1 /*
   2  * Copyright (c) 2016, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016 SAP SE. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "c1/c1_MacroAssembler.hpp"
  29 #include "c1/c1_Runtime1.hpp"
  30 #include "classfile/systemDictionary.hpp"
  31 #include "gc/shared/collectedHeap.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "oops/arrayOop.hpp"
  34 #include "oops/markOop.hpp"
  35 #include "runtime/basicLock.hpp"
  36 #include "runtime/biasedLocking.hpp"
  37 #include "runtime/os.hpp"
  38 #include "runtime/sharedRuntime.hpp"
  39 #include "runtime/stubRoutines.hpp"
  40 
  41 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
  42   Label ic_miss, ic_hit;
  43   verify_oop(receiver);
  44   int klass_offset = oopDesc::klass_offset_in_bytes();
  45 
  46   if (!ImplicitNullChecks || MacroAssembler::needs_explicit_null_check(klass_offset)) {
  47     if (VM_Version::has_CompareBranch()) {
  48       z_cgij(receiver, 0, Assembler::bcondEqual, ic_miss);
  49     } else {
  50       z_ltgr(receiver, receiver);
  51       z_bre(ic_miss);
  52     }
  53   }
  54 
  55   compare_klass_ptr(iCache, klass_offset, receiver, false);
  56   z_bre(ic_hit);
  57 
  58   // If icache check fails, then jump to runtime routine.
  59   // Note: RECEIVER must still contain the receiver!
  60   load_const_optimized(Z_R1_scratch, AddressLiteral(SharedRuntime::get_ic_miss_stub()));
  61   z_br(Z_R1_scratch);
  62   align(CodeEntryAlignment);
  63   bind(ic_hit);
  64 }
  65 
  66 void C1_MacroAssembler::explicit_null_check(Register base) {
  67   ShouldNotCallThis(); // unused
  68 }
  69 
  70 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
  71   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
  72   generate_stack_overflow_check(bang_size_in_bytes);
  73   save_return_pc();
  74   push_frame(frame_size_in_bytes);
  75 }
  76 
  77 void C1_MacroAssembler::verified_entry() {
  78   if (C1Breakpoint) z_illtrap(0xC1);
  79 }
  80 
  81 void C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
  82   const int hdr_offset = oopDesc::mark_offset_in_bytes();
  83   assert_different_registers(hdr, obj, disp_hdr);
  84   NearLabel done;
  85 
  86   verify_oop(obj);
  87 
  88   // Load object header.
  89   z_lg(hdr, Address(obj, hdr_offset));
  90 
  91   // Save object being locked into the BasicObjectLock...
  92   z_stg(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
  93 
  94   if (UseBiasedLocking) {
  95     biased_locking_enter(obj, hdr, Z_R1_scratch, Z_R0_scratch, done, &slow_case);
  96   }
  97 
  98   // and mark it as unlocked.
  99   z_oill(hdr, markOopDesc::unlocked_value);
 100   // Save unlocked object header into the displaced header location on the stack.
 101   z_stg(hdr, Address(disp_hdr, (intptr_t)0));
 102   // Test if object header is still the same (i.e. unlocked), and if so, store the
 103   // displaced header address in the object header. If it is not the same, get the
 104   // object header instead.
 105   z_csg(hdr, disp_hdr, hdr_offset, obj);
 106   // If the object header was the same, we're done.
 107   if (PrintBiasedLockingStatistics) {
 108     Unimplemented();
 109 #if 0
 110     cond_inc32(Assembler::equal,
 111                ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
 112 #endif
 113   }
 114   branch_optimized(Assembler::bcondEqual, done);
 115   // If the object header was not the same, it is now in the hdr register.
 116   // => Test if it is a stack pointer into the same stack (recursive locking), i.e.:
 117   //
 118   // 1) (hdr & markOopDesc::lock_mask_in_place) == 0
 119   // 2) rsp <= hdr
 120   // 3) hdr <= rsp + page_size
 121   //
 122   // These 3 tests can be done by evaluating the following expression:
 123   //
 124   // (hdr - Z_SP) & (~(page_size-1) | markOopDesc::lock_mask_in_place)
 125   //
 126   // assuming both the stack pointer and page_size have their least
 127   // significant 2 bits cleared and page_size is a power of 2
 128   z_sgr(hdr, Z_SP);
 129 
 130   load_const_optimized(Z_R0_scratch, (~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place));
 131   z_ngr(hdr, Z_R0_scratch); // AND sets CC (result eq/ne 0).
 132   // For recursive locking, the result is zero. => Save it in the displaced header
 133   // location (NULL in the displaced hdr location indicates recursive locking).
 134   z_stg(hdr, Address(disp_hdr, (intptr_t)0));
 135   // Otherwise we don't care about the result and handle locking via runtime call.
 136   branch_optimized(Assembler::bcondNotZero, slow_case);
 137   // done
 138   bind(done);
 139 }
 140 
 141 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
 142   const int aligned_mask = BytesPerWord -1;
 143   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 144   assert_different_registers(hdr, obj, disp_hdr);
 145   NearLabel done;
 146 
 147   if (UseBiasedLocking) {
 148     // Load object.
 149     z_lg(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 150     biased_locking_exit(obj, hdr, done);
 151   }
 152 
 153   // Load displaced header.
 154   z_ltg(hdr, Address(disp_hdr, (intptr_t)0));
 155   // If the loaded hdr is NULL we had recursive locking, and we are done.
 156   z_bre(done);
 157   if (!UseBiasedLocking) {
 158     // Load object.
 159     z_lg(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 160   }
 161   verify_oop(obj);
 162   // Test if object header is pointing to the displaced header, and if so, restore
 163   // the displaced header in the object. If the object header is not pointing to
 164   // the displaced header, get the object header instead.
 165   z_csg(disp_hdr, hdr, hdr_offset, obj);
 166   // If the object header was not pointing to the displaced header,
 167   // we do unlocking via runtime call.
 168   branch_optimized(Assembler::bcondNotEqual, slow_case);
 169   // done
 170   bind(done);
 171 }
 172 
 173 void C1_MacroAssembler::try_allocate(
 174   Register obj,                        // result: Pointer to object after successful allocation.
 175   Register var_size_in_bytes,          // Object size in bytes if unknown at compile time; invalid otherwise.
 176   int      con_size_in_bytes,          // Object size in bytes if   known at compile time.
 177   Register t1,                         // Temp register: Must be global register for incr_allocated_bytes.
 178   Label&   slow_case                   // Continuation point if fast allocation fails.
 179 ) {
 180   if (UseTLAB) {
 181     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
 182   } else {
 183     // Allocation in shared Eden not implemented, because sapjvm allocation trace does not allow it.
 184     z_brul(slow_case);
 185   }
 186 }
 187 
 188 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register Rzero, Register t1) {
 189   assert_different_registers(obj, klass, len, t1, Rzero);
 190   if (UseBiasedLocking && !len->is_valid()) {
 191     assert_different_registers(obj, klass, len, t1);
 192     z_lg(t1, Address(klass, Klass::prototype_header_offset()));
 193   } else {
 194     // This assumes that all prototype bits fit in an int32_t.
 195     load_const_optimized(t1, (intx)markOopDesc::prototype());
 196   }
 197   z_stg(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
 198 
 199   if (len->is_valid()) {
 200     // Length will be in the klass gap, if one exists.
 201     z_st(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
 202   } else if (UseCompressedClassPointers) {
 203     store_klass_gap(Rzero, obj);  // Zero klass gap for compressed oops.
 204   }
 205   store_klass(klass, obj, t1);
 206 }
 207 
 208 void C1_MacroAssembler::initialize_body(Register objectFields, Register len_in_bytes, Register Rzero) {
 209   Label done;
 210   assert_different_registers(objectFields, len_in_bytes, Rzero);
 211 
 212   // Initialize object fields.
 213   // See documentation for MVCLE instruction!!!
 214   assert(objectFields->encoding()%2==0, "objectFields must be an even register");
 215   assert(len_in_bytes->encoding() == (objectFields->encoding()+1), "objectFields and len_in_bytes must be a register pair");
 216   assert(Rzero->encoding()%2==1, "Rzero must be an odd register");
 217 
 218   // Use Rzero as src length, then mvcle will copy nothing
 219   // and fill the object with the padding value 0.
 220   move_long_ext(objectFields, as_Register(Rzero->encoding()-1), 0);
 221   bind(done);
 222 }
 223 
 224 void C1_MacroAssembler::allocate_object(
 225   Register obj,                        // Result: pointer to object after successful allocation.
 226   Register t1,                         // temp register
 227   Register t2,                         // temp register: Must be a global register for try_allocate.
 228   int      hdr_size,                   // object header size in words
 229   int      obj_size,                   // object size in words
 230   Register klass,                      // object klass
 231   Label&   slow_case                   // Continuation point if fast allocation fails.
 232 ) {
 233   assert_different_registers(obj, t1, t2, klass);
 234 
 235   // Allocate space and initialize header.
 236   try_allocate(obj, noreg, obj_size * wordSize, t1, slow_case);
 237 
 238   initialize_object(obj, klass, noreg, obj_size * HeapWordSize, t1, t2);
 239 }
 240 
 241 void C1_MacroAssembler::initialize_object(
 242   Register obj,                        // result: Pointer to object after successful allocation.
 243   Register klass,                      // object klass
 244   Register var_size_in_bytes,          // Object size in bytes if unknown at compile time; invalid otherwise.
 245   int      con_size_in_bytes,          // Object size in bytes if   known at compile time.
 246   Register t1,                         // temp register
 247   Register t2                          // temp register
 248  ) {
 249   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
 250          "con_size_in_bytes is not multiple of alignment");
 251   assert(var_size_in_bytes == noreg, "not implemented");
 252   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
 253 
 254   const Register Rzero = t2;
 255 
 256   z_xgr(Rzero, Rzero);
 257   initialize_header(obj, klass, noreg, Rzero, t1);
 258 
 259   // Clear rest of allocated space.
 260   const int threshold = 4 * BytesPerWord;
 261   if (con_size_in_bytes <= threshold) {
 262     // Use explicit null stores.
 263     // code size = 6*n bytes (n = number of fields to clear)
 264     for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
 265       z_stg(Rzero, Address(obj, i));
 266   } else {
 267     // Code size generated by initialize_body() is 16.
 268     Register object_fields = Z_R0_scratch;
 269     Register len_in_bytes  = Z_R1_scratch;
 270     z_la(object_fields, hdr_size_in_bytes, obj);
 271     load_const_optimized(len_in_bytes, con_size_in_bytes - hdr_size_in_bytes);
 272     initialize_body(object_fields, len_in_bytes, Rzero);
 273   }
 274 
 275   // Dtrace support is unimplemented.
 276   //  if (CURRENT_ENV->dtrace_alloc_probes()) {
 277   //    assert(obj == rax, "must be");
 278   //    call(RuntimeAddress(Runtime1::entry_for (Runtime1::dtrace_object_alloc_id)));
 279   //  }
 280 
 281   verify_oop(obj);
 282 }
 283 
 284 void C1_MacroAssembler::allocate_array(
 285   Register obj,                        // result: Pointer to array after successful allocation.
 286   Register len,                        // array length
 287   Register t1,                         // temp register
 288   Register t2,                         // temp register
 289   int      hdr_size,                   // object header size in words
 290   int      elt_size,                   // element size in bytes
 291   Register klass,                      // object klass
 292   Label&   slow_case                   // Continuation point if fast allocation fails.
 293 ) {
 294   assert_different_registers(obj, len, t1, t2, klass);
 295 
 296   // Determine alignment mask.
 297   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
 298 
 299   // Check for negative or excessive length.
 300   compareU64_and_branch(len, (int32_t)max_array_allocation_length, bcondHigh, slow_case);
 301 
 302   // Compute array size.
 303   // Note: If 0 <= len <= max_length, len*elt_size + header + alignment is
 304   // smaller or equal to the largest integer. Also, since top is always
 305   // aligned, we can do the alignment here instead of at the end address
 306   // computation.
 307   const Register arr_size = t2;
 308   switch (elt_size) {
 309     case  1: lgr_if_needed(arr_size, len); break;
 310     case  2: z_sllg(arr_size, len, 1); break;
 311     case  4: z_sllg(arr_size, len, 2); break;
 312     case  8: z_sllg(arr_size, len, 3); break;
 313     default: ShouldNotReachHere();
 314   }
 315   add2reg(arr_size, hdr_size * wordSize + MinObjAlignmentInBytesMask); // Add space for header & alignment.
 316   z_nill(arr_size, (~MinObjAlignmentInBytesMask) & 0xffff);            // Align array size.
 317 
 318   try_allocate(obj, arr_size, 0, t1, slow_case);
 319 
 320   initialize_header(obj, klass, len, noreg, t1);
 321 
 322   // Clear rest of allocated space.
 323   Label done;
 324   Register object_fields = t1;
 325   Register Rzero = Z_R1_scratch;
 326   z_aghi(arr_size, -(hdr_size * BytesPerWord));
 327   z_bre(done); // Jump if size of fields is zero.
 328   z_la(object_fields, hdr_size * BytesPerWord, obj);
 329   z_xgr(Rzero, Rzero);
 330   initialize_body(object_fields, arr_size, Rzero);
 331   bind(done);
 332 
 333   // Dtrace support is unimplemented.
 334   // if (CURRENT_ENV->dtrace_alloc_probes()) {
 335   //   assert(obj == rax, "must be");
 336   //   call(RuntimeAddress(Runtime1::entry_for (Runtime1::dtrace_object_alloc_id)));
 337   // }
 338 
 339   verify_oop(obj);
 340 }
 341 
 342 
 343 #ifndef PRODUCT
 344 
 345 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
 346   Unimplemented();
 347   // if (!VerifyOops) return;
 348   // verify_oop_addr(Address(SP, stack_offset + STACK_BIAS));
 349 }
 350 
 351 void C1_MacroAssembler::verify_not_null_oop(Register r) {
 352   if (!VerifyOops) return;
 353   NearLabel not_null;
 354   compareU64_and_branch(r, (intptr_t)0, bcondNotEqual, not_null);
 355   stop("non-null oop required");
 356   bind(not_null);
 357   verify_oop(r);
 358 }
 359 
 360 void C1_MacroAssembler::invalidate_registers(Register preserve1,
 361                                              Register preserve2,
 362                                              Register preserve3) {
 363   Register dead_value = noreg;
 364   for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {
 365     Register r = as_Register(i);
 366     if (r != preserve1 && r != preserve2 && r != preserve3 && r != Z_SP && r != Z_thread) {
 367       if (dead_value == noreg) {
 368         load_const_optimized(r, 0xc1dead);
 369         dead_value = r;
 370       } else {
 371         z_lgr(r, dead_value);
 372       }
 373     }
 374   }
 375 }
 376 
 377 #endif // !PRODUCT