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   // 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 - (int)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   address tpc = zero_words(t1, t2);
203 
204   bind(done);
205   if (tpc == nullptr) {
206     Compilation::current()->bailout("no space for trampoline stub");
207   }
208 }
209 
210 
211 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
212   assert_different_registers(obj, t1, t2); // XXX really?
213   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
214 
215   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
216 
217   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
218 }
219 
220 // Scratch registers: t1 = r10, t2 = r11
221 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) {
222   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
223          "con_size_in_bytes is not multiple of alignment");
224   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
225 
226   initialize_header(obj, klass, noreg, t1, t2);
227 
228   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
229      // clear rest of allocated space
230      const Register index = t2;
231      if (var_size_in_bytes != noreg) {
232        mov(index, var_size_in_bytes);
233        initialize_body(obj, index, hdr_size_in_bytes, t1, t2);
234        if (Compilation::current()->bailed_out()) {
235          return;
236        }
237      } else if (con_size_in_bytes > hdr_size_in_bytes) {
238        con_size_in_bytes -= hdr_size_in_bytes;
239        lea(t1, Address(obj, hdr_size_in_bytes));
240        address tpc = zero_words(t1, con_size_in_bytes / BytesPerWord);
241        if (tpc == nullptr) {
242          Compilation::current()->bailout("no space for trampoline stub");
243          return;
244        }
245      }
246   }
247 
248   membar(StoreStore);
249 
250   if (CURRENT_ENV->dtrace_alloc_probes()) {
251     assert(obj == r0, "must be");
252     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
253   }
254 
255   verify_oop(obj);
256 }
257 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, int f, Register klass, Label& slow_case) {
258   assert_different_registers(obj, len, t1, t2, klass);
259 
260   // determine alignment mask
261   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
262 
263   // check for negative or excessive length
264   mov(rscratch1, (int32_t)max_array_allocation_length);
265   cmp(len, rscratch1);
266   br(Assembler::HS, slow_case);
267 
268   const Register arr_size = t2; // okay to be the same
269   // align object end
270   mov(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
271   add(arr_size, arr_size, len, ext::uxtw, f);
272   andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
273 
274   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
275 
276   initialize_header(obj, klass, len, t1, t2);
277 
278   // clear rest of allocated space
279   initialize_body(obj, arr_size, header_size * BytesPerWord, t1, t2);
280   if (Compilation::current()->bailed_out()) {
281     return;
282   }
283 
284   membar(StoreStore);
285 
286   if (CURRENT_ENV->dtrace_alloc_probes()) {
287     assert(obj == r0, "must be");
288     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
289   }
290 
291   verify_oop(obj);
292 }
293 
294 
295 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
296   verify_oop(receiver);
297   // explicit NULL check not needed since load from [klass_offset] causes a trap
298   // check against inline cache
299   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
300 
301   cmp_klass(receiver, iCache, rscratch1);
302 }
303 
304 
305 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) {
306   assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
307   // Make sure there is enough stack space for this method's activation.
308   // Note that we do this before creating a frame.
309   generate_stack_overflow_check(bang_size_in_bytes);
310   MacroAssembler::build_frame(framesize);
311 
312   // Insert nmethod entry barrier into frame.
313   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
314   bs->nmethod_entry_barrier(this, NULL /* slow_path */, NULL /* continuation */, NULL /* guard */);
315 }
316 
317 void C1_MacroAssembler::remove_frame(int framesize) {
318   MacroAssembler::remove_frame(framesize);
319 }
320 
321 
322 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
323   // If we have to make this method not-entrant we'll overwrite its
324   // first instruction with a jump.  For this action to be legal we
325   // must ensure that this first instruction is a B, BL, NOP, BKPT,
326   // SVC, HVC, or SMC.  Make it a NOP.
327   nop();
328 }
329 
330 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
331   // rfp, + 0: link
332   //     + 1: return address
333   //     + 2: argument with offset 0
334   //     + 3: argument with offset 1
335   //     + 4: ...
336 
337   ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
338 }
339 
340 #ifndef PRODUCT
341 
342 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
343   if (!VerifyOops) return;
344   verify_oop_addr(Address(sp, stack_offset));
345 }
346 
347 void C1_MacroAssembler::verify_not_null_oop(Register r) {
348   if (!VerifyOops) return;
349   Label not_null;
350   cbnz(r, not_null);
351   stop("non-null oop required");
352   bind(not_null);
353   verify_oop(r);
354 }
355 
356 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
357 #ifdef ASSERT
358   static int nn;
359   if (inv_r0) mov(r0, 0xDEAD);
360   if (inv_r19) mov(r19, 0xDEAD);
361   if (inv_r2) mov(r2, nn++);
362   if (inv_r3) mov(r3, 0xDEAD);
363   if (inv_r4) mov(r4, 0xDEAD);
364   if (inv_r5) mov(r5, 0xDEAD);
365 #endif
366 }
367 #endif // ifndef PRODUCT