1 /*
   2  * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, Red Hat Inc. 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 "c1/c1_MacroAssembler.hpp"
  28 #include "c1/c1_Runtime1.hpp"
  29 #include "classfile/systemDictionary.hpp"
  30 #include "gc/shared/collectedHeap.hpp"
  31 #include "gc/shared/barrierSet.hpp"
  32 #include "gc/shared/barrierSetAssembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "oops/arrayOop.hpp"
  35 #include "oops/markWord.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 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
  43                                   FloatRegister f0, FloatRegister f1,
  44                                   Register result)
  45 {
  46   Label done;
  47   if (is_float) {
  48     fcmps(f0, f1);
  49   } else {
  50     fcmpd(f0, f1);
  51   }
  52   if (unordered_result < 0) {
  53     // we want -1 for unordered or less than, 0 for equal and 1 for
  54     // greater than.
  55     cset(result, NE);  // Not equal or unordered
  56     cneg(result, result, LT);  // Less than or unordered
  57   } else {
  58     // we want -1 for less than, 0 for equal and 1 for unordered or
  59     // greater than.
  60     cset(result, NE);  // Not equal or unordered
  61     cneg(result, result, LO);  // Less than
  62   }
  63 }
  64 
  65 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
  66   const int aligned_mask = BytesPerWord -1;
  67   const int hdr_offset = oopDesc::mark_offset_in_bytes();
  68   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
  69   Label done;
  70   int null_check_offset = -1;
  71 
  72   verify_oop(obj);
  73 
  74   // save object being locked into the BasicObjectLock
  75   str(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
  76 
  77   if (UseBiasedLocking) {
  78     assert(scratch != noreg, "should have scratch register at this point");
  79     null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case);
  80   } else {
  81     null_check_offset = offset();
  82   }
  83 
  84   // Load object header
  85   ldr(hdr, Address(obj, hdr_offset));
  86   // and mark it as unlocked
  87   orr(hdr, hdr, markWord::unlocked_value);
  88 
  89   if (EnableValhalla && !UseBiasedLocking) {
  90     // Mask always_locked bit such that we go to the slow path if object is a value type
  91     andr(hdr, hdr, ~markWord::biased_lock_bit_in_place);
  92   }
  93 
  94   // save unlocked object header into the displaced header location on the stack
  95   str(hdr, Address(disp_hdr, 0));
  96   // test if object header is still the same (i.e. unlocked), and if so, store the
  97   // displaced header address in the object header - if it is not the same, get the
  98   // object header instead
  99   lea(rscratch2, Address(obj, hdr_offset));
 100   cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/NULL);
 101   // if the object header was the same, we're done
 102   // if the object header was not the same, it is now in the hdr register
 103   // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
 104   //
 105   // 1) (hdr & aligned_mask) == 0
 106   // 2) sp <= hdr
 107   // 3) hdr <= sp + page_size
 108   //
 109   // these 3 tests can be done by evaluating the following expression:
 110   //
 111   // (hdr - sp) & (aligned_mask - page_size)
 112   //
 113   // assuming both the stack pointer and page_size have their least
 114   // significant 2 bits cleared and page_size is a power of 2
 115   mov(rscratch1, sp);
 116   sub(hdr, hdr, rscratch1);
 117   ands(hdr, hdr, aligned_mask - os::vm_page_size());
 118   // for recursive locking, the result is zero => save it in the displaced header
 119   // location (NULL in the displaced hdr location indicates recursive locking)
 120   str(hdr, Address(disp_hdr, 0));
 121   // otherwise we don't care about the result and handle locking via runtime call
 122   cbnz(hdr, slow_case);
 123   // done
 124   bind(done);
 125   if (PrintBiasedLockingStatistics) {
 126     lea(rscratch2, ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
 127     addmw(Address(rscratch2, 0), 1, rscratch1);
 128   }
 129   return null_check_offset;
 130 }
 131 
 132 
 133 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
 134   const int aligned_mask = BytesPerWord -1;
 135   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 136   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
 137   Label done;
 138 
 139   if (UseBiasedLocking) {
 140     // load object
 141     ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 142     biased_locking_exit(obj, hdr, done);
 143   }
 144 
 145   // load displaced header
 146   ldr(hdr, Address(disp_hdr, 0));
 147   // if the loaded hdr is NULL we had recursive locking
 148   // if we had recursive locking, we are done
 149   cbz(hdr, done);
 150   if (!UseBiasedLocking) {
 151     // load object
 152     ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 153   }
 154   verify_oop(obj);
 155   // test if object header is pointing to the displaced header, and if so, restore
 156   // the displaced header in the object - if the object header is not pointing to
 157   // the displaced header, get the object header instead
 158   // if the object header was not pointing to the displaced header,
 159   // we do unlocking via runtime call
 160   if (hdr_offset) {
 161     lea(rscratch1, Address(obj, hdr_offset));
 162     cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case);
 163   } else {
 164     cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
 165   }
 166   // done
 167   bind(done);
 168 }
 169 
 170 
 171 // Defines obj, preserves var_size_in_bytes
 172 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
 173   if (UseTLAB) {
 174     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 175   } else {
 176     eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
 177   }
 178 }
 179 
 180 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
 181   assert_different_registers(obj, klass, len);
 182   if (UseBiasedLocking && !len->is_valid()) {
 183     assert_different_registers(obj, klass, len, t1, t2);
 184     ldr(t1, Address(klass, Klass::prototype_header_offset()));
 185   } else {
 186     // This assumes that all prototype bits fit in an int32_t
 187     mov(t1, (int32_t)(intptr_t)markWord::prototype().value());
 188   }
 189   str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
 190 
 191   if (UseCompressedClassPointers) { // Take care not to kill klass
 192     encode_klass_not_null(t1, klass);
 193     strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
 194   } else {
 195     str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 196   }
 197 
 198   if (len->is_valid()) {
 199     strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
 200   } else if (UseCompressedClassPointers) {
 201     store_klass_gap(obj, zr);
 202   }
 203 }
 204 
 205 // preserves obj, destroys len_in_bytes
 206 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
 207   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
 208   Label done;
 209 
 210   // len_in_bytes is positive and ptr sized
 211   subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
 212   br(Assembler::EQ, done);
 213 
 214   // Preserve obj
 215   if (hdr_size_in_bytes)
 216     add(obj, obj, hdr_size_in_bytes);
 217   zero_memory(obj, len_in_bytes, t1);
 218   if (hdr_size_in_bytes)
 219     sub(obj, obj, hdr_size_in_bytes);
 220 
 221   bind(done);
 222 }
 223 
 224 
 225 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
 226   assert_different_registers(obj, t1, t2); // XXX really?
 227   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
 228 
 229   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
 230 
 231   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
 232 }
 233 
 234 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) {
 235   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
 236          "con_size_in_bytes is not multiple of alignment");
 237   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
 238 
 239   initialize_header(obj, klass, noreg, t1, t2);
 240 
 241   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
 242      // clear rest of allocated space
 243      const Register index = t2;
 244      const int threshold = 16 * BytesPerWord;   // approximate break even point for code size (see comments below)
 245      if (var_size_in_bytes != noreg) {
 246        mov(index, var_size_in_bytes);
 247        initialize_body(obj, index, hdr_size_in_bytes, t1);
 248      } else if (con_size_in_bytes <= threshold) {
 249        // use explicit null stores
 250        int i = hdr_size_in_bytes;
 251        if (i < con_size_in_bytes && (con_size_in_bytes % (2 * BytesPerWord))) {
 252          str(zr, Address(obj, i));
 253          i += BytesPerWord;
 254        }
 255        for (; i < con_size_in_bytes; i += 2 * BytesPerWord)
 256          stp(zr, zr, Address(obj, i));
 257      } else if (con_size_in_bytes > hdr_size_in_bytes) {
 258        block_comment("zero memory");
 259       // use loop to null out the fields
 260 
 261        int words = (con_size_in_bytes - hdr_size_in_bytes) / BytesPerWord;
 262        mov(index,  words / 8);
 263 
 264        const int unroll = 8; // Number of str(zr) instructions we'll unroll
 265        int remainder = words % unroll;
 266        lea(rscratch1, Address(obj, hdr_size_in_bytes + remainder * BytesPerWord));
 267 
 268        Label entry_point, loop;
 269        b(entry_point);
 270 
 271        bind(loop);
 272        sub(index, index, 1);
 273        for (int i = -unroll; i < 0; i++) {
 274          if (-i == remainder)
 275            bind(entry_point);
 276          str(zr, Address(rscratch1, i * wordSize));
 277        }
 278        if (remainder == 0)
 279          bind(entry_point);
 280        add(rscratch1, rscratch1, unroll * wordSize);
 281        cbnz(index, loop);
 282 
 283      }
 284   }
 285 
 286   membar(StoreStore);
 287 
 288   if (CURRENT_ENV->dtrace_alloc_probes()) {
 289     assert(obj == r0, "must be");
 290     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
 291   }
 292 
 293   verify_oop(obj);
 294 }
 295 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, int f, Register klass, Label& slow_case) {
 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   mov(rscratch1, (int32_t)max_array_allocation_length);
 303   cmp(len, rscratch1);
 304   br(Assembler::HS, slow_case);
 305 
 306   const Register arr_size = t2; // okay to be the same
 307   // align object end
 308   mov(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
 309   add(arr_size, arr_size, len, ext::uxtw, f);
 310   andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
 311 
 312   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
 313 
 314   initialize_header(obj, klass, len, t1, t2);
 315 
 316   // clear rest of allocated space
 317   const Register len_zero = len;
 318   initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
 319 
 320   membar(StoreStore);
 321 
 322   if (CURRENT_ENV->dtrace_alloc_probes()) {
 323     assert(obj == r0, "must be");
 324     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
 325   }
 326 
 327   verify_oop(obj);
 328 }
 329 
 330 
 331 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
 332   verify_oop(receiver);
 333   // explicit NULL check not needed since load from [klass_offset] causes a trap
 334   // check against inline cache
 335   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
 336 
 337   cmp_klass(receiver, iCache, rscratch1);
 338 }
 339 
 340 
 341 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes, bool needs_stack_repair, Label* verified_value_entry_label) {
 342   assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
 343   // Make sure there is enough stack space for this method's activation.
 344   // Note that we do this before doing an enter().
 345   generate_stack_overflow_check(bang_size_in_bytes);
 346 
 347   guarantee(needs_stack_repair == false, "Stack repair should not be true");
 348   if (verified_value_entry_label != NULL) {
 349     bind(*verified_value_entry_label);
 350   }
 351 
 352   MacroAssembler::build_frame(framesize + 2 * wordSize);
 353 }
 354 
 355 void C1_MacroAssembler::remove_frame(int framesize, bool needs_stack_repair) {
 356 
 357   guarantee(needs_stack_repair == false, "Stack repair should not be true");
 358 
 359   MacroAssembler::remove_frame(framesize + 2 * wordSize);
 360 }
 361 
 362 void C1_MacroAssembler::verified_value_entry() {
 363   if (C1Breakpoint || VerifyFPU || !UseStackBanging) {
 364     // Verified Entry first instruction should be 5 bytes long for correct
 365     // patching by patch_verified_entry().
 366     //
 367     // C1Breakpoint and VerifyFPU have one byte first instruction.
 368     // Also first instruction will be one byte "push(rbp)" if stack banging
 369     // code is not generated (see build_frame() above).
 370     // For all these cases generate long instruction first.
 371     nop();
 372   }
 373   
 374   // build frame
 375   // verify_FPU(0, "method_entry");
 376 }
 377 
 378 int C1_MacroAssembler::scalarized_entry(const CompiledEntrySignature *ces, int frame_size_in_bytes, int bang_size_in_bytes, Label& verified_value_entry_label, bool is_value_ro_entry) {
 379   // This function required to support for ValueTypePassFieldsAsArgs
 380   if (C1Breakpoint || VerifyFPU || !UseStackBanging) {
 381     // Verified Entry first instruction should be 5 bytes long for correct
 382     // patching by patch_verified_entry().
 383     //
 384     // C1Breakpoint and VerifyFPU have one byte first instruction.
 385     // Also first instruction will be one byte "push(rbp)" if stack banging
 386     // code is not generated (see build_frame() above).
 387     // For all these cases generate long instruction first.
 388     nop();
 389   }
 390 
 391   // verify_FPU(0, "method_entry");
 392 
 393   assert(ValueTypePassFieldsAsArgs, "sanity");
 394 
 395   GrowableArray<SigEntry>* sig   = &ces->sig();
 396   GrowableArray<SigEntry>* sig_cc = is_value_ro_entry ? &ces->sig_cc_ro() : &ces->sig_cc();
 397   VMRegPair* regs      = ces->regs();
 398   VMRegPair* regs_cc   = is_value_ro_entry ? ces->regs_cc_ro() : ces->regs_cc();
 399   int args_on_stack    = ces->args_on_stack();
 400   int args_on_stack_cc = is_value_ro_entry ? ces->args_on_stack_cc_ro() : ces->args_on_stack_cc();
 401 
 402   assert(sig->length() <= sig_cc->length(), "Zero-sized value class not allowed!");
 403   BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sig_cc->length());
 404   int args_passed = sig->length();
 405   int args_passed_cc = SigEntry::fill_sig_bt(sig_cc, sig_bt);
 406 
 407   int extra_stack_offset = wordSize; // tos is return address.
 408 
 409   // Create a temp frame so we can call into runtime. It must be properly set up to accomodate GC.
 410   int sp_inc = (args_on_stack - args_on_stack_cc) * VMRegImpl::stack_slot_size;
 411   if (sp_inc > 0) {
 412     sp_inc = align_up(sp_inc, StackAlignmentInBytes);
 413     sub(sp, sp, sp_inc);
 414   } else {
 415     sp_inc = 0;
 416   }
 417 
 418   sub(sp, sp, frame_size_in_bytes);
 419   if (sp_inc > 0) {
 420     int real_frame_size = frame_size_in_bytes +
 421            + wordSize  // pushed rbp
 422            + wordSize  // returned address pushed by the stack extension code
 423            + sp_inc;   // stack extension
 424     mov(rscratch1, real_frame_size);
 425     str(rscratch1, Address(sp, frame_size_in_bytes - wordSize));
 426   }
 427 
 428   // FIXME -- call runtime only if we cannot in-line allocate all the incoming value args.
 429   mov(r1, (intptr_t) ces->method());
 430   if (is_value_ro_entry) {
 431     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::buffer_value_args_no_receiver_id)));
 432   } else {
 433     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::buffer_value_args_id)));
 434   }
 435   int rt_call_offset = offset();
 436 
 437   // Remove the temp frame
 438   add(sp, sp, frame_size_in_bytes);
 439 
 440   int n = shuffle_value_args(true, is_value_ro_entry, extra_stack_offset, sig_bt, sig_cc,
 441                              args_passed_cc, args_on_stack_cc, regs_cc, // from
 442                              args_passed, args_on_stack, regs);         // to
 443   assert(sp_inc == n, "must be");
 444 
 445   if (sp_inc != 0) {
 446     // Do the stack banging here, and skip over the stack repair code in the
 447     // verified_value_entry (which has a different real_frame_size).
 448     assert(sp_inc > 0, "stack should not shrink");
 449     generate_stack_overflow_check(bang_size_in_bytes);
 450     decrement(sp, frame_size_in_bytes);
 451   }
 452 
 453   b(verified_value_entry_label);
 454   return rt_call_offset;
 455 }
 456 
 457 
 458 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
 459   // rbp, + 0: link
 460   //     + 1: return address
 461   //     + 2: argument with offset 0
 462   //     + 3: argument with offset 1
 463   //     + 4: ...
 464 
 465   ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
 466 }
 467 
 468 #ifndef PRODUCT
 469 
 470 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
 471   if (!VerifyOops) return;
 472   verify_oop_addr(Address(sp, stack_offset), "oop");
 473 }
 474 
 475 void C1_MacroAssembler::verify_not_null_oop(Register r) {
 476   if (!VerifyOops) return;
 477   Label not_null;
 478   cbnz(r, not_null);
 479   stop("non-null oop required");
 480   bind(not_null);
 481   verify_oop(r);
 482 }
 483 
 484 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
 485 #ifdef ASSERT
 486   static int nn;
 487   if (inv_r0) mov(r0, 0xDEAD);
 488   if (inv_r19) mov(r19, 0xDEAD);
 489   if (inv_r2) mov(r2, nn++);
 490   if (inv_r3) mov(r3, 0xDEAD);
 491   if (inv_r4) mov(r4, 0xDEAD);
 492   if (inv_r5) mov(r5, 0xDEAD);
 493 #endif
 494 }
 495 #endif // ifndef PRODUCT