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 const Register cache = R11_scratch1;
176 const Register size = R12_scratch2;
177 __ get_cache_and_index_at_bcp(cache, 1, index_size);
178
179 // Big Endian (get least significant byte of 64 bit value):
180 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
181 __ sldi(size, size, Interpreter::logStackElementSize);
182 __ add(R15_esp, R15_esp, size);
183 __ dispatch_next(state, step);
184 return entry;
185 }
186
187 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
188 address entry = __ pc();
189 // If state != vtos, we're returning from a native method, which put it's result
190 // into the result register. So move the value out of the return register back
191 // to the TOS cache of current frame.
192
193 switch (state) {
194 case ltos:
195 case btos:
196 case ctos:
197 case stos:
198 case atos:
199 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache
200 case ftos:
201 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
202 case vtos: break; // Nothing to do, this was a void return.
203 default : ShouldNotReachHere();
204 }
205
206 // Load LcpoolCache @@@ should be already set!
207 __ get_constant_pool_cache(R27_constPoolCache);
208
209 // Handle a pending exception, fall through if none.
210 __ check_and_forward_exception(R11_scratch1, R12_scratch2);
211
212 // Start executing bytecodes.
213 __ dispatch_next(state, step);
214
215 return entry;
216 }
217
218 // A result handler converts the native result into java format.
219 // Use the shared code between c++ and template interpreter.
220 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
221 return AbstractInterpreterGenerator::generate_result_handler_for(type);
222 }
223
224 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
225 address entry = __ pc();
226
227 __ push(state);
228 __ call_VM(noreg, runtime_entry);
229 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
230
231 return entry;
232 }
233
234 // Helpers for commoning out cases in the various type of method entries.
235
236 // Increment invocation count & check for overflow.
237 //
238 // Note: checking for negative value instead of overflow
239 // so we have a 'sticky' overflow test.
240 //
241 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
242 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
243 Register Rscratch1 = R11_scratch1;
244 Register Rscratch2 = R12_scratch2;
245 Register R3_counters = R3_ARG1;
246 Label done;
247
248 if (TieredCompilation) {
249 const int increment = InvocationCounter::count_increment;
250 const int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
251 Label no_mdo;
252 if (ProfileInterpreter) {
253 const Register Rmdo = Rscratch1;
254 // If no method data exists, go to profile_continue.
255 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
256 __ cmpdi(CCR0, Rmdo, 0);
257 __ beq(CCR0, no_mdo);
258
259 // Increment backedge counter in the MDO.
260 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
261 __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
262 __ addi(Rscratch2, Rscratch2, increment);
263 __ stw(Rscratch2, mdo_bc_offs, Rmdo);
264 __ load_const_optimized(Rscratch1, mask, R0);
265 __ and_(Rscratch1, Rscratch2, Rscratch1);
266 __ bne(CCR0, done);
267 __ b(*overflow);
268 }
269
270 // Increment counter in MethodCounters*.
271 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
272 __ bind(no_mdo);
273 __ get_method_counters(R19_method, R3_counters, done);
274 __ lwz(Rscratch2, mo_bc_offs, R3_counters);
275 __ addi(Rscratch2, Rscratch2, increment);
276 __ stw(Rscratch2, mo_bc_offs, R3_counters);
277 __ load_const_optimized(Rscratch1, mask, R0);
278 __ and_(Rscratch1, Rscratch2, Rscratch1);
279 __ beq(CCR0, *overflow);
280
281 __ bind(done);
282
283 } else {
284
285 // Update standard invocation counters.
286 Register Rsum_ivc_bec = R4_ARG2;
287 __ get_method_counters(R19_method, R3_counters, done);
288 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2);
289 // Increment interpreter invocation counter.
290 if (ProfileInterpreter) { // %%% Merge this into methodDataOop.
291 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
292 __ addi(R12_scratch2, R12_scratch2, 1);
293 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
294 }
295 // Check if we must create a method data obj.
296 if (ProfileInterpreter && profile_method != NULL) {
297 const Register profile_limit = Rscratch1;
298 int pl_offs = __ load_const_optimized(profile_limit, &InvocationCounter::InterpreterProfileLimit, R0, true);
299 __ lwz(profile_limit, pl_offs, profile_limit);
300 // Test to see if we should create a method data oop.
301 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit);
302 __ blt(CCR0, *profile_method_continue);
303 // If no method data exists, go to profile_method.
304 __ test_method_data_pointer(*profile_method);
305 }
306 // Finally check for counter overflow.
307 if (overflow) {
308 const Register invocation_limit = Rscratch1;
309 int il_offs = __ load_const_optimized(invocation_limit, &InvocationCounter::InterpreterInvocationLimit, R0, true);
310 __ lwz(invocation_limit, il_offs, invocation_limit);
311 assert(4 == sizeof(InvocationCounter::InterpreterInvocationLimit), "unexpected field size");
312 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit);
313 __ bge(CCR0, *overflow);
314 }
315
316 __ bind(done);
317 }
318 }
319
320 // Generate code to initiate compilation on invocation counter overflow.
321 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
322 // Generate code to initiate compilation on the counter overflow.
323
324 // InterpreterRuntime::frequency_counter_overflow takes one arguments,
325 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp)
326 // We pass zero in.
327 // The call returns the address of the verified entry point for the method or NULL
328 // if the compilation did not complete (either went background or bailed out).
329 //
330 // Unlike the C++ interpreter above: Check exceptions!
331 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
332 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
333
334 __ li(R4_ARG2, 0);
335 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
336
337 // Returns verified_entry_point or NULL.
338 // We ignore it in any case.
339 __ b(continue_entry);
340 }
341
342 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
343 assert_different_registers(Rmem_frame_size, Rscratch1);
344 __ generate_stack_overflow_check_with_compare_and_throw(Rmem_frame_size, Rscratch1);
345 }
346
347 void TemplateInterpreterGenerator::unlock_method(bool check_exceptions) {
348 __ unlock_object(R26_monitor, check_exceptions);
349 }
350
351 // Lock the current method, interpreter register window must be set up!
352 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
353 const Register Robj_to_lock = Rscratch2;
354
355 {
356 if (!flags_preloaded) {
357 __ lwz(Rflags, method_(access_flags));
358 }
359
360 #ifdef ASSERT
361 // Check if methods needs synchronization.
362 {
363 Label Lok;
364 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
365 __ btrue(CCR0,Lok);
366 __ stop("method doesn't need synchronization");
367 __ bind(Lok);
368 }
369 #endif // ASSERT
370 }
371
372 // Get synchronization object to Rscratch2.
373 {
374 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
375 Label Lstatic;
376 Label Ldone;
377
378 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
379 __ btrue(CCR0, Lstatic);
380
381 // Non-static case: load receiver obj from stack and we're done.
382 __ ld(Robj_to_lock, R18_locals);
383 __ b(Ldone);
384
385 __ bind(Lstatic); // Static case: Lock the java mirror
386 __ ld(Robj_to_lock, in_bytes(Method::const_offset()), R19_method);
387 __ ld(Robj_to_lock, in_bytes(ConstMethod::constants_offset()), Robj_to_lock);
388 __ ld(Robj_to_lock, ConstantPool::pool_holder_offset_in_bytes(), Robj_to_lock);
389 __ ld(Robj_to_lock, mirror_offset, Robj_to_lock);
390
391 __ bind(Ldone);
392 __ verify_oop(Robj_to_lock);
393 }
394
395 // Got the oop to lock => execute!
396 __ add_monitor_to_stack(true, Rscratch1, R0);
397
398 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
399 __ lock_object(R26_monitor, Robj_to_lock);
400 }
401
402 // Generate a fixed interpreter frame for pure interpreter
403 // and I2N native transition frames.
404 //
405 // Before (stack grows downwards):
406 //
407 // | ... |
408 // |------------- |
409 // | java arg0 |
410 // | ... |
411 // | java argn |
412 // | | <- R15_esp
413 // | |
414 // |--------------|
415 // | abi_112 |
416 // | | <- R1_SP
417 // |==============|
418 //
419 //
420 // After:
421 //
422 // | ... |
423 // | java arg0 |<- R18_locals
424 // | ... |
425 // | java argn |
426 // |--------------|
427 // | |
428 // | java locals |
429 // | |
430 // |--------------|
431 // | abi_48 |
432 // |==============|
433 // | |
434 // | istate |
435 // | |
436 // |--------------|
437 // | monitor |<- R26_monitor
438 // |--------------|
439 // | |<- R15_esp
440 // | expression |
441 // | stack |
442 // | |
443 // |--------------|
444 // | |
445 // | abi_112 |<- R1_SP
446 // |==============|
447 //
448 // The top most frame needs an abi space of 112 bytes. This space is needed,
449 // since we call to c. The c function may spill their arguments to the caller
450 // frame. When we call to java, we don't need these spill slots. In order to save
451 // space on the stack, we resize the caller. However, java local reside in
452 // the caller frame and the frame has to be increased. The frame_size for the
453 // current frame was calculated based on max_stack as size for the expression
454 // stack. At the call, just a part of the expression stack might be used.
455 // We don't want to waste this space and cut the frame back accordingly.
456 // The resulting amount for resizing is calculated as follows:
457 // resize = (number_of_locals - number_of_arguments) * slot_size
458 // + (R1_SP - R15_esp) + 48
459 //
460 // The size for the callee frame is calculated:
461 // framesize = 112 + max_stack + monitor + state_size
462 //
463 // maxstack: Max number of slots on the expression stack, loaded from the method.
464 // monitor: We statically reserve room for one monitor object.
465 // state_size: We save the current state of the interpreter to this area.
466 //
467 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
468 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
469 top_frame_size = R7_ARG5,
470 Rconst_method = R8_ARG6;
471
472 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size);
473
474 __ ld(Rconst_method, method_(const));
475 __ lhz(Rsize_of_parameters /* number of params */,
476 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
477 if (native_call) {
478 // If we're calling a native method, we reserve space for the worst-case signature
479 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2).
480 // We add two slots to the parameter_count, one for the jni
481 // environment and one for a possible native mirror.
482 Label skip_native_calculate_max_stack;
483 __ addi(top_frame_size, Rsize_of_parameters, 2);
484 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters);
485 __ bge(CCR0, skip_native_calculate_max_stack);
486 __ li(top_frame_size, Argument::n_register_parameters);
487 __ bind(skip_native_calculate_max_stack);
488 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
489 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
490 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
491 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
492 } else {
493 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
494 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
495 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
496 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
497 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
498 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
499 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
500 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1);
501 }
502
503 // Compute top frame size.
504 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size);
505
506 // Cut back area between esp and max_stack.
507 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize);
508
509 __ round_to(top_frame_size, frame::alignment_in_bytes);
510 __ round_to(parent_frame_resize, frame::alignment_in_bytes);
511 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
512 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
513
514 {
515 // --------------------------------------------------------------------------
516 // Stack overflow check
517
518 Label cont;
519 __ add(R11_scratch1, parent_frame_resize, top_frame_size);
520 generate_stack_overflow_check(R11_scratch1, R12_scratch2);
521 }
522
523 // Set up interpreter state registers.
524
525 __ add(R18_locals, R15_esp, Rsize_of_parameters);
526 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
527 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache);
528
529 // Set method data pointer.
530 if (ProfileInterpreter) {
531 Label zero_continue;
532 __ ld(R28_mdx, method_(method_data));
533 __ cmpdi(CCR0, R28_mdx, 0);
534 __ beq(CCR0, zero_continue);
535 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
536 __ bind(zero_continue);
537 }
538
539 if (native_call) {
540 __ li(R14_bcp, 0); // Must initialize.
541 } else {
542 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method);
543 }
544
545 // Resize parent frame.
546 __ mflr(R12_scratch2);
547 __ neg(parent_frame_resize, parent_frame_resize);
548 __ resize_frame(parent_frame_resize, R11_scratch1);
549 __ std(R12_scratch2, _abi(lr), R1_SP);
550
551 __ addi(R26_monitor, R1_SP, - frame::ijava_state_size);
552 __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
553
554 // Store values.
555 // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls
556 // in InterpreterMacroAssembler::call_from_interpreter.
557 __ std(R19_method, _ijava_state_neg(method), R1_SP);
558 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
559 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
560 __ std(R18_locals, _ijava_state_neg(locals), R1_SP);
561
562 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
563 // be found in the frame after save_interpreter_state is done. This is always true
564 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
565 // because e.g. frame::interpreter_frame_bcp() will not access the correct value
566 // (Enhanced Stack Trace).
567 // The signal handler does not save the interpreter state into the frame.
568 __ li(R0, 0);
569 #ifdef ASSERT
570 // Fill remaining slots with constants.
571 __ load_const_optimized(R11_scratch1, 0x5afe);
572 __ load_const_optimized(R12_scratch2, 0xdead);
573 #endif
574 // We have to initialize some frame slots for native calls (accessed by GC).
575 if (native_call) {
576 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP);
577 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
578 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
579 }
580 #ifdef ASSERT
581 else {
582 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
583 __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP);
584 __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP);
585 }
586 __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP);
587 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
588 __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP);
589 __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP);
590 #endif
591 __ subf(R12_scratch2, top_frame_size, R1_SP);
592 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP);
593 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
594
595 // Push top frame.
596 __ push_frame(top_frame_size, R11_scratch1);
597 }
598
599 // End of helpers
600
601 // ============================================================================
602 // Various method entries
603 //
604
605 // Empty method, generate a very fast return. We must skip this entry if
606 // someone's debugging, indicated by the flag
607 // "interp_mode" in the Thread obj.
608 // Note: empty methods are generated mostly methods that do assertions, which are
609 // disabled in the "java opt build".
610 address TemplateInterpreterGenerator::generate_empty_entry(void) {
611 if (!UseFastEmptyMethods) {
612 NOT_PRODUCT(__ should_not_reach_here();)
613 return Interpreter::entry_for_kind(Interpreter::zerolocals);
614 }
615
616 Label Lslow_path;
617 const Register Rjvmti_mode = R11_scratch1;
618 address entry = __ pc();
619
620 __ lwz(Rjvmti_mode, thread_(interp_only_mode));
621 __ cmpwi(CCR0, Rjvmti_mode, 0);
622 __ bne(CCR0, Lslow_path); // jvmti_mode!=0
623
624 // Noone's debuggin: Simply return.
625 // Pop c2i arguments (if any) off when we return.
626 #ifdef ASSERT
627 __ ld(R9_ARG7, 0, R1_SP);
628 __ ld(R10_ARG8, 0, R21_sender_SP);
629 __ cmpd(CCR0, R9_ARG7, R10_ARG8);
630 __ asm_assert_eq("backlink", 0x545);
631 #endif // ASSERT
632 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
633
634 // And we're done.
635 __ blr();
636
637 __ bind(Lslow_path);
638 __ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
639 __ flush();
640
641 return entry;
642 }
643
644 // Support abs and sqrt like in compiler.
645 // For others we can use a normal (native) entry.
646
647 inline bool math_entry_available(AbstractInterpreter::MethodKind kind) {
648 // Provide math entry with debugging on demand.
649 // Note: Debugging changes which code will get executed:
650 // Debugging or disabled InlineIntrinsics: java method will get interpreted and performs a native call.
651 // Not debugging and enabled InlineIntrinics: processor instruction will get used.
652 // Result might differ slightly due to rounding etc.
653 if (!InlineIntrinsics && (!FLAG_IS_ERGO(InlineIntrinsics))) return false; // Generate a vanilla entry.
654
655 return ((kind==Interpreter::java_lang_math_sqrt && VM_Version::has_fsqrt()) ||
656 (kind==Interpreter::java_lang_math_abs));
657 }
658
659 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
660 if (!math_entry_available(kind)) {
661 NOT_PRODUCT(__ should_not_reach_here();)
662 return Interpreter::entry_for_kind(Interpreter::zerolocals);
663 }
664
665 Label Lslow_path;
666 const Register Rjvmti_mode = R11_scratch1;
667 address entry = __ pc();
668
669 // Provide math entry with debugging on demand.
670 __ lwz(Rjvmti_mode, thread_(interp_only_mode));
671 __ cmpwi(CCR0, Rjvmti_mode, 0);
672 __ bne(CCR0, Lslow_path); // jvmti_mode!=0
673
674 __ lfd(F1_RET, Interpreter::stackElementSize, R15_esp);
675
676 // Pop c2i arguments (if any) off when we return.
677 #ifdef ASSERT
678 __ ld(R9_ARG7, 0, R1_SP);
679 __ ld(R10_ARG8, 0, R21_sender_SP);
680 __ cmpd(CCR0, R9_ARG7, R10_ARG8);
681 __ asm_assert_eq("backlink", 0x545);
682 #endif // ASSERT
683 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
684
685 if (kind == Interpreter::java_lang_math_sqrt) {
686 __ fsqrt(F1_RET, F1_RET);
687 } else if (kind == Interpreter::java_lang_math_abs) {
688 __ fabs(F1_RET, F1_RET);
689 } else {
690 ShouldNotReachHere();
691 }
692
693 // And we're done.
694 __ blr();
695
696 // Provide slow path for JVMTI case.
697 __ bind(Lslow_path);
698 __ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R12_scratch2);
699 __ flush();
700
701 return entry;
702 }
703
704 // Interpreter stub for calling a native method. (asm interpreter)
705 // This sets up a somewhat different looking stack for calling the
706 // native method than the typical interpreter frame setup.
707 //
708 // On entry:
709 // R19_method - method
710 // R16_thread - JavaThread*
711 // R15_esp - intptr_t* sender tos
712 //
713 // abstract stack (grows up)
714 // [ IJava (caller of JNI callee) ] <-- ASP
715 // ...
716 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
717
718 address entry = __ pc();
719
720 const bool inc_counter = UseCompiler || CountCompiledCalls;
721
722 // -----------------------------------------------------------------------------
723 // Allocate a new frame that represents the native callee (i2n frame).
724 // This is not a full-blown interpreter frame, but in particular, the
725 // following registers are valid after this:
726 // - R19_method
727 // - R18_local (points to start of argumuments to native function)
728 //
729 // abstract stack (grows up)
730 // [ IJava (caller of JNI callee) ] <-- ASP
731 // ...
732
733 const Register signature_handler_fd = R11_scratch1;
734 const Register pending_exception = R0;
735 const Register result_handler_addr = R31;
736 const Register native_method_fd = R11_scratch1;
737 const Register access_flags = R22_tmp2;
738 const Register active_handles = R11_scratch1; // R26_monitor saved to state.
739 const Register sync_state = R12_scratch2;
740 const Register sync_state_addr = sync_state; // Address is dead after use.
741 const Register suspend_flags = R11_scratch1;
742
743 //=============================================================================
744 // Allocate new frame and initialize interpreter state.
745
746 Label exception_return;
747 Label exception_return_sync_check;
748 Label stack_overflow_return;
749
750 // Generate new interpreter state and jump to stack_overflow_return in case of
751 // a stack overflow.
752 //generate_compute_interpreter_state(stack_overflow_return);
753
754 Register size_of_parameters = R22_tmp2;
755
756 generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
757
758 //=============================================================================
759 // Increment invocation counter. On overflow, entry to JNI method
760 // will be compiled.
761 Label invocation_counter_overflow, continue_after_compile;
762 if (inc_counter) {
763 if (synchronized) {
764 // Since at this point in the method invocation the exception handler
765 // would try to exit the monitor of synchronized methods which hasn't
766 // been entered yet, we set the thread local variable
767 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
768 // runtime, exception handling i.e. unlock_if_synchronized_method will
769 // check this thread local flag.
770 // This flag has two effects, one is to force an unwind in the topmost
771 // interpreter frame and not perform an unlock while doing so.
772 __ li(R0, 1);
773 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
774 }
775 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
776
777 __ BIND(continue_after_compile);
778 // Reset the _do_not_unlock_if_synchronized flag.
779 if (synchronized) {
780 __ li(R0, 0);
781 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
782 }
783 }
784
785 // access_flags = method->access_flags();
786 // Load access flags.
787 assert(access_flags->is_nonvolatile(),
788 "access_flags must be in a non-volatile register");
789 // Type check.
790 assert(4 == sizeof(AccessFlags), "unexpected field size");
791 __ lwz(access_flags, method_(access_flags));
792
793 // We don't want to reload R19_method and access_flags after calls
794 // to some helper functions.
795 assert(R19_method->is_nonvolatile(),
796 "R19_method must be a non-volatile register");
797
798 // Check for synchronized methods. Must happen AFTER invocation counter
799 // check, so method is not locked if counter overflows.
800
801 if (synchronized) {
802 lock_method(access_flags, R11_scratch1, R12_scratch2, true);
803
804 // Update monitor in state.
805 __ ld(R11_scratch1, 0, R1_SP);
806 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1);
807 }
808
809 // jvmti/jvmpi support
810 __ notify_method_entry();
811
812 //=============================================================================
813 // Get and call the signature handler.
814
815 __ ld(signature_handler_fd, method_(signature_handler));
816 Label call_signature_handler;
817
818 __ cmpdi(CCR0, signature_handler_fd, 0);
819 __ bne(CCR0, call_signature_handler);
820
821 // Method has never been called. Either generate a specialized
822 // handler or point to the slow one.
823 //
824 // Pass parameter 'false' to avoid exception check in call_VM.
825 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
826
827 // Check for an exception while looking up the target method. If we
828 // incurred one, bail.
829 __ ld(pending_exception, thread_(pending_exception));
830 __ cmpdi(CCR0, pending_exception, 0);
831 __ bne(CCR0, exception_return_sync_check); // Has pending exception.
832
833 // Reload signature handler, it may have been created/assigned in the meanwhile.
834 __ ld(signature_handler_fd, method_(signature_handler));
835 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
836
837 __ BIND(call_signature_handler);
838
839 // Before we call the signature handler we push a new frame to
840 // protect the interpreter frame volatile registers when we return
841 // from jni but before we can get back to Java.
842
843 // First set the frame anchor while the SP/FP registers are
844 // convenient and the slow signature handler can use this same frame
845 // anchor.
846
847 // We have a TOP_IJAVA_FRAME here, which belongs to us.
848 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
849
850 // Now the interpreter frame (and its call chain) have been
851 // invalidated and flushed. We are now protected against eager
852 // being enabled in native code. Even if it goes eager the
853 // registers will be reloaded as clean and we will invalidate after
854 // the call so no spurious flush should be possible.
855
856 // Call signature handler and pass locals address.
857 //
858 // Our signature handlers copy required arguments to the C stack
859 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
860 __ mr(R3_ARG1, R18_locals);
861 __ ld(signature_handler_fd, 0, signature_handler_fd);
862
863 __ call_stub(signature_handler_fd);
864
865 // Remove the register parameter varargs slots we allocated in
866 // compute_interpreter_state. SP+16 ends up pointing to the ABI
867 // outgoing argument area.
868 //
869 // Not needed on PPC64.
870 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord);
871
872 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
873 // Save across call to native method.
874 __ mr(result_handler_addr, R3_RET);
875
876 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
877
878 // Set up fixed parameters and call the native method.
879 // If the method is static, get mirror into R4_ARG2.
880 {
881 Label method_is_not_static;
882 // Access_flags is non-volatile and still, no need to restore it.
883
884 // Restore access flags.
885 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
886 __ bfalse(CCR0, method_is_not_static);
887
888 // constants = method->constants();
889 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
890 __ ld(R11_scratch1, in_bytes(ConstMethod::constants_offset()), R11_scratch1);
891 // pool_holder = method->constants()->pool_holder();
892 __ ld(R11_scratch1/*pool_holder*/, ConstantPool::pool_holder_offset_in_bytes(),
893 R11_scratch1/*constants*/);
894
895 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
896
897 // mirror = pool_holder->klass_part()->java_mirror();
898 __ ld(R0/*mirror*/, mirror_offset, R11_scratch1/*pool_holder*/);
899 // state->_native_mirror = mirror;
900
901 __ ld(R11_scratch1, 0, R1_SP);
902 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
903 // R4_ARG2 = &state->_oop_temp;
904 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
905 __ BIND(method_is_not_static);
906 }
907
908 // At this point, arguments have been copied off the stack into
909 // their JNI positions. Oops are boxed in-place on the stack, with
910 // handles copied to arguments. The result handler address is in a
911 // register.
912
913 // Pass JNIEnv address as first parameter.
914 __ addir(R3_ARG1, thread_(jni_environment));
915
916 // Load the native_method entry before we change the thread state.
917 __ ld(native_method_fd, method_(native_function));
918
919 //=============================================================================
920 // Transition from _thread_in_Java to _thread_in_native. As soon as
921 // we make this change the safepoint code needs to be certain that
922 // the last Java frame we established is good. The pc in that frame
923 // just needs to be near here not an actual return address.
924
925 // We use release_store_fence to update values like the thread state, where
926 // we don't want the current thread to continue until all our prior memory
927 // accesses (including the new thread state) are visible to other threads.
928 __ li(R0, _thread_in_native);
929 __ release();
930
931 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
932 __ stw(R0, thread_(thread_state));
933
934 if (UseMembar) {
935 __ fence();
936 }
937
938 //=============================================================================
939 // Call the native method. Argument registers must not have been
940 // overwritten since "__ call_stub(signature_handler);" (except for
941 // ARG1 and ARG2 for static methods).
942 __ call_c(native_method_fd);
943
944 __ li(R0, 0);
945 __ ld(R11_scratch1, 0, R1_SP);
946 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
947 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
948 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
949
950 // Note: C++ interpreter needs the following here:
951 // The frame_manager_lr field, which we use for setting the last
952 // java frame, gets overwritten by the signature handler. Restore
953 // it now.
954 //__ get_PC_trash_LR(R11_scratch1);
955 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
956
957 // Because of GC R19_method may no longer be valid.
958
959 // Block, if necessary, before resuming in _thread_in_Java state.
960 // In order for GC to work, don't clear the last_Java_sp until after
961 // blocking.
962
963 //=============================================================================
964 // Switch thread to "native transition" state before reading the
965 // synchronization state. This additional state is necessary
966 // because reading and testing the synchronization state is not
967 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
968 // in _thread_in_native state, loads _not_synchronized and is
969 // preempted. VM thread changes sync state to synchronizing and
970 // suspends threads for GC. Thread A is resumed to finish this
971 // native method, but doesn't block here since it didn't see any
972 // synchronization in progress, and escapes.
973
974 // We use release_store_fence to update values like the thread state, where
975 // we don't want the current thread to continue until all our prior memory
976 // accesses (including the new thread state) are visible to other threads.
977 __ li(R0/*thread_state*/, _thread_in_native_trans);
978 __ release();
979 __ stw(R0/*thread_state*/, thread_(thread_state));
980 if (UseMembar) {
981 __ fence();
982 }
983 // Write serialization page so that the VM thread can do a pseudo remote
984 // membar. We use the current thread pointer to calculate a thread
985 // specific offset to write to within the page. This minimizes bus
986 // traffic due to cache line collision.
987 else {
988 __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2);
989 }
990
991 // Now before we return to java we must look for a current safepoint
992 // (a new safepoint can not start since we entered native_trans).
993 // We must check here because a current safepoint could be modifying
994 // the callers registers right this moment.
995
996 // Acquire isn't strictly necessary here because of the fence, but
997 // sync_state is declared to be volatile, so we do it anyway
998 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
999 int sync_state_offs = __ load_const_optimized(sync_state_addr, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
1000
1001 // TODO PPC port assert(4 == SafepointSynchronize::sz_state(), "unexpected field size");
1002 __ lwz(sync_state, sync_state_offs, sync_state_addr);
1003
1004 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1005 __ lwz(suspend_flags, thread_(suspend_flags));
1006
1007 Label sync_check_done;
1008 Label do_safepoint;
1009 // No synchronization in progress nor yet synchronized.
1010 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
1011 // Not suspended.
1012 __ cmpwi(CCR1, suspend_flags, 0);
1013
1014 __ bne(CCR0, do_safepoint);
1015 __ beq(CCR1, sync_check_done);
1016 __ bind(do_safepoint);
1017 __ isync();
1018 // Block. We do the call directly and leave the current
1019 // last_Java_frame setup undisturbed. We must save any possible
1020 // native result across the call. No oop is present.
1021
1022 __ mr(R3_ARG1, R16_thread);
1023 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
1024 relocInfo::none);
1025
1026 __ bind(sync_check_done);
1027
1028 //=============================================================================
1029 // <<<<<< Back in Interpreter Frame >>>>>
1030
1031 // We are in thread_in_native_trans here and back in the normal
1032 // interpreter frame. We don't have to do anything special about
1033 // safepoints and we can switch to Java mode anytime we are ready.
1034
1035 // Note: frame::interpreter_frame_result has a dependency on how the
1036 // method result is saved across the call to post_method_exit. For
1037 // native methods it assumes that the non-FPU/non-void result is
1038 // saved in _native_lresult and a FPU result in _native_fresult. If
1039 // this changes then the interpreter_frame_result implementation
1040 // will need to be updated too.
1041
1042 // On PPC64, we have stored the result directly after the native call.
1043
1044 //=============================================================================
1045 // Back in Java
1046
1047 // We use release_store_fence to update values like the thread state, where
1048 // we don't want the current thread to continue until all our prior memory
1049 // accesses (including the new thread state) are visible to other threads.
1050 __ li(R0/*thread_state*/, _thread_in_Java);
1051 __ release();
1052 __ stw(R0/*thread_state*/, thread_(thread_state));
1053 if (UseMembar) {
1054 __ fence();
1055 }
1056
1057 __ reset_last_Java_frame();
1058
1059 // Jvmdi/jvmpi support. Whether we've got an exception pending or
1060 // not, and whether unlocking throws an exception or not, we notify
1061 // on native method exit. If we do have an exception, we'll end up
1062 // in the caller's context to handle it, so if we don't do the
1063 // notify here, we'll drop it on the floor.
1064 __ notify_method_exit(true/*native method*/,
1065 ilgl /*illegal state (not used for native methods)*/,
1066 InterpreterMacroAssembler::NotifyJVMTI,
1067 false /*check_exceptions*/);
1068
1069 //=============================================================================
1070 // Handle exceptions
1071
1072 if (synchronized) {
1073 // Don't check for exceptions since we're still in the i2n frame. Do that
1074 // manually afterwards.
1075 unlock_method(false);
1076 }
1077
1078 // Reset active handles after returning from native.
1079 // thread->active_handles()->clear();
1080 __ ld(active_handles, thread_(active_handles));
1081 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1082 __ li(R0, 0);
1083 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles);
1084
1085 Label exception_return_sync_check_already_unlocked;
1086 __ ld(R0/*pending_exception*/, thread_(pending_exception));
1087 __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1088 __ bne(CCR0, exception_return_sync_check_already_unlocked);
1089
1090 //-----------------------------------------------------------------------------
1091 // No exception pending.
1092
1093 // Move native method result back into proper registers and return.
1094 // Invoke result handler (may unbox/promote).
1095 __ ld(R11_scratch1, 0, R1_SP);
1096 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1097 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1098 __ call_stub(result_handler_addr);
1099
1100 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1101
1102 // Must use the return pc which was loaded from the caller's frame
1103 // as the VM uses return-pc-patching for deoptimization.
1104 __ mtlr(R0);
1105 __ blr();
1106
1107 //-----------------------------------------------------------------------------
1108 // An exception is pending. We call into the runtime only if the
1109 // caller was not interpreted. If it was interpreted the
1110 // interpreter will do the correct thing. If it isn't interpreted
1111 // (call stub/compiled code) we will change our return and continue.
1112
1113 __ BIND(exception_return_sync_check);
1114
1115 if (synchronized) {
1116 // Don't check for exceptions since we're still in the i2n frame. Do that
1117 // manually afterwards.
1118 unlock_method(false);
1119 }
1120 __ BIND(exception_return_sync_check_already_unlocked);
1121
1122 const Register return_pc = R31;
1123
1124 __ ld(return_pc, 0, R1_SP);
1125 __ ld(return_pc, _abi(lr), return_pc);
1126
1127 // Get the address of the exception handler.
1128 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1129 R16_thread,
1130 return_pc /* return pc */);
1131 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1132
1133 // Load the PC of the the exception handler into LR.
1134 __ mtlr(R3_RET);
1135
1136 // Load exception into R3_ARG1 and clear pending exception in thread.
1137 __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1138 __ li(R4_ARG2, 0);
1139 __ std(R4_ARG2, thread_(pending_exception));
1140
1141 // Load the original return pc into R4_ARG2.
1142 __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1143
1144 // Return to exception handler.
1145 __ blr();
1146
1147 //=============================================================================
1148 // Counter overflow.
1149
1150 if (inc_counter) {
1151 // Handle invocation counter overflow.
1152 __ bind(invocation_counter_overflow);
1153
1154 generate_counter_overflow(continue_after_compile);
1155 }
1156
1157 return entry;
1158 }
1159
1160 // Generic interpreted method entry to (asm) interpreter.
1161 //
1162 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1163 bool inc_counter = UseCompiler || CountCompiledCalls;
1164 address entry = __ pc();
1165 // Generate the code to allocate the interpreter stack frame.
1166 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1167 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame.
1168
1169 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1170
1171 #ifdef FAST_DISPATCH
1172 __ unimplemented("Fast dispatch in generate_normal_entry");
1173 #if 0
1174 __ set((intptr_t)Interpreter::dispatch_table(), IdispatchTables);
1175 // Set bytecode dispatch table base.
1176 #endif
1177 #endif
1178
1179 // --------------------------------------------------------------------------
1180 // Zero out non-parameter locals.
1181 // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1182 // worth to ask the flag, just do it.
1183 Register Rslot_addr = R6_ARG4,
1184 Rnum = R7_ARG5;
1185 Label Lno_locals, Lzero_loop;
1186
1187 // Set up the zeroing loop.
1188 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1189 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1190 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1191 __ beq(CCR0, Lno_locals);
1192 __ li(R0, 0);
1193 __ mtctr(Rnum);
1194
1195 // The zero locals loop.
1196 __ bind(Lzero_loop);
1197 __ std(R0, 0, Rslot_addr);
1198 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1199 __ bdnz(Lzero_loop);
1200
1201 __ bind(Lno_locals);
1202
1203 // --------------------------------------------------------------------------
1204 // Counter increment and overflow check.
1205 Label invocation_counter_overflow,
1206 profile_method,
1207 profile_method_continue;
1208 if (inc_counter || ProfileInterpreter) {
1209
1210 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1211 if (synchronized) {
1212 // Since at this point in the method invocation the exception handler
1213 // would try to exit the monitor of synchronized methods which hasn't
1214 // been entered yet, we set the thread local variable
1215 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1216 // runtime, exception handling i.e. unlock_if_synchronized_method will
1217 // check this thread local flag.
1218 // This flag has two effects, one is to force an unwind in the topmost
1219 // interpreter frame and not perform an unlock while doing so.
1220 __ li(R0, 1);
1221 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1222 }
1223 // Increment invocation counter and check for overflow.
1224 if (inc_counter) {
1225 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1226 }
1227
1228 __ bind(profile_method_continue);
1229
1230 // Reset the _do_not_unlock_if_synchronized flag.
1231 if (synchronized) {
1232 __ li(R0, 0);
1233 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1234 }
1235 }
1236
1237 // --------------------------------------------------------------------------
1238 // Locking of synchronized methods. Must happen AFTER invocation_counter
1239 // check and stack overflow check, so method is not locked if overflows.
1240 if (synchronized) {
1241 lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1242 }
1243 #ifdef ASSERT
1244 else {
1245 Label Lok;
1246 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1247 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1248 __ asm_assert_eq("method needs synchronization", 0x8521);
1249 __ bind(Lok);
1250 }
1251 #endif // ASSERT
1252
1253 __ verify_thread();
1254
1255 // --------------------------------------------------------------------------
1256 // JVMTI support
1257 __ notify_method_entry();
1258
1259 // --------------------------------------------------------------------------
1260 // Start executing instructions.
1261 __ dispatch_next(vtos);
1262
1263 // --------------------------------------------------------------------------
1264 // Out of line counter overflow and MDO creation code.
1265 if (ProfileInterpreter) {
1266 // We have decided to profile this method in the interpreter.
1267 __ bind(profile_method);
1268 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1269 __ set_method_data_pointer_for_bcp();
1270 __ b(profile_method_continue);
1271 }
1272
1273 if (inc_counter) {
1274 // Handle invocation counter overflow.
1275 __ bind(invocation_counter_overflow);
1276 generate_counter_overflow(profile_method_continue);
1277 }
1278 return entry;
1279 }
1280
1281 // =============================================================================
1282 // Entry points
1283
1284 address AbstractInterpreterGenerator::generate_method_entry(
1285 AbstractInterpreter::MethodKind kind) {
1286 // Determine code generation flags.
1287 bool synchronized = false;
1288 address entry_point = NULL;
1289
1290 switch (kind) {
1291 case Interpreter::zerolocals : break;
1292 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1293 case Interpreter::native : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1294 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true); break;
1295 case Interpreter::empty : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry(); break;
1296 case Interpreter::accessor : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry(); break;
1297 case Interpreter::abstract : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry(); break;
1298
1299 case Interpreter::java_lang_math_sin : // fall thru
1300 case Interpreter::java_lang_math_cos : // fall thru
1301 case Interpreter::java_lang_math_tan : // fall thru
1302 case Interpreter::java_lang_math_abs : // fall thru
1303 case Interpreter::java_lang_math_log : // fall thru
1304 case Interpreter::java_lang_math_log10 : // fall thru
1305 case Interpreter::java_lang_math_sqrt : // fall thru
1306 case Interpreter::java_lang_math_pow : // fall thru
1307 case Interpreter::java_lang_math_exp : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind); break;
1308 case Interpreter::java_lang_ref_reference_get
1309 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1310 default : ShouldNotReachHere(); break;
1311 }
1312
1313 if (entry_point) {
1314 return entry_point;
1315 }
1316
1317 return ((InterpreterGenerator*) this)->generate_normal_entry(synchronized);
1318 }
1319
1320 // These should never be compiled since the interpreter will prefer
1321 // the compiled version to the intrinsic version.
1322 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1323 return !math_entry_available(method_kind(m));
1324 }
1325
1326 // How much stack a method activation needs in stack slots.
1327 // We must calc this exactly like in generate_fixed_frame.
1328 // Note: This returns the conservative size assuming maximum alignment.
1329 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1330 const int max_alignment_size = 2;
1331 const int abi_scratch = frame::abi_reg_args_size;
1332 return method->max_locals() + method->max_stack() +
1333 frame::interpreter_frame_monitor_size() + max_alignment_size + abi_scratch;
1334 }
1335
1336 // Returns number of stackElementWords needed for the interpreter frame with the
1337 // given sections.
1338 // This overestimates the stack by one slot in case of alignments.
1339 int AbstractInterpreter::size_activation(int max_stack,
1340 int temps,
1341 int extra_args,
1342 int monitors,
1343 int callee_params,
1344 int callee_locals,
1345 bool is_top_frame) {
1346 // Note: This calculation must exactly parallel the frame setup
1347 // in AbstractInterpreterGenerator::generate_method_entry.
1348 assert(Interpreter::stackElementWords == 1, "sanity");
1349 const int max_alignment_space = StackAlignmentInBytes / Interpreter::stackElementSize;
1350 const int abi_scratch = is_top_frame ? (frame::abi_reg_args_size / Interpreter::stackElementSize) :
1351 (frame::abi_minframe_size / Interpreter::stackElementSize);
1352 const int size =
1353 max_stack +
1354 (callee_locals - callee_params) +
1355 monitors * frame::interpreter_frame_monitor_size() +
1356 max_alignment_space +
1357 abi_scratch +
1358 frame::ijava_state_size / Interpreter::stackElementSize;
1359
1360 // Fixed size of an interpreter frame, align to 16-byte.
1361 return (size & -2);
1362 }
1363
1364 // Fills a sceletal interpreter frame generated during deoptimizations.
1365 //
1366 // Parameters:
1367 //
1368 // interpreter_frame != NULL:
1369 // set up the method, locals, and monitors.
1370 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1371 // right size, as determined by a previous call to this method.
1372 // It is also guaranteed to be walkable even though it is in a skeletal state
1373 //
1374 // is_top_frame == true:
1375 // We're processing the *oldest* interpreter frame!
1376 //
1377 // pop_frame_extra_args:
1378 // If this is != 0 we are returning to a deoptimized frame by popping
1379 // off the callee frame. We want to re-execute the call that called the
1380 // callee interpreted, but since the return to the interpreter would pop
1381 // the arguments off advance the esp by dummy popframe_extra_args slots.
1382 // Popping off those will establish the stack layout as it was before the call.
1383 //
1384 void AbstractInterpreter::layout_activation(Method* method,
1385 int tempcount,
1386 int popframe_extra_args,
1387 int moncount,
1388 int caller_actual_parameters,
1389 int callee_param_count,
1390 int callee_locals_count,
1391 frame* caller,
1392 frame* interpreter_frame,
1393 bool is_top_frame,
1394 bool is_bottom_frame) {
1395
1396 const int abi_scratch = is_top_frame ? (frame::abi_reg_args_size / Interpreter::stackElementSize) :
1397 (frame::abi_minframe_size / Interpreter::stackElementSize);
1398
1399 intptr_t* locals_base = (caller->is_interpreted_frame()) ?
1400 caller->interpreter_frame_esp() + caller_actual_parameters :
1401 caller->sp() + method->max_locals() - 1 + (frame::abi_minframe_size / Interpreter::stackElementSize) ;
1402
1403 intptr_t* monitor_base = caller->sp() - frame::ijava_state_size / Interpreter::stackElementSize ;
1404 intptr_t* monitor = monitor_base - (moncount * frame::interpreter_frame_monitor_size());
1405 intptr_t* esp_base = monitor - 1;
1406 intptr_t* esp = esp_base - tempcount - popframe_extra_args;
1407 intptr_t* sp = (intptr_t *) (((intptr_t) (esp_base - callee_locals_count + callee_param_count - method->max_stack()- abi_scratch)) & -StackAlignmentInBytes);
1408 intptr_t* sender_sp = caller->sp() + (frame::abi_minframe_size - frame::abi_reg_args_size) / Interpreter::stackElementSize;
1409 intptr_t* top_frame_sp = is_top_frame ? sp : sp + (frame::abi_minframe_size - frame::abi_reg_args_size) / Interpreter::stackElementSize;
1410
1411 interpreter_frame->interpreter_frame_set_method(method);
1412 interpreter_frame->interpreter_frame_set_locals(locals_base);
1413 interpreter_frame->interpreter_frame_set_cpcache(method->constants()->cache());
1414 interpreter_frame->interpreter_frame_set_esp(esp);
1415 interpreter_frame->interpreter_frame_set_monitor_end((BasicObjectLock *)monitor);
1416 interpreter_frame->interpreter_frame_set_top_frame_sp(top_frame_sp);
1417 if (!is_bottom_frame) {
1418 interpreter_frame->interpreter_frame_set_sender_sp(sender_sp);
1419 }
1420 }
1421
1422 // =============================================================================
1423 // Exceptions
1424
1425 void TemplateInterpreterGenerator::generate_throw_exception() {
1426 Register Rexception = R17_tos,
1427 Rcontinuation = R3_RET;
1428
1429 // --------------------------------------------------------------------------
1430 // Entry point if an method returns with a pending exception (rethrow).
1431 Interpreter::_rethrow_exception_entry = __ pc();
1432 {
1433 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
1434 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
1435 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
1436
1437 // Compiled code destroys templateTableBase, reload.
1438 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
1439 }
1440
1441 // Entry point if a interpreted method throws an exception (throw).
1442 Interpreter::_throw_exception_entry = __ pc();
1443 {
1444 __ mr(Rexception, R3_RET);
1445
1446 __ verify_thread();
1447 __ verify_oop(Rexception);
1448
1449 // Expression stack must be empty before entering the VM in case of an exception.
1450 __ empty_expression_stack();
1451 // Find exception handler address and preserve exception oop.
1452 // Call C routine to find handler and jump to it.
1453 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
1454 __ mtctr(Rcontinuation);
1455 // Push exception for exception handler bytecodes.
1456 __ push_ptr(Rexception);
1457
1458 // Jump to exception handler (may be remove activation entry!).
1459 __ bctr();
1460 }
1461
1462 // If the exception is not handled in the current frame the frame is
1463 // removed and the exception is rethrown (i.e. exception
1464 // continuation is _rethrow_exception).
1465 //
1466 // Note: At this point the bci is still the bxi for the instruction
1467 // which caused the exception and the expression stack is
1468 // empty. Thus, for any VM calls at this point, GC will find a legal
1469 // oop map (with empty expression stack).
1470
1471 // In current activation
1472 // tos: exception
1473 // bcp: exception bcp
1474
1475 // --------------------------------------------------------------------------
1476 // JVMTI PopFrame support
1477
1478 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1479 {
1480 // Set the popframe_processing bit in popframe_condition indicating that we are
1481 // currently handling popframe, so that call_VMs that may happen later do not
1482 // trigger new popframe handling cycles.
1483 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1484 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
1485 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1486
1487 // Empty the expression stack, as in normal exception handling.
1488 __ empty_expression_stack();
1489 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
1490
1491 // Check to see whether we are returning to a deoptimized frame.
1492 // (The PopFrame call ensures that the caller of the popped frame is
1493 // either interpreted or compiled and deoptimizes it if compiled.)
1494 // Note that we don't compare the return PC against the
1495 // deoptimization blob's unpack entry because of the presence of
1496 // adapter frames in C2.
1497 Label Lcaller_not_deoptimized;
1498 Register return_pc = R3_ARG1;
1499 __ ld(return_pc, 0, R1_SP);
1500 __ ld(return_pc, _abi(lr), return_pc);
1501 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
1502 __ cmpdi(CCR0, R3_RET, 0);
1503 __ bne(CCR0, Lcaller_not_deoptimized);
1504
1505 // The deoptimized case.
1506 // In this case, we can't call dispatch_next() after the frame is
1507 // popped, but instead must save the incoming arguments and restore
1508 // them after deoptimization has occurred.
1509 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
1510 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
1511 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
1512 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
1513 __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
1514 // Save these arguments.
1515 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
1516
1517 // Inform deoptimization that it is responsible for restoring these arguments.
1518 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1519 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1520
1521 // Return from the current method into the deoptimization blob. Will eventually
1522 // end up in the deopt interpeter entry, deoptimization prepared everything that
1523 // we will reexecute the call that called us.
1524 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
1525 __ mtlr(return_pc);
1526 __ blr();
1527
1528 // The non-deoptimized case.
1529 __ bind(Lcaller_not_deoptimized);
1530
1531 // Clear the popframe condition flag.
1532 __ li(R0, 0);
1533 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1534
1535 // Get out of the current method and re-execute the call that called us.
1536 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
1537 __ restore_interpreter_state(R11_scratch1);
1538 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
1539 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
1540 if (ProfileInterpreter) {
1541 __ set_method_data_pointer_for_bcp();
1542 }
1543 #if INCLUDE_JVMTI
1544 Label L_done;
1545
1546 __ lbz(R11_scratch1, 0, R14_bcp);
1547 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
1548 __ bne(CCR0, L_done);
1549
1550 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1551 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1552 __ ld(R4_ARG2, 0, R18_locals);
1553 __ call_VM(R11_scratch1, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null),
1554 R4_ARG2, R19_method, R14_bcp);
1555
1556 __ cmpdi(CCR0, R11_scratch1, 0);
1557 __ beq(CCR0, L_done);
1558
1559 __ std(R11_scratch1, wordSize, R15_esp);
1560 __ bind(L_done);
1561 #endif // INCLUDE_JVMTI
1562 __ dispatch_next(vtos);
1563 }
1564 // end of JVMTI PopFrame support
1565
1566 // --------------------------------------------------------------------------
1567 // Remove activation exception entry.
1568 // This is jumped to if an interpreted method can't handle an exception itself
1569 // (we come from the throw/rethrow exception entry above). We're going to call
1570 // into the VM to find the exception handler in the caller, pop the current
1571 // frame and return the handler we calculated.
1572 Interpreter::_remove_activation_entry = __ pc();
1573 {
1574 __ pop_ptr(Rexception);
1575 __ verify_thread();
1576 __ verify_oop(Rexception);
1577 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
1578
1579 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
1580 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
1581
1582 __ get_vm_result(Rexception);
1583
1584 // We are done with this activation frame; find out where to go next.
1585 // The continuation point will be an exception handler, which expects
1586 // the following registers set up:
1587 //
1588 // RET: exception oop
1589 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
1590
1591 Register return_pc = R31; // Needs to survive the runtime call.
1592 __ ld(return_pc, 0, R1_SP);
1593 __ ld(return_pc, _abi(lr), return_pc);
1594 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
1595
1596 // Remove the current activation.
1597 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
1598
1599 __ mr(R4_ARG2, return_pc);
1600 __ mtlr(R3_RET);
1601 __ mr(R3_RET, Rexception);
1602 __ blr();
1603 }
1604 }
1605
1606 // JVMTI ForceEarlyReturn support.
1607 // Returns "in the middle" of a method with a "fake" return value.
1608 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1609
1610 Register Rscratch1 = R11_scratch1,
1611 Rscratch2 = R12_scratch2;
1612
1613 address entry = __ pc();
1614 __ empty_expression_stack();
1615
1616 __ load_earlyret_value(state, Rscratch1);
1617
1618 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
1619 // Clear the earlyret state.
1620 __ li(R0, 0);
1621 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
1622
1623 __ remove_activation(state, false, false);
1624 // Copied from TemplateTable::_return.
1625 // Restoration of lr done by remove_activation.
1626 switch (state) {
1627 case ltos:
1628 case btos:
1629 case ctos:
1630 case stos:
1631 case atos:
1632 case itos: __ mr(R3_RET, R17_tos); break;
1633 case ftos:
1634 case dtos: __ fmr(F1_RET, F15_ftos); break;
1635 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
1636 // to get visible before the reference to the object gets stored anywhere.
1637 __ membar(Assembler::StoreStore); break;
1638 default : ShouldNotReachHere();
1639 }
1640 __ blr();
1641
1642 return entry;
1643 } // end of ForceEarlyReturn support
1644
1645 //-----------------------------------------------------------------------------
1646 // Helper for vtos entry point generation
1647
1648 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1649 address& bep,
1650 address& cep,
1651 address& sep,
1652 address& aep,
1653 address& iep,
1654 address& lep,
1655 address& fep,
1656 address& dep,
1657 address& vep) {
1658 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1659 Label L;
1660
1661 aep = __ pc(); __ push_ptr(); __ b(L);
1662 fep = __ pc(); __ push_f(); __ b(L);
1663 dep = __ pc(); __ push_d(); __ b(L);
1664 lep = __ pc(); __ push_l(); __ b(L);
1665 __ align(32, 12, 24); // align L
1666 bep = cep = sep =
1667 iep = __ pc(); __ push_i();
1668 vep = __ pc();
1669 __ bind(L);
1670 generate_and_dispatch(t);
1671 }
1672
1673 //-----------------------------------------------------------------------------
1674 // Generation of individual instructions
1675
1676 // helpers for generate_and_dispatch
1677
1678 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1679 : TemplateInterpreterGenerator(code) {
1680 generate_all(); // Down here so it can be "virtual".
1681 }
1682
1683 //-----------------------------------------------------------------------------
1684
1685 // Non-product code
1686 #ifndef PRODUCT
1687 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1688 //__ flush_bundle();
1689 address entry = __ pc();
1690
1691 const char *bname = NULL;
1692 uint tsize = 0;
1693 switch(state) {
1694 case ftos:
1695 bname = "trace_code_ftos {";
1696 tsize = 2;
1697 break;
1698 case btos:
1699 bname = "trace_code_btos {";
1700 tsize = 2;
1701 break;
1702 case ctos:
1703 bname = "trace_code_ctos {";
1704 tsize = 2;
1705 break;
1706 case stos:
1707 bname = "trace_code_stos {";
1708 tsize = 2;
1709 break;
1710 case itos:
1711 bname = "trace_code_itos {";
1712 tsize = 2;
1713 break;
1714 case ltos:
1715 bname = "trace_code_ltos {";
1716 tsize = 3;
1717 break;
1718 case atos:
1719 bname = "trace_code_atos {";
1720 tsize = 2;
1721 break;
1722 case vtos:
1723 // Note: In case of vtos, the topmost of stack value could be a int or doubl
1724 // In case of a double (2 slots) we won't see the 2nd stack value.
1725 // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
1726 bname = "trace_code_vtos {";
1727 tsize = 2;
1728
1729 break;
1730 case dtos:
1731 bname = "trace_code_dtos {";
1732 tsize = 3;
1733 break;
1734 default:
1735 ShouldNotReachHere();
1736 }
1737 BLOCK_COMMENT(bname);
1738
1739 // Support short-cut for TraceBytecodesAt.
1740 // Don't call into the VM if we don't want to trace to speed up things.
1741 Label Lskip_vm_call;
1742 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
1743 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
1744 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
1745 __ ld(R11_scratch1, offs1, R11_scratch1);
1746 __ lwa(R12_scratch2, offs2, R12_scratch2);
1747 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
1748 __ blt(CCR0, Lskip_vm_call);
1749 }
1750
1751 __ push(state);
1752 // Load 2 topmost expression stack values.
1753 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
1754 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
1755 __ mflr(R31);
1756 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
1757 __ mtlr(R31);
1758 __ pop(state);
1759
1760 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
1761 __ bind(Lskip_vm_call);
1762 }
1763 __ blr();
1764 BLOCK_COMMENT("} trace_code");
1765 return entry;
1766 }
1767
1768 void TemplateInterpreterGenerator::count_bytecode() {
1769 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
1770 __ lwz(R12_scratch2, offs, R11_scratch1);
1771 __ addi(R12_scratch2, R12_scratch2, 1);
1772 __ stw(R12_scratch2, offs, R11_scratch1);
1773 }
1774
1775 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1776 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
1777 __ lwz(R12_scratch2, offs, R11_scratch1);
1778 __ addi(R12_scratch2, R12_scratch2, 1);
1779 __ stw(R12_scratch2, offs, R11_scratch1);
1780 }
1781
1782 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1783 const Register addr = R11_scratch1,
1784 tmp = R12_scratch2;
1785 // Get index, shift out old bytecode, bring in new bytecode, and store it.
1786 // _index = (_index >> log2_number_of_codes) |
1787 // (bytecode << log2_number_of_codes);
1788 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
1789 __ lwz(tmp, offs1, addr);
1790 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
1791 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1792 __ stw(tmp, offs1, addr);
1793
1794 // Bump bucket contents.
1795 // _counters[_index] ++;
1796 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
1797 __ sldi(tmp, tmp, LogBytesPerInt);
1798 __ add(addr, tmp, addr);
1799 __ lwz(tmp, offs2, addr);
1800 __ addi(tmp, tmp, 1);
1801 __ stw(tmp, offs2, addr);
1802 }
1803
1804 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1805 // Call a little run-time stub to avoid blow-up for each bytecode.
1806 // The run-time runtime saves the right registers, depending on
1807 // the tosca in-state for the given template.
1808
1809 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1810 "entry must have been generated");
1811
1812 // Note: we destroy LR here.
1813 __ bl(Interpreter::trace_code(t->tos_in()));
1814 }
1815
1816 void TemplateInterpreterGenerator::stop_interpreter_at() {
1817 Label L;
1818 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
1819 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
1820 __ ld(R11_scratch1, offs1, R11_scratch1);
1821 __ lwa(R12_scratch2, offs2, R12_scratch2);
1822 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
1823 __ bne(CCR0, L);
1824 __ illtrap();
1825 __ bind(L);
1826 }
1827
1828 #endif // !PRODUCT
1829 #endif // !CC_INTERP