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 "ci/ciMethod.hpp"
30 #include "interpreter/bytecodeHistogram.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "interpreter/interpreterGenerator.hpp"
33 #include "interpreter/interpreterRuntime.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(const char* name) {
86 address entry = __ pc();
87
88 // Target class oop is in register R6_ARG4 by convention!
89
90 // Expression stack must be empty before entering the VM if an
91 // exception happened.
92 __ empty_expression_stack();
93 // Setup parameters.
94 // Thread will be loaded to R3_ARG1.
95 __ load_const_optimized(R4_ARG2, (address) name);
96 __ mr(R5_ARG3, R17_tos);
97 // R6_ARG4 contains specified class.
98 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose));
99 #ifdef ASSERT
100 // Above call must not return here since exception pending.
101 __ should_not_reach_here();
102 #endif
103 return entry;
104 }
105 #endif
106
107 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
108 address entry = __ pc();
109 // Expression stack must be empty before entering the VM if an
110 // exception happened.
111 __ empty_expression_stack();
112
113 // Load exception object.
114 // Thread will be loaded to R3_ARG1.
115 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
116 #ifdef ASSERT
117 // Above call must not return here since exception pending.
118 __ should_not_reach_here();
119 #endif
120 return entry;
121 }
122
123 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
124 address entry = __ pc();
125 //__ untested("generate_exception_handler_common");
126 Register Rexception = R17_tos;
127
128 // Expression stack must be empty before entering the VM if an exception happened.
129 __ empty_expression_stack();
130
131 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
132 if (pass_oop) {
133 __ mr(R5_ARG3, Rexception);
134 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false);
135 } else {
136 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
137 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false);
138 }
139
140 // Throw exception.
141 __ mr(R3_ARG1, Rexception);
142 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
143 __ mtctr(R11_scratch1);
144 __ bctr();
145
146 return entry;
147 }
148
149 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
150 address entry = __ pc();
151 __ unimplemented("generate_continuation_for");
152 return entry;
153 }
154
155 // This entry is returned to when a call returns to the interpreter.
156 // When we arrive here, we expect that the callee stack frame is already popped.
157 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
158 address entry = __ pc();
159
160 // Move the value out of the return register back to the TOS cache of current frame.
161 switch (state) {
162 case ltos:
163 case btos:
164 case ctos:
165 case stos:
166 case atos:
167 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache
168 case ftos:
169 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
170 case vtos: break; // Nothing to do, this was a void return.
171 default : ShouldNotReachHere();
172 }
173
174 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
175 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
176 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
177
178 // Compiled code destroys templateTableBase, reload.
179 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
180
181 const Register cache = R11_scratch1;
182 const Register size = R12_scratch2;
183 __ get_cache_and_index_at_bcp(cache, 1, index_size);
184
185 // Big Endian (get least significant byte of 64 bit value):
186 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
187 __ sldi(size, size, Interpreter::logStackElementSize);
188 __ add(R15_esp, R15_esp, size);
189 __ dispatch_next(state, step);
190 return entry;
191 }
192
193 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
194 address entry = __ pc();
195 // If state != vtos, we're returning from a native method, which put it's result
196 // into the result register. So move the value out of the return register back
197 // to the TOS cache of current frame.
198
199 switch (state) {
200 case ltos:
201 case btos:
202 case ctos:
203 case stos:
204 case atos:
205 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache
206 case ftos:
207 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
208 case vtos: break; // Nothing to do, this was a void return.
209 default : ShouldNotReachHere();
210 }
211
212 // Load LcpoolCache @@@ should be already set!
213 __ get_constant_pool_cache(R27_constPoolCache);
214
215 // Handle a pending exception, fall through if none.
216 __ check_and_forward_exception(R11_scratch1, R12_scratch2);
217
218 // Start executing bytecodes.
219 __ dispatch_next(state, step);
220
221 return entry;
222 }
223
224 // A result handler converts the native result into java format.
225 // Use the shared code between c++ and template interpreter.
226 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
227 return AbstractInterpreterGenerator::generate_result_handler_for(type);
228 }
229
230 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
231 address entry = __ pc();
232
233 __ push(state);
234 __ call_VM(noreg, runtime_entry);
235 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
236
237 return entry;
238 }
239
240 // Helpers for commoning out cases in the various type of method entries.
241
242 // Increment invocation count & check for overflow.
243 //
244 // Note: checking for negative value instead of overflow
245 // so we have a 'sticky' overflow test.
246 //
247 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
248 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
249 Register Rscratch1 = R11_scratch1;
250 Register Rscratch2 = R12_scratch2;
251 Register R3_counters = R3_ARG1;
252 Label done;
253
254 if (TieredCompilation) {
255 const int increment = InvocationCounter::count_increment;
256 const int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
257 Label no_mdo;
258 if (ProfileInterpreter) {
259 const Register Rmdo = Rscratch1;
260 // If no method data exists, go to profile_continue.
261 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
262 __ cmpdi(CCR0, Rmdo, 0);
263 __ beq(CCR0, no_mdo);
264
265 // Increment backedge counter in the MDO.
266 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
267 __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
268 __ addi(Rscratch2, Rscratch2, increment);
269 __ stw(Rscratch2, mdo_bc_offs, Rmdo);
270 __ load_const_optimized(Rscratch1, mask, R0);
271 __ and_(Rscratch1, Rscratch2, Rscratch1);
272 __ bne(CCR0, done);
273 __ b(*overflow);
274 }
275
276 // Increment counter in MethodCounters*.
277 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
278 __ bind(no_mdo);
279 __ get_method_counters(R19_method, R3_counters, done);
280 __ lwz(Rscratch2, mo_bc_offs, R3_counters);
281 __ addi(Rscratch2, Rscratch2, increment);
282 __ stw(Rscratch2, mo_bc_offs, R3_counters);
283 __ load_const_optimized(Rscratch1, mask, R0);
284 __ and_(Rscratch1, Rscratch2, Rscratch1);
285 __ beq(CCR0, *overflow);
286
287 __ bind(done);
288
289 } else {
290
291 // Update standard invocation counters.
292 Register Rsum_ivc_bec = R4_ARG2;
293 __ get_method_counters(R19_method, R3_counters, done);
294 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2);
295 // Increment interpreter invocation counter.
296 if (ProfileInterpreter) { // %%% Merge this into methodDataOop.
297 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
298 __ addi(R12_scratch2, R12_scratch2, 1);
299 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
300 }
301 // Check if we must create a method data obj.
302 if (ProfileInterpreter && profile_method != NULL) {
303 const Register profile_limit = Rscratch1;
304 int pl_offs = __ load_const_optimized(profile_limit, &InvocationCounter::InterpreterProfileLimit, R0, true);
305 __ lwz(profile_limit, pl_offs, profile_limit);
306 // Test to see if we should create a method data oop.
307 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit);
308 __ blt(CCR0, *profile_method_continue);
309 // If no method data exists, go to profile_method.
310 __ test_method_data_pointer(*profile_method);
311 }
312 // Finally check for counter overflow.
313 if (overflow) {
314 const Register invocation_limit = Rscratch1;
315 int il_offs = __ load_const_optimized(invocation_limit, &InvocationCounter::InterpreterInvocationLimit, R0, true);
316 __ lwz(invocation_limit, il_offs, invocation_limit);
317 assert(4 == sizeof(InvocationCounter::InterpreterInvocationLimit), "unexpected field size");
318 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit);
319 __ bge(CCR0, *overflow);
320 }
321
322 __ bind(done);
323 }
324 }
325
326 // Generate code to initiate compilation on invocation counter overflow.
327 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
328 // Generate code to initiate compilation on the counter overflow.
329
330 // InterpreterRuntime::frequency_counter_overflow takes one arguments,
331 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp)
332 // We pass zero in.
333 // The call returns the address of the verified entry point for the method or NULL
334 // if the compilation did not complete (either went background or bailed out).
335 //
336 // Unlike the C++ interpreter above: Check exceptions!
337 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
338 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
339
340 __ li(R4_ARG2, 0);
341 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
342
343 // Returns verified_entry_point or NULL.
344 // We ignore it in any case.
345 __ b(continue_entry);
346 }
347
348 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
349 assert_different_registers(Rmem_frame_size, Rscratch1);
350 __ generate_stack_overflow_check_with_compare_and_throw(Rmem_frame_size, Rscratch1);
351 }
352
353 void TemplateInterpreterGenerator::unlock_method(bool check_exceptions) {
354 __ unlock_object(R26_monitor, check_exceptions);
355 }
356
357 // Lock the current method, interpreter register window must be set up!
358 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
359 const Register Robj_to_lock = Rscratch2;
360
361 {
362 if (!flags_preloaded) {
363 __ lwz(Rflags, method_(access_flags));
364 }
365
366 #ifdef ASSERT
367 // Check if methods needs synchronization.
368 {
369 Label Lok;
370 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
371 __ btrue(CCR0,Lok);
372 __ stop("method doesn't need synchronization");
373 __ bind(Lok);
374 }
375 #endif // ASSERT
376 }
377
378 // Get synchronization object to Rscratch2.
379 {
380 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
381 Label Lstatic;
382 Label Ldone;
383
384 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
385 __ btrue(CCR0, Lstatic);
386
387 // Non-static case: load receiver obj from stack and we're done.
388 __ ld(Robj_to_lock, R18_locals);
389 __ b(Ldone);
390
391 __ bind(Lstatic); // Static case: Lock the java mirror
392 __ ld(Robj_to_lock, in_bytes(Method::const_offset()), R19_method);
393 __ ld(Robj_to_lock, in_bytes(ConstMethod::constants_offset()), Robj_to_lock);
394 __ ld(Robj_to_lock, ConstantPool::pool_holder_offset_in_bytes(), Robj_to_lock);
395 __ ld(Robj_to_lock, mirror_offset, Robj_to_lock);
396
397 __ bind(Ldone);
398 __ verify_oop(Robj_to_lock);
399 }
400
401 // Got the oop to lock => execute!
402 __ add_monitor_to_stack(true, Rscratch1, R0);
403
404 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
405 __ lock_object(R26_monitor, Robj_to_lock);
406 }
407
408 // Generate a fixed interpreter frame for pure interpreter
409 // and I2N native transition frames.
410 //
411 // Before (stack grows downwards):
412 //
413 // | ... |
414 // |------------- |
415 // | java arg0 |
416 // | ... |
417 // | java argn |
418 // | | <- R15_esp
419 // | |
420 // |--------------|
421 // | abi_112 |
422 // | | <- R1_SP
423 // |==============|
424 //
425 //
426 // After:
427 //
428 // | ... |
429 // | java arg0 |<- R18_locals
430 // | ... |
431 // | java argn |
432 // |--------------|
433 // | |
434 // | java locals |
435 // | |
436 // |--------------|
437 // | abi_48 |
438 // |==============|
439 // | |
440 // | istate |
441 // | |
442 // |--------------|
443 // | monitor |<- R26_monitor
444 // |--------------|
445 // | |<- R15_esp
446 // | expression |
447 // | stack |
448 // | |
449 // |--------------|
450 // | |
451 // | abi_112 |<- R1_SP
452 // |==============|
453 //
454 // The top most frame needs an abi space of 112 bytes. This space is needed,
455 // since we call to c. The c function may spill their arguments to the caller
456 // frame. When we call to java, we don't need these spill slots. In order to save
457 // space on the stack, we resize the caller. However, java local reside in
458 // the caller frame and the frame has to be increased. The frame_size for the
459 // current frame was calculated based on max_stack as size for the expression
460 // stack. At the call, just a part of the expression stack might be used.
461 // We don't want to waste this space and cut the frame back accordingly.
462 // The resulting amount for resizing is calculated as follows:
463 // resize = (number_of_locals - number_of_arguments) * slot_size
464 // + (R1_SP - R15_esp) + 48
465 //
466 // The size for the callee frame is calculated:
467 // framesize = 112 + max_stack + monitor + state_size
468 //
469 // maxstack: Max number of slots on the expression stack, loaded from the method.
470 // monitor: We statically reserve room for one monitor object.
471 // state_size: We save the current state of the interpreter to this area.
472 //
473 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
474 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
475 top_frame_size = R7_ARG5,
476 Rconst_method = R8_ARG6;
477
478 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size);
479
480 __ ld(Rconst_method, method_(const));
481 __ lhz(Rsize_of_parameters /* number of params */,
482 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
483 if (native_call) {
484 // If we're calling a native method, we reserve space for the worst-case signature
485 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2).
486 // We add two slots to the parameter_count, one for the jni
487 // environment and one for a possible native mirror.
488 Label skip_native_calculate_max_stack;
489 __ addi(top_frame_size, Rsize_of_parameters, 2);
490 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters);
491 __ bge(CCR0, skip_native_calculate_max_stack);
492 __ li(top_frame_size, Argument::n_register_parameters);
493 __ bind(skip_native_calculate_max_stack);
494 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
495 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
496 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
497 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
498 } else {
499 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
500 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
501 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
502 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
503 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
504 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
505 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
506 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1);
507 }
508
509 // Compute top frame size.
510 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size);
511
512 // Cut back area between esp and max_stack.
513 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize);
514
515 __ round_to(top_frame_size, frame::alignment_in_bytes);
516 __ round_to(parent_frame_resize, frame::alignment_in_bytes);
517 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
518 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
519
520 {
521 // --------------------------------------------------------------------------
522 // Stack overflow check
523
524 Label cont;
525 __ add(R11_scratch1, parent_frame_resize, top_frame_size);
526 generate_stack_overflow_check(R11_scratch1, R12_scratch2);
527 }
528
529 // Set up interpreter state registers.
530
531 __ add(R18_locals, R15_esp, Rsize_of_parameters);
532 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
533 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache);
534
535 // Set method data pointer.
536 if (ProfileInterpreter) {
537 Label zero_continue;
538 __ ld(R28_mdx, method_(method_data));
539 __ cmpdi(CCR0, R28_mdx, 0);
540 __ beq(CCR0, zero_continue);
541 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
542 __ bind(zero_continue);
543 }
544
545 if (native_call) {
546 __ li(R14_bcp, 0); // Must initialize.
547 } else {
548 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method);
549 }
550
551 // Resize parent frame.
552 __ mflr(R12_scratch2);
553 __ neg(parent_frame_resize, parent_frame_resize);
554 __ resize_frame(parent_frame_resize, R11_scratch1);
555 __ std(R12_scratch2, _abi(lr), R1_SP);
556
557 __ addi(R26_monitor, R1_SP, - frame::ijava_state_size);
558 __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
559
560 // Store values.
561 // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls
562 // in InterpreterMacroAssembler::call_from_interpreter.
563 __ std(R19_method, _ijava_state_neg(method), R1_SP);
564 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
565 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
566 __ std(R18_locals, _ijava_state_neg(locals), R1_SP);
567
568 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
569 // be found in the frame after save_interpreter_state is done. This is always true
570 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
571 // because e.g. frame::interpreter_frame_bcp() will not access the correct value
572 // (Enhanced Stack Trace).
573 // The signal handler does not save the interpreter state into the frame.
574 __ li(R0, 0);
575 #ifdef ASSERT
576 // Fill remaining slots with constants.
577 __ load_const_optimized(R11_scratch1, 0x5afe);
578 __ load_const_optimized(R12_scratch2, 0xdead);
579 #endif
580 // We have to initialize some frame slots for native calls (accessed by GC).
581 if (native_call) {
582 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP);
583 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
584 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
585 }
586 #ifdef ASSERT
587 else {
588 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
589 __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP);
590 __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP);
591 }
592 __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP);
593 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
594 __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP);
595 __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP);
596 #endif
597 __ subf(R12_scratch2, top_frame_size, R1_SP);
598 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP);
599 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
600
601 // Push top frame.
602 __ push_frame(top_frame_size, R11_scratch1);
603 }
604
605 // End of helpers
606
607 // ============================================================================
608 // Various method entries
609 //
610
611 // Empty method, generate a very fast return. We must skip this entry if
612 // someone's debugging, indicated by the flag
613 // "interp_mode" in the Thread obj.
614 // Note: empty methods are generated mostly methods that do assertions, which are
615 // disabled in the "java opt build".
616 address TemplateInterpreterGenerator::generate_empty_entry(void) {
617 if (!UseFastEmptyMethods) {
618 NOT_PRODUCT(__ should_not_reach_here();)
619 return Interpreter::entry_for_kind(Interpreter::zerolocals);
620 }
621
622 Label Lslow_path;
623 const Register Rjvmti_mode = R11_scratch1;
624 address entry = __ pc();
625
626 __ lwz(Rjvmti_mode, thread_(interp_only_mode));
627 __ cmpwi(CCR0, Rjvmti_mode, 0);
628 __ bne(CCR0, Lslow_path); // jvmti_mode!=0
629
630 // Noone's debuggin: Simply return.
631 // Pop c2i arguments (if any) off when we return.
632 #ifdef ASSERT
633 __ ld(R9_ARG7, 0, R1_SP);
634 __ ld(R10_ARG8, 0, R21_sender_SP);
635 __ cmpd(CCR0, R9_ARG7, R10_ARG8);
636 __ asm_assert_eq("backlink", 0x545);
637 #endif // ASSERT
638 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
639
640 // And we're done.
641 __ blr();
642
643 __ bind(Lslow_path);
644 __ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
645 __ flush();
646
647 return entry;
648 }
649
650 // Support abs and sqrt like in compiler.
651 // For others we can use a normal (native) entry.
652
653 inline bool math_entry_available(AbstractInterpreter::MethodKind kind) {
654 // Provide math entry with debugging on demand.
655 // Note: Debugging changes which code will get executed:
656 // Debugging or disabled InlineIntrinsics: java method will get interpreted and performs a native call.
657 // Not debugging and enabled InlineIntrinics: processor instruction will get used.
658 // Result might differ slightly due to rounding etc.
659 if (!InlineIntrinsics && (!FLAG_IS_ERGO(InlineIntrinsics))) return false; // Generate a vanilla entry.
660
661 return ((kind==Interpreter::java_lang_math_sqrt && VM_Version::has_fsqrt()) ||
662 (kind==Interpreter::java_lang_math_abs));
663 }
664
665 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
666 if (!math_entry_available(kind)) {
667 NOT_PRODUCT(__ should_not_reach_here();)
668 return Interpreter::entry_for_kind(Interpreter::zerolocals);
669 }
670
671 Label Lslow_path;
672 const Register Rjvmti_mode = R11_scratch1;
673 address entry = __ pc();
674
675 // Provide math entry with debugging on demand.
676 __ lwz(Rjvmti_mode, thread_(interp_only_mode));
677 __ cmpwi(CCR0, Rjvmti_mode, 0);
678 __ bne(CCR0, Lslow_path); // jvmti_mode!=0
679
680 __ lfd(F1_RET, Interpreter::stackElementSize, R15_esp);
681
682 // Pop c2i arguments (if any) off when we return.
683 #ifdef ASSERT
684 __ ld(R9_ARG7, 0, R1_SP);
685 __ ld(R10_ARG8, 0, R21_sender_SP);
686 __ cmpd(CCR0, R9_ARG7, R10_ARG8);
687 __ asm_assert_eq("backlink", 0x545);
688 #endif // ASSERT
689 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
690
691 if (kind == Interpreter::java_lang_math_sqrt) {
692 __ fsqrt(F1_RET, F1_RET);
693 } else if (kind == Interpreter::java_lang_math_abs) {
694 __ fabs(F1_RET, F1_RET);
695 } else {
696 ShouldNotReachHere();
697 }
698
699 // And we're done.
700 __ blr();
701
702 // Provide slow path for JVMTI case.
703 __ bind(Lslow_path);
704 __ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R12_scratch2);
705 __ flush();
706
707 return entry;
708 }
709
710 // Interpreter stub for calling a native method. (asm interpreter)
711 // This sets up a somewhat different looking stack for calling the
712 // native method than the typical interpreter frame setup.
713 //
714 // On entry:
715 // R19_method - method
716 // R16_thread - JavaThread*
717 // R15_esp - intptr_t* sender tos
718 //
719 // abstract stack (grows up)
720 // [ IJava (caller of JNI callee) ] <-- ASP
721 // ...
722 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
723
724 address entry = __ pc();
725
726 const bool inc_counter = UseCompiler || CountCompiledCalls;
727
728 // -----------------------------------------------------------------------------
729 // Allocate a new frame that represents the native callee (i2n frame).
730 // This is not a full-blown interpreter frame, but in particular, the
731 // following registers are valid after this:
732 // - R19_method
733 // - R18_local (points to start of argumuments to native function)
734 //
735 // abstract stack (grows up)
736 // [ IJava (caller of JNI callee) ] <-- ASP
737 // ...
738
739 const Register signature_handler_fd = R11_scratch1;
740 const Register pending_exception = R0;
741 const Register result_handler_addr = R31;
742 const Register native_method_fd = R11_scratch1;
743 const Register access_flags = R22_tmp2;
744 const Register active_handles = R11_scratch1; // R26_monitor saved to state.
745 const Register sync_state = R12_scratch2;
746 const Register sync_state_addr = sync_state; // Address is dead after use.
747 const Register suspend_flags = R11_scratch1;
748
749 //=============================================================================
750 // Allocate new frame and initialize interpreter state.
751
752 Label exception_return;
753 Label exception_return_sync_check;
754 Label stack_overflow_return;
755
756 // Generate new interpreter state and jump to stack_overflow_return in case of
757 // a stack overflow.
758 //generate_compute_interpreter_state(stack_overflow_return);
759
760 Register size_of_parameters = R22_tmp2;
761
762 generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
763
764 //=============================================================================
765 // Increment invocation counter. On overflow, entry to JNI method
766 // will be compiled.
767 Label invocation_counter_overflow, continue_after_compile;
768 if (inc_counter) {
769 if (synchronized) {
770 // Since at this point in the method invocation the exception handler
771 // would try to exit the monitor of synchronized methods which hasn't
772 // been entered yet, we set the thread local variable
773 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
774 // runtime, exception handling i.e. unlock_if_synchronized_method will
775 // check this thread local flag.
776 // This flag has two effects, one is to force an unwind in the topmost
777 // interpreter frame and not perform an unlock while doing so.
778 __ li(R0, 1);
779 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
780 }
781 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
782
783 __ BIND(continue_after_compile);
784 // Reset the _do_not_unlock_if_synchronized flag.
785 if (synchronized) {
786 __ li(R0, 0);
787 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
788 }
789 }
790
791 // access_flags = method->access_flags();
792 // Load access flags.
793 assert(access_flags->is_nonvolatile(),
794 "access_flags must be in a non-volatile register");
795 // Type check.
796 assert(4 == sizeof(AccessFlags), "unexpected field size");
797 __ lwz(access_flags, method_(access_flags));
798
799 // We don't want to reload R19_method and access_flags after calls
800 // to some helper functions.
801 assert(R19_method->is_nonvolatile(),
802 "R19_method must be a non-volatile register");
803
804 // Check for synchronized methods. Must happen AFTER invocation counter
805 // check, so method is not locked if counter overflows.
806
807 if (synchronized) {
808 lock_method(access_flags, R11_scratch1, R12_scratch2, true);
809
810 // Update monitor in state.
811 __ ld(R11_scratch1, 0, R1_SP);
812 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1);
813 }
814
815 // jvmti/jvmpi support
816 __ notify_method_entry();
817
818 //=============================================================================
819 // Get and call the signature handler.
820
821 __ ld(signature_handler_fd, method_(signature_handler));
822 Label call_signature_handler;
823
824 __ cmpdi(CCR0, signature_handler_fd, 0);
825 __ bne(CCR0, call_signature_handler);
826
827 // Method has never been called. Either generate a specialized
828 // handler or point to the slow one.
829 //
830 // Pass parameter 'false' to avoid exception check in call_VM.
831 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
832
833 // Check for an exception while looking up the target method. If we
834 // incurred one, bail.
835 __ ld(pending_exception, thread_(pending_exception));
836 __ cmpdi(CCR0, pending_exception, 0);
837 __ bne(CCR0, exception_return_sync_check); // Has pending exception.
838
839 // Reload signature handler, it may have been created/assigned in the meanwhile.
840 __ ld(signature_handler_fd, method_(signature_handler));
841 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
842
843 __ BIND(call_signature_handler);
844
845 // Before we call the signature handler we push a new frame to
846 // protect the interpreter frame volatile registers when we return
847 // from jni but before we can get back to Java.
848
849 // First set the frame anchor while the SP/FP registers are
850 // convenient and the slow signature handler can use this same frame
851 // anchor.
852
853 // We have a TOP_IJAVA_FRAME here, which belongs to us.
854 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
855
856 // Now the interpreter frame (and its call chain) have been
857 // invalidated and flushed. We are now protected against eager
858 // being enabled in native code. Even if it goes eager the
859 // registers will be reloaded as clean and we will invalidate after
860 // the call so no spurious flush should be possible.
861
862 // Call signature handler and pass locals address.
863 //
864 // Our signature handlers copy required arguments to the C stack
865 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
866 __ mr(R3_ARG1, R18_locals);
867 __ ld(signature_handler_fd, 0, signature_handler_fd);
868
869 __ call_stub(signature_handler_fd);
870
871 // Remove the register parameter varargs slots we allocated in
872 // compute_interpreter_state. SP+16 ends up pointing to the ABI
873 // outgoing argument area.
874 //
875 // Not needed on PPC64.
876 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord);
877
878 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
879 // Save across call to native method.
880 __ mr(result_handler_addr, R3_RET);
881
882 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
883
884 // Set up fixed parameters and call the native method.
885 // If the method is static, get mirror into R4_ARG2.
886 {
887 Label method_is_not_static;
888 // Access_flags is non-volatile and still, no need to restore it.
889
890 // Restore access flags.
891 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
892 __ bfalse(CCR0, method_is_not_static);
893
894 // constants = method->constants();
895 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
896 __ ld(R11_scratch1, in_bytes(ConstMethod::constants_offset()), R11_scratch1);
897 // pool_holder = method->constants()->pool_holder();
898 __ ld(R11_scratch1/*pool_holder*/, ConstantPool::pool_holder_offset_in_bytes(),
899 R11_scratch1/*constants*/);
900
901 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
902
903 // mirror = pool_holder->klass_part()->java_mirror();
904 __ ld(R0/*mirror*/, mirror_offset, R11_scratch1/*pool_holder*/);
905 // state->_native_mirror = mirror;
906
907 __ ld(R11_scratch1, 0, R1_SP);
908 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
909 // R4_ARG2 = &state->_oop_temp;
910 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
911 __ BIND(method_is_not_static);
912 }
913
914 // At this point, arguments have been copied off the stack into
915 // their JNI positions. Oops are boxed in-place on the stack, with
916 // handles copied to arguments. The result handler address is in a
917 // register.
918
919 // Pass JNIEnv address as first parameter.
920 __ addir(R3_ARG1, thread_(jni_environment));
921
922 // Load the native_method entry before we change the thread state.
923 __ ld(native_method_fd, method_(native_function));
924
925 //=============================================================================
926 // Transition from _thread_in_Java to _thread_in_native. As soon as
927 // we make this change the safepoint code needs to be certain that
928 // the last Java frame we established is good. The pc in that frame
929 // just needs to be near here not an actual return address.
930
931 // We use release_store_fence to update values like the thread state, where
932 // we don't want the current thread to continue until all our prior memory
933 // accesses (including the new thread state) are visible to other threads.
934 __ li(R0, _thread_in_native);
935 __ release();
936
937 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
938 __ stw(R0, thread_(thread_state));
939
940 if (UseMembar) {
941 __ fence();
942 }
943
944 //=============================================================================
945 // Call the native method. Argument registers must not have been
946 // overwritten since "__ call_stub(signature_handler);" (except for
947 // ARG1 and ARG2 for static methods).
948 __ call_c(native_method_fd);
949
950 __ li(R0, 0);
951 __ ld(R11_scratch1, 0, R1_SP);
952 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
953 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
954 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
955
956 // Note: C++ interpreter needs the following here:
957 // The frame_manager_lr field, which we use for setting the last
958 // java frame, gets overwritten by the signature handler. Restore
959 // it now.
960 //__ get_PC_trash_LR(R11_scratch1);
961 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
962
963 // Because of GC R19_method may no longer be valid.
964
965 // Block, if necessary, before resuming in _thread_in_Java state.
966 // In order for GC to work, don't clear the last_Java_sp until after
967 // blocking.
968
969 //=============================================================================
970 // Switch thread to "native transition" state before reading the
971 // synchronization state. This additional state is necessary
972 // because reading and testing the synchronization state is not
973 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
974 // in _thread_in_native state, loads _not_synchronized and is
975 // preempted. VM thread changes sync state to synchronizing and
976 // suspends threads for GC. Thread A is resumed to finish this
977 // native method, but doesn't block here since it didn't see any
978 // synchronization in progress, and escapes.
979
980 // We use release_store_fence to update values like the thread state, where
981 // we don't want the current thread to continue until all our prior memory
982 // accesses (including the new thread state) are visible to other threads.
983 __ li(R0/*thread_state*/, _thread_in_native_trans);
984 __ release();
985 __ stw(R0/*thread_state*/, thread_(thread_state));
986 if (UseMembar) {
987 __ fence();
988 }
989 // Write serialization page so that the VM thread can do a pseudo remote
990 // membar. We use the current thread pointer to calculate a thread
991 // specific offset to write to within the page. This minimizes bus
992 // traffic due to cache line collision.
993 else {
994 __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2);
995 }
996
997 // Now before we return to java we must look for a current safepoint
998 // (a new safepoint can not start since we entered native_trans).
999 // We must check here because a current safepoint could be modifying
1000 // the callers registers right this moment.
1001
1002 // Acquire isn't strictly necessary here because of the fence, but
1003 // sync_state is declared to be volatile, so we do it anyway
1004 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1005 int sync_state_offs = __ load_const_optimized(sync_state_addr, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
1006
1007 // TODO PPC port assert(4 == SafepointSynchronize::sz_state(), "unexpected field size");
1008 __ lwz(sync_state, sync_state_offs, sync_state_addr);
1009
1010 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1011 __ lwz(suspend_flags, thread_(suspend_flags));
1012
1013 Label sync_check_done;
1014 Label do_safepoint;
1015 // No synchronization in progress nor yet synchronized.
1016 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
1017 // Not suspended.
1018 __ cmpwi(CCR1, suspend_flags, 0);
1019
1020 __ bne(CCR0, do_safepoint);
1021 __ beq(CCR1, sync_check_done);
1022 __ bind(do_safepoint);
1023 __ isync();
1024 // Block. We do the call directly and leave the current
1025 // last_Java_frame setup undisturbed. We must save any possible
1026 // native result across the call. No oop is present.
1027
1028 __ mr(R3_ARG1, R16_thread);
1029 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
1030 relocInfo::none);
1031
1032 __ bind(sync_check_done);
1033
1034 //=============================================================================
1035 // <<<<<< Back in Interpreter Frame >>>>>
1036
1037 // We are in thread_in_native_trans here and back in the normal
1038 // interpreter frame. We don't have to do anything special about
1039 // safepoints and we can switch to Java mode anytime we are ready.
1040
1041 // Note: frame::interpreter_frame_result has a dependency on how the
1042 // method result is saved across the call to post_method_exit. For
1043 // native methods it assumes that the non-FPU/non-void result is
1044 // saved in _native_lresult and a FPU result in _native_fresult. If
1045 // this changes then the interpreter_frame_result implementation
1046 // will need to be updated too.
1047
1048 // On PPC64, we have stored the result directly after the native call.
1049
1050 //=============================================================================
1051 // Back in Java
1052
1053 // We use release_store_fence to update values like the thread state, where
1054 // we don't want the current thread to continue until all our prior memory
1055 // accesses (including the new thread state) are visible to other threads.
1056 __ li(R0/*thread_state*/, _thread_in_Java);
1057 __ release();
1058 __ stw(R0/*thread_state*/, thread_(thread_state));
1059 if (UseMembar) {
1060 __ fence();
1061 }
1062
1063 __ reset_last_Java_frame();
1064
1065 // Jvmdi/jvmpi support. Whether we've got an exception pending or
1066 // not, and whether unlocking throws an exception or not, we notify
1067 // on native method exit. If we do have an exception, we'll end up
1068 // in the caller's context to handle it, so if we don't do the
1069 // notify here, we'll drop it on the floor.
1070 __ notify_method_exit(true/*native method*/,
1071 ilgl /*illegal state (not used for native methods)*/,
1072 InterpreterMacroAssembler::NotifyJVMTI,
1073 false /*check_exceptions*/);
1074
1075 //=============================================================================
1076 // Handle exceptions
1077
1078 if (synchronized) {
1079 // Don't check for exceptions since we're still in the i2n frame. Do that
1080 // manually afterwards.
1081 unlock_method(false);
1082 }
1083
1084 // Reset active handles after returning from native.
1085 // thread->active_handles()->clear();
1086 __ ld(active_handles, thread_(active_handles));
1087 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1088 __ li(R0, 0);
1089 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles);
1090
1091 Label exception_return_sync_check_already_unlocked;
1092 __ ld(R0/*pending_exception*/, thread_(pending_exception));
1093 __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1094 __ bne(CCR0, exception_return_sync_check_already_unlocked);
1095
1096 //-----------------------------------------------------------------------------
1097 // No exception pending.
1098
1099 // Move native method result back into proper registers and return.
1100 // Invoke result handler (may unbox/promote).
1101 __ ld(R11_scratch1, 0, R1_SP);
1102 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1103 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1104 __ call_stub(result_handler_addr);
1105
1106 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1107
1108 // Must use the return pc which was loaded from the caller's frame
1109 // as the VM uses return-pc-patching for deoptimization.
1110 __ mtlr(R0);
1111 __ blr();
1112
1113 //-----------------------------------------------------------------------------
1114 // An exception is pending. We call into the runtime only if the
1115 // caller was not interpreted. If it was interpreted the
1116 // interpreter will do the correct thing. If it isn't interpreted
1117 // (call stub/compiled code) we will change our return and continue.
1118
1119 __ BIND(exception_return_sync_check);
1120
1121 if (synchronized) {
1122 // Don't check for exceptions since we're still in the i2n frame. Do that
1123 // manually afterwards.
1124 unlock_method(false);
1125 }
1126 __ BIND(exception_return_sync_check_already_unlocked);
1127
1128 const Register return_pc = R31;
1129
1130 __ ld(return_pc, 0, R1_SP);
1131 __ ld(return_pc, _abi(lr), return_pc);
1132
1133 // Get the address of the exception handler.
1134 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1135 R16_thread,
1136 return_pc /* return pc */);
1137 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1138
1139 // Load the PC of the the exception handler into LR.
1140 __ mtlr(R3_RET);
1141
1142 // Load exception into R3_ARG1 and clear pending exception in thread.
1143 __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1144 __ li(R4_ARG2, 0);
1145 __ std(R4_ARG2, thread_(pending_exception));
1146
1147 // Load the original return pc into R4_ARG2.
1148 __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1149
1150 // Return to exception handler.
1151 __ blr();
1152
1153 //=============================================================================
1154 // Counter overflow.
1155
1156 if (inc_counter) {
1157 // Handle invocation counter overflow.
1158 __ bind(invocation_counter_overflow);
1159
1160 generate_counter_overflow(continue_after_compile);
1161 }
1162
1163 return entry;
1164 }
1165
1166 // Generic interpreted method entry to (asm) interpreter.
1167 //
1168 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1169 bool inc_counter = UseCompiler || CountCompiledCalls;
1170 address entry = __ pc();
1171 // Generate the code to allocate the interpreter stack frame.
1172 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1173 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame.
1174
1175 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1176
1177 #ifdef FAST_DISPATCH
1178 __ unimplemented("Fast dispatch in generate_normal_entry");
1179 #if 0
1180 __ set((intptr_t)Interpreter::dispatch_table(), IdispatchTables);
1181 // Set bytecode dispatch table base.
1182 #endif
1183 #endif
1184
1185 // --------------------------------------------------------------------------
1186 // Zero out non-parameter locals.
1187 // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1188 // worth to ask the flag, just do it.
1189 Register Rslot_addr = R6_ARG4,
1190 Rnum = R7_ARG5;
1191 Label Lno_locals, Lzero_loop;
1192
1193 // Set up the zeroing loop.
1194 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1195 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1196 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1197 __ beq(CCR0, Lno_locals);
1198 __ li(R0, 0);
1199 __ mtctr(Rnum);
1200
1201 // The zero locals loop.
1202 __ bind(Lzero_loop);
1203 __ std(R0, 0, Rslot_addr);
1204 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1205 __ bdnz(Lzero_loop);
1206
1207 __ bind(Lno_locals);
1208
1209 // --------------------------------------------------------------------------
1210 // Counter increment and overflow check.
1211 Label invocation_counter_overflow,
1212 profile_method,
1213 profile_method_continue;
1214 if (inc_counter || ProfileInterpreter) {
1215
1216 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1217 if (synchronized) {
1218 // Since at this point in the method invocation the exception handler
1219 // would try to exit the monitor of synchronized methods which hasn't
1220 // been entered yet, we set the thread local variable
1221 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1222 // runtime, exception handling i.e. unlock_if_synchronized_method will
1223 // check this thread local flag.
1224 // This flag has two effects, one is to force an unwind in the topmost
1225 // interpreter frame and not perform an unlock while doing so.
1226 __ li(R0, 1);
1227 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1228 }
1229 // Increment invocation counter and check for overflow.
1230 if (inc_counter) {
1231 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1232 }
1233
1234 __ bind(profile_method_continue);
1235
1236 // Reset the _do_not_unlock_if_synchronized flag.
1237 if (synchronized) {
1238 __ li(R0, 0);
1239 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1240 }
1241 }
1242
1243 // --------------------------------------------------------------------------
1244 // Locking of synchronized methods. Must happen AFTER invocation_counter
1245 // check and stack overflow check, so method is not locked if overflows.
1246 if (synchronized) {
1247 lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1248 }
1249 #ifdef ASSERT
1250 else {
1251 Label Lok;
1252 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1253 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1254 __ asm_assert_eq("method needs synchronization", 0x8521);
1255 __ bind(Lok);
1256 }
1257 #endif // ASSERT
1258
1259 __ verify_thread();
1260
1261 // --------------------------------------------------------------------------
1262 // JVMTI support
1263 __ notify_method_entry();
1264
1265 // --------------------------------------------------------------------------
1266 // Start executing instructions.
1267 __ dispatch_next(vtos);
1268
1269 // --------------------------------------------------------------------------
1270 // Out of line counter overflow and MDO creation code.
1271 if (ProfileInterpreter) {
1272 // We have decided to profile this method in the interpreter.
1273 __ bind(profile_method);
1274 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1275 __ set_method_data_pointer_for_bcp();
1276 __ b(profile_method_continue);
1277 }
1278
1279 if (inc_counter) {
1280 // Handle invocation counter overflow.
1281 __ bind(invocation_counter_overflow);
1282 generate_counter_overflow(profile_method_continue);
1283 }
1284 return entry;
1285 }
1286
1287 // =============================================================================
1288 // Entry points
1289
1290 address AbstractInterpreterGenerator::generate_method_entry(
1291 AbstractInterpreter::MethodKind kind) {
1292 // Determine code generation flags.
1293 bool synchronized = false;
1294 address entry_point = NULL;
1295
1296 switch (kind) {
1297 case Interpreter::zerolocals : break;
1298 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1299 case Interpreter::native : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1300 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true); break;
1301 case Interpreter::empty : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry(); break;
1302 case Interpreter::accessor : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry(); break;
1303 case Interpreter::abstract : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry(); break;
1304
1305 case Interpreter::java_lang_math_sin : // fall thru
1306 case Interpreter::java_lang_math_cos : // fall thru
1307 case Interpreter::java_lang_math_tan : // fall thru
1308 case Interpreter::java_lang_math_abs : // fall thru
1309 case Interpreter::java_lang_math_log : // fall thru
1310 case Interpreter::java_lang_math_log10 : // fall thru
1311 case Interpreter::java_lang_math_sqrt : // fall thru
1312 case Interpreter::java_lang_math_pow : // fall thru
1313 case Interpreter::java_lang_math_exp : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind); break;
1314 case Interpreter::java_lang_ref_reference_get
1315 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1316 default : ShouldNotReachHere(); break;
1317 }
1318
1319 if (entry_point) {
1320 return entry_point;
1321 }
1322
1323 return ((InterpreterGenerator*) this)->generate_normal_entry(synchronized);
1324 }
1325
1326 // These should never be compiled since the interpreter will prefer
1327 // the compiled version to the intrinsic version.
1328 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1329 return !math_entry_available(method_kind(m));
1330 }
1331
1332 // How much stack a method activation needs in stack slots.
1333 // We must calc this exactly like in generate_fixed_frame.
1334 // Note: This returns the conservative size assuming maximum alignment.
1335 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1336 const int max_alignment_size = 2;
1337 const int abi_scratch = frame::abi_reg_args_size;
1338 return method->max_locals() + method->max_stack() +
1339 frame::interpreter_frame_monitor_size() + max_alignment_size + abi_scratch;
1340 }
1341
1342 // Returns number of stackElementWords needed for the interpreter frame with the
1343 // given sections.
1344 // This overestimates the stack by one slot in case of alignments.
1345 int AbstractInterpreter::size_activation(int max_stack,
1346 int temps,
1347 int popframe_args,
1348 int monitors,
1349 int callee_params,
1350 int callee_locals,
1351 bool is_top_frame) {
1352 // Note: This calculation must exactly parallel the frame setup
1353 // in AbstractInterpreterGenerator::generate_method_entry.
1354 assert(Interpreter::stackElementWords == 1, "sanity");
1355 const int max_alignment_space = StackAlignmentInBytes / Interpreter::stackElementSize;
1356 const int abi_scratch = is_top_frame ? (frame::abi_reg_args_size / Interpreter::stackElementSize) :
1357 (frame::abi_minframe_size / Interpreter::stackElementSize);
1358 const int size =
1359 max_stack +
1360 (callee_locals - callee_params) +
1361 monitors * frame::interpreter_frame_monitor_size() +
1362 max_alignment_space +
1363 abi_scratch +
1364 frame::ijava_state_size / Interpreter::stackElementSize;
1365
1366 // Fixed size of an interpreter frame, align to 16-byte.
1367 return (size & -2);
1368 }
1369
1370 // Fills a sceletal interpreter frame generated during deoptimizations.
1371 //
1372 // Parameters:
1373 //
1374 // interpreter_frame != NULL:
1375 // set up the method, locals, and monitors.
1376 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1377 // right size, as determined by a previous call to this method.
1378 // It is also guaranteed to be walkable even though it is in a skeletal state
1379 //
1380 // is_top_frame == true:
1381 // We're processing the *oldest* interpreter frame!
1382 //
1383 // pop_frame_extra_args:
1384 // If this is != 0 we are returning to a deoptimized frame by popping
1385 // off the callee frame. We want to re-execute the call that called the
1386 // callee interpreted, but since the return to the interpreter would pop
1387 // the arguments off advance the esp by dummy popframe_extra_args slots.
1388 // Popping off those will establish the stack layout as it was before the call.
1389 //
1390 void AbstractInterpreter::layout_activation(Method* method,
1391 int tempcount,
1392 int popframe_extra_args,
1393 int moncount,
1394 int caller_actual_parameters,
1395 int callee_param_count,
1396 int callee_locals_count,
1397 frame* caller,
1398 frame* interpreter_frame,
1399 bool is_top_frame,
1400 bool is_bottom_frame) {
1401
1402 const int abi_scratch = is_top_frame ? (frame::abi_reg_args_size / Interpreter::stackElementSize) :
1403 (frame::abi_minframe_size / Interpreter::stackElementSize);
1404
1405 intptr_t* locals_base = (caller->is_interpreted_frame()) ?
1406 caller->interpreter_frame_esp() + caller_actual_parameters :
1407 caller->sp() + method->max_locals() - 1 + (frame::abi_minframe_size / Interpreter::stackElementSize) ;
1408
1409 intptr_t* monitor_base = caller->sp() - frame::ijava_state_size / Interpreter::stackElementSize ;
1410 intptr_t* monitor = monitor_base - (moncount * frame::interpreter_frame_monitor_size());
1411 intptr_t* esp_base = monitor - 1;
1412 intptr_t* esp = esp_base - tempcount - popframe_extra_args;
1413 intptr_t* sp = (intptr_t *) (((intptr_t) (esp_base - callee_locals_count + callee_param_count - method->max_stack()- abi_scratch)) & -StackAlignmentInBytes);
1414 intptr_t* sender_sp = caller->sp() + (frame::abi_minframe_size - frame::abi_reg_args_size) / Interpreter::stackElementSize;
1415 intptr_t* top_frame_sp = is_top_frame ? sp : sp + (frame::abi_minframe_size - frame::abi_reg_args_size) / Interpreter::stackElementSize;
1416
1417 interpreter_frame->interpreter_frame_set_method(method);
1418 interpreter_frame->interpreter_frame_set_locals(locals_base);
1419 interpreter_frame->interpreter_frame_set_cpcache(method->constants()->cache());
1420 interpreter_frame->interpreter_frame_set_esp(esp);
1421 interpreter_frame->interpreter_frame_set_monitor_end((BasicObjectLock *)monitor);
1422 interpreter_frame->interpreter_frame_set_top_frame_sp(top_frame_sp);
1423 if (!is_bottom_frame) {
1424 interpreter_frame->interpreter_frame_set_sender_sp(sender_sp);
1425 }
1426 }
1427
1428 // =============================================================================
1429 // Exceptions
1430
1431 void TemplateInterpreterGenerator::generate_throw_exception() {
1432 Register Rexception = R17_tos,
1433 Rcontinuation = R3_RET;
1434
1435 // --------------------------------------------------------------------------
1436 // Entry point if an method returns with a pending exception (rethrow).
1437 Interpreter::_rethrow_exception_entry = __ pc();
1438 {
1439 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
1440 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
1441 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
1442
1443 // Compiled code destroys templateTableBase, reload.
1444 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
1445 }
1446
1447 // Entry point if a interpreted method throws an exception (throw).
1448 Interpreter::_throw_exception_entry = __ pc();
1449 {
1450 __ mr(Rexception, R3_RET);
1451
1452 __ verify_thread();
1453 __ verify_oop(Rexception);
1454
1455 // Expression stack must be empty before entering the VM in case of an exception.
1456 __ empty_expression_stack();
1457 // Find exception handler address and preserve exception oop.
1458 // Call C routine to find handler and jump to it.
1459 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
1460 __ mtctr(Rcontinuation);
1461 // Push exception for exception handler bytecodes.
1462 __ push_ptr(Rexception);
1463
1464 // Jump to exception handler (may be remove activation entry!).
1465 __ bctr();
1466 }
1467
1468 // If the exception is not handled in the current frame the frame is
1469 // removed and the exception is rethrown (i.e. exception
1470 // continuation is _rethrow_exception).
1471 //
1472 // Note: At this point the bci is still the bxi for the instruction
1473 // which caused the exception and the expression stack is
1474 // empty. Thus, for any VM calls at this point, GC will find a legal
1475 // oop map (with empty expression stack).
1476
1477 // In current activation
1478 // tos: exception
1479 // bcp: exception bcp
1480
1481 // --------------------------------------------------------------------------
1482 // JVMTI PopFrame support
1483
1484 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1485 {
1486 // Set the popframe_processing bit in popframe_condition indicating that we are
1487 // currently handling popframe, so that call_VMs that may happen later do not
1488 // trigger new popframe handling cycles.
1489 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1490 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
1491 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1492
1493 // Empty the expression stack, as in normal exception handling.
1494 __ empty_expression_stack();
1495 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
1496
1497 // Check to see whether we are returning to a deoptimized frame.
1498 // (The PopFrame call ensures that the caller of the popped frame is
1499 // either interpreted or compiled and deoptimizes it if compiled.)
1500 // Note that we don't compare the return PC against the
1501 // deoptimization blob's unpack entry because of the presence of
1502 // adapter frames in C2.
1503 Label Lcaller_not_deoptimized;
1504 Register return_pc = R3_ARG1;
1505 __ ld(return_pc, 0, R1_SP);
1506 __ ld(return_pc, _abi(lr), return_pc);
1507 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
1508 __ cmpdi(CCR0, R3_RET, 0);
1509 __ bne(CCR0, Lcaller_not_deoptimized);
1510
1511 // The deoptimized case.
1512 // In this case, we can't call dispatch_next() after the frame is
1513 // popped, but instead must save the incoming arguments and restore
1514 // them after deoptimization has occurred.
1515 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
1516 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
1517 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
1518 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
1519 __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
1520 // Save these arguments.
1521 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
1522
1523 // Inform deoptimization that it is responsible for restoring these arguments.
1524 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1525 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1526
1527 // Return from the current method into the deoptimization blob. Will eventually
1528 // end up in the deopt interpeter entry, deoptimization prepared everything that
1529 // we will reexecute the call that called us.
1530 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
1531 __ mtlr(return_pc);
1532 __ blr();
1533
1534 // The non-deoptimized case.
1535 __ bind(Lcaller_not_deoptimized);
1536
1537 // Clear the popframe condition flag.
1538 __ li(R0, 0);
1539 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1540
1541 // Get out of the current method and re-execute the call that called us.
1542 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ return_pc, R11_scratch1, R12_scratch2);
1543 __ restore_interpreter_state(R11_scratch1);
1544 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
1545 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
1546 __ mtlr(return_pc);
1547 if (ProfileInterpreter) {
1548 __ set_method_data_pointer_for_bcp();
1549 }
1550 __ dispatch_next(vtos);
1551 }
1552 // end of JVMTI PopFrame support
1553
1554 // --------------------------------------------------------------------------
1555 // Remove activation exception entry.
1556 // This is jumped to if an interpreted method can't handle an exception itself
1557 // (we come from the throw/rethrow exception entry above). We're going to call
1558 // into the VM to find the exception handler in the caller, pop the current
1559 // frame and return the handler we calculated.
1560 Interpreter::_remove_activation_entry = __ pc();
1561 {
1562 __ pop_ptr(Rexception);
1563 __ verify_thread();
1564 __ verify_oop(Rexception);
1565 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
1566
1567 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
1568 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
1569
1570 __ get_vm_result(Rexception);
1571
1572 // We are done with this activation frame; find out where to go next.
1573 // The continuation point will be an exception handler, which expects
1574 // the following registers set up:
1575 //
1576 // RET: exception oop
1577 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
1578
1579 Register return_pc = R31; // Needs to survive the runtime call.
1580 __ ld(return_pc, 0, R1_SP);
1581 __ ld(return_pc, _abi(lr), return_pc);
1582 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
1583
1584 // Remove the current activation.
1585 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
1586
1587 __ mr(R4_ARG2, return_pc);
1588 __ mtlr(R3_RET);
1589 __ mr(R3_RET, Rexception);
1590 __ blr();
1591 }
1592 }
1593
1594 // JVMTI ForceEarlyReturn support.
1595 // Returns "in the middle" of a method with a "fake" return value.
1596 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1597
1598 Register Rscratch1 = R11_scratch1,
1599 Rscratch2 = R12_scratch2;
1600
1601 address entry = __ pc();
1602 __ empty_expression_stack();
1603
1604 __ load_earlyret_value(state, Rscratch1);
1605
1606 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
1607 // Clear the earlyret state.
1608 __ li(R0, 0);
1609 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
1610
1611 __ remove_activation(state, false, false);
1612 // Copied from TemplateTable::_return.
1613 // Restoration of lr done by remove_activation.
1614 switch (state) {
1615 case ltos:
1616 case btos:
1617 case ctos:
1618 case stos:
1619 case atos:
1620 case itos: __ mr(R3_RET, R17_tos); break;
1621 case ftos:
1622 case dtos: __ fmr(F1_RET, F15_ftos); break;
1623 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
1624 // to get visible before the reference to the object gets stored anywhere.
1625 __ membar(Assembler::StoreStore); break;
1626 default : ShouldNotReachHere();
1627 }
1628 __ blr();
1629
1630 return entry;
1631 } // end of ForceEarlyReturn support
1632
1633 //-----------------------------------------------------------------------------
1634 // Helper for vtos entry point generation
1635
1636 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1637 address& bep,
1638 address& cep,
1639 address& sep,
1640 address& aep,
1641 address& iep,
1642 address& lep,
1643 address& fep,
1644 address& dep,
1645 address& vep) {
1646 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1647 Label L;
1648
1649 aep = __ pc(); __ push_ptr(); __ b(L);
1650 fep = __ pc(); __ push_f(); __ b(L);
1651 dep = __ pc(); __ push_d(); __ b(L);
1652 lep = __ pc(); __ push_l(); __ b(L);
1653 __ align(32, 12, 24); // align L
1654 bep = cep = sep =
1655 iep = __ pc(); __ push_i();
1656 vep = __ pc();
1657 __ bind(L);
1658 generate_and_dispatch(t);
1659 }
1660
1661 //-----------------------------------------------------------------------------
1662 // Generation of individual instructions
1663
1664 // helpers for generate_and_dispatch
1665
1666 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1667 : TemplateInterpreterGenerator(code) {
1668 generate_all(); // Down here so it can be "virtual".
1669 }
1670
1671 //-----------------------------------------------------------------------------
1672
1673 // Non-product code
1674 #ifndef PRODUCT
1675 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1676 //__ flush_bundle();
1677 address entry = __ pc();
1678
1679 char *bname = NULL;
1680 uint tsize = 0;
1681 switch(state) {
1682 case ftos:
1683 bname = "trace_code_ftos {";
1684 tsize = 2;
1685 break;
1686 case btos:
1687 bname = "trace_code_btos {";
1688 tsize = 2;
1689 break;
1690 case ctos:
1691 bname = "trace_code_ctos {";
1692 tsize = 2;
1693 break;
1694 case stos:
1695 bname = "trace_code_stos {";
1696 tsize = 2;
1697 break;
1698 case itos:
1699 bname = "trace_code_itos {";
1700 tsize = 2;
1701 break;
1702 case ltos:
1703 bname = "trace_code_ltos {";
1704 tsize = 3;
1705 break;
1706 case atos:
1707 bname = "trace_code_atos {";
1708 tsize = 2;
1709 break;
1710 case vtos:
1711 // Note: In case of vtos, the topmost of stack value could be a int or doubl
1712 // In case of a double (2 slots) we won't see the 2nd stack value.
1713 // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
1714 bname = "trace_code_vtos {";
1715 tsize = 2;
1716
1717 break;
1718 case dtos:
1719 bname = "trace_code_dtos {";
1720 tsize = 3;
1721 break;
1722 default:
1723 ShouldNotReachHere();
1724 }
1725 BLOCK_COMMENT(bname);
1726
1727 // Support short-cut for TraceBytecodesAt.
1728 // Don't call into the VM if we don't want to trace to speed up things.
1729 Label Lskip_vm_call;
1730 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
1731 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
1732 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
1733 __ ld(R11_scratch1, offs1, R11_scratch1);
1734 __ lwa(R12_scratch2, offs2, R12_scratch2);
1735 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
1736 __ blt(CCR0, Lskip_vm_call);
1737 }
1738
1739 __ push(state);
1740 // Load 2 topmost expression stack values.
1741 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
1742 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
1743 __ mflr(R31);
1744 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
1745 __ mtlr(R31);
1746 __ pop(state);
1747
1748 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
1749 __ bind(Lskip_vm_call);
1750 }
1751 __ blr();
1752 BLOCK_COMMENT("} trace_code");
1753 return entry;
1754 }
1755
1756 void TemplateInterpreterGenerator::count_bytecode() {
1757 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
1758 __ lwz(R12_scratch2, offs, R11_scratch1);
1759 __ addi(R12_scratch2, R12_scratch2, 1);
1760 __ stw(R12_scratch2, offs, R11_scratch1);
1761 }
1762
1763 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1764 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
1765 __ lwz(R12_scratch2, offs, R11_scratch1);
1766 __ addi(R12_scratch2, R12_scratch2, 1);
1767 __ stw(R12_scratch2, offs, R11_scratch1);
1768 }
1769
1770 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1771 const Register addr = R11_scratch1,
1772 tmp = R12_scratch2;
1773 // Get index, shift out old bytecode, bring in new bytecode, and store it.
1774 // _index = (_index >> log2_number_of_codes) |
1775 // (bytecode << log2_number_of_codes);
1776 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
1777 __ lwz(tmp, offs1, addr);
1778 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
1779 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1780 __ stw(tmp, offs1, addr);
1781
1782 // Bump bucket contents.
1783 // _counters[_index] ++;
1784 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
1785 __ sldi(tmp, tmp, LogBytesPerInt);
1786 __ add(addr, tmp, addr);
1787 __ lwz(tmp, offs2, addr);
1788 __ addi(tmp, tmp, 1);
1789 __ stw(tmp, offs2, addr);
1790 }
1791
1792 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1793 // Call a little run-time stub to avoid blow-up for each bytecode.
1794 // The run-time runtime saves the right registers, depending on
1795 // the tosca in-state for the given template.
1796
1797 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1798 "entry must have been generated");
1799
1800 // Note: we destroy LR here.
1801 __ bl(Interpreter::trace_code(t->tos_in()));
1802 }
1803
1804 void TemplateInterpreterGenerator::stop_interpreter_at() {
1805 Label L;
1806 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
1807 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
1808 __ ld(R11_scratch1, offs1, R11_scratch1);
1809 __ lwa(R12_scratch2, offs2, R12_scratch2);
1810 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
1811 __ bne(CCR0, L);
1812 __ illtrap();
1813 __ bind(L);
1814 }
1815
1816 #endif // !PRODUCT
1817 #endif // !CC_INTERP