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 
 52   verify_oop(obj);
 53 
 54   // save object being locked into the BasicObjectLock
 55   movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
 56 
 57   null_check_offset = offset();
 58 
 59   if (DiagnoseSyncOnValueBasedClasses != 0) {
 60     load_klass(hdr, obj, rscratch1);
 61     movl(hdr, Address(hdr, Klass::access_flags_offset()));
 62     testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
 63     jcc(Assembler::notZero, slow_case);
 64   }
 65 
 66   if (LockingMode == LM_LIGHTWEIGHT) {
 67 #ifdef _LP64
 68     const Register thread = r15_thread;
 69 #else
 70     const Register thread = disp_hdr;
 71     get_thread(thread);
 72 #endif
 73     lightweight_lock(obj, hdr, thread, tmp, slow_case);
 74   } else  if (LockingMode == LM_LEGACY) {
 75     Label done;
 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, done);
 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     bind(done);
111   }
112 
113   inc_held_monitor_count();
114 
115   return null_check_offset;
116 }
117 
118 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
119   const int aligned_mask = BytesPerWord -1;
120   const int hdr_offset = oopDesc::mark_offset_in_bytes();
121   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
122   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
123   Label done;
124 
125   if (LockingMode != LM_LIGHTWEIGHT) {
126     // load displaced header
127     movptr(hdr, Address(disp_hdr, 0));
128     // if the loaded hdr is null we had recursive locking
129     testptr(hdr, hdr);
130     // if we had recursive locking, we are done
131     jcc(Assembler::zero, done);
132   }
133 
134   // load object
135   movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
136   verify_oop(obj);
137 
138   if (LockingMode == LM_LIGHTWEIGHT) {
139 #ifdef _LP64
140     lightweight_unlock(obj, disp_hdr, r15_thread, hdr, slow_case);
141 #else
142     // This relies on the implementation of lightweight_unlock being able to handle
143     // that the reg_rax and thread Register parameters may alias each other.
144     get_thread(disp_hdr);
145     lightweight_unlock(obj, disp_hdr, disp_hdr, hdr, slow_case);
146 #endif
147   } else if (LockingMode == LM_LEGACY) {
148     // test if object header is pointing to the displaced header, and if so, restore
149     // the displaced header in the object - if the object header is not pointing to
150     // the displaced header, get the object header instead
151     MacroAssembler::lock(); // must be immediately before cmpxchg!
152     cmpxchgptr(hdr, Address(obj, hdr_offset));
153     // if the object header was not pointing to the displaced header,
154     // we do unlocking via runtime call
155     jcc(Assembler::notEqual, slow_case);
156     // done
157   }
158   bind(done);
159   dec_held_monitor_count();
160 }
161 
162 
163 // Defines obj, preserves var_size_in_bytes
164 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
165   if (UseTLAB) {
166     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
167   } else {
168     jmp(slow_case);
169   }
170 }
171 
172 
173 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
174   assert_different_registers(obj, klass, len, t1, t2);
175 #ifdef _LP64
176   if (UseCompactObjectHeaders) {
177     movptr(t1, Address(klass, Klass::prototype_header_offset()));
178     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
179   } else if (UseCompressedClassPointers) { // Take care not to kill klass
180     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
181     movptr(t1, klass);
182     encode_klass_not_null(t1, rscratch1);
183     movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
184   } else
185 #endif
186   {
187     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
188     movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
189   }
190   if (len->is_valid()) {
191     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
192   }
193 #ifdef _LP64
194   else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
195     xorptr(t1, t1);
196     store_klass_gap(obj, t1);
197   }
198 #endif
199 }
200 
201 
202 // preserves obj, destroys len_in_bytes
203 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
204   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
205   Label done;
206 
207   // len_in_bytes is positive and ptr sized
208   subptr(len_in_bytes, hdr_size_in_bytes);
209   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
210   bind(done);
211 }
212 
213 
214 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
215   assert(obj == rax, "obj must be in rax, for cmpxchg");
216   assert_different_registers(obj, t1, t2); // XXX really?
217   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
218 
219   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
220 
221   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
222 }
223 
224 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) {
225   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
226          "con_size_in_bytes is not multiple of alignment");
227   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
228   if (UseCompactObjectHeaders) {
229     assert(hdr_size_in_bytes == 8, "check object headers size");
230   }
231   initialize_header(obj, klass, noreg, t1, t2);
232 
233   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
234     // clear rest of allocated space
235     const Register t1_zero = t1;
236     const Register index = t2;
237     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
238     if (var_size_in_bytes != noreg) {
239       mov(index, var_size_in_bytes);
240       initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
241     } else if (con_size_in_bytes <= threshold) {
242       // use explicit null stores
243       // code size = 2 + 3*n bytes (n = number of fields to clear)
244       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
245       for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
246         movptr(Address(obj, i), t1_zero);
247     } else if (con_size_in_bytes > hdr_size_in_bytes) {
248       // use loop to null out the fields
249       // code size = 16 bytes for even n (n = number of fields to clear)
250       // initialize last object field first if odd number of fields
251       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
252       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
253       // initialize last object field if constant size is odd
254       if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
255         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
256       // initialize remaining object fields: rdx is a multiple of 2
257       { Label loop;
258         bind(loop);
259         movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
260                t1_zero);
261         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
262                t1_zero);)
263         decrement(index);
264         jcc(Assembler::notZero, loop);
265       }
266     }
267   }
268 
269   if (CURRENT_ENV->dtrace_alloc_probes()) {
270     assert(obj == rax, "must be");
271     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
272   }
273 
274   verify_oop(obj);
275 }
276 
277 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int base_offset_in_bytes, Address::ScaleFactor f, Register klass, Label& slow_case) {
278   assert(obj == rax, "obj must be in rax, for cmpxchg");
279   assert_different_registers(obj, len, t1, t2, klass);
280 
281   // determine alignment mask
282   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
283 
284   // check for negative or excessive length
285   cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
286   jcc(Assembler::above, slow_case);
287 
288   const Register arr_size = t2; // okay to be the same
289   // align object end
290   movptr(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
291   lea(arr_size, Address(arr_size, len, f));
292   andptr(arr_size, ~MinObjAlignmentInBytesMask);
293 
294   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
295 
296   initialize_header(obj, klass, len, t1, t2);
297 
298   // Clear leading 4 bytes, if necessary.
299   // TODO: This could perhaps go into initialize_body() and also clear the leading 4 bytes
300   // for non-array objects, thereby replacing the klass-gap clearing code in initialize_header().
301   int base_offset = base_offset_in_bytes;
302 #ifdef _LP64
303   if (!is_aligned(base_offset, BytesPerWord)) {
304     assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
305     movl(Address(obj, base_offset), 0);
306     base_offset += BytesPerInt;
307   }
308 #endif
309   assert(is_aligned(base_offset, BytesPerWord), "must be word aligned");
310 
311   // clear rest of allocated space
312   const Register len_zero = len;
313   initialize_body(obj, arr_size, base_offset, len_zero);
314 
315   if (CURRENT_ENV->dtrace_alloc_probes()) {
316     assert(obj == rax, "must be");
317     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
318   }
319 
320   verify_oop(obj);
321 }
322 
323 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
324   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
325   // Make sure there is enough stack space for this method's activation.
326   // Note that we do this before doing an enter(). This matches the
327   // ordering of C2's stack overflow check / rsp decrement and allows
328   // the SharedRuntime stack overflow handling to be consistent
329   // between the two compilers.
330   generate_stack_overflow_check(bang_size_in_bytes);
331 
332   push(rbp);
333   if (PreserveFramePointer) {
334     mov(rbp, rsp);
335   }
336 #if !defined(_LP64) && defined(COMPILER2)
337   if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
338     // c2 leaves fpu stack dirty. Clean it on entry
339     empty_FPU_stack();
340   }
341 #endif // !_LP64 && COMPILER2
342   decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
343 
344   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
345   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
346   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
347 }
348 
349 
350 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
351   increment(rsp, frame_size_in_bytes);  // Does not emit code for frame_size == 0
352   pop(rbp);
353 }
354 
355 
356 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
357   if (breakAtEntry || VerifyFPU) {
358     // Verified Entry first instruction should be 5 bytes long for correct
359     // patching by patch_verified_entry().
360     //
361     // Breakpoint and VerifyFPU have one byte first instruction.
362     // Also first instruction will be one byte "push(rbp)" if stack banging
363     // code is not generated (see build_frame() above).
364     // For all these cases generate long instruction first.
365     fat_nop();
366   }
367   if (breakAtEntry) int3();
368   // build frame
369   IA32_ONLY( verify_FPU(0, "method_entry"); )
370 }
371 
372 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
373   // rbp, + 0: link
374   //     + 1: return address
375   //     + 2: argument with offset 0
376   //     + 3: argument with offset 1
377   //     + 4: ...
378 
379   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
380 }
381 
382 #ifndef PRODUCT
383 
384 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
385   if (!VerifyOops) return;
386   verify_oop_addr(Address(rsp, stack_offset));
387 }
388 
389 void C1_MacroAssembler::verify_not_null_oop(Register r) {
390   if (!VerifyOops) return;
391   Label not_null;
392   testptr(r, r);
393   jcc(Assembler::notZero, not_null);
394   stop("non-null oop required");
395   bind(not_null);
396   verify_oop(r);
397 }
398 
399 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
400 #ifdef ASSERT
401   if (inv_rax) movptr(rax, 0xDEAD);
402   if (inv_rbx) movptr(rbx, 0xDEAD);
403   if (inv_rcx) movptr(rcx, 0xDEAD);
404   if (inv_rdx) movptr(rdx, 0xDEAD);
405   if (inv_rsi) movptr(rsi, 0xDEAD);
406   if (inv_rdi) movptr(rdi, 0xDEAD);
407 #endif
408 }
409 
410 #endif // ifndef PRODUCT