of actual arguments at the
 193                                           // interpreted caller of the deoptimized frame
 194     int       _unpack_kind;               // exec_mode that can be changed during fetch_unroll_info
 195 
 196     // The following fields are used as temps during the unpacking phase
 197     // (which is tight on registers, especially on x86). They really ought
 198     // to be PD variables but that involves moving this class into its own
 199     // file to use the pd include mechanism. Maybe in a later cleanup ...
 200     intptr_t  _counter_temp;              // SHOULD BE PD VARIABLE (x86 frame count temp)
 201     intptr_t  _sender_sp_temp;            // SHOULD BE PD VARIABLE (x86 sender_sp)
 202    public:
 203     // Constructor
 204     UnrollBlock(int  size_of_deoptimized_frame,
 205                 int  caller_adjustment,
 206                 int  caller_actual_parameters,
 207                 int  number_of_frames,
 208                 intptr_t* frame_sizes,
 209                 address* frames_pcs,
 210                 BasicType return_type,
 211                 int unpack_kind);
 212     ~UnrollBlock();
 213 
 214     // Returns where a register is located.
 215     intptr_t* value_addr_at(int register_number) const;
 216 
 217     // Accessors
 218     intptr_t* frame_sizes()  const { return _frame_sizes; }
 219     int number_of_frames()  const { return _number_of_frames; }
 220     address*  frame_pcs()   const { return _frame_pcs ; }
 221     int  unpack_kind()   const { return _unpack_kind; }
 222 
 223     // Returns the total size of frames
 224     int size_of_frames() const;
 225 
 226     void set_initial_info(intptr_t info) { _initial_info = info; }
 227 
 228     int caller_actual_parameters() const { return _caller_actual_parameters; }
 229 
 230     // Accessors used by the code generator for the unpack stub.
 231     static int size_of_deoptimized_frame_offset_in_bytes() { return offset_of(UnrollBlock, _size_of_deoptimized_frame); }
 232     static int caller_adjustment_offset_in_bytes()         { return offset_of(UnrollBlock, _caller_adjustment);         }
 233     static int number_of_frames_offset_in_bytes()          { return offset_of(UnrollBlock, _number_of_frames);          }
 234     static int frame_sizes_offset_in_bytes()               { return offset_of(UnrollBlock, _frame_sizes);               }
 235     static int total_frame_sizes_offset_in_bytes()         { return offset_of(UnrollBlock, _total_frame_sizes);         }
 236     static int frame_pcs_offset_in_bytes()                 { return offset_of(UnrollBlock, _frame_pcs);                 }
 237     static int register_block_offset_in_bytes()            { return offset_of(UnrollBlock, _register_block);            }
 238     static int return_type_offset_in_bytes()               { return offset_of(UnrollBlock, _return_type);               }
 239     static int counter_temp_offset_in_bytes()              { return offset_of(UnrollBlock, _counter_temp);              }
 240     static int initial_info_offset_in_bytes()              { return offset_of(UnrollBlock, _initial_info);              }
 241     static int unpack_kind_offset_in_bytes()               { return offset_of(UnrollBlock, _unpack_kind);               }
 242     static int sender_sp_temp_offset_in_bytes()            { return offset_of(UnrollBlock, _sender_sp_temp);            }
 243 
 244     BasicType return_type() const { return _return_type; }
 245     void print();
 246   };
 247 
 248   //** Returns an UnrollBlock continuing information
 249   // how to make room for the resulting interpreter frames.
 250   // Called by assembly stub after execution has returned to
 251   // deoptimized frame.
 252   // @argument thread.     Thread where stub_frame resides.
 253   // @see OptoRuntime::deoptimization_fetch_unroll_info_C
 254   static UnrollBlock* fetch_unroll_info(JavaThread* thread, int exec_mode);
 255 
 256   //** Unpacks vframeArray onto execution stack
 257   // Called by assembly stub after execution has returned to
 258   // deoptimized frame and after the stack unrolling.
 259   // @argument thread.     Thread where stub_frame resides.
 260   // @argument exec_mode.  Determines how execution should be continued in top frame.
 261   //                       0 means continue after current byte code
 262   //                       1 means exception has happened, handle exception
 263   //                       2 means reexecute current bytecode (for uncommon traps).
 264   // @see OptoRuntime::deoptimization_unpack_frames_C
 265   // Return BasicType of call return type, if any
 266   static BasicType unpack_frames(JavaThread* thread, int exec_mode);
 267 
 268   // Cleans up deoptimization bits on thread after unpacking or in the
 269   // case of an exception.
 270   static void cleanup_deopt_info(JavaThread  *thread,
 271                                  vframeArray * array);
 272 
 273   // Restores callee saved values from deoptimized frame into oldest interpreter frame
 274   // so caller of the deoptimized frame will get back the values it expects.
 275   static void unwind_callee_save_values(frame* f, vframeArray* vframe_array);
 276 
 277   //** Performs an uncommon trap for compiled code.
 278   // The top most compiler frame is converted into interpreter frames
 279   static UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index, jint exec_mode);
 280   // Helper routine that enters the VM and may block
 281   static void uncommon_trap_inner(JavaThread* thread, jint unloaded_class_index);
 282 
 283   //** Deoptimizes the frame identified by id.
 284   // Only called from VMDeoptimizeFrame
 285   // @argument thread.     Thread where stub_frame resides.
 286   // @argument id.         id of frame that should be deoptimized.
 287   static void deoptimize_frame_internal(JavaThread* thread, intptr_t* id, DeoptReason reason);
 288 
 289   // if thread is not the current thread then execute
 290   // VM_DeoptimizeFrame otherwise deoptimize directly.
 291   static void deoptimize_frame(JavaThread* thread, intptr_t* id, DeoptReason reason);
 292   static void deoptimize_frame(JavaThread* thread, intptr_t* id);
 293 
 294   // Statistics
 295   static void gather_statistics(DeoptReason reason, DeoptAction action,
 296                                 Bytecodes::Code bc = Bytecodes::_illegal);
 297   static void print_statistics();
 298 
 299   // How much room to adjust the last frame's SP by, to make space for
 300   // the callee's interpreter frame (which expects locals to be next to
 301   // incoming arguments)
 302   static int last_frame_adjust(int callee_parameters, int callee_locals);
 303 
 304   // trap_request codes
 305   static DeoptReason trap_request_reason(int trap_request) {
 306     if (trap_request < 0)
 307       return (DeoptReason)
 308         ((~(trap_request) >> _reason_shift) & right_n_bits(_reason_bits));
 309     else
 310       // standard reason for unloaded CP entry
 311       return Reason_unloaded;
 312   }
 313   static DeoptAction trap_request_action(int trap_request) {
 314     if (trap_request < 0)
 315       return (DeoptAction)
 316         ((~(trap_request) >> _action_shift) & right_n_bits(_action_bits));
 317     else
 318       // standard action for unloaded CP entry
 319       return _unloaded_action;
 320   }
 321   static int trap_request_debug_id(int trap_request) {
 322     if (trap_request < 0) {
 323       return ((~(trap_request) >> _debug_id_shift) & right_n_bits(_debug_id_bits));
 324     } else {
 325       // standard action for unloaded CP entry
 326       return 0;
 327     }
 328   }
 329   static int trap_request_index(int trap_request) {
 330     if (trap_request < 0)
 331       return -1;
 332     else
 333       return trap_request;
 334   }
 335   static int make_trap_request(DeoptReason reason, DeoptAction action,
 336                                int index = -1) {
 337     assert((1 << _reason_bits) >= Reason_LIMIT, "enough bits");
 338     assert((1 << _action_bits) >= Action_LIMIT, "enough bits");
 339     int trap_request;
 340     if (index != -1)
 341       trap_request = index;
 342     else
 343       trap_request = (~(((reason) << _reason_shift)
 344                         + ((action) << _action_shift)));
 345     assert(reason == trap_request_reason(trap_request), "valid reason");
 346     assert(action == trap_request_action(trap_request), "valid action");
 347     assert(index  == trap_request_index(trap_request),  "valid index");
 348     return trap_request;
 349   }
 350 
 351   // The trap_state stored in a MDO is decoded here.
 352   // It records two items of information.
 353   //  reason:  If a deoptimization happened here, what its reason was,
 354   //           or if there were multiple deopts with differing reasons.
 355   //  recompiled: If a deoptimization here triggered a recompilation.
 356   // Note that not all reasons are recorded per-bci.
 357   static DeoptReason trap_state_reason(int trap_state);
 358   static int  trap_state_has_reason(int trap_state, int reason);
 359   static int  trap_state_add_reason(int trap_state, int reason);
 360   static bool trap_state_is_recompiled(int trap_state);
 361   static int  trap_state_set_recompiled(int trap_state, bool z);
 362   static const char* format_trap_state(char* buf, size_t buflen,
 363                                        int trap_state);
 364 
 365   static bool reason_is_recorded_per_bytecode(DeoptReason reason) {
 366     return reason > Reason_none && reason <= Reason_RECORDED_LIMIT;
 367   }
 368 
 369   static DeoptReason reason_recorded_per_bytecode_if_any(DeoptReason reason) {
 370     if (reason_is_recorded_per_bytecode(reason))
 371       return reason;
 372     else if (reason == Reason_div0_check) // null check due to divide-by-zero?
 373       return Reason_null_check;           // recorded per BCI as a null check
 374     else if (reason == Reason_speculate_class_check)
 375       return Reason_class_check;
 376     else if (reason == Reason_speculate_null_check)
 377       return Reason_null_check;
 378     else if (reason == Reason_speculate_null_assert)
 379       return Reason_null_assert;
 380     else if (reason == Reason_unstable_if)
 381       return Reason_intrinsic;
 382     else if (reason == Reason_unstable_fused_if)
 383       return Reason_range_check;
 384     else
 385       return Reason_none;
 386   }
 387 
 388   static bool reason_is_speculate(int reason) {
 389     if (reason == Reason_speculate_class_check ||
 390         reason == Reason_speculate_null_check ||
 391         reason == Reason_speculate_null_assert) {
 392       return true;
 393     }
 394     return false;
 395   }
 396 
 397   static DeoptReason reason_null_check(bool speculative) {
 398     return speculative ? Deoptimization::Reason_speculate_null_check : Deoptimization::Reason_null_check;
 399   }
 400 
 401   static DeoptReason reason_class_check(bool speculative) {
 402     return speculative ? Deoptimization::Reason_speculate_class_check : Deoptimization::Reason_class_check;
 403   }
 404 
 405   static DeoptReason reason_null_assert(bool speculative) {
 406     return speculative ? Deoptimization::Reason_speculate_null_assert : Deoptimization::Reason_null_assert;
 407   }
 408 
 409   static uint per_method_trap_limit(int reason) {
 410     return reason_is_speculate(reason) ? (uint)PerMethodSpecTrapLimit : (uint)PerMethodTrapLimit;
 411   }
 412 
 413   static const char* trap_reason_name(int reason);
 414   static const char* trap_action_name(int action);
 415   // Format like reason='foo' action='bar' index='123'.
 416   // This is suitable both for XML and for tty output.
 417   static const char* format_trap_request(char* buf, size_t buflen,
 418                                          int trap_request);
 419 
 420   static jint total_deoptimization_count();
 421 
 422   // JVMTI PopFrame support
 423 
 424   // Preserves incoming arguments to the popped frame when it is
 425   // returning to a deoptimized caller
 426   static void popframe_preserve_args(JavaThread* thread, int bytes_to_save, void* start_address);
 427 
 428   static MethodData* get_method_data(JavaThread* thread, const methodHandle& m, bool create_if_missing);
 429  private:
 430   // Update the mdo's count and per-BCI reason bits, returning previous state:
 431   static ProfileData* query_update_method_data(MethodData* trap_mdo,
 432                                                int trap_bci,
 433                                                DeoptReason reason,
 434                                                bool update_total_trap_count,
 435 #if INCLUDE_JVMCI
 436                                                bool is_osr,
 437 #endif
 438                                                Method* compiled_method,
 439                                                //outputs:
 440                                                uint& ret_this_trap_count,
 441                                                bool& ret_maybe_prior_trap,
 442                                                bool& ret_maybe_prior_recompile);
 443   // class loading support for uncommon trap
 444   static void load_class_by_index(const constantPoolHandle& constant_pool, int index, TRAPS);
 445   static void load_class_by_index(const constantPoolHandle& constant_pool, int index);
 446 
 447   static UnrollBlock* fetch_unroll_info_helper(JavaThread* thread, int exec_mode);
 448 
 449   static DeoptAction _unloaded_action; // == Action_reinterpret;
 450   static const char* _trap_reason_name[];
 451   static const char* _trap_action_name[];
 452 
 453   static juint _deoptimization_hist[Reason_LIMIT][1+Action_LIMIT][BC_CASE_LIMIT];
 454   // Note:  Histogram array size is 1-2 Kb.
 455 
 456  public:
 457   static void update_method_data_from_interpreter(MethodData* trap_mdo, int trap_bci, int reason);
 458 };
 459 
 460 class DeoptimizationMarker : StackObj {  // for profiling
 461   static bool _is_active;
 462 public:
 463   DeoptimizationMarker()  { _is_active = true; }
 464   ~DeoptimizationMarker() { _is_active = false; }
 465   static bool is_active() { return _is_active; }
 466 };
 467 
 468 #endif // SHARE_RUNTIME_DEOPTIMIZATION_HPP