1 /*
  2  * Copyright (c) 1999, 2023, 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/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 tmp, Label& slow_case) {
 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_different_registers(hdr, obj, disp_hdr, tmp);
 46   int null_check_offset = -1;
 47 
 48   verify_oop(obj);
 49 
 50   // save object being locked into the BasicObjectLock
 51   movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj);
 52 
 53   null_check_offset = offset();
 54 
 55   if (DiagnoseSyncOnValueBasedClasses != 0) {
 56     load_klass(hdr, obj, rscratch1);
 57     movl(hdr, Address(hdr, Klass::access_flags_offset()));
 58     testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
 59     jcc(Assembler::notZero, slow_case);
 60   }
 61 
 62   // Load object header
 63   movptr(hdr, Address(obj, hdr_offset));
 64 
 65   if (UseFastLocking) {
 66 #ifdef _LP64
 67     const Register thread = r15_thread;
 68 #else
 69     const Register thread = disp_hdr;
 70     get_thread(thread);
 71 #endif
 72     fast_lock_impl(obj, hdr, thread, tmp, slow_case, LP64_ONLY(false) NOT_LP64(true));
 73   } else {
 74     Label done;
 75     orptr(hdr, markWord::unlocked_value);
 76     // save unlocked object header into the displaced header location on the stack
 77     movptr(Address(disp_hdr, 0), hdr);
 78     // test if object header is still the same (i.e. unlocked), and if so, store the
 79     // displaced header address in the object header - if it is not the same, get the
 80     // object header instead
 81     MacroAssembler::lock(); // must be immediately before cmpxchg!
 82     cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
 83     // if the object header was the same, we're done
 84     jcc(Assembler::equal, done);
 85     // if the object header was not the same, it is now in the hdr register
 86     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
 87     //
 88     // 1) (hdr & aligned_mask) == 0
 89     // 2) rsp <= hdr
 90     // 3) hdr <= rsp + page_size
 91     //
 92     // these 3 tests can be done by evaluating the following expression:
 93     //
 94     // (hdr - rsp) & (aligned_mask - page_size)
 95     //
 96     // assuming both the stack pointer and page_size have their least
 97     // significant 2 bits cleared and page_size is a power of 2
 98     subptr(hdr, rsp);
 99     andptr(hdr, aligned_mask - (int)os::vm_page_size());
100     // for recursive locking, the result is zero => save it in the displaced header
101     // location (NULL in the displaced hdr location indicates recursive locking)
102     movptr(Address(disp_hdr, 0), hdr);
103     // otherwise we don't care about the result and handle locking via runtime call
104     jcc(Assembler::notZero, slow_case);
105     // done
106     bind(done);
107   }
108 
109   inc_held_monitor_count();
110 
111   return null_check_offset;
112 }
113 
114 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
115   const int aligned_mask = BytesPerWord -1;
116   const int hdr_offset = oopDesc::mark_offset_in_bytes();
117   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
118   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
119   Label done;
120 
121   if (!UseFastLocking) {
122     // load displaced header
123     movptr(hdr, Address(disp_hdr, 0));
124     // if the loaded hdr is NULL we had recursive locking
125     testptr(hdr, hdr);
126     // if we had recursive locking, we are done
127     jcc(Assembler::zero, done);
128   }
129 
130   // load object
131   movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
132   verify_oop(obj);
133 
134   if (UseFastLocking) {
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     // test if object header is pointing to the displaced header, and if so, restore
140     // the displaced header in the object - if the object header is not pointing to
141     // the displaced header, get the object header instead
142     MacroAssembler::lock(); // must be immediately before cmpxchg!
143     cmpxchgptr(hdr, Address(obj, hdr_offset));
144     // if the object header was not pointing to the displaced header,
145     // we do unlocking via runtime call
146     jcc(Assembler::notEqual, slow_case);
147     // done
148   }
149   bind(done);
150   dec_held_monitor_count();
151 }
152 
153 
154 // Defines obj, preserves var_size_in_bytes
155 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
156   if (UseTLAB) {
157     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
158   } else {
159     jmp(slow_case);
160   }
161 }
162 
163 
164 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
165   assert_different_registers(obj, klass, len, t1, t2);
166   if (UseCompactObjectHeaders) {
167     movptr(t1, Address(klass, Klass::prototype_header_offset()));
168     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
169   } else {
170     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
171 #ifdef _LP64
172     if (UseCompressedClassPointers) { // Take care not to kill klass
173       movptr(t1, klass);
174       encode_klass_not_null(t1, rscratch1);
175       movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
176     } else
177 #endif
178     {
179       movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
180     }
181   }
182   if (len->is_valid()) {
183     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
184   }
185 #ifdef _LP64
186   else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
187     xorptr(t1, t1);
188     store_klass_gap(obj, t1);
189   }
190 #endif
191 }
192 
193 
194 // preserves obj, destroys len_in_bytes
195 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
196   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
197   Label done;
198 
199   // len_in_bytes is positive and ptr sized
200   subptr(len_in_bytes, hdr_size_in_bytes);
201   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
202   bind(done);
203 }
204 
205 
206 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
207   assert(obj == rax, "obj must be in rax, for cmpxchg");
208   assert_different_registers(obj, t1, t2); // XXX really?
209   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
210 
211   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
212 
213   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
214 }
215 
216 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) {
217   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
218          "con_size_in_bytes is not multiple of alignment");
219   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
220 
221   initialize_header(obj, klass, noreg, t1, t2);
222 
223   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
224     // clear rest of allocated space
225     const Register t1_zero = t1;
226     const Register index = t2;
227     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
228     int hdr_size_aligned = align_up(hdr_size_in_bytes, BytesPerWord); // klass gap is already cleared by init_header().
229     if (var_size_in_bytes != noreg) {
230       mov(index, var_size_in_bytes);
231       initialize_body(obj, index, hdr_size_aligned, t1_zero);
232     } else if (con_size_in_bytes <= threshold) {
233       // use explicit null stores
234       // code size = 2 + 3*n bytes (n = number of fields to clear)
235       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
236       for (int i = hdr_size_aligned; i < con_size_in_bytes; i += BytesPerWord)
237         movptr(Address(obj, i), t1_zero);
238     } else if (con_size_in_bytes > hdr_size_aligned) {
239       // use loop to null out the fields
240       // code size = 16 bytes for even n (n = number of fields to clear)
241       // initialize last object field first if odd number of fields
242       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
243       movptr(index, (con_size_in_bytes - hdr_size_aligned) >> 3);
244       // initialize last object field if constant size is odd
245       if (((con_size_in_bytes - hdr_size_aligned) & 4) != 0)
246         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
247       // initialize remaining object fields: rdx is a multiple of 2
248       { Label loop;
249         bind(loop);
250         movptr(Address(obj, index, Address::times_8, hdr_size_aligned - (1*BytesPerWord)),
251                t1_zero);
252         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_aligned - (2*BytesPerWord)),
253                t1_zero);)
254         decrement(index);
255         jcc(Assembler::notZero, loop);
256       }
257     }
258   }
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 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) {
269   assert(obj == rax, "obj must be in rax, for cmpxchg");
270   assert_different_registers(obj, len, t1, t2, klass);
271 
272   // determine alignment mask
273   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
274 
275   // check for negative or excessive length
276   cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
277   jcc(Assembler::above, slow_case);
278 
279   const Register arr_size = t2; // okay to be the same
280   // align object end
281   movptr(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
282   lea(arr_size, Address(arr_size, len, f));
283   andptr(arr_size, ~MinObjAlignmentInBytesMask);
284 
285   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
286 
287   initialize_header(obj, klass, len, t1, t2);
288 
289   // clear rest of allocated space
290   const Register len_zero = len;
291   initialize_body(obj, arr_size, base_offset_in_bytes, len_zero);
292 
293   if (CURRENT_ENV->dtrace_alloc_probes()) {
294     assert(obj == rax, "must be");
295     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
296   }
297 
298   verify_oop(obj);
299 }
300 
301 
302 
303 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
304   verify_oop(receiver);
305   // explicit NULL check not needed since load from [klass_offset] causes a trap
306   // check against inline cache
307   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
308   int start_offset = offset();
309 
310   if (UseCompressedClassPointers) {
311     load_klass(rscratch1, receiver, rscratch2);
312     cmpptr(rscratch1, iCache);
313   } else {
314     cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
315   }
316   // if icache check fails, then jump to runtime routine
317   // Note: RECEIVER must still contain the receiver!
318   jump_cc(Assembler::notEqual,
319           RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
320   const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
321   assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
322 }
323 
324 
325 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes, int max_monitors) {
326   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
327   // Make sure there is enough stack space for this method's activation.
328   // Note that we do this before doing an enter(). This matches the
329   // ordering of C2's stack overflow check / rsp decrement and allows
330   // the SharedRuntime stack overflow handling to be consistent
331   // between the two compilers.
332   generate_stack_overflow_check(bang_size_in_bytes);
333 
334   push(rbp);
335   if (PreserveFramePointer) {
336     mov(rbp, rsp);
337   }
338 #if !defined(_LP64) && defined(COMPILER2)
339   if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
340     // c2 leaves fpu stack dirty. Clean it on entry
341     empty_FPU_stack();
342   }
343 #endif // !_LP64 && COMPILER2
344   decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
345 
346 #ifdef _LP64
347   if (UseFastLocking && max_monitors > 0) {
348     Label ok;
349     movptr(rax, Address(r15_thread, JavaThread::lock_stack_current_offset()));
350     addptr(rax, max_monitors * wordSize);
351     cmpptr(rax, Address(r15_thread, JavaThread::lock_stack_limit_offset()));
352     jcc(Assembler::less, ok);
353     assert(StubRoutines::x86::check_lock_stack() != NULL, "need runtime call stub");
354     call(RuntimeAddress(StubRoutines::x86::check_lock_stack()));
355     bind(ok);
356   }
357 #endif
358 
359   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
360   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
361   bs->nmethod_entry_barrier(this, NULL /* slow_path */, NULL /* continuation */);
362 }
363 
364 
365 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
366   increment(rsp, frame_size_in_bytes);  // Does not emit code for frame_size == 0
367   pop(rbp);
368 }
369 
370 
371 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
372   if (breakAtEntry || VerifyFPU) {
373     // Verified Entry first instruction should be 5 bytes long for correct
374     // patching by patch_verified_entry().
375     //
376     // Breakpoint and VerifyFPU have one byte first instruction.
377     // Also first instruction will be one byte "push(rbp)" if stack banging
378     // code is not generated (see build_frame() above).
379     // For all these cases generate long instruction first.
380     fat_nop();
381   }
382   if (breakAtEntry) int3();
383   // build frame
384   IA32_ONLY( verify_FPU(0, "method_entry"); )
385 }
386 
387 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
388   // rbp, + 0: link
389   //     + 1: return address
390   //     + 2: argument with offset 0
391   //     + 3: argument with offset 1
392   //     + 4: ...
393 
394   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
395 }
396 
397 #ifndef PRODUCT
398 
399 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
400   if (!VerifyOops) return;
401   verify_oop_addr(Address(rsp, stack_offset));
402 }
403 
404 void C1_MacroAssembler::verify_not_null_oop(Register r) {
405   if (!VerifyOops) return;
406   Label not_null;
407   testptr(r, r);
408   jcc(Assembler::notZero, not_null);
409   stop("non-null oop required");
410   bind(not_null);
411   verify_oop(r);
412 }
413 
414 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
415 #ifdef ASSERT
416   if (inv_rax) movptr(rax, 0xDEAD);
417   if (inv_rbx) movptr(rbx, 0xDEAD);
418   if (inv_rcx) movptr(rcx, 0xDEAD);
419   if (inv_rdx) movptr(rdx, 0xDEAD);
420   if (inv_rsi) movptr(rsi, 0xDEAD);
421   if (inv_rdi) movptr(rdi, 0xDEAD);
422 #endif
423 }
424 
425 #endif // ifndef PRODUCT