1 /*
  2  * Copyright (c) 2008, 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 "gc/shared/barrierSet.hpp"
 29 #include "gc/shared/barrierSetAssembler.hpp"
 30 #include "gc/shared/collectedHeap.hpp"
 31 #include "gc/shared/tlab_globals.hpp"
 32 #include "interpreter/interpreter.hpp"
 33 #include "oops/arrayOop.hpp"
 34 #include "oops/markWord.hpp"
 35 #include "runtime/basicLock.hpp"
 36 #include "runtime/os.hpp"
 37 #include "runtime/sharedRuntime.hpp"
 38 #include "runtime/stubRoutines.hpp"
 39 #include "utilities/powerOfTwo.hpp"
 40 
 41 // Note: Rtemp usage is this file should not impact C2 and should be
 42 // correct as long as it is not implicitly used in lower layers (the
 43 // arm [macro]assembler) and used with care in the other C1 specific
 44 // files.
 45 
 46 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
 47   Label verified;
 48   load_klass(Rtemp, receiver);
 49   cmp(Rtemp, iCache);
 50   b(verified, eq); // jump over alignment no-ops
 51   jump(SharedRuntime::get_ic_miss_stub(), relocInfo::runtime_call_type);
 52   align(CodeEntryAlignment);
 53   bind(verified);
 54 }
 55 
 56 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
 57   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
 58   assert((frame_size_in_bytes % StackAlignmentInBytes) == 0, "frame size should be aligned");
 59 
 60 
 61   arm_stack_overflow_check(bang_size_in_bytes, Rtemp);
 62 
 63   // FP can no longer be used to memorize SP. It may be modified
 64   // if this method contains a methodHandle call site
 65   raw_push(FP, LR);
 66   sub_slow(SP, SP, frame_size_in_bytes);
 67 
 68   // Insert nmethod entry barrier into frame.
 69   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 70   bs->nmethod_entry_barrier(this);
 71 }
 72 
 73 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
 74   add_slow(SP, SP, frame_size_in_bytes);
 75   raw_pop(FP, LR);
 76 }
 77 
 78 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
 79   if (breakAtEntry) {
 80     breakpoint();
 81   }
 82 }
 83 
 84 // Puts address of allocated object into register `obj` and end of allocated object into register `obj_end`.
 85 void C1_MacroAssembler::try_allocate(Register obj, Register obj_end, Register tmp1, Register tmp2,
 86                                      RegisterOrConstant size_expression, Label& slow_case) {
 87   if (UseTLAB) {
 88     tlab_allocate(obj, obj_end, tmp1, size_expression, slow_case);
 89   } else {
 90     b(slow_case);
 91   }
 92 }
 93 
 94 
 95 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register tmp) {
 96   assert_different_registers(obj, klass, len, tmp);
 97 
 98   mov(tmp, (intptr_t)markWord::prototype().value());
 99 
100   str(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
101   str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
102 
103   if (len->is_valid()) {
104     str_32(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
105   }
106 }
107 
108 
109 // Cleans object body [base..obj_end]. Clobbers `base` and `tmp` registers.
110 void C1_MacroAssembler::initialize_body(Register base, Register obj_end, Register tmp) {
111   zero_memory(base, obj_end, tmp);
112 }
113 
114 
115 void C1_MacroAssembler::initialize_object(Register obj, Register obj_end, Register klass,
116                                           Register len, Register tmp1, Register tmp2,
117                                           RegisterOrConstant header_size, int obj_size_in_bytes,
118                                           bool is_tlab_allocated)
119 {
120   assert_different_registers(obj, obj_end, klass, len, tmp1, tmp2);
121   initialize_header(obj, klass, len, tmp1);
122 
123   const Register ptr = tmp2;
124 
125   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
126     if (obj_size_in_bytes >= 0 && obj_size_in_bytes <= 8 * BytesPerWord) {
127       mov(tmp1, 0);
128       const int base = instanceOopDesc::header_size() * HeapWordSize;
129       for (int i = base; i < obj_size_in_bytes; i += wordSize) {
130         str(tmp1, Address(obj, i));
131       }
132     } else {
133       assert(header_size.is_constant() || header_size.as_register() == ptr, "code assumption");
134       add(ptr, obj, header_size);
135       initialize_body(ptr, obj_end, tmp1);
136     }
137   }
138 
139   // StoreStore barrier required after complete initialization
140   // (headers + content zeroing), before the object may escape.
141   membar(MacroAssembler::StoreStore, tmp1);
142 }
143 
144 void C1_MacroAssembler::allocate_object(Register obj, Register tmp1, Register tmp2, Register tmp3,
145                                         int header_size, int object_size,
146                                         Register klass, Label& slow_case) {
147   assert_different_registers(obj, tmp1, tmp2, tmp3, klass, Rtemp);
148   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
149   const int object_size_in_bytes = object_size * BytesPerWord;
150 
151   const Register obj_end = tmp1;
152   const Register len = noreg;
153 
154   if (Assembler::is_arith_imm_in_range(object_size_in_bytes)) {
155     try_allocate(obj, obj_end, tmp2, tmp3, object_size_in_bytes, slow_case);
156   } else {
157     // Rtemp should be free at c1 LIR level
158     mov_slow(Rtemp, object_size_in_bytes);
159     try_allocate(obj, obj_end, tmp2, tmp3, Rtemp, slow_case);
160   }
161   initialize_object(obj, obj_end, klass, len, tmp2, tmp3, instanceOopDesc::header_size() * HeapWordSize, object_size_in_bytes, /* is_tlab_allocated */ UseTLAB);
162 }
163 
164 void C1_MacroAssembler::allocate_array(Register obj, Register len,
165                                        Register tmp1, Register tmp2, Register tmp3,
166                                        int header_size_in_bytes, int element_size,
167                                        Register klass, Label& slow_case) {
168   assert_different_registers(obj, len, tmp1, tmp2, tmp3, klass, Rtemp);
169   const int scale_shift = exact_log2(element_size);
170   const Register obj_size = Rtemp; // Rtemp should be free at c1 LIR level
171 
172   cmp_32(len, max_array_allocation_length);
173   b(slow_case, hs);
174 
175   bool align_header = ((header_size_in_bytes | element_size) & MinObjAlignmentInBytesMask) != 0;
176   assert(align_header || ((header_size_in_bytes & MinObjAlignmentInBytesMask) == 0), "must be");
177   assert(align_header || ((element_size & MinObjAlignmentInBytesMask) == 0), "must be");
178 
179   mov(obj_size, header_size_in_bytes + (align_header ? (MinObjAlignmentInBytes - 1) : 0));
180   add_ptr_scaled_int32(obj_size, obj_size, len, scale_shift);
181 
182   if (align_header) {
183     align_reg(obj_size, obj_size, MinObjAlignmentInBytes);
184   }
185 
186   try_allocate(obj, tmp1, tmp2, tmp3, obj_size, slow_case);
187   initialize_object(obj, tmp1, klass, len, tmp2, tmp3, header_size_in_bytes, -1, /* is_tlab_allocated */ UseTLAB);
188 }
189 
190 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
191   Label done, fast_lock, fast_lock_done;
192   int null_check_offset = 0;
193 
194   const Register tmp2 = Rtemp; // Rtemp should be free at c1 LIR level
195   assert_different_registers(hdr, obj, disp_hdr, tmp2);
196 
197   assert(BasicObjectLock::lock_offset() == 0, "adjust this code");
198   const ByteSize obj_offset = BasicObjectLock::obj_offset();
199   const int mark_offset = BasicLock::displaced_header_offset_in_bytes();
200 
201   // save object being locked into the BasicObjectLock
202   str(obj, Address(disp_hdr, obj_offset));
203 
204   null_check_offset = offset();
205 
206   if (DiagnoseSyncOnValueBasedClasses != 0) {
207     load_klass(tmp2, obj);
208     ldr_u32(tmp2, Address(tmp2, Klass::access_flags_offset()));
209     tst(tmp2, JVM_ACC_IS_VALUE_BASED_CLASS);
210     b(slow_case, ne);
211   }
212 
213   assert(oopDesc::mark_offset_in_bytes() == 0, "Required by atomic instructions");
214 
215   if (LockingMode == LM_LIGHTWEIGHT) {
216 
217     Register t1 = disp_hdr; // Needs saving, probably
218     Register t2 = hdr;      // blow
219     Register t3 = Rtemp;    // blow
220 
221     lightweight_lock(obj /* obj */, t1, t2, t3, 1 /* savemask - save t1 */, slow_case);
222     // Success: fall through
223 
224   } else if (LockingMode == LM_LEGACY) {
225 
226     // On MP platforms the next load could return a 'stale' value if the memory location has been modified by another thread.
227     // That would be acceptable as ether CAS or slow case path is taken in that case.
228 
229     // Must be the first instruction here, because implicit null check relies on it
230     ldr(hdr, Address(obj, oopDesc::mark_offset_in_bytes()));
231 
232     tst(hdr, markWord::unlocked_value);
233     b(fast_lock, ne);
234 
235     // Check for recursive locking
236     // See comments in InterpreterMacroAssembler::lock_object for
237     // explanations on the fast recursive locking check.
238     // -1- test low 2 bits
239     movs(tmp2, AsmOperand(hdr, lsl, 30));
240     // -2- test (hdr - SP) if the low two bits are 0
241     sub(tmp2, hdr, SP, eq);
242     movs(tmp2, AsmOperand(tmp2, lsr, exact_log2(os::vm_page_size())), eq);
243     // If still 'eq' then recursive locking OK
244     // set to zero if recursive lock, set to non zero otherwise (see discussion in JDK-8267042)
245     str(tmp2, Address(disp_hdr, mark_offset));
246     b(fast_lock_done, eq);
247     // else need slow case
248     b(slow_case);
249 
250 
251     bind(fast_lock);
252     // Save previous object header in BasicLock structure and update the header
253     str(hdr, Address(disp_hdr, mark_offset));
254 
255     cas_for_lock_acquire(hdr, disp_hdr, obj, tmp2, slow_case);
256 
257     bind(fast_lock_done);
258   }
259   bind(done);
260 
261   return null_check_offset;
262 }
263 
264 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
265   assert_different_registers(hdr, obj, disp_hdr, Rtemp);
266   Register tmp2 = Rtemp;
267 
268   assert(BasicObjectLock::lock_offset() == 0, "adjust this code");
269   const ByteSize obj_offset = BasicObjectLock::obj_offset();
270   const int mark_offset = BasicLock::displaced_header_offset_in_bytes();
271 
272   Label done;
273 
274   assert(oopDesc::mark_offset_in_bytes() == 0, "Required by atomic instructions");
275 
276   if (LockingMode == LM_LIGHTWEIGHT) {
277 
278     ldr(obj, Address(disp_hdr, obj_offset));
279 
280     Register t1 = disp_hdr; // Needs saving, probably
281     Register t2 = hdr;      // blow
282     Register t3 = Rtemp;    // blow
283 
284     lightweight_unlock(obj /* object */, t1, t2, t3, 1 /* savemask (save t1) */,
285                        slow_case);
286     // Success: Fall through
287 
288   } else if (LockingMode == LM_LEGACY) {
289 
290     // Load displaced header and object from the lock
291     ldr(hdr, Address(disp_hdr, mark_offset));
292     // If hdr is null, we've got recursive locking and there's nothing more to do
293     cbz(hdr, done);
294 
295     // load object
296     ldr(obj, Address(disp_hdr, obj_offset));
297 
298     // Restore the object header
299     cas_for_lock_release(disp_hdr, hdr, obj, tmp2, slow_case);
300   }
301   bind(done);
302 }
303 
304 #ifndef PRODUCT
305 
306 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
307   if (!VerifyOops) return;
308   verify_oop_addr(Address(SP, stack_offset));
309 }
310 
311 void C1_MacroAssembler::verify_not_null_oop(Register r) {
312   Label not_null;
313   cbnz(r, not_null);
314   stop("non-null oop required");
315   bind(not_null);
316   if (!VerifyOops) return;
317   verify_oop(r);
318 }
319 
320 #endif // !PRODUCT