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_MacroAssembler.hpp"
 26 #include "c1/c1_Runtime1.hpp"
 27 #include "code/compiledIC.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/globals.hpp"
 39 #include "runtime/os.hpp"
 40 #include "runtime/sharedRuntime.hpp"
 41 #include "runtime/stubRoutines.hpp"
 42 #include "utilities/checkedCast.hpp"
 43 #include "utilities/globalDefinitions.hpp"
 44 
 45 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
 46   const int aligned_mask = BytesPerWord -1;
 47   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 48   assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
 49   assert_different_registers(hdr, obj, disp_hdr, tmp);
 50   int null_check_offset = -1;
 51 
 52   verify_oop(obj);
 53 
 54   // save object being locked into the BasicObjectLock
 55   movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
 56 
 57   null_check_offset = offset();
 58 
 59   if (LockingMode == LM_LIGHTWEIGHT) {
 60     lightweight_lock(disp_hdr, obj, hdr, tmp, slow_case);
 61   } else  if (LockingMode == LM_LEGACY) {
 62     Label done;
 63 
 64     if (DiagnoseSyncOnValueBasedClasses != 0) {
 65       load_klass(hdr, obj, rscratch1);
 66       testb(Address(hdr, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
 67       jcc(Assembler::notZero, slow_case);
 68     }
 69 
 70     // Load object header
 71     movptr(hdr, Address(obj, hdr_offset));
 72     // and mark it as unlocked
 73     orptr(hdr, markWord::unlocked_value);
 74     if (EnableValhalla) {
 75       // Mask inline_type bit such that we go to the slow path if object is an inline type
 76       andptr(hdr, ~((int) markWord::inline_type_bit_in_place));
 77     }
 78     // save unlocked object header into the displaced header location on the stack
 79     movptr(Address(disp_hdr, 0), hdr);
 80     // test if object header is still the same (i.e. unlocked), and if so, store the
 81     // displaced header address in the object header - if it is not the same, get the
 82     // object header instead
 83     MacroAssembler::lock(); // must be immediately before cmpxchg!
 84     cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
 85     // if the object header was the same, we're done
 86     jcc(Assembler::equal, done);
 87     // if the object header was not the same, it is now in the hdr register
 88     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
 89     //
 90     // 1) (hdr & aligned_mask) == 0
 91     // 2) rsp <= hdr
 92     // 3) hdr <= rsp + page_size
 93     //
 94     // these 3 tests can be done by evaluating the following expression:
 95     //
 96     // (hdr - rsp) & (aligned_mask - page_size)
 97     //
 98     // assuming both the stack pointer and page_size have their least
 99     // significant 2 bits cleared and page_size is a power of 2
100     subptr(hdr, rsp);
101     andptr(hdr, aligned_mask - (int)os::vm_page_size());
102     // for recursive locking, the result is zero => save it in the displaced header
103     // location (null in the displaced hdr location indicates recursive locking)
104     movptr(Address(disp_hdr, 0), hdr);
105     // otherwise we don't care about the result and handle locking via runtime call
106     jcc(Assembler::notZero, slow_case);
107     // done
108     bind(done);
109     inc_held_monitor_count();
110   }
111 
112   return null_check_offset;
113 }
114 
115 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
116   const int aligned_mask = BytesPerWord -1;
117   const int hdr_offset = oopDesc::mark_offset_in_bytes();
118   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
119   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
120   Label done;
121 
122   if (LockingMode != LM_LIGHTWEIGHT) {
123     // load displaced header
124     movptr(hdr, Address(disp_hdr, 0));
125     // if the loaded hdr is null we had recursive locking
126     testptr(hdr, hdr);
127     // if we had recursive locking, we are done
128     jcc(Assembler::zero, done);
129   }
130 
131   // load object
132   movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
133   verify_oop(obj);
134 
135   if (LockingMode == LM_LIGHTWEIGHT) {
136     lightweight_unlock(obj, disp_hdr, hdr, slow_case);
137   } else if (LockingMode == LM_LEGACY) {
138     // test if object header is pointing to the displaced header, and if so, restore
139     // the displaced header in the object - if the object header is not pointing to
140     // the displaced header, get the object header instead
141     MacroAssembler::lock(); // must be immediately before cmpxchg!
142     cmpxchgptr(hdr, Address(obj, hdr_offset));
143     // if the object header was not pointing to the displaced header,
144     // we do unlocking via runtime call
145     jcc(Assembler::notEqual, slow_case);
146     // done
147     bind(done);
148     dec_held_monitor_count();
149   }
150 }
151 
152 
153 // Defines obj, preserves var_size_in_bytes
154 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
155   if (UseTLAB) {
156     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
157   } else {
158     jmp(slow_case);
159   }
160 }
161 
162 
163 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
164   assert_different_registers(obj, klass, len, t1, t2);
165   if (UseCompactObjectHeaders || EnableValhalla) {
166     movptr(t1, Address(klass, Klass::prototype_header_offset()));
167     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
168   } else {
169     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
170   }
171   if (UseCompressedClassPointers) { // Take care not to kill klass
172     movptr(t1, klass);
173     encode_klass_not_null(t1, rscratch1);
174     movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
175   } else if (!UseCompactObjectHeaders) {
176     movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
177   }
178 
179   if (len->is_valid()) {
180     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
181     int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
182     if (!is_aligned(base_offset, BytesPerWord)) {
183       assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
184       // Clear gap/first 4 bytes following the length field.
185       xorl(t1, t1);
186       movl(Address(obj, base_offset), t1);
187     }
188   } else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
189     xorptr(t1, t1);
190     store_klass_gap(obj, t1);
191   }
192 }
193 
194 
195 // preserves obj, destroys len_in_bytes
196 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
197   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
198   Label done;
199 
200   // len_in_bytes is positive and ptr sized
201   subptr(len_in_bytes, hdr_size_in_bytes);
202   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
203   bind(done);
204 }
205 
206 
207 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
208   assert(obj == rax, "obj must be in rax, for cmpxchg");
209   assert_different_registers(obj, t1, t2); // XXX really?
210   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
211 
212   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
213 
214   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
215 }
216 
217 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) {
218   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
219          "con_size_in_bytes is not multiple of alignment");
220   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
221   if (UseCompactObjectHeaders) {
222     assert(hdr_size_in_bytes == 8, "check object headers size");
223   }
224   initialize_header(obj, klass, noreg, t1, t2);
225 
226   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
227     // clear rest of allocated space
228     const Register t1_zero = t1;
229     const Register index = t2;
230     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
231     if (var_size_in_bytes != noreg) {
232       mov(index, var_size_in_bytes);
233       initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
234     } else if (con_size_in_bytes <= threshold) {
235       // use explicit null stores
236       // code size = 2 + 3*n bytes (n = number of fields to clear)
237       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
238       for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
239         movptr(Address(obj, i), t1_zero);
240     } else if (con_size_in_bytes > hdr_size_in_bytes) {
241       // use loop to null out the fields
242       // code size = 16 bytes for even n (n = number of fields to clear)
243       // initialize last object field first if odd number of fields
244       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
245       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
246       // initialize last object field if constant size is odd
247       if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
248         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
249       // initialize remaining object fields: rdx is a multiple of 2
250       { Label loop;
251         bind(loop);
252         movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
253                t1_zero);
254         decrement(index);
255         jcc(Assembler::notZero, loop);
256       }
257     }
258   }
259 
260   if (CURRENT_ENV->dtrace_alloc_probes()) {
261     assert(obj == rax, "must be");
262     call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
263   }
264 
265   verify_oop(obj);
266 }
267 
268 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int base_offset_in_bytes, Address::ScaleFactor f, Register klass, Label& slow_case, bool zero_array) {
269   assert(obj == rax, "obj must be in rax, for cmpxchg");
270   assert_different_registers(obj, len, t1, t2, klass);
271 
272   // determine alignment mask
273   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
274 
275   // check for negative or excessive length
276   cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
277   jcc(Assembler::above, slow_case);
278 
279   const Register arr_size = t2; // okay to be the same
280   // align object end
281   movptr(arr_size, base_offset_in_bytes + MinObjAlignmentInBytesMask);
282   lea(arr_size, Address(arr_size, len, f));
283   andptr(arr_size, ~MinObjAlignmentInBytesMask);
284 
285   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
286 
287   initialize_header(obj, klass, len, t1, t2);
288 
289   // clear rest of allocated space
290   if (zero_array) {
291     const Register len_zero = len;
292     // Align-up to word boundary, because we clear the 4 bytes potentially
293     // following the length field in initialize_header().
294     int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
295     initialize_body(obj, arr_size, base_offset, len_zero);
296   }
297 
298   if (CURRENT_ENV->dtrace_alloc_probes()) {
299     assert(obj == rax, "must be");
300     call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
301   }
302 
303   verify_oop(obj);
304 }
305 
306 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) {
307   push(rbp);
308   if (PreserveFramePointer) {
309     mov(rbp, rsp);
310   }
311   decrement(rsp, frame_size_in_bytes);
312 
313   if (needs_stack_repair) {
314     // Save stack increment (also account for fixed framesize and rbp)
315     assert((sp_inc & (StackAlignmentInBytes-1)) == 0, "stack increment not aligned");
316     int real_frame_size = sp_inc + frame_size_in_bytes + wordSize;
317     movptr(Address(rsp, frame_size_in_bytes - wordSize), real_frame_size);
318   }
319   if (reset_orig_pc) {
320     // Zero orig_pc to detect deoptimization during buffering in the entry points
321     movptr(Address(rsp, sp_offset_for_orig_pc), 0);
322   }
323 }
324 
325 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) {
326   // Make sure there is enough stack space for this method's activation.
327   // Note that we do this before doing an enter(). This matches the
328   // ordering of C2's stack overflow check / rsp decrement and allows
329   // the SharedRuntime stack overflow handling to be consistent
330   // between the two compilers.
331   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
332   generate_stack_overflow_check(bang_size_in_bytes);
333 
334   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, has_scalarized_args, needs_stack_repair);
335 
336   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
337   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
338   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
339 
340   if (verified_inline_entry_label != nullptr) {
341     // Jump here from the scalarized entry points that already created the frame.
342     bind(*verified_inline_entry_label);
343   }
344 }
345 
346 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
347   if (breakAtEntry) {
348     // Verified Entry first instruction should be 5 bytes long for correct
349     // patching by patch_verified_entry().
350     //
351     // Breakpoint has one byte first instruction.
352     // Also first instruction will be one byte "push(rbp)" if stack banging
353     // code is not generated (see build_frame() above).
354     // For all these cases generate long instruction first.
355     fat_nop();
356   }
357   if (breakAtEntry) int3();
358   // build frame
359 }
360 
361 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) {
362   assert(InlineTypePassFieldsAsArgs, "sanity");
363   // Make sure there is enough stack space for this method's activation.
364   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
365   generate_stack_overflow_check(bang_size_in_bytes);
366 
367   GrowableArray<SigEntry>* sig    = ces->sig();
368   GrowableArray<SigEntry>* sig_cc = is_inline_ro_entry ? ces->sig_cc_ro() : ces->sig_cc();
369   VMRegPair* regs      = ces->regs();
370   VMRegPair* regs_cc   = is_inline_ro_entry ? ces->regs_cc_ro() : ces->regs_cc();
371   int args_on_stack    = ces->args_on_stack();
372   int args_on_stack_cc = is_inline_ro_entry ? ces->args_on_stack_cc_ro() : ces->args_on_stack_cc();
373 
374   assert(sig->length() <= sig_cc->length(), "Zero-sized inline class not allowed!");
375   BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sig_cc->length());
376   int args_passed = sig->length();
377   int args_passed_cc = SigEntry::fill_sig_bt(sig_cc, sig_bt);
378 
379   // Create a temp frame so we can call into the runtime. It must be properly set up to accommodate GC.
380   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, true, ces->c1_needs_stack_repair());
381 
382   // The runtime call might safepoint, make sure nmethod entry barrier is executed
383   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
384   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
385   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
386 
387   // FIXME -- call runtime only if we cannot in-line allocate all the incoming inline type args.
388   movptr(rbx, (intptr_t)(ces->method()));
389   if (is_inline_ro_entry) {
390     call(RuntimeAddress(Runtime1::entry_for(C1StubId::buffer_inline_args_no_receiver_id)));
391   } else {
392     call(RuntimeAddress(Runtime1::entry_for(C1StubId::buffer_inline_args_id)));
393   }
394   int rt_call_offset = offset();
395 
396   // Remove the temp frame
397   addptr(rsp, frame_size_in_bytes);
398   pop(rbp);
399 
400   // Check if we need to extend the stack for packing
401   int sp_inc = 0;
402   if (args_on_stack > args_on_stack_cc) {
403     sp_inc = extend_stack_for_inline_args(args_on_stack);
404   }
405 
406   shuffle_inline_args(true, is_inline_ro_entry, sig_cc,
407                       args_passed_cc, args_on_stack_cc, regs_cc, // from
408                       args_passed, args_on_stack, regs,          // to
409                       sp_inc, rax);
410 
411   // Create the real frame. Below jump will then skip over the stack banging and frame
412   // setup code in the verified_inline_entry (which has a different real_frame_size).
413   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, sp_inc, false, ces->c1_needs_stack_repair());
414 
415   jmp(verified_inline_entry_label);
416   return rt_call_offset;
417 }
418 
419 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
420   // rbp, + 0: link
421   //     + 1: return address
422   //     + 2: argument with offset 0
423   //     + 3: argument with offset 1
424   //     + 4: ...
425 
426   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
427 }
428 
429 #ifndef PRODUCT
430 
431 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
432   if (!VerifyOops) return;
433   verify_oop_addr(Address(rsp, stack_offset));
434 }
435 
436 void C1_MacroAssembler::verify_not_null_oop(Register r) {
437   if (!VerifyOops) return;
438   Label not_null;
439   testptr(r, r);
440   jcc(Assembler::notZero, not_null);
441   stop("non-null oop required");
442   bind(not_null);
443   verify_oop(r);
444 }
445 
446 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
447 #ifdef ASSERT
448   if (inv_rax) movptr(rax, 0xDEAD);
449   if (inv_rbx) movptr(rbx, 0xDEAD);
450   if (inv_rcx) movptr(rcx, 0xDEAD);
451   if (inv_rdx) movptr(rdx, 0xDEAD);
452   if (inv_rsi) movptr(rsi, 0xDEAD);
453   if (inv_rdi) movptr(rdi, 0xDEAD);
454 #endif
455 }
456 
457 #endif // ifndef PRODUCT