1 /*
  2  * Copyright (c) 1999, 2020, 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() != NULL,
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 RangeCheckStub::RangeCheckStub(CodeEmitInfo* info, LIR_Opr index, LIR_Opr array)
120   : _index(index), _array(array), _throw_index_out_of_bounds_exception(false) {
121   assert(info != NULL, "must have info");
122   _info = new CodeEmitInfo(info);
123 }
124 
125 RangeCheckStub::RangeCheckStub(CodeEmitInfo* info, LIR_Opr index)
126   : _index(index), _array(), _throw_index_out_of_bounds_exception(true) {
127   assert(info != NULL, "must have info");
128   _info = new CodeEmitInfo(info);
129 }
130 
131 void RangeCheckStub::emit_code(LIR_Assembler* ce) {
132   __ bind(_entry);
133   if (_info->deoptimize_on_exception()) {
134     address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id);
135     __ call(RuntimeAddress(a));
136     ce->add_call_info_here(_info);
137     ce->verify_oop_map(_info);
138     debug_only(__ should_not_reach_here());
139     return;
140   }
141 
142   // pass the array index on stack because all registers must be preserved
143   if (_index->is_cpu_register()) {
144     ce->store_parameter(_index->as_register(), 0);
145   } else {
146     ce->store_parameter(_index->as_jint(), 0);
147   }
148   Runtime1::StubID stub_id;
149   if (_throw_index_out_of_bounds_exception) {
150     stub_id = Runtime1::throw_index_exception_id;
151   } else {
152     stub_id = Runtime1::throw_range_check_failed_id;
153     ce->store_parameter(_array->as_pointer_register(), 1);
154   }
155   __ call(RuntimeAddress(Runtime1::entry_for(stub_id)));
156   ce->add_call_info_here(_info);
157   ce->verify_oop_map(_info);
158   debug_only(__ should_not_reach_here());
159 }
160 
161 PredicateFailedStub::PredicateFailedStub(CodeEmitInfo* info) {
162   _info = new CodeEmitInfo(info);
163 }
164 
165 void PredicateFailedStub::emit_code(LIR_Assembler* ce) {
166   __ bind(_entry);
167   address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id);
168   __ call(RuntimeAddress(a));
169   ce->add_call_info_here(_info);
170   ce->verify_oop_map(_info);
171   debug_only(__ should_not_reach_here());
172 }
173 
174 void DivByZeroStub::emit_code(LIR_Assembler* ce) {
175   if (_offset != -1) {
176     ce->compilation()->implicit_exception_table()->append(_offset, __ offset());
177   }
178   __ bind(_entry);
179   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::throw_div0_exception_id)));
180   ce->add_call_info_here(_info);
181   debug_only(__ should_not_reach_here());
182 }
183 
184 
185 // Implementation of NewInstanceStub
186 
187 NewInstanceStub::NewInstanceStub(LIR_Opr klass_reg, LIR_Opr result, ciInstanceKlass* klass, CodeEmitInfo* info, Runtime1::StubID stub_id) {
188   _result = result;
189   _klass = klass;
190   _klass_reg = klass_reg;
191   _info = new CodeEmitInfo(info);
192   assert(stub_id == Runtime1::new_instance_id                 ||
193          stub_id == Runtime1::fast_new_instance_id            ||
194          stub_id == Runtime1::fast_new_instance_init_check_id,
195          "need new_instance id");
196   _stub_id   = stub_id;
197 }
198 
199 
200 void NewInstanceStub::emit_code(LIR_Assembler* ce) {
201   assert(__ rsp_offset() == 0, "frame size should be fixed");
202   __ bind(_entry);
203   __ movptr(rdx, _klass_reg->as_register());
204   __ call(RuntimeAddress(Runtime1::entry_for(_stub_id)));
205   ce->add_call_info_here(_info);
206   ce->verify_oop_map(_info);
207   assert(_result->as_register() == rax, "result must in rax,");
208   __ jmp(_continuation);
209 }
210 
211 
212 // Implementation of NewTypeArrayStub
213 
214 NewTypeArrayStub::NewTypeArrayStub(LIR_Opr klass_reg, LIR_Opr length, LIR_Opr result, CodeEmitInfo* info) {
215   _klass_reg = klass_reg;
216   _length = length;
217   _result = result;
218   _info = new CodeEmitInfo(info);
219 }
220 
221 
222 void NewTypeArrayStub::emit_code(LIR_Assembler* ce) {
223   assert(__ rsp_offset() == 0, "frame size should be fixed");
224   __ bind(_entry);
225   assert(_length->as_register() == rbx, "length must in rbx,");
226   assert(_klass_reg->as_register() == rdx, "klass_reg must in rdx");
227   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::new_type_array_id)));
228   ce->add_call_info_here(_info);
229   ce->verify_oop_map(_info);
230   assert(_result->as_register() == rax, "result must in rax,");
231   __ jmp(_continuation);
232 }
233 
234 
235 // Implementation of NewObjectArrayStub
236 
237 NewObjectArrayStub::NewObjectArrayStub(LIR_Opr klass_reg, LIR_Opr length, LIR_Opr result, CodeEmitInfo* info) {
238   _klass_reg = klass_reg;
239   _result = result;
240   _length = length;
241   _info = new CodeEmitInfo(info);
242 }
243 
244 
245 void NewObjectArrayStub::emit_code(LIR_Assembler* ce) {
246   assert(__ rsp_offset() == 0, "frame size should be fixed");
247   __ bind(_entry);
248   assert(_length->as_register() == rbx, "length must in rbx,");
249   assert(_klass_reg->as_register() == rdx, "klass_reg must in rdx");
250   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::new_object_array_id)));
251   ce->add_call_info_here(_info);
252   ce->verify_oop_map(_info);
253   assert(_result->as_register() == rax, "result must in rax,");
254   __ jmp(_continuation);
255 }
256 
257 
258 // Implementation of MonitorAccessStubs
259 
260 MonitorEnterStub::MonitorEnterStub(LIR_Opr obj_reg, CodeEmitInfo* info)
261 : MonitorAccessStub(obj_reg)
262 {
263   _info = new CodeEmitInfo(info);
264 }
265 
266 
267 void MonitorEnterStub::emit_code(LIR_Assembler* ce) {
268   assert(__ rsp_offset() == 0, "frame size should be fixed");
269   __ bind(_entry);
270   ce->store_parameter(_obj_reg->as_register(),  0);
271   Runtime1::StubID enter_id;
272   if (ce->compilation()->has_fpu_code()) {
273     enter_id = Runtime1::monitorenter_id;
274   } else {
275     enter_id = Runtime1::monitorenter_nofpu_id;
276   }
277   __ call(RuntimeAddress(Runtime1::entry_for(enter_id)));
278   ce->add_call_info_here(_info);
279   ce->verify_oop_map(_info);
280   __ jmp(_continuation);
281 }
282 
283 
284 void MonitorExitStub::emit_code(LIR_Assembler* ce) {
285   __ bind(_entry);
286   ce->store_parameter(_obj_reg->as_register(), 0);
287   // note: non-blocking leaf routine => no call info needed
288   Runtime1::StubID exit_id;
289   if (ce->compilation()->has_fpu_code()) {
290     exit_id = Runtime1::monitorexit_id;
291   } else {
292     exit_id = Runtime1::monitorexit_nofpu_id;
293   }
294   __ call(RuntimeAddress(Runtime1::entry_for(exit_id)));
295   __ jmp(_continuation);
296 }
297 
298 void LoadKlassStub::emit_code(LIR_Assembler* ce) {
299   __ bind(_entry);
300 #ifdef _LP64
301   Register res = _result->as_register();
302   ce->store_parameter(_obj->as_register(), 0);
303   if (res != rax) {
304     // This preserves rax and allows it to be used as return-register,
305     // without messing with the stack.
306     __ xchgptr(rax, res);
307   }
308   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::load_klass_id)));
309   if (res != rax) {
310     // Swap back rax, and move result to correct register.
311     __ xchgptr(rax, res);
312   }
313   __ jmp(_continuation);
314 #else
315   __ should_not_reach_here();
316 #endif
317 }
318 
319 // Implementation of patching:
320 // - Copy the code at given offset to an inlined buffer (first the bytes, then the number of bytes)
321 // - Replace original code with a call to the stub
322 // At Runtime:
323 // - call to stub, jump to runtime
324 // - in runtime: preserve all registers (rspecially objects, i.e., source and destination object)
325 // - in runtime: after initializing class, restore original code, reexecute instruction
326 
327 int PatchingStub::_patch_info_offset = -NativeGeneralJump::instruction_size;
328 
329 void PatchingStub::align_patch_site(MacroAssembler* masm) {
330   // We're patching a 5-7 byte instruction on intel and we need to
331   // make sure that we don't see a piece of the instruction.  It
332   // appears mostly impossible on Intel to simply invalidate other
333   // processors caches and since they may do aggressive prefetch it's
334   // very hard to make a guess about what code might be in the icache.
335   // Force the instruction to be double word aligned so that it
336   // doesn't span a cache line.
337   masm->align(align_up((int)NativeGeneralJump::instruction_size, wordSize));
338 }
339 
340 void PatchingStub::emit_code(LIR_Assembler* ce) {
341   assert(NativeCall::instruction_size <= _bytes_to_copy && _bytes_to_copy <= 0xFF, "not enough room for call");
342 
343   Label call_patch;
344 
345   // static field accesses have special semantics while the class
346   // initializer is being run so we emit a test which can be used to
347   // check that this code is being executed by the initializing
348   // thread.
349   address being_initialized_entry = __ pc();
350   if (CommentedAssembly) {
351     __ block_comment(" patch template");
352   }
353   if (_id == load_klass_id) {
354     // produce a copy of the load klass instruction for use by the being initialized case
355 #ifdef ASSERT
356     address start = __ pc();
357 #endif
358     Metadata* o = NULL;
359     __ mov_metadata(_obj, o);
360 #ifdef ASSERT
361     for (int i = 0; i < _bytes_to_copy; i++) {
362       address ptr = (address)(_pc_start + i);
363       int a_byte = (*ptr) & 0xFF;
364       assert(a_byte == *start++, "should be the same code");
365     }
366 #endif
367   } else if (_id == load_mirror_id) {
368     // produce a copy of the load mirror instruction for use by the being
369     // initialized case
370 #ifdef ASSERT
371     address start = __ pc();
372 #endif
373     jobject o = NULL;
374     __ movoop(_obj, o);
375 #ifdef ASSERT
376     for (int i = 0; i < _bytes_to_copy; i++) {
377       address ptr = (address)(_pc_start + i);
378       int a_byte = (*ptr) & 0xFF;
379       assert(a_byte == *start++, "should be the same code");
380     }
381 #endif
382   } else {
383     // make a copy the code which is going to be patched.
384     for (int i = 0; i < _bytes_to_copy; i++) {
385       address ptr = (address)(_pc_start + i);
386       int a_byte = (*ptr) & 0xFF;
387       __ emit_int8(a_byte);
388       *ptr = 0x90; // make the site look like a nop
389     }
390   }
391 
392   address end_of_patch = __ pc();
393   int bytes_to_skip = 0;
394   if (_id == load_mirror_id) {
395     int offset = __ offset();
396     if (CommentedAssembly) {
397       __ block_comment(" being_initialized check");
398     }
399     assert(_obj != noreg, "must be a valid register");
400     Register tmp = rax;
401     Register tmp2 = rbx;
402     __ push(tmp);
403     __ push(tmp2);
404     // Load without verification to keep code size small. We need it because
405     // begin_initialized_entry_offset has to fit in a byte. Also, we know it's not null.
406     __ movptr(tmp2, Address(_obj, java_lang_Class::klass_offset()));
407     __ get_thread(tmp);
408     __ cmpptr(tmp, Address(tmp2, InstanceKlass::init_thread_offset()));
409     __ pop(tmp2);
410     __ pop(tmp);
411     __ jcc(Assembler::notEqual, call_patch);
412 
413     // access_field patches may execute the patched code before it's
414     // copied back into place so we need to jump back into the main
415     // code of the nmethod to continue execution.
416     __ jmp(_patch_site_continuation);
417 
418     // make sure this extra code gets skipped
419     bytes_to_skip += __ offset() - offset;
420   }
421   if (CommentedAssembly) {
422     __ block_comment("patch data encoded as movl");
423   }
424   // Now emit the patch record telling the runtime how to find the
425   // pieces of the patch.  We only need 3 bytes but for readability of
426   // the disassembly we make the data look like a movl reg, imm32,
427   // which requires 5 bytes
428   int sizeof_patch_record = 5;
429   bytes_to_skip += sizeof_patch_record;
430 
431   // emit the offsets needed to find the code to patch
432   int being_initialized_entry_offset = __ pc() - being_initialized_entry + sizeof_patch_record;
433 
434   __ emit_int8((unsigned char)0xB8);
435   __ emit_int8(0);
436   __ emit_int8(being_initialized_entry_offset);
437   __ emit_int8(bytes_to_skip);
438   __ emit_int8(_bytes_to_copy);
439   address patch_info_pc = __ pc();
440   assert(patch_info_pc - end_of_patch == bytes_to_skip, "incorrect patch info");
441 
442   address entry = __ pc();
443   NativeGeneralJump::insert_unconditional((address)_pc_start, entry);
444   address target = NULL;
445   relocInfo::relocType reloc_type = relocInfo::none;
446   switch (_id) {
447     case access_field_id:  target = Runtime1::entry_for(Runtime1::access_field_patching_id); break;
448     case load_klass_id:    target = Runtime1::entry_for(Runtime1::load_klass_patching_id); reloc_type = relocInfo::metadata_type; break;
449     case load_mirror_id:   target = Runtime1::entry_for(Runtime1::load_mirror_patching_id); reloc_type = relocInfo::oop_type; break;
450     case load_appendix_id:      target = Runtime1::entry_for(Runtime1::load_appendix_patching_id); reloc_type = relocInfo::oop_type; break;
451     default: ShouldNotReachHere();
452   }
453   __ bind(call_patch);
454 
455   if (CommentedAssembly) {
456     __ block_comment("patch entry point");
457   }
458   __ call(RuntimeAddress(target));
459   assert(_patch_info_offset == (patch_info_pc - __ pc()), "must not change");
460   ce->add_call_info_here(_info);
461   int jmp_off = __ offset();
462   __ jmp(_patch_site_entry);
463   // Add enough nops so deoptimization can overwrite the jmp above with a call
464   // and not destroy the world. We cannot use fat nops here, since the concurrent
465   // code rewrite may transiently create the illegal instruction sequence.
466   for (int j = __ offset() ; j < jmp_off + 5 ; j++ ) {
467     __ nop();
468   }
469   if (_id == load_klass_id || _id == load_mirror_id || _id == load_appendix_id) {
470     CodeSection* cs = __ code_section();
471     RelocIterator iter(cs, (address)_pc_start, (address)(_pc_start + 1));
472     relocInfo::change_reloc_info_for_address(&iter, (address) _pc_start, reloc_type, relocInfo::none);
473   }
474 }
475 
476 
477 void DeoptimizeStub::emit_code(LIR_Assembler* ce) {
478   __ bind(_entry);
479   ce->store_parameter(_trap_request, 0);
480   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::deoptimize_id)));
481   ce->add_call_info_here(_info);
482   DEBUG_ONLY(__ should_not_reach_here());
483 }
484 
485 
486 void ImplicitNullCheckStub::emit_code(LIR_Assembler* ce) {
487   address a;
488   if (_info->deoptimize_on_exception()) {
489     // Deoptimize, do not throw the exception, because it is probably wrong to do it here.
490     a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id);
491   } else {
492     a = Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id);
493   }
494 
495   ce->compilation()->implicit_exception_table()->append(_offset, __ offset());
496   __ bind(_entry);
497   __ call(RuntimeAddress(a));
498   ce->add_call_info_here(_info);
499   ce->verify_oop_map(_info);
500   debug_only(__ should_not_reach_here());
501 }
502 
503 
504 void SimpleExceptionStub::emit_code(LIR_Assembler* ce) {
505   assert(__ rsp_offset() == 0, "frame size should be fixed");
506 
507   __ bind(_entry);
508   // pass the object on stack because all registers must be preserved
509   if (_obj->is_cpu_register()) {
510     ce->store_parameter(_obj->as_register(), 0);
511   }
512   __ call(RuntimeAddress(Runtime1::entry_for(_stub)));
513   ce->add_call_info_here(_info);
514   debug_only(__ should_not_reach_here());
515 }
516 
517 
518 void ArrayCopyStub::emit_code(LIR_Assembler* ce) {
519   //---------------slow case: call to native-----------------
520   __ bind(_entry);
521   // Figure out where the args should go
522   // This should really convert the IntrinsicID to the Method* and signature
523   // but I don't know how to do that.
524   //
525   VMRegPair args[5];
526   BasicType signature[5] = { T_OBJECT, T_INT, T_OBJECT, T_INT, T_INT};
527   SharedRuntime::java_calling_convention(signature, args, 5);
528 
529   // push parameters
530   // (src, src_pos, dest, destPos, length)
531   Register r[5];
532   r[0] = src()->as_register();
533   r[1] = src_pos()->as_register();
534   r[2] = dst()->as_register();
535   r[3] = dst_pos()->as_register();
536   r[4] = length()->as_register();
537 
538   // next registers will get stored on the stack
539   for (int i = 0; i < 5 ; i++ ) {
540     VMReg r_1 = args[i].first();
541     if (r_1->is_stack()) {
542       int st_off = r_1->reg2stack() * wordSize;
543       __ movptr (Address(rsp, st_off), r[i]);
544     } else {
545       assert(r[i] == args[i].first()->as_Register(), "Wrong register for arg ");
546     }
547   }
548 
549   ce->align_call(lir_static_call);
550 
551   ce->emit_static_call_stub();
552   if (ce->compilation()->bailed_out()) {
553     return; // CodeCache is full
554   }
555   AddressLiteral resolve(SharedRuntime::get_resolve_static_call_stub(),
556                          relocInfo::static_call_type);
557   __ call(resolve);
558   ce->add_call_info_here(info());
559 
560 #ifndef PRODUCT
561   if (PrintC1Statistics) {
562     __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_slowcase_cnt), rscratch1);
563   }
564 #endif
565 
566   __ jmp(_continuation);
567 }
568 
569 #undef __