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, 2023, 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.hpp"
29 #include "asm/macroAssembler.inline.hpp"
30 #include "code/compiledIC.hpp"
31 #include "code/debugInfoRec.hpp"
32 #include "code/icBuffer.hpp"
33 #include "code/vtableStubs.hpp"
34 #include "compiler/oopMap.hpp"
35 #include "gc/shared/barrierSetAssembler.hpp"
36 #include "interpreter/interp_masm.hpp"
37 #include "interpreter/interpreter.hpp"
38 #include "logging/log.hpp"
39 #include "memory/resourceArea.hpp"
40 #include "nativeInst_riscv.hpp"
41 #include "oops/compiledICHolder.hpp"
42 #include "oops/klass.inline.hpp"
43 #include "oops/method.inline.hpp"
44 #include "prims/methodHandles.hpp"
45 #include "runtime/continuation.hpp"
46 #include "runtime/continuationEntry.inline.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/jniHandles.hpp"
49 #include "runtime/safepointMechanism.hpp"
50 #include "runtime/sharedRuntime.hpp"
51 #include "runtime/signature.hpp"
52 #include "runtime/stubRoutines.hpp"
53 #include "runtime/vframeArray.hpp"
54 #include "utilities/align.hpp"
55 #include "utilities/formatBuffer.hpp"
56 #include "vmreg_riscv.inline.hpp"
57 #ifdef COMPILER1
58 #include "c1/c1_Runtime1.hpp"
59 #endif
60 #ifdef COMPILER2
61 #include "adfiles/ad_riscv.hpp"
62 #include "opto/runtime.hpp"
63 #endif
64 #if INCLUDE_JVMCI
65 #include "jvmci/jvmciJavaClasses.hpp"
66 #endif
67
68 #define __ masm->
69
70 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
71
72 class SimpleRuntimeFrame {
73 public:
74
75 // Most of the runtime stubs have this simple frame layout.
76 // This class exists to make the layout shared in one place.
77 // Offsets are for compiler stack slots, which are jints.
78 enum layout {
79 // The frame sender code expects that fp will be in the "natural" place and
80 // will override any oopMap setting for it. We must therefore force the layout
81 // so that it agrees with the frame sender code.
82 // we don't expect any arg reg save area so riscv asserts that
83 // frame::arg_reg_save_area_bytes == 0
84 fp_off = 0, fp_off2,
85 return_off, return_off2,
86 framesize
87 };
88 };
89
90 class RegisterSaver {
91 const bool _save_vectors;
92 public:
93 RegisterSaver(bool save_vectors) : _save_vectors(UseRVV && save_vectors) {}
94 ~RegisterSaver() {}
95 OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words);
96 void restore_live_registers(MacroAssembler* masm);
97
98 // Offsets into the register save area
99 // Used by deoptimization when it is managing result register
100 // values on its own
101 // gregs:28, float_register:32; except: x1(ra) & x2(sp) & gp(x3) & tp(x4)
102 // |---v0---|<---SP
103 // |---v1---|save vectors only in generate_handler_blob
104 // |-- .. --|
105 // |---v31--|-----
106 // |---f0---|
107 // |---f1---|
108 // | .. |
109 // |---f31--|
110 // |---reserved slot for stack alignment---|
111 // |---x5---|
112 // | x6 |
113 // |---.. --|
114 // |---x31--|
115 // |---fp---|
116 // |---ra---|
117 int v0_offset_in_bytes(void) { return 0; }
118 int f0_offset_in_bytes(void) {
119 int f0_offset = 0;
120 #ifdef COMPILER2
121 if (_save_vectors) {
122 f0_offset += Matcher::scalable_vector_reg_size(T_INT) * VectorRegister::number_of_registers *
123 BytesPerInt;
124 }
125 #endif
126 return f0_offset;
127 }
128 int reserved_slot_offset_in_bytes(void) {
129 return f0_offset_in_bytes() +
130 FloatRegister::max_slots_per_register *
131 FloatRegister::number_of_registers *
132 BytesPerInt;
133 }
134
135 int reg_offset_in_bytes(Register r) {
136 assert (r->encoding() > 4, "ra, sp, gp and tp not saved");
137 return reserved_slot_offset_in_bytes() + (r->encoding() - 4 /* x1, x2, x3, x4 */) * wordSize;
138 }
139
140 int freg_offset_in_bytes(FloatRegister f) {
141 return f0_offset_in_bytes() + f->encoding() * wordSize;
142 }
143
144 int ra_offset_in_bytes(void) {
145 return reserved_slot_offset_in_bytes() +
146 (Register::number_of_registers - 3) *
147 Register::max_slots_per_register *
148 BytesPerInt;
149 }
150 };
151
152 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words) {
153 int vector_size_in_bytes = 0;
154 int vector_size_in_slots = 0;
155 #ifdef COMPILER2
156 if (_save_vectors) {
157 vector_size_in_bytes += Matcher::scalable_vector_reg_size(T_BYTE);
158 vector_size_in_slots += Matcher::scalable_vector_reg_size(T_INT);
159 }
160 #endif
161
162 int frame_size_in_bytes = align_up(additional_frame_words * wordSize + ra_offset_in_bytes() + wordSize, 16);
163 // OopMap frame size is in compiler stack slots (jint's) not bytes or words
164 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
165 // The caller will allocate additional_frame_words
166 int additional_frame_slots = additional_frame_words * wordSize / BytesPerInt;
167 // CodeBlob frame size is in words.
168 int frame_size_in_words = frame_size_in_bytes / wordSize;
169 *total_frame_words = frame_size_in_words;
170
171 // Save Integer, Float and Vector registers.
172 __ enter();
173 __ push_CPU_state(_save_vectors, vector_size_in_bytes);
174
175 // Set an oopmap for the call site. This oopmap will map all
176 // oop-registers and debug-info registers as callee-saved. This
177 // will allow deoptimization at this safepoint to find all possible
178 // debug-info recordings, as well as let GC find all oops.
179
180 OopMapSet *oop_maps = new OopMapSet();
181 OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
182 assert_cond(oop_maps != nullptr && oop_map != nullptr);
183
184 int sp_offset_in_slots = 0;
185 int step_in_slots = 0;
186 if (_save_vectors) {
187 step_in_slots = vector_size_in_slots;
188 for (int i = 0; i < VectorRegister::number_of_registers; i++, sp_offset_in_slots += step_in_slots) {
189 VectorRegister r = as_VectorRegister(i);
190 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset_in_slots), r->as_VMReg());
191 }
192 }
193
194 step_in_slots = FloatRegister::max_slots_per_register;
195 for (int i = 0; i < FloatRegister::number_of_registers; i++, sp_offset_in_slots += step_in_slots) {
196 FloatRegister r = as_FloatRegister(i);
197 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset_in_slots), r->as_VMReg());
198 }
199
200 step_in_slots = Register::max_slots_per_register;
201 // skip the slot reserved for alignment, see MacroAssembler::push_reg;
202 // also skip x5 ~ x6 on the stack because they are caller-saved registers.
203 sp_offset_in_slots += Register::max_slots_per_register * 3;
204 // besides, we ignore x0 ~ x4 because push_CPU_state won't push them on the stack.
205 for (int i = 7; i < Register::number_of_registers; i++, sp_offset_in_slots += step_in_slots) {
206 Register r = as_Register(i);
207 if (r != xthread) {
208 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset_in_slots + additional_frame_slots), r->as_VMReg());
209 }
210 }
211
212 return oop_map;
213 }
214
215 void RegisterSaver::restore_live_registers(MacroAssembler* masm) {
216 #ifdef COMPILER2
217 __ pop_CPU_state(_save_vectors, Matcher::scalable_vector_reg_size(T_BYTE));
218 #else
219 #if !INCLUDE_JVMCI
220 assert(!_save_vectors, "vectors are generated only by C2 and JVMCI");
221 #endif
222 __ pop_CPU_state(_save_vectors);
223 #endif
224 __ leave();
225 }
226
227 // Is vector's size (in bytes) bigger than a size saved by default?
228 // riscv does not ovlerlay the floating-point registers on vector registers like aarch64.
229 bool SharedRuntime::is_wide_vector(int size) {
230 return UseRVV;
231 }
232
233 // ---------------------------------------------------------------------------
234 // Read the array of BasicTypes from a signature, and compute where the
235 // arguments should go. Values in the VMRegPair regs array refer to 4-byte
236 // quantities. Values less than VMRegImpl::stack0 are registers, those above
237 // refer to 4-byte stack slots. All stack slots are based off of the stack pointer
238 // as framesizes are fixed.
239 // VMRegImpl::stack0 refers to the first slot 0(sp).
240 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher.
241 // Register up to Register::number_of_registers) are the 64-bit
242 // integer registers.
243
244 // Note: the INPUTS in sig_bt are in units of Java argument words,
245 // which are 64-bit. The OUTPUTS are in 32-bit units.
246
247 // The Java calling convention is a "shifted" version of the C ABI.
248 // By skipping the first C ABI register we can call non-static jni
249 // methods with small numbers of arguments without having to shuffle
250 // the arguments at all. Since we control the java ABI we ought to at
251 // least get some advantage out of it.
252
253 int SharedRuntime::java_calling_convention(const BasicType *sig_bt,
254 VMRegPair *regs,
255 int total_args_passed) {
256 // Create the mapping between argument positions and
257 // registers.
258 static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = {
259 j_rarg0, j_rarg1, j_rarg2, j_rarg3,
260 j_rarg4, j_rarg5, j_rarg6, j_rarg7
261 };
262 static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters_j] = {
263 j_farg0, j_farg1, j_farg2, j_farg3,
264 j_farg4, j_farg5, j_farg6, j_farg7
265 };
266
267 uint int_args = 0;
268 uint fp_args = 0;
269 uint stk_args = 0;
270
271 for (int i = 0; i < total_args_passed; i++) {
272 switch (sig_bt[i]) {
273 case T_BOOLEAN: // fall through
274 case T_CHAR: // fall through
275 case T_BYTE: // fall through
276 case T_SHORT: // fall through
277 case T_INT:
278 if (int_args < Argument::n_int_register_parameters_j) {
279 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
280 } else {
281 stk_args = align_up(stk_args, 2);
282 regs[i].set1(VMRegImpl::stack2reg(stk_args));
283 stk_args += 1;
284 }
285 break;
286 case T_VOID:
287 // halves of T_LONG or T_DOUBLE
288 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
289 regs[i].set_bad();
290 break;
291 case T_LONG: // fall through
292 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
293 case T_OBJECT: // fall through
294 case T_ARRAY: // fall through
295 case T_ADDRESS:
296 if (int_args < Argument::n_int_register_parameters_j) {
297 regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
298 } else {
299 stk_args = align_up(stk_args, 2);
300 regs[i].set2(VMRegImpl::stack2reg(stk_args));
301 stk_args += 2;
302 }
303 break;
304 case T_FLOAT:
305 if (fp_args < Argument::n_float_register_parameters_j) {
306 regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
307 } else {
308 stk_args = align_up(stk_args, 2);
309 regs[i].set1(VMRegImpl::stack2reg(stk_args));
310 stk_args += 1;
311 }
312 break;
313 case T_DOUBLE:
314 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
315 if (fp_args < Argument::n_float_register_parameters_j) {
316 regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
317 } else {
318 stk_args = align_up(stk_args, 2);
319 regs[i].set2(VMRegImpl::stack2reg(stk_args));
320 stk_args += 2;
321 }
322 break;
323 default:
324 ShouldNotReachHere();
325 }
326 }
327
328 return stk_args;
329 }
330
331 // Patch the callers callsite with entry to compiled code if it exists.
332 static void patch_callers_callsite(MacroAssembler *masm) {
333 Label L;
334 __ ld(t0, Address(xmethod, in_bytes(Method::code_offset())));
335 __ beqz(t0, L);
336
337 __ enter();
338 __ push_CPU_state();
339
340 // VM needs caller's callsite
341 // VM needs target method
342 // This needs to be a long call since we will relocate this adapter to
343 // the codeBuffer and it may not reach
344
345 #ifndef PRODUCT
346 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
347 #endif
348
349 __ mv(c_rarg0, xmethod);
350 __ mv(c_rarg1, ra);
351 RuntimeAddress target(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite));
352 __ relocate(target.rspec(), [&] {
353 int32_t offset;
354 __ la_patchable(t0, target, offset);
355 __ jalr(x1, t0, offset);
356 });
357
358 __ pop_CPU_state();
359 // restore sp
360 __ leave();
361 __ bind(L);
362 }
363
364 static void gen_c2i_adapter(MacroAssembler *masm,
365 int total_args_passed,
366 int comp_args_on_stack,
367 const BasicType *sig_bt,
368 const VMRegPair *regs,
369 Label& skip_fixup) {
370 // Before we get into the guts of the C2I adapter, see if we should be here
371 // at all. We've come from compiled code and are attempting to jump to the
372 // interpreter, which means the caller made a static call to get here
373 // (vcalls always get a compiled target if there is one). Check for a
374 // compiled target. If there is one, we need to patch the caller's call.
375 patch_callers_callsite(masm);
376
377 __ bind(skip_fixup);
378
379 int words_pushed = 0;
380
381 // Since all args are passed on the stack, total_args_passed *
382 // Interpreter::stackElementSize is the space we need.
383
384 int extraspace = total_args_passed * Interpreter::stackElementSize;
385
386 __ mv(x19_sender_sp, sp);
387
388 // stack is aligned, keep it that way
389 extraspace = align_up(extraspace, 2 * wordSize);
390
391 if (extraspace) {
392 __ sub(sp, sp, extraspace);
393 }
394
395 // Now write the args into the outgoing interpreter space
396 for (int i = 0; i < total_args_passed; i++) {
397 if (sig_bt[i] == T_VOID) {
398 assert(i > 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "missing half");
399 continue;
400 }
401
402 // offset to start parameters
403 int st_off = (total_args_passed - i - 1) * Interpreter::stackElementSize;
404 int next_off = st_off - Interpreter::stackElementSize;
405
406 // Say 4 args:
407 // i st_off
408 // 0 32 T_LONG
409 // 1 24 T_VOID
410 // 2 16 T_OBJECT
411 // 3 8 T_BOOL
412 // - 0 return address
413 //
414 // However to make thing extra confusing. Because we can fit a Java long/double in
415 // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
416 // leaves one slot empty and only stores to a single slot. In this case the
417 // slot that is occupied is the T_VOID slot. See I said it was confusing.
418
419 VMReg r_1 = regs[i].first();
420 VMReg r_2 = regs[i].second();
421 if (!r_1->is_valid()) {
422 assert(!r_2->is_valid(), "");
423 continue;
424 }
425 if (r_1->is_stack()) {
426 // memory to memory use t0
427 int ld_off = (r_1->reg2stack() * VMRegImpl::stack_slot_size
428 + extraspace
429 + words_pushed * wordSize);
430 if (!r_2->is_valid()) {
431 __ lwu(t0, Address(sp, ld_off));
432 __ sd(t0, Address(sp, st_off), /*temp register*/esp);
433 } else {
434 __ ld(t0, Address(sp, ld_off), /*temp register*/esp);
435
436 // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
437 // T_DOUBLE and T_LONG use two slots in the interpreter
438 if (sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
439 // ld_off == LSW, ld_off+wordSize == MSW
440 // st_off == MSW, next_off == LSW
441 __ sd(t0, Address(sp, next_off), /*temp register*/esp);
442 #ifdef ASSERT
443 // Overwrite the unused slot with known junk
444 __ mv(t0, 0xdeadffffdeadaaaaul);
445 __ sd(t0, Address(sp, st_off), /*temp register*/esp);
446 #endif /* ASSERT */
447 } else {
448 __ sd(t0, Address(sp, st_off), /*temp register*/esp);
449 }
450 }
451 } else if (r_1->is_Register()) {
452 Register r = r_1->as_Register();
453 if (!r_2->is_valid()) {
454 // must be only an int (or less ) so move only 32bits to slot
455 __ sd(r, Address(sp, st_off));
456 } else {
457 // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
458 // T_DOUBLE and T_LONG use two slots in the interpreter
459 if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
460 // long/double in gpr
461 #ifdef ASSERT
462 // Overwrite the unused slot with known junk
463 __ mv(t0, 0xdeadffffdeadaaabul);
464 __ sd(t0, Address(sp, st_off), /*temp register*/esp);
465 #endif /* ASSERT */
466 __ sd(r, Address(sp, next_off));
467 } else {
468 __ sd(r, Address(sp, st_off));
469 }
470 }
471 } else {
472 assert(r_1->is_FloatRegister(), "");
473 if (!r_2->is_valid()) {
474 // only a float use just part of the slot
475 __ fsw(r_1->as_FloatRegister(), Address(sp, st_off));
476 } else {
477 #ifdef ASSERT
478 // Overwrite the unused slot with known junk
479 __ mv(t0, 0xdeadffffdeadaaacul);
480 __ sd(t0, Address(sp, st_off), /*temp register*/esp);
481 #endif /* ASSERT */
482 __ fsd(r_1->as_FloatRegister(), Address(sp, next_off));
483 }
484 }
485 }
486
487 __ mv(esp, sp); // Interp expects args on caller's expression stack
488
489 __ ld(t0, Address(xmethod, in_bytes(Method::interpreter_entry_offset())));
490 __ jr(t0);
491 }
492
493 void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,
494 int total_args_passed,
495 int comp_args_on_stack,
496 const BasicType *sig_bt,
497 const VMRegPair *regs) {
498 // Note: x19_sender_sp contains the senderSP on entry. We must
499 // preserve it since we may do a i2c -> c2i transition if we lose a
500 // race where compiled code goes non-entrant while we get args
501 // ready.
502
503 // Cut-out for having no stack args.
504 int comp_words_on_stack = align_up(comp_args_on_stack * VMRegImpl::stack_slot_size, wordSize) >> LogBytesPerWord;
505 if (comp_args_on_stack != 0) {
506 __ sub(t0, sp, comp_words_on_stack * wordSize);
507 __ andi(sp, t0, -16);
508 }
509
510 // Will jump to the compiled code just as if compiled code was doing it.
511 // Pre-load the register-jump target early, to schedule it better.
512 __ ld(t1, Address(xmethod, in_bytes(Method::from_compiled_offset())));
513
514 #if INCLUDE_JVMCI
515 if (EnableJVMCI) {
516 // check if this call should be routed towards a specific entry point
517 __ ld(t0, Address(xthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
518 Label no_alternative_target;
519 __ beqz(t0, no_alternative_target);
520 __ mv(t1, t0);
521 __ sd(zr, Address(xthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
522 __ bind(no_alternative_target);
523 }
524 #endif // INCLUDE_JVMCI
525
526 // Now generate the shuffle code.
527 for (int i = 0; i < total_args_passed; i++) {
528 if (sig_bt[i] == T_VOID) {
529 assert(i > 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "missing half");
530 continue;
531 }
532
533 // Pick up 0, 1 or 2 words from SP+offset.
534
535 assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
536 "scrambled load targets?");
537 // Load in argument order going down.
538 int ld_off = (total_args_passed - i - 1) * Interpreter::stackElementSize;
539 // Point to interpreter value (vs. tag)
540 int next_off = ld_off - Interpreter::stackElementSize;
541
542 VMReg r_1 = regs[i].first();
543 VMReg r_2 = regs[i].second();
544 if (!r_1->is_valid()) {
545 assert(!r_2->is_valid(), "");
546 continue;
547 }
548 if (r_1->is_stack()) {
549 // Convert stack slot to an SP offset (+ wordSize to account for return address )
550 int st_off = regs[i].first()->reg2stack() * VMRegImpl::stack_slot_size;
551 if (!r_2->is_valid()) {
552 __ lw(t0, Address(esp, ld_off));
553 __ sd(t0, Address(sp, st_off), /*temp register*/t2);
554 } else {
555 //
556 // We are using two optoregs. This can be either T_OBJECT,
557 // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
558 // two slots but only uses one for thr T_LONG or T_DOUBLE case
559 // So we must adjust where to pick up the data to match the
560 // interpreter.
561 //
562 // Interpreter local[n] == MSW, local[n+1] == LSW however locals
563 // are accessed as negative so LSW is at LOW address
564
565 // ld_off is MSW so get LSW
566 const int offset = (sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) ?
567 next_off : ld_off;
568 __ ld(t0, Address(esp, offset));
569 // st_off is LSW (i.e. reg.first())
570 __ sd(t0, Address(sp, st_off), /*temp register*/t2);
571 }
572 } else if (r_1->is_Register()) { // Register argument
573 Register r = r_1->as_Register();
574 if (r_2->is_valid()) {
575 //
576 // We are using two VMRegs. This can be either T_OBJECT,
577 // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
578 // two slots but only uses one for thr T_LONG or T_DOUBLE case
579 // So we must adjust where to pick up the data to match the
580 // interpreter.
581
582 const int offset = (sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) ?
583 next_off : ld_off;
584
585 // this can be a misaligned move
586 __ ld(r, Address(esp, offset));
587 } else {
588 // sign extend and use a full word?
589 __ lw(r, Address(esp, ld_off));
590 }
591 } else {
592 if (!r_2->is_valid()) {
593 __ flw(r_1->as_FloatRegister(), Address(esp, ld_off));
594 } else {
595 __ fld(r_1->as_FloatRegister(), Address(esp, next_off));
596 }
597 }
598 }
599
600 __ push_cont_fastpath(xthread); // Set JavaThread::_cont_fastpath to the sp of the oldest interpreted frame we know about
601
602 // 6243940 We might end up in handle_wrong_method if
603 // the callee is deoptimized as we race thru here. If that
604 // happens we don't want to take a safepoint because the
605 // caller frame will look interpreted and arguments are now
606 // "compiled" so it is much better to make this transition
607 // invisible to the stack walking code. Unfortunately if
608 // we try and find the callee by normal means a safepoint
609 // is possible. So we stash the desired callee in the thread
610 // and the vm will find there should this case occur.
611
612 __ sd(xmethod, Address(xthread, JavaThread::callee_target_offset()));
613
614 __ jr(t1);
615 }
616
617 // ---------------------------------------------------------------
618 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
619 int total_args_passed,
620 int comp_args_on_stack,
621 const BasicType *sig_bt,
622 const VMRegPair *regs,
623 AdapterFingerPrint* fingerprint) {
624 address i2c_entry = __ pc();
625 gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
626
627 address c2i_unverified_entry = __ pc();
628 Label skip_fixup;
629
630 Label ok;
631
632 const Register holder = t1;
633 const Register receiver = j_rarg0;
634 const Register tmp = t2; // A call-clobbered register not used for arg passing
635
636 // -------------------------------------------------------------------------
637 // Generate a C2I adapter. On entry we know xmethod holds the Method* during calls
638 // to the interpreter. The args start out packed in the compiled layout. They
639 // need to be unpacked into the interpreter layout. This will almost always
640 // require some stack space. We grow the current (compiled) stack, then repack
641 // the args. We finally end in a jump to the generic interpreter entry point.
642 // On exit from the interpreter, the interpreter will restore our SP (lest the
643 // compiled code, which relies solely on SP and not FP, get sick).
644
645 {
646 __ block_comment("c2i_unverified_entry {");
647 __ load_klass(t0, receiver, tmp);
648 __ ld(tmp, Address(holder, CompiledICHolder::holder_klass_offset()));
649 __ ld(xmethod, Address(holder, CompiledICHolder::holder_metadata_offset()));
650 __ beq(t0, tmp, ok);
651 __ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
652
653 __ bind(ok);
654 // Method might have been compiled since the call site was patched to
655 // interpreted; if that is the case treat it as a miss so we can get
656 // the call site corrected.
657 __ ld(t0, Address(xmethod, in_bytes(Method::code_offset())));
658 __ beqz(t0, skip_fixup);
659 __ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
660 __ block_comment("} c2i_unverified_entry");
661 }
662
663 address c2i_entry = __ pc();
664
665 // Class initialization barrier for static methods
666 address c2i_no_clinit_check_entry = nullptr;
667 if (VM_Version::supports_fast_class_init_checks()) {
668 Label L_skip_barrier;
669
670 { // Bypass the barrier for non-static methods
671 __ lwu(t0, Address(xmethod, Method::access_flags_offset()));
672 __ test_bit(t1, t0, exact_log2(JVM_ACC_STATIC));
673 __ beqz(t1, L_skip_barrier); // non-static
674 }
675
676 __ load_method_holder(t1, xmethod);
677 __ clinit_barrier(t1, t0, &L_skip_barrier);
678 __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
679
680 __ bind(L_skip_barrier);
681 c2i_no_clinit_check_entry = __ pc();
682 }
683
684 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
685 bs->c2i_entry_barrier(masm);
686
687 gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
688
689 return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry, c2i_no_clinit_check_entry);
690 }
691
692 int SharedRuntime::vector_calling_convention(VMRegPair *regs,
693 uint num_bits,
694 uint total_args_passed) {
695 Unimplemented();
696 return 0;
697 }
698
699 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
700 VMRegPair *regs,
701 VMRegPair *regs2,
702 int total_args_passed) {
703 assert(regs2 == nullptr, "not needed on riscv");
704
705 // We return the amount of VMRegImpl stack slots we need to reserve for all
706 // the arguments NOT counting out_preserve_stack_slots.
707
708 static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
709 c_rarg0, c_rarg1, c_rarg2, c_rarg3,
710 c_rarg4, c_rarg5, c_rarg6, c_rarg7
711 };
712 static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
713 c_farg0, c_farg1, c_farg2, c_farg3,
714 c_farg4, c_farg5, c_farg6, c_farg7
715 };
716
717 uint int_args = 0;
718 uint fp_args = 0;
719 uint stk_args = 0; // inc by 2 each time
720
721 for (int i = 0; i < total_args_passed; i++) {
722 switch (sig_bt[i]) {
723 case T_BOOLEAN: // fall through
724 case T_CHAR: // fall through
725 case T_BYTE: // fall through
726 case T_SHORT: // fall through
727 case T_INT:
728 if (int_args < Argument::n_int_register_parameters_c) {
729 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
730 } else {
731 regs[i].set1(VMRegImpl::stack2reg(stk_args));
732 stk_args += 2;
733 }
734 break;
735 case T_LONG: // fall through
736 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
737 case T_OBJECT: // fall through
738 case T_ARRAY: // fall through
739 case T_ADDRESS: // fall through
740 case T_METADATA:
741 if (int_args < Argument::n_int_register_parameters_c) {
742 regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
743 } else {
744 regs[i].set2(VMRegImpl::stack2reg(stk_args));
745 stk_args += 2;
746 }
747 break;
748 case T_FLOAT:
749 if (fp_args < Argument::n_float_register_parameters_c) {
750 regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
751 } else if (int_args < Argument::n_int_register_parameters_c) {
752 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
753 } else {
754 regs[i].set1(VMRegImpl::stack2reg(stk_args));
755 stk_args += 2;
756 }
757 break;
758 case T_DOUBLE:
759 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
760 if (fp_args < Argument::n_float_register_parameters_c) {
761 regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
762 } else if (int_args < Argument::n_int_register_parameters_c) {
763 regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
764 } else {
765 regs[i].set2(VMRegImpl::stack2reg(stk_args));
766 stk_args += 2;
767 }
768 break;
769 case T_VOID: // Halves of longs and doubles
770 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
771 regs[i].set_bad();
772 break;
773 default:
774 ShouldNotReachHere();
775 }
776 }
777
778 return stk_args;
779 }
780
781 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
782 // We always ignore the frame_slots arg and just use the space just below frame pointer
783 // which by this time is free to use
784 switch (ret_type) {
785 case T_FLOAT:
786 __ fsw(f10, Address(fp, -3 * wordSize));
787 break;
788 case T_DOUBLE:
789 __ fsd(f10, Address(fp, -3 * wordSize));
790 break;
791 case T_VOID: break;
792 default: {
793 __ sd(x10, Address(fp, -3 * wordSize));
794 }
795 }
796 }
797
798 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
799 // We always ignore the frame_slots arg and just use the space just below frame pointer
800 // which by this time is free to use
801 switch (ret_type) {
802 case T_FLOAT:
803 __ flw(f10, Address(fp, -3 * wordSize));
804 break;
805 case T_DOUBLE:
806 __ fld(f10, Address(fp, -3 * wordSize));
807 break;
808 case T_VOID: break;
809 default: {
810 __ ld(x10, Address(fp, -3 * wordSize));
811 }
812 }
813 }
814
815 static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
816 RegSet x;
817 for ( int i = first_arg ; i < arg_count ; i++ ) {
818 if (args[i].first()->is_Register()) {
819 x = x + args[i].first()->as_Register();
820 } else if (args[i].first()->is_FloatRegister()) {
821 __ addi(sp, sp, -2 * wordSize);
822 __ fsd(args[i].first()->as_FloatRegister(), Address(sp, 0));
823 }
824 }
825 __ push_reg(x, sp);
826 }
827
828 static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
829 RegSet x;
830 for ( int i = first_arg ; i < arg_count ; i++ ) {
831 if (args[i].first()->is_Register()) {
832 x = x + args[i].first()->as_Register();
833 } else {
834 ;
835 }
836 }
837 __ pop_reg(x, sp);
838 for ( int i = arg_count - 1 ; i >= first_arg ; i-- ) {
839 if (args[i].first()->is_Register()) {
840 ;
841 } else if (args[i].first()->is_FloatRegister()) {
842 __ fld(args[i].first()->as_FloatRegister(), Address(sp, 0));
843 __ add(sp, sp, 2 * wordSize);
844 }
845 }
846 }
847
848 static void verify_oop_args(MacroAssembler* masm,
849 const methodHandle& method,
850 const BasicType* sig_bt,
851 const VMRegPair* regs) {
852 const Register temp_reg = x9; // not part of any compiled calling seq
853 if (VerifyOops) {
854 for (int i = 0; i < method->size_of_parameters(); i++) {
855 if (sig_bt[i] == T_OBJECT ||
856 sig_bt[i] == T_ARRAY) {
857 VMReg r = regs[i].first();
858 assert(r->is_valid(), "bad oop arg");
859 if (r->is_stack()) {
860 __ ld(temp_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
861 __ verify_oop(temp_reg);
862 } else {
863 __ verify_oop(r->as_Register());
864 }
865 }
866 }
867 }
868 }
869
870 // on exit, sp points to the ContinuationEntry
871 static OopMap* continuation_enter_setup(MacroAssembler* masm, int& stack_slots) {
872 assert(ContinuationEntry::size() % VMRegImpl::stack_slot_size == 0, "");
873 assert(in_bytes(ContinuationEntry::cont_offset()) % VMRegImpl::stack_slot_size == 0, "");
874 assert(in_bytes(ContinuationEntry::chunk_offset()) % VMRegImpl::stack_slot_size == 0, "");
875
876 stack_slots += (int)ContinuationEntry::size() / wordSize;
877 __ sub(sp, sp, (int)ContinuationEntry::size()); // place Continuation metadata
878
879 OopMap* map = new OopMap(((int)ContinuationEntry::size() + wordSize) / VMRegImpl::stack_slot_size, 0 /* arg_slots*/);
880
881 __ ld(t0, Address(xthread, JavaThread::cont_entry_offset()));
882 __ sd(t0, Address(sp, ContinuationEntry::parent_offset()));
883 __ sd(sp, Address(xthread, JavaThread::cont_entry_offset()));
884
885 return map;
886 }
887
888 // on entry c_rarg1 points to the continuation
889 // sp points to ContinuationEntry
890 // c_rarg3 -- isVirtualThread
891 static void fill_continuation_entry(MacroAssembler* masm) {
892 #ifdef ASSERT
893 __ mv(t0, ContinuationEntry::cookie_value());
894 __ sw(t0, Address(sp, ContinuationEntry::cookie_offset()));
895 #endif
896
897 __ sd(c_rarg1, Address(sp, ContinuationEntry::cont_offset()));
898 __ sw(c_rarg3, Address(sp, ContinuationEntry::flags_offset()));
899 __ sd(zr, Address(sp, ContinuationEntry::chunk_offset()));
900 __ sw(zr, Address(sp, ContinuationEntry::argsize_offset()));
901 __ sw(zr, Address(sp, ContinuationEntry::pin_count_offset()));
902
903 __ ld(t0, Address(xthread, JavaThread::cont_fastpath_offset()));
904 __ sd(t0, Address(sp, ContinuationEntry::parent_cont_fastpath_offset()));
905 __ ld(t0, Address(xthread, JavaThread::held_monitor_count_offset()));
906 __ sd(t0, Address(sp, ContinuationEntry::parent_held_monitor_count_offset()));
907
908 __ sd(zr, Address(xthread, JavaThread::cont_fastpath_offset()));
909 __ sd(zr, Address(xthread, JavaThread::held_monitor_count_offset()));
910 }
911
912 // on entry, sp points to the ContinuationEntry
913 // on exit, fp points to the spilled fp + 2 * wordSize in the entry frame
914 static void continuation_enter_cleanup(MacroAssembler* masm) {
915 #ifndef PRODUCT
916 Label OK;
917 __ ld(t0, Address(xthread, JavaThread::cont_entry_offset()));
918 __ beq(sp, t0, OK);
919 __ stop("incorrect sp");
920 __ bind(OK);
921 #endif
922
923 __ ld(t0, Address(sp, ContinuationEntry::parent_cont_fastpath_offset()));
924 __ sd(t0, Address(xthread, JavaThread::cont_fastpath_offset()));
925 __ ld(t0, Address(sp, ContinuationEntry::parent_held_monitor_count_offset()));
926 __ sd(t0, Address(xthread, JavaThread::held_monitor_count_offset()));
927
928 __ ld(t0, Address(sp, ContinuationEntry::parent_offset()));
929 __ sd(t0, Address(xthread, JavaThread::cont_entry_offset()));
930 __ add(fp, sp, (int)ContinuationEntry::size() + 2 * wordSize /* 2 extra words to match up with leave() */);
931 }
932
933 // enterSpecial(Continuation c, boolean isContinue, boolean isVirtualThread)
934 // On entry: c_rarg1 -- the continuation object
935 // c_rarg2 -- isContinue
936 // c_rarg3 -- isVirtualThread
937 static void gen_continuation_enter(MacroAssembler* masm,
938 const methodHandle& method,
939 const BasicType* sig_bt,
940 const VMRegPair* regs,
941 int& exception_offset,
942 OopMapSet*oop_maps,
943 int& frame_complete,
944 int& stack_slots,
945 int& interpreted_entry_offset,
946 int& compiled_entry_offset) {
947 // verify_oop_args(masm, method, sig_bt, regs);
948 Address resolve(SharedRuntime::get_resolve_static_call_stub(), relocInfo::static_call_type);
949
950 address start = __ pc();
951
952 Label call_thaw, exit;
953
954 // i2i entry used at interp_only_mode only
955 interpreted_entry_offset = __ pc() - start;
956 {
957 #ifdef ASSERT
958 Label is_interp_only;
959 __ lw(t0, Address(xthread, JavaThread::interp_only_mode_offset()));
960 __ bnez(t0, is_interp_only);
961 __ stop("enterSpecial interpreter entry called when not in interp_only_mode");
962 __ bind(is_interp_only);
963 #endif
964
965 // Read interpreter arguments into registers (this is an ad-hoc i2c adapter)
966 __ ld(c_rarg1, Address(esp, Interpreter::stackElementSize * 2));
967 __ ld(c_rarg2, Address(esp, Interpreter::stackElementSize * 1));
968 __ ld(c_rarg3, Address(esp, Interpreter::stackElementSize * 0));
969 __ push_cont_fastpath(xthread);
970
971 __ enter();
972 stack_slots = 2; // will be adjusted in setup
973 OopMap* map = continuation_enter_setup(masm, stack_slots);
974 // The frame is complete here, but we only record it for the compiled entry, so the frame would appear unsafe,
975 // but that's okay because at the very worst we'll miss an async sample, but we're in interp_only_mode anyway.
976
977 fill_continuation_entry(masm);
978
979 __ bnez(c_rarg2, call_thaw);
980
981 // Make sure the call is patchable
982 __ align(NativeInstruction::instruction_size);
983
984 const address tr_call = __ trampoline_call(resolve);
985 if (tr_call == nullptr) {
986 fatal("CodeCache is full at gen_continuation_enter");
987 }
988
989 oop_maps->add_gc_map(__ pc() - start, map);
990 __ post_call_nop();
991
992 __ j(exit);
993
994 CodeBuffer* cbuf = masm->code_section()->outer();
995 address stub = CompiledStaticCall::emit_to_interp_stub(*cbuf, tr_call);
996 if (stub == nullptr) {
997 fatal("CodeCache is full at gen_continuation_enter");
998 }
999 }
1000
1001 // compiled entry
1002 __ align(CodeEntryAlignment);
1003 compiled_entry_offset = __ pc() - start;
1004
1005 __ enter();
1006 stack_slots = 2; // will be adjusted in setup
1007 OopMap* map = continuation_enter_setup(masm, stack_slots);
1008 frame_complete = __ pc() - start;
1009
1010 fill_continuation_entry(masm);
1011
1012 __ bnez(c_rarg2, call_thaw);
1013
1014 // Make sure the call is patchable
1015 __ align(NativeInstruction::instruction_size);
1016
1017 const address tr_call = __ trampoline_call(resolve);
1018 if (tr_call == nullptr) {
1019 fatal("CodeCache is full at gen_continuation_enter");
1020 }
1021
1022 oop_maps->add_gc_map(__ pc() - start, map);
1023 __ post_call_nop();
1024
1025 __ j(exit);
1026
1027 __ bind(call_thaw);
1028
1029 __ rt_call(CAST_FROM_FN_PTR(address, StubRoutines::cont_thaw()));
1030 oop_maps->add_gc_map(__ pc() - start, map->deep_copy());
1031 ContinuationEntry::_return_pc_offset = __ pc() - start;
1032 __ post_call_nop();
1033
1034 __ bind(exit);
1035 continuation_enter_cleanup(masm);
1036 __ leave();
1037 __ ret();
1038
1039 // exception handling
1040 exception_offset = __ pc() - start;
1041 {
1042 __ mv(x9, x10); // save return value contaning the exception oop in callee-saved x9
1043
1044 continuation_enter_cleanup(masm);
1045
1046 __ ld(c_rarg1, Address(fp, -1 * wordSize)); // return address
1047 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), xthread, c_rarg1);
1048
1049 // see OptoRuntime::generate_exception_blob: x10 -- exception oop, x13 -- exception pc
1050
1051 __ mv(x11, x10); // the exception handler
1052 __ mv(x10, x9); // restore return value contaning the exception oop
1053 __ verify_oop(x10);
1054
1055 __ leave();
1056 __ mv(x13, ra);
1057 __ jr(x11); // the exception handler
1058 }
1059
1060 CodeBuffer* cbuf = masm->code_section()->outer();
1061 address stub = CompiledStaticCall::emit_to_interp_stub(*cbuf, tr_call);
1062 if (stub == nullptr) {
1063 fatal("CodeCache is full at gen_continuation_enter");
1064 }
1065 }
1066
1067 static void gen_continuation_yield(MacroAssembler* masm,
1068 const methodHandle& method,
1069 const BasicType* sig_bt,
1070 const VMRegPair* regs,
1071 OopMapSet* oop_maps,
1072 int& frame_complete,
1073 int& stack_slots,
1074 int& compiled_entry_offset) {
1075 enum layout {
1076 fp_off,
1077 fp_off2,
1078 return_off,
1079 return_off2,
1080 framesize // inclusive of return address
1081 };
1082 // assert(is_even(framesize/2), "sp not 16-byte aligned");
1083
1084 stack_slots = framesize / VMRegImpl::slots_per_word;
1085 assert(stack_slots == 2, "recheck layout");
1086
1087 address start = __ pc();
1088
1089 compiled_entry_offset = __ pc() - start;
1090 __ enter();
1091
1092 __ mv(c_rarg1, sp);
1093
1094 frame_complete = __ pc() - start;
1095 address the_pc = __ pc();
1096
1097 __ post_call_nop(); // this must be exactly after the pc value that is pushed into the frame info, we use this nop for fast CodeBlob lookup
1098
1099 __ mv(c_rarg0, xthread);
1100 __ set_last_Java_frame(sp, fp, the_pc, t0);
1101 __ call_VM_leaf(Continuation::freeze_entry(), 2);
1102 __ reset_last_Java_frame(true);
1103
1104 Label pinned;
1105
1106 __ bnez(x10, pinned);
1107
1108 // We've succeeded, set sp to the ContinuationEntry
1109 __ ld(sp, Address(xthread, JavaThread::cont_entry_offset()));
1110 continuation_enter_cleanup(masm);
1111
1112 __ bind(pinned); // pinned -- return to caller
1113
1114 // handle pending exception thrown by freeze
1115 __ ld(t0, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1116 Label ok;
1117 __ beqz(t0, ok);
1118 __ leave();
1119 __ la(t0, RuntimeAddress(StubRoutines::forward_exception_entry()));
1120 __ jr(t0);
1121 __ bind(ok);
1122
1123 __ leave();
1124 __ ret();
1125
1126 OopMap* map = new OopMap(framesize, 1);
1127 oop_maps->add_gc_map(the_pc - start, map);
1128 }
1129
1130 static void gen_special_dispatch(MacroAssembler* masm,
1131 const methodHandle& method,
1132 const BasicType* sig_bt,
1133 const VMRegPair* regs) {
1134 verify_oop_args(masm, method, sig_bt, regs);
1135 vmIntrinsics::ID iid = method->intrinsic_id();
1136
1137 // Now write the args into the outgoing interpreter space
1138 bool has_receiver = false;
1139 Register receiver_reg = noreg;
1140 int member_arg_pos = -1;
1141 Register member_reg = noreg;
1142 int ref_kind = MethodHandles::signature_polymorphic_intrinsic_ref_kind(iid);
1143 if (ref_kind != 0) {
1144 member_arg_pos = method->size_of_parameters() - 1; // trailing MemberName argument
1145 member_reg = x9; // known to be free at this point
1146 has_receiver = MethodHandles::ref_kind_has_receiver(ref_kind);
1147 } else if (iid == vmIntrinsics::_invokeBasic) {
1148 has_receiver = true;
1149 } else if (iid == vmIntrinsics::_linkToNative) {
1150 member_arg_pos = method->size_of_parameters() - 1; // trailing NativeEntryPoint argument
1151 member_reg = x9; // known to be free at this point
1152 } else {
1153 fatal("unexpected intrinsic id %d", vmIntrinsics::as_int(iid));
1154 }
1155
1156 if (member_reg != noreg) {
1157 // Load the member_arg into register, if necessary.
1158 SharedRuntime::check_member_name_argument_is_last_argument(method, sig_bt, regs);
1159 VMReg r = regs[member_arg_pos].first();
1160 if (r->is_stack()) {
1161 __ ld(member_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
1162 } else {
1163 // no data motion is needed
1164 member_reg = r->as_Register();
1165 }
1166 }
1167
1168 if (has_receiver) {
1169 // Make sure the receiver is loaded into a register.
1170 assert(method->size_of_parameters() > 0, "oob");
1171 assert(sig_bt[0] == T_OBJECT, "receiver argument must be an object");
1172 VMReg r = regs[0].first();
1173 assert(r->is_valid(), "bad receiver arg");
1174 if (r->is_stack()) {
1175 // Porting note: This assumes that compiled calling conventions always
1176 // pass the receiver oop in a register. If this is not true on some
1177 // platform, pick a temp and load the receiver from stack.
1178 fatal("receiver always in a register");
1179 receiver_reg = x12; // known to be free at this point
1180 __ ld(receiver_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
1181 } else {
1182 // no data motion is needed
1183 receiver_reg = r->as_Register();
1184 }
1185 }
1186
1187 // Figure out which address we are really jumping to:
1188 MethodHandles::generate_method_handle_dispatch(masm, iid,
1189 receiver_reg, member_reg, /*for_compiler_entry:*/ true);
1190 }
1191
1192 // ---------------------------------------------------------------------------
1193 // Generate a native wrapper for a given method. The method takes arguments
1194 // in the Java compiled code convention, marshals them to the native
1195 // convention (handlizes oops, etc), transitions to native, makes the call,
1196 // returns to java state (possibly blocking), unhandlizes any result and
1197 // returns.
1198 //
1199 // Critical native functions are a shorthand for the use of
1200 // GetPrimtiveArrayCritical and disallow the use of any other JNI
1201 // functions. The wrapper is expected to unpack the arguments before
1202 // passing them to the callee and perform checks before and after the
1203 // native call to ensure that they GCLocker
1204 // lock_critical/unlock_critical semantics are followed. Some other
1205 // parts of JNI setup are skipped like the tear down of the JNI handle
1206 // block and the check for pending exceptions it's impossible for them
1207 // to be thrown.
1208 //
1209 // They are roughly structured like this:
1210 // if (GCLocker::needs_gc()) SharedRuntime::block_for_jni_critical()
1211 // tranistion to thread_in_native
1212 // unpack array arguments and call native entry point
1213 // check for safepoint in progress
1214 // check if any thread suspend flags are set
1215 // call into JVM and possible unlock the JNI critical
1216 // if a GC was suppressed while in the critical native.
1217 // transition back to thread_in_Java
1218 // return to caller
1219 //
1220 nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
1221 const methodHandle& method,
1222 int compile_id,
1223 BasicType* in_sig_bt,
1224 VMRegPair* in_regs,
1225 BasicType ret_type) {
1226 if (method->is_continuation_native_intrinsic()) {
1227 int exception_offset = -1;
1228 OopMapSet* oop_maps = new OopMapSet();
1229 int frame_complete = -1;
1230 int stack_slots = -1;
1231 int interpreted_entry_offset = -1;
1232 int vep_offset = -1;
1233 if (method->is_continuation_enter_intrinsic()) {
1234 gen_continuation_enter(masm,
1235 method,
1236 in_sig_bt,
1237 in_regs,
1238 exception_offset,
1239 oop_maps,
1240 frame_complete,
1241 stack_slots,
1242 interpreted_entry_offset,
1243 vep_offset);
1244 } else if (method->is_continuation_yield_intrinsic()) {
1245 gen_continuation_yield(masm,
1246 method,
1247 in_sig_bt,
1248 in_regs,
1249 oop_maps,
1250 frame_complete,
1251 stack_slots,
1252 vep_offset);
1253 } else {
1254 guarantee(false, "Unknown Continuation native intrinsic");
1255 }
1256
1257 #ifdef ASSERT
1258 if (method->is_continuation_enter_intrinsic()) {
1259 assert(interpreted_entry_offset != -1, "Must be set");
1260 assert(exception_offset != -1, "Must be set");
1261 } else {
1262 assert(interpreted_entry_offset == -1, "Must be unset");
1263 assert(exception_offset == -1, "Must be unset");
1264 }
1265 assert(frame_complete != -1, "Must be set");
1266 assert(stack_slots != -1, "Must be set");
1267 assert(vep_offset != -1, "Must be set");
1268 #endif
1269
1270 __ flush();
1271 nmethod* nm = nmethod::new_native_nmethod(method,
1272 compile_id,
1273 masm->code(),
1274 vep_offset,
1275 frame_complete,
1276 stack_slots,
1277 in_ByteSize(-1),
1278 in_ByteSize(-1),
1279 oop_maps,
1280 exception_offset);
1281 if (method->is_continuation_enter_intrinsic()) {
1282 ContinuationEntry::set_enter_code(nm, interpreted_entry_offset);
1283 } else if (method->is_continuation_yield_intrinsic()) {
1284 _cont_doYield_stub = nm;
1285 } else {
1286 guarantee(false, "Unknown Continuation native intrinsic");
1287 }
1288 return nm;
1289 }
1290
1291 if (method->is_method_handle_intrinsic()) {
1292 vmIntrinsics::ID iid = method->intrinsic_id();
1293 intptr_t start = (intptr_t)__ pc();
1294 int vep_offset = ((intptr_t)__ pc()) - start;
1295
1296 // First instruction must be a nop as it may need to be patched on deoptimisation
1297 {
1298 Assembler::IncompressibleRegion ir(masm); // keep the nop as 4 bytes for patching.
1299 MacroAssembler::assert_alignment(__ pc());
1300 __ nop(); // 4 bytes
1301 }
1302 gen_special_dispatch(masm,
1303 method,
1304 in_sig_bt,
1305 in_regs);
1306 int frame_complete = ((intptr_t)__ pc()) - start; // not complete, period
1307 __ flush();
1308 int stack_slots = SharedRuntime::out_preserve_stack_slots(); // no out slots at all, actually
1309 return nmethod::new_native_nmethod(method,
1310 compile_id,
1311 masm->code(),
1312 vep_offset,
1313 frame_complete,
1314 stack_slots / VMRegImpl::slots_per_word,
1315 in_ByteSize(-1),
1316 in_ByteSize(-1),
1317 (OopMapSet*)nullptr);
1318 }
1319 address native_func = method->native_function();
1320 assert(native_func != nullptr, "must have function");
1321
1322 // An OopMap for lock (and class if static)
1323 OopMapSet *oop_maps = new OopMapSet();
1324 assert_cond(oop_maps != nullptr);
1325 intptr_t start = (intptr_t)__ pc();
1326
1327 // We have received a description of where all the java arg are located
1328 // on entry to the wrapper. We need to convert these args to where
1329 // the jni function will expect them. To figure out where they go
1330 // we convert the java signature to a C signature by inserting
1331 // the hidden arguments as arg[0] and possibly arg[1] (static method)
1332
1333 const int total_in_args = method->size_of_parameters();
1334 int total_c_args = total_in_args + (method->is_static() ? 2 : 1);
1335
1336 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args);
1337 VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args);
1338 BasicType* in_elem_bt = nullptr;
1339
1340 int argc = 0;
1341 out_sig_bt[argc++] = T_ADDRESS;
1342 if (method->is_static()) {
1343 out_sig_bt[argc++] = T_OBJECT;
1344 }
1345
1346 for (int i = 0; i < total_in_args ; i++) {
1347 out_sig_bt[argc++] = in_sig_bt[i];
1348 }
1349
1350 // Now figure out where the args must be stored and how much stack space
1351 // they require.
1352 int out_arg_slots = c_calling_convention(out_sig_bt, out_regs, nullptr, total_c_args);
1353
1354 // Compute framesize for the wrapper. We need to handlize all oops in
1355 // incoming registers
1356
1357 // Calculate the total number of stack slots we will need.
1358
1359 // First count the abi requirement plus all of the outgoing args
1360 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
1361
1362 // Now the space for the inbound oop handle area
1363 int total_save_slots = 8 * VMRegImpl::slots_per_word; // 8 arguments passed in registers
1364
1365 int oop_handle_offset = stack_slots;
1366 stack_slots += total_save_slots;
1367
1368 // Now any space we need for handlizing a klass if static method
1369
1370 int klass_slot_offset = 0;
1371 int klass_offset = -1;
1372 int lock_slot_offset = 0;
1373 bool is_static = false;
1374
1375 if (method->is_static()) {
1376 klass_slot_offset = stack_slots;
1377 stack_slots += VMRegImpl::slots_per_word;
1378 klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size;
1379 is_static = true;
1380 }
1381
1382 // Plus a lock if needed
1383
1384 if (method->is_synchronized()) {
1385 lock_slot_offset = stack_slots;
1386 stack_slots += VMRegImpl::slots_per_word;
1387 }
1388
1389 // Now a place (+2) to save return values or temp during shuffling
1390 // + 4 for return address (which we own) and saved fp
1391 stack_slots += 6;
1392
1393 // Ok The space we have allocated will look like:
1394 //
1395 //
1396 // FP-> | |
1397 // | 2 slots (ra) |
1398 // | 2 slots (fp) |
1399 // |---------------------|
1400 // | 2 slots for moves |
1401 // |---------------------|
1402 // | lock box (if sync) |
1403 // |---------------------| <- lock_slot_offset
1404 // | klass (if static) |
1405 // |---------------------| <- klass_slot_offset
1406 // | oopHandle area |
1407 // |---------------------| <- oop_handle_offset (8 java arg registers)
1408 // | outbound memory |
1409 // | based arguments |
1410 // | |
1411 // |---------------------|
1412 // | |
1413 // SP-> | out_preserved_slots |
1414 //
1415 //
1416
1417
1418 // Now compute actual number of stack words we need rounding to make
1419 // stack properly aligned.
1420 stack_slots = align_up(stack_slots, StackAlignmentInSlots);
1421
1422 int stack_size = stack_slots * VMRegImpl::stack_slot_size;
1423
1424 // First thing make an ic check to see if we should even be here
1425
1426 // We are free to use all registers as temps without saving them and
1427 // restoring them except fp. fp is the only callee save register
1428 // as far as the interpreter and the compiler(s) are concerned.
1429
1430
1431 const Register ic_reg = t1;
1432 const Register receiver = j_rarg0;
1433
1434 Label hit;
1435 Label exception_pending;
1436
1437 __ verify_oop(receiver);
1438 assert_different_registers(ic_reg, receiver, t0, t2);
1439 __ cmp_klass(receiver, ic_reg, t0, t2 /* call-clobbered t2 as a tmp */, hit);
1440
1441 __ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1442
1443 // Verified entry point must be aligned
1444 __ align(8);
1445
1446 __ bind(hit);
1447
1448 int vep_offset = ((intptr_t)__ pc()) - start;
1449
1450 // If we have to make this method not-entrant we'll overwrite its
1451 // first instruction with a jump.
1452 {
1453 Assembler::IncompressibleRegion ir(masm); // keep the nop as 4 bytes for patching.
1454 MacroAssembler::assert_alignment(__ pc());
1455 __ nop(); // 4 bytes
1456 }
1457
1458 if (VM_Version::supports_fast_class_init_checks() && method->needs_clinit_barrier()) {
1459 Label L_skip_barrier;
1460 __ mov_metadata(t1, method->method_holder()); // InstanceKlass*
1461 __ clinit_barrier(t1, t0, &L_skip_barrier);
1462 __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
1463
1464 __ bind(L_skip_barrier);
1465 }
1466
1467 // Generate stack overflow check
1468 __ bang_stack_with_offset(checked_cast<int>(StackOverflow::stack_shadow_zone_size()));
1469
1470 // Generate a new frame for the wrapper.
1471 __ enter();
1472 // -2 because return address is already present and so is saved fp
1473 __ sub(sp, sp, stack_size - 2 * wordSize);
1474
1475 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
1476 assert_cond(bs != nullptr);
1477 bs->nmethod_entry_barrier(masm, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
1478
1479 // Frame is now completed as far as size and linkage.
1480 int frame_complete = ((intptr_t)__ pc()) - start;
1481
1482 // We use x18 as the oop handle for the receiver/klass
1483 // It is callee save so it survives the call to native
1484
1485 const Register oop_handle_reg = x18;
1486
1487 //
1488 // We immediately shuffle the arguments so that any vm call we have to
1489 // make from here on out (sync slow path, jvmti, etc.) we will have
1490 // captured the oops from our caller and have a valid oopMap for
1491 // them.
1492
1493 // -----------------
1494 // The Grand Shuffle
1495
1496 // The Java calling convention is either equal (linux) or denser (win64) than the
1497 // c calling convention. However the because of the jni_env argument the c calling
1498 // convention always has at least one more (and two for static) arguments than Java.
1499 // Therefore if we move the args from java -> c backwards then we will never have
1500 // a register->register conflict and we don't have to build a dependency graph
1501 // and figure out how to break any cycles.
1502 //
1503
1504 // Record esp-based slot for receiver on stack for non-static methods
1505 int receiver_offset = -1;
1506
1507 // This is a trick. We double the stack slots so we can claim
1508 // the oops in the caller's frame. Since we are sure to have
1509 // more args than the caller doubling is enough to make
1510 // sure we can capture all the incoming oop args from the
1511 // caller.
1512 //
1513 OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1514 assert_cond(map != nullptr);
1515
1516 int float_args = 0;
1517 int int_args = 0;
1518
1519 #ifdef ASSERT
1520 bool reg_destroyed[Register::number_of_registers];
1521 bool freg_destroyed[FloatRegister::number_of_registers];
1522 for ( int r = 0 ; r < Register::number_of_registers ; r++ ) {
1523 reg_destroyed[r] = false;
1524 }
1525 for ( int f = 0 ; f < FloatRegister::number_of_registers ; f++ ) {
1526 freg_destroyed[f] = false;
1527 }
1528
1529 #endif /* ASSERT */
1530
1531 // For JNI natives the incoming and outgoing registers are offset upwards.
1532 GrowableArray<int> arg_order(2 * total_in_args);
1533 VMRegPair tmp_vmreg;
1534 tmp_vmreg.set2(x9->as_VMReg());
1535
1536 for (int i = total_in_args - 1, c_arg = total_c_args - 1; i >= 0; i--, c_arg--) {
1537 arg_order.push(i);
1538 arg_order.push(c_arg);
1539 }
1540
1541 int temploc = -1;
1542 for (int ai = 0; ai < arg_order.length(); ai += 2) {
1543 int i = arg_order.at(ai);
1544 int c_arg = arg_order.at(ai + 1);
1545 __ block_comment(err_msg("mv %d -> %d", i, c_arg));
1546 assert(c_arg != -1 && i != -1, "wrong order");
1547 #ifdef ASSERT
1548 if (in_regs[i].first()->is_Register()) {
1549 assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!");
1550 } else if (in_regs[i].first()->is_FloatRegister()) {
1551 assert(!freg_destroyed[in_regs[i].first()->as_FloatRegister()->encoding()], "destroyed reg!");
1552 }
1553 if (out_regs[c_arg].first()->is_Register()) {
1554 reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
1555 } else if (out_regs[c_arg].first()->is_FloatRegister()) {
1556 freg_destroyed[out_regs[c_arg].first()->as_FloatRegister()->encoding()] = true;
1557 }
1558 #endif /* ASSERT */
1559 switch (in_sig_bt[i]) {
1560 case T_ARRAY:
1561 case T_OBJECT:
1562 __ object_move(map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
1563 ((i == 0) && (!is_static)),
1564 &receiver_offset);
1565 int_args++;
1566 break;
1567 case T_VOID:
1568 break;
1569
1570 case T_FLOAT:
1571 __ float_move(in_regs[i], out_regs[c_arg]);
1572 float_args++;
1573 break;
1574
1575 case T_DOUBLE:
1576 assert( i + 1 < total_in_args &&
1577 in_sig_bt[i + 1] == T_VOID &&
1578 out_sig_bt[c_arg + 1] == T_VOID, "bad arg list");
1579 __ double_move(in_regs[i], out_regs[c_arg]);
1580 float_args++;
1581 break;
1582
1583 case T_LONG :
1584 __ long_move(in_regs[i], out_regs[c_arg]);
1585 int_args++;
1586 break;
1587
1588 case T_ADDRESS:
1589 assert(false, "found T_ADDRESS in java args");
1590 break;
1591
1592 default:
1593 __ move32_64(in_regs[i], out_regs[c_arg]);
1594 int_args++;
1595 }
1596 }
1597
1598 // point c_arg at the first arg that is already loaded in case we
1599 // need to spill before we call out
1600 int c_arg = total_c_args - total_in_args;
1601
1602 // Pre-load a static method's oop into c_rarg1.
1603 if (method->is_static()) {
1604
1605 // load oop into a register
1606 __ movoop(c_rarg1,
1607 JNIHandles::make_local(method->method_holder()->java_mirror()));
1608
1609 // Now handlize the static class mirror it's known not-null.
1610 __ sd(c_rarg1, Address(sp, klass_offset));
1611 map->set_oop(VMRegImpl::stack2reg(klass_slot_offset));
1612
1613 // Now get the handle
1614 __ la(c_rarg1, Address(sp, klass_offset));
1615 // and protect the arg if we must spill
1616 c_arg--;
1617 }
1618
1619 // Change state to native (we save the return address in the thread, since it might not
1620 // be pushed on the stack when we do a stack traversal).
1621 // We use the same pc/oopMap repeatedly when we call out
1622
1623 Label native_return;
1624 __ set_last_Java_frame(sp, noreg, native_return, t0);
1625
1626 Label dtrace_method_entry, dtrace_method_entry_done;
1627 {
1628 ExternalAddress target((address)&DTraceMethodProbes);
1629 __ relocate(target.rspec(), [&] {
1630 int32_t offset;
1631 __ la_patchable(t0, target, offset);
1632 __ lbu(t0, Address(t0, offset));
1633 });
1634 __ bnez(t0, dtrace_method_entry);
1635 __ bind(dtrace_method_entry_done);
1636 }
1637
1638 // RedefineClasses() tracing support for obsolete method entry
1639 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1640 // protect the args we've loaded
1641 save_args(masm, total_c_args, c_arg, out_regs);
1642 __ mov_metadata(c_rarg1, method());
1643 __ call_VM_leaf(
1644 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1645 xthread, c_rarg1);
1646 restore_args(masm, total_c_args, c_arg, out_regs);
1647 }
1648
1649 // Lock a synchronized method
1650
1651 // Register definitions used by locking and unlocking
1652
1653 const Register swap_reg = x10;
1654 const Register obj_reg = x9; // Will contain the oop
1655 const Register lock_reg = x30; // Address of compiler lock object (BasicLock)
1656 const Register old_hdr = x30; // value of old header at unlock time
1657 const Register lock_tmp = x31; // Temporary used by lightweight_lock/unlock
1658 const Register tmp = ra;
1659
1660 Label slow_path_lock;
1661 Label lock_done;
1662
1663 if (method->is_synchronized()) {
1664 Label count;
1665
1666 const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes();
1667
1668 // Get the handle (the 2nd argument)
1669 __ mv(oop_handle_reg, c_rarg1);
1670
1671 // Get address of the box
1672
1673 __ la(lock_reg, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1674
1675 // Load the oop from the handle
1676 __ ld(obj_reg, Address(oop_handle_reg, 0));
1677
1678 if (LockingMode == LM_MONITOR) {
1679 __ j(slow_path_lock);
1680 } else if (LockingMode == LM_LEGACY) {
1681 // Load (object->mark() | 1) into swap_reg % x10
1682 __ ld(t0, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1683 __ ori(swap_reg, t0, 1);
1684
1685 // Save (object->mark() | 1) into BasicLock's displaced header
1686 __ sd(swap_reg, Address(lock_reg, mark_word_offset));
1687
1688 // src -> dest if dest == x10 else x10 <- dest
1689 __ cmpxchg_obj_header(x10, lock_reg, obj_reg, t0, count, /*fallthrough*/nullptr);
1690
1691 // Test if the oopMark is an obvious stack pointer, i.e.,
1692 // 1) (mark & 3) == 0, and
1693 // 2) sp <= mark < mark + os::pagesize()
1694 // These 3 tests can be done by evaluating the following
1695 // expression: ((mark - sp) & (3 - os::vm_page_size())),
1696 // assuming both stack pointer and pagesize have their
1697 // least significant 2 bits clear.
1698 // NOTE: the oopMark is in swap_reg % 10 as the result of cmpxchg
1699
1700 __ sub(swap_reg, swap_reg, sp);
1701 __ andi(swap_reg, swap_reg, 3 - (int)os::vm_page_size());
1702
1703 // Save the test result, for recursive case, the result is zero
1704 __ sd(swap_reg, Address(lock_reg, mark_word_offset));
1705 __ bnez(swap_reg, slow_path_lock);
1706 } else {
1707 assert(LockingMode == LM_LIGHTWEIGHT, "");
1708 __ ld(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1709 __ lightweight_lock(obj_reg, swap_reg, tmp, lock_tmp, slow_path_lock);
1710 }
1711
1712 __ bind(count);
1713 __ increment(Address(xthread, JavaThread::held_monitor_count_offset()));
1714
1715 // Slow path will re-enter here
1716 __ bind(lock_done);
1717 }
1718
1719
1720 // Finally just about ready to make the JNI call
1721
1722 // get JNIEnv* which is first argument to native
1723 __ la(c_rarg0, Address(xthread, in_bytes(JavaThread::jni_environment_offset())));
1724
1725 // Now set thread in native
1726 __ la(t1, Address(xthread, JavaThread::thread_state_offset()));
1727 __ mv(t0, _thread_in_native);
1728 __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1729 __ sw(t0, Address(t1));
1730
1731 __ rt_call(native_func);
1732
1733 __ bind(native_return);
1734
1735 intptr_t return_pc = (intptr_t) __ pc();
1736 oop_maps->add_gc_map(return_pc - start, map);
1737
1738 // Unpack native results.
1739 if (ret_type != T_OBJECT && ret_type != T_ARRAY) {
1740 __ cast_primitive_type(ret_type, x10);
1741 }
1742
1743 Label safepoint_in_progress, safepoint_in_progress_done;
1744 Label after_transition;
1745
1746 // Switch thread to "native transition" state before reading the synchronization state.
1747 // This additional state is necessary because reading and testing the synchronization
1748 // state is not atomic w.r.t. GC, as this scenario demonstrates:
1749 // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
1750 // VM thread changes sync state to synchronizing and suspends threads for GC.
1751 // Thread A is resumed to finish this native method, but doesn't block here since it
1752 // didn't see any synchronization is progress, and escapes.
1753 __ mv(t0, _thread_in_native_trans);
1754
1755 __ sw(t0, Address(xthread, JavaThread::thread_state_offset()));
1756
1757 // Force this write out before the read below
1758 if (!UseSystemMemoryBarrier) {
1759 __ membar(MacroAssembler::AnyAny);
1760 }
1761
1762 // check for safepoint operation in progress and/or pending suspend requests
1763 {
1764 // We need an acquire here to ensure that any subsequent load of the
1765 // global SafepointSynchronize::_state flag is ordered after this load
1766 // of the thread-local polling word. We don't want this poll to
1767 // return false (i.e. not safepointing) and a later poll of the global
1768 // SafepointSynchronize::_state spuriously to return true.
1769 // This is to avoid a race when we're in a native->Java transition
1770 // racing the code which wakes up from a safepoint.
1771
1772 __ safepoint_poll(safepoint_in_progress, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1773 __ lwu(t0, Address(xthread, JavaThread::suspend_flags_offset()));
1774 __ bnez(t0, safepoint_in_progress);
1775 __ bind(safepoint_in_progress_done);
1776 }
1777
1778 // change thread state
1779 __ la(t1, Address(xthread, JavaThread::thread_state_offset()));
1780 __ mv(t0, _thread_in_Java);
1781 __ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
1782 __ sw(t0, Address(t1));
1783 __ bind(after_transition);
1784
1785 Label reguard;
1786 Label reguard_done;
1787 __ lbu(t0, Address(xthread, JavaThread::stack_guard_state_offset()));
1788 __ mv(t1, StackOverflow::stack_guard_yellow_reserved_disabled);
1789 __ beq(t0, t1, reguard);
1790 __ bind(reguard_done);
1791
1792 // native result if any is live
1793
1794 // Unlock
1795 Label unlock_done;
1796 Label slow_path_unlock;
1797 if (method->is_synchronized()) {
1798
1799 // Get locked oop from the handle we passed to jni
1800 __ ld(obj_reg, Address(oop_handle_reg, 0));
1801
1802 Label done, not_recursive;
1803
1804 if (LockingMode == LM_LEGACY) {
1805 // Simple recursive lock?
1806 __ ld(t0, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1807 __ bnez(t0, not_recursive);
1808 __ decrement(Address(xthread, JavaThread::held_monitor_count_offset()));
1809 __ j(done);
1810 }
1811
1812 __ bind(not_recursive);
1813
1814 // Must save x10 if if it is live now because cmpxchg must use it
1815 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1816 save_native_result(masm, ret_type, stack_slots);
1817 }
1818
1819 if (LockingMode == LM_MONITOR) {
1820 __ j(slow_path_unlock);
1821 } else if (LockingMode == LM_LEGACY) {
1822 // get address of the stack lock
1823 __ la(x10, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1824 // get old displaced header
1825 __ ld(old_hdr, Address(x10, 0));
1826
1827 // Atomic swap old header if oop still contains the stack lock
1828 Label count;
1829 __ cmpxchg_obj_header(x10, old_hdr, obj_reg, t0, count, &slow_path_unlock);
1830 __ bind(count);
1831 __ decrement(Address(xthread, JavaThread::held_monitor_count_offset()));
1832 } else {
1833 assert(LockingMode == LM_LIGHTWEIGHT, "");
1834 __ ld(old_hdr, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1835 __ test_bit(t0, old_hdr, exact_log2(markWord::monitor_value));
1836 __ bnez(t0, slow_path_unlock);
1837 __ lightweight_unlock(obj_reg, old_hdr, swap_reg, lock_tmp, slow_path_unlock);
1838 __ decrement(Address(xthread, JavaThread::held_monitor_count_offset()));
1839 }
1840
1841 // slow path re-enters here
1842 __ bind(unlock_done);
1843 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1844 restore_native_result(masm, ret_type, stack_slots);
1845 }
1846
1847 __ bind(done);
1848 }
1849
1850 Label dtrace_method_exit, dtrace_method_exit_done;
1851 {
1852 ExternalAddress target((address)&DTraceMethodProbes);
1853 __ relocate(target.rspec(), [&] {
1854 int32_t offset;
1855 __ la_patchable(t0, target, offset);
1856 __ lbu(t0, Address(t0, offset));
1857 });
1858 __ bnez(t0, dtrace_method_exit);
1859 __ bind(dtrace_method_exit_done);
1860 }
1861
1862 __ reset_last_Java_frame(false);
1863
1864 // Unbox oop result, e.g. JNIHandles::resolve result.
1865 if (is_reference_type(ret_type)) {
1866 __ resolve_jobject(x10, x11, x12);
1867 }
1868
1869 if (CheckJNICalls) {
1870 // clear_pending_jni_exception_check
1871 __ sd(zr, Address(xthread, JavaThread::pending_jni_exception_check_fn_offset()));
1872 }
1873
1874 // reset handle block
1875 __ ld(x12, Address(xthread, JavaThread::active_handles_offset()));
1876 __ sd(zr, Address(x12, JNIHandleBlock::top_offset()));
1877
1878 __ leave();
1879
1880 // Any exception pending?
1881 __ ld(t0, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1882 __ bnez(t0, exception_pending);
1883
1884 // We're done
1885 __ ret();
1886
1887 // Unexpected paths are out of line and go here
1888
1889 // forward the exception
1890 __ bind(exception_pending);
1891
1892 // and forward the exception
1893 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1894
1895 // Slow path locking & unlocking
1896 if (method->is_synchronized()) {
1897
1898 __ block_comment("Slow path lock {");
1899 __ bind(slow_path_lock);
1900
1901 // has last_Java_frame setup. No exceptions so do vanilla call not call_VM
1902 // args are (oop obj, BasicLock* lock, JavaThread* thread)
1903
1904 // protect the args we've loaded
1905 save_args(masm, total_c_args, c_arg, out_regs);
1906
1907 __ mv(c_rarg0, obj_reg);
1908 __ mv(c_rarg1, lock_reg);
1909 __ mv(c_rarg2, xthread);
1910
1911 // Not a leaf but we have last_Java_frame setup as we want
1912 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), 3);
1913 restore_args(masm, total_c_args, c_arg, out_regs);
1914
1915 #ifdef ASSERT
1916 { Label L;
1917 __ ld(t0, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1918 __ beqz(t0, L);
1919 __ stop("no pending exception allowed on exit from monitorenter");
1920 __ bind(L);
1921 }
1922 #endif
1923 __ j(lock_done);
1924
1925 __ block_comment("} Slow path lock");
1926
1927 __ block_comment("Slow path unlock {");
1928 __ bind(slow_path_unlock);
1929
1930 if (ret_type == T_FLOAT || ret_type == T_DOUBLE) {
1931 save_native_result(masm, ret_type, stack_slots);
1932 }
1933
1934 __ mv(c_rarg2, xthread);
1935 __ la(c_rarg1, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1936 __ mv(c_rarg0, obj_reg);
1937
1938 // Save pending exception around call to VM (which contains an EXCEPTION_MARK)
1939 // NOTE that obj_reg == x9 currently
1940 __ ld(x9, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1941 __ sd(zr, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1942
1943 __ rt_call(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C));
1944
1945 #ifdef ASSERT
1946 {
1947 Label L;
1948 __ ld(t0, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1949 __ beqz(t0, L);
1950 __ stop("no pending exception allowed on exit complete_monitor_unlocking_C");
1951 __ bind(L);
1952 }
1953 #endif /* ASSERT */
1954
1955 __ sd(x9, Address(xthread, in_bytes(Thread::pending_exception_offset())));
1956
1957 if (ret_type == T_FLOAT || ret_type == T_DOUBLE) {
1958 restore_native_result(masm, ret_type, stack_slots);
1959 }
1960 __ j(unlock_done);
1961
1962 __ block_comment("} Slow path unlock");
1963
1964 } // synchronized
1965
1966 // SLOW PATH Reguard the stack if needed
1967
1968 __ bind(reguard);
1969 save_native_result(masm, ret_type, stack_slots);
1970 __ rt_call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1971 restore_native_result(masm, ret_type, stack_slots);
1972 // and continue
1973 __ j(reguard_done);
1974
1975 // SLOW PATH safepoint
1976 {
1977 __ block_comment("safepoint {");
1978 __ bind(safepoint_in_progress);
1979
1980 // Don't use call_VM as it will see a possible pending exception and forward it
1981 // and never return here preventing us from clearing _last_native_pc down below.
1982 //
1983 save_native_result(masm, ret_type, stack_slots);
1984 __ mv(c_rarg0, xthread);
1985 #ifndef PRODUCT
1986 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
1987 #endif
1988 RuntimeAddress target(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1989 __ relocate(target.rspec(), [&] {
1990 int32_t offset;
1991 __ la_patchable(t0, target, offset);
1992 __ jalr(x1, t0, offset);
1993 });
1994
1995 // Restore any method result value
1996 restore_native_result(masm, ret_type, stack_slots);
1997
1998 __ j(safepoint_in_progress_done);
1999 __ block_comment("} safepoint");
2000 }
2001
2002 // SLOW PATH dtrace support
2003 {
2004 __ block_comment("dtrace entry {");
2005 __ bind(dtrace_method_entry);
2006
2007 // We have all of the arguments setup at this point. We must not touch any register
2008 // argument registers at this point (what if we save/restore them there are no oop?
2009
2010 save_args(masm, total_c_args, c_arg, out_regs);
2011 __ mov_metadata(c_rarg1, method());
2012 __ call_VM_leaf(
2013 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2014 xthread, c_rarg1);
2015 restore_args(masm, total_c_args, c_arg, out_regs);
2016 __ j(dtrace_method_entry_done);
2017 __ block_comment("} dtrace entry");
2018 }
2019
2020 {
2021 __ block_comment("dtrace exit {");
2022 __ bind(dtrace_method_exit);
2023 save_native_result(masm, ret_type, stack_slots);
2024 __ mov_metadata(c_rarg1, method());
2025 __ call_VM_leaf(
2026 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2027 xthread, c_rarg1);
2028 restore_native_result(masm, ret_type, stack_slots);
2029 __ j(dtrace_method_exit_done);
2030 __ block_comment("} dtrace exit");
2031 }
2032
2033 __ flush();
2034
2035 nmethod *nm = nmethod::new_native_nmethod(method,
2036 compile_id,
2037 masm->code(),
2038 vep_offset,
2039 frame_complete,
2040 stack_slots / VMRegImpl::slots_per_word,
2041 (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)),
2042 in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size),
2043 oop_maps);
2044 assert(nm != nullptr, "create native nmethod fail!");
2045 return nm;
2046 }
2047
2048 // this function returns the adjust size (in number of words) to a c2i adapter
2049 // activation for use during deoptimization
2050 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals) {
2051 assert(callee_locals >= callee_parameters,
2052 "test and remove; got more parms than locals");
2053 if (callee_locals < callee_parameters) {
2054 return 0; // No adjustment for negative locals
2055 }
2056 int diff = (callee_locals - callee_parameters) * Interpreter::stackElementWords;
2057 // diff is counted in stack words
2058 return align_up(diff, 2);
2059 }
2060
2061 //------------------------------generate_deopt_blob----------------------------
2062 void SharedRuntime::generate_deopt_blob() {
2063 // Allocate space for the code
2064 ResourceMark rm;
2065 // Setup code generation tools
2066 int pad = 0;
2067 #if INCLUDE_JVMCI
2068 if (EnableJVMCI) {
2069 pad += 512; // Increase the buffer size when compiling for JVMCI
2070 }
2071 #endif
2072 CodeBuffer buffer("deopt_blob", 2048 + pad, 1024);
2073 MacroAssembler* masm = new MacroAssembler(&buffer);
2074 int frame_size_in_words = -1;
2075 OopMap* map = nullptr;
2076 OopMapSet *oop_maps = new OopMapSet();
2077 assert_cond(masm != nullptr && oop_maps != nullptr);
2078 RegisterSaver reg_saver(COMPILER2_OR_JVMCI != 0);
2079
2080 // -------------
2081 // This code enters when returning to a de-optimized nmethod. A return
2082 // address has been pushed on the stack, and return values are in
2083 // registers.
2084 // If we are doing a normal deopt then we were called from the patched
2085 // nmethod from the point we returned to the nmethod. So the return
2086 // address on the stack is wrong by NativeCall::instruction_size
2087 // We will adjust the value so it looks like we have the original return
2088 // address on the stack (like when we eagerly deoptimized).
2089 // In the case of an exception pending when deoptimizing, we enter
2090 // with a return address on the stack that points after the call we patched
2091 // into the exception handler. We have the following register state from,
2092 // e.g., the forward exception stub (see stubGenerator_riscv.cpp).
2093 // x10: exception oop
2094 // x9: exception handler
2095 // x13: throwing pc
2096 // So in this case we simply jam x13 into the useless return address and
2097 // the stack looks just like we want.
2098 //
2099 // At this point we need to de-opt. We save the argument return
2100 // registers. We call the first C routine, fetch_unroll_info(). This
2101 // routine captures the return values and returns a structure which
2102 // describes the current frame size and the sizes of all replacement frames.
2103 // The current frame is compiled code and may contain many inlined
2104 // functions, each with their own JVM state. We pop the current frame, then
2105 // push all the new frames. Then we call the C routine unpack_frames() to
2106 // populate these frames. Finally unpack_frames() returns us the new target
2107 // address. Notice that callee-save registers are BLOWN here; they have
2108 // already been captured in the vframeArray at the time the return PC was
2109 // patched.
2110 address start = __ pc();
2111 Label cont;
2112
2113 // Prolog for non exception case!
2114
2115 // Save everything in sight.
2116 map = reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2117
2118 // Normal deoptimization. Save exec mode for unpack_frames.
2119 __ mv(xcpool, Deoptimization::Unpack_deopt); // callee-saved
2120 __ j(cont);
2121
2122 int reexecute_offset = __ pc() - start;
2123 #if INCLUDE_JVMCI && !defined(COMPILER1)
2124 if (EnableJVMCI && UseJVMCICompiler) {
2125 // JVMCI does not use this kind of deoptimization
2126 __ should_not_reach_here();
2127 }
2128 #endif
2129
2130 // Reexecute case
2131 // return address is the pc describes what bci to do re-execute at
2132
2133 // No need to update map as each call to save_live_registers will produce identical oopmap
2134 (void) reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2135
2136 __ mv(xcpool, Deoptimization::Unpack_reexecute); // callee-saved
2137 __ j(cont);
2138
2139 #if INCLUDE_JVMCI
2140 Label after_fetch_unroll_info_call;
2141 int implicit_exception_uncommon_trap_offset = 0;
2142 int uncommon_trap_offset = 0;
2143
2144 if (EnableJVMCI) {
2145 implicit_exception_uncommon_trap_offset = __ pc() - start;
2146
2147 __ ld(ra, Address(xthread, in_bytes(JavaThread::jvmci_implicit_exception_pc_offset())));
2148 __ sd(zr, Address(xthread, in_bytes(JavaThread::jvmci_implicit_exception_pc_offset())));
2149
2150 uncommon_trap_offset = __ pc() - start;
2151
2152 // Save everything in sight.
2153 reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2154 // fetch_unroll_info needs to call last_java_frame()
2155 Label retaddr;
2156 __ set_last_Java_frame(sp, noreg, retaddr, t0);
2157
2158 __ lw(c_rarg1, Address(xthread, in_bytes(JavaThread::pending_deoptimization_offset())));
2159 __ mv(t0, -1);
2160 __ sw(t0, Address(xthread, in_bytes(JavaThread::pending_deoptimization_offset())));
2161
2162 __ mv(xcpool, Deoptimization::Unpack_reexecute);
2163 __ mv(c_rarg0, xthread);
2164 __ orrw(c_rarg2, zr, xcpool); // exec mode
2165 RuntimeAddress target(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap));
2166 __ relocate(target.rspec(), [&] {
2167 int32_t offset;
2168 __ la_patchable(t0, target, offset);
2169 __ jalr(x1, t0, offset);
2170 });
2171 __ bind(retaddr);
2172 oop_maps->add_gc_map( __ pc()-start, map->deep_copy());
2173
2174 __ reset_last_Java_frame(false);
2175
2176 __ j(after_fetch_unroll_info_call);
2177 } // EnableJVMCI
2178 #endif // INCLUDE_JVMCI
2179
2180 int exception_offset = __ pc() - start;
2181
2182 // Prolog for exception case
2183
2184 // all registers are dead at this entry point, except for x10, and
2185 // x13 which contain the exception oop and exception pc
2186 // respectively. Set them in TLS and fall thru to the
2187 // unpack_with_exception_in_tls entry point.
2188
2189 __ sd(x13, Address(xthread, JavaThread::exception_pc_offset()));
2190 __ sd(x10, Address(xthread, JavaThread::exception_oop_offset()));
2191
2192 int exception_in_tls_offset = __ pc() - start;
2193
2194 // new implementation because exception oop is now passed in JavaThread
2195
2196 // Prolog for exception case
2197 // All registers must be preserved because they might be used by LinearScan
2198 // Exceptiop oop and throwing PC are passed in JavaThread
2199 // tos: stack at point of call to method that threw the exception (i.e. only
2200 // args are on the stack, no return address)
2201
2202 // The return address pushed by save_live_registers will be patched
2203 // later with the throwing pc. The correct value is not available
2204 // now because loading it from memory would destroy registers.
2205
2206 // NB: The SP at this point must be the SP of the method that is
2207 // being deoptimized. Deoptimization assumes that the frame created
2208 // here by save_live_registers is immediately below the method's SP.
2209 // This is a somewhat fragile mechanism.
2210
2211 // Save everything in sight.
2212 map = reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2213
2214 // Now it is safe to overwrite any register
2215
2216 // Deopt during an exception. Save exec mode for unpack_frames.
2217 __ mv(xcpool, Deoptimization::Unpack_exception); // callee-saved
2218
2219 // load throwing pc from JavaThread and patch it as the return address
2220 // of the current frame. Then clear the field in JavaThread
2221
2222 __ ld(x13, Address(xthread, JavaThread::exception_pc_offset()));
2223 __ sd(x13, Address(fp, frame::return_addr_offset * wordSize));
2224 __ sd(zr, Address(xthread, JavaThread::exception_pc_offset()));
2225
2226 #ifdef ASSERT
2227 // verify that there is really an exception oop in JavaThread
2228 __ ld(x10, Address(xthread, JavaThread::exception_oop_offset()));
2229 __ verify_oop(x10);
2230
2231 // verify that there is no pending exception
2232 Label no_pending_exception;
2233 __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
2234 __ beqz(t0, no_pending_exception);
2235 __ stop("must not have pending exception here");
2236 __ bind(no_pending_exception);
2237 #endif
2238
2239 __ bind(cont);
2240
2241 // Call C code. Need thread and this frame, but NOT official VM entry
2242 // crud. We cannot block on this call, no GC can happen.
2243 //
2244 // UnrollBlock* fetch_unroll_info(JavaThread* thread)
2245
2246 // fetch_unroll_info needs to call last_java_frame().
2247
2248 Label retaddr;
2249 __ set_last_Java_frame(sp, noreg, retaddr, t0);
2250 #ifdef ASSERT
2251 {
2252 Label L;
2253 __ ld(t0, Address(xthread,
2254 JavaThread::last_Java_fp_offset()));
2255 __ beqz(t0, L);
2256 __ stop("SharedRuntime::generate_deopt_blob: last_Java_fp not cleared");
2257 __ bind(L);
2258 }
2259 #endif // ASSERT
2260 __ mv(c_rarg0, xthread);
2261 __ mv(c_rarg1, xcpool);
2262 RuntimeAddress target(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info));
2263 __ relocate(target.rspec(), [&] {
2264 int32_t offset;
2265 __ la_patchable(t0, target, offset);
2266 __ jalr(x1, t0, offset);
2267 });
2268 __ bind(retaddr);
2269
2270 // Need to have an oopmap that tells fetch_unroll_info where to
2271 // find any register it might need.
2272 oop_maps->add_gc_map(__ pc() - start, map);
2273
2274 __ reset_last_Java_frame(false);
2275
2276 #if INCLUDE_JVMCI
2277 if (EnableJVMCI) {
2278 __ bind(after_fetch_unroll_info_call);
2279 }
2280 #endif
2281
2282 // Load UnrollBlock* into x15
2283 __ mv(x15, x10);
2284
2285 __ lwu(xcpool, Address(x15, Deoptimization::UnrollBlock::unpack_kind_offset()));
2286 Label noException;
2287 __ mv(t0, Deoptimization::Unpack_exception);
2288 __ bne(xcpool, t0, noException); // Was exception pending?
2289 __ ld(x10, Address(xthread, JavaThread::exception_oop_offset()));
2290 __ ld(x13, Address(xthread, JavaThread::exception_pc_offset()));
2291 __ sd(zr, Address(xthread, JavaThread::exception_oop_offset()));
2292 __ sd(zr, Address(xthread, JavaThread::exception_pc_offset()));
2293
2294 __ verify_oop(x10);
2295
2296 // Overwrite the result registers with the exception results.
2297 __ sd(x10, Address(sp, reg_saver.reg_offset_in_bytes(x10)));
2298
2299 __ bind(noException);
2300
2301 // Only register save data is on the stack.
2302 // Now restore the result registers. Everything else is either dead
2303 // or captured in the vframeArray.
2304
2305 // Restore fp result register
2306 __ fld(f10, Address(sp, reg_saver.freg_offset_in_bytes(f10)));
2307 // Restore integer result register
2308 __ ld(x10, Address(sp, reg_saver.reg_offset_in_bytes(x10)));
2309
2310 // Pop all of the register save area off the stack
2311 __ add(sp, sp, frame_size_in_words * wordSize);
2312
2313 // All of the register save area has been popped of the stack. Only the
2314 // return address remains.
2315
2316 // Pop all the frames we must move/replace.
2317 //
2318 // Frame picture (youngest to oldest)
2319 // 1: self-frame (no frame link)
2320 // 2: deopting frame (no frame link)
2321 // 3: caller of deopting frame (could be compiled/interpreted).
2322 //
2323 // Note: by leaving the return address of self-frame on the stack
2324 // and using the size of frame 2 to adjust the stack
2325 // when we are done the return to frame 3 will still be on the stack.
2326
2327 // Pop deoptimized frame
2328 __ lwu(x12, Address(x15, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset()));
2329 __ sub(x12, x12, 2 * wordSize);
2330 __ add(sp, sp, x12);
2331 __ ld(fp, Address(sp, 0));
2332 __ ld(ra, Address(sp, wordSize));
2333 __ addi(sp, sp, 2 * wordSize);
2334 // RA should now be the return address to the caller (3)
2335
2336 #ifdef ASSERT
2337 // Compilers generate code that bang the stack by as much as the
2338 // interpreter would need. So this stack banging should never
2339 // trigger a fault. Verify that it does not on non product builds.
2340 __ lwu(x9, Address(x15, Deoptimization::UnrollBlock::total_frame_sizes_offset()));
2341 __ bang_stack_size(x9, x12);
2342 #endif
2343 // Load address of array of frame pcs into x12
2344 __ ld(x12, Address(x15, Deoptimization::UnrollBlock::frame_pcs_offset()));
2345
2346 // Load address of array of frame sizes into x14
2347 __ ld(x14, Address(x15, Deoptimization::UnrollBlock::frame_sizes_offset()));
2348
2349 // Load counter into x13
2350 __ lwu(x13, Address(x15, Deoptimization::UnrollBlock::number_of_frames_offset()));
2351
2352 // Now adjust the caller's stack to make up for the extra locals
2353 // but record the original sp so that we can save it in the skeletal interpreter
2354 // frame and the stack walking of interpreter_sender will get the unextended sp
2355 // value and not the "real" sp value.
2356
2357 const Register sender_sp = x16;
2358
2359 __ mv(sender_sp, sp);
2360 __ lwu(x9, Address(x15,
2361 Deoptimization::UnrollBlock::
2362 caller_adjustment_offset()));
2363 __ sub(sp, sp, x9);
2364
2365 // Push interpreter frames in a loop
2366 __ mv(t0, 0xDEADDEAD); // Make a recognizable pattern
2367 __ mv(t1, t0);
2368 Label loop;
2369 __ bind(loop);
2370 __ ld(x9, Address(x14, 0)); // Load frame size
2371 __ addi(x14, x14, wordSize);
2372 __ sub(x9, x9, 2 * wordSize); // We'll push pc and fp by hand
2373 __ ld(ra, Address(x12, 0)); // Load pc
2374 __ addi(x12, x12, wordSize);
2375 __ enter(); // Save old & set new fp
2376 __ sub(sp, sp, x9); // Prolog
2377 // This value is corrected by layout_activation_impl
2378 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
2379 __ sd(sender_sp, Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize)); // Make it walkable
2380 __ mv(sender_sp, sp); // Pass sender_sp to next frame
2381 __ addi(x13, x13, -1); // Decrement counter
2382 __ bnez(x13, loop);
2383
2384 // Re-push self-frame
2385 __ ld(ra, Address(x12));
2386 __ enter();
2387
2388 // Allocate a full sized register save area. We subtract 2 because
2389 // enter() just pushed 2 words
2390 __ sub(sp, sp, (frame_size_in_words - 2) * wordSize);
2391
2392 // Restore frame locals after moving the frame
2393 __ fsd(f10, Address(sp, reg_saver.freg_offset_in_bytes(f10)));
2394 __ sd(x10, Address(sp, reg_saver.reg_offset_in_bytes(x10)));
2395
2396 // Call C code. Need thread but NOT official VM entry
2397 // crud. We cannot block on this call, no GC can happen. Call should
2398 // restore return values to their stack-slots with the new SP.
2399 //
2400 // void Deoptimization::unpack_frames(JavaThread* thread, int exec_mode)
2401
2402 // Use fp because the frames look interpreted now
2403 // Don't need the precise return PC here, just precise enough to point into this code blob.
2404 address the_pc = __ pc();
2405 __ set_last_Java_frame(sp, fp, the_pc, t0);
2406
2407 __ mv(c_rarg0, xthread);
2408 __ mv(c_rarg1, xcpool); // second arg: exec_mode
2409 target = RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames));
2410 __ relocate(target.rspec(), [&] {
2411 int32_t offset;
2412 __ la_patchable(t0, target, offset);
2413 __ jalr(x1, t0, offset);
2414 });
2415
2416 // Set an oopmap for the call site
2417 // Use the same PC we used for the last java frame
2418 oop_maps->add_gc_map(the_pc - start,
2419 new OopMap(frame_size_in_words, 0));
2420
2421 // Clear fp AND pc
2422 __ reset_last_Java_frame(true);
2423
2424 // Collect return values
2425 __ fld(f10, Address(sp, reg_saver.freg_offset_in_bytes(f10)));
2426 __ ld(x10, Address(sp, reg_saver.reg_offset_in_bytes(x10)));
2427
2428 // Pop self-frame.
2429 __ leave(); // Epilog
2430
2431 // Jump to interpreter
2432 __ ret();
2433
2434 // Make sure all code is generated
2435 masm->flush();
2436
2437 _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words);
2438 assert(_deopt_blob != nullptr, "create deoptimization blob fail!");
2439 _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset);
2440 #if INCLUDE_JVMCI
2441 if (EnableJVMCI) {
2442 _deopt_blob->set_uncommon_trap_offset(uncommon_trap_offset);
2443 _deopt_blob->set_implicit_exception_uncommon_trap_offset(implicit_exception_uncommon_trap_offset);
2444 }
2445 #endif
2446 }
2447
2448 // Number of stack slots between incoming argument block and the start of
2449 // a new frame. The PROLOG must add this many slots to the stack. The
2450 // EPILOG must remove this many slots.
2451 // RISCV needs two words for RA (return address) and FP (frame pointer).
2452 uint SharedRuntime::in_preserve_stack_slots() {
2453 return 2 * VMRegImpl::slots_per_word;
2454 }
2455
2456 uint SharedRuntime::out_preserve_stack_slots() {
2457 return 0;
2458 }
2459
2460 #ifdef COMPILER2
2461 //------------------------------generate_uncommon_trap_blob--------------------
2462 void SharedRuntime::generate_uncommon_trap_blob() {
2463 // Allocate space for the code
2464 ResourceMark rm;
2465 // Setup code generation tools
2466 CodeBuffer buffer("uncommon_trap_blob", 2048, 1024);
2467 MacroAssembler* masm = new MacroAssembler(&buffer);
2468 assert_cond(masm != nullptr);
2469
2470 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
2471
2472 address start = __ pc();
2473
2474 // Push self-frame. We get here with a return address in RA
2475 // and sp should be 16 byte aligned
2476 // push fp and retaddr by hand
2477 __ addi(sp, sp, -2 * wordSize);
2478 __ sd(ra, Address(sp, wordSize));
2479 __ sd(fp, Address(sp, 0));
2480 // we don't expect an arg reg save area
2481 #ifndef PRODUCT
2482 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
2483 #endif
2484 // compiler left unloaded_class_index in j_rarg0 move to where the
2485 // runtime expects it.
2486 __ sign_extend(c_rarg1, j_rarg0, 32);
2487
2488 // we need to set the past SP to the stack pointer of the stub frame
2489 // and the pc to the address where this runtime call will return
2490 // although actually any pc in this code blob will do).
2491 Label retaddr;
2492 __ set_last_Java_frame(sp, noreg, retaddr, t0);
2493
2494 // Call C code. Need thread but NOT official VM entry
2495 // crud. We cannot block on this call, no GC can happen. Call should
2496 // capture callee-saved registers as well as return values.
2497 //
2498 // UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index, jint exec_mode)
2499 //
2500 // n.b. 3 gp args, 0 fp args, integral return type
2501
2502 __ mv(c_rarg0, xthread);
2503 __ mv(c_rarg2, Deoptimization::Unpack_uncommon_trap);
2504 RuntimeAddress target(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap));
2505 __ relocate(target.rspec(), [&] {
2506 int32_t offset;
2507 __ la_patchable(t0, target, offset);
2508 __ jalr(x1, t0, offset);
2509 });
2510 __ bind(retaddr);
2511
2512 // Set an oopmap for the call site
2513 OopMapSet* oop_maps = new OopMapSet();
2514 OopMap* map = new OopMap(SimpleRuntimeFrame::framesize, 0);
2515 assert_cond(oop_maps != nullptr && map != nullptr);
2516
2517 // location of fp is known implicitly by the frame sender code
2518
2519 oop_maps->add_gc_map(__ pc() - start, map);
2520
2521 __ reset_last_Java_frame(false);
2522
2523 // move UnrollBlock* into x14
2524 __ mv(x14, x10);
2525
2526 #ifdef ASSERT
2527 { Label L;
2528 __ lwu(t0, Address(x14, Deoptimization::UnrollBlock::unpack_kind_offset()));
2529 __ mv(t1, Deoptimization::Unpack_uncommon_trap);
2530 __ beq(t0, t1, L);
2531 __ stop("SharedRuntime::generate_uncommon_trap_blob: expected Unpack_uncommon_trap");
2532 __ bind(L);
2533 }
2534 #endif
2535
2536 // Pop all the frames we must move/replace.
2537 //
2538 // Frame picture (youngest to oldest)
2539 // 1: self-frame (no frame link)
2540 // 2: deopting frame (no frame link)
2541 // 3: caller of deopting frame (could be compiled/interpreted).
2542
2543 __ add(sp, sp, (SimpleRuntimeFrame::framesize) << LogBytesPerInt); // Epilog!
2544
2545 // Pop deoptimized frame (int)
2546 __ lwu(x12, Address(x14,
2547 Deoptimization::UnrollBlock::
2548 size_of_deoptimized_frame_offset()));
2549 __ sub(x12, x12, 2 * wordSize);
2550 __ add(sp, sp, x12);
2551 __ ld(fp, Address(sp, 0));
2552 __ ld(ra, Address(sp, wordSize));
2553 __ addi(sp, sp, 2 * wordSize);
2554 // RA should now be the return address to the caller (3) frame
2555
2556 #ifdef ASSERT
2557 // Compilers generate code that bang the stack by as much as the
2558 // interpreter would need. So this stack banging should never
2559 // trigger a fault. Verify that it does not on non product builds.
2560 __ lwu(x11, Address(x14,
2561 Deoptimization::UnrollBlock::
2562 total_frame_sizes_offset()));
2563 __ bang_stack_size(x11, x12);
2564 #endif
2565
2566 // Load address of array of frame pcs into x12 (address*)
2567 __ ld(x12, Address(x14,
2568 Deoptimization::UnrollBlock::frame_pcs_offset()));
2569
2570 // Load address of array of frame sizes into x15 (intptr_t*)
2571 __ ld(x15, Address(x14,
2572 Deoptimization::UnrollBlock::
2573 frame_sizes_offset()));
2574
2575 // Counter
2576 __ lwu(x13, Address(x14,
2577 Deoptimization::UnrollBlock::
2578 number_of_frames_offset())); // (int)
2579
2580 // Now adjust the caller's stack to make up for the extra locals but
2581 // record the original sp so that we can save it in the skeletal
2582 // interpreter frame and the stack walking of interpreter_sender
2583 // will get the unextended sp value and not the "real" sp value.
2584
2585 const Register sender_sp = t1; // temporary register
2586
2587 __ lwu(x11, Address(x14,
2588 Deoptimization::UnrollBlock::
2589 caller_adjustment_offset())); // (int)
2590 __ mv(sender_sp, sp);
2591 __ sub(sp, sp, x11);
2592
2593 // Push interpreter frames in a loop
2594 Label loop;
2595 __ bind(loop);
2596 __ ld(x11, Address(x15, 0)); // Load frame size
2597 __ sub(x11, x11, 2 * wordSize); // We'll push pc and fp by hand
2598 __ ld(ra, Address(x12, 0)); // Save return address
2599 __ enter(); // and old fp & set new fp
2600 __ sub(sp, sp, x11); // Prolog
2601 __ sd(sender_sp, Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize)); // Make it walkable
2602 // This value is corrected by layout_activation_impl
2603 __ sd(zr, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
2604 __ mv(sender_sp, sp); // Pass sender_sp to next frame
2605 __ add(x15, x15, wordSize); // Bump array pointer (sizes)
2606 __ add(x12, x12, wordSize); // Bump array pointer (pcs)
2607 __ subw(x13, x13, 1); // Decrement counter
2608 __ bgtz(x13, loop);
2609 __ ld(ra, Address(x12, 0)); // save final return address
2610 // Re-push self-frame
2611 __ enter(); // & old fp & set new fp
2612
2613 // Use fp because the frames look interpreted now
2614 // Save "the_pc" since it cannot easily be retrieved using the last_java_SP after we aligned SP.
2615 // Don't need the precise return PC here, just precise enough to point into this code blob.
2616 address the_pc = __ pc();
2617 __ set_last_Java_frame(sp, fp, the_pc, t0);
2618
2619 // Call C code. Need thread but NOT official VM entry
2620 // crud. We cannot block on this call, no GC can happen. Call should
2621 // restore return values to their stack-slots with the new SP.
2622 //
2623 // BasicType unpack_frames(JavaThread* thread, int exec_mode)
2624 //
2625
2626 // n.b. 2 gp args, 0 fp args, integral return type
2627
2628 // sp should already be aligned
2629 __ mv(c_rarg0, xthread);
2630 __ mv(c_rarg1, Deoptimization::Unpack_uncommon_trap);
2631 target = RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames));
2632 __ relocate(target.rspec(), [&] {
2633 int32_t offset;
2634 __ la_patchable(t0, target, offset);
2635 __ jalr(x1, t0, offset);
2636 });
2637
2638 // Set an oopmap for the call site
2639 // Use the same PC we used for the last java frame
2640 oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
2641
2642 // Clear fp AND pc
2643 __ reset_last_Java_frame(true);
2644
2645 // Pop self-frame.
2646 __ leave(); // Epilog
2647
2648 // Jump to interpreter
2649 __ ret();
2650
2651 // Make sure all code is generated
2652 masm->flush();
2653
2654 _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps,
2655 SimpleRuntimeFrame::framesize >> 1);
2656 }
2657 #endif // COMPILER2
2658
2659 //------------------------------generate_handler_blob------
2660 //
2661 // Generate a special Compile2Runtime blob that saves all registers,
2662 // and setup oopmap.
2663 //
2664 SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, int poll_type) {
2665 ResourceMark rm;
2666 OopMapSet *oop_maps = new OopMapSet();
2667 assert_cond(oop_maps != nullptr);
2668 OopMap* map = nullptr;
2669
2670 // Allocate space for the code. Setup code generation tools.
2671 CodeBuffer buffer("handler_blob", 2048, 1024);
2672 MacroAssembler* masm = new MacroAssembler(&buffer);
2673 assert_cond(masm != nullptr);
2674
2675 address start = __ pc();
2676 address call_pc = nullptr;
2677 int frame_size_in_words = -1;
2678 bool cause_return = (poll_type == POLL_AT_RETURN);
2679 RegisterSaver reg_saver(poll_type == POLL_AT_VECTOR_LOOP /* save_vectors */);
2680
2681 // Save Integer and Float registers.
2682 map = reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2683
2684 // The following is basically a call_VM. However, we need the precise
2685 // address of the call in order to generate an oopmap. Hence, we do all the
2686 // work ourselves.
2687
2688 Label retaddr;
2689 __ set_last_Java_frame(sp, noreg, retaddr, t0);
2690
2691 // The return address must always be correct so that frame constructor never
2692 // sees an invalid pc.
2693
2694 if (!cause_return) {
2695 // overwrite the return address pushed by save_live_registers
2696 // Additionally, x18 is a callee-saved register so we can look at
2697 // it later to determine if someone changed the return address for
2698 // us!
2699 __ ld(x18, Address(xthread, JavaThread::saved_exception_pc_offset()));
2700 __ sd(x18, Address(fp, frame::return_addr_offset * wordSize));
2701 }
2702
2703 // Do the call
2704 __ mv(c_rarg0, xthread);
2705 RuntimeAddress target(call_ptr);
2706 __ relocate(target.rspec(), [&] {
2707 int32_t offset;
2708 __ la_patchable(t0, target, offset);
2709 __ jalr(x1, t0, offset);
2710 });
2711 __ bind(retaddr);
2712
2713 // Set an oopmap for the call site. This oopmap will map all
2714 // oop-registers and debug-info registers as callee-saved. This
2715 // will allow deoptimization at this safepoint to find all possible
2716 // debug-info recordings, as well as let GC find all oops.
2717
2718 oop_maps->add_gc_map( __ pc() - start, map);
2719
2720 Label noException;
2721
2722 __ reset_last_Java_frame(false);
2723
2724 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2725
2726 __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
2727 __ beqz(t0, noException);
2728
2729 // Exception pending
2730
2731 reg_saver.restore_live_registers(masm);
2732
2733 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2734
2735 // No exception case
2736 __ bind(noException);
2737
2738 Label no_adjust, bail;
2739 if (!cause_return) {
2740 // If our stashed return pc was modified by the runtime we avoid touching it
2741 __ ld(t0, Address(fp, frame::return_addr_offset * wordSize));
2742 __ bne(x18, t0, no_adjust);
2743
2744 #ifdef ASSERT
2745 // Verify the correct encoding of the poll we're about to skip.
2746 // See NativeInstruction::is_lwu_to_zr()
2747 __ lwu(t0, Address(x18));
2748 __ andi(t1, t0, 0b0000011);
2749 __ mv(t2, 0b0000011);
2750 __ bne(t1, t2, bail); // 0-6:0b0000011
2751 __ srli(t1, t0, 7);
2752 __ andi(t1, t1, 0b00000);
2753 __ bnez(t1, bail); // 7-11:0b00000
2754 __ srli(t1, t0, 12);
2755 __ andi(t1, t1, 0b110);
2756 __ mv(t2, 0b110);
2757 __ bne(t1, t2, bail); // 12-14:0b110
2758 #endif
2759 // Adjust return pc forward to step over the safepoint poll instruction
2760 __ add(x18, x18, NativeInstruction::instruction_size);
2761 __ sd(x18, Address(fp, frame::return_addr_offset * wordSize));
2762 }
2763
2764 __ bind(no_adjust);
2765 // Normal exit, restore registers and exit.
2766
2767 reg_saver.restore_live_registers(masm);
2768 __ ret();
2769
2770 #ifdef ASSERT
2771 __ bind(bail);
2772 __ stop("Attempting to adjust pc to skip safepoint poll but the return point is not what we expected");
2773 #endif
2774
2775 // Make sure all code is generated
2776 masm->flush();
2777
2778 // Fill-out other meta info
2779 return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words);
2780 }
2781
2782 //
2783 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss
2784 //
2785 // Generate a stub that calls into vm to find out the proper destination
2786 // of a java call. All the argument registers are live at this point
2787 // but since this is generic code we don't know what they are and the caller
2788 // must do any gc of the args.
2789 //
2790 RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) {
2791 assert(StubRoutines::forward_exception_entry() != nullptr, "must be generated before");
2792
2793 // allocate space for the code
2794 ResourceMark rm;
2795
2796 CodeBuffer buffer(name, 1000, 512);
2797 MacroAssembler* masm = new MacroAssembler(&buffer);
2798 assert_cond(masm != nullptr);
2799
2800 int frame_size_in_words = -1;
2801 RegisterSaver reg_saver(false /* save_vectors */);
2802
2803 OopMapSet *oop_maps = new OopMapSet();
2804 assert_cond(oop_maps != nullptr);
2805 OopMap* map = nullptr;
2806
2807 int start = __ offset();
2808
2809 map = reg_saver.save_live_registers(masm, 0, &frame_size_in_words);
2810
2811 int frame_complete = __ offset();
2812
2813 {
2814 Label retaddr;
2815 __ set_last_Java_frame(sp, noreg, retaddr, t0);
2816
2817 __ mv(c_rarg0, xthread);
2818 RuntimeAddress target(destination);
2819 __ relocate(target.rspec(), [&] {
2820 int32_t offset;
2821 __ la_patchable(t0, target, offset);
2822 __ jalr(x1, t0, offset);
2823 });
2824 __ bind(retaddr);
2825 }
2826
2827 // Set an oopmap for the call site.
2828 // We need this not only for callee-saved registers, but also for volatile
2829 // registers that the compiler might be keeping live across a safepoint.
2830
2831 oop_maps->add_gc_map( __ offset() - start, map);
2832
2833 // x10 contains the address we are going to jump to assuming no exception got installed
2834
2835 // clear last_Java_sp
2836 __ reset_last_Java_frame(false);
2837 // check for pending exceptions
2838 Label pending;
2839 __ ld(t0, Address(xthread, Thread::pending_exception_offset()));
2840 __ bnez(t0, pending);
2841
2842 // get the returned Method*
2843 __ get_vm_result_2(xmethod, xthread);
2844 __ sd(xmethod, Address(sp, reg_saver.reg_offset_in_bytes(xmethod)));
2845
2846 // x10 is where we want to jump, overwrite t0 which is saved and temporary
2847 __ sd(x10, Address(sp, reg_saver.reg_offset_in_bytes(t0)));
2848 reg_saver.restore_live_registers(masm);
2849
2850 // We are back to the original state on entry and ready to go.
2851
2852 __ jr(t0);
2853
2854 // Pending exception after the safepoint
2855
2856 __ bind(pending);
2857
2858 reg_saver.restore_live_registers(masm);
2859
2860 // exception pending => remove activation and forward to exception handler
2861
2862 __ sd(zr, Address(xthread, JavaThread::vm_result_offset()));
2863
2864 __ ld(x10, Address(xthread, Thread::pending_exception_offset()));
2865 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2866
2867 // -------------
2868 // make sure all code is generated
2869 masm->flush();
2870
2871 // return the blob
2872 return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true);
2873 }
2874
2875 #ifdef COMPILER2
2876 //------------------------------generate_exception_blob---------------------------
2877 // creates exception blob at the end
2878 // Using exception blob, this code is jumped from a compiled method.
2879 // (see emit_exception_handler in riscv.ad file)
2880 //
2881 // Given an exception pc at a call we call into the runtime for the
2882 // handler in this method. This handler might merely restore state
2883 // (i.e. callee save registers) unwind the frame and jump to the
2884 // exception handler for the nmethod if there is no Java level handler
2885 // for the nmethod.
2886 //
2887 // This code is entered with a jmp.
2888 //
2889 // Arguments:
2890 // x10: exception oop
2891 // x13: exception pc
2892 //
2893 // Results:
2894 // x10: exception oop
2895 // x13: exception pc in caller
2896 // destination: exception handler of caller
2897 //
2898 // Note: the exception pc MUST be at a call (precise debug information)
2899 // Registers x10, x13, x12, x14, x15, t0 are not callee saved.
2900 //
2901
2902 void OptoRuntime::generate_exception_blob() {
2903 assert(!OptoRuntime::is_callee_saved_register(R13_num), "");
2904 assert(!OptoRuntime::is_callee_saved_register(R10_num), "");
2905 assert(!OptoRuntime::is_callee_saved_register(R12_num), "");
2906
2907 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
2908
2909 // Allocate space for the code
2910 ResourceMark rm;
2911 // Setup code generation tools
2912 CodeBuffer buffer("exception_blob", 2048, 1024);
2913 MacroAssembler* masm = new MacroAssembler(&buffer);
2914 assert_cond(masm != nullptr);
2915
2916 // TODO check various assumptions made here
2917 //
2918 // make sure we do so before running this
2919
2920 address start = __ pc();
2921
2922 // push fp and retaddr by hand
2923 // Exception pc is 'return address' for stack walker
2924 __ addi(sp, sp, -2 * wordSize);
2925 __ sd(ra, Address(sp, wordSize));
2926 __ sd(fp, Address(sp));
2927 // there are no callee save registers and we don't expect an
2928 // arg reg save area
2929 #ifndef PRODUCT
2930 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
2931 #endif
2932 // Store exception in Thread object. We cannot pass any arguments to the
2933 // handle_exception call, since we do not want to make any assumption
2934 // about the size of the frame where the exception happened in.
2935 __ sd(x10, Address(xthread, JavaThread::exception_oop_offset()));
2936 __ sd(x13, Address(xthread, JavaThread::exception_pc_offset()));
2937
2938 // This call does all the hard work. It checks if an exception handler
2939 // exists in the method.
2940 // If so, it returns the handler address.
2941 // If not, it prepares for stack-unwinding, restoring the callee-save
2942 // registers of the frame being removed.
2943 //
2944 // address OptoRuntime::handle_exception_C(JavaThread* thread)
2945 //
2946 // n.b. 1 gp arg, 0 fp args, integral return type
2947
2948 // the stack should always be aligned
2949 address the_pc = __ pc();
2950 __ set_last_Java_frame(sp, noreg, the_pc, t0);
2951 __ mv(c_rarg0, xthread);
2952 RuntimeAddress target(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C));
2953 __ relocate(target.rspec(), [&] {
2954 int32_t offset;
2955 __ la_patchable(t0, target, offset);
2956 __ jalr(x1, t0, offset);
2957 });
2958
2959
2960 // handle_exception_C is a special VM call which does not require an explicit
2961 // instruction sync afterwards.
2962
2963 // Set an oopmap for the call site. This oopmap will only be used if we
2964 // are unwinding the stack. Hence, all locations will be dead.
2965 // Callee-saved registers will be the same as the frame above (i.e.,
2966 // handle_exception_stub), since they were restored when we got the
2967 // exception.
2968
2969 OopMapSet* oop_maps = new OopMapSet();
2970 assert_cond(oop_maps != nullptr);
2971
2972 oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
2973
2974 __ reset_last_Java_frame(false);
2975
2976 // Restore callee-saved registers
2977
2978 // fp is an implicitly saved callee saved register (i.e. the calling
2979 // convention will save restore it in prolog/epilog) Other than that
2980 // there are no callee save registers now that adapter frames are gone.
2981 // and we dont' expect an arg reg save area
2982 __ ld(fp, Address(sp));
2983 __ ld(x13, Address(sp, wordSize));
2984 __ addi(sp, sp , 2 * wordSize);
2985
2986 // x10: exception handler
2987
2988 // We have a handler in x10 (could be deopt blob).
2989 __ mv(t0, x10);
2990
2991 // Get the exception oop
2992 __ ld(x10, Address(xthread, JavaThread::exception_oop_offset()));
2993 // Get the exception pc in case we are deoptimized
2994 __ ld(x14, Address(xthread, JavaThread::exception_pc_offset()));
2995 #ifdef ASSERT
2996 __ sd(zr, Address(xthread, JavaThread::exception_handler_pc_offset()));
2997 __ sd(zr, Address(xthread, JavaThread::exception_pc_offset()));
2998 #endif
2999 // Clear the exception oop so GC no longer processes it as a root.
3000 __ sd(zr, Address(xthread, JavaThread::exception_oop_offset()));
3001
3002 // x10: exception oop
3003 // t0: exception handler
3004 // x14: exception pc
3005 // Jump to handler
3006
3007 __ jr(t0);
3008
3009 // Make sure all code is generated
3010 masm->flush();
3011
3012 // Set exception blob
3013 _exception_blob = ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
3014 }
3015 #endif // COMPILER2