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 (UseFastLocking) {
 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     fast_lock_impl(obj, hdr, thread, scratch, slow_case, LP64_ONLY(false) NOT_LP64(true));
 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 (UseFastLocking) {
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     fast_unlock_impl(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 // Defines obj, preserves var_size_in_bytes
172 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
173   if (UseTLAB) {
174     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
175   } else {
176     eden_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
177   }
178 }
179 
180 
181 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
182   assert_different_registers(obj, klass, len, t1, t2);
183   if (UseCompactObjectHeaders) {
184     movptr(t1, Address(klass, Klass::prototype_header_offset()));
185     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
186   } else {
187     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
188 #ifdef _LP64
189     if (UseCompressedClassPointers) { // Take care not to kill klass
190       movptr(t1, klass);
191       encode_klass_not_null(t1, rscratch1);
192       movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
193     } else
194 #endif
195     {
196       movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
197     }
198   }
199   if (len->is_valid()) {
200     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
201   }
202 #ifdef _LP64
203   else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
204     xorptr(t1, t1);
205     store_klass_gap(obj, t1);
206   }
207 #endif
208 }
209 
210 
211 // preserves obj, destroys len_in_bytes
212 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
213   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
214   Label done;
215 
216   // len_in_bytes is positive and ptr sized
217   subptr(len_in_bytes, hdr_size_in_bytes);
218   jcc(Assembler::zero, done);
219   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
220   bind(done);
221 }
222 
223 
224 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
225   assert(obj == rax, "obj must be in rax, for cmpxchg");
226   assert_different_registers(obj, t1, t2); // XXX really?
227   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
228 
229   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
230 
231   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
232 }
233 
234 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) {
235   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
236          "con_size_in_bytes is not multiple of alignment");
237   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
238 
239   initialize_header(obj, klass, noreg, t1, t2);
240 
241   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
242     // clear rest of allocated space
243     const Register t1_zero = t1;
244     const Register index = t2;
245     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
246     int hdr_size_aligned = align_up(hdr_size_in_bytes, BytesPerWord); // klass gap is already cleared by init_header().
247     if (var_size_in_bytes != noreg) {
248       mov(index, var_size_in_bytes);
249       initialize_body(obj, index, hdr_size_aligned, t1_zero);
250     } else if (con_size_in_bytes <= threshold) {
251       // use explicit null stores
252       // code size = 2 + 3*n bytes (n = number of fields to clear)
253       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
254       for (int i = hdr_size_aligned; i < con_size_in_bytes; i += BytesPerWord)
255         movptr(Address(obj, i), t1_zero);
256     } else if (con_size_in_bytes > hdr_size_aligned) {
257       // use loop to null out the fields
258       // code size = 16 bytes for even n (n = number of fields to clear)
259       // initialize last object field first if odd number of fields
260       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
261       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
262       // initialize last object field if constant size is odd
263       if (((con_size_in_bytes - hdr_size_aligned) & 4) != 0)
264         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
265       // initialize remaining object fields: rdx is a multiple of 2
266       { Label loop;
267         bind(loop);
268         movptr(Address(obj, index, Address::times_8, hdr_size_aligned - (1*BytesPerWord)),
269                t1_zero);
270         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_aligned - (2*BytesPerWord)),
271                t1_zero);)
272         decrement(index);
273         jcc(Assembler::notZero, loop);
274       }
275     }
276   }
277 
278   if (CURRENT_ENV->dtrace_alloc_probes()) {
279     assert(obj == rax, "must be");
280     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
281   }
282 
283   verify_oop(obj);
284 }
285 
286 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) {
287   assert(obj == rax, "obj must be in rax, for cmpxchg");
288   assert_different_registers(obj, len, t1, t2, klass);
289 
290   // determine alignment mask
291   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
292 
293   // check for negative or excessive length
294   cmpptr(len, (int32_t)max_array_allocation_length);
295   jcc(Assembler::above, slow_case);
296 
297   const Register arr_size = t2; // okay to be the same
298   // align object end
299   movptr(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
300   lea(arr_size, Address(arr_size, len, f));
301   andptr(arr_size, ~MinObjAlignmentInBytesMask);
302 
303   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
304 
305   initialize_header(obj, klass, len, t1, t2);
306 
307   // clear rest of allocated space
308   const Register len_zero = len;
309   initialize_body(obj, arr_size, base_offset_in_bytes, len_zero);
310 
311   if (CURRENT_ENV->dtrace_alloc_probes()) {
312     assert(obj == rax, "must be");
313     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
314   }
315 
316   verify_oop(obj);
317 }
318 
319 
320 
321 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
322   verify_oop(receiver);
323   // explicit NULL check not needed since load from [klass_offset] causes a trap
324   // check against inline cache
325   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
326   int start_offset = offset();
327   Register tmp_load_klass = LP64_ONLY(rscratch2) NOT_LP64(noreg);
328 
329   if (UseCompressedClassPointers) {
330     load_klass(rscratch1, receiver, tmp_load_klass);
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, int max_monitors) {
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 #ifdef _LP64
366   if (UseFastLocking && max_monitors > 0) {
367     Label ok;
368     movptr(rax, Address(r15_thread, JavaThread::lock_stack_current_offset()));
369     addptr(rax, max_monitors * wordSize);
370     cmpptr(rax, Address(r15_thread, JavaThread::lock_stack_limit_offset()));
371     jcc(Assembler::less, ok);
372     assert(StubRoutines::x86::check_lock_stack() != NULL, "need runtime call stub");
373     call(RuntimeAddress(StubRoutines::x86::check_lock_stack()));
374     bind(ok);
375   }
376 #endif
377 
378   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
379   bs->nmethod_entry_barrier(this);
380 }
381 
382 
383 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
384   increment(rsp, frame_size_in_bytes);  // Does not emit code for frame_size == 0
385   pop(rbp);
386 }
387 
388 
389 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
390   if (breakAtEntry || VerifyFPU) {
391     // Verified Entry first instruction should be 5 bytes long for correct
392     // patching by patch_verified_entry().
393     //
394     // Breakpoint and VerifyFPU have one byte first instruction.
395     // Also first instruction will be one byte "push(rbp)" if stack banging
396     // code is not generated (see build_frame() above).
397     // For all these cases generate long instruction first.
398     fat_nop();
399   }
400   if (breakAtEntry) int3();
401   // build frame
402   IA32_ONLY( verify_FPU(0, "method_entry"); )
403 }
404 
405 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
406   // rbp, + 0: link
407   //     + 1: return address
408   //     + 2: argument with offset 0
409   //     + 3: argument with offset 1
410   //     + 4: ...
411 
412   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
413 }
414 
415 #ifndef PRODUCT
416 
417 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
418   if (!VerifyOops) return;
419   verify_oop_addr(Address(rsp, stack_offset));
420 }
421 
422 void C1_MacroAssembler::verify_not_null_oop(Register r) {
423   if (!VerifyOops) return;
424   Label not_null;
425   testptr(r, r);
426   jcc(Assembler::notZero, not_null);
427   stop("non-null oop required");
428   bind(not_null);
429   verify_oop(r);
430 }
431 
432 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
433 #ifdef ASSERT
434   if (inv_rax) movptr(rax, 0xDEAD);
435   if (inv_rbx) movptr(rbx, 0xDEAD);
436   if (inv_rcx) movptr(rcx, 0xDEAD);
437   if (inv_rdx) movptr(rdx, 0xDEAD);
438   if (inv_rsi) movptr(rsi, 0xDEAD);
439   if (inv_rdi) movptr(rdi, 0xDEAD);
440 #endif
441 }
442 
443 #endif // ifndef PRODUCT