1 /*
  2  * Copyright (c) 1997, 2021, 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 "interpreter/bytecodes.hpp"
 29 #include "memory/allocation.hpp"
 30 #include "runtime/frame.hpp"
 31 
 32 class ProfileData;
 33 class vframeArray;
 34 class MonitorInfo;
 35 class MonitorValue;
 36 class ObjectValue;
 37 class AutoBoxObjectValue;
 38 class ScopeValue;
 39 class compiledVFrame;
 40 
 41 template<class E> class GrowableArray;
 42 
 43 class Deoptimization : AllStatic {
 44   friend class VMStructs;
 45   friend class EscapeBarrier;
 46 
 47  public:
 48   // What condition caused the deoptimization?
 49   enum DeoptReason {
 50     Reason_many = -1,             // indicates presence of several reasons
 51     Reason_none = 0,              // indicates absence of a relevant deopt.
 52     // Next 8 reasons are recorded per bytecode in DataLayout::trap_bits.
 53     // This is more complicated for JVMCI as JVMCI may deoptimize to *some* bytecode before the
 54     // bytecode that actually caused the deopt (with inlining, JVMCI may even deoptimize to a
 55     // bytecode in another method):
 56     //  - bytecode y in method b() causes deopt
 57     //  - JVMCI deoptimizes to bytecode x in method a()
 58     // -> the deopt reason will be recorded for method a() at bytecode x
 59     Reason_null_check,            // saw unexpected null or zero divisor (@bci)
 60     Reason_null_assert,           // saw unexpected non-null or non-zero (@bci)
 61     Reason_range_check,           // saw unexpected array index (@bci)
 62     Reason_class_check,           // saw unexpected object class (@bci)
 63     Reason_array_check,           // saw unexpected array class (aastore @bci)
 64     Reason_intrinsic,             // saw unexpected operand to intrinsic (@bci)
 65     Reason_bimorphic,             // saw unexpected object class in bimorphic inlining (@bci)
 66 
 67 #if INCLUDE_JVMCI
 68     Reason_unreached0             = Reason_null_assert,
 69     Reason_type_checked_inlining  = Reason_intrinsic,
 70     Reason_optimized_type_check   = Reason_bimorphic,
 71 #endif
 72 
 73     Reason_profile_predicate,     // compiler generated predicate moved from frequent branch in a loop failed
 74 
 75     // recorded per method
 76     Reason_unloaded,              // unloaded class or constant pool entry
 77     Reason_uninitialized,         // bad class state (uninitialized)
 78     Reason_initialized,           // class has been fully initialized
 79     Reason_unreached,             // code is not reached, compiler
 80     Reason_unhandled,             // arbitrary compiler limitation
 81     Reason_constraint,            // arbitrary runtime constraint violated
 82     Reason_div0_check,            // a null_check due to division by zero
 83     Reason_age,                   // nmethod too old; tier threshold reached
 84     Reason_predicate,             // compiler generated predicate failed
 85     Reason_loop_limit_check,      // compiler generated loop limits check failed
 86     Reason_speculate_class_check, // saw unexpected object class from type speculation
 87     Reason_speculate_null_check,  // saw unexpected null from type speculation
 88     Reason_speculate_null_assert, // saw unexpected null from type speculation
 89     Reason_rtm_state_change,      // rtm state change detected
 90     Reason_unstable_if,           // a branch predicted always false was taken
 91     Reason_unstable_fused_if,     // fused two ifs that had each one untaken branch. One is now taken.
 92     Reason_receiver_constraint,   // receiver subtype check failed
 93 #if INCLUDE_JVMCI
 94     Reason_aliasing,              // optimistic assumption about aliasing failed
 95     Reason_transfer_to_interpreter, // explicit transferToInterpreter()
 96     Reason_not_compiled_exception_handler,
 97     Reason_unresolved,
 98     Reason_jsr_mismatch,
 99 #endif
100 
101     // Reason_tenured is counted separately, add normal counted Reasons above.
102     // Related to MethodData::_trap_hist_limit where Reason_tenured isn't included
103     Reason_tenured,               // age of the code has reached the limit
104     Reason_LIMIT,
105 
106     // Note:  Keep this enum in sync. with _trap_reason_name.
107     Reason_RECORDED_LIMIT = Reason_profile_predicate  // some are not recorded per bc
108     // Note:  Reason_RECORDED_LIMIT should fit into 31 bits of
109     // DataLayout::trap_bits.  This dependency is enforced indirectly
110     // via asserts, to avoid excessive direct header-to-header dependencies.
111     // See Deoptimization::trap_state_reason and class DataLayout.
112   };
113 
114   // What action must be taken by the runtime?
115   enum DeoptAction {
116     Action_none,                  // just interpret, do not invalidate nmethod
117     Action_maybe_recompile,       // recompile the nmethod; need not invalidate
118     Action_reinterpret,           // invalidate the nmethod, reset IC, maybe recompile
119     Action_make_not_entrant,      // invalidate the nmethod, recompile (probably)
120     Action_make_not_compilable,   // invalidate the nmethod and do not compile
121     Action_LIMIT
122     // Note:  Keep this enum in sync. with _trap_action_name.
123   };
124 
125   enum {
126     _action_bits = 3,
127     _reason_bits = 5,
128     _debug_id_bits = 23,
129     _action_shift = 0,
130     _reason_shift = _action_shift+_action_bits,
131     _debug_id_shift = _reason_shift+_reason_bits,
132     BC_CASE_LIMIT = PRODUCT_ONLY(1) NOT_PRODUCT(4) // for _deoptimization_hist
133   };
134 
135   enum UnpackType {
136     Unpack_deopt                = 0, // normal deoptimization, use pc computed in unpack_vframe_on_stack
137     Unpack_exception            = 1, // exception is pending
138     Unpack_uncommon_trap        = 2, // redo last byte code (C2 only)
139     Unpack_reexecute            = 3, // reexecute bytecode (C1 only)
140     Unpack_none                 = 4, // not deoptimizing the frame, just reallocating/relocking for JVMTI
141     Unpack_LIMIT                = 5
142   };
143 
144 #if INCLUDE_JVMCI
145   // Can reconstruct virtualized unsafe large accesses to byte arrays.
146   static const int _support_large_access_byte_array_virtualization = 1;
147 #endif
148 
149   // Make all nmethods that are marked_for_deoptimization not_entrant and deoptimize any live
150   // activations using those nmethods.  If an nmethod is passed as an argument then it is
151   // marked_for_deoptimization and made not_entrant.  Otherwise a scan of the code cache is done to
152   // find all marked nmethods and they are made not_entrant.
153   static void deoptimize_all_marked(nmethod* nmethod_only = NULL);
154 
155  public:
156   // Deoptimizes a frame lazily. Deopt happens on return to the frame.
157   static void deoptimize(JavaThread* thread, frame fr, DeoptReason reason = Reason_constraint);
158 
159 #if INCLUDE_JVMCI
160   static address deoptimize_for_missing_exception_handler(CompiledMethod* cm);
161 #endif
162 
163   static oop get_cached_box(AutoBoxObjectValue* bv, frame* fr, RegisterMap* reg_map, TRAPS);
164 
165   private:
166   // Does the actual work for deoptimizing a single frame
167   static void deoptimize_single_frame(JavaThread* thread, frame fr, DeoptReason reason);
168 
169 #if COMPILER2_OR_JVMCI
170   // Deoptimize objects, that is reallocate and relock them, just before they
171   // escape through JVMTI.  The given vframes cover one physical frame.
172   static bool deoptimize_objects_internal(JavaThread* thread, GrowableArray<compiledVFrame*>* chunk,
173                                           bool& realloc_failures);
174 
175  public:
176 
177   // Support for restoring non-escaping objects
178   static bool realloc_objects(JavaThread* thread, frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, TRAPS);
179   static void reassign_type_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, typeArrayOop obj, BasicType type);
180   static void reassign_object_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, objArrayOop obj);
181   static void reassign_fields(frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, bool realloc_failures, bool skip_internal);
182   static bool relock_objects(JavaThread* thread, GrowableArray<MonitorInfo*>* monitors,
183                              JavaThread* deoptee_thread, frame& fr, int exec_mode, bool realloc_failures);
184   static void pop_frames_failed_reallocs(JavaThread* thread, vframeArray* array);
185 #endif // COMPILER2_OR_JVMCI
186 
187   public:
188   static vframeArray* create_vframeArray(JavaThread* thread, frame fr, RegisterMap *reg_map, GrowableArray<compiledVFrame*>* chunk, bool realloc_failures);
189 
190   // Interface used for unpacking deoptimized frames
191 
192   // UnrollBlock is returned by fetch_unroll_info() to the deoptimization handler (blob).
193   // This is only a CheapObj to ease debugging after a deopt failure
194   class UnrollBlock : public CHeapObj<mtCompiler> {
195     friend class VMStructs;
196     friend class JVMCIVMStructs;
197    private:
198     int       _size_of_deoptimized_frame; // Size, in bytes, of current deoptimized frame
199     int       _caller_adjustment;         // Adjustment, in bytes, to caller's SP by initial interpreted frame
200     int       _number_of_frames;          // Number frames to unroll
201     int       _total_frame_sizes;         // Total of number*sizes frames
202     intptr_t* _frame_sizes;               // Array of frame sizes, in bytes, for unrolling the stack
203     address*  _frame_pcs;                 // Array of frame pc's, in bytes, for unrolling the stack
204     intptr_t* _register_block;            // Block for storing callee-saved registers.
205     BasicType _return_type;               // Tells if we have to restore double or long return value
206     intptr_t  _initial_info;              // Platform dependent data for the sender frame (was FP on x86)
207     int       _caller_actual_parameters;  // The number of actual arguments at the
208                                           // interpreted caller of the deoptimized frame
209     int       _unpack_kind;               // exec_mode that can be changed during fetch_unroll_info
210 
211     // The following fields are used as temps during the unpacking phase
212     // (which is tight on registers, especially on x86). They really ought
213     // to be PD variables but that involves moving this class into its own
214     // file to use the pd include mechanism. Maybe in a later cleanup ...
215     intptr_t  _counter_temp;              // SHOULD BE PD VARIABLE (x86 frame count temp)
216     intptr_t  _sender_sp_temp;            // SHOULD BE PD VARIABLE (x86 sender_sp)
217    public:
218     // Constructor
219     UnrollBlock(int  size_of_deoptimized_frame,
220                 int  caller_adjustment,
221                 int  caller_actual_parameters,
222                 int  number_of_frames,
223                 intptr_t* frame_sizes,
224                 address* frames_pcs,
225                 BasicType return_type,
226                 int unpack_kind);
227     ~UnrollBlock();
228 
229     // Returns where a register is located.
230     intptr_t* value_addr_at(int register_number) const;
231 
232     // Accessors
233     intptr_t* frame_sizes()  const { return _frame_sizes; }
234     int number_of_frames()  const { return _number_of_frames; }
235     address*  frame_pcs()   const { return _frame_pcs ; }
236     int  unpack_kind()   const { return _unpack_kind; }
237 
238     // Returns the total size of frames
239     int size_of_frames() const;
240 
241     void set_initial_info(intptr_t info) { _initial_info = info; }
242 
243     int caller_actual_parameters() const { return _caller_actual_parameters; }
244 
245     // Accessors used by the code generator for the unpack stub.
246     static int size_of_deoptimized_frame_offset_in_bytes() { return offset_of(UnrollBlock, _size_of_deoptimized_frame); }
247     static int caller_adjustment_offset_in_bytes()         { return offset_of(UnrollBlock, _caller_adjustment);         }
248     static int number_of_frames_offset_in_bytes()          { return offset_of(UnrollBlock, _number_of_frames);          }
249     static int frame_sizes_offset_in_bytes()               { return offset_of(UnrollBlock, _frame_sizes);               }
250     static int total_frame_sizes_offset_in_bytes()         { return offset_of(UnrollBlock, _total_frame_sizes);         }
251     static int frame_pcs_offset_in_bytes()                 { return offset_of(UnrollBlock, _frame_pcs);                 }
252     static int register_block_offset_in_bytes()            { return offset_of(UnrollBlock, _register_block);            }
253     static int return_type_offset_in_bytes()               { return offset_of(UnrollBlock, _return_type);               }
254     static int counter_temp_offset_in_bytes()              { return offset_of(UnrollBlock, _counter_temp);              }
255     static int initial_info_offset_in_bytes()              { return offset_of(UnrollBlock, _initial_info);              }
256     static int unpack_kind_offset_in_bytes()               { return offset_of(UnrollBlock, _unpack_kind);               }
257     static int sender_sp_temp_offset_in_bytes()            { return offset_of(UnrollBlock, _sender_sp_temp);            }
258 
259     BasicType return_type() const { return _return_type; }
260     void print();
261   };
262 
263   //** Returns an UnrollBlock continuing information
264   // how to make room for the resulting interpreter frames.
265   // Called by assembly stub after execution has returned to
266   // deoptimized frame.
267   // @argument thread.     Thread where stub_frame resides.
268   // @see OptoRuntime::deoptimization_fetch_unroll_info_C
269   static UnrollBlock* fetch_unroll_info(JavaThread* current, int exec_mode);
270 
271   //** Unpacks vframeArray onto execution stack
272   // Called by assembly stub after execution has returned to
273   // deoptimized frame and after the stack unrolling.
274   // @argument thread.     Thread where stub_frame resides.
275   // @argument exec_mode.  Determines how execution should be continued in top frame.
276   //                       0 means continue after current byte code
277   //                       1 means exception has happened, handle exception
278   //                       2 means reexecute current bytecode (for uncommon traps).
279   // @see OptoRuntime::deoptimization_unpack_frames_C
280   // Return BasicType of call return type, if any
281   static BasicType unpack_frames(JavaThread* thread, int exec_mode);
282 
283   // Cleans up deoptimization bits on thread after unpacking or in the
284   // case of an exception.
285   static void cleanup_deopt_info(JavaThread  *thread,
286                                  vframeArray * array);
287 
288   // Restores callee saved values from deoptimized frame into oldest interpreter frame
289   // so caller of the deoptimized frame will get back the values it expects.
290   static void unwind_callee_save_values(frame* f, vframeArray* vframe_array);
291 
292   //** Performs an uncommon trap for compiled code.
293   // The top most compiler frame is converted into interpreter frames
294   static UnrollBlock* uncommon_trap(JavaThread* current, jint unloaded_class_index, jint exec_mode);
295   // Helper routine that enters the VM and may block
296   static void uncommon_trap_inner(JavaThread* current, jint unloaded_class_index);
297 
298   //** Deoptimizes the frame identified by id.
299   // Only called from VMDeoptimizeFrame
300   // @argument thread.     Thread where stub_frame resides.
301   // @argument id.         id of frame that should be deoptimized.
302   static void deoptimize_frame_internal(JavaThread* thread, intptr_t* id, DeoptReason reason);
303 
304   // if thread is not the current thread then execute
305   // VM_DeoptimizeFrame otherwise deoptimize directly.
306   static void deoptimize_frame(JavaThread* thread, intptr_t* id, DeoptReason reason);
307   static void deoptimize_frame(JavaThread* thread, intptr_t* id);
308 
309   // Statistics
310   static void gather_statistics(DeoptReason reason, DeoptAction action,
311                                 Bytecodes::Code bc = Bytecodes::_illegal);
312   static void print_statistics();
313 
314   // How much room to adjust the last frame's SP by, to make space for
315   // the callee's interpreter frame (which expects locals to be next to
316   // incoming arguments)
317   static int last_frame_adjust(int callee_parameters, int callee_locals);
318 
319   // trap_request codes
320   static DeoptReason trap_request_reason(int trap_request) {
321     if (trap_request < 0)
322       return (DeoptReason)
323         ((~(trap_request) >> _reason_shift) & right_n_bits(_reason_bits));
324     else
325       // standard reason for unloaded CP entry
326       return Reason_unloaded;
327   }
328   static DeoptAction trap_request_action(int trap_request) {
329     if (trap_request < 0)
330       return (DeoptAction)
331         ((~(trap_request) >> _action_shift) & right_n_bits(_action_bits));
332     else
333       // standard action for unloaded CP entry
334       return _unloaded_action;
335   }
336   static int trap_request_debug_id(int trap_request) {
337     if (trap_request < 0) {
338       return ((~(trap_request) >> _debug_id_shift) & right_n_bits(_debug_id_bits));
339     } else {
340       // standard action for unloaded CP entry
341       return 0;
342     }
343   }
344   static int trap_request_index(int trap_request) {
345     if (trap_request < 0)
346       return -1;
347     else
348       return trap_request;
349   }
350   static int make_trap_request(DeoptReason reason, DeoptAction action,
351                                int index = -1) {
352     assert((1 << _reason_bits) >= Reason_LIMIT, "enough bits");
353     assert((1 << _action_bits) >= Action_LIMIT, "enough bits");
354     int trap_request;
355     if (index != -1)
356       trap_request = index;
357     else
358       trap_request = (~(((reason) << _reason_shift)
359                         + ((action) << _action_shift)));
360     assert(reason == trap_request_reason(trap_request), "valid reason");
361     assert(action == trap_request_action(trap_request), "valid action");
362     assert(index  == trap_request_index(trap_request),  "valid index");
363     return trap_request;
364   }
365 
366   // The trap_state stored in a MDO is decoded here.
367   // It records two items of information.
368   //  reason:  If a deoptimization happened here, what its reason was,
369   //           or if there were multiple deopts with differing reasons.
370   //  recompiled: If a deoptimization here triggered a recompilation.
371   // Note that not all reasons are recorded per-bci.
372   static DeoptReason trap_state_reason(int trap_state);
373   static int  trap_state_has_reason(int trap_state, int reason);
374   static int  trap_state_add_reason(int trap_state, int reason);
375   static bool trap_state_is_recompiled(int trap_state);
376   static int  trap_state_set_recompiled(int trap_state, bool z);
377   static const char* format_trap_state(char* buf, size_t buflen,
378                                        int trap_state);
379 
380   static bool reason_is_recorded_per_bytecode(DeoptReason reason) {
381     return reason > Reason_none && reason <= Reason_RECORDED_LIMIT;
382   }
383 
384   static DeoptReason reason_recorded_per_bytecode_if_any(DeoptReason reason) {
385     if (reason_is_recorded_per_bytecode(reason))
386       return reason;
387     else if (reason == Reason_div0_check) // null check due to divide-by-zero?
388       return Reason_null_check;           // recorded per BCI as a null check
389     else if (reason == Reason_speculate_class_check)
390       return Reason_class_check;
391     else if (reason == Reason_speculate_null_check)
392       return Reason_null_check;
393     else if (reason == Reason_speculate_null_assert)
394       return Reason_null_assert;
395     else if (reason == Reason_unstable_if)
396       return Reason_intrinsic;
397     else if (reason == Reason_unstable_fused_if)
398       return Reason_range_check;
399     else
400       return Reason_none;
401   }
402 
403   static bool reason_is_speculate(int reason) {
404     if (reason == Reason_speculate_class_check ||
405         reason == Reason_speculate_null_check ||
406         reason == Reason_speculate_null_assert) {
407       return true;
408     }
409     return false;
410   }
411 
412   static DeoptReason reason_null_check(bool speculative) {
413     return speculative ? Deoptimization::Reason_speculate_null_check : Deoptimization::Reason_null_check;
414   }
415 
416   static DeoptReason reason_class_check(bool speculative) {
417     return speculative ? Deoptimization::Reason_speculate_class_check : Deoptimization::Reason_class_check;
418   }
419 
420   static DeoptReason reason_null_assert(bool speculative) {
421     return speculative ? Deoptimization::Reason_speculate_null_assert : Deoptimization::Reason_null_assert;
422   }
423 
424   static uint per_method_trap_limit(int reason) {
425     return reason_is_speculate(reason) ? (uint)PerMethodSpecTrapLimit : (uint)PerMethodTrapLimit;
426   }
427 
428   static const char* trap_reason_name(int reason);
429   static const char* trap_action_name(int action);
430   // Format like reason='foo' action='bar' index='123'.
431   // This is suitable both for XML and for tty output.
432   static const char* format_trap_request(char* buf, size_t buflen,
433                                          int trap_request);
434 
435   static jint total_deoptimization_count();
436 
437   // JVMTI PopFrame support
438 
439   // Preserves incoming arguments to the popped frame when it is
440   // returning to a deoptimized caller
441   static void popframe_preserve_args(JavaThread* thread, int bytes_to_save, void* start_address);
442 
443   static MethodData* get_method_data(JavaThread* thread, const methodHandle& m, bool create_if_missing);
444  private:
445   // Update the mdo's count and per-BCI reason bits, returning previous state:
446   static ProfileData* query_update_method_data(MethodData* trap_mdo,
447                                                int trap_bci,
448                                                DeoptReason reason,
449                                                bool update_total_trap_count,
450 #if INCLUDE_JVMCI
451                                                bool is_osr,
452 #endif
453                                                Method* compiled_method,
454                                                //outputs:
455                                                uint& ret_this_trap_count,
456                                                bool& ret_maybe_prior_trap,
457                                                bool& ret_maybe_prior_recompile);
458   // class loading support for uncommon trap
459   static void load_class_by_index(const constantPoolHandle& constant_pool, int index, TRAPS);
460 
461   static UnrollBlock* fetch_unroll_info_helper(JavaThread* current, int exec_mode);
462 
463   static DeoptAction _unloaded_action; // == Action_reinterpret;
464   static const char* _trap_reason_name[];
465   static const char* _trap_action_name[];
466 
467   static juint _deoptimization_hist[Reason_LIMIT][1+Action_LIMIT][BC_CASE_LIMIT];
468   // Note:  Histogram array size is 1-2 Kb.
469 
470  public:
471   static void update_method_data_from_interpreter(MethodData* trap_mdo, int trap_bci, int reason);
472 };
473 
474 
475 class DeoptimizationMarker : StackObj {  // for profiling
476   static bool _is_active;
477 public:
478   DeoptimizationMarker()  { _is_active = true; }
479   ~DeoptimizationMarker() { _is_active = false; }
480   static bool is_active() { return _is_active; }
481 };
482 
483 #endif // SHARE_RUNTIME_DEOPTIMIZATION_HPP