1 /*
  2  * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
  3  * Copyright (c) 2014, 2021, Red Hat Inc. All rights reserved.
  4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  5  *
  6  * This code is free software; you can redistribute it and/or modify it
  7  * under the terms of the GNU General Public License version 2 only, as
  8  * published by the Free Software Foundation.
  9  *
 10  * This code is distributed in the hope that it will be useful, but WITHOUT
 11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 13  * version 2 for more details (a copy is included in the LICENSE file that
 14  * accompanied this code).
 15  *
 16  * You should have received a copy of the GNU General Public License version
 17  * 2 along with this work; if not, write to the Free Software Foundation,
 18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 19  *
 20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 21  * or visit www.oracle.com if you need additional information or have any
 22  * questions.
 23  *
 24  */
 25 
 26 #include "c1/c1_MacroAssembler.hpp"
 27 #include "c1/c1_Runtime1.hpp"
 28 #include "gc/shared/barrierSetAssembler.hpp"
 29 #include "gc/shared/collectedHeap.hpp"
 30 #include "gc/shared/tlab_globals.hpp"
 31 #include "interpreter/interpreter.hpp"
 32 #include "oops/arrayOop.hpp"
 33 #include "oops/markWord.hpp"
 34 #include "runtime/basicLock.hpp"
 35 #include "runtime/os.hpp"
 36 #include "runtime/sharedRuntime.hpp"
 37 #include "runtime/stubRoutines.hpp"
 38 
 39 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
 40                                   FloatRegister f0, FloatRegister f1,
 41                                   Register result)
 42 {
 43   Label done;
 44   if (is_float) {
 45     fcmps(f0, f1);
 46   } else {
 47     fcmpd(f0, f1);
 48   }
 49   if (unordered_result < 0) {
 50     // we want -1 for unordered or less than, 0 for equal and 1 for
 51     // greater than.
 52     cset(result, NE);  // Not equal or unordered
 53     cneg(result, result, LT);  // Less than or unordered
 54   } else {
 55     // we want -1 for less than, 0 for equal and 1 for unordered or
 56     // greater than.
 57     cset(result, NE);  // Not equal or unordered
 58     cneg(result, result, LO);  // Less than
 59   }
 60 }
 61 
 62 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register temp, Label& slow_case) {
 63   const int aligned_mask = BytesPerWord -1;
 64   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 65   assert_different_registers(hdr, obj, disp_hdr, temp, rscratch2);
 66   int null_check_offset = -1;
 67 
 68   verify_oop(obj);
 69 
 70   // save object being locked into the BasicObjectLock
 71   str(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
 72 
 73   null_check_offset = offset();
 74 
 75   if (DiagnoseSyncOnValueBasedClasses != 0) {
 76     load_klass(hdr, obj);
 77     ldrb(hdr, Address(hdr, Klass::misc_flags_offset()));
 78     tst(hdr, KlassFlags::_misc_is_value_based_class);
 79     br(Assembler::NE, slow_case);
 80   }
 81 
 82   if (LockingMode == LM_LIGHTWEIGHT) {
 83     lightweight_lock(disp_hdr, obj, hdr, temp, rscratch2, slow_case);
 84   } else if (LockingMode == LM_LEGACY) {
 85     Label done;
 86     // Load object header
 87     ldr(hdr, Address(obj, hdr_offset));
 88     // and mark it as unlocked
 89     orr(hdr, hdr, markWord::unlocked_value);
 90     // save unlocked object header into the displaced header location on the stack
 91     str(hdr, Address(disp_hdr, 0));
 92     // test if object header is still the same (i.e. unlocked), and if so, store the
 93     // displaced header address in the object header - if it is not the same, get the
 94     // object header instead
 95     lea(rscratch2, Address(obj, hdr_offset));
 96     cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/nullptr);
 97     // if the object header was the same, we're done
 98     // if the object header was not the same, it is now in the hdr register
 99     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
100     //
101     // 1) (hdr & aligned_mask) == 0
102     // 2) sp <= hdr
103     // 3) hdr <= sp + page_size
104     //
105     // these 3 tests can be done by evaluating the following expression:
106     //
107     // (hdr - sp) & (aligned_mask - page_size)
108     //
109     // assuming both the stack pointer and page_size have their least
110     // significant 2 bits cleared and page_size is a power of 2
111     mov(rscratch1, sp);
112     sub(hdr, hdr, rscratch1);
113     ands(hdr, hdr, aligned_mask - (int)os::vm_page_size());
114     // for recursive locking, the result is zero => save it in the displaced header
115     // location (null in the displaced hdr location indicates recursive locking)
116     str(hdr, Address(disp_hdr, 0));
117     // otherwise we don't care about the result and handle locking via runtime call
118     cbnz(hdr, slow_case);
119     // done
120     bind(done);
121     inc_held_monitor_count(rscratch1);
122   }
123   return null_check_offset;
124 }
125 
126 
127 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Register temp, Label& slow_case) {
128   const int aligned_mask = BytesPerWord -1;
129   const int hdr_offset = oopDesc::mark_offset_in_bytes();
130   assert_different_registers(hdr, obj, disp_hdr, temp, rscratch2);
131   Label done;
132 
133   if (LockingMode != LM_LIGHTWEIGHT) {
134     // load displaced header
135     ldr(hdr, Address(disp_hdr, 0));
136     // if the loaded hdr is null we had recursive locking
137     // if we had recursive locking, we are done
138     cbz(hdr, done);
139   }
140 
141   // load object
142   ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
143   verify_oop(obj);
144 
145   if (LockingMode == LM_LIGHTWEIGHT) {
146     lightweight_unlock(obj, hdr, temp, rscratch2, slow_case);
147   } else if (LockingMode == LM_LEGACY) {
148     // test if object header is pointing to the displaced header, and if so, restore
149     // the displaced header in the object - if the object header is not pointing to
150     // the displaced header, get the object header instead
151     // if the object header was not pointing to the displaced header,
152     // we do unlocking via runtime call
153     if (hdr_offset) {
154       lea(rscratch1, Address(obj, hdr_offset));
155       cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case);
156     } else {
157       cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
158     }
159     // done
160     bind(done);
161     dec_held_monitor_count(rscratch1);
162   }
163 }
164 
165 
166 // Defines obj, preserves var_size_in_bytes
167 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
168   if (UseTLAB) {
169     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
170   } else {
171     b(slow_case);
172   }
173 }
174 
175 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
176   assert_different_registers(obj, klass, len);
177 
178   if (UseCompactObjectHeaders) {
179     ldr(t1, Address(klass, Klass::prototype_header_offset()));
180     str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
181   } else {
182     mov(t1, checked_cast<int32_t>(markWord::prototype().value()));
183     str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
184     if (UseCompressedClassPointers) { // Take care not to kill klass
185       encode_klass_not_null(t1, klass);
186       strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
187     } else {
188       str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
189     }
190   }
191 
192   if (len->is_valid()) {
193     strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
194     int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
195     if (!is_aligned(base_offset, BytesPerWord)) {
196       assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
197       // Clear gap/first 4 bytes following the length field.
198       strw(zr, Address(obj, base_offset));
199     }
200   } else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
201     store_klass_gap(obj, zr);
202   }
203 }
204 
205 // preserves obj, destroys len_in_bytes
206 //
207 // Scratch registers: t1 = r10, t2 = r11
208 //
209 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1, Register t2) {
210   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
211   assert(t1 == r10 && t2 == r11, "must be");
212 
213   Label done;
214 
215   // len_in_bytes is positive and ptr sized
216   subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
217   br(Assembler::EQ, done);
218 
219   // zero_words() takes ptr in r10 and count in words in r11
220   mov(rscratch1, len_in_bytes);
221   lea(t1, Address(obj, hdr_size_in_bytes));
222   lsr(t2, rscratch1, LogBytesPerWord);
223   address tpc = zero_words(t1, t2);
224 
225   bind(done);
226   if (tpc == nullptr) {
227     Compilation::current()->bailout("no space for trampoline stub");
228   }
229 }
230 
231 
232 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
233   assert_different_registers(obj, t1, t2); // XXX really?
234   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
235 
236   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
237 
238   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
239 }
240 
241 // Scratch registers: t1 = r10, t2 = r11
242 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) {
243   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
244          "con_size_in_bytes is not multiple of alignment");
245   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
246 
247   initialize_header(obj, klass, noreg, t1, t2);
248 
249   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
250      // clear rest of allocated space
251      const Register index = t2;
252      if (var_size_in_bytes != noreg) {
253        mov(index, var_size_in_bytes);
254        initialize_body(obj, index, hdr_size_in_bytes, t1, t2);
255        if (Compilation::current()->bailed_out()) {
256          return;
257        }
258      } else if (con_size_in_bytes > hdr_size_in_bytes) {
259        con_size_in_bytes -= hdr_size_in_bytes;
260        lea(t1, Address(obj, hdr_size_in_bytes));
261        address tpc = zero_words(t1, con_size_in_bytes / BytesPerWord);
262        if (tpc == nullptr) {
263          Compilation::current()->bailout("no space for trampoline stub");
264          return;
265        }
266      }
267   }
268 
269   membar(StoreStore);
270 
271   if (CURRENT_ENV->dtrace_alloc_probes()) {
272     assert(obj == r0, "must be");
273     far_call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
274   }
275 
276   verify_oop(obj);
277 }
278 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int base_offset_in_bytes, int f, Register klass, Label& slow_case, bool zero_array) {
279   assert_different_registers(obj, len, t1, t2, klass);
280 
281   // determine alignment mask
282   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
283 
284   // check for negative or excessive length
285   mov(rscratch1, (int32_t)max_array_allocation_length);
286   cmp(len, rscratch1);
287   br(Assembler::HS, slow_case);
288 
289   const Register arr_size = t2; // okay to be the same
290   // align object end
291   mov(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
292   add(arr_size, arr_size, len, ext::uxtw, f);
293   andr(arr_size, 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   // Align-up to word boundary, because we clear the 4 bytes potentially
300   // following the length field in initialize_header().
301   int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
302   // clear rest of allocated space
303   if (zero_array) {
304     initialize_body(obj, arr_size, base_offset, t1, t2);
305   }
306   if (Compilation::current()->bailed_out()) {
307     return;
308   }
309 
310   membar(StoreStore);
311 
312   if (CURRENT_ENV->dtrace_alloc_probes()) {
313     assert(obj == r0, "must be");
314     far_call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
315   }
316 
317   verify_oop(obj);
318 }
319 
320 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) {
321   assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
322   // Make sure there is enough stack space for this method's activation.
323   // Note that we do this before creating a frame.
324   generate_stack_overflow_check(bang_size_in_bytes);
325   MacroAssembler::build_frame(framesize);
326 
327   // Insert nmethod entry barrier into frame.
328   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
329   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
330 }
331 
332 void C1_MacroAssembler::remove_frame(int framesize) {
333   MacroAssembler::remove_frame(framesize);
334 }
335 
336 
337 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
338   // If we have to make this method not-entrant we'll overwrite its
339   // first instruction with a jump.  For this action to be legal we
340   // must ensure that this first instruction is a B, BL, NOP, BKPT,
341   // SVC, HVC, or SMC.  Make it a NOP.
342   nop();
343 }
344 
345 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
346   // rfp, + 0: link
347   //     + 1: return address
348   //     + 2: argument with offset 0
349   //     + 3: argument with offset 1
350   //     + 4: ...
351 
352   ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
353 }
354 
355 #ifndef PRODUCT
356 
357 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
358   if (!VerifyOops) return;
359   verify_oop_addr(Address(sp, stack_offset));
360 }
361 
362 void C1_MacroAssembler::verify_not_null_oop(Register r) {
363   if (!VerifyOops) return;
364   Label not_null;
365   cbnz(r, not_null);
366   stop("non-null oop required");
367   bind(not_null);
368   verify_oop(r);
369 }
370 
371 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
372 #ifdef ASSERT
373   static int nn;
374   if (inv_r0) mov(r0, 0xDEAD);
375   if (inv_r19) mov(r19, 0xDEAD);
376   if (inv_r2) mov(r2, nn++);
377   if (inv_r3) mov(r3, 0xDEAD);
378   if (inv_r4) mov(r4, 0xDEAD);
379   if (inv_r5) mov(r5, 0xDEAD);
380 #endif
381 }
382 #endif // ifndef PRODUCT