1 /*
  2  * Copyright (c) 1999, 2023, 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 "precompiled.hpp"
 26 #include "c1/c1_CodeStubs.hpp"
 27 #include "c1/c1_FrameMap.hpp"
 28 #include "c1/c1_LIRAssembler.hpp"
 29 #include "c1/c1_MacroAssembler.hpp"
 30 #include "c1/c1_Runtime1.hpp"
 31 #include "classfile/javaClasses.hpp"
 32 #include "nativeInst_x86.hpp"
 33 #include "runtime/sharedRuntime.hpp"
 34 #include "utilities/align.hpp"
 35 #include "utilities/macros.hpp"
 36 #include "vmreg_x86.inline.hpp"
 37 
 38 
 39 #define __ ce->masm()->
 40 
 41 #ifndef _LP64
 42 float ConversionStub::float_zero = 0.0;
 43 double ConversionStub::double_zero = 0.0;
 44 
 45 void ConversionStub::emit_code(LIR_Assembler* ce) {
 46   __ bind(_entry);
 47   assert(bytecode() == Bytecodes::_f2i || bytecode() == Bytecodes::_d2i, "other conversions do not require stub");
 48 
 49 
 50   if (input()->is_single_xmm()) {
 51     __ comiss(input()->as_xmm_float_reg(),
 52               ExternalAddress((address)&float_zero));
 53   } else if (input()->is_double_xmm()) {
 54     __ comisd(input()->as_xmm_double_reg(),
 55               ExternalAddress((address)&double_zero));
 56   } else {
 57     __ push(rax);
 58     __ ftst();
 59     __ fnstsw_ax();
 60     __ sahf();
 61     __ pop(rax);
 62   }
 63 
 64   Label NaN, do_return;
 65   __ jccb(Assembler::parity, NaN);
 66   __ jccb(Assembler::below, do_return);
 67 
 68   // input is > 0 -> return maxInt
 69   // result register already contains 0x80000000, so subtracting 1 gives 0x7fffffff
 70   __ decrement(result()->as_register());
 71   __ jmpb(do_return);
 72 
 73   // input is NaN -> return 0
 74   __ bind(NaN);
 75   __ xorptr(result()->as_register(), result()->as_register());
 76 
 77   __ bind(do_return);
 78   __ jmp(_continuation);
 79 }
 80 #endif // !_LP64
 81 
 82 void C1SafepointPollStub::emit_code(LIR_Assembler* ce) {
 83   __ bind(_entry);
 84   InternalAddress safepoint_pc(ce->masm()->pc() - ce->masm()->offset() + safepoint_offset());
 85 #ifdef _LP64
 86   __ lea(rscratch1, safepoint_pc);
 87   __ movptr(Address(r15_thread, JavaThread::saved_exception_pc_offset()), rscratch1);
 88 #else
 89   const Register tmp1 = rcx;
 90   const Register tmp2 = rdx;
 91   __ push(tmp1);
 92   __ push(tmp2);
 93 
 94   __ lea(tmp1, safepoint_pc);
 95   __ get_thread(tmp2);
 96   __ movptr(Address(tmp2, JavaThread::saved_exception_pc_offset()), tmp1);
 97 
 98   __ pop(tmp2);
 99   __ pop(tmp1);
100 #endif /* _LP64 */
101   assert(SharedRuntime::polling_page_return_handler_blob() != nullptr,
102          "polling page return stub not created yet");
103 
104   address stub = SharedRuntime::polling_page_return_handler_blob()->entry_point();
105   __ jump(RuntimeAddress(stub));
106 }
107 
108 void CounterOverflowStub::emit_code(LIR_Assembler* ce) {
109   __ bind(_entry);
110   Metadata *m = _method->as_constant_ptr()->as_metadata();
111   ce->store_parameter(m, 1);
112   ce->store_parameter(_bci, 0);
113   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::counter_overflow_id)));
114   ce->add_call_info_here(_info);
115   ce->verify_oop_map(_info);
116   __ jmp(_continuation);
117 }
118 
119 void RangeCheckStub::emit_code(LIR_Assembler* ce) {
120   __ bind(_entry);
121   if (_info->deoptimize_on_exception()) {
122     address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id);
123     __ call(RuntimeAddress(a));
124     ce->add_call_info_here(_info);
125     ce->verify_oop_map(_info);
126     debug_only(__ should_not_reach_here());
127     return;
128   }
129 
130   // pass the array index on stack because all registers must be preserved
131   if (_index->is_cpu_register()) {
132     ce->store_parameter(_index->as_register(), 0);
133   } else {
134     ce->store_parameter(_index->as_jint(), 0);
135   }
136   Runtime1::StubID stub_id;
137   if (_throw_index_out_of_bounds_exception) {
138     stub_id = Runtime1::throw_index_exception_id;
139   } else {
140     stub_id = Runtime1::throw_range_check_failed_id;
141     ce->store_parameter(_array->as_pointer_register(), 1);
142   }
143   __ call(RuntimeAddress(Runtime1::entry_for(stub_id)));
144   ce->add_call_info_here(_info);
145   ce->verify_oop_map(_info);
146   debug_only(__ should_not_reach_here());
147 }
148 
149 PredicateFailedStub::PredicateFailedStub(CodeEmitInfo* info) {
150   _info = new CodeEmitInfo(info);
151 }
152 
153 void PredicateFailedStub::emit_code(LIR_Assembler* ce) {
154   __ bind(_entry);
155   address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id);
156   __ call(RuntimeAddress(a));
157   ce->add_call_info_here(_info);
158   ce->verify_oop_map(_info);
159   debug_only(__ should_not_reach_here());
160 }
161 
162 void DivByZeroStub::emit_code(LIR_Assembler* ce) {
163   if (_offset != -1) {
164     ce->compilation()->implicit_exception_table()->append(_offset, __ offset());
165   }
166   __ bind(_entry);
167   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::throw_div0_exception_id)));
168   ce->add_call_info_here(_info);
169   debug_only(__ should_not_reach_here());
170 }
171 
172 
173 // Implementation of NewInstanceStub
174 
175 NewInstanceStub::NewInstanceStub(LIR_Opr klass_reg, LIR_Opr result, ciInstanceKlass* klass, CodeEmitInfo* info, Runtime1::StubID stub_id) {
176   _result = result;
177   _klass = klass;
178   _klass_reg = klass_reg;
179   _info = new CodeEmitInfo(info);
180   assert(stub_id == Runtime1::new_instance_id                 ||
181          stub_id == Runtime1::fast_new_instance_id            ||
182          stub_id == Runtime1::fast_new_instance_init_check_id,
183          "need new_instance id");
184   _stub_id   = stub_id;
185 }
186 
187 
188 void NewInstanceStub::emit_code(LIR_Assembler* ce) {
189   assert(__ rsp_offset() == 0, "frame size should be fixed");
190   __ bind(_entry);
191   __ movptr(rdx, _klass_reg->as_register());
192   __ call(RuntimeAddress(Runtime1::entry_for(_stub_id)));
193   ce->add_call_info_here(_info);
194   ce->verify_oop_map(_info);
195   assert(_result->as_register() == rax, "result must in rax,");
196   __ jmp(_continuation);
197 }
198 
199 
200 // Implementation of NewTypeArrayStub
201 
202 NewTypeArrayStub::NewTypeArrayStub(LIR_Opr klass_reg, LIR_Opr length, LIR_Opr result, CodeEmitInfo* info) {
203   _klass_reg = klass_reg;
204   _length = length;
205   _result = result;
206   _info = new CodeEmitInfo(info);
207 }
208 
209 
210 void NewTypeArrayStub::emit_code(LIR_Assembler* ce) {
211   assert(__ rsp_offset() == 0, "frame size should be fixed");
212   __ bind(_entry);
213   assert(_length->as_register() == rbx, "length must in rbx,");
214   assert(_klass_reg->as_register() == rdx, "klass_reg must in rdx");
215   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::new_type_array_id)));
216   ce->add_call_info_here(_info);
217   ce->verify_oop_map(_info);
218   assert(_result->as_register() == rax, "result must in rax,");
219   __ jmp(_continuation);
220 }
221 
222 
223 // Implementation of NewObjectArrayStub
224 
225 NewObjectArrayStub::NewObjectArrayStub(LIR_Opr klass_reg, LIR_Opr length, LIR_Opr result, CodeEmitInfo* info) {
226   _klass_reg = klass_reg;
227   _result = result;
228   _length = length;
229   _info = new CodeEmitInfo(info);
230 }
231 
232 
233 void NewObjectArrayStub::emit_code(LIR_Assembler* ce) {
234   assert(__ rsp_offset() == 0, "frame size should be fixed");
235   __ bind(_entry);
236   assert(_length->as_register() == rbx, "length must in rbx,");
237   assert(_klass_reg->as_register() == rdx, "klass_reg must in rdx");
238   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::new_object_array_id)));
239   ce->add_call_info_here(_info);
240   ce->verify_oop_map(_info);
241   assert(_result->as_register() == rax, "result must in rax,");
242   __ jmp(_continuation);
243 }
244 
245 void MonitorEnterStub::emit_code(LIR_Assembler* ce) {
246   assert(__ rsp_offset() == 0, "frame size should be fixed");
247   __ bind(_entry);
248   ce->store_parameter(_obj_reg->as_register(),  1);
249   ce->store_parameter(_lock_reg->as_register(), 0);
250   Runtime1::StubID enter_id;
251   if (ce->compilation()->has_fpu_code()) {
252     enter_id = Runtime1::monitorenter_id;
253   } else {
254     enter_id = Runtime1::monitorenter_nofpu_id;
255   }
256   __ call(RuntimeAddress(Runtime1::entry_for(enter_id)));
257   ce->add_call_info_here(_info);
258   ce->verify_oop_map(_info);
259   __ jmp(_continuation);
260 }
261 
262 
263 void MonitorExitStub::emit_code(LIR_Assembler* ce) {
264   __ bind(_entry);
265   if (_compute_lock) {
266     // lock_reg was destroyed by fast unlocking attempt => recompute it
267     ce->monitor_address(_monitor_ix, _lock_reg);
268   }
269   ce->store_parameter(_lock_reg->as_register(), 0);
270   // note: non-blocking leaf routine => no call info needed
271   Runtime1::StubID exit_id;
272   if (ce->compilation()->has_fpu_code()) {
273     exit_id = Runtime1::monitorexit_id;
274   } else {
275     exit_id = Runtime1::monitorexit_nofpu_id;
276   }
277   __ call(RuntimeAddress(Runtime1::entry_for(exit_id)));
278   __ jmp(_continuation);
279 }
280 
281 
282 // Implementation of patching:
283 // - Copy the code at given offset to an inlined buffer (first the bytes, then the number of bytes)
284 // - Replace original code with a call to the stub
285 // At Runtime:
286 // - call to stub, jump to runtime
287 // - in runtime: preserve all registers (rspecially objects, i.e., source and destination object)
288 // - in runtime: after initializing class, restore original code, reexecute instruction
289 
290 int PatchingStub::_patch_info_offset = -NativeGeneralJump::instruction_size;
291 
292 void PatchingStub::align_patch_site(MacroAssembler* masm) {
293   // We're patching a 5-7 byte instruction on intel and we need to
294   // make sure that we don't see a piece of the instruction.  It
295   // appears mostly impossible on Intel to simply invalidate other
296   // processors caches and since they may do aggressive prefetch it's
297   // very hard to make a guess about what code might be in the icache.
298   // Force the instruction to be double word aligned so that it
299   // doesn't span a cache line.
300   masm->align(align_up((int)NativeGeneralJump::instruction_size, wordSize));
301 }
302 
303 void PatchingStub::emit_code(LIR_Assembler* ce) {
304   assert(NativeCall::instruction_size <= _bytes_to_copy && _bytes_to_copy <= 0xFF, "not enough room for call");
305 
306   Label call_patch;
307 
308   // static field accesses have special semantics while the class
309   // initializer is being run so we emit a test which can be used to
310   // check that this code is being executed by the initializing
311   // thread.
312   address being_initialized_entry = __ pc();
313   if (CommentedAssembly) {
314     __ block_comment(" patch template");
315   }
316   if (_id == load_klass_id) {
317     // produce a copy of the load klass instruction for use by the being initialized case
318 #ifdef ASSERT
319     address start = __ pc();
320 #endif
321     Metadata* o = nullptr;
322     __ mov_metadata(_obj, o);
323 #ifdef ASSERT
324     for (int i = 0; i < _bytes_to_copy; i++) {
325       address ptr = (address)(_pc_start + i);
326       int a_byte = (*ptr) & 0xFF;
327       assert(a_byte == *start++, "should be the same code");
328     }
329 #endif
330   } else if (_id == load_mirror_id) {
331     // produce a copy of the load mirror instruction for use by the being
332     // initialized case
333 #ifdef ASSERT
334     address start = __ pc();
335 #endif
336     jobject o = nullptr;
337     __ movoop(_obj, o);
338 #ifdef ASSERT
339     for (int i = 0; i < _bytes_to_copy; i++) {
340       address ptr = (address)(_pc_start + i);
341       int a_byte = (*ptr) & 0xFF;
342       assert(a_byte == *start++, "should be the same code");
343     }
344 #endif
345   } else {
346     // make a copy the code which is going to be patched.
347     for (int i = 0; i < _bytes_to_copy; i++) {
348       address ptr = (address)(_pc_start + i);
349       int a_byte = (*ptr) & 0xFF;
350       __ emit_int8(a_byte);
351       *ptr = 0x90; // make the site look like a nop
352     }
353   }
354 
355   address end_of_patch = __ pc();
356   int bytes_to_skip = 0;
357   if (_id == load_mirror_id) {
358     int offset = __ offset();
359     if (CommentedAssembly) {
360       __ block_comment(" being_initialized check");
361     }
362     assert(_obj != noreg, "must be a valid register");
363     Register tmp = rax;
364     Register tmp2 = rbx;
365     __ push(tmp);
366     __ push(tmp2);
367     // Load without verification to keep code size small. We need it because
368     // begin_initialized_entry_offset has to fit in a byte. Also, we know it's not null.
369     __ movptr(tmp2, Address(_obj, java_lang_Class::klass_offset()));
370     __ get_thread(tmp);
371     __ cmpptr(tmp, Address(tmp2, InstanceKlass::init_thread_offset()));
372     __ pop(tmp2);
373     __ pop(tmp);
374     __ jcc(Assembler::notEqual, call_patch);
375 
376     // access_field patches may execute the patched code before it's
377     // copied back into place so we need to jump back into the main
378     // code of the nmethod to continue execution.
379     __ jmp(_patch_site_continuation);
380 
381     // make sure this extra code gets skipped
382     bytes_to_skip += __ offset() - offset;
383   }
384   if (CommentedAssembly) {
385     __ block_comment("patch data encoded as movl");
386   }
387   // Now emit the patch record telling the runtime how to find the
388   // pieces of the patch.  We only need 3 bytes but for readability of
389   // the disassembly we make the data look like a movl reg, imm32,
390   // which requires 5 bytes
391   int sizeof_patch_record = 5;
392   bytes_to_skip += sizeof_patch_record;
393 
394   // emit the offsets needed to find the code to patch
395   int being_initialized_entry_offset = __ pc() - being_initialized_entry + sizeof_patch_record;
396 
397   __ emit_int8((unsigned char)0xB8);
398   __ emit_int8(0);
399   __ emit_int8(being_initialized_entry_offset);
400   __ emit_int8(bytes_to_skip);
401   __ emit_int8(_bytes_to_copy);
402   address patch_info_pc = __ pc();
403   assert(patch_info_pc - end_of_patch == bytes_to_skip, "incorrect patch info");
404 
405   address entry = __ pc();
406   NativeGeneralJump::insert_unconditional((address)_pc_start, entry);
407   address target = nullptr;
408   relocInfo::relocType reloc_type = relocInfo::none;
409   switch (_id) {
410     case access_field_id:  target = Runtime1::entry_for(Runtime1::access_field_patching_id); break;
411     case load_klass_id:    target = Runtime1::entry_for(Runtime1::load_klass_patching_id); reloc_type = relocInfo::metadata_type; break;
412     case load_mirror_id:   target = Runtime1::entry_for(Runtime1::load_mirror_patching_id); reloc_type = relocInfo::oop_type; break;
413     case load_appendix_id:      target = Runtime1::entry_for(Runtime1::load_appendix_patching_id); reloc_type = relocInfo::oop_type; break;
414     default: ShouldNotReachHere();
415   }
416   __ bind(call_patch);
417 
418   if (CommentedAssembly) {
419     __ block_comment("patch entry point");
420   }
421   __ call(RuntimeAddress(target));
422   assert(_patch_info_offset == (patch_info_pc - __ pc()), "must not change");
423   ce->add_call_info_here(_info);
424   int jmp_off = __ offset();
425   __ jmp(_patch_site_entry);
426   // Add enough nops so deoptimization can overwrite the jmp above with a call
427   // and not destroy the world. We cannot use fat nops here, since the concurrent
428   // code rewrite may transiently create the illegal instruction sequence.
429   for (int j = __ offset() ; j < jmp_off + 5 ; j++ ) {
430     __ nop();
431   }
432   if (_id == load_klass_id || _id == load_mirror_id || _id == load_appendix_id) {
433     CodeSection* cs = __ code_section();
434     RelocIterator iter(cs, (address)_pc_start, (address)(_pc_start + 1));
435     relocInfo::change_reloc_info_for_address(&iter, (address) _pc_start, reloc_type, relocInfo::none);
436   }
437 }
438 
439 
440 void DeoptimizeStub::emit_code(LIR_Assembler* ce) {
441   __ bind(_entry);
442   ce->store_parameter(_trap_request, 0);
443   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::deoptimize_id)));
444   ce->add_call_info_here(_info);
445   DEBUG_ONLY(__ should_not_reach_here());
446 }
447 
448 
449 void ImplicitNullCheckStub::emit_code(LIR_Assembler* ce) {
450   address a;
451   if (_info->deoptimize_on_exception()) {
452     // Deoptimize, do not throw the exception, because it is probably wrong to do it here.
453     a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id);
454   } else {
455     a = Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id);
456   }
457 
458   ce->compilation()->implicit_exception_table()->append(_offset, __ offset());
459   __ bind(_entry);
460   __ call(RuntimeAddress(a));
461   ce->add_call_info_here(_info);
462   ce->verify_oop_map(_info);
463   debug_only(__ should_not_reach_here());
464 }
465 
466 
467 void SimpleExceptionStub::emit_code(LIR_Assembler* ce) {
468   assert(__ rsp_offset() == 0, "frame size should be fixed");
469 
470   __ bind(_entry);
471   // pass the object on stack because all registers must be preserved
472   if (_obj->is_cpu_register()) {
473     ce->store_parameter(_obj->as_register(), 0);
474   }
475   __ call(RuntimeAddress(Runtime1::entry_for(_stub)));
476   ce->add_call_info_here(_info);
477   debug_only(__ should_not_reach_here());
478 }
479 
480 
481 void ArrayCopyStub::emit_code(LIR_Assembler* ce) {
482   //---------------slow case: call to native-----------------
483   __ bind(_entry);
484   // Figure out where the args should go
485   // This should really convert the IntrinsicID to the Method* and signature
486   // but I don't know how to do that.
487   //
488   VMRegPair args[5];
489   BasicType signature[5] = { T_OBJECT, T_INT, T_OBJECT, T_INT, T_INT};
490   SharedRuntime::java_calling_convention(signature, args, 5);
491 
492   // push parameters
493   // (src, src_pos, dest, destPos, length)
494   Register r[5];
495   r[0] = src()->as_register();
496   r[1] = src_pos()->as_register();
497   r[2] = dst()->as_register();
498   r[3] = dst_pos()->as_register();
499   r[4] = length()->as_register();
500 
501   // next registers will get stored on the stack
502   for (int i = 0; i < 5 ; i++ ) {
503     VMReg r_1 = args[i].first();
504     if (r_1->is_stack()) {
505       int st_off = r_1->reg2stack() * wordSize;
506       __ movptr (Address(rsp, st_off), r[i]);
507     } else {
508       assert(r[i] == args[i].first()->as_Register(), "Wrong register for arg ");
509     }
510   }
511 
512   ce->align_call(lir_static_call);
513 
514   ce->emit_static_call_stub();
515   if (ce->compilation()->bailed_out()) {
516     return; // CodeCache is full
517   }
518   AddressLiteral resolve(SharedRuntime::get_resolve_static_call_stub(),
519                          relocInfo::static_call_type);
520   __ call(resolve);
521   ce->add_call_info_here(info());
522 
523 #ifndef PRODUCT
524   if (PrintC1Statistics) {
525     __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_slowcase_cnt), rscratch1);
526   }
527 #endif
528 
529   __ jmp(_continuation);
530 }
531 
532 #undef __