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 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
  39   Label L;
  40   const Register temp_reg = G3_scratch;
  41   // Note: needs more testing of out-of-line vs. inline slow case
  42   verify_oop(receiver);
  43   ld_ptr(receiver, oopDesc::klass_offset_in_bytes(), temp_reg);
  44   cmp(temp_reg, iCache);
  45   brx(Assembler::equal, true, Assembler::pt, L);
  46   delayed()->nop();
  47   AddressLiteral ic_miss(SharedRuntime::get_ic_miss_stub());
  48   jump_to(ic_miss, temp_reg);
  49   delayed()->nop();
  50   align(CodeEntryAlignment);
  51   bind(L);
  52 }
  53 
  54 
  55 void C1_MacroAssembler::explicit_null_check(Register base) {
  56   Unimplemented();
  57 }
  58 
  59 
  60 void C1_MacroAssembler::build_frame(int frame_size_in_bytes) {
  61 
  62   generate_stack_overflow_check(frame_size_in_bytes);
  63   // Create the frame.
  64   save_frame_c1(frame_size_in_bytes);
  65 }
  66 
  67 
  68 void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) {
  69   if (C1Breakpoint) breakpoint_trap();
  70   inline_cache_check(receiver, ic_klass);
  71 }
  72 
  73 
  74 void C1_MacroAssembler::verified_entry() {
  75   if (C1Breakpoint) breakpoint_trap();
  76   // build frame
  77   verify_FPU(0, "method_entry");
  78 }
  79 
  80 
  81 void C1_MacroAssembler::lock_object(Register Rmark, Register Roop, Register Rbox, Register Rscratch, Label& slow_case) {
  82   assert_different_registers(Rmark, Roop, Rbox, Rscratch);
  83 
  84   Label done;
  85 
  86   Address mark_addr(Roop, oopDesc::mark_offset_in_bytes());
  87 
  88   // The following move must be the first instruction of emitted since debug
  89   // information may be generated for it.
  90   // Load object header
  91   ld_ptr(mark_addr, Rmark);
  92 
  93   verify_oop(Roop);
  94 
  95   // save object being locked into the BasicObjectLock
  96   st_ptr(Roop, Rbox, BasicObjectLock::obj_offset_in_bytes());
  97 
  98   if (UseBiasedLocking) {
  99     biased_locking_enter(Roop, Rmark, Rscratch, done, &slow_case);
 100   }
 101 
 102   // Save Rbox in Rscratch to be used for the cas operation
 103   mov(Rbox, Rscratch);
 104 
 105   // and mark it unlocked
 106   or3(Rmark, markOopDesc::unlocked_value, Rmark);
 107 
 108   // save unlocked object header into the displaced header location on the stack
 109   st_ptr(Rmark, Rbox, BasicLock::displaced_header_offset_in_bytes());
 110 
 111   // compare object markOop with Rmark and if equal exchange Rscratch with object markOop
 112   assert(mark_addr.disp() == 0, "cas must take a zero displacement");
 113   casx_under_lock(mark_addr.base(), Rmark, Rscratch, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
 114   // if compare/exchange succeeded we found an unlocked object and we now have locked it
 115   // hence we are done
 116   cmp(Rmark, Rscratch);
 117   brx(Assembler::equal, false, Assembler::pt, done);
 118   delayed()->sub(Rscratch, SP, Rscratch);  //pull next instruction into delay slot
 119   // we did not find an unlocked object so see if this is a recursive case
 120   // sub(Rscratch, SP, Rscratch);
 121   assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
 122   andcc(Rscratch, 0xfffff003, Rscratch);
 123   brx(Assembler::notZero, false, Assembler::pn, slow_case);
 124   delayed()->st_ptr(Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes());
 125   bind(done);
 126 }
 127 
 128 
 129 void C1_MacroAssembler::unlock_object(Register Rmark, Register Roop, Register Rbox, Label& slow_case) {
 130   assert_different_registers(Rmark, Roop, Rbox);
 131 
 132   Label done;
 133 
 134   Address mark_addr(Roop, oopDesc::mark_offset_in_bytes());
 135   assert(mark_addr.disp() == 0, "cas must take a zero displacement");
 136 
 137   if (UseBiasedLocking) {
 138     // load the object out of the BasicObjectLock
 139     ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop);
 140     verify_oop(Roop);
 141     biased_locking_exit(mark_addr, Rmark, done);
 142   }
 143   // Test first it it is a fast recursive unlock
 144   ld_ptr(Rbox, BasicLock::displaced_header_offset_in_bytes(), Rmark);
 145   br_null(Rmark, false, Assembler::pt, done);
 146   delayed()->nop();
 147   if (!UseBiasedLocking) {
 148     // load object
 149     ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop);
 150     verify_oop(Roop);
 151   }
 152 
 153   // Check if it is still a light weight lock, this is is true if we see
 154   // the stack address of the basicLock in the markOop of the object
 155   casx_under_lock(mark_addr.base(), Rbox, Rmark, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
 156   cmp(Rbox, Rmark);
 157 
 158   brx(Assembler::notEqual, false, Assembler::pn, slow_case);
 159   delayed()->nop();
 160   // Done
 161   bind(done);
 162 }
 163 
 164 
 165 void C1_MacroAssembler::try_allocate(
 166   Register obj,                        // result: pointer to object after successful allocation
 167   Register var_size_in_bytes,          // object size in bytes if unknown at compile time; invalid otherwise
 168   int      con_size_in_bytes,          // object size in bytes if   known at compile time
 169   Register t1,                         // temp register
 170   Register t2,                         // temp register
 171   Label&   slow_case                   // continuation point if fast allocation fails
 172 ) {
 173   if (UseTLAB) {
 174     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
 175   } else {
 176     eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, 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 (UseBiasedLocking && !len->is_valid()) {
 184     ld_ptr(klass, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes(), t1);
 185   } else {
 186     set((intx)markOopDesc::prototype(), t1);
 187   }
 188   st_ptr(t1  , obj, oopDesc::mark_offset_in_bytes       ());
 189   st_ptr(klass, obj, oopDesc::klass_offset_in_bytes      ());
 190   if (len->is_valid()) st(len  , obj, arrayOopDesc::length_offset_in_bytes());
 191 }
 192 
 193 
 194 void C1_MacroAssembler::initialize_body(Register base, Register index) {
 195   assert_different_registers(base, index);
 196   Label loop;
 197   bind(loop);
 198   subcc(index, HeapWordSize, index);
 199   brx(Assembler::greaterEqual, true, Assembler::pt, loop);
 200   delayed()->st_ptr(G0, base, index);
 201 }
 202 
 203 
 204 void C1_MacroAssembler::allocate_object(
 205   Register obj,                        // result: pointer to object after successful allocation
 206   Register t1,                         // temp register
 207   Register t2,                         // temp register
 208   Register t3,                         // temp register
 209   int      hdr_size,                   // object header size in words
 210   int      obj_size,                   // object size in words
 211   Register klass,                      // object klass
 212   Label&   slow_case                   // continuation point if fast allocation fails
 213 ) {
 214   assert_different_registers(obj, t1, t2, t3, klass);
 215   assert(klass == G5, "must be G5");
 216 
 217   // allocate space & initialize header
 218   if (!is_simm13(obj_size * wordSize)) {
 219     // would need to use extra register to load
 220     // object size => go the slow case for now
 221     br(Assembler::always, false, Assembler::pt, slow_case);
 222     delayed()->nop();
 223     return;
 224   }
 225   try_allocate(obj, noreg, obj_size * wordSize, t2, t3, slow_case);
 226 
 227   initialize_object(obj, klass, noreg, obj_size * HeapWordSize, t1, t2);
 228 }
 229 
 230 void C1_MacroAssembler::initialize_object(
 231   Register obj,                        // result: pointer to object after successful allocation
 232   Register klass,                      // object klass
 233   Register var_size_in_bytes,          // object size in bytes if unknown at compile time; invalid otherwise
 234   int      con_size_in_bytes,          // object size in bytes if   known at compile time
 235   Register t1,                         // temp register
 236   Register t2                          // temp register
 237   ) {
 238   const int hdr_size_in_bytes = instanceOopDesc::base_offset_in_bytes();
 239 
 240   initialize_header(obj, klass, noreg, t1, t2);
 241 
 242 #ifdef ASSERT
 243   {
 244     Label ok;
 245     ld(klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), t1);
 246     if (var_size_in_bytes != noreg) {
 247       cmp(t1, var_size_in_bytes);
 248     } else {
 249       cmp(t1, con_size_in_bytes);
 250     }
 251     brx(Assembler::equal, false, Assembler::pt, ok);
 252     delayed()->nop();
 253     stop("bad size in initialize_object");
 254     should_not_reach_here();
 255 
 256     bind(ok);
 257   }
 258 
 259 #endif
 260 
 261   // initialize body
 262   const int threshold = 5 * HeapWordSize;              // approximate break even point for code size
 263   if (var_size_in_bytes != noreg) {
 264     // use a loop
 265     add(obj, hdr_size_in_bytes, t1);               // compute address of first element
 266     sub(var_size_in_bytes, hdr_size_in_bytes, t2); // compute size of body
 267     initialize_body(t1, t2);
 268 #ifndef _LP64
 269   } else if (VM_Version::v9_instructions_work() && con_size_in_bytes < threshold * 2) {
 270     // on v9 we can do double word stores to fill twice as much space.
 271     assert(hdr_size_in_bytes % 8 == 0, "double word aligned");
 272     assert(con_size_in_bytes % 8 == 0, "double word aligned");
 273     for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += 2 * HeapWordSize) stx(G0, obj, i);
 274 #endif
 275   } else if (con_size_in_bytes <= threshold) {
 276     // use explicit NULL stores
 277     for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += HeapWordSize)     st_ptr(G0, obj, i);
 278   } else if (con_size_in_bytes > hdr_size_in_bytes) {
 279     // use a loop
 280     const Register base  = t1;
 281     const Register index = t2;
 282     add(obj, hdr_size_in_bytes, base);               // compute address of first element
 283     // compute index = number of words to clear
 284     set(con_size_in_bytes - hdr_size_in_bytes, index);
 285     initialize_body(base, index);
 286   }
 287 
 288   if (CURRENT_ENV->dtrace_alloc_probes()) {
 289     assert(obj == O0, "must be");
 290     call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
 291          relocInfo::runtime_call_type);
 292     delayed()->nop();
 293   }
 294 
 295   verify_oop(obj);
 296 }
 297 
 298 
 299 void C1_MacroAssembler::allocate_array(
 300   Register obj,                        // result: pointer to array after successful allocation
 301   Register len,                        // array length
 302   Register t1,                         // temp register
 303   Register t2,                         // temp register
 304   Register t3,                         // temp register
 305   int      hdr_size,                   // object header size in words
 306   int      elt_size,                   // element size in bytes
 307   Register klass,                      // object klass
 308   Label&   slow_case                   // continuation point if fast allocation fails
 309 ) {
 310   assert_different_registers(obj, len, t1, t2, t3, klass);
 311   assert(klass == G5, "must be G5");
 312   assert(t1 == G1, "must be G1");
 313 
 314   // determine alignment mask
 315   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
 316 
 317   // check for negative or excessive length
 318   // note: the maximum length allowed is chosen so that arrays of any
 319   //       element size with this length are always smaller or equal
 320   //       to the largest integer (i.e., array size computation will
 321   //       not overflow)
 322   set(max_array_allocation_length, t1);
 323   cmp(len, t1);
 324   br(Assembler::greaterUnsigned, false, Assembler::pn, slow_case);
 325 
 326   // compute array size
 327   // note: if 0 <= len <= max_length, len*elt_size + header + alignment is
 328   //       smaller or equal to the largest integer; also, since top is always
 329   //       aligned, we can do the alignment here instead of at the end address
 330   //       computation
 331   const Register arr_size = t1;
 332   switch (elt_size) {
 333     case  1: delayed()->mov(len,    arr_size); break;
 334     case  2: delayed()->sll(len, 1, arr_size); break;
 335     case  4: delayed()->sll(len, 2, arr_size); break;
 336     case  8: delayed()->sll(len, 3, arr_size); break;
 337     default: ShouldNotReachHere();
 338   }
 339   add(arr_size, hdr_size * wordSize + MinObjAlignmentInBytesMask, arr_size); // add space for header & alignment
 340   and3(arr_size, ~MinObjAlignmentInBytesMask, arr_size);                     // align array size
 341 
 342   // allocate space & initialize header
 343   if (UseTLAB) {
 344     tlab_allocate(obj, arr_size, 0, t2, slow_case);
 345   } else {
 346     eden_allocate(obj, arr_size, 0, t2, t3, slow_case);
 347   }
 348   initialize_header(obj, klass, len, t2, t3);
 349 
 350   // initialize body
 351   const Register base  = t2;
 352   const Register index = t3;
 353   add(obj, hdr_size * wordSize, base);               // compute address of first element
 354   sub(arr_size, hdr_size * wordSize, index);         // compute index = number of words to clear
 355   initialize_body(base, index);
 356 
 357   if (CURRENT_ENV->dtrace_alloc_probes()) {
 358     assert(obj == O0, "must be");
 359     call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
 360          relocInfo::runtime_call_type);
 361     delayed()->nop();
 362   }
 363 
 364   verify_oop(obj);
 365 }
 366 
 367 
 368 #ifndef PRODUCT
 369 
 370 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
 371   if (!VerifyOops) return;
 372   verify_oop_addr(Address(SP, stack_offset + STACK_BIAS));
 373 }
 374 
 375 void C1_MacroAssembler::verify_not_null_oop(Register r) {
 376   Label not_null;
 377   br_zero(Assembler::notEqual, false, Assembler::pt, r, not_null);
 378   delayed()->nop();
 379   stop("non-null oop required");
 380   bind(not_null);
 381   if (!VerifyOops) return;
 382   verify_oop(r);
 383 }
 384 
 385 void C1_MacroAssembler::invalidate_registers(bool iregisters, bool lregisters, bool oregisters,
 386                                              Register preserve1, Register preserve2) {
 387   if (iregisters) {
 388     for (int i = 0; i < 6; i++) {
 389       Register r = as_iRegister(i);
 390       if (r != preserve1 && r != preserve2)  set(0xdead, r);
 391     }
 392   }
 393   if (oregisters) {
 394     for (int i = 0; i < 6; i++) {
 395       Register r = as_oRegister(i);
 396       if (r != preserve1 && r != preserve2)  set(0xdead, r);
 397     }
 398   }
 399   if (lregisters) {
 400     for (int i = 0; i < 8; i++) {
 401       Register r = as_lRegister(i);
 402       if (r != preserve1 && r != preserve2)  set(0xdead, r);
 403     }
 404   }
 405 }
 406 
 407 
 408 #endif