1 /*
2 * Copyright (c) 2003, 2017, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2013, Red Hat Inc. 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 #include "asm/macroAssembler.hpp"
28 #include "interpreter/bytecodeHistogram.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterGenerator.hpp"
31 #include "interpreter/interpreterRuntime.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "interpreter/bytecodeTracer.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 "runtime/arguments.hpp"
41 #include "runtime/deoptimization.hpp"
42 #include "runtime/frame.inline.hpp"
43 #include "runtime/sharedRuntime.hpp"
44 #include "runtime/stubRoutines.hpp"
45 #include "runtime/synchronizer.hpp"
46 #include "runtime/timer.hpp"
47 #include "runtime/vframeArray.hpp"
48 #include "utilities/debug.hpp"
49 #include <sys/types.h>
50
51 #ifndef PRODUCT
52 #include "oops/method.hpp"
53 #endif // !PRODUCT
54
55 #define __ _masm->
56
57 #ifndef CC_INTERP
58
59 //-----------------------------------------------------------------------------
60
61 extern "C" void entry(CodeBuffer*);
62
63 //-----------------------------------------------------------------------------
64
65 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
66 address entry = __ pc();
67
68 #ifdef ASSERT
69 {
70 Label L;
71 __ ldr(rscratch1, Address(rfp,
72 frame::interpreter_frame_monitor_block_top_offset *
73 wordSize));
74 __ mov(rscratch2, sp);
75 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
76 // grows negative)
77 __ br(Assembler::HS, L); // check if frame is complete
78 __ stop ("interpreter frame not set up");
79 __ bind(L);
80 }
81 #endif // ASSERT
82 // Restore bcp under the assumption that the current frame is still
83 // interpreted
84 __ restore_bcp();
85
86 // expression stack must be empty before entering the VM if an
87 // exception happened
88 __ empty_expression_stack();
89 // throw exception
90 __ call_VM(noreg,
91 CAST_FROM_FN_PTR(address,
92 InterpreterRuntime::throw_StackOverflowError));
93 return entry;
94 }
95
96 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
97 const char* name) {
98 address entry = __ pc();
99 // expression stack must be empty before entering the VM if an
100 // exception happened
101 __ empty_expression_stack();
102 // setup parameters
103 // ??? convention: expect aberrant index in register r1
104 __ movw(c_rarg2, r1);
105 __ mov(c_rarg1, (address)name);
106 __ call_VM(noreg,
107 CAST_FROM_FN_PTR(address,
108 InterpreterRuntime::
109 throw_ArrayIndexOutOfBoundsException),
110 c_rarg1, c_rarg2);
111 return entry;
112 }
113
114 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
115 address entry = __ pc();
116
117 // object is at TOS
118 __ pop(c_rarg1);
119
120 // expression stack must be empty before entering the VM if an
121 // exception happened
122 __ empty_expression_stack();
123
124 __ call_VM(noreg,
125 CAST_FROM_FN_PTR(address,
126 InterpreterRuntime::
127 throw_ClassCastException),
128 c_rarg1);
129 return entry;
130 }
131
132 address TemplateInterpreterGenerator::generate_exception_handler_common(
133 const char* name, const char* message, bool pass_oop) {
134 assert(!pass_oop || message == NULL, "either oop or message but not both");
135 address entry = __ pc();
136 if (pass_oop) {
137 // object is at TOS
138 __ pop(c_rarg2);
139 }
140 // expression stack must be empty before entering the VM if an
141 // exception happened
142 __ empty_expression_stack();
143 // setup parameters
144 __ lea(c_rarg1, Address((address)name));
145 if (pass_oop) {
146 __ call_VM(r0, CAST_FROM_FN_PTR(address,
147 InterpreterRuntime::
148 create_klass_exception),
149 c_rarg1, c_rarg2);
150 } else {
151 // kind of lame ExternalAddress can't take NULL because
152 // external_word_Relocation will assert.
153 if (message != NULL) {
154 __ lea(c_rarg2, Address((address)message));
155 } else {
156 __ mov(c_rarg2, NULL_WORD);
157 }
158 __ call_VM(r0,
159 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
160 c_rarg1, c_rarg2);
161 }
162 // throw exception
163 __ b(address(Interpreter::throw_exception_entry()));
164 return entry;
165 }
166
167 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
168 address entry = __ pc();
169 // NULL last_sp until next java call
170 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
171 __ dispatch_next(state);
172 return entry;
173 }
174
175 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
176 address entry = __ pc();
177
178 // Restore stack bottom in case i2c adjusted stack
179 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
180 // and NULL it as marker that esp is now tos until next java call
181 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
182 __ restore_bcp();
183 __ restore_locals();
184 __ restore_constant_pool_cache();
185 __ get_method(rmethod);
186
187 // Pop N words from the stack
188 __ get_cache_and_index_at_bcp(r1, r2, 1, index_size);
189 __ ldr(r1, Address(r1, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
190 __ andr(r1, r1, ConstantPoolCacheEntry::parameter_size_mask);
191
192 __ add(esp, esp, r1, Assembler::LSL, 3);
193
194 // Restore machine SP
195 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
196 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
197 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
198 __ ldr(rscratch2,
199 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
200 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
201 __ andr(sp, rscratch1, -16);
202
203 __ get_dispatch();
204 __ dispatch_next(state, step);
205
206 return entry;
207 }
208
209 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
210 int step) {
211 address entry = __ pc();
212 __ restore_bcp();
213 __ restore_locals();
214 __ restore_constant_pool_cache();
215 __ get_method(rmethod);
216 __ get_dispatch();
217
218 // Calculate stack limit
219 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
220 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
221 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size()
222 + (EnableInvokeDynamic ? 2 : 0));
223 __ ldr(rscratch2,
224 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
225 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
226 __ andr(sp, rscratch1, -16);
227
228 // Restore expression stack pointer
229 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
230 // NULL last_sp until next java call
231 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
232
233 // handle exceptions
234 {
235 Label L;
236 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
237 __ cbz(rscratch1, L);
238 __ call_VM(noreg,
239 CAST_FROM_FN_PTR(address,
240 InterpreterRuntime::throw_pending_exception));
241 __ should_not_reach_here();
242 __ bind(L);
243 }
244
245 __ dispatch_next(state, step);
246 return entry;
247 }
248
249
250 int AbstractInterpreter::BasicType_as_index(BasicType type) {
251 int i = 0;
252 switch (type) {
253 case T_BOOLEAN: i = 0; break;
254 case T_CHAR : i = 1; break;
255 case T_BYTE : i = 2; break;
256 case T_SHORT : i = 3; break;
257 case T_INT : i = 4; break;
258 case T_LONG : i = 5; break;
259 case T_VOID : i = 6; break;
260 case T_FLOAT : i = 7; break;
261 case T_DOUBLE : i = 8; break;
262 case T_OBJECT : i = 9; break;
263 case T_ARRAY : i = 9; break;
264 default : ShouldNotReachHere();
265 }
266 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
267 "index out of bounds");
268 return i;
269 }
270
271
272 address TemplateInterpreterGenerator::generate_result_handler_for(
273 BasicType type) {
274 address entry = __ pc();
275 switch (type) {
276 case T_BOOLEAN: __ c2bool(r0); break;
277 case T_CHAR : __ uxth(r0, r0); break;
278 case T_BYTE : __ sxtb(r0, r0); break;
279 case T_SHORT : __ sxth(r0, r0); break;
280 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
281 case T_LONG : /* nothing to do */ break;
282 case T_VOID : /* nothing to do */ break;
283 case T_FLOAT : /* nothing to do */ break;
284 case T_DOUBLE : /* nothing to do */ break;
285 case T_OBJECT :
286 // retrieve result from frame
287 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
288 // and verify it
289 __ verify_oop(r0);
290 break;
291 default : ShouldNotReachHere();
292 }
293 __ ret(lr); // return from result handler
294 return entry;
295 }
296
297 address TemplateInterpreterGenerator::generate_safept_entry_for(
298 TosState state,
299 address runtime_entry) {
300 address entry = __ pc();
301 __ push(state);
302 __ call_VM(noreg, runtime_entry);
303 __ membar(Assembler::AnyAny);
304 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
305 return entry;
306 }
307
308 // Helpers for commoning out cases in the various type of method entries.
309 //
310
311
312 // increment invocation count & check for overflow
313 //
314 // Note: checking for negative value instead of overflow
315 // so we have a 'sticky' overflow test
316 //
317 // rmethod: method
318 //
319 void InterpreterGenerator::generate_counter_incr(
320 Label* overflow,
321 Label* profile_method,
322 Label* profile_method_continue) {
323 Label done;
324 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
325 if (TieredCompilation) {
326 int increment = InvocationCounter::count_increment;
327 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
328 Label no_mdo;
329 if (ProfileInterpreter) {
330 // Are we profiling?
331 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
332 __ cbz(r0, no_mdo);
333 // Increment counter in the MDO
334 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
335 in_bytes(InvocationCounter::counter_offset()));
336 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
337 __ b(done);
338 }
339 __ bind(no_mdo);
340 // Increment counter in MethodCounters
341 const Address invocation_counter(rscratch2,
342 MethodCounters::invocation_counter_offset() +
343 InvocationCounter::counter_offset());
344 __ get_method_counters(rmethod, rscratch2, done);
345 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
346 __ bind(done);
347 } else {
348 const Address backedge_counter(rscratch2,
349 MethodCounters::backedge_counter_offset() +
350 InvocationCounter::counter_offset());
351 const Address invocation_counter(rscratch2,
352 MethodCounters::invocation_counter_offset() +
353 InvocationCounter::counter_offset());
354
355 __ get_method_counters(rmethod, rscratch2, done);
356
357 if (ProfileInterpreter) { // %%% Merge this into MethodData*
358 __ ldrw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
359 __ addw(r1, r1, 1);
360 __ strw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
361 }
362 // Update standard invocation counters
363 __ ldrw(r1, invocation_counter);
364 __ ldrw(r0, backedge_counter);
365
366 __ addw(r1, r1, InvocationCounter::count_increment);
367 __ andw(r0, r0, InvocationCounter::count_mask_value);
368
369 __ strw(r1, invocation_counter);
370 __ addw(r0, r0, r1); // add both counters
371
372 // profile_method is non-null only for interpreted method so
373 // profile_method != NULL == !native_call
374
375 if (ProfileInterpreter && profile_method != NULL) {
376 // Test to see if we should create a method data oop
377 unsigned long offset;
378 __ adrp(rscratch2, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit),
379 offset);
380 __ ldrw(rscratch2, Address(rscratch2, offset));
381 __ cmp(r0, rscratch2);
382 __ br(Assembler::LT, *profile_method_continue);
383
384 // if no method data exists, go to profile_method
385 __ test_method_data_pointer(rscratch2, *profile_method);
386 }
387
388 {
389 unsigned long offset;
390 __ adrp(rscratch2,
391 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit),
392 offset);
393 __ ldrw(rscratch2, Address(rscratch2, offset));
394 __ cmpw(r0, rscratch2);
395 __ br(Assembler::HS, *overflow);
396 }
397 __ bind(done);
398 }
399 }
400
401 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
402
403 // Asm interpreter on entry
404 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
405 // Everything as it was on entry
406
407 // InterpreterRuntime::frequency_counter_overflow takes two
408 // arguments, the first (thread) is passed by call_VM, the second
409 // indicates if the counter overflow occurs at a backwards branch
410 // (NULL bcp). We pass zero for it. The call returns the address
411 // of the verified entry point for the method or NULL if the
412 // compilation did not complete (either went background or bailed
413 // out).
414 __ mov(c_rarg1, 0);
415 __ call_VM(noreg,
416 CAST_FROM_FN_PTR(address,
417 InterpreterRuntime::frequency_counter_overflow),
418 c_rarg1);
419
420 __ b(*do_continue);
421 }
422
423 // See if we've got enough room on the stack for locals plus overhead.
424 // The expression stack grows down incrementally, so the normal guard
425 // page mechanism will work for that.
426 //
427 // NOTE: Since the additional locals are also always pushed (wasn't
428 // obvious in generate_method_entry) so the guard should work for them
429 // too.
430 //
431 // Args:
432 // r3: number of additional locals this frame needs (what we must check)
433 // rmethod: Method*
434 //
435 // Kills:
436 // r0
437 void InterpreterGenerator::generate_stack_overflow_check(void) {
438
439 // monitor entry size: see picture of stack set
440 // (generate_method_entry) and frame_amd64.hpp
441 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
442
443 // total overhead size: entry_size + (saved rbp through expr stack
444 // bottom). be sure to change this if you add/subtract anything
445 // to/from the overhead area
446 const int overhead_size =
447 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
448
449 const int page_size = os::vm_page_size();
450
451 Label after_frame_check;
452
453 // see if the frame is greater than one page in size. If so,
454 // then we need to verify there is enough stack space remaining
455 // for the additional locals.
456 //
457 // Note that we use SUBS rather than CMP here because the immediate
458 // field of this instruction may overflow. SUBS can cope with this
459 // because it is a macro that will expand to some number of MOV
460 // instructions and a register operation.
461 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
462 __ br(Assembler::LS, after_frame_check);
463
464 // compute rsp as if this were going to be the last frame on
465 // the stack before the red zone
466
467 const Address stack_base(rthread, Thread::stack_base_offset());
468 const Address stack_size(rthread, Thread::stack_size_offset());
469
470 // locals + overhead, in bytes
471 __ mov(r0, overhead_size);
472 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
473
474 __ ldr(rscratch1, stack_base);
475 __ ldr(rscratch2, stack_size);
476
477 #ifdef ASSERT
478 Label stack_base_okay, stack_size_okay;
479 // verify that thread stack base is non-zero
480 __ cbnz(rscratch1, stack_base_okay);
481 __ stop("stack base is zero");
482 __ bind(stack_base_okay);
483 // verify that thread stack size is non-zero
484 __ cbnz(rscratch2, stack_size_okay);
485 __ stop("stack size is zero");
486 __ bind(stack_size_okay);
487 #endif
488
489 // Add stack base to locals and subtract stack size
490 __ sub(rscratch1, rscratch1, rscratch2); // Stack limit
491 __ add(r0, r0, rscratch1);
492
493 // Use the maximum number of pages we might bang.
494 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
495 (StackRedPages+StackYellowPages);
496
497 // add in the red and yellow zone sizes
498 __ add(r0, r0, max_pages * page_size * 2);
499
500 // check against the current stack bottom
501 __ cmp(sp, r0);
502 __ br(Assembler::HI, after_frame_check);
503
504 // Remove the incoming args, peeling the machine SP back to where it
505 // was in the caller. This is not strictly necessary, but unless we
506 // do so the stack frame may have a garbage FP; this ensures a
507 // correct call stack that we can always unwind. The ANDR should be
508 // unnecessary because the sender SP in r13 is always aligned, but
509 // it doesn't hurt.
510 __ andr(sp, r13, -16);
511
512 // Note: the restored frame is not necessarily interpreted.
513 // Use the shared runtime version of the StackOverflowError.
514 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
515 __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
516
517 // all done with frame size check
518 __ bind(after_frame_check);
519 }
520
521 // Allocate monitor and lock method (asm interpreter)
522 //
523 // Args:
524 // rmethod: Method*
525 // rlocals: locals
526 //
527 // Kills:
528 // r0
529 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
530 // rscratch1, rscratch2 (scratch regs)
531 void InterpreterGenerator::lock_method(void) {
532 // synchronize method
533 const Address access_flags(rmethod, Method::access_flags_offset());
534 const Address monitor_block_top(
535 rfp,
536 frame::interpreter_frame_monitor_block_top_offset * wordSize);
537 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
538
539 #ifdef ASSERT
540 {
541 Label L;
542 __ ldrw(r0, access_flags);
543 __ tst(r0, JVM_ACC_SYNCHRONIZED);
544 __ br(Assembler::NE, L);
545 __ stop("method doesn't need synchronization");
546 __ bind(L);
547 }
548 #endif // ASSERT
549
550 // get synchronization object
551 {
552 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
553 Label done;
554 __ ldrw(r0, access_flags);
555 __ tst(r0, JVM_ACC_STATIC);
556 // get receiver (assume this is frequent case)
557 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
558 __ br(Assembler::EQ, done);
559 __ ldr(r0, Address(rmethod, Method::const_offset()));
560 __ ldr(r0, Address(r0, ConstMethod::constants_offset()));
561 __ ldr(r0, Address(r0,
562 ConstantPool::pool_holder_offset_in_bytes()));
563 __ ldr(r0, Address(r0, mirror_offset));
564
565 #ifdef ASSERT
566 {
567 Label L;
568 __ cbnz(r0, L);
569 __ stop("synchronization object is NULL");
570 __ bind(L);
571 }
572 #endif // ASSERT
573
574 __ bind(done);
575 }
576
577 // add space for monitor & lock
578 __ sub(sp, sp, entry_size); // add space for a monitor entry
579 __ sub(esp, esp, entry_size);
580 __ mov(rscratch1, esp);
581 __ str(rscratch1, monitor_block_top); // set new monitor block top
582 // store object
583 __ str(r0, Address(esp, BasicObjectLock::obj_offset_in_bytes()));
584 __ mov(c_rarg1, esp); // object address
585 __ lock_object(c_rarg1);
586 }
587
588 // Generate a fixed interpreter frame. This is identical setup for
589 // interpreted methods and for native methods hence the shared code.
590 //
591 // Args:
592 // lr: return address
593 // rmethod: Method*
594 // rlocals: pointer to locals
595 // rcpool: cp cache
596 // stack_pointer: previous sp
597 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
598 // initialize fixed part of activation frame
599 if (native_call) {
600 __ sub(esp, sp, 12 * wordSize);
601 __ mov(rbcp, zr);
602 __ stp(esp, zr, Address(__ pre(sp, -12 * wordSize)));
603 // add 2 zero-initialized slots for native calls
604 __ stp(zr, zr, Address(sp, 10 * wordSize));
605 } else {
606 __ sub(esp, sp, 10 * wordSize);
607 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
608 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
609 __ stp(esp, rbcp, Address(__ pre(sp, -10 * wordSize)));
610 }
611
612 if (ProfileInterpreter) {
613 Label method_data_continue;
614 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
615 __ cbz(rscratch1, method_data_continue);
616 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
617 __ bind(method_data_continue);
618 __ stp(rscratch1, rmethod, Address(sp, 4 * wordSize)); // save Method* and mdp (method data pointer)
619 } else {
620 __ stp(zr, rmethod, Address(sp, 4 * wordSize)); // save Method* (no mdp)
621 }
622
623 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
624 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
625 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset_in_bytes()));
626 __ stp(rlocals, rcpool, Address(sp, 2 * wordSize));
627
628 __ stp(rfp, lr, Address(sp, 8 * wordSize));
629 __ lea(rfp, Address(sp, 8 * wordSize));
630
631 // set sender sp
632 // leave last_sp as null
633 __ stp(zr, r13, Address(sp, 6 * wordSize));
634
635 // Move SP out of the way
636 if (! native_call) {
637 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
638 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
639 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size()
640 + (EnableInvokeDynamic ? 2 : 0));
641 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
642 __ andr(sp, rscratch1, -16);
643 }
644 }
645
646 // End of helpers
647
648 // Various method entries
649 //------------------------------------------------------------------------------------------------------------------------
650 //
651 //
652
653 // Call an accessor method (assuming it is resolved, otherwise drop
654 // into vanilla (slow path) entry
655 address InterpreterGenerator::generate_accessor_entry(void) {
656 return NULL;
657 }
658
659 // Method entry for java.lang.ref.Reference.get.
660 address InterpreterGenerator::generate_Reference_get_entry(void) {
661 #if INCLUDE_ALL_GCS
662 // Code: _aload_0, _getfield, _areturn
663 // parameter size = 1
664 //
665 // The code that gets generated by this routine is split into 2 parts:
666 // 1. The "intrinsified" code for G1 (or any SATB based GC),
667 // 2. The slow path - which is an expansion of the regular method entry.
668 //
669 // Notes:-
670 // * In the G1 code we do not check whether we need to block for
671 // a safepoint. If G1 is enabled then we must execute the specialized
672 // code for Reference.get (except when the Reference object is null)
673 // so that we can log the value in the referent field with an SATB
674 // update buffer.
675 // If the code for the getfield template is modified so that the
676 // G1 pre-barrier code is executed when the current method is
677 // Reference.get() then going through the normal method entry
678 // will be fine.
679 // * The G1 code can, however, check the receiver object (the instance
680 // of java.lang.Reference) and jump to the slow path if null. If the
681 // Reference object is null then we obviously cannot fetch the referent
682 // and so we don't need to call the G1 pre-barrier. Thus we can use the
683 // regular method entry code to generate the NPE.
684 //
685 // This code is based on generate_accessor_entry.
686 //
687 // rmethod: Method*
688 // r13: senderSP must preserve for slow path, set SP to it on fast path
689
690 address entry = __ pc();
691
692 const int referent_offset = java_lang_ref_Reference::referent_offset;
693 guarantee(referent_offset > 0, "referent offset not initialized");
694
695 if (UseG1GC || (UseShenandoahGC && ShenandoahSATBBarrier)) {
696 Label slow_path;
697 const Register local_0 = c_rarg0;
698 // Check if local 0 != NULL
699 // If the receiver is null then it is OK to jump to the slow path.
700 __ ldr(local_0, Address(esp, 0));
701 __ mov(r19, r13); // First call-saved register
702 __ cbz(local_0, slow_path);
703
704 // Load the value of the referent field.
705 const Address field_address(local_0, referent_offset);
706 __ load_heap_oop(local_0, field_address);
707
708 __ mov(r19, r13); // Move senderSP to a callee-saved register
709 // Generate the G1 pre-barrier code to log the value of
710 // the referent field in an SATB buffer.
711 __ enter(); // g1_write may call runtime
712 __ g1_write_barrier_pre(noreg /* obj */,
713 local_0 /* pre_val */,
714 rthread /* thread */,
715 rscratch2 /* tmp */,
716 true /* tosca_live */,
717 true /* expand_call */);
718 __ leave();
719 // areturn
720 __ andr(sp, r19, -16); // done with stack
721 __ ret(lr);
722
723 // generate a vanilla interpreter entry as the slow path
724 __ bind(slow_path);
725 (void) generate_normal_entry(false);
726
727 return entry;
728 }
729 #endif // INCLUDE_ALL_GCS
730
731 // If G1 is not enabled then attempt to go through the accessor entry point
732 // Reference.get is an accessor
733 return generate_accessor_entry();
734 }
735
736 /**
737 * Method entry for static native methods:
738 * int java.util.zip.CRC32.update(int crc, int b)
739 */
740 address InterpreterGenerator::generate_CRC32_update_entry() {
741 if (UseCRC32Intrinsics) {
742 address entry = __ pc();
743
744 // rmethod: Method*
745 // r13: senderSP must preserved for slow path
746 // esp: args
747
748 Label slow_path;
749 // If we need a safepoint check, generate full interpreter entry.
750 ExternalAddress state(SafepointSynchronize::address_of_state());
751 unsigned long offset;
752 __ adrp(rscratch1, ExternalAddress(SafepointSynchronize::address_of_state()), offset);
753 __ ldrw(rscratch1, Address(rscratch1, offset));
754 assert(SafepointSynchronize::_not_synchronized == 0, "rewrite this code");
755 __ cbnz(rscratch1, slow_path);
756
757 // We don't generate local frame and don't align stack because
758 // we call stub code and there is no safepoint on this path.
759
760 // Load parameters
761 const Register crc = c_rarg0; // crc
762 const Register val = c_rarg1; // source java byte value
763 const Register tbl = c_rarg2; // scratch
764
765 // Arguments are reversed on java expression stack
766 __ ldrw(val, Address(esp, 0)); // byte value
767 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
768
769 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
770 __ add(tbl, tbl, offset);
771
772 __ ornw(crc, zr, crc); // ~crc
773 __ update_byte_crc32(crc, val, tbl);
774 __ ornw(crc, zr, crc); // ~crc
775
776 // result in c_rarg0
777
778 __ andr(sp, r13, -16);
779 __ ret(lr);
780
781 // generate a vanilla native entry as the slow path
782 __ bind(slow_path);
783
784 (void) generate_native_entry(false);
785
786 return entry;
787 }
788 return generate_native_entry(false);
789 }
790
791 /**
792 * Method entry for static native methods:
793 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
794 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
795 */
796 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
797 if (UseCRC32Intrinsics) {
798 address entry = __ pc();
799
800 // rmethod,: Method*
801 // r13: senderSP must preserved for slow path
802
803 Label slow_path;
804 // If we need a safepoint check, generate full interpreter entry.
805 ExternalAddress state(SafepointSynchronize::address_of_state());
806 unsigned long offset;
807 __ adrp(rscratch1, ExternalAddress(SafepointSynchronize::address_of_state()), offset);
808 __ ldrw(rscratch1, Address(rscratch1, offset));
809 assert(SafepointSynchronize::_not_synchronized == 0, "rewrite this code");
810 __ cbnz(rscratch1, slow_path);
811
812 // We don't generate local frame and don't align stack because
813 // we call stub code and there is no safepoint on this path.
814
815 // Load parameters
816 const Register crc = c_rarg0; // crc
817 const Register buf = c_rarg1; // source java byte array address
818 const Register len = c_rarg2; // length
819 const Register off = len; // offset (never overlaps with 'len')
820
821 // Arguments are reversed on java expression stack
822 // Calculate address of start element
823 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
824 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
825 __ ldrw(off, Address(esp, wordSize)); // offset
826 __ add(buf, buf, off); // + offset
827 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
828 } else {
829 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
830 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
831 __ ldrw(off, Address(esp, wordSize)); // offset
832 __ add(buf, buf, off); // + offset
833 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
834 }
835 // Can now load 'len' since we're finished with 'off'
836 __ ldrw(len, Address(esp, 0x0)); // Length
837
838 __ andr(sp, r13, -16); // Restore the caller's SP
839
840 // We are frameless so we can just jump to the stub.
841 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
842
843 // generate a vanilla native entry as the slow path
844 __ bind(slow_path);
845
846 (void) generate_native_entry(false);
847
848 return entry;
849 }
850 return generate_native_entry(false);
851 }
852
853 void InterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
854 // Bang each page in the shadow zone. We can't assume it's been done for
855 // an interpreter frame with greater than a page of locals, so each page
856 // needs to be checked. Only true for non-native.
857 if (UseStackBanging) {
858 const int start_page = native_call ? StackShadowPages : 1;
859 const int page_size = os::vm_page_size();
860 for (int pages = start_page; pages <= StackShadowPages ; pages++) {
861 __ sub(rscratch2, sp, pages*page_size);
862 __ str(zr, Address(rscratch2));
863 }
864 }
865 }
866
867
868 // Interpreter stub for calling a native method. (asm interpreter)
869 // This sets up a somewhat different looking stack for calling the
870 // native method than the typical interpreter frame setup.
871 address InterpreterGenerator::generate_native_entry(bool synchronized) {
872 // determine code generation flags
873 bool inc_counter = UseCompiler || CountCompiledCalls;
874
875 // r1: Method*
876 // rscratch1: sender sp
877
878 address entry_point = __ pc();
879
880 const Address constMethod (rmethod, Method::const_offset());
881 const Address access_flags (rmethod, Method::access_flags_offset());
882 const Address size_of_parameters(r2, ConstMethod::
883 size_of_parameters_offset());
884
885 // get parameter size (always needed)
886 __ ldr(r2, constMethod);
887 __ load_unsigned_short(r2, size_of_parameters);
888
889 // native calls don't need the stack size check since they have no
890 // expression stack and the arguments are already on the stack and
891 // we only add a handful of words to the stack
892
893 // rmethod: Method*
894 // r2: size of parameters
895 // rscratch1: sender sp
896
897 // for natives the size of locals is zero
898
899 // compute beginning of parameters (rlocals)
900 __ add(rlocals, esp, r2, ext::uxtx, 3);
901 __ add(rlocals, rlocals, -wordSize);
902
903 // Pull SP back to minimum size: this avoids holes in the stack
904 __ andr(sp, esp, -16);
905
906 // initialize fixed part of activation frame
907 generate_fixed_frame(true);
908
909 // make sure method is native & not abstract
910 #ifdef ASSERT
911 __ ldrw(r0, access_flags);
912 {
913 Label L;
914 __ tst(r0, JVM_ACC_NATIVE);
915 __ br(Assembler::NE, L);
916 __ stop("tried to execute non-native method as native");
917 __ bind(L);
918 }
919 {
920 Label L;
921 __ tst(r0, JVM_ACC_ABSTRACT);
922 __ br(Assembler::EQ, L);
923 __ stop("tried to execute abstract method in interpreter");
924 __ bind(L);
925 }
926 #endif
927
928 // Since at this point in the method invocation the exception
929 // handler would try to exit the monitor of synchronized methods
930 // which hasn't been entered yet, we set the thread local variable
931 // _do_not_unlock_if_synchronized to true. The remove_activation
932 // will check this flag.
933
934 const Address do_not_unlock_if_synchronized(rthread,
935 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
936 __ mov(rscratch2, true);
937 __ strb(rscratch2, do_not_unlock_if_synchronized);
938
939 // increment invocation count & check for overflow
940 Label invocation_counter_overflow;
941 if (inc_counter) {
942 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
943 }
944
945 Label continue_after_compile;
946 __ bind(continue_after_compile);
947
948 bang_stack_shadow_pages(true);
949
950 // reset the _do_not_unlock_if_synchronized flag
951 __ strb(zr, do_not_unlock_if_synchronized);
952
953 // check for synchronized methods
954 // Must happen AFTER invocation_counter check and stack overflow check,
955 // so method is not locked if overflows.
956 if (synchronized) {
957 lock_method();
958 } else {
959 // no synchronization necessary
960 #ifdef ASSERT
961 {
962 Label L;
963 __ ldrw(r0, access_flags);
964 __ tst(r0, JVM_ACC_SYNCHRONIZED);
965 __ br(Assembler::EQ, L);
966 __ stop("method needs synchronization");
967 __ bind(L);
968 }
969 #endif
970 }
971
972 // start execution
973 #ifdef ASSERT
974 {
975 Label L;
976 const Address monitor_block_top(rfp,
977 frame::interpreter_frame_monitor_block_top_offset * wordSize);
978 __ ldr(rscratch1, monitor_block_top);
979 __ cmp(esp, rscratch1);
980 __ br(Assembler::EQ, L);
981 __ stop("broken stack frame setup in interpreter");
982 __ bind(L);
983 }
984 #endif
985
986 // jvmti support
987 __ notify_method_entry();
988
989 // work registers
990 const Register t = r17;
991 const Register result_handler = r19;
992
993 // allocate space for parameters
994 __ ldr(t, Address(rmethod, Method::const_offset()));
995 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
996
997 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
998 __ andr(sp, rscratch1, -16);
999 __ mov(esp, rscratch1);
1000
1001 // get signature handler
1002 {
1003 Label L;
1004 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1005 __ cbnz(t, L);
1006 __ call_VM(noreg,
1007 CAST_FROM_FN_PTR(address,
1008 InterpreterRuntime::prepare_native_call),
1009 rmethod);
1010 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1011 __ bind(L);
1012 }
1013
1014 // call signature handler
1015 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1016 "adjust this code");
1017 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1018 "adjust this code");
1019 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1020 "adjust this code");
1021
1022 // The generated handlers do not touch rmethod (the method).
1023 // However, large signatures cannot be cached and are generated
1024 // each time here. The slow-path generator can do a GC on return,
1025 // so we must reload it after the call.
1026 __ blr(t);
1027 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1028
1029
1030 // result handler is in r0
1031 // set result handler
1032 __ mov(result_handler, r0);
1033 // pass mirror handle if static call
1034 {
1035 Label L;
1036 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1037 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1038 __ tst(t, JVM_ACC_STATIC);
1039 __ br(Assembler::EQ, L);
1040 // get mirror
1041 __ ldr(t, Address(rmethod, Method::const_offset()));
1042 __ ldr(t, Address(t, ConstMethod::constants_offset()));
1043 __ ldr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1044 __ ldr(t, Address(t, mirror_offset));
1045 // copy mirror into activation frame
1046 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1047 // pass handle to mirror
1048 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1049 __ bind(L);
1050 }
1051
1052 // get native function entry point in r10
1053 {
1054 Label L;
1055 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1056 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1057 __ mov(rscratch2, unsatisfied);
1058 __ ldr(rscratch2, rscratch2);
1059 __ cmp(r10, rscratch2);
1060 __ br(Assembler::NE, L);
1061 __ call_VM(noreg,
1062 CAST_FROM_FN_PTR(address,
1063 InterpreterRuntime::prepare_native_call),
1064 rmethod);
1065 __ get_method(rmethod);
1066 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1067 __ bind(L);
1068 }
1069
1070 // pass JNIEnv
1071 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1072
1073 // Set the last Java PC in the frame anchor to be the return address from
1074 // the call to the native method: this will allow the debugger to
1075 // generate an accurate stack trace.
1076 Label native_return;
1077 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1078
1079 // change thread state
1080 #ifdef ASSERT
1081 {
1082 Label L;
1083 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1084 __ cmp(t, _thread_in_Java);
1085 __ br(Assembler::EQ, L);
1086 __ stop("Wrong thread state in native stub");
1087 __ bind(L);
1088 }
1089 #endif
1090
1091 // Change state to native
1092 __ mov(rscratch1, _thread_in_native);
1093 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1094 __ stlrw(rscratch1, rscratch2);
1095
1096 // Call the native method.
1097 __ blr(r10);
1098 __ bind(native_return);
1099 __ maybe_isb();
1100 __ get_method(rmethod);
1101 // result potentially in r0 or v0
1102
1103 // make room for the pushes we're about to do
1104 __ sub(rscratch1, esp, 4 * wordSize);
1105 __ andr(sp, rscratch1, -16);
1106
1107 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1108 // in order to extract the result of a method call. If the order of these
1109 // pushes change or anything else is added to the stack then the code in
1110 // interpreter_frame_result must also change.
1111 __ push(dtos);
1112 __ push(ltos);
1113
1114 // change thread state
1115 __ mov(rscratch1, _thread_in_native_trans);
1116 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1117 __ stlrw(rscratch1, rscratch2);
1118
1119 if (os::is_MP()) {
1120 if (UseMembar) {
1121 // Force this write out before the read below
1122 __ dsb(Assembler::SY);
1123 } else {
1124 // Write serialization page so VM thread can do a pseudo remote membar.
1125 // We use the current thread pointer to calculate a thread specific
1126 // offset to write to within the page. This minimizes bus traffic
1127 // due to cache line collision.
1128 __ serialize_memory(rthread, rscratch2);
1129 }
1130 }
1131
1132 // check for safepoint operation in progress and/or pending suspend requests
1133 {
1134 Label Continue;
1135 {
1136 unsigned long offset;
1137 __ adrp(rscratch2, SafepointSynchronize::address_of_state(), offset);
1138 __ ldrw(rscratch2, Address(rscratch2, offset));
1139 }
1140 assert(SafepointSynchronize::_not_synchronized == 0,
1141 "SafepointSynchronize::_not_synchronized");
1142 Label L;
1143 __ cbnz(rscratch2, L);
1144 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1145 __ cbz(rscratch2, Continue);
1146 __ bind(L);
1147
1148 // Don't use call_VM as it will see a possible pending exception
1149 // and forward it and never return here preventing us from
1150 // clearing _last_native_pc down below. Also can't use
1151 // call_VM_leaf either as it will check to see if r13 & r14 are
1152 // preserved and correspond to the bcp/locals pointers. So we do a
1153 // runtime call by hand.
1154 //
1155 __ mov(c_rarg0, rthread);
1156 __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1157 __ blr(rscratch2);
1158 __ maybe_isb();
1159 __ get_method(rmethod);
1160 __ reinit_heapbase();
1161 __ bind(Continue);
1162 }
1163
1164 // change thread state
1165 __ mov(rscratch1, _thread_in_Java);
1166 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1167 __ stlrw(rscratch1, rscratch2);
1168
1169 // reset_last_Java_frame
1170 __ reset_last_Java_frame(true);
1171
1172 // reset handle block
1173 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1174 __ str(zr, Address(t, JNIHandleBlock::top_offset_in_bytes()));
1175
1176 // If result is an oop unbox and store it in frame where gc will see it
1177 // and result handler will pick it up
1178
1179 {
1180 Label no_oop, not_weak, store_result;
1181 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1182 __ cmp(t, result_handler);
1183 __ br(Assembler::NE, no_oop);
1184 // Unbox oop result, e.g. JNIHandles::resolve result.
1185 __ pop(ltos);
1186 __ cbz(r0, store_result); // Use NULL as-is.
1187 STATIC_ASSERT(JNIHandles::weak_tag_mask == 1u);
1188 __ tbz(r0, 0, not_weak); // Test for jweak tag.
1189 // Resolve jweak.
1190 __ ldr(r0, Address(r0, -JNIHandles::weak_tag_value));
1191 #if INCLUDE_ALL_GCS
1192 if (UseG1GC || (UseShenandoahGC && ShenandoahSATBBarrier)) {
1193 __ enter(); // Barrier may call runtime.
1194 __ g1_write_barrier_pre(noreg /* obj */,
1195 r0 /* pre_val */,
1196 rthread /* thread */,
1197 t /* tmp */,
1198 true /* tosca_live */,
1199 true /* expand_call */);
1200 __ leave();
1201 }
1202 #endif // INCLUDE_ALL_GCS
1203 __ b(store_result);
1204 __ bind(not_weak);
1205 // Resolve (untagged) jobject.
1206 __ ldr(r0, Address(r0, 0));
1207 __ bind(store_result);
1208 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1209 // keep stack depth as expected by pushing oop which will eventually be discarded
1210 __ push(ltos);
1211 __ bind(no_oop);
1212 }
1213
1214 {
1215 Label no_reguard;
1216 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1217 __ ldrb(rscratch1, Address(rscratch1));
1218 __ cmp(rscratch1, JavaThread::stack_guard_yellow_disabled);
1219 __ br(Assembler::NE, no_reguard);
1220
1221 __ pusha(); // XXX only save smashed registers
1222 __ mov(c_rarg0, rthread);
1223 __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1224 __ blr(rscratch2);
1225 __ popa(); // XXX only restore smashed registers
1226 __ bind(no_reguard);
1227 }
1228
1229 // The method register is junk from after the thread_in_native transition
1230 // until here. Also can't call_VM until the bcp has been
1231 // restored. Need bcp for throwing exception below so get it now.
1232 __ get_method(rmethod);
1233
1234 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1235 // rbcp == code_base()
1236 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1237 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1238 // handle exceptions (exception handling will handle unlocking!)
1239 {
1240 Label L;
1241 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1242 __ cbz(rscratch1, L);
1243 // Note: At some point we may want to unify this with the code
1244 // used in call_VM_base(); i.e., we should use the
1245 // StubRoutines::forward_exception code. For now this doesn't work
1246 // here because the rsp is not correctly set at this point.
1247 __ MacroAssembler::call_VM(noreg,
1248 CAST_FROM_FN_PTR(address,
1249 InterpreterRuntime::throw_pending_exception));
1250 __ should_not_reach_here();
1251 __ bind(L);
1252 }
1253
1254 // do unlocking if necessary
1255 {
1256 Label L;
1257 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1258 __ tst(t, JVM_ACC_SYNCHRONIZED);
1259 __ br(Assembler::EQ, L);
1260 // the code below should be shared with interpreter macro
1261 // assembler implementation
1262 {
1263 Label unlock;
1264 // BasicObjectLock will be first in list, since this is a
1265 // synchronized method. However, need to check that the object
1266 // has not been unlocked by an explicit monitorexit bytecode.
1267
1268 // monitor expect in c_rarg1 for slow unlock path
1269 __ lea (c_rarg1, Address(rfp, // address of first monitor
1270 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1271 wordSize - sizeof(BasicObjectLock))));
1272
1273 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1274 __ cbnz(t, unlock);
1275
1276 // Entry already unlocked, need to throw exception
1277 __ MacroAssembler::call_VM(noreg,
1278 CAST_FROM_FN_PTR(address,
1279 InterpreterRuntime::throw_illegal_monitor_state_exception));
1280 __ should_not_reach_here();
1281
1282 __ bind(unlock);
1283 __ unlock_object(c_rarg1);
1284 }
1285 __ bind(L);
1286 }
1287
1288 // jvmti support
1289 // Note: This must happen _after_ handling/throwing any exceptions since
1290 // the exception handler code notifies the runtime of method exits
1291 // too. If this happens before, method entry/exit notifications are
1292 // not properly paired (was bug - gri 11/22/99).
1293 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1294
1295 // restore potential result in r0:d0, call result handler to
1296 // restore potential result in ST0 & handle result
1297
1298 __ pop(ltos);
1299 __ pop(dtos);
1300
1301 __ blr(result_handler);
1302
1303 // remove activation
1304 __ ldr(esp, Address(rfp,
1305 frame::interpreter_frame_sender_sp_offset *
1306 wordSize)); // get sender sp
1307 // remove frame anchor
1308 __ leave();
1309
1310 // resture sender sp
1311 __ mov(sp, esp);
1312
1313 __ ret(lr);
1314
1315 if (inc_counter) {
1316 // Handle overflow of counter and compile method
1317 __ bind(invocation_counter_overflow);
1318 generate_counter_overflow(&continue_after_compile);
1319 }
1320
1321 return entry_point;
1322 }
1323
1324 //
1325 // Generic interpreted method entry to (asm) interpreter
1326 //
1327 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1328 // determine code generation flags
1329 bool inc_counter = UseCompiler || CountCompiledCalls;
1330
1331 // rscratch1: sender sp
1332 address entry_point = __ pc();
1333
1334 const Address constMethod(rmethod, Method::const_offset());
1335 const Address access_flags(rmethod, Method::access_flags_offset());
1336 const Address size_of_parameters(r3,
1337 ConstMethod::size_of_parameters_offset());
1338 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1339
1340 // get parameter size (always needed)
1341 // need to load the const method first
1342 __ ldr(r3, constMethod);
1343 __ load_unsigned_short(r2, size_of_parameters);
1344
1345 // r2: size of parameters
1346
1347 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1348 __ sub(r3, r3, r2); // r3 = no. of additional locals
1349
1350 // see if we've got enough room on the stack for locals plus overhead.
1351 generate_stack_overflow_check();
1352
1353 // compute beginning of parameters (rlocals)
1354 __ add(rlocals, esp, r2, ext::uxtx, 3);
1355 __ sub(rlocals, rlocals, wordSize);
1356
1357 // Make room for locals
1358 __ sub(rscratch1, esp, r3, ext::uxtx, 3);
1359 __ andr(sp, rscratch1, -16);
1360
1361 // r3 - # of additional locals
1362 // allocate space for locals
1363 // explicitly initialize locals
1364 {
1365 Label exit, loop;
1366 __ ands(zr, r3, r3);
1367 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1368 __ bind(loop);
1369 __ str(zr, Address(__ post(rscratch1, wordSize)));
1370 __ sub(r3, r3, 1); // until everything initialized
1371 __ cbnz(r3, loop);
1372 __ bind(exit);
1373 }
1374
1375 // And the base dispatch table
1376 __ get_dispatch();
1377
1378 // initialize fixed part of activation frame
1379 generate_fixed_frame(false);
1380
1381 // make sure method is not native & not abstract
1382 #ifdef ASSERT
1383 __ ldrw(r0, access_flags);
1384 {
1385 Label L;
1386 __ tst(r0, JVM_ACC_NATIVE);
1387 __ br(Assembler::EQ, L);
1388 __ stop("tried to execute native method as non-native");
1389 __ bind(L);
1390 }
1391 {
1392 Label L;
1393 __ tst(r0, JVM_ACC_ABSTRACT);
1394 __ br(Assembler::EQ, L);
1395 __ stop("tried to execute abstract method in interpreter");
1396 __ bind(L);
1397 }
1398 #endif
1399
1400 // Since at this point in the method invocation the exception
1401 // handler would try to exit the monitor of synchronized methods
1402 // which hasn't been entered yet, we set the thread local variable
1403 // _do_not_unlock_if_synchronized to true. The remove_activation
1404 // will check this flag.
1405
1406 const Address do_not_unlock_if_synchronized(rthread,
1407 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1408 __ mov(rscratch2, true);
1409 __ strb(rscratch2, do_not_unlock_if_synchronized);
1410
1411 // increment invocation count & check for overflow
1412 Label invocation_counter_overflow;
1413 Label profile_method;
1414 Label profile_method_continue;
1415 if (inc_counter) {
1416 generate_counter_incr(&invocation_counter_overflow,
1417 &profile_method,
1418 &profile_method_continue);
1419 if (ProfileInterpreter) {
1420 __ bind(profile_method_continue);
1421 }
1422 }
1423
1424 Label continue_after_compile;
1425 __ bind(continue_after_compile);
1426
1427 bang_stack_shadow_pages(false);
1428
1429 // reset the _do_not_unlock_if_synchronized flag
1430 __ strb(zr, do_not_unlock_if_synchronized);
1431
1432 // check for synchronized methods
1433 // Must happen AFTER invocation_counter check and stack overflow check,
1434 // so method is not locked if overflows.
1435 if (synchronized) {
1436 // Allocate monitor and lock method
1437 lock_method();
1438 } else {
1439 // no synchronization necessary
1440 #ifdef ASSERT
1441 {
1442 Label L;
1443 __ ldrw(r0, access_flags);
1444 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1445 __ br(Assembler::EQ, L);
1446 __ stop("method needs synchronization");
1447 __ bind(L);
1448 }
1449 #endif
1450 }
1451
1452 // start execution
1453 #ifdef ASSERT
1454 {
1455 Label L;
1456 const Address monitor_block_top (rfp,
1457 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1458 __ ldr(rscratch1, monitor_block_top);
1459 __ cmp(esp, rscratch1);
1460 __ br(Assembler::EQ, L);
1461 __ stop("broken stack frame setup in interpreter");
1462 __ bind(L);
1463 }
1464 #endif
1465
1466 // jvmti support
1467 __ notify_method_entry();
1468
1469 __ dispatch_next(vtos);
1470
1471 // invocation counter overflow
1472 if (inc_counter) {
1473 if (ProfileInterpreter) {
1474 // We have decided to profile this method in the interpreter
1475 __ bind(profile_method);
1476 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1477 __ set_method_data_pointer_for_bcp();
1478 // don't think we need this
1479 __ get_method(r1);
1480 __ b(profile_method_continue);
1481 }
1482 // Handle overflow of counter and compile method
1483 __ bind(invocation_counter_overflow);
1484 generate_counter_overflow(&continue_after_compile);
1485 }
1486
1487 return entry_point;
1488 }
1489
1490 // Entry points
1491 //
1492 // Here we generate the various kind of entries into the interpreter.
1493 // The two main entry type are generic bytecode methods and native
1494 // call method. These both come in synchronized and non-synchronized
1495 // versions but the frame layout they create is very similar. The
1496 // other method entry types are really just special purpose entries
1497 // that are really entry and interpretation all in one. These are for
1498 // trivial methods like accessor, empty, or special math methods.
1499 //
1500 // When control flow reaches any of the entry types for the interpreter
1501 // the following holds ->
1502 //
1503 // Arguments:
1504 //
1505 // rmethod: Method*
1506 //
1507 // Stack layout immediately at entry
1508 //
1509 // [ return address ] <--- rsp
1510 // [ parameter n ]
1511 // ...
1512 // [ parameter 1 ]
1513 // [ expression stack ] (caller's java expression stack)
1514
1515 // Assuming that we don't go to one of the trivial specialized entries
1516 // the stack will look like below when we are ready to execute the
1517 // first bytecode (or call the native routine). The register usage
1518 // will be as the template based interpreter expects (see
1519 // interpreter_aarch64.hpp).
1520 //
1521 // local variables follow incoming parameters immediately; i.e.
1522 // the return address is moved to the end of the locals).
1523 //
1524 // [ monitor entry ] <--- esp
1525 // ...
1526 // [ monitor entry ]
1527 // [ expr. stack bottom ]
1528 // [ saved rbcp ]
1529 // [ current rlocals ]
1530 // [ Method* ]
1531 // [ saved rfp ] <--- rfp
1532 // [ return address ]
1533 // [ local variable m ]
1534 // ...
1535 // [ local variable 1 ]
1536 // [ parameter n ]
1537 // ...
1538 // [ parameter 1 ] <--- rlocals
1539
1540 address AbstractInterpreterGenerator::generate_method_entry(
1541 AbstractInterpreter::MethodKind kind) {
1542 // determine code generation flags
1543 bool synchronized = false;
1544 address entry_point = NULL;
1545
1546 switch (kind) {
1547 case Interpreter::zerolocals : break;
1548 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1549 case Interpreter::native : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1550 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true); break;
1551 case Interpreter::empty : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry(); break;
1552 case Interpreter::accessor : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry(); break;
1553 case Interpreter::abstract : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry(); break;
1554
1555 case Interpreter::java_lang_math_sin : // fall thru
1556 case Interpreter::java_lang_math_cos : // fall thru
1557 case Interpreter::java_lang_math_tan : // fall thru
1558 case Interpreter::java_lang_math_abs : // fall thru
1559 case Interpreter::java_lang_math_log : // fall thru
1560 case Interpreter::java_lang_math_log10 : // fall thru
1561 case Interpreter::java_lang_math_sqrt : // fall thru
1562 case Interpreter::java_lang_math_pow : // fall thru
1563 case Interpreter::java_lang_math_exp : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind); break;
1564 case Interpreter::java_lang_ref_reference_get
1565 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1566 case Interpreter::java_util_zip_CRC32_update
1567 : entry_point = ((InterpreterGenerator*)this)->generate_CRC32_update_entry(); break;
1568 case Interpreter::java_util_zip_CRC32_updateBytes
1569 : // fall thru
1570 case Interpreter::java_util_zip_CRC32_updateByteBuffer
1571 : entry_point = ((InterpreterGenerator*)this)->generate_CRC32_updateBytes_entry(kind); break;
1572 default : ShouldNotReachHere(); break;
1573 }
1574
1575 if (entry_point) {
1576 return entry_point;
1577 }
1578
1579 return ((InterpreterGenerator*) this)->
1580 generate_normal_entry(synchronized);
1581 }
1582
1583
1584 // These should never be compiled since the interpreter will prefer
1585 // the compiled version to the intrinsic version.
1586 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1587 switch (method_kind(m)) {
1588 case Interpreter::java_lang_math_sin : // fall thru
1589 case Interpreter::java_lang_math_cos : // fall thru
1590 case Interpreter::java_lang_math_tan : // fall thru
1591 case Interpreter::java_lang_math_abs : // fall thru
1592 case Interpreter::java_lang_math_log : // fall thru
1593 case Interpreter::java_lang_math_log10 : // fall thru
1594 case Interpreter::java_lang_math_sqrt : // fall thru
1595 case Interpreter::java_lang_math_pow : // fall thru
1596 case Interpreter::java_lang_math_exp :
1597 return false;
1598 default:
1599 return true;
1600 }
1601 }
1602
1603 // How much stack a method activation needs in words.
1604 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1605 const int entry_size = frame::interpreter_frame_monitor_size();
1606
1607 // total overhead size: entry_size + (saved rfp thru expr stack
1608 // bottom). be sure to change this if you add/subtract anything
1609 // to/from the overhead area
1610 const int overhead_size =
1611 -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1612
1613 const int stub_code = frame::entry_frame_after_call_words;
1614 const int method_stack = (method->max_locals() + method->max_stack()) *
1615 Interpreter::stackElementWords;
1616 return (overhead_size + method_stack + stub_code);
1617 }
1618
1619 // asm based interpreter deoptimization helpers
1620 int AbstractInterpreter::size_activation(int max_stack,
1621 int temps,
1622 int extra_args,
1623 int monitors,
1624 int callee_params,
1625 int callee_locals,
1626 bool is_top_frame) {
1627 // Note: This calculation must exactly parallel the frame setup
1628 // in AbstractInterpreterGenerator::generate_method_entry.
1629
1630 // fixed size of an interpreter frame:
1631 int overhead = frame::sender_sp_offset -
1632 frame::interpreter_frame_initial_sp_offset;
1633 // Our locals were accounted for by the caller (or last_frame_adjust
1634 // on the transistion) Since the callee parameters already account
1635 // for the callee's params we only need to account for the extra
1636 // locals.
1637 int size = overhead +
1638 (callee_locals - callee_params) +
1639 monitors * frame::interpreter_frame_monitor_size() +
1640 // On the top frame, at all times SP <= ESP, and SP is
1641 // 16-aligned. We ensure this by adjusting SP on method
1642 // entry and re-entry to allow room for the maximum size of
1643 // the expression stack. When we call another method we bump
1644 // SP so that no stack space is wasted. So, only on the top
1645 // frame do we need to allow max_stack words.
1646 (is_top_frame ? max_stack : temps + extra_args);
1647
1648 // On AArch64 we always keep the stack pointer 16-aligned, so we
1649 // must round up here.
1650 size = round_to(size, 2);
1651
1652 return size;
1653 }
1654
1655 void AbstractInterpreter::layout_activation(Method* method,
1656 int tempcount,
1657 int popframe_extra_args,
1658 int moncount,
1659 int caller_actual_parameters,
1660 int callee_param_count,
1661 int callee_locals,
1662 frame* caller,
1663 frame* interpreter_frame,
1664 bool is_top_frame,
1665 bool is_bottom_frame) {
1666 // The frame interpreter_frame is guaranteed to be the right size,
1667 // as determined by a previous call to the size_activation() method.
1668 // It is also guaranteed to be walkable even though it is in a
1669 // skeletal state
1670
1671 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1672 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1673 Interpreter::stackElementWords;
1674
1675 #ifdef ASSERT
1676 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable");
1677 #endif
1678
1679 interpreter_frame->interpreter_frame_set_method(method);
1680 // NOTE the difference in using sender_sp and
1681 // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1682 // the original sp of the caller (the unextended_sp) and
1683 // sender_sp is fp+8/16 (32bit/64bit) XXX
1684 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1685
1686 #ifdef ASSERT
1687 if (caller->is_interpreted_frame()) {
1688 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1689 }
1690 #endif
1691
1692 interpreter_frame->interpreter_frame_set_locals(locals);
1693 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1694 BasicObjectLock* monbot = montop - moncount;
1695 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1696
1697 // Set last_sp
1698 intptr_t* esp = (intptr_t*) monbot -
1699 tempcount*Interpreter::stackElementWords -
1700 popframe_extra_args;
1701 interpreter_frame->interpreter_frame_set_last_sp(esp);
1702
1703 // All frames but the initial (oldest) interpreter frame we fill in have
1704 // a value for sender_sp that allows walking the stack but isn't
1705 // truly correct. Correct the value here.
1706 if (extra_locals != 0 &&
1707 interpreter_frame->sender_sp() ==
1708 interpreter_frame->interpreter_frame_sender_sp()) {
1709 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1710 extra_locals);
1711 }
1712 *interpreter_frame->interpreter_frame_cache_addr() =
1713 method->constants()->cache();
1714 }
1715
1716
1717 //-----------------------------------------------------------------------------
1718 // Exceptions
1719
1720 void TemplateInterpreterGenerator::generate_throw_exception() {
1721 // Entry point in previous activation (i.e., if the caller was
1722 // interpreted)
1723 Interpreter::_rethrow_exception_entry = __ pc();
1724 // Restore sp to interpreter_frame_last_sp even though we are going
1725 // to empty the expression stack for the exception processing.
1726 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1727 // r0: exception
1728 // r3: return address/pc that threw exception
1729 __ restore_bcp(); // rbcp points to call/send
1730 __ restore_locals();
1731 __ restore_constant_pool_cache();
1732 __ reinit_heapbase(); // restore rheapbase as heapbase.
1733 __ get_dispatch();
1734
1735 // Entry point for exceptions thrown within interpreter code
1736 Interpreter::_throw_exception_entry = __ pc();
1737 // If we came here via a NullPointerException on the receiver of a
1738 // method, rmethod may be corrupt.
1739 __ get_method(rmethod);
1740 // expression stack is undefined here
1741 // r0: exception
1742 // rbcp: exception bcp
1743 __ verify_oop(r0);
1744 __ mov(c_rarg1, r0);
1745
1746 // expression stack must be empty before entering the VM in case of
1747 // an exception
1748 __ empty_expression_stack();
1749 // find exception handler address and preserve exception oop
1750 __ call_VM(r3,
1751 CAST_FROM_FN_PTR(address,
1752 InterpreterRuntime::exception_handler_for_exception),
1753 c_rarg1);
1754
1755 // Calculate stack limit
1756 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1757 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1758 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size()
1759 + (EnableInvokeDynamic ? 2 : 0) + 2);
1760 __ ldr(rscratch2,
1761 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1762 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
1763 __ andr(sp, rscratch1, -16);
1764
1765 // r0: exception handler entry point
1766 // r3: preserved exception oop
1767 // rbcp: bcp for exception handler
1768 __ push_ptr(r3); // push exception which is now the only value on the stack
1769 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1770
1771 // If the exception is not handled in the current frame the frame is
1772 // removed and the exception is rethrown (i.e. exception
1773 // continuation is _rethrow_exception).
1774 //
1775 // Note: At this point the bci is still the bxi for the instruction
1776 // which caused the exception and the expression stack is
1777 // empty. Thus, for any VM calls at this point, GC will find a legal
1778 // oop map (with empty expression stack).
1779
1780 //
1781 // JVMTI PopFrame support
1782 //
1783
1784 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1785 __ empty_expression_stack();
1786 // Set the popframe_processing bit in pending_popframe_condition
1787 // indicating that we are currently handling popframe, so that
1788 // call_VMs that may happen later do not trigger new popframe
1789 // handling cycles.
1790 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1791 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1792 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1793
1794 {
1795 // Check to see whether we are returning to a deoptimized frame.
1796 // (The PopFrame call ensures that the caller of the popped frame is
1797 // either interpreted or compiled and deoptimizes it if compiled.)
1798 // In this case, we can't call dispatch_next() after the frame is
1799 // popped, but instead must save the incoming arguments and restore
1800 // them after deoptimization has occurred.
1801 //
1802 // Note that we don't compare the return PC against the
1803 // deoptimization blob's unpack entry because of the presence of
1804 // adapter frames in C2.
1805 Label caller_not_deoptimized;
1806 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1807 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1808 InterpreterRuntime::interpreter_contains), c_rarg1);
1809 __ cbnz(r0, caller_not_deoptimized);
1810
1811 // Compute size of arguments for saving when returning to
1812 // deoptimized caller
1813 __ get_method(r0);
1814 __ ldr(r0, Address(r0, Method::const_offset()));
1815 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1816 size_of_parameters_offset())));
1817 __ lsl(r0, r0, Interpreter::logStackElementSize);
1818 __ restore_locals(); // XXX do we need this?
1819 __ sub(rlocals, rlocals, r0);
1820 __ add(rlocals, rlocals, wordSize);
1821 // Save these arguments
1822 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1823 Deoptimization::
1824 popframe_preserve_args),
1825 rthread, r0, rlocals);
1826
1827 __ remove_activation(vtos,
1828 /* throw_monitor_exception */ false,
1829 /* install_monitor_exception */ false,
1830 /* notify_jvmdi */ false);
1831
1832 // Inform deoptimization that it is responsible for restoring
1833 // these arguments
1834 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1835 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1836
1837 // Continue in deoptimization handler
1838 __ ret(lr);
1839
1840 __ bind(caller_not_deoptimized);
1841 }
1842
1843 __ remove_activation(vtos,
1844 /* throw_monitor_exception */ false,
1845 /* install_monitor_exception */ false,
1846 /* notify_jvmdi */ false);
1847
1848 // Restore the last_sp and null it out
1849 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1850 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1851
1852 __ restore_bcp();
1853 __ restore_locals();
1854 __ restore_constant_pool_cache();
1855 __ get_method(rmethod);
1856 __ get_dispatch();
1857
1858 // The method data pointer was incremented already during
1859 // call profiling. We have to restore the mdp for the current bcp.
1860 if (ProfileInterpreter) {
1861 __ set_method_data_pointer_for_bcp();
1862 }
1863
1864 // Clear the popframe condition flag
1865 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1866 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1867
1868 #if INCLUDE_JVMTI
1869 if (EnableInvokeDynamic) {
1870 Label L_done;
1871
1872 __ ldrb(rscratch1, Address(rbcp, 0));
1873 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1874 __ br(Assembler::NE, L_done);
1875
1876 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1877 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1878
1879 __ ldr(c_rarg0, Address(rlocals, 0));
1880 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1881
1882 __ cbz(r0, L_done);
1883
1884 __ str(r0, Address(esp, 0));
1885 __ bind(L_done);
1886 }
1887 #endif // INCLUDE_JVMTI
1888
1889 // Restore machine SP
1890 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1891 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1892 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size()
1893 + (EnableInvokeDynamic ? 2 : 0));
1894 __ ldr(rscratch2,
1895 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1896 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
1897 __ andr(sp, rscratch1, -16);
1898
1899 __ dispatch_next(vtos);
1900 // end of PopFrame support
1901
1902 Interpreter::_remove_activation_entry = __ pc();
1903
1904 // preserve exception over this code sequence
1905 __ pop_ptr(r0);
1906 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1907 // remove the activation (without doing throws on illegalMonitorExceptions)
1908 __ remove_activation(vtos, false, true, false);
1909 // restore exception
1910 __ get_vm_result(r0, rthread);
1911
1912 // In between activations - previous activation type unknown yet
1913 // compute continuation point - the continuation point expects the
1914 // following registers set up:
1915 //
1916 // r0: exception
1917 // lr: return address/pc that threw exception
1918 // esp: expression stack of caller
1919 // rfp: fp of caller
1920 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
1921 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1922 SharedRuntime::exception_handler_for_return_address),
1923 rthread, lr);
1924 __ mov(r1, r0); // save exception handler
1925 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
1926 // We might be returning to a deopt handler that expects r3 to
1927 // contain the exception pc
1928 __ mov(r3, lr);
1929 // Note that an "issuing PC" is actually the next PC after the call
1930 __ br(r1); // jump to exception
1931 // handler of caller
1932 }
1933
1934
1935 //
1936 // JVMTI ForceEarlyReturn support
1937 //
1938 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1939 address entry = __ pc();
1940
1941 __ restore_bcp();
1942 __ restore_locals();
1943 __ empty_expression_stack();
1944 __ load_earlyret_value(state);
1945
1946 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
1947 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
1948
1949 // Clear the earlyret state
1950 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1951 __ str(zr, cond_addr);
1952
1953 __ remove_activation(state,
1954 false, /* throw_monitor_exception */
1955 false, /* install_monitor_exception */
1956 true); /* notify_jvmdi */
1957 __ ret(lr);
1958
1959 return entry;
1960 } // end of ForceEarlyReturn support
1961
1962
1963
1964 //-----------------------------------------------------------------------------
1965 // Helper for vtos entry point generation
1966
1967 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1968 address& bep,
1969 address& cep,
1970 address& sep,
1971 address& aep,
1972 address& iep,
1973 address& lep,
1974 address& fep,
1975 address& dep,
1976 address& vep) {
1977 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1978 Label L;
1979 aep = __ pc(); __ push_ptr(); __ b(L);
1980 fep = __ pc(); __ push_f(); __ b(L);
1981 dep = __ pc(); __ push_d(); __ b(L);
1982 lep = __ pc(); __ push_l(); __ b(L);
1983 bep = cep = sep =
1984 iep = __ pc(); __ push_i();
1985 vep = __ pc();
1986 __ bind(L);
1987 generate_and_dispatch(t);
1988 }
1989
1990 //-----------------------------------------------------------------------------
1991 // Generation of individual instructions
1992
1993 // helpers for generate_and_dispatch
1994
1995
1996 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1997 : TemplateInterpreterGenerator(code) {
1998 generate_all(); // down here so it can be "virtual"
1999 }
2000
2001 //-----------------------------------------------------------------------------
2002
2003 // Non-product code
2004 #ifndef PRODUCT
2005 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2006 address entry = __ pc();
2007
2008 __ push(lr);
2009 __ push(state);
2010 __ push(RegSet::range(r0, r15), sp);
2011 __ mov(c_rarg2, r0); // Pass itos
2012 __ call_VM(noreg,
2013 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
2014 c_rarg1, c_rarg2, c_rarg3);
2015 __ pop(RegSet::range(r0, r15), sp);
2016 __ pop(state);
2017 __ pop(lr);
2018 __ ret(lr); // return from result handler
2019
2020 return entry;
2021 }
2022
2023 void TemplateInterpreterGenerator::count_bytecode() {
2024 Register rscratch3 = r0;
2025 __ push(rscratch1);
2026 __ push(rscratch2);
2027 __ push(rscratch3);
2028 __ mov(rscratch3, (address) &BytecodeCounter::_counter_value);
2029 __ atomic_add(noreg, 1, rscratch3);
2030 __ pop(rscratch3);
2031 __ pop(rscratch2);
2032 __ pop(rscratch1);
2033 }
2034
2035 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { ; }
2036
2037 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { ; }
2038
2039
2040 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2041 // Call a little run-time stub to avoid blow-up for each bytecode.
2042 // The run-time runtime saves the right registers, depending on
2043 // the tosca in-state for the given template.
2044
2045 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2046 "entry must have been generated");
2047 __ bl(Interpreter::trace_code(t->tos_in()));
2048 __ reinit_heapbase();
2049 }
2050
2051
2052 void TemplateInterpreterGenerator::stop_interpreter_at() {
2053 Label L;
2054 __ push(rscratch1);
2055 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2056 __ ldr(rscratch1, Address(rscratch1));
2057 __ mov(rscratch2, StopInterpreterAt);
2058 __ cmpw(rscratch1, rscratch2);
2059 __ br(Assembler::NE, L);
2060 __ brk(0);
2061 __ bind(L);
2062 __ pop(rscratch1);
2063 }
2064
2065 #endif // !PRODUCT
2066 #endif // ! CC_INTERP