1 /*
2 * Copyright (c) 2014, 2015, Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2013, 2015 SAP AG. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #ifndef CC_INTERP
28 #include "asm/macroAssembler.inline.hpp"
29 #include "interpreter/bytecodeHistogram.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "interpreter/interpreterGenerator.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "interpreter/interp_masm.hpp"
34 #include "interpreter/templateTable.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/methodData.hpp"
37 #include "oops/method.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "prims/jvmtiExport.hpp"
40 #include "prims/jvmtiThreadState.hpp"
41 #include "runtime/arguments.hpp"
42 #include "runtime/deoptimization.hpp"
43 #include "runtime/frame.inline.hpp"
44 #include "runtime/sharedRuntime.hpp"
45 #include "runtime/stubRoutines.hpp"
46 #include "runtime/synchronizer.hpp"
47 #include "runtime/timer.hpp"
48 #include "runtime/vframeArray.hpp"
49 #include "utilities/debug.hpp"
50 #include "utilities/macros.hpp"
51
52 #undef __
53 #define __ _masm->
54
55 #ifdef PRODUCT
56 #define BLOCK_COMMENT(str) /* nothing */
57 #else
58 #define BLOCK_COMMENT(str) __ block_comment(str)
59 #endif
60
61 #define BIND(label) __ bind(label); BLOCK_COMMENT(#label ":")
62
63 //-----------------------------------------------------------------------------
64
65 // Actually we should never reach here since we do stack overflow checks before pushing any frame.
66 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
67 address entry = __ pc();
68 __ unimplemented("generate_StackOverflowError_handler");
69 return entry;
70 }
71
72 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
73 address entry = __ pc();
74 __ empty_expression_stack();
75 __ load_const_optimized(R4_ARG2, (address) name);
76 // Index is in R17_tos.
77 __ mr(R5_ARG3, R17_tos);
78 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException));
79 return entry;
80 }
81
82 #if 0
83 // Call special ClassCastException constructor taking object to cast
84 // and target class as arguments.
85 address TemplateInterpreterGenerator::generate_ClassCastException_verbose_handler() {
86 address entry = __ pc();
87
88 // Expression stack must be empty before entering the VM if an
89 // exception happened.
90 __ empty_expression_stack();
91
92 // Thread will be loaded to R3_ARG1.
93 // Target class oop is in register R5_ARG3 by convention!
94 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose), R17_tos, R5_ARG3);
95 // Above call must not return here since exception pending.
96 DEBUG_ONLY(__ should_not_reach_here();)
97 return entry;
98 }
99 #endif
100
101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
102 address entry = __ pc();
103 // Expression stack must be empty before entering the VM if an
104 // exception happened.
105 __ empty_expression_stack();
106
107 // Load exception object.
108 // Thread will be loaded to R3_ARG1.
109 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
110 #ifdef ASSERT
111 // Above call must not return here since exception pending.
112 __ should_not_reach_here();
113 #endif
114 return entry;
115 }
116
117 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
118 address entry = __ pc();
119 //__ untested("generate_exception_handler_common");
120 Register Rexception = R17_tos;
121
122 // Expression stack must be empty before entering the VM if an exception happened.
123 __ empty_expression_stack();
124
125 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
126 if (pass_oop) {
127 __ mr(R5_ARG3, Rexception);
128 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false);
129 } else {
130 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
131 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false);
132 }
133
134 // Throw exception.
135 __ mr(R3_ARG1, Rexception);
136 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
137 __ mtctr(R11_scratch1);
138 __ bctr();
139
140 return entry;
141 }
142
143 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
144 address entry = __ pc();
145 __ unimplemented("generate_continuation_for");
146 return entry;
147 }
148
149 // This entry is returned to when a call returns to the interpreter.
150 // When we arrive here, we expect that the callee stack frame is already popped.
151 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
152 address entry = __ pc();
153
154 // Move the value out of the return register back to the TOS cache of current frame.
155 switch (state) {
156 case ltos:
157 case btos:
158 case ctos:
159 case stos:
160 case atos:
161 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache
162 case ftos:
163 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
164 case vtos: break; // Nothing to do, this was a void return.
165 default : ShouldNotReachHere();
166 }
167
168 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
169 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
170 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
171
172 // Compiled code destroys templateTableBase, reload.
173 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
174
175 if (state == atos) {
176 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
177 }
178
179 const Register cache = R11_scratch1;
180 const Register size = R12_scratch2;
181 __ get_cache_and_index_at_bcp(cache, 1, index_size);
182
183 // Get least significant byte of 64 bit value:
184 #if defined(VM_LITTLE_ENDIAN)
185 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache);
186 #else
187 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
188 #endif
189 __ sldi(size, size, Interpreter::logStackElementSize);
190 __ add(R15_esp, R15_esp, size);
191 __ dispatch_next(state, step);
192 return entry;
193 }
194
195 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
196 address entry = __ pc();
197 // If state != vtos, we're returning from a native method, which put it's result
198 // into the result register. So move the value out of the return register back
199 // to the TOS cache of current frame.
200
201 switch (state) {
202 case ltos:
203 case btos:
204 case ctos:
205 case stos:
206 case atos:
207 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache
208 case ftos:
209 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
210 case vtos: break; // Nothing to do, this was a void return.
211 default : ShouldNotReachHere();
212 }
213
214 // Load LcpoolCache @@@ should be already set!
215 __ get_constant_pool_cache(R27_constPoolCache);
216
217 // Handle a pending exception, fall through if none.
218 __ check_and_forward_exception(R11_scratch1, R12_scratch2);
219
220 // Start executing bytecodes.
221 __ dispatch_next(state, step);
222
223 return entry;
224 }
225
226 // A result handler converts the native result into java format.
227 // Use the shared code between c++ and template interpreter.
228 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
229 return AbstractInterpreterGenerator::generate_result_handler_for(type);
230 }
231
232 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
233 address entry = __ pc();
234
235 __ push(state);
236 __ call_VM(noreg, runtime_entry);
237 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
238
239 return entry;
240 }
241
242 // Helpers for commoning out cases in the various type of method entries.
243
244 // Increment invocation count & check for overflow.
245 //
246 // Note: checking for negative value instead of overflow
247 // so we have a 'sticky' overflow test.
248 //
249 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
250 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
251 Register Rscratch1 = R11_scratch1;
252 Register Rscratch2 = R12_scratch2;
253 Register R3_counters = R3_ARG1;
254 Label done;
255
256 if (TieredCompilation) {
257 const int increment = InvocationCounter::count_increment;
258 const int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
259 Label no_mdo;
260 if (ProfileInterpreter) {
261 const Register Rmdo = Rscratch1;
262 // If no method data exists, go to profile_continue.
263 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
264 __ cmpdi(CCR0, Rmdo, 0);
265 __ beq(CCR0, no_mdo);
266
267 // Increment backedge counter in the MDO.
268 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
269 __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
270 __ addi(Rscratch2, Rscratch2, increment);
271 __ stw(Rscratch2, mdo_bc_offs, Rmdo);
272 __ load_const_optimized(Rscratch1, mask, R0);
273 __ and_(Rscratch1, Rscratch2, Rscratch1);
274 __ bne(CCR0, done);
275 __ b(*overflow);
276 }
277
278 // Increment counter in MethodCounters*.
279 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
280 __ bind(no_mdo);
281 __ get_method_counters(R19_method, R3_counters, done);
282 __ lwz(Rscratch2, mo_bc_offs, R3_counters);
283 __ addi(Rscratch2, Rscratch2, increment);
284 __ stw(Rscratch2, mo_bc_offs, R3_counters);
285 __ load_const_optimized(Rscratch1, mask, R0);
286 __ and_(Rscratch1, Rscratch2, Rscratch1);
287 __ beq(CCR0, *overflow);
288
289 __ bind(done);
290
291 } else {
292
293 // Update standard invocation counters.
294 Register Rsum_ivc_bec = R4_ARG2;
295 __ get_method_counters(R19_method, R3_counters, done);
296 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2);
297 // Increment interpreter invocation counter.
298 if (ProfileInterpreter) { // %%% Merge this into methodDataOop.
299 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
300 __ addi(R12_scratch2, R12_scratch2, 1);
301 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
302 }
303 // Check if we must create a method data obj.
304 if (ProfileInterpreter && profile_method != NULL) {
305 const Register profile_limit = Rscratch1;
306 int pl_offs = __ load_const_optimized(profile_limit, &InvocationCounter::InterpreterProfileLimit, R0, true);
307 __ lwz(profile_limit, pl_offs, profile_limit);
308 // Test to see if we should create a method data oop.
309 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit);
310 __ blt(CCR0, *profile_method_continue);
311 // If no method data exists, go to profile_method.
312 __ test_method_data_pointer(*profile_method);
313 }
314 // Finally check for counter overflow.
315 if (overflow) {
316 const Register invocation_limit = Rscratch1;
317 int il_offs = __ load_const_optimized(invocation_limit, &InvocationCounter::InterpreterInvocationLimit, R0, true);
318 __ lwz(invocation_limit, il_offs, invocation_limit);
319 assert(4 == sizeof(InvocationCounter::InterpreterInvocationLimit), "unexpected field size");
320 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit);
321 __ bge(CCR0, *overflow);
322 }
323
324 __ bind(done);
325 }
326 }
327
328 // Generate code to initiate compilation on invocation counter overflow.
329 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
330 // Generate code to initiate compilation on the counter overflow.
331
332 // InterpreterRuntime::frequency_counter_overflow takes one arguments,
333 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp)
334 // We pass zero in.
335 // The call returns the address of the verified entry point for the method or NULL
336 // if the compilation did not complete (either went background or bailed out).
337 //
338 // Unlike the C++ interpreter above: Check exceptions!
339 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
340 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
341
342 __ li(R4_ARG2, 0);
343 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
344
345 // Returns verified_entry_point or NULL.
346 // We ignore it in any case.
347 __ b(continue_entry);
348 }
349
350 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
351 assert_different_registers(Rmem_frame_size, Rscratch1);
352 __ generate_stack_overflow_check_with_compare_and_throw(Rmem_frame_size, Rscratch1);
353 }
354
355 void TemplateInterpreterGenerator::unlock_method(bool check_exceptions) {
356 __ unlock_object(R26_monitor, check_exceptions);
357 }
358
359 // Lock the current method, interpreter register window must be set up!
360 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
361 const Register Robj_to_lock = Rscratch2;
362
363 {
364 if (!flags_preloaded) {
365 __ lwz(Rflags, method_(access_flags));
366 }
367
368 #ifdef ASSERT
369 // Check if methods needs synchronization.
370 {
371 Label Lok;
372 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
373 __ btrue(CCR0,Lok);
374 __ stop("method doesn't need synchronization");
375 __ bind(Lok);
376 }
377 #endif // ASSERT
378 }
379
380 // Get synchronization object to Rscratch2.
381 {
382 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
383 Label Lstatic;
384 Label Ldone;
385
386 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
387 __ btrue(CCR0, Lstatic);
388
389 // Non-static case: load receiver obj from stack and we're done.
390 __ ld(Robj_to_lock, R18_locals);
391 __ b(Ldone);
392
393 __ bind(Lstatic); // Static case: Lock the java mirror
394 __ ld(Robj_to_lock, in_bytes(Method::const_offset()), R19_method);
395 __ ld(Robj_to_lock, in_bytes(ConstMethod::constants_offset()), Robj_to_lock);
396 __ ld(Robj_to_lock, ConstantPool::pool_holder_offset_in_bytes(), Robj_to_lock);
397 __ ld(Robj_to_lock, mirror_offset, Robj_to_lock);
398
399 __ bind(Ldone);
400 __ verify_oop(Robj_to_lock);
401 }
402
403 // Got the oop to lock => execute!
404 __ add_monitor_to_stack(true, Rscratch1, R0);
405
406 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
407 __ lock_object(R26_monitor, Robj_to_lock);
408 }
409
410 // Generate a fixed interpreter frame for pure interpreter
411 // and I2N native transition frames.
412 //
413 // Before (stack grows downwards):
414 //
415 // | ... |
416 // |------------- |
417 // | java arg0 |
418 // | ... |
419 // | java argn |
420 // | | <- R15_esp
421 // | |
422 // |--------------|
423 // | abi_112 |
424 // | | <- R1_SP
425 // |==============|
426 //
427 //
428 // After:
429 //
430 // | ... |
431 // | java arg0 |<- R18_locals
432 // | ... |
433 // | java argn |
434 // |--------------|
435 // | |
436 // | java locals |
437 // | |
438 // |--------------|
439 // | abi_48 |
440 // |==============|
441 // | |
442 // | istate |
443 // | |
444 // |--------------|
445 // | monitor |<- R26_monitor
446 // |--------------|
447 // | |<- R15_esp
448 // | expression |
449 // | stack |
450 // | |
451 // |--------------|
452 // | |
453 // | abi_112 |<- R1_SP
454 // |==============|
455 //
456 // The top most frame needs an abi space of 112 bytes. This space is needed,
457 // since we call to c. The c function may spill their arguments to the caller
458 // frame. When we call to java, we don't need these spill slots. In order to save
459 // space on the stack, we resize the caller. However, java local reside in
460 // the caller frame and the frame has to be increased. The frame_size for the
461 // current frame was calculated based on max_stack as size for the expression
462 // stack. At the call, just a part of the expression stack might be used.
463 // We don't want to waste this space and cut the frame back accordingly.
464 // The resulting amount for resizing is calculated as follows:
465 // resize = (number_of_locals - number_of_arguments) * slot_size
466 // + (R1_SP - R15_esp) + 48
467 //
468 // The size for the callee frame is calculated:
469 // framesize = 112 + max_stack + monitor + state_size
470 //
471 // maxstack: Max number of slots on the expression stack, loaded from the method.
472 // monitor: We statically reserve room for one monitor object.
473 // state_size: We save the current state of the interpreter to this area.
474 //
475 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
476 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
477 top_frame_size = R7_ARG5,
478 Rconst_method = R8_ARG6;
479
480 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size);
481
482 __ ld(Rconst_method, method_(const));
483 __ lhz(Rsize_of_parameters /* number of params */,
484 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
485 if (native_call) {
486 // If we're calling a native method, we reserve space for the worst-case signature
487 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2).
488 // We add two slots to the parameter_count, one for the jni
489 // environment and one for a possible native mirror.
490 Label skip_native_calculate_max_stack;
491 __ addi(top_frame_size, Rsize_of_parameters, 2);
492 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters);
493 __ bge(CCR0, skip_native_calculate_max_stack);
494 __ li(top_frame_size, Argument::n_register_parameters);
495 __ bind(skip_native_calculate_max_stack);
496 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
497 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
498 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
499 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
500 } else {
501 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
502 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
503 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
504 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
505 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
506 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
507 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
508 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1);
509 }
510
511 // Compute top frame size.
512 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size);
513
514 // Cut back area between esp and max_stack.
515 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize);
516
517 __ round_to(top_frame_size, frame::alignment_in_bytes);
518 __ round_to(parent_frame_resize, frame::alignment_in_bytes);
519 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
520 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
521
522 {
523 // --------------------------------------------------------------------------
524 // Stack overflow check
525
526 Label cont;
527 __ add(R11_scratch1, parent_frame_resize, top_frame_size);
528 generate_stack_overflow_check(R11_scratch1, R12_scratch2);
529 }
530
531 // Set up interpreter state registers.
532
533 __ add(R18_locals, R15_esp, Rsize_of_parameters);
534 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
535 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache);
536
537 // Set method data pointer.
538 if (ProfileInterpreter) {
539 Label zero_continue;
540 __ ld(R28_mdx, method_(method_data));
541 __ cmpdi(CCR0, R28_mdx, 0);
542 __ beq(CCR0, zero_continue);
543 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
544 __ bind(zero_continue);
545 }
546
547 if (native_call) {
548 __ li(R14_bcp, 0); // Must initialize.
549 } else {
550 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method);
551 }
552
553 // Resize parent frame.
554 __ mflr(R12_scratch2);
555 __ neg(parent_frame_resize, parent_frame_resize);
556 __ resize_frame(parent_frame_resize, R11_scratch1);
557 __ std(R12_scratch2, _abi(lr), R1_SP);
558
559 __ addi(R26_monitor, R1_SP, - frame::ijava_state_size);
560 __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
561
562 // Store values.
563 // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls
564 // in InterpreterMacroAssembler::call_from_interpreter.
565 __ std(R19_method, _ijava_state_neg(method), R1_SP);
566 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
567 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
568 __ std(R18_locals, _ijava_state_neg(locals), R1_SP);
569
570 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
571 // be found in the frame after save_interpreter_state is done. This is always true
572 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
573 // because e.g. frame::interpreter_frame_bcp() will not access the correct value
574 // (Enhanced Stack Trace).
575 // The signal handler does not save the interpreter state into the frame.
576 __ li(R0, 0);
577 #ifdef ASSERT
578 // Fill remaining slots with constants.
579 __ load_const_optimized(R11_scratch1, 0x5afe);
580 __ load_const_optimized(R12_scratch2, 0xdead);
581 #endif
582 // We have to initialize some frame slots for native calls (accessed by GC).
583 if (native_call) {
584 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP);
585 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
586 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
587 }
588 #ifdef ASSERT
589 else {
590 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
591 __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP);
592 __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP);
593 }
594 __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP);
595 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
596 __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP);
597 __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP);
598 #endif
599 __ subf(R12_scratch2, top_frame_size, R1_SP);
600 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP);
601 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
602
603 // Push top frame.
604 __ push_frame(top_frame_size, R11_scratch1);
605 }
606
607 // End of helpers
608
609
610 // Support abs and sqrt like in compiler.
611 // For others we can use a normal (native) entry.
612
613 inline bool math_entry_available(AbstractInterpreter::MethodKind kind) {
614 if (!InlineIntrinsics) return false;
615
616 return ((kind==Interpreter::java_lang_math_sqrt && VM_Version::has_fsqrt()) ||
617 (kind==Interpreter::java_lang_math_abs));
618 }
619
620 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
621 if (!math_entry_available(kind)) {
622 NOT_PRODUCT(__ should_not_reach_here();)
623 return Interpreter::entry_for_kind(Interpreter::zerolocals);
624 }
625
626 address entry = __ pc();
627
628 __ lfd(F1_RET, Interpreter::stackElementSize, R15_esp);
629
630 // Pop c2i arguments (if any) off when we return.
631 #ifdef ASSERT
632 __ ld(R9_ARG7, 0, R1_SP);
633 __ ld(R10_ARG8, 0, R21_sender_SP);
634 __ cmpd(CCR0, R9_ARG7, R10_ARG8);
635 __ asm_assert_eq("backlink", 0x545);
636 #endif // ASSERT
637 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
638
639 if (kind == Interpreter::java_lang_math_sqrt) {
640 __ fsqrt(F1_RET, F1_RET);
641 } else if (kind == Interpreter::java_lang_math_abs) {
642 __ fabs(F1_RET, F1_RET);
643 } else {
644 ShouldNotReachHere();
645 }
646
647 // And we're done.
648 __ blr();
649
650 __ flush();
651
652 return entry;
653 }
654
655 // Interpreter stub for calling a native method. (asm interpreter)
656 // This sets up a somewhat different looking stack for calling the
657 // native method than the typical interpreter frame setup.
658 //
659 // On entry:
660 // R19_method - method
661 // R16_thread - JavaThread*
662 // R15_esp - intptr_t* sender tos
663 //
664 // abstract stack (grows up)
665 // [ IJava (caller of JNI callee) ] <-- ASP
666 // ...
667 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
668
669 address entry = __ pc();
670
671 const bool inc_counter = UseCompiler || CountCompiledCalls;
672
673 // -----------------------------------------------------------------------------
674 // Allocate a new frame that represents the native callee (i2n frame).
675 // This is not a full-blown interpreter frame, but in particular, the
676 // following registers are valid after this:
677 // - R19_method
678 // - R18_local (points to start of argumuments to native function)
679 //
680 // abstract stack (grows up)
681 // [ IJava (caller of JNI callee) ] <-- ASP
682 // ...
683
684 const Register signature_handler_fd = R11_scratch1;
685 const Register pending_exception = R0;
686 const Register result_handler_addr = R31;
687 const Register native_method_fd = R11_scratch1;
688 const Register access_flags = R22_tmp2;
689 const Register active_handles = R11_scratch1; // R26_monitor saved to state.
690 const Register sync_state = R12_scratch2;
691 const Register sync_state_addr = sync_state; // Address is dead after use.
692 const Register suspend_flags = R11_scratch1;
693
694 //=============================================================================
695 // Allocate new frame and initialize interpreter state.
696
697 Label exception_return;
698 Label exception_return_sync_check;
699 Label stack_overflow_return;
700
701 // Generate new interpreter state and jump to stack_overflow_return in case of
702 // a stack overflow.
703 //generate_compute_interpreter_state(stack_overflow_return);
704
705 Register size_of_parameters = R22_tmp2;
706
707 generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
708
709 //=============================================================================
710 // Increment invocation counter. On overflow, entry to JNI method
711 // will be compiled.
712 Label invocation_counter_overflow, continue_after_compile;
713 if (inc_counter) {
714 if (synchronized) {
715 // Since at this point in the method invocation the exception handler
716 // would try to exit the monitor of synchronized methods which hasn't
717 // been entered yet, we set the thread local variable
718 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
719 // runtime, exception handling i.e. unlock_if_synchronized_method will
720 // check this thread local flag.
721 // This flag has two effects, one is to force an unwind in the topmost
722 // interpreter frame and not perform an unlock while doing so.
723 __ li(R0, 1);
724 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
725 }
726 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
727
728 BIND(continue_after_compile);
729 // Reset the _do_not_unlock_if_synchronized flag.
730 if (synchronized) {
731 __ li(R0, 0);
732 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
733 }
734 }
735
736 // access_flags = method->access_flags();
737 // Load access flags.
738 assert(access_flags->is_nonvolatile(),
739 "access_flags must be in a non-volatile register");
740 // Type check.
741 assert(4 == sizeof(AccessFlags), "unexpected field size");
742 __ lwz(access_flags, method_(access_flags));
743
744 // We don't want to reload R19_method and access_flags after calls
745 // to some helper functions.
746 assert(R19_method->is_nonvolatile(),
747 "R19_method must be a non-volatile register");
748
749 // Check for synchronized methods. Must happen AFTER invocation counter
750 // check, so method is not locked if counter overflows.
751
752 if (synchronized) {
753 lock_method(access_flags, R11_scratch1, R12_scratch2, true);
754
755 // Update monitor in state.
756 __ ld(R11_scratch1, 0, R1_SP);
757 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1);
758 }
759
760 // jvmti/jvmpi support
761 __ notify_method_entry();
762
763 //=============================================================================
764 // Get and call the signature handler.
765
766 __ ld(signature_handler_fd, method_(signature_handler));
767 Label call_signature_handler;
768
769 __ cmpdi(CCR0, signature_handler_fd, 0);
770 __ bne(CCR0, call_signature_handler);
771
772 // Method has never been called. Either generate a specialized
773 // handler or point to the slow one.
774 //
775 // Pass parameter 'false' to avoid exception check in call_VM.
776 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
777
778 // Check for an exception while looking up the target method. If we
779 // incurred one, bail.
780 __ ld(pending_exception, thread_(pending_exception));
781 __ cmpdi(CCR0, pending_exception, 0);
782 __ bne(CCR0, exception_return_sync_check); // Has pending exception.
783
784 // Reload signature handler, it may have been created/assigned in the meanwhile.
785 __ ld(signature_handler_fd, method_(signature_handler));
786 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
787
788 BIND(call_signature_handler);
789
790 // Before we call the signature handler we push a new frame to
791 // protect the interpreter frame volatile registers when we return
792 // from jni but before we can get back to Java.
793
794 // First set the frame anchor while the SP/FP registers are
795 // convenient and the slow signature handler can use this same frame
796 // anchor.
797
798 // We have a TOP_IJAVA_FRAME here, which belongs to us.
799 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
800
801 // Now the interpreter frame (and its call chain) have been
802 // invalidated and flushed. We are now protected against eager
803 // being enabled in native code. Even if it goes eager the
804 // registers will be reloaded as clean and we will invalidate after
805 // the call so no spurious flush should be possible.
806
807 // Call signature handler and pass locals address.
808 //
809 // Our signature handlers copy required arguments to the C stack
810 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
811 __ mr(R3_ARG1, R18_locals);
812 #if !defined(ABI_ELFv2)
813 __ ld(signature_handler_fd, 0, signature_handler_fd);
814 #endif
815
816 __ call_stub(signature_handler_fd);
817
818 // Remove the register parameter varargs slots we allocated in
819 // compute_interpreter_state. SP+16 ends up pointing to the ABI
820 // outgoing argument area.
821 //
822 // Not needed on PPC64.
823 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord);
824
825 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
826 // Save across call to native method.
827 __ mr(result_handler_addr, R3_RET);
828
829 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
830
831 // Set up fixed parameters and call the native method.
832 // If the method is static, get mirror into R4_ARG2.
833 {
834 Label method_is_not_static;
835 // Access_flags is non-volatile and still, no need to restore it.
836
837 // Restore access flags.
838 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
839 __ bfalse(CCR0, method_is_not_static);
840
841 // constants = method->constants();
842 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
843 __ ld(R11_scratch1, in_bytes(ConstMethod::constants_offset()), R11_scratch1);
844 // pool_holder = method->constants()->pool_holder();
845 __ ld(R11_scratch1/*pool_holder*/, ConstantPool::pool_holder_offset_in_bytes(),
846 R11_scratch1/*constants*/);
847
848 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
849
850 // mirror = pool_holder->klass_part()->java_mirror();
851 __ ld(R0/*mirror*/, mirror_offset, R11_scratch1/*pool_holder*/);
852 // state->_native_mirror = mirror;
853
854 __ ld(R11_scratch1, 0, R1_SP);
855 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
856 // R4_ARG2 = &state->_oop_temp;
857 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
858 BIND(method_is_not_static);
859 }
860
861 // At this point, arguments have been copied off the stack into
862 // their JNI positions. Oops are boxed in-place on the stack, with
863 // handles copied to arguments. The result handler address is in a
864 // register.
865
866 // Pass JNIEnv address as first parameter.
867 __ addir(R3_ARG1, thread_(jni_environment));
868
869 // Load the native_method entry before we change the thread state.
870 __ ld(native_method_fd, method_(native_function));
871
872 //=============================================================================
873 // Transition from _thread_in_Java to _thread_in_native. As soon as
874 // we make this change the safepoint code needs to be certain that
875 // the last Java frame we established is good. The pc in that frame
876 // just needs to be near here not an actual return address.
877
878 // We use release_store_fence to update values like the thread state, where
879 // we don't want the current thread to continue until all our prior memory
880 // accesses (including the new thread state) are visible to other threads.
881 __ li(R0, _thread_in_native);
882 __ release();
883
884 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
885 __ stw(R0, thread_(thread_state));
886
887 if (UseMembar) {
888 __ fence();
889 }
890
891 //=============================================================================
892 // Call the native method. Argument registers must not have been
893 // overwritten since "__ call_stub(signature_handler);" (except for
894 // ARG1 and ARG2 for static methods).
895 __ call_c(native_method_fd);
896
897 __ li(R0, 0);
898 __ ld(R11_scratch1, 0, R1_SP);
899 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
900 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
901 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
902
903 // Note: C++ interpreter needs the following here:
904 // The frame_manager_lr field, which we use for setting the last
905 // java frame, gets overwritten by the signature handler. Restore
906 // it now.
907 //__ get_PC_trash_LR(R11_scratch1);
908 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
909
910 // Because of GC R19_method may no longer be valid.
911
912 // Block, if necessary, before resuming in _thread_in_Java state.
913 // In order for GC to work, don't clear the last_Java_sp until after
914 // blocking.
915
916 //=============================================================================
917 // Switch thread to "native transition" state before reading the
918 // synchronization state. This additional state is necessary
919 // because reading and testing the synchronization state is not
920 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
921 // in _thread_in_native state, loads _not_synchronized and is
922 // preempted. VM thread changes sync state to synchronizing and
923 // suspends threads for GC. Thread A is resumed to finish this
924 // native method, but doesn't block here since it didn't see any
925 // synchronization in progress, and escapes.
926
927 // We use release_store_fence to update values like the thread state, where
928 // we don't want the current thread to continue until all our prior memory
929 // accesses (including the new thread state) are visible to other threads.
930 __ li(R0/*thread_state*/, _thread_in_native_trans);
931 __ release();
932 __ stw(R0/*thread_state*/, thread_(thread_state));
933 if (UseMembar) {
934 __ fence();
935 }
936 // Write serialization page so that the VM thread can do a pseudo remote
937 // membar. We use the current thread pointer to calculate a thread
938 // specific offset to write to within the page. This minimizes bus
939 // traffic due to cache line collision.
940 else {
941 __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2);
942 }
943
944 // Now before we return to java we must look for a current safepoint
945 // (a new safepoint can not start since we entered native_trans).
946 // We must check here because a current safepoint could be modifying
947 // the callers registers right this moment.
948
949 // Acquire isn't strictly necessary here because of the fence, but
950 // sync_state is declared to be volatile, so we do it anyway
951 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
952 int sync_state_offs = __ load_const_optimized(sync_state_addr, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
953
954 // TODO PPC port assert(4 == SafepointSynchronize::sz_state(), "unexpected field size");
955 __ lwz(sync_state, sync_state_offs, sync_state_addr);
956
957 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
958 __ lwz(suspend_flags, thread_(suspend_flags));
959
960 Label sync_check_done;
961 Label do_safepoint;
962 // No synchronization in progress nor yet synchronized.
963 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
964 // Not suspended.
965 __ cmpwi(CCR1, suspend_flags, 0);
966
967 __ bne(CCR0, do_safepoint);
968 __ beq(CCR1, sync_check_done);
969 __ bind(do_safepoint);
970 __ isync();
971 // Block. We do the call directly and leave the current
972 // last_Java_frame setup undisturbed. We must save any possible
973 // native result across the call. No oop is present.
974
975 __ mr(R3_ARG1, R16_thread);
976 #if defined(ABI_ELFv2)
977 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
978 relocInfo::none);
979 #else
980 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
981 relocInfo::none);
982 #endif
983
984 __ bind(sync_check_done);
985
986 //=============================================================================
987 // <<<<<< Back in Interpreter Frame >>>>>
988
989 // We are in thread_in_native_trans here and back in the normal
990 // interpreter frame. We don't have to do anything special about
991 // safepoints and we can switch to Java mode anytime we are ready.
992
993 // Note: frame::interpreter_frame_result has a dependency on how the
994 // method result is saved across the call to post_method_exit. For
995 // native methods it assumes that the non-FPU/non-void result is
996 // saved in _native_lresult and a FPU result in _native_fresult. If
997 // this changes then the interpreter_frame_result implementation
998 // will need to be updated too.
999
1000 // On PPC64, we have stored the result directly after the native call.
1001
1002 //=============================================================================
1003 // Back in Java
1004
1005 // We use release_store_fence to update values like the thread state, where
1006 // we don't want the current thread to continue until all our prior memory
1007 // accesses (including the new thread state) are visible to other threads.
1008 __ li(R0/*thread_state*/, _thread_in_Java);
1009 __ release();
1010 __ stw(R0/*thread_state*/, thread_(thread_state));
1011 if (UseMembar) {
1012 __ fence();
1013 }
1014
1015 __ reset_last_Java_frame();
1016
1017 // Jvmdi/jvmpi support. Whether we've got an exception pending or
1018 // not, and whether unlocking throws an exception or not, we notify
1019 // on native method exit. If we do have an exception, we'll end up
1020 // in the caller's context to handle it, so if we don't do the
1021 // notify here, we'll drop it on the floor.
1022 __ notify_method_exit(true/*native method*/,
1023 ilgl /*illegal state (not used for native methods)*/,
1024 InterpreterMacroAssembler::NotifyJVMTI,
1025 false /*check_exceptions*/);
1026
1027 //=============================================================================
1028 // Handle exceptions
1029
1030 if (synchronized) {
1031 // Don't check for exceptions since we're still in the i2n frame. Do that
1032 // manually afterwards.
1033 unlock_method(false);
1034 }
1035
1036 // Reset active handles after returning from native.
1037 // thread->active_handles()->clear();
1038 __ ld(active_handles, thread_(active_handles));
1039 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1040 __ li(R0, 0);
1041 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles);
1042
1043 Label exception_return_sync_check_already_unlocked;
1044 __ ld(R0/*pending_exception*/, thread_(pending_exception));
1045 __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1046 __ bne(CCR0, exception_return_sync_check_already_unlocked);
1047
1048 //-----------------------------------------------------------------------------
1049 // No exception pending.
1050
1051 // Move native method result back into proper registers and return.
1052 // Invoke result handler (may unbox/promote).
1053 __ ld(R11_scratch1, 0, R1_SP);
1054 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1055 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1056 __ call_stub(result_handler_addr);
1057
1058 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1059
1060 // Must use the return pc which was loaded from the caller's frame
1061 // as the VM uses return-pc-patching for deoptimization.
1062 __ mtlr(R0);
1063 __ blr();
1064
1065 //-----------------------------------------------------------------------------
1066 // An exception is pending. We call into the runtime only if the
1067 // caller was not interpreted. If it was interpreted the
1068 // interpreter will do the correct thing. If it isn't interpreted
1069 // (call stub/compiled code) we will change our return and continue.
1070
1071 BIND(exception_return_sync_check);
1072
1073 if (synchronized) {
1074 // Don't check for exceptions since we're still in the i2n frame. Do that
1075 // manually afterwards.
1076 unlock_method(false);
1077 }
1078 BIND(exception_return_sync_check_already_unlocked);
1079
1080 const Register return_pc = R31;
1081
1082 __ ld(return_pc, 0, R1_SP);
1083 __ ld(return_pc, _abi(lr), return_pc);
1084
1085 // Get the address of the exception handler.
1086 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1087 R16_thread,
1088 return_pc /* return pc */);
1089 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1090
1091 // Load the PC of the the exception handler into LR.
1092 __ mtlr(R3_RET);
1093
1094 // Load exception into R3_ARG1 and clear pending exception in thread.
1095 __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1096 __ li(R4_ARG2, 0);
1097 __ std(R4_ARG2, thread_(pending_exception));
1098
1099 // Load the original return pc into R4_ARG2.
1100 __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1101
1102 // Return to exception handler.
1103 __ blr();
1104
1105 //=============================================================================
1106 // Counter overflow.
1107
1108 if (inc_counter) {
1109 // Handle invocation counter overflow.
1110 __ bind(invocation_counter_overflow);
1111
1112 generate_counter_overflow(continue_after_compile);
1113 }
1114
1115 return entry;
1116 }
1117
1118 // Generic interpreted method entry to (asm) interpreter.
1119 //
1120 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1121 bool inc_counter = UseCompiler || CountCompiledCalls;
1122 address entry = __ pc();
1123 // Generate the code to allocate the interpreter stack frame.
1124 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1125 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame.
1126
1127 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1128
1129 #ifdef FAST_DISPATCH
1130 __ unimplemented("Fast dispatch in generate_normal_entry");
1131 #if 0
1132 __ set((intptr_t)Interpreter::dispatch_table(), IdispatchTables);
1133 // Set bytecode dispatch table base.
1134 #endif
1135 #endif
1136
1137 // --------------------------------------------------------------------------
1138 // Zero out non-parameter locals.
1139 // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1140 // worth to ask the flag, just do it.
1141 Register Rslot_addr = R6_ARG4,
1142 Rnum = R7_ARG5;
1143 Label Lno_locals, Lzero_loop;
1144
1145 // Set up the zeroing loop.
1146 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1147 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1148 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1149 __ beq(CCR0, Lno_locals);
1150 __ li(R0, 0);
1151 __ mtctr(Rnum);
1152
1153 // The zero locals loop.
1154 __ bind(Lzero_loop);
1155 __ std(R0, 0, Rslot_addr);
1156 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1157 __ bdnz(Lzero_loop);
1158
1159 __ bind(Lno_locals);
1160
1161 // --------------------------------------------------------------------------
1162 // Counter increment and overflow check.
1163 Label invocation_counter_overflow,
1164 profile_method,
1165 profile_method_continue;
1166 if (inc_counter || ProfileInterpreter) {
1167
1168 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1169 if (synchronized) {
1170 // Since at this point in the method invocation the exception handler
1171 // would try to exit the monitor of synchronized methods which hasn't
1172 // been entered yet, we set the thread local variable
1173 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1174 // runtime, exception handling i.e. unlock_if_synchronized_method will
1175 // check this thread local flag.
1176 // This flag has two effects, one is to force an unwind in the topmost
1177 // interpreter frame and not perform an unlock while doing so.
1178 __ li(R0, 1);
1179 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1180 }
1181
1182 // Argument and return type profiling.
1183 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1184
1185 // Increment invocation counter and check for overflow.
1186 if (inc_counter) {
1187 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1188 }
1189
1190 __ bind(profile_method_continue);
1191
1192 // Reset the _do_not_unlock_if_synchronized flag.
1193 if (synchronized) {
1194 __ li(R0, 0);
1195 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1196 }
1197 }
1198
1199 // --------------------------------------------------------------------------
1200 // Locking of synchronized methods. Must happen AFTER invocation_counter
1201 // check and stack overflow check, so method is not locked if overflows.
1202 if (synchronized) {
1203 lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1204 }
1205 #ifdef ASSERT
1206 else {
1207 Label Lok;
1208 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1209 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1210 __ asm_assert_eq("method needs synchronization", 0x8521);
1211 __ bind(Lok);
1212 }
1213 #endif // ASSERT
1214
1215 __ verify_thread();
1216
1217 // --------------------------------------------------------------------------
1218 // JVMTI support
1219 __ notify_method_entry();
1220
1221 // --------------------------------------------------------------------------
1222 // Start executing instructions.
1223 __ dispatch_next(vtos);
1224
1225 // --------------------------------------------------------------------------
1226 // Out of line counter overflow and MDO creation code.
1227 if (ProfileInterpreter) {
1228 // We have decided to profile this method in the interpreter.
1229 __ bind(profile_method);
1230 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1231 __ set_method_data_pointer_for_bcp();
1232 __ b(profile_method_continue);
1233 }
1234
1235 if (inc_counter) {
1236 // Handle invocation counter overflow.
1237 __ bind(invocation_counter_overflow);
1238 generate_counter_overflow(profile_method_continue);
1239 }
1240 return entry;
1241 }
1242 // CRC32 Intrinsics.
1243 //
1244 // Contract on scratch and work registers.
1245 // =======================================
1246 //
1247 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1248 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1249 // You can't rely on these registers across calls.
1250 //
1251 // The generators for CRC32_update and for CRC32_updateBytes use the
1252 // scratch/work register set internally, passing the work registers
1253 // as arguments to the MacroAssembler emitters as required.
1254 //
1255 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1256 // Their contents is not constant but may change according to the requirements
1257 // of the emitted code.
1258 //
1259 // All other registers from the scratch/work register set are used "internally"
1260 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1261 // Basically, only R3_RET contains a defined value which is the function result.
1262 //
1263 /**
1264 * Method entry for static native methods:
1265 * int java.util.zip.CRC32.update(int crc, int b)
1266 */
1267 address InterpreterGenerator::generate_CRC32_update_entry() {
1268 address start = __ function_entry(); // Remember stub start address (is rtn value).
1269
1270 if (UseCRC32Intrinsics) {
1271 Label slow_path;
1272
1273 // Safepoint check
1274 const Register sync_state = R11_scratch1;
1275 int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
1276 __ lwz(sync_state, sync_state_offs, sync_state);
1277 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
1278 __ bne(CCR0, slow_path);
1279
1280 // We don't generate local frame and don't align stack because
1281 // we not even call stub code (we generate the code inline)
1282 // and there is no safepoint on this path.
1283
1284 // Load java parameters.
1285 // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1286 const Register argP = R15_esp;
1287 const Register crc = R3_ARG1; // crc value
1288 const Register data = R4_ARG2; // address of java byte value (kernel_crc32 needs address)
1289 const Register dataLen = R5_ARG3; // source data len (1 byte). Not used because calling the single-byte emitter.
1290 const Register table = R6_ARG4; // address of crc32 table
1291 const Register tmp = dataLen; // Reuse unused len register to show we don't actually need a separate tmp here.
1292
1293 BLOCK_COMMENT("CRC32_update {");
1294
1295 // Arguments are reversed on java expression stack
1296 #ifdef VM_LITTLE_ENDIAN
1297 __ addi(data, argP, 0+1*wordSize); // (stack) address of byte value. Emitter expects address, not value.
1298 // Being passed as an int, the single byte is at offset +0.
1299 #else
1300 __ addi(data, argP, 3+1*wordSize); // (stack) address of byte value. Emitter expects address, not value.
1301 // Being passed from java as an int, the single byte is at offset +3.
1302 #endif
1303 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1304
1305 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
1306 __ kernel_crc32_singleByte(crc, data, dataLen, table, tmp);
1307
1308 // Restore caller sp for c2i case and return.
1309 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
1310 __ blr();
1311
1312 // Generate a vanilla native entry as the slow path.
1313 BLOCK_COMMENT("} CRC32_update");
1314 BIND(slow_path);
1315 }
1316
1317 (void) generate_native_entry(false);
1318
1319 #if defined(ABI_ELFv2)
1320 return start;
1321 #else
1322 return *(address*)start;
1323 #endif
1324 }
1325
1326 // CRC32 Intrinsics.
1327 /**
1328 * Method entry for static native methods:
1329 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len)
1330 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1331 */
1332 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1333 address start = __ function_entry(); // Remember stub start address (is rtn value).
1334
1335 if (UseCRC32Intrinsics) {
1336 Label slow_path;
1337
1338 // Safepoint check
1339 const Register sync_state = R11_scratch1;
1340 int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
1341 __ lwz(sync_state, sync_state_offs, sync_state);
1342 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
1343 __ bne(CCR0, slow_path);
1344
1345 // We don't generate local frame and don't align stack because
1346 // we not even call stub code (we generate the code inline)
1347 // and there is no safepoint on this path.
1348
1349 // Load parameters.
1350 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1351 const Register argP = R15_esp;
1352 const Register crc = R3_ARG1; // crc value
1353 const Register data = R4_ARG2; // address of java byte array
1354 const Register dataLen = R5_ARG3; // source data len
1355 const Register table = R6_ARG4; // address of crc32 table
1356
1357 const Register t0 = R9; // scratch registers for crc calculation
1358 const Register t1 = R10;
1359 const Register t2 = R11;
1360 const Register t3 = R12;
1361
1362 const Register tc0 = R2; // registers to hold pre-calculated column addresses
1363 const Register tc1 = R7;
1364 const Register tc2 = R8;
1365 const Register tc3 = table; // table address is reconstructed at the end of kernel_crc32_* emitters
1366
1367 const Register tmp = t0; // Only used very locally to calculate byte buffer address.
1368
1369 // Arguments are reversed on java expression stack.
1370 // Calculate address of start element.
1371 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1372 BLOCK_COMMENT("CRC32_updateByteBuffer {");
1373 // crc @ (SP + 5W) (32bit)
1374 // buf @ (SP + 3W) (64bit ptr to long array)
1375 // off @ (SP + 2W) (32bit)
1376 // dataLen @ (SP + 1W) (32bit)
1377 // data = buf + off
1378 __ ld( data, 3*wordSize, argP); // start of byte buffer
1379 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1380 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1381 __ lwz( crc, 5*wordSize, argP); // current crc state
1382 __ add( data, data, tmp); // Add byte buffer offset.
1383 } else { // Used for "updateBytes update".
1384 BLOCK_COMMENT("CRC32_updateBytes {");
1385 // crc @ (SP + 4W) (32bit)
1386 // buf @ (SP + 3W) (64bit ptr to byte array)
1387 // off @ (SP + 2W) (32bit)
1388 // dataLen @ (SP + 1W) (32bit)
1389 // data = buf + off + base_offset
1390 __ ld( data, 3*wordSize, argP); // start of byte buffer
1391 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1392 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1393 __ add( data, data, tmp); // add byte buffer offset
1394 __ lwz( crc, 4*wordSize, argP); // current crc state
1395 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1396 }
1397
1398 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
1399
1400 #if 1
1401 // Performance measurements show the 1word and 2word variants to be almost equivalent,
1402 // with very light advantages for the 1word variant. We chose the 1word variant for
1403 // code compactness.
1404 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3);
1405 #else
1406 if (UseNewCode) {
1407 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3);
1408 } else if (UseNewCode2) {
1409 __ kernel_crc32_2word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3);
1410 } else if (UseNewCode3) {
1411 __ kernel_crc32_1byte(crc, data, dataLen, table, t0, t1, t2, t3);
1412 } else {
1413 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3);
1414 }
1415 #endif
1416
1417 // Restore caller sp for c2i case and return.
1418 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
1419 __ blr();
1420
1421 // Generate a vanilla native entry as the slow path.
1422 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1423 BIND(slow_path);
1424 }
1425
1426 (void) generate_native_entry(false);
1427
1428 #if defined(ABI_ELFv2)
1429 return start;
1430 #else
1431 return *(address*)start;
1432 #endif
1433 }
1434
1435 // These should never be compiled since the interpreter will prefer
1436 // the compiled version to the intrinsic version.
1437 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1438 return !math_entry_available(method_kind(m));
1439 }
1440
1441 // How much stack a method activation needs in stack slots.
1442 // We must calc this exactly like in generate_fixed_frame.
1443 // Note: This returns the conservative size assuming maximum alignment.
1444 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1445 const int max_alignment_size = 2;
1446 const int abi_scratch = frame::abi_reg_args_size;
1447 return method->max_locals() + method->max_stack() +
1448 frame::interpreter_frame_monitor_size() + max_alignment_size + abi_scratch;
1449 }
1450
1451 // Returns number of stackElementWords needed for the interpreter frame with the
1452 // given sections.
1453 // This overestimates the stack by one slot in case of alignments.
1454 int AbstractInterpreter::size_activation(int max_stack,
1455 int temps,
1456 int extra_args,
1457 int monitors,
1458 int callee_params,
1459 int callee_locals,
1460 bool is_top_frame) {
1461 // Note: This calculation must exactly parallel the frame setup
1462 // in InterpreterGenerator::generate_fixed_frame.
1463 assert(Interpreter::stackElementWords == 1, "sanity");
1464 const int max_alignment_space = StackAlignmentInBytes / Interpreter::stackElementSize;
1465 const int abi_scratch = is_top_frame ? (frame::abi_reg_args_size / Interpreter::stackElementSize) :
1466 (frame::abi_minframe_size / Interpreter::stackElementSize);
1467 const int size =
1468 max_stack +
1469 (callee_locals - callee_params) +
1470 monitors * frame::interpreter_frame_monitor_size() +
1471 max_alignment_space +
1472 abi_scratch +
1473 frame::ijava_state_size / Interpreter::stackElementSize;
1474
1475 // Fixed size of an interpreter frame, align to 16-byte.
1476 return (size & -2);
1477 }
1478
1479 // Fills a sceletal interpreter frame generated during deoptimizations.
1480 //
1481 // Parameters:
1482 //
1483 // interpreter_frame != NULL:
1484 // set up the method, locals, and monitors.
1485 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1486 // right size, as determined by a previous call to this method.
1487 // It is also guaranteed to be walkable even though it is in a skeletal state
1488 //
1489 // is_top_frame == true:
1490 // We're processing the *oldest* interpreter frame!
1491 //
1492 // pop_frame_extra_args:
1493 // If this is != 0 we are returning to a deoptimized frame by popping
1494 // off the callee frame. We want to re-execute the call that called the
1495 // callee interpreted, but since the return to the interpreter would pop
1496 // the arguments off advance the esp by dummy popframe_extra_args slots.
1497 // Popping off those will establish the stack layout as it was before the call.
1498 //
1499 void AbstractInterpreter::layout_activation(Method* method,
1500 int tempcount,
1501 int popframe_extra_args,
1502 int moncount,
1503 int caller_actual_parameters,
1504 int callee_param_count,
1505 int callee_locals_count,
1506 frame* caller,
1507 frame* interpreter_frame,
1508 bool is_top_frame,
1509 bool is_bottom_frame) {
1510
1511 const int abi_scratch = is_top_frame ? (frame::abi_reg_args_size / Interpreter::stackElementSize) :
1512 (frame::abi_minframe_size / Interpreter::stackElementSize);
1513
1514 intptr_t* locals_base = (caller->is_interpreted_frame()) ?
1515 caller->interpreter_frame_esp() + caller_actual_parameters :
1516 caller->sp() + method->max_locals() - 1 + (frame::abi_minframe_size / Interpreter::stackElementSize) ;
1517
1518 intptr_t* monitor_base = caller->sp() - frame::ijava_state_size / Interpreter::stackElementSize ;
1519 intptr_t* monitor = monitor_base - (moncount * frame::interpreter_frame_monitor_size());
1520 intptr_t* esp_base = monitor - 1;
1521 intptr_t* esp = esp_base - tempcount - popframe_extra_args;
1522 intptr_t* sp = (intptr_t *) (((intptr_t) (esp_base - callee_locals_count + callee_param_count - method->max_stack()- abi_scratch)) & -StackAlignmentInBytes);
1523 intptr_t* sender_sp = caller->sp() + (frame::abi_minframe_size - frame::abi_reg_args_size) / Interpreter::stackElementSize;
1524 intptr_t* top_frame_sp = is_top_frame ? sp : sp + (frame::abi_minframe_size - frame::abi_reg_args_size) / Interpreter::stackElementSize;
1525
1526 interpreter_frame->interpreter_frame_set_method(method);
1527 interpreter_frame->interpreter_frame_set_locals(locals_base);
1528 interpreter_frame->interpreter_frame_set_cpcache(method->constants()->cache());
1529 interpreter_frame->interpreter_frame_set_esp(esp);
1530 interpreter_frame->interpreter_frame_set_monitor_end((BasicObjectLock *)monitor);
1531 interpreter_frame->interpreter_frame_set_top_frame_sp(top_frame_sp);
1532 if (!is_bottom_frame) {
1533 interpreter_frame->interpreter_frame_set_sender_sp(sender_sp);
1534 }
1535 }
1536
1537 // =============================================================================
1538 // Exceptions
1539
1540 void TemplateInterpreterGenerator::generate_throw_exception() {
1541 Register Rexception = R17_tos,
1542 Rcontinuation = R3_RET;
1543
1544 // --------------------------------------------------------------------------
1545 // Entry point if an method returns with a pending exception (rethrow).
1546 Interpreter::_rethrow_exception_entry = __ pc();
1547 {
1548 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
1549 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
1550 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
1551
1552 // Compiled code destroys templateTableBase, reload.
1553 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
1554 }
1555
1556 // Entry point if a interpreted method throws an exception (throw).
1557 Interpreter::_throw_exception_entry = __ pc();
1558 {
1559 __ mr(Rexception, R3_RET);
1560
1561 __ verify_thread();
1562 __ verify_oop(Rexception);
1563
1564 // Expression stack must be empty before entering the VM in case of an exception.
1565 __ empty_expression_stack();
1566 // Find exception handler address and preserve exception oop.
1567 // Call C routine to find handler and jump to it.
1568 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
1569 __ mtctr(Rcontinuation);
1570 // Push exception for exception handler bytecodes.
1571 __ push_ptr(Rexception);
1572
1573 // Jump to exception handler (may be remove activation entry!).
1574 __ bctr();
1575 }
1576
1577 // If the exception is not handled in the current frame the frame is
1578 // removed and the exception is rethrown (i.e. exception
1579 // continuation is _rethrow_exception).
1580 //
1581 // Note: At this point the bci is still the bxi for the instruction
1582 // which caused the exception and the expression stack is
1583 // empty. Thus, for any VM calls at this point, GC will find a legal
1584 // oop map (with empty expression stack).
1585
1586 // In current activation
1587 // tos: exception
1588 // bcp: exception bcp
1589
1590 // --------------------------------------------------------------------------
1591 // JVMTI PopFrame support
1592
1593 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1594 {
1595 // Set the popframe_processing bit in popframe_condition indicating that we are
1596 // currently handling popframe, so that call_VMs that may happen later do not
1597 // trigger new popframe handling cycles.
1598 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1599 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
1600 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1601
1602 // Empty the expression stack, as in normal exception handling.
1603 __ empty_expression_stack();
1604 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
1605
1606 // Check to see whether we are returning to a deoptimized frame.
1607 // (The PopFrame call ensures that the caller of the popped frame is
1608 // either interpreted or compiled and deoptimizes it if compiled.)
1609 // Note that we don't compare the return PC against the
1610 // deoptimization blob's unpack entry because of the presence of
1611 // adapter frames in C2.
1612 Label Lcaller_not_deoptimized;
1613 Register return_pc = R3_ARG1;
1614 __ ld(return_pc, 0, R1_SP);
1615 __ ld(return_pc, _abi(lr), return_pc);
1616 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
1617 __ cmpdi(CCR0, R3_RET, 0);
1618 __ bne(CCR0, Lcaller_not_deoptimized);
1619
1620 // The deoptimized case.
1621 // In this case, we can't call dispatch_next() after the frame is
1622 // popped, but instead must save the incoming arguments and restore
1623 // them after deoptimization has occurred.
1624 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
1625 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
1626 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
1627 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
1628 __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
1629 // Save these arguments.
1630 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
1631
1632 // Inform deoptimization that it is responsible for restoring these arguments.
1633 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1634 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1635
1636 // Return from the current method into the deoptimization blob. Will eventually
1637 // end up in the deopt interpeter entry, deoptimization prepared everything that
1638 // we will reexecute the call that called us.
1639 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
1640 __ mtlr(return_pc);
1641 __ blr();
1642
1643 // The non-deoptimized case.
1644 __ bind(Lcaller_not_deoptimized);
1645
1646 // Clear the popframe condition flag.
1647 __ li(R0, 0);
1648 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1649
1650 // Get out of the current method and re-execute the call that called us.
1651 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
1652 __ restore_interpreter_state(R11_scratch1);
1653 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
1654 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
1655 if (ProfileInterpreter) {
1656 __ set_method_data_pointer_for_bcp();
1657 __ ld(R11_scratch1, 0, R1_SP);
1658 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
1659 }
1660 #if INCLUDE_JVMTI
1661 Label L_done;
1662
1663 __ lbz(R11_scratch1, 0, R14_bcp);
1664 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
1665 __ bne(CCR0, L_done);
1666
1667 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1668 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1669 __ ld(R4_ARG2, 0, R18_locals);
1670 __ MacroAssembler::call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp, false);
1671 __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
1672 __ cmpdi(CCR0, R4_ARG2, 0);
1673 __ beq(CCR0, L_done);
1674 __ std(R4_ARG2, wordSize, R15_esp);
1675 __ bind(L_done);
1676 #endif // INCLUDE_JVMTI
1677 __ dispatch_next(vtos);
1678 }
1679 // end of JVMTI PopFrame support
1680
1681 // --------------------------------------------------------------------------
1682 // Remove activation exception entry.
1683 // This is jumped to if an interpreted method can't handle an exception itself
1684 // (we come from the throw/rethrow exception entry above). We're going to call
1685 // into the VM to find the exception handler in the caller, pop the current
1686 // frame and return the handler we calculated.
1687 Interpreter::_remove_activation_entry = __ pc();
1688 {
1689 __ pop_ptr(Rexception);
1690 __ verify_thread();
1691 __ verify_oop(Rexception);
1692 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
1693
1694 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
1695 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
1696
1697 __ get_vm_result(Rexception);
1698
1699 // We are done with this activation frame; find out where to go next.
1700 // The continuation point will be an exception handler, which expects
1701 // the following registers set up:
1702 //
1703 // RET: exception oop
1704 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
1705
1706 Register return_pc = R31; // Needs to survive the runtime call.
1707 __ ld(return_pc, 0, R1_SP);
1708 __ ld(return_pc, _abi(lr), return_pc);
1709 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
1710
1711 // Remove the current activation.
1712 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
1713
1714 __ mr(R4_ARG2, return_pc);
1715 __ mtlr(R3_RET);
1716 __ mr(R3_RET, Rexception);
1717 __ blr();
1718 }
1719 }
1720
1721 // JVMTI ForceEarlyReturn support.
1722 // Returns "in the middle" of a method with a "fake" return value.
1723 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1724
1725 Register Rscratch1 = R11_scratch1,
1726 Rscratch2 = R12_scratch2;
1727
1728 address entry = __ pc();
1729 __ empty_expression_stack();
1730
1731 __ load_earlyret_value(state, Rscratch1);
1732
1733 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
1734 // Clear the earlyret state.
1735 __ li(R0, 0);
1736 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
1737
1738 __ remove_activation(state, false, false);
1739 // Copied from TemplateTable::_return.
1740 // Restoration of lr done by remove_activation.
1741 switch (state) {
1742 case ltos:
1743 case btos:
1744 case ctos:
1745 case stos:
1746 case atos:
1747 case itos: __ mr(R3_RET, R17_tos); break;
1748 case ftos:
1749 case dtos: __ fmr(F1_RET, F15_ftos); break;
1750 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
1751 // to get visible before the reference to the object gets stored anywhere.
1752 __ membar(Assembler::StoreStore); break;
1753 default : ShouldNotReachHere();
1754 }
1755 __ blr();
1756
1757 return entry;
1758 } // end of ForceEarlyReturn support
1759
1760 //-----------------------------------------------------------------------------
1761 // Helper for vtos entry point generation
1762
1763 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1764 address& bep,
1765 address& cep,
1766 address& sep,
1767 address& aep,
1768 address& iep,
1769 address& lep,
1770 address& fep,
1771 address& dep,
1772 address& vep) {
1773 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1774 Label L;
1775
1776 aep = __ pc(); __ push_ptr(); __ b(L);
1777 fep = __ pc(); __ push_f(); __ b(L);
1778 dep = __ pc(); __ push_d(); __ b(L);
1779 lep = __ pc(); __ push_l(); __ b(L);
1780 __ align(32, 12, 24); // align L
1781 bep = cep = sep =
1782 iep = __ pc(); __ push_i();
1783 vep = __ pc();
1784 __ bind(L);
1785 generate_and_dispatch(t);
1786 }
1787
1788 //-----------------------------------------------------------------------------
1789 // Generation of individual instructions
1790
1791 // helpers for generate_and_dispatch
1792
1793 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1794 : TemplateInterpreterGenerator(code) {
1795 generate_all(); // Down here so it can be "virtual".
1796 }
1797
1798 //-----------------------------------------------------------------------------
1799
1800 // Non-product code
1801 #ifndef PRODUCT
1802 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1803 //__ flush_bundle();
1804 address entry = __ pc();
1805
1806 const char *bname = NULL;
1807 uint tsize = 0;
1808 switch(state) {
1809 case ftos:
1810 bname = "trace_code_ftos {";
1811 tsize = 2;
1812 break;
1813 case btos:
1814 bname = "trace_code_btos {";
1815 tsize = 2;
1816 break;
1817 case ctos:
1818 bname = "trace_code_ctos {";
1819 tsize = 2;
1820 break;
1821 case stos:
1822 bname = "trace_code_stos {";
1823 tsize = 2;
1824 break;
1825 case itos:
1826 bname = "trace_code_itos {";
1827 tsize = 2;
1828 break;
1829 case ltos:
1830 bname = "trace_code_ltos {";
1831 tsize = 3;
1832 break;
1833 case atos:
1834 bname = "trace_code_atos {";
1835 tsize = 2;
1836 break;
1837 case vtos:
1838 // Note: In case of vtos, the topmost of stack value could be a int or doubl
1839 // In case of a double (2 slots) we won't see the 2nd stack value.
1840 // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
1841 bname = "trace_code_vtos {";
1842 tsize = 2;
1843
1844 break;
1845 case dtos:
1846 bname = "trace_code_dtos {";
1847 tsize = 3;
1848 break;
1849 default:
1850 ShouldNotReachHere();
1851 }
1852 BLOCK_COMMENT(bname);
1853
1854 // Support short-cut for TraceBytecodesAt.
1855 // Don't call into the VM if we don't want to trace to speed up things.
1856 Label Lskip_vm_call;
1857 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
1858 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
1859 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
1860 __ ld(R11_scratch1, offs1, R11_scratch1);
1861 __ lwa(R12_scratch2, offs2, R12_scratch2);
1862 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
1863 __ blt(CCR0, Lskip_vm_call);
1864 }
1865
1866 __ push(state);
1867 // Load 2 topmost expression stack values.
1868 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
1869 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
1870 __ mflr(R31);
1871 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
1872 __ mtlr(R31);
1873 __ pop(state);
1874
1875 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
1876 __ bind(Lskip_vm_call);
1877 }
1878 __ blr();
1879 BLOCK_COMMENT("} trace_code");
1880 return entry;
1881 }
1882
1883 void TemplateInterpreterGenerator::count_bytecode() {
1884 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
1885 __ lwz(R12_scratch2, offs, R11_scratch1);
1886 __ addi(R12_scratch2, R12_scratch2, 1);
1887 __ stw(R12_scratch2, offs, R11_scratch1);
1888 }
1889
1890 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1891 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
1892 __ lwz(R12_scratch2, offs, R11_scratch1);
1893 __ addi(R12_scratch2, R12_scratch2, 1);
1894 __ stw(R12_scratch2, offs, R11_scratch1);
1895 }
1896
1897 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1898 const Register addr = R11_scratch1,
1899 tmp = R12_scratch2;
1900 // Get index, shift out old bytecode, bring in new bytecode, and store it.
1901 // _index = (_index >> log2_number_of_codes) |
1902 // (bytecode << log2_number_of_codes);
1903 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
1904 __ lwz(tmp, offs1, addr);
1905 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
1906 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1907 __ stw(tmp, offs1, addr);
1908
1909 // Bump bucket contents.
1910 // _counters[_index] ++;
1911 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
1912 __ sldi(tmp, tmp, LogBytesPerInt);
1913 __ add(addr, tmp, addr);
1914 __ lwz(tmp, offs2, addr);
1915 __ addi(tmp, tmp, 1);
1916 __ stw(tmp, offs2, addr);
1917 }
1918
1919 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1920 // Call a little run-time stub to avoid blow-up for each bytecode.
1921 // The run-time runtime saves the right registers, depending on
1922 // the tosca in-state for the given template.
1923
1924 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1925 "entry must have been generated");
1926
1927 // Note: we destroy LR here.
1928 __ bl(Interpreter::trace_code(t->tos_in()));
1929 }
1930
1931 void TemplateInterpreterGenerator::stop_interpreter_at() {
1932 Label L;
1933 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
1934 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
1935 __ ld(R11_scratch1, offs1, R11_scratch1);
1936 __ lwa(R12_scratch2, offs2, R12_scratch2);
1937 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
1938 __ bne(CCR0, L);
1939 __ illtrap();
1940 __ bind(L);
1941 }
1942
1943 #endif // !PRODUCT
1944 #endif // !CC_INTERP