1 /*
2 * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * Copyright (c) 2020, 2022, Huawei Technologies Co., Ltd. All rights reserved.
5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 *
7 * This code is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 only, as
9 * published by the Free Software Foundation.
10 *
11 * This code is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * version 2 for more details (a copy is included in the LICENSE file that
15 * accompanied this code).
16 *
17 * You should have received a copy of the GNU General Public License version
18 * 2 along with this work; if not, write to the Free Software Foundation,
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 *
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
22 * or visit www.oracle.com if you need additional information or have any
23 * questions.
24 *
25 */
26
27 #include "precompiled.hpp"
28 #include "asm/macroAssembler.inline.hpp"
29 #include "classfile/javaClasses.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interpreter/bytecodeHistogram.hpp"
32 #include "interpreter/bytecodeTracer.hpp"
33 #include "interpreter/interp_masm.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "interpreter/interpreterRuntime.hpp"
36 #include "interpreter/templateInterpreterGenerator.hpp"
37 #include "interpreter/templateTable.hpp"
38 #include "memory/resourceArea.hpp"
39 #include "oops/arrayOop.hpp"
40 #include "oops/method.inline.hpp"
41 #include "oops/methodData.hpp"
42 #include "oops/oop.inline.hpp"
43 #include "oops/resolvedIndyEntry.hpp"
44 #include "oops/resolvedMethodEntry.hpp"
45 #include "prims/jvmtiExport.hpp"
46 #include "prims/jvmtiThreadState.hpp"
47 #include "runtime/arguments.hpp"
48 #include "runtime/deoptimization.hpp"
49 #include "runtime/frame.inline.hpp"
50 #include "runtime/globals.hpp"
51 #include "runtime/jniHandles.hpp"
52 #include "runtime/sharedRuntime.hpp"
53 #include "runtime/stubRoutines.hpp"
54 #include "runtime/synchronizer.hpp"
55 #include "runtime/timer.hpp"
56 #include "runtime/vframeArray.hpp"
57 #include "utilities/checkedCast.hpp"
58 #include "utilities/debug.hpp"
59 #include "utilities/powerOfTwo.hpp"
60 #include <sys/types.h>
61
62 #ifndef PRODUCT
63 #include "oops/method.hpp"
64 #endif // !PRODUCT
65
66 // Size of interpreter code. Increase if too small. Interpreter will
67 // fail with a guarantee ("not enough space for interpreter generation");
68 // if too small.
69 // Run with +PrintInterpreter to get the VM to print out the size.
70 // Max size with JVMTI
71 int TemplateInterpreter::InterpreterCodeSize = 256 * 1024;
72
73 #define __ _masm->
74
75 //-----------------------------------------------------------------------------
76
77 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
78 address entry = __ pc();
79
80 __ andi(esp, esp, -16);
81 __ mv(c_rarg3, esp);
82 // xmethod
83 // xlocals
84 // c_rarg3: first stack arg - wordSize
85 // adjust sp
86
87 __ addi(sp, c_rarg3, -18 * wordSize);
88 __ addi(sp, sp, -2 * wordSize);
89 __ sd(ra, Address(sp, 0));
90
91 __ call_VM(noreg,
92 CAST_FROM_FN_PTR(address,
93 InterpreterRuntime::slow_signature_handler),
94 xmethod, xlocals, c_rarg3);
95
96 // x10: result handler
97
98 // Stack layout:
99 // sp: return address <- sp
100 // 1 garbage
101 // 8 integer args (if static first is unused)
102 // 1 float/double identifiers
103 // 8 double args
104 // stack args <- esp
105 // garbage
106 // expression stack bottom
107 // bcp (null)
108 // ...
109
110 // Restore ra
111 __ ld(ra, Address(sp, 0));
112 __ addi(sp, sp , 2 * wordSize);
113
114 // Do FP first so we can use c_rarg3 as temp
115 __ lwu(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
116
117 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
118 const FloatRegister r = g_FPArgReg[i];
119 Label d, done;
120
121 __ test_bit(t0, c_rarg3, i);
122 __ bnez(t0, d);
123 __ flw(r, Address(sp, (10 + i) * wordSize));
124 __ j(done);
125 __ bind(d);
126 __ fld(r, Address(sp, (10 + i) * wordSize));
127 __ bind(done);
128 }
129
130 // c_rarg0 contains the result from the call of
131 // InterpreterRuntime::slow_signature_handler so we don't touch it
132 // here. It will be loaded with the JNIEnv* later.
133 for (int i = 1; i < Argument::n_int_register_parameters_c; i++) {
134 const Register rm = g_INTArgReg[i];
135 __ ld(rm, Address(sp, i * wordSize));
136 }
137
138 __ addi(sp, sp, 18 * wordSize);
139 __ ret();
140
141 return entry;
142 }
143
144 // Various method entries
145 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
146 // xmethod: Method*
147 // x19_sender_sp: sender sp
148 // esp: args
149
150 // These don't need a safepoint check because they aren't virtually
151 // callable. We won't enter these intrinsics from compiled code.
152 // If in the future we added an intrinsic which was virtually callable
153 // we'd have to worry about how to safepoint so that this code is used.
154
155 // mathematical functions inlined by compiler
156 // (interpreter must provide identical implementation
157 // in order to avoid monotonicity bugs when switching
158 // from interpreter to compiler in the middle of some
159 // computation)
160 //
161 // stack:
162 // [ arg ] <-- esp
163 // [ arg ]
164 // retaddr in ra
165
166 address fn = nullptr;
167 address entry_point = nullptr;
168 Register continuation = ra;
169 switch (kind) {
170 case Interpreter::java_lang_math_abs:
171 entry_point = __ pc();
172 __ fld(f10, Address(esp));
173 __ fabs_d(f10, f10);
174 __ mv(sp, x19_sender_sp); // Restore caller's SP
175 break;
176 case Interpreter::java_lang_math_sqrt:
177 entry_point = __ pc();
178 __ fld(f10, Address(esp));
179 __ fsqrt_d(f10, f10);
180 __ mv(sp, x19_sender_sp);
181 break;
182 case Interpreter::java_lang_math_sin :
183 entry_point = __ pc();
184 __ fld(f10, Address(esp));
185 __ mv(sp, x19_sender_sp);
186 __ mv(x9, ra);
187 continuation = x9; // The first callee-saved register
188 if (StubRoutines::dsin() == nullptr) {
189 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
190 } else {
191 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
192 }
193 __ call(fn);
194 break;
195 case Interpreter::java_lang_math_cos :
196 entry_point = __ pc();
197 __ fld(f10, Address(esp));
198 __ mv(sp, x19_sender_sp);
199 __ mv(x9, ra);
200 continuation = x9; // The first callee-saved register
201 if (StubRoutines::dcos() == nullptr) {
202 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
203 } else {
204 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
205 }
206 __ call(fn);
207 break;
208 case Interpreter::java_lang_math_tan :
209 entry_point = __ pc();
210 __ fld(f10, Address(esp));
211 __ mv(sp, x19_sender_sp);
212 __ mv(x9, ra);
213 continuation = x9; // The first callee-saved register
214 if (StubRoutines::dtan() == nullptr) {
215 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
216 } else {
217 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
218 }
219 __ call(fn);
220 break;
221 case Interpreter::java_lang_math_log :
222 entry_point = __ pc();
223 __ fld(f10, Address(esp));
224 __ mv(sp, x19_sender_sp);
225 __ mv(x9, ra);
226 continuation = x9; // The first callee-saved register
227 if (StubRoutines::dlog() == nullptr) {
228 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
229 } else {
230 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
231 }
232 __ call(fn);
233 break;
234 case Interpreter::java_lang_math_log10 :
235 entry_point = __ pc();
236 __ fld(f10, Address(esp));
237 __ mv(sp, x19_sender_sp);
238 __ mv(x9, ra);
239 continuation = x9; // The first callee-saved register
240 if (StubRoutines::dlog10() == nullptr) {
241 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
242 } else {
243 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
244 }
245 __ call(fn);
246 break;
247 case Interpreter::java_lang_math_exp :
248 entry_point = __ pc();
249 __ fld(f10, Address(esp));
250 __ mv(sp, x19_sender_sp);
251 __ mv(x9, ra);
252 continuation = x9; // The first callee-saved register
253 if (StubRoutines::dexp() == nullptr) {
254 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
255 } else {
256 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
257 }
258 __ call(fn);
259 break;
260 case Interpreter::java_lang_math_pow :
261 entry_point = __ pc();
262 __ mv(x9, ra);
263 continuation = x9;
264 __ fld(f10, Address(esp, 2 * Interpreter::stackElementSize));
265 __ fld(f11, Address(esp));
266 __ mv(sp, x19_sender_sp);
267 if (StubRoutines::dpow() == nullptr) {
268 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
269 } else {
270 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
271 }
272 __ call(fn);
273 break;
274 case Interpreter::java_lang_math_fmaD :
275 if (UseFMA) {
276 entry_point = __ pc();
277 __ fld(f10, Address(esp, 4 * Interpreter::stackElementSize));
278 __ fld(f11, Address(esp, 2 * Interpreter::stackElementSize));
279 __ fld(f12, Address(esp));
280 __ fmadd_d(f10, f10, f11, f12);
281 __ mv(sp, x19_sender_sp); // Restore caller's SP
282 }
283 break;
284 case Interpreter::java_lang_math_fmaF :
285 if (UseFMA) {
286 entry_point = __ pc();
287 __ flw(f10, Address(esp, 2 * Interpreter::stackElementSize));
288 __ flw(f11, Address(esp, Interpreter::stackElementSize));
289 __ flw(f12, Address(esp));
290 __ fmadd_s(f10, f10, f11, f12);
291 __ mv(sp, x19_sender_sp); // Restore caller's SP
292 }
293 break;
294 default:
295 ;
296 }
297 if (entry_point != nullptr) {
298 __ jr(continuation);
299 }
300
301 return entry_point;
302 }
303
304 // Abstract method entry
305 // Attempt to execute abstract method. Throw exception
306 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
307 // xmethod: Method*
308 // x19_sender_sp: sender SP
309
310 address entry_point = __ pc();
311
312 // abstract method entry
313
314 // pop return address, reset last_sp to null
315 __ empty_expression_stack();
316 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
317 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
318
319 // throw exception
320 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
321 InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
322 xmethod);
323 // the call_VM checks for exception, so we should never return here.
324 __ should_not_reach_here();
325
326 return entry_point;
327 }
328
329 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
330 address entry = __ pc();
331
332 #ifdef ASSERT
333 {
334 Label L;
335 __ ld(t0, Address(fp, frame::interpreter_frame_monitor_block_top_offset * wordSize));
336 __ shadd(t0, t0, fp, t0, LogBytesPerWord);
337 // maximal sp for current fp (stack grows negative)
338 // check if frame is complete
339 __ bge(t0, sp, L);
340 __ stop ("interpreter frame not set up");
341 __ bind(L);
342 }
343 #endif // ASSERT
344 // Restore bcp under the assumption that the current frame is still
345 // interpreted
346 __ restore_bcp();
347
348 // expression stack must be empty before entering the VM if an
349 // exception happened
350 __ empty_expression_stack();
351 // throw exception
352 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
353 return entry;
354 }
355
356 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
357 address entry = __ pc();
358 // expression stack must be empty before entering the VM if an
359 // exception happened
360 __ empty_expression_stack();
361 // setup parameters
362
363 // convention: expect aberrant index in register x11
364 __ zero_extend(c_rarg2, x11, 32);
365 // convention: expect array in register x13
366 __ mv(c_rarg1, x13);
367 __ call_VM(noreg,
368 CAST_FROM_FN_PTR(address,
369 InterpreterRuntime::
370 throw_ArrayIndexOutOfBoundsException),
371 c_rarg1, c_rarg2);
372 return entry;
373 }
374
375 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
376 address entry = __ pc();
377
378 // object is at TOS
379 __ pop_reg(c_rarg1);
380
381 // expression stack must be empty before entering the VM if an
382 // exception happened
383 __ empty_expression_stack();
384
385 __ call_VM(noreg,
386 CAST_FROM_FN_PTR(address,
387 InterpreterRuntime::
388 throw_ClassCastException),
389 c_rarg1);
390 return entry;
391 }
392
393 address TemplateInterpreterGenerator::generate_exception_handler_common(
394 const char* name, const char* message, bool pass_oop) {
395 assert(!pass_oop || message == nullptr, "either oop or message but not both");
396 address entry = __ pc();
397 if (pass_oop) {
398 // object is at TOS
399 __ pop_reg(c_rarg2);
400 }
401 // expression stack must be empty before entering the VM if an
402 // exception happened
403 __ empty_expression_stack();
404 // setup parameters
405 __ la(c_rarg1, Address((address)name));
406 if (pass_oop) {
407 __ call_VM(x10, CAST_FROM_FN_PTR(address,
408 InterpreterRuntime::
409 create_klass_exception),
410 c_rarg1, c_rarg2);
411 } else {
412 // kind of lame ExternalAddress can't take null because
413 // external_word_Relocation will assert.
414 if (message != nullptr) {
415 __ la(c_rarg2, Address((address)message));
416 } else {
417 __ mv(c_rarg2, NULL_WORD);
418 }
419 __ call_VM(x10,
420 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
421 c_rarg1, c_rarg2);
422 }
423 // throw exception
424 __ j(address(Interpreter::throw_exception_entry()));
425 return entry;
426 }
427
428 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
429 address entry = __ pc();
430
431 // Restore stack bottom in case i2c adjusted stack
432 __ ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
433 __ shadd(esp, t0, fp, t0, LogBytesPerWord);
434 // and null it as marker that esp is now tos until next java call
435 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
436 __ restore_bcp();
437 __ restore_locals();
438 __ restore_constant_pool_cache();
439 __ get_method(xmethod);
440
441 if (state == atos) {
442 Register obj = x10;
443 Register mdp = x11;
444 Register tmp = x12;
445 __ ld(mdp, Address(xmethod, Method::method_data_offset()));
446 __ profile_return_type(mdp, obj, tmp);
447 }
448
449 const Register cache = x11;
450 const Register index = x12;
451
452 if (index_size == sizeof(u4)) {
453 __ load_resolved_indy_entry(cache, index);
454 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
455 __ shadd(esp, cache, esp, t0, 3);
456 } else {
457 // Pop N words from the stack
458 assert(index_size == sizeof(u2), "Can only be u2");
459 __ load_method_entry(cache, index);
460 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
461
462 __ shadd(esp, cache, esp, t0, 3);
463 }
464
465 // Restore machine SP
466 __ restore_sp_after_call();
467
468 __ check_and_handle_popframe(xthread);
469 __ check_and_handle_earlyret(xthread);
470
471 __ get_dispatch();
472 __ dispatch_next(state, step);
473
474 return entry;
475 }
476
477 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
478 int step,
479 address continuation) {
480 address entry = __ pc();
481 __ restore_bcp();
482 __ restore_locals();
483 __ restore_constant_pool_cache();
484 __ get_method(xmethod);
485 __ get_dispatch();
486
487 __ restore_sp_after_call(); // Restore SP to extended SP
488
489 // Restore expression stack pointer
490 __ ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
491 __ shadd(esp, t0, fp, t0, LogBytesPerWord);
492 // null last_sp until next java call
493 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
494
495 // handle exceptions
496 {
497 Label L;
498 __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
499 __ beqz(t0, L);
500 __ call_VM(noreg,
501 CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
502 __ should_not_reach_here();
503 __ bind(L);
504 }
505
506 if (continuation == nullptr) {
507 __ dispatch_next(state, step);
508 } else {
509 __ jump_to_entry(continuation);
510 }
511 return entry;
512 }
513
514 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
515 address entry = __ pc();
516 if (type == T_OBJECT) {
517 // retrieve result from frame
518 __ ld(x10, Address(fp, frame::interpreter_frame_oop_temp_offset * wordSize));
519 // and verify it
520 __ verify_oop(x10);
521 } else {
522 __ cast_primitive_type(type, x10);
523 }
524
525 __ ret(); // return from result handler
526 return entry;
527 }
528
529 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state,
530 address runtime_entry) {
531 assert_cond(runtime_entry != nullptr);
532 address entry = __ pc();
533 __ push(state);
534 __ push_cont_fastpath(xthread);
535 __ call_VM(noreg, runtime_entry);
536 __ pop_cont_fastpath(xthread);
537 __ membar(MacroAssembler::AnyAny);
538 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
539 return entry;
540 }
541
542 address TemplateInterpreterGenerator::generate_cont_preempt_rerun_interpreter_adapter() {
543 return nullptr;
544 }
545
546
547 // Helpers for commoning out cases in the various type of method entries.
548 //
549
550
551 // increment invocation count & check for overflow
552 //
553 // Note: checking for negative value instead of overflow
554 // so we have a 'sticky' overflow test
555 //
556 // xmethod: method
557 //
558 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
559 Label done;
560 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
561 int increment = InvocationCounter::count_increment;
562 Label no_mdo;
563 if (ProfileInterpreter) {
564 // Are we profiling?
565 __ ld(x10, Address(xmethod, Method::method_data_offset()));
566 __ beqz(x10, no_mdo);
567 // Increment counter in the MDO
568 const Address mdo_invocation_counter(x10, in_bytes(MethodData::invocation_counter_offset()) +
569 in_bytes(InvocationCounter::counter_offset()));
570 const Address mask(x10, in_bytes(MethodData::invoke_mask_offset()));
571 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, t0, t1, false, overflow);
572 __ j(done);
573 }
574 __ bind(no_mdo);
575 // Increment counter in MethodCounters
576 const Address invocation_counter(t1,
577 MethodCounters::invocation_counter_offset() +
578 InvocationCounter::counter_offset());
579 __ get_method_counters(xmethod, t1, done);
580 const Address mask(t1, in_bytes(MethodCounters::invoke_mask_offset()));
581 __ increment_mask_and_jump(invocation_counter, increment, mask, t0, x11, false, overflow);
582 __ bind(done);
583 }
584
585 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
586 __ mv(c_rarg1, zr);
587 __ call_VM(noreg,
588 CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), c_rarg1);
589 __ j(do_continue);
590 }
591
592 // See if we've got enough room on the stack for locals plus overhead
593 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
594 // without going through the signal handler, i.e., reserved and yellow zones
595 // will not be made usable. The shadow zone must suffice to handle the
596 // overflow.
597 // The expression stack grows down incrementally, so the normal guard
598 // page mechanism will work for that.
599 //
600 // NOTE: Since the additional locals are also always pushed (wasn't
601 // obvious in generate_method_entry) so the guard should work for them
602 // too.
603 //
604 // Args:
605 // x13: number of additional locals this frame needs (what we must check)
606 // xmethod: Method*
607 //
608 // Kills:
609 // x10
610 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
611
612 // monitor entry size: see picture of stack set
613 // (generate_method_entry) and frame_amd64.hpp
614 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
615
616 // total overhead size: entry_size + (saved fp through expr stack
617 // bottom). be sure to change this if you add/subtract anything
618 // to/from the overhead area
619 const int overhead_size =
620 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
621
622 const int page_size = (int)os::vm_page_size();
623
624 Label after_frame_check;
625
626 // see if the frame is greater than one page in size. If so,
627 // then we need to verify there is enough stack space remaining
628 // for the additional locals.
629 __ mv(t0, (page_size - overhead_size) / Interpreter::stackElementSize);
630 __ bleu(x13, t0, after_frame_check);
631
632 // compute sp as if this were going to be the last frame on
633 // the stack before the red zone
634
635 // locals + overhead, in bytes
636 __ mv(x10, overhead_size);
637 __ shadd(x10, x13, x10, t0, Interpreter::logStackElementSize); // 2 slots per parameter.
638
639 const Address stack_limit(xthread, JavaThread::stack_overflow_limit_offset());
640 __ ld(t0, stack_limit);
641
642 #ifdef ASSERT
643 Label limit_okay;
644 // Verify that thread stack limit is non-zero.
645 __ bnez(t0, limit_okay);
646 __ stop("stack overflow limit is zero");
647 __ bind(limit_okay);
648 #endif
649
650 // Add stack limit to locals.
651 __ add(x10, x10, t0);
652
653 // Check against the current stack bottom.
654 __ bgtu(sp, x10, after_frame_check);
655
656 // Remove the incoming args, peeling the machine SP back to where it
657 // was in the caller. This is not strictly necessary, but unless we
658 // do so the stack frame may have a garbage FP; this ensures a
659 // correct call stack that we can always unwind. The ANDI should be
660 // unnecessary because the sender SP in x19 is always aligned, but
661 // it doesn't hurt.
662 __ andi(sp, x19_sender_sp, -16);
663
664 // Note: the restored frame is not necessarily interpreted.
665 // Use the shared runtime version of the StackOverflowError.
666 assert(StubRoutines::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
667 __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
668
669 // all done with frame size check
670 __ bind(after_frame_check);
671 }
672
673 // Allocate monitor and lock method (asm interpreter)
674 //
675 // Args:
676 // xmethod: Method*
677 // xlocals: locals
678 //
679 // Kills:
680 // x10
681 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
682 // t0, t1 (temporary regs)
683 void TemplateInterpreterGenerator::lock_method() {
684 // synchronize method
685 const Address access_flags(xmethod, Method::access_flags_offset());
686 const Address monitor_block_top(fp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
687 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
688
689 #ifdef ASSERT
690 __ lwu(x10, access_flags);
691 __ verify_access_flags(x10, JVM_ACC_SYNCHRONIZED, "method doesn't need synchronization", false);
692 #endif // ASSERT
693
694 // get synchronization object
695 {
696 Label done;
697 __ lwu(x10, access_flags);
698 __ andi(t0, x10, JVM_ACC_STATIC);
699 // get receiver (assume this is frequent case)
700 __ ld(x10, Address(xlocals, Interpreter::local_offset_in_bytes(0)));
701 __ beqz(t0, done);
702 __ load_mirror(x10, xmethod, x15, t1);
703
704 #ifdef ASSERT
705 {
706 Label L;
707 __ bnez(x10, L);
708 __ stop("synchronization object is null");
709 __ bind(L);
710 }
711 #endif // ASSERT
712
713 __ bind(done);
714 }
715
716 // add space for monitor & lock
717 __ check_extended_sp();
718 __ add(sp, sp, - entry_size); // add space for a monitor entry
719 __ add(esp, esp, - entry_size);
720 __ sub(t0, sp, fp);
721 __ srai(t0, t0, Interpreter::logStackElementSize);
722 __ sd(t0, Address(fp, frame::interpreter_frame_extended_sp_offset * wordSize));
723 __ sub(t0, esp, fp);
724 __ srai(t0, t0, Interpreter::logStackElementSize);
725 __ sd(t0, monitor_block_top); // set new monitor block top
726 // store object
727 __ sd(x10, Address(esp, BasicObjectLock::obj_offset()));
728 __ mv(c_rarg1, esp); // object address
729 __ lock_object(c_rarg1);
730 }
731
732 // Generate a fixed interpreter frame. This is identical setup for
733 // interpreted methods and for native methods hence the shared code.
734 //
735 // Args:
736 // ra: return address
737 // xmethod: Method*
738 // xlocals: pointer to locals
739 // xcpool: cp cache
740 // stack_pointer: previous sp
741 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
742 // initialize fixed part of activation frame
743 if (native_call) {
744 __ add(esp, sp, - 14 * wordSize);
745 __ mv(xbcp, zr);
746 __ add(sp, sp, - 14 * wordSize);
747 // add 2 zero-initialized slots for native calls
748 __ sd(zr, Address(sp, 13 * wordSize));
749 __ sd(zr, Address(sp, 12 * wordSize));
750 } else {
751 __ add(esp, sp, - 12 * wordSize);
752 __ ld(t0, Address(xmethod, Method::const_offset())); // get ConstMethod
753 __ add(xbcp, t0, in_bytes(ConstMethod::codes_offset())); // get codebase
754 __ add(sp, sp, - 12 * wordSize);
755 }
756 __ sd(xbcp, Address(sp, wordSize));
757 __ mv(t0, frame::interpreter_frame_initial_sp_offset);
758 __ sd(t0, Address(sp, 0));
759
760 if (ProfileInterpreter) {
761 Label method_data_continue;
762 __ ld(t0, Address(xmethod, Method::method_data_offset()));
763 __ beqz(t0, method_data_continue);
764 __ la(t0, Address(t0, in_bytes(MethodData::data_offset())));
765 __ bind(method_data_continue);
766 }
767
768 __ sd(xmethod, Address(sp, 7 * wordSize));
769 __ sd(ProfileInterpreter ? t0 : zr, Address(sp, 6 * wordSize));
770
771 __ sd(ra, Address(sp, 11 * wordSize));
772 __ sd(fp, Address(sp, 10 * wordSize));
773 __ la(fp, Address(sp, 12 * wordSize)); // include ra & fp
774
775 __ ld(xcpool, Address(xmethod, Method::const_offset()));
776 __ ld(xcpool, Address(xcpool, ConstMethod::constants_offset()));
777 __ ld(xcpool, Address(xcpool, ConstantPool::cache_offset()));
778 __ sd(xcpool, Address(sp, 3 * wordSize));
779 __ sub(t0, xlocals, fp);
780 __ srai(t0, t0, Interpreter::logStackElementSize); // t0 = xlocals - fp();
781 // Store relativized xlocals, see frame::interpreter_frame_locals().
782 __ sd(t0, Address(sp, 2 * wordSize));
783
784 // set sender sp
785 // leave last_sp as null
786 __ sd(x19_sender_sp, Address(sp, 9 * wordSize));
787 __ sd(zr, Address(sp, 8 * wordSize));
788
789 // Get mirror and store it in the frame as GC root for this Method*
790 __ load_mirror(t2, xmethod, x15, t1);
791 __ sd(t2, Address(sp, 4 * wordSize));
792
793 if (!native_call) {
794 __ ld(t0, Address(xmethod, Method::const_offset()));
795 __ lhu(t0, Address(t0, ConstMethod::max_stack_offset()));
796 __ add(t0, t0, MAX2(3, Method::extra_stack_entries()));
797 __ slli(t0, t0, 3);
798 __ sub(t0, sp, t0);
799 __ andi(t0, t0, -16);
800 __ sub(t1, t0, fp);
801 __ srai(t1, t1, Interpreter::logStackElementSize);
802 // Store extended SP
803 __ sd(t1, Address(sp, 5 * wordSize));
804 // Move SP out of the way
805 __ mv(sp, t0);
806 } else {
807 // Make sure there is room for the exception oop pushed in case method throws
808 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
809 __ sub(t0, sp, 2 * wordSize);
810 __ sub(t1, t0, fp);
811 __ srai(t1, t1, Interpreter::logStackElementSize);
812 __ sd(t1, Address(sp, 5 * wordSize));
813 __ mv(sp, t0);
814 }
815 }
816
817 // End of helpers
818
819 // Various method entries
820 //------------------------------------------------------------------------------------------------------------------------
821 //
822 //
823
824 // Method entry for java.lang.ref.Reference.get.
825 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
826 // Code: _aload_0, _getfield, _areturn
827 // parameter size = 1
828 //
829 // The code that gets generated by this routine is split into 2 parts:
830 // 1. The "intrinsified" code for G1 (or any SATB based GC),
831 // 2. The slow path - which is an expansion of the regular method entry.
832 //
833 // Notes:-
834 // * In the G1 code we do not check whether we need to block for
835 // a safepoint. If G1 is enabled then we must execute the specialized
836 // code for Reference.get (except when the Reference object is null)
837 // so that we can log the value in the referent field with an SATB
838 // update buffer.
839 // If the code for the getfield template is modified so that the
840 // G1 pre-barrier code is executed when the current method is
841 // Reference.get() then going through the normal method entry
842 // will be fine.
843 // * The G1 code can, however, check the receiver object (the instance
844 // of java.lang.Reference) and jump to the slow path if null. If the
845 // Reference object is null then we obviously cannot fetch the referent
846 // and so we don't need to call the G1 pre-barrier. Thus we can use the
847 // regular method entry code to generate the NPE.
848 //
849 // This code is based on generate_accessor_entry.
850 //
851 // xmethod: Method*
852 // x19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
853
854 // ra is live. It must be saved around calls.
855
856 address entry = __ pc();
857
858 const int referent_offset = java_lang_ref_Reference::referent_offset();
859 guarantee(referent_offset > 0, "referent offset not initialized");
860
861 Label slow_path;
862 const Register local_0 = c_rarg0;
863 // Check if local 0 isn't null
864 // If the receiver is null then it is OK to jump to the slow path.
865 __ ld(local_0, Address(esp, 0));
866 __ beqz(local_0, slow_path);
867
868 // Load the value of the referent field.
869 const Address field_address(local_0, referent_offset);
870 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
871 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ t0, /*tmp2*/ t1);
872
873 // areturn
874 __ andi(sp, x19_sender_sp, -16); // done with stack
875 __ ret();
876
877 // generate a vanilla interpreter entry as the slow path
878 __ bind(slow_path);
879 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
880 return entry;
881 }
882
883 /**
884 * Method entry for static native methods:
885 * int java.util.zip.CRC32.update(int crc, int b)
886 */
887 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
888 // TODO: Unimplemented generate_CRC32_update_entry
889 return nullptr;
890 }
891
892 /**
893 * Method entry for static native methods:
894 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
895 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
896 */
897 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
898 // TODO: Unimplemented generate_CRC32_updateBytes_entry
899 return nullptr;
900 }
901
902 /**
903 * Method entry for intrinsic-candidate (non-native) methods:
904 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
905 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
906 * Unlike CRC32, CRC32C does not have any methods marked as native
907 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
908 */
909 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
910 // TODO: Unimplemented generate_CRC32C_updateBytes_entry
911 return nullptr;
912 }
913
914 // Not supported
915 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
916 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
917 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
918 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
919 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() { return nullptr; }
920 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() { return nullptr; }
921
922 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
923 // See more discussion in stackOverflow.hpp.
924
925 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
926 const int page_size = (int)os::vm_page_size();
927 const int n_shadow_pages = shadow_zone_size / page_size;
928
929 #ifdef ASSERT
930 Label L_good_limit;
931 __ ld(t0, Address(xthread, JavaThread::shadow_zone_safe_limit()));
932 __ bnez(t0, L_good_limit);
933 __ stop("shadow zone safe limit is not initialized");
934 __ bind(L_good_limit);
935
936 Label L_good_watermark;
937 __ ld(t0, Address(xthread, JavaThread::shadow_zone_growth_watermark()));
938 __ bnez(t0, L_good_watermark);
939 __ stop("shadow zone growth watermark is not initialized");
940 __ bind(L_good_watermark);
941 #endif
942
943 Label L_done;
944
945 __ ld(t0, Address(xthread, JavaThread::shadow_zone_growth_watermark()));
946 __ bgtu(sp, t0, L_done);
947
948 for (int p = 1; p <= n_shadow_pages; p++) {
949 __ bang_stack_with_offset(p * page_size);
950 }
951
952 // Record the new watermark, but only if the update is above the safe limit.
953 // Otherwise, the next time around the check above would pass the safe limit.
954 __ ld(t0, Address(xthread, JavaThread::shadow_zone_safe_limit()));
955 __ bleu(sp, t0, L_done);
956 __ sd(sp, Address(xthread, JavaThread::shadow_zone_growth_watermark()));
957
958 __ bind(L_done);
959 }
960
961 // Interpreter stub for calling a native method. (asm interpreter)
962 // This sets up a somewhat different looking stack for calling the
963 // native method than the typical interpreter frame setup.
964 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
965 // determine code generation flags
966 bool inc_counter = UseCompiler || CountCompiledCalls;
967
968 // x11: Method*
969 // x30: sender sp
970
971 address entry_point = __ pc();
972
973 const Address constMethod (xmethod, Method::const_offset());
974 const Address access_flags (xmethod, Method::access_flags_offset());
975 const Address size_of_parameters(x12, ConstMethod::
976 size_of_parameters_offset());
977
978 // get parameter size (always needed)
979 __ ld(x12, constMethod);
980 __ load_unsigned_short(x12, size_of_parameters);
981
982 // Native calls don't need the stack size check since they have no
983 // expression stack and the arguments are already on the stack and
984 // we only add a handful of words to the stack.
985
986 // xmethod: Method*
987 // x12: size of parameters
988 // x30: sender sp
989
990 // for natives the size of locals is zero
991
992 // compute beginning of parameters (xlocals)
993 __ shadd(xlocals, x12, esp, xlocals, 3);
994 __ addi(xlocals, xlocals, -wordSize);
995
996 // Pull SP back to minimum size: this avoids holes in the stack
997 __ andi(sp, esp, -16);
998
999 // initialize fixed part of activation frame
1000 generate_fixed_frame(true);
1001
1002 // make sure method is native & not abstract
1003 #ifdef ASSERT
1004 __ lwu(x10, access_flags);
1005 __ verify_access_flags(x10, JVM_ACC_NATIVE, "tried to execute non-native method as native", false);
1006 __ verify_access_flags(x10, JVM_ACC_ABSTRACT, "tried to execute abstract method in interpreter");
1007 #endif
1008
1009 // Since at this point in the method invocation the exception
1010 // handler would try to exit the monitor of synchronized methods
1011 // which hasn't been entered yet, we set the thread local variable
1012 // _do_not_unlock_if_synchronized to true. The remove_activation
1013 // will check this flag.
1014
1015 const Address do_not_unlock_if_synchronized(xthread,
1016 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1017 __ mv(t1, true);
1018 __ sb(t1, do_not_unlock_if_synchronized);
1019
1020 // increment invocation count & check for overflow
1021 Label invocation_counter_overflow;
1022 if (inc_counter) {
1023 generate_counter_incr(&invocation_counter_overflow);
1024 }
1025
1026 Label continue_after_compile;
1027 __ bind(continue_after_compile);
1028
1029 bang_stack_shadow_pages(true);
1030
1031 // reset the _do_not_unlock_if_synchronized flag
1032 __ sb(zr, do_not_unlock_if_synchronized);
1033
1034 // check for synchronized methods
1035 // Must happen AFTER invocation_counter check and stack overflow check,
1036 // so method is not locked if overflows.
1037 if (synchronized) {
1038 lock_method();
1039 } else {
1040 // no synchronization necessary
1041 #ifdef ASSERT
1042 __ lwu(x10, access_flags);
1043 __ verify_access_flags(x10, JVM_ACC_SYNCHRONIZED, "method needs synchronization");
1044 #endif
1045 }
1046
1047 // start execution
1048 #ifdef ASSERT
1049 __ verify_frame_setup();
1050 #endif
1051
1052 // jvmti support
1053 __ notify_method_entry();
1054
1055 // work registers
1056 const Register t = x18;
1057 const Register result_handler = x19;
1058
1059 // allocate space for parameters
1060 __ ld(t, Address(xmethod, Method::const_offset()));
1061 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1062
1063 __ slli(t, t, Interpreter::logStackElementSize);
1064 __ sub(x30, esp, t);
1065 __ andi(sp, x30, -16);
1066 __ mv(esp, x30);
1067
1068 // get signature handler
1069 {
1070 Label L;
1071 __ ld(t, Address(xmethod, Method::signature_handler_offset()));
1072 __ bnez(t, L);
1073 __ call_VM(noreg,
1074 CAST_FROM_FN_PTR(address,
1075 InterpreterRuntime::prepare_native_call),
1076 xmethod);
1077 __ ld(t, Address(xmethod, Method::signature_handler_offset()));
1078 __ bind(L);
1079 }
1080
1081 // call signature handler
1082 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == xlocals,
1083 "adjust this code");
1084 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1085 "adjust this code");
1086 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t0,
1087 "adjust this code");
1088
1089 // The generated handlers do not touch xmethod (the method).
1090 // However, large signatures cannot be cached and are generated
1091 // each time here. The slow-path generator can do a GC on return,
1092 // so we must reload it after the call.
1093 __ jalr(t);
1094 __ get_method(xmethod); // slow path can do a GC, reload xmethod
1095
1096
1097 // result handler is in x10
1098 // set result handler
1099 __ mv(result_handler, x10);
1100 // pass mirror handle if static call
1101 {
1102 Label L;
1103 __ lwu(t, Address(xmethod, Method::access_flags_offset()));
1104 __ test_bit(t0, t, exact_log2(JVM_ACC_STATIC));
1105 __ beqz(t0, L);
1106 // get mirror
1107 __ load_mirror(t, xmethod, x28, t1);
1108 // copy mirror into activation frame
1109 __ sd(t, Address(fp, frame::interpreter_frame_oop_temp_offset * wordSize));
1110 // pass handle to mirror
1111 __ addi(c_rarg1, fp, frame::interpreter_frame_oop_temp_offset * wordSize);
1112 __ bind(L);
1113 }
1114
1115 // get native function entry point in x28
1116 {
1117 Label L;
1118 __ ld(x28, Address(xmethod, Method::native_function_offset()));
1119 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1120 __ mv(t, unsatisfied);
1121 __ load_long_misaligned(t1, Address(t, 0), t0, 2); // 2 bytes aligned, but not 4 or 8
1122
1123 __ bne(x28, t1, L);
1124 __ call_VM(noreg,
1125 CAST_FROM_FN_PTR(address,
1126 InterpreterRuntime::prepare_native_call),
1127 xmethod);
1128 __ get_method(xmethod);
1129 __ ld(x28, Address(xmethod, Method::native_function_offset()));
1130 __ bind(L);
1131 }
1132
1133 // pass JNIEnv
1134 __ add(c_rarg0, xthread, in_bytes(JavaThread::jni_environment_offset()));
1135
1136 // It is enough that the pc() points into the right code
1137 // segment. It does not have to be the correct return pc.
1138 Label native_return;
1139 __ set_last_Java_frame(esp, fp, native_return, x30);
1140
1141 // change thread state
1142 #ifdef ASSERT
1143 {
1144 Label L;
1145 __ lwu(t, Address(xthread, JavaThread::thread_state_offset()));
1146 __ addi(t0, zr, (u1)_thread_in_Java);
1147 __ beq(t, t0, L);
1148 __ stop("Wrong thread state in native stub");
1149 __ bind(L);
1150 }
1151 #endif
1152
1153 // Change state to native
1154 __ la(t1, Address(xthread, JavaThread::thread_state_offset()));
1155 __ mv(t0, _thread_in_native);
1156 __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1157 __ sw(t0, Address(t1));
1158
1159 // Call the native method.
1160 __ jalr(x28);
1161 __ bind(native_return);
1162 __ get_method(xmethod);
1163 // result potentially in x10 or f10
1164
1165 // Restore cpu control state after JNI call
1166 __ restore_cpu_control_state_after_jni(t0);
1167
1168 // make room for the pushes we're about to do
1169 __ sub(t0, esp, 4 * wordSize);
1170 __ andi(sp, t0, -16);
1171
1172 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1173 // in order to extract the result of a method call. If the order of these
1174 // pushes change or anything else is added to the stack then the code in
1175 // interpreter_frame_result must also change.
1176 __ push(dtos);
1177 __ push(ltos);
1178
1179 // change thread state
1180 // Force all preceding writes to be observed prior to thread state change
1181 __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1182
1183 __ mv(t0, _thread_in_native_trans);
1184 __ sw(t0, Address(xthread, JavaThread::thread_state_offset()));
1185
1186 // Force this write out before the read below
1187 if (!UseSystemMemoryBarrier) {
1188 __ membar(MacroAssembler::AnyAny);
1189 }
1190
1191 // check for safepoint operation in progress and/or pending suspend requests
1192 {
1193 Label L, Continue;
1194
1195 // We need an acquire here to ensure that any subsequent load of the
1196 // global SafepointSynchronize::_state flag is ordered after this load
1197 // of the thread-local polling word. We don't want this poll to
1198 // return false (i.e. not safepointing) and a later poll of the global
1199 // SafepointSynchronize::_state spuriously to return true.
1200 //
1201 // This is to avoid a race when we're in a native->Java transition
1202 // racing the code which wakes up from a safepoint.
1203 __ safepoint_poll(L, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1204 __ lwu(t1, Address(xthread, JavaThread::suspend_flags_offset()));
1205 __ beqz(t1, Continue);
1206 __ bind(L);
1207
1208 // Don't use call_VM as it will see a possible pending exception
1209 // and forward it and never return here preventing us from
1210 // clearing _last_native_pc down below. So we do a runtime call by
1211 // hand.
1212 //
1213 __ mv(c_rarg0, xthread);
1214 __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1215 __ get_method(xmethod);
1216 __ reinit_heapbase();
1217 __ bind(Continue);
1218 }
1219
1220 // change thread state
1221 // Force all preceding writes to be observed prior to thread state change
1222 __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1223
1224 __ mv(t0, _thread_in_Java);
1225 __ sw(t0, Address(xthread, JavaThread::thread_state_offset()));
1226
1227 // reset_last_Java_frame
1228 __ reset_last_Java_frame(true);
1229
1230 if (CheckJNICalls) {
1231 // clear_pending_jni_exception_check
1232 __ sd(zr, Address(xthread, JavaThread::pending_jni_exception_check_fn_offset()));
1233 }
1234
1235 // reset handle block
1236 __ ld(t, Address(xthread, JavaThread::active_handles_offset()));
1237 __ sd(zr, Address(t, JNIHandleBlock::top_offset()));
1238
1239 // If result is an oop unbox and store it in frame where gc will see it
1240 // and result handler will pick it up
1241
1242 {
1243 Label no_oop;
1244 __ la(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1245 __ bne(t, result_handler, no_oop);
1246 // Unbox oop result, e.g. JNIHandles::resolve result.
1247 __ pop(ltos);
1248 __ resolve_jobject(x10, t, t1);
1249 __ sd(x10, Address(fp, frame::interpreter_frame_oop_temp_offset * wordSize));
1250 // keep stack depth as expected by pushing oop which will eventually be discarded
1251 __ push(ltos);
1252 __ bind(no_oop);
1253 }
1254
1255 {
1256 Label no_reguard;
1257 __ lwu(t0, Address(xthread, in_bytes(JavaThread::stack_guard_state_offset())));
1258 __ addi(t1, zr, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1259 __ bne(t0, t1, no_reguard);
1260
1261 __ push_call_clobbered_registers();
1262 __ mv(c_rarg0, xthread);
1263 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1264 __ pop_call_clobbered_registers();
1265 __ bind(no_reguard);
1266 }
1267
1268 // The method register is junk from after the thread_in_native transition
1269 // until here. Also can't call_VM until the bcp has been
1270 // restored. Need bcp for throwing exception below so get it now.
1271 __ get_method(xmethod);
1272
1273 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1274 // xbcp == code_base()
1275 __ ld(xbcp, Address(xmethod, Method::const_offset())); // get ConstMethod*
1276 __ add(xbcp, xbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1277 // handle exceptions (exception handling will handle unlocking!)
1278 {
1279 Label L;
1280 __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
1281 __ beqz(t0, L);
1282 // Note: At some point we may want to unify this with the code
1283 // used in call_VM_base(); i.e., we should use the
1284 // StubRoutines::forward_exception code. For now this doesn't work
1285 // here because the sp is not correctly set at this point.
1286 __ MacroAssembler::call_VM(noreg,
1287 CAST_FROM_FN_PTR(address,
1288 InterpreterRuntime::throw_pending_exception));
1289 __ should_not_reach_here();
1290 __ bind(L);
1291 }
1292
1293 // do unlocking if necessary
1294 {
1295 Label L;
1296 __ lwu(t, Address(xmethod, Method::access_flags_offset()));
1297 __ test_bit(t0, t, exact_log2(JVM_ACC_SYNCHRONIZED));
1298 __ beqz(t0, L);
1299 // the code below should be shared with interpreter macro
1300 // assembler implementation
1301 {
1302 Label unlock;
1303 // BasicObjectLock will be first in list, since this is a
1304 // synchronized method. However, need to check that the object
1305 // has not been unlocked by an explicit monitorexit bytecode.
1306
1307 // monitor expect in c_rarg1 for slow unlock path
1308 __ la(c_rarg1, Address(fp, // address of first monitor
1309 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1310 wordSize - sizeof(BasicObjectLock))));
1311
1312 __ ld(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1313 __ bnez(t, unlock);
1314
1315 // Entry already unlocked, need to throw exception
1316 __ MacroAssembler::call_VM(noreg,
1317 CAST_FROM_FN_PTR(address,
1318 InterpreterRuntime::throw_illegal_monitor_state_exception));
1319 __ should_not_reach_here();
1320
1321 __ bind(unlock);
1322 __ unlock_object(c_rarg1);
1323 }
1324 __ bind(L);
1325 }
1326
1327 // jvmti support
1328 // Note: This must happen _after_ handling/throwing any exceptions since
1329 // the exception handler code notifies the runtime of method exits
1330 // too. If this happens before, method entry/exit notifications are
1331 // not properly paired (was bug - gri 11/22/99).
1332 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1333
1334 __ pop(ltos);
1335 __ pop(dtos);
1336
1337 __ jalr(result_handler);
1338
1339 // remove activation
1340 __ ld(esp, Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1341 // remove frame anchor
1342 __ leave();
1343
1344 // restore sender sp
1345 __ mv(sp, esp);
1346
1347 __ ret();
1348
1349 if (inc_counter) {
1350 // Handle overflow of counter and compile method
1351 __ bind(invocation_counter_overflow);
1352 generate_counter_overflow(continue_after_compile);
1353 }
1354
1355 return entry_point;
1356 }
1357
1358 //
1359 // Generic interpreted method entry to (asm) interpreter
1360 //
1361 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1362
1363 // determine code generation flags
1364 const bool inc_counter = UseCompiler || CountCompiledCalls;
1365
1366 // t0: sender sp
1367 address entry_point = __ pc();
1368
1369 const Address constMethod(xmethod, Method::const_offset());
1370 const Address access_flags(xmethod, Method::access_flags_offset());
1371 const Address size_of_parameters(x13,
1372 ConstMethod::size_of_parameters_offset());
1373 const Address size_of_locals(x13, ConstMethod::size_of_locals_offset());
1374
1375 // get parameter size (always needed)
1376 // need to load the const method first
1377 __ ld(x13, constMethod);
1378 __ load_unsigned_short(x12, size_of_parameters);
1379
1380 // x12: size of parameters
1381
1382 __ load_unsigned_short(x13, size_of_locals); // get size of locals in words
1383 __ sub(x13, x13, x12); // x13 = no. of additional locals
1384
1385 // see if we've got enough room on the stack for locals plus overhead.
1386 generate_stack_overflow_check();
1387
1388 // compute beginning of parameters (xlocals)
1389 __ shadd(xlocals, x12, esp, t1, 3);
1390 __ add(xlocals, xlocals, -wordSize);
1391
1392 // Make room for additional locals
1393 __ slli(t1, x13, 3);
1394 __ sub(t0, esp, t1);
1395
1396 // Padding between locals and fixed part of activation frame to ensure
1397 // SP is always 16-byte aligned.
1398 __ andi(sp, t0, -16);
1399
1400 // x13 - # of additional locals
1401 // allocate space for locals
1402 // explicitly initialize locals
1403 {
1404 Label exit, loop;
1405 __ blez(x13, exit); // do nothing if x13 <= 0
1406 __ bind(loop);
1407 __ sd(zr, Address(t0));
1408 __ add(t0, t0, wordSize);
1409 __ add(x13, x13, -1); // until everything initialized
1410 __ bnez(x13, loop);
1411 __ bind(exit);
1412 }
1413
1414 // And the base dispatch table
1415 __ get_dispatch();
1416
1417 // initialize fixed part of activation frame
1418 generate_fixed_frame(false);
1419
1420 // make sure method is not native & not abstract
1421 #ifdef ASSERT
1422 __ lwu(x10, access_flags);
1423 __ verify_access_flags(x10, JVM_ACC_NATIVE, "tried to execute native method as non-native");
1424 __ verify_access_flags(x10, JVM_ACC_ABSTRACT, "tried to execute abstract method in interpreter");
1425 #endif
1426
1427 // Since at this point in the method invocation the exception
1428 // handler would try to exit the monitor of synchronized methods
1429 // which hasn't been entered yet, we set the thread local variable
1430 // _do_not_unlock_if_synchronized to true. The remove_activation
1431 // will check this flag.
1432
1433 const Address do_not_unlock_if_synchronized(xthread,
1434 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1435 __ mv(t1, true);
1436 __ sb(t1, do_not_unlock_if_synchronized);
1437
1438 Label no_mdp;
1439 const Register mdp = x13;
1440 __ ld(mdp, Address(xmethod, Method::method_data_offset()));
1441 __ beqz(mdp, no_mdp);
1442 __ add(mdp, mdp, in_bytes(MethodData::data_offset()));
1443 __ profile_parameters_type(mdp, x11, x12, x14); // use x11, x12, x14 as tmp registers
1444 __ bind(no_mdp);
1445
1446 // increment invocation count & check for overflow
1447 Label invocation_counter_overflow;
1448 if (inc_counter) {
1449 generate_counter_incr(&invocation_counter_overflow);
1450 }
1451
1452 Label continue_after_compile;
1453 __ bind(continue_after_compile);
1454
1455 bang_stack_shadow_pages(false);
1456
1457 // reset the _do_not_unlock_if_synchronized flag
1458 __ sb(zr, do_not_unlock_if_synchronized);
1459
1460 // check for synchronized methods
1461 // Must happen AFTER invocation_counter check and stack overflow check,
1462 // so method is not locked if overflows.
1463 if (synchronized) {
1464 // Allocate monitor and lock method
1465 lock_method();
1466 } else {
1467 // no synchronization necessary
1468 #ifdef ASSERT
1469 __ lwu(x10, access_flags);
1470 __ verify_access_flags(x10, JVM_ACC_SYNCHRONIZED, "method needs synchronization");
1471 #endif
1472 }
1473
1474 // start execution
1475 #ifdef ASSERT
1476 __ verify_frame_setup();
1477 #endif
1478
1479 // jvmti support
1480 __ notify_method_entry();
1481
1482 __ dispatch_next(vtos);
1483
1484 // invocation counter overflow
1485 if (inc_counter) {
1486 // Handle overflow of counter and compile method
1487 __ bind(invocation_counter_overflow);
1488 generate_counter_overflow(continue_after_compile);
1489 }
1490
1491 return entry_point;
1492 }
1493
1494 // Method entry for java.lang.Thread.currentThread
1495 address TemplateInterpreterGenerator::generate_currentThread() {
1496 address entry_point = __ pc();
1497
1498 __ ld(x10, Address(xthread, JavaThread::vthread_offset()));
1499 __ resolve_oop_handle(x10, t0, t1);
1500 __ ret();
1501
1502 return entry_point;
1503 }
1504
1505 //-----------------------------------------------------------------------------
1506 // Exceptions
1507
1508 void TemplateInterpreterGenerator::generate_throw_exception() {
1509 // Entry point in previous activation (i.e., if the caller was
1510 // interpreted)
1511 Interpreter::_rethrow_exception_entry = __ pc();
1512 // Restore sp to interpreter_frame_last_sp even though we are going
1513 // to empty the expression stack for the exception processing.
1514 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1515 // x10: exception
1516 // x13: return address/pc that threw exception
1517 __ restore_bcp(); // xbcp points to call/send
1518 __ restore_locals();
1519 __ restore_constant_pool_cache();
1520 __ reinit_heapbase(); // restore xheapbase as heapbase.
1521 __ get_dispatch();
1522
1523 // Entry point for exceptions thrown within interpreter code
1524 Interpreter::_throw_exception_entry = __ pc();
1525 // If we came here via a NullPointerException on the receiver of a
1526 // method, xthread may be corrupt.
1527 __ get_method(xmethod);
1528 // expression stack is undefined here
1529 // x10: exception
1530 // xbcp: exception bcp
1531 __ verify_oop(x10);
1532 __ mv(c_rarg1, x10);
1533
1534 // expression stack must be empty before entering the VM in case of
1535 // an exception
1536 __ empty_expression_stack();
1537 // find exception handler address and preserve exception oop
1538 __ call_VM(x13,
1539 CAST_FROM_FN_PTR(address,
1540 InterpreterRuntime::exception_handler_for_exception),
1541 c_rarg1);
1542
1543 // Restore machine SP
1544 __ restore_sp_after_call();
1545
1546 // x10: exception handler entry point
1547 // x13: preserved exception oop
1548 // xbcp: bcp for exception handler
1549 __ push_ptr(x13); // push exception which is now the only value on the stack
1550 __ jr(x10); // jump to exception handler (may be _remove_activation_entry!)
1551
1552 // If the exception is not handled in the current frame the frame is
1553 // removed and the exception is rethrown (i.e. exception
1554 // continuation is _rethrow_exception).
1555 //
1556 // Note: At this point the bci is still the bxi for the instruction
1557 // which caused the exception and the expression stack is
1558 // empty. Thus, for any VM calls at this point, GC will find a legal
1559 // oop map (with empty expression stack).
1560
1561 //
1562 // JVMTI PopFrame support
1563 //
1564
1565 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1566 __ empty_expression_stack();
1567 // Set the popframe_processing bit in pending_popframe_condition
1568 // indicating that we are currently handling popframe, so that
1569 // call_VMs that may happen later do not trigger new popframe
1570 // handling cycles.
1571 __ lwu(x13, Address(xthread, JavaThread::popframe_condition_offset()));
1572 __ ori(x13, x13, JavaThread::popframe_processing_bit);
1573 __ sw(x13, Address(xthread, JavaThread::popframe_condition_offset()));
1574
1575 {
1576 // Check to see whether we are returning to a deoptimized frame.
1577 // (The PopFrame call ensures that the caller of the popped frame is
1578 // either interpreted or compiled and deoptimizes it if compiled.)
1579 // In this case, we can't call dispatch_next() after the frame is
1580 // popped, but instead must save the incoming arguments and restore
1581 // them after deoptimization has occurred.
1582 //
1583 // Note that we don't compare the return PC against the
1584 // deoptimization blob's unpack entry because of the presence of
1585 // adapter frames in C2.
1586 Label caller_not_deoptimized;
1587 __ ld(c_rarg1, Address(fp, frame::return_addr_offset * wordSize));
1588 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), c_rarg1);
1589 __ bnez(x10, caller_not_deoptimized);
1590
1591 // Compute size of arguments for saving when returning to
1592 // deoptimized caller
1593 __ get_method(x10);
1594 __ ld(x10, Address(x10, Method::const_offset()));
1595 __ load_unsigned_short(x10, Address(x10, in_bytes(ConstMethod::
1596 size_of_parameters_offset())));
1597 __ slli(x10, x10, Interpreter::logStackElementSize);
1598 __ restore_locals();
1599 __ sub(xlocals, xlocals, x10);
1600 __ add(xlocals, xlocals, wordSize);
1601 // Save these arguments
1602 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1603 Deoptimization::
1604 popframe_preserve_args),
1605 xthread, x10, xlocals);
1606
1607 __ remove_activation(vtos,
1608 /* throw_monitor_exception */ false,
1609 /* install_monitor_exception */ false,
1610 /* notify_jvmdi */ false);
1611
1612 // Inform deoptimization that it is responsible for restoring
1613 // these arguments
1614 __ mv(t0, JavaThread::popframe_force_deopt_reexecution_bit);
1615 __ sw(t0, Address(xthread, JavaThread::popframe_condition_offset()));
1616
1617 // Continue in deoptimization handler
1618 __ ret();
1619
1620 __ bind(caller_not_deoptimized);
1621 }
1622
1623 __ remove_activation(vtos,
1624 /* throw_monitor_exception */ false,
1625 /* install_monitor_exception */ false,
1626 /* notify_jvmdi */ false);
1627
1628 // Restore the last_sp and null it out
1629 __ ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1630 __ shadd(esp, t0, fp, t0, LogBytesPerWord);
1631 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1632
1633 __ restore_bcp();
1634 __ restore_locals();
1635 __ restore_constant_pool_cache();
1636 __ get_method(xmethod);
1637 __ get_dispatch();
1638
1639 // The method data pointer was incremented already during
1640 // call profiling. We have to restore the mdp for the current bcp.
1641 if (ProfileInterpreter) {
1642 __ set_method_data_pointer_for_bcp();
1643 }
1644
1645 // Clear the popframe condition flag
1646 __ sw(zr, Address(xthread, JavaThread::popframe_condition_offset()));
1647 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1648
1649 #if INCLUDE_JVMTI
1650 {
1651 Label L_done;
1652
1653 __ lbu(t0, Address(xbcp, 0));
1654 __ mv(t1, Bytecodes::_invokestatic);
1655 __ bne(t1, t0, L_done);
1656
1657 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1658 // Detect such a case in the InterpreterRuntime function and return the member name argument,or null.
1659
1660 __ ld(c_rarg0, Address(xlocals, 0));
1661 __ call_VM(x10, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null),c_rarg0, xmethod, xbcp);
1662
1663 __ beqz(x10, L_done);
1664
1665 __ sd(x10, Address(esp, 0));
1666 __ bind(L_done);
1667 }
1668 #endif // INCLUDE_JVMTI
1669
1670 // Restore machine SP
1671 __ restore_sp_after_call();
1672
1673 __ dispatch_next(vtos);
1674 // end of PopFrame support
1675
1676 Interpreter::_remove_activation_entry = __ pc();
1677
1678 // preserve exception over this code sequence
1679 __ pop_ptr(x10);
1680 __ sd(x10, Address(xthread, JavaThread::vm_result_offset()));
1681 // remove the activation (without doing throws on illegalMonitorExceptions)
1682 __ remove_activation(vtos, false, true, false);
1683 // restore exception
1684 __ get_vm_result(x10, xthread);
1685
1686 // In between activations - previous activation type unknown yet
1687 // compute continuation point - the continuation point expects the
1688 // following registers set up:
1689 //
1690 // x10: exception
1691 // ra: return address/pc that threw exception
1692 // sp: expression stack of caller
1693 // fp: fp of caller
1694 // FIXME: There's no point saving ra here because VM calls don't trash it
1695 __ sub(sp, sp, 2 * wordSize);
1696 __ sd(x10, Address(sp, 0)); // save exception
1697 __ sd(ra, Address(sp, wordSize)); // save return address
1698 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1699 SharedRuntime::exception_handler_for_return_address),
1700 xthread, ra);
1701 __ mv(x11, x10); // save exception handler
1702 __ ld(x10, Address(sp, 0)); // restore exception
1703 __ ld(ra, Address(sp, wordSize)); // restore return address
1704 __ add(sp, sp, 2 * wordSize);
1705 // We might be returning to a deopt handler that expects x13 to
1706 // contain the exception pc
1707 __ mv(x13, ra);
1708 // Note that an "issuing PC" is actually the next PC after the call
1709 __ jr(x11); // jump to exception
1710 // handler of caller
1711 }
1712
1713 //
1714 // JVMTI ForceEarlyReturn support
1715 //
1716 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1717 address entry = __ pc();
1718
1719 __ restore_bcp();
1720 __ restore_locals();
1721 __ empty_expression_stack();
1722 __ load_earlyret_value(state);
1723
1724 __ ld(t0, Address(xthread, JavaThread::jvmti_thread_state_offset()));
1725 Address cond_addr(t0, JvmtiThreadState::earlyret_state_offset());
1726
1727 // Clear the earlyret state
1728 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1729 __ sd(zr, cond_addr);
1730
1731 __ remove_activation(state,
1732 false, /* throw_monitor_exception */
1733 false, /* install_monitor_exception */
1734 true); /* notify_jvmdi */
1735 __ ret();
1736
1737 return entry;
1738 }
1739 // end of ForceEarlyReturn support
1740
1741 //-----------------------------------------------------------------------------
1742 // Helper for vtos entry point generation
1743
1744 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1745 address& bep,
1746 address& cep,
1747 address& sep,
1748 address& aep,
1749 address& iep,
1750 address& lep,
1751 address& fep,
1752 address& dep,
1753 address& vep) {
1754 assert(t != nullptr && t->is_valid() && t->tos_in() == vtos, "illegal template");
1755 Label L;
1756 aep = __ pc(); __ push_ptr(); __ j(L);
1757 fep = __ pc(); __ push_f(); __ j(L);
1758 dep = __ pc(); __ push_d(); __ j(L);
1759 lep = __ pc(); __ push_l(); __ j(L);
1760 bep = cep = sep =
1761 iep = __ pc(); __ push_i();
1762 vep = __ pc();
1763 __ bind(L);
1764 generate_and_dispatch(t);
1765 }
1766
1767 //-----------------------------------------------------------------------------
1768
1769 // Non-product code
1770 #ifndef PRODUCT
1771 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1772 address entry = __ pc();
1773
1774 __ push_reg(ra);
1775 __ push(state);
1776 __ push_reg(RegSet::range(x10, x17) + RegSet::range(x5, x7) + RegSet::range(x28, x31), sp);
1777 __ mv(c_rarg2, x10); // Pass itos
1778 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), c_rarg1, c_rarg2, c_rarg3);
1779 __ pop_reg(RegSet::range(x10, x17) + RegSet::range(x5, x7) + RegSet::range(x28, x31), sp);
1780 __ pop(state);
1781 __ pop_reg(ra);
1782 __ ret(); // return from result handler
1783
1784 return entry;
1785 }
1786
1787 void TemplateInterpreterGenerator::count_bytecode() {
1788 __ mv(x7, (address) &BytecodeCounter::_counter_value);
1789 __ atomic_addw(noreg, 1, x7);
1790 }
1791
1792 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1793 __ mv(x7, (address) &BytecodeHistogram::_counters[t->bytecode()]);
1794 __ atomic_addw(noreg, 1, x7);
1795 }
1796
1797 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1798 // Calculate new index for counter:
1799 // _index = (_index >> log2_number_of_codes) |
1800 // (bytecode << log2_number_of_codes);
1801 Register index_addr = t1;
1802 Register index = t0;
1803 __ mv(index_addr, (address) &BytecodePairHistogram::_index);
1804 __ lw(index, index_addr);
1805 __ mv(x7, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1806 __ srli(index, index, BytecodePairHistogram::log2_number_of_codes);
1807 __ orrw(index, x7, index);
1808 __ sw(index, index_addr);
1809 // Bump bucket contents:
1810 // _counters[_index] ++;
1811 Register counter_addr = t1;
1812 __ mv(x7, (address) &BytecodePairHistogram::_counters);
1813 __ shadd(counter_addr, index, x7, counter_addr, LogBytesPerInt);
1814 __ atomic_addw(noreg, 1, counter_addr);
1815 }
1816
1817 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1818 // Call a little run-time stub to avoid blow-up for each bytecode.
1819 // The run-time runtime saves the right registers, depending on
1820 // the tosca in-state for the given template.
1821
1822 assert(Interpreter::trace_code(t->tos_in()) != nullptr, "entry must have been generated");
1823 __ jal(Interpreter::trace_code(t->tos_in()));
1824 __ reinit_heapbase();
1825 }
1826
1827 void TemplateInterpreterGenerator::stop_interpreter_at() {
1828 Label L;
1829 __ push_reg(t0);
1830 __ mv(t0, (address) &BytecodeCounter::_counter_value);
1831 __ ld(t0, Address(t0));
1832 __ mv(t1, StopInterpreterAt);
1833 __ bne(t0, t1, L);
1834 __ ebreak();
1835 __ bind(L);
1836 __ pop_reg(t0);
1837 }
1838
1839 #endif // !PRODUCT