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_MacroAssembler.hpp"
 27 #include "c1/c1_Runtime1.hpp"
 28 #include "compiler/compilerDefinitions.inline.hpp"
 29 #include "gc/shared/barrierSet.hpp"
 30 #include "gc/shared/barrierSetAssembler.hpp"
 31 #include "gc/shared/collectedHeap.hpp"
 32 #include "gc/shared/tlab_globals.hpp"
 33 #include "interpreter/interpreter.hpp"
 34 #include "oops/arrayOop.hpp"
 35 #include "oops/markWord.hpp"
 36 #include "runtime/basicLock.hpp"
 37 #include "runtime/frame.inline.hpp"
 38 #include "runtime/os.hpp"
 39 #include "runtime/sharedRuntime.hpp"
 40 #include "runtime/stubRoutines.hpp"
 41 #include "utilities/checkedCast.hpp"
 42 
 43 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
 44   const int aligned_mask = BytesPerWord -1;
 45   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 46   assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
 47   assert_different_registers(hdr, obj, disp_hdr, tmp);
 48   int null_check_offset = -1;
 49 
 50   verify_oop(obj);
 51 
 52   // save object being locked into the BasicObjectLock
 53   movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
 54 
 55   null_check_offset = offset();
 56 
 57   if (DiagnoseSyncOnValueBasedClasses != 0) {
 58     load_klass(hdr, obj, rscratch1);
 59     movl(hdr, Address(hdr, Klass::access_flags_offset()));
 60     testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
 61     jcc(Assembler::notZero, slow_case);
 62   }
 63 
 64   // Load object header
 65   movptr(hdr, Address(obj, hdr_offset));
 66 
 67   if (LockingMode == LM_LIGHTWEIGHT) {
 68 #ifdef _LP64
 69     const Register thread = r15_thread;
 70 #else
 71     const Register thread = disp_hdr;
 72     get_thread(thread);
 73 #endif
 74     lightweight_lock(obj, hdr, thread, tmp, slow_case);
 75   } else  if (LockingMode == LM_LEGACY) {
 76     Label done;
 77     // and mark it as unlocked
 78     orptr(hdr, markWord::unlocked_value);
 79     if (EnableValhalla) {
 80       // Mask inline_type bit such that we go to the slow path if object is an inline type
 81       andptr(hdr, ~((int) markWord::inline_type_bit_in_place));
 82     }
 83     // save unlocked object header into the displaced header location on the stack
 84     movptr(Address(disp_hdr, 0), hdr);
 85     // test if object header is still the same (i.e. unlocked), and if so, store the
 86     // displaced header address in the object header - if it is not the same, get the
 87     // object header instead
 88     MacroAssembler::lock(); // must be immediately before cmpxchg!
 89     cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
 90     // if the object header was the same, we're done
 91     jcc(Assembler::equal, done);
 92     // if the object header was not the same, it is now in the hdr register
 93     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
 94     //
 95     // 1) (hdr & aligned_mask) == 0
 96     // 2) rsp <= hdr
 97     // 3) hdr <= rsp + page_size
 98     //
 99     // these 3 tests can be done by evaluating the following expression:
100     //
101     // (hdr - rsp) & (aligned_mask - page_size)
102     //
103     // assuming both the stack pointer and page_size have their least
104     // significant 2 bits cleared and page_size is a power of 2
105     subptr(hdr, rsp);
106     andptr(hdr, aligned_mask - (int)os::vm_page_size());
107     // for recursive locking, the result is zero => save it in the displaced header
108     // location (null in the displaced hdr location indicates recursive locking)
109     movptr(Address(disp_hdr, 0), hdr);
110     // otherwise we don't care about the result and handle locking via runtime call
111     jcc(Assembler::notZero, slow_case);
112     // done
113     bind(done);
114   }
115 
116   inc_held_monitor_count();
117 
118   return null_check_offset;
119 }
120 
121 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
122   const int aligned_mask = BytesPerWord -1;
123   const int hdr_offset = oopDesc::mark_offset_in_bytes();
124   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
125   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
126   Label done;
127 
128   if (LockingMode != LM_LIGHTWEIGHT) {
129     // load displaced header
130     movptr(hdr, Address(disp_hdr, 0));
131     // if the loaded hdr is null we had recursive locking
132     testptr(hdr, hdr);
133     // if we had recursive locking, we are done
134     jcc(Assembler::zero, done);
135   }
136 
137   // load object
138   movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
139   verify_oop(obj);
140 
141   if (LockingMode == LM_LIGHTWEIGHT) {
142     movptr(disp_hdr, Address(obj, hdr_offset));
143     andptr(disp_hdr, ~(int32_t)markWord::lock_mask_in_place);
144     lightweight_unlock(obj, disp_hdr, hdr, slow_case);
145   } else if (LockingMode == LM_LEGACY) {
146     // test if object header is pointing to the displaced header, and if so, restore
147     // the displaced header in the object - if the object header is not pointing to
148     // the displaced header, get the object header instead
149     MacroAssembler::lock(); // must be immediately before cmpxchg!
150     cmpxchgptr(hdr, Address(obj, hdr_offset));
151     // if the object header was not pointing to the displaced header,
152     // we do unlocking via runtime call
153     jcc(Assembler::notEqual, slow_case);
154     // done
155   }
156   bind(done);
157   dec_held_monitor_count();
158 }
159 
160 
161 // Defines obj, preserves var_size_in_bytes
162 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
163   if (UseTLAB) {
164     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
165   } else {
166     jmp(slow_case);
167   }
168 }
169 
170 
171 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
172   assert_different_registers(obj, klass, len);
173   if (EnableValhalla) {
174     // Need to copy markWord::prototype header for klass
175     assert_different_registers(obj, klass, len, t1, t2);
176     movptr(t1, Address(klass, Klass::prototype_header_offset()));
177     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
178   } else {
179     // This assumes that all prototype bits fit in an int32_t
180     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
181   }
182 #ifdef _LP64
183   if (UseCompressedClassPointers) { // Take care not to kill klass
184     movptr(t1, klass);
185     encode_klass_not_null(t1, rscratch1);
186     movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
187   } else
188 #endif
189   {
190     movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
191   }
192 
193   if (len->is_valid()) {
194     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
195   }
196 #ifdef _LP64
197   else if (UseCompressedClassPointers) {
198     xorptr(t1, t1);
199     store_klass_gap(obj, t1);
200   }
201 #endif
202 }
203 
204 
205 // preserves obj, destroys len_in_bytes
206 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
207   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
208   Label done;
209 
210   // len_in_bytes is positive and ptr sized
211   subptr(len_in_bytes, hdr_size_in_bytes);
212   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
213   bind(done);
214 }
215 
216 
217 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
218   assert(obj == rax, "obj must be in rax, for cmpxchg");
219   assert_different_registers(obj, t1, t2); // XXX really?
220   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
221 
222   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
223 
224   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
225 }
226 
227 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) {
228   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
229          "con_size_in_bytes is not multiple of alignment");
230   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
231 
232   initialize_header(obj, klass, noreg, t1, t2);
233 
234   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
235     // clear rest of allocated space
236     const Register t1_zero = t1;
237     const Register index = t2;
238     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
239     if (var_size_in_bytes != noreg) {
240       mov(index, var_size_in_bytes);
241       initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
242     } else if (con_size_in_bytes <= threshold) {
243       // use explicit null stores
244       // code size = 2 + 3*n bytes (n = number of fields to clear)
245       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
246       for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
247         movptr(Address(obj, i), t1_zero);
248     } else if (con_size_in_bytes > hdr_size_in_bytes) {
249       // use loop to null out the fields
250       // code size = 16 bytes for even n (n = number of fields to clear)
251       // initialize last object field first if odd number of fields
252       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
253       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
254       // initialize last object field if constant size is odd
255       if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
256         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
257       // initialize remaining object fields: rdx is a multiple of 2
258       { Label loop;
259         bind(loop);
260         movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
261                t1_zero);
262         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
263                t1_zero);)
264         decrement(index);
265         jcc(Assembler::notZero, loop);
266       }
267     }
268   }
269 
270   if (CURRENT_ENV->dtrace_alloc_probes()) {
271     assert(obj == rax, "must be");
272     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
273   }
274 
275   verify_oop(obj);
276 }
277 
278 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
279   assert(obj == rax, "obj must be in rax, for cmpxchg");
280   assert_different_registers(obj, len, t1, t2, klass);
281 
282   // determine alignment mask
283   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
284 
285   // check for negative or excessive length
286   cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
287   jcc(Assembler::above, slow_case);
288 
289   const Register arr_size = t2; // okay to be the same
290   // align object end
291   movptr(arr_size, header_size * BytesPerWord + MinObjAlignmentInBytesMask);
292   lea(arr_size, Address(arr_size, len, f));
293   andptr(arr_size, ~MinObjAlignmentInBytesMask);
294 
295   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
296 
297   initialize_header(obj, klass, len, t1, t2);
298 
299   // clear rest of allocated space
300   const Register len_zero = len;
301   initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
302 
303   if (CURRENT_ENV->dtrace_alloc_probes()) {
304     assert(obj == rax, "must be");
305     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
306   }
307 
308   verify_oop(obj);
309 }
310 
311 
312 
313 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
314   verify_oop(receiver);
315   // explicit null check not needed since load from [klass_offset] causes a trap
316   // check against inline cache
317   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
318   int start_offset = offset();
319 
320   if (UseCompressedClassPointers) {
321     load_klass(rscratch1, receiver, rscratch2);
322     cmpptr(rscratch1, iCache);
323   } else {
324     cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
325   }
326   // if icache check fails, then jump to runtime routine
327   // Note: RECEIVER must still contain the receiver!
328   jump_cc(Assembler::notEqual,
329           RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
330   const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
331   assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
332 }
333 
334 void C1_MacroAssembler::build_frame_helper(int frame_size_in_bytes, int sp_offset_for_orig_pc, int sp_inc, bool reset_orig_pc, bool needs_stack_repair) {
335   push(rbp);
336   if (PreserveFramePointer) {
337     mov(rbp, rsp);
338   }
339 #if !defined(_LP64) && defined(COMPILER2)
340   if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
341       // c2 leaves fpu stack dirty. Clean it on entry
342       empty_FPU_stack();
343     }
344 #endif // !_LP64 && COMPILER2
345   decrement(rsp, frame_size_in_bytes);
346 
347   if (needs_stack_repair) {
348     // Save stack increment (also account for fixed framesize and rbp)
349     assert((sp_inc & (StackAlignmentInBytes-1)) == 0, "stack increment not aligned");
350     int real_frame_size = sp_inc + frame_size_in_bytes + wordSize;
351     movptr(Address(rsp, frame_size_in_bytes - wordSize), real_frame_size);
352   }
353   if (reset_orig_pc) {
354     // Zero orig_pc to detect deoptimization during buffering in the entry points
355     movptr(Address(rsp, sp_offset_for_orig_pc), 0);
356   }
357 }
358 
359 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes, int sp_offset_for_orig_pc, bool needs_stack_repair, bool has_scalarized_args, Label* verified_inline_entry_label) {
360   // Make sure there is enough stack space for this method's activation.
361   // Note that we do this before doing an enter(). This matches the
362   // ordering of C2's stack overflow check / rsp decrement and allows
363   // the SharedRuntime stack overflow handling to be consistent
364   // between the two compilers.
365   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
366   generate_stack_overflow_check(bang_size_in_bytes);
367 
368   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, has_scalarized_args, needs_stack_repair);
369 
370   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
371   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
372   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
373 
374   if (verified_inline_entry_label != nullptr) {
375     // Jump here from the scalarized entry points that already created the frame.
376     bind(*verified_inline_entry_label);
377   }
378 }
379 
380 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
381   if (breakAtEntry || VerifyFPU) {
382     // Verified Entry first instruction should be 5 bytes long for correct
383     // patching by patch_verified_entry().
384     //
385     // Breakpoint and VerifyFPU have one byte first instruction.
386     // Also first instruction will be one byte "push(rbp)" if stack banging
387     // code is not generated (see build_frame() above).
388     // For all these cases generate long instruction first.
389     fat_nop();
390   }
391   if (breakAtEntry) int3();
392   // build frame
393   IA32_ONLY( verify_FPU(0, "method_entry"); )
394 }
395 
396 int C1_MacroAssembler::scalarized_entry(const CompiledEntrySignature* ces, int frame_size_in_bytes, int bang_size_in_bytes, int sp_offset_for_orig_pc, Label& verified_inline_entry_label, bool is_inline_ro_entry) {
397   assert(InlineTypePassFieldsAsArgs, "sanity");
398   // Make sure there is enough stack space for this method's activation.
399   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
400   generate_stack_overflow_check(bang_size_in_bytes);
401 
402   GrowableArray<SigEntry>* sig    = ces->sig();
403   GrowableArray<SigEntry>* sig_cc = is_inline_ro_entry ? ces->sig_cc_ro() : ces->sig_cc();
404   VMRegPair* regs      = ces->regs();
405   VMRegPair* regs_cc   = is_inline_ro_entry ? ces->regs_cc_ro() : ces->regs_cc();
406   int args_on_stack    = ces->args_on_stack();
407   int args_on_stack_cc = is_inline_ro_entry ? ces->args_on_stack_cc_ro() : ces->args_on_stack_cc();
408 
409   assert(sig->length() <= sig_cc->length(), "Zero-sized inline class not allowed!");
410   BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sig_cc->length());
411   int args_passed = sig->length();
412   int args_passed_cc = SigEntry::fill_sig_bt(sig_cc, sig_bt);
413 
414   // Create a temp frame so we can call into the runtime. It must be properly set up to accommodate GC.
415   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, true, ces->c1_needs_stack_repair());
416 
417   // The runtime call might safepoint, make sure nmethod entry barrier is executed
418   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
419   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
420   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
421 
422   // FIXME -- call runtime only if we cannot in-line allocate all the incoming inline type args.
423   movptr(rbx, (intptr_t)(ces->method()));
424   if (is_inline_ro_entry) {
425     call(RuntimeAddress(Runtime1::entry_for(Runtime1::buffer_inline_args_no_receiver_id)));
426   } else {
427     call(RuntimeAddress(Runtime1::entry_for(Runtime1::buffer_inline_args_id)));
428   }
429   int rt_call_offset = offset();
430 
431   // Remove the temp frame
432   addptr(rsp, frame_size_in_bytes);
433   pop(rbp);
434 
435   // Check if we need to extend the stack for packing
436   int sp_inc = 0;
437   if (args_on_stack > args_on_stack_cc) {
438     sp_inc = extend_stack_for_inline_args(args_on_stack);
439   }
440 
441   shuffle_inline_args(true, is_inline_ro_entry, sig_cc,
442                       args_passed_cc, args_on_stack_cc, regs_cc, // from
443                       args_passed, args_on_stack, regs,          // to
444                       sp_inc, rax);
445 
446   // Create the real frame. Below jump will then skip over the stack banging and frame
447   // setup code in the verified_inline_entry (which has a different real_frame_size).
448   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, sp_inc, false, ces->c1_needs_stack_repair());
449 
450   jmp(verified_inline_entry_label);
451   return rt_call_offset;
452 }
453 
454 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
455   // rbp, + 0: link
456   //     + 1: return address
457   //     + 2: argument with offset 0
458   //     + 3: argument with offset 1
459   //     + 4: ...
460 
461   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
462 }
463 
464 #ifndef PRODUCT
465 
466 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
467   if (!VerifyOops) return;
468   verify_oop_addr(Address(rsp, stack_offset));
469 }
470 
471 void C1_MacroAssembler::verify_not_null_oop(Register r) {
472   if (!VerifyOops) return;
473   Label not_null;
474   testptr(r, r);
475   jcc(Assembler::notZero, not_null);
476   stop("non-null oop required");
477   bind(not_null);
478   verify_oop(r);
479 }
480 
481 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
482 #ifdef ASSERT
483   if (inv_rax) movptr(rax, 0xDEAD);
484   if (inv_rbx) movptr(rbx, 0xDEAD);
485   if (inv_rcx) movptr(rcx, 0xDEAD);
486   if (inv_rdx) movptr(rdx, 0xDEAD);
487   if (inv_rsi) movptr(rsi, 0xDEAD);
488   if (inv_rdi) movptr(rdi, 0xDEAD);
489 #endif
490 }
491 
492 #endif // ifndef PRODUCT