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