1 /* |
|
2 * Copyright (c) 1997, 2014, 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 #include "precompiled.hpp" |
|
26 #include "interp_masm_x86.hpp" |
|
27 #include "interpreter/interpreter.hpp" |
|
28 #include "interpreter/interpreterRuntime.hpp" |
|
29 #include "oops/arrayOop.hpp" |
|
30 #include "oops/markOop.hpp" |
|
31 #include "oops/methodData.hpp" |
|
32 #include "oops/method.hpp" |
|
33 #include "prims/jvmtiExport.hpp" |
|
34 #include "prims/jvmtiRedefineClassesTrace.hpp" |
|
35 #include "prims/jvmtiThreadState.hpp" |
|
36 #include "runtime/basicLock.hpp" |
|
37 #include "runtime/biasedLocking.hpp" |
|
38 #include "runtime/sharedRuntime.hpp" |
|
39 #include "runtime/thread.inline.hpp" |
|
40 |
|
41 |
|
42 // Implementation of InterpreterMacroAssembler |
|
43 #ifdef CC_INTERP |
|
44 void InterpreterMacroAssembler::get_method(Register reg) { |
|
45 movptr(reg, Address(rbp, -(sizeof(BytecodeInterpreter) + 2 * wordSize))); |
|
46 movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method))); |
|
47 } |
|
48 #endif // CC_INTERP |
|
49 |
|
50 |
|
51 #ifndef CC_INTERP |
|
52 void InterpreterMacroAssembler::call_VM_leaf_base( |
|
53 address entry_point, |
|
54 int number_of_arguments |
|
55 ) { |
|
56 // interpreter specific |
|
57 // |
|
58 // Note: No need to save/restore bcp & locals (rsi & rdi) pointer |
|
59 // since these are callee saved registers and no blocking/ |
|
60 // GC can happen in leaf calls. |
|
61 // Further Note: DO NOT save/restore bcp/locals. If a caller has |
|
62 // already saved them so that it can use rsi/rdi as temporaries |
|
63 // then a save/restore here will DESTROY the copy the caller |
|
64 // saved! There used to be a save_bcp() that only happened in |
|
65 // the ASSERT path (no restore_bcp). Which caused bizarre failures |
|
66 // when jvm built with ASSERTs. |
|
67 #ifdef ASSERT |
|
68 { Label L; |
|
69 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); |
|
70 jcc(Assembler::equal, L); |
|
71 stop("InterpreterMacroAssembler::call_VM_leaf_base: last_sp != NULL"); |
|
72 bind(L); |
|
73 } |
|
74 #endif |
|
75 // super call |
|
76 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); |
|
77 // interpreter specific |
|
78 |
|
79 // Used to ASSERT that rsi/rdi were equal to frame's bcp/locals |
|
80 // but since they may not have been saved (and we don't want to |
|
81 // save them here (see note above) the assert is invalid. |
|
82 } |
|
83 |
|
84 |
|
85 void InterpreterMacroAssembler::call_VM_base( |
|
86 Register oop_result, |
|
87 Register java_thread, |
|
88 Register last_java_sp, |
|
89 address entry_point, |
|
90 int number_of_arguments, |
|
91 bool check_exceptions |
|
92 ) { |
|
93 #ifdef ASSERT |
|
94 { Label L; |
|
95 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); |
|
96 jcc(Assembler::equal, L); |
|
97 stop("InterpreterMacroAssembler::call_VM_base: last_sp != NULL"); |
|
98 bind(L); |
|
99 } |
|
100 #endif /* ASSERT */ |
|
101 // interpreter specific |
|
102 // |
|
103 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't |
|
104 // really make a difference for these runtime calls, since they are |
|
105 // slow anyway. Btw., bcp must be saved/restored since it may change |
|
106 // due to GC. |
|
107 assert(java_thread == noreg , "not expecting a precomputed java thread"); |
|
108 save_bcp(); |
|
109 // super call |
|
110 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exceptions); |
|
111 // interpreter specific |
|
112 restore_bcp(); |
|
113 restore_locals(); |
|
114 } |
|
115 |
|
116 |
|
117 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) { |
|
118 if (JvmtiExport::can_pop_frame()) { |
|
119 Label L; |
|
120 // Initiate popframe handling only if it is not already being processed. If the flag |
|
121 // has the popframe_processing bit set, it means that this code is called *during* popframe |
|
122 // handling - we don't want to reenter. |
|
123 Register pop_cond = java_thread; // Not clear if any other register is available... |
|
124 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset())); |
|
125 testl(pop_cond, JavaThread::popframe_pending_bit); |
|
126 jcc(Assembler::zero, L); |
|
127 testl(pop_cond, JavaThread::popframe_processing_bit); |
|
128 jcc(Assembler::notZero, L); |
|
129 // Call Interpreter::remove_activation_preserving_args_entry() to get the |
|
130 // address of the same-named entrypoint in the generated interpreter code. |
|
131 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); |
|
132 jmp(rax); |
|
133 bind(L); |
|
134 get_thread(java_thread); |
|
135 } |
|
136 } |
|
137 |
|
138 |
|
139 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { |
|
140 get_thread(rcx); |
|
141 movl(rcx, Address(rcx, JavaThread::jvmti_thread_state_offset())); |
|
142 const Address tos_addr (rcx, JvmtiThreadState::earlyret_tos_offset()); |
|
143 const Address oop_addr (rcx, JvmtiThreadState::earlyret_oop_offset()); |
|
144 const Address val_addr (rcx, JvmtiThreadState::earlyret_value_offset()); |
|
145 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset() |
|
146 + in_ByteSize(wordSize)); |
|
147 switch (state) { |
|
148 case atos: movptr(rax, oop_addr); |
|
149 movptr(oop_addr, NULL_WORD); |
|
150 verify_oop(rax, state); break; |
|
151 case ltos: |
|
152 movl(rdx, val_addr1); // fall through |
|
153 case btos: // fall through |
|
154 case ctos: // fall through |
|
155 case stos: // fall through |
|
156 case itos: movl(rax, val_addr); break; |
|
157 case ftos: fld_s(val_addr); break; |
|
158 case dtos: fld_d(val_addr); break; |
|
159 case vtos: /* nothing to do */ break; |
|
160 default : ShouldNotReachHere(); |
|
161 } |
|
162 // Clean up tos value in the thread object |
|
163 movl(tos_addr, (int32_t) ilgl); |
|
164 movptr(val_addr, NULL_WORD); |
|
165 NOT_LP64(movptr(val_addr1, NULL_WORD)); |
|
166 } |
|
167 |
|
168 |
|
169 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) { |
|
170 if (JvmtiExport::can_force_early_return()) { |
|
171 Label L; |
|
172 Register tmp = java_thread; |
|
173 movptr(tmp, Address(tmp, JavaThread::jvmti_thread_state_offset())); |
|
174 testptr(tmp, tmp); |
|
175 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit; |
|
176 |
|
177 // Initiate earlyret handling only if it is not already being processed. |
|
178 // If the flag has the earlyret_processing bit set, it means that this code |
|
179 // is called *during* earlyret handling - we don't want to reenter. |
|
180 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset())); |
|
181 cmpl(tmp, JvmtiThreadState::earlyret_pending); |
|
182 jcc(Assembler::notEqual, L); |
|
183 |
|
184 // Call Interpreter::remove_activation_early_entry() to get the address of the |
|
185 // same-named entrypoint in the generated interpreter code. |
|
186 get_thread(java_thread); |
|
187 movptr(tmp, Address(java_thread, JavaThread::jvmti_thread_state_offset())); |
|
188 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset())); |
|
189 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1); |
|
190 jmp(rax); |
|
191 bind(L); |
|
192 get_thread(java_thread); |
|
193 } |
|
194 } |
|
195 |
|
196 |
|
197 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) { |
|
198 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode"); |
|
199 load_unsigned_short(reg, Address(rsi, bcp_offset)); |
|
200 bswapl(reg); |
|
201 shrl(reg, 16); |
|
202 } |
|
203 |
|
204 |
|
205 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register reg, int bcp_offset, size_t index_size) { |
|
206 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); |
|
207 if (index_size == sizeof(u2)) { |
|
208 load_unsigned_short(reg, Address(rsi, bcp_offset)); |
|
209 } else if (index_size == sizeof(u4)) { |
|
210 movl(reg, Address(rsi, bcp_offset)); |
|
211 // Check if the secondary index definition is still ~x, otherwise |
|
212 // we have to change the following assembler code to calculate the |
|
213 // plain index. |
|
214 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); |
|
215 notl(reg); // convert to plain index |
|
216 } else if (index_size == sizeof(u1)) { |
|
217 load_unsigned_byte(reg, Address(rsi, bcp_offset)); |
|
218 } else { |
|
219 ShouldNotReachHere(); |
|
220 } |
|
221 } |
|
222 |
|
223 |
|
224 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register index, |
|
225 int bcp_offset, size_t index_size) { |
|
226 assert_different_registers(cache, index); |
|
227 get_cache_index_at_bcp(index, bcp_offset, index_size); |
|
228 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); |
|
229 assert(sizeof(ConstantPoolCacheEntry) == 4*wordSize, "adjust code below"); |
|
230 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line"); |
|
231 shlptr(index, 2); // convert from field index to ConstantPoolCacheEntry index |
|
232 } |
|
233 |
|
234 |
|
235 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, |
|
236 Register index, |
|
237 Register bytecode, |
|
238 int byte_no, |
|
239 int bcp_offset, |
|
240 size_t index_size) { |
|
241 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size); |
|
242 movptr(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset())); |
|
243 const int shift_count = (1 + byte_no) * BitsPerByte; |
|
244 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) || |
|
245 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift), |
|
246 "correct shift count"); |
|
247 shrptr(bytecode, shift_count); |
|
248 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask"); |
|
249 andptr(bytecode, ConstantPoolCacheEntry::bytecode_1_mask); |
|
250 } |
|
251 |
|
252 |
|
253 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp, |
|
254 int bcp_offset, size_t index_size) { |
|
255 assert(cache != tmp, "must use different register"); |
|
256 get_cache_index_at_bcp(tmp, bcp_offset, index_size); |
|
257 assert(sizeof(ConstantPoolCacheEntry) == 4*wordSize, "adjust code below"); |
|
258 // convert from field index to ConstantPoolCacheEntry index |
|
259 // and from word offset to byte offset |
|
260 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line"); |
|
261 shll(tmp, 2 + LogBytesPerWord); |
|
262 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); |
|
263 // skip past the header |
|
264 addptr(cache, in_bytes(ConstantPoolCache::base_offset())); |
|
265 addptr(cache, tmp); // construct pointer to cache entry |
|
266 } |
|
267 |
|
268 // Load object from cpool->resolved_references(index) |
|
269 void InterpreterMacroAssembler::load_resolved_reference_at_index( |
|
270 Register result, Register index) { |
|
271 assert_different_registers(result, index); |
|
272 // convert from field index to resolved_references() index and from |
|
273 // word index to byte offset. Since this is a java object, it can be compressed |
|
274 Register tmp = index; // reuse |
|
275 shll(tmp, LogBytesPerHeapOop); |
|
276 |
|
277 get_constant_pool(result); |
|
278 // load pointer for resolved_references[] objArray |
|
279 movptr(result, Address(result, ConstantPool::resolved_references_offset_in_bytes())); |
|
280 // JNIHandles::resolve(obj); |
|
281 movptr(result, Address(result, 0)); |
|
282 // Add in the index |
|
283 addptr(result, tmp); |
|
284 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
|
285 } |
|
286 |
|
287 // Generate a subtype check: branch to ok_is_subtype if sub_klass is |
|
288 // a subtype of super_klass. EAX holds the super_klass. Blows ECX. |
|
289 // Resets EDI to locals. Register sub_klass cannot be any of the above. |
|
290 void InterpreterMacroAssembler::gen_subtype_check( Register Rsub_klass, Label &ok_is_subtype ) { |
|
291 assert( Rsub_klass != rax, "rax, holds superklass" ); |
|
292 assert( Rsub_klass != rcx, "used as a temp" ); |
|
293 assert( Rsub_klass != rdi, "used as a temp, restored from locals" ); |
|
294 |
|
295 // Profile the not-null value's klass. |
|
296 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi |
|
297 |
|
298 // Do the check. |
|
299 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx |
|
300 |
|
301 // Profile the failure of the check. |
|
302 profile_typecheck_failed(rcx); // blows rcx |
|
303 } |
|
304 |
|
305 void InterpreterMacroAssembler::f2ieee() { |
|
306 if (IEEEPrecision) { |
|
307 fstp_s(Address(rsp, 0)); |
|
308 fld_s(Address(rsp, 0)); |
|
309 } |
|
310 } |
|
311 |
|
312 |
|
313 void InterpreterMacroAssembler::d2ieee() { |
|
314 if (IEEEPrecision) { |
|
315 fstp_d(Address(rsp, 0)); |
|
316 fld_d(Address(rsp, 0)); |
|
317 } |
|
318 } |
|
319 |
|
320 // Java Expression Stack |
|
321 |
|
322 void InterpreterMacroAssembler::pop_ptr(Register r) { |
|
323 pop(r); |
|
324 } |
|
325 |
|
326 void InterpreterMacroAssembler::pop_i(Register r) { |
|
327 pop(r); |
|
328 } |
|
329 |
|
330 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) { |
|
331 pop(lo); |
|
332 pop(hi); |
|
333 } |
|
334 |
|
335 void InterpreterMacroAssembler::pop_f() { |
|
336 fld_s(Address(rsp, 0)); |
|
337 addptr(rsp, 1 * wordSize); |
|
338 } |
|
339 |
|
340 void InterpreterMacroAssembler::pop_d() { |
|
341 fld_d(Address(rsp, 0)); |
|
342 addptr(rsp, 2 * wordSize); |
|
343 } |
|
344 |
|
345 |
|
346 void InterpreterMacroAssembler::pop(TosState state) { |
|
347 switch (state) { |
|
348 case atos: pop_ptr(rax); break; |
|
349 case btos: // fall through |
|
350 case ctos: // fall through |
|
351 case stos: // fall through |
|
352 case itos: pop_i(rax); break; |
|
353 case ltos: pop_l(rax, rdx); break; |
|
354 case ftos: pop_f(); break; |
|
355 case dtos: pop_d(); break; |
|
356 case vtos: /* nothing to do */ break; |
|
357 default : ShouldNotReachHere(); |
|
358 } |
|
359 verify_oop(rax, state); |
|
360 } |
|
361 |
|
362 void InterpreterMacroAssembler::push_ptr(Register r) { |
|
363 push(r); |
|
364 } |
|
365 |
|
366 void InterpreterMacroAssembler::push_i(Register r) { |
|
367 push(r); |
|
368 } |
|
369 |
|
370 void InterpreterMacroAssembler::push_l(Register lo, Register hi) { |
|
371 push(hi); |
|
372 push(lo); |
|
373 } |
|
374 |
|
375 void InterpreterMacroAssembler::push_f() { |
|
376 // Do not schedule for no AGI! Never write beyond rsp! |
|
377 subptr(rsp, 1 * wordSize); |
|
378 fstp_s(Address(rsp, 0)); |
|
379 } |
|
380 |
|
381 void InterpreterMacroAssembler::push_d(Register r) { |
|
382 // Do not schedule for no AGI! Never write beyond rsp! |
|
383 subptr(rsp, 2 * wordSize); |
|
384 fstp_d(Address(rsp, 0)); |
|
385 } |
|
386 |
|
387 |
|
388 void InterpreterMacroAssembler::push(TosState state) { |
|
389 verify_oop(rax, state); |
|
390 switch (state) { |
|
391 case atos: push_ptr(rax); break; |
|
392 case btos: // fall through |
|
393 case ctos: // fall through |
|
394 case stos: // fall through |
|
395 case itos: push_i(rax); break; |
|
396 case ltos: push_l(rax, rdx); break; |
|
397 case ftos: push_f(); break; |
|
398 case dtos: push_d(rax); break; |
|
399 case vtos: /* nothing to do */ break; |
|
400 default : ShouldNotReachHere(); |
|
401 } |
|
402 } |
|
403 |
|
404 |
|
405 // Helpers for swap and dup |
|
406 void InterpreterMacroAssembler::load_ptr(int n, Register val) { |
|
407 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n))); |
|
408 } |
|
409 |
|
410 void InterpreterMacroAssembler::store_ptr(int n, Register val) { |
|
411 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val); |
|
412 } |
|
413 |
|
414 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() { |
|
415 // set sender sp |
|
416 lea(rsi, Address(rsp, wordSize)); |
|
417 // record last_sp |
|
418 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rsi); |
|
419 } |
|
420 |
|
421 |
|
422 // Jump to from_interpreted entry of a call unless single stepping is possible |
|
423 // in this thread in which case we must call the i2i entry |
|
424 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { |
|
425 prepare_to_jump_from_interpreted(); |
|
426 |
|
427 if (JvmtiExport::can_post_interpreter_events()) { |
|
428 Label run_compiled_code; |
|
429 // JVMTI events, such as single-stepping, are implemented partly by avoiding running |
|
430 // compiled code in threads for which the event is enabled. Check here for |
|
431 // interp_only_mode if these events CAN be enabled. |
|
432 get_thread(temp); |
|
433 // interp_only is an int, on little endian it is sufficient to test the byte only |
|
434 // Is a cmpl faster? |
|
435 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0); |
|
436 jccb(Assembler::zero, run_compiled_code); |
|
437 jmp(Address(method, Method::interpreter_entry_offset())); |
|
438 bind(run_compiled_code); |
|
439 } |
|
440 |
|
441 jmp(Address(method, Method::from_interpreted_offset())); |
|
442 |
|
443 } |
|
444 |
|
445 |
|
446 // The following two routines provide a hook so that an implementation |
|
447 // can schedule the dispatch in two parts. Intel does not do this. |
|
448 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) { |
|
449 // Nothing Intel-specific to be done here. |
|
450 } |
|
451 |
|
452 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { |
|
453 dispatch_next(state, step); |
|
454 } |
|
455 |
|
456 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table, |
|
457 bool verifyoop) { |
|
458 verify_FPU(1, state); |
|
459 if (VerifyActivationFrameSize) { |
|
460 Label L; |
|
461 mov(rcx, rbp); |
|
462 subptr(rcx, rsp); |
|
463 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize; |
|
464 cmpptr(rcx, min_frame_size); |
|
465 jcc(Assembler::greaterEqual, L); |
|
466 stop("broken stack frame"); |
|
467 bind(L); |
|
468 } |
|
469 if (verifyoop) verify_oop(rax, state); |
|
470 Address index(noreg, rbx, Address::times_ptr); |
|
471 ExternalAddress tbl((address)table); |
|
472 ArrayAddress dispatch(tbl, index); |
|
473 jump(dispatch); |
|
474 } |
|
475 |
|
476 |
|
477 void InterpreterMacroAssembler::dispatch_only(TosState state) { |
|
478 dispatch_base(state, Interpreter::dispatch_table(state)); |
|
479 } |
|
480 |
|
481 |
|
482 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) { |
|
483 dispatch_base(state, Interpreter::normal_table(state)); |
|
484 } |
|
485 |
|
486 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) { |
|
487 dispatch_base(state, Interpreter::normal_table(state), false); |
|
488 } |
|
489 |
|
490 |
|
491 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) { |
|
492 // load next bytecode (load before advancing rsi to prevent AGI) |
|
493 load_unsigned_byte(rbx, Address(rsi, step)); |
|
494 // advance rsi |
|
495 increment(rsi, step); |
|
496 dispatch_base(state, Interpreter::dispatch_table(state)); |
|
497 } |
|
498 |
|
499 |
|
500 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { |
|
501 // load current bytecode |
|
502 load_unsigned_byte(rbx, Address(rsi, 0)); |
|
503 dispatch_base(state, table); |
|
504 } |
|
505 |
|
506 // remove activation |
|
507 // |
|
508 // Unlock the receiver if this is a synchronized method. |
|
509 // Unlock any Java monitors from syncronized blocks. |
|
510 // Remove the activation from the stack. |
|
511 // |
|
512 // If there are locked Java monitors |
|
513 // If throw_monitor_exception |
|
514 // throws IllegalMonitorStateException |
|
515 // Else if install_monitor_exception |
|
516 // installs IllegalMonitorStateException |
|
517 // Else |
|
518 // no error processing |
|
519 void InterpreterMacroAssembler::remove_activation(TosState state, Register ret_addr, |
|
520 bool throw_monitor_exception, |
|
521 bool install_monitor_exception, |
|
522 bool notify_jvmdi) { |
|
523 // Note: Registers rax, rdx and FPU ST(0) may be in use for the result |
|
524 // check if synchronized method |
|
525 Label unlocked, unlock, no_unlock; |
|
526 |
|
527 get_thread(rcx); |
|
528 const Address do_not_unlock_if_synchronized(rcx, |
|
529 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); |
|
530 |
|
531 movbool(rbx, do_not_unlock_if_synchronized); |
|
532 mov(rdi,rbx); |
|
533 movbool(do_not_unlock_if_synchronized, false); // reset the flag |
|
534 |
|
535 movptr(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); // get method access flags |
|
536 movl(rcx, Address(rbx, Method::access_flags_offset())); |
|
537 |
|
538 testl(rcx, JVM_ACC_SYNCHRONIZED); |
|
539 jcc(Assembler::zero, unlocked); |
|
540 |
|
541 // Don't unlock anything if the _do_not_unlock_if_synchronized flag |
|
542 // is set. |
|
543 mov(rcx,rdi); |
|
544 testbool(rcx); |
|
545 jcc(Assembler::notZero, no_unlock); |
|
546 |
|
547 // unlock monitor |
|
548 push(state); // save result |
|
549 |
|
550 // BasicObjectLock will be first in list, since this is a synchronized method. However, need |
|
551 // to check that the object has not been unlocked by an explicit monitorexit bytecode. |
|
552 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock)); |
|
553 lea (rdx, monitor); // address of first monitor |
|
554 |
|
555 movptr (rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); |
|
556 testptr(rax, rax); |
|
557 jcc (Assembler::notZero, unlock); |
|
558 |
|
559 pop(state); |
|
560 if (throw_monitor_exception) { |
|
561 empty_FPU_stack(); // remove possible return value from FPU-stack, otherwise stack could overflow |
|
562 |
|
563 // Entry already unlocked, need to throw exception |
|
564 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); |
|
565 should_not_reach_here(); |
|
566 } else { |
|
567 // Monitor already unlocked during a stack unroll. |
|
568 // If requested, install an illegal_monitor_state_exception. |
|
569 // Continue with stack unrolling. |
|
570 if (install_monitor_exception) { |
|
571 empty_FPU_stack(); // remove possible return value from FPU-stack, otherwise stack could overflow |
|
572 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); |
|
573 } |
|
574 jmp(unlocked); |
|
575 } |
|
576 |
|
577 bind(unlock); |
|
578 unlock_object(rdx); |
|
579 pop(state); |
|
580 |
|
581 // Check that for block-structured locking (i.e., that all locked objects has been unlocked) |
|
582 bind(unlocked); |
|
583 |
|
584 // rax, rdx: Might contain return value |
|
585 |
|
586 // Check that all monitors are unlocked |
|
587 { |
|
588 Label loop, exception, entry, restart; |
|
589 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; |
|
590 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); |
|
591 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize); |
|
592 |
|
593 bind(restart); |
|
594 movptr(rcx, monitor_block_top); // points to current entry, starting with top-most entry |
|
595 lea(rbx, monitor_block_bot); // points to word before bottom of monitor block |
|
596 jmp(entry); |
|
597 |
|
598 // Entry already locked, need to throw exception |
|
599 bind(exception); |
|
600 |
|
601 if (throw_monitor_exception) { |
|
602 empty_FPU_stack(); // remove possible return value from FPU-stack, otherwise stack could overflow |
|
603 |
|
604 // Throw exception |
|
605 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); |
|
606 should_not_reach_here(); |
|
607 } else { |
|
608 // Stack unrolling. Unlock object and install illegal_monitor_exception |
|
609 // Unlock does not block, so don't have to worry about the frame |
|
610 |
|
611 push(state); |
|
612 mov(rdx, rcx); |
|
613 unlock_object(rdx); |
|
614 pop(state); |
|
615 |
|
616 if (install_monitor_exception) { |
|
617 empty_FPU_stack(); // remove possible return value from FPU-stack, otherwise stack could overflow |
|
618 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); |
|
619 } |
|
620 |
|
621 jmp(restart); |
|
622 } |
|
623 |
|
624 bind(loop); |
|
625 cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); // check if current entry is used |
|
626 jcc(Assembler::notEqual, exception); |
|
627 |
|
628 addptr(rcx, entry_size); // otherwise advance to next entry |
|
629 bind(entry); |
|
630 cmpptr(rcx, rbx); // check if bottom reached |
|
631 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry |
|
632 } |
|
633 |
|
634 bind(no_unlock); |
|
635 |
|
636 // jvmti support |
|
637 if (notify_jvmdi) { |
|
638 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA |
|
639 } else { |
|
640 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA |
|
641 } |
|
642 |
|
643 // remove activation |
|
644 movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp |
|
645 leave(); // remove frame anchor |
|
646 pop(ret_addr); // get return address |
|
647 mov(rsp, rbx); // set sp to sender sp |
|
648 if (UseSSE) { |
|
649 // float and double are returned in xmm register in SSE-mode |
|
650 if (state == ftos && UseSSE >= 1) { |
|
651 subptr(rsp, wordSize); |
|
652 fstp_s(Address(rsp, 0)); |
|
653 movflt(xmm0, Address(rsp, 0)); |
|
654 addptr(rsp, wordSize); |
|
655 } else if (state == dtos && UseSSE >= 2) { |
|
656 subptr(rsp, 2*wordSize); |
|
657 fstp_d(Address(rsp, 0)); |
|
658 movdbl(xmm0, Address(rsp, 0)); |
|
659 addptr(rsp, 2*wordSize); |
|
660 } |
|
661 } |
|
662 } |
|
663 |
|
664 #endif /* !CC_INTERP */ |
|
665 |
|
666 void InterpreterMacroAssembler::get_method_counters(Register method, |
|
667 Register mcs, Label& skip) { |
|
668 Label has_counters; |
|
669 movptr(mcs, Address(method, Method::method_counters_offset())); |
|
670 testptr(mcs, mcs); |
|
671 jcc(Assembler::notZero, has_counters); |
|
672 call_VM(noreg, CAST_FROM_FN_PTR(address, |
|
673 InterpreterRuntime::build_method_counters), method); |
|
674 movptr(mcs, Address(method,Method::method_counters_offset())); |
|
675 testptr(mcs, mcs); |
|
676 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory |
|
677 bind(has_counters); |
|
678 } |
|
679 |
|
680 |
|
681 // Lock object |
|
682 // |
|
683 // Argument: rdx : Points to BasicObjectLock to be used for locking. Must |
|
684 // be initialized with object to lock |
|
685 void InterpreterMacroAssembler::lock_object(Register lock_reg) { |
|
686 assert(lock_reg == rdx, "The argument is only for looks. It must be rdx"); |
|
687 |
|
688 if (UseHeavyMonitors) { |
|
689 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); |
|
690 } else { |
|
691 |
|
692 Label done; |
|
693 |
|
694 const Register swap_reg = rax; // Must use rax, for cmpxchg instruction |
|
695 const Register obj_reg = rcx; // Will contain the oop |
|
696 |
|
697 const int obj_offset = BasicObjectLock::obj_offset_in_bytes(); |
|
698 const int lock_offset = BasicObjectLock::lock_offset_in_bytes (); |
|
699 const int mark_offset = lock_offset + BasicLock::displaced_header_offset_in_bytes(); |
|
700 |
|
701 Label slow_case; |
|
702 |
|
703 // Load object pointer into obj_reg %rcx |
|
704 movptr(obj_reg, Address(lock_reg, obj_offset)); |
|
705 |
|
706 if (UseBiasedLocking) { |
|
707 // Note: we use noreg for the temporary register since it's hard |
|
708 // to come up with a free register on all incoming code paths |
|
709 biased_locking_enter(lock_reg, obj_reg, swap_reg, noreg, false, done, &slow_case); |
|
710 } |
|
711 |
|
712 // Load immediate 1 into swap_reg %rax, |
|
713 movptr(swap_reg, (int32_t)1); |
|
714 |
|
715 // Load (object->mark() | 1) into swap_reg %rax, |
|
716 orptr(swap_reg, Address(obj_reg, 0)); |
|
717 |
|
718 // Save (object->mark() | 1) into BasicLock's displaced header |
|
719 movptr(Address(lock_reg, mark_offset), swap_reg); |
|
720 |
|
721 assert(lock_offset == 0, "displached header must be first word in BasicObjectLock"); |
|
722 if (os::is_MP()) { |
|
723 lock(); |
|
724 } |
|
725 cmpxchgptr(lock_reg, Address(obj_reg, 0)); |
|
726 if (PrintBiasedLockingStatistics) { |
|
727 cond_inc32(Assembler::zero, |
|
728 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr())); |
|
729 } |
|
730 jcc(Assembler::zero, done); |
|
731 |
|
732 // Test if the oopMark is an obvious stack pointer, i.e., |
|
733 // 1) (mark & 3) == 0, and |
|
734 // 2) rsp <= mark < mark + os::pagesize() |
|
735 // |
|
736 // These 3 tests can be done by evaluating the following |
|
737 // expression: ((mark - rsp) & (3 - os::vm_page_size())), |
|
738 // assuming both stack pointer and pagesize have their |
|
739 // least significant 2 bits clear. |
|
740 // NOTE: the oopMark is in swap_reg %rax, as the result of cmpxchg |
|
741 subptr(swap_reg, rsp); |
|
742 andptr(swap_reg, 3 - os::vm_page_size()); |
|
743 |
|
744 // Save the test result, for recursive case, the result is zero |
|
745 movptr(Address(lock_reg, mark_offset), swap_reg); |
|
746 |
|
747 if (PrintBiasedLockingStatistics) { |
|
748 cond_inc32(Assembler::zero, |
|
749 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr())); |
|
750 } |
|
751 jcc(Assembler::zero, done); |
|
752 |
|
753 bind(slow_case); |
|
754 |
|
755 // Call the runtime routine for slow case |
|
756 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); |
|
757 |
|
758 bind(done); |
|
759 } |
|
760 } |
|
761 |
|
762 |
|
763 // Unlocks an object. Used in monitorexit bytecode and remove_activation. |
|
764 // |
|
765 // Argument: rdx : Points to BasicObjectLock structure for lock |
|
766 // Throw an IllegalMonitorException if object is not locked by current thread |
|
767 // |
|
768 // Uses: rax, rbx, rcx, rdx |
|
769 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { |
|
770 assert(lock_reg == rdx, "The argument is only for looks. It must be rdx"); |
|
771 |
|
772 if (UseHeavyMonitors) { |
|
773 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); |
|
774 } else { |
|
775 Label done; |
|
776 |
|
777 const Register swap_reg = rax; // Must use rax, for cmpxchg instruction |
|
778 const Register header_reg = rbx; // Will contain the old oopMark |
|
779 const Register obj_reg = rcx; // Will contain the oop |
|
780 |
|
781 save_bcp(); // Save in case of exception |
|
782 |
|
783 // Convert from BasicObjectLock structure to object and BasicLock structure |
|
784 // Store the BasicLock address into %rax, |
|
785 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes())); |
|
786 |
|
787 // Load oop into obj_reg(%rcx) |
|
788 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes ())); |
|
789 |
|
790 // Free entry |
|
791 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), NULL_WORD); |
|
792 |
|
793 if (UseBiasedLocking) { |
|
794 biased_locking_exit(obj_reg, header_reg, done); |
|
795 } |
|
796 |
|
797 // Load the old header from BasicLock structure |
|
798 movptr(header_reg, Address(swap_reg, BasicLock::displaced_header_offset_in_bytes())); |
|
799 |
|
800 // Test for recursion |
|
801 testptr(header_reg, header_reg); |
|
802 |
|
803 // zero for recursive case |
|
804 jcc(Assembler::zero, done); |
|
805 |
|
806 // Atomic swap back the old header |
|
807 if (os::is_MP()) lock(); |
|
808 cmpxchgptr(header_reg, Address(obj_reg, 0)); |
|
809 |
|
810 // zero for recursive case |
|
811 jcc(Assembler::zero, done); |
|
812 |
|
813 // Call the runtime routine for slow case. |
|
814 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), obj_reg); // restore obj |
|
815 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); |
|
816 |
|
817 bind(done); |
|
818 |
|
819 restore_bcp(); |
|
820 } |
|
821 } |
|
822 |
|
823 |
|
824 #ifndef CC_INTERP |
|
825 |
|
826 // Test ImethodDataPtr. If it is null, continue at the specified label |
|
827 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) { |
|
828 assert(ProfileInterpreter, "must be profiling interpreter"); |
|
829 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize)); |
|
830 testptr(mdp, mdp); |
|
831 jcc(Assembler::zero, zero_continue); |
|
832 } |
|
833 |
|
834 |
|
835 // Set the method data pointer for the current bcp. |
|
836 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { |
|
837 assert(ProfileInterpreter, "must be profiling interpreter"); |
|
838 Label set_mdp; |
|
839 push(rax); |
|
840 push(rbx); |
|
841 |
|
842 get_method(rbx); |
|
843 // Test MDO to avoid the call if it is NULL. |
|
844 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset()))); |
|
845 testptr(rax, rax); |
|
846 jcc(Assembler::zero, set_mdp); |
|
847 // rbx,: method |
|
848 // rsi: bcp |
|
849 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, rsi); |
|
850 // rax,: mdi |
|
851 // mdo is guaranteed to be non-zero here, we checked for it before the call. |
|
852 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset()))); |
|
853 addptr(rbx, in_bytes(MethodData::data_offset())); |
|
854 addptr(rax, rbx); |
|
855 bind(set_mdp); |
|
856 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax); |
|
857 pop(rbx); |
|
858 pop(rax); |
|
859 } |
|
860 |
|
861 void InterpreterMacroAssembler::verify_method_data_pointer() { |
|
862 assert(ProfileInterpreter, "must be profiling interpreter"); |
|
863 #ifdef ASSERT |
|
864 Label verify_continue; |
|
865 push(rax); |
|
866 push(rbx); |
|
867 push(rcx); |
|
868 push(rdx); |
|
869 test_method_data_pointer(rcx, verify_continue); // If mdp is zero, continue |
|
870 get_method(rbx); |
|
871 |
|
872 // If the mdp is valid, it will point to a DataLayout header which is |
|
873 // consistent with the bcp. The converse is highly probable also. |
|
874 load_unsigned_short(rdx, Address(rcx, in_bytes(DataLayout::bci_offset()))); |
|
875 addptr(rdx, Address(rbx, Method::const_offset())); |
|
876 lea(rdx, Address(rdx, ConstMethod::codes_offset())); |
|
877 cmpptr(rdx, rsi); |
|
878 jcc(Assembler::equal, verify_continue); |
|
879 // rbx,: method |
|
880 // rsi: bcp |
|
881 // rcx: mdp |
|
882 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), rbx, rsi, rcx); |
|
883 bind(verify_continue); |
|
884 pop(rdx); |
|
885 pop(rcx); |
|
886 pop(rbx); |
|
887 pop(rax); |
|
888 #endif // ASSERT |
|
889 } |
|
890 |
|
891 |
|
892 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) { |
|
893 // %%% this seems to be used to store counter data which is surely 32bits |
|
894 // however 64bit side stores 64 bits which seems wrong |
|
895 assert(ProfileInterpreter, "must be profiling interpreter"); |
|
896 Address data(mdp_in, constant); |
|
897 movptr(data, value); |
|
898 } |
|
899 |
|
900 |
|
901 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, |
|
902 int constant, |
|
903 bool decrement) { |
|
904 // Counter address |
|
905 Address data(mdp_in, constant); |
|
906 |
|
907 increment_mdp_data_at(data, decrement); |
|
908 } |
|
909 |
|
910 |
|
911 void InterpreterMacroAssembler::increment_mdp_data_at(Address data, |
|
912 bool decrement) { |
|
913 |
|
914 assert( DataLayout::counter_increment==1, "flow-free idiom only works with 1" ); |
|
915 assert(ProfileInterpreter, "must be profiling interpreter"); |
|
916 |
|
917 // %%% 64bit treats this as 64 bit which seems unlikely |
|
918 if (decrement) { |
|
919 // Decrement the register. Set condition codes. |
|
920 addl(data, -DataLayout::counter_increment); |
|
921 // If the decrement causes the counter to overflow, stay negative |
|
922 Label L; |
|
923 jcc(Assembler::negative, L); |
|
924 addl(data, DataLayout::counter_increment); |
|
925 bind(L); |
|
926 } else { |
|
927 assert(DataLayout::counter_increment == 1, |
|
928 "flow-free idiom only works with 1"); |
|
929 // Increment the register. Set carry flag. |
|
930 addl(data, DataLayout::counter_increment); |
|
931 // If the increment causes the counter to overflow, pull back by 1. |
|
932 sbbl(data, 0); |
|
933 } |
|
934 } |
|
935 |
|
936 |
|
937 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, |
|
938 Register reg, |
|
939 int constant, |
|
940 bool decrement) { |
|
941 Address data(mdp_in, reg, Address::times_1, constant); |
|
942 |
|
943 increment_mdp_data_at(data, decrement); |
|
944 } |
|
945 |
|
946 |
|
947 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, int flag_byte_constant) { |
|
948 assert(ProfileInterpreter, "must be profiling interpreter"); |
|
949 int header_offset = in_bytes(DataLayout::header_offset()); |
|
950 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant); |
|
951 // Set the flag |
|
952 orl(Address(mdp_in, header_offset), header_bits); |
|
953 } |
|
954 |
|
955 |
|
956 |
|
957 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, |
|
958 int offset, |
|
959 Register value, |
|
960 Register test_value_out, |
|
961 Label& not_equal_continue) { |
|
962 assert(ProfileInterpreter, "must be profiling interpreter"); |
|
963 if (test_value_out == noreg) { |
|
964 cmpptr(value, Address(mdp_in, offset)); |
|
965 } else { |
|
966 // Put the test value into a register, so caller can use it: |
|
967 movptr(test_value_out, Address(mdp_in, offset)); |
|
968 cmpptr(test_value_out, value); |
|
969 } |
|
970 jcc(Assembler::notEqual, not_equal_continue); |
|
971 } |
|
972 |
|
973 |
|
974 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) { |
|
975 assert(ProfileInterpreter, "must be profiling interpreter"); |
|
976 Address disp_address(mdp_in, offset_of_disp); |
|
977 addptr(mdp_in,disp_address); |
|
978 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); |
|
979 } |
|
980 |
|
981 |
|
982 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, Register reg, int offset_of_disp) { |
|
983 assert(ProfileInterpreter, "must be profiling interpreter"); |
|
984 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp); |
|
985 addptr(mdp_in, disp_address); |
|
986 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); |
|
987 } |
|
988 |
|
989 |
|
990 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) { |
|
991 assert(ProfileInterpreter, "must be profiling interpreter"); |
|
992 addptr(mdp_in, constant); |
|
993 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); |
|
994 } |
|
995 |
|
996 |
|
997 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { |
|
998 assert(ProfileInterpreter, "must be profiling interpreter"); |
|
999 push(return_bci); // save/restore across call_VM |
|
1000 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); |
|
1001 pop(return_bci); |
|
1002 } |
|
1003 |
|
1004 |
|
1005 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) { |
|
1006 if (ProfileInterpreter) { |
|
1007 Label profile_continue; |
|
1008 |
|
1009 // If no method data exists, go to profile_continue. |
|
1010 // Otherwise, assign to mdp |
|
1011 test_method_data_pointer(mdp, profile_continue); |
|
1012 |
|
1013 // We are taking a branch. Increment the taken count. |
|
1014 // We inline increment_mdp_data_at to return bumped_count in a register |
|
1015 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); |
|
1016 Address data(mdp, in_bytes(JumpData::taken_offset())); |
|
1017 |
|
1018 // %%% 64bit treats these cells as 64 bit but they seem to be 32 bit |
|
1019 movl(bumped_count,data); |
|
1020 assert( DataLayout::counter_increment==1, "flow-free idiom only works with 1" ); |
|
1021 addl(bumped_count, DataLayout::counter_increment); |
|
1022 sbbl(bumped_count, 0); |
|
1023 movl(data,bumped_count); // Store back out |
|
1024 |
|
1025 // The method data pointer needs to be updated to reflect the new target. |
|
1026 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); |
|
1027 bind (profile_continue); |
|
1028 } |
|
1029 } |
|
1030 |
|
1031 |
|
1032 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { |
|
1033 if (ProfileInterpreter) { |
|
1034 Label profile_continue; |
|
1035 |
|
1036 // If no method data exists, go to profile_continue. |
|
1037 test_method_data_pointer(mdp, profile_continue); |
|
1038 |
|
1039 // We are taking a branch. Increment the not taken count. |
|
1040 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset())); |
|
1041 |
|
1042 // The method data pointer needs to be updated to correspond to the next bytecode |
|
1043 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); |
|
1044 bind (profile_continue); |
|
1045 } |
|
1046 } |
|
1047 |
|
1048 void InterpreterMacroAssembler::profile_call(Register mdp) { |
|
1049 if (ProfileInterpreter) { |
|
1050 Label profile_continue; |
|
1051 |
|
1052 // If no method data exists, go to profile_continue. |
|
1053 test_method_data_pointer(mdp, profile_continue); |
|
1054 |
|
1055 // We are making a call. Increment the count. |
|
1056 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); |
|
1057 |
|
1058 // The method data pointer needs to be updated to reflect the new target. |
|
1059 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); |
|
1060 bind (profile_continue); |
|
1061 } |
|
1062 } |
|
1063 |
|
1064 |
|
1065 void InterpreterMacroAssembler::profile_final_call(Register mdp) { |
|
1066 if (ProfileInterpreter) { |
|
1067 Label profile_continue; |
|
1068 |
|
1069 // If no method data exists, go to profile_continue. |
|
1070 test_method_data_pointer(mdp, profile_continue); |
|
1071 |
|
1072 // We are making a call. Increment the count. |
|
1073 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); |
|
1074 |
|
1075 // The method data pointer needs to be updated to reflect the new target. |
|
1076 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); |
|
1077 bind (profile_continue); |
|
1078 } |
|
1079 } |
|
1080 |
|
1081 |
|
1082 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, Register mdp, |
|
1083 Register reg2, |
|
1084 bool receiver_can_be_null) { |
|
1085 if (ProfileInterpreter) { |
|
1086 Label profile_continue; |
|
1087 |
|
1088 // If no method data exists, go to profile_continue. |
|
1089 test_method_data_pointer(mdp, profile_continue); |
|
1090 |
|
1091 Label skip_receiver_profile; |
|
1092 if (receiver_can_be_null) { |
|
1093 Label not_null; |
|
1094 testptr(receiver, receiver); |
|
1095 jccb(Assembler::notZero, not_null); |
|
1096 // We are making a call. Increment the count for null receiver. |
|
1097 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); |
|
1098 jmp(skip_receiver_profile); |
|
1099 bind(not_null); |
|
1100 } |
|
1101 |
|
1102 // Record the receiver type. |
|
1103 record_klass_in_profile(receiver, mdp, reg2, true); |
|
1104 bind(skip_receiver_profile); |
|
1105 |
|
1106 // The method data pointer needs to be updated to reflect the new target. |
|
1107 update_mdp_by_constant(mdp, |
|
1108 in_bytes(VirtualCallData:: |
|
1109 virtual_call_data_size())); |
|
1110 bind(profile_continue); |
|
1111 } |
|
1112 } |
|
1113 |
|
1114 |
|
1115 void InterpreterMacroAssembler::record_klass_in_profile_helper( |
|
1116 Register receiver, Register mdp, |
|
1117 Register reg2, int start_row, |
|
1118 Label& done, bool is_virtual_call) { |
|
1119 if (TypeProfileWidth == 0) { |
|
1120 if (is_virtual_call) { |
|
1121 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); |
|
1122 } |
|
1123 return; |
|
1124 } |
|
1125 |
|
1126 int last_row = VirtualCallData::row_limit() - 1; |
|
1127 assert(start_row <= last_row, "must be work left to do"); |
|
1128 // Test this row for both the receiver and for null. |
|
1129 // Take any of three different outcomes: |
|
1130 // 1. found receiver => increment count and goto done |
|
1131 // 2. found null => keep looking for case 1, maybe allocate this cell |
|
1132 // 3. found something else => keep looking for cases 1 and 2 |
|
1133 // Case 3 is handled by a recursive call. |
|
1134 for (int row = start_row; row <= last_row; row++) { |
|
1135 Label next_test; |
|
1136 bool test_for_null_also = (row == start_row); |
|
1137 |
|
1138 // See if the receiver is receiver[n]. |
|
1139 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); |
|
1140 test_mdp_data_at(mdp, recvr_offset, receiver, |
|
1141 (test_for_null_also ? reg2 : noreg), |
|
1142 next_test); |
|
1143 // (Reg2 now contains the receiver from the CallData.) |
|
1144 |
|
1145 // The receiver is receiver[n]. Increment count[n]. |
|
1146 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); |
|
1147 increment_mdp_data_at(mdp, count_offset); |
|
1148 jmp(done); |
|
1149 bind(next_test); |
|
1150 |
|
1151 if (row == start_row) { |
|
1152 Label found_null; |
|
1153 // Failed the equality check on receiver[n]... Test for null. |
|
1154 testptr(reg2, reg2); |
|
1155 if (start_row == last_row) { |
|
1156 // The only thing left to do is handle the null case. |
|
1157 if (is_virtual_call) { |
|
1158 jccb(Assembler::zero, found_null); |
|
1159 // Receiver did not match any saved receiver and there is no empty row for it. |
|
1160 // Increment total counter to indicate polymorphic case. |
|
1161 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); |
|
1162 jmp(done); |
|
1163 bind(found_null); |
|
1164 } else { |
|
1165 jcc(Assembler::notZero, done); |
|
1166 } |
|
1167 break; |
|
1168 } |
|
1169 // Since null is rare, make it be the branch-taken case. |
|
1170 jcc(Assembler::zero, found_null); |
|
1171 |
|
1172 // Put all the "Case 3" tests here. |
|
1173 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call); |
|
1174 |
|
1175 // Found a null. Keep searching for a matching receiver, |
|
1176 // but remember that this is an empty (unused) slot. |
|
1177 bind(found_null); |
|
1178 } |
|
1179 } |
|
1180 |
|
1181 // In the fall-through case, we found no matching receiver, but we |
|
1182 // observed the receiver[start_row] is NULL. |
|
1183 |
|
1184 // Fill in the receiver field and increment the count. |
|
1185 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); |
|
1186 set_mdp_data_at(mdp, recvr_offset, receiver); |
|
1187 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); |
|
1188 movptr(reg2, (intptr_t)DataLayout::counter_increment); |
|
1189 set_mdp_data_at(mdp, count_offset, reg2); |
|
1190 if (start_row > 0) { |
|
1191 jmp(done); |
|
1192 } |
|
1193 } |
|
1194 |
|
1195 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, |
|
1196 Register mdp, Register reg2, |
|
1197 bool is_virtual_call) { |
|
1198 assert(ProfileInterpreter, "must be profiling"); |
|
1199 Label done; |
|
1200 |
|
1201 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); |
|
1202 |
|
1203 bind (done); |
|
1204 } |
|
1205 |
|
1206 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) { |
|
1207 if (ProfileInterpreter) { |
|
1208 Label profile_continue; |
|
1209 uint row; |
|
1210 |
|
1211 // If no method data exists, go to profile_continue. |
|
1212 test_method_data_pointer(mdp, profile_continue); |
|
1213 |
|
1214 // Update the total ret count. |
|
1215 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); |
|
1216 |
|
1217 for (row = 0; row < RetData::row_limit(); row++) { |
|
1218 Label next_test; |
|
1219 |
|
1220 // See if return_bci is equal to bci[n]: |
|
1221 test_mdp_data_at(mdp, in_bytes(RetData::bci_offset(row)), return_bci, |
|
1222 noreg, next_test); |
|
1223 |
|
1224 // return_bci is equal to bci[n]. Increment the count. |
|
1225 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); |
|
1226 |
|
1227 // The method data pointer needs to be updated to reflect the new target. |
|
1228 update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row))); |
|
1229 jmp(profile_continue); |
|
1230 bind(next_test); |
|
1231 } |
|
1232 |
|
1233 update_mdp_for_ret(return_bci); |
|
1234 |
|
1235 bind (profile_continue); |
|
1236 } |
|
1237 } |
|
1238 |
|
1239 |
|
1240 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { |
|
1241 if (ProfileInterpreter) { |
|
1242 Label profile_continue; |
|
1243 |
|
1244 // If no method data exists, go to profile_continue. |
|
1245 test_method_data_pointer(mdp, profile_continue); |
|
1246 |
|
1247 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); |
|
1248 |
|
1249 // The method data pointer needs to be updated. |
|
1250 int mdp_delta = in_bytes(BitData::bit_data_size()); |
|
1251 if (TypeProfileCasts) { |
|
1252 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); |
|
1253 } |
|
1254 update_mdp_by_constant(mdp, mdp_delta); |
|
1255 |
|
1256 bind (profile_continue); |
|
1257 } |
|
1258 } |
|
1259 |
|
1260 |
|
1261 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) { |
|
1262 if (ProfileInterpreter && TypeProfileCasts) { |
|
1263 Label profile_continue; |
|
1264 |
|
1265 // If no method data exists, go to profile_continue. |
|
1266 test_method_data_pointer(mdp, profile_continue); |
|
1267 |
|
1268 int count_offset = in_bytes(CounterData::count_offset()); |
|
1269 // Back up the address, since we have already bumped the mdp. |
|
1270 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); |
|
1271 |
|
1272 // *Decrement* the counter. We expect to see zero or small negatives. |
|
1273 increment_mdp_data_at(mdp, count_offset, true); |
|
1274 |
|
1275 bind (profile_continue); |
|
1276 } |
|
1277 } |
|
1278 |
|
1279 |
|
1280 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) |
|
1281 { |
|
1282 if (ProfileInterpreter) { |
|
1283 Label profile_continue; |
|
1284 |
|
1285 // If no method data exists, go to profile_continue. |
|
1286 test_method_data_pointer(mdp, profile_continue); |
|
1287 |
|
1288 // The method data pointer needs to be updated. |
|
1289 int mdp_delta = in_bytes(BitData::bit_data_size()); |
|
1290 if (TypeProfileCasts) { |
|
1291 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); |
|
1292 |
|
1293 // Record the object type. |
|
1294 record_klass_in_profile(klass, mdp, reg2, false); |
|
1295 assert(reg2 == rdi, "we know how to fix this blown reg"); |
|
1296 restore_locals(); // Restore EDI |
|
1297 } |
|
1298 update_mdp_by_constant(mdp, mdp_delta); |
|
1299 |
|
1300 bind(profile_continue); |
|
1301 } |
|
1302 } |
|
1303 |
|
1304 |
|
1305 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { |
|
1306 if (ProfileInterpreter) { |
|
1307 Label profile_continue; |
|
1308 |
|
1309 // If no method data exists, go to profile_continue. |
|
1310 test_method_data_pointer(mdp, profile_continue); |
|
1311 |
|
1312 // Update the default case count |
|
1313 increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset())); |
|
1314 |
|
1315 // The method data pointer needs to be updated. |
|
1316 update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset())); |
|
1317 |
|
1318 bind (profile_continue); |
|
1319 } |
|
1320 } |
|
1321 |
|
1322 |
|
1323 void InterpreterMacroAssembler::profile_switch_case(Register index, Register mdp, Register reg2) { |
|
1324 if (ProfileInterpreter) { |
|
1325 Label profile_continue; |
|
1326 |
|
1327 // If no method data exists, go to profile_continue. |
|
1328 test_method_data_pointer(mdp, profile_continue); |
|
1329 |
|
1330 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() |
|
1331 movptr(reg2, (intptr_t)in_bytes(MultiBranchData::per_case_size())); |
|
1332 // index is positive and so should have correct value if this code were |
|
1333 // used on 64bits |
|
1334 imulptr(index, reg2); |
|
1335 addptr(index, in_bytes(MultiBranchData::case_array_offset())); |
|
1336 |
|
1337 // Update the case count |
|
1338 increment_mdp_data_at(mdp, index, in_bytes(MultiBranchData::relative_count_offset())); |
|
1339 |
|
1340 // The method data pointer needs to be updated. |
|
1341 update_mdp_by_offset(mdp, index, in_bytes(MultiBranchData::relative_displacement_offset())); |
|
1342 |
|
1343 bind (profile_continue); |
|
1344 } |
|
1345 } |
|
1346 |
|
1347 #endif // !CC_INTERP |
|
1348 |
|
1349 |
|
1350 |
|
1351 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { |
|
1352 if (state == atos) MacroAssembler::verify_oop(reg); |
|
1353 } |
|
1354 |
|
1355 |
|
1356 #ifndef CC_INTERP |
|
1357 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { |
|
1358 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); |
|
1359 } |
|
1360 |
|
1361 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. |
|
1362 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, |
|
1363 int increment, Address mask, |
|
1364 Register scratch, bool preloaded, |
|
1365 Condition cond, Label* where) { |
|
1366 if (!preloaded) { |
|
1367 movl(scratch, counter_addr); |
|
1368 } |
|
1369 incrementl(scratch, increment); |
|
1370 movl(counter_addr, scratch); |
|
1371 andl(scratch, mask); |
|
1372 jcc(cond, *where); |
|
1373 } |
|
1374 #endif /* CC_INTERP */ |
|
1375 |
|
1376 |
|
1377 void InterpreterMacroAssembler::notify_method_entry() { |
|
1378 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to |
|
1379 // track stack depth. If it is possible to enter interp_only_mode we add |
|
1380 // the code to check if the event should be sent. |
|
1381 if (JvmtiExport::can_post_interpreter_events()) { |
|
1382 Label L; |
|
1383 get_thread(rcx); |
|
1384 movl(rcx, Address(rcx, JavaThread::interp_only_mode_offset())); |
|
1385 testl(rcx,rcx); |
|
1386 jcc(Assembler::zero, L); |
|
1387 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); |
|
1388 bind(L); |
|
1389 } |
|
1390 |
|
1391 { |
|
1392 SkipIfEqual skip_if(this, &DTraceMethodProbes, 0); |
|
1393 get_thread(rcx); |
|
1394 get_method(rbx); |
|
1395 call_VM_leaf( |
|
1396 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), rcx, rbx); |
|
1397 } |
|
1398 |
|
1399 // RedefineClasses() tracing support for obsolete method entry |
|
1400 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { |
|
1401 get_thread(rcx); |
|
1402 get_method(rbx); |
|
1403 call_VM_leaf( |
|
1404 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), |
|
1405 rcx, rbx); |
|
1406 } |
|
1407 } |
|
1408 |
|
1409 |
|
1410 void InterpreterMacroAssembler::notify_method_exit( |
|
1411 TosState state, NotifyMethodExitMode mode) { |
|
1412 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to |
|
1413 // track stack depth. If it is possible to enter interp_only_mode we add |
|
1414 // the code to check if the event should be sent. |
|
1415 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { |
|
1416 Label L; |
|
1417 // Note: frame::interpreter_frame_result has a dependency on how the |
|
1418 // method result is saved across the call to post_method_exit. If this |
|
1419 // is changed then the interpreter_frame_result implementation will |
|
1420 // need to be updated too. |
|
1421 |
|
1422 // For c++ interpreter the result is always stored at a known location in the frame |
|
1423 // template interpreter will leave it on the top of the stack. |
|
1424 NOT_CC_INTERP(push(state);) |
|
1425 get_thread(rcx); |
|
1426 movl(rcx, Address(rcx, JavaThread::interp_only_mode_offset())); |
|
1427 testl(rcx,rcx); |
|
1428 jcc(Assembler::zero, L); |
|
1429 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); |
|
1430 bind(L); |
|
1431 NOT_CC_INTERP(pop(state);) |
|
1432 } |
|
1433 |
|
1434 { |
|
1435 SkipIfEqual skip_if(this, &DTraceMethodProbes, 0); |
|
1436 NOT_CC_INTERP(push(state)); |
|
1437 get_thread(rbx); |
|
1438 get_method(rcx); |
|
1439 call_VM_leaf( |
|
1440 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), |
|
1441 rbx, rcx); |
|
1442 NOT_CC_INTERP(pop(state)); |
|
1443 } |
|
1444 } |
|