1 /*
   2  * Copyright (c) 1999, 2010, 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 "classfile/systemDictionary.hpp"
  29 #include "gc_interface/collectedHeap.hpp"
  30 #include "interpreter/interpreter.hpp"
  31 #include "oops/arrayOop.hpp"
  32 #include "oops/markOop.hpp"
  33 #include "runtime/basicLock.hpp"
  34 #include "runtime/biasedLocking.hpp"
  35 #include "runtime/os.hpp"
  36 #include "runtime/stubRoutines.hpp"
  37 
  38 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
  39   const int aligned_mask = BytesPerWord -1;
  40   const int hdr_offset = oopDesc::mark_offset_in_bytes();
  41   assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
  42   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
  43   Label done;
  44   int null_check_offset = -1;
  45 
  46   verify_oop(obj);
  47 
  48   // save object being locked into the BasicObjectLock
  49   movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj);
  50 
  51   if (UseBiasedLocking) {
  52     assert(scratch != noreg, "should have scratch register at this point");
  53     null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case);
  54   } else {
  55     null_check_offset = offset();
  56   }
  57 
  58   // Load object header
  59   movptr(hdr, Address(obj, hdr_offset));
  60   // and mark it as unlocked
  61   orptr(hdr, markOopDesc::unlocked_value);
  62   // save unlocked object header into the displaced header location on the stack
  63   movptr(Address(disp_hdr, 0), hdr);
  64   // test if object header is still the same (i.e. unlocked), and if so, store the
  65   // displaced header address in the object header - if it is not the same, get the
  66   // object header instead
  67   if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg!
  68   cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
  69   // if the object header was the same, we're done
  70   if (PrintBiasedLockingStatistics) {
  71     cond_inc32(Assembler::equal,
  72                ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
  73   }
  74   jcc(Assembler::equal, done);
  75   // if the object header was not the same, it is now in the hdr register
  76   // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
  77   //
  78   // 1) (hdr & aligned_mask) == 0
  79   // 2) rsp <= hdr
  80   // 3) hdr <= rsp + page_size
  81   //
  82   // these 3 tests can be done by evaluating the following expression:
  83   //
  84   // (hdr - rsp) & (aligned_mask - page_size)
  85   //
  86   // assuming both the stack pointer and page_size have their least
  87   // significant 2 bits cleared and page_size is a power of 2
  88   subptr(hdr, rsp);
  89   andptr(hdr, aligned_mask - os::vm_page_size());
  90   // for recursive locking, the result is zero => save it in the displaced header
  91   // location (NULL in the displaced hdr location indicates recursive locking)
  92   movptr(Address(disp_hdr, 0), hdr);
  93   // otherwise we don't care about the result and handle locking via runtime call
  94   jcc(Assembler::notZero, slow_case);
  95   // done
  96   bind(done);
  97   return null_check_offset;
  98 }
  99 
 100 
 101 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
 102   const int aligned_mask = BytesPerWord -1;
 103   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 104   assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
 105   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
 106   Label done;
 107 
 108   if (UseBiasedLocking) {
 109     // load object
 110     movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 111     biased_locking_exit(obj, hdr, done);
 112   }
 113 
 114   // load displaced header
 115   movptr(hdr, Address(disp_hdr, 0));
 116   // if the loaded hdr is NULL we had recursive locking
 117   testptr(hdr, hdr);
 118   // if we had recursive locking, we are done
 119   jcc(Assembler::zero, done);
 120   if (!UseBiasedLocking) {
 121     // load object
 122     movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 123   }
 124   verify_oop(obj);
 125   // test if object header is pointing to the displaced header, and if so, restore
 126   // the displaced header in the object - if the object header is not pointing to
 127   // the displaced header, get the object header instead
 128   if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg!
 129   cmpxchgptr(hdr, Address(obj, hdr_offset));
 130   // if the object header was not pointing to the displaced header,
 131   // we do unlocking via runtime call
 132   jcc(Assembler::notEqual, slow_case);
 133   // done
 134   bind(done);
 135 }
 136 
 137 
 138 // Defines obj, preserves var_size_in_bytes
 139 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
 140   if (UseTLAB) {
 141     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 142   } else {
 143     eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
 144   }
 145 }
 146 
 147 
 148 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
 149   assert_different_registers(obj, klass, len);
 150   if (UseBiasedLocking && !len->is_valid()) {
 151     assert_different_registers(obj, klass, len, t1, t2);
 152     movptr(t1, Address(klass, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
 153     movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
 154   } else {
 155     // This assumes that all prototype bits fit in an int32_t
 156     movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markOopDesc::prototype());
 157   }
 158 #ifdef _LP64
 159   if (UseCompressedOops) { // Take care not to kill klass
 160     movptr(t1, klass);
 161     encode_heap_oop_not_null(t1);
 162     movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
 163   } else 
 164 #endif
 165   {
 166     movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
 167   }
 168 
 169   if (len->is_valid()) {
 170     movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
 171   }
 172 #ifdef _LP64 
 173   else if (UseCompressedOops) {
 174     xorptr(t1, t1);
 175     store_klass_gap(obj, t1);
 176   }
 177 #endif
 178 }
 179 
 180 
 181 // preserves obj, destroys len_in_bytes
 182 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
 183   Label done;
 184   assert(obj != len_in_bytes && obj != t1 && t1 != len_in_bytes, "registers must be different");
 185   assert((hdr_size_in_bytes & (BytesPerWord - 1)) == 0, "header size is not a multiple of BytesPerWord");
 186   Register index = len_in_bytes;
 187   // index is positive and ptr sized
 188   subptr(index, hdr_size_in_bytes);
 189   jcc(Assembler::zero, done);
 190   // initialize topmost word, divide index by 2, check if odd and test if zero
 191   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 192 #ifdef ASSERT
 193   { Label L;
 194     testptr(index, BytesPerWord - 1);
 195     jcc(Assembler::zero, L);
 196     stop("index is not a multiple of BytesPerWord");
 197     bind(L);
 198   }
 199 #endif
 200   xorptr(t1, t1);    // use _zero reg to clear memory (shorter code)
 201   if (UseIncDec) {
 202     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 203   } else {
 204     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 205     shrptr(index, 1);
 206   }
 207 #ifndef _LP64
 208   // index could have been not a multiple of 8 (i.e., bit 2 was set)
 209   { Label even;
 210     // note: if index was a multiple of 8, than it cannot
 211     //       be 0 now otherwise it must have been 0 before
 212     //       => if it is even, we don't need to check for 0 again
 213     jcc(Assembler::carryClear, even);
 214     // clear topmost word (no jump needed if conditional assignment would work here)
 215     movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 0*BytesPerWord), t1);
 216     // index could be 0 now, need to check again
 217     jcc(Assembler::zero, done);
 218     bind(even);
 219   }
 220 #endif // !_LP64
 221   // initialize remaining object fields: rdx is a multiple of 2 now
 222   { Label loop;
 223     bind(loop);
 224     movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 1*BytesPerWord), t1);
 225     NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 2*BytesPerWord), t1);)
 226     decrement(index);
 227     jcc(Assembler::notZero, loop);
 228   }
 229 
 230   // done
 231   bind(done);
 232 }
 233 
 234 
 235 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
 236   assert(obj == rax, "obj must be in rax, for cmpxchg");
 237   assert(obj != t1 && obj != t2 && t1 != t2, "registers must be different"); // XXX really?
 238   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
 239 
 240   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
 241 
 242   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2);
 243 }
 244 
 245 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2) {
 246   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
 247          "con_size_in_bytes is not multiple of alignment");
 248   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
 249 
 250   initialize_header(obj, klass, noreg, t1, t2);
 251 
 252   // clear rest of allocated space
 253   const Register t1_zero = t1;
 254   const Register index = t2;
 255   const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
 256   if (var_size_in_bytes != noreg) {
 257     mov(index, var_size_in_bytes);
 258     initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
 259   } else if (con_size_in_bytes <= threshold) {
 260     // use explicit null stores
 261     // code size = 2 + 3*n bytes (n = number of fields to clear)
 262     xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
 263     for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
 264       movptr(Address(obj, i), t1_zero);
 265   } else if (con_size_in_bytes > hdr_size_in_bytes) {
 266     // use loop to null out the fields
 267     // code size = 16 bytes for even n (n = number of fields to clear)
 268     // initialize last object field first if odd number of fields
 269     xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
 270     movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
 271     // initialize last object field if constant size is odd
 272     if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
 273       movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
 274     // initialize remaining object fields: rdx is a multiple of 2
 275     { Label loop;
 276       bind(loop);
 277       movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
 278              t1_zero);
 279       NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
 280              t1_zero);)
 281       decrement(index);
 282       jcc(Assembler::notZero, loop);
 283     }
 284   }
 285 
 286   if (CURRENT_ENV->dtrace_alloc_probes()) {
 287     assert(obj == rax, "must be");
 288     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
 289   }
 290 
 291   verify_oop(obj);
 292 }
 293 
 294 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
 295   assert(obj == rax, "obj must be in rax, for cmpxchg");
 296   assert_different_registers(obj, len, t1, t2, klass);
 297 
 298   // determine alignment mask
 299   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
 300 
 301   // check for negative or excessive length
 302   cmpptr(len, (int32_t)max_array_allocation_length);
 303   jcc(Assembler::above, slow_case);
 304 
 305   const Register arr_size = t2; // okay to be the same
 306   // align object end
 307   movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
 308   lea(arr_size, Address(arr_size, len, f));
 309   andptr(arr_size, ~MinObjAlignmentInBytesMask);
 310 
 311   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
 312 
 313   initialize_header(obj, klass, len, t1, t2);
 314 
 315   // clear rest of allocated space
 316   const Register len_zero = len;
 317   initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
 318 
 319   if (CURRENT_ENV->dtrace_alloc_probes()) {
 320     assert(obj == rax, "must be");
 321     call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
 322   }
 323 
 324   verify_oop(obj);
 325 }
 326 
 327 
 328 
 329 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
 330   verify_oop(receiver);
 331   // explicit NULL check not needed since load from [klass_offset] causes a trap
 332   // check against inline cache
 333   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
 334   int start_offset = offset();
 335 
 336   if (UseCompressedOops) {
 337     load_klass(rscratch1, receiver);
 338     cmpptr(rscratch1, iCache);
 339   } else {
 340     cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
 341   }
 342   // if icache check fails, then jump to runtime routine
 343   // Note: RECEIVER must still contain the receiver!
 344   jump_cc(Assembler::notEqual,
 345           RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 346   const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
 347   assert(UseCompressedOops || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
 348 }
 349 
 350 
 351 void C1_MacroAssembler::build_frame(int frame_size_in_bytes) {
 352   // Make sure there is enough stack space for this method's activation.
 353   // Note that we do this before doing an enter(). This matches the
 354   // ordering of C2's stack overflow check / rsp decrement and allows
 355   // the SharedRuntime stack overflow handling to be consistent
 356   // between the two compilers.
 357   generate_stack_overflow_check(frame_size_in_bytes);
 358 
 359   push(rbp);
 360 #ifdef TIERED
 361   // c2 leaves fpu stack dirty. Clean it on entry
 362   if (UseSSE < 2 ) {
 363     empty_FPU_stack();
 364   }
 365 #endif // TIERED
 366   decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
 367 }
 368 
 369 
 370 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
 371   increment(rsp, frame_size_in_bytes);  // Does not emit code for frame_size == 0
 372   pop(rbp);
 373 }
 374 
 375 
 376 void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) {
 377   if (C1Breakpoint) int3();
 378   inline_cache_check(receiver, ic_klass);
 379 }
 380 
 381 
 382 void C1_MacroAssembler::verified_entry() {
 383   if (C1Breakpoint)int3();
 384   // build frame
 385   verify_FPU(0, "method_entry");
 386 }
 387 
 388 
 389 #ifndef PRODUCT
 390 
 391 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
 392   if (!VerifyOops) return;
 393   verify_oop_addr(Address(rsp, stack_offset));
 394 }
 395 
 396 void C1_MacroAssembler::verify_not_null_oop(Register r) {
 397   if (!VerifyOops) return;
 398   Label not_null;
 399   testptr(r, r);
 400   jcc(Assembler::notZero, not_null);
 401   stop("non-null oop required");
 402   bind(not_null);
 403   verify_oop(r);
 404 }
 405 
 406 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
 407 #ifdef ASSERT
 408   if (inv_rax) movptr(rax, 0xDEAD);
 409   if (inv_rbx) movptr(rbx, 0xDEAD);
 410   if (inv_rcx) movptr(rcx, 0xDEAD);
 411   if (inv_rdx) movptr(rdx, 0xDEAD);
 412   if (inv_rsi) movptr(rsi, 0xDEAD);
 413   if (inv_rdi) movptr(rdi, 0xDEAD);
 414 #endif
 415 }
 416 
 417 #endif // ifndef PRODUCT