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_IR.hpp"
 27 #include "c1/c1_MacroAssembler.hpp"
 28 #include "c1/c1_Runtime1.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, 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(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
 47   Label done;
 48   int null_check_offset = -1;
 49 
 50   verify_oop(obj);
 51 
 52   // save object being locked into the BasicObjectLock
 53   movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj);
 54 
 55   null_check_offset = offset();
 56 
 57   if (DiagnoseSyncOnValueBasedClasses != 0) {
 58     load_klass(hdr, obj, rklass_decode_tmp);
 59     movl(hdr, Address(hdr, Klass::access_flags_offset()));
 60     testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
 61     jcc(Assembler::notZero, slow_case);
 62   }
 63 
 64   // Load object header
 65   movptr(hdr, Address(obj, hdr_offset));
 66   // and mark it as unlocked
 67   orptr(hdr, markWord::unlocked_value);
 68   // save unlocked object header into the displaced header location on the stack
 69   movptr(Address(disp_hdr, 0), hdr);
 70   // test if object header is still the same (i.e. unlocked), and if so, store the
 71   // displaced header address in the object header - if it is not the same, get the
 72   // object header instead
 73   MacroAssembler::lock(); // must be immediately before cmpxchg!
 74   cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
 75   // if the object header was the same, we're done
 76   jcc(Assembler::equal, done);
 77   // if the object header was not the same, it is now in the hdr register
 78   // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
 79   //
 80   // 1) (hdr & aligned_mask) == 0
 81   // 2) rsp <= hdr
 82   // 3) hdr <= rsp + page_size
 83   //
 84   // these 3 tests can be done by evaluating the following expression:
 85   //
 86   // (hdr - rsp) & (aligned_mask - page_size)
 87   //
 88   // assuming both the stack pointer and page_size have their least
 89   // significant 2 bits cleared and page_size is a power of 2
 90   subptr(hdr, rsp);
 91   andptr(hdr, aligned_mask - os::vm_page_size());
 92   // for recursive locking, the result is zero => save it in the displaced header
 93   // location (NULL in the displaced hdr location indicates recursive locking)
 94   movptr(Address(disp_hdr, 0), hdr);
 95   // otherwise we don't care about the result and handle locking via runtime call
 96   jcc(Assembler::notZero, slow_case);
 97   // done
 98   bind(done);
 99   
100   return null_check_offset;
101 }
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 
133 
134 // Defines obj, preserves var_size_in_bytes
135 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
136   if (UseTLAB) {
137     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
138   } else {
139     eden_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
140   }
141 }
142 
143 
144 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
145   assert_different_registers(obj, klass, len);
146   Register tmp_encode_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
147   // This assumes that all prototype bits fit in an int32_t
148   movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markWord::prototype().value());
149 #ifdef _LP64
150   if (UseCompressedClassPointers) { // Take care not to kill klass
151     movptr(t1, klass);
152     encode_klass_not_null(t1, tmp_encode_klass);
153     movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
154   } else
155 #endif
156   {
157     movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
158   }
159 
160   if (len->is_valid()) {
161     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
162   }
163 #ifdef _LP64
164   else if (UseCompressedClassPointers) {
165     xorptr(t1, t1);
166     store_klass_gap(obj, t1);
167   }
168 #endif
169 }
170 
171 
172 // preserves obj, destroys len_in_bytes
173 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
174   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
175   Label done;
176 
177   // len_in_bytes is positive and ptr sized
178   subptr(len_in_bytes, hdr_size_in_bytes);
179   jcc(Assembler::zero, done);
180   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
181   bind(done);
182 }
183 
184 
185 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
186   assert(obj == rax, "obj must be in rax, for cmpxchg");
187   assert_different_registers(obj, t1, t2); // XXX really?
188   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
189 
190   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
191 
192   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
193 }
194 
195 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) {
196   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
197          "con_size_in_bytes is not multiple of alignment");
198   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
199 
200   initialize_header(obj, klass, noreg, t1, t2);
201 
202   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
203     // clear rest of allocated space
204     const Register t1_zero = t1;
205     const Register index = t2;
206     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
207     if (var_size_in_bytes != noreg) {
208       mov(index, var_size_in_bytes);
209       initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
210     } else if (con_size_in_bytes <= threshold) {
211       // use explicit null stores
212       // code size = 2 + 3*n bytes (n = number of fields to clear)
213       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
214       for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
215         movptr(Address(obj, i), t1_zero);
216     } else if (con_size_in_bytes > hdr_size_in_bytes) {
217       // use loop to null out the fields
218       // code size = 16 bytes for even n (n = number of fields to clear)
219       // initialize last object field first if odd number of fields
220       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
221       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
222       // initialize last object field if constant size is odd
223       if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
224         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
225       // initialize remaining object fields: rdx is a multiple of 2
226       { Label loop;
227         bind(loop);
228         movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
229                t1_zero);
230         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
231                t1_zero);)
232         decrement(index);
233         jcc(Assembler::notZero, loop);
234       }
235     }
236   }
237 
238   if (CURRENT_ENV->dtrace_alloc_probes()) {
239     assert(obj == rax, "must be");
240     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
241   }
242 
243   verify_oop(obj);
244 }
245 
246 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
247   assert(obj == rax, "obj must be in rax, for cmpxchg");
248   assert_different_registers(obj, len, t1, t2, klass);
249 
250   // determine alignment mask
251   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
252 
253   // check for negative or excessive length
254   cmpptr(len, (int32_t)max_array_allocation_length);
255   jcc(Assembler::above, slow_case);
256 
257   const Register arr_size = t2; // okay to be the same
258   // align object end
259   movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
260   lea(arr_size, Address(arr_size, len, f));
261   andptr(arr_size, ~MinObjAlignmentInBytesMask);
262 
263   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
264 
265   initialize_header(obj, klass, len, t1, t2);
266 
267   // clear rest of allocated space
268   const Register len_zero = len;
269   initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
270 
271   if (CURRENT_ENV->dtrace_alloc_probes()) {
272     assert(obj == rax, "must be");
273     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
274   }
275 
276   verify_oop(obj);
277 }
278 
279 
280 
281 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
282   verify_oop(receiver);
283   // explicit NULL check not needed since load from [klass_offset] causes a trap
284   // check against inline cache
285   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
286   int start_offset = offset();
287   Register tmp_load_klass = LP64_ONLY(rscratch2) NOT_LP64(noreg);
288 
289   if (UseCompressedClassPointers) {
290     load_klass(rscratch1, receiver, tmp_load_klass);
291     cmpptr(rscratch1, iCache);
292   } else {
293     cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
294   }
295   // if icache check fails, then jump to runtime routine
296   // Note: RECEIVER must still contain the receiver!
297   jump_cc(Assembler::notEqual,
298           RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
299   const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
300   assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
301 }
302 
303 
304 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
305   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
306   // Make sure there is enough stack space for this method's activation.
307   // Note that we do this before doing an enter(). This matches the
308   // ordering of C2's stack overflow check / rsp decrement and allows
309   // the SharedRuntime stack overflow handling to be consistent
310   // between the two compilers.
311   generate_stack_overflow_check(bang_size_in_bytes);
312 
313   push(rbp);
314   if (PreserveFramePointer) {
315     mov(rbp, rsp);
316   }
317 #if !defined(_LP64) && defined(COMPILER2)
318   if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
319     // c2 leaves fpu stack dirty. Clean it on entry
320     empty_FPU_stack();
321   }
322 #endif // !_LP64 && COMPILER2
323   decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
324 
325   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
326   bs->nmethod_entry_barrier(this);
327 }
328 
329 
330 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
331   increment(rsp, frame_size_in_bytes);  // Does not emit code for frame_size == 0
332   pop(rbp);
333 }
334 
335 
336 void C1_MacroAssembler::verified_entry() {
337   if (C1Breakpoint || VerifyFPU) {
338     // Verified Entry first instruction should be 5 bytes long for correct
339     // patching by patch_verified_entry().
340     //
341     // C1Breakpoint and VerifyFPU have one byte first instruction.
342     // Also first instruction will be one byte "push(rbp)" if stack banging
343     // code is not generated (see build_frame() above).
344     // For all these cases generate long instruction first.
345     fat_nop();
346   }
347   if (C1Breakpoint)int3();
348   // build frame
349   IA32_ONLY( verify_FPU(0, "method_entry"); )
350 }
351 
352 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
353   // rbp, + 0: link
354   //     + 1: return address
355   //     + 2: argument with offset 0
356   //     + 3: argument with offset 1
357   //     + 4: ...
358 
359   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
360 }
361 
362 void C1_MacroAssembler::oopmap_metadata(CodeEmitInfo* info) {
363   MacroAssembler::oopmap_metadata(info != NULL ? info->oop_map()->index() : -1);
364 }
365 
366 #ifndef PRODUCT
367 
368 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
369   if (!VerifyOops) return;
370   verify_oop_addr(Address(rsp, stack_offset));
371 }
372 
373 void C1_MacroAssembler::verify_not_null_oop(Register r) {
374   if (!VerifyOops) return;
375   Label not_null;
376   testptr(r, r);
377   jcc(Assembler::notZero, not_null);
378   stop("non-null oop required");
379   bind(not_null);
380   verify_oop(r);
381 }
382 
383 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
384 #ifdef ASSERT
385   if (inv_rax) movptr(rax, 0xDEAD);
386   if (inv_rbx) movptr(rbx, 0xDEAD);
387   if (inv_rcx) movptr(rcx, 0xDEAD);
388   if (inv_rdx) movptr(rdx, 0xDEAD);
389   if (inv_rsi) movptr(rsi, 0xDEAD);
390   if (inv_rdi) movptr(rdi, 0xDEAD);
391 #endif
392 }
393 
394 #endif // ifndef PRODUCT