1 /*
  2  * Copyright (c) 1999, 2021, 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 "gc/shared/barrierSet.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 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
 41   const Register rklass_decode_tmp = LP64_ONLY(rscratch1) NOT_LP64(noreg);
 42   const int aligned_mask = BytesPerWord -1;
 43   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 44   assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
 45   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
 46   Label done;
 47   int null_check_offset = -1;
 48 
 49   verify_oop(obj);
 50 
 51   // save object being locked into the BasicObjectLock
 52   movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj);
 53 
 54   null_check_offset = offset();
 55 
 56   if (DiagnoseSyncOnValueBasedClasses != 0) {
 57     load_klass(hdr, obj, rklass_decode_tmp);
 58     movl(hdr, Address(hdr, Klass::access_flags_offset()));
 59     testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
 60     jcc(Assembler::notZero, slow_case);
 61   }
 62 
 63   // Load object header
 64   movptr(hdr, Address(obj, hdr_offset));
 65   // and mark it as unlocked
 66   orptr(hdr, markWord::unlocked_value);
 67   // save unlocked object header into the displaced header location on the stack
 68   movptr(Address(disp_hdr, 0), hdr);
 69   // test if object header is still the same (i.e. unlocked), and if so, store the
 70   // displaced header address in the object header - if it is not the same, get the
 71   // object header instead
 72   MacroAssembler::lock(); // must be immediately before cmpxchg!
 73   cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
 74   // if the object header was the same, we're done
 75   jcc(Assembler::equal, done);
 76   // if the object header was not the same, it is now in the hdr register
 77   // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
 78   //
 79   // 1) (hdr & aligned_mask) == 0
 80   // 2) rsp <= hdr
 81   // 3) hdr <= rsp + page_size
 82   //
 83   // these 3 tests can be done by evaluating the following expression:
 84   //
 85   // (hdr - rsp) & (aligned_mask - page_size)
 86   //
 87   // assuming both the stack pointer and page_size have their least
 88   // significant 2 bits cleared and page_size is a power of 2
 89   subptr(hdr, rsp);
 90   andptr(hdr, aligned_mask - os::vm_page_size());
 91   // for recursive locking, the result is zero => save it in the displaced header
 92   // location (NULL in the displaced hdr location indicates recursive locking)
 93   movptr(Address(disp_hdr, 0), hdr);
 94   // otherwise we don't care about the result and handle locking via runtime call
 95   jcc(Assembler::notZero, slow_case);
 96   // done
 97   bind(done);
 98 
 99   inc_held_monitor_count();
100 
101   return null_check_offset;
102 }
103 
104 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
105   const int aligned_mask = BytesPerWord -1;
106   const int hdr_offset = oopDesc::mark_offset_in_bytes();
107   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
108   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
109   Label done;
110 
111   // load displaced header
112   movptr(hdr, Address(disp_hdr, 0));
113   // if the loaded hdr is NULL we had recursive locking
114   testptr(hdr, hdr);
115   // if we had recursive locking, we are done
116   jcc(Assembler::zero, done);
117   // load object
118   movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
119 
120   verify_oop(obj);
121   // test if object header is pointing to the displaced header, and if so, restore
122   // the displaced header in the object - if the object header is not pointing to
123   // the displaced header, get the object header instead
124   MacroAssembler::lock(); // must be immediately before cmpxchg!
125   cmpxchgptr(hdr, Address(obj, hdr_offset));
126   // if the object header was not pointing to the displaced header,
127   // we do unlocking via runtime call
128   jcc(Assembler::notEqual, slow_case);
129   // done
130   bind(done);
131 
132   dec_held_monitor_count();
133 }
134 
135 
136 // Defines obj, preserves var_size_in_bytes
137 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
138   if (UseTLAB) {
139     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
140   } else {
141     jmp(slow_case);
142   }
143 }
144 
145 
146 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
147   Register tmp_encode_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
148   assert_different_registers(obj, klass, len, t1, t2);
149   movptr(t1, Address(klass, Klass::prototype_header_offset()));
150   movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
151 #ifndef _LP64
152   movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
153 #endif
154 
155   if (len->is_valid()) {
156     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
157   }
158 }
159 
160 
161 // preserves obj, destroys len_in_bytes
162 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
163   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
164   Label done;
165 
166   // len_in_bytes is positive and ptr sized
167   subptr(len_in_bytes, hdr_size_in_bytes);
168   jcc(Assembler::zero, done);
169   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
170   bind(done);
171 }
172 
173 
174 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
175   assert(obj == rax, "obj must be in rax, for cmpxchg");
176   assert_different_registers(obj, t1, t2); // XXX really?
177   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
178 
179   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
180 
181   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
182 }
183 
184 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) {
185   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
186          "con_size_in_bytes is not multiple of alignment");
187   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
188 
189   initialize_header(obj, klass, noreg, t1, t2);
190 
191   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
192     // clear rest of allocated space
193     const Register t1_zero = t1;
194     const Register index = t2;
195     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
196     if (var_size_in_bytes != noreg) {
197       mov(index, var_size_in_bytes);
198       initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
199     } else if (con_size_in_bytes <= threshold) {
200       // use explicit null stores
201       // code size = 2 + 3*n bytes (n = number of fields to clear)
202       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
203       for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
204         movptr(Address(obj, i), t1_zero);
205     } else if (con_size_in_bytes > hdr_size_in_bytes) {
206       // use loop to null out the fields
207       // code size = 16 bytes for even n (n = number of fields to clear)
208       // initialize last object field first if odd number of fields
209       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
210       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
211       // initialize last object field if constant size is odd
212       if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
213         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
214       // initialize remaining object fields: rdx is a multiple of 2
215       { Label loop;
216         bind(loop);
217         movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
218                t1_zero);
219         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
220                t1_zero);)
221         decrement(index);
222         jcc(Assembler::notZero, loop);
223       }
224     }
225   }
226 
227   if (CURRENT_ENV->dtrace_alloc_probes()) {
228     assert(obj == rax, "must be");
229     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
230   }
231 
232   verify_oop(obj);
233 }
234 
235 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) {
236   assert(obj == rax, "obj must be in rax, for cmpxchg");
237   assert_different_registers(obj, len, t1, t2, klass);
238 
239   // determine alignment mask
240   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
241 
242   // check for negative or excessive length
243   cmpptr(len, (int32_t)max_array_allocation_length);
244   jcc(Assembler::above, slow_case);
245 
246   const Register arr_size = t2; // okay to be the same
247   // align object end
248   movptr(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
249   lea(arr_size, Address(arr_size, len, f));
250   andptr(arr_size, ~MinObjAlignmentInBytesMask);
251 
252   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
253 
254   initialize_header(obj, klass, len, t1, t2);
255 
256   // clear rest of allocated space
257   const Register len_zero = len;
258   initialize_body(obj, arr_size, base_offset_in_bytes, len_zero);
259 
260   if (CURRENT_ENV->dtrace_alloc_probes()) {
261     assert(obj == rax, "must be");
262     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
263   }
264 
265   verify_oop(obj);
266 }
267 
268 
269 
270 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
271   verify_oop(receiver);
272   // explicit NULL check not needed since load from [klass_offset] causes a trap
273   // check against inline cache
274   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
275   int start_offset = offset();
276   Register tmp_load_klass = LP64_ONLY(rscratch2) NOT_LP64(noreg);
277 
278   if (UseCompressedClassPointers) {
279     load_klass(rscratch1, receiver, tmp_load_klass);
280     cmpptr(rscratch1, iCache);
281   } else {
282     cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
283   }
284   // if icache check fails, then jump to runtime routine
285   // Note: RECEIVER must still contain the receiver!
286   jump_cc(Assembler::notEqual,
287           RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
288   const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
289   assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
290 }
291 
292 
293 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
294   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
295   // Make sure there is enough stack space for this method's activation.
296   // Note that we do this before doing an enter(). This matches the
297   // ordering of C2's stack overflow check / rsp decrement and allows
298   // the SharedRuntime stack overflow handling to be consistent
299   // between the two compilers.
300   generate_stack_overflow_check(bang_size_in_bytes);
301 
302   push(rbp);
303   if (PreserveFramePointer) {
304     mov(rbp, rsp);
305   }
306 #if !defined(_LP64) && defined(COMPILER2)
307   if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
308     // c2 leaves fpu stack dirty. Clean it on entry
309     empty_FPU_stack();
310   }
311 #endif // !_LP64 && COMPILER2
312   decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
313 
314   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
315   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
316   bs->nmethod_entry_barrier(this, NULL /* slow_path */, NULL /* continuation */);
317 }
318 
319 
320 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
321   increment(rsp, frame_size_in_bytes);  // Does not emit code for frame_size == 0
322   pop(rbp);
323 }
324 
325 
326 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
327   if (breakAtEntry || VerifyFPU) {
328     // Verified Entry first instruction should be 5 bytes long for correct
329     // patching by patch_verified_entry().
330     //
331     // Breakpoint and VerifyFPU have one byte first instruction.
332     // Also first instruction will be one byte "push(rbp)" if stack banging
333     // code is not generated (see build_frame() above).
334     // For all these cases generate long instruction first.
335     fat_nop();
336   }
337   if (breakAtEntry) int3();
338   // build frame
339   IA32_ONLY( verify_FPU(0, "method_entry"); )
340 }
341 
342 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
343   // rbp, + 0: link
344   //     + 1: return address
345   //     + 2: argument with offset 0
346   //     + 3: argument with offset 1
347   //     + 4: ...
348 
349   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
350 }
351 
352 #ifndef PRODUCT
353 
354 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
355   if (!VerifyOops) return;
356   verify_oop_addr(Address(rsp, stack_offset));
357 }
358 
359 void C1_MacroAssembler::verify_not_null_oop(Register r) {
360   if (!VerifyOops) return;
361   Label not_null;
362   testptr(r, r);
363   jcc(Assembler::notZero, not_null);
364   stop("non-null oop required");
365   bind(not_null);
366   verify_oop(r);
367 }
368 
369 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
370 #ifdef ASSERT
371   if (inv_rax) movptr(rax, 0xDEAD);
372   if (inv_rbx) movptr(rbx, 0xDEAD);
373   if (inv_rcx) movptr(rcx, 0xDEAD);
374   if (inv_rdx) movptr(rdx, 0xDEAD);
375   if (inv_rsi) movptr(rsi, 0xDEAD);
376   if (inv_rdi) movptr(rdi, 0xDEAD);
377 #endif
378 }
379 
380 #endif // ifndef PRODUCT