1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "c1/c1_CodeStubs.hpp"
26 #include "c1/c1_FrameMap.hpp"
27 #include "c1/c1_LIRAssembler.hpp"
28 #include "c1/c1_MacroAssembler.hpp"
29 #include "c1/c1_Runtime1.hpp"
30 #include "classfile/javaClasses.hpp"
31 #include "nativeInst_x86.hpp"
32 #include "runtime/sharedRuntime.hpp"
33 #include "utilities/align.hpp"
34 #include "utilities/macros.hpp"
35 #include "vmreg_x86.inline.hpp"
36
37
38 #define __ ce->masm()->
39
40 #ifndef _LP64
41 float ConversionStub::float_zero = 0.0;
42 double ConversionStub::double_zero = 0.0;
43
44 void ConversionStub::emit_code(LIR_Assembler* ce) {
45 __ bind(_entry);
46 assert(bytecode() == Bytecodes::_f2i || bytecode() == Bytecodes::_d2i, "other conversions do not require stub");
47
48
49 if (input()->is_single_xmm()) {
50 __ comiss(input()->as_xmm_float_reg(),
51 ExternalAddress((address)&float_zero));
52 } else if (input()->is_double_xmm()) {
53 __ comisd(input()->as_xmm_double_reg(),
54 ExternalAddress((address)&double_zero));
55 } else {
56 __ push(rax);
57 __ ftst();
58 __ fnstsw_ax();
59 __ sahf();
60 __ pop(rax);
61 }
62
63 Label NaN, do_return;
64 __ jccb(Assembler::parity, NaN);
65 __ jccb(Assembler::below, do_return);
66
67 // input is > 0 -> return maxInt
68 // result register already contains 0x80000000, so subtracting 1 gives 0x7fffffff
69 __ decrement(result()->as_register());
70 __ jmpb(do_return);
71
72 // input is NaN -> return 0
73 __ bind(NaN);
74 __ xorptr(result()->as_register(), result()->as_register());
75
76 __ bind(do_return);
77 __ jmp(_continuation);
78 }
79 #endif // !_LP64
80
81 void C1SafepointPollStub::emit_code(LIR_Assembler* ce) {
82 __ bind(_entry);
83 InternalAddress safepoint_pc(ce->masm()->pc() - ce->masm()->offset() + safepoint_offset());
84 #ifdef _LP64
85 __ lea(rscratch1, safepoint_pc);
86 __ movptr(Address(r15_thread, JavaThread::saved_exception_pc_offset()), rscratch1);
87 #else
88 const Register tmp1 = rcx;
89 const Register tmp2 = rdx;
90 __ push(tmp1);
91 __ push(tmp2);
92
93 __ lea(tmp1, safepoint_pc);
94 __ get_thread(tmp2);
95 __ movptr(Address(tmp2, JavaThread::saved_exception_pc_offset()), tmp1);
96
97 __ pop(tmp2);
98 __ pop(tmp1);
99 #endif /* _LP64 */
100 assert(SharedRuntime::polling_page_return_handler_blob() != nullptr,
101 "polling page return stub not created yet");
102
103 address stub = SharedRuntime::polling_page_return_handler_blob()->entry_point();
104 __ jump(RuntimeAddress(stub));
105 }
106
107 void CounterOverflowStub::emit_code(LIR_Assembler* ce) {
108 __ bind(_entry);
109 Metadata *m = _method->as_constant_ptr()->as_metadata();
110 ce->store_parameter(m, 1);
111 ce->store_parameter(_bci, 0);
112 __ call(RuntimeAddress(Runtime1::entry_for(C1StubId::counter_overflow_id)));
113 ce->add_call_info_here(_info);
114 ce->verify_oop_map(_info);
115 __ jmp(_continuation);
116 }
117
118 void RangeCheckStub::emit_code(LIR_Assembler* ce) {
119 __ bind(_entry);
120 if (_info->deoptimize_on_exception()) {
121 address a = Runtime1::entry_for(C1StubId::predicate_failed_trap_id);
122 __ call(RuntimeAddress(a));
123 ce->add_call_info_here(_info);
124 ce->verify_oop_map(_info);
125 debug_only(__ should_not_reach_here());
126 return;
127 }
128
129 // pass the array index on stack because all registers must be preserved
130 if (_index->is_cpu_register()) {
131 ce->store_parameter(_index->as_register(), 0);
132 } else {
133 ce->store_parameter(_index->as_jint(), 0);
134 }
135 C1StubId stub_id;
136 if (_throw_index_out_of_bounds_exception) {
137 stub_id = C1StubId::throw_index_exception_id;
138 } else {
139 stub_id = C1StubId::throw_range_check_failed_id;
140 ce->store_parameter(_array->as_pointer_register(), 1);
141 }
142 __ call(RuntimeAddress(Runtime1::entry_for(stub_id)));
143 ce->add_call_info_here(_info);
144 ce->verify_oop_map(_info);
145 debug_only(__ should_not_reach_here());
146 }
147
148 PredicateFailedStub::PredicateFailedStub(CodeEmitInfo* info) {
149 _info = new CodeEmitInfo(info);
150 }
151
152 void PredicateFailedStub::emit_code(LIR_Assembler* ce) {
153 __ bind(_entry);
154 address a = Runtime1::entry_for(C1StubId::predicate_failed_trap_id);
155 __ call(RuntimeAddress(a));
156 ce->add_call_info_here(_info);
157 ce->verify_oop_map(_info);
158 debug_only(__ should_not_reach_here());
159 }
160
161 void DivByZeroStub::emit_code(LIR_Assembler* ce) {
162 if (_offset != -1) {
163 ce->compilation()->implicit_exception_table()->append(_offset, __ offset());
164 }
165 __ bind(_entry);
166 __ call(RuntimeAddress(Runtime1::entry_for(C1StubId::throw_div0_exception_id)));
167 ce->add_call_info_here(_info);
168 debug_only(__ should_not_reach_here());
169 }
170
171
172 // Implementation of NewInstanceStub
173
174 NewInstanceStub::NewInstanceStub(LIR_Opr klass_reg, LIR_Opr result, ciInstanceKlass* klass, CodeEmitInfo* info, C1StubId stub_id) {
175 _result = result;
176 _klass = klass;
177 _klass_reg = klass_reg;
178 _info = new CodeEmitInfo(info);
179 assert(stub_id == C1StubId::new_instance_id ||
180 stub_id == C1StubId::fast_new_instance_id ||
181 stub_id == C1StubId::fast_new_instance_init_check_id,
182 "need new_instance id");
183 _stub_id = stub_id;
184 }
185
186
187 void NewInstanceStub::emit_code(LIR_Assembler* ce) {
188 assert(__ rsp_offset() == 0, "frame size should be fixed");
189 __ bind(_entry);
190 __ movptr(rdx, _klass_reg->as_register());
191 __ call(RuntimeAddress(Runtime1::entry_for(_stub_id)));
192 ce->add_call_info_here(_info);
193 ce->verify_oop_map(_info);
194 assert(_result->as_register() == rax, "result must in rax,");
195 __ jmp(_continuation);
196 }
197
198
199 // Implementation of NewTypeArrayStub
200
201 NewTypeArrayStub::NewTypeArrayStub(LIR_Opr klass_reg, LIR_Opr length, LIR_Opr result, CodeEmitInfo* info) {
202 _klass_reg = klass_reg;
203 _length = length;
204 _result = result;
205 _info = new CodeEmitInfo(info);
206 }
207
208
209 void NewTypeArrayStub::emit_code(LIR_Assembler* ce) {
210 assert(__ rsp_offset() == 0, "frame size should be fixed");
211 __ bind(_entry);
212 assert(_length->as_register() == rbx, "length must in rbx,");
213 assert(_klass_reg->as_register() == rdx, "klass_reg must in rdx");
214 __ call(RuntimeAddress(Runtime1::entry_for(C1StubId::new_type_array_id)));
215 ce->add_call_info_here(_info);
216 ce->verify_oop_map(_info);
217 assert(_result->as_register() == rax, "result must in rax,");
218 __ jmp(_continuation);
219 }
220
221
222 // Implementation of NewObjectArrayStub
223
224 NewObjectArrayStub::NewObjectArrayStub(LIR_Opr klass_reg, LIR_Opr length, LIR_Opr result, CodeEmitInfo* info) {
225 _klass_reg = klass_reg;
226 _result = result;
227 _length = length;
228 _info = new CodeEmitInfo(info);
229 }
230
231
232 void NewObjectArrayStub::emit_code(LIR_Assembler* ce) {
233 assert(__ rsp_offset() == 0, "frame size should be fixed");
234 __ bind(_entry);
235 assert(_length->as_register() == rbx, "length must in rbx,");
236 assert(_klass_reg->as_register() == rdx, "klass_reg must in rdx");
237 __ call(RuntimeAddress(Runtime1::entry_for(C1StubId::new_object_array_id)));
238 ce->add_call_info_here(_info);
239 ce->verify_oop_map(_info);
240 assert(_result->as_register() == rax, "result must in rax,");
241 __ jmp(_continuation);
242 }
243
244 void MonitorEnterStub::emit_code(LIR_Assembler* ce) {
245 assert(__ rsp_offset() == 0, "frame size should be fixed");
246 __ bind(_entry);
247 ce->store_parameter(_obj_reg->as_register(), 1);
248 ce->store_parameter(_lock_reg->as_register(), 0);
249 C1StubId enter_id;
250 if (ce->compilation()->has_fpu_code()) {
251 enter_id = C1StubId::monitorenter_id;
252 } else {
253 enter_id = C1StubId::monitorenter_nofpu_id;
254 }
255 __ call(RuntimeAddress(Runtime1::entry_for(enter_id)));
256 ce->add_call_info_here(_info);
257 ce->verify_oop_map(_info);
258 __ jmp(_continuation);
259 }
260
261
262 void MonitorExitStub::emit_code(LIR_Assembler* ce) {
263 __ bind(_entry);
264 if (_compute_lock) {
265 // lock_reg was destroyed by fast unlocking attempt => recompute it
266 ce->monitor_address(_monitor_ix, _lock_reg);
267 }
268 ce->store_parameter(_lock_reg->as_register(), 0);
269 // note: non-blocking leaf routine => no call info needed
270 C1StubId exit_id;
271 if (ce->compilation()->has_fpu_code()) {
272 exit_id = C1StubId::monitorexit_id;
273 } else {
274 exit_id = C1StubId::monitorexit_nofpu_id;
275 }
276 __ call(RuntimeAddress(Runtime1::entry_for(exit_id)));
277 __ jmp(_continuation);
278 }
279
280
281 // Implementation of patching:
282 // - Copy the code at given offset to an inlined buffer (first the bytes, then the number of bytes)
283 // - Replace original code with a call to the stub
284 // At Runtime:
285 // - call to stub, jump to runtime
286 // - in runtime: preserve all registers (rspecially objects, i.e., source and destination object)
287 // - in runtime: after initializing class, restore original code, reexecute instruction
288
289 int PatchingStub::_patch_info_offset = -NativeGeneralJump::instruction_size;
290
291 void PatchingStub::align_patch_site(MacroAssembler* masm) {
292 // We're patching a 5-7 byte instruction on intel and we need to
293 // make sure that we don't see a piece of the instruction. It
294 // appears mostly impossible on Intel to simply invalidate other
295 // processors caches and since they may do aggressive prefetch it's
296 // very hard to make a guess about what code might be in the icache.
297 // Force the instruction to be double word aligned so that it
298 // doesn't span a cache line.
299 masm->align(align_up((int)NativeGeneralJump::instruction_size, wordSize));
300 }
301
302 void PatchingStub::emit_code(LIR_Assembler* ce) {
303 assert(NativeCall::instruction_size <= _bytes_to_copy && _bytes_to_copy <= 0xFF, "not enough room for call");
304
305 Label call_patch;
306
307 // static field accesses have special semantics while the class
308 // initializer is being run so we emit a test which can be used to
309 // check that this code is being executed by the initializing
310 // thread.
311 address being_initialized_entry = __ pc();
312 if (CommentedAssembly) {
313 __ block_comment(" patch template");
314 }
315 if (_id == load_klass_id) {
316 // produce a copy of the load klass instruction for use by the being initialized case
317 #ifdef ASSERT
318 address start = __ pc();
319 #endif
320 Metadata* o = nullptr;
321 __ mov_metadata(_obj, o);
322 #ifdef ASSERT
323 for (int i = 0; i < _bytes_to_copy; i++) {
324 address ptr = (address)(_pc_start + i);
325 int a_byte = (*ptr) & 0xFF;
326 assert(a_byte == *start++, "should be the same code");
327 }
328 #endif
329 } else if (_id == load_mirror_id) {
330 // produce a copy of the load mirror instruction for use by the being
331 // initialized case
332 #ifdef ASSERT
333 address start = __ pc();
334 #endif
335 jobject o = nullptr;
336 __ movoop(_obj, o);
337 #ifdef ASSERT
338 for (int i = 0; i < _bytes_to_copy; i++) {
339 address ptr = (address)(_pc_start + i);
340 int a_byte = (*ptr) & 0xFF;
341 assert(a_byte == *start++, "should be the same code");
342 }
343 #endif
344 } else {
345 // make a copy the code which is going to be patched.
346 for (int i = 0; i < _bytes_to_copy; i++) {
347 address ptr = (address)(_pc_start + i);
348 int a_byte = (*ptr) & 0xFF;
349 __ emit_int8(a_byte);
350 *ptr = 0x90; // make the site look like a nop
351 }
352 }
353
354 address end_of_patch = __ pc();
355 int bytes_to_skip = 0;
356 if (_id == load_mirror_id) {
357 int offset = __ offset();
358 if (CommentedAssembly) {
359 __ block_comment(" being_initialized check");
360 }
361 assert(_obj != noreg, "must be a valid register");
362 Register tmp = rax;
363 __ push(tmp);
364 __ movptr(tmp, Address(_obj, java_lang_Class::klass_offset()));
365 __ cmpptr(r15_thread, Address(tmp, InstanceKlass::init_thread_offset()));
366 __ pop(tmp); // pop it right away, no matter which path we take
367 __ jccb(Assembler::notEqual, call_patch);
368
369 // access_field patches may execute the patched code before it's
370 // copied back into place so we need to jump back into the main
371 // code of the nmethod to continue execution.
372 __ jmp(_patch_site_continuation);
373
374 // make sure this extra code gets skipped
375 bytes_to_skip += __ offset() - offset;
376 }
377 if (CommentedAssembly) {
378 __ block_comment("patch data encoded as movl");
379 }
380 // Now emit the patch record telling the runtime how to find the
381 // pieces of the patch. We only need 3 bytes but for readability of
382 // the disassembly we make the data look like a movl reg, imm32,
383 // which requires 5 bytes
384 int sizeof_patch_record = 5;
385 bytes_to_skip += sizeof_patch_record;
386
387 // emit the offsets needed to find the code to patch
388 int being_initialized_entry_offset = __ pc() - being_initialized_entry + sizeof_patch_record;
389
390 __ emit_int8((unsigned char)0xB8);
391 __ emit_int8(0);
392 __ emit_int8(being_initialized_entry_offset);
393 __ emit_int8(bytes_to_skip);
394 __ emit_int8(_bytes_to_copy);
395 address patch_info_pc = __ pc();
396 assert(patch_info_pc - end_of_patch == bytes_to_skip, "incorrect patch info");
397
398 address entry = __ pc();
399 NativeGeneralJump::insert_unconditional((address)_pc_start, entry);
400 address target = nullptr;
401 relocInfo::relocType reloc_type = relocInfo::none;
402 switch (_id) {
403 case access_field_id: target = Runtime1::entry_for(C1StubId::access_field_patching_id); break;
404 case load_klass_id: target = Runtime1::entry_for(C1StubId::load_klass_patching_id); reloc_type = relocInfo::metadata_type; break;
405 case load_mirror_id: target = Runtime1::entry_for(C1StubId::load_mirror_patching_id); reloc_type = relocInfo::oop_type; break;
406 case load_appendix_id: target = Runtime1::entry_for(C1StubId::load_appendix_patching_id); reloc_type = relocInfo::oop_type; break;
407 default: ShouldNotReachHere();
408 }
409 __ bind(call_patch);
410
411 if (CommentedAssembly) {
412 __ block_comment("patch entry point");
413 }
414 __ call(RuntimeAddress(target));
415 assert(_patch_info_offset == (patch_info_pc - __ pc()), "must not change");
416 ce->add_call_info_here(_info);
417 int jmp_off = __ offset();
418 __ jmp(_patch_site_entry);
419 // Add enough nops so deoptimization can overwrite the jmp above with a call
420 // and not destroy the world. We cannot use fat nops here, since the concurrent
421 // code rewrite may transiently create the illegal instruction sequence.
422 for (int j = __ offset() ; j < jmp_off + 5 ; j++ ) {
423 __ nop();
424 }
425 if (_id == load_klass_id || _id == load_mirror_id || _id == load_appendix_id) {
426 CodeSection* cs = __ code_section();
427 RelocIterator iter(cs, (address)_pc_start, (address)(_pc_start + 1));
428 relocInfo::change_reloc_info_for_address(&iter, (address) _pc_start, reloc_type, relocInfo::none);
429 }
430 }
431
432
433 void DeoptimizeStub::emit_code(LIR_Assembler* ce) {
434 __ bind(_entry);
435 ce->store_parameter(_trap_request, 0);
436 __ call(RuntimeAddress(Runtime1::entry_for(C1StubId::deoptimize_id)));
437 ce->add_call_info_here(_info);
438 DEBUG_ONLY(__ should_not_reach_here());
439 }
440
441
442 void ImplicitNullCheckStub::emit_code(LIR_Assembler* ce) {
443 address a;
444 if (_info->deoptimize_on_exception()) {
445 // Deoptimize, do not throw the exception, because it is probably wrong to do it here.
446 a = Runtime1::entry_for(C1StubId::predicate_failed_trap_id);
447 } else {
448 a = Runtime1::entry_for(C1StubId::throw_null_pointer_exception_id);
449 }
450
451 ce->compilation()->implicit_exception_table()->append(_offset, __ offset());
452 __ bind(_entry);
453 __ call(RuntimeAddress(a));
454 ce->add_call_info_here(_info);
455 ce->verify_oop_map(_info);
456 debug_only(__ should_not_reach_here());
457 }
458
459
460 void SimpleExceptionStub::emit_code(LIR_Assembler* ce) {
461 assert(__ rsp_offset() == 0, "frame size should be fixed");
462
463 __ bind(_entry);
464 // pass the object on stack because all registers must be preserved
465 if (_obj->is_cpu_register()) {
466 ce->store_parameter(_obj->as_register(), 0);
467 }
468 __ call(RuntimeAddress(Runtime1::entry_for(_stub)));
469 ce->add_call_info_here(_info);
470 debug_only(__ should_not_reach_here());
471 }
472
473
474 void ArrayCopyStub::emit_code(LIR_Assembler* ce) {
475 //---------------slow case: call to native-----------------
476 __ bind(_entry);
477 // Figure out where the args should go
478 // This should really convert the IntrinsicID to the Method* and signature
479 // but I don't know how to do that.
480 //
481 VMRegPair args[5];
482 BasicType signature[5] = { T_OBJECT, T_INT, T_OBJECT, T_INT, T_INT};
483 SharedRuntime::java_calling_convention(signature, args, 5);
484
485 // push parameters
486 // (src, src_pos, dest, destPos, length)
487 Register r[5];
488 r[0] = src()->as_register();
489 r[1] = src_pos()->as_register();
490 r[2] = dst()->as_register();
491 r[3] = dst_pos()->as_register();
492 r[4] = length()->as_register();
493
494 // next registers will get stored on the stack
495 for (int i = 0; i < 5 ; i++ ) {
496 VMReg r_1 = args[i].first();
497 if (r_1->is_stack()) {
498 int st_off = r_1->reg2stack() * wordSize;
499 __ movptr (Address(rsp, st_off), r[i]);
500 } else {
501 assert(r[i] == args[i].first()->as_Register(), "Wrong register for arg ");
502 }
503 }
504
505 ce->align_call(lir_static_call);
506
507 ce->emit_static_call_stub();
508 if (ce->compilation()->bailed_out()) {
509 return; // CodeCache is full
510 }
511 AddressLiteral resolve(SharedRuntime::get_resolve_static_call_stub(),
512 relocInfo::static_call_type);
513 __ call(resolve);
514 ce->add_call_info_here(info());
515
516 #ifndef PRODUCT
517 if (PrintC1Statistics) {
518 __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_slowcase_cnt), rscratch1);
519 }
520 #endif
521
522 __ jmp(_continuation);
523 }
524
525 #undef __