1 /*
2 * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2021, Red Hat Inc. All rights reserved.
4 * Copyright (c) 2021, Azul Systems, Inc. All rights reserved.
5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 *
7 * This code is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 only, as
9 * published by the Free Software Foundation.
10 *
11 * This code is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * version 2 for more details (a copy is included in the LICENSE file that
15 * accompanied this code).
16 *
17 * You should have received a copy of the GNU General Public License version
18 * 2 along with this work; if not, write to the Free Software Foundation,
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 *
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
22 * or visit www.oracle.com if you need additional information or have any
23 * questions.
24 *
25 */
26
27 #include "asm/macroAssembler.hpp"
28 #include "asm/macroAssembler.inline.hpp"
29 #include "code/aotCodeCache.hpp"
30 #include "code/codeCache.hpp"
31 #include "code/compiledIC.hpp"
32 #include "code/debugInfoRec.hpp"
33 #include "code/vtableStubs.hpp"
34 #include "compiler/oopMap.hpp"
35 #include "gc/shared/barrierSetAssembler.hpp"
36 #include "interpreter/interpreter.hpp"
37 #include "interpreter/interp_masm.hpp"
38 #include "logging/log.hpp"
39 #include "memory/resourceArea.hpp"
40 #include "nativeInst_aarch64.hpp"
41 #include "oops/klass.inline.hpp"
42 #include "oops/method.inline.hpp"
43 #include "prims/methodHandles.hpp"
44 #include "runtime/continuation.hpp"
45 #include "runtime/continuationEntry.inline.hpp"
46 #include "runtime/globals.hpp"
47 #include "runtime/jniHandles.hpp"
48 #include "runtime/safepointMechanism.hpp"
49 #include "runtime/sharedRuntime.hpp"
50 #include "runtime/signature.hpp"
51 #include "runtime/stubRoutines.hpp"
52 #include "runtime/timerTrace.hpp"
53 #include "runtime/vframeArray.hpp"
54 #include "utilities/align.hpp"
55 #include "utilities/formatBuffer.hpp"
56 #include "vmreg_aarch64.inline.hpp"
57 #ifdef COMPILER1
58 #include "c1/c1_Runtime1.hpp"
59 #endif
60 #ifdef COMPILER2
61 #include "adfiles/ad_aarch64.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 #ifdef PRODUCT
71 #define BLOCK_COMMENT(str) /* nothing */
72 #else
73 #define BLOCK_COMMENT(str) __ block_comment(str)
74 #endif
75
76 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
77
78 // FIXME -- this is used by C1
79 class RegisterSaver {
80 const bool _save_vectors;
81 public:
82 RegisterSaver(bool save_vectors) : _save_vectors(save_vectors) {}
83
84 OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words);
85 void restore_live_registers(MacroAssembler* masm);
86
87 // Offsets into the register save area
88 // Used by deoptimization when it is managing result register
89 // values on its own
90
91 int reg_offset_in_bytes(Register r);
92 int r0_offset_in_bytes() { return reg_offset_in_bytes(r0); }
93 int rscratch1_offset_in_bytes() { return reg_offset_in_bytes(rscratch1); }
94 int v0_offset_in_bytes();
95
96 // Total stack size in bytes for saving sve predicate registers.
97 int total_sve_predicate_in_bytes();
98
99 // Capture info about frame layout
100 // Note this is only correct when not saving full vectors.
101 enum layout {
102 fpu_state_off = 0,
103 fpu_state_end = fpu_state_off + FPUStateSizeInWords - 1,
104 // The frame sender code expects that rfp will be in
105 // the "natural" place and will override any oopMap
106 // setting for it. We must therefore force the layout
107 // so that it agrees with the frame sender code.
108 r0_off = fpu_state_off + FPUStateSizeInWords,
109 rfp_off = r0_off + (Register::number_of_registers - 2) * Register::max_slots_per_register,
110 return_off = rfp_off + Register::max_slots_per_register, // slot for return address
111 reg_save_size = return_off + Register::max_slots_per_register};
112
113 };
114
115 int RegisterSaver::reg_offset_in_bytes(Register r) {
116 // The integer registers are located above the floating point
117 // registers in the stack frame pushed by save_live_registers() so the
118 // offset depends on whether we are saving full vectors, and whether
119 // those vectors are NEON or SVE.
120
121 int slots_per_vect = FloatRegister::save_slots_per_register;
122
123 #if COMPILER2_OR_JVMCI
124 if (_save_vectors) {
125 slots_per_vect = FloatRegister::slots_per_neon_register;
126
127 #ifdef COMPILER2
128 if (Matcher::supports_scalable_vector()) {
129 slots_per_vect = Matcher::scalable_vector_reg_size(T_FLOAT);
130 }
131 #endif
132 }
133 #endif
134
135 int r0_offset = v0_offset_in_bytes() + (slots_per_vect * FloatRegister::number_of_registers) * BytesPerInt;
136 return r0_offset + r->encoding() * wordSize;
137 }
138
139 int RegisterSaver::v0_offset_in_bytes() {
140 // The floating point registers are located above the predicate registers if
141 // they are present in the stack frame pushed by save_live_registers(). So the
142 // offset depends on the saved total predicate vectors in the stack frame.
143 return (total_sve_predicate_in_bytes() / VMRegImpl::stack_slot_size) * BytesPerInt;
144 }
145
146 int RegisterSaver::total_sve_predicate_in_bytes() {
147 #ifdef COMPILER2
148 if (_save_vectors && Matcher::supports_scalable_vector()) {
149 return (Matcher::scalable_vector_reg_size(T_BYTE) >> LogBitsPerByte) *
150 PRegister::number_of_registers;
151 }
152 #endif
153 return 0;
154 }
155
156 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words) {
157 bool use_sve = false;
158 int sve_vector_size_in_bytes = 0;
159 int sve_vector_size_in_slots = 0;
160 int sve_predicate_size_in_slots = 0;
161 int total_predicate_in_bytes = total_sve_predicate_in_bytes();
162 int total_predicate_in_slots = total_predicate_in_bytes / VMRegImpl::stack_slot_size;
163
164 #ifdef COMPILER2
165 use_sve = Matcher::supports_scalable_vector();
166 if (use_sve) {
167 sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
168 sve_vector_size_in_slots = Matcher::scalable_vector_reg_size(T_FLOAT);
169 sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
170 }
171 #endif
172
173 #if COMPILER2_OR_JVMCI
174 if (_save_vectors) {
175 int extra_save_slots_per_register = 0;
176 // Save upper half of vector registers
177 if (use_sve) {
178 extra_save_slots_per_register = sve_vector_size_in_slots - FloatRegister::save_slots_per_register;
179 } else {
180 extra_save_slots_per_register = FloatRegister::extra_save_slots_per_neon_register;
181 }
182 int extra_vector_bytes = extra_save_slots_per_register *
183 VMRegImpl::stack_slot_size *
184 FloatRegister::number_of_registers;
185 additional_frame_words += ((extra_vector_bytes + total_predicate_in_bytes) / wordSize);
186 }
187 #else
188 assert(!_save_vectors, "vectors are generated only by C2 and JVMCI");
189 #endif
190
191 int frame_size_in_bytes = align_up(additional_frame_words * wordSize +
192 reg_save_size * BytesPerInt, 16);
193 // OopMap frame size is in compiler stack slots (jint's) not bytes or words
194 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
195 // The caller will allocate additional_frame_words
196 int additional_frame_slots = additional_frame_words * wordSize / BytesPerInt;
197 // CodeBlob frame size is in words.
198 int frame_size_in_words = frame_size_in_bytes / wordSize;
199 *total_frame_words = frame_size_in_words;
200
201 // Save Integer and Float registers.
202 __ enter();
203 __ push_CPU_state(_save_vectors, use_sve, sve_vector_size_in_bytes, total_predicate_in_bytes);
204
205 // Set an oopmap for the call site. This oopmap will map all
206 // oop-registers and debug-info registers as callee-saved. This
207 // will allow deoptimization at this safepoint to find all possible
208 // debug-info recordings, as well as let GC find all oops.
209
210 OopMapSet *oop_maps = new OopMapSet();
211 OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
212
213 for (int i = 0; i < Register::number_of_registers; i++) {
214 Register r = as_Register(i);
215 if (i <= rfp->encoding() && r != rscratch1 && r != rscratch2) {
216 // SP offsets are in 4-byte words.
217 // Register slots are 8 bytes wide, 32 floating-point registers.
218 int sp_offset = Register::max_slots_per_register * i +
219 FloatRegister::save_slots_per_register * FloatRegister::number_of_registers;
220 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset + additional_frame_slots), r->as_VMReg());
221 }
222 }
223
224 for (int i = 0; i < FloatRegister::number_of_registers; i++) {
225 FloatRegister r = as_FloatRegister(i);
226 int sp_offset = 0;
227 if (_save_vectors) {
228 sp_offset = use_sve ? (total_predicate_in_slots + sve_vector_size_in_slots * i) :
229 (FloatRegister::slots_per_neon_register * i);
230 } else {
231 sp_offset = FloatRegister::save_slots_per_register * i;
232 }
233 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset), r->as_VMReg());
234 }
235
236 return oop_map;
237 }
238
239 void RegisterSaver::restore_live_registers(MacroAssembler* masm) {
240 #ifdef COMPILER2
241 __ pop_CPU_state(_save_vectors, Matcher::supports_scalable_vector(),
242 Matcher::scalable_vector_reg_size(T_BYTE), total_sve_predicate_in_bytes());
243 #else
244 #if !INCLUDE_JVMCI
245 assert(!_save_vectors, "vectors are generated only by C2 and JVMCI");
246 #endif
247 __ pop_CPU_state(_save_vectors);
248 #endif
249 __ ldp(rfp, lr, Address(__ post(sp, 2 * wordSize)));
250 __ authenticate_return_address();
251 }
252
253 // Is vector's size (in bytes) bigger than a size saved by default?
254 // 8 bytes vector registers are saved by default on AArch64.
255 // The SVE supported min vector size is 8 bytes and we need to save
256 // predicate registers when the vector size is 8 bytes as well.
257 bool SharedRuntime::is_wide_vector(int size) {
258 return size > 8 || (UseSVE > 0 && size >= 8);
259 }
260
261 // ---------------------------------------------------------------------------
262 // Read the array of BasicTypes from a signature, and compute where the
263 // arguments should go. Values in the VMRegPair regs array refer to 4-byte
264 // quantities. Values less than VMRegImpl::stack0 are registers, those above
265 // refer to 4-byte stack slots. All stack slots are based off of the stack pointer
266 // as framesizes are fixed.
267 // VMRegImpl::stack0 refers to the first slot 0(sp).
268 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher.
269 // Register up to Register::number_of_registers are the 64-bit
270 // integer registers.
271
272 // Note: the INPUTS in sig_bt are in units of Java argument words,
273 // which are 64-bit. The OUTPUTS are in 32-bit units.
274
275 // The Java calling convention is a "shifted" version of the C ABI.
276 // By skipping the first C ABI register we can call non-static jni
277 // methods with small numbers of arguments without having to shuffle
278 // the arguments at all. Since we control the java ABI we ought to at
279 // least get some advantage out of it.
280
281 int SharedRuntime::java_calling_convention(const BasicType *sig_bt,
282 VMRegPair *regs,
283 int total_args_passed) {
284
285 // Create the mapping between argument positions and
286 // registers.
287 static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = {
288 j_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4, j_rarg5, j_rarg6, j_rarg7
289 };
290 static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters_j] = {
291 j_farg0, j_farg1, j_farg2, j_farg3,
292 j_farg4, j_farg5, j_farg6, j_farg7
293 };
294
295
296 uint int_args = 0;
297 uint fp_args = 0;
298 uint stk_args = 0;
299
300 for (int i = 0; i < total_args_passed; i++) {
301 switch (sig_bt[i]) {
302 case T_BOOLEAN:
303 case T_CHAR:
304 case T_BYTE:
305 case T_SHORT:
306 case T_INT:
307 if (int_args < Argument::n_int_register_parameters_j) {
308 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
309 } else {
310 stk_args = align_up(stk_args, 2);
311 regs[i].set1(VMRegImpl::stack2reg(stk_args));
312 stk_args += 1;
313 }
314 break;
315 case T_VOID:
316 // halves of T_LONG or T_DOUBLE
317 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
318 regs[i].set_bad();
319 break;
320 case T_LONG:
321 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
322 // fall through
323 case T_OBJECT:
324 case T_ARRAY:
325 case T_ADDRESS:
326 if (int_args < Argument::n_int_register_parameters_j) {
327 regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
328 } else {
329 stk_args = align_up(stk_args, 2);
330 regs[i].set2(VMRegImpl::stack2reg(stk_args));
331 stk_args += 2;
332 }
333 break;
334 case T_FLOAT:
335 if (fp_args < Argument::n_float_register_parameters_j) {
336 regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
337 } else {
338 stk_args = align_up(stk_args, 2);
339 regs[i].set1(VMRegImpl::stack2reg(stk_args));
340 stk_args += 1;
341 }
342 break;
343 case T_DOUBLE:
344 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
345 if (fp_args < Argument::n_float_register_parameters_j) {
346 regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
347 } else {
348 stk_args = align_up(stk_args, 2);
349 regs[i].set2(VMRegImpl::stack2reg(stk_args));
350 stk_args += 2;
351 }
352 break;
353 default:
354 ShouldNotReachHere();
355 break;
356 }
357 }
358
359 return stk_args;
360 }
361
362 // Patch the callers callsite with entry to compiled code if it exists.
363 static void patch_callers_callsite(MacroAssembler *masm) {
364 Label L;
365 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::code_offset())));
366 __ cbz(rscratch1, L);
367
368 __ enter();
369 __ push_CPU_state();
370
371 // VM needs caller's callsite
372 // VM needs target method
373 // This needs to be a long call since we will relocate this adapter to
374 // the codeBuffer and it may not reach
375
376 #ifndef PRODUCT
377 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
378 #endif
379
380 __ mov(c_rarg0, rmethod);
381 __ mov(c_rarg1, lr);
382 __ authenticate_return_address(c_rarg1);
383 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)));
384 __ blr(rscratch1);
385
386 // Explicit isb required because fixup_callers_callsite may change the code
387 // stream.
388 __ safepoint_isb();
389
390 __ pop_CPU_state();
391 // restore sp
392 __ leave();
393 __ bind(L);
394 }
395
396 static void gen_c2i_adapter(MacroAssembler *masm,
397 int total_args_passed,
398 int comp_args_on_stack,
399 const BasicType *sig_bt,
400 const VMRegPair *regs,
401 Label& skip_fixup) {
402 // Before we get into the guts of the C2I adapter, see if we should be here
403 // at all. We've come from compiled code and are attempting to jump to the
404 // interpreter, which means the caller made a static call to get here
405 // (vcalls always get a compiled target if there is one). Check for a
406 // compiled target. If there is one, we need to patch the caller's call.
407 patch_callers_callsite(masm);
408
409 __ bind(skip_fixup);
410
411 int words_pushed = 0;
412
413 // Since all args are passed on the stack, total_args_passed *
414 // Interpreter::stackElementSize is the space we need.
415
416 int extraspace = total_args_passed * Interpreter::stackElementSize;
417
418 __ mov(r19_sender_sp, sp);
419
420 // stack is aligned, keep it that way
421 extraspace = align_up(extraspace, 2*wordSize);
422
423 if (extraspace)
424 __ sub(sp, sp, extraspace);
425
426 // Now write the args into the outgoing interpreter space
427 for (int i = 0; i < total_args_passed; i++) {
428 if (sig_bt[i] == T_VOID) {
429 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
430 continue;
431 }
432
433 // offset to start parameters
434 int st_off = (total_args_passed - i - 1) * Interpreter::stackElementSize;
435 int next_off = st_off - Interpreter::stackElementSize;
436
437 // Say 4 args:
438 // i st_off
439 // 0 32 T_LONG
440 // 1 24 T_VOID
441 // 2 16 T_OBJECT
442 // 3 8 T_BOOL
443 // - 0 return address
444 //
445 // However to make thing extra confusing. Because we can fit a Java long/double in
446 // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
447 // leaves one slot empty and only stores to a single slot. In this case the
448 // slot that is occupied is the T_VOID slot. See I said it was confusing.
449
450 VMReg r_1 = regs[i].first();
451 VMReg r_2 = regs[i].second();
452 if (!r_1->is_valid()) {
453 assert(!r_2->is_valid(), "");
454 continue;
455 }
456 if (r_1->is_stack()) {
457 // memory to memory use rscratch1
458 int ld_off = (r_1->reg2stack() * VMRegImpl::stack_slot_size
459 + extraspace
460 + words_pushed * wordSize);
461 if (!r_2->is_valid()) {
462 // sign extend??
463 __ ldrw(rscratch1, Address(sp, ld_off));
464 __ str(rscratch1, Address(sp, st_off));
465
466 } else {
467
468 __ ldr(rscratch1, Address(sp, ld_off));
469
470 // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
471 // T_DOUBLE and T_LONG use two slots in the interpreter
472 if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
473 // ld_off == LSW, ld_off+wordSize == MSW
474 // st_off == MSW, next_off == LSW
475 __ str(rscratch1, Address(sp, next_off));
476 #ifdef ASSERT
477 // Overwrite the unused slot with known junk
478 __ mov(rscratch1, (uint64_t)0xdeadffffdeadaaaaull);
479 __ str(rscratch1, Address(sp, st_off));
480 #endif /* ASSERT */
481 } else {
482 __ str(rscratch1, Address(sp, st_off));
483 }
484 }
485 } else if (r_1->is_Register()) {
486 Register r = r_1->as_Register();
487 if (!r_2->is_valid()) {
488 // must be only an int (or less ) so move only 32bits to slot
489 // why not sign extend??
490 __ str(r, Address(sp, st_off));
491 } else {
492 // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
493 // T_DOUBLE and T_LONG use two slots in the interpreter
494 if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
495 // jlong/double in gpr
496 #ifdef ASSERT
497 // Overwrite the unused slot with known junk
498 __ mov(rscratch1, (uint64_t)0xdeadffffdeadaaabull);
499 __ str(rscratch1, Address(sp, st_off));
500 #endif /* ASSERT */
501 __ str(r, Address(sp, next_off));
502 } else {
503 __ str(r, Address(sp, st_off));
504 }
505 }
506 } else {
507 assert(r_1->is_FloatRegister(), "");
508 if (!r_2->is_valid()) {
509 // only a float use just part of the slot
510 __ strs(r_1->as_FloatRegister(), Address(sp, st_off));
511 } else {
512 #ifdef ASSERT
513 // Overwrite the unused slot with known junk
514 __ mov(rscratch1, (uint64_t)0xdeadffffdeadaaacull);
515 __ str(rscratch1, Address(sp, st_off));
516 #endif /* ASSERT */
517 __ strd(r_1->as_FloatRegister(), Address(sp, next_off));
518 }
519 }
520 }
521
522 __ mov(esp, sp); // Interp expects args on caller's expression stack
523
524 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::interpreter_entry_offset())));
525 __ br(rscratch1);
526 }
527
528
529 void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,
530 int total_args_passed,
531 int comp_args_on_stack,
532 const BasicType *sig_bt,
533 const VMRegPair *regs) {
534
535 // Note: r19_sender_sp contains the senderSP on entry. We must
536 // preserve it since we may do a i2c -> c2i transition if we lose a
537 // race where compiled code goes non-entrant while we get args
538 // ready.
539
540 // Adapters are frameless.
541
542 // An i2c adapter is frameless because the *caller* frame, which is
543 // interpreted, routinely repairs its own esp (from
544 // interpreter_frame_last_sp), even if a callee has modified the
545 // stack pointer. It also recalculates and aligns sp.
546
547 // A c2i adapter is frameless because the *callee* frame, which is
548 // interpreted, routinely repairs its caller's sp (from sender_sp,
549 // which is set up via the senderSP register).
550
551 // In other words, if *either* the caller or callee is interpreted, we can
552 // get the stack pointer repaired after a call.
553
554 // This is why c2i and i2c adapters cannot be indefinitely composed.
555 // In particular, if a c2i adapter were to somehow call an i2c adapter,
556 // both caller and callee would be compiled methods, and neither would
557 // clean up the stack pointer changes performed by the two adapters.
558 // If this happens, control eventually transfers back to the compiled
559 // caller, but with an uncorrected stack, causing delayed havoc.
560
561 // Cut-out for having no stack args.
562 int comp_words_on_stack = align_up(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord;
563 if (comp_args_on_stack) {
564 __ sub(rscratch1, sp, comp_words_on_stack * wordSize);
565 __ andr(sp, rscratch1, -16);
566 }
567
568 // Will jump to the compiled code just as if compiled code was doing it.
569 // Pre-load the register-jump target early, to schedule it better.
570 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::from_compiled_offset())));
571
572 #if INCLUDE_JVMCI
573 if (EnableJVMCI) {
574 // check if this call should be routed towards a specific entry point
575 __ ldr(rscratch2, Address(rthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
576 Label no_alternative_target;
577 __ cbz(rscratch2, no_alternative_target);
578 __ mov(rscratch1, rscratch2);
579 __ str(zr, Address(rthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
580 __ bind(no_alternative_target);
581 }
582 #endif // INCLUDE_JVMCI
583
584 // Now generate the shuffle code.
585 for (int i = 0; i < total_args_passed; i++) {
586 if (sig_bt[i] == T_VOID) {
587 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
588 continue;
589 }
590
591 // Pick up 0, 1 or 2 words from SP+offset.
592
593 assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
594 "scrambled load targets?");
595 // Load in argument order going down.
596 int ld_off = (total_args_passed - i - 1)*Interpreter::stackElementSize;
597 // Point to interpreter value (vs. tag)
598 int next_off = ld_off - Interpreter::stackElementSize;
599 //
600 //
601 //
602 VMReg r_1 = regs[i].first();
603 VMReg r_2 = regs[i].second();
604 if (!r_1->is_valid()) {
605 assert(!r_2->is_valid(), "");
606 continue;
607 }
608 if (r_1->is_stack()) {
609 // Convert stack slot to an SP offset (+ wordSize to account for return address )
610 int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size;
611 if (!r_2->is_valid()) {
612 // sign extend???
613 __ ldrsw(rscratch2, Address(esp, ld_off));
614 __ str(rscratch2, Address(sp, st_off));
615 } else {
616 //
617 // We are using two optoregs. This can be either T_OBJECT,
618 // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
619 // two slots but only uses one for thr T_LONG or T_DOUBLE case
620 // So we must adjust where to pick up the data to match the
621 // interpreter.
622 //
623 // Interpreter local[n] == MSW, local[n+1] == LSW however locals
624 // are accessed as negative so LSW is at LOW address
625
626 // ld_off is MSW so get LSW
627 const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
628 next_off : ld_off;
629 __ ldr(rscratch2, Address(esp, offset));
630 // st_off is LSW (i.e. reg.first())
631 __ str(rscratch2, Address(sp, st_off));
632 }
633 } else if (r_1->is_Register()) { // Register argument
634 Register r = r_1->as_Register();
635 if (r_2->is_valid()) {
636 //
637 // We are using two VMRegs. This can be either T_OBJECT,
638 // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
639 // two slots but only uses one for thr T_LONG or T_DOUBLE case
640 // So we must adjust where to pick up the data to match the
641 // interpreter.
642
643 const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
644 next_off : ld_off;
645
646 // this can be a misaligned move
647 __ ldr(r, Address(esp, offset));
648 } else {
649 // sign extend and use a full word?
650 __ ldrw(r, Address(esp, ld_off));
651 }
652 } else {
653 if (!r_2->is_valid()) {
654 __ ldrs(r_1->as_FloatRegister(), Address(esp, ld_off));
655 } else {
656 __ ldrd(r_1->as_FloatRegister(), Address(esp, next_off));
657 }
658 }
659 }
660
661 __ mov(rscratch2, rscratch1);
662 __ push_cont_fastpath(rthread); // Set JavaThread::_cont_fastpath to the sp of the oldest interpreted frame we know about; kills rscratch1
663 __ mov(rscratch1, rscratch2);
664
665 // 6243940 We might end up in handle_wrong_method if
666 // the callee is deoptimized as we race thru here. If that
667 // happens we don't want to take a safepoint because the
668 // caller frame will look interpreted and arguments are now
669 // "compiled" so it is much better to make this transition
670 // invisible to the stack walking code. Unfortunately if
671 // we try and find the callee by normal means a safepoint
672 // is possible. So we stash the desired callee in the thread
673 // and the vm will find there should this case occur.
674
675 __ str(rmethod, Address(rthread, JavaThread::callee_target_offset()));
676
677 __ br(rscratch1);
678 }
679
680 // ---------------------------------------------------------------
681 void SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
682 int total_args_passed,
683 int comp_args_on_stack,
684 const BasicType *sig_bt,
685 const VMRegPair *regs,
686 address entry_address[AdapterBlob::ENTRY_COUNT]) {
687 entry_address[AdapterBlob::I2C] = __ pc();
688
689 gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
690
691 entry_address[AdapterBlob::C2I_Unverified] = __ pc();
692 Label skip_fixup;
693
694 Register data = rscratch2;
695 Register receiver = j_rarg0;
696 Register tmp = r10; // A call-clobbered register not used for arg passing
697
698 // -------------------------------------------------------------------------
699 // Generate a C2I adapter. On entry we know rmethod holds the Method* during calls
700 // to the interpreter. The args start out packed in the compiled layout. They
701 // need to be unpacked into the interpreter layout. This will almost always
702 // require some stack space. We grow the current (compiled) stack, then repack
703 // the args. We finally end in a jump to the generic interpreter entry point.
704 // On exit from the interpreter, the interpreter will restore our SP (lest the
705 // compiled code, which relies solely on SP and not FP, get sick).
706
707 {
708 __ block_comment("c2i_unverified_entry {");
709 // Method might have been compiled since the call site was patched to
710 // interpreted; if that is the case treat it as a miss so we can get
711 // the call site corrected.
712 __ ic_check(1 /* end_alignment */);
713 __ ldr(rmethod, Address(data, CompiledICData::speculated_method_offset()));
714
715 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::code_offset())));
716 __ cbz(rscratch1, skip_fixup);
717 __ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
718 __ block_comment("} c2i_unverified_entry");
719 }
720
721 entry_address[AdapterBlob::C2I] = __ pc();
722
723 // Class initialization barrier for static methods
724 entry_address[AdapterBlob::C2I_No_Clinit_Check] = nullptr;
725 if (VM_Version::supports_fast_class_init_checks()) {
726 Label L_skip_barrier;
727
728 { // Bypass the barrier for non-static methods
729 __ ldrh(rscratch1, Address(rmethod, Method::access_flags_offset()));
730 __ andsw(zr, rscratch1, JVM_ACC_STATIC);
731 __ br(Assembler::EQ, L_skip_barrier); // non-static
732 }
733
734 __ load_method_holder(rscratch2, rmethod);
735 __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
736 __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
737
738 __ bind(L_skip_barrier);
739 entry_address[AdapterBlob::C2I_No_Clinit_Check] = __ pc();
740 }
741
742 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
743 bs->c2i_entry_barrier(masm);
744
745 gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
746 return;
747 }
748
749 static int c_calling_convention_priv(const BasicType *sig_bt,
750 VMRegPair *regs,
751 int total_args_passed) {
752
753 // We return the amount of VMRegImpl stack slots we need to reserve for all
754 // the arguments NOT counting out_preserve_stack_slots.
755
756 static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
757 c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, c_rarg5, c_rarg6, c_rarg7
758 };
759 static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
760 c_farg0, c_farg1, c_farg2, c_farg3,
761 c_farg4, c_farg5, c_farg6, c_farg7
762 };
763
764 uint int_args = 0;
765 uint fp_args = 0;
766 uint stk_args = 0; // inc by 2 each time
767
768 for (int i = 0; i < total_args_passed; i++) {
769 switch (sig_bt[i]) {
770 case T_BOOLEAN:
771 case T_CHAR:
772 case T_BYTE:
773 case T_SHORT:
774 case T_INT:
775 if (int_args < Argument::n_int_register_parameters_c) {
776 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
777 } else {
778 #ifdef __APPLE__
779 // Less-than word types are stored one after another.
780 // The code is unable to handle this so bailout.
781 return -1;
782 #endif
783 regs[i].set1(VMRegImpl::stack2reg(stk_args));
784 stk_args += 2;
785 }
786 break;
787 case T_LONG:
788 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
789 // fall through
790 case T_OBJECT:
791 case T_ARRAY:
792 case T_ADDRESS:
793 case T_METADATA:
794 if (int_args < Argument::n_int_register_parameters_c) {
795 regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
796 } else {
797 regs[i].set2(VMRegImpl::stack2reg(stk_args));
798 stk_args += 2;
799 }
800 break;
801 case T_FLOAT:
802 if (fp_args < Argument::n_float_register_parameters_c) {
803 regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
804 } else {
805 #ifdef __APPLE__
806 // Less-than word types are stored one after another.
807 // The code is unable to handle this so bailout.
808 return -1;
809 #endif
810 regs[i].set1(VMRegImpl::stack2reg(stk_args));
811 stk_args += 2;
812 }
813 break;
814 case T_DOUBLE:
815 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
816 if (fp_args < Argument::n_float_register_parameters_c) {
817 regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
818 } else {
819 regs[i].set2(VMRegImpl::stack2reg(stk_args));
820 stk_args += 2;
821 }
822 break;
823 case T_VOID: // Halves of longs and doubles
824 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
825 regs[i].set_bad();
826 break;
827 default:
828 ShouldNotReachHere();
829 break;
830 }
831 }
832
833 return stk_args;
834 }
835
836 int SharedRuntime::vector_calling_convention(VMRegPair *regs,
837 uint num_bits,
838 uint total_args_passed) {
839 // More than 8 argument inputs are not supported now.
840 assert(total_args_passed <= Argument::n_float_register_parameters_c, "unsupported");
841 assert(num_bits >= 64 && num_bits <= 2048 && is_power_of_2(num_bits), "unsupported");
842
843 static const FloatRegister VEC_ArgReg[Argument::n_float_register_parameters_c] = {
844 v0, v1, v2, v3, v4, v5, v6, v7
845 };
846
847 // On SVE, we use the same vector registers with 128-bit vector registers on NEON.
848 int next_reg_val = num_bits == 64 ? 1 : 3;
849 for (uint i = 0; i < total_args_passed; i++) {
850 VMReg vmreg = VEC_ArgReg[i]->as_VMReg();
851 regs[i].set_pair(vmreg->next(next_reg_val), vmreg);
852 }
853 return 0;
854 }
855
856 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
857 VMRegPair *regs,
858 int total_args_passed)
859 {
860 int result = c_calling_convention_priv(sig_bt, regs, total_args_passed);
861 guarantee(result >= 0, "Unsupported arguments configuration");
862 return result;
863 }
864
865
866 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
867 // We always ignore the frame_slots arg and just use the space just below frame pointer
868 // which by this time is free to use
869 switch (ret_type) {
870 case T_FLOAT:
871 __ strs(v0, Address(rfp, -wordSize));
872 break;
873 case T_DOUBLE:
874 __ strd(v0, Address(rfp, -wordSize));
875 break;
876 case T_VOID: break;
877 default: {
878 __ str(r0, Address(rfp, -wordSize));
879 }
880 }
881 }
882
883 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
884 // We always ignore the frame_slots arg and just use the space just below frame pointer
885 // which by this time is free to use
886 switch (ret_type) {
887 case T_FLOAT:
888 __ ldrs(v0, Address(rfp, -wordSize));
889 break;
890 case T_DOUBLE:
891 __ ldrd(v0, Address(rfp, -wordSize));
892 break;
893 case T_VOID: break;
894 default: {
895 __ ldr(r0, Address(rfp, -wordSize));
896 }
897 }
898 }
899 static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
900 RegSet x;
901 for ( int i = first_arg ; i < arg_count ; i++ ) {
902 if (args[i].first()->is_Register()) {
903 x = x + args[i].first()->as_Register();
904 } else if (args[i].first()->is_FloatRegister()) {
905 __ strd(args[i].first()->as_FloatRegister(), Address(__ pre(sp, -2 * wordSize)));
906 }
907 }
908 __ push(x, sp);
909 }
910
911 static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
912 RegSet x;
913 for ( int i = first_arg ; i < arg_count ; i++ ) {
914 if (args[i].first()->is_Register()) {
915 x = x + args[i].first()->as_Register();
916 } else {
917 ;
918 }
919 }
920 __ pop(x, sp);
921 for ( int i = arg_count - 1 ; i >= first_arg ; i-- ) {
922 if (args[i].first()->is_Register()) {
923 ;
924 } else if (args[i].first()->is_FloatRegister()) {
925 __ ldrd(args[i].first()->as_FloatRegister(), Address(__ post(sp, 2 * wordSize)));
926 }
927 }
928 }
929
930 static void verify_oop_args(MacroAssembler* masm,
931 const methodHandle& method,
932 const BasicType* sig_bt,
933 const VMRegPair* regs) {
934 Register temp_reg = r19; // not part of any compiled calling seq
935 if (VerifyOops) {
936 for (int i = 0; i < method->size_of_parameters(); i++) {
937 if (sig_bt[i] == T_OBJECT ||
938 sig_bt[i] == T_ARRAY) {
939 VMReg r = regs[i].first();
940 assert(r->is_valid(), "bad oop arg");
941 if (r->is_stack()) {
942 __ ldr(temp_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
943 __ verify_oop(temp_reg);
944 } else {
945 __ verify_oop(r->as_Register());
946 }
947 }
948 }
949 }
950 }
951
952 // on exit, sp points to the ContinuationEntry
953 static OopMap* continuation_enter_setup(MacroAssembler* masm, int& stack_slots) {
954 assert(ContinuationEntry::size() % VMRegImpl::stack_slot_size == 0, "");
955 assert(in_bytes(ContinuationEntry::cont_offset()) % VMRegImpl::stack_slot_size == 0, "");
956 assert(in_bytes(ContinuationEntry::chunk_offset()) % VMRegImpl::stack_slot_size == 0, "");
957
958 stack_slots += (int)ContinuationEntry::size()/wordSize;
959 __ sub(sp, sp, (int)ContinuationEntry::size()); // place Continuation metadata
960
961 OopMap* map = new OopMap(((int)ContinuationEntry::size() + wordSize)/ VMRegImpl::stack_slot_size, 0 /* arg_slots*/);
962
963 __ ldr(rscratch1, Address(rthread, JavaThread::cont_entry_offset()));
964 __ str(rscratch1, Address(sp, ContinuationEntry::parent_offset()));
965 __ mov(rscratch1, sp); // we can't use sp as the source in str
966 __ str(rscratch1, Address(rthread, JavaThread::cont_entry_offset()));
967
968 return map;
969 }
970
971 // on entry c_rarg1 points to the continuation
972 // sp points to ContinuationEntry
973 // c_rarg3 -- isVirtualThread
974 static void fill_continuation_entry(MacroAssembler* masm) {
975 #ifdef ASSERT
976 __ movw(rscratch1, ContinuationEntry::cookie_value());
977 __ strw(rscratch1, Address(sp, ContinuationEntry::cookie_offset()));
978 #endif
979
980 __ str (c_rarg1, Address(sp, ContinuationEntry::cont_offset()));
981 __ strw(c_rarg3, Address(sp, ContinuationEntry::flags_offset()));
982 __ str (zr, Address(sp, ContinuationEntry::chunk_offset()));
983 __ strw(zr, Address(sp, ContinuationEntry::argsize_offset()));
984 __ strw(zr, Address(sp, ContinuationEntry::pin_count_offset()));
985
986 __ ldr(rscratch1, Address(rthread, JavaThread::cont_fastpath_offset()));
987 __ str(rscratch1, Address(sp, ContinuationEntry::parent_cont_fastpath_offset()));
988 __ ldr(rscratch1, Address(rthread, JavaThread::held_monitor_count_offset()));
989 __ str(rscratch1, Address(sp, ContinuationEntry::parent_held_monitor_count_offset()));
990
991 __ str(zr, Address(rthread, JavaThread::cont_fastpath_offset()));
992 __ str(zr, Address(rthread, JavaThread::held_monitor_count_offset()));
993 }
994
995 // on entry, sp points to the ContinuationEntry
996 // on exit, rfp points to the spilled rfp in the entry frame
997 static void continuation_enter_cleanup(MacroAssembler* masm) {
998 #ifndef PRODUCT
999 Label OK;
1000 __ ldr(rscratch1, Address(rthread, JavaThread::cont_entry_offset()));
1001 __ cmp(sp, rscratch1);
1002 __ br(Assembler::EQ, OK);
1003 __ stop("incorrect sp1");
1004 __ bind(OK);
1005 #endif
1006 __ ldr(rscratch1, Address(sp, ContinuationEntry::parent_cont_fastpath_offset()));
1007 __ str(rscratch1, Address(rthread, JavaThread::cont_fastpath_offset()));
1008
1009 if (CheckJNICalls) {
1010 // Check if this is a virtual thread continuation
1011 Label L_skip_vthread_code;
1012 __ ldrw(rscratch1, Address(sp, ContinuationEntry::flags_offset()));
1013 __ cbzw(rscratch1, L_skip_vthread_code);
1014
1015 // If the held monitor count is > 0 and this vthread is terminating then
1016 // it failed to release a JNI monitor. So we issue the same log message
1017 // that JavaThread::exit does.
1018 __ ldr(rscratch1, Address(rthread, JavaThread::jni_monitor_count_offset()));
1019 __ cbz(rscratch1, L_skip_vthread_code);
1020
1021 // Save return value potentially containing the exception oop in callee-saved R19.
1022 __ mov(r19, r0);
1023 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::log_jni_monitor_still_held));
1024 // Restore potential return value.
1025 __ mov(r0, r19);
1026
1027 // For vthreads we have to explicitly zero the JNI monitor count of the carrier
1028 // on termination. The held count is implicitly zeroed below when we restore from
1029 // the parent held count (which has to be zero).
1030 __ str(zr, Address(rthread, JavaThread::jni_monitor_count_offset()));
1031
1032 __ bind(L_skip_vthread_code);
1033 }
1034 #ifdef ASSERT
1035 else {
1036 // Check if this is a virtual thread continuation
1037 Label L_skip_vthread_code;
1038 __ ldrw(rscratch1, Address(sp, ContinuationEntry::flags_offset()));
1039 __ cbzw(rscratch1, L_skip_vthread_code);
1040
1041 // See comment just above. If not checking JNI calls the JNI count is only
1042 // needed for assertion checking.
1043 __ str(zr, Address(rthread, JavaThread::jni_monitor_count_offset()));
1044
1045 __ bind(L_skip_vthread_code);
1046 }
1047 #endif
1048
1049 __ ldr(rscratch1, Address(sp, ContinuationEntry::parent_held_monitor_count_offset()));
1050 __ str(rscratch1, Address(rthread, JavaThread::held_monitor_count_offset()));
1051
1052 __ ldr(rscratch2, Address(sp, ContinuationEntry::parent_offset()));
1053 __ str(rscratch2, Address(rthread, JavaThread::cont_entry_offset()));
1054 __ add(rfp, sp, (int)ContinuationEntry::size());
1055 }
1056
1057 // enterSpecial(Continuation c, boolean isContinue, boolean isVirtualThread)
1058 // On entry: c_rarg1 -- the continuation object
1059 // c_rarg2 -- isContinue
1060 // c_rarg3 -- isVirtualThread
1061 static void gen_continuation_enter(MacroAssembler* masm,
1062 const methodHandle& method,
1063 const BasicType* sig_bt,
1064 const VMRegPair* regs,
1065 int& exception_offset,
1066 OopMapSet*oop_maps,
1067 int& frame_complete,
1068 int& stack_slots,
1069 int& interpreted_entry_offset,
1070 int& compiled_entry_offset) {
1071 //verify_oop_args(masm, method, sig_bt, regs);
1072 Address resolve(SharedRuntime::get_resolve_static_call_stub(), relocInfo::static_call_type);
1073
1074 address start = __ pc();
1075
1076 Label call_thaw, exit;
1077
1078 // i2i entry used at interp_only_mode only
1079 interpreted_entry_offset = __ pc() - start;
1080 {
1081
1082 #ifdef ASSERT
1083 Label is_interp_only;
1084 __ ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
1085 __ cbnzw(rscratch1, is_interp_only);
1086 __ stop("enterSpecial interpreter entry called when not in interp_only_mode");
1087 __ bind(is_interp_only);
1088 #endif
1089
1090 // Read interpreter arguments into registers (this is an ad-hoc i2c adapter)
1091 __ ldr(c_rarg1, Address(esp, Interpreter::stackElementSize*2));
1092 __ ldr(c_rarg2, Address(esp, Interpreter::stackElementSize*1));
1093 __ ldr(c_rarg3, Address(esp, Interpreter::stackElementSize*0));
1094 __ push_cont_fastpath(rthread);
1095
1096 __ enter();
1097 stack_slots = 2; // will be adjusted in setup
1098 OopMap* map = continuation_enter_setup(masm, stack_slots);
1099 // The frame is complete here, but we only record it for the compiled entry, so the frame would appear unsafe,
1100 // but that's okay because at the very worst we'll miss an async sample, but we're in interp_only_mode anyway.
1101
1102 fill_continuation_entry(masm);
1103
1104 __ cbnz(c_rarg2, call_thaw);
1105
1106 const address tr_call = __ trampoline_call(resolve);
1107 if (tr_call == nullptr) {
1108 fatal("CodeCache is full at gen_continuation_enter");
1109 }
1110
1111 oop_maps->add_gc_map(__ pc() - start, map);
1112 __ post_call_nop();
1113
1114 __ b(exit);
1115
1116 address stub = CompiledDirectCall::emit_to_interp_stub(masm, tr_call);
1117 if (stub == nullptr) {
1118 fatal("CodeCache is full at gen_continuation_enter");
1119 }
1120 }
1121
1122 // compiled entry
1123 __ align(CodeEntryAlignment);
1124 compiled_entry_offset = __ pc() - start;
1125
1126 __ enter();
1127 stack_slots = 2; // will be adjusted in setup
1128 OopMap* map = continuation_enter_setup(masm, stack_slots);
1129 frame_complete = __ pc() - start;
1130
1131 fill_continuation_entry(masm);
1132
1133 __ cbnz(c_rarg2, call_thaw);
1134
1135 const address tr_call = __ trampoline_call(resolve);
1136 if (tr_call == nullptr) {
1137 fatal("CodeCache is full at gen_continuation_enter");
1138 }
1139
1140 oop_maps->add_gc_map(__ pc() - start, map);
1141 __ post_call_nop();
1142
1143 __ b(exit);
1144
1145 __ bind(call_thaw);
1146
1147 ContinuationEntry::_thaw_call_pc_offset = __ pc() - start;
1148 __ rt_call(CAST_FROM_FN_PTR(address, StubRoutines::cont_thaw()));
1149 oop_maps->add_gc_map(__ pc() - start, map->deep_copy());
1150 ContinuationEntry::_return_pc_offset = __ pc() - start;
1151 __ post_call_nop();
1152
1153 __ bind(exit);
1154 ContinuationEntry::_cleanup_offset = __ pc() - start;
1155 continuation_enter_cleanup(masm);
1156 __ leave();
1157 __ ret(lr);
1158
1159 /// exception handling
1160
1161 exception_offset = __ pc() - start;
1162 {
1163 __ mov(r19, r0); // save return value contaning the exception oop in callee-saved R19
1164
1165 continuation_enter_cleanup(masm);
1166
1167 __ ldr(c_rarg1, Address(rfp, wordSize)); // return address
1168 __ authenticate_return_address(c_rarg1);
1169 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rthread, c_rarg1);
1170
1171 // see OptoRuntime::generate_exception_blob: r0 -- exception oop, r3 -- exception pc
1172
1173 __ mov(r1, r0); // the exception handler
1174 __ mov(r0, r19); // restore return value contaning the exception oop
1175 __ verify_oop(r0);
1176
1177 __ leave();
1178 __ mov(r3, lr);
1179 __ br(r1); // the exception handler
1180 }
1181
1182 address stub = CompiledDirectCall::emit_to_interp_stub(masm, tr_call);
1183 if (stub == nullptr) {
1184 fatal("CodeCache is full at gen_continuation_enter");
1185 }
1186 }
1187
1188 static void gen_continuation_yield(MacroAssembler* masm,
1189 const methodHandle& method,
1190 const BasicType* sig_bt,
1191 const VMRegPair* regs,
1192 OopMapSet* oop_maps,
1193 int& frame_complete,
1194 int& stack_slots,
1195 int& compiled_entry_offset) {
1196 enum layout {
1197 rfp_off1,
1198 rfp_off2,
1199 lr_off,
1200 lr_off2,
1201 framesize // inclusive of return address
1202 };
1203 // assert(is_even(framesize/2), "sp not 16-byte aligned");
1204 stack_slots = framesize / VMRegImpl::slots_per_word;
1205 assert(stack_slots == 2, "recheck layout");
1206
1207 address start = __ pc();
1208
1209 compiled_entry_offset = __ pc() - start;
1210 __ enter();
1211
1212 __ mov(c_rarg1, sp);
1213
1214 frame_complete = __ pc() - start;
1215 address the_pc = __ pc();
1216
1217 __ 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
1218
1219 __ mov(c_rarg0, rthread);
1220 __ set_last_Java_frame(sp, rfp, the_pc, rscratch1);
1221 __ call_VM_leaf(Continuation::freeze_entry(), 2);
1222 __ reset_last_Java_frame(true);
1223
1224 Label pinned;
1225
1226 __ cbnz(r0, pinned);
1227
1228 // We've succeeded, set sp to the ContinuationEntry
1229 __ ldr(rscratch1, Address(rthread, JavaThread::cont_entry_offset()));
1230 __ mov(sp, rscratch1);
1231 continuation_enter_cleanup(masm);
1232
1233 __ bind(pinned); // pinned -- return to caller
1234
1235 // handle pending exception thrown by freeze
1236 __ ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
1237 Label ok;
1238 __ cbz(rscratch1, ok);
1239 __ leave();
1240 __ lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
1241 __ br(rscratch1);
1242 __ bind(ok);
1243
1244 __ leave();
1245 __ ret(lr);
1246
1247 OopMap* map = new OopMap(framesize, 1);
1248 oop_maps->add_gc_map(the_pc - start, map);
1249 }
1250
1251 void SharedRuntime::continuation_enter_cleanup(MacroAssembler* masm) {
1252 ::continuation_enter_cleanup(masm);
1253 }
1254
1255 static void gen_special_dispatch(MacroAssembler* masm,
1256 const methodHandle& method,
1257 const BasicType* sig_bt,
1258 const VMRegPair* regs) {
1259 verify_oop_args(masm, method, sig_bt, regs);
1260 vmIntrinsics::ID iid = method->intrinsic_id();
1261
1262 // Now write the args into the outgoing interpreter space
1263 bool has_receiver = false;
1264 Register receiver_reg = noreg;
1265 int member_arg_pos = -1;
1266 Register member_reg = noreg;
1267 int ref_kind = MethodHandles::signature_polymorphic_intrinsic_ref_kind(iid);
1268 if (ref_kind != 0) {
1269 member_arg_pos = method->size_of_parameters() - 1; // trailing MemberName argument
1270 member_reg = r19; // known to be free at this point
1271 has_receiver = MethodHandles::ref_kind_has_receiver(ref_kind);
1272 } else if (iid == vmIntrinsics::_invokeBasic) {
1273 has_receiver = true;
1274 } else if (iid == vmIntrinsics::_linkToNative) {
1275 member_arg_pos = method->size_of_parameters() - 1; // trailing NativeEntryPoint argument
1276 member_reg = r19; // known to be free at this point
1277 } else {
1278 fatal("unexpected intrinsic id %d", vmIntrinsics::as_int(iid));
1279 }
1280
1281 if (member_reg != noreg) {
1282 // Load the member_arg into register, if necessary.
1283 SharedRuntime::check_member_name_argument_is_last_argument(method, sig_bt, regs);
1284 VMReg r = regs[member_arg_pos].first();
1285 if (r->is_stack()) {
1286 __ ldr(member_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
1287 } else {
1288 // no data motion is needed
1289 member_reg = r->as_Register();
1290 }
1291 }
1292
1293 if (has_receiver) {
1294 // Make sure the receiver is loaded into a register.
1295 assert(method->size_of_parameters() > 0, "oob");
1296 assert(sig_bt[0] == T_OBJECT, "receiver argument must be an object");
1297 VMReg r = regs[0].first();
1298 assert(r->is_valid(), "bad receiver arg");
1299 if (r->is_stack()) {
1300 // Porting note: This assumes that compiled calling conventions always
1301 // pass the receiver oop in a register. If this is not true on some
1302 // platform, pick a temp and load the receiver from stack.
1303 fatal("receiver always in a register");
1304 receiver_reg = r2; // known to be free at this point
1305 __ ldr(receiver_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
1306 } else {
1307 // no data motion is needed
1308 receiver_reg = r->as_Register();
1309 }
1310 }
1311
1312 // Figure out which address we are really jumping to:
1313 MethodHandles::generate_method_handle_dispatch(masm, iid,
1314 receiver_reg, member_reg, /*for_compiler_entry:*/ true);
1315 }
1316
1317 // ---------------------------------------------------------------------------
1318 // Generate a native wrapper for a given method. The method takes arguments
1319 // in the Java compiled code convention, marshals them to the native
1320 // convention (handlizes oops, etc), transitions to native, makes the call,
1321 // returns to java state (possibly blocking), unhandlizes any result and
1322 // returns.
1323 //
1324 // Critical native functions are a shorthand for the use of
1325 // GetPrimtiveArrayCritical and disallow the use of any other JNI
1326 // functions. The wrapper is expected to unpack the arguments before
1327 // passing them to the callee. Critical native functions leave the state _in_Java,
1328 // since they block out GC.
1329 // Some other parts of JNI setup are skipped like the tear down of the JNI handle
1330 // block and the check for pending exceptions it's impossible for them
1331 // to be thrown.
1332 //
1333 nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
1334 const methodHandle& method,
1335 int compile_id,
1336 BasicType* in_sig_bt,
1337 VMRegPair* in_regs,
1338 BasicType ret_type) {
1339 if (method->is_continuation_native_intrinsic()) {
1340 int exception_offset = -1;
1341 OopMapSet* oop_maps = new OopMapSet();
1342 int frame_complete = -1;
1343 int stack_slots = -1;
1344 int interpreted_entry_offset = -1;
1345 int vep_offset = -1;
1346 if (method->is_continuation_enter_intrinsic()) {
1347 gen_continuation_enter(masm,
1348 method,
1349 in_sig_bt,
1350 in_regs,
1351 exception_offset,
1352 oop_maps,
1353 frame_complete,
1354 stack_slots,
1355 interpreted_entry_offset,
1356 vep_offset);
1357 } else if (method->is_continuation_yield_intrinsic()) {
1358 gen_continuation_yield(masm,
1359 method,
1360 in_sig_bt,
1361 in_regs,
1362 oop_maps,
1363 frame_complete,
1364 stack_slots,
1365 vep_offset);
1366 } else {
1367 guarantee(false, "Unknown Continuation native intrinsic");
1368 }
1369
1370 #ifdef ASSERT
1371 if (method->is_continuation_enter_intrinsic()) {
1372 assert(interpreted_entry_offset != -1, "Must be set");
1373 assert(exception_offset != -1, "Must be set");
1374 } else {
1375 assert(interpreted_entry_offset == -1, "Must be unset");
1376 assert(exception_offset == -1, "Must be unset");
1377 }
1378 assert(frame_complete != -1, "Must be set");
1379 assert(stack_slots != -1, "Must be set");
1380 assert(vep_offset != -1, "Must be set");
1381 #endif
1382
1383 __ flush();
1384 nmethod* nm = nmethod::new_native_nmethod(method,
1385 compile_id,
1386 masm->code(),
1387 vep_offset,
1388 frame_complete,
1389 stack_slots,
1390 in_ByteSize(-1),
1391 in_ByteSize(-1),
1392 oop_maps,
1393 exception_offset);
1394 if (nm == nullptr) return nm;
1395 if (method->is_continuation_enter_intrinsic()) {
1396 ContinuationEntry::set_enter_code(nm, interpreted_entry_offset);
1397 } else if (method->is_continuation_yield_intrinsic()) {
1398 _cont_doYield_stub = nm;
1399 } else {
1400 guarantee(false, "Unknown Continuation native intrinsic");
1401 }
1402 return nm;
1403 }
1404
1405 if (method->is_method_handle_intrinsic()) {
1406 vmIntrinsics::ID iid = method->intrinsic_id();
1407 intptr_t start = (intptr_t)__ pc();
1408 int vep_offset = ((intptr_t)__ pc()) - start;
1409
1410 // First instruction must be a nop as it may need to be patched on deoptimisation
1411 __ nop();
1412 gen_special_dispatch(masm,
1413 method,
1414 in_sig_bt,
1415 in_regs);
1416 int frame_complete = ((intptr_t)__ pc()) - start; // not complete, period
1417 __ flush();
1418 int stack_slots = SharedRuntime::out_preserve_stack_slots(); // no out slots at all, actually
1419 return nmethod::new_native_nmethod(method,
1420 compile_id,
1421 masm->code(),
1422 vep_offset,
1423 frame_complete,
1424 stack_slots / VMRegImpl::slots_per_word,
1425 in_ByteSize(-1),
1426 in_ByteSize(-1),
1427 nullptr);
1428 }
1429 address native_func = method->native_function();
1430 assert(native_func != nullptr, "must have function");
1431
1432 // An OopMap for lock (and class if static)
1433 OopMapSet *oop_maps = new OopMapSet();
1434 intptr_t start = (intptr_t)__ pc();
1435
1436 // We have received a description of where all the java arg are located
1437 // on entry to the wrapper. We need to convert these args to where
1438 // the jni function will expect them. To figure out where they go
1439 // we convert the java signature to a C signature by inserting
1440 // the hidden arguments as arg[0] and possibly arg[1] (static method)
1441
1442 const int total_in_args = method->size_of_parameters();
1443 int total_c_args = total_in_args + (method->is_static() ? 2 : 1);
1444
1445 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args);
1446 VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args);
1447
1448 int argc = 0;
1449 out_sig_bt[argc++] = T_ADDRESS;
1450 if (method->is_static()) {
1451 out_sig_bt[argc++] = T_OBJECT;
1452 }
1453
1454 for (int i = 0; i < total_in_args ; i++ ) {
1455 out_sig_bt[argc++] = in_sig_bt[i];
1456 }
1457
1458 // Now figure out where the args must be stored and how much stack space
1459 // they require.
1460 int out_arg_slots;
1461 out_arg_slots = c_calling_convention_priv(out_sig_bt, out_regs, total_c_args);
1462
1463 if (out_arg_slots < 0) {
1464 return nullptr;
1465 }
1466
1467 // Compute framesize for the wrapper. We need to handlize all oops in
1468 // incoming registers
1469
1470 // Calculate the total number of stack slots we will need.
1471
1472 // First count the abi requirement plus all of the outgoing args
1473 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
1474
1475 // Now the space for the inbound oop handle area
1476 int total_save_slots = 8 * VMRegImpl::slots_per_word; // 8 arguments passed in registers
1477
1478 int oop_handle_offset = stack_slots;
1479 stack_slots += total_save_slots;
1480
1481 // Now any space we need for handlizing a klass if static method
1482
1483 int klass_slot_offset = 0;
1484 int klass_offset = -1;
1485 int lock_slot_offset = 0;
1486 bool is_static = false;
1487
1488 if (method->is_static()) {
1489 klass_slot_offset = stack_slots;
1490 stack_slots += VMRegImpl::slots_per_word;
1491 klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size;
1492 is_static = true;
1493 }
1494
1495 // Plus a lock if needed
1496
1497 if (method->is_synchronized()) {
1498 lock_slot_offset = stack_slots;
1499 stack_slots += VMRegImpl::slots_per_word;
1500 }
1501
1502 // Now a place (+2) to save return values or temp during shuffling
1503 // + 4 for return address (which we own) and saved rfp
1504 stack_slots += 6;
1505
1506 // Ok The space we have allocated will look like:
1507 //
1508 //
1509 // FP-> | |
1510 // |---------------------|
1511 // | 2 slots for moves |
1512 // |---------------------|
1513 // | lock box (if sync) |
1514 // |---------------------| <- lock_slot_offset
1515 // | klass (if static) |
1516 // |---------------------| <- klass_slot_offset
1517 // | oopHandle area |
1518 // |---------------------| <- oop_handle_offset (8 java arg registers)
1519 // | outbound memory |
1520 // | based arguments |
1521 // | |
1522 // |---------------------|
1523 // | |
1524 // SP-> | out_preserved_slots |
1525 //
1526 //
1527
1528
1529 // Now compute actual number of stack words we need rounding to make
1530 // stack properly aligned.
1531 stack_slots = align_up(stack_slots, StackAlignmentInSlots);
1532
1533 int stack_size = stack_slots * VMRegImpl::stack_slot_size;
1534
1535 // First thing make an ic check to see if we should even be here
1536
1537 // We are free to use all registers as temps without saving them and
1538 // restoring them except rfp. rfp is the only callee save register
1539 // as far as the interpreter and the compiler(s) are concerned.
1540
1541 const Register receiver = j_rarg0;
1542
1543 Label exception_pending;
1544
1545 assert_different_registers(receiver, rscratch1);
1546 __ verify_oop(receiver);
1547 __ ic_check(8 /* end_alignment */);
1548
1549 // Verified entry point must be aligned
1550 int vep_offset = ((intptr_t)__ pc()) - start;
1551
1552 // If we have to make this method not-entrant we'll overwrite its
1553 // first instruction with a jump. For this action to be legal we
1554 // must ensure that this first instruction is a B, BL, NOP, BKPT,
1555 // SVC, HVC, or SMC. Make it a NOP.
1556 __ nop();
1557
1558 if (VM_Version::supports_fast_class_init_checks() && method->needs_clinit_barrier()) {
1559 Label L_skip_barrier;
1560 __ mov_metadata(rscratch2, method->method_holder()); // InstanceKlass*
1561 __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
1562 __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
1563
1564 __ bind(L_skip_barrier);
1565 }
1566
1567 // Generate stack overflow check
1568 __ bang_stack_with_offset(checked_cast<int>(StackOverflow::stack_shadow_zone_size()));
1569
1570 // Generate a new frame for the wrapper.
1571 __ enter();
1572 // -2 because return address is already present and so is saved rfp
1573 __ sub(sp, sp, stack_size - 2*wordSize);
1574
1575 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
1576 bs->nmethod_entry_barrier(masm, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
1577
1578 // Frame is now completed as far as size and linkage.
1579 int frame_complete = ((intptr_t)__ pc()) - start;
1580
1581 // We use r20 as the oop handle for the receiver/klass
1582 // It is callee save so it survives the call to native
1583
1584 const Register oop_handle_reg = r20;
1585
1586 //
1587 // We immediately shuffle the arguments so that any vm call we have to
1588 // make from here on out (sync slow path, jvmti, etc.) we will have
1589 // captured the oops from our caller and have a valid oopMap for
1590 // them.
1591
1592 // -----------------
1593 // The Grand Shuffle
1594
1595 // The Java calling convention is either equal (linux) or denser (win64) than the
1596 // c calling convention. However the because of the jni_env argument the c calling
1597 // convention always has at least one more (and two for static) arguments than Java.
1598 // Therefore if we move the args from java -> c backwards then we will never have
1599 // a register->register conflict and we don't have to build a dependency graph
1600 // and figure out how to break any cycles.
1601 //
1602
1603 // Record esp-based slot for receiver on stack for non-static methods
1604 int receiver_offset = -1;
1605
1606 // This is a trick. We double the stack slots so we can claim
1607 // the oops in the caller's frame. Since we are sure to have
1608 // more args than the caller doubling is enough to make
1609 // sure we can capture all the incoming oop args from the
1610 // caller.
1611 //
1612 OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1613
1614 // Mark location of rfp (someday)
1615 // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, vmreg(rfp));
1616
1617
1618 int float_args = 0;
1619 int int_args = 0;
1620
1621 #ifdef ASSERT
1622 bool reg_destroyed[Register::number_of_registers];
1623 bool freg_destroyed[FloatRegister::number_of_registers];
1624 for ( int r = 0 ; r < Register::number_of_registers ; r++ ) {
1625 reg_destroyed[r] = false;
1626 }
1627 for ( int f = 0 ; f < FloatRegister::number_of_registers ; f++ ) {
1628 freg_destroyed[f] = false;
1629 }
1630
1631 #endif /* ASSERT */
1632
1633 // For JNI natives the incoming and outgoing registers are offset upwards.
1634 GrowableArray<int> arg_order(2 * total_in_args);
1635
1636 for (int i = total_in_args - 1, c_arg = total_c_args - 1; i >= 0; i--, c_arg--) {
1637 arg_order.push(i);
1638 arg_order.push(c_arg);
1639 }
1640
1641 for (int ai = 0; ai < arg_order.length(); ai += 2) {
1642 int i = arg_order.at(ai);
1643 int c_arg = arg_order.at(ai + 1);
1644 __ block_comment(err_msg("move %d -> %d", i, c_arg));
1645 assert(c_arg != -1 && i != -1, "wrong order");
1646 #ifdef ASSERT
1647 if (in_regs[i].first()->is_Register()) {
1648 assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!");
1649 } else if (in_regs[i].first()->is_FloatRegister()) {
1650 assert(!freg_destroyed[in_regs[i].first()->as_FloatRegister()->encoding()], "destroyed reg!");
1651 }
1652 if (out_regs[c_arg].first()->is_Register()) {
1653 reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
1654 } else if (out_regs[c_arg].first()->is_FloatRegister()) {
1655 freg_destroyed[out_regs[c_arg].first()->as_FloatRegister()->encoding()] = true;
1656 }
1657 #endif /* ASSERT */
1658 switch (in_sig_bt[i]) {
1659 case T_ARRAY:
1660 case T_OBJECT:
1661 __ object_move(map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
1662 ((i == 0) && (!is_static)),
1663 &receiver_offset);
1664 int_args++;
1665 break;
1666 case T_VOID:
1667 break;
1668
1669 case T_FLOAT:
1670 __ float_move(in_regs[i], out_regs[c_arg]);
1671 float_args++;
1672 break;
1673
1674 case T_DOUBLE:
1675 assert( i + 1 < total_in_args &&
1676 in_sig_bt[i + 1] == T_VOID &&
1677 out_sig_bt[c_arg+1] == T_VOID, "bad arg list");
1678 __ double_move(in_regs[i], out_regs[c_arg]);
1679 float_args++;
1680 break;
1681
1682 case T_LONG :
1683 __ long_move(in_regs[i], out_regs[c_arg]);
1684 int_args++;
1685 break;
1686
1687 case T_ADDRESS: assert(false, "found T_ADDRESS in java args");
1688
1689 default:
1690 __ move32_64(in_regs[i], out_regs[c_arg]);
1691 int_args++;
1692 }
1693 }
1694
1695 // point c_arg at the first arg that is already loaded in case we
1696 // need to spill before we call out
1697 int c_arg = total_c_args - total_in_args;
1698
1699 // Pre-load a static method's oop into c_rarg1.
1700 if (method->is_static()) {
1701
1702 // load oop into a register
1703 __ movoop(c_rarg1,
1704 JNIHandles::make_local(method->method_holder()->java_mirror()));
1705
1706 // Now handlize the static class mirror it's known not-null.
1707 __ str(c_rarg1, Address(sp, klass_offset));
1708 map->set_oop(VMRegImpl::stack2reg(klass_slot_offset));
1709
1710 // Now get the handle
1711 __ lea(c_rarg1, Address(sp, klass_offset));
1712 // and protect the arg if we must spill
1713 c_arg--;
1714 }
1715
1716 // Change state to native (we save the return address in the thread, since it might not
1717 // be pushed on the stack when we do a stack traversal). It is enough that the pc()
1718 // points into the right code segment. It does not have to be the correct return pc.
1719 // We use the same pc/oopMap repeatedly when we call out.
1720
1721 Label native_return;
1722 if (method->is_object_wait0()) {
1723 // For convenience we use the pc we want to resume to in case of preemption on Object.wait.
1724 __ set_last_Java_frame(sp, noreg, native_return, rscratch1);
1725 } else {
1726 intptr_t the_pc = (intptr_t) __ pc();
1727 oop_maps->add_gc_map(the_pc - start, map);
1728
1729 __ set_last_Java_frame(sp, noreg, __ pc(), rscratch1);
1730 }
1731
1732 Label dtrace_method_entry, dtrace_method_entry_done;
1733 if (DTraceMethodProbes) {
1734 __ b(dtrace_method_entry);
1735 __ bind(dtrace_method_entry_done);
1736 }
1737
1738 // RedefineClasses() tracing support for obsolete method entry
1739 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1740 // protect the args we've loaded
1741 save_args(masm, total_c_args, c_arg, out_regs);
1742 __ mov_metadata(c_rarg1, method());
1743 __ call_VM_leaf(
1744 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1745 rthread, c_rarg1);
1746 restore_args(masm, total_c_args, c_arg, out_regs);
1747 }
1748
1749 // Lock a synchronized method
1750
1751 // Register definitions used by locking and unlocking
1752
1753 const Register swap_reg = r0;
1754 const Register obj_reg = r19; // Will contain the oop
1755 const Register lock_reg = r13; // Address of compiler lock object (BasicLock)
1756 const Register old_hdr = r13; // value of old header at unlock time
1757 const Register lock_tmp = r14; // Temporary used by lightweight_lock/unlock
1758 const Register tmp = lr;
1759
1760 Label slow_path_lock;
1761 Label lock_done;
1762
1763 if (method->is_synchronized()) {
1764 // Get the handle (the 2nd argument)
1765 __ mov(oop_handle_reg, c_rarg1);
1766
1767 // Get address of the box
1768
1769 __ lea(lock_reg, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1770
1771 // Load the oop from the handle
1772 __ ldr(obj_reg, Address(oop_handle_reg, 0));
1773
1774 __ lightweight_lock(lock_reg, obj_reg, swap_reg, tmp, lock_tmp, slow_path_lock);
1775
1776 // Slow path will re-enter here
1777 __ bind(lock_done);
1778 }
1779
1780
1781 // Finally just about ready to make the JNI call
1782
1783 // get JNIEnv* which is first argument to native
1784 __ lea(c_rarg0, Address(rthread, in_bytes(JavaThread::jni_environment_offset())));
1785
1786 // Now set thread in native
1787 __ mov(rscratch1, _thread_in_native);
1788 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1789 __ stlrw(rscratch1, rscratch2);
1790
1791 __ rt_call(native_func);
1792
1793 // Verify or restore cpu control state after JNI call
1794 __ restore_cpu_control_state_after_jni(rscratch1, rscratch2);
1795
1796 // Unpack native results.
1797 switch (ret_type) {
1798 case T_BOOLEAN: __ c2bool(r0); break;
1799 case T_CHAR : __ ubfx(r0, r0, 0, 16); break;
1800 case T_BYTE : __ sbfx(r0, r0, 0, 8); break;
1801 case T_SHORT : __ sbfx(r0, r0, 0, 16); break;
1802 case T_INT : __ sbfx(r0, r0, 0, 32); break;
1803 case T_DOUBLE :
1804 case T_FLOAT :
1805 // Result is in v0 we'll save as needed
1806 break;
1807 case T_ARRAY: // Really a handle
1808 case T_OBJECT: // Really a handle
1809 break; // can't de-handlize until after safepoint check
1810 case T_VOID: break;
1811 case T_LONG: break;
1812 default : ShouldNotReachHere();
1813 }
1814
1815 Label safepoint_in_progress, safepoint_in_progress_done;
1816
1817 // Switch thread to "native transition" state before reading the synchronization state.
1818 // This additional state is necessary because reading and testing the synchronization
1819 // state is not atomic w.r.t. GC, as this scenario demonstrates:
1820 // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
1821 // VM thread changes sync state to synchronizing and suspends threads for GC.
1822 // Thread A is resumed to finish this native method, but doesn't block here since it
1823 // didn't see any synchronization is progress, and escapes.
1824 __ mov(rscratch1, _thread_in_native_trans);
1825
1826 __ strw(rscratch1, Address(rthread, JavaThread::thread_state_offset()));
1827
1828 // Force this write out before the read below
1829 if (!UseSystemMemoryBarrier) {
1830 __ dmb(Assembler::ISH);
1831 }
1832
1833 __ verify_sve_vector_length();
1834
1835 // Check for safepoint operation in progress and/or pending suspend requests.
1836 {
1837 // No need for acquire as Java threads always disarm themselves.
1838 __ safepoint_poll(safepoint_in_progress, true /* at_return */, false /* in_nmethod */);
1839 __ ldrw(rscratch1, Address(rthread, JavaThread::suspend_flags_offset()));
1840 __ cbnzw(rscratch1, safepoint_in_progress);
1841 __ bind(safepoint_in_progress_done);
1842 }
1843
1844 // change thread state
1845 __ mov(rscratch1, _thread_in_Java);
1846 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1847 __ stlrw(rscratch1, rscratch2);
1848
1849 if (method->is_object_wait0()) {
1850 // Check preemption for Object.wait()
1851 __ ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1852 __ cbz(rscratch1, native_return);
1853 __ str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1854 __ br(rscratch1);
1855 __ bind(native_return);
1856
1857 intptr_t the_pc = (intptr_t) __ pc();
1858 oop_maps->add_gc_map(the_pc - start, map);
1859 }
1860
1861 Label reguard;
1862 Label reguard_done;
1863 __ ldrb(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
1864 __ cmpw(rscratch1, StackOverflow::stack_guard_yellow_reserved_disabled);
1865 __ br(Assembler::EQ, reguard);
1866 __ bind(reguard_done);
1867
1868 // native result if any is live
1869
1870 // Unlock
1871 Label unlock_done;
1872 Label slow_path_unlock;
1873 if (method->is_synchronized()) {
1874
1875 // Get locked oop from the handle we passed to jni
1876 __ ldr(obj_reg, Address(oop_handle_reg, 0));
1877
1878 // Must save r0 if if it is live now because cmpxchg must use it
1879 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1880 save_native_result(masm, ret_type, stack_slots);
1881 }
1882
1883 __ lightweight_unlock(obj_reg, old_hdr, swap_reg, lock_tmp, slow_path_unlock);
1884
1885 // slow path re-enters here
1886 __ bind(unlock_done);
1887 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1888 restore_native_result(masm, ret_type, stack_slots);
1889 }
1890 }
1891
1892 Label dtrace_method_exit, dtrace_method_exit_done;
1893 if (DTraceMethodProbes) {
1894 __ b(dtrace_method_exit);
1895 __ bind(dtrace_method_exit_done);
1896 }
1897
1898 __ reset_last_Java_frame(false);
1899
1900 // Unbox oop result, e.g. JNIHandles::resolve result.
1901 if (is_reference_type(ret_type)) {
1902 __ resolve_jobject(r0, r1, r2);
1903 }
1904
1905 if (CheckJNICalls) {
1906 // clear_pending_jni_exception_check
1907 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1908 }
1909
1910 // reset handle block
1911 __ ldr(r2, Address(rthread, JavaThread::active_handles_offset()));
1912 __ str(zr, Address(r2, JNIHandleBlock::top_offset()));
1913
1914 __ leave();
1915
1916 #if INCLUDE_JFR
1917 // We need to do a poll test after unwind in case the sampler
1918 // managed to sample the native frame after returning to Java.
1919 Label L_return;
1920 __ ldr(rscratch1, Address(rthread, JavaThread::polling_word_offset()));
1921 address poll_test_pc = __ pc();
1922 __ relocate(relocInfo::poll_return_type);
1923 __ tbz(rscratch1, log2i_exact(SafepointMechanism::poll_bit()), L_return);
1924 assert(SharedRuntime::polling_page_return_handler_blob() != nullptr,
1925 "polling page return stub not created yet");
1926 address stub = SharedRuntime::polling_page_return_handler_blob()->entry_point();
1927 __ adr(rscratch1, InternalAddress(poll_test_pc));
1928 __ str(rscratch1, Address(rthread, JavaThread::saved_exception_pc_offset()));
1929 __ far_jump(RuntimeAddress(stub));
1930 __ bind(L_return);
1931 #endif // INCLUDE_JFR
1932
1933 // Any exception pending?
1934 __ ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
1935 __ cbnz(rscratch1, exception_pending);
1936
1937 // We're done
1938 __ ret(lr);
1939
1940 // Unexpected paths are out of line and go here
1941
1942 // forward the exception
1943 __ bind(exception_pending);
1944
1945 // and forward the exception
1946 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1947
1948 // Slow path locking & unlocking
1949 if (method->is_synchronized()) {
1950
1951 __ block_comment("Slow path lock {");
1952 __ bind(slow_path_lock);
1953
1954 // has last_Java_frame setup. No exceptions so do vanilla call not call_VM
1955 // args are (oop obj, BasicLock* lock, JavaThread* thread)
1956
1957 // protect the args we've loaded
1958 save_args(masm, total_c_args, c_arg, out_regs);
1959
1960 __ mov(c_rarg0, obj_reg);
1961 __ mov(c_rarg1, lock_reg);
1962 __ mov(c_rarg2, rthread);
1963
1964 // Not a leaf but we have last_Java_frame setup as we want.
1965 // We don't want to unmount in case of contention since that would complicate preserving
1966 // the arguments that had already been marshalled into the native convention. So we force
1967 // the freeze slow path to find this native wrapper frame (see recurse_freeze_native_frame())
1968 // and pin the vthread. Otherwise the fast path won't find it since we don't walk the stack.
1969 __ push_cont_fastpath();
1970 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), 3);
1971 __ pop_cont_fastpath();
1972 restore_args(masm, total_c_args, c_arg, out_regs);
1973
1974 #ifdef ASSERT
1975 { Label L;
1976 __ ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
1977 __ cbz(rscratch1, L);
1978 __ stop("no pending exception allowed on exit from monitorenter");
1979 __ bind(L);
1980 }
1981 #endif
1982 __ b(lock_done);
1983
1984 __ block_comment("} Slow path lock");
1985
1986 __ block_comment("Slow path unlock {");
1987 __ bind(slow_path_unlock);
1988
1989 // If we haven't already saved the native result we must save it now as xmm registers
1990 // are still exposed.
1991
1992 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
1993 save_native_result(masm, ret_type, stack_slots);
1994 }
1995
1996 __ mov(c_rarg2, rthread);
1997 __ lea(c_rarg1, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1998 __ mov(c_rarg0, obj_reg);
1999
2000 // Save pending exception around call to VM (which contains an EXCEPTION_MARK)
2001 // NOTE that obj_reg == r19 currently
2002 __ ldr(r19, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2003 __ str(zr, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2004
2005 __ rt_call(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C));
2006
2007 #ifdef ASSERT
2008 {
2009 Label L;
2010 __ ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2011 __ cbz(rscratch1, L);
2012 __ stop("no pending exception allowed on exit complete_monitor_unlocking_C");
2013 __ bind(L);
2014 }
2015 #endif /* ASSERT */
2016
2017 __ str(r19, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2018
2019 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
2020 restore_native_result(masm, ret_type, stack_slots);
2021 }
2022 __ b(unlock_done);
2023
2024 __ block_comment("} Slow path unlock");
2025
2026 } // synchronized
2027
2028 // SLOW PATH Reguard the stack if needed
2029
2030 __ bind(reguard);
2031 save_native_result(masm, ret_type, stack_slots);
2032 __ rt_call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
2033 restore_native_result(masm, ret_type, stack_slots);
2034 // and continue
2035 __ b(reguard_done);
2036
2037 // SLOW PATH safepoint
2038 {
2039 __ block_comment("safepoint {");
2040 __ bind(safepoint_in_progress);
2041
2042 // Don't use call_VM as it will see a possible pending exception and forward it
2043 // and never return here preventing us from clearing _last_native_pc down below.
2044 //
2045 save_native_result(masm, ret_type, stack_slots);
2046 __ mov(c_rarg0, rthread);
2047 #ifndef PRODUCT
2048 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
2049 #endif
2050 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
2051 __ blr(rscratch1);
2052
2053 // Restore any method result value
2054 restore_native_result(masm, ret_type, stack_slots);
2055
2056 __ b(safepoint_in_progress_done);
2057 __ block_comment("} safepoint");
2058 }
2059
2060 // SLOW PATH dtrace support
2061 if (DTraceMethodProbes) {
2062 {
2063 __ block_comment("dtrace entry {");
2064 __ bind(dtrace_method_entry);
2065
2066 // We have all of the arguments setup at this point. We must not touch any register
2067 // argument registers at this point (what if we save/restore them there are no oop?
2068
2069 save_args(masm, total_c_args, c_arg, out_regs);
2070 __ mov_metadata(c_rarg1, method());
2071 __ call_VM_leaf(
2072 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2073 rthread, c_rarg1);
2074 restore_args(masm, total_c_args, c_arg, out_regs);
2075 __ b(dtrace_method_entry_done);
2076 __ block_comment("} dtrace entry");
2077 }
2078
2079 {
2080 __ block_comment("dtrace exit {");
2081 __ bind(dtrace_method_exit);
2082 save_native_result(masm, ret_type, stack_slots);
2083 __ mov_metadata(c_rarg1, method());
2084 __ call_VM_leaf(
2085 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2086 rthread, c_rarg1);
2087 restore_native_result(masm, ret_type, stack_slots);
2088 __ b(dtrace_method_exit_done);
2089 __ block_comment("} dtrace exit");
2090 }
2091 }
2092
2093 __ flush();
2094
2095 nmethod *nm = nmethod::new_native_nmethod(method,
2096 compile_id,
2097 masm->code(),
2098 vep_offset,
2099 frame_complete,
2100 stack_slots / VMRegImpl::slots_per_word,
2101 (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)),
2102 in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size),
2103 oop_maps);
2104
2105 return nm;
2106 }
2107
2108 // this function returns the adjust size (in number of words) to a c2i adapter
2109 // activation for use during deoptimization
2110 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals) {
2111 assert(callee_locals >= callee_parameters,
2112 "test and remove; got more parms than locals");
2113 if (callee_locals < callee_parameters)
2114 return 0; // No adjustment for negative locals
2115 int diff = (callee_locals - callee_parameters) * Interpreter::stackElementWords;
2116 // diff is counted in stack words
2117 return align_up(diff, 2);
2118 }
2119
2120
2121 //------------------------------generate_deopt_blob----------------------------
2122 void SharedRuntime::generate_deopt_blob() {
2123 // Allocate space for the code
2124 ResourceMark rm;
2125 // Setup code generation tools
2126 int pad = 0;
2127 #if INCLUDE_JVMCI
2128 if (EnableJVMCI) {
2129 pad += 512; // Increase the buffer size when compiling for JVMCI
2130 }
2131 #endif
2132 const char* name = SharedRuntime::stub_name(StubId::shared_deopt_id);
2133 CodeBlob* blob = AOTCodeCache::load_code_blob(AOTCodeEntry::SharedBlob, BlobId::shared_deopt_id);
2134 if (blob != nullptr) {
2135 _deopt_blob = blob->as_deoptimization_blob();
2136 return;
2137 }
2138
2139 CodeBuffer buffer(name, 2048+pad, 1024);
2140 MacroAssembler* masm = new MacroAssembler(&buffer);
2141 int frame_size_in_words;
2142 OopMap* map = nullptr;
2143 OopMapSet *oop_maps = new OopMapSet();
2144 RegisterSaver reg_save(COMPILER2_OR_JVMCI != 0);
2145
2146 // -------------
2147 // This code enters when returning to a de-optimized nmethod. A return
2148 // address has been pushed on the stack, and return values are in
2149 // registers.
2150 // If we are doing a normal deopt then we were called from the patched
2151 // nmethod from the point we returned to the nmethod. So the return
2152 // address on the stack is wrong by NativeCall::instruction_size
2153 // We will adjust the value so it looks like we have the original return
2154 // address on the stack (like when we eagerly deoptimized).
2155 // In the case of an exception pending when deoptimizing, we enter
2156 // with a return address on the stack that points after the call we patched
2157 // into the exception handler. We have the following register state from,
2158 // e.g., the forward exception stub (see stubGenerator_x86_64.cpp).
2159 // r0: exception oop
2160 // r19: exception handler
2161 // r3: throwing pc
2162 // So in this case we simply jam r3 into the useless return address and
2163 // the stack looks just like we want.
2164 //
2165 // At this point we need to de-opt. We save the argument return
2166 // registers. We call the first C routine, fetch_unroll_info(). This
2167 // routine captures the return values and returns a structure which
2168 // describes the current frame size and the sizes of all replacement frames.
2169 // The current frame is compiled code and may contain many inlined
2170 // functions, each with their own JVM state. We pop the current frame, then
2171 // push all the new frames. Then we call the C routine unpack_frames() to
2172 // populate these frames. Finally unpack_frames() returns us the new target
2173 // address. Notice that callee-save registers are BLOWN here; they have
2174 // already been captured in the vframeArray at the time the return PC was
2175 // patched.
2176 address start = __ pc();
2177 Label cont;
2178
2179 // Prolog for non exception case!
2180
2181 // Save everything in sight.
2182 map = reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2183
2184 // Normal deoptimization. Save exec mode for unpack_frames.
2185 __ movw(rcpool, Deoptimization::Unpack_deopt); // callee-saved
2186 __ b(cont);
2187
2188 int reexecute_offset = __ pc() - start;
2189 #if INCLUDE_JVMCI && !defined(COMPILER1)
2190 if (UseJVMCICompiler) {
2191 // JVMCI does not use this kind of deoptimization
2192 __ should_not_reach_here();
2193 }
2194 #endif
2195
2196 // Reexecute case
2197 // return address is the pc describes what bci to do re-execute at
2198
2199 // No need to update map as each call to save_live_registers will produce identical oopmap
2200 (void) reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2201
2202 __ movw(rcpool, Deoptimization::Unpack_reexecute); // callee-saved
2203 __ b(cont);
2204
2205 #if INCLUDE_JVMCI
2206 Label after_fetch_unroll_info_call;
2207 int implicit_exception_uncommon_trap_offset = 0;
2208 int uncommon_trap_offset = 0;
2209
2210 if (EnableJVMCI) {
2211 implicit_exception_uncommon_trap_offset = __ pc() - start;
2212
2213 __ ldr(lr, Address(rthread, in_bytes(JavaThread::jvmci_implicit_exception_pc_offset())));
2214 __ str(zr, Address(rthread, in_bytes(JavaThread::jvmci_implicit_exception_pc_offset())));
2215
2216 uncommon_trap_offset = __ pc() - start;
2217
2218 // Save everything in sight.
2219 reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2220 // fetch_unroll_info needs to call last_java_frame()
2221 Label retaddr;
2222 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2223
2224 __ ldrw(c_rarg1, Address(rthread, in_bytes(JavaThread::pending_deoptimization_offset())));
2225 __ movw(rscratch1, -1);
2226 __ strw(rscratch1, Address(rthread, in_bytes(JavaThread::pending_deoptimization_offset())));
2227
2228 __ movw(rcpool, (int32_t)Deoptimization::Unpack_reexecute);
2229 __ mov(c_rarg0, rthread);
2230 __ movw(c_rarg2, rcpool); // exec mode
2231 __ lea(rscratch1,
2232 RuntimeAddress(CAST_FROM_FN_PTR(address,
2233 Deoptimization::uncommon_trap)));
2234 __ blr(rscratch1);
2235 __ bind(retaddr);
2236 oop_maps->add_gc_map( __ pc()-start, map->deep_copy());
2237
2238 __ reset_last_Java_frame(false);
2239
2240 __ b(after_fetch_unroll_info_call);
2241 } // EnableJVMCI
2242 #endif // INCLUDE_JVMCI
2243
2244 int exception_offset = __ pc() - start;
2245
2246 // Prolog for exception case
2247
2248 // all registers are dead at this entry point, except for r0, and
2249 // r3 which contain the exception oop and exception pc
2250 // respectively. Set them in TLS and fall thru to the
2251 // unpack_with_exception_in_tls entry point.
2252
2253 __ str(r3, Address(rthread, JavaThread::exception_pc_offset()));
2254 __ str(r0, Address(rthread, JavaThread::exception_oop_offset()));
2255
2256 int exception_in_tls_offset = __ pc() - start;
2257
2258 // new implementation because exception oop is now passed in JavaThread
2259
2260 // Prolog for exception case
2261 // All registers must be preserved because they might be used by LinearScan
2262 // Exceptiop oop and throwing PC are passed in JavaThread
2263 // tos: stack at point of call to method that threw the exception (i.e. only
2264 // args are on the stack, no return address)
2265
2266 // The return address pushed by save_live_registers will be patched
2267 // later with the throwing pc. The correct value is not available
2268 // now because loading it from memory would destroy registers.
2269
2270 // NB: The SP at this point must be the SP of the method that is
2271 // being deoptimized. Deoptimization assumes that the frame created
2272 // here by save_live_registers is immediately below the method's SP.
2273 // This is a somewhat fragile mechanism.
2274
2275 // Save everything in sight.
2276 map = reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2277
2278 // Now it is safe to overwrite any register
2279
2280 // Deopt during an exception. Save exec mode for unpack_frames.
2281 __ mov(rcpool, Deoptimization::Unpack_exception); // callee-saved
2282
2283 // load throwing pc from JavaThread and patch it as the return address
2284 // of the current frame. Then clear the field in JavaThread
2285 __ ldr(r3, Address(rthread, JavaThread::exception_pc_offset()));
2286 __ protect_return_address(r3);
2287 __ str(r3, Address(rfp, wordSize));
2288 __ str(zr, Address(rthread, JavaThread::exception_pc_offset()));
2289
2290 #ifdef ASSERT
2291 // verify that there is really an exception oop in JavaThread
2292 __ ldr(r0, Address(rthread, JavaThread::exception_oop_offset()));
2293 __ verify_oop(r0);
2294
2295 // verify that there is no pending exception
2296 Label no_pending_exception;
2297 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
2298 __ cbz(rscratch1, no_pending_exception);
2299 __ stop("must not have pending exception here");
2300 __ bind(no_pending_exception);
2301 #endif
2302
2303 __ bind(cont);
2304
2305 // Call C code. Need thread and this frame, but NOT official VM entry
2306 // crud. We cannot block on this call, no GC can happen.
2307 //
2308 // UnrollBlock* fetch_unroll_info(JavaThread* thread)
2309
2310 // fetch_unroll_info needs to call last_java_frame().
2311
2312 Label retaddr;
2313 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2314 #ifdef ASSERT
2315 { Label L;
2316 __ ldr(rscratch1, Address(rthread, JavaThread::last_Java_fp_offset()));
2317 __ cbz(rscratch1, L);
2318 __ stop("SharedRuntime::generate_deopt_blob: last_Java_fp not cleared");
2319 __ bind(L);
2320 }
2321 #endif // ASSERT
2322 __ mov(c_rarg0, rthread);
2323 __ mov(c_rarg1, rcpool);
2324 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info)));
2325 __ blr(rscratch1);
2326 __ bind(retaddr);
2327
2328 // Need to have an oopmap that tells fetch_unroll_info where to
2329 // find any register it might need.
2330 oop_maps->add_gc_map(__ pc() - start, map);
2331
2332 __ reset_last_Java_frame(false);
2333
2334 #if INCLUDE_JVMCI
2335 if (EnableJVMCI) {
2336 __ bind(after_fetch_unroll_info_call);
2337 }
2338 #endif
2339
2340 // Load UnrollBlock* into r5
2341 __ mov(r5, r0);
2342
2343 __ ldrw(rcpool, Address(r5, Deoptimization::UnrollBlock::unpack_kind_offset()));
2344 Label noException;
2345 __ cmpw(rcpool, Deoptimization::Unpack_exception); // Was exception pending?
2346 __ br(Assembler::NE, noException);
2347 __ ldr(r0, Address(rthread, JavaThread::exception_oop_offset()));
2348 // QQQ this is useless it was null above
2349 __ ldr(r3, Address(rthread, JavaThread::exception_pc_offset()));
2350 __ str(zr, Address(rthread, JavaThread::exception_oop_offset()));
2351 __ str(zr, Address(rthread, JavaThread::exception_pc_offset()));
2352
2353 __ verify_oop(r0);
2354
2355 // Overwrite the result registers with the exception results.
2356 __ str(r0, Address(sp, reg_save.r0_offset_in_bytes()));
2357 // I think this is useless
2358 // __ str(r3, Address(sp, RegisterSaver::r3_offset_in_bytes()));
2359
2360 __ bind(noException);
2361
2362 // Only register save data is on the stack.
2363 // Now restore the result registers. Everything else is either dead
2364 // or captured in the vframeArray.
2365
2366 // Restore fp result register
2367 __ ldrd(v0, Address(sp, reg_save.v0_offset_in_bytes()));
2368 // Restore integer result register
2369 __ ldr(r0, Address(sp, reg_save.r0_offset_in_bytes()));
2370
2371 // Pop all of the register save area off the stack
2372 __ add(sp, sp, frame_size_in_words * wordSize);
2373
2374 // All of the register save area has been popped of the stack. Only the
2375 // return address remains.
2376
2377 // Pop all the frames we must move/replace.
2378 //
2379 // Frame picture (youngest to oldest)
2380 // 1: self-frame (no frame link)
2381 // 2: deopting frame (no frame link)
2382 // 3: caller of deopting frame (could be compiled/interpreted).
2383 //
2384 // Note: by leaving the return address of self-frame on the stack
2385 // and using the size of frame 2 to adjust the stack
2386 // when we are done the return to frame 3 will still be on the stack.
2387
2388 // Pop deoptimized frame
2389 __ ldrw(r2, Address(r5, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset()));
2390 __ sub(r2, r2, 2 * wordSize);
2391 __ add(sp, sp, r2);
2392 __ ldp(rfp, zr, __ post(sp, 2 * wordSize));
2393
2394 #ifdef ASSERT
2395 // Compilers generate code that bang the stack by as much as the
2396 // interpreter would need. So this stack banging should never
2397 // trigger a fault. Verify that it does not on non product builds.
2398 __ ldrw(r19, Address(r5, Deoptimization::UnrollBlock::total_frame_sizes_offset()));
2399 __ bang_stack_size(r19, r2);
2400 #endif
2401 // Load address of array of frame pcs into r2
2402 __ ldr(r2, Address(r5, Deoptimization::UnrollBlock::frame_pcs_offset()));
2403
2404 // Trash the old pc
2405 // __ addptr(sp, wordSize); FIXME ????
2406
2407 // Load address of array of frame sizes into r4
2408 __ ldr(r4, Address(r5, Deoptimization::UnrollBlock::frame_sizes_offset()));
2409
2410 // Load counter into r3
2411 __ ldrw(r3, Address(r5, Deoptimization::UnrollBlock::number_of_frames_offset()));
2412
2413 // Now adjust the caller's stack to make up for the extra locals
2414 // but record the original sp so that we can save it in the skeletal interpreter
2415 // frame and the stack walking of interpreter_sender will get the unextended sp
2416 // value and not the "real" sp value.
2417
2418 const Register sender_sp = r6;
2419
2420 __ mov(sender_sp, sp);
2421 __ ldrw(r19, Address(r5,
2422 Deoptimization::UnrollBlock::
2423 caller_adjustment_offset()));
2424 __ sub(sp, sp, r19);
2425
2426 // Push interpreter frames in a loop
2427 __ mov(rscratch1, (uint64_t)0xDEADDEAD); // Make a recognizable pattern
2428 __ mov(rscratch2, rscratch1);
2429 Label loop;
2430 __ bind(loop);
2431 __ ldr(r19, Address(__ post(r4, wordSize))); // Load frame size
2432 __ sub(r19, r19, 2*wordSize); // We'll push pc and fp by hand
2433 __ ldr(lr, Address(__ post(r2, wordSize))); // Load pc
2434 __ enter(); // Save old & set new fp
2435 __ sub(sp, sp, r19); // Prolog
2436 // This value is corrected by layout_activation_impl
2437 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
2438 __ str(sender_sp, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize)); // Make it walkable
2439 __ mov(sender_sp, sp); // Pass sender_sp to next frame
2440 __ sub(r3, r3, 1); // Decrement counter
2441 __ cbnz(r3, loop);
2442
2443 // Re-push self-frame
2444 __ ldr(lr, Address(r2));
2445 __ enter();
2446
2447 // Allocate a full sized register save area. We subtract 2 because
2448 // enter() just pushed 2 words
2449 __ sub(sp, sp, (frame_size_in_words - 2) * wordSize);
2450
2451 // Restore frame locals after moving the frame
2452 __ strd(v0, Address(sp, reg_save.v0_offset_in_bytes()));
2453 __ str(r0, Address(sp, reg_save.r0_offset_in_bytes()));
2454
2455 // Call C code. Need thread but NOT official VM entry
2456 // crud. We cannot block on this call, no GC can happen. Call should
2457 // restore return values to their stack-slots with the new SP.
2458 //
2459 // void Deoptimization::unpack_frames(JavaThread* thread, int exec_mode)
2460
2461 // Use rfp because the frames look interpreted now
2462 // Don't need the precise return PC here, just precise enough to point into this code blob.
2463 address the_pc = __ pc();
2464 __ set_last_Java_frame(sp, rfp, the_pc, rscratch1);
2465
2466 __ mov(c_rarg0, rthread);
2467 __ movw(c_rarg1, rcpool); // second arg: exec_mode
2468 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
2469 __ blr(rscratch1);
2470
2471 // Set an oopmap for the call site
2472 // Use the same PC we used for the last java frame
2473 oop_maps->add_gc_map(the_pc - start,
2474 new OopMap( frame_size_in_words, 0 ));
2475
2476 // Clear fp AND pc
2477 __ reset_last_Java_frame(true);
2478
2479 // Collect return values
2480 __ ldrd(v0, Address(sp, reg_save.v0_offset_in_bytes()));
2481 __ ldr(r0, Address(sp, reg_save.r0_offset_in_bytes()));
2482 // I think this is useless (throwing pc?)
2483 // __ ldr(r3, Address(sp, RegisterSaver::r3_offset_in_bytes()));
2484
2485 // Pop self-frame.
2486 __ leave(); // Epilog
2487
2488 // Jump to interpreter
2489 __ ret(lr);
2490
2491 // Make sure all code is generated
2492 masm->flush();
2493
2494 _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words);
2495 _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset);
2496 #if INCLUDE_JVMCI
2497 if (EnableJVMCI) {
2498 _deopt_blob->set_uncommon_trap_offset(uncommon_trap_offset);
2499 _deopt_blob->set_implicit_exception_uncommon_trap_offset(implicit_exception_uncommon_trap_offset);
2500 }
2501 #endif
2502
2503 AOTCodeCache::store_code_blob(*_deopt_blob, AOTCodeEntry::SharedBlob, BlobId::shared_deopt_id);
2504 }
2505
2506 // Number of stack slots between incoming argument block and the start of
2507 // a new frame. The PROLOG must add this many slots to the stack. The
2508 // EPILOG must remove this many slots. aarch64 needs two slots for
2509 // return address and fp.
2510 // TODO think this is correct but check
2511 uint SharedRuntime::in_preserve_stack_slots() {
2512 return 4;
2513 }
2514
2515 uint SharedRuntime::out_preserve_stack_slots() {
2516 return 0;
2517 }
2518
2519
2520 VMReg SharedRuntime::thread_register() {
2521 return rthread->as_VMReg();
2522 }
2523
2524 //------------------------------generate_handler_blob------
2525 //
2526 // Generate a special Compile2Runtime blob that saves all registers,
2527 // and setup oopmap.
2528 //
2529 SafepointBlob* SharedRuntime::generate_handler_blob(StubId id, address call_ptr) {
2530 assert(is_polling_page_id(id), "expected a polling page stub id");
2531
2532 // Allocate space for the code. Setup code generation tools.
2533 const char* name = SharedRuntime::stub_name(id);
2534 CodeBlob* blob = AOTCodeCache::load_code_blob(AOTCodeEntry::SharedBlob, StubInfo::blob(id));
2535 if (blob != nullptr) {
2536 return blob->as_safepoint_blob();
2537 }
2538
2539 ResourceMark rm;
2540 OopMapSet *oop_maps = new OopMapSet();
2541 OopMap* map;
2542 CodeBuffer buffer(name, 2048, 1024);
2543 MacroAssembler* masm = new MacroAssembler(&buffer);
2544
2545 address start = __ pc();
2546 address call_pc = nullptr;
2547 int frame_size_in_words;
2548 bool cause_return = (id == StubId::shared_polling_page_return_handler_id);
2549 RegisterSaver reg_save(id == StubId::shared_polling_page_vectors_safepoint_handler_id /* save_vectors */);
2550
2551 // When the signal occurred, the LR was either signed and stored on the stack (in which
2552 // case it will be restored from the stack before being used) or unsigned and not stored
2553 // on the stack. Stipping ensures we get the right value.
2554 __ strip_return_address();
2555
2556 // Save Integer and Float registers.
2557 map = reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2558
2559 // The following is basically a call_VM. However, we need the precise
2560 // address of the call in order to generate an oopmap. Hence, we do all the
2561 // work ourselves.
2562
2563 Label retaddr;
2564 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2565
2566 // The return address must always be correct so that frame constructor never
2567 // sees an invalid pc.
2568
2569 if (!cause_return) {
2570 // overwrite the return address pushed by save_live_registers
2571 // Additionally, r20 is a callee-saved register so we can look at
2572 // it later to determine if someone changed the return address for
2573 // us!
2574 __ ldr(r20, Address(rthread, JavaThread::saved_exception_pc_offset()));
2575 __ protect_return_address(r20);
2576 __ str(r20, Address(rfp, wordSize));
2577 }
2578
2579 // Do the call
2580 __ mov(c_rarg0, rthread);
2581 __ lea(rscratch1, RuntimeAddress(call_ptr));
2582 __ blr(rscratch1);
2583 __ bind(retaddr);
2584
2585 // Set an oopmap for the call site. This oopmap will map all
2586 // oop-registers and debug-info registers as callee-saved. This
2587 // will allow deoptimization at this safepoint to find all possible
2588 // debug-info recordings, as well as let GC find all oops.
2589
2590 oop_maps->add_gc_map( __ pc() - start, map);
2591
2592 Label noException;
2593
2594 __ reset_last_Java_frame(false);
2595
2596 __ membar(Assembler::LoadLoad | Assembler::LoadStore);
2597
2598 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
2599 __ cbz(rscratch1, noException);
2600
2601 // Exception pending
2602
2603 reg_save.restore_live_registers(masm);
2604
2605 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2606
2607 // No exception case
2608 __ bind(noException);
2609
2610 Label no_adjust, bail;
2611 if (!cause_return) {
2612 // If our stashed return pc was modified by the runtime we avoid touching it
2613 __ ldr(rscratch1, Address(rfp, wordSize));
2614 __ cmp(r20, rscratch1);
2615 __ br(Assembler::NE, no_adjust);
2616 __ authenticate_return_address(r20);
2617
2618 #ifdef ASSERT
2619 // Verify the correct encoding of the poll we're about to skip.
2620 // See NativeInstruction::is_ldrw_to_zr()
2621 __ ldrw(rscratch1, Address(r20));
2622 __ ubfx(rscratch2, rscratch1, 22, 10);
2623 __ cmpw(rscratch2, 0b1011100101);
2624 __ br(Assembler::NE, bail);
2625 __ ubfx(rscratch2, rscratch1, 0, 5);
2626 __ cmpw(rscratch2, 0b11111);
2627 __ br(Assembler::NE, bail);
2628 #endif
2629 // Adjust return pc forward to step over the safepoint poll instruction
2630 __ add(r20, r20, NativeInstruction::instruction_size);
2631 __ protect_return_address(r20);
2632 __ str(r20, Address(rfp, wordSize));
2633 }
2634
2635 __ bind(no_adjust);
2636 // Normal exit, restore registers and exit.
2637 reg_save.restore_live_registers(masm);
2638
2639 __ ret(lr);
2640
2641 #ifdef ASSERT
2642 __ bind(bail);
2643 __ stop("Attempting to adjust pc to skip safepoint poll but the return point is not what we expected");
2644 #endif
2645
2646 // Make sure all code is generated
2647 masm->flush();
2648
2649 // Fill-out other meta info
2650 SafepointBlob* sp_blob = SafepointBlob::create(&buffer, oop_maps, frame_size_in_words);
2651
2652 AOTCodeCache::store_code_blob(*sp_blob, AOTCodeEntry::SharedBlob, StubInfo::blob(id));
2653 return sp_blob;
2654 }
2655
2656 //
2657 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss
2658 //
2659 // Generate a stub that calls into vm to find out the proper destination
2660 // of a java call. All the argument registers are live at this point
2661 // but since this is generic code we don't know what they are and the caller
2662 // must do any gc of the args.
2663 //
2664 RuntimeStub* SharedRuntime::generate_resolve_blob(StubId id, address destination) {
2665 assert (StubRoutines::forward_exception_entry() != nullptr, "must be generated before");
2666 assert(is_resolve_id(id), "expected a resolve stub id");
2667
2668 const char* name = SharedRuntime::stub_name(id);
2669 CodeBlob* blob = AOTCodeCache::load_code_blob(AOTCodeEntry::SharedBlob, StubInfo::blob(id));
2670 if (blob != nullptr) {
2671 return blob->as_runtime_stub();
2672 }
2673
2674 // allocate space for the code
2675 ResourceMark rm;
2676 CodeBuffer buffer(name, 1000, 512);
2677 MacroAssembler* masm = new MacroAssembler(&buffer);
2678
2679 int frame_size_in_words;
2680 RegisterSaver reg_save(false /* save_vectors */);
2681
2682 OopMapSet *oop_maps = new OopMapSet();
2683 OopMap* map = nullptr;
2684
2685 int start = __ offset();
2686
2687 map = reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2688
2689 int frame_complete = __ offset();
2690
2691 {
2692 Label retaddr;
2693 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2694
2695 __ mov(c_rarg0, rthread);
2696 __ lea(rscratch1, RuntimeAddress(destination));
2697
2698 __ blr(rscratch1);
2699 __ bind(retaddr);
2700 }
2701
2702 // Set an oopmap for the call site.
2703 // We need this not only for callee-saved registers, but also for volatile
2704 // registers that the compiler might be keeping live across a safepoint.
2705
2706 oop_maps->add_gc_map( __ offset() - start, map);
2707
2708 // r0 contains the address we are going to jump to assuming no exception got installed
2709
2710 // clear last_Java_sp
2711 __ reset_last_Java_frame(false);
2712 // check for pending exceptions
2713 Label pending;
2714 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
2715 __ cbnz(rscratch1, pending);
2716
2717 // get the returned Method*
2718 __ get_vm_result_metadata(rmethod, rthread);
2719 __ str(rmethod, Address(sp, reg_save.reg_offset_in_bytes(rmethod)));
2720
2721 // r0 is where we want to jump, overwrite rscratch1 which is saved and scratch
2722 __ str(r0, Address(sp, reg_save.rscratch1_offset_in_bytes()));
2723 reg_save.restore_live_registers(masm);
2724
2725 // We are back to the original state on entry and ready to go.
2726
2727 __ br(rscratch1);
2728
2729 // Pending exception after the safepoint
2730
2731 __ bind(pending);
2732
2733 reg_save.restore_live_registers(masm);
2734
2735 // exception pending => remove activation and forward to exception handler
2736
2737 __ str(zr, Address(rthread, JavaThread::vm_result_oop_offset()));
2738
2739 __ ldr(r0, Address(rthread, Thread::pending_exception_offset()));
2740 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2741
2742 // -------------
2743 // make sure all code is generated
2744 masm->flush();
2745
2746 // return the blob
2747 // frame_size_words or bytes??
2748 RuntimeStub* rs_blob = RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true);
2749
2750 AOTCodeCache::store_code_blob(*rs_blob, AOTCodeEntry::SharedBlob, StubInfo::blob(id));
2751 return rs_blob;
2752 }
2753
2754 // Continuation point for throwing of implicit exceptions that are
2755 // not handled in the current activation. Fabricates an exception
2756 // oop and initiates normal exception dispatching in this
2757 // frame. Since we need to preserve callee-saved values (currently
2758 // only for C2, but done for C1 as well) we need a callee-saved oop
2759 // map and therefore have to make these stubs into RuntimeStubs
2760 // rather than BufferBlobs. If the compiler needs all registers to
2761 // be preserved between the fault point and the exception handler
2762 // then it must assume responsibility for that in
2763 // AbstractCompiler::continuation_for_implicit_null_exception or
2764 // continuation_for_implicit_division_by_zero_exception. All other
2765 // implicit exceptions (e.g., NullPointerException or
2766 // AbstractMethodError on entry) are either at call sites or
2767 // otherwise assume that stack unwinding will be initiated, so
2768 // caller saved registers were assumed volatile in the compiler.
2769
2770 RuntimeStub* SharedRuntime::generate_throw_exception(StubId id, address runtime_entry) {
2771 assert(is_throw_id(id), "expected a throw stub id");
2772
2773 const char* name = SharedRuntime::stub_name(id);
2774
2775 // Information about frame layout at time of blocking runtime call.
2776 // Note that we only have to preserve callee-saved registers since
2777 // the compilers are responsible for supplying a continuation point
2778 // if they expect all registers to be preserved.
2779 // n.b. aarch64 asserts that frame::arg_reg_save_area_bytes == 0
2780 enum layout {
2781 rfp_off = 0,
2782 rfp_off2,
2783 return_off,
2784 return_off2,
2785 framesize // inclusive of return address
2786 };
2787
2788 int insts_size = 512;
2789 int locs_size = 64;
2790
2791 const char* timer_msg = "SharedRuntime generate_throw_exception";
2792 TraceTime timer(timer_msg, TRACETIME_LOG(Info, startuptime));
2793
2794 CodeBlob* blob = AOTCodeCache::load_code_blob(AOTCodeEntry::SharedBlob, StubInfo::blob(id));
2795 if (blob != nullptr) {
2796 return blob->as_runtime_stub();
2797 }
2798
2799 ResourceMark rm;
2800 CodeBuffer code(name, insts_size, locs_size);
2801 OopMapSet* oop_maps = new OopMapSet();
2802 MacroAssembler* masm = new MacroAssembler(&code);
2803
2804 address start = __ pc();
2805
2806 // This is an inlined and slightly modified version of call_VM
2807 // which has the ability to fetch the return PC out of
2808 // thread-local storage and also sets up last_Java_sp slightly
2809 // differently than the real call_VM
2810
2811 __ enter(); // Save FP and LR before call
2812
2813 assert(is_even(framesize/2), "sp not 16-byte aligned");
2814
2815 // lr and fp are already in place
2816 __ sub(sp, rfp, ((uint64_t)framesize-4) << LogBytesPerInt); // prolog
2817
2818 int frame_complete = __ pc() - start;
2819
2820 // Set up last_Java_sp and last_Java_fp
2821 address the_pc = __ pc();
2822 __ set_last_Java_frame(sp, rfp, the_pc, rscratch1);
2823
2824 __ mov(c_rarg0, rthread);
2825 BLOCK_COMMENT("call runtime_entry");
2826 __ lea(rscratch1, RuntimeAddress(runtime_entry));
2827 __ blr(rscratch1);
2828
2829 // Generate oop map
2830 OopMap* map = new OopMap(framesize, 0);
2831
2832 oop_maps->add_gc_map(the_pc - start, map);
2833
2834 __ reset_last_Java_frame(true);
2835
2836 // Reinitialize the ptrue predicate register, in case the external runtime
2837 // call clobbers ptrue reg, as we may return to SVE compiled code.
2838 __ reinitialize_ptrue();
2839
2840 __ leave();
2841
2842 // check for pending exceptions
2843 #ifdef ASSERT
2844 Label L;
2845 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
2846 __ cbnz(rscratch1, L);
2847 __ should_not_reach_here();
2848 __ bind(L);
2849 #endif // ASSERT
2850 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2851
2852 // codeBlob framesize is in words (not VMRegImpl::slot_size)
2853 RuntimeStub* stub =
2854 RuntimeStub::new_runtime_stub(name,
2855 &code,
2856 frame_complete,
2857 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
2858 oop_maps, false);
2859 AOTCodeCache::store_code_blob(*stub, AOTCodeEntry::SharedBlob, StubInfo::blob(id));
2860
2861 return stub;
2862 }
2863
2864 #if INCLUDE_JFR
2865
2866 static void jfr_prologue(address the_pc, MacroAssembler* masm, Register thread) {
2867 __ set_last_Java_frame(sp, rfp, the_pc, rscratch1);
2868 __ mov(c_rarg0, thread);
2869 }
2870
2871 // The handle is dereferenced through a load barrier.
2872 static void jfr_epilogue(MacroAssembler* masm) {
2873 __ reset_last_Java_frame(true);
2874 }
2875
2876 // For c2: c_rarg0 is junk, call to runtime to write a checkpoint.
2877 // It returns a jobject handle to the event writer.
2878 // The handle is dereferenced and the return value is the event writer oop.
2879 RuntimeStub* SharedRuntime::generate_jfr_write_checkpoint() {
2880 enum layout {
2881 rbp_off,
2882 rbpH_off,
2883 return_off,
2884 return_off2,
2885 framesize // inclusive of return address
2886 };
2887
2888 int insts_size = 1024;
2889 int locs_size = 64;
2890 const char* name = SharedRuntime::stub_name(StubId::shared_jfr_write_checkpoint_id);
2891 CodeBuffer code(name, insts_size, locs_size);
2892 OopMapSet* oop_maps = new OopMapSet();
2893 MacroAssembler* masm = new MacroAssembler(&code);
2894
2895 address start = __ pc();
2896 __ enter();
2897 int frame_complete = __ pc() - start;
2898 address the_pc = __ pc();
2899 jfr_prologue(the_pc, masm, rthread);
2900 __ call_VM_leaf(CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::write_checkpoint), 1);
2901 jfr_epilogue(masm);
2902 __ resolve_global_jobject(r0, rscratch1, rscratch2);
2903 __ leave();
2904 __ ret(lr);
2905
2906 OopMap* map = new OopMap(framesize, 1); // rfp
2907 oop_maps->add_gc_map(the_pc - start, map);
2908
2909 RuntimeStub* stub = // codeBlob framesize is in words (not VMRegImpl::slot_size)
2910 RuntimeStub::new_runtime_stub(name, &code, frame_complete,
2911 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
2912 oop_maps, false);
2913 return stub;
2914 }
2915
2916 // For c2: call to return a leased buffer.
2917 RuntimeStub* SharedRuntime::generate_jfr_return_lease() {
2918 enum layout {
2919 rbp_off,
2920 rbpH_off,
2921 return_off,
2922 return_off2,
2923 framesize // inclusive of return address
2924 };
2925
2926 int insts_size = 1024;
2927 int locs_size = 64;
2928
2929 const char* name = SharedRuntime::stub_name(StubId::shared_jfr_return_lease_id);
2930 CodeBuffer code(name, insts_size, locs_size);
2931 OopMapSet* oop_maps = new OopMapSet();
2932 MacroAssembler* masm = new MacroAssembler(&code);
2933
2934 address start = __ pc();
2935 __ enter();
2936 int frame_complete = __ pc() - start;
2937 address the_pc = __ pc();
2938 jfr_prologue(the_pc, masm, rthread);
2939 __ call_VM_leaf(CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::return_lease), 1);
2940 jfr_epilogue(masm);
2941
2942 __ leave();
2943 __ ret(lr);
2944
2945 OopMap* map = new OopMap(framesize, 1); // rfp
2946 oop_maps->add_gc_map(the_pc - start, map);
2947
2948 RuntimeStub* stub = // codeBlob framesize is in words (not VMRegImpl::slot_size)
2949 RuntimeStub::new_runtime_stub(name, &code, frame_complete,
2950 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
2951 oop_maps, false);
2952 return stub;
2953 }
2954
2955 #endif // INCLUDE_JFR