1 /*
  2  * Copyright (c) 1999, 2022, 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   // and mark it as unlocked
 87   orr(hdr, hdr, markWord::unlocked_value);
 88   // save unlocked object header into the displaced header location on the stack
 89   str(hdr, Address(disp_hdr, 0));
 90   // test if object header is still the same (i.e. unlocked), and if so, store the
 91   // displaced header address in the object header - if it is not the same, get the
 92   // object header instead
 93   lea(rscratch2, Address(obj, hdr_offset));
 94   cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/NULL);
 95   // if the object header was the same, we're done
 96   // if the object header was not the same, it is now in the hdr register
 97   // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
 98   //
 99   // 1) (hdr & aligned_mask) == 0
100   // 2) sp <= hdr
101   // 3) hdr <= sp + page_size
102   //
103   // these 3 tests can be done by evaluating the following expression:
104   //
105   // (hdr - sp) & (aligned_mask - page_size)
106   //
107   // assuming both the stack pointer and page_size have their least
108   // significant 2 bits cleared and page_size is a power of 2
109   mov(rscratch1, sp);
110   sub(hdr, hdr, rscratch1);
111   ands(hdr, hdr, aligned_mask - os::vm_page_size());
112   // for recursive locking, the result is zero => save it in the displaced header
113   // location (NULL in the displaced hdr location indicates recursive locking)
114   str(hdr, Address(disp_hdr, 0));
115   // otherwise we don't care about the result and handle locking via runtime call
116   cbnz(hdr, slow_case);
117   // done
118   bind(done);
119   increment(Address(rthread, JavaThread::held_monitor_count_offset()));
120   return null_check_offset;
121 }
122 
123 
124 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
125   const int aligned_mask = BytesPerWord -1;
126   const int hdr_offset = oopDesc::mark_offset_in_bytes();
127   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
128   Label done;
129 
130   // load displaced header
131   ldr(hdr, Address(disp_hdr, 0));
132   // if the loaded hdr is NULL we had recursive locking
133   // if we had recursive locking, we are done
134   cbz(hdr, done);
135   // load object
136   ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
137   verify_oop(obj);
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   // if the object header was not pointing to the displaced header,
142   // we do unlocking via runtime call
143   if (hdr_offset) {
144     lea(rscratch1, Address(obj, hdr_offset));
145     cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case);
146   } else {
147     cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
148   }
149   // done
150   bind(done);
151   decrement(Address(rthread, JavaThread::held_monitor_count_offset()));
152 }
153 
154 
155 // Defines obj, preserves var_size_in_bytes
156 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
157   if (UseTLAB) {
158     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
159   } else {
160     b(slow_case);
161   }
162 }
163 
164 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
165   assert_different_registers(obj, klass, len);
166   // This assumes that all prototype bits fit in an int32_t
167   mov(t1, (int32_t)(intptr_t)markWord::prototype().value());
168   str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
169 
170   if (UseCompressedClassPointers) { // Take care not to kill klass
171     encode_klass_not_null(t1, klass);
172     strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
173   } else {
174     str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
175   }
176 
177   if (len->is_valid()) {
178     strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
179   } else if (UseCompressedClassPointers) {
180     store_klass_gap(obj, zr);
181   }
182 }
183 
184 // preserves obj, destroys len_in_bytes
185 //
186 // Scratch registers: t1 = r10, t2 = r11
187 //
188 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1, Register t2) {
189   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
190   assert(t1 == r10 && t2 == r11, "must be");
191 
192   Label done;
193 
194   // len_in_bytes is positive and ptr sized
195   subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
196   br(Assembler::EQ, done);
197 
198   // zero_words() takes ptr in r10 and count in words in r11
199   mov(rscratch1, len_in_bytes);
200   lea(t1, Address(obj, hdr_size_in_bytes));
201   lsr(t2, rscratch1, LogBytesPerWord);
202   zero_words(t1, t2);
203 
204   bind(done);
205 }
206 
207 
208 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
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 // Scratch registers: t1 = r10, t2 = r11
218 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) {
219   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
220          "con_size_in_bytes is not multiple of alignment");
221   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
222 
223   initialize_header(obj, klass, noreg, t1, t2);
224 
225   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
226      // clear rest of allocated space
227      const Register index = t2;
228      if (var_size_in_bytes != noreg) {
229        mov(index, var_size_in_bytes);
230        initialize_body(obj, index, hdr_size_in_bytes, t1, t2);
231      } else if (con_size_in_bytes > hdr_size_in_bytes) {
232        con_size_in_bytes -= hdr_size_in_bytes;
233        lea(t1, Address(obj, hdr_size_in_bytes));
234        zero_words(t1, con_size_in_bytes / BytesPerWord);
235      }
236   }
237 
238   membar(StoreStore);
239 
240   if (CURRENT_ENV->dtrace_alloc_probes()) {
241     assert(obj == r0, "must be");
242     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
243   }
244 
245   verify_oop(obj);
246 }
247 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, int f, Register klass, Label& slow_case) {
248   assert_different_registers(obj, len, t1, t2, klass);
249 
250   // determine alignment mask
251   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
252 
253   // check for negative or excessive length
254   mov(rscratch1, (int32_t)max_array_allocation_length);
255   cmp(len, rscratch1);
256   br(Assembler::HS, slow_case);
257 
258   const Register arr_size = t2; // okay to be the same
259   // align object end
260   mov(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
261   add(arr_size, arr_size, len, ext::uxtw, f);
262   andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
263 
264   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
265 
266   initialize_header(obj, klass, len, t1, t2);
267 
268   // clear rest of allocated space
269   initialize_body(obj, arr_size, header_size * BytesPerWord, t1, t2);
270 
271   membar(StoreStore);
272 
273   if (CURRENT_ENV->dtrace_alloc_probes()) {
274     assert(obj == r0, "must be");
275     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
276   }
277 
278   verify_oop(obj);
279 }
280 
281 
282 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
283   verify_oop(receiver);
284   // explicit NULL check not needed since load from [klass_offset] causes a trap
285   // check against inline cache
286   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
287 
288   cmp_klass(receiver, iCache, rscratch1);
289 }
290 
291 
292 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) {
293   assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
294   // Make sure there is enough stack space for this method's activation.
295   // Note that we do this before creating a frame.
296   generate_stack_overflow_check(bang_size_in_bytes);
297   MacroAssembler::build_frame(framesize);
298 
299   // Insert nmethod entry barrier into frame.
300   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
301   bs->nmethod_entry_barrier(this, NULL /* slow_path */, NULL /* continuation */, NULL /* guard */);
302 }
303 
304 void C1_MacroAssembler::remove_frame(int framesize) {
305   MacroAssembler::remove_frame(framesize);
306 }
307 
308 
309 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
310   // If we have to make this method not-entrant we'll overwrite its
311   // first instruction with a jump.  For this action to be legal we
312   // must ensure that this first instruction is a B, BL, NOP, BKPT,
313   // SVC, HVC, or SMC.  Make it a NOP.
314   nop();
315 }
316 
317 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
318   // rfp, + 0: link
319   //     + 1: return address
320   //     + 2: argument with offset 0
321   //     + 3: argument with offset 1
322   //     + 4: ...
323 
324   ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
325 }
326 
327 #ifndef PRODUCT
328 
329 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
330   if (!VerifyOops) return;
331   verify_oop_addr(Address(sp, stack_offset));
332 }
333 
334 void C1_MacroAssembler::verify_not_null_oop(Register r) {
335   if (!VerifyOops) return;
336   Label not_null;
337   cbnz(r, not_null);
338   stop("non-null oop required");
339   bind(not_null);
340   verify_oop(r);
341 }
342 
343 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
344 #ifdef ASSERT
345   static int nn;
346   if (inv_r0) mov(r0, 0xDEAD);
347   if (inv_r19) mov(r19, 0xDEAD);
348   if (inv_r2) mov(r2, nn++);
349   if (inv_r3) mov(r3, 0xDEAD);
350   if (inv_r4) mov(r4, 0xDEAD);
351   if (inv_r5) mov(r5, 0xDEAD);
352 #endif
353 }
354 #endif // ifndef PRODUCT