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/biasedLocking.hpp"
 37 #include "runtime/os.hpp"
 38 #include "runtime/sharedRuntime.hpp"
 39 #include "runtime/stubRoutines.hpp"
 40 
 41 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
 42   const Register rklass_decode_tmp = LP64_ONLY(rscratch1) NOT_LP64(noreg);
 43   const int aligned_mask = BytesPerWord -1;
 44   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 45   assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
 46   assert_different_registers(hdr, obj, disp_hdr, scratch);
 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   if (LockingMode == LM_LIGHTWEIGHT) {
 64 #ifdef _LP64
 65     const Register thread = r15_thread;
 66 #else
 67     const Register thread = disp_hdr;
 68     get_thread(thread);
 69 #endif
 70     // Load object header
 71     movptr(hdr, Address(obj, hdr_offset));
 72     lightweight_lock(obj, hdr, thread, scratch, slow_case);
 73   } else {
 74     Label done;
 75 
 76     if (UseBiasedLocking) {
 77       assert(scratch != noreg, "should have scratch register at this point");
 78       biased_locking_enter(disp_hdr, obj, hdr, scratch, rklass_decode_tmp, false, done, &slow_case);
 79     }
 80 
 81     // Load object header
 82     movptr(hdr, Address(obj, hdr_offset));
 83     // and mark it as unlocked
 84     orptr(hdr, markWord::unlocked_value);
 85     // save unlocked object header into the displaced header location on the stack
 86     movptr(Address(disp_hdr, 0), hdr);
 87     // test if object header is still the same (i.e. unlocked), and if so, store the
 88     // displaced header address in the object header - if it is not the same, get the
 89     // object header instead
 90     MacroAssembler::lock(); // must be immediately before cmpxchg!
 91     cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
 92     // if the object header was the same, we're done
 93     if (PrintBiasedLockingStatistics) {
 94       cond_inc32(Assembler::equal,
 95                  ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
 96     }
 97     jcc(Assembler::equal, done);
 98     // if the object header was not the same, it is now in the hdr register
 99     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
100     //
101     // 1) (hdr & aligned_mask) == 0
102     // 2) rsp <= hdr
103     // 3) hdr <= rsp + page_size
104     //
105     // these 3 tests can be done by evaluating the following expression:
106     //
107     // (hdr - rsp) & (aligned_mask - page_size)
108     //
109     // assuming both the stack pointer and page_size have their least
110     // significant 2 bits cleared and page_size is a power of 2
111     subptr(hdr, rsp);
112     andptr(hdr, aligned_mask - os::vm_page_size());
113     // for recursive locking, the result is zero => save it in the displaced header
114     // location (NULL in the displaced hdr location indicates recursive locking)
115     movptr(Address(disp_hdr, 0), hdr);
116     // otherwise we don't care about the result and handle locking via runtime call
117     jcc(Assembler::notZero, slow_case);
118     // done
119     bind(done);
120   }
121   return null_check_offset;
122 }
123 
124 
125 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
126   const int aligned_mask = BytesPerWord -1;
127   const int hdr_offset = oopDesc::mark_offset_in_bytes();
128   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
129   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
130 
131   if (LockingMode == LM_LIGHTWEIGHT) {
132     // load object
133     movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
134     verify_oop(obj);
135     movptr(disp_hdr, Address(obj, hdr_offset));
136     andptr(disp_hdr, ~(int32_t)markWord::lock_mask_in_place);
137     lightweight_unlock(obj, disp_hdr, hdr, slow_case);
138   } else {
139     Label done;
140 
141     if (UseBiasedLocking) {
142       // load object
143       movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
144       biased_locking_exit(obj, hdr, done);
145     }
146 
147     // load displaced header
148     movptr(hdr, Address(disp_hdr, 0));
149     // if the loaded hdr is NULL we had recursive locking
150     testptr(hdr, hdr);
151     // if we had recursive locking, we are done
152     jcc(Assembler::zero, done);
153     if (!UseBiasedLocking) {
154       // load object
155       movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
156     }
157     verify_oop(obj);
158     // test if object header is pointing to the displaced header, and if so, restore
159     // the displaced header in the object - if the object header is not pointing to
160     // the displaced header, get the object header instead
161     MacroAssembler::lock(); // must be immediately before cmpxchg!
162     cmpxchgptr(hdr, Address(obj, hdr_offset));
163     // if the object header was not pointing to the displaced header,
164     // we do unlocking via runtime call
165     jcc(Assembler::notEqual, slow_case);
166     // done
167     bind(done);
168   }
169 }
170 
171 
172 // Defines obj, preserves var_size_in_bytes
173 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
174   if (UseTLAB) {
175     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
176   } else {
177     eden_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
178   }
179 }
180 
181 
182 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
183   assert_different_registers(obj, klass, len, t1, t2);
184   if (UseCompactObjectHeaders || (UseBiasedLocking && !len->is_valid())) {
185     movptr(t1, Address(klass, Klass::prototype_header_offset()));
186     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
187   } else {
188     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
189   }
190 
191   if (!UseCompactObjectHeaders) {
192 #ifdef _LP64
193     if (UseCompressedClassPointers) { // Take care not to kill klass
194       movptr(t1, klass);
195       encode_klass_not_null(t1, rscratch1);
196       movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
197     } else
198 #endif
199     {
200       movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
201     }
202   }
203   if (len->is_valid()) {
204     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
205   }
206 #ifdef _LP64
207   else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
208     xorptr(t1, t1);
209     store_klass_gap(obj, t1);
210   }
211 #endif
212 }
213 
214 
215 // preserves obj, destroys len_in_bytes
216 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
217   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
218   Label done;
219 
220   // len_in_bytes is positive and ptr sized
221   subptr(len_in_bytes, hdr_size_in_bytes);
222   jcc(Assembler::zero, done);
223   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
224   bind(done);
225 }
226 
227 
228 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
229   assert(obj == rax, "obj must be in rax, for cmpxchg");
230   assert_different_registers(obj, t1, t2); // XXX really?
231   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
232 
233   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
234 
235   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
236 }
237 
238 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) {
239   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
240          "con_size_in_bytes is not multiple of alignment");
241   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
242 
243   initialize_header(obj, klass, noreg, t1, t2);
244 
245   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
246     // clear rest of allocated space
247     const Register t1_zero = t1;
248     const Register index = t2;
249     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
250     int hdr_size_aligned = align_up(hdr_size_in_bytes, BytesPerWord); // klass gap is already cleared by init_header().
251     if (var_size_in_bytes != noreg) {
252       mov(index, var_size_in_bytes);
253       initialize_body(obj, index, hdr_size_aligned, t1_zero);
254     } else if (con_size_in_bytes <= threshold) {
255       // use explicit null stores
256       // code size = 2 + 3*n bytes (n = number of fields to clear)
257       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
258       for (int i = hdr_size_aligned; i < con_size_in_bytes; i += BytesPerWord)
259         movptr(Address(obj, i), t1_zero);
260     } else if (con_size_in_bytes > hdr_size_aligned) {
261       // use loop to null out the fields
262       // code size = 16 bytes for even n (n = number of fields to clear)
263       // initialize last object field first if odd number of fields
264       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
265       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
266       // initialize last object field if constant size is odd
267       if (((con_size_in_bytes - hdr_size_aligned) & 4) != 0)
268         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
269       // initialize remaining object fields: rdx is a multiple of 2
270       { Label loop;
271         bind(loop);
272         movptr(Address(obj, index, Address::times_8, hdr_size_aligned - (1*BytesPerWord)),
273                t1_zero);
274         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_aligned - (2*BytesPerWord)),
275                t1_zero);)
276         decrement(index);
277         jcc(Assembler::notZero, loop);
278       }
279     }
280   }
281 
282   if (CURRENT_ENV->dtrace_alloc_probes()) {
283     assert(obj == rax, "must be");
284     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
285   }
286 
287   verify_oop(obj);
288 }
289 
290 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) {
291   assert(obj == rax, "obj must be in rax, for cmpxchg");
292   assert_different_registers(obj, len, t1, t2, klass);
293 
294   // determine alignment mask
295   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
296 
297   // check for negative or excessive length
298   cmpptr(len, (int32_t)max_array_allocation_length);
299   jcc(Assembler::above, slow_case);
300 
301   const Register arr_size = t2; // okay to be the same
302   // align object end
303   movptr(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
304   lea(arr_size, Address(arr_size, len, f));
305   andptr(arr_size, ~MinObjAlignmentInBytesMask);
306 
307   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
308 
309   initialize_header(obj, klass, len, t1, t2);
310 
311   // clear rest of allocated space
312   const Register len_zero = len;
313   initialize_body(obj, arr_size, base_offset_in_bytes, 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 
324 
325 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
326   verify_oop(receiver);
327   // explicit NULL check not needed since load from [klass_offset] causes a trap
328   // check against inline cache
329   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
330   int start_offset = offset();
331   Register tmp_load_klass = LP64_ONLY(rscratch2) NOT_LP64(noreg);
332 
333   if (UseCompressedClassPointers) {
334     load_klass(rscratch1, receiver, tmp_load_klass);
335     cmpptr(rscratch1, iCache);
336   } else {
337     cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
338   }
339   // if icache check fails, then jump to runtime routine
340   // Note: RECEIVER must still contain the receiver!
341   jump_cc(Assembler::notEqual,
342           RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
343   const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
344   assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
345 }
346 
347 
348 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
349   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
350   // Make sure there is enough stack space for this method's activation.
351   // Note that we do this before doing an enter(). This matches the
352   // ordering of C2's stack overflow check / rsp decrement and allows
353   // the SharedRuntime stack overflow handling to be consistent
354   // between the two compilers.
355   generate_stack_overflow_check(bang_size_in_bytes);
356 
357   push(rbp);
358   if (PreserveFramePointer) {
359     mov(rbp, rsp);
360   }
361 #if !defined(_LP64) && defined(COMPILER2)
362   if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
363     // c2 leaves fpu stack dirty. Clean it on entry
364     empty_FPU_stack();
365   }
366 #endif // !_LP64 && COMPILER2
367   decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
368 
369   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
370   bs->nmethod_entry_barrier(this);
371 }
372 
373 
374 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
375   increment(rsp, frame_size_in_bytes);  // Does not emit code for frame_size == 0
376   pop(rbp);
377 }
378 
379 
380 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
381   if (breakAtEntry || VerifyFPU) {
382     // Verified Entry first instruction should be 5 bytes long for correct
383     // patching by patch_verified_entry().
384     //
385     // Breakpoint and VerifyFPU have one byte first instruction.
386     // Also first instruction will be one byte "push(rbp)" if stack banging
387     // code is not generated (see build_frame() above).
388     // For all these cases generate long instruction first.
389     fat_nop();
390   }
391   if (breakAtEntry) int3();
392   // build frame
393   IA32_ONLY( verify_FPU(0, "method_entry"); )
394 }
395 
396 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
397   // rbp, + 0: link
398   //     + 1: return address
399   //     + 2: argument with offset 0
400   //     + 3: argument with offset 1
401   //     + 4: ...
402 
403   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
404 }
405 
406 #ifndef PRODUCT
407 
408 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
409   if (!VerifyOops) return;
410   verify_oop_addr(Address(rsp, stack_offset));
411 }
412 
413 void C1_MacroAssembler::verify_not_null_oop(Register r) {
414   if (!VerifyOops) return;
415   Label not_null;
416   testptr(r, r);
417   jcc(Assembler::notZero, not_null);
418   stop("non-null oop required");
419   bind(not_null);
420   verify_oop(r);
421 }
422 
423 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
424 #ifdef ASSERT
425   if (inv_rax) movptr(rax, 0xDEAD);
426   if (inv_rbx) movptr(rbx, 0xDEAD);
427   if (inv_rcx) movptr(rcx, 0xDEAD);
428   if (inv_rdx) movptr(rdx, 0xDEAD);
429   if (inv_rsi) movptr(rsi, 0xDEAD);
430   if (inv_rdi) movptr(rdi, 0xDEAD);
431 #endif
432 }
433 
434 #endif // ifndef PRODUCT