src/share/vm/runtime/deoptimization.hpp

Thu, 24 Nov 2016 11:27:57 +0100

author
tschatzl
date
Thu, 24 Nov 2016 11:27:57 +0100
changeset 9982
72053ed6f8d4
parent 7419
d3f3f7677537
child 7535
7ae4e26cb1e0
permissions
-rw-r--r--

8057003: Large reference arrays cause extremely long synchronization times
Summary: Slice large object arrays into parts so that the synchronization of marking threads with an STW pause request does not take long.
Reviewed-by: ehelin, pliden
Contributed-by: maoliang.ml@alibaba-inc.com

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

mercurial