1 /*
  2  * Copyright (c) 1999, 2021, 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/biasedLocking.hpp"
 37 #include "runtime/os.hpp"
 38 #include "runtime/sharedRuntime.hpp"
 39 #include "runtime/stubRoutines.hpp"
 40 
 41 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
 42                                   FloatRegister f0, FloatRegister f1,
 43                                   Register result)
 44 {
 45   Label done;
 46   if (is_float) {
 47     fcmps(f0, f1);
 48   } else {
 49     fcmpd(f0, f1);
 50   }
 51   if (unordered_result < 0) {
 52     // we want -1 for unordered or less than, 0 for equal and 1 for
 53     // greater than.
 54     cset(result, NE);  // Not equal or unordered
 55     cneg(result, result, LT);  // Less than or unordered
 56   } else {
 57     // we want -1 for less than, 0 for equal and 1 for unordered or
 58     // greater than.
 59     cset(result, NE);  // Not equal or unordered
 60     cneg(result, result, LO);  // Less than
 61   }
 62 }
 63 
 64 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
 65   const int aligned_mask = BytesPerWord -1;
 66   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 67   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
 68   Label done;
 69   int null_check_offset = -1;
 70 
 71   verify_oop(obj);
 72 
 73   // save object being locked into the BasicObjectLock
 74   str(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 75 
 76   null_check_offset = offset();
 77 
 78   if (DiagnoseSyncOnValueBasedClasses != 0) {
 79     load_klass(hdr, obj);
 80     ldrw(hdr, Address(hdr, Klass::access_flags_offset()));
 81     tstw(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
 82     br(Assembler::NE, slow_case);
 83   }
 84 
 85   if (UseBiasedLocking) {
 86     assert(scratch != noreg, "should have scratch register at this point");
 87     biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case);
 88   }
 89 
 90   // Load object header
 91   ldr(hdr, Address(obj, hdr_offset));
 92   if (UseFastLocking) {
 93     fast_lock(obj, hdr, rscratch1, rscratch2, slow_case, false);
 94   } else {
 95     // and mark it as unlocked
 96     orr(hdr, hdr, markWord::unlocked_value);
 97     // save unlocked object header into the displaced header location on the stack
 98     str(hdr, Address(disp_hdr, 0));
 99     // test if object header is still the same (i.e. unlocked), and if so, store the
100     // displaced header address in the object header - if it is not the same, get the
101     // object header instead
102     lea(rscratch2, Address(obj, hdr_offset));
103     cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/NULL);
104     // if the object header was the same, we're done
105     // if the object header was not the same, it is now in the hdr register
106     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
107     //
108     // 1) (hdr & aligned_mask) == 0
109     // 2) sp <= hdr
110     // 3) hdr <= sp + page_size
111     //
112     // these 3 tests can be done by evaluating the following expression:
113     //
114     // (hdr - sp) & (aligned_mask - page_size)
115     //
116     // assuming both the stack pointer and page_size have their least
117     // significant 2 bits cleared and page_size is a power of 2
118     mov(rscratch1, sp);
119     sub(hdr, hdr, rscratch1);
120     ands(hdr, hdr, aligned_mask - os::vm_page_size());
121     // for recursive locking, the result is zero => save it in the displaced header
122     // location (NULL in the displaced hdr location indicates recursive locking)
123     str(hdr, Address(disp_hdr, 0));
124     // otherwise we don't care about the result and handle locking via runtime call
125     cbnz(hdr, slow_case);
126     // done
127     bind(done);
128   }
129   if (PrintBiasedLockingStatistics) {
130     lea(rscratch2, ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
131     addmw(Address(rscratch2, 0), 1, rscratch1);
132   }
133   return null_check_offset;
134 }
135 
136 
137 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
138   const int aligned_mask = BytesPerWord -1;
139   const int hdr_offset = oopDesc::mark_offset_in_bytes();
140   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
141   Label done;
142 
143   if (UseBiasedLocking) {
144     // load object
145     ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
146     biased_locking_exit(obj, hdr, done);
147   }
148 
149   if (UseFastLocking) {
150     // load object
151     ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
152     verify_oop(obj);
153     ldr(hdr, Address(obj, oopDesc::mark_offset_in_bytes()));
154     fast_unlock(obj, hdr, rscratch1, rscratch2, slow_case);
155   } else {
156     // load displaced header
157     ldr(hdr, Address(disp_hdr, 0));
158     // if the loaded hdr is NULL we had recursive locking
159     // if we had recursive locking, we are done
160     cbz(hdr, done);
161     if (!UseBiasedLocking) {
162       // load object
163       ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
164     }
165     verify_oop(obj);
166     // test if object header is pointing to the displaced header, and if so, restore
167     // the displaced header in the object - if the object header is not pointing to
168     // the displaced header, get the object header instead
169     // if the object header was not pointing to the displaced header,
170     // we do unlocking via runtime call
171     if (hdr_offset) {
172       lea(rscratch1, Address(obj, hdr_offset));
173       cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case);
174     } else {
175       cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
176     }
177   }
178   // done
179   bind(done);
180 }
181 
182 
183 // Defines obj, preserves var_size_in_bytes
184 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
185   if (UseTLAB) {
186     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
187   } else {
188     eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
189   }
190 }
191 
192 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
193   assert_different_registers(obj, klass, len);
194   if (UseCompactObjectHeaders || (UseBiasedLocking && !len->is_valid())) {
195     assert_different_registers(obj, klass, len, t1, t2);
196     ldr(t1, Address(klass, Klass::prototype_header_offset()));
197   } else {
198     // This assumes that all prototype bits fit in an int32_t
199     mov(t1, (int32_t)(intptr_t)markWord::prototype().value());
200   }
201   str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
202 
203   if (!UseCompactObjectHeaders) {
204     if (UseCompressedClassPointers) { // Take care not to kill klass
205       encode_klass_not_null(t1, klass);
206       strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
207     } else {
208       str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
209     }
210   }
211 
212   if (len->is_valid()) {
213     strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
214   } else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
215     store_klass_gap(obj, zr);
216   }
217 }
218 
219 // preserves obj, destroys len_in_bytes
220 //
221 // Scratch registers: t1 = r10, t2 = r11
222 //
223 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1, Register t2) {
224   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
225   assert(t1 == r10 && t2 == r11, "must be");
226 
227   Label done;
228 
229   // len_in_bytes is positive and ptr sized
230   subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
231   br(Assembler::EQ, done);
232 
233   // Zero first 4 bytes, if start offset is not word aligned.
234   if (!is_aligned(hdr_size_in_bytes, BytesPerWord)) {
235     strw(zr, Address(obj, hdr_size_in_bytes));
236     hdr_size_in_bytes += BytesPerInt;
237   }
238 
239   // zero_words() takes ptr in r10 and count in words in r11
240   mov(rscratch1, len_in_bytes);
241   lea(t1, Address(obj, hdr_size_in_bytes));
242   lsr(t2, rscratch1, LogBytesPerWord);
243   zero_words(t1, t2);
244 
245   bind(done);
246 }
247 
248 
249 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
250   assert_different_registers(obj, t1, t2); // XXX really?
251   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
252 
253   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
254 
255   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
256 }
257 
258 // Scratch registers: t1 = r10, t2 = r11
259 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) {
260   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
261          "con_size_in_bytes is not multiple of alignment");
262   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
263 
264   initialize_header(obj, klass, noreg, t1, t2);
265 
266   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
267      // clear rest of allocated space
268      const Register index = t2;
269      if (var_size_in_bytes != noreg) {
270        mov(index, var_size_in_bytes);
271        initialize_body(obj, index, hdr_size_in_bytes, t1, t2);
272      } else if (con_size_in_bytes > hdr_size_in_bytes) {
273        con_size_in_bytes -= hdr_size_in_bytes;
274        lea(t1, Address(obj, hdr_size_in_bytes));
275        zero_words(t1, con_size_in_bytes / BytesPerWord);
276      }
277   }
278 
279   membar(StoreStore);
280 
281   if (CURRENT_ENV->dtrace_alloc_probes()) {
282     assert(obj == r0, "must be");
283     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
284   }
285 
286   verify_oop(obj);
287 }
288 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) {
289   assert_different_registers(obj, len, t1, t2, klass);
290 
291   // determine alignment mask
292   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
293 
294   // check for negative or excessive length
295   mov(rscratch1, (int32_t)max_array_allocation_length);
296   cmp(len, rscratch1);
297   br(Assembler::HS, slow_case);
298 
299   const Register arr_size = t2; // okay to be the same
300   // align object end
301   mov(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
302   add(arr_size, arr_size, len, ext::uxtw, f);
303   andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
304 
305   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
306 
307   initialize_header(obj, klass, len, t1, t2);
308 
309   // clear rest of allocated space
310   initialize_body(obj, arr_size, base_offset_in_bytes, t1, t2);
311 
312   membar(StoreStore);
313 
314   if (CURRENT_ENV->dtrace_alloc_probes()) {
315     assert(obj == r0, "must be");
316     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
317   }
318 
319   verify_oop(obj);
320 }
321 
322 
323 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
324   verify_oop(receiver);
325   // explicit NULL check not needed since load from [klass_offset] causes a trap
326   // check against inline cache
327   if (UseCompactObjectHeaders) {
328     assert(!MacroAssembler::needs_explicit_null_check(oopDesc::mark_offset_in_bytes()), "must add explicit null check");
329   } else {
330     assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
331   }
332   cmp_klass(receiver, iCache, rscratch1);
333 }
334 
335 
336 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes, int max_monitors) {
337   assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
338   // Make sure there is enough stack space for this method's activation.
339   // Note that we do this before creating a frame.
340   generate_stack_overflow_check(bang_size_in_bytes);
341   MacroAssembler::build_frame(framesize);
342 
343   if (UseFastLocking && max_monitors > 0) {
344     Label ok;
345     ldr(r9, Address(rthread, JavaThread::lock_stack_current_offset()));
346     add(r9, r9, max_monitors * oopSize);
347     ldr(r10, Address(rthread, JavaThread::lock_stack_limit_offset()));
348     cmp(r9, r10);
349     br(Assembler::LT, ok);
350     assert(StubRoutines::aarch64::check_lock_stack() != NULL, "need runtime call stub");
351     movptr(rscratch1, (uintptr_t) StubRoutines::aarch64::check_lock_stack());
352     blr(rscratch1);
353     bind(ok);
354   }
355 
356   // Insert nmethod entry barrier into frame.
357   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
358   bs->nmethod_entry_barrier(this);
359 }
360 
361 void C1_MacroAssembler::remove_frame(int framesize) {
362   MacroAssembler::remove_frame(framesize);
363 }
364 
365 
366 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
367   // If we have to make this method not-entrant we'll overwrite its
368   // first instruction with a jump.  For this action to be legal we
369   // must ensure that this first instruction is a B, BL, NOP, BKPT,
370   // SVC, HVC, or SMC.  Make it a NOP.
371   nop();
372 }
373 
374 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
375   // rbp, + 0: link
376   //     + 1: return address
377   //     + 2: argument with offset 0
378   //     + 3: argument with offset 1
379   //     + 4: ...
380 
381   ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
382 }
383 
384 #ifndef PRODUCT
385 
386 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
387   if (!VerifyOops) return;
388   verify_oop_addr(Address(sp, stack_offset), "oop");
389 }
390 
391 void C1_MacroAssembler::verify_not_null_oop(Register r) {
392   if (!VerifyOops) return;
393   Label not_null;
394   cbnz(r, not_null);
395   stop("non-null oop required");
396   bind(not_null);
397   verify_oop(r);
398 }
399 
400 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
401 #ifdef ASSERT
402   static int nn;
403   if (inv_r0) mov(r0, 0xDEAD);
404   if (inv_r19) mov(r19, 0xDEAD);
405   if (inv_r2) mov(r2, nn++);
406   if (inv_r3) mov(r3, 0xDEAD);
407   if (inv_r4) mov(r4, 0xDEAD);
408   if (inv_r5) mov(r5, 0xDEAD);
409 #endif
410 }
411 #endif // ifndef PRODUCT