Mercurial > hg > truffle
annotate src/share/vm/runtime/deoptimization.hpp @ 1711:a6bff45449bc
6973570: OrderAccess::storestore() scales poorly on multi-socket x64 and sparc: cache-line ping-ponging
Summary: volatile store to static variable removed in favour of a volatile store to stack to avoid excessive cache coherency traffic; verified that the volatile store is not elided by any of our current compilers.
Reviewed-by: dholmes, dice, jcoomes, kvn
author | ysr |
---|---|
date | Tue, 10 Aug 2010 14:53:35 -0700 |
parents | c18cbe5936b8 |
children | d5d065957597 |
rev | line source |
---|---|
0 | 1 /* |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1206
diff
changeset
|
2 * Copyright (c) 1997, 2008, Oracle and/or its affiliates. All rights reserved. |
0 | 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 * | |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1206
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1206
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1206
diff
changeset
|
21 * questions. |
0 | 22 * |
23 */ | |
24 | |
25 class ProfileData; | |
26 class vframeArray; | |
27 class MonitorValue; | |
28 class ObjectValue; | |
29 | |
30 class Deoptimization : AllStatic { | |
31 public: | |
32 // What condition caused the deoptimization? | |
33 enum DeoptReason { | |
34 Reason_many = -1, // indicates presence of several reasons | |
35 Reason_none = 0, // indicates absence of a relevant deopt. | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
1172
diff
changeset
|
36 // Next 7 reasons are recorded per bytecode in DataLayout::trap_bits |
0 | 37 Reason_null_check, // saw unexpected null or zero divisor (@bci) |
38 Reason_null_assert, // saw unexpected non-null or non-zero (@bci) | |
39 Reason_range_check, // saw unexpected array index (@bci) | |
40 Reason_class_check, // saw unexpected object class (@bci) | |
41 Reason_array_check, // saw unexpected array class (aastore @bci) | |
42 Reason_intrinsic, // saw unexpected operand to intrinsic (@bci) | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
1172
diff
changeset
|
43 Reason_bimorphic, // saw unexpected object class in bimorphic inlining (@bci) |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
1172
diff
changeset
|
44 |
0 | 45 Reason_unloaded, // unloaded class or constant pool entry |
46 Reason_uninitialized, // bad class state (uninitialized) | |
47 Reason_unreached, // code is not reached, compiler | |
48 Reason_unhandled, // arbitrary compiler limitation | |
49 Reason_constraint, // arbitrary runtime constraint violated | |
50 Reason_div0_check, // a null_check due to division by zero | |
51 Reason_age, // nmethod too old; tier threshold reached | |
1172 | 52 Reason_predicate, // compiler generated predicate failed |
0 | 53 Reason_LIMIT, |
54 // Note: Keep this enum in sync. with _trap_reason_name. | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
1172
diff
changeset
|
55 Reason_RECORDED_LIMIT = Reason_bimorphic // some are not recorded per bc |
0 | 56 // Note: Reason_RECORDED_LIMIT should be < 8 to fit into 3 bits of |
57 // DataLayout::trap_bits. This dependency is enforced indirectly | |
58 // via asserts, to avoid excessive direct header-to-header dependencies. | |
59 // See Deoptimization::trap_state_reason and class DataLayout. | |
60 }; | |
61 | |
62 // What action must be taken by the runtime? | |
63 enum DeoptAction { | |
64 Action_none, // just interpret, do not invalidate nmethod | |
65 Action_maybe_recompile, // recompile the nmethod; need not invalidate | |
66 Action_reinterpret, // invalidate the nmethod, reset IC, maybe recompile | |
67 Action_make_not_entrant, // invalidate the nmethod, recompile (probably) | |
68 Action_make_not_compilable, // invalidate the nmethod and do not compile | |
69 Action_LIMIT | |
70 // Note: Keep this enum in sync. with _trap_action_name. | |
71 }; | |
72 | |
73 enum { | |
74 _action_bits = 3, | |
75 _reason_bits = 4, | |
76 _action_shift = 0, | |
77 _reason_shift = _action_shift+_action_bits, | |
78 BC_CASE_LIMIT = PRODUCT_ONLY(1) NOT_PRODUCT(4) // for _deoptimization_hist | |
79 }; | |
80 | |
81 enum UnpackType { | |
82 Unpack_deopt = 0, // normal deoptimization, use pc computed in unpack_vframe_on_stack | |
83 Unpack_exception = 1, // exception is pending | |
84 Unpack_uncommon_trap = 2, // redo last byte code (C2 only) | |
85 Unpack_reexecute = 3 // reexecute bytecode (C1 only) | |
86 }; | |
87 | |
88 // Checks all compiled methods. Invalid methods are deleted and | |
89 // corresponding activations are deoptimized. | |
90 static int deoptimize_dependents(); | |
91 | |
92 // Deoptimizes a frame lazily. nmethod gets patched deopt happens on return to the frame | |
93 static void deoptimize(JavaThread* thread, frame fr, RegisterMap *reg_map); | |
94 | |
95 private: | |
96 // Does the actual work for deoptimizing a single frame | |
97 static void deoptimize_single_frame(JavaThread* thread, frame fr); | |
98 | |
99 // Helper function to revoke biases of all monitors in frame if UseBiasedLocking | |
100 // is enabled | |
101 static void revoke_biases_of_monitors(JavaThread* thread, frame fr, RegisterMap* map); | |
102 // Helper function to revoke biases of all monitors in frames | |
103 // executing in a particular CodeBlob if UseBiasedLocking is enabled | |
104 static void revoke_biases_of_monitors(CodeBlob* cb); | |
105 | |
106 #ifdef COMPILER2 | |
107 // Support for restoring non-escaping objects | |
108 static bool realloc_objects(JavaThread* thread, frame* fr, GrowableArray<ScopeValue*>* objects, TRAPS); | |
109 static void reassign_type_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, typeArrayOop obj, BasicType type); | |
110 static void reassign_object_array_elements(frame* fr, RegisterMap* reg_map, ObjectValue* sv, objArrayOop obj); | |
111 static void reassign_fields(frame* fr, RegisterMap* reg_map, GrowableArray<ScopeValue*>* objects); | |
83
d3cd40645d0d
6681646: Relocking of a scalar replaced object during deoptimization is broken
kvn
parents:
0
diff
changeset
|
112 static void relock_objects(GrowableArray<MonitorInfo*>* monitors, JavaThread* thread); |
0 | 113 NOT_PRODUCT(static void print_objects(GrowableArray<ScopeValue*>* objects);) |
114 #endif // COMPILER2 | |
115 | |
116 public: | |
117 static vframeArray* create_vframeArray(JavaThread* thread, frame fr, RegisterMap *reg_map, GrowableArray<compiledVFrame*>* chunk); | |
118 | |
119 // Interface used for unpacking deoptimized frames | |
120 | |
121 // UnrollBlock is returned by fetch_unroll_info() to the deoptimization handler (blob). | |
122 // This is only a CheapObj to ease debugging after a deopt failure | |
123 class UnrollBlock : public CHeapObj { | |
124 private: | |
125 int _size_of_deoptimized_frame; // Size, in bytes, of current deoptimized frame | |
126 int _caller_adjustment; // Adjustment, in bytes, to caller's SP by initial interpreted frame | |
127 int _number_of_frames; // Number frames to unroll | |
128 int _total_frame_sizes; // Total of number*sizes frames | |
129 intptr_t* _frame_sizes; // Array of frame sizes, in bytes, for unrolling the stack | |
130 address* _frame_pcs; // Array of frame pc's, in bytes, for unrolling the stack | |
131 intptr_t* _register_block; // Block for storing callee-saved registers. | |
132 BasicType _return_type; // Tells if we have to restore double or long return value | |
133 // The following fields are used as temps during the unpacking phase | |
134 // (which is tight on registers, especially on x86). They really ought | |
135 // to be PD variables but that involves moving this class into its own | |
136 // file to use the pd include mechanism. Maybe in a later cleanup ... | |
137 intptr_t _counter_temp; // SHOULD BE PD VARIABLE (x86 frame count temp) | |
138 intptr_t _initial_fp; // SHOULD BE PD VARIABLE (x86/c2 initial ebp) | |
139 intptr_t _unpack_kind; // SHOULD BE PD VARIABLE (x86 unpack kind) | |
140 intptr_t _sender_sp_temp; // SHOULD BE PD VARIABLE (x86 sender_sp) | |
141 public: | |
142 // Constructor | |
143 UnrollBlock(int size_of_deoptimized_frame, | |
144 int caller_adjustment, | |
145 int number_of_frames, | |
146 intptr_t* frame_sizes, | |
147 address* frames_pcs, | |
148 BasicType return_type); | |
149 ~UnrollBlock(); | |
150 | |
151 // Returns where a register is located. | |
152 intptr_t* value_addr_at(int register_number) const; | |
153 | |
154 // Accessors | |
155 intptr_t* frame_sizes() const { return _frame_sizes; } | |
156 int number_of_frames() const { return _number_of_frames; } | |
157 address* frame_pcs() const { return _frame_pcs ; } | |
158 | |
159 // Returns the total size of frames | |
160 int size_of_frames() const; | |
161 | |
162 // Accessors used by the code generator for the unpack stub. | |
163 static int size_of_deoptimized_frame_offset_in_bytes() { return offset_of(UnrollBlock, _size_of_deoptimized_frame); } | |
164 static int caller_adjustment_offset_in_bytes() { return offset_of(UnrollBlock, _caller_adjustment); } | |
165 static int number_of_frames_offset_in_bytes() { return offset_of(UnrollBlock, _number_of_frames); } | |
166 static int frame_sizes_offset_in_bytes() { return offset_of(UnrollBlock, _frame_sizes); } | |
167 static int total_frame_sizes_offset_in_bytes() { return offset_of(UnrollBlock, _total_frame_sizes); } | |
168 static int frame_pcs_offset_in_bytes() { return offset_of(UnrollBlock, _frame_pcs); } | |
169 static int register_block_offset_in_bytes() { return offset_of(UnrollBlock, _register_block); } | |
170 static int return_type_offset_in_bytes() { return offset_of(UnrollBlock, _return_type); } | |
171 static int counter_temp_offset_in_bytes() { return offset_of(UnrollBlock, _counter_temp); } | |
172 static int initial_fp_offset_in_bytes() { return offset_of(UnrollBlock, _initial_fp); } | |
173 static int unpack_kind_offset_in_bytes() { return offset_of(UnrollBlock, _unpack_kind); } | |
174 static int sender_sp_temp_offset_in_bytes() { return offset_of(UnrollBlock, _sender_sp_temp); } | |
175 | |
176 BasicType return_type() const { return _return_type; } | |
177 void print(); | |
178 }; | |
179 | |
180 //** Returns an UnrollBlock continuing information | |
181 // how to make room for the resulting interpreter frames. | |
182 // Called by assembly stub after execution has returned to | |
183 // deoptimized frame. | |
184 // @argument thread. Thread where stub_frame resides. | |
185 // @see OptoRuntime::deoptimization_fetch_unroll_info_C | |
186 static UnrollBlock* fetch_unroll_info(JavaThread* thread); | |
187 | |
188 //** Unpacks vframeArray onto execution stack | |
189 // Called by assembly stub after execution has returned to | |
190 // deoptimized frame and after the stack unrolling. | |
191 // @argument thread. Thread where stub_frame resides. | |
192 // @argument exec_mode. Determines how execution should be continuted in top frame. | |
193 // 0 means continue after current byte code | |
194 // 1 means exception has happened, handle exception | |
195 // 2 means reexecute current bytecode (for uncommon traps). | |
196 // @see OptoRuntime::deoptimization_unpack_frames_C | |
197 // Return BasicType of call return type, if any | |
198 static BasicType unpack_frames(JavaThread* thread, int exec_mode); | |
199 | |
200 // Cleans up deoptimization bits on thread after unpacking or in the | |
201 // case of an exception. | |
202 static void cleanup_deopt_info(JavaThread *thread, | |
203 vframeArray * array); | |
204 | |
205 // Restores callee saved values from deoptimized frame into oldest interpreter frame | |
206 // so caller of the deoptimized frame will get back the values it expects. | |
207 static void unwind_callee_save_values(frame* f, vframeArray* vframe_array); | |
208 | |
209 //** Performs an uncommon trap for compiled code. | |
210 // The top most compiler frame is converted into interpreter frames | |
211 static UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index); | |
212 // Helper routine that enters the VM and may block | |
213 static void uncommon_trap_inner(JavaThread* thread, jint unloaded_class_index); | |
214 | |
215 //** Deoptimizes the frame identified by id. | |
216 // Only called from VMDeoptimizeFrame | |
217 // @argument thread. Thread where stub_frame resides. | |
218 // @argument id. id of frame that should be deoptimized. | |
219 static void deoptimize_frame(JavaThread* thread, intptr_t* id); | |
220 | |
221 // Statistics | |
222 static void gather_statistics(DeoptReason reason, DeoptAction action, | |
223 Bytecodes::Code bc = Bytecodes::_illegal); | |
224 static void print_statistics(); | |
225 | |
226 // How much room to adjust the last frame's SP by, to make space for | |
227 // the callee's interpreter frame (which expects locals to be next to | |
228 // incoming arguments) | |
229 static int last_frame_adjust(int callee_parameters, int callee_locals); | |
230 | |
231 // trap_request codes | |
232 static DeoptReason trap_request_reason(int trap_request) { | |
233 if (trap_request < 0) | |
234 return (DeoptReason) | |
235 ((~(trap_request) >> _reason_shift) & right_n_bits(_reason_bits)); | |
236 else | |
237 // standard reason for unloaded CP entry | |
238 return Reason_unloaded; | |
239 } | |
240 static DeoptAction trap_request_action(int trap_request) { | |
241 if (trap_request < 0) | |
242 return (DeoptAction) | |
243 ((~(trap_request) >> _action_shift) & right_n_bits(_action_bits)); | |
244 else | |
245 // standard action for unloaded CP entry | |
246 return _unloaded_action; | |
247 } | |
248 static int trap_request_index(int trap_request) { | |
249 if (trap_request < 0) | |
250 return -1; | |
251 else | |
252 return trap_request; | |
253 } | |
254 static int make_trap_request(DeoptReason reason, DeoptAction action, | |
255 int index = -1) { | |
256 assert((1 << _reason_bits) >= Reason_LIMIT, "enough bits"); | |
257 assert((1 << _action_bits) >= Action_LIMIT, "enough bits"); | |
258 int trap_request; | |
259 if (index != -1) | |
260 trap_request = index; | |
261 else | |
262 trap_request = (~(((reason) << _reason_shift) | |
263 + ((action) << _action_shift))); | |
264 assert(reason == trap_request_reason(trap_request), "valid reason"); | |
265 assert(action == trap_request_action(trap_request), "valid action"); | |
266 assert(index == trap_request_index(trap_request), "valid index"); | |
267 return trap_request; | |
268 } | |
269 | |
270 // The trap_state stored in a MDO is decoded here. | |
271 // It records two items of information. | |
272 // reason: If a deoptimization happened here, what its reason was, | |
273 // or if there were multiple deopts with differing reasons. | |
274 // recompiled: If a deoptimization here triggered a recompilation. | |
275 // Note that not all reasons are recorded per-bci. | |
276 static DeoptReason trap_state_reason(int trap_state); | |
277 static int trap_state_has_reason(int trap_state, int reason); | |
278 static int trap_state_add_reason(int trap_state, int reason); | |
279 static bool trap_state_is_recompiled(int trap_state); | |
280 static int trap_state_set_recompiled(int trap_state, bool z); | |
281 static const char* format_trap_state(char* buf, size_t buflen, | |
282 int trap_state); | |
283 | |
284 static bool reason_is_recorded_per_bytecode(DeoptReason reason) { | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
1172
diff
changeset
|
285 return reason > Reason_none && reason <= Reason_RECORDED_LIMIT; |
0 | 286 } |
287 | |
288 static DeoptReason reason_recorded_per_bytecode_if_any(DeoptReason reason) { | |
289 if (reason_is_recorded_per_bytecode(reason)) | |
290 return reason; | |
291 else if (reason == Reason_div0_check) // null check due to divide-by-zero? | |
292 return Reason_null_check; // recorded per BCI as a null check | |
293 else | |
294 return Reason_none; | |
295 } | |
296 | |
297 static const char* trap_reason_name(int reason); | |
298 static const char* trap_action_name(int action); | |
299 // Format like reason='foo' action='bar' index='123'. | |
300 // This is suitable both for XML and for tty output. | |
301 static const char* format_trap_request(char* buf, size_t buflen, | |
302 int trap_request); | |
303 | |
304 static jint total_deoptimization_count(); | |
305 static jint deoptimization_count(DeoptReason reason); | |
306 | |
307 // JVMTI PopFrame support | |
308 | |
309 // Preserves incoming arguments to the popped frame when it is | |
310 // returning to a deoptimized caller | |
311 static void popframe_preserve_args(JavaThread* thread, int bytes_to_save, void* start_address); | |
312 | |
313 private: | |
314 enum { | |
315 _no_count = -1 | |
316 }; | |
317 | |
318 static void reset_invocation_counter(ScopeDesc* trap_scope, jint count = _no_count); | |
319 | |
320 static methodDataOop get_method_data(JavaThread* thread, methodHandle m, bool create_if_missing); | |
321 // Update the mdo's count and per-BCI reason bits, returning previous state: | |
322 static ProfileData* query_update_method_data(methodDataHandle trap_mdo, | |
323 int trap_bci, | |
324 DeoptReason reason, | |
325 //outputs: | |
326 uint& ret_this_trap_count, | |
327 bool& ret_maybe_prior_trap, | |
328 bool& ret_maybe_prior_recompile); | |
329 // class loading support for uncommon trap | |
330 static void load_class_by_index(constantPoolHandle constant_pool, int index, TRAPS); | |
331 static void load_class_by_index(constantPoolHandle constant_pool, int index); | |
332 | |
333 static UnrollBlock* fetch_unroll_info_helper(JavaThread* thread); | |
334 | |
335 static DeoptAction _unloaded_action; // == Action_reinterpret; | |
336 static const char* _trap_reason_name[Reason_LIMIT]; | |
337 static const char* _trap_action_name[Action_LIMIT]; | |
338 | |
339 static juint _deoptimization_hist[Reason_LIMIT][1+Action_LIMIT][BC_CASE_LIMIT]; | |
340 // Note: Histogram array size is 1-2 Kb. | |
341 | |
342 public: | |
343 static void update_method_data_from_interpreter(methodDataHandle trap_mdo, int trap_bci, int reason); | |
344 }; | |
345 | |
346 class DeoptimizationMarker : StackObj { // for profiling | |
347 static bool _is_active; | |
348 public: | |
349 DeoptimizationMarker() { _is_active = true; } | |
350 ~DeoptimizationMarker() { _is_active = false; } | |
351 static bool is_active() { return _is_active; } | |
352 }; |