1 /*
2 * Copyright (c) 2003, 2024, 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 "precompiled.hpp"
28 #include "asm/macroAssembler.hpp"
29 #include "asm/macroAssembler.inline.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/vframeArray.hpp"
53 #include "utilities/align.hpp"
54 #include "utilities/formatBuffer.hpp"
55 #include "vmreg_aarch64.inline.hpp"
56 #ifdef COMPILER1
57 #include "c1/c1_Runtime1.hpp"
58 #endif
59 #ifdef COMPILER2
60 #include "adfiles/ad_aarch64.hpp"
61 #include "opto/runtime.hpp"
62 #endif
63 #if INCLUDE_JVMCI
64 #include "jvmci/jvmciJavaClasses.hpp"
65 #endif
66
67 #define __ masm->
68
69 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
70
71 class SimpleRuntimeFrame {
72
73 public:
74
75 // Most of the runtime stubs have this simple frame layout.
76 // This class exists to make the layout shared in one place.
77 // Offsets are for compiler stack slots, which are jints.
78 enum layout {
79 // The frame sender code expects that rbp will be in the "natural" place and
80 // will override any oopMap setting for it. We must therefore force the layout
81 // so that it agrees with the frame sender code.
82 // we don't expect any arg reg save area so aarch64 asserts that
83 // frame::arg_reg_save_area_bytes == 0
84 rfp_off = 0,
85 rfp_off2,
86 return_off, return_off2,
87 framesize
88 };
89 };
90
91 // FIXME -- this is used by C1
92 class RegisterSaver {
93 const bool _save_vectors;
94 public:
95 RegisterSaver(bool save_vectors) : _save_vectors(save_vectors) {}
96
97 OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words);
98 void restore_live_registers(MacroAssembler* masm);
99
100 // Offsets into the register save area
101 // Used by deoptimization when it is managing result register
102 // values on its own
103
104 int reg_offset_in_bytes(Register r);
105 int r0_offset_in_bytes() { return reg_offset_in_bytes(r0); }
106 int rscratch1_offset_in_bytes() { return reg_offset_in_bytes(rscratch1); }
107 int v0_offset_in_bytes();
108
109 // Total stack size in bytes for saving sve predicate registers.
110 int total_sve_predicate_in_bytes();
111
112 // Capture info about frame layout
113 // Note this is only correct when not saving full vectors.
114 enum layout {
115 fpu_state_off = 0,
116 fpu_state_end = fpu_state_off + FPUStateSizeInWords - 1,
117 // The frame sender code expects that rfp will be in
118 // the "natural" place and will override any oopMap
119 // setting for it. We must therefore force the layout
120 // so that it agrees with the frame sender code.
121 r0_off = fpu_state_off + FPUStateSizeInWords,
122 rfp_off = r0_off + (Register::number_of_registers - 2) * Register::max_slots_per_register,
123 return_off = rfp_off + Register::max_slots_per_register, // slot for return address
124 reg_save_size = return_off + Register::max_slots_per_register};
125
126 };
127
128 int RegisterSaver::reg_offset_in_bytes(Register r) {
129 // The integer registers are located above the floating point
130 // registers in the stack frame pushed by save_live_registers() so the
131 // offset depends on whether we are saving full vectors, and whether
132 // those vectors are NEON or SVE.
133
134 int slots_per_vect = FloatRegister::save_slots_per_register;
135
136 #if COMPILER2_OR_JVMCI
137 if (_save_vectors) {
138 slots_per_vect = FloatRegister::slots_per_neon_register;
139
140 #ifdef COMPILER2
141 if (Matcher::supports_scalable_vector()) {
142 slots_per_vect = Matcher::scalable_vector_reg_size(T_FLOAT);
143 }
144 #endif
145 }
146 #endif
147
148 int r0_offset = v0_offset_in_bytes() + (slots_per_vect * FloatRegister::number_of_registers) * BytesPerInt;
149 return r0_offset + r->encoding() * wordSize;
150 }
151
152 int RegisterSaver::v0_offset_in_bytes() {
153 // The floating point registers are located above the predicate registers if
154 // they are present in the stack frame pushed by save_live_registers(). So the
155 // offset depends on the saved total predicate vectors in the stack frame.
156 return (total_sve_predicate_in_bytes() / VMRegImpl::stack_slot_size) * BytesPerInt;
157 }
158
159 int RegisterSaver::total_sve_predicate_in_bytes() {
160 #ifdef COMPILER2
161 if (_save_vectors && Matcher::supports_scalable_vector()) {
162 return (Matcher::scalable_vector_reg_size(T_BYTE) >> LogBitsPerByte) *
163 PRegister::number_of_registers;
164 }
165 #endif
166 return 0;
167 }
168
169 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words) {
170 bool use_sve = false;
171 int sve_vector_size_in_bytes = 0;
172 int sve_vector_size_in_slots = 0;
173 int sve_predicate_size_in_slots = 0;
174 int total_predicate_in_bytes = total_sve_predicate_in_bytes();
175 int total_predicate_in_slots = total_predicate_in_bytes / VMRegImpl::stack_slot_size;
176
177 #ifdef COMPILER2
178 use_sve = Matcher::supports_scalable_vector();
179 if (use_sve) {
180 sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
181 sve_vector_size_in_slots = Matcher::scalable_vector_reg_size(T_FLOAT);
182 sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
183 }
184 #endif
185
186 #if COMPILER2_OR_JVMCI
187 if (_save_vectors) {
188 int extra_save_slots_per_register = 0;
189 // Save upper half of vector registers
190 if (use_sve) {
191 extra_save_slots_per_register = sve_vector_size_in_slots - FloatRegister::save_slots_per_register;
192 } else {
193 extra_save_slots_per_register = FloatRegister::extra_save_slots_per_neon_register;
194 }
195 int extra_vector_bytes = extra_save_slots_per_register *
196 VMRegImpl::stack_slot_size *
197 FloatRegister::number_of_registers;
198 additional_frame_words += ((extra_vector_bytes + total_predicate_in_bytes) / wordSize);
199 }
200 #else
201 assert(!_save_vectors, "vectors are generated only by C2 and JVMCI");
202 #endif
203
204 int frame_size_in_bytes = align_up(additional_frame_words * wordSize +
205 reg_save_size * BytesPerInt, 16);
206 // OopMap frame size is in compiler stack slots (jint's) not bytes or words
207 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
208 // The caller will allocate additional_frame_words
209 int additional_frame_slots = additional_frame_words * wordSize / BytesPerInt;
210 // CodeBlob frame size is in words.
211 int frame_size_in_words = frame_size_in_bytes / wordSize;
212 *total_frame_words = frame_size_in_words;
213
214 // Save Integer and Float registers.
215 __ enter();
216 __ push_CPU_state(_save_vectors, use_sve, sve_vector_size_in_bytes, total_predicate_in_bytes);
217
218 // Set an oopmap for the call site. This oopmap will map all
219 // oop-registers and debug-info registers as callee-saved. This
220 // will allow deoptimization at this safepoint to find all possible
221 // debug-info recordings, as well as let GC find all oops.
222
223 OopMapSet *oop_maps = new OopMapSet();
224 OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
225
226 for (int i = 0; i < Register::number_of_registers; i++) {
227 Register r = as_Register(i);
228 if (i <= rfp->encoding() && r != rscratch1 && r != rscratch2) {
229 // SP offsets are in 4-byte words.
230 // Register slots are 8 bytes wide, 32 floating-point registers.
231 int sp_offset = Register::max_slots_per_register * i +
232 FloatRegister::save_slots_per_register * FloatRegister::number_of_registers;
233 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset + additional_frame_slots), r->as_VMReg());
234 }
235 }
236
237 for (int i = 0; i < FloatRegister::number_of_registers; i++) {
238 FloatRegister r = as_FloatRegister(i);
239 int sp_offset = 0;
240 if (_save_vectors) {
241 sp_offset = use_sve ? (total_predicate_in_slots + sve_vector_size_in_slots * i) :
242 (FloatRegister::slots_per_neon_register * i);
243 } else {
244 sp_offset = FloatRegister::save_slots_per_register * i;
245 }
246 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset), r->as_VMReg());
247 }
248
249 return oop_map;
250 }
251
252 void RegisterSaver::restore_live_registers(MacroAssembler* masm) {
253 #ifdef COMPILER2
254 __ pop_CPU_state(_save_vectors, Matcher::supports_scalable_vector(),
255 Matcher::scalable_vector_reg_size(T_BYTE), total_sve_predicate_in_bytes());
256 #else
257 #if !INCLUDE_JVMCI
258 assert(!_save_vectors, "vectors are generated only by C2 and JVMCI");
259 #endif
260 __ pop_CPU_state(_save_vectors);
261 #endif
262 __ ldp(rfp, lr, Address(__ post(sp, 2 * wordSize)));
263 __ authenticate_return_address();
264 }
265
266 // Is vector's size (in bytes) bigger than a size saved by default?
267 // 8 bytes vector registers are saved by default on AArch64.
268 // The SVE supported min vector size is 8 bytes and we need to save
269 // predicate registers when the vector size is 8 bytes as well.
270 bool SharedRuntime::is_wide_vector(int size) {
271 return size > 8 || (UseSVE > 0 && size >= 8);
272 }
273
274 // ---------------------------------------------------------------------------
275 // Read the array of BasicTypes from a signature, and compute where the
276 // arguments should go. Values in the VMRegPair regs array refer to 4-byte
277 // quantities. Values less than VMRegImpl::stack0 are registers, those above
278 // refer to 4-byte stack slots. All stack slots are based off of the stack pointer
279 // as framesizes are fixed.
280 // VMRegImpl::stack0 refers to the first slot 0(sp).
281 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher.
282 // Register up to Register::number_of_registers are the 64-bit
283 // integer registers.
284
285 // Note: the INPUTS in sig_bt are in units of Java argument words,
286 // which are 64-bit. The OUTPUTS are in 32-bit units.
287
288 // The Java calling convention is a "shifted" version of the C ABI.
289 // By skipping the first C ABI register we can call non-static jni
290 // methods with small numbers of arguments without having to shuffle
291 // the arguments at all. Since we control the java ABI we ought to at
292 // least get some advantage out of it.
293
294 int SharedRuntime::java_calling_convention(const BasicType *sig_bt,
295 VMRegPair *regs,
296 int total_args_passed) {
297
298 // Create the mapping between argument positions and
299 // registers.
300 static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = {
301 j_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4, j_rarg5, j_rarg6, j_rarg7
302 };
303 static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters_j] = {
304 j_farg0, j_farg1, j_farg2, j_farg3,
305 j_farg4, j_farg5, j_farg6, j_farg7
306 };
307
308
309 uint int_args = 0;
310 uint fp_args = 0;
311 uint stk_args = 0;
312
313 for (int i = 0; i < total_args_passed; i++) {
314 switch (sig_bt[i]) {
315 case T_BOOLEAN:
316 case T_CHAR:
317 case T_BYTE:
318 case T_SHORT:
319 case T_INT:
320 if (int_args < Argument::n_int_register_parameters_j) {
321 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
322 } else {
323 stk_args = align_up(stk_args, 2);
324 regs[i].set1(VMRegImpl::stack2reg(stk_args));
325 stk_args += 1;
326 }
327 break;
328 case T_VOID:
329 // halves of T_LONG or T_DOUBLE
330 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
331 regs[i].set_bad();
332 break;
333 case T_LONG:
334 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
335 // fall through
336 case T_OBJECT:
337 case T_ARRAY:
338 case T_ADDRESS:
339 if (int_args < Argument::n_int_register_parameters_j) {
340 regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
341 } else {
342 stk_args = align_up(stk_args, 2);
343 regs[i].set2(VMRegImpl::stack2reg(stk_args));
344 stk_args += 2;
345 }
346 break;
347 case T_FLOAT:
348 if (fp_args < Argument::n_float_register_parameters_j) {
349 regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
350 } else {
351 stk_args = align_up(stk_args, 2);
352 regs[i].set1(VMRegImpl::stack2reg(stk_args));
353 stk_args += 1;
354 }
355 break;
356 case T_DOUBLE:
357 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
358 if (fp_args < Argument::n_float_register_parameters_j) {
359 regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
360 } else {
361 stk_args = align_up(stk_args, 2);
362 regs[i].set2(VMRegImpl::stack2reg(stk_args));
363 stk_args += 2;
364 }
365 break;
366 default:
367 ShouldNotReachHere();
368 break;
369 }
370 }
371
372 return stk_args;
373 }
374
375 // Patch the callers callsite with entry to compiled code if it exists.
376 static void patch_callers_callsite(MacroAssembler *masm) {
377 Label L;
378 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::code_offset())));
379 __ cbz(rscratch1, L);
380
381 __ enter();
382 __ push_CPU_state();
383
384 // VM needs caller's callsite
385 // VM needs target method
386 // This needs to be a long call since we will relocate this adapter to
387 // the codeBuffer and it may not reach
388
389 #ifndef PRODUCT
390 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
391 #endif
392
393 __ mov(c_rarg0, rmethod);
394 __ mov(c_rarg1, lr);
395 __ authenticate_return_address(c_rarg1);
396 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)));
397 __ blr(rscratch1);
398
399 // Explicit isb required because fixup_callers_callsite may change the code
400 // stream.
401 __ safepoint_isb();
402
403 __ pop_CPU_state();
404 // restore sp
405 __ leave();
406 __ bind(L);
407 }
408
409 static void gen_c2i_adapter(MacroAssembler *masm,
410 int total_args_passed,
411 int comp_args_on_stack,
412 const BasicType *sig_bt,
413 const VMRegPair *regs,
414 Label& skip_fixup) {
415 // Before we get into the guts of the C2I adapter, see if we should be here
416 // at all. We've come from compiled code and are attempting to jump to the
417 // interpreter, which means the caller made a static call to get here
418 // (vcalls always get a compiled target if there is one). Check for a
419 // compiled target. If there is one, we need to patch the caller's call.
420 patch_callers_callsite(masm);
421
422 __ bind(skip_fixup);
423
424 int words_pushed = 0;
425
426 // Since all args are passed on the stack, total_args_passed *
427 // Interpreter::stackElementSize is the space we need.
428
429 int extraspace = total_args_passed * Interpreter::stackElementSize;
430
431 __ mov(r19_sender_sp, sp);
432
433 // stack is aligned, keep it that way
434 extraspace = align_up(extraspace, 2*wordSize);
435
436 if (extraspace)
437 __ sub(sp, sp, extraspace);
438
439 // Now write the args into the outgoing interpreter space
440 for (int i = 0; i < total_args_passed; i++) {
441 if (sig_bt[i] == T_VOID) {
442 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
443 continue;
444 }
445
446 // offset to start parameters
447 int st_off = (total_args_passed - i - 1) * Interpreter::stackElementSize;
448 int next_off = st_off - Interpreter::stackElementSize;
449
450 // Say 4 args:
451 // i st_off
452 // 0 32 T_LONG
453 // 1 24 T_VOID
454 // 2 16 T_OBJECT
455 // 3 8 T_BOOL
456 // - 0 return address
457 //
458 // However to make thing extra confusing. Because we can fit a Java long/double in
459 // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
460 // leaves one slot empty and only stores to a single slot. In this case the
461 // slot that is occupied is the T_VOID slot. See I said it was confusing.
462
463 VMReg r_1 = regs[i].first();
464 VMReg r_2 = regs[i].second();
465 if (!r_1->is_valid()) {
466 assert(!r_2->is_valid(), "");
467 continue;
468 }
469 if (r_1->is_stack()) {
470 // memory to memory use rscratch1
471 int ld_off = (r_1->reg2stack() * VMRegImpl::stack_slot_size
472 + extraspace
473 + words_pushed * wordSize);
474 if (!r_2->is_valid()) {
475 // sign extend??
476 __ ldrw(rscratch1, Address(sp, ld_off));
477 __ str(rscratch1, Address(sp, st_off));
478
479 } else {
480
481 __ ldr(rscratch1, Address(sp, ld_off));
482
483 // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
484 // T_DOUBLE and T_LONG use two slots in the interpreter
485 if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
486 // ld_off == LSW, ld_off+wordSize == MSW
487 // st_off == MSW, next_off == LSW
488 __ str(rscratch1, Address(sp, next_off));
489 #ifdef ASSERT
490 // Overwrite the unused slot with known junk
491 __ mov(rscratch1, (uint64_t)0xdeadffffdeadaaaaull);
492 __ str(rscratch1, Address(sp, st_off));
493 #endif /* ASSERT */
494 } else {
495 __ str(rscratch1, Address(sp, st_off));
496 }
497 }
498 } else if (r_1->is_Register()) {
499 Register r = r_1->as_Register();
500 if (!r_2->is_valid()) {
501 // must be only an int (or less ) so move only 32bits to slot
502 // why not sign extend??
503 __ str(r, Address(sp, st_off));
504 } else {
505 // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
506 // T_DOUBLE and T_LONG use two slots in the interpreter
507 if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
508 // jlong/double in gpr
509 #ifdef ASSERT
510 // Overwrite the unused slot with known junk
511 __ mov(rscratch1, (uint64_t)0xdeadffffdeadaaabull);
512 __ str(rscratch1, Address(sp, st_off));
513 #endif /* ASSERT */
514 __ str(r, Address(sp, next_off));
515 } else {
516 __ str(r, Address(sp, st_off));
517 }
518 }
519 } else {
520 assert(r_1->is_FloatRegister(), "");
521 if (!r_2->is_valid()) {
522 // only a float use just part of the slot
523 __ strs(r_1->as_FloatRegister(), Address(sp, st_off));
524 } else {
525 #ifdef ASSERT
526 // Overwrite the unused slot with known junk
527 __ mov(rscratch1, (uint64_t)0xdeadffffdeadaaacull);
528 __ str(rscratch1, Address(sp, st_off));
529 #endif /* ASSERT */
530 __ strd(r_1->as_FloatRegister(), Address(sp, next_off));
531 }
532 }
533 }
534
535 __ mov(esp, sp); // Interp expects args on caller's expression stack
536
537 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::interpreter_entry_offset())));
538 __ br(rscratch1);
539 }
540
541
542 void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,
543 int total_args_passed,
544 int comp_args_on_stack,
545 const BasicType *sig_bt,
546 const VMRegPair *regs) {
547
548 // Note: r19_sender_sp contains the senderSP on entry. We must
549 // preserve it since we may do a i2c -> c2i transition if we lose a
550 // race where compiled code goes non-entrant while we get args
551 // ready.
552
553 // Adapters are frameless.
554
555 // An i2c adapter is frameless because the *caller* frame, which is
556 // interpreted, routinely repairs its own esp (from
557 // interpreter_frame_last_sp), even if a callee has modified the
558 // stack pointer. It also recalculates and aligns sp.
559
560 // A c2i adapter is frameless because the *callee* frame, which is
561 // interpreted, routinely repairs its caller's sp (from sender_sp,
562 // which is set up via the senderSP register).
563
564 // In other words, if *either* the caller or callee is interpreted, we can
565 // get the stack pointer repaired after a call.
566
567 // This is why c2i and i2c adapters cannot be indefinitely composed.
568 // In particular, if a c2i adapter were to somehow call an i2c adapter,
569 // both caller and callee would be compiled methods, and neither would
570 // clean up the stack pointer changes performed by the two adapters.
571 // If this happens, control eventually transfers back to the compiled
572 // caller, but with an uncorrected stack, causing delayed havoc.
573
574 if (VerifyAdapterCalls &&
575 (Interpreter::code() != nullptr || StubRoutines::final_stubs_code() != nullptr)) {
576 #if 0
577 // So, let's test for cascading c2i/i2c adapters right now.
578 // assert(Interpreter::contains($return_addr) ||
579 // StubRoutines::contains($return_addr),
580 // "i2c adapter must return to an interpreter frame");
581 __ block_comment("verify_i2c { ");
582 Label L_ok;
583 if (Interpreter::code() != nullptr) {
584 range_check(masm, rax, r11,
585 Interpreter::code()->code_start(), Interpreter::code()->code_end(),
586 L_ok);
587 }
588 if (StubRoutines::initial_stubs_code() != nullptr) {
589 range_check(masm, rax, r11,
590 StubRoutines::initial_stubs_code()->code_begin(),
591 StubRoutines::initial_stubs_code()->code_end(),
592 L_ok);
593 }
594 if (StubRoutines::final_stubs_code() != nullptr) {
595 range_check(masm, rax, r11,
596 StubRoutines::final_stubs_code()->code_begin(),
597 StubRoutines::final_stubs_code()->code_end(),
598 L_ok);
599 }
600 const char* msg = "i2c adapter must return to an interpreter frame";
601 __ block_comment(msg);
602 __ stop(msg);
603 __ bind(L_ok);
604 __ block_comment("} verify_i2ce ");
605 #endif
606 }
607
608 // Cut-out for having no stack args.
609 int comp_words_on_stack = align_up(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord;
610 if (comp_args_on_stack) {
611 __ sub(rscratch1, sp, comp_words_on_stack * wordSize);
612 __ andr(sp, rscratch1, -16);
613 }
614
615 // Will jump to the compiled code just as if compiled code was doing it.
616 // Pre-load the register-jump target early, to schedule it better.
617 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::from_compiled_offset())));
618
619 #if INCLUDE_JVMCI
620 if (EnableJVMCI) {
621 // check if this call should be routed towards a specific entry point
622 __ ldr(rscratch2, Address(rthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
623 Label no_alternative_target;
624 __ cbz(rscratch2, no_alternative_target);
625 __ mov(rscratch1, rscratch2);
626 __ str(zr, Address(rthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
627 __ bind(no_alternative_target);
628 }
629 #endif // INCLUDE_JVMCI
630
631 // Now generate the shuffle code.
632 for (int i = 0; i < total_args_passed; i++) {
633 if (sig_bt[i] == T_VOID) {
634 assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
635 continue;
636 }
637
638 // Pick up 0, 1 or 2 words from SP+offset.
639
640 assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
641 "scrambled load targets?");
642 // Load in argument order going down.
643 int ld_off = (total_args_passed - i - 1)*Interpreter::stackElementSize;
644 // Point to interpreter value (vs. tag)
645 int next_off = ld_off - Interpreter::stackElementSize;
646 //
647 //
648 //
649 VMReg r_1 = regs[i].first();
650 VMReg r_2 = regs[i].second();
651 if (!r_1->is_valid()) {
652 assert(!r_2->is_valid(), "");
653 continue;
654 }
655 if (r_1->is_stack()) {
656 // Convert stack slot to an SP offset (+ wordSize to account for return address )
657 int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size;
658 if (!r_2->is_valid()) {
659 // sign extend???
660 __ ldrsw(rscratch2, Address(esp, ld_off));
661 __ str(rscratch2, Address(sp, st_off));
662 } else {
663 //
664 // We are using two optoregs. This can be either T_OBJECT,
665 // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
666 // two slots but only uses one for thr T_LONG or T_DOUBLE case
667 // So we must adjust where to pick up the data to match the
668 // interpreter.
669 //
670 // Interpreter local[n] == MSW, local[n+1] == LSW however locals
671 // are accessed as negative so LSW is at LOW address
672
673 // ld_off is MSW so get LSW
674 const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
675 next_off : ld_off;
676 __ ldr(rscratch2, Address(esp, offset));
677 // st_off is LSW (i.e. reg.first())
678 __ str(rscratch2, Address(sp, st_off));
679 }
680 } else if (r_1->is_Register()) { // Register argument
681 Register r = r_1->as_Register();
682 if (r_2->is_valid()) {
683 //
684 // We are using two VMRegs. This can be either T_OBJECT,
685 // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
686 // two slots but only uses one for thr T_LONG or T_DOUBLE case
687 // So we must adjust where to pick up the data to match the
688 // interpreter.
689
690 const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
691 next_off : ld_off;
692
693 // this can be a misaligned move
694 __ ldr(r, Address(esp, offset));
695 } else {
696 // sign extend and use a full word?
697 __ ldrw(r, Address(esp, ld_off));
698 }
699 } else {
700 if (!r_2->is_valid()) {
701 __ ldrs(r_1->as_FloatRegister(), Address(esp, ld_off));
702 } else {
703 __ ldrd(r_1->as_FloatRegister(), Address(esp, next_off));
704 }
705 }
706 }
707
708 __ mov(rscratch2, rscratch1);
709 __ push_cont_fastpath(rthread); // Set JavaThread::_cont_fastpath to the sp of the oldest interpreted frame we know about; kills rscratch1
710 __ mov(rscratch1, rscratch2);
711
712 // 6243940 We might end up in handle_wrong_method if
713 // the callee is deoptimized as we race thru here. If that
714 // happens we don't want to take a safepoint because the
715 // caller frame will look interpreted and arguments are now
716 // "compiled" so it is much better to make this transition
717 // invisible to the stack walking code. Unfortunately if
718 // we try and find the callee by normal means a safepoint
719 // is possible. So we stash the desired callee in the thread
720 // and the vm will find there should this case occur.
721
722 __ str(rmethod, Address(rthread, JavaThread::callee_target_offset()));
723
724 __ br(rscratch1);
725 }
726
727 // ---------------------------------------------------------------
728 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
729 int total_args_passed,
730 int comp_args_on_stack,
731 const BasicType *sig_bt,
732 const VMRegPair *regs,
733 AdapterFingerPrint* fingerprint) {
734 address i2c_entry = __ pc();
735
736 gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
737
738 address c2i_unverified_entry = __ pc();
739 Label skip_fixup;
740
741 Register data = rscratch2;
742 Register receiver = j_rarg0;
743 Register tmp = r10; // A call-clobbered register not used for arg passing
744
745 // -------------------------------------------------------------------------
746 // Generate a C2I adapter. On entry we know rmethod holds the Method* during calls
747 // to the interpreter. The args start out packed in the compiled layout. They
748 // need to be unpacked into the interpreter layout. This will almost always
749 // require some stack space. We grow the current (compiled) stack, then repack
750 // the args. We finally end in a jump to the generic interpreter entry point.
751 // On exit from the interpreter, the interpreter will restore our SP (lest the
752 // compiled code, which relies solely on SP and not FP, get sick).
753
754 {
755 __ block_comment("c2i_unverified_entry {");
756 // Method might have been compiled since the call site was patched to
757 // interpreted; if that is the case treat it as a miss so we can get
758 // the call site corrected.
759 __ ic_check(1 /* end_alignment */);
760 __ ldr(rmethod, Address(data, CompiledICData::speculated_method_offset()));
761
762 __ ldr(rscratch1, Address(rmethod, in_bytes(Method::code_offset())));
763 __ cbz(rscratch1, skip_fixup);
764 __ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
765 __ block_comment("} c2i_unverified_entry");
766 }
767
768 address c2i_entry = __ pc();
769
770 // Class initialization barrier for static methods
771 address c2i_no_clinit_check_entry = nullptr;
772 if (VM_Version::supports_fast_class_init_checks()) {
773 Label L_skip_barrier;
774
775 { // Bypass the barrier for non-static methods
776 __ ldrw(rscratch1, Address(rmethod, Method::access_flags_offset()));
777 __ andsw(zr, rscratch1, JVM_ACC_STATIC);
778 __ br(Assembler::EQ, L_skip_barrier); // non-static
779 }
780
781 __ load_method_holder(rscratch2, rmethod);
782 __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
783 __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
784
785 __ bind(L_skip_barrier);
786 c2i_no_clinit_check_entry = __ pc();
787 }
788
789 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
790 bs->c2i_entry_barrier(masm);
791
792 gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
793
794 return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry, c2i_no_clinit_check_entry);
795 }
796
797 static int c_calling_convention_priv(const BasicType *sig_bt,
798 VMRegPair *regs,
799 int total_args_passed) {
800
801 // We return the amount of VMRegImpl stack slots we need to reserve for all
802 // the arguments NOT counting out_preserve_stack_slots.
803
804 static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
805 c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, c_rarg5, c_rarg6, c_rarg7
806 };
807 static const FloatRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
808 c_farg0, c_farg1, c_farg2, c_farg3,
809 c_farg4, c_farg5, c_farg6, c_farg7
810 };
811
812 uint int_args = 0;
813 uint fp_args = 0;
814 uint stk_args = 0; // inc by 2 each time
815
816 for (int i = 0; i < total_args_passed; i++) {
817 switch (sig_bt[i]) {
818 case T_BOOLEAN:
819 case T_CHAR:
820 case T_BYTE:
821 case T_SHORT:
822 case T_INT:
823 if (int_args < Argument::n_int_register_parameters_c) {
824 regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
825 } else {
826 #ifdef __APPLE__
827 // Less-than word types are stored one after another.
828 // The code is unable to handle this so bailout.
829 return -1;
830 #endif
831 regs[i].set1(VMRegImpl::stack2reg(stk_args));
832 stk_args += 2;
833 }
834 break;
835 case T_LONG:
836 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
837 // fall through
838 case T_OBJECT:
839 case T_ARRAY:
840 case T_ADDRESS:
841 case T_METADATA:
842 if (int_args < Argument::n_int_register_parameters_c) {
843 regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
844 } else {
845 regs[i].set2(VMRegImpl::stack2reg(stk_args));
846 stk_args += 2;
847 }
848 break;
849 case T_FLOAT:
850 if (fp_args < Argument::n_float_register_parameters_c) {
851 regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
852 } else {
853 #ifdef __APPLE__
854 // Less-than word types are stored one after another.
855 // The code is unable to handle this so bailout.
856 return -1;
857 #endif
858 regs[i].set1(VMRegImpl::stack2reg(stk_args));
859 stk_args += 2;
860 }
861 break;
862 case T_DOUBLE:
863 assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
864 if (fp_args < Argument::n_float_register_parameters_c) {
865 regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
866 } else {
867 regs[i].set2(VMRegImpl::stack2reg(stk_args));
868 stk_args += 2;
869 }
870 break;
871 case T_VOID: // Halves of longs and doubles
872 assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
873 regs[i].set_bad();
874 break;
875 default:
876 ShouldNotReachHere();
877 break;
878 }
879 }
880
881 return stk_args;
882 }
883
884 int SharedRuntime::vector_calling_convention(VMRegPair *regs,
885 uint num_bits,
886 uint total_args_passed) {
887 Unimplemented();
888 return 0;
889 }
890
891 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
892 VMRegPair *regs,
893 int total_args_passed)
894 {
895 int result = c_calling_convention_priv(sig_bt, regs, total_args_passed);
896 guarantee(result >= 0, "Unsupported arguments configuration");
897 return result;
898 }
899
900
901 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
902 // We always ignore the frame_slots arg and just use the space just below frame pointer
903 // which by this time is free to use
904 switch (ret_type) {
905 case T_FLOAT:
906 __ strs(v0, Address(rfp, -wordSize));
907 break;
908 case T_DOUBLE:
909 __ strd(v0, Address(rfp, -wordSize));
910 break;
911 case T_VOID: break;
912 default: {
913 __ str(r0, Address(rfp, -wordSize));
914 }
915 }
916 }
917
918 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
919 // We always ignore the frame_slots arg and just use the space just below frame pointer
920 // which by this time is free to use
921 switch (ret_type) {
922 case T_FLOAT:
923 __ ldrs(v0, Address(rfp, -wordSize));
924 break;
925 case T_DOUBLE:
926 __ ldrd(v0, Address(rfp, -wordSize));
927 break;
928 case T_VOID: break;
929 default: {
930 __ ldr(r0, Address(rfp, -wordSize));
931 }
932 }
933 }
934 static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
935 RegSet x;
936 for ( int i = first_arg ; i < arg_count ; i++ ) {
937 if (args[i].first()->is_Register()) {
938 x = x + args[i].first()->as_Register();
939 } else if (args[i].first()->is_FloatRegister()) {
940 __ strd(args[i].first()->as_FloatRegister(), Address(__ pre(sp, -2 * wordSize)));
941 }
942 }
943 __ push(x, sp);
944 }
945
946 static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
947 RegSet x;
948 for ( int i = first_arg ; i < arg_count ; i++ ) {
949 if (args[i].first()->is_Register()) {
950 x = x + args[i].first()->as_Register();
951 } else {
952 ;
953 }
954 }
955 __ pop(x, sp);
956 for ( int i = arg_count - 1 ; i >= first_arg ; i-- ) {
957 if (args[i].first()->is_Register()) {
958 ;
959 } else if (args[i].first()->is_FloatRegister()) {
960 __ ldrd(args[i].first()->as_FloatRegister(), Address(__ post(sp, 2 * wordSize)));
961 }
962 }
963 }
964
965 static void verify_oop_args(MacroAssembler* masm,
966 const methodHandle& method,
967 const BasicType* sig_bt,
968 const VMRegPair* regs) {
969 Register temp_reg = r19; // not part of any compiled calling seq
970 if (VerifyOops) {
971 for (int i = 0; i < method->size_of_parameters(); i++) {
972 if (sig_bt[i] == T_OBJECT ||
973 sig_bt[i] == T_ARRAY) {
974 VMReg r = regs[i].first();
975 assert(r->is_valid(), "bad oop arg");
976 if (r->is_stack()) {
977 __ ldr(temp_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
978 __ verify_oop(temp_reg);
979 } else {
980 __ verify_oop(r->as_Register());
981 }
982 }
983 }
984 }
985 }
986
987 // on exit, sp points to the ContinuationEntry
988 static OopMap* continuation_enter_setup(MacroAssembler* masm, int& stack_slots) {
989 assert(ContinuationEntry::size() % VMRegImpl::stack_slot_size == 0, "");
990 assert(in_bytes(ContinuationEntry::cont_offset()) % VMRegImpl::stack_slot_size == 0, "");
991 assert(in_bytes(ContinuationEntry::chunk_offset()) % VMRegImpl::stack_slot_size == 0, "");
992
993 stack_slots += (int)ContinuationEntry::size()/wordSize;
994 __ sub(sp, sp, (int)ContinuationEntry::size()); // place Continuation metadata
995
996 OopMap* map = new OopMap(((int)ContinuationEntry::size() + wordSize)/ VMRegImpl::stack_slot_size, 0 /* arg_slots*/);
997
998 __ ldr(rscratch1, Address(rthread, JavaThread::cont_entry_offset()));
999 __ str(rscratch1, Address(sp, ContinuationEntry::parent_offset()));
1000 __ mov(rscratch1, sp); // we can't use sp as the source in str
1001 __ str(rscratch1, Address(rthread, JavaThread::cont_entry_offset()));
1002
1003 return map;
1004 }
1005
1006 // on entry c_rarg1 points to the continuation
1007 // sp points to ContinuationEntry
1008 // c_rarg3 -- isVirtualThread
1009 static void fill_continuation_entry(MacroAssembler* masm) {
1010 #ifdef ASSERT
1011 __ movw(rscratch1, ContinuationEntry::cookie_value());
1012 __ strw(rscratch1, Address(sp, ContinuationEntry::cookie_offset()));
1013 #endif
1014
1015 __ str (c_rarg1, Address(sp, ContinuationEntry::cont_offset()));
1016 __ strw(c_rarg3, Address(sp, ContinuationEntry::flags_offset()));
1017 __ str (zr, Address(sp, ContinuationEntry::chunk_offset()));
1018 __ strw(zr, Address(sp, ContinuationEntry::argsize_offset()));
1019 __ strw(zr, Address(sp, ContinuationEntry::pin_count_offset()));
1020
1021 __ ldr(rscratch1, Address(rthread, JavaThread::cont_fastpath_offset()));
1022 __ str(rscratch1, Address(sp, ContinuationEntry::parent_cont_fastpath_offset()));
1023 __ ldr(rscratch1, Address(rthread, JavaThread::held_monitor_count_offset()));
1024 __ str(rscratch1, Address(sp, ContinuationEntry::parent_held_monitor_count_offset()));
1025
1026 __ str(zr, Address(rthread, JavaThread::cont_fastpath_offset()));
1027 __ str(zr, Address(rthread, JavaThread::held_monitor_count_offset()));
1028 }
1029
1030 // on entry, sp points to the ContinuationEntry
1031 // on exit, rfp points to the spilled rfp in the entry frame
1032 static void continuation_enter_cleanup(MacroAssembler* masm) {
1033 #ifndef PRODUCT
1034 Label OK;
1035 __ ldr(rscratch1, Address(rthread, JavaThread::cont_entry_offset()));
1036 __ cmp(sp, rscratch1);
1037 __ br(Assembler::EQ, OK);
1038 __ stop("incorrect sp1");
1039 __ bind(OK);
1040 #endif
1041
1042 __ ldr(rscratch1, Address(sp, ContinuationEntry::parent_cont_fastpath_offset()));
1043 __ str(rscratch1, Address(rthread, JavaThread::cont_fastpath_offset()));
1044 __ ldr(rscratch1, Address(sp, ContinuationEntry::parent_held_monitor_count_offset()));
1045 __ str(rscratch1, Address(rthread, JavaThread::held_monitor_count_offset()));
1046
1047 __ ldr(rscratch2, Address(sp, ContinuationEntry::parent_offset()));
1048 __ str(rscratch2, Address(rthread, JavaThread::cont_entry_offset()));
1049 __ add(rfp, sp, (int)ContinuationEntry::size());
1050 }
1051
1052 // enterSpecial(Continuation c, boolean isContinue, boolean isVirtualThread)
1053 // On entry: c_rarg1 -- the continuation object
1054 // c_rarg2 -- isContinue
1055 // c_rarg3 -- isVirtualThread
1056 static void gen_continuation_enter(MacroAssembler* masm,
1057 const methodHandle& method,
1058 const BasicType* sig_bt,
1059 const VMRegPair* regs,
1060 int& exception_offset,
1061 OopMapSet*oop_maps,
1062 int& frame_complete,
1063 int& stack_slots,
1064 int& interpreted_entry_offset,
1065 int& compiled_entry_offset) {
1066 //verify_oop_args(masm, method, sig_bt, regs);
1067 Address resolve(SharedRuntime::get_resolve_static_call_stub(), relocInfo::static_call_type);
1068
1069 address start = __ pc();
1070
1071 Label call_thaw, exit;
1072
1073 // i2i entry used at interp_only_mode only
1074 interpreted_entry_offset = __ pc() - start;
1075 {
1076
1077 #ifdef ASSERT
1078 Label is_interp_only;
1079 __ ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
1080 __ cbnzw(rscratch1, is_interp_only);
1081 __ stop("enterSpecial interpreter entry called when not in interp_only_mode");
1082 __ bind(is_interp_only);
1083 #endif
1084
1085 // Read interpreter arguments into registers (this is an ad-hoc i2c adapter)
1086 __ ldr(c_rarg1, Address(esp, Interpreter::stackElementSize*2));
1087 __ ldr(c_rarg2, Address(esp, Interpreter::stackElementSize*1));
1088 __ ldr(c_rarg3, Address(esp, Interpreter::stackElementSize*0));
1089 __ push_cont_fastpath(rthread);
1090
1091 __ enter();
1092 stack_slots = 2; // will be adjusted in setup
1093 OopMap* map = continuation_enter_setup(masm, stack_slots);
1094 // The frame is complete here, but we only record it for the compiled entry, so the frame would appear unsafe,
1095 // but that's okay because at the very worst we'll miss an async sample, but we're in interp_only_mode anyway.
1096
1097 fill_continuation_entry(masm);
1098
1099 __ cbnz(c_rarg2, call_thaw);
1100
1101 const address tr_call = __ trampoline_call(resolve);
1102 if (tr_call == nullptr) {
1103 fatal("CodeCache is full at gen_continuation_enter");
1104 }
1105
1106 oop_maps->add_gc_map(__ pc() - start, map);
1107 __ post_call_nop();
1108
1109 __ b(exit);
1110
1111 CodeBuffer* cbuf = masm->code_section()->outer();
1112 address stub = CompiledDirectCall::emit_to_interp_stub(*cbuf, tr_call);
1113 if (stub == nullptr) {
1114 fatal("CodeCache is full at gen_continuation_enter");
1115 }
1116 }
1117
1118 // compiled entry
1119 __ align(CodeEntryAlignment);
1120 compiled_entry_offset = __ pc() - start;
1121
1122 __ enter();
1123 stack_slots = 2; // will be adjusted in setup
1124 OopMap* map = continuation_enter_setup(masm, stack_slots);
1125 frame_complete = __ pc() - start;
1126
1127 fill_continuation_entry(masm);
1128
1129 __ cbnz(c_rarg2, call_thaw);
1130
1131 const address tr_call = __ trampoline_call(resolve);
1132 if (tr_call == nullptr) {
1133 fatal("CodeCache is full at gen_continuation_enter");
1134 }
1135
1136 oop_maps->add_gc_map(__ pc() - start, map);
1137 __ post_call_nop();
1138
1139 __ b(exit);
1140
1141 __ bind(call_thaw);
1142
1143 __ rt_call(CAST_FROM_FN_PTR(address, StubRoutines::cont_thaw()));
1144 oop_maps->add_gc_map(__ pc() - start, map->deep_copy());
1145 ContinuationEntry::_return_pc_offset = __ pc() - start;
1146 __ post_call_nop();
1147
1148 __ bind(exit);
1149 continuation_enter_cleanup(masm);
1150 __ leave();
1151 __ ret(lr);
1152
1153 /// exception handling
1154
1155 exception_offset = __ pc() - start;
1156 {
1157 __ mov(r19, r0); // save return value contaning the exception oop in callee-saved R19
1158
1159 continuation_enter_cleanup(masm);
1160
1161 __ ldr(c_rarg1, Address(rfp, wordSize)); // return address
1162 __ authenticate_return_address(c_rarg1);
1163 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rthread, c_rarg1);
1164
1165 // see OptoRuntime::generate_exception_blob: r0 -- exception oop, r3 -- exception pc
1166
1167 __ mov(r1, r0); // the exception handler
1168 __ mov(r0, r19); // restore return value contaning the exception oop
1169 __ verify_oop(r0);
1170
1171 __ leave();
1172 __ mov(r3, lr);
1173 __ br(r1); // the exception handler
1174 }
1175
1176 CodeBuffer* cbuf = masm->code_section()->outer();
1177 address stub = CompiledDirectCall::emit_to_interp_stub(*cbuf, tr_call);
1178 if (stub == nullptr) {
1179 fatal("CodeCache is full at gen_continuation_enter");
1180 }
1181 }
1182
1183 static void gen_continuation_yield(MacroAssembler* masm,
1184 const methodHandle& method,
1185 const BasicType* sig_bt,
1186 const VMRegPair* regs,
1187 OopMapSet* oop_maps,
1188 int& frame_complete,
1189 int& stack_slots,
1190 int& compiled_entry_offset) {
1191 enum layout {
1192 rfp_off1,
1193 rfp_off2,
1194 lr_off,
1195 lr_off2,
1196 framesize // inclusive of return address
1197 };
1198 // assert(is_even(framesize/2), "sp not 16-byte aligned");
1199 stack_slots = framesize / VMRegImpl::slots_per_word;
1200 assert(stack_slots == 2, "recheck layout");
1201
1202 address start = __ pc();
1203
1204 compiled_entry_offset = __ pc() - start;
1205 __ enter();
1206
1207 __ mov(c_rarg1, sp);
1208
1209 frame_complete = __ pc() - start;
1210 address the_pc = __ pc();
1211
1212 __ 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
1213
1214 __ mov(c_rarg0, rthread);
1215 __ set_last_Java_frame(sp, rfp, the_pc, rscratch1);
1216 __ call_VM_leaf(Continuation::freeze_entry(), 2);
1217 __ reset_last_Java_frame(true);
1218
1219 Label pinned;
1220
1221 __ cbnz(r0, pinned);
1222
1223 // We've succeeded, set sp to the ContinuationEntry
1224 __ ldr(rscratch1, Address(rthread, JavaThread::cont_entry_offset()));
1225 __ mov(sp, rscratch1);
1226 continuation_enter_cleanup(masm);
1227
1228 __ bind(pinned); // pinned -- return to caller
1229
1230 // handle pending exception thrown by freeze
1231 __ ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
1232 Label ok;
1233 __ cbz(rscratch1, ok);
1234 __ leave();
1235 __ lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
1236 __ br(rscratch1);
1237 __ bind(ok);
1238
1239 __ leave();
1240 __ ret(lr);
1241
1242 OopMap* map = new OopMap(framesize, 1);
1243 oop_maps->add_gc_map(the_pc - start, map);
1244 }
1245
1246 static void gen_special_dispatch(MacroAssembler* masm,
1247 const methodHandle& method,
1248 const BasicType* sig_bt,
1249 const VMRegPair* regs) {
1250 verify_oop_args(masm, method, sig_bt, regs);
1251 vmIntrinsics::ID iid = method->intrinsic_id();
1252
1253 // Now write the args into the outgoing interpreter space
1254 bool has_receiver = false;
1255 Register receiver_reg = noreg;
1256 int member_arg_pos = -1;
1257 Register member_reg = noreg;
1258 int ref_kind = MethodHandles::signature_polymorphic_intrinsic_ref_kind(iid);
1259 if (ref_kind != 0) {
1260 member_arg_pos = method->size_of_parameters() - 1; // trailing MemberName argument
1261 member_reg = r19; // known to be free at this point
1262 has_receiver = MethodHandles::ref_kind_has_receiver(ref_kind);
1263 } else if (iid == vmIntrinsics::_invokeBasic) {
1264 has_receiver = true;
1265 } else if (iid == vmIntrinsics::_linkToNative) {
1266 member_arg_pos = method->size_of_parameters() - 1; // trailing NativeEntryPoint argument
1267 member_reg = r19; // known to be free at this point
1268 } else {
1269 fatal("unexpected intrinsic id %d", vmIntrinsics::as_int(iid));
1270 }
1271
1272 if (member_reg != noreg) {
1273 // Load the member_arg into register, if necessary.
1274 SharedRuntime::check_member_name_argument_is_last_argument(method, sig_bt, regs);
1275 VMReg r = regs[member_arg_pos].first();
1276 if (r->is_stack()) {
1277 __ ldr(member_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
1278 } else {
1279 // no data motion is needed
1280 member_reg = r->as_Register();
1281 }
1282 }
1283
1284 if (has_receiver) {
1285 // Make sure the receiver is loaded into a register.
1286 assert(method->size_of_parameters() > 0, "oob");
1287 assert(sig_bt[0] == T_OBJECT, "receiver argument must be an object");
1288 VMReg r = regs[0].first();
1289 assert(r->is_valid(), "bad receiver arg");
1290 if (r->is_stack()) {
1291 // Porting note: This assumes that compiled calling conventions always
1292 // pass the receiver oop in a register. If this is not true on some
1293 // platform, pick a temp and load the receiver from stack.
1294 fatal("receiver always in a register");
1295 receiver_reg = r2; // known to be free at this point
1296 __ ldr(receiver_reg, Address(sp, r->reg2stack() * VMRegImpl::stack_slot_size));
1297 } else {
1298 // no data motion is needed
1299 receiver_reg = r->as_Register();
1300 }
1301 }
1302
1303 // Figure out which address we are really jumping to:
1304 MethodHandles::generate_method_handle_dispatch(masm, iid,
1305 receiver_reg, member_reg, /*for_compiler_entry:*/ true);
1306 }
1307
1308 // ---------------------------------------------------------------------------
1309 // Generate a native wrapper for a given method. The method takes arguments
1310 // in the Java compiled code convention, marshals them to the native
1311 // convention (handlizes oops, etc), transitions to native, makes the call,
1312 // returns to java state (possibly blocking), unhandlizes any result and
1313 // returns.
1314 //
1315 // Critical native functions are a shorthand for the use of
1316 // GetPrimtiveArrayCritical and disallow the use of any other JNI
1317 // functions. The wrapper is expected to unpack the arguments before
1318 // passing them to the callee. Critical native functions leave the state _in_Java,
1319 // since they block out GC.
1320 // Some other parts of JNI setup are skipped like the tear down of the JNI handle
1321 // block and the check for pending exceptions it's impossible for them
1322 // to be thrown.
1323 //
1324 nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
1325 const methodHandle& method,
1326 int compile_id,
1327 BasicType* in_sig_bt,
1328 VMRegPair* in_regs,
1329 BasicType ret_type) {
1330 if (method->is_continuation_native_intrinsic()) {
1331 int exception_offset = -1;
1332 OopMapSet* oop_maps = new OopMapSet();
1333 int frame_complete = -1;
1334 int stack_slots = -1;
1335 int interpreted_entry_offset = -1;
1336 int vep_offset = -1;
1337 if (method->is_continuation_enter_intrinsic()) {
1338 gen_continuation_enter(masm,
1339 method,
1340 in_sig_bt,
1341 in_regs,
1342 exception_offset,
1343 oop_maps,
1344 frame_complete,
1345 stack_slots,
1346 interpreted_entry_offset,
1347 vep_offset);
1348 } else if (method->is_continuation_yield_intrinsic()) {
1349 gen_continuation_yield(masm,
1350 method,
1351 in_sig_bt,
1352 in_regs,
1353 oop_maps,
1354 frame_complete,
1355 stack_slots,
1356 vep_offset);
1357 } else {
1358 guarantee(false, "Unknown Continuation native intrinsic");
1359 }
1360
1361 #ifdef ASSERT
1362 if (method->is_continuation_enter_intrinsic()) {
1363 assert(interpreted_entry_offset != -1, "Must be set");
1364 assert(exception_offset != -1, "Must be set");
1365 } else {
1366 assert(interpreted_entry_offset == -1, "Must be unset");
1367 assert(exception_offset == -1, "Must be unset");
1368 }
1369 assert(frame_complete != -1, "Must be set");
1370 assert(stack_slots != -1, "Must be set");
1371 assert(vep_offset != -1, "Must be set");
1372 #endif
1373
1374 __ flush();
1375 nmethod* nm = nmethod::new_native_nmethod(method,
1376 compile_id,
1377 masm->code(),
1378 vep_offset,
1379 frame_complete,
1380 stack_slots,
1381 in_ByteSize(-1),
1382 in_ByteSize(-1),
1383 oop_maps,
1384 exception_offset);
1385 if (nm == nullptr) return nm;
1386 if (method->is_continuation_enter_intrinsic()) {
1387 ContinuationEntry::set_enter_code(nm, interpreted_entry_offset);
1388 } else if (method->is_continuation_yield_intrinsic()) {
1389 _cont_doYield_stub = nm;
1390 } else {
1391 guarantee(false, "Unknown Continuation native intrinsic");
1392 }
1393 return nm;
1394 }
1395
1396 if (method->is_method_handle_intrinsic()) {
1397 vmIntrinsics::ID iid = method->intrinsic_id();
1398 intptr_t start = (intptr_t)__ pc();
1399 int vep_offset = ((intptr_t)__ pc()) - start;
1400
1401 // First instruction must be a nop as it may need to be patched on deoptimisation
1402 __ nop();
1403 gen_special_dispatch(masm,
1404 method,
1405 in_sig_bt,
1406 in_regs);
1407 int frame_complete = ((intptr_t)__ pc()) - start; // not complete, period
1408 __ flush();
1409 int stack_slots = SharedRuntime::out_preserve_stack_slots(); // no out slots at all, actually
1410 return nmethod::new_native_nmethod(method,
1411 compile_id,
1412 masm->code(),
1413 vep_offset,
1414 frame_complete,
1415 stack_slots / VMRegImpl::slots_per_word,
1416 in_ByteSize(-1),
1417 in_ByteSize(-1),
1418 nullptr);
1419 }
1420 address native_func = method->native_function();
1421 assert(native_func != nullptr, "must have function");
1422
1423 // An OopMap for lock (and class if static)
1424 OopMapSet *oop_maps = new OopMapSet();
1425 intptr_t start = (intptr_t)__ pc();
1426
1427 // We have received a description of where all the java arg are located
1428 // on entry to the wrapper. We need to convert these args to where
1429 // the jni function will expect them. To figure out where they go
1430 // we convert the java signature to a C signature by inserting
1431 // the hidden arguments as arg[0] and possibly arg[1] (static method)
1432
1433 const int total_in_args = method->size_of_parameters();
1434 int total_c_args = total_in_args + (method->is_static() ? 2 : 1);
1435
1436 BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args);
1437 VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args);
1438 BasicType* in_elem_bt = nullptr;
1439
1440 int argc = 0;
1441 out_sig_bt[argc++] = T_ADDRESS;
1442 if (method->is_static()) {
1443 out_sig_bt[argc++] = T_OBJECT;
1444 }
1445
1446 for (int i = 0; i < total_in_args ; i++ ) {
1447 out_sig_bt[argc++] = in_sig_bt[i];
1448 }
1449
1450 // Now figure out where the args must be stored and how much stack space
1451 // they require.
1452 int out_arg_slots;
1453 out_arg_slots = c_calling_convention_priv(out_sig_bt, out_regs, total_c_args);
1454
1455 if (out_arg_slots < 0) {
1456 return nullptr;
1457 }
1458
1459 // Compute framesize for the wrapper. We need to handlize all oops in
1460 // incoming registers
1461
1462 // Calculate the total number of stack slots we will need.
1463
1464 // First count the abi requirement plus all of the outgoing args
1465 int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
1466
1467 // Now the space for the inbound oop handle area
1468 int total_save_slots = 8 * VMRegImpl::slots_per_word; // 8 arguments passed in registers
1469
1470 int oop_handle_offset = stack_slots;
1471 stack_slots += total_save_slots;
1472
1473 // Now any space we need for handlizing a klass if static method
1474
1475 int klass_slot_offset = 0;
1476 int klass_offset = -1;
1477 int lock_slot_offset = 0;
1478 bool is_static = false;
1479
1480 if (method->is_static()) {
1481 klass_slot_offset = stack_slots;
1482 stack_slots += VMRegImpl::slots_per_word;
1483 klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size;
1484 is_static = true;
1485 }
1486
1487 // Plus a lock if needed
1488
1489 if (method->is_synchronized()) {
1490 lock_slot_offset = stack_slots;
1491 stack_slots += VMRegImpl::slots_per_word;
1492 }
1493
1494 // Now a place (+2) to save return values or temp during shuffling
1495 // + 4 for return address (which we own) and saved rfp
1496 stack_slots += 6;
1497
1498 // Ok The space we have allocated will look like:
1499 //
1500 //
1501 // FP-> | |
1502 // |---------------------|
1503 // | 2 slots for moves |
1504 // |---------------------|
1505 // | lock box (if sync) |
1506 // |---------------------| <- lock_slot_offset
1507 // | klass (if static) |
1508 // |---------------------| <- klass_slot_offset
1509 // | oopHandle area |
1510 // |---------------------| <- oop_handle_offset (8 java arg registers)
1511 // | outbound memory |
1512 // | based arguments |
1513 // | |
1514 // |---------------------|
1515 // | |
1516 // SP-> | out_preserved_slots |
1517 //
1518 //
1519
1520
1521 // Now compute actual number of stack words we need rounding to make
1522 // stack properly aligned.
1523 stack_slots = align_up(stack_slots, StackAlignmentInSlots);
1524
1525 int stack_size = stack_slots * VMRegImpl::stack_slot_size;
1526
1527 // First thing make an ic check to see if we should even be here
1528
1529 // We are free to use all registers as temps without saving them and
1530 // restoring them except rfp. rfp is the only callee save register
1531 // as far as the interpreter and the compiler(s) are concerned.
1532
1533 const Register receiver = j_rarg0;
1534
1535 Label exception_pending;
1536
1537 assert_different_registers(receiver, rscratch1);
1538 __ verify_oop(receiver);
1539 __ ic_check(8 /* end_alignment */);
1540
1541 // Verified entry point must be aligned
1542 int vep_offset = ((intptr_t)__ pc()) - start;
1543
1544 // If we have to make this method not-entrant we'll overwrite its
1545 // first instruction with a jump. For this action to be legal we
1546 // must ensure that this first instruction is a B, BL, NOP, BKPT,
1547 // SVC, HVC, or SMC. Make it a NOP.
1548 __ nop();
1549
1550 if (VM_Version::supports_fast_class_init_checks() && method->needs_clinit_barrier()) {
1551 Label L_skip_barrier;
1552 __ mov_metadata(rscratch2, method->method_holder()); // InstanceKlass*
1553 __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
1554 __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
1555
1556 __ bind(L_skip_barrier);
1557 }
1558
1559 // Generate stack overflow check
1560 __ bang_stack_with_offset(checked_cast<int>(StackOverflow::stack_shadow_zone_size()));
1561
1562 // Generate a new frame for the wrapper.
1563 __ enter();
1564 // -2 because return address is already present and so is saved rfp
1565 __ sub(sp, sp, stack_size - 2*wordSize);
1566
1567 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
1568 bs->nmethod_entry_barrier(masm, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
1569
1570 // Frame is now completed as far as size and linkage.
1571 int frame_complete = ((intptr_t)__ pc()) - start;
1572
1573 // We use r20 as the oop handle for the receiver/klass
1574 // It is callee save so it survives the call to native
1575
1576 const Register oop_handle_reg = r20;
1577
1578 //
1579 // We immediately shuffle the arguments so that any vm call we have to
1580 // make from here on out (sync slow path, jvmti, etc.) we will have
1581 // captured the oops from our caller and have a valid oopMap for
1582 // them.
1583
1584 // -----------------
1585 // The Grand Shuffle
1586
1587 // The Java calling convention is either equal (linux) or denser (win64) than the
1588 // c calling convention. However the because of the jni_env argument the c calling
1589 // convention always has at least one more (and two for static) arguments than Java.
1590 // Therefore if we move the args from java -> c backwards then we will never have
1591 // a register->register conflict and we don't have to build a dependency graph
1592 // and figure out how to break any cycles.
1593 //
1594
1595 // Record esp-based slot for receiver on stack for non-static methods
1596 int receiver_offset = -1;
1597
1598 // This is a trick. We double the stack slots so we can claim
1599 // the oops in the caller's frame. Since we are sure to have
1600 // more args than the caller doubling is enough to make
1601 // sure we can capture all the incoming oop args from the
1602 // caller.
1603 //
1604 OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1605
1606 // Mark location of rfp (someday)
1607 // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, vmreg(rfp));
1608
1609
1610 int float_args = 0;
1611 int int_args = 0;
1612
1613 #ifdef ASSERT
1614 bool reg_destroyed[Register::number_of_registers];
1615 bool freg_destroyed[FloatRegister::number_of_registers];
1616 for ( int r = 0 ; r < Register::number_of_registers ; r++ ) {
1617 reg_destroyed[r] = false;
1618 }
1619 for ( int f = 0 ; f < FloatRegister::number_of_registers ; f++ ) {
1620 freg_destroyed[f] = false;
1621 }
1622
1623 #endif /* ASSERT */
1624
1625 // For JNI natives the incoming and outgoing registers are offset upwards.
1626 GrowableArray<int> arg_order(2 * total_in_args);
1627 VMRegPair tmp_vmreg;
1628 tmp_vmreg.set2(r19->as_VMReg());
1629
1630 for (int i = total_in_args - 1, c_arg = total_c_args - 1; i >= 0; i--, c_arg--) {
1631 arg_order.push(i);
1632 arg_order.push(c_arg);
1633 }
1634
1635 int temploc = -1;
1636 for (int ai = 0; ai < arg_order.length(); ai += 2) {
1637 int i = arg_order.at(ai);
1638 int c_arg = arg_order.at(ai + 1);
1639 __ block_comment(err_msg("move %d -> %d", i, c_arg));
1640 assert(c_arg != -1 && i != -1, "wrong order");
1641 #ifdef ASSERT
1642 if (in_regs[i].first()->is_Register()) {
1643 assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!");
1644 } else if (in_regs[i].first()->is_FloatRegister()) {
1645 assert(!freg_destroyed[in_regs[i].first()->as_FloatRegister()->encoding()], "destroyed reg!");
1646 }
1647 if (out_regs[c_arg].first()->is_Register()) {
1648 reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
1649 } else if (out_regs[c_arg].first()->is_FloatRegister()) {
1650 freg_destroyed[out_regs[c_arg].first()->as_FloatRegister()->encoding()] = true;
1651 }
1652 #endif /* ASSERT */
1653 switch (in_sig_bt[i]) {
1654 case T_ARRAY:
1655 case T_OBJECT:
1656 __ object_move(map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
1657 ((i == 0) && (!is_static)),
1658 &receiver_offset);
1659 int_args++;
1660 break;
1661 case T_VOID:
1662 break;
1663
1664 case T_FLOAT:
1665 __ float_move(in_regs[i], out_regs[c_arg]);
1666 float_args++;
1667 break;
1668
1669 case T_DOUBLE:
1670 assert( i + 1 < total_in_args &&
1671 in_sig_bt[i + 1] == T_VOID &&
1672 out_sig_bt[c_arg+1] == T_VOID, "bad arg list");
1673 __ double_move(in_regs[i], out_regs[c_arg]);
1674 float_args++;
1675 break;
1676
1677 case T_LONG :
1678 __ long_move(in_regs[i], out_regs[c_arg]);
1679 int_args++;
1680 break;
1681
1682 case T_ADDRESS: assert(false, "found T_ADDRESS in java args");
1683
1684 default:
1685 __ move32_64(in_regs[i], out_regs[c_arg]);
1686 int_args++;
1687 }
1688 }
1689
1690 // point c_arg at the first arg that is already loaded in case we
1691 // need to spill before we call out
1692 int c_arg = total_c_args - total_in_args;
1693
1694 // Pre-load a static method's oop into c_rarg1.
1695 if (method->is_static()) {
1696
1697 // load oop into a register
1698 __ movoop(c_rarg1,
1699 JNIHandles::make_local(method->method_holder()->java_mirror()));
1700
1701 // Now handlize the static class mirror it's known not-null.
1702 __ str(c_rarg1, Address(sp, klass_offset));
1703 map->set_oop(VMRegImpl::stack2reg(klass_slot_offset));
1704
1705 // Now get the handle
1706 __ lea(c_rarg1, Address(sp, klass_offset));
1707 // and protect the arg if we must spill
1708 c_arg--;
1709 }
1710
1711 // Change state to native (we save the return address in the thread, since it might not
1712 // be pushed on the stack when we do a stack traversal).
1713 // We use the same pc/oopMap repeatedly when we call out
1714
1715 Label native_return;
1716 __ set_last_Java_frame(sp, noreg, native_return, rscratch1);
1717
1718 Label dtrace_method_entry, dtrace_method_entry_done;
1719 {
1720 uint64_t offset;
1721 __ adrp(rscratch1, ExternalAddress((address)&DTraceMethodProbes), offset);
1722 __ ldrb(rscratch1, Address(rscratch1, offset));
1723 __ cbnzw(rscratch1, dtrace_method_entry);
1724 __ bind(dtrace_method_entry_done);
1725 }
1726
1727 // RedefineClasses() tracing support for obsolete method entry
1728 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1729 // protect the args we've loaded
1730 save_args(masm, total_c_args, c_arg, out_regs);
1731 __ mov_metadata(c_rarg1, method());
1732 __ call_VM_leaf(
1733 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1734 rthread, c_rarg1);
1735 restore_args(masm, total_c_args, c_arg, out_regs);
1736 }
1737
1738 // Lock a synchronized method
1739
1740 // Register definitions used by locking and unlocking
1741
1742 const Register swap_reg = r0;
1743 const Register obj_reg = r19; // Will contain the oop
1744 const Register lock_reg = r13; // Address of compiler lock object (BasicLock)
1745 const Register old_hdr = r13; // value of old header at unlock time
1746 const Register lock_tmp = r14; // Temporary used by lightweight_lock/unlock
1747 const Register tmp = lr;
1748
1749 Label slow_path_lock;
1750 Label lock_done;
1751
1752 if (method->is_synchronized()) {
1753 Label count;
1754 const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes();
1755
1756 // Get the handle (the 2nd argument)
1757 __ mov(oop_handle_reg, c_rarg1);
1758
1759 // Get address of the box
1760
1761 __ lea(lock_reg, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1762
1763 // Load the oop from the handle
1764 __ ldr(obj_reg, Address(oop_handle_reg, 0));
1765
1766 if (LockingMode == LM_MONITOR) {
1767 __ b(slow_path_lock);
1768 } else if (LockingMode == LM_LEGACY) {
1769 // Load (object->mark() | 1) into swap_reg %r0
1770 __ ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1771 __ orr(swap_reg, rscratch1, 1);
1772
1773 // Save (object->mark() | 1) into BasicLock's displaced header
1774 __ str(swap_reg, Address(lock_reg, mark_word_offset));
1775
1776 // src -> dest iff dest == r0 else r0 <- dest
1777 __ cmpxchg_obj_header(r0, lock_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
1778
1779 // Hmm should this move to the slow path code area???
1780
1781 // Test if the oopMark is an obvious stack pointer, i.e.,
1782 // 1) (mark & 3) == 0, and
1783 // 2) sp <= mark < mark + os::pagesize()
1784 // These 3 tests can be done by evaluating the following
1785 // expression: ((mark - sp) & (3 - os::vm_page_size())),
1786 // assuming both stack pointer and pagesize have their
1787 // least significant 2 bits clear.
1788 // NOTE: the oopMark is in swap_reg %r0 as the result of cmpxchg
1789
1790 __ sub(swap_reg, sp, swap_reg);
1791 __ neg(swap_reg, swap_reg);
1792 __ ands(swap_reg, swap_reg, 3 - (int)os::vm_page_size());
1793
1794 // Save the test result, for recursive case, the result is zero
1795 __ str(swap_reg, Address(lock_reg, mark_word_offset));
1796 __ br(Assembler::NE, slow_path_lock);
1797 } else {
1798 assert(LockingMode == LM_LIGHTWEIGHT, "must be");
1799 __ lightweight_lock(obj_reg, swap_reg, tmp, lock_tmp, slow_path_lock);
1800 }
1801 __ bind(count);
1802 __ increment(Address(rthread, JavaThread::held_monitor_count_offset()));
1803
1804 // Slow path will re-enter here
1805 __ bind(lock_done);
1806 }
1807
1808
1809 // Finally just about ready to make the JNI call
1810
1811 // get JNIEnv* which is first argument to native
1812 __ lea(c_rarg0, Address(rthread, in_bytes(JavaThread::jni_environment_offset())));
1813
1814 // Now set thread in native
1815 __ mov(rscratch1, _thread_in_native);
1816 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1817 __ stlrw(rscratch1, rscratch2);
1818
1819 __ rt_call(native_func);
1820
1821 __ bind(native_return);
1822
1823 intptr_t return_pc = (intptr_t) __ pc();
1824 oop_maps->add_gc_map(return_pc - start, map);
1825
1826 // Verify or restore cpu control state after JNI call
1827 __ restore_cpu_control_state_after_jni(rscratch1, rscratch2);
1828
1829 // Unpack native results.
1830 switch (ret_type) {
1831 case T_BOOLEAN: __ c2bool(r0); break;
1832 case T_CHAR : __ ubfx(r0, r0, 0, 16); break;
1833 case T_BYTE : __ sbfx(r0, r0, 0, 8); break;
1834 case T_SHORT : __ sbfx(r0, r0, 0, 16); break;
1835 case T_INT : __ sbfx(r0, r0, 0, 32); break;
1836 case T_DOUBLE :
1837 case T_FLOAT :
1838 // Result is in v0 we'll save as needed
1839 break;
1840 case T_ARRAY: // Really a handle
1841 case T_OBJECT: // Really a handle
1842 break; // can't de-handlize until after safepoint check
1843 case T_VOID: break;
1844 case T_LONG: break;
1845 default : ShouldNotReachHere();
1846 }
1847
1848 Label safepoint_in_progress, safepoint_in_progress_done;
1849 Label after_transition;
1850
1851 // Switch thread to "native transition" state before reading the synchronization state.
1852 // This additional state is necessary because reading and testing the synchronization
1853 // state is not atomic w.r.t. GC, as this scenario demonstrates:
1854 // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
1855 // VM thread changes sync state to synchronizing and suspends threads for GC.
1856 // Thread A is resumed to finish this native method, but doesn't block here since it
1857 // didn't see any synchronization is progress, and escapes.
1858 __ mov(rscratch1, _thread_in_native_trans);
1859
1860 __ strw(rscratch1, Address(rthread, JavaThread::thread_state_offset()));
1861
1862 // Force this write out before the read below
1863 if (!UseSystemMemoryBarrier) {
1864 __ dmb(Assembler::ISH);
1865 }
1866
1867 __ verify_sve_vector_length();
1868
1869 // Check for safepoint operation in progress and/or pending suspend requests.
1870 {
1871 // We need an acquire here to ensure that any subsequent load of the
1872 // global SafepointSynchronize::_state flag is ordered after this load
1873 // of the thread-local polling word. We don't want this poll to
1874 // return false (i.e. not safepointing) and a later poll of the global
1875 // SafepointSynchronize::_state spuriously to return true.
1876 //
1877 // This is to avoid a race when we're in a native->Java transition
1878 // racing the code which wakes up from a safepoint.
1879
1880 __ safepoint_poll(safepoint_in_progress, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1881 __ ldrw(rscratch1, Address(rthread, JavaThread::suspend_flags_offset()));
1882 __ cbnzw(rscratch1, safepoint_in_progress);
1883 __ bind(safepoint_in_progress_done);
1884 }
1885
1886 // change thread state
1887 __ mov(rscratch1, _thread_in_Java);
1888 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1889 __ stlrw(rscratch1, rscratch2);
1890 __ bind(after_transition);
1891
1892 Label reguard;
1893 Label reguard_done;
1894 __ ldrb(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
1895 __ cmpw(rscratch1, StackOverflow::stack_guard_yellow_reserved_disabled);
1896 __ br(Assembler::EQ, reguard);
1897 __ bind(reguard_done);
1898
1899 // native result if any is live
1900
1901 // Unlock
1902 Label unlock_done;
1903 Label slow_path_unlock;
1904 if (method->is_synchronized()) {
1905
1906 // Get locked oop from the handle we passed to jni
1907 __ ldr(obj_reg, Address(oop_handle_reg, 0));
1908
1909 Label done, not_recursive;
1910
1911 if (LockingMode == LM_LEGACY) {
1912 // Simple recursive lock?
1913 __ ldr(rscratch1, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1914 __ cbnz(rscratch1, not_recursive);
1915 __ decrement(Address(rthread, JavaThread::held_monitor_count_offset()));
1916 __ b(done);
1917 }
1918
1919 __ bind(not_recursive);
1920
1921 // Must save r0 if if it is live now because cmpxchg must use it
1922 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1923 save_native_result(masm, ret_type, stack_slots);
1924 }
1925
1926 if (LockingMode == LM_MONITOR) {
1927 __ b(slow_path_unlock);
1928 } else if (LockingMode == LM_LEGACY) {
1929 // get address of the stack lock
1930 __ lea(r0, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
1931 // get old displaced header
1932 __ ldr(old_hdr, Address(r0, 0));
1933
1934 // Atomic swap old header if oop still contains the stack lock
1935 Label count;
1936 __ cmpxchg_obj_header(r0, old_hdr, obj_reg, rscratch1, count, &slow_path_unlock);
1937 __ bind(count);
1938 __ decrement(Address(rthread, JavaThread::held_monitor_count_offset()));
1939 } else {
1940 assert(LockingMode == LM_LIGHTWEIGHT, "");
1941 __ lightweight_unlock(obj_reg, old_hdr, swap_reg, lock_tmp, slow_path_unlock);
1942 __ decrement(Address(rthread, JavaThread::held_monitor_count_offset()));
1943 }
1944
1945 // slow path re-enters here
1946 __ bind(unlock_done);
1947 if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1948 restore_native_result(masm, ret_type, stack_slots);
1949 }
1950
1951 __ bind(done);
1952 }
1953
1954 Label dtrace_method_exit, dtrace_method_exit_done;
1955 {
1956 uint64_t offset;
1957 __ adrp(rscratch1, ExternalAddress((address)&DTraceMethodProbes), offset);
1958 __ ldrb(rscratch1, Address(rscratch1, offset));
1959 __ cbnzw(rscratch1, dtrace_method_exit);
1960 __ bind(dtrace_method_exit_done);
1961 }
1962
1963 __ reset_last_Java_frame(false);
1964
1965 // Unbox oop result, e.g. JNIHandles::resolve result.
1966 if (is_reference_type(ret_type)) {
1967 __ resolve_jobject(r0, r1, r2);
1968 }
1969
1970 if (CheckJNICalls) {
1971 // clear_pending_jni_exception_check
1972 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1973 }
1974
1975 // reset handle block
1976 __ ldr(r2, Address(rthread, JavaThread::active_handles_offset()));
1977 __ str(zr, Address(r2, JNIHandleBlock::top_offset()));
1978
1979 __ leave();
1980
1981 // Any exception pending?
1982 __ ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
1983 __ cbnz(rscratch1, exception_pending);
1984
1985 // We're done
1986 __ ret(lr);
1987
1988 // Unexpected paths are out of line and go here
1989
1990 // forward the exception
1991 __ bind(exception_pending);
1992
1993 // and forward the exception
1994 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1995
1996 // Slow path locking & unlocking
1997 if (method->is_synchronized()) {
1998
1999 __ block_comment("Slow path lock {");
2000 __ bind(slow_path_lock);
2001
2002 // has last_Java_frame setup. No exceptions so do vanilla call not call_VM
2003 // args are (oop obj, BasicLock* lock, JavaThread* thread)
2004
2005 // protect the args we've loaded
2006 save_args(masm, total_c_args, c_arg, out_regs);
2007
2008 __ mov(c_rarg0, obj_reg);
2009 __ mov(c_rarg1, lock_reg);
2010 __ mov(c_rarg2, rthread);
2011
2012 // Not a leaf but we have last_Java_frame setup as we want
2013 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), 3);
2014 restore_args(masm, total_c_args, c_arg, out_regs);
2015
2016 #ifdef ASSERT
2017 { Label L;
2018 __ ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2019 __ cbz(rscratch1, L);
2020 __ stop("no pending exception allowed on exit from monitorenter");
2021 __ bind(L);
2022 }
2023 #endif
2024 __ b(lock_done);
2025
2026 __ block_comment("} Slow path lock");
2027
2028 __ block_comment("Slow path unlock {");
2029 __ bind(slow_path_unlock);
2030
2031 // If we haven't already saved the native result we must save it now as xmm registers
2032 // are still exposed.
2033
2034 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
2035 save_native_result(masm, ret_type, stack_slots);
2036 }
2037
2038 __ mov(c_rarg2, rthread);
2039 __ lea(c_rarg1, Address(sp, lock_slot_offset * VMRegImpl::stack_slot_size));
2040 __ mov(c_rarg0, obj_reg);
2041
2042 // Save pending exception around call to VM (which contains an EXCEPTION_MARK)
2043 // NOTE that obj_reg == r19 currently
2044 __ ldr(r19, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2045 __ str(zr, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2046
2047 __ rt_call(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C));
2048
2049 #ifdef ASSERT
2050 {
2051 Label L;
2052 __ ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2053 __ cbz(rscratch1, L);
2054 __ stop("no pending exception allowed on exit complete_monitor_unlocking_C");
2055 __ bind(L);
2056 }
2057 #endif /* ASSERT */
2058
2059 __ str(r19, Address(rthread, in_bytes(Thread::pending_exception_offset())));
2060
2061 if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
2062 restore_native_result(masm, ret_type, stack_slots);
2063 }
2064 __ b(unlock_done);
2065
2066 __ block_comment("} Slow path unlock");
2067
2068 } // synchronized
2069
2070 // SLOW PATH Reguard the stack if needed
2071
2072 __ bind(reguard);
2073 save_native_result(masm, ret_type, stack_slots);
2074 __ rt_call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
2075 restore_native_result(masm, ret_type, stack_slots);
2076 // and continue
2077 __ b(reguard_done);
2078
2079 // SLOW PATH safepoint
2080 {
2081 __ block_comment("safepoint {");
2082 __ bind(safepoint_in_progress);
2083
2084 // Don't use call_VM as it will see a possible pending exception and forward it
2085 // and never return here preventing us from clearing _last_native_pc down below.
2086 //
2087 save_native_result(masm, ret_type, stack_slots);
2088 __ mov(c_rarg0, rthread);
2089 #ifndef PRODUCT
2090 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
2091 #endif
2092 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
2093 __ blr(rscratch1);
2094
2095 // Restore any method result value
2096 restore_native_result(masm, ret_type, stack_slots);
2097
2098 __ b(safepoint_in_progress_done);
2099 __ block_comment("} safepoint");
2100 }
2101
2102 // SLOW PATH dtrace support
2103 {
2104 __ block_comment("dtrace entry {");
2105 __ bind(dtrace_method_entry);
2106
2107 // We have all of the arguments setup at this point. We must not touch any register
2108 // argument registers at this point (what if we save/restore them there are no oop?
2109
2110 save_args(masm, total_c_args, c_arg, out_regs);
2111 __ mov_metadata(c_rarg1, method());
2112 __ call_VM_leaf(
2113 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2114 rthread, c_rarg1);
2115 restore_args(masm, total_c_args, c_arg, out_regs);
2116 __ b(dtrace_method_entry_done);
2117 __ block_comment("} dtrace entry");
2118 }
2119
2120 {
2121 __ block_comment("dtrace exit {");
2122 __ bind(dtrace_method_exit);
2123 save_native_result(masm, ret_type, stack_slots);
2124 __ mov_metadata(c_rarg1, method());
2125 __ call_VM_leaf(
2126 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2127 rthread, c_rarg1);
2128 restore_native_result(masm, ret_type, stack_slots);
2129 __ b(dtrace_method_exit_done);
2130 __ block_comment("} dtrace exit");
2131 }
2132
2133
2134 __ flush();
2135
2136 nmethod *nm = nmethod::new_native_nmethod(method,
2137 compile_id,
2138 masm->code(),
2139 vep_offset,
2140 frame_complete,
2141 stack_slots / VMRegImpl::slots_per_word,
2142 (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)),
2143 in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size),
2144 oop_maps);
2145
2146 return nm;
2147 }
2148
2149 // this function returns the adjust size (in number of words) to a c2i adapter
2150 // activation for use during deoptimization
2151 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals) {
2152 assert(callee_locals >= callee_parameters,
2153 "test and remove; got more parms than locals");
2154 if (callee_locals < callee_parameters)
2155 return 0; // No adjustment for negative locals
2156 int diff = (callee_locals - callee_parameters) * Interpreter::stackElementWords;
2157 // diff is counted in stack words
2158 return align_up(diff, 2);
2159 }
2160
2161
2162 //------------------------------generate_deopt_blob----------------------------
2163 void SharedRuntime::generate_deopt_blob() {
2164 // Allocate space for the code
2165 ResourceMark rm;
2166 // Setup code generation tools
2167 int pad = 0;
2168 #if INCLUDE_JVMCI
2169 if (EnableJVMCI) {
2170 pad += 512; // Increase the buffer size when compiling for JVMCI
2171 }
2172 #endif
2173 CodeBuffer buffer("deopt_blob", 2048+pad, 1024);
2174 MacroAssembler* masm = new MacroAssembler(&buffer);
2175 int frame_size_in_words;
2176 OopMap* map = nullptr;
2177 OopMapSet *oop_maps = new OopMapSet();
2178 RegisterSaver reg_save(COMPILER2_OR_JVMCI != 0);
2179
2180 // -------------
2181 // This code enters when returning to a de-optimized nmethod. A return
2182 // address has been pushed on the stack, and return values are in
2183 // registers.
2184 // If we are doing a normal deopt then we were called from the patched
2185 // nmethod from the point we returned to the nmethod. So the return
2186 // address on the stack is wrong by NativeCall::instruction_size
2187 // We will adjust the value so it looks like we have the original return
2188 // address on the stack (like when we eagerly deoptimized).
2189 // In the case of an exception pending when deoptimizing, we enter
2190 // with a return address on the stack that points after the call we patched
2191 // into the exception handler. We have the following register state from,
2192 // e.g., the forward exception stub (see stubGenerator_x86_64.cpp).
2193 // r0: exception oop
2194 // r19: exception handler
2195 // r3: throwing pc
2196 // So in this case we simply jam r3 into the useless return address and
2197 // the stack looks just like we want.
2198 //
2199 // At this point we need to de-opt. We save the argument return
2200 // registers. We call the first C routine, fetch_unroll_info(). This
2201 // routine captures the return values and returns a structure which
2202 // describes the current frame size and the sizes of all replacement frames.
2203 // The current frame is compiled code and may contain many inlined
2204 // functions, each with their own JVM state. We pop the current frame, then
2205 // push all the new frames. Then we call the C routine unpack_frames() to
2206 // populate these frames. Finally unpack_frames() returns us the new target
2207 // address. Notice that callee-save registers are BLOWN here; they have
2208 // already been captured in the vframeArray at the time the return PC was
2209 // patched.
2210 address start = __ pc();
2211 Label cont;
2212
2213 // Prolog for non exception case!
2214
2215 // Save everything in sight.
2216 map = reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2217
2218 // Normal deoptimization. Save exec mode for unpack_frames.
2219 __ movw(rcpool, Deoptimization::Unpack_deopt); // callee-saved
2220 __ b(cont);
2221
2222 int reexecute_offset = __ pc() - start;
2223 #if INCLUDE_JVMCI && !defined(COMPILER1)
2224 if (EnableJVMCI && UseJVMCICompiler) {
2225 // JVMCI does not use this kind of deoptimization
2226 __ should_not_reach_here();
2227 }
2228 #endif
2229
2230 // Reexecute case
2231 // return address is the pc describes what bci to do re-execute at
2232
2233 // No need to update map as each call to save_live_registers will produce identical oopmap
2234 (void) reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2235
2236 __ movw(rcpool, Deoptimization::Unpack_reexecute); // callee-saved
2237 __ b(cont);
2238
2239 #if INCLUDE_JVMCI
2240 Label after_fetch_unroll_info_call;
2241 int implicit_exception_uncommon_trap_offset = 0;
2242 int uncommon_trap_offset = 0;
2243
2244 if (EnableJVMCI) {
2245 implicit_exception_uncommon_trap_offset = __ pc() - start;
2246
2247 __ ldr(lr, Address(rthread, in_bytes(JavaThread::jvmci_implicit_exception_pc_offset())));
2248 __ str(zr, Address(rthread, in_bytes(JavaThread::jvmci_implicit_exception_pc_offset())));
2249
2250 uncommon_trap_offset = __ pc() - start;
2251
2252 // Save everything in sight.
2253 reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2254 // fetch_unroll_info needs to call last_java_frame()
2255 Label retaddr;
2256 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2257
2258 __ ldrw(c_rarg1, Address(rthread, in_bytes(JavaThread::pending_deoptimization_offset())));
2259 __ movw(rscratch1, -1);
2260 __ strw(rscratch1, Address(rthread, in_bytes(JavaThread::pending_deoptimization_offset())));
2261
2262 __ movw(rcpool, (int32_t)Deoptimization::Unpack_reexecute);
2263 __ mov(c_rarg0, rthread);
2264 __ movw(c_rarg2, rcpool); // exec mode
2265 __ lea(rscratch1,
2266 RuntimeAddress(CAST_FROM_FN_PTR(address,
2267 Deoptimization::uncommon_trap)));
2268 __ blr(rscratch1);
2269 __ bind(retaddr);
2270 oop_maps->add_gc_map( __ pc()-start, map->deep_copy());
2271
2272 __ reset_last_Java_frame(false);
2273
2274 __ b(after_fetch_unroll_info_call);
2275 } // EnableJVMCI
2276 #endif // INCLUDE_JVMCI
2277
2278 int exception_offset = __ pc() - start;
2279
2280 // Prolog for exception case
2281
2282 // all registers are dead at this entry point, except for r0, and
2283 // r3 which contain the exception oop and exception pc
2284 // respectively. Set them in TLS and fall thru to the
2285 // unpack_with_exception_in_tls entry point.
2286
2287 __ str(r3, Address(rthread, JavaThread::exception_pc_offset()));
2288 __ str(r0, Address(rthread, JavaThread::exception_oop_offset()));
2289
2290 int exception_in_tls_offset = __ pc() - start;
2291
2292 // new implementation because exception oop is now passed in JavaThread
2293
2294 // Prolog for exception case
2295 // All registers must be preserved because they might be used by LinearScan
2296 // Exceptiop oop and throwing PC are passed in JavaThread
2297 // tos: stack at point of call to method that threw the exception (i.e. only
2298 // args are on the stack, no return address)
2299
2300 // The return address pushed by save_live_registers will be patched
2301 // later with the throwing pc. The correct value is not available
2302 // now because loading it from memory would destroy registers.
2303
2304 // NB: The SP at this point must be the SP of the method that is
2305 // being deoptimized. Deoptimization assumes that the frame created
2306 // here by save_live_registers is immediately below the method's SP.
2307 // This is a somewhat fragile mechanism.
2308
2309 // Save everything in sight.
2310 map = reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2311
2312 // Now it is safe to overwrite any register
2313
2314 // Deopt during an exception. Save exec mode for unpack_frames.
2315 __ mov(rcpool, Deoptimization::Unpack_exception); // callee-saved
2316
2317 // load throwing pc from JavaThread and patch it as the return address
2318 // of the current frame. Then clear the field in JavaThread
2319 __ ldr(r3, Address(rthread, JavaThread::exception_pc_offset()));
2320 __ protect_return_address(r3);
2321 __ str(r3, Address(rfp, wordSize));
2322 __ str(zr, Address(rthread, JavaThread::exception_pc_offset()));
2323
2324 #ifdef ASSERT
2325 // verify that there is really an exception oop in JavaThread
2326 __ ldr(r0, Address(rthread, JavaThread::exception_oop_offset()));
2327 __ verify_oop(r0);
2328
2329 // verify that there is no pending exception
2330 Label no_pending_exception;
2331 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
2332 __ cbz(rscratch1, no_pending_exception);
2333 __ stop("must not have pending exception here");
2334 __ bind(no_pending_exception);
2335 #endif
2336
2337 __ bind(cont);
2338
2339 // Call C code. Need thread and this frame, but NOT official VM entry
2340 // crud. We cannot block on this call, no GC can happen.
2341 //
2342 // UnrollBlock* fetch_unroll_info(JavaThread* thread)
2343
2344 // fetch_unroll_info needs to call last_java_frame().
2345
2346 Label retaddr;
2347 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2348 #ifdef ASSERT
2349 { Label L;
2350 __ ldr(rscratch1, Address(rthread, JavaThread::last_Java_fp_offset()));
2351 __ cbz(rscratch1, L);
2352 __ stop("SharedRuntime::generate_deopt_blob: last_Java_fp not cleared");
2353 __ bind(L);
2354 }
2355 #endif // ASSERT
2356 __ mov(c_rarg0, rthread);
2357 __ mov(c_rarg1, rcpool);
2358 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info)));
2359 __ blr(rscratch1);
2360 __ bind(retaddr);
2361
2362 // Need to have an oopmap that tells fetch_unroll_info where to
2363 // find any register it might need.
2364 oop_maps->add_gc_map(__ pc() - start, map);
2365
2366 __ reset_last_Java_frame(false);
2367
2368 #if INCLUDE_JVMCI
2369 if (EnableJVMCI) {
2370 __ bind(after_fetch_unroll_info_call);
2371 }
2372 #endif
2373
2374 // Load UnrollBlock* into r5
2375 __ mov(r5, r0);
2376
2377 __ ldrw(rcpool, Address(r5, Deoptimization::UnrollBlock::unpack_kind_offset()));
2378 Label noException;
2379 __ cmpw(rcpool, Deoptimization::Unpack_exception); // Was exception pending?
2380 __ br(Assembler::NE, noException);
2381 __ ldr(r0, Address(rthread, JavaThread::exception_oop_offset()));
2382 // QQQ this is useless it was null above
2383 __ ldr(r3, Address(rthread, JavaThread::exception_pc_offset()));
2384 __ str(zr, Address(rthread, JavaThread::exception_oop_offset()));
2385 __ str(zr, Address(rthread, JavaThread::exception_pc_offset()));
2386
2387 __ verify_oop(r0);
2388
2389 // Overwrite the result registers with the exception results.
2390 __ str(r0, Address(sp, reg_save.r0_offset_in_bytes()));
2391 // I think this is useless
2392 // __ str(r3, Address(sp, RegisterSaver::r3_offset_in_bytes()));
2393
2394 __ bind(noException);
2395
2396 // Only register save data is on the stack.
2397 // Now restore the result registers. Everything else is either dead
2398 // or captured in the vframeArray.
2399
2400 // Restore fp result register
2401 __ ldrd(v0, Address(sp, reg_save.v0_offset_in_bytes()));
2402 // Restore integer result register
2403 __ ldr(r0, Address(sp, reg_save.r0_offset_in_bytes()));
2404
2405 // Pop all of the register save area off the stack
2406 __ add(sp, sp, frame_size_in_words * wordSize);
2407
2408 // All of the register save area has been popped of the stack. Only the
2409 // return address remains.
2410
2411 // Pop all the frames we must move/replace.
2412 //
2413 // Frame picture (youngest to oldest)
2414 // 1: self-frame (no frame link)
2415 // 2: deopting frame (no frame link)
2416 // 3: caller of deopting frame (could be compiled/interpreted).
2417 //
2418 // Note: by leaving the return address of self-frame on the stack
2419 // and using the size of frame 2 to adjust the stack
2420 // when we are done the return to frame 3 will still be on the stack.
2421
2422 // Pop deoptimized frame
2423 __ ldrw(r2, Address(r5, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset()));
2424 __ sub(r2, r2, 2 * wordSize);
2425 __ add(sp, sp, r2);
2426 __ ldp(rfp, zr, __ post(sp, 2 * wordSize));
2427
2428 #ifdef ASSERT
2429 // Compilers generate code that bang the stack by as much as the
2430 // interpreter would need. So this stack banging should never
2431 // trigger a fault. Verify that it does not on non product builds.
2432 __ ldrw(r19, Address(r5, Deoptimization::UnrollBlock::total_frame_sizes_offset()));
2433 __ bang_stack_size(r19, r2);
2434 #endif
2435 // Load address of array of frame pcs into r2
2436 __ ldr(r2, Address(r5, Deoptimization::UnrollBlock::frame_pcs_offset()));
2437
2438 // Trash the old pc
2439 // __ addptr(sp, wordSize); FIXME ????
2440
2441 // Load address of array of frame sizes into r4
2442 __ ldr(r4, Address(r5, Deoptimization::UnrollBlock::frame_sizes_offset()));
2443
2444 // Load counter into r3
2445 __ ldrw(r3, Address(r5, Deoptimization::UnrollBlock::number_of_frames_offset()));
2446
2447 // Now adjust the caller's stack to make up for the extra locals
2448 // but record the original sp so that we can save it in the skeletal interpreter
2449 // frame and the stack walking of interpreter_sender will get the unextended sp
2450 // value and not the "real" sp value.
2451
2452 const Register sender_sp = r6;
2453
2454 __ mov(sender_sp, sp);
2455 __ ldrw(r19, Address(r5,
2456 Deoptimization::UnrollBlock::
2457 caller_adjustment_offset()));
2458 __ sub(sp, sp, r19);
2459
2460 // Push interpreter frames in a loop
2461 __ mov(rscratch1, (uint64_t)0xDEADDEAD); // Make a recognizable pattern
2462 __ mov(rscratch2, rscratch1);
2463 Label loop;
2464 __ bind(loop);
2465 __ ldr(r19, Address(__ post(r4, wordSize))); // Load frame size
2466 __ sub(r19, r19, 2*wordSize); // We'll push pc and fp by hand
2467 __ ldr(lr, Address(__ post(r2, wordSize))); // Load pc
2468 __ enter(); // Save old & set new fp
2469 __ sub(sp, sp, r19); // Prolog
2470 // This value is corrected by layout_activation_impl
2471 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
2472 __ str(sender_sp, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize)); // Make it walkable
2473 __ mov(sender_sp, sp); // Pass sender_sp to next frame
2474 __ sub(r3, r3, 1); // Decrement counter
2475 __ cbnz(r3, loop);
2476
2477 // Re-push self-frame
2478 __ ldr(lr, Address(r2));
2479 __ enter();
2480
2481 // Allocate a full sized register save area. We subtract 2 because
2482 // enter() just pushed 2 words
2483 __ sub(sp, sp, (frame_size_in_words - 2) * wordSize);
2484
2485 // Restore frame locals after moving the frame
2486 __ strd(v0, Address(sp, reg_save.v0_offset_in_bytes()));
2487 __ str(r0, Address(sp, reg_save.r0_offset_in_bytes()));
2488
2489 // Call C code. Need thread but NOT official VM entry
2490 // crud. We cannot block on this call, no GC can happen. Call should
2491 // restore return values to their stack-slots with the new SP.
2492 //
2493 // void Deoptimization::unpack_frames(JavaThread* thread, int exec_mode)
2494
2495 // Use rfp because the frames look interpreted now
2496 // Don't need the precise return PC here, just precise enough to point into this code blob.
2497 address the_pc = __ pc();
2498 __ set_last_Java_frame(sp, rfp, the_pc, rscratch1);
2499
2500 __ mov(c_rarg0, rthread);
2501 __ movw(c_rarg1, rcpool); // second arg: exec_mode
2502 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
2503 __ blr(rscratch1);
2504
2505 // Set an oopmap for the call site
2506 // Use the same PC we used for the last java frame
2507 oop_maps->add_gc_map(the_pc - start,
2508 new OopMap( frame_size_in_words, 0 ));
2509
2510 // Clear fp AND pc
2511 __ reset_last_Java_frame(true);
2512
2513 // Collect return values
2514 __ ldrd(v0, Address(sp, reg_save.v0_offset_in_bytes()));
2515 __ ldr(r0, Address(sp, reg_save.r0_offset_in_bytes()));
2516 // I think this is useless (throwing pc?)
2517 // __ ldr(r3, Address(sp, RegisterSaver::r3_offset_in_bytes()));
2518
2519 // Pop self-frame.
2520 __ leave(); // Epilog
2521
2522 // Jump to interpreter
2523 __ ret(lr);
2524
2525 // Make sure all code is generated
2526 masm->flush();
2527
2528 _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words);
2529 _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset);
2530 #if INCLUDE_JVMCI
2531 if (EnableJVMCI) {
2532 _deopt_blob->set_uncommon_trap_offset(uncommon_trap_offset);
2533 _deopt_blob->set_implicit_exception_uncommon_trap_offset(implicit_exception_uncommon_trap_offset);
2534 }
2535 #endif
2536 }
2537
2538 // Number of stack slots between incoming argument block and the start of
2539 // a new frame. The PROLOG must add this many slots to the stack. The
2540 // EPILOG must remove this many slots. aarch64 needs two slots for
2541 // return address and fp.
2542 // TODO think this is correct but check
2543 uint SharedRuntime::in_preserve_stack_slots() {
2544 return 4;
2545 }
2546
2547 uint SharedRuntime::out_preserve_stack_slots() {
2548 return 0;
2549 }
2550
2551 #ifdef COMPILER2
2552 //------------------------------generate_uncommon_trap_blob--------------------
2553 void SharedRuntime::generate_uncommon_trap_blob() {
2554 // Allocate space for the code
2555 ResourceMark rm;
2556 // Setup code generation tools
2557 CodeBuffer buffer("uncommon_trap_blob", 2048, 1024);
2558 MacroAssembler* masm = new MacroAssembler(&buffer);
2559
2560 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
2561
2562 address start = __ pc();
2563
2564 // Push self-frame. We get here with a return address in LR
2565 // and sp should be 16 byte aligned
2566 // push rfp and retaddr by hand
2567 __ protect_return_address();
2568 __ stp(rfp, lr, Address(__ pre(sp, -2 * wordSize)));
2569 // we don't expect an arg reg save area
2570 #ifndef PRODUCT
2571 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
2572 #endif
2573 // compiler left unloaded_class_index in j_rarg0 move to where the
2574 // runtime expects it.
2575 if (c_rarg1 != j_rarg0) {
2576 __ movw(c_rarg1, j_rarg0);
2577 }
2578
2579 // we need to set the past SP to the stack pointer of the stub frame
2580 // and the pc to the address where this runtime call will return
2581 // although actually any pc in this code blob will do).
2582 Label retaddr;
2583 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2584
2585 // Call C code. Need thread but NOT official VM entry
2586 // crud. We cannot block on this call, no GC can happen. Call should
2587 // capture callee-saved registers as well as return values.
2588 // Thread is in rdi already.
2589 //
2590 // UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index);
2591 //
2592 // n.b. 2 gp args, 0 fp args, integral return type
2593
2594 __ mov(c_rarg0, rthread);
2595 __ movw(c_rarg2, (unsigned)Deoptimization::Unpack_uncommon_trap);
2596 __ lea(rscratch1,
2597 RuntimeAddress(CAST_FROM_FN_PTR(address,
2598 Deoptimization::uncommon_trap)));
2599 __ blr(rscratch1);
2600 __ bind(retaddr);
2601
2602 // Set an oopmap for the call site
2603 OopMapSet* oop_maps = new OopMapSet();
2604 OopMap* map = new OopMap(SimpleRuntimeFrame::framesize, 0);
2605
2606 // location of rfp is known implicitly by the frame sender code
2607
2608 oop_maps->add_gc_map(__ pc() - start, map);
2609
2610 __ reset_last_Java_frame(false);
2611
2612 // move UnrollBlock* into r4
2613 __ mov(r4, r0);
2614
2615 #ifdef ASSERT
2616 { Label L;
2617 __ ldrw(rscratch1, Address(r4, Deoptimization::UnrollBlock::unpack_kind_offset()));
2618 __ cmpw(rscratch1, (unsigned)Deoptimization::Unpack_uncommon_trap);
2619 __ br(Assembler::EQ, L);
2620 __ stop("SharedRuntime::generate_uncommon_trap_blob: expected Unpack_uncommon_trap");
2621 __ bind(L);
2622 }
2623 #endif
2624
2625 // Pop all the frames we must move/replace.
2626 //
2627 // Frame picture (youngest to oldest)
2628 // 1: self-frame (no frame link)
2629 // 2: deopting frame (no frame link)
2630 // 3: caller of deopting frame (could be compiled/interpreted).
2631
2632 // Pop self-frame. We have no frame, and must rely only on r0 and sp.
2633 __ add(sp, sp, (SimpleRuntimeFrame::framesize) << LogBytesPerInt); // Epilog!
2634
2635 // Pop deoptimized frame (int)
2636 __ ldrw(r2, Address(r4,
2637 Deoptimization::UnrollBlock::
2638 size_of_deoptimized_frame_offset()));
2639 __ sub(r2, r2, 2 * wordSize);
2640 __ add(sp, sp, r2);
2641 __ ldp(rfp, zr, __ post(sp, 2 * wordSize));
2642
2643 #ifdef ASSERT
2644 // Compilers generate code that bang the stack by as much as the
2645 // interpreter would need. So this stack banging should never
2646 // trigger a fault. Verify that it does not on non product builds.
2647 __ ldrw(r1, Address(r4,
2648 Deoptimization::UnrollBlock::
2649 total_frame_sizes_offset()));
2650 __ bang_stack_size(r1, r2);
2651 #endif
2652
2653 // Load address of array of frame pcs into r2 (address*)
2654 __ ldr(r2, Address(r4,
2655 Deoptimization::UnrollBlock::frame_pcs_offset()));
2656
2657 // Load address of array of frame sizes into r5 (intptr_t*)
2658 __ ldr(r5, Address(r4,
2659 Deoptimization::UnrollBlock::
2660 frame_sizes_offset()));
2661
2662 // Counter
2663 __ ldrw(r3, Address(r4,
2664 Deoptimization::UnrollBlock::
2665 number_of_frames_offset())); // (int)
2666
2667 // Now adjust the caller's stack to make up for the extra locals but
2668 // record the original sp so that we can save it in the skeletal
2669 // interpreter frame and the stack walking of interpreter_sender
2670 // will get the unextended sp value and not the "real" sp value.
2671
2672 const Register sender_sp = r8;
2673
2674 __ mov(sender_sp, sp);
2675 __ ldrw(r1, Address(r4,
2676 Deoptimization::UnrollBlock::
2677 caller_adjustment_offset())); // (int)
2678 __ sub(sp, sp, r1);
2679
2680 // Push interpreter frames in a loop
2681 Label loop;
2682 __ bind(loop);
2683 __ ldr(r1, Address(r5, 0)); // Load frame size
2684 __ sub(r1, r1, 2 * wordSize); // We'll push pc and rfp by hand
2685 __ ldr(lr, Address(r2, 0)); // Save return address
2686 __ enter(); // and old rfp & set new rfp
2687 __ sub(sp, sp, r1); // Prolog
2688 __ str(sender_sp, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize)); // Make it walkable
2689 // This value is corrected by layout_activation_impl
2690 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
2691 __ mov(sender_sp, sp); // Pass sender_sp to next frame
2692 __ add(r5, r5, wordSize); // Bump array pointer (sizes)
2693 __ add(r2, r2, wordSize); // Bump array pointer (pcs)
2694 __ subsw(r3, r3, 1); // Decrement counter
2695 __ br(Assembler::GT, loop);
2696 __ ldr(lr, Address(r2, 0)); // save final return address
2697 // Re-push self-frame
2698 __ enter(); // & old rfp & set new rfp
2699
2700 // Use rfp because the frames look interpreted now
2701 // Save "the_pc" since it cannot easily be retrieved using the last_java_SP after we aligned SP.
2702 // Don't need the precise return PC here, just precise enough to point into this code blob.
2703 address the_pc = __ pc();
2704 __ set_last_Java_frame(sp, rfp, the_pc, rscratch1);
2705
2706 // Call C code. Need thread but NOT official VM entry
2707 // crud. We cannot block on this call, no GC can happen. Call should
2708 // restore return values to their stack-slots with the new SP.
2709 // Thread is in rdi already.
2710 //
2711 // BasicType unpack_frames(JavaThread* thread, int exec_mode);
2712 //
2713 // n.b. 2 gp args, 0 fp args, integral return type
2714
2715 // sp should already be aligned
2716 __ mov(c_rarg0, rthread);
2717 __ movw(c_rarg1, (unsigned)Deoptimization::Unpack_uncommon_trap);
2718 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
2719 __ blr(rscratch1);
2720
2721 // Set an oopmap for the call site
2722 // Use the same PC we used for the last java frame
2723 oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
2724
2725 // Clear fp AND pc
2726 __ reset_last_Java_frame(true);
2727
2728 // Pop self-frame.
2729 __ leave(); // Epilog
2730
2731 // Jump to interpreter
2732 __ ret(lr);
2733
2734 // Make sure all code is generated
2735 masm->flush();
2736
2737 _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps,
2738 SimpleRuntimeFrame::framesize >> 1);
2739 }
2740 #endif // COMPILER2
2741
2742
2743 //------------------------------generate_handler_blob------
2744 //
2745 // Generate a special Compile2Runtime blob that saves all registers,
2746 // and setup oopmap.
2747 //
2748 SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, int poll_type) {
2749 ResourceMark rm;
2750 OopMapSet *oop_maps = new OopMapSet();
2751 OopMap* map;
2752
2753 // Allocate space for the code. Setup code generation tools.
2754 CodeBuffer buffer("handler_blob", 2048, 1024);
2755 MacroAssembler* masm = new MacroAssembler(&buffer);
2756
2757 address start = __ pc();
2758 address call_pc = nullptr;
2759 int frame_size_in_words;
2760 bool cause_return = (poll_type == POLL_AT_RETURN);
2761 RegisterSaver reg_save(poll_type == POLL_AT_VECTOR_LOOP /* save_vectors */);
2762
2763 // When the signal occurred, the LR was either signed and stored on the stack (in which
2764 // case it will be restored from the stack before being used) or unsigned and not stored
2765 // on the stack. Stipping ensures we get the right value.
2766 __ strip_return_address();
2767
2768 // Save Integer and Float registers.
2769 map = reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2770
2771 // The following is basically a call_VM. However, we need the precise
2772 // address of the call in order to generate an oopmap. Hence, we do all the
2773 // work ourselves.
2774
2775 Label retaddr;
2776 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2777
2778 // The return address must always be correct so that frame constructor never
2779 // sees an invalid pc.
2780
2781 if (!cause_return) {
2782 // overwrite the return address pushed by save_live_registers
2783 // Additionally, r20 is a callee-saved register so we can look at
2784 // it later to determine if someone changed the return address for
2785 // us!
2786 __ ldr(r20, Address(rthread, JavaThread::saved_exception_pc_offset()));
2787 __ protect_return_address(r20);
2788 __ str(r20, Address(rfp, wordSize));
2789 }
2790
2791 // Do the call
2792 __ mov(c_rarg0, rthread);
2793 __ lea(rscratch1, RuntimeAddress(call_ptr));
2794 __ blr(rscratch1);
2795 __ bind(retaddr);
2796
2797 // Set an oopmap for the call site. This oopmap will map all
2798 // oop-registers and debug-info registers as callee-saved. This
2799 // will allow deoptimization at this safepoint to find all possible
2800 // debug-info recordings, as well as let GC find all oops.
2801
2802 oop_maps->add_gc_map( __ pc() - start, map);
2803
2804 Label noException;
2805
2806 __ reset_last_Java_frame(false);
2807
2808 __ membar(Assembler::LoadLoad | Assembler::LoadStore);
2809
2810 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
2811 __ cbz(rscratch1, noException);
2812
2813 // Exception pending
2814
2815 reg_save.restore_live_registers(masm);
2816
2817 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2818
2819 // No exception case
2820 __ bind(noException);
2821
2822 Label no_adjust, bail;
2823 if (!cause_return) {
2824 // If our stashed return pc was modified by the runtime we avoid touching it
2825 __ ldr(rscratch1, Address(rfp, wordSize));
2826 __ cmp(r20, rscratch1);
2827 __ br(Assembler::NE, no_adjust);
2828 __ authenticate_return_address(r20);
2829
2830 #ifdef ASSERT
2831 // Verify the correct encoding of the poll we're about to skip.
2832 // See NativeInstruction::is_ldrw_to_zr()
2833 __ ldrw(rscratch1, Address(r20));
2834 __ ubfx(rscratch2, rscratch1, 22, 10);
2835 __ cmpw(rscratch2, 0b1011100101);
2836 __ br(Assembler::NE, bail);
2837 __ ubfx(rscratch2, rscratch1, 0, 5);
2838 __ cmpw(rscratch2, 0b11111);
2839 __ br(Assembler::NE, bail);
2840 #endif
2841 // Adjust return pc forward to step over the safepoint poll instruction
2842 __ add(r20, r20, NativeInstruction::instruction_size);
2843 __ protect_return_address(r20);
2844 __ str(r20, Address(rfp, wordSize));
2845 }
2846
2847 __ bind(no_adjust);
2848 // Normal exit, restore registers and exit.
2849 reg_save.restore_live_registers(masm);
2850
2851 __ ret(lr);
2852
2853 #ifdef ASSERT
2854 __ bind(bail);
2855 __ stop("Attempting to adjust pc to skip safepoint poll but the return point is not what we expected");
2856 #endif
2857
2858 // Make sure all code is generated
2859 masm->flush();
2860
2861 // Fill-out other meta info
2862 return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words);
2863 }
2864
2865 //
2866 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss
2867 //
2868 // Generate a stub that calls into vm to find out the proper destination
2869 // of a java call. All the argument registers are live at this point
2870 // but since this is generic code we don't know what they are and the caller
2871 // must do any gc of the args.
2872 //
2873 RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) {
2874 assert (StubRoutines::forward_exception_entry() != nullptr, "must be generated before");
2875
2876 // allocate space for the code
2877 ResourceMark rm;
2878
2879 CodeBuffer buffer(name, 1000, 512);
2880 MacroAssembler* masm = new MacroAssembler(&buffer);
2881
2882 int frame_size_in_words;
2883 RegisterSaver reg_save(false /* save_vectors */);
2884
2885 OopMapSet *oop_maps = new OopMapSet();
2886 OopMap* map = nullptr;
2887
2888 int start = __ offset();
2889
2890 map = reg_save.save_live_registers(masm, 0, &frame_size_in_words);
2891
2892 int frame_complete = __ offset();
2893
2894 {
2895 Label retaddr;
2896 __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
2897
2898 __ mov(c_rarg0, rthread);
2899 __ lea(rscratch1, RuntimeAddress(destination));
2900
2901 __ blr(rscratch1);
2902 __ bind(retaddr);
2903 }
2904
2905 // Set an oopmap for the call site.
2906 // We need this not only for callee-saved registers, but also for volatile
2907 // registers that the compiler might be keeping live across a safepoint.
2908
2909 oop_maps->add_gc_map( __ offset() - start, map);
2910
2911 // r0 contains the address we are going to jump to assuming no exception got installed
2912
2913 // clear last_Java_sp
2914 __ reset_last_Java_frame(false);
2915 // check for pending exceptions
2916 Label pending;
2917 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
2918 __ cbnz(rscratch1, pending);
2919
2920 // get the returned Method*
2921 __ get_vm_result_2(rmethod, rthread);
2922 __ str(rmethod, Address(sp, reg_save.reg_offset_in_bytes(rmethod)));
2923
2924 // r0 is where we want to jump, overwrite rscratch1 which is saved and scratch
2925 __ str(r0, Address(sp, reg_save.rscratch1_offset_in_bytes()));
2926 reg_save.restore_live_registers(masm);
2927
2928 // We are back to the original state on entry and ready to go.
2929
2930 __ br(rscratch1);
2931
2932 // Pending exception after the safepoint
2933
2934 __ bind(pending);
2935
2936 reg_save.restore_live_registers(masm);
2937
2938 // exception pending => remove activation and forward to exception handler
2939
2940 __ str(zr, Address(rthread, JavaThread::vm_result_offset()));
2941
2942 __ ldr(r0, Address(rthread, Thread::pending_exception_offset()));
2943 __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2944
2945 // -------------
2946 // make sure all code is generated
2947 masm->flush();
2948
2949 // return the blob
2950 // frame_size_words or bytes??
2951 return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true);
2952 }
2953
2954 #ifdef COMPILER2
2955 // This is here instead of runtime_aarch64_64.cpp because it uses SimpleRuntimeFrame
2956 //
2957 //------------------------------generate_exception_blob---------------------------
2958 // creates exception blob at the end
2959 // Using exception blob, this code is jumped from a compiled method.
2960 // (see emit_exception_handler in x86_64.ad file)
2961 //
2962 // Given an exception pc at a call we call into the runtime for the
2963 // handler in this method. This handler might merely restore state
2964 // (i.e. callee save registers) unwind the frame and jump to the
2965 // exception handler for the nmethod if there is no Java level handler
2966 // for the nmethod.
2967 //
2968 // This code is entered with a jmp.
2969 //
2970 // Arguments:
2971 // r0: exception oop
2972 // r3: exception pc
2973 //
2974 // Results:
2975 // r0: exception oop
2976 // r3: exception pc in caller or ???
2977 // destination: exception handler of caller
2978 //
2979 // Note: the exception pc MUST be at a call (precise debug information)
2980 // Registers r0, r3, r2, r4, r5, r8-r11 are not callee saved.
2981 //
2982
2983 void OptoRuntime::generate_exception_blob() {
2984 assert(!OptoRuntime::is_callee_saved_register(R3_num), "");
2985 assert(!OptoRuntime::is_callee_saved_register(R0_num), "");
2986 assert(!OptoRuntime::is_callee_saved_register(R2_num), "");
2987
2988 assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
2989
2990 // Allocate space for the code
2991 ResourceMark rm;
2992 // Setup code generation tools
2993 CodeBuffer buffer("exception_blob", 2048, 1024);
2994 MacroAssembler* masm = new MacroAssembler(&buffer);
2995
2996 // TODO check various assumptions made here
2997 //
2998 // make sure we do so before running this
2999
3000 address start = __ pc();
3001
3002 // push rfp and retaddr by hand
3003 // Exception pc is 'return address' for stack walker
3004 __ protect_return_address();
3005 __ stp(rfp, lr, Address(__ pre(sp, -2 * wordSize)));
3006 // there are no callee save registers and we don't expect an
3007 // arg reg save area
3008 #ifndef PRODUCT
3009 assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
3010 #endif
3011 // Store exception in Thread object. We cannot pass any arguments to the
3012 // handle_exception call, since we do not want to make any assumption
3013 // about the size of the frame where the exception happened in.
3014 __ str(r0, Address(rthread, JavaThread::exception_oop_offset()));
3015 __ str(r3, Address(rthread, JavaThread::exception_pc_offset()));
3016
3017 // This call does all the hard work. It checks if an exception handler
3018 // exists in the method.
3019 // If so, it returns the handler address.
3020 // If not, it prepares for stack-unwinding, restoring the callee-save
3021 // registers of the frame being removed.
3022 //
3023 // address OptoRuntime::handle_exception_C(JavaThread* thread)
3024 //
3025 // n.b. 1 gp arg, 0 fp args, integral return type
3026
3027 // the stack should always be aligned
3028 address the_pc = __ pc();
3029 __ set_last_Java_frame(sp, noreg, the_pc, rscratch1);
3030 __ mov(c_rarg0, rthread);
3031 __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C)));
3032 __ blr(rscratch1);
3033 // handle_exception_C is a special VM call which does not require an explicit
3034 // instruction sync afterwards.
3035
3036 // May jump to SVE compiled code
3037 __ reinitialize_ptrue();
3038
3039 // Set an oopmap for the call site. This oopmap will only be used if we
3040 // are unwinding the stack. Hence, all locations will be dead.
3041 // Callee-saved registers will be the same as the frame above (i.e.,
3042 // handle_exception_stub), since they were restored when we got the
3043 // exception.
3044
3045 OopMapSet* oop_maps = new OopMapSet();
3046
3047 oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
3048
3049 __ reset_last_Java_frame(false);
3050
3051 // Restore callee-saved registers
3052
3053 // rfp is an implicitly saved callee saved register (i.e. the calling
3054 // convention will save restore it in prolog/epilog) Other than that
3055 // there are no callee save registers now that adapter frames are gone.
3056 // and we dont' expect an arg reg save area
3057 __ ldp(rfp, r3, Address(__ post(sp, 2 * wordSize)));
3058 __ authenticate_return_address(r3);
3059
3060 // r0: exception handler
3061
3062 // We have a handler in r0 (could be deopt blob).
3063 __ mov(r8, r0);
3064
3065 // Get the exception oop
3066 __ ldr(r0, Address(rthread, JavaThread::exception_oop_offset()));
3067 // Get the exception pc in case we are deoptimized
3068 __ ldr(r4, Address(rthread, JavaThread::exception_pc_offset()));
3069 #ifdef ASSERT
3070 __ str(zr, Address(rthread, JavaThread::exception_handler_pc_offset()));
3071 __ str(zr, Address(rthread, JavaThread::exception_pc_offset()));
3072 #endif
3073 // Clear the exception oop so GC no longer processes it as a root.
3074 __ str(zr, Address(rthread, JavaThread::exception_oop_offset()));
3075
3076 // r0: exception oop
3077 // r8: exception handler
3078 // r4: exception pc
3079 // Jump to handler
3080
3081 __ br(r8);
3082
3083 // Make sure all code is generated
3084 masm->flush();
3085
3086 // Set exception blob
3087 _exception_blob = ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
3088 }
3089
3090 #endif // COMPILER2