1 /*
  2  * Copyright (c) 1999, 2023, 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 "precompiled.hpp"
 27 #include "c1/c1_MacroAssembler.hpp"
 28 #include "c1/c1_Runtime1.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 
 40 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
 41                                   FloatRegister f0, FloatRegister f1,
 42                                   Register result)
 43 {
 44   Label done;
 45   if (is_float) {
 46     fcmps(f0, f1);
 47   } else {
 48     fcmpd(f0, f1);
 49   }
 50   if (unordered_result < 0) {
 51     // we want -1 for unordered or less than, 0 for equal and 1 for
 52     // greater than.
 53     cset(result, NE);  // Not equal or unordered
 54     cneg(result, result, LT);  // Less than or unordered
 55   } else {
 56     // we want -1 for less than, 0 for equal and 1 for unordered or
 57     // greater than.
 58     cset(result, NE);  // Not equal or unordered
 59     cneg(result, result, LO);  // Less than
 60   }
 61 }
 62 
 63 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
 64   const int aligned_mask = BytesPerWord -1;
 65   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 66   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
 67   Label done;
 68   int null_check_offset = -1;
 69 
 70   verify_oop(obj);
 71 
 72   // save object being locked into the BasicObjectLock
 73   str(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 74 
 75   null_check_offset = offset();
 76 
 77   if (DiagnoseSyncOnValueBasedClasses != 0) {
 78     load_klass(hdr, obj);
 79     ldrw(hdr, Address(hdr, Klass::access_flags_offset()));
 80     tstw(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
 81     br(Assembler::NE, slow_case);
 82   }
 83 
 84   // Load object header
 85   ldr(hdr, Address(obj, hdr_offset));
 86   if (UseFastLocking) {
 87     fast_lock(obj, hdr, rscratch1, rscratch2, slow_case, false);
 88   } else {
 89     // and mark it as unlocked
 90     orr(hdr, hdr, markWord::unlocked_value);
 91     // save unlocked object header into the displaced header location on the stack
 92     str(hdr, Address(disp_hdr, 0));
 93     // test if object header is still the same (i.e. unlocked), and if so, store the
 94     // displaced header address in the object header - if it is not the same, get the
 95     // object header instead
 96     lea(rscratch2, Address(obj, hdr_offset));
 97     cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/NULL);
 98     // if the object header was the same, we're done
 99     // if the object header was not the same, it is now in the hdr register
100     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
101     //
102     // 1) (hdr & aligned_mask) == 0
103     // 2) sp <= hdr
104     // 3) hdr <= sp + page_size
105     //
106     // these 3 tests can be done by evaluating the following expression:
107     //
108     // (hdr - sp) & (aligned_mask - page_size)
109     //
110     // assuming both the stack pointer and page_size have their least
111     // significant 2 bits cleared and page_size is a power of 2
112     mov(rscratch1, sp);
113     sub(hdr, hdr, rscratch1);
114     ands(hdr, hdr, aligned_mask - (int)os::vm_page_size());
115     // for recursive locking, the result is zero => save it in the displaced header
116     // location (NULL in the displaced hdr location indicates recursive locking)
117     str(hdr, Address(disp_hdr, 0));
118     // otherwise we don't care about the result and handle locking via runtime call
119     cbnz(hdr, slow_case);
120   }
121   // done
122   bind(done);
123   increment(Address(rthread, JavaThread::held_monitor_count_offset()));
124   return null_check_offset;
125 }
126 
127 
128 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
129   const int aligned_mask = BytesPerWord -1;
130   const int hdr_offset = oopDesc::mark_offset_in_bytes();
131   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
132   Label done;
133 
134   if (!UseFastLocking) {
135     // load displaced header
136     ldr(hdr, Address(disp_hdr, 0));
137     // if the loaded hdr is NULL we had recursive locking
138     // if we had recursive locking, we are done
139     cbz(hdr, done);
140   }
141 
142   // load object
143   ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
144   verify_oop(obj);
145 
146   if (UseFastLocking) {
147     ldr(hdr, Address(obj, oopDesc::mark_offset_in_bytes()));
148     fast_unlock(obj, hdr, rscratch1, rscratch2, slow_case);
149   } else {
150     // test if object header is pointing to the displaced header, and if so, restore
151     // the displaced header in the object - if the object header is not pointing to
152     // the displaced header, get the object header instead
153     // if the object header was not pointing to the displaced header,
154     // we do unlocking via runtime call
155     if (hdr_offset) {
156       lea(rscratch1, Address(obj, hdr_offset));
157       cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case);
158     } else {
159       cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
160     }
161     // done
162     bind(done);
163   }
164   decrement(Address(rthread, JavaThread::held_monitor_count_offset()));
165 }
166 
167 
168 // Defines obj, preserves var_size_in_bytes
169 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
170   if (UseTLAB) {
171     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
172   } else {
173     b(slow_case);
174   }
175 }
176 
177 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
178   assert_different_registers(obj, klass, len);
179   if (UseCompactObjectHeaders) {
180     ldr(t1, Address(klass, Klass::prototype_header_offset()));
181     str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
182   } else {
183     // This assumes that all prototype bits fit in an int32_t
184     mov(t1, (int32_t)(intptr_t)markWord::prototype().value());
185     str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
186 
187     if (UseCompressedClassPointers) { // Take care not to kill klass
188       encode_klass_not_null(t1, klass);
189       strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
190     } else {
191       str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
192     }
193   }
194 
195   if (len->is_valid()) {
196     strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
197   } else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
198     store_klass_gap(obj, zr);
199   }
200 }
201 
202 // preserves obj, destroys len_in_bytes
203 //
204 // Scratch registers: t1 = r10, t2 = r11
205 //
206 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1, Register t2) {
207   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
208   assert(t1 == r10 && t2 == r11, "must be");
209 
210   Label done;
211 
212   // len_in_bytes is positive and ptr sized
213   subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
214   br(Assembler::EQ, done);
215 
216   // Zero first 4 bytes, if start offset is not word aligned.
217   if (!is_aligned(hdr_size_in_bytes, BytesPerWord)) {
218     strw(zr, Address(obj, hdr_size_in_bytes));
219     hdr_size_in_bytes += BytesPerInt;
220   }
221 
222   // zero_words() takes ptr in r10 and count in words in r11
223   mov(rscratch1, len_in_bytes);
224   lea(t1, Address(obj, hdr_size_in_bytes));
225   lsr(t2, rscratch1, LogBytesPerWord);
226   address tpc = zero_words(t1, t2);
227 
228   bind(done);
229   if (tpc == nullptr) {
230     Compilation::current()->bailout("no space for trampoline stub");
231   }
232 }
233 
234 
235 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
236   assert_different_registers(obj, t1, t2); // XXX really?
237   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
238 
239   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
240 
241   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
242 }
243 
244 // Scratch registers: t1 = r10, t2 = r11
245 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) {
246   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
247          "con_size_in_bytes is not multiple of alignment");
248   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
249 
250   initialize_header(obj, klass, noreg, t1, t2);
251 
252   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
253      // clear rest of allocated space
254      const Register index = t2;
255      if (var_size_in_bytes != noreg) {
256        mov(index, var_size_in_bytes);
257        initialize_body(obj, index, hdr_size_in_bytes, t1, t2);
258        if (Compilation::current()->bailed_out()) {
259          return;
260        }
261      } else if (con_size_in_bytes > hdr_size_in_bytes) {
262        con_size_in_bytes -= hdr_size_in_bytes;
263        lea(t1, Address(obj, hdr_size_in_bytes));
264        address tpc = zero_words(t1, con_size_in_bytes / BytesPerWord);
265        if (tpc == nullptr) {
266          Compilation::current()->bailout("no space for trampoline stub");
267          return;
268        }
269      }
270   }
271 
272   membar(StoreStore);
273 
274   if (CURRENT_ENV->dtrace_alloc_probes()) {
275     assert(obj == r0, "must be");
276     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
277   }
278 
279   verify_oop(obj);
280 }
281 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) {
282   assert_different_registers(obj, len, t1, t2, klass);
283 
284   // determine alignment mask
285   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
286 
287   // check for negative or excessive length
288   mov(rscratch1, (int32_t)max_array_allocation_length);
289   cmp(len, rscratch1);
290   br(Assembler::HS, slow_case);
291 
292   const Register arr_size = t2; // okay to be the same
293   // align object end
294   mov(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
295   add(arr_size, arr_size, len, ext::uxtw, f);
296   andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
297 
298   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
299 
300   initialize_header(obj, klass, len, t1, t2);
301 
302   // clear rest of allocated space
303   initialize_body(obj, arr_size, base_offset_in_bytes, t1, t2);
304   if (Compilation::current()->bailed_out()) {
305     return;
306   }
307 
308   membar(StoreStore);
309 
310   if (CURRENT_ENV->dtrace_alloc_probes()) {
311     assert(obj == r0, "must be");
312     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
313   }
314 
315   verify_oop(obj);
316 }
317 
318 
319 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
320   verify_oop(receiver);
321   // explicit NULL check not needed since load from [klass_offset] causes a trap
322   // check against inline cache
323   if (UseCompactObjectHeaders) {
324     assert(!MacroAssembler::needs_explicit_null_check(oopDesc::mark_offset_in_bytes()), "must add explicit null check");
325   } else {
326     assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
327   }
328 
329   cmp_klass(receiver, iCache, rscratch1);
330 }
331 
332 
333 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes, int max_monitors) {
334   assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
335   // Make sure there is enough stack space for this method's activation.
336   // Note that we do this before creating a frame.
337   generate_stack_overflow_check(bang_size_in_bytes);
338   MacroAssembler::build_frame(framesize);
339 
340   if (UseFastLocking && max_monitors > 0) {
341     Label ok;
342     ldr(r9, Address(rthread, JavaThread::lock_stack_current_offset()));
343     ldr(r10, Address(rthread, JavaThread::lock_stack_limit_offset()));
344     add(r9, r9, max_monitors * oopSize);
345     cmp(r9, r10);
346     br(Assembler::LT, ok);
347     assert(StubRoutines::aarch64::check_lock_stack() != NULL, "need runtime call stub");
348     far_call(StubRoutines::aarch64::check_lock_stack());
349     bind(ok);
350   }
351 
352   // Insert nmethod entry barrier into frame.
353   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
354   bs->nmethod_entry_barrier(this, NULL /* slow_path */, NULL /* continuation */, NULL /* guard */);
355 }
356 
357 void C1_MacroAssembler::remove_frame(int framesize) {
358   MacroAssembler::remove_frame(framesize);
359 }
360 
361 
362 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
363   // If we have to make this method not-entrant we'll overwrite its
364   // first instruction with a jump.  For this action to be legal we
365   // must ensure that this first instruction is a B, BL, NOP, BKPT,
366   // SVC, HVC, or SMC.  Make it a NOP.
367   nop();
368 }
369 
370 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
371   // rfp, + 0: link
372   //     + 1: return address
373   //     + 2: argument with offset 0
374   //     + 3: argument with offset 1
375   //     + 4: ...
376 
377   ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
378 }
379 
380 #ifndef PRODUCT
381 
382 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
383   if (!VerifyOops) return;
384   verify_oop_addr(Address(sp, stack_offset));
385 }
386 
387 void C1_MacroAssembler::verify_not_null_oop(Register r) {
388   if (!VerifyOops) return;
389   Label not_null;
390   cbnz(r, not_null);
391   stop("non-null oop required");
392   bind(not_null);
393   verify_oop(r);
394 }
395 
396 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
397 #ifdef ASSERT
398   static int nn;
399   if (inv_r0) mov(r0, 0xDEAD);
400   if (inv_r19) mov(r19, 0xDEAD);
401   if (inv_r2) mov(r2, nn++);
402   if (inv_r3) mov(r3, 0xDEAD);
403   if (inv_r4) mov(r4, 0xDEAD);
404   if (inv_r5) mov(r5, 0xDEAD);
405 #endif
406 }
407 #endif // ifndef PRODUCT