1 /*
  2  * Copyright (c) 1999, 2024, 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 "code/compiledIC.hpp"
 29 #include "compiler/compilerDefinitions.inline.hpp"
 30 #include "gc/shared/barrierSet.hpp"
 31 #include "gc/shared/barrierSetAssembler.hpp"
 32 #include "gc/shared/collectedHeap.hpp"
 33 #include "gc/shared/tlab_globals.hpp"
 34 #include "interpreter/interpreter.hpp"
 35 #include "oops/arrayOop.hpp"
 36 #include "oops/markWord.hpp"
 37 #include "runtime/basicLock.hpp"
 38 #include "runtime/globals.hpp"
 39 #include "runtime/os.hpp"
 40 #include "runtime/sharedRuntime.hpp"
 41 #include "runtime/stubRoutines.hpp"
 42 #include "utilities/checkedCast.hpp"
 43 #include "utilities/globalDefinitions.hpp"
 44 
 45 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
 46   const int aligned_mask = BytesPerWord -1;
 47   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 48   assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
 49   assert_different_registers(hdr, obj, disp_hdr, tmp);
 50   int null_check_offset = -1;
 51   Label count_locking, done;
 52 
 53   verify_oop(obj);
 54 
 55   // save object being locked into the BasicObjectLock
 56   movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
 57 
 58   null_check_offset = offset();
 59 
 60   if (DiagnoseSyncOnValueBasedClasses != 0) {
 61     load_klass(hdr, obj, rscratch1);
 62     movl(hdr, Address(hdr, Klass::access_flags_offset()));
 63     testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
 64     jcc(Assembler::notZero, slow_case);
 65   }
 66 
 67   if (LockingMode == LM_LIGHTWEIGHT) {
 68 #ifdef _LP64
 69     const Register thread = r15_thread;
 70 #else
 71     const Register thread = disp_hdr;
 72     get_thread(thread);
 73 #endif
 74     lightweight_lock(obj, hdr, thread, tmp, slow_case);
 75   } else  if (LockingMode == LM_LEGACY) {
 76     // Load object header
 77     movptr(hdr, Address(obj, hdr_offset));
 78     // and mark it as unlocked
 79     orptr(hdr, markWord::unlocked_value);
 80     // save unlocked object header into the displaced header location on the stack
 81     movptr(Address(disp_hdr, 0), hdr);
 82     // test if object header is still the same (i.e. unlocked), and if so, store the
 83     // displaced header address in the object header - if it is not the same, get the
 84     // object header instead
 85     MacroAssembler::lock(); // must be immediately before cmpxchg!
 86     cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
 87     // if the object header was the same, we're done
 88     jcc(Assembler::equal, count_locking);
 89     // if the object header was not the same, it is now in the hdr register
 90     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
 91     //
 92     // 1) (hdr & aligned_mask) == 0
 93     // 2) rsp <= hdr
 94     // 3) hdr <= rsp + page_size
 95     //
 96     // these 3 tests can be done by evaluating the following expression:
 97     //
 98     // (hdr - rsp) & (aligned_mask - page_size)
 99     //
100     // assuming both the stack pointer and page_size have their least
101     // significant 2 bits cleared and page_size is a power of 2
102     subptr(hdr, rsp);
103     andptr(hdr, aligned_mask - (int)os::vm_page_size());
104     // for recursive locking, the result is zero => save it in the displaced header
105     // location (null in the displaced hdr location indicates recursive locking)
106     movptr(Address(disp_hdr, 0), hdr);
107     // otherwise we don't care about the result and handle locking via runtime call
108     jcc(Assembler::notZero, slow_case);
109     // done
110   }
111   jmp(done);
112 
113   bind(count_locking);
114   inc_held_monitor_count();
115   bind(done);
116 
117   return null_check_offset;
118 }
119 
120 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
121   const int aligned_mask = BytesPerWord -1;
122   const int hdr_offset = oopDesc::mark_offset_in_bytes();
123   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
124   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
125   Label done;
126 
127   if (LockingMode != LM_LIGHTWEIGHT) {
128     // load displaced header
129     movptr(hdr, Address(disp_hdr, 0));
130     // if the loaded hdr is null we had recursive locking
131     testptr(hdr, hdr);
132     // if we had recursive locking, we are done
133     jcc(Assembler::zero, done);
134   }
135 
136   // load object
137   movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
138   verify_oop(obj);
139 
140   if (LockingMode == LM_LIGHTWEIGHT) {
141 #ifdef _LP64
142     lightweight_unlock(obj, disp_hdr, r15_thread, hdr, slow_case);
143 #else
144     // This relies on the implementation of lightweight_unlock being able to handle
145     // that the reg_rax and thread Register parameters may alias each other.
146     get_thread(disp_hdr);
147     lightweight_unlock(obj, disp_hdr, disp_hdr, hdr, slow_case);
148 #endif
149   } else if (LockingMode == LM_LEGACY) {
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     MacroAssembler::lock(); // must be immediately before cmpxchg!
154     cmpxchgptr(hdr, Address(obj, hdr_offset));
155     // if the object header was not pointing to the displaced header,
156     // we do unlocking via runtime call
157     jcc(Assembler::notEqual, slow_case);
158     dec_held_monitor_count();
159     // done
160   }
161   bind(done);
162 }
163 
164 
165 // Defines obj, preserves var_size_in_bytes
166 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
167   if (UseTLAB) {
168     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
169   } else {
170     jmp(slow_case);
171   }
172 }
173 
174 
175 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
176   assert_different_registers(obj, klass, len);
177   movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
178 #ifdef _LP64
179   if (UseCompressedClassPointers) { // Take care not to kill klass
180     movptr(t1, klass);
181     encode_klass_not_null(t1, rscratch1);
182     movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
183   } else
184 #endif
185   {
186     movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
187   }
188 
189   if (len->is_valid()) {
190     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
191   }
192 #ifdef _LP64
193   else if (UseCompressedClassPointers) {
194     xorptr(t1, t1);
195     store_klass_gap(obj, t1);
196   }
197 #endif
198 }
199 
200 
201 // preserves obj, destroys len_in_bytes
202 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
203   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
204   Label done;
205 
206   // len_in_bytes is positive and ptr sized
207   subptr(len_in_bytes, hdr_size_in_bytes);
208   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
209   bind(done);
210 }
211 
212 
213 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
214   assert(obj == rax, "obj must be in rax, for cmpxchg");
215   assert_different_registers(obj, t1, t2); // XXX really?
216   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
217 
218   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
219 
220   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
221 }
222 
223 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) {
224   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
225          "con_size_in_bytes is not multiple of alignment");
226   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
227 
228   initialize_header(obj, klass, noreg, t1, t2);
229 
230   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
231     // clear rest of allocated space
232     const Register t1_zero = t1;
233     const Register index = t2;
234     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
235     if (var_size_in_bytes != noreg) {
236       mov(index, var_size_in_bytes);
237       initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
238     } else if (con_size_in_bytes <= threshold) {
239       // use explicit null stores
240       // code size = 2 + 3*n bytes (n = number of fields to clear)
241       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
242       for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
243         movptr(Address(obj, i), t1_zero);
244     } else if (con_size_in_bytes > hdr_size_in_bytes) {
245       // use loop to null out the fields
246       // code size = 16 bytes for even n (n = number of fields to clear)
247       // initialize last object field first if odd number of fields
248       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
249       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
250       // initialize last object field if constant size is odd
251       if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
252         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
253       // initialize remaining object fields: rdx is a multiple of 2
254       { Label loop;
255         bind(loop);
256         movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
257                t1_zero);
258         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
259                t1_zero);)
260         decrement(index);
261         jcc(Assembler::notZero, loop);
262       }
263     }
264   }
265 
266   if (CURRENT_ENV->dtrace_alloc_probes()) {
267     assert(obj == rax, "must be");
268     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
269   }
270 
271   verify_oop(obj);
272 }
273 
274 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
275   assert(obj == rax, "obj must be in rax, for cmpxchg");
276   assert_different_registers(obj, len, t1, t2, klass);
277 
278   // determine alignment mask
279   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
280 
281   // check for negative or excessive length
282   cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
283   jcc(Assembler::above, slow_case);
284 
285   const Register arr_size = t2; // okay to be the same
286   // align object end
287   movptr(arr_size, header_size * BytesPerWord + MinObjAlignmentInBytesMask);
288   lea(arr_size, Address(arr_size, len, f));
289   andptr(arr_size, ~MinObjAlignmentInBytesMask);
290 
291   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
292 
293   initialize_header(obj, klass, len, t1, t2);
294 
295   // clear rest of allocated space
296   const Register len_zero = len;
297   initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
298 
299   if (CURRENT_ENV->dtrace_alloc_probes()) {
300     assert(obj == rax, "must be");
301     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
302   }
303 
304   verify_oop(obj);
305 }
306 
307 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
308   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
309   // Make sure there is enough stack space for this method's activation.
310   // Note that we do this before doing an enter(). This matches the
311   // ordering of C2's stack overflow check / rsp decrement and allows
312   // the SharedRuntime stack overflow handling to be consistent
313   // between the two compilers.
314   generate_stack_overflow_check(bang_size_in_bytes);
315 
316   push(rbp);
317   if (PreserveFramePointer) {
318     mov(rbp, rsp);
319   }
320 #if !defined(_LP64) && defined(COMPILER2)
321   if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
322     // c2 leaves fpu stack dirty. Clean it on entry
323     empty_FPU_stack();
324   }
325 #endif // !_LP64 && COMPILER2
326   decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
327 
328   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
329   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
330   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
331 }
332 
333 
334 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
335   increment(rsp, frame_size_in_bytes);  // Does not emit code for frame_size == 0
336   pop(rbp);
337 }
338 
339 
340 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
341   if (breakAtEntry || VerifyFPU) {
342     // Verified Entry first instruction should be 5 bytes long for correct
343     // patching by patch_verified_entry().
344     //
345     // Breakpoint and VerifyFPU have one byte first instruction.
346     // Also first instruction will be one byte "push(rbp)" if stack banging
347     // code is not generated (see build_frame() above).
348     // For all these cases generate long instruction first.
349     fat_nop();
350   }
351   if (breakAtEntry) int3();
352   // build frame
353   IA32_ONLY( verify_FPU(0, "method_entry"); )
354 }
355 
356 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
357   // rbp, + 0: link
358   //     + 1: return address
359   //     + 2: argument with offset 0
360   //     + 3: argument with offset 1
361   //     + 4: ...
362 
363   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
364 }
365 
366 #ifndef PRODUCT
367 
368 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
369   if (!VerifyOops) return;
370   verify_oop_addr(Address(rsp, stack_offset));
371 }
372 
373 void C1_MacroAssembler::verify_not_null_oop(Register r) {
374   if (!VerifyOops) return;
375   Label not_null;
376   testptr(r, r);
377   jcc(Assembler::notZero, not_null);
378   stop("non-null oop required");
379   bind(not_null);
380   verify_oop(r);
381 }
382 
383 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
384 #ifdef ASSERT
385   if (inv_rax) movptr(rax, 0xDEAD);
386   if (inv_rbx) movptr(rbx, 0xDEAD);
387   if (inv_rcx) movptr(rcx, 0xDEAD);
388   if (inv_rdx) movptr(rdx, 0xDEAD);
389   if (inv_rsi) movptr(rsi, 0xDEAD);
390   if (inv_rdi) movptr(rdi, 0xDEAD);
391 #endif
392 }
393 
394 #endif // ifndef PRODUCT