1 /*
2 * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2013, 2014 SAP AG. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #ifndef CC_INTERP
28 #include "asm/macroAssembler.inline.hpp"
29 #include "interpreter/bytecodeHistogram.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "interpreter/interpreterGenerator.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "interpreter/interp_masm.hpp"
34 #include "interpreter/templateTable.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/methodData.hpp"
37 #include "oops/method.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "prims/jvmtiExport.hpp"
40 #include "prims/jvmtiThreadState.hpp"
41 #include "runtime/arguments.hpp"
42 #include "runtime/deoptimization.hpp"
43 #include "runtime/frame.inline.hpp"
44 #include "runtime/sharedRuntime.hpp"
45 #include "runtime/stubRoutines.hpp"
46 #include "runtime/synchronizer.hpp"
47 #include "runtime/timer.hpp"
48 #include "runtime/vframeArray.hpp"
49 #include "utilities/debug.hpp"
50 #include "utilities/macros.hpp"
51
52 #undef __
53 #define __ _masm->
54
55 #ifdef PRODUCT
56 #define BLOCK_COMMENT(str) /* nothing */
57 #else
58 #define BLOCK_COMMENT(str) __ block_comment(str)
59 #endif
60
61 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
62
63 //-----------------------------------------------------------------------------
64
65 // Actually we should never reach here since we do stack overflow checks before pushing any frame.
66 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
67 address entry = __ pc();
68 __ unimplemented("generate_StackOverflowError_handler");
69 return entry;
70 }
71
72 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
73 address entry = __ pc();
74 __ empty_expression_stack();
75 __ load_const_optimized(R4_ARG2, (address) name);
76 // Index is in R17_tos.
77 __ mr(R5_ARG3, R17_tos);
78 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException));
79 return entry;
80 }
81
82 #if 0
83 // Call special ClassCastException constructor taking object to cast
84 // and target class as arguments.
85 address TemplateInterpreterGenerator::generate_ClassCastException_verbose_handler() {
86 address entry = __ pc();
87
88 // Expression stack must be empty before entering the VM if an
89 // exception happened.
90 __ empty_expression_stack();
91
92 // Thread will be loaded to R3_ARG1.
93 // Target class oop is in register R5_ARG3 by convention!
94 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose), R17_tos, R5_ARG3);
95 // Above call must not return here since exception pending.
96 DEBUG_ONLY(__ should_not_reach_here();)
97 return entry;
98 }
99 #endif
100
101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
102 address entry = __ pc();
103 // Expression stack must be empty before entering the VM if an
104 // exception happened.
105 __ empty_expression_stack();
106
107 // Load exception object.
108 // Thread will be loaded to R3_ARG1.
109 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
110 #ifdef ASSERT
111 // Above call must not return here since exception pending.
112 __ should_not_reach_here();
113 #endif
114 return entry;
115 }
116
117 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
118 address entry = __ pc();
119 //__ untested("generate_exception_handler_common");
120 Register Rexception = R17_tos;
121
122 // Expression stack must be empty before entering the VM if an exception happened.
123 __ empty_expression_stack();
124
125 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
126 if (pass_oop) {
127 __ mr(R5_ARG3, Rexception);
128 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false);
129 } else {
130 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
131 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false);
132 }
133
134 // Throw exception.
135 __ mr(R3_ARG1, Rexception);
136 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
137 __ mtctr(R11_scratch1);
138 __ bctr();
139
140 return entry;
141 }
142
143 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
144 address entry = __ pc();
145 __ unimplemented("generate_continuation_for");
146 return entry;
147 }
148
149 // This entry is returned to when a call returns to the interpreter.
150 // When we arrive here, we expect that the callee stack frame is already popped.
151 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
152 address entry = __ pc();
153
154 // Move the value out of the return register back to the TOS cache of current frame.
155 switch (state) {
156 case ltos:
157 case btos:
158 case ctos:
159 case stos:
160 case atos:
161 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache
162 case ftos:
163 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
164 case vtos: break; // Nothing to do, this was a void return.
165 default : ShouldNotReachHere();
166 }
167
168 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
169 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
170 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
171
172 // Compiled code destroys templateTableBase, reload.
173 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
174
175 if (state == atos) {
176 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
177 }
178
179 const Register cache = R11_scratch1;
180 const Register size = R12_scratch2;
181 __ get_cache_and_index_at_bcp(cache, 1, index_size);
182
183 // Get least significant byte of 64 bit value:
184 #if defined(VM_LITTLE_ENDIAN)
185 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache);
186 #else
187 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
188 #endif
189 __ sldi(size, size, Interpreter::logStackElementSize);
190 __ add(R15_esp, R15_esp, size);
191 __ dispatch_next(state, step);
192 return entry;
193 }
194
195 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
196 address entry = __ pc();
197 // If state != vtos, we're returning from a native method, which put it's result
198 // into the result register. So move the value out of the return register back
199 // to the TOS cache of current frame.
200
201 switch (state) {
202 case ltos:
203 case btos:
204 case ctos:
205 case stos:
206 case atos:
207 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache
208 case ftos:
209 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
210 case vtos: break; // Nothing to do, this was a void return.
211 default : ShouldNotReachHere();
212 }
213
214 // Load LcpoolCache @@@ should be already set!
215 __ get_constant_pool_cache(R27_constPoolCache);
216
217 // Handle a pending exception, fall through if none.
218 __ check_and_forward_exception(R11_scratch1, R12_scratch2);
219
220 // Start executing bytecodes.
221 __ dispatch_next(state, step);
222
223 return entry;
224 }
225
226 // A result handler converts the native result into java format.
227 // Use the shared code between c++ and template interpreter.
228 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
229 return AbstractInterpreterGenerator::generate_result_handler_for(type);
230 }
231
232 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
233 address entry = __ pc();
234
235 __ push(state);
236 __ call_VM(noreg, runtime_entry);
237 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
238
239 return entry;
240 }
241
242 // Helpers for commoning out cases in the various type of method entries.
243
244 // Increment invocation count & check for overflow.
245 //
246 // Note: checking for negative value instead of overflow
247 // so we have a 'sticky' overflow test.
248 //
249 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
250 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
251 Register Rscratch1 = R11_scratch1;
252 Register Rscratch2 = R12_scratch2;
253 Register R3_counters = R3_ARG1;
254 Label done;
255
256 if (TieredCompilation) {
257 const int increment = InvocationCounter::count_increment;
258 const int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
259 Label no_mdo;
260 if (ProfileInterpreter) {
261 const Register Rmdo = Rscratch1;
262 // If no method data exists, go to profile_continue.
263 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
264 __ cmpdi(CCR0, Rmdo, 0);
265 __ beq(CCR0, no_mdo);
266
267 // Increment invocation counter in the MDO.
268 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
269 __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
270 __ addi(Rscratch2, Rscratch2, increment);
271 __ stw(Rscratch2, mdo_ic_offs, Rmdo);
272 __ load_const_optimized(Rscratch1, mask, R0);
273 __ and_(Rscratch1, Rscratch2, Rscratch1);
274 __ bne(CCR0, done);
275 __ b(*overflow);
276 }
277
278 // Increment counter in MethodCounters*.
279 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
280 __ bind(no_mdo);
281 __ get_method_counters(R19_method, R3_counters, done);
282 __ lwz(Rscratch2, mo_ic_offs, R3_counters);
283 __ addi(Rscratch2, Rscratch2, increment);
284 __ stw(Rscratch2, mo_ic_offs, R3_counters);
285 __ load_const_optimized(Rscratch1, mask, R0);
286 __ and_(Rscratch1, Rscratch2, Rscratch1);
287 __ beq(CCR0, *overflow);
288
289 __ bind(done);
290
291 } else {
292
293 // Update standard invocation counters.
294 Register Rsum_ivc_bec = R4_ARG2;
295 __ get_method_counters(R19_method, R3_counters, done);
296 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2);
297 // Increment interpreter invocation counter.
298 if (ProfileInterpreter) { // %%% Merge this into methodDataOop.
299 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
300 __ addi(R12_scratch2, R12_scratch2, 1);
301 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
302 }
303 // Check if we must create a method data obj.
304 if (ProfileInterpreter && profile_method != NULL) {
305 const Register profile_limit = Rscratch1;
306 int pl_offs = __ load_const_optimized(profile_limit, &InvocationCounter::InterpreterProfileLimit, R0, true);
307 __ lwz(profile_limit, pl_offs, profile_limit);
308 // Test to see if we should create a method data oop.
309 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit);
310 __ blt(CCR0, *profile_method_continue);
311 // If no method data exists, go to profile_method.
312 __ test_method_data_pointer(*profile_method);
313 }
314 // Finally check for counter overflow.
315 if (overflow) {
316 const Register invocation_limit = Rscratch1;
317 int il_offs = __ load_const_optimized(invocation_limit, &InvocationCounter::InterpreterInvocationLimit, R0, true);
318 __ lwz(invocation_limit, il_offs, invocation_limit);
319 assert(4 == sizeof(InvocationCounter::InterpreterInvocationLimit), "unexpected field size");
320 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit);
321 __ bge(CCR0, *overflow);
322 }
323
324 __ bind(done);
325 }
326 }
327
328 // Generate code to initiate compilation on invocation counter overflow.
329 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
330 // Generate code to initiate compilation on the counter overflow.
331
332 // InterpreterRuntime::frequency_counter_overflow takes one arguments,
333 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp)
334 // We pass zero in.
335 // The call returns the address of the verified entry point for the method or NULL
336 // if the compilation did not complete (either went background or bailed out).
337 //
338 // Unlike the C++ interpreter above: Check exceptions!
339 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
340 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
341
342 __ li(R4_ARG2, 0);
343 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
344
345 // Returns verified_entry_point or NULL.
346 // We ignore it in any case.
347 __ b(continue_entry);
348 }
349
350 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
351 assert_different_registers(Rmem_frame_size, Rscratch1);
352 __ generate_stack_overflow_check_with_compare_and_throw(Rmem_frame_size, Rscratch1);
353 }
354
355 void TemplateInterpreterGenerator::unlock_method(bool check_exceptions) {
356 __ unlock_object(R26_monitor, check_exceptions);
357 }
358
359 // Lock the current method, interpreter register window must be set up!
360 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
361 const Register Robj_to_lock = Rscratch2;
362
363 {
364 if (!flags_preloaded) {
365 __ lwz(Rflags, method_(access_flags));
366 }
367
368 #ifdef ASSERT
369 // Check if methods needs synchronization.
370 {
371 Label Lok;
372 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
373 __ btrue(CCR0,Lok);
374 __ stop("method doesn't need synchronization");
375 __ bind(Lok);
376 }
377 #endif // ASSERT
378 }
379
380 // Get synchronization object to Rscratch2.
381 {
382 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
383 Label Lstatic;
384 Label Ldone;
385
386 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
387 __ btrue(CCR0, Lstatic);
388
389 // Non-static case: load receiver obj from stack and we're done.
390 __ ld(Robj_to_lock, R18_locals);
391 __ b(Ldone);
392
393 __ bind(Lstatic); // Static case: Lock the java mirror
394 __ ld(Robj_to_lock, in_bytes(Method::const_offset()), R19_method);
395 __ ld(Robj_to_lock, in_bytes(ConstMethod::constants_offset()), Robj_to_lock);
396 __ ld(Robj_to_lock, ConstantPool::pool_holder_offset_in_bytes(), Robj_to_lock);
397 __ ld(Robj_to_lock, mirror_offset, Robj_to_lock);
398
399 __ bind(Ldone);
400 __ verify_oop(Robj_to_lock);
401 }
402
403 // Got the oop to lock => execute!
404 __ add_monitor_to_stack(true, Rscratch1, R0);
405
406 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
407 __ lock_object(R26_monitor, Robj_to_lock);
408 }
409
410 // Generate a fixed interpreter frame for pure interpreter
411 // and I2N native transition frames.
412 //
413 // Before (stack grows downwards):
414 //
415 // | ... |
416 // |------------- |
417 // | java arg0 |
418 // | ... |
419 // | java argn |
420 // | | <- R15_esp
421 // | |
422 // |--------------|
423 // | abi_112 |
424 // | | <- R1_SP
425 // |==============|
426 //
427 //
428 // After:
429 //
430 // | ... |
431 // | java arg0 |<- R18_locals
432 // | ... |
433 // | java argn |
434 // |--------------|
435 // | |
436 // | java locals |
437 // | |
438 // |--------------|
439 // | abi_48 |
440 // |==============|
441 // | |
442 // | istate |
443 // | |
444 // |--------------|
445 // | monitor |<- R26_monitor
446 // |--------------|
447 // | |<- R15_esp
448 // | expression |
449 // | stack |
450 // | |
451 // |--------------|
452 // | |
453 // | abi_112 |<- R1_SP
454 // |==============|
455 //
456 // The top most frame needs an abi space of 112 bytes. This space is needed,
457 // since we call to c. The c function may spill their arguments to the caller
458 // frame. When we call to java, we don't need these spill slots. In order to save
459 // space on the stack, we resize the caller. However, java local reside in
460 // the caller frame and the frame has to be increased. The frame_size for the
461 // current frame was calculated based on max_stack as size for the expression
462 // stack. At the call, just a part of the expression stack might be used.
463 // We don't want to waste this space and cut the frame back accordingly.
464 // The resulting amount for resizing is calculated as follows:
465 // resize = (number_of_locals - number_of_arguments) * slot_size
466 // + (R1_SP - R15_esp) + 48
467 //
468 // The size for the callee frame is calculated:
469 // framesize = 112 + max_stack + monitor + state_size
470 //
471 // maxstack: Max number of slots on the expression stack, loaded from the method.
472 // monitor: We statically reserve room for one monitor object.
473 // state_size: We save the current state of the interpreter to this area.
474 //
475 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
476 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
477 top_frame_size = R7_ARG5,
478 Rconst_method = R8_ARG6;
479
480 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size);
481
482 __ ld(Rconst_method, method_(const));
483 __ lhz(Rsize_of_parameters /* number of params */,
484 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
485 if (native_call) {
486 // If we're calling a native method, we reserve space for the worst-case signature
487 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2).
488 // We add two slots to the parameter_count, one for the jni
489 // environment and one for a possible native mirror.
490 Label skip_native_calculate_max_stack;
491 __ addi(top_frame_size, Rsize_of_parameters, 2);
492 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters);
493 __ bge(CCR0, skip_native_calculate_max_stack);
494 __ li(top_frame_size, Argument::n_register_parameters);
495 __ bind(skip_native_calculate_max_stack);
496 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
497 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
498 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
499 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
500 } else {
501 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
502 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
503 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
504 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
505 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
506 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
507 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
508 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1);
509 }
510
511 // Compute top frame size.
512 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size);
513
514 // Cut back area between esp and max_stack.
515 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize);
516
517 __ round_to(top_frame_size, frame::alignment_in_bytes);
518 __ round_to(parent_frame_resize, frame::alignment_in_bytes);
519 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
520 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
521
522 {
523 // --------------------------------------------------------------------------
524 // Stack overflow check
525
526 Label cont;
527 __ add(R11_scratch1, parent_frame_resize, top_frame_size);
528 generate_stack_overflow_check(R11_scratch1, R12_scratch2);
529 }
530
531 // Set up interpreter state registers.
532
533 __ add(R18_locals, R15_esp, Rsize_of_parameters);
534 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
535 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache);
536
537 // Set method data pointer.
538 if (ProfileInterpreter) {
539 Label zero_continue;
540 __ ld(R28_mdx, method_(method_data));
541 __ cmpdi(CCR0, R28_mdx, 0);
542 __ beq(CCR0, zero_continue);
543 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
544 __ bind(zero_continue);
545 }
546
547 if (native_call) {
548 __ li(R14_bcp, 0); // Must initialize.
549 } else {
550 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method);
551 }
552
553 // Resize parent frame.
554 __ mflr(R12_scratch2);
555 __ neg(parent_frame_resize, parent_frame_resize);
556 __ resize_frame(parent_frame_resize, R11_scratch1);
557 __ std(R12_scratch2, _abi(lr), R1_SP);
558
559 __ addi(R26_monitor, R1_SP, - frame::ijava_state_size);
560 __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
561
562 // Store values.
563 // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls
564 // in InterpreterMacroAssembler::call_from_interpreter.
565 __ std(R19_method, _ijava_state_neg(method), R1_SP);
566 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
567 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
568 __ std(R18_locals, _ijava_state_neg(locals), R1_SP);
569
570 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
571 // be found in the frame after save_interpreter_state is done. This is always true
572 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
573 // because e.g. frame::interpreter_frame_bcp() will not access the correct value
574 // (Enhanced Stack Trace).
575 // The signal handler does not save the interpreter state into the frame.
576 __ li(R0, 0);
577 #ifdef ASSERT
578 // Fill remaining slots with constants.
579 __ load_const_optimized(R11_scratch1, 0x5afe);
580 __ load_const_optimized(R12_scratch2, 0xdead);
581 #endif
582 // We have to initialize some frame slots for native calls (accessed by GC).
583 if (native_call) {
584 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP);
585 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
586 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
587 }
588 #ifdef ASSERT
589 else {
590 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
591 __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP);
592 __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP);
593 }
594 __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP);
595 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
596 __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP);
597 __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP);
598 #endif
599 __ subf(R12_scratch2, top_frame_size, R1_SP);
600 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP);
601 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
602
603 // Push top frame.
604 __ push_frame(top_frame_size, R11_scratch1);
605 }
606
607 // End of helpers
608
609
610 // Support abs and sqrt like in compiler.
611 // For others we can use a normal (native) entry.
612
613 inline bool math_entry_available(AbstractInterpreter::MethodKind kind) {
614 // Provide math entry with debugging on demand.
615 // Note: Debugging changes which code will get executed:
616 // Debugging or disabled InlineIntrinsics: java method will get interpreted and performs a native call.
617 // Not debugging and enabled InlineIntrinics: processor instruction will get used.
618 // Result might differ slightly due to rounding etc.
619 if (!InlineIntrinsics && (!FLAG_IS_ERGO(InlineIntrinsics))) return false; // Generate a vanilla entry.
620
621 return ((kind==Interpreter::java_lang_math_sqrt && VM_Version::has_fsqrt()) ||
622 (kind==Interpreter::java_lang_math_abs));
623 }
624
625 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
626 if (!math_entry_available(kind)) {
627 NOT_PRODUCT(__ should_not_reach_here();)
628 return Interpreter::entry_for_kind(Interpreter::zerolocals);
629 }
630
631 Label Lslow_path;
632 const Register Rjvmti_mode = R11_scratch1;
633 address entry = __ pc();
634
635 // Provide math entry with debugging on demand.
636 __ lwz(Rjvmti_mode, thread_(interp_only_mode));
637 __ cmpwi(CCR0, Rjvmti_mode, 0);
638 __ bne(CCR0, Lslow_path); // jvmti_mode!=0
639
640 __ lfd(F1_RET, Interpreter::stackElementSize, R15_esp);
641
642 // Pop c2i arguments (if any) off when we return.
643 #ifdef ASSERT
644 __ ld(R9_ARG7, 0, R1_SP);
645 __ ld(R10_ARG8, 0, R21_sender_SP);
646 __ cmpd(CCR0, R9_ARG7, R10_ARG8);
647 __ asm_assert_eq("backlink", 0x545);
648 #endif // ASSERT
649 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
650
651 if (kind == Interpreter::java_lang_math_sqrt) {
652 __ fsqrt(F1_RET, F1_RET);
653 } else if (kind == Interpreter::java_lang_math_abs) {
654 __ fabs(F1_RET, F1_RET);
655 } else {
656 ShouldNotReachHere();
657 }
658
659 // And we're done.
660 __ blr();
661
662 // Provide slow path for JVMTI case.
663 __ bind(Lslow_path);
664 __ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R12_scratch2);
665 __ flush();
666
667 return entry;
668 }
669
670 // Interpreter stub for calling a native method. (asm interpreter)
671 // This sets up a somewhat different looking stack for calling the
672 // native method than the typical interpreter frame setup.
673 //
674 // On entry:
675 // R19_method - method
676 // R16_thread - JavaThread*
677 // R15_esp - intptr_t* sender tos
678 //
679 // abstract stack (grows up)
680 // [ IJava (caller of JNI callee) ] <-- ASP
681 // ...
682 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
683
684 address entry = __ pc();
685
686 const bool inc_counter = UseCompiler || CountCompiledCalls;
687
688 // -----------------------------------------------------------------------------
689 // Allocate a new frame that represents the native callee (i2n frame).
690 // This is not a full-blown interpreter frame, but in particular, the
691 // following registers are valid after this:
692 // - R19_method
693 // - R18_local (points to start of argumuments to native function)
694 //
695 // abstract stack (grows up)
696 // [ IJava (caller of JNI callee) ] <-- ASP
697 // ...
698
699 const Register signature_handler_fd = R11_scratch1;
700 const Register pending_exception = R0;
701 const Register result_handler_addr = R31;
702 const Register native_method_fd = R11_scratch1;
703 const Register access_flags = R22_tmp2;
704 const Register active_handles = R11_scratch1; // R26_monitor saved to state.
705 const Register sync_state = R12_scratch2;
706 const Register sync_state_addr = sync_state; // Address is dead after use.
707 const Register suspend_flags = R11_scratch1;
708
709 //=============================================================================
710 // Allocate new frame and initialize interpreter state.
711
712 Label exception_return;
713 Label exception_return_sync_check;
714 Label stack_overflow_return;
715
716 // Generate new interpreter state and jump to stack_overflow_return in case of
717 // a stack overflow.
718 //generate_compute_interpreter_state(stack_overflow_return);
719
720 Register size_of_parameters = R22_tmp2;
721
722 generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
723
724 //=============================================================================
725 // Increment invocation counter. On overflow, entry to JNI method
726 // will be compiled.
727 Label invocation_counter_overflow, continue_after_compile;
728 if (inc_counter) {
729 if (synchronized) {
730 // Since at this point in the method invocation the exception handler
731 // would try to exit the monitor of synchronized methods which hasn't
732 // been entered yet, we set the thread local variable
733 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
734 // runtime, exception handling i.e. unlock_if_synchronized_method will
735 // check this thread local flag.
736 // This flag has two effects, one is to force an unwind in the topmost
737 // interpreter frame and not perform an unlock while doing so.
738 __ li(R0, 1);
739 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
740 }
741 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
742
743 __ BIND(continue_after_compile);
744 // Reset the _do_not_unlock_if_synchronized flag.
745 if (synchronized) {
746 __ li(R0, 0);
747 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
748 }
749 }
750
751 // access_flags = method->access_flags();
752 // Load access flags.
753 assert(access_flags->is_nonvolatile(),
754 "access_flags must be in a non-volatile register");
755 // Type check.
756 assert(4 == sizeof(AccessFlags), "unexpected field size");
757 __ lwz(access_flags, method_(access_flags));
758
759 // We don't want to reload R19_method and access_flags after calls
760 // to some helper functions.
761 assert(R19_method->is_nonvolatile(),
762 "R19_method must be a non-volatile register");
763
764 // Check for synchronized methods. Must happen AFTER invocation counter
765 // check, so method is not locked if counter overflows.
766
767 if (synchronized) {
768 lock_method(access_flags, R11_scratch1, R12_scratch2, true);
769
770 // Update monitor in state.
771 __ ld(R11_scratch1, 0, R1_SP);
772 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1);
773 }
774
775 // jvmti/jvmpi support
776 __ notify_method_entry();
777
778 //=============================================================================
779 // Get and call the signature handler.
780
781 __ ld(signature_handler_fd, method_(signature_handler));
782 Label call_signature_handler;
783
784 __ cmpdi(CCR0, signature_handler_fd, 0);
785 __ bne(CCR0, call_signature_handler);
786
787 // Method has never been called. Either generate a specialized
788 // handler or point to the slow one.
789 //
790 // Pass parameter 'false' to avoid exception check in call_VM.
791 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
792
793 // Check for an exception while looking up the target method. If we
794 // incurred one, bail.
795 __ ld(pending_exception, thread_(pending_exception));
796 __ cmpdi(CCR0, pending_exception, 0);
797 __ bne(CCR0, exception_return_sync_check); // Has pending exception.
798
799 // Reload signature handler, it may have been created/assigned in the meanwhile.
800 __ ld(signature_handler_fd, method_(signature_handler));
801 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
802
803 __ BIND(call_signature_handler);
804
805 // Before we call the signature handler we push a new frame to
806 // protect the interpreter frame volatile registers when we return
807 // from jni but before we can get back to Java.
808
809 // First set the frame anchor while the SP/FP registers are
810 // convenient and the slow signature handler can use this same frame
811 // anchor.
812
813 // We have a TOP_IJAVA_FRAME here, which belongs to us.
814 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
815
816 // Now the interpreter frame (and its call chain) have been
817 // invalidated and flushed. We are now protected against eager
818 // being enabled in native code. Even if it goes eager the
819 // registers will be reloaded as clean and we will invalidate after
820 // the call so no spurious flush should be possible.
821
822 // Call signature handler and pass locals address.
823 //
824 // Our signature handlers copy required arguments to the C stack
825 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
826 __ mr(R3_ARG1, R18_locals);
827 #if !defined(ABI_ELFv2)
828 __ ld(signature_handler_fd, 0, signature_handler_fd);
829 #endif
830
831 __ call_stub(signature_handler_fd);
832
833 // Remove the register parameter varargs slots we allocated in
834 // compute_interpreter_state. SP+16 ends up pointing to the ABI
835 // outgoing argument area.
836 //
837 // Not needed on PPC64.
838 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord);
839
840 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
841 // Save across call to native method.
842 __ mr(result_handler_addr, R3_RET);
843
844 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
845
846 // Set up fixed parameters and call the native method.
847 // If the method is static, get mirror into R4_ARG2.
848 {
849 Label method_is_not_static;
850 // Access_flags is non-volatile and still, no need to restore it.
851
852 // Restore access flags.
853 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
854 __ bfalse(CCR0, method_is_not_static);
855
856 // constants = method->constants();
857 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
858 __ ld(R11_scratch1, in_bytes(ConstMethod::constants_offset()), R11_scratch1);
859 // pool_holder = method->constants()->pool_holder();
860 __ ld(R11_scratch1/*pool_holder*/, ConstantPool::pool_holder_offset_in_bytes(),
861 R11_scratch1/*constants*/);
862
863 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
864
865 // mirror = pool_holder->klass_part()->java_mirror();
866 __ ld(R0/*mirror*/, mirror_offset, R11_scratch1/*pool_holder*/);
867 // state->_native_mirror = mirror;
868
869 __ ld(R11_scratch1, 0, R1_SP);
870 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
871 // R4_ARG2 = &state->_oop_temp;
872 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
873 __ BIND(method_is_not_static);
874 }
875
876 // At this point, arguments have been copied off the stack into
877 // their JNI positions. Oops are boxed in-place on the stack, with
878 // handles copied to arguments. The result handler address is in a
879 // register.
880
881 // Pass JNIEnv address as first parameter.
882 __ addir(R3_ARG1, thread_(jni_environment));
883
884 // Load the native_method entry before we change the thread state.
885 __ ld(native_method_fd, method_(native_function));
886
887 //=============================================================================
888 // Transition from _thread_in_Java to _thread_in_native. As soon as
889 // we make this change the safepoint code needs to be certain that
890 // the last Java frame we established is good. The pc in that frame
891 // just needs to be near here not an actual return address.
892
893 // We use release_store_fence to update values like the thread state, where
894 // we don't want the current thread to continue until all our prior memory
895 // accesses (including the new thread state) are visible to other threads.
896 __ li(R0, _thread_in_native);
897 __ release();
898
899 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
900 __ stw(R0, thread_(thread_state));
901
902 if (UseMembar) {
903 __ fence();
904 }
905
906 //=============================================================================
907 // Call the native method. Argument registers must not have been
908 // overwritten since "__ call_stub(signature_handler);" (except for
909 // ARG1 and ARG2 for static methods).
910 __ call_c(native_method_fd);
911
912 __ li(R0, 0);
913 __ ld(R11_scratch1, 0, R1_SP);
914 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
915 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
916 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
917
918 // Note: C++ interpreter needs the following here:
919 // The frame_manager_lr field, which we use for setting the last
920 // java frame, gets overwritten by the signature handler. Restore
921 // it now.
922 //__ get_PC_trash_LR(R11_scratch1);
923 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
924
925 // Because of GC R19_method may no longer be valid.
926
927 // Block, if necessary, before resuming in _thread_in_Java state.
928 // In order for GC to work, don't clear the last_Java_sp until after
929 // blocking.
930
931 //=============================================================================
932 // Switch thread to "native transition" state before reading the
933 // synchronization state. This additional state is necessary
934 // because reading and testing the synchronization state is not
935 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
936 // in _thread_in_native state, loads _not_synchronized and is
937 // preempted. VM thread changes sync state to synchronizing and
938 // suspends threads for GC. Thread A is resumed to finish this
939 // native method, but doesn't block here since it didn't see any
940 // synchronization in progress, and escapes.
941
942 // We use release_store_fence to update values like the thread state, where
943 // we don't want the current thread to continue until all our prior memory
944 // accesses (including the new thread state) are visible to other threads.
945 __ li(R0/*thread_state*/, _thread_in_native_trans);
946 __ release();
947 __ stw(R0/*thread_state*/, thread_(thread_state));
948 if (UseMembar) {
949 __ fence();
950 }
951 // Write serialization page so that the VM thread can do a pseudo remote
952 // membar. We use the current thread pointer to calculate a thread
953 // specific offset to write to within the page. This minimizes bus
954 // traffic due to cache line collision.
955 else {
956 __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2);
957 }
958
959 // Now before we return to java we must look for a current safepoint
960 // (a new safepoint can not start since we entered native_trans).
961 // We must check here because a current safepoint could be modifying
962 // the callers registers right this moment.
963
964 // Acquire isn't strictly necessary here because of the fence, but
965 // sync_state is declared to be volatile, so we do it anyway
966 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
967 int sync_state_offs = __ load_const_optimized(sync_state_addr, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
968
969 // TODO PPC port assert(4 == SafepointSynchronize::sz_state(), "unexpected field size");
970 __ lwz(sync_state, sync_state_offs, sync_state_addr);
971
972 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
973 __ lwz(suspend_flags, thread_(suspend_flags));
974
975 Label sync_check_done;
976 Label do_safepoint;
977 // No synchronization in progress nor yet synchronized.
978 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
979 // Not suspended.
980 __ cmpwi(CCR1, suspend_flags, 0);
981
982 __ bne(CCR0, do_safepoint);
983 __ beq(CCR1, sync_check_done);
984 __ bind(do_safepoint);
985 __ isync();
986 // Block. We do the call directly and leave the current
987 // last_Java_frame setup undisturbed. We must save any possible
988 // native result across the call. No oop is present.
989
990 __ mr(R3_ARG1, R16_thread);
991 #if defined(ABI_ELFv2)
992 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
993 relocInfo::none);
994 #else
995 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
996 relocInfo::none);
997 #endif
998
999 __ bind(sync_check_done);
1000
1001 //=============================================================================
1002 // <<<<<< Back in Interpreter Frame >>>>>
1003
1004 // We are in thread_in_native_trans here and back in the normal
1005 // interpreter frame. We don't have to do anything special about
1006 // safepoints and we can switch to Java mode anytime we are ready.
1007
1008 // Note: frame::interpreter_frame_result has a dependency on how the
1009 // method result is saved across the call to post_method_exit. For
1010 // native methods it assumes that the non-FPU/non-void result is
1011 // saved in _native_lresult and a FPU result in _native_fresult. If
1012 // this changes then the interpreter_frame_result implementation
1013 // will need to be updated too.
1014
1015 // On PPC64, we have stored the result directly after the native call.
1016
1017 //=============================================================================
1018 // Back in Java
1019
1020 // We use release_store_fence to update values like the thread state, where
1021 // we don't want the current thread to continue until all our prior memory
1022 // accesses (including the new thread state) are visible to other threads.
1023 __ li(R0/*thread_state*/, _thread_in_Java);
1024 __ release();
1025 __ stw(R0/*thread_state*/, thread_(thread_state));
1026 if (UseMembar) {
1027 __ fence();
1028 }
1029
1030 __ reset_last_Java_frame();
1031
1032 // Jvmdi/jvmpi support. Whether we've got an exception pending or
1033 // not, and whether unlocking throws an exception or not, we notify
1034 // on native method exit. If we do have an exception, we'll end up
1035 // in the caller's context to handle it, so if we don't do the
1036 // notify here, we'll drop it on the floor.
1037 __ notify_method_exit(true/*native method*/,
1038 ilgl /*illegal state (not used for native methods)*/,
1039 InterpreterMacroAssembler::NotifyJVMTI,
1040 false /*check_exceptions*/);
1041
1042 //=============================================================================
1043 // Handle exceptions
1044
1045 if (synchronized) {
1046 // Don't check for exceptions since we're still in the i2n frame. Do that
1047 // manually afterwards.
1048 unlock_method(false);
1049 }
1050
1051 // Reset active handles after returning from native.
1052 // thread->active_handles()->clear();
1053 __ ld(active_handles, thread_(active_handles));
1054 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1055 __ li(R0, 0);
1056 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles);
1057
1058 Label exception_return_sync_check_already_unlocked;
1059 __ ld(R0/*pending_exception*/, thread_(pending_exception));
1060 __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1061 __ bne(CCR0, exception_return_sync_check_already_unlocked);
1062
1063 //-----------------------------------------------------------------------------
1064 // No exception pending.
1065
1066 // Move native method result back into proper registers and return.
1067 // Invoke result handler (may unbox/promote).
1068 __ ld(R11_scratch1, 0, R1_SP);
1069 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1070 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1071 __ call_stub(result_handler_addr);
1072
1073 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1074
1075 // Must use the return pc which was loaded from the caller's frame
1076 // as the VM uses return-pc-patching for deoptimization.
1077 __ mtlr(R0);
1078 __ blr();
1079
1080 //-----------------------------------------------------------------------------
1081 // An exception is pending. We call into the runtime only if the
1082 // caller was not interpreted. If it was interpreted the
1083 // interpreter will do the correct thing. If it isn't interpreted
1084 // (call stub/compiled code) we will change our return and continue.
1085
1086 __ BIND(exception_return_sync_check);
1087
1088 if (synchronized) {
1089 // Don't check for exceptions since we're still in the i2n frame. Do that
1090 // manually afterwards.
1091 unlock_method(false);
1092 }
1093 __ BIND(exception_return_sync_check_already_unlocked);
1094
1095 const Register return_pc = R31;
1096
1097 __ ld(return_pc, 0, R1_SP);
1098 __ ld(return_pc, _abi(lr), return_pc);
1099
1100 // Get the address of the exception handler.
1101 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1102 R16_thread,
1103 return_pc /* return pc */);
1104 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1105
1106 // Load the PC of the the exception handler into LR.
1107 __ mtlr(R3_RET);
1108
1109 // Load exception into R3_ARG1 and clear pending exception in thread.
1110 __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1111 __ li(R4_ARG2, 0);
1112 __ std(R4_ARG2, thread_(pending_exception));
1113
1114 // Load the original return pc into R4_ARG2.
1115 __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1116
1117 // Return to exception handler.
1118 __ blr();
1119
1120 //=============================================================================
1121 // Counter overflow.
1122
1123 if (inc_counter) {
1124 // Handle invocation counter overflow.
1125 __ bind(invocation_counter_overflow);
1126
1127 generate_counter_overflow(continue_after_compile);
1128 }
1129
1130 return entry;
1131 }
1132
1133 // Generic interpreted method entry to (asm) interpreter.
1134 //
1135 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1136 bool inc_counter = UseCompiler || CountCompiledCalls;
1137 address entry = __ pc();
1138 // Generate the code to allocate the interpreter stack frame.
1139 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1140 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame.
1141
1142 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1143
1144 #ifdef FAST_DISPATCH
1145 __ unimplemented("Fast dispatch in generate_normal_entry");
1146 #if 0
1147 __ set((intptr_t)Interpreter::dispatch_table(), IdispatchTables);
1148 // Set bytecode dispatch table base.
1149 #endif
1150 #endif
1151
1152 // --------------------------------------------------------------------------
1153 // Zero out non-parameter locals.
1154 // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1155 // worth to ask the flag, just do it.
1156 Register Rslot_addr = R6_ARG4,
1157 Rnum = R7_ARG5;
1158 Label Lno_locals, Lzero_loop;
1159
1160 // Set up the zeroing loop.
1161 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1162 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1163 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1164 __ beq(CCR0, Lno_locals);
1165 __ li(R0, 0);
1166 __ mtctr(Rnum);
1167
1168 // The zero locals loop.
1169 __ bind(Lzero_loop);
1170 __ std(R0, 0, Rslot_addr);
1171 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1172 __ bdnz(Lzero_loop);
1173
1174 __ bind(Lno_locals);
1175
1176 // --------------------------------------------------------------------------
1177 // Counter increment and overflow check.
1178 Label invocation_counter_overflow,
1179 profile_method,
1180 profile_method_continue;
1181 if (inc_counter || ProfileInterpreter) {
1182
1183 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1184 if (synchronized) {
1185 // Since at this point in the method invocation the exception handler
1186 // would try to exit the monitor of synchronized methods which hasn't
1187 // been entered yet, we set the thread local variable
1188 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1189 // runtime, exception handling i.e. unlock_if_synchronized_method will
1190 // check this thread local flag.
1191 // This flag has two effects, one is to force an unwind in the topmost
1192 // interpreter frame and not perform an unlock while doing so.
1193 __ li(R0, 1);
1194 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1195 }
1196
1197 // Argument and return type profiling.
1198 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1199
1200 // Increment invocation counter and check for overflow.
1201 if (inc_counter) {
1202 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1203 }
1204
1205 __ bind(profile_method_continue);
1206
1207 // Reset the _do_not_unlock_if_synchronized flag.
1208 if (synchronized) {
1209 __ li(R0, 0);
1210 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1211 }
1212 }
1213
1214 // --------------------------------------------------------------------------
1215 // Locking of synchronized methods. Must happen AFTER invocation_counter
1216 // check and stack overflow check, so method is not locked if overflows.
1217 if (synchronized) {
1218 lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1219 }
1220 #ifdef ASSERT
1221 else {
1222 Label Lok;
1223 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1224 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1225 __ asm_assert_eq("method needs synchronization", 0x8521);
1226 __ bind(Lok);
1227 }
1228 #endif // ASSERT
1229
1230 __ verify_thread();
1231
1232 // --------------------------------------------------------------------------
1233 // JVMTI support
1234 __ notify_method_entry();
1235
1236 // --------------------------------------------------------------------------
1237 // Start executing instructions.
1238 __ dispatch_next(vtos);
1239
1240 // --------------------------------------------------------------------------
1241 // Out of line counter overflow and MDO creation code.
1242 if (ProfileInterpreter) {
1243 // We have decided to profile this method in the interpreter.
1244 __ bind(profile_method);
1245 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1246 __ set_method_data_pointer_for_bcp();
1247 __ b(profile_method_continue);
1248 }
1249
1250 if (inc_counter) {
1251 // Handle invocation counter overflow.
1252 __ bind(invocation_counter_overflow);
1253 generate_counter_overflow(profile_method_continue);
1254 }
1255 return entry;
1256 }
1257
1258 // These should never be compiled since the interpreter will prefer
1259 // the compiled version to the intrinsic version.
1260 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1261 return !math_entry_available(method_kind(m));
1262 }
1263
1264 // How much stack a method activation needs in stack slots.
1265 // We must calc this exactly like in generate_fixed_frame.
1266 // Note: This returns the conservative size assuming maximum alignment.
1267 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1268 const int max_alignment_size = 2;
1269 const int abi_scratch = frame::abi_reg_args_size;
1270 return method->max_locals() + method->max_stack() +
1271 frame::interpreter_frame_monitor_size() + max_alignment_size + abi_scratch;
1272 }
1273
1274 // Returns number of stackElementWords needed for the interpreter frame with the
1275 // given sections.
1276 // This overestimates the stack by one slot in case of alignments.
1277 int AbstractInterpreter::size_activation(int max_stack,
1278 int temps,
1279 int extra_args,
1280 int monitors,
1281 int callee_params,
1282 int callee_locals,
1283 bool is_top_frame) {
1284 // Note: This calculation must exactly parallel the frame setup
1285 // in InterpreterGenerator::generate_fixed_frame.
1286 assert(Interpreter::stackElementWords == 1, "sanity");
1287 const int max_alignment_space = StackAlignmentInBytes / Interpreter::stackElementSize;
1288 const int abi_scratch = is_top_frame ? (frame::abi_reg_args_size / Interpreter::stackElementSize) :
1289 (frame::abi_minframe_size / Interpreter::stackElementSize);
1290 const int size =
1291 max_stack +
1292 (callee_locals - callee_params) +
1293 monitors * frame::interpreter_frame_monitor_size() +
1294 max_alignment_space +
1295 abi_scratch +
1296 frame::ijava_state_size / Interpreter::stackElementSize;
1297
1298 // Fixed size of an interpreter frame, align to 16-byte.
1299 return (size & -2);
1300 }
1301
1302 // Fills a sceletal interpreter frame generated during deoptimizations.
1303 //
1304 // Parameters:
1305 //
1306 // interpreter_frame != NULL:
1307 // set up the method, locals, and monitors.
1308 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1309 // right size, as determined by a previous call to this method.
1310 // It is also guaranteed to be walkable even though it is in a skeletal state
1311 //
1312 // is_top_frame == true:
1313 // We're processing the *oldest* interpreter frame!
1314 //
1315 // pop_frame_extra_args:
1316 // If this is != 0 we are returning to a deoptimized frame by popping
1317 // off the callee frame. We want to re-execute the call that called the
1318 // callee interpreted, but since the return to the interpreter would pop
1319 // the arguments off advance the esp by dummy popframe_extra_args slots.
1320 // Popping off those will establish the stack layout as it was before the call.
1321 //
1322 void AbstractInterpreter::layout_activation(Method* method,
1323 int tempcount,
1324 int popframe_extra_args,
1325 int moncount,
1326 int caller_actual_parameters,
1327 int callee_param_count,
1328 int callee_locals_count,
1329 frame* caller,
1330 frame* interpreter_frame,
1331 bool is_top_frame,
1332 bool is_bottom_frame) {
1333
1334 const int abi_scratch = is_top_frame ? (frame::abi_reg_args_size / Interpreter::stackElementSize) :
1335 (frame::abi_minframe_size / Interpreter::stackElementSize);
1336
1337 intptr_t* locals_base = (caller->is_interpreted_frame()) ?
1338 caller->interpreter_frame_esp() + caller_actual_parameters :
1339 caller->sp() + method->max_locals() - 1 + (frame::abi_minframe_size / Interpreter::stackElementSize) ;
1340
1341 intptr_t* monitor_base = caller->sp() - frame::ijava_state_size / Interpreter::stackElementSize ;
1342 intptr_t* monitor = monitor_base - (moncount * frame::interpreter_frame_monitor_size());
1343 intptr_t* esp_base = monitor - 1;
1344 intptr_t* esp = esp_base - tempcount - popframe_extra_args;
1345 intptr_t* sp = (intptr_t *) (((intptr_t) (esp_base - callee_locals_count + callee_param_count - method->max_stack()- abi_scratch)) & -StackAlignmentInBytes);
1346 intptr_t* sender_sp = caller->sp() + (frame::abi_minframe_size - frame::abi_reg_args_size) / Interpreter::stackElementSize;
1347 intptr_t* top_frame_sp = is_top_frame ? sp : sp + (frame::abi_minframe_size - frame::abi_reg_args_size) / Interpreter::stackElementSize;
1348
1349 interpreter_frame->interpreter_frame_set_method(method);
1350 interpreter_frame->interpreter_frame_set_locals(locals_base);
1351 interpreter_frame->interpreter_frame_set_cpcache(method->constants()->cache());
1352 interpreter_frame->interpreter_frame_set_esp(esp);
1353 interpreter_frame->interpreter_frame_set_monitor_end((BasicObjectLock *)monitor);
1354 interpreter_frame->interpreter_frame_set_top_frame_sp(top_frame_sp);
1355 if (!is_bottom_frame) {
1356 interpreter_frame->interpreter_frame_set_sender_sp(sender_sp);
1357 }
1358 }
1359
1360 // =============================================================================
1361 // Exceptions
1362
1363 void TemplateInterpreterGenerator::generate_throw_exception() {
1364 Register Rexception = R17_tos,
1365 Rcontinuation = R3_RET;
1366
1367 // --------------------------------------------------------------------------
1368 // Entry point if an method returns with a pending exception (rethrow).
1369 Interpreter::_rethrow_exception_entry = __ pc();
1370 {
1371 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
1372 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
1373 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
1374
1375 // Compiled code destroys templateTableBase, reload.
1376 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
1377 }
1378
1379 // Entry point if a interpreted method throws an exception (throw).
1380 Interpreter::_throw_exception_entry = __ pc();
1381 {
1382 __ mr(Rexception, R3_RET);
1383
1384 __ verify_thread();
1385 __ verify_oop(Rexception);
1386
1387 // Expression stack must be empty before entering the VM in case of an exception.
1388 __ empty_expression_stack();
1389 // Find exception handler address and preserve exception oop.
1390 // Call C routine to find handler and jump to it.
1391 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
1392 __ mtctr(Rcontinuation);
1393 // Push exception for exception handler bytecodes.
1394 __ push_ptr(Rexception);
1395
1396 // Jump to exception handler (may be remove activation entry!).
1397 __ bctr();
1398 }
1399
1400 // If the exception is not handled in the current frame the frame is
1401 // removed and the exception is rethrown (i.e. exception
1402 // continuation is _rethrow_exception).
1403 //
1404 // Note: At this point the bci is still the bxi for the instruction
1405 // which caused the exception and the expression stack is
1406 // empty. Thus, for any VM calls at this point, GC will find a legal
1407 // oop map (with empty expression stack).
1408
1409 // In current activation
1410 // tos: exception
1411 // bcp: exception bcp
1412
1413 // --------------------------------------------------------------------------
1414 // JVMTI PopFrame support
1415
1416 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1417 {
1418 // Set the popframe_processing bit in popframe_condition indicating that we are
1419 // currently handling popframe, so that call_VMs that may happen later do not
1420 // trigger new popframe handling cycles.
1421 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1422 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
1423 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1424
1425 // Empty the expression stack, as in normal exception handling.
1426 __ empty_expression_stack();
1427 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
1428
1429 // Check to see whether we are returning to a deoptimized frame.
1430 // (The PopFrame call ensures that the caller of the popped frame is
1431 // either interpreted or compiled and deoptimizes it if compiled.)
1432 // Note that we don't compare the return PC against the
1433 // deoptimization blob's unpack entry because of the presence of
1434 // adapter frames in C2.
1435 Label Lcaller_not_deoptimized;
1436 Register return_pc = R3_ARG1;
1437 __ ld(return_pc, 0, R1_SP);
1438 __ ld(return_pc, _abi(lr), return_pc);
1439 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
1440 __ cmpdi(CCR0, R3_RET, 0);
1441 __ bne(CCR0, Lcaller_not_deoptimized);
1442
1443 // The deoptimized case.
1444 // In this case, we can't call dispatch_next() after the frame is
1445 // popped, but instead must save the incoming arguments and restore
1446 // them after deoptimization has occurred.
1447 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
1448 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
1449 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
1450 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
1451 __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
1452 // Save these arguments.
1453 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
1454
1455 // Inform deoptimization that it is responsible for restoring these arguments.
1456 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1457 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1458
1459 // Return from the current method into the deoptimization blob. Will eventually
1460 // end up in the deopt interpeter entry, deoptimization prepared everything that
1461 // we will reexecute the call that called us.
1462 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
1463 __ mtlr(return_pc);
1464 __ blr();
1465
1466 // The non-deoptimized case.
1467 __ bind(Lcaller_not_deoptimized);
1468
1469 // Clear the popframe condition flag.
1470 __ li(R0, 0);
1471 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1472
1473 // Get out of the current method and re-execute the call that called us.
1474 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
1475 __ restore_interpreter_state(R11_scratch1);
1476 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
1477 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
1478 if (ProfileInterpreter) {
1479 __ set_method_data_pointer_for_bcp();
1480 __ ld(R11_scratch1, 0, R1_SP);
1481 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
1482 }
1483 #if INCLUDE_JVMTI
1484 Label L_done;
1485
1486 __ lbz(R11_scratch1, 0, R14_bcp);
1487 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
1488 __ bne(CCR0, L_done);
1489
1490 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1491 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1492 __ ld(R4_ARG2, 0, R18_locals);
1493 __ MacroAssembler::call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp, false);
1494 __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
1495 __ cmpdi(CCR0, R4_ARG2, 0);
1496 __ beq(CCR0, L_done);
1497 __ std(R4_ARG2, wordSize, R15_esp);
1498 __ bind(L_done);
1499 #endif // INCLUDE_JVMTI
1500 __ dispatch_next(vtos);
1501 }
1502 // end of JVMTI PopFrame support
1503
1504 // --------------------------------------------------------------------------
1505 // Remove activation exception entry.
1506 // This is jumped to if an interpreted method can't handle an exception itself
1507 // (we come from the throw/rethrow exception entry above). We're going to call
1508 // into the VM to find the exception handler in the caller, pop the current
1509 // frame and return the handler we calculated.
1510 Interpreter::_remove_activation_entry = __ pc();
1511 {
1512 __ pop_ptr(Rexception);
1513 __ verify_thread();
1514 __ verify_oop(Rexception);
1515 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
1516
1517 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
1518 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
1519
1520 __ get_vm_result(Rexception);
1521
1522 // We are done with this activation frame; find out where to go next.
1523 // The continuation point will be an exception handler, which expects
1524 // the following registers set up:
1525 //
1526 // RET: exception oop
1527 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
1528
1529 Register return_pc = R31; // Needs to survive the runtime call.
1530 __ ld(return_pc, 0, R1_SP);
1531 __ ld(return_pc, _abi(lr), return_pc);
1532 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
1533
1534 // Remove the current activation.
1535 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
1536
1537 __ mr(R4_ARG2, return_pc);
1538 __ mtlr(R3_RET);
1539 __ mr(R3_RET, Rexception);
1540 __ blr();
1541 }
1542 }
1543
1544 // JVMTI ForceEarlyReturn support.
1545 // Returns "in the middle" of a method with a "fake" return value.
1546 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1547
1548 Register Rscratch1 = R11_scratch1,
1549 Rscratch2 = R12_scratch2;
1550
1551 address entry = __ pc();
1552 __ empty_expression_stack();
1553
1554 __ load_earlyret_value(state, Rscratch1);
1555
1556 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
1557 // Clear the earlyret state.
1558 __ li(R0, 0);
1559 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
1560
1561 __ remove_activation(state, false, false);
1562 // Copied from TemplateTable::_return.
1563 // Restoration of lr done by remove_activation.
1564 switch (state) {
1565 case ltos:
1566 case btos:
1567 case ctos:
1568 case stos:
1569 case atos:
1570 case itos: __ mr(R3_RET, R17_tos); break;
1571 case ftos:
1572 case dtos: __ fmr(F1_RET, F15_ftos); break;
1573 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
1574 // to get visible before the reference to the object gets stored anywhere.
1575 __ membar(Assembler::StoreStore); break;
1576 default : ShouldNotReachHere();
1577 }
1578 __ blr();
1579
1580 return entry;
1581 } // end of ForceEarlyReturn support
1582
1583 //-----------------------------------------------------------------------------
1584 // Helper for vtos entry point generation
1585
1586 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1587 address& bep,
1588 address& cep,
1589 address& sep,
1590 address& aep,
1591 address& iep,
1592 address& lep,
1593 address& fep,
1594 address& dep,
1595 address& vep) {
1596 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1597 Label L;
1598
1599 aep = __ pc(); __ push_ptr(); __ b(L);
1600 fep = __ pc(); __ push_f(); __ b(L);
1601 dep = __ pc(); __ push_d(); __ b(L);
1602 lep = __ pc(); __ push_l(); __ b(L);
1603 __ align(32, 12, 24); // align L
1604 bep = cep = sep =
1605 iep = __ pc(); __ push_i();
1606 vep = __ pc();
1607 __ bind(L);
1608 generate_and_dispatch(t);
1609 }
1610
1611 //-----------------------------------------------------------------------------
1612 // Generation of individual instructions
1613
1614 // helpers for generate_and_dispatch
1615
1616 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1617 : TemplateInterpreterGenerator(code) {
1618 generate_all(); // Down here so it can be "virtual".
1619 }
1620
1621 //-----------------------------------------------------------------------------
1622
1623 // Non-product code
1624 #ifndef PRODUCT
1625 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1626 //__ flush_bundle();
1627 address entry = __ pc();
1628
1629 const char *bname = NULL;
1630 uint tsize = 0;
1631 switch(state) {
1632 case ftos:
1633 bname = "trace_code_ftos {";
1634 tsize = 2;
1635 break;
1636 case btos:
1637 bname = "trace_code_btos {";
1638 tsize = 2;
1639 break;
1640 case ctos:
1641 bname = "trace_code_ctos {";
1642 tsize = 2;
1643 break;
1644 case stos:
1645 bname = "trace_code_stos {";
1646 tsize = 2;
1647 break;
1648 case itos:
1649 bname = "trace_code_itos {";
1650 tsize = 2;
1651 break;
1652 case ltos:
1653 bname = "trace_code_ltos {";
1654 tsize = 3;
1655 break;
1656 case atos:
1657 bname = "trace_code_atos {";
1658 tsize = 2;
1659 break;
1660 case vtos:
1661 // Note: In case of vtos, the topmost of stack value could be a int or doubl
1662 // In case of a double (2 slots) we won't see the 2nd stack value.
1663 // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
1664 bname = "trace_code_vtos {";
1665 tsize = 2;
1666
1667 break;
1668 case dtos:
1669 bname = "trace_code_dtos {";
1670 tsize = 3;
1671 break;
1672 default:
1673 ShouldNotReachHere();
1674 }
1675 BLOCK_COMMENT(bname);
1676
1677 // Support short-cut for TraceBytecodesAt.
1678 // Don't call into the VM if we don't want to trace to speed up things.
1679 Label Lskip_vm_call;
1680 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
1681 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
1682 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
1683 __ ld(R11_scratch1, offs1, R11_scratch1);
1684 __ lwa(R12_scratch2, offs2, R12_scratch2);
1685 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
1686 __ blt(CCR0, Lskip_vm_call);
1687 }
1688
1689 __ push(state);
1690 // Load 2 topmost expression stack values.
1691 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
1692 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
1693 __ mflr(R31);
1694 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
1695 __ mtlr(R31);
1696 __ pop(state);
1697
1698 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
1699 __ bind(Lskip_vm_call);
1700 }
1701 __ blr();
1702 BLOCK_COMMENT("} trace_code");
1703 return entry;
1704 }
1705
1706 void TemplateInterpreterGenerator::count_bytecode() {
1707 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
1708 __ lwz(R12_scratch2, offs, R11_scratch1);
1709 __ addi(R12_scratch2, R12_scratch2, 1);
1710 __ stw(R12_scratch2, offs, R11_scratch1);
1711 }
1712
1713 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1714 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
1715 __ lwz(R12_scratch2, offs, R11_scratch1);
1716 __ addi(R12_scratch2, R12_scratch2, 1);
1717 __ stw(R12_scratch2, offs, R11_scratch1);
1718 }
1719
1720 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1721 const Register addr = R11_scratch1,
1722 tmp = R12_scratch2;
1723 // Get index, shift out old bytecode, bring in new bytecode, and store it.
1724 // _index = (_index >> log2_number_of_codes) |
1725 // (bytecode << log2_number_of_codes);
1726 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
1727 __ lwz(tmp, offs1, addr);
1728 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
1729 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1730 __ stw(tmp, offs1, addr);
1731
1732 // Bump bucket contents.
1733 // _counters[_index] ++;
1734 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
1735 __ sldi(tmp, tmp, LogBytesPerInt);
1736 __ add(addr, tmp, addr);
1737 __ lwz(tmp, offs2, addr);
1738 __ addi(tmp, tmp, 1);
1739 __ stw(tmp, offs2, addr);
1740 }
1741
1742 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1743 // Call a little run-time stub to avoid blow-up for each bytecode.
1744 // The run-time runtime saves the right registers, depending on
1745 // the tosca in-state for the given template.
1746
1747 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1748 "entry must have been generated");
1749
1750 // Note: we destroy LR here.
1751 __ bl(Interpreter::trace_code(t->tos_in()));
1752 }
1753
1754 void TemplateInterpreterGenerator::stop_interpreter_at() {
1755 Label L;
1756 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
1757 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
1758 __ ld(R11_scratch1, offs1, R11_scratch1);
1759 __ lwa(R12_scratch2, offs2, R12_scratch2);
1760 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
1761 __ bne(CCR0, L);
1762 __ illtrap();
1763 __ bind(L);
1764 }
1765
1766 #endif // !PRODUCT
1767 #endif // !CC_INTERP