1 /*
   2  * Copyright (c) 1999, 2018, 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 "classfile/systemDictionary.hpp"
  30 #include "gc/shared/barrierSet.hpp"
  31 #include "gc/shared/barrierSetAssembler.hpp"
  32 #include "gc/shared/collectedHeap.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "oops/arrayOop.hpp"
  35 #include "oops/markOop.hpp"
  36 #include "runtime/basicLock.hpp"
  37 #include "runtime/biasedLocking.hpp"
  38 #include "runtime/os.hpp"
  39 #include "runtime/sharedRuntime.hpp"
  40 #include "runtime/stubRoutines.hpp"
  41 
  42 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
  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   if (UseBiasedLocking) {
  56     assert(scratch != noreg, "should have scratch register at this point");
  57     null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case);
  58   } else {
  59     null_check_offset = offset();
  60   }
  61 
  62   // Load object header
  63   movptr(hdr, Address(obj, hdr_offset));
  64   // and mark it as unlocked
  65   orptr(hdr, markOopDesc::unlocked_value);
  66   // save unlocked object header into the displaced header location on the stack
  67   movptr(Address(disp_hdr, 0), hdr);
  68   // test if object header is still the same (i.e. unlocked), and if so, store the
  69   // displaced header address in the object header - if it is not the same, get the
  70   // object header instead
  71   MacroAssembler::lock(); // must be immediately before cmpxchg!
  72   cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
  73   // if the object header was the same, we're done
  74   if (PrintBiasedLockingStatistics) {
  75     cond_inc32(Assembler::equal,
  76                ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
  77   }
  78   jcc(Assembler::equal, done);
  79   // if the object header was not the same, it is now in the hdr register
  80   // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
  81   //
  82   // 1) (hdr & aligned_mask) == 0
  83   // 2) rsp <= hdr
  84   // 3) hdr <= rsp + page_size
  85   //
  86   // these 3 tests can be done by evaluating the following expression:
  87   //
  88   // (hdr - rsp) & (aligned_mask - page_size)
  89   //
  90   // assuming both the stack pointer and page_size have their least
  91   // significant 2 bits cleared and page_size is a power of 2
  92   subptr(hdr, rsp);
  93   andptr(hdr, aligned_mask - os::vm_page_size());
  94   // for recursive locking, the result is zero => save it in the displaced header
  95   // location (NULL in the displaced hdr location indicates recursive locking)
  96   movptr(Address(disp_hdr, 0), hdr);
  97   // otherwise we don't care about the result and handle locking via runtime call
  98   jcc(Assembler::notZero, slow_case);
  99   // done
 100   bind(done);
 101   return null_check_offset;
 102 }
 103 
 104 
 105 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
 106   const int aligned_mask = BytesPerWord -1;
 107   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 108   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
 109   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
 110   Label done;
 111 
 112   if (UseBiasedLocking) {
 113     // load object
 114     movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 115     biased_locking_exit(obj, hdr, done);
 116   }
 117 
 118   // load displaced header
 119   movptr(hdr, Address(disp_hdr, 0));
 120   // if the loaded hdr is NULL we had recursive locking
 121   testptr(hdr, hdr);
 122   // if we had recursive locking, we are done
 123   jcc(Assembler::zero, done);
 124   if (!UseBiasedLocking) {
 125     // load object
 126     movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 127   }
 128   verify_oop(obj);
 129   // test if object header is pointing to the displaced header, and if so, restore
 130   // the displaced header in the object - if the object header is not pointing to
 131   // the displaced header, get the object header instead
 132   MacroAssembler::lock(); // must be immediately before cmpxchg!
 133   cmpxchgptr(hdr, Address(obj, hdr_offset));
 134   // if the object header was not pointing to the displaced header,
 135   // we do unlocking via runtime call
 136   jcc(Assembler::notEqual, slow_case);
 137   // done
 138   bind(done);
 139 }
 140 
 141 
 142 // Defines obj, preserves var_size_in_bytes
 143 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
 144   if (UseTLAB) {
 145     tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 146   } else {
 147     eden_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
 148   }
 149 }
 150 
 151 
 152 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
 153   assert_different_registers(obj, klass, len);
 154   if (UseBiasedLocking && !len->is_valid()) {
 155     assert_different_registers(obj, klass, len, t1, t2);
 156     movptr(t1, Address(klass, Klass::prototype_header_offset()));
 157     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
 158   } else {
 159     // This assumes that all prototype bits fit in an int32_t
 160     movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markOopDesc::prototype());
 161   }
 162 #ifdef _LP64
 163   if (UseCompressedClassPointers) { // Take care not to kill klass
 164     movptr(t1, klass);
 165     encode_klass_not_null(t1);
 166     movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
 167   } else
 168 #endif
 169   {
 170     movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
 171   }
 172 
 173   if (len->is_valid()) {
 174     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
 175   }
 176 #ifdef _LP64
 177   else if (UseCompressedClassPointers) {
 178     xorptr(t1, t1);
 179     store_klass_gap(obj, t1);
 180   }
 181 #endif
 182 }
 183 
 184 
 185 // preserves obj, destroys len_in_bytes
 186 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
 187   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
 188   Label done;
 189 
 190   // len_in_bytes is positive and ptr sized
 191   subptr(len_in_bytes, hdr_size_in_bytes);
 192   jcc(Assembler::zero, done);
 193   zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
 194   bind(done);
 195 }
 196 
 197 
 198 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
 199   assert(obj == rax, "obj must be in rax, for cmpxchg");
 200   assert_different_registers(obj, t1, t2); // XXX really?
 201   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
 202 
 203   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
 204 
 205   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
 206 }
 207 
 208 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) {
 209   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
 210          "con_size_in_bytes is not multiple of alignment");
 211   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
 212 
 213   initialize_header(obj, klass, noreg, t1, t2);
 214 
 215   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
 216     // clear rest of allocated space
 217     const Register t1_zero = t1;
 218     const Register index = t2;
 219     const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
 220     if (var_size_in_bytes != noreg) {
 221       mov(index, var_size_in_bytes);
 222       initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
 223     } else if (con_size_in_bytes <= threshold) {
 224       // use explicit null stores
 225       // code size = 2 + 3*n bytes (n = number of fields to clear)
 226       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
 227       for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
 228         movptr(Address(obj, i), t1_zero);
 229     } else if (con_size_in_bytes > hdr_size_in_bytes) {
 230       // use loop to null out the fields
 231       // code size = 16 bytes for even n (n = number of fields to clear)
 232       // initialize last object field first if odd number of fields
 233       xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
 234       movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
 235       // initialize last object field if constant size is odd
 236       if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
 237         movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
 238       // initialize remaining object fields: rdx is a multiple of 2
 239       { Label loop;
 240         bind(loop);
 241         movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
 242                t1_zero);
 243         NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
 244                t1_zero);)
 245         decrement(index);
 246         jcc(Assembler::notZero, loop);
 247       }
 248     }
 249   }
 250 
 251   if (CURRENT_ENV->dtrace_alloc_probes()) {
 252     assert(obj == rax, "must be");
 253     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
 254   }
 255 
 256   verify_oop(obj);
 257 }
 258 
 259 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
 260   assert(obj == rax, "obj must be in rax, for cmpxchg");
 261   assert_different_registers(obj, len, t1, t2, klass);
 262 
 263   // determine alignment mask
 264   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
 265 
 266   // check for negative or excessive length
 267   cmpptr(len, (int32_t)max_array_allocation_length);
 268   jcc(Assembler::above, slow_case);
 269 
 270   const Register arr_size = t2; // okay to be the same
 271   // align object end
 272   movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
 273   lea(arr_size, Address(arr_size, len, f));
 274   andptr(arr_size, ~MinObjAlignmentInBytesMask);
 275 
 276   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
 277 
 278   initialize_header(obj, klass, len, t1, t2);
 279 
 280   // clear rest of allocated space
 281   const Register len_zero = len;
 282   initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
 283 
 284   if (CURRENT_ENV->dtrace_alloc_probes()) {
 285     assert(obj == rax, "must be");
 286     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
 287   }
 288 
 289   verify_oop(obj);
 290 }
 291 
 292 
 293 
 294 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
 295   verify_oop(receiver);
 296   // explicit NULL check not needed since load from [klass_offset] causes a trap
 297   // check against inline cache
 298   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
 299   int start_offset = offset();
 300 
 301   if (UseCompressedClassPointers) {
 302     load_klass(rscratch1, receiver);
 303     cmpptr(rscratch1, iCache);
 304   } else {
 305     cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
 306   }
 307   // if icache check fails, then jump to runtime routine
 308   // Note: RECEIVER must still contain the receiver!
 309   jump_cc(Assembler::notEqual,
 310           RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 311   const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
 312   assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
 313 }
 314 
 315 
 316 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
 317   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
 318   // Make sure there is enough stack space for this method's activation.
 319   // Note that we do this before doing an enter(). This matches the
 320   // ordering of C2's stack overflow check / rsp decrement and allows
 321   // the SharedRuntime stack overflow handling to be consistent
 322   // between the two compilers.
 323   generate_stack_overflow_check(bang_size_in_bytes);
 324 
 325   push(rbp);
 326   if (PreserveFramePointer) {
 327     mov(rbp, rsp);
 328   }
 329 #ifdef TIERED
 330   // c2 leaves fpu stack dirty. Clean it on entry
 331   if (UseSSE < 2 ) {
 332     empty_FPU_stack();
 333   }
 334 #endif // TIERED
 335   decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
 336 
 337   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 338   bs->nmethod_entry_barrier(this);
 339 }
 340 
 341 
 342 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
 343   increment(rsp, frame_size_in_bytes);  // Does not emit code for frame_size == 0
 344   pop(rbp);
 345 }
 346 
 347 
 348 void C1_MacroAssembler::verified_entry() {
 349   if (C1Breakpoint || VerifyFPU || !UseStackBanging) {
 350     // Verified Entry first instruction should be 5 bytes long for correct
 351     // patching by patch_verified_entry().
 352     //
 353     // C1Breakpoint and VerifyFPU have one byte first instruction.
 354     // Also first instruction will be one byte "push(rbp)" if stack banging
 355     // code is not generated (see build_frame() above).
 356     // For all these cases generate long instruction first.
 357     fat_nop();
 358   }
 359   if (C1Breakpoint)int3();
 360   // build frame
 361   verify_FPU(0, "method_entry");
 362 }
 363 
 364 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
 365   // rbp, + 0: link
 366   //     + 1: return address
 367   //     + 2: argument with offset 0
 368   //     + 3: argument with offset 1
 369   //     + 4: ...
 370 
 371   movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
 372 }
 373 
 374 void C1_MacroAssembler::oopmap_metadata(CodeEmitInfo* info) {
 375   MacroAssembler::oopmap_metadata(info != NULL ? info->oop_map()->index() : -1);
 376 }
 377 
 378 #ifndef PRODUCT
 379 
 380 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
 381   if (!VerifyOops) return;
 382   verify_oop_addr(Address(rsp, stack_offset));
 383 }
 384 
 385 void C1_MacroAssembler::verify_not_null_oop(Register r) {
 386   if (!VerifyOops) return;
 387   Label not_null;
 388   testptr(r, r);
 389   jcc(Assembler::notZero, not_null);
 390   stop("non-null oop required");
 391   bind(not_null);
 392   verify_oop(r);
 393 }
 394 
 395 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
 396 #ifdef ASSERT
 397   if (inv_rax) movptr(rax, 0xDEAD);
 398   if (inv_rbx) movptr(rbx, 0xDEAD);
 399   if (inv_rcx) movptr(rcx, 0xDEAD);
 400   if (inv_rdx) movptr(rdx, 0xDEAD);
 401   if (inv_rsi) movptr(rsi, 0xDEAD);
 402   if (inv_rdi) movptr(rdi, 0xDEAD);
 403 #endif
 404 }
 405 
 406 #endif // ifndef PRODUCT