1 /*
2 * Copyright (c) 1997, 2019, 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 "memory/allocation.hpp"
29 #include "runtime/frame.hpp"
30
31 class ProfileData;
32 class vframeArray;
33 class MonitorInfo;
34 class MonitorValue;
35 class ObjectValue;
36 class AutoBoxObjectValue;
37 class ScopeValue;
38 class compiledVFrame;
39
40 template<class E> class GrowableArray;
41
42 class Deoptimization : AllStatic {
43 friend class VMStructs;
44
45 public:
46 // What condition caused the deoptimization?
47 enum DeoptReason {
48 Reason_many = -1, // indicates presence of several reasons
49 Reason_none = 0, // indicates absence of a relevant deopt.
50 // Next 8 reasons are recorded per bytecode in DataLayout::trap_bits.
51 // This is more complicated for JVMCI as JVMCI may deoptimize to *some* bytecode before the
52 // bytecode that actually caused the deopt (with inlining, JVMCI may even deoptimize to a
53 // bytecode in another method):
54 // - bytecode y in method b() causes deopt
55 // - JVMCI deoptimizes to bytecode x in method a()
56 // -> the deopt reason will be recorded for method a() at bytecode x
57 Reason_null_check, // saw unexpected null or zero divisor (@bci)
58 Reason_null_assert, // saw unexpected non-null or non-zero (@bci)
59 Reason_range_check, // saw unexpected array index (@bci)
60 Reason_class_check, // saw unexpected object class (@bci)
61 Reason_array_check, // saw unexpected array class (aastore @bci)
62 Reason_intrinsic, // saw unexpected operand to intrinsic (@bci)
63 Reason_bimorphic, // saw unexpected object class in bimorphic inlining (@bci)
64
65 #if INCLUDE_JVMCI
66 Reason_unreached0 = Reason_null_assert,
67 Reason_type_checked_inlining = Reason_intrinsic,
68 Reason_optimized_type_check = Reason_bimorphic,
69 #endif
70
71 Reason_profile_predicate, // compiler generated predicate moved from frequent branch in a loop failed
72
73 // recorded per method
74 Reason_unloaded, // unloaded class or constant pool entry
75 Reason_uninitialized, // bad class state (uninitialized)
76 Reason_initialized, // class has been fully initialized
77 Reason_unreached, // code is not reached, compiler
78 Reason_unhandled, // arbitrary compiler limitation
79 Reason_constraint, // arbitrary runtime constraint violated
80 Reason_div0_check, // a null_check due to division by zero
81 Reason_age, // nmethod too old; tier threshold reached
82 Reason_predicate, // compiler generated predicate failed
83 Reason_loop_limit_check, // compiler generated loop limits check failed
84 Reason_speculate_class_check, // saw unexpected object class from type speculation
85 Reason_speculate_null_check, // saw unexpected null from type speculation
86 Reason_speculate_null_assert, // saw unexpected null from type speculation
87 Reason_rtm_state_change, // rtm state change detected
88 Reason_unstable_if, // a branch predicted always false was taken
89 Reason_unstable_fused_if, // fused two ifs that had each one untaken branch. One is now taken.
90 #if INCLUDE_JVMCI
91 Reason_aliasing, // optimistic assumption about aliasing failed
92 Reason_transfer_to_interpreter, // explicit transferToInterpreter()
93 Reason_not_compiled_exception_handler,
94 Reason_unresolved,
95 Reason_jsr_mismatch,
96 #endif
97
98 // Reason_tenured is counted separately, add normal counted Reasons above.
99 // Related to MethodData::_trap_hist_limit where Reason_tenured isn't included
100 Reason_tenured, // age of the code has reached the limit
101 Reason_LIMIT,
102
103 // Note: Keep this enum in sync. with _trap_reason_name.
104 Reason_RECORDED_LIMIT = Reason_profile_predicate // some are not recorded per bc
105 // Note: Reason_RECORDED_LIMIT should fit into 31 bits of
106 // DataLayout::trap_bits. This dependency is enforced indirectly
107 // via asserts, to avoid excessive direct header-to-header dependencies.
108 // See Deoptimization::trap_state_reason and class DataLayout.
109 };
110
111 // What action must be taken by the runtime?
112 enum DeoptAction {
113 Action_none, // just interpret, do not invalidate nmethod
114 Action_maybe_recompile, // recompile the nmethod; need not invalidate
115 Action_reinterpret, // invalidate the nmethod, reset IC, maybe recompile
116 Action_make_not_entrant, // invalidate the nmethod, recompile (probably)
117 Action_make_not_compilable, // invalidate the nmethod and do not compile
118 Action_LIMIT
119 // Note: Keep this enum in sync. with _trap_action_name.
120 };
121
122 enum {
123 _action_bits = 3,
124 _reason_bits = 5,
125 _debug_id_bits = 23,
126 _action_shift = 0,
127 _reason_shift = _action_shift+_action_bits,
128 _debug_id_shift = _reason_shift+_reason_bits,
129 BC_CASE_LIMIT = PRODUCT_ONLY(1) NOT_PRODUCT(4) // for _deoptimization_hist
130 };
131
132 enum UnpackType {
133 Unpack_deopt = 0, // normal deoptimization, use pc computed in unpack_vframe_on_stack
134 Unpack_exception = 1, // exception is pending
135 Unpack_uncommon_trap = 2, // redo last byte code (C2 only)
136 Unpack_reexecute = 3, // reexecute bytecode (C1 only)
137 Unpack_LIMIT = 4
138 };
139
140 #if INCLUDE_JVMCI
141 // Can reconstruct virtualized unsafe large accesses to byte arrays.
142 static const int _support_large_access_byte_array_virtualization = 1;
143 #endif
144
145 // Make all nmethods that are marked_for_deoptimization not_entrant and deoptimize any live
146 // activations using those nmethods. If an nmethod is passed as an argument then it is
147 // marked_for_deoptimization and made not_entrant. Otherwise a scan of the code cache is done to
148 // find all marked nmethods and they are made not_entrant.
149 static void deoptimize_all_marked(nmethod* nmethod_only = NULL);
150
151 private:
152 // Revoke biased locks at deopt.
153 static void revoke_from_deopt_handler(JavaThread* thread, frame fr, RegisterMap* map);
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 static oop get_cached_box(AutoBoxObjectValue* bv, frame* fr, RegisterMap* reg_map, TRAPS);
162 #endif
163
164 private:
165 // Does the actual work for deoptimizing a single frame
166 static void deoptimize_single_frame(JavaThread* thread, frame fr, DeoptReason reason);
167
168 #if COMPILER2_OR_JVMCI
169 public:
170
171 // Support for restoring non-escaping objects
172 static bool realloc_objects(JavaThread* thread, frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, TRAPS);
173 static void reassign_stack_allocated_fields(frame *fr, GrowableArray<ScopeValue*>* objects, ObjectValue *sv, Handle obj, Klass* k);
174 static void reassign_stack_allocated_type_array_elements(oop orig, oop newly_allocated, Klass *k);
175 static void reassign_stack_allocated_object_array_elements(oop orig, oop newly_allocated, intptr_t *sp_base, intptr_t *sp_top, GrowableArray<ScopeValue*>* objects);
176 static void reassign_scalar_replaced_fields(frame *fr, RegisterMap *reg_map, GrowableArray<ScopeValue*>* objects, ObjectValue *sv, Handle obj, Klass* k, bool skip_internal);
177 static void reassign_scalar_replaced_type_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, typeArrayOop obj, BasicType type);
178 static void reassign_scalar_replaced_object_array_elements(frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, ObjectValue* sv, objArrayOop obj);
179 static ScopeValue *get_scope_value(frame* fr, RegisterMap* reg_map, ScopeValue* sv, GrowableArray<ScopeValue*>* objects);
180 static ScopeValue *match_object_to_stack_oop(intptr_t *oop_ptr, intptr_t *sp_base, GrowableArray<ScopeValue*>* objects);
181 static void reassign_scalar_replaced_fields_by_klass(InstanceKlass* klass, frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, ObjectValue* sv, int svIndex, oop obj, bool skip_internal);
182 static void reassign_fields(frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects, bool realloc_failures, bool skip_internal);
183 static void relock_objects(GrowableArray<MonitorInfo*>* monitors, JavaThread* thread, bool realloc_failures);
184 static void pop_frames_failed_reallocs(JavaThread* thread, vframeArray* array);
185 NOT_PRODUCT(static void print_objects(GrowableArray<ScopeValue*>* objects, bool realloc_failures);)
186 #endif // COMPILER2_OR_JVMCI
187
188 public:
189 static vframeArray* create_vframeArray(JavaThread* thread, frame fr, RegisterMap *reg_map, GrowableArray<compiledVFrame*>* chunk, bool realloc_failures);
190
191 // Interface used for unpacking deoptimized frames
192
193 // UnrollBlock is returned by fetch_unroll_info() to the deoptimization handler (blob).
194 // This is only a CheapObj to ease debugging after a deopt failure
195 class UnrollBlock : public CHeapObj<mtCompiler> {
196 friend class VMStructs;
197 friend class JVMCIVMStructs;
198 private:
199 int _size_of_deoptimized_frame; // Size, in bytes, of current deoptimized frame
200 int _caller_adjustment; // Adjustment, in bytes, to caller's SP by initial interpreted frame
201 int _number_of_frames; // Number frames to unroll
202 int _total_frame_sizes; // Total of number*sizes frames
203 intptr_t* _frame_sizes; // Array of frame sizes, in bytes, for unrolling the stack
204 address* _frame_pcs; // Array of frame pc's, in bytes, for unrolling the stack
205 intptr_t* _register_block; // Block for storing callee-saved registers.
206 BasicType _return_type; // Tells if we have to restore double or long return value
207 intptr_t _initial_info; // Platform dependent data for the sender frame (was FP on x86)
208 int _caller_actual_parameters; // The number of actual arguments at the
209 // interpreted caller of the deoptimized frame
210 int _unpack_kind; // exec_mode that can be changed during fetch_unroll_info
211
212 // The following fields are used as temps during the unpacking phase
213 // (which is tight on registers, especially on x86). They really ought
214 // to be PD variables but that involves moving this class into its own
215 // file to use the pd include mechanism. Maybe in a later cleanup ...
216 intptr_t _counter_temp; // SHOULD BE PD VARIABLE (x86 frame count temp)
217 intptr_t _sender_sp_temp; // SHOULD BE PD VARIABLE (x86 sender_sp)
218 public:
219 // Constructor
220 UnrollBlock(int size_of_deoptimized_frame,
221 int caller_adjustment,
222 int caller_actual_parameters,
223 int number_of_frames,
224 intptr_t* frame_sizes,
225 address* frames_pcs,
226 BasicType return_type,
227 int unpack_kind);
228 ~UnrollBlock();
229
230 // Returns where a register is located.
231 intptr_t* value_addr_at(int register_number) const;
232
233 // Accessors
234 intptr_t* frame_sizes() const { return _frame_sizes; }
235 int number_of_frames() const { return _number_of_frames; }
236 address* frame_pcs() const { return _frame_pcs ; }
237 int unpack_kind() const { return _unpack_kind; }
238
239 // Returns the total size of frames
240 int size_of_frames() const;
241
242 void set_initial_info(intptr_t info) { _initial_info = info; }
243
244 int caller_actual_parameters() const { return _caller_actual_parameters; }
245
246 // Accessors used by the code generator for the unpack stub.
247 static int size_of_deoptimized_frame_offset_in_bytes() { return offset_of(UnrollBlock, _size_of_deoptimized_frame); }
248 static int caller_adjustment_offset_in_bytes() { return offset_of(UnrollBlock, _caller_adjustment); }
249 static int number_of_frames_offset_in_bytes() { return offset_of(UnrollBlock, _number_of_frames); }
250 static int frame_sizes_offset_in_bytes() { return offset_of(UnrollBlock, _frame_sizes); }
251 static int total_frame_sizes_offset_in_bytes() { return offset_of(UnrollBlock, _total_frame_sizes); }
252 static int frame_pcs_offset_in_bytes() { return offset_of(UnrollBlock, _frame_pcs); }
253 static int register_block_offset_in_bytes() { return offset_of(UnrollBlock, _register_block); }
254 static int return_type_offset_in_bytes() { return offset_of(UnrollBlock, _return_type); }
255 static int counter_temp_offset_in_bytes() { return offset_of(UnrollBlock, _counter_temp); }
256 static int initial_info_offset_in_bytes() { return offset_of(UnrollBlock, _initial_info); }
257 static int unpack_kind_offset_in_bytes() { return offset_of(UnrollBlock, _unpack_kind); }
258 static int sender_sp_temp_offset_in_bytes() { return offset_of(UnrollBlock, _sender_sp_temp); }
259
260 BasicType return_type() const { return _return_type; }
261 void print();
262 };
263
264 //** Returns an UnrollBlock continuing information
265 // how to make room for the resulting interpreter frames.
266 // Called by assembly stub after execution has returned to
267 // deoptimized frame.
268 // @argument thread. Thread where stub_frame resides.
269 // @see OptoRuntime::deoptimization_fetch_unroll_info_C
270 static UnrollBlock* fetch_unroll_info(JavaThread* thread, int exec_mode);
271
272 //** Unpacks vframeArray onto execution stack
273 // Called by assembly stub after execution has returned to
274 // deoptimized frame and after the stack unrolling.
275 // @argument thread. Thread where stub_frame resides.
276 // @argument exec_mode. Determines how execution should be continued in top frame.
277 // 0 means continue after current byte code
278 // 1 means exception has happened, handle exception
279 // 2 means reexecute current bytecode (for uncommon traps).
280 // @see OptoRuntime::deoptimization_unpack_frames_C
281 // Return BasicType of call return type, if any
282 static BasicType unpack_frames(JavaThread* thread, int exec_mode);
283
284 // Cleans up deoptimization bits on thread after unpacking or in the
285 // case of an exception.
286 static void cleanup_deopt_info(JavaThread *thread,
287 vframeArray * array);
288
289 // Restores callee saved values from deoptimized frame into oldest interpreter frame
290 // so caller of the deoptimized frame will get back the values it expects.
291 static void unwind_callee_save_values(frame* f, vframeArray* vframe_array);
292
293 //** Performs an uncommon trap for compiled code.
294 // The top most compiler frame is converted into interpreter frames
295 static UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index, jint exec_mode);
296 // Helper routine that enters the VM and may block
297 static void uncommon_trap_inner(JavaThread* thread, jint unloaded_class_index);
298
299 //** Deoptimizes the frame identified by id.
300 // Only called from VMDeoptimizeFrame
301 // @argument thread. Thread where stub_frame resides.
302 // @argument id. id of frame that should be deoptimized.
303 static void deoptimize_frame_internal(JavaThread* thread, intptr_t* id, DeoptReason reason);
304
305 // if thread is not the current thread then execute
306 // VM_DeoptimizeFrame otherwise deoptimize directly.
307 static void deoptimize_frame(JavaThread* thread, intptr_t* id, DeoptReason reason);
308 static void deoptimize_frame(JavaThread* thread, intptr_t* id);
309
310 // Statistics
311 static void gather_statistics(DeoptReason reason, DeoptAction action,
312 Bytecodes::Code bc = Bytecodes::_illegal);
313 static void print_statistics();
314
315 // How much room to adjust the last frame's SP by, to make space for
316 // the callee's interpreter frame (which expects locals to be next to
317 // incoming arguments)
318 static int last_frame_adjust(int callee_parameters, int callee_locals);
319
320 // trap_request codes
321 static DeoptReason trap_request_reason(int trap_request) {
322 if (trap_request < 0)
323 return (DeoptReason)
324 ((~(trap_request) >> _reason_shift) & right_n_bits(_reason_bits));
325 else
326 // standard reason for unloaded CP entry
327 return Reason_unloaded;
328 }
329 static DeoptAction trap_request_action(int trap_request) {
330 if (trap_request < 0)
331 return (DeoptAction)
332 ((~(trap_request) >> _action_shift) & right_n_bits(_action_bits));
333 else
334 // standard action for unloaded CP entry
335 return _unloaded_action;
336 }
337 static int trap_request_debug_id(int trap_request) {
338 if (trap_request < 0) {
339 return ((~(trap_request) >> _debug_id_shift) & right_n_bits(_debug_id_bits));
340 } else {
341 // standard action for unloaded CP entry
342 return 0;
343 }
344 }
345 static int trap_request_index(int trap_request) {
346 if (trap_request < 0)
347 return -1;
348 else
349 return trap_request;
350 }
351 static int make_trap_request(DeoptReason reason, DeoptAction action,
352 int index = -1) {
353 assert((1 << _reason_bits) >= Reason_LIMIT, "enough bits");
354 assert((1 << _action_bits) >= Action_LIMIT, "enough bits");
355 int trap_request;
356 if (index != -1)
357 trap_request = index;
358 else
359 trap_request = (~(((reason) << _reason_shift)
360 + ((action) << _action_shift)));
361 assert(reason == trap_request_reason(trap_request), "valid reason");
362 assert(action == trap_request_action(trap_request), "valid action");
363 assert(index == trap_request_index(trap_request), "valid index");
364 return trap_request;
365 }
366
367 // The trap_state stored in a MDO is decoded here.
368 // It records two items of information.
369 // reason: If a deoptimization happened here, what its reason was,
370 // or if there were multiple deopts with differing reasons.
371 // recompiled: If a deoptimization here triggered a recompilation.
372 // Note that not all reasons are recorded per-bci.
373 static DeoptReason trap_state_reason(int trap_state);
374 static int trap_state_has_reason(int trap_state, int reason);
375 static int trap_state_add_reason(int trap_state, int reason);
376 static bool trap_state_is_recompiled(int trap_state);
377 static int trap_state_set_recompiled(int trap_state, bool z);
378 static const char* format_trap_state(char* buf, size_t buflen,
379 int trap_state);
380
381 static bool reason_is_recorded_per_bytecode(DeoptReason reason) {
382 return reason > Reason_none && reason <= Reason_RECORDED_LIMIT;
383 }
384
385 static DeoptReason reason_recorded_per_bytecode_if_any(DeoptReason reason) {
386 if (reason_is_recorded_per_bytecode(reason))
387 return reason;
388 else if (reason == Reason_div0_check) // null check due to divide-by-zero?
389 return Reason_null_check; // recorded per BCI as a null check
390 else if (reason == Reason_speculate_class_check)
391 return Reason_class_check;
392 else if (reason == Reason_speculate_null_check)
393 return Reason_null_check;
394 else if (reason == Reason_speculate_null_assert)
395 return Reason_null_assert;
396 else if (reason == Reason_unstable_if)
397 return Reason_intrinsic;
398 else if (reason == Reason_unstable_fused_if)
399 return Reason_range_check;
400 else
401 return Reason_none;
402 }
403
404 static bool reason_is_speculate(int reason) {
405 if (reason == Reason_speculate_class_check ||
406 reason == Reason_speculate_null_check ||
407 reason == Reason_speculate_null_assert) {
408 return true;
409 }
410 return false;
411 }
412
413 static DeoptReason reason_null_check(bool speculative) {
414 return speculative ? Deoptimization::Reason_speculate_null_check : Deoptimization::Reason_null_check;
415 }
416
417 static DeoptReason reason_class_check(bool speculative) {
418 return speculative ? Deoptimization::Reason_speculate_class_check : Deoptimization::Reason_class_check;
419 }
420
421 static DeoptReason reason_null_assert(bool speculative) {
422 return speculative ? Deoptimization::Reason_speculate_null_assert : Deoptimization::Reason_null_assert;
423 }
424
425 static uint per_method_trap_limit(int reason) {
426 return reason_is_speculate(reason) ? (uint)PerMethodSpecTrapLimit : (uint)PerMethodTrapLimit;
427 }
428
429 static const char* trap_reason_name(int reason);
430 static const char* trap_action_name(int action);
431 // Format like reason='foo' action='bar' index='123'.
432 // This is suitable both for XML and for tty output.
433 static const char* format_trap_request(char* buf, size_t buflen,
434 int trap_request);
435
436 static jint total_deoptimization_count();
437
438 // JVMTI PopFrame support
439
440 // Preserves incoming arguments to the popped frame when it is
441 // returning to a deoptimized caller
442 static void popframe_preserve_args(JavaThread* thread, int bytes_to_save, void* start_address);
443
444 static MethodData* get_method_data(JavaThread* thread, const methodHandle& m, bool create_if_missing);
445 private:
446 // Update the mdo's count and per-BCI reason bits, returning previous state:
447 static ProfileData* query_update_method_data(MethodData* trap_mdo,
448 int trap_bci,
449 DeoptReason reason,
450 bool update_total_trap_count,
451 #if INCLUDE_JVMCI
452 bool is_osr,
453 #endif
454 Method* compiled_method,
455 //outputs:
456 uint& ret_this_trap_count,
457 bool& ret_maybe_prior_trap,
458 bool& ret_maybe_prior_recompile);
459 // class loading support for uncommon trap
460 static void load_class_by_index(const constantPoolHandle& constant_pool, int index, TRAPS);
461 static void load_class_by_index(const constantPoolHandle& constant_pool, int index);
462
463 static UnrollBlock* fetch_unroll_info_helper(JavaThread* thread, int exec_mode);
464
465 static DeoptAction _unloaded_action; // == Action_reinterpret;
466 static const char* _trap_reason_name[];
467 static const char* _trap_action_name[];
468
469 static juint _deoptimization_hist[Reason_LIMIT][1+Action_LIMIT][BC_CASE_LIMIT];
470 // Note: Histogram array size is 1-2 Kb.
471
472 public:
473 static void update_method_data_from_interpreter(MethodData* trap_mdo, int trap_bci, int reason);
474 };
475
476 class DeoptimizationMarker : StackObj { // for profiling
477 static bool _is_active;
478 public:
479 DeoptimizationMarker() { _is_active = true; }
480 ~DeoptimizationMarker() { _is_active = false; }
481 static bool is_active() { return _is_active; }
482 };
483
484 #endif // SHARE_RUNTIME_DEOPTIMIZATION_HPP