1 /*
   2  * Copyright (c) 1997, 2015, 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_speculate_null_assert, // saw unexpected null from type speculation
  78     Reason_rtm_state_change,      // rtm state change detected
  79     Reason_unstable_if,           // a branch predicted always false was taken
  80     Reason_unstable_fused_if,     // fused two ifs that had each one untaken branch. One is now taken.
  81 #if INCLUDE_JVMCI
  82     Reason_aliasing,              // optimistic assumption about aliasing failed
  83     Reason_transfer_to_interpreter, // explicit transferToInterpreter()
  84     Reason_not_compiled_exception_handler,
  85     Reason_unresolved,
  86     Reason_jsr_mismatch,
  87 #endif
  88 
  89     // Reason_tenured is counted separately, add normal counted Reasons above.
  90     // Related to MethodData::_trap_hist_limit where Reason_tenured isn't included
  91     Reason_tenured,               // age of the code has reached the limit
  92     Reason_LIMIT,
  93 
  94     // Note:  Keep this enum in sync. with _trap_reason_name.
  95     Reason_RECORDED_LIMIT = Reason_bimorphic  // some are not recorded per bc
  96     // Note:  Reason_RECORDED_LIMIT should be < 8 to fit into 3 bits of
  97     // DataLayout::trap_bits.  This dependency is enforced indirectly
  98     // via asserts, to avoid excessive direct header-to-header dependencies.
  99     // See Deoptimization::trap_state_reason and class DataLayout.
 100   };
 101 
 102   // What action must be taken by the runtime?
 103   enum DeoptAction {
 104     Action_none,                  // just interpret, do not invalidate nmethod
 105     Action_maybe_recompile,       // recompile the nmethod; need not invalidate
 106     Action_reinterpret,           // invalidate the nmethod, reset IC, maybe recompile
 107     Action_make_not_entrant,      // invalidate the nmethod, recompile (probably)
 108     Action_make_not_compilable,   // invalidate the nmethod and do not compile
 109     Action_LIMIT
 110     // Note:  Keep this enum in sync. with _trap_action_name.
 111   };
 112 
 113   enum {
 114     _action_bits = 3,
 115     _reason_bits = 5,
 116     _debug_id_bits = 23,
 117     _action_shift = 0,
 118     _reason_shift = _action_shift+_action_bits,
 119     _debug_id_shift = _reason_shift+_reason_bits,
 120     BC_CASE_LIMIT = PRODUCT_ONLY(1) NOT_PRODUCT(4) // for _deoptimization_hist
 121   };
 122 
 123   enum UnpackType {
 124     Unpack_deopt                = 0, // normal deoptimization, use pc computed in unpack_vframe_on_stack
 125     Unpack_exception            = 1, // exception is pending
 126     Unpack_uncommon_trap        = 2, // redo last byte code (C2 only)
 127     Unpack_reexecute            = 3, // reexecute bytecode (C1 only)
 128     Unpack_LIMIT                = 4
 129   };
 130 
 131   // Checks all compiled methods. Invalid methods are deleted and
 132   // corresponding activations are deoptimized.
 133   static int deoptimize_dependents();
 134 
 135   // Deoptimizes a frame lazily. nmethod gets patched deopt happens on return to the frame
 136   static void deoptimize(JavaThread* thread, frame fr, RegisterMap *reg_map);
 137   static void deoptimize(JavaThread* thread, frame fr, RegisterMap *reg_map, DeoptReason reason);
 138 
 139   private:
 140   // Does the actual work for deoptimizing a single frame
 141   static void deoptimize_single_frame(JavaThread* thread, frame fr, DeoptReason reason);
 142 
 143   // Helper function to revoke biases of all monitors in frame if UseBiasedLocking
 144   // is enabled
 145   static void revoke_biases_of_monitors(JavaThread* thread, frame fr, RegisterMap* map);
 146   // Helper function to revoke biases of all monitors in frames
 147   // executing in a particular CodeBlob if UseBiasedLocking is enabled
 148   static void revoke_biases_of_monitors(CodeBlob* cb);
 149 
 150 #if defined(COMPILER2) || INCLUDE_JVMCI
 151 JVMCI_ONLY(public:)
 152 
 153   // Support for restoring non-escaping objects
 154   static bool realloc_objects(JavaThread* thread, frame* fr, GrowableArray<ScopeValue*>* objects, TRAPS);
 155   static void reassign_type_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, typeArrayOop obj, BasicType type);
 156   static void reassign_object_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, objArrayOop obj);
 157   static void reassign_fields(frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, bool realloc_failures, bool skip_internal);
 158   static void relock_objects(GrowableArray<MonitorInfo*>* monitors, JavaThread* thread, bool realloc_failures);
 159   static void pop_frames_failed_reallocs(JavaThread* thread, vframeArray* array);
 160   NOT_PRODUCT(static void print_objects(GrowableArray<ScopeValue*>* objects, bool realloc_failures);)
 161 #endif // COMPILER2 || INCLUDE_JVMCI
 162 
 163   public:
 164   static vframeArray* create_vframeArray(JavaThread* thread, frame fr, RegisterMap *reg_map, GrowableArray<compiledVFrame*>* chunk, bool realloc_failures);
 165 
 166   // Interface used for unpacking deoptimized frames
 167 
 168   // UnrollBlock is returned by fetch_unroll_info() to the deoptimization handler (blob).
 169   // This is only a CheapObj to ease debugging after a deopt failure
 170   class UnrollBlock : public CHeapObj<mtCompiler> {
 171     friend class VMStructs;
 172     friend class JVMCIVMStructs;
 173    private:
 174     int       _size_of_deoptimized_frame; // Size, in bytes, of current deoptimized frame
 175     int       _caller_adjustment;         // Adjustment, in bytes, to caller's SP by initial interpreted frame
 176     int       _number_of_frames;          // Number frames to unroll
 177     int       _total_frame_sizes;         // Total of number*sizes frames
 178     intptr_t* _frame_sizes;               // Array of frame sizes, in bytes, for unrolling the stack
 179     address*  _frame_pcs;                 // Array of frame pc's, in bytes, for unrolling the stack
 180     intptr_t* _register_block;            // Block for storing callee-saved registers.
 181     BasicType _return_type;               // Tells if we have to restore double or long return value
 182     intptr_t  _initial_info;              // Platform dependent data for the sender frame (was FP on x86)
 183     int       _caller_actual_parameters;  // The number of actual arguments at the
 184                                           // interpreted caller of the deoptimized frame
 185     int       _unpack_kind;               // exec_mode that can be changed during fetch_unroll_info
 186 
 187     // The following fields are used as temps during the unpacking phase
 188     // (which is tight on registers, especially on x86). They really ought
 189     // to be PD variables but that involves moving this class into its own
 190     // file to use the pd include mechanism. Maybe in a later cleanup ...
 191     intptr_t  _counter_temp;              // SHOULD BE PD VARIABLE (x86 frame count temp)
 192     intptr_t  _sender_sp_temp;            // SHOULD BE PD VARIABLE (x86 sender_sp)
 193    public:
 194     // Constructor
 195     UnrollBlock(int  size_of_deoptimized_frame,
 196                 int  caller_adjustment,
 197                 int  caller_actual_parameters,
 198                 int  number_of_frames,
 199                 intptr_t* frame_sizes,
 200                 address* frames_pcs,
 201                 BasicType return_type,
 202                 int unpack_kind);
 203     ~UnrollBlock();
 204 
 205     // Returns where a register is located.
 206     intptr_t* value_addr_at(int register_number) const;
 207 
 208     // Accessors
 209     intptr_t* frame_sizes()  const { return _frame_sizes; }
 210     int number_of_frames()  const { return _number_of_frames; }
 211     address*  frame_pcs()   const { return _frame_pcs ; }
 212     int  unpack_kind()   const { return _unpack_kind; }
 213 
 214     // Returns the total size of frames
 215     int size_of_frames() const;
 216 
 217     void set_initial_info(intptr_t info) { _initial_info = info; }
 218 
 219     int caller_actual_parameters() const { return _caller_actual_parameters; }
 220 
 221     // Accessors used by the code generator for the unpack stub.
 222     static int size_of_deoptimized_frame_offset_in_bytes() { return offset_of(UnrollBlock, _size_of_deoptimized_frame); }
 223     static int caller_adjustment_offset_in_bytes()         { return offset_of(UnrollBlock, _caller_adjustment);         }
 224     static int number_of_frames_offset_in_bytes()          { return offset_of(UnrollBlock, _number_of_frames);          }
 225     static int frame_sizes_offset_in_bytes()               { return offset_of(UnrollBlock, _frame_sizes);               }
 226     static int total_frame_sizes_offset_in_bytes()         { return offset_of(UnrollBlock, _total_frame_sizes);         }
 227     static int frame_pcs_offset_in_bytes()                 { return offset_of(UnrollBlock, _frame_pcs);                 }
 228     static int register_block_offset_in_bytes()            { return offset_of(UnrollBlock, _register_block);            }
 229     static int return_type_offset_in_bytes()               { return offset_of(UnrollBlock, _return_type);               }
 230     static int counter_temp_offset_in_bytes()              { return offset_of(UnrollBlock, _counter_temp);              }
 231     static int initial_info_offset_in_bytes()              { return offset_of(UnrollBlock, _initial_info);              }
 232     static int unpack_kind_offset_in_bytes()               { return offset_of(UnrollBlock, _unpack_kind);               }
 233     static int sender_sp_temp_offset_in_bytes()            { return offset_of(UnrollBlock, _sender_sp_temp);            }
 234 
 235     BasicType return_type() const { return _return_type; }
 236     void print();
 237   };
 238 
 239   //** Returns an UnrollBlock continuing information
 240   // how to make room for the resulting interpreter frames.
 241   // Called by assembly stub after execution has returned to
 242   // deoptimized frame.
 243   // @argument thread.     Thread where stub_frame resides.
 244   // @see OptoRuntime::deoptimization_fetch_unroll_info_C
 245   static UnrollBlock* fetch_unroll_info(JavaThread* thread, int exec_mode);
 246 
 247   //** Unpacks vframeArray onto execution stack
 248   // Called by assembly stub after execution has returned to
 249   // deoptimized frame and after the stack unrolling.
 250   // @argument thread.     Thread where stub_frame resides.
 251   // @argument exec_mode.  Determines how execution should be continued in top frame.
 252   //                       0 means continue after current byte code
 253   //                       1 means exception has happened, handle exception
 254   //                       2 means reexecute current bytecode (for uncommon traps).
 255   // @see OptoRuntime::deoptimization_unpack_frames_C
 256   // Return BasicType of call return type, if any
 257   static BasicType unpack_frames(JavaThread* thread, int exec_mode);
 258 
 259   // Cleans up deoptimization bits on thread after unpacking or in the
 260   // case of an exception.
 261   static void cleanup_deopt_info(JavaThread  *thread,
 262                                  vframeArray * array);
 263 
 264   // Restores callee saved values from deoptimized frame into oldest interpreter frame
 265   // so caller of the deoptimized frame will get back the values it expects.
 266   static void unwind_callee_save_values(frame* f, vframeArray* vframe_array);
 267 
 268   //** Performs an uncommon trap for compiled code.
 269   // The top most compiler frame is converted into interpreter frames
 270   static UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index, jint exec_mode);
 271   // Helper routine that enters the VM and may block
 272   static void uncommon_trap_inner(JavaThread* thread, jint unloaded_class_index);
 273 
 274   //** Deoptimizes the frame identified by id.
 275   // Only called from VMDeoptimizeFrame
 276   // @argument thread.     Thread where stub_frame resides.
 277   // @argument id.         id of frame that should be deoptimized.
 278   static void deoptimize_frame_internal(JavaThread* thread, intptr_t* id, DeoptReason reason);
 279 
 280   // if thread is not the current thread then execute
 281   // VM_DeoptimizeFrame otherwise deoptimize directly.
 282   static void deoptimize_frame(JavaThread* thread, intptr_t* id, DeoptReason reason);
 283   static void deoptimize_frame(JavaThread* thread, intptr_t* id);
 284 
 285   // Statistics
 286   static void gather_statistics(DeoptReason reason, DeoptAction action,
 287                                 Bytecodes::Code bc = Bytecodes::_illegal);
 288   static void print_statistics();
 289 
 290   // How much room to adjust the last frame's SP by, to make space for
 291   // the callee's interpreter frame (which expects locals to be next to
 292   // incoming arguments)
 293   static int last_frame_adjust(int callee_parameters, int callee_locals);
 294 
 295   // trap_request codes
 296   static DeoptReason trap_request_reason(int trap_request) {
 297     if (trap_request < 0)
 298       return (DeoptReason)
 299         ((~(trap_request) >> _reason_shift) & right_n_bits(_reason_bits));
 300     else
 301       // standard reason for unloaded CP entry
 302       return Reason_unloaded;
 303   }
 304   static DeoptAction trap_request_action(int trap_request) {
 305     if (trap_request < 0)
 306       return (DeoptAction)
 307         ((~(trap_request) >> _action_shift) & right_n_bits(_action_bits));
 308     else
 309       // standard action for unloaded CP entry
 310       return _unloaded_action;
 311   }
 312   static int trap_request_debug_id(int trap_request) {
 313     if (trap_request < 0) {
 314       return ((~(trap_request) >> _debug_id_shift) & right_n_bits(_debug_id_bits));
 315     } else {
 316       // standard action for unloaded CP entry
 317       return 0;
 318     }
 319   }
 320   static int trap_request_index(int trap_request) {
 321     if (trap_request < 0)
 322       return -1;
 323     else
 324       return trap_request;
 325   }
 326   static int make_trap_request(DeoptReason reason, DeoptAction action,
 327                                int index = -1) {
 328     assert((1 << _reason_bits) >= Reason_LIMIT, "enough bits");
 329     assert((1 << _action_bits) >= Action_LIMIT, "enough bits");
 330     int trap_request;
 331     if (index != -1)
 332       trap_request = index;
 333     else
 334       trap_request = (~(((reason) << _reason_shift)
 335                         + ((action) << _action_shift)));
 336     assert(reason == trap_request_reason(trap_request), "valid reason");
 337     assert(action == trap_request_action(trap_request), "valid action");
 338     assert(index  == trap_request_index(trap_request),  "valid index");
 339     return trap_request;
 340   }
 341 
 342   // The trap_state stored in a MDO is decoded here.
 343   // It records two items of information.
 344   //  reason:  If a deoptimization happened here, what its reason was,
 345   //           or if there were multiple deopts with differing reasons.
 346   //  recompiled: If a deoptimization here triggered a recompilation.
 347   // Note that not all reasons are recorded per-bci.
 348   static DeoptReason trap_state_reason(int trap_state);
 349   static int  trap_state_has_reason(int trap_state, int reason);
 350   static int  trap_state_add_reason(int trap_state, int reason);
 351   static bool trap_state_is_recompiled(int trap_state);
 352   static int  trap_state_set_recompiled(int trap_state, bool z);
 353   static const char* format_trap_state(char* buf, size_t buflen,
 354                                        int trap_state);
 355 
 356   static bool reason_is_recorded_per_bytecode(DeoptReason reason) {
 357     return reason > Reason_none && reason <= Reason_RECORDED_LIMIT;
 358   }
 359 
 360   static DeoptReason reason_recorded_per_bytecode_if_any(DeoptReason reason) {
 361     if (reason_is_recorded_per_bytecode(reason))
 362       return reason;
 363     else if (reason == Reason_div0_check) // null check due to divide-by-zero?
 364       return Reason_null_check;           // recorded per BCI as a null check
 365     else if (reason == Reason_speculate_class_check)
 366       return Reason_class_check;
 367     else if (reason == Reason_speculate_null_check)
 368       return Reason_null_check;
 369     else if (reason == Reason_speculate_null_assert)
 370       return Reason_null_assert;
 371     else if (reason == Reason_unstable_if)
 372       return Reason_intrinsic;
 373     else if (reason == Reason_unstable_fused_if)
 374       return Reason_range_check;
 375     else
 376       return Reason_none;
 377   }
 378 
 379   static bool reason_is_speculate(int reason) {
 380     if (reason == Reason_speculate_class_check ||
 381         reason == Reason_speculate_null_check ||
 382         reason == Reason_speculate_null_assert) {
 383       return true;
 384     }
 385     return false;
 386   }
 387 
 388   static DeoptReason reason_null_check(bool speculative) {
 389     return speculative ? Deoptimization::Reason_speculate_null_check : Deoptimization::Reason_null_check;
 390   }
 391 
 392   static DeoptReason reason_class_check(bool speculative) {
 393     return speculative ? Deoptimization::Reason_speculate_class_check : Deoptimization::Reason_class_check;
 394   }
 395 
 396   static DeoptReason reason_null_assert(bool speculative) {
 397     return speculative ? Deoptimization::Reason_speculate_null_assert : Deoptimization::Reason_null_assert;
 398   }
 399 
 400   static uint per_method_trap_limit(int reason) {
 401     return reason_is_speculate(reason) ? (uint)PerMethodSpecTrapLimit : (uint)PerMethodTrapLimit;
 402   }
 403 
 404   static const char* trap_reason_name(int reason);
 405   static const char* trap_action_name(int action);
 406   // Format like reason='foo' action='bar' index='123'.
 407   // This is suitable both for XML and for tty output.
 408   static const char* format_trap_request(char* buf, size_t buflen,
 409                                          int trap_request);
 410 
 411   static jint total_deoptimization_count();
 412   static jint deoptimization_count(DeoptReason reason);
 413 
 414   // JVMTI PopFrame support
 415 
 416   // Preserves incoming arguments to the popped frame when it is
 417   // returning to a deoptimized caller
 418   static void popframe_preserve_args(JavaThread* thread, int bytes_to_save, void* start_address);
 419 
 420   static MethodData* get_method_data(JavaThread* thread, const methodHandle& m, bool create_if_missing);
 421  private:
 422   // Update the mdo's count and per-BCI reason bits, returning previous state:
 423   static ProfileData* query_update_method_data(MethodData* trap_mdo,
 424                                                int trap_bci,
 425                                                DeoptReason reason,
 426                                                bool update_total_trap_count,
 427 #if INCLUDE_JVMCI
 428                                                bool is_osr,
 429 #endif
 430                                                Method* compiled_method,
 431                                                //outputs:
 432                                                uint& ret_this_trap_count,
 433                                                bool& ret_maybe_prior_trap,
 434                                                bool& ret_maybe_prior_recompile);
 435   // class loading support for uncommon trap
 436   static void load_class_by_index(const constantPoolHandle& constant_pool, int index, TRAPS);
 437   static void load_class_by_index(const constantPoolHandle& constant_pool, int index);
 438 
 439   static UnrollBlock* fetch_unroll_info_helper(JavaThread* thread, int exec_mode);
 440 
 441   static DeoptAction _unloaded_action; // == Action_reinterpret;
 442   static const char* _trap_reason_name[];
 443   static const char* _trap_action_name[];
 444 
 445   static juint _deoptimization_hist[Reason_LIMIT][1+Action_LIMIT][BC_CASE_LIMIT];
 446   // Note:  Histogram array size is 1-2 Kb.
 447 
 448  public:
 449   static void update_method_data_from_interpreter(MethodData* trap_mdo, int trap_bci, int reason);
 450 };
 451 
 452 class DeoptimizationMarker : StackObj {  // for profiling
 453   static bool _is_active;
 454 public:
 455   DeoptimizationMarker()  { _is_active = true; }
 456   ~DeoptimizationMarker() { _is_active = false; }
 457   static bool is_active() { return _is_active; }
 458 };
 459 
 460 #endif // SHARE_VM_RUNTIME_DEOPTIMIZATION_HPP