/*
 * Copyright (c) 2018, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2015-2018, Azul Systems, Inc. All rights reserved.
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
 *
 */

#include "precompiled.hpp"
#include "asm/macroAssembler.inline.hpp"
#include "c1/c1_LIRAssembler.hpp"
#include "c1/c1_MacroAssembler.hpp"
#include "gc/g1/c1/g1BarrierSetC1.hpp"
#include "gc/g1/g1BarrierSet.hpp"
#include "gc/g1/g1BarrierSetAssembler.hpp"
#include "gc/g1/g1BarrierSetRuntime.hpp"
#include "gc/g1/g1CardTable.hpp"
#include "gc/g1/g1ThreadLocalData.hpp"
#include "gc/g1/heapRegion.hpp"
#include "gc/shared/collectedHeap.hpp"
#include "runtime/sharedRuntime.hpp"
#include "runtime/thread.hpp"
#include "interpreter/interp_masm.hpp"
#include "runtime/sharedRuntime.hpp"

#define __ masm->

void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators,
                                                            Register addr, Register count) {
  bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
  // With G1, don't generate the call if we statically know that the target in uninitialized
  if (!dest_uninitialized) {
    __ push(RegSet::range(r0, r3), sp);
    if (count == c_rarg0) {
      if (addr == c_rarg1) {
        // exactly backwards!!
        __ eor(c_rarg0, c_rarg0, c_rarg1);
        __ eor(c_rarg1, c_rarg0, c_rarg1);
        __ eor(c_rarg0, c_rarg0, c_rarg1);
      } else {
        __ mov(c_rarg1, count);
        __ mov(c_rarg0, addr);
      }
    } else {
      __ mov(c_rarg0, addr);
      __ mov(c_rarg1, count);
    }
    __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry), 2);
    __ pop(RegSet::range(r0, r3), sp);
  }
}

void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
                                                             Register start, Register end, Register scratch) {
  // must compute element count unless barrier set interface is changed (other platforms supply count)
  assert_different_registers(start, end, scratch);
  __ lea(scratch, Address(end, BytesPerHeapOop));
  __ sub(scratch, scratch, start);               // subtract start to get #bytes
  __ lsr(scratch, scratch, LogBytesPerHeapOop);  // convert to element count
  __ mov(c_rarg0, start);
  __ mov(c_rarg1, scratch);
  __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry), 2);
}


void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm,
                                                 Address obj,
                                                 Register pre_val,
                                                 Register thread,
                                                 Register tmp,
                                                 bool tosca_live,
                                                 bool expand_call) {
  // If expand_call is true then we expand the call_VM_leaf macro
  // directly to skip generating the check by
  // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.

  assert(thread == rthread, "must be");

  Label done;
  Label runtime;

  assert(pre_val != noreg, "check this code");

  assert_different_registers(pre_val, tmp);
  if (obj.get_mode() != Address::no_mode)
    assert(!obj.uses(pre_val) && !obj.uses(tmp), "destroys register");

  Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
  Address index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
  Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));


  // Is marking active?
  if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
    __ ldr(tmp, in_progress);
  } else {
    assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
    __ ldrb(tmp, in_progress);
  }
  __ cbz(tmp, done);

  // Do we need to load the previous value?
  if (obj.get_mode() != Address::no_mode) {
    __ load_heap_oop(pre_val, obj, noreg, noreg, AS_RAW);
  }

  // Is the previous value null?
  __ cbz(pre_val, done);

  // Can we store original value in the thread's buffer?
  // Is index == 0?
  // (The index field is typed as size_t.)

  __ ldr(tmp, index);                      // tmp := *index_adr
  __ cbz(tmp, runtime);                    // tmp == 0?
                                        // If yes, goto runtime

  __ sub(tmp, tmp, wordSize);              // tmp := tmp - wordSize
  __ str(tmp, index);                      // *index_adr := tmp
  __ ldr(rscratch1, buffer);
  __ add(tmp, tmp, rscratch1);             // tmp := tmp + *buffer_adr

  // Record the previous value
  __ str(pre_val, Address(tmp));
  __ b(done);

  __ bind(runtime);
  // save the live input values
  __ push(r0->bit(tosca_live) | obj.reg_bits() | pre_val->bit(true) | lr->bit(true), sp);

  // Calling the runtime using the regular call_VM_leaf mechanism generates
  // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
  // that checks that the *(rfp+frame::interpreter_frame_last_sp) == NULL.
  //
  // If we care generating the pre-barrier without a frame (e.g. in the
  // intrinsified Reference.get() routine) then ebp might be pointing to
  // the caller frame and so this check will most likely fail at runtime.
  //
  // Expanding the call directly bypasses the generation of the check.
  // So when we do not have have a full interpreter frame on the stack
  // expand_call should be passed true.

  if (expand_call) {
    assert(pre_val != c_rarg1, "smashed arg");
    __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
  } else {
    __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
  }

  __ pop(r0->bit(tosca_live) | obj.reg_bits() | pre_val->bit(true) | lr->bit(true), sp);

  __ bind(done);
}

void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
                                                  Address store_addr,
                                                  Register new_val,
                                                  Register thread,
                                                  Register tmp,
                                                  Register tmp2) {
  assert(thread == rthread, "must be");

  Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
  Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));

  BarrierSet* bs = BarrierSet::barrier_set();
  CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
  CardTable* ct = ctbs->card_table();
  assert(sizeof(*ct->byte_map_base()) == sizeof(jbyte), "adjust this code");

  Label done;
  Label runtime;

  // Does store cross heap regions?

  __ lea(tmp2, store_addr);
  __ eor(tmp, tmp2, new_val);
  __ lsrs(tmp, tmp, HeapRegion::LogOfHRGrainBytes);
  __ b(done, Assembler::EQ);

  // crosses regions, storing NULL?

  __ cbz(new_val, done);

  // storing region crossing non-NULL, is card already dirty?

  assert(sizeof(*ct->byte_map_base()) == sizeof(jbyte), "adjust this code");
  const Register card_addr = tmp;

  __ lsr(card_addr, tmp2, CardTable::card_shift);

  //ExternalAddress cardtable((address) ct->byte_map_base());
  __ mov(tmp2, (unsigned)ct->byte_map_base());

  // get the address of the card
  __ add(card_addr, card_addr, tmp2);
  __ ldrb(tmp2, Address(card_addr));
  __ cmp(tmp2, (int)G1CardTable::g1_young_card_val());
  __ b(done, Assembler::EQ);

  assert((int)CardTable::dirty_card_val() == 0, "must be 0");

  __ membar(Assembler::StoreLoad);

  __ ldrb(tmp2, Address(card_addr));
  __ cbz(tmp2, done);

  // storing a region crossing, non-NULL oop, card is clean.
  // dirty card and log.
  __ mov(rscratch1, 0);
  __ strb(rscratch1, Address(card_addr));

  __ ldr(rscratch1, queue_index);
  __ cbz(rscratch1, runtime);
  __ sub(rscratch1, rscratch1, wordSize);
  __ str(rscratch1, queue_index);

  __ ldr(tmp2, buffer);
  __ str(card_addr, Address(tmp2, rscratch1));
  __ b(done);

  __ bind(runtime);
  // save the live input values
  __ push(store_addr.reg_bits() | new_val->bit(true), sp);
  __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
  __ pop(store_addr.reg_bits() | new_val->bit(true), sp);

  __ bind(done);
}

void G1BarrierSetAssembler::load_word_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
                                         Register dst, Address src, Register tmp1, Register tmp_thread) {
  bool on_oop = type == T_OBJECT || type == T_ARRAY;
  bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
  bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
  bool on_reference = on_weak || on_phantom;
  ModRefBarrierSetAssembler::load_word_at(masm, decorators, type, dst, src, tmp1, tmp_thread);
  if (on_oop && on_reference) {
    // LR is live.  It must be saved around calls.
    // Generate the G1 pre-barrier code to log the value of
    // the referent field in an SATB buffer.
    g1_write_barrier_pre(masm /* masm */,
                         Address() /* obj */,
                         dst /* pre_val */,
                         rthread /* thread */,
                         tmp1 /* tmp */,
                         true /* tosca_live */,
                         true /* expand_call */);
  }
}

void G1BarrierSetAssembler::load_tos_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
                                        Address src, Register tmp1, Register tmp_thread) {
  bool on_oop = type == T_OBJECT || type == T_ARRAY;
  bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
  bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
  bool on_reference = on_weak || on_phantom;
  ModRefBarrierSetAssembler::load_tos_at(masm, decorators, type, src, tmp1, tmp_thread);
  if (on_oop && on_reference) {
    // Generate the G1 pre-barrier code to log the value of
    // the referent field in an SATB buffer.
    g1_write_barrier_pre(masm /* masm */,
                         Address() /* obj */,
                         r0 /* pre_val */, // atos is in r0
                         rthread /* thread */,
                         tmp1 /* tmp */,
                         true /* tosca_live */,
                         true /* expand_call */);
  }
}

void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
                                         Address dst, Register val, Register tmp1, Register tmp2) {
  g1_write_barrier_pre(masm,
                       dst /* obj */,
                       tmp2 /* pre_val */,
                       rthread /* thread */,
                       tmp1  /* tmp */,
                       val != noreg /* tosca_live */,
                       false /* expand_call */);

  if (val == noreg) {
    BarrierSetAssembler::store_word_at(masm, decorators, type, dst, noreg, tmp1, noreg);
  } else {
    BarrierSetAssembler::store_word_at(masm, decorators, type, dst, val, noreg, noreg);
    g1_write_barrier_post(masm,
                          dst /* store_adr */,
                          val /* new_val */,
                          rthread /* thread */,
                          tmp1 /* tmp */,
                          tmp2 /* tmp2 */);
  }

}

#ifdef COMPILER1

#undef __
#define __ ce->masm()->

void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
  G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
  // At this point we know that marking is in progress.
  // If do_load() is true then we have to emit the
  // load of the previous value; otherwise it has already
  // been loaded into _pre_val.

  __ bind(*stub->entry());

  assert(stub->pre_val()->is_register(), "Precondition.");

  Register pre_val_reg = stub->pre_val()->as_register();

  if (stub->do_load()) {
    ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/, false /*unaligned*/);
  }
  __ cbz(pre_val_reg, *stub->continuation());
  ce->store_parameter(stub->pre_val()->as_register(), 0);
  __ far_call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
  __ b(*stub->continuation());
}

void G1BarrierSetAssembler::gen_post_barrier_stub(LIR_Assembler* ce, G1PostBarrierStub* stub) {
  G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
  __ bind(*stub->entry());
  assert(stub->addr()->is_register(), "Precondition.");
  assert(stub->new_val()->is_register(), "Precondition.");
  Register new_val_reg = stub->new_val()->as_register();
  __ cbz(new_val_reg, *stub->continuation());
  ce->store_parameter(stub->addr()->as_pointer_register(), 0);
  __ far_call(RuntimeAddress(bs->post_barrier_c1_runtime_code_blob()->code_begin()));
  __ b(*stub->continuation());
}

#undef __

#define __ sasm->

void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
  __ prologue("g1_pre_barrier", false);

  // arg0 : previous value of memory

  BarrierSet* bs = BarrierSet::barrier_set();

  const Register pre_val = r0;
  const Register thread = rthread;
  const Register tmp = rscratch1;

  Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
  Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
  Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));

  Label done;
  Label runtime;

  // Is marking still active?
  if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
    __ ldr(tmp, in_progress);
  } else {
    assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
    __ ldrb(tmp, in_progress);
  }
  __ cbz(tmp, done);

  // Can we store original value in the thread's buffer?
  __ ldr(tmp, queue_index);
  __ cbz(tmp, runtime);

  __ sub(tmp, tmp, wordSize);
  __ str(tmp, queue_index);
  __ ldr(rscratch2, buffer);
  __ add(tmp, tmp, rscratch2);
  __ load_parameter(0, rscratch2);
  __ str(rscratch2, Address(tmp, 0));
  __ b(done);

  __ bind(runtime);
  __ push_call_clobbered_registers();
  __ load_parameter(0, pre_val);
  __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
  __ pop_call_clobbered_registers();
  __ bind(done);

  __ epilogue();
}

void G1BarrierSetAssembler::generate_c1_post_barrier_runtime_stub(StubAssembler* sasm) {
  __ prologue("g1_post_barrier", false);

  // arg0: store_address
  Address store_addr(rfp, 2*BytesPerWord);

  BarrierSet* bs = BarrierSet::barrier_set();
  CardTableBarrierSet* ctbs = barrier_set_cast<CardTableBarrierSet>(bs);
  CardTable* ct = ctbs->card_table();
  assert(sizeof(*ct->byte_map_base()) == sizeof(jbyte), "adjust this code");

  Label done;
  Label runtime;

  // At this point we know new_value is non-NULL and the new_value crosses regions.
  // Must check to see if card is already dirty

  const Register thread = rthread;

  Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
  Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));

  const Register card_addr = rscratch2;
  ExternalAddress cardtable((address) ct->byte_map_base());

  __ load_parameter(0, card_addr);
  __ lsr(card_addr, card_addr, CardTable::card_shift);
  __ mov(rscratch1, cardtable);
  __ add(card_addr, card_addr, rscratch1);
  __ ldrb(rscratch1, Address(card_addr));
  __ cmp(rscratch1, (int)G1CardTable::g1_young_card_val());
  __ b(done, Assembler::EQ);

  assert((int)CardTable::dirty_card_val() == 0, "must be 0");

  __ membar(Assembler::StoreLoad);
  __ ldrb(rscratch1, Address(card_addr));
  __ cbz(rscratch1, done);

  // storing region crossing non-NULL, card is clean.
  // dirty card and log.
  __ mov(rscratch1, 0);
  __ strb(rscratch1, Address(card_addr));

  __ ldr(rscratch1, queue_index);
  __ cbz(rscratch1, runtime);
  __ sub(rscratch1, rscratch1, wordSize);
  __ str(rscratch1, queue_index);

  // Reuse LR to hold buffer_addr
  const Register buffer_addr = lr;

  __ ldr(buffer_addr, buffer);
  __ str(card_addr, Address(buffer_addr, rscratch1));
  __ b(done);

  __ bind(runtime);
  __ push_call_clobbered_registers();
  __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
  __ pop_call_clobbered_registers();
  __ bind(done);
  __ epilogue();
}

#undef __

#endif // COMPILER1