1 /*
2 * Copyright (c) 2008, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "interpreter/bytecodeHistogram.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "interpreter/interpreterRuntime.hpp"
32 #include "interpreter/templateInterpreterGenerator.hpp"
33 #include "interpreter/templateTable.hpp"
34 #include "oops/arrayOop.hpp"
35 #include "oops/methodData.hpp"
36 #include "oops/method.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "prims/jvmtiExport.hpp"
39 #include "prims/jvmtiThreadState.hpp"
40 #include "prims/methodHandles.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/align.hpp"
50 #include "utilities/debug.hpp"
51 #include "utilities/macros.hpp"
52
53 // Size of interpreter code. Increase if too small. Interpreter will
54 // fail with a guarantee ("not enough space for interpreter generation");
55 // if too small.
56 // Run with +PrintInterpreter to get the VM to print out the size.
57 // Max size with JVMTI
58 int TemplateInterpreter::InterpreterCodeSize = 180 * 1024;
59
60 #define __ _masm->
61
62 //------------------------------------------------------------------------------------------------------------------------
63
64 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
65 address entry = __ pc();
66
67 // callee-save register for saving LR, shared with generate_native_entry
68 const Register Rsaved_ret_addr = Rtmp_save0;
69
70 __ mov(Rsaved_ret_addr, LR);
71
72 __ mov(R1, Rmethod);
73 __ mov(R2, Rlocals);
74 __ mov(R3, SP);
75
76
77 // Safer to save R9 (when scratched) since callers may have been
78 // written assuming R9 survives. This is suboptimal but
79 // probably not important for this slow case call site.
80 // Note for R9 saving: slow_signature_handler may copy register
81 // arguments above the current SP (passed as R3). It is safe for
82 // call_VM to use push and pop to protect additional values on the
83 // stack if needed.
84 __ call_VM(CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), true /* save R9 if needed*/);
85 __ add(SP, SP, wordSize); // Skip R0
86 __ pop(RegisterSet(R1, R3)); // Load arguments passed in registers
87 #ifdef __ABI_HARD__
88 // Few alternatives to an always-load-FP-registers approach:
89 // - parse method signature to detect FP arguments
90 // - keep a counter/flag on a stack indicationg number of FP arguments in the method.
91 // The later has been originally implemented and tested but a conditional path could
92 // eliminate any gain imposed by avoiding 8 double word loads.
93 __ fldmiad(SP, FloatRegisterSet(D0, 8), writeback);
94 #endif // __ABI_HARD__
95
96 __ ret(Rsaved_ret_addr);
97
98 return entry;
99 }
100
101
102 //
103 // Various method entries (that c++ and asm interpreter agree upon)
104 //------------------------------------------------------------------------------------------------------------------------
105 //
106 //
107
108 // Abstract method entry
109 // Attempt to execute abstract method. Throw exception
110 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
111 address entry_point = __ pc();
112
113
114 __ empty_expression_stack();
115
116 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
117
118 DEBUG_ONLY(STOP("generate_abstract_entry");) // Should not reach here
119 return entry_point;
120 }
121
122 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
123 if (!InlineIntrinsics) return NULL; // Generate a vanilla entry
124
125 // TODO: ARM
126 return NULL;
127
128 address entry_point = __ pc();
129 STOP("generate_math_entry");
130 return entry_point;
131 }
132
133 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
134 address entry = __ pc();
135
136 // Note: There should be a minimal interpreter frame set up when stack
137 // overflow occurs since we check explicitly for it now.
138 //
139 #ifdef ASSERT
140 { Label L;
141 __ sub(Rtemp, FP, - frame::interpreter_frame_monitor_block_top_offset * wordSize);
142 __ cmp(SP, Rtemp); // Rtemp = maximal SP for current FP,
143 // (stack grows negative)
144 __ b(L, ls); // check if frame is complete
145 __ stop ("interpreter frame not set up");
146 __ bind(L);
147 }
148 #endif // ASSERT
149
150 // Restore bcp under the assumption that the current frame is still
151 // interpreted
152 __ restore_bcp();
153
154 // expression stack must be empty before entering the VM if an exception
155 // happened
156 __ empty_expression_stack();
157
158 // throw exception
159 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
160
161 __ should_not_reach_here();
162
163 return entry;
164 }
165
166 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
167 address entry = __ pc();
168
169 // index is in R4_ArrayIndexOutOfBounds_index
170
171 // expression stack must be empty before entering the VM if an exception happened
172 __ empty_expression_stack();
173
174 // setup parameters
175 // Array expected in R1.
176 __ mov(R2, R4_ArrayIndexOutOfBounds_index);
177
178 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R1, R2);
179
180 __ nop(); // to avoid filling CPU pipeline with invalid instructions
181 __ nop();
182 __ should_not_reach_here();
183
184 return entry;
185 }
186
187 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
188 address entry = __ pc();
189
190 // object is in R2_ClassCastException_obj
191
192 // expression stack must be empty before entering the VM if an exception
193 // happened
194 __ empty_expression_stack();
195
196 __ mov(R1, R2_ClassCastException_obj);
197 __ call_VM(noreg,
198 CAST_FROM_FN_PTR(address,
199 InterpreterRuntime::throw_ClassCastException),
200 R1);
201
202 __ should_not_reach_here();
203
204 return entry;
205 }
206
207 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
208 assert(!pass_oop || message == NULL, "either oop or message but not both");
209 address entry = __ pc();
210
211 InlinedString Lname(name);
212 InlinedString Lmessage(message);
213
214 if (pass_oop) {
215 // object is at TOS
216 __ pop_ptr(R2);
217 }
218
219 // expression stack must be empty before entering the VM if an exception happened
220 __ empty_expression_stack();
221
222 // setup parameters
223 __ ldr_literal(R1, Lname);
224
225 if (pass_oop) {
226 __ call_VM(Rexception_obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), R1, R2);
227 } else {
228 if (message != NULL) {
229 __ ldr_literal(R2, Lmessage);
230 } else {
231 __ mov(R2, 0);
232 }
233 __ call_VM(Rexception_obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), R1, R2);
234 }
235
236 // throw exception
237 __ b(Interpreter::throw_exception_entry());
238
239 __ nop(); // to avoid filling CPU pipeline with invalid instructions
240 __ nop();
241 __ bind_literal(Lname);
242 if (!pass_oop && (message != NULL)) {
243 __ bind_literal(Lmessage);
244 }
245
246 return entry;
247 }
248
249 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
250 address entry = __ pc();
251
252 __ interp_verify_oop(R0_tos, state, __FILE__, __LINE__);
253
254 // Restore stack bottom in case i2c adjusted stack
255 __ ldr(SP, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
256 // and NULL it as marker that SP is now tos until next java call
257 __ mov(Rtemp, (int)NULL_WORD);
258 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
259
260 __ restore_method();
261 __ restore_bcp();
262 __ restore_dispatch();
263 __ restore_locals();
264
265 const Register Rcache = R2_tmp;
266 const Register Rindex = R3_tmp;
267 __ get_cache_and_index_at_bcp(Rcache, Rindex, 1, index_size);
268
269 __ add(Rtemp, Rcache, AsmOperand(Rindex, lsl, LogBytesPerWord));
270 __ ldrb(Rtemp, Address(Rtemp, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
271 __ check_stack_top();
272 __ add(Rstack_top, Rstack_top, AsmOperand(Rtemp, lsl, Interpreter::logStackElementSize));
273
274 __ convert_retval_to_tos(state);
275
276 __ check_and_handle_popframe();
277 __ check_and_handle_earlyret();
278
279 __ dispatch_next(state, step);
280
281 return entry;
282 }
283
284
285 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
286 address entry = __ pc();
287
288 __ interp_verify_oop(R0_tos, state, __FILE__, __LINE__);
289
290 // The stack is not extended by deopt but we must NULL last_sp as this
291 // entry is like a "return".
292 __ mov(Rtemp, 0);
293 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
294
295 __ restore_method();
296 __ restore_bcp();
297 __ restore_dispatch();
298 __ restore_locals();
299
300 // handle exceptions
301 { Label L;
302 __ ldr(Rtemp, Address(Rthread, Thread::pending_exception_offset()));
303 __ cbz(Rtemp, L);
304 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
305 __ should_not_reach_here();
306 __ bind(L);
307 }
308
309 if (continuation == NULL) {
310 __ dispatch_next(state, step);
311 } else {
312 __ jump_to_entry(continuation);
313 }
314
315 return entry;
316 }
317
318 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
319 // Result handlers are not used on 32-bit ARM
320 // since the returned value is already in appropriate format.
321 __ should_not_reach_here(); // to avoid empty code block
322
323 // The result handler non-zero indicates an object is returned and this is
324 // used in the native entry code.
325 return type == T_OBJECT ? (address)(-1) : NULL;
326 }
327
328 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
329 address entry = __ pc();
330 __ push(state);
331 __ call_VM(noreg, runtime_entry);
332
333 // load current bytecode
334 __ ldrb(R3_bytecode, Address(Rbcp));
335 __ dispatch_only_normal(vtos);
336 return entry;
337 }
338
339
340 // Helpers for commoning out cases in the various type of method entries.
341 //
342
343 // increment invocation count & check for overflow
344 //
345 // Note: checking for negative value instead of overflow
346 // so we have a 'sticky' overflow test
347 //
348 // In: Rmethod.
349 //
350 // Uses R0, R1, Rtemp.
351 //
352 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow,
353 Label* profile_method,
354 Label* profile_method_continue) {
355 Label done;
356 const Register Rcounters = Rtemp;
357 const Address invocation_counter(Rcounters,
358 MethodCounters::invocation_counter_offset() +
359 InvocationCounter::counter_offset());
360
361 // Note: In tiered we increment either counters in MethodCounters* or
362 // in MDO depending if we're profiling or not.
363 if (TieredCompilation) {
364 int increment = InvocationCounter::count_increment;
365 Label no_mdo;
366 if (ProfileInterpreter) {
367 // Are we profiling?
368 __ ldr(R1_tmp, Address(Rmethod, Method::method_data_offset()));
369 __ cbz(R1_tmp, no_mdo);
370 // Increment counter in the MDO
371 const Address mdo_invocation_counter(R1_tmp,
372 in_bytes(MethodData::invocation_counter_offset()) +
373 in_bytes(InvocationCounter::counter_offset()));
374 const Address mask(R1_tmp, in_bytes(MethodData::invoke_mask_offset()));
375 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, R0_tmp, Rtemp, eq, overflow);
376 __ b(done);
377 }
378 __ bind(no_mdo);
379 __ get_method_counters(Rmethod, Rcounters, done);
380 const Address mask(Rcounters, in_bytes(MethodCounters::invoke_mask_offset()));
381 __ increment_mask_and_jump(invocation_counter, increment, mask, R0_tmp, R1_tmp, eq, overflow);
382 __ bind(done);
383 } else { // not TieredCompilation
384 const Address backedge_counter(Rcounters,
385 MethodCounters::backedge_counter_offset() +
386 InvocationCounter::counter_offset());
387
388 const Register Ricnt = R0_tmp; // invocation counter
389 const Register Rbcnt = R1_tmp; // backedge counter
390
391 __ get_method_counters(Rmethod, Rcounters, done);
392
393 if (ProfileInterpreter) {
394 const Register Riic = R1_tmp;
395 __ ldr_s32(Riic, Address(Rcounters, MethodCounters::interpreter_invocation_counter_offset()));
396 __ add(Riic, Riic, 1);
397 __ str_32(Riic, Address(Rcounters, MethodCounters::interpreter_invocation_counter_offset()));
398 }
399
400 // Update standard invocation counters
401
402 __ ldr_u32(Ricnt, invocation_counter);
403 __ ldr_u32(Rbcnt, backedge_counter);
404
405 __ add(Ricnt, Ricnt, InvocationCounter::count_increment);
406
407 __ bic(Rbcnt, Rbcnt, ~InvocationCounter::count_mask_value); // mask out the status bits
408
409 __ str_32(Ricnt, invocation_counter); // save invocation count
410 __ add(Ricnt, Ricnt, Rbcnt); // add both counters
411
412 // profile_method is non-null only for interpreted method so
413 // profile_method != NULL == !native_call
414 // BytecodeInterpreter only calls for native so code is elided.
415
416 if (ProfileInterpreter && profile_method != NULL) {
417 assert(profile_method_continue != NULL, "should be non-null");
418
419 // Test to see if we should create a method data oop
420 // Reuse R1_tmp as we don't need backedge counters anymore.
421 Address profile_limit(Rcounters, in_bytes(MethodCounters::interpreter_profile_limit_offset()));
422 __ ldr_s32(R1_tmp, profile_limit);
423 __ cmp_32(Ricnt, R1_tmp);
424 __ b(*profile_method_continue, lt);
425
426 // if no method data exists, go to profile_method
427 __ test_method_data_pointer(R1_tmp, *profile_method);
428 }
429
430 Address invoke_limit(Rcounters, in_bytes(MethodCounters::interpreter_invocation_limit_offset()));
431 __ ldr_s32(R1_tmp, invoke_limit);
432 __ cmp_32(Ricnt, R1_tmp);
433 __ b(*overflow, hs);
434 __ bind(done);
435 }
436 }
437
438 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
439 // InterpreterRuntime::frequency_counter_overflow takes one argument
440 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
441 // The call returns the address of the verified entry point for the method or NULL
442 // if the compilation did not complete (either went background or bailed out).
443 __ mov(R1, (int)false);
444 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R1);
445
446 // jump to the interpreted entry.
447 __ b(do_continue);
448 }
449
450 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
451 // Check if we've got enough room on the stack for
452 // - overhead;
453 // - locals;
454 // - expression stack.
455 //
456 // Registers on entry:
457 //
458 // R3 = number of additional locals
459 // Rthread
460 // Rmethod
461 // Registers used: R0, R1, R2, Rtemp.
462
463 const Register Radditional_locals = R3;
464 const Register RmaxStack = R2;
465
466 // monitor entry size
467 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
468
469 // total overhead size: entry_size + (saved registers, thru expr stack bottom).
470 // be sure to change this if you add/subtract anything to/from the overhead area
471 const int overhead_size = (frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset)*wordSize + entry_size;
472
473 // Pages reserved for VM runtime calls and subsequent Java calls.
474 const int reserved_pages = JavaThread::stack_shadow_zone_size();
475
476 // Thread::stack_size() includes guard pages, and they should not be touched.
477 const int guard_pages = JavaThread::stack_guard_zone_size();
478
479 __ ldr(R0, Address(Rthread, Thread::stack_base_offset()));
480 __ ldr(R1, Address(Rthread, Thread::stack_size_offset()));
481 __ ldr(Rtemp, Address(Rmethod, Method::const_offset()));
482 __ ldrh(RmaxStack, Address(Rtemp, ConstMethod::max_stack_offset()));
483 __ sub_slow(Rtemp, SP, overhead_size + reserved_pages + guard_pages + Method::extra_stack_words());
484
485 // reserve space for additional locals
486 __ sub(Rtemp, Rtemp, AsmOperand(Radditional_locals, lsl, Interpreter::logStackElementSize));
487
488 // stack size
489 __ sub(R0, R0, R1);
490
491 // reserve space for expression stack
492 __ sub(Rtemp, Rtemp, AsmOperand(RmaxStack, lsl, Interpreter::logStackElementSize));
493
494 __ cmp(Rtemp, R0);
495
496 __ mov(SP, Rsender_sp, ls); // restore SP
497 __ b(StubRoutines::throw_StackOverflowError_entry(), ls);
498 }
499
500
501 // Allocate monitor and lock method (asm interpreter)
502 //
503 void TemplateInterpreterGenerator::lock_method() {
504 // synchronize method
505
506 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
507 assert ((entry_size % StackAlignmentInBytes) == 0, "should keep stack alignment");
508
509 #ifdef ASSERT
510 { Label L;
511 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
512 __ tbnz(Rtemp, JVM_ACC_SYNCHRONIZED_BIT, L);
513 __ stop("method doesn't need synchronization");
514 __ bind(L);
515 }
516 #endif // ASSERT
517
518 // get synchronization object
519 { Label done;
520 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
521 __ tst(Rtemp, JVM_ACC_STATIC);
522 __ ldr(R0, Address(Rlocals, Interpreter::local_offset_in_bytes(0)), eq); // get receiver (assume this is frequent case)
523 __ b(done, eq);
524 __ load_mirror(R0, Rmethod, Rtemp);
525 __ bind(done);
526 }
527
528 // add space for monitor & lock
529
530
531 __ sub(Rstack_top, Rstack_top, entry_size);
532 __ check_stack_top_on_expansion();
533 // add space for a monitor entry
534 __ str(Rstack_top, Address(FP, frame::interpreter_frame_monitor_block_top_offset * wordSize));
535 // set new monitor block top
536 __ str(R0, Address(Rstack_top, BasicObjectLock::obj_offset_in_bytes()));
537 // store object
538 __ mov(R1, Rstack_top); // monitor entry address
539 __ lock_object(R1);
540 }
541
542
543 //
544 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
545 // and for native methods hence the shared code.
546
547 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
548 // Generates the following stack layout:
549 //
550 // [ expr. stack bottom ]
551 // [ saved Rbcp ]
552 // [ current Rlocals ]
553 // [ cache ]
554 // [ mdx ]
555 // [ Method* ]
556 // [ last_sp ]
557 // [ sender_sp ]
558 // [ saved FP ] <--- FP
559 // [ saved LR ]
560
561 // initialize fixed part of activation frame
562 __ push(LR); // save return address
563 __ push(FP); // save FP
564 __ mov(FP, SP); // establish new FP
565
566 __ push(Rsender_sp);
567
568 __ mov(R0, 0);
569 __ push(R0); // leave last_sp as null
570
571 // setup Rbcp
572 if (native_call) {
573 __ mov(Rbcp, 0); // bcp = 0 for native calls
574 } else {
575 __ ldr(Rtemp, Address(Rmethod, Method::const_offset())); // get ConstMethod*
576 __ add(Rbcp, Rtemp, ConstMethod::codes_offset()); // get codebase
577 }
578
579 __ push(Rmethod); // save Method*
580 // Get mirror and store it in the frame as GC root for this Method*
581 __ load_mirror(Rtemp, Rmethod, Rtemp);
582 __ push(Rtemp);
583
584 if (ProfileInterpreter) {
585 __ ldr(Rtemp, Address(Rmethod, Method::method_data_offset()));
586 __ tst(Rtemp, Rtemp);
587 __ add(Rtemp, Rtemp, in_bytes(MethodData::data_offset()), ne);
588 __ push(Rtemp); // set the mdp (method data pointer)
589 } else {
590 __ push(R0);
591 }
592
593 __ ldr(Rtemp, Address(Rmethod, Method::const_offset()));
594 __ ldr(Rtemp, Address(Rtemp, ConstMethod::constants_offset()));
595 __ ldr(Rtemp, Address(Rtemp, ConstantPool::cache_offset_in_bytes()));
596 __ push(Rtemp); // set constant pool cache
597 __ push(Rlocals); // set locals pointer
598 __ push(Rbcp); // set bcp
599 __ push(R0); // reserve word for pointer to expression stack bottom
600 __ str(SP, Address(SP, 0)); // set expression stack bottom
601 }
602
603
604 // End of helpers
605
606 //------------------------------------------------------------------------------------------------------------------------
607 // Entry points
608 //
609 // Here we generate the various kind of entries into the interpreter.
610 // The two main entry type are generic bytecode methods and native call method.
611 // These both come in synchronized and non-synchronized versions but the
612 // frame layout they create is very similar. The other method entry
613 // types are really just special purpose entries that are really entry
614 // and interpretation all in one. These are for trivial methods like
615 // accessor, empty, or special math methods.
616 //
617 // When control flow reaches any of the entry types for the interpreter
618 // the following holds ->
619 //
620 // Arguments:
621 //
622 // Rmethod: Method*
623 // Rthread: thread
624 // Rsender_sp: sender sp
625 // Rparams (SP on 32-bit ARM): pointer to method parameters
626 //
627 // LR: return address
628 //
629 // Stack layout immediately at entry
630 //
631 // [ parameter n ] <--- Rparams (SP on 32-bit ARM)
632 // ...
633 // [ parameter 1 ]
634 // [ expression stack ] (caller's java expression stack)
635
636 // Assuming that we don't go to one of the trivial specialized
637 // entries the stack will look like below when we are ready to execute
638 // the first bytecode (or call the native routine). The register usage
639 // will be as the template based interpreter expects.
640 //
641 // local variables follow incoming parameters immediately; i.e.
642 // the return address is saved at the end of the locals.
643 //
644 // [ expr. stack ] <--- Rstack_top (SP on 32-bit ARM)
645 // [ monitor entry ]
646 // ...
647 // [ monitor entry ]
648 // [ expr. stack bottom ]
649 // [ saved Rbcp ]
650 // [ current Rlocals ]
651 // [ cache ]
652 // [ mdx ]
653 // [ mirror ]
654 // [ Method* ]
655 //
656 // 32-bit ARM:
657 // [ last_sp ]
658 //
659 // [ sender_sp ]
660 // [ saved FP ] <--- FP
661 // [ saved LR ]
662 // [ optional padding(*)]
663 // [ local variable m ]
664 // ...
665 // [ local variable 1 ]
666 // [ parameter n ]
667 // ...
668 // [ parameter 1 ] <--- Rlocals
669 //
670
671 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
672 // Code: _aload_0, _getfield, _areturn
673 // parameter size = 1
674 //
675 // The code that gets generated by this routine is split into 2 parts:
676 // 1. The "intrinsified" code performing an ON_WEAK_OOP_REF load,
677 // 2. The slow path - which is an expansion of the regular method entry.
678 //
679 // Notes:-
680 // * An intrinsic is always executed, where an ON_WEAK_OOP_REF load is performed.
681 // * We may jump to the slow path iff the receiver is null. If the
682 // Reference object is null then we no longer perform an ON_WEAK_OOP_REF load
683 // Thus we can use the regular method entry code to generate the NPE.
684 //
685 // Rmethod: Method*
686 // Rthread: thread
687 // Rsender_sp: sender sp, must be preserved for slow path, set SP to it on fast path
688 // Rparams: parameters
689
690 address entry = __ pc();
691 Label slow_path;
692 const Register Rthis = R0;
693 const Register Rret_addr = Rtmp_save1;
694 assert_different_registers(Rthis, Rret_addr, Rsender_sp);
695
696 const int referent_offset = java_lang_ref_Reference::referent_offset;
697 guarantee(referent_offset > 0, "referent offset not initialized");
698
699 // Check if local 0 != NULL
700 // If the receiver is null then it is OK to jump to the slow path.
701 __ ldr(Rthis, Address(Rparams));
702 __ cbz(Rthis, slow_path);
703
704 // Preserve LR
705 __ mov(Rret_addr, LR);
706
707 // Load the value of the referent field.
708 const Address field_address(Rthis, referent_offset);
709 __ load_heap_oop(R0, field_address, Rtemp, R1_tmp, R2_tmp, ON_WEAK_OOP_REF);
710
711 // _areturn
712 __ mov(SP, Rsender_sp);
713 __ ret(Rret_addr);
714
715 // generate a vanilla interpreter entry as the slow path
716 __ bind(slow_path);
717 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
718 return entry;
719 }
720
721 // Not supported
722 address TemplateInterpreterGenerator::generate_CRC32_update_entry() { return NULL; }
723 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { return NULL; }
724 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) { return NULL; }
725
726 //
727 // Interpreter stub for calling a native method. (asm interpreter)
728 // This sets up a somewhat different looking stack for calling the native method
729 // than the typical interpreter frame setup.
730 //
731
732 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
733 // determine code generation flags
734 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
735
736 // Incoming registers:
737 //
738 // Rmethod: Method*
739 // Rthread: thread
740 // Rsender_sp: sender sp
741 // Rparams: parameters
742
743 address entry_point = __ pc();
744
745 // Register allocation
746 const Register Rsize_of_params = R6;
747 const Register Rsig_handler = Rtmp_save0;
748 const Register Rnative_code = Rtmp_save1;
749 const Register Rresult_handler = R6;
750
751 const Register Rsaved_result_lo = Rtmp_save0; // R4
752 const Register Rsaved_result_hi = Rtmp_save1; // R5
753 FloatRegister saved_result_fp;
754
755
756 __ ldr(Rsize_of_params, Address(Rmethod, Method::const_offset()));
757 __ ldrh(Rsize_of_params, Address(Rsize_of_params, ConstMethod::size_of_parameters_offset()));
758
759 // native calls don't need the stack size check since they have no expression stack
760 // and the arguments are already on the stack and we only add a handful of words
761 // to the stack
762
763 // compute beginning of parameters (Rlocals)
764 __ sub(Rlocals, Rparams, wordSize);
765 __ add(Rlocals, Rlocals, AsmOperand(Rsize_of_params, lsl, Interpreter::logStackElementSize));
766
767 // reserve stack space for oop_temp
768 __ mov(R0, 0);
769 __ push(R0);
770
771 generate_fixed_frame(true); // Note: R9 is now saved in the frame
772
773 // make sure method is native & not abstract
774 #ifdef ASSERT
775 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
776 {
777 Label L;
778 __ tbnz(Rtemp, JVM_ACC_NATIVE_BIT, L);
779 __ stop("tried to execute non-native method as native");
780 __ bind(L);
781 }
782 { Label L;
783 __ tbz(Rtemp, JVM_ACC_ABSTRACT_BIT, L);
784 __ stop("tried to execute abstract method in interpreter");
785 __ bind(L);
786 }
787 #endif
788
789 // increment invocation count & check for overflow
790 Label invocation_counter_overflow;
791 if (inc_counter) {
792 if (synchronized) {
793 // Avoid unlocking method's monitor in case of exception, as it has not
794 // been locked yet.
795 __ set_do_not_unlock_if_synchronized(true, Rtemp);
796 }
797 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
798 }
799
800 Label continue_after_compile;
801 __ bind(continue_after_compile);
802
803 if (inc_counter && synchronized) {
804 __ set_do_not_unlock_if_synchronized(false, Rtemp);
805 }
806
807 // check for synchronized methods
808 // Must happen AFTER invocation_counter check and stack overflow check,
809 // so method is not locked if overflows.
810 //
811 if (synchronized) {
812 lock_method();
813 } else {
814 // no synchronization necessary
815 #ifdef ASSERT
816 { Label L;
817 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
818 __ tbz(Rtemp, JVM_ACC_SYNCHRONIZED_BIT, L);
819 __ stop("method needs synchronization");
820 __ bind(L);
821 }
822 #endif
823 }
824
825 // start execution
826 #ifdef ASSERT
827 { Label L;
828 __ ldr(Rtemp, Address(FP, frame::interpreter_frame_monitor_block_top_offset * wordSize));
829 __ cmp(Rtemp, Rstack_top);
830 __ b(L, eq);
831 __ stop("broken stack frame setup in interpreter");
832 __ bind(L);
833 }
834 #endif
835 __ check_extended_sp(Rtemp);
836
837 // jvmti/dtrace support
838 __ notify_method_entry();
839 #if R9_IS_SCRATCHED
840 __ restore_method();
841 #endif
842
843 {
844 Label L;
845 __ ldr(Rsig_handler, Address(Rmethod, Method::signature_handler_offset()));
846 __ cbnz(Rsig_handler, L);
847 __ mov(R1, Rmethod);
848 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R1, true);
849 __ ldr(Rsig_handler, Address(Rmethod, Method::signature_handler_offset()));
850 __ bind(L);
851 }
852
853 {
854 Label L;
855 __ ldr(Rnative_code, Address(Rmethod, Method::native_function_offset()));
856 __ cbnz(Rnative_code, L);
857 __ mov(R1, Rmethod);
858 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R1);
859 __ ldr(Rnative_code, Address(Rmethod, Method::native_function_offset()));
860 __ bind(L);
861 }
862
863 // Allocate stack space for arguments
864
865
866 // C functions need aligned stack
867 __ bic(SP, SP, StackAlignmentInBytes - 1);
868 // Multiply by BytesPerLong instead of BytesPerWord, because calling convention
869 // may require empty slots due to long alignment, e.g. func(int, jlong, int, jlong)
870 __ sub(SP, SP, AsmOperand(Rsize_of_params, lsl, LogBytesPerLong));
871
872 #ifdef __ABI_HARD__
873 // Allocate more stack space to accomodate all GP as well as FP registers:
874 // 4 * wordSize
875 // 8 * BytesPerLong
876 int reg_arguments = align_up((4*wordSize) + (8*BytesPerLong), StackAlignmentInBytes);
877 #else
878 // Reserve at least 4 words on the stack for loading
879 // of parameters passed on registers (R0-R3).
880 // See generate_slow_signature_handler().
881 // It is also used for JNIEnv & class additional parameters.
882 int reg_arguments = 4 * wordSize;
883 #endif // __ABI_HARD__
884
885 __ sub(SP, SP, reg_arguments);
886
887
888 // Note: signature handler blows R4 besides all scratch registers.
889 // See AbstractInterpreterGenerator::generate_slow_signature_handler().
890 __ call(Rsig_handler);
891 #if R9_IS_SCRATCHED
892 __ restore_method();
893 #endif
894 __ mov(Rresult_handler, R0);
895
896 // Pass JNIEnv and mirror for static methods
897 {
898 Label L;
899 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
900 __ add(R0, Rthread, in_bytes(JavaThread::jni_environment_offset()));
901 __ tbz(Rtemp, JVM_ACC_STATIC_BIT, L);
902 __ load_mirror(Rtemp, Rmethod, Rtemp);
903 __ add(R1, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
904 __ str(Rtemp, Address(R1, 0));
905 __ bind(L);
906 }
907
908 __ set_last_Java_frame(SP, FP, true, Rtemp);
909
910 // Changing state to _thread_in_native must be the last thing to do
911 // before the jump to native code. At this moment stack must be
912 // safepoint-safe and completely prepared for stack walking.
913 #ifdef ASSERT
914 {
915 Label L;
916 __ ldr_u32(Rtemp, Address(Rthread, JavaThread::thread_state_offset()));
917 __ cmp_32(Rtemp, _thread_in_Java);
918 __ b(L, eq);
919 __ stop("invalid thread state");
920 __ bind(L);
921 }
922 #endif
923
924 // Force all preceding writes to be observed prior to thread state change
925 __ membar(MacroAssembler::StoreStore, Rtemp);
926
927 __ mov(Rtemp, _thread_in_native);
928 __ str(Rtemp, Address(Rthread, JavaThread::thread_state_offset()));
929
930 __ call(Rnative_code);
931 #if R9_IS_SCRATCHED
932 __ restore_method();
933 #endif
934
935 // Set FPSCR/FPCR to a known state
936 if (AlwaysRestoreFPU) {
937 __ restore_default_fp_mode();
938 }
939
940 // Do safepoint check
941 __ mov(Rtemp, _thread_in_native_trans);
942 __ str_32(Rtemp, Address(Rthread, JavaThread::thread_state_offset()));
943
944 // Force this write out before the read below
945 __ membar(MacroAssembler::StoreLoad, Rtemp);
946
947 __ ldr_global_s32(Rtemp, SafepointSynchronize::address_of_state());
948
949 // Protect the return value in the interleaved code: save it to callee-save registers.
950 __ mov(Rsaved_result_lo, R0);
951 __ mov(Rsaved_result_hi, R1);
952 #ifdef __ABI_HARD__
953 // preserve native FP result in a callee-saved register
954 saved_result_fp = D8;
955 __ fcpyd(saved_result_fp, D0);
956 #else
957 saved_result_fp = fnoreg;
958 #endif // __ABI_HARD__
959
960 {
961 __ ldr_u32(R3, Address(Rthread, JavaThread::suspend_flags_offset()));
962 __ cmp(Rtemp, SafepointSynchronize::_not_synchronized);
963 __ cond_cmp(R3, 0, eq);
964
965 __ mov(R0, Rthread, ne);
966 __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), relocInfo::none, ne);
967 #if R9_IS_SCRATCHED
968 __ restore_method();
969 #endif
970 }
971
972 // Perform Native->Java thread transition
973 __ mov(Rtemp, _thread_in_Java);
974 __ str_32(Rtemp, Address(Rthread, JavaThread::thread_state_offset()));
975
976 // Zero handles and last_java_sp
977 __ reset_last_Java_frame(Rtemp);
978 __ ldr(R3, Address(Rthread, JavaThread::active_handles_offset()));
979 __ str_32(__ zero_register(Rtemp), Address(R3, JNIHandleBlock::top_offset_in_bytes()));
980 if (CheckJNICalls) {
981 __ str(__ zero_register(Rtemp), Address(Rthread, JavaThread::pending_jni_exception_check_fn_offset()));
982 }
983
984 // Unbox oop result, e.g. JNIHandles::resolve result if it's an oop.
985 {
986 Label Lnot_oop;
987 // For ARM32, Rresult_handler is -1 for oop result, 0 otherwise.
988 __ cbz(Rresult_handler, Lnot_oop);
989 Register value = Rsaved_result_lo;
990 __ resolve_jobject(value, // value
991 Rtemp, // tmp1
992 R1_tmp); // tmp2
993 // Store resolved result in frame for GC visibility.
994 __ str(value, Address(FP, frame::interpreter_frame_oop_temp_offset * wordSize));
995 __ bind(Lnot_oop);
996 }
997
998
999 // reguard stack if StackOverflow exception happened while in native.
1000 {
1001 __ ldr_u32(Rtemp, Address(Rthread, JavaThread::stack_guard_state_offset()));
1002 __ cmp_32(Rtemp, JavaThread::stack_guard_yellow_reserved_disabled);
1003 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), relocInfo::none, eq);
1004 #if R9_IS_SCRATCHED
1005 __ restore_method();
1006 #endif
1007 }
1008
1009 // check pending exceptions
1010 {
1011 __ ldr(Rtemp, Address(Rthread, Thread::pending_exception_offset()));
1012 __ cmp(Rtemp, 0);
1013 __ mov(Rexception_pc, PC, ne);
1014 __ b(StubRoutines::forward_exception_entry(), ne);
1015 }
1016
1017 if (synchronized) {
1018 // address of first monitor
1019 __ sub(R1, FP, - (frame::interpreter_frame_monitor_block_bottom_offset - frame::interpreter_frame_monitor_size()) * wordSize);
1020 __ unlock_object(R1);
1021 }
1022
1023 // jvmti/dtrace support
1024 // Note: This must happen _after_ handling/throwing any exceptions since
1025 // the exception handler code notifies the runtime of method exits
1026 // too. If this happens before, method entry/exit notifications are
1027 // not properly paired (was bug - gri 11/22/99).
1028 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI, true, Rsaved_result_lo, Rsaved_result_hi, saved_result_fp);
1029
1030 // Restore the result. Oop result is restored from the stack.
1031 __ cmp(Rresult_handler, 0);
1032 __ ldr(R0, Address(FP, frame::interpreter_frame_oop_temp_offset * wordSize), ne);
1033 __ mov(R0, Rsaved_result_lo, eq);
1034 __ mov(R1, Rsaved_result_hi);
1035
1036 #ifdef __ABI_HARD__
1037 // reload native FP result
1038 __ fcpyd(D0, D8);
1039 #endif // __ABI_HARD__
1040
1041 #ifdef ASSERT
1042 if (VerifyOops) {
1043 Label L;
1044 __ cmp(Rresult_handler, 0);
1045 __ b(L, eq);
1046 __ verify_oop(R0);
1047 __ bind(L);
1048 }
1049 #endif // ASSERT
1050
1051 // Restore FP/LR, sender_sp and return
1052 __ mov(Rtemp, FP);
1053 __ ldmia(FP, RegisterSet(FP) | RegisterSet(LR));
1054 __ ldr(SP, Address(Rtemp, frame::interpreter_frame_sender_sp_offset * wordSize));
1055
1056 __ ret();
1057
1058 if (inc_counter) {
1059 // Handle overflow of counter and compile method
1060 __ bind(invocation_counter_overflow);
1061 generate_counter_overflow(continue_after_compile);
1062 }
1063
1064 return entry_point;
1065 }
1066
1067 //
1068 // Generic interpreted method entry to (asm) interpreter
1069 //
1070 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1071 // determine code generation flags
1072 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1073
1074 // Rmethod: Method*
1075 // Rthread: thread
1076 // Rsender_sp: sender sp (could differ from SP if we were called via c2i)
1077 // Rparams: pointer to the last parameter in the stack
1078
1079 address entry_point = __ pc();
1080
1081 const Register RconstMethod = R3;
1082
1083
1084 __ ldr(RconstMethod, Address(Rmethod, Method::const_offset()));
1085
1086 __ ldrh(R2, Address(RconstMethod, ConstMethod::size_of_parameters_offset()));
1087 __ ldrh(R3, Address(RconstMethod, ConstMethod::size_of_locals_offset()));
1088
1089 // setup Rlocals
1090 __ sub(Rlocals, Rparams, wordSize);
1091 __ add(Rlocals, Rlocals, AsmOperand(R2, lsl, Interpreter::logStackElementSize));
1092
1093 __ sub(R3, R3, R2); // number of additional locals
1094
1095
1096 // see if we've got enough room on the stack for locals plus overhead.
1097 generate_stack_overflow_check();
1098
1099 // allocate space for locals
1100 // explicitly initialize locals
1101
1102 // Loop is unrolled 4 times
1103 Label loop;
1104 __ mov(R0, 0);
1105 __ bind(loop);
1106
1107 // #1
1108 __ subs(R3, R3, 1);
1109 __ push(R0, ge);
1110
1111 // #2
1112 __ subs(R3, R3, 1, ge);
1113 __ push(R0, ge);
1114
1115 // #3
1116 __ subs(R3, R3, 1, ge);
1117 __ push(R0, ge);
1118
1119 // #4
1120 __ subs(R3, R3, 1, ge);
1121 __ push(R0, ge);
1122
1123 __ b(loop, gt);
1124
1125 // initialize fixed part of activation frame
1126 generate_fixed_frame(false);
1127
1128 __ restore_dispatch();
1129
1130 // make sure method is not native & not abstract
1131 #ifdef ASSERT
1132 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
1133 {
1134 Label L;
1135 __ tbz(Rtemp, JVM_ACC_NATIVE_BIT, L);
1136 __ stop("tried to execute native method as non-native");
1137 __ bind(L);
1138 }
1139 { Label L;
1140 __ tbz(Rtemp, JVM_ACC_ABSTRACT_BIT, L);
1141 __ stop("tried to execute abstract method in interpreter");
1142 __ bind(L);
1143 }
1144 #endif
1145
1146 // increment invocation count & check for overflow
1147 Label invocation_counter_overflow;
1148 Label profile_method;
1149 Label profile_method_continue;
1150 if (inc_counter) {
1151 if (synchronized) {
1152 // Avoid unlocking method's monitor in case of exception, as it has not
1153 // been locked yet.
1154 __ set_do_not_unlock_if_synchronized(true, Rtemp);
1155 }
1156 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1157 if (ProfileInterpreter) {
1158 __ bind(profile_method_continue);
1159 }
1160 }
1161 Label continue_after_compile;
1162 __ bind(continue_after_compile);
1163
1164 if (inc_counter && synchronized) {
1165 __ set_do_not_unlock_if_synchronized(false, Rtemp);
1166 }
1167 #if R9_IS_SCRATCHED
1168 __ restore_method();
1169 #endif
1170
1171 // check for synchronized methods
1172 // Must happen AFTER invocation_counter check and stack overflow check,
1173 // so method is not locked if overflows.
1174 //
1175 if (synchronized) {
1176 // Allocate monitor and lock method
1177 lock_method();
1178 } else {
1179 // no synchronization necessary
1180 #ifdef ASSERT
1181 { Label L;
1182 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
1183 __ tbz(Rtemp, JVM_ACC_SYNCHRONIZED_BIT, L);
1184 __ stop("method needs synchronization");
1185 __ bind(L);
1186 }
1187 #endif
1188 }
1189
1190 // start execution
1191 #ifdef ASSERT
1192 { Label L;
1193 __ ldr(Rtemp, Address(FP, frame::interpreter_frame_monitor_block_top_offset * wordSize));
1194 __ cmp(Rtemp, Rstack_top);
1195 __ b(L, eq);
1196 __ stop("broken stack frame setup in interpreter");
1197 __ bind(L);
1198 }
1199 #endif
1200 __ check_extended_sp(Rtemp);
1201
1202 // jvmti support
1203 __ notify_method_entry();
1204 #if R9_IS_SCRATCHED
1205 __ restore_method();
1206 #endif
1207
1208 __ dispatch_next(vtos);
1209
1210 // invocation counter overflow
1211 if (inc_counter) {
1212 if (ProfileInterpreter) {
1213 // We have decided to profile this method in the interpreter
1214 __ bind(profile_method);
1215
1216 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1217 __ set_method_data_pointer_for_bcp();
1218
1219 __ b(profile_method_continue);
1220 }
1221
1222 // Handle overflow of counter and compile method
1223 __ bind(invocation_counter_overflow);
1224 generate_counter_overflow(continue_after_compile);
1225 }
1226
1227 return entry_point;
1228 }
1229
1230 //------------------------------------------------------------------------------------------------------------------------
1231 // Exceptions
1232
1233 void TemplateInterpreterGenerator::generate_throw_exception() {
1234 // Entry point in previous activation (i.e., if the caller was interpreted)
1235 Interpreter::_rethrow_exception_entry = __ pc();
1236 // Rexception_obj: exception
1237
1238 // Clear interpreter_frame_last_sp.
1239 __ mov(Rtemp, 0);
1240 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
1241
1242 #if R9_IS_SCRATCHED
1243 __ restore_method();
1244 #endif
1245 __ restore_bcp();
1246 __ restore_dispatch();
1247 __ restore_locals();
1248
1249
1250 // Entry point for exceptions thrown within interpreter code
1251 Interpreter::_throw_exception_entry = __ pc();
1252
1253 // expression stack is undefined here
1254 // Rexception_obj: exception
1255 // Rbcp: exception bcp
1256 __ verify_oop(Rexception_obj);
1257
1258 // expression stack must be empty before entering the VM in case of an exception
1259 __ empty_expression_stack();
1260 // find exception handler address and preserve exception oop
1261 __ mov(R1, Rexception_obj);
1262 __ call_VM(Rexception_obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), R1);
1263 // R0: exception handler entry point
1264 // Rexception_obj: preserved exception oop
1265 // Rbcp: bcp for exception handler
1266 __ push_ptr(Rexception_obj); // push exception which is now the only value on the stack
1267 __ jump(R0); // jump to exception handler (may be _remove_activation_entry!)
1268
1269 // If the exception is not handled in the current frame the frame is removed and
1270 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1271 //
1272 // Note: At this point the bci is still the bxi for the instruction which caused
1273 // the exception and the expression stack is empty. Thus, for any VM calls
1274 // at this point, GC will find a legal oop map (with empty expression stack).
1275
1276 // In current activation
1277 // tos: exception
1278 // Rbcp: exception bcp
1279
1280 //
1281 // JVMTI PopFrame support
1282 //
1283 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1284
1285
1286 __ empty_expression_stack();
1287
1288 // Set the popframe_processing bit in _popframe_condition indicating that we are
1289 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1290 // popframe handling cycles.
1291
1292 __ ldr_s32(Rtemp, Address(Rthread, JavaThread::popframe_condition_offset()));
1293 __ orr(Rtemp, Rtemp, (unsigned)JavaThread::popframe_processing_bit);
1294 __ str_32(Rtemp, Address(Rthread, JavaThread::popframe_condition_offset()));
1295
1296 {
1297 // Check to see whether we are returning to a deoptimized frame.
1298 // (The PopFrame call ensures that the caller of the popped frame is
1299 // either interpreted or compiled and deoptimizes it if compiled.)
1300 // In this case, we can't call dispatch_next() after the frame is
1301 // popped, but instead must save the incoming arguments and restore
1302 // them after deoptimization has occurred.
1303 //
1304 // Note that we don't compare the return PC against the
1305 // deoptimization blob's unpack entry because of the presence of
1306 // adapter frames in C2.
1307 Label caller_not_deoptimized;
1308 __ ldr(R0, Address(FP, frame::return_addr_offset * wordSize));
1309 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), R0);
1310 __ cbnz_32(R0, caller_not_deoptimized);
1311
1312 // Compute size of arguments for saving when returning to deoptimized caller
1313 __ restore_method();
1314 __ ldr(R0, Address(Rmethod, Method::const_offset()));
1315 __ ldrh(R0, Address(R0, ConstMethod::size_of_parameters_offset()));
1316
1317 __ logical_shift_left(R1, R0, Interpreter::logStackElementSize);
1318 // Save these arguments
1319 __ restore_locals();
1320 __ sub(R2, Rlocals, R1);
1321 __ add(R2, R2, wordSize);
1322 __ mov(R0, Rthread);
1323 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R0, R1, R2);
1324
1325 __ remove_activation(vtos, LR,
1326 /* throw_monitor_exception */ false,
1327 /* install_monitor_exception */ false,
1328 /* notify_jvmdi */ false);
1329
1330 // Inform deoptimization that it is responsible for restoring these arguments
1331 __ mov(Rtemp, JavaThread::popframe_force_deopt_reexecution_bit);
1332 __ str_32(Rtemp, Address(Rthread, JavaThread::popframe_condition_offset()));
1333
1334 // Continue in deoptimization handler
1335 __ ret();
1336
1337 __ bind(caller_not_deoptimized);
1338 }
1339
1340 __ remove_activation(vtos, R4,
1341 /* throw_monitor_exception */ false,
1342 /* install_monitor_exception */ false,
1343 /* notify_jvmdi */ false);
1344
1345 // Finish with popframe handling
1346 // A previous I2C followed by a deoptimization might have moved the
1347 // outgoing arguments further up the stack. PopFrame expects the
1348 // mutations to those outgoing arguments to be preserved and other
1349 // constraints basically require this frame to look exactly as
1350 // though it had previously invoked an interpreted activation with
1351 // no space between the top of the expression stack (current
1352 // last_sp) and the top of stack. Rather than force deopt to
1353 // maintain this kind of invariant all the time we call a small
1354 // fixup routine to move the mutated arguments onto the top of our
1355 // expression stack if necessary.
1356 __ mov(R1, SP);
1357 __ ldr(R2, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
1358 // PC must point into interpreter here
1359 __ set_last_Java_frame(SP, FP, true, Rtemp);
1360 __ mov(R0, Rthread);
1361 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), R0, R1, R2);
1362 __ reset_last_Java_frame(Rtemp);
1363
1364 // Restore the last_sp and null it out
1365 __ ldr(SP, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
1366 __ mov(Rtemp, (int)NULL_WORD);
1367 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
1368
1369 __ restore_bcp();
1370 __ restore_dispatch();
1371 __ restore_locals();
1372 __ restore_method();
1373
1374 // The method data pointer was incremented already during
1375 // call profiling. We have to restore the mdp for the current bcp.
1376 if (ProfileInterpreter) {
1377 __ set_method_data_pointer_for_bcp();
1378 }
1379
1380 // Clear the popframe condition flag
1381 assert(JavaThread::popframe_inactive == 0, "adjust this code");
1382 __ str_32(__ zero_register(Rtemp), Address(Rthread, JavaThread::popframe_condition_offset()));
1383
1384 #if INCLUDE_JVMTI
1385 {
1386 Label L_done;
1387
1388 __ ldrb(Rtemp, Address(Rbcp, 0));
1389 __ cmp(Rtemp, Bytecodes::_invokestatic);
1390 __ b(L_done, ne);
1391
1392 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1393 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1394
1395 // get local0
1396 __ ldr(R1, Address(Rlocals, 0));
1397 __ mov(R2, Rmethod);
1398 __ mov(R3, Rbcp);
1399 __ call_VM(R0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R1, R2, R3);
1400
1401 __ cbz(R0, L_done);
1402
1403 __ str(R0, Address(Rstack_top));
1404 __ bind(L_done);
1405 }
1406 #endif // INCLUDE_JVMTI
1407
1408 __ dispatch_next(vtos);
1409 // end of PopFrame support
1410
1411 Interpreter::_remove_activation_entry = __ pc();
1412
1413 // preserve exception over this code sequence
1414 __ pop_ptr(R0_tos);
1415 __ str(R0_tos, Address(Rthread, JavaThread::vm_result_offset()));
1416 // remove the activation (without doing throws on illegalMonitorExceptions)
1417 __ remove_activation(vtos, Rexception_pc, false, true, false);
1418 // restore exception
1419 __ get_vm_result(Rexception_obj, Rtemp);
1420
1421 // Inbetween activations - previous activation type unknown yet
1422 // compute continuation point - the continuation point expects
1423 // the following registers set up:
1424 //
1425 // Rexception_obj: exception
1426 // Rexception_pc: return address/pc that threw exception
1427 // SP: expression stack of caller
1428 // FP: frame pointer of caller
1429 __ mov(c_rarg0, Rthread);
1430 __ mov(c_rarg1, Rexception_pc);
1431 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), c_rarg0, c_rarg1);
1432 // Note that an "issuing PC" is actually the next PC after the call
1433
1434 __ jump(R0); // jump to exception handler of caller
1435 }
1436
1437
1438 //
1439 // JVMTI ForceEarlyReturn support
1440 //
1441 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1442 address entry = __ pc();
1443
1444
1445 __ restore_bcp();
1446 __ restore_dispatch();
1447 __ restore_locals();
1448
1449 __ empty_expression_stack();
1450
1451 __ load_earlyret_value(state);
1452
1453 // Clear the earlyret state
1454 __ ldr(Rtemp, Address(Rthread, JavaThread::jvmti_thread_state_offset()));
1455
1456 assert(JvmtiThreadState::earlyret_inactive == 0, "adjust this code");
1457 __ str_32(__ zero_register(R2), Address(Rtemp, JvmtiThreadState::earlyret_state_offset()));
1458
1459 __ remove_activation(state, LR,
1460 false, /* throw_monitor_exception */
1461 false, /* install_monitor_exception */
1462 true); /* notify_jvmdi */
1463
1464 // According to interpreter calling conventions, result is returned in R0/R1,
1465 // so ftos (S0) and dtos (D0) are moved to R0/R1.
1466 // This conversion should be done after remove_activation, as it uses
1467 // push(state) & pop(state) to preserve return value.
1468 __ convert_tos_to_retval(state);
1469 __ ret();
1470
1471 return entry;
1472 } // end of ForceEarlyReturn support
1473
1474
1475 //------------------------------------------------------------------------------------------------------------------------
1476 // Helper for vtos entry point generation
1477
1478 void TemplateInterpreterGenerator::set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) {
1479 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1480 Label L;
1481
1482 #ifdef __SOFTFP__
1483 dep = __ pc(); // fall through
1484 #else
1485 fep = __ pc(); __ push(ftos); __ b(L);
1486 dep = __ pc(); __ push(dtos); __ b(L);
1487 #endif // __SOFTFP__
1488
1489 lep = __ pc(); __ push(ltos); __ b(L);
1490
1491 if (VerifyOops) { // can't share atos entry if VerifyOops
1492 aep = __ pc(); __ push(atos); __ b(L);
1493 } else {
1494 aep = __ pc(); // fall through
1495 }
1496
1497 #ifdef __SOFTFP__
1498 fep = __ pc(); // fall through
1499 #endif // __SOFTFP__
1500
1501 bep = cep = sep = // fall through
1502 iep = __ pc(); __ push(itos); // fall through
1503 vep = __ pc(); __ bind(L); // fall through
1504 generate_and_dispatch(t);
1505 }
1506
1507 //------------------------------------------------------------------------------------------------------------------------
1508
1509 // Non-product code
1510 #ifndef PRODUCT
1511 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1512 address entry = __ pc();
1513
1514 // prepare expression stack
1515 __ push(state); // save tosca
1516
1517 // pass tosca registers as arguments
1518 __ mov(R2, R0_tos);
1519 __ mov(R3, R1_tos_hi);
1520 __ mov(R1, LR); // save return address
1521
1522 // call tracer
1523 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), R1, R2, R3);
1524
1525 __ mov(LR, R0); // restore return address
1526 __ pop(state); // restore tosca
1527
1528 // return
1529 __ ret();
1530
1531 return entry;
1532 }
1533
1534
1535 void TemplateInterpreterGenerator::count_bytecode() {
1536 __ inc_global_counter((address) &BytecodeCounter::_counter_value, 0, Rtemp, R2_tmp, true);
1537 }
1538
1539
1540 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1541 __ inc_global_counter((address)&BytecodeHistogram::_counters[0], sizeof(BytecodeHistogram::_counters[0]) * t->bytecode(), Rtemp, R2_tmp, true);
1542 }
1543
1544
1545 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1546 const Register Rindex_addr = R2_tmp;
1547 Label Lcontinue;
1548 InlinedAddress Lcounters((address)BytecodePairHistogram::_counters);
1549 InlinedAddress Lindex((address)&BytecodePairHistogram::_index);
1550 const Register Rcounters_addr = R2_tmp;
1551 const Register Rindex = R4_tmp;
1552
1553 // calculate new index for counter:
1554 // index = (_index >> log2_number_of_codes) | (bytecode << log2_number_of_codes).
1555 // (_index >> log2_number_of_codes) is previous bytecode
1556
1557 __ ldr_literal(Rindex_addr, Lindex);
1558 __ ldr_s32(Rindex, Address(Rindex_addr));
1559 __ mov_slow(Rtemp, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1560 __ orr(Rindex, Rtemp, AsmOperand(Rindex, lsr, BytecodePairHistogram::log2_number_of_codes));
1561 __ str_32(Rindex, Address(Rindex_addr));
1562
1563 // Rindex (R4) contains index of counter
1564
1565 __ ldr_literal(Rcounters_addr, Lcounters);
1566 __ ldr_s32(Rtemp, Address::indexed_32(Rcounters_addr, Rindex));
1567 __ adds_32(Rtemp, Rtemp, 1);
1568 __ b(Lcontinue, mi); // avoid overflow
1569 __ str_32(Rtemp, Address::indexed_32(Rcounters_addr, Rindex));
1570
1571 __ b(Lcontinue);
1572
1573 __ bind_literal(Lindex);
1574 __ bind_literal(Lcounters);
1575
1576 __ bind(Lcontinue);
1577 }
1578
1579
1580 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1581 // Call a little run-time stub to avoid blow-up for each bytecode.
1582 // The run-time runtime saves the right registers, depending on
1583 // the tosca in-state for the given template.
1584 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1585 "entry must have been generated");
1586 address trace_entry = Interpreter::trace_code(t->tos_in());
1587 __ call(trace_entry, relocInfo::none);
1588 }
1589
1590
1591 void TemplateInterpreterGenerator::stop_interpreter_at() {
1592 Label Lcontinue;
1593 const Register stop_at = R2_tmp;
1594
1595 __ ldr_global_s32(Rtemp, (address) &BytecodeCounter::_counter_value);
1596 __ mov_slow(stop_at, StopInterpreterAt);
1597
1598 // test bytecode counter
1599 __ cmp(Rtemp, stop_at);
1600 __ b(Lcontinue, ne);
1601
1602 __ trace_state("stop_interpreter_at");
1603 __ breakpoint();
1604
1605 __ bind(Lcontinue);
1606 }
1607 #endif // !PRODUCT