1 /*
  2  * Copyright (c) 1997, 2022, 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   // Note: Keep this enum in sync. with Deoptimization::_trap_reason_name.
 50   enum DeoptReason {
 51     Reason_many = -1,             // indicates presence of several reasons
 52     Reason_none = 0,              // indicates absence of a relevant deopt.
 53     // Next 8 reasons are recorded per bytecode in DataLayout::trap_bits.
 54     // This is more complicated for JVMCI as JVMCI may deoptimize to *some* bytecode before the
 55     // bytecode that actually caused the deopt (with inlining, JVMCI may even deoptimize to a
 56     // bytecode in another method):
 57     //  - bytecode y in method b() causes deopt
 58     //  - JVMCI deoptimizes to bytecode x in method a()
 59     // -> the deopt reason will be recorded for method a() at bytecode x
 60     Reason_null_check,            // saw unexpected null or zero divisor (@bci)
 61     Reason_null_assert,           // saw unexpected non-null or non-zero (@bci)
 62     Reason_range_check,           // saw unexpected array index (@bci)
 63     Reason_class_check,           // saw unexpected object class (@bci)
 64     Reason_array_check,           // saw unexpected array class (aastore @bci)
 65     Reason_intrinsic,             // saw unexpected operand to intrinsic (@bci)
 66     Reason_bimorphic,             // saw unexpected object class in bimorphic inlining (@bci)
 67 
 68 #if INCLUDE_JVMCI
 69     Reason_unreached0             = Reason_null_assert,
 70     Reason_type_checked_inlining  = Reason_intrinsic,
 71     Reason_optimized_type_check   = Reason_bimorphic,
 72 #endif
 73 
 74     Reason_profile_predicate,     // compiler generated predicate moved from frequent branch in a loop failed
 75 
 76     // recorded per method
 77     Reason_unloaded,              // unloaded class or constant pool entry
 78     Reason_uninitialized,         // bad class state (uninitialized)
 79     Reason_initialized,           // class has been fully initialized
 80     Reason_unreached,             // code is not reached, compiler
 81     Reason_unhandled,             // arbitrary compiler limitation
 82     Reason_constraint,            // arbitrary runtime constraint violated
 83     Reason_div0_check,            // a null_check due to division by zero
 84     Reason_age,                   // nmethod too old; tier threshold reached
 85     Reason_predicate,             // compiler generated predicate failed
 86     Reason_loop_limit_check,      // compiler generated loop limits check failed
 87     Reason_speculate_class_check, // saw unexpected object class from type speculation
 88     Reason_speculate_null_check,  // saw unexpected null from type speculation
 89     Reason_speculate_null_assert, // saw unexpected null from type speculation
 90     Reason_rtm_state_change,      // rtm state change detected
 91     Reason_unstable_if,           // a branch predicted always false was taken
 92     Reason_unstable_fused_if,     // fused two ifs that had each one untaken branch. One is now taken.
 93     Reason_receiver_constraint,   // receiver subtype check failed
 94 #if INCLUDE_JVMCI
 95     Reason_aliasing,              // optimistic assumption about aliasing failed
 96     Reason_transfer_to_interpreter, // explicit transferToInterpreter()
 97     Reason_not_compiled_exception_handler,
 98     Reason_unresolved,
 99     Reason_jsr_mismatch,
100 #endif
101 
102     // Used to define MethodData::_trap_hist_limit where Reason_tenured isn't included
103     Reason_TRAP_HISTORY_LENGTH,
104 
105     // Reason_tenured is counted separately, add normal counted Reasons above.
106     Reason_tenured = Reason_TRAP_HISTORY_LENGTH, // age of the code has reached the limit
107     Reason_LIMIT,
108 
109     // Note:  Reason_RECORDED_LIMIT should fit into 31 bits of
110     // DataLayout::trap_bits.  This dependency is enforced indirectly
111     // via asserts, to avoid excessive direct header-to-header dependencies.
112     // See Deoptimization::trap_state_reason and class DataLayout.
113     Reason_RECORDED_LIMIT = Reason_profile_predicate,  // some are not recorded per bc
114   };
115 
116   // What action must be taken by the runtime?
117   // Note: Keep this enum in sync. with Deoptimization::_trap_action_name.
118   enum DeoptAction {
119     Action_none,                  // just interpret, do not invalidate nmethod
120     Action_maybe_recompile,       // recompile the nmethod; need not invalidate
121     Action_reinterpret,           // invalidate the nmethod, reset IC, maybe recompile
122     Action_make_not_entrant,      // invalidate the nmethod, recompile (probably)
123     Action_make_not_compilable,   // invalidate the nmethod and do not compile
124     Action_LIMIT
125   };
126 
127   enum {
128     _action_bits = 3,
129     _reason_bits = 5,
130     _debug_id_bits = 23,
131     _action_shift = 0,
132     _reason_shift = _action_shift+_action_bits,
133     _debug_id_shift = _reason_shift+_reason_bits,
134     BC_CASE_LIMIT = PRODUCT_ONLY(1) NOT_PRODUCT(4) // for _deoptimization_hist
135   };
136 
137   enum UnpackType {
138     Unpack_deopt                = 0, // normal deoptimization, use pc computed in unpack_vframe_on_stack
139     Unpack_exception            = 1, // exception is pending
140     Unpack_uncommon_trap        = 2, // redo last byte code (C2 only)
141     Unpack_reexecute            = 3, // reexecute bytecode (C1 only)
142     Unpack_none                 = 4, // not deoptimizing the frame, just reallocating/relocking for JVMTI
143     Unpack_LIMIT                = 5
144   };
145 
146 #if INCLUDE_JVMCI
147   // Can reconstruct virtualized unsafe large accesses to byte arrays.
148   static const int _support_large_access_byte_array_virtualization = 1;
149 #endif
150 
151   // Make all nmethods that are marked_for_deoptimization not_entrant and deoptimize any live
152   // activations using those nmethods.  If an nmethod is passed as an argument then it is
153   // marked_for_deoptimization and made not_entrant.  Otherwise a scan of the code cache is done to
154   // find all marked nmethods and they are made not_entrant.
155   static void deoptimize_all_marked(nmethod* nmethod_only = NULL);
156 
157  public:
158   // Deoptimizes a frame lazily. Deopt happens on return to the frame.
159   static void deoptimize(JavaThread* thread, frame fr, DeoptReason reason = Reason_constraint);
160 
161 #if INCLUDE_JVMCI
162   static address deoptimize_for_missing_exception_handler(CompiledMethod* cm);
163   static oop get_cached_box(AutoBoxObjectValue* bv, frame* fr, RegisterMap* reg_map, TRAPS);
164 #endif
165 
166   private:
167   // Does the actual work for deoptimizing a single frame
168   static void deoptimize_single_frame(JavaThread* thread, frame fr, DeoptReason reason);
169 
170 #if COMPILER2_OR_JVMCI
171   // Deoptimize objects, that is reallocate and relock them, just before they
172   // escape through JVMTI.  The given vframes cover one physical frame.
173   static bool deoptimize_objects_internal(JavaThread* thread, GrowableArray<compiledVFrame*>* chunk,
174                                           bool& realloc_failures);
175 
176  public:
177 
178   // Support for restoring non-escaping objects
179   static bool realloc_objects(JavaThread* thread, frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, TRAPS);
180   static bool realloc_inline_type_result(InlineKlass* vk, const RegisterMap& map, GrowableArray<Handle>& return_oops, TRAPS);
181   static void reassign_type_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, typeArrayOop obj, BasicType type);
182   static void reassign_object_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, objArrayOop obj);
183   static void reassign_flat_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, flatArrayOop obj, FlatArrayKlass* vak, bool skip_internal, TRAPS);
184   static void reassign_fields(frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, bool realloc_failures, bool skip_internal, TRAPS);
185   static bool relock_objects(JavaThread* thread, GrowableArray<MonitorInfo*>* monitors,
186                              JavaThread* deoptee_thread, frame& fr, int exec_mode, bool realloc_failures);
187   static void pop_frames_failed_reallocs(JavaThread* thread, vframeArray* array);
188 #endif // COMPILER2_OR_JVMCI
189 
190   public:
191   static vframeArray* create_vframeArray(JavaThread* thread, frame fr, RegisterMap *reg_map, GrowableArray<compiledVFrame*>* chunk, bool realloc_failures);
192 
193   // Interface used for unpacking deoptimized frames
194 
195   // UnrollBlock is returned by fetch_unroll_info() to the deoptimization handler (blob).
196   // This is only a CheapObj to ease debugging after a deopt failure
197   class UnrollBlock : public CHeapObj<mtCompiler> {
198     friend class VMStructs;
199     friend class JVMCIVMStructs;
200    private:
201     int       _size_of_deoptimized_frame; // Size, in bytes, of current deoptimized frame
202     int       _caller_adjustment;         // Adjustment, in bytes, to caller's SP by initial interpreted frame
203     int       _number_of_frames;          // Number frames to unroll
204     int       _total_frame_sizes;         // Total of number*sizes frames
205     intptr_t* _frame_sizes;               // Array of frame sizes, in bytes, for unrolling the stack
206     address*  _frame_pcs;                 // Array of frame pc's, in bytes, for unrolling the stack
207     intptr_t* _register_block;            // Block for storing callee-saved registers.
208     BasicType _return_type;               // Tells if we have to restore double or long return value
209     intptr_t  _initial_info;              // Platform dependent data for the sender frame (was FP on x86)
210     int       _caller_actual_parameters;  // The number of actual arguments at the
211                                           // interpreted caller of the deoptimized frame
212     int       _unpack_kind;               // exec_mode that can be changed during fetch_unroll_info
213 
214     // The following fields are used as temps during the unpacking phase
215     // (which is tight on registers, especially on x86). They really ought
216     // to be PD variables but that involves moving this class into its own
217     // file to use the pd include mechanism. Maybe in a later cleanup ...
218     intptr_t  _counter_temp;              // SHOULD BE PD VARIABLE (x86 frame count temp)
219     intptr_t  _sender_sp_temp;            // SHOULD BE PD VARIABLE (x86 sender_sp)
220    public:
221     // Constructor
222     UnrollBlock(int  size_of_deoptimized_frame,
223                 int  caller_adjustment,
224                 int  caller_actual_parameters,
225                 int  number_of_frames,
226                 intptr_t* frame_sizes,
227                 address* frames_pcs,
228                 BasicType return_type,
229                 int unpack_kind);
230     ~UnrollBlock();
231 
232     // Returns where a register is located.
233     intptr_t* value_addr_at(int register_number) const;
234 
235     // Accessors
236     intptr_t* frame_sizes()  const { return _frame_sizes; }
237     int number_of_frames()  const { return _number_of_frames; }
238     address*  frame_pcs()   const { return _frame_pcs ; }
239     int  unpack_kind()   const { return _unpack_kind; }
240 
241     // Returns the total size of frames
242     int size_of_frames() const;
243 
244     void set_initial_info(intptr_t info) { _initial_info = info; }
245 
246     int caller_actual_parameters() const { return _caller_actual_parameters; }
247 
248     // Accessors used by the code generator for the unpack stub.
249     static int size_of_deoptimized_frame_offset_in_bytes() { return offset_of(UnrollBlock, _size_of_deoptimized_frame); }
250     static int caller_adjustment_offset_in_bytes()         { return offset_of(UnrollBlock, _caller_adjustment);         }
251     static int number_of_frames_offset_in_bytes()          { return offset_of(UnrollBlock, _number_of_frames);          }
252     static int frame_sizes_offset_in_bytes()               { return offset_of(UnrollBlock, _frame_sizes);               }
253     static int total_frame_sizes_offset_in_bytes()         { return offset_of(UnrollBlock, _total_frame_sizes);         }
254     static int frame_pcs_offset_in_bytes()                 { return offset_of(UnrollBlock, _frame_pcs);                 }
255     static int register_block_offset_in_bytes()            { return offset_of(UnrollBlock, _register_block);            }
256     static int return_type_offset_in_bytes()               { return offset_of(UnrollBlock, _return_type);               }
257     static int counter_temp_offset_in_bytes()              { return offset_of(UnrollBlock, _counter_temp);              }
258     static int initial_info_offset_in_bytes()              { return offset_of(UnrollBlock, _initial_info);              }
259     static int unpack_kind_offset_in_bytes()               { return offset_of(UnrollBlock, _unpack_kind);               }
260     static int sender_sp_temp_offset_in_bytes()            { return offset_of(UnrollBlock, _sender_sp_temp);            }
261 
262     BasicType return_type() const { return _return_type; }
263     void print();
264   };
265 
266   //** Returns an UnrollBlock continuing information
267   // how to make room for the resulting interpreter frames.
268   // Called by assembly stub after execution has returned to
269   // deoptimized frame.
270   // @argument thread.     Thread where stub_frame resides.
271   // @see OptoRuntime::deoptimization_fetch_unroll_info_C
272   static UnrollBlock* fetch_unroll_info(JavaThread* current, int exec_mode);
273 
274   //** Unpacks vframeArray onto execution stack
275   // Called by assembly stub after execution has returned to
276   // deoptimized frame and after the stack unrolling.
277   // @argument thread.     Thread where stub_frame resides.
278   // @argument exec_mode.  Determines how execution should be continued in top frame.
279   //                       0 means continue after current byte code
280   //                       1 means exception has happened, handle exception
281   //                       2 means reexecute current bytecode (for uncommon traps).
282   // @see OptoRuntime::deoptimization_unpack_frames_C
283   // Return BasicType of call return type, if any
284   static BasicType unpack_frames(JavaThread* thread, int exec_mode);
285 
286   // Cleans up deoptimization bits on thread after unpacking or in the
287   // case of an exception.
288   static void cleanup_deopt_info(JavaThread  *thread,
289                                  vframeArray * array);
290 
291   // Restores callee saved values from deoptimized frame into oldest interpreter frame
292   // so caller of the deoptimized frame will get back the values it expects.
293   static void unwind_callee_save_values(frame* f, vframeArray* vframe_array);
294 
295   //** Performs an uncommon trap for compiled code.
296   // The top most compiler frame is converted into interpreter frames
297   static UnrollBlock* uncommon_trap(JavaThread* current, jint unloaded_class_index, jint exec_mode);
298   // Helper routine that enters the VM and may block
299   static void uncommon_trap_inner(JavaThread* current, jint unloaded_class_index);
300 
301   //** Deoptimizes the frame identified by id.
302   // Only called from VMDeoptimizeFrame
303   // @argument thread.     Thread where stub_frame resides.
304   // @argument id.         id of frame that should be deoptimized.
305   static void deoptimize_frame_internal(JavaThread* thread, intptr_t* id, DeoptReason reason);
306 
307   // if thread is not the current thread then execute
308   // VM_DeoptimizeFrame otherwise deoptimize directly.
309   static void deoptimize_frame(JavaThread* thread, intptr_t* id, DeoptReason reason);
310   static void deoptimize_frame(JavaThread* thread, intptr_t* id);
311 
312   // Statistics
313   static void gather_statistics(DeoptReason reason, DeoptAction action,
314                                 Bytecodes::Code bc = Bytecodes::_illegal);
315   static void print_statistics();
316 
317   // How much room to adjust the last frame's SP by, to make space for
318   // the callee's interpreter frame (which expects locals to be next to
319   // incoming arguments)
320   static int last_frame_adjust(int callee_parameters, int callee_locals);
321 
322   // trap_request codes
323   static DeoptReason trap_request_reason(int trap_request) {
324     if (trap_request < 0)
325       return (DeoptReason)
326         ((~(trap_request) >> _reason_shift) & right_n_bits(_reason_bits));
327     else
328       // standard reason for unloaded CP entry
329       return Reason_unloaded;
330   }
331   static DeoptAction trap_request_action(int trap_request) {
332     if (trap_request < 0)
333       return (DeoptAction)
334         ((~(trap_request) >> _action_shift) & right_n_bits(_action_bits));
335     else
336       // standard action for unloaded CP entry
337       return _unloaded_action;
338   }
339   static int trap_request_debug_id(int trap_request) {
340     if (trap_request < 0) {
341       return ((~(trap_request) >> _debug_id_shift) & right_n_bits(_debug_id_bits));
342     } else {
343       // standard action for unloaded CP entry
344       return 0;
345     }
346   }
347   static int trap_request_index(int trap_request) {
348     if (trap_request < 0)
349       return -1;
350     else
351       return trap_request;
352   }
353   static int make_trap_request(DeoptReason reason, DeoptAction action,
354                                int index = -1) {
355     assert((1 << _reason_bits) >= Reason_LIMIT, "enough bits");
356     assert((1 << _action_bits) >= Action_LIMIT, "enough bits");
357     int trap_request;
358     if (index != -1)
359       trap_request = index;
360     else
361       trap_request = (~(((reason) << _reason_shift)
362                         + ((action) << _action_shift)));
363     assert(reason == trap_request_reason(trap_request), "valid reason");
364     assert(action == trap_request_action(trap_request), "valid action");
365     assert(index  == trap_request_index(trap_request),  "valid index");
366     return trap_request;
367   }
368 
369   // The trap_state stored in a MDO is decoded here.
370   // It records two items of information.
371   //  reason:  If a deoptimization happened here, what its reason was,
372   //           or if there were multiple deopts with differing reasons.
373   //  recompiled: If a deoptimization here triggered a recompilation.
374   // Note that not all reasons are recorded per-bci.
375   static DeoptReason trap_state_reason(int trap_state);
376   static int  trap_state_has_reason(int trap_state, int reason);
377   static int  trap_state_add_reason(int trap_state, int reason);
378   static bool trap_state_is_recompiled(int trap_state);
379   static int  trap_state_set_recompiled(int trap_state, bool z);
380   static const char* format_trap_state(char* buf, size_t buflen,
381                                        int trap_state);
382 
383   static bool reason_is_recorded_per_bytecode(DeoptReason reason) {
384     return reason > Reason_none && reason <= Reason_RECORDED_LIMIT;
385   }
386 
387   static DeoptReason reason_recorded_per_bytecode_if_any(DeoptReason reason) {
388     if (reason_is_recorded_per_bytecode(reason))
389       return reason;
390     else if (reason == Reason_div0_check) // null check due to divide-by-zero?
391       return Reason_null_check;           // recorded per BCI as a null check
392     else if (reason == Reason_speculate_class_check)
393       return Reason_class_check;
394     else if (reason == Reason_speculate_null_check)
395       return Reason_null_check;
396     else if (reason == Reason_speculate_null_assert)
397       return Reason_null_assert;
398     else if (reason == Reason_unstable_if)
399       return Reason_intrinsic;
400     else if (reason == Reason_unstable_fused_if)
401       return Reason_range_check;
402     else
403       return Reason_none;
404   }
405 
406   static bool reason_is_speculate(int reason) {
407     if (reason == Reason_speculate_class_check ||
408         reason == Reason_speculate_null_check ||
409         reason == Reason_speculate_null_assert) {
410       return true;
411     }
412     return false;
413   }
414 
415   static DeoptReason reason_null_check(bool speculative) {
416     return speculative ? Deoptimization::Reason_speculate_null_check : Deoptimization::Reason_null_check;
417   }
418 
419   static DeoptReason reason_class_check(bool speculative) {
420     return speculative ? Deoptimization::Reason_speculate_class_check : Deoptimization::Reason_class_check;
421   }
422 
423   static DeoptReason reason_null_assert(bool speculative) {
424     return speculative ? Deoptimization::Reason_speculate_null_assert : Deoptimization::Reason_null_assert;
425   }
426 
427   static uint per_method_trap_limit(int reason) {
428     return reason_is_speculate(reason) ? (uint)PerMethodSpecTrapLimit : (uint)PerMethodTrapLimit;
429   }
430 
431   static const char* trap_reason_name(int reason);
432   static const char* trap_action_name(int action);
433   // Format like reason='foo' action='bar' index='123'.
434   // This is suitable both for XML and for tty output.
435   static const char* format_trap_request(char* buf, size_t buflen,
436                                          int trap_request);
437 
438   static jint total_deoptimization_count();
439   static jint deoptimization_count(const char* reason_str, const char* action_str);
440 
441   // JVMTI PopFrame support
442 
443   // Preserves incoming arguments to the popped frame when it is
444   // returning to a deoptimized caller
445   static void popframe_preserve_args(JavaThread* thread, int bytes_to_save, void* start_address);
446 
447   static MethodData* get_method_data(JavaThread* thread, const methodHandle& m, bool create_if_missing);
448  private:
449   // Update the mdo's count and per-BCI reason bits, returning previous state:
450   static ProfileData* query_update_method_data(MethodData* trap_mdo,
451                                                int trap_bci,
452                                                DeoptReason reason,
453                                                bool update_total_trap_count,
454 #if INCLUDE_JVMCI
455                                                bool is_osr,
456 #endif
457                                                Method* compiled_method,
458                                                //outputs:
459                                                uint& ret_this_trap_count,
460                                                bool& ret_maybe_prior_trap,
461                                                bool& ret_maybe_prior_recompile);
462   // class loading support for uncommon trap
463   static void load_class_by_index(const constantPoolHandle& constant_pool, int index, TRAPS);
464 
465   static UnrollBlock* fetch_unroll_info_helper(JavaThread* current, int exec_mode);
466 
467   static DeoptAction _unloaded_action; // == Action_reinterpret;
468   static const char* _trap_reason_name[];
469   static const char* _trap_action_name[];
470 
471   static juint _deoptimization_hist[Reason_LIMIT][1+Action_LIMIT][BC_CASE_LIMIT];
472   // Note:  Histogram array size is 1-2 Kb.
473 
474  public:
475   static void update_method_data_from_interpreter(MethodData* trap_mdo, int trap_bci, int reason);
476 };
477 
478 
479 class DeoptimizationMarker : StackObj {  // for profiling
480   static bool _is_active;
481 public:
482   DeoptimizationMarker()  { _is_active = true; }
483   ~DeoptimizationMarker() { _is_active = false; }
484   static bool is_active() { return _is_active; }
485 };
486 
487 #endif // SHARE_RUNTIME_DEOPTIMIZATION_HPP