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 "compiler/compilerDefinitions.inline.hpp"
 29 #include "gc/shared/barrierSet.hpp"
 30 #include "gc/shared/barrierSetAssembler.hpp"
 31 #include "gc/shared/collectedHeap.hpp"
 32 #include "gc/shared/tlab_globals.hpp"
 33 #include "interpreter/interpreter.hpp"
 34 #include "oops/arrayOop.hpp"
 35 #include "oops/markWord.hpp"
 36 #include "runtime/basicLock.hpp"
 37 #include "runtime/globals.hpp"
 38 #include "runtime/os.hpp"
 39 #include "runtime/sharedRuntime.hpp"
 40 #include "runtime/stubRoutines.hpp"
 41 #include "utilities/globalDefinitions.hpp"
 42 
 43 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
 44   const int aligned_mask = BytesPerWord -1;
 45   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 46   assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
 47   assert_different_registers(hdr, obj, disp_hdr, tmp);
 48   int null_check_offset = -1;
 49 
 50   verify_oop(obj);
 51 
 52   // save object being locked into the BasicObjectLock
 53   movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
 54 
 55   null_check_offset = offset();
 56 
 57   if (DiagnoseSyncOnValueBasedClasses != 0) {
 58     load_klass(hdr, obj, rscratch1);
 59     movl(hdr, Address(hdr, Klass::access_flags_offset()));
 60     testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
 61     jcc(Assembler::notZero, slow_case);
 62   }
 63 



 64   if (LockingMode == LM_LIGHTWEIGHT) {
 65 #ifdef _LP64
 66     const Register thread = r15_thread;
 67 #else
 68     const Register thread = disp_hdr;
 69     get_thread(thread);
 70 #endif
 71     lightweight_lock(obj, hdr, thread, tmp, slow_case);
 72   } else  if (LockingMode == LM_LEGACY) {
 73     Label done;
 74     // Load object header
 75     movptr(hdr, Address(obj, hdr_offset));
 76     // and mark it as unlocked
 77     orptr(hdr, markWord::unlocked_value);
 78     // save unlocked object header into the displaced header location on the stack
 79     movptr(Address(disp_hdr, 0), hdr);
 80     // test if object header is still the same (i.e. unlocked), and if so, store the
 81     // displaced header address in the object header - if it is not the same, get the
 82     // object header instead
 83     MacroAssembler::lock(); // must be immediately before cmpxchg!
 84     cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
 85     // if the object header was the same, we're done
 86     jcc(Assembler::equal, done);
 87     // if the object header was not the same, it is now in the hdr register
 88     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
 89     //
 90     // 1) (hdr & aligned_mask) == 0
 91     // 2) rsp <= hdr
 92     // 3) hdr <= rsp + page_size
 93     //
 94     // these 3 tests can be done by evaluating the following expression:
 95     //
 96     // (hdr - rsp) & (aligned_mask - page_size)
 97     //
 98     // assuming both the stack pointer and page_size have their least
 99     // significant 2 bits cleared and page_size is a power of 2
100     subptr(hdr, rsp);
101     andptr(hdr, aligned_mask - (int)os::vm_page_size());
102     // for recursive locking, the result is zero => save it in the displaced header
103     // location (null in the displaced hdr location indicates recursive locking)
104     movptr(Address(disp_hdr, 0), hdr);
105     // otherwise we don't care about the result and handle locking via runtime call
106     jcc(Assembler::notZero, slow_case);
107     // done
108     bind(done);
109   }
110 
111   inc_held_monitor_count();
112 
113   return null_check_offset;
114 }
115 
116 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
117   const int aligned_mask = BytesPerWord -1;
118   const int hdr_offset = oopDesc::mark_offset_in_bytes();
119   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
120   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
121   Label done;
122 
123   if (LockingMode != LM_LIGHTWEIGHT) {
124     // load displaced header
125     movptr(hdr, Address(disp_hdr, 0));
126     // if the loaded hdr is null we had recursive locking
127     testptr(hdr, hdr);
128     // if we had recursive locking, we are done
129     jcc(Assembler::zero, done);
130   }
131 
132   // load object
133   movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
134   verify_oop(obj);
135 
136   if (LockingMode == LM_LIGHTWEIGHT) {
137 #ifdef _LP64
138     lightweight_unlock(obj, disp_hdr, r15_thread, hdr, slow_case);
139 #else
140     // This relies on the implementation of lightweight_unlock being able to handle
141     // that the reg_rax and thread Register parameters may alias each other.
142     get_thread(disp_hdr);
143     lightweight_unlock(obj, disp_hdr, disp_hdr, hdr, slow_case);
144 #endif
145   } else if (LockingMode == LM_LEGACY) {
146     // test if object header is pointing to the displaced header, and if so, restore
147     // the displaced header in the object - if the object header is not pointing to
148     // the displaced header, get the object header instead
149     MacroAssembler::lock(); // must be immediately before cmpxchg!
150     cmpxchgptr(hdr, Address(obj, hdr_offset));
151     // if the object header was not pointing to the displaced header,
152     // we do unlocking via runtime call
153     jcc(Assembler::notEqual, slow_case);
154     // done
155   }
156   bind(done);
157   dec_held_monitor_count();
158 }
159 
160 
161 // Defines obj, preserves var_size_in_bytes
162 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
163   if (UseTLAB) {
164     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
165   } else {
166     jmp(slow_case);
167   }
168 }
169 
170 
171 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
172   assert_different_registers(obj, klass, len, t1, t2);

173 #ifdef _LP64
174   if (UseCompactObjectHeaders) {
175     movptr(t1, Address(klass, Klass::prototype_header_offset()));
176     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
177   } else if (UseCompressedClassPointers) { // Take care not to kill klass
178     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
179     movptr(t1, klass);
180     encode_klass_not_null(t1, rscratch1);
181     movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
182   } else
183 #endif
184   {
185     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
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 && !UseCompactObjectHeaders) {
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   assert(!UseCompactObjectHeaders || hdr_size_in_bytes == 8, "check object headers size");
228 
229   initialize_header(obj, klass, noreg, t1, t2);
230 
231   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
232     // clear rest of allocated space
233     const Register t1_zero = t1;
234     const Register index = t2;
235     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
236     if (var_size_in_bytes != noreg) {
237       mov(index, var_size_in_bytes);
238       initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
239     } else if (con_size_in_bytes <= threshold) {
240       // use explicit null stores
241       // code size = 2 + 3*n bytes (n = number of fields to clear)
242       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
243       for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
244         movptr(Address(obj, i), t1_zero);
245     } else if (con_size_in_bytes > hdr_size_in_bytes) {
246       // use loop to null out the fields
247       // code size = 16 bytes for even n (n = number of fields to clear)
248       // initialize last object field first if odd number of fields
249       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
250       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
251       // initialize last object field if constant size is odd
252       if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
253         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
254       // initialize remaining object fields: rdx is a multiple of 2
255       { Label loop;
256         bind(loop);
257         movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
258                t1_zero);
259         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
260                t1_zero);)
261         decrement(index);
262         jcc(Assembler::notZero, loop);
263       }
264     }
265   }
266 
267   if (CURRENT_ENV->dtrace_alloc_probes()) {
268     assert(obj == rax, "must be");
269     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
270   }
271 
272   verify_oop(obj);
273 }
274 
275 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) {
276   assert(obj == rax, "obj must be in rax, for cmpxchg");
277   assert_different_registers(obj, len, t1, t2, klass);
278 
279   // determine alignment mask
280   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
281 
282   // check for negative or excessive length
283   cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
284   jcc(Assembler::above, slow_case);
285 
286   const Register arr_size = t2; // okay to be the same
287   // align object end
288   movptr(arr_size, base_offset_in_bytes + MinObjAlignmentInBytesMask);
289   lea(arr_size, Address(arr_size, len, f));
290   andptr(arr_size, ~MinObjAlignmentInBytesMask);
291 
292   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
293 
294   initialize_header(obj, klass, len, t1, t2);
295 
296   // Clear leading 4 bytes, if necessary.
297   // TODO: This could perhaps go into initialize_body() and also clear the leading 4 bytes
298   // for non-array objects, thereby replacing the klass-gap clearing code in initialize_header().
299   int base_offset = base_offset_in_bytes;
300 #ifdef _LP64
301   if (!is_aligned(base_offset, BytesPerWord)) {
302     assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
303     movl(Address(obj, base_offset), 0);
304     base_offset += BytesPerInt;
305   }
306 #endif
307   assert(is_aligned(base_offset, BytesPerWord), "must be word aligned");
308 
309   // clear rest of allocated space
310   const Register len_zero = len;
311   initialize_body(obj, arr_size, base_offset, len_zero);
312 
313   if (CURRENT_ENV->dtrace_alloc_probes()) {
314     assert(obj == rax, "must be");
315     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
316   }
317 
318   verify_oop(obj);
319 }
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. This is checked in Universe::genesis().

327   int start_offset = offset();
328 
329   if (UseCompressedClassPointers) {
330     load_klass(rscratch1, receiver, rscratch2);
331     cmpptr(rscratch1, iCache);
332   } else {
333     cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
334   }
335   // if icache check fails, then jump to runtime routine
336   // Note: RECEIVER must still contain the receiver!
337   jump_cc(Assembler::notEqual,
338           RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
339   const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
340   assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
341 }
342 
343 
344 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
345   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
346   // Make sure there is enough stack space for this method's activation.
347   // Note that we do this before doing an enter(). This matches the
348   // ordering of C2's stack overflow check / rsp decrement and allows
349   // the SharedRuntime stack overflow handling to be consistent
350   // between the two compilers.
351   generate_stack_overflow_check(bang_size_in_bytes);
352 
353   push(rbp);
354   if (PreserveFramePointer) {
355     mov(rbp, rsp);
356   }
357 #if !defined(_LP64) && defined(COMPILER2)
358   if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
359     // c2 leaves fpu stack dirty. Clean it on entry
360     empty_FPU_stack();
361   }
362 #endif // !_LP64 && COMPILER2
363   decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
364 
365   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
366   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
367   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
368 }
369 
370 
371 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
372   increment(rsp, frame_size_in_bytes);  // Does not emit code for frame_size == 0
373   pop(rbp);
374 }
375 
376 
377 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
378   if (breakAtEntry || VerifyFPU) {
379     // Verified Entry first instruction should be 5 bytes long for correct
380     // patching by patch_verified_entry().
381     //
382     // Breakpoint and VerifyFPU have one byte first instruction.
383     // Also first instruction will be one byte "push(rbp)" if stack banging
384     // code is not generated (see build_frame() above).
385     // For all these cases generate long instruction first.
386     fat_nop();
387   }
388   if (breakAtEntry) int3();
389   // build frame
390   IA32_ONLY( verify_FPU(0, "method_entry"); )
391 }
392 
393 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
394   // rbp, + 0: link
395   //     + 1: return address
396   //     + 2: argument with offset 0
397   //     + 3: argument with offset 1
398   //     + 4: ...
399 
400   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
401 }
402 
403 #ifndef PRODUCT
404 
405 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
406   if (!VerifyOops) return;
407   verify_oop_addr(Address(rsp, stack_offset));
408 }
409 
410 void C1_MacroAssembler::verify_not_null_oop(Register r) {
411   if (!VerifyOops) return;
412   Label not_null;
413   testptr(r, r);
414   jcc(Assembler::notZero, not_null);
415   stop("non-null oop required");
416   bind(not_null);
417   verify_oop(r);
418 }
419 
420 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
421 #ifdef ASSERT
422   if (inv_rax) movptr(rax, 0xDEAD);
423   if (inv_rbx) movptr(rbx, 0xDEAD);
424   if (inv_rcx) movptr(rcx, 0xDEAD);
425   if (inv_rdx) movptr(rdx, 0xDEAD);
426   if (inv_rsi) movptr(rsi, 0xDEAD);
427   if (inv_rdi) movptr(rdi, 0xDEAD);
428 #endif
429 }
430 
431 #endif // ifndef PRODUCT
--- EOF ---