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