1 /*
2 * Copyright (c) 2003, 2025, 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 "asm/macroAssembler.inline.hpp"
28 #include "classfile/javaClasses.hpp"
29 #include "compiler/disassembler.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 __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _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 __ subi(sp, c_rarg3, 18 * wordSize);
88 __ subi(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 switch (kind) {
169 case Interpreter::java_lang_math_abs:
170 entry_point = __ pc();
171 __ fld(f10, Address(esp));
172 __ fabs_d(f10, f10);
173 __ mv(sp, x19_sender_sp); // Restore caller's SP
174 break;
175 case Interpreter::java_lang_math_sqrt:
176 entry_point = __ pc();
177 __ fld(f10, Address(esp));
178 __ fsqrt_d(f10, f10);
179 __ mv(sp, x19_sender_sp);
180 break;
181 case Interpreter::java_lang_math_sin :
182 entry_point = __ pc();
183 __ fld(f10, Address(esp));
184 __ mv(sp, x19_sender_sp);
185 __ mv(x9, ra);
186 if (StubRoutines::dsin() == nullptr) {
187 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
188 } else {
189 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
190 }
191 __ rt_call(fn);
192 __ mv(ra, x9);
193 break;
194 case Interpreter::java_lang_math_cos :
195 entry_point = __ pc();
196 __ fld(f10, Address(esp));
197 __ mv(sp, x19_sender_sp);
198 __ mv(x9, ra);
199 if (StubRoutines::dcos() == nullptr) {
200 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
201 } else {
202 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
203 }
204 __ rt_call(fn);
205 __ mv(ra, x9);
206 break;
207 case Interpreter::java_lang_math_tan :
208 entry_point = __ pc();
209 __ fld(f10, Address(esp));
210 __ mv(sp, x19_sender_sp);
211 __ mv(x9, ra);
212 if (StubRoutines::dtan() == nullptr) {
213 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
214 } else {
215 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
216 }
217 __ rt_call(fn);
218 __ mv(ra, x9);
219 break;
220 case Interpreter::java_lang_math_log :
221 entry_point = __ pc();
222 __ fld(f10, Address(esp));
223 __ mv(sp, x19_sender_sp);
224 __ mv(x9, ra);
225 if (StubRoutines::dlog() == nullptr) {
226 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
227 } else {
228 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
229 }
230 __ rt_call(fn);
231 __ mv(ra, x9);
232 break;
233 case Interpreter::java_lang_math_log10 :
234 entry_point = __ pc();
235 __ fld(f10, Address(esp));
236 __ mv(sp, x19_sender_sp);
237 __ mv(x9, ra);
238 if (StubRoutines::dlog10() == nullptr) {
239 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
240 } else {
241 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
242 }
243 __ rt_call(fn);
244 __ mv(ra, x9);
245 break;
246 case Interpreter::java_lang_math_exp :
247 entry_point = __ pc();
248 __ fld(f10, Address(esp));
249 __ mv(sp, x19_sender_sp);
250 __ mv(x9, ra);
251 if (StubRoutines::dexp() == nullptr) {
252 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
253 } else {
254 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
255 }
256 __ rt_call(fn);
257 __ mv(ra, x9);
258 break;
259 case Interpreter::java_lang_math_pow :
260 entry_point = __ pc();
261 __ mv(x9, ra);
262 __ fld(f10, Address(esp, 2 * Interpreter::stackElementSize));
263 __ fld(f11, Address(esp));
264 __ mv(sp, x19_sender_sp);
265 if (StubRoutines::dpow() == nullptr) {
266 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
267 } else {
268 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
269 }
270 __ rt_call(fn);
271 __ mv(ra, x9);
272 break;
273 case Interpreter::java_lang_math_fmaD :
274 if (UseFMA) {
275 entry_point = __ pc();
276 __ fld(f10, Address(esp, 4 * Interpreter::stackElementSize));
277 __ fld(f11, Address(esp, 2 * Interpreter::stackElementSize));
278 __ fld(f12, Address(esp));
279 __ fmadd_d(f10, f10, f11, f12);
280 __ mv(sp, x19_sender_sp); // Restore caller's SP
281 }
282 break;
283 case Interpreter::java_lang_math_fmaF :
284 if (UseFMA) {
285 entry_point = __ pc();
286 __ flw(f10, Address(esp, 2 * Interpreter::stackElementSize));
287 __ flw(f11, Address(esp, Interpreter::stackElementSize));
288 __ flw(f12, Address(esp));
289 __ fmadd_s(f10, f10, f11, f12);
290 __ mv(sp, x19_sender_sp); // Restore caller's SP
291 }
292 break;
293 default:
294 ;
295 }
296 if (entry_point != nullptr) {
297 __ ret();
298 }
299
300 return entry_point;
301 }
302
303 // Abstract method entry
304 // Attempt to execute abstract method. Throw exception
305 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
306 // xmethod: Method*
307 // x19_sender_sp: sender SP
308
309 address entry_point = __ pc();
310
311 // abstract method entry
312
313 // pop return address, reset last_sp to null
314 __ empty_expression_stack();
315 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
316 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
317
318 // throw exception
319 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
320 InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
321 xmethod);
322 // the call_VM checks for exception, so we should never return here.
323 __ should_not_reach_here();
324
325 return entry_point;
326 }
327
328 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
329 address entry = __ pc();
330
331 #ifdef ASSERT
332 {
333 Label L;
334 __ ld(t0, Address(fp, frame::interpreter_frame_monitor_block_top_offset * wordSize));
335 __ shadd(t0, t0, fp, t0, LogBytesPerWord);
336 // maximal sp for current fp (stack grows negative)
337 // check if frame is complete
338 __ bge(t0, sp, L);
339 __ stop ("interpreter frame not set up");
340 __ bind(L);
341 }
342 #endif // ASSERT
343 // Restore bcp under the assumption that the current frame is still
344 // interpreted
345 __ restore_bcp();
346
347 // expression stack must be empty before entering the VM if an
348 // exception happened
349 __ empty_expression_stack();
350 // throw exception
351 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
352 return entry;
353 }
354
355 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
356 address entry = __ pc();
357 // expression stack must be empty before entering the VM if an
358 // exception happened
359 __ empty_expression_stack();
360 // setup parameters
361
362 // convention: expect aberrant index in register x11
363 __ zext(c_rarg2, x11, 32);
364 // convention: expect array in register x13
365 __ mv(c_rarg1, x13);
366 __ call_VM(noreg,
367 CAST_FROM_FN_PTR(address,
368 InterpreterRuntime::
369 throw_ArrayIndexOutOfBoundsException),
370 c_rarg1, c_rarg2);
371 return entry;
372 }
373
374 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
375 address entry = __ pc();
376
377 // object is at TOS
378 __ pop_reg(c_rarg1);
379
380 // expression stack must be empty before entering the VM if an
381 // exception happened
382 __ empty_expression_stack();
383
384 __ call_VM(noreg,
385 CAST_FROM_FN_PTR(address,
386 InterpreterRuntime::
387 throw_ClassCastException),
388 c_rarg1);
389 return entry;
390 }
391
392 address TemplateInterpreterGenerator::generate_exception_handler_common(
393 const char* name, const char* message, bool pass_oop) {
394 assert(!pass_oop || message == nullptr, "either oop or message but not both");
395 address entry = __ pc();
396 if (pass_oop) {
397 // object is at TOS
398 __ pop_reg(c_rarg2);
399 }
400 // expression stack must be empty before entering the VM if an
401 // exception happened
402 __ empty_expression_stack();
403 // setup parameters
404 __ la(c_rarg1, Address((address)name));
405 if (pass_oop) {
406 __ call_VM(x10, CAST_FROM_FN_PTR(address,
407 InterpreterRuntime::
408 create_klass_exception),
409 c_rarg1, c_rarg2);
410 } else {
411 // kind of lame ExternalAddress can't take null because
412 // external_word_Relocation will assert.
413 if (message != nullptr) {
414 __ la(c_rarg2, Address((address)message));
415 } else {
416 __ mv(c_rarg2, NULL_WORD);
417 }
418 __ call_VM(x10,
419 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
420 c_rarg1, c_rarg2);
421 }
422 // throw exception
423 __ j(RuntimeAddress(Interpreter::throw_exception_entry()));
424 return entry;
425 }
426
427 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
428 address entry = __ pc();
429
430 // Restore stack bottom in case i2c adjusted stack
431 __ ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
432 __ shadd(esp, t0, fp, t0, LogBytesPerWord);
433 // and null it as marker that esp is now tos until next java call
434 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
435 __ restore_bcp();
436 __ restore_locals();
437 __ restore_constant_pool_cache();
438 __ get_method(xmethod);
439
440 if (state == atos) {
441 Register obj = x10;
442 Register mdp = x11;
443 Register tmp = x12;
444 __ ld(mdp, Address(xmethod, Method::method_data_offset()));
445 __ profile_return_type(mdp, obj, tmp);
446 }
447
448 const Register cache = x11;
449 const Register index = x12;
450
451 if (index_size == sizeof(u4)) {
452 __ load_resolved_indy_entry(cache, index);
453 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
454 __ shadd(esp, cache, esp, t0, 3);
455 } else {
456 // Pop N words from the stack
457 assert(index_size == sizeof(u2), "Can only be u2");
458 __ load_method_entry(cache, index);
459 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
460
461 __ shadd(esp, cache, esp, t0, 3);
462 }
463
464 // Restore machine SP
465 __ restore_sp_after_call();
466
467 __ check_and_handle_popframe(xthread);
468 __ check_and_handle_earlyret(xthread);
469
470 __ get_dispatch();
471 __ dispatch_next(state, step);
472
473 return entry;
474 }
475
476 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
477 int step,
478 address continuation) {
479 address entry = __ pc();
480 __ restore_bcp();
481 __ restore_locals();
482 __ restore_constant_pool_cache();
483 __ get_method(xmethod);
484 __ get_dispatch();
485
486 __ restore_sp_after_call(); // Restore SP to extended SP
487
488 // Restore expression stack pointer
489 __ ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
490 __ shadd(esp, t0, fp, t0, LogBytesPerWord);
491 // null last_sp until next java call
492 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
493
494 // handle exceptions
495 {
496 Label L;
497 __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
498 __ beqz(t0, L);
499 __ call_VM(noreg,
500 CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
501 __ should_not_reach_here();
502 __ bind(L);
503 }
504
505 if (continuation == nullptr) {
506 __ dispatch_next(state, step);
507 } else {
508 __ jump_to_entry(continuation);
509 }
510 return entry;
511 }
512
513 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
514 address entry = __ pc();
515 if (type == T_OBJECT) {
516 // retrieve result from frame
517 __ ld(x10, Address(fp, frame::interpreter_frame_oop_temp_offset * wordSize));
518 // and verify it
519 __ verify_oop(x10);
520 } else {
521 __ cast_primitive_type(type, x10);
522 }
523
524 __ ret(); // return from result handler
525 return entry;
526 }
527
528 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state,
529 address runtime_entry) {
530 assert_cond(runtime_entry != nullptr);
531 address entry = __ pc();
532 __ push(state);
533 __ push_cont_fastpath(xthread);
534 __ call_VM(noreg, runtime_entry);
535 __ pop_cont_fastpath(xthread);
536 __ membar(MacroAssembler::AnyAny);
537 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
538 return entry;
539 }
540
541 address TemplateInterpreterGenerator::generate_cont_resume_interpreter_adapter() {
542 if (!Continuations::enabled()) return nullptr;
543 address start = __ pc();
544
545 __ restore_bcp();
546 __ restore_locals();
547
548 // Restore constant pool cache
549 __ ld(xcpool, Address(fp, frame::interpreter_frame_cache_offset * wordSize));
550
551 // Restore Java expression stack pointer
552 __ ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
553 __ shadd(esp, t0, fp, t0, Interpreter::logStackElementSize);
554 // and null it as marker that esp is now tos until next java call
555 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
556
557 // Restore machine SP
558 __ ld(t0, Address(fp, frame::interpreter_frame_extended_sp_offset * wordSize));
559 __ shadd(sp, t0, fp, t0, LogBytesPerWord);
560
561 // Restore method
562 __ ld(xmethod, Address(fp, frame::interpreter_frame_method_offset * wordSize));
563
564 // Restore dispatch
565 __ la(xdispatch, ExternalAddress((address)Interpreter::dispatch_table()));
566
567 __ ret();
568
569 return start;
570 }
571
572
573 // Helpers for commoning out cases in the various type of method entries.
574 //
575
576
577 // increment invocation count & check for overflow
578 //
579 // Note: checking for negative value instead of overflow
580 // so we have a 'sticky' overflow test
581 //
582 // xmethod: method
583 //
584 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
585 Label done;
586 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
587 int increment = InvocationCounter::count_increment;
588 Label no_mdo;
589 if (ProfileInterpreter) {
590 // Are we profiling?
591 __ ld(x10, Address(xmethod, Method::method_data_offset()));
592 __ beqz(x10, no_mdo);
593 // Increment counter in the MDO
594 const Address mdo_invocation_counter(x10, in_bytes(MethodData::invocation_counter_offset()) +
595 in_bytes(InvocationCounter::counter_offset()));
596 const Address mask(x10, in_bytes(MethodData::invoke_mask_offset()));
597 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, t0, t1, false, overflow);
598 __ j(done);
599 }
600 __ bind(no_mdo);
601 // Increment counter in MethodCounters
602 const Address invocation_counter(t1,
603 MethodCounters::invocation_counter_offset() +
604 InvocationCounter::counter_offset());
605 __ get_method_counters(xmethod, t1, done);
606 const Address mask(t1, in_bytes(MethodCounters::invoke_mask_offset()));
607 __ increment_mask_and_jump(invocation_counter, increment, mask, t0, x11, false, overflow);
608 __ bind(done);
609 }
610
611 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
612 __ mv(c_rarg1, zr);
613 __ call_VM(noreg,
614 CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), c_rarg1);
615 __ j(do_continue);
616 }
617
618 // See if we've got enough room on the stack for locals plus overhead
619 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
620 // without going through the signal handler, i.e., reserved and yellow zones
621 // will not be made usable. The shadow zone must suffice to handle the
622 // overflow.
623 // The expression stack grows down incrementally, so the normal guard
624 // page mechanism will work for that.
625 //
626 // NOTE: Since the additional locals are also always pushed (wasn't
627 // obvious in generate_method_entry) so the guard should work for them
628 // too.
629 //
630 // Args:
631 // x13: number of additional locals this frame needs (what we must check)
632 // xmethod: Method*
633 //
634 // Kills:
635 // x10
636 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
637
638 // monitor entry size: see picture of stack set
639 // (generate_method_entry) and frame_amd64.hpp
640 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
641
642 // total overhead size: entry_size + (saved fp through expr stack
643 // bottom). be sure to change this if you add/subtract anything
644 // to/from the overhead area
645 const int overhead_size =
646 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
647
648 const int page_size = (int)os::vm_page_size();
649
650 Label after_frame_check;
651
652 // see if the frame is greater than one page in size. If so,
653 // then we need to verify there is enough stack space remaining
654 // for the additional locals.
655 __ mv(t0, (page_size - overhead_size) / Interpreter::stackElementSize);
656 __ bleu(x13, t0, after_frame_check);
657
658 // compute sp as if this were going to be the last frame on
659 // the stack before the red zone
660
661 // locals + overhead, in bytes
662 __ mv(x10, overhead_size);
663 __ shadd(x10, x13, x10, t0, Interpreter::logStackElementSize); // 2 slots per parameter.
664
665 const Address stack_limit(xthread, JavaThread::stack_overflow_limit_offset());
666 __ ld(t0, stack_limit);
667
668 #ifdef ASSERT
669 Label limit_okay;
670 // Verify that thread stack limit is non-zero.
671 __ bnez(t0, limit_okay);
672 __ stop("stack overflow limit is zero");
673 __ bind(limit_okay);
674 #endif
675
676 // Add stack limit to locals.
677 __ add(x10, x10, t0);
678
679 // Check against the current stack bottom.
680 __ bgtu(sp, x10, after_frame_check);
681
682 // Remove the incoming args, peeling the machine SP back to where it
683 // was in the caller. This is not strictly necessary, but unless we
684 // do so the stack frame may have a garbage FP; this ensures a
685 // correct call stack that we can always unwind. The ANDI should be
686 // unnecessary because the sender SP in x19 is always aligned, but
687 // it doesn't hurt.
688 __ andi(sp, x19_sender_sp, -16);
689
690 // Note: the restored frame is not necessarily interpreted.
691 // Use the shared runtime version of the StackOverflowError.
692 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
693 __ far_jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry()));
694
695 // all done with frame size check
696 __ bind(after_frame_check);
697 }
698
699 // Allocate monitor and lock method (asm interpreter)
700 //
701 // Args:
702 // xmethod: Method*
703 // xlocals: locals
704 //
705 // Kills:
706 // x10
707 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
708 // t0, t1 (temporary regs)
709 void TemplateInterpreterGenerator::lock_method() {
710 // synchronize method
711 const Address access_flags(xmethod, Method::access_flags_offset());
712 const Address monitor_block_top(fp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
713 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
714
715 #ifdef ASSERT
716 __ load_unsigned_short(x10, access_flags);
717 __ verify_access_flags(x10, JVM_ACC_SYNCHRONIZED, "method doesn't need synchronization", false);
718 #endif // ASSERT
719
720 // get synchronization object
721 {
722 Label done;
723 __ load_unsigned_short(x10, access_flags);
724 __ andi(t0, x10, JVM_ACC_STATIC);
725 // get receiver (assume this is frequent case)
726 __ ld(x10, Address(xlocals, Interpreter::local_offset_in_bytes(0)));
727 __ beqz(t0, done);
728 __ load_mirror(x10, xmethod, x15, t1);
729
730 #ifdef ASSERT
731 {
732 Label L;
733 __ bnez(x10, L);
734 __ stop("synchronization object is null");
735 __ bind(L);
736 }
737 #endif // ASSERT
738
739 __ bind(done);
740 }
741
742 // add space for monitor & lock
743 __ check_extended_sp();
744 __ sub(sp, sp, entry_size); // add space for a monitor entry
745 __ sub(esp, esp, entry_size);
746 __ sub(t0, sp, fp);
747 __ srai(t0, t0, Interpreter::logStackElementSize);
748 __ sd(t0, Address(fp, frame::interpreter_frame_extended_sp_offset * wordSize));
749 __ sub(t0, esp, fp);
750 __ srai(t0, t0, Interpreter::logStackElementSize);
751 __ sd(t0, monitor_block_top); // set new monitor block top
752 // store object
753 __ sd(x10, Address(esp, BasicObjectLock::obj_offset()));
754 __ mv(c_rarg1, esp); // object address
755 __ lock_object(c_rarg1);
756 }
757
758 // Generate a fixed interpreter frame. This is identical setup for
759 // interpreted methods and for native methods hence the shared code.
760 //
761 // Args:
762 // ra: return address
763 // xmethod: Method*
764 // xlocals: pointer to locals
765 // xcpool: cp cache
766 // stack_pointer: previous sp
767 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
768 // initialize fixed part of activation frame
769 if (native_call) {
770 __ subi(esp, sp, 14 * wordSize);
771 __ mv(xbcp, zr);
772 __ subi(sp, sp, 14 * wordSize);
773 // add 2 zero-initialized slots for native calls
774 __ sd(zr, Address(sp, 13 * wordSize));
775 __ sd(zr, Address(sp, 12 * wordSize));
776 } else {
777 __ subi(esp, sp, 12 * wordSize);
778 __ ld(t0, Address(xmethod, Method::const_offset())); // get ConstMethod
779 __ add(xbcp, t0, in_bytes(ConstMethod::codes_offset())); // get codebase
780 __ subi(sp, sp, 12 * wordSize);
781 }
782 __ sd(xbcp, Address(sp, wordSize));
783 __ mv(t0, frame::interpreter_frame_initial_sp_offset);
784 __ sd(t0, Address(sp, 0));
785
786 if (ProfileInterpreter) {
787 Label method_data_continue;
788 __ ld(t0, Address(xmethod, Method::method_data_offset()));
789 __ beqz(t0, method_data_continue);
790 __ la(t0, Address(t0, in_bytes(MethodData::data_offset())));
791 __ bind(method_data_continue);
792 }
793
794 __ sd(xmethod, Address(sp, 7 * wordSize));
795 __ sd(ProfileInterpreter ? t0 : zr, Address(sp, 6 * wordSize));
796
797 __ sd(ra, Address(sp, 11 * wordSize));
798 __ sd(fp, Address(sp, 10 * wordSize));
799 __ la(fp, Address(sp, 12 * wordSize)); // include ra & fp
800
801 __ ld(xcpool, Address(xmethod, Method::const_offset()));
802 __ ld(xcpool, Address(xcpool, ConstMethod::constants_offset()));
803 __ ld(xcpool, Address(xcpool, ConstantPool::cache_offset()));
804 __ sd(xcpool, Address(sp, 3 * wordSize));
805 __ sub(t0, xlocals, fp);
806 __ srai(t0, t0, Interpreter::logStackElementSize); // t0 = xlocals - fp();
807 // Store relativized xlocals, see frame::interpreter_frame_locals().
808 __ sd(t0, Address(sp, 2 * wordSize));
809
810 // set sender sp
811 // leave last_sp as null
812 __ sd(x19_sender_sp, Address(sp, 9 * wordSize));
813 __ sd(zr, Address(sp, 8 * wordSize));
814
815 // Get mirror and store it in the frame as GC root for this Method*
816 __ load_mirror(t2, xmethod, x15, t1);
817 __ sd(t2, Address(sp, 4 * wordSize));
818
819 if (!native_call) {
820 __ ld(t0, Address(xmethod, Method::const_offset()));
821 __ lhu(t0, Address(t0, ConstMethod::max_stack_offset()));
822 __ add(t0, t0, MAX2(3, Method::extra_stack_entries()));
823 __ slli(t0, t0, 3);
824 __ sub(t0, sp, t0);
825 __ andi(t0, t0, -16);
826 __ sub(t1, t0, fp);
827 __ srai(t1, t1, Interpreter::logStackElementSize);
828 // Store extended SP
829 __ sd(t1, Address(sp, 5 * wordSize));
830 // Move SP out of the way
831 __ mv(sp, t0);
832 } else {
833 // Make sure there is room for the exception oop pushed in case method throws
834 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
835 __ subi(t0, sp, 2 * wordSize);
836 __ sub(t1, t0, fp);
837 __ srai(t1, t1, Interpreter::logStackElementSize);
838 __ sd(t1, Address(sp, 5 * wordSize));
839 __ mv(sp, t0);
840 }
841 }
842
843 // End of helpers
844
845 // Various method entries
846 //------------------------------------------------------------------------------------------------------------------------
847 //
848 //
849
850 // Method entry for java.lang.ref.Reference.get.
851 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
852 // Code: _aload_0, _getfield, _areturn
853 // parameter size = 1
854 //
855 // The code that gets generated by this routine is split into 2 parts:
856 // 1. The "intrinsified" code for G1 (or any SATB based GC),
857 // 2. The slow path - which is an expansion of the regular method entry.
858 //
859 // Notes:-
860 // * In the G1 code we do not check whether we need to block for
861 // a safepoint. If G1 is enabled then we must execute the specialized
862 // code for Reference.get (except when the Reference object is null)
863 // so that we can log the value in the referent field with an SATB
864 // update buffer.
865 // If the code for the getfield template is modified so that the
866 // G1 pre-barrier code is executed when the current method is
867 // Reference.get() then going through the normal method entry
868 // will be fine.
869 // * The G1 code can, however, check the receiver object (the instance
870 // of java.lang.Reference) and jump to the slow path if null. If the
871 // Reference object is null then we obviously cannot fetch the referent
872 // and so we don't need to call the G1 pre-barrier. Thus we can use the
873 // regular method entry code to generate the NPE.
874 //
875 // This code is based on generate_accessor_entry.
876 //
877 // xmethod: Method*
878 // x19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
879
880 // ra is live. It must be saved around calls.
881
882 address entry = __ pc();
883
884 const int referent_offset = java_lang_ref_Reference::referent_offset();
885 guarantee(referent_offset > 0, "referent offset not initialized");
886
887 Label slow_path;
888 const Register local_0 = c_rarg0;
889 // Check if local 0 isn't null
890 // If the receiver is null then it is OK to jump to the slow path.
891 __ ld(local_0, Address(esp, 0));
892 __ beqz(local_0, slow_path);
893
894 // Load the value of the referent field.
895 const Address field_address(local_0, referent_offset);
896 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
897 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ t0, /*tmp2*/ t1);
898
899 // areturn
900 __ andi(sp, x19_sender_sp, -16); // done with stack
901 __ ret();
902
903 // generate a vanilla interpreter entry as the slow path
904 __ bind(slow_path);
905 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
906 return entry;
907 }
908
909 /**
910 * Method entry for static native methods:
911 * int java.util.zip.CRC32.update(int crc, int b)
912 */
913 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
914 // TODO: Unimplemented generate_CRC32_update_entry
915 return nullptr;
916 }
917
918 /**
919 * Method entry for static native methods:
920 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
921 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
922 */
923 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
924 // TODO: Unimplemented generate_CRC32_updateBytes_entry
925 return nullptr;
926 }
927
928 /**
929 * Method entry for intrinsic-candidate (non-native) methods:
930 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
931 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
932 * Unlike CRC32, CRC32C does not have any methods marked as native
933 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
934 */
935 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
936 // TODO: Unimplemented generate_CRC32C_updateBytes_entry
937 return nullptr;
938 }
939
940 // Not supported
941 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
942 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
943 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
944 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
945 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() { return nullptr; }
946 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() { return nullptr; }
947
948 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
949 // See more discussion in stackOverflow.hpp.
950
951 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
952 const int page_size = (int)os::vm_page_size();
953 const int n_shadow_pages = shadow_zone_size / page_size;
954
955 #ifdef ASSERT
956 Label L_good_limit;
957 __ ld(t0, Address(xthread, JavaThread::shadow_zone_safe_limit()));
958 __ bnez(t0, L_good_limit);
959 __ stop("shadow zone safe limit is not initialized");
960 __ bind(L_good_limit);
961
962 Label L_good_watermark;
963 __ ld(t0, Address(xthread, JavaThread::shadow_zone_growth_watermark()));
964 __ bnez(t0, L_good_watermark);
965 __ stop("shadow zone growth watermark is not initialized");
966 __ bind(L_good_watermark);
967 #endif
968
969 Label L_done;
970
971 __ ld(t0, Address(xthread, JavaThread::shadow_zone_growth_watermark()));
972 __ bgtu(sp, t0, L_done);
973
974 for (int p = 1; p <= n_shadow_pages; p++) {
975 __ bang_stack_with_offset(p * page_size);
976 }
977
978 // Record the new watermark, but only if the update is above the safe limit.
979 // Otherwise, the next time around the check above would pass the safe limit.
980 __ ld(t0, Address(xthread, JavaThread::shadow_zone_safe_limit()));
981 __ bleu(sp, t0, L_done);
982 __ sd(sp, Address(xthread, JavaThread::shadow_zone_growth_watermark()));
983
984 __ bind(L_done);
985 }
986
987 // Interpreter stub for calling a native method. (asm interpreter)
988 // This sets up a somewhat different looking stack for calling the
989 // native method than the typical interpreter frame setup.
990 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized, bool runtime_upcalls) {
991 // determine code generation flags
992 bool inc_counter = (UseCompiler || CountCompiledCalls) && !PreloadOnly;
993
994 // x11: Method*
995 // x30: sender sp
996
997 address entry_point = __ pc();
998
999 const Address constMethod (xmethod, Method::const_offset());
1000 const Address access_flags (xmethod, Method::access_flags_offset());
1001 const Address size_of_parameters(x12, ConstMethod::
1002 size_of_parameters_offset());
1003
1004 // get parameter size (always needed)
1005 __ ld(x12, constMethod);
1006 __ load_unsigned_short(x12, size_of_parameters);
1007
1008 // Native calls don't need the stack size check since they have no
1009 // expression stack and the arguments are already on the stack and
1010 // we only add a handful of words to the stack.
1011
1012 // xmethod: Method*
1013 // x12: size of parameters
1014 // x30: sender sp
1015
1016 // for natives the size of locals is zero
1017
1018 // compute beginning of parameters (xlocals)
1019 __ shadd(xlocals, x12, esp, xlocals, 3);
1020 __ subi(xlocals, xlocals, wordSize);
1021
1022 // Pull SP back to minimum size: this avoids holes in the stack
1023 __ andi(sp, esp, -16);
1024
1025 // initialize fixed part of activation frame
1026 generate_fixed_frame(true);
1027
1028 // make sure method is native & not abstract
1029 #ifdef ASSERT
1030 __ load_unsigned_short(x10, access_flags);
1031 __ verify_access_flags(x10, JVM_ACC_NATIVE, "tried to execute non-native method as native", false);
1032 __ verify_access_flags(x10, JVM_ACC_ABSTRACT, "tried to execute abstract method in interpreter");
1033 #endif
1034
1035 // Since at this point in the method invocation the exception
1036 // handler would try to exit the monitor of synchronized methods
1037 // which hasn't been entered yet, we set the thread local variable
1038 // _do_not_unlock_if_synchronized to true. The remove_activation
1039 // will check this flag.
1040
1041 const Address do_not_unlock_if_synchronized(xthread,
1042 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1043 __ mv(t1, true);
1044 __ sb(t1, do_not_unlock_if_synchronized);
1045
1046 // increment invocation count & check for overflow
1047 Label invocation_counter_overflow;
1048 if (inc_counter) {
1049 generate_counter_incr(&invocation_counter_overflow);
1050 }
1051
1052 Label continue_after_compile;
1053 __ bind(continue_after_compile);
1054
1055 bang_stack_shadow_pages(true);
1056
1057 // reset the _do_not_unlock_if_synchronized flag
1058 __ sb(zr, do_not_unlock_if_synchronized);
1059
1060 // check for synchronized methods
1061 // Must happen AFTER invocation_counter check and stack overflow check,
1062 // so method is not locked if overflows.
1063 if (synchronized) {
1064 lock_method();
1065 } else {
1066 // no synchronization necessary
1067 #ifdef ASSERT
1068 __ load_unsigned_short(x10, access_flags);
1069 __ verify_access_flags(x10, JVM_ACC_SYNCHRONIZED, "method needs synchronization");
1070 #endif
1071 }
1072
1073 // start execution
1074 #ifdef ASSERT
1075 __ verify_frame_setup();
1076 #endif
1077
1078 // jvmti support
1079 __ notify_method_entry();
1080
1081 // work registers
1082 const Register t = x18;
1083 const Register result_handler = x19;
1084
1085 // allocate space for parameters
1086 __ ld(t, Address(xmethod, Method::const_offset()));
1087 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1088
1089 __ slli(t, t, Interpreter::logStackElementSize);
1090 __ sub(x30, esp, t);
1091 __ andi(sp, x30, -16);
1092 __ mv(esp, x30);
1093
1094 // get signature handler
1095 {
1096 Label L;
1097 __ ld(t, Address(xmethod, Method::signature_handler_offset()));
1098 __ bnez(t, L);
1099 __ call_VM(noreg,
1100 CAST_FROM_FN_PTR(address,
1101 InterpreterRuntime::prepare_native_call),
1102 xmethod);
1103 __ ld(t, Address(xmethod, Method::signature_handler_offset()));
1104 __ bind(L);
1105 }
1106
1107 // call signature handler
1108 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == xlocals,
1109 "adjust this code");
1110 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1111 "adjust this code");
1112 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t0,
1113 "adjust this code");
1114
1115 // The generated handlers do not touch xmethod (the method).
1116 // However, large signatures cannot be cached and are generated
1117 // each time here. The slow-path generator can do a GC on return,
1118 // so we must reload it after the call.
1119 __ jalr(t);
1120 __ get_method(xmethod); // slow path can do a GC, reload xmethod
1121
1122
1123 // result handler is in x10
1124 // set result handler
1125 __ mv(result_handler, x10);
1126 // Save it in the frame in case of preemption; we cannot rely on callee saved registers.
1127 __ sd(x10, Address(fp, frame::interpreter_frame_result_handler_offset * wordSize));
1128
1129 // pass mirror handle if static call
1130 {
1131 Label L;
1132 __ load_unsigned_short(t, Address(xmethod, Method::access_flags_offset()));
1133 __ test_bit(t0, t, exact_log2(JVM_ACC_STATIC));
1134 __ beqz(t0, L);
1135 // get mirror
1136 __ load_mirror(t, xmethod, x28, t1);
1137 // copy mirror into activation frame
1138 __ sd(t, Address(fp, frame::interpreter_frame_oop_temp_offset * wordSize));
1139 // pass handle to mirror
1140 __ addi(c_rarg1, fp, frame::interpreter_frame_oop_temp_offset * wordSize);
1141 __ bind(L);
1142 }
1143
1144 // get native function entry point in x28
1145 {
1146 Label L;
1147 __ ld(x28, Address(xmethod, Method::native_function_offset()));
1148 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1149 __ la(t, unsatisfied);
1150 __ load_long_misaligned(t1, Address(t, 0), t0, 2); // 2 bytes aligned, but not 4 or 8
1151
1152 __ bne(x28, t1, L);
1153 __ call_VM(noreg,
1154 CAST_FROM_FN_PTR(address,
1155 InterpreterRuntime::prepare_native_call),
1156 xmethod);
1157 __ get_method(xmethod);
1158 __ ld(x28, Address(xmethod, Method::native_function_offset()));
1159 __ bind(L);
1160 }
1161
1162 // pass JNIEnv
1163 __ add(c_rarg0, xthread, in_bytes(JavaThread::jni_environment_offset()));
1164
1165 // It is enough that the pc() points into the right code
1166 // segment. It does not have to be the correct return pc.
1167 // For convenience we use the pc we want to resume to in
1168 // case of preemption on Object.wait.
1169 Label native_return;
1170 __ set_last_Java_frame(esp, fp, native_return, x30);
1171
1172 // change thread state
1173 #ifdef ASSERT
1174 {
1175 Label L;
1176 __ lwu(t, Address(xthread, JavaThread::thread_state_offset()));
1177 __ mv(t0, (u1)_thread_in_Java);
1178 __ beq(t, t0, L);
1179 __ stop("Wrong thread state in native stub");
1180 __ bind(L);
1181 }
1182 #endif
1183
1184 // Change state to native
1185 __ la(t1, Address(xthread, JavaThread::thread_state_offset()));
1186 __ mv(t0, _thread_in_native);
1187 __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1188 __ sw(t0, Address(t1));
1189
1190 __ push_cont_fastpath();
1191
1192 // Call the native method.
1193 __ jalr(x28);
1194
1195 __ pop_cont_fastpath();
1196
1197 __ get_method(xmethod);
1198 // result potentially in x10 or f10
1199
1200 // Restore cpu control state after JNI call
1201 __ restore_cpu_control_state_after_jni(t0);
1202
1203 // make room for the pushes we're about to do
1204 __ subi(t0, esp, 4 * wordSize);
1205 __ andi(sp, t0, -16);
1206
1207 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1208 // in order to extract the result of a method call. If the order of these
1209 // pushes change or anything else is added to the stack then the code in
1210 // interpreter_frame_result must also change.
1211 __ push(dtos);
1212 __ push(ltos);
1213
1214 // change thread state
1215 // Force all preceding writes to be observed prior to thread state change
1216 __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1217
1218 __ mv(t0, _thread_in_native_trans);
1219 __ sw(t0, Address(xthread, JavaThread::thread_state_offset()));
1220
1221 // Force this write out before the read below
1222 if (!UseSystemMemoryBarrier) {
1223 __ membar(MacroAssembler::AnyAny);
1224 }
1225
1226 // check for safepoint operation in progress and/or pending suspend requests
1227 {
1228 Label L, Continue;
1229
1230 // We need an acquire here to ensure that any subsequent load of the
1231 // global SafepointSynchronize::_state flag is ordered after this load
1232 // of the thread-local polling word. We don't want this poll to
1233 // return false (i.e. not safepointing) and a later poll of the global
1234 // SafepointSynchronize::_state spuriously to return true.
1235 //
1236 // This is to avoid a race when we're in a native->Java transition
1237 // racing the code which wakes up from a safepoint.
1238 __ safepoint_poll(L, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1239 __ lwu(t1, Address(xthread, JavaThread::suspend_flags_offset()));
1240 __ beqz(t1, Continue);
1241 __ bind(L);
1242
1243 // Don't use call_VM as it will see a possible pending exception
1244 // and forward it and never return here preventing us from
1245 // clearing _last_native_pc down below. So we do a runtime call by
1246 // hand.
1247 //
1248 __ mv(c_rarg0, xthread);
1249 __ rt_call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1250 __ get_method(xmethod);
1251 __ reinit_heapbase();
1252 __ bind(Continue);
1253 }
1254
1255 // change thread state
1256 // Force all preceding writes to be observed prior to thread state change
1257 __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1258
1259 __ mv(t0, _thread_in_Java);
1260 __ sw(t0, Address(xthread, JavaThread::thread_state_offset()));
1261
1262 if (LockingMode != LM_LEGACY) {
1263 // Check preemption for Object.wait()
1264 Label not_preempted;
1265 __ ld(t1, Address(xthread, JavaThread::preempt_alternate_return_offset()));
1266 __ beqz(t1, not_preempted);
1267 __ sd(zr, Address(xthread, JavaThread::preempt_alternate_return_offset()));
1268 __ jr(t1);
1269 __ bind(native_return);
1270 __ restore_after_resume(true /* is_native */);
1271 // reload result_handler
1272 __ ld(result_handler, Address(fp, frame::interpreter_frame_result_handler_offset * wordSize));
1273 __ bind(not_preempted);
1274 } else {
1275 // any pc will do so just use this one for LM_LEGACY to keep code together.
1276 __ bind(native_return);
1277 }
1278
1279 // reset_last_Java_frame
1280 __ reset_last_Java_frame(true);
1281
1282 if (CheckJNICalls) {
1283 // clear_pending_jni_exception_check
1284 __ sd(zr, Address(xthread, JavaThread::pending_jni_exception_check_fn_offset()));
1285 }
1286
1287 // reset handle block
1288 __ ld(t, Address(xthread, JavaThread::active_handles_offset()));
1289 __ sd(zr, Address(t, JNIHandleBlock::top_offset()));
1290
1291 // If result is an oop unbox and store it in frame where gc will see it
1292 // and result handler will pick it up
1293
1294 {
1295 Label no_oop;
1296 __ la(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1297 __ bne(t, result_handler, no_oop);
1298 // Unbox oop result, e.g. JNIHandles::resolve result.
1299 __ pop(ltos);
1300 __ resolve_jobject(x10, t, t1);
1301 __ sd(x10, Address(fp, frame::interpreter_frame_oop_temp_offset * wordSize));
1302 // keep stack depth as expected by pushing oop which will eventually be discarded
1303 __ push(ltos);
1304 __ bind(no_oop);
1305 }
1306
1307 {
1308 Label no_reguard;
1309 __ lwu(t0, Address(xthread, in_bytes(JavaThread::stack_guard_state_offset())));
1310 __ mv(t1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1311 __ bne(t0, t1, no_reguard);
1312
1313 __ push_call_clobbered_registers();
1314 __ mv(c_rarg0, xthread);
1315 __ rt_call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1316 __ pop_call_clobbered_registers();
1317 __ bind(no_reguard);
1318 }
1319
1320 // The method register is junk from after the thread_in_native transition
1321 // until here. Also can't call_VM until the bcp has been
1322 // restored. Need bcp for throwing exception below so get it now.
1323 __ get_method(xmethod);
1324
1325 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1326 // xbcp == code_base()
1327 __ ld(xbcp, Address(xmethod, Method::const_offset())); // get ConstMethod*
1328 __ add(xbcp, xbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1329 // handle exceptions (exception handling will handle unlocking!)
1330 {
1331 Label L;
1332 __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
1333 __ beqz(t0, L);
1334 // Note: At some point we may want to unify this with the code
1335 // used in call_VM_base(); i.e., we should use the
1336 // StubRoutines::forward_exception code. For now this doesn't work
1337 // here because the sp is not correctly set at this point.
1338 __ MacroAssembler::call_VM(noreg,
1339 CAST_FROM_FN_PTR(address,
1340 InterpreterRuntime::throw_pending_exception));
1341 __ should_not_reach_here();
1342 __ bind(L);
1343 }
1344
1345 // do unlocking if necessary
1346 {
1347 Label L;
1348 __ load_unsigned_short(t, Address(xmethod, Method::access_flags_offset()));
1349 __ test_bit(t0, t, exact_log2(JVM_ACC_SYNCHRONIZED));
1350 __ beqz(t0, L);
1351 // the code below should be shared with interpreter macro
1352 // assembler implementation
1353 {
1354 Label unlock;
1355 // BasicObjectLock will be first in list, since this is a
1356 // synchronized method. However, need to check that the object
1357 // has not been unlocked by an explicit monitorexit bytecode.
1358
1359 // monitor expect in c_rarg1 for slow unlock path
1360 __ la(c_rarg1, Address(fp, // address of first monitor
1361 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1362 wordSize - sizeof(BasicObjectLock))));
1363
1364 __ ld(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1365 __ bnez(t, unlock);
1366
1367 // Entry already unlocked, need to throw exception
1368 __ MacroAssembler::call_VM(noreg,
1369 CAST_FROM_FN_PTR(address,
1370 InterpreterRuntime::throw_illegal_monitor_state_exception));
1371 __ should_not_reach_here();
1372
1373 __ bind(unlock);
1374 __ unlock_object(c_rarg1);
1375 }
1376 __ bind(L);
1377 }
1378
1379 // jvmti support
1380 // Note: This must happen _after_ handling/throwing any exceptions since
1381 // the exception handler code notifies the runtime of method exits
1382 // too. If this happens before, method entry/exit notifications are
1383 // not properly paired (was bug - gri 11/22/99).
1384 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1385
1386 __ pop(ltos);
1387 __ pop(dtos);
1388
1389 __ jalr(result_handler);
1390
1391 // remove activation
1392 __ ld(esp, Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1393 // remove frame anchor
1394 __ leave();
1395
1396 // restore sender sp
1397 __ mv(sp, esp);
1398
1399 __ ret();
1400
1401 if (inc_counter) {
1402 // Handle overflow of counter and compile method
1403 __ bind(invocation_counter_overflow);
1404 generate_counter_overflow(continue_after_compile);
1405 }
1406
1407 return entry_point;
1408 }
1409
1410 //
1411 // Generic interpreted method entry to (asm) interpreter
1412 //
1413 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized, bool runtime_upcalls) {
1414
1415 // determine code generation flags
1416 const bool inc_counter = (UseCompiler || CountCompiledCalls) && !PreloadOnly;
1417
1418 // t0: sender sp
1419 address entry_point = __ pc();
1420
1421 const Address constMethod(xmethod, Method::const_offset());
1422 const Address access_flags(xmethod, Method::access_flags_offset());
1423 const Address size_of_parameters(x13,
1424 ConstMethod::size_of_parameters_offset());
1425 const Address size_of_locals(x13, ConstMethod::size_of_locals_offset());
1426
1427 // get parameter size (always needed)
1428 // need to load the const method first
1429 __ ld(x13, constMethod);
1430 __ load_unsigned_short(x12, size_of_parameters);
1431
1432 // x12: size of parameters
1433
1434 __ load_unsigned_short(x13, size_of_locals); // get size of locals in words
1435 __ sub(x13, x13, x12); // x13 = no. of additional locals
1436
1437 // see if we've got enough room on the stack for locals plus overhead.
1438 generate_stack_overflow_check();
1439
1440 // compute beginning of parameters (xlocals)
1441 __ shadd(xlocals, x12, esp, t1, 3);
1442 __ subi(xlocals, xlocals, wordSize);
1443
1444 // Make room for additional locals
1445 __ slli(t1, x13, 3);
1446 __ sub(t0, esp, t1);
1447
1448 // Padding between locals and fixed part of activation frame to ensure
1449 // SP is always 16-byte aligned.
1450 __ andi(sp, t0, -16);
1451
1452 // x13 - # of additional locals
1453 // allocate space for locals
1454 // explicitly initialize locals
1455 {
1456 Label exit, loop;
1457 __ blez(x13, exit); // do nothing if x13 <= 0
1458 __ bind(loop);
1459 __ sd(zr, Address(t0));
1460 __ addi(t0, t0, wordSize);
1461 __ subi(x13, x13, 1); // until everything initialized
1462 __ bnez(x13, loop);
1463 __ bind(exit);
1464 }
1465
1466 // And the base dispatch table
1467 __ get_dispatch();
1468
1469 // initialize fixed part of activation frame
1470 generate_fixed_frame(false);
1471
1472 // make sure method is not native & not abstract
1473 #ifdef ASSERT
1474 __ load_unsigned_short(x10, access_flags);
1475 __ verify_access_flags(x10, JVM_ACC_NATIVE, "tried to execute native method as non-native");
1476 __ verify_access_flags(x10, JVM_ACC_ABSTRACT, "tried to execute abstract method in interpreter");
1477 #endif
1478
1479 // Since at this point in the method invocation the exception
1480 // handler would try to exit the monitor of synchronized methods
1481 // which hasn't been entered yet, we set the thread local variable
1482 // _do_not_unlock_if_synchronized to true. The remove_activation
1483 // will check this flag.
1484
1485 const Address do_not_unlock_if_synchronized(xthread,
1486 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1487 __ mv(t1, true);
1488 __ sb(t1, do_not_unlock_if_synchronized);
1489
1490 Label no_mdp;
1491 const Register mdp = x13;
1492 __ ld(mdp, Address(xmethod, Method::method_data_offset()));
1493 __ beqz(mdp, no_mdp);
1494 __ add(mdp, mdp, in_bytes(MethodData::data_offset()));
1495 __ profile_parameters_type(mdp, x11, x12, x14); // use x11, x12, x14 as tmp registers
1496 __ bind(no_mdp);
1497
1498 // increment invocation count & check for overflow
1499 Label invocation_counter_overflow;
1500 if (inc_counter) {
1501 generate_counter_incr(&invocation_counter_overflow);
1502 }
1503
1504 Label continue_after_compile;
1505 __ bind(continue_after_compile);
1506
1507 bang_stack_shadow_pages(false);
1508
1509 // reset the _do_not_unlock_if_synchronized flag
1510 __ sb(zr, do_not_unlock_if_synchronized);
1511
1512 // check for synchronized methods
1513 // Must happen AFTER invocation_counter check and stack overflow check,
1514 // so method is not locked if overflows.
1515 if (synchronized) {
1516 // Allocate monitor and lock method
1517 lock_method();
1518 } else {
1519 // no synchronization necessary
1520 #ifdef ASSERT
1521 __ load_unsigned_short(x10, access_flags);
1522 __ verify_access_flags(x10, JVM_ACC_SYNCHRONIZED, "method needs synchronization");
1523 #endif
1524 }
1525
1526 // start execution
1527 #ifdef ASSERT
1528 __ verify_frame_setup();
1529 #endif
1530
1531 // jvmti support
1532 __ notify_method_entry();
1533
1534 __ dispatch_next(vtos);
1535
1536 // invocation counter overflow
1537 if (inc_counter) {
1538 // Handle overflow of counter and compile method
1539 __ bind(invocation_counter_overflow);
1540 generate_counter_overflow(continue_after_compile);
1541 }
1542
1543 return entry_point;
1544 }
1545
1546 // Method entry for java.lang.Thread.currentThread
1547 address TemplateInterpreterGenerator::generate_currentThread() {
1548 address entry_point = __ pc();
1549
1550 __ ld(x10, Address(xthread, JavaThread::vthread_offset()));
1551 __ resolve_oop_handle(x10, t0, t1);
1552 __ ret();
1553
1554 return entry_point;
1555 }
1556
1557 //-----------------------------------------------------------------------------
1558 // Exceptions
1559
1560 void TemplateInterpreterGenerator::generate_throw_exception() {
1561 // Entry point in previous activation (i.e., if the caller was
1562 // interpreted)
1563 Interpreter::_rethrow_exception_entry = __ pc();
1564 // Restore sp to interpreter_frame_last_sp even though we are going
1565 // to empty the expression stack for the exception processing.
1566 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1567 // x10: exception
1568 // x13: return address/pc that threw exception
1569 __ restore_bcp(); // xbcp points to call/send
1570 __ restore_locals();
1571 __ restore_constant_pool_cache();
1572 __ reinit_heapbase(); // restore xheapbase as heapbase.
1573 __ get_dispatch();
1574
1575 // Entry point for exceptions thrown within interpreter code
1576 Interpreter::_throw_exception_entry = __ pc();
1577 // If we came here via a NullPointerException on the receiver of a
1578 // method, xthread may be corrupt.
1579 __ get_method(xmethod);
1580 // expression stack is undefined here
1581 // x10: exception
1582 // xbcp: exception bcp
1583 __ verify_oop(x10);
1584 __ mv(c_rarg1, x10);
1585
1586 // expression stack must be empty before entering the VM in case of
1587 // an exception
1588 __ empty_expression_stack();
1589 // find exception handler address and preserve exception oop
1590 __ call_VM(x13,
1591 CAST_FROM_FN_PTR(address,
1592 InterpreterRuntime::exception_handler_for_exception),
1593 c_rarg1);
1594
1595 // Restore machine SP
1596 __ restore_sp_after_call();
1597
1598 // x10: exception handler entry point
1599 // x13: preserved exception oop
1600 // xbcp: bcp for exception handler
1601 __ push_ptr(x13); // push exception which is now the only value on the stack
1602 __ jr(x10); // jump to exception handler (may be _remove_activation_entry!)
1603
1604 // If the exception is not handled in the current frame the frame is
1605 // removed and the exception is rethrown (i.e. exception
1606 // continuation is _rethrow_exception).
1607 //
1608 // Note: At this point the bci is still the bxi for the instruction
1609 // which caused the exception and the expression stack is
1610 // empty. Thus, for any VM calls at this point, GC will find a legal
1611 // oop map (with empty expression stack).
1612
1613 //
1614 // JVMTI PopFrame support
1615 //
1616
1617 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1618 __ empty_expression_stack();
1619 // Set the popframe_processing bit in pending_popframe_condition
1620 // indicating that we are currently handling popframe, so that
1621 // call_VMs that may happen later do not trigger new popframe
1622 // handling cycles.
1623 __ lwu(x13, Address(xthread, JavaThread::popframe_condition_offset()));
1624 __ ori(x13, x13, JavaThread::popframe_processing_bit);
1625 __ sw(x13, Address(xthread, JavaThread::popframe_condition_offset()));
1626
1627 {
1628 // Check to see whether we are returning to a deoptimized frame.
1629 // (The PopFrame call ensures that the caller of the popped frame is
1630 // either interpreted or compiled and deoptimizes it if compiled.)
1631 // In this case, we can't call dispatch_next() after the frame is
1632 // popped, but instead must save the incoming arguments and restore
1633 // them after deoptimization has occurred.
1634 //
1635 // Note that we don't compare the return PC against the
1636 // deoptimization blob's unpack entry because of the presence of
1637 // adapter frames in C2.
1638 Label caller_not_deoptimized;
1639 __ ld(c_rarg1, Address(fp, frame::return_addr_offset * wordSize));
1640 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), c_rarg1);
1641 __ bnez(x10, caller_not_deoptimized);
1642
1643 // Compute size of arguments for saving when returning to
1644 // deoptimized caller
1645 __ get_method(x10);
1646 __ ld(x10, Address(x10, Method::const_offset()));
1647 __ load_unsigned_short(x10, Address(x10, in_bytes(ConstMethod::
1648 size_of_parameters_offset())));
1649 __ slli(x10, x10, Interpreter::logStackElementSize);
1650 __ restore_locals();
1651 __ sub(xlocals, xlocals, x10);
1652 __ addi(xlocals, xlocals, wordSize);
1653 // Save these arguments
1654 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1655 Deoptimization::
1656 popframe_preserve_args),
1657 xthread, x10, xlocals);
1658
1659 __ remove_activation(vtos,
1660 /* throw_monitor_exception */ false,
1661 /* install_monitor_exception */ false,
1662 /* notify_jvmdi */ false);
1663
1664 // Inform deoptimization that it is responsible for restoring
1665 // these arguments
1666 __ mv(t0, JavaThread::popframe_force_deopt_reexecution_bit);
1667 __ sw(t0, Address(xthread, JavaThread::popframe_condition_offset()));
1668
1669 // Continue in deoptimization handler
1670 __ ret();
1671
1672 __ bind(caller_not_deoptimized);
1673 }
1674
1675 __ remove_activation(vtos,
1676 /* throw_monitor_exception */ false,
1677 /* install_monitor_exception */ false,
1678 /* notify_jvmdi */ false);
1679
1680 // Restore the last_sp and null it out
1681 __ ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1682 __ shadd(esp, t0, fp, t0, LogBytesPerWord);
1683 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1684
1685 __ restore_bcp();
1686 __ restore_locals();
1687 __ restore_constant_pool_cache();
1688 __ get_method(xmethod);
1689 __ get_dispatch();
1690
1691 // The method data pointer was incremented already during
1692 // call profiling. We have to restore the mdp for the current bcp.
1693 if (ProfileInterpreter) {
1694 __ set_method_data_pointer_for_bcp();
1695 }
1696
1697 // Clear the popframe condition flag
1698 __ sw(zr, Address(xthread, JavaThread::popframe_condition_offset()));
1699 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1700
1701 #if INCLUDE_JVMTI
1702 {
1703 Label L_done;
1704
1705 __ lbu(t0, Address(xbcp, 0));
1706 __ mv(t1, Bytecodes::_invokestatic);
1707 __ bne(t1, t0, L_done);
1708
1709 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1710 // Detect such a case in the InterpreterRuntime function and return the member name argument,or null.
1711
1712 __ ld(c_rarg0, Address(xlocals, 0));
1713 __ call_VM(x10, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null),c_rarg0, xmethod, xbcp);
1714
1715 __ beqz(x10, L_done);
1716
1717 __ sd(x10, Address(esp, 0));
1718 __ bind(L_done);
1719 }
1720 #endif // INCLUDE_JVMTI
1721
1722 // Restore machine SP
1723 __ restore_sp_after_call();
1724
1725 __ dispatch_next(vtos);
1726 // end of PopFrame support
1727
1728 Interpreter::_remove_activation_entry = __ pc();
1729
1730 // preserve exception over this code sequence
1731 __ pop_ptr(x10);
1732 __ sd(x10, Address(xthread, JavaThread::vm_result_offset()));
1733 // remove the activation (without doing throws on illegalMonitorExceptions)
1734 __ remove_activation(vtos, false, true, false);
1735 // restore exception
1736 __ get_vm_result(x10, xthread);
1737
1738 // In between activations - previous activation type unknown yet
1739 // compute continuation point - the continuation point expects the
1740 // following registers set up:
1741 //
1742 // x10: exception
1743 // ra: return address/pc that threw exception
1744 // sp: expression stack of caller
1745 // fp: fp of caller
1746 // FIXME: There's no point saving ra here because VM calls don't trash it
1747 __ subi(sp, sp, 2 * wordSize);
1748 __ sd(x10, Address(sp, 0)); // save exception
1749 __ sd(ra, Address(sp, wordSize)); // save return address
1750 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1751 SharedRuntime::exception_handler_for_return_address),
1752 xthread, ra);
1753 __ mv(x11, x10); // save exception handler
1754 __ ld(x10, Address(sp, 0)); // restore exception
1755 __ ld(ra, Address(sp, wordSize)); // restore return address
1756 __ addi(sp, sp, 2 * wordSize);
1757 // We might be returning to a deopt handler that expects x13 to
1758 // contain the exception pc
1759 __ mv(x13, ra);
1760 // Note that an "issuing PC" is actually the next PC after the call
1761 __ jr(x11); // jump to exception
1762 // handler of caller
1763 }
1764
1765 //
1766 // JVMTI ForceEarlyReturn support
1767 //
1768 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1769 address entry = __ pc();
1770
1771 __ restore_bcp();
1772 __ restore_locals();
1773 __ empty_expression_stack();
1774 __ load_earlyret_value(state);
1775
1776 __ ld(t0, Address(xthread, JavaThread::jvmti_thread_state_offset()));
1777 Address cond_addr(t0, JvmtiThreadState::earlyret_state_offset());
1778
1779 // Clear the earlyret state
1780 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1781 __ sd(zr, cond_addr);
1782
1783 __ remove_activation(state,
1784 false, /* throw_monitor_exception */
1785 false, /* install_monitor_exception */
1786 true); /* notify_jvmdi */
1787 __ ret();
1788
1789 return entry;
1790 }
1791 // end of ForceEarlyReturn support
1792
1793 //-----------------------------------------------------------------------------
1794 // Helper for vtos entry point generation
1795
1796 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1797 address& bep,
1798 address& cep,
1799 address& sep,
1800 address& aep,
1801 address& iep,
1802 address& lep,
1803 address& fep,
1804 address& dep,
1805 address& vep) {
1806 assert(t != nullptr && t->is_valid() && t->tos_in() == vtos, "illegal template");
1807 Label L;
1808 aep = __ pc(); // atos entry point
1809 __ push_ptr();
1810 __ j(L);
1811 fep = __ pc(); // ftos entry point
1812 __ push_f();
1813 __ j(L);
1814 dep = __ pc(); // dtos entry point
1815 __ push_d();
1816 __ j(L);
1817 lep = __ pc(); // ltos entry point
1818 __ push_l();
1819 __ j(L);
1820 bep = cep = sep = iep = __ pc(); // [bcsi]tos entry point
1821 __ push_i();
1822 vep = __ pc(); // vtos entry point
1823 __ bind(L);
1824 generate_and_dispatch(t);
1825 }
1826
1827 //-----------------------------------------------------------------------------
1828
1829 void TemplateInterpreterGenerator::count_bytecode() {
1830 __ mv(x7, (address) &BytecodeCounter::_counter_value);
1831 __ atomic_addw(noreg, 1, x7);
1832 }
1833
1834 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1835 __ mv(x7, (address) &BytecodeHistogram::_counters[t->bytecode()]);
1836 __ atomic_addw(noreg, 1, x7);
1837 }
1838
1839 // Non-product code
1840 #ifndef PRODUCT
1841 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1842 address entry = __ pc();
1843
1844 __ push_reg(ra);
1845 __ push(state);
1846 __ push_reg(RegSet::range(x10, x17) + RegSet::range(x5, x7) + RegSet::range(x28, x31), sp);
1847 __ mv(c_rarg2, x10); // Pass itos
1848 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), c_rarg1, c_rarg2, c_rarg3);
1849 __ pop_reg(RegSet::range(x10, x17) + RegSet::range(x5, x7) + RegSet::range(x28, x31), sp);
1850 __ pop(state);
1851 __ pop_reg(ra);
1852 __ ret(); // return from result handler
1853
1854 return entry;
1855 }
1856
1857 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1858 // Calculate new index for counter:
1859 // _index = (_index >> log2_number_of_codes) |
1860 // (bytecode << log2_number_of_codes);
1861 Register index_addr = t1;
1862 Register index = t0;
1863 __ mv(index_addr, (address) &BytecodePairHistogram::_index);
1864 __ lw(index, index_addr);
1865 __ mv(x7, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1866 __ srli(index, index, BytecodePairHistogram::log2_number_of_codes);
1867 __ orrw(index, x7, index);
1868 __ sw(index, index_addr);
1869 // Bump bucket contents:
1870 // _counters[_index] ++;
1871 Register counter_addr = t1;
1872 __ mv(x7, (address) &BytecodePairHistogram::_counters);
1873 __ shadd(counter_addr, index, x7, counter_addr, LogBytesPerInt);
1874 __ atomic_addw(noreg, 1, counter_addr);
1875 }
1876
1877 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1878 // Call a little run-time stub to avoid blow-up for each bytecode.
1879 // The run-time runtime saves the right registers, depending on
1880 // the tosca in-state for the given template.
1881
1882 assert(Interpreter::trace_code(t->tos_in()) != nullptr, "entry must have been generated");
1883 __ rt_call(Interpreter::trace_code(t->tos_in()));
1884 __ reinit_heapbase();
1885 }
1886
1887 void TemplateInterpreterGenerator::stop_interpreter_at() {
1888 Label L;
1889 __ push_reg(t0);
1890 __ mv(t0, (address) &BytecodeCounter::_counter_value);
1891 __ ld(t0, Address(t0));
1892 __ mv(t1, StopInterpreterAt);
1893 __ bne(t0, t1, L);
1894 __ ebreak();
1895 __ bind(L);
1896 __ pop_reg(t0);
1897 }
1898
1899 #endif // !PRODUCT