rev 5968 : 8031320: Use Intel RTM instructions for locks
Summary: Use RTM for inflated locks and stack locks.
Reviewed-by: iveresov, twisti, roland, dcubed

   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     Reason_null_check,            // saw unexpected null or zero divisor (@bci)
  46     Reason_null_assert,           // saw unexpected non-null or non-zero (@bci)
  47     Reason_range_check,           // saw unexpected array index (@bci)
  48     Reason_class_check,           // saw unexpected object class (@bci)
  49     Reason_array_check,           // saw unexpected array class (aastore @bci)
  50     Reason_intrinsic,             // saw unexpected operand to intrinsic (@bci)
  51     Reason_bimorphic,             // saw unexpected object class in bimorphic inlining (@bci)
  52 
  53     Reason_unloaded,              // unloaded class or constant pool entry
  54     Reason_uninitialized,         // bad class state (uninitialized)
  55     Reason_unreached,             // code is not reached, compiler
  56     Reason_unhandled,             // arbitrary compiler limitation
  57     Reason_constraint,            // arbitrary runtime constraint violated
  58     Reason_div0_check,            // a null_check due to division by zero
  59     Reason_age,                   // nmethod too old; tier threshold reached
  60     Reason_predicate,             // compiler generated predicate failed
  61     Reason_loop_limit_check,      // compiler generated loop limits check failed
  62     Reason_speculate_class_check, // saw unexpected object class from type speculation
  63     Reason_rtm_state_change,      // rtm state change detected
  64     Reason_LIMIT,
  65     // Note:  Keep this enum in sync. with _trap_reason_name.
  66     Reason_RECORDED_LIMIT = Reason_bimorphic  // some are not recorded per bc
  67     // Note:  Reason_RECORDED_LIMIT should be < 8 to fit into 3 bits of
  68     // DataLayout::trap_bits.  This dependency is enforced indirectly
  69     // via asserts, to avoid excessive direct header-to-header dependencies.
  70     // See Deoptimization::trap_state_reason and class DataLayout.
  71   };
  72 
  73   // What action must be taken by the runtime?
  74   enum DeoptAction {
  75     Action_none,                  // just interpret, do not invalidate nmethod
  76     Action_maybe_recompile,       // recompile the nmethod; need not invalidate
  77     Action_reinterpret,           // invalidate the nmethod, reset IC, maybe recompile
  78     Action_make_not_entrant,      // invalidate the nmethod, recompile (probably)
  79     Action_make_not_compilable,   // invalidate the nmethod and do not compile
  80     Action_LIMIT
  81     // Note:  Keep this enum in sync. with _trap_action_name.
  82   };
  83 
  84   enum {
  85     _action_bits = 3,
  86     _reason_bits = 5,
  87     _action_shift = 0,
  88     _reason_shift = _action_shift+_action_bits,
  89     BC_CASE_LIMIT = PRODUCT_ONLY(1) NOT_PRODUCT(4) // for _deoptimization_hist
  90   };
  91 
  92   enum UnpackType {
  93     Unpack_deopt                = 0, // normal deoptimization, use pc computed in unpack_vframe_on_stack
  94     Unpack_exception            = 1, // exception is pending
  95     Unpack_uncommon_trap        = 2, // redo last byte code (C2 only)
  96     Unpack_reexecute            = 3  // reexecute bytecode (C1 only)
  97   };
  98 
  99   // Checks all compiled methods. Invalid methods are deleted and
 100   // corresponding activations are deoptimized.
 101   static int deoptimize_dependents();
 102 
 103   // Deoptimizes a frame lazily. nmethod gets patched deopt happens on return to the frame
 104   static void deoptimize(JavaThread* thread, frame fr, RegisterMap *reg_map);
 105 
 106   private:
 107   // Does the actual work for deoptimizing a single frame
 108   static void deoptimize_single_frame(JavaThread* thread, frame fr);
 109 
 110   // Helper function to revoke biases of all monitors in frame if UseBiasedLocking
 111   // is enabled
 112   static void revoke_biases_of_monitors(JavaThread* thread, frame fr, RegisterMap* map);
 113   // Helper function to revoke biases of all monitors in frames
 114   // executing in a particular CodeBlob if UseBiasedLocking is enabled
 115   static void revoke_biases_of_monitors(CodeBlob* cb);
 116 
 117 #ifdef COMPILER2
 118   // Support for restoring non-escaping objects
 119   static bool realloc_objects(JavaThread* thread, frame* fr, GrowableArray<ScopeValue*>* objects, TRAPS);
 120   static void reassign_type_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, typeArrayOop obj, BasicType type);
 121   static void reassign_object_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, objArrayOop obj);
 122   static void reassign_fields(frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects);
 123   static void relock_objects(GrowableArray<MonitorInfo*>* monitors, JavaThread* thread);
 124   NOT_PRODUCT(static void print_objects(GrowableArray<ScopeValue*>* objects);)
 125 #endif // COMPILER2
 126 
 127   public:
 128   static vframeArray* create_vframeArray(JavaThread* thread, frame fr, RegisterMap *reg_map, GrowableArray<compiledVFrame*>* chunk);
 129 
 130   // Interface used for unpacking deoptimized frames
 131 
 132   // UnrollBlock is returned by fetch_unroll_info() to the deoptimization handler (blob).
 133   // This is only a CheapObj to ease debugging after a deopt failure
 134   class UnrollBlock : public CHeapObj<mtCompiler> {
 135    private:
 136     int       _size_of_deoptimized_frame; // Size, in bytes, of current deoptimized frame
 137     int       _caller_adjustment;         // Adjustment, in bytes, to caller's SP by initial interpreted frame
 138     int       _number_of_frames;          // Number frames to unroll
 139     int       _total_frame_sizes;         // Total of number*sizes frames
 140     intptr_t* _frame_sizes;               // Array of frame sizes, in bytes, for unrolling the stack
 141     address*  _frame_pcs;                 // Array of frame pc's, in bytes, for unrolling the stack
 142     intptr_t* _register_block;            // Block for storing callee-saved registers.
 143     BasicType _return_type;               // Tells if we have to restore double or long return value
 144     intptr_t  _initial_info;              // Platform dependent data for the sender frame (was FP on x86)
 145     int       _caller_actual_parameters;  // The number of actual arguments at the
 146                                           // interpreted caller of the deoptimized frame
 147 
 148     // The following fields are used as temps during the unpacking phase
 149     // (which is tight on registers, especially on x86). They really ought
 150     // to be PD variables but that involves moving this class into its own
 151     // file to use the pd include mechanism. Maybe in a later cleanup ...
 152     intptr_t  _counter_temp;              // SHOULD BE PD VARIABLE (x86 frame count temp)
 153     intptr_t  _unpack_kind;               // SHOULD BE PD VARIABLE (x86 unpack kind)
 154     intptr_t  _sender_sp_temp;            // SHOULD BE PD VARIABLE (x86 sender_sp)
 155    public:
 156     // Constructor
 157     UnrollBlock(int  size_of_deoptimized_frame,
 158                 int  caller_adjustment,
 159                 int  caller_actual_parameters,
 160                 int  number_of_frames,
 161                 intptr_t* frame_sizes,
 162                 address* frames_pcs,
 163                 BasicType return_type);
 164     ~UnrollBlock();
 165 
 166     // Returns where a register is located.
 167     intptr_t* value_addr_at(int register_number) const;
 168 
 169     // Accessors
 170     intptr_t* frame_sizes()  const { return _frame_sizes; }
 171     int number_of_frames()  const { return _number_of_frames; }
 172     address*  frame_pcs()   const { return _frame_pcs ; }
 173 
 174     // Returns the total size of frames
 175     int size_of_frames() const;
 176 
 177     void set_initial_info(intptr_t info) { _initial_info = info; }
 178 
 179     int caller_actual_parameters() const { return _caller_actual_parameters; }
 180 
 181     // Accessors used by the code generator for the unpack stub.
 182     static int size_of_deoptimized_frame_offset_in_bytes() { return offset_of(UnrollBlock, _size_of_deoptimized_frame); }
 183     static int caller_adjustment_offset_in_bytes()         { return offset_of(UnrollBlock, _caller_adjustment);         }
 184     static int number_of_frames_offset_in_bytes()          { return offset_of(UnrollBlock, _number_of_frames);          }
 185     static int frame_sizes_offset_in_bytes()               { return offset_of(UnrollBlock, _frame_sizes);               }
 186     static int total_frame_sizes_offset_in_bytes()         { return offset_of(UnrollBlock, _total_frame_sizes);         }
 187     static int frame_pcs_offset_in_bytes()                 { return offset_of(UnrollBlock, _frame_pcs);                 }
 188     static int register_block_offset_in_bytes()            { return offset_of(UnrollBlock, _register_block);            }
 189     static int return_type_offset_in_bytes()               { return offset_of(UnrollBlock, _return_type);               }
 190     static int counter_temp_offset_in_bytes()              { return offset_of(UnrollBlock, _counter_temp);              }
 191     static int initial_info_offset_in_bytes()              { return offset_of(UnrollBlock, _initial_info);              }
 192     static int unpack_kind_offset_in_bytes()               { return offset_of(UnrollBlock, _unpack_kind);               }
 193     static int sender_sp_temp_offset_in_bytes()            { return offset_of(UnrollBlock, _sender_sp_temp);            }
 194 
 195     BasicType return_type() const { return _return_type; }
 196     void print();
 197   };
 198 
 199   //** Returns an UnrollBlock continuing information
 200   // how to make room for the resulting interpreter frames.
 201   // Called by assembly stub after execution has returned to
 202   // deoptimized frame.
 203   // @argument thread.     Thread where stub_frame resides.
 204   // @see OptoRuntime::deoptimization_fetch_unroll_info_C
 205   static UnrollBlock* fetch_unroll_info(JavaThread* thread);
 206 
 207   //** Unpacks vframeArray onto execution stack
 208   // Called by assembly stub after execution has returned to
 209   // deoptimized frame and after the stack unrolling.
 210   // @argument thread.     Thread where stub_frame resides.
 211   // @argument exec_mode.  Determines how execution should be continuted in top frame.
 212   //                       0 means continue after current byte code
 213   //                       1 means exception has happened, handle exception
 214   //                       2 means reexecute current bytecode (for uncommon traps).
 215   // @see OptoRuntime::deoptimization_unpack_frames_C
 216   // Return BasicType of call return type, if any
 217   static BasicType unpack_frames(JavaThread* thread, int exec_mode);
 218 
 219   // Cleans up deoptimization bits on thread after unpacking or in the
 220   // case of an exception.
 221   static void cleanup_deopt_info(JavaThread  *thread,
 222                                  vframeArray * array);
 223 
 224   // Restores callee saved values from deoptimized frame into oldest interpreter frame
 225   // so caller of the deoptimized frame will get back the values it expects.
 226   static void unwind_callee_save_values(frame* f, vframeArray* vframe_array);
 227 
 228   //** Performs an uncommon trap for compiled code.
 229   // The top most compiler frame is converted into interpreter frames
 230   static UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index);
 231   // Helper routine that enters the VM and may block
 232   static void uncommon_trap_inner(JavaThread* thread, jint unloaded_class_index);
 233 
 234   //** Deoptimizes the frame identified by id.
 235   // Only called from VMDeoptimizeFrame
 236   // @argument thread.     Thread where stub_frame resides.
 237   // @argument id.         id of frame that should be deoptimized.
 238   static void deoptimize_frame_internal(JavaThread* thread, intptr_t* id);
 239 
 240   // If thread is not the current thread then execute
 241   // VM_DeoptimizeFrame otherwise deoptimize directly.
 242   static void deoptimize_frame(JavaThread* thread, intptr_t* id);
 243 
 244   // Statistics
 245   static void gather_statistics(DeoptReason reason, DeoptAction action,
 246                                 Bytecodes::Code bc = Bytecodes::_illegal);
 247   static void print_statistics();
 248 
 249   // How much room to adjust the last frame's SP by, to make space for
 250   // the callee's interpreter frame (which expects locals to be next to
 251   // incoming arguments)
 252   static int last_frame_adjust(int callee_parameters, int callee_locals);
 253 
 254   // trap_request codes
 255   static DeoptReason trap_request_reason(int trap_request) {
 256     if (trap_request < 0)
 257       return (DeoptReason)
 258         ((~(trap_request) >> _reason_shift) & right_n_bits(_reason_bits));
 259     else
 260       // standard reason for unloaded CP entry
 261       return Reason_unloaded;
 262   }
 263   static DeoptAction trap_request_action(int trap_request) {
 264     if (trap_request < 0)
 265       return (DeoptAction)
 266         ((~(trap_request) >> _action_shift) & right_n_bits(_action_bits));
 267     else
 268       // standard action for unloaded CP entry
 269       return _unloaded_action;
 270   }
 271   static int trap_request_index(int trap_request) {
 272     if (trap_request < 0)
 273       return -1;
 274     else
 275       return trap_request;
 276   }
 277   static int make_trap_request(DeoptReason reason, DeoptAction action,
 278                                int index = -1) {
 279     assert((1 << _reason_bits) >= Reason_LIMIT, "enough bits");
 280     assert((1 << _action_bits) >= Action_LIMIT, "enough bits");
 281     int trap_request;
 282     if (index != -1)
 283       trap_request = index;
 284     else
 285       trap_request = (~(((reason) << _reason_shift)
 286                         + ((action) << _action_shift)));
 287     assert(reason == trap_request_reason(trap_request), "valid reason");
 288     assert(action == trap_request_action(trap_request), "valid action");
 289     assert(index  == trap_request_index(trap_request),  "valid index");
 290     return trap_request;
 291   }
 292 
 293   // The trap_state stored in a MDO is decoded here.
 294   // It records two items of information.
 295   //  reason:  If a deoptimization happened here, what its reason was,
 296   //           or if there were multiple deopts with differing reasons.
 297   //  recompiled: If a deoptimization here triggered a recompilation.
 298   // Note that not all reasons are recorded per-bci.
 299   static DeoptReason trap_state_reason(int trap_state);
 300   static int  trap_state_has_reason(int trap_state, int reason);
 301   static int  trap_state_add_reason(int trap_state, int reason);
 302   static bool trap_state_is_recompiled(int trap_state);
 303   static int  trap_state_set_recompiled(int trap_state, bool z);
 304   static const char* format_trap_state(char* buf, size_t buflen,
 305                                        int trap_state);
 306 
 307   static bool reason_is_recorded_per_bytecode(DeoptReason reason) {
 308     return reason > Reason_none && reason <= Reason_RECORDED_LIMIT;
 309   }
 310 
 311   static DeoptReason reason_recorded_per_bytecode_if_any(DeoptReason reason) {
 312     if (reason_is_recorded_per_bytecode(reason))
 313       return reason;
 314     else if (reason == Reason_div0_check) // null check due to divide-by-zero?
 315       return Reason_null_check;           // recorded per BCI as a null check
 316     else if (reason == Reason_speculate_class_check)
 317       return Reason_class_check;
 318     else
 319       return Reason_none;
 320   }
 321 
 322   static bool reason_is_speculate(int reason) {
 323     if (reason == Reason_speculate_class_check) {
 324       return true;
 325     }
 326     return false;
 327   }
 328 
 329   static uint per_method_trap_limit(int reason) {
 330     return reason_is_speculate(reason) ? (uint)PerMethodSpecTrapLimit : (uint)PerMethodTrapLimit;
 331   }
 332 
 333   static const char* trap_reason_name(int reason);
 334   static const char* trap_action_name(int action);
 335   // Format like reason='foo' action='bar' index='123'.
 336   // This is suitable both for XML and for tty output.
 337   static const char* format_trap_request(char* buf, size_t buflen,
 338                                          int trap_request);
 339 
 340   static jint total_deoptimization_count();
 341   static jint deoptimization_count(DeoptReason reason);
 342 
 343   // JVMTI PopFrame support
 344 
 345   // Preserves incoming arguments to the popped frame when it is
 346   // returning to a deoptimized caller
 347   static void popframe_preserve_args(JavaThread* thread, int bytes_to_save, void* start_address);
 348 
 349  private:
 350   static MethodData* get_method_data(JavaThread* thread, methodHandle m, bool create_if_missing);
 351   // Update the mdo's count and per-BCI reason bits, returning previous state:
 352   static ProfileData* query_update_method_data(MethodData* trap_mdo,
 353                                                int trap_bci,
 354                                                DeoptReason reason,
 355                                                Method* compiled_method,
 356                                                //outputs:
 357                                                uint& ret_this_trap_count,
 358                                                bool& ret_maybe_prior_trap,
 359                                                bool& ret_maybe_prior_recompile);
 360   // class loading support for uncommon trap
 361   static void load_class_by_index(constantPoolHandle constant_pool, int index, TRAPS);
 362   static void load_class_by_index(constantPoolHandle constant_pool, int index);
 363 
 364   static UnrollBlock* fetch_unroll_info_helper(JavaThread* thread);
 365 
 366   static DeoptAction _unloaded_action; // == Action_reinterpret;
 367   static const char* _trap_reason_name[Reason_LIMIT];
 368   static const char* _trap_action_name[Action_LIMIT];
 369 
 370   static juint _deoptimization_hist[Reason_LIMIT][1+Action_LIMIT][BC_CASE_LIMIT];
 371   // Note:  Histogram array size is 1-2 Kb.
 372 
 373  public:
 374   static void update_method_data_from_interpreter(MethodData* trap_mdo, int trap_bci, int reason);
 375 };
 376 
 377 class DeoptimizationMarker : StackObj {  // for profiling
 378   static bool _is_active;
 379 public:
 380   DeoptimizationMarker()  { _is_active = true; }
 381   ~DeoptimizationMarker() { _is_active = false; }
 382   static bool is_active() { return _is_active; }
 383 };
 384 
 385 #endif // SHARE_VM_RUNTIME_DEOPTIMIZATION_HPP
--- EOF ---